irmd: Pass MTU from IPCP to process for FRCT

FRCT needs to know the MTU for fragmentation. The MTU is now passed
from the layer serving the flow to the process as part of flow
allocation.

Signed-off-by: Dimitri Staessens <dimitri@ouroboros.rocks>
Signed-off-by: Sander Vrijders <sander@ouroboros.rocks>

15 files changed, 123 insertions, 13 deletions

diff --git a/src/ipcpd/broadcast/dt.c b/src/ipcpd/broadcast/dt.c
index 30e89a4f..95483e33 100644
--- a/src/ipcpd/broadcast/dt.c
+++ b/src/ipcpd/broadcast/dt.c
@@ -28,7 +28,7 @@
 
 #include "config.h"
 
-#define BROADCAST_MTU    1400 /* FIXME: avoid packet copy. */
+#define BROADCAST_MTU    IPCP_BROADCAST_MTU /* FIXME: avoid packet copy. */
 
 #define DT               "dt"
 #define OUROBOROS_PREFIX DT
diff --git a/src/ipcpd/broadcast/main.c b/src/ipcpd/broadcast/main.c
index b3cbdc79..77e22531 100644
--- a/src/ipcpd/broadcast/main.c
+++ b/src/ipcpd/broadcast/main.c
@@ -242,7 +242,7 @@ static int broadcast_ipcp_join(int             fd,
 
         notifier_event(NOTIFY_DT_CONN_ADD, &conn);
 
-        ipcp_flow_alloc_reply(fd, 0, mpl, &data);
+        ipcp_flow_alloc_reply(fd, 0, mpl, IPCP_BROADCAST_MTU, &data);
 
         return 0;
 }
diff --git a/src/ipcpd/config.h.in b/src/ipcpd/config.h.in
index 37ee2c5e..a719fa3b 100644
--- a/src/ipcpd/config.h.in
+++ b/src/ipcpd/config.h.in
@@ -46,6 +46,7 @@
 #define IPCP_SCHED_THR_MUL  @IPCP_SCHED_THR_MUL@
 #define PFT_SIZE            @PFT_SIZE@
 #define IPCP_UNICAST_MPL    @IPCP_UNICAST_MPL@
+#define IPCP_UNICAST_MTU    @IPCP_UNICAST_MTU@
 #define CONNMGR_RCV_TIMEOUT @CONNMGR_RCV_TIMEOUT@
 
 #cmakedefine DISABLE_CORE_LOCK
@@ -66,6 +67,8 @@
 #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
@@ -80,7 +83,8 @@
 
 /* local */
 #define IPCP_LOCAL_MPL      @IPCP_LOCAL_MPL@
+#define IPCP_LOCAL_MTU      @IPCP_LOCAL_MTU@
 
 /* broadcast */
-/* local */
 #define IPCP_BROADCAST_MPL  @IPCP_BROADCAST_MPL@
+#define IPCP_BROADCAST_MTU  @IPCP_BROADCAST_MTU@
diff --git a/src/ipcpd/eth/eth.c b/src/ipcpd/eth/eth.c
index 10a8b338..1dc2b8fe 100644
--- a/src/ipcpd/eth/eth.c
+++ b/src/ipcpd/eth/eth.c
@@ -841,7 +841,8 @@ static int eth_ipcp_req(uint8_t *        r_addr,
 {
         int fd;
 
-        fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_ETH_MPL, data);
+        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;
@@ -913,7 +914,8 @@ static int eth_ipcp_alloc_reply(uint8_t *        r_addr,
 #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, data)) < 0) {
+        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;
         }
diff --git a/src/ipcpd/ipcp.c b/src/ipcpd/ipcp.c
index 5ad2401f..1052a686 100644
--- a/src/ipcpd/ipcp.c
+++ b/src/ipcpd/ipcp.c
@@ -363,6 +363,7 @@ static void * acceptloop(void * o)
 int ipcp_wait_flow_req_arr(const uint8_t *  dst,
                            qosspec_t        qs,
                            time_t           mpl,
+                           uint32_t         mtu,
                            const buffer_t * data)
 {
         struct timespec ts = TIMESPEC_INIT_MS(ALLOC_TIMEOUT);
@@ -392,7 +393,7 @@ int ipcp_wait_flow_req_arr(const uint8_t *  dst,
 
         assert(ipcpd.alloc_id == -1);
 
-        fd = ipcp_flow_req_arr(&hash, qs, mpl, data);
+        fd = ipcp_flow_req_arr(&hash, qs, mpl, mtu, data);
         if (fd < 0) {
                 pthread_mutex_unlock(&ipcpd.alloc_lock);
                 log_err("Failed to get fd for flow.");
diff --git a/src/ipcpd/ipcp.h b/src/ipcpd/ipcp.h
index 26a780a3..0adcc694 100644
--- a/src/ipcpd/ipcp.h
+++ b/src/ipcpd/ipcp.h
@@ -98,6 +98,7 @@ enum ipcp_state ipcp_get_state(void);
 int             ipcp_wait_flow_req_arr(const uint8_t *  dst,
                                        qosspec_t        qs,
                                        time_t           mpl,
+                                       uint32_t         mtu,
                                        const buffer_t * data);
 
 int             ipcp_wait_flow_resp(const int fd);
diff --git a/src/ipcpd/local/main.c b/src/ipcpd/local/main.c
index 2c867317..eb9836f2 100644
--- a/src/ipcpd/local/main.c
+++ b/src/ipcpd/local/main.c
@@ -203,7 +203,8 @@ static int local_ipcp_flow_alloc(int              fd,
                 HASH_VAL32(dst), fd);
         assert(dst);
 
-        out_fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_LOCAL_MPL, data);
+        out_fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_LOCAL_MPL,
+                                        IPCP_LOCAL_MTU, data);
         if (out_fd < 0) {
                 log_dbg("Flow allocation failed: %d", out_fd);
                 return -1;
@@ -255,14 +256,16 @@ static int local_ipcp_flow_alloc_resp(int              fd,
         }
 
         if (response < 0) {
-                ipcp_flow_alloc_reply(out_fd, response, mpl, data);
+                ipcp_flow_alloc_reply(out_fd, response, mpl,
+                                      IPCP_LOCAL_MTU, data);
                 log_info("Flow allocation rejected, fds (%d, %d).", out_fd, fd);
                 return 0;
         }
 
         fset_add(local_data.flows, fd);
 
-        if (ipcp_flow_alloc_reply(out_fd, response, mpl, data) < 0) {
+        if (ipcp_flow_alloc_reply(out_fd, response, mpl,
+                                  IPCP_LOCAL_MTU, data) < 0) {
                 log_err("Failed to reply to allocation");
                 fset_del(local_data.flows, fd);
                 return -1;
diff --git a/src/ipcpd/udp/udp.c b/src/ipcpd/udp/udp.c
index 168ea718..8fba5a77 100644
--- a/src/ipcpd/udp/udp.c
+++ b/src/ipcpd/udp/udp.c
@@ -47,6 +47,10 @@
 #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
@@ -130,6 +134,53 @@ static const char * __inet_ntop(const struct __ADDR * addr,
         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;
@@ -285,7 +336,8 @@ static int udp_ipcp_port_req(struct __SOCKADDR * c_saddr,
 {
         int fd;
 
-        fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UDP_MPL, data);
+        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;
@@ -332,7 +384,8 @@ static int udp_ipcp_port_alloc_reply(const struct __SOCKADDR * saddr,
 
         pthread_rwlock_unlock(&udp_data.flows_lock);
 
-        if (ipcp_flow_alloc_reply(s_eid, response, mpl, data) < 0) {
+        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;
         }
diff --git a/src/ipcpd/unicast/fa.c b/src/ipcpd/unicast/fa.c
index 39928ab2..ca27108e 100644
--- a/src/ipcpd/unicast/fa.c
+++ b/src/ipcpd/unicast/fa.c
@@ -500,7 +500,8 @@ static int fa_handle_flow_req(struct fa_msg * msg,
         qs.max_gap      = ntoh32(msg->max_gap);
         qs.timeout      = ntoh32(msg->timeout);
 
-        fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UNICAST_MPL, &data);
+        fd = ipcp_wait_flow_req_arr(dst, qs, IPCP_UNICAST_MPL,
+                                    IPCP_UNICAST_MTU, &data);
         if (fd < 0)
                 return fd;
 
@@ -559,7 +560,8 @@ static int fa_handle_flow_reply(struct fa_msg * msg,
 
         pthread_rwlock_unlock(&fa.flows_lock);
 
-        if (ipcp_flow_alloc_reply(fd, response, mpl, &data) < 0) {
+        if (ipcp_flow_alloc_reply(fd, response, mpl,
+                                  IPCP_UNICAST_MTU, &data) < 0) {
                 log_err("Failed to reply for flow allocation on fd %d.", fd);
                 return -EIRMD;
         }
diff --git a/src/irmd/main.c b/src/irmd/main.c
index 5a6245e0..ffb07334 100644
--- a/src/irmd/main.c
+++ b/src/irmd/main.c
@@ -87,6 +87,8 @@
 #define OAP_SEEN_TIMER     20   /*  s */
 #define DEALLOC_TIME       300  /*  s */
 #define DIRECT_MPL         1    /*  s */
+/* bytes; in-process, bounded only by PUP/GSPP. */
+#define DIRECT_MTU         65000
 
 enum irm_state {
         IRMD_NULL = 0,
@@ -1216,6 +1218,7 @@ static int flow_alloc_direct(const char *       dst,
 
         acc.n_1_pid = flow->n_pid;
         acc.mpl     = DIRECT_MPL;
+        acc.mtu     = DIRECT_MTU;
         acc.qs      = flow->qs;
         acc.state   = FLOW_ALLOCATED;
 
@@ -1251,6 +1254,7 @@ static int flow_alloc_direct(const char *       dst,
         flow->id      = acc.id;
         flow->n_1_pid = acc.n_pid;
         flow->mpl     = DIRECT_MPL;
+        flow->mtu     = DIRECT_MTU;
         flow->state   = FLOW_ALLOCATED;
 
         log_info("Flow %d allocated (direct) for %d to %s.",
diff --git a/src/irmd/reg/flow.c b/src/irmd/reg/flow.c
index 93c3e128..5c709dea 100644
--- a/src/irmd/reg/flow.c
+++ b/src/irmd/reg/flow.c
@@ -42,6 +42,7 @@ struct reg_flow * reg_flow_create(const struct flow_info * info)
         assert(info->n_pid != 0);
         assert(info->n_1_pid == 0);
         assert(info->mpl == 0);
+        assert(info->mtu == 0);
         assert(info->state == FLOW_INIT);
 
         flow = malloc(sizeof(*flow));
@@ -160,6 +161,7 @@ int reg_flow_update(struct reg_flow *  flow,
                 assert(info->mpl != 0);
 
                 flow->info.mpl = info->mpl;
+                flow->info.mtu = info->mtu;
 
                 if (flow->info.state == FLOW_ALLOC_PENDING)
                         break;
diff --git a/src/irmd/reg/tests/flow_test.c b/src/irmd/reg/tests/flow_test.c
index 7e1c1360..10cd9f3d 100644
--- a/src/irmd/reg/tests/flow_test.c
+++ b/src/irmd/reg/tests/flow_test.c
@@ -122,6 +122,21 @@ static int test_reg_flow_create_has_mpl(void) {
         return TEST_RC_SUCCESS;
 }
 
+static int test_reg_flow_create_has_mtu(void) {
+        struct flow_info info = {
+                .id      = 1,
+                .n_pid   = 1,
+                .n_1_pid = 0,
+                .mtu     = 1400,
+                .qs      = qos_raw,
+                .state   = FLOW_ALLOC_PENDING
+        };
+
+        reg_flow_create(&info); /* assert fail */
+
+        return TEST_RC_SUCCESS;
+}
+
 static int test_reg_flow_update(void)
 {
         struct reg_flow * f;
@@ -210,6 +225,7 @@ static int test_reg_flow_assert_fails(void)
         ret |= test_assert_fail(test_reg_flow_create_has_n_1_pid);
         ret |= test_assert_fail(test_reg_flow_create_wrong_state);
         ret |= test_assert_fail(test_reg_flow_create_has_mpl);
+        ret |= test_assert_fail(test_reg_flow_create_has_mtu);
         ret |= test_assert_fail(test_reg_flow_update_wrong_id);
 
         return ret;
diff --git a/src/irmd/reg/tests/reg_test.c b/src/irmd/reg/tests/reg_test.c
index 5a5178c2..dcc26783 100644
--- a/src/irmd/reg/tests/reg_test.c
+++ b/src/irmd/reg/tests/reg_test.c
@@ -31,6 +31,7 @@
 #define TEST_N_1_PID  3999
 #define TEST_FAKE_ID  9128349
 #define TEST_MPL      5
+#define TEST_MTU      1400
 #define TEST_PROG     "reg_test" /* own binary for binary check */
 #define TEST_IPCP     "testipcp"
 #define TEST_NAME     "testname"
@@ -266,6 +267,7 @@ static int test_reg_accept_flow_success(void)
 
         n_1_info.id  = info.id;
         n_1_info.mpl = 1;
+        n_1_info.mtu = TEST_MTU;
 
         pthread_create(&thr, NULL, test_flow_respond_accept, &n_1_info);
 
@@ -284,6 +286,11 @@ static int test_reg_accept_flow_success(void)
                 goto fail;
         }
 
+        if (info.mtu != TEST_MTU) {
+                printf("MTU not propagated.\n");
+                goto fail;
+        }
+
         if (rbuf.data == NULL) {
                 printf("rbuf data not returned.\n");
                 goto fail;
@@ -363,6 +370,7 @@ static int test_reg_accept_flow_success_no_crypt(void)
 
         n_1_info.id  = info.id;
         n_1_info.mpl = 1;
+        n_1_info.mtu = TEST_MTU;
 
         pthread_create(&thr, NULL, test_flow_respond_accept, &n_1_info);
 
@@ -381,6 +389,11 @@ static int test_reg_accept_flow_success_no_crypt(void)
                 goto fail;
         }
 
+        if (info.mtu != TEST_MTU) {
+                printf("MTU not propagated.\n");
+                goto fail;
+        }
+
         if (rbuf.data == NULL) {
                 printf("rbuf data was not returned.\n");
                 goto fail;
@@ -533,6 +546,7 @@ static int test_reg_respond_alloc_duplicate(void)
 
         n_1_info.id  = info.id;
         n_1_info.mpl = 1;
+        n_1_info.mtu = TEST_MTU;
 
         pthread_create(&thr, NULL, test_flow_respond_alloc, &n_1_info);
 
@@ -547,6 +561,11 @@ static int test_reg_respond_alloc_duplicate(void)
         pthread_join(thr, NULL);
         freebuf(rbuf);
 
+        if (info.mtu != TEST_MTU) {
+                printf("MTU not propagated.\n");
+                goto fail;
+        }
+
         /* Duplicate reply on an already-ALLOCATED flow must not assert. */
         dup_info       = n_1_info;
         dup_info.state = FLOW_DEALLOCATED;
diff --git a/src/lib/pb/model.proto b/src/lib/pb/model.proto
index f1382f3d..51bea760 100644
--- a/src/lib/pb/model.proto
+++ b/src/lib/pb/model.proto
@@ -41,6 +41,7 @@ message flow_info_msg {
         required uint32      state      = 5;
         required qosspec_msg qos        = 6;
         required uint32      uid        = 7;
+        required uint32      mtu        = 8; /* Layer MTU (bytes).       */
 }
 
 message name_info_msg {
diff --git a/src/lib/protobuf.c b/src/lib/protobuf.c
index 28b3aab2..75c71dfd 100644
--- a/src/lib/protobuf.c
+++ b/src/lib/protobuf.c
@@ -81,6 +81,7 @@ flow_info_msg_t * flow_info_s_to_msg(const struct flow_info * s)
         msg->mpl      = s->mpl;
         msg->state    = s->state;
         msg->uid      = s->uid;
+        msg->mtu      = s->mtu;
         msg->qos      = qos_spec_s_to_msg(&s->qs);
         if (msg->qos == NULL)
                 goto fail_msg;
@@ -107,6 +108,7 @@ struct flow_info flow_info_msg_to_s(const flow_info_msg_t * msg)
         s.mpl     = msg->mpl;
         s.state   = msg->state;
         s.uid     = msg->uid;
+        s.mtu     = msg->mtu;
         s.qs      = qos_spec_msg_to_s(msg->qos);
 
         return s;