diff options
Diffstat (limited to 'src/ipcpd/unicast')
73 files changed, 13204 insertions, 6030 deletions
diff --git a/src/ipcpd/unicast/CMakeLists.txt b/src/ipcpd/unicast/CMakeLists.txt index 035ee5f4..2373f877 100644 --- a/src/ipcpd/unicast/CMakeLists.txt +++ b/src/ipcpd/unicast/CMakeLists.txt @@ -1,74 +1,52 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) +# Unicast IPCP build configuration -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) +protobuf_generate_c(DHT_PROTO_SRCS DHT_PROTO_HDRS + "${CMAKE_CURRENT_SOURCE_DIR}/dir/dht.proto") -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -set(IPCP_UNICAST_TARGET ipcpd-unicast CACHE INTERNAL "") - -protobuf_generate_c(KAD_PROTO_SRCS KAD_PROTO_HDRS kademlia.proto) - -math(EXPR PFT_EXPR "1 << 12") -set(PFT_SIZE ${PFT_EXPR} CACHE STRING - "Size of the PDU forwarding table") -if (HAVE_FUSE) - set(IPCP_FLOW_STATS TRUE CACHE BOOL - "Enable flow statistics tracking in IPCP") - if (IPCP_FLOW_STATS) - message(STATUS "IPCP flow statistics enabled") - else () - message(STATUS "IPCP flow statistics disabled") - endif () -endif () - - -set(SOURCE_FILES - # Add source files here - addr_auth.c +set(UNICAST_SOURCES + addr-auth.c ca.c + cap.c connmgr.c - dht.c dir.c dt.c - enroll.c fa.c main.c pff.c routing.c psched.c - # Add policies last - pol/pft.c - pol/flat.c - pol/link_state.c - pol/graph.c - pol/simple_pff.c - pol/alternate_pff.c - pol/multipath_pff.c - pol/ca-mb-ecn.c - pol/ca-nop.c - ) - -add_executable(ipcpd-unicast ${SOURCE_FILES} ${IPCP_SOURCES} - ${KAD_PROTO_SRCS} ${LAYER_CONFIG_PROTO_SRCS}) -target_link_libraries(ipcpd-unicast LINK_PUBLIC ouroboros-dev) - -include(AddCompileFlags) -if (CMAKE_BUILD_TYPE MATCHES "Debug*") - add_compile_flags(ipcpd-unicast -DCONFIG_OUROBOROS_DEBUG) -endif () - -install(TARGETS ipcpd-unicast RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) - -add_subdirectory(pol/tests) - -if (NOT GNU) + addr-auth/flat.c + ca/mb-ecn.c + ca/nop.c + dir/dht.c + pff/simple.c + pff/alternate.c + pff/multipath.c + pff/pft.c + routing/link-state.c + routing/graph.c +) + +add_executable(${IPCP_UNICAST_TARGET} + ${UNICAST_SOURCES} + ${IPCP_SOURCES} + ${COMMON_SOURCES} + ${DHT_PROTO_SRCS} +) + +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${IPCP_INCLUDE_DIRS}) +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}) +target_include_directories(${IPCP_UNICAST_TARGET} PRIVATE ${CMAKE_CURRENT_BINARY_DIR}) +target_link_libraries(${IPCP_UNICAST_TARGET} PRIVATE ouroboros-dev) + +ouroboros_target_debug_definitions(${IPCP_UNICAST_TARGET}) + +install(TARGETS ${IPCP_UNICAST_TARGET} RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) + +if(BUILD_TESTS) + add_subdirectory(ca/tests) + add_subdirectory(dir/tests) + add_subdirectory(pff/tests) + add_subdirectory(routing/tests) add_subdirectory(tests) -endif () +endif() diff --git a/src/ipcpd/unicast/addr_auth.c b/src/ipcpd/unicast/addr-auth.c index e82ea254..27671d09 100644 --- a/src/ipcpd/unicast/addr_auth.c +++ b/src/ipcpd/unicast/addr-auth.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -24,13 +24,12 @@ #include <ouroboros/logs.h> -#include "addr_auth.h" -#include "pol-addr-auth-ops.h" -#include "pol/flat.h" +#include "addr-auth.h" +#include "addr-auth/pol.h" #include <stdlib.h> -struct pol_addr_auth_ops * ops; +struct addr_auth_ops * ops; int addr_auth_init(enum pol_addr_auth type, const void * info) diff --git a/src/ipcpd/unicast/addr_auth.h b/src/ipcpd/unicast/addr-auth.h index 6bedf420..65567dc3 100644 --- a/src/ipcpd/unicast/addr_auth.h +++ b/src/ipcpd/unicast/addr-auth.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -27,6 +27,14 @@ #include <stdint.h> +#define ADDR_FMT32 "%02x.%02x.%02x.%02x" +#define ADDR_VAL32(a) \ + ((uint8_t *) a)[0], ((uint8_t *) a)[1], \ + ((uint8_t *) a)[2], ((uint8_t *) a)[3] + +#define ADDR_FMT64 ADDR_FMT32 "." ADDR_FMT32 +#define ADDR_VAL64(a) ADDR_VAL32(a), ADDR_VAL32(a + 4) + int addr_auth_init(enum pol_addr_auth type, const void * info); diff --git a/src/ipcpd/unicast/pol/flat.c b/src/ipcpd/unicast/addr-auth/flat.c index 6e5c96ab..cf832279 100644 --- a/src/ipcpd/unicast/pol/flat.c +++ b/src/ipcpd/unicast/addr-auth/flat.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for flat addresses in a distributed way * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -29,28 +29,21 @@ #define OUROBOROS_PREFIX "flat-addr-auth" #include <ouroboros/logs.h> -#include <ouroboros/errno.h> -#include <ouroboros/time_utils.h> -#include <ouroboros/utils.h> +#include <ouroboros/random.h> +#include "addr-auth.h" #include "ipcp.h" #include "flat.h" -#include <time.h> -#include <stdlib.h> -#include <math.h> -#include <string.h> -#include <assert.h> - -#define NAME_LEN 8 +#define NAME_LEN 8 +#define INVALID_ADDRESS 0 struct { - uint8_t addr_size; + uint8_t addr_size; + uint32_t addr; } flat; -#define INVALID_ADDRESS 0 - -struct pol_addr_auth_ops flat_ops = { +struct addr_auth_ops flat_ops = { .init = flat_init, .fini = flat_fini, .address = flat_address @@ -65,6 +58,15 @@ int flat_init(const void * info) return -1; } +#if defined (CONFIG_OUROBOROS_DEBUG) && defined (IPCP_DEBUG_LOCAL) + flat.addr = getpid(); +#else + while (flat.addr == INVALID_ADDRESS) + random_buffer(&flat.addr,sizeof(flat.addr)); +#endif + log_dbg("Flat address initialized to " ADDR_FMT32 ".", + ADDR_VAL32((uint8_t *) &flat.addr)); + return 0; } @@ -75,13 +77,5 @@ int flat_fini(void) uint64_t flat_address(void) { - struct timespec t; - uint32_t addr; - - clock_gettime(CLOCK_REALTIME, &t); - srand(t.tv_nsec); - - addr = (rand() % (RAND_MAX - 1) + 1) & 0xFFFFFFFF; - - return addr; + return (uint64_t) flat.addr; } diff --git a/src/ipcpd/unicast/pol/flat.h b/src/ipcpd/unicast/addr-auth/flat.h index a542d7b7..b5c7e525 100644 --- a/src/ipcpd/unicast/pol/flat.h +++ b/src/ipcpd/unicast/addr-auth/flat.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for flat addresses in a distributed way * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,7 +23,7 @@ #ifndef OUROBOROS_IPCPD_UNICAST_FLAT_H #define OUROBOROS_IPCPD_UNICAST_FLAT_H -#include "pol-addr-auth-ops.h" +#include "ops.h" int flat_init(const void * info); @@ -31,6 +31,6 @@ int flat_fini(void); uint64_t flat_address(void); -extern struct pol_addr_auth_ops flat_ops; +extern struct addr_auth_ops flat_ops; #endif /* OUROBOROS_IPCPD_UNICAST_FLAT_H */ diff --git a/src/ipcpd/unicast/pol-addr-auth-ops.h b/src/ipcpd/unicast/addr-auth/ops.h index 1096eecb..f5c4c611 100644 --- a/src/ipcpd/unicast/pol-addr-auth-ops.h +++ b/src/ipcpd/unicast/addr-auth/ops.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Address authority policy ops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,10 +20,10 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#ifndef OUROBOROS_IPCPD_UNICAST_POL_ADDR_AUTH_OPS_H -#define OUROBOROS_IPCPD_UNICAST_POL_ADDR_AUTH_OPS_H +#ifndef OUROBOROS_IPCPD_UNICAST_ADDR_AUTH_OPS_H +#define OUROBOROS_IPCPD_UNICAST_ADDR_AUTH_OPS_H -struct pol_addr_auth_ops { +struct addr_auth_ops { int (* init)(const void * info); int (* fini)(void); @@ -31,4 +31,4 @@ struct pol_addr_auth_ops { uint64_t (* address)(void); }; -#endif /* OUROBOROS_IPCPD_UNICAST_POL_ADDR_AUTH_OPS_H */ +#endif /* OUROBOROS_IPCPD_UNICAST_ADDR_AUTH_OPS_H */ diff --git a/src/ipcpd/unicast/addr-auth/pol.h b/src/ipcpd/unicast/addr-auth/pol.h new file mode 100644 index 00000000..a33f1dd6 --- /dev/null +++ b/src/ipcpd/unicast/addr-auth/pol.h @@ -0,0 +1,23 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Address Authority policies + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "flat.h" diff --git a/src/ipcpd/unicast/ca.c b/src/ipcpd/unicast/ca.c index 3add2a7d..d0ee2f73 100644 --- a/src/ipcpd/unicast/ca.c +++ b/src/ipcpd/unicast/ca.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion Avoidance * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -22,19 +22,47 @@ #define OUROBOROS_PREFIX "ca" +#include "config.h" + +#include <ouroboros/list.h> #include <ouroboros/logs.h> #include "ca.h" -#include "pol-ca-ops.h" -#include "pol/ca-mb-ecn.h" -#include "pol/ca-nop.h" +#include "ca/pol.h" + +#include <pthread.h> +#include <stdlib.h> + +/* + * A ca_ctx holds congestion state for a (peer address, qos cube) PATH, + * not for a flow. In the default build the façade interns one ctx per + * (addr, qc) and shares it across every flow on that path; the policy + * runs on the shared ctx and cannot tell one flow from many. Per-flow + * ctx (IPCP_CA_PER_FLOW) is a testing build only: it skips interning so + * every flow gets its own ctx. + */ + +struct ca_ctx { + uint64_t addr; + qoscube_t qc; + size_t refs; + void * pol; /* policy ctx (ops->ctx_create result) */ + struct list_head next; +}; struct { - struct pol_ca_ops * ops; + struct ca_ops * ops; +#ifndef IPCP_CA_PER_FLOW + struct list_head buckets[CA_BUCKETS]; + pthread_mutex_t mtx; +#endif } ca; int ca_init(enum pol_cong_avoid pol) { +#ifndef IPCP_CA_PER_FLOW + size_t i; +#endif switch(pol) { case CA_NONE: log_dbg("Disabling congestion control."); @@ -48,63 +76,167 @@ int ca_init(enum pol_cong_avoid pol) return -1; } +#ifndef IPCP_CA_PER_FLOW + for (i = 0; i < CA_BUCKETS; i++) + list_head_init(&ca.buckets[i]); + + if (pthread_mutex_init(&ca.mtx, NULL) != 0) + return -1; +#endif return 0; } - void ca_fini(void) { +#ifndef IPCP_CA_PER_FLOW + size_t i; + + /* Data path is stopped; drain any ctx a flow left interned. */ + for (i = 0; i < CA_BUCKETS; i++) { + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, &ca.buckets[i]) { + struct ca_ctx * ctx; + ctx = list_entry(p, struct ca_ctx, next); + list_del(&ctx->next); + ca.ops->ctx_destroy(ctx->pol); + free(ctx); + } + } + + pthread_mutex_destroy(&ca.mtx); +#endif ca.ops = NULL; } -void * ca_ctx_create(void) +#ifndef IPCP_CA_PER_FLOW +static size_t ca_bucket(uint64_t addr, + qoscube_t qc) { - return ca.ops->ctx_create(); + return (addr ^ (addr >> 32) ^ (uint64_t) qc) & (CA_BUCKETS - 1); } +#endif -void ca_ctx_destroy(void * ctx) +void * ca_ctx_get(uint64_t addr, + qoscube_t qc) { - return ca.ops->ctx_destroy(ctx); + struct ca_ctx * ctx; +#ifndef IPCP_CA_PER_FLOW + struct list_head * p; + size_t b = ca_bucket(addr, qc); + + pthread_mutex_lock(&ca.mtx); + + list_for_each(p, &ca.buckets[b]) { + ctx = list_entry(p, struct ca_ctx, next); + if (ctx->addr == addr && ctx->qc == qc) { + ctx->refs++; + pthread_mutex_unlock(&ca.mtx); + return ctx; + } + } +#endif + ctx = malloc(sizeof(*ctx)); + if (ctx == NULL) + goto fail_ctx; + + ctx->pol = ca.ops->ctx_create(); + if (ctx->pol == NULL) + goto fail_pol; + + ctx->addr = addr; + ctx->qc = qc; + ctx->refs = 1; + +#ifndef IPCP_CA_PER_FLOW + list_add(&ctx->next, &ca.buckets[b]); + + pthread_mutex_unlock(&ca.mtx); +#endif + return ctx; + fail_pol: + free(ctx); + fail_ctx: +#ifndef IPCP_CA_PER_FLOW + pthread_mutex_unlock(&ca.mtx); +#endif + return NULL; } -ca_wnd_t ca_ctx_update_snd(void * ctx, - size_t len) +void ca_ctx_put(void * _ctx) { - return ca.ops->ctx_update_snd(ctx, len); + struct ca_ctx * ctx = _ctx; + +#ifndef IPCP_CA_PER_FLOW + pthread_mutex_lock(&ca.mtx); + + if (--ctx->refs > 0) { + pthread_mutex_unlock(&ca.mtx); + return; + } + + list_del(&ctx->next); + + pthread_mutex_unlock(&ca.mtx); +#endif + ca.ops->ctx_destroy(ctx->pol); + + free(ctx); } -bool ca_ctx_update_rcv(void * ctx, - size_t len, - uint8_t ecn, - uint16_t * ece) +time_t ca_ctx_update_snd(void * _ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag) { - return ca.ops->ctx_update_rcv(ctx, len, ecn, ece); + struct ca_ctx * ctx = _ctx; + + return ca.ops->ctx_update_snd(ctx->pol, len, lecn, ftag); } -void ca_ctx_update_ece(void * ctx, - uint16_t ece) +bool ca_ctx_update_rcv(void * _ctx, + size_t len, + uint8_t ecn, + uint8_t cap, + uint16_t * ece, + uint8_t * fcap) { - return ca.ops->ctx_update_ece(ctx, ece); + struct ca_ctx * ctx = _ctx; + + return ca.ops->ctx_update_rcv(ctx->pol, len, ecn, cap, ece, fcap); } -void ca_wnd_wait(ca_wnd_t wnd) +void ca_ctx_update_ece(void * _ctx, + uint16_t ece, + uint8_t cap) { - return ca.ops->wnd_wait(wnd); + struct ca_ctx * ctx = _ctx; + + return ca.ops->ctx_update_ece(ctx->pol, ece, cap); } -int ca_calc_ecn(int fd, +int ca_calc_ecn(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len) { - return ca.ops->calc_ecn(fd, ecn, len); + return ca.ops->calc_ecn(queued, ecn, qc, len); } -ssize_t ca_print_stats(void * ctx, +bool ca_marks_ecn(void) +{ + return ca.ops->marks_ecn; +} + +ssize_t ca_print_stats(void * _ctx, char * buf, size_t len) { + struct ca_ctx * ctx = _ctx; + if (ca.ops->print_stats == NULL) return 0; - return ca.ops->print_stats(ctx, buf, len); + return ca.ops->print_stats(ctx->pol, buf, len); } diff --git a/src/ipcpd/unicast/ca.h b/src/ipcpd/unicast/ca.h index 0604a161..d73d35f5 100644 --- a/src/ipcpd/unicast/ca.h +++ b/src/ipcpd/unicast/ca.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion avoidance * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -24,41 +24,45 @@ #define OUROBOROS_IPCPD_UNICAST_CA_H #include <ouroboros/ipcp.h> +#include <ouroboros/qoscube.h> #include <stdbool.h> #include <sys/types.h> -typedef union { - time_t wait; -} ca_wnd_t; - int ca_init(enum pol_cong_avoid ca); void ca_fini(void); /* OPS */ -void * ca_ctx_create(void); +void * ca_ctx_get(uint64_t addr, + qoscube_t qc); -void ca_ctx_destroy(void * ctx); +void ca_ctx_put(void * ctx); -ca_wnd_t ca_ctx_update_snd(void * ctx, - size_t len); +time_t ca_ctx_update_snd(void * ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag); bool ca_ctx_update_rcv(void * ctx, size_t len, uint8_t ecn, - uint16_t * ece); + uint8_t cap, + uint16_t * ece, + uint8_t * fcap); void ca_ctx_update_ece(void * ctx, - uint16_t ece); + uint16_t ece, + uint8_t cap); -void ca_wnd_wait(ca_wnd_t wnd); - -int ca_calc_ecn(int fd, +int ca_calc_ecn(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len); +bool ca_marks_ecn(void); + ssize_t ca_print_stats(void * ctx, char * buf, size_t len); diff --git a/src/ipcpd/unicast/ca/mb-ecn.c b/src/ipcpd/unicast/ca/mb-ecn.c new file mode 100644 index 00000000..e59aac88 --- /dev/null +++ b/src/ipcpd/unicast/ca/mb-ecn.c @@ -0,0 +1,818 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Multi-bit ECN Congestion Avoidance + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <ouroboros/time.h> +#include <ouroboros/utils.h> + +#include "cap.h" +#include "mb-ecn.h" + +#include <inttypes.h> +#include <stdlib.h> +#include <string.h> +#include <stdio.h> + +/* + * The sender paces with a token bucket at a rate driven by Δt-scaled + * AIMD: each step changes the rate proportional to elapsed wall-clock + * time, so the per-second dynamics do not depend on how often packets + * arrive. The receiver's averaging window and the sender's feedback + * staleness both stretch with the flow's byte rate, so a slow flow + * is measured and controlled like a fast one; CA_RATE_MIN only + * bounds those horizons (window <= CA_TW_ABSMAX, TTL ~8 s). There + * is no per-flow timer; the control runs on packet sends. + * + * The floor and the AI slope scale with the path: forwarders stamp + * their measured link capacity into the PCI (cap.c), the receiver + * feeds the path MIN back with the ece, and the sender derives + * rate_min = ai_rate = C / 32, clamped to [CA_RATE_MIN, CA_RMIN_MAX], + * falling back to those defaults when the signal goes stale. + */ + +#define CA_SHFT 5 /* ece fixed point: 32 * ecn */ +#define CA_TW_MIN (1ULL << 20) /* min mean window ~1.05 ms */ +#define CA_TW_INIT (1ULL << 26) /* initial mean window ~67ms */ +#define CA_TW_ABSMAX (1ULL << 32) /* window ceiling ~4.3 s */ +/* Quiet horizon, in windows (1 << shift): gap restart and the TTLs. */ +#define CA_TW_GAP_SHFT 2 +#define CA_N_TARGET 16 /* target packets per window */ +#define CA_RX_WBYTES (CA_N_TARGET * 1000ULL) /* target bytes/window */ +#define CA_RX_WCLOSE (2 * CA_RX_WBYTES) /* byte-triggered early close */ +#define CA_TW_SM_SHFT 2 /* window EWMA weight 1/4 */ +#define CA_MARK_Q 4 /* mark quantum (packets) */ + +#define CA_RATE_MIN (1ULL << 13) /* 8 KiB/s rate floor */ +#define CA_RATE_INIT (1ULL << 16) /* slow start seed 64 KiB/s */ +/* Rate cap; also keeps rate * dt and rate * rise below 2^64. */ +#define CA_RATE_MAX (1ULL << 37) +#define CA_INV_SHFT 32 /* reciprocal-rate fixp */ +#define CA_AI_RATE (1ULL << 16) /* 64 KiB/s^2 additive inc */ +#define CA_PROBE_TC (8ULL * BILLION) /* proportional probe TC 8s */ +#define CA_ECE_REF (16 << CA_SHFT) /* full congestion: ecn 16 */ +#define CA_MD_KD_DIV 4 /* one-sided lead gain 1/4 */ +/* Must stay >= 1 ms: the MD term scales by dtc / MILLION (truncates). */ +#define CA_DT_CTRL (BILLION / 1000) +#define CA_DT_CAP (BILLION / 20) /* idle-resume Δt clamp 50ms */ +/* Floor of the rate-relative feedback staleness (ctx->ece_ttl). */ +#define CA_ECE_TTL ((1 << CA_TW_GAP_SHFT) * CA_TW_INIT) +#define CA_SS_TC (BILLION / 50) +#define CA_WASH_BKT (BILLION / 32) /* washout bucket ~31 ms */ +#define CA_WASH_SHFT 2 /* damp 1/4 of bucket change */ + +#define CA_CAP_SHFT 5 /* floor = capacity / 32 */ +#define CA_CAP_SM_SHFT 1 /* capacity EWMA weight 1/2 */ +/* Outlives ece_ttl 16x: onset-fresh fcap re-seeds each episode. */ +#define CA_CAP_TTL_SHFT 4 +#define CA_RMIN_MAX (1ULL << 32) /* derived floor ceiling */ + +#define CA_SND_WIN CA_TW_INIT /* sender util window ~67ms */ +#define CA_USE_NUM 3 /* backlogged: offered >= */ +#define CA_USE_DEN 4 /* 3/4 * window-start rate */ +#define CA_HDRM_MARKS 4 /* ceiling ~2x offered load */ +#define CA_SND_DEC_SHFT 4 /* offered max-filter 1/16 */ +#define CA_SND_DEC_CAP 16 /* bound gapped-close decay */ +#define CA_SND_BYT_MAX (1ULL << 33) /* offered-byte saturation */ + +/* + * Retuning invariants (pinned by the unit tests): + * - (1 << CA_TW_GAP_SHFT) * CA_TW_INIT > S * BILLION / CA_RATE_MIN, or + * a floor-rate flow's onset restart-loops (S ~ one MTU; both ns). + * - CA_RX_WBYTES * BILLION / CA_RATE_MIN < CA_TW_ABSMAX: the + * floor-rate window must clear the ceiling. + * - CA_RATE_MAX * CA_DT_CAP, the folded lead * inv_rate at + * CA_RATE_MIN, and owed * BILLION (owed clamped in mb_ecn_snd) all + * keep the pacer arithmetic below 2^64. + * - CA_DT_CTRL < CA_WASH_BKT < CA_DT_CAP: control cadence under the + * washout bucket under the sparse-step cutoff. + * - CA_RATE_MIN <= CA_RATE_INIT and CA_RMIN_MAX < CA_RATE_MAX. + */ + +struct mb_ecn_ctx { + uint16_t rx_ece; /* smoothed congestion echo (32 * ecn) */ + uint64_t rx_acc; /* window integral of ecn * dt */ + uint64_t rx_byt; /* bytes arrived in current window */ + uint64_t rx_ts; /* last packet arrival (ns) */ + uint64_t rx_win; /* window start (ns) */ + uint64_t rx_tw; /* adaptive averaging window (ns) */ + uint8_t rx_cap; /* window bottleneck capacity code */ + + uint16_t tx_ece; /* congestion reported from downstream */ + uint16_t tx_ecp; /* previous tx_ece (rise detection) */ + uint8_t tx_loc; /* local first-hop ecn mark (fallback) */ + uint8_t tx_cap; /* path capacity code fed back to us */ + bool tx_cav; /* past slow start */ + uint64_t rate; /* paced send rate (bytes/s) */ + uint64_t rate_min; /* capacity-derived rate floor (B/s) */ + uint64_t ai_rate; /* additive-increase slope (B/s^2) */ + uint64_t ece_ttl; /* how long feedback stays valid (ns) */ + uint64_t r_bkt; /* rate snapshot at last washout bucket */ + uint64_t wash_acc; /* washout bucket time accumulator (ns) */ + uint64_t inv_rate; /* fixed-point 1/rate for pacing */ + uint64_t vt; /* virtual service clock (bytes) */ + uint64_t lead; /* pacer lead of last send (bytes) */ + uint64_t last_ts; /* last clock advance (ns) */ + uint64_t last_ctrl; /* last rate update (ns) */ + uint64_t last_fb; /* last feedback applied (ns) */ + uint64_t last_loc; /* last local mark seen (ns) */ + uint64_t last_cap; /* last capacity applied (ns) */ + + uint64_t n_ctrl; /* control steps taken */ + uint64_t t_ctrl; /* wall time covered by steps (ns) */ + uint64_t t_bank; /* increase time banked in steps (ns) */ + uint64_t n_fb; /* feedback updates received */ + uint64_t n_ttl; /* feedback aged out (TTL) */ + uint64_t n_cap; /* capacity updates applied */ + uint64_t ss_peak; /* peak rate in slow start (bytes/s) */ + + uint64_t snd_byt; /* bytes offered this window (capped) */ + uint64_t snd_win; /* utilisation window start (ns) */ + uint64_t snd_r0; /* rate at window start */ + uint64_t snd_rate; /* max-filter of offered rate (B/s) */ + bool backlogged; /* offered load keeps the pacer busy */ + bool src_limited; /* rate held at offered-load ceiling */ + bool started; /* a real send has occurred */ +}; + +struct ca_ops mb_ecn_ca_ops = { + .ctx_create = mb_ecn_ctx_create, + .ctx_destroy = mb_ecn_ctx_destroy, + .ctx_update_snd = mb_ecn_ctx_update_snd, + .ctx_update_rcv = mb_ecn_ctx_update_rcv, + .ctx_update_ece = mb_ecn_ctx_update_ece, + .calc_ecn = mb_ecn_calc_ecn, + .marks_ecn = true, + .print_stats = mb_ecn_print_stats +}; + +static uint64_t mb_ecn_rate_inv(uint64_t rate) +{ + return ((uint64_t) BILLION << CA_INV_SHFT) / rate; +} + +/* + * Feedback arrives once per receiver window, and the window tracks + * the flow's byte rate. Mirror it: age the signal out only past the + * quiet horizon at the current rate, floored for fast flows. + */ +static uint64_t mb_ecn_ece_ttl(uint64_t rate) +{ + uint64_t ttl; + + ttl = (1 << CA_TW_GAP_SHFT) * CA_RX_WBYTES * BILLION / rate; + + return ttl > (uint64_t) CA_ECE_TTL ? ttl : (uint64_t) CA_ECE_TTL; +} + +void * mb_ecn_ctx_create(void) +{ + struct timespec now; + uint64_t t; + struct mb_ecn_ctx * ctx; + + ctx = malloc(sizeof(*ctx)); + if (ctx == NULL) + return NULL; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + memset(ctx, 0, sizeof(*ctx)); + + t = TS_TO_UINT64(now); + + ctx->rate = CA_RATE_INIT; + ctx->rate_min = CA_RATE_MIN; + ctx->ai_rate = CA_AI_RATE; + ctx->ece_ttl = mb_ecn_ece_ttl(CA_RATE_INIT); + ctx->r_bkt = CA_RATE_INIT; + ctx->inv_rate = mb_ecn_rate_inv(CA_RATE_INIT); + ctx->rx_ts = t; + ctx->rx_win = t; + ctx->rx_tw = CA_TW_INIT; + ctx->last_ts = t; + ctx->last_ctrl = t; + ctx->last_fb = t; + ctx->last_loc = t; + ctx->last_cap = t; + + /* snd_win/last_ts re-seeded lazily on the first real send. */ + ctx->snd_r0 = CA_RATE_INIT; + ctx->snd_rate = CA_RATE_INIT; + ctx->backlogged = true; + + return (void *) ctx; +} + +void mb_ecn_ctx_destroy(void * ctx) +{ + free(ctx); +} + +/* Local first-hop mark exits slow start and covers dead feedback. */ +static void mb_ecn_loc(struct mb_ecn_ctx * ctx, + uint8_t lecn, + uint64_t t) +{ + if (lecn == 0) + return; + + ctx->tx_loc = lecn; + ctx->tx_cav = true; + ctx->last_loc = t; +} + +/* Slow start: ramp only while backlogged. */ +static void mb_ecn_slow_start(struct mb_ecn_ctx * ctx, + uint64_t dta) +{ + if (ctx->backlogged) + ctx->rate += ctx->rate * dta / CA_SS_TC; + + ctx->r_bkt = ctx->rate; +} + +/* Additive increase plus a rate-independent proportional probe. */ +static void mb_ecn_increase(struct mb_ecn_ctx * ctx, + uint64_t dta) +{ + if (!ctx->backlogged) + return; + + ctx->rate += ctx->ai_rate * dta / BILLION; + ctx->rate += ctx->rate * dta / CA_PROBE_TC; +} + +/* Multiplicative decrease: proportional cut plus a one-sided lead. */ +static void mb_ecn_decrease(struct mb_ecn_ctx * ctx, + uint64_t dtc) +{ + uint64_t dtm; + uint64_t mark; + uint64_t rise; + uint64_t cut; + uint16_t m; + + m = ctx->tx_ece > 0 ? ctx->tx_ece + : (uint16_t) (ctx->tx_loc << CA_SHFT); + if (m == 0) { + ctx->tx_ecp = 0; + return; + } + + mark = MIN(m, CA_ECE_REF); + rise = 0; + if (m > ctx->tx_ecp) + rise = MIN(m - ctx->tx_ecp, CA_ECE_REF); + + cut = ctx->rate * rise / (CA_ECE_REF * CA_MD_KD_DIV); + + /* Honest elapsed ms, so a starved sender still cuts. */ + dtm = dtc / MILLION; + if (mark * dtm >= CA_ECE_REF * 500) + cut += ctx->rate / 2; + else + cut += ctx->rate * mark * dtm / (CA_ECE_REF * 1000); + + if (cut > ctx->rate / 2) + cut = ctx->rate / 2; + + ctx->rate -= cut; + ctx->tx_ecp = m; +} + +/* + * Washout: once per wall-clock bucket, damp a fixed fraction of the + * rate change over that bucket. Bucketed (not per-step) so it stays + * cadence-independent; bounded so it cannot reverse a ramp. A sparse + * step resets it, so a starved sender keeps its cut. + */ +static void mb_ecn_washout(struct mb_ecn_ctx * ctx, + uint64_t dtc, + uint64_t dta) +{ + if (dtc > (uint64_t) CA_DT_CAP) { + ctx->r_bkt = ctx->rate; + ctx->wash_acc = 0; + return; + } + + ctx->wash_acc += dta; + if (ctx->wash_acc < (uint64_t) CA_WASH_BKT) + return; + + if (ctx->rate > ctx->r_bkt) + ctx->rate -= (ctx->rate - ctx->r_bkt) >> CA_WASH_SHFT; + else + ctx->rate += (ctx->r_bkt - ctx->rate) >> CA_WASH_SHFT; + + ctx->r_bkt = ctx->rate; + ctx->wash_acc = 0; +} + +/* Offered-load ceiling backstop while source-limited. */ +static void mb_ecn_ceiling(struct mb_ecn_ctx * ctx) +{ + unsigned code; + uint64_t hi; + + if (ctx->backlogged) { + ctx->src_limited = false; + return; + } + + code = (unsigned) cap_enc(ctx->snd_rate) + CA_HDRM_MARKS; + if (code > UINT8_MAX) /* keep the cast lossless */ + code = UINT8_MAX; + + hi = cap_dec((uint8_t) code); + if (hi > CA_RATE_MAX) + hi = CA_RATE_MAX; + + if (hi < CA_RATE_MIN) + hi = CA_RATE_MIN; + + ctx->src_limited = ctx->rate > hi; + if (ctx->src_limited) { + ctx->rate = hi; + ctx->r_bkt = ctx->rate; + } +} + +static void mb_ecn_ctrl(struct mb_ecn_ctx * ctx, + uint64_t dtc) +{ + uint64_t dta; + uint64_t lo; + + /* AI and slow start bank at most CA_DT_CAP of idle time. */ + dta = MIN(dtc, (uint64_t) CA_DT_CAP); + + ctx->n_ctrl++; + ctx->t_ctrl += dtc; + ctx->t_bank += dta; + + if (ctx->tx_cav) { + mb_ecn_increase(ctx, dta); + mb_ecn_decrease(ctx, dtc); + mb_ecn_washout(ctx, dtc, dta); + } else { + mb_ecn_slow_start(ctx, dta); + } + + mb_ecn_ceiling(ctx); + + /* Capacity floor only while backlogged; else the absolute floor. */ + lo = ctx->backlogged ? ctx->rate_min : (uint64_t) CA_RATE_MIN; + if (ctx->rate < lo) + ctx->rate = lo; + + if (ctx->rate > CA_RATE_MAX) + ctx->rate = CA_RATE_MAX; + + ctx->inv_rate = mb_ecn_rate_inv(ctx->rate); + ctx->ece_ttl = mb_ecn_ece_ttl(ctx->rate); + + if (!ctx->tx_cav && ctx->rate > ctx->ss_peak) + ctx->ss_peak = ctx->rate; +} + +/* Fold offered into the max filter: rise at once, decay 1/16 per window. */ +static void mb_ecn_offered(struct mb_ecn_ctx * ctx, + uint64_t offered, + uint64_t elapsed) +{ + uint64_t n; + + if (offered >= ctx->snd_rate) { + ctx->snd_rate = offered; + return; + } + + n = MIN(elapsed / CA_SND_WIN, CA_SND_DEC_CAP); + while (n-- > 0 && ctx->snd_rate > offered) + ctx->snd_rate -= (ctx->snd_rate - offered) >> CA_SND_DEC_SHFT; +} + +/* + * Close the utilisation window: set backlogged from the level test, + * fold offered into the max filter, then reset the window. + */ +static void mb_ecn_win(struct mb_ecn_ctx * ctx, + uint64_t t) +{ + uint64_t elapsed = t - ctx->snd_win; + uint64_t offered; + + offered = ctx->snd_byt * BILLION / elapsed; + + ctx->backlogged = offered * CA_USE_DEN >= ctx->snd_r0 * CA_USE_NUM; + + mb_ecn_offered(ctx, offered, elapsed); + + if (ctx->backlogged) + ctx->src_limited = false; + + ctx->snd_win = t; + ctx->snd_byt = 0; + ctx->snd_r0 = ctx->rate; +} + +/* Age out congestion, local-mark and capacity signals once stale. */ +static void mb_ecn_age(struct mb_ecn_ctx * ctx, + uint64_t t) +{ + if (t - ctx->last_fb > ctx->ece_ttl) { + if (ctx->tx_ece > 0) + ctx->n_ttl++; + ctx->tx_ece = 0; + } + + if (t - ctx->last_loc > ctx->ece_ttl) + ctx->tx_loc = 0; + + /* Stale capacity: fall back to the compile-time defaults. */ + if (t - ctx->last_cap > ctx->ece_ttl << CA_CAP_TTL_SHFT) { + ctx->rate_min = CA_RATE_MIN; + ctx->ai_rate = CA_AI_RATE; + ctx->tx_cap = 0; + } +} + +/* Advance the virtual clock; a gap past CA_DT_CAP credits a burst. */ +static void mb_ecn_advance(struct mb_ecn_ctx * ctx, + uint64_t dt, + size_t len, + uint64_t ftag) +{ + uint64_t burst; + uint64_t owed; + + if (dt <= (uint64_t) CA_DT_CAP) { + ctx->vt += ctx->rate * dt / BILLION; + return; + } + + burst = ctx->rate * CA_DT_CAP / BILLION; + if (burst < (uint64_t) len) + burst = len; + + owed = ftag > ctx->vt ? ftag - ctx->vt + burst : burst; + + /* Clamp so owed * BILLION cannot wrap (2^33 B backlog). */ + if (owed > (1ULL << 33)) + owed = 1ULL << 33; + + if (dt >= owed * BILLION / ctx->rate) + ctx->vt += owed; + else + ctx->vt += ctx->rate * dt / BILLION; +} + +static time_t mb_ecn_snd(struct mb_ecn_ctx * ctx, + size_t len, + uint64_t t, + uint64_t * ftag) +{ + uint64_t dt; + uint64_t dtc; + uint64_t s; + + /* Lazy warm-up seed: packet #1 is never an idle resume. */ + if (!ctx->started) { + ctx->started = true; + ctx->last_ts = t; + ctx->snd_win = t; + ctx->snd_r0 = ctx->rate; + } + + dt = t - ctx->last_ts; + ctx->last_ts = t; + + mb_ecn_age(ctx, t); + + /* Offered-load estimator: accumulate, gate growth, size ceiling. */ + ctx->snd_byt += len; + if (ctx->snd_byt > (uint64_t) CA_SND_BYT_MAX) + ctx->snd_byt = CA_SND_BYT_MAX; + + if (dt > (uint64_t) CA_DT_CAP) + ctx->backlogged = false; + + if (t - ctx->snd_win >= (uint64_t) CA_SND_WIN) + mb_ecn_win(ctx, t); + + /* Rate update before the vt advance: burst uses the clamped rate. */ + dtc = t - ctx->last_ctrl; + if (dtc >= (uint64_t) CA_DT_CTRL) { + ctx->last_ctrl = t; + mb_ecn_ctrl(ctx, dtc); + } + + mb_ecn_advance(ctx, dt, len, *ftag); + + /* SFQ start tag: behind the clock starts now, ahead waits. */ + s = *ftag > ctx->vt ? *ftag : ctx->vt; + *ftag = s + len; + + ctx->lead = s - ctx->vt; + + /* Reciprocal pacing; folded so any lead * rate stays in range. */ + if (s > ctx->vt) + return (time_t) ((ctx->lead * (ctx->inv_rate >> 16)) + >> (CA_INV_SHFT - 16)); + + return 0; +} + +time_t mb_ecn_ctx_update_snd(void * _ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag) +{ + struct timespec now; + uint64_t t; + struct mb_ecn_ctx * ctx = _ctx; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + t = TS_TO_UINT64(now); + + mb_ecn_loc(ctx, lecn, t); + + return mb_ecn_snd(ctx, len, t, ftag); +} + +/* Estimator idle, or a gap past ~4 current windows: restart fresh. */ +static bool mb_ecn_rcv_fresh(const struct mb_ecn_ctx * ctx, + uint64_t dt) +{ + if (ctx->rx_ece == 0 && ctx->rx_acc == 0) + return true; + + return dt > ctx->rx_tw << CA_TW_GAP_SHFT; +} + +/* Size the next averaging window to ~CA_N_TARGET packets at this rate. */ +static void mb_ecn_resize(struct mb_ecn_ctx * ctx, + uint64_t win) +{ + uint64_t tw = CA_RX_WBYTES * win / ctx->rx_byt; + + if (tw > ctx->rx_tw) + ctx->rx_tw += (tw - ctx->rx_tw) >> CA_TW_SM_SHFT; + else + ctx->rx_tw -= (ctx->rx_tw - tw) >> CA_TW_SM_SHFT; + + if (ctx->rx_tw < CA_TW_MIN) + ctx->rx_tw = CA_TW_MIN; + + if (ctx->rx_tw > CA_TW_ABSMAX) + ctx->rx_tw = CA_TW_ABSMAX; +} + +static bool mb_ecn_rcv(struct mb_ecn_ctx * ctx, + size_t len, + uint8_t ecn, + uint8_t cap, + uint16_t * ece, + uint8_t * fcap, + uint64_t t) +{ + uint64_t dt; + uint64_t win; + + dt = t - ctx->rx_ts; + ctx->rx_ts = t; + + if (ctx->rx_ece == 0 && ctx->rx_acc == 0 && ecn == 0) + return false; + + /* Onset, or ~4 windows of silence: emit fresh, undiluted. */ + if (mb_ecn_rcv_fresh(ctx, dt)) { + ctx->rx_win = t; + ctx->rx_acc = 0; + ctx->rx_byt = len; + ctx->rx_cap = cap; /* fresh, seeds the new window */ + ctx->rx_ece = (uint16_t) (ecn << CA_SHFT); + *ece = ctx->rx_ece; + *fcap = ctx->rx_cap; + return true; + } + + /* Dwell clamp: one packet weighs at most one window of mark. */ + ctx->rx_acc += ecn * MIN(dt, ctx->rx_tw); + ctx->rx_byt += len; + ctx->rx_cap = cap_min(ctx->rx_cap, cap); + + *ece = ctx->rx_ece; + + win = t - ctx->rx_win; + if (win < ctx->rx_tw) { + /* Early close once 2x target bytes arrive (speed-up). */ + if (ctx->rx_byt < CA_RX_WCLOSE) + return false; + + if (win < CA_TW_MIN) + return false; + } + + /* Time-integral mean over the actual window elapsed (never rx_tw). */ + ctx->rx_ece = (uint16_t) ((ctx->rx_acc << CA_SHFT) / win); + + if (ctx->rx_byt > 0) + mb_ecn_resize(ctx, win); + + *fcap = ctx->rx_cap; + + ctx->rx_win = t; + ctx->rx_acc = 0; + ctx->rx_byt = 0; + ctx->rx_cap = 0; /* the next window starts unknown */ + + *ece = ctx->rx_ece; + + return true; +} + +bool mb_ecn_ctx_update_rcv(void * _ctx, + size_t len, + uint8_t ecn, + uint8_t cap, + uint16_t * ece, + uint8_t * fcap) +{ + struct timespec now; + struct mb_ecn_ctx * ctx = _ctx; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + return mb_ecn_rcv(ctx, len, ecn, cap, ece, fcap, TS_TO_UINT64(now)); +} + +static void mb_ecn_ece(struct mb_ecn_ctx * ctx, + uint16_t ece, + uint8_t cap, + uint64_t t) +{ + uint64_t tgt; + + ctx->tx_ece = ece; + ctx->tx_cav = true; + ctx->last_fb = t; + ctx->n_fb++; + + /* Scale the floor and AI slope to the path bottleneck. */ + if (cap != 0) { + tgt = cap_dec(cap) >> CA_CAP_SHFT; + if (tgt < CA_RATE_MIN) + tgt = CA_RATE_MIN; + + if (tgt > CA_RMIN_MAX) + tgt = CA_RMIN_MAX; + + if (tgt > ctx->rate_min) + ctx->rate_min += (tgt - ctx->rate_min) + >> CA_CAP_SM_SHFT; + else + ctx->rate_min -= (ctx->rate_min - tgt) + >> CA_CAP_SM_SHFT; + + ctx->ai_rate = ctx->rate_min; + ctx->tx_cap = cap; + ctx->last_cap = t; + ctx->n_cap++; + } + + /* Control from the feedback path: a starved sender recovers. */ + if (t - ctx->last_ctrl < (uint64_t) CA_DT_CTRL) + return; + + mb_ecn_ctrl(ctx, t - ctx->last_ctrl); + + ctx->last_ctrl = t; +} + +void mb_ecn_ctx_update_ece(void * _ctx, + uint16_t ece, + uint8_t cap) +{ + struct timespec now; + struct mb_ecn_ctx * ctx = _ctx; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + mb_ecn_ece(ctx, ece, cap, TS_TO_UINT64(now)); +} + +int mb_ecn_calc_ecn(size_t queued, + uint8_t * ecn, + qoscube_t qc, + size_t len) +{ + size_t q; + uint8_t mark; + + (void) len; + (void) qc; + + /* Saturate: a queue past 255 quanta must not wrap to a low mark. */ + q = queued / CA_MARK_Q; + mark = q > 255 ? (uint8_t) 255 : (uint8_t) q; + + if (mark > *ecn) + *ecn = mark; + + return 0; +} + +ssize_t mb_ecn_print_stats(void * _ctx, + char * buf, + size_t len) +{ + struct mb_ecn_ctx * ctx = _ctx; + char * regime; + uint64_t rate; + uint64_t peak; + int code; + uint16_t m; + + if (len < 1024) + return 0; + + /* No signal seen: the rate is unconstrained drift, not a target. */ + rate = ctx->tx_cav ? ctx->rate : 0; + peak = ctx->tx_cav ? ctx->ss_peak : 0; + + /* Match the controller: MD fires on m, incl. the local fallback. */ + m = ctx->tx_ece > 0 ? ctx->tx_ece + : (uint16_t) (ctx->tx_loc << CA_SHFT); + + if (!ctx->tx_cav) { + regime = "Slow start"; + code = 0; + } else if (ctx->src_limited) { + regime = "Source limited"; + code = 3; + } else if (m > 0) { + regime = "Proportional dec"; + code = 2; + } else { + regime = "Additive inc"; + code = 1; + } + + sprintf(buf, + "Congestion avoidance algorithm: %20s\n" + "Upstream congestion level: %20u\n" + "Downstream congestion level: %20u\n" + "Paced rate (bytes/s): %20" PRIu64 "\n" + "Pacer lead (bytes): %20" PRIu64 "\n" + "Congestion regime (code): %20d\n" + "Current congestion regime: %20s\n" + "Control steps (count): %20" PRIu64 "\n" + "Control time elapsed (ns): %20" PRIu64 "\n" + "Control time banked (ns): %20" PRIu64 "\n" + "Feedback updates (count): %20" PRIu64 "\n" + "Feedback timeouts (count): %20" PRIu64 "\n" + "Path capacity (bytes/s): %20" PRIu64 "\n" + "Capacity rate floor (bytes/s): %20" PRIu64 "\n" + "Capacity updates (count): %20" PRIu64 "\n" + "Slow start peak rate (bytes/s): %20" PRIu64 "\n", + "Multi-bit ECN", + ctx->tx_ece, + ctx->rx_ece, + rate, ctx->lead, code, + regime, + ctx->n_ctrl, ctx->t_ctrl, ctx->t_bank, + ctx->n_fb, ctx->n_ttl, + cap_dec(ctx->tx_cap), ctx->rate_min, ctx->n_cap, + peak); + + return strlen(buf); +} diff --git a/src/ipcpd/unicast/pol/ca-mb-ecn.h b/src/ipcpd/unicast/ca/mb-ecn.h index 30dba4a2..7bf0b29f 100644 --- a/src/ipcpd/unicast/pol/ca-mb-ecn.h +++ b/src/ipcpd/unicast/ca/mb-ecn.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Multi-bit ECN Congestion Avoidance * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,33 +23,37 @@ #ifndef OUROBOROS_IPCPD_UNICAST_CA_MB_ECN_H #define OUROBOROS_IPCPD_UNICAST_CA_MB_ECN_H -#include "pol-ca-ops.h" +#include "ops.h" void * mb_ecn_ctx_create(void); void mb_ecn_ctx_destroy(void * ctx); -ca_wnd_t mb_ecn_ctx_update_snd(void * ctx, - size_t len); +time_t mb_ecn_ctx_update_snd(void * ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag); bool mb_ecn_ctx_update_rcv(void * ctx, size_t len, uint8_t ecn, - uint16_t * ece); + uint8_t cap, + uint16_t * ece, + uint8_t * fcap); void mb_ecn_ctx_update_ece(void * ctx, - uint16_t ece); - -void mb_ecn_wnd_wait(ca_wnd_t wnd); + uint16_t ece, + uint8_t cap); -int mb_ecn_calc_ecn(int fd, +int mb_ecn_calc_ecn(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len); ssize_t mb_ecn_print_stats(void * ctx, char * buf, size_t len); -extern struct pol_ca_ops mb_ecn_ca_ops; +extern struct ca_ops mb_ecn_ca_ops; #endif /* OUROBOROS_IPCPD_UNICAST_CA_MB_ECN_H */ diff --git a/src/ipcpd/unicast/pol/ca-nop.c b/src/ipcpd/unicast/ca/nop.c index ce210cfb..e6965297 100644 --- a/src/ipcpd/unicast/pol/ca-nop.c +++ b/src/ipcpd/unicast/ca/nop.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Dummy Congestion Avoidance * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,18 +20,18 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#include "ca-nop.h" +#include "nop.h" #include <string.h> -struct pol_ca_ops nop_ca_ops = { +struct ca_ops nop_ca_ops = { .ctx_create = nop_ctx_create, .ctx_destroy = nop_ctx_destroy, .ctx_update_snd = nop_ctx_update_snd, .ctx_update_rcv = nop_ctx_update_rcv, .ctx_update_ece = nop_ctx_update_ece, - .wnd_wait = nop_wnd_wait, .calc_ecn = nop_calc_ecn, + .marks_ecn = false, .print_stats = NULL }; @@ -45,52 +45,55 @@ void nop_ctx_destroy(void * ctx) (void) ctx; } -ca_wnd_t nop_ctx_update_snd(void * ctx, - size_t len) +time_t nop_ctx_update_snd(void * ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag) { - ca_wnd_t wnd; - (void) ctx; (void) len; + (void) lecn; + (void) ftag; - memset(&wnd, 0, sizeof(wnd)); - - return wnd; -} - -void nop_wnd_wait(ca_wnd_t wnd) -{ - (void) wnd; + return 0; } bool nop_ctx_update_rcv(void * ctx, size_t len, uint8_t ecn, - uint16_t * ece) + uint8_t cap, + uint16_t * ece, + uint8_t * fcap) { (void) ctx; (void) len; (void) ecn; + (void) cap; (void) ece; + (void) fcap; return false; } void nop_ctx_update_ece(void * ctx, - uint16_t ece) + uint16_t ece, + uint8_t cap) { (void) ctx; (void) ece; + (void) cap; } -int nop_calc_ecn(int fd, +int nop_calc_ecn(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len) { - (void) fd; + (void) queued; (void) len; (void) ecn; + (void) qc; return 0; } diff --git a/src/ipcpd/unicast/pol/ca-nop.h b/src/ipcpd/unicast/ca/nop.h index 859b52dc..6ca206df 100644 --- a/src/ipcpd/unicast/pol/ca-nop.h +++ b/src/ipcpd/unicast/ca/nop.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Dummy Congestion Avoidance * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,29 +23,33 @@ #ifndef OUROBOROS_IPCPD_UNICAST_CA_NOP_H #define OUROBOROS_IPCPD_UNICAST_CA_NOP_H -#include "pol-ca-ops.h" +#include "ops.h" void * nop_ctx_create(void); void nop_ctx_destroy(void * ctx); -ca_wnd_t nop_ctx_update_snd(void * ctx, - size_t len); +time_t nop_ctx_update_snd(void * ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag); bool nop_ctx_update_rcv(void * ctx, size_t len, uint8_t ecn, - uint16_t * ece); + uint8_t cap, + uint16_t * ece, + uint8_t * fcap); void nop_ctx_update_ece(void * ctx, - uint16_t ece); - -void nop_wnd_wait(ca_wnd_t wnd); + uint16_t ece, + uint8_t cap); -int nop_calc_ecn(int fd, +int nop_calc_ecn(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len); -extern struct pol_ca_ops nop_ca_ops; +extern struct ca_ops nop_ca_ops; #endif /* OUROBOROS_IPCPD_UNICAST_CA_NOP_H */ diff --git a/src/ipcpd/unicast/pol-ca-ops.h b/src/ipcpd/unicast/ca/ops.h index 1e7c48b9..a86c1b3b 100644 --- a/src/ipcpd/unicast/pol-ca-ops.h +++ b/src/ipcpd/unicast/ca/ops.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Congestion avoidance policy ops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,33 +20,40 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#ifndef OUROBOROS_IPCPD_UNICAST_POL_CA_OPS_H -#define OUROBOROS_IPCPD_UNICAST_POL_CA_OPS_H +#ifndef OUROBOROS_IPCPD_UNICAST_CA_OPS_H +#define OUROBOROS_IPCPD_UNICAST_CA_OPS_H #include "ca.h" -struct pol_ca_ops { +struct ca_ops { void * (* ctx_create)(void); void (* ctx_destroy)(void * ctx); - ca_wnd_t (* ctx_update_snd)(void * ctx, - size_t len); + time_t (* ctx_update_snd)(void * ctx, + size_t len, + uint8_t lecn, + uint64_t * ftag); bool (* ctx_update_rcv)(void * ctx, size_t len, uint8_t ecn, - uint16_t * ece); + uint8_t cap, + uint16_t * ece, + uint8_t * fcap); void (* ctx_update_ece)(void * ctx, - uint16_t ece); - - void (* wnd_wait)(ca_wnd_t wnd); + uint16_t ece, + uint8_t cap); - int (* calc_ecn)(int fd, + int (* calc_ecn)(size_t queued, uint8_t * ecn, + qoscube_t qc, size_t len); + /* True if calc_ecn inspects the queue; gates the lookup. */ + bool marks_ecn; + /* Optional, can be NULL */ ssize_t (* print_stats)(void * ctx, char * buf, @@ -54,4 +61,4 @@ struct pol_ca_ops { }; -#endif /* OUROBOROS_IPCPD_UNICAST_POL_CA_OPS_H */ +#endif /* OUROBOROS_IPCPD_UNICAST_CA_OPS_H */ diff --git a/src/ipcpd/unicast/ca/pol.h b/src/ipcpd/unicast/ca/pol.h new file mode 100644 index 00000000..bfb9cc2d --- /dev/null +++ b/src/ipcpd/unicast/ca/pol.h @@ -0,0 +1,24 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Congestion avoidance policies + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "mb-ecn.h" +#include "nop.h" diff --git a/src/ipcpd/unicast/ca/tests/CMakeLists.txt b/src/ipcpd/unicast/ca/tests/CMakeLists.txt new file mode 100644 index 00000000..6e42163d --- /dev/null +++ b/src/ipcpd/unicast/ca/tests/CMakeLists.txt @@ -0,0 +1,44 @@ +get_filename_component(CURRENT_SOURCE_PARENT_DIR + ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(CURRENT_BINARY_PARENT_DIR + ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) + +get_filename_component(UNICAST_SOURCE_DIR ${CURRENT_SOURCE_PARENT_DIR} DIRECTORY) +get_filename_component(UNICAST_BINARY_DIR ${CURRENT_BINARY_PARENT_DIR} DIRECTORY) + +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + mb_ecn_test.c + ca_test.c + ) + +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests} + ${UNICAST_SOURCE_DIR}/ca.c + ${UNICAST_SOURCE_DIR}/cap.c + ${CURRENT_SOURCE_PARENT_DIR}/nop.c + ) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${UNICAST_SOURCE_DIR} + ${UNICAST_BINARY_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) + +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/ca/tests/ca_test.c b/src/ipcpd/unicast/ca/tests/ca_test.c new file mode 100644 index 00000000..4c44d29e --- /dev/null +++ b/src/ipcpd/unicast/ca/tests/ca_test.c @@ -0,0 +1,392 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests for the congestion-avoidance interface + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "config.h" + +#include "ca.h" + +#include <test/test.h> + +#define ADDR_A 0x1111ULL +#define ADDR_B 0x2222ULL + +static const struct { + enum pol_cong_avoid pol; + const char * name; +} ca_pols[] = { + { CA_NONE, "none" }, + { CA_MB_ECN, "mb-ecn" } +}; + +#define CA_POLS (sizeof(ca_pols) / sizeof(ca_pols[0])) + +static int test_ca_init_fini(enum pol_cong_avoid pol, + const char * name) +{ + TEST_START("(%s)", name); + + if (ca_init(pol) < 0) { + printf("Failed to init ca for %s.\n", name); + goto fail; + } + + ca_fini(); + + TEST_SUCCESS("(%s)", name); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL("(%s)", name); + return TEST_RC_FAIL; +} + +static int test_ca_init_fini_all(void) +{ + int ret = 0; + size_t i; + + for (i = 0; i < CA_POLS; i++) + ret |= test_ca_init_fini(ca_pols[i].pol, ca_pols[i].name); + + return ret; +} + +static int test_ca_init_invalid(void) +{ + TEST_START(); + + if (ca_init(CA_INVALID) == 0) { + printf("Init accepted an invalid policy.\n"); + ca_fini(); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_ca_ctx_share(enum pol_cong_avoid pol, + const char * name) +{ + void * c1; + void * c2; + + TEST_START("(%s)", name); + + if (ca_init(pol) < 0) { + printf("Failed to init ca for %s.\n", name); + goto fail; + } + + c1 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); + if (c1 == NULL) { + printf("Failed to get ctx.\n"); + goto fail_init; + } + + c2 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); + if (c2 == NULL) { + printf("Failed to get second ctx.\n"); + goto fail_c1; + } + +#ifdef IPCP_CA_PER_FLOW + if (c1 == c2) { + printf("Per-flow build shared a ctx across flows.\n"); + goto fail_c2; + } +#else + if (c1 != c2) { + printf("Aggregate build did not share ctx per (addr, qc).\n"); + goto fail_c2; + } +#endif + ca_ctx_put(c2); + ca_ctx_put(c1); + + ca_fini(); + + TEST_SUCCESS("(%s)", name); + + return TEST_RC_SUCCESS; + fail_c2: + ca_ctx_put(c2); + fail_c1: + ca_ctx_put(c1); + fail_init: + ca_fini(); + fail: + TEST_FAIL("(%s)", name); + return TEST_RC_FAIL; +} + +static int test_ca_ctx_share_all(void) +{ + int ret = 0; + size_t i; + + for (i = 0; i < CA_POLS; i++) + ret |= test_ca_ctx_share(ca_pols[i].pol, ca_pols[i].name); + + return ret; +} + +static int test_ca_ctx_distinct(void) +{ + void * a_be; + void * b_be; + void * a_video; + + TEST_START(); + + if (ca_init(CA_NONE) < 0) { + printf("Failed to init ca.\n"); + goto fail; + } + + a_be = ca_ctx_get(ADDR_A, QOS_CUBE_BE); + if (a_be == NULL) { + printf("Failed to get ctx.\n"); + goto fail_init; + } + + b_be = ca_ctx_get(ADDR_B, QOS_CUBE_BE); + if (b_be == NULL) { + printf("Failed to get ctx.\n"); + goto fail_a_be; + } + + a_video = ca_ctx_get(ADDR_A, QOS_CUBE_VIDEO); + if (a_video == NULL) { + printf("Failed to get ctx.\n"); + goto fail_b_be; + } + + if (a_be == b_be) { + printf("Distinct addresses shared a ctx.\n"); + goto fail_a_video; + } + + if (a_be == a_video) { + printf("Distinct qos cubes shared a ctx.\n"); + goto fail_a_video; + } + + ca_ctx_put(a_video); + ca_ctx_put(b_be); + ca_ctx_put(a_be); + + ca_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_a_video: + ca_ctx_put(a_video); + fail_b_be: + ca_ctx_put(b_be); + fail_a_be: + ca_ctx_put(a_be); + fail_init: + ca_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Refcount survival is an aggregate-only property. */ +#ifndef IPCP_CA_PER_FLOW +static int test_ca_ctx_refcount(void) +{ + void * c1; + void * c3; + + TEST_START(); + + if (ca_init(CA_NONE) < 0) { + printf("Failed to init ca.\n"); + goto fail; + } + + c1 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 1 */ + if (c1 == NULL) { + printf("Failed to get ctx.\n"); + goto fail_init; + } + + if (ca_ctx_get(ADDR_A, QOS_CUBE_BE) == NULL) { /* refs = 2 */ + printf("Failed to get second ref.\n"); + goto fail_c1; + } + + ca_ctx_put(c1); /* refs = 1 */ + + c3 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 2 */ + if (c3 == NULL) { + printf("Failed to get third ref.\n"); + goto fail_c1; + } + + if (c3 != c1) { + printf("Refcounted ctx freed while still referenced.\n"); + goto fail_c3; + } + + ca_ctx_put(c3); + ca_ctx_put(c1); + + ca_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_c3: + ca_ctx_put(c3); + fail_c1: + ca_ctx_put(c1); + fail_init: + ca_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The last put frees the interned ctx; a fresh get recreates it. */ +static int test_ca_ctx_recreate(void) +{ + void * c1; + void * c2; + void * c3; + + TEST_START(); + + if (ca_init(CA_NONE) < 0) { + printf("Failed to init ca.\n"); + goto fail; + } + + c1 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 1 */ + if (c1 == NULL) { + printf("Failed to get ctx.\n"); + goto fail_init; + } + + ca_ctx_put(c1); /* refs = 0: freed and de-interned */ + + c2 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* fresh entry */ + if (c2 == NULL) { + printf("Get after release did not recreate.\n"); + goto fail_init; + } + + c3 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 2: shares */ + if (c3 == NULL) { + printf("Failed to share recreated ctx.\n"); + goto fail_c2; + } + + if (c3 != c2) { + printf("Recreated ctx did not intern.\n"); + goto fail_c3; + } + + ca_ctx_put(c3); + ca_ctx_put(c2); + + ca_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_c3: + ca_ctx_put(c3); + fail_c2: + ca_ctx_put(c2); + fail_init: + ca_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* ca_fini drains a ctx a flow left interned, with no leak. */ +static int test_ca_fini_drains(void) +{ + void * c1; + void * c2; + + TEST_START(); + + if (ca_init(CA_NONE) < 0) { + printf("Failed to init ca.\n"); + goto fail; + } + + c1 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 1 */ + if (c1 == NULL) { + printf("Failed to get ctx.\n"); + goto fail_init; + } + + c2 = ca_ctx_get(ADDR_A, QOS_CUBE_BE); /* refs = 2 */ + if (c2 == NULL) { + printf("Failed to get second ref.\n"); + goto fail_init; + } + + /* Leave both refs live: ca_fini must drain and free the ctx. */ + ca_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + ca_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} +#endif /* !IPCP_CA_PER_FLOW */ + +int ca_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_ca_init_fini_all(); + ret |= test_ca_init_invalid(); + ret |= test_ca_ctx_share_all(); + ret |= test_ca_ctx_distinct(); +#ifndef IPCP_CA_PER_FLOW + ret |= test_ca_ctx_refcount(); + ret |= test_ca_ctx_recreate(); + ret |= test_ca_fini_drains(); +#endif + return ret; +} diff --git a/src/ipcpd/unicast/ca/tests/mb_ecn_test.c b/src/ipcpd/unicast/ca/tests/mb_ecn_test.c new file mode 100644 index 00000000..8e3a73da --- /dev/null +++ b/src/ipcpd/unicast/ca/tests/mb_ecn_test.c @@ -0,0 +1,2588 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests for multi-bit ECN congestion avoidance + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "mb-ecn.c" + +#include <test/test.h> + +#define MS (MILLION) /* one millisecond in ns */ +#define LEN 1000 /* default packet size (bytes) */ + +/* Create a context with the clock zeroed for deterministic time steps. */ +static struct mb_ecn_ctx * mk_ctx(void) +{ + struct mb_ecn_ctx * ctx; + + ctx = mb_ecn_ctx_create(); + if (ctx == NULL) + return NULL; + + ctx->rx_ts = 0; + ctx->rx_win = 0; + ctx->last_ts = 0; + ctx->last_ctrl = 0; + ctx->last_fb = 0; + ctx->last_loc = 0; + ctx->last_cap = 0; + + ctx->snd_byt = 0; + ctx->snd_win = 0; + ctx->snd_r0 = CA_RATE_INIT; + ctx->snd_rate = CA_RATE_INIT; + ctx->backlogged = true; + ctx->src_limited = false; + ctx->started = false; + + return ctx; +} + +/* + * Drive ctx as a fully backlogged flow: offer a packet every paced + * wait, so the offered load tracks the paced rate. Returns end time. + */ +static uint64_t drive_backlogged(struct mb_ecn_ctx * ctx, + uint64_t * ftag, + uint64_t t, + uint64_t dur, + size_t len) +{ + uint64_t end = t + dur; + time_t w; + + while (t < end) { + w = mb_ecn_snd(ctx, len, t, ftag); + t += w > 0 ? (uint64_t) w : 1; + } + + return t; +} + +static int test_mb_ecn_ctx_create_destroy(void) +{ + struct mb_ecn_ctx * ctx; + + TEST_START(); + + ctx = mb_ecn_ctx_create(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + if (ctx->rate != CA_RATE_INIT) { + printf("Bad initial rate %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + if (ctx->rate_min != CA_RATE_MIN) { + printf("Bad initial floor %" PRIu64 ".\n", ctx->rate_min); + goto fail_ctx; + } + + if (ctx->vt != 0) { + printf("Bad initial virtual clock %" PRIu64 ".\n", ctx->vt); + goto fail_ctx; + } + + if (ctx->tx_cav) { + printf("Context did not start in slow start.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_calc_ecn(void) +{ + uint8_t ecn; + + TEST_START(); + + /* A queue below one ECN quantum marks nothing. */ + ecn = 0; + mb_ecn_calc_ecn(CA_MARK_Q - 1, &ecn, QOS_CUBE_BE, 0); + if (ecn != 0) { + printf("Sub-quantum queue marked %u.\n", ecn); + goto fail; + } + + /* Queue depth maps to ecn = queued / CA_MARK_Q. */ + ecn = 0; + mb_ecn_calc_ecn(5 * CA_MARK_Q, &ecn, QOS_CUBE_BE, 0); + if (ecn != 5) { + printf("Expected ecn 5, got %u.\n", ecn); + goto fail; + } + + /* MAX keeps the larger value; a smaller mark cannot raise it. */ + ecn = 0x80; + mb_ecn_calc_ecn(CA_MARK_Q, &ecn, QOS_CUBE_BE, 0); + if (ecn != 0x80) { + printf("Expected ecn 0x80, got 0x%x.\n", ecn); + goto fail; + } + + ecn = 3; + mb_ecn_calc_ecn(4 * CA_MARK_Q, &ecn, QOS_CUBE_BE, 0); + if (ecn != 4) { + printf("Expected ecn 4, got %u.\n", ecn); + goto fail; + } + + /* A queue past 255 quanta saturates, it does not wrap to a low mark. */ + ecn = 0; + mb_ecn_calc_ecn(256 * CA_MARK_Q, &ecn, QOS_CUBE_BE, 0); + if (ecn != 255) { + printf("Deep queue wrapped: exp 255, got %u.\n", ecn); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The first mark after idle emits the raw value with zero latency. */ +static int test_mb_ecn_rcv_onset_immediate(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + if (!mb_ecn_rcv(ctx, LEN, 4, 0, &ece, &fcap, MS)) { + printf("Onset did not update.\n"); + goto fail_ctx; + } + + if (ece != 4 << CA_SHFT) { + printf("Onset ece: exp %u, got %u.\n", 4 << CA_SHFT, ece); + goto fail_ctx; + } + + if (mb_ecn_rcv(ctx, LEN, 4, 0, &ece, &fcap, 2 * MS)) { + printf("Mid-window packet updated.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A 50% duty mark square wave emits the time mean, not the last peak. */ +static int test_mb_ecn_rcv_window_mean(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + size_t upd; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, MS); + + /* The byte trigger closes every ~32 packets: two windows. */ + upd = 0; + for (i = 1; i <= 68; i++) { + time_t ecn = (i & 1) ? 8 : 0; + time_t t = MS + i * MS; + if (mb_ecn_rcv(ctx, LEN, ecn, 0, &ece, &fcap, t)) + upd++; + } + + if (upd != 2) { + printf("%zu updates in two windows.\n", upd); + goto fail_ctx; + } + + if (ece < 120 || ece > 136) { + printf("window mean: exp ~128, got %u.\n", ece); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Two flows on the same wall-clock mark timeline, 15x apart in byte + * rate: the same congestion estimate, but the faster flow's window is + * shorter, so it feeds back more often (cadence tracks byte rate). + */ +static int test_mb_ecn_rcv_rate_independent(void) +{ + struct mb_ecn_ctx * a; + struct mb_ecn_ctx * b; + uint16_t ea; + uint16_t eb; + uint8_t fcap; + size_t ua; + size_t ub; + size_t i; + + TEST_START(); + + a = mk_ctx(); + b = mk_ctx(); + if (a == NULL || b == NULL) { + printf("Failed to create contexts.\n"); + goto fail_ctx; + } + + ea = 0; + eb = 0; + ua = 0; + ub = 0; + + /* 300 ms of sustained mark 8; a at 1 kpps, b at ~66 pps. */ + for (i = 1; i <= 300; i++) { + ua += mb_ecn_rcv(a, LEN, 8, 0, &ea, &fcap, i * MS) ? 1 : 0; + if (i % 15 != 0) + continue; + + ub += mb_ecn_rcv(b, LEN, 8, 0, &eb, &fcap, i * MS) ? 1 : 0; + } + + if (ea > eb + 32 || eb > ea + 32) { + printf("estimates diverge: %u vs %u.\n", ea, eb); + goto fail_ctx; + } + + if (ua < ub + 2) { + printf("cadence not rate-scaled: %zu vs %zu.\n", ua, ub); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * Two flows on one bottleneck, equal byte rate but 7.5x apart in + * packet size: the same congestion estimate and the same feedback + * cadence. Framing does not skew the control signal (fair share). + */ +static int test_mb_ecn_rcv_size_fair(void) +{ + struct mb_ecn_ctx * a; + struct mb_ecn_ctx * b; + uint16_t ea; + uint16_t eb; + uint8_t fcap; + size_t ua; + size_t ub; + uint64_t ta; + uint64_t tb; + + TEST_START(); + + a = mk_ctx(); + b = mk_ctx(); + if (a == NULL || b == NULL) { + printf("Failed to create contexts.\n"); + goto fail_ctx; + } + + ea = 0; + eb = 0; + ua = 0; + ub = 0; + ta = 0; + tb = 0; + + /* 1 MB/s each: a at 200 B / 200 us, b at 1500 B / 1.5 ms. */ + while (ta < 500 * MS) { + ta += 200 * 1000; + ua += mb_ecn_rcv(a, 200, 8, 0, &ea, &fcap, ta) ? 1 : 0; + } + + while (tb < 500 * MS) { + tb += 1500 * 1000; + ub += mb_ecn_rcv(b, 1500, 8, 0, &eb, &fcap, tb) ? 1 : 0; + } + + if (ea != 8 << CA_SHFT || eb != 8 << CA_SHFT) { + printf("size-skewed estimate: %u vs %u.\n", ea, eb); + goto fail_ctx; + } + + if (ua > ub + 3 || ub > ua + 3) { + printf("cadence skewed by size: %zu vs %zu.\n", ua, ub); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Release emits exactly one 0 and leaves the estimator fully idle. */ +static int test_mb_ecn_rcv_release_exact_zero(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + size_t ends; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + for (i = 1; i <= 140; i++) + mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, i * MS); + + ends = 0; + for (i = 141; i <= 350; i++) { + if (!mb_ecn_rcv(ctx, LEN, 0, 0, &ece, &fcap, i * MS)) + continue; + + if (ece == 0) + ends++; + } + + if (ends != 1) { + printf("end of congestion fired %zu times.\n", ends); + goto fail_ctx; + } + + if (ctx->rx_ece != 0 || ctx->rx_acc != 0) { + printf("estimator not idle: ece %u acc %" PRIu64 ".\n", + ctx->rx_ece, ctx->rx_acc); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A gap past the window restarts fresh: no stale, diluted estimate. */ +static int test_mb_ecn_rcv_gap_restart(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint64_t t; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_rcv(ctx, LEN, 6, 0, &ece, &fcap, MS); + mb_ecn_rcv(ctx, LEN, 6, 0, &ece, &fcap, 2 * MS); + + t = 2 * MS + 10 * CA_TW_INIT; + if (!mb_ecn_rcv(ctx, LEN, 5, 0, &ece, &fcap, t)) { + printf("gap restart did not update.\n"); + goto fail_ctx; + } + + if (ece != 5 << CA_SHFT) { + printf("gap restart: exp %u, got %u.\n", 5 << CA_SHFT, ece); + goto fail_ctx; + } + + t += 10 * CA_TW_INIT; + if (!mb_ecn_rcv(ctx, LEN, 0, 0, &ece, &fcap, t) || ece != 0) { + printf("gap with clean packet did not end: %u.\n", ece); + goto fail_ctx; + } + + if (mb_ecn_rcv(ctx, LEN, 0, 0, &ece, &fcap, t + MS)) { + printf("idle packet updated.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Max marks at max gaps: exact ceiling, no overflow past the edge. */ +static int test_mb_ecn_rcv_accum_bounds(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint64_t t; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, MS); + + /* Two packets at dt just under CA_TW_INIT straddle the boundary. */ + t = MS + CA_TW_INIT - 1; + if (mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, t)) { + printf("update before the window closed.\n"); + goto fail_ctx; + } + + t += CA_TW_INIT - 1; + if (!mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, t)) { + printf("no update at the window boundary.\n"); + goto fail_ctx; + } + + if (ece != 15 << CA_SHFT) { + printf("ceiling: exp %u, got %u.\n", 15 << CA_SHFT, ece); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Scale-free density: the window holds ~CA_N_TARGET packets at any rate. */ +static int test_mb_ecn_rcv_window_holds_target(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint16_t last_ece; + uint64_t rates[4]; + uint64_t ia; + uint64_t t; + size_t closes; + size_t since; + size_t count; + size_t ri; + + TEST_START(); + + rates[0] = 5000000; + rates[1] = 10000000; + rates[2] = 50000000; + rates[3] = 100000000; + + ctx = NULL; + for (ri = 0; ri < 4; ri++) { + ia = 8000ULL * BILLION / rates[ri]; + t = 0; + closes = 0; + since = 0; + count = 0; + last_ece = 0; + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Warm past the ramp from CA_TW_INIT, then time one gap. */ + while (closes < 42) { + t += ia; + since++; + if (!mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, t)) + continue; + closes++; + if (closes == 41) { + since = 0; + } else if (closes == 42) { + count = since - 1; + last_ece = ece; + } + } + + if (count < 8 || count > 32) { + printf("rate %" PRIu64 ": %zu pkts/window.\n", + rates[ri], count); + goto fail_ctx; + } + + if (last_ece < 224 || last_ece > 288) { + printf("rate %" PRIu64 ": ece %u ~256.\n", + rates[ri], last_ece); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + ctx = NULL; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * The window floors at CA_TW_MIN, tracks the rate below the old knee, + * and only a pathological fold hits the CA_TW_ABSMAX ceiling. + */ +static int test_mb_ecn_rcv_window_clip_bounds(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint64_t ia; + uint64_t t; + size_t closes; + + TEST_START(); + + /* 1 GbE is above the high knee: the window floors at CA_TW_MIN. */ + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ia = 8000ULL * BILLION / 1000000000ULL; + t = 0; + closes = 0; + while (closes < 40) { + t += ia; + if (mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, t)) + closes++; + } + + if (ctx->rx_tw != CA_TW_MIN) { + printf("high-rate window: exp %" PRIu64 ", got %" PRIu64 + ".\n", (uint64_t) CA_TW_MIN, ctx->rx_tw); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + /* 1 Mbps: past the old knee, ~16 pkts = 16 * 8 ms = 131 ms. */ + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ia = 8000ULL * BILLION / 1000000ULL; + t = 0; + closes = 0; + while (closes < 40) { + t += ia; + if (mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, t)) + closes++; + } + + if (ctx->rx_tw < 120 * MS || ctx->rx_tw > 140 * MS) { + printf("low-rate window: exp ~131 ms, got %" PRIu64 ".\n", + ctx->rx_tw); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + /* A near-empty window folds a huge target: ceiling holds. */ + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_rcv(ctx, 10, 8, 0, &ece, &fcap, MS); + mb_ecn_rcv(ctx, 10, 8, 0, &ece, &fcap, MS + CA_TW_INIT); + + if (ctx->rx_tw != CA_TW_ABSMAX) { + printf("window ceiling breached: %" PRIu64 ".\n", + ctx->rx_tw); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A CA-limited slow flow grows its window to hold ~16 packets. */ +static int test_mb_ecn_rcv_slow_window(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* 1400 B every 171 ms (~8 KB/s), sustained mark 8. */ + for (i = 1; i <= 100; i++) + mb_ecn_rcv(ctx, 1400, 8, 0, &ece, &fcap, i * 171 * MS); + + /* Target window 16 * 1000 B at 8187 B/s ~= 2.0 s. */ + if (ctx->rx_tw < 1400 * MS || ctx->rx_tw > 2800 * MS) { + printf("slow window: exp ~2 s, got %" PRIu64 ".\n", + ctx->rx_tw); + goto fail_ctx; + } + + /* Steady mark 8 emits exactly 256 once the window settles. */ + if (ece != 8 << CA_SHFT) { + printf("slow-flow estimate: exp %u, got %u.\n", + 8 << CA_SHFT, ece); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A physically maximal window must fold without overflow or wrap. */ +static int test_mb_ecn_rcv_no_overflow_highrate(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + bool ok; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Open a window, then inject a maximal byte count and span. */ + mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, 0); + ctx->rx_byt = CA_RATE_MAX / 8; + ctx->rx_ts = 2 * CA_TW_INIT - 2; + + ok = mb_ecn_rcv(ctx, LEN, 15, 0, &ece, &fcap, 2 * CA_TW_INIT - 1); + + if (!ok) { + printf("max-window close did not fire.\n"); + goto fail_ctx; + } + + if (ece > 8160) { + printf("estimate %u wrapped.\n", ece); + goto fail_ctx; + } + + /* A wrapped numerator drives rx_tw to MAX; it must descend. */ + if (ctx->rx_tw >= CA_TW_INIT || ctx->rx_tw < CA_TW_MIN) { + printf("window %" PRIu64 " did not descend.\n", ctx->rx_tw); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * The sender holds a mark across the full inter-feedback gap (TTL > + * 2 * CA_TW_INIT) and a repeated mark must not re-fire the one-sided lead. + */ +static int test_mb_ecn_ece_ttl_covers_cadence(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t t; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 100 << 20; + ctx->tx_cav = true; + + mb_ecn_ece(ctx, 100, 0, MS); + mb_ecn_snd(ctx, LEN, MS, &ftag); + + /* Sends between feedbacks spaced 2 * CA_TW_INIT + 5 ms apart. */ + t = MS; + for (i = 0; i < 4; i++) { + t += (2 * CA_TW_INIT + 5 * MS) / 4; + mb_ecn_snd(ctx, LEN, t, &ftag); + if (ctx->tx_ece == 0) { + printf("mark cleared inside the feedback gap.\n"); + goto fail_ctx; + } + } + + /* Same mark again: rise 0, so only the proportional cut. */ + prev = ctx->rate; + t += MS; + mb_ecn_ece(ctx, 100, 0, t); + mb_ecn_snd(ctx, LEN, t, &ftag); + + if (prev - ctx->rate > prev / 100) { + printf("phantom lead cut: %" PRIu64 " -> %" PRIu64 ".\n", + prev, ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_slow_start(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t t; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + prev = ctx->rate; + t = 0; + + /* No feedback: the flow stays in slow start and grows each step. */ + for (i = 0; i < 16; i++) { + t += MS; + mb_ecn_snd(ctx, LEN, t, &ftag); + if (ctx->rate <= prev) { + printf("rate did not grow: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + prev = ctx->rate; + } + + /* Exponential ramp doubles in ~ln2 * CA_SS_TC ~= 14 ms. */ + if (ctx->rate < 2 * CA_RATE_INIT) { + printf("slow start too slow: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_dt_scaling_invariant(void) +{ + struct mb_ecn_ctx * a; + struct mb_ecn_ctx * b; + uint64_t inc_a; + uint64_t inc_b; + uint64_t t; + uint64_t fta = 0; + uint64_t ftb = 0; + size_t i; + + TEST_START(); + + a = mk_ctx(); + b = mk_ctx(); + if (a == NULL || b == NULL) { + printf("Failed to create contexts.\n"); + goto fail_ctx; + } + + /* Leave slow start; seed a realistic rate (truncation-free). */ + mb_ecn_ece(a, 0, 0, 0); + mb_ecn_ece(b, 0, 0, 0); + a->rate = (uint64_t) 10 << 20; + b->rate = (uint64_t) 10 << 20; + a->r_bkt = a->rate; + b->r_bkt = b->rate; + + /* a: one 30 ms step (under one washout bucket, so it stays out). */ + mb_ecn_snd(a, LEN, 30 * MS, &fta); + + /* b: thirty 1 ms steps over the same 30 ms. */ + t = 0; + for (i = 0; i < 30; i++) { + t += MS; + mb_ecn_snd(b, LEN, t, &ftb); + } + + inc_a = a->rate - ((uint64_t) 10 << 20); + inc_b = b->rate - ((uint64_t) 10 << 20); + + /* Equal within 1 %; the small gap is per-step integer truncation. */ + if (inc_a == 0 || inc_b == 0) { + printf("no additive increase: %" PRIu64 " %" PRIu64 ".\n", + inc_a, inc_b); + goto fail_ctx; + } + + if (inc_a > inc_b + inc_a / 100 || inc_b > inc_a + inc_a / 100) { + printf("cadence-dependent AI: %" PRIu64 " vs %" PRIu64 ".\n", + inc_a, inc_b); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_multiplicative_decrease(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t t; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 100 << 20; + prev = ctx->rate; + t = 0; + + for (i = 0; i < 10; i++) { + t += MS; + mb_ecn_ece(ctx, CA_ECE_REF, 0, t); + mb_ecn_snd(ctx, LEN, t, &ftag); + if (ctx->rate >= prev) { + printf("rate did not shrink: %" PRIu64 ".\n", + ctx->rate); + goto fail_ctx; + } + + prev = ctx->rate; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_rate_floor(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Cut larger than headroom must clamp; stay inside CA_ECE_TTL. */ + ctx->rate = CA_RATE_MIN + 1000; + mb_ecn_ece(ctx, CA_ECE_REF, 0, 0); + mb_ecn_snd(ctx, LEN, 30 * MS, &ftag); + + if (ctx->rate != CA_RATE_MIN) { + printf("rate floor breached: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_mb_ecn_fixed_point(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t exp; + uint64_t t; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + exp = CA_AI_RATE * CA_ECE_REF / + (128 - CA_ECE_REF * BILLION / CA_PROBE_TC); + t = 0; + + /* ~20 s: the probe raises the loop time constant to CA_PROBE_TC. */ + for (i = 0; i < 20000; i++) { + t += MS; + mb_ecn_ece(ctx, 128, 0, t); + mb_ecn_snd(ctx, LEN, t, &ftag); + } + + if (ctx->rate < exp - exp / 4 || ctx->rate > exp + exp / 4) { + printf("no fixed point: exp ~%" PRIu64 ", got %" PRIu64 ".\n", + exp, ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * One-sided lead: a mark rise cuts rate * rise / (REF * DEN) once; + * a flat or falling mark leaves only the dt-scaled proportional cut. + */ +static int test_mb_ecn_lead_cut(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t drop; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 100 << 20; + ctx->tx_cav = true; + + /* Rise 0 -> 256: lead cuts ~rise/(REF*DEN) = 1/8 of the rate. */ + prev = ctx->rate; + mb_ecn_ece(ctx, 256, 0, MS); + mb_ecn_snd(ctx, LEN, MS, &ftag); + drop = prev - ctx->rate; + if (drop < prev / 10) { + printf("lead cut missing: dropped %" PRIu64 ".\n", drop); + goto fail_ctx; + } + + /* Flat mark: rise 0, only the ~0.05%% proportional cut. */ + prev = ctx->rate; + mb_ecn_ece(ctx, 256, 0, 2 * MS); + mb_ecn_snd(ctx, LEN, 2 * MS, &ftag); + drop = prev - ctx->rate; + if (drop > prev / 100) { + printf("flat mark over-cut: dropped %" PRIu64 ".\n", drop); + goto fail_ctx; + } + + /* Falling mark: one-sided lead must not fire. */ + prev = ctx->rate; + mb_ecn_ece(ctx, 64, 0, 3 * MS); + mb_ecn_snd(ctx, LEN, 3 * MS, &ftag); + drop = prev > ctx->rate ? prev - ctx->rate : 0; + if (drop > prev / 100) { + printf("falling mark cut: dropped %" PRIu64 ".\n", drop); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A local first-hop mark exits slow start with no feedback needed. */ +static int test_mb_ecn_slow_start_local_brake(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + for (i = 1; i <= 50; i++) + mb_ecn_snd(ctx, LEN, i * MS, &ftag); + + if (ctx->tx_cav) { + printf("Left slow start without any signal.\n"); + goto fail_ctx; + } + + prev = ctx->rate; + mb_ecn_loc(ctx, 1, 50 * MS); + if (!ctx->tx_cav) { + printf("Local mark did not exit slow start.\n"); + goto fail_ctx; + } + + mb_ecn_snd(ctx, LEN, 51 * MS, &ftag); + if (ctx->rate > prev + prev / 20) { + printf("SS ramp survived the brake: %" PRIu64 ".\n", + ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A clean path still ramps to line rate in well under a second. */ +static int test_mb_ecn_slow_start_clean_ramp(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Backlogged, no marks: slow start sprints in a couple windows. */ + drive_backlogged(ctx, &ftag, MS, 2 * CA_SND_WIN, LEN); + + if (ctx->rate < (1ULL << 24)) { + printf("backlogged slow start too slow: %" PRIu64 ".\n", + ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * A sender starved of send-path control steps recovers through the + * feedback path: honest elapsed time, at most a 50% cut per step. + */ +static int test_mb_ecn_starved_decrease_escape(void) +{ + struct mb_ecn_ctx * ctx; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 100 << 20; + ctx->tx_cav = true; + + /* Feedback arrives once per second; no sends at all. */ + for (i = 1; i <= 6; i++) + mb_ecn_ece(ctx, 480, 0, i * BILLION); + + if (ctx->rate > (5ULL << 19)) { + printf("still wedged at %" PRIu64 " B/s.\n", ctx->rate); + goto fail_ctx; + } + + if (ctx->rate < CA_RATE_MIN) { + printf("rate floor breached: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* With feedback fully dead, the local mark alone recovers the rate. */ +static int test_mb_ecn_starved_local_fallback(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t t; + uint64_t ftag = 0; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 100 << 20; + + for (i = 1; i <= 7; i++) { + t = i * BILLION; + mb_ecn_loc(ctx, 15, t); + mb_ecn_snd(ctx, LEN, t, &ftag); + } + + if (!ctx->tx_cav) { + printf("Local mark did not exit slow start.\n"); + goto fail_ctx; + } + + if (ctx->rate > (5ULL << 19)) { + printf("still wedged at %" PRIu64 " B/s.\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * MD Δt-invariance: the same elapsed time under the same mark cuts + * the same, in one big step or five small ones. + */ +static int test_mb_ecn_decrease_dt_invariant(void) +{ + struct mb_ecn_ctx * a; + struct mb_ecn_ctx * b; + uint64_t cut_a; + uint64_t cut_b; + uint64_t r0; + uint64_t fta = 0; + uint64_t ftb = 0; + size_t i; + + TEST_START(); + + a = mk_ctx(); + b = mk_ctx(); + if (a == NULL || b == NULL) { + printf("Failed to create contexts.\n"); + goto fail_ctx; + } + + r0 = (uint64_t) 100 << 20; + a->rate = r0; + b->rate = r0; + + mb_ecn_ece(a, 256, 0, 0); + mb_ecn_ece(b, 256, 0, 0); + + /* a: one 50 ms step; b: five 10 ms steps (both within DT_CAP). */ + mb_ecn_snd(a, LEN, 50 * MS, &fta); + + for (i = 1; i <= 5; i++) + mb_ecn_snd(b, LEN, i * 10 * MS, &ftb); + + cut_a = r0 - a->rate; + cut_b = r0 - b->rate; + + if (cut_a == 0 || cut_b == 0) { + printf("no cut: %" PRIu64 " %" PRIu64 ".\n", cut_a, cut_b); + goto fail_ctx; + } + + /* Within 10%: residual is Euler compounding of MD and the probe. */ + if (cut_a > cut_b + cut_a / 10 || cut_b > cut_a + cut_a / 10) { + printf("cadence-dependent MD: %" PRIu64 " vs %" PRIu64 + ".\n", cut_a, cut_b); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Resuming after a long idle gap: bounded AI, no cut from stale marks. */ +static int test_mb_ecn_idle_resume_bounded(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t prev; + uint64_t bump; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 10 << 20; + + mb_ecn_loc(ctx, 15, MS); + mb_ecn_ece(ctx, 480, 0, MS); + + prev = ctx->rate; + + /* 600 s later: both signals stale; one capped AI + probe step. */ + mb_ecn_snd(ctx, LEN, 600 * BILLION, &ftag); + + if (ctx->rate < prev) { + printf("stale mark cut the rate: %" PRIu64 ".\n", + ctx->rate); + goto fail_ctx; + } + + bump = CA_AI_RATE * CA_DT_CAP / BILLION; + bump += (prev + bump) * CA_DT_CAP / CA_PROBE_TC; + + if (ctx->rate > prev + bump + 2) { + printf("idle resume cap breached: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The congestion signal ages out on wall-clock time, not packet count. */ +static int test_mb_ecn_ece_staleness(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* A fast flow's control step caches the floor TTL. */ + ctx->rate = (uint64_t) 1 << 20; + ctx->tx_cav = true; + mb_ecn_snd(ctx, LEN, MS, &ftag); + + if (ctx->ece_ttl != CA_ECE_TTL) { + printf("Fast-flow TTL: exp %" PRIu64 ", got %" PRIu64 ".\n", + (uint64_t) CA_ECE_TTL, ctx->ece_ttl); + goto fail_ctx; + } + + mb_ecn_ece(ctx, CA_ECE_REF, 0, 2 * MS); + + /* Just inside the TTL: the signal is still held. */ + mb_ecn_snd(ctx, LEN, 2 * MS + ctx->ece_ttl, &ftag); + if (ctx->tx_ece == 0) { + printf("signal aged out too early.\n"); + goto fail_ctx; + } + + /* Past the TTL without feedback: the signal is cleared. */ + mb_ecn_snd(ctx, LEN, 2 * MS + ctx->ece_ttl + 1, &ftag); + if (ctx->tx_ece != 0) { + printf("stale signal not cleared: %u.\n", ctx->tx_ece); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The staleness horizon stretches with a slow flow's window. */ +static int test_mb_ecn_ece_ttl_tracks_rate(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t want; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = 8192; + ctx->rate_min = 8192; + ctx->ai_rate = 0; + ctx->tx_cav = true; + mb_ecn_snd(ctx, 1400, MS, &ftag); + + want = (1 << CA_TW_GAP_SHFT) * CA_RX_WBYTES * BILLION / ctx->rate; + if (ctx->ece_ttl != want) { + printf("Slow-flow TTL: exp %" PRIu64 ", got %" PRIu64 ".\n", + want, ctx->ece_ttl); + goto fail_ctx; + } + + if (ctx->ece_ttl < 7 * (uint64_t) BILLION) { + printf("TTL did not stretch: %" PRIu64 ".\n", + ctx->ece_ttl); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * First packet of a flow starts at the clock; a same-instant second + * packet leads by its length and is paced by lead / rate. + */ +static int test_mb_ecn_sfq_pace(void) +{ + struct mb_ecn_ctx * ctx; + time_t wait; + uint64_t ftag = 0; + time_t want; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = 1U << 20; + ctx->inv_rate = mb_ecn_rate_inv(ctx->rate); + + /* First send: start tag equals the clock, so no wait. */ + wait = mb_ecn_snd(ctx, 1500, 0, &ftag); + if (wait != 0) { + printf("first packet waited %ld, expected 0.\n", (long) wait); + goto fail_ctx; + } + + /* Same instant (dt = 0): the flow now leads by 1500 B. */ + wait = mb_ecn_snd(ctx, 1500, 0, &ftag); + want = (time_t) ((uint64_t) 1500 * BILLION / ctx->rate); + + if (wait != want) { + printf("paced wait %ld, expected %ld.\n", + (long) wait, (long) want); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * The proportional probe grows the rate by the same fraction per unit + * time regardless of the absolute rate: two clean flows 100x apart in + * rate grow by the same ratio. Deleting the probe leaves only the tiny + * additive increase, failing the growth floor. + */ +static int test_mb_ecn_probe_scale_invariant(void) +{ + struct mb_ecn_ctx * a; + struct mb_ecn_ctx * b; + uint64_t ra0; + uint64_t rb0; + double ga; + double gb; + uint64_t t; + size_t i; + + TEST_START(); + + a = mk_ctx(); + b = mk_ctx(); + if (a == NULL || b == NULL) { + printf("Failed to create contexts.\n"); + goto fail_ctx; + } + + /* Clean path (mark 0), out of slow start, backlogged, 100x apart. */ + mb_ecn_ece(a, 0, 0, 0); + mb_ecn_ece(b, 0, 0, 0); + a->rate = (uint64_t) 10 << 20; + b->rate = (uint64_t) 1000 << 20; + a->r_bkt = a->rate; + b->r_bkt = b->rate; + a->backlogged = true; + b->backlogged = true; + ra0 = a->rate; + rb0 = b->rate; + + /* Drive control via the feedback path so backlogged stays set. */ + t = 0; + for (i = 0; i < 500; i++) { + t += MS; + mb_ecn_ece(a, 0, 0, t); + mb_ecn_ece(b, 0, 0, t); + } + + ga = (double) a->rate / ra0; + gb = (double) b->rate / rb0; + + if (ga < gb - gb / 50 || gb < ga - ga / 50) { + printf("probe not scale-invariant: %.4f vs %.4f.\n", ga, gb); + goto fail_ctx; + } + + /* And it must actually grow: the probe is present, not deleted. */ + if (ga < 1.05) { + printf("probe did not grow the rate: %.4f.\n", ga); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(a); + mb_ecn_ctx_destroy(b); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * The proportional probe e-folds the rate over CA_PROBE_TC: a clean flow + * grows by ~e in 8 s. Pinned to a literal e-band so a mistuned + * CA_PROBE_TC (e.g. 4 s gives e^2) fails. + */ +static int test_mb_ecn_probe_time_constant(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t r0; + double ratio; + uint64_t t; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Clean path, out of slow start, backlogged, below the ceiling. */ + mb_ecn_ece(ctx, 0, 0, 0); + ctx->rate = (uint64_t) 1 << 30; + ctx->r_bkt = ctx->rate; + ctx->backlogged = true; + r0 = ctx->rate; + + /* 8000 x 1 ms of clean growth, driven via the feedback path. */ + t = 0; + for (i = 0; i < 8000; i++) { + t += MS; + mb_ecn_ece(ctx, 0, 0, t); + } + + /* Washout damps the probe to ~4/3 TC, so 8 s -> ~2.12x. */ + ratio = (double) ctx->rate / r0; + if (ratio < 2.0 || ratio > 2.25) { + printf("probe TC off: exp ~2.12, got %.3fx over 8 s.\n", + ratio); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The fed-back capacity is the MIN of the nonzero caps in the window. */ +static int test_mb_ecn_rcv_cap_window_min(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint64_t t; + bool upd; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Onset packet carries no capacity: feed back unknown. */ + if (!mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, MS)) { + printf("Onset did not update.\n"); + goto fail_ctx; + } + + if (fcap != 0) { + printf("Onset fed back cap: exp 0, got %u.\n", fcap); + goto fail_ctx; + } + + mb_ecn_rcv(ctx, LEN, 8, 40, &ece, &fcap, 2 * MS); + mb_ecn_rcv(ctx, LEN, 8, 36, &ece, &fcap, 3 * MS); + + t = 3 * MS + CA_TW_INIT; + if (!mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, t)) { + printf("Window did not close.\n"); + goto fail_ctx; + } + + if (fcap != 36) { + printf("Window min cap: exp 36, got %u.\n", fcap); + goto fail_ctx; + } + + /* The next window starts unknown; follow the adapted rx_tw. */ + upd = false; + for (i = 0; i < 128 && !upd; i++) { + t += CA_TW_INIT; + upd = mb_ecn_rcv(ctx, LEN, 8, 0, &ece, &fcap, t); + } + + if (!upd) { + printf("Second window did not close.\n"); + goto fail_ctx; + } + + if (fcap != 0) { + printf("Stale cap %u leaked into the next window.\n", fcap); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Onset and gap restarts emit the triggering packet's cap, fresh. */ +static int test_mb_ecn_rcv_cap_onset_fresh(void) +{ + struct mb_ecn_ctx * ctx; + uint16_t ece; + uint8_t fcap; + uint64_t t; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + if (!mb_ecn_rcv(ctx, LEN, 4, 77, &ece, &fcap, MS)) { + printf("Onset did not update.\n"); + goto fail_ctx; + } + + if (fcap != 77) { + printf("Onset cap: exp 77, got %u.\n", fcap); + goto fail_ctx; + } + + mb_ecn_rcv(ctx, LEN, 4, 50, &ece, &fcap, 2 * MS); + + /* A gap restart must not fold in the stale window min. */ + t = 2 * MS + 5 * CA_TW_INIT; + if (!mb_ecn_rcv(ctx, LEN, 4, 90, &ece, &fcap, t)) { + printf("Gap restart did not update.\n"); + goto fail_ctx; + } + + if (fcap != 90) { + printf("Gap restart cap: exp 90, got %u.\n", fcap); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Fed-back capacity derives the floor and slope: EWMA toward C/32. */ +static int test_mb_ecn_ece_cap_derives_rates(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t tgt; + uint64_t want; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Code 120 = 2^30 B/s; target floor = 2^25 B/s. */ + tgt = cap_dec(120) >> CA_CAP_SHFT; + + mb_ecn_ece(ctx, 100, 120, MS); + + want = CA_RATE_MIN + ((tgt - CA_RATE_MIN) >> CA_CAP_SM_SHFT); + if (ctx->rate_min != want) { + printf("Floor: exp %" PRIu64 ", got %" PRIu64 ".\n", + want, ctx->rate_min); + goto fail_ctx; + } + + if (ctx->ai_rate != ctx->rate_min) { + printf("AI slope did not track the floor.\n"); + goto fail_ctx; + } + + mb_ecn_ece(ctx, 100, 120, 2 * MS); + + want += (tgt - want) >> CA_CAP_SM_SHFT; + if (ctx->rate_min != want) { + printf("Floor EWMA: exp %" PRIu64 ", got %" PRIu64 ".\n", + want, ctx->rate_min); + goto fail_ctx; + } + + if (ctx->n_cap != 2) { + printf("Capacity updates: exp 2, got %" PRIu64 ".\n", + ctx->n_cap); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Feedback without a capacity leaves the derived rates untouched. */ +static int test_mb_ecn_ece_cap_zero_keeps_rates(void) +{ + struct mb_ecn_ctx * ctx; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_ece(ctx, 100, 0, MS); + + if (ctx->rate_min != CA_RATE_MIN || ctx->ai_rate != CA_AI_RATE) { + printf("Unknown cap moved the derived rates.\n"); + goto fail_ctx; + } + + if (ctx->n_cap != 0) { + printf("Unknown cap counted as an update.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The derived floor clamps to [CA_RATE_MIN, CA_RMIN_MAX]. */ +static int test_mb_ecn_ece_cap_clamps(void) +{ + struct mb_ecn_ctx * ctx; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* A path slower than the default floor cannot lower it. */ + mb_ecn_ece(ctx, 100, 1, MS); + + if (ctx->rate_min != CA_RATE_MIN) { + printf("Slow path lowered the floor: %" PRIu64 ".\n", + ctx->rate_min); + goto fail_ctx; + } + + /* A absurdly fast path saturates at the ceiling. */ + for (i = 1; i <= 40; i++) + mb_ecn_ece(ctx, 100, 255, (1 + i) * MS); + + if (ctx->rate_min > CA_RMIN_MAX) { + printf("Floor above the ceiling: %" PRIu64 ".\n", + ctx->rate_min); + goto fail_ctx; + } + + if (ctx->rate_min < CA_RMIN_MAX - 4) { + printf("Floor did not reach the ceiling: %" PRIu64 ".\n", + ctx->rate_min); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Stale capacity reverts the derived rates to the defaults. */ +static int test_mb_ecn_cap_ttl_reverts(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + mb_ecn_ece(ctx, 100, 120, MS); + + if (ctx->rate_min == CA_RATE_MIN) { + printf("Capacity did not derive a floor.\n"); + goto fail_ctx; + } + + /* Just inside the TTL: the derived rates hold. */ + mb_ecn_snd(ctx, LEN, MS + (ctx->ece_ttl << CA_CAP_TTL_SHFT), &ftag); + + if (ctx->rate_min == CA_RATE_MIN) { + printf("Derived rates reverted too early.\n"); + goto fail_ctx; + } + + /* Past the TTL: back to the defaults. */ + mb_ecn_snd(ctx, LEN, MS + (ctx->ece_ttl << CA_CAP_TTL_SHFT) + 1, + &ftag); + + if (ctx->rate_min != CA_RATE_MIN || ctx->ai_rate != CA_AI_RATE) { + printf("Stale capacity kept the derived rates.\n"); + goto fail_ctx; + } + + if (ctx->tx_cap != 0) { + printf("Stale capacity code not cleared.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The control law uses the per-ctx AI slope. */ +static int test_mb_ecn_ctrl_per_ctx_ai(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t want; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* Leave slow start; raise the slope as capacity would. */ + mb_ecn_ece(ctx, 0, 0, 0); + ctx->rate = (uint64_t) 10 << 20; + ctx->r_bkt = ctx->rate; + ctx->ai_rate = 16 * CA_AI_RATE; + + want = ctx->rate + ctx->ai_rate * (30 * MS) / BILLION; + want += want * (30 * MS) / CA_PROBE_TC; + + mb_ecn_snd(ctx, LEN, 30 * MS, &ftag); + + if (ctx->rate != want) { + printf("AI not per-ctx: exp %" PRIu64 ", got %" PRIu64 ".\n", + want, ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The rate clamp honours the per-ctx derived floor. */ +static int test_mb_ecn_ctrl_per_ctx_floor(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate_min = (uint64_t) 1 << 20; + ctx->rate = ((uint64_t) 1 << 20) + 1000; + + mb_ecn_ece(ctx, CA_ECE_REF, 0, 0); + mb_ecn_snd(ctx, LEN, 30 * MS, &ftag); + + if (ctx->rate != ctx->rate_min) { + printf("Floor not per-ctx: %" PRIu64 ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * A paced flow slower than one packet per CA_DT_CAP must not decay: + * the gap credit law grants its true elapsed service, so the lead + * stays pinned at ~one packet instead of growing without bound. + */ +static int test_mb_ecn_snd_slow_rate_paced(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + uint64_t t = 0; + time_t wait; + size_t i; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* 8 KB/s, 1400 B packets: inter-send gap ~171 ms > CA_DT_CAP. */ + ctx->rate = 8192; + ctx->rate_min = 8192; + ctx->ai_rate = 0; + ctx->inv_rate = mb_ecn_rate_inv(8192); + ctx->tx_cav = true; + + for (i = 0; i < 50; i++) { + wait = mb_ecn_snd(ctx, 1400, t, &ftag); + t += wait > 0 ? (uint64_t) wait : 1; + } + + if (ctx->lead > 2 * 1400) { + printf("Pacer starves a slow flow: lead %" PRIu64 ".\n", + ctx->lead); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * A long idle makes the aggregate source-limited, so the offered-load + * ceiling bounds the resume rate (hence the burst) well below the + * pre-idle rate. + */ +static int test_mb_ecn_snd_idle_burst_bound(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = (uint64_t) 1 << 20; + ctx->inv_rate = mb_ecn_rate_inv(ctx->rate); + ctx->tx_cav = true; + + mb_ecn_snd(ctx, 1400, 0, &ftag); /* warm-up: sets started */ + + /* 600 s idle. */ + mb_ecn_snd(ctx, 1400, 600 * BILLION, &ftag); + + if (ctx->backlogged) { + printf("long idle did not clear backlogged.\n"); + goto fail_ctx; + } + + if (ctx->rate >= ((uint64_t) 1 << 20)) { + printf("idle resume rate not ceiling-bounded: %" PRIu64 + ".\n", ctx->rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A deep backlog at a low rate must not wrap the wait computation. */ +static int test_mb_ecn_snd_wait_no_overflow(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t want; + uint64_t ftag; + time_t wait; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->rate = 8192; + ctx->inv_rate = mb_ecn_rate_inv(8192); + + /* 128 flows x 1400 B of SFQ lead at the floor rate. */ + ftag = 128 * 1400; + + wait = mb_ecn_snd(ctx, 1400, 0, &ftag); + want = (uint64_t) 128 * 1400 * BILLION / 8192; + + if ((uint64_t) wait < want - want / 100 || + (uint64_t) wait > want + want / 100) { + printf("Wait wrapped: exp ~%" PRIu64 ", got %" PRIu64 ".\n", + want, (uint64_t) wait); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A fully paced-backlogged flow reads backlogged after a window. */ +static int test_mb_ecn_backlogged_paced(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->backlogged = false; /* prove a window close re-earns it */ + + drive_backlogged(ctx, &ftag, MS, 4 * CA_SND_WIN, LEN); + + if (!ctx->backlogged) { + printf("paced-backlogged flow read source-limited.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* + * A source-limited flow is capped to the next quarter-log2 headroom + * above the offered estimate, and NOT re-floored to a high capacity + * rate_min. + */ +static int test_mb_ecn_source_limited_ceiling(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + uint64_t t = 10 * MS; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->tx_cav = true; + ctx->started = true; + ctx->backlogged = false; + ctx->rate = (uint64_t) 100 << 20; + ctx->inv_rate = mb_ecn_rate_inv(ctx->rate); + ctx->rate_min = (uint64_t) 50 << 20; + ctx->snd_rate = (uint64_t) 1 << 20; + ctx->snd_r0 = ctx->rate; + ctx->snd_win = t; + ctx->last_ts = t; + ctx->last_ctrl = t; + + mb_ecn_snd(ctx, LEN, t + 2 * MS, &ftag); + + if (ctx->rate != ((uint64_t) 2 << 20)) { + printf("ceiling: exp %" PRIu64 ", got %" PRIu64 ".\n", + (uint64_t) 2 << 20, ctx->rate); + goto fail_ctx; + } + + if (!ctx->src_limited) { + printf("ceiling bound but src_limited not set.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* One quiet window must not collapse the max-filter; it decays ~1/16. */ +static int test_mb_ecn_max_filter(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + uint64_t hi = (uint64_t) 10 << 20; + uint64_t t = 10 * MS; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->started = true; + ctx->snd_rate = hi; + ctx->snd_r0 = hi; + ctx->snd_byt = 0; + ctx->snd_win = t; + ctx->last_ts = t; + ctx->last_ctrl = t; + + /* Close one window with almost no bytes offered. */ + mb_ecn_snd(ctx, LEN, t + CA_SND_WIN + 1, &ftag); + + if (ctx->snd_rate >= hi || ctx->snd_rate < hi - hi / 8) { + printf("max-filter: exp ~15/16 of %" PRIu64 ", got %" + PRIu64 " after one quiet window.\n", + hi, ctx->snd_rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* A >CA_DT_CAP gap clears backlogged without touching the estimate. */ +static int test_mb_ecn_idle_clears_backlogged(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + uint64_t snd_rate = (uint64_t) 5 << 20; + uint64_t t = 10 * MS; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + ctx->started = true; + ctx->backlogged = true; + ctx->snd_rate = snd_rate; + ctx->snd_win = t; + ctx->last_ts = t; + ctx->last_ctrl = t; + + /* 55 ms gap: past CA_DT_CAP, under CA_SND_WIN (no window close). */ + mb_ecn_snd(ctx, LEN, t + 55 * MS, &ftag); + + if (ctx->backlogged) { + printf("idle gap did not clear backlogged.\n"); + goto fail_ctx; + } + + if (ctx->snd_rate != snd_rate) { + printf("idle step altered snd_rate %" PRIu64 ".\n", + ctx->snd_rate); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* The first send is never misread as idle, whatever the wall clock. */ +static int test_mb_ecn_first_send_warmup(void) +{ + struct mb_ecn_ctx * ctx; + uint64_t ftag = 0; + + TEST_START(); + + ctx = mk_ctx(); + if (ctx == NULL) { + printf("Failed to create context.\n"); + goto fail; + } + + /* started == false; a large first timestamp must not look idle. */ + mb_ecn_snd(ctx, LEN, 500 * MS, &ftag); + + if (!ctx->started) { + printf("first send did not set the warm-up sentinel.\n"); + goto fail_ctx; + } + + if (!ctx->backlogged) { + printf("first send misclassified as idle.\n"); + goto fail_ctx; + } + + mb_ecn_ctx_destroy(ctx); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_ctx: + mb_ecn_ctx_destroy(ctx); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int mb_ecn_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_mb_ecn_ctx_create_destroy(); + ret |= test_mb_ecn_calc_ecn(); + ret |= test_mb_ecn_rcv_onset_immediate(); + ret |= test_mb_ecn_rcv_window_mean(); + ret |= test_mb_ecn_rcv_rate_independent(); + ret |= test_mb_ecn_rcv_size_fair(); + ret |= test_mb_ecn_rcv_release_exact_zero(); + ret |= test_mb_ecn_rcv_gap_restart(); + ret |= test_mb_ecn_rcv_accum_bounds(); + ret |= test_mb_ecn_rcv_window_holds_target(); + ret |= test_mb_ecn_rcv_window_clip_bounds(); + ret |= test_mb_ecn_rcv_slow_window(); + ret |= test_mb_ecn_rcv_no_overflow_highrate(); + ret |= test_mb_ecn_ece_ttl_covers_cadence(); + ret |= test_mb_ecn_slow_start(); + ret |= test_mb_ecn_dt_scaling_invariant(); + ret |= test_mb_ecn_probe_scale_invariant(); + ret |= test_mb_ecn_multiplicative_decrease(); + ret |= test_mb_ecn_fixed_point(); + ret |= test_mb_ecn_lead_cut(); + ret |= test_mb_ecn_slow_start_local_brake(); + ret |= test_mb_ecn_slow_start_clean_ramp(); + ret |= test_mb_ecn_starved_decrease_escape(); + ret |= test_mb_ecn_starved_local_fallback(); + ret |= test_mb_ecn_decrease_dt_invariant(); + ret |= test_mb_ecn_idle_resume_bounded(); + ret |= test_mb_ecn_rate_floor(); + ret |= test_mb_ecn_ece_staleness(); + ret |= test_mb_ecn_ece_ttl_tracks_rate(); + ret |= test_mb_ecn_sfq_pace(); + ret |= test_mb_ecn_probe_time_constant(); + ret |= test_mb_ecn_rcv_cap_window_min(); + ret |= test_mb_ecn_rcv_cap_onset_fresh(); + ret |= test_mb_ecn_ece_cap_derives_rates(); + ret |= test_mb_ecn_ece_cap_zero_keeps_rates(); + ret |= test_mb_ecn_ece_cap_clamps(); + ret |= test_mb_ecn_cap_ttl_reverts(); + ret |= test_mb_ecn_ctrl_per_ctx_ai(); + ret |= test_mb_ecn_ctrl_per_ctx_floor(); + ret |= test_mb_ecn_snd_slow_rate_paced(); + ret |= test_mb_ecn_snd_idle_burst_bound(); + ret |= test_mb_ecn_snd_wait_no_overflow(); + ret |= test_mb_ecn_backlogged_paced(); + ret |= test_mb_ecn_source_limited_ceiling(); + ret |= test_mb_ecn_max_filter(); + ret |= test_mb_ecn_idle_clears_backlogged(); + ret |= test_mb_ecn_first_send_warmup(); + + return ret; +} diff --git a/src/ipcpd/unicast/cap.c b/src/ipcpd/unicast/cap.c new file mode 100644 index 00000000..0d823dc6 --- /dev/null +++ b/src/ipcpd/unicast/cap.c @@ -0,0 +1,291 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Link capacity estimation + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include "config.h" + +#include <ouroboros/atomics.h> +#include <ouroboros/time.h> + +#include "cap.h" + +#include <string.h> + +/* + * Link-capacity estimation by watching the egress queue drain. + * + * A saturated link drains its queue at exactly its capacity, so we + * estimate capacity by measuring the drain rate of the ring buffer + * toward an n-1 flow (the flow to the layer below) while that ring + * is backlogged. + * + * Sampling is lock-free and off the fast path: the ring depth is + * read only at enqueue time, concurrently by many sender threads. + * Each enqueue bumps relaxed counters (packets, bytes, empty-ring + * hits). At most once per CAP_T_MIN, one thread wins a try-lock and + * closes a measurement window. + * + * Over a window, packet conservation gives the slots that drained: + * drained = queue at start (q0) + enqueued - queue now (q1) + * A window stays open until CAP_N_MIN slots have drained, so its + * length self-scales with the link rate (~1 ms at 1 Gbit, ~19 ms at + * 10 Mbit). CAP_T_MAX discards a window that spanned a traffic gap. + * + * Only a backlogged link measures its own capacity, so a window + * whose ring ran mostly idle is discarded (a few empty samples, as + * from a token-bucket shaper, are tolerated). The drain rate feeds a + * max filter that jumps up at once but decays slowly, converging on + * the capacity from below. A window that touched an empty ring at + * either edge may have drained into downstream buffers faster than + * the wire, so it may only lower the estimate, never raise it. + * + * The estimate is published as a quarter-log2 code: capacity is only + * ever needed to order-of-magnitude accuracy. + */ + +#define CAP_T_MIN (BILLION / 1000) /* min fold spacing ~1 ms */ +#define CAP_T_MAX (1ULL << 27) /* stale window cap ~134 ms */ +#define CAP_N_MIN 16 /* drained slots to close */ +#define CAP_DEC_SHFT 4 /* max-filter decay 1/16 */ +#define CAP_IDL_SHFT 3 /* idle tolerance 1/8 */ + +/* Try-lock on the busy flag: test-and-set acquire, store release. */ +#define CAP_TRY(p) (__atomic_exchange_n(p, 1, __ATOMIC_ACQUIRE) == 0) +#define CAP_REL(p) (__atomic_store_n(p, 0, __ATOMIC_RELEASE)) + +struct cap_est { + uint64_t c_pkt; /* total packets enqueued (relaxed) */ + uint64_t c_byt; /* total bytes enqueued (relaxed) */ + uint64_t c_idl; /* times ring seen empty (relaxed) */ + + uint64_t t_gate; /* last fold timestamp (ns) */ + uint8_t busy; /* fold in progress (try-lock) */ + + uint64_t t0; /* window start (ns), 0 = no window */ + uint64_t q0; /* ring occupancy at window start */ + uint64_t pkt0; /* c_pkt snapshot at window start */ + uint64_t byt0; /* c_byt snapshot at window start */ + uint64_t idl0; /* c_idl snapshot at window start */ + uint64_t rate; /* filtered drain rate (bytes/s) */ + + uint8_t cap; /* published capacity code (0=none) */ +}; + +struct { + struct cap_est est[PROC_MAX_FLOWS]; +} cap; + +int cap_init(void) +{ + memset(&cap, 0, sizeof(cap)); + + return 0; +} + +void cap_fini(void) +{ +} + +void cap_reset(int fd) +{ + /* A racing update seeds one bogus window; the filter absorbs. */ + memset(&cap.est[fd], 0, sizeof(cap.est[fd])); +} + +/* + * Rate <-> 8-bit code (cap_enc / cap_dec). The code is a tiny float: + * the high 6 bits are a band e = floor(log2 rate), the low 2 bits a + * quarter k that splits each band [2^e, 2^(e+1)) into four, so + * code = 4 * e + k. Each step is ~19% in rate; that coarseness is + * deliberate, capacity only needs order-of-magnitude accuracy. + * + * The quarter cut points are 256 * 2^(k/4) rounded to an integer: + * {256, 304, 362, 431} over the normalized range [256, 512). + */ +uint8_t cap_enc(uint64_t rate) +{ + static const uint16_t thr[3] = {304, 362, 431}; + uint64_t r = rate; /* copy halved to find band */ + unsigned e = 0; /* band: floor log2 rate */ + unsigned k = 0; /* quarter within band 0..3 */ + unsigned c; /* code = 4 * band + quarter */ + uint16_t top; /* rate scaled to [256, 512) */ + + if (rate == 0) + return 0; + + while (r > 1) { + r >>= 1; + e++; + } + + /* Top 9 bits: rate normalized to [256, 512). */ + top = e >= 8 ? (uint16_t) (rate >> (e - 8)) + : (uint16_t) (rate << (8 - e)); + + while (k < 3 && top >= thr[k]) + k++; + + c = 4 * e + k; + if (c == 0) + c = 1; /* 0 means unknown */ + + return (uint8_t) c; +} + +uint64_t cap_dec(uint8_t c) +{ + static const uint16_t m[4] = {256, 304, 362, 431}; + unsigned e = c >> 2; /* band = c >> 2 */ + unsigned k = c & 3; /* quarter = c & 3 */ + + if (c == 0) + return 0; + + if (e >= 8) + return (uint64_t) m[k] << (e - 8); + + return ((uint64_t) m[k] << e) >> 8; +} + +uint8_t cap_min(uint8_t a, + uint8_t b) +{ + if (a == 0) + return b; + + if (b == 0) + return a; + + return a < b ? a : b; +} + +void cap_stamp(uint8_t * pci, + uint8_t own) +{ + if (own == 0) + return; + + if (*pci == 0 || own < *pci) + *pci = own; +} + +/* Fold flag held; q1 is the caller's pre-write ring sample. */ +static void cap_fold(struct cap_est * e, + uint64_t q1, + uint64_t now) +{ + uint64_t pkt; /* current c_pkt snapshot */ + uint64_t byt; /* current c_byt snapshot */ + uint64_t idl; /* current c_idl snapshot */ + uint64_t dt; /* window duration (ns) */ + uint64_t enq; /* packets enqueued in window */ + uint64_t avg; /* mean packet size (bytes) */ + uint64_t r; /* window drain rate (bytes/s) */ + int64_t drained; /* slots drained over window */ + + pkt = LOAD_RELAXED(&e->c_pkt); + byt = LOAD_RELAXED(&e->c_byt); + idl = LOAD_RELAXED(&e->c_idl); + + dt = now - e->t0; + enq = pkt - e->pkt0; + + drained = (int64_t) (e->q0 + enq - q1); + + if (e->t0 == 0 || dt > CAP_T_MAX || enq == 0) + goto reopen; + + if (drained < (int64_t) CAP_N_MIN) + return; /* extend the window until enough drains */ + + if ((idl - e->idl0) << CAP_IDL_SHFT > enq) + goto reopen; /* mostly idle ring: not saturated */ + + avg = (byt - e->byt0) / enq; + r = (uint64_t) drained * avg * MILLION / (dt / 1000); + + if (r >= e->rate) { + /* Empty-edged windows drain into buffers below. */ + if (e->q0 > 0 && q1 > 0) + e->rate = r; + } else { + e->rate -= (e->rate - r) >> CAP_DEC_SHFT; + } + + STORE_RELAXED(&e->cap, cap_enc(e->rate)); + reopen: + e->t0 = now; + e->q0 = q1; + e->pkt0 = pkt; + e->byt0 = byt; + e->idl0 = idl; +} + +/* Internal, timestamped entry point; tests drive this directly. */ +static void cap_update_at(int fd, + size_t qlen, + size_t len, + uint64_t now) +{ + struct cap_est * e = &cap.est[fd]; /* this flow's estimator */ + + FETCH_ADD_RELAXED(&e->c_pkt, 1); + FETCH_ADD_RELAXED(&e->c_byt, len); + + if (qlen == 0) + FETCH_ADD_RELAXED(&e->c_idl, 1); + + if (now - LOAD_RELAXED(&e->t_gate) < CAP_T_MIN) + return; + + if (!CAP_TRY(&e->busy)) + return; + + if (now - e->t_gate >= CAP_T_MIN) { + cap_fold(e, qlen, now); + STORE_RELAXED(&e->t_gate, now); + } + + CAP_REL(&e->busy); +} + +void cap_update(int fd, + size_t qlen, + size_t len) +{ + struct timespec now; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + cap_update_at(fd, qlen, len, TS_TO_UINT64(now)); +} + +uint8_t cap_get(int fd) +{ + return LOAD_RELAXED(&cap.est[fd].cap); +} diff --git a/src/ipcpd/unicast/cap.h b/src/ipcpd/unicast/cap.h new file mode 100644 index 00000000..df8c2ec1 --- /dev/null +++ b/src/ipcpd/unicast/cap.h @@ -0,0 +1,54 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Link capacity estimation + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IPCPD_UNICAST_CAP_H +#define OUROBOROS_IPCPD_UNICAST_CAP_H + +#include <stddef.h> +#include <stdint.h> + +int cap_init(void); + +void cap_fini(void); + +/* Account an egress packet; qlen is sampled before the write. */ +void cap_update(int fd, + size_t qlen, + size_t len); + +uint8_t cap_get(int fd); + +void cap_reset(int fd); + +/* Quarter-log2 capacity code: ~2^(c / 4) bytes/s, 0 = unknown. */ +uint8_t cap_enc(uint64_t rate); + +uint64_t cap_dec(uint8_t c); + +uint8_t cap_min(uint8_t a, + uint8_t b); + +/* MIN-combine the own link code into the PCI byte. */ +void cap_stamp(uint8_t * pci, + uint8_t own); + +#endif /* OUROBOROS_IPCPD_UNICAST_CAP_H */ diff --git a/src/ipcpd/unicast/comp.h b/src/ipcpd/unicast/comp.h deleted file mode 100644 index 42367833..00000000 --- a/src/ipcpd/unicast/comp.h +++ /dev/null @@ -1,48 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Components for the unicast IPC process - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#ifndef OUROBOROS_IPCPD_UNICAST_COMP_H -#define OUROBOROS_IPCPD_UNICAST_COMP_H - -#include <ouroboros/cacep.h> - -#include "dt.h" - -#define DST_MAX_STRLEN 64 - -enum comp_id { - COMPID_DT = 0, - COMPID_ENROLL, - COMPID_MGMT, - COMPID_MAX -}; - -struct conn { - struct conn_info conn_info; - struct { - char dst[DST_MAX_STRLEN + 1]; - int fd; - qosspec_t qs; - } flow_info; -}; - -#endif /* OUROBOROS_IPCPD_UNICAST_COMP_H */ diff --git a/src/ipcpd/unicast/connmgr.c b/src/ipcpd/unicast/connmgr.c index 3ebef7f9..f60f6fca 100644 --- a/src/ipcpd/unicast/connmgr.c +++ b/src/ipcpd/unicast/connmgr.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Handles connections between components * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,506 +20,16 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ +#include "config.h" + #if defined(__linux__) || defined(__CYGWIN__) #define _DEFAULT_SOURCE #else #define _POSIX_C_SOURCE 200112L #endif -#define OUROBOROS_PREFIX "connection-manager" - -#include <ouroboros/dev.h> -#include <ouroboros/cacep.h> -#include <ouroboros/errno.h> -#include <ouroboros/list.h> -#include <ouroboros/logs.h> -#include <ouroboros/notifier.h> - -#include "comp.h" -#include "connmgr.h" -#include "dir.h" -#include "enroll.h" -#include "ipcp.h" - -#include <pthread.h> -#include <string.h> -#include <stdlib.h> -#include <assert.h> - -enum connmgr_state { - CONNMGR_NULL = 0, - CONNMGR_INIT, - CONNMGR_RUNNING -}; - -struct conn_el { - struct list_head next; - struct conn conn; -}; - -struct comp { - struct conn_info info; - - struct list_head conns; - struct list_head pending; - - pthread_cond_t cond; - pthread_mutex_t lock; -}; - -struct { - struct comp comps[COMPID_MAX]; - enum connmgr_state state; - - pthread_t acceptor; -} connmgr; - -static int get_id_by_name(const char * name) -{ - enum comp_id i; - - for (i = 0; i < COMPID_MAX; ++i) - if (strcmp(name, connmgr.comps[i].info.comp_name) == 0) - return i; - - return -1; -} - -static int get_conn_by_fd(int fd, - enum comp_id id, - struct conn * conn) -{ - struct list_head * p; - - pthread_mutex_lock(&connmgr.comps[id].lock); - - list_for_each(p, &connmgr.comps[id].conns) { - struct conn_el * c = - list_entry(p, struct conn_el, next); - if (c->conn.flow_info.fd == fd) { - *conn = c->conn; - pthread_mutex_unlock(&connmgr.comps[id].lock); - return 0; - } - } - - pthread_mutex_unlock(&connmgr.comps[id].lock); - - return -1; -} - -static int add_comp_conn(enum comp_id id, - int fd, - qosspec_t qs, - struct conn_info * rcv_info) -{ - struct conn_el * el; - - el = malloc(sizeof(*el)); - if (el == NULL) { - log_err("Not enough memory."); - return -1; - } - - el->conn.conn_info = *rcv_info; - el->conn.flow_info.fd = fd; - el->conn.flow_info.qs = qs; - - pthread_mutex_lock(&connmgr.comps[id].lock); - - list_add(&el->next, &connmgr.comps[id].pending); - pthread_cond_signal(&connmgr.comps[id].cond); - - pthread_mutex_unlock(&connmgr.comps[id].lock); - - return 0; -} - -static void * flow_acceptor(void * o) -{ - int fd; - qosspec_t qs; - struct conn_info rcv_info; - struct conn_info fail_info; - - (void) o; - - memset(&fail_info, 0, sizeof(fail_info)); - - while (true) { - int id; - - fd = flow_accept(&qs, NULL); - if (fd < 0) { - if (fd != -EIRMD) - log_warn("Flow accept failed: %d", fd); - continue; - } - - if (cacep_rcv(fd, &rcv_info)) { - log_dbg("Error establishing application connection."); - flow_dealloc(fd); - continue; - } - - id = get_id_by_name(rcv_info.comp_name); - if (id < 0) { - log_dbg("Connection request for unknown component %s.", - rcv_info.comp_name); - cacep_snd(fd, &fail_info); - flow_dealloc(fd); - continue; - } - - assert(id < COMPID_MAX); - - if (cacep_snd(fd, &connmgr.comps[id].info)) { - log_dbg("Failed to respond to request."); - flow_dealloc(fd); - continue; - } - - if (add_comp_conn(id, fd, qs, &rcv_info)) { - log_dbg("Failed to add new connection."); - flow_dealloc(fd); - continue; - } - } - - return (void *) 0; -} - -static void handle_event(void * self, - int event, - const void * o) -{ - struct conn conn; - - (void) self; - - if (!(event == NOTIFY_DT_FLOW_UP || - event == NOTIFY_DT_FLOW_DOWN || - event == NOTIFY_DT_FLOW_DEALLOC)) - return; - - if (get_conn_by_fd(*((int *) o), COMPID_DT, &conn)) - return; - - switch (event) { - case NOTIFY_DT_FLOW_UP: - notifier_event(NOTIFY_DT_CONN_UP, &conn); - break; - case NOTIFY_DT_FLOW_DOWN: - notifier_event(NOTIFY_DT_CONN_DOWN, &conn); - break; - case NOTIFY_DT_FLOW_DEALLOC: - notifier_event(NOTIFY_DT_CONN_DEL, &conn); - break; - default: - break; - } -} - -int connmgr_init(void) -{ - connmgr.state = CONNMGR_INIT; - - if (notifier_reg(handle_event, NULL)) - return -1; - - return 0; -} - -void connmgr_fini(void) -{ - int i; - - notifier_unreg(handle_event); - - if (connmgr.state == CONNMGR_RUNNING) - pthread_join(connmgr.acceptor, NULL); - - for (i = 0; i < COMPID_MAX; ++i) - connmgr_comp_fini(i); -} - -int connmgr_start(void) -{ - if (pthread_create(&connmgr.acceptor, NULL, flow_acceptor, NULL)) - return -1; - - connmgr.state = CONNMGR_RUNNING; - - return 0; -} - -void connmgr_stop(void) -{ - if (connmgr.state == CONNMGR_RUNNING) - pthread_cancel(connmgr.acceptor); -} - -int connmgr_comp_init(enum comp_id id, - const struct conn_info * info) -{ - struct comp * comp; - - assert(id >= 0 && id < COMPID_MAX); - - comp = connmgr.comps + id; - - if (pthread_mutex_init(&comp->lock, NULL)) - return -1; - - if (pthread_cond_init(&comp->cond, NULL)) { - pthread_mutex_destroy(&comp->lock); - return -1; - } - - list_head_init(&comp->conns); - list_head_init(&comp->pending); - - memcpy(&connmgr.comps[id].info, info, sizeof(connmgr.comps[id].info)); - - return 0; -} - -void connmgr_comp_fini(enum comp_id id) -{ - struct list_head * p; - struct list_head * h; - struct comp * comp; - - assert(id >= 0 && id < COMPID_MAX); - - if (strlen(connmgr.comps[id].info.comp_name) == 0) - return; - - comp = connmgr.comps + id; - - pthread_mutex_lock(&comp->lock); - - list_for_each_safe(p, h, &comp->conns) { - struct conn_el * e = list_entry(p, struct conn_el, next); - list_del(&e->next); - free(e); - } - - list_for_each_safe(p, h, &comp->pending) { - struct conn_el * e = list_entry(p, struct conn_el, next); - list_del(&e->next); - free(e); - } - - pthread_mutex_unlock(&comp->lock); - - pthread_cond_destroy(&comp->cond); - pthread_mutex_destroy(&comp->lock); - - memset(&connmgr.comps[id].info, 0, sizeof(connmgr.comps[id].info)); -} - -int connmgr_ipcp_connect(const char * dst, - const char * component, - qosspec_t qs) -{ - struct conn_el * ce; - int id; - - assert(dst); - assert(component); - - ce = malloc(sizeof(*ce)); - if (ce == NULL) { - log_dbg("Out of memory."); - return -1; - } - - id = get_id_by_name(component); - if (id < 0) { - log_dbg("No such component: %s", component); - free(ce); - return -1; - } - - if (connmgr_alloc(id, dst, &qs, &ce->conn)) { - free(ce); - return -1; - } - - if (strlen(dst) > DST_MAX_STRLEN) { - log_warn("Truncating dst length for connection."); - memcpy(ce->conn.flow_info.dst, dst, DST_MAX_STRLEN); - ce->conn.flow_info.dst[DST_MAX_STRLEN] = '\0'; - } else { - strcpy(ce->conn.flow_info.dst, dst); - } - - pthread_mutex_lock(&connmgr.comps[id].lock); - - list_add(&ce->next, &connmgr.comps[id].conns); - - pthread_mutex_unlock(&connmgr.comps[id].lock); - - return 0; -} - -int connmgr_ipcp_disconnect(const char * dst, - const char * component) -{ - struct list_head * p; - struct list_head * h; - int id; - - assert(dst); - assert(component); - - id = get_id_by_name(component); - if (id < 0) - return -1; - - pthread_mutex_lock(&connmgr.comps[id].lock); - - list_for_each_safe(p,h, &connmgr.comps[id].conns) { - struct conn_el * el = list_entry(p, struct conn_el, next); - if (strcmp(el->conn.flow_info.dst, dst) == 0) { - int ret; - pthread_mutex_unlock(&connmgr.comps[id].lock); - list_del(&el->next); - ret = connmgr_dealloc(id, &el->conn); - free(el); - return ret; - } - } - - pthread_mutex_unlock(&connmgr.comps[id].lock); - - return 0; -} - -int connmgr_alloc(enum comp_id id, - const char * dst, - qosspec_t * qs, - struct conn * conn) -{ - assert(id >= 0 && id < COMPID_MAX); - assert(dst); - - conn->flow_info.fd = flow_alloc(dst, qs, NULL); - if (conn->flow_info.fd < 0) { - log_dbg("Failed to allocate flow to %s.", dst); - return -1; - } - - if (qs != NULL) - conn->flow_info.qs = *qs; - else - memset(&conn->flow_info.qs, 0, sizeof(conn->flow_info.qs)); - - log_dbg("Sending cacep info for protocol %s to fd %d.", - connmgr.comps[id].info.protocol, conn->flow_info.fd); - - if (cacep_snd(conn->flow_info.fd, &connmgr.comps[id].info)) { - log_dbg("Failed to create application connection."); - flow_dealloc(conn->flow_info.fd); - return -1; - } - - if (cacep_rcv(conn->flow_info.fd, &conn->conn_info)) { - log_dbg("Failed to connect to application."); - flow_dealloc(conn->flow_info.fd); - return -1; - } - - if (strcmp(connmgr.comps[id].info.protocol, conn->conn_info.protocol)) { - log_dbg("Unknown protocol (requested %s, got %s).", - connmgr.comps[id].info.protocol, - conn->conn_info.protocol); - flow_dealloc(conn->flow_info.fd); - return -1; - } - - if (connmgr.comps[id].info.pref_version != - conn->conn_info.pref_version) { - log_dbg("Unknown protocol version."); - flow_dealloc(conn->flow_info.fd); - return -1; - } - - if (connmgr.comps[id].info.pref_syntax != conn->conn_info.pref_syntax) { - log_dbg("Unknown protocol syntax."); - flow_dealloc(conn->flow_info.fd); - return -1; - } - - switch (id) { - case COMPID_DT: - notifier_event(NOTIFY_DT_CONN_ADD, conn); -#ifdef IPCP_CONN_WAIT_DIR - dir_wait_running(); -#endif - break; - case COMPID_MGMT: - notifier_event(NOTIFY_MGMT_CONN_ADD, conn); - break; - default: - break; - } - - return 0; -} - -int connmgr_dealloc(enum comp_id id, - struct conn * conn) -{ - switch (id) { - case COMPID_DT: - notifier_event(NOTIFY_DT_CONN_DEL, conn); - break; - case COMPID_MGMT: - notifier_event(NOTIFY_MGMT_CONN_DEL, conn); - break; - default: - break; - } - - return flow_dealloc(conn->flow_info.fd); -} - - -int connmgr_wait(enum comp_id id, - struct conn * conn) -{ - struct conn_el * el; - struct comp * comp; - - assert(id >= 0 && id < COMPID_MAX); - assert(conn); - - comp = connmgr.comps + id; - - pthread_mutex_lock(&comp->lock); - - pthread_cleanup_push((void(*)(void *))pthread_mutex_unlock, - (void *) &comp->lock); - - while (list_is_empty(&comp->pending)) - pthread_cond_wait(&comp->cond, &comp->lock); - - pthread_cleanup_pop(false); - - el = list_first_entry((&comp->pending), struct conn_el, next); - if (el == NULL) { - pthread_mutex_unlock(&comp->lock); - return -1; - } - - *conn = el->conn; - - list_del(&el->next); - list_add(&el->next, &connmgr.comps[id].conns); +#include <ouroboros/ipcp.h> - pthread_mutex_unlock(&comp->lock); +#define BUILD_IPCP_UNICAST - return 0; -} +#include "common/connmgr.c" diff --git a/src/ipcpd/unicast/connmgr.h b/src/ipcpd/unicast/connmgr.h deleted file mode 100644 index 17f78245..00000000 --- a/src/ipcpd/unicast/connmgr.h +++ /dev/null @@ -1,74 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Handles the different AP connections - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#ifndef OUROBOROS_IPCPD_UNICAST_CONNMGR_H -#define OUROBOROS_IPCPD_UNICAST_CONNMGR_H - -#include <ouroboros/cacep.h> -#include <ouroboros/qos.h> - -#include "comp.h" - -#define NOTIFY_DT_CONN_ADD 0x00D0 -#define NOTIFY_DT_CONN_DEL 0x00D1 -#define NOTIFY_DT_CONN_QOS 0x00D2 -#define NOTIFY_DT_CONN_UP 0x00D3 -#define NOTIFY_DT_CONN_DOWN 0x00D4 -#define NOTIFY_DT_FLOW_UP 0x00D5 -#define NOTIFY_DT_FLOW_DOWN 0x00D6 -#define NOTIFY_DT_FLOW_DEALLOC 0x00D7 - -#define NOTIFY_MGMT_CONN_ADD 0x00F0 -#define NOTIFY_MGMT_CONN_DEL 0x00F1 - -int connmgr_init(void); - -void connmgr_fini(void); - -int connmgr_start(void); - -void connmgr_stop(void); - -int connmgr_comp_init(enum comp_id id, - const struct conn_info * info); - -void connmgr_comp_fini(enum comp_id id); - -int connmgr_ipcp_connect(const char * dst, - const char * component, - qosspec_t qs); - -int connmgr_ipcp_disconnect(const char * dst, - const char * component); - -int connmgr_alloc(enum comp_id id, - const char * dst, - qosspec_t * qs, - struct conn * conn); - -int connmgr_dealloc(enum comp_id id, - struct conn * conn); - -int connmgr_wait(enum comp_id id, - struct conn * conn); - -#endif /* OUROBOROS_IPCPD_UNICAST_CONNMGR_H */ diff --git a/src/ipcpd/unicast/dht.c b/src/ipcpd/unicast/dht.c deleted file mode 100644 index b048580d..00000000 --- a/src/ipcpd/unicast/dht.c +++ /dev/null @@ -1,2846 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Distributed Hash Table based on Kademlia - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#if defined(__linux__) || defined(__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 200112L -#endif - -#include "config.h" - -#define DHT "dht" -#define OUROBOROS_PREFIX DHT - -#include <ouroboros/hash.h> -#include <ouroboros/ipcp-dev.h> -#include <ouroboros/bitmap.h> -#include <ouroboros/errno.h> -#include <ouroboros/logs.h> -#include <ouroboros/list.h> -#include <ouroboros/notifier.h> -#include <ouroboros/random.h> -#include <ouroboros/time_utils.h> -#include <ouroboros/tpm.h> -#include <ouroboros/utils.h> - -#include "connmgr.h" -#include "dht.h" -#include "dt.h" - -#include <pthread.h> -#include <stdlib.h> -#include <string.h> -#include <assert.h> -#include <inttypes.h> -#include <limits.h> - -#include "kademlia.pb-c.h" -typedef KadMsg kad_msg_t; -typedef KadContactMsg kad_contact_msg_t; - -#ifndef CLOCK_REALTIME_COARSE -#define CLOCK_REALTIME_COARSE CLOCK_REALTIME -#endif - -#define DHT_MAX_REQS 2048 /* KAD recommends rnd(), bmp can be changed. */ -#define KAD_ALPHA 3 /* Parallel factor, proven optimal value. */ -#define KAD_K 8 /* Replication factor, MDHT value. */ -#define KAD_T_REPL 900 /* Replication time, tied to k. MDHT value. */ -#define KAD_T_REFR 900 /* Refresh time stale bucket, MDHT value. */ -#define KAD_T_JOIN 8 /* Response time to wait for a join. */ -#define KAD_T_RESP 5 /* Response time to wait for a response. */ -#define KAD_R_PING 2 /* Ping retries before declaring peer dead. */ -#define KAD_QUEER 15 /* Time to declare peer questionable. */ -#define KAD_BETA 8 /* Bucket split factor, must be 1, 2, 4 or 8. */ -#define KAD_RESP_RETR 6 /* Number of retries on sending a response. */ -#define KAD_JOIN_RETR 8 /* Number of retries sending a join. */ -#define KAD_JOIN_INTV 1 /* Time (seconds) between join retries. */ -#define HANDLE_TIMEO 1000 /* Timeout for dht_handle_packet tpm check (ms) */ -#define DHT_RETR_ADDR 1 /* Number of addresses to return on retrieve */ - -enum dht_state { - DHT_INIT = 0, - DHT_SHUTDOWN, - DHT_JOINING, - DHT_RUNNING, -}; - -enum kad_code { - KAD_JOIN = 0, - KAD_FIND_NODE, - KAD_FIND_VALUE, - /* Messages without a response below. */ - KAD_STORE, - KAD_RESPONSE -}; - -enum kad_req_state { - REQ_NULL = 0, - REQ_INIT, - REQ_PENDING, - REQ_RESPONSE, - REQ_DONE, - REQ_DESTROY -}; - -enum lookup_state { - LU_NULL = 0, - LU_INIT, - LU_PENDING, - LU_UPDATE, - LU_COMPLETE, - LU_DESTROY -}; - -struct kad_req { - struct list_head next; - - uint32_t cookie; - enum kad_code code; - uint8_t * key; - uint64_t addr; - - enum kad_req_state state; - pthread_cond_t cond; - pthread_mutex_t lock; - - time_t t_exp; -}; - -struct cookie_el { - struct list_head next; - - uint32_t cookie; -}; - -struct lookup { - struct list_head next; - - struct list_head cookies; - - uint8_t * key; - - struct list_head contacts; - size_t n_contacts; - - uint64_t * addrs; - size_t n_addrs; - - enum lookup_state state; - pthread_cond_t cond; - pthread_mutex_t lock; -}; - -struct val { - struct list_head next; - - uint64_t addr; - - time_t t_exp; - time_t t_rep; -}; - -struct ref_entry { - struct list_head next; - - uint8_t * key; - - time_t t_rep; -}; - -struct dht_entry { - struct list_head next; - - uint8_t * key; - size_t n_vals; - struct list_head vals; -}; - -struct contact { - struct list_head next; - - uint8_t * id; - uint64_t addr; - - size_t fails; - time_t t_seen; -}; - -struct bucket { - struct list_head contacts; - size_t n_contacts; - - struct list_head alts; - size_t n_alts; - - time_t t_refr; - - size_t depth; - uint8_t mask; - - struct bucket * parent; - struct bucket * children[1L << KAD_BETA]; -}; - -struct cmd { - struct list_head next; - - struct shm_du_buff * sdb; -}; - -struct dht { - size_t alpha; - size_t b; - size_t k; - - time_t t_expire; - time_t t_refresh; - time_t t_replic; - time_t t_repub; - - uint8_t * id; - uint64_t addr; - - struct bucket * buckets; - - struct list_head entries; - - struct list_head refs; - - struct list_head lookups; - - struct list_head requests; - struct bmp * cookies; - - enum dht_state state; - struct list_head cmds; - pthread_cond_t cond; - pthread_mutex_t mtx; - - pthread_rwlock_t lock; - - uint64_t eid; - - struct tpm * tpm; - - pthread_t worker; -}; - -struct join_info { - struct dht * dht; - uint64_t addr; -}; - -struct packet_info { - struct dht * dht; - struct shm_du_buff * sdb; -}; - -static uint8_t * dht_dup_key(const uint8_t * key, - size_t len) -{ - uint8_t * dup; - - dup = malloc(sizeof(*dup) * len); - if (dup == NULL) - return NULL; - - memcpy(dup, key, len); - - return dup; -} - -static enum dht_state dht_get_state(struct dht * dht) -{ - enum dht_state state; - - pthread_mutex_lock(&dht->mtx); - - state = dht->state; - - pthread_mutex_unlock(&dht->mtx); - - return state; -} - -static int dht_set_state(struct dht * dht, - enum dht_state state) -{ - pthread_mutex_lock(&dht->mtx); - - if (state == DHT_JOINING && dht->state != DHT_INIT) { - pthread_mutex_unlock(&dht->mtx); - return -1; - } - - dht->state = state; - - pthread_cond_broadcast(&dht->cond); - - pthread_mutex_unlock(&dht->mtx); - - return 0; -} - -int dht_wait_running(struct dht * dht) -{ - int ret = 0; - - pthread_mutex_lock(&dht->mtx); - - pthread_cleanup_push((void *)(void *) pthread_mutex_unlock, - &dht->mtx); - - while (dht->state == DHT_JOINING) - pthread_cond_wait(&dht->cond, &dht->mtx); - - if (dht->state != DHT_RUNNING) - ret = -1; - - pthread_cleanup_pop(true); - - return ret; -} - -static uint8_t * create_id(size_t len) -{ - uint8_t * id; - - id = malloc(len); - if (id == NULL) - return NULL; - - if (random_buffer(id, len) < 0) { - free(id); - return NULL; - } - - return id; -} - -static void kad_req_create(struct dht * dht, - kad_msg_t * msg, - uint64_t addr) -{ - struct kad_req * req; - pthread_condattr_t cattr; - struct timespec t; - size_t b; - - req = malloc(sizeof(*req)); - if (req == NULL) - return; - - list_head_init(&req->next); - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - req->t_exp = t.tv_sec + KAD_T_RESP; - req->addr = addr; - req->state = REQ_INIT; - req->cookie = msg->cookie; - req->code = msg->code; - req->key = NULL; - - pthread_rwlock_rdlock(&dht->lock); - b = dht->b; - pthread_rwlock_unlock(&dht->lock); - - if (msg->has_key) { - req->key = dht_dup_key(msg->key.data, b); - if (req->key == NULL) { - free(req); - return; - } - } - - if (pthread_mutex_init(&req->lock, NULL)) { - free(req->key); - free(req); - return; - } - - pthread_condattr_init(&cattr); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - - if (pthread_cond_init(&req->cond, &cattr)) { - pthread_condattr_destroy(&cattr); - pthread_mutex_destroy(&req->lock); - free(req->key); - free(req); - return; - } - - pthread_condattr_destroy(&cattr); - - pthread_rwlock_wrlock(&dht->lock); - - list_add(&req->next, &dht->requests); - - pthread_rwlock_unlock(&dht->lock); -} - -static void cancel_req_destroy(void * o) -{ - struct kad_req * req = (struct kad_req *) o; - - pthread_mutex_unlock(&req->lock); - - pthread_cond_destroy(&req->cond); - pthread_mutex_destroy(&req->lock); - - if (req->key != NULL) - free(req->key); - - free(req); -} - -static void kad_req_destroy(struct kad_req * req) -{ - assert(req); - - pthread_mutex_lock(&req->lock); - - switch (req->state) { - case REQ_DESTROY: - pthread_mutex_unlock(&req->lock); - return; - case REQ_PENDING: - req->state = REQ_DESTROY; - pthread_cond_signal(&req->cond); - break; - case REQ_INIT: - case REQ_DONE: - req->state = REQ_NULL; - break; - case REQ_RESPONSE: - case REQ_NULL: - default: - break; - } - - pthread_cleanup_push(cancel_req_destroy, req); - - while (req->state != REQ_NULL && req->state != REQ_DONE) - pthread_cond_wait(&req->cond, &req->lock); - - pthread_cleanup_pop(true); -} - -static int kad_req_wait(struct kad_req * req, - time_t t) -{ - struct timespec timeo = {t, 0}; - struct timespec abs; - int ret = 0; - - assert(req); - - clock_gettime(PTHREAD_COND_CLOCK, &abs); - - ts_add(&abs, &timeo, &abs); - - pthread_mutex_lock(&req->lock); - - req->state = REQ_PENDING; - - pthread_cleanup_push((void *)(void *) pthread_mutex_unlock, - &req->lock); - - while (req->state == REQ_PENDING && ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(&req->cond, &req->lock, &abs); - - switch(req->state) { - case REQ_DESTROY: - ret = -1; - req->state = REQ_NULL; - pthread_cond_signal(&req->cond); - break; - case REQ_PENDING: /* ETIMEDOUT */ - case REQ_RESPONSE: - req->state = REQ_DONE; - pthread_cond_broadcast(&req->cond); - break; - default: - break; - } - - pthread_cleanup_pop(true); - - return ret; -} - -static void kad_req_respond(struct kad_req * req) -{ - pthread_mutex_lock(&req->lock); - - req->state = REQ_RESPONSE; - pthread_cond_signal(&req->cond); - - pthread_mutex_unlock(&req->lock); -} - -static struct contact * contact_create(const uint8_t * id, - size_t len, - uint64_t addr) -{ - struct contact * c; - struct timespec t; - - c = malloc(sizeof(*c)); - if (c == NULL) - return NULL; - - list_head_init(&c->next); - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - c->addr = addr; - c->fails = 0; - c->t_seen = t.tv_sec; - c->id = dht_dup_key(id, len); - if (c->id == NULL) { - free(c); - return NULL; - } - - return c; -} - -static void contact_destroy(struct contact * c) -{ - if (c != NULL) - free(c->id); - - free(c); -} - -static struct bucket * iter_bucket(struct bucket * b, - const uint8_t * id) -{ - uint8_t byte; - uint8_t mask; - - assert(b); - - if (b->children[0] == NULL) - return b; - - byte = id[(b->depth * KAD_BETA) / CHAR_BIT]; - - mask = ((1L << KAD_BETA) - 1) & 0xFF; - - byte >>= (CHAR_BIT - KAD_BETA) - - (((b->depth) * KAD_BETA) & (CHAR_BIT - 1)); - - return iter_bucket(b->children[(byte & mask)], id); -} - -static struct bucket * dht_get_bucket(struct dht * dht, - const uint8_t * id) -{ - assert(dht->buckets); - - return iter_bucket(dht->buckets, id); -} - -/* - * If someone builds a network where the n (n > k) closest nodes all - * have IDs starting with the same 64 bits: by all means, change this. - */ -static uint64_t dist(const uint8_t * src, - const uint8_t * dst) -{ - return betoh64(*((uint64_t *) src) ^ *((uint64_t *) dst)); -} - -static size_t list_add_sorted(struct list_head * l, - struct contact * c, - const uint8_t * key) -{ - struct list_head * p; - - assert(l); - assert(c); - assert(key); - assert(c->id); - - list_for_each(p, l) { - struct contact * e = list_entry(p, struct contact, next); - if (dist(c->id, key) > dist(e->id, key)) - break; - } - - list_add_tail(&c->next, p); - - return 1; -} - -static size_t dht_contact_list(struct dht * dht, - struct list_head * l, - const uint8_t * key) -{ - struct list_head * p; - struct bucket * b; - size_t len = 0; - size_t i; - struct timespec t; - - assert(l); - assert(dht); - assert(key); - assert(list_is_empty(l)); - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - b = dht_get_bucket(dht, key); - if (b == NULL) - return 0; - - b->t_refr = t.tv_sec + KAD_T_REFR; - - if (b->n_contacts == dht->k || b->parent == NULL) { - list_for_each(p, &b->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - c = contact_create(c->id, dht->b, c->addr); - if (list_add_sorted(l, c, key) == 1) - if (++len == dht->k) - break; - } - } else { - struct bucket * d = b->parent; - for (i = 0; i < (1L << KAD_BETA) && len < dht->k; ++i) { - list_for_each(p, &d->children[i]->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - c = contact_create(c->id, dht->b, c->addr); - if (c == NULL) - continue; - if (list_add_sorted(l, c, key) == 1) - if (++len == dht->k) - break; - } - } - } - - assert(len == dht->k || b->parent == NULL); - - return len; -} - -static struct lookup * lookup_create(struct dht * dht, - const uint8_t * id) -{ - struct lookup * lu; - pthread_condattr_t cattr; - - assert(dht); - assert(id); - - lu = malloc(sizeof(*lu)); - if (lu == NULL) - goto fail_malloc; - - list_head_init(&lu->contacts); - list_head_init(&lu->cookies); - - lu->state = LU_INIT; - lu->addrs = NULL; - lu->n_addrs = 0; - lu->key = dht_dup_key(id, dht->b); - if (lu->key == NULL) - goto fail_id; - - if (pthread_mutex_init(&lu->lock, NULL)) - goto fail_mutex; - - pthread_condattr_init(&cattr); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - - if (pthread_cond_init(&lu->cond, &cattr)) - goto fail_cond; - - pthread_condattr_destroy(&cattr); - - pthread_rwlock_wrlock(&dht->lock); - - list_add(&lu->next, &dht->lookups); - - lu->n_contacts = dht_contact_list(dht, &lu->contacts, id); - - pthread_rwlock_unlock(&dht->lock); - - return lu; - - fail_cond: - pthread_condattr_destroy(&cattr); - pthread_mutex_destroy(&lu->lock); - fail_mutex: - free(lu->key); - fail_id: - free(lu); - fail_malloc: - return NULL; -} - -static void cancel_lookup_destroy(void * o) -{ - struct lookup * lu; - struct list_head * p; - struct list_head * h; - - lu = (struct lookup *) o; - - if (lu->key != NULL) - free(lu->key); - if (lu->addrs != NULL) - free(lu->addrs); - - list_for_each_safe(p, h, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); - contact_destroy(c); - } - - list_for_each_safe(p, h, &lu->cookies) { - struct cookie_el * c = list_entry(p, struct cookie_el, next); - list_del(&c->next); - free(c); - } - - pthread_mutex_unlock(&lu->lock); - - pthread_mutex_destroy(&lu->lock); - - free(lu); -} - -static void lookup_destroy(struct lookup * lu) -{ - assert(lu); - - pthread_mutex_lock(&lu->lock); - - switch (lu->state) { - case LU_DESTROY: - pthread_mutex_unlock(&lu->lock); - return; - case LU_PENDING: - lu->state = LU_DESTROY; - pthread_cond_broadcast(&lu->cond); - break; - case LU_INIT: - case LU_UPDATE: - case LU_COMPLETE: - lu->state = LU_NULL; - break; - case LU_NULL: - default: - break; - } - - pthread_cleanup_push(cancel_lookup_destroy, lu); - - while (lu->state != LU_NULL) - pthread_cond_wait(&lu->cond, &lu->lock); - - pthread_cleanup_pop(true); -} - -static void lookup_update(struct dht * dht, - struct lookup * lu, - kad_msg_t * msg) -{ - struct list_head * p = NULL; - struct list_head * h; - struct contact * c = NULL; - size_t n; - size_t pos = 0; - bool mod = false; - - assert(lu); - assert(msg); - - if (dht_get_state(dht) != DHT_RUNNING) - return; - - pthread_mutex_lock(&lu->lock); - - list_for_each_safe(p, h, &lu->cookies) { - struct cookie_el * e = list_entry(p, struct cookie_el, next); - if (e->cookie == msg->cookie) { - list_del(&e->next); - free(e); - break; - } - } - - if (lu->state == LU_COMPLETE) { - pthread_mutex_unlock(&lu->lock); - return; - } - - if (msg->n_addrs > 0) { - if (lu->addrs == NULL) { - lu->addrs = malloc(sizeof(*lu->addrs) * msg->n_addrs); - for (n = 0; n < msg->n_addrs; ++n) - lu->addrs[n] = msg->addrs[n]; - lu->n_addrs = msg->n_addrs; - } - - lu->state = LU_COMPLETE; - pthread_cond_broadcast(&lu->cond); - pthread_mutex_unlock(&lu->lock); - return; - } - - pthread_cleanup_push((void *)(void *) pthread_mutex_unlock, - &lu->lock); - - while (lu->state == LU_INIT) { - pthread_rwlock_unlock(&dht->lock); - pthread_cond_wait(&lu->cond, &lu->lock); - pthread_rwlock_rdlock(&dht->lock); - } - - pthread_cleanup_pop(false); - - for (n = 0; n < msg->n_contacts; ++n) { - c = contact_create(msg->contacts[n]->id.data, - dht->b, msg->contacts[n]->addr); - if (c == NULL) - continue; - - pos = 0; - - list_for_each(p, &lu->contacts) { - struct contact * e; - e = list_entry(p, struct contact, next); - if (!memcmp(e->id, c->id, dht->b)) { - contact_destroy(c); - c = NULL; - break; - } - - if (dist(c->id, lu->key) > dist(e->id, lu->key)) - break; - - pos++; - } - - if (c == NULL) - continue; - - if (lu->n_contacts < dht->k) { - list_add_tail(&c->next, p); - ++lu->n_contacts; - mod = true; - } else if (pos == dht->k) { - contact_destroy(c); - } else { - struct contact * d; - list_add_tail(&c->next, p); - d = list_last_entry(&lu->contacts, - struct contact, next); - list_del(&d->next); - assert(lu->contacts.prv != &d->next); - contact_destroy(d); - mod = true; - } - } - - if (list_is_empty(&lu->cookies) && !mod) - lu->state = LU_COMPLETE; - else - lu->state = LU_UPDATE; - - pthread_cond_broadcast(&lu->cond); - pthread_mutex_unlock(&lu->lock); - return; -} - -static ssize_t lookup_get_addrs(struct lookup * lu, - uint64_t * addrs) -{ - ssize_t n; - - assert(lu); - - pthread_mutex_lock(&lu->lock); - - for (n = 0; (size_t) n < lu->n_addrs; ++n) - addrs[n] = lu->addrs[n]; - - assert((size_t) n == lu->n_addrs); - - pthread_mutex_unlock(&lu->lock); - - return n; -} - -static ssize_t lookup_contact_addrs(struct lookup * lu, - uint64_t * addrs) -{ - struct list_head * p; - ssize_t n = 0; - - assert(lu); - assert(addrs); - - pthread_mutex_lock(&lu->lock); - - list_for_each(p, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - addrs[n] = c->addr; - n++; - } - - pthread_mutex_unlock(&lu->lock); - - return n; -} - -static void lookup_new_addrs(struct lookup * lu, - uint64_t * addrs) -{ - struct list_head * p; - size_t n = 0; - - assert(lu); - assert(addrs); - - pthread_mutex_lock(&lu->lock); - - /* Uses fails to check if the contact has been contacted. */ - list_for_each(p, &lu->contacts) { - struct contact * c = list_entry(p, struct contact, next); - if (c->fails == 0) { - c->fails = 1; - addrs[n] = c->addr; - n++; - } - - if (n == KAD_ALPHA) - break; - } - - assert(n <= KAD_ALPHA); - - addrs[n] = 0; - - pthread_mutex_unlock(&lu->lock); -} - -static void lookup_set_state(struct lookup * lu, - enum lookup_state state) -{ - pthread_mutex_lock(&lu->lock); - - lu->state = state; - pthread_cond_broadcast(&lu->cond); - - pthread_mutex_unlock(&lu->lock); -} - -static void cleanup_wait(void * o) -{ - struct lookup * lu = (struct lookup *) o; - lu->state = LU_NULL; - pthread_mutex_unlock(&lu->lock); - lookup_destroy(lu); -} - -static enum lookup_state lookup_wait(struct lookup * lu) -{ - struct timespec timeo = {KAD_T_RESP, 0}; - struct timespec abs; - enum lookup_state state; - int ret = 0; - - clock_gettime(PTHREAD_COND_CLOCK, &abs); - - ts_add(&abs, &timeo, &abs); - - pthread_mutex_lock(&lu->lock); - - if (lu->state == LU_INIT || lu->state == LU_UPDATE) - lu->state = LU_PENDING; - - pthread_cleanup_push(cleanup_wait, lu); - - while (lu->state == LU_PENDING && ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(&lu->cond, &lu->lock, &abs); - - pthread_cleanup_pop(false); - - if (ret == -ETIMEDOUT) - lu->state = LU_COMPLETE; - - state = lu->state; - - pthread_mutex_unlock(&lu->lock); - - return state; -} - -static struct kad_req * dht_find_request(struct dht * dht, - kad_msg_t * msg) -{ - struct list_head * p; - - assert(dht); - assert(msg); - - list_for_each(p, &dht->requests) { - struct kad_req * r = list_entry(p, struct kad_req, next); - if (r->cookie == msg->cookie) - return r; - } - - return NULL; -} - -static struct lookup * dht_find_lookup(struct dht * dht, - uint32_t cookie) -{ - struct list_head * p; - struct list_head * p2; - struct list_head * h2; - - assert(dht); - assert(cookie > 0); - - list_for_each(p, &dht->lookups) { - struct lookup * l = list_entry(p, struct lookup, next); - pthread_mutex_lock(&l->lock); - list_for_each_safe(p2, h2, &l->cookies) { - struct cookie_el * e; - e = list_entry(p2, struct cookie_el, next); - if (e->cookie == cookie) { - list_del(&e->next); - free(e); - pthread_mutex_unlock(&l->lock); - return l; - } - } - pthread_mutex_unlock(&l->lock); - } - - return NULL; -} - -static struct val * val_create(uint64_t addr, - time_t exp) -{ - struct val * v; - struct timespec t; - - v = malloc(sizeof(*v)); - if (v == NULL) - return NULL; - - list_head_init(&v->next); - v->addr = addr; - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - v->t_exp = t.tv_sec + exp; - v->t_rep = t.tv_sec + KAD_T_REPL; - - return v; -} - -static void val_destroy(struct val * v) -{ - assert(v); - - free(v); -} - -static struct ref_entry * ref_entry_create(struct dht * dht, - const uint8_t * key) -{ - struct ref_entry * e; - struct timespec t; - - assert(dht); - assert(key); - - e = malloc(sizeof(*e)); - if (e == NULL) - return NULL; - - e->key = dht_dup_key(key, dht->b); - if (e->key == NULL) { - free(e); - return NULL; - } - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - e->t_rep = t.tv_sec + dht->t_repub; - - return e; -} - -static void ref_entry_destroy(struct ref_entry * e) -{ - free(e->key); - free(e); -} - -static struct dht_entry * dht_entry_create(struct dht * dht, - const uint8_t * key) -{ - struct dht_entry * e; - - assert(dht); - assert(key); - - e = malloc(sizeof(*e)); - if (e == NULL) - return NULL; - - list_head_init(&e->next); - list_head_init(&e->vals); - - e->n_vals = 0; - - e->key = dht_dup_key(key, dht->b); - if (e->key == NULL) { - free(e); - return NULL; - } - - return e; -} - -static void dht_entry_destroy(struct dht_entry * e) -{ - struct list_head * p; - struct list_head * h; - - assert(e); - - list_for_each_safe(p, h, &e->vals) { - struct val * v = list_entry(p, struct val, next); - list_del(&v->next); - val_destroy(v); - } - - free(e->key); - - free(e); -} - -static int dht_entry_add_addr(struct dht_entry * e, - uint64_t addr, - time_t exp) -{ - struct list_head * p; - struct val * val; - struct timespec t; - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr == addr) { - if (v->t_exp < t.tv_sec + exp) { - v->t_exp = t.tv_sec + exp; - v->t_rep = t.tv_sec + KAD_T_REPL; - } - - return 0; - } - } - - val = val_create(addr, exp); - if (val == NULL) - return -ENOMEM; - - list_add(&val->next, &e->vals); - ++e->n_vals; - - return 0; -} - - -static void dht_entry_del_addr(struct dht_entry * e, - uint64_t addr) -{ - struct list_head * p; - struct list_head * h; - - assert(e); - - list_for_each_safe(p, h, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr == addr) { - list_del(&v->next); - val_destroy(v); - --e->n_vals; - } - } - - if (e->n_vals == 0) { - list_del(&e->next); - dht_entry_destroy(e); - } -} - -static uint64_t dht_entry_get_addr(struct dht * dht, - struct dht_entry * e) -{ - struct list_head * p; - - assert(e); - assert(!list_is_empty(&e->vals)); - - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - if (v->addr != dht->addr) - return v->addr; - } - - return 0; -} - -/* Forward declaration. */ -static struct lookup * kad_lookup(struct dht * dht, - const uint8_t * key, - enum kad_code code); - - -/* Build a refresh list. */ -static void bucket_refresh(struct dht * dht, - struct bucket * b, - time_t t, - struct list_head * r) -{ - size_t i; - - if (*b->children != NULL) - for (i = 0; i < (1L << KAD_BETA); ++i) - bucket_refresh(dht, b->children[i], t, r); - - if (b->n_contacts == 0) - return; - - if (t > b->t_refr) { - struct contact * c; - struct contact * d; - c = list_first_entry(&b->contacts, struct contact, next); - d = contact_create(c->id, dht->b, c->addr); - if (c != NULL) - list_add(&d->next, r); - return; - } -} - - -static struct bucket * bucket_create(void) -{ - struct bucket * b; - struct timespec t; - size_t i; - - b = malloc(sizeof(*b)); - if (b == NULL) - return NULL; - - list_head_init(&b->contacts); - b->n_contacts = 0; - - list_head_init(&b->alts); - b->n_alts = 0; - - clock_gettime(CLOCK_REALTIME_COARSE, &t); - b->t_refr = t.tv_sec + KAD_T_REFR; - - for (i = 0; i < (1L << KAD_BETA); ++i) - b->children[i] = NULL; - - b->parent = NULL; - b->depth = 0; - - return b; -} - -static void bucket_destroy(struct bucket * b) -{ - struct list_head * p; - struct list_head * h; - size_t i; - - assert(b); - - for (i = 0; i < (1L << KAD_BETA); ++i) - if (b->children[i] != NULL) - bucket_destroy(b->children[i]); - - list_for_each_safe(p, h, &b->contacts) { - struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); - contact_destroy(c); - --b->n_contacts; - } - - list_for_each_safe(p, h, &b->alts) { - struct contact * c = list_entry(p, struct contact, next); - list_del(&c->next); - contact_destroy(c); - --b->n_contacts; - } - - free(b); -} - -static bool bucket_has_id(struct bucket * b, - const uint8_t * id) -{ - uint8_t mask; - uint8_t byte; - - if (b->depth == 0) - return true; - - byte = id[(b->depth * KAD_BETA) / CHAR_BIT]; - - mask = ((1L << KAD_BETA) - 1) & 0xFF; - - byte >>= (CHAR_BIT - KAD_BETA) - - (((b->depth - 1) * KAD_BETA) & (CHAR_BIT - 1)); - - return ((byte & mask) == b->mask); -} - -static int split_bucket(struct bucket * b) -{ - struct list_head * p; - struct list_head * h; - uint8_t mask = 0; - size_t i; - size_t c; - - assert(b); - assert(b->n_alts == 0); - assert(b->n_contacts); - assert(b->children[0] == NULL); - - c = b->n_contacts; - - for (i = 0; i < (1L << KAD_BETA); ++i) { - b->children[i] = bucket_create(); - if (b->children[i] == NULL) { - size_t j; - for (j = 0; j < i; ++j) - bucket_destroy(b->children[j]); - return -1; - } - - b->children[i]->depth = b->depth + 1; - b->children[i]->mask = mask; - b->children[i]->parent = b; - - list_for_each_safe(p, h, &b->contacts) { - struct contact * c; - c = list_entry(p, struct contact, next); - if (bucket_has_id(b->children[i], c->id)) { - list_del(&c->next); - --b->n_contacts; - list_add(&c->next, &b->children[i]->contacts); - ++b->children[i]->n_contacts; - } - } - - mask++; - } - - for (i = 0; i < (1L << KAD_BETA); ++i) - if (b->children[i]->n_contacts == c) - split_bucket(b->children[i]); - - return 0; -} - -/* Locked externally to mandate update as (final) part of join transaction. */ -static int dht_update_bucket(struct dht * dht, - const uint8_t * id, - uint64_t addr) -{ - struct list_head * p; - struct list_head * h; - struct bucket * b; - struct contact * c; - - assert(dht); - - b = dht_get_bucket(dht, id); - if (b == NULL) - return -1; - - c = contact_create(id, dht->b, addr); - if (c == NULL) - return -1; - - list_for_each_safe(p, h, &b->contacts) { - struct contact * d = list_entry(p, struct contact, next); - if (d->addr == addr) { - list_del(&d->next); - contact_destroy(d); - --b->n_contacts; - } - } - - if (b->n_contacts == dht->k) { - if (bucket_has_id(b, dht->id)) { - list_add_tail(&c->next, &b->contacts); - ++b->n_contacts; - if (split_bucket(b)) { - list_del(&c->next); - contact_destroy(c); - --b->n_contacts; - } - } else if (b->n_alts == dht->k) { - struct contact * d; - d = list_first_entry(&b->alts, struct contact, next); - list_del(&d->next); - contact_destroy(d); - list_add_tail(&c->next, &b->alts); - } else { - list_add_tail(&c->next, &b->alts); - ++b->n_alts; - } - } else { - list_add_tail(&c->next, &b->contacts); - ++b->n_contacts; - } - - return 0; -} - -static int send_msg(struct dht * dht, - kad_msg_t * msg, - uint64_t addr) -{ -#ifndef __DHT_TEST__ - struct shm_du_buff * sdb; - size_t len; -#endif - int retr = 0; - - if (msg->code == KAD_RESPONSE) - retr = KAD_RESP_RETR; - - pthread_rwlock_wrlock(&dht->lock); - - if (dht->id != NULL) { - msg->has_s_id = true; - msg->s_id.data = dht->id; - msg->s_id.len = dht->b; - } - - msg->s_addr = dht->addr; - - if (msg->code < KAD_STORE) { - msg->cookie = bmp_allocate(dht->cookies); - if (!bmp_is_id_valid(dht->cookies, msg->cookie)) { - pthread_rwlock_unlock(&dht->lock); - goto fail_bmp_alloc; - } - } - - pthread_rwlock_unlock(&dht->lock); - -#ifndef __DHT_TEST__ - len = kad_msg__get_packed_size(msg); - if (len == 0) - goto fail_msg; - - while (true) { - if (ipcp_sdb_reserve(&sdb, len)) - goto fail_msg; - - kad_msg__pack(msg, shm_du_buff_head(sdb)); - - if (dt_write_packet(addr, QOS_CUBE_BE, dht->eid, sdb) == 0) - break; - - ipcp_sdb_release(sdb); - - sleep(1); - - if (--retr < 0) - goto fail_msg; - } - -#else - (void) addr; - (void) retr; -#endif /* __DHT_TEST__ */ - - if (msg->code < KAD_STORE && dht_get_state(dht) != DHT_SHUTDOWN) - kad_req_create(dht, msg, addr); - - return msg->cookie; -#ifndef __DHT_TEST__ - fail_msg: - pthread_rwlock_wrlock(&dht->lock); - bmp_release(dht->cookies, msg->cookie); - pthread_rwlock_unlock(&dht->lock); -#endif /* !__DHT_TEST__ */ - fail_bmp_alloc: - return -1; -} - -static struct dht_entry * dht_find_entry(struct dht * dht, - const uint8_t * key) -{ - struct list_head * p; - - list_for_each(p, &dht->entries) { - struct dht_entry * e = list_entry(p, struct dht_entry, next); - if (!memcmp(key, e->key, dht->b)) - return e; - } - - return NULL; -} - -static int kad_add(struct dht * dht, - const kad_contact_msg_t * contacts, - ssize_t n, - time_t exp) -{ - struct dht_entry * e; - - pthread_rwlock_wrlock(&dht->lock); - - while (n-- > 0) { - if (contacts[n].id.len != dht->b) - log_warn("Bad key length in contact data."); - - e = dht_find_entry(dht, contacts[n].id.data); - if (e != NULL) { - if (dht_entry_add_addr(e, contacts[n].addr, exp)) - goto fail; - } else { - e = dht_entry_create(dht, contacts[n].id.data); - if (e == NULL) - goto fail; - - if (dht_entry_add_addr(e, contacts[n].addr, exp)) { - dht_entry_destroy(e); - goto fail; - } - - list_add(&e->next, &dht->entries); - } - } - - pthread_rwlock_unlock(&dht->lock); - return 0; - - fail: - pthread_rwlock_unlock(&dht->lock); - return -ENOMEM; -} - -static int wait_resp(struct dht * dht, - kad_msg_t * msg, - time_t timeo) -{ - struct kad_req * req; - - assert(dht); - assert(msg); - - pthread_rwlock_rdlock(&dht->lock); - - req = dht_find_request(dht, msg); - if (req == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -EPERM; - } - - pthread_rwlock_unlock(&dht->lock); - - return kad_req_wait(req, timeo); -} - -static int kad_store(struct dht * dht, - const uint8_t * key, - uint64_t addr, - uint64_t r_addr, - time_t ttl) -{ - kad_msg_t msg = KAD_MSG__INIT; - kad_contact_msg_t cmsg = KAD_CONTACT_MSG__INIT; - kad_contact_msg_t * cmsgp[1]; - - cmsg.id.data = (uint8_t *) key; - cmsg.addr = addr; - - pthread_rwlock_rdlock(&dht->lock); - - cmsg.id.len = dht->b; - - pthread_rwlock_unlock(&dht->lock); - - cmsgp[0] = &cmsg; - - msg.code = KAD_STORE; - msg.has_t_expire = true; - msg.t_expire = ttl; - msg.n_contacts = 1; - msg.contacts = cmsgp; - - if (send_msg(dht, &msg, r_addr) < 0) - return -1; - - return 0; -} - -static ssize_t kad_find(struct dht * dht, - struct lookup * lu, - const uint64_t * addrs, - enum kad_code code) -{ - kad_msg_t msg = KAD_MSG__INIT; - ssize_t sent = 0; - - assert(dht); - assert(lu->key); - - msg.code = code; - - msg.has_key = true; - msg.key.data = (uint8_t *) lu->key; - msg.key.len = dht->b; - - while (*addrs != 0) { - struct cookie_el * c; - int ret; - - if (*addrs == dht->addr) { - ++addrs; - continue; - } - - ret = send_msg(dht, &msg, *addrs); - if (ret < 0) - break; - - c = malloc(sizeof(*c)); - if (c == NULL) - break; - - c->cookie = (uint32_t) ret; - - pthread_mutex_lock(&lu->lock); - - list_add_tail(&c->next, &lu->cookies); - - pthread_mutex_unlock(&lu->lock); - - ++sent; - ++addrs; - } - - return sent; -} - -static void lookup_detach(struct dht * dht, - struct lookup * lu) -{ - pthread_rwlock_wrlock(&dht->lock); - - list_del(&lu->next); - - pthread_rwlock_unlock(&dht->lock); -} - -static struct lookup * kad_lookup(struct dht * dht, - const uint8_t * id, - enum kad_code code) -{ - uint64_t addrs[KAD_ALPHA + 1]; - enum lookup_state state; - struct lookup * lu; - - lu = lookup_create(dht, id); - if (lu == NULL) - return NULL; - - lookup_new_addrs(lu, addrs); - - if (addrs[0] == 0) { - lookup_detach(dht, lu); - lookup_destroy(lu); - return NULL; - } - - if (kad_find(dht, lu, addrs, code) == 0) { - lookup_detach(dht, lu); - return lu; - } - - while ((state = lookup_wait(lu)) != LU_COMPLETE) { - switch (state) { - case LU_UPDATE: - lookup_new_addrs(lu, addrs); - if (addrs[0] == 0) - break; - - kad_find(dht, lu, addrs, code); - break; - case LU_DESTROY: - lookup_detach(dht, lu); - lookup_set_state(lu, LU_NULL); - return NULL; - default: - break; - } - } - - assert(state == LU_COMPLETE); - - lookup_detach(dht, lu); - - return lu; -} - -static void kad_publish(struct dht * dht, - const uint8_t * key, - uint64_t addr, - time_t exp) -{ - struct lookup * lu; - uint64_t * addrs; - ssize_t n; - size_t k; - time_t t_expire; - - - assert(dht); - assert(key); - - pthread_rwlock_rdlock(&dht->lock); - - k = dht->k; - t_expire = dht->t_expire; - - pthread_rwlock_unlock(&dht->lock); - - addrs = malloc(k * sizeof(*addrs)); - if (addrs == NULL) - return; - - lu = kad_lookup(dht, key, KAD_FIND_NODE); - if (lu == NULL) { - free(addrs); - return; - } - - n = lookup_contact_addrs(lu, addrs); - - while (n-- > 0) { - if (addrs[n] == dht->addr) { - kad_contact_msg_t msg = KAD_CONTACT_MSG__INIT; - msg.id.data = (uint8_t *) key; - msg.id.len = dht->b; - msg.addr = addr; - kad_add(dht, &msg, 1, exp); - } else { - if (kad_store(dht, key, addr, addrs[n], t_expire)) - log_warn("Failed to send store message."); - } - } - - lookup_destroy(lu); - - free(addrs); -} - -static int kad_join(struct dht * dht, - uint64_t addr) -{ - kad_msg_t msg = KAD_MSG__INIT; - - msg.code = KAD_JOIN; - - msg.has_alpha = true; - msg.has_b = true; - msg.has_k = true; - msg.has_t_refresh = true; - msg.has_t_replicate = true; - msg.alpha = KAD_ALPHA; - msg.k = KAD_K; - msg.t_refresh = KAD_T_REFR; - msg.t_replicate = KAD_T_REPL; - - pthread_rwlock_rdlock(&dht->lock); - - msg.b = dht->b; - - pthread_rwlock_unlock(&dht->lock); - - if (send_msg(dht, &msg, addr) < 0) - return -1; - - if (wait_resp(dht, &msg, KAD_T_JOIN) < 0) - return -1; - - dht->id = create_id(dht->b); - if (dht->id == NULL) - return -1; - - pthread_rwlock_wrlock(&dht->lock); - - dht_update_bucket(dht, dht->id, dht->addr); - - pthread_rwlock_unlock(&dht->lock); - - return 0; -} - -static void dht_dead_peer(struct dht * dht, - uint8_t * key, - uint64_t addr) -{ - struct list_head * p; - struct list_head * h; - struct bucket * b; - - b = dht_get_bucket(dht, key); - - list_for_each_safe(p, h, &b->contacts) { - struct contact * c = list_entry(p, struct contact, next); - if (b->n_contacts + b->n_alts <= dht->k) { - ++c->fails; - return; - } - - if (c->addr == addr) { - list_del(&c->next); - contact_destroy(c); - --b->n_contacts; - break; - } - } - - while (b->n_contacts < dht->k && b->n_alts > 0) { - struct contact * c; - c = list_first_entry(&b->alts, struct contact, next); - list_del(&c->next); - --b->n_alts; - list_add(&c->next, &b->contacts); - ++b->n_contacts; - } -} - -static int dht_del(struct dht * dht, - const uint8_t * key, - uint64_t addr) -{ - struct dht_entry * e; - - pthread_rwlock_wrlock(&dht->lock); - - e = dht_find_entry(dht, key); - if (e == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -EPERM; - } - - dht_entry_del_addr(e, addr); - - pthread_rwlock_unlock(&dht->lock); - - return 0; -} - -static buffer_t dht_retrieve(struct dht * dht, - const uint8_t * key) -{ - struct dht_entry * e; - struct list_head * p; - buffer_t buf; - uint64_t * pos; - size_t addrs = 0; - - pthread_rwlock_rdlock(&dht->lock); - - e = dht_find_entry(dht, key); - if (e == NULL) - goto fail; - - buf.len = MIN(DHT_RETR_ADDR, e->n_vals); - if (buf.len == 0) - goto fail; - - pos = malloc(sizeof(dht->addr) * buf.len); - if (pos == NULL) - goto fail; - - buf.data = (uint8_t *) pos; - - list_for_each(p, &e->vals) { - struct val * v = list_entry(p, struct val, next); - *pos++ = v->addr; - if (++addrs >= buf.len) - break; - } - - pthread_rwlock_unlock(&dht->lock); - - return buf; - - fail: - pthread_rwlock_unlock(&dht->lock); - buf.len = 0; - - return buf; -} - -static ssize_t dht_get_contacts(struct dht * dht, - const uint8_t * key, - kad_contact_msg_t *** msgs) -{ - struct list_head l; - struct list_head * p; - struct list_head * h; - size_t len; - size_t i = 0; - - list_head_init(&l); - - pthread_rwlock_wrlock(&dht->lock); - - len = dht_contact_list(dht, &l, key); - if (len == 0) { - pthread_rwlock_unlock(&dht->lock); - *msgs = NULL; - return 0; - } - - *msgs = malloc(len * sizeof(**msgs)); - if (*msgs == NULL) { - pthread_rwlock_unlock(&dht->lock); - return 0; - } - - list_for_each_safe(p, h, &l) { - struct contact * c = list_entry(p, struct contact, next); - (*msgs)[i] = malloc(sizeof(***msgs)); - if ((*msgs)[i] == NULL) { - pthread_rwlock_unlock(&dht->lock); - while (i > 0) - free(*msgs[--i]); - free(*msgs); - *msgs = NULL; - return 0; - } - - kad_contact_msg__init((*msgs)[i]); - - (*msgs)[i]->id.data = c->id; - (*msgs)[i]->id.len = dht->b; - (*msgs)[i++]->addr = c->addr; - list_del(&c->next); - free(c); - } - - pthread_rwlock_unlock(&dht->lock); - - return i; -} - -static time_t gcd(time_t a, - time_t b) -{ - if (a == 0) - return b; - - return gcd(b % a, a); -} - -static void * work(void * o) -{ - struct dht * dht; - struct timespec now; - struct list_head * p; - struct list_head * h; - struct list_head reflist; - time_t intv; - struct lookup * lu; - - dht = (struct dht *) o; - - pthread_rwlock_rdlock(&dht->lock); - - intv = gcd(dht->t_expire, dht->t_repub); - intv = gcd(intv, gcd(KAD_T_REPL, KAD_T_REFR)) / 2; - - pthread_rwlock_unlock(&dht->lock); - - list_head_init(&reflist); - - while (true) { - clock_gettime(CLOCK_REALTIME_COARSE, &now); - - pthread_rwlock_wrlock(&dht->lock); - - /* Republish registered hashes. */ - list_for_each(p, &dht->refs) { - struct ref_entry * e; - uint8_t * key; - uint64_t addr; - time_t t_expire; - e = list_entry(p, struct ref_entry, next); - if (now.tv_sec > e->t_rep) { - key = dht_dup_key(e->key, dht->b); - if (key == NULL) - continue; - addr = dht->addr; - t_expire = dht->t_expire; - e->t_rep = now.tv_sec + dht->t_repub; - - pthread_rwlock_unlock(&dht->lock); - kad_publish(dht, key, addr, t_expire); - pthread_rwlock_wrlock(&dht->lock); - free(key); - } - } - - /* Remove stale entries and republish if necessary. */ - list_for_each_safe(p, h, &dht->entries) { - struct list_head * p1; - struct list_head * h1; - struct dht_entry * e; - uint8_t * key; - time_t t_expire; - e = list_entry (p, struct dht_entry, next); - list_for_each_safe(p1, h1, &e->vals) { - struct val * v; - uint64_t addr; - v = list_entry(p1, struct val, next); - if (now.tv_sec > v->t_exp) { - list_del(&v->next); - val_destroy(v); - continue; - } - - if (now.tv_sec > v->t_rep) { - key = dht_dup_key(e->key, dht->b); - addr = v->addr; - t_expire = dht->t_expire = now.tv_sec; - v->t_rep = now.tv_sec + dht->t_replic; - pthread_rwlock_unlock(&dht->lock); - kad_publish(dht, key, addr, t_expire); - pthread_rwlock_wrlock(&dht->lock); - free(key); - } - } - } - - /* Check the requests list for unresponsive nodes. */ - list_for_each_safe(p, h, &dht->requests) { - struct kad_req * r; - r = list_entry(p, struct kad_req, next); - if (now.tv_sec > r->t_exp) { - list_del(&r->next); - bmp_release(dht->cookies, r->cookie); - dht_dead_peer(dht, r->key, r->addr); - kad_req_destroy(r); - } - } - - /* Refresh unaccessed buckets. */ - bucket_refresh(dht, dht->buckets, now.tv_sec, &reflist); - - pthread_rwlock_unlock(&dht->lock); - - list_for_each_safe(p, h, &reflist) { - struct contact * c; - c = list_entry(p, struct contact, next); - lu = kad_lookup(dht, c->id, KAD_FIND_NODE); - if (lu != NULL) - lookup_destroy(lu); - list_del(&c->next); - contact_destroy(c); - } - - sleep(intv); - } - - return (void *) 0; -} - -static int kad_handle_join_resp(struct dht * dht, - struct kad_req * req, - kad_msg_t * msg) -{ - assert(dht); - assert(req); - assert(msg); - - /* We might send version numbers later to warn of updates if needed. */ - if (!(msg->has_alpha && msg->has_b && msg->has_k && msg->has_t_expire && - msg->has_t_refresh && msg->has_t_replicate)) { - log_warn("Join refused by remote."); - return -1; - } - - if (msg->b < sizeof(uint64_t)) { - log_err("Hash sizes less than 8 bytes unsupported."); - return -1; - } - - pthread_rwlock_wrlock(&dht->lock); - - dht->buckets = bucket_create(); - if (dht->buckets == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -1; - } - - /* Likely corrupt packet. The member will refuse, we might here too. */ - if (msg->alpha != KAD_ALPHA || msg->k != KAD_K) - log_warn("Different kademlia parameters detected."); - - if (msg->t_replicate != KAD_T_REPL) - log_warn("Different kademlia replication time detected."); - - if (msg->t_refresh != KAD_T_REFR) - log_warn("Different kademlia refresh time detected."); - - dht->k = msg->k; - dht->b = msg->b; - dht->t_expire = msg->t_expire; - dht->t_repub = MAX(1, dht->t_expire - 10); - - if (pthread_create(&dht->worker, NULL, work, dht)) { - bucket_destroy(dht->buckets); - pthread_rwlock_unlock(&dht->lock); - return -1; - } - - kad_req_respond(req); - - dht_update_bucket(dht, msg->s_id.data, msg->s_addr); - - pthread_rwlock_unlock(&dht->lock); - - log_dbg("Enrollment of DHT completed."); - - return 0; -} - -static int kad_handle_find_resp(struct dht * dht, - struct kad_req * req, - kad_msg_t * msg) -{ - struct lookup * lu; - - assert(dht); - assert(req); - assert(msg); - - pthread_rwlock_rdlock(&dht->lock); - - lu = dht_find_lookup(dht, req->cookie); - if (lu == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -1; - } - - lookup_update(dht, lu, msg); - - pthread_rwlock_unlock(&dht->lock); - - return 0; -} - -static void kad_handle_response(struct dht * dht, - kad_msg_t * msg) -{ - struct kad_req * req; - - assert(dht); - assert(msg); - - pthread_rwlock_wrlock(&dht->lock); - - req = dht_find_request(dht, msg); - if (req == NULL) { - pthread_rwlock_unlock(&dht->lock); - return; - } - - bmp_release(dht->cookies, req->cookie); - list_del(&req->next); - - pthread_rwlock_unlock(&dht->lock); - - switch(req->code) { - case KAD_JOIN: - if (kad_handle_join_resp(dht, req, msg)) - log_err("Enrollment of DHT failed."); - break; - case KAD_FIND_VALUE: - case KAD_FIND_NODE: - if (dht_get_state(dht) != DHT_RUNNING) - break; - kad_handle_find_resp(dht, req, msg); - break; - default: - break; - } - - kad_req_destroy(req); -} - -int dht_bootstrap(struct dht * dht, - size_t b, - time_t t_expire) -{ - assert(dht); - - pthread_rwlock_wrlock(&dht->lock); - - dht->id = create_id(b); - if (dht->id == NULL) - goto fail_id; - - dht->buckets = bucket_create(); - if (dht->buckets == NULL) - goto fail_buckets; - - dht->buckets->depth = 0; - dht->buckets->mask = 0; - - dht->b = b / CHAR_BIT; - dht->t_expire = MAX(2, t_expire); - dht->t_repub = MAX(1, t_expire - 10); - dht->k = KAD_K; - - if (pthread_create(&dht->worker, NULL, work, dht)) - goto fail_pthread_create; - - dht->state = DHT_RUNNING; - - dht_update_bucket(dht, dht->id, dht->addr); - - pthread_rwlock_unlock(&dht->lock); - - return 0; - - fail_pthread_create: - bucket_destroy(dht->buckets); - dht->buckets = NULL; - fail_buckets: - free(dht->id); - dht->id = NULL; - fail_id: - pthread_rwlock_unlock(&dht->lock); - return -1; -} - -static struct ref_entry * ref_entry_get(struct dht * dht, - const uint8_t * key) -{ - struct list_head * p; - - list_for_each(p, &dht->refs) { - struct ref_entry * r = list_entry(p, struct ref_entry, next); - if (!memcmp(key, r->key, dht-> b) ) - return r; - } - - return NULL; -} - -int dht_reg(struct dht * dht, - const uint8_t * key) -{ - struct ref_entry * e; - uint64_t addr; - time_t t_expire; - - assert(dht); - assert(key); - assert(dht->addr != 0); - - if (dht_wait_running(dht)) - return -1; - - pthread_rwlock_wrlock(&dht->lock); - - if (ref_entry_get(dht, key) != NULL) { - log_dbg("Name already registered."); - pthread_rwlock_unlock(&dht->lock); - return 0; - } - - e = ref_entry_create(dht, key); - if (e == NULL) { - pthread_rwlock_unlock(&dht->lock); - return -ENOMEM; - } - - list_add(&e->next, &dht->refs); - - t_expire = dht->t_expire; - addr = dht->addr; - - pthread_rwlock_unlock(&dht->lock); - - kad_publish(dht, key, addr, t_expire); - - return 0; -} - -int dht_unreg(struct dht * dht, - const uint8_t * key) -{ - struct list_head * p; - struct list_head * h; - - assert(dht); - assert(key); - - if (dht_get_state(dht) != DHT_RUNNING) - return -1; - - pthread_rwlock_wrlock(&dht->lock); - - list_for_each_safe(p, h, &dht->refs) { - struct ref_entry * r = list_entry(p, struct ref_entry, next); - if (!memcmp(key, r->key, dht-> b) ) { - list_del(&r->next); - ref_entry_destroy(r); - } - } - - dht_del(dht, key, dht->addr); - - pthread_rwlock_unlock(&dht->lock); - - return 0; -} - -uint64_t dht_query(struct dht * dht, - const uint8_t * key) -{ - struct dht_entry * e; - struct lookup * lu; - uint64_t addrs[KAD_K]; - size_t n; - - addrs[0] = 0; - - if (dht_wait_running(dht)) - return 0; - - pthread_rwlock_rdlock(&dht->lock); - - e = dht_find_entry(dht, key); - if (e != NULL) - addrs[0] = dht_entry_get_addr(dht, e); - - pthread_rwlock_unlock(&dht->lock); - - if (addrs[0] != 0) - return addrs[0]; - - lu = kad_lookup(dht, key, KAD_FIND_VALUE); - if (lu == NULL) - return 0; - - n = lookup_get_addrs(lu, addrs); - if (n == 0) { - lookup_destroy(lu); - return 0; - } - - lookup_destroy(lu); - - /* Current behaviour is anycast and return the first peer address. */ - if (addrs[0] != dht->addr) - return addrs[0]; - - if (n > 1) - return addrs[1]; - - return 0; -} - -static void * dht_handle_packet(void * o) -{ - struct dht * dht = (struct dht *) o; - - assert(dht); - - while (true) { - kad_msg_t * msg; - kad_contact_msg_t ** cmsgs; - kad_msg_t resp_msg = KAD_MSG__INIT; - uint64_t addr; - buffer_t buf; - size_t i; - size_t b; - size_t t_expire; - struct cmd * cmd; - - pthread_mutex_lock(&dht->mtx); - - pthread_cleanup_push((void *)(void *) pthread_mutex_unlock, - &dht->mtx); - - while (list_is_empty(&dht->cmds)) - pthread_cond_wait(&dht->cond, &dht->mtx); - - cmd = list_last_entry(&dht->cmds, struct cmd, next); - list_del(&cmd->next); - - pthread_cleanup_pop(true); - - i = shm_du_buff_tail(cmd->sdb) - shm_du_buff_head(cmd->sdb); - - msg = kad_msg__unpack(NULL, i, shm_du_buff_head(cmd->sdb)); -#ifndef __DHT_TEST__ - ipcp_sdb_release(cmd->sdb); -#endif - free(cmd); - - if (msg == NULL) { - log_err("Failed to unpack message."); - continue; - } - - if (msg->code != KAD_RESPONSE && dht_wait_running(dht)) { - kad_msg__free_unpacked(msg, NULL); - log_dbg("Got a request message when not running."); - continue; - } - - pthread_rwlock_rdlock(&dht->lock); - - b = dht->b; - t_expire = dht->t_expire; - - pthread_rwlock_unlock(&dht->lock); - - if (msg->has_key && msg->key.len != b) { - kad_msg__free_unpacked(msg, NULL); - log_warn("Bad key in message."); - continue; - } - - if (msg->has_s_id && !msg->has_b && msg->s_id.len != b) { - kad_msg__free_unpacked(msg, NULL); - log_warn("Bad source ID in message of type %d.", - msg->code); - continue; - } - - tpm_dec(dht->tpm); - - addr = msg->s_addr; - - resp_msg.code = KAD_RESPONSE; - resp_msg.cookie = msg->cookie; - - switch(msg->code) { - case KAD_JOIN: - /* Refuse enrollee on check fails. */ - if (msg->alpha != KAD_ALPHA || msg->k != KAD_K) { - log_warn("Parameter mismatch. " - "DHT enrolment refused."); - break; - } - - if (msg->t_replicate != KAD_T_REPL) { - log_warn("Replication time mismatch. " - "DHT enrolment refused."); - - break; - } - - if (msg->t_refresh != KAD_T_REFR) { - log_warn("Refresh time mismatch. " - "DHT enrolment refused."); - break; - } - - resp_msg.has_alpha = true; - resp_msg.has_b = true; - resp_msg.has_k = true; - resp_msg.has_t_expire = true; - resp_msg.has_t_refresh = true; - resp_msg.has_t_replicate = true; - resp_msg.alpha = KAD_ALPHA; - resp_msg.b = b; - resp_msg.k = KAD_K; - resp_msg.t_expire = t_expire; - resp_msg.t_refresh = KAD_T_REFR; - resp_msg.t_replicate = KAD_T_REPL; - break; - case KAD_FIND_VALUE: - buf = dht_retrieve(dht, msg->key.data); - if (buf.len != 0) { - resp_msg.n_addrs = buf.len; - resp_msg.addrs = (uint64_t *) buf.data; - break; - } - /* FALLTHRU */ - case KAD_FIND_NODE: - /* Return k closest contacts. */ - resp_msg.n_contacts = - dht_get_contacts(dht, msg->key.data, &cmsgs); - resp_msg.contacts = cmsgs; - break; - case KAD_STORE: - if (msg->n_contacts < 1) { - log_warn("No contacts in store message."); - break; - } - - if (!msg->has_t_expire) { - log_warn("No expiry time in store message."); - break; - } - - kad_add(dht, *msg->contacts, msg->n_contacts, - msg->t_expire); - break; - case KAD_RESPONSE: - kad_handle_response(dht, msg); - break; - default: - assert(false); - break; - } - - if (msg->code != KAD_JOIN) { - pthread_rwlock_wrlock(&dht->lock); - if (dht_get_state(dht) == DHT_JOINING && - dht->buckets == NULL) { - pthread_rwlock_unlock(&dht->lock); - goto finish; - } - - if (dht_update_bucket(dht, msg->s_id.data, addr)) - log_warn("Failed to update bucket."); - pthread_rwlock_unlock(&dht->lock); - } - - if (msg->code < KAD_STORE && send_msg(dht, &resp_msg, addr) < 0) - log_warn("Failed to send response."); - - finish: - kad_msg__free_unpacked(msg, NULL); - - if (resp_msg.n_addrs > 0) - free(resp_msg.addrs); - - if (resp_msg.n_contacts == 0) { - tpm_inc(dht->tpm); - continue; - } - - for (i = 0; i < resp_msg.n_contacts; ++i) - kad_contact_msg__free_unpacked(resp_msg.contacts[i], - NULL); - free(resp_msg.contacts); - - tpm_inc(dht->tpm); - } - - return (void *) 0; -} - -static void dht_post_packet(void * comp, - struct shm_du_buff * sdb) -{ - struct cmd * cmd; - struct dht * dht = (struct dht *) comp; - - if (dht_get_state(dht) == DHT_SHUTDOWN) { -#ifndef __DHT_TEST__ - ipcp_sdb_release(sdb); -#endif - return; - } - - cmd = malloc(sizeof(*cmd)); - if (cmd == NULL) { - log_err("Command failed. Out of memory."); - return; - } - - cmd->sdb = sdb; - - pthread_mutex_lock(&dht->mtx); - - list_add(&cmd->next, &dht->cmds); - - pthread_cond_signal(&dht->cond); - - pthread_mutex_unlock(&dht->mtx); -} - -void dht_destroy(struct dht * dht) -{ - struct list_head * p; - struct list_head * h; - - if (dht == NULL) - return; - -#ifndef __DHT_TEST__ - tpm_stop(dht->tpm); - - tpm_destroy(dht->tpm); -#endif - if (dht_get_state(dht) == DHT_RUNNING) { - dht_set_state(dht, DHT_SHUTDOWN); - pthread_cancel(dht->worker); - pthread_join(dht->worker, NULL); - } - - pthread_rwlock_wrlock(&dht->lock); - - list_for_each_safe(p, h, &dht->cmds) { - struct cmd * c = list_entry(p, struct cmd, next); - list_del(&c->next); -#ifndef __DHT_TEST__ - ipcp_sdb_release(c->sdb); -#endif - free(c); - } - - list_for_each_safe(p, h, &dht->entries) { - struct dht_entry * e = list_entry(p, struct dht_entry, next); - list_del(&e->next); - dht_entry_destroy(e); - } - - list_for_each_safe(p, h, &dht->requests) { - struct kad_req * r = list_entry(p, struct kad_req, next); - list_del(&r->next); - kad_req_destroy(r); - } - - list_for_each_safe(p, h, &dht->refs) { - struct ref_entry * e = list_entry(p, struct ref_entry, next); - list_del(&e->next); - ref_entry_destroy(e); - } - - list_for_each_safe(p, h, &dht->lookups) { - struct lookup * l = list_entry(p, struct lookup, next); - list_del(&l->next); - lookup_destroy(l); - } - - pthread_rwlock_unlock(&dht->lock); - - if (dht->buckets != NULL) - bucket_destroy(dht->buckets); - - bmp_destroy(dht->cookies); - - pthread_mutex_destroy(&dht->mtx); - - pthread_rwlock_destroy(&dht->lock); - - free(dht->id); - - free(dht); -} - -static void * join_thr(void * o) -{ - struct join_info * info = (struct join_info *) o; - struct lookup * lu; - size_t retr = 0; - - assert(info); - - while (kad_join(info->dht, info->addr)) { - if (dht_get_state(info->dht) == DHT_SHUTDOWN) { - log_dbg("DHT enrollment aborted."); - goto finish; - } - - if (retr++ == KAD_JOIN_RETR) { - dht_set_state(info->dht, DHT_INIT); - log_warn("DHT enrollment attempt failed."); - goto finish; - } - - sleep(KAD_JOIN_INTV); - } - - dht_set_state(info->dht, DHT_RUNNING); - - lu = kad_lookup(info->dht, info->dht->id, KAD_FIND_NODE); - if (lu != NULL) - lookup_destroy(lu); - - finish: - free(info); - - return (void *) 0; -} - -static void handle_event(void * self, - int event, - const void * o) -{ - struct dht * dht = (struct dht *) self; - - if (event == NOTIFY_DT_CONN_ADD) { - pthread_t thr; - struct join_info * inf; - struct conn * c = (struct conn *) o; - struct timespec slack = {0, DHT_ENROLL_SLACK * MILLION}; - - /* Give the pff some time to update for the new link. */ - nanosleep(&slack, NULL); - - switch(dht_get_state(dht)) { - case DHT_INIT: - inf = malloc(sizeof(*inf)); - if (inf == NULL) - break; - - inf->dht = dht; - inf->addr = c->conn_info.addr; - - if (dht_set_state(dht, DHT_JOINING) == 0 || - dht_wait_running(dht)) { - if (pthread_create(&thr, NULL, join_thr, inf)) { - dht_set_state(dht, DHT_INIT); - free(inf); - return; - } - pthread_detach(thr); - } else { - free(inf); - } - break; - case DHT_RUNNING: - /* - * FIXME: this lookup for effiency reasons - * causes a SEGV when stressed with rapid - * enrollments. - * lu = kad_lookup(dht, dht->id, KAD_FIND_NODE); - * if (lu != NULL) - * lookup_destroy(lu); - */ - break; - default: - break; - } - } -} - -struct dht * dht_create(uint64_t addr) -{ - struct dht * dht; - - dht = malloc(sizeof(*dht)); - if (dht == NULL) - goto fail_malloc; - - dht->buckets = NULL; - - list_head_init(&dht->entries); - list_head_init(&dht->requests); - list_head_init(&dht->refs); - list_head_init(&dht->lookups); - list_head_init(&dht->cmds); - - if (pthread_rwlock_init(&dht->lock, NULL)) - goto fail_rwlock; - - if (pthread_mutex_init(&dht->mtx, NULL)) - goto fail_mutex; - - if (pthread_cond_init(&dht->cond, NULL)) - goto fail_cond; - - dht->cookies = bmp_create(DHT_MAX_REQS, 1); - if (dht->cookies == NULL) - goto fail_bmp; - - dht->b = 0; - dht->addr = addr; - dht->id = NULL; -#ifndef __DHT_TEST__ - dht->tpm = tpm_create(2, 1, dht_handle_packet, dht); - if (dht->tpm == NULL) - goto fail_tpm_create; - - if (tpm_start(dht->tpm)) - goto fail_tpm_start; - - dht->eid = dt_reg_comp(dht, &dht_post_packet, DHT); - if ((int) dht->eid < 0) - goto fail_tpm_start; - - notifier_reg(handle_event, dht); -#else - (void) handle_event; - (void) dht_handle_packet; - (void) dht_post_packet; -#endif - dht->state = DHT_INIT; - - return dht; -#ifndef __DHT_TEST__ - fail_tpm_start: - tpm_destroy(dht->tpm); - fail_tpm_create: - bmp_destroy(dht->cookies); -#endif - fail_bmp: - pthread_cond_destroy(&dht->cond); - fail_cond: - pthread_mutex_destroy(&dht->mtx); - fail_mutex: - pthread_rwlock_destroy(&dht->lock); - fail_rwlock: - free(dht); - fail_malloc: - return NULL; -} diff --git a/src/ipcpd/unicast/dht.h b/src/ipcpd/unicast/dht.h deleted file mode 100644 index 39dfc07e..00000000 --- a/src/ipcpd/unicast/dht.h +++ /dev/null @@ -1,52 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Distributed Hash Table based on Kademlia - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#ifndef OUROBOROS_IPCPD_UNICAST_DHT_H -#define OUROBOROS_IPCPD_UNICAST_DHT_H - -#include <ouroboros/ipcp-dev.h> - -#include <stdint.h> -#include <sys/types.h> - -struct dht; - -struct dht * dht_create(uint64_t addr); - -int dht_bootstrap(struct dht * dht, - size_t b, - time_t t_expire); - -void dht_destroy(struct dht * dht); - -int dht_reg(struct dht * dht, - const uint8_t * key); - -int dht_unreg(struct dht * dht, - const uint8_t * key); - -uint64_t dht_query(struct dht * dht, - const uint8_t * key); - -int dht_wait_running(struct dht * dht); - -#endif /* OUROBOROS_IPCPD_UNICAST_DHT_H */ diff --git a/src/ipcpd/unicast/dir.c b/src/ipcpd/unicast/dir.c index b99380fe..a31a562f 100644 --- a/src/ipcpd/unicast/dir.c +++ b/src/ipcpd/unicast/dir.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * - * Directory + * Directory Management * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -34,8 +34,7 @@ #include <ouroboros/utils.h> #include "dir.h" -#include "dht.h" -#include "ipcp.h" +#include "dir/pol.h" #include <stdlib.h> #include <string.h> @@ -43,60 +42,59 @@ #include <inttypes.h> #include <limits.h> -#define KAD_B (hash_len(ipcpi.dir_hash_algo) * CHAR_BIT) +struct { + struct dir_ops * ops; +} dir; -struct ipcp icpci; -struct dht * dht; - -int dir_init(void) +int dir_init(struct dir_config * conf) { - dht = dht_create(ipcpi.dt_addr); - if (dht == NULL) - return -ENOMEM; + void * cfg; + + assert(conf != NULL); + + switch (conf->pol) { + case DIR_DHT: + log_info("Using DHT policy."); + dir.ops = &dht_dir_ops; + cfg = &conf->dht; + break; + default: /* DIR_INVALID */ + log_err("Invalid directory policy %d.", conf->pol); + return -EINVAL; + } - return 0; + assert(dir.ops->init != NULL); + + return dir.ops->init(cfg); } void dir_fini(void) { - dht_destroy(dht); + dir.ops->fini(); + dir.ops = NULL; } -int dir_bootstrap(void) { - log_dbg("Bootstrapping directory."); - - /* TODO: get parameters for bootstrap from IRM tool. */ - if (dht_bootstrap(dht, KAD_B, 86400)) { - dht_destroy(dht); - return -ENOMEM; - } - - log_info("Directory bootstrapped."); +int dir_start(void) +{ + return dir.ops->start(); +} - return 0; +void dir_stop(void) +{ + dir.ops->stop(); } int dir_reg(const uint8_t * hash) { - return dht_reg(dht, hash); + return dir.ops->reg(hash); } int dir_unreg(const uint8_t * hash) { - return dht_unreg(dht, hash); + return dir.ops->unreg(hash); } uint64_t dir_query(const uint8_t * hash) { - return dht_query(dht, hash); -} - -int dir_wait_running(void) -{ - if (dht_wait_running(dht)) { - log_warn("Directory did not bootstrap."); - return -1; - } - - return 0; + return dir.ops->query(hash); } diff --git a/src/ipcpd/unicast/dir.h b/src/ipcpd/unicast/dir.h index 1b67a08c..2be7b10f 100644 --- a/src/ipcpd/unicast/dir.h +++ b/src/ipcpd/unicast/dir.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Directory * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,11 +23,16 @@ #ifndef OUROBOROS_IPCPD_UNICAST_DIR_H #define OUROBOROS_IPCPD_UNICAST_DIR_H -int dir_init(void); +#include <inttypes.h> + +/* may update the config! */ +int dir_init(struct dir_config * conf); void dir_fini(void); -int dir_bootstrap(void); +int dir_start(void); + +void dir_stop(void); int dir_reg(const uint8_t * hash); @@ -35,6 +40,4 @@ int dir_unreg(const uint8_t * hash); uint64_t dir_query(const uint8_t * hash); -int dir_wait_running(void); - #endif /* OUROBOROS_IPCPD_UNICAST_DIR_H */ diff --git a/src/ipcpd/unicast/dir/dht.c b/src/ipcpd/unicast/dir/dht.c new file mode 100644 index 00000000..c2cd33aa --- /dev/null +++ b/src/ipcpd/unicast/dir/dht.c @@ -0,0 +1,4043 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Distributed Hash Table based on Kademlia + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#if !defined (__DHT_TEST__) + #if defined(__linux__) || defined(__CYGWIN__) + #define _DEFAULT_SOURCE + #else + #define _POSIX_C_SOURCE 200112L + #endif +#endif + +#include "config.h" + +#define DHT "dht" +#define OUROBOROS_PREFIX DHT + +#include <ouroboros/endian.h> +#include <ouroboros/hash.h> +#include <ouroboros/ipcp-dev.h> +#include <ouroboros/bitmap.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> +#include <ouroboros/list.h> +#include <ouroboros/random.h> +#include <ouroboros/rib.h> +#include <ouroboros/time.h> +#include <ouroboros/tpm.h> +#include <ouroboros/utils.h> +#include <ouroboros/pthread.h> + +#include "addr-auth.h" +#include "common/connmgr.h" +#include "dht.h" +#include "dt.h" +#include "ipcp.h" +#include "ops.h" + +#include <stdlib.h> +#include <string.h> +#include <assert.h> +#include <inttypes.h> +#include <limits.h> + +#include "dht.pb-c.h" +typedef DhtMsg dht_msg_t; +typedef DhtContactMsg dht_contact_msg_t; +typedef DhtStoreMsg dht_store_msg_t; +typedef DhtFindReqMsg dht_find_req_msg_t; +typedef DhtFindNodeRspMsg dht_find_node_rsp_msg_t; +typedef DhtFindValueRspMsg dht_find_value_rsp_msg_t; +typedef ProtobufCBinaryData binary_data_t; + +#ifndef CLOCK_REALTIME_COARSE +#define CLOCK_REALTIME_COARSE CLOCK_REALTIME +#endif + +#define DHT_MAX_REQS 128 /* KAD recommends rnd(), bmp can be changed. */ +#define DHT_WARN_REQS 100 /* Warn if number of requests exceeds this. */ +#define DHT_MAX_VALS 8 /* Max number of values to return for a key. */ +#define DHT_T_CACHE 60 /* Max cache time for values (s) */ +#define DHT_T_RESP 2 /* Response time to wait for a response (s). */ +#define DHT_N_REPUB 5 /* Republish if expiry within n replications. */ +#define DHT_R_PING 2 /* Ping retries before declaring peer dead. */ +#define DHT_QUEER 15 /* Time to declare peer questionable. */ +#define DHT_BETA 8 /* Bucket split factor, must be 1, 2, 4 or 8. */ +#define DHT_RESP_RETR 6 /* Number of retries on sending a response. */ +#define HANDLE_TIMEO 1000 /* Timeout for dht_handle_packet tpm check (ms) */ +#define DHT_INVALID 0 /* Invalid cookie value. */ + +#define KEY_FMT "K<" HASH_FMT64 ">" +#define KEY_VAL(key) HASH_VAL64(key) + +#define VAL_FMT "V<" HASH_FMT64 ">" +#define VAL_VAL(val) HASH_VAL64((val).data) + +#define KV_FMT "<" HASH_FMT64 ", " HASH_FMT64 ">" +#define KV_VAL(key, val) HASH_VAL64(key), HASH_VAL64((val).data) + +#define PEER_FMT "[" HASH_FMT64 "|" ADDR_FMT32 "]" +#define PEER_VAL(id, addr) HASH_VAL64(id), ADDR_VAL32(&(addr)) + +#define DHT_CODE(msg) dht_code_str[(msg)->code] + +#define TX_HDR_FMT "%s --> " PEER_FMT +#define TX_HDR_VAL(msg, id, addr) DHT_CODE(msg), PEER_VAL(id, addr) + +#define RX_HDR_FMT "%s <-- " PEER_FMT +#define RX_HDR_VAL(msg) DHT_CODE(msg), \ + PEER_VAL(msg->src->id.data, msg->src->addr) + +#define CK_FMT "|" HASH_FMT64 "|" +#define CK_VAL(cookie) HASH_VAL64(&(cookie)) + +#define IS_REQUEST(code) \ + (code == DHT_FIND_NODE_REQ || code == DHT_FIND_VALUE_REQ) + +enum dht_code { + DHT_STORE, + DHT_FIND_NODE_REQ, + DHT_FIND_NODE_RSP, + DHT_FIND_VALUE_REQ, + DHT_FIND_VALUE_RSP +}; + +const char * dht_code_str[] = { + "DHT_STORE", + "DHT_FIND_NODE_REQ", + "DHT_FIND_NODE_RSP", + "DHT_FIND_VALUE_REQ", + "DHT_FIND_VALUE_RSP" +}; + +enum dht_state { + DHT_NULL = 0, + DHT_INIT, + DHT_RUNNING +}; + +struct val_entry { + struct list_head next; + + buffer_t val; + + time_t t_exp; /* Expiry time */ + time_t t_repl; /* Last replication time */ +}; + +struct dht_entry { + struct list_head next; + + uint8_t * key; + + struct llist vals; /* We don't own these, only replicate */ + struct llist lvals; /* We own these, must be republished */ +}; + +struct contact { + struct list_head next; + + uint8_t * id; + uint64_t addr; + + size_t fails; + time_t t_seen; +}; + +struct peer_entry { + struct list_head next; + + uint64_t cookie; + uint8_t * id; + uint64_t addr; + enum dht_code code; + + time_t t_sent; +}; + +struct dht_req { + struct list_head next; + + uint8_t * key; + time_t t_exp; + + struct llist peers; + struct llist cache; +}; + +struct bucket { + struct llist contacts; + struct llist alts; + + time_t t_refr; + + size_t depth; + uint8_t mask; + + struct bucket * parent; + struct bucket * children[1L << DHT_BETA]; +}; + +struct cmd { + struct list_head next; + buffer_t cbuf; +}; + +struct dir_ops dht_dir_ops = { + .init = (int (*)(void *)) dht_init, + .fini = dht_fini, + .start = dht_start, + .stop = dht_stop, + .reg = dht_reg, + .unreg = dht_unreg, + .query = dht_query +}; + +struct { + struct { /* Kademlia parameters */ + uint32_t alpha; /* Number of concurrent requests */ + size_t k; /* Number of replicas to store */ + time_t t_exp; /* Expiry time for values (s) */ + time_t t_refr; /* Refresh time for contacts (s) */ + time_t t_repl; /* Replication time for values (s) */ + }; + + buffer_t id; + + time_t t0; /* Creation time */ + uint64_t addr; /* Our own address */ + uint64_t peer; /* Enrollment peer address */ + uint64_t magic; /* Magic cookie for retransmit */ + + uint64_t eid; /* Entity ID */ + + struct tpm * tpm; + pthread_t worker; + + enum dht_state state; + + struct { + struct { + struct bucket * root; + } contacts; + + struct { + struct llist ll; + size_t vals; + size_t lvals; + } kv; + + pthread_rwlock_t lock; + } db; + + struct { + struct llist ll; + pthread_cond_t cond; + pthread_mutex_t mtx; + } reqs; + + struct { + struct list_head list; + pthread_cond_t cond; + pthread_mutex_t mtx; + } cmds; +} dht; + + +/* DHT RIB */ + +static const char * dht_dir[] = { + "database", + "stats", + NULL +}; + +const char * dht_stats = \ + "DHT: " HASH_FMT64 "\n" + " Created: %s\n" + " Address: " ADDR_FMT32 "\n" + " Kademlia parameters:\n" + " Number of concurrent requests (alpha): %10zu\n" + " Number of replicas (k): %10zu\n" + " Expiry time for values (s): %10ld\n" + " Refresh time for contacts (s): %10ld\n" + " Replication time for values (s): %10ld\n" + " Number of keys: %10zu\n" + " Number of local values: %10zu\n" + " Number of non-local values: %10zu\n"; + +static int dht_rib_statfile(char * buf, + size_t len) +{ + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + size_t keys; + size_t vals; + size_t lvals; + + assert(buf != NULL); + assert(len > 0); + + pthread_rwlock_rdlock(&dht.db.lock); + + keys = dht.db.kv.ll.len; + lvals = dht.db.kv.lvals; + vals = dht.db.kv.vals; + + pthread_rwlock_unlock(&dht.db.lock); + + tm = gmtime(&dht.t0); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + snprintf(buf, len, dht_stats, + HASH_VAL64(dht.id.data), + tmstr, + ADDR_VAL32(&dht.addr), + dht.alpha, dht.k, + dht.t_exp, dht.t_refr, dht.t_repl, + keys, vals, lvals); + + return strlen(buf); +} + +static size_t dht_db_file_len(void) +{ + size_t sz; + size_t vals; + + sz = 18; /* DHT database + 2 * \n */ + + pthread_rwlock_rdlock(&dht.db.lock); + + if (llist_is_empty(&dht.db.kv.ll)) { + pthread_rwlock_unlock(&dht.db.lock); + sz += 14; /* No entries */ + return sz; + } + + sz += 39 * 3 + 1; /* tally + extra newline */ + sz += dht.db.kv.ll.len * (25 + 19 + 23 + 1); + + vals = dht.db.kv.vals + dht.db.kv.lvals; + + sz += vals * (48 + 2 * RIB_TM_STRLEN); + + pthread_rwlock_unlock(&dht.db.lock); + + return sz; +} + +static int dht_rib_dbfile(char * buf, + size_t len) +{ + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + char exstr[RIB_TM_STRLEN]; + size_t i = 0; + struct list_head * p; + + assert(buf != NULL); + assert(len > 0); + + pthread_rwlock_rdlock(&dht.db.lock); + + if (llist_is_empty(&dht.db.kv.ll)) { + i += snprintf(buf, len, " No entries.\n"); + pthread_rwlock_unlock(&dht.db.lock); + return i; + } + + i += snprintf(buf + i, len - i, "DHT database:\n\n"); + i += snprintf(buf + i, len - i, + "Number of keys: %10zu\n" + "Number of local values: %10zu\n" + "Number of non-local values: %10zu\n\n", + dht.db.kv.ll.len, dht.db.kv.vals, dht.db.kv.lvals); + + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + struct list_head * h; + + i += snprintf(buf + i, len - i, "Key: " KEY_FMT "\n", + KEY_VAL(e->key)); + i += snprintf(buf + i, len - i, " Local entries:\n"); + + llist_for_each(h, &e->vals) { + struct val_entry * v; + + v = list_entry(h, struct val_entry, next); + + tm = gmtime(&v->t_repl); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + tm = gmtime(&v->t_exp); + strftime(exstr, sizeof(exstr), RIB_TM_FORMAT, tm); + + i += snprintf(buf + i, len - i, + " " VAL_FMT + ", t_replicated=%.*s, t_exp=%.*s\n", + VAL_VAL(v->val), + RIB_TM_STRLEN, tmstr, + RIB_TM_STRLEN, exstr); + } + + i += snprintf(buf + i, len - i, "\n"); + + i += snprintf(buf + i, len - i, " Non-local entries:\n"); + + llist_for_each(h, &e->lvals) { + struct val_entry * v; + + v= list_entry(h, struct val_entry, next); + + tm = gmtime(&v->t_repl); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + + tm = gmtime(&v->t_exp); + strftime(exstr, sizeof(exstr), RIB_TM_FORMAT, tm); + + i += snprintf(buf + i, len - i, + " " VAL_FMT + ", t_replicated=%.*s, t_exp=%.*s\n", + VAL_VAL(v->val), + RIB_TM_STRLEN, tmstr, + RIB_TM_STRLEN, exstr); + + } + } + + pthread_rwlock_unlock(&dht.db.lock); + + printf("DHT RIB DB file generated (%zu bytes).\n", i); + + return i; +} + +static int dht_rib_read(const char * path, + char * buf, + size_t len) +{ + char * entry; + + entry = strstr(path, RIB_SEPARATOR) + 1; + + if (strcmp(entry, "database") == 0) { + return dht_rib_dbfile(buf, len); + } else if (strcmp(entry, "stats") == 0) { + return dht_rib_statfile(buf, len); + } + + return 0; +} + +static int dht_rib_readdir(char *** buf) +{ + int i = 0; + + while (dht_dir[i++] != NULL); + + *buf = malloc(sizeof(**buf) * i); + if (*buf == NULL) + goto fail_buf; + + i = 0; + + while (dht_dir[i] != NULL) { + (*buf)[i] = strdup(dht_dir[i]); + if ((*buf)[i] == NULL) + goto fail_dup; + i++; + } + + return i; + fail_dup: + freepp(char, *buf, i); + fail_buf: + return -ENOMEM; +} + +static int dht_rib_getattr(const char * path, + struct rib_attr * attr) +{ + struct timespec now; + char * entry; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + attr->mtime = now.tv_sec; + + entry = strstr(path, RIB_SEPARATOR) + 1; + + if (strcmp(entry, "database") == 0) { + attr->size = dht_db_file_len(); + } else if (strcmp(entry, "stats") == 0) { + attr->size = 545; + } + + return 0; +} + +static struct rib_ops r_ops = { + .read = dht_rib_read, + .readdir = dht_rib_readdir, + .getattr = dht_rib_getattr +}; + +/* Helper functions */ + +static uint8_t * generate_id(void) +{ + uint8_t * id; + + if(dht.id.len < sizeof(uint64_t)) { + log_err("DHT ID length is too short (%zu < %zu).", + dht.id.len, sizeof(uint64_t)); + return NULL; + } + + id = malloc(dht.id.len); + if (id == NULL) { + log_err("Failed to malloc ID."); + goto fail_id; + } + + if (random_buffer(id, dht.id.len) < 0) { + log_err("Failed to generate random ID."); + goto fail_rnd; + } + + return id; + fail_rnd: + free(id); + fail_id: + return NULL; +} + +static uint64_t generate_cookie(void) +{ + uint64_t cookie = DHT_INVALID; + + while (cookie == DHT_INVALID) + random_buffer((uint8_t *) &cookie, sizeof(cookie)); + + return cookie; +} + +/* + * If someone builds a network where the n (n > k) closest nodes all + * have IDs starting with the same 64 bits: by all means, change this. + */ +static uint64_t dist(const uint8_t * src, + const uint8_t * dst) +{ + assert(dht.id.len >= sizeof(uint64_t)); + + return betoh64(*((uint64_t *) src) ^ *((uint64_t *) dst)); +} + +#define IS_CLOSER(x, y) (dist((x), dht.id.data) < dist((y), dht.id.data)) + +static int addr_to_buf(const uint64_t addr, + buffer_t * buf) +{ + size_t len; + uint64_t _addr; + + len = sizeof(addr); + _addr = hton64(addr); + + assert(buf != NULL); + + buf->data = malloc(len); + if (buf->data == NULL) + goto fail_malloc; + + buf->len = sizeof(_addr); + memcpy(buf->data, &_addr, sizeof(_addr)); + + return 0; + fail_malloc: + return -ENOMEM; +} + +static int buf_to_addr(const buffer_t buf, + uint64_t * addr) +{ + assert(addr != NULL); + assert(buf.data != NULL); + + if (buf.len != sizeof(*addr)) + return - EINVAL; + + *addr = ntoh64(*((uint64_t *) buf.data)); + + if (*addr == dht.addr) + *addr = INVALID_ADDR; + + return 0; +} + +static uint8_t * dht_dup_key(const uint8_t * key) +{ + uint8_t * dup; + + assert(key != NULL); + assert(dht.id.len != 0); + + dup = malloc(dht.id.len); + if (dup == NULL) + return NULL; + + memcpy(dup, key, dht.id.len); + + return dup; +} + +/* DHT */ + +static struct val_entry * val_entry_create(const buffer_t val, + time_t exp) +{ + struct val_entry * e; + struct timespec now; + + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + +#ifndef __DHT_TEST_ALLOW_EXPIRED__ + if (exp < now.tv_sec) + return NULL; /* Refuse to add expired values */ +#endif + e = malloc(sizeof(*e)); + if (e == NULL) + goto fail_entry; + + list_head_init(&e->next); + + e->val.len = val.len; + e->val.data = malloc(val.len); + if (e->val.data == NULL) + goto fail_val; + + memcpy(e->val.data, val.data, val.len); + + e->t_repl = 0; + e->t_exp = exp; + + return e; + + fail_val: + free(e); + fail_entry: + return NULL; +} + +static void val_entry_destroy(struct val_entry * v) +{ + assert(v->val.data != NULL); + + freebuf(v->val); + free(v); +} + +static struct dht_entry * dht_entry_create(const uint8_t * key) +{ + struct dht_entry * e; + + assert(key != NULL); + + e = malloc(sizeof(*e)); + if (e == NULL) + goto fail_entry; + + list_head_init(&e->next); + llist_init(&e->vals); + llist_init(&e->lvals); + + e->key = dht_dup_key(key); + if (e->key == NULL) + goto fail_key; + + return e; + fail_key: + free(e); + fail_entry: + return NULL; +} + +static void dht_entry_destroy(struct dht_entry * e) +{ + struct list_head * p; + struct list_head * h; + + assert(e != NULL); + + llist_for_each_safe(p, h, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + llist_del(&v->next, &e->vals); + val_entry_destroy(v); + --dht.db.kv.vals; + } + + llist_for_each_safe(p, h, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + llist_del(&v->next, &e->lvals); + val_entry_destroy(v); + --dht.db.kv.lvals; + } + + free(e->key); + + assert(llist_is_empty(&e->vals) && llist_is_empty(&e->lvals)); + + free(e); +} + +static struct val_entry * dht_entry_get_lval(const struct dht_entry * e, + const buffer_t val) +{ + struct list_head * p; + + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + llist_for_each(p, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (bufcmp(&v->val, &val) == 0) + return v; + } + + return NULL; +} + +static struct val_entry * dht_entry_get_val(const struct dht_entry * e, + const buffer_t val) +{ + struct list_head * p; + + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + llist_for_each(p, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (bufcmp(&v->val, &val) == 0) + return v; + + } + + return NULL; +} + +static int dht_entry_update_val(struct dht_entry * e, + buffer_t val, + time_t exp) +{ + struct val_entry * v; + struct timespec now; + + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (exp < now.tv_sec) + return -EINVAL; /* Refuse to add expired values */ + + if (dht_entry_get_lval(e, val) != NULL) { + log_dbg(KV_FMT " Val already in lvals.", KV_VAL(e->key, val)); + return 0; /* Refuse to add local values */ + } + + v = dht_entry_get_val(e, val); + if (v == NULL) { + v = val_entry_create(val, exp); + if (v == NULL) + return -ENOMEM; + + llist_add_tail(&v->next, &e->vals); + ++dht.db.kv.vals; + + return 0; + } + + if (v->t_exp < exp) + v->t_exp = exp; + + return 0; +} + +static int dht_entry_update_lval(struct dht_entry * e, + buffer_t val) +{ + struct val_entry * v; + struct timespec now; + + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + v = dht_entry_get_lval(e, val); + if (v == NULL) { + log_dbg(KV_FMT " Adding lval.", KV_VAL(e->key, val)); + v = val_entry_create(val, now.tv_sec + dht.t_exp); + if (v == NULL) + return -ENOMEM; + + llist_add_tail(&v->next, &e->lvals); + ++dht.db.kv.lvals; + + return 0; + } + + return 0; +} + +static int dht_entry_remove_lval(struct dht_entry * e, + buffer_t val) +{ + struct val_entry * v; + + assert(e != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + v = dht_entry_get_lval(e, val); + if (v == NULL) + return -ENOENT; + + log_dbg(KV_FMT " Removing lval.", KV_VAL(e->key, val)); + + llist_del(&v->next, &e->lvals); + val_entry_destroy(v); + --dht.db.kv.lvals; + + return 0; +} + +#define IS_EXPIRED(v, now) ((now)->tv_sec > (v)->t_exp) +static void dht_entry_remove_expired_vals(struct dht_entry * e) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + + assert(e != NULL); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + llist_for_each_safe(p, h, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (!IS_EXPIRED(v, &now)) + continue; + + log_dbg(KV_FMT " Value expired." , KV_VAL(e->key, v->val)); + llist_del(&v->next, &e->vals); + val_entry_destroy(v); + --dht.db.kv.vals; + } +} + +static struct dht_entry * __dht_kv_find_entry(const uint8_t * key) +{ + struct list_head * p; + + assert(key != NULL); + + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + if (!memcmp(key, e->key, dht.id.len)) + return e; + } + + return NULL; +} + +static void dht_kv_remove_expired_entries(void) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_rwlock_wrlock(&dht.db.lock); + + llist_for_each_safe(p, h, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + dht_entry_remove_expired_vals(e); + if (e->lvals.len > 0 || e->vals.len > 0) + continue; + + log_dbg(KEY_FMT " Entry removed. ", KEY_VAL(e->key)); + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + + pthread_rwlock_unlock(&dht.db.lock); +} + + +static struct contact * contact_create(const uint8_t * id, + uint64_t addr) +{ + struct contact * c; + struct timespec t; + + c = malloc(sizeof(*c)); + if (c == NULL) + return NULL; + + list_head_init(&c->next); + + clock_gettime(CLOCK_REALTIME_COARSE, &t); + + c->addr = addr; + c->fails = 0; + c->t_seen = t.tv_sec; + c->id = dht_dup_key(id); + if (c->id == NULL) { + free(c); + return NULL; + } + + return c; +} + +static void contact_destroy(struct contact * c) +{ + assert(c != NULL); + assert(list_is_empty(&c->next)); + + free(c->id); + free(c); +} + +static struct dht_req * dht_req_create(const uint8_t * key) +{ + struct dht_req * req; + struct timespec now; + + assert(key != NULL); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + req = malloc(sizeof(*req)); + if (req == NULL) + goto fail_malloc; + + list_head_init(&req->next); + + req->t_exp = now.tv_sec + DHT_T_RESP; + + llist_init(&req->peers); + + req->key = dht_dup_key(key); + if (req->key == NULL) + goto fail_dup_key; + + llist_init(&req->cache); + + return req; + + fail_dup_key: + free(req); + fail_malloc: + return NULL; +} + +static void dht_req_destroy(struct dht_req * req) +{ + struct list_head * p; + struct list_head * h; + + assert(req); + assert(req->key); + + llist_for_each_safe(p, h, &req->peers) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + llist_del(&e->next, &req->peers); + free(e->id); + free(e); + } + + llist_for_each_safe(p, h, &req->cache) { + struct val_entry * e = list_entry(p, struct val_entry, next); + llist_del(&e->next, &req->cache); + val_entry_destroy(e); + } + + free(req->key); + + assert(llist_is_empty(&req->peers)); + + free(req); +} + +static struct peer_entry * dht_req_get_peer(struct dht_req * req, + const struct peer_entry * e) +{ + struct list_head * p; + + llist_for_each(p, &req->peers) { + struct peer_entry * x = list_entry(p, struct peer_entry, next); + if (x->addr == e->addr) + return x; + } + + return NULL; +} + +#define IS_MAGIC(peer) ((peer)->cookie == dht.magic) +static int dht_req_add_peer(struct dht_req * req, + const struct peer_entry * e) +{ + struct peer_entry * x; /* existing */ + struct list_head * p; /* iterator */ + size_t pos = 0; + + assert(req != NULL); + assert(e != NULL); + assert(e->id != NULL); + + /* + * Dedupe messages to the same peer, unless + * 1) The previous was FIND_NODE and now it's FIND_VALUE + * 2) We urgently need contacts (magic cookie) + */ + x = dht_req_get_peer(req, e); + if (x != NULL && x->code >= e->code && !IS_MAGIC(e)) + return -1; + + /* Find how this contact ranks in distance to the key */ + llist_for_each(p, &req->peers) { + struct peer_entry * y; + y = list_entry(p, struct peer_entry, next); + if (IS_CLOSER(y->id, e->id)) { + pos++; + continue; + } + break; + } + + if (pos >= dht.alpha && IS_MAGIC(e)) + return -1; + + x = malloc(sizeof(*x)); + if (x == NULL) { + log_err("Failed to malloc peer entry."); + return -1; + } + + x->cookie = e->cookie; + x->addr = e->addr; + x->code = e->code; + x->t_sent = e->t_sent; + x->id = dht_dup_key(e->id); + if (x->id == NULL) { + log_err("Failed to dup peer ID."); + free(x); + return -1; + } + + if (IS_MAGIC(e)) + llist_add_at(&x->next, p, &req->peers); + else + llist_add_tail_at(&x->next, p, &req->peers); + + return 0; +} + +static size_t dht_req_add_peers(struct dht_req * req, + struct list_head * pl) +{ + struct list_head * p; + struct list_head * h; + size_t n = 0; + + assert(req != NULL); + assert(pl != NULL); + + list_for_each_safe(p, h, pl) { + struct peer_entry * e; + e = list_entry(p, struct peer_entry, next); + if (dht_req_add_peer(req, e) < 0) { + list_del(&e->next); + free(e->id); + free(e); + } + } + + return n; +} + +static bool dht_req_has_peer(struct dht_req * req, + uint64_t cookie) +{ + struct list_head * p; + + assert(req != NULL); + + llist_for_each(p, &req->peers) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + if (e->cookie == cookie) + return true; + } + + return false; +} + +static void peer_list_destroy(struct list_head * pl) +{ + struct list_head * p; + struct list_head * h; + + assert(pl != NULL); + + list_for_each_safe(p, h, pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + list_del(&e->next); + free(e->id); + free(e); + } +} + +static int dht_kv_create_peer_list(struct list_head * cl, + struct list_head * pl, + enum dht_code code) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + size_t len; + + assert(cl != NULL); + assert(pl != NULL); + assert(list_is_empty(pl)); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + len = 0; + + list_for_each_safe(p, h, cl) { + struct contact * c = list_entry(p, struct contact, next); + struct peer_entry * e; + if (len++ == dht.alpha) + break; + + e = malloc(sizeof(*e)); + if (e == NULL) + return -ENOMEM; + + e->cookie = generate_cookie(); + e->code = code; + e->addr = c->addr; + e->t_sent = now.tv_sec; + + e->id = c->id; + + list_add_tail(&e->next, pl); + + list_del(&c->next); + c->id = NULL; /* we stole the id */ + contact_destroy(c); + } + + return 0; +} + +static struct dht_req * __dht_kv_req_get_req(const uint8_t * key) +{ + struct list_head * p; + + llist_for_each(p, &dht.reqs.ll) { + struct dht_req * r = list_entry(p, struct dht_req, next); + if (memcmp(r->key, key, dht.id.len) == 0) + return r; + } + + return NULL; +} + +static struct dht_req * __dht_kv_get_req_cache(const uint8_t * key) +{ + struct dht_req * req; + + assert(key != NULL); + + req = __dht_kv_req_get_req(key); + if (req == NULL) + return NULL; + + if (llist_is_empty(&req->cache)) + return NULL; + + return req; +} + +static void __dht_kv_req_remove(const uint8_t * key) +{ + struct dht_req * req; + + assert(key != NULL); + + req = __dht_kv_req_get_req(key); + if (req == NULL) + return; + + llist_del(&req->next, &dht.reqs.ll); + + dht_req_destroy(req); +} + +static struct dht_req * __dht_kv_get_req_peer(const uint8_t * key, + uint64_t cookie) +{ + struct dht_req * req; + + assert(key != NULL); + + req = __dht_kv_req_get_req(key); + if (req == NULL) + return NULL; + + if (!dht_req_has_peer(req, cookie)) + return NULL; + + return req; +} + +static bool dht_kv_has_req(const uint8_t * key, + uint64_t cookie) +{ + bool found; + + pthread_mutex_lock(&dht.reqs.mtx); + + found = __dht_kv_get_req_peer(key, cookie) != NULL; + + pthread_mutex_unlock(&dht.reqs.mtx); + + return found; +} + +/* + * This will filter the peer list for addresses that still need to be + * contacted. + */ +static int dht_kv_update_req(const uint8_t * key, + struct list_head * pl) +{ + struct dht_req * req; + struct timespec now; + + assert(key != NULL); + assert(pl != NULL); + assert(!list_is_empty(pl)); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + pthread_mutex_lock(&dht.reqs.mtx); + + req = __dht_kv_req_get_req(key); + if (req == NULL) { + if (dht.reqs.ll.len == DHT_MAX_REQS) { + log_err(KEY_FMT " Max reqs reached (%zu).", + KEY_VAL(key), dht.reqs.ll.len); + peer_list_destroy(pl); + goto fail_req; + } + req = dht_req_create(key); + if (req == NULL) { + log_err(KEY_FMT "Failed to create req.", KEY_VAL(key)); + goto fail_req; + } + llist_add_tail(&req->next, &dht.reqs.ll); + } + + if (req->cache.len > 0) /* Already have values */ + peer_list_destroy(pl); + + dht_req_add_peers(req, pl); + req->t_exp = now.tv_sec + DHT_T_RESP; + + if (dht.reqs.ll.len > DHT_WARN_REQS) { + log_warn("Number of outstanding requests (%zu) exceeds %u.", + dht.reqs.ll.len, DHT_WARN_REQS); + } + + pthread_mutex_unlock(&dht.reqs.mtx); + + return 0; + fail_req: + pthread_mutex_unlock(&dht.reqs.mtx); + return -1; +} + +static int dht_kv_respond_req(uint8_t * key, + binary_data_t * vals, + size_t len) +{ + struct dht_req * req; + struct timespec now; + size_t i; + + assert(key != NULL); + assert(vals != NULL); + assert(len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_mutex_lock(&dht.reqs.mtx); + + req = __dht_kv_req_get_req(key); + if (req == NULL) { + log_dbg(KEY_FMT " Failed to find req.", KEY_VAL(key)); + goto fail_req; + } + + for (i = 0; i < len; ++i) { + struct val_entry * e; + buffer_t val; + val.data = vals[i].data; + val.len = vals[i].len; + e = val_entry_create(val, now.tv_sec + DHT_T_CACHE); + if (e == NULL) { + log_err(" Failed to create val_entry."); + continue; + } + + llist_add_tail(&e->next, &req->cache); + } + + pthread_cond_broadcast(&dht.reqs.cond); + + pthread_mutex_unlock(&dht.reqs.mtx); + + return 0; + fail_req: + pthread_mutex_unlock(&dht.reqs.mtx); + return -1; +} + +static ssize_t dht_kv_wait_req(const uint8_t * key, + buffer_t ** vals) +{ + struct list_head * p; + struct dht_req * req; + struct timespec t; +#ifdef __DHT_TEST__ + struct timespec intv = TIMESPEC_INIT_MS(10); +#else + struct timespec intv = TIMESPEC_INIT_S(DHT_T_RESP); +#endif + size_t max; + size_t i = 0; + int ret = 0; + + assert(key != NULL); + assert(vals != NULL); + + clock_gettime(PTHREAD_COND_CLOCK, &t); + + ts_add(&t, &intv, &t); + + pthread_mutex_lock(&dht.reqs.mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, &dht.reqs.mtx); + + while ((req = __dht_kv_get_req_cache(key)) == NULL) { + ret = pthread_cond_timedwait(&dht.reqs.cond, &dht.reqs.mtx, &t); + if (ret == ETIMEDOUT) + break; + } + + pthread_cleanup_pop(false); + + if (ret == ETIMEDOUT) { + log_warn(KEY_FMT " Req timed out.", KEY_VAL(key)); + __dht_kv_req_remove(key); + goto timedout; + } + + max = MIN(req->cache.len, DHT_MAX_VALS); + if (max == 0) + goto no_vals; + + *vals = malloc(max * sizeof(**vals)); + if (*vals == NULL) { + log_err(KEY_FMT "Failed to malloc val buffer.", KEY_VAL(key)); + goto fail_vals; + } + + memset(*vals, 0, max * sizeof(**vals)); + + llist_for_each(p, &req->cache) { + struct val_entry * v; + if (i == max) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + + pthread_mutex_unlock(&dht.reqs.mtx); + + return i; + no_vals: + pthread_mutex_unlock(&dht.reqs.mtx); + *vals = NULL; + return 0; + fail_val_data: + freebufs(*vals, i); + fail_vals: + pthread_mutex_unlock(&dht.reqs.mtx); + return -ENOMEM; + timedout: + pthread_mutex_unlock(&dht.reqs.mtx); + return -ETIMEDOUT; +} + +static struct bucket * iter_bucket(struct bucket * b, + const uint8_t * id) +{ + uint8_t byte; + uint8_t mask; + + assert(b != NULL); + + if (b->children[0] == NULL) + return b; + + byte = id[(b->depth * DHT_BETA) / CHAR_BIT]; + + mask = ((1L << DHT_BETA) - 1) & 0xFF; + + byte >>= (CHAR_BIT - DHT_BETA) - + (((b->depth) * DHT_BETA) & (CHAR_BIT - 1)); + + return iter_bucket(b->children[(byte & mask)], id); +} + +static struct bucket * __dht_kv_get_bucket(const uint8_t * id) +{ + assert(dht.db.contacts.root != NULL); + + return iter_bucket(dht.db.contacts.root, id); +} + +static void contact_list_add(struct list_head * l, + struct contact * c) +{ + struct list_head * p; + + assert(l != NULL); + assert(c != NULL); + + list_for_each(p, l) { + struct contact * e = list_entry(p, struct contact, next); + if (IS_CLOSER(e->id, c->id)) + continue; + } + + list_add_tail(&c->next, p); +} + +static ssize_t dht_kv_contact_list(const uint8_t * key, + struct list_head * l, + size_t max) +{ + struct list_head * p; + struct bucket * b; + struct timespec t; + size_t i; + size_t len = 0; + + assert(l != NULL); + assert(key != NULL); + assert(list_is_empty(l)); + + clock_gettime(CLOCK_REALTIME_COARSE, &t); + + max = MIN(max, dht.k); + + pthread_rwlock_rdlock(&dht.db.lock); + + b = __dht_kv_get_bucket(key); + if (b == NULL) { + log_err(KEY_FMT " Failed to get bucket.", KEY_VAL(key)); + goto fail_bucket; + } + + b->t_refr = t.tv_sec + dht.t_refr; + + if (b->contacts.len == dht.k || b->parent == NULL) { + llist_for_each(p, &b->contacts) { + struct contact * c; + struct contact * d; + c = list_entry(p, struct contact, next); + if (c->addr == dht.addr) + continue; + d = contact_create(c->id, c->addr); + if (d == NULL) + continue; + contact_list_add(l, d); + if (++len == max) + break; + } + } else { + struct bucket * d = b->parent; + for (i = 0; i < (1L << DHT_BETA) && len < dht.k; ++i) { + llist_for_each(p, &d->children[i]->contacts) { + struct contact * c; + struct contact * d; + c = list_entry(p, struct contact, next); + if (c->addr == dht.addr) + continue; + d = contact_create(c->id, c->addr); + if (d == NULL) + continue; + contact_list_add(l, d); + if (++len == max) + break; + } + } + } + + pthread_rwlock_unlock(&dht.db.lock); + + return len; + fail_bucket: + pthread_rwlock_unlock(&dht.db.lock); + return -1; +} + +static void contact_list_destroy(struct list_head * l) +{ + struct list_head * p; + struct list_head * h; + + assert(l != NULL); + + list_for_each_safe(p, h, l) { + struct contact * c = list_entry(p, struct contact, next); + list_del(&c->next); + contact_destroy(c); + } +} + +static ssize_t dht_kv_get_contacts(const uint8_t * key, + dht_contact_msg_t *** msgs) +{ + struct list_head cl; + struct list_head * p; + struct list_head * h; + size_t len; + size_t i = 0; + + assert(key != NULL); + assert(msgs != NULL); + + list_head_init(&cl); + + len = dht_kv_contact_list(key, &cl, dht.k); + if (len == 0) { + *msgs = NULL; + return 0; + } + + *msgs = malloc(len * sizeof(**msgs)); + if (*msgs == NULL) + goto fail_msgs; + + list_for_each_safe(p, h, &cl) { + struct contact * c; + (*msgs)[i] = malloc(sizeof(***msgs)); + if ((*msgs)[i] == NULL) + goto fail_contact; + + dht_contact_msg__init((*msgs)[i]); + c = list_entry(p, struct contact, next); + list_del(&c->next); + (*msgs)[i]->id.data = c->id; + (*msgs)[i]->id.len = dht.id.len; + (*msgs)[i++]->addr = c->addr; + free(c); + } + + return i; + fail_contact: + while (i-- > 0) + dht_contact_msg__free_unpacked((*msgs)[i], NULL); + + free(*msgs); + *msgs = NULL; + fail_msgs: + contact_list_destroy(&cl); + return -ENOMEM; +} + +/* Build a refresh list. */ +static void __dht_kv_bucket_refresh_list(struct bucket * b, + time_t t, + struct list_head * r) +{ + struct contact * c; + struct contact * d; + + assert(b != NULL); + + if (t < b->t_refr) + return; + + if (*b->children != NULL) { + size_t i; + for (i = 0; i < (1L << DHT_BETA); ++i) + __dht_kv_bucket_refresh_list(b->children[i], t, r); + } + + if (llist_is_empty(&b->contacts)) + return; + + c = llist_first_entry(&b->contacts, struct contact, next); + if (t > c->t_seen + dht.t_refr) { + d = contact_create(c->id, c->addr); + if (d != NULL) + list_add(&d->next, r); + } +} + +static struct bucket * bucket_create(void) +{ + struct bucket * b; + struct timespec t; + size_t i; + + b = malloc(sizeof(*b)); + if (b == NULL) + return NULL; + + llist_init(&b->contacts); + + llist_init(&b->alts); + + clock_gettime(CLOCK_REALTIME_COARSE, &t); + b->t_refr = t.tv_sec + dht.t_refr; + + for (i = 0; i < (1L << DHT_BETA); ++i) + b->children[i] = NULL; + + b->parent = NULL; + b->depth = 0; + b->mask = 0; + + return b; +} + +static void bucket_destroy(struct bucket * b) +{ + struct list_head * p; + struct list_head * h; + size_t i; + + assert(b != NULL); + + for (i = 0; i < (1L << DHT_BETA); ++i) + if (b->children[i] != NULL) + bucket_destroy(b->children[i]); + + llist_for_each_safe(p, h, &b->contacts) { + struct contact * c = list_entry(p, struct contact, next); + llist_del(&c->next, &b->contacts); + contact_destroy(c); + } + + llist_for_each_safe(p, h, &b->alts) { + struct contact * c = list_entry(p, struct contact, next); + llist_del(&c->next, &b->alts); + contact_destroy(c); + } + + free(b); +} + +static bool bucket_has_id(struct bucket * b, + const uint8_t * id) +{ + uint8_t mask; + uint8_t byte; + + if (b->depth == 0) + return true; + + byte = id[(b->depth * DHT_BETA) / CHAR_BIT]; + + mask = ((1L << DHT_BETA) - 1) & 0xFF; + + byte >>= (CHAR_BIT - DHT_BETA) - + (((b->depth - 1) * DHT_BETA) & (CHAR_BIT - 1)); + + return ((byte & mask) == b->mask); +} + +static int move_contacts(struct bucket * b, + struct bucket * c) +{ + struct list_head * p; + struct list_head * h; + struct contact * d; + + assert(b != NULL); + assert(c != NULL); + + llist_for_each_safe(p, h, &b->contacts) { + d = list_entry(p, struct contact, next); + if (bucket_has_id(c, d->id)) { + llist_del(&d->next, &b->contacts); + llist_add_tail(&d->next, &c->contacts); + } + } + + return 0; +} + +static int split_bucket(struct bucket * b) +{ + uint8_t mask = 0; + size_t i; + size_t b_len; + + assert(b); + assert(llist_is_empty(&b->alts)); + assert(!llist_is_empty(&b->contacts)); + assert(b->children[0] == NULL); + + b_len = b->contacts.len; + + for (i = 0; i < (1L << DHT_BETA); ++i) { + b->children[i] = bucket_create(); + if (b->children[i] == NULL) + goto fail_child; + + b->children[i]->depth = b->depth + 1; + b->children[i]->mask = mask; + b->children[i]->parent = b; + + move_contacts(b, b->children[i]); + + mask++; + } + + for (i = 0; i < (1L << DHT_BETA); ++i) + if (b->children[i]->contacts.len == b_len) + split_bucket(b->children[i]); + + return 0; + fail_child: + while (i-- > 0) + bucket_destroy(b->children[i]); + + return -1; +} + +static int dht_kv_update_contacts(const uint8_t * id, + uint64_t addr) +{ + struct list_head * p; + struct list_head * h; + struct bucket * b; + struct contact * c; + + assert(id != NULL); + assert(addr != INVALID_ADDR); + + pthread_rwlock_wrlock(&dht.db.lock); + + b = __dht_kv_get_bucket(id); + if (b == NULL) { + log_err(PEER_FMT " Failed to get bucket.", PEER_VAL(id, addr)); + goto fail_update; + } + + c = contact_create(id, addr); + if (c == NULL) { + log_err(PEER_FMT " Failed to create contact.", + PEER_VAL(id, addr)); + goto fail_update; + } + + llist_for_each_safe(p, h, &b->contacts) { + struct contact * d = list_entry(p, struct contact, next); + if (d->addr == addr) { + llist_del(&d->next, &b->contacts); + contact_destroy(d); + } + } + + if (b->contacts.len == dht.k) { + if (bucket_has_id(b, dht.id.data)) { + llist_add_tail(&c->next, &b->contacts); + if (split_bucket(b)) { + llist_del(&c->next, &b->contacts); + contact_destroy(c); + } + } else if (b->alts.len == dht.k) { + struct contact * d; + d = llist_first_entry(&b->alts, + struct contact, next); + llist_del(&d->next, &b->alts); + contact_destroy(d); + llist_add_tail(&c->next, &b->alts); + } else { + llist_add_tail(&c->next, &b->alts); + } + } else { + llist_add_tail(&c->next, &b->contacts); + } + + pthread_rwlock_unlock(&dht.db.lock); + + return 0; + fail_update: + pthread_rwlock_unlock(&dht.db.lock); + return -1; +} + +static time_t gcd(time_t a, + time_t b) +{ + if (a == 0) + return b; + + return gcd(b % a, a); +} + +static dht_contact_msg_t * dht_kv_src_contact_msg(void) +{ + dht_contact_msg_t * src; + + src = malloc(sizeof(*src)); + if (src == NULL) + goto fail_malloc; + + dht_contact_msg__init(src); + + src->id.data = dht_dup_key(dht.id.data); + if (src->id.data == NULL) + goto fail_id; + + src->id.len = dht.id.len; + src->addr = dht.addr; + + return src; + fail_id: + dht_contact_msg__free_unpacked(src, NULL); + fail_malloc: + return NULL; +} + +static dht_msg_t * dht_kv_find_req_msg(const uint8_t * key, + enum dht_code code) +{ + dht_msg_t * msg; + + assert(key != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + dht_msg__init(msg); + msg->code = code; + + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; + + msg->find = malloc(sizeof(*msg->find)); + if (msg->find == NULL) + goto fail_msg; + + dht_find_req_msg__init(msg->find); + + msg->find->key.data = dht_dup_key(key); + if (msg->find->key.data == NULL) + goto fail_msg; + + msg->find->key.len = dht.id.len; + msg->find->cookie = DHT_INVALID; + + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +static dht_msg_t * dht_kv_find_node_req_msg(const uint8_t * key) +{ + return dht_kv_find_req_msg(key, DHT_FIND_NODE_REQ); +} + +static dht_msg_t * dht_kv_find_value_req_msg(const uint8_t * key) +{ + return dht_kv_find_req_msg(key, DHT_FIND_VALUE_REQ); +} + +static dht_msg_t * dht_kv_find_node_rsp_msg(uint8_t * key, + uint64_t cookie, + dht_contact_msg_t *** contacts, + size_t len) +{ + dht_msg_t * msg; + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + dht_msg__init(msg); + msg->code = DHT_FIND_NODE_RSP; + + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; + + msg->node = malloc(sizeof(*msg->node)); + if (msg->node == NULL) + goto fail_msg; + + dht_find_node_rsp_msg__init(msg->node); + + msg->node->key.data = dht_dup_key(key); + if (msg->node->key.data == NULL) + goto fail_msg; + + msg->node->cookie = cookie; + msg->node->key.len = dht.id.len; + msg->node->n_contacts = len; + if (len != 0) { /* Steal the ptr */ + msg->node->contacts = *contacts; + *contacts = NULL; + } + + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +static dht_msg_t * dht_kv_find_value_rsp_msg(uint8_t * key, + uint64_t cookie, + dht_contact_msg_t *** contacts, + size_t n_contacts, + buffer_t ** vals, + size_t n_vals) +{ + dht_msg_t * msg; + + msg = dht_kv_find_node_rsp_msg(key, cookie, contacts, n_contacts); + if (msg == NULL) + goto fail_node_rsp; + + msg->code = DHT_FIND_VALUE_RSP; + + msg->val = malloc(sizeof(*msg->val)); + if (msg->val == NULL) + goto fail_msg; + + dht_find_value_rsp_msg__init(msg->val); + + msg->val->n_values = n_vals; + if (n_vals != 0) /* Steal the ptr */ + msg->val->values = (binary_data_t *) *vals; + + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_node_rsp: + return NULL; +} + +static dht_msg_t * dht_kv_store_msg(const uint8_t * key, + const buffer_t val, + time_t exp) +{ + dht_msg_t * msg; + + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + dht_msg__init(msg); + + msg->code = DHT_STORE; + + msg->src = dht_kv_src_contact_msg(); + if (msg->src == NULL) + goto fail_msg; + + msg->store = malloc(sizeof(*msg->store)); + if (msg->store == NULL) + goto fail_msg; + + dht_store_msg__init(msg->store); + + msg->store->key.data = dht_dup_key(key); + if (msg->store->key.data == NULL) + goto fail_msg; + + msg->store->key.len = dht.id.len; + msg->store->val.data = malloc(val.len); + if (msg->store->val.data == NULL) + goto fail_msg; + + memcpy(msg->store->val.data, val.data, val.len); + + msg->store->val.len = val.len; + msg->store->exp = exp; + + return msg; + + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +static ssize_t dht_kv_retrieve(const uint8_t * key, + buffer_t ** vals) +{ + struct dht_entry * e; + struct list_head * p; + size_t n; + size_t i; + + assert(key != NULL); + + pthread_rwlock_rdlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); + if (e == NULL) + goto no_vals; + + n = MIN(DHT_MAX_VALS, e->vals.len + e->lvals.len); + if (n == 0) + goto no_vals; + + *vals = malloc(n * sizeof(**vals)); + if (*vals == NULL) + goto fail_vals; + + memset(*vals, 0, n * sizeof(**vals)); + + i = 0; + + llist_for_each(p, &e->vals) { + struct val_entry * v; + if (i == n) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + + llist_for_each(p, &e->lvals) { + struct val_entry * v; + if (i == n) + break; /* We have enough values */ + v = list_entry(p, struct val_entry, next); + (*vals)[i].data = malloc(v->val.len); + if ((*vals)[i].data == NULL) + goto fail_val_data; + + (*vals)[i].len = v->val.len; + memcpy((*vals)[i++].data, v->val.data, v->val.len); + } + + pthread_rwlock_unlock(&dht.db.lock); + + return (ssize_t) i; + + fail_val_data: + pthread_rwlock_unlock(&dht.db.lock); + freebufs(*vals, i); + *vals = NULL; + return -ENOMEM; + fail_vals: + pthread_rwlock_unlock(&dht.db.lock); + return -ENOMEM; + no_vals: + pthread_rwlock_unlock(&dht.db.lock); + *vals = NULL; + return 0; +} + +static void __cleanup_dht_msg(void * msg) +{ + dht_msg__free_unpacked((dht_msg_t *) msg, NULL); +} + +#ifdef DEBUG_PROTO_DHT +static void dht_kv_debug_msg(dht_msg_t * msg) +{ + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + time_t stamp; + size_t i; + + if (msg == NULL) + return; + + pthread_cleanup_push(__cleanup_dht_msg, msg); + + switch (msg->code) { + case DHT_STORE: + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->store->key.data), + msg->store->key.len); + log_proto(" val: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->store->val.data), + msg->store->val.len); + stamp = msg->store->exp; + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + log_proto(" exp: %s.", tmstr); + break; + case DHT_FIND_NODE_REQ: + /* FALLTHRU */ + case DHT_FIND_VALUE_REQ: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->find->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->find->key.data), + msg->find->key.len); + break; + case DHT_FIND_VALUE_RSP: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->node->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->node->key.data), + msg->node->key.len); + log_proto(" values: [%zd]", msg->val->n_values); + for (i = 0; i < msg->val->n_values; i++) + log_proto(" " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->val->values[i].data), + msg->val->values[i].len); + log_proto(" contacts: [%zd]", msg->node->n_contacts); + for (i = 0; i < msg->node->n_contacts; i++) { + dht_contact_msg_t * c = msg->node->contacts[i]; + log_proto(" " PEER_FMT, + PEER_VAL(c->id.data, c->addr)); + } + break; + case DHT_FIND_NODE_RSP: + log_proto(" cookie: " HASH_FMT64, + HASH_VAL64(&msg->node->cookie)); + log_proto(" key: " HASH_FMT64 " [%zu bytes]", + HASH_VAL64(msg->node->key.data), msg->node->key.len); + log_proto(" contacts: [%zd]", msg->node->n_contacts); + for (i = 0; i < msg->node->n_contacts; i++) { + dht_contact_msg_t * c = msg->node->contacts[i]; + log_proto(" " PEER_FMT, + PEER_VAL(c->id.data, c->addr)); + } + + break; + default: + break; + } + + pthread_cleanup_pop(false); +} + +static void dht_kv_debug_msg_snd(dht_msg_t * msg, + uint8_t * id, + uint64_t addr) +{ + if (msg == NULL) + return; + + log_proto(TX_HDR_FMT ".", TX_HDR_VAL(msg, id, addr)); + + dht_kv_debug_msg(msg); +} + +static void dht_kv_debug_msg_rcv(dht_msg_t * msg) +{ + if (msg == NULL) + return; + + log_proto(RX_HDR_FMT ".", RX_HDR_VAL(msg)); + + dht_kv_debug_msg(msg); +} +#endif + +#ifndef __DHT_TEST__ +static int dht_send_msg(dht_msg_t * msg, + uint64_t addr) +{ + size_t len; + struct ssm_pk_buff * spb; + + if (msg == NULL) + return 0; + + assert(addr != INVALID_ADDR && addr != dht.addr); + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + log_warn("%s failed to pack.", DHT_CODE(msg)); + goto fail_msg; + } + + if (ipcp_spb_reserve(&spb, len)) { + log_warn("%s failed to get spb.", DHT_CODE(msg)); + goto fail_msg; + } + + dht_msg__pack(msg, ssm_pk_buff_head(spb)); + + if (dt_write_packet(addr, QOS_CUBE_BE, dht.eid, spb, NULL) < 0) { + log_warn("%s write failed", DHT_CODE(msg)); + goto fail_send; + } + + return 0; + fail_send: + ipcp_spb_release(spb); + fail_msg: + return -1; +} +#else /* funtion for testing */ +static int dht_send_msg(dht_msg_t * msg, + uint64_t addr) +{ + buffer_t buf; + + assert(msg != NULL); + assert(addr != INVALID_ADDR && addr != dht.addr); + + buf.len = dht_msg__get_packed_size(msg); + if (buf.len == 0) { + log_warn("%s failed to pack.", DHT_CODE(msg)); + goto fail_msg; + } + + buf.data = malloc(buf.len); + if (buf.data == NULL) { + log_warn("%s failed to malloc buf.", DHT_CODE(msg)); + goto fail_msg; + } + + dht_msg__pack(msg, buf.data); + + if (sink_send_msg(&buf, addr) < 0) { + log_warn("%s write failed", DHT_CODE(msg)); + goto fail_send; + } + + return 0; + fail_send: + freebuf(buf); + fail_msg: + return -1; +} +#endif /* __DHT_TEST__ */ + +static void __cleanup_peer_list(void * pl) +{ + struct list_head * p; + struct list_head * h; + + assert(pl != NULL); + + list_for_each_safe(p, h, (struct list_head *) pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + list_del(&e->next); + free(e->id); + free(e); + } +} + + +static int dht_kv_send_msgs(dht_msg_t * msg, + struct list_head * pl) +{ + struct list_head * p; + struct list_head * h; + + pthread_cleanup_push(__cleanup_dht_msg, msg); + pthread_cleanup_push(__cleanup_peer_list, pl); + + list_for_each_safe(p, h, pl) { + struct peer_entry * e = list_entry(p, struct peer_entry, next); + if (IS_REQUEST(msg->code)) { + msg->find->cookie = e->cookie; + assert(msg->find->cookie != DHT_INVALID); + } + if (dht_send_msg(msg, e->addr) < 0) + continue; + +#ifdef DEBUG_PROTO_DHT + dht_kv_debug_msg_snd(msg, e->id, e->addr); +#endif + list_del(&e->next); + free(e->id); + free(e); + } + + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + return list_is_empty(pl) ? 0 : -1; +} + +static int dht_kv_get_peer_list_for_msg(dht_msg_t * msg, + struct list_head * pl) +{ + struct list_head cl; /* contact list */ + uint8_t * key; /* key in the request */ + size_t max; + + assert(msg != NULL); + + assert(list_is_empty(pl)); + + max = msg->code == DHT_STORE ? dht.k : dht.alpha; + + switch (msg->code) { + case DHT_FIND_NODE_REQ: + /* FALLTHRU */ + case DHT_FIND_VALUE_REQ: + key = msg->find->key.data; + break; + case DHT_STORE: + key = msg->store->key.data; + break; + default: + log_err("Invalid DHT msg code (%d).", msg->code); + return -1; + } + + list_head_init(&cl); + + if (dht_kv_contact_list(key, &cl, max) < 0) { + log_err(KEY_FMT " Failed to get contact list.", KEY_VAL(key)); + goto fail_contacts; + } + + if (list_is_empty(&cl)) { + log_warn(KEY_FMT " No available contacts.", KEY_VAL(key)); + goto fail_contacts; + } + + if (dht_kv_create_peer_list(&cl, pl, msg->code) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peers; + } + + contact_list_destroy(&cl); + return 0; + fail_peers: + contact_list_destroy(&cl); + fail_contacts: + return -1; +} + +static int dht_kv_store_remote(const uint8_t * key, + const buffer_t val, + time_t exp) +{ + dht_msg_t * msg; + struct timespec now; + struct list_head pl; + + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + msg = dht_kv_store_msg(key, val, exp); + if (msg == NULL) { + log_err(KV_FMT " Failed to create %s.", + KV_VAL(key, val), dht_code_str[DHT_STORE]); + goto fail_msg; + } + + list_head_init(&pl); + + if (dht_kv_get_peer_list_for_msg(msg, &pl) < 0) { + log_dbg(KV_FMT " Failed to get peer list.", KV_VAL(key, val)); + goto fail_peer_list; + } + + if (dht_kv_send_msgs(msg, &pl) < 0) { + log_warn(KV_FMT " Failed to send any %s msg.", + KV_VAL(key, val), DHT_CODE(msg)); + goto fail_msgs; + } + + dht_msg__free_unpacked(msg, NULL); + + return 0; + fail_msgs: + peer_list_destroy(&pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; +} + +/* recursive lookup, start with pl NULL */ +static int dht_kv_query_contacts(const uint8_t * key, + struct list_head * pl) +{ + struct list_head p; + + dht_msg_t * msg; + + assert(key != NULL); + + msg = dht_kv_find_node_req_msg(key); + if (msg == NULL) { + log_err(KEY_FMT " Failed to create %s msg.", + KEY_VAL(key), dht_code_str[DHT_FIND_NODE_REQ]); + goto fail_msg; + } + + if (pl == NULL) { + list_head_init(&p); + pl = &p; + } + + if (list_is_empty(pl) && dht_kv_get_peer_list_for_msg(msg, pl) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peer_list; + } + + if (dht_kv_update_req(key, pl) < 0) { + log_warn(KEY_FMT " Failed to update req.", KEY_VAL(key)); + goto fail_update; + } + + if (dht_kv_send_msgs(msg, pl)) { + log_warn(KEY_FMT " Failed to send any %s msg.", + KEY_VAL(key), DHT_CODE(msg)); + goto fail_update; + } + + dht_msg__free_unpacked(msg, NULL); + + return 0; + fail_update: + peer_list_destroy(pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; +} + +/* recursive lookup, start with pl NULL */ +static ssize_t dht_kv_query_remote(const uint8_t * key, + buffer_t ** vals, + struct list_head * pl) +{ + struct list_head p; + dht_msg_t * msg; + + assert(key != NULL); + + msg = dht_kv_find_value_req_msg(key); + if (msg == NULL) { + log_err(KEY_FMT " Failed to create value req.", KEY_VAL(key)); + goto fail_msg; + } + + if (pl == NULL) { + list_head_init(&p); + pl = &p; + } + + if (list_is_empty(pl) && dht_kv_get_peer_list_for_msg(msg, pl) < 0) { + log_warn(KEY_FMT " Failed to get peer list.", KEY_VAL(key)); + goto fail_peer_list; + } + + if (dht_kv_update_req(key, pl) < 0) { + log_err(KEY_FMT " Failed to update request.", KEY_VAL(key)); + goto fail_update; + } + + if (dht_kv_send_msgs(msg, pl)) { + log_warn(KEY_FMT " Failed to send %s msg.", + KEY_VAL(key), DHT_CODE(msg)); + goto fail_update; + } + + dht_msg__free_unpacked(msg, NULL); + + if (vals == NULL) /* recursive lookup, already waiting */ + return 0; + + return dht_kv_wait_req(key, vals); + fail_update: + peer_list_destroy(pl); + fail_peer_list: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + return -1; +} + +static void __add_dht_kv_entry(struct dht_entry * e) +{ + struct list_head * p; + + assert(e != NULL); + + llist_for_each(p, &dht.db.kv.ll) { + struct dht_entry * d = list_entry(p, struct dht_entry, next); + if (IS_CLOSER(d->key, e->key)) + continue; + break; + } + + llist_add_tail_at(&e->next, p, &dht.db.kv.ll); +} + +/* incoming store message */ +static int dht_kv_store(const uint8_t * key, + const buffer_t val, + time_t exp) +{ + struct dht_entry * e; + bool new = false; + + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); + if (e == NULL) { + log_dbg(KV_FMT " Adding entry (store).", KV_VAL(key, val)); + e = dht_entry_create(key); + if (e == NULL) + goto fail; + + new = true; + + __add_dht_kv_entry(e); + } + + if (dht_entry_update_val(e, val, exp) < 0) + goto fail_add; + + pthread_rwlock_unlock(&dht.db.lock); + + return 0; + fail_add: + if (new) { + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + fail: + pthread_rwlock_unlock(&dht.db.lock); + return -1; +} + +static int dht_kv_publish(const uint8_t * key, + const buffer_t val) +{ + struct dht_entry * e; + struct timespec now; + bool new = false; + + assert(key != NULL); + assert(val.data != NULL); + assert(val.len > 0); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); + if (e == NULL) { + log_dbg(KV_FMT " Adding entry (publish).", KV_VAL(key, val)); + e = dht_entry_create(key); + if (e == NULL) + goto fail; + + __add_dht_kv_entry(e); + new = true; + } + + if (dht_entry_update_lval(e, val) < 0) + goto fail_add; + + pthread_rwlock_unlock(&dht.db.lock); + + dht_kv_store_remote(key, val, now.tv_sec + dht.t_exp); + + return 0; + fail_add: + if (new) { + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + fail: + pthread_rwlock_unlock(&dht.db.lock); + return -1; +} + +static int dht_kv_unpublish(const uint8_t * key, + const buffer_t val) +{ + struct dht_entry * e; + int rc; + + assert(key != NULL); + + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); + if (e == NULL) + goto no_entry; + + rc = dht_entry_remove_lval(e, val); + + pthread_rwlock_unlock(&dht.db.lock); + + return rc; + no_entry: + pthread_rwlock_unlock(&dht.db.lock); + return -ENOENT; + +} + +/* message validation */ +static int dht_kv_validate_store_msg(const dht_store_msg_t * store) +{ + if (store == NULL) { + log_warn("Store in msg is NULL."); + return -EINVAL; + } + + if (store->key.data == NULL || store->key.len == 0) { + log_warn("Invalid key in DHT store msg."); + return -EINVAL; + } + + if (store->key.len != dht.id.len) { + log_warn("Invalid key length in DHT store msg."); + return -EINVAL; + } + + if (store->val.data == NULL || store->val.len == 0) { + log_warn("Invalid value in DHT store msg."); + return -EINVAL; + } + + return 0; +} + +static int validate_find_req_msg(const dht_find_req_msg_t * req) +{ + if (req == NULL) { + log_warn("Request in msg is NULL."); + return -EINVAL; + } + + if (req->key.data == NULL || req->key.len == 0) { + log_warn("Find request without key."); + return -EINVAL; + } + + if (req->key.len != dht.id.len) { + log_warn("Invalid key length in request msg."); + return -EINVAL; + } + + return 0; +} + +static int validate_node_rsp_msg(const dht_find_node_rsp_msg_t * rsp) +{ + if (rsp == NULL) { + log_warn("Node rsp in msg is NULL."); + return -EINVAL; + } + + if (rsp->key.data == NULL) { + log_warn("Invalid key in DHT response msg."); + return -EINVAL; + } + + if (rsp->key.len != dht.id.len) { + log_warn("Invalid key length in DHT response msg."); + return -EINVAL; + } + + if (!dht_kv_has_req(rsp->key.data, rsp->cookie)) { + log_warn(KEY_FMT " No request " CK_FMT ".", + KEY_VAL(rsp->key.data), CK_VAL(rsp->cookie)); + + return -EINVAL; + } + + return 0; +} + +static int validate_value_rsp_msg(const dht_find_value_rsp_msg_t * rsp) +{ + if (rsp == NULL) { + log_warn("Invalid DHT find value response msg."); + return -EINVAL; + } + + if (rsp->values == NULL && rsp->n_values > 0) { + log_dbg("No values in DHT response msg."); + return 0; + } + + if (rsp->n_values == 0 && rsp->values != NULL) { + log_dbg("DHT response did not set values NULL."); + return 0; + } + + return 0; +} + +static int dht_kv_validate_msg(dht_msg_t * msg) +{ + + assert(msg != NULL); + + if (msg->src->id.len != dht.id.len) { + log_warn("%s Invalid source contact ID.", DHT_CODE(msg)); + return -EINVAL; + } + + if (msg->src->addr == INVALID_ADDR) { + log_warn("%s Invalid source address.", DHT_CODE(msg)); + return -EINVAL; + } + + switch (msg->code) { + case DHT_FIND_VALUE_REQ: + /* FALLTHRU */ + case DHT_FIND_NODE_REQ: + if (validate_find_req_msg(msg->find) < 0) + return -EINVAL; + break; + case DHT_FIND_VALUE_RSP: + if (validate_value_rsp_msg(msg->val) < 0) + return -EINVAL; + /* FALLTHRU */ + case DHT_FIND_NODE_RSP: + if (validate_node_rsp_msg(msg->node) < 0) + return -EINVAL; + break; + case DHT_STORE: + if (dht_kv_validate_store_msg(msg->store) < 0) + return -EINVAL; + break; + default: + log_warn("Invalid DHT msg code (%d).", msg->code); + return -ENOENT; + } + + return 0; +} + +static void do_dht_kv_store(const dht_store_msg_t * store) +{ + struct tm * tm; + char tmstr[RIB_TM_STRLEN]; + buffer_t val; + uint8_t * key; + time_t exp; + + assert(store != NULL); + + val.data = store->val.data; + val.len = store->val.len; + key = store->key.data; + exp = store->exp; + + if (dht_kv_store(key, val, store->exp) < 0) { + log_err(KV_FMT " Failed to store.", KV_VAL(key, val)); + return; + } + + tm = gmtime(&exp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); + log_dbg(KV_FMT " Stored value until %s.", KV_VAL(key, val), tmstr); +} + +static dht_msg_t * do_dht_kv_find_node_req(const dht_find_req_msg_t * req) +{ + dht_contact_msg_t ** contacts; + dht_msg_t * rsp; + uint8_t * key; + uint64_t cookie; + ssize_t len; + + assert(req != NULL); + + key = req->key.data; + cookie = req->cookie; + + len = dht_kv_get_contacts(key, &contacts); + if (len < 0) { + log_warn(KEY_FMT " Failed to get contacts.", KEY_VAL(key)); + goto fail_contacts; + } + + rsp = dht_kv_find_node_rsp_msg(key, cookie, &contacts, len); + if (rsp == NULL) { + log_err(KEY_FMT " Failed to create %s.", KEY_VAL(key), + dht_code_str[DHT_FIND_NODE_RSP]); + goto fail_msg; + } + + assert(rsp->code == DHT_FIND_NODE_RSP); + + log_info(KEY_FMT " Responding with %zd contacts", KEY_VAL(key), len); + + return rsp; + fail_msg: + while (len-- > 0) + dht_contact_msg__free_unpacked(contacts[len], NULL); + + free(contacts); + fail_contacts: + return NULL; +} + +static void dht_kv_process_node_rsp(dht_contact_msg_t ** contacts, + size_t len, + struct list_head * pl, + enum dht_code code) +{ + struct timespec now; + size_t i; + + assert(contacts != NULL); + assert(len > 0); + assert(pl != NULL); + assert(list_is_empty(pl)); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + for (i = 0; i < len; i++) { + dht_contact_msg_t * c = contacts[i]; + struct peer_entry * e; + if (c->addr == dht.addr) + continue; + + if (dht_kv_update_contacts(c->id.data, c->addr) < 0) + log_warn(PEER_FMT " Failed to update contacts.", + PEER_VAL(c->id.data, c->addr)); + + e = malloc(sizeof(*e)); + if (e == NULL) { + log_err(PEER_FMT " Failed to malloc entry.", + PEER_VAL(c->id.data, c->addr)); + continue; + } + + e->id = dht_dup_key(c->id.data); + if (e->id == NULL) { + log_warn(PEER_FMT " Failed to duplicate id.", + PEER_VAL(c->id.data, c->addr)); + free(e); + continue; + } + + e->cookie = generate_cookie(); + e->code = code; + e->addr = c->addr; + e->t_sent = now.tv_sec; + + list_add_tail(&e->next, pl); + } +} + +static dht_msg_t * do_dht_kv_find_value_req(const dht_find_req_msg_t * req) +{ + dht_contact_msg_t ** contacts; + ssize_t n_contacts; + buffer_t * vals; + ssize_t n_vals; + dht_msg_t * rsp; + uint8_t * key; + uint64_t cookie; + + assert(req != NULL); + + key = req->key.data; + cookie = req->cookie; + + n_contacts = dht_kv_get_contacts(key, &contacts); + if (n_contacts < 0) { + log_warn(KEY_FMT " Failed to get contacts.", KEY_VAL(key)); + goto fail_contacts; + } + + assert(n_contacts > 0 || contacts == NULL); + + n_vals = dht_kv_retrieve(key, &vals); + if (n_vals < 0) { + log_dbg(KEY_FMT " Failed to get values.", KEY_VAL(key)); + goto fail_vals; + } + + if (n_vals == 0) + log_dbg(KEY_FMT " No values found.", KEY_VAL(key)); + + rsp = dht_kv_find_value_rsp_msg(key, cookie, &contacts, n_contacts, + &vals, n_vals); + if (rsp == NULL) { + log_err(KEY_FMT " Failed to create %s msg.", + KEY_VAL(key), dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_msg; + } + + log_info(KEY_FMT " Responding with %zd contacts, %zd values.", + KEY_VAL(req->key.data), n_contacts, n_vals); + + return rsp; + + fail_msg: + freebufs(vals, n_vals); + fail_vals: + while (contacts != NULL && n_contacts-- > 0) + dht_contact_msg__free_unpacked(contacts[n_contacts], NULL); + + free(contacts); + fail_contacts: + return NULL; +} + +static void do_dht_kv_find_node_rsp(const dht_find_node_rsp_msg_t * rsp) +{ + struct list_head pl; + + assert(rsp != NULL); + + list_head_init(&pl); + + dht_kv_process_node_rsp(rsp->contacts, rsp->n_contacts, &pl, + DHT_FIND_NODE_REQ); + + if (list_is_empty(&pl)) + goto no_contacts; + + if (dht_kv_update_req(rsp->key.data, &pl) < 0) { + log_err(KEY_FMT " Failed to update request.", + KEY_VAL(rsp->key.data)); + goto fail_update; + } + + dht_kv_query_contacts(rsp->key.data, &pl); + + return; + + fail_update: + peer_list_destroy(&pl); + no_contacts: + return; +} + +static void do_dht_kv_find_value_rsp(const dht_find_node_rsp_msg_t * node, + const dht_find_value_rsp_msg_t * val) +{ + struct list_head pl; + uint8_t * key; + + assert(node != NULL); + assert(val != NULL); + + list_head_init(&pl); + + key = node->key.data; + + dht_kv_process_node_rsp(node->contacts, node->n_contacts, &pl, + DHT_FIND_VALUE_REQ); + + if (val->n_values > 0) { + log_dbg(KEY_FMT " %zd new values received.", + KEY_VAL(key), val->n_values); + if (dht_kv_respond_req(key, val->values, val->n_values) < 0) + log_warn(KEY_FMT " Failed to respond to request.", + KEY_VAL(key)); + peer_list_destroy(&pl); + return; /* done! */ + } + + if (list_is_empty(&pl)) + goto no_contacts; + + if (dht_kv_update_req(key, &pl) < 0) { + log_err(KEY_FMT " Failed to update request.", KEY_VAL(key)); + goto fail_update; + } + + dht_kv_query_remote(key, NULL, &pl); + + return; + fail_update: + peer_list_destroy(&pl); + no_contacts: + return; +} + +static dht_msg_t * dht_wait_for_dht_msg(void) +{ + dht_msg_t * msg; + struct cmd * cmd; + + pthread_mutex_lock(&dht.cmds.mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, &dht.cmds.mtx); + + while (list_is_empty(&dht.cmds.list)) + pthread_cond_wait(&dht.cmds.cond, &dht.cmds.mtx); + + cmd = list_last_entry(&dht.cmds.list, struct cmd, next); + list_del(&cmd->next); + + pthread_cleanup_pop(true); + + msg = dht_msg__unpack(NULL, cmd->cbuf.len, cmd->cbuf.data); + if (msg == NULL) + log_warn("Failed to unpack DHT msg."); + + freebuf(cmd->cbuf); + free(cmd); + + return msg; +} + +static void do_dht_msg(dht_msg_t * msg) +{ + dht_msg_t * rsp = NULL; + uint8_t * id; + uint64_t addr; + +#ifdef DEBUG_PROTO_DHT + dht_kv_debug_msg_rcv(msg); +#endif + if (dht_kv_validate_msg(msg) == -EINVAL) { + log_warn("%s Validation failed.", DHT_CODE(msg)); + dht_msg__free_unpacked(msg, NULL); + return; + } + + id = msg->src->id.data; + addr = msg->src->addr; + + if (dht_kv_update_contacts(id, addr) < 0) + log_warn(PEER_FMT " Failed to update contact from msg src.", + PEER_VAL(id, addr)); + + pthread_cleanup_push(__cleanup_dht_msg, msg); + + switch(msg->code) { + case DHT_FIND_VALUE_REQ: + rsp = do_dht_kv_find_value_req(msg->find); + break; + case DHT_FIND_NODE_REQ: + rsp = do_dht_kv_find_node_req(msg->find); + break; + case DHT_STORE: + do_dht_kv_store(msg->store); + break; + case DHT_FIND_NODE_RSP: + do_dht_kv_find_node_rsp(msg->node); + break; + case DHT_FIND_VALUE_RSP: + do_dht_kv_find_value_rsp(msg->node, msg->val); + break; + default: + assert(false); /* already validated */ + } + + pthread_cleanup_pop(true); + + if (rsp == NULL) + return; + + pthread_cleanup_push(__cleanup_dht_msg, rsp); + + dht_send_msg(rsp, addr); + + pthread_cleanup_pop(true); /* free rsp */ +} + +static void * dht_handle_packet(void * o) +{ + (void) o; + + while (true) { + dht_msg_t * msg; + + msg = dht_wait_for_dht_msg(); + if (msg == NULL) + continue; + + tpm_begin_work(dht.tpm); + + do_dht_msg(msg); + + tpm_end_work(dht.tpm); + } + + return (void *) 0; +} +#ifndef __DHT_TEST__ +static void dht_post_packet(void * comp, + struct ssm_pk_buff * spb) +{ + struct cmd * cmd; + + (void) comp; + + cmd = malloc(sizeof(*cmd)); + if (cmd == NULL) { + log_err("Command malloc failed."); + goto fail_cmd; + } + + cmd->cbuf.data = malloc(ssm_pk_buff_len(spb)); + if (cmd->cbuf.data == NULL) { + log_err("Command buffer malloc failed."); + goto fail_buf; + } + + cmd->cbuf.len = ssm_pk_buff_len(spb); + + memcpy(cmd->cbuf.data, ssm_pk_buff_head(spb), cmd->cbuf.len); + + ipcp_spb_release(spb); + + pthread_mutex_lock(&dht.cmds.mtx); + + list_add(&cmd->next, &dht.cmds.list); + + pthread_cond_signal(&dht.cmds.cond); + + pthread_mutex_unlock(&dht.cmds.mtx); + + return; + + fail_buf: + free(cmd); + fail_cmd: + ipcp_spb_release(spb); + return; +} +#endif + +int dht_reg(const uint8_t * key) +{ + buffer_t val; + + if (addr_to_buf(dht.addr, &val) < 0) { + log_err("Failed to convert address to buffer."); + goto fail_a2b; + } + + if (dht_kv_publish(key, val)) { + log_err(KV_FMT " Failed to publish.", KV_VAL(key, val)); + goto fail_publish; + } + + freebuf(val); + + return 0; + fail_publish: + freebuf(val); + fail_a2b: + return -1; +} + +int dht_unreg(const uint8_t * key) +{ + buffer_t val; + + if (addr_to_buf(dht.addr, &val) < 0) { + log_err("Failed to convert address to buffer."); + goto fail_a2b; + } + + if (dht_kv_unpublish(key, val)) { + log_err(KV_FMT " Failed to unpublish.", KV_VAL(key, val)); + goto fail_unpublish; + } + + freebuf(val); + + return 0; + fail_unpublish: + freebuf(val); + fail_a2b: + return -ENOMEM; +} + +uint64_t dht_query(const uint8_t * key) +{ + buffer_t * vals; + ssize_t n; + uint64_t addr; + + n = dht_kv_retrieve(key, &vals); + if (n < 0) { + log_err(KEY_FMT " Failed to query db.", KEY_VAL(key)); + goto fail_vals; + } + + if (n == 0) { + assert(vals == NULL); + + log_dbg(KEY_FMT " No local values.", KEY_VAL(key)); + n = dht_kv_query_remote(key, &vals, NULL); + if (n < 0) { + log_warn(KEY_FMT " Failed to query DHT.", KEY_VAL(key)); + goto fail_vals; + } + if (n == 0) { + log_dbg(KEY_FMT " No values.", KEY_VAL(key)); + goto no_vals; + } + } + + if (buf_to_addr(vals[0], &addr) < 0) { + log_err(VAL_FMT " Failed addr conversion.", VAL_VAL(vals[0])); + goto fail_b2a; + } + + if (n > 1 && addr == INVALID_ADDR && buf_to_addr(vals[1], &addr) < 0) { + log_err(VAL_FMT " Failed addr conversion.", VAL_VAL(vals[1])); + goto fail_b2a; + } + + freebufs(vals, n); + + return addr; + fail_b2a: + freebufs(vals, n); + return INVALID_ADDR; + no_vals: + free(vals); + fail_vals: + return INVALID_ADDR; +} + +static int emergency_peer(struct list_head * pl) +{ + struct peer_entry * e; + struct timespec now; + + assert(pl != NULL); + assert(list_is_empty(pl)); + + if (dht.peer == INVALID_ADDR) + return -1; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + e = malloc(sizeof(*e)); + if (e == NULL) { + log_err("Failed to malloc emergency peer entry."); + goto fail_malloc; + } + + e->id = dht_dup_key(dht.id.data); + if (e->id == NULL) { + log_err("Failed to duplicate DHT ID for emergency peer."); + goto fail_id; + } + + e->addr = dht.peer; + e->cookie = dht.magic; + e->code = DHT_FIND_NODE_REQ; + e->t_sent = now.tv_sec; + + list_add_tail(&e->next, pl); + + return 0; + fail_id: + free(e); + fail_malloc: + return -ENOMEM; +} + +static bool __dht_kv_bucket_has_addr(struct bucket * b, + uint64_t addr) +{ + struct list_head * p; + size_t i; + + assert(b != NULL); + + if (*b->children != NULL) + for (i = 0; i < (1L << DHT_BETA); ++i) + if (__dht_kv_bucket_has_addr(b->children[i], addr)) + return true; + + llist_for_each(p, &b->contacts) { + struct contact * c; + c = list_entry(p, struct contact, next); + if (c->addr == addr) + return true; + } + + return false; +} + +static bool dht_kv_knows_peer(void) +{ + bool found; + + pthread_rwlock_rdlock(&dht.db.lock); + + found = __dht_kv_bucket_has_addr(dht.db.contacts.root, dht.peer); + + pthread_rwlock_unlock(&dht.db.lock); + + return found; +} + +static int dht_kv_seed_bootstrap_peer(void) +{ + struct list_head pl; + + list_head_init(&pl); + + if (dht.peer == INVALID_ADDR) { + log_dbg("No-one to contact."); + return 0; + } + + if (dht_kv_knows_peer()) + return 0; + + if (emergency_peer(&pl) < 0) { + log_err("Could not create emergency peer."); + goto fail_peer; + } + + log_dbg("Pinging emergency peer " ADDR_FMT32 ".", + ADDR_VAL32(&dht.peer)); + + if (dht_kv_query_contacts(dht.id.data, &pl) < 0) { + log_warn("Failed to bootstrap peer."); + goto fail_query; + } + + peer_list_destroy(&pl); + + /* Sent, but not bootstrapped until the peer is in the DHT. */ + return -EAGAIN; + fail_query: + peer_list_destroy(&pl); + fail_peer: + return -EAGAIN; +} + +static void dht_kv_check_contacts(void) +{ + struct list_head cl; + struct list_head pl; + + list_head_init(&cl); + + dht_kv_contact_list(dht.id.data, &cl, dht.k); + + if (!list_is_empty(&cl)) + goto success; + + contact_list_destroy(&cl); + + list_head_init(&pl); + + if (dht.peer == INVALID_ADDR) { + log_dbg("No-one to contact."); + return; + } + + if (emergency_peer(&pl) < 0) { + log_err("Could not create emergency peer."); + goto fail_peer; + } + + log_dbg("No contacts found, using emergency peer " ADDR_FMT32 ".", + ADDR_VAL32(&dht.peer)); + + dht_kv_query_contacts(dht.id.data, &pl); + + peer_list_destroy(&pl); + + return; + success: + contact_list_destroy(&cl); + return; + fail_peer: + return; +} + +static void dht_kv_remove_expired_reqs(void) +{ + struct list_head * p; + struct list_head * h; + struct timespec now; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + pthread_mutex_lock(&dht.reqs.mtx); + + llist_for_each_safe(p, h, &dht.reqs.ll) { + struct dht_req * e; + e = list_entry(p, struct dht_req, next); + if (IS_EXPIRED(e, &now)) { + log_dbg(KEY_FMT " Removing expired request.", + KEY_VAL(e->key)); + llist_del(&e->next, &dht.reqs.ll); + dht_req_destroy(e); + } + } + + pthread_mutex_unlock(&dht.reqs.mtx); +} + +static void value_list_destroy(struct list_head * vl) +{ + struct list_head * p; + struct list_head * h; + + assert(vl != NULL); + + list_for_each_safe(p, h, vl) { + struct val_entry * v = list_entry(p, struct val_entry, next); + list_del(&v->next); + val_entry_destroy(v); + } +} + +#define MUST_REPLICATE(v, now) ((now)->tv_sec > (v)->t_repl + dht.t_repl) +#define MUST_REPUBLISH(v, now) /* Close to expiry deadline */ \ + (((v)->t_exp - (now)->tv_sec) < (DHT_N_REPUB * dht.t_repl)) +/* A local value must be (re)stored if near expiry or never stored. */ +#define MUST_STORE_LVAL(v, now) (MUST_REPUBLISH(v, now) || (v)->t_repl == 0) +static void dht_entry_get_repl_lists(const struct dht_entry * e, + struct list_head * repl, + struct list_head * rebl, + struct timespec * now) +{ + struct list_head * p; + struct val_entry * n; + + llist_for_each(p, &e->vals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (MUST_REPLICATE(v, now) && !IS_EXPIRED(v, now)) { + n = val_entry_create(v->val, v->t_exp); + if (n == NULL) + continue; + + list_add_tail(&n->next, repl); + } + } + + llist_for_each(p, &e->lvals) { + struct val_entry * v = list_entry(p, struct val_entry, next); + if (MUST_REPLICATE(v, now) && MUST_STORE_LVAL(v, now)) { + /* Add expire time here, to allow creating val_entry */ + n = val_entry_create(v->val, now->tv_sec + dht.t_exp); + if (n == NULL) + continue; + + list_add_tail(&n->next, rebl); + } + } +} + +static int dht_kv_next_values(uint8_t * key, + struct list_head * repl, + struct list_head * rebl) +{ + struct timespec now; + struct list_head * p; + struct list_head * h; + struct dht_entry * e = NULL; + + assert(key != NULL); + assert(repl != NULL); + assert(rebl != NULL); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + assert(list_is_empty(repl)); + assert(list_is_empty(rebl)); + + pthread_rwlock_rdlock(&dht.db.lock); + + if (llist_is_empty(&dht.db.kv.ll)) + goto no_entries; + + llist_for_each_safe(p, h, &dht.db.kv.ll) { + e = list_entry(p, struct dht_entry, next); + if (IS_CLOSER(e->key, key)) + continue; /* Already processed */ + } + + if (e != NULL) { + memcpy(key, e->key, dht.id.len); + dht_entry_get_repl_lists(e, repl, rebl, &now); + } + no_entries: + pthread_rwlock_unlock(&dht.db.lock); + + return list_is_empty(repl) && list_is_empty(rebl) ? -ENOENT : 0; +} + +static void dht_kv_replicate_value(const uint8_t * key, + struct val_entry * v, + const struct timespec * now) +{ + assert(MUST_REPLICATE(v, now)); + + (void) now; + + if (dht_kv_store_remote(key, v->val, v->t_exp) == 0) { + log_dbg(KV_FMT " Replicated.", KV_VAL(key, v->val)); + return; + } + + log_dbg(KV_FMT " Replication failed.", KV_VAL(key, v->val)); + + list_del(&v->next); + val_entry_destroy(v); +} + +static void dht_kv_republish_value(const uint8_t * key, + struct val_entry * v, + const struct timespec * now) +{ + assert(MUST_REPLICATE(v, now)); + + if (MUST_REPUBLISH(v, now)) + assert(v->t_exp >= now->tv_sec + dht.t_exp); + + if (dht_kv_store_remote(key, v->val, v->t_exp) == 0) { + log_dbg(KV_FMT " Republished.", KV_VAL(key, v->val)); + return; + } + + if (MUST_REPUBLISH(v, now)) + log_warn(KV_FMT " Republish failed.", KV_VAL(key, v->val)); + else + log_dbg(KV_FMT " Replication failed.", KV_VAL(key, v->val)); + + list_del(&v->next); + val_entry_destroy(v); +} + +static void dht_kv_update_replication_times(const uint8_t * key, + struct list_head * repl, + struct list_head * rebl, + const struct timespec * now) +{ + struct dht_entry * e; + struct list_head * p; + struct list_head * h; + struct val_entry * v; + + assert(key != NULL); + assert(repl != NULL); + assert(rebl != NULL); + assert(now != NULL); + + pthread_rwlock_wrlock(&dht.db.lock); + + e = __dht_kv_find_entry(key); + if (e == NULL) { + pthread_rwlock_unlock(&dht.db.lock); + return; + } + + list_for_each_safe(p, h, repl) { + struct val_entry * x; + v = list_entry(p, struct val_entry, next); + x = dht_entry_get_val(e, v->val); + if (x == NULL) { + log_err(KV_FMT " Not in vals.", KV_VAL(key, v->val)); + continue; + } + + x->t_repl = now->tv_sec; + + list_del(&v->next); + val_entry_destroy(v); + } + + list_for_each_safe(p, h, rebl) { + struct val_entry * x; + v = list_entry(p, struct val_entry, next); + x = dht_entry_get_lval(e, v->val); + if (x == NULL) { + log_err(KV_FMT " Not in lvals.", KV_VAL(key, v->val)); + continue; + } + + x->t_repl = now->tv_sec; + if (v->t_exp > x->t_exp) { + x->t_exp = v->t_exp; /* update expiration time */ + } + + list_del(&v->next); + val_entry_destroy(v); + } + + pthread_rwlock_unlock(&dht.db.lock); +} + +static void __cleanup_value_list(void * o) +{ + return value_list_destroy((struct list_head *) o); +} + +static void dht_kv_replicate_values(const uint8_t * key, + struct list_head * repl, + struct list_head * rebl) +{ + struct timespec now; + struct list_head * p; + struct list_head * h; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_cleanup_push(__cleanup_value_list, repl); + pthread_cleanup_push(__cleanup_value_list, rebl); + + list_for_each_safe(p, h, repl) { + struct val_entry * v; + v = list_entry(p, struct val_entry, next); + dht_kv_replicate_value(key, v, &now); + } + + list_for_each_safe(p, h, rebl) { + struct val_entry * v; + v = list_entry(p, struct val_entry, next); + dht_kv_republish_value(key, v, &now); + } + + pthread_cleanup_pop(false); + pthread_cleanup_pop(false); + + /* removes non-replicated items from the list */ + dht_kv_update_replication_times(key, repl, rebl, &now); + + if (list_is_empty(repl) && list_is_empty(rebl)) + return; + + log_warn(KEY_FMT " Failed to update replication times.", KEY_VAL(key)); +} + +static void dht_kv_replicate(void) +{ + struct list_head repl; /* list of values to replicate */ + struct list_head rebl; /* list of local values to republish */ + uint8_t * key; + + key = dht_dup_key(dht.id.data); /* dist == 0 */ + if (key == NULL) { + log_err("Replicate: Failed to duplicate DHT ID."); + return; + } + + list_head_init(&repl); + list_head_init(&rebl); + + pthread_cleanup_push(free, key); + + while (dht_kv_next_values(key, &repl, &rebl) == 0) { + dht_kv_replicate_values(key, &repl, &rebl); + if (!list_is_empty(&repl)) { + log_warn(KEY_FMT " Replication items left.", + KEY_VAL(key)); + value_list_destroy(&repl); + } + + if (!list_is_empty(&rebl)) { + log_warn(KEY_FMT " Republish items left.", + KEY_VAL(key)); + value_list_destroy(&rebl); + } + } + + pthread_cleanup_pop(true); +} + +static void dht_kv_refresh_contacts(void) +{ + struct list_head * p; + struct list_head * h; + struct list_head rl; /* refresh list */ + struct timespec now; + + list_head_init(&rl); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + pthread_rwlock_rdlock(&dht.db.lock); + + __dht_kv_bucket_refresh_list(dht.db.contacts.root, now.tv_sec, &rl); + + pthread_rwlock_unlock(&dht.db.lock); + + list_for_each_safe(p, h, &rl) { + struct contact * c; + c = list_entry(p, struct contact, next); + log_dbg(PEER_FMT " Refreshing contact.", + PEER_VAL(c->id, c->addr)); + dht_kv_query_contacts(c->id, NULL); + list_del(&c->next); + contact_destroy(c); + } + + assert(list_is_empty(&rl)); +} + +static void (*tasks[])(void) = { + dht_kv_check_contacts, + dht_kv_remove_expired_entries, + dht_kv_remove_expired_reqs, + dht_kv_replicate, + dht_kv_refresh_contacts, + NULL +}; + +static void * work(void * o) +{ + struct timespec now = TIMESPEC_INIT_MS(1); + time_t intv; + size_t n; /* number of tasks */ + + n = sizeof(tasks) / sizeof(tasks[0]) - 1; /* last is NULL */ + + (void) o; + + while (dht_kv_seed_bootstrap_peer() == -EAGAIN) { + ts_add(&now, &now, &now); /* exponential backoff */ + if (now.tv_sec > 1) /* cap at 1 second */ + now.tv_sec = 1; + nanosleep(&now, NULL); + } + + intv = gcd(dht.t_exp, (dht.t_exp - DHT_N_REPUB * dht.t_repl)); + intv = gcd(intv, gcd(dht.t_repl, dht.t_refr)) / 2; + intv = MAX(1, intv / n); + + log_dbg("DHT worker starting %ld seconds interval.", intv * n); + + /* Flush names registered before we had peers to store them. */ + dht_kv_replicate(); + + while (true) { + int i = 0; + while (tasks[i] != NULL) { + tasks[i++](); + sleep(intv); + } + } + + return (void *) 0; +} + +int dht_start(void) +{ + dht.state = DHT_RUNNING; + + if (tpm_start(dht.tpm)) + goto fail_tpm_start; + +#ifndef __DHT_TEST__ + if (pthread_create(&dht.worker, NULL, work, NULL)) { + log_err("Failed to create DHT worker thread."); + goto fail_worker; + } + + dht.eid = dt_reg_comp(&dht, &dht_post_packet, DHT); + if ((int) dht.eid < 0) { + log_err("Failed to register DHT component."); + goto fail_reg; + } +#else + (void) work; +#endif + return 0; +#ifndef __DHT_TEST__ + fail_reg: + pthread_cancel(dht.worker); + pthread_join(dht.worker, NULL); + fail_worker: + tpm_stop(dht.tpm); +#endif + fail_tpm_start: + dht.state = DHT_INIT; + return -1; +} + +void dht_stop(void) +{ + assert(dht.state == DHT_RUNNING); + +#ifndef __DHT_TEST__ + dt_unreg_comp(dht.eid); + + pthread_cancel(dht.worker); + pthread_join(dht.worker, NULL); +#endif + tpm_stop(dht.tpm); + + dht.state = DHT_INIT; +} + +int dht_init(struct dir_dht_config * conf) +{ + struct timespec now; + pthread_condattr_t cattr; + + assert(conf != NULL); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + +#ifndef __DHT_TEST__ + dht.id.len = ipcp_dir_hash_len(); + dht.addr = addr_auth_address(); +#else + dht.id.len = DHT_TEST_KEY_LEN; + dht.addr = DHT_TEST_ADDR; +#endif + dht.t0 = now.tv_sec; + dht.alpha = conf->params.alpha; + dht.k = conf->params.k; + dht.t_exp = conf->params.t_expire; + dht.t_refr = conf->params.t_refresh; + dht.t_repl = conf->params.t_replicate; + dht.peer = conf->peer; + + dht.magic = generate_cookie(); + + /* Send my address on enrollment */ + conf->peer = dht.addr; + + dht.id.data = generate_id(); + if (dht.id.data == NULL) { + log_err("Failed to create DHT ID."); + goto fail_id; + } + + list_head_init(&dht.cmds.list); + + if (pthread_mutex_init(&dht.cmds.mtx, NULL)) { + log_err("Failed to initialize command mutex."); + goto fail_cmds_mutex; + } + + if (pthread_cond_init(&dht.cmds.cond, NULL)) { + log_err("Failed to initialize command condvar."); + goto fail_cmds_cond; + } + + llist_init(&dht.reqs.ll); + + if (pthread_mutex_init(&dht.reqs.mtx, NULL)) { + log_err("Failed to initialize request mutex."); + goto fail_reqs_mutex; + } + + if (pthread_condattr_init(&cattr)) { + log_err("Failed to initialize request condvar attributes."); + goto fail_cattr; + } +#ifndef __APPLE__ + if (pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK)) { + log_err("Failed to set request condvar clock."); + goto fail_cattr; + } +#endif + if (pthread_cond_init(&dht.reqs.cond, &cattr)) { + log_err("Failed to initialize request condvar."); + goto fail_reqs_cond; + } + + llist_init(&dht.db.kv.ll); + dht.db.kv.vals = 0; + dht.db.kv.lvals = 0; + + if (pthread_rwlock_init(&dht.db.lock, NULL)) { + log_err("Failed to initialize store rwlock."); + goto fail_rwlock; + } + + dht.db.contacts.root = bucket_create(); + if (dht.db.contacts.root == NULL) { + log_err("Failed to create DHT buckets."); + goto fail_buckets; + } + + if (rib_reg(DHT, &r_ops) < 0) { + log_err("Failed to register DHT RIB operations."); + goto fail_rib_reg; + } + + dht.tpm = tpm_create(2, 1, dht_handle_packet, NULL); + if (dht.tpm == NULL) { + log_err("Failed to create TPM for DHT."); + goto fail_tpm_create; + } + + if (dht_kv_update_contacts(dht.id.data, dht.addr) < 0) + log_warn("Failed to update contacts with DHT ID."); + + pthread_condattr_destroy(&cattr); +#ifndef __DHT_TEST__ + log_info("DHT initialized."); + log_dbg(" ID: " HASH_FMT64 " [%zu bytes].", + HASH_VAL64(dht.id.data), dht.id.len); + log_dbg(" address: " ADDR_FMT32 ".", ADDR_VAL32(&dht.addr)); + log_dbg(" peer: " ADDR_FMT32 ".", ADDR_VAL32(&dht.peer)); + log_dbg(" magic cookie: " HASH_FMT64 ".", HASH_VAL64(&dht.magic)); + log_info(" parameters: alpha=%u, k=%zu, t_exp=%ld, " + "t_refr=%ld, t_replicate=%ld.", + dht.alpha, dht.k, dht.t_exp, dht.t_refr, dht.t_repl); +#endif + dht.state = DHT_INIT; + + return 0; + + fail_tpm_create: + rib_unreg(DHT); + fail_rib_reg: + bucket_destroy(dht.db.contacts.root); + fail_buckets: + pthread_rwlock_destroy(&dht.db.lock); + fail_rwlock: + pthread_cond_destroy(&dht.reqs.cond); + fail_reqs_cond: + pthread_condattr_destroy(&cattr); + fail_cattr: + pthread_mutex_destroy(&dht.reqs.mtx); + fail_reqs_mutex: + pthread_cond_destroy(&dht.cmds.cond); + fail_cmds_cond: + pthread_mutex_destroy(&dht.cmds.mtx); + fail_cmds_mutex: + freebuf(dht.id); + fail_id: + return -1; +} + +void dht_fini(void) +{ + struct list_head * p; + struct list_head * h; + + rib_unreg(DHT); + + tpm_destroy(dht.tpm); + + pthread_mutex_lock(&dht.cmds.mtx); + + list_for_each_safe(p, h, &dht.cmds.list) { + struct cmd * c = list_entry(p, struct cmd, next); + list_del(&c->next); + freebuf(c->cbuf); + free(c); + } + + pthread_mutex_unlock(&dht.cmds.mtx); + + pthread_cond_destroy(&dht.cmds.cond); + pthread_mutex_destroy(&dht.cmds.mtx); + + pthread_mutex_lock(&dht.reqs.mtx); + + llist_for_each_safe(p, h, &dht.reqs.ll) { + struct dht_req * r = list_entry(p, struct dht_req, next); + llist_del(&r->next, &dht.reqs.ll); + dht_req_destroy(r); + } + + pthread_mutex_unlock(&dht.reqs.mtx); + + pthread_cond_destroy(&dht.reqs.cond); + pthread_mutex_destroy(&dht.reqs.mtx); + + pthread_rwlock_wrlock(&dht.db.lock); + + llist_for_each_safe(p, h, &dht.db.kv.ll) { + struct dht_entry * e = list_entry(p, struct dht_entry, next); + llist_del(&e->next, &dht.db.kv.ll); + dht_entry_destroy(e); + } + + if (dht.db.contacts.root != NULL) + bucket_destroy(dht.db.contacts.root); + + pthread_rwlock_unlock(&dht.db.lock); + + pthread_rwlock_destroy(&dht.db.lock); + + assert(llist_is_empty(&dht.db.kv.ll)); + assert(dht.db.kv.vals == 0); + assert(dht.db.kv.lvals == 0); + assert(llist_is_empty(&dht.reqs.ll)); + + freebuf(dht.id); +} diff --git a/src/ipcpd/unicast/dir/dht.h b/src/ipcpd/unicast/dir/dht.h new file mode 100644 index 00000000..ed3bb9f0 --- /dev/null +++ b/src/ipcpd/unicast/dir/dht.h @@ -0,0 +1,49 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Distributed Hash Table based on Kademlia + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IPCPD_UNICAST_DIR_DHT_H +#define OUROBOROS_IPCPD_UNICAST_DIR_DHT_H + +#include <ouroboros/ipcp-dev.h> + +#include "ops.h" + +#include <stdint.h> +#include <sys/types.h> + +int dht_init(struct dir_dht_config * conf); + +void dht_fini(void); + +int dht_start(void); + +void dht_stop(void); + +int dht_reg(const uint8_t * key); + +int dht_unreg(const uint8_t * key); + +uint64_t dht_query(const uint8_t * key); + +extern struct dir_ops dht_dir_ops; + +#endif /* OUROBOROS_IPCPD_UNICAST_DIR_DHT_H */ diff --git a/src/ipcpd/unicast/dir/dht.proto b/src/ipcpd/unicast/dir/dht.proto new file mode 100644 index 00000000..02b6b341 --- /dev/null +++ b/src/ipcpd/unicast/dir/dht.proto @@ -0,0 +1,58 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * DHT protocol, based on Kademlia + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public License + * version 2.1 as published by the Free Software Foundation. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +syntax = "proto2"; + +message dht_contact_msg { + required bytes id = 1; + required uint64 addr = 2; +} + +message dht_find_req_msg { + required uint64 cookie = 1; + required bytes key = 2; +} + +message dht_find_node_rsp_msg { + required uint64 cookie = 1; + required bytes key = 2; + repeated dht_contact_msg contacts = 3; +} + +message dht_find_value_rsp_msg { + repeated bytes values = 1; +} + +message dht_store_msg { + required bytes key = 1; + required bytes val = 2; + required uint32 exp = 3; +} + +message dht_msg { + required uint32 code = 1; + required dht_contact_msg src = 2; + optional dht_store_msg store = 3; + optional dht_find_req_msg find = 4; + optional dht_find_node_rsp_msg node = 5; + optional dht_find_value_rsp_msg val = 6; +} diff --git a/src/ipcpd/unicast/dir/ops.h b/src/ipcpd/unicast/dir/ops.h new file mode 100644 index 00000000..6c336ee0 --- /dev/null +++ b/src/ipcpd/unicast/dir/ops.h @@ -0,0 +1,42 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Directory policy ops + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#ifndef OUROBOROS_IPCPD_UNICAST_DIR_OPS_H +#define OUROBOROS_IPCPD_UNICAST_DIR_OPS_H + +struct dir_ops { + int (* init)(void * config); + + void (* fini)(void); + + int (* start)(void); + + void (* stop)(void); + + int (* reg)(const uint8_t * hash); + + int (* unreg)(const uint8_t * hash); + + uint64_t (* query)(const uint8_t * hash); +}; + +#endif /* OUROBOROS_IPCPD_UNICAST_DIR_OPS_H */ diff --git a/src/ipcpd/unicast/dir/pol.h b/src/ipcpd/unicast/dir/pol.h new file mode 100644 index 00000000..8ccf4f95 --- /dev/null +++ b/src/ipcpd/unicast/dir/pol.h @@ -0,0 +1,23 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Directory policies + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "dht.h" diff --git a/src/ipcpd/unicast/dir/tests/CMakeLists.txt b/src/ipcpd/unicast/dir/tests/CMakeLists.txt new file mode 100644 index 00000000..eded823f --- /dev/null +++ b/src/ipcpd/unicast/dir/tests/CMakeLists.txt @@ -0,0 +1,38 @@ +get_filename_component(CURRENT_SOURCE_PARENT_DIR + ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(CURRENT_BINARY_PARENT_DIR + ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) + +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + dht_test.c +) + +protobuf_generate_c(DHT_PROTO_SRCS KAD_PROTO_HDRS ${CURRENT_SOURCE_PARENT_DIR}/dht.proto) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests} + ${DHT_PROTO_SRCS}) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd + ${CMAKE_SOURCE_DIR}/src/ipcpd/unicast + ${CMAKE_BINARY_DIR}/src/ipcpd/unicast +) + +disable_test_logging_for_target(${PARENT_DIR}_test) + +target_link_libraries(${PARENT_DIR}_test ouroboros-common) +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/dir/tests/dht_test.c b/src/ipcpd/unicast/dir/tests/dht_test.c new file mode 100644 index 00000000..1f7026b3 --- /dev/null +++ b/src/ipcpd/unicast/dir/tests/dht_test.c @@ -0,0 +1,1914 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests of the DHT + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#define __DHT_TEST__ + +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200112L +#endif + +#include <test/test.h> +#include <ouroboros/list.h> +#include <ouroboros/utils.h> + +#include "dht.pb-c.h" + +#include <assert.h> +#include <inttypes.h> +#include <time.h> +#include <stdlib.h> +#include <stdio.h> + +#define DHT_MAX_RAND_SIZE 64 +#define DHT_TEST_KEY_LEN 32 +#define DHT_TEST_ADDR 0x1234567890abcdefULL + +/* forward declare for use in the dht code */ +/* Packet sink for DHT tests */ +struct { + bool enabled; + + struct llist msgs; +} sink; + +struct message { + struct list_head next; + void * msg; + uint64_t dst; +}; + +static int sink_send_msg(buffer_t * pkt, + uint64_t addr) +{ + struct message * m; + + assert(pkt != NULL); + assert(addr != 0); + + if (!sink.enabled) + goto finish; + + m = malloc(sizeof(*m)); + if (m == NULL) { + printf("Failed to malloc message."); + goto fail_malloc; + } + + m->msg = dht_msg__unpack(NULL, pkt->len, pkt->data); + if (m->msg == NULL) + goto fail_unpack; + + m->dst = addr; + + llist_add_tail(&m->next, &sink.msgs); + + finish: + freebuf(*pkt); + + return 0; + fail_unpack: + free(m); + fail_malloc: + freebuf(*pkt); + return -1; +} + +#include "dht.c" + +/* Test helpers */ + +static void sink_init(void) +{ + llist_init(&sink.msgs); + sink.enabled = true; +} + +static void sink_clear(void) +{ + struct list_head * p; + struct list_head * h; + + llist_for_each_safe(p, h, &sink.msgs) { + struct message * m = list_entry(p, struct message, next); + llist_del(&m->next, &sink.msgs); + dht_msg__free_unpacked((dht_msg_t *) m->msg, NULL); + free(m); + } + + assert(llist_is_empty(&sink.msgs)); +} + +static void sink_fini(void) +{ + sink_clear(); + sink.enabled = false; +} + +static dht_msg_t * sink_read(void) +{ + struct message * m; + dht_msg_t * msg; + + if (llist_is_empty(&sink.msgs)) + return NULL; + + m = llist_first_entry(&sink.msgs, struct message, next); + + llist_del(&m->next, &sink.msgs); + + msg = m->msg; + + free(m); + + return (dht_msg_t *) msg; +} + +static const buffer_t test_val = { + .data = (uint8_t *) "test_value", + .len = 10 +}; + +static const buffer_t test_val2 = { + .data = (uint8_t *) "test_value_2", + .len = 12 +}; + +static int random_value_len(buffer_t * b) +{ + assert(b != NULL); + assert(b->len > 0 && b->len <= DHT_MAX_RAND_SIZE); + + b->data = malloc(b->len); + if (b->data == NULL) + goto fail_malloc; + + random_buffer(b->data, b->len); + + return 0; + + fail_malloc: + return -ENOMEM; +} + +static int random_value(buffer_t * b) +{ + assert(b != NULL); + + b->len = rand() % DHT_MAX_RAND_SIZE + 1; + + return random_value_len(b); +} + +static int fill_dht_with_contacts(size_t n) +{ + size_t i; + uint8_t * id; + + for (i = 0; i < n; i++) { + uint64_t addr = generate_cookie(); + id = generate_id(); + if (id == NULL) + goto fail_id; + + if (dht_kv_update_contacts(id, addr) < 0) + goto fail_update; + free(id); + } + + return 0; + + fail_update: + free(id); + fail_id: + return -1; +} + +static int fill_store_with_random_values(const uint8_t * key, + size_t len, + size_t n_values) +{ + buffer_t val; + struct timespec now; + size_t i; + uint8_t * _key; + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + for (i = 0; i < n_values; ++i) { + if (key != NULL) + _key = (uint8_t *) key; + else { + _key = generate_id(); + if (_key == NULL) + goto fail_key; + } + + if (len == 0) + val.len = rand() % DHT_MAX_RAND_SIZE + 1; + else + val.len = len; + + if (random_value_len(&val) < 0) + goto fail_value; + + if (dht_kv_store(_key, val, now.tv_sec + 10) < 0) + goto fail_store; + + freebuf(val); + if (key == NULL) + free(_key); + } + + return 0; + + fail_store: + freebuf(val); + fail_value: + free(_key); + fail_key: + return -1; +} + +static int random_contact_list(dht_contact_msg_t *** contacts, + size_t max) +{ + size_t i; + + assert(contacts != NULL); + assert(max > 0); + + *contacts = malloc(max * sizeof(**contacts)); + if (*contacts == NULL) + goto fail_malloc; + + for (i = 0; i < max; i++) { + (*contacts)[i] = malloc(sizeof(*(*contacts)[i])); + if ((*contacts)[i] == NULL) + goto fail_contacts; + + dht_contact_msg__init((*contacts)[i]); + + (*contacts)[i]->id.data = generate_id(); + if ((*contacts)[i]->id.data == NULL) + goto fail_contact; + + (*contacts)[i]->id.len = dht.id.len; + (*contacts)[i]->addr = generate_cookie(); + } + + return 0; + + fail_contact: + dht_contact_msg__free_unpacked((*contacts)[i], NULL); + fail_contacts: + while (i-- > 0) + free((*contacts)[i]); + free(*contacts); + fail_malloc: + return -ENOMEM; +} + +static void clear_contacts(dht_contact_msg_t ** contacts, + size_t len) +{ + size_t i; + + assert(contacts != NULL); + if (*contacts == NULL) + return; + + for (i = 0; i < len; ++i) + dht_contact_msg__free_unpacked((contacts)[i], NULL); + + free(*contacts); + *contacts = NULL; +} + +/* Start of actual tests */ +static struct dir_dht_config test_dht_config = { + .params = { + .alpha = 3, + .k = 8, + .t_expire = 86400, + .t_refresh = 900, + .t_replicate = 900 + } +}; + +static int test_dht_init_fini(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_start_stop(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (dht_start() < 0) { + printf("Failed to start dht.\n"); + goto fail_start; + } + + dht_stop(); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_start: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_val_entry_create_destroy(void) +{ + struct val_entry * e; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = val_entry_create(test_val, now.tv_sec + 10); + if (e == NULL) { + printf("Failed to create val entry.\n"); + goto fail_entry; + } + + val_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_create_destroy(void) +{ + struct dht_entry * e; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_update_get_val(void) +{ + struct dht_entry * e; + struct val_entry * v; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + if (dht_entry_get_val(e, test_val) != NULL) { + printf("Found value in empty dht entry.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 10) < 0) { + printf("Failed to update dht entry value.\n"); + goto fail_get; + } + + if (dht_entry_get_val(e, test_val2) != NULL) { + printf("Found value in dht entry with different key.\n"); + goto fail_get; + } + + v = dht_entry_get_val(e, test_val); + if (v == NULL) { + printf("Failed to get value from dht entry.\n"); + goto fail_get; + } + + if (v->val.len != test_val.len) { + printf("Length in dht entry does not match expected.\n"); + goto fail_get; + } + + if(memcmp(v->val.data, test_val.data, test_val.len) != 0) { + printf("Data in dht entry does not match expected.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 15) < 0) { + printf("Failed to update exsting dht entry value.\n"); + goto fail_get; + } + + if (v->t_exp != now.tv_sec + 15) { + printf("Expiration time in dht entry value not updated.\n"); + goto fail_get; + } + + if (dht_entry_update_val(e, test_val, now.tv_sec + 5) < 0) { + printf("Failed to update existing dht entry value (5).\n"); + goto fail_get; + } + + if (v->t_exp != now.tv_sec + 15) { + printf("Expiration time in dht entry shortened.\n"); + goto fail_get; + } + + if (dht_entry_get_val(e, test_val) != v) { + printf("Wrong value in dht entry found after update.\n"); + goto fail_get; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_entry_destroy(e); + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_entry_update_get_lval(void) +{ + struct dht_entry * e; + struct val_entry * v; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + e = dht_entry_create(dht.id.data); + if (e == NULL) { + printf("Failed to create dht entry.\n"); + goto fail_entry; + } + + if (dht_entry_get_lval(e, test_val) != NULL) { + printf("Found value in empty dht entry.\n"); + goto fail_get; + } + + if (dht_entry_update_lval(e, test_val) < 0) { + printf("Failed to update dht entry value.\n"); + goto fail_get; + } + + v = dht_entry_get_lval(e, test_val); + if (v== NULL) { + printf("Failed to get value from dht entry.\n"); + goto fail_get; + } + + if (dht_entry_get_lval(e, test_val2) != NULL) { + printf("Found value in dht entry in vals.\n"); + goto fail_get; + } + + if (v->val.len != test_val.len) { + printf("Length in dht entry does not match expected.\n"); + goto fail_get; + } + + if(memcmp(v->val.data, test_val.data, test_val.len) != 0) { + printf("Data in dht entry does not match expected.\n"); + goto fail_get; + } + + if (dht_entry_update_lval(e, test_val) < 0) { + printf("Failed to update existing dht entry value.\n"); + goto fail_get; + } + + if (dht_entry_get_lval(e, test_val) != v) { + printf("Wrong value in dht entry found after update.\n"); + goto fail_get; + } + + dht_entry_destroy(e); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_entry_destroy(e); + fail_entry: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_contact_create_destroy(void) +{ + struct contact * c; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + c = contact_create(dht.id.data, dht.addr); + if (c == NULL) { + printf("Failed to create contact.\n"); + goto fail_contact; + } + + contact_destroy(c); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_contact: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_update_bucket(void) +{ + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(1000) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_update; + } + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_update: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_contact_list(void) +{ + struct list_head cl; + ssize_t len; + ssize_t items; + + TEST_START(); + + list_head_init(&cl); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + items = 5; + + if (fill_dht_with_contacts(items) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + len = dht_kv_contact_list(dht.id.data, &cl, dht.k); + if (len < 0) { + printf("Failed to get contact list.\n"); + goto fail_fill; + } + + if (len != items) { + printf("Failed to get contacts (%zu != %zu).\n", len, items); + goto fail_contact_list; + } + + contact_list_destroy(&cl); + + items = 100; + + if (fill_dht_with_contacts(items) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + len = dht_kv_contact_list(dht.id.data, &cl, items); + if (len < 0) { + printf("Failed to get contact list.\n"); + goto fail_fill; + } + + if ((size_t) len < dht.k) { + printf("Failed to get contacts (%zu < %zu).\n", len, dht.k); + goto fail_contact_list; + } + + contact_list_destroy(&cl); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_contact_list: + contact_list_destroy(&cl); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_get_values(void) +{ + buffer_t * vals; + ssize_t len; + size_t n = sizeof(uint64_t); + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_store_with_random_values(dht.id.data, n, 3) < 0) { + printf("Failed to fill store with random values.\n"); + goto fail_fill; + } + + len = dht_kv_retrieve(dht.id.data, &vals); + if (len < 0) { + printf("Failed to get values from store.\n"); + goto fail_fill; + } + + if (len != 3) { + printf("Failed to get %ld values (%zu).\n", 3L, len); + goto fail_get_values; + } + + freebufs(vals, len); + + if (fill_store_with_random_values(dht.id.data, n, 20) < 0) { + printf("Failed to fill store with random values.\n"); + goto fail_fill; + } + + len = dht_kv_retrieve(dht.id.data, &vals); + if (len < 0) { + printf("Failed to get values from store.\n"); + goto fail_fill; + } + + if (len != DHT_MAX_VALS) { + printf("Failed to get %d values.\n", DHT_MAX_VALS); + goto fail_get_values; + } + + freebufs(vals, len); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get_values: + freebufs(vals, len); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_req_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_node_req_msg(dht.id.data); + if (msg == NULL) { + printf("Failed to get find node request message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_NODE_REQ) { + printf("Wrong code in find_node_req message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_NODE_REQ]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_req.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_req buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_req message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_req message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_rsp_msg(void) +{ + dht_contact_msg_t ** contacts; + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_node_rsp_msg(dht.id.data, 0, &contacts, 0); + if (msg == NULL) { + printf("Failed to get find node response message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_NODE_RSP) { + printf("Wrong code in find_node_rsp message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_NODE_RSP]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_node_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_node_rsp_msg_contacts(void) +{ + dht_contact_msg_t ** contacts; + dht_msg_t * msg; + dht_msg_t * upk; + uint8_t * buf; + size_t len; + ssize_t n; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(100) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + n = dht_kv_get_contacts(dht.id.data, &contacts); + if (n < 0) { + printf("Failed to get contacts.\n"); + goto fail_fill; + } + + if ((size_t) n < dht.k) { + printf("Failed to get all contacts (%zu < %zu).\n", n, dht.k); + goto fail_fill; + } + + msg = dht_kv_find_node_rsp_msg(dht.id.data, 0, &contacts, n); + if (msg == NULL) { + printf("Failed to build find node response message.\n"); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_node_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_node_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_node_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + clear_contacts(contacts, n); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_req_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_value_req_msg(dht.id.data); + if (msg == NULL) { + printf("Failed to build find value request message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_REQ) { + printf("Wrong code in find_value_req message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_VALUE_REQ]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_req.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_node_req buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_req message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_req message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_find_value_rsp_msg(dht.id.data, 0, NULL, 0, NULL, 0); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_RSP) { + printf("Wrong code in find_value_rsp message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg_contacts(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + dht_contact_msg_t ** contacts; + ssize_t n; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(100) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_fill; + } + + n = dht_kv_get_contacts(dht.id.data, &contacts); + if (n < 0) { + printf("Failed to get contacts.\n"); + goto fail_fill; + } + + if ((size_t) n < dht.k) { + printf("Failed to get all contacts (%zu < %zu).\n", n, dht.k); + goto fail_fill; + } + + msg = dht_kv_find_value_rsp_msg(dht.id.data, 0, &contacts, n, NULL, 0); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + clear_contacts(contacts, n); + fail_fill: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_find_value_rsp_msg_values(void) +{ + dht_msg_t * msg; + dht_msg_t * upk; + size_t len; + uint8_t * buf; + buffer_t * values; + size_t i; + uint64_t ck; + + TEST_START(); + + ck = generate_cookie(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + values = malloc(sizeof(*values) * 8); + if (values == NULL) { + printf("Failed to malloc values.\n"); + goto fail_values; + } + + for (i = 0; i < 8; i++) { + if (random_value(&values[i]) < 0) { + printf("Failed to create random value.\n"); + goto fail_fill; + } + } + + msg = dht_kv_find_value_rsp_msg(dht.id.data, ck, NULL, 0, &values, 8); + if (msg == NULL) { + printf("Failed to build find value response message.\n"); + goto fail_msg; + } + + values = NULL; /* msg owns the values now */ + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed length of find_value_rsp.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc find_value_rsp buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack find_value_rsp message.\n"); + goto fail_pack; + } + + upk = dht_msg__unpack(NULL, len, buf); + if (upk == NULL) { + printf("Failed to unpack find_value_rsp message.\n"); + goto fail_unpack; + } + + if (upk->code != DHT_FIND_VALUE_RSP) { + printf("Wrong code in find_value_rsp message (%s != %s).\n", + dht_code_str[upk->code], + dht_code_str[DHT_FIND_VALUE_RSP]); + goto fail_unpack; + } + + if (upk->val == NULL) { + printf("No values in find_value_rsp message.\n"); + goto fail_unpack; + } + + if (upk->val->n_values != 8) { + printf("Not enough values in find_value_rsp (%zu != %lu).\n", + upk->val->n_values, 8UL); + goto fail_unpack; + } + + free(buf); + dht_msg__free_unpacked(msg, NULL); + dht_msg__free_unpacked(upk, NULL); + + free(values); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_unpack: + dht_msg__free_unpacked(msg, NULL); + fail_pack: + free(buf); + fail_msg: + fail_fill: + while((i--) > 0) + freebuf(values[i]); + free(values); + fail_values: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_store_msg(void) +{ + dht_msg_t * msg; + size_t len; + uint8_t * buf; + struct timespec now; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + msg = dht_kv_store_msg(dht.id.data, test_val, now.tv_sec + 10); + if (msg == NULL) { + printf("Failed to get store message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_STORE) { + printf("Wrong code in store message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_STORE]); + goto fail_store_msg; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate store message.\n"); + goto fail_store_msg; + } + + len = dht_msg__get_packed_size(msg); + if (len == 0) { + printf("Failed to get packed msg length.\n"); + goto fail_msg; + } + + buf = malloc(len); + if (buf == NULL) { + printf("Failed to malloc store msg buf.\n"); + goto fail_msg; + } + + if (dht_msg__pack(msg, buf) != len) { + printf("Failed to pack store message.\n"); + goto fail_pack; + } + + free(buf); + + dht_msg__free_unpacked(msg, NULL); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_pack: + free(buf); + fail_store_msg: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_kv_query_contacts_req_rsp(void) +{ + dht_msg_t * req; + dht_msg_t * rsp; + dht_contact_msg_t ** contacts; + size_t len = 2; + + uint8_t * key; + + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(1) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_prep; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key.\n"); + goto fail_prep; + } + + if (dht_kv_query_contacts(key, NULL) < 0) { + printf("Failed to query contacts.\n"); + goto fail_query; + } + + req = sink_read(); + if (req == NULL) { + printf("Failed to read request from sink.\n"); + goto fail_query; + } + + if (dht_kv_validate_msg(req) < 0) { + printf("Failed to validate find node req.\n"); + goto fail_val_req; + } + + if (random_contact_list(&contacts, len) < 0) { + printf("Failed to create random contact.\n"); + goto fail_val_req; + } + + rsp = dht_kv_find_node_rsp_msg(key, req->find->cookie, &contacts, len); + if (rsp == NULL) { + printf("Failed to create find node response message.\n"); + goto fail_rsp; + } + + memcpy(rsp->src->id.data, dht.id.data, dht.id.len); + rsp->src->addr = generate_cookie(); + + if (dht_kv_validate_msg(rsp) < 0) { + printf("Failed to validate find node response message.\n"); + goto fail_val_rsp; + } + + do_dht_kv_find_node_rsp(rsp->node); + + /* dht_contact_msg__free_unpacked(contacts[0], NULL); set to NULL */ + + free(contacts); + + dht_msg__free_unpacked(rsp, NULL); + + free(key); + + dht_msg__free_unpacked(req, NULL); + + sink_fini(); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_val_rsp: + dht_msg__free_unpacked(rsp, NULL); + fail_rsp: + while (len-- > 0) + dht_contact_msg__free_unpacked(contacts[len], NULL); + free(contacts); + fail_val_req: + dht_msg__free_unpacked(req, NULL); + fail_query: + free(key); + fail_prep: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_req_create_destroy(void) +{ + struct dht_req * req; + + TEST_START(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + req = dht_req_create(dht.id.data); + if (req == NULL) { + printf("Failed to create kad request.\n"); + goto fail_req; + } + + dht_req_destroy(req); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_req: + dht_fini(); + fail_init: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_unreg(void) +{ + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (!llist_is_empty(&sink.msgs)) { + printf("Packet sent without contacts!"); + goto fail_msg; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_msg; + } + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_msg: + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_unreg_contacts(void) +{ + dht_msg_t * msg; + + TEST_START(); + + sink_init(); + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(4) < 0) { + printf("Failed to fill bucket with contacts.\n"); + goto fail_reg; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (sink.msgs.len != dht.alpha) { + printf("Packet sent to too few contacts!\n"); + goto fail_msg; + } + + msg = sink_read(); + if (msg == NULL) { + printf("Failed to read message from sink.\n"); + goto fail_msg; + } + + if (msg->code != DHT_STORE) { + printf("Wrong code in dht reg message (%s != %s).\n", + dht_code_str[msg->code], + dht_code_str[DHT_STORE]); + goto fail_validation; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate dht message.\n"); + goto fail_validation; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_validation; + } + + dht_msg__free_unpacked(msg, NULL); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_validation: + dht_msg__free_unpacked(msg, NULL); + fail_msg: + sink_clear(); + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + sink_fini(); + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_reg_query_local(void) +{ + struct timespec now; + buffer_t test_addr; + + TEST_START(); + + clock_gettime(CLOCK_REALTIME_COARSE, &now); + + if (addr_to_buf(1234321, &test_addr) < 0) { + printf("Failed to convert test address to buffer.\n"); + goto fail_buf; + } + + if (dht_init(&test_dht_config) < 0) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (dht_reg(dht.id.data) < 0) { + printf("Failed to register own id.\n"); + goto fail_reg; + } + + if (dht_query(dht.id.data) == dht.addr) { + printf("Succeeded to query own id.\n"); + goto fail_get; + } + + if (dht_kv_store(dht.id.data, test_addr, now.tv_sec + 5) < 0) { + printf("Failed to publish value.\n"); + goto fail_get; + } + + if (dht_query(dht.id.data) != 1234321) { + printf("Failed to return remote addr.\n"); + goto fail_get; + } + + if (dht_unreg(dht.id.data) < 0) { + printf("Failed to unregister own id.\n"); + goto fail_get; + } + + freebuf(test_addr); + + dht_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_get: + dht_unreg(dht.id.data); + fail_reg: + dht_fini(); + fail_init: + freebuf(test_addr); + fail_buf: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_dht_query(void) +{ + uint8_t * key; + struct dir_dht_config cfg; + + TEST_START(); + + sink_init(); + + cfg = test_dht_config; + cfg.peer = generate_cookie(); + + if (dht_init(&cfg)) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key.\n"); + goto fail_key; + } + + if (dht_query(key) != INVALID_ADDR) { + printf("Succeeded to get address without contacts.\n"); + goto fail_get; + } + + if (!llist_is_empty(&sink.msgs)) { + printf("Packet sent without contacts!"); + goto fail_test; + } + + free(key); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + + fail_test: + sink_clear(); + fail_get: + free(key); + fail_key: + dht_fini(); + fail_init: + sink_fini(); + return TEST_RC_FAIL; +} + +static int test_dht_query_contacts(void) +{ + dht_msg_t * msg; + uint8_t * key; + struct dir_dht_config cfg; + + + TEST_START(); + + sink_init(); + + cfg = test_dht_config; + cfg.peer = generate_cookie(); + + if (dht_init(&cfg)) { + printf("Failed to create dht.\n"); + goto fail_init; + } + + if (fill_dht_with_contacts(10) < 0) { + printf("Failed to fill with contacts!"); + goto fail_contacts; + } + + key = generate_id(); + if (key == NULL) { + printf("Failed to generate key."); + goto fail_contacts; + } + + if (dht_query(key) != INVALID_ADDR) { + printf("Succeeded to get address for random id.\n"); + goto fail_query; + } + + msg = sink_read(); + if (msg == NULL) { + printf("Failed to read message.!\n"); + goto fail_read; + } + + if (dht_kv_validate_msg(msg) < 0) { + printf("Failed to validate dht message.\n"); + goto fail_msg; + } + + if (msg->code != DHT_FIND_VALUE_REQ) { + printf("Failed to validate dht message.\n"); + goto fail_msg; + } + + dht_msg__free_unpacked(msg, NULL); + + free(key); + + sink_clear(); + + dht_fini(); + + sink_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_msg: + dht_msg__free_unpacked(msg, NULL); + fail_read: + sink_clear(); + fail_query: + free(key); + fail_contacts: + dht_fini(); + fail_init: + sink_fini(); + return TEST_RC_FAIL; +} + +int dht_test(int argc, + char ** argv) +{ + int rc = 0; + + (void) argc; + (void) argv; + + rc |= test_dht_init_fini(); + rc |= test_dht_start_stop(); + rc |= test_val_entry_create_destroy(); + rc |= test_dht_entry_create_destroy(); + rc |= test_dht_entry_update_get_val(); + rc |= test_dht_entry_update_get_lval(); + rc |= test_dht_kv_contact_create_destroy(); + rc |= test_dht_kv_contact_list(); + rc |= test_dht_kv_update_bucket(); + rc |= test_dht_kv_get_values(); + rc |= test_dht_kv_find_node_req_msg(); + rc |= test_dht_kv_find_node_rsp_msg(); + rc |= test_dht_kv_find_node_rsp_msg_contacts(); + rc |= test_dht_kv_query_contacts_req_rsp(); + rc |= test_dht_kv_find_value_req_msg(); + rc |= test_dht_kv_find_value_rsp_msg(); + rc |= test_dht_kv_find_value_rsp_msg_contacts(); + rc |= test_dht_kv_find_value_rsp_msg_values(); + rc |= test_dht_kv_store_msg(); + rc |= test_dht_req_create_destroy(); + rc |= test_dht_reg_unreg(); + rc |= test_dht_reg_unreg_contacts(); + rc |= test_dht_reg_query_local(); + rc |= test_dht_query(); + rc |= test_dht_query_contacts(); + + return rc; +} diff --git a/src/ipcpd/unicast/dt.c b/src/ipcpd/unicast/dt.c index 1360371a..381ed815 100644 --- a/src/ipcpd/unicast/dt.c +++ b/src/ipcpd/unicast/dt.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Data Transfer Component * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -31,24 +31,28 @@ #define DT "dt" #define OUROBOROS_PREFIX DT +#include <ouroboros/atomics.h> #include <ouroboros/bitmap.h> #include <ouroboros/errno.h> #include <ouroboros/logs.h> #include <ouroboros/dev.h> +#include <ouroboros/ipcp-dev.h> #include <ouroboros/notifier.h> #include <ouroboros/rib.h> #ifdef IPCP_FLOW_STATS #include <ouroboros/fccntl.h> #endif +#include "addr-auth.h" +#include "common/comp.h" +#include "common/connmgr.h" #include "ca.h" -#include "connmgr.h" +#include "cap.h" #include "ipcp.h" #include "dt.h" #include "pff.h" #include "routing.h" #include "psched.h" -#include "comp.h" #include "fa.h" #include <stdlib.h> @@ -58,15 +62,16 @@ #include <inttypes.h> #include <assert.h> -#define QOS_BLOCK_LEN 672 -#define STAT_FILE_LEN (189 + QOS_BLOCK_LEN * QOS_CUBE_MAX) +#define QOS_BLOCK_LEN 672 +#define RIB_FILE_STRLEN (169 + RIB_TM_STRLEN + QOS_BLOCK_LEN * QOS_CUBE_MAX) +#define RIB_NAME_STRLEN 256 #ifndef CLOCK_REALTIME_COARSE #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif struct comp_info { - void (* post_packet)(void * comp, struct shm_du_buff * sdb); + void (* post_packet)(void * comp, struct ssm_pk_buff * spb); void * comp; char * name; }; @@ -75,12 +80,14 @@ struct comp_info { #define TTL_LEN 1 #define QOS_LEN 1 #define ECN_LEN 1 +#define CAP_LEN 1 struct dt_pci { uint64_t dst_addr; qoscube_t qc; uint8_t ttl; uint8_t ecn; + uint8_t cap; uint64_t eid; }; @@ -93,6 +100,7 @@ struct { size_t qc_o; size_t ttl_o; size_t ecn_o; + size_t cap_o; size_t eid_o; /* Initial TTL value */ @@ -112,6 +120,7 @@ static void dt_pci_ser(uint8_t * head, memcpy(head + dt_pci_info.qc_o, &dt_pci->qc, QOS_LEN); memcpy(head + dt_pci_info.ttl_o, &ttl, TTL_LEN); memcpy(head + dt_pci_info.ecn_o, &dt_pci->ecn, ECN_LEN); + memcpy(head + dt_pci_info.cap_o, &dt_pci->cap, CAP_LEN); memcpy(head + dt_pci_info.eid_o, &dt_pci->eid, dt_pci_info.eid_size); } @@ -130,22 +139,26 @@ static void dt_pci_des(uint8_t * head, memcpy(&dt_pci->qc, head + dt_pci_info.qc_o, QOS_LEN); memcpy(&dt_pci->ttl, head + dt_pci_info.ttl_o, TTL_LEN); memcpy(&dt_pci->ecn, head + dt_pci_info.ecn_o, ECN_LEN); + memcpy(&dt_pci->cap, head + dt_pci_info.cap_o, CAP_LEN); memcpy(&dt_pci->eid, head + dt_pci_info.eid_o, dt_pci_info.eid_size); } -static void dt_pci_shrink(struct shm_du_buff * sdb) +static void dt_pci_shrink(struct ssm_pk_buff * spb) { - assert(sdb); + assert(spb); - shm_du_buff_head_release(sdb, dt_pci_info.head_size); + ssm_pk_buff_pop(spb, dt_pci_info.head_size); } struct { struct psched * psched; + uint64_t addr; + struct pff * pff[QOS_CUBE_MAX]; struct routing_i * routing[QOS_CUBE_MAX]; #ifdef IPCP_FLOW_STATS + /* Flow stats use lock-free atomics; stamp is the validity flag. */ struct { time_t stamp; uint64_t addr; @@ -163,66 +176,77 @@ struct { size_t w_drp_bytes[QOS_CUBE_MAX]; size_t f_nhp_pkt[QOS_CUBE_MAX]; size_t f_nhp_bytes[QOS_CUBE_MAX]; - pthread_mutex_t lock; - } stat[PROG_MAX_FLOWS]; + } stat[PROC_MAX_FLOWS]; size_t n_flows; #endif struct bmp * res_fds; - struct comp_info comps[PROG_RES_FDS]; + struct comp_info comps[PROC_RES_FDS]; pthread_rwlock_t lock; pthread_t listener; } dt; -static int dt_stat_read(const char * path, - char * buf, - size_t len) -{ #ifdef IPCP_FLOW_STATS +#define dt_stat_inc(idx, name, qc, len) \ + do { \ + FETCH_ADD_RELAXED(&dt.stat[idx].name ## _pkt[qc], 1); \ + FETCH_ADD_RELAXED(&dt.stat[idx].name ## _bytes[qc], (len)); \ + } while (0) +#define dt_stat_load(idx, field, qc) LOAD_RELAXED(&dt.stat[idx].field[qc]) + +static int dt_rib_read(const char * path, + char * buf, + size_t len) +{ int fd; int i; char str[QOS_BLOCK_LEN + 1]; char addrstr[20]; - char tmstr[20]; + char * entry; + char tmstr[RIB_TM_STRLEN]; size_t rxqlen = 0; size_t txqlen = 0; + time_t stamp; + uint64_t addr; struct tm * tm; /* NOTE: we may need stronger checks. */ - fd = atoi(path); + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); - if (len < STAT_FILE_LEN) + fd = atoi(entry); + + if (len < RIB_FILE_STRLEN) return 0; buf[0] = '\0'; - pthread_mutex_lock(&dt.stat[fd].lock); - - if (dt.stat[fd].stamp == 0) { - pthread_mutex_unlock(&dt.stat[fd].lock); + stamp = LOAD_ACQUIRE(&dt.stat[fd].stamp); + if (stamp == 0) return 0; - } - if (dt.stat[fd].addr == ipcpi.dt_addr) + addr = LOAD_RELAXED(&dt.stat[fd].addr); + + if (addr == dt.addr) sprintf(addrstr, "%s", dt.comps[fd].name); else - sprintf(addrstr, "%" PRIu64, dt.stat[fd].addr); + sprintf(addrstr, ADDR_FMT32, ADDR_VAL32(&addr)); - tm = localtime(&dt.stat[fd].stamp); - strftime(tmstr, sizeof(tmstr), "%F %T", tm); + tm = gmtime(&stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - if (fd >= PROG_RES_FDS) { + if (fd >= PROC_RES_FDS) { fccntl(fd, FLOWGRXQLEN, &rxqlen); fccntl(fd, FLOWGTXQLEN, &txqlen); } sprintf(buf, - "Flow established at: %20s\n" + "Flow established at: %.*s\n" "Endpoint address: %20s\n" "Queued packets (rx): %20zu\n" "Queued packets (tx): %20zu\n\n", - tmstr, addrstr, rxqlen, txqlen); + RIB_TM_STRLEN - 1, tmstr, addrstr, rxqlen, txqlen); for (i = 0; i < QOS_CUBE_MAX; ++i) { sprintf(str, "Qos cube %3d:\n" @@ -241,38 +265,29 @@ static int dt_stat_read(const char * path, " failed nhop (packets): %20zu\n" " failed nhop (bytes): %20zu\n", i, - dt.stat[fd].snd_pkt[i], - dt.stat[fd].snd_bytes[i], - dt.stat[fd].rcv_pkt[i], - dt.stat[fd].rcv_bytes[i], - dt.stat[fd].lcl_w_pkt[i], - dt.stat[fd].lcl_w_bytes[i], - dt.stat[fd].lcl_r_pkt[i], - dt.stat[fd].lcl_r_bytes[i], - dt.stat[fd].r_drp_pkt[i], - dt.stat[fd].r_drp_bytes[i], - dt.stat[fd].w_drp_pkt[i], - dt.stat[fd].w_drp_bytes[i], - dt.stat[fd].f_nhp_pkt[i], - dt.stat[fd].f_nhp_bytes[i] + dt_stat_load(fd, snd_pkt, i), + dt_stat_load(fd, snd_bytes, i), + dt_stat_load(fd, rcv_pkt, i), + dt_stat_load(fd, rcv_bytes, i), + dt_stat_load(fd, lcl_w_pkt, i), + dt_stat_load(fd, lcl_w_bytes, i), + dt_stat_load(fd, lcl_r_pkt, i), + dt_stat_load(fd, lcl_r_bytes, i), + dt_stat_load(fd, r_drp_pkt, i), + dt_stat_load(fd, r_drp_bytes, i), + dt_stat_load(fd, w_drp_pkt, i), + dt_stat_load(fd, w_drp_bytes, i), + dt_stat_load(fd, f_nhp_pkt, i), + dt_stat_load(fd, f_nhp_bytes, i) ); strcat(buf, str); } - pthread_mutex_unlock(&dt.stat[fd].lock); - - return STAT_FILE_LEN; -#else - (void) path; - (void) buf; - (void) len; - return 0; -#endif + return RIB_FILE_STRLEN; } -static int dt_stat_readdir(char *** buf) +static int dt_rib_readdir(char *** buf) { -#ifdef IPCP_FLOW_STATS char entry[RIB_PATH_LEN + 1]; size_t i; int idx = 0; @@ -280,114 +295,116 @@ static int dt_stat_readdir(char *** buf) pthread_rwlock_rdlock(&dt.lock); if (dt.n_flows < 1) { - pthread_rwlock_unlock(&dt.lock); - return 0; + *buf = NULL; + goto no_flows; } *buf = malloc(sizeof(**buf) * dt.n_flows); - if (*buf == NULL) { - pthread_rwlock_unlock(&dt.lock); - return -ENOMEM; - } - - for (i = 0; i < PROG_MAX_FLOWS; ++i) { - pthread_mutex_lock(&dt.stat[i].lock); + if (*buf == NULL) + goto fail_entries; - if (dt.stat[i].stamp == 0) { - pthread_mutex_unlock(&dt.stat[i].lock); - /* Optimization: skip unused res_fds. */ - if (i < PROG_RES_FDS) - i = PROG_RES_FDS; - continue; - } + for (i = 0; i < PROC_MAX_FLOWS && idx < (int) dt.n_flows; ++i) { + if (LOAD_RELAXED(&dt.stat[i].stamp) == 0) + continue; /* n-1 flows start at PROC_RES_FDS */ sprintf(entry, "%zu", i); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(buf); - pthread_mutex_unlock(&dt.stat[i].lock); - pthread_rwlock_unlock(&dt.lock); - return -ENOMEM; - } + if ((*buf)[idx] == NULL) + goto fail_entry; strcpy((*buf)[idx++], entry); - pthread_mutex_unlock(&dt.stat[i].lock); } - assert((size_t) idx == dt.n_flows); - + no_flows: pthread_rwlock_unlock(&dt.lock); return idx; -#else - (void) buf; - return 0; -#endif + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + + free(*buf); + fail_entries: + pthread_rwlock_unlock(&dt.lock); + return -ENOMEM; } -static int dt_stat_getattr(const char * path, - struct stat * st) +static int dt_rib_getattr(const char * path, + struct rib_attr * attr) { -#ifdef IPCP_FLOW_STATS - int fd; + int fd; + char * entry; + time_t stamp; - fd = atoi(path); + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); - st->st_mode = S_IFREG | 0755; - st->st_nlink = 1; - st->st_uid = getuid(); - st->st_gid = getgid(); + fd = atoi(entry); - pthread_mutex_lock(&dt.stat[fd].lock); + stamp = LOAD_ACQUIRE(&dt.stat[fd].stamp); - if (dt.stat[fd].stamp != -1) { - st->st_size = STAT_FILE_LEN; - st->st_mtime = dt.stat[fd].stamp; + if (stamp != -1) { + attr->size = RIB_FILE_STRLEN; + attr->mtime = stamp; } else { - st->st_size = 0; - st->st_mtime = 0; + attr->size = 0; + attr->mtime = 0; } - pthread_mutex_unlock(&dt.stat[fd].lock); -#else - (void) path; - (void) st; -#endif return 0; } static struct rib_ops r_ops = { - .read = dt_stat_read, - .readdir = dt_stat_readdir, - .getattr = dt_stat_getattr + .read = dt_rib_read, + .readdir = dt_rib_readdir, + .getattr = dt_rib_getattr }; -#ifdef IPCP_FLOW_STATS static void stat_used(int fd, uint64_t addr) { struct timespec now; + int i; clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_mutex_lock(&dt.stat[fd].lock); - - memset(&dt.stat[fd], 0, sizeof(dt.stat[fd])); + pthread_rwlock_wrlock(&dt.lock); - dt.stat[fd].stamp = (addr != INVALID_ADDR) ? now.tv_sec : 0; - dt.stat[fd].addr = addr; + STORE_RELEASE(&dt.stat[fd].stamp, 0); - pthread_mutex_unlock(&dt.stat[fd].lock); + /* Don't memset: incremented without locks in fast path. */ + for (i = 0; i < QOS_CUBE_MAX; ++i) { + STORE_RELAXED(&dt.stat[fd].snd_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].snd_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].rcv_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].rcv_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].lcl_r_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].lcl_r_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].lcl_w_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].lcl_w_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].r_drp_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].r_drp_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].w_drp_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].w_drp_bytes[i], 0); + STORE_RELAXED(&dt.stat[fd].f_nhp_pkt[i], 0); + STORE_RELAXED(&dt.stat[fd].f_nhp_bytes[i], 0); + } - pthread_rwlock_wrlock(&dt.lock); + STORE_RELAXED(&dt.stat[fd].addr, addr); - (addr != INVALID_ADDR) ? ++dt.n_flows : --dt.n_flows; + if (addr != INVALID_ADDR) { + STORE_RELEASE(&dt.stat[fd].stamp, now.tv_sec); + ++dt.n_flows; + } else { + --dt.n_flows; + } pthread_rwlock_unlock(&dt.lock); } +#else +#define dt_stat_inc(idx, name, qc, len) ((void) 0) #endif static void handle_event(void * self, @@ -395,6 +412,7 @@ static void handle_event(void * self, const void * o) { struct conn * c; + int fd; (void) self; @@ -402,65 +420,57 @@ static void handle_event(void * self, switch (event) { case NOTIFY_DT_CONN_ADD: + fd = c->flow_info.fd; #ifdef IPCP_FLOW_STATS - stat_used(c->flow_info.fd, c->conn_info.addr); + stat_used(fd, c->conn_info.addr); #endif - psched_add(dt.psched, c->flow_info.fd); - log_dbg("Added fd %d to packet scheduler.", c->flow_info.fd); + cap_reset(fd); + psched_add(dt.psched, fd); + log_dbg("Added fd %d to packet scheduler.", fd); break; case NOTIFY_DT_CONN_DEL: + fd = c->flow_info.fd; #ifdef IPCP_FLOW_STATS - stat_used(c->flow_info.fd, INVALID_ADDR); + stat_used(fd, INVALID_ADDR); #endif - psched_del(dt.psched, c->flow_info.fd); - log_dbg("Removed fd %d from " - "packet scheduler.", c->flow_info.fd); + psched_del(dt.psched, fd); + log_dbg("Removed fd %d from packet scheduler.", fd); break; default: break; } } -static void packet_handler(int fd, - qoscube_t qc, - struct shm_du_buff * sdb) +static time_t packet_handler(int fd, + qoscube_t qc, + struct ssm_pk_buff * spb) { struct dt_pci dt_pci; int ret; int ofd; uint8_t * head; size_t len; + size_t qlen; + bool marks; - len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); + len = ssm_pk_buff_len(spb); #ifndef IPCP_FLOW_STATS - (void) fd; -#else - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].rcv_pkt[qc]; - dt.stat[fd].rcv_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); + (void) fd; #endif + dt_stat_inc(fd, rcv, qc, len); + memset(&dt_pci, 0, sizeof(dt_pci)); - head = shm_du_buff_head(sdb); + head = ssm_pk_buff_head(spb); dt_pci_des(head, &dt_pci); - if (dt_pci.dst_addr != ipcpi.dt_addr) { + if (dt_pci.dst_addr != dt.addr) { if (dt_pci.ttl == 0) { log_dbg("TTL was zero."); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].r_drp_pkt[qc]; - dt.stat[fd].r_drp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[fd].lock); -#endif - return; + ipcp_spb_release(spb); + dt_stat_inc(fd, r_drp, qc, len); + return 0; } /* FIXME: Use qoscube from PCI instead of incoming flow. */ @@ -468,75 +478,56 @@ static void packet_handler(int fd, if (ofd < 0) { log_dbg("No next hop for %" PRIu64 ".", dt_pci.dst_addr); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); + ipcp_spb_release(spb); + dt_stat_inc(fd, f_nhp, qc, len); + return 0; + } - ++dt.stat[fd].f_nhp_pkt[qc]; - dt.stat[fd].f_nhp_bytes[qc] += len; + marks = ca_marks_ecn(); + qlen = marks ? ipcp_flow_queued(ofd) : 0; - pthread_mutex_unlock(&dt.stat[fd].lock); -#endif - return; - } + (void) ca_calc_ecn(qlen, head + dt_pci_info.ecn_o, qc, len); - (void) ca_calc_ecn(ofd, head + dt_pci_info.ecn_o, len); + if (marks) + cap_stamp(head + dt_pci_info.cap_o, cap_get(ofd)); - ret = ipcp_flow_write(ofd, sdb); + ret = ipcp_flow_write(ofd, spb); if (ret < 0) { log_dbg("Failed to write packet to fd %d.", ofd); if (ret == -EFLOWDOWN) notifier_event(NOTIFY_DT_FLOW_DOWN, &ofd); - ipcp_sdb_release(sdb); -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[ofd].lock); - - ++dt.stat[ofd].w_drp_pkt[qc]; - dt.stat[ofd].w_drp_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[ofd].lock); -#endif - return; + ipcp_spb_release(spb); + dt_stat_inc(ofd, w_drp, qc, len); + return 0; } -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[ofd].lock); - ++dt.stat[ofd].snd_pkt[qc]; - dt.stat[ofd].snd_bytes[qc] += len; + dt_stat_inc(ofd, snd, qc, len); - pthread_mutex_unlock(&dt.stat[ofd].lock); -#endif + if (marks) + cap_update(ofd, qlen, len); } else { - dt_pci_shrink(sdb); - if (dt_pci.eid >= PROG_RES_FDS) { + dt_pci_shrink(spb); + if (dt_pci.eid >= PROC_RES_FDS) { uint8_t ecn = *(head + dt_pci_info.ecn_o); - fa_np1_rcv(dt_pci.eid, ecn, sdb); - return; + uint8_t cap = *(head + dt_pci_info.cap_o); + fa_np1_rcv(dt_pci.eid, ecn, cap, spb); + return 0; } if (dt.comps[dt_pci.eid].post_packet == NULL) { log_err("No registered component on eid %" PRIu64 ".", dt_pci.eid); - ipcp_sdb_release(sdb); - return; + ipcp_spb_release(spb); + return 0; } -#ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - ++dt.stat[fd].lcl_r_pkt[qc]; - dt.stat[fd].lcl_r_bytes[qc] += len; + dt_stat_inc(fd, lcl_r, qc, len); + dt_stat_inc(dt_pci.eid, snd, qc, len); - pthread_mutex_unlock(&dt.stat[fd].lock); - pthread_mutex_lock(&dt.stat[dt_pci.eid].lock); - - ++dt.stat[dt_pci.eid].snd_pkt[qc]; - dt.stat[dt_pci.eid].snd_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[dt_pci.eid].lock); -#endif dt.comps[dt_pci.eid].post_packet(dt.comps[dt_pci.eid].comp, - sdb); + spb); } + + return 0; } static void * dt_conn_handle(void * o) @@ -559,43 +550,49 @@ static void * dt_conn_handle(void * o) return 0; } -int dt_init(enum pol_routing pr, - uint8_t addr_size, - uint8_t eid_size, - uint8_t max_ttl) +int dt_init(struct dt_config cfg) { int i; int j; - char dtstr[256]; - int pp; +#ifdef IPCP_FLOW_STATS + char dtstr[RIB_NAME_STRLEN + 1]; +#endif + enum pol_pff pp; struct conn_info info; memset(&info, 0, sizeof(info)); + dt.addr = addr_auth_address(); + if (dt.addr == INVALID_ADDR) { + log_err("Failed to get address"); + return -1; + } + strcpy(info.comp_name, DT_COMP); strcpy(info.protocol, DT_PROTO); info.pref_version = 1; info.pref_syntax = PROTO_FIXED; - info.addr = ipcpi.dt_addr; + info.addr = dt.addr; - if (eid_size != 8) { /* only support 64 bits from now */ + if (cfg.eid_size != 8) { /* only support 64 bits from now */ log_warn("Invalid EID size. Only 64 bit is supported."); - eid_size = 8; + cfg.eid_size = 8; } - dt_pci_info.addr_size = addr_size; - dt_pci_info.eid_size = eid_size; - dt_pci_info.max_ttl = max_ttl; + dt_pci_info.addr_size = cfg.addr_size; + dt_pci_info.eid_size = cfg.eid_size; + dt_pci_info.max_ttl = cfg.max_ttl; dt_pci_info.qc_o = dt_pci_info.addr_size; dt_pci_info.ttl_o = dt_pci_info.qc_o + QOS_LEN; dt_pci_info.ecn_o = dt_pci_info.ttl_o + TTL_LEN; - dt_pci_info.eid_o = dt_pci_info.ecn_o + ECN_LEN; + dt_pci_info.cap_o = dt_pci_info.ecn_o + ECN_LEN; + dt_pci_info.eid_o = dt_pci_info.cap_o + CAP_LEN; dt_pci_info.head_size = dt_pci_info.eid_o + dt_pci_info.eid_size; - if (notifier_reg(handle_event, NULL)) { - log_err("Failed to register with notifier."); - goto fail_notifier_reg; + if (cap_init() < 0) { + log_err("Failed to init capacity estimator."); + goto fail_cap; } if (connmgr_comp_init(COMPID_DT, &info)) { @@ -603,8 +600,7 @@ int dt_init(enum pol_routing pr, goto fail_connmgr_comp_init; } - pp = routing_init(pr); - if (pp < 0) { + if (routing_init(&cfg.routing, &pp) < 0) { log_err("Failed to init routing."); goto fail_routing; } @@ -633,34 +629,27 @@ int dt_init(enum pol_routing pr, goto fail_rwlock_init; } - dt.res_fds = bmp_create(PROG_RES_FDS, 0); + dt.res_fds = bmp_create(PROC_RES_FDS, 0); if (dt.res_fds == NULL) goto fail_res_fds; #ifdef IPCP_FLOW_STATS memset(dt.stat, 0, sizeof(dt.stat)); - for (i = 0; i < PROG_MAX_FLOWS; ++i) - if (pthread_mutex_init(&dt.stat[i].lock, NULL)) { - for (j = 0; j < i; ++j) - pthread_mutex_destroy(&dt.stat[j].lock); - goto fail_stat_lock; - } - dt.n_flows = 0; -#endif - sprintf(dtstr, "%s.%" PRIu64, DT, ipcpi.dt_addr); - if (rib_reg(dtstr, &r_ops)) + + sprintf(dtstr, "%s." ADDR_FMT32, DT, ADDR_VAL32(&dt.addr)); + if (rib_reg(dtstr, &r_ops)) { + log_err("Failed to register RIB."); goto fail_rib_reg; + } +#endif return 0; - fail_rib_reg: #ifdef IPCP_FLOW_STATS - for (i = 0; i < PROG_MAX_FLOWS; ++i) - pthread_mutex_destroy(&dt.stat[i].lock); - fail_stat_lock: -#endif + fail_rib_reg: bmp_destroy(dt.res_fds); +#endif fail_res_fds: pthread_rwlock_destroy(&dt.lock); fail_rwlock_init: @@ -674,19 +663,21 @@ int dt_init(enum pol_routing pr, fail_routing: connmgr_comp_fini(COMPID_DT); fail_connmgr_comp_init: - notifier_unreg(&handle_event); - fail_notifier_reg: + cap_fini(); + fail_cap: return -1; } void dt_fini(void) { +#ifdef IPCP_FLOW_STATS + char dtstr[RIB_NAME_STRLEN + 1]; +#endif int i; - rib_unreg(DT); #ifdef IPCP_FLOW_STATS - for (i = 0; i < PROG_MAX_FLOWS; ++i) - pthread_mutex_destroy(&dt.stat[i].lock); + sprintf(dtstr, "%s.%" PRIu64, DT, dt.addr); + rib_unreg(dtstr); #endif bmp_destroy(dt.res_fds); @@ -702,46 +693,70 @@ void dt_fini(void) connmgr_comp_fini(COMPID_DT); - notifier_unreg(&handle_event); + cap_fini(); } int dt_start(void) { - dt.psched = psched_create(packet_handler); + dt.psched = psched_create(packet_handler, ipcp_flow_read); if (dt.psched == NULL) { log_err("Failed to create N-1 packet scheduler."); - return -1; + goto fail_psched; + } + + if (notifier_reg(handle_event, NULL)) { + log_err("Failed to register with notifier."); + goto fail_notifier_reg; } if (pthread_create(&dt.listener, NULL, dt_conn_handle, NULL)) { log_err("Failed to create listener thread."); - psched_destroy(dt.psched); - return -1; + goto fail_listener; + } + + if (routing_start() < 0) { + log_err("Failed to start routing."); + goto fail_routing; } return 0; + + fail_routing: + pthread_cancel(dt.listener); + pthread_join(dt.listener, NULL); + fail_listener: + notifier_unreg(&handle_event); + fail_notifier_reg: + psched_destroy(dt.psched); + fail_psched: + return -1; } void dt_stop(void) { + routing_stop(); + pthread_cancel(dt.listener); pthread_join(dt.listener, NULL); + + notifier_unreg(&handle_event); + psched_destroy(dt.psched); } int dt_reg_comp(void * comp, - void (* func)(void * func, struct shm_du_buff *), + void (* func)(void * func, struct ssm_pk_buff *), char * name) { int eid; - assert(func); + assert(func != NULL); pthread_rwlock_wrlock(&dt.lock); eid = bmp_allocate(dt.res_fds); if (!bmp_is_id_valid(dt.res_fds, eid)) { - log_warn("Reserved EIDs depleted."); + log_err("Cannot allocate EID."); pthread_rwlock_unlock(&dt.lock); return -EBADF; } @@ -756,71 +771,90 @@ int dt_reg_comp(void * comp, pthread_rwlock_unlock(&dt.lock); #ifdef IPCP_FLOW_STATS - stat_used(eid, ipcpi.dt_addr); + stat_used(eid, dt.addr); #endif return eid; } +void dt_unreg_comp(int eid) +{ + assert(eid >= 0 && eid < PROC_RES_FDS); + + pthread_rwlock_wrlock(&dt.lock); + + assert(dt.comps[eid].post_packet != NULL); + + dt.comps[eid].post_packet = NULL; + dt.comps[eid].comp = NULL; + dt.comps[eid].name = NULL; + + pthread_rwlock_unlock(&dt.lock); + + return; +} + int dt_write_packet(uint64_t dst_addr, qoscube_t qc, uint64_t eid, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb, + uint8_t * ecn) { struct dt_pci dt_pci; int fd; int ret; uint8_t * head; size_t len; + size_t qlen; + bool marks; - assert(sdb); - assert(dst_addr != ipcpi.dt_addr); - - len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); + assert(spb); + assert(dst_addr != dt.addr); #ifdef IPCP_FLOW_STATS - if (eid < PROG_RES_FDS) { - pthread_mutex_lock(&dt.stat[eid].lock); - - ++dt.stat[eid].lcl_r_pkt[qc]; - dt.stat[eid].lcl_r_bytes[qc] += len; + len = ssm_pk_buff_len(spb); - pthread_mutex_unlock(&dt.stat[eid].lock); - } + if (eid < PROC_RES_FDS) + dt_stat_inc(eid, lcl_r, qc, len); #endif fd = pff_nhop(dt.pff[qc], dst_addr); if (fd < 0) { - log_dbg("Could not get nhop for addr %" PRIu64 ".", dst_addr); + log_dbg("Could not get nhop for " ADDR_FMT32 ".", + ADDR_VAL32(&dst_addr)); #ifdef IPCP_FLOW_STATS - if (eid < PROG_RES_FDS) { - pthread_mutex_lock(&dt.stat[eid].lock); - - ++dt.stat[eid].lcl_r_pkt[qc]; - dt.stat[eid].lcl_r_bytes[qc] += len; - - pthread_mutex_unlock(&dt.stat[eid].lock); - } + if (eid < PROC_RES_FDS) + dt_stat_inc(eid, lcl_r, qc, len); #endif return -EPERM; } - head = shm_du_buff_head_alloc(sdb, dt_pci_info.head_size); + head = ssm_pk_buff_push(spb, dt_pci_info.head_size); if (head == NULL) { log_dbg("Failed to allocate DT header."); goto fail_write; } - len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); + len = ssm_pk_buff_len(spb); dt_pci.dst_addr = dst_addr; dt_pci.qc = qc; dt_pci.eid = eid; dt_pci.ecn = 0; + dt_pci.cap = 0; - (void) ca_calc_ecn(fd, &dt_pci.ecn, len); + marks = ca_marks_ecn(); + qlen = marks ? ipcp_flow_queued(fd) : 0; + + (void) ca_calc_ecn(qlen, &dt_pci.ecn, qc, len); + + if (marks) + dt_pci.cap = cap_get(fd); + + if (ecn != NULL) + *ecn = dt_pci.ecn; dt_pci_ser(head, &dt_pci); - ret = ipcp_flow_write(fd, sdb); + ret = ipcp_flow_write(fd, spb); if (ret < 0) { log_dbg("Failed to write packet to fd %d.", fd); if (ret == -EFLOWDOWN) @@ -828,31 +862,22 @@ int dt_write_packet(uint64_t dst_addr, goto fail_write; } #ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - if (dt_pci.eid < PROG_RES_FDS) { - ++dt.stat[fd].lcl_w_pkt[qc]; - dt.stat[fd].lcl_w_bytes[qc] += len; - } - ++dt.stat[fd].snd_pkt[qc]; - dt.stat[fd].snd_bytes[qc] += len; + if (dt_pci.eid < PROC_RES_FDS) + dt_stat_inc(fd, lcl_w, qc, len); - pthread_mutex_unlock(&dt.stat[fd].lock); + dt_stat_inc(fd, snd, qc, len); #endif + if (marks) + cap_update(fd, qlen, len); + return 0; fail_write: #ifdef IPCP_FLOW_STATS - pthread_mutex_lock(&dt.stat[fd].lock); - - if (eid < PROG_RES_FDS) { - ++dt.stat[fd].lcl_w_pkt[qc]; - dt.stat[fd].lcl_w_bytes[qc] += len; - } - ++dt.stat[fd].w_drp_pkt[qc]; - dt.stat[fd].w_drp_bytes[qc] += len; + if (eid < PROC_RES_FDS) + dt_stat_inc(fd, lcl_w, qc, len); - pthread_mutex_unlock(&dt.stat[fd].lock); + dt_stat_inc(fd, w_drp, qc, len); #endif return -1; } diff --git a/src/ipcpd/unicast/dt.h b/src/ipcpd/unicast/dt.h index 1fd8f295..a055efea 100644 --- a/src/ipcpd/unicast/dt.h +++ b/src/ipcpd/unicast/dt.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Data Transfer component * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -25,17 +25,13 @@ #include <ouroboros/ipcp.h> #include <ouroboros/qoscube.h> -#include <ouroboros/shm_rdrbuff.h> +#include <ouroboros/ssm_pool.h> #define DT_COMP "Data Transfer" #define DT_PROTO "dtp" #define INVALID_ADDR 0 -int dt_init(enum pol_routing pr, - uint8_t addr_size, - uint8_t eid_size, - uint8_t max_ttl -); +int dt_init(struct dt_config cfg); void dt_fini(void); @@ -43,13 +39,16 @@ int dt_start(void); void dt_stop(void); -int dt_reg_comp(void * comp, - void (* func)(void * comp, struct shm_du_buff * sdb), - char * name); +int dt_reg_comp(void * comp, + void (* func)(void * comp, struct ssm_pk_buff * spb), + char * name); + +void dt_unreg_comp(int eid); int dt_write_packet(uint64_t dst_addr, qoscube_t qc, uint64_t eid, - struct shm_du_buff * sdb); + struct ssm_pk_buff * spb, + uint8_t * ecn); #endif /* OUROBOROS_IPCPD_UNICAST_DT_H */ diff --git a/src/ipcpd/unicast/enroll.c b/src/ipcpd/unicast/enroll.c deleted file mode 100644 index d47ebb40..00000000 --- a/src/ipcpd/unicast/enroll.c +++ /dev/null @@ -1,382 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Enrollment Task - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#if defined(__linux__) || defined(__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 199309L -#endif - -#define OUROBOROS_PREFIX "enrollment" - -#include <ouroboros/endian.h> -#include <ouroboros/errno.h> -#include <ouroboros/time_utils.h> -#include <ouroboros/dev.h> -#include <ouroboros/logs.h> -#include <ouroboros/errno.h> -#include <ouroboros/sockets.h> - -#include "connmgr.h" -#include "enroll.h" -#include "ipcp.h" - -#include <assert.h> -#include <stdlib.h> -#include <string.h> -#include <pthread.h> - -#include "ipcp_config.pb-c.h" -typedef EnrollMsg enroll_msg_t; - -#define ENROLL_COMP "Enrollment" -#define ENROLL_PROTO "OEP" /* Ouroboros enrollment protocol */ -#define ENROLL_WARN_TIME_OFFSET 20 -#define ENROLL_BUF_LEN 1024 - -enum enroll_state { - ENROLL_NULL = 0, - ENROLL_INIT, - ENROLL_RUNNING -}; - -struct { - struct ipcp_config conf; - enum enroll_state state; - pthread_t listener; -} enroll; - -static int send_rcv_enroll_msg(int fd) -{ - enroll_msg_t req = ENROLL_MSG__INIT; - enroll_msg_t * reply; - uint8_t buf[ENROLL_BUF_LEN]; - ssize_t len; - ssize_t delta_t; - struct timespec t0; - struct timespec rtt; - - req.code = ENROLL_CODE__ENROLL_REQ; - - len = enroll_msg__get_packed_size(&req); - if (len < 0) { - log_dbg("Failed pack request message."); - return -1; - } - - enroll_msg__pack(&req, buf); - - clock_gettime(CLOCK_REALTIME, &t0); - - if (flow_write(fd, buf, len) < 0) { - log_dbg("Failed to send request message."); - return -1; - } - - len = flow_read(fd, buf, ENROLL_BUF_LEN); - if (len < 0) { - log_dbg("No enrollment reply received."); - return -1; - } - - log_dbg("Received enrollment info (%zd bytes).", len); - - reply = enroll_msg__unpack(NULL, len, buf); - if (reply == NULL) { - log_dbg("No enrollment response."); - return -1; - } - - if (reply->code != ENROLL_CODE__ENROLL_BOOT) { - log_dbg("Failed to unpack enrollment response."); - enroll_msg__free_unpacked(reply, NULL); - return -1; - } - - if (!(reply->has_t_sec && reply->has_t_nsec)) { - log_dbg("No time in response message."); - enroll_msg__free_unpacked(reply, NULL); - return -1; - } - - clock_gettime(CLOCK_REALTIME, &rtt); - - delta_t = ts_diff_ms(&t0, &rtt); - - rtt.tv_sec = reply->t_sec; - rtt.tv_nsec = reply->t_nsec; - - if (labs(ts_diff_ms(&t0, &rtt)) - delta_t > ENROLL_WARN_TIME_OFFSET) - log_warn("Clock offset above threshold."); - - strcpy(enroll.conf.layer_info.layer_name, - reply->conf->layer_info->layer_name); - enroll.conf.type = reply->conf->ipcp_type; - enroll.conf.addr_size = reply->conf->addr_size; - enroll.conf.eid_size = reply->conf->eid_size; - enroll.conf.max_ttl = reply->conf->max_ttl; - enroll.conf.addr_auth_type = reply->conf->addr_auth_type; - enroll.conf.routing_type = reply->conf->routing_type; - enroll.conf.cong_avoid = reply->conf->cong_avoid; - enroll.conf.layer_info.dir_hash_algo - = reply->conf->layer_info->dir_hash_algo; - - enroll_msg__free_unpacked(reply, NULL); - - return 0; -} - -static ssize_t enroll_pack(uint8_t ** buf) -{ - enroll_msg_t msg = ENROLL_MSG__INIT; - ipcp_config_msg_t config = IPCP_CONFIG_MSG__INIT; - layer_info_msg_t layer_info = LAYER_INFO_MSG__INIT; - struct timespec now; - ssize_t len; - - clock_gettime(CLOCK_REALTIME, &now); - - msg.code = ENROLL_CODE__ENROLL_BOOT; - msg.has_t_sec = true; - msg.t_sec = now.tv_sec; - msg.has_t_nsec = true; - msg.t_nsec = now.tv_nsec; - msg.conf = &config; - - config.ipcp_type = enroll.conf.type; - config.has_addr_size = true; - config.addr_size = enroll.conf.addr_size; - config.has_eid_size = true; - config.eid_size = enroll.conf.eid_size; - config.has_max_ttl = true; - config.max_ttl = enroll.conf.max_ttl; - config.has_addr_auth_type = true; - config.addr_auth_type = enroll.conf.addr_auth_type; - config.has_routing_type = true; - config.routing_type = enroll.conf.routing_type; - config.has_cong_avoid = true; - config.cong_avoid = enroll.conf.cong_avoid; - config.layer_info = &layer_info; - - layer_info.layer_name = (char *) enroll.conf.layer_info.layer_name; - layer_info.dir_hash_algo = enroll.conf.layer_info.dir_hash_algo; - - len = enroll_msg__get_packed_size(&msg); - - *buf = malloc(len); - if (*buf == NULL) - return -1; - - enroll_msg__pack(&msg, *buf); - - return len; -} - -static void * enroll_handle(void * o) -{ - struct conn conn; - uint8_t buf[ENROLL_BUF_LEN]; - uint8_t * reply; - ssize_t len; - enroll_msg_t * msg; - - (void) o; - - while (true) { - if (connmgr_wait(COMPID_ENROLL, &conn)) { - log_err("Failed to get next connection."); - continue; - } - - len = flow_read(conn.flow_info.fd, buf, ENROLL_BUF_LEN); - if (len < 0) { - log_err("Failed to read from flow."); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - msg = enroll_msg__unpack(NULL, len, buf); - if (msg == NULL) { - log_err("Failed to unpack message."); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - if (msg->code != ENROLL_CODE__ENROLL_REQ) { - log_err("Wrong message type."); - connmgr_dealloc(COMPID_ENROLL, &conn); - enroll_msg__free_unpacked(msg, NULL); - continue; - } - - log_dbg("Enrolling a new neighbor."); - - enroll_msg__free_unpacked(msg, NULL); - - len = enroll_pack(&reply); - if (reply == NULL) { - log_err("Failed to pack enrollment message."); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - log_dbg("Sending enrollment info (%zd bytes).", len); - - if (flow_write(conn.flow_info.fd, reply, len) < 0) { - log_err("Failed respond to enrollment request."); - connmgr_dealloc(COMPID_ENROLL, &conn); - free(reply); - continue; - } - - free(reply); - - len = flow_read(conn.flow_info.fd, buf, ENROLL_BUF_LEN); - if (len < 0) { - log_err("Failed to read from flow."); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - msg = enroll_msg__unpack(NULL, len, buf); - if (msg == NULL) { - log_err("Failed to unpack message."); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - if (msg->code != ENROLL_CODE__ENROLL_DONE || !msg->has_result) { - log_err("Wrong message type."); - enroll_msg__free_unpacked(msg, NULL); - connmgr_dealloc(COMPID_ENROLL, &conn); - continue; - } - - if (msg->result == 0) - log_dbg("Neighbor enrollment successful."); - else - log_dbg("Neigbor reported failed enrollment."); - - enroll_msg__free_unpacked(msg, NULL); - - connmgr_dealloc(COMPID_ENROLL, &conn); - } - - return 0; -} - -int enroll_boot(struct conn * conn) -{ - log_dbg("Getting boot information."); - - if (send_rcv_enroll_msg(conn->flow_info.fd)) { - log_err("Failed to enroll."); - return -1; - } - - return 0; -} - -int enroll_done(struct conn * conn, - int result) -{ - enroll_msg_t msg = ENROLL_MSG__INIT; - uint8_t buf[ENROLL_BUF_LEN]; - ssize_t len; - - msg.code = ENROLL_CODE__ENROLL_DONE; - msg.has_result = true; - msg.result = result; - - len = enroll_msg__get_packed_size(&msg); - if (len < 0) { - log_dbg("Failed pack request message."); - return -1; - } - - enroll_msg__pack(&msg, buf); - - if (flow_write(conn->flow_info.fd, buf, len) < 0) { - log_dbg("Failed to send acknowledgment."); - return -1; - } - - return 0; -} - -void enroll_bootstrap(const struct ipcp_config * conf) -{ - assert(conf); - - memcpy(&enroll.conf, conf, sizeof(enroll.conf)); -} - -struct ipcp_config * enroll_get_conf(void) -{ - return &enroll.conf; -} - -int enroll_init(void) -{ - struct conn_info info; - - memset(&info, 0, sizeof(info)); - - strcpy(info.comp_name, ENROLL_COMP); - strcpy(info.protocol, ENROLL_PROTO); - info.pref_version = 1; - info.pref_syntax = PROTO_GPB; - info.addr = 0; - - if (connmgr_comp_init(COMPID_ENROLL, &info)) { - log_err("Failed to register with connmgr."); - return -1; - } - - enroll.state = ENROLL_INIT; - - return 0; -} - -void enroll_fini(void) -{ - if (enroll.state == ENROLL_RUNNING) - pthread_join(enroll.listener, NULL); - - connmgr_comp_fini(COMPID_ENROLL); -} - -int enroll_start(void) -{ - if (pthread_create(&enroll.listener, NULL, enroll_handle, NULL)) - return -1; - - enroll.state = ENROLL_RUNNING; - - return 0; -} - -void enroll_stop(void) -{ - if (enroll.state == ENROLL_RUNNING) - pthread_cancel(enroll.listener); -} diff --git a/src/ipcpd/unicast/enroll.h b/src/ipcpd/unicast/enroll.h deleted file mode 100644 index 804f5d5b..00000000 --- a/src/ipcpd/unicast/enroll.h +++ /dev/null @@ -1,47 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Enrollment Task - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#ifndef OUROBOROS_IPCPD_UNICAST_ENROLL_H -#define OUROBOROS_IPCPD_UNICAST_ENROLL_H - -#include <ouroboros/ipcp.h> - -#include "comp.h" - -int enroll_init(void); - -void enroll_fini(void); - -int enroll_start(void); - -void enroll_stop(void); - -void enroll_bootstrap(const struct ipcp_config * conf); - -int enroll_boot(struct conn * conn); - -int enroll_done(struct conn * conn, - int result); - -struct ipcp_config * enroll_get_conf(void); - -#endif /* OUROBOROS_IPCPD_UNICAST_ENROLL_H */ diff --git a/src/ipcpd/unicast/fa.c b/src/ipcpd/unicast/fa.c index 0d132619..ac2ecaea 100644 --- a/src/ipcpd/unicast/fa.c +++ b/src/ipcpd/unicast/fa.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Flow allocator of the IPC Process * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -31,32 +31,43 @@ #define FA "flow-allocator" #define OUROBOROS_PREFIX FA +#include <ouroboros/atomics.h> +#include <ouroboros/endian.h> #include <ouroboros/logs.h> #include <ouroboros/fqueue.h> #include <ouroboros/errno.h> #include <ouroboros/dev.h> #include <ouroboros/ipcp-dev.h> +#include <ouroboros/np1_flow.h> +#include <ouroboros/qoscube.h> #include <ouroboros/rib.h> #include <ouroboros/random.h> +#include <ouroboros/pthread.h> +#include "addr-auth.h" #include "dir.h" #include "fa.h" #include "psched.h" #include "ipcp.h" #include "dt.h" #include "ca.h" +#include "np1.h" #include <inttypes.h> -#include <pthread.h> #include <stdlib.h> #include <string.h> -#define TIMEOUT 10000 /* nanoseconds */ +#if defined (IPCP_FLOW_STATS) && !defined(CLOCK_REALTIME_COARSE) +#define CLOCK_REALTIME_COARSE CLOCK_REALTIME +#endif + +#define TIMEOUT 10 * MILLION /* nanoseconds */ +#define MSGBUFSZ 32768 -#define FLOW_REQ 0 -#define FLOW_REPLY 1 -#define FLOW_UPDATE 2 -#define MSGBUFSZ 2048 +#define FLOW_REQ 0 +#define FLOW_REPLY 1 +#define FLOW_UPDATE 2 +#define FLOW_IRM_UPDATE 3 #define STAT_FILE_LEN 0 @@ -64,23 +75,23 @@ struct fa_msg { uint64_t s_addr; uint64_t r_eid; uint64_t s_eid; - uint8_t code; - int8_t response; - uint16_t ece; - /* QoS parameters from spec, aligned */ - uint8_t availability; - uint8_t in_order; - uint32_t delay; uint64_t bandwidth; + int32_t response; + uint32_t delay; uint32_t loss; uint32_t ber; uint32_t max_gap; - uint16_t cypher_s; + uint32_t timeout; + uint16_t ece; + uint8_t cap; + uint8_t code; + uint8_t availability; + uint8_t service; } __attribute__((packed)); struct cmd { struct list_head next; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; }; struct fa_flow { @@ -94,16 +105,20 @@ struct fa_flow { size_t p_rcv_f; /* Packets received fail */ size_t b_rcv; /* Bytes received */ size_t b_rcv_f; /* Bytes received fail */ + size_t u_snd; /* Flow updates sent */ + size_t u_rcv; /* Flow updates received */ #endif uint64_t s_eid; /* Local endpoint id */ uint64_t r_eid; /* Remote endpoint id */ uint64_t r_addr; /* Remote address */ void * ctx; /* Congestion avoidance context */ + uint64_t fair; /* SFQ virtual finish tag (bytes) */ + uint8_t l_ecn; /* Local first-hop mark (relaxed) */ }; struct { pthread_rwlock_t flows_lock; - struct fa_flow flows[PROG_MAX_FLOWS]; + struct fa_flow flows[PROC_MAX_FLOWS]; #ifdef IPCP_FLOW_STATS size_t n_flows; #endif @@ -117,23 +132,27 @@ struct { struct psched * psched; } fa; -static int fa_stat_read(const char * path, - char * buf, - size_t len) -{ #ifdef IPCP_FLOW_STATS +static int fa_rib_read(const char * path, + char * buf, + size_t len) +{ struct fa_flow * flow; int fd; - char r_addrstr[20]; - char s_eidstr[20]; - char r_eidstr[20]; - char tmstr[20]; + char r_addrstr[21]; + char s_eidstr[21]; + char r_eidstr[21]; + char tmstr[RIB_TM_STRLEN]; char castr[1024]; + char * entry; struct tm * tm; - fd = atoi(path); + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); + + fd = atoi(entry); - if (fd < 0 || fd > PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -1; if (len < 1536) @@ -154,8 +173,8 @@ static int fa_stat_read(const char * path, sprintf(s_eidstr, "%" PRIu64, flow->s_eid); sprintf(r_eidstr, "%" PRIu64, flow->r_eid); - tm = localtime(&flow->stamp); - strftime(tmstr, sizeof(tmstr), "%F %T", tm); + tm = gmtime(&flow->stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); ca_print_stats(flow->ctx, castr, 1024); @@ -172,6 +191,8 @@ static int fa_stat_read(const char * path, "Received (bytes): %20zu\n" "Receive failed (packets): %20zu\n" "Receive failed (bytes): %20zu\n" + "Sent flow updates (packets): %20zu\n" + "Received flow updates (packets): %20zu\n" "%s", tmstr, r_addrstr, s_eidstr, r_eidstr, @@ -179,22 +200,16 @@ static int fa_stat_read(const char * path, flow->p_snd_f, flow->b_snd_f, flow->p_rcv, flow->b_rcv, flow->b_rcv_f, flow->b_rcv_f, + flow->u_snd, flow->u_rcv, castr); pthread_rwlock_unlock(&fa.flows_lock); return strlen(buf); -#else - (void) path; - (void) buf; - (void) len; - return 0; -#endif } -static int fa_stat_readdir(char *** buf) +static int fa_rib_readdir(char *** buf) { -#ifdef IPCP_FLOW_STATS char entry[RIB_PATH_LEN + 1]; size_t i; int idx = 0; @@ -202,17 +217,15 @@ static int fa_stat_readdir(char *** buf) pthread_rwlock_rdlock(&fa.flows_lock); if (fa.n_flows < 1) { - pthread_rwlock_unlock(&fa.flows_lock); - return 0; + *buf = NULL; + goto no_flows; } *buf = malloc(sizeof(**buf) * fa.n_flows); - if (*buf == NULL) { - pthread_rwlock_unlock(&fa.flows_lock); - return -ENOMEM; - } + if (*buf == NULL) + goto fail_entries; - for (i = 0; i < PROG_MAX_FLOWS; ++i) { + for (i = 0; i < PROC_MAX_FLOWS; ++i) { struct fa_flow * flow; flow = &fa.flows[i]; @@ -222,67 +235,63 @@ static int fa_stat_readdir(char *** buf) sprintf(entry, "%zu", i); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(buf); - pthread_rwlock_unlock(&fa.flows_lock); - return -ENOMEM; - } + if ((*buf)[idx] == NULL) + goto fail_entry; strcpy((*buf)[idx++], entry); } assert((size_t) idx == fa.n_flows); - + no_flows: pthread_rwlock_unlock(&fa.flows_lock); return idx; -#else - (void) buf; - return 0; -#endif + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + + free(*buf); + fail_entries: + pthread_rwlock_unlock(&fa.flows_lock); + return -ENOMEM; } -static int fa_stat_getattr(const char * path, - struct stat * st) +static int fa_rib_getattr(const char * path, + struct rib_attr * attr) { -#ifdef IPCP_FLOW_STATS int fd; + char * entry; struct fa_flow * flow; - fd = atoi(path); + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); - st->st_mode = S_IFREG | 0755; - st->st_nlink = 1; - st->st_uid = getuid(); - st->st_gid = getgid(); + fd = atoi(entry); flow = &fa.flows[fd]; pthread_rwlock_rdlock(&fa.flows_lock); if (flow->stamp != 0) { - st->st_size = 1536; - st->st_mtime = flow->stamp; + attr->size = 1536; + attr->mtime = flow->stamp; } else { - st->st_size = 0; - st->st_mtime = 0; + attr->size = 0; + attr->mtime = 0; } pthread_rwlock_unlock(&fa.flows_lock); -#else - (void) path; - (void) st; -#endif + return 0; } static struct rib_ops r_ops = { - .read = fa_stat_read, - .readdir = fa_stat_readdir, - .getattr = fa_stat_getattr + .read = fa_rib_read, + .readdir = fa_rib_readdir, + .getattr = fa_rib_getattr }; +#endif /* IPCP_FLOW_STATS */ static int eid_to_fd(uint64_t eid) { @@ -291,7 +300,7 @@ static int eid_to_fd(uint64_t eid) fd = eid & 0xFFFFFFFF; - if (fd < 0 || fd > PROG_MAX_FLOWS) + if (fd < 0 || fd >= PROC_MAX_FLOWS) return -1; flow = &fa.flows[fd]; @@ -314,46 +323,53 @@ static uint64_t gen_eid(int fd) return ((uint64_t) rnd << 32) + fd; } -static void packet_handler(int fd, - qoscube_t qc, - struct shm_du_buff * sdb) +static time_t packet_handler(int fd, + qoscube_t qc, + struct ssm_pk_buff * spb) { struct fa_flow * flow; uint64_t r_addr; uint64_t r_eid; - ca_wnd_t wnd; + time_t wait; size_t len; + uint8_t ecn; flow = &fa.flows[fd]; + ecn = 0; + pthread_rwlock_wrlock(&fa.flows_lock); - len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); + len = ssm_pk_buff_len(spb); #ifdef IPCP_FLOW_STATS ++flow->p_snd; flow->b_snd += len; #endif - wnd = ca_ctx_update_snd(flow->ctx, len); + wait = ca_ctx_update_snd(flow->ctx, len, + LOAD_RELAXED(&flow->l_ecn), &flow->fair); r_addr = flow->r_addr; r_eid = flow->r_eid; pthread_rwlock_unlock(&fa.flows_lock); - ca_wnd_wait(wnd); - - if (dt_write_packet(r_addr, qc, r_eid, sdb)) { - ipcp_sdb_release(sdb); - log_warn("Failed to forward packet."); + if (dt_write_packet(r_addr, qc, r_eid, spb, &ecn)) { + STORE_RELAXED(&flow->l_ecn, ecn); + ipcp_spb_release(spb); + log_dbg("Failed to forward packet."); #ifdef IPCP_FLOW_STATS pthread_rwlock_wrlock(&fa.flows_lock); ++flow->p_snd_f; flow->b_snd_f += len; pthread_rwlock_unlock(&fa.flows_lock); #endif - return; + return wait; } + + STORE_RELAXED(&flow->l_ecn, ecn); + + return wait; } static int fa_flow_init(struct fa_flow * flow) @@ -367,9 +383,7 @@ static int fa_flow_init(struct fa_flow * flow) flow->s_eid = -1; flow->r_addr = INVALID_ADDR; - flow->ctx = ca_ctx_create(); - if (flow->ctx == NULL) - return -1; + /* ctx is acquired once (r_addr, qc) are known (ca_ctx_get). */ #ifdef IPCP_FLOW_STATS clock_gettime(CLOCK_REALTIME_COARSE, &now); @@ -383,7 +397,8 @@ static int fa_flow_init(struct fa_flow * flow) static void fa_flow_fini(struct fa_flow * flow) { - ca_ctx_destroy(flow->ctx); + if (flow->ctx != NULL) + ca_ctx_put(flow->ctx); memset(flow, 0, sizeof(*flow)); @@ -397,7 +412,7 @@ static void fa_flow_fini(struct fa_flow * flow) } static void fa_post_packet(void * comp, - struct shm_du_buff * sdb) + struct ssm_pk_buff * spb) { struct cmd * cmd; @@ -408,11 +423,11 @@ static void fa_post_packet(void * comp, cmd = malloc(sizeof(*cmd)); if (cmd == NULL) { log_err("Command failed. Out of memory."); - ipcp_sdb_release(sdb); + ipcp_spb_release(spb); return; } - cmd->sdb = sdb; + cmd->spb = spb; pthread_mutex_lock(&fa.mtx); @@ -423,175 +438,255 @@ static void fa_post_packet(void * comp, pthread_mutex_unlock(&fa.mtx); } -static void * fa_handle_packet(void * o) +static size_t fa_wait_for_fa_msg(struct fa_msg * msg) { - struct timespec ts = {0, TIMEOUT * 1000}; + struct cmd * cmd; + size_t len; - (void) o; + pthread_mutex_lock(&fa.mtx); - while (true) { - struct timespec abstime; - int fd; - uint8_t buf[MSGBUFSZ]; - struct fa_msg * msg; - qosspec_t qs; - struct cmd * cmd; - size_t len; - size_t msg_len; - struct fa_flow * flow; + pthread_cleanup_push(__cleanup_mutex_unlock, &fa.mtx); - pthread_mutex_lock(&fa.mtx); + while (list_is_empty(&fa.cmds)) + pthread_cond_wait(&fa.cond, &fa.mtx); - pthread_cleanup_push((void (*)(void *)) pthread_mutex_unlock, - &fa.mtx); + cmd = list_last_entry(&fa.cmds, struct cmd, next); + list_del(&cmd->next); - while (list_is_empty(&fa.cmds)) - pthread_cond_wait(&fa.cond, &fa.mtx); + pthread_cleanup_pop(true); - cmd = list_last_entry(&fa.cmds, struct cmd, next); - list_del(&cmd->next); + len = ssm_pk_buff_len(cmd->spb); + if (len > MSGBUFSZ || len < sizeof(*msg)) { + log_warn("Invalid flow allocation message (len: %zd).", len); + free(cmd); + return 0; /* No valid message */ + } - pthread_cleanup_pop(true); + memcpy(msg, ssm_pk_buff_head(cmd->spb), len); - len = shm_du_buff_tail(cmd->sdb) - shm_du_buff_head(cmd->sdb); + ipcp_spb_release(cmd->spb); - if (len > MSGBUFSZ) { - log_err("Message over buffer size."); - free(cmd); - continue; - } + free(cmd); - msg = (struct fa_msg *) buf; + return len; +} - /* Depending on the message call the function in ipcp-dev.h */ +static int fa_handle_flow_req(struct fa_msg * msg, + size_t len) +{ + size_t msg_len; + int fd; + qosspec_t qs; + struct fa_flow * flow; + uint8_t * dst; + buffer_t data; /* Piggbacked data on flow alloc request. */ - memcpy(msg, shm_du_buff_head(cmd->sdb), len); + msg_len = sizeof(*msg) + ipcp_dir_hash_len(); + if (len < msg_len) { + log_err("Invalid flow allocation request"); + return -EPERM; + } - ipcp_sdb_release(cmd->sdb); + dst = (uint8_t *)(msg + 1); + data.data = (uint8_t *) msg + msg_len; + data.len = len - msg_len; + + qs.delay = ntoh32(msg->delay); + qs.bandwidth = ntoh64(msg->bandwidth); + qs.availability = msg->availability; + qs.loss = ntoh32(msg->loss); + qs.ber = ntoh32(msg->ber); + qs.service = msg->service; + qs.max_gap = ntoh32(msg->max_gap); + qs.timeout = ntoh32(msg->timeout); + + fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UNICAST_MPL, + IPCP_UNICAST_MTU, &data); + if (fd < 0) + return fd; - free(cmd); + flow = &fa.flows[fd]; - switch (msg->code) { - case FLOW_REQ: - msg_len = sizeof(*msg) + ipcp_dir_hash_len(); + pthread_rwlock_wrlock(&fa.flows_lock); - assert(len >= msg_len); + fa_flow_init(flow); - clock_gettime(PTHREAD_COND_CLOCK, &abstime); + flow->s_eid = gen_eid(fd); + flow->r_eid = ntoh64(msg->s_eid); + flow->r_addr = ntoh64(msg->s_addr); - pthread_mutex_lock(&ipcpi.alloc_lock); + flow->ctx = ca_ctx_get(flow->r_addr, qos_spec_to_cube(qs)); + if (flow->ctx == NULL) { + fa_flow_fini(flow); + pthread_rwlock_unlock(&fa.flows_lock); + return -ENOMEM; + } - while (ipcpi.alloc_id != -1 && - ipcp_get_state() == IPCP_OPERATIONAL) { - ts_add(&abstime, &ts, &abstime); - pthread_cond_timedwait(&ipcpi.alloc_cond, - &ipcpi.alloc_lock, - &abstime); - } + pthread_rwlock_unlock(&fa.flows_lock); - if (ipcp_get_state() != IPCP_OPERATIONAL) { - pthread_mutex_unlock(&ipcpi.alloc_lock); - log_dbg("Won't allocate over non-operational" - "IPCP."); - continue; - } + return fd; +} - assert(ipcpi.alloc_id == -1); +static int fa_handle_flow_reply(struct fa_msg * msg, + size_t len) +{ + int fd; + struct fa_flow * flow; + buffer_t data; /* Piggbacked data on flow alloc request. */ + time_t mpl = IPCP_UNICAST_MPL; + int response; + + if (len < sizeof(*msg)) + return -EINVAL; + + data.data = (uint8_t *) msg + sizeof(*msg); + data.len = len - sizeof(*msg); + + pthread_rwlock_wrlock(&fa.flows_lock); + + fd = eid_to_fd(ntoh64(msg->r_eid)); + if (fd < 0) { + pthread_rwlock_unlock(&fa.flows_lock); + log_err("Flow reply for unknown EID %" PRIu64 ".", + ntoh64(msg->r_eid)); + return -ENOTALLOC; + } - qs.delay = ntoh32(msg->delay); - qs.bandwidth = ntoh64(msg->bandwidth); - qs.availability = msg->availability; - qs.loss = ntoh32(msg->loss); - qs.ber = ntoh32(msg->ber); - qs.in_order = msg->in_order; - qs.max_gap = ntoh32(msg->max_gap); - qs.cypher_s = ntoh16(msg->cypher_s); + flow = &fa.flows[fd]; - fd = ipcp_flow_req_arr((uint8_t *) (msg + 1), - ipcp_dir_hash_len(), - qs, - buf + msg_len, - len - msg_len); - if (fd < 0) { - pthread_mutex_unlock(&ipcpi.alloc_lock); - log_err("Failed to get fd for flow."); - continue; - } + flow->r_eid = ntoh64(msg->s_eid); + response = ntoh32(msg->response); - flow = &fa.flows[fd]; + log_dbg("IPCP received msg response %d for flow on fd %d.", + response, fd); - pthread_rwlock_wrlock(&fa.flows_lock); + if (response < 0) + fa_flow_fini(flow); + else + psched_add(fa.psched, fd); - fa_flow_init(flow); + pthread_rwlock_unlock(&fa.flows_lock); - flow->s_eid = gen_eid(fd); - flow->r_eid = ntoh64(msg->s_eid); - flow->r_addr = ntoh64(msg->s_addr); + if (ipcp_flow_alloc_reply(fd, response, mpl, + IPCP_UNICAST_MTU, &data) < 0) { + log_err("Failed to reply for flow allocation on fd %d.", fd); + return -EIRMD; + } - pthread_rwlock_unlock(&fa.flows_lock); + return 0; +} - ipcpi.alloc_id = fd; - pthread_cond_broadcast(&ipcpi.alloc_cond); +static int fa_handle_flow_update(struct fa_msg * msg, + size_t len) +{ + struct fa_flow * flow; + int fd; - pthread_mutex_unlock(&ipcpi.alloc_lock); + if (len < sizeof(*msg)) + return -EINVAL; - break; - case FLOW_REPLY: - assert(len >= sizeof(*msg)); + pthread_rwlock_wrlock(&fa.flows_lock); - pthread_rwlock_wrlock(&fa.flows_lock); + fd = eid_to_fd(ntoh64(msg->r_eid)); + if (fd < 0) { + pthread_rwlock_unlock(&fa.flows_lock); + log_err("Flow update for unknown EID %" PRIu64 ".", + ntoh64(msg->r_eid)); + return -EPERM; + } - fd = eid_to_fd(ntoh64(msg->r_eid)); - if (fd < 0) { - pthread_rwlock_unlock(&fa.flows_lock); - break; - } + flow = &fa.flows[fd]; +#ifdef IPCP_FLOW_STATS + flow->u_rcv++; +#endif + ca_ctx_update_ece(flow->ctx, ntoh16(msg->ece), msg->cap); - flow = &fa.flows[fd]; + pthread_rwlock_unlock(&fa.flows_lock); - flow->r_eid = ntoh64(msg->s_eid); + return 0; +} - if (msg->response < 0) - fa_flow_fini(flow); - else - psched_add(fa.psched, fd); +static int fa_handle_flow_irm_update(struct fa_msg * msg, + size_t len) +{ + buffer_t data; + int fd; + int flow_id; - pthread_rwlock_unlock(&fa.flows_lock); + if (len < sizeof(*msg)) + return -EINVAL; - ipcp_flow_alloc_reply(fd, - msg->response, - buf + sizeof(*msg), - len - sizeof(*msg)); - break; - case FLOW_UPDATE: - assert(len >= sizeof(*msg)); + data.data = (uint8_t *) msg + sizeof(*msg); + data.len = len - sizeof(*msg); - pthread_rwlock_wrlock(&fa.flows_lock); + pthread_rwlock_rdlock(&fa.flows_lock); + + fd = eid_to_fd(ntoh64(msg->r_eid)); + + pthread_rwlock_unlock(&fa.flows_lock); + + if (fd < 0) { + log_err("Flow update for unknown EID %" PRIu64 ".", + ntoh64(msg->r_eid)); + return -ENOTALLOC; + } - fd = eid_to_fd(ntoh64(msg->r_eid)); - if (fd < 0) { - pthread_rwlock_unlock(&fa.flows_lock); - break; - } + flow_id = np1_flow_id(fd); + if (flow_id < 0) + return -ENOTALLOC; - flow = &fa.flows[fd]; + if (ipcp_flow_update_arr(flow_id, &data) < 0) { + log_err("Failed to relay flow update on fd %d.", fd); + return -EIRMD; + } - ca_ctx_update_ece(flow->ctx, ntoh16(msg->ece)); + return 0; +} - pthread_rwlock_unlock(&fa.flows_lock); +static void * fa_handle_packet(void * o) +{ + (void) o; + while (true) { + uint8_t buf[MSGBUFSZ]; + struct fa_msg * msg; + size_t len; + + msg = (struct fa_msg *) buf; + + len = fa_wait_for_fa_msg(msg); + if (len == 0) + continue; + + switch (msg->code) { + case FLOW_REQ: + if (fa_handle_flow_req(msg, len) < 0) + log_err("Error handling flow alloc request."); + break; + case FLOW_REPLY: + if (fa_handle_flow_reply(msg, len) < 0) + log_err("Error handling flow reply."); + break; + case FLOW_UPDATE: + if (fa_handle_flow_update(msg, len) < 0) + log_err("Error handling flow update."); + break; + case FLOW_IRM_UPDATE: + if (fa_handle_flow_irm_update(msg, len) < 0) + log_err("Error handling flow update."); break; default: - log_err("Got an unknown flow allocation message."); + log_warn("Recieved unknown flow allocation message."); break; } } + + return (void *) 0; } int fa_init(void) { pthread_condattr_t cattr; - char fastr[256]; if (pthread_rwlock_init(&fa.flows_lock, NULL)) goto fail_rwlock; @@ -608,21 +703,26 @@ int fa_init(void) if (pthread_cond_init(&fa.cond, &cattr)) goto fail_cond; - pthread_condattr_destroy(&cattr); - - list_head_init(&fa.cmds); - - sprintf(fastr, "%s", FA); - if (rib_reg(fastr, &r_ops)) +#ifdef IPCP_FLOW_STATS + if (rib_reg(FA, &r_ops)) goto fail_rib_reg; +#endif fa.eid = dt_reg_comp(&fa, &fa_post_packet, FA); if ((int) fa.eid < 0) - goto fail_rib_reg; + goto fail_dt_reg; + + list_head_init(&fa.cmds); + + pthread_condattr_destroy(&cattr); return 0; + fail_dt_reg: +#ifdef IPCP_FLOW_STATS + rib_unreg(FA); fail_rib_reg: +#endif pthread_cond_destroy(&fa.cond); fail_cond: pthread_condattr_destroy(&cattr); @@ -631,24 +731,34 @@ int fa_init(void) fail_mtx: pthread_rwlock_destroy(&fa.flows_lock); fail_rwlock: - log_err("Failed to initialize flow allocator."); return -1; } void fa_fini(void) { +#ifdef IPCP_FLOW_STATS + rib_unreg(FA); +#endif pthread_cond_destroy(&fa.cond);; pthread_mutex_destroy(&fa.mtx); pthread_rwlock_destroy(&fa.flows_lock); } +static int np1_flow_read_fa(int fd, + struct ssm_pk_buff ** spb) +{ + return np1_flow_read(fd, spb, NP1_GET_POOL(fd)); +} + int fa_start(void) { +#ifndef BUILD_CONTAINER struct sched_param par; int pol; int max; +#endif - fa.psched = psched_create(packet_handler); + fa.psched = psched_create(packet_handler, np1_flow_read_fa); if (fa.psched == NULL) { log_err("Failed to start packet scheduler."); goto fail_psched; @@ -659,6 +769,7 @@ int fa_start(void) goto fail_thread; } +#ifndef BUILD_CONTAINER if (pthread_getschedparam(fa.worker, &pol, &par)) { log_err("Failed to get worker thread scheduling parameters."); goto fail_sched; @@ -676,16 +787,18 @@ int fa_start(void) log_err("Failed to set scheduler priority to maximum."); goto fail_sched; } +#endif return 0; +#ifndef BUILD_CONTAINER fail_sched: pthread_cancel(fa.worker); pthread_join(fa.worker, NULL); +#endif fail_thread: psched_destroy(fa.psched); fail_psched: - log_err("Failed to start flow allocator."); return -1; } @@ -697,14 +810,13 @@ void fa_stop(void) psched_destroy(fa.psched); } -int fa_alloc(int fd, - const uint8_t * dst, - qosspec_t qs, - const void * data, - size_t dlen) +int fa_alloc(int fd, + const uint8_t * dst, + qosspec_t qs, + const buffer_t * data) { struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; struct fa_flow * flow; uint64_t addr; qoscube_t qc = QOS_CUBE_BE; @@ -717,31 +829,33 @@ int fa_alloc(int fd, len = sizeof(*msg) + ipcp_dir_hash_len(); - if (ipcp_sdb_reserve(&sdb, len + dlen)) + if (ipcp_spb_reserve(&spb, len + data->len)) return -1; - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); eid = gen_eid(fd); msg->code = FLOW_REQ; msg->s_eid = hton64(eid); - msg->s_addr = hton64(ipcpi.dt_addr); + msg->s_addr = hton64(addr_auth_address()); msg->delay = hton32(qs.delay); msg->bandwidth = hton64(qs.bandwidth); msg->availability = qs.availability; msg->loss = hton32(qs.loss); msg->ber = hton32(qs.ber); - msg->in_order = qs.in_order; + msg->service = qs.service; msg->max_gap = hton32(qs.max_gap); - msg->cypher_s = hton16(qs.cypher_s); + msg->timeout = hton32(qs.timeout); memcpy(msg + 1, dst, ipcp_dir_hash_len()); - memcpy(shm_du_buff_head(sdb) + len, data, dlen); + if (data->len > 0) + memcpy(ssm_pk_buff_head(spb) + len, data->data, data->len); - if (dt_write_packet(addr, qc, fa.eid, sdb)) { - ipcp_sdb_release(sdb); + if (dt_write_packet(addr, qc, fa.eid, spb, NULL)) { + log_err("Failed to send flow allocation request packet."); + ipcp_spb_release(spb); return -1; } @@ -753,79 +867,115 @@ int fa_alloc(int fd, flow->r_addr = addr; flow->s_eid = eid; + flow->ctx = ca_ctx_get(addr, qos_spec_to_cube(qs)); + if (flow->ctx == NULL) { + fa_flow_fini(flow); + pthread_rwlock_unlock(&fa.flows_lock); + return -1; + } + pthread_rwlock_unlock(&fa.flows_lock); return 0; } -int fa_alloc_resp(int fd, - int response, - const void * data, - size_t len) +int fa_alloc_resp(int fd, + int response, + const buffer_t * data) { - struct timespec ts = {0, TIMEOUT * 1000}; - struct timespec abstime; struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; struct fa_flow * flow; qoscube_t qc = QOS_CUBE_BE; - clock_gettime(PTHREAD_COND_CLOCK, &abstime); - flow = &fa.flows[fd]; - pthread_mutex_lock(&ipcpi.alloc_lock); - - while (ipcpi.alloc_id != fd && ipcp_get_state() == IPCP_OPERATIONAL) { - ts_add(&abstime, &ts, &abstime); - pthread_cond_timedwait(&ipcpi.alloc_cond, - &ipcpi.alloc_lock, - &abstime); - } - - if (ipcp_get_state() != IPCP_OPERATIONAL) { - pthread_mutex_unlock(&ipcpi.alloc_lock); - return -1; + if (ipcp_wait_flow_resp(fd) < 0) { + log_err("Failed to wait for flow response."); + goto fail_alloc_resp; } - ipcpi.alloc_id = -1; - pthread_cond_broadcast(&ipcpi.alloc_cond); - - pthread_mutex_unlock(&ipcpi.alloc_lock); - - if (ipcp_sdb_reserve(&sdb, sizeof(*msg) + len)) { - fa_flow_fini(flow); - return -1; + if (ipcp_spb_reserve(&spb, sizeof(*msg) + data->len)) { + log_err("Failed to reserve spb (%zu bytes).", + sizeof(*msg) + data->len); + goto fail_reserve; } - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); - pthread_rwlock_wrlock(&fa.flows_lock); - msg->code = FLOW_REPLY; + msg->response = hton32(response); + if (data->len > 0) + memcpy(msg + 1, data->data, data->len); + + pthread_rwlock_rdlock(&fa.flows_lock); + msg->r_eid = hton64(flow->r_eid); msg->s_eid = hton64(flow->s_eid); - msg->response = response; - memcpy(msg + 1, data, len); + pthread_rwlock_unlock(&fa.flows_lock); + + if (dt_write_packet(flow->r_addr, qc, fa.eid, spb, NULL)) { + log_err("Failed to send flow allocation response packet."); + goto fail_packet; + } if (response < 0) { + pthread_rwlock_wrlock(&fa.flows_lock); fa_flow_fini(flow); - ipcp_sdb_release(sdb); + pthread_rwlock_unlock(&fa.flows_lock); } else { psched_add(fa.psched, fd); } - if (dt_write_packet(flow->r_addr, qc, fa.eid, sdb)) { - fa_flow_fini(flow); - pthread_rwlock_unlock(&fa.flows_lock); - ipcp_sdb_release(sdb); + return 0; + + fail_packet: + ipcp_spb_release(spb); + fail_reserve: + pthread_rwlock_wrlock(&fa.flows_lock); + fa_flow_fini(flow); + pthread_rwlock_unlock(&fa.flows_lock); + fail_alloc_resp: + return -1; +} + +int fa_irm_update(int fd, + const buffer_t * data) +{ + struct fa_msg * msg; + struct ssm_pk_buff * spb; + struct fa_flow * flow; + qoscube_t qc = QOS_CUBE_BE; + uint64_t r_addr; + + flow = &fa.flows[fd]; + + if (ipcp_spb_reserve(&spb, sizeof(*msg) + data->len)) return -1; - } + + msg = (struct fa_msg *) ssm_pk_buff_head(spb); + memset(msg, 0, sizeof(*msg)); + + msg->code = FLOW_IRM_UPDATE; + if (data->len > 0) + memcpy(msg + 1, data->data, data->len); + + pthread_rwlock_rdlock(&fa.flows_lock); + + msg->r_eid = hton64(flow->r_eid); + msg->s_eid = hton64(flow->s_eid); + r_addr = flow->r_addr; pthread_rwlock_unlock(&fa.flows_lock); + if (dt_write_packet(r_addr, qc, fa.eid, spb, NULL)) { + log_err("Failed to send flow update packet."); + ipcp_spb_release(spb); + return -1; + } + return 0; } @@ -842,43 +992,49 @@ int fa_dealloc(int fd) pthread_rwlock_unlock(&fa.flows_lock); - flow_dealloc(fd); + ipcp_flow_dealloc(fd); return 0; } static int fa_update_remote(int fd, - uint16_t ece) + uint16_t ece, + uint8_t cap) { struct fa_msg * msg; - struct shm_du_buff * sdb; + struct ssm_pk_buff * spb; qoscube_t qc = QOS_CUBE_BE; struct fa_flow * flow; uint64_t r_addr; - if (ipcp_sdb_reserve(&sdb, sizeof(*msg))) { + if (ipcp_spb_reserve(&spb, sizeof(*msg))) { + log_err("Failed to reserve spb (%zu bytes).", sizeof(*msg)); return -1; } - msg = (struct fa_msg *) shm_du_buff_head(sdb); + msg = (struct fa_msg *) ssm_pk_buff_head(spb); memset(msg, 0, sizeof(*msg)); flow = &fa.flows[fd]; - pthread_rwlock_rdlock(&fa.flows_lock); + pthread_rwlock_wrlock(&fa.flows_lock); msg->code = FLOW_UPDATE; msg->r_eid = hton64(flow->r_eid); msg->ece = hton16(ece); + msg->cap = cap; r_addr = flow->r_addr; - +#ifdef IPCP_FLOW_STATS + flow->u_snd++; +#endif pthread_rwlock_unlock(&fa.flows_lock); - if (dt_write_packet(r_addr, qc, fa.eid, sdb)) { - ipcp_sdb_release(sdb); + if (dt_write_packet(r_addr, qc, fa.eid, spb, NULL)) { + log_err("Failed to send flow update packet."); + ipcp_spb_release(spb); return -1; } @@ -887,22 +1043,25 @@ static int fa_update_remote(int fd, void fa_np1_rcv(uint64_t eid, uint8_t ecn, - struct shm_du_buff * sdb) + uint8_t cap, + struct ssm_pk_buff * spb) { struct fa_flow * flow; bool update; uint16_t ece; + uint8_t fcap; int fd; size_t len; - len = shm_du_buff_tail(sdb) - shm_du_buff_head(sdb); + len = ssm_pk_buff_len(spb); pthread_rwlock_wrlock(&fa.flows_lock); fd = eid_to_fd(eid); if (fd < 0) { pthread_rwlock_unlock(&fa.flows_lock); - ipcp_sdb_release(sdb); + log_dbg("Received packet for unknown EID %" PRIu64 ".", eid); + ipcp_spb_release(spb); return; } @@ -912,12 +1071,13 @@ void fa_np1_rcv(uint64_t eid, ++flow->p_rcv; flow->b_rcv += len; #endif - update = ca_ctx_update_rcv(flow->ctx, len, ecn, &ece); + update = ca_ctx_update_rcv(flow->ctx, len, ecn, cap, &ece, &fcap); pthread_rwlock_unlock(&fa.flows_lock); - if (ipcp_flow_write(fd, sdb) < 0) { - ipcp_sdb_release(sdb); + if (np1_flow_write(fd, spb, NP1_GET_POOL(fd)) < 0) { + log_dbg("Failed to write to flow %d.", fd); + ipcp_spb_release(spb); #ifdef IPCP_FLOW_STATS pthread_rwlock_wrlock(&fa.flows_lock); ++flow->p_rcv_f; @@ -927,5 +1087,5 @@ void fa_np1_rcv(uint64_t eid, } if (update) - fa_update_remote(eid, ece); + fa_update_remote(eid, ece, fcap); } diff --git a/src/ipcpd/unicast/fa.h b/src/ipcpd/unicast/fa.h index ea06a64e..1d0012b0 100644 --- a/src/ipcpd/unicast/fa.h +++ b/src/ipcpd/unicast/fa.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Flow allocator of the IPC Process * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -34,21 +34,23 @@ int fa_start(void); void fa_stop(void); -int fa_alloc(int fd, - const uint8_t * dst, - qosspec_t qs, - const void * data, - size_t len); +int fa_alloc(int fd, + const uint8_t * dst, + qosspec_t qs, + const buffer_t * data); -int fa_alloc_resp(int fd, - int response, - const void * data, - size_t len); +int fa_alloc_resp(int fd, + int response, + const buffer_t * data); int fa_dealloc(int fd); +int fa_irm_update(int fd, + const buffer_t * data); + void fa_np1_rcv(uint64_t eid, uint8_t ecn, - struct shm_du_buff * sdb); + uint8_t cap, + struct ssm_pk_buff * spb); #endif /* OUROBOROS_IPCPD_UNICAST_FA_H */ diff --git a/src/ipcpd/unicast/kademlia.proto b/src/ipcpd/unicast/kademlia.proto deleted file mode 100644 index 4f807fd3..00000000 --- a/src/ipcpd/unicast/kademlia.proto +++ /dev/null @@ -1,45 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * KAD protocol - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public License - * version 2.1 as published by the Free Software Foundation. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -syntax = "proto2"; - -message kad_contact_msg { - required bytes id = 1; - required uint64 addr = 2; -}; - -message kad_msg { - required uint32 code = 1; - required uint32 cookie = 2; - required uint64 s_addr = 3; - optional bytes s_id = 4; - optional bytes key = 5; - repeated uint64 addrs = 6; - repeated kad_contact_msg contacts = 7; - // enrolment parameters - optional uint32 alpha = 8; - optional uint32 b = 9; - optional uint32 k = 10; - optional uint32 t_expire = 11; - optional uint32 t_refresh = 12; - optional uint32 t_replicate = 13; -};
\ No newline at end of file diff --git a/src/ipcpd/unicast/main.c b/src/ipcpd/unicast/main.c index 1b2cc14e..1155b88b 100644 --- a/src/ipcpd/unicast/main.c +++ b/src/ipcpd/unicast/main.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Unicast IPC Process * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -29,21 +29,22 @@ #include "config.h" #define OUROBOROS_PREFIX "unicast-ipcp" +#define THIS_TYPE IPCP_UNICAST #include <ouroboros/errno.h> -#include <ouroboros/hash.h> #include <ouroboros/ipcp-dev.h> #include <ouroboros/logs.h> #include <ouroboros/notifier.h> +#include <ouroboros/random.h> #include <ouroboros/rib.h> -#include <ouroboros/time_utils.h> +#include <ouroboros/time.h> -#include "addr_auth.h" +#include "common/connmgr.h" +#include "common/enroll.h" +#include "addr-auth.h" #include "ca.h" -#include "connmgr.h" #include "dir.h" #include "dt.h" -#include "enroll.h" #include "fa.h" #include "ipcp.h" @@ -54,239 +55,203 @@ #include <assert.h> #include <inttypes.h> -#define THIS_TYPE IPCP_UNICAST - -struct ipcp ipcpi; - -static int initialize_components(const struct ipcp_config * conf) +static int initialize_components(struct ipcp_config * conf) { - ipcpi.layer_name = strdup(conf->layer_info.layer_name); - if (ipcpi.layer_name == NULL) { - log_err("Failed to set layer name."); - goto fail_layer_name; - } - - ipcpi.dir_hash_algo = conf->layer_info.dir_hash_algo; - assert(ipcp_dir_hash_len() != 0); - if (addr_auth_init(conf->addr_auth_type, - &conf->addr_size)) { + if (addr_auth_init(conf->unicast.addr_auth_type, + &conf->unicast.dt.addr_size)) { log_err("Failed to init address authority."); goto fail_addr_auth; } - ipcpi.dt_addr = addr_auth_address(); - if (ipcpi.dt_addr == 0) { - log_err("Failed to get a valid address."); - goto fail_addr_auth; - } - - log_dbg("IPCP got address %" PRIu64 ".", ipcpi.dt_addr); + log_info("IPCP got address %" PRIu64 ".", addr_auth_address()); - if (ca_init(conf->cong_avoid)) { + if (ca_init(conf->unicast.cong_avoid)) { log_err("Failed to initialize congestion avoidance."); goto fail_ca; } - if (dt_init(conf->routing_type, - conf->addr_size, - conf->eid_size, - conf->max_ttl)) { + if (dt_init(conf->unicast.dt)) { log_err("Failed to initialize data transfer component."); goto fail_dt; } - if (fa_init()) { - log_err("Failed to initialize flow allocator component."); - goto fail_fa; - } + ipcp_set_dir_hash_algo((enum hash_algo) conf->layer_info.dir_hash_algo); - if (dir_init()) { + if (dir_init(&conf->unicast.dir)) { log_err("Failed to initialize directory."); goto fail_dir; } + if (fa_init()) { + log_err("Failed to initialize flow allocator component."); + goto fail_fa; + } + ipcp_set_state(IPCP_INIT); return 0; - fail_dir: - fa_fini(); fail_fa: + dir_fini(); + fail_dir: dt_fini(); fail_dt: ca_fini(); fail_ca: addr_auth_fini(); fail_addr_auth: - free(ipcpi.layer_name); - fail_layer_name: return -1; } static void finalize_components(void) { - dir_fini(); - fa_fini(); + dir_fini(); + dt_fini(); ca_fini(); addr_auth_fini(); - - free(ipcpi.layer_name); } static int start_components(void) { - assert(ipcp_get_state() == IPCP_INIT); + if (connmgr_start() < 0) { + log_err("Failed to start AP connection manager."); + goto fail_connmgr_start; + } - ipcp_set_state(IPCP_OPERATIONAL); + if (dt_start() < 0) { + log_err("Failed to start data transfer."); + goto fail_dt_start; + } - if (fa_start()) { + if (fa_start() < 0) { log_err("Failed to start flow allocator."); goto fail_fa_start; } - if (enroll_start()) { + if (enroll_start() < 0) { log_err("Failed to start enrollment."); goto fail_enroll_start; } - if (connmgr_start()) { - log_err("Failed to start AP connection manager."); - goto fail_connmgr_start; + if (dir_start() < 0) { + log_err("Failed to start directory."); + goto fail_dir_start; } return 0; - fail_connmgr_start: + fail_dir_start: enroll_stop(); fail_enroll_start: fa_stop(); fail_fa_start: + dt_stop(); + fail_dt_start: + connmgr_stop(); + fail_connmgr_start: ipcp_set_state(IPCP_INIT); return -1; } static void stop_components(void) { - assert(ipcp_get_state() == IPCP_OPERATIONAL || - ipcp_get_state() == IPCP_SHUTDOWN); - - connmgr_stop(); + dir_stop(); enroll_stop(); fa_stop(); - ipcp_set_state(IPCP_INIT); -} + dt_stop(); -static int bootstrap_components(void) -{ - if (dir_bootstrap()) { - log_err("Failed to bootstrap directory."); - dt_stop(); - return -1; - } + connmgr_stop(); - return 0; + ipcp_set_state(IPCP_BOOT); } static int unicast_ipcp_enroll(const char * dst, struct layer_info * info) { - struct conn conn; + struct ipcp_config * conf; + struct conn conn; + uint8_t id[ENROLL_ID_LEN]; - if (connmgr_alloc(COMPID_ENROLL, dst, NULL, &conn)) { - log_err("Failed to get connection."); - goto fail_er_flow; + if (random_buffer(id, ENROLL_ID_LEN) < 0) { + log_err("Failed to generate enrollment ID."); + goto fail_id; } - /* Get boot state from peer. */ - if (enroll_boot(&conn)) { - log_err("Failed to get boot information."); - goto fail_enroll_boot; + log_info_id(id, "Requesting enrollment."); + + if (connmgr_alloc(COMPID_ENROLL, dst, NULL, &conn) < 0) { + log_err_id(id, "Failed to get connection."); + goto fail_id; } - if (initialize_components(enroll_get_conf())) { - log_err("Failed to initialize IPCP components."); + /* Get boot state from peer. */ + if (enroll_boot(&conn, id) < 0) { + log_err_id(id, "Failed to get boot information."); goto fail_enroll_boot; } - if (dt_start()) { - log_err("Failed to initialize IPCP components."); - goto fail_dt_start; + conf = enroll_get_conf(); + + *info = conf->layer_info; + + if (initialize_components(conf) < 0) { + log_err_id(id, "Failed to initialize components."); + goto fail_enroll_boot; } - if (start_components()) { - log_err("Failed to start components."); + if (start_components() < 0) { + log_err_id(id, "Failed to start components."); goto fail_start_comp; } - if (enroll_done(&conn, 0)) - log_warn("Failed to confirm enrollment with peer."); - - if (connmgr_dealloc(COMPID_ENROLL, &conn)) - log_warn("Failed to deallocate enrollment flow."); + if (enroll_ack(&conn, id, 0) < 0) + log_err_id(id, "Failed to confirm enrollment."); - log_info("Enrolled with %s.", dst); + if (connmgr_dealloc(COMPID_ENROLL, &conn) < 0) + log_warn_id(id, "Failed to dealloc enrollment flow."); - info->dir_hash_algo = ipcpi.dir_hash_algo; - strcpy(info->layer_name, ipcpi.layer_name); + log_info_id(id, "Enrolled with %s.", dst); return 0; fail_start_comp: - dt_stop(); - fail_dt_start: finalize_components(); fail_enroll_boot: connmgr_dealloc(COMPID_ENROLL, &conn); - fail_er_flow: + fail_id: return -1; } -static int unicast_ipcp_bootstrap(const struct ipcp_config * conf) +static int unicast_ipcp_bootstrap(struct ipcp_config * conf) { assert(conf); assert(conf->type == THIS_TYPE); - enroll_bootstrap(conf); - - if (initialize_components(conf)) { + if (initialize_components(conf) < 0) { log_err("Failed to init IPCP components."); goto fail_init; } - if (dt_start()) { - log_err("Failed to initialize IPCP components."); - goto fail_dt_start; - }; + enroll_bootstrap(conf); - if (start_components()) { + if (start_components() < 0) { log_err("Failed to init IPCP components."); goto fail_start; } - if (bootstrap_components()) { - log_err("Failed to bootstrap IPCP components."); - goto fail_bootstrap; - } - - log_dbg("Bootstrapped in layer %s.", conf->layer_info.layer_name); - return 0; - fail_bootstrap: - stop_components(); fail_start: - dt_stop(); - fail_dt_start: finalize_components(); fail_init: return -1; @@ -308,55 +273,46 @@ static struct ipcp_ops unicast_ops = { .ipcp_flow_alloc = fa_alloc, .ipcp_flow_join = NULL, .ipcp_flow_alloc_resp = fa_alloc_resp, - .ipcp_flow_dealloc = fa_dealloc + .ipcp_flow_dealloc = fa_dealloc, + .ipcp_flow_update = fa_irm_update }; int main(int argc, char * argv[]) { - if (ipcp_init(argc, argv, &unicast_ops) < 0) { + if (ipcp_init(argc, argv, &unicast_ops, THIS_TYPE) < 0) { log_err("Failed to init IPCP."); goto fail_init; } - /* These components must be init at creation. */ - if (rib_init(ipcpi.name)) { - log_err("Failed to initialize RIB."); - goto fail_rib_init; - } - - if (notifier_init()) { + if (notifier_init() < 0) { log_err("Failed to initialize notifier component."); goto fail_notifier_init; } - if (connmgr_init()) { + if (connmgr_init() < 0) { log_err("Failed to initialize connection manager."); goto fail_connmgr_init; } - if (enroll_init()) { + if (enroll_init() < 0) { log_err("Failed to initialize enrollment component."); goto fail_enroll_init; } - if (ipcp_boot() < 0) { - log_err("Failed to boot IPCP."); - goto fail_boot; - } - - if (ipcp_create_r(0)) { - log_err("Failed to notify IRMd we are initialized."); - ipcp_set_state(IPCP_NULL); - goto fail_create_r; + if (ipcp_start() < 0) { + log_err("Failed to start IPCP."); + goto fail_start; } - ipcp_shutdown(); + ipcp_sigwait(); if (ipcp_get_state() == IPCP_SHUTDOWN) { - dt_stop(); + ipcp_stop(); stop_components(); finalize_components(); + } else { + ipcp_stop(); } enroll_fini(); @@ -365,25 +321,18 @@ int main(int argc, notifier_fini(); - rib_fini(); - ipcp_fini(); exit(EXIT_SUCCESS); - fail_create_r: - ipcp_shutdown(); - fail_boot: + fail_start: enroll_fini(); fail_enroll_init: connmgr_fini(); fail_connmgr_init: notifier_fini(); fail_notifier_init: - rib_fini(); - fail_rib_init: - ipcp_fini(); + ipcp_fini(); fail_init: - ipcp_create_r(-1); exit(EXIT_FAILURE); } diff --git a/src/ipcpd/unicast/pff.c b/src/ipcpd/unicast/pff.c index 19432972..c8c3126f 100644 --- a/src/ipcpd/unicast/pff.c +++ b/src/ipcpd/unicast/pff.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * PDU Forwarding Function * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -26,14 +26,11 @@ #include <ouroboros/logs.h> #include "pff.h" -#include "pol-pff-ops.h" -#include "pol/alternate_pff.h" -#include "pol/multipath_pff.h" -#include "pol/simple_pff.h" +#include "pff/pol.h" struct pff { - struct pol_pff_ops * ops; - struct pff_i * pff_i; + struct pff_ops * ops; + struct pff_i * pff_i; }; struct pff * pff_create(enum pol_pff pol) @@ -62,8 +59,10 @@ struct pff * pff_create(enum pol_pff pol) } pff->pff_i = pff->ops->create(); - if (pff->pff_i == NULL) + if (pff->pff_i == NULL) { + log_err("Failed to create PFF instance."); goto err; + } return pff; err: diff --git a/src/ipcpd/unicast/pff.h b/src/ipcpd/unicast/pff.h index 962ae594..2eb42ce6 100644 --- a/src/ipcpd/unicast/pff.h +++ b/src/ipcpd/unicast/pff.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * PDU Forwarding Function * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as diff --git a/src/ipcpd/unicast/pol/alternate_pff.c b/src/ipcpd/unicast/pff/alternate.c index f26bb047..1c508c1b 100644 --- a/src/ipcpd/unicast/pol/alternate_pff.c +++ b/src/ipcpd/unicast/pff/alternate.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF with alternate next hops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -28,7 +28,7 @@ #include <ouroboros/list.h> #include "pft.h" -#include "alternate_pff.h" +#include "alternate.h" #include <string.h> #include <assert.h> @@ -54,7 +54,7 @@ struct pff_i { pthread_rwlock_t lock; }; -struct pol_pff_ops alternate_pff_ops = { +struct pff_ops alternate_pff_ops = { .create = alternate_pff_create, .destroy = alternate_pff_destroy, .lock = alternate_pff_lock, @@ -211,7 +211,7 @@ struct pff_i * alternate_pff_create(void) if (pthread_rwlock_init(&tmp->lock, NULL)) goto fail_lock; - tmp->pft = pft_create(PFT_SIZE, false); + tmp->pft = pft_create(PFT_SIZE); if (tmp->pft == NULL) goto fail_pft; diff --git a/src/ipcpd/unicast/pol/alternate_pff.h b/src/ipcpd/unicast/pff/alternate.h index b3d9c83c..ae3758dc 100644 --- a/src/ipcpd/unicast/pol/alternate_pff.h +++ b/src/ipcpd/unicast/pff/alternate.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF with alternate next hops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,7 +23,7 @@ #ifndef OUROBOROS_IPCPD_UNICAST_ALTERNATE_PFF_H #define OUROBOROS_IPCPD_UNICAST_ALTERNATE_PFF_H -#include "pol-pff-ops.h" +#include "ops.h" struct pff_i * alternate_pff_create(void); @@ -56,6 +56,6 @@ int alternate_flow_state_change(struct pff_i * pff_i, int fd, bool up); -extern struct pol_pff_ops alternate_pff_ops; +extern struct pff_ops alternate_pff_ops; #endif /* OUROBOROS_IPCPD_UNICAST_ALTERNATE_PFF_H */ diff --git a/src/ipcpd/unicast/pol/multipath_pff.c b/src/ipcpd/unicast/pff/multipath.c index 0fe101bc..9ba59592 100644 --- a/src/ipcpd/unicast/pol/multipath_pff.c +++ b/src/ipcpd/unicast/pff/multipath.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF supporting multipath routing * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * Nick Aerts <nick.aerts@ugent.be> * * This program is free software; you can redistribute it and/or modify @@ -28,7 +28,7 @@ #include <ouroboros/errno.h> #include "pft.h" -#include "multipath_pff.h" +#include "multipath.h" #include <string.h> #include <assert.h> @@ -39,7 +39,7 @@ struct pff_i { pthread_rwlock_t lock; }; -struct pol_pff_ops multipath_pff_ops = { +struct pff_ops multipath_pff_ops = { .create = multipath_pff_create, .destroy = multipath_pff_destroy, .lock = multipath_pff_lock, @@ -58,21 +58,23 @@ struct pff_i * multipath_pff_create(void) tmp = malloc(sizeof(*tmp)); if (tmp == NULL) - return NULL; + goto fail_malloc; - if (pthread_rwlock_init(&tmp->lock, NULL)) { - free(tmp); - return NULL; - } + if (pthread_rwlock_init(&tmp->lock, NULL)) + goto fail_rwlock; - tmp->pft = pft_create(PFT_SIZE, false); - if (tmp->pft == NULL) { - pthread_rwlock_destroy(&tmp->lock); - free(tmp); - return NULL; - } + tmp->pft = pft_create(PFT_SIZE); + if (tmp->pft == NULL) + goto fail_pft; return tmp; + + fail_pft: + pthread_rwlock_destroy(&tmp->lock); + fail_rwlock: + free(tmp); + fail_malloc: + return NULL; } void multipath_pff_destroy(struct pff_i * pff_i) @@ -80,8 +82,8 @@ void multipath_pff_destroy(struct pff_i * pff_i) assert(pff_i); pft_destroy(pff_i->pft); - pthread_rwlock_destroy(&pff_i->lock); + free(pff_i); } @@ -177,7 +179,7 @@ int multipath_pff_nhop(struct pff_i * pff_i, assert(pff_i); - pthread_rwlock_rdlock(&pff_i->lock); + pthread_rwlock_wrlock(&pff_i->lock); if (pft_lookup(pff_i->pft, addr, &fds, &len)) { pthread_rwlock_unlock(&pff_i->lock); @@ -189,7 +191,7 @@ int multipath_pff_nhop(struct pff_i * pff_i, assert(len > 0); /* Rotate fds left. */ - memcpy(fds, fds + 1, (len - 1) * sizeof(*fds)); + memmove(fds, fds + 1, (len - 1) * sizeof(*fds)); fds[len - 1] = fd; pthread_rwlock_unlock(&pff_i->lock); diff --git a/src/ipcpd/unicast/pol/multipath_pff.h b/src/ipcpd/unicast/pff/multipath.h index ecbb664a..5329f7fc 100644 --- a/src/ipcpd/unicast/pol/multipath_pff.h +++ b/src/ipcpd/unicast/pff/multipath.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Policy for PFF supporting multipath routing * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * Nick Aerts <nick.aerts@ugent.be> * * This program is free software; you can redistribute it and/or modify @@ -24,7 +24,7 @@ #ifndef OUROBOROS_IPCPD_UNICAST_MULTIPATH_PFF_H #define OUROBOROS_IPCPD_UNICAST_MULTIPATH_PFF_H -#include "pol-pff-ops.h" +#include "ops.h" struct pff_i * multipath_pff_create(void); @@ -53,6 +53,6 @@ void multipath_pff_flush(struct pff_i * pff_i); int multipath_pff_nhop(struct pff_i * pff_i, uint64_t addr); -extern struct pol_pff_ops multipath_pff_ops; +extern struct pff_ops multipath_pff_ops; #endif /* OUROBOROS_IPCPD_UNICAST_MULTIPATH_PFF_H */ diff --git a/src/ipcpd/unicast/pol-pff-ops.h b/src/ipcpd/unicast/pff/ops.h index 9e126cb5..e4cabd11 100644 --- a/src/ipcpd/unicast/pol-pff-ops.h +++ b/src/ipcpd/unicast/pff/ops.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Pff policy ops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,14 +20,14 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#ifndef OUROBOROS_IPCPD_UNICAST_POL_PFF_OPS_H -#define OUROBOROS_IPCPD_UNICAST_POL_PFF_OPS_H +#ifndef OUROBOROS_IPCPD_UNICAST_PFF_OPS_H +#define OUROBOROS_IPCPD_UNICAST_PFF_OPS_H #include <stdbool.h> struct pff_i; -struct pol_pff_ops { +struct pff_ops { struct pff_i * (* create)(void); void (* destroy)(struct pff_i * pff_i); @@ -60,4 +60,4 @@ struct pol_pff_ops { bool up); }; -#endif /* OUROBOROS_IPCPD_UNICAST_POL_PFF_OPS_H */ +#endif /* OUROBOROS_IPCPD_UNICAST_PFF_OPS_H */ diff --git a/src/ipcpd/unicast/pol/pft.c b/src/ipcpd/unicast/pff/pft.c index 53acc08e..d0e562d6 100644 --- a/src/ipcpd/unicast/pol/pft.c +++ b/src/ipcpd/unicast/pff/pft.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet forwarding table (PFT) with chaining on collisions * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -43,12 +43,10 @@ struct pft_entry { struct pft { struct list_head * buckets; - bool hash_key; uint64_t buckets_size; }; -struct pft * pft_create(uint64_t buckets, - bool hash_key) +struct pft * pft_create(uint64_t buckets) { struct pft * tmp; unsigned int i; @@ -69,7 +67,6 @@ struct pft * pft_create(uint64_t buckets, if (tmp == NULL) return NULL; - tmp->hash_key = hash_key; tmp->buckets_size = buckets; tmp->buckets = malloc(buckets * sizeof(*tmp->buckets)); @@ -113,30 +110,10 @@ void pft_flush(struct pft * pft) } } -static uint64_t hash(uint64_t key) -{ - void * res; - uint64_t ret; - uint8_t keys[4]; - - memcpy(keys, &key, 4); - - mem_hash(HASH_MD5, &res, keys, 4); - - ret = (* (uint64_t *) res); - - free(res); - - return ret; -} - static uint64_t calc_key(struct pft * pft, uint64_t dst) { - if (pft->hash_key) - dst = hash(dst); - - return (dst & (pft->buckets_size - 1)); + return hash_mix64(dst) & (pft->buckets_size - 1); } int pft_insert(struct pft * pft, diff --git a/src/ipcpd/unicast/pol/pft.h b/src/ipcpd/unicast/pff/pft.h index aed4dba8..15bbe451 100644 --- a/src/ipcpd/unicast/pol/pft.h +++ b/src/ipcpd/unicast/pff/pft.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet forwarding table (PFT) with chaining on collisions * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -24,14 +24,12 @@ #define OUROBOROS_PFT_H #include <stdint.h> -#include <stdbool.h> #include <stdlib.h> struct pft; /* Buckets is rounded up to the nearest power of 2 */ -struct pft * pft_create(uint64_t buckets, - bool hash_key); +struct pft * pft_create(uint64_t buckets); void pft_destroy(struct pft * table); diff --git a/src/ipcpd/unicast/pff/pol.h b/src/ipcpd/unicast/pff/pol.h new file mode 100644 index 00000000..e31c2794 --- /dev/null +++ b/src/ipcpd/unicast/pff/pol.h @@ -0,0 +1,25 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * PDU Forwarding Function policies + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "alternate.h" +#include "multipath.h" +#include "simple.h" diff --git a/src/ipcpd/unicast/pol/simple_pff.c b/src/ipcpd/unicast/pff/simple.c index 5bd73d8a..7befa42f 100644 --- a/src/ipcpd/unicast/pol/simple_pff.c +++ b/src/ipcpd/unicast/pff/simple.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Simple PDU Forwarding Function * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -27,7 +27,7 @@ #include <ouroboros/errno.h> #include "pft.h" -#include "simple_pff.h" +#include "simple.h" #include <assert.h> #include <pthread.h> @@ -37,7 +37,7 @@ struct pff_i { pthread_rwlock_t lock; }; -struct pol_pff_ops simple_pff_ops = { +struct pff_ops simple_pff_ops = { .create = simple_pff_create, .destroy = simple_pff_destroy, .lock = simple_pff_lock, @@ -63,7 +63,7 @@ struct pff_i * simple_pff_create(void) return NULL; } - tmp->pft = pft_create(PFT_SIZE, false); + tmp->pft = pft_create(PFT_SIZE); if (tmp->pft == NULL) { pthread_rwlock_destroy(&tmp->lock); free(tmp); diff --git a/src/ipcpd/unicast/pol/simple_pff.h b/src/ipcpd/unicast/pff/simple.h index 5c972347..1046e4c4 100644 --- a/src/ipcpd/unicast/pol/simple_pff.h +++ b/src/ipcpd/unicast/pff/simple.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Simple policy for PFF * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -23,7 +23,7 @@ #ifndef OUROBOROS_IPCPD_UNICAST_SIMPLE_PFF_H #define OUROBOROS_IPCPD_UNICAST_SIMPLE_PFF_H -#include "pol-pff-ops.h" +#include "ops.h" struct pff_i * simple_pff_create(void); @@ -52,6 +52,6 @@ void simple_pff_flush(struct pff_i * pff_i); int simple_pff_nhop(struct pff_i * pff_i, uint64_t addr); -extern struct pol_pff_ops simple_pff_ops; +extern struct pff_ops simple_pff_ops; #endif /* OUROBOROS_IPCPD_UNICAST_SIMPLE_PFF_H */ diff --git a/src/ipcpd/unicast/pff/tests/CMakeLists.txt b/src/ipcpd/unicast/pff/tests/CMakeLists.txt new file mode 100644 index 00000000..8c0e3d51 --- /dev/null +++ b/src/ipcpd/unicast/pff/tests/CMakeLists.txt @@ -0,0 +1,34 @@ +get_filename_component(CURRENT_SOURCE_PARENT_DIR + ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(CURRENT_BINARY_PARENT_DIR + ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) + +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + pft_test.c + ) + +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) + +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/pol/tests/pft_test.c b/src/ipcpd/unicast/pff/tests/pft_test.c index 4e23898b..20e73a94 100644 --- a/src/ipcpd/unicast/pol/tests/pft_test.c +++ b/src/ipcpd/unicast/pff/tests/pft_test.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the hash table * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -38,15 +38,7 @@ int pft_test(int argc, (void) argc; (void) argv; - pft = pft_create(TBL_SIZE, true); - if (pft == NULL) { - printf("Failed to create.\n"); - return -1; - } - - pft_destroy(pft); - - pft = pft_create(TBL_SIZE, false); + pft = pft_create(TBL_SIZE); if (pft == NULL) { printf("Failed to create.\n"); return -1; diff --git a/src/ipcpd/unicast/pol/ca-mb-ecn.c b/src/ipcpd/unicast/pol/ca-mb-ecn.c deleted file mode 100644 index 9560e211..00000000 --- a/src/ipcpd/unicast/pol/ca-mb-ecn.c +++ /dev/null @@ -1,271 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Multi-bit ECN Congestion Avoidance - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#if defined(__linux__) || defined(__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 200809L -#endif - -#include "config.h" - -#include <ouroboros/ipcp-dev.h> -#include <ouroboros/time_utils.h> - -#include "ca-mb-ecn.h" - -#include <stdlib.h> -#include <string.h> -#include <stdio.h> - -/* congestion avoidance constants */ -#define CA_SHFT 5 /* Average over 32 pkts */ -#define CA_WND (1 << CA_SHFT) /* 32 pkts receiver wnd */ -#define CA_UPD (1 << (CA_SHFT - 3)) /* Update snd every 8 pkt */ -#define CA_SLOT 24 /* Initial slot = 16 ms */ -#define CA_INC 1 << 16 /* ~4MiB/s^2 additive inc */ -#define CA_IWL 1 << 16 /* Initial limit ~4MiB/s */ -#define CA_MINPS 8 /* Mimimum pkts / slot */ -#define CA_MAXPS 64 /* Maximum pkts / slot */ -#define ECN_Q_SHFT 4 -#define ts_to_ns(ts) (ts.tv_sec * BILLION + ts.tv_nsec) - -struct mb_ecn_ctx { - uint16_t rx_ece; /* Level of congestion (upstream) */ - size_t rx_ctr; /* Receiver side packet counter */ - - uint16_t tx_ece; /* Level of congestion (downstream) */ - size_t tx_ctr; /* Sender side packet counter */ - size_t tx_wbc; /* Window byte count */ - size_t tx_wpc; /* Window packet count */ - size_t tx_wbl; /* Window byte limit */ - bool tx_cav; /* Congestion avoidance */ - size_t tx_mul; /* Slot size multiplier */ - size_t tx_inc; /* Additive increase */ - size_t tx_slot; -}; - -struct pol_ca_ops mb_ecn_ca_ops = { - .ctx_create = mb_ecn_ctx_create, - .ctx_destroy = mb_ecn_ctx_destroy, - .ctx_update_snd = mb_ecn_ctx_update_snd, - .ctx_update_rcv = mb_ecn_ctx_update_rcv, - .ctx_update_ece = mb_ecn_ctx_update_ece, - .wnd_wait = mb_ecn_wnd_wait, - .calc_ecn = mb_ecn_calc_ecn, - .print_stats = mb_ecn_print_stats -}; - -void * mb_ecn_ctx_create(void) -{ - struct timespec now; - struct mb_ecn_ctx * ctx; - - ctx = malloc(sizeof(*ctx)); - if (ctx == NULL) - return NULL; - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - memset(ctx, 0, sizeof(*ctx)); - - ctx->tx_mul = CA_SLOT; - ctx->tx_wbl = CA_IWL; - ctx->tx_inc = CA_INC; - ctx->tx_slot = ts_to_ns(now) >> ctx->tx_mul; - - return (void *) ctx; -} - -void mb_ecn_ctx_destroy(void * ctx) -{ - free(ctx); -} - -ca_wnd_t mb_ecn_ctx_update_snd(void * _ctx, - size_t len) -{ - struct timespec now; - size_t slot; - ca_wnd_t wnd; - struct mb_ecn_ctx * ctx = _ctx; - - clock_gettime(PTHREAD_COND_CLOCK, &now); - - slot = ts_to_ns(now) >> ctx->tx_mul; - - ctx->tx_ctr++; - ctx->tx_wpc++; - ctx->tx_wbc += len; - - if (ctx->tx_ctr > CA_WND) - ctx->tx_ece = 0; - - if (slot > ctx->tx_slot) { - ctx->tx_slot = slot; - - if (!ctx->tx_cav) { /* Slow start */ - if (ctx->tx_wbc > ctx->tx_wbl) - ctx->tx_wbl <<= 1; - } else { - if (ctx->tx_ece) /* Mult. Decrease */ - ctx->tx_wbl -= (ctx->tx_wbl * ctx->tx_ece) - >> (CA_SHFT + 8); - else if (ctx->tx_wbc > ctx->tx_wbl) /* Add. Increase */ - ctx->tx_wbl = ctx->tx_wbl + ctx->tx_inc; - } - - /* Window scaling */ - if (ctx->tx_wpc < CA_MINPS) { - ++ctx->tx_mul; - ctx->tx_slot >>= 1; - ctx->tx_wbl <<= 1; - ctx->tx_inc <<= 1; - } - - if (ctx->tx_wpc > CA_MAXPS) { - --ctx->tx_mul; /* Underflows at ~CA_MAXPS billion pps */ - ctx->tx_slot <<= 1; - ctx->tx_wbl >>= 1; - ctx->tx_inc >>= 1; - } - - ctx->tx_wbc = 0; - ctx->tx_wpc = 0; - } - - if (ctx->tx_wbc > ctx->tx_wbl) - wnd.wait = ((ctx->tx_slot + 1) << ctx->tx_mul) - ts_to_ns(now); - else - wnd.wait = 0; - - return wnd; -} - -void mb_ecn_wnd_wait(ca_wnd_t wnd) -{ - if (wnd.wait > 0) { - struct timespec s = {0, 0}; - if (wnd.wait > BILLION) /* Don't care throttling < 1s */ - s.tv_sec = 1; - else - s.tv_nsec = wnd.wait; - - nanosleep(&s, NULL); - } -} - -bool mb_ecn_ctx_update_rcv(void * _ctx, - size_t len, - uint8_t ecn, - uint16_t * ece) -{ - struct mb_ecn_ctx* ctx = _ctx; - bool update; - - (void) len; - - if ((ctx->rx_ece | ecn) == 0) - return false; - - if (ecn == 0) { /* End of congestion */ - ctx->rx_ece >>= 2; - update = ctx->rx_ece == 0; - } else { - if (ctx->rx_ece == 0) { /* Start of congestion */ - ctx->rx_ece = ecn; - ctx->rx_ctr = 0; - update = true; - } else { /* Congestion update */ - ctx->rx_ece -= ctx->rx_ece >> CA_SHFT; - ctx->rx_ece += ecn; - update = (ctx->rx_ctr++ & (CA_UPD - 1)) == true; - } - } - - *ece = ctx->rx_ece; - - return update; -} - - -void mb_ecn_ctx_update_ece(void * _ctx, - uint16_t ece) -{ - struct mb_ecn_ctx* ctx = _ctx; - - ctx->tx_ece = ece; - ctx->tx_ctr = 0; - ctx->tx_cav = true; -} - -int mb_ecn_calc_ecn(int fd, - uint8_t * ecn, - size_t len) -{ - size_t q; - - (void) len; - - q = ipcp_flow_queued(fd); - - *ecn |= (uint8_t) (q >> ECN_Q_SHFT); - - return 0; -} - -ssize_t mb_ecn_print_stats(void * _ctx, - char * buf, - size_t len) -{ - struct mb_ecn_ctx* ctx = _ctx; - char * regime; - - if (len < 1024) - return 0; - - if (!ctx->tx_cav) - regime = "Slow start"; - else if (ctx->tx_ece) - regime = "Multiplicative dec"; - else - regime = "Additive inc"; - - sprintf(buf, - "Congestion avoidance algorithm: %20s\n" - "Upstream congestion level: %20u\n" - "Upstream packet counter: %20zu\n" - "Downstream congestion level: %20u\n" - "Downstream packet counter: %20zu\n" - "Congestion window size (ns): %20zu\n" - "Packets in this window: %20zu\n" - "Bytes in this window: %20zu\n" - "Max bytes in this window: %20zu\n" - "Current congestion regime: %20s\n", - "Multi-bit ECN", - ctx->rx_ece, ctx->rx_ctr, - ctx->tx_ece, ctx->tx_ctr, (size_t) (1 << ctx->tx_mul), - ctx->tx_wpc, ctx->tx_wbc, ctx->tx_wbl, - regime); - - return strlen(buf); -} diff --git a/src/ipcpd/unicast/pol/tests/CMakeLists.txt b/src/ipcpd/unicast/pol/tests/CMakeLists.txt deleted file mode 100644 index 34d80e8d..00000000 --- a/src/ipcpd/unicast/pol/tests/CMakeLists.txt +++ /dev/null @@ -1,35 +0,0 @@ -get_filename_component(CURRENT_SOURCE_PARENT_DIR - ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(CURRENT_BINARY_PARENT_DIR - ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) - -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - -get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) -get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) - -create_test_sourcelist(${PARENT_DIR}_tests test_suite.c - # Add new tests here - graph_test.c - pft_test.c - ) - -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) - -add_dependencies(check ${PARENT_DIR}_test) - -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) - -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) diff --git a/src/ipcpd/unicast/psched.c b/src/ipcpd/unicast/psched.c index 898e6b27..2b535d67 100644 --- a/src/ipcpd/unicast/psched.c +++ b/src/ipcpd/unicast/psched.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet scheduler component * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -30,10 +30,11 @@ #include <ouroboros/errno.h> #include <ouroboros/notifier.h> +#include <ouroboros/time.h> +#include "common/connmgr.h" #include "ipcp.h" #include "psched.h" -#include "connmgr.h" #include <assert.h> #include <sched.h> @@ -41,38 +42,157 @@ #include <stdlib.h> #include <string.h> +#ifndef BUILD_CONTAINER static int qos_prio [] = { QOS_PRIO_BE, QOS_PRIO_VIDEO, QOS_PRIO_VOICE, }; +#endif struct psched { - fset_t * set[QOS_CUBE_MAX]; + fset_t * set[QOS_CUBE_MAX * IPCP_SCHED_THR_MUL]; next_packet_fn_t callback; + read_fn_t read; pthread_t readers[QOS_CUBE_MAX * IPCP_SCHED_THR_MUL]; }; struct sched_info { struct psched * sch; qoscube_t qc; + size_t idx; }; +/* Map an FD to one reader's set: one FD, one thread (no shared FDs). */ +static size_t fd_set_idx(int fd, qoscube_t qc) +{ + return qc + ((size_t) fd % IPCP_SCHED_THR_MUL) * QOS_CUBE_MAX; +} + static void cleanup_reader(void * o) { fqueue_destroy((fqueue_t *) o); } +/* + * Per-reader deadline scheduler: a paced flow is served, then deferred + * to its next-send deadline instead of blocking the thread, so it never + * stalls its thread-mates. + */ +struct dsched { + uint64_t deadline[PROC_MAX_FLOWS]; /* absolute ns, per tracked fd */ + int active[PROC_MAX_FLOWS]; /* compact list of tracked fds */ + int posn[PROC_MAX_FLOWS]; /* fd -> active index, -1 = none */ + size_t n; +}; + +static void cleanup_dsched(void * o) +{ + free(o); +} + +static void dsched_track(struct dsched * d, + int fd, + uint64_t deadline) +{ + if (d->posn[fd] >= 0) + return; + + d->deadline[fd] = deadline; + d->posn[fd] = (int) d->n; + d->active[d->n++] = fd; +} + +static void dsched_untrack(struct dsched * d, + int fd) +{ + int i = d->posn[fd]; + + if (i < 0) + return; + + d->active[i] = d->active[--d->n]; + d->posn[d->active[i]] = i; + d->posn[fd] = -1; +} + +static uint64_t dsched_serve(struct dsched * d, + struct psched * sched, + qoscube_t qc, + uint64_t now) +{ + struct ssm_pk_buff * spb; + uint64_t dmin = 0; + size_t i; + int fd; + time_t wait; + + for (i = 0; i < d->n; ) { + fd = d->active[i]; + + if (d->deadline[fd] > now) { + if (dmin == 0 || d->deadline[fd] < dmin) + dmin = d->deadline[fd]; + ++i; + continue; + } + + if (sched->read(fd, &spb) < 0) { + dsched_untrack(d, fd); + continue; + } + + wait = sched->callback(fd, qc, spb); + if (wait == 0) + continue; + + d->deadline[fd] = now + (uint64_t) wait; + if (dmin == 0 || d->deadline[fd] < dmin) + dmin = d->deadline[fd]; + ++i; + } + + return dmin; +} + +static void dsched_events(struct dsched * d, + fqueue_t * fq, + uint64_t now) +{ + int fd; + + while ((fd = fqueue_next(fq)) >= 0) { + switch (fqueue_type(fq)) { + case FLOW_DEALLOC: + dsched_untrack(d, fd); + notifier_event(NOTIFY_DT_FLOW_DEALLOC, &fd); + break; + case FLOW_DOWN: + notifier_event(NOTIFY_DT_FLOW_DOWN, &fd); + break; + case FLOW_UP: + notifier_event(NOTIFY_DT_FLOW_UP, &fd); + break; + case FLOW_PKT: + dsched_track(d, fd, now); + break; + default: + break; + } + } +} + static void * packet_reader(void * o) { - struct psched * sched; - struct shm_du_buff * sdb; - int fd; - fqueue_t * fq; - qoscube_t qc; + struct psched * sched; + struct dsched * d; + fqueue_t * fq; + qoscube_t qc; + size_t idx; sched = ((struct sched_info *) o)->sch; qc = ((struct sched_info *) o)->qc; + idx = ((struct sched_info *) o)->idx; ipcp_lock_to_core(); @@ -82,42 +202,57 @@ static void * packet_reader(void * o) if (fq == NULL) return (void *) -1; + d = malloc(sizeof(*d)); + if (d == NULL) { + fqueue_destroy(fq); + return (void *) -1; + } + + memset(d, 0, sizeof(*d)); + memset(d->posn, 0xFF, sizeof(d->posn)); /* -1: nothing tracked yet */ + + pthread_cleanup_push(cleanup_dsched, d); pthread_cleanup_push(cleanup_reader, fq); while (true) { - int ret = fevent(sched->set[qc], fq, NULL); + struct timespec now_ts; + struct timespec to; + struct timespec * timeo; + uint64_t now; + uint64_t dmin; + uint64_t delta; + int ret; + + clock_gettime(PTHREAD_COND_CLOCK, &now_ts); + + now = TS_TO_UINT64(now_ts); + + dmin = dsched_serve(d, sched, qc, now); + + if (dmin == 0) { + timeo = NULL; + } else { + delta = dmin > now ? dmin - now : 1; + to.tv_sec = (time_t) (delta / BILLION); + to.tv_nsec = (long) (delta % BILLION); + timeo = &to; + } + + ret = fevent(sched->set[idx], fq, timeo); if (ret < 0) continue; - while ((fd = fqueue_next(fq)) >= 0) { - switch (fqueue_type(fq)) { - case FLOW_DEALLOC: - notifier_event(NOTIFY_DT_FLOW_DEALLOC, &fd); - break; - case FLOW_DOWN: - notifier_event(NOTIFY_DT_FLOW_DOWN, &fd); - break; - case FLOW_UP: - notifier_event(NOTIFY_DT_FLOW_UP, &fd); - break; - case FLOW_PKT: - if (ipcp_flow_read(fd, &sdb)) - continue; - - sched->callback(fd, qc, sdb); - break; - default: - break; - } - } + dsched_events(d, fq, now); } pthread_cleanup_pop(true); + pthread_cleanup_pop(true); return (void *) 0; } -struct psched * psched_create(next_packet_fn_t callback) +struct psched * psched_create(next_packet_fn_t callback, + read_fn_t read) { struct psched * psched; struct sched_info * infos[QOS_CUBE_MAX * IPCP_SCHED_THR_MUL]; @@ -131,8 +266,9 @@ struct psched * psched_create(next_packet_fn_t callback) goto fail_malloc; psched->callback = callback; + psched->read = read; - for (i = 0; i < QOS_CUBE_MAX; ++i) { + for (i = 0; i < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++i) { psched->set[i] = fset_create(); if (psched->set[i] == NULL) { for (j = 0; j < i; ++j) @@ -150,6 +286,7 @@ struct psched * psched_create(next_packet_fn_t callback) } infos[i]->sch = psched; infos[i]->qc = i % QOS_CUBE_MAX; + infos[i]->idx = i; } for (i = 0; i < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++i) { @@ -165,6 +302,7 @@ struct psched * psched_create(next_packet_fn_t callback) } } +#ifndef BUILD_CONTAINER for (i = 0; i < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++i) { struct sched_param par; int pol = SCHED_RR; @@ -182,16 +320,20 @@ struct psched * psched_create(next_packet_fn_t callback) if (pthread_setschedparam(psched->readers[i], pol, &par)) goto fail_sched; } +#endif return psched; +#ifndef BUILD_CONTAINER fail_sched: for (j = 0; j < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++j) pthread_cancel(psched->readers[j]); + for (j = 0; j < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++j) pthread_join(psched->readers[j], NULL); +#endif fail_infos: - for (j = 0; j < QOS_CUBE_MAX; ++j) + for (j = 0; j < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++j) fset_destroy(psched->set[j]); fail_flow_set: free(psched); @@ -210,7 +352,7 @@ void psched_destroy(struct psched * psched) pthread_join(psched->readers[i], NULL); } - for (i = 0; i < QOS_CUBE_MAX; ++i) + for (i = 0; i < QOS_CUBE_MAX * IPCP_SCHED_THR_MUL; ++i) fset_destroy(psched->set[i]); free(psched); @@ -224,7 +366,7 @@ void psched_add(struct psched * psched, assert(psched); ipcp_flow_get_qoscube(fd, &qc); - fset_add(psched->set[qc], fd); + fset_add(psched->set[fd_set_idx(fd, qc)], fd); } void psched_del(struct psched * psched, @@ -235,5 +377,5 @@ void psched_del(struct psched * psched, assert(psched); ipcp_flow_get_qoscube(fd, &qc); - fset_del(psched->set[qc], fd); + fset_del(psched->set[fd_set_idx(fd, qc)], fd); } diff --git a/src/ipcpd/unicast/psched.h b/src/ipcpd/unicast/psched.h index 85f32b9a..8c2914b3 100644 --- a/src/ipcpd/unicast/psched.h +++ b/src/ipcpd/unicast/psched.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Packet scheduler component * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -26,11 +26,15 @@ #include <ouroboros/ipcp-dev.h> #include <ouroboros/fqueue.h> -typedef void (* next_packet_fn_t)(int fd, - qoscube_t qc, - struct shm_du_buff * sdb); +typedef time_t (* next_packet_fn_t)(int fd, + qoscube_t qc, + struct ssm_pk_buff * spb); -struct psched * psched_create(next_packet_fn_t callback); +typedef int (* read_fn_t)(int fd, + struct ssm_pk_buff ** spb); + +struct psched * psched_create(next_packet_fn_t callback, + read_fn_t read); void psched_destroy(struct psched * psched); diff --git a/src/ipcpd/unicast/routing.c b/src/ipcpd/unicast/routing.c index 0ac43f9f..1a4e4372 100644 --- a/src/ipcpd/unicast/routing.c +++ b/src/ipcpd/unicast/routing.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing component of the IPCP * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -26,35 +26,30 @@ #include "pff.h" #include "routing.h" -#include "pol/link_state.h" +#include "routing/pol.h" -struct pol_routing_ops * r_ops; +struct routing_ops * r_ops; -int routing_init(enum pol_routing pr) +int routing_init(struct routing_config * conf, + enum pol_pff * pff_type) { - enum pol_pff pff_type; + void * cfg; - switch (pr) { + switch (conf->pol) { case ROUTING_LINK_STATE: - pff_type = PFF_SIMPLE; - r_ops = &link_state_ops; - break; - case ROUTING_LINK_STATE_LFA: - pff_type = PFF_ALTERNATE; - r_ops = &link_state_ops; - break; - case ROUTING_LINK_STATE_ECMP: - pff_type=PFF_MULTIPATH; r_ops = &link_state_ops; + cfg = &conf->ls; break; default: return -ENOTSUP; } - if (r_ops->init(pr)) - return -1; + return r_ops->init(cfg, pff_type); +} - return pff_type; +int routing_start(void) +{ + return r_ops->start(); } struct routing_i * routing_i_create(struct pff * pff) @@ -67,6 +62,11 @@ void routing_i_destroy(struct routing_i * instance) return r_ops->routing_i_destroy(instance); } +void routing_stop(void) +{ + r_ops->stop(); +} + void routing_fini(void) { r_ops->fini(); diff --git a/src/ipcpd/unicast/routing.h b/src/ipcpd/unicast/routing.h index 38e875e7..8d721095 100644 --- a/src/ipcpd/unicast/routing.h +++ b/src/ipcpd/unicast/routing.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing component of the IPCP * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -30,10 +30,15 @@ #include <stdint.h> -int routing_init(enum pol_routing pr); +int routing_init(struct routing_config * conf, + enum pol_pff * pff_type); void routing_fini(void); +int routing_start(void); + +void routing_stop(void); + struct routing_i * routing_i_create(struct pff * pff); void routing_i_destroy(struct routing_i * instance); diff --git a/src/ipcpd/unicast/pol/graph.c b/src/ipcpd/unicast/routing/graph.c index ba2ce553..c168eb7d 100644 --- a/src/ipcpd/unicast/pol/graph.c +++ b/src/ipcpd/unicast/routing/graph.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Undirected graph structure * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * Nick Aerts <nick.aerts@ugent.be> * * This program is free software; you can redistribute it and/or modify @@ -57,8 +57,8 @@ struct edge { }; struct graph { - size_t nr_vertices; - struct list_head vertices; + struct llist vertices; + pthread_mutex_t lock; }; @@ -67,7 +67,7 @@ static struct edge * find_edge_by_addr(struct vertex * vertex, { struct list_head * p; - assert(vertex); + assert(vertex != NULL); list_for_each(p, &vertex->edges) { struct edge * e = list_entry(p, struct edge, next); @@ -85,7 +85,7 @@ static struct vertex * find_vertex_by_addr(struct graph * graph, assert(graph); - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * e = list_entry(p, struct vertex, next); if (e->addr == addr) return e; @@ -99,8 +99,8 @@ static struct edge * add_edge(struct vertex * vertex, { struct edge * edge; - assert(vertex); - assert(nb); + assert(vertex != NULL); + assert(nb != NULL); edge = malloc(sizeof(*edge)); if (edge == NULL) @@ -139,7 +139,7 @@ static struct vertex * add_vertex(struct graph * graph, vertex->addr = addr; /* Keep them ordered on address. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > addr) break; @@ -148,46 +148,50 @@ static struct vertex * add_vertex(struct graph * graph, vertex->index = i; - list_add_tail(&vertex->next, p); + llist_add_tail_at(&vertex->next, p, &graph->vertices); /* Increase the index of the vertices to the right. */ - list_for_each(p, &graph->vertices) { + list_for_each(p, &vertex->next) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > addr) v->index++; } - graph->nr_vertices++; - return vertex; } +static void free_edges(struct list_head * edges) +{ + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, edges) { + struct edge * e = list_entry(p, struct edge, next); + list_del(&e->next); + free(e); + } +} + static void del_vertex(struct graph * graph, struct vertex * vertex) { struct list_head * p; - struct list_head * h; - assert(graph); - assert(vertex); + assert(graph != NULL); + assert(vertex != NULL); - list_del(&vertex->next); + llist_del(&vertex->next, &graph->vertices); /* Decrease the index of the vertices to the right. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { struct vertex * v = list_entry(p, struct vertex, next); if (v->addr > vertex->addr) v->index--; } - list_for_each_safe(p, h, &vertex->edges) { - struct edge * e = list_entry(p, struct edge, next); - del_edge(e); - } + free_edges(&vertex->edges); free(vertex); - - graph->nr_vertices--; } struct graph * graph_create(void) @@ -203,8 +207,7 @@ struct graph * graph_create(void) return NULL; } - graph->nr_vertices = 0; - list_head_init(&graph->vertices); + llist_init(&graph->vertices); return graph; } @@ -218,7 +221,7 @@ void graph_destroy(struct graph * graph) pthread_mutex_lock(&graph->lock); - list_for_each_safe(p, n, &graph->vertices) { + llist_for_each_safe(p, n, &graph->vertices) { struct vertex * e = list_entry(p, struct vertex, next); del_vertex(graph, e); } @@ -227,6 +230,8 @@ void graph_destroy(struct graph * graph) pthread_mutex_destroy(&graph->lock); + assert(llist_is_empty(&graph->vertices)); + free(graph); } @@ -240,63 +245,35 @@ int graph_update_edge(struct graph * graph, struct vertex * nb; struct edge * nb_e; - assert(graph); + assert(graph != NULL); pthread_mutex_lock(&graph->lock); v = find_vertex_by_addr(graph, s_addr); - if (v == NULL) { - v = add_vertex(graph, s_addr); - if (v == NULL) { - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add vertex."); - return -ENOMEM; - } + if (v == NULL && ((v = add_vertex(graph, s_addr)) == NULL)) {; + log_err("Failed to add src vertex."); + goto fail_add_s; } nb = find_vertex_by_addr(graph, d_addr); - if (nb == NULL) { - nb = add_vertex(graph, d_addr); - if (nb == NULL) { - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add vertex."); - return -ENOMEM; - } + if (nb == NULL && ((nb = add_vertex(graph, d_addr)) == NULL)) { + log_err("Failed to add dst vertex."); + goto fail_add_d; } e = find_edge_by_addr(v, d_addr); - if (e == NULL) { - e = add_edge(v, nb); - if (e == NULL) { - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - if (list_is_empty(&nb->edges)) - del_vertex(graph, nb); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add edge."); - return -ENOMEM; - } + if (e == NULL && ((e = add_edge(v, nb)) == NULL)) { + log_err("Failed to add edge to dst."); + goto fail_add_edge_d; } e->announced++; e->qs = qs; nb_e = find_edge_by_addr(nb, s_addr); - if (nb_e == NULL) { - nb_e = add_edge(nb, v); - if (nb_e == NULL) { - if (--e->announced == 0) - del_edge(e); - if (list_is_empty(&v->edges)) - del_vertex(graph, v); - if (list_is_empty(&nb->edges)) - del_vertex(graph, nb); - pthread_mutex_unlock(&graph->lock); - log_err("Failed to add edge."); - return -ENOMEM; - } + if (nb_e == NULL && ((nb_e = add_edge(nb, v)) == NULL)) {; + log_err("Failed to add edge to src."); + goto fail_add_edge_s; } nb_e->announced++; @@ -305,6 +282,19 @@ int graph_update_edge(struct graph * graph, pthread_mutex_unlock(&graph->lock); return 0; + fail_add_edge_s: + if (--e->announced == 0) + del_edge(e); + fail_add_edge_d: + if (list_is_empty(&nb->edges)) + del_vertex(graph, nb); + fail_add_d: + if (list_is_empty(&v->edges)) + del_vertex(graph, v); + fail_add_s: + pthread_mutex_unlock(&graph->lock); + return -ENOMEM; + } int graph_del_edge(struct graph * graph, @@ -322,30 +312,26 @@ int graph_del_edge(struct graph * graph, v = find_vertex_by_addr(graph, s_addr); if (v == NULL) { - pthread_mutex_unlock(&graph->lock); - log_err("No such source vertex."); - return -1; + log_err("Failed to find src vertex."); + goto fail; } nb = find_vertex_by_addr(graph, d_addr); if (nb == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such destination vertex."); - return -1; + goto fail; } e = find_edge_by_addr(v, d_addr); if (e == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such source edge."); - return -1; + goto fail; } nb_e = find_edge_by_addr(nb, s_addr); if (nb_e == NULL) { - pthread_mutex_unlock(&graph->lock); log_err("No such destination edge."); - return -1; + goto fail; } if (--e->announced == 0) @@ -362,6 +348,10 @@ int graph_del_edge(struct graph * graph, pthread_mutex_unlock(&graph->lock); return 0; + + fail: + pthread_mutex_unlock(&graph->lock); + return -1; } static int get_min_vertex(struct graph * graph, @@ -381,7 +371,7 @@ static int get_min_vertex(struct graph * graph, *v = NULL; - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { if (!used[i] && dist[i] < min) { min = dist[i]; index = i; @@ -413,24 +403,24 @@ static int dijkstra(struct graph * graph, assert(nhops); assert(dist); - *nhops = malloc(sizeof(**nhops) * graph->nr_vertices); + *nhops = malloc(sizeof(**nhops) * graph->vertices.len); if (*nhops == NULL) goto fail_pnhops; - *dist = malloc(sizeof(**dist) * graph->nr_vertices); + *dist = malloc(sizeof(**dist) * graph->vertices.len); if (*dist == NULL) goto fail_pdist; - used = malloc(sizeof(*used) * graph->nr_vertices); + used = malloc(sizeof(*used) * graph->vertices.len); if (used == NULL) goto fail_used; /* Init the data structures */ - memset(used, 0, sizeof(*used) * graph->nr_vertices); - memset(*nhops, 0, sizeof(**nhops) * graph->nr_vertices); - memset(*dist, 0, sizeof(**dist) * graph->nr_vertices); + memset(used, 0, sizeof(*used) * graph->vertices.len); + memset(*nhops, 0, sizeof(**nhops) * graph->vertices.len); + memset(*dist, 0, sizeof(**dist) * graph->vertices.len); - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); (*dist)[i++] = (v->addr == src) ? 0 : INT_MAX; } @@ -527,7 +517,7 @@ static int graph_routing_table_simple(struct graph * graph, assert(dist); /* We need at least 2 vertices for a table */ - if (graph->nr_vertices < 2) + if (graph->vertices.len < 2) goto fail_vertices; if (dijkstra(graph, s_addr, &nhops, dist)) @@ -536,7 +526,7 @@ static int graph_routing_table_simple(struct graph * graph, list_head_init(table); /* Now construct the routing table from the nhops. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); /* This is the src */ @@ -613,9 +603,9 @@ static int graph_routing_table_lfa(struct graph * graph, struct list_head * table, int ** dist) { - int * n_dist[PROG_MAX_FLOWS]; - uint64_t addrs[PROG_MAX_FLOWS]; - int n_index[PROG_MAX_FLOWS]; + int * n_dist[PROC_MAX_FLOWS]; + uint64_t addrs[PROC_MAX_FLOWS]; + int n_index[PROC_MAX_FLOWS]; struct list_head * p; struct list_head * q; struct vertex * v; @@ -628,13 +618,13 @@ static int graph_routing_table_lfa(struct graph * graph, if (graph_routing_table_simple(graph, s_addr, table, dist)) goto fail_table; - for (j = 0; j < PROG_MAX_FLOWS; j++) { + for (j = 0; j < PROC_MAX_FLOWS; j++) { n_dist[j] = NULL; n_index[j] = -1; addrs[j] = -1; } - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); if (v->addr != s_addr) @@ -660,7 +650,7 @@ static int graph_routing_table_lfa(struct graph * graph, } /* Loop though all nodes to see if we have a LFA for them. */ - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); if (v->addr == s_addr) @@ -705,7 +695,6 @@ static int graph_routing_table_ecmp(struct graph * graph, { struct vertex ** nhops; struct list_head * p; - struct list_head * h; size_t i; struct vertex * v; struct vertex * src_v; @@ -717,14 +706,14 @@ static int graph_routing_table_ecmp(struct graph * graph, assert(graph); assert(dist); - if (graph-> nr_vertices < 2) + if (graph->vertices.len < 2) goto fail_vertices; - forwarding = malloc(sizeof(*forwarding) * graph->nr_vertices); + forwarding = malloc(sizeof(*forwarding) * graph->vertices.len); if (forwarding == NULL) goto fail_vertices; - for (i = 0; i < graph->nr_vertices; ++i) + for (i = 0; i < graph->vertices.len; ++i) list_head_init(&forwarding[i]); if (dijkstra(graph, s_addr, &nhops, dist)) @@ -745,16 +734,15 @@ static int graph_routing_table_ecmp(struct graph * graph, free(nhops); - list_for_each(h, &graph->vertices) { - v = list_entry(h, struct vertex, next); - if (tmp_dist[v->index] + 1 == (*dist)[v->index]) { + for (i = 0; i < graph->vertices.len; ++i) { + if (tmp_dist[i] + 1 == (*dist)[i]) { n = malloc(sizeof(*n)); if (n == NULL) { free(tmp_dist); goto fail_src_v; } n->nhop = e->nb->addr; - list_add_tail(&n->next, &forwarding[v->index]); + list_add_tail(&n->next, &forwarding[i]); } } @@ -763,38 +751,34 @@ static int graph_routing_table_ecmp(struct graph * graph, list_head_init(table); i = 0; - list_for_each(p, &graph->vertices) { + llist_for_each(p, &graph->vertices) { v = list_entry(p, struct vertex, next); - if (v->addr == s_addr) { + if (v->addr == s_addr || list_is_empty(&forwarding[i])) { ++i; continue; } t = malloc(sizeof(*t)); if (t == NULL) - goto fail_t; + goto fail_malloc; t->dst = v->addr; list_head_init(&t->nhops); - if (&forwarding[i] != forwarding[i].nxt) { - t->nhops.nxt = forwarding[i].nxt; - t->nhops.prv = forwarding[i].prv; - forwarding[i].prv->nxt = &t->nhops; - forwarding[i].nxt->prv = &t->nhops; - } + t->nhops.nxt = forwarding[i].nxt; + t->nhops.prv = forwarding[i].prv; + forwarding[i].prv->nxt = &t->nhops; + forwarding[i].nxt->prv = &t->nhops; list_add(&t->next, table); ++i; } - free(*dist); - *dist = NULL; free(forwarding); return 0; - fail_t: + fail_malloc: free_routing_table(table); fail_src_v: free(*dist); @@ -834,7 +818,7 @@ int graph_routing_table(struct graph * graph, break; default: log_err("Unsupported algorithm."); - goto fail_algo; + goto fail_table; } pthread_mutex_unlock(&graph->lock); @@ -843,8 +827,6 @@ int graph_routing_table(struct graph * graph, return 0; - fail_algo: - free(s_dist); fail_table: pthread_mutex_unlock(&graph->lock); return -1; diff --git a/src/ipcpd/unicast/pol/graph.h b/src/ipcpd/unicast/routing/graph.h index 473a5163..f3766771 100644 --- a/src/ipcpd/unicast/pol/graph.h +++ b/src/ipcpd/unicast/routing/graph.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Undirected graph structure * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as diff --git a/src/ipcpd/unicast/pol/link_state.c b/src/ipcpd/unicast/routing/link-state.c index ca8a7c50..c4ea9e1c 100644 --- a/src/ipcpd/unicast/pol/link_state.c +++ b/src/ipcpd/unicast/routing/link-state.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Link state routing policy * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -38,32 +38,36 @@ #include <ouroboros/list.h> #include <ouroboros/logs.h> #include <ouroboros/notifier.h> +#include <ouroboros/pthread.h> #include <ouroboros/rib.h> #include <ouroboros/utils.h> -#include "comp.h" -#include "connmgr.h" +#include "addr-auth.h" +#include "common/comp.h" +#include "common/connmgr.h" #include "graph.h" #include "ipcp.h" -#include "link_state.h" +#include "link-state.h" #include "pff.h" #include <assert.h> #include <stdlib.h> #include <inttypes.h> #include <string.h> -#include <pthread.h> -#define RECALC_TIME 4 -#define LS_UPDATE_TIME 15 -#define LS_TIMEO 60 #define LS_ENTRY_SIZE 104 -#define LSDB "lsdb" +#define lsdb "lsdb" #ifndef CLOCK_REALTIME_COARSE #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif +#define LINK_FMT ADDR_FMT32 "--" ADDR_FMT32 +#define LINK_VAL(src, dst) ADDR_VAL32(&src), ADDR_VAL32(&dst) + +#define LSU_FMT "LSU ["ADDR_FMT32 " -- " ADDR_FMT32 " seq: %09" PRIu64 "]" +#define LSU_VAL(src, dst, seqno) ADDR_VAL32(&src), ADDR_VAL32(&dst), seqno + struct lsa { uint64_t d_addr; uint64_t s_addr; @@ -106,30 +110,37 @@ struct nb { }; struct { - struct list_head nbs; - size_t nbs_len; + uint64_t addr; + + enum routing_algo routing_algo; + + struct ls_config conf; + fset_t * mgmt_set; - struct list_head db; - size_t db_len; + struct graph * graph; - pthread_rwlock_t db_lock; + struct { + struct llist nbs; + struct llist db; + pthread_rwlock_t lock; + }; - struct graph * graph; + struct { + struct list_head list; + pthread_mutex_t mtx; + } instances; pthread_t lsupdate; pthread_t lsreader; pthread_t listener; - - struct list_head routing_instances; - pthread_mutex_t routing_i_lock; - - enum routing_algo routing_algo; } ls; -struct pol_routing_ops link_state_ops = { - .init = link_state_init, +struct routing_ops link_state_ops = { + .init = (int (*)(void *, enum pol_pff *)) link_state_init, .fini = link_state_fini, + .start = link_state_start, + .stop = link_state_stop, .routing_i_create = link_state_routing_i_create, .routing_i_destroy = link_state_routing_i_destroy }; @@ -138,7 +149,7 @@ static int str_adj(struct adjacency * adj, char * buf, size_t len) { - char tmbuf[64]; + char tmstr[RIB_TM_STRLEN]; char srcbuf[64]; char dstbuf[64]; char seqnobuf[64]; @@ -149,15 +160,16 @@ static int str_adj(struct adjacency * adj, if (len < LS_ENTRY_SIZE) return -1; - tm = localtime(&adj->stamp); - strftime(tmbuf, sizeof(tmbuf), "%F %T", tm); /* 19 chars */ + tm = gmtime(&adj->stamp); + strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - sprintf(srcbuf, "%" PRIu64, adj->src); - sprintf(dstbuf, "%" PRIu64, adj->dst); + sprintf(srcbuf, ADDR_FMT32, ADDR_VAL32(&adj->src)); + sprintf(dstbuf, ADDR_FMT32, ADDR_VAL32(&adj->dst)); sprintf(seqnobuf, "%" PRIu64, adj->seqno); - sprintf(buf, "src: %20s\ndst: %20s\nseqno: %18s\nupd: %20s\n", - srcbuf, dstbuf, seqnobuf, tmbuf); + sprintf(buf, "src: %20s\ndst: %20s\nseqno: %18s\n" + "upd: %s\n", + srcbuf, dstbuf, seqnobuf, tmstr); return LS_ENTRY_SIZE; } @@ -169,9 +181,9 @@ static struct adjacency * get_adj(const char * path) assert(path); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - sprintf(entry, "%" PRIu64 ".%" PRIu64, a->src, a->dst); + sprintf(entry, LINK_FMT, LINK_VAL(a->src, a->dst)); if (strcmp(entry, path) == 0) return a; } @@ -179,53 +191,56 @@ static struct adjacency * get_adj(const char * path) return NULL; } -static int lsdb_getattr(const char * path, - struct stat * st) +static int lsdb_rib_getattr(const char * path, + struct rib_attr * attr) { struct adjacency * adj; struct timespec now; + char * entry; assert(path); - assert(st); + assert(attr); + + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - adj = get_adj(path); + adj = get_adj(entry); if (adj != NULL) { - st->st_mtime = adj->stamp; - st->st_size = LS_ENTRY_SIZE; + attr->mtime = adj->stamp; + attr->size = LS_ENTRY_SIZE; } else { - st->st_mtime = now.tv_sec; - st->st_size = 0; + attr->mtime = now.tv_sec; + attr->size = 0; } - st->st_mode = S_IFREG | 0755; - st->st_nlink = 1; - st->st_uid = getuid(); - st->st_gid = getgid(); - - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return 0; } -static int lsdb_read(const char * path, - char * buf, - size_t len) +static int lsdb_rib_read(const char * path, + char * buf, + size_t len) { struct adjacency * a; + char * entry; int size; assert(path); - pthread_rwlock_rdlock(&ls.db_lock); + entry = strstr(path, RIB_SEPARATOR) + 1; + assert(entry); - if (ls.db_len + ls.nbs_len == 0) + pthread_rwlock_rdlock(&ls.lock); + + if (llist_is_empty(&ls.db) && llist_is_empty(&ls.nbs)) goto fail; - a = get_adj(path); + a = get_adj(entry); if (a == NULL) goto fail; @@ -233,80 +248,72 @@ static int lsdb_read(const char * path, if (size < 0) goto fail; - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return size; fail: - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -1; } -static int lsdb_readdir(char *** buf) +static int lsdb_rib_readdir(char *** buf) { struct list_head * p; char entry[RIB_PATH_LEN + 1]; ssize_t idx = 0; - assert(buf); + assert(buf != NULL); - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - if (ls.db_len + ls.nbs_len == 0) { - pthread_rwlock_unlock(&ls.db_lock); - return 0; + if (llist_is_empty(&ls.db) && llist_is_empty(&ls.nbs)) { + *buf = NULL; + goto no_entries; } - *buf = malloc(sizeof(**buf) * (ls.db_len + ls.nbs_len)); - if (*buf == NULL) { - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } - list_for_each(p, &ls.nbs) { + *buf = malloc(sizeof(**buf) * (ls.db.len + ls.nbs.len)); + if (*buf == NULL) + goto fail_entries; + + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - char * str = (nb->type == NB_DT ? "dt." : "mgmt."); - sprintf(entry, "%s%" PRIu64, str, nb->addr); + char * str = (nb->type == NB_DT ? ".dt " : ".mgmt "); + sprintf(entry, "%s" ADDR_FMT32 , str, ADDR_VAL32(&nb->addr)); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - while (idx-- > 0) - free((*buf)[idx]); - free(buf); - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } - - strcpy((*buf)[idx], entry); + if ((*buf)[idx] == NULL) + goto fail_entry; - idx++; + strcpy((*buf)[idx++], entry); } - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - sprintf(entry, "%" PRIu64 ".%" PRIu64, a->src, a->dst); + sprintf(entry, LINK_FMT, LINK_VAL(a->src, a->dst)); (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) { - ssize_t j; - for (j = 0; j < idx; ++j) - free(*buf[j]); - free(buf); - pthread_rwlock_unlock(&ls.db_lock); - return -ENOMEM; - } + if ((*buf)[idx] == NULL) + goto fail_entry; - strcpy((*buf)[idx], entry); - - idx++; + strcpy((*buf)[idx++], entry); } - - pthread_rwlock_unlock(&ls.db_lock); + no_entries: + pthread_rwlock_unlock(&ls.lock); return idx; + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + free(*buf); + fail_entries: + pthread_rwlock_unlock(&ls.lock); + return -ENOMEM; } static struct rib_ops r_ops = { - .read = lsdb_read, - .readdir = lsdb_readdir, - .getattr = lsdb_getattr + .read = lsdb_rib_read, + .readdir = lsdb_rib_readdir, + .getattr = lsdb_rib_getattr }; static int lsdb_add_nb(uint64_t addr, @@ -316,28 +323,28 @@ static int lsdb_add_nb(uint64_t addr, struct list_head * p; struct nb * nb; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * el = list_entry(p, struct nb, next); - if (el->addr == addr && el->type == type) { - log_dbg("Already know %s neighbor %" PRIu64 ".", - type == NB_DT ? "dt" : "mgmt", addr); - if (el->fd != fd) { - log_warn("Existing neighbor assigned new fd."); - el->fd = fd; - } - pthread_rwlock_unlock(&ls.db_lock); - return -EPERM; - } - if (addr > el->addr) break; + if (el->addr != addr || el->type != type) + continue; + + log_dbg("Already know %s neighbor " ADDR_FMT32 ".", + type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); + if (el->fd != fd) { + log_warn("Existing neighbor assigned new fd."); + el->fd = fd; + } + pthread_rwlock_unlock(&ls.lock); + return -EPERM; } nb = malloc(sizeof(*nb)); if (nb == NULL) { - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -ENOMEM; } @@ -345,14 +352,12 @@ static int lsdb_add_nb(uint64_t addr, nb->fd = fd; nb->type = type; - list_add_tail(&nb->next, p); - - ++ls.nbs_len; + llist_add_tail_at(&nb->next, p, &ls.nbs); - log_dbg("Type %s neighbor %" PRIu64 " added.", - nb->type == NB_DT ? "dt" : "mgmt", addr); + log_dbg("Type %s neighbor " ADDR_FMT32 " added.", + nb->type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return 0; } @@ -363,22 +368,22 @@ static int lsdb_del_nb(uint64_t addr, struct list_head * p; struct list_head * h; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.nbs) { + llist_for_each_safe(p, h, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->addr == addr && nb->fd == fd) { - list_del(&nb->next); - --ls.nbs_len; - pthread_rwlock_unlock(&ls.db_lock); - log_dbg("Type %s neighbor %" PRIu64 " deleted.", - nb->type == NB_DT ? "dt" : "mgmt", addr); - free(nb); - return 0; - } + if (nb->addr != addr || nb->fd != fd) + continue; + + llist_del(&nb->next, &ls.nbs); + pthread_rwlock_unlock(&ls.lock); + log_dbg("Type %s neighbor " ADDR_FMT32 " deleted.", + nb->type == NB_DT ? "dt" : "mgmt", ADDR_VAL32(&addr)); + free(nb); + return 0; } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -EPERM; } @@ -388,18 +393,18 @@ static int nbr_to_fd(uint64_t addr) struct list_head * p; int fd; - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); if (nb->addr == addr && nb->type == NB_DT) { fd = nb->fd; - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return fd; } } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -1; } @@ -410,12 +415,11 @@ static void calculate_pff(struct routing_i * instance) struct list_head table; struct list_head * p; struct list_head * q; - int fds[PROG_MAX_FLOWS]; + int fds[PROC_MAX_FLOWS]; assert(instance); - if (graph_routing_table(ls.graph, ls.routing_algo, - ipcpi.dt_addr, &table)) + if (graph_routing_table(ls.graph, ls.routing_algo, ls.addr, &table)) return; pff_lock(instance->pff); @@ -450,8 +454,8 @@ static void set_pff_modified(bool calc) { struct list_head * p; - pthread_mutex_lock(&ls.routing_i_lock); - list_for_each(p, &ls.routing_instances) { + pthread_mutex_lock(&ls.instances.mtx); + list_for_each(p, &ls.instances.list) { struct routing_i * inst = list_entry(p, struct routing_i, next); pthread_mutex_lock(&inst->lock); @@ -460,7 +464,7 @@ static void set_pff_modified(bool calc) if (calc) calculate_pff(inst); } - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); } static int lsdb_add_link(uint64_t src, @@ -477,9 +481,9 @@ static int lsdb_add_link(uint64_t src, clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); if (a->dst == dst && a->src == src) { if (a->seqno < seqno) { @@ -487,7 +491,7 @@ static int lsdb_add_link(uint64_t src, a->seqno = seqno; ret = 0; } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return ret; } @@ -497,7 +501,7 @@ static int lsdb_add_link(uint64_t src, adj = malloc(sizeof(*adj)); if (adj == NULL) { - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -ENOMEM; } @@ -506,14 +510,12 @@ static int lsdb_add_link(uint64_t src, adj->seqno = seqno; adj->stamp = now.tv_sec; - list_add_tail(&adj->next, p); - - ls.db_len++; + llist_add_tail_at(&adj->next, p, &ls.db); if (graph_update_edge(ls.graph, src, dst, *qs)) log_warn("Failed to add edge to graph."); - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); set_pff_modified(true); @@ -526,25 +528,23 @@ static int lsdb_del_link(uint64_t src, struct list_head * p; struct list_head * h; - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); if (a->dst == dst && a->src == src) { - list_del(&a->next); + llist_del(&a->next, &ls.db); if (graph_del_edge(ls.graph, src, dst)) log_warn("Failed to delete edge from graph."); - ls.db_len--; - - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); set_pff_modified(false); free(a); return 0; } } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); return -EPERM; } @@ -567,7 +567,7 @@ static void * periodic_recalc_pff(void * o) if (modified) calculate_pff(inst); - sleep(RECALC_TIME); + sleep(ls.conf.t_recalc); } return (void *) 0; @@ -584,10 +584,20 @@ static void send_lsm(uint64_t src, lsm.s_addr = hton64(src); lsm.seqno = hton64(seqno); - list_for_each(p, &ls.nbs) { + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->type == NB_MGMT) - flow_write(nb->fd, &lsm, sizeof(lsm)); + if (nb->type != NB_MGMT) + continue; + + if (flow_write(nb->fd, &lsm, sizeof(lsm)) < 0) + log_err("Failed to send LSM to " ADDR_FMT32, + ADDR_VAL32(&nb->addr)); +#ifdef DEBUG_PROTO_LS + else + log_proto(LSU_FMT " --> " ADDR_FMT32, + LSU_VAL(src, dst, seqno), + ADDR_VAL32(&nb->addr)); +#endif } } @@ -601,9 +611,9 @@ static void lsdb_replicate(int fd) list_head_init(©); /* Lock the lsdb, copy the lsms and send outside of lock. */ - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - list_for_each(p, &ls.db) { + llist_for_each(p, &ls.db) { struct adjacency * adj; struct adjacency * cpy; adj = list_entry(p, struct adjacency, next); @@ -620,7 +630,7 @@ static void lsdb_replicate(int fd) list_add_tail(&cpy->next, ©); } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); list_for_each_safe(p, h, ©) { struct lsa lsm; @@ -646,18 +656,17 @@ static void * lsupdate(void * o) while (true) { clock_gettime(CLOCK_REALTIME_COARSE, &now); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - pthread_cleanup_push((void (*) (void *)) pthread_rwlock_unlock, - (void *) &ls.db_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * adj; adj = list_entry(p, struct adjacency, next); - if (now.tv_sec - adj->stamp > LS_TIMEO) { - list_del(&adj->next); - log_dbg("%" PRIu64 " - %" PRIu64" timed out.", - adj->src, adj->dst); + if (now.tv_sec > adj->stamp + ls.conf.t_timeo) { + llist_del(&adj->next, &ls.db); + log_dbg(LINK_FMT " timed out.", + LINK_VAL(adj->src, adj->dst)); if (graph_del_edge(ls.graph, adj->src, adj->dst)) log_err("Failed to del edge."); @@ -665,7 +674,7 @@ static void * lsupdate(void * o) continue; } - if (adj->src == ipcpi.dt_addr) { + if (adj->src == ls.addr) { adj->seqno++; send_lsm(adj->src, adj->dst, adj->seqno); adj->stamp = now.tv_sec; @@ -674,7 +683,7 @@ static void * lsupdate(void * o) pthread_cleanup_pop(true); - sleep(LS_UPDATE_TIME); + sleep(ls.conf.t_update); } return (void *) 0; @@ -706,30 +715,55 @@ static void forward_lsm(uint8_t * buf, int in_fd) { struct list_head * p; +#ifdef DEBUG_PROTO_LS + struct lsa lsm; + + assert(buf); + assert(len >= sizeof(struct lsa)); - pthread_rwlock_rdlock(&ls.db_lock); + memcpy(&lsm, buf, sizeof(lsm)); - pthread_cleanup_push((void (*))(void *) pthread_rwlock_unlock, - &ls.db_lock); + lsm.s_addr = ntoh64(lsm.s_addr); + lsm.d_addr = ntoh64(lsm.d_addr); + lsm.seqno = ntoh64(lsm.seqno); +#endif + pthread_rwlock_rdlock(&ls.lock); - list_for_each(p, &ls.nbs) { + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); + + llist_for_each(p, &ls.nbs) { struct nb * nb = list_entry(p, struct nb, next); - if (nb->type == NB_MGMT && nb->fd != in_fd) - flow_write(nb->fd, buf, len); + if (nb->type != NB_MGMT || nb->fd == in_fd) + continue; + + if (flow_write(nb->fd, buf, len) < 0) + log_err("Failed to forward LSM to " ADDR_FMT32, + ADDR_VAL32(&nb->addr)); +#ifdef DEBUG_PROTO_LS + else + log_proto(LSU_FMT " --> " ADDR_FMT32 " [forwarded]", + LSU_VAL(lsm.s_addr, lsm.d_addr, lsm.seqno), + ADDR_VAL32(&nb->addr)); +#endif } pthread_cleanup_pop(true); } +static void cleanup_fqueue(void * fq) +{ + fqueue_destroy((fqueue_t *) fq); +} + static void * lsreader(void * o) { - fqueue_t * fq; - int ret; - uint8_t buf[sizeof(struct lsa)]; - int fd; - qosspec_t qs; - struct lsa * msg; - size_t len; + fqueue_t * fq; + int ret; + uint8_t buf[sizeof(struct lsa)]; + int fd; + qosspec_t qs; + struct lsa msg; + size_t len; (void) o; @@ -739,8 +773,7 @@ static void * lsreader(void * o) if (fq == NULL) return (void *) -1; - pthread_cleanup_push((void (*) (void *)) fqueue_destroy, - (void *) fq); + pthread_cleanup_push(cleanup_fqueue, fq); while (true) { ret = fevent(ls.mgmt_set, fq, NULL); @@ -753,15 +786,22 @@ static void * lsreader(void * o) if (fqueue_type(fq) != FLOW_PKT) continue; - len = flow_read(fd, buf, sizeof(*msg)); - if (len <= 0 || len != sizeof(*msg)) + len = flow_read(fd, buf, sizeof(msg)); + if (len <= 0 || len != sizeof(msg)) continue; - msg = (struct lsa *) buf; - - if (lsdb_add_link(ntoh64(msg->s_addr), - ntoh64(msg->d_addr), - ntoh64(msg->seqno), + memcpy(&msg, buf, sizeof(msg)); + msg.s_addr = ntoh64(msg.s_addr); + msg.d_addr = ntoh64(msg.d_addr); + msg.seqno = ntoh64(msg.seqno); +#ifdef DEBUG_PROTO_LS + log_proto(LSU_FMT " <-- " ADDR_FMT32, + LSU_VAL(msg.s_addr, msg.d_addr, msg.seqno), + ADDR_VAL32(&ls.addr)); +#endif + if (lsdb_add_link(msg.s_addr, + msg.d_addr, + msg.seqno, &qs)) continue; @@ -782,14 +822,14 @@ static void flow_event(int fd, log_dbg("Notifying routing instances of flow event."); - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); - list_for_each(p, &ls.routing_instances) { + list_for_each(p, &ls.instances.list) { struct routing_i * ri = list_entry(p, struct routing_i, next); pff_flow_state_change(ri->pff, fd, up); } - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); } static void handle_event(void * self, @@ -811,28 +851,27 @@ static void handle_event(void * self, switch (event) { case NOTIFY_DT_CONN_ADD: - pthread_rwlock_rdlock(&ls.db_lock); + pthread_rwlock_rdlock(&ls.lock); - pthread_cleanup_push((void (*) (void *)) pthread_rwlock_unlock, - (void *) &ls.db_lock); + pthread_cleanup_push(__cleanup_rwlock_unlock, &ls.lock); - send_lsm(ipcpi.dt_addr, c->conn_info.addr, 0); + send_lsm(ls.addr, c->conn_info.addr, 0); pthread_cleanup_pop(true); if (lsdb_add_nb(c->conn_info.addr, c->flow_info.fd, NB_DT)) - log_dbg("Failed to add neighbor to LSDB."); + log_dbg("Failed to add neighbor to lsdb."); - if (lsdb_add_link(ipcpi.dt_addr, c->conn_info.addr, 0, &qs)) - log_dbg("Failed to add new adjacency to LSDB."); + if (lsdb_add_link(ls.addr, c->conn_info.addr, 0, &qs)) + log_dbg("Failed to add new adjacency to lsdb."); break; case NOTIFY_DT_CONN_DEL: flow_event(c->flow_info.fd, false); if (lsdb_del_nb(c->conn_info.addr, c->flow_info.fd)) - log_dbg("Failed to delete neighbor from LSDB."); + log_dbg("Failed to delete neighbor from lsdb."); - if (lsdb_del_link(ipcpi.dt_addr, c->conn_info.addr)) - log_dbg("Local link was not in LSDB."); + if (lsdb_del_link(ls.addr, c->conn_info.addr)) + log_dbg("Local link was not in lsdb."); break; case NOTIFY_DT_CONN_QOS: log_dbg("QoS changes currently unsupported."); @@ -848,14 +887,14 @@ static void handle_event(void * self, fccntl(c->flow_info.fd, FLOWSFLAGS, flags | FLOWFRNOPART); fset_add(ls.mgmt_set, c->flow_info.fd); if (lsdb_add_nb(c->conn_info.addr, c->flow_info.fd, NB_MGMT)) - log_warn("Failed to add mgmt neighbor to LSDB."); + log_warn("Failed to add mgmt neighbor to lsdb."); /* replicate the entire lsdb */ lsdb_replicate(c->flow_info.fd); break; case NOTIFY_MGMT_CONN_DEL: fset_del(ls.mgmt_set, c->flow_info.fd); if (lsdb_del_nb(c->conn_info.addr, c->flow_info.fd)) - log_warn("Failed to delete mgmt neighbor from LSDB."); + log_warn("Failed to delete mgmt neighbor from lsdb."); break; default: break; @@ -882,11 +921,11 @@ struct routing_i * link_state_routing_i_create(struct pff * pff) periodic_recalc_pff, tmp)) goto fail_pthread_create_lsupdate; - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); - list_add(&tmp->next, &ls.routing_instances); + list_add(&tmp->next, &ls.instances.list); - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); return tmp; @@ -902,11 +941,11 @@ void link_state_routing_i_destroy(struct routing_i * instance) { assert(instance); - pthread_mutex_lock(&ls.routing_i_lock); + pthread_mutex_lock(&ls.instances.mtx); list_del(&instance->next); - pthread_mutex_unlock(&ls.routing_i_lock); + pthread_mutex_unlock(&ls.instances.mtx); pthread_cancel(instance->calculator); @@ -917,96 +956,143 @@ void link_state_routing_i_destroy(struct routing_i * instance) free(instance); } -int link_state_init(enum pol_routing pr) +int link_state_start(void) +{ + if (notifier_reg(handle_event, NULL)) { + log_err("Failed to register link-state with notifier."); + goto fail_notifier_reg; + } + + if (pthread_create(&ls.lsupdate, NULL, lsupdate, NULL)) { + log_err("Failed to create lsupdate thread."); + goto fail_pthread_create_lsupdate; + } + + if (pthread_create(&ls.lsreader, NULL, lsreader, NULL)) { + log_err("Failed to create lsreader thread."); + goto fail_pthread_create_lsreader; + } + + if (pthread_create(&ls.listener, NULL, ls_conn_handle, NULL)) { + log_err("Failed to create listener thread."); + goto fail_pthread_create_listener; + } + + return 0; + + fail_pthread_create_listener: + pthread_cancel(ls.lsreader); + pthread_join(ls.lsreader, NULL); + fail_pthread_create_lsreader: + pthread_cancel(ls.lsupdate); + pthread_join(ls.lsupdate, NULL); + fail_pthread_create_lsupdate: + notifier_unreg(handle_event); + fail_notifier_reg: + return -1; +} + +void link_state_stop(void) +{ + pthread_cancel(ls.listener); + pthread_cancel(ls.lsreader); + pthread_cancel(ls.lsupdate); + + pthread_join(ls.listener, NULL); + pthread_join(ls.lsreader, NULL); + pthread_join(ls.lsupdate, NULL); + + notifier_unreg(handle_event); +} + + +int link_state_init(struct ls_config * conf, + enum pol_pff * pff_type) { struct conn_info info; + assert(conf != NULL); + assert(pff_type != NULL); + memset(&info, 0, sizeof(info)); + ls.addr = addr_auth_address(); + strcpy(info.comp_name, LS_COMP); strcpy(info.protocol, LS_PROTO); info.pref_version = 1; info.pref_syntax = PROTO_GPB; - info.addr = ipcpi.dt_addr; + info.addr = ls.addr; + + ls.conf = *conf; - switch (pr) { - case ROUTING_LINK_STATE: - log_dbg("Using link state routing policy."); + switch (conf->pol) { + case LS_SIMPLE: + *pff_type = PFF_SIMPLE; ls.routing_algo = ROUTING_SIMPLE; + log_dbg("Using Link State Routing policy."); break; - case ROUTING_LINK_STATE_LFA: - log_dbg("Using Loop-Free Alternates policy."); + case LS_LFA: ls.routing_algo = ROUTING_LFA; + *pff_type = PFF_ALTERNATE; + log_dbg("Using Loop-Free Alternates policy."); break; - case ROUTING_LINK_STATE_ECMP: - log_dbg("Using Equal-Cost Multipath policy."); + case LS_ECMP: ls.routing_algo = ROUTING_ECMP; + *pff_type = PFF_MULTIPATH; + log_dbg("Using Equal-Cost Multipath policy."); break; default: goto fail_graph; } + log_dbg("LS update interval: %ld seconds.", ls.conf.t_update); + log_dbg("LS link timeout : %ld seconds.", ls.conf.t_timeo); + log_dbg("LS recalc interval: %ld seconds.", ls.conf.t_recalc); + ls.graph = graph_create(); if (ls.graph == NULL) goto fail_graph; - if (notifier_reg(handle_event, NULL)) - goto fail_notifier_reg; - - if (pthread_rwlock_init(&ls.db_lock, NULL)) - goto fail_db_lock_init; + if (pthread_rwlock_init(&ls.lock, NULL)) { + log_err("Failed to init lock."); + goto fail_lock_init; + } - if (pthread_mutex_init(&ls.routing_i_lock, NULL)) + if (pthread_mutex_init(&ls.instances.mtx, NULL)) { + log_err("Failed to init instances mutex."); goto fail_routing_i_lock_init; + } - if (connmgr_comp_init(COMPID_MGMT, &info)) + if (connmgr_comp_init(COMPID_MGMT, &info)) { + log_err("Failed to init connmgr."); goto fail_connmgr_comp_init; + } ls.mgmt_set = fset_create(); - if (ls.mgmt_set == NULL) + if (ls.mgmt_set == NULL) { + log_err("Failed to create fset."); goto fail_fset_create; + } - list_head_init(&ls.db); - list_head_init(&ls.nbs); - list_head_init(&ls.routing_instances); - - if (pthread_create(&ls.lsupdate, NULL, lsupdate, NULL)) - goto fail_pthread_create_lsupdate; - - if (pthread_create(&ls.lsreader, NULL, lsreader, NULL)) - goto fail_pthread_create_lsreader; - - if (pthread_create(&ls.listener, NULL, ls_conn_handle, NULL)) - goto fail_pthread_create_listener; + llist_init(&ls.db); + llist_init(&ls.nbs); + list_head_init(&ls.instances.list); - if (rib_reg(LSDB, &r_ops)) + if (rib_reg(lsdb, &r_ops)) goto fail_rib_reg; - ls.db_len = 0; - ls.nbs_len = 0; - return 0; fail_rib_reg: - pthread_cancel(ls.listener); - pthread_join(ls.listener, NULL); - fail_pthread_create_listener: - pthread_cancel(ls.lsreader); - pthread_join(ls.lsreader, NULL); - fail_pthread_create_lsreader: - pthread_cancel(ls.lsupdate); - pthread_join(ls.lsupdate, NULL); - fail_pthread_create_lsupdate: fset_destroy(ls.mgmt_set); fail_fset_create: connmgr_comp_fini(COMPID_MGMT); fail_connmgr_comp_init: - pthread_mutex_destroy(&ls.routing_i_lock); + pthread_mutex_destroy(&ls.instances.mtx); fail_routing_i_lock_init: - pthread_rwlock_destroy(&ls.db_lock); - fail_db_lock_init: - notifier_unreg(handle_event); - fail_notifier_reg: + pthread_rwlock_destroy(&ls.lock); + fail_lock_init: graph_destroy(ls.graph); fail_graph: return -1; @@ -1017,17 +1103,7 @@ void link_state_fini(void) struct list_head * p; struct list_head * h; - rib_unreg(LSDB); - - notifier_unreg(handle_event); - - pthread_cancel(ls.listener); - pthread_cancel(ls.lsreader); - pthread_cancel(ls.lsupdate); - - pthread_join(ls.listener, NULL); - pthread_join(ls.lsreader, NULL); - pthread_join(ls.lsupdate, NULL); + rib_unreg(lsdb); fset_destroy(ls.mgmt_set); @@ -1035,17 +1111,17 @@ void link_state_fini(void) graph_destroy(ls.graph); - pthread_rwlock_wrlock(&ls.db_lock); + pthread_rwlock_wrlock(&ls.lock); - list_for_each_safe(p, h, &ls.db) { + llist_for_each_safe(p, h, &ls.db) { struct adjacency * a = list_entry(p, struct adjacency, next); - list_del(&a->next); + llist_del(&a->next, &ls.db); free(a); } - pthread_rwlock_unlock(&ls.db_lock); + pthread_rwlock_unlock(&ls.lock); - pthread_rwlock_destroy(&ls.db_lock); + pthread_rwlock_destroy(&ls.lock); - pthread_mutex_destroy(&ls.routing_i_lock); + pthread_mutex_destroy(&ls.instances.mtx); } diff --git a/src/ipcpd/unicast/pol/link_state.h b/src/ipcpd/unicast/routing/link-state.h index 9a1ada1b..38e19065 100644 --- a/src/ipcpd/unicast/pol/link_state.h +++ b/src/ipcpd/unicast/routing/link-state.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Link state routing policy * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -26,16 +26,21 @@ #define LS_COMP "Management" #define LS_PROTO "LSP" -#include "pol-routing-ops.h" +#include "ops.h" -int link_state_init(enum pol_routing pr); +int link_state_init(struct ls_config * ls, + enum pol_pff * pff_type); void link_state_fini(void); +int link_state_start(void); + +void link_state_stop(void); + struct routing_i * link_state_routing_i_create(struct pff * pff); void link_state_routing_i_destroy(struct routing_i * instance); -extern struct pol_routing_ops link_state_ops; +extern struct routing_ops link_state_ops; #endif /* OUROBOROS_IPCPD_UNICAST_POL_LINK_STATE_H */ diff --git a/src/ipcpd/unicast/pol-routing-ops.h b/src/ipcpd/unicast/routing/ops.h index 3fa1d573..b19c5176 100644 --- a/src/ipcpd/unicast/pol-routing-ops.h +++ b/src/ipcpd/unicast/routing/ops.h @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Routing policy ops * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as @@ -20,19 +20,24 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#ifndef OUROBOROS_IPCPD_UNICAST_POL_ROUTING_OPS_H -#define OUROBOROS_IPCPD_UNICAST_POL_ROUTING_OPS_H +#ifndef OUROBOROS_IPCPD_UNICAST_ROUTING_OPS_H +#define OUROBOROS_IPCPD_UNICAST_ROUTING_OPS_H #include "pff.h" -struct pol_routing_ops { - int (* init)(enum pol_routing pr); +struct routing_ops { + int (* init)(void * conf, + enum pol_pff * pff_type); void (* fini)(void); + int (* start)(void); + + void (* stop)(void); + struct routing_i * (* routing_i_create)(struct pff * pff); void (* routing_i_destroy)(struct routing_i * instance); }; -#endif /* OUROBOROS_IPCPD_UNICAST_POL_ROUTING_OPS_H */ +#endif /* OUROBOROS_IPCPD_UNICAST_ROUTING_OPS_H */ diff --git a/src/ipcpd/unicast/routing/pol.h b/src/ipcpd/unicast/routing/pol.h new file mode 100644 index 00000000..545f5df2 --- /dev/null +++ b/src/ipcpd/unicast/routing/pol.h @@ -0,0 +1,23 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Routing policies + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "link-state.h" diff --git a/src/ipcpd/unicast/routing/tests/CMakeLists.txt b/src/ipcpd/unicast/routing/tests/CMakeLists.txt new file mode 100644 index 00000000..be2de72c --- /dev/null +++ b/src/ipcpd/unicast/routing/tests/CMakeLists.txt @@ -0,0 +1,34 @@ +get_filename_component(CURRENT_SOURCE_PARENT_DIR + ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(CURRENT_BINARY_PARENT_DIR + ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) + +get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) +get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) + +compute_test_prefix() + +create_test_sourcelist(${PARENT_DIR}_tests test_suite.c + # Add new tests here + graph_test.c + ) + +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) + +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) + +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) + +add_dependencies(build_tests ${PARENT_DIR}_test) + +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/pol/tests/graph_test.c b/src/ipcpd/unicast/routing/tests/graph_test.c index ea4c0e59..40a744ff 100644 --- a/src/ipcpd/unicast/pol/tests/graph_test.c +++ b/src/ipcpd/unicast/routing/tests/graph_test.c @@ -1,10 +1,10 @@ /* - * Ouroboros - Copyright (C) 2016 - 2020 + * Ouroboros - Copyright (C) 2016 - 2026 * * Test of the graph structure * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as diff --git a/src/ipcpd/unicast/tests/CMakeLists.txt b/src/ipcpd/unicast/tests/CMakeLists.txt index 482711d5..2e35ed66 100644 --- a/src/ipcpd/unicast/tests/CMakeLists.txt +++ b/src/ipcpd/unicast/tests/CMakeLists.txt @@ -3,35 +3,32 @@ get_filename_component(CURRENT_SOURCE_PARENT_DIR get_filename_component(CURRENT_BINARY_PARENT_DIR ${CMAKE_CURRENT_BINARY_DIR} DIRECTORY) -include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -include_directories(${CMAKE_CURRENT_BINARY_DIR}) - -include_directories(${CURRENT_SOURCE_PARENT_DIR}) -include_directories(${CURRENT_BINARY_PARENT_DIR}) - -include_directories(${CMAKE_SOURCE_DIR}/include) -include_directories(${CMAKE_BINARY_DIR}/include) - get_filename_component(PARENT_PATH ${CMAKE_CURRENT_SOURCE_DIR} DIRECTORY) get_filename_component(PARENT_DIR ${PARENT_PATH} NAME) +compute_test_prefix() + create_test_sourcelist(${PARENT_DIR}_tests test_suite.c # Add new tests here - dht_test.c + cap_test.c ) -protobuf_generate_c(KAD_PROTO_SRCS KAD_PROTO_HDRS ../kademlia.proto) +add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) -add_executable(${PARENT_DIR}_test EXCLUDE_FROM_ALL ${${PARENT_DIR}_tests} - ${KAD_PROTO_SRCS}) -target_link_libraries(${PARENT_DIR}_test ouroboros-common) +target_include_directories(${PARENT_DIR}_test PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR} + ${CURRENT_SOURCE_PARENT_DIR} + ${CURRENT_BINARY_PARENT_DIR} + ${CMAKE_SOURCE_DIR}/include + ${CMAKE_BINARY_DIR}/include + ${CMAKE_SOURCE_DIR}/src/ipcpd + ${CMAKE_BINARY_DIR}/src/ipcpd +) -add_dependencies(check ${PARENT_DIR}_test) +disable_test_logging_for_target(${PARENT_DIR}_test) +target_link_libraries(${PARENT_DIR}_test PRIVATE ouroboros-common) -set(tests_to_run ${${PARENT_DIR}_tests}) -remove(tests_to_run test_suite.c) +add_dependencies(build_tests ${PARENT_DIR}_test) -foreach (test ${tests_to_run}) - get_filename_component(test_name ${test} NAME_WE) - add_test(${test_name} ${C_TEST_PATH}/${PARENT_DIR}_test ${test_name}) -endforeach (test) +ouroboros_register_tests(TARGET ${PARENT_DIR}_test TESTS ${${PARENT_DIR}_tests}) diff --git a/src/ipcpd/unicast/tests/cap_test.c b/src/ipcpd/unicast/tests/cap_test.c new file mode 100644 index 00000000..7867b490 --- /dev/null +++ b/src/ipcpd/unicast/tests/cap_test.c @@ -0,0 +1,593 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Unit tests for link capacity estimation + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., http://www.fsf.org/about/contact/. + */ + +#include "cap.c" + +#include <test/test.h> + +#define TICK (50 * 1000ULL) /* 50 us between packets */ +#define LEN 1000ULL /* default packet size (B) */ +#define QLEN 8 /* steady ring backlog */ +#define RATE (LEN * BILLION / TICK) /* LEN per TICK = 20 MB/s */ + +#define SHP_LEN 1250ULL /* shaped-link packet (B) */ +#define SHP_STEP 20 /* packets per shaped window */ +#define SHP_RATE (SHP_LEN * BILLION / (SHP_STEP * TICK)) + +static int test_cap_init_fini(void) +{ + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + if (cap_get(0) != 0 || cap_get(PROC_MAX_FLOWS - 1) != 0) { + printf("Fresh estimator not unknown.\n"); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Exact roundtrip holds for codes >= 32 (rates >= 256 B/s). */ +static int test_cap_codec_roundtrip(void) +{ + unsigned c; + + TEST_START(); + + for (c = 32; c <= 255; c++) { + if (cap_enc(cap_dec((uint8_t) c)) != c) { + printf("Code %u does not roundtrip.\n", c); + goto fail; + } + + if (cap_dec((uint8_t) c) <= cap_dec((uint8_t) (c - 1))) { + printf("Decode not monotone at %u.\n", c); + goto fail; + } + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_codec_bounds(void) +{ + TEST_START(); + + if (cap_enc(0) != 0 || cap_dec(0) != 0) { + printf("Zero is not unknown.\n"); + goto fail; + } + + if (cap_enc(1) != 1) { + printf("Rate 1 encoded as %u.\n", cap_enc(1)); + goto fail; + } + + if (cap_enc(UINT64_MAX) != 255) { + printf("Max rate encoded as %u.\n", cap_enc(UINT64_MAX)); + goto fail; + } + + if (cap_dec(255) <= cap_dec(254)) { + printf("Top code does not decode.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_min(void) +{ + TEST_START(); + + if (cap_min(0, 42) != 42 || cap_min(42, 0) != 42) { + printf("Unknown not skipped in min.\n"); + goto fail; + } + + if (cap_min(0, 0) != 0) { + printf("Two unknowns not unknown.\n"); + goto fail; + } + + if (cap_min(97, 42) != 42 || cap_min(42, 97) != 42) { + printf("Min not taken.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_stamp(void) +{ + uint8_t pci; + + TEST_START(); + + pci = 42; + cap_stamp(&pci, 0); + if (pci != 42) { + printf("Unknown own code overwrote the byte.\n"); + goto fail; + } + + pci = 0; + cap_stamp(&pci, 97); + if (pci != 97) { + printf("Own code not written into unknown.\n"); + goto fail; + } + + pci = 97; + cap_stamp(&pci, 42); + if (pci != 42) { + printf("Lower own code did not lower the byte.\n"); + goto fail; + } + + pci = 42; + cap_stamp(&pci, 97); + if (pci != 42) { + printf("Higher own code raised the byte.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_busy_window(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + /* 1000 B every 50 us, ring steady at 8: drain = 20 MB/s. */ + for (i = 1; i <= 40; i++) + cap_update_at(0, QLEN, LEN, i * TICK); + + if (cap_get(0) != cap_enc(RATE)) { + printf("Estimated code: exp %u, got %u.\n", + cap_enc(RATE), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_idle_tolerated(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + for (i = 1; i <= 40; i++) + cap_update_at(0, i == 21 ? 0 : QLEN, LEN, i * TICK); + + if (cap_get(0) != cap_enc(RATE)) { + printf("Grazed window: exp %u, got %u.\n", + cap_enc(RATE), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_mostly_idle_rejects(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + for (i = 1; i <= 100; i++) + cap_update_at(0, 0, LEN, i * TICK); + + if (cap_get(0) != 0) { + printf("Idle ring estimated %u.\n", cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_slow_link_extends(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + /* 1000 B every 100 us: 10 slots/ms closes on a 2 ms window. */ + for (i = 1; i <= 30; i++) + cap_update_at(0, QLEN, LEN, i * 2 * TICK); + + if (cap_get(0) != cap_enc(RATE / 2)) { + printf("Slow link: exp %u, got %u.\n", + cap_enc(RATE / 2), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_shaped_link(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + /* 1250 B every ms; one empty observation per 20 packets. */ + for (i = 1; i <= 100; i++) + cap_update_at(0, i % SHP_STEP == 0 ? 0 : 6, SHP_LEN, + i * SHP_STEP * TICK); + + if (cap_get(0) != cap_enc(SHP_RATE)) { + printf("Shaped link: exp %u, got %u.\n", + cap_enc(SHP_RATE), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_stale_discard(void) +{ + uint64_t t; + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + /* Open a window, trickle 4 slots, then ~200 ms of silence. */ + for (i = 1; i <= 5; i++) + cap_update_at(0, QLEN, LEN, i * CAP_T_MIN); + + t = 205 * CAP_T_MIN; + + cap_update_at(0, QLEN, LEN, t); + + if (cap_get(0) != 0) { + printf("Gap window estimated %u.\n", cap_get(0)); + goto fail_init; + } + + for (i = 1; i <= 40; i++) + cap_update_at(0, QLEN, LEN, t + i * TICK); + + if (cap_get(0) != cap_enc(RATE)) { + printf("Post-gap: exp %u, got %u.\n", + cap_enc(RATE), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_est_empty_start_no_raise(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + cap_update_at(0, 0, LEN, CAP_T_MIN); + + for (i = 1; i <= 40; i++) + cap_update_at(0, QLEN, LEN, CAP_T_MIN + i * TICK); + + if (cap_get(0) != 0) { + printf("Empty-start window raised to %u.\n", + cap_get(0)); + goto fail_init; + } + + for (i = 41; i <= 60; i++) + cap_update_at(0, QLEN, LEN, CAP_T_MIN + i * TICK); + + if (cap_get(0) != cap_enc(RATE)) { + printf("Backlogged window: exp %u, got %u.\n", + cap_enc(RATE), cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* Max filter: fast attack on a high sample, slow release on lower. */ +static int test_cap_est_max_filter(void) +{ + uint8_t high; + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + for (i = 1; i <= 40; i++) + cap_update_at(0, QLEN, LEN, i * TICK); + + high = cap_get(0); + if (high != cap_enc(RATE)) { + printf("Attack missed: exp %u, got %u.\n", cap_enc(RATE), + high); + goto fail_init; + } + + /* Halved packet size: valid samples at 10 MB/s. */ + for (i = 41; i <= 80; i++) + cap_update_at(0, QLEN, LEN / 2, i * TICK); + + if (cap_get(0) >= high) { + printf("Release did not decay: %u.\n", cap_get(0)); + goto fail_init; + } + + if (cap_get(0) <= cap_enc(RATE / 2)) { + printf("Release collapsed to %u.\n", cap_get(0)); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +/* No fold within CAP_T_MIN of the previous one. */ +static int test_cap_est_gate(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + cap_update_at(0, QLEN, LEN, CAP_T_MIN); + + for (i = 0; i < 5; i++) + cap_update_at(0, QLEN, LEN, CAP_T_MIN + CAP_T_MIN / 2); + + if (cap.est[0].t_gate != CAP_T_MIN) { + printf("Fold ran inside the gate.\n"); + goto fail_init; + } + + if (LOAD_RELAXED(&cap.est[0].c_pkt) != 6) { + printf("Gated packets not counted.\n"); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_cap_reset(void) +{ + size_t i; + + TEST_START(); + + if (cap_init() < 0) { + printf("Failed to init cap.\n"); + goto fail; + } + + for (i = 1; i <= 40; i++) + cap_update_at(0, QLEN, LEN, i * TICK); + + if (cap_get(0) == 0) { + printf("No estimate to reset.\n"); + goto fail_init; + } + + cap_reset(0); + + if (cap_get(0) != 0) { + printf("Reset did not clear the estimate.\n"); + goto fail_init; + } + + cap_fini(); + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail_init: + cap_fini(); + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +int cap_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_cap_init_fini(); + ret |= test_cap_codec_roundtrip(); + ret |= test_cap_codec_bounds(); + ret |= test_cap_min(); + ret |= test_cap_stamp(); + ret |= test_cap_est_busy_window(); + ret |= test_cap_est_idle_tolerated(); + ret |= test_cap_est_mostly_idle_rejects(); + ret |= test_cap_est_slow_link_extends(); + ret |= test_cap_est_shaped_link(); + ret |= test_cap_est_stale_discard(); + ret |= test_cap_est_empty_start_no_raise(); + ret |= test_cap_est_max_filter(); + ret |= test_cap_est_gate(); + ret |= test_cap_reset(); + + return ret; +} diff --git a/src/ipcpd/unicast/tests/dht_test.c b/src/ipcpd/unicast/tests/dht_test.c deleted file mode 100644 index 21ecd564..00000000 --- a/src/ipcpd/unicast/tests/dht_test.c +++ /dev/null @@ -1,99 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2020 - * - * Unit tests of the DHT - * - * Dimitri Staessens <dimitri.staessens@ugent.be> - * Sander Vrijders <sander.vrijders@ugent.be> - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., http://www.fsf.org/about/contact/. - */ - -#define __DHT_TEST__ - -#include "dht.c" - -#include <pthread.h> -#include <time.h> -#include <stdlib.h> -#include <stdio.h> - -#define KEY_LEN 32 - -#define EXP 86400 -#define CONTACTS 1000 - -int dht_test(int argc, - char ** argv) -{ - struct dht * dht; - uint64_t addr = 0x0D1F; - uint8_t key[KEY_LEN]; - size_t i; - - (void) argc; - (void) argv; - - dht = dht_create(addr); - if (dht == NULL) { - printf("Failed to create dht.\n"); - return -1; - } - - dht_destroy(dht); - - dht = dht_create(addr); - if (dht == NULL) { - printf("Failed to re-create dht.\n"); - return -1; - } - - if (dht_bootstrap(dht, KEY_LEN, EXP)) { - printf("Failed to bootstrap dht.\n"); - dht_destroy(dht); - return -1; - } - - dht_destroy(dht); - - dht = dht_create(addr); - if (dht == NULL) { - printf("Failed to re-create dht.\n"); - return -1; - } - - if (dht_bootstrap(dht, KEY_LEN, EXP)) { - printf("Failed to bootstrap dht.\n"); - dht_destroy(dht); - return -1; - } - - for (i = 0; i < CONTACTS; ++i) { - uint64_t addr; - random_buffer(&addr, sizeof(addr)); - random_buffer(key, KEY_LEN); - pthread_rwlock_wrlock(&dht->lock); - if (dht_update_bucket(dht, key, addr)) { - pthread_rwlock_unlock(&dht->lock); - printf("Failed to update bucket.\n"); - dht_destroy(dht); - return -1; - } - pthread_rwlock_unlock(&dht->lock); - } - - dht_destroy(dht); - - return 0; -} |