diff options
91 files changed, 10267 insertions, 5989 deletions
diff --git a/CMakeLists.txt b/CMakeLists.txt index bfabd711..e624c22c 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -51,15 +51,16 @@ include(config/global) include(dependencies) -include(config/lib) -include(config/ssm) +include(config/lib/common) +include(config/lib/crypt) +include(config/lib/poa) +include(config/lib/frct) +include(config/lib/ssm) include(config/irmd) include(config/ipcp/common) include(config/ipcp/unicast) include(config/ipcp/broadcast) include(config/ipcp/local) -include(config/ipcp/eth) -include(config/ipcp/udp) include(tests) include(include) diff --git a/cmake/config/ipcp/common.cmake b/cmake/config/ipcp/common.cmake index 7dbc252b..79c822a3 100644 --- a/cmake/config/ipcp/common.cmake +++ b/cmake/config/ipcp/common.cmake @@ -49,10 +49,4 @@ if(HAVE_FUSE) if(IPCP_FLOW_STATS) message(STATUS "IPCP flow statistics enabled") endif() - - set(IPCP_ETH_FLOW_STATS FALSE CACHE BOOL - "Enable ipcpd-eth flow statistics via RIB") - if(IPCP_ETH_FLOW_STATS) - message(STATUS "ipcpd-eth flow statistics enabled") - endif() endif() diff --git a/cmake/config/ipcp/eth.cmake b/cmake/config/ipcp/eth.cmake deleted file mode 100644 index 6a044d4b..00000000 --- a/cmake/config/ipcp/eth.cmake +++ /dev/null @@ -1,22 +0,0 @@ -# Ethernet IPCP configuration options for Ouroboros -# Options for eth-llc and eth-dix IPCPs - -set(IPCP_ETH_LLC_TARGET ipcpd-eth-llc) -set(IPCP_ETH_DIX_TARGET ipcpd-eth-dix) - -set(IPCP_ETH_RD_THR 1 CACHE STRING - "Number of reader threads in Ethernet IPCP") -set(IPCP_ETH_WR_THR 1 CACHE STRING - "Number of writer threads in Ethernet IPCP") -set(IPCP_ETH_QDISC_BYPASS false CACHE BOOL - "Bypass the Qdisc in the kernel when using raw sockets") -set(IPCP_ETH_SNDBUF 4096 CACHE STRING - "Raw socket SO_SNDBUF in bytes (floored to one frame); 0 = kernel default") -set(IPCP_ETH_RCVBUF 0 CACHE STRING - "Raw socket SO_RCVBUF in bytes; 0 = leave kernel default (rmem_default)") -set(IPCP_ETH_LO_MTU 9000 CACHE STRING - "Restrict Ethernet MTU over loopback interfaces") -set(IPCP_ETH_MGMT_FRAME_SIZE 9000 CACHE STRING - "Management frame buffer size for Ethernet IPCPs") -set(IPCP_ETH_MPL 100 CACHE STRING - "Default maximum packet lifetime for the Ethernet IPCPs, in ms") diff --git a/cmake/config/ipcp/udp.cmake b/cmake/config/ipcp/udp.cmake deleted file mode 100644 index af84a844..00000000 --- a/cmake/config/ipcp/udp.cmake +++ /dev/null @@ -1,16 +0,0 @@ -# UDP IPCP configuration options for Ouroboros -# Options for udp4 and udp6 IPCPs - -set(IPCP_UDP4_TARGET ipcpd-udp4) -set(IPCP_UDP6_TARGET ipcpd-udp6) - -set(IPCP_UDP_RD_THR 3 CACHE STRING - "Number of reader threads in UDP IPCPs") -set(IPCP_UDP_WR_THR 3 CACHE STRING - "Number of writer threads in UDP IPCPs") -set(IPCP_UDP_MPL 5000 CACHE STRING - "Default maximum packet lifetime for the UDP IPCPs, in ms") -set(IPCP_UDP4_MTU 1472 CACHE STRING - "Fallback UDP4 layer MTU when getsockopt(IP_MTU) is unavailable, in bytes") -set(IPCP_UDP6_MTU 1452 CACHE STRING - "Fallback UDP6 layer MTU when getsockopt(IPV6_MTU) is unavailable, in bytes") diff --git a/cmake/config/irmd.cmake b/cmake/config/irmd.cmake index 79e24bae..bbb73bec 100644 --- a/cmake/config/irmd.cmake +++ b/cmake/config/irmd.cmake @@ -11,7 +11,7 @@ set(ENROLL_TIMEOUT 20000 CACHE STRING set(REG_TIMEOUT 20000 CACHE STRING "Timeout for registering a name (ms)") set(QUERY_TIMEOUT 2000 CACHE STRING - "Timeout to query a name with an IPCP (ms); must exceed shim retry budget") + "Timeout to query a name with an IPCP (ms); must exceed PoA retries") set(CONNECT_TIMEOUT 20000 CACHE STRING "Timeout to connect an IPCP to another IPCP (ms)") set(FLOW_ALLOC_TIMEOUT 20000 CACHE STRING diff --git a/cmake/config/lib.cmake b/cmake/config/lib.cmake deleted file mode 100644 index 81a7d6ba..00000000 --- a/cmake/config/lib.cmake +++ /dev/null @@ -1,139 +0,0 @@ -# Library configuration options for Ouroboros -# Options affecting libouroboros-common, libouroboros-dev, libouroboros-irm - -# Flow limits -set(SYS_MAX_FLOWS 10240 CACHE STRING - "Maximum number of total flows for this system") -set(PROC_MAX_FLOWS 4096 CACHE STRING - "Maximum number of flows in an application") -set(PROC_RES_FDS 64 CACHE STRING - "Number of reserved flow descriptors per application") -set(PROC_MAX_FQUEUES 32 CACHE STRING - "Maximum number of flow sets per application") - -# Threading -if(NOT APPLE) - set(PTHREAD_COND_CLOCK "CLOCK_MONOTONIC" CACHE STRING - "Clock to use for condition variable timing") -else() - set(PTHREAD_COND_CLOCK "CLOCK_REALTIME" CACHE INTERNAL - "Clock to use for condition variable timing") -endif() - -# Timeouts -set(SOCKET_TIMEOUT 500 CACHE STRING - "Default timeout for responses from IPCPs (ms)") - -# QoS settings -set(QOS_DISABLE_CRC TRUE CACHE BOOL - "Ignores ber setting on all QoS cubes") - -include(utils/CPUUtils) -detect_pclmul() -detect_pmull() -if(HAVE_PCLMUL) - message(STATUS "CRC-64/NVMe backend: PCLMUL (x86 SSE4.1+PCLMUL)") -elseif(HAVE_PMULL) - message(STATUS "CRC-64/NVMe backend: PMULL (aarch64 crypto)") -else() - message(STATUS "CRC-64/NVMe backend: byte table (no acceleration)") -endif() - -# Delta-t protocol timers (Watson bound: 3*MPL + A + R). -# MPL is reported per IPCP (IPCP_*_MPL); A and R are FRCT-wide. -set(DELTA_T_ACK 1000 CACHE STRING - "Maximum time to acknowledge a packet (ms)") -set(DELTA_T_RTX 30000 CACHE STRING - "Maximum time to retransmit a packet (ms)") - -# FRCT configuration -set(FRCT_REORDER_QUEUE_SIZE 128 CACHE STRING - "Size of the reordering queue, must be a power of 2") -set(FRCT_START_WINDOW 128 CACHE STRING - "Start window, must be a power of 2") -set(FRCT_LINUX_RTT_ESTIMATOR TRUE CACHE BOOL - "Use Linux RTT estimator formula instead of the TCP RFC formula") -set(FRCT_RTO_MDEV_MULTIPLIER 2 CACHE STRING - "Multiplier for deviation term in the RTO: RTO = sRTT + (mdev << X)") -set(FRCT_RTO_INC_FACTOR 0 CACHE STRING - "Divisor for RTO increase after timeout: RTO += RTX >> X, 0: Karn/Partridge") -set(FRCT_RTO_MIN 250 CACHE STRING - "Hard floor for Retransmission Timeout (RTO) for FRCT (us)") -set(FRCT_TICK_TIME 5000 CACHE STRING - "Tick time for FRCT activity (retransmission, acknowledgments) (us)") -set(FRCT_DEBUG_STDOUT FALSE CACHE BOOL - "Print FRCT final counters to stdout at flow teardown") - -# Retransmission (RXM) configuration -set(RXM_MIN_RESOLUTION 20 CACHE STRING - "Minimum retransmission delay (ns), as a power to 2") -set(RXM_WHEEL_MULTIPLIER 4 CACHE STRING - "Factor for retransmission wheel levels as a power to 2") -set(RXM_WHEEL_LEVELS 3 CACHE STRING - "Number of levels in the retransmission wheel") -set(RXM_WHEEL_SLOTS_PER_LEVEL 256 CACHE STRING - "Number of slots per level in the retransmission wheel, must be a power of 2") - -# Acknowledgment wheel configuration -set(ACK_WHEEL_SLOTS 256 CACHE STRING - "Number of slots in the acknowledgment wheel, must be a power of 2") -set(ACK_WHEEL_RESOLUTION 18 CACHE STRING - "Minimum acknowledgment delay (ns), as a power to 2") - -# Thread pool manager (TPM) debugging -set(TPM_DEBUG_REPORT_INTERVAL 0 CACHE STRING - "Interval at wich the TPM will report long running threads (s), 0 disables") -set(TPM_DEBUG_ABORT_TIMEOUT 0 CACHE STRING - "TPM abort process after a thread reaches this timeout (s), 0 disables") - -# Encryption -set(KEY_LEAF_BITS 20 CACHE STRING - "Packets per leaf key as a power of two (2^20 = AEAD-safe default)") -set(KEY_NODE_BITS 6 CACHE STRING - "Leaf keys per node key, power of two (2^6 = 64; leak compartment)") -set(KEY_NODE_COUNT 128 CACHE STRING - "Node keys per batch (N); <= 4096, the 12-bit on-wire node index") -set(KEY_REKEY_WATERMARK 4 CACHE STRING - "Re-key when this many node keys remain; 0 disables the count trigger") -set(KEY_REPLAY_WINDOW 2048 CACHE STRING - "RX replay window in packets; power of two, >= 128") -set(KEY_REKEY_WM_CHECK_BITS 16 CACHE STRING - "Re-key watermark is consulted once per 2^n flow writes") -if(NOT KEY_REPLAY_WINDOW MATCHES "^[0-9]+$") - message(FATAL_ERROR "KEY_REPLAY_WINDOW must be a positive integer") -endif() -math(EXPR _krw_p2 "${KEY_REPLAY_WINDOW} & (${KEY_REPLAY_WINDOW} - 1)") -if(KEY_REPLAY_WINDOW LESS 128 OR NOT _krw_p2 EQUAL 0) - message(FATAL_ERROR "KEY_REPLAY_WINDOW must be a power of two >= 128") -endif() - -# Re-key must finish within its lead window - KEY_REKEY_WATERMARK node keys -# worth of packets - before the batch exhausts and TX fails closed. dev.c only -# evaluates the watermark once per FLOW_WM_CHECK writes, so a lead below ~2x -# that leaves a high-rate flow no room to complete the exchange. Production -# defaults are vast; this guards under-sized (test) geometries. -if(KEY_REKEY_WATERMARK GREATER 0) - math(EXPR _rk_wm_check "1 << ${KEY_REKEY_WM_CHECK_BITS}") - math(EXPR _rk_lead - "${KEY_REKEY_WATERMARK} << (${KEY_LEAF_BITS} + ${KEY_NODE_BITS})") - math(EXPR _rk_min "2 * ${_rk_wm_check}") - if(_rk_lead LESS _rk_min) - message(WARNING - "Re-key lead is ${_rk_lead} packets vs the watermark check interval " - "${_rk_wm_check}; a high-rate flow may exhaust its key batch before the " - "re-key completes (TX fails closed until it does). Raise KEY_LEAF_BITS, " - "KEY_NODE_BITS, or KEY_REKEY_WATERMARK.") - endif() -endif() - -# Flow statistics (requires FUSE) -if(HAVE_FUSE) - set(PROC_FLOW_STATS TRUE CACHE BOOL - "Enable flow statistics tracking for application flows") - if(PROC_FLOW_STATS) - message(STATUS "Application flow statistics enabled") - else() - message(STATUS "Application flow statistics disabled") - endif() -endif() - diff --git a/cmake/config/lib/common.cmake b/cmake/config/lib/common.cmake new file mode 100644 index 00000000..ebdc3196 --- /dev/null +++ b/cmake/config/lib/common.cmake @@ -0,0 +1,58 @@ +# Library configuration options for Ouroboros Options affecting libouroboros-common, +# libouroboros-dev and libouroboros-irm as a whole; per-subsystem options sit beside this file + +# Flow limits +set(SYS_MAX_FLOWS 10240 CACHE STRING + "Maximum number of total flows for this system") +set(PROC_MAX_FLOWS 4096 CACHE STRING + "Maximum number of flows in an application") +set(PROC_RES_FDS 64 CACHE STRING + "Number of reserved flow descriptors per application") +set(PROC_MAX_FQUEUES 32 CACHE STRING + "Maximum number of flow sets per application") + +# Threading +if(NOT APPLE) + set(PTHREAD_COND_CLOCK "CLOCK_MONOTONIC" CACHE STRING + "Clock to use for condition variable timing") +else() + set(PTHREAD_COND_CLOCK "CLOCK_REALTIME" CACHE INTERNAL + "Clock to use for condition variable timing") +endif() + +# Timeouts +set(SOCKET_TIMEOUT 500 CACHE STRING + "Default timeout for responses from IPCPs (ms)") + +# QoS settings +set(QOS_DISABLE_CRC TRUE CACHE BOOL + "Ignores ber setting on all QoS cubes") + +include(utils/CPUUtils) +detect_pclmul() +detect_pmull() +if(HAVE_PCLMUL) + message(STATUS "CRC-64/NVMe backend: PCLMUL (x86 SSE4.1+PCLMUL)") +elseif(HAVE_PMULL) + message(STATUS "CRC-64/NVMe backend: PMULL (aarch64 crypto)") +else() + message(STATUS "CRC-64/NVMe backend: byte table (no acceleration)") +endif() + +# Thread pool manager (TPM) debugging +set(TPM_DEBUG_REPORT_INTERVAL 0 CACHE STRING + "Interval at wich the TPM will report long running threads (s), 0 disables") +set(TPM_DEBUG_ABORT_TIMEOUT 0 CACHE STRING + "TPM abort process after a thread reaches this timeout (s), 0 disables") + +# Flow statistics (requires FUSE) +if(HAVE_FUSE) + set(PROC_FLOW_STATS TRUE CACHE BOOL + "Enable flow statistics tracking for application flows") + if(PROC_FLOW_STATS) + message(STATUS "Application flow statistics enabled") + else() + message(STATUS "Application flow statistics disabled") + endif() +endif() + diff --git a/cmake/config/lib/crypt.cmake b/cmake/config/lib/crypt.cmake new file mode 100644 index 00000000..45fdaaed --- /dev/null +++ b/cmake/config/lib/crypt.cmake @@ -0,0 +1,41 @@ +# Encryption configuration for Ouroboros Options for the key schedule and packet protection in +# src/lib/crypt/ + +# Encryption +set(KEY_LEAF_BITS 20 CACHE STRING + "Packets per leaf key as a power of two (2^20 = AEAD-safe default)") +set(KEY_NODE_BITS 6 CACHE STRING + "Leaf keys per node key, power of two (2^6 = 64; leak compartment)") +set(KEY_NODE_COUNT 128 CACHE STRING + "Node keys per batch (N); <= 4096, the 12-bit on-wire node index") +set(KEY_REKEY_WATERMARK 4 CACHE STRING + "Re-key when this many node keys remain; 0 disables the count trigger") +set(KEY_REPLAY_WINDOW 2048 CACHE STRING + "RX replay window in packets; power of two, >= 128") +set(KEY_REKEY_WM_CHECK_BITS 16 CACHE STRING + "Re-key watermark is consulted once per 2^n flow writes") +if(NOT KEY_REPLAY_WINDOW MATCHES "^[0-9]+$") + message(FATAL_ERROR "KEY_REPLAY_WINDOW must be a positive integer") +endif() +math(EXPR _krw_p2 "${KEY_REPLAY_WINDOW} & (${KEY_REPLAY_WINDOW} - 1)") +if(KEY_REPLAY_WINDOW LESS 128 OR NOT _krw_p2 EQUAL 0) + message(FATAL_ERROR "KEY_REPLAY_WINDOW must be a power of two >= 128") +endif() + +# Re-key must finish within its lead window - KEY_REKEY_WATERMARK node keys worth of packets - +# before the batch exhausts and TX fails closed. dev.c only evaluates the watermark once per +# FLOW_WM_CHECK writes, so a lead below ~2x that leaves a high-rate flow no room to complete the +# exchange. Production defaults are vast; this guards under-sized (test) geometries. +if(KEY_REKEY_WATERMARK GREATER 0) + math(EXPR _rk_wm_check "1 << ${KEY_REKEY_WM_CHECK_BITS}") + math(EXPR _rk_lead + "${KEY_REKEY_WATERMARK} << (${KEY_LEAF_BITS} + ${KEY_NODE_BITS})") + math(EXPR _rk_min "2 * ${_rk_wm_check}") + if(_rk_lead LESS _rk_min) + message(WARNING + "Re-key lead is ${_rk_lead} packets vs the watermark check interval " + "${_rk_wm_check}; a high-rate flow may exhaust its key batch before the " + "re-key completes (TX fails closed until it does). Raise KEY_LEAF_BITS, " + "KEY_NODE_BITS, or KEY_REKEY_WATERMARK.") + endif() +endif() diff --git a/cmake/config/lib/frct.cmake b/cmake/config/lib/frct.cmake new file mode 100644 index 00000000..1bd60aa0 --- /dev/null +++ b/cmake/config/lib/frct.cmake @@ -0,0 +1,43 @@ +# FRCT configuration for Ouroboros Options for the flow and retransmission control protocol in +# src/lib/frct.c + +# Delta-t protocol timers (Watson bound: 3*MPL + A + R). +# MPL is reported per IPCP (IPCP_*_MPL); A and R are FRCT-wide. +set(DELTA_T_ACK 1000 CACHE STRING + "Maximum time to acknowledge a packet (ms)") +set(DELTA_T_RTX 32000 CACHE STRING + "Maximum time to retransmit a packet (ms)") + +# FRCT configuration +set(FRCT_REORDER_QUEUE_SIZE 128 CACHE STRING + "Size of the reordering queue, must be a power of 2") +set(FRCT_START_WINDOW 128 CACHE STRING + "Start window, must be a power of 2") +set(FRCT_LINUX_RTT_ESTIMATOR TRUE CACHE BOOL + "Use Linux RTT estimator formula instead of the TCP RFC formula") +set(FRCT_RTO_MDEV_MULTIPLIER 2 CACHE STRING + "Multiplier for deviation term in the RTO: RTO = sRTT + (mdev << X)") +set(FRCT_RTO_INC_FACTOR 0 CACHE STRING + "Divisor for RTO increase after timeout: RTO += RTX >> X, 0: Karn/Partridge") +set(FRCT_RTO_MIN 250 CACHE STRING + "Hard floor for Retransmission Timeout (RTO) for FRCT (us)") +set(FRCT_TICK_TIME 5000 CACHE STRING + "Tick time for FRCT activity (retransmission, acknowledgments) (us)") +set(FRCT_DEBUG_STDOUT FALSE CACHE BOOL + "Print FRCT final counters to stdout at flow teardown") + +# Retransmission (RXM) configuration +set(RXM_MIN_RESOLUTION 20 CACHE STRING + "Minimum retransmission delay (ns), as a power to 2") +set(RXM_WHEEL_MULTIPLIER 4 CACHE STRING + "Factor for retransmission wheel levels as a power to 2") +set(RXM_WHEEL_LEVELS 3 CACHE STRING + "Number of levels in the retransmission wheel") +set(RXM_WHEEL_SLOTS_PER_LEVEL 256 CACHE STRING + "Number of slots per level in the retransmission wheel, must be a power of 2") + +# Acknowledgment wheel configuration +set(ACK_WHEEL_SLOTS 256 CACHE STRING + "Number of slots in the acknowledgment wheel, must be a power of 2") +set(ACK_WHEEL_RESOLUTION 18 CACHE STRING + "Minimum acknowledgment delay (ns), as a power to 2") diff --git a/cmake/config/lib/poa.cmake b/cmake/config/lib/poa.cmake new file mode 100644 index 00000000..c1ca10a8 --- /dev/null +++ b/cmake/config/lib/poa.cmake @@ -0,0 +1,37 @@ +# Flow Point of attachment (poa) configuration for Ouroboros + +set(POA_MGMT_FRAME_SIZE 16384 CACHE STRING + "Maximum size of a flow endpoint management frame, in bytes") + +# Management frames are retried or repeated, so a send never waits long. +set(POA_MGMT_SND_TIMEO 100 CACHE STRING + "Deadline for sending a management frame, in ms") + +# Flows on one PoA share its transmit queue, so congestion avoidance keys its link estimator on the +# PoA and needs a bound on their number. +set(POA_MAX_POAS 16 CACHE STRING + "Maximum number of points of attachment per IPCP") + +# UDP endpoints +set(POA_UDP_MPL 5000 CACHE STRING + "Default maximum packet lifetime for UDP flow endpoints, in ms") +set(POA_UDP4_MTU 1472 CACHE STRING + "Fallback UDP4 endpoint MTU when getsockopt(IP_MTU) is unavailable") +set(POA_UDP6_MTU 1452 CACHE STRING + "Fallback UDP6 endpoint MTU when getsockopt(IPV6_MTU) is unavailable") +set(POA_UDP_RD_BUF 65535 CACHE STRING + "UDP endpoint receive buffer in bytes. Bounds the advertised MTU") + +# Ethernet endpoints +set(POA_ETH_MPL 100 CACHE STRING + "Default maximum packet lifetime for Ethernet flow endpoints, in ms") +set(POA_ETH_QDISC_BYPASS false CACHE BOOL + "Bypass the Qdisc in the kernel when using raw sockets") +set(POA_ETH_SNDBUF 0 CACHE STRING + "Raw socket SO_SNDBUF in bytes (floored to one frame). 0 = kernel default") +set(POA_ETH_RCVBUF 0 CACHE STRING + "Raw socket SO_RCVBUF in bytes. 0 = kernel default") +set(POA_ETH_LO_MTU 9000 CACHE STRING + "Restrict Ethernet flow endpoint MTU over loopback interfaces") +set(POA_ETH_RD_BUF 16384 CACHE STRING + "Cap on the Ethernet endpoint receive buffer and MTU, in bytes") diff --git a/cmake/config/ssm.cmake b/cmake/config/lib/ssm.cmake index 589171ea..a9ac35c9 100644 --- a/cmake/config/ssm.cmake +++ b/cmake/config/lib/ssm.cmake @@ -1,6 +1,5 @@ -# Secure Shared Memory (SSM) pool configuration for Ouroboros -# This file defines the allocation parameters for the secure shared memory -# pool allocator +# Secure Shared Memory (SSM) pool configuration for Ouroboros This file defines the allocation +# parameters for the secure shared memory pool allocator # Shared memory pool naming configuration set(SSM_PREFIX "ouroboros" CACHE STRING @@ -21,6 +20,8 @@ set(SSM_PK_BUFF_TAILSPACE 32 CACHE STRING "Bytes of tailspace to reserve for future tails") set(SSM_RBUFF_SIZE 1024 CACHE STRING "Number of slots in a flow's rbuff ring; must be a power of 2") +set(SSM_RBUFF_TXQ_DELAY 10 CACHE STRING + "Queueing delay a flow's tx ring may hold (ms); 0 is unlimited") set(SSM_RBUFF_PREFIX "/${SHM_PREFIX}.rbuff." CACHE INTERNAL "Prefix for rbuff POSIX shared memory filenames") set(SSM_FLOW_SET_PREFIX "/${SHM_PREFIX}.set." CACHE INTERNAL @@ -32,8 +33,8 @@ set(SSM_POOL_SHARDS 4 CACHE STRING set(SSM_POOL_RECLAIM_AGE_S 60 CACHE STRING "Minimum age in seconds before a block is presumed stale and reclaimed") -# Global Shared Packet Pool (GSPP) - for privileged processes -# Shared by all processes in 'ouroboros' group (~60 MB total) +# Global Shared Packet Pool (GSPP) - for privileged processes Shared by all processes in 'ouroboros' +# group (~60 MB total) set(SSM_GSPP_256_BLOCKS 1024 CACHE STRING "GSPP: Number of 256B blocks") set(SSM_GSPP_512_BLOCKS 2048 CACHE STRING @@ -53,8 +54,8 @@ set(SSM_GSPP_256K_BLOCKS 32 CACHE STRING set(SSM_GSPP_1M_BLOCKS 16 CACHE STRING "GSPP: Number of 1MB blocks") -# Per-User Pool (PUP) - for unprivileged applications -# Each unprivileged app gets its own smaller pool (~7.5 MB total) +# Per-User Pool (PUP) - for unprivileged applications Each unprivileged app gets its own smaller +# pool (~7.5 MB total) set(SSM_PUP_256_BLOCKS 512 CACHE STRING "PUP: Number of 256B blocks") set(SSM_PUP_512_BLOCKS 512 CACHE STRING @@ -147,10 +148,9 @@ message(STATUS " Blocks: ${SSM_PUP_256_BLOCKS}, ${SSM_PUP_512_BLOCKS}, " "${SSM_PUP_16K_BLOCKS}, ${SSM_PUP_64K_BLOCKS}, ${SSM_PUP_256K_BLOCKS}, " "${SSM_PUP_1M_BLOCKS}") -# FRCT reorder queue must fit in every enabled size class. If RQ_SIZE -# >= any backing pool, the receiver advertises a window the pool -# cannot back; np1_flow_write fails under load and a single dropped -# fragment wedges the flow. Auto-zeroed classes are skipped. +# FRCT reorder queue must fit in every enabled size class. If RQ_SIZE >= any backing pool, the +# receiver advertises a window the pool cannot back; np1_flow_write fails under load and a single +# dropped fragment wedges the flow. Auto-zeroed classes are skipped. foreach(_class 256 512 1K 2K) if(SSM_PUP_${_class}_BLOCKS GREATER 0 AND NOT FRCT_REORDER_QUEUE_SIZE LESS SSM_PUP_${_class}_BLOCKS) diff --git a/cmake/dependencies.cmake b/cmake/dependencies.cmake index ff44ad68..93c9c0de 100644 --- a/cmake/dependencies.cmake +++ b/cmake/dependencies.cmake @@ -17,19 +17,19 @@ include(dependencies/crypt/libgcrypt) # IRMd include(dependencies/irmd/libtoml) -# Ethernet IPCP backends +# Ethernet PoA backends. netmap is last: it overrides the others. include(dependencies/eth/rawsockets) include(dependencies/eth/bpf) include(dependencies/eth/netmap) + +# The Ethernet PoA needs a backend to send and receive frames. if(HAVE_RAW_SOCKETS OR HAVE_BPF OR HAVE_NETMAP) - set(HAVE_ETH TRUE CACHE INTERNAL "Ethernet IPCP support available") + set(HAVE_ETH TRUE) else() - unset(HAVE_ETH CACHE) + message(STATUS "No Ethernet backend, Ethernet PoAs disabled") + unset(HAVE_ETH) endif() -# UDP IPCP -include(dependencies/udp/ddns) - # Coverage tools include(dependencies/coverage/gcov) include(dependencies/coverage/lcov) diff --git a/cmake/dependencies/eth/netmap.cmake b/cmake/dependencies/eth/netmap.cmake index 94ecd634..b31c4d03 100644 --- a/cmake/dependencies/eth/netmap.cmake +++ b/cmake/dependencies/eth/netmap.cmake @@ -1,18 +1,22 @@ -# netmap support (optional acceleration) +# netmap support (kernel bypass). Explicit opt-in. find_path(NETMAP_C_INCLUDE_DIR net/netmap_user.h HINTS /usr/include /usr/local/include) mark_as_advanced(NETMAP_C_INCLUDE_DIR) -if(NOT HAVE_RAW_SOCKETS AND NOT HAVE_BPF AND NETMAP_C_INCLUDE_DIR) - set(DISABLE_NETMAP FALSE CACHE BOOL - "Disable netmap support for ETH IPCPs") - if(NOT DISABLE_NETMAP) - message(STATUS "Netmap support for Ethernet IPCPs enabled") - set(HAVE_NETMAP TRUE) - else() - message(STATUS "Netmap support for Ethernet IPCPs disabled by user") - unset(HAVE_NETMAP) +set(ENABLE_NETMAP FALSE CACHE BOOL + "Use netmap for Ethernet PoAs, overriding raw sockets or BPF") + +if(ENABLE_NETMAP) + if(NOT NETMAP_C_INCLUDE_DIR) + message(FATAL_ERROR "ENABLE_NETMAP is set, but netmap was not found.") endif() + + message(STATUS "Netmap support for Ethernet PoAs enabled") + + set(HAVE_NETMAP TRUE) + + unset(HAVE_RAW_SOCKETS) + unset(HAVE_BPF) endif() diff --git a/cmake/dependencies/udp/ddns.cmake b/cmake/dependencies/udp/ddns.cmake deleted file mode 100644 index e8208e47..00000000 --- a/cmake/dependencies/udp/ddns.cmake +++ /dev/null @@ -1,31 +0,0 @@ -# DDNS (Dynamic DNS) support detection -# Requires nsupdate and nslookup tools - -find_program(NSUPDATE_EXECUTABLE - NAMES nsupdate - DOC "The nsupdate tool that enables DDNS") - -find_program(NSLOOKUP_EXECUTABLE - NAMES nslookup - DOC "The nslookup tool that resolves DNS names") - -mark_as_advanced(NSLOOKUP_EXECUTABLE NSUPDATE_EXECUTABLE) - -if(NSLOOKUP_EXECUTABLE AND NSUPDATE_EXECUTABLE) - set(DISABLE_DDNS FALSE CACHE BOOL "Disable DDNS support") - if(NOT DISABLE_DDNS) - message(STATUS "DDNS support enabled") - set(HAVE_DDNS TRUE CACHE INTERNAL "Dynamic DNS support available") - else() - message(STATUS "DDNS support disabled by user") - unset(HAVE_DDNS CACHE) - endif() -else() - if(NSLOOKUP_EXECUTABLE) - message(STATUS "Install nsupdate to enable DDNS support") - elseif(NSUPDATE_EXECUTABLE) - message(STATUS "Install nslookup to enable DDNS support") - else() - message(STATUS "Install nslookup and nsupdate to enable DDNS support") - endif() -endif() diff --git a/doc/man/ouroboros-tutorial.7 b/doc/man/ouroboros-tutorial.7 index 1fc02a02..54abba5b 100644 --- a/doc/man/ouroboros-tutorial.7 +++ b/doc/man/ouroboros-tutorial.7 @@ -47,37 +47,56 @@ The output should be .SH PINGING A SERVER APPLICATION OVER THE LOOPBACK ADAPTER With a running irmd, let's create an IPC process. For this tutorial, -we will create and bootstrap an eth-llc IPCP over the loopback -interface. To observe what's going on, open another terminal -window. Note that "ipcp bootstrap" will create an IPCP if an IPCP by -that name does not yet exist (See \fBouroboros\fR(8)). +we will create a unicast IPCP, attach it to the loopback interface and +bootstrap it. To observe what's going on, open another terminal +window. .RS 4 -$ irm ipcp bootstrap type eth-llc name llc layer llc if lo +$ irm ipcp create name lan type unicast .RE .RS 4 ==23918== irmd(II): Created IPCP 23932. -.br -==23932== ipcpd/eth-llc(II): Using raw socket device. -.br -==23918== irmd(II): Bootstrapped IPCP 23932 in layer llc. +.RE + +An IPCP needs a \fIpoint of attachment\fR: the transmission technology +it sends and receives on. Attach it to the loopback interface (See +\fBouroboros\fR(8)). + +.RS 4 +$ irm ipcp poa attach name lan eth dev lo +.RE + +.RS 4 +==23918== irmd(II): Attached IPCP 23932. +.RE + +.RS 4 +$ irm ipcp bootstrap name lan layer lan +.RE + +.RS 4 +==23918== irmd(II): Bootstrapped IPCP 23932. .RE Now that we have the IPCP bootstrapped, it can act as a local network layer that can provide full connectivity between all processes in the system. Let's test it using the oping application. First, let's choose -a name for the server ("my.oping.server") and register in the llc +a name for the server ("my.oping.server") and register it in the lan layer. .RS 4 -$ irm reg name my.oping.server layer llc +$ irm name create my.oping.server +.br +$ irm name register my.oping.server layer lan .RE The IRMd should respond with .RS 4 -==23918== irmd(II): Registered my.oping.server in llc as 716016b1. +==23918== irmd(II): Created new name: my.oping.server. +.br +==23918== irmd(II): Registered my.oping.server with IPCP 23932 as 716016b1. .RE Now start a server of oping in the background (or in a different diff --git a/doc/man/ouroboros.8 b/doc/man/ouroboros.8 index 759b1433..4c83e5a9 100644 --- a/doc/man/ouroboros.8 +++ b/doc/man/ouroboros.8 @@ -98,10 +98,13 @@ accessed by other processes. In order to enroll an IPC process in a layer, some other member will have to be reachable over a lower layer. IPCPs that wrap a legacy transmission technology are all bootstrapped and thus need not enroll -as they work directly over a physical connection. Ouroboros currently -supports IPCPs over shared memory (local), L2 (eth-llc and eth-dix) -and L3 (udp). The unicast and broadcast layers require connections to -be established between IPCP components for its operation. +as they work directly over a physical connection. A unicast or +broadcast IPCP attaches to a transmission technology itself, as a +\fIpoint of attachment\fR (PoA); Ouroboros currently supports PoAs on +L2 (Ethernet) and L3 (UDP/IPv4 and UDP/IPv6), next to IPCPs over +shared memory (local). The unicast and broadcast layers require +connections to be established between IPCP components for its +operation. \fBConnecting the management components\fR using \fImanagement flows\fR allows management information to be sent between IPCPs so @@ -123,14 +126,6 @@ creates an IPCP process of type \fItype\fR in the system with name .PP \fBlocal\fR - create a loopback IPCP. .PP -\fBeth-llc\fR - create an IPCP that attaches to Ethernet using LLC frames. -.PP -\fBeth-dix\fR - create an IPCP that attaches to Ethernet using DIX frames. -.PP -\fBudp4\fR - create an IPCP that attaches to a UDP/IPv4 socket. -.PP -\fBudp6\fR - create an IPCP that attaches to a UDP/IPv6 socket. -.PP \fBunicast\fR - create a unicast IPCP that uses lower level layers. .PP \fBbroadcast\fR - create a broadcast IPCP that uses lower level layers. @@ -161,69 +156,9 @@ Values for [\fIparam\fR] are dependent on \fItype\fR: default: SHA3_256. .RE -.PP -\fBeth-llc\fR -.RS 4 -.PP -dev \fIinterface\fR specifies the interface to bind the IPCP to. -.PP -[hash \fIpolicy\fR] specifies the hash function used for the directory, -.br -\fIpolicy\fR: SHA3_224, SHA3_256, SHA3_384, SHA3_512. -.br -default: SHA3_256. -.RE -.PP -\fBeth-dix\fR -.RS 4 -.PP -dev \fIinterface\fR specifies the interface to bind the IPCP to. -.PP -[ethertype \fIethertype\fR] specifies the ethertype used for the layer. -.br -default: 0xA000. -.PP -[hash \fIpolicy\fR] specifies the hash function used for the directory, -.br -\fIpolicy\fR: SHA3_224, SHA3_256, SHA3_384, SHA3_512. -.br -default: SHA3_256. -.RE -.PP -\fBudp4\fR -.RS 4 -.PP -ip \fIip\fR specifies the local IPv4 address to bind to -.PP -[dns \fIdns\fR] specifies an optional DDNS server that will be used for -the directory. -.PP -[port \fIport\fR] specifies a UDP port that is used for sending and -receiving ouroboros traffic. This must be the same for the entire UDP4 -layer. Parallel UDP4 layers should use different ports. This UDP port -needs to be forwarded if the server is behind a NAT and wants to -receive incoming requests. -.br -default: 3435 -.RE -.PP -\fBudp6\fR -.RS 4 -.PP -ip \fIip\fR specifies the local IPv6 address to bind to -.PP -[dns \fIdns\fR] specifies an optional DDNS server that will be used for -the directory. -.PP -[port \fIport\fR] specifies a UDP port that is used for sending and -receiving ouroboros traffic. This must be the same for the entire UDP6 -layer. Parallel UDP6 layers should use different ports. -.br -default: 3435 -.RE .PP \fBunicast\fR @@ -293,6 +228,38 @@ default: SHA3_256. .RE .PP +\fBirm ipcp poa attach\fR name \fIname\fR \fIpoa\fR +.RS 4 +attaches the IPCP with name \fIname\fR to a point of attachment. +Repeat to attach more than one. Exactly one \fIpoa\fR is given: +.PP +udp \fIip\fR[:\fIport\fR] attaches to a local IPv4 or IPv6 address. +IPv6 addresses need brackets when a port is given: [\fIip\fR]:\fIport\fR. +.br +default port: 3435. +.PP +eth dev \fIdevice\fR [ethertype \fIethertype\fR] attaches to an Ethernet +device. +.br +default: 0xA000. +.RE + +.PP +\fBirm ipcp poa detach\fR name \fIname\fR \fIpoa\fR +.RS 4 +releases a point of attachment that the IPCP with name \fIname\fR is +attached to. Flows over it are brought down first. \fIpoa\fR takes the +same values as for \fBattach\fR. +.RE + +.PP +\fBirm ipcp poa list\fR name \fIname\fR +.RS 4 +lists the points of attachment that the IPCP with name \fIname\fR is +attached to. +.RE + +.PP \fBirm ipcp enroll\fR name \fIname\fR [type \fItype\fR] [dst \fIdst\fR] \ [layer \fIlayer\fR] [\fIautobind\fR] .RS 4 @@ -308,6 +275,15 @@ the layer name is a shorthand for the destination name being the same as the layer name. .PP [autobind] will automatically bind this IPCP to its name and the layer name. +.PP +[udp \fIip\fR[:\fIport\fR]] enrols over a point of attachment this +IPCP is attached to, at a peer reachable at this IPv4 or IPv6 address +or host name. +.PP +[eth [dev \fIdevice\fR] [ethertype \fIethertype\fR]] enrols over a +point of attachment this IPCP is attached to, resolving the peer by a +name query over Ethernet. [dev \fIdevice\fR] restricts the query to +one attached device. .RE \fBirm ipcp connect\fR name \fIname\fR component \fIcomponent\fR dst @@ -316,6 +292,18 @@ as the layer name. connects a \fIcomponent\fR (\fBdt\fR or \fBmgmt\fR) of a unicast or broadcast IPCP with name \fIname\fR to that component of the destination IPCP within the same layer. +.PP +[qos \fIqos\fR] specifies the QoS cube for a \fBdt\fR flow: raw, safe, +rt, rt-safe or msg. +.PP +[udp \fIip\fR[:\fIport\fR]] connects over a point of attachment +this IPCP is attached to, to a peer reachable at this IPv4 or IPv6 +address or host name. +.PP +[eth [dev \fIdevice\fR] [ethertype \fIethertype\fR]] connects over a +point of attachment this IPCP is attached to, resolving the peer by a +name query over Ethernet. [dev \fIdevice\fR] restricts the query to +one attached device. .RE \fBirm ipcp disconnect\fR name \fIname\fR component \fIcomponent\fR dst diff --git a/include/ouroboros/ipcp-dev.h b/include/ouroboros/ipcp-dev.h index d00d6f08..8677adf0 100644 --- a/include/ouroboros/ipcp-dev.h +++ b/include/ouroboros/ipcp-dev.h @@ -24,55 +24,106 @@ #define OUROBOROS_LIB_IPCP_DEV_H #include <ouroboros/ipcp.h> +#include <ouroboros/qos.h> #include <ouroboros/qoscube.h> #include <ouroboros/ssm_pool.h> #include <ouroboros/utils.h> +#include <stdbool.h> #include <stdint.h> +#include <time.h> -int ipcp_create_r(const struct ipcp_info * info); +int ipcp_create_r(const struct ipcp_info * info); -int ipcp_flow_req_arr(const buffer_t * dst, - qosspec_t qs, - time_t mpl, - uint32_t mtu, - const buffer_t * data); +int ipcp_flow_req_arr(const buffer_t * dst, + qosspec_t qs, + time_t mpl, + uint32_t mtu, + const buffer_t * data); -int ipcp_flow_update_arr(int flow_id, - const buffer_t * data); - -int ipcp_flow_alloc_reply(int fd, - int response, - time_t mpl, - uint32_t mtu, +int ipcp_flow_update_arr(int flow_id, const buffer_t * data); -int ipcp_flow_read(int fd, - struct ssm_pk_buff ** spb); +int ipcp_flow_alloc_reply(int fd, + int response, + time_t mpl, + uint32_t mtu, + const buffer_t * data); + +int ipcp_flow_read(int fd, + struct ssm_pk_buff ** spb); + +int ipcp_flow_write(int fd, + struct ssm_pk_buff * spb); + +int np1_flow_read(int fd, + struct ssm_pk_buff ** spb, + struct ssm_pool * pool); + +int np1_flow_write(int fd, + struct ssm_pk_buff * spb, + struct ssm_pool * pool); + +int ipcp_flow_dealloc(int fd); + +int ipcp_flow_fini(int fd); + +int ipcp_flow_get_qoscube(int fd, + qoscube_t * cube); + +size_t ipcp_flow_queued(int fd); -int ipcp_flow_write(int fd, - struct ssm_pk_buff * spb); +/* + * Identifies the transmit queue behind the flow. Flows attached to + * one point of attachment share a queue and answer the same id. + */ +int ipcp_flow_queue_id(int fd); + +int ipcp_spb_reserve(struct ssm_pk_buff ** spb, + size_t len); + +void ipcp_spb_release(struct ssm_pk_buff * spb); + +/* PoA lifecycle; an IPCP owns its PoAs, applications have none. */ +int poa_init(const char * name); -int np1_flow_read(int fd, - struct ssm_pk_buff ** spb, - struct ssm_pool * pool); +int poa_start(void); -int np1_flow_write(int fd, - struct ssm_pk_buff * spb, - struct ssm_pool * pool); +void poa_stop(void); -int ipcp_flow_dealloc(int fd); +void poa_fini(void); -int ipcp_flow_fini(int fd); +/* Also answer name queries for the layer; set at bootstrap/enroll. */ +int poa_set_layer(const char * layer); -int ipcp_flow_get_qoscube(int fd, - qoscube_t * cube); +/* Attach or release a PoA; the type in the spec picks the transport. */ +int poa_attach(const struct poa_spec * poa); -size_t ipcp_flow_queued(int fd); +int poa_detach(const struct poa_spec * poa); + +/* The PoAs this process has attached. */ +ssize_t poa_list(struct poa_spec * specs, + size_t max); + +/* Allocate a flow to dst over the PoA that carries addr. */ +int poa_flow_alloc(const char * dst, + const struct poa_addr * addr, + qosspec_t * qs, + const struct timespec * timeo); + +/* Resolve dst on the attached PoAs; no flow is created. */ +int poa_query(const char * dst, + const struct timespec * timeo, + struct poa_addr * addr); + +/* PoA counterparts of the ipcp_flow_* operations */ +int poa_flow_alloc_resp(int flow_id, + int response, + const buffer_t * data); -int ipcp_spb_reserve(struct ssm_pk_buff ** spb, - size_t len); +int poa_flow_update(int flow_id, + const buffer_t * data); -void ipcp_spb_release(struct ssm_pk_buff * spb); +int poa_flow_dealloc(int flow_id); #endif /* OUROBOROS_LIB_IPCP_DEV_H */ diff --git a/include/ouroboros/ipcp.h b/include/ouroboros/ipcp.h index 135b8fcb..4da3b81d 100644 --- a/include/ouroboros/ipcp.h +++ b/include/ouroboros/ipcp.h @@ -48,13 +48,19 @@ enum ipcp_type { /* IRMd uses order to select an IPCP for flow allocation. */ IPCP_LOCAL = 0, IPCP_UNICAST, IPCP_BROADCAST, - IPCP_ETH_LLC, - IPCP_ETH_DIX, - IPCP_UDP4, - IPCP_UDP6, IPCP_INVALID }; +/* The medium a PoA sits on. */ +enum poa_type { + POA_UDP4 = 0, + POA_UDP6, + /* A name to resolve; the family is DNS's pick. Port in udp4. */ + POA_UDP, + POA_ETH, + POA_INVALID +}; + struct ipcp_info { enum ipcp_type type; pid_t pid; @@ -256,23 +262,52 @@ static const struct uni_config default_uni_config = { .cong_avoid = CA_MB_ECN }; -struct eth_config { +#define POA_UDP_PORT 3435 /* default UDP PoA port */ +#define POA_ETHERTYPE 0xA000 /* default Ethertype */ +#define POA_MAC_SIZE 6 +#define POA_HOST_STRLEN 255 + +struct eth_poa { char dev[DEV_NAME_SIZE + 1]; - uint16_t ethertype; /* DIX only*/ + uint16_t ethertype; + uint8_t mac[POA_MAC_SIZE]; }; -struct udp4_config { +struct udp4_poa { struct in_addr ip_addr; - struct in_addr dns_addr; uint16_t port; }; -struct udp6_config { +struct udp6_poa { struct in6_addr ip_addr; - struct in6_addr dns_addr; uint16_t port; }; +/* PoA a unicast or broadcast IPCP attaches to. */ +struct poa_spec { + enum poa_type type; + union { + struct udp4_poa udp4; + struct udp6_poa udp6; + struct eth_poa eth; + }; +}; + + +/* Peer PoA to connect to. */ +struct poa_addr { + enum poa_type type; + char hostname[POA_HOST_STRLEN + 1]; /* resolved by IRMd */ + union { + struct udp4_poa udp4; + struct udp6_poa udp6; + struct { + struct eth_poa src; + struct eth_poa dst; + } eth; + }; +}; + /* Layers */ struct layer_info { char name[LAYER_NAME_SIZE + 1]; @@ -285,58 +320,22 @@ struct ipcp_config { struct layer_info layer_info; enum ipcp_type type; - union { - struct uni_config unicast; - struct udp4_config udp4; - struct udp6_config udp6; - struct eth_config eth; - }; + struct uni_config unicast; }; /* default configurations */ static const struct ipcp_config local_default_conf = { - .type = IPCP_LOCAL, - .layer_info = { - .dir_hash_algo = DIR_HASH_SHA3_256 - } -}; - -static const struct ipcp_config eth_dix_default_conf = { - .type = IPCP_ETH_DIX, .layer_info = { .dir_hash_algo = DIR_HASH_SHA3_256 }, - .eth = { - .ethertype=0xA000, - } -}; - -static const struct ipcp_config eth_llc_default_conf = { - .type = IPCP_ETH_LLC, - .layer_info = { - .dir_hash_algo = DIR_HASH_SHA3_256 - } -}; - -static const struct ipcp_config udp4_default_conf = { - .type = IPCP_UDP4, - .udp4 = { - .port = 3435 - } -}; - -static const struct ipcp_config udp6_default_conf = { - .type = IPCP_UDP6, - .udp6 = { - .port = 3435 - } + .type = IPCP_LOCAL }; static const struct ipcp_config uni_default_conf = { - .type = IPCP_UNICAST, .layer_info = { .dir_hash_algo = DIR_HASH_SHA3_256 }, + .type = IPCP_UNICAST, .unicast = { .dt = { .addr_size = 4, diff --git a/include/ouroboros/irm.h b/include/ouroboros/irm.h index 7cb71c21..3e8a24af 100644 --- a/include/ouroboros/irm.h +++ b/include/ouroboros/irm.h @@ -46,16 +46,29 @@ int irm_destroy_ipcp(pid_t pid); ssize_t irm_list_ipcps(struct ipcp_list_info ** ipcps); -int irm_enroll_ipcp(pid_t pid, - const char * dst); +int irm_enroll_ipcp(pid_t pid, + const char * dst, + const struct poa_addr * addr); int irm_bootstrap_ipcp(pid_t pid, const struct ipcp_config * conf); -int irm_connect_ipcp(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs); +/* Attach an IPCP to a medium; repeat for more than one PoA. */ +int irm_attach_ipcp(pid_t pid, + const struct poa_spec * poa); + +int irm_detach_ipcp(pid_t pid, + const struct poa_spec * poa); + +/* Caller frees *poas. */ +ssize_t irm_list_poas(pid_t pid, + struct poa_spec ** poas); + +int irm_connect_ipcp(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr); int irm_disconnect_ipcp(pid_t pid, const char * dst, diff --git a/include/ouroboros/protobuf.h b/include/ouroboros/protobuf.h index 951e8fd6..7937f86b 100644 --- a/include/ouroboros/protobuf.h +++ b/include/ouroboros/protobuf.h @@ -24,9 +24,9 @@ #define OUROBOROS_LIB_PROTOBUF_H #include <ouroboros/flow.h> -#include <ouroboros/qos.h> #include <ouroboros/ipcp.h> #include <ouroboros/irm.h> +#include <ouroboros/qos.h> #include <ouroboros/serdes-irm.h> #include <ouroboros/serdes-oep.h> @@ -37,9 +37,10 @@ typedef RoutingConfigMsg routing_config_msg_t; typedef DtConfigMsg dt_config_msg_t; typedef DirConfigMsg dir_config_msg_t; typedef DirDhtConfigMsg dir_dht_config_msg_t; -typedef EthConfigMsg eth_config_msg_t; -typedef Udp4ConfigMsg udp4_config_msg_t; -typedef Udp6ConfigMsg udp6_config_msg_t; +typedef EthPoaMsg eth_poa_msg_t; +typedef Udp4PoaMsg udp4_poa_msg_t; +typedef PoaSpecMsg poa_spec_msg_t; +typedef Udp6PoaMsg udp6_poa_msg_t; typedef UniConfigMsg uni_config_msg_t; #include "ipcp.pb-c.h" @@ -56,6 +57,7 @@ typedef FlowInfoMsg flow_info_msg_t; typedef NameInfoMsg name_info_msg_t; typedef LayerInfoMsg layer_info_msg_t; typedef QosspecMsg qosspec_msg_t; +typedef PoaAddrMsg poa_addr_msg_t; #include "enroll.pb-c.h" typedef EnrollReqMsg enroll_req_msg_t; @@ -79,6 +81,10 @@ layer_info_msg_t * layer_info_s_to_msg(const struct layer_info * s); struct layer_info layer_info_msg_to_s(const layer_info_msg_t * msg); +poa_addr_msg_t * poa_addr_s_to_msg(const struct poa_addr * s); + +struct poa_addr poa_addr_msg_to_s(const poa_addr_msg_t * msg); + ipcp_info_msg_t * ipcp_info_s_to_msg(const struct ipcp_info * s); struct ipcp_info ipcp_info_msg_to_s(const ipcp_info_msg_t * msg); @@ -91,20 +97,24 @@ uni_config_msg_t * uni_config_s_to_msg(const struct uni_config * s); struct uni_config uni_config_msg_to_s(const uni_config_msg_t * msg); -eth_config_msg_t * eth_config_s_to_msg(const struct eth_config * s); +eth_poa_msg_t * eth_poa_s_to_msg(const struct eth_poa * s); -struct eth_config eth_config_msg_to_s(const eth_config_msg_t * msg); +struct eth_poa eth_poa_msg_to_s(const eth_poa_msg_t * msg); -udp4_config_msg_t * udp4_config_s_to_msg(const struct udp4_config * s); +udp4_poa_msg_t * udp4_poa_s_to_msg(const struct udp4_poa * s); -struct udp4_config udp4_config_msg_to_s(const udp4_config_msg_t * msg); +struct udp4_poa udp4_poa_msg_to_s(const udp4_poa_msg_t * msg); -udp6_config_msg_t * udp6_config_s_to_msg(const struct udp6_config * s); +udp6_poa_msg_t * udp6_poa_s_to_msg(const struct udp6_poa * s); -struct udp6_config udp6_config_msg_to_s(const udp6_config_msg_t * msg); +struct udp6_poa udp6_poa_msg_to_s(const udp6_poa_msg_t * msg); ipcp_config_msg_t * ipcp_config_s_to_msg(const struct ipcp_config * s); +poa_spec_msg_t * poa_spec_s_to_msg(const struct poa_spec * s); + +struct poa_spec poa_spec_msg_to_s(const poa_spec_msg_t * msg); + struct ipcp_config ipcp_config_msg_to_s(const ipcp_config_msg_t * msg); /* QoS */ diff --git a/include/ouroboros/qos.h b/include/ouroboros/qos.h index 7980ad00..26fc2637 100644 --- a/include/ouroboros/qos.h +++ b/include/ouroboros/qos.h @@ -30,9 +30,9 @@ /* qos_spec.service: framing / reliability class. */ enum qos_service { - SVC_RAW = 0, /* No FRCT; best-effort raw messages */ - SVC_MESSAGE = 1, /* FRCT, reliable ordered messages */ - SVC_STREAM = 2, /* FRCT, reliable ordered byte stream */ + SVC_RAW = 0, /* No FRCT; best-effort raw messages */ + SVC_MESSAGE = 1, /* FRCT, ordered messages, optional reliability */ + SVC_STREAM = 2, /* FRCT, reliable ordered byte stream */ }; typedef struct qos_spec { diff --git a/include/ouroboros/serdes-irm.h b/include/ouroboros/serdes-irm.h index a5854d5b..bc3c46d2 100644 --- a/include/ouroboros/serdes-irm.h +++ b/include/ouroboros/serdes-irm.h @@ -52,6 +52,23 @@ int ipcp_flow_req_arr__irm_req_ser(buffer_t * buf, const struct flow_info * flow, const buffer_t * data); +int poa_flow_alloc__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const char * dst); + +int poa_flow_alloc_r__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data, + int response); + +int ipcp_poa_flow_req_arr__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data); + +int poa_flow__irm_result_des(buffer_t * buf, + struct flow_info * flow, + buffer_t * data); + int ipcp_flow_update_arr__irm_req_ser(buffer_t * buf, const struct flow_info * flow, const buffer_t * data); diff --git a/irmd.conf.in b/irmd.conf.in index 02cb1da3..c3cfe5e5 100644 --- a/irmd.conf.in +++ b/irmd.conf.in @@ -7,10 +7,6 @@ # # [name.<name of service>] add a new name to the system. # [local.<name for IPCP>] add a new local IPCP to the system. -# [eth-llc.<name for IPCP>] add a new IPCP over Ethernet (LLC) to the system. -# [eth-dix.<name for IPCP>] add a new IPCP over Ethernet (DIX) to the system. -# [udp4.<name for IPCP>] add a new IPCP over UDP/IPv4 to the system. -# [udp6.<name for IPCP>] add a new IPCP over UDP/IPv6 to the system. # [broadcast.<name of IPCP>] add a new broadcast IPCP to the system. # [unicast.<name of IPCP>] add a new unicast IPCP to the system. # @@ -32,6 +28,49 @@ # enrol=<dst>: Enrol with this neigbor (broadcast and unicast only). # conn=[<dst>]: Create mgmt and data flows with these neighbors. # reg=[<names>]: Register these names with the IPCP. +# +# enrol, and each element of conn, also accept a table instead of a bare +# name, to reach that peer over one specific point of attachment: +# +# {dst=<name>, eth={}} Resolve dst by query over every +# attached eth PoA. +# {dst=<name>, eth={dev=<device>}} Same, pinned to one device. eth +# accepts an optional +# ethertype=<ethertype>, default +# 0xA000. +# {dst=<name>, udp=<addr>[:<port>]} Literal peer; IPv4, bracketed IPv6 +# ("[::1]:3435"), bare IPv6, or a +# host name the IRMd resolves. Port +# defaults to 3435. +# +# A table names at most one of eth or udp; dst is required. +# +# A unicast or broadcast IPCP binds its own points of attachment. udp +# and eth are each an array of strings or tables, one udp for both +# address families: +# +# udp = [<addr>, ...] <addr> is a literal IPv4 or +# IPv6 address, with an optional +# port ("<addr>:<port>", +# bracketed for IPv6: +# "[::1]:3435"). Port defaults +# to 3435. A table, +# {addr=<addr>[:<port>]}, is +# also accepted. +# eth = [<device>, ...] <device> is a bare device +# name, or a table +# {dev=<device>[, ethertype= +# <ethertype>]}. Ethertype +# defaults to 0xA000. +# +# Because a table element is accepted, the array-of-tables spelling +# still works too, e.g. [[unicast.lan1.udp]] with addr="<addr>" as a +# separate stanza; repeat the stanza, or the array, to attach more +# than one PoA. +# +# Unlike conn/enrol above, a PoA binds locally: only literal addresses +# are accepted here, not host names. +# # Bootstrap options can be set as in the examples below. # For more details on the configuration options for each of the IPCP types, # please refer to the Ouroboros man page. @@ -57,11 +96,11 @@ prog=["@INSTALL_DIR@/oping"] # Defaults to []. args=["--listen"] # Defaults to disabled. Autostart server with these args. lb="round-robin" # Defaults to spill (load-balancing options: spill, round-robin). # server_sec_file=/path/to/sec.conf Default: @OUROBOROS_SRV_CRT_DIR@/<name>/sec.conf -# server_crt_file=/path/to/crt.pem Default: @OUROBOROS_SRV_CRT_DIR@/<name>/crt.pem -# server_key_file=/path/to/key.pem Default: @OUROBOROS_SRV_CRT_DIR@/<name>/key.pem +# server_crt_file=/path/to/crt.pem Default: @OUROBOROS_SRV_CRT_DIR@/<name>/crt.pem +# server_key_file=/path/to/key.pem Default: @OUROBOROS_SRV_CRT_DIR@/<name>/key.pem # client_sec_file=/path/to/sec.conf Default: @OUROBOROS_CLI_CRT_DIR@/<name>/sec.conf -# client_crt_file=/path/to/crt.pem Default: @OUROBOROS_CLI_CRT_DIR@/<name>/crt.pem -# client_key_file=/path/to/key.pem Default: @OUROBOROS_CLI_CRT_DIR@/<name>/key.pem +# client_crt_file=/path/to/crt.pem Default: @OUROBOROS_CLI_CRT_DIR@/<name>/crt.pem +# client_key_file=/path/to/key.pem Default: @OUROBOROS_CLI_CRT_DIR@/<name>/key.pem [local.local1] bootstrap="local1" # Defaults to not set. @@ -70,34 +109,6 @@ bootstrap="local1" # Defaults to not set. # NAMES KNOWN reg=["bc1", "LAN"] # Defaults to []. -[eth-llc.ethl] -bootstrap="ethl1" # Defaults to not set. - # BOOTSTRAP CONFIGURATION -dev="lo" -# hash="SHA3_224" # Defaults to SHA3_256. - -[eth-dix.eth1] -bootstrap="eth1" # Defaults to not set. - # BOOTSTRAP CONFIGURATION -dev="lo" -# ethertype=0xA007 # Defaults to 0xA000. -# hash="SHA3_224" # Defaults to SHA3_256. -reg=["lan1"] - -[udp4.udp1] -bootstrap="udp" # Defaults to not set. - # BOOTSTRAP CONFIGURATION -ip="127.0.0.1" -# port=9000 # Defaults to 3435. -# dns="127.0.0.1" # Requires a DDNS server. Disables DDNS support if not set. - -[udp6.udp2] -bootstrap="udp2" # Defaults to not set. - # BOOTSTRAP CONFIGURATION -ip="::1" -# port=9000 # Defaults to 3435. -# dns="::1" # Requires a DDNS server. Disables DDNS support if not set. - [broadcast.bc1] bootstrap="broadcast" # Defaults to not set. # autobind=true # Defaults to false. @@ -130,7 +141,19 @@ autobind=true # Defaults to false. # NAMES KNOWN reg=["oping"] # Defaults to []. + # POINTS OF ATTACHMENT +udp = ["127.0.0.1", "127.0.0.1:3436"] # Two PoAs; port defaults to 3435. +# udp = ["127.0.0.1:9000"] # Or pick a port explicitly. + +[[unicast.lan1.eth]] +dev="lo" +# ethertype=0xA007 # Defaults to 0xA000. + [unicast.lan2] enrol="LAN" # Defaults to not set. +# enrol={dst="LAN", eth={dev="eth0"}} # Or enrol over a specific PoA. conn=["lan1"] # Defaults to []. +# conn=["lan1", +# {dst="lan3", eth={}}, +# {dst="lan4", udp="10.0.0.1:3435"}] # Or connect over specific PoAs. autobind=true # Defaults to false. diff --git a/src/ipcpd/CMakeLists.txt b/src/ipcpd/CMakeLists.txt index 609da54a..a84e5369 100644 --- a/src/ipcpd/CMakeLists.txt +++ b/src/ipcpd/CMakeLists.txt @@ -4,7 +4,6 @@ # Common sources shared by all IPCPs (absolute paths for subdirectories) set(IPCP_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/ipcp.c - ${CMAKE_CURRENT_SOURCE_DIR}/shim-data.c ) set(COMMON_SOURCES @@ -24,7 +23,3 @@ configure_file("${CMAKE_CURRENT_SOURCE_DIR}/config.h.in" add_subdirectory(local) add_subdirectory(broadcast) add_subdirectory(unicast) -if(HAVE_ETH) - add_subdirectory(eth) -endif() -add_subdirectory(udp) diff --git a/src/ipcpd/broadcast/main.c b/src/ipcpd/broadcast/main.c index d18cac82..22b34a76 100644 --- a/src/ipcpd/broadcast/main.c +++ b/src/ipcpd/broadcast/main.c @@ -37,6 +37,7 @@ #include <ouroboros/logs.h> #include <ouroboros/notifier.h> #include <ouroboros/np1_flow.h> +#include <ouroboros/qos.h> #include <ouroboros/random.h> #include <ouroboros/rib.h> #include <ouroboros/time.h> @@ -100,12 +101,14 @@ static void stop_components(void) enroll_stop(); } -static int broadcast_ipcp_enroll(const char * dst, - struct layer_info * info) +static int broadcast_ipcp_enroll(const char * dst, + const struct poa_addr * addr, + struct layer_info * info) { struct ipcp_config * conf; - struct conn conn; - uint8_t id[ENROLL_ID_LEN]; + struct conn conn; + uint8_t id[ENROLL_ID_LEN]; + qosspec_t qs = qos_msg; if (random_buffer(id, ENROLL_ID_LEN) < 0) { log_err("Failed to generate enrollment ID."); @@ -114,7 +117,7 @@ static int broadcast_ipcp_enroll(const char * dst, log_info_id(id, "Requesting enrollment."); - if (connmgr_alloc(COMPID_ENROLL, dst, NULL, &conn) < 0) { + if (connmgr_alloc(COMPID_ENROLL, dst, &qs, addr, &conn) < 0) { log_err_id(id, "Failed to get connection."); goto fail_id; } diff --git a/src/ipcpd/common/connmgr.c b/src/ipcpd/common/connmgr.c index 6384a45c..e0ad80cb 100644 --- a/src/ipcpd/common/connmgr.c +++ b/src/ipcpd/common/connmgr.c @@ -26,6 +26,7 @@ #include <ouroboros/dev.h> #include <ouroboros/errno.h> #include <ouroboros/fccntl.h> +#include <ouroboros/ipcp-dev.h> #include <ouroboros/list.h> #include <ouroboros/logs.h> #include <ouroboros/notifier.h> @@ -39,6 +40,10 @@ #include <stdlib.h> #include <string.h> +#define CONNMGR_ETH_PROBE_TIMEO 20 /* ms, one query attempt */ +#define CONNMGR_ETH_RETRY_TIMEO 1500 /* ms, the remaining tries */ +#define CONNMGR_DHT_TIMEO 1000 /* ms, bounded lower-layer */ + struct conn_el { struct list_head next; struct conn conn; @@ -60,6 +65,14 @@ struct { pthread_t acceptor; } connmgr; +static bool is_eth_query(const struct poa_addr * addr) +{ + static const uint8_t zero[POA_MAC_SIZE] = { 0 }; + + return addr->type == POA_ETH && + memcmp(addr->eth.dst.mac, zero, POA_MAC_SIZE) == 0; +} + static int get_id_by_name(const char * name) { enum comp_id i; @@ -121,6 +134,7 @@ static int add_comp_conn(enum comp_id id, return 0; } +/* qs is also an in-parameter, and flow_accept writes it back. */ static void * flow_acceptor(void * o) { int fd; @@ -323,9 +337,10 @@ void connmgr_comp_fini(enum comp_id id) memset(&connmgr.comps[id].info, 0, sizeof(connmgr.comps[id].info)); } -int connmgr_ipcp_connect(const char * dst, - const char * component, - qosspec_t qs) +int connmgr_ipcp_connect(const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr) { struct conn_el * ce; int id; @@ -334,6 +349,11 @@ int connmgr_ipcp_connect(const char * dst, assert(dst); assert(component); + if (qs.service == SVC_STREAM) { + log_err("No stream service on component flows."); + return -ENOTSUP; + } + ce = malloc(sizeof(*ce)); if (ce == NULL) { log_err("Out of memory."); @@ -348,7 +368,7 @@ int connmgr_ipcp_connect(const char * dst, pthread_cleanup_push(free, ce); - ret = connmgr_alloc(id, dst, &qs, &ce->conn); + ret = connmgr_alloc(id, dst, &qs, addr, &ce->conn); pthread_cleanup_pop(false); @@ -414,10 +434,46 @@ int connmgr_ipcp_disconnect(const char * dst, return 0; } -int connmgr_alloc(enum comp_id id, - const char * dst, - qosspec_t * qs, - struct conn * conn) +/* + * Without an address, a peer may be on the wire or reachable through + * the layer below. A PoA query is cheap and creates no flow, so it + * goes first; the layer below gets a bounded try before the query + * retries, and the last try is unbounded. + */ +static int alloc_any(const char * dst, + qosspec_t * qs) +{ + struct timespec probe = TIMESPEC_INIT_MS(CONNMGR_ETH_PROBE_TIMEO); + struct timespec retry = TIMESPEC_INIT_MS(CONNMGR_ETH_RETRY_TIMEO); + struct timespec below = TIMESPEC_INIT_MS(CONNMGR_DHT_TIMEO); + struct poa_addr addr; + int fd; + + if (poa_query(dst, &probe, &addr) == 0) { + fd = poa_flow_alloc(dst, &addr, qs, NULL); + if (fd >= 0) + return fd; + } + + fd = flow_alloc(dst, qs, &below); + if (fd >= 0) + return fd; + + if (poa_query(dst, &retry, &addr) == 0) { + fd = poa_flow_alloc(dst, &addr, qs, NULL); + if (fd >= 0) + return fd; + } + + return flow_alloc(dst, qs, NULL); +} + +/* A literal peer address bypasses the layer below. */ +int connmgr_alloc(enum comp_id id, + const char * dst, + qosspec_t * qs, + const struct poa_addr * addr, + struct conn * conn) { struct comp * comp; int fd; @@ -428,7 +484,26 @@ int connmgr_alloc(enum comp_id id, comp = connmgr.comps + id; - fd = flow_alloc(dst, qs, NULL); + if (addr != NULL) + fd = poa_flow_alloc(dst, addr, qs, NULL); + else + fd = alloc_any(dst, qs); + + if (fd == -EPERM && addr != NULL) { + log_err("No PoA attached to reach %s.", dst); + goto fail_alloc; + } + + if (fd == -EINVAL && addr != NULL) { + log_err("More than one PoA could reach %s", dst); + goto fail_alloc; + } + + if (fd == -ETIMEDOUT && addr != NULL && is_eth_query(addr)) { + log_err("No answer to name query for %s.", dst); + goto fail_alloc; + } + if (fd < 0) { log_err("Failed to allocate flow to %s.", dst); goto fail_alloc; diff --git a/src/ipcpd/common/connmgr.h b/src/ipcpd/common/connmgr.h index f48ecd1b..86f80fb0 100644 --- a/src/ipcpd/common/connmgr.h +++ b/src/ipcpd/common/connmgr.h @@ -24,6 +24,7 @@ #define OUROBOROS_IPCPD_COMMON_CONNMGR_H #include <ouroboros/cep.h> +#include <ouroboros/ipcp.h> #include <ouroboros/qos.h> #include "comp.h" @@ -53,17 +54,19 @@ int connmgr_comp_init(enum comp_id id, void connmgr_comp_fini(enum comp_id id); -int connmgr_ipcp_connect(const char * dst, - const char * component, - qosspec_t qs); +int connmgr_ipcp_connect(const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr); 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_alloc(enum comp_id id, + const char * dst, + qosspec_t * qs, + const struct poa_addr * addr, + struct conn * conn); int connmgr_dealloc(enum comp_id id, struct conn * conn); diff --git a/src/ipcpd/config.h.in b/src/ipcpd/config.h.in index 517016cc..55b0120c 100644 --- a/src/ipcpd/config.h.in +++ b/src/ipcpd/config.h.in @@ -25,6 +25,7 @@ #define SYS_MAX_FLOWS @SYS_MAX_FLOWS@ #define PROC_RES_FDS @PROC_RES_FDS@ #define PROC_MAX_FLOWS @PROC_MAX_FLOWS@ +#define POA_MAX_POAS @POA_MAX_POAS@ #define SOCKET_TIMEOUT @SOCKET_TIMEOUT@ #define CONNECT_TIMEOUT @CONNECT_TIMEOUT@ @@ -54,7 +55,6 @@ #cmakedefine DISABLE_CORE_LOCK #cmakedefine BUILD_CONTAINER #cmakedefine IPCP_FLOW_STATS -#cmakedefine IPCP_ETH_FLOW_STATS #cmakedefine IPCP_DEBUG_LOCAL #ifdef CONFIG_OUROBOROS_DEBUG #cmakedefine DEBUG_PROTO_DHT @@ -62,29 +62,6 @@ #cmakedefine DEBUG_PROTO_LS #endif -/* udp */ -#cmakedefine HAVE_DDNS -#define NSUPDATE_EXEC "@NSUPDATE_EXECUTABLE@" -#define NSLOOKUP_EXEC "@NSLOOKUP_EXECUTABLE@" -#define IPCP_UDP_RD_THR @IPCP_UDP_RD_THR@ -#define IPCP_UDP_WR_THR @IPCP_UDP_WR_THR@ -#define IPCP_UDP_MPL @IPCP_UDP_MPL@ -#define IPCP_UDP4_MTU @IPCP_UDP4_MTU@ -#define IPCP_UDP6_MTU @IPCP_UDP6_MTU@ - -/* eth */ -#cmakedefine HAVE_NETMAP -#cmakedefine HAVE_BPF -#cmakedefine HAVE_RAW_SOCKETS -#cmakedefine IPCP_ETH_QDISC_BYPASS -#define IPCP_ETH_RD_THR @IPCP_ETH_RD_THR@ -#define IPCP_ETH_WR_THR @IPCP_ETH_WR_THR@ -#define IPCP_ETH_LO_MTU @IPCP_ETH_LO_MTU@ -#define IPCP_ETH_MGMT_FRAME_SIZE @IPCP_ETH_MGMT_FRAME_SIZE@ -#define IPCP_ETH_MPL @IPCP_ETH_MPL@ -#define IPCP_ETH_SNDBUF @IPCP_ETH_SNDBUF@ -#define IPCP_ETH_RCVBUF @IPCP_ETH_RCVBUF@ - /* local */ #define IPCP_LOCAL_MPL @IPCP_LOCAL_MPL@ #define IPCP_LOCAL_MTU @IPCP_LOCAL_MTU@ diff --git a/src/ipcpd/eth/CMakeLists.txt b/src/ipcpd/eth/CMakeLists.txt deleted file mode 100644 index 5a36352d..00000000 --- a/src/ipcpd/eth/CMakeLists.txt +++ /dev/null @@ -1,21 +0,0 @@ -# Ethernet IPCPs build configuration (LLC and DIX) -# HAVE_ETH detection is in cmake/dependencies.cmake - -add_executable(${IPCP_ETH_LLC_TARGET} llc.c ${IPCP_SOURCES}) -add_executable(${IPCP_ETH_DIX_TARGET} dix.c ${IPCP_SOURCES}) - -foreach(target ${IPCP_ETH_LLC_TARGET} ${IPCP_ETH_DIX_TARGET}) - target_include_directories(${target} PRIVATE ${IPCP_INCLUDE_DIRS}) - if(HAVE_BPF AND NOT APPLE) - target_include_directories(${target} PRIVATE ${BPF_C_INCLUDE_DIR}) - endif() - if(HAVE_NETMAP AND NOT APPLE) - target_compile_options(${target} PRIVATE -std=c99) - target_include_directories(${target} PRIVATE ${NETMAP_C_INCLUDE_DIR}) - endif() - target_link_libraries(${target} PRIVATE ouroboros-dev) - ouroboros_target_debug_definitions(${target}) -endforeach() - -install(TARGETS ${IPCP_ETH_LLC_TARGET} ${IPCP_ETH_DIX_TARGET} - RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/eth/dix.c b/src/ipcpd/eth/dix.c deleted file mode 100644 index cf8253bd..00000000 --- a/src/ipcpd/eth/dix.c +++ /dev/null @@ -1,26 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC processes over Ethernet - DIX - * - * 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/. - */ - -#define BUILD_ETH_DIX -#define OUROBOROS_PREFIX "ipcpd/eth-dix" - -#include "eth.c" diff --git a/src/ipcpd/eth/eth.c b/src/ipcpd/eth/eth.c deleted file mode 100644 index 2d3bd249..00000000 --- a/src/ipcpd/eth/eth.c +++ /dev/null @@ -1,2468 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC processes over Ethernet - * - * 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(BUILD_ETH_DIX) && !defined(BUILD_ETH_LLC) -#error Define BUILD_ETH_DIX or BUILD_ETH_LLC to build an Ethernet IPCP -#endif - -#if defined(__APPLE__) -#define _BSD_SOURCE -#define _DARWIN_C_SOURCE -#elif defined(__FreeBSD__) -#define __BSD_VISIBLE 1 -#elif defined (__linux__) || defined (__CYGWIN__) -#define _DEFAULT_SOURCE -#else -#define _POSIX_C_SOURCE 200112L -#endif - -#include "config.h" - -#include <ouroboros/atomics.h> -#include <ouroboros/endian.h> -#include <ouroboros/hash.h> -#include <ouroboros/errno.h> -#include <ouroboros/list.h> -#include <ouroboros/utils.h> -#include <ouroboros/bitmap.h> -#include <ouroboros/crc8.h> -#include <ouroboros/dev.h> -#include <ouroboros/ipcp-dev.h> -#include <ouroboros/fqueue.h> -#include <ouroboros/logs.h> -#include <ouroboros/np1_flow.h> -#include <ouroboros/time.h> -#include <ouroboros/fccntl.h> -#include <ouroboros/pthread.h> -#include <ouroboros/rib.h> - -#ifndef IPCP_ETH_FLOW_STATS -#undef FETCH_ADD_RELAXED -#define FETCH_ADD_RELAXED(p, v) ((void) 0) -#undef FETCH_SUB_RELAXED -#define FETCH_SUB_RELAXED(p, v) ((void) 0) -#endif - -#include "ipcp.h" -#include "np1.h" -#include "shim-data.h" - -#include <signal.h> -#include <stdlib.h> -#include <fcntl.h> -#include <unistd.h> -#include <string.h> -#include <sys/socket.h> -#include <sys/types.h> -#include <sys/ioctl.h> - -#include <net/if.h> -#include <netinet/in.h> - -#ifdef __linux__ -#include <linux/if_packet.h> -#include <linux/if_ether.h> -#include <linux/netlink.h> -#include <linux/rtnetlink.h> -#include <linux/sockios.h> -#endif - -#ifdef __FreeBSD__ -#include <net/if_dl.h> -#include <netinet/if_ether.h> -#include <ifaddrs.h> -#endif - -#ifdef __APPLE__ -#include <net/if_dl.h> -#include <ifaddrs.h> -#endif - -#include <poll.h> -#include <sys/mman.h> - -#if defined(HAVE_NETMAP) - #define NETMAP_WITH_LIBS - #include <net/netmap_user.h> -#elif defined(HAVE_BPF) - #define BPF_DEV_MAX 256 - #define BPF_BLEN sysconf(_SC_PAGESIZE) - #include <net/bpf.h> -#endif - -#define MAC_FMT "%02x:%02x:%02x:%02x:%02x:%02x" -#define MAC_VAL(a) \ - (uint8_t)(a)[0], (uint8_t)(a)[1], (uint8_t)(a)[2], \ - (uint8_t)(a)[3], (uint8_t)(a)[4], (uint8_t)(a)[5] - - -#ifndef ETH_MAX_MTU /* In if_ether.h as of Linux 4.10. */ - #define ETH_MAX_MTU 0xFFFFU -#endif /* ETH_MAX_MTU */ -#ifdef BUILD_ETH_DIX - #define ETH_MTU eth_data.mtu - #define ETH_MTU_MAX ETH_MAX_MTU -#else - #define ETH_MTU eth_data.mtu - #define ETH_MTU_MAX 1500 -#endif /* BUILD_ETH_DIX */ - -#define ETH_TYPE_LENGTH_SIZE sizeof(uint16_t) -#define ETH_HEADER_SIZE (2 * MAC_SIZE + ETH_TYPE_LENGTH_SIZE) - -#if defined(BUILD_ETH_DIX) -#define THIS_TYPE IPCP_ETH_DIX -#define MGMT_EID 0 -#define DIX_EID_SIZE sizeof(uint16_t) -#define DIX_LENGTH_SIZE sizeof(uint16_t) -#define DIX_HCS_SIZE CRC8_HASH_LEN -#define DIX_HEADER_SIZE (DIX_EID_SIZE + DIX_LENGTH_SIZE + DIX_HCS_SIZE) -#define ETH_HEADER_TOT_SIZE (ETH_HEADER_SIZE + DIX_HEADER_SIZE) -#define MAX_EIDS (1 << (8 * DIX_EID_SIZE)) -#define ETH_MAX_PACKET_SIZE (ETH_MTU - DIX_HEADER_SIZE) -#define ETH_FRAME_SIZE (ETH_HEADER_SIZE + ETH_MTU_MAX) -#elif defined(BUILD_ETH_LLC) -#define THIS_TYPE IPCP_ETH_LLC -#define MGMT_SAP 0x01 -#define LLC_FIELDS_SIZE 3 -#define LLC_HCS_SIZE CRC8_HASH_LEN -#define LLC_HEADER_SIZE (LLC_FIELDS_SIZE + LLC_HCS_SIZE) -#define ETH_HEADER_TOT_SIZE (ETH_HEADER_SIZE + LLC_HEADER_SIZE) -#define MAX_SAPS 64 -#define ETH_MAX_PACKET_SIZE (ETH_MTU - LLC_HEADER_SIZE) -#define ETH_FRAME_SIZE (ETH_HEADER_SIZE + ETH_MTU_MAX) -#endif - -#define NAME_QUERY_TIMEO 1900 /* ms total budget */ -#define NAME_QUERY_RETRIES 3 /* retransmits, 4 attempts total */ -#define MGMT_TIMEO 100 /* ms */ -#define MGMT_FRAME_SIZE IPCP_ETH_MGMT_FRAME_SIZE -#define ETH_RIB_PATH "eth" - -#define FLOW_REQ 0 -#define FLOW_REPLY 1 -#define NAME_QUERY_REQ 2 -#define NAME_QUERY_REPLY 3 -#define FLOW_IRM_UPDATE 4 - -struct mgmt_msg { -#if defined(BUILD_ETH_DIX) - uint16_t seid; - uint16_t deid; -#elif defined(BUILD_ETH_LLC) - uint8_t ssap; - uint8_t dsap; - /* QoS here for alignment */ - uint8_t code; - uint8_t availability; -#endif - /* QoS parameters from spec, aligned */ - uint32_t loss; - uint64_t bandwidth; - uint32_t ber; - uint32_t max_gap; - uint32_t delay; - uint32_t timeout; - int32_t response; - uint8_t service; -#if defined (BUILD_ETH_DIX) - uint8_t code; - uint8_t availability; -#endif -} __attribute__((packed)); - -struct eth_frame { - uint8_t dst_hwaddr[MAC_SIZE]; - uint8_t src_hwaddr[MAC_SIZE]; -#if defined(BUILD_ETH_DIX) - uint16_t ethertype; - uint16_t eid; - uint16_t length; -#elif defined(BUILD_ETH_LLC) - uint16_t length; - uint8_t dsap; - uint8_t ssap; - uint8_t cf; -#endif - uint8_t hcs; - uint8_t payload; -} __attribute__((packed)); - -struct ef { -#if defined(BUILD_ETH_DIX) - int32_t r_eid; -#elif defined(BUILD_ETH_LLC) - int8_t sap; - int8_t r_sap; -#endif - uint8_t r_addr[MAC_SIZE]; -#ifdef IPCP_ETH_FLOW_STATS - struct { - time_t stamp; - size_t p_rcv; - size_t b_rcv; - size_t p_dlv_f; - size_t p_snd; - size_t b_snd; - size_t p_snd_f; - } stat; -#endif -}; - -struct mgmt_frame { - struct list_head next; - uint8_t r_addr[MAC_SIZE]; - uint8_t buf[MGMT_FRAME_SIZE]; - size_t len; -}; - -struct { - struct shim_data * shim_data; - - int mtu; -#ifdef __linux__ - int if_idx; -#endif -#if defined(HAVE_NETMAP) - struct nm_desc * nmd; - uint8_t hw_addr[MAC_SIZE]; - struct pollfd poll_in; - struct pollfd poll_out; -#elif defined(HAVE_BPF) - int bpf; - uint8_t hw_addr[MAC_SIZE]; -#elif defined(HAVE_RAW_SOCKETS) - int s_fd; - struct sockaddr_ll device; -#endif /* HAVE_NETMAP */ -#if defined (BUILD_ETH_DIX) - uint16_t ethertype; -#elif defined(BUILD_ETH_LLC) - struct bmp * saps; - int * ef_to_fd; -#endif - struct ef * fd_to_ef; - fset_t * np1_flows; - pthread_rwlock_t flows_lock; -#ifdef IPCP_ETH_FLOW_STATS - struct { - size_t n_flows; - size_t n_rcv; - size_t n_snd; - size_t n_mgmt_rcv; - size_t n_mgmt_snd; - size_t n_bad_id; - size_t n_dlv_f; - size_t n_buf_f; - size_t n_rcv_f; - size_t n_snd_f; - size_t kern_rcv; - size_t kern_drp; - } stat; -#endif - - pthread_t packet_writer[IPCP_ETH_WR_THR]; - pthread_t packet_reader[IPCP_ETH_RD_THR]; - -#ifdef __linux__ - pthread_t if_monitor; -#endif - - /* Handle mgmt frames in a different thread */ - pthread_t mgmt_handler; - pthread_mutex_t mgmt_lock; - pthread_cond_t mgmt_cond; - struct list_head mgmt_frames; -} eth_data; - -static int eth_data_init(void) -{ - int i; - int ret = -ENOMEM; - pthread_condattr_t cattr; - - eth_data.fd_to_ef = - malloc(sizeof(*eth_data.fd_to_ef) * SYS_MAX_FLOWS); - if (eth_data.fd_to_ef == NULL) - goto fail_fd_to_ef; - -#ifdef BUILD_ETH_LLC - eth_data.ef_to_fd = - malloc(sizeof(*eth_data.ef_to_fd) * MAX_SAPS); - if (eth_data.ef_to_fd == NULL) - goto fail_ef_to_fd; - - for (i = 0; i < MAX_SAPS; ++i) - eth_data.ef_to_fd[i] = -1; - - eth_data.saps = bmp_create(MAX_SAPS, 2); - if (eth_data.saps == NULL) - goto fail_saps; -#endif - eth_data.np1_flows = fset_create(); - if (eth_data.np1_flows == NULL) - goto fail_np1_flows; - - for (i = 0; i < SYS_MAX_FLOWS; ++i) { -#if defined(BUILD_ETH_DIX) - eth_data.fd_to_ef[i].r_eid = -1; -#elif defined(BUILD_ETH_LLC) - eth_data.fd_to_ef[i].sap = -1; - eth_data.fd_to_ef[i].r_sap = -1; -#endif - memset(ð_data.fd_to_ef[i].r_addr, 0, MAC_SIZE); -#ifdef IPCP_ETH_FLOW_STATS - memset(ð_data.fd_to_ef[i].stat, 0, - sizeof(eth_data.fd_to_ef[i].stat)); -#endif - } -#ifdef IPCP_ETH_FLOW_STATS - memset(ð_data.stat, 0, sizeof(eth_data.stat)); -#endif - - eth_data.shim_data = shim_data_create(); - if (eth_data.shim_data == NULL) - goto fail_shim_data; - - ret = -1; - - if (pthread_rwlock_init(ð_data.flows_lock, NULL)) - goto fail_flows_lock; - - if (pthread_mutex_init(ð_data.mgmt_lock, NULL)) - goto fail_mgmt_lock; - - if (pthread_condattr_init(&cattr)) - goto fail_condattr; - -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - - if (pthread_cond_init(ð_data.mgmt_cond, &cattr)) - goto fail_mgmt_cond; - - pthread_condattr_destroy(&cattr); - - list_head_init(ð_data.mgmt_frames); - - return 0; - - fail_mgmt_cond: - pthread_condattr_destroy(&cattr); - fail_condattr: - pthread_mutex_destroy(ð_data.mgmt_lock); - fail_mgmt_lock: - pthread_rwlock_destroy(ð_data.flows_lock); - fail_flows_lock: - shim_data_destroy(eth_data.shim_data); - fail_shim_data: - fset_destroy(eth_data.np1_flows); - fail_np1_flows: -#ifdef BUILD_ETH_LLC - bmp_destroy(eth_data.saps); - fail_saps: - free(eth_data.ef_to_fd); - fail_ef_to_fd: -#endif - free(eth_data.fd_to_ef); - fail_fd_to_ef: - return ret; -} - -static void eth_data_fini(void) -{ -#if defined(HAVE_NETMAP) - nm_close(eth_data.nmd); -#elif defined(HAVE_BPF) - close(eth_data.bpf); -#elif defined(HAVE_RAW_SOCKETS) - close(eth_data.s_fd); -#endif - pthread_cond_destroy(ð_data.mgmt_cond); - pthread_mutex_destroy(ð_data.mgmt_lock); - pthread_rwlock_destroy(ð_data.flows_lock); - shim_data_destroy(eth_data.shim_data); - fset_destroy(eth_data.np1_flows); -#ifdef BUILD_ETH_LLC - bmp_destroy(eth_data.saps); - free(eth_data.ef_to_fd); -#endif - free(eth_data.fd_to_ef); -} - -#ifdef IPCP_ETH_FLOW_STATS -static int eth_rib_read(const char * path, - char * buf, - size_t len) -{ - struct ef * flow; - int fd; - char tmstr[RIB_TM_STRLEN]; - struct tm * tm; - time_t stamp; - char * entry; - - entry = strstr(path, RIB_SEPARATOR) + 1; - assert(entry); - - if (len < 2048) - return 0; - - buf[0] = '\0'; - - if (strcmp(entry, "summary") == 0) { - int n; -#if defined(HAVE_RAW_SOCKETS) - int rcvbuf = 0; - int sndbuf = 0; - int queued = 0; - socklen_t optlen = sizeof(rcvbuf); -# if defined(__linux__) - int outq = 0; - struct tpacket_stats tp_stats; - socklen_t tp_len = sizeof(tp_stats); -# endif - - getsockopt(eth_data.s_fd, SOL_SOCKET, - SO_RCVBUF, &rcvbuf, &optlen); - optlen = sizeof(sndbuf); - getsockopt(eth_data.s_fd, SOL_SOCKET, - SO_SNDBUF, &sndbuf, &optlen); - ioctl(eth_data.s_fd, FIONREAD, &queued); -# if defined(__linux__) - ioctl(eth_data.s_fd, SIOCOUTQ, &outq); - if (getsockopt(eth_data.s_fd, SOL_PACKET, - PACKET_STATISTICS, - &tp_stats, &tp_len) == 0) { - FETCH_ADD_RELAXED(ð_data.stat.kern_rcv, - tp_stats.tp_packets); - FETCH_ADD_RELAXED(ð_data.stat.kern_drp, - tp_stats.tp_drops); - } -# endif -#endif - n = sprintf(buf, - "Active flows: %20zu\n" - "Total frames received: %20zu\n" - "Total frames sent: %20zu\n" - "Management frames received: %20zu\n" - "Management frames sent: %20zu\n" - "Bad EID/SAP frames: %20zu\n" - "Delivery (N+1) failures: %20zu\n" - "Buffer alloc failures: %20zu\n" - "Frame read failures: %20zu\n" - "Frame send failures: %20zu\n", - LOAD_RELAXED(ð_data.stat.n_flows), - LOAD_RELAXED(ð_data.stat.n_rcv), - LOAD_RELAXED(ð_data.stat.n_snd), - LOAD_RELAXED(ð_data.stat.n_mgmt_rcv), - LOAD_RELAXED(ð_data.stat.n_mgmt_snd), - LOAD_RELAXED(ð_data.stat.n_bad_id), - LOAD_RELAXED(ð_data.stat.n_dlv_f), - LOAD_RELAXED(ð_data.stat.n_buf_f), - LOAD_RELAXED(ð_data.stat.n_rcv_f), - LOAD_RELAXED(ð_data.stat.n_snd_f)); -#if defined(HAVE_RAW_SOCKETS) - n += sprintf(buf + n, - "Socket rcvbuf (bytes): %20d\n" - "Socket sndbuf (bytes): %20d\n" - "Socket ingress (bytes): %20d\n", - rcvbuf, sndbuf, queued); -# if defined(__linux__) - n += sprintf(buf + n, - "Socket egress (bytes): %20d\n" - "Kernel frames received: %20zu\n" - "Kernel frames dropped: %20zu\n", - outq, - LOAD_RELAXED(ð_data.stat.kern_rcv), - LOAD_RELAXED(ð_data.stat.kern_drp)); -# endif -#endif - return n; - } - - fd = atoi(entry); - - if (fd < 0 || fd >= SYS_MAX_FLOWS) - return -1; - - flow = ð_data.fd_to_ef[fd]; - - pthread_rwlock_rdlock(ð_data.flows_lock); - - stamp = flow->stat.stamp; - if (stamp == 0) { - pthread_rwlock_unlock(ð_data.flows_lock); - return 0; - } - - pthread_rwlock_unlock(ð_data.flows_lock); - - tm = gmtime(&stamp); - strftime(tmstr, sizeof(tmstr), RIB_TM_FORMAT, tm); - - sprintf(buf, - "Flow established at: %20s\n" - "Sent (packets): %20zu\n" - "Sent (bytes): %20zu\n" - "Send failed (packets): %20zu\n" - "Received (packets): %20zu\n" - "Received (bytes): %20zu\n" - "Delivery (N+1) failures: %20zu\n", - tmstr, - LOAD_RELAXED(&flow->stat.p_snd), - LOAD_RELAXED(&flow->stat.b_snd), - LOAD_RELAXED(&flow->stat.p_snd_f), - LOAD_RELAXED(&flow->stat.p_rcv), - LOAD_RELAXED(&flow->stat.b_rcv), - LOAD_RELAXED(&flow->stat.p_dlv_f)); - - return strlen(buf); -} - -static int eth_rib_readdir(char *** buf) -{ - char entry[RIB_PATH_LEN + 1]; - size_t i; - int idx = 0; - int n_entries; - - pthread_rwlock_rdlock(ð_data.flows_lock); - - n_entries = (int) LOAD_RELAXED(ð_data.stat.n_flows) + 1; - - *buf = malloc(sizeof(**buf) * n_entries); - if (*buf == NULL) - goto fail_entries; - - (*buf)[idx] = malloc(strlen("summary") + 1); - if ((*buf)[idx] == NULL) - goto fail_entry; - - strcpy((*buf)[idx++], "summary"); - - for (i = 0; i < SYS_MAX_FLOWS && idx < n_entries; ++i) { - if (eth_data.fd_to_ef[i].stat.stamp == 0) - continue; - - sprintf(entry, "%zu", i); - - (*buf)[idx] = malloc(strlen(entry) + 1); - if ((*buf)[idx] == NULL) - goto fail_entry; - - strcpy((*buf)[idx++], entry); - } - - pthread_rwlock_unlock(ð_data.flows_lock); - - return idx; - - fail_entry: - while (idx-- > 0) - free((*buf)[idx]); - - free(*buf); - fail_entries: - pthread_rwlock_unlock(ð_data.flows_lock); - return -ENOMEM; -} - -static int eth_rib_getattr(const char * path, - struct rib_attr * attr) -{ - int fd; - char * entry; - struct ef * flow; - - entry = strstr(path, RIB_SEPARATOR) + 1; - assert(entry); - - if (strcmp(entry, "summary") == 0) { - attr->size = 2048; - attr->mtime = 0; - return 0; - } - - fd = atoi(entry); - - if (fd < 0 || fd >= SYS_MAX_FLOWS) { - attr->size = 0; - attr->mtime = 0; - return 0; - } - - flow = ð_data.fd_to_ef[fd]; - - pthread_rwlock_rdlock(ð_data.flows_lock); - - if (flow->stat.stamp != 0) { - attr->size = 2048; - attr->mtime = flow->stat.stamp; - } else { - attr->size = 0; - attr->mtime = 0; - } - - pthread_rwlock_unlock(ð_data.flows_lock); - - return 0; -} - -static struct rib_ops eth_r_ops = { - .read = eth_rib_read, - .readdir = eth_rib_readdir, - .getattr = eth_rib_getattr -}; -#endif /* IPCP_ETH_FLOW_STATS */ - -#ifdef BUILD_ETH_LLC -static uint8_t reverse_bits(uint8_t b) -{ - b = (b & 0xF0) >> 4 | (b & 0x0F) << 4; - b = (b & 0xCC) >> 2 | (b & 0x33) << 2; - b = (b & 0xAA) >> 1 | (b & 0x55) << 1; - - return b; -} -#endif - -/* Pass a buffer that contains space for the header. */ -static int eth_ipcp_send_frame(const uint8_t * dst_addr, -#if defined(BUILD_ETH_DIX) - uint16_t deid, -#elif defined(BUILD_ETH_LLC) - uint8_t dsap, - uint8_t ssap, -#endif - const uint8_t * frame, - size_t len) -{ - uint32_t frame_len = 0; -#ifdef BUILD_ETH_LLC - uint8_t cf = 0x03; -#endif - struct eth_frame * e_frame; -#ifdef HAVE_RAW_SOCKETS - fd_set fds; - - FD_ZERO(&fds); -#endif - - assert(frame); - - if (len > (size_t) ETH_MAX_PACKET_SIZE) - return -1; - - e_frame = (struct eth_frame *) frame; - - memcpy(e_frame->dst_hwaddr, dst_addr, MAC_SIZE); - memcpy(e_frame->src_hwaddr, -#if defined(HAVE_NETMAP) || defined(HAVE_BPF) - eth_data.hw_addr, -#elif defined(HAVE_RAW_SOCKETS) - eth_data.device.sll_addr, -#endif /* HAVE_NETMAP */ - MAC_SIZE); -#if defined(BUILD_ETH_DIX) - e_frame->ethertype = eth_data.ethertype; - e_frame->eid = htons(deid); - e_frame->length = htons(len); - mem_hash(HASH_CRC8, &e_frame->hcs, - (uint8_t *) &e_frame->eid, - DIX_EID_SIZE + DIX_LENGTH_SIZE); - frame_len = ETH_HEADER_TOT_SIZE + len; -#elif defined(BUILD_ETH_LLC) - e_frame->length = htons(LLC_HEADER_SIZE + len); - e_frame->dsap = dsap; - e_frame->ssap = ssap; - e_frame->cf = cf; - mem_hash(HASH_CRC8, &e_frame->hcs, - (uint8_t *) &e_frame->dsap, - LLC_FIELDS_SIZE); - frame_len = ETH_HEADER_TOT_SIZE + len; -#endif - -#if defined(HAVE_NETMAP) - if (poll(ð_data.poll_out, 1, -1) < 0) - return -1; - - if (nm_inject(eth_data.nmd, frame, frame_len) != (int) frame_len) { - log_dbg("Failed to send message."); - return -1; - } -#elif defined(HAVE_BPF) - if (write(eth_data.bpf, frame, frame_len) < 0) { - log_dbg("Failed to send message."); - return -1; - } - -#elif defined(HAVE_RAW_SOCKETS) - FD_SET(eth_data.s_fd, &fds); - if (select(eth_data.s_fd + 1, NULL, &fds, NULL, NULL) < 0) { - log_dbg("Select() failed: %s.", strerror(errno)); - return -1; - } - assert(FD_ISSET(eth_data.s_fd, &fds)); - - if (sendto(eth_data.s_fd, frame, frame_len, 0, - (struct sockaddr *) ð_data.device, - sizeof(eth_data.device)) <= 0) { - log_dbg("Failed to send message: %s.", strerror(errno)); - return -1; - } -#endif /* HAVE_NETMAP */ - - FETCH_ADD_RELAXED(ð_data.stat.n_snd, 1); - - return 0; -} - -static int eth_ipcp_alloc(const uint8_t * dst_addr, -#if defined(BUILD_ETH_DIX) - uint16_t eid, -#elif defined(BUILD_ETH_LLC) - uint8_t ssap, -#endif - const uint8_t * hash, - qosspec_t qs, - const buffer_t * data) -{ - uint8_t * buf; - struct mgmt_msg * msg; - size_t len; - int ret; - - len = sizeof(*msg) + ipcp_dir_hash_len(); - - buf = malloc(len + ETH_HEADER_TOT_SIZE + data->len); - if (buf == NULL) - return -1; - - memset(buf, 0, len + ETH_HEADER_TOT_SIZE + data->len); - - msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); - msg->code = FLOW_REQ; -#if defined(BUILD_ETH_DIX) - msg->seid = htons(eid); -#elif defined(BUILD_ETH_LLC) - msg->ssap = ssap; -#endif - - 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->service = qs.service; - msg->max_gap = hton32(qs.max_gap); - msg->timeout = hton32(qs.timeout); - - memcpy(msg + 1, hash, ipcp_dir_hash_len()); - if (data->len > 0) - memcpy(buf + len + ETH_HEADER_TOT_SIZE, data->data, data->len); - - ret = eth_ipcp_send_frame(dst_addr, -#if defined(BUILD_ETH_DIX) - MGMT_EID, -#elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), -#endif - buf, len + data->len); - free(buf); - - if (ret == 0) - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); - - return ret; -} - -static int eth_ipcp_alloc_resp(uint8_t * dst_addr, -#if defined(BUILD_ETH_DIX) - uint16_t seid, - uint16_t deid, -#elif defined(BUILD_ETH_LLC) - uint8_t ssap, - uint8_t dsap, -#endif - int response, - const buffer_t * data) -{ - struct mgmt_msg * msg; - uint8_t * buf; - - buf = malloc(sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); - if (buf == NULL) - return -1; - - memset(buf, 0, sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); - - msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); - - msg->code = FLOW_REPLY; -#if defined(BUILD_ETH_DIX) - msg->seid = htons(seid); - msg->deid = htons(deid); -#elif defined(BUILD_ETH_LLC) - msg->ssap = ssap; - msg->dsap = dsap; -#endif - msg->response = hton32(response); - - if (data->len > 0) - memcpy(msg + 1, data->data, data->len); - - if (eth_ipcp_send_frame(dst_addr, -#if defined(BUILD_ETH_DIX) - MGMT_EID, -#elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), -#endif - buf, sizeof(*msg) + data->len)) { - free(buf); - return -1; - } - - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); - - free(buf); - - return 0; -} - -static int eth_ipcp_flow_update(int fd, - const buffer_t * data) -{ - struct mgmt_msg * msg; - struct ef * flow; - uint8_t * buf; - uint8_t r_addr[MAC_SIZE]; - int ret; - - buf = malloc(sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); - if (buf == NULL) - return -1; - - memset(buf, 0, sizeof(*msg) + ETH_HEADER_TOT_SIZE + data->len); - - msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); - - msg->code = FLOW_IRM_UPDATE; - - pthread_rwlock_rdlock(ð_data.flows_lock); - - flow = ð_data.fd_to_ef[fd]; -#if defined(BUILD_ETH_DIX) - msg->seid = htons((uint16_t) fd); - msg->deid = htons((uint16_t) flow->r_eid); -#elif defined(BUILD_ETH_LLC) - msg->ssap = flow->sap; - msg->dsap = (uint8_t) flow->r_sap; -#endif - memcpy(r_addr, flow->r_addr, MAC_SIZE); - - pthread_rwlock_unlock(ð_data.flows_lock); - - if (data->len > 0) - memcpy(msg + 1, data->data, data->len); - - ret = eth_ipcp_send_frame(r_addr, -#if defined(BUILD_ETH_DIX) - MGMT_EID, -#elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), -#endif - buf, sizeof(*msg) + data->len); - free(buf); - - if (ret == 0) - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); - - return ret; -} - -static int eth_ipcp_req(uint8_t * r_addr, -#if defined(BUILD_ETH_DIX) - uint16_t r_eid, -#elif defined(BUILD_ETH_LLC) - uint8_t r_sap, -#endif - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) -{ - int fd; - - fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_ETH_MPL, - ETH_MAX_PACKET_SIZE, data); - if (fd < 0) { - log_err("Could not get new flow from IRMd."); - return -1; - } - - pthread_rwlock_wrlock(ð_data.flows_lock); -#if defined(BUILD_ETH_DIX) - eth_data.fd_to_ef[fd].r_eid = r_eid; -#elif defined(BUILD_ETH_LLC) - eth_data.fd_to_ef[fd].r_sap = r_sap; -#endif - memcpy(eth_data.fd_to_ef[fd].r_addr, r_addr, MAC_SIZE); - - pthread_rwlock_unlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - log_dbg("New flow request, fd %d, remote endpoint %d.", fd, r_eid); -#elif defined(BUILD_ETH_LLC) - log_dbg("New flow request, fd %d, remote SAP %d.", fd, r_sap); -#endif - return 0; -} - -static int eth_ipcp_alloc_reply(uint8_t * r_addr, -#if defined(BUILD_ETH_DIX) - uint16_t seid, - uint16_t deid, -#elif defined(BUILD_ETH_LLC) - uint8_t ssap, - int dsap, -#endif - int response, - const buffer_t * data) -{ - int ret = 0; - int fd = -1; - time_t mpl = IPCP_ETH_MPL; - - pthread_rwlock_wrlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - fd = deid; -#elif defined(BUILD_ETH_LLC) - fd = eth_data.ef_to_fd[dsap]; -#endif - if (fd < 0) { - pthread_rwlock_unlock(ð_data.flows_lock); - log_err("No flow found with that SAP."); - return -1; /* -EFLOWNOTFOUND */ - } - - if (response) { -#ifdef BUILD_ETH_LLC - bmp_release(eth_data.saps, eth_data.fd_to_ef[fd].sap); -#endif - } else { -#if defined(BUILD_ETH_DIX) - eth_data.fd_to_ef[fd].r_eid = seid; -#elif defined(BUILD_ETH_LLC) - eth_data.fd_to_ef[fd].r_sap = ssap; -#endif - memcpy(eth_data.fd_to_ef[fd].r_addr, r_addr, MAC_SIZE); - } - - pthread_rwlock_unlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - log_dbg("Flow reply, fd %d, src eid %d, dst eid %d.", fd, seid, deid); -#elif defined(BUILD_ETH_LLC) - log_dbg("Flow reply, fd %d, SSAP %d, DSAP %d.", fd, ssap, dsap); -#endif - if ((ret = ipcp_flow_alloc_reply(fd, response, mpl, - ETH_MAX_PACKET_SIZE, data)) < 0) { - log_err("Failed to reply to flow allocation."); - return -1; - } - - return ret; -} - -static int eth_ipcp_name_query_req(const uint8_t * hash, - uint8_t * r_addr) -{ - uint8_t * buf; - struct mgmt_msg * msg; - size_t len; - - if (shim_data_reg_has(eth_data.shim_data, hash)) { - len = sizeof(*msg) + ipcp_dir_hash_len(); - - buf = malloc(len + ETH_HEADER_TOT_SIZE); - if (buf == NULL) - return -1; - - memset(buf, 0, len + ETH_HEADER_TOT_SIZE); - - msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); - msg->code = NAME_QUERY_REPLY; - - memcpy(msg + 1, hash, ipcp_dir_hash_len()); - - if (eth_ipcp_send_frame(r_addr, -#if defined(BUILD_ETH_DIX) - MGMT_EID, -#elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), -#endif - buf, len)) { - log_err("Failed to send management frame."); - free(buf); - return -1; - } - - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); - - free(buf); - } - - return 0; -} - -static int eth_ipcp_name_query_reply(const uint8_t * hash, - uint8_t * r_addr) -{ - struct addr addr; - - memcpy(&addr.mac, r_addr, MAC_SIZE); - - shim_data_dir_add_entry(eth_data.shim_data, hash, addr); - - shim_data_dir_query_respond(eth_data.shim_data, hash); - - return 0; -} - -static int eth_ipcp_flow_update_arr(const uint8_t * buf, - size_t len) -{ - struct mgmt_msg * msg; - buffer_t data; - int fd; - int flow_id; - - msg = (struct mgmt_msg *) buf; - - data.data = (uint8_t *) buf + sizeof(*msg); - data.len = len - sizeof(*msg); - - pthread_rwlock_rdlock(ð_data.flows_lock); -#if defined(BUILD_ETH_DIX) - fd = ntohs(msg->deid); -#elif defined(BUILD_ETH_LLC) - fd = eth_data.ef_to_fd[msg->dsap]; -#endif - pthread_rwlock_unlock(ð_data.flows_lock); - - if (fd < 0 || fd >= SYS_MAX_FLOWS) { - log_err("Flow update for unknown endpoint."); - return -1; - } - - flow_id = np1_flow_id(fd); - if (flow_id < 0) - return -1; - - if (ipcp_flow_update_arr(flow_id, &data) < 0) { - log_err("Failed to relay flow update on fd %d.", fd); - return -1; - } - - return 0; -} - -static int eth_ipcp_mgmt_frame(const uint8_t * buf, - size_t len, - uint8_t * r_addr) -{ - struct mgmt_msg * msg; - size_t msg_len; - qosspec_t qs; - buffer_t data; - - if (len < sizeof(*msg)) - return -1; - - msg = (struct mgmt_msg *) buf; - - switch (msg->code) { - case FLOW_REQ: - msg_len = sizeof(*msg) + ipcp_dir_hash_len(); - - if (len < msg_len) - return -1; - - 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); - - data.data = (uint8_t *) buf + msg_len; - data.len = len - msg_len; - - if (shim_data_reg_has(eth_data.shim_data, - buf + sizeof(*msg))) { - eth_ipcp_req(r_addr, -#if defined(BUILD_ETH_DIX) - ntohs(msg->seid), -#elif defined(BUILD_ETH_LLC) - msg->ssap, -#endif - buf + sizeof(*msg), - qs, - &data); - } - break; - case FLOW_REPLY: - data.data = (uint8_t *) buf + sizeof(*msg); - data.len = len - sizeof(*msg); - - eth_ipcp_alloc_reply(r_addr, -#if defined(BUILD_ETH_DIX) - ntohs(msg->seid), - ntohs(msg->deid), -#elif defined(BUILD_ETH_LLC) - msg->ssap, - msg->dsap, -#endif - ntoh32(msg->response), - &data); - break; - case FLOW_IRM_UPDATE: - eth_ipcp_flow_update_arr(buf, len); - break; - case NAME_QUERY_REQ: - if (len < sizeof(*msg) + ipcp_dir_hash_len()) - return -1; - eth_ipcp_name_query_req(buf + sizeof(*msg), r_addr); - break; - case NAME_QUERY_REPLY: - if (len < sizeof(*msg) + ipcp_dir_hash_len()) - return -1; - eth_ipcp_name_query_reply(buf + sizeof(*msg), r_addr); - break; - default: - log_err("Unknown message received %d.", msg->code); - return -1; - } - - return 0; -} - -static void * eth_ipcp_mgmt_handler(void * o) -{ - (void) o; - - pthread_cleanup_push(__cleanup_mutex_unlock, ð_data.mgmt_lock); - - while (true) { - int ret = 0; - struct timespec timeout = TIMESPEC_INIT_MS(MGMT_TIMEO); - struct timespec abstime; - struct mgmt_frame * frame = NULL; - - clock_gettime(PTHREAD_COND_CLOCK, &abstime); - ts_add(&abstime, &timeout, &abstime); - - pthread_mutex_lock(ð_data.mgmt_lock); - - while (list_is_empty(ð_data.mgmt_frames) && - ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(ð_data.mgmt_cond, - ð_data.mgmt_lock, - &abstime); - if (ret != -ETIMEDOUT) - frame = list_first_entry((ð_data.mgmt_frames), - struct mgmt_frame, next); - if (frame != NULL) - list_del(&frame->next); - - pthread_mutex_unlock(ð_data.mgmt_lock); - - if (frame == NULL) - continue; - - eth_ipcp_mgmt_frame(frame->buf, frame->len, frame->r_addr); - - free(frame); - } - - pthread_cleanup_pop(false); - - return (void *) 0; -} - -static void * eth_ipcp_packet_reader(void * o) -{ - uint8_t br_addr[MAC_SIZE]; -#if defined(BUILD_ETH_DIX) - uint16_t deid; -#elif defined(BUILD_ETH_LLC) - uint8_t dsap; - uint8_t ssap; -#endif - uint16_t length; - int fd; - uint8_t * buf; -#if defined(HAVE_NETMAP) - struct nm_pkthdr hdr; -#else - struct ssm_pk_buff * spb; - fd_set fds; - int frame_len; -#endif -#if defined(HAVE_RAW_SOCKETS) - struct sockaddr_ll src; - socklen_t slen; -#endif - size_t eth_len; - uint8_t hcs; - struct eth_frame * e_frame; - struct mgmt_frame * frame; - - (void) o; - - ipcp_lock_to_core(); - - memset(br_addr, 0xff, MAC_SIZE * sizeof(uint8_t)); - - while (true) { -#if defined(HAVE_NETMAP) - if (poll(ð_data.poll_in, 1, -1) < 0) - continue; - if (eth_data.poll_in.revents == 0) /* TIMED OUT */ - continue; - - buf = nm_nextpkt(eth_data.nmd, &hdr); - if (buf == NULL) { - log_dbg("Bad read from netmap device."); - continue; - } -#else - FD_ZERO(&fds); - #if defined(HAVE_BPF) - FD_SET(eth_data.bpf, &fds); - if (select(eth_data.bpf + 1, &fds, NULL, NULL, NULL)) - continue; - assert(FD_ISSET(eth_data.bpf, &fds)); - if (ipcp_spb_reserve(&spb, BPF_LEN)) - continue; - buf = ssm_pk_buff_head(spb); - frame_len = read(eth_data.bpf, buf, BPF_BLEN); - #elif defined(HAVE_RAW_SOCKETS) - FD_SET(eth_data.s_fd, &fds); - if (select(eth_data.s_fd + 1, &fds, NULL, NULL, NULL) < 0) - continue; - assert(FD_ISSET(eth_data.s_fd, &fds)); - if (ipcp_spb_reserve(&spb, ETH_MTU)) { - FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); - continue; - } - buf = ssm_pk_buff_push(spb, ETH_HEADER_TOT_SIZE); - if (buf == NULL) { - log_dbg("Failed to allocate header."); - ipcp_spb_release(spb); - FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); - continue; - } - slen = sizeof(src); - /* MSG_DONTWAIT: RD_THR>1 race-loser bails with EAGAIN. */ - frame_len = recvfrom(eth_data.s_fd, buf, - ETH_MTU + ETH_HEADER_TOT_SIZE, - MSG_DONTWAIT, - (struct sockaddr *) &src, &slen); - #endif - if (frame_len == 0) { - ipcp_spb_release(spb); - continue; /* Spurious */ - } - - if (frame_len < 0) { - ipcp_spb_release(spb); - - if (errno == EAGAIN || errno == EWOULDBLOCK) - continue; - - log_dbg("Failed to rcv frame: %s.", strerror(errno)); - FETCH_ADD_RELAXED(ð_data.stat.n_rcv_f, 1); - continue; - } -#endif - -#if defined(HAVE_NETMAP) - eth_len = hdr.len; -#elif defined(HAVE_BPF) - eth_len = ((struct bpf_hdr *) buf)->bh_caplen; -#else - eth_len = (size_t) frame_len; -#endif - /* Defense in depth: reject before parsing dereferences. */ - if (eth_len < ETH_HEADER_TOT_SIZE) - goto fail_frame; - -#if defined(HAVE_RAW_SOCKETS) - /* Drop our own egress. */ - if (src.sll_pkttype == PACKET_OUTGOING) - goto fail_frame; -#endif - -#if defined(HAVE_BPF) && !defined(HAVE_NETMAP) - e_frame = (struct eth_frame *) - (buf + ((struct bpf_hdr *) buf)->bh_hdrlen); -#else - e_frame = (struct eth_frame *) buf; -#endif - assert(e_frame->dst_hwaddr); - -#if !defined(HAVE_BPF) - #if defined(HAVE_NETMAP) - if (memcmp(eth_data.hw_addr, - #elif defined(HAVE_RAW_SOCKETS) - if (memcmp(eth_data.device.sll_addr, - #endif /* HAVE_NETMAP */ - e_frame->dst_hwaddr, - MAC_SIZE) && - memcmp(br_addr, e_frame->dst_hwaddr, MAC_SIZE)) { - FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); - goto fail_frame; - } -#endif - length = ntohs(e_frame->length); -#if defined(BUILD_ETH_DIX) - if (e_frame->ethertype != eth_data.ethertype) - goto fail_frame; - - if (length > ETH_MTU) - goto fail_frame; - - deid = ntohs(e_frame->eid); -#elif defined (BUILD_ETH_LLC) - if (length > 0x05FF) /* DIX */ - goto fail_frame; - - if (length < LLC_HEADER_SIZE || length > ETH_MTU) - goto fail_frame; - - length -= LLC_HEADER_SIZE; - - dsap = reverse_bits(e_frame->dsap); - ssap = reverse_bits(e_frame->ssap); -#endif - - if (eth_len < ETH_HEADER_TOT_SIZE + (size_t) length) - goto fail_frame; - -#if defined(BUILD_ETH_DIX) - mem_hash(HASH_CRC8, &hcs, - (uint8_t *) &e_frame->eid, - DIX_EID_SIZE + DIX_LENGTH_SIZE); -#elif defined(BUILD_ETH_LLC) - mem_hash(HASH_CRC8, &hcs, - (uint8_t *) &e_frame->dsap, - LLC_FIELDS_SIZE); -#endif - if (hcs != e_frame->hcs) - goto fail_frame; - -#if defined(BUILD_ETH_DIX) - if (deid == MGMT_EID) { -#elif defined (BUILD_ETH_LLC) - if (ssap == MGMT_SAP && dsap == MGMT_SAP) { -#endif - ipcp_spb_release(spb); /* No need for the N+1 buffer. */ - - if (length > MGMT_FRAME_SIZE) { - log_warn("Management frame size %u exceeds %u.", - length, MGMT_FRAME_SIZE); - continue; - } - - frame = malloc(sizeof(*frame)); - if (frame == NULL) { - log_err("Failed to allocate frame."); - continue; - } - - memcpy(frame->buf, &e_frame->payload, length); - memcpy(frame->r_addr, e_frame->src_hwaddr, MAC_SIZE); - frame->len = length; - - pthread_mutex_lock(ð_data.mgmt_lock); - list_add(&frame->next, ð_data.mgmt_frames); - pthread_cond_signal(ð_data.mgmt_cond); - pthread_mutex_unlock(ð_data.mgmt_lock); - FETCH_ADD_RELAXED(ð_data.stat.n_rcv, 1); - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_rcv, 1); - } else { - pthread_rwlock_rdlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - fd = deid; -#elif defined(BUILD_ETH_LLC) - fd = eth_data.ef_to_fd[dsap]; -#endif - if (fd < 0) { - pthread_rwlock_unlock(ð_data.flows_lock); - FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); - goto fail_frame; - } - -#ifdef BUILD_ETH_LLC - if (eth_data.fd_to_ef[fd].r_sap != ssap - || memcmp(eth_data.fd_to_ef[fd].r_addr, - e_frame->src_hwaddr, MAC_SIZE)) { - pthread_rwlock_unlock(ð_data.flows_lock); - FETCH_ADD_RELAXED(ð_data.stat.n_bad_id, 1); - goto fail_frame; - } -#endif - FETCH_ADD_RELAXED(ð_data.fd_to_ef[fd].stat.p_rcv, 1); - FETCH_ADD_RELAXED(ð_data.fd_to_ef[fd].stat.b_rcv, - length); - FETCH_ADD_RELAXED(ð_data.stat.n_rcv, 1); - pthread_rwlock_unlock(ð_data.flows_lock); - -#ifndef HAVE_NETMAP - ssm_pk_buff_pop(spb, ETH_HEADER_TOT_SIZE); - ssm_pk_buff_truncate(spb, length); -#else - if (ipcp_spb_reserve(&spb, length)) - continue; - - buf = ssm_pk_buff_head(spb); - memcpy(buf, &e_frame->payload, length); -#endif - if (np1_flow_write(fd, spb, NP1_GET_POOL(fd)) < 0) { - ipcp_spb_release(spb); - FETCH_ADD_RELAXED( - ð_data.fd_to_ef[fd].stat.p_dlv_f, - 1); - FETCH_ADD_RELAXED(ð_data.stat.n_dlv_f, 1); - } - - continue; - fail_frame: -#ifndef HAVE_NETMAP - ipcp_spb_release(spb); -#endif - } - } - - return (void *) 0; -} - -static void cleanup_writer(void * o) -{ - fqueue_destroy((fqueue_t *) o); -} - -static void * eth_ipcp_packet_writer(void * o) -{ - int fd; - struct ssm_pk_buff * spb; - size_t len; -#if defined(BUILD_ETH_DIX) - uint16_t deid; -#elif defined(BUILD_ETH_LLC) - uint8_t dsap; - uint8_t ssap; -#endif - uint8_t r_addr[MAC_SIZE]; - - fqueue_t * fq; - - fq = fqueue_create(); - if (fq == NULL) - return (void *) -1; - - (void) o; - - ipcp_lock_to_core(); - - pthread_cleanup_push(cleanup_writer, fq); - - while (true) { - fevent(eth_data.np1_flows, fq, NULL); - while ((fd = fqueue_next(fq)) >= 0) { - if (fqueue_type(fq) != FLOW_PKT) - continue; - - if (np1_flow_read(fd, &spb, NP1_GET_POOL(fd))) { - log_dbg("Bad read from fd %d.", fd); - continue; - } - - len = ssm_pk_buff_len(spb); - - if (ssm_pk_buff_push(spb, ETH_HEADER_TOT_SIZE) - == NULL) { - log_dbg("Failed to allocate header."); - ipcp_spb_release(spb); - FETCH_ADD_RELAXED(ð_data.stat.n_buf_f, 1); - continue; - } - - pthread_rwlock_rdlock(ð_data.flows_lock); -#if defined(BUILD_ETH_DIX) - deid = eth_data.fd_to_ef[fd].r_eid; -#elif defined(BUILD_ETH_LLC) - dsap = reverse_bits(eth_data.fd_to_ef[fd].r_sap); - ssap = reverse_bits(eth_data.fd_to_ef[fd].sap); -#endif - memcpy(r_addr, - eth_data.fd_to_ef[fd].r_addr, - MAC_SIZE); - - pthread_rwlock_unlock(ð_data.flows_lock); - - if (eth_ipcp_send_frame(r_addr, -#if defined(BUILD_ETH_DIX) - deid, -#elif defined(BUILD_ETH_LLC) - dsap, ssap, -#endif - ssm_pk_buff_head(spb), - len)) { - log_dbg("Failed to send frame."); - FETCH_ADD_RELAXED( - ð_data.fd_to_ef[fd].stat.p_snd_f, - 1); - FETCH_ADD_RELAXED(ð_data.stat.n_snd_f, 1); - } else { - FETCH_ADD_RELAXED( - ð_data.fd_to_ef[fd].stat.p_snd, - 1); - FETCH_ADD_RELAXED( - ð_data.fd_to_ef[fd].stat.b_snd, - len); - } - ipcp_spb_release(spb); - } - } - - pthread_cleanup_pop(true); - - return (void *) 1; -} - -#ifdef __linux__ -static int open_netlink_socket(void) -{ - struct sockaddr_nl sa; - int fd; - - memset(&sa, 0, sizeof(sa)); - sa.nl_family = AF_NETLINK; - sa.nl_pid = getpid(); - sa.nl_groups = RTMGRP_LINK; - - fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE); - if (fd < 0) - return -1; - - if (bind(fd, (struct sockaddr *) &sa, sizeof(sa))) { - close(fd); - return -1; - } - - return fd; -} - -static void change_flows_state(bool up) -{ - int i; - uint32_t flags; - - pthread_rwlock_rdlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - for (i = 0; i < SYS_MAX_FLOWS; ++i) - if (eth_data.fd_to_ef[i].r_eid != -1) { - fccntl(i, FLOWGFLAGS, &flags); - if (up) - fccntl(i, FLOWSFLAGS, flags & ~FLOWFDOWN); - else - fccntl(i, FLOWSFLAGS, flags | FLOWFDOWN); - } -#elif defined(BUILD_ETH_LLC) - for (i = 0; i < MAX_SAPS; i++) - if (eth_data.ef_to_fd[i] != -1) { - fccntl(eth_data.ef_to_fd[i], FLOWGFLAGS, &flags); - if (up) - fccntl(eth_data.ef_to_fd[i], - FLOWSFLAGS, flags & ~FLOWFDOWN); - else - fccntl(eth_data.ef_to_fd[i], - FLOWSFLAGS, flags | FLOWFDOWN); - } -#endif - - pthread_rwlock_unlock(ð_data.flows_lock); -} - -static void * eth_ipcp_if_monitor(void * o) -{ - int fd; - int status; - char buf[4096]; - struct iovec iov = {buf, sizeof(buf)}; - struct sockaddr_nl snl; - struct msghdr msg = {(void *) &snl, sizeof(snl), - &iov, 1, NULL, 0, 0}; - struct nlmsghdr * h; - struct ifinfomsg * ifi; - - (void ) o; - - fd = open_netlink_socket(); - if (fd < 0) { - log_err("Failed to open socket."); - return (void *) -1; - } - - pthread_cleanup_push(__cleanup_close_ptr, &fd); - - while (true) { - status = recvmsg(fd, &msg, 0); - if (status < 0) - continue; - - for (h = (struct nlmsghdr *) buf; - NLMSG_OK(h, (unsigned int) status); - h = NLMSG_NEXT(h, status)) { - - /* Finish reading */ - if (h->nlmsg_type == NLMSG_DONE) - break; - - /* Message is some kind of error */ - if (h->nlmsg_type == NLMSG_ERROR) - continue; - - /* Only interested in link up/down */ - if (h->nlmsg_type != RTM_NEWLINK) - continue; - - ifi = NLMSG_DATA(h); - - /* Not our interface */ - if (ifi->ifi_index != eth_data.if_idx) - continue; - - if (ifi->ifi_flags & IFF_UP) { - change_flows_state(true); - log_dbg("Interface up."); - } else { - change_flows_state(false); - log_dbg("Interface down."); - } - } - } - - pthread_cleanup_pop(true); - - return (void *) 0; -} -#endif - -#if defined (HAVE_BPF) && !defined(HAVE_NETMAP) -static int open_bpf_device(void) -{ - char dev[32]; - size_t i = 0; - - for (i = 0; i < BPF_DEV_MAX; i++) { - int fd = -1; - - snprintf(dev, sizeof(dev), "/dev/bpf%zu", i); - - fd = open(dev, O_RDWR); - if (fd > -1) - return fd; - } - - return -1; -} -#endif - -#if defined(__FreeBSD__) || defined(__APPLE__) -static int ifr_hwaddr_from_ifaddrs(struct ifreq * ifr) -{ - struct ifaddrs * ifaddr; - struct ifaddrs * ifa; - int idx; - - if (getifaddrs(&ifaddr) < 0) { - log_err("Could not get interfaces."); - goto fail_ifaddrs; - } - - for (ifa = ifaddr, idx = 0; ifa != NULL; ifa = ifa->ifa_next, ++idx) { - if (strcmp(ifa->ifa_name, ifr->ifr_name) == 0) - break; - } - - if (ifa == NULL) { - log_err("Interface not found."); - goto fail_ifa; - } - - memcpy(&ifr->ifr_addr, ifa->ifa_addr, sizeof(*ifa->ifa_addr)); - - log_dbg("Interface %s hwaddr " MAC_FMT ".", ifr->ifr_name, - MAC_VAL(ifr->ifr_addr.sa_data)); - - freeifaddrs(ifaddr); - - return 0; - fail_ifa: - freeifaddrs(ifaddr); - fail_ifaddrs: - return -1; - -} -#elif defined(__linux__) -static int ifr_hwaddr_from_socket(struct ifreq * ifr) -{ - int skfd; - - skfd = socket(AF_UNIX, SOCK_STREAM, 0); - if (skfd < 0) { - log_err("Failed to open socket."); - goto fail_socket; - } - - if (ioctl(skfd, SIOCGIFHWADDR, ifr)) { - log_err("Failed to get hwaddr."); - goto fail_ifr; - } - - log_dbg("Interface %s hwaddr " MAC_FMT ".", ifr->ifr_name, - MAC_VAL(ifr->ifr_hwaddr.sa_data)); - - close(skfd); - - return 0; - - fail_ifr: - close(skfd); - fail_socket: - return -1; -} -#endif - -static int eth_ifr_hwaddr(struct ifreq * ifr) -{ -#if defined(__FreeBSD__) || defined(__APPLE__) - return ifr_hwaddr_from_ifaddrs(ifr); -#elif defined(__linux__) - return ifr_hwaddr_from_socket(ifr); -#else - return -1; -#endif -} - -static int eth_ifr_mtu(struct ifreq * ifr) -{ - int skfd; - - skfd = socket(AF_UNIX, SOCK_STREAM, 0); - if (skfd < 0) { - log_err("Failed to open socket."); - goto fail_socket; - } - - if (ioctl(skfd, SIOCGIFMTU, ifr) < 0) { - log_err("Failed to get MTU."); - goto fail_mtu; - } - close(skfd); - - return 0; - - fail_mtu: - close(skfd); - fail_socket: - return -1; -} - -static int eth_set_mtu(struct ifreq * ifr) -{ - if (eth_ifr_mtu(ifr) < 0) { - log_err("Failed to get interface MTU."); - return -1; - } - - log_dbg("Device MTU is %d.", ifr->ifr_mtu); - - eth_data.mtu = MIN((int) ETH_MTU_MAX, ifr->ifr_mtu); - if (memcmp(ifr->ifr_name, "lo", 2) == 0 && - eth_data.mtu > IPCP_ETH_LO_MTU) { - log_dbg("Using loopback interface. MTU restricted to %d.", - IPCP_ETH_LO_MTU); - eth_data.mtu = IPCP_ETH_LO_MTU; - } - - log_dbg("Layer MTU is %d.", eth_data.mtu); - - return 0; -} -#if defined(HAVE_NETMAP) -static int eth_init_nmd(struct ifreq * ifr) -{ - strcpy(ifn, "netmap:"); - strcat(ifn, ifr->ifr_name); - - eth_data.nmd = nm_open(ifn, NULL, 0, NULL); - if (eth_data.nmd == NULL) { - log_err("Failed to open netmap device."); - goto fail_nmd; - } - - memset(ð_data.poll_in, 0, sizeof(eth_data.poll_in)); - memset(ð_data.poll_out, 0, sizeof(eth_data.poll_out)); - - eth_data.poll_in.fd = NETMAP_FD(eth_data.nmd); - eth_data.poll_in.events = POLLIN; - eth_data.poll_out.fd = NETMAP_FD(eth_data.nmd); - eth_data.poll_out.events = POLLOUT; - - log_info("Using netmap device."); - - return 0; - fail_nmd: - return -1; -} -#elif defined (HAVE_BPF) -static int eth_init_bpf(struct ifreq * ifr) -{ - int enable = 1; - int disable = 0; - int blen; - - eth_data.bpf = open_bpf_device(); - if (eth_data.bpf < 0) { - log_err("Failed to open bpf device."); - goto fail_bpf; - } - - ioctl(eth_data.bpf, BIOCGBLEN, &blen); - if (BPF_BLEN < blen) { - log_err("BPF buffer too small (is: %ld must be: %d).", - BPF_BLEN, blen); - goto fail_device; - } - - if (ioctl(eth_data.bpf, BIOCSETIF, ifr) < 0) { - log_err("Failed to set interface."); - goto fail_device; - } - - if (ioctl(eth_data.bpf, BIOCSHDRCMPLT, &enable) < 0) { - log_err("Failed to set BIOCSHDRCMPLT."); - goto fail_device; - } - - if (ioctl(eth_data.bpf, BIOCSSEESENT, &disable) < 0) { - log_err("Failed to set BIOCSSEESENT."); - goto fail_device; - } - - if (ioctl(eth_data.bpf, BIOCIMMEDIATE, &enable) < 0) { - log_err("Failed to set BIOCIMMEDIATE."); - goto fail_device; - } - - log_info("Using Berkeley Packet Filter."); - - return 0; - - fail_device: - close(eth_data.bpf); - fail_bpf: - return -1; -} -#elif defined(HAVE_RAW_SOCKETS) -#define SOCKOPT() -static int eth_init_raw_socket(struct ifreq * ifr) -{ - int idx; - int sndbuf; - int rcvbuf; -#if defined(IPCP_ETH_QDISC_BYPASS) - int qdisc_bypass = 1; -#endif /* ENABLE_QDISC_BYPASS */ - - idx = if_nametoindex(ifr->ifr_name); - if (idx == 0) { - log_err("Failed to retrieve interface index."); - return -1; - } - - memset(&(eth_data.device), 0, sizeof(eth_data.device)); - eth_data.device.sll_ifindex = idx; - eth_data.device.sll_family = AF_PACKET; - memcpy(eth_data.device.sll_addr, ifr->ifr_hwaddr.sa_data, MAC_SIZE); - eth_data.device.sll_halen = MAC_SIZE; - eth_data.device.sll_protocol = htons(ETH_P_ALL); -#if defined (BUILD_ETH_DIX) - eth_data.s_fd = socket(AF_PACKET, SOCK_RAW, eth_data.ethertype); -#elif defined (BUILD_ETH_LLC) - eth_data.s_fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_802_2)); -#endif - if (eth_data.s_fd < 0) { - log_err("Failed to create socket."); - goto fail_socket; - } - -#if defined(IPCP_ETH_QDISC_BYPASS) - if (setsockopt(eth_data.s_fd, SOL_PACKET, PACKET_QDISC_BYPASS, - &qdisc_bypass, sizeof(qdisc_bypass))) { - log_info("Qdisc bypass not supported."); - } -#endif - - sndbuf = IPCP_ETH_SNDBUF; - if (sndbuf > 0) { - /* Never below one frame or sendto() cannot fit an SDU. */ - sndbuf = MAX(sndbuf, (int) (ETH_HEADER_TOT_SIZE + ETH_MTU)); - if (setsockopt(eth_data.s_fd, SOL_SOCKET, SO_SNDBUF, - &sndbuf, sizeof(sndbuf))) - log_info("Failed to set SO_SNDBUF to %d.", sndbuf); - } - - rcvbuf = IPCP_ETH_RCVBUF; - if (rcvbuf > 0) { - /* SO_RCVBUFFORCE bypasses rmem_max; SO_RCVBUF is fallback. */ - if (setsockopt(eth_data.s_fd, SOL_SOCKET, SO_RCVBUFFORCE, - &rcvbuf, sizeof(rcvbuf)) - && setsockopt(eth_data.s_fd, SOL_SOCKET, SO_RCVBUF, - &rcvbuf, sizeof(rcvbuf))) - log_info("Failed to set SO_RCVBUF to %d.", rcvbuf); - } - - if (bind(eth_data.s_fd, (struct sockaddr *) ð_data.device, - sizeof(eth_data.device)) < 0) { - log_err("Failed to bind socket to interface."); - goto fail_device; - } -#ifdef __linux__ - eth_data.if_idx = idx; -#endif - log_info("Using raw socket device."); - - return 0; - fail_device: - close(eth_data.s_fd); - fail_socket: - return -1; -} -#endif - -static int eth_ipcp_bootstrap(struct ipcp_config * conf) -{ - struct ifreq ifr; - int i; -#if defined(HAVE_NETMAP) - char ifn[IFNAMSIZ]; -#endif /* HAVE_NETMAP */ - - assert(conf); - assert(conf->type == THIS_TYPE); - - memset(&ifr, 0, sizeof(ifr)); - strcpy(ifr.ifr_name, conf->eth.dev); - - if (strlen(conf->eth.dev) >= IFNAMSIZ) { - log_err("Invalid device name: %s.", conf->eth.dev); - return -1; - } -#ifdef BUILD_ETH_DIX - if (conf->eth.ethertype < 0x0600 || conf->eth.ethertype == 0xFFFF) { - log_err("Invalid Ethertype: %d.", conf->eth.ethertype); - return -1; - } - eth_data.ethertype = htons(conf->eth.ethertype); -#endif - if (eth_set_mtu(&ifr) < 0) { - log_err("Failed to set MTU."); - return -1; - } - - if (eth_ifr_hwaddr(&ifr) < 0) { - log_err("Failed to get hardware addr."); - return -1; - } -#if defined(HAVE_NETMAP) || defined(HAVE_BPF) - memcpy(eth_data.hw_addr, LLADDR((struct sockaddr_dl *) &ifr.ifr_addr), - MAC_SIZE); -#endif -#if defined(HAVE_NETMAP) - if (eth_init_nmd(&ifr) < 0) { - log_err("Failed to initialize netmap device."); - return -1; - } -#elif defined(HAVE_BPF) /* !HAVE_NETMAP */ - if (eth_init_bpf(&ifr) < 0) { - log_err("Failed to initialize BPF device."); - return -1; - } -#elif defined(HAVE_RAW_SOCKETS) - if (eth_init_raw_socket(&ifr) < 0) { - log_err("Failed to initialize raw socket device."); - return -1; - } -#endif /* HAVE_NETMAP */ -#ifdef IPCP_ETH_FLOW_STATS - if (rib_reg(ETH_RIB_PATH, ð_r_ops)) { - log_err("Failed to register RIB."); - goto fail_rib_reg; - } -#endif -#if defined(__linux__) - if (pthread_create(ð_data.if_monitor, NULL, - eth_ipcp_if_monitor, NULL)) { - log_err("Failed to create monitor thread: %s.", - strerror(errno)); - goto fail_monitor; - } -#endif - if (pthread_create(ð_data.mgmt_handler, NULL, - eth_ipcp_mgmt_handler, NULL)) { - log_err("Failed to create mgmt handler thread: %s.", - strerror(errno)); - goto fail_mgmt_handler; - } - - for (i = 0; i < IPCP_ETH_RD_THR; i++) { - if (pthread_create(ð_data.packet_reader[i], NULL, - eth_ipcp_packet_reader, NULL)) { - log_err("Failed to create packet reader thread: %s", - strerror(errno)); - goto fail_packet_reader; - } - } - - for (i = 0; i < IPCP_ETH_WR_THR; i++) { - if (pthread_create(ð_data.packet_writer[i], NULL, - eth_ipcp_packet_writer, NULL)) { - log_err("Failed to create packet writer thread: %s", - strerror(errno)); - goto fail_packet_writer; - } - } - -#if defined(BUILD_ETH_DIX) - log_dbg("Bootstrapped IPCP over DIX Ethernet with pid %d " - "and Ethertype 0x%X.", getpid(), conf->eth.ethertype); -#elif defined(BUILD_ETH_LLC) - log_dbg("Bootstrapped IPCP over Ethernet with LLC with pid %d.", - getpid()); -#endif - return 0; - - fail_packet_writer: - while (i-- > 0) { - pthread_cancel(eth_data.packet_writer[i]); - pthread_join(eth_data.packet_writer[i], NULL); - } - i = IPCP_ETH_RD_THR; - fail_packet_reader: - while (i-- > 0) { - pthread_cancel(eth_data.packet_reader[i]); - pthread_join(eth_data.packet_reader[i], NULL); - } - pthread_cancel(eth_data.mgmt_handler); - pthread_join(eth_data.mgmt_handler, NULL); - fail_mgmt_handler: -#if defined(__linux__) - pthread_cancel(eth_data.if_monitor); - pthread_join(eth_data.if_monitor, NULL); -#endif -#if defined(__linux__) - fail_monitor: -#endif -#ifdef IPCP_ETH_FLOW_STATS - rib_unreg(ETH_RIB_PATH); - fail_rib_reg: -#endif -#if defined(HAVE_NETMAP) - nm_close(eth_data.nmd); -#elif defined(HAVE_BPF) - close(eth_data.bpf); -#elif defined(HAVE_RAW_SOCKETS) - close(eth_data.s_fd); -#endif - return -1; -} - -static int eth_ipcp_reg(const uint8_t * hash) -{ - if (shim_data_reg_add_entry(eth_data.shim_data, hash)) { - log_err("Failed to add " HASH_FMT32 " to local registry.", - HASH_VAL32(hash)); - return -1; - } - - return 0; -} - -static int eth_ipcp_unreg(const uint8_t * hash) -{ - shim_data_reg_del_entry(eth_data.shim_data, hash); - - return 0; -} - -static int eth_ipcp_query(const uint8_t * hash) -{ - uint8_t r_addr[MAC_SIZE]; - struct timespec timeout; - struct dir_query * query; - int ret; - int attempt; - uint8_t * buf; - struct mgmt_msg * msg; - size_t len; - long per_ms; - - if (shim_data_dir_has(eth_data.shim_data, hash)) - return 0; - - len = sizeof(*msg) + ipcp_dir_hash_len(); - - buf = malloc(len + ETH_HEADER_TOT_SIZE); - if (buf == NULL) - return -1; - - memset(buf, 0, len + ETH_HEADER_TOT_SIZE); - - msg = (struct mgmt_msg *) (buf + ETH_HEADER_TOT_SIZE); - msg->code = NAME_QUERY_REQ; - - memcpy(msg + 1, hash, ipcp_dir_hash_len()); - - memset(r_addr, 0xff, MAC_SIZE); - - per_ms = NAME_QUERY_TIMEO / (NAME_QUERY_RETRIES + 1); - - ret = -1; - for (attempt = 0; attempt <= NAME_QUERY_RETRIES; ++attempt) { - query = shim_data_dir_query_create(eth_data.shim_data, hash); - if (query == NULL) { - ret = -1; - break; - } - - if (eth_ipcp_send_frame(r_addr, -#if defined(BUILD_ETH_DIX) - MGMT_EID, -#elif defined(BUILD_ETH_LLC) - reverse_bits(MGMT_SAP), - reverse_bits(MGMT_SAP), -#endif - buf, len)) { - log_err("Failed to send management frame."); - shim_data_dir_query_destroy(eth_data.shim_data, - query); - ret = -1; - break; - } - - FETCH_ADD_RELAXED(ð_data.stat.n_mgmt_snd, 1); - - timeout.tv_sec = per_ms / 1000; - timeout.tv_nsec = (per_ms % 1000) * 1000000L; - - ret = shim_data_dir_query_wait(query, &timeout); - - shim_data_dir_query_destroy(eth_data.shim_data, query); - - if (ret != -ETIMEDOUT) - break; - } - - free(buf); - - return ret; -} - -static int eth_ipcp_flow_alloc(int fd, - const uint8_t * hash, - qosspec_t qs, - const buffer_t * data) -{ -#ifdef BUILD_ETH_LLC - uint8_t ssap = 0; -#endif - uint8_t r_addr[MAC_SIZE]; - struct addr addr; - - assert(hash); - - if (!shim_data_dir_has(eth_data.shim_data, hash)) { - log_err("Destination "HASH_FMT32 "unreachable.", - HASH_VAL32(hash)); - return -1; - } - - addr = shim_data_dir_get_addr(eth_data.shim_data, hash); - memcpy(r_addr, &addr.mac, MAC_SIZE); - -#ifdef BUILD_ETH_LLC - pthread_rwlock_wrlock(ð_data.flows_lock); - ssap = bmp_allocate(eth_data.saps); - if (!bmp_is_id_valid(eth_data.saps, ssap)) { - pthread_rwlock_unlock(ð_data.flows_lock); - log_err("Failed to allocate SSAP."); - return -1; - } - - eth_data.fd_to_ef[fd].sap = ssap; - eth_data.ef_to_fd[ssap] = fd; - pthread_rwlock_unlock(ð_data.flows_lock); -#endif - - if (eth_ipcp_alloc(r_addr, -#if defined(BUILD_ETH_DIX) - fd, -#elif defined(BUILD_ETH_LLC) - ssap, -#endif - hash, - qs, - data) < 0) { -#ifdef BUILD_ETH_LLC - pthread_rwlock_wrlock(ð_data.flows_lock); - bmp_release(eth_data.saps, eth_data.fd_to_ef[fd].sap); - eth_data.fd_to_ef[fd].sap = -1; - eth_data.ef_to_fd[ssap] = -1; - pthread_rwlock_unlock(ð_data.flows_lock); - log_err("Failed to allocate with peer."); -#endif - return -1; - } - - fset_add(eth_data.np1_flows, fd); -#ifdef IPCP_ETH_FLOW_STATS - pthread_rwlock_wrlock(ð_data.flows_lock); - memset(ð_data.fd_to_ef[fd].stat, 0, - sizeof(eth_data.fd_to_ef[fd].stat)); - eth_data.fd_to_ef[fd].stat.stamp = time(NULL); - FETCH_ADD_RELAXED(ð_data.stat.n_flows, 1); - pthread_rwlock_unlock(ð_data.flows_lock); -#endif -#if defined(BUILD_ETH_LLC) - log_dbg("Assigned SAP %d for fd %d.", ssap, fd); -#endif - return 0; -} - -static int eth_ipcp_flow_alloc_resp(int fd, - int response, - const buffer_t * data) -{ -#if defined(BUILD_ETH_DIX) - uint16_t r_eid; -#elif defined(BUILD_ETH_LLC) - uint8_t ssap; - uint8_t r_sap; -#endif - uint8_t r_addr[MAC_SIZE]; - - if (ipcp_wait_flow_resp(fd) < 0) { - log_err("Failed to wait for flow response."); - return -1; - } - - pthread_rwlock_wrlock(ð_data.flows_lock); -#if defined(BUILD_ETH_DIX) - r_eid = eth_data.fd_to_ef[fd].r_eid; -#elif defined(BUILD_ETH_LLC) - ssap = bmp_allocate(eth_data.saps); - if (!bmp_is_id_valid(eth_data.saps, ssap)) { - pthread_rwlock_unlock(ð_data.flows_lock); - log_err("Failed to allocate SSAP."); - return -1; - } - - eth_data.fd_to_ef[fd].sap = ssap; - r_sap = eth_data.fd_to_ef[fd].r_sap; - eth_data.ef_to_fd[ssap] = fd; -#endif - memcpy(r_addr, eth_data.fd_to_ef[fd].r_addr, MAC_SIZE); - - pthread_rwlock_unlock(ð_data.flows_lock); - - if (eth_ipcp_alloc_resp(r_addr, -#if defined(BUILD_ETH_DIX) - fd, r_eid, -#elif defined(BUILD_ETH_LLC) - ssap, r_sap, -#endif - response, - data) < 0) { -#ifdef BUILD_ETH_LLC - pthread_rwlock_wrlock(ð_data.flows_lock); - bmp_release(eth_data.saps, eth_data.fd_to_ef[fd].sap); - pthread_rwlock_unlock(ð_data.flows_lock); -#endif - log_err("Failed to respond to peer."); - return -1; - } - - fset_add(eth_data.np1_flows, fd); -#ifdef IPCP_ETH_FLOW_STATS - pthread_rwlock_wrlock(ð_data.flows_lock); - memset(ð_data.fd_to_ef[fd].stat, 0, - sizeof(eth_data.fd_to_ef[fd].stat)); - eth_data.fd_to_ef[fd].stat.stamp = time(NULL); - FETCH_ADD_RELAXED(ð_data.stat.n_flows, 1); - pthread_rwlock_unlock(ð_data.flows_lock); -#endif -#if defined(BUILD_ETH_LLC) - log_dbg("Assigned SAP %d for fd %d.", ssap, fd); -#endif - return 0; -} - -static int eth_ipcp_flow_dealloc(int fd) -{ -#ifdef BUILD_ETH_LLC - uint8_t sap; -#endif - ipcp_flow_fini(fd); - - fset_del(eth_data.np1_flows, fd); - - pthread_rwlock_wrlock(ð_data.flows_lock); - -#if defined(BUILD_ETH_DIX) - eth_data.fd_to_ef[fd].r_eid = -1; -#elif defined BUILD_ETH_LLC - sap = eth_data.fd_to_ef[fd].sap; - bmp_release(eth_data.saps, sap); - eth_data.fd_to_ef[fd].sap = -1; - eth_data.fd_to_ef[fd].r_sap = -1; - eth_data.ef_to_fd[sap] = -1; -#endif - memset(ð_data.fd_to_ef[fd].r_addr, 0, MAC_SIZE); - -#ifdef IPCP_ETH_FLOW_STATS - memset(ð_data.fd_to_ef[fd].stat, 0, - sizeof(eth_data.fd_to_ef[fd].stat)); - FETCH_SUB_RELAXED(ð_data.stat.n_flows, 1); -#endif - - pthread_rwlock_unlock(ð_data.flows_lock); - - ipcp_flow_dealloc(fd); - - return 0; -} - -static struct ipcp_ops eth_ops = { - .ipcp_bootstrap = eth_ipcp_bootstrap, - .ipcp_enroll = NULL, - .ipcp_connect = NULL, - .ipcp_disconnect = NULL, - .ipcp_reg = eth_ipcp_reg, - .ipcp_unreg = eth_ipcp_unreg, - .ipcp_query = eth_ipcp_query, - .ipcp_flow_alloc = eth_ipcp_flow_alloc, - .ipcp_flow_join = NULL, - .ipcp_flow_alloc_resp = eth_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = eth_ipcp_flow_dealloc, - .ipcp_flow_update = eth_ipcp_flow_update -}; - -int main(int argc, - char * argv[]) -{ - int i; - - if (eth_data_init() < 0) { -#if defined(BUILD_ETH_DIX) - log_err("Failed to init eth-llc data."); -#elif defined(BUILD_ETH_LLC) - log_err("Failed to init eth-dix data."); -#endif - goto fail_data_init; - } - - if (ipcp_init(argc, argv, ð_ops, THIS_TYPE) < 0) { - log_err("Failed to initialize IPCP."); - goto fail_init; - } - - if (ipcp_start() < 0) { - log_err("Failed to start IPCP."); - goto fail_start; - } - - ipcp_sigwait(); - - if (ipcp_get_state() == IPCP_SHUTDOWN) { - for (i = 0; i < IPCP_ETH_WR_THR; ++i) - pthread_cancel(eth_data.packet_writer[i]); - for (i = 0; i < IPCP_ETH_RD_THR; ++i) - pthread_cancel(eth_data.packet_reader[i]); - - pthread_cancel(eth_data.mgmt_handler); -#ifdef __linux__ - pthread_cancel(eth_data.if_monitor); -#endif - for (i = 0; i < IPCP_ETH_WR_THR; ++i) - pthread_join(eth_data.packet_writer[i], NULL); - for (i = 0; i < IPCP_ETH_RD_THR; ++i) - pthread_join(eth_data.packet_reader[i], NULL); - - pthread_join(eth_data.mgmt_handler, NULL); -#ifdef __linux__ - pthread_join(eth_data.if_monitor, NULL); -#endif -#ifdef IPCP_ETH_FLOW_STATS - rib_unreg(ETH_RIB_PATH); -#endif - } - - ipcp_stop(); - - ipcp_fini(); - - eth_data_fini(); - - exit(EXIT_SUCCESS); - - fail_start: - ipcp_fini(); - fail_init: - eth_data_fini(); - fail_data_init: - exit(EXIT_FAILURE); -} diff --git a/src/ipcpd/ipcp.c b/src/ipcpd/ipcp.c index dcee4b9c..b25b7d04 100644 --- a/src/ipcpd/ipcp.c +++ b/src/ipcpd/ipcp.c @@ -36,6 +36,7 @@ #define OUROBOROS_PREFIX "ipcpd/ipcp" #define IPCP_INFO "info" #define ALLOC_TIMEOUT 50 /* ms */ +#define HAS_POA (ipcpd.type != IPCP_LOCAL) #include <ouroboros/bitmap.h> #include <ouroboros/dev.h> @@ -54,6 +55,7 @@ #include "ipcp.h" #include "np1.h" +#include <arpa/inet.h> #include <signal.h> #include <string.h> #include <sys/socket.h> @@ -69,14 +71,14 @@ #define CLOCK_REALTIME_COARSE CLOCK_REALTIME #endif +/* Fits "eth <dev> 0x<type>", the longest PoA rendering. */ +/* Matches src/tools/irm/irm_utils.h; keep in sync. */ +#define POA_STRLEN (DEV_NAME_SIZE + 11) + static char * ipcp_type_str[] = { "local", "unicast", - "broadcast", - "eth-llc", - "eth-dix", - "udp4", - "udp6" + "broadcast" }; static char * dir_hash_str[] = { @@ -237,14 +239,6 @@ static int ipcp_rib_read(const char * path, strcpy(buf, "unicast\n"); else if (ipcpd.type == IPCP_BROADCAST) strcpy(buf, "broadcast\n"); - else if (ipcpd.type == IPCP_ETH_LLC) - strcpy(buf, "eth-llc\n"); - else if (ipcpd.type == IPCP_ETH_DIX) - strcpy(buf, "eth-dix\n"); - else if (ipcpd.type == IPCP_UDP4) - strcpy(buf, "udp4\n"); - else if (ipcpd.type == IPCP_UDP6) - strcpy(buf, "udp6\n"); else strcpy(buf, "bug\n"); } @@ -316,8 +310,7 @@ static void * acceptloop(void * o) (void) o; - while (ipcp_get_state() != IPCP_SHUTDOWN && - ipcp_get_state() != IPCP_INIT) { + while (ipcp_get_state() != IPCP_SHUTDOWN) { struct cmd * cmd; csockfd = accept(ipcpd.sockfd, 0, 0); @@ -471,11 +464,6 @@ static void do_bootstrap(ipcp_config_msg_t * conf_msg, conf = ipcp_config_msg_to_s(conf_msg); switch(conf.type) { /* FIXED algorithms */ - case IPCP_UDP4: - /* FALLTHRU */ - case IPCP_UDP6: - conf.layer_info.dir_hash_algo = (enum pol_dir_hash) HASH_MD5; - break; case IPCP_BROADCAST: conf.layer_info.dir_hash_algo = DIR_HASH_SHA3_256; break; @@ -493,6 +481,8 @@ static void do_bootstrap(ipcp_config_msg_t * conf_msg, strcpy(ipcpd.layer_name, info->name); ipcpd.dir_hash_algo = (enum hash_algo) info->dir_hash_algo; + if (poa_set_layer(info->name) < 0) + log_warn("Failed to set layer name for PoA queries."); ret_msg->layer_info = layer_info_s_to_msg(info); ipcp_set_state(IPCP_OPERATIONAL); @@ -503,10 +493,13 @@ static void do_bootstrap(ipcp_config_msg_t * conf_msg, ipcp_dir_hash_len()); } -static void do_enroll(const char * dst, - ipcp_msg_t * ret_msg) +static void do_enroll(const char * dst, + const poa_addr_msg_t * peer, + ipcp_msg_t * ret_msg) { struct layer_info info; + struct poa_addr addr; + struct poa_addr * pa = NULL; log_info("Enrolling with %s...", dst); @@ -524,7 +517,16 @@ static void do_enroll(const char * dst, return; } - ret_msg->result = ipcpd.ops->ipcp_enroll(dst, &info); + if (peer != NULL) { + addr = poa_addr_msg_to_s(peer); + if (addr.type == POA_INVALID || addr.type == POA_UDP) { + ret_msg->result = -EINVAL; + return; + } + pa = &addr; + } + + ret_msg->result = ipcpd.ops->ipcp_enroll(dst, pa, &info); if (ret_msg->result < 0) { log_err("Failed to bootstrap IPCP."); return; @@ -532,6 +534,8 @@ static void do_enroll(const char * dst, strcpy(ipcpd.layer_name, info.name); ipcpd.dir_hash_algo = (enum hash_algo) info.dir_hash_algo; + if (poa_set_layer(info.name) < 0) + log_warn("Failed to set layer name for PoA queries."); ret_msg->layer_info = layer_info_s_to_msg(&info); ipcp_set_state(IPCP_OPERATIONAL); @@ -542,11 +546,136 @@ static void do_enroll(const char * dst, ipcp_dir_hash_len()); } -static void do_connect(const char * dst, - const char * comp, - qosspec_t qs, - ipcp_msg_t * ret_msg) +/* Bounded so one oversized reply cannot be built; 64 is generous. */ +#define POA_LIST_MAX 64 + +static void do_list_poas(ipcp_msg_t * ret_msg) +{ + struct poa_spec eps[POA_LIST_MAX]; + ssize_t n; + ssize_t i; + + if (ipcpd.type != IPCP_UNICAST && ipcpd.type != IPCP_BROADCAST) { + ret_msg->result = -ENOTSUP; + return; + } + + n = poa_list(eps, POA_LIST_MAX); + if (n < 0) { + ret_msg->result = (int) n; + return; + } + + if (n > POA_LIST_MAX) { + log_warn("Listing %d of %zd PoAs.", POA_LIST_MAX, n); + + n = POA_LIST_MAX; + } + + if (n == 0) { + ret_msg->result = 0; + return; + } + + ret_msg->poas = malloc(n * sizeof(*ret_msg->poas)); + if (ret_msg->poas == NULL) { + ret_msg->result = -ENOMEM; + return; + } + + for (i = 0; i < n; i++) { + ret_msg->poas[i] = poa_spec_s_to_msg(&eps[i]); + if (ret_msg->poas[i] == NULL) { + ret_msg->result = -ENOMEM; + return; + } + ret_msg->n_poas++; + } + + ret_msg->result = 0; +} + +/* Names a PoA the way "irm ipcp poa list" prints it. */ +/* Matches src/tools/irm/irm_utils.c; keep in sync. */ +static void poa_spec_str(const struct poa_spec * poa, + char * buf, + size_t len) +{ + char addr[INET6_ADDRSTRLEN]; + + switch (poa->type) { + case POA_UDP4: + if (inet_ntop(AF_INET, &poa->udp4.ip_addr, + addr, sizeof(addr)) == NULL) + break; + + snprintf(buf, len, "udp4 %s:%u", addr, poa->udp4.port); + return; + case POA_UDP6: + if (inet_ntop(AF_INET6, &poa->udp6.ip_addr, + addr, sizeof(addr)) == NULL) + break; + + snprintf(buf, len, "udp6 [%s]:%u", addr, poa->udp6.port); + return; + case POA_ETH: + snprintf(buf, len, "eth %s 0x%04X", poa->eth.dev, + poa->eth.ethertype); + return; + default: + break; + } + + snprintf(buf, len, "(unknown)"); +} + +static void do_attach(poa_spec_msg_t * msg, + ipcp_msg_t * ret_msg) { + struct poa_spec poa; + char str[POA_STRLEN + 1]; + + poa = poa_spec_msg_to_s(msg); + + poa_spec_str(&poa, str, sizeof(str)); + + ret_msg->result = poa_attach(&poa); + if (ret_msg->result < 0) { + log_err("Failed to attach %s.", str); + return; + } + + log_info("Attached %s.", str); +} + +static void do_detach(poa_spec_msg_t * msg, + ipcp_msg_t * ret_msg) +{ + struct poa_spec poa; + char str[POA_STRLEN + 1]; + + poa = poa_spec_msg_to_s(msg); + + poa_spec_str(&poa, str, sizeof(str)); + + ret_msg->result = poa_detach(&poa); + if (ret_msg->result < 0) { + log_err("Failed to detach %s.", str); + return; + } + + log_info("Detached %s.", str); +} + +static void do_connect(const char * dst, + const char * comp, + qosspec_t qs, + const poa_addr_msg_t * peer, + ipcp_msg_t * ret_msg) +{ + struct poa_addr addr; + struct poa_addr * pa = NULL; + log_info("Connecting %s to %s...", comp, dst); if (ipcpd.ops->ipcp_connect == NULL) { @@ -555,7 +684,16 @@ static void do_connect(const char * dst, return; } - ret_msg->result = ipcpd.ops->ipcp_connect(dst, comp, qs); + if (peer != NULL) { + addr = poa_addr_msg_to_s(peer); + if (addr.type == POA_INVALID || addr.type == POA_UDP) { + ret_msg->result = -EINVAL; + return; + } + pa = &addr; + } + + ret_msg->result = ipcpd.ops->ipcp_connect(dst, comp, qs, pa); log_info("Finished connecting."); } @@ -734,9 +872,14 @@ static void do_flow_join(pid_t pid, log_info("Finished joining layer " HASH_FMT32 ".", HASH_VAL32(dst)); } +/* + * The IRMd says whether the flow is on a PoA, as it may not be known + * here yet. PoA flows answer without the IPCP type's flow machinery. + */ static void do_flow_alloc_resp(int resp, int flow_id, uid_t uid, + bool is_poa, const buffer_t * data, ipcp_msg_t * ret_msg) { @@ -745,13 +888,6 @@ static void do_flow_alloc_resp(int resp, log_info("Responding %d to alloc on flow_id %d.", resp, flow_id); - if (ipcpd.ops->ipcp_flow_alloc_resp == NULL) { - log_err("Failed to respond on flow %d: operation unsupported.", - flow_id); - ret_msg->result = -ENOTSUP; - return; - } - if (ipcp_get_state() != IPCP_OPERATIONAL) { log_err("Failed to respond to flow %d:" "IPCP in state <%s>, need <%s>.", @@ -762,6 +898,20 @@ static void do_flow_alloc_resp(int resp, return; } + if (is_poa) { + ret_msg->result = poa_flow_alloc_resp(flow_id, resp, data); + log_info("Finished responding %d on PoA flow %d.", + ret_msg->result, flow_id); + return; + } + + if (ipcpd.ops->ipcp_flow_alloc_resp == NULL) { + log_err("Failed to respond on flow %d: operation unsupported.", + flow_id); + ret_msg->result = -ENOTSUP; + return; + } + fd = np1_flow_resp(flow_id, resp); if (fd < 0) { log_warn("Flow_id %d is not known.", flow_id); @@ -788,18 +938,13 @@ static void do_flow_alloc_resp(int resp, static void do_flow_dealloc(int flow_id, int timeo_sec, + bool is_poa, ipcp_msg_t * ret_msg) { int fd; log_info("Deallocating flow %d.", flow_id); - if (ipcpd.ops->ipcp_flow_dealloc == NULL) { - log_err("Failed to dealloc: operation unsupported."); - ret_msg->result = -ENOTSUP; - return; - } - if (ipcp_get_state() != IPCP_OPERATIONAL) { log_err("Failed to enroll: IPCP in state <%s>, need <%s>.", ipcp_state_str[ipcp_get_state()], @@ -808,6 +953,20 @@ static void do_flow_dealloc(int flow_id, return; } + if (is_poa) { + ret_msg->result = poa_flow_dealloc(flow_id); + + log_info("Finished deallocating PoA flow %d.", flow_id); + return; + } + + if (ipcpd.ops->ipcp_flow_dealloc == NULL) { + log_err("Failed to dealloc: operation unsupported."); + + ret_msg->result = -ENOTSUP; + return; + } + fd = np1_flow_dealloc(flow_id, timeo_sec); if (fd < 0) { log_warn("Could not deallocate flow_id %d.", flow_id); @@ -822,18 +981,25 @@ static void do_flow_dealloc(int flow_id, static void do_flow_update(int flow_id, const buffer_t * data, + bool is_poa, ipcp_msg_t * ret_msg) { int fd; - if (ipcpd.ops->ipcp_flow_update == NULL) { - log_err("Failed to update flow: operation unsupported."); - ret_msg->result = -ENOTSUP; + if (ipcp_get_state() != IPCP_OPERATIONAL) { + ret_msg->result = -EIPCPSTATE; return; } - if (ipcp_get_state() != IPCP_OPERATIONAL) { - ret_msg->result = -EIPCPSTATE; + if (is_poa) { + ret_msg->result = poa_flow_update(flow_id, data); + return; + } + + if (ipcpd.ops->ipcp_flow_update == NULL) { + log_err("Failed to update flow: operation unsupported."); + + ret_msg->result = -ENOTSUP; return; } @@ -881,6 +1047,7 @@ static void * mainloop(void * o) free(cmd); if (msg == NULL) { + log_err("Failed to unpack command message."); close(sfd); continue; } @@ -897,11 +1064,23 @@ static void * mainloop(void * o) do_bootstrap(msg->conf, &ret_msg); break; case IPCP_MSG_CODE__IPCP_ENROLL: - do_enroll(msg->dst, &ret_msg); + do_enroll(msg->dst, msg->peer, &ret_msg); + break; + case IPCP_MSG_CODE__IPCP_ATTACH: + assert(HAS_POA); + do_attach(msg->poa, &ret_msg); + break; + case IPCP_MSG_CODE__IPCP_DETACH: + assert(HAS_POA); + do_detach(msg->poa, &ret_msg); + break; + case IPCP_MSG_CODE__IPCP_LIST_POAS: + do_list_poas(&ret_msg); break; case IPCP_MSG_CODE__IPCP_CONNECT: qs = qos_spec_msg_to_s(msg->qosspec); - do_connect(msg->dst, msg->comp, qs, &ret_msg); + do_connect(msg->dst, msg->comp, qs, msg->peer, + &ret_msg); break; case IPCP_MSG_CODE__IPCP_DISCONNECT: do_disconnect(msg->dst, msg->comp, &ret_msg); @@ -940,17 +1119,20 @@ static void * mainloop(void * o) data.len = msg->pk.len; data.data = msg->pk.data; do_flow_alloc_resp(msg->response, msg->flow_id, - msg->uid, &data, &ret_msg); + msg->uid, msg->is_poa, + &data, &ret_msg); break; case IPCP_MSG_CODE__IPCP_FLOW_DEALLOC: - do_flow_dealloc(msg->flow_id, msg->timeo_sec, &ret_msg); + do_flow_dealloc(msg->flow_id, msg->timeo_sec, + msg->is_poa, &ret_msg); break; case IPCP_MSG_CODE__IPCP_FLOW_UPDATE: assert(msg->pk.len > 0 ? msg->pk.data != NULL : msg->pk.data == NULL); data.len = msg->pk.len; data.data = msg->pk.data; - do_flow_update(msg->flow_id, &data, &ret_msg); + do_flow_update(msg->flow_id, &data, msg->is_poa, + &ret_msg); break; default: ret_msg.result = -1; @@ -988,7 +1170,7 @@ static void * mainloop(void * o) if (write(sfd, buffer.data, buffer.len) == -1) log_warn("Failed to send reply message"); - pthread_cleanup_pop(true); /* close sfd */ + pthread_cleanup_pop(true); /* close sfd */ pthread_cleanup_pop(true); /* free buffer.data */ tpm_end_work(ipcpd.tpm); @@ -1101,6 +1283,11 @@ int ipcp_init(int argc, goto fail_rib_reg; } + if (poa_init(ipcpd.name) < 0) { + log_err("Failed to initialize PoAs."); + goto fail_poa_init; + } + list_head_init(&ipcpd.cmds); ipcpd.tpm = tpm_create(IPCP_MIN_THREADS, IPCP_ADD_THREADS, @@ -1124,6 +1311,8 @@ int ipcp_init(int argc, return 0; fail_tpm_create: + poa_fini(); + fail_poa_init: rib_unreg(IPCP_INFO); fail_rib_reg: rib_fini(); @@ -1149,6 +1338,7 @@ int ipcp_init(int argc, return -1; } +/* Enrolment runs over a PoA, so poa_start() precedes any RPC. */ int ipcp_start(void) { sigset_t sigset; @@ -1169,6 +1359,11 @@ int ipcp_start(void) ipcp_set_state(IPCP_BOOT); + if (poa_start() < 0) { + log_err("Failed to start PoAs."); + goto fail_poa_start; + } + if (tpm_start(ipcpd.tpm)) { log_err("Failed to start threadpool manager."); goto fail_tpm_start; @@ -1192,6 +1387,8 @@ int ipcp_start(void) fail_acceptor: tpm_stop(ipcpd.tpm); fail_tpm_start: + poa_stop(); + fail_poa_start: tpm_destroy(ipcpd.tpm); ipcp_set_state(IPCP_INIT); ipcp_create_r(&info); @@ -1263,6 +1460,8 @@ void ipcp_stop(void) tpm_stop(ipcpd.tpm); + poa_stop(); + ipcp_set_state(IPCP_INIT); } @@ -1271,6 +1470,8 @@ void ipcp_fini(void) tpm_destroy(ipcpd.tpm); + poa_fini(); + rib_unreg(IPCP_INFO); rib_fini(); diff --git a/src/ipcpd/ipcp.h b/src/ipcpd/ipcp.h index 210157ec..e0aab291 100644 --- a/src/ipcpd/ipcp.h +++ b/src/ipcpd/ipcp.h @@ -36,15 +36,18 @@ #define ipcp_dir_hash_strlen() (ipcp_dir_hash_len() * 2) +/* Attach or release one flow PoA on a unicast or broadcast IPCP. */ struct ipcp_ops { int (* ipcp_bootstrap)(struct ipcp_config * conf); - int (* ipcp_enroll)(const char * dst, - struct layer_info * info); + int (* ipcp_enroll)(const char * dst, + const struct poa_addr * addr, + struct layer_info * info); - int (* ipcp_connect)(const char * dst, - const char * component, - qosspec_t qs); + int (* ipcp_connect)(const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr); int (* ipcp_disconnect)(const char * dst, const char * component); diff --git a/src/ipcpd/local/CMakeLists.txt b/src/ipcpd/local/CMakeLists.txt index 91f300a3..af433d01 100644 --- a/src/ipcpd/local/CMakeLists.txt +++ b/src/ipcpd/local/CMakeLists.txt @@ -2,6 +2,7 @@ add_executable(${IPCP_LOCAL_TARGET} main.c + reg.c ${IPCP_SOURCES} ) diff --git a/src/ipcpd/local/main.c b/src/ipcpd/local/main.c index c0aeb51e..69eac8a6 100644 --- a/src/ipcpd/local/main.c +++ b/src/ipcpd/local/main.c @@ -42,7 +42,7 @@ #include "ipcp.h" #include "np1.h" -#include "shim-data.h" +#include "reg.h" #include <string.h> #include <stdlib.h> @@ -53,14 +53,14 @@ #define THIS_TYPE IPCP_LOCAL struct { - struct shim_data * shim_data; + struct reg * reg; - int in_out[SYS_MAX_FLOWS]; - fset_t * flows; - fqueue_t * fq; + int in_out[SYS_MAX_FLOWS]; + fset_t * flows; + fqueue_t * fq; - pthread_rwlock_t lock; - pthread_t packet_loop; + pthread_rwlock_t lock; + pthread_t packet_loop; } local_data; static int local_data_init(void) @@ -77,9 +77,9 @@ static int local_data_init(void) if (local_data.fq == NULL) goto fail_fqueue; - local_data.shim_data = shim_data_create(); - if (local_data.shim_data == NULL) - goto fail_shim_data; + local_data.reg = reg_create(); + if (local_data.reg == NULL) + goto fail_reg; if (pthread_rwlock_init(&local_data.lock, NULL) < 0) goto fail_rwlock_init; @@ -87,8 +87,8 @@ static int local_data_init(void) return 0; fail_rwlock_init: - shim_data_destroy(local_data.shim_data); - fail_shim_data: + reg_destroy(local_data.reg); + fail_reg: fqueue_destroy(local_data.fq); fail_fqueue: fset_destroy(local_data.flows); @@ -98,7 +98,7 @@ static int local_data_init(void) static void local_data_fini(void){ pthread_rwlock_destroy(&local_data.lock); - shim_data_destroy(local_data.shim_data); + reg_destroy(local_data.reg); fqueue_destroy(local_data.fq); fset_destroy(local_data.flows); } @@ -166,7 +166,7 @@ static int local_ipcp_bootstrap(struct ipcp_config * conf) static int local_ipcp_reg(const uint8_t * hash) { - if (shim_data_reg_add_entry(local_data.shim_data, hash)) { + if (reg_add(local_data.reg, hash) < 0) { log_err("Failed to add " HASH_FMT32 " to local registry.", HASH_VAL32(hash)); return -1; @@ -177,7 +177,7 @@ static int local_ipcp_reg(const uint8_t * hash) static int local_ipcp_unreg(const uint8_t * hash) { - shim_data_reg_del_entry(local_data.shim_data, hash); + reg_del(local_data.reg, hash); log_info("Unregistered " HASH_FMT32 ".", HASH_VAL32(hash)); @@ -188,7 +188,7 @@ static int local_ipcp_query(const uint8_t * hash) { int ret; - ret = (shim_data_reg_has(local_data.shim_data, hash) ? 0 : -1); + ret = (reg_has(local_data.reg, hash) ? 0 : -1); return ret; } diff --git a/src/ipcpd/local/reg.c b/src/ipcpd/local/reg.c new file mode 100644 index 00000000..36f19b16 --- /dev/null +++ b/src/ipcpd/local/reg.c @@ -0,0 +1,217 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Names registered with the local IPCP + * + * 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 200112L +#endif + +#define OUROBOROS_PREFIX "local-reg" + +#include <ouroboros/hash.h> +#include <ouroboros/list.h> +#include <ouroboros/logs.h> + +#include "reg.h" +#include "ipcp.h" + +#include <assert.h> +#include <pthread.h> +#include <stdlib.h> +#include <string.h> + +struct reg_entry { + struct list_head list; + uint8_t * hash; +}; + +struct reg { + struct list_head names; + pthread_rwlock_t lock; +}; + +static struct reg_entry * reg_entry_create(uint8_t * hash) +{ + struct reg_entry * entry; + + entry = malloc(sizeof(*entry)); + if (entry == NULL) + return NULL; + + list_head_init(&entry->list); + + entry->hash = hash; + + return entry; +} + +static void reg_entry_destroy(struct reg_entry * entry) +{ + assert(entry); + + free(entry->hash); + free(entry); +} + +/* Call with the lock held. */ +static struct reg_entry * reg_find(struct reg * reg, + const uint8_t * hash) +{ + struct list_head * p; + + list_for_each(p, ®->names) { + struct reg_entry * e; + + e = list_entry(p, struct reg_entry, list); + if (memcmp(e->hash, hash, ipcp_dir_hash_len()) == 0) + return e; + } + + return NULL; +} + +struct reg * reg_create(void) +{ + struct reg * reg; + + reg = malloc(sizeof(*reg)); + if (reg == NULL) + goto fail_malloc; + + list_head_init(®->names); + + if (pthread_rwlock_init(®->lock, NULL) < 0) + goto fail_lock; + + return reg; + + fail_lock: + free(reg); + fail_malloc: + return NULL; +} + +void reg_destroy(struct reg * reg) +{ + if (reg == NULL) + return; + + pthread_rwlock_wrlock(®->lock); + + while (!list_is_empty(®->names)) { + struct reg_entry * e; + + e = list_first_entry(®->names, struct reg_entry, list); + + list_del(&e->list); + + reg_entry_destroy(e); + } + + pthread_rwlock_unlock(®->lock); + + pthread_rwlock_destroy(®->lock); + + free(reg); +} + +int reg_add(struct reg * reg, + const uint8_t * hash) +{ + struct reg_entry * entry; + uint8_t * dup; + + assert(reg); + assert(hash); + + pthread_rwlock_wrlock(®->lock); + + if (reg_find(reg, hash) != NULL) { + pthread_rwlock_unlock(®->lock); + log_dbg(HASH_FMT32 " was already registered.", + HASH_VAL32(hash)); + return 0; + } + + dup = ipcp_hash_dup(hash); + if (dup == NULL) + goto fail; + + entry = reg_entry_create(dup); + if (entry == NULL) { + free(dup); + goto fail; + } + + list_add(&entry->list, ®->names); + + pthread_rwlock_unlock(®->lock); + + return 0; + + fail: + pthread_rwlock_unlock(®->lock); + return -1; +} + +int reg_del(struct reg * reg, + const uint8_t * hash) +{ + struct reg_entry * e; + + if (reg == NULL) + return -1; + + pthread_rwlock_wrlock(®->lock); + + e = reg_find(reg, hash); + if (e == NULL) { + pthread_rwlock_unlock(®->lock); + return 0; /* nothing to do */ + } + + list_del(&e->list); + + pthread_rwlock_unlock(®->lock); + + reg_entry_destroy(e); + + return 0; +} + +bool reg_has(struct reg * reg, + const uint8_t * hash) +{ + bool ret; + + assert(reg); + assert(hash); + + pthread_rwlock_rdlock(®->lock); + + ret = reg_find(reg, hash) != NULL; + + pthread_rwlock_unlock(®->lock); + + return ret; +} diff --git a/src/ipcpd/eth/llc.c b/src/ipcpd/local/reg.h index a772e86e..2c6142bb 100644 --- a/src/ipcpd/eth/llc.c +++ b/src/ipcpd/local/reg.h @@ -1,7 +1,7 @@ /* * Ouroboros - Copyright (C) 2016 - 2026 * - * IPC processes over Ethernet - LLC + * Names registered with the local IPCP * * Dimitri Staessens <dimitri@ouroboros.rocks> * Sander Vrijders <sander@ouroboros.rocks> @@ -20,7 +20,26 @@ * Foundation, Inc., http://www.fsf.org/about/contact/. */ -#define BUILD_ETH_LLC -#define OUROBOROS_PREFIX "ipcpd/eth-llc" +#ifndef OUROBOROS_IPCPD_LOCAL_REG_H +#define OUROBOROS_IPCPD_LOCAL_REG_H -#include "eth.c" +#include <stdbool.h> +#include <stdint.h> + +/* The hashes of the names registered with this IPCP. */ +struct reg; + +struct reg * reg_create(void); + +void reg_destroy(struct reg * reg); + +int reg_add(struct reg * reg, + const uint8_t * hash); + +int reg_del(struct reg * reg, + const uint8_t * hash); + +bool reg_has(struct reg * reg, + const uint8_t * hash); + +#endif /* OUROBOROS_IPCPD_LOCAL_REG_H */ diff --git a/src/ipcpd/shim-data.c b/src/ipcpd/shim-data.c deleted file mode 100644 index 90a676da..00000000 --- a/src/ipcpd/shim-data.c +++ /dev/null @@ -1,582 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC process utilities - * - * 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 200112L -#endif - -#include "config.h" - -#define OUROBOROS_PREFIX "shim-data" - -#include <ouroboros/errno.h> -#include <ouroboros/hash.h> -#include <ouroboros/list.h> -#include <ouroboros/logs.h> -#include <ouroboros/time.h> - -#include "shim-data.h" -#include "ipcp.h" - -#include <assert.h> -#include <stdlib.h> -#include <string.h> - -struct reg_entry { - struct list_head list; - uint8_t * hash; -}; - -struct dir_entry { - struct list_head list; - uint8_t * hash; - struct addr addr; -}; - -static void destroy_dir_query(struct dir_query * query) -{ - assert(query); - - pthread_mutex_lock(&query->lock); - - switch (query->state) { - case QUERY_INIT: - query->state = QUERY_DONE; - break; - case QUERY_PENDING: - query->state = QUERY_DESTROY; - pthread_cond_broadcast(&query->cond); - break; - case QUERY_RESPONSE: - case QUERY_DONE: - break; - case QUERY_DESTROY: - pthread_mutex_unlock(&query->lock); - return; - default: - assert(false); - return; - } - - while (query->state != QUERY_DONE) - pthread_cond_wait(&query->cond, &query->lock); - - pthread_mutex_unlock(&query->lock); - - pthread_cond_destroy(&query->cond); - pthread_mutex_destroy(&query->lock); - - free(query->hash); - free(query); -} - -static struct reg_entry * reg_entry_create(uint8_t * hash) -{ - struct reg_entry * entry = malloc(sizeof(*entry)); - if (entry == NULL) - return NULL; - - assert(hash); - - entry->hash = hash; - - return entry; -} - -static void reg_entry_destroy(struct reg_entry * entry) -{ - assert(entry); - - free(entry->hash); - free(entry); -} - -static struct dir_entry * dir_entry_create(uint8_t * hash, - struct addr addr) -{ - struct dir_entry * entry = malloc(sizeof(*entry)); - if (entry == NULL) - return NULL; - - assert(hash); - - entry->addr = addr; - entry->hash = hash; - - return entry; -} - -static void dir_entry_destroy(struct dir_entry * entry) -{ - assert(entry); - - free(entry->hash); - free(entry); -} - -struct shim_data * shim_data_create(void) -{ - struct shim_data * sd; - - sd = malloc(sizeof(*sd)); - if (sd == NULL) - return NULL; - - /* init the lists */ - list_head_init(&sd->registry); - list_head_init(&sd->directory); - list_head_init(&sd->dir_queries); - - /* init the locks */ - if (pthread_rwlock_init(&sd->reg_lock, NULL) < 0) - goto fail_reg_lock_init; - - if (pthread_rwlock_init(&sd->dir_lock, NULL) < 0) - goto fail_dir_lock_init; - - if (pthread_mutex_init(&sd->dir_queries_lock, NULL) < 0) - goto fail_mutex_init; - - return sd; - - fail_mutex_init: - pthread_rwlock_destroy(&sd->dir_lock); - fail_dir_lock_init: - pthread_rwlock_destroy(&sd->reg_lock); - fail_reg_lock_init: - return NULL; -} - -static void clear_registry(struct shim_data * data) -{ - struct list_head * h; - struct list_head * t; - - assert(data); - - list_for_each_safe(h, t, &data->registry) { - struct reg_entry * e = list_entry(h, struct reg_entry, list); - list_del(&e->list); - reg_entry_destroy(e); - } -} - -static void clear_directory(struct shim_data * data) -{ - struct list_head * h; - struct list_head * t; - - assert(data); - - list_for_each_safe(h, t, &data->directory) { - struct dir_entry * e = list_entry(h, struct dir_entry, list); - list_del(&e->list); - dir_entry_destroy(e); - } -} - -static void clear_dir_queries(struct shim_data * data) -{ - struct list_head * h; - struct list_head * t; - - assert(data); - - list_for_each_safe(h, t, &data->dir_queries) { - struct dir_query * e = list_entry(h, struct dir_query, next); - list_del(&e->next); - destroy_dir_query(e); - } -} - -void shim_data_destroy(struct shim_data * data) -{ - if (data == NULL) - return; - - /* clear the lists */ - pthread_rwlock_wrlock(&data->reg_lock); - clear_registry(data); - pthread_rwlock_unlock(&data->reg_lock); - - pthread_rwlock_wrlock(&data->dir_lock); - clear_directory(data); - pthread_rwlock_unlock(&data->dir_lock); - - pthread_mutex_lock(&data->dir_queries_lock); - clear_dir_queries(data); - pthread_mutex_unlock(&data->dir_queries_lock); - - pthread_rwlock_destroy(&data->dir_lock); - pthread_rwlock_destroy(&data->reg_lock); - pthread_mutex_destroy(&data->dir_queries_lock); - - free(data); -} - -static struct reg_entry * find_reg_entry_by_hash(struct shim_data * data, - const uint8_t * hash) -{ - struct list_head * h; - - assert(data); - assert(hash); - - list_for_each(h, &data->registry) { - struct reg_entry * e = list_entry(h, struct reg_entry, list); - if (!memcmp(e->hash, hash, ipcp_dir_hash_len())) - return e; - } - - return NULL; -} - -static struct dir_entry * find_dir_entry(struct shim_data * data, - const uint8_t * hash, - struct addr addr) -{ - struct list_head * h; - list_for_each(h, &data->directory) { - struct dir_entry * e = list_entry(h, struct dir_entry, list); - if (memcmp(&e->addr, &addr, sizeof(addr)) != 0) - continue; - - if (memcmp(e->hash, hash, ipcp_dir_hash_len()) == 0) - return e; - } - - return NULL; -} - -static struct dir_entry * find_dir_entry_any(struct shim_data * data, - const uint8_t * hash) -{ - struct list_head * h; - list_for_each(h, &data->directory) { - struct dir_entry * e = list_entry(h, struct dir_entry, list); - if (!memcmp(e->hash, hash, ipcp_dir_hash_len())) - return e; - } - - return NULL; -} - -int shim_data_reg_add_entry(struct shim_data * data, - const uint8_t * hash) -{ - struct reg_entry * entry; - uint8_t * hash_dup; - - assert(data); - assert(hash); - - pthread_rwlock_wrlock(&data->reg_lock); - - if (find_reg_entry_by_hash(data, hash)) { - pthread_rwlock_unlock(&data->reg_lock); - log_dbg(HASH_FMT32 " was already in the directory.", - HASH_VAL32(hash)); - return 0; - } - - hash_dup = ipcp_hash_dup(hash); - if (hash_dup == NULL) { - pthread_rwlock_unlock(&data->reg_lock); - return -1; - } - - entry = reg_entry_create(hash_dup); - if (entry == NULL) { - pthread_rwlock_unlock(&data->reg_lock); - return -1; - } - - list_add(&entry->list, &data->registry); - - pthread_rwlock_unlock(&data->reg_lock); - - return 0; -} - -int shim_data_reg_del_entry(struct shim_data * data, - const uint8_t * hash) -{ - struct reg_entry * e; - if (data == NULL) - return -1; - - pthread_rwlock_wrlock(&data->reg_lock); - - e = find_reg_entry_by_hash(data, hash); - if (e == NULL) { - pthread_rwlock_unlock(&data->reg_lock); - return 0; /* nothing to do */ - } - - list_del(&e->list); - - pthread_rwlock_unlock(&data->reg_lock); - - reg_entry_destroy(e); - - return 0; -} - -bool shim_data_reg_has(struct shim_data * data, - const uint8_t * hash) -{ - bool ret = false; - - assert(data); - assert(hash); - - pthread_rwlock_rdlock(&data->reg_lock); - - ret = (find_reg_entry_by_hash(data, hash) != NULL); - - pthread_rwlock_unlock(&data->reg_lock); - - return ret; -} - -int shim_data_dir_add_entry(struct shim_data * data, - const uint8_t * hash, - struct addr addr) -{ - struct dir_entry * entry; - uint8_t * entry_hash; - - assert(data); - assert(hash); - - pthread_rwlock_wrlock(&data->dir_lock); - - if (find_dir_entry(data, hash, addr) != NULL) { - pthread_rwlock_unlock(&data->dir_lock); - return -1; - } - - entry_hash = ipcp_hash_dup(hash); - if (entry_hash == NULL) { - pthread_rwlock_unlock(&data->dir_lock); - return -1; - } - - entry = dir_entry_create(entry_hash, addr); - if (entry == NULL) { - pthread_rwlock_unlock(&data->dir_lock); - return -1; - } - - list_add(&entry->list,&data->directory); - - pthread_rwlock_unlock(&data->dir_lock); - - return 0; -} - -int shim_data_dir_del_entry(struct shim_data * data, - const uint8_t * hash, - struct addr addr) -{ - struct dir_entry * e; - if (data == NULL) - return -1; - - pthread_rwlock_wrlock(&data->dir_lock); - - e = find_dir_entry(data, hash, addr); - if (e == NULL) { - pthread_rwlock_unlock(&data->dir_lock); - return 0; /* nothing to do */ - } - - list_del(&e->list); - - pthread_rwlock_unlock(&data->dir_lock); - - dir_entry_destroy(e); - - return 0; -} - -bool shim_data_dir_has(struct shim_data * data, - const uint8_t * hash) -{ - bool ret = false; - - pthread_rwlock_rdlock(&data->dir_lock); - - ret = (find_dir_entry_any(data, hash) != NULL); - - pthread_rwlock_unlock(&data->dir_lock); - - return ret; -} - -struct addr shim_data_dir_get_addr(struct shim_data * data, - const uint8_t * hash) -{ - struct dir_entry * entry; - struct addr addr = {0}; - - pthread_rwlock_rdlock(&data->dir_lock); - - entry = find_dir_entry_any(data, hash); - if (entry == NULL) { - pthread_rwlock_unlock(&data->dir_lock); - log_warn("No address for " HASH_FMT32 ".", HASH_VAL32(hash)); - return addr; /* undefined behaviour, 0 may be a valid address */ - } - - addr = entry->addr; - - pthread_rwlock_unlock(&data->dir_lock); - - return addr; -} - -struct dir_query * shim_data_dir_query_create(struct shim_data * data, - const uint8_t * hash) -{ - struct dir_query * query; - pthread_condattr_t cattr; - - query = malloc(sizeof(*query)); - if (query == NULL) - return NULL; - - query->hash = ipcp_hash_dup(hash); - if (query->hash == NULL) { - free(query); - return NULL; - } - - query->state = QUERY_INIT; - - pthread_condattr_init(&cattr); -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - pthread_cond_init(&query->cond, &cattr); - pthread_mutex_init(&query->lock, NULL); - - list_head_init(&query->next); - - pthread_mutex_lock(&data->dir_queries_lock); - list_add(&query->next, &data->dir_queries); - pthread_mutex_unlock(&data->dir_queries_lock); - - return query; -} - -void shim_data_dir_query_respond(struct shim_data * data, - const uint8_t * hash) -{ - struct dir_query * e = NULL; - struct list_head * pos; - bool found = false; - - pthread_mutex_lock(&data->dir_queries_lock); - - list_for_each(pos, &data->dir_queries) { - e = list_entry(pos, struct dir_query, next); - - if (memcmp(e->hash, hash, ipcp_dir_hash_len()) == 0) { - found = true; - break; - } - } - - if (!found) { - pthread_mutex_unlock(&data->dir_queries_lock); - return; - } - - pthread_mutex_lock(&e->lock); - - if (e->state != QUERY_PENDING) { - pthread_mutex_unlock(&e->lock); - pthread_mutex_unlock(&data->dir_queries_lock); - return; - } - - e->state = QUERY_RESPONSE; - pthread_cond_broadcast(&e->cond); - - while (e->state == QUERY_RESPONSE) - pthread_cond_wait(&e->cond, &e->lock); - - pthread_mutex_unlock(&e->lock); - - pthread_mutex_unlock(&data->dir_queries_lock); -} - -void shim_data_dir_query_destroy(struct shim_data * data, - struct dir_query * query) -{ - pthread_mutex_lock(&data->dir_queries_lock); - - list_del(&query->next); - destroy_dir_query(query); - - pthread_mutex_unlock(&data->dir_queries_lock); -} - -int shim_data_dir_query_wait(struct dir_query * query, - const struct timespec * timeout) -{ - struct timespec abstime; - int ret = 0; - - assert(query); - assert(timeout); - - clock_gettime(PTHREAD_COND_CLOCK, &abstime); - ts_add(&abstime, timeout, &abstime); - - pthread_mutex_lock(&query->lock); - - if (query->state != QUERY_INIT) { - pthread_mutex_unlock(&query->lock); - return -EINVAL; - } - - query->state = QUERY_PENDING; - - while (query->state == QUERY_PENDING && ret != -ETIMEDOUT) - ret = -pthread_cond_timedwait(&query->cond, - &query->lock, - &abstime); - - if (query->state == QUERY_DESTROY) - ret = -1; - - query->state = QUERY_DONE; - pthread_cond_broadcast(&query->cond); - - pthread_mutex_unlock(&query->lock); - - return ret; -} diff --git a/src/ipcpd/shim-data.h b/src/ipcpd/shim-data.h deleted file mode 100644 index fbadb4d4..00000000 --- a/src/ipcpd/shim-data.h +++ /dev/null @@ -1,109 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * Utitilies for building IPC processes - * - * 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_IPCP_DATA_H -#define OUROBOROS_IPCPD_IPCP_DATA_H - -#include <ouroboros/list.h> - -#include <pthread.h> -#include <stdint.h> -#include <netinet/in.h> -#include <sys/types.h> - -#define MAC_SIZE 6 - -enum dir_query_state { - QUERY_INIT = 0, - QUERY_PENDING, - QUERY_RESPONSE, - QUERY_DONE, - QUERY_DESTROY -}; - -struct dir_query { - struct list_head next; - uint8_t * hash; - enum dir_query_state state; - - pthread_mutex_t lock; - pthread_cond_t cond; -}; - -struct addr { - union { - uint8_t mac[MAC_SIZE]; - struct in_addr ip4; - struct in6_addr ip6; - }; -}; - -struct shim_data { - struct list_head registry; - pthread_rwlock_t reg_lock; - - struct list_head directory; - pthread_rwlock_t dir_lock; - - struct list_head dir_queries; - pthread_mutex_t dir_queries_lock; -}; - -struct shim_data * shim_data_create(void); - -void shim_data_destroy(struct shim_data * data); - -int shim_data_reg_add_entry(struct shim_data * data, - const uint8_t * hash); - -int shim_data_reg_del_entry(struct shim_data * data, - const uint8_t * hash); - -bool shim_data_reg_has(struct shim_data * data, - const uint8_t * hash); - -int shim_data_dir_add_entry(struct shim_data * data, - const uint8_t * hash, - struct addr addr); - -int shim_data_dir_del_entry(struct shim_data * data, - const uint8_t * hash, - struct addr addr); - -bool shim_data_dir_has(struct shim_data * data, - const uint8_t * hash); - -struct addr shim_data_dir_get_addr(struct shim_data * data, - const uint8_t * hash); - -struct dir_query * shim_data_dir_query_create(struct shim_data * data, - const uint8_t * hash); - -void shim_data_dir_query_destroy(struct shim_data * data, - struct dir_query * query); - -void shim_data_dir_query_respond(struct shim_data * data, - const uint8_t * hash); - -int shim_data_dir_query_wait(struct dir_query * query, - const struct timespec * timeout); -#endif /* OUROBOROS_IPCPD_SHIM_DATA_H */ diff --git a/src/ipcpd/udp/CMakeLists.txt b/src/ipcpd/udp/CMakeLists.txt deleted file mode 100644 index a98f0919..00000000 --- a/src/ipcpd/udp/CMakeLists.txt +++ /dev/null @@ -1,14 +0,0 @@ -# UDP IPCPs build configuration (UDP4 and UDP6) -# DDNS detection is in cmake/dependencies/udp/ddns.cmake - -add_executable(${IPCP_UDP4_TARGET} udp4.c ${IPCP_SOURCES}) -add_executable(${IPCP_UDP6_TARGET} udp6.c ${IPCP_SOURCES}) - -foreach(target ${IPCP_UDP4_TARGET} ${IPCP_UDP6_TARGET}) - target_include_directories(${target} PRIVATE ${IPCP_INCLUDE_DIRS}) - target_link_libraries(${target} PRIVATE ouroboros-dev) - ouroboros_target_debug_definitions(${target}) -endforeach() - -install(TARGETS ${IPCP_UDP4_TARGET} ${IPCP_UDP6_TARGET} - RUNTIME DESTINATION ${CMAKE_INSTALL_SBINDIR}) diff --git a/src/ipcpd/udp/udp.c b/src/ipcpd/udp/udp.c deleted file mode 100644 index db57e2f4..00000000 --- a/src/ipcpd/udp/udp.c +++ /dev/null @@ -1,1333 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC process over UDP - * - * 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 <ouroboros/bitmap.h> -#include <ouroboros/endian.h> -#include <ouroboros/hash.h> -#include <ouroboros/list.h> -#include <ouroboros/utils.h> -#include <ouroboros/dev.h> -#include <ouroboros/ipcp-dev.h> -#include <ouroboros/np1_flow.h> -#include <ouroboros/fqueue.h> -#include <ouroboros/errno.h> -#include <ouroboros/logs.h> -#include <ouroboros/pthread.h> - -#include "ipcp.h" -#include "np1.h" -#include "shim-data.h" - -#include <string.h> -#include <sys/socket.h> -#include <sys/select.h> -#include <arpa/inet.h> -#include <netdb.h> -#include <netinet/in.h> -#include <signal.h> -#include <stdlib.h> -#include <sys/wait.h> -#include <fcntl.h> -#include <unistd.h> -#if defined(__linux__) -#include <netinet/ip.h> -#endif - -#define FLOW_REQ 1 -#define FLOW_REPLY 2 -#define FLOW_IRM_UPDATE 3 - -#define OUR_HEADER_LEN sizeof(uint32_t) /* adds eid */ - -#define IPCP_UDP_BUF_SIZE IPCP_UDP_MAX_PACKET_SIZE -#define IPCP_UDP_MSG_SIZE IPCP_UDP_MAX_PACKET_SIZE - -#define DNS_TTL 86400 - -#define SADDR ((struct sockaddr *) &udp_data.s_saddr) -#define SADDR_SIZE (sizeof(udp_data.s_saddr)) -#define LOCAL_IP (udp_data.s_saddr.sin_addr.s_addr) - -#define MGMT_EID 0 -#define MGMT_FRAME_SIZE (sizeof(struct mgmt_msg)) -#define MGMT_FRAME_BUF_SIZE 2048 - -#ifdef __linux__ -#define SENDTO_FLAGS MSG_CONFIRM -#else -#define SENDTO_FLAGS 0 -#endif - -/* Keep order for alignment. */ -struct mgmt_msg { - uint32_t eid; - uint32_t s_eid; - uint32_t d_eid; - int32_t response; - uint64_t bandwidth; - uint32_t delay; - uint32_t loss; - uint32_t ber; - uint32_t max_gap; - uint32_t timeout; - uint8_t code; - /* QoS parameters from spec */ - uint8_t availability; - uint8_t service; -} __attribute__((packed)); - -struct mgmt_frame { - struct list_head next; - struct __SOCKADDR r_saddr; - uint8_t buf[MGMT_FRAME_BUF_SIZE]; - size_t len; -}; - -/* UDP flow */ -struct uf { - int d_eid; - struct __SOCKADDR r_saddr; -}; - -struct { - struct shim_data * shim_data; - - struct __ADDR dns_addr; - struct __SOCKADDR s_saddr; - int s_fd; - - fset_t * np1_flows; - struct uf fd_to_uf[SYS_MAX_FLOWS]; - pthread_rwlock_t flows_lock; - - pthread_t packet_writer[IPCP_UDP_WR_THR]; - pthread_t packet_reader[IPCP_UDP_RD_THR]; - - /* Handle mgmt frames in a different thread */ - pthread_t mgmt_handler; - pthread_mutex_t mgmt_lock; - pthread_cond_t mgmt_cond; - struct list_head mgmt_frames; -} udp_data; - -static const char * __inet_ntop(const struct __ADDR * addr, - char * buf) -{ - return inet_ntop(__AF, addr, buf, __ADDRSTRLEN); -} - -#if defined(BUILD_IPCP_UDP4) -#define UDP_MTU_FALLBACK IPCP_UDP4_MTU -#define UDP_IP_OVERHEAD 28U /* IPv4 + UDP */ -#else -#define UDP_MTU_FALLBACK IPCP_UDP6_MTU -#define UDP_IP_OVERHEAD 48U /* IPv6 + UDP */ -#endif - -static uint32_t udp_query_mtu(const struct __SOCKADDR * saddr) -{ -#if defined(__linux__) && (defined(IP_MTU) || defined(IPV6_MTU)) - int sock; - int mtu = 0; - socklen_t len = sizeof(mtu); - - sock = socket(__AF, SOCK_DGRAM, IPPROTO_UDP); - if (sock < 0) - return UDP_MTU_FALLBACK; - - if (connect(sock, (const struct sockaddr *) saddr, - sizeof(*saddr)) < 0) - goto fallback; - -#if defined(BUILD_IPCP_UDP4) && defined(IP_MTU) - if (getsockopt(sock, IPPROTO_IP, IP_MTU, &mtu, &len) < 0) - goto fallback; -#elif defined(BUILD_IPCP_UDP6) && defined(IPV6_MTU) - if (getsockopt(sock, IPPROTO_IPV6, IPV6_MTU, &mtu, &len) < 0) - goto fallback; -#else - goto fallback; -#endif - close(sock); - - if (mtu <= (int) UDP_IP_OVERHEAD) - return UDP_MTU_FALLBACK; - - return (uint32_t) mtu - UDP_IP_OVERHEAD; - - fallback: - close(sock); -#else - (void) saddr; -#endif - return UDP_MTU_FALLBACK; -} - -static int udp_data_init(void) -{ - int i; - pthread_condattr_t cattr; - - if (pthread_rwlock_init(&udp_data.flows_lock, NULL)) - goto fail_rwlock_init; - - if (pthread_condattr_init(&cattr)) - goto fail_condattr; -#ifndef __APPLE__ - pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); -#endif - if (pthread_cond_init(&udp_data.mgmt_cond, &cattr)) - goto fail_mgmt_cond; - - if (pthread_mutex_init(&udp_data.mgmt_lock, NULL)) - goto fail_mgmt_lock; - - for (i = 0; i < SYS_MAX_FLOWS; ++i) - udp_data.fd_to_uf[i].d_eid = -1; - - udp_data.np1_flows = fset_create(); - if (udp_data.np1_flows == NULL) - goto fail_fset; - - udp_data.shim_data = shim_data_create(); - if (udp_data.shim_data == NULL) - goto fail_data; - - pthread_condattr_destroy(&cattr); - - list_head_init(&udp_data.mgmt_frames); - - return 0; - - fail_data: - fset_destroy(udp_data.np1_flows); - fail_fset: - pthread_mutex_destroy(&udp_data.mgmt_lock); - fail_mgmt_lock: - pthread_cond_destroy(&udp_data.mgmt_cond); - fail_mgmt_cond: - pthread_condattr_destroy(&cattr); - fail_condattr: - pthread_rwlock_destroy(&udp_data.flows_lock); - fail_rwlock_init: - return -1; -} - -static void udp_data_fini(void) -{ - shim_data_destroy(udp_data.shim_data); - - fset_destroy(udp_data.np1_flows); - - pthread_rwlock_destroy(&udp_data.flows_lock); - pthread_cond_destroy(&udp_data.mgmt_cond); - pthread_mutex_destroy(&udp_data.mgmt_lock); -} - -static int udp_ipcp_port_alloc(const struct __SOCKADDR * r_saddr, - uint32_t s_eid, - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) -{ - uint8_t * buf; - struct mgmt_msg * msg; - size_t len; - - assert(data->len > 0 ? data->data != NULL : data->data == NULL); - - len = sizeof(*msg) + ipcp_dir_hash_len(); - - buf = malloc(len + data->len); - if (buf == NULL) - return -1; - - memset(buf, 0, len + data->len); - - msg = (struct mgmt_msg *) buf; - msg->eid = hton32(MGMT_EID); - msg->code = FLOW_REQ; - msg->s_eid = hton32(s_eid); - 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->service = qs.service; - msg->max_gap = hton32(qs.max_gap); - msg->timeout = hton32(qs.timeout); - - memcpy(msg + 1, dst, ipcp_dir_hash_len()); - if (data->len > 0) - memcpy(buf + len, data->data, data->len); - - if (sendto(udp_data.s_fd, msg, len + data->len, - SENDTO_FLAGS, - (const struct sockaddr *) r_saddr, sizeof(*r_saddr)) < 0) { - log_err("Failed to send flow allocation request: %s.", - strerror(errno)); - free(buf); - return -1; - } - - free(buf); - - return 0; -} - -static int udp_ipcp_port_alloc_resp(const struct __SOCKADDR * r_saddr, - uint32_t s_eid, - uint32_t d_eid, - int32_t response, - const buffer_t * data) -{ - struct mgmt_msg * msg; - - msg = malloc(sizeof(*msg) + data->len); - if (msg == NULL) - return -1; - - memset(msg, 0, sizeof(*msg) + data->len); - - msg->eid = hton32(MGMT_EID); - msg->code = FLOW_REPLY; - msg->s_eid = hton32(s_eid); - msg->d_eid = hton32(d_eid); - msg->response = hton32(response); - - if (data->len > 0) - memcpy(msg + 1, data->data, data->len); - - if (sendto(udp_data.s_fd, msg, sizeof(*msg) + data->len, - SENDTO_FLAGS, - (const struct sockaddr *) r_saddr, sizeof(*r_saddr)) < 0 ) { - free(msg); - return -1; - } - - free(msg); - - return 0; -} - -static int udp_ipcp_flow_update(int fd, - const buffer_t * data) -{ - struct mgmt_msg * msg; - struct __SOCKADDR r_saddr; - uint32_t d_eid; - - msg = malloc(sizeof(*msg) + data->len); - if (msg == NULL) - return -1; - - memset(msg, 0, sizeof(*msg) + data->len); - - pthread_rwlock_rdlock(&udp_data.flows_lock); - - r_saddr = udp_data.fd_to_uf[fd].r_saddr; - d_eid = (uint32_t) udp_data.fd_to_uf[fd].d_eid; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - msg->eid = hton32(MGMT_EID); - msg->code = FLOW_IRM_UPDATE; - msg->s_eid = hton32(d_eid); - msg->d_eid = hton32((uint32_t) fd); - - if (data->len > 0) - memcpy(msg + 1, data->data, data->len); - - if (sendto(udp_data.s_fd, msg, sizeof(*msg) + data->len, - SENDTO_FLAGS, - (const struct sockaddr *) &r_saddr, - sizeof(r_saddr)) < 0) { - log_err("Failed to send flow update: %s.", strerror(errno)); - free(msg); - return -1; - } - - free(msg); - - return 0; -} - -static int udp_ipcp_port_req(struct __SOCKADDR * c_saddr, - int d_eid, - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) -{ - int fd; - - fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UDP_MPL, - udp_query_mtu(c_saddr), data); - if (fd < 0) { - log_err("Could not get new flow from IRMd."); - return -1; - } - - pthread_rwlock_wrlock(&udp_data.flows_lock); - - udp_data.fd_to_uf[fd].r_saddr = *c_saddr; - udp_data.fd_to_uf[fd].d_eid = d_eid; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - log_dbg("Pending allocation request, fd %d, remote eid %d.", - fd, d_eid); - - return 0; -} - -static int udp_ipcp_port_alloc_reply(const struct __SOCKADDR * saddr, - uint32_t s_eid, - uint32_t d_eid, - int32_t response, - const buffer_t * data) -{ - time_t mpl = IPCP_UDP_MPL; - - pthread_rwlock_wrlock(&udp_data.flows_lock); - - if (memcmp(&udp_data.fd_to_uf[s_eid].r_saddr, saddr, sizeof(*saddr))) { - char ipstr[__ADDRSTRLEN]; - pthread_rwlock_unlock(&udp_data.flows_lock); - #ifdef BUILD_IPCP_UDP4 - __inet_ntop(&saddr->sin_addr, ipstr); - #else - __inet_ntop(&saddr->sin6_addr, ipstr); - #endif - log_err("Flow allocation reply for %u from wrong source %s.", - s_eid, ipstr); - return -1; - } - - if (response == 0) - udp_data.fd_to_uf[s_eid].d_eid = d_eid; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - if (ipcp_flow_alloc_reply(s_eid, response, mpl, - udp_query_mtu(saddr), data) < 0) { - log_err("Failed to reply to flow allocation."); - return -1; - } - - log_dbg("Flow allocation completed on eids (%d, %d).", - s_eid, d_eid); - - return 0; -} - -static int udp_ipcp_flow_update_arr(const uint8_t * buf, - size_t len) -{ - struct mgmt_msg * msg; - buffer_t data; - int fd; - int flow_id; - - msg = (struct mgmt_msg *) buf; - - fd = (int) ntoh32(msg->s_eid); - if (fd < 0 || fd >= SYS_MAX_FLOWS) { - log_err("Flow update for invalid eid %d.", fd); - return -1; - } - - data.len = len - sizeof(*msg); - data.data = (uint8_t *) buf + sizeof(*msg); - - flow_id = np1_flow_id(fd); - if (flow_id < 0) - return -1; - - if (ipcp_flow_update_arr(flow_id, &data) < 0) { - log_err("Failed to relay flow update on fd %d.", fd); - return -1; - } - - return 0; -} - -static int udp_ipcp_mgmt_frame(struct __SOCKADDR c_saddr, - const uint8_t * buf, - size_t len) -{ - struct mgmt_msg * msg; - size_t msg_len; - qosspec_t qs; - buffer_t data; - - /* Defence against malformed/corrupted wire input. */ - if (len < sizeof(*msg)) - return -1; - - msg = (struct mgmt_msg *) buf; - - switch (msg->code) { - case FLOW_REQ: - msg_len = sizeof(*msg) + ipcp_dir_hash_len(); - - if (len < msg_len) - return -1; - - data.len = len - msg_len; - data.data = (uint8_t *) buf + 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); - - return udp_ipcp_port_req(&c_saddr, ntoh32(msg->s_eid), - (uint8_t *) (msg + 1), qs, - &data); - case FLOW_REPLY: - data.len = len - sizeof(*msg); - data.data = (uint8_t *) buf + sizeof(*msg); - - return udp_ipcp_port_alloc_reply(&c_saddr, - ntoh32(msg->s_eid), - ntoh32(msg->d_eid), - ntoh32(msg->response), - &data); - case FLOW_IRM_UPDATE: - return udp_ipcp_flow_update_arr(buf, len); - default: - log_err("Unknown message received %d.", msg->code); - return -1; - } -} - -static void * udp_ipcp_mgmt_handler(void * o) -{ - (void) o; - - pthread_cleanup_push(__cleanup_mutex_unlock, &udp_data.mgmt_lock); - - while (true) { - struct mgmt_frame * frame; - - pthread_mutex_lock(&udp_data.mgmt_lock); - - while (list_is_empty(&udp_data.mgmt_frames)) - pthread_cond_wait(&udp_data.mgmt_cond, - &udp_data.mgmt_lock); - - frame = list_first_entry((&udp_data.mgmt_frames), - struct mgmt_frame, next); - assert(frame != NULL); - list_del(&frame->next); - - pthread_mutex_unlock(&udp_data.mgmt_lock); - - udp_ipcp_mgmt_frame(frame->r_saddr, frame->buf, frame->len); - - free(frame); - } - - pthread_cleanup_pop(false); - - return (void *) 0; -} - -static void * udp_ipcp_packet_reader(void * o) -{ - uint8_t buf[IPCP_UDP_MAX_PACKET_SIZE]; - uint8_t * data; - ssize_t n; - uint32_t eid; - uint32_t * eid_p; - - (void) o; - - ipcp_lock_to_core(); - - data = buf + sizeof(uint32_t); - eid_p = (uint32_t *) buf; - - while (true) { - struct mgmt_frame * frame; - struct __SOCKADDR r_saddr; - socklen_t len; - struct ssm_pk_buff * spb; - uint8_t * head; - - len = sizeof(r_saddr); - - n = recvfrom(udp_data.s_fd, buf, IPCP_UDP_MAX_PACKET_SIZE, 0, - (struct sockaddr *) &r_saddr, &len); - if (n < 0) - continue; - - if (n == 0) - log_dbg("Got a 0 frame."); - - if ((size_t) n < sizeof(eid)) { - log_dbg("Dropped bad frame."); - continue; - } - - eid = ntoh32(*eid_p); - - /* pass onto mgmt queue */ - if (eid == MGMT_EID) { - if ((size_t) n < MGMT_FRAME_SIZE) { - log_warn("Dropped runt mgmt frame."); - continue; - } - - frame = malloc(sizeof(*frame)); - if (frame == NULL) - continue; - - memcpy(frame->buf, buf, n); - memcpy(&frame->r_saddr, &r_saddr, sizeof(r_saddr)); - frame->len = n; - - pthread_mutex_lock(&udp_data.mgmt_lock); - list_add(&frame->next, &udp_data.mgmt_frames); - pthread_cond_signal(&udp_data.mgmt_cond); - pthread_mutex_unlock(&udp_data.mgmt_lock); - continue; - } - - n-= sizeof(eid); - - if (ipcp_spb_reserve(&spb, n)) - continue; - - head = ssm_pk_buff_head(spb); - memcpy(head, data, n); - if (np1_flow_write(eid, spb, NP1_GET_POOL(eid)) < 0) - ipcp_spb_release(spb); - } - - return (void *) 0; -} - -static void cleanup_fqueue(void * fq) -{ - fqueue_destroy((fqueue_t *) fq); -} - -static void cleanup_spb(void * spb) -{ - ipcp_spb_release((struct ssm_pk_buff *) spb); -} - -static void * udp_ipcp_packet_writer(void * o) -{ - fqueue_t * fq; - - fq = fqueue_create(); - if (fq == NULL) - return (void *) -1; - - (void) o; - - ipcp_lock_to_core(); - - pthread_cleanup_push(cleanup_fqueue, fq); - - while (true) { - struct __SOCKADDR saddr; - int eid; - int fd; - fevent(udp_data.np1_flows, fq, NULL); - while ((fd = fqueue_next(fq)) >= 0) { - struct ssm_pk_buff * spb; - uint8_t * buf; - uint16_t len; - - if (fqueue_type(fq) != FLOW_PKT) - continue; - - if (np1_flow_read(fd, &spb, NP1_GET_POOL(fd))) { - log_dbg("Bad read from fd %d.", fd); - continue; - } - - len = ssm_pk_buff_len(spb); - if (len > IPCP_UDP_MAX_PACKET_SIZE) { - log_dbg("Packet length exceeds MTU."); - ipcp_spb_release(spb); - continue; - } - - buf = ssm_pk_buff_push(spb, OUR_HEADER_LEN); - if (buf == NULL) { - log_dbg("Failed to allocate header."); - ipcp_spb_release(spb); - continue; - } - - pthread_rwlock_rdlock(&udp_data.flows_lock); - - eid = hton32(udp_data.fd_to_uf[fd].d_eid); - saddr = udp_data.fd_to_uf[fd].r_saddr; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - memcpy(buf, &eid, sizeof(eid)); - - pthread_cleanup_push(cleanup_spb, spb); - - if (sendto(udp_data.s_fd, buf, len + OUR_HEADER_LEN, - SENDTO_FLAGS, - (const struct sockaddr *) &saddr, - sizeof(saddr)) < 0) - log_err("Failed to send packet."); - - pthread_cleanup_pop(true); - } - } - - pthread_cleanup_pop(true); - - return (void *) 1; -} - -static bool is_addr_specified(const struct __ADDR * addr) -{ -#ifdef BUILD_IPCP_UDP4 - return addr->s_addr != 0; -#else - return !IN6_IS_ADDR_UNSPECIFIED(addr); -#endif -} - -static int udp_ipcp_bootstrap(struct ipcp_config * conf) -{ - char ipstr[__ADDRSTRLEN]; - char dnsstr[__ADDRSTRLEN]; - int i = 1; -#ifdef BUILD_IPCP_UDP4 - struct udp4_config * udp; - udp = &conf->udp4; -#else - struct udp6_config * udp; - udp = &conf->udp6; -#endif - - assert(conf != NULL); - assert(conf->type == THIS_TYPE); - assert(conf->layer_info.dir_hash_algo == (enum pol_dir_hash) HASH_MD5); - - if (__inet_ntop(&udp->ip_addr, ipstr) == NULL) { - log_err("Failed to convert IP address."); - return -1; - } - - if (is_addr_specified(&udp->dns_addr)) { - if (__inet_ntop(&udp->dns_addr, dnsstr) == NULL) { - log_err("Failed to convert DNS address."); - return -1; - } -#ifndef HAVE_DDNS - log_warn("DNS disabled at compile time, address ignored."); -#endif - } else { - strcpy(dnsstr, "not set"); - } - - /* UDP listen server */ - udp_data.s_fd = socket(__AF, SOCK_DGRAM, IPPROTO_UDP); - if (udp_data.s_fd < 0) { - log_err("Can't create socket: %s", strerror(errno)); - goto fail_socket; - } - - memset((char *) &udp_data.s_saddr, 0, sizeof(udp_data.s_saddr)); -#ifdef BUILD_IPCP_UDP4 - udp_data.s_saddr.sin_family = AF_INET; - udp_data.s_saddr.sin_addr = udp->ip_addr; - udp_data.s_saddr.sin_port = htons(udp->port); -#else - udp_data.s_saddr.sin6_family = AF_INET6; - udp_data.s_saddr.sin6_addr = udp->ip_addr; - udp_data.s_saddr.sin6_port = htons(udp->port); -#endif - if (bind(udp_data.s_fd, SADDR, SADDR_SIZE) < 0) { - log_err("Couldn't bind to %s:%d. %s.", - ipstr, udp->port, strerror(errno)); - goto fail_bind; - } - - udp_data.dns_addr = udp->dns_addr; - - if (pthread_create(&udp_data.mgmt_handler, NULL, - udp_ipcp_mgmt_handler, NULL)) { - log_err("Failed to create management thread."); - goto fail_bind; - } - - for (i = 0; i < IPCP_UDP_RD_THR; ++i) { - if (pthread_create(&udp_data.packet_reader[i], NULL, - udp_ipcp_packet_reader, NULL)) { - log_err("Failed to create reader thread."); - goto fail_packet_reader; - } - } - - for (i = 0; i < IPCP_UDP_WR_THR; ++i) { - if (pthread_create(&udp_data.packet_writer[i], NULL, - udp_ipcp_packet_writer, NULL)) { - log_err("Failed to create writer thread."); - goto fail_packet_writer; - } - } - - log_dbg("Bootstrapped " TYPE_STR " with pid %d.", getpid()); - log_dbg("Bound to IP address %s.", ipstr); - log_dbg("Using port %u.", udp->port); - if (is_addr_specified(&udp_data.dns_addr)) - log_dbg("DNS server address is %s.", dnsstr); - else - log_dbg("DNS server not in use."); - - return 0; - - fail_packet_writer: - while (i-- > 0) { - pthread_cancel(udp_data.packet_writer[i]); - pthread_join(udp_data.packet_writer[i], NULL); - } - i = IPCP_UDP_RD_THR; - fail_packet_reader: - while (i-- > 0) { - pthread_cancel(udp_data.packet_reader[i]); - pthread_join(udp_data.packet_reader[i], NULL); - } - pthread_cancel(udp_data.mgmt_handler); - pthread_join(udp_data.mgmt_handler, NULL); - fail_bind: - close(udp_data.s_fd); - fail_socket: - return -1; -} - -#ifdef HAVE_DDNS -/* FIXME: Dependency on nsupdate to be removed in the end */ -/* NOTE: Disgusted with this crap */ -static int ddns_send(char * cmd) -{ - pid_t pid; - int wstatus; - int pipe_fd[2]; - char * argv[] = {NSUPDATE_EXEC, 0}; - char * envp[] = {0}; - - if (pipe(pipe_fd)) { - log_err("Failed to create pipe: %s.", strerror(errno)); - return -1; - } - - pid = fork(); - if (pid == -1) { - log_err("Failed to fork: %s.", strerror(errno)); - close(pipe_fd[0]); - close(pipe_fd[1]); - return -1; - } - - if (pid == 0) { - close(pipe_fd[1]); - dup2(pipe_fd[0], 0); - execve(argv[0], &argv[0], envp); - log_err("Failed to execute: %s", strerror(errno)); - exit(1); - } - - close(pipe_fd[0]); - - if (write(pipe_fd[1], cmd, strlen(cmd)) == -1) { - log_err("Failed to communicate with nsupdate: %s.", - strerror(errno)); - close(pipe_fd[1]); - return -1; - } - - waitpid(pid, &wstatus, 0); - if (WIFEXITED(wstatus) && WEXITSTATUS(wstatus) == 0) - log_dbg("Succesfully communicated with DNS server."); - else - log_err("Failed to register with DNS server."); - - close(pipe_fd[1]); - - return 0; -} - -static struct __ADDR ddns_resolve(char * name, - struct __ADDR dns_addr) -{ - pid_t pid = -1; - int wstatus; - int pipe_fd[2]; - char dnsstr[__ADDRSTRLEN]; - char buf[IPCP_UDP_BUF_SIZE]; - ssize_t count = 0; - char * substr = NULL; - char * substr2 = NULL; - char * addr_str = "Address:"; - struct __ADDR ip_addr = __ADDR_ANY_INIT; - - if (__inet_ntop(&dns_addr, dnsstr) == NULL) - return ip_addr; - - if (pipe(pipe_fd)) { - log_err("Failed to create pipe: %s.", strerror(errno)); - return ip_addr; - } - - pid = fork(); - if (pid == -1) { - log_err("Failed to fork: %s.", strerror(errno)); - close(pipe_fd[0]); - close(pipe_fd[1]); - return ip_addr; - } - - if (pid == 0) { - char * argv[] = {NSLOOKUP_EXEC, name, dnsstr, 0}; - char * envp[] = {0}; - - close(pipe_fd[0]); - dup2(pipe_fd[1], 1); - execve(argv[0], &argv[0], envp); - log_err("Failed to execute: %s", strerror(errno)); - exit(1); - } - - close(pipe_fd[1]); - - count = read(pipe_fd[0], buf, IPCP_UDP_BUF_SIZE - 1); - if (count <= 0) { - log_err("Failed to communicate with nslookup."); - close(pipe_fd[0]); - return ip_addr; - } - - close(pipe_fd[0]); - - waitpid(pid, &wstatus, 0); - if (WIFEXITED(wstatus) && WEXITSTATUS(wstatus) == 0 && - count != IPCP_UDP_BUF_SIZE - 1) - log_dbg("Succesfully communicated with nslookup."); - else - log_err("Failed to resolve DNS address."); - - buf[count] = '\0'; - substr = strtok(buf, "\n"); - while (substr != NULL) { - substr2 = substr; - substr = strtok(NULL, "\n"); - } - - if (substr2 == NULL || strstr(substr2, addr_str) == NULL) { - log_err("Failed to resolve DNS address."); - return ip_addr; - } - - if (inet_pton(__AF, substr2 + strlen(addr_str) + 1, &ip_addr) != 1) { - log_err("Failed to resolve DNS address."); - assert(!is_addr_specified(&ip_addr)); - return ip_addr; - } - - return ip_addr; -} -#endif - -static int udp_ipcp_reg(const uint8_t * hash) -{ -#ifdef HAVE_DDNS - char ipstr[__ADDRSTRLEN]; - char dnsstr[__ADDRSTRLEN]; - char cmd[1000]; - struct __ADDR dns_addr; - struct __ADDR ip_addr; -#endif - char * hashstr; - - hashstr = malloc(ipcp_dir_hash_strlen() + 1); - if (hashstr == NULL) { - log_err("Failed to malloc hashstr."); - return -1; - } - - assert(hash); - - ipcp_hash_str(hashstr, hash); - - if (shim_data_reg_add_entry(udp_data.shim_data, hash)) { - log_err("Failed to add " HASH_FMT32 " to local registry.", - HASH_VAL32(hash)); - free(hashstr); - return -1; - } - -#ifdef HAVE_DDNS - /* register application with DNS server */ - - dns_addr = udp_data.dns_addr; - - if (is_addr_specified(&dns_addr)) { -#ifdef BUILD_IPCP_UDP4 - ip_addr = udp_data.s_saddr.sin_addr; -#else - ip_addr = udp_data.s_saddr.sin6_addr; -#endif - if (__inet_ntop(&ip_addr, ipstr) == NULL) { - log_err("Failed to convert IP address to string."); - free(hashstr); - return -1; - } - - if (__inet_ntop(&dns_addr, dnsstr) == NULL) { - log_err("Failed to convert DNS address to string."); - free(hashstr); - return -1; - } - - sprintf(cmd, "server %s\nupdate add %s %d A %s\nsend\nquit\n", - dnsstr, hashstr, DNS_TTL, ipstr); - - if (ddns_send(cmd)) { - log_err("Failed to send DDNS message."); - shim_data_reg_del_entry(udp_data.shim_data, hash); - free(hashstr); - return -1; - } - } -#endif - free(hashstr); - - return 0; -} - -static int udp_ipcp_unreg(const uint8_t * hash) -{ -#ifdef HAVE_DDNS - char dnsstr[__ADDRSTRLEN]; - /* max DNS name length + max IP length + max command length */ - char cmd[100]; - struct __ADDR dns_addr; -#endif - char * hashstr; - - assert(hash); - - hashstr = malloc(ipcp_dir_hash_strlen() + 1); - if (hashstr == NULL) { - log_err("Failed to malloc hashstr."); - return -1; - } - - ipcp_hash_str(hashstr, hash); - -#ifdef HAVE_DDNS - /* unregister application with DNS server */ - - dns_addr = udp_data.dns_addr; - - if (is_addr_specified(&dns_addr)) { - if (__inet_ntop(&dns_addr, dnsstr) == NULL) { - log_err("Failed to convert DNS address to string."); - free(hashstr); - return -1; - } - sprintf(cmd, "server %s\nupdate delete %s A\nsend\nquit\n", - dnsstr, hashstr); - - ddns_send(cmd); - } -#endif - - shim_data_reg_del_entry(udp_data.shim_data, hash); - - free(hashstr); - - return 0; -} - -static int udp_ipcp_query(const uint8_t * hash) -{ - struct addr addr = {}; - char * hashstr; - struct addrinfo hints; - struct addrinfo * ai; -#ifdef HAVE_DDNS - struct __ADDR dns_addr = __ADDR_ANY_INIT; - struct __ADDR ip_addr = __ADDR_ANY_INIT; -#endif - assert(hash); - - hashstr = malloc(ipcp_dir_hash_strlen() + 1); - if (hashstr == NULL) { - log_err("Failed to malloc hashstr."); - return -ENOMEM; - } - - ipcp_hash_str(hashstr, hash); - - if (shim_data_dir_has(udp_data.shim_data, hash)) { - free(hashstr); - return 0; - } - -#ifdef HAVE_DDNS - dns_addr = udp_data.dns_addr; - - if (is_addr_specified(&dns_addr)) { - ip_addr = ddns_resolve(hashstr, dns_addr); - if (!is_addr_specified(&ip_addr)) { - log_err("Could not resolve %s.", hashstr); - free(hashstr); - return -1; - } - } else { -#endif - memset(&hints, 0, sizeof(hints)); - - hints.ai_family = __AF; - if (getaddrinfo(hashstr, NULL, &hints, &ai) != 0) { - log_err("Could not resolve %s: %s.", hashstr, - gai_strerror(errno)); - free(hashstr); - return -1; - } - - if (ai->ai_family != __AF) { - log_err("Wrong addres family for %s.", hashstr); - freeaddrinfo(ai); - free(hashstr); - return -1; - } - - #ifdef BUILD_IPCP_UDP4 - addr.ip4 = ((struct sockaddr_in *) (ai->ai_addr))->sin_addr; - #else - addr.ip6 = ((struct sockaddr_in6 *) (ai->ai_addr))->sin6_addr; - #endif - freeaddrinfo(ai); -#ifdef HAVE_DDNS - } -#endif - if (shim_data_dir_add_entry(udp_data.shim_data, hash, addr)) { - log_err("Failed to add directory entry."); - free(hashstr); - return -1; - } - - free(hashstr); - - return 0; -} - -static int udp_ipcp_flow_alloc(int fd, - const uint8_t * dst, - qosspec_t qs, - const buffer_t * data) -{ - struct __SOCKADDR r_saddr; /* Server address */ - struct __ADDR ip_addr; - struct addr addr; - char ipstr[__ADDRSTRLEN]; - - (void) qs; - - assert(dst); - - if (!shim_data_dir_has(udp_data.shim_data, dst)) { - log_err("Could not resolve destination."); - return -1; - } - - addr = shim_data_dir_get_addr(udp_data.shim_data, dst); -#ifdef BUILD_IPCP_UDP4 - ip_addr = addr.ip4; -#else - ip_addr = addr.ip6; -#endif - if (__inet_ntop(&ip_addr, ipstr) == NULL) { - log_err("Could not convert IP address."); - return -1; - } - - log_dbg("Destination " HASH_FMT32 " resolved at IP %s.", - HASH_VAL32(dst), ipstr); - - memset((char *) &r_saddr, 0, sizeof(r_saddr)); -#ifdef BUILD_IPCP_UDP4 - r_saddr.sin_family = AF_INET; - r_saddr.sin_addr = addr.ip4; - r_saddr.sin_port = udp_data.s_saddr.sin_port; -#else - r_saddr.sin6_family = AF_INET6; - r_saddr.sin6_addr = addr.ip6; - r_saddr.sin6_port = udp_data.s_saddr.sin6_port; -#endif - - if (udp_ipcp_port_alloc(&r_saddr, fd, dst, qs, data) < 0) { - log_err("Could not allocate port."); - return -1; - } - - pthread_rwlock_wrlock(&udp_data.flows_lock); - - udp_data.fd_to_uf[fd].d_eid = -1; - udp_data.fd_to_uf[fd].r_saddr = r_saddr; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - fset_add(udp_data.np1_flows, fd); - - return 0; -} - -static int udp_ipcp_flow_alloc_resp(int fd, - int resp, - const buffer_t * data) -{ - struct __SOCKADDR saddr; - int d_eid; - - if (ipcp_wait_flow_resp(fd) < 0) { - log_err("Failed to wait for flow response."); - return -1; - } - - pthread_rwlock_rdlock(&udp_data.flows_lock); - - saddr = udp_data.fd_to_uf[fd].r_saddr; - d_eid = udp_data.fd_to_uf[fd].d_eid; - - pthread_rwlock_unlock(&udp_data.flows_lock); - - if (udp_ipcp_port_alloc_resp(&saddr, d_eid, fd, resp, data) < 0) { - fset_del(udp_data.np1_flows, fd); - log_err("Failed to respond to flow request."); - return -1; - } - - fset_add(udp_data.np1_flows, fd); - - return 0; -} - -static int udp_ipcp_flow_dealloc(int fd) -{ - ipcp_flow_fini(fd); - - fset_del(udp_data.np1_flows, fd); - - pthread_rwlock_wrlock(&udp_data.flows_lock); - - udp_data.fd_to_uf[fd].d_eid = -1; - memset(&udp_data.fd_to_uf[fd].r_saddr, 0, SADDR_SIZE); - - pthread_rwlock_unlock(&udp_data.flows_lock); - - ipcp_flow_dealloc(fd); - - return 0; -} - -static struct ipcp_ops udp_ops = { - .ipcp_bootstrap = udp_ipcp_bootstrap, - .ipcp_enroll = NULL, - .ipcp_connect = NULL, - .ipcp_disconnect = NULL, - .ipcp_reg = udp_ipcp_reg, - .ipcp_unreg = udp_ipcp_unreg, - .ipcp_query = udp_ipcp_query, - .ipcp_flow_alloc = udp_ipcp_flow_alloc, - .ipcp_flow_join = NULL, - .ipcp_flow_alloc_resp = udp_ipcp_flow_alloc_resp, - .ipcp_flow_dealloc = udp_ipcp_flow_dealloc, - .ipcp_flow_update = udp_ipcp_flow_update -}; - -int main(int argc, - char * argv[]) -{ - int i; - - - if (udp_data_init() < 0) { - log_err("Failed to init udp data."); - goto fail_data_init; - } - - if (ipcp_init(argc, argv, &udp_ops, THIS_TYPE) < 0) { - log_err("Failed to initialize IPCP."); - goto fail_init; - } - - if (ipcp_start() < 0) { - log_err("Failed to start IPCP."); - goto fail_start; - } - - ipcp_sigwait(); - - if (ipcp_get_state() == IPCP_SHUTDOWN) { - for (i = 0; i < IPCP_UDP_WR_THR; ++i) - pthread_cancel(udp_data.packet_writer[i]); - for (i = 0; i < IPCP_UDP_RD_THR; ++i) - pthread_cancel(udp_data.packet_reader[i]); - pthread_cancel(udp_data.mgmt_handler); - - for (i = 0; i < IPCP_UDP_WR_THR; ++i) - pthread_join(udp_data.packet_writer[i], NULL); - for (i = 0; i < IPCP_UDP_RD_THR; ++i) - pthread_join(udp_data.packet_reader[i], NULL); - pthread_join(udp_data.mgmt_handler, NULL); - close(udp_data.s_fd); - } - - ipcp_stop(); - - ipcp_fini(); - - udp_data_fini(); - - exit(EXIT_SUCCESS); - - fail_start: - ipcp_fini(); - fail_init: - udp_data_fini(); - fail_data_init: - exit(EXIT_FAILURE); -} diff --git a/src/ipcpd/udp/udp4.c b/src/ipcpd/udp/udp4.c deleted file mode 100644 index ff57bc09..00000000 --- a/src/ipcpd/udp/udp4.c +++ /dev/null @@ -1,42 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC process over UDP/IPv4 - * - * 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 200112L -#endif - -#include <ouroboros/ipcp-dev.h> - -#define BUILD_IPCP_UDP4 -#define THIS_TYPE IPCP_UDP4 -#define TYPE_STR "IPCP over UDP/IPv4" -#define OUROBOROS_PREFIX "ipcpd/udp4" -#define IPCP_UDP_MAX_PACKET_SIZE 8980 -#define __AF AF_INET -#define __ADDRSTRLEN INET_ADDRSTRLEN -#define __SOCKADDR sockaddr_in -#define __ADDR in_addr -#define __ADDR_ANY_INIT { .s_addr = INADDR_ANY } - -#include "udp.c" diff --git a/src/ipcpd/udp/udp6.c b/src/ipcpd/udp/udp6.c deleted file mode 100644 index 2ceb95f0..00000000 --- a/src/ipcpd/udp/udp6.c +++ /dev/null @@ -1,42 +0,0 @@ -/* - * Ouroboros - Copyright (C) 2016 - 2026 - * - * IPC process over UDP/IPv6 - * - * 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 200112L -#endif - -#include <ouroboros/ipcp-dev.h> - -#define BUILD_IPCP_UDP6 -#define THIS_TYPE IPCP_UDP6 -#define TYPE_STR "IPCP over UDP/IPv6" -#define OUROBOROS_PREFIX "ipcpd/udp6" -#define IPCP_UDP_MAX_PACKET_SIZE 8952 -#define __AF AF_INET6 -#define __ADDRSTRLEN INET6_ADDRSTRLEN -#define __SOCKADDR sockaddr_in6 -#define __ADDR in6_addr -#define __ADDR_ANY_INIT IN6ADDR_ANY_INIT - -#include "udp.c" diff --git a/src/ipcpd/unicast/fa.c b/src/ipcpd/unicast/fa.c index 596b101b..744eb316 100644 --- a/src/ipcpd/unicast/fa.c +++ b/src/ipcpd/unicast/fa.c @@ -130,8 +130,8 @@ struct fa_flow { 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 s_eid; /* Local PoA id */ + uint64_t r_eid; /* Remote PoA id */ uint64_t r_addr; /* Remote address */ void * ctx; /* Congestion avoidance context */ uint64_t fair; /* SFQ virtual finish tag (bytes) */ @@ -208,8 +208,8 @@ static int fa_rib_read(const char * path, sprintf(buf, "Flow established at: %20s\n" "Remote address: %20s\n" - "Local endpoint ID: %20s\n" - "Remote endpoint ID: %20s\n" + "Local PoA ID: %20s\n" + "Remote PoA ID: %20s\n" "Sent (packets): %20zu\n" "Sent (bytes): %20zu\n" "Send failed (packets): %20zu\n" diff --git a/src/ipcpd/unicast/main.c b/src/ipcpd/unicast/main.c index 86cb1f06..3ee53401 100644 --- a/src/ipcpd/unicast/main.c +++ b/src/ipcpd/unicast/main.c @@ -35,6 +35,7 @@ #include <ouroboros/ipcp-dev.h> #include <ouroboros/logs.h> #include <ouroboros/notifier.h> +#include <ouroboros/qos.h> #include <ouroboros/random.h> #include <ouroboros/rib.h> #include <ouroboros/time.h> @@ -175,12 +176,14 @@ static void stop_components(void) ipcp_set_state(IPCP_BOOT); } -static int unicast_ipcp_enroll(const char * dst, - struct layer_info * info) +static int unicast_ipcp_enroll(const char * dst, + const struct poa_addr * addr, + struct layer_info * info) { struct ipcp_config * conf; struct conn conn; uint8_t id[ENROLL_ID_LEN]; + qosspec_t qs = qos_msg; if (random_buffer(id, ENROLL_ID_LEN) < 0) { log_err("Failed to generate enrollment ID."); @@ -189,7 +192,7 @@ static int unicast_ipcp_enroll(const char * dst, log_info_id(id, "Requesting enrollment."); - if (connmgr_alloc(COMPID_ENROLL, dst, NULL, &conn) < 0) { + if (connmgr_alloc(COMPID_ENROLL, dst, &qs, addr, &conn) < 0) { log_err_id(id, "Failed to get connection."); goto fail_id; } diff --git a/src/irmd/config.h.in b/src/irmd/config.h.in index 53f80ca4..2440e180 100644 --- a/src/irmd/config.h.in +++ b/src/irmd/config.h.in @@ -21,10 +21,6 @@ */ -#define IPCP_UDP4_EXEC "@IPCP_UDP4_TARGET@" -#define IPCP_UDP6_EXEC "@IPCP_UDP6_TARGET@" -#define IPCP_ETH_LLC_EXEC "@IPCP_ETH_LLC_TARGET@" -#define IPCP_ETH_DIX_EXEC "@IPCP_ETH_DIX_TARGET@" #define IPCP_UNICAST_EXEC "@IPCP_UNICAST_TARGET@" #define IPCP_BROADCAST_EXEC "@IPCP_BROADCAST_TARGET@" #define IPCP_LOCAL_EXEC "@IPCP_LOCAL_TARGET@" @@ -75,6 +71,8 @@ #define OUROBOROS_CLI_CRT_DIR "@OUROBOROS_CLI_CRT_DIR@" #define OUROBOROS_CHAIN_DIR "@OUROBOROS_UNTRUSTED_DIR@" +/* Endpoint peers are keyed on their address, not on a service name. */ + #define IRMD_PKILL_TIMEOUT @IRMD_PKILL_TIMEOUT@ #cmakedefine DISABLE_DIRECT_IPC diff --git a/src/irmd/configfile.c b/src/irmd/configfile.c index e2e1e554..de02242f 100644 --- a/src/irmd/configfile.c +++ b/src/irmd/configfile.c @@ -43,6 +43,7 @@ #include <assert.h> #include <errno.h> #include <inttypes.h> +#include <stdio.h> #include <stdlib.h> #include <string.h> #include <toml.h> @@ -92,171 +93,173 @@ static int toml_local(toml_table_t * table, return toml_hash(table, &conf->layer_info); } -static int toml_eth_dev(toml_table_t * table, - struct eth_config * conf) -{ - toml_datum_t dev; - dev = toml_string_in(table, "dev"); - if (!dev.ok) { - log_err("Missing device."); - return -1; - } +/* Defined with the peer helpers below; shared grammar for both paths. */ +static int toml_udp_str(const char * str, + char * host, + int * port); - if (strlen(dev.u.s) > DEV_NAME_SIZE) { - log_err("Device name too long: %s", dev.u.s); - free(dev.u.s); +/* Attach the PoAs an IPCP is given; one call per PoA. */ +static int poa_eth_set(struct poa_spec * poa, + const char * dev, + uint16_t ethertype) +{ + if (strlen(dev) > DEV_NAME_SIZE) { + log_err("Invalid device name %s.", dev); return -1; } - strcpy(conf->dev, dev.u.s); - free(dev.u.s); + memset(poa, 0, sizeof(*poa)); + + poa->type = POA_ETH; + poa->eth.ethertype = ethertype; + + strcpy(poa->eth.dev, dev); return 0; } -static int toml_eth_llc(toml_table_t * table, - struct ipcp_config * conf) +static int toml_poa_eth(toml_table_t * tbl, + struct poa_spec * poa) { - *conf = eth_llc_default_conf; - - if (toml_hash(table, &conf->layer_info) < 0) - return -1; - - return toml_eth_dev(table, &conf->eth); -} + toml_datum_t dev; + toml_datum_t et; + uint16_t ethertype = POA_ETHERTYPE; + int ret = -1; + dev = toml_string_in(tbl, "dev"); + if (!dev.ok) { + log_err("An eth PoA needs a device."); + goto fail; + } -static int toml_ethertype(toml_table_t * table, - struct eth_config * conf) -{ - toml_datum_t ethertype; + et = toml_int_in(tbl, "ethertype"); + if (et.ok) { + if (et.u.i < 0 || et.u.i > 0xFFFF) { + log_err("Invalid ethertype."); + goto fail; + } - ethertype = toml_int_in(table, "ethertype"); - if (ethertype.ok) - conf->ethertype = ethertype.u.i; + ethertype = et.u.i; + } - if (conf->ethertype < 0x0600 || conf->ethertype == 0xFFFF) - return -1; + ret = poa_eth_set(poa, dev.u.s, ethertype); + fail: + if (dev.ok) + free(dev.u.s); - return 0; + return ret; } -static int toml_eth_dix(toml_table_t * table, - struct ipcp_config * conf) +/* A PoA is bound locally: only literal addresses are accepted here. */ +static int toml_poa_udp(const char * str, + struct poa_spec * poa) { - *conf = eth_dix_default_conf; + char host[POA_HOST_STRLEN + 1]; + int port; - if (toml_hash(table, &conf->layer_info) < 0) + if (toml_udp_str(str, host, &port) < 0) return -1; - if (toml_eth_dev(table, &conf->eth) < 0) - return -1; + memset(poa, 0, sizeof(*poa)); - if (toml_ethertype(table, &conf->eth) < 0) { - log_err("Ethertype not in valid range."); - return -1; + if (inet_pton(AF_INET, host, &poa->udp4.ip_addr.s_addr) == 1) { + poa->type = POA_UDP4; + poa->udp4.port = port; + return 0; } - return 0; -} - -static int toml_udp4(toml_table_t * table, - struct ipcp_config * conf) -{ - struct udp4_config * udp4; - toml_datum_t ip; - toml_datum_t port; - toml_datum_t dns; - - *conf = udp4_default_conf; - udp4 = &conf->udp4; - - ip = toml_string_in(table, "ip"); - if (!ip.ok) { - log_err("No IP address specified!"); - goto fail_ip; + if (inet_pton(AF_INET6, host, &poa->udp6.ip_addr) == 1) { + poa->type = POA_UDP6; + poa->udp6.port = port; + return 0; } - if (inet_pton (AF_INET, ip.u.s, &udp4->ip_addr.s_addr) != 1) { - log_err("Failed to parse IPv4 address %s.", ip.u.s); - goto fail_addr; - } + log_err("Invalid IP address %s.", host); - port = toml_int_in(table, "port"); - if (port.ok) - udp4->port = port.u.i; + return -1; +} - dns = toml_string_in(table, "dns"); - if (dns.ok) { - if (inet_pton(AF_INET, dns.u.s, &udp4->dns_addr.s_addr) < 0) { - log_err("Failed to parse DNS address %s.", ip.u.s); - goto fail_dns; +/* Attaches every PoA in the "udp" or "eth" array; string or table. */ +static int toml_poa_array(toml_table_t * table, + pid_t pid, + const char * key) +{ + struct poa_spec poa; + toml_array_t * arr; + int n; + int i; + bool is_eth; + + arr = toml_array_in(table, key); + if (arr == NULL) { + /* A table here would otherwise attach nothing, silently. */ + if (toml_table_in(table, key) != NULL) { + log_err("A %s PoA is an array: %s = [...] " + "or [[%s]].", key, key, key); + return -1; } - free(dns.u.s); + return 0; } - free(ip.u.s); + is_eth = strcmp(key, "eth") == 0; - return 0; + n = toml_array_nelem(arr); - fail_dns: - free(dns.u.s); - fail_addr: - free(ip.u.s); - fail_ip: - return -1; -} + for (i = 0; i < n; i++) { + toml_datum_t s; + int ret; -static int toml_udp6(toml_table_t * table, - struct ipcp_config * conf) -{ - struct in6_addr ip6; - struct in6_addr dns6; - toml_datum_t ip; - toml_datum_t port; - toml_datum_t dns; - - *conf = udp6_default_conf; - ip6 = conf->udp6.ip_addr; - dns6 = conf->udp6.dns_addr; - - ip = toml_string_in(table, "ip"); - if (!ip.ok) { - log_err("No IP address specified!"); - goto fail_ip; - } + s = toml_string_at(arr, i); + if (s.ok) { + if (is_eth) + ret = poa_eth_set(&poa, s.u.s, + POA_ETHERTYPE); + else + ret = toml_poa_udp(s.u.s, &poa); - if (inet_pton (AF_INET6, ip.u.s, &ip6.s6_addr) != 1) { - log_err("Failed to parse IPv4 address %s.", ip.u.s); - goto fail_addr; - } + free(s.u.s); + } else { + toml_table_t * d = toml_table_at(arr, i); - port = toml_int_in(table, "port"); - if (port.ok) - conf->udp6.port = port.u.i; + if (d == NULL) { + log_err("Invalid %s PoA entry.", key); + return -1; + } + + if (is_eth) { + ret = toml_poa_eth(d, &poa); + } else { + toml_datum_t addr = toml_string_in(d, "addr"); + + if (!addr.ok) { + log_err("No addr for udp PoA."); + return -1; + } - dns = toml_string_in(table, "dns"); - if (dns.ok) { - if (inet_pton(AF_INET6, dns.u.s, &dns6.s6_addr) < 0) { - log_err("Failed to parse DNS address %s.", ip.u.s); - goto fail_dns; + ret = toml_poa_udp(addr.u.s, &poa); + free(addr.u.s); + } } - free(dns.u.s); - } + if (ret < 0) + return -1; - free(ip.u.s); + if (attach_ipcp(pid, &poa, true) < 0) + return -1; + } return 0; +} - fail_dns: - free(dns.u.s); - fail_addr: - free(ip.u.s); - fail_ip: - return -1; +static int toml_attach(toml_table_t * table, + pid_t pid) +{ + if (toml_poa_array(table, pid, "udp") < 0) + return -1; + + return toml_poa_array(table, pid, "eth"); } static int toml_broadcast(toml_table_t * table, @@ -265,15 +268,13 @@ static int toml_broadcast(toml_table_t * table, (void) table; (void) conf; - /* Nothing to do here. */ - return 0; } #define BETWEEN(a, b, c) ((a) >= (b) && (a) <= (c)) #define DHT(conf, x) (conf)->dht.params.x static int toml_dir(toml_table_t * table, - struct dir_config * conf) + struct dir_config * conf) { toml_datum_t dir; toml_datum_t alpha; @@ -594,6 +595,207 @@ static int toml_register(toml_table_t * table, return ret; } +/* Fills in a peer's eth PoA; dst MAC stays zero, the resolve marker. */ +static int toml_peer_eth(toml_table_t * tbl, + struct poa_addr * addr) +{ + toml_datum_t dev; + toml_datum_t et; + int ret = -1; + + dev = toml_string_in(tbl, "dev"); + + memset(addr, 0, sizeof(*addr)); + + addr->type = POA_ETH; + addr->eth.src.ethertype = POA_ETHERTYPE; + addr->eth.dst.ethertype = POA_ETHERTYPE; + + et = toml_int_in(tbl, "ethertype"); + if (et.ok) { + if (et.u.i < 0 || et.u.i > 0xFFFF) { + log_err("Invalid ethertype."); + goto fail; + } + + addr->eth.src.ethertype = et.u.i; + addr->eth.dst.ethertype = et.u.i; + } + + if (dev.ok) { + if (strlen(dev.u.s) > DEV_NAME_SIZE) { + log_err("Invalid device name %s.", dev.u.s); + goto fail; + } + + strcpy(addr->eth.src.dev, dev.u.s); + } + + ret = 0; + fail: + if (dev.ok) + free(dev.u.s); + + return ret; +} + +/* Same grammar as the CLI's udp keyword; see irm_utils.c parse_udp_str. */ +static int toml_udp_str(const char * str, + char * host, + int * port) +{ + struct in6_addr v6; + char buf[POA_HOST_STRLEN + 1]; + char * p; + char * end; + long n; + + *port = POA_UDP_PORT; + + if (strlen(str) > POA_HOST_STRLEN) + goto fail; + + strcpy(buf, str); + + if (buf[0] == '[') { + p = strchr(buf, ']'); + if (p == NULL) + goto fail; + + *p++ = '\0'; + + strcpy(host, buf + 1); + + if (*p == '\0') + return 0; + + if (*p != ':') + goto fail; + + ++p; + } else if (inet_pton(AF_INET6, buf, &v6) == 1) { + strcpy(host, buf); + return 0; + } else { + p = strrchr(buf, ':'); + if (p == NULL) { + strcpy(host, buf); + return 0; + } + + *p++ = '\0'; + + strcpy(host, buf); + } + + n = strtol(p, &end, 10); + if (*p == '\0' || *end != '\0' || n < 1 || n > 65535) + goto fail; + + *port = (int) n; + + return 0; + fail: + log_err("Invalid UDP address: %s.", str); + return -1; +} + +/* Fills in a peer's udp PoA; an unresolved name is left for the IRMd. */ +static int toml_peer_udp(const char * str, + struct poa_addr * addr) +{ + char host[POA_HOST_STRLEN + 1]; + int port; + + if (toml_udp_str(str, host, &port) < 0) + return -1; + + memset(addr, 0, sizeof(*addr)); + + if (inet_pton(AF_INET, host, &addr->udp4.ip_addr) == 1) { + addr->type = POA_UDP4; + addr->udp4.port = port; + return 0; + } + + if (inet_pton(AF_INET6, host, &addr->udp6.ip_addr) == 1) { + addr->type = POA_UDP6; + addr->udp6.port = port; + return 0; + } + + addr->type = POA_UDP; + addr->udp4.port = port; + + strcpy(addr->hostname, host); + + return 0; +} + +/* + * An entry is a name, or a table naming a PoA to reach it over. + * On entry, *paddr already points at the caller's struct poa_addr to + * fill in. Set to NULL wherever there is no PoA to dial: a bare + * dst-only table (recursive lookup) or any parse failure. + */ +static int toml_peer(toml_table_t * tbl, + char * dst, + struct poa_addr ** paddr) +{ + struct poa_addr * addr = *paddr; + toml_table_t * eth; + toml_datum_t name; + toml_datum_t udp; + int ret = -1; + + name = toml_string_in(tbl, "dst"); + if (!name.ok) { + log_err("PoA table entry has no dst."); + + *paddr = NULL; + return -1; + } + + if (strlen(name.u.s) > LAYER_NAME_SIZE) { + log_err("Destination name too long: %s.", name.u.s); + free(name.u.s); + + *paddr = NULL; + return -1; + } + + strcpy(dst, name.u.s); + free(name.u.s); + + eth = toml_table_in(tbl, "eth"); + udp = toml_string_in(tbl, "udp"); + if (eth != NULL && udp.ok) { + log_err("A PoA is eth or udp, not both."); + + *paddr = NULL; + goto fail; + } + + if (eth == NULL && !udp.ok) { + *paddr = NULL; + ret = 0; + goto fail; + } + + if (eth != NULL) + ret = toml_peer_eth(eth, addr); + else + ret = toml_peer_udp(udp.u.s, addr); + + if (ret < 0) + *paddr = NULL; + fail: + if (udp.ok) + free(udp.u.s); + + return ret; +} + static int toml_connect(toml_table_t * table, pid_t pid) { @@ -605,21 +807,40 @@ static int toml_connect(toml_table_t * table, if (conn == NULL) return 0; - for (i=0; ret == 0; i++) { - toml_datum_t dst; - qosspec_t qs = qos_raw; + for (i = 0; ret == 0; i++) { + toml_datum_t dst; + toml_table_t * tbl; + struct poa_addr addr; + struct poa_addr * paddr = &addr; + char buf[LAYER_NAME_SIZE + 1]; + const char * d; + qosspec_t qs = qos_raw; dst = toml_string_at(conn, i); - if (!dst.ok) - break; + if (dst.ok) { + d = dst.u.s; + paddr = NULL; + } else { + tbl = toml_table_at(conn, i); + if (tbl == NULL) + break; + + ret = toml_peer(tbl, buf, &paddr); + if (ret < 0) + break; + + d = buf; + } + + log_dbg("Connecting %d to %s", pid, d); - log_dbg("Connecting %d to %s", pid, dst.u.s); + ret = connect_ipcp_resolve(pid, d, MGMT_COMP, qs, paddr); - ret = connect_ipcp(pid, dst.u.s, MGMT_COMP, qs); if (ret == 0) - ret = connect_ipcp(pid, dst.u.s, DT_COMP, qs); + ret = connect_ipcp_resolve(pid, d, DT_COMP, qs, paddr); - free(dst.u.s); + if (dst.ok) + free(dst.u.s); } return ret; @@ -629,9 +850,11 @@ static int toml_ipcp(toml_table_t * table, struct ipcp_info * info, struct ipcp_config * conf) { - toml_datum_t bootstrap; - toml_datum_t enrol; - int ret; + toml_datum_t bootstrap; + toml_datum_t enrol; + toml_table_t * enrol_tbl; + bool have_enrol; + int ret; log_dbg("Found IPCP %s in configuration file.", info->name); @@ -643,21 +866,49 @@ static int toml_ipcp(toml_table_t * table, bootstrap = toml_string_in(table, "bootstrap"); enrol = toml_string_in(table, "enrol"); - if (bootstrap.ok && enrol.ok) { + enrol_tbl = enrol.ok ? NULL : toml_table_in(table, "enrol"); + + have_enrol = enrol.ok || enrol_tbl != NULL; + if (bootstrap.ok && have_enrol) { log_err("Ignoring bootstrap for IPCP %s.", info->name); free(bootstrap.u.s); bootstrap.ok = false; } - if (!bootstrap.ok && !enrol.ok) { + if (!bootstrap.ok && !have_enrol) { log_dbg("Nothing more to do for %s.", info->name); return 0; } - if (enrol.ok) { + /* Endpoints come first: enrolment reaches the peer over one. */ + if (toml_attach(table, info->pid) < 0) { + log_err("Failed to attach PoAs for %s.", info->name); + return -1; + } + + if (have_enrol) { struct layer_info layer; - ret = enroll_ipcp(info->pid, enrol.u.s); - free(enrol.u.s); + struct poa_addr addr; + struct poa_addr * paddr = &addr; + char buf[LAYER_NAME_SIZE + 1]; + const char * dst; + + if (enrol.ok) { + dst = enrol.u.s; + paddr = NULL; + } else { + if (toml_peer(enrol_tbl, buf, &paddr) < 0) { + log_err("Invalid enrol table for %s.", + info->name); + return -1; + } + dst = buf; + } + + ret = enroll_ipcp_resolve(info->pid, dst, paddr); + + if (enrol.ok) + free(enrol.u.s); if (ret < 0) { log_err("Failed to enrol %s.", info->name); return -1; @@ -694,18 +945,6 @@ static int toml_ipcp(toml_table_t * table, case IPCP_LOCAL: ret = toml_local(table, conf); break; - case IPCP_ETH_DIX: - ret = toml_eth_dix(table, conf); - break; - case IPCP_ETH_LLC: - ret = toml_eth_llc(table, conf); - break; - case IPCP_UDP4: - ret = toml_udp4(table, conf); - break; - case IPCP_UDP6: - ret = toml_udp6(table, conf); - break; case IPCP_BROADCAST: ret = toml_broadcast(table, conf); break; @@ -1031,14 +1270,6 @@ static int toml_toplevel(toml_table_t * table, return toml_name_list(subtable); else if (strcmp(key, "local") == 0) return toml_ipcp_list(subtable, IPCP_LOCAL); - else if (strcmp(key, "eth-dix") == 0) - return toml_ipcp_list(subtable, IPCP_ETH_DIX); - else if (strcmp(key, "eth-llc") == 0) - return toml_ipcp_list(subtable, IPCP_ETH_LLC); - else if (strcmp(key, "udp4") == 0) - return toml_ipcp_list(subtable, IPCP_UDP4); - else if (strcmp(key, "udp6") == 0) - return toml_ipcp_list(subtable, IPCP_UDP6); else if (strcmp(key, "broadcast") == 0) return toml_ipcp_list(subtable, IPCP_BROADCAST); else if (strcmp(key, "unicast") == 0) diff --git a/src/irmd/ipcp.c b/src/irmd/ipcp.c index 7eccfc80..cd662221 100644 --- a/src/irmd/ipcp.c +++ b/src/irmd/ipcp.c @@ -34,6 +34,7 @@ #include <ouroboros/utils.h> #include "ipcp.h" +#include "reg/reg.h" #include <fcntl.h> #include <pthread.h> @@ -72,6 +73,10 @@ static char * str_ipcp_cmd(int code) return "alloc_resp"; case IPCP_MSG_CODE__IPCP_FLOW_DEALLOC: return "dealloc"; + case IPCP_MSG_CODE__IPCP_FLOW_UPDATE: + return "flow_update"; + case IPCP_MSG_CODE__IPCP_REPLY: + return "reply"; default: assert(false); return "unknown"; @@ -196,7 +201,9 @@ int ipcp_bootstrap(pid_t pid, msg.conf = ipcp_config_s_to_msg(conf); recv_msg = send_recv_ipcp_msg(pid, &msg); + ipcp_config_msg__free_unpacked(msg.conf, NULL); + if (recv_msg == NULL) return -EIPCP; @@ -225,9 +232,88 @@ int ipcp_bootstrap(pid_t pid, return ret; } -int ipcp_enroll(pid_t pid, - const char * dst, - struct layer_info * info) +ssize_t ipcp_list_poas(pid_t pid, + struct poa_spec ** eps) +{ + ipcp_msg_t msg = IPCP_MSG__INIT; + ipcp_msg_t * recv_msg; + size_t nr; + size_t i; + + if (eps == NULL) + return -EINVAL; + + *eps = NULL; + + msg.code = IPCP_MSG_CODE__IPCP_LIST_POAS; + + recv_msg = send_recv_ipcp_msg(pid, &msg); + if (recv_msg == NULL) + return -EIPCP; + + nr = recv_msg->n_poas; + if (nr == 0) { + ipcp_msg__free_unpacked(recv_msg, NULL); + return 0; + } + + *eps = malloc(nr * sizeof(**eps)); + if (*eps == NULL) { + ipcp_msg__free_unpacked(recv_msg, NULL); + return -ENOMEM; + } + + for (i = 0; i < nr; i++) + (*eps)[i] = poa_spec_msg_to_s(recv_msg->poas[i]); + + ipcp_msg__free_unpacked(recv_msg, NULL); + + return (ssize_t) nr; +} + +int ipcp_attach(pid_t pid, + const struct poa_spec * poa, + bool attach) +{ + ipcp_msg_t msg = IPCP_MSG__INIT; + ipcp_msg_t * recv_msg; + int ret; + + if (poa == NULL) + return -EINVAL; + + if (attach) + msg.code = IPCP_MSG_CODE__IPCP_ATTACH; + else + msg.code = IPCP_MSG_CODE__IPCP_DETACH; + + msg.poa = poa_spec_s_to_msg(poa); + if (msg.poa == NULL) + return -EINVAL; + + recv_msg = send_recv_ipcp_msg(pid, &msg); + + poa_spec_msg__free_unpacked(msg.poa, NULL); + + if (recv_msg == NULL) + return -EIPCP; + + if (!recv_msg->has_result) { + ipcp_msg__free_unpacked(recv_msg, NULL); + return -EIPCP; + } + + ret = recv_msg->result; + + ipcp_msg__free_unpacked(recv_msg, NULL); + + return ret; +} + +int ipcp_enroll(pid_t pid, + const char * dst, + const struct poa_addr * addr, + struct layer_info * info) { ipcp_msg_t msg = IPCP_MSG__INIT; ipcp_msg_t * recv_msg; @@ -239,7 +325,16 @@ int ipcp_enroll(pid_t pid, msg.code = IPCP_MSG_CODE__IPCP_ENROLL; msg.dst = (char *) dst; + if (addr != NULL) { + msg.peer = poa_addr_s_to_msg(addr); + if (msg.peer == NULL) + return -ENOMEM; + } + recv_msg = send_recv_ipcp_msg(pid, &msg); + + if (msg.peer != NULL) + poa_addr_msg__free_unpacked(msg.peer, NULL); if (recv_msg == NULL) return -EIPCP; @@ -267,10 +362,11 @@ int ipcp_enroll(pid_t pid, return 0; } -int ipcp_connect(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs) +int ipcp_connect(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr) { ipcp_msg_t msg = IPCP_MSG__INIT; ipcp_msg_t * recv_msg; @@ -283,8 +379,21 @@ int ipcp_connect(pid_t pid, msg.pid = pid; msg.qosspec = qos_spec_s_to_msg(&qs); + if (addr != NULL) { + msg.peer = poa_addr_s_to_msg(addr); + if (msg.peer == NULL) { + free(msg.qosspec); + return -ENOMEM; + } + } + recv_msg = send_recv_ipcp_msg(pid, &msg); + + if (msg.peer != NULL) + poa_addr_msg__free_unpacked(msg.peer, NULL); + free(msg.qosspec); + if (recv_msg == NULL) return -EIPCP; @@ -457,6 +566,8 @@ int ipcp_flow_update(const struct flow_info * flow, msg.has_pk = true; msg.pk.data = data.data; msg.pk.len = data.len; + msg.has_is_poa = true; + msg.is_poa = reg_flow_is_poa(flow->id); recv_msg = send_recv_ipcp_msg(flow->n_1_pid, &msg); if (recv_msg == NULL) { @@ -538,6 +649,8 @@ int ipcp_flow_alloc_resp(const struct flow_info * flow, msg.has_pk = response == 0; msg.pk.data = data.data; msg.pk.len = data.len; + msg.has_is_poa = true; + msg.is_poa = reg_flow_is_poa(flow->id); recv_msg = send_recv_ipcp_msg(flow->n_1_pid, &msg); if (recv_msg == NULL) @@ -567,6 +680,8 @@ int ipcp_flow_dealloc(pid_t pid, msg.flow_id = flow_id; msg.has_timeo_sec = true; msg.timeo_sec = timeo; + msg.has_is_poa = true; + msg.is_poa = reg_flow_is_poa(flow_id); recv_msg = send_recv_ipcp_msg(pid, &msg); if (recv_msg == NULL) diff --git a/src/irmd/ipcp.h b/src/irmd/ipcp.h index 8d06623c..1f257e7d 100644 --- a/src/irmd/ipcp.h +++ b/src/irmd/ipcp.h @@ -27,48 +27,57 @@ #ifndef OUROBOROS_IRMD_IPCP_H #define OUROBOROS_IRMD_IPCP_H -int ipcp_enroll(pid_t pid, - const char * dst, - struct layer_info * info); +int ipcp_enroll(pid_t pid, + const char * dst, + const struct poa_addr * addr, + struct layer_info * info); -int ipcp_bootstrap(pid_t pid, - struct ipcp_config * conf, - struct layer_info * info); +int ipcp_attach(pid_t pid, + const struct poa_spec * poa, + bool attach); -int ipcp_connect(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs); +ssize_t ipcp_list_poas(pid_t pid, + struct poa_spec ** eps); -int ipcp_disconnect(pid_t pid, - const char * dst, - const char * component); +int ipcp_bootstrap(pid_t pid, + struct ipcp_config * conf, + struct layer_info * info); -int ipcp_reg(pid_t pid, - const buffer_t hash); +int ipcp_connect(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr); -int ipcp_unreg(pid_t pid, +int ipcp_disconnect(pid_t pid, + const char * dst, + const char * component); + +int ipcp_reg(pid_t pid, const buffer_t hash); -int ipcp_query(pid_t pid, - const buffer_t dst); +int ipcp_unreg(pid_t pid, + const buffer_t hash); + +int ipcp_query(pid_t pid, + const buffer_t dst); -int ipcp_flow_alloc(const struct flow_info * flow, - const buffer_t hash, - const buffer_t data); +int ipcp_flow_alloc(const struct flow_info * flow, + const buffer_t hash, + const buffer_t data); -int ipcp_flow_join(const struct flow_info * flow, - const buffer_t dst); +int ipcp_flow_join(const struct flow_info * flow, + const buffer_t dst); -int ipcp_flow_alloc_resp(const struct flow_info * flow, - int response, - const buffer_t data); +int ipcp_flow_alloc_resp(const struct flow_info * flow, + int response, + const buffer_t data); -int ipcp_flow_dealloc(pid_t pid, - int flow_id, - time_t timeo); +int ipcp_flow_dealloc(pid_t pid, + int flow_id, + time_t timeo); -int ipcp_flow_update(const struct flow_info * flow, - const buffer_t data); +int ipcp_flow_update(const struct flow_info * flow, + const buffer_t data); #endif /* OUROBOROS_IRMD_IPCP_H */ diff --git a/src/irmd/irmd.h b/src/irmd/irmd.h index f88378ad..9d42e248 100644 --- a/src/irmd/irmd.h +++ b/src/irmd/irmd.h @@ -26,29 +26,48 @@ #include <ouroboros/ipcp.h> #include <ouroboros/irm.h> -int create_ipcp(struct ipcp_info * info); +int create_ipcp(struct ipcp_info * info); -int bootstrap_ipcp(pid_t pid, - struct ipcp_config * conf); +int bootstrap_ipcp(pid_t pid, + struct ipcp_config * conf); -int enroll_ipcp(pid_t pid, - const char * dst); +int enroll_ipcp(pid_t pid, + const char * dst, + const struct poa_addr * addr); -int connect_ipcp(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs); +int enroll_ipcp_resolve(pid_t pid, + const char * dst, + struct poa_addr * addr); -int name_create(struct name_info * info); +int attach_ipcp(pid_t pid, + const struct poa_spec * poa, + bool attach); -int name_reg(const char * name, - pid_t pid); +ssize_t list_poas(pid_t pid, + struct poa_spec ** eps); -int bind_process(pid_t pid, - const char * name); +int connect_ipcp(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr); -int bind_program(char ** exec, - const char * name, - uint8_t flags); +int connect_ipcp_resolve(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + struct poa_addr * addr); + +int name_create(struct name_info * info); + +int name_reg(const char * name, + pid_t pid); + +int bind_process(pid_t pid, + const char * name); + +int bind_program(char ** exec, + const char * name, + uint8_t flags); #endif /* OUROBOROS_IRMD_H*/ diff --git a/src/irmd/main.c b/src/irmd/main.c index 19be4ab9..e587a552 100644 --- a/src/irmd/main.c +++ b/src/irmd/main.c @@ -61,6 +61,7 @@ #include <dirent.h> #include <grp.h> +#include <netdb.h> #include <pwd.h> #include <signal.h> #include <spawn.h> @@ -225,18 +226,6 @@ static pid_t spawn_ipcp(struct ipcp_info * info) case IPCP_BROADCAST: exec_name = IPCP_BROADCAST_EXEC; break; - case IPCP_UDP4: - exec_name = IPCP_UDP4_EXEC; - break; - case IPCP_UDP6: - exec_name = IPCP_UDP6_EXEC; - break; - case IPCP_ETH_LLC: - exec_name = IPCP_ETH_LLC_EXEC; - break; - case IPCP_ETH_DIX: - exec_name = IPCP_ETH_DIX_EXEC; - break; case IPCP_LOCAL: exec_name = IPCP_LOCAL_EXEC; break; @@ -367,9 +356,6 @@ int bootstrap_ipcp(pid_t pid, goto fail; } - if (conf->type == IPCP_UDP4 || conf->type == IPCP_UDP6) - conf->layer_info.dir_hash_algo = (enum pol_dir_hash) HASH_MD5; - if (ipcp_bootstrap(pid, conf, &layer)) { log_err("Could not bootstrap IPCP."); goto fail; @@ -389,8 +375,55 @@ int bootstrap_ipcp(pid_t pid, return -1; } -int enroll_ipcp(pid_t pid, - const char * dst) +ssize_t list_poas(pid_t pid, + struct poa_spec ** eps) +{ + struct ipcp_info info; + + info.pid = pid; + if (reg_get_ipcp(&info, NULL) < 0) { + log_err("Could not find IPCP %d.", pid); + return -1; + } + + if (info.type != IPCP_UNICAST && info.type != IPCP_BROADCAST) + return 0; /* nothing it could be attached to */ + + return ipcp_list_poas(pid, eps); +} + +/* Only the types that can carry a PoA may take one. */ +int attach_ipcp(pid_t pid, + const struct poa_spec * poa, + bool attach) +{ + struct ipcp_info info; + + info.pid = pid; + if (reg_get_ipcp(&info, NULL) < 0) { + log_err("Could not find IPCP %d.", pid); + return -1; + } + + if (info.type != IPCP_UNICAST && info.type != IPCP_BROADCAST) { + log_err("IPCP %d does not support PoAs.", pid); + return -1; + } + + if (ipcp_attach(pid, poa, attach) < 0) { + log_err("Could not %s IPCP %d.", + attach ? "attach" : "detach", pid); + return -1; + } + + log_info("%s IPCP %d.", attach ? "Attached" : "Detached", pid); + + return 0; +} + +int enroll_ipcp(pid_t pid, + const char * dst, + const struct poa_addr * addr) { struct layer_info layer; struct ipcp_info info; @@ -402,7 +435,7 @@ int enroll_ipcp(pid_t pid, goto fail; } - if (ipcp_enroll(pid, dst, &layer) < 0) { + if (ipcp_enroll(pid, dst, addr, &layer) < 0) { log_err("Could not enroll IPCP %d.", pid); goto fail; } @@ -421,10 +454,11 @@ int enroll_ipcp(pid_t pid, return -1; } -int connect_ipcp(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs) +int connect_ipcp(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr) { struct ipcp_info info; @@ -442,7 +476,7 @@ int connect_ipcp(pid_t pid, log_dbg("Connecting %s to %s.", component, dst); - if (ipcp_connect(pid, dst, component, qs)) { + if (ipcp_connect(pid, dst, component, qs, addr) < 0) { log_err("Could not connect IPCP %d to %s.", pid, dst); return -EPERM; } @@ -453,6 +487,172 @@ int connect_ipcp(pid_t pid, return 0; } +/* Resolve the dial name, if any, and fill in one record. */ +static int poa_addr_resolve(struct poa_addr * addr, + struct addrinfo ** res) +{ + struct addrinfo hints; + + *res = NULL; + + if (addr == NULL) + return 0; + + if (addr->hostname[0] == '\0') + return addr->type == POA_UDP ? -EINVAL : 0; + + if (addr->type != POA_UDP) + return -EINVAL; + + memset(&hints, 0, sizeof(hints)); + + hints.ai_family = AF_UNSPEC; + hints.ai_socktype = SOCK_DGRAM; + if (getaddrinfo(addr->hostname, NULL, &hints, res) != 0) { + log_err("Failed to resolve %s.", addr->hostname); + return -EADDRNOTAVAIL; + } + + return 0; +} + +static void poa_addr_from_ai(struct poa_addr * addr, + const struct addrinfo * ai, + uint16_t port) +{ + struct sockaddr_in * in; + struct sockaddr_in6 * in6; + + if (ai->ai_family == AF_INET) { + in = (struct sockaddr_in *) ai->ai_addr; + addr->type = POA_UDP4; + addr->udp4.ip_addr = in->sin_addr; + addr->udp4.port = port; + } else { + in6 = (struct sockaddr_in6 *) ai->ai_addr; + addr->type = POA_UDP6; + addr->udp6.ip_addr = in6->sin6_addr; + addr->udp6.port = port; + } +} + +/* Skip families without an attached PoA; keep all if none reported. */ +static void poa_families(pid_t pid, + bool * v4, + bool * v6) +{ + struct poa_spec * eps; + ssize_t n; + ssize_t i; + + eps = NULL; + + *v4 = false; + *v6 = false; + + n = list_poas(pid, &eps); + for (i = 0; i < n; i++) { + if (eps[i].type == POA_UDP4) + *v4 = true; + + if (eps[i].type == POA_UDP6) + *v6 = true; + } + + free(eps); + + if (!*v4 && !*v6) { + *v4 = true; + *v6 = true; + } +} + +/* Try each resolved record in order until an enrolment succeeds. */ +int enroll_ipcp_resolve(pid_t pid, + const char * dst, + struct poa_addr * addr) +{ + struct addrinfo * res; + struct addrinfo * ai; + uint16_t port; + bool v4; + bool v6; + int ret; + + ret = poa_addr_resolve(addr, &res); + if (ret < 0) + return ret; + + if (res == NULL) + return enroll_ipcp(pid, dst, addr); + + port = addr->udp4.port; /* POA_UDP parks it there */ + + poa_families(pid, &v4, &v6); + + ret = -EADDRNOTAVAIL; + + for (ai = res; ai != NULL; ai = ai->ai_next) { + if ((ai->ai_family == AF_INET && !v4) + || (ai->ai_family == AF_INET6 && !v6)) + continue; + + poa_addr_from_ai(addr, ai, port); + + ret = enroll_ipcp(pid, dst, addr); + if (ret == 0) + break; + } + + freeaddrinfo(res); + + return ret; +} + +/* Try each resolved record in order until a connect succeeds. */ +int connect_ipcp_resolve(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + struct poa_addr * addr) +{ + struct addrinfo * res; + struct addrinfo * ai; + uint16_t port; + bool v4; + bool v6; + int ret; + + ret = poa_addr_resolve(addr, &res); + if (ret < 0) + return ret; + + if (res == NULL) + return connect_ipcp(pid, dst, component, qs, addr); + + port = addr->udp4.port; /* POA_UDP parks it there */ + + poa_families(pid, &v4, &v6); + + ret = -EADDRNOTAVAIL; + + for (ai = res; ai != NULL; ai = ai->ai_next) { + if ((ai->ai_family == AF_INET && !v4) + || (ai->ai_family == AF_INET6 && !v6)) + continue; + + poa_addr_from_ai(addr, ai, port); + + ret = connect_ipcp(pid, dst, component, qs, addr); + if (ret == 0) + break; + } + + freeaddrinfo(res); + + return ret; +} + static int disconnect_ipcp(pid_t pid, const char * dst, const char * component) @@ -882,6 +1082,22 @@ static void __cleanup_flow(void * flow) reg_destroy_flow(((struct flow_info *) flow)->id); } +/* + * A PoA flow is secured as the IPCP that owns it: trust is in the + * peer itself. Without a security config for that name the flow + * stays in the clear. + */ +static void poa_name_info(const char * name, + struct name_info * info) +{ + if (reg_get_name_info(name, info) == 0) + return; + + memset(info, 0, sizeof(*info)); + + strcpy(info->name, name); +} + static int flow_accept(struct flow_info * flow, buffer_t * data, struct timespec * abstime, @@ -942,7 +1158,9 @@ static int flow_accept(struct flow_info * flow, goto fail_oap; } - if (reg_get_name_info(name, &info) < 0) { + if (reg_flow_is_poa(flow->id)) { + poa_name_info(name, &info); + } else if (reg_get_name_info(name, &info) < 0) { log_err("Failed to get name info for %s.", name); err = -ENAME; goto fail_oap; @@ -1262,7 +1480,7 @@ static int flow_alloc_direct(const char * dst, return -EAGAIN; } - if (oap_cli_prepare(&ctx, info, &req_hdr, *data, false) < 0) { + if (oap_cli_prepare(&ctx, info, NULL, &req_hdr, *data, false) < 0) { log_err("Failed to prepare OAP for %s.", dst); return -EBADF; } @@ -1295,7 +1513,7 @@ static int flow_alloc_direct(const char * dst, return -ETIMEDOUT; } - err = oap_cli_complete(ctx, info, resp_hdr, data, sk, NULL, NULL); + err = oap_cli_complete(ctx, resp_hdr, data, sk, NULL, NULL); if (err < 0) { log_err("OAP completion failed for %s.", dst); freebuf(resp_hdr); @@ -1378,7 +1596,7 @@ static int flow_alloc(const char * dst, goto fail_prepare; } - if (oap_cli_prepare(&ctx, &info, &req_hdr, *data, false) < 0) { + if (oap_cli_prepare(&ctx, &info, NULL, &req_hdr, *data, false) < 0) { log_err("Failed to prepare OAP request for %s.", dst); err = -EBADF; goto fail_prepare; @@ -1410,7 +1628,7 @@ static int flow_alloc(const char * dst, goto fail_peer; } - err = oap_cli_complete(ctx, &info, resp_hdr, data, sk, NULL, &peer_crt); + err = oap_cli_complete(ctx, resp_hdr, data, sk, NULL, &peer_crt); if (err < 0) { log_err("OAP completion failed for %s.", dst); goto fail_complete; @@ -1444,6 +1662,199 @@ static int flow_alloc(const char * dst, return err; } +/* Creates the flow and prepares the key exchange the IPCP will carry. */ +/* The N-1 is only known once the flow exists. */ +static int poa_alloc(struct flow_info * flow, + const char * dst, + buffer_t * data) +{ + struct ipcp_info ipcp; + struct layer_info layer; + struct name_info info; + buffer_t req_hdr = BUF_INIT; + void * ctx; + int err; + + ipcp.pid = flow->n_pid; + if (reg_get_ipcp(&ipcp, &layer) < 0) { + log_err("No IPCP with pid %d.", flow->n_pid); + + err = -EIPCP; + goto fail_flow; + } + + flow->n_1_pid = 0; + if (reg_create_flow(flow) < 0) { + log_err("Failed to create PoA flow."); + + err = -EBADF; + goto fail_flow; + } + + reg_set_name_for_flow_id(ipcp.name, flow->id); + + if (reg_flow_set_poa(flow->id) < 0) { + err = -EBADF; + goto fail_prepare; + } + + flow->uid = reg_get_proc_uid(flow->n_pid); + flow->n_1_pid = flow->n_pid; /* the IPCP is its own N-1 */ + if (reg_prepare_flow_alloc(flow) < 0) { + log_err("Failed to prepare PoA flow allocation."); + + err = -EBADF; + goto fail_prepare; + } + + poa_name_info(ipcp.name, &info); + + if (oap_cli_prepare(&ctx, &info, dst, &req_hdr, *data, false) < 0) { + log_err("Failed to prepare OAP request for %s.", ipcp.name); + + err = -EBADF; + goto fail_prepare; + } + + if (reg_flow_set_oap_ctx(flow->id, ctx) < 0) { + err = -EBADF; + goto fail_ctx; + } + + log_info("Allocating PoA flow %d for %d as %s.", + flow->id, flow->n_pid, ipcp.name); + + *data = req_hdr; + + return 0; + + fail_ctx: + freebuf(req_hdr); + oap_ctx_free(ctx); + fail_prepare: + reg_destroy_flow(flow->id); + fail_flow: + return err; +} + +/* Completes the key exchange once the IPCP has the peer's response. */ +/* + * We present the PoA's own credentials and expect the peer to present + * dst's: the address only says how to reach it. oap_cli_complete + * frees the ctx on every path. + */ +static int poa_complete(struct flow_info * flow, + int response, + buffer_t * data, + struct crypt_sk * sk) +{ + struct name_info info; + buffer_t peer_crt = BUF_INIT; + buffer_t out = BUF_INIT; + buffer_t empty = BUF_INIT; + char name[NAME_SIZE + 1]; + void * ctx; + int err; + + ctx = reg_flow_take_oap_ctx(flow->id); + if (ctx == NULL) { + log_err("No pending PoA flow %d.", flow->id); + return -EBADF; + } + + if (response < 0) { + log_dbg("PoA flow %d refused: %d.", flow->id, response); + + err = response; + goto fail; + } + + if (reg_get_name_for_flow_id(name, flow->id) < 0) { + err = -EBADF; + goto fail; + } + + poa_name_info(name, &info); + + err = oap_cli_complete(ctx, *data, &out, sk, NULL, &peer_crt); + if (err < 0) { + log_err("OAP completion failed for %s.", name); + + ctx = NULL; + goto fail; + } + + if (sk->nid != NID_undef) + reg_flow_set_rekey(flow->id, true, peer_crt); + + flow->state = FLOW_ALLOCATED; + if (reg_respond_alloc(flow, &empty, 0) < 0) { + log_err("Failed to update PoA flow %d.", flow->id); + + err = -EBADF; + goto fail_crt; + } + + log_info("PoA flow %d allocated to %s.", flow->id, name); + + freebuf(peer_crt); + freebuf(out); + + return 0; + + fail_crt: + freebuf(peer_crt); + freebuf(out); + fail: + oap_ctx_free(ctx); + reg_destroy_flow(flow->id); + return err; +} + +/* A peer requested a flow on a PoA of this IPCP. */ +static int poa_req_arr(struct flow_info * flow, + buffer_t * data) +{ + struct ipcp_info ipcp; + struct layer_info layer; + int ret; + + ipcp.pid = flow->n_1_pid; + if (reg_get_ipcp(&ipcp, &layer) < 0) { + log_err("No IPCP with pid %d.", flow->n_1_pid); + + ret = -EIPCP; + goto fail; + } + + log_info("PoA flow request arrived for %s.", ipcp.name); + + ret = wait_for_accept(ipcp.name); + if (ret < 0) { + log_err("No active process for %s.", ipcp.name); + goto fail; + } + + flow->id = ret; + flow->state = FLOW_ALLOCATED; + if (reg_flow_set_poa(flow->id) < 0) { + ret = -EBADF; + goto fail; + } + + reg_set_name_for_flow_id(ipcp.name, flow->id); + + ret = reg_respond_accept(flow, data); + if (ret < 0) { + log_err("Failed to respond to PoA flow %d.", flow->id); + goto fail; + } + + return 0; + fail: + return ret; +} + static int flow_alloc_reply(struct flow_info * flow, int response, buffer_t * data) @@ -1596,6 +2007,11 @@ static int rekey_name_info(int flow_id, if (reg_get_name_for_flow_id(name, flow_id) < 0) return -1; + if (reg_flow_is_poa(flow_id)) { + poa_name_info(name, info); + return 0; + } + return reg_get_name_info(name, info); } @@ -1667,7 +2083,7 @@ static void rekey_do_initiate(struct list_head * tbl, goto fail; } - if (oap_cli_prepare(&ctx, &name, &req, data, true) < 0) { + if (oap_cli_prepare(&ctx, &name, NULL, &req, data, true) < 0) { log_err("Failed to prepare re-key for flow %d.", flow_id); goto fail; } @@ -1758,7 +2174,7 @@ static void rekey_do_complete(struct list_head * tbl, reg_flow_get_peer_crt(flow_id, &crt); /* oap_cli_complete frees the ctx on every path. */ - if (oap_cli_complete(e->ctx, &info, buf, &data, &sk, &crt, NULL) < 0) { + if (oap_cli_complete(e->ctx, buf, &data, &sk, &crt, NULL) < 0) { log_warn("Failed to complete re-key for flow %d.", flow_id); e->ctx = NULL; goto finish_clear; @@ -1933,7 +2349,7 @@ static void rekey_do_direct(int flow_id) return; } - if (oap_cli_prepare(&ctx, &info, &req, data, true) < 0) { + if (oap_cli_prepare(&ctx, &info, NULL, &req, data, true) < 0) { log_err("Failed to prepare re-key for flow %d.", flow_id); reg_flow_clear_in_flight(flow_id); return; @@ -2307,6 +2723,10 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg, pid_t cpid; irm_msg_t * ret_msg; buffer_t data; + struct poa_addr poa_addr; + struct poa_addr * pa; + struct poa_spec poa_spec; + struct poa_spec * eps = NULL; memset(&flow, 0, sizeof(flow)); @@ -2350,11 +2770,74 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg, res = bootstrap_ipcp(msg->pid, &conf); break; case IRM_MSG_CODE__IRM_ENROLL_IPCP: - res = enroll_ipcp(msg->pid, msg->dst); + if (msg->peer != NULL) { + poa_addr = poa_addr_msg_to_s(msg->peer); + if (poa_addr.type == POA_INVALID) { + res = -EINVAL; + break; + } + } + + if (msg->conf != NULL) + conf = ipcp_config_msg_to_s(msg->conf); + res = enroll_ipcp_resolve(msg->pid, msg->dst, + msg->peer != NULL ? &poa_addr : NULL); + break; + case IRM_MSG_CODE__IRM_ATTACH_IPCP: + poa_spec = poa_spec_msg_to_s(msg->poa); + if (poa_spec.type == POA_INVALID) { + res = -EINVAL; + break; + } + + res = attach_ipcp(msg->pid, &poa_spec, true); + break; + case IRM_MSG_CODE__IRM_DETACH_IPCP: + poa_spec = poa_spec_msg_to_s(msg->poa); + if (poa_spec.type == POA_INVALID) { + res = -EINVAL; + break; + } + + res = attach_ipcp(msg->pid, &poa_spec, false); + break; + case IRM_MSG_CODE__IRM_LIST_POAS: + res = list_poas(msg->pid, &eps); + if (res > 0) { + ssize_t i; + ret_msg->poas = malloc(res * sizeof(*ret_msg->poas)); + if (ret_msg->poas == NULL) { + free(eps); + + res = -ENOMEM; + break; + } + for (i = 0; i < res; i++) { + ret_msg->poas[i] = poa_spec_s_to_msg(&eps[i]); + if (ret_msg->poas[i] == NULL) + break; + ret_msg->n_poas++; + } + res = i == res ? 0 : -ENOMEM; + } + free(eps); break; case IRM_MSG_CODE__IRM_CONNECT_IPCP: flow.qs = qos_spec_msg_to_s(msg->qosspec); - res = connect_ipcp(msg->pid, msg->dst, msg->comp, flow.qs); + + pa = NULL; + + if (msg->peer != NULL) { + poa_addr = poa_addr_msg_to_s(msg->peer); + if (poa_addr.type == POA_INVALID) { + res = -EINVAL; + break; + } + pa = &poa_addr; + } + + res = connect_ipcp_resolve(msg->pid, msg->dst, msg->comp, + flow.qs, pa); break; case IRM_MSG_CODE__IRM_DISCONNECT_IPCP: res = disconnect_ipcp(msg->pid, msg->dst, msg->comp); @@ -2499,6 +2982,67 @@ static irm_msg_t * do_command_msg(irm_msg_t * msg, if (res == 0) ret_msg->flow_info = flow_info_s_to_msg(&flow); break; + case IRM_MSG_CODE__IRM_POA_FLOW_ALLOC: + flow = flow_info_msg_to_s(msg->flow_info); + + clrbuf(data); + + res = poa_alloc(&flow, msg->dst, &data); + if (res == 0) { + ret_msg->flow_info = flow_info_s_to_msg(&flow); + ret_msg->has_pk = true; + ret_msg->pk.len = data.len; + ret_msg->pk.data = data.data; + + clrbuf(data); + } + break; + case IRM_MSG_CODE__IRM_POA_FLOW_ALLOC_R: + data.len = msg->pk.len; + data.data = msg->pk.data; + msg->pk.data = NULL; /* pass data */ + msg->pk.len = 0; + flow = flow_info_msg_to_s(msg->flow_info); + sk.key = kbuf; + res = poa_complete(&flow, msg->response, &data, &sk); + + freebuf(data); + + if (res != 0) + break; + + ret_msg->flow_info = flow_info_s_to_msg(&flow); + ret_msg->has_cipher_nid = true; + ret_msg->cipher_nid = sk.nid; + + if (sk.nid == NID_undef) + break; + + hbuf = malloc(SYMMKEYSZ); + if (hbuf == NULL) { + log_err("Failed to malloc key buf"); + + res = -ENOMEM; + break; + } + + memcpy(hbuf, kbuf, SYMMKEYSZ); + + ret_msg->sym_key.data = hbuf; + ret_msg->sym_key.len = SYMMKEYSZ; + ret_msg->has_sym_key = true; + break; + case IRM_MSG_CODE__IPCP_POA_FLOW_REQ_ARR: + data.len = msg->pk.len; + data.data = msg->pk.data; + msg->pk.data = NULL; /* pass data */ + msg->pk.len = 0; + flow = flow_info_msg_to_s(msg->flow_info); + + res = poa_req_arr(&flow, &data); + if (res == 0) + ret_msg->flow_info = flow_info_s_to_msg(&flow); + break; case IRM_MSG_CODE__IPCP_FLOW_ALLOC_REPLY: data.len = msg->pk.len; data.data = msg->pk.data; diff --git a/src/irmd/oap.h b/src/irmd/oap.h index 86f11e21..e9d7511b 100644 --- a/src/irmd/oap.h +++ b/src/irmd/oap.h @@ -40,21 +40,21 @@ int oap_auth_add_ca_crt(void * crt); int oap_auth_add_chain_crt(void * crt); /* -* Prepare OAP request header for server, returns context -* Passes client data for srv, returns srv data for client -* rekey forces ephemeral server-encap KEX (no client-encap; preserves FS/PCS) -*/ + * Prepares the request header and returns the context. info holds the + * credentials we present, peer the name the server certificate must + * carry (NULL expects info->name); rekey forces an ephemeral KEX. + */ int oap_cli_prepare(void ** ctx, const struct name_info * info, + const char * peer, buffer_t * req_buf, buffer_t data, bool rekey); /* - * Server processes header, creates response header, returns secret key. - * data is in/out: input=srv data to send, output=cli data received. - * rekey drops the cert and verifies against cached_crt; peer_crt (or NULL) - * receives a copy of the peer cert to cache at the initial handshake. + * Answers the request header and returns the secret key. data is + * in/out; rekey verifies against cached_crt, and peer_crt takes a copy + * of the peer cert to cache at the initial handshake. */ int oap_srv_process(const struct name_info * info, buffer_t req_buf, @@ -70,13 +70,12 @@ int oap_srv_process(const struct name_info * info, * cached_crt verifies a cert-less re-key; peer_crt (or NULL) receives a * copy of the peer cert to cache at the initial handshake. */ -int oap_cli_complete(void * ctx, - const struct name_info * info, - buffer_t rsp_buf, - buffer_t * data, - struct crypt_sk * sk, - const buffer_t * cached_crt, - buffer_t * peer_crt); +int oap_cli_complete(void * ctx, + buffer_t rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + const buffer_t * cached_crt, + buffer_t * peer_crt); /* Free OAP state (on failure before complete) */ void oap_ctx_free(void * ctx); diff --git a/src/irmd/oap/cli.c b/src/irmd/oap/cli.c index ebfcd71f..2203596f 100644 --- a/src/irmd/oap/cli.c +++ b/src/irmd/oap/cli.c @@ -50,6 +50,7 @@ struct oap_cli_ctx { uint8_t __id[OAP_ID_SIZE]; buffer_t id; + char peer[NAME_SIZE + 1]; /* expected server name */ uint8_t kex_buf[CRYPT_KEY_BUFSZ]; uint8_t req_hash[MAX_HASH_SIZE]; size_t req_hash_len; @@ -241,6 +242,7 @@ static int do_client_kex_prepare(const char * server_name, int oap_cli_prepare(void ** ctx, const struct name_info * info, + const char * peer, buffer_t * req_buf, buffer_t data, bool rekey) @@ -259,6 +261,11 @@ int oap_cli_prepare(void ** ctx, clrbuf(*req_buf); *ctx = NULL; + if (peer != NULL && strlen(peer) > NAME_SIZE) { + log_err("Peer name too long."); + return -EINVAL; + } + /* Allocate ctx to carry between prepare and complete */ s = malloc(sizeof(*s)); if (s == NULL) { @@ -269,13 +276,15 @@ int oap_cli_prepare(void ** ctx, memset(s, 0, sizeof(*s)); OAP_CLI_CTX_INIT(s); + strcpy(s->peer, peer != NULL ? peer : info->name); + /* Generate session ID */ if (random_buffer(s->__id, OAP_ID_SIZE) < 0) { log_err("Failed to generate OAP session ID."); goto fail_id; } - log_dbg_id(s->id.data, "Preparing OAP request for %s.", info->name); + log_dbg_id(s->id.data, "Preparing OAP request for %s.", s->peer); /* Load client credentials */ if (load_cli_credentials(info, &pkp, &crt) < 0) { @@ -315,7 +324,7 @@ int oap_cli_prepare(void ** ctx, oap_hdr_init(&s->local_hdr, s->id, s->kex_buf, data, s->scfg.c.nid); - if (do_client_kex_prepare(info->name, s) < 0) { + if (do_client_kex_prepare(s->peer, s) < 0) { log_err_id(s->id.data, "Failed to prepare client KEX."); goto fail_kex; } @@ -482,7 +491,8 @@ static int do_client_kex_complete(struct oap_cli_ctx * s, } SET_KEX_CIPHER(scfg, peer_hdr->cipher_str); - if (crypt_validate_nid(scfg->c.nid) < 0) { + + if (crypt_cipher_rank(scfg->c.nid) < 0) { log_err_id(id, "Server cipher '%s' not supported.", peer_hdr->cipher_str); return -ENOTSUP; @@ -518,13 +528,12 @@ static int do_client_kex_complete(struct oap_cli_ctx * s, return do_client_kex_complete_dhe(s, peer_hdr, sk); } -int oap_cli_complete(void * ctx, - const struct name_info * info, - buffer_t rsp_buf, - buffer_t * data, - struct crypt_sk * sk, - const buffer_t * cached_crt, - buffer_t * peer_crt) +int oap_cli_complete(void * ctx, + buffer_t rsp_buf, + buffer_t * data, + struct crypt_sk * sk, + const buffer_t * cached_crt, + buffer_t * peer_crt) { struct oap_cli_ctx * s = ctx; struct oap_hdr peer_hdr; @@ -538,7 +547,6 @@ int oap_cli_complete(void * ctx, int rc; assert(ctx != NULL); - assert(info != NULL); assert(data != NULL); assert(sk != NULL); @@ -550,7 +558,7 @@ int oap_cli_complete(void * ctx, id = s->id.data; - log_dbg_id(id, "Completing OAP for %s.", info->name); + log_dbg_id(id, "Completing OAP for %s.", s->peer); /* Decode response header using client's md_nid for hash length */ if (oap_hdr_decode(&peer_hdr, rsp_buf, s->req_md_nid, false) < 0) { @@ -616,9 +624,9 @@ int oap_cli_complete(void * ctx, } /* Verify peer certificate name matches expected destination */ - if (peer_hdr.crt.len > 0 && strcmp(peer, info->name) != 0) { + if (peer_hdr.crt.len > 0 && strcmp(peer, s->peer) != 0) { log_err_id(id, "Peer crt for '%s' does not match '%s'.", - peer, info->name); + peer, s->peer); goto fail_oap; } @@ -667,7 +675,7 @@ int oap_cli_complete(void * ctx, goto fail_oap; } - log_info_id(id, "OAP completed for %s.", info->name); + log_info_id(id, "OAP completed for %s.", s->peer); freebuf(peer_hdr.sealed_pt); diff --git a/src/irmd/oap/tests/common.c b/src/irmd/oap/tests/common.c index 16d52c63..b65f3997 100644 --- a/src/irmd/oap/tests/common.c +++ b/src/irmd/oap/tests/common.c @@ -159,7 +159,7 @@ void oap_test_teardown(struct oap_test_ctx * ctx) if (ctx->cli.state != NULL) { res.key = ctx->cli.key; - oap_cli_complete(ctx->cli.state, &ctx->cli.info, dummy, + oap_cli_complete(ctx->cli.state, dummy, &ctx->data, &res, NULL, NULL); ctx->cli.state = NULL; } @@ -179,8 +179,8 @@ void oap_test_teardown(struct oap_test_ctx * ctx) int oap_cli_prepare_ctx(struct oap_test_ctx * ctx) { - return oap_cli_prepare(&ctx->cli.state, &ctx->cli.info, &ctx->req_hdr, - ctx->data, ctx->rekey); + return oap_cli_prepare(&ctx->cli.state, &ctx->cli.info, NULL, + &ctx->req_hdr, ctx->data, ctx->rekey); } int oap_srv_process_ctx(struct oap_test_ctx * ctx) @@ -203,8 +203,7 @@ int oap_cli_complete_ctx(struct oap_test_ctx * ctx) struct crypt_sk res = { .nid = NID_undef, .key = ctx->cli.key }; int ret; - ret = oap_cli_complete(ctx->cli.state, &ctx->cli.info, ctx->resp_hdr, - &ctx->data, &res, + ret = oap_cli_complete(ctx->cli.state, ctx->resp_hdr, &ctx->data, &res, ctx->rekey ? &ctx->cli_crt : NULL, ctx->rekey ? NULL : &ctx->cli_crt); ctx->cli.state = NULL; @@ -527,7 +526,7 @@ int roundtrip_kex_only(void) goto fail; } - if (oap_cli_prepare(&cli_state, &cli_info, &req_hdr, + if (oap_cli_prepare(&cli_state, &cli_info, NULL, &req_hdr, data, false) < 0) { printf("Client prepare failed.\n"); goto fail_cleanup; @@ -545,7 +544,7 @@ int roundtrip_kex_only(void) res.key = cli_key; - if (oap_cli_complete(cli_state, &cli_info, resp_hdr, &data, &res, + if (oap_cli_complete(cli_state, resp_hdr, &data, &res, NULL, NULL) < 0) { printf("Client complete failed.\n"); cli_state = NULL; @@ -575,8 +574,8 @@ int roundtrip_kex_only(void) fail_cleanup: if (cli_state != NULL) { res.key = cli_key; - oap_cli_complete(cli_state, &cli_info, resp_hdr, &data, - &res, NULL, NULL); + + oap_cli_complete(cli_state, resp_hdr, &data, &res, NULL, NULL); } freebuf(resp_hdr); freebuf(req_hdr); @@ -655,7 +654,7 @@ int corrupted_response(const char * root_ca, res.key = ctx.cli.key; - if (oap_cli_complete(ctx.cli.state, &ctx.cli.info, ctx.resp_hdr, + if (oap_cli_complete(ctx.cli.state, ctx.resp_hdr, &ctx.data, &res, NULL, NULL) == 0) { printf("Client should reject corrupted response.\n"); ctx.cli.state = NULL; diff --git a/src/irmd/reg/flow.c b/src/irmd/reg/flow.c index 8be2dfc7..63c9199b 100644 --- a/src/irmd/reg/flow.c +++ b/src/irmd/reg/flow.c @@ -126,6 +126,9 @@ static int create_rbuffs(struct reg_flow * flow, assert(flow->n_1_rb == NULL); flow->info.n_1_pid = info->n_1_pid; + if (flow->poa) + return 0; + flow->n_1_rb = ssm_rbuff_create(info->n_1_pid, info->id); if (flow->n_1_rb == NULL) goto fail_n_1_rb; diff --git a/src/irmd/reg/flow.h b/src/irmd/reg/flow.h index 166bed61..c7021a0f 100644 --- a/src/irmd/reg/flow.h +++ b/src/irmd/reg/flow.h @@ -48,6 +48,8 @@ struct reg_flow { char name[NAME_SIZE + 1]; bool direct; + bool poa; /* transport is a point of attachment */ + void * oap_ctx; /* key exchange, prepare -> complete */ /* Tier-2 re-key state (encrypted flows only) */ struct { diff --git a/src/irmd/reg/name.c b/src/irmd/reg/name.c index a3621fc3..08426033 100644 --- a/src/irmd/reg/name.c +++ b/src/irmd/reg/name.c @@ -1,4 +1,3 @@ - /* * Ouroboros - Copyright (C) 2016 - 2026 * diff --git a/src/irmd/reg/reg.c b/src/irmd/reg/reg.c index ebf3959d..a302fa15 100644 --- a/src/irmd/reg/reg.c +++ b/src/irmd/reg/reg.c @@ -2119,6 +2119,77 @@ bool reg_flow_is_direct(int flow_id) return ret; } +bool reg_flow_is_poa(int flow_id) +{ + struct reg_flow * flow; + bool ret; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + + ret = flow != NULL && flow->poa; + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +int reg_flow_set_poa(int flow_id) +{ + struct reg_flow * flow; + int ret = -1; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + flow->poa = true; + ret = 0; + } + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +int reg_flow_set_oap_ctx(int flow_id, + void * ctx) +{ + struct reg_flow * flow; + int ret = -1; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + flow->oap_ctx = ctx; + ret = 0; + } + + pthread_mutex_unlock(®.mtx); + + return ret; +} + +void * reg_flow_take_oap_ctx(int flow_id) +{ + struct reg_flow * flow; + void * ctx = NULL; + + pthread_mutex_lock(®.mtx); + + flow = __reg_get_flow(flow_id); + if (flow != NULL) { + ctx = flow->oap_ctx; + flow->oap_ctx = NULL; + } + + pthread_mutex_unlock(®.mtx); + + return ctx; +} + void reg_flow_set_rekey(int flow_id, bool initiator, buffer_t peer_crt) @@ -2607,7 +2678,7 @@ void reg_notify_flow(int flow_id, pthread_mutex_unlock(®.mtx); } -/* Wake both endpoints of a direct flow (acceptor and allocator). */ +/* Wake both PoAs of a direct flow (acceptor and allocator). */ void reg_notify_flow_peers(int flow_id, int event) { diff --git a/src/irmd/reg/reg.h b/src/irmd/reg/reg.h index 8a313d46..6882532c 100644 --- a/src/irmd/reg/reg.h +++ b/src/irmd/reg/reg.h @@ -33,138 +33,147 @@ #include "pool.h" -int reg_init(void); +int reg_init(void); -void reg_clear(void); +void reg_clear(void); -void reg_fini(void); +void reg_fini(void); -int reg_create_flow(struct flow_info * info); +int reg_create_flow(struct flow_info * info); -int reg_destroy_flow(int flow_id); +int reg_destroy_flow(int flow_id); -bool reg_has_flow(int flow_id); +bool reg_has_flow(int flow_id); -int reg_create_proc(const struct proc_info * info); +int reg_create_proc(const struct proc_info * info); /* Use this for all processes, including ipcps */ -int reg_destroy_proc(pid_t pid); +int reg_destroy_proc(pid_t pid); -bool reg_has_proc(pid_t pid); +bool reg_has_proc(pid_t pid); -bool reg_is_proc_privileged(pid_t pid); +bool reg_is_proc_privileged(pid_t pid); -int reg_prepare_pool(uid_t uid, - gid_t gid); +int reg_prepare_pool(uid_t uid, + gid_t gid); -uid_t reg_get_proc_uid(pid_t pid); +uid_t reg_get_proc_uid(pid_t pid); -void reg_kill_all_proc(int signal); +void reg_kill_all_proc(int signal); -pid_t reg_get_dead_proc(void); +pid_t reg_get_dead_proc(void); -int reg_create_spawned(pid_t pid); +int reg_create_spawned(pid_t pid); -bool reg_has_spawned(pid_t pid); +bool reg_has_spawned(pid_t pid); -void reg_kill_all_spawned(int signal); +void reg_kill_all_spawned(int signal); -int reg_first_spawned(void); +int reg_first_spawned(void); -int reg_bind_proc(const char * name, - pid_t proc); +int reg_bind_proc(const char * name, + pid_t proc); -int reg_unbind_proc(const char * name, - pid_t proc); +int reg_unbind_proc(const char * name, + pid_t proc); -int reg_create_ipcp(const struct ipcp_info * info); +int reg_create_ipcp(const struct ipcp_info * info); -bool reg_has_ipcp(pid_t pid); +bool reg_has_ipcp(pid_t pid); -int reg_set_layer_for_ipcp(struct ipcp_info * info, - const struct layer_info * layer); +int reg_set_layer_for_ipcp(struct ipcp_info * info, + const struct layer_info * layer); -int reg_get_ipcp(struct ipcp_info * info, - struct layer_info * layer); +int reg_get_ipcp(struct ipcp_info * info, + struct layer_info * layer); -int reg_get_ipcp_by_layer(struct ipcp_info * info, - struct layer_info * layer); +int reg_get_ipcp_by_layer(struct ipcp_info * info, + struct layer_info * layer); /* TODO don't rely on protobuf here */ -int reg_list_ipcps(ipcp_list_msg_t *** msg); +int reg_list_ipcps(ipcp_list_msg_t *** msg); -int reg_create_name(const struct name_info * info); +int reg_create_name(const struct name_info * info); -int reg_destroy_name(const char * name); +int reg_destroy_name(const char * name); -bool reg_has_name(const char * name); +bool reg_has_name(const char * name); -int reg_get_name_info(const char * name, - struct name_info * info); +int reg_get_name_info(const char * name, + struct name_info * info); -int reg_get_name_for_hash(char * buf, - enum hash_algo algo, - const uint8_t * hash); +int reg_get_name_for_hash(char * buf, + enum hash_algo algo, + const uint8_t * hash); -int reg_get_name_for_flow_id(char * buf, - int flow_id); +int reg_get_name_for_flow_id(char * buf, + int flow_id); -void reg_set_name_for_flow_id(const char * name, - int flow_id); +void reg_set_name_for_flow_id(const char * name, + int flow_id); /* TODO don't rely on protobuf here */ -int reg_list_names(name_info_msg_t *** names); +int reg_list_names(name_info_msg_t *** names); -int reg_create_prog(const struct prog_info * info); +int reg_create_prog(const struct prog_info * info); -int reg_destroy_prog(const char * name); +int reg_destroy_prog(const char * name); -bool reg_has_prog(const char * name); +bool reg_has_prog(const char * name); -int reg_get_exec(const char * name, - char *** exec); +int reg_get_exec(const char * name, + char *** exec); -int reg_bind_prog(const char * name, - char ** exec, - uint8_t flags); +int reg_bind_prog(const char * name, + char ** exec, + uint8_t flags); -int reg_unbind_prog(const char * name, - const char * prog); +int reg_unbind_prog(const char * name, + const char * prog); -int reg_prepare_flow_alloc(struct flow_info * info); +int reg_prepare_flow_alloc(struct flow_info * info); -int reg_wait_flow_allocated(struct flow_info * info, +int reg_wait_flow_allocated(struct flow_info * info, + buffer_t * pbuf, + const struct timespec * abstime); + +int reg_respond_alloc(struct flow_info * info, + buffer_t * pbuf, + int response); + +int reg_prepare_flow_accept(struct flow_info * info); + +int reg_wait_flow_accepted(struct flow_info * info, buffer_t * pbuf, const struct timespec * abstime); -int reg_respond_alloc(struct flow_info * info, - buffer_t * pbuf, - int response); +int reg_wait_flow_accepting(const char * name, + const struct timespec * abstime); -int reg_prepare_flow_accept(struct flow_info * info); +int reg_respond_accept(struct flow_info * info, + buffer_t * pbuf); -int reg_wait_flow_accepted(struct flow_info * info, - buffer_t * pbuf, - const struct timespec * abstime); +int reg_prepare_flow_direct(struct flow_info * info, + buffer_t * pbuf, + uid_t alloc_uid); -int reg_wait_flow_accepting(const char * name, - const struct timespec * abstime); +int reg_respond_flow_direct(int flow_id, + buffer_t * pbuf); + +int reg_wait_flow_direct(int flow_id, + buffer_t * pbuf, + const struct timespec * abstime); -int reg_respond_accept(struct flow_info * info, - buffer_t * pbuf); +bool reg_flow_is_direct(int flow_id); -int reg_prepare_flow_direct(struct flow_info * info, - buffer_t * pbuf, - uid_t alloc_uid); +bool reg_flow_is_poa(int flow_id); -int reg_respond_flow_direct(int flow_id, - buffer_t * pbuf); +int reg_flow_set_poa(int flow_id); -int reg_wait_flow_direct(int flow_id, - buffer_t * pbuf, - const struct timespec * abstime); +int reg_flow_set_oap_ctx(int flow_id, + void * ctx); -bool reg_flow_is_direct(int flow_id); +void * reg_flow_take_oap_ctx(int flow_id); /* Per-flow snapshot for the re-key timer */ struct rekey_info { @@ -176,70 +185,70 @@ struct rekey_info { bool direct; }; -void reg_flow_set_rekey(int flow_id, - bool initiator, - buffer_t peer_crt); +void reg_flow_set_rekey(int flow_id, + bool initiator, + buffer_t peer_crt); -int reg_flow_get_peer_crt(int flow_id, - buffer_t * crt); +int reg_flow_get_peer_crt(int flow_id, + buffer_t * crt); -int reg_flow_get_epoch(int flow_id); +int reg_flow_get_epoch(int flow_id); -bool reg_flow_rekey_pending(int flow_id); +bool reg_flow_rekey_pending(int flow_id); -pid_t reg_flow_get_n_1_pid(int flow_id); +pid_t reg_flow_get_n_1_pid(int flow_id); -int reg_flow_snapshot_rekey_due(struct rekey_info * snap, - int max); +int reg_flow_snapshot_rekey_due(struct rekey_info * snap, + int max); -void reg_flow_clear_in_flight(int flow_id); +void reg_flow_clear_in_flight(int flow_id); -bool reg_flow_rekey_begin(int flow_id); +bool reg_flow_rekey_begin(int flow_id); -bool reg_flow_rekey_should_yield(int flow_id); +bool reg_flow_rekey_should_yield(int flow_id); -int reg_flow_store_pending(int flow_id, - const uint8_t * seed, - uint8_t epoch, - bool initiator); +int reg_flow_store_pending(int flow_id, + const uint8_t * seed, + uint8_t epoch, + bool initiator); -int reg_flow_store_pending_direct(int flow_id, - const uint8_t * seed, - uint8_t epoch); +int reg_flow_store_pending_direct(int flow_id, + const uint8_t * seed, + uint8_t epoch); -int reg_flow_take_pending(int flow_id, - uid_t uid, - pid_t cpid, - uint8_t * seed, - uint8_t * epoch, - bool * initiator); +int reg_flow_take_pending(int flow_id, + uid_t uid, + pid_t cpid, + uint8_t * seed, + uint8_t * epoch, + bool * initiator); -bool reg_flow_rekey_arr_admit(int flow_id, - pid_t n_1_pid, - bool is_req); +bool reg_flow_rekey_arr_admit(int flow_id, + pid_t n_1_pid, + bool is_req); -void reg_flow_rekey_arr_done(int flow_id, - bool is_req); +void reg_flow_rekey_arr_done(int flow_id, + bool is_req); -bool reg_flow_owned_by(int flow_id, - uid_t uid); +bool reg_flow_owned_by(int flow_id, + uid_t uid); -void reg_notify_flow(int flow_id, - int event); +void reg_notify_flow(int flow_id, + int event); -void reg_notify_flow_peers(int flow_id, - int event); +void reg_notify_flow_peers(int flow_id, + int event); -void reg_dealloc_flow(struct flow_info * info); +void reg_dealloc_flow(struct flow_info * info); -void reg_dealloc_flow_resp(struct flow_info * info); +void reg_dealloc_flow_resp(struct flow_info * info); -int reg_wait_proc(pid_t pid, - const struct timespec * abstime); +int reg_wait_proc(pid_t pid, + const struct timespec * abstime); -int reg_wait_ipcp_boot(struct ipcp_info * ipcp, - const struct timespec * abstime); +int reg_wait_ipcp_boot(struct ipcp_info * ipcp, + const struct timespec * abstime); -int reg_respond_ipcp(const struct ipcp_info * info); +int reg_respond_ipcp(const struct ipcp_info * info); #endif /* OUROBOROS_IRMD_REG_H */ diff --git a/src/irmd/reg/tests/reg_test.c b/src/irmd/reg/tests/reg_test.c index a8c1b1fa..ab57241c 100644 --- a/src/irmd/reg/tests/reg_test.c +++ b/src/irmd/reg/tests/reg_test.c @@ -746,7 +746,7 @@ static int test_reg_direct_flow_success(void) reg_dealloc_flow(&info); if (info.state != FLOW_DEALLOC_PENDING) { - printf("Same endpoint dealloc changed state.\n"); + printf("Same PoA dealloc changed state.\n"); goto fail; } diff --git a/src/lib/CMakeLists.txt b/src/lib/CMakeLists.txt index 3abf39d0..48064ce4 100644 --- a/src/lib/CMakeLists.txt +++ b/src/lib/CMakeLists.txt @@ -95,9 +95,9 @@ endif() if(HAVE_LIBURCU) target_link_libraries(ouroboros-common PRIVATE Urcu::Urcu) - # urcu headers require C99; override the global -std=c89 for this TU only. - set_source_files_properties(crypt/keyrot.c PROPERTIES - COMPILE_OPTIONS "-std=gnu99") + # urcu headers require C99; override the global -std=c89 for these TUs. + set_source_files_properties(crypt/keyrot.c dev.c poa/addr.c poa/eth.c + poa/udp.c PROPERTIES COMPILE_OPTIONS "-std=gnu99") endif() install(TARGETS ouroboros-common @@ -107,8 +107,14 @@ install(TARGETS ouroboros-common set(SOURCE_FILES_DEV cep.c dev.c + poa/addr.c + poa/udp.c ) +if(HAVE_ETH) + list(APPEND SOURCE_FILES_DEV poa/eth.c) +endif() + add_library(ouroboros-dev SHARED ${SOURCE_FILES_DEV} ${CEP_PROTO_SRCS}) @@ -131,6 +137,11 @@ target_include_directories(ouroboros-dev target_link_libraries(ouroboros-dev PUBLIC ouroboros-common) +if(HAVE_LIBURCU) + # dev.c calls urcu directly; do not rely on transitive linkage. + target_link_libraries(ouroboros-dev PRIVATE Urcu::Urcu) +endif() + install(TARGETS ouroboros-dev EXPORT OuroborosTargets LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}) diff --git a/src/lib/config.h.in b/src/lib/config.h.in index 4ebaa194..3c6985c3 100644 --- a/src/lib/config.h.in +++ b/src/lib/config.h.in @@ -79,6 +79,25 @@ #define PROC_RES_FDS @PROC_RES_FDS@ #define PROC_MAX_FQUEUES @PROC_MAX_FQUEUES@ +/* Flow endpoints */ +#define POA_MGMT_FRAME_SIZE @POA_MGMT_FRAME_SIZE@ +#define POA_MGMT_SND_TIMEO @POA_MGMT_SND_TIMEO@ +#define POA_MAX_POAS @POA_MAX_POAS@ +#define POA_UDP_MPL @POA_UDP_MPL@ +#define POA_UDP4_MTU @POA_UDP4_MTU@ +#define POA_UDP6_MTU @POA_UDP6_MTU@ +#define POA_UDP_RD_BUF @POA_UDP_RD_BUF@ +#cmakedefine HAVE_ETH +#cmakedefine HAVE_RAW_SOCKETS +#cmakedefine HAVE_BPF +#cmakedefine HAVE_NETMAP +#cmakedefine POA_ETH_QDISC_BYPASS +#define POA_ETH_MPL @POA_ETH_MPL@ +#define POA_ETH_SNDBUF @POA_ETH_SNDBUF@ +#define POA_ETH_RCVBUF @POA_ETH_RCVBUF@ +#define POA_ETH_LO_MTU @POA_ETH_LO_MTU@ +#define POA_ETH_RD_BUF @POA_ETH_RD_BUF@ + /* Default Delta-t parameters */ #cmakedefine FRCT_LINUX_RTT_ESTIMATOR #define DELT_A (@DELTA_T_ACK@) /* ms */ diff --git a/src/lib/dev.c b/src/lib/dev.c index 3fb8d831..eb706691 100644 --- a/src/lib/dev.c +++ b/src/lib/dev.c @@ -28,6 +28,9 @@ #include "config.h" #include "ssm.h" +#include "poa/poa.h" + +#define OUROBOROS_PREFIX "libouroboros" #include <ouroboros/atomics.h> #include <ouroboros/bitmap.h> @@ -45,6 +48,7 @@ #include <ouroboros/ipcp-dev.h> #include <ouroboros/list.h> #include <ouroboros/local-dev.h> +#include <ouroboros/logs.h> #include <ouroboros/np1_flow.h> #include <ouroboros/pthread.h> #include <ouroboros/random.h> @@ -63,8 +67,10 @@ #ifdef HAVE_LIBGCRYPT #include <gcrypt.h> #endif +#include <arpa/inet.h> #include <stdarg.h> #include <stdbool.h> +#include <stddef.h> #include <inttypes.h> #include <stdio.h> #include <stdlib.h> @@ -98,8 +104,8 @@ struct flow { ssize_t part_idx; struct crypt_ctx * crypt; - int headsz; /* Selector */ - int tailsz; /* Tag + CRC */ + int headsz; /* Selector */ + int tailsz; /* Tag + CRC */ struct timespec rk_grace; /* TX-promote deadline */ struct timespec rk_attempt; /* Last re-key attempt */ @@ -116,6 +122,8 @@ struct flow { struct timespec rcv_timeo; struct frcti * frcti; + + struct poa_flow * poa; /* NULL for shared memory flows */ }; struct flow_set { @@ -696,6 +704,9 @@ static void do_flow_fini(int fd) { assert(fd >= 0 && fd < PROC_MAX_FLOWS); + if (proc.flows[fd].poa != NULL) + poa_flow_detach(proc.flows[fd].poa); + if (proc.flows[fd].frcti != NULL) frcti_destroy(proc.flows[fd].frcti); @@ -750,9 +761,11 @@ static __inline__ size_t flow_user_mtu(const struct flow * flow, return raw > hdr ? raw - hdr : 0; } +/* A PoA flow transmits on its own socket; it has no tx ring. */ static int flow_init(struct flow_info * info, struct crypt_sk * sk, - time_t rtt_hint) + time_t rtt_hint, + struct poa_flow * pf) { struct timespec now; struct flow * flow; @@ -777,13 +790,15 @@ static int flow_init(struct flow_info * info, if (flow->rx_rb == NULL) goto fail_rx_rb; - flow->tx_rb = ssm_rbuff_open(info->n_1_pid, info->id); - if (flow->tx_rb == NULL) - goto fail_tx_rb; + if (pf == NULL) { + flow->tx_rb = ssm_rbuff_open(info->n_1_pid, info->id); + if (flow->tx_rb == NULL) + goto fail_tx_rb; - flow->set = ssm_flow_set_open(info->n_1_pid); - if (flow->set == NULL) - goto fail_set; + flow->set = ssm_flow_set_open(info->n_1_pid); + if (flow->set == NULL) + goto fail_set; + } flow->oflags = FLOWFDEFAULT; flow->part_idx = NO_PART; @@ -792,11 +807,14 @@ static int flow_init(struct flow_info * info, flow->crypt = NULL; flow->headsz = 0; flow->tailsz = 0; + flow->poa = pf; if (IS_ENCRYPTED(sk)) { flow->crypt = crypt_create_ctx(sk); - if (flow->crypt == NULL) + if (flow->crypt == NULL) { + err = -ECRYPT; goto fail_crypt; + } flow->headsz = crypt_get_headsz(flow->crypt); flow->tailsz = crypt_get_tagsz(flow->crypt); } @@ -815,6 +833,9 @@ static int flow_init(struct flow_info * info, proc.id_to_fd[info->id].fd = fd; + if (pf != NULL) + poa_flow_attach(pf, info->id, flow->rx_rb); + flow_set_state(&proc.id_to_fd[info->id], FLOW_ALLOCATED); pthread_rwlock_unlock(&proc.lock); @@ -824,9 +845,11 @@ static int flow_init(struct flow_info * info, fail_frcti: crypt_destroy_ctx(flow->crypt); fail_crypt: - ssm_flow_set_close(flow->set); + if (flow->set != NULL) + ssm_flow_set_close(flow->set); fail_set: - ssm_rbuff_close(flow->tx_rb); + if (flow->tx_rb != NULL) + ssm_rbuff_close(flow->tx_rb); fail_tx_rb: ssm_rbuff_close(flow->rx_rb); fail_rx_rb: @@ -1077,14 +1100,20 @@ static void fini(void) __attribute__((section(INIT_SECTION))) __typeof__(init) * __init = init; __attribute__((section(FINI_SECTION))) __typeof__(fini) * __fini = fini; +/* + * A PoA flow is announced by its peer before the accept; from the + * reply on, the peer may transmit, so the flow must be able to + * receive. + */ int flow_accept(qosspec_t * qs, const struct timespec * timeo) { struct flow_info flow; - struct crypt_sk crypt; + struct crypt_sk crypt; uint8_t buf[SOCK_BUF_SIZE]; buffer_t msg = {SOCK_BUF_SIZE, buf}; uint8_t key[SYMMKEYSZ]; + struct poa_flow * pf; int fd; int err; @@ -1116,8 +1145,13 @@ int flow_accept(qosspec_t * qs, if (err < 0) return err; - /* No RTT in accept; rtt_hint=0 bootstraps from first ACK. */ - fd = flow_init(&flow, &crypt, 0); + pf = poa_flow_take_pending(flow.id); + + fd = flow_init(&flow, &crypt, 0, pf); + if (fd >= 0) + poa_flow_ready(pf); + else if (pf != NULL) + poa_flow_detach(pf); crypt_secure_clear(key, SYMMKEYSZ); @@ -1173,7 +1207,7 @@ int flow_alloc(const char * dst, if (err < 0) return err; - fd = flow_init(&flow, &crypt, ts_diff_ns(&t1, &t0)); + fd = flow_init(&flow, &crypt, ts_diff_ns(&t1, &t0), NULL); crypt_secure_clear(key, SYMMKEYSZ); @@ -1214,7 +1248,7 @@ int flow_join(const char * dst, if (err < 0) return err; - fd = flow_init(&flow, &crypt, 0); + fd = flow_init(&flow, &crypt, 0, NULL); crypt_secure_clear(key, SYMMKEYSZ); @@ -1275,7 +1309,8 @@ int flow_dealloc(int fd) pthread_cleanup_push(__cleanup_rwlock_unlock, &proc.lock); - ssm_rbuff_fini(flow->tx_rb); + if (flow->tx_rb != NULL) + ssm_rbuff_fini(flow->tx_rb); pthread_cleanup_pop(true); @@ -1428,7 +1463,11 @@ int fccntl(int fd, break; case FLOWGTXQLEN: qlen = va_arg(l, size_t *); - *qlen = ssm_rbuff_queued(flow->tx_rb); + + if (flow->poa != NULL) + *qlen = poa_flow_qpkts(flow->poa); + else + *qlen = ssm_rbuff_queued(flow->tx_rb); break; case FLOWGMTU: maxp = va_arg(l, size_t *); @@ -1456,16 +1495,18 @@ int fccntl(int fd, if (flow->oflags & FLOWFDOWN) { ssm_rbuff_set_bits(flow->rx_rb, RB_FLOWDOWN); - ssm_rbuff_set_bits(flow->tx_rb, RB_FLOWDOWN); - ssm_flow_set_notify(flow->set, - flow->info.id, - FLOW_DOWN); + if (flow->tx_rb != NULL) + ssm_rbuff_set_bits(flow->tx_rb, RB_FLOWDOWN); + if (flow->set != NULL) + ssm_flow_set_notify(flow->set, flow->info.id, + FLOW_DOWN); } else { ssm_rbuff_clr_bits(flow->rx_rb, RB_FLOWDOWN); - ssm_rbuff_clr_bits(flow->tx_rb, RB_FLOWDOWN); - ssm_flow_set_notify(flow->set, - flow->info.id, - FLOW_UP); + if (flow->tx_rb != NULL) + ssm_rbuff_clr_bits(flow->tx_rb, RB_FLOWDOWN); + if (flow->set != NULL) + ssm_flow_set_notify(flow->set, flow->info.id, + FLOW_UP); } break; @@ -1599,6 +1640,9 @@ static int flow_tx_spb(struct flow * flow, goto enomem; } + if (flow->poa != NULL) + return poa_flow_tx(flow->poa, spb, block, abstime); + if (!block) ret = ssm_rbuff_write(flow->tx_rb, idx); else @@ -2450,7 +2494,7 @@ int np1_flow_alloc(pid_t n_pid, /* np1 flow: n_1_pid is the upper. */ flow.n_1_pid = n_pid; - return flow_init(&flow, &crypt, 0); + return flow_init(&flow, &crypt, 0, NULL); } int np1_flow_dealloc(int flow_id, @@ -2584,7 +2628,7 @@ int ipcp_flow_req_arr(const buffer_t * dst, crypt.nid = NID_undef; - return flow_init(&flow, &crypt, 0); + return flow_init(&flow, &crypt, 0, NULL); } int ipcp_flow_update_arr(int flow_id, @@ -2853,11 +2897,13 @@ int ipcp_flow_fini(int fd) } ssm_rbuff_set_bits(proc.flows[fd].rx_rb, RB_FLOWDOWN); - ssm_rbuff_set_bits(proc.flows[fd].tx_rb, RB_FLOWDOWN); - ssm_flow_set_notify(proc.flows[fd].set, - proc.flows[fd].info.id, - FLOW_DEALLOC); + if (proc.flows[fd].tx_rb != NULL) + ssm_rbuff_set_bits(proc.flows[fd].tx_rb, RB_FLOWDOWN); + + if (proc.flows[fd].set != NULL) + ssm_flow_set_notify(proc.flows[fd].set, proc.flows[fd].info.id, + FLOW_DEALLOC); rx_rb = proc.flows[fd].rx_rb; @@ -2892,9 +2938,30 @@ size_t ipcp_flow_queued(int fd) assert(fd >= 0 && fd < PROC_MAX_FLOWS); assert(proc.flows[fd].info.id >= 0); + if (proc.flows[fd].poa != NULL) + return poa_flow_qlen(proc.flows[fd].poa); + return ssm_rbuff_queued(proc.flows[fd].tx_rb); } +int ipcp_flow_queue_id(int fd) +{ + int qid; + + assert(fd >= 0 && fd < PROC_MAX_FLOWS); + assert(proc.flows[fd].info.id >= 0); + + if (proc.flows[fd].poa == NULL) + return fd; + + /* An unidentified PoA answers for itself, never for an fd. */ + qid = poa_flow_qid(proc.flows[fd].poa); + if (qid < 0 || qid >= POA_MAX_POAS) + return fd; + + return PROC_MAX_FLOWS + qid; +} + int local_flow_transfer(int src_fd, int dst_fd, struct ssm_pool * src_pool, @@ -2960,3 +3027,5 @@ int local_flow_transfer(int src_fd, return ret; } + +#include "poa/poa.c" diff --git a/src/lib/irm.c b/src/lib/irm.c index c62701aa..5d53841f 100644 --- a/src/lib/irm.c +++ b/src/lib/irm.c @@ -118,6 +118,8 @@ int irm_bootstrap_ipcp(pid_t pid, msg.has_pid = true; msg.pid = pid; msg.conf = ipcp_config_s_to_msg(conf); + if (msg.conf == NULL) + return -ENOMEM; recv_msg = send_recv_irm_msg(&msg); ipcp_config_msg__free_unpacked(msg.conf, NULL); @@ -135,10 +137,11 @@ int irm_bootstrap_ipcp(pid_t pid, return ret; } -int irm_connect_ipcp(pid_t pid, - const char * dst, - const char * component, - qosspec_t qs) +int irm_connect_ipcp(pid_t pid, + const char * dst, + const char * component, + qosspec_t qs, + const struct poa_addr * addr) { irm_msg_t msg = IRM_MSG__INIT; irm_msg_t * recv_msg; @@ -152,7 +155,18 @@ int irm_connect_ipcp(pid_t pid, msg.pid = pid; msg.qosspec = qos_spec_s_to_msg(&qs); + if (addr != NULL) { + msg.peer = poa_addr_s_to_msg(addr); + if (msg.peer == NULL) { + qosspec_msg__free_unpacked(msg.qosspec, NULL); + return -ENOMEM; + } + } + recv_msg = send_recv_irm_msg(&msg); + + if (msg.peer != NULL) + poa_addr_msg__free_unpacked(msg.peer, NULL); qosspec_msg__free_unpacked(msg.qosspec, NULL); if (recv_msg == NULL) @@ -245,8 +259,99 @@ ssize_t irm_list_ipcps(struct ipcp_list_info ** ipcps) return nr; } -int irm_enroll_ipcp(pid_t pid, - const char * dst) +static int irm_attach_detach(pid_t pid, + const struct poa_spec * poa, + int code) +{ + irm_msg_t msg = IRM_MSG__INIT; + irm_msg_t * recv_msg; + int ret; + + if (pid == -1 || poa == NULL) + return -EINVAL; + + msg.code = code; + msg.has_pid = true; + msg.pid = pid; + msg.poa = poa_spec_s_to_msg(poa); + if (msg.poa == NULL) + return -EINVAL; + + recv_msg = send_recv_irm_msg(&msg); + + poa_spec_msg__free_unpacked(msg.poa, NULL); + + if (recv_msg == NULL) + return -EIRMD; + + if (!recv_msg->has_result) { + irm_msg__free_unpacked(recv_msg, NULL); + return -EIRMD; + } + + ret = recv_msg->result; + + irm_msg__free_unpacked(recv_msg, NULL); + + return ret; +} + +ssize_t irm_list_poas(pid_t pid, + struct poa_spec ** poas) +{ + irm_msg_t msg = IRM_MSG__INIT; + irm_msg_t * recv_msg; + size_t nr; + size_t i; + + if (pid == -1 || poas == NULL) + return -EINVAL; + + *poas = NULL; + + msg.code = IRM_MSG_CODE__IRM_LIST_POAS; + msg.has_pid = true; + msg.pid = pid; + + recv_msg = send_recv_irm_msg(&msg); + if (recv_msg == NULL) + return -EIRMD; + + nr = recv_msg->n_poas; + if (nr == 0) { + irm_msg__free_unpacked(recv_msg, NULL); + return 0; + } + + *poas = malloc(nr * sizeof(**poas)); + if (*poas == NULL) { + irm_msg__free_unpacked(recv_msg, NULL); + return -ENOMEM; + } + + for (i = 0; i < nr; i++) + (*poas)[i] = poa_spec_msg_to_s(recv_msg->poas[i]); + + irm_msg__free_unpacked(recv_msg, NULL); + + return (ssize_t) nr; +} + +int irm_attach_ipcp(pid_t pid, + const struct poa_spec * poa) +{ + return irm_attach_detach(pid, poa, IRM_MSG_CODE__IRM_ATTACH_IPCP); +} + +int irm_detach_ipcp(pid_t pid, + const struct poa_spec * poa) +{ + return irm_attach_detach(pid, poa, IRM_MSG_CODE__IRM_DETACH_IPCP); +} + +int irm_enroll_ipcp(pid_t pid, + const char * dst, + const struct poa_addr * addr) { irm_msg_t msg = IRM_MSG__INIT; irm_msg_t * recv_msg = NULL; @@ -260,7 +365,17 @@ int irm_enroll_ipcp(pid_t pid, msg.pid = pid; msg.dst = (char *) dst; + if (addr != NULL) { + msg.peer = poa_addr_s_to_msg(addr); + if (msg.peer == NULL) + return -ENOMEM; + } + recv_msg = send_recv_irm_msg(&msg); + + if (msg.peer != NULL) + poa_addr_msg__free_unpacked(msg.peer, NULL); + if (recv_msg == NULL) return -EIRMD; diff --git a/src/lib/pb/ipcp.proto b/src/lib/pb/ipcp.proto index afee4f91..298a15e2 100644 --- a/src/lib/pb/ipcp.proto +++ b/src/lib/pb/ipcp.proto @@ -40,6 +40,9 @@ enum ipcp_msg_code { IPCP_DISCONNECT = 11; IPCP_REPLY = 12; IPCP_FLOW_UPDATE = 13; + IPCP_ATTACH = 14; + IPCP_DETACH = 15; + IPCP_LIST_POAS = 16; } message ipcp_msg { @@ -58,4 +61,8 @@ message ipcp_msg { optional sint32 mpl = 13; /* MPL in ms. */ optional int32 result = 14; optional uint32 uid = 15; /* 0 = GSPP, >0 = PUP uid */ + optional poa_addr_msg peer = 16; /* peer PoA address */ + optional bool is_poa = 17; /* flow rides a PoA */ + optional poa_spec_msg poa = 18; /* PoA to attach */ + repeated poa_spec_msg poas = 19; /* PoAs attached */ } diff --git a/src/lib/pb/ipcp_config.proto b/src/lib/pb/ipcp_config.proto index a111b6d8..f6d313ac 100644 --- a/src/lib/pb/ipcp_config.proto +++ b/src/lib/pb/ipcp_config.proto @@ -66,29 +66,32 @@ message uni_config_msg { required uint32 cong_avoid = 4; } -message eth_config_msg { +message eth_poa_msg { required string dev = 1; required uint32 ethertype = 2; + optional bytes mac = 3; } -message udp4_config_msg { - required uint32 ip_addr = 1; - required uint32 port = 2; - required uint32 dns_addr = 3; /* set to 0 if unused */ +message udp4_poa_msg { + required uint32 ip_addr = 1; + required uint32 port = 2; } -message udp6_config_msg { - required bytes ip_addr = 1; - required uint32 port = 2; - required bytes dns_addr = 3; /* set to NULL if unused */ +message udp6_poa_msg { + required bytes ip_addr = 1; + required uint32 port = 2; } +/* Exactly one field is set; presence is the PoA type. */ +message poa_spec_msg { + optional udp4_poa_msg udp4 = 1; + optional udp6_poa_msg udp6 = 2; + optional eth_poa_msg eth = 3; +} + message ipcp_config_msg { required layer_info_msg layer_info = 1; required uint32 ipcp_type = 2; optional uni_config_msg unicast = 3; - optional udp4_config_msg udp4 = 4; - optional udp6_config_msg udp6 = 5; - optional eth_config_msg eth = 6; } diff --git a/src/lib/pb/irm.proto b/src/lib/pb/irm.proto index f54bc9ea..8f594fa3 100644 --- a/src/lib/pb/irm.proto +++ b/src/lib/pb/irm.proto @@ -55,6 +55,12 @@ enum irm_msg_code { IRM_REPLY = 27; IRM_FLOW_UPDATE = 28; IPCP_FLOW_UPDATE_ARR = 29; + IRM_POA_FLOW_ALLOC = 30; + IRM_POA_FLOW_ALLOC_R = 31; + IPCP_POA_FLOW_REQ_ARR = 32; + IRM_ATTACH_IPCP = 33; + IRM_DETACH_IPCP = 34; + IRM_LIST_POAS = 35; } message timespec_msg { @@ -101,4 +107,7 @@ message irm_msg { optional uint32 generation = 28; /* re-key batch generation */ optional bool rekey = 29; /* re-key watermark trigger */ optional bool rk_initiator = 30; /* re-key proof-holder side */ + optional poa_addr_msg peer = 32; /* peer PoA address */ + optional poa_spec_msg poa = 33; /* PoA to attach */ + repeated poa_spec_msg poas = 34; /* PoAs attached */ } diff --git a/src/lib/pb/model.proto b/src/lib/pb/model.proto index 4c1564a5..e00f3e12 100644 --- a/src/lib/pb/model.proto +++ b/src/lib/pb/model.proto @@ -58,6 +58,19 @@ message layer_info_msg { required uint32 dir_hash_algo = 2; } +/* Address of a flow PoA peer. */ +message poa_addr_msg { + required uint32 type = 1; /* enum poa_type */ + optional uint32 ip4 = 2; + optional bytes ip6 = 3; + optional uint32 port = 4; + optional bytes mac = 5; + optional uint32 ethertype = 6; + optional string dev = 7; /* our device, eth only */ + optional string hostname = 8; /* IRMd resolves, UDP only */ + optional bytes src_mac = 9; /* our MAC, eth only */ +} + message ipcp_info_msg { required uint32 type = 1; required string name = 2; diff --git a/src/lib/poa/addr.c b/src/lib/poa/addr.c new file mode 100644 index 00000000..d8811244 --- /dev/null +++ b/src/lib/poa/addr.c @@ -0,0 +1,142 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Points of attachment (PoA) - addresses and management messages + * + * 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/. + */ + +#if defined(__linux__) || defined(__CYGWIN__) +#ifndef _DEFAULT_SOURCE /* Test include source */ +#define _DEFAULT_SOURCE +#endif +#endif + +#include "config.h" + +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> + +#include "poa.h" + +#include <arpa/inet.h> +#include <stdio.h> +#include <string.h> + +int poa_addr_cmp(const struct poa_addr * a, + const struct poa_addr * b) +{ + if (a->type != b->type) + return 1; + + switch (a->type) { + case POA_UDP4: + if (a->udp4.port != b->udp4.port) + return 1; + return memcmp(&a->udp4.ip_addr, &b->udp4.ip_addr, + sizeof(a->udp4.ip_addr)); + case POA_UDP6: + if (a->udp6.port != b->udp6.port) + return 1; + return memcmp(&a->udp6.ip_addr, &b->udp6.ip_addr, + sizeof(a->udp6.ip_addr)); + case POA_ETH: + if (a->eth.dst.ethertype != b->eth.dst.ethertype) + return 1; + return memcmp(a->eth.dst.mac, b->eth.dst.mac, POA_MAC_SIZE); + default: + return 1; + } +} + +/* Display/RIB entry name, e.g. "udp4.<ip>.<port>". */ +int poa_addr_name(const struct poa_addr * a, + char * buf, + size_t len) +{ + char ip[INET6_ADDRSTRLEN]; + int ret; + + switch (a->type) { + case POA_UDP4: + if (inet_ntop(AF_INET, &a->udp4.ip_addr, + ip, sizeof(ip)) == NULL) + return -EINVAL; + ret = snprintf(buf, len, "udp4.%s.%u", ip, a->udp4.port); + break; + case POA_UDP6: + if (inet_ntop(AF_INET6, &a->udp6.ip_addr, + ip, sizeof(ip)) == NULL) + return -EINVAL; + ret = snprintf(buf, len, "udp6.%s.%u", ip, a->udp6.port); + break; + case POA_ETH: + ret = snprintf(buf, len, "eth.%s.%04X", + a->eth.src.dev, a->eth.src.ethertype); + break; + default: + return -EINVAL; + } + + if (ret < 0 || (size_t) ret >= len) + return -EMSGSIZE; + + return 0; +} + +void poa_mgmt_msg_ser(struct poa_mgmt_msg * msg, + uint8_t code, + uint32_t s_eid, + uint32_t d_eid, + qosspec_t qs, + int response, + size_t data_len) +{ + memset(msg, 0, sizeof(*msg)); + + msg->code = code; + msg->s_eid = hton32(s_eid); + msg->d_eid = hton32(d_eid); + msg->response = hton32(response); + msg->data_len = hton16((uint16_t) data_len); + + if (code != POA_FLOW_REQ) + return; + + msg->bandwidth = hton64(qs.bandwidth); + msg->delay = hton32(qs.delay); + msg->loss = hton32(qs.loss); + msg->ber = hton32(qs.ber); + msg->max_gap = hton32(qs.max_gap); + msg->timeout = hton32(qs.timeout); + msg->availability = qs.availability; + msg->service = qs.service; +} + +void poa_mgmt_msg_qos(const struct poa_mgmt_msg * msg, + qosspec_t * qs) +{ + qs->bandwidth = ntoh64(msg->bandwidth); + qs->delay = ntoh32(msg->delay); + qs->loss = ntoh32(msg->loss); + qs->ber = ntoh32(msg->ber); + qs->max_gap = ntoh32(msg->max_gap); + qs->timeout = ntoh32(msg->timeout); + qs->availability = msg->availability; + qs->service = msg->service; +} + diff --git a/src/lib/poa/eth.c b/src/lib/poa/eth.c new file mode 100644 index 00000000..5e959ef6 --- /dev/null +++ b/src/lib/poa/eth.c @@ -0,0 +1,1987 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Points of attachment (PoA) - Ethernet transport + * + * 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/. + */ + +#if defined(__APPLE__) +#define _BSD_SOURCE +#define _DARWIN_C_SOURCE +#elif defined(__FreeBSD__) +#define __BSD_VISIBLE 1 +#elif defined(__linux__) || defined(__CYGWIN__) +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE +#endif +#else +#ifndef _POSIX_C_SOURCE +#define _POSIX_C_SOURCE 200809L +#endif +#endif + +#include "config.h" + +#define OUROBOROS_PREFIX "poa-eth" + +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> +#include <ouroboros/hash.h> +#include <ouroboros/logs.h> +#include <ouroboros/pthread.h> +#include <ouroboros/sockets.h> +#include <ouroboros/time.h> + +#include "poa.h" + +#ifdef HAVE_RAW_SOCKETS +#include <net/if.h> +#include <netinet/in.h> +#include <linux/if_ether.h> +#include <linux/if_packet.h> +#include <linux/netlink.h> +#include <linux/gen_stats.h> +#include <linux/pkt_sched.h> +#include <linux/rtnetlink.h> +#include <linux/sockios.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/uio.h> + +#include <ifaddrs.h> +#elif defined(HAVE_BPF) +#include <net/bpf.h> +#include <net/if.h> +#include <net/if_dl.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/uio.h> + +#include <fcntl.h> +#include <ifaddrs.h> +#elif defined(HAVE_NETMAP) +#define NETMAP_WITH_LIBS +#include <net/netmap_user.h> +#include <net/if.h> +#include <sys/ioctl.h> +#include <sys/socket.h> + +#include <poll.h> +#ifndef __linux__ +#include <net/if_dl.h> +#include <ifaddrs.h> +#endif +#endif + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#define ETH_TYPE_LEN_SIZE sizeof(uint16_t) +#define ETH_HDR_SIZE (2 * POA_MAC_SIZE + ETH_TYPE_LEN_SIZE) +#define POA_HDR_SIZE (sizeof(struct poa_hdr)) +#define ETH_HDR_TOT_SIZE (ETH_HDR_SIZE + POA_HDR_SIZE) + +#define BPF_DEV_MAX 256 /* /dev/bpfN units to try */ +/* Wait for the link to come back before reading it again. */ +#define ETH_DOWN_TIMEO 100 /* ms */ +/* Budget for a qdisc dump; the send path must not wait on netlink. */ +#define POA_QDISC_TIMEO 5 /* ms */ + +#ifndef ETH_MAX_MTU /* In if_ether.h as of Linux 4.10. */ +#define ETH_MAX_MTU 0xFFFFU +#endif + +struct poa_hdr { + uint16_t eid; + uint16_t len; + uint8_t hcs; +} __attribute__((packed)); + +struct eth_hdr { + uint8_t dst[POA_MAC_SIZE]; + uint8_t src[POA_MAC_SIZE]; + uint16_t ethertype; + struct poa_hdr poa; +} __attribute__((packed)); + + +struct eth_priv { + struct poa * poa; + + int s_fd; /* raw socket or bpf device */ + uint16_t ethertype; /* network order */ + uint8_t hw_addr[POA_MAC_SIZE]; + uint32_t mtu; /* device payload capacity */ + int ifindex; /* link id; 0 where unknown */ + + pthread_t reader; + bool running; +#ifdef HAVE_RAW_SOCKETS + struct sockaddr_ll device; +#elif defined(HAVE_BPF) + size_t blen; /* bpf device buffer size */ +#elif defined(HAVE_NETMAP) + struct nm_desc * nmd; + struct pollfd poll_in; + struct pollfd poll_out; +#endif + /* The kernel zeroes its counters on read, so accumulate. */ + size_t kern_rcv; + size_t kern_drp; +#ifdef HAVE_RAW_SOCKETS + /* Qdisc depth: netlink descriptor, its try-lock and cache. */ + int nl_fd; + uint8_t nl_busy; + size_t nl_pkt; + size_t nl_byt; + uint64_t nl_time; +#endif +}; + +struct eth_query { + struct list_head next; + uint8_t hash[POA_QUERY_HLEN]; + /* The resolve's constraints; replies must satisfy them. */ + uint16_t ethertype; + char c_dev[DEV_NAME_SIZE + 1]; + uint8_t c_mac[POA_MAC_SIZE]; + bool replied; + char dev[DEV_NAME_SIZE + 1]; + uint8_t src_mac[POA_MAC_SIZE]; + uint8_t mac[POA_MAC_SIZE]; + uint16_t r_ethertype; +}; + +static struct { + pthread_once_t once; + + struct llist pending; + pthread_mutex_t mtx; + pthread_cond_t cond; +} queries = { .once = PTHREAD_ONCE_INIT }; + +static void eth_hdr_ser(const struct eth_priv * priv, + struct eth_hdr * hdr, + const uint8_t * dst, + uint32_t eid, + size_t len) +{ + memcpy(hdr->dst, dst, POA_MAC_SIZE); + memcpy(hdr->src, priv->hw_addr, POA_MAC_SIZE); + + hdr->ethertype = priv->ethertype; + hdr->poa.eid = htons((uint16_t) eid); + hdr->poa.len = htons((uint16_t) len); + mem_hash(HASH_CRC8, &hdr->poa.hcs, + (uint8_t *) &hdr->poa.eid, 2 * sizeof(uint16_t)); +} + +/* Oriented from us to the sender: our PoA, then their MAC. */ +static void frame_to_addr(const struct eth_priv * priv, + const struct eth_hdr * hdr, + struct poa_addr * addr) +{ + memset(addr, 0, sizeof(*addr)); + + addr->type = priv->poa->type; + + addr->eth.src = priv->poa->local.eth.src; + + memcpy(addr->eth.dst.mac, hdr->src, POA_MAC_SIZE); + + addr->eth.dst.ethertype = ntohs(priv->ethertype); +} + +static bool frame_is_for_us(const struct eth_priv * priv, + const uint8_t * dst) +{ + static const uint8_t bc[POA_MAC_SIZE] = + { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; + + if (memcmp(dst, priv->hw_addr, POA_MAC_SIZE) == 0) + return true; + + return memcmp(dst, bc, POA_MAC_SIZE) == 0; +} + +/* + * Parse and validate a frame; on success *eid and *plen locate the + * payload. Errors are silent: anyone can spray an interface. + */ +static int frame_parse(const struct eth_priv * priv, + const uint8_t * buf, + size_t len, + uint32_t * eid, + size_t * plen) +{ + const struct eth_hdr * hdr = (const struct eth_hdr *) buf; + uint8_t hcs; + + if (len < ETH_HDR_TOT_SIZE) + return -1; + + if (!frame_is_for_us(priv, hdr->dst)) + return -1; + + if (hdr->ethertype != priv->ethertype) + return -1; + + mem_hash(HASH_CRC8, &hcs, + (const uint8_t *) &hdr->poa.eid, 2 * sizeof(uint16_t)); + + if (hcs != hdr->poa.hcs) + return -1; + + *eid = ntohs(hdr->poa.eid); + + *plen = ntohs(hdr->poa.len); + if (*plen > len - ETH_HDR_TOT_SIZE) + return -1; + + return 0; +} + +#define ETH_QUERY_TIMEO 1900 /* ms total budget */ +#define ETH_QUERY_RETRIES 3 /* retransmits, 4 attempts total */ + +/* A resolve in progress; a reply fills every matching entry. */ +static void queries_init(void) +{ + pthread_condattr_t cattr; + + llist_init(&queries.pending); + + pthread_mutex_init(&queries.mtx, NULL); + + pthread_condattr_init(&cattr); +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + pthread_cond_init(&queries.cond, &cattr); + pthread_condattr_destroy(&cattr); +} + +static void eth_query_cleanup(void * o) +{ + struct eth_query * q = (struct eth_query *) o; + + pthread_mutex_lock(&queries.mtx); + llist_del(&q->next, &queries.pending); + pthread_mutex_unlock(&queries.mtx); +} + +static void eth_query_arr(struct poa * poa, + const struct poa_addr * src, + const uint8_t * hash) +{ + uint8_t buf[sizeof(struct poa_mgmt_msg) + + POA_QUERY_HLEN]; + struct poa_mgmt_msg * msg = (struct poa_mgmt_msg *) buf; + + if (!poa_has_name(hash)) + return; + + poa_mgmt_msg_ser(msg, POA_NAME_REPLY, 0, 0, qos_raw, 0, POA_QUERY_HLEN); + + memcpy(buf + sizeof(*msg), hash, POA_QUERY_HLEN); + + if (poa->ops->poa_send_mgmt(poa, src, buf, sizeof(buf)) < 0) + return; /* the requester retransmits */ + + POA_STAT_BUMP(poa, rep_tx); +} + +static bool mac_is_zero(const uint8_t * mac) +{ + static const uint8_t zero[POA_MAC_SIZE] = { 0 }; + + return memcmp(mac, zero, POA_MAC_SIZE) == 0; +} + +static bool eth_query_is_match(const struct eth_query * q, + const struct poa * poa) +{ + if (q->ethertype != 0 && + q->ethertype != poa->local.eth.src.ethertype) + return false; + + if (q->c_dev[0] != '\0' && + strcmp(q->c_dev, poa->local.eth.src.dev) != 0) + return false; + + if (mac_is_zero(q->c_mac)) + return true; + + return memcmp(q->c_mac, poa->local.eth.src.mac, POA_MAC_SIZE) == 0; +} + +static void eth_reply_arr(struct poa * poa, + const struct poa_addr * src, + const uint8_t * hash) +{ + struct list_head * p; + + pthread_once(&queries.once, queries_init); + + pthread_mutex_lock(&queries.mtx); + + list_for_each(p, &queries.pending.list) { + struct eth_query * q; + + q = list_entry(p, struct eth_query, next); + if (q->replied || memcmp(q->hash, hash, POA_QUERY_HLEN) != 0) + continue; + + if (!eth_query_is_match(q, poa)) + continue; + + memcpy(q->mac, src->eth.dst.mac, POA_MAC_SIZE); + memcpy(q->src_mac, src->eth.src.mac, POA_MAC_SIZE); + strcpy(q->dev, src->eth.src.dev); + + q->r_ethertype = poa->local.eth.src.ethertype; + + q->replied = true; + } + + pthread_cond_broadcast(&queries.cond); + + pthread_mutex_unlock(&queries.mtx); +} + +/* Name query frames are handled in the transport. */ +static void eth_rx_mgmt(struct poa * poa, + const struct poa_addr * src, + const uint8_t * buf, + size_t len) +{ + const struct poa_mgmt_msg * msg; + const uint8_t * hash; + + msg = (const struct poa_mgmt_msg *) buf; + if (len < sizeof(*msg)) { + poa_rx_mgmt(poa, src, buf, len); + return; + } + + if (msg->code != POA_NAME_QUERY && msg->code != POA_NAME_REPLY) { + poa_rx_mgmt(poa, src, buf, len); + return; + } + + if (ntoh16(msg->data_len) != POA_QUERY_HLEN) + return; /* malformed */ + + if (len < sizeof(*msg) + POA_QUERY_HLEN) + return; /* malformed */ + + hash = buf + sizeof(*msg); + + if (msg->code == POA_NAME_QUERY) { + POA_STAT_BUMP(poa, qry_rx); + eth_query_arr(poa, src, hash); + } else { + POA_STAT_BUMP(poa, rep_rx); + eth_reply_arr(poa, src, hash); + } +} + +static int eth_query_send(const uint8_t * hash, + const struct poa_addr * addr) +{ + uint8_t buf[sizeof(struct poa_mgmt_msg) + + POA_QUERY_HLEN]; + struct poa_mgmt_msg * msg = (struct poa_mgmt_msg *) buf; + struct poa_addr bcast; + + bcast = *addr; + + memset(bcast.eth.dst.mac, 0xff, POA_MAC_SIZE); + + poa_mgmt_msg_ser(msg, POA_NAME_QUERY, 0, 0, qos_raw, 0, POA_QUERY_HLEN); + + memcpy(buf + sizeof(*msg), hash, POA_QUERY_HLEN); + + return poa_bcast_mgmt(&bcast, buf, sizeof(buf)); +} + +/* Complete addr for dst by broadcast query; the poa_query op. */ +static int eth_query(const char * dst, + const struct timespec * timeo, + struct poa_addr * addr) +{ + struct timespec dflt = TIMESPEC_INIT_MS(ETH_QUERY_TIMEO); + struct timespec rintv = TIMESPEC_INIT_MS(ETH_QUERY_TIMEO + / (ETH_QUERY_RETRIES + 1)); + struct eth_query q; + struct timespec abstime; + struct timespec now; + struct timespec dl; + time_t rintv_ns = (time_t) TS_TO_UINT64(rintv); + time_t diff; + uint8_t hash[POA_QUERY_HLEN]; + int n; + int err = -ETIMEDOUT; + + if (strlen(addr->eth.src.dev) > DEV_NAME_SIZE) + return -EINVAL; + + addr->eth.src.ethertype = addr->eth.dst.ethertype; + + /* The destination is set; a zero ethertype cannot be sent. */ + if (!mac_is_zero(addr->eth.dst.mac)) + return addr->eth.dst.ethertype < 0x0600 ? -EINVAL : 0; + + str_hash(HASH_SHA3_256, hash, dst); + + pthread_once(&queries.once, queries_init); + + memset(&q, 0, sizeof(q)); + memcpy(q.hash, hash, POA_QUERY_HLEN); + + q.ethertype = addr->eth.dst.ethertype; + + memcpy(q.c_mac, addr->eth.src.mac, POA_MAC_SIZE); + strcpy(q.c_dev, addr->eth.src.dev); + + pthread_mutex_lock(&queries.mtx); + llist_add(&q.next, &queries.pending); + pthread_mutex_unlock(&queries.mtx); + + pthread_cleanup_push(eth_query_cleanup, &q); + + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, timeo != NULL ? timeo : &dflt, &abstime); + + while (ts_diff_ns(&now, &abstime) < 0) { + n = eth_query_send(hash, addr); + if (n < 0) { + err = n; + break; + } + + if (n == 0) { + err = -EPERM; + break; + } + + ts_add(&now, &rintv, &dl); + + if (ts_diff_ns(&dl, &abstime) > 0) + dl = abstime; + + pthread_mutex_lock(&queries.mtx); + + pthread_cleanup_push(__cleanup_mutex_unlock, &queries.mtx); + + while (!q.replied) { + if (pthread_cond_timedwait(&queries.cond, &queries.mtx, + &dl) == ETIMEDOUT) + break; + } + + if (q.replied) { + memcpy(addr->eth.dst.mac, q.mac, POA_MAC_SIZE); + memcpy(addr->eth.src.mac, q.src_mac, POA_MAC_SIZE); + strcpy(addr->eth.src.dev, q.dev); + + addr->eth.dst.ethertype = q.r_ethertype; + addr->eth.src.ethertype = q.r_ethertype; + + err = 0; + } + + pthread_cleanup_pop(true); + + if (err == 0) + break; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + diff = ts_diff_ns(&now, &abstime); + if (diff > -rintv_ns) /* skip the runt attempt */ + break; + } + + pthread_cleanup_pop(true); + + return err; +} + +static void eth_spec(const struct poa * poa, + struct poa_spec * spec) +{ + spec->type = poa->type; + spec->eth = poa->local.eth.src; +} + +static bool eth_has_id(const struct poa * poa, + const struct poa_spec * spec) +{ + if (strnlen(spec->eth.dev, sizeof(spec->eth.dev)) > DEV_NAME_SIZE) + return false; + + if (strcmp(poa->local.eth.src.dev, spec->eth.dev) != 0) + return false; + + return poa->local.eth.src.ethertype == spec->eth.ethertype; +} + +/* Our end of the link; an unnamed one matches any. */ +static bool eth_is_src(const struct poa * poa, + const struct eth_poa * src) +{ + if (src->dev[0] != '\0' && + strcmp(poa->local.eth.src.dev, src->dev) != 0) + return false; + + if (mac_is_zero(src->mac)) + return true; + + return memcmp(poa->local.eth.src.mac, src->mac, + POA_MAC_SIZE) == 0; +} + +/* An ethertype of 0 in dst matches any, for broadcast queries. */ +static bool eth_match(const struct poa * poa, + const struct poa_addr * dst) +{ + uint16_t et = dst->eth.dst.ethertype; + + if (et != 0 && et != poa->local.eth.src.ethertype) + return false; + + return eth_is_src(poa, &dst->eth.src); +} + +static bool eth_link_match(const struct poa * poa, + int id) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + return priv->ifindex == id; +} + +#ifdef HAVE_RAW_SOCKETS + +/* MSG_DONTWAIT: the reader blocks on this socket. */ +static int eth_sendv(struct eth_priv * priv, + const uint8_t * dst, + uint32_t eid, + const uint8_t * body, + size_t len, + bool block, + const struct timespec * abstime) +{ + struct eth_hdr hdr; + struct msghdr msg; + struct iovec iov[2]; + int ret; + + if (len > priv->mtu - POA_HDR_SIZE) + return -EMSGSIZE; + + eth_hdr_ser(priv, &hdr, dst, eid, len); + + iov[0].iov_base = &hdr; + iov[0].iov_len = ETH_HDR_TOT_SIZE; + iov[1].iov_base = (void *) body; + iov[1].iov_len = len; + + memset(&msg, 0, sizeof(msg)); + + msg.msg_name = &priv->device; + msg.msg_namelen = sizeof(priv->device); + msg.msg_iov = iov; + msg.msg_iovlen = len > 0 ? 2 : 1; + while (sendmsg(priv->s_fd, &msg, MSG_DONTWAIT) < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) + return -EIO; + + if (!block) + return -EAGAIN; + + ret = poa_wait_out(priv->s_fd, abstime); + if (ret < 0) + return ret; + } + + return 0; +} + +static void * eth_reader(void * o) +{ + struct poa * poa = (struct poa *) o; + struct eth_priv * priv = (struct eth_priv *) poa->priv; + struct timespec down = TIMESPEC_INIT_MS(ETH_DOWN_TIMEO); + uint8_t * buf; + size_t bufsz; + + bufsz = ETH_HDR_SIZE + priv->mtu; + + buf = malloc(bufsz); + if (buf == NULL) + return (void *) -1; + + pthread_cleanup_push(free, buf); + + while (true) { + struct ssm_pk_buff * spb; + struct sockaddr_ll from; + socklen_t flen; + struct poa_addr src; + ssize_t n; + uint32_t eid; + size_t plen; + const uint8_t * body; + + flen = sizeof(from); + + n = recvfrom(priv->s_fd, buf, bufsz, 0, + (struct sockaddr *) &from, &flen); + + if (n < 0) { + if (errno == EINTR) + continue; + + POA_STAT_BUMP(poa, rcv_fail); + + if (errno == ENETDOWN) { + nanosleep(&down, NULL); + continue; + } + + log_err("Reader on %s stopped: %s.", + poa->local.eth.src.dev, + strerror(errno)); + break; + } + + if (from.sll_pkttype == PACKET_OUTGOING) + continue; + + if (frame_parse(priv, buf, (size_t) n, &eid, &plen) < 0) + continue; + + body = buf + ETH_HDR_TOT_SIZE; + + if (eid == POA_MGMT_EID) { + frame_to_addr(priv, (struct eth_hdr *) buf, &src); + eth_rx_mgmt(poa, &src, body, plen); + continue; + } + + if (poa_spb_reserve(&spb, plen) < 0) { + POA_STAT_BUMP(poa, buf_fail); + continue; + } + + memcpy(ssm_pk_buff_head(spb), body, plen); + + poa_rx_pkt(poa, eid, spb); + } + + pthread_cleanup_pop(true); + + return (void *) 0; +} + +/* + * One netlink socket for the whole subsystem: RTMGRP_LINK delivers the + * events of every interface anyway, so a socket per PoA only added + * discards. + */ +int poa_monitor_open(void) +{ + struct sockaddr_nl sa; + int fd; + + memset(&sa, 0, sizeof(sa)); + + sa.nl_family = AF_NETLINK; + sa.nl_groups = RTMGRP_LINK; + + fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE); + if (fd < 0) + return -1; + + if (bind(fd, (struct sockaddr *) &sa, sizeof(sa)) < 0) { + close(fd); + return -1; + } + + return fd; +} + +void poa_monitor_read(int fd) +{ + uint8_t buf[4096]; + struct nlmsghdr * h; + ssize_t n; + + n = recv(fd, buf, sizeof(buf), 0); + if (n < 0) + return; + + for (h = (struct nlmsghdr *) buf; + NLMSG_OK(h, (unsigned int) n); + h = NLMSG_NEXT(h, n)) { + struct ifinfomsg * ifi; + unsigned int usable; + bool up; + size_t cnt; + + if (h->nlmsg_type == NLMSG_DONE) + break; + + if (h->nlmsg_type != RTM_NEWLINK) + continue; + + ifi = NLMSG_DATA(h); + + usable = ifi->ifi_flags & (IFF_UP | IFF_RUNNING); + up = usable == (IFF_UP | IFF_RUNNING); + + cnt = poa_link_updown(ifi->ifi_index, up); + if (cnt > 0) + log_info("Link %d %s, %zu flows.", ifi->ifi_index, + up ? "up" : "down", cnt); + } +} + +static int eth_dev_info(const char * dev, + int * idx, + uint8_t * mac, + uint32_t * mtu) +{ + struct ifreq ifr; + int fd; + + if (strlen(dev) >= IFNAMSIZ) + return -EINVAL; + + *idx = if_nametoindex(dev); + if (*idx == 0) { + log_err("Failed to find device %s.", dev); + return -ENODEV; + } + + fd = socket(AF_UNIX, SOCK_STREAM, 0); + if (fd < 0) + return -EIO; + + memset(&ifr, 0, sizeof(ifr)); + strcpy(ifr.ifr_name, dev); + + if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) { + log_err("Failed to get hardware address of %s.", dev); + goto fail_ioctl; + } + + memcpy(mac, ifr.ifr_hwaddr.sa_data, POA_MAC_SIZE); + + if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) { + log_err("Failed to get MTU of %s.", dev); + goto fail_ioctl; + } + + close(fd); + + *mtu = MIN(MIN(ETH_MAX_MTU, POA_ETH_RD_BUF), (uint32_t) ifr.ifr_mtu); + if (memcmp(dev, "lo", 2) == 0 && *mtu > POA_ETH_LO_MTU) + *mtu = POA_ETH_LO_MTU; + + return 0; + + fail_ioctl: + close(fd); + return -EIO; +} + +/* SO_RCVBUFFORCE bypasses rmem_max; SO_RCVBUF is the fallback. */ +static void eth_set_rcvbuf(int fd, + int rcvbuf) +{ + if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, + &rcvbuf, sizeof(rcvbuf)) == 0) + return; + + if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, sizeof(rcvbuf)) < 0) + log_info("Failed to set SO_RCVBUF to %d.", rcvbuf); +} + +/* The send buffer holds at least one frame, or sendmsg cannot fit a PDU. */ +static int eth_socket(struct eth_priv * priv, + int idx, + const uint8_t * mac) +{ + int sndbuf; + int rcvbuf; +#ifdef POA_ETH_QDISC_BYPASS + int bypass = 1; +#endif + + memset(&priv->device, 0, sizeof(priv->device)); + + priv->device.sll_ifindex = idx; + priv->device.sll_family = AF_PACKET; + priv->device.sll_halen = POA_MAC_SIZE; + priv->device.sll_protocol = htons(ETH_P_ALL); + + memcpy(priv->device.sll_addr, mac, POA_MAC_SIZE); + memcpy(priv->hw_addr, mac, POA_MAC_SIZE); + + priv->s_fd = socket(AF_PACKET, SOCK_RAW, priv->ethertype); + if (priv->s_fd < 0) { + log_err("Failed to create socket: %s.", strerror(errno)); + return -1; + } + +#ifdef POA_ETH_QDISC_BYPASS + if (setsockopt(priv->s_fd, SOL_PACKET, PACKET_QDISC_BYPASS, + &bypass, sizeof(bypass)) < 0) + log_info("Qdisc bypass not supported."); +#endif + + sndbuf = POA_ETH_SNDBUF; + if (sndbuf > 0) { + sndbuf = MAX(sndbuf, (int) (ETH_HDR_SIZE + priv->mtu)); + + if (setsockopt(priv->s_fd, SOL_SOCKET, SO_SNDBUF, + &sndbuf, sizeof(sndbuf)) < 0) + log_info("Failed to set SO_SNDBUF to %d.", sndbuf); + } + + rcvbuf = POA_ETH_RCVBUF; + if (rcvbuf > 0) + eth_set_rcvbuf(priv->s_fd, rcvbuf); + + if (bind(priv->s_fd, (struct sockaddr *) &priv->device, + sizeof(priv->device)) < 0) { + log_err("Failed to bind socket to %d.", idx); + close(priv->s_fd); + return -1; + } + + return 0; +} + +/* + * Pull qlen and backlog from the nested TCA_STATS2. The top-level + * TCA_STATS shares TCA_STATS_QUEUE's id but carries a wholly + * different struct, so descend first; never match on id alone. + */ +static void eth_qdisc_parse(struct rtattr * rta, + size_t rlen, + size_t * byt, + size_t * pkt) +{ + struct gnet_stats_queue q; + struct rtattr * in; + size_t ilen; + + for (; RTA_OK(rta, rlen); rta = RTA_NEXT(rta, rlen)) { + if (rta->rta_type != TCA_STATS2) + continue; + + in = (struct rtattr *) RTA_DATA(rta); + ilen = RTA_PAYLOAD(rta); + for (; RTA_OK(in, ilen); in = RTA_NEXT(in, ilen)) { + if (in->rta_type != TCA_STATS_QUEUE) + continue; + + if (RTA_PAYLOAD(in) < sizeof(q)) + continue; + + memcpy(&q, RTA_DATA(in), sizeof(q)); + + *byt = q.backlog; + *pkt = q.qlen; + } + } +} + +/* + * Egress backlog of the device's root qdisc, in bytes and packets, + * as the kernel queues them. The caller owns the netlink descriptor + * and serialises the query. An early end of dump reports failure. + */ +static int eth_qdisc_backlog(int fd, + int ifindex, + size_t * byt, + size_t * pkt) +{ + struct { + struct nlmsghdr nh; + struct tcmsg tc; + } req; + struct nlmsghdr * nh; + struct rtattr * rta; + struct tcmsg * tc; + char buf[16384]; + ssize_t len; + int ret = -1; + + if (fd < 0) + goto fail; + + memset(&req, 0, sizeof(req)); + + req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(req.tc)); + req.nh.nlmsg_type = RTM_GETQDISC; + req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.nh.nlmsg_seq = 1; + req.tc.tcm_family = AF_UNSPEC; + req.tc.tcm_ifindex = ifindex; + if (send(fd, &req, req.nh.nlmsg_len, 0) < 0) + goto fail; + + *byt = 0; + *pkt = 0; + + while ((len = recv(fd, buf, sizeof(buf), 0)) > 0) { + nh = (struct nlmsghdr *) buf; + for (; NLMSG_OK(nh, len); nh = NLMSG_NEXT(nh, len)) { + if (nh->nlmsg_type == NLMSG_DONE) + goto done; + + if (nh->nlmsg_type == NLMSG_ERROR) + goto fail; + + if (nh->nlmsg_type != RTM_NEWQDISC) + continue; + + tc = (struct tcmsg *) NLMSG_DATA(nh); + if (tc->tcm_ifindex != ifindex) + continue; + + if (tc->tcm_parent != TC_H_ROOT) + continue; + + rta = (struct rtattr *) + ((char *) tc + NLMSG_ALIGN(sizeof(*tc))); + + eth_qdisc_parse(rta, nh->nlmsg_len + - NLMSG_LENGTH(sizeof(*tc)), + byt, pkt); + } + } + + goto fail; /* early end: a zero would read as empty */ + done: + ret = 0; + fail: + return ret; +} + +/* + * A netlink descriptor for the qdisc query. SO_RCVTIMEO bounds the + * dump: this is read from the send path, and a reply that never + * arrives must not park a sender thread. + */ +static int eth_qdisc_open(void) +{ + struct sockaddr_nl sa; + struct timeval tv = TIMEVAL_INIT_MS(POA_QDISC_TIMEO); + int fd; + + fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE); + if (fd < 0) + return -1; + + memset(&sa, 0, sizeof(sa)); + + sa.nl_family = AF_NETLINK; + if (bind(fd, (struct sockaddr *) &sa, sizeof(sa)) < 0) + goto fail; + + if (setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)) < 0) + goto fail; + + return fd; + fail: + close(fd); + + return -1; +} + +/* + * Queue depth in packets, straight from the qdisc. Gated like the + * transport's own depth and skipped when another sender is already + * asking, so the send path never waits on netlink. + */ +static int eth_qpkts(struct poa * poa, + size_t * pkts, + size_t * byts) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + struct timespec now; + uint64_t ns; + size_t byt; + size_t pkt; + + if (priv->nl_fd < 0) + return -1; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + ns = TS_TO_UINT64(now); + if (ns - LOAD_RELAXED(&priv->nl_time) < POA_QLEN_GATE) + goto cached; + + if (__atomic_exchange_n(&priv->nl_busy, 1, __ATOMIC_ACQUIRE) != 0) + goto cached; + + if (eth_qdisc_backlog(priv->nl_fd, priv->ifindex, &byt, &pkt) == 0) { + STORE_RELAXED(&priv->nl_pkt, pkt); + STORE_RELAXED(&priv->nl_byt, byt); + STORE_RELAXED(&priv->nl_time, ns); + } + + __atomic_store_n(&priv->nl_busy, 0, __ATOMIC_RELEASE); + cached: + if (LOAD_RELAXED(&priv->nl_time) == 0) /* nothing measured yet */ + return -1; + + *pkts = LOAD_RELAXED(&priv->nl_pkt); + *byts = LOAD_RELAXED(&priv->nl_byt); + + return 0; +} + +static int eth_attach(struct poa * poa, + const struct poa_spec * spec) +{ + const struct eth_poa * c = &spec->eth; + struct eth_priv * priv; + uint8_t mac[POA_MAC_SIZE]; + uint32_t mtu; + int idx; + int err; + + if (strnlen(c->dev, sizeof(c->dev)) > DEV_NAME_SIZE) + return -EINVAL; + + if (c->ethertype < 0x0600 || c->ethertype == 0xFFFF) { + log_err("Invalid Ethertype 0x%x.", c->ethertype); + return -EINVAL; + } + + priv = malloc(sizeof(*priv)); + if (priv == NULL) + return -ENOMEM; + + memset(priv, 0, sizeof(*priv)); + + priv->poa = poa; + priv->nl_fd = eth_qdisc_open(); /* optional; socket fallback */ + + priv->ethertype = htons(c->ethertype); + + err = eth_dev_info(c->dev, &idx, mac, &mtu); + if (err < 0) + goto fail_conf; + + priv->mtu = mtu; + if (eth_socket(priv, idx, mac) < 0) + goto fail_conf; + + poa->priv = priv; + poa->local.type = poa->type; + poa->local.eth.src.ethertype = c->ethertype; + priv->ifindex = idx; + + memcpy(poa->local.eth.src.mac, mac, POA_MAC_SIZE); + strcpy(poa->local.eth.src.dev, c->dev); + + return 0; + + fail_conf: + if (priv->nl_fd >= 0) + close(priv->nl_fd); + + free(priv); + + return -EIO; +} + +static void eth_detach(struct poa * poa) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + if (priv == NULL) + return; + + close(priv->s_fd); + + if (priv->nl_fd >= 0) + close(priv->nl_fd); + + free(priv); + + poa->priv = NULL; +} + +static uint32_t eth_mtu(struct poa * poa, + const struct poa_addr * dst) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + (void) dst; + + return priv->mtu - POA_HDR_SIZE; +} + +/* All flows on the PoA share the socket, so this is aggregate. */ +static size_t eth_qlen(struct poa * poa) +{ +#ifdef SIOCOUTQ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + int qlen; + + qlen = 0; + if (ioctl(priv->s_fd, SIOCOUTQ, &qlen) < 0) + return 0; + + return (size_t) qlen; +#else + (void) poa; + + return 0; +#endif +} + +/* + * PACKET_STATISTICS zeroes the kernel counters on read, so totals + * accumulate here; relaxed atomics allow concurrent RIB reads. + * The qdisc depth rides the sender's guarded path: one dump at a time. + */ +static int eth_rib(struct poa * poa, + char * buf, + size_t len) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + struct tpacket_stats ts; + socklen_t optlen; + size_t sndbuf = 0; + size_t rcvbuf = 0; + size_t qd_byt; + size_t qd_pkt; + int val; + int size; + + optlen = sizeof(val); + if (getsockopt(priv->s_fd, SOL_SOCKET, SO_SNDBUF, &val, &optlen) == 0) + sndbuf = (size_t) val; + + optlen = sizeof(val); + if (getsockopt(priv->s_fd, SOL_SOCKET, SO_RCVBUF, &val, &optlen) == 0) + rcvbuf = (size_t) val; + + optlen = sizeof(ts); + + if (getsockopt(priv->s_fd, SOL_PACKET, PACKET_STATISTICS, + &ts, &optlen) == 0) { + FETCH_ADD_RELAXED(&priv->kern_rcv, ts.tp_packets); + FETCH_ADD_RELAXED(&priv->kern_drp, ts.tp_drops); + } + + if (eth_qpkts(poa, &qd_pkt, &qd_byt) < 0) { + qd_byt = 0; + qd_pkt = 0; + } + + size = snprintf(buf, len, + "Socket sndbuf (bytes): %zu\n" + "Socket rcvbuf (bytes): %zu\n" + "Kernel packets received: %zu\n" + "Kernel packets dropped: %zu\n" + "Qdisc backlog (bytes): %zu\n" + "Qdisc backlog (packets): %zu\n", + sndbuf, rcvbuf, + LOAD_RELAXED(&priv->kern_rcv), + LOAD_RELAXED(&priv->kern_drp), + qd_byt, qd_pkt); + if (size < 0 || (size_t) size >= len) + return -1; + + return size; +} + +#elif defined(HAVE_BPF) + +/* + * BSD and macOS reach the link layer through a cloned /dev/bpf + * device, bound to an interface with BIOCSETIF. One device per PoA. + */ + +static int eth_sendv(struct eth_priv * priv, + const uint8_t * dst, + uint32_t eid, + const uint8_t * body, + size_t len, + bool block, + const struct timespec * abstime) +{ + struct eth_hdr hdr; + struct iovec iov[2]; + int n; + int ret; + + if (len > priv->mtu - POA_HDR_SIZE) + return -EMSGSIZE; + + eth_hdr_ser(priv, &hdr, dst, eid, len); + + iov[0].iov_base = &hdr; + iov[0].iov_len = ETH_HDR_TOT_SIZE; + iov[1].iov_base = (void *) body; + iov[1].iov_len = len; + n = len > 0 ? 2 : 1; + while (writev(priv->s_fd, iov, n) < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) + return -EIO; + + if (!block) + return -EAGAIN; + + ret = poa_wait_out(priv->s_fd, abstime); + if (ret < 0) + return ret; + } + + return 0; +} + +/* One read yields a batch of BPF_WORDALIGN'ed frames; walk all of them. */ +static void * eth_reader(void * o) +{ + struct poa * poa = (struct poa *) o; + struct eth_priv * priv = (struct eth_priv *) poa->priv; + uint8_t * buf; + + buf = malloc(priv->blen); + if (buf == NULL) + return (void *) -1; + + pthread_cleanup_push(free, buf); + + while (true) { + uint8_t * p; + uint8_t * end; + ssize_t n; + + n = read(priv->s_fd, buf, priv->blen); + if (n < 0) { + if (errno == EINTR) + continue; + + POA_STAT_BUMP(poa, rcv_fail); + break; + } + + p = buf; + end = buf + n; + while (p + sizeof(struct bpf_hdr) <= end) { + struct bpf_hdr * bh = (struct bpf_hdr *) p; + struct ssm_pk_buff * spb; + struct poa_addr src; + const uint8_t * frame; + const uint8_t * body; + uint32_t eid; + size_t plen; + + frame = p + bh->bh_hdrlen; + if (frame + bh->bh_caplen > end) + break; + + if (frame_parse(priv, frame, bh->bh_caplen, + &eid, &plen) < 0) + goto next; + + body = frame + ETH_HDR_TOT_SIZE; + + if (eid == POA_MGMT_EID) { + frame_to_addr(priv, + (const struct eth_hdr *) frame, + &src); + eth_rx_mgmt(poa, &src, body, plen); + goto next; + } + + if (poa_spb_reserve(&spb, plen) < 0) { + POA_STAT_BUMP(poa, buf_fail); + goto next; + } + + memcpy(ssm_pk_buff_head(spb), body, plen); + + poa_rx_pkt(poa, eid, spb); + next: + p += BPF_WORDALIGN(bh->bh_hdrlen + bh->bh_caplen); + } + } + + pthread_cleanup_pop(true); + + return (void *) 0; +} + +static int eth_dev_info(const char * dev, + uint8_t * mac, + uint32_t * mtu) +{ + struct ifaddrs * ifas; + struct ifaddrs * ifa; + struct ifreq ifr; + int fd; + int found = 0; + + if (getifaddrs(&ifas) < 0) { + log_err("Failed to list interfaces."); + return -1; + } + + for (ifa = ifas; ifa != NULL; ifa = ifa->ifa_next) { + struct sockaddr_dl * dl; + + if (ifa->ifa_addr == NULL) + continue; + + if (ifa->ifa_addr->sa_family != AF_LINK) + continue; + + if (strcmp(ifa->ifa_name, dev) != 0) + continue; + + dl = (struct sockaddr_dl *) ifa->ifa_addr; + if (dl->sdl_alen != POA_MAC_SIZE) + continue; + + memcpy(mac, LLADDR(dl), POA_MAC_SIZE); + + found = 1; + break; + } + + freeifaddrs(ifas); + + if (!found) { + log_err("No such device: %s.", dev); + return -1; + } + + fd = socket(AF_INET, SOCK_DGRAM, 0); + if (fd < 0) + return -1; + + memset(&ifr, 0, sizeof(ifr)); + + strcpy(ifr.ifr_name, dev); + + if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) { + log_err("Failed to get MTU of %s.", dev); + close(fd); + return -1; + } + + close(fd); + + *mtu = MIN(MIN(ETH_MAX_MTU, POA_ETH_RD_BUF), (uint32_t) ifr.ifr_mtu); + if (memcmp(dev, "lo", 2) == 0 && *mtu > POA_ETH_LO_MTU) + *mtu = POA_ETH_LO_MTU; + + return 0; +} + +static int eth_bpf_open(void) +{ + char dev[32]; + size_t i; + + for (i = 0; i < BPF_DEV_MAX; ++i) { + int fd; + + sprintf(dev, "/dev/bpf%zu", i); + + fd = open(dev, O_RDWR); + if (fd >= 0) + return fd; + } + + return -1; +} + +/* + * BIOCSHDRCMPLT: we fill in the source address ourselves. + * BIOCSSEESENT: our own egress must not come back at us. + * BIOCIMMEDIATE: deliver on arrival, do not wait for a full buffer. + */ +static int eth_bpf_setup(struct eth_priv * priv, + const char * dev) +{ + struct ifreq ifr; + int enable = 1; + int disable = 0; + int blen = 0; + + memset(&ifr, 0, sizeof(ifr)); + + strcpy(ifr.ifr_name, dev); + + if (ioctl(priv->s_fd, BIOCSETIF, &ifr) < 0) { + log_err("Failed to bind bpf device to %s.", dev); + return -1; + } + + if (ioctl(priv->s_fd, BIOCGBLEN, &blen) < 0 || blen <= 0) { + log_err("Failed to get the bpf buffer length."); + return -1; + } + + priv->blen = (size_t) blen; + if (ioctl(priv->s_fd, BIOCSHDRCMPLT, &enable) < 0) { + log_err("Failed to set BIOCSHDRCMPLT."); + return -1; + } + + if (ioctl(priv->s_fd, BIOCSSEESENT, &disable) < 0) { + log_err("Failed to set BIOCSSEESENT."); + return -1; + } + + if (ioctl(priv->s_fd, BIOCIMMEDIATE, &enable) < 0) { + log_err("Failed to set BIOCIMMEDIATE."); + return -1; + } + + return 0; +} + +static int eth_attach(struct poa * poa, + const struct poa_spec * spec) +{ + const struct eth_poa * c = &spec->eth; + struct eth_priv * priv; + uint8_t mac[POA_MAC_SIZE]; + uint32_t mtu; + + if (strnlen(c->dev, sizeof(c->dev)) > DEV_NAME_SIZE) + return -EINVAL; + + if (c->ethertype < 0x0600 || c->ethertype == 0xFFFF) { + log_err("Invalid Ethertype 0x%x.", c->ethertype); + return -EINVAL; + } + + priv = malloc(sizeof(*priv)); + if (priv == NULL) + return -ENOMEM; + + memset(priv, 0, sizeof(*priv)); + + priv->poa = poa; + priv->s_fd = -1; + + priv->ethertype = htons(c->ethertype); + + if (eth_dev_info(c->dev, mac, &mtu) < 0) + goto fail_conf; + + priv->mtu = mtu; + + memcpy(priv->hw_addr, mac, POA_MAC_SIZE); + + priv->s_fd = eth_bpf_open(); + if (priv->s_fd < 0) { + log_err("Failed to open a bpf device."); + goto fail_conf; + } + + if (eth_bpf_setup(priv, c->dev) < 0) + goto fail_device; + + poa->priv = priv; + poa->local.type = poa->type; + poa->local.eth.src.ethertype = c->ethertype; + + memcpy(poa->local.eth.src.mac, mac, POA_MAC_SIZE); + strcpy(poa->local.eth.src.dev, c->dev); + + log_info("Using Berkeley Packet Filter on %s.", c->dev); + + return 0; + + fail_device: + close(priv->s_fd); + fail_conf: + free(priv); + + return -EIO; +} + +static void eth_detach(struct poa * poa) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + if (priv == NULL) + return; + + close(priv->s_fd); + + free(priv); + + poa->priv = NULL; +} + +static uint32_t eth_mtu(struct poa * poa, + const struct poa_addr * dst) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + (void) dst; + + return priv->mtu - POA_HDR_SIZE; +} + +/* A bpf device has no send queue to report; mb-ECN cannot mark here. */ +static size_t eth_qlen(struct poa * poa) +{ + (void) poa; + + return 0; +} + +/* The bpf device buffer is all the receive queue there is. */ +static int eth_rib(struct poa * poa, + char * buf, + size_t len) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + struct bpf_stat bs; + int size; + + if (ioctl(priv->s_fd, BIOCGSTATS, &bs) == 0) { + FETCH_ADD_RELAXED(&priv->kern_rcv, bs.bs_recv); + FETCH_ADD_RELAXED(&priv->kern_drp, bs.bs_drop); + } + + size = snprintf(buf, len, + "Socket rcvbuf (bytes): %zu\n" + "Kernel packets received: %zu\n" + "Kernel packets dropped: %zu\n", + priv->blen, + LOAD_RELAXED(&priv->kern_rcv), + LOAD_RELAXED(&priv->kern_drp)); + if (size < 0 || (size_t) size >= len) + return -1; + + return size; +} + +#elif defined(HAVE_NETMAP) + +/* + * netmap gives one PoA the whole port: reads copy out of the NIC + * ring, writes inject into it. + */ + +/* nm_inject takes one contiguous frame, so the header is copied in. */ +static int eth_sendv(struct eth_priv * priv, + const uint8_t * dst, + uint32_t eid, + const uint8_t * body, + size_t len, + bool block, + const struct timespec * abstime) +{ + uint8_t * frame; + size_t flen; + int ret; + + if (len > priv->mtu - POA_HDR_SIZE) + return -EMSGSIZE; + + flen = ETH_HDR_TOT_SIZE + len; + + frame = malloc(flen); + if (frame == NULL) + return -ENOMEM; + + eth_hdr_ser(priv, (struct eth_hdr *) frame, dst, eid, len); + + if (len > 0) + memcpy(frame + ETH_HDR_TOT_SIZE, body, len); + + if (block) + ret = poa_wait_out(priv->poll_out.fd, abstime); + else + ret = poll(&priv->poll_out, 1, 0) > 0 ? 0 : -EAGAIN; + + if (ret < 0) + goto fail; + + ret = nm_inject(priv->nmd, frame, flen) == (int) flen ? 0 : -EIO; + fail: + free(frame); + + return ret; +} + +/* A slot stays owned by the ring, so each frame is copied out. */ +static void * eth_reader(void * o) +{ + struct poa * poa = (struct poa *) o; + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + while (true) { + struct ssm_pk_buff * spb; + struct poa_addr src; + struct nm_pkthdr hdr; + const uint8_t * frame; + const uint8_t * body; + uint32_t eid; + size_t plen; + + if (poll(&priv->poll_in, 1, -1) < 0) { + if (errno == EINTR) + continue; + + POA_STAT_BUMP(poa, rcv_fail); + break; + } + + if (priv->poll_in.revents == 0) + continue; + + frame = nm_nextpkt(priv->nmd, &hdr); + if (frame == NULL) + continue; + + if (frame_parse(priv, frame, hdr.len, &eid, &plen) < 0) + continue; + + body = frame + ETH_HDR_TOT_SIZE; + + if (eid == POA_MGMT_EID) { + frame_to_addr(priv, (const struct eth_hdr *) frame, + &src); + eth_rx_mgmt(poa, &src, body, plen); + continue; + } + + if (poa_spb_reserve(&spb, plen) < 0) { + POA_STAT_BUMP(poa, buf_fail); + continue; + } + + memcpy(ssm_pk_buff_head(spb), body, plen); + + poa_rx_pkt(poa, eid, spb); + } + + return (void *) 0; +} + +static int eth_dev_info(const char * dev, + uint8_t * mac, + uint32_t * mtu) +{ + struct ifreq ifr; +#ifndef __linux__ + struct ifaddrs * ifas; + struct ifaddrs * ifa; + int found = 0; +#endif + int fd; + + if (strlen(dev) >= IFNAMSIZ) + return -EINVAL; + + fd = socket(AF_INET, SOCK_DGRAM, 0); + if (fd < 0) + return -EIO; + + memset(&ifr, 0, sizeof(ifr)); + + strcpy(ifr.ifr_name, dev); + +#ifdef __linux__ + if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) { + log_err("Failed to get hardware address of %s.", dev); + goto fail_ioctl; + } + + memcpy(mac, ifr.ifr_hwaddr.sa_data, POA_MAC_SIZE); +#else + if (getifaddrs(&ifas) < 0) + goto fail_ioctl; + + for (ifa = ifas; ifa != NULL; ifa = ifa->ifa_next) { + struct sockaddr_dl * dl; + + if (ifa->ifa_addr == NULL) + continue; + + if (ifa->ifa_addr->sa_family != AF_LINK) + continue; + + if (strcmp(ifa->ifa_name, dev) != 0) + continue; + + dl = (struct sockaddr_dl *) ifa->ifa_addr; + if (dl->sdl_alen != POA_MAC_SIZE) + continue; + + memcpy(mac, LLADDR(dl), POA_MAC_SIZE); + + found = 1; + break; + } + + freeifaddrs(ifas); + + if (!found) { + log_err("No hardware address for %s.", dev); + goto fail_ioctl; + } +#endif + if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) { + log_err("Failed to get MTU of %s.", dev); + goto fail_ioctl; + } + + close(fd); + + *mtu = MIN(MIN(ETH_MAX_MTU, POA_ETH_RD_BUF), (uint32_t) ifr.ifr_mtu); + if (memcmp(dev, "lo", 2) == 0 && *mtu > POA_ETH_LO_MTU) + *mtu = POA_ETH_LO_MTU; + + return 0; + + fail_ioctl: + close(fd); + + return -EIO; +} + +static int eth_nm_open(struct eth_priv * priv, + const char * dev) +{ + char ifn[IFNAMSIZ + sizeof("netmap:")]; + + strcpy(ifn, "netmap:"); + strcat(ifn, dev); + + priv->nmd = nm_open(ifn, NULL, 0, NULL); + if (priv->nmd == NULL) { + log_err("Failed to open netmap device for %s.", dev); + return -1; + } + + memset(&priv->poll_in, 0, sizeof(priv->poll_in)); + memset(&priv->poll_out, 0, sizeof(priv->poll_out)); + + priv->poll_in.fd = NETMAP_FD(priv->nmd); + priv->poll_in.events = POLLIN; + priv->poll_out.fd = NETMAP_FD(priv->nmd); + priv->poll_out.events = POLLOUT; + + return 0; +} + +static int eth_attach(struct poa * poa, + const struct poa_spec * spec) +{ + const struct eth_poa * c = &spec->eth; + struct eth_priv * priv; + uint8_t mac[POA_MAC_SIZE]; + uint32_t mtu; + + if (strnlen(c->dev, sizeof(c->dev)) > DEV_NAME_SIZE) + return -EINVAL; + + if (c->ethertype < 0x0600 || c->ethertype == 0xFFFF) { + log_err("Invalid Ethertype 0x%x.", c->ethertype); + return -EINVAL; + } + + priv = malloc(sizeof(*priv)); + if (priv == NULL) + return -ENOMEM; + + memset(priv, 0, sizeof(*priv)); + + priv->poa = poa; + + priv->ethertype = htons(c->ethertype); + + if (eth_dev_info(c->dev, mac, &mtu) < 0) + goto fail_conf; + + priv->mtu = mtu; + + memcpy(priv->hw_addr, mac, POA_MAC_SIZE); + + if (eth_nm_open(priv, c->dev) < 0) + goto fail_conf; + + poa->priv = priv; + poa->local.type = poa->type; + poa->local.eth.src.ethertype = c->ethertype; + + memcpy(poa->local.eth.src.mac, mac, POA_MAC_SIZE); + strcpy(poa->local.eth.src.dev, c->dev); + + log_info("Using netmap on %s.", c->dev); + + return 0; + + fail_conf: + free(priv); + + return -EIO; +} + +static void eth_detach(struct poa * poa) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + if (priv == NULL) + return; + + nm_close(priv->nmd); + + free(priv); + + poa->priv = NULL; +} + +static uint32_t eth_mtu(struct poa * poa, + const struct poa_addr * dst) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + (void) dst; + + return priv->mtu - POA_HDR_SIZE; +} + +/* The ring is drained by the NIC; there is no queue to report. */ +static size_t eth_qlen(struct poa * poa) +{ + (void) poa; + + return 0; +} + +#endif /* HAVE_RAW_SOCKETS */ + +#ifndef HAVE_RAW_SOCKETS + +/* Only netlink reports link events; no other backend has a monitor. */ +int poa_monitor_open(void) +{ + return -1; +} + +void poa_monitor_read(int fd) +{ + (void) fd; +} + +#endif + +/* One reader per socket, so a flow cannot be reordered on receive. */ +static int eth_start(struct poa * poa) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + if (pthread_create(&priv->reader, NULL, eth_reader, poa) != 0) + return -1; + + priv->running = true; + + return 0; +} + +static void eth_stop(struct poa * poa) +{ + struct eth_priv * priv = (struct eth_priv *) poa->priv; + + if (!priv->running) + return; + + pthread_cancel(priv->reader); + pthread_join(priv->reader, NULL); + + priv->running = false; +} + +static int eth_send(struct poa * poa, + const struct poa_addr * dst, + uint32_t eid, + struct ssm_pk_buff * spb, + bool block, + const struct timespec * abstime) +{ + return eth_sendv((struct eth_priv *) poa->priv, dst->eth.dst.mac, eid, + ssm_pk_buff_head(spb), ssm_pk_buff_len(spb), + block, abstime); +} + +static int eth_send_mgmt(struct poa * poa, + const struct poa_addr * dst, + const uint8_t * buf, + size_t len) +{ + struct timespec timeo = TIMESPEC_INIT_MS(POA_MGMT_SND_TIMEO); + struct timespec abstime; + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + ts_add(&abstime, &timeo, &abstime); + + return eth_sendv((struct eth_priv *) poa->priv, dst->eth.dst.mac, + POA_MGMT_EID, buf, len, true, &abstime); +} + +const struct poa_ops eth_poa_ops = { + .poa_attach = eth_attach, + .poa_detach = eth_detach, + .poa_start = eth_start, + .poa_stop = eth_stop, + .poa_send = eth_send, + .poa_send_mgmt = eth_send_mgmt, + .poa_query = eth_query, + .poa_mtu = eth_mtu, + .poa_qlen = eth_qlen, +#ifdef HAVE_RAW_SOCKETS + .poa_qpkts = eth_qpkts, +#endif +#ifndef HAVE_NETMAP + .poa_rib = eth_rib, +#endif + .poa_spec = eth_spec, + .poa_has_id = eth_has_id, + .poa_match = eth_match, + .poa_link_match = eth_link_match, + .mpl = POA_ETH_MPL +}; diff --git a/src/lib/poa/poa.c b/src/lib/poa/poa.c new file mode 100644 index 00000000..3ad17c4f --- /dev/null +++ b/src/lib/poa/poa.c @@ -0,0 +1,2515 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Points of attachment (PoA) - transport independent core + * + * Included by dev.c; uses dev.c statics (proc, flow_init, ...). + * + * 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/. + */ + +#define POA_MAX_EIDS PROC_MAX_FLOWS +#define POA_ALLOC_TIMEO 10000 /* ms, overall FLOW_REQ deadline */ +#define POA_RETRY_TIMEO 300 /* ms, FLOW_REQ retransmit period */ +/* Must fit a certificate chain: post-quantum ones are large. */ +#define POA_MGMT_BUFSZ POA_MGMT_FRAME_SIZE +#define POA_MGMT_QMAX 64 /* queued management frames per PoA */ +#define POA_PEND_TIMEO 10 /* s, reap a request that never completes */ +#define POA_SWEEP_TIMEO 1000 /* ms, sweep interval */ +#define POA_SWEEP_MAX 16 /* requests reaped per sweep */ +#define POA_DEFER_MAX 64 /* replies waiting for their flow id */ + +/* EWMA over 8 samples. */ +#define POA_AVG_SHIFT 3 +/* Queue cost is sampled every 64th packet: qlen is a syscall. */ +#define POA_COST_MASK 63 +/* Reuse a qlen read for this long; the mark moves on doublings. */ + +#define POA_RIB "poa" +/* Fits the RIB labels below with 20-digit counters. */ +#define POA_RIB_STRLEN 2048 + +enum poa_flow_state { + POA_FLOW_NULL = 0, + POA_FLOW_PENDING, + POA_FLOW_ALLOCATED, + POA_FLOW_DEAD +}; + +enum poa_state { + POA_NULL = 0, + POA_INIT, /* poa_init(); this process may attach */ + POA_RUNNING, /* the threads are up */ + POA_OPERATIONAL /* bootstrapped or enrolled in a layer */ +}; + +struct poa_flow { + struct poa * poa; + + int flow_id; + uint32_t eid; + uint32_t r_eid; + struct poa_addr r_addr; + + enum poa_flow_state state; + + struct ssm_rbuff * rx_rb; + + struct list_head pend; /* on poa->pending while unanswered */ + struct timespec t0; + + /* Our answer, kept to re-send when a request is repeated. */ + bool answered; + bool answer_sent; + + /* Handed to a caller that will attach it; not the sweeper's. */ + bool claimed; + int answer; + buffer_t answer_data; + + /* Handshake rendezvous with the reader thread. */ + pthread_mutex_t mtx; + pthread_cond_t cond; + bool replied; + bool pending; + int response; + buffer_t resp_data; +}; + +/* + * An answer can be ready before the flow it answers has an id, and + * the caller must not be kept waiting for one: it answers to the IRMd, + * which gives up long before we would. + */ +struct poa_deferred { + struct list_head next; + struct timespec t0; + int flow_id; + int response; + buffer_t data; +}; + +struct poa_mgmt_frame { + struct list_head next; + struct poa * poa; + struct poa_addr src; + size_t len; + uint8_t buf[POA_MGMT_BUFSZ]; +}; + +/* + * A detach is performed by the management thread, so that it cannot + * run alongside the accept side. The caller waits for the result. + */ +struct poa_detach_req { + struct list_head next; + struct poa_spec spec; + int result; + bool done; +}; + +static struct { + struct list_head list; + + struct poa_flow * id_to_pf[SYS_MAX_FLOWS]; + + struct llist deferred; + + pthread_mutex_t mtx; /* guards id_to_pf */ + pthread_cond_t cond; + + /* One management thread and one link monitor for all PoAs. */ + struct llist mgmt_frames; + struct list_head detach; + pthread_mutex_t mgmt_mtx; + pthread_cond_t mgmt_cond; /* work for the handler */ + pthread_cond_t done_cond; /* a detach has completed */ + pthread_t mgmt_handler; + bool mgmt_stop; + + pthread_t monitor; + int mon_fd; + + enum poa_state state; + + /* Hashes of the names this process answers queries for. */ + uint8_t name_hash[POA_QUERY_HLEN]; + uint8_t layer_hash[POA_QUERY_HLEN]; + + pthread_rwlock_t lock; /* the PoAs and their flows */ +} poas; + +static int mgmt_send(struct poa * poa, + const struct poa_addr * dst, + uint8_t code, + uint32_t s_eid, + uint32_t d_eid, + qosspec_t qs, + int response, + const buffer_t * data) +{ + uint8_t buf[POA_MGMT_BUFSZ]; + struct poa_mgmt_msg * msg = (struct poa_mgmt_msg *) buf; + size_t len; + + len = sizeof(*msg); + if (data != NULL && data->len > 0) { + if (len + data->len > sizeof(buf)) + return -EMSGSIZE; + memcpy(buf + len, data->data, data->len); + + len += data->len; + } + + poa_mgmt_msg_ser(msg, code, s_eid, d_eid, qs, response, + data != NULL ? data->len : 0); + + POA_STAT_BUMP(poa, mgmt_tx); + + return poa->ops->poa_send_mgmt(poa, dst, buf, len); +} + +/* Caller holds poas.lock for writing. */ +static struct poa_flow * pf_create(struct poa * poa, + const struct poa_addr * r_addr) +{ + struct poa_flow * pf; + pthread_condattr_t cattr; + int eid; + + pf = malloc(sizeof(*pf)); + if (pf == NULL) + goto fail_malloc; + + memset(pf, 0, sizeof(*pf)); + + if (pthread_mutex_init(&pf->mtx, NULL) != 0) + goto fail_mtx; + + if (pthread_condattr_init(&cattr) != 0) + goto fail_cond; +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + if (pthread_cond_init(&pf->cond, &cattr) != 0) { + pthread_condattr_destroy(&cattr); + goto fail_cond; + } + + pthread_condattr_destroy(&cattr); + + eid = bmp_allocate(poa->eids); + if (!bmp_is_id_valid(poa->eids, eid)) + goto fail_eid; + + pf->poa = poa; + pf->eid = (uint32_t) eid; + pf->flow_id = -1; + pf->state = POA_FLOW_PENDING; + pf->r_addr = *r_addr; + + clock_gettime(PTHREAD_COND_CLOCK, &pf->t0); + + rcu_wrlock(&poa->guard); + + rcu_publish(pf); + rcu_assign(poa->eid_to_pf[eid], pf); + + rcu_wrunlock(&poa->guard); + + list_add_tail(&pf->pend, &poa->flows); + + POA_STAT_BUMP(poa, n_flows); + + return pf; + + fail_eid: + pthread_cond_destroy(&pf->cond); + fail_cond: + pthread_mutex_destroy(&pf->mtx); + fail_mtx: + free(pf); + fail_malloc: + return NULL; +} + +/* Caller holds poas.mtx. */ +static void deferred_purge(int flow_id) +{ + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, &poas.deferred.list) { + struct poa_deferred * d; + + d = list_entry(p, struct poa_deferred, next); + if (d->flow_id != flow_id) + continue; + + llist_del(&d->next, &poas.deferred); + freebuf(d->data); + free(d); + } +} + +/* The id may have moved to a newer flow; only its owner clears it. */ +static void pf_destroy(struct poa_flow * pf) +{ + struct poa * poa = pf->poa; + + pthread_rwlock_wrlock(&poas.lock); + + rcu_wrlock(&poa->guard); + + rcu_assign(poa->eid_to_pf[pf->eid], NULL); + rcu_assign(pf->rx_rb, NULL); + + rcu_wrunlock(&poa->guard); + + bmp_release(poa->eids, pf->eid); + + if (!list_is_empty(&pf->pend)) { + list_del(&pf->pend); + POA_STAT_SUB(poa, n_flows, 1); + } + + pthread_rwlock_unlock(&poas.lock); + + pthread_mutex_lock(&poas.mtx); + + if (pf->flow_id >= 0 && poas.id_to_pf[pf->flow_id] == pf) { + poas.id_to_pf[pf->flow_id] = NULL; + + deferred_purge(pf->flow_id); + } + + pthread_mutex_unlock(&poas.mtx); + + rcu_reclaim(&poa->guard); /* a receive may still hold pf */ + + freebuf(pf->resp_data); + freebuf(pf->answer_data); + + pthread_cond_destroy(&pf->cond); + pthread_mutex_destroy(&pf->mtx); + + free(pf); +} + +/* + * Keeps the answer for a repeat, then sends it. An accept must wait + * until the flow can receive; a refusal needs no receiver. + */ +static int pf_answer(struct poa_flow * pf, + int response, + const buffer_t * data) +{ + int err; + + pthread_rwlock_wrlock(&poas.lock); + + freebuf(pf->answer_data); + + if (data != NULL && data->len > 0) { + pf->answer_data.data = malloc(data->len); + if (pf->answer_data.data != NULL) { + memcpy(pf->answer_data.data, data->data, data->len); + + pf->answer_data.len = data->len; + } + } + + pf->answer = response; + pf->answered = true; + if (response == 0 && pf->state != POA_FLOW_ALLOCATED) { + pthread_rwlock_unlock(&poas.lock); + return 0; + } + + pf->answer_sent = true; + + pthread_rwlock_unlock(&poas.lock); + + err = mgmt_send(pf->poa, &pf->r_addr, POA_FLOW_REPLY, pf->eid, + pf->r_eid, qos_raw, response, data); + + if (err == -ETIMEDOUT || err == -EAGAIN) + err = 0; /* stored; a repeat request resends it */ + + return err; +} + +/* Takes an answer left for a flow that had no id yet. */ +static struct poa_deferred * deferred_take(int flow_id) +{ + struct list_head * p; + struct list_head * h; + + list_for_each_safe(p, h, &poas.deferred.list) { + struct poa_deferred * d; + + d = list_entry(p, struct poa_deferred, next); + if (d->flow_id != flow_id) + continue; + + llist_del(&d->next, &poas.deferred); + + return d; + } + + return NULL; +} + +/* Publishes the flow_id so an answer can find this flow. */ +static void pf_set_flow_id(struct poa_flow * pf, + int flow_id) +{ + struct poa_deferred * d; + + pthread_mutex_lock(&poas.mtx); + + pf->flow_id = flow_id; + poas.id_to_pf[flow_id] = pf; + + d = deferred_take(flow_id); + + pthread_cond_broadcast(&poas.cond); + + pthread_mutex_unlock(&poas.mtx); + + if (d != NULL) { + pf_answer(pf, d->response, &d->data); + freebuf(d->data); + free(d); + } +} + +/* + * Between the request arriving and the accept returning, the flow has + * an id but no fd yet; flow_init claims the PoA here. + */ +static void pf_set_pending(struct poa_flow * pf) +{ + pthread_mutex_lock(&poas.mtx); + + pf->pending = true; + + pthread_mutex_unlock(&poas.mtx); +} + +static void pf_clr_pending(struct poa_flow * pf) +{ + pthread_mutex_lock(&poas.mtx); + + pf->pending = false; + + pthread_mutex_unlock(&poas.mtx); +} + +/* A process that attaches no PoA has nothing pending. */ +struct poa_flow * poa_flow_take_pending(int flow_id) +{ + struct poa_flow * pf; + + if (poas.state == POA_NULL) + return NULL; + + if (flow_id < 0 || flow_id >= SYS_MAX_FLOWS) + return NULL; + + pthread_mutex_lock(&poas.mtx); + + pf = poas.id_to_pf[flow_id]; + if (pf != NULL && pf->pending) + pf->pending = false; + else + pf = NULL; + + pthread_mutex_unlock(&poas.mtx); + + return pf; +} + +static struct poa_flow * pf_get(int flow_id) +{ + struct poa_flow * pf; + + if (flow_id < 0 || flow_id >= SYS_MAX_FLOWS) + return NULL; + + pthread_mutex_lock(&poas.mtx); + + pf = poas.id_to_pf[flow_id]; + + pthread_mutex_unlock(&poas.mtx); + + return pf; +} + +#ifdef PROC_FLOW_STATS + +/* Caller holds poas.lock. */ +static struct poa * poa_by_rib_name(const char * name) +{ + struct list_head * p; + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (strcmp(poa->name, name) == 0) + return poa; + } + + return NULL; +} + +static int poa_rib_read(const char * path, + char * buf, + size_t len) +{ + struct poa * poa; + const char * entry; + size_t qlen; + size_t avg; + size_t cost; + int size; + int ret; + + entry = strstr(path, RIB_SEPARATOR) + 1; + + if (len < POA_RIB_STRLEN) + return -1; + + pthread_rwlock_rdlock(&poas.lock); + + poa = poa_by_rib_name(entry); + if (poa == NULL) + goto fail; + + qlen = poa->ops->poa_qlen(poa); + avg = poa->avg_len; + cost = poa->avg_len + poa->avg_ovh; + + size = snprintf(buf, len, + "Active flows: %zu\n" + "Packets received: %zu\n" + "Bytes received: %zu\n" + "Packets sent: %zu\n" + "Bytes sent: %zu\n" + "Management frames rcvd: %zu\n" + "Management frames sent: %zu\n" + "Bad EID packets: %zu\n" + "Delivery (N+1) failures: %zu\n" + "Buffer alloc failures: %zu\n" + "Packet read failures: %zu\n" + "Packet send failures: %zu\n" + "Name queries sent: %zu\n" + "Name queries received: %zu\n" + "Name replies sent: %zu\n" + "Name replies received: %zu\n" + "Queued (transport): %zu\n" + "Queued (packets): %zu\n" + "Mean packet size (bytes): %zu\n" + "Mean packet cost: %zu\n", + POA_STAT_LOAD(poa, n_flows), + POA_STAT_LOAD(poa, rx_pkts), + POA_STAT_LOAD(poa, rx_bytes), + POA_STAT_LOAD(poa, tx_pkts), + POA_STAT_LOAD(poa, tx_bytes), + POA_STAT_LOAD(poa, mgmt_rx), + POA_STAT_LOAD(poa, mgmt_tx), + POA_STAT_LOAD(poa, bad_eid), + POA_STAT_LOAD(poa, dlv_fail), + POA_STAT_LOAD(poa, buf_fail), + POA_STAT_LOAD(poa, rcv_fail), + POA_STAT_LOAD(poa, snd_fail), + POA_STAT_LOAD(poa, qry_tx), + POA_STAT_LOAD(poa, qry_rx), + POA_STAT_LOAD(poa, rep_tx), + POA_STAT_LOAD(poa, rep_rx), + qlen, cost > 0 ? qlen / cost : 0, avg, cost); + if (size < 0 || (size_t) size >= len) + goto fail; + + if (poa->ops->poa_rib != NULL) { + ret = poa->ops->poa_rib(poa, buf + size, len - size); + if (ret < 0) + goto fail; + + size += ret; + } + + pthread_rwlock_unlock(&poas.lock); + + return size; + + fail: + pthread_rwlock_unlock(&poas.lock); + + return -1; +} + +static int poa_rib_readdir(char *** buf) +{ + struct list_head * p; + size_t n = 0; + int idx = 0; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) + ++n; + + if (n == 0) { + *buf = NULL; + goto no_poas; + } + + *buf = malloc(sizeof(**buf) * n); + if (*buf == NULL) + goto fail_entries; + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + (*buf)[idx] = malloc(strlen(poa->name) + 1); + if ((*buf)[idx] == NULL) + goto fail_entry; + + strcpy((*buf)[idx++], poa->name); + } + no_poas: + pthread_rwlock_unlock(&poas.lock); + + return idx; + + fail_entry: + while (idx-- > 0) + free((*buf)[idx]); + + free(*buf); + fail_entries: + pthread_rwlock_unlock(&poas.lock); + + return -ENOMEM; +} + +static int poa_rib_getattr(const char * path, + struct rib_attr * attr) +{ + (void) path; + + attr->size = POA_RIB_STRLEN; + attr->mtime = 0; + + return 0; +} + +static struct rib_ops poa_r_ops = { + .read = poa_rib_read, + .readdir = poa_rib_readdir, + .getattr = poa_rib_getattr +}; + +#endif /* PROC_FLOW_STATS */ + +int poa_init(const char * name) +{ + pthread_condattr_t cattr; + + assert(name != NULL); + + memset(&poas, 0, sizeof(poas)); + + str_hash(HASH_SHA3_256, poas.name_hash, name); + + poas.mon_fd = -1; + if (pthread_mutex_init(&poas.mtx, NULL) != 0) + goto fail_mtx; + + if (pthread_cond_init(&poas.cond, NULL) != 0) + goto fail_cond; + + if (pthread_mutex_init(&poas.mgmt_mtx, NULL) != 0) + goto fail_mgmt_mtx; + + if (pthread_condattr_init(&cattr) != 0) + goto fail_cattr; +#ifndef __APPLE__ + pthread_condattr_setclock(&cattr, PTHREAD_COND_CLOCK); +#endif + if (pthread_cond_init(&poas.mgmt_cond, &cattr) != 0) { + pthread_condattr_destroy(&cattr); + goto fail_cattr; + } + + pthread_condattr_destroy(&cattr); + + if (pthread_cond_init(&poas.done_cond, NULL) != 0) + goto fail_done_cond; + + if (pthread_rwlock_init(&poas.lock, NULL) != 0) + goto fail_lock; + + list_head_init(&poas.list); + llist_init(&poas.deferred); + llist_init(&poas.mgmt_frames); + list_head_init(&poas.detach); + + poas.state = POA_INIT; + +#ifdef PROC_FLOW_STATS + if (rib_reg(POA_RIB, &poa_r_ops) < 0) + goto fail_rib; +#endif + return 0; + +#ifdef PROC_FLOW_STATS + fail_rib: + pthread_rwlock_destroy(&poas.lock); +#endif + + fail_lock: + pthread_cond_destroy(&poas.done_cond); + fail_done_cond: + pthread_cond_destroy(&poas.mgmt_cond); + fail_cattr: + pthread_mutex_destroy(&poas.mgmt_mtx); + fail_mgmt_mtx: + pthread_cond_destroy(&poas.cond); + fail_cond: + pthread_mutex_destroy(&poas.mtx); + fail_mtx: + return -1; +} + +int poa_set_layer(const char * layer) +{ + if (layer == NULL) + return -EINVAL; + + pthread_rwlock_wrlock(&poas.lock); + + str_hash(HASH_SHA3_256, poas.layer_hash, layer); + + poas.state = POA_OPERATIONAL; + + pthread_rwlock_unlock(&poas.lock); + + return 0; +} + +bool poa_has_name(const uint8_t * hash) +{ + bool match = false; + + pthread_rwlock_rdlock(&poas.lock); + + if (poas.state >= POA_INIT) + match = memcmp(hash, poas.name_hash, POA_QUERY_HLEN) == 0; + + if (!match && poas.state >= POA_OPERATIONAL) + match = memcmp(hash, poas.layer_hash, POA_QUERY_HLEN) == 0; + + pthread_rwlock_unlock(&poas.lock); + + return match; +} + +int poa_spb_reserve(struct ssm_pk_buff ** spb, + size_t len) +{ + return ssm_pool_alloc_b(proc.pool, len, NULL, spb, NULL) < 0 ? -1 : 0; +} + +void poa_spb_release(struct ssm_pk_buff * spb) +{ + ssm_pool_remove(proc.pool, ssm_pk_buff_get_off(spb)); +} + +void poa_rx_pkt(struct poa * poa, + uint32_t eid, + struct ssm_pk_buff * spb) +{ + struct poa_flow * pf; + struct ssm_rbuff * rx_rb; + size_t len; + + len = ssm_pk_buff_len(spb); /* the ring write takes it over */ + + if (eid >= poa->n_eids) { + POA_STAT_BUMP(poa, bad_eid); + poa_spb_release(spb); + return; + } + + rcu_rdlock(&poa->guard); + + pf = rcu_deref(poa->eid_to_pf[eid]); + + rcu_consume(pf); + + if (pf == NULL) + goto fail_eid; + + rx_rb = rcu_deref(pf->rx_rb); + + rcu_consume(rx_rb); + + if (rx_rb == NULL) + goto fail_eid; + + if (ssm_rbuff_write(rx_rb, ssm_pk_buff_get_off(spb)) < 0) { + POA_STAT_BUMP(poa, dlv_fail); + rcu_rdunlock(&poa->guard); + poa_spb_release(spb); + return; + } + + POA_STAT_BUMP(poa, rx_pkts); + POA_STAT_ADD(poa, rx_bytes, len); + + ssm_flow_set_notify(proc.fqset, pf->flow_id, FLOW_PKT); + + rcu_rdunlock(&poa->guard); + + return; + + fail_eid: + POA_STAT_BUMP(poa, bad_eid); + + rcu_rdunlock(&poa->guard); + + poa_spb_release(spb); +} + +static int poa_flow_req_arr(struct poa_flow * pf, + qosspec_t qs, + uint32_t mtu, + const buffer_t * data) +{ + struct flow_info flow; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + buffer_t out = BUF_INIT; + int err; + + memset(&flow, 0, sizeof(flow)); + + flow.n_pid = getpid(); + flow.n_1_pid = getpid(); + flow.qs = qs; + flow.mpl = pf->poa->mpl; + flow.mtu = mtu; + if (ipcp_poa_flow_req_arr__irm_req_ser(&msg, &flow, data) < 0) + return -ENOMEM; + + pf_set_pending(pf); + + err = send_recv_msg(&msg); + if (err < 0) + goto fail; + + err = poa_flow__irm_result_des(&msg, &flow, &out); + if (err < 0) + goto fail; + + freebuf(out); + + if (flow.id < 0 || flow.id >= SYS_MAX_FLOWS) { + err = -EBADF; + goto fail; + } + + pf_set_flow_id(pf, flow.id); + + return 0; + fail: + pf_clr_pending(pf); + return err; +} + +static void handle_flow_req(struct poa * poa, + const struct poa_addr * src, + const struct poa_mgmt_msg * msg, + const uint8_t * data, + size_t data_len) +{ + struct list_head * p; + struct poa_flow * pf = NULL; + qosspec_t qs; + buffer_t buf; + buffer_t answer_data = BUF_INIT; + uint32_t r_eid; + uint32_t mtu; + uint32_t eid = 0; + int answer = 0; + bool found = false; + bool served = false; + + r_eid = ntoh32(msg->s_eid); + + poa_mgmt_msg_qos(msg, &qs); + + pthread_rwlock_wrlock(&poas.lock); + + list_for_each(p, &poa->flows) { + pf = list_entry(p, struct poa_flow, pend); + if (pf->r_eid != r_eid || poa_addr_cmp(&pf->r_addr, src) != 0) + continue; + + answer = pf->answer; + eid = pf->eid; + served = true; + + found = pf->answer_sent; + if (found && pf->answer_data.len > 0) { + answer_data.data = malloc(pf->answer_data.len); + if (answer_data.data != NULL) { + memcpy(answer_data.data, pf->answer_data.data, + pf->answer_data.len); + answer_data.len = pf->answer_data.len; + } + } + break; + } + + if (!served) { + pf = pf_create(poa, src); + if (pf != NULL) + pf->r_eid = r_eid; + } + + pthread_rwlock_unlock(&poas.lock); + + if (found) { + mgmt_send(poa, src, POA_FLOW_REPLY, eid, r_eid, qos_raw, answer, + answer_data.len > 0 ? &answer_data : NULL); + freebuf(answer_data); + } + + if (served || pf == NULL) + return; + + buf.len = data_len; + buf.data = (uint8_t *) data; + mtu = poa->ops->poa_mtu(poa, src); + if (poa_flow_req_arr(pf, qs, mtu, &buf) < 0) { + mgmt_send(poa, src, POA_FLOW_REPLY, pf->eid, r_eid, qos_raw, + -1, NULL); + pf_destroy(pf); + } +} + +static void handle_flow_reply(struct poa * poa, + const struct poa_addr * src, + const struct poa_mgmt_msg * msg, + const uint8_t * data, + size_t data_len) +{ + struct poa_flow * pf; + uint32_t eid; + + eid = ntoh32(msg->d_eid); + + pthread_rwlock_rdlock(&poas.lock); + + pf = eid < poa->n_eids ? poa->eid_to_pf[eid] : NULL; + if (pf == NULL || pf->state != POA_FLOW_PENDING) { + pthread_rwlock_unlock(&poas.lock); + return; + } + + if (poa_addr_cmp(&pf->r_addr, src) != 0) { + pthread_rwlock_unlock(&poas.lock); + return; + } + + pthread_mutex_lock(&pf->mtx); + + if (pf->replied) { + pthread_mutex_unlock(&pf->mtx); + pthread_rwlock_unlock(&poas.lock); + return; + } + + if (data_len > 0) { + pf->resp_data.data = malloc(data_len); + if (pf->resp_data.data != NULL) { + memcpy(pf->resp_data.data, data, data_len); + + pf->resp_data.len = data_len; + } + } + + pf->r_eid = ntoh32(msg->s_eid); + pf->response = ntoh32(msg->response); + pf->replied = true; + + pthread_cond_broadcast(&pf->cond); + + pthread_mutex_unlock(&pf->mtx); + + pthread_rwlock_unlock(&poas.lock); +} + +static void handle_flow_update(struct poa * poa, + const struct poa_addr * src, + const struct poa_mgmt_msg * msg, + const uint8_t * data, + size_t data_len) +{ + struct poa_flow * pf; + buffer_t buf; + uint32_t eid; + + eid = ntoh32(msg->d_eid); + + pthread_rwlock_rdlock(&poas.lock); + + pf = eid < poa->n_eids ? poa->eid_to_pf[eid] : NULL; + if (pf == NULL || pf->state != POA_FLOW_ALLOCATED) { + pthread_rwlock_unlock(&poas.lock); + return; + } + + if (poa_addr_cmp(&pf->r_addr, src) != 0) { + pthread_rwlock_unlock(&poas.lock); + return; + } + + eid = (uint32_t) pf->flow_id; + + pthread_rwlock_unlock(&poas.lock); + + buf.len = data_len; + buf.data = (uint8_t *) data; + + ipcp_flow_update_arr((int) eid, &buf); +} + +static void mgmt_frame_handle(struct poa_mgmt_frame * frame) +{ + const struct poa_mgmt_msg * msg; + const uint8_t * data; + size_t data_len; + + msg = (const struct poa_mgmt_msg *) frame->buf; + if (frame->len < sizeof(*msg)) + return; + + data_len = ntoh16(msg->data_len); + if (data_len > frame->len - sizeof(*msg)) + return; + + data = frame->buf + sizeof(*msg); + + switch (msg->code) { + case POA_FLOW_REQ: + handle_flow_req(frame->poa, &frame->src, msg, data, data_len); + break; + case POA_FLOW_REPLY: + handle_flow_reply(frame->poa, &frame->src, msg, data, data_len); + break; + case POA_FLOW_UPDATE: + handle_flow_update(frame->poa, &frame->src, msg, data, + data_len); + break; + default: + break; + } +} + +static bool pf_steal(struct poa_flow * pf) +{ + bool stolen = false; + + pthread_mutex_lock(&poas.mtx); + + if (pf->pending) { + pf->pending = false; + poas.id_to_pf[pf->flow_id] = NULL; + + deferred_purge(pf->flow_id); + + stolen = true; + } + + pthread_mutex_unlock(&poas.mtx); + + return stolen; +} + +static void sweep_pending(void) +{ + struct poa_flow * dead[POA_SWEEP_MAX]; + struct list_head * p; + struct list_head * q; + struct timespec now; + size_t n = 0; + size_t i; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + list_for_each(q, &poa->flows) { + struct poa_flow * pf; + + if (n == POA_SWEEP_MAX) + break; + + pf = list_entry(q, struct poa_flow, pend); + if (pf->state != POA_FLOW_PENDING || pf->claimed) + continue; + + if (now.tv_sec - pf->t0.tv_sec < POA_PEND_TIMEO) + continue; + + if (pf->flow_id >= 0 && !pf_steal(pf)) + continue; + + dead[n++] = pf; + } + + if (n == POA_SWEEP_MAX) + break; + } + + pthread_rwlock_unlock(&poas.lock); + + for (i = 0; i < n; ++i) + pf_destroy(dead[i]); + + pthread_mutex_lock(&poas.mtx); + + list_for_each_safe(p, q, &poas.deferred.list) { + struct poa_deferred * d; + + d = list_entry(p, struct poa_deferred, next); + if (now.tv_sec - d->t0.tv_sec < POA_PEND_TIMEO) + continue; + + llist_del(&d->next, &poas.deferred); + freebuf(d->data); + free(d); + } + + pthread_mutex_unlock(&poas.mtx); +} + +void poa_rx_mgmt(struct poa * poa, + const struct poa_addr * src, + const uint8_t * buf, + size_t len) +{ + struct poa_mgmt_frame * frame; + + if (len < sizeof(struct poa_mgmt_msg) || len > POA_MGMT_BUFSZ) + return; + + POA_STAT_BUMP(poa, mgmt_rx); + + frame = malloc(offsetof(struct poa_mgmt_frame, buf) + len); + if (frame == NULL) + return; + + frame->poa = poa; + frame->src = *src; + frame->len = len; + + memcpy(frame->buf, buf, len); + + pthread_mutex_lock(&poas.mgmt_mtx); + + if (poa->n_mgmt >= POA_MGMT_QMAX) { + pthread_mutex_unlock(&poas.mgmt_mtx); + free(frame); + return; + } + + ++poa->n_mgmt; + + llist_add_tail(&frame->next, &poas.mgmt_frames); + + pthread_cond_signal(&poas.mgmt_cond); + + pthread_mutex_unlock(&poas.mgmt_mtx); +} + +static void mgmt_frames_purge(const struct poa * poa) +{ + struct list_head * p; + struct list_head * h; + + pthread_mutex_lock(&poas.mgmt_mtx); + + list_for_each_safe(p, h, &poas.mgmt_frames.list) { + struct poa_mgmt_frame * frame; + + frame = list_entry(p, struct poa_mgmt_frame, next); + if (frame->poa != poa) + continue; + + --frame->poa->n_mgmt; + + llist_del(&frame->next, &poas.mgmt_frames); + + free(frame); + } + + pthread_mutex_unlock(&poas.mgmt_mtx); +} + +static bool poa_has_id(const struct poa * poa, + const struct poa_spec * spec) +{ + if (poa->type != spec->type) + return false; + + return poa->ops->poa_has_id(poa, spec); +} + +/* + * The PoA carrying dst, as the transport judges it. -EPERM if none + * matches, -EINVAL if several do. Caller holds poas.lock. + */ +static int poa_lookup(const struct poa_addr * dst, + struct poa ** out) +{ + struct list_head * p; + struct poa * found = NULL; + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->type != dst->type) + continue; + + if (!poa->ops->poa_match(poa, dst)) + continue; + + if (found != NULL) /* nothing given, two candidates */ + return -EINVAL; + + found = poa; + } + + if (found == NULL) + return -EPERM; + + *out = found; + + return 0; +} + +/* Sends are deadlined, bounding the lock hold on a full queue. */ +int poa_bcast_mgmt(const struct poa_addr * dst, + const uint8_t * buf, + size_t len) +{ + struct list_head * p; + int n = 0; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->type != dst->type) + continue; + + if (!poa->ops->poa_match(poa, dst)) + continue; + + if (poa->ops->poa_send_mgmt(poa, dst, buf, len) < 0) + continue; + + /* All management broadcasts are name queries. */ + POA_STAT_BUMP(poa, qry_tx); + ++n; + } + + pthread_rwlock_unlock(&poas.lock); + + return n; +} + +static bool deadline_is_malformed(const struct timespec * timeo) +{ + if (timeo == NULL) + return false; + + if (timeo->tv_sec < 0 || timeo->tv_nsec < 0) + return true; + + return timeo->tv_nsec >= BILLION; +} + +/* + * Complete addr for dst on any backend that can query. The ops are + * collected under poas.lock but called outside it: a query blocks up + * to its deadline and takes the lock again to broadcast. The tables + * are static, so nothing dangles; a struct poa cannot be carried + * across the unlock. The deadline applies per backend. + */ +int poa_query(const char * dst, + const struct timespec * timeo, + struct poa_addr * addr) +{ + const struct poa_ops * cand[POA_MAX_POAS]; + enum poa_type type[POA_MAX_POAS]; + struct list_head * p; + size_t n = 0; + size_t i; + int err = -ENOTSUP; + + if (dst == NULL || addr == NULL) + return -EINVAL; + + if (deadline_is_malformed(timeo)) + return -EINVAL; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->ops->poa_query == NULL) + continue; + + for (i = 0; i < n && cand[i] != poa->ops; i++) + ; + if (i < n) + continue; + + /* One type per backend: eth. Revisit if that changes. */ + cand[n] = poa->ops; + type[n++] = poa->type; + } + + pthread_rwlock_unlock(&poas.lock); + + for (i = 0; i < n; i++) { + memset(addr, 0, sizeof(*addr)); + + addr->type = type[i]; + + err = cand[i]->poa_query(dst, timeo, addr); + if (err == 0) + return 0; + } + + return err; +} + +static int poa_check(const struct poa_addr * dst) +{ + struct poa * poa; + int err; + + pthread_rwlock_rdlock(&poas.lock); + + err = poa_lookup(dst, &poa); + + pthread_rwlock_unlock(&poas.lock); + + return err; +} + +static int poa_alloc(const struct poa_addr * dst, + qosspec_t qs, + const buffer_t * req, + buffer_t * resp, + struct poa_flow ** pf_out, + uint32_t * mtu, + const struct timespec * timeo) +{ + struct timespec dflt = TIMESPEC_INIT_MS(POA_ALLOC_TIMEO); + struct timespec rintv = TIMESPEC_INIT_MS(POA_RETRY_TIMEO); + struct poa_flow * pf; + struct poa * poa; + struct timespec abstime; + struct timespec now; + struct timespec retry; + int err; + + pthread_rwlock_wrlock(&poas.lock); + + err = poa_lookup(dst, &poa); + if (err < 0) { + pthread_rwlock_unlock(&poas.lock); + return err; + } + + pf = pf_create(poa, dst); + if (pf == NULL) { + pthread_rwlock_unlock(&poas.lock); + return -ENOMEM; + } + + pf->claimed = true; + + pthread_rwlock_unlock(&poas.lock); + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + ts_add(&abstime, timeo != NULL ? timeo : &dflt, &abstime); + + pthread_mutex_lock(&pf->mtx); + + while (!pf->replied) { + pthread_mutex_unlock(&pf->mtx); + + err = mgmt_send(poa, dst, POA_FLOW_REQ, pf->eid, 0, qs, 0, req); + + pthread_mutex_lock(&pf->mtx); + + if (err < 0 && err != -ETIMEDOUT && err != -EAGAIN) { + err = -EIO; + goto fail; + } + + if (pf->replied) + break; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + ts_add(&now, &rintv, &retry); + + if (ts_diff_ns(&retry, &abstime) > 0) + retry = abstime; + + pthread_cond_timedwait(&pf->cond, &pf->mtx, &retry); + + if (pf->replied) + break; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + if (ts_diff_ns(&now, &abstime) >= 0) { + err = -ETIMEDOUT; + goto fail; + } + } + + if (pf->response != 0) { + err = -ECONNREFUSED; + goto fail; + } + + *resp = pf->resp_data; + pf->resp_data.len = 0; + pf->resp_data.data = NULL; + + pthread_mutex_unlock(&pf->mtx); + + *mtu = poa->ops->poa_mtu(poa, dst); + *pf_out = pf; + + return 0; + + fail: + pthread_mutex_unlock(&pf->mtx); + pf_destroy(pf); + return err; +} + +static void poa_alloc_fail(struct poa_flow * pf) +{ + pf_destroy(pf); +} + +void poa_flow_attach(struct poa_flow * pf, + int flow_id, + struct ssm_rbuff * rx_rb) +{ + struct poa * poa = pf->poa; + + if (pf->flow_id != flow_id) + pf_set_flow_id(pf, flow_id); + + pthread_rwlock_wrlock(&poas.lock); + + pf->state = POA_FLOW_ALLOCATED; + + rcu_wrlock(&poa->guard); + + rcu_publish(rx_rb); + rcu_assign(pf->rx_rb, rx_rb); + + rcu_wrunlock(&poa->guard); + + pthread_rwlock_unlock(&poas.lock); +} + +void poa_flow_detach(struct poa_flow * pf) +{ + struct poa * poa = pf->poa; + + pthread_rwlock_wrlock(&poas.lock); + + pf->state = POA_FLOW_DEAD; + + rcu_wrlock(&poa->guard); + + rcu_assign(pf->rx_rb, NULL); + + rcu_wrunlock(&poa->guard); + + pthread_rwlock_unlock(&poas.lock); + + pf_destroy(pf); +} + +static size_t flows_updown(struct poa * poa, + bool up) +{ + struct list_head * p; + size_t n = 0; + + list_for_each(p, &poa->flows) { + struct poa_flow * pf; + struct flow * flow; + + pf = list_entry(p, struct poa_flow, pend); + if (pf->state != POA_FLOW_ALLOCATED || pf->flow_id < 0) + continue; + + flow = &proc.flows[proc.id_to_fd[pf->flow_id].fd]; + if (flow->info.id != pf->flow_id) + continue; + + if (((flow->oflags & FLOWFDOWN) != 0) != !up) + ++n; + + if (up) { + flow->oflags &= ~FLOWFDOWN; + + ssm_rbuff_clr_bits(flow->rx_rb, RB_FLOWDOWN); + } else { + flow->oflags |= FLOWFDOWN; + + ssm_rbuff_set_bits(flow->rx_rb, RB_FLOWDOWN); + } + + ssm_flow_set_notify(proc.fqset, pf->flow_id, + up ? FLOW_UP : FLOW_DOWN); + } + + return n; +} + +size_t poa_link_updown(int id, + bool up) +{ + struct list_head * p; + size_t n = 0; + + pthread_rwlock_wrlock(&proc.lock); + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->ops->poa_link_match == NULL) + continue; + + if (!poa->ops->poa_link_match(poa, id)) + continue; + + n += flows_updown(poa, up); + } + + pthread_rwlock_unlock(&poas.lock); + pthread_rwlock_unlock(&proc.lock); + + return n; +} + +static size_t poa_ewma(size_t avg, + size_t sz) +{ + if (avg == 0) + return sz; + + avg = avg + (sz >> POA_AVG_SHIFT) - (avg >> POA_AVG_SHIFT); + + return avg == 0 ? 1 : avg; +} + +static void poa_avg_len_update(struct poa * poa, + size_t sz) +{ + STORE_RELAXED(&poa->avg_len, poa_ewma(LOAD_RELAXED(&poa->avg_len), sz)); +} + +static size_t poa_qlen(struct poa * poa) +{ + struct timespec now; + uint64_t ns; + size_t qlen; + + clock_gettime(PTHREAD_COND_CLOCK, &now); + + ns = TS_TO_UINT64(now); + if (ns - LOAD_RELAXED(&poa->q_time) < POA_QLEN_GATE) + return LOAD_RELAXED(&poa->q_cache); + + qlen = poa->ops->poa_qlen(poa); + + STORE_RELAXED(&poa->q_cache, qlen); + STORE_RELAXED(&poa->q_time, ns); + + return qlen; +} + +static void poa_cost_sample(struct poa * poa, + size_t before, + size_t len) +{ + size_t after; + + after = poa->ops->poa_qlen(poa); + if (after <= before) + return; /* drained; nothing to learn */ + + after -= before; + if (after < len || after > (len << 2) + 1024) + return; + + STORE_RELAXED(&poa->avg_ovh, + poa_ewma(LOAD_RELAXED(&poa->avg_ovh), after - len)); +} + +int poa_flow_tx(struct poa_flow * pf, + struct ssm_pk_buff * spb, + bool block, + const struct timespec * abstime) +{ + struct poa * poa = pf->poa; + size_t len = ssm_pk_buff_len(spb); + size_t before = 0; + bool sample; + int ret; + + sample = (LOAD_RELAXED(&poa->n_tx) & POA_COST_MASK) == 0; + if (sample) + before = poa->ops->poa_qlen(poa); + + ret = poa->ops->poa_send(poa, &pf->r_addr, pf->r_eid, spb, block, + abstime); + if (ret < 0) { /* the caller releases the buffer */ + POA_STAT_BUMP(poa, snd_fail); + return ret; + } + + POA_STAT_BUMP(poa, tx_pkts); + POA_STAT_ADD(poa, tx_bytes, len); + + FETCH_ADD_RELAXED(&poa->n_tx, 1); + + poa_avg_len_update(poa, len); + + if (sample) + poa_cost_sample(poa, before, len); + + poa_spb_release(spb); + + return 0; +} + +size_t poa_flow_qlen(const struct poa_flow * pf) +{ + struct poa * poa = pf->poa; + uint64_t bytes; + size_t cost; + size_t pkts; + size_t byts; + + if (poa->ops->poa_qpkts != NULL && + poa->ops->poa_qpkts(poa, &pkts, &byts) == 0) + return byts; + + cost = LOAD_RELAXED(&poa->avg_len) + LOAD_RELAXED(&poa->avg_ovh); + if (LOAD_RELAXED(&poa->avg_ovh) == 0 || cost == 0) + return poa_qlen(poa); /* overstated beats false empty */ + + bytes = (uint64_t) poa_qlen(poa) * LOAD_RELAXED(&poa->avg_len); + + return (size_t) (bytes / cost); +} + +size_t poa_flow_qpkts(const struct poa_flow * pf) +{ + struct poa * poa = pf->poa; + size_t cost; + size_t pkts; + size_t byts; + + if (poa->ops->poa_qpkts != NULL && + poa->ops->poa_qpkts(poa, &pkts, &byts) == 0) + return pkts; + + cost = LOAD_RELAXED(&poa->avg_len) + LOAD_RELAXED(&poa->avg_ovh); + if (LOAD_RELAXED(&poa->avg_ovh) == 0 || cost == 0) + return 0; + + return poa_qlen(poa) / cost; +} + +size_t poa_flow_mean_len(const struct poa_flow * pf) +{ + return LOAD_RELAXED(&pf->poa->avg_len); +} + +int poa_flow_qid(const struct poa_flow * pf) +{ + return pf->poa->qid; +} + +void poa_flow_ready(struct poa_flow * pf) +{ + buffer_t data; + int answer; + + if (pf == NULL) + return; + + clrbuf(data); + + pthread_rwlock_wrlock(&poas.lock); + + if (!pf->answered || pf->answer_sent) { + pthread_rwlock_unlock(&poas.lock); + return; + } + + answer = pf->answer; + + if (pf->answer_data.len > 0) { + data.data = malloc(pf->answer_data.len); + if (data.data != NULL) { + memcpy(data.data, pf->answer_data.data, + pf->answer_data.len); + data.len = pf->answer_data.len; + } + } + + pf->answer_sent = true; + + pthread_rwlock_unlock(&poas.lock); + + mgmt_send(pf->poa, &pf->r_addr, POA_FLOW_REPLY, pf->eid, + pf->r_eid, qos_raw, answer, &data); + + freebuf(data); +} + +int poa_flow_alloc_resp(int flow_id, + int response, + const buffer_t * data) +{ + struct poa_deferred * d; + struct poa_flow * pf; + + if (flow_id < 0 || flow_id >= SYS_MAX_FLOWS) + return -EPERM; + + pthread_mutex_lock(&poas.mtx); + + pf = poas.id_to_pf[flow_id]; + if (pf != NULL) { + pthread_mutex_unlock(&poas.mtx); + return pf_answer(pf, response, data); + } + + if (poas.deferred.len >= POA_DEFER_MAX) { + pthread_mutex_unlock(&poas.mtx); + return -ENOMEM; + } + + d = malloc(sizeof(*d)); + if (d == NULL) { + pthread_mutex_unlock(&poas.mtx); + return -ENOMEM; + } + + memset(d, 0, sizeof(*d)); + + clock_gettime(PTHREAD_COND_CLOCK, &d->t0); + + d->flow_id = flow_id; + d->response = response; + + if (data != NULL && data->len > 0) { + d->data.data = malloc(data->len); + if (d->data.data == NULL) { + free(d); + pthread_mutex_unlock(&poas.mtx); + return -ENOMEM; + } + memcpy(d->data.data, data->data, data->len); + + d->data.len = data->len; + } + + llist_add_tail(&d->next, &poas.deferred); + + pthread_mutex_unlock(&poas.mtx); + + return 0; +} + +int poa_flow_update(int flow_id, + const buffer_t * data) +{ + struct poa_flow * pf; + + pf = pf_get(flow_id); + if (pf == NULL) + return -EPERM; + + return mgmt_send(pf->poa, &pf->r_addr, POA_FLOW_UPDATE, pf->eid, + pf->r_eid, qos_raw, 0, data); +} + +/* The PoA state is released when the flow itself is torn down. */ +int poa_flow_dealloc(int flow_id) +{ + (void) flow_id; + + return 0; +} + +/* PoA id 0 = management channel. */ +static struct poa * poa_create(enum poa_type type, + const struct poa_ops * ops, + size_t n_eids) +{ + struct poa * poa; + + poa = malloc(sizeof(*poa)); + if (poa == NULL) + goto fail_malloc; + + memset(poa, 0, sizeof(*poa)); + + poa->eid_to_pf = malloc(sizeof(*poa->eid_to_pf) * n_eids); + if (poa->eid_to_pf == NULL) + goto fail_map; + + memset(poa->eid_to_pf, 0, sizeof(*poa->eid_to_pf) * n_eids); + + poa->eids = bmp_create(n_eids - 1, 1); + if (poa->eids == NULL) + goto fail_bmp; + + if (rcu_guard_init(&poa->guard) != 0) + goto fail_guard; + + list_head_init(&poa->next); + list_head_init(&poa->flows); + + poa->type = type; + poa->ops = ops; + poa->mpl = ops->mpl; + poa->n_eids = n_eids; + poa->qid = -1; + + return poa; + + fail_guard: + bmp_destroy(poa->eids); + fail_bmp: + free(poa->eid_to_pf); + fail_map: + free(poa); + fail_malloc: + return NULL; +} + +static void poa_destroy(struct poa * poa) +{ + rcu_guard_fini(&poa->guard); + + bmp_destroy(poa->eids); + + free(poa->eid_to_pf); + free(poa); +} + +static void poa_teardown(struct poa * poa) +{ + if (poas.state >= POA_RUNNING) + poa->ops->poa_stop(poa); + + mgmt_frames_purge(poa); + + poa->ops->poa_detach(poa); + + poa_destroy(poa); +} + +static void poa_detach_all(void) +{ + pthread_rwlock_wrlock(&poas.lock); + + while (!list_is_empty(&poas.list)) { + struct poa * poa; + + poa = list_first_entry(&poas.list, struct poa, next); + + list_del(&poa->next); + + pthread_rwlock_unlock(&poas.lock); + + poa_teardown(poa); + + pthread_rwlock_wrlock(&poas.lock); + } + + pthread_rwlock_unlock(&poas.lock); +} + +static int poa_do_detach(const struct poa_detach_req * req) +{ + struct list_head * p; + struct poa * found = NULL; + + pthread_rwlock_wrlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (!poa_has_id(poa, &req->spec)) + continue; + + found = poa; + break; + } + + if (found == NULL) { + pthread_rwlock_unlock(&poas.lock); + return -ENOENT; + } + + if (!list_is_empty(&found->flows)) { + pthread_rwlock_unlock(&poas.lock); + return -EBUSY; + } + + list_del(&found->next); + + pthread_rwlock_unlock(&poas.lock); + + poa_teardown(found); + + return 0; +} + +static int poa_del(const struct poa_spec * spec) +{ + struct poa_detach_req req; + int cs; + int ret; + + memset(&req, 0, sizeof(req)); + + req.spec = *spec; + + pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cs); + + if (poas.state < POA_RUNNING) { /* set before workers run */ + ret = poa_do_detach(&req); + goto out; + } + + pthread_mutex_lock(&poas.mgmt_mtx); + + if (poas.mgmt_stop) { /* stopped: poa_fini reaps these */ + pthread_mutex_unlock(&poas.mgmt_mtx); + + ret = -EBUSY; + goto out; + } + + list_add_tail(&req.next, &poas.detach); + + pthread_cond_signal(&poas.mgmt_cond); + + while (!req.done) + pthread_cond_wait(&poas.done_cond, &poas.mgmt_mtx); + + pthread_mutex_unlock(&poas.mgmt_mtx); + + ret = req.result; + out: + pthread_setcancelstate(cs, NULL); + + return ret; +} + +static __inline__ bool mgmt_idle(void) +{ + if (poas.mgmt_stop) + return false; + + if (!llist_is_empty(&poas.mgmt_frames)) + return false; + + return list_is_empty(&poas.detach); +} + +static void detach_run(void) +{ + while (!list_is_empty(&poas.detach)) { + struct poa_detach_req * req; + + req = list_first_entry(&poas.detach, struct poa_detach_req, + next); + list_del(&req->next); + + pthread_mutex_unlock(&poas.mgmt_mtx); + + req->result = poa_do_detach(req); + + pthread_mutex_lock(&poas.mgmt_mtx); + + req->done = true; + + pthread_cond_broadcast(&poas.done_cond); + } +} + +static void * mgmt_handler(void * o) +{ + struct timespec intv = TIMESPEC_INIT_MS(POA_SWEEP_TIMEO); + + (void) o; + + while (true) { + struct poa_mgmt_frame * frame; + struct timespec abstime; + + pthread_mutex_lock(&poas.mgmt_mtx); + + detach_run(); + + while (mgmt_idle()) { + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + ts_add(&abstime, &intv, &abstime); + + if (pthread_cond_timedwait(&poas.mgmt_cond, + &poas.mgmt_mtx, + &abstime) == ETIMEDOUT) { + pthread_mutex_unlock(&poas.mgmt_mtx); + sweep_pending(); + pthread_mutex_lock(&poas.mgmt_mtx); + } + } + + if (poas.mgmt_stop) { + detach_run(); /* nobody else serves these */ + + pthread_mutex_unlock(&poas.mgmt_mtx); + break; + } + + if (llist_is_empty(&poas.mgmt_frames)) { + pthread_mutex_unlock(&poas.mgmt_mtx); + continue; + } + + frame = llist_first_entry(&poas.mgmt_frames, + struct poa_mgmt_frame, next); + llist_del(&frame->next, &poas.mgmt_frames); + + --frame->poa->n_mgmt; + + pthread_mutex_unlock(&poas.mgmt_mtx); + + mgmt_frame_handle(frame); + + free(frame); + } + + return (void *) 0; +} + +#ifndef HAVE_ETH +/* Only the Ethernet transport reports link events. */ +int poa_monitor_open(void) +{ + return -1; +} + +void poa_monitor_read(int fd) +{ + (void) fd; +} +#endif + +static void * poa_monitor(void * o) +{ + (void) o; + + while (true) + poa_monitor_read(poas.mon_fd); + + return (void *) 0; +} + +static int monitor_start(void) +{ + poas.mon_fd = poa_monitor_open(); + if (poas.mon_fd < 0) + return 0; + + if (pthread_create(&poas.monitor, NULL, poa_monitor, NULL) == 0) + return 0; + + close(poas.mon_fd); + + poas.mon_fd = -1; + + return -1; +} + +static void monitor_stop(void) +{ + if (poas.mon_fd < 0) + return; + + pthread_cancel(poas.monitor); + pthread_join(poas.monitor, NULL); + + close(poas.mon_fd); + + poas.mon_fd = -1; +} + +int poa_start(void) +{ + struct list_head * p; + struct list_head * h; + + if (poas.state == POA_NULL) + return 0; + + if (pthread_create(&poas.mgmt_handler, NULL, mgmt_handler, NULL) != 0) + goto fail_mgmt; + + if (monitor_start() < 0) + goto fail_monitor; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->ops->poa_start(poa) < 0) + goto fail_reader; + } + + poas.state = POA_RUNNING; + + pthread_rwlock_unlock(&poas.lock); + + return 0; + + fail_reader: + list_for_each(h, &poas.list) { + struct poa * poa = list_entry(h, struct poa, next); + + if (h == p) + break; + + poa->ops->poa_stop(poa); + } + + pthread_rwlock_unlock(&poas.lock); + + monitor_stop(); + fail_monitor: + pthread_mutex_lock(&poas.mgmt_mtx); + + poas.mgmt_stop = true; + + pthread_cond_broadcast(&poas.mgmt_cond); + pthread_mutex_unlock(&poas.mgmt_mtx); + + pthread_join(poas.mgmt_handler, NULL); + fail_mgmt: + return -1; +} + +void poa_stop(void) +{ + struct list_head * p; + + if (poas.state < POA_RUNNING) + return; + + pthread_mutex_lock(&poas.mgmt_mtx); + + poas.mgmt_stop = true; + + pthread_cond_broadcast(&poas.mgmt_cond); + pthread_mutex_unlock(&poas.mgmt_mtx); + + pthread_join(poas.mgmt_handler, NULL); + + monitor_stop(); + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + poa->ops->poa_stop(poa); + } + + pthread_rwlock_unlock(&poas.lock); + + poas.state = POA_INIT; +} + +void poa_fini(void) +{ + if (poas.state == POA_NULL) + return; + + poa_stop(); + + poas.state = POA_NULL; + + poa_detach_all(); + +#ifdef PROC_FLOW_STATS + rib_unreg(POA_RIB); +#endif + + pthread_rwlock_destroy(&poas.lock); + pthread_cond_destroy(&poas.done_cond); + pthread_cond_destroy(&poas.mgmt_cond); + pthread_mutex_destroy(&poas.mgmt_mtx); + pthread_cond_destroy(&poas.cond); + pthread_mutex_destroy(&poas.mtx); +} + +/* + * Lowest queue id no attached PoA holds; detaching frees it by + * leaving the list. Caller holds poas.lock. + */ +static int poa_qid_alloc(void) +{ + struct list_head * p; + bool used[POA_MAX_POAS]; + int i; + + memset(used, 0, sizeof(used)); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->qid >= 0 && poa->qid < POA_MAX_POAS) + used[poa->qid] = true; + } + + for (i = 0; i < POA_MAX_POAS; i++) + if (!used[i]) + return i; + + return -1; +} + +static int poa_add(const struct poa_spec * spec, + const struct poa_ops * ops, + size_t n_eids) +{ + struct list_head * p; + struct poa * poa; + int err; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + poa = list_entry(p, struct poa, next); + if (poa_has_id(poa, spec)) { + pthread_rwlock_unlock(&poas.lock); + return -EPERM; + } + } + + pthread_rwlock_unlock(&poas.lock); + + poa = poa_create(spec->type, ops, n_eids); + if (poa == NULL) + return -ENOMEM; + + err = poa->ops->poa_attach(poa, spec); + if (err < 0) + goto fail_bind; + + err = poa_addr_name(&poa->local, poa->name, sizeof(poa->name)); + if (err < 0) + goto fail_start; + + err = -1; + + pthread_rwlock_wrlock(&poas.lock); + + poa->qid = poa_qid_alloc(); + if (poa->qid < 0) { + pthread_rwlock_unlock(&poas.lock); + goto fail_start; + } + + if (poas.state >= POA_RUNNING && poa->ops->poa_start(poa) < 0) { + pthread_rwlock_unlock(&poas.lock); + goto fail_start; + } + + list_add_tail(&poa->next, &poas.list); + + pthread_rwlock_unlock(&poas.lock); + + return 0; + + fail_start: + poa->ops->poa_detach(poa); + fail_bind: + poa_destroy(poa); + return err; +} + +/* The single place a type is bound to its transport. */ +int poa_attach(const struct poa_spec * poa) +{ + if (poa == NULL) + return -EINVAL; + + switch (poa->type) { + case POA_UDP4: + /* FALLTHRU */ + case POA_UDP6: + return poa_add(poa, &udp_poa_ops, POA_MAX_EIDS); + case POA_ETH: +#ifdef HAVE_ETH + return poa_add(poa, ð_poa_ops, POA_MAX_EIDS); +#else + return -ENOTSUP; +#endif + default: + return -ENOTSUP; + } +} + +int poa_detach(const struct poa_spec * poa) +{ + if (poa == NULL) + return -EINVAL; + + return poa_del(poa); +} + +ssize_t poa_list(struct poa_spec * specs, + size_t max) +{ + struct list_head * p; + size_t n = 0; + + if (specs == NULL) + return -EINVAL; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (n++ >= max) + continue; + + memset(specs, 0, sizeof(*specs)); + + poa->ops->poa_spec(poa, specs); + + specs++; + } + + pthread_rwlock_unlock(&poas.lock); + + return (ssize_t) n; +} + +/* + * Complete peer for dst on the backend serving its type. The ops are + * borrowed under poas.lock and called outside it (see poa_query); if + * every PoA of the type detaches in between, the query's broadcast + * reaches nothing and reports -EPERM, as the lookup would. + */ +static int poa_peer_resolve(const char * dst, + struct poa_addr * peer) +{ + const struct poa_ops * ops = NULL; + struct list_head * p; + + pthread_rwlock_rdlock(&poas.lock); + + list_for_each(p, &poas.list) { + struct poa * poa = list_entry(p, struct poa, next); + + if (poa->type == peer->type) { + ops = poa->ops; + break; + } + } + + pthread_rwlock_unlock(&poas.lock); + + if (ops == NULL) /* nothing could carry the flow */ + return -EPERM; + + if (ops->poa_query == NULL) /* these addresses arrive complete */ + return 0; + + return ops->poa_query(dst, NULL, peer); +} + +/* + * Three steps: the IRMd creates the flow and prepares the key exchange, + * the PoA handshakes with the peer, the IRMd completes the + * exchange and hands us the key. + */ +int poa_flow_alloc(const char * dst, + const struct poa_addr * addr, + qosspec_t * qs, + const struct timespec * timeo) +{ + struct flow_info flow; + struct poa_flow * pf; + struct poa_addr peer; + struct crypt_sk crypt; + struct timespec t0; + struct timespec t1; + uint8_t key[SYMMKEYSZ]; + uint8_t buf[SOCK_BUF_SIZE]; + buffer_t msg = {SOCK_BUF_SIZE, buf}; + buffer_t req; + buffer_t resp; + uint32_t mtu = 0; + int err; + + if (addr == NULL) + return -EINVAL; + + if (qs != NULL && qs->service == SVC_STREAM && qs->loss != 0) + return -EINVAL; + + peer = *addr; + + err = poa_peer_resolve(dst, &peer); + if (err < 0) + return err; + + addr = &peer; + + err = poa_check(addr); + if (err < 0) + return err; + + memset(&flow, 0, sizeof(flow)); + + flow.n_pid = getpid(); + flow.n_1_pid = getpid(); + flow.qs = qs == NULL ? qos_raw : *qs; + if (poa_flow_alloc__irm_req_ser(&msg, &flow, dst) < 0) + return -ENOMEM; + + err = send_recv_msg(&msg); + if (err < 0) + return err; + + clrbuf(req); + clrbuf(resp); + + err = poa_flow__irm_result_des(&msg, &flow, &req); + if (err < 0) + return err; + + clock_gettime(PTHREAD_COND_CLOCK, &t0); + + err = poa_alloc(addr, flow.qs, &req, &resp, &pf, &mtu, timeo); + + freebuf(req); + + if (err < 0) + goto fail_alloc; + + clock_gettime(PTHREAD_COND_CLOCK, &t1); + + flow.mtu = mtu; + flow.mpl = pf->poa->mpl; + msg.len = SOCK_BUF_SIZE; + msg.data = buf; + if (poa_flow_alloc_r__irm_req_ser(&msg, &flow, &resp, 0) < 0) { + err = -ENOMEM; + goto fail_resp; + } + + freebuf(resp); + + err = send_recv_msg(&msg); + if (err < 0) + goto fail_msg; + + crypt.key = key; + crypt.epoch = 0; + crypt.role = CRYPT_ROLE_INIT; + + err = flow__irm_result_des(&msg, &flow, &crypt); + if (err < 0) + goto fail_msg; + + err = flow_init(&flow, &crypt, ts_diff_ns(&t1, &t0), pf); + + crypt_secure_clear(key, SYMMKEYSZ); + + if (err < 0) + goto fail_msg; + + if (qs != NULL) + *qs = flow.qs; + + return err; + + fail_resp: + freebuf(resp); + fail_msg: + poa_alloc_fail(pf); + return err; + fail_alloc: + msg.len = SOCK_BUF_SIZE; + msg.data = buf; + if (poa_flow_alloc_r__irm_req_ser(&msg, &flow, NULL, err) == 0) + send_recv_msg(&msg); + + return err; +} + diff --git a/src/lib/poa/poa.h b/src/lib/poa/poa.h new file mode 100644 index 00000000..014986a3 --- /dev/null +++ b/src/lib/poa/poa.h @@ -0,0 +1,364 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Points of attachment (PoA) - internal API + * + * 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_LIB_POA_POA_H +#define OUROBOROS_LIB_POA_POA_H + +#include <ouroboros/atomics.h> +#include <ouroboros/ipcp-dev.h> +#include <ouroboros/list.h> +#include <ouroboros/qos.h> +#include <ouroboros/rcu.h> +#include <ouroboros/ssm_flow_set.h> +#include <ouroboros/ssm_pool.h> +#include <ouroboros/ssm_rbuff.h> +#include <ouroboros/time.h> +#include <ouroboros/utils.h> + +#include <errno.h> +#include <limits.h> +#include <poll.h> +#include <pthread.h> +#include <stdbool.h> +#include <stdint.h> + +#define POA_MGMT_EID 0 /* reserved for the mgmt channel */ + +#define POA_FLOW_REQ 1 +#define POA_FLOW_REPLY 2 +#define POA_FLOW_UPDATE 3 +#define POA_NAME_QUERY 4 +#define POA_NAME_REPLY 5 + +#define POA_QUERY_HLEN 32 /* SHA3-256, the query hash algorithm */ + +/* Fits "udp6.<ip6>.<port>", the longest display/RIB entry name. */ +#define POA_NAME_STRLEN 63 + +/* Management message; every transport frames it the same way. */ +struct poa_mgmt_msg { + uint8_t code; + uint8_t resv[3]; + uint32_t s_eid; + uint32_t d_eid; + int32_t response; + uint64_t bandwidth; + uint32_t delay; + uint32_t loss; + uint32_t ber; + uint32_t max_gap; + uint32_t timeout; + uint8_t availability; + uint8_t service; + uint16_t data_len; +} __attribute__((packed)); + +struct poa; +struct poa_flow; + +#ifdef PROC_FLOW_STATS +#define POA_STAT_BUMP(poa, field) FETCH_ADD_RELAXED(&(poa)->stat.field, 1) +#define POA_STAT_ADD(poa, field, v) FETCH_ADD_RELAXED(&(poa)->stat.field, (v)) +#define POA_STAT_SUB(poa, field, v) FETCH_SUB_RELAXED(&(poa)->stat.field, (v)) +#define POA_STAT_LOAD(poa, field) LOAD_RELAXED(&(poa)->stat.field) +#else +#define POA_STAT_BUMP(poa, field) ((void) (poa)) +#define POA_STAT_ADD(poa, field, v) ((void) (poa), (void) (v)) +#define POA_STAT_SUB(poa, field, v) ((void) (poa), (void) (v)) +#define POA_STAT_LOAD(poa, field) ((void) (poa), (size_t) 0) +#endif + +struct poa_stat { + size_t n_flows; /* gauge; the RIB reads it without poas.lock */ + size_t rx_pkts; /* packets delivered to a flow */ + size_t rx_bytes; /* payload bytes delivered */ + size_t tx_pkts; /* packets handed to the transport */ + size_t tx_bytes; /* payload bytes handed to the transport */ + size_t mgmt_rx; /* frames queued for the mgmt handler */ + size_t mgmt_tx; /* management frames sent */ + size_t bad_eid; /* no flow on the EID a peer sent */ + size_t dlv_fail; /* the rx ring above would not take it */ + size_t buf_fail; /* no buffer to receive into */ + size_t rcv_fail; /* transport read failed; the reader exits */ + size_t snd_fail; /* transport send failed */ + size_t qry_tx; /* name queries broadcast */ + size_t qry_rx; /* name queries received */ + size_t rep_tx; /* name replies sent, query matched */ + size_t rep_rx; /* name replies received */ +}; + +/* Spacing between transmit-depth samples; a depth costs a syscall. */ +#define POA_QLEN_GATE (100 * 1000) /* ns */ + + +/* Transport operations; public poa_X() dispatches to ops->poa_X. */ +struct poa_ops { + /* Parse own arm of the spec; validate; fill local and priv. */ + int (* poa_attach)(struct poa * poa, + const struct poa_spec * spec); + + void (* poa_detach)(struct poa * poa); + + int (* poa_start)(struct poa * poa); + + void (* poa_stop)(struct poa * poa); + + /* Full queue: -EAGAIN unless block; then wait, to abstime if set. */ + int (* poa_send)(struct poa * poa, + const struct poa_addr * dst, + uint32_t eid, + struct ssm_pk_buff * spb, + bool block, + const struct timespec * abstime); + + int (* poa_send_mgmt)(struct poa * poa, + const struct poa_addr * dst, + const uint8_t * buf, + size_t len); + + int (* poa_query)(const char * dst, + const struct timespec * timeo, + struct poa_addr * addr); + + uint32_t (* poa_mtu)(struct poa * poa, + const struct poa_addr * dst); + + /* Bytes queued in the transmit path of the PoA. */ + size_t (* poa_qlen)(struct poa * poa); + + /* Depth from the queue itself; NULL infers it from qlen. */ + int (* poa_qpkts)(struct poa * poa, + size_t * pkts, + size_t * byts); + + int (* poa_rib)(struct poa * poa, + char * buf, + size_t len); + + /* Identity as a spec, e.g. for poa_list. */ + void (* poa_spec)(const struct poa * poa, + struct poa_spec * spec); + + /* Same identity as spec? Caller matched poa->type already. */ + bool (* poa_has_id)(const struct poa * poa, + const struct poa_spec * spec); + + /* Carries dst? Caller matched poa->type already. */ + bool (* poa_match)(const struct poa * poa, + const struct poa_addr * dst); + + /* + * Flows ride the link this id names; NULL: no link events. + * Ids are meaningful only to the backend whose monitor + * produced them; a single backend owns the monitor. + */ + bool (* poa_link_match)(const struct poa * poa, + int id); + + /* Maximum packet lifetime in the transport, seconds. */ + time_t mpl; +}; +struct poa { + struct list_head next; + + enum poa_type type; + const struct poa_ops * ops; + void * priv; + + struct poa_addr local; /* what peers dial us on */ + + /* Display/RIB entry name for local, e.g. "udp4.<ip>.<port>". */ + char name[POA_NAME_STRLEN + 1]; + + time_t mpl; + + /* Identifies the transmit queue the flows on this PoA share. */ + int qid; + + /* Mean sent packet size (bytes), EWMA over the send path. */ + size_t avg_len; + /* Cost of one packet in the queue, in the transport's terms. */ + size_t avg_ovh; + size_t n_tx; + + /* Last queue depth read, and when, in the transport's terms. */ + size_t q_cache; + uint64_t q_time; + + /* Queued management frames, capped; poas.mgmt_mtx guards. */ + size_t n_mgmt; + +#ifdef PROC_FLOW_STATS + struct poa_stat stat; +#endif + + struct bmp * eids; + struct poa_flow ** eid_to_pf; + size_t n_eids; + + struct list_head flows; /* live flows, for repeats */ + + /* Keeps a flow and its ring alive under the receive path. */ + struct rcu_guard guard; +}; + +/* poa/poa.c is part of the dev.c translation unit. */ +int poa_init(const char * name); + +int poa_start(void); + +void poa_stop(void); + +void poa_fini(void); + +/* Also answer name queries for the layer once enrolled. */ +int poa_set_layer(const char * layer); + +/* Hash of a name this process answers queries for? */ +bool poa_has_name(const uint8_t * hash); + +int poa_flow_tx(struct poa_flow * pf, + struct ssm_pk_buff * spb, + bool block, + const struct timespec * abstime); + +size_t poa_flow_qlen(const struct poa_flow * pf); + +size_t poa_flow_qpkts(const struct poa_flow * pf); + +int poa_flow_qid(const struct poa_flow * pf); + +size_t poa_flow_mean_len(const struct poa_flow * pf); + +void poa_flow_attach(struct poa_flow * pf, + int flow_id, + struct ssm_rbuff * rx_rb); + +void poa_flow_ready(struct poa_flow * pf); + +void poa_flow_detach(struct poa_flow * pf); + +struct poa_flow * poa_flow_take_pending(int flow_id); + +/* Addresses and management messages (poa/addr.c). */ +int poa_addr_cmp(const struct poa_addr * a, + const struct poa_addr * b); + +/* Display/RIB entry name, e.g. "udp4.<ip>.<port>". */ +int poa_addr_name(const struct poa_addr * a, + char * buf, + size_t len); + +void poa_mgmt_msg_ser(struct poa_mgmt_msg * msg, + uint8_t code, + uint32_t s_eid, + uint32_t d_eid, + qosspec_t qs, + int response, + size_t data_len); + +void poa_mgmt_msg_qos(const struct poa_mgmt_msg * msg, + qosspec_t * qs); + +/* Called by the transports. */ +void poa_rx_pkt(struct poa * poa, + uint32_t eid, + struct ssm_pk_buff * spb); + +void poa_rx_mgmt(struct poa * poa, + const struct poa_addr * src, + const uint8_t * buf, + size_t len); + +/* Reserve a buffer for a received packet, with transport headroom. */ +int poa_spb_reserve(struct ssm_pk_buff ** spb, + size_t len); + +void poa_spb_release(struct ssm_pk_buff * spb); + +/* + * All flows on PoAs whose poa_link_match reports this link id go up + * or down with it. Returns the number of flows whose state changed. + */ +size_t poa_link_updown(int id, + bool up); + +/* + * Link monitor: one socket for the whole subsystem, opened by + * poa_start(). Returns -1 where the transport has no monitor. + */ +int poa_monitor_open(void); + +/* Reads one batch of link events; cancellation point. */ +void poa_monitor_read(int fd); + +/* Broadcast a mgmt frame on every PoA matching dst; # sent. */ +int poa_bcast_mgmt(const struct poa_addr * dst, + const uint8_t * buf, + size_t len); + +/* Transport op tables. */ +extern const struct poa_ops udp_poa_ops; +extern const struct poa_ops eth_poa_ops; + +/* + * Waits for a descriptor to take another packet, up to abstime. + * A NULL deadline waits indefinitely. Transports call this when + * their send reports the transmit queue full. + */ +static __inline__ int poa_wait_out(int fd, + const struct timespec * abstime) +{ + struct pollfd pfd; + struct timespec now; + long ms = -1; + bool clamped = false; + int ret; + + if (abstime != NULL) { + clock_gettime(PTHREAD_COND_CLOCK, &now); + + if (ts_diff_ns(abstime, &now) <= 0) + return -ETIMEDOUT; + + ms = ts_diff_ms(abstime, &now) + 1; /* sub-ms must wait */ + if (ms > INT_MAX) { /* poll takes an int */ + ms = INT_MAX; + clamped = true; + } + } + + pfd.fd = fd; + pfd.events = POLLOUT; + pfd.revents = 0; + + ret = poll(&pfd, 1, (int) ms); + if (ret < 0) + return errno == EINTR ? 0 : -EIO; + + if (ret == 0) + return clamped ? 0 : -ETIMEDOUT; /* clamped: retry */ + + return 0; +} + +#endif /* OUROBOROS_LIB_POA_POA_H */ diff --git a/src/lib/poa/udp.c b/src/lib/poa/udp.c new file mode 100644 index 00000000..6753347a --- /dev/null +++ b/src/lib/poa/udp.c @@ -0,0 +1,633 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Points of attachment (PoA) - UDP transport + * + * 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/. + */ + +#if defined(__APPLE__) +#define _BSD_SOURCE +#define _DARWIN_C_SOURCE +#elif defined(__FreeBSD__) +#define __BSD_VISIBLE 1 +#elif defined(__linux__) || defined(__CYGWIN__) +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE +#endif +#else +#ifndef _POSIX_C_SOURCE +#define _POSIX_C_SOURCE 200809L +#endif +#endif + +#include "config.h" + +#define OUROBOROS_PREFIX "poa-udp" + +#include <ouroboros/endian.h> +#include <ouroboros/errno.h> +#include <ouroboros/logs.h> +#include <ouroboros/time.h> +#include <ouroboros/utils.h> + +#include "poa.h" + +#ifdef __linux__ +#include <linux/sockios.h> +#endif +#include <arpa/inet.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/uio.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#define UDP_HDR_LEN sizeof(uint32_t) /* PoA id */ +#define UDP_MAX_PAYLOAD (POA_UDP_RD_BUF - UDP_HDR_LEN) +/* The reader buffer must fit a full mgmt frame at any tuning. */ +#define UDP_MAX_PACKET MAX(POA_UDP_RD_BUF, POA_MGMT_FRAME_SIZE + UDP_HDR_LEN) +#define UDP_IP4_OVERH 28U /* IPv4 + UDP */ +#define UDP_IP6_OVERH 48U /* IPv6 + UDP */ +/* Wait for the link to come back before reading it again. */ +#define UDP_DOWN_TIMEO 100 /* ms */ + +union udp_saddr { + struct sockaddr sa; + struct sockaddr_in in; + struct sockaddr_in6 in6; +}; + +struct udp_priv { + int s_fd; + int af; + union udp_saddr s_saddr; + pthread_t reader; + bool running; +}; + +static socklen_t saddr_len(int af) +{ + if (af == AF_INET) + return sizeof(struct sockaddr_in); + + return sizeof(struct sockaddr_in6); +} + +static void addr_to_saddr(const struct poa_addr * addr, + union udp_saddr * saddr) +{ + memset(saddr, 0, sizeof(*saddr)); + + if (addr->type == POA_UDP4) { + saddr->in.sin_family = AF_INET; + saddr->in.sin_addr = addr->udp4.ip_addr; + saddr->in.sin_port = htons(addr->udp4.port); + } else { + saddr->in6.sin6_family = AF_INET6; + saddr->in6.sin6_addr = addr->udp6.ip_addr; + saddr->in6.sin6_port = htons(addr->udp6.port); + } +} + +static void saddr_to_addr(const union udp_saddr * saddr, + struct poa_addr * addr) +{ + memset(addr, 0, sizeof(*addr)); + + if (saddr->sa.sa_family == AF_INET) { + addr->type = POA_UDP4; + addr->udp4.ip_addr = saddr->in.sin_addr; + addr->udp4.port = ntohs(saddr->in.sin_port); + } else { + addr->type = POA_UDP6; + addr->udp6.ip_addr = saddr->in6.sin6_addr; + addr->udp6.port = ntohs(saddr->in6.sin6_port); + } +} + +/* A datagram longer than the buffer arrives truncated: drop it. */ +static void * udp_reader(void * o) +{ + struct poa * poa = (struct poa *) o; + struct udp_priv * priv = (struct udp_priv *) poa->priv; + struct timespec down = TIMESPEC_INIT_MS(UDP_DOWN_TIMEO); + uint8_t * buf; + + buf = malloc(UDP_MAX_PACKET); + if (buf == NULL) + return (void *) -1; + + pthread_cleanup_push(free, buf); + + while (true) { + struct ssm_pk_buff * spb; + union udp_saddr r_saddr; + struct poa_addr src; + struct msghdr mh; + struct iovec iov; + ssize_t n; + uint32_t eid; + size_t plen; + + iov.iov_base = buf; + iov.iov_len = UDP_MAX_PACKET; + + memset(&mh, 0, sizeof(mh)); + + mh.msg_name = &r_saddr; + mh.msg_namelen = sizeof(r_saddr); + mh.msg_iov = &iov; + mh.msg_iovlen = 1; + + n = recvmsg(priv->s_fd, &mh, 0); + if (n < 0) { + if (errno == EINTR) + continue; + + POA_STAT_BUMP(poa, rcv_fail); + + if (errno == ENETDOWN) { + nanosleep(&down, NULL); + continue; + } + + log_err("Reader stopped: %s.", strerror(errno)); + break; + } + + if ((mh.msg_flags & MSG_TRUNC) != 0) { + POA_STAT_BUMP(poa, buf_fail); + continue; + } + + if ((size_t) n < UDP_HDR_LEN) + continue; + + eid = ntoh32(*(uint32_t *) buf); + plen = (size_t) n - UDP_HDR_LEN; + + saddr_to_addr(&r_saddr, &src); + + if (eid == POA_MGMT_EID) { + poa_rx_mgmt(poa, &src, buf + UDP_HDR_LEN, plen); + continue; + } + + if (poa_spb_reserve(&spb, plen) < 0) { + POA_STAT_BUMP(poa, buf_fail); + continue; + } + + memcpy(ssm_pk_buff_head(spb), buf + UDP_HDR_LEN, plen); + + poa_rx_pkt(poa, eid, spb); + } + + pthread_cleanup_pop(true); + + return (void *) 0; +} + +/* Reads the bound address back: an ephemeral port is only known after. */ +static int udp_attach(struct poa * poa, + const struct poa_spec * spec) +{ + struct udp_priv * priv; + socklen_t len; + int af; + + af = spec->type == POA_UDP4 ? AF_INET : AF_INET6; + + priv = malloc(sizeof(*priv)); + if (priv == NULL) + return -ENOMEM; + + memset(priv, 0, sizeof(*priv)); + + priv->af = af; + priv->s_fd = socket(af, SOCK_DGRAM, IPPROTO_UDP); + if (priv->s_fd < 0) { + log_err("Failed to create socket: %s.", strerror(errno)); + goto fail_socket; + } + + if (af == AF_INET) { + priv->s_saddr.in.sin_family = AF_INET; + priv->s_saddr.in.sin_addr = spec->udp4.ip_addr; + priv->s_saddr.in.sin_port = htons(spec->udp4.port); + } else { + int on = 1; + + if (setsockopt(priv->s_fd, IPPROTO_IPV6, IPV6_V6ONLY, + &on, sizeof(on)) < 0) { + log_err("Failed to set IPV6_V6ONLY: %s.", + strerror(errno)); + goto fail_bind; + } + + priv->s_saddr.in6.sin6_family = AF_INET6; + priv->s_saddr.in6.sin6_addr = spec->udp6.ip_addr; + priv->s_saddr.in6.sin6_port = htons(spec->udp6.port); + } + + if (bind(priv->s_fd, &priv->s_saddr.sa, saddr_len(af)) < 0) { + log_err("Failed to bind: %s.", strerror(errno)); + goto fail_bind; + } + + poa->priv = priv; + len = saddr_len(af); + if (getsockname(priv->s_fd, &priv->s_saddr.sa, &len) < 0) + log_warn("Failed to read the bound address: %s.", + strerror(errno)); + + saddr_to_addr(&priv->s_saddr, &poa->local); + + return 0; + + fail_bind: + close(priv->s_fd); + fail_socket: + poa->priv = NULL; + + free(priv); + + return -EIO; +} + +static void udp_detach(struct poa * poa) +{ + struct udp_priv * priv = (struct udp_priv *) poa->priv; + + if (priv == NULL) + return; + + close(priv->s_fd); + + free(priv); + + poa->priv = NULL; +} + +/* One reader per socket, so a flow cannot be reordered on receive. */ +static int udp_start(struct poa * poa) +{ + struct udp_priv * priv = (struct udp_priv *) poa->priv; + + if (pthread_create(&priv->reader, NULL, udp_reader, poa) != 0) + return -1; + + priv->running = true; + + return 0; +} + +static void udp_stop(struct poa * poa) +{ + struct udp_priv * priv = (struct udp_priv *) poa->priv; + + if (!priv->running) + return; + + pthread_cancel(priv->reader); + pthread_join(priv->reader, NULL); + + priv->running = false; +} + +/* + * The PoA id is a separate iovec, so the buffer needs no headroom. + * MSG_DONTWAIT: the reader blocks on this socket. + */ +static int udp_sendv(struct udp_priv * priv, + const struct poa_addr * dst, + uint32_t eid, + const uint8_t * body, + size_t len, + bool block, + const struct timespec * abstime) +{ + union udp_saddr saddr; + struct msghdr msg; + struct iovec iov[2]; + uint32_t hdr; + int ret; + + if (len > UDP_MAX_PAYLOAD) + return -EMSGSIZE; + + addr_to_saddr(dst, &saddr); + + hdr = hton32(eid); + + iov[0].iov_base = &hdr; + iov[0].iov_len = sizeof(hdr); + iov[1].iov_base = (void *) body; + iov[1].iov_len = len; + + memset(&msg, 0, sizeof(msg)); + + msg.msg_name = &saddr; + msg.msg_namelen = saddr_len(priv->af); + msg.msg_iov = iov; + msg.msg_iovlen = len > 0 ? 2 : 1; + while (sendmsg(priv->s_fd, &msg, MSG_DONTWAIT) < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) + return -EIO; + + if (!block) + return -EAGAIN; + + ret = poa_wait_out(priv->s_fd, abstime); + if (ret < 0) + return ret; + } + + return 0; +} + +static int udp_send(struct poa * poa, + const struct poa_addr * dst, + uint32_t eid, + struct ssm_pk_buff * spb, + bool block, + const struct timespec * abstime) +{ + return udp_sendv((struct udp_priv *) poa->priv, dst, eid, + ssm_pk_buff_head(spb), ssm_pk_buff_len(spb), + block, abstime); +} + +static int udp_send_mgmt(struct poa * poa, + const struct poa_addr * dst, + const uint8_t * buf, + size_t len) +{ + struct timespec timeo = TIMESPEC_INIT_MS(POA_MGMT_SND_TIMEO); + struct timespec abstime; + + clock_gettime(PTHREAD_COND_CLOCK, &abstime); + ts_add(&abstime, &timeo, &abstime); + + return udp_sendv((struct udp_priv *) poa->priv, dst, POA_MGMT_EID, + buf, len, true, &abstime); +} + +/* The PoA id header eats into the usable MTU. */ +static uint32_t udp_mtu(struct poa * poa, + const struct poa_addr * dst) +{ + struct udp_priv * priv = (struct udp_priv *) poa->priv; + uint32_t fallback; + uint32_t overh; +#if defined(__linux__) && (defined(IP_MTU) || defined(IPV6_MTU)) + union udp_saddr saddr; + socklen_t len; + int sock; + int mtu = 0; +#endif + if (priv->af == AF_INET) { + fallback = POA_UDP4_MTU; + overh = UDP_IP4_OVERH; + } else { + fallback = POA_UDP6_MTU; + overh = UDP_IP6_OVERH; + } + + fallback -= UDP_HDR_LEN; + if (fallback > UDP_MAX_PAYLOAD) + fallback = UDP_MAX_PAYLOAD; + +#if defined(__linux__) && (defined(IP_MTU) || defined(IPV6_MTU)) + + addr_to_saddr(dst, &saddr); + + sock = socket(priv->af, SOCK_DGRAM, IPPROTO_UDP); + if (sock < 0) + return fallback; + + if (connect(sock, &saddr.sa, saddr_len(priv->af)) < 0) + goto fallback; + + len = sizeof(mtu); + +#if defined(IP_MTU) + if (priv->af == AF_INET) { + if (getsockopt(sock, IPPROTO_IP, IP_MTU, &mtu, &len) < 0) + goto fallback; + } +#endif +#if defined(IPV6_MTU) + if (priv->af == AF_INET6) { + if (getsockopt(sock, IPPROTO_IPV6, IPV6_MTU, &mtu, &len) < 0) + goto fallback; + } +#endif + close(sock); + + if (mtu <= (int) (overh + UDP_HDR_LEN)) + return fallback; + + return MIN((uint32_t) mtu - overh - UDP_HDR_LEN, UDP_MAX_PAYLOAD); + + fallback: + close(sock); +#else + (void) dst; + (void) overh; +#endif + return fallback; +} + +/* All flows on the PoA share the socket, so this is aggregate. */ +static size_t udp_qlen(struct poa * poa) +{ +#if defined(__linux__) && defined(SIOCOUTQ) + struct udp_priv * priv = (struct udp_priv *) poa->priv; + int qlen; + + qlen = 0; + if (ioctl(priv->s_fd, SIOCOUTQ, &qlen) < 0) + return 0; + + return (size_t) qlen; +#else + (void) poa; + + return 0; +#endif +} + +/* The kernel keeps no per-socket drop count for UDP. */ +static int udp_rib(struct poa * poa, + char * buf, + size_t len) +{ + struct udp_priv * priv = (struct udp_priv *) poa->priv; + socklen_t optlen; + size_t sndbuf = 0; + size_t rcvbuf = 0; + int val; + int size; + + optlen = sizeof(val); + if (getsockopt(priv->s_fd, SOL_SOCKET, SO_SNDBUF, &val, &optlen) == 0) + sndbuf = (size_t) val; + + optlen = sizeof(val); + if (getsockopt(priv->s_fd, SOL_SOCKET, SO_RCVBUF, &val, &optlen) == 0) + rcvbuf = (size_t) val; + + size = snprintf(buf, len, + "Socket sndbuf (bytes): %zu\n" + "Socket rcvbuf (bytes): %zu\n", + sndbuf, rcvbuf); + if (size < 0 || (size_t) size >= len) + return -1; + + return size; +} + +/* + * Asks the kernel which address it would send from: connect() does the + * real route lookup and sends nothing, so this honours the default + * route, metrics and policy rules alike. + */ +static int udp_src_addr(const struct poa_addr * dst, + struct poa_addr * src) +{ + union udp_saddr saddr; + socklen_t len; + int af; + int fd; + + if (dst->type != POA_UDP4 && dst->type != POA_UDP6) + return -EINVAL; + + af = dst->type == POA_UDP4 ? AF_INET : AF_INET6; + + addr_to_saddr(dst, &saddr); + + fd = socket(af, SOCK_DGRAM, IPPROTO_UDP); + if (fd < 0) + return -EIO; + + if (connect(fd, &saddr.sa, saddr_len(af)) < 0) + goto fail; + + len = saddr_len(af); + if (getsockname(fd, &saddr.sa, &len) < 0) + goto fail; + + close(fd); + + saddr_to_addr(&saddr, src); + + return 0; + + fail: + close(fd); + + return -EIO; +} + +static void udp_spec(const struct poa * poa, + struct poa_spec * spec) +{ + spec->type = poa->type; + + if (poa->type == POA_UDP4) + spec->udp4 = poa->local.udp4; + else + spec->udp6 = poa->local.udp6; +} + +static bool udp_has_id(const struct poa * poa, + const struct poa_spec * spec) +{ + if (poa->type == POA_UDP4) { + if (poa->local.udp4.port != spec->udp4.port) + return false; + + return memcmp(&poa->local.udp4.ip_addr, + &spec->udp4.ip_addr, + sizeof(spec->udp4.ip_addr)) == 0; + } + + if (poa->local.udp6.port != spec->udp6.port) + return false; + + return memcmp(&poa->local.udp6.ip_addr, &spec->udp6.ip_addr, + sizeof(spec->udp6.ip_addr)) == 0; +} + +static bool udp_addr_is_any(const struct poa_addr * addr) +{ + static const struct in6_addr any6 = IN6ADDR_ANY_INIT; + + if (addr->type == POA_UDP4) + return addr->udp4.ip_addr.s_addr == htonl(INADDR_ANY); + + return memcmp(&addr->udp6.ip_addr, &any6, sizeof(any6)) == 0; +} + +/* + * Our end of the flow: the IP the kernel would send to dst from. A + * PoA bound to the wildcard is the catch-all and a failed probe + * matches any. Ports are not compared: the probe's is ephemeral. + */ +static bool udp_match(const struct poa * poa, + const struct poa_addr * dst) +{ + struct poa_addr src; + + if (udp_addr_is_any(&poa->local)) + return true; + + if (udp_src_addr(dst, &src) < 0) + return true; + + if (poa->type == POA_UDP4) + return memcmp(&poa->local.udp4.ip_addr, + &src.udp4.ip_addr, + sizeof(src.udp4.ip_addr)) == 0; + + return memcmp(&poa->local.udp6.ip_addr, &src.udp6.ip_addr, + sizeof(src.udp6.ip_addr)) == 0; +} + +const struct poa_ops udp_poa_ops = { + .poa_attach = udp_attach, + .poa_detach = udp_detach, + .poa_start = udp_start, + .poa_stop = udp_stop, + .poa_send = udp_send, + .poa_send_mgmt = udp_send_mgmt, + .poa_mtu = udp_mtu, + .poa_qlen = udp_qlen, + .poa_rib = udp_rib, + .poa_spec = udp_spec, + .poa_has_id = udp_has_id, + .poa_match = udp_match, + .mpl = POA_UDP_MPL +}; diff --git a/src/lib/protobuf.c b/src/lib/protobuf.c index 6beae000..aa247755 100644 --- a/src/lib/protobuf.c +++ b/src/lib/protobuf.c @@ -22,6 +22,7 @@ #define _DEFAULT_SOURCE +#include <ouroboros/errno.h> #include <ouroboros/protobuf.h> #include <ouroboros/crypt.h> #include <ouroboros/proc.h> @@ -215,6 +216,166 @@ struct layer_info layer_info_msg_to_s(const layer_info_msg_t * msg) return s; } + +static bool mac_is_zero(const uint8_t * mac) +{ + static const uint8_t zero[POA_MAC_SIZE] = { 0 }; + + return memcmp(mac, zero, POA_MAC_SIZE) == 0; +} + +poa_addr_msg_t * poa_addr_s_to_msg(const struct poa_addr * s) +{ + poa_addr_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + poa_addr_msg__init(msg); + + msg->type = s->type; + + if (s->hostname[0] != '\0') { + msg->hostname = strdup(s->hostname); + if (msg->hostname == NULL) + goto fail_msg; + } + + switch (s->type) { + case POA_UDP4: + msg->has_ip4 = true; + msg->ip4 = s->udp4.ip_addr.s_addr; + msg->has_port = true; + msg->port = s->udp4.port; + break; + case POA_UDP6: + msg->has_ip6 = true; + msg->ip6.len = sizeof(s->udp6.ip_addr); + msg->ip6.data = malloc(msg->ip6.len); + if (msg->ip6.data == NULL) + goto fail_msg; + memcpy(msg->ip6.data, &s->udp6.ip_addr, msg->ip6.len); + + msg->has_port = true; + msg->port = s->udp6.port; + break; + case POA_ETH: + msg->has_ethertype = true; + msg->ethertype = s->eth.dst.ethertype; + msg->has_mac = true; + msg->mac.len = POA_MAC_SIZE; + msg->mac.data = malloc(POA_MAC_SIZE); + if (msg->mac.data == NULL) + goto fail_msg; + memcpy(msg->mac.data, s->eth.dst.mac, POA_MAC_SIZE); + + if (!mac_is_zero(s->eth.src.mac)) { + msg->has_src_mac = true; + msg->src_mac.len = POA_MAC_SIZE; + msg->src_mac.data = malloc(POA_MAC_SIZE); + if (msg->src_mac.data == NULL) + goto fail_msg; + memcpy(msg->src_mac.data, s->eth.src.mac, POA_MAC_SIZE); + } + + if (s->eth.src.dev[0] != '\0') { + msg->dev = strdup(s->eth.src.dev); + if (msg->dev == NULL) + goto fail_msg; + } + break; + case POA_UDP: + msg->has_port = true; + msg->port = s->udp4.port; + break; + default: + goto fail_msg; + } + + return msg; + + fail_msg: + poa_addr_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +struct poa_addr poa_addr_msg_to_s(const poa_addr_msg_t * msg) +{ + struct poa_addr s; + + assert(msg != NULL); + + memset(&s, 0, sizeof(s)); + + s.type = POA_INVALID; + + if (msg->hostname != NULL) { + if (strlen(msg->hostname) > POA_HOST_STRLEN) + return s; + strcpy(s.hostname, msg->hostname); + } + + switch (msg->type) { + case POA_UDP4: + if (!msg->has_ip4 || !msg->has_port) + return s; + + s.udp4.ip_addr.s_addr = msg->ip4; + s.udp4.port = msg->port; + break; + case POA_UDP6: + if (!msg->has_ip6 || !msg->has_port) + return s; + + if (msg->ip6.len != sizeof(s.udp6.ip_addr) + || msg->ip6.data == NULL) + return s; + + memcpy(&s.udp6.ip_addr, msg->ip6.data, msg->ip6.len); + + s.udp6.port = msg->port; + break; + case POA_ETH: + if (!msg->has_ethertype) + return s; + + s.eth.src.ethertype = msg->ethertype; + s.eth.dst.ethertype = msg->ethertype; + + if (msg->mac.len != POA_MAC_SIZE || msg->mac.data == NULL) + return s; + + memcpy(s.eth.dst.mac, msg->mac.data, POA_MAC_SIZE); + + if (msg->src_mac.len == POA_MAC_SIZE) + memcpy(s.eth.src.mac, msg->src_mac.data, POA_MAC_SIZE); + + if (msg->dev != NULL) { + if (strlen(msg->dev) > DEV_NAME_SIZE) + return s; + strcpy(s.eth.src.dev, msg->dev); + } + break; + case POA_UDP: + if (msg->hostname == NULL || msg->hostname[0] == '\0' + || !msg->has_port) + return s; + + s.udp4.port = msg->port; + break; + default: + return s; + } + + s.type = msg->type; + + return s; +} + ipcp_info_msg_t * ipcp_info_s_to_msg(const struct ipcp_info * s) { ipcp_info_msg_t * msg; @@ -389,22 +550,6 @@ struct dt_config dt_config_msg_to_s(const dt_config_msg_t * msg) return s; } -struct dir_dht_config dir_dht_config_msg_to_s(const dir_dht_config_msg_t * msg) -{ - struct dir_dht_config s; - - assert(msg != NULL); - - s.params.alpha = msg->alpha; - s.params.k = msg->k; - s.params.t_expire = msg->t_expire; - s.params.t_refresh = msg->t_refresh; - s.params.t_replicate = msg->t_replicate; - s.peer = msg->peer; - - return s; -} - dir_dht_config_msg_t * dir_dht_config_s_to_msg(const struct dir_dht_config * s) { dir_dht_config_msg_t * msg; @@ -427,22 +572,18 @@ dir_dht_config_msg_t * dir_dht_config_s_to_msg(const struct dir_dht_config * s) return msg; } -struct dir_config dir_config_msg_to_s(const dir_config_msg_t * msg) +struct dir_dht_config dir_dht_config_msg_to_s(const dir_dht_config_msg_t * msg) { - struct dir_config s; + struct dir_dht_config s; assert(msg != NULL); - switch (msg->pol) { - case DIR_DHT: - s.dht = dir_dht_config_msg_to_s(msg->dht); - break; - default: - /* No checks here */ - break; - } - - s.pol = msg->pol; + s.params.alpha = msg->alpha; + s.params.k = msg->k; + s.params.t_expire = msg->t_expire; + s.params.t_refresh = msg->t_refresh; + s.params.t_replicate = msg->t_replicate; + s.peer = msg->peer; return s; } @@ -479,6 +620,25 @@ dir_config_msg_t * dir_config_s_to_msg(const struct dir_config * s) return NULL; } +struct dir_config dir_config_msg_to_s(const dir_config_msg_t * msg) +{ + struct dir_config s; + + assert(msg != NULL); + + switch (msg->pol) { + case DIR_DHT: + s.dht = dir_dht_config_msg_to_s(msg->dht); + break; + default: + /* No checks here */ + break; + } + + s.pol = msg->pol; + + return s; +} uni_config_msg_t * uni_config_s_to_msg(const struct uni_config * s) { @@ -525,9 +685,9 @@ struct uni_config uni_config_msg_to_s(const uni_config_msg_t * msg) return s; } -udp4_config_msg_t * udp4_config_s_to_msg(const struct udp4_config * s) +udp4_poa_msg_t * udp4_poa_s_to_msg(const struct udp4_poa * s) { - udp4_config_msg_t * msg; + udp4_poa_msg_t * msg; assert(s != NULL); @@ -535,32 +695,30 @@ udp4_config_msg_t * udp4_config_s_to_msg(const struct udp4_config * s) if (msg == NULL) return NULL; - udp4_config_msg__init(msg); + udp4_poa_msg__init(msg); - msg->ip_addr = s->ip_addr.s_addr; - msg->dns_addr = s->dns_addr.s_addr; - msg->port = s->port; + msg->ip_addr = s->ip_addr.s_addr; + msg->port = s->port; return msg; } -struct udp4_config udp4_config_msg_to_s(const udp4_config_msg_t * msg) +struct udp4_poa udp4_poa_msg_to_s(const udp4_poa_msg_t * msg) { - struct udp4_config s; + struct udp4_poa s; assert(msg != NULL); - s.ip_addr.s_addr = msg->ip_addr; - s.dns_addr.s_addr = msg->dns_addr; - s.port = msg->port; + s.ip_addr.s_addr = msg->ip_addr; + s.port = msg->port; return s; } #define IN6_LEN (size_t) sizeof(struct in6_addr) -udp6_config_msg_t * udp6_config_s_to_msg(const struct udp6_config * s) +udp6_poa_msg_t * udp6_poa_s_to_msg(const struct udp6_poa * s) { - udp6_config_msg_t * msg; + udp6_poa_msg_t * msg; assert(s != NULL); @@ -568,7 +726,7 @@ udp6_config_msg_t * udp6_config_s_to_msg(const struct udp6_config * s) if (msg == NULL) goto fail_malloc; - udp6_config_msg__init(msg); + udp6_poa_msg__init(msg); msg->ip_addr.data = malloc(IN6_LEN); if (msg->ip_addr.data == NULL) @@ -577,42 +735,33 @@ udp6_config_msg_t * udp6_config_s_to_msg(const struct udp6_config * s) msg->ip_addr.len = IN6_LEN; memcpy(msg->ip_addr.data, &s->ip_addr.s6_addr, IN6_LEN); - msg->dns_addr.data = malloc(IN6_LEN); - if (msg->dns_addr.data == NULL) - goto fail_msg; - - msg->dns_addr.len = IN6_LEN; - memcpy(msg->dns_addr.data, &s->dns_addr.s6_addr, IN6_LEN); - msg->port = s->port; return msg; fail_msg: - udp6_config_msg__free_unpacked(msg, NULL); + udp6_poa_msg__free_unpacked(msg, NULL); fail_malloc: return NULL; } -struct udp6_config udp6_config_msg_to_s(const udp6_config_msg_t * msg) +struct udp6_poa udp6_poa_msg_to_s(const udp6_poa_msg_t * msg) { - struct udp6_config s; + struct udp6_poa s; assert(msg != NULL); assert(msg->ip_addr.len == IN6_LEN); - assert(msg->dns_addr.len == IN6_LEN); memcpy(&s.ip_addr.s6_addr, msg->ip_addr.data, IN6_LEN); - memcpy(&s.dns_addr.s6_addr, msg->dns_addr.data, IN6_LEN); s.port = msg->port; return s; } -eth_config_msg_t * eth_config_s_to_msg(const struct eth_config * s) +eth_poa_msg_t * eth_poa_s_to_msg(const struct eth_poa * s) { - eth_config_msg_t * msg; + eth_poa_msg_t * msg; assert(s != NULL); @@ -620,7 +769,7 @@ eth_config_msg_t * eth_config_s_to_msg(const struct eth_config * s) if (msg == NULL) goto fail_malloc; - eth_config_msg__init(msg); + eth_poa_msg__init(msg); msg->dev = strdup(s->dev); if (msg->dev == NULL) @@ -628,28 +777,115 @@ eth_config_msg_t * eth_config_s_to_msg(const struct eth_config * s) msg->ethertype = s->ethertype; + msg->has_mac = true; + msg->mac.len = POA_MAC_SIZE; + msg->mac.data = malloc(POA_MAC_SIZE); + if (msg->mac.data == NULL) + goto fail_msg; + + memcpy(msg->mac.data, s->mac, POA_MAC_SIZE); + return msg; fail_msg: - eth_config_msg__free_unpacked(msg, NULL); + eth_poa_msg__free_unpacked(msg, NULL); fail_malloc: return NULL; } -struct eth_config eth_config_msg_to_s(const eth_config_msg_t * msg) +struct eth_poa eth_poa_msg_to_s(const eth_poa_msg_t * msg) { - struct eth_config s; + struct eth_poa s; assert(msg != NULL); assert(strlen(msg->dev) <= DEV_NAME_SIZE); + memset(&s, 0, sizeof(s)); + strcpy(s.dev, msg->dev); + s.ethertype = msg->ethertype; + if (msg->mac.len == POA_MAC_SIZE) + memcpy(s.mac, msg->mac.data, POA_MAC_SIZE); + return s; } +poa_spec_msg_t * poa_spec_s_to_msg(const struct poa_spec * s) +{ + poa_spec_msg_t * msg; + + assert(s != NULL); + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + poa_spec_msg__init(msg); + + switch (s->type) { + case POA_UDP4: + msg->udp4 = udp4_poa_s_to_msg(&s->udp4); + if (msg->udp4 == NULL) + goto fail_msg; + break; + case POA_UDP6: + msg->udp6 = udp6_poa_s_to_msg(&s->udp6); + if (msg->udp6 == NULL) + goto fail_msg; + break; + case POA_ETH: + msg->eth = eth_poa_s_to_msg(&s->eth); + if (msg->eth == NULL) + goto fail_msg; + break; + default: + goto fail_msg; + } + + return msg; + + fail_msg: + poa_spec_msg__free_unpacked(msg, NULL); + fail_malloc: + return NULL; +} + +struct poa_spec poa_spec_msg_to_s(const poa_spec_msg_t * msg) +{ + struct poa_spec s; + + memset(&s, 0, sizeof(s)); + + s.type = POA_INVALID; + + if (msg == NULL) + return s; + + if (msg->udp4 != NULL) { + s.type = POA_UDP4; + s.udp4 = udp4_poa_msg_to_s(msg->udp4); + } else if (msg->udp6 != NULL) { + if (msg->udp6->ip_addr.len != IN6_LEN + || msg->udp6->ip_addr.data == NULL) + return s; + + s.type = POA_UDP6; + s.udp6 = udp6_poa_msg_to_s(msg->udp6); + } else if (msg->eth != NULL) { + if (msg->eth->dev == NULL + || strlen(msg->eth->dev) > DEV_NAME_SIZE) + return s; + + s.type = POA_ETH; + s.eth = eth_poa_msg_to_s(msg->eth); + } + + return s; +} + ipcp_config_msg_t * ipcp_config_s_to_msg(const struct ipcp_config * s) { ipcp_config_msg_t * msg; @@ -672,23 +908,6 @@ ipcp_config_msg_t * ipcp_config_s_to_msg(const struct ipcp_config * s) break; case IPCP_BROADCAST: break; - case IPCP_ETH_LLC: - /* FALLTHRU */ - case IPCP_ETH_DIX: - msg->eth = eth_config_s_to_msg(&s->eth); - if (msg->eth == NULL) - goto fail_msg; - break; - case IPCP_UDP4: - msg->udp4 = udp4_config_s_to_msg(&s->udp4); - if (msg->udp4 == NULL) - goto fail_msg; - break; - case IPCP_UDP6: - msg->udp6 = udp6_config_s_to_msg(&s->udp6); - if (msg->udp6 == NULL) - goto fail_msg; - break; default: /* No checks here */ break; @@ -714,6 +933,8 @@ struct ipcp_config ipcp_config_msg_to_s(const ipcp_config_msg_t * msg) assert(msg != NULL); + memset(&s, 0, sizeof(s)); + s.type = msg->ipcp_type; s.layer_info = layer_info_msg_to_s(msg->layer_info); @@ -724,17 +945,6 @@ struct ipcp_config ipcp_config_msg_to_s(const ipcp_config_msg_t * msg) case IPCP_UNICAST: s.unicast = uni_config_msg_to_s(msg->unicast); break; - case IPCP_ETH_LLC: - /* FALLTHRU */ - case IPCP_ETH_DIX: - s.eth = eth_config_msg_to_s(msg->eth); - break; - case IPCP_UDP4: - s.udp4 = udp4_config_msg_to_s(msg->udp4); - break; - case IPCP_UDP6: - s.udp6 = udp6_config_msg_to_s(msg->udp6); - break; case IPCP_BROADCAST: break; default: diff --git a/src/lib/serdes-irm.c b/src/lib/serdes-irm.c index 1d9b4dec..20b25a1b 100644 --- a/src/lib/serdes-irm.c +++ b/src/lib/serdes-irm.c @@ -441,6 +441,7 @@ int proc_exit__irm_req_ser(buffer_t * buf) return -ENOMEM; } +/* data is borrowed from the caller; detach before free. */ int ipcp_flow_req_arr__irm_req_ser(buffer_t * buf, const buffer_t * dst, const struct flow_info * flow, @@ -494,6 +495,138 @@ int ipcp_flow_req_arr__irm_req_ser(buffer_t * buf, return -ENOMEM; } +static int __ep_flow_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data, + int response, + const char * dst, + int code) +{ + irm_msg_t * msg; + size_t len; + int err = -ENOMEM; + + msg = malloc(sizeof(*msg)); + if (msg == NULL) + goto fail_malloc; + + irm_msg__init(msg); + + msg->code = code; + msg->flow_info = flow_info_s_to_msg(flow); + if (msg->flow_info == NULL) + goto fail_msg; + + if (dst != NULL) { + msg->dst = strdup(dst); + if (msg->dst == NULL) + goto fail_msg; + } + + if (data != NULL) { + msg->has_pk = true; + msg->pk.len = data->len; + msg->pk.data = data->data; + } + + msg->has_response = true; + msg->response = response; + + len = irm_msg__get_packed_size(msg); + if (len == 0 || len > buf->len) + goto fail_msg; + + buf->len = len; + + irm_msg__pack(msg, buf->data); + + err = 0; + fail_msg: + msg->pk.len = 0; + msg->pk.data = NULL; + + irm_msg__free_unpacked(msg, NULL); + fail_malloc: + return err; +} + +int poa_flow_alloc__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const char * dst) +{ + return __ep_flow_ser(buf, flow, NULL, 0, dst, + IRM_MSG_CODE__IRM_POA_FLOW_ALLOC); +} + +int poa_flow_alloc_r__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data, + int response) +{ + return __ep_flow_ser(buf, flow, data, response, NULL, + IRM_MSG_CODE__IRM_POA_FLOW_ALLOC_R); +} + +int ipcp_poa_flow_req_arr__irm_req_ser(buffer_t * buf, + const struct flow_info * flow, + const buffer_t * data) +{ + return __ep_flow_ser(buf, flow, data, 0, NULL, + IRM_MSG_CODE__IPCP_POA_FLOW_REQ_ARR); +} + +int poa_flow__irm_result_des(buffer_t * buf, + struct flow_info * flow, + buffer_t * data) +{ + irm_msg_t * msg; + int err; + + msg = irm_msg__unpack(NULL, buf->len, buf->data); + if (msg == NULL) { + err = -EIRMD; + goto fail_msg; + } + + if (!msg->has_result) { + err = -EIRMD; + goto fail; + } + + if (msg->result < 0) { + err = msg->result; + goto fail; + } + + if (msg->flow_info == NULL) { + err = -EBADF; + goto fail; + } + + *flow = flow_info_msg_to_s(msg->flow_info); + + clrbuf(*data); + + if (msg->has_pk && msg->pk.len > 0) { + data->data = malloc(msg->pk.len); + if (data->data == NULL) { + err = -ENOMEM; + goto fail; + } + memcpy(data->data, msg->pk.data, msg->pk.len); + + data->len = msg->pk.len; + } + + irm_msg__free_unpacked(msg, NULL); + + return 0; + fail: + irm_msg__free_unpacked(msg, NULL); + fail_msg: + return err; +} + int ipcp_flow_update_arr__irm_req_ser(buffer_t * buf, const struct flow_info * flow, const buffer_t * data) diff --git a/src/lib/tests/CMakeLists.txt b/src/lib/tests/CMakeLists.txt index 002d94af..1f2e9ba2 100644 --- a/src/lib/tests/CMakeLists.txt +++ b/src/lib/tests/CMakeLists.txt @@ -11,6 +11,7 @@ create_test_sourcelist(${PARENT_DIR}_tests test_suite.c bitmap_test.c btree_test.c crypt_test.c + poa_test.c hash_test.c kex_test.c kex_test_ml_kem.c @@ -25,6 +26,12 @@ create_test_sourcelist(${PARENT_DIR}_tests test_suite.c add_executable(${PARENT_DIR}_test ${${PARENT_DIR}_tests}) +if(HAVE_LIBURCU) + # poa_test.c pulls in poa.h, whose urcu guard needs C99. + set_source_files_properties(poa_test.c PROPERTIES + COMPILE_OPTIONS "-std=gnu99") +endif() + target_include_directories(${PARENT_DIR}_test PRIVATE ${CMAKE_SOURCE_DIR}/src/lib) diff --git a/src/lib/tests/poa_test.c b/src/lib/tests/poa_test.c new file mode 100644 index 00000000..99886769 --- /dev/null +++ b/src/lib/tests/poa_test.c @@ -0,0 +1,307 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Flow PoA tests + * + * 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 <test/test.h> + +#include "poa/addr.c" +#ifdef HAVE_RAW_SOCKETS +#include "poa/eth.c" +#endif + +#include <arpa/inet.h> +#include <stdio.h> + +#ifdef HAVE_RAW_SOCKETS +static const uint8_t eth_our_mac[POA_MAC_SIZE] = + { 0x02, 0x00, 0x00, 0x00, 0x00, 0x01 }; +static const uint8_t eth_far_mac[POA_MAC_SIZE] = + { 0x02, 0x00, 0x00, 0x00, 0x00, 0x02 }; +static const uint8_t eth_bc_mac[POA_MAC_SIZE] = + { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; +#endif + +/* PoA core stubs: the reader threads never run in this test. */ +void poa_rx_pkt(struct poa * poa, + uint32_t eid, + struct ssm_pk_buff * spb) +{ + (void) poa; + (void) eid; + (void) spb; +} + +void poa_rx_mgmt(struct poa * poa, + const struct poa_addr * src, + const uint8_t * buf, + size_t len) +{ + (void) poa; + (void) src; + (void) buf; + (void) len; +} + +int poa_spb_reserve(struct ssm_pk_buff ** spb, + size_t len) +{ + (void) spb; + (void) len; + + return -1; +} + +size_t poa_link_updown(int ifindex, + bool up) +{ + (void) ifindex; + (void) up; + + return 0; +} + +bool poa_has_name(const uint8_t * hash) +{ + (void) hash; + + return false; +} + +int poa_bcast_mgmt(const struct poa_addr * dst, + const uint8_t * buf, + size_t len) +{ + (void) dst; + (void) buf; + (void) len; + + return 0; +} + +static void udp4_addr(struct poa_addr * addr, + const char * ip, + uint16_t port) +{ + memset(addr, 0, sizeof(*addr)); + + addr->type = POA_UDP4; + addr->udp4.port = port; + + inet_pton(AF_INET, ip, &addr->udp4.ip_addr); +} + +static int test_poa_addr_cmp(void) +{ + struct poa_addr a; + struct poa_addr b; + + TEST_START(); + + udp4_addr(&a, "10.0.0.10", 3435); + udp4_addr(&b, "10.0.0.10", 3435); + + if (poa_addr_cmp(&a, &b) != 0) { + printf("Identical addresses did not match.\n"); + goto fail; + } + + udp4_addr(&b, "10.0.0.10", 3436); + + if (poa_addr_cmp(&a, &b) == 0) { + printf("Addresses with a different port matched.\n"); + goto fail; + } + + udp4_addr(&b, "10.0.0.11", 3435); + + if (poa_addr_cmp(&a, &b) == 0) { + printf("Addresses with a different ip matched.\n"); + goto fail; + } + + b.type = POA_UDP6; + if (poa_addr_cmp(&a, &b) == 0) { + printf("Addresses of a different type matched.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +static int test_poa_mgmt_msg_qos(void) +{ + struct poa_mgmt_msg msg; + qosspec_t qs; + + TEST_START(); + + poa_mgmt_msg_ser(&msg, POA_FLOW_REQ, 1, 2, qos_stream, 0, 0); + + if (msg.code != POA_FLOW_REQ) { + printf("Wrong code in management message.\n"); + goto fail; + } + + if (ntoh32(msg.s_eid) != 1 || ntoh32(msg.d_eid) != 2) { + printf("Wrong PoA ids in management message.\n"); + goto fail; + } + + memset(&qs, 0, sizeof(qs)); + + poa_mgmt_msg_qos(&msg, &qs); + + if (memcmp(&qs, &qos_stream, sizeof(qs)) != 0) { + printf("QoS did not survive the management message.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +#ifdef HAVE_RAW_SOCKETS + +static void eth_fake_priv(struct eth_priv * priv, + struct poa * e) +{ + memset(priv, 0, sizeof(*priv)); + memset(e, 0, sizeof(*e)); + + e->type = POA_ETH; + + strcpy(e->local.eth.src.dev, "test0"); + + priv->poa = e; + priv->ethertype = htons(0xA000); + priv->mtu = 1500; + + memcpy(priv->hw_addr, eth_our_mac, POA_MAC_SIZE); +} + +static int test_eth_frame(void) +{ + struct eth_priv priv; + struct poa e; + uint8_t buf[64]; + struct eth_hdr * hdr = (struct eth_hdr *) buf; + const char * body = "hello"; + size_t n; + size_t plen; + uint32_t eid; + + TEST_START(); + + eth_fake_priv(&priv, &e); + + eth_hdr_ser(&priv, hdr, eth_our_mac, 7, strlen(body)); + + memcpy(buf + ETH_HDR_TOT_SIZE, body, strlen(body)); + + n = ETH_HDR_TOT_SIZE + strlen(body); + if (frame_parse(&priv, buf, n, &eid, &plen) < 0) { + printf("Failed to parse a valid frame.\n"); + goto fail; + } + + if (eid != 7 || plen != strlen(body)) { + printf("Expected eid 7 len %zu, got %u len %zu.\n", + strlen(body), eid, plen); + goto fail; + } + + hdr->poa.hcs ^= 0xFF; + + if (frame_parse(&priv, buf, n, &eid, &plen) == 0) { + printf("Accepted a corrupt header checksum.\n"); + goto fail; + } + + hdr->poa.hcs ^= 0xFF; + hdr->ethertype ^= 0xFF; + + if (frame_parse(&priv, buf, n, &eid, &plen) == 0) { + printf("Accepted a foreign Ethertype.\n"); + goto fail; + } + + hdr->ethertype ^= 0xFF; + + if (frame_parse(&priv, buf, ETH_HDR_SIZE + 2, &eid, &plen) == 0) { + printf("Accepted a truncated frame.\n"); + goto fail; + } + + eth_hdr_ser(&priv, hdr, eth_far_mac, 7, strlen(body)); + + if (frame_parse(&priv, buf, n, &eid, &plen) == 0) { + printf("Accepted a frame for another host.\n"); + goto fail; + } + + eth_hdr_ser(&priv, hdr, eth_bc_mac, 7, strlen(body)); + + if (frame_parse(&priv, buf, n, &eid, &plen) < 0) { + printf("Rejected a broadcast frame.\n"); + goto fail; + } + + TEST_SUCCESS(); + + return TEST_RC_SUCCESS; + fail: + TEST_FAIL(); + return TEST_RC_FAIL; +} + +#endif /* HAVE_RAW_SOCKETS */ + +int poa_test(int argc, + char ** argv) +{ + int ret = 0; + + (void) argc; + (void) argv; + + ret |= test_poa_addr_cmp(); + ret |= test_poa_mgmt_msg_qos(); +#ifdef HAVE_RAW_SOCKETS + ret |= test_eth_frame(); +#endif + + return ret; +} diff --git a/src/tools/CMakeLists.txt b/src/tools/CMakeLists.txt index 6b418838..bb601733 100644 --- a/src/tools/CMakeLists.txt +++ b/src/tools/CMakeLists.txt @@ -15,6 +15,10 @@ set(IRM_SOURCES irm/irm_ipcp_destroy.c irm/irm_ipcp_bootstrap.c irm/irm_ipcp_enroll.c + irm/irm_ipcp_poa.c + irm/irm_ipcp_poa_attach.c + irm/irm_ipcp_poa_detach.c + irm/irm_ipcp_poa_list.c irm/irm_ipcp_list.c irm/irm_ipcp_connect.c irm/irm_ipcp_disconnect.c diff --git a/src/tools/irm/irm_ipcp.c b/src/tools/irm/irm_ipcp.c index 34458a20..5c0db9cf 100644 --- a/src/tools/irm/irm_ipcp.c +++ b/src/tools/irm/irm_ipcp.c @@ -46,6 +46,7 @@ static void usage(void) printf("Usage: irm ipcp [OPERATION]\n\n" "where OPERATION in {create destroy\n" " bootstrap enroll\n" + " poa\n" " connect disconnect\n" " list\n" " help}\n"); @@ -68,6 +69,7 @@ static const struct cmd { { "destroy", do_destroy_ipcp }, { "bootstrap", do_bootstrap_ipcp }, { "enroll", do_enroll_ipcp }, + { "poa", poa_cmd }, { "connect", do_connect_ipcp }, { "disconnect", do_disconnect_ipcp }, { "list", do_list_ipcp}, diff --git a/src/tools/irm/irm_ipcp_bootstrap.c b/src/tools/irm/irm_ipcp_bootstrap.c index cc8bf8fa..bbcf2312 100644 --- a/src/tools/irm/irm_ipcp_bootstrap.c +++ b/src/tools/irm/irm_ipcp_bootstrap.c @@ -53,10 +53,6 @@ #define UNICAST "unicast" #define BROADCAST "broadcast" -#define IP_UDP4 "udp4" -#define IP_UDP6 "udp6" -#define ETH_LLC "eth-llc" -#define ETH_DIX "eth-dix" #define LOCAL "local" #define MD5 "MD5" @@ -76,18 +72,13 @@ #define DT(x) default_dt_config.x #define DHT(x) default_dht_config.params.x #define UNI(x) default_uni_config.x -#define DIX(x) eth_dix_default_conf.eth.x -#define LLC(x) eth_llc_default_conf.eth.x -#define UD4(x) udp4_default_conf.udp4.x -#define UD6(x) udp6_default_conf.udp6.x static char * usage_str = \ "Usage: irm ipcp bootstrap\n" " name <ipcp name>\n" " layer <layer name>\n" " [type [TYPE]]\n" - "where TYPE in {" UNICAST " " BROADCAST " " LOCAL " " - IP_UDP4 " " IP_UDP6 " " ETH_LLC " " ETH_DIX "},\n\n" + "where TYPE in {" UNICAST " " BROADCAST " " LOCAL "},\n\n" "if TYPE == " UNICAST "\n" " [addr_auth <ADDRESS_POLICY> (default: %s)]\n" " [directory <DIRECTORY_POLICY> (default: %s)]\n" @@ -117,28 +108,6 @@ static char * usage_str = \ " [ls_t_recalc <pff recalc interval (s)> (default: %ld)]\n" " [ls_t_update <LSA update interval (s)> (default: %ld)]\n" " [ls_t_timeo <link timeout (s)> (default: %ld)]\n\n" - "if TYPE == " IP_UDP4 "\n" - " ip <IP address in dotted notation>\n" - " [port <UDP port> (default: %d)]\n" - " [dns <DDNS IPv4 address in dotted notation>" - " (default: none)]\n\n" - "if TYPE == " IP_UDP6 "\n" - " ip <IPv6 address>\n" - " [port <UDP port> (default: %d)]\n" - " [dns <DDNS IPv6 address>" - " (default: none)]\n\n" - - "if TYPE == " ETH_LLC "\n" - " dev <interface name>\n" - " [hash [ALGORITHM] (default: %s)]\n" - "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" - "if TYPE == " ETH_DIX "\n" - " dev <interface name>\n" - " [ethertype <ethertype> (default: 0x%4X)]\n" - " [hash [ALGORITHM] (default: %s)]\n" - "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " - SHA3_384 " " SHA3_512 "}\n\n" "if TYPE == " LOCAL "\n" " [hash [ALGORITHM] (default: %s)]\n" "where ALGORITHM in {" SHA3_224 " " SHA3_256 " " @@ -160,15 +129,6 @@ static void usage(void) /* ls */ default_ls_config.t_recalc, default_ls_config.t_update, default_ls_config.t_timeo, - /* udp4 */ - UD4(port), - /* udp6 */ - UD6(port), - /* eth_llc */ - SHA3_256, - /* eth_dix */ - DIX(ethertype), - SHA3_256, /* local */ SHA3_256, /* broadcast */ @@ -190,19 +150,9 @@ int do_bootstrap_ipcp(int argc, enum pol_addr_auth addr_auth_type = UNI(addr_auth_type); enum pol_cong_avoid cong_avoid = UNI(cong_avoid); enum pol_dir_hash hash_algo = DIR_HASH_SHA3_256; - char * ipstr = NULL; - char * dnsstr = NULL; - struct in_addr ip4_addr = {.s_addr = INADDR_ANY}; - struct in_addr dns4_addr = UD4(dns_addr); - int port4 = UD4(port); - struct in6_addr ip6_addr = IN6ADDR_ANY_INIT; - struct in6_addr dns6_addr = UD6(dns_addr); - int port6 = UD6(port); char * ipcp_type = NULL; enum ipcp_type type = IPCP_INVALID; char * layer = NULL; - char * dev = NULL; - uint16_t ethertype = DIX(ethertype); struct ipcp_list_info * ipcps; ssize_t len = 0; int i = 0; @@ -211,6 +161,14 @@ int do_bootstrap_ipcp(int argc, while (argc > 0) { cargs = 2; + if (matches(*argv, "autobind") == 0) + cargs = 1; + + if (argc < cargs) { + usage(); + return -1; + } + if (matches(*argv, "type") == 0) { ipcp_type = *(argv + 1); } else if (matches(*argv, "layer") == 0) { @@ -228,24 +186,6 @@ int do_bootstrap_ipcp(int argc, hash_algo = DIR_HASH_SHA3_512; else goto unknown_param; - } else if (matches(*argv, "ip") == 0) { - ipstr = *(argv + 1); - } else if (matches(*argv, "dns") == 0) { - dnsstr = *(argv + 1); - } else if (matches(*argv, "device") == 0) { - dev = *(argv + 1); - } else if (matches(*argv, "ethertype") == 0) { - /* NOTE: We might do some more checks on strtol. */ - if (matches(*(argv + 1), "0x") == 0) - ethertype = strtol(*(argv + 1), NULL, 0); - else - ethertype = strtol(*(argv + 1), NULL, 16); - if (ethertype < 0x0600 || ethertype >= 0xFFFF) { - printf("Invalid Ethertype: \"%s\".\n" - "Recommended range: 0xA000-0xEFFF.\n", - *(argv + 1)); - return -1; - } } else if (matches(*argv, "addr") == 0) { addr_size = atoi(*(argv + 1)); } else if (matches(*argv, "eid") == 0) { @@ -254,9 +194,6 @@ int do_bootstrap_ipcp(int argc, max_ttl = atoi(*(argv + 1)); } else if (matches(*argv, "rtt") == 0) { max_rtt = atoi(*(argv + 1)); - } else if (matches(*argv, "port") == 0) { - port4 = atoi(*(argv + 1)); - port6 = port4; } else if (matches(*argv, "autobind") == 0) { autobind = true; cargs = 1; @@ -334,55 +271,11 @@ int do_bootstrap_ipcp(int argc, type = IPCP_UNICAST; else if (matches(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (matches(ipcp_type, IP_UDP4) == 0) - type = IPCP_UDP4; - else if (matches(ipcp_type, IP_UDP6) == 0) - type = IPCP_UDP6; - else if (matches(ipcp_type, ETH_DIX) == 0) - type = IPCP_ETH_DIX; - else if (matches(ipcp_type, ETH_LLC) == 0) - type = IPCP_ETH_LLC; else if (matches(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; else goto fail_usage; } - if (type == IPCP_UDP4) { - if (inet_pton (AF_INET, ipstr, &ip4_addr) != 1) { - printf("Invalid IPv4 address: \"%s\".\n", ipstr); - goto fail_usage; - } - - if (ip4_addr.s_addr == INADDR_ANY) { - printf("Cannot use IPv4 address: \"%s\".\n", ipstr); - goto fail_usage; - } - - if (dnsstr != NULL && - inet_pton(AF_INET, dnsstr, &dns4_addr) != 1) { - printf("Invalid DNS IPv4 address: \"%s\".\n", dnsstr); - goto fail_usage; - } - } - - if (type == IPCP_UDP6) { - if (inet_pton(AF_INET6, ipstr, &ip6_addr) != 1) { - printf("Invalid IPv6 address: \"%s\".\n", ipstr); - goto fail_usage; - } - - if (IN6_IS_ADDR_UNSPECIFIED(&ip6_addr)) { - printf("Cannot use IPv6 address: \"%s\".\n", ipstr); - goto fail_usage; - } - - if (dnsstr != NULL && - inet_pton(AF_INET6, dnsstr, &dns6_addr) != 1) { - printf("Invalid DNS IPv6 address: \"%s\".\n", dnsstr); - goto fail_usage; - } - } - if (pid == -1) { if (ipcp_type == NULL) { printf("No IPCPs matching %s found.\n\n", ipcp); @@ -432,29 +325,6 @@ int do_bootstrap_ipcp(int argc, conf.unicast.cong_avoid = cong_avoid; conf.unicast.dir = dir_config; break; - case IPCP_UDP4: - conf.udp4.ip_addr = ip4_addr; - conf.udp4.dns_addr = dns4_addr; - conf.udp4.port = port4; - break; - case IPCP_UDP6: - conf.udp6.ip_addr = ip6_addr; - conf.udp6.dns_addr = dns6_addr; - conf.udp6.port = port6; - break; - case IPCP_ETH_DIX: - conf.eth.ethertype = ethertype; - /* FALLTHRU */ - case IPCP_ETH_LLC: - if (dev == NULL) - goto fail_usage; - if (strlen(dev) > DEV_NAME_SIZE) { - printf("Device name too long.\n\n"); - goto fail_usage; - } - - strcpy(conf.eth.dev, dev); - break; case IPCP_BROADCAST: /* FALLTHRU */ case IPCP_LOCAL: diff --git a/src/tools/irm/irm_ipcp_connect.c b/src/tools/irm/irm_ipcp_connect.c index fb21faec..66646e7d 100644 --- a/src/tools/irm/irm_ipcp_connect.c +++ b/src/tools/irm/irm_ipcp_connect.c @@ -44,6 +44,7 @@ #include <stdio.h> #include <stdlib.h> +#include <arpa/inet.h> #include <string.h> #define DT "dt" @@ -55,10 +56,14 @@ static void usage(void) " name <ipcp name>\n" " dst <name of destination IPCP>\n" " [component [COMPONENT]]\n" - "where COMPONENT in {" DT " " MGMT "}\n\n" + " [udp [UDP_PEER]]\n" + " [eth [dev <device>]" + " [ethertype <ethertype>]]\n" + "where COMPONENT in {" DT " " MGMT "}\n" + "and UDP_PEER is <IP address or host>[:<port>]\n\n" "if COMPONENT == " DT "\n" - " [qos [QOS]\n" - "where QOS in {raw, best, voice, video, data}\n"); + " [qos [QOS]]\n" + "where QOS in {raw, safe, rt, rt-safe, msg}\n"); } int do_connect_ipcp(int argc, @@ -69,13 +74,29 @@ int do_connect_ipcp(int argc, char * comp = "*"; char * component = NULL; char * qos = NULL; + char * udpstr = NULL; + char * devstr = NULL; + uint16_t ethertype = POA_ETHERTYPE; + bool eth = false; + struct poa_addr addr; + struct poa_addr * pa = NULL; struct ipcp_list_info * ipcps; ssize_t len = 0; pid_t pid = -1; ssize_t i; + int cargs; qosspec_t qs = qos_raw; while (argc > 0) { + cargs = 2; + if (strcmp(*argv, "eth") == 0) + cargs = 1; + + if (argc < cargs) { + usage(); + return -1; + } + if (matches(*argv, "name") == 0) { ipcp = *(argv + 1); } else if (matches(*argv, "dst") == 0) { @@ -84,21 +105,66 @@ int do_connect_ipcp(int argc, comp = *(argv + 1); } else if (matches(*argv, "qos") == 0) { qos = *(argv + 1); + } else if (strcmp(*argv, "udp") == 0) { + udpstr = *(argv + 1); + } else if (strcmp(*argv, "dev") == 0) { + devstr = *(argv + 1); + } else if (strcmp(*argv, "ethertype") == 0) { + if (parse_ethertype(*(argv + 1), ðertype) < 0) { + printf("Invalid ethertype: \"%s\".\n", + *(argv + 1)); + return -1; + } + } else if (strcmp(*argv, "eth") == 0) { + eth = true; + cargs = 1; } else { printf("\"%s\" is unknown, try \"irm " "ipcp connect\".\n", *argv); return -1; } - argc -= 2; - argv += 2; + argc -= cargs; + argv += cargs; } - if (ipcp == NULL || dst == NULL || comp == NULL) { + if (ipcp == NULL || comp == NULL) { usage(); return -1; } + memset(&addr, 0, sizeof(addr)); + + if (udpstr != NULL) { + if (poa_addr_set_udp(&addr, udpstr) < 0) + return -1; + pa = &addr; + } + + if (eth) { + if (udpstr != NULL) { + printf("Connect via udp or eth, not both.\n"); + return -1; + } + + if (dst == NULL) { + usage(); + return -1; + } + + if (poa_addr_set_eth(&addr, devstr, ethertype) < 0) + return -1; + pa = &addr; + } + + if (dst == NULL && pa == NULL) { + usage(); + return -1; + } + + if (dst == NULL) + dst = ""; + if (qos != NULL) { if (strcmp(qos, "raw") == 0) qs = qos_raw; @@ -110,9 +176,11 @@ int do_connect_ipcp(int argc, qs = qos_rt_safe; else if (strcmp(qos, "msg") == 0) qs = qos_msg; - else if (strcmp(qos, "stream") == 0) - qs = qos_stream; - else + else if (strcmp(qos, "stream") == 0) { + printf("Stream QoS is not allowed on " + "IPCP component flows.\n"); + return -1; + } else printf("Unknown QoS cube, defaulting to raw.\n"); } @@ -129,13 +197,13 @@ int do_connect_ipcp(int argc, if (wildcard_match(comp, MGMT) == 0) { component = MGMT_COMP; /* FIXME: move to qos_msg when stable */ - if (irm_connect_ipcp(pid, dst, component, qos_raw)) + if (irm_connect_ipcp(pid, dst, component, qos_raw, pa)) return -1; } if (wildcard_match(comp, DT) == 0) { component = DT_COMP; - if (irm_connect_ipcp(pid, dst, component, qs)) + if (irm_connect_ipcp(pid, dst, component, qs, pa)) return -1; } diff --git a/src/tools/irm/irm_ipcp_create.c b/src/tools/irm/irm_ipcp_create.c index c6b2074b..15cfedde 100644 --- a/src/tools/irm/irm_ipcp_create.c +++ b/src/tools/irm/irm_ipcp_create.c @@ -46,10 +46,6 @@ #define UNICAST "unicast" #define BROADCAST "broadcast" -#define UDP4 "udp4" -#define UDP6 "udp6" -#define ETH_LLC "eth-llc" -#define ETH_DIX "eth-dix" #define LOCAL "local" static void usage(void) @@ -57,8 +53,7 @@ static void usage(void) printf("Usage: irm ipcp create\n" " name <ipcp name>\n" " type [TYPE]\n\n" - "where TYPE in {" UNICAST " " BROADCAST " " LOCAL " " - UDP4 " " UDP6 " " ETH_LLC " " ETH_DIX "}\n"); + "where TYPE in {" UNICAST " " BROADCAST " " LOCAL "}\n"); } int do_create_ipcp(int argc, @@ -93,16 +88,8 @@ int do_create_ipcp(int argc, type = IPCP_UNICAST; else if (strcmp(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (strcmp(ipcp_type, UDP4) == 0) - type = IPCP_UDP4; - else if (strcmp(ipcp_type, UDP6) == 0) - type = IPCP_UDP6; else if (strcmp(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; - else if (strcmp(ipcp_type, ETH_LLC) == 0) - type = IPCP_ETH_LLC; - else if (strcmp(ipcp_type, ETH_DIX) == 0) - type = IPCP_ETH_DIX; else { printf("IPCP type \"%s\" is unknown.\n", ipcp_type); usage(); diff --git a/src/tools/irm/irm_ipcp_enroll.c b/src/tools/irm/irm_ipcp_enroll.c index 350b536e..54fed022 100644 --- a/src/tools/irm/irm_ipcp_enroll.c +++ b/src/tools/irm/irm_ipcp_enroll.c @@ -44,6 +44,7 @@ #include "irm_ops.h" #include "irm_utils.h" +#include <arpa/inet.h> #include <string.h> #define UNICAST "unicast" @@ -57,7 +58,10 @@ static void usage(void) " [dst <destination to enroll with>]\n" " [type [TYPE], default = " UNICAST "]\n" " [autobind]\n" - "where TYPE in {" UNICAST " " BROADCAST "}\n"); + " [udp [UDP_PEER]]\n" + " [eth [dev <device>] [ethertype <ethertype>]]\n" + "where TYPE in {" UNICAST " " BROADCAST "}\n" + "and UDP_PEER is <IP address or host>[:<port>]\n"); } static int get_layer_name(const char * ipcp, @@ -86,6 +90,12 @@ int do_enroll_ipcp(int argc, char * ipcp = NULL; char * layer = NULL; char * dst = NULL; + char * udpstr = NULL; + char * devstr = NULL; + uint16_t ethertype = POA_ETHERTYPE; + bool eth = false; + struct poa_addr addr; + struct poa_addr * pa = NULL; struct ipcp_list_info * ipcps; pid_t pid = -1; ssize_t len = 0; @@ -97,14 +107,37 @@ int do_enroll_ipcp(int argc, while (argc > 0) { cargs = 2; + if (strcmp(*argv, "eth") == 0) + cargs = 1; + else if (matches(*argv, "autobind") == 0) + cargs = 1; + + if (argc < cargs) { + usage(); + return -1; + } + if (matches(*argv, "name") == 0) { ipcp = *(argv + 1); } else if (matches(*argv, "type") == 0) { ipcp_type = *(argv + 1); } else if (matches(*argv, "layer") == 0) { layer = *(argv + 1); - } else if (matches(*argv, "dst") == 0) { + } else if (strcmp(*argv, "dst") == 0) { dst = *(argv + 1); + } else if (strcmp(*argv, "udp") == 0) { + udpstr = *(argv + 1); + } else if (strcmp(*argv, "dev") == 0) { + devstr = *(argv + 1); + } else if (strcmp(*argv, "ethertype") == 0) { + if (parse_ethertype(*(argv + 1), ðertype) < 0) { + printf("Invalid ethertype: \"%s\".\n", + *(argv + 1)); + return -1; + } + } else if (strcmp(*argv, "eth") == 0) { + eth = true; + cargs = 1; } else if (matches(*argv, "autobind") == 0) { autobind = true; cargs = 1; @@ -126,6 +159,25 @@ int do_enroll_ipcp(int argc, if (dst == NULL) dst = layer; + memset(&addr, 0, sizeof(addr)); + + if (udpstr != NULL) { + if (poa_addr_set_udp(&addr, udpstr) < 0) + return -1; + pa = &addr; + } + + if (eth) { + if (udpstr != NULL) { + printf("Enroll via udp or eth, not both.\n"); + return -1; + } + + if (poa_addr_set_eth(&addr, devstr, ethertype) < 0) + return -1; + pa = &addr; + } + if (strcmp(ipcp_type, UNICAST) == 0) type = IPCP_UNICAST; else if (strcmp(ipcp_type, BROADCAST) == 0) @@ -155,7 +207,7 @@ int do_enroll_ipcp(int argc, pid = ipcps[i].pid; - if (irm_enroll_ipcp(pid, dst)) { + if (irm_enroll_ipcp(pid, dst, pa) < 0) { printf("Failed to enroll IPCP.\n"); goto fail; } diff --git a/src/tools/irm/irm_ipcp_list.c b/src/tools/irm/irm_ipcp_list.c index a211a02b..60154efb 100644 --- a/src/tools/irm/irm_ipcp_list.c +++ b/src/tools/irm/irm_ipcp_list.c @@ -48,10 +48,6 @@ #define UNICAST "unicast" #define BROADCAST "broadcast" -#define UDP4 "udp4" -#define UDP6 "udp6" -#define ETH_LLC "eth-llc" -#define ETH_DIX "eth-dix" #define LOCAL "local" static void usage(void) @@ -60,8 +56,7 @@ static void usage(void) " [name <ipcp name>]\n" " [layer <layer_name>]\n\n" " [type [TYPE]]\n\n" - "where TYPE = {" UNICAST " " LOCAL " " - UDP4 " " UDP6 " " ETH_LLC " " ETH_DIX "}\n"); + "where TYPE = {" UNICAST " " BROADCAST " " LOCAL "}\n"); } static char * str_type(enum ipcp_type type) @@ -71,14 +66,6 @@ static char * str_type(enum ipcp_type type) return UNICAST; case IPCP_BROADCAST: return BROADCAST; - case IPCP_ETH_LLC: - return ETH_LLC; - case IPCP_ETH_DIX: - return ETH_DIX; - case IPCP_UDP4: - return UDP4; - case IPCP_UDP6: - return UDP6; case IPCP_LOCAL: return LOCAL; default: @@ -116,16 +103,8 @@ int do_list_ipcp(int argc, type = IPCP_UNICAST; else if (strcmp(ipcp_type, BROADCAST) == 0) type = IPCP_BROADCAST; - else if (strcmp(ipcp_type, UDP4) == 0) - type = IPCP_UDP4; - else if (strcmp(ipcp_type, UDP6) == 0) - type = IPCP_UDP6; else if (strcmp(ipcp_type, LOCAL) == 0) type = IPCP_LOCAL; - else if (strcmp(ipcp_type, ETH_LLC) == 0) - type = IPCP_ETH_LLC; - else if (strcmp(ipcp_type, ETH_DIX) == 0) - type = IPCP_ETH_DIX; else { usage(); return -1; diff --git a/src/tools/irm/irm_ipcp_poa.c b/src/tools/irm/irm_ipcp_poa.c new file mode 100644 index 00000000..cd939020 --- /dev/null +++ b/src/tools/irm/irm_ipcp_poa.c @@ -0,0 +1,98 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Connect components of unicast or broadcast IPC processes + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + * + * 3. Neither the name of the copyright holder nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "irm_ops.h" +#include "irm_utils.h" + +#include <stdio.h> +#include <string.h> + +static void usage(void) +{ + printf("Usage: irm ipcp poa [OPERATION]\n\n" + "where OPERATION in {attach detach list help}\n"); +} + +static int do_help(int argc, + char ** argv) +{ + (void) argc; + (void) argv; + + usage(); + + return 0; +} + +static const struct cmd { + const char * cmd; + int (* func)(int argc, char ** argv); +} cmds[] = { + { "attach", do_attach_ipcp }, + { "detach", do_detach_ipcp }, + { "list", do_list_poas }, + { "help", do_help }, + { NULL, NULL } +}; + +static int do_cmd(const char * argv0, + int argc, + char ** argv) +{ + const struct cmd * c; + + for (c = cmds; c->cmd != NULL; ++c) + if (matches(argv0, c->cmd) == 0) + return c->func(argc - 1, argv + 1); + + fprintf(stderr, "\"%s\" is unknown, try \"irm ipcp poa help\".\n", + argv0); + + return -1; +} + +int poa_cmd(int argc, + char ** argv) +{ + if (argc < 1) { + usage(); + return -1; + } + + return do_cmd(argv[0], argc, argv); +} diff --git a/src/tools/irm/irm_ipcp_poa_attach.c b/src/tools/irm/irm_ipcp_poa_attach.c new file mode 100644 index 00000000..74a83344 --- /dev/null +++ b/src/tools/irm/irm_ipcp_poa_attach.c @@ -0,0 +1,153 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Connect components of unicast or broadcast IPC processes + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + * + * 3. Neither the name of the copyright holder nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <ouroboros/irm.h> + +#include "irm_ops.h" +#include "irm_utils.h" + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +static void usage(void) +{ + printf("Usage: irm ipcp poa attach\n" + " name <ipcp name>\n" + " [udp [UDP_POA]]\n" + " [eth [ETH_POA]]\n" + "where exactly one of udp or eth is given\n" + "and UDP_POA is <local IP address>[:<port>]\n" + "and ETH_POA is dev <device> [ethertype <ethertype>]\n"); +} + +int do_attach_ipcp(int argc, + char ** argv) +{ + char * ipcp = NULL; + char * udpstr = NULL; + char * devstr = NULL; + uint16_t ethertype = POA_ETHERTYPE; + bool eth = false; + struct poa_spec poa; + char str[POA_STRLEN + 1]; + struct ipcp_list_info * ipcps; + ssize_t len = 0; + pid_t pid = -1; + ssize_t i; + int cargs; + + while (argc > 0) { + cargs = 2; + + if (strcmp(*argv, "eth") == 0) + cargs = 1; + + if (argc < cargs) { + usage(); + return -1; + } + + if (matches(*argv, "name") == 0) { + ipcp = *(argv + 1); + } else if (strcmp(*argv, "udp") == 0) { + udpstr = *(argv + 1); + } else if (strcmp(*argv, "dev") == 0) { + devstr = *(argv + 1); + } else if (strcmp(*argv, "ethertype") == 0) { + if (parse_ethertype(*(argv + 1), ðertype) < 0) { + printf("Invalid ethertype: \"%s\".\n", + *(argv + 1)); + return -1; + } + } else if (strcmp(*argv, "eth") == 0) { + eth = true; + cargs = 1; + } else { + printf("\"%s\" is unknown, try \"irm ipcp " + "poa attach\".\n", *argv); + return -1; + } + + argc -= cargs; + argv += cargs; + } + + if (ipcp == NULL) { + usage(); + return -1; + } + + if (eth && udpstr != NULL) { + printf("A PoA is udp or eth, not both.\n"); + return -1; + } + + if (eth && devstr == NULL) { + printf("An eth PoA needs a device.\n"); + return -1; + } + + if (!eth && udpstr == NULL) { + usage(); + return -1; + } + + if (poa_spec_set(&poa, udpstr, devstr, ethertype) < 0) + return -1; + + len = irm_list_ipcps(&ipcps); + for (i = 0; i < len; i++) + if (strcmp(ipcps[i].name, ipcp) == 0) + pid = ipcps[i].pid; + + free(ipcps); + + if (pid == -1) { + printf("No such IPCP: \"%s\".\n", ipcp); + return -1; + } + + if (irm_attach_ipcp(pid, &poa) < 0) { + poa_spec_str(&poa, str, sizeof(str)); + printf("Failed to attach PoA %s on IPCP %s.\n", str, ipcp); + return -1; + } + + return 0; +} diff --git a/src/tools/irm/irm_ipcp_poa_detach.c b/src/tools/irm/irm_ipcp_poa_detach.c new file mode 100644 index 00000000..ce6bef13 --- /dev/null +++ b/src/tools/irm/irm_ipcp_poa_detach.c @@ -0,0 +1,153 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * Connect components of unicast or broadcast IPC processes + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + * + * 3. Neither the name of the copyright holder nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <ouroboros/irm.h> + +#include "irm_ops.h" +#include "irm_utils.h" + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +static void usage(void) +{ + printf("Usage: irm ipcp poa detach\n" + " name <ipcp name>\n" + " [udp [UDP_POA]]\n" + " [eth [ETH_POA]]\n" + "where exactly one of udp or eth is given\n" + "and UDP_POA is <local IP address>[:<port>]\n" + "and ETH_POA is dev <device> [ethertype <ethertype>]\n"); +} + +int do_detach_ipcp(int argc, + char ** argv) +{ + char * ipcp = NULL; + char * udpstr = NULL; + char * devstr = NULL; + uint16_t ethertype = POA_ETHERTYPE; + bool eth = false; + struct poa_spec poa; + char str[POA_STRLEN + 1]; + struct ipcp_list_info * ipcps; + ssize_t len = 0; + pid_t pid = -1; + ssize_t i; + int cargs; + + while (argc > 0) { + cargs = 2; + + if (strcmp(*argv, "eth") == 0) + cargs = 1; + + if (argc < cargs) { + usage(); + return -1; + } + + if (matches(*argv, "name") == 0) { + ipcp = *(argv + 1); + } else if (strcmp(*argv, "udp") == 0) { + udpstr = *(argv + 1); + } else if (strcmp(*argv, "dev") == 0) { + devstr = *(argv + 1); + } else if (strcmp(*argv, "ethertype") == 0) { + if (parse_ethertype(*(argv + 1), ðertype) < 0) { + printf("Invalid ethertype: \"%s\".\n", + *(argv + 1)); + return -1; + } + } else if (strcmp(*argv, "eth") == 0) { + eth = true; + cargs = 1; + } else { + printf("\"%s\" is unknown, try \"irm ipcp " + "poa detach\".\n", *argv); + return -1; + } + + argc -= cargs; + argv += cargs; + } + + if (ipcp == NULL) { + usage(); + return -1; + } + + if (eth && udpstr != NULL) { + printf("A PoA is udp or eth, not both.\n"); + return -1; + } + + if (eth && devstr == NULL) { + printf("An eth PoA needs a device.\n"); + return -1; + } + + if (!eth && udpstr == NULL) { + usage(); + return -1; + } + + if (poa_spec_set(&poa, udpstr, devstr, ethertype) < 0) + return -1; + + len = irm_list_ipcps(&ipcps); + for (i = 0; i < len; i++) + if (strcmp(ipcps[i].name, ipcp) == 0) + pid = ipcps[i].pid; + + free(ipcps); + + if (pid == -1) { + printf("No such IPCP: \"%s\".\n", ipcp); + return -1; + } + + if (irm_detach_ipcp(pid, &poa) < 0) { + poa_spec_str(&poa, str, sizeof(str)); + printf("Failed to detach PoA %s on IPCP %s.\n", str, ipcp); + return -1; + } + + return 0; +} diff --git a/src/tools/irm/irm_ipcp_poa_list.c b/src/tools/irm/irm_ipcp_poa_list.c new file mode 100644 index 00000000..8797c3c1 --- /dev/null +++ b/src/tools/irm/irm_ipcp_poa_list.c @@ -0,0 +1,136 @@ +/* + * Ouroboros - Copyright (C) 2016 - 2026 + * + * List the points of attachment of an IPC process + * + * Dimitri Staessens <dimitri@ouroboros.rocks> + * Sander Vrijders <sander@ouroboros.rocks> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + * + * 3. Neither the name of the copyright holder nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <ouroboros/irm.h> + +#include "irm_ops.h" +#include "irm_utils.h" + +#include <arpa/inet.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +static void usage(void) +{ + printf("Usage: irm ipcp poa list\n" + " name <ipcp name>\n"); +} + +static void print_poa(const struct poa_spec * poa) +{ + char buf[INET6_ADDRSTRLEN]; + + switch (poa->type) { + case POA_UDP4: + if (inet_ntop(AF_INET, &poa->udp4.ip_addr, buf, + sizeof(buf)) == NULL) + return; + printf("%-6s %s:%u\n", "udp4", buf, poa->udp4.port); + break; + case POA_UDP6: + if (inet_ntop(AF_INET6, &poa->udp6.ip_addr, buf, + sizeof(buf)) == NULL) + return; + printf("%-6s [%s]:%u\n", "udp6", buf, poa->udp6.port); + break; + case POA_ETH: + printf("%-6s %s 0x%04X %02x:%02x:%02x:%02x:%02x:%02x\n", + "eth", poa->eth.dev, poa->eth.ethertype, + poa->eth.mac[0], poa->eth.mac[1], poa->eth.mac[2], + poa->eth.mac[3], poa->eth.mac[4], poa->eth.mac[5]); + break; + default: + break; + } +} + +int do_list_poas(int argc, + char ** argv) +{ + char * ipcp = NULL; + struct poa_spec * poas = NULL; + struct ipcp_list_info * ipcps; + ssize_t len = 0; + pid_t pid = -1; + ssize_t n; + ssize_t i; + + while (argc > 1) { + if (matches(*argv, "name") == 0) { + ipcp = *(argv + 1); + } else { + printf("\"%s\" is unknown, try \"irm ipcp " + "poa list\".\n", *argv); + return -1; + } + + argc -= 2; + argv += 2; + } + + if (ipcp == NULL) { + usage(); + return -1; + } + + len = irm_list_ipcps(&ipcps); + for (i = 0; i < len; i++) + if (strcmp(ipcps[i].name, ipcp) == 0) + pid = ipcps[i].pid; + + free(ipcps); + + if (pid == -1) { + printf("No such IPCP: \"%s\".\n", ipcp); + return -1; + } + + n = irm_list_poas(pid, &poas); + if (n < 0) { + printf("Failed to list PoAs of %s.\n", ipcp); + return -1; + } + + for (i = 0; i < n; i++) + print_poa(&poas[i]); + + free(poas); + + return 0; +} diff --git a/src/tools/irm/irm_ops.h b/src/tools/irm/irm_ops.h index 195c5cbc..7f6c65cd 100644 --- a/src/tools/irm/irm_ops.h +++ b/src/tools/irm/irm_ops.h @@ -51,6 +51,18 @@ int do_bootstrap_ipcp(int argc, int do_enroll_ipcp(int argc, char ** argv); +int poa_cmd(int argc, + char ** argv); + +int do_attach_ipcp(int argc, + char ** argv); + +int do_detach_ipcp(int argc, + char ** argv); + +int do_list_poas(int argc, + char ** argv); + int do_connect_ipcp(int argc, char ** argv); diff --git a/src/tools/irm/irm_utils.c b/src/tools/irm/irm_utils.c index 69873097..c43accec 100644 --- a/src/tools/irm/irm_utils.c +++ b/src/tools/irm/irm_utils.c @@ -77,10 +77,21 @@ */ -#include <string.h> +#if defined(__linux__) || defined(__CYGWIN__) +#define _DEFAULT_SOURCE +#else +#define _POSIX_C_SOURCE 200809L +#endif + +#include <ouroboros/ipcp.h> #include "irm_utils.h" +#include <arpa/inet.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + int matches(const char * cmd, const char * pattern) { @@ -123,3 +134,223 @@ int wildcard_match(const char * pattern, } } } + +/* Splits "<addr>[:<port>]"; bare IPv6 needs no brackets. */ +static int parse_udp_str(const char * str, + char * host, + int * port) +{ + struct in6_addr v6; + char buf[POA_HOST_STRLEN + 1]; + char * p; + char * end; + long n; + + *port = POA_UDP_PORT; + + if (strlen(str) > POA_HOST_STRLEN) + goto fail; + + strcpy(buf, str); + + if (buf[0] == '[') { + p = strchr(buf, ']'); + if (p == NULL) + goto fail; + + *p++ = '\0'; + + strcpy(host, buf + 1); + + if (*p == '\0') + return 0; + + if (*p != ':') + goto fail; + + ++p; + } else if (inet_pton(AF_INET6, buf, &v6) == 1) { + strcpy(host, buf); + return 0; + } else { + p = strrchr(buf, ':'); + if (p == NULL) { + strcpy(host, buf); + return 0; + } + + *p++ = '\0'; + + strcpy(host, buf); + } + + n = strtol(p, &end, 10); + if (*p == '\0' || *end != '\0' || n < 1 || n > 65535) + goto fail; + + *port = (int) n; + + return 0; + fail: + printf("Invalid UDP address: \"%s\".\n", str); + return -1; +} + +/* An unresolved name is left for the IRMd, which picks the family. */ +int poa_addr_set_udp(struct poa_addr * addr, + const char * str) +{ + char host[POA_HOST_STRLEN + 1]; + int port; + + if (parse_udp_str(str, host, &port) < 0) + return -1; + + if (inet_pton(AF_INET, host, &addr->udp4.ip_addr) == 1) { + addr->type = POA_UDP4; + addr->udp4.port = port; + return 0; + } + + if (inet_pton(AF_INET6, host, &addr->udp6.ip_addr) == 1) { + addr->type = POA_UDP6; + addr->udp6.port = port; + return 0; + } + + addr->type = POA_UDP; + addr->udp4.port = port; + + strcpy(addr->hostname, host); + + return 0; +} + +/* + * Parses a hex ethertype; rejects garbage and out-of-range values. + * Overflow clamps to LONG_MAX and lands in the range check. + */ +int parse_ethertype(const char * str, + uint16_t * ethertype) +{ + char * end; + long val; + + val = strtol(str, &end, 16); + + if (end == str || *end != '\0') + return -1; + + if (val < 0 || val > 0xFFFF) + return -1; + + *ethertype = (uint16_t) val; + + return 0; +} + +int poa_addr_set_eth(struct poa_addr * addr, + const char * devstr, + uint16_t ethertype) +{ + addr->type = POA_ETH; + + addr->eth.src.ethertype = ethertype; + addr->eth.dst.ethertype = ethertype; + + if (devstr != NULL) { + if (strlen(devstr) > DEV_NAME_SIZE) { + printf("Invalid device name: \"%s\".\n", devstr); + return -1; + } + + strcpy(addr->eth.src.dev, devstr); + } + + return 0; +} + +/* Matches src/ipcpd/ipcp.c; keep in sync. */ +void poa_spec_str(const struct poa_spec * poa, + char * buf, + size_t len) +{ + char addr[INET6_ADDRSTRLEN]; + + switch (poa->type) { + case POA_UDP4: + if (inet_ntop(AF_INET, &poa->udp4.ip_addr, + addr, sizeof(addr)) == NULL) + break; + + snprintf(buf, len, "udp4 %s:%u", addr, poa->udp4.port); + return; + case POA_UDP6: + if (inet_ntop(AF_INET6, &poa->udp6.ip_addr, + addr, sizeof(addr)) == NULL) + break; + + snprintf(buf, len, "udp6 [%s]:%u", addr, poa->udp6.port); + return; + case POA_ETH: + snprintf(buf, len, "eth %s 0x%04X", poa->eth.dev, + poa->eth.ethertype); + return; + default: + break; + } + + snprintf(buf, len, "(unknown)"); +} + +int poa_spec_set(struct poa_spec * poa, + const char * udpstr, + const char * devstr, + uint16_t ethertype) +{ + char host[POA_HOST_STRLEN + 1]; + int port; + + memset(poa, 0, sizeof(*poa)); + + if ((udpstr != NULL) + (devstr != NULL) > 1) { + printf("A PoA is an address or a device.\n"); + return -1; + } + + if (udpstr != NULL) { + if (parse_udp_str(udpstr, host, &port) < 0) + return -1; + + if (inet_pton(AF_INET, host, &poa->udp4.ip_addr) == 1) { + poa->type = POA_UDP4; + poa->udp4.port = port; + return 0; + } + + if (inet_pton(AF_INET6, host, &poa->udp6.ip_addr) == 1) { + poa->type = POA_UDP6; + poa->udp6.port = port; + return 0; + } + + printf("Invalid IP address: \"%s\".\n", udpstr); + return -1; + } + + if (devstr != NULL) { + if (strlen(devstr) > DEV_NAME_SIZE) { + printf("Invalid device name: \"%s\".\n", devstr); + return -1; + } + + poa->type = POA_ETH; + poa->eth.ethertype = ethertype; + + strcpy(poa->eth.dev, devstr); + + return 0; + } + + return -1; +} diff --git a/src/tools/irm/irm_utils.h b/src/tools/irm/irm_utils.h index c6d4bf18..3760b4e4 100644 --- a/src/tools/irm/irm_utils.h +++ b/src/tools/irm/irm_utils.h @@ -79,10 +79,43 @@ #ifndef OUROBOROS_TOOLS_IRM_UTILS_H #define OUROBOROS_TOOLS_IRM_UTILS_H -int matches(const char * cmd, - const char * pattern); +#include <ouroboros/ipcp.h> -int wildcard_match(const char * pattern, - const char * string); +#include <arpa/inet.h> + +#include <stdint.h> + +int matches(const char * cmd, + const char * pattern); + +int wildcard_match(const char * pattern, + const char * string); + +/* Fill one PoA to attach or detach: an address or a device. */ +int poa_spec_set(struct poa_spec * poa, + const char * udpstr, + const char * devstr, + uint16_t ethertype); + +/* Fill a peer PoA address; a host name is resolved by the IRMd. */ +int poa_addr_set_udp(struct poa_addr * addr, + const char * str); + +/* Fits "eth <dev> 0x<type>", the longest PoA rendering. */ +/* Matches src/ipcpd/ipcp.c; keep in sync. */ +#define POA_STRLEN (DEV_NAME_SIZE + 11) + +/* Names a PoA the way the IPCP logs it. */ +void poa_spec_str(const struct poa_spec * poa, + char * buf, + size_t len); + +int poa_addr_set_eth(struct poa_addr * addr, + const char * devstr, + uint16_t ethertype); + +/* Parses a hex ethertype; rejects garbage and out-of-range values. */ +int parse_ethertype(const char * str, + uint16_t * ethertype); #endif /* OUROBOROS_TOOLS_IRM_UTILS_H */ |