Index: head/sys/kern/uipc_ktls.c =================================================================== --- head/sys/kern/uipc_ktls.c (revision 360291) +++ head/sys/kern/uipc_ktls.c (revision 360292) @@ -1,1591 +1,1592 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2014-2019 Netflix Inc. * * 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 REGENTS 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 __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__aarch64__) || defined(__amd64__) || defined(__i386__) #include #endif #include #include #include #ifdef RSS #include +#include #include #endif #if defined(INET) || defined(INET6) #include #include #endif #include #ifdef TCP_OFFLOAD #include #endif #include #include #include #include #include struct ktls_wq { struct mtx mtx; STAILQ_HEAD(, mbuf_ext_pgs) head; bool running; } __aligned(CACHE_LINE_SIZE); static struct ktls_wq *ktls_wq; static struct proc *ktls_proc; LIST_HEAD(, ktls_crypto_backend) ktls_backends; static struct rmlock ktls_backends_lock; static uma_zone_t ktls_session_zone; static uint16_t ktls_cpuid_lookup[MAXCPU]; SYSCTL_NODE(_kern_ipc, OID_AUTO, tls, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Kernel TLS offload"); SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, stats, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Kernel TLS offload stats"); static int ktls_allow_unload; SYSCTL_INT(_kern_ipc_tls, OID_AUTO, allow_unload, CTLFLAG_RDTUN, &ktls_allow_unload, 0, "Allow software crypto modules to unload"); #ifdef RSS static int ktls_bind_threads = 1; #else static int ktls_bind_threads; #endif SYSCTL_INT(_kern_ipc_tls, OID_AUTO, bind_threads, CTLFLAG_RDTUN, &ktls_bind_threads, 0, "Bind crypto threads to cores or domains at boot"); static u_int ktls_maxlen = 16384; SYSCTL_UINT(_kern_ipc_tls, OID_AUTO, maxlen, CTLFLAG_RWTUN, &ktls_maxlen, 0, "Maximum TLS record size"); static int ktls_number_threads; SYSCTL_INT(_kern_ipc_tls_stats, OID_AUTO, threads, CTLFLAG_RD, &ktls_number_threads, 0, "Number of TLS threads in thread-pool"); static bool ktls_offload_enable; SYSCTL_BOOL(_kern_ipc_tls, OID_AUTO, enable, CTLFLAG_RW, &ktls_offload_enable, 0, "Enable support for kernel TLS offload"); static bool ktls_cbc_enable = true; SYSCTL_BOOL(_kern_ipc_tls, OID_AUTO, cbc_enable, CTLFLAG_RW, &ktls_cbc_enable, 1, "Enable Support of AES-CBC crypto for kernel TLS"); static counter_u64_t ktls_tasks_active; SYSCTL_COUNTER_U64(_kern_ipc_tls, OID_AUTO, tasks_active, CTLFLAG_RD, &ktls_tasks_active, "Number of active tasks"); static counter_u64_t ktls_cnt_on; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, so_inqueue, CTLFLAG_RD, &ktls_cnt_on, "Number of TLS records in queue to tasks for SW crypto"); static counter_u64_t ktls_offload_total; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, offload_total, CTLFLAG_RD, &ktls_offload_total, "Total successful TLS setups (parameters set)"); static counter_u64_t ktls_offload_enable_calls; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, enable_calls, CTLFLAG_RD, &ktls_offload_enable_calls, "Total number of TLS enable calls made"); static counter_u64_t ktls_offload_active; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, active, CTLFLAG_RD, &ktls_offload_active, "Total Active TLS sessions"); static counter_u64_t ktls_offload_failed_crypto; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, failed_crypto, CTLFLAG_RD, &ktls_offload_failed_crypto, "Total TLS crypto failures"); static counter_u64_t ktls_switch_to_ifnet; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_to_ifnet, CTLFLAG_RD, &ktls_switch_to_ifnet, "TLS sessions switched from SW to ifnet"); static counter_u64_t ktls_switch_to_sw; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_to_sw, CTLFLAG_RD, &ktls_switch_to_sw, "TLS sessions switched from ifnet to SW"); static counter_u64_t ktls_switch_failed; SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_failed, CTLFLAG_RD, &ktls_switch_failed, "TLS sessions unable to switch between SW and ifnet"); SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, sw, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Software TLS session stats"); SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, ifnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Hardware (ifnet) TLS session stats"); #ifdef TCP_OFFLOAD SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, toe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "TOE TLS session stats"); #endif static counter_u64_t ktls_sw_cbc; SYSCTL_COUNTER_U64(_kern_ipc_tls_sw, OID_AUTO, cbc, CTLFLAG_RD, &ktls_sw_cbc, "Active number of software TLS sessions using AES-CBC"); static counter_u64_t ktls_sw_gcm; SYSCTL_COUNTER_U64(_kern_ipc_tls_sw, OID_AUTO, gcm, CTLFLAG_RD, &ktls_sw_gcm, "Active number of software TLS sessions using AES-GCM"); static counter_u64_t ktls_ifnet_cbc; SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, cbc, CTLFLAG_RD, &ktls_ifnet_cbc, "Active number of ifnet TLS sessions using AES-CBC"); static counter_u64_t ktls_ifnet_gcm; SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, gcm, CTLFLAG_RD, &ktls_ifnet_gcm, "Active number of ifnet TLS sessions using AES-GCM"); static counter_u64_t ktls_ifnet_reset; SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset, CTLFLAG_RD, &ktls_ifnet_reset, "TLS sessions updated to a new ifnet send tag"); static counter_u64_t ktls_ifnet_reset_dropped; SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset_dropped, CTLFLAG_RD, &ktls_ifnet_reset_dropped, "TLS sessions dropped after failing to update ifnet send tag"); static counter_u64_t ktls_ifnet_reset_failed; SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset_failed, CTLFLAG_RD, &ktls_ifnet_reset_failed, "TLS sessions that failed to allocate a new ifnet send tag"); static int ktls_ifnet_permitted; SYSCTL_UINT(_kern_ipc_tls_ifnet, OID_AUTO, permitted, CTLFLAG_RWTUN, &ktls_ifnet_permitted, 1, "Whether to permit hardware (ifnet) TLS sessions"); #ifdef TCP_OFFLOAD static counter_u64_t ktls_toe_cbc; SYSCTL_COUNTER_U64(_kern_ipc_tls_toe, OID_AUTO, cbc, CTLFLAG_RD, &ktls_toe_cbc, "Active number of TOE TLS sessions using AES-CBC"); static counter_u64_t ktls_toe_gcm; SYSCTL_COUNTER_U64(_kern_ipc_tls_toe, OID_AUTO, gcm, CTLFLAG_RD, &ktls_toe_gcm, "Active number of TOE TLS sessions using AES-GCM"); #endif static MALLOC_DEFINE(M_KTLS, "ktls", "Kernel TLS"); static void ktls_cleanup(struct ktls_session *tls); #if defined(INET) || defined(INET6) static void ktls_reset_send_tag(void *context, int pending); #endif static void ktls_work_thread(void *ctx); int ktls_crypto_backend_register(struct ktls_crypto_backend *be) { struct ktls_crypto_backend *curr_be, *tmp; if (be->api_version != KTLS_API_VERSION) { printf("KTLS: API version mismatch (%d vs %d) for %s\n", be->api_version, KTLS_API_VERSION, be->name); return (EINVAL); } rm_wlock(&ktls_backends_lock); printf("KTLS: Registering crypto method %s with prio %d\n", be->name, be->prio); if (LIST_EMPTY(&ktls_backends)) { LIST_INSERT_HEAD(&ktls_backends, be, next); } else { LIST_FOREACH_SAFE(curr_be, &ktls_backends, next, tmp) { if (curr_be->prio < be->prio) { LIST_INSERT_BEFORE(curr_be, be, next); break; } if (LIST_NEXT(curr_be, next) == NULL) { LIST_INSERT_AFTER(curr_be, be, next); break; } } } rm_wunlock(&ktls_backends_lock); return (0); } int ktls_crypto_backend_deregister(struct ktls_crypto_backend *be) { struct ktls_crypto_backend *tmp; /* * Don't error if the backend isn't registered. This permits * MOD_UNLOAD handlers to use this function unconditionally. */ rm_wlock(&ktls_backends_lock); LIST_FOREACH(tmp, &ktls_backends, next) { if (tmp == be) break; } if (tmp == NULL) { rm_wunlock(&ktls_backends_lock); return (0); } if (!ktls_allow_unload) { rm_wunlock(&ktls_backends_lock); printf( "KTLS: Deregistering crypto method %s is not supported\n", be->name); return (EBUSY); } if (be->use_count) { rm_wunlock(&ktls_backends_lock); return (EBUSY); } LIST_REMOVE(be, next); rm_wunlock(&ktls_backends_lock); return (0); } #if defined(INET) || defined(INET6) static u_int ktls_get_cpu(struct socket *so) { struct inpcb *inp; u_int cpuid; inp = sotoinpcb(so); #ifdef RSS cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype); if (cpuid != NETISR_CPUID_NONE) return (cpuid); #endif /* * Just use the flowid to shard connections in a repeatable * fashion. Note that some crypto backends rely on the * serialization provided by having the same connection use * the same queue. */ cpuid = ktls_cpuid_lookup[inp->inp_flowid % ktls_number_threads]; return (cpuid); } #endif static void ktls_init(void *dummy __unused) { struct thread *td; struct pcpu *pc; cpuset_t mask; int error, i; ktls_tasks_active = counter_u64_alloc(M_WAITOK); ktls_cnt_on = counter_u64_alloc(M_WAITOK); ktls_offload_total = counter_u64_alloc(M_WAITOK); ktls_offload_enable_calls = counter_u64_alloc(M_WAITOK); ktls_offload_active = counter_u64_alloc(M_WAITOK); ktls_offload_failed_crypto = counter_u64_alloc(M_WAITOK); ktls_switch_to_ifnet = counter_u64_alloc(M_WAITOK); ktls_switch_to_sw = counter_u64_alloc(M_WAITOK); ktls_switch_failed = counter_u64_alloc(M_WAITOK); ktls_sw_cbc = counter_u64_alloc(M_WAITOK); ktls_sw_gcm = counter_u64_alloc(M_WAITOK); ktls_ifnet_cbc = counter_u64_alloc(M_WAITOK); ktls_ifnet_gcm = counter_u64_alloc(M_WAITOK); ktls_ifnet_reset = counter_u64_alloc(M_WAITOK); ktls_ifnet_reset_dropped = counter_u64_alloc(M_WAITOK); ktls_ifnet_reset_failed = counter_u64_alloc(M_WAITOK); #ifdef TCP_OFFLOAD ktls_toe_cbc = counter_u64_alloc(M_WAITOK); ktls_toe_gcm = counter_u64_alloc(M_WAITOK); #endif rm_init(&ktls_backends_lock, "ktls backends"); LIST_INIT(&ktls_backends); ktls_wq = malloc(sizeof(*ktls_wq) * (mp_maxid + 1), M_KTLS, M_WAITOK | M_ZERO); ktls_session_zone = uma_zcreate("ktls_session", sizeof(struct ktls_session), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); /* * Initialize the workqueues to run the TLS work. We create a * work queue for each CPU. */ CPU_FOREACH(i) { STAILQ_INIT(&ktls_wq[i].head); mtx_init(&ktls_wq[i].mtx, "ktls work queue", NULL, MTX_DEF); error = kproc_kthread_add(ktls_work_thread, &ktls_wq[i], &ktls_proc, &td, 0, 0, "KTLS", "thr_%d", i); if (error) panic("Can't add KTLS thread %d error %d", i, error); /* * Bind threads to cores. If ktls_bind_threads is > * 1, then we bind to the NUMA domain. */ if (ktls_bind_threads) { if (ktls_bind_threads > 1) { pc = pcpu_find(i); CPU_COPY(&cpuset_domain[pc->pc_domain], &mask); } else { CPU_SETOF(i, &mask); } error = cpuset_setthread(td->td_tid, &mask); if (error) panic( "Unable to bind KTLS thread for CPU %d error %d", i, error); } ktls_cpuid_lookup[ktls_number_threads] = i; ktls_number_threads++; } printf("KTLS: Initialized %d threads\n", ktls_number_threads); } SYSINIT(ktls, SI_SUB_SMP + 1, SI_ORDER_ANY, ktls_init, NULL); #if defined(INET) || defined(INET6) static int ktls_create_session(struct socket *so, struct tls_enable *en, struct ktls_session **tlsp) { struct ktls_session *tls; int error; /* Only TLS 1.0 - 1.3 are supported. */ if (en->tls_vmajor != TLS_MAJOR_VER_ONE) return (EINVAL); if (en->tls_vminor < TLS_MINOR_VER_ZERO || en->tls_vminor > TLS_MINOR_VER_THREE) return (EINVAL); if (en->auth_key_len < 0 || en->auth_key_len > TLS_MAX_PARAM_SIZE) return (EINVAL); if (en->cipher_key_len < 0 || en->cipher_key_len > TLS_MAX_PARAM_SIZE) return (EINVAL); if (en->iv_len < 0 || en->iv_len > sizeof(tls->params.iv)) return (EINVAL); /* All supported algorithms require a cipher key. */ if (en->cipher_key_len == 0) return (EINVAL); /* No flags are currently supported. */ if (en->flags != 0) return (EINVAL); /* Common checks for supported algorithms. */ switch (en->cipher_algorithm) { case CRYPTO_AES_NIST_GCM_16: /* * auth_algorithm isn't used, but permit GMAC values * for compatibility. */ switch (en->auth_algorithm) { case 0: #ifdef COMPAT_FREEBSD12 /* XXX: Really 13.0-current COMPAT. */ case CRYPTO_AES_128_NIST_GMAC: case CRYPTO_AES_192_NIST_GMAC: case CRYPTO_AES_256_NIST_GMAC: #endif break; default: return (EINVAL); } if (en->auth_key_len != 0) return (EINVAL); if ((en->tls_vminor == TLS_MINOR_VER_TWO && en->iv_len != TLS_AEAD_GCM_LEN) || (en->tls_vminor == TLS_MINOR_VER_THREE && en->iv_len != TLS_1_3_GCM_IV_LEN)) return (EINVAL); break; case CRYPTO_AES_CBC: switch (en->auth_algorithm) { case CRYPTO_SHA1_HMAC: /* * TLS 1.0 requires an implicit IV. TLS 1.1+ * all use explicit IVs. */ if (en->tls_vminor == TLS_MINOR_VER_ZERO) { if (en->iv_len != TLS_CBC_IMPLICIT_IV_LEN) return (EINVAL); break; } /* FALLTHROUGH */ case CRYPTO_SHA2_256_HMAC: case CRYPTO_SHA2_384_HMAC: /* Ignore any supplied IV. */ en->iv_len = 0; break; default: return (EINVAL); } if (en->auth_key_len == 0) return (EINVAL); break; default: return (EINVAL); } tls = uma_zalloc(ktls_session_zone, M_WAITOK | M_ZERO); counter_u64_add(ktls_offload_active, 1); refcount_init(&tls->refcount, 1); TASK_INIT(&tls->reset_tag_task, 0, ktls_reset_send_tag, tls); tls->wq_index = ktls_get_cpu(so); tls->params.cipher_algorithm = en->cipher_algorithm; tls->params.auth_algorithm = en->auth_algorithm; tls->params.tls_vmajor = en->tls_vmajor; tls->params.tls_vminor = en->tls_vminor; tls->params.flags = en->flags; tls->params.max_frame_len = min(TLS_MAX_MSG_SIZE_V10_2, ktls_maxlen); /* Set the header and trailer lengths. */ tls->params.tls_hlen = sizeof(struct tls_record_layer); switch (en->cipher_algorithm) { case CRYPTO_AES_NIST_GCM_16: /* * TLS 1.2 uses a 4 byte implicit IV with an explicit 8 byte * nonce. TLS 1.3 uses a 12 byte implicit IV. */ if (en->tls_vminor < TLS_MINOR_VER_THREE) tls->params.tls_hlen += sizeof(uint64_t); tls->params.tls_tlen = AES_GMAC_HASH_LEN; /* * TLS 1.3 includes optional padding which we * do not support, and also puts the "real" record * type at the end of the encrypted data. */ if (en->tls_vminor == TLS_MINOR_VER_THREE) tls->params.tls_tlen += sizeof(uint8_t); tls->params.tls_bs = 1; break; case CRYPTO_AES_CBC: switch (en->auth_algorithm) { case CRYPTO_SHA1_HMAC: if (en->tls_vminor == TLS_MINOR_VER_ZERO) { /* Implicit IV, no nonce. */ } else { tls->params.tls_hlen += AES_BLOCK_LEN; } tls->params.tls_tlen = AES_BLOCK_LEN + SHA1_HASH_LEN; break; case CRYPTO_SHA2_256_HMAC: tls->params.tls_hlen += AES_BLOCK_LEN; tls->params.tls_tlen = AES_BLOCK_LEN + SHA2_256_HASH_LEN; break; case CRYPTO_SHA2_384_HMAC: tls->params.tls_hlen += AES_BLOCK_LEN; tls->params.tls_tlen = AES_BLOCK_LEN + SHA2_384_HASH_LEN; break; default: panic("invalid hmac"); } tls->params.tls_bs = AES_BLOCK_LEN; break; default: panic("invalid cipher"); } KASSERT(tls->params.tls_hlen <= MBUF_PEXT_HDR_LEN, ("TLS header length too long: %d", tls->params.tls_hlen)); KASSERT(tls->params.tls_tlen <= MBUF_PEXT_TRAIL_LEN, ("TLS trailer length too long: %d", tls->params.tls_tlen)); if (en->auth_key_len != 0) { tls->params.auth_key_len = en->auth_key_len; tls->params.auth_key = malloc(en->auth_key_len, M_KTLS, M_WAITOK); error = copyin(en->auth_key, tls->params.auth_key, en->auth_key_len); if (error) goto out; } tls->params.cipher_key_len = en->cipher_key_len; tls->params.cipher_key = malloc(en->cipher_key_len, M_KTLS, M_WAITOK); error = copyin(en->cipher_key, tls->params.cipher_key, en->cipher_key_len); if (error) goto out; /* * This holds the implicit portion of the nonce for GCM and * the initial implicit IV for TLS 1.0. The explicit portions * of the IV are generated in ktls_frame(). */ if (en->iv_len != 0) { tls->params.iv_len = en->iv_len; error = copyin(en->iv, tls->params.iv, en->iv_len); if (error) goto out; /* * For TLS 1.2, generate an 8-byte nonce as a counter * to generate unique explicit IVs. * * Store this counter in the last 8 bytes of the IV * array so that it is 8-byte aligned. */ if (en->cipher_algorithm == CRYPTO_AES_NIST_GCM_16 && en->tls_vminor == TLS_MINOR_VER_TWO) arc4rand(tls->params.iv + 8, sizeof(uint64_t), 0); } *tlsp = tls; return (0); out: ktls_cleanup(tls); return (error); } static struct ktls_session * ktls_clone_session(struct ktls_session *tls) { struct ktls_session *tls_new; tls_new = uma_zalloc(ktls_session_zone, M_WAITOK | M_ZERO); counter_u64_add(ktls_offload_active, 1); refcount_init(&tls_new->refcount, 1); /* Copy fields from existing session. */ tls_new->params = tls->params; tls_new->wq_index = tls->wq_index; /* Deep copy keys. */ if (tls_new->params.auth_key != NULL) { tls_new->params.auth_key = malloc(tls->params.auth_key_len, M_KTLS, M_WAITOK); memcpy(tls_new->params.auth_key, tls->params.auth_key, tls->params.auth_key_len); } tls_new->params.cipher_key = malloc(tls->params.cipher_key_len, M_KTLS, M_WAITOK); memcpy(tls_new->params.cipher_key, tls->params.cipher_key, tls->params.cipher_key_len); return (tls_new); } #endif static void ktls_cleanup(struct ktls_session *tls) { counter_u64_add(ktls_offload_active, -1); switch (tls->mode) { case TCP_TLS_MODE_SW: MPASS(tls->be != NULL); switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_sw_cbc, -1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_sw_gcm, -1); break; } tls->free(tls); break; case TCP_TLS_MODE_IFNET: switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_ifnet_cbc, -1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_ifnet_gcm, -1); break; } m_snd_tag_rele(tls->snd_tag); break; #ifdef TCP_OFFLOAD case TCP_TLS_MODE_TOE: switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_toe_cbc, -1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_toe_gcm, -1); break; } break; #endif } if (tls->params.auth_key != NULL) { explicit_bzero(tls->params.auth_key, tls->params.auth_key_len); free(tls->params.auth_key, M_KTLS); tls->params.auth_key = NULL; tls->params.auth_key_len = 0; } if (tls->params.cipher_key != NULL) { explicit_bzero(tls->params.cipher_key, tls->params.cipher_key_len); free(tls->params.cipher_key, M_KTLS); tls->params.cipher_key = NULL; tls->params.cipher_key_len = 0; } explicit_bzero(tls->params.iv, sizeof(tls->params.iv)); } #if defined(INET) || defined(INET6) #ifdef TCP_OFFLOAD static int ktls_try_toe(struct socket *so, struct ktls_session *tls) { struct inpcb *inp; struct tcpcb *tp; int error; inp = so->so_pcb; INP_WLOCK(inp); if (inp->inp_flags2 & INP_FREED) { INP_WUNLOCK(inp); return (ECONNRESET); } if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_WUNLOCK(inp); return (ECONNRESET); } if (inp->inp_socket == NULL) { INP_WUNLOCK(inp); return (ECONNRESET); } tp = intotcpcb(inp); if (tp->tod == NULL) { INP_WUNLOCK(inp); return (EOPNOTSUPP); } error = tcp_offload_alloc_tls_session(tp, tls); INP_WUNLOCK(inp); if (error == 0) { tls->mode = TCP_TLS_MODE_TOE; switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_toe_cbc, 1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_toe_gcm, 1); break; } } return (error); } #endif /* * Common code used when first enabling ifnet TLS on a connection or * when allocating a new ifnet TLS session due to a routing change. * This function allocates a new TLS send tag on whatever interface * the connection is currently routed over. */ static int ktls_alloc_snd_tag(struct inpcb *inp, struct ktls_session *tls, bool force, struct m_snd_tag **mstp) { union if_snd_tag_alloc_params params; struct ifnet *ifp; - struct rtentry *rt; + struct nhop_object *nh; struct tcpcb *tp; int error; INP_RLOCK(inp); if (inp->inp_flags2 & INP_FREED) { INP_RUNLOCK(inp); return (ECONNRESET); } if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_RUNLOCK(inp); return (ECONNRESET); } if (inp->inp_socket == NULL) { INP_RUNLOCK(inp); return (ECONNRESET); } tp = intotcpcb(inp); /* * Check administrative controls on ifnet TLS to determine if * ifnet TLS should be denied. * * - Always permit 'force' requests. * - ktls_ifnet_permitted == 0: always deny. */ if (!force && ktls_ifnet_permitted == 0) { INP_RUNLOCK(inp); return (ENXIO); } /* * XXX: Use the cached route in the inpcb to find the * interface. This should perhaps instead use * rtalloc1_fib(dst, 0, 0, fibnum). Since KTLS is only * enabled after a connection has completed key negotiation in * userland, the cached route will be present in practice. */ - rt = inp->inp_route.ro_rt; - if (rt == NULL || rt->rt_ifp == NULL) { + nh = inp->inp_route.ro_nh; + if (nh == NULL) { INP_RUNLOCK(inp); return (ENXIO); } - ifp = rt->rt_ifp; + ifp = nh->nh_ifp; if_ref(ifp); params.hdr.type = IF_SND_TAG_TYPE_TLS; params.hdr.flowid = inp->inp_flowid; params.hdr.flowtype = inp->inp_flowtype; params.hdr.numa_domain = inp->inp_numa_domain; params.tls.inp = inp; params.tls.tls = tls; INP_RUNLOCK(inp); if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; goto out; } if ((ifp->if_capenable & IFCAP_NOMAP) == 0) { error = EOPNOTSUPP; goto out; } if (inp->inp_vflag & INP_IPV6) { if ((ifp->if_capenable & IFCAP_TXTLS6) == 0) { error = EOPNOTSUPP; goto out; } } else { if ((ifp->if_capenable & IFCAP_TXTLS4) == 0) { error = EOPNOTSUPP; goto out; } } error = ifp->if_snd_tag_alloc(ifp, ¶ms, mstp); out: if_rele(ifp); return (error); } static int ktls_try_ifnet(struct socket *so, struct ktls_session *tls, bool force) { struct m_snd_tag *mst; int error; error = ktls_alloc_snd_tag(so->so_pcb, tls, force, &mst); if (error == 0) { tls->mode = TCP_TLS_MODE_IFNET; tls->snd_tag = mst; switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_ifnet_cbc, 1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_ifnet_gcm, 1); break; } } return (error); } static int ktls_try_sw(struct socket *so, struct ktls_session *tls) { struct rm_priotracker prio; struct ktls_crypto_backend *be; /* * Choose the best software crypto backend. Backends are * stored in sorted priority order (larget value == most * important at the head of the list), so this just stops on * the first backend that claims the session by returning * success. */ if (ktls_allow_unload) rm_rlock(&ktls_backends_lock, &prio); LIST_FOREACH(be, &ktls_backends, next) { if (be->try(so, tls) == 0) break; KASSERT(tls->cipher == NULL, ("ktls backend leaked a cipher pointer")); } if (be != NULL) { if (ktls_allow_unload) be->use_count++; tls->be = be; } if (ktls_allow_unload) rm_runlock(&ktls_backends_lock, &prio); if (be == NULL) return (EOPNOTSUPP); tls->mode = TCP_TLS_MODE_SW; switch (tls->params.cipher_algorithm) { case CRYPTO_AES_CBC: counter_u64_add(ktls_sw_cbc, 1); break; case CRYPTO_AES_NIST_GCM_16: counter_u64_add(ktls_sw_gcm, 1); break; } return (0); } int ktls_enable_tx(struct socket *so, struct tls_enable *en) { struct ktls_session *tls; int error; if (!ktls_offload_enable) return (ENOTSUP); counter_u64_add(ktls_offload_enable_calls, 1); /* * This should always be true since only the TCP socket option * invokes this function. */ if (so->so_proto->pr_protocol != IPPROTO_TCP) return (EINVAL); /* * XXX: Don't overwrite existing sessions. We should permit * this to support rekeying in the future. */ if (so->so_snd.sb_tls_info != NULL) return (EALREADY); if (en->cipher_algorithm == CRYPTO_AES_CBC && !ktls_cbc_enable) return (ENOTSUP); /* TLS requires ext pgs */ if (mb_use_ext_pgs == 0) return (ENXIO); error = ktls_create_session(so, en, &tls); if (error) return (error); /* Prefer TOE -> ifnet TLS -> software TLS. */ #ifdef TCP_OFFLOAD error = ktls_try_toe(so, tls); if (error) #endif error = ktls_try_ifnet(so, tls, false); if (error) error = ktls_try_sw(so, tls); if (error) { ktls_cleanup(tls); return (error); } error = sblock(&so->so_snd, SBL_WAIT); if (error) { ktls_cleanup(tls); return (error); } SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_tls_info = tls; if (tls->mode != TCP_TLS_MODE_SW) so->so_snd.sb_flags |= SB_TLS_IFNET; SOCKBUF_UNLOCK(&so->so_snd); sbunlock(&so->so_snd); counter_u64_add(ktls_offload_total, 1); return (0); } int ktls_get_tx_mode(struct socket *so) { struct ktls_session *tls; struct inpcb *inp; int mode; inp = so->so_pcb; INP_WLOCK_ASSERT(inp); SOCKBUF_LOCK(&so->so_snd); tls = so->so_snd.sb_tls_info; if (tls == NULL) mode = TCP_TLS_MODE_NONE; else mode = tls->mode; SOCKBUF_UNLOCK(&so->so_snd); return (mode); } /* * Switch between SW and ifnet TLS sessions as requested. */ int ktls_set_tx_mode(struct socket *so, int mode) { struct ktls_session *tls, *tls_new; struct inpcb *inp; int error; switch (mode) { case TCP_TLS_MODE_SW: case TCP_TLS_MODE_IFNET: break; default: return (EINVAL); } inp = so->so_pcb; INP_WLOCK_ASSERT(inp); SOCKBUF_LOCK(&so->so_snd); tls = so->so_snd.sb_tls_info; if (tls == NULL) { SOCKBUF_UNLOCK(&so->so_snd); return (0); } if (tls->mode == mode) { SOCKBUF_UNLOCK(&so->so_snd); return (0); } tls = ktls_hold(tls); SOCKBUF_UNLOCK(&so->so_snd); INP_WUNLOCK(inp); tls_new = ktls_clone_session(tls); if (mode == TCP_TLS_MODE_IFNET) error = ktls_try_ifnet(so, tls_new, true); else error = ktls_try_sw(so, tls_new); if (error) { counter_u64_add(ktls_switch_failed, 1); ktls_free(tls_new); ktls_free(tls); INP_WLOCK(inp); return (error); } error = sblock(&so->so_snd, SBL_WAIT); if (error) { counter_u64_add(ktls_switch_failed, 1); ktls_free(tls_new); ktls_free(tls); INP_WLOCK(inp); return (error); } /* * If we raced with another session change, keep the existing * session. */ if (tls != so->so_snd.sb_tls_info) { counter_u64_add(ktls_switch_failed, 1); sbunlock(&so->so_snd); ktls_free(tls_new); ktls_free(tls); INP_WLOCK(inp); return (EBUSY); } SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_tls_info = tls_new; if (tls_new->mode != TCP_TLS_MODE_SW) so->so_snd.sb_flags |= SB_TLS_IFNET; SOCKBUF_UNLOCK(&so->so_snd); sbunlock(&so->so_snd); /* * Drop two references on 'tls'. The first is for the * ktls_hold() above. The second drops the reference from the * socket buffer. */ KASSERT(tls->refcount >= 2, ("too few references on old session")); ktls_free(tls); ktls_free(tls); if (mode == TCP_TLS_MODE_IFNET) counter_u64_add(ktls_switch_to_ifnet, 1); else counter_u64_add(ktls_switch_to_sw, 1); INP_WLOCK(inp); return (0); } /* * Try to allocate a new TLS send tag. This task is scheduled when * ip_output detects a route change while trying to transmit a packet * holding a TLS record. If a new tag is allocated, replace the tag * in the TLS session. Subsequent packets on the connection will use * the new tag. If a new tag cannot be allocated, drop the * connection. */ static void ktls_reset_send_tag(void *context, int pending) { struct epoch_tracker et; struct ktls_session *tls; struct m_snd_tag *old, *new; struct inpcb *inp; struct tcpcb *tp; int error; MPASS(pending == 1); tls = context; inp = tls->inp; /* * Free the old tag first before allocating a new one. * ip[6]_output_send() will treat a NULL send tag the same as * an ifp mismatch and drop packets until a new tag is * allocated. * * Write-lock the INP when changing tls->snd_tag since * ip[6]_output_send() holds a read-lock when reading the * pointer. */ INP_WLOCK(inp); old = tls->snd_tag; tls->snd_tag = NULL; INP_WUNLOCK(inp); if (old != NULL) m_snd_tag_rele(old); error = ktls_alloc_snd_tag(inp, tls, true, &new); if (error == 0) { INP_WLOCK(inp); tls->snd_tag = new; mtx_pool_lock(mtxpool_sleep, tls); tls->reset_pending = false; mtx_pool_unlock(mtxpool_sleep, tls); if (!in_pcbrele_wlocked(inp)) INP_WUNLOCK(inp); counter_u64_add(ktls_ifnet_reset, 1); /* * XXX: Should we kick tcp_output explicitly now that * the send tag is fixed or just rely on timers? */ } else { NET_EPOCH_ENTER(et); INP_WLOCK(inp); if (!in_pcbrele_wlocked(inp)) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED)) { tp = intotcpcb(inp); CURVNET_SET(tp->t_vnet); tp = tcp_drop(tp, ECONNABORTED); CURVNET_RESTORE(); if (tp != NULL) INP_WUNLOCK(inp); counter_u64_add(ktls_ifnet_reset_dropped, 1); } else INP_WUNLOCK(inp); } NET_EPOCH_EXIT(et); counter_u64_add(ktls_ifnet_reset_failed, 1); /* * Leave reset_pending true to avoid future tasks while * the socket goes away. */ } ktls_free(tls); } int ktls_output_eagain(struct inpcb *inp, struct ktls_session *tls) { if (inp == NULL) return (ENOBUFS); INP_LOCK_ASSERT(inp); /* * See if we should schedule a task to update the send tag for * this session. */ mtx_pool_lock(mtxpool_sleep, tls); if (!tls->reset_pending) { (void) ktls_hold(tls); in_pcbref(inp); tls->inp = inp; tls->reset_pending = true; taskqueue_enqueue(taskqueue_thread, &tls->reset_tag_task); } mtx_pool_unlock(mtxpool_sleep, tls); return (ENOBUFS); } #endif void ktls_destroy(struct ktls_session *tls) { struct rm_priotracker prio; ktls_cleanup(tls); if (tls->be != NULL && ktls_allow_unload) { rm_rlock(&ktls_backends_lock, &prio); tls->be->use_count--; rm_runlock(&ktls_backends_lock, &prio); } uma_zfree(ktls_session_zone, tls); } void ktls_seq(struct sockbuf *sb, struct mbuf *m) { struct mbuf_ext_pgs *pgs; for (; m != NULL; m = m->m_next) { KASSERT((m->m_flags & M_NOMAP) != 0, ("ktls_seq: mapped mbuf %p", m)); pgs = &m->m_ext_pgs; pgs->seqno = sb->sb_tls_seqno; sb->sb_tls_seqno++; } } /* * Add TLS framing (headers and trailers) to a chain of mbufs. Each * mbuf in the chain must be an unmapped mbuf. The payload of the * mbuf must be populated with the payload of each TLS record. * * The record_type argument specifies the TLS record type used when * populating the TLS header. * * The enq_count argument on return is set to the number of pages of * payload data for this entire chain that need to be encrypted via SW * encryption. The returned value should be passed to ktls_enqueue * when scheduling encryption of this chain of mbufs. */ void ktls_frame(struct mbuf *top, struct ktls_session *tls, int *enq_cnt, uint8_t record_type) { struct tls_record_layer *tlshdr; struct mbuf *m; struct mbuf_ext_pgs *pgs; uint64_t *noncep; uint16_t tls_len; int maxlen; maxlen = tls->params.max_frame_len; *enq_cnt = 0; for (m = top; m != NULL; m = m->m_next) { /* * All mbufs in the chain should be non-empty TLS * records whose payload does not exceed the maximum * frame length. */ KASSERT(m->m_len <= maxlen && m->m_len > 0, ("ktls_frame: m %p len %d\n", m, m->m_len)); /* * TLS frames require unmapped mbufs to store session * info. */ KASSERT((m->m_flags & M_NOMAP) != 0, ("ktls_frame: mapped mbuf %p (top = %p)\n", m, top)); tls_len = m->m_len; pgs = &m->m_ext_pgs; /* Save a reference to the session. */ pgs->tls = ktls_hold(tls); pgs->hdr_len = tls->params.tls_hlen; pgs->trail_len = tls->params.tls_tlen; if (tls->params.cipher_algorithm == CRYPTO_AES_CBC) { int bs, delta; /* * AES-CBC pads messages to a multiple of the * block size. Note that the padding is * applied after the digest and the encryption * is done on the "plaintext || mac || padding". * At least one byte of padding is always * present. * * Compute the final trailer length assuming * at most one block of padding. * tls->params.sb_tls_tlen is the maximum * possible trailer length (padding + digest). * delta holds the number of excess padding * bytes if the maximum were used. Those * extra bytes are removed. */ bs = tls->params.tls_bs; delta = (tls_len + tls->params.tls_tlen) & (bs - 1); pgs->trail_len -= delta; } m->m_len += pgs->hdr_len + pgs->trail_len; /* Populate the TLS header. */ tlshdr = (void *)pgs->m_epg_hdr; tlshdr->tls_vmajor = tls->params.tls_vmajor; /* * TLS 1.3 masquarades as TLS 1.2 with a record type * of TLS_RLTYPE_APP. */ if (tls->params.tls_vminor == TLS_MINOR_VER_THREE && tls->params.tls_vmajor == TLS_MAJOR_VER_ONE) { tlshdr->tls_vminor = TLS_MINOR_VER_TWO; tlshdr->tls_type = TLS_RLTYPE_APP; /* save the real record type for later */ pgs->record_type = record_type; pgs->m_epg_trail[0] = record_type; } else { tlshdr->tls_vminor = tls->params.tls_vminor; tlshdr->tls_type = record_type; } tlshdr->tls_length = htons(m->m_len - sizeof(*tlshdr)); /* * Store nonces / explicit IVs after the end of the * TLS header. * * For GCM with TLS 1.2, an 8 byte nonce is copied * from the end of the IV. The nonce is then * incremented for use by the next record. * * For CBC, a random nonce is inserted for TLS 1.1+. */ if (tls->params.cipher_algorithm == CRYPTO_AES_NIST_GCM_16 && tls->params.tls_vminor == TLS_MINOR_VER_TWO) { noncep = (uint64_t *)(tls->params.iv + 8); be64enc(tlshdr + 1, *noncep); (*noncep)++; } else if (tls->params.cipher_algorithm == CRYPTO_AES_CBC && tls->params.tls_vminor >= TLS_MINOR_VER_ONE) arc4rand(tlshdr + 1, AES_BLOCK_LEN, 0); /* * When using SW encryption, mark the mbuf not ready. * It will be marked ready via sbready() after the * record has been encrypted. * * When using ifnet TLS, unencrypted TLS records are * sent down the stack to the NIC. */ if (tls->mode == TCP_TLS_MODE_SW) { m->m_flags |= M_NOTREADY; pgs->nrdy = pgs->npgs; *enq_cnt += pgs->npgs; } } } void ktls_enqueue_to_free(struct mbuf_ext_pgs *pgs) { struct ktls_wq *wq; bool running; /* Mark it for freeing. */ pgs->mbuf = NULL; wq = &ktls_wq[pgs->tls->wq_index]; mtx_lock(&wq->mtx); STAILQ_INSERT_TAIL(&wq->head, pgs, stailq); running = wq->running; mtx_unlock(&wq->mtx); if (!running) wakeup(wq); } void ktls_enqueue(struct mbuf *m, struct socket *so, int page_count) { struct mbuf_ext_pgs *pgs; struct ktls_wq *wq; bool running; KASSERT(((m->m_flags & (M_NOMAP | M_NOTREADY)) == (M_NOMAP | M_NOTREADY)), ("ktls_enqueue: %p not unready & nomap mbuf\n", m)); KASSERT(page_count != 0, ("enqueueing TLS mbuf with zero page count")); pgs = &m->m_ext_pgs; KASSERT(pgs->tls->mode == TCP_TLS_MODE_SW, ("!SW TLS mbuf")); pgs->enc_cnt = page_count; pgs->mbuf = m; /* * Save a pointer to the socket. The caller is responsible * for taking an additional reference via soref(). */ pgs->so = so; wq = &ktls_wq[pgs->tls->wq_index]; mtx_lock(&wq->mtx); STAILQ_INSERT_TAIL(&wq->head, pgs, stailq); running = wq->running; mtx_unlock(&wq->mtx); if (!running) wakeup(wq); counter_u64_add(ktls_cnt_on, 1); } static __noinline void ktls_encrypt(struct mbuf_ext_pgs *pgs) { struct ktls_session *tls; struct socket *so; struct mbuf *m, *top; vm_paddr_t parray[1 + btoc(TLS_MAX_MSG_SIZE_V10_2)]; struct iovec src_iov[1 + btoc(TLS_MAX_MSG_SIZE_V10_2)]; struct iovec dst_iov[1 + btoc(TLS_MAX_MSG_SIZE_V10_2)]; vm_page_t pg; int error, i, len, npages, off, total_pages; bool is_anon; so = pgs->so; tls = pgs->tls; top = pgs->mbuf; KASSERT(tls != NULL, ("tls = NULL, top = %p, pgs = %p\n", top, pgs)); KASSERT(so != NULL, ("so = NULL, top = %p, pgs = %p\n", top, pgs)); #ifdef INVARIANTS pgs->so = NULL; pgs->mbuf = NULL; #endif total_pages = pgs->enc_cnt; npages = 0; /* * Encrypt the TLS records in the chain of mbufs starting with * 'top'. 'total_pages' gives us a total count of pages and is * used to know when we have finished encrypting the TLS * records originally queued with 'top'. * * NB: These mbufs are queued in the socket buffer and * 'm_next' is traversing the mbufs in the socket buffer. The * socket buffer lock is not held while traversing this chain. * Since the mbufs are all marked M_NOTREADY their 'm_next' * pointers should be stable. However, the 'm_next' of the * last mbuf encrypted is not necessarily NULL. It can point * to other mbufs appended while 'top' was on the TLS work * queue. * * Each mbuf holds an entire TLS record. */ error = 0; for (m = top; npages != total_pages; m = m->m_next) { pgs = &m->m_ext_pgs; KASSERT(pgs->tls == tls, ("different TLS sessions in a single mbuf chain: %p vs %p", tls, pgs->tls)); KASSERT((m->m_flags & (M_NOMAP | M_NOTREADY)) == (M_NOMAP | M_NOTREADY), ("%p not unready & nomap mbuf (top = %p)\n", m, top)); KASSERT(npages + pgs->npgs <= total_pages, ("page count mismatch: top %p, total_pages %d, m %p", top, total_pages, m)); /* * Generate source and destination ivoecs to pass to * the SW encryption backend. For writable mbufs, the * destination iovec is a copy of the source and * encryption is done in place. For file-backed mbufs * (from sendfile), anonymous wired pages are * allocated and assigned to the destination iovec. */ is_anon = (pgs->flags & MBUF_PEXT_FLAG_ANON) != 0; off = pgs->first_pg_off; for (i = 0; i < pgs->npgs; i++, off = 0) { len = mbuf_ext_pg_len(pgs, i, off); src_iov[i].iov_len = len; src_iov[i].iov_base = (char *)(void *)PHYS_TO_DMAP(pgs->m_epg_pa[i]) + off; if (is_anon) { dst_iov[i].iov_base = src_iov[i].iov_base; dst_iov[i].iov_len = src_iov[i].iov_len; continue; } retry_page: pg = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | VM_ALLOC_NODUMP | VM_ALLOC_WIRED); if (pg == NULL) { vm_wait(NULL); goto retry_page; } parray[i] = VM_PAGE_TO_PHYS(pg); dst_iov[i].iov_base = (char *)(void *)PHYS_TO_DMAP(parray[i]) + off; dst_iov[i].iov_len = len; } npages += i; error = (*tls->sw_encrypt)(tls, (const struct tls_record_layer *)pgs->m_epg_hdr, pgs->m_epg_trail, src_iov, dst_iov, i, pgs->seqno, pgs->record_type); if (error) { counter_u64_add(ktls_offload_failed_crypto, 1); break; } /* * For file-backed mbufs, release the file-backed * pages and replace them in the ext_pgs array with * the anonymous wired pages allocated above. */ if (!is_anon) { /* Free the old pages. */ m->m_ext.ext_free(m); /* Replace them with the new pages. */ for (i = 0; i < pgs->npgs; i++) pgs->m_epg_pa[i] = parray[i]; /* Use the basic free routine. */ m->m_ext.ext_free = mb_free_mext_pgs; /* Pages are now writable. */ pgs->flags |= MBUF_PEXT_FLAG_ANON; } /* * Drop a reference to the session now that it is no * longer needed. Existing code depends on encrypted * records having no associated session vs * yet-to-be-encrypted records having an associated * session. */ pgs->tls = NULL; ktls_free(tls); } CURVNET_SET(so->so_vnet); if (error == 0) { (void)(*so->so_proto->pr_usrreqs->pru_ready)(so, top, npages); } else { so->so_proto->pr_usrreqs->pru_abort(so); so->so_error = EIO; mb_free_notready(top, total_pages); } SOCK_LOCK(so); sorele(so); CURVNET_RESTORE(); } static void ktls_work_thread(void *ctx) { struct ktls_wq *wq = ctx; struct mbuf_ext_pgs *p, *n; struct ktls_session *tls; struct mbuf *m; STAILQ_HEAD(, mbuf_ext_pgs) local_head; #if defined(__aarch64__) || defined(__amd64__) || defined(__i386__) fpu_kern_thread(0); #endif for (;;) { mtx_lock(&wq->mtx); while (STAILQ_EMPTY(&wq->head)) { wq->running = false; mtx_sleep(wq, &wq->mtx, 0, "-", 0); wq->running = true; } STAILQ_INIT(&local_head); STAILQ_CONCAT(&local_head, &wq->head); mtx_unlock(&wq->mtx); STAILQ_FOREACH_SAFE(p, &local_head, stailq, n) { if (p->mbuf != NULL) { ktls_encrypt(p); counter_u64_add(ktls_cnt_on, -1); } else { tls = p->tls; ktls_free(tls); m = __containerof(p, struct mbuf, m_ext_pgs); uma_zfree(zone_mbuf, m); } } } } Index: head/sys/net/radix_mpath.c =================================================================== --- head/sys/net/radix_mpath.c (revision 360291) +++ head/sys/net/radix_mpath.c (revision 360292) @@ -1,314 +1,315 @@ /* $KAME: radix_mpath.c,v 1.17 2004/11/08 10:29:39 itojun Exp $ */ /* * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2001 WIDE Project. * All rights reserved. * * 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 project 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 PROJECT 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 PROJECT 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. * THE AUTHORS DO NOT GUARANTEE THAT THIS SOFTWARE DOES NOT INFRINGE * ANY OTHERS' INTELLECTUAL PROPERTIES. IN NO EVENT SHALL THE AUTHORS * BE LIABLE FOR ANY INFRINGEMENT OF ANY OTHERS' INTELLECTUAL * PROPERTIES. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include /* * give some jitter to hash, to avoid synchronization between routers */ static uint32_t hashjitter; int rt_mpath_capable(struct rib_head *rnh) { return rnh->rnh_multipath; } int rn_mpath_capable(struct radix_head *rh) { return (rt_mpath_capable((struct rib_head *)rh)); } struct radix_node * rn_mpath_next(struct radix_node *rn) { struct radix_node *next; if (!rn->rn_dupedkey) return NULL; next = rn->rn_dupedkey; if (rn->rn_mask == next->rn_mask) return next; else return NULL; } uint32_t rn_mpath_count(struct radix_node *rn) { uint32_t i = 0; struct rtentry *rt; while (rn != NULL) { rt = (struct rtentry *)rn; i += rt->rt_weight; rn = rn_mpath_next(rn); } return (i); } struct rtentry * rt_mpath_matchgate(struct rtentry *rt, struct sockaddr *gate) { struct radix_node *rn; struct nhop_object *nh; if (gate == NULL) return (NULL); /* beyond here, we use rn as the master copy */ rn = (struct radix_node *)rt; do { rt = (struct rtentry *)rn; nh = rt->rt_nhop; /* * we are removing an address alias that has * the same prefix as another address * we need to compare the interface address because * gateway is a special sockaddr_dl structure */ if (nh->gw_sa.sa_family == AF_LINK) { if (!memcmp(nh->nh_ifa->ifa_addr, gate, gate->sa_len)) break; } /* * Check for other options: * 1) Routes with 'real' IPv4/IPv6 gateway * 2) Loopback host routes (another AF_LINK/sockadd_dl check) * */ if (nh->gw_sa.sa_len == gate->sa_len && !memcmp(&nh->gw_sa, gate, gate->sa_len)) break; } while ((rn = rn_mpath_next(rn)) != NULL); return (struct rtentry *)rn; } /* * go through the chain and unlink "rt" from the list * the caller will free "rt" */ int rt_mpath_deldup(struct rtentry *headrt, struct rtentry *rt) { struct radix_node *t, *tt; if (!headrt || !rt) return (0); t = (struct radix_node *)headrt; tt = rn_mpath_next(t); while (tt) { if (tt == (struct radix_node *)rt) { t->rn_dupedkey = tt->rn_dupedkey; tt->rn_dupedkey = NULL; tt->rn_flags &= ~RNF_ACTIVE; tt[1].rn_flags &= ~RNF_ACTIVE; return (1); } t = tt; tt = rn_mpath_next((struct radix_node *)t); } return (0); } /* * check if we have the same key/mask/gateway on the table already. * Assume @rt rt_key host bits are cleared according to @netmask */ int rt_mpath_conflict(struct rib_head *rnh, struct rtentry *rt, struct sockaddr *netmask) { struct radix_node *rn, *rn1; struct rtentry *rt1; rn = (struct radix_node *)rt; rn1 = rnh->rnh_lookup(rt_key(rt), netmask, &rnh->head); if (!rn1 || rn1->rn_flags & RNF_ROOT) return (0); /* key/mask are the same. compare gateway for all multipaths */ do { rt1 = (struct rtentry *)rn1; /* sanity: no use in comparing the same thing */ if (rn1 == rn) continue; if (rt1->rt_gateway->sa_family == AF_LINK) { if (rt1->rt_ifa->ifa_addr->sa_len != rt->rt_ifa->ifa_addr->sa_len || bcmp(rt1->rt_ifa->ifa_addr, rt->rt_ifa->ifa_addr, rt1->rt_ifa->ifa_addr->sa_len)) continue; } else { if (rt1->rt_gateway->sa_len != rt->rt_gateway->sa_len || bcmp(rt1->rt_gateway, rt->rt_gateway, rt1->rt_gateway->sa_len)) continue; } /* all key/mask/gateway are the same. conflicting entry. */ return (EEXIST); } while ((rn1 = rn_mpath_next(rn1)) != NULL); return (0); } struct rtentry * rt_mpath_selectrte(struct rtentry *rte, uint32_t hash) { struct radix_node *rn0, *rn; uint32_t total_weight; struct rtentry *rt; int64_t weight; /* beyond here, we use rn as the master copy */ rn0 = rn = (struct radix_node *)rte; rt = rte; /* gw selection by Modulo-N Hash (RFC2991) XXX need improvement? */ total_weight = rn_mpath_count(rn0); hash += hashjitter; hash %= total_weight; for (weight = abs((int32_t)hash); rt != NULL && weight >= rt->rt_weight; weight -= (rt == NULL) ? 0 : rt->rt_weight) { /* stay within the multipath routes */ if (rn->rn_dupedkey && rn->rn_mask != rn->rn_dupedkey->rn_mask) break; rn = rn->rn_dupedkey; rt = (struct rtentry *)rn; } return (rt); } struct rtentry * rt_mpath_select(struct rtentry *rte, uint32_t hash) { if (rn_mpath_next((struct radix_node *)rte) == NULL) return (rte); return (rt_mpath_selectrte(rte, hash)); } void rtalloc_mpath_fib(struct route *ro, uint32_t hash, u_int fibnum) { - struct rtentry *rt; + struct rtentry *rt, *rt_tmp; /* * XXX we don't attempt to lookup cached route again; what should * be done for sendto(3) case? */ - if (ro->ro_rt && ro->ro_rt->rt_ifp && (ro->ro_rt->rt_flags & RTF_UP) - && RT_LINK_IS_UP(ro->ro_rt->rt_ifp)) + if (ro->ro_nh && RT_LINK_IS_UP(ro->ro_nh->nh_ifp)) return; - ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, 0, fibnum); + ro->ro_nh = NULL; + rt_tmp = rtalloc1_fib(&ro->ro_dst, 1, 0, fibnum); /* if the route does not exist or it is not multipath, don't care */ - if (ro->ro_rt == NULL) + if (rt_tmp == NULL) return; - if (rn_mpath_next((struct radix_node *)ro->ro_rt) == NULL) { - RT_UNLOCK(ro->ro_rt); + if (rn_mpath_next((struct radix_node *)rt_tmp) == NULL) { + ro->ro_nh = rt_tmp->rt_nhop; + nhop_ref_object(ro->ro_nh); + RT_UNLOCK(rt_tmp); return; } - rt = rt_mpath_selectrte(ro->ro_rt, hash); + rt = rt_mpath_selectrte(rt_tmp, hash); /* XXX try filling rt_gwroute and avoid unreachable gw */ /* gw selection has failed - there must be only zero weight routes */ if (!rt) { - RT_UNLOCK(ro->ro_rt); - ro->ro_rt = NULL; + RT_UNLOCK(rt_tmp); return; } - if (ro->ro_rt != rt) { - RTFREE_LOCKED(ro->ro_rt); - ro->ro_rt = rt; - RT_LOCK(ro->ro_rt); - RT_ADDREF(ro->ro_rt); - - } - RT_UNLOCK(ro->ro_rt); + if (rt_tmp != rt) { + RTFREE_LOCKED(rt_tmp); + ro->ro_nh = rt->rt_nhop; + nhop_ref_object(ro->ro_nh); + } else + RT_UNLOCK(rt_tmp); } void rt_mpath_init_rnh(struct rib_head *rnh) { rnh->rnh_multipath = 1; } #ifdef RADIX_MPATH static void mpath_init(void) { hashjitter = arc4random(); } SYSINIT(mpath_init, SI_SUB_LAST, SI_ORDER_ANY, mpath_init, NULL); #endif Index: head/sys/net/route.c =================================================================== --- head/sys/net/route.c (revision 360291) +++ head/sys/net/route.c (revision 360292) @@ -1,2387 +1,2390 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1980, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)route.c 8.3.1.1 (Berkeley) 2/23/95 * $FreeBSD$ */ /************************************************************************ * Note: In this file a 'fib' is a "forwarding information base" * * Which is the new name for an in kernel routing (next hop) table. * ***********************************************************************/ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_mrouting.h" #include "opt_mpath.h" #include "opt_route.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include #include #include #define RT_MAXFIBS UINT16_MAX /* Kernel config default option. */ #ifdef ROUTETABLES #if ROUTETABLES <= 0 #error "ROUTETABLES defined too low" #endif #if ROUTETABLES > RT_MAXFIBS #error "ROUTETABLES defined too big" #endif #define RT_NUMFIBS ROUTETABLES #endif /* ROUTETABLES */ /* Initialize to default if not otherwise set. */ #ifndef RT_NUMFIBS #define RT_NUMFIBS 1 #endif /* This is read-only.. */ u_int rt_numfibs = RT_NUMFIBS; SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, ""); /* * By default add routes to all fibs for new interfaces. * Once this is set to 0 then only allocate routes on interface * changes for the FIB of the caller when adding a new set of addresses * to an interface. XXX this is a shotgun aproach to a problem that needs * a more fine grained solution.. that will come. * XXX also has the problems getting the FIB from curthread which will not * always work given the fib can be overridden and prefixes can be added * from the network stack context. */ VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1; SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(rt_add_addr_allfibs), 0, ""); VNET_PCPUSTAT_DEFINE(struct rtstat, rtstat); VNET_PCPUSTAT_SYSINIT(rtstat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(rtstat); #endif VNET_DEFINE(struct rib_head *, rt_tables); #define V_rt_tables VNET(rt_tables) VNET_DEFINE(int, rttrash); /* routes not in table but not freed */ #define V_rttrash VNET(rttrash) /* * Convert a 'struct radix_node *' to a 'struct rtentry *'. * The operation can be done safely (in this code) because a * 'struct rtentry' starts with two 'struct radix_node''s, the first * one representing leaf nodes in the routing tree, which is * what the code in radix.c passes us as a 'struct radix_node'. * * But because there are a lot of assumptions in this conversion, * do not cast explicitly, but always use the macro below. */ #define RNTORT(p) ((struct rtentry *)(p)) VNET_DEFINE_STATIC(uma_zone_t, rtzone); /* Routing table UMA zone. */ #define V_rtzone VNET(rtzone) EVENTHANDLER_LIST_DEFINE(rt_addrmsg); static int rt_getifa_fib(struct rt_addrinfo *, u_int); static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *); static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *, void *arg); static struct rtentry *rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror); static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info); #ifdef RADIX_MPATH static struct radix_node *rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry *rto, int *perror); #endif static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags); static int add_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt); static int del_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt); static int change_route(struct rib_head *, struct rt_addrinfo *, struct rtentry **); struct if_mtuinfo { struct ifnet *ifp; int mtu; }; static int if_updatemtu_cb(struct radix_node *, void *); /* * handler for net.my_fibnum */ static int sysctl_my_fibnum(SYSCTL_HANDLER_ARGS) { int fibnum; int error; fibnum = curthread->td_proc->p_fibnum; error = sysctl_handle_int(oidp, &fibnum, 0, req); return (error); } SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller"); static __inline struct rib_head ** rt_tables_get_rnh_ptr(int table, int fam) { struct rib_head **rnh; KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds (0 <= %d < %d)", __func__, table, rt_numfibs)); KASSERT(fam >= 0 && fam < (AF_MAX + 1), ("%s: fam out of bounds (0 <= %d < %d)", __func__, fam, AF_MAX+1)); /* rnh is [fib=0][af=0]. */ rnh = (struct rib_head **)V_rt_tables; /* Get the offset to the requested table and fam. */ rnh += table * (AF_MAX+1) + fam; return (rnh); } struct rib_head * rt_tables_get_rnh(int table, int fam) { return (*rt_tables_get_rnh_ptr(table, fam)); } u_int rt_tables_get_gen(int table, int fam) { struct rib_head *rnh; rnh = *rt_tables_get_rnh_ptr(table, fam); KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d", __func__, table, fam)); return (rnh->rnh_gen); } /* * route initialization must occur before ip6_init2(), which happenas at * SI_ORDER_MIDDLE. */ static void route_init(void) { /* whack the tunable ints into line. */ if (rt_numfibs > RT_MAXFIBS) rt_numfibs = RT_MAXFIBS; if (rt_numfibs == 0) rt_numfibs = 1; nhops_init(); } SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL); static int rtentry_zinit(void *mem, int size, int how) { struct rtentry *rt = mem; rt->rt_pksent = counter_u64_alloc(how); if (rt->rt_pksent == NULL) return (ENOMEM); RT_LOCK_INIT(rt); return (0); } static void rtentry_zfini(void *mem, int size) { struct rtentry *rt = mem; RT_LOCK_DESTROY(rt); counter_u64_free(rt->rt_pksent); } static int rtentry_ctor(void *mem, int size, void *arg, int how) { struct rtentry *rt = mem; bzero(rt, offsetof(struct rtentry, rt_endzero)); counter_u64_zero(rt->rt_pksent); rt->rt_chain = NULL; return (0); } static void rtentry_dtor(void *mem, int size, void *arg) { struct rtentry *rt = mem; RT_UNLOCK_COND(rt); } static void vnet_route_init(const void *unused __unused) { struct domain *dom; struct rib_head **rnh; int table; int fam; V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) * sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO); V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), rtentry_ctor, rtentry_dtor, rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0); for (dom = domains; dom; dom = dom->dom_next) { if (dom->dom_rtattach == NULL) continue; for (table = 0; table < rt_numfibs; table++) { fam = dom->dom_family; if (table != 0 && fam != AF_INET6 && fam != AF_INET) break; rnh = rt_tables_get_rnh_ptr(table, fam); if (rnh == NULL) panic("%s: rnh NULL", __func__); dom->dom_rtattach((void **)rnh, 0, table); } } } VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, vnet_route_init, 0); #ifdef VIMAGE static void vnet_route_uninit(const void *unused __unused) { int table; int fam; struct domain *dom; struct rib_head **rnh; for (dom = domains; dom; dom = dom->dom_next) { if (dom->dom_rtdetach == NULL) continue; for (table = 0; table < rt_numfibs; table++) { fam = dom->dom_family; if (table != 0 && fam != AF_INET6 && fam != AF_INET) break; rnh = rt_tables_get_rnh_ptr(table, fam); if (rnh == NULL) panic("%s: rnh NULL", __func__); dom->dom_rtdetach((void **)rnh, 0); } } free(V_rt_tables, M_RTABLE); uma_zdestroy(V_rtzone); } VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, vnet_route_uninit, 0); #endif struct rib_head * rt_table_init(int offset, int family, u_int fibnum) { struct rib_head *rh; rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO); /* TODO: These details should be hidded inside radix.c */ /* Init masks tree */ rn_inithead_internal(&rh->head, rh->rnh_nodes, offset); rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0); rh->head.rnh_masks = &rh->rmhead; /* Save metadata associated with this routing table. */ rh->rib_family = family; rh->rib_fibnum = fibnum; #ifdef VIMAGE rh->rib_vnet = curvnet; #endif tmproutes_init(rh); /* Init locks */ RIB_LOCK_INIT(rh); nhops_init_rib(rh); /* Finally, set base callbacks */ rh->rnh_addaddr = rn_addroute; rh->rnh_deladdr = rn_delete; rh->rnh_matchaddr = rn_match; rh->rnh_lookup = rn_lookup; rh->rnh_walktree = rn_walktree; rh->rnh_walktree_from = rn_walktree_from; return (rh); } static int rt_freeentry(struct radix_node *rn, void *arg) { struct radix_head * const rnh = arg; struct radix_node *x; x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh); if (x != NULL) R_Free(x); return (0); } void rt_table_destroy(struct rib_head *rh) { tmproutes_destroy(rh); rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head); nhops_destroy_rib(rh); /* Assume table is already empty */ RIB_LOCK_DESTROY(rh); free(rh, M_RTABLE); } #ifndef _SYS_SYSPROTO_H_ struct setfib_args { int fibnum; }; #endif int sys_setfib(struct thread *td, struct setfib_args *uap) { if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs) return EINVAL; td->td_proc->p_fibnum = uap->fibnum; return (0); } /* * Packet routing routines. */ void rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum) { struct rtentry *rt; - if ((rt = ro->ro_rt) != NULL) { - if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP) + if (ro->ro_nh != NULL) { + if (NH_IS_VALID(ro->ro_nh)) return; - RTFREE(rt); - ro->ro_rt = NULL; + NH_FREE(ro->ro_nh); + ro->ro_nh = NULL; } - ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum); - if (ro->ro_rt) - RT_UNLOCK(ro->ro_rt); + rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum); + if (rt != NULL) { + ro->ro_nh = rt->rt_nhop; + nhop_ref_object(rt->rt_nhop); + RT_UNLOCK(rt); + } } /* * Look up the route that matches the address given * Or, at least try.. Create a cloned route if needed. * * The returned route, if any, is locked. */ struct rtentry * rtalloc1(struct sockaddr *dst, int report, u_long ignflags) { return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB)); } struct rtentry * rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *newrt; struct rt_addrinfo info; int err = 0, msgtype = RTM_MISS; KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum")); rh = rt_tables_get_rnh(fibnum, dst->sa_family); newrt = NULL; if (rh == NULL) goto miss; /* * Look up the address in the table for that Address Family */ if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RLOCK(rh); #ifdef INVARIANTS else RIB_LOCK_ASSERT(rh); #endif rn = rh->rnh_matchaddr(dst, &rh->head); if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) { newrt = RNTORT(rn); RT_LOCK(newrt); RT_ADDREF(newrt); if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RUNLOCK(rh); return (newrt); } else if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RUNLOCK(rh); /* * Either we hit the root or could not find any match, * which basically means: "cannot get there from here". */ miss: RTSTAT_INC(rts_unreach); if (report) { /* * If required, report the failure to the supervising * Authorities. * For a delete, this is not an error. (report == 0) */ bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; rt_missmsg_fib(msgtype, &info, 0, err, fibnum); } return (newrt); } /* * Remove a reference count from an rtentry. * If the count gets low enough, take it out of the routing table */ void rtfree(struct rtentry *rt) { struct rib_head *rnh; KASSERT(rt != NULL,("%s: NULL rt", __func__)); rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family); KASSERT(rnh != NULL,("%s: NULL rnh", __func__)); RT_LOCK_ASSERT(rt); /* * The callers should use RTFREE_LOCKED() or RTFREE(), so * we should come here exactly with the last reference. */ RT_REMREF(rt); if (rt->rt_refcnt > 0) { log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt); goto done; } /* * On last reference give the "close method" a chance * to cleanup private state. This also permits (for * IPv4 and IPv6) a chance to decide if the routing table * entry should be purged immediately or at a later time. * When an immediate purge is to happen the close routine * typically calls rtexpunge which clears the RTF_UP flag * on the entry so that the code below reclaims the storage. */ if (rt->rt_refcnt == 0 && rnh->rnh_close) rnh->rnh_close((struct radix_node *)rt, &rnh->head); /* * If we are no longer "up" (and ref == 0) * then we can free the resources associated * with the route. */ if ((rt->rt_flags & RTF_UP) == 0) { if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT)) panic("rtfree 2"); /* * the rtentry must have been removed from the routing table * so it is represented in rttrash.. remove that now. */ V_rttrash--; #ifdef DIAGNOSTIC if (rt->rt_refcnt < 0) { printf("rtfree: %p not freed (neg refs)\n", rt); goto done; } #endif /* * release references on items we hold them on.. * e.g other routes and ifaddrs. */ if (rt->rt_ifa) ifa_free(rt->rt_ifa); /* * The key is separatly alloc'd so free it (see rt_setgate()). * This also frees the gateway, as they are always malloc'd * together. */ R_Free(rt_key(rt)); /* Unreference nexthop */ nhop_free(rt->rt_nhop); /* * and the rtentry itself of course */ uma_zfree(V_rtzone, rt); return; } done: RT_UNLOCK(rt); } /* * Temporary RTFREE() function wrapper. * Intended to use in control plane code to * avoid exposing internal layout of 'struct rtentry'. */ void rtfree_func(struct rtentry *rt) { RTFREE(rt); } /* * Adds a temporal redirect entry to the routing table. * @fibnum: fib number * @dst: destination to install redirect to * @gateway: gateway to go via * @author: sockaddr of originating router, can be NULL * @ifp: interface to use for the redirected route * @flags: set of flags to add. Allowed: RTF_GATEWAY * @lifetime_sec: time in seconds to expire this redirect. * * Retuns 0 on success, errno otherwise. */ int rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway, struct sockaddr *author, struct ifnet *ifp, int flags, int lifetime_sec) { struct rtentry *rt; int error; struct rt_addrinfo info; struct rt_metrics rti_rmx; struct ifaddr *ifa; NET_EPOCH_ASSERT(); if (rt_tables_get_rnh(fibnum, dst->sa_family) == NULL) return (EAFNOSUPPORT); /* Verify the allowed flag mask. */ KASSERT(((flags & ~(RTF_GATEWAY)) == 0), ("invalid redirect flags: %x", flags)); /* Get the best ifa for the given interface and gateway. */ if ((ifa = ifaof_ifpforaddr(gateway, ifp)) == NULL) return (ENETUNREACH); ifa_ref(ifa); bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_ifa = ifa; info.rti_ifp = ifp; info.rti_flags = flags | RTF_HOST | RTF_DYNAMIC; /* Setup route metrics to define expire time. */ bzero(&rti_rmx, sizeof(rti_rmx)); /* Set expire time as absolute. */ rti_rmx.rmx_expire = lifetime_sec + time_second; info.rti_mflags |= RTV_EXPIRE; info.rti_rmx = &rti_rmx; error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum); ifa_free(ifa); if (error != 0) { /* TODO: add per-fib redirect stats. */ return (error); } RT_LOCK(rt); flags = rt->rt_flags; RTFREE_LOCKED(rt); RTSTAT_INC(rts_dynamic); /* Send notification of a route addition to userland. */ bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_info[RTAX_AUTHOR] = author; rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum); return (0); } /* * Routing table ioctl interface. */ int rtioctl_fib(u_long req, caddr_t data, u_int fibnum) { /* * If more ioctl commands are added here, make sure the proper * super-user checks are being performed because it is possible for * prison-root to make it this far if raw sockets have been enabled * in jails. */ #ifdef INET /* Multicast goop, grrr... */ return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP; #else /* INET */ return ENXIO; #endif /* INET */ } struct ifaddr * ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway, u_int fibnum) { struct ifaddr *ifa; int not_found = 0; NET_EPOCH_ASSERT(); if ((flags & RTF_GATEWAY) == 0) { /* * If we are adding a route to an interface, * and the interface is a pt to pt link * we should search for the destination * as our clue to the interface. Otherwise * we can use the local address. */ ifa = NULL; if (flags & RTF_HOST) ifa = ifa_ifwithdstaddr(dst, fibnum); if (ifa == NULL) ifa = ifa_ifwithaddr(gateway); } else { /* * If we are adding a route to a remote net * or host, the gateway may still be on the * other end of a pt to pt link. */ ifa = ifa_ifwithdstaddr(gateway, fibnum); } if (ifa == NULL) ifa = ifa_ifwithnet(gateway, 0, fibnum); if (ifa == NULL) { struct rtentry *rt; rt = rtalloc1_fib(gateway, 0, flags, fibnum); if (rt == NULL) goto out; /* * dismiss a gateway that is reachable only * through the default router */ switch (gateway->sa_family) { case AF_INET: if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) not_found = 1; break; case AF_INET6: if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr)) not_found = 1; break; default: break; } if (!not_found && rt->rt_ifa != NULL) { ifa = rt->rt_ifa; } RT_REMREF(rt); RT_UNLOCK(rt); if (not_found || ifa == NULL) goto out; } if (ifa->ifa_addr->sa_family != dst->sa_family) { struct ifaddr *oifa = ifa; ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); if (ifa == NULL) ifa = oifa; } out: return (ifa); } /* * Do appropriate manipulations of a routing tree given * all the bits of info needed */ int rtrequest_fib(int req, struct sockaddr *dst, struct sockaddr *gateway, struct sockaddr *netmask, int flags, struct rtentry **ret_nrt, u_int fibnum) { struct rt_addrinfo info; if (dst->sa_len == 0) return(EINVAL); bzero((caddr_t)&info, sizeof(info)); info.rti_flags = flags; info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_info[RTAX_NETMASK] = netmask; return rtrequest1_fib(req, &info, ret_nrt, fibnum); } /* * Copy most of @rt data into @info. * * If @flags contains NHR_COPY, copies dst,netmask and gw to the * pointers specified by @info structure. Assume such pointers * are zeroed sockaddr-like structures with sa_len field initialized * to reflect size of the provided buffer. if no NHR_COPY is specified, * point dst,netmask and gw @info fields to appropriate @rt values. * * if @flags contains NHR_REF, do refcouting on rt_ifp and rt_ifa. * * Returns 0 on success. */ int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags) { struct rt_metrics *rmx; struct sockaddr *src, *dst; struct nhop_object *nh; int sa_len; if (flags & NHR_COPY) { /* Copy destination if dst is non-zero */ src = rt_key(rt); dst = info->rti_info[RTAX_DST]; sa_len = src->sa_len; if (dst != NULL) { if (src->sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_DST; } /* Copy mask if set && dst is non-zero */ src = rt_mask(rt); dst = info->rti_info[RTAX_NETMASK]; if (src != NULL && dst != NULL) { /* * Radix stores different value in sa_len, * assume rt_mask() to have the same length * as rt_key() */ if (sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_NETMASK; } /* Copy gateway is set && dst is non-zero */ src = &rt->rt_nhop->gw_sa; dst = info->rti_info[RTAX_GATEWAY]; if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){ if (src->sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_GATEWAY; } } else { info->rti_info[RTAX_DST] = rt_key(rt); info->rti_addrs |= RTA_DST; if (rt_mask(rt) != NULL) { info->rti_info[RTAX_NETMASK] = rt_mask(rt); info->rti_addrs |= RTA_NETMASK; } if (rt->rt_flags & RTF_GATEWAY) { info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa; info->rti_addrs |= RTA_GATEWAY; } } nh = rt->rt_nhop; rmx = info->rti_rmx; if (rmx != NULL) { info->rti_mflags |= RTV_MTU; rmx->rmx_mtu = nh->nh_mtu; } info->rti_flags = rt->rt_flags | nhop_get_rtflags(nh); info->rti_ifp = nh->nh_ifp; info->rti_ifa = nh->nh_ifa; if (flags & NHR_REF) { if_ref(info->rti_ifp); ifa_ref(info->rti_ifa); } return (0); } /* * Lookups up route entry for @dst in RIB database for fib @fibnum. * Exports entry data to @info using rt_exportinfo(). * * If @flags contains NHR_REF, refcouting is performed on rt_ifp and rt_ifa. * All references can be released later by calling rib_free_info(). * * Returns 0 on success. * Returns ENOENT for lookup failure, ENOMEM for export failure. */ int rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags, uint32_t flowid, struct rt_addrinfo *info) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *rt; int error; KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum")); rh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rh == NULL) return (ENOENT); RIB_RLOCK(rh); rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { rt = RNTORT(rn); /* Ensure route & ifp is UP */ if (RT_LINK_IS_UP(rt->rt_ifp)) { flags = (flags & NHR_REF) | NHR_COPY; error = rt_exportinfo(rt, info, flags); RIB_RUNLOCK(rh); return (error); } } RIB_RUNLOCK(rh); return (ENOENT); } /* * Releases all references acquired by rib_lookup_info() when * called with NHR_REF flags. */ void rib_free_info(struct rt_addrinfo *info) { ifa_free(info->rti_ifa); if_rele(info->rti_ifp); } /* * Iterates over all existing fibs in system calling * @setwa_f function prior to traversing each fib. * Calls @wa_f function for each element in current fib. * If af is not AF_UNSPEC, iterates over fibs in particular * address family. */ void rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f, void *arg) { struct rib_head *rnh; uint32_t fibnum; int i; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { /* Do we want some specific family? */ if (af != AF_UNSPEC) { rnh = rt_tables_get_rnh(fibnum, af); if (rnh == NULL) continue; if (setwa_f != NULL) setwa_f(rnh, fibnum, af, arg); RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg); RIB_WUNLOCK(rnh); continue; } for (i = 1; i <= AF_MAX; i++) { rnh = rt_tables_get_rnh(fibnum, i); if (rnh == NULL) continue; if (setwa_f != NULL) setwa_f(rnh, fibnum, i, arg); RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg); RIB_WUNLOCK(rnh); } } } struct rt_delinfo { struct rt_addrinfo info; struct rib_head *rnh; struct rtentry *head; }; /* * Conditionally unlinks @rn from radix tree based * on info data passed in @arg. */ static int rt_checkdelroute(struct radix_node *rn, void *arg) { struct rt_delinfo *di; struct rt_addrinfo *info; struct rtentry *rt; int error; di = (struct rt_delinfo *)arg; rt = (struct rtentry *)rn; info = &di->info; error = 0; info->rti_info[RTAX_DST] = rt_key(rt); info->rti_info[RTAX_NETMASK] = rt_mask(rt); info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa; rt = rt_unlinkrte(di->rnh, info, &error); if (rt == NULL) { /* Either not allowed or not matched. Skip entry */ return (0); } /* Entry was unlinked. Add to the list and return */ rt->rt_chain = di->head; di->head = rt; return (0); } /* * Iterates over a routing table specified by @fibnum and @family and * deletes elements marked by @filter_f. * @fibnum: rtable id * @family: AF_ address family * @filter_f: function returning non-zero value for items to delete * @arg: data to pass to the @filter_f function * @report: true if rtsock notification is needed. */ void rib_walk_del(u_int fibnum, int family, rt_filter_f_t *filter_f, void *arg, bool report) { struct rib_head *rnh; struct rt_delinfo di; struct rtentry *rt; rnh = rt_tables_get_rnh(fibnum, family); if (rnh == NULL) return; bzero(&di, sizeof(di)); di.info.rti_filter = filter_f; di.info.rti_filterdata = arg; di.rnh = rnh; RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di); RIB_WUNLOCK(rnh); if (di.head == NULL) return; /* We might have something to reclaim. */ while (di.head != NULL) { rt = di.head; di.head = rt->rt_chain; rt->rt_chain = NULL; /* TODO std rt -> rt_addrinfo export */ di.info.rti_info[RTAX_DST] = rt_key(rt); di.info.rti_info[RTAX_NETMASK] = rt_mask(rt); rt_notifydelete(rt, &di.info); if (report) rt_routemsg(RTM_DELETE, rt, rt->rt_ifp, 0, fibnum); RTFREE_LOCKED(rt); } } /* * Iterates over all existing fibs in system and deletes each element * for which @filter_f function returns non-zero value. * If @family is not AF_UNSPEC, iterates over fibs in particular * address family. */ void rt_foreach_fib_walk_del(int family, rt_filter_f_t *filter_f, void *arg) { u_int fibnum; int i, start, end; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { /* Do we want some specific family? */ if (family != AF_UNSPEC) { start = family; end = family; } else { start = 1; end = AF_MAX; } for (i = start; i <= end; i++) { if (rt_tables_get_rnh(fibnum, i) == NULL) continue; rib_walk_del(fibnum, i, filter_f, arg, 0); } } } /* * Delete Routes for a Network Interface * * Called for each routing entry via the rnh->rnh_walktree() call above * to delete all route entries referencing a detaching network interface. * * Arguments: * rt pointer to rtentry * nh pointer to nhop * arg argument passed to rnh->rnh_walktree() - detaching interface * * Returns: * 0 successful * errno failed - reason indicated */ static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *nh, void *arg) { struct ifnet *ifp = arg; if (nh->nh_ifp != ifp) return (0); /* * Protect (sorta) against walktree recursion problems * with cloned routes */ if ((rt->rt_flags & RTF_UP) == 0) return (0); return (1); } /* * Delete all remaining routes using this interface * Unfortuneatly the only way to do this is to slog through * the entire routing table looking for routes which point * to this interface...oh well... */ void rt_flushifroutes_af(struct ifnet *ifp, int af) { KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d", __func__, af, AF_MAX)); rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp); } void rt_flushifroutes(struct ifnet *ifp) { rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp); } /* * Conditionally unlinks rtentry matching data inside @info from @rnh. * Returns unlinked, locked and referenced @rtentry on success, * Returns NULL and sets @perror to: * ESRCH - if prefix was not found, * EADDRINUSE - if trying to delete PINNED route without appropriate flag. * ENOENT - if supplied filter function returned 0 (not matched). */ static struct rtentry * rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror) { struct sockaddr *dst, *netmask; struct rtentry *rt; struct radix_node *rn; dst = info->rti_info[RTAX_DST]; netmask = info->rti_info[RTAX_NETMASK]; rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head); if (rt == NULL) { *perror = ESRCH; return (NULL); } if ((info->rti_flags & RTF_PINNED) == 0) { /* Check if target route can be deleted */ if (rt->rt_flags & RTF_PINNED) { *perror = EADDRINUSE; return (NULL); } } if (info->rti_filter != NULL) { if (info->rti_filter(rt, rt->rt_nhop, info->rti_filterdata)==0){ /* Not matched */ *perror = ENOENT; return (NULL); } /* * Filter function requested rte deletion. * Ease the caller work by filling in remaining info * from that particular entry. */ info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa; } /* * Remove the item from the tree and return it. * Complain if it is not there and do no more processing. */ *perror = ESRCH; #ifdef RADIX_MPATH if (rt_mpath_capable(rnh)) rn = rt_mpath_unlink(rnh, info, rt, perror); else #endif rn = rnh->rnh_deladdr(dst, netmask, &rnh->head); if (rn == NULL) return (NULL); if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) panic ("rtrequest delete"); rt = RNTORT(rn); RT_LOCK(rt); RT_ADDREF(rt); rt->rt_flags &= ~RTF_UP; *perror = 0; return (rt); } static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info) { struct ifaddr *ifa; /* * give the protocol a chance to keep things in sync. */ ifa = rt->rt_ifa; if (ifa != NULL && ifa->ifa_rtrequest != NULL) ifa->ifa_rtrequest(RTM_DELETE, rt, info); /* * One more rtentry floating around that is not * linked to the routing table. rttrash will be decremented * when RTFREE(rt) is eventually called. */ V_rttrash++; } /* * These (questionable) definitions of apparent local variables apply * to the next two functions. XXXXXX!!! */ #define dst info->rti_info[RTAX_DST] #define gateway info->rti_info[RTAX_GATEWAY] #define netmask info->rti_info[RTAX_NETMASK] #define ifaaddr info->rti_info[RTAX_IFA] #define ifpaddr info->rti_info[RTAX_IFP] #define flags info->rti_flags /* * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined, * it will be referenced so the caller must free it. * * Assume basic consistency checks are executed by callers: * RTAX_DST exists, if RTF_GATEWAY is set, RTAX_GATEWAY exists as well. */ int rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum) { struct epoch_tracker et; int needref, error; /* * ifp may be specified by sockaddr_dl * when protocol address is ambiguous. */ error = 0; needref = (info->rti_ifa == NULL); NET_EPOCH_ENTER(et); /* If we have interface specified by the ifindex in the address, use it */ if (info->rti_ifp == NULL && ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)ifpaddr; if (sdl->sdl_index != 0) info->rti_ifp = ifnet_byindex(sdl->sdl_index); } /* * If we have source address specified, try to find it * TODO: avoid enumerating all ifas on all interfaces. */ if (info->rti_ifa == NULL && ifaaddr != NULL) info->rti_ifa = ifa_ifwithaddr(ifaaddr); if (info->rti_ifa == NULL) { struct sockaddr *sa; /* * Most common use case for the userland-supplied routes. * * Choose sockaddr to select ifa. * -- if ifp is set -- * Order of preference: * 1) IFA address * 2) gateway address * Note: for interface routes link-level gateway address * is specified to indicate the interface index without * specifying RTF_GATEWAY. In this case, ignore gateway * Note: gateway AF may be different from dst AF. In this case, * ignore gateway * 3) final destination. * 4) if all of these fails, try to get at least link-level ifa. * -- else -- * try to lookup gateway or dst in the routing table to get ifa */ if (info->rti_info[RTAX_IFA] != NULL) sa = info->rti_info[RTAX_IFA]; else if ((info->rti_flags & RTF_GATEWAY) != 0 && gateway->sa_family == dst->sa_family) sa = gateway; else sa = dst; if (info->rti_ifp != NULL) { info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp); /* Case 4 */ if (info->rti_ifa == NULL && gateway != NULL) info->rti_ifa = ifaof_ifpforaddr(gateway, info->rti_ifp); } else if (dst != NULL && gateway != NULL) info->rti_ifa = ifa_ifwithroute(flags, dst, gateway, fibnum); else if (sa != NULL) info->rti_ifa = ifa_ifwithroute(flags, sa, sa, fibnum); } if (needref && info->rti_ifa != NULL) { if (info->rti_ifp == NULL) info->rti_ifp = info->rti_ifa->ifa_ifp; ifa_ref(info->rti_ifa); } else error = ENETUNREACH; NET_EPOCH_EXIT(et); return (error); } static int if_updatemtu_cb(struct radix_node *rn, void *arg) { struct rtentry *rt; struct if_mtuinfo *ifmtu; rt = (struct rtentry *)rn; ifmtu = (struct if_mtuinfo *)arg; if (rt->rt_ifp != ifmtu->ifp) return (0); if (rt->rt_mtu >= ifmtu->mtu) { /* We have to decrease mtu regardless of flags */ rt->rt_mtu = ifmtu->mtu; return (0); } /* * New MTU is bigger. Check if are allowed to alter it */ if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) { /* * Skip routes with user-supplied MTU and * non-interface routes */ return (0); } /* We are safe to update route MTU */ rt->rt_mtu = ifmtu->mtu; return (0); } void rt_updatemtu(struct ifnet *ifp) { struct if_mtuinfo ifmtu; struct rib_head *rnh; int i, j; ifmtu.ifp = ifp; /* * Try to update rt_mtu for all routes using this interface * Unfortunately the only way to do this is to traverse all * routing tables in all fibs/domains. */ for (i = 1; i <= AF_MAX; i++) { ifmtu.mtu = if_getmtu_family(ifp, i); for (j = 0; j < rt_numfibs; j++) { rnh = rt_tables_get_rnh(j, i); if (rnh == NULL) continue; RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu); RIB_WUNLOCK(rnh); nhops_update_ifmtu(rnh, ifp, ifmtu.mtu); } } } #if 0 int p_sockaddr(char *buf, int buflen, struct sockaddr *s); int rt_print(char *buf, int buflen, struct rtentry *rt); int p_sockaddr(char *buf, int buflen, struct sockaddr *s) { void *paddr = NULL; switch (s->sa_family) { case AF_INET: paddr = &((struct sockaddr_in *)s)->sin_addr; break; case AF_INET6: paddr = &((struct sockaddr_in6 *)s)->sin6_addr; break; } if (paddr == NULL) return (0); if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL) return (0); return (strlen(buf)); } int rt_print(char *buf, int buflen, struct rtentry *rt) { struct sockaddr *addr, *mask; int i = 0; addr = rt_key(rt); mask = rt_mask(rt); i = p_sockaddr(buf, buflen, addr); if (!(rt->rt_flags & RTF_HOST)) { buf[i++] = '/'; i += p_sockaddr(buf + i, buflen - i, mask); } if (rt->rt_flags & RTF_GATEWAY) { buf[i++] = '>'; i += p_sockaddr(buf + i, buflen - i, &rt->rt_nhop->gw_sa); } return (i); } #endif #ifdef RADIX_MPATH /* * Deletes key for single-path routes, unlinks rtentry with * gateway specified in @info from multi-path routes. * * Returnes unlinked entry. In case of failure, returns NULL * and sets @perror to ESRCH. */ static struct radix_node * rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry *rto, int *perror) { /* * if we got multipath routes, we require users to specify * a matching RTAX_GATEWAY. */ struct rtentry *rt; // *rto = NULL; struct radix_node *rn; struct sockaddr *gw; gw = info->rti_info[RTAX_GATEWAY]; rt = rt_mpath_matchgate(rto, gw); if (rt == NULL) { *perror = ESRCH; return (NULL); } /* * this is the first entry in the chain */ if (rto == rt) { rn = rn_mpath_next((struct radix_node *)rt); /* * there is another entry, now it's active */ if (rn) { rto = RNTORT(rn); RT_LOCK(rto); rto->rt_flags |= RTF_UP; RT_UNLOCK(rto); } else if (rt->rt_flags & RTF_GATEWAY) { /* * For gateway routes, we need to * make sure that we we are deleting * the correct gateway. * rt_mpath_matchgate() does not * check the case when there is only * one route in the chain. */ if (gw && (rt->rt_nhop->gw_sa.sa_len != gw->sa_len || memcmp(&rt->rt_nhop->gw_sa, gw, gw->sa_len))) { *perror = ESRCH; return (NULL); } } /* * use the normal delete code to remove * the first entry */ rn = rnh->rnh_deladdr(dst, netmask, &rnh->head); *perror = 0; return (rn); } /* * if the entry is 2nd and on up */ if (rt_mpath_deldup(rto, rt) == 0) panic ("rtrequest1: rt_mpath_deldup"); *perror = 0; rn = (struct radix_node *)rt; return (rn); } #endif #undef dst #undef gateway #undef netmask #undef ifaaddr #undef ifpaddr #undef flags int rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt, u_int fibnum) { struct epoch_tracker et; const struct sockaddr *dst; struct rib_head *rnh; int error; KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum")); KASSERT((info->rti_flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked")); dst = info->rti_info[RTAX_DST]; switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We support multiple FIBs. */ break; default: fibnum = RT_DEFAULT_FIB; break; } /* * Find the correct routing tree to use for this Address Family */ rnh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rnh == NULL) return (EAFNOSUPPORT); /* * If we are adding a host route then we don't want to put * a netmask in the tree, nor do we want to clone it. */ if (info->rti_flags & RTF_HOST) info->rti_info[RTAX_NETMASK] = NULL; error = 0; switch (req) { case RTM_DELETE: error = del_route(rnh, info, ret_nrt); break; case RTM_RESOLVE: /* * resolve was only used for route cloning * here for compat */ break; case RTM_ADD: error = add_route(rnh, info, ret_nrt); break; case RTM_CHANGE: NET_EPOCH_ENTER(et); RIB_WLOCK(rnh); error = change_route(rnh, info, ret_nrt); RIB_WUNLOCK(rnh); NET_EPOCH_EXIT(et); break; default: error = EOPNOTSUPP; } return (error); } static int add_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt) { struct sockaddr *dst, *ndst, *gateway, *netmask; struct rtentry *rt, *rt_old; struct nhop_object *nh; struct radix_node *rn; struct ifaddr *ifa; int error, flags; struct epoch_tracker et; dst = info->rti_info[RTAX_DST]; gateway = info->rti_info[RTAX_GATEWAY]; netmask = info->rti_info[RTAX_NETMASK]; flags = info->rti_flags; if ((flags & RTF_GATEWAY) && !gateway) return (EINVAL); if (dst && gateway && (dst->sa_family != gateway->sa_family) && (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) return (EINVAL); if (info->rti_ifa == NULL) { error = rt_getifa_fib(info, rnh->rib_fibnum); if (error) return (error); } else { ifa_ref(info->rti_ifa); } NET_EPOCH_ENTER(et); error = nhop_create_from_info(rnh, info, &nh); NET_EPOCH_EXIT(et); if (error != 0) { ifa_free(info->rti_ifa); return (error); } rt = uma_zalloc(V_rtzone, M_NOWAIT); if (rt == NULL) { ifa_free(info->rti_ifa); nhop_free(nh); return (ENOBUFS); } rt->rt_flags = RTF_UP | flags; rt->rt_fibnum = rnh->rib_fibnum; rt->rt_nhop = nh; /* * Add the gateway. Possibly re-malloc-ing the storage for it. */ if ((error = rt_setgate(rt, dst, gateway)) != 0) { ifa_free(info->rti_ifa); nhop_free(nh); uma_zfree(V_rtzone, rt); return (error); } /* * point to the (possibly newly malloc'd) dest address. */ ndst = (struct sockaddr *)rt_key(rt); /* * make sure it contains the value we want (masked if needed). */ if (netmask) { rt_maskedcopy(dst, ndst, netmask); } else bcopy(dst, ndst, dst->sa_len); /* * We use the ifa reference returned by rt_getifa_fib(). * This moved from below so that rnh->rnh_addaddr() can * examine the ifa and ifa->ifa_ifp if it so desires. */ ifa = info->rti_ifa; rt->rt_ifa = ifa; rt->rt_ifp = ifa->ifa_ifp; rt->rt_weight = 1; rt_setmetrics(info, rt); RIB_WLOCK(rnh); RT_LOCK(rt); #ifdef RADIX_MPATH /* do not permit exactly the same dst/mask/gw pair */ if (rt_mpath_capable(rnh) && rt_mpath_conflict(rnh, rt, netmask)) { RIB_WUNLOCK(rnh); ifa_free(rt->rt_ifa); R_Free(rt_key(rt)); nhop_free(nh); uma_zfree(V_rtzone, rt); return (EEXIST); } #endif /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes); if (rn != NULL && rt->rt_expire > 0) tmproutes_update(rnh, rt); rt_old = NULL; if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) { /* * Force removal and re-try addition * TODO: better multipath&pinned support */ struct sockaddr *info_dst = info->rti_info[RTAX_DST]; info->rti_info[RTAX_DST] = ndst; /* Do not delete existing PINNED(interface) routes */ info->rti_flags &= ~RTF_PINNED; rt_old = rt_unlinkrte(rnh, info, &error); info->rti_flags |= RTF_PINNED; info->rti_info[RTAX_DST] = info_dst; if (rt_old != NULL) rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes); } RIB_WUNLOCK(rnh); if (rt_old != NULL) RT_UNLOCK(rt_old); /* * If it still failed to go into the tree, * then un-make it (this should be a function) */ if (rn == NULL) { ifa_free(rt->rt_ifa); R_Free(rt_key(rt)); nhop_free(nh); uma_zfree(V_rtzone, rt); return (EEXIST); } if (rt_old != NULL) { rt_notifydelete(rt_old, info); RTFREE(rt_old); } /* * If this protocol has something to add to this then * allow it to do that as well. */ if (ifa->ifa_rtrequest) ifa->ifa_rtrequest(RTM_ADD, rt, info); /* * actually return a resultant rtentry and * give the caller a single reference. */ if (ret_nrt) { *ret_nrt = rt; RT_ADDREF(rt); } rnh->rnh_gen++; /* Routing table updated */ RT_UNLOCK(rt); return (0); } static int del_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt) { struct sockaddr *dst, *netmask; struct sockaddr_storage mdst; struct rtentry *rt; int error; dst = info->rti_info[RTAX_DST]; netmask = info->rti_info[RTAX_NETMASK]; if (netmask) { if (dst->sa_len > sizeof(mdst)) return (EINVAL); rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask); dst = (struct sockaddr *)&mdst; } RIB_WLOCK(rnh); rt = rt_unlinkrte(rnh, info, &error); RIB_WUNLOCK(rnh); if (error != 0) return (error); rt_notifydelete(rt, info); /* * If the caller wants it, then it can have it, * but it's up to it to free the rtentry as we won't be * doing it. */ if (ret_nrt) { *ret_nrt = rt; RT_UNLOCK(rt); } else RTFREE_LOCKED(rt); return (0); } static int change_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt) { struct rtentry *rt = NULL; int error = 0; int free_ifa = 0; int family, mtu; struct nhop_object *nh; struct if_mtuinfo ifmtu; RIB_WLOCK_ASSERT(rnh); rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], &rnh->head); if (rt == NULL) return (ESRCH); #ifdef RADIX_MPATH /* * If we got multipath routes, * we require users to specify a matching RTAX_GATEWAY. */ if (rt_mpath_capable(rnh)) { rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]); if (rt == NULL) return (ESRCH); } #endif nh = NULL; RT_LOCK(rt); rt_setmetrics(info, rt); /* * New gateway could require new ifaddr, ifp; * flags may also be different; ifp may be specified * by ll sockaddr when protocol address is ambiguous */ if (((rt->rt_flags & RTF_GATEWAY) && info->rti_info[RTAX_GATEWAY] != NULL) || info->rti_info[RTAX_IFP] != NULL || (info->rti_info[RTAX_IFA] != NULL && !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) { /* * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags * to avoid rlock in the ifa_ifwithroute(). */ info->rti_flags |= RTF_RNH_LOCKED; error = rt_getifa_fib(info, rnh->rib_fibnum); info->rti_flags &= ~RTF_RNH_LOCKED; if (info->rti_ifa != NULL) free_ifa = 1; if (error != 0) goto bad; } error = nhop_create_from_nhop(rnh, rt->rt_nhop, info, &nh); if (error != 0) goto bad; /* Check if outgoing interface has changed */ if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa && rt->rt_ifa != NULL) { if (rt->rt_ifa->ifa_rtrequest != NULL) rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info); ifa_free(rt->rt_ifa); rt->rt_ifa = NULL; } /* Update gateway address */ if (info->rti_info[RTAX_GATEWAY] != NULL) { error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]); if (error != 0) goto bad; rt->rt_flags &= ~RTF_GATEWAY; rt->rt_flags |= (RTF_GATEWAY & info->rti_flags); } if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) { ifa_ref(info->rti_ifa); rt->rt_ifa = info->rti_ifa; rt->rt_ifp = info->rti_ifp; } /* Allow some flags to be toggled on change. */ rt->rt_flags &= ~RTF_FMASK; rt->rt_flags |= info->rti_flags & RTF_FMASK; if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL) rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info); /* Alter route MTU if necessary */ if (rt->rt_ifp != NULL) { family = info->rti_info[RTAX_DST]->sa_family; mtu = if_getmtu_family(rt->rt_ifp, family); /* Set default MTU */ if (rt->rt_mtu == 0) rt->rt_mtu = mtu; if (rt->rt_mtu != mtu) { /* Check if we really need to update */ ifmtu.ifp = rt->rt_ifp; ifmtu.mtu = mtu; if_updatemtu_cb(rt->rt_nodes, &ifmtu); } } /* Update nexthop */ nhop_free(rt->rt_nhop); rt->rt_nhop = nh; nh = NULL; /* * This route change may have modified the route's gateway. In that * case, any inpcbs that have cached this route need to invalidate their * llentry cache. */ rnh->rnh_gen++; if (ret_nrt) { *ret_nrt = rt; RT_ADDREF(rt); } bad: RT_UNLOCK(rt); if (nh != NULL) nhop_free(nh); if (free_ifa != 0) { ifa_free(info->rti_ifa); info->rti_ifa = NULL; } return (error); } static void rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt) { if (info->rti_mflags & RTV_MTU) { if (info->rti_rmx->rmx_mtu != 0) { /* * MTU was explicitly provided by user. * Keep it. */ rt->rt_flags |= RTF_FIXEDMTU; } else { /* * User explicitly sets MTU to 0. * Assume rollback to default. */ rt->rt_flags &= ~RTF_FIXEDMTU; } rt->rt_mtu = info->rti_rmx->rmx_mtu; } if (info->rti_mflags & RTV_WEIGHT) rt->rt_weight = info->rti_rmx->rmx_weight; /* Kernel -> userland timebase conversion. */ if (info->rti_mflags & RTV_EXPIRE) rt->rt_expire = info->rti_rmx->rmx_expire ? info->rti_rmx->rmx_expire - time_second + time_uptime : 0; } int rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate) { /* XXX dst may be overwritten, can we move this to below */ int dlen = SA_SIZE(dst), glen = SA_SIZE(gate); /* * Prepare to store the gateway in rt->rt_gateway. * Both dst and gateway are stored one after the other in the same * malloc'd chunk. If we have room, we can reuse the old buffer, * rt_gateway already points to the right place. * Otherwise, malloc a new block and update the 'dst' address. */ if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) { caddr_t new; R_Malloc(new, caddr_t, dlen + glen); if (new == NULL) return ENOBUFS; /* * XXX note, we copy from *dst and not *rt_key(rt) because * rt_setgate() can be called to initialize a newly * allocated route entry, in which case rt_key(rt) == NULL * (and also rt->rt_gateway == NULL). * Free()/free() handle a NULL argument just fine. */ bcopy(dst, new, dlen); R_Free(rt_key(rt)); /* free old block, if any */ rt_key(rt) = (struct sockaddr *)new; rt->rt_gateway = (struct sockaddr *)(new + dlen); } /* * Copy the new gateway value into the memory chunk. */ bcopy(gate, rt->rt_gateway, glen); return (0); } void rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask) { u_char *cp1 = (u_char *)src; u_char *cp2 = (u_char *)dst; u_char *cp3 = (u_char *)netmask; u_char *cplim = cp2 + *cp3; u_char *cplim2 = cp2 + *cp1; *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ cp3 += 2; if (cplim > cplim2) cplim = cplim2; while (cp2 < cplim) *cp2++ = *cp1++ & *cp3++; if (cp2 < cplim2) bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2)); } /* * Set up a routing table entry, normally * for an interface. */ #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */ static inline int rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum) { RIB_RLOCK_TRACKER; struct sockaddr *dst; struct sockaddr *netmask; struct rtentry *rt = NULL; struct rt_addrinfo info; int error = 0; int startfib, endfib; char tempbuf[_SOCKADDR_TMPSIZE]; int didwork = 0; int a_failure = 0; struct sockaddr_dl_short *sdl = NULL; struct rib_head *rnh; if (flags & RTF_HOST) { dst = ifa->ifa_dstaddr; netmask = NULL; } else { dst = ifa->ifa_addr; netmask = ifa->ifa_netmask; } if (dst->sa_len == 0) return(EINVAL); switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We support multiple FIBs. */ break; default: fibnum = RT_DEFAULT_FIB; break; } if (fibnum == RT_ALL_FIBS) { if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) startfib = endfib = ifa->ifa_ifp->if_fib; else { startfib = 0; endfib = rt_numfibs - 1; } } else { KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum")); startfib = fibnum; endfib = fibnum; } /* * If it's a delete, check that if it exists, * it's on the correct interface or we might scrub * a route to another ifa which would * be confusing at best and possibly worse. */ if (cmd == RTM_DELETE) { /* * It's a delete, so it should already exist.. * If it's a net, mask off the host bits * (Assuming we have a mask) * XXX this is kinda inet specific.. */ if (netmask != NULL) { rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask); dst = (struct sockaddr *)tempbuf; } } else if (cmd == RTM_ADD) { sdl = (struct sockaddr_dl_short *)tempbuf; bzero(sdl, sizeof(struct sockaddr_dl_short)); sdl->sdl_family = AF_LINK; sdl->sdl_len = sizeof(struct sockaddr_dl_short); sdl->sdl_type = ifa->ifa_ifp->if_type; sdl->sdl_index = ifa->ifa_ifp->if_index; } /* * Now go through all the requested tables (fibs) and do the * requested action. Realistically, this will either be fib 0 * for protocols that don't do multiple tables or all the * tables for those that do. */ for ( fibnum = startfib; fibnum <= endfib; fibnum++) { if (cmd == RTM_DELETE) { struct radix_node *rn; /* * Look up an rtentry that is in the routing tree and * contains the correct info. */ rnh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rnh == NULL) /* this table doesn't exist but others might */ continue; RIB_RLOCK(rnh); rn = rnh->rnh_lookup(dst, netmask, &rnh->head); #ifdef RADIX_MPATH if (rt_mpath_capable(rnh)) { if (rn == NULL) error = ESRCH; else { rt = RNTORT(rn); /* * for interface route the * rt->rt_gateway is sockaddr_dl, so * rt_mpath_matchgate must use the * interface address */ rt = rt_mpath_matchgate(rt, ifa->ifa_addr); if (rt == NULL) error = ESRCH; } } #endif error = (rn == NULL || (rn->rn_flags & RNF_ROOT) || RNTORT(rn)->rt_ifa != ifa); RIB_RUNLOCK(rnh); if (error) { /* this is only an error if bad on ALL tables */ continue; } } /* * Do the actual request */ bzero((caddr_t)&info, sizeof(info)); info.rti_ifa = ifa; info.rti_flags = flags | (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED; info.rti_info[RTAX_DST] = dst; /* * doing this for compatibility reasons */ if (cmd == RTM_ADD) info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)sdl; else info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; info.rti_info[RTAX_NETMASK] = netmask; error = rtrequest1_fib(cmd, &info, &rt, fibnum); if (error == 0 && rt != NULL) { /* * notify any listening routing agents of the change */ RT_LOCK(rt); #ifdef RADIX_MPATH /* * in case address alias finds the first address * e.g. ifconfig bge0 192.0.2.246/24 * e.g. ifconfig bge0 192.0.2.247/24 * the address set in the route is 192.0.2.246 * so we need to replace it with 192.0.2.247 */ if (memcmp(rt->rt_ifa->ifa_addr, ifa->ifa_addr, ifa->ifa_addr->sa_len)) { ifa_free(rt->rt_ifa); ifa_ref(ifa); rt->rt_ifp = ifa->ifa_ifp; rt->rt_ifa = ifa; } #endif RT_ADDREF(rt); RT_UNLOCK(rt); rt_newaddrmsg_fib(cmd, ifa, rt, fibnum); RT_LOCK(rt); RT_REMREF(rt); if (cmd == RTM_DELETE) { /* * If we are deleting, and we found an entry, * then it's been removed from the tree.. * now throw it away. */ RTFREE_LOCKED(rt); } else { if (cmd == RTM_ADD) { /* * We just wanted to add it.. * we don't actually need a reference. */ RT_REMREF(rt); } RT_UNLOCK(rt); } didwork = 1; } if (error) a_failure = error; } if (cmd == RTM_DELETE) { if (didwork) { error = 0; } else { /* we only give an error if it wasn't in any table */ error = ((flags & RTF_HOST) ? EHOSTUNREACH : ENETUNREACH); } } else { if (a_failure) { /* return an error if any of them failed */ error = a_failure; } } return (error); } /* * Set up a routing table entry, normally * for an interface. */ int rtinit(struct ifaddr *ifa, int cmd, int flags) { struct sockaddr *dst; int fib = RT_DEFAULT_FIB; if (flags & RTF_HOST) { dst = ifa->ifa_dstaddr; } else { dst = ifa->ifa_addr; } switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We do support multiple FIBs. */ fib = RT_ALL_FIBS; break; } return (rtinit1(ifa, cmd, flags, fib)); } /* * Announce interface address arrival/withdraw * Returns 0 on success. */ int rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %d", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); EVENTHANDLER_DIRECT_INVOKE(rt_addrmsg, ifa, cmd); return (rtsock_addrmsg(cmd, ifa, fibnum)); } /* * Announce kernel-originated route addition/removal to rtsock based on @rt data. * cmd: RTM_ cmd * @rt: valid rtentry * @ifp: target route interface * @fibnum: fib id or RT_ALL_FIBS * * Returns 0 on success. */ int rt_routemsg(int cmd, struct rtentry *rt, struct ifnet *ifp, int rti_addrs, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %d", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__)); return (rtsock_routemsg(cmd, rt, ifp, 0, fibnum)); } /* * Announce kernel-originated route addition/removal to rtsock based on @rt data. * cmd: RTM_ cmd * @info: addrinfo structure with valid data. * @fibnum: fib id or RT_ALL_FIBS * * Returns 0 on success. */ int rt_routemsg_info(int cmd, struct rt_addrinfo *info, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE || cmd == RTM_CHANGE, ("unexpected cmd %d", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); KASSERT(info->rti_info[RTAX_DST] != NULL, (":%s: RTAX_DST must be supplied", __func__)); return (rtsock_routemsg_info(cmd, info, fibnum)); } /* * This is called to generate messages from the routing socket * indicating a network interface has had addresses associated with it. */ void rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, struct rtentry *rt, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %u", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); if (cmd == RTM_ADD) { rt_addrmsg(cmd, ifa, fibnum); if (rt != NULL) rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum); } else { if (rt != NULL) rt_routemsg(cmd, rt, ifa->ifa_ifp, 0, fibnum); rt_addrmsg(cmd, ifa, fibnum); } } Index: head/sys/net/route.h =================================================================== --- head/sys/net/route.h (revision 360291) +++ head/sys/net/route.h (revision 360292) @@ -1,527 +1,522 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1980, 1986, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)route.h 8.4 (Berkeley) 1/9/95 * $FreeBSD$ */ #ifndef _NET_ROUTE_H_ #define _NET_ROUTE_H_ #include #include /* * Kernel resident routing tables. * * The routing tables are initialized when interface addresses * are set by making entries for all directly connected interfaces. */ /* * Struct route consiste of a destination address, * a route entry pointer, link-layer prepend data pointer along * with its length. */ struct route { - struct rtentry *ro_rt; + struct nhop_object *ro_nh; struct llentry *ro_lle; /* * ro_prepend and ro_plen are only used for bpf to pass in a * preformed header. They are not cacheable. */ char *ro_prepend; uint16_t ro_plen; uint16_t ro_flags; uint16_t ro_mtu; /* saved ro_rt mtu */ uint16_t spare; struct sockaddr ro_dst; }; #define RT_L2_ME_BIT 2 /* dst L2 addr is our address */ #define RT_MAY_LOOP_BIT 3 /* dst may require loop copy */ #define RT_HAS_HEADER_BIT 4 /* mbuf already have its header prepended */ #define RT_L2_ME (1 << RT_L2_ME_BIT) /* 0x0004 */ #define RT_MAY_LOOP (1 << RT_MAY_LOOP_BIT) /* 0x0008 */ #define RT_HAS_HEADER (1 << RT_HAS_HEADER_BIT) /* 0x0010 */ #define RT_REJECT 0x0020 /* Destination is reject */ #define RT_BLACKHOLE 0x0040 /* Destination is blackhole */ #define RT_HAS_GW 0x0080 /* Destination has GW */ #define RT_LLE_CACHE 0x0100 /* Cache link layer */ struct rt_metrics { u_long rmx_locks; /* Kernel must leave these values alone */ u_long rmx_mtu; /* MTU for this path */ u_long rmx_hopcount; /* max hops expected */ u_long rmx_expire; /* lifetime for route, e.g. redirect */ u_long rmx_recvpipe; /* inbound delay-bandwidth product */ u_long rmx_sendpipe; /* outbound delay-bandwidth product */ u_long rmx_ssthresh; /* outbound gateway buffer limit */ u_long rmx_rtt; /* estimated round trip time */ u_long rmx_rttvar; /* estimated rtt variance */ u_long rmx_pksent; /* packets sent using this route */ u_long rmx_weight; /* route weight */ u_long rmx_nhidx; /* route nexhop index */ u_long rmx_filler[2]; /* will be used for T/TCP later */ }; /* * rmx_rtt and rmx_rttvar are stored as microseconds; * RTTTOPRHZ(rtt) converts to a value suitable for use * by a protocol slowtimo counter. */ #define RTM_RTTUNIT 1000000 /* units for rtt, rttvar, as units per sec */ #define RTTTOPRHZ(r) ((r) / (RTM_RTTUNIT / PR_SLOWHZ)) /* lle state is exported in rmx_state rt_metrics field */ #define rmx_state rmx_weight /* * Keep a generation count of routing table, incremented on route addition, * so we can invalidate caches. This is accessed without a lock, as precision * is not required. */ typedef volatile u_int rt_gen_t; /* tree generation (for adds) */ #define RT_GEN(fibnum, af) rt_tables_get_gen(fibnum, af) #define RT_DEFAULT_FIB 0 /* Explicitly mark fib=0 restricted cases */ #define RT_ALL_FIBS -1 /* Announce event for every fib */ #ifdef _KERNEL extern u_int rt_numfibs; /* number of usable routing tables */ VNET_DECLARE(u_int, rt_add_addr_allfibs); /* Announce interfaces to all fibs */ #define V_rt_add_addr_allfibs VNET(rt_add_addr_allfibs) #endif /* * We distinguish between routes to hosts and routes to networks, * preferring the former if available. For each route we infer * the interface to use from the gateway address supplied when * the route was entered. Routes that forward packets through * gateways are marked so that the output routines know to address the * gateway rather than the ultimate destination. */ #ifndef RNF_NORMAL #include #ifdef RADIX_MPATH #include #endif #endif #if defined(_KERNEL) struct rtentry { struct radix_node rt_nodes[2]; /* tree glue, and other values */ /* * XXX struct rtentry must begin with a struct radix_node (or two!) * because the code does some casts of a 'struct radix_node *' * to a 'struct rtentry *' */ #define rt_key(r) (*((struct sockaddr **)(&(r)->rt_nodes->rn_key))) #define rt_mask(r) (*((struct sockaddr **)(&(r)->rt_nodes->rn_mask))) #define rt_key_const(r) (*((const struct sockaddr * const *)(&(r)->rt_nodes->rn_key))) #define rt_mask_const(r) (*((const struct sockaddr * const *)(&(r)->rt_nodes->rn_mask))) struct sockaddr *rt_gateway; /* value */ struct ifnet *rt_ifp; /* the answer: interface to use */ struct ifaddr *rt_ifa; /* the answer: interface address to use */ struct nhop_object *rt_nhop; /* nexthop data */ int rt_flags; /* up/down?, host/net */ int rt_refcnt; /* # held references */ u_int rt_fibnum; /* which FIB */ u_long rt_mtu; /* MTU for this path */ u_long rt_weight; /* absolute weight */ u_long rt_expire; /* lifetime for route, e.g. redirect */ #define rt_endzero rt_pksent counter_u64_t rt_pksent; /* packets sent using this route */ struct mtx rt_mtx; /* mutex for routing entry */ struct rtentry *rt_chain; /* pointer to next rtentry to delete */ }; #endif /* _KERNEL */ #define RTF_UP 0x1 /* route usable */ #define RTF_GATEWAY 0x2 /* destination is a gateway */ #define RTF_HOST 0x4 /* host entry (net otherwise) */ #define RTF_REJECT 0x8 /* host or net unreachable */ #define RTF_DYNAMIC 0x10 /* created dynamically (by redirect) */ #define RTF_MODIFIED 0x20 /* modified dynamically (by redirect) */ #define RTF_DONE 0x40 /* message confirmed */ /* 0x80 unused, was RTF_DELCLONE */ /* 0x100 unused, was RTF_CLONING */ #define RTF_XRESOLVE 0x200 /* external daemon resolves name */ #define RTF_LLINFO 0x400 /* DEPRECATED - exists ONLY for backward compatibility */ #define RTF_LLDATA 0x400 /* used by apps to add/del L2 entries */ #define RTF_STATIC 0x800 /* manually added */ #define RTF_BLACKHOLE 0x1000 /* just discard pkts (during updates) */ #define RTF_PROTO2 0x4000 /* protocol specific routing flag */ #define RTF_PROTO1 0x8000 /* protocol specific routing flag */ /* 0x10000 unused, was RTF_PRCLONING */ /* 0x20000 unused, was RTF_WASCLONED */ #define RTF_PROTO3 0x40000 /* protocol specific routing flag */ #define RTF_FIXEDMTU 0x80000 /* MTU was explicitly specified */ #define RTF_PINNED 0x100000 /* route is immutable */ #define RTF_LOCAL 0x200000 /* route represents a local address */ #define RTF_BROADCAST 0x400000 /* route represents a bcast address */ #define RTF_MULTICAST 0x800000 /* route represents a mcast address */ /* 0x8000000 and up unassigned */ #define RTF_STICKY 0x10000000 /* always route dst->src */ #define RTF_RNH_LOCKED 0x40000000 /* radix node head is locked */ #define RTF_GWFLAG_COMPAT 0x80000000 /* a compatibility bit for interacting with existing routing apps */ /* Mask of RTF flags that are allowed to be modified by RTM_CHANGE. */ #define RTF_FMASK \ (RTF_PROTO1 | RTF_PROTO2 | RTF_PROTO3 | RTF_BLACKHOLE | \ RTF_REJECT | RTF_STATIC | RTF_STICKY) /* * fib_ nexthop API flags. */ /* Consumer-visible nexthop info flags */ #define NHF_REJECT 0x0010 /* RTF_REJECT */ #define NHF_BLACKHOLE 0x0020 /* RTF_BLACKHOLE */ #define NHF_REDIRECT 0x0040 /* RTF_DYNAMIC|RTF_MODIFIED */ #define NHF_DEFAULT 0x0080 /* Default route */ #define NHF_BROADCAST 0x0100 /* RTF_BROADCAST */ #define NHF_GATEWAY 0x0200 /* RTF_GATEWAY */ #define NHF_HOST 0x0400 /* RTF_HOST */ /* Nexthop request flags */ #define NHR_NONE 0x00 /* empty flags field */ #define NHR_IFAIF 0x01 /* Return ifa_ifp interface */ #define NHR_REF 0x02 /* For future use */ /* uRPF */ #define NHR_NODEFAULT 0x04 /* do not consider default route */ /* Control plane route request flags */ #define NHR_COPY 0x100 /* Copy rte data */ -#ifdef _KERNEL -/* rte<>ro_flags translation */ -static inline void -rt_update_ro_flags(struct route *ro) -{ - int rt_flags = ro->ro_rt->rt_flags; - - ro->ro_flags &= ~ (RT_REJECT|RT_BLACKHOLE|RT_HAS_GW); - - ro->ro_flags |= (rt_flags & RTF_REJECT) ? RT_REJECT : 0; - ro->ro_flags |= (rt_flags & RTF_BLACKHOLE) ? RT_BLACKHOLE : 0; - ro->ro_flags |= (rt_flags & RTF_GATEWAY) ? RT_HAS_GW : 0; -} -#endif - /* * Routing statistics. */ struct rtstat { uint64_t rts_badredirect; /* bogus redirect calls */ uint64_t rts_dynamic; /* routes created by redirects */ uint64_t rts_newgateway; /* routes modified by redirects */ uint64_t rts_unreach; /* lookups which failed */ uint64_t rts_wildcard; /* lookups satisfied by a wildcard */ uint64_t rts_nh_idx_alloc_failure; /* nexthop index alloc failure*/ uint64_t rts_nh_alloc_failure; /* nexthop allocation failure*/ }; /* * Structures for routing messages. */ struct rt_msghdr { u_short rtm_msglen; /* to skip over non-understood messages */ u_char rtm_version; /* future binary compatibility */ u_char rtm_type; /* message type */ u_short rtm_index; /* index for associated ifp */ u_short _rtm_spare1; int rtm_flags; /* flags, incl. kern & message, e.g. DONE */ int rtm_addrs; /* bitmask identifying sockaddrs in msg */ pid_t rtm_pid; /* identify sender */ int rtm_seq; /* for sender to identify action */ int rtm_errno; /* why failed */ int rtm_fmask; /* bitmask used in RTM_CHANGE message */ u_long rtm_inits; /* which metrics we are initializing */ struct rt_metrics rtm_rmx; /* metrics themselves */ }; #define RTM_VERSION 5 /* Up the ante and ignore older versions */ /* * Message types. * * The format for each message is annotated below using the following * identifiers: * * (1) struct rt_msghdr * (2) struct ifa_msghdr * (3) struct if_msghdr * (4) struct ifma_msghdr * (5) struct if_announcemsghdr * */ #define RTM_ADD 0x1 /* (1) Add Route */ #define RTM_DELETE 0x2 /* (1) Delete Route */ #define RTM_CHANGE 0x3 /* (1) Change Metrics or flags */ #define RTM_GET 0x4 /* (1) Report Metrics */ #define RTM_LOSING 0x5 /* (1) Kernel Suspects Partitioning */ #define RTM_REDIRECT 0x6 /* (1) Told to use different route */ #define RTM_MISS 0x7 /* (1) Lookup failed on this address */ #define RTM_LOCK 0x8 /* (1) fix specified metrics */ /* 0x9 */ /* 0xa */ #define RTM_RESOLVE 0xb /* (1) req to resolve dst to LL addr */ #define RTM_NEWADDR 0xc /* (2) address being added to iface */ #define RTM_DELADDR 0xd /* (2) address being removed from iface */ #define RTM_IFINFO 0xe /* (3) iface going up/down etc. */ #define RTM_NEWMADDR 0xf /* (4) mcast group membership being added to if */ #define RTM_DELMADDR 0x10 /* (4) mcast group membership being deleted */ #define RTM_IFANNOUNCE 0x11 /* (5) iface arrival/departure */ #define RTM_IEEE80211 0x12 /* (5) IEEE80211 wireless event */ /* * Bitmask values for rtm_inits and rmx_locks. */ #define RTV_MTU 0x1 /* init or lock _mtu */ #define RTV_HOPCOUNT 0x2 /* init or lock _hopcount */ #define RTV_EXPIRE 0x4 /* init or lock _expire */ #define RTV_RPIPE 0x8 /* init or lock _recvpipe */ #define RTV_SPIPE 0x10 /* init or lock _sendpipe */ #define RTV_SSTHRESH 0x20 /* init or lock _ssthresh */ #define RTV_RTT 0x40 /* init or lock _rtt */ #define RTV_RTTVAR 0x80 /* init or lock _rttvar */ #define RTV_WEIGHT 0x100 /* init or lock _weight */ /* * Bitmask values for rtm_addrs. */ #define RTA_DST 0x1 /* destination sockaddr present */ #define RTA_GATEWAY 0x2 /* gateway sockaddr present */ #define RTA_NETMASK 0x4 /* netmask sockaddr present */ #define RTA_GENMASK 0x8 /* cloning mask sockaddr present */ #define RTA_IFP 0x10 /* interface name sockaddr present */ #define RTA_IFA 0x20 /* interface addr sockaddr present */ #define RTA_AUTHOR 0x40 /* sockaddr for author of redirect */ #define RTA_BRD 0x80 /* for NEWADDR, broadcast or p-p dest addr */ /* * Index offsets for sockaddr array for alternate internal encoding. */ #define RTAX_DST 0 /* destination sockaddr present */ #define RTAX_GATEWAY 1 /* gateway sockaddr present */ #define RTAX_NETMASK 2 /* netmask sockaddr present */ #define RTAX_GENMASK 3 /* cloning mask sockaddr present */ #define RTAX_IFP 4 /* interface name sockaddr present */ #define RTAX_IFA 5 /* interface addr sockaddr present */ #define RTAX_AUTHOR 6 /* sockaddr for author of redirect */ #define RTAX_BRD 7 /* for NEWADDR, broadcast or p-p dest addr */ #define RTAX_MAX 8 /* size of array to allocate */ struct nhop_object; typedef int rt_filter_f_t(const struct rtentry *, const struct nhop_object *, void *); struct rt_addrinfo { int rti_addrs; /* Route RTF_ flags */ int rti_flags; /* Route RTF_ flags */ struct sockaddr *rti_info[RTAX_MAX]; /* Sockaddr data */ struct ifaddr *rti_ifa; /* value of rt_ifa addr */ struct ifnet *rti_ifp; /* route interface */ rt_filter_f_t *rti_filter; /* filter function */ void *rti_filterdata; /* filter paramenters */ u_long rti_mflags; /* metrics RTV_ flags */ u_long rti_spare; /* Will be used for fib */ struct rt_metrics *rti_rmx; /* Pointer to route metrics */ }; /* * This macro returns the size of a struct sockaddr when passed * through a routing socket. Basically we round up sa_len to * a multiple of sizeof(long), with a minimum of sizeof(long). * The case sa_len == 0 should only apply to empty structures. */ #define SA_SIZE(sa) \ ( (((struct sockaddr *)(sa))->sa_len == 0) ? \ sizeof(long) : \ 1 + ( (((struct sockaddr *)(sa))->sa_len - 1) | (sizeof(long) - 1) ) ) #define sa_equal(a, b) ( \ (((const struct sockaddr *)(a))->sa_len == ((const struct sockaddr *)(b))->sa_len) && \ (bcmp((a), (b), ((const struct sockaddr *)(b))->sa_len) == 0)) #ifdef _KERNEL #define RT_LINK_IS_UP(ifp) (!((ifp)->if_capabilities & IFCAP_LINKSTATE) \ || (ifp)->if_link_state == LINK_STATE_UP) #define RT_LOCK_INIT(_rt) \ mtx_init(&(_rt)->rt_mtx, "rtentry", NULL, MTX_DEF | MTX_DUPOK | MTX_NEW) #define RT_LOCK(_rt) mtx_lock(&(_rt)->rt_mtx) #define RT_UNLOCK(_rt) mtx_unlock(&(_rt)->rt_mtx) #define RT_LOCK_DESTROY(_rt) mtx_destroy(&(_rt)->rt_mtx) #define RT_LOCK_ASSERT(_rt) mtx_assert(&(_rt)->rt_mtx, MA_OWNED) #define RT_UNLOCK_COND(_rt) do { \ if (mtx_owned(&(_rt)->rt_mtx)) \ mtx_unlock(&(_rt)->rt_mtx); \ } while (0) #define RT_ADDREF(_rt) do { \ RT_LOCK_ASSERT(_rt); \ KASSERT((_rt)->rt_refcnt >= 0, \ ("negative refcnt %d", (_rt)->rt_refcnt)); \ (_rt)->rt_refcnt++; \ } while (0) #define RT_REMREF(_rt) do { \ RT_LOCK_ASSERT(_rt); \ KASSERT((_rt)->rt_refcnt > 0, \ ("bogus refcnt %d", (_rt)->rt_refcnt)); \ (_rt)->rt_refcnt--; \ } while (0) #define RTFREE_LOCKED(_rt) do { \ if ((_rt)->rt_refcnt <= 1) \ rtfree(_rt); \ else { \ RT_REMREF(_rt); \ RT_UNLOCK(_rt); \ } \ /* guard against invalid refs */ \ _rt = 0; \ } while (0) #define RTFREE(_rt) do { \ RT_LOCK(_rt); \ RTFREE_LOCKED(_rt); \ } while (0) #define RTFREE_FUNC(_rt) rtfree_func(_rt) #define RO_RTFREE(_ro) do { \ if ((_ro)->ro_rt) \ RTFREE((_ro)->ro_rt); \ } while (0) +#define RO_NHFREE(_ro) do { \ + if ((_ro)->ro_nh) { \ + NH_FREE((_ro)->ro_nh); \ + (_ro)->ro_nh = NULL; \ + } \ +} while (0) + #define RO_INVALIDATE_CACHE(ro) do { \ - RO_RTFREE(ro); \ if ((ro)->ro_lle != NULL) { \ LLE_FREE((ro)->ro_lle); \ (ro)->ro_lle = NULL; \ } \ + if ((ro)->ro_nh != NULL) { \ + NH_FREE((ro)->ro_nh); \ + (ro)->ro_nh = NULL; \ + } \ } while (0) /* * Validate a cached route based on a supplied cookie. If there is an * out-of-date cache, simply free it. Update the generation number * for the new allocation */ -#define RT_VALIDATE(ro, cookiep, fibnum) do { \ +#define NH_VALIDATE(ro, cookiep, fibnum) do { \ rt_gen_t cookie = RT_GEN(fibnum, (ro)->ro_dst.sa_family); \ if (*(cookiep) != cookie) { \ RO_INVALIDATE_CACHE(ro); \ *(cookiep) = cookie; \ } \ } while (0) struct ifmultiaddr; struct rib_head; void rt_ieee80211msg(struct ifnet *, int, void *, size_t); void rt_ifannouncemsg(struct ifnet *, int); void rt_ifmsg(struct ifnet *); void rt_missmsg(int, struct rt_addrinfo *, int, int); void rt_missmsg_fib(int, struct rt_addrinfo *, int, int, int); void rt_newaddrmsg_fib(int, struct ifaddr *, struct rtentry *, int); int rt_addrmsg(int, struct ifaddr *, int); int rt_routemsg(int, struct rtentry *, struct ifnet *ifp, int, int); int rt_routemsg_info(int, struct rt_addrinfo *, int); void rt_newmaddrmsg(int, struct ifmultiaddr *); int rt_setgate(struct rtentry *, struct sockaddr *, struct sockaddr *); void rt_maskedcopy(struct sockaddr *, struct sockaddr *, struct sockaddr *); struct rib_head *rt_table_init(int, int, u_int); void rt_table_destroy(struct rib_head *); u_int rt_tables_get_gen(int table, int fam); int rtsock_addrmsg(int, struct ifaddr *, int); int rtsock_routemsg(int, struct rtentry *, struct ifnet *ifp, int, int); int rtsock_routemsg_info(int, struct rt_addrinfo *, int); /* * Note the following locking behavior: * * rtalloc1() returns a locked rtentry * * rtfree() and RTFREE_LOCKED() require a locked rtentry * * RTFREE() uses an unlocked entry. */ void rtfree(struct rtentry *); void rtfree_func(struct rtentry *); void rt_updatemtu(struct ifnet *); typedef int rt_walktree_f_t(struct rtentry *, void *); typedef void rt_setwarg_t(struct rib_head *, uint32_t, int, void *); void rib_walk_del(u_int fibnum, int family, rt_filter_f_t *filter_f, void *arg, bool report); void rt_foreach_fib_walk(int af, rt_setwarg_t *, rt_walktree_f_t *, void *); void rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg); void rt_flushifroutes_af(struct ifnet *, int); void rt_flushifroutes(struct ifnet *ifp); /* XXX MRT COMPAT VERSIONS THAT SET UNIVERSE to 0 */ /* Thes are used by old code not yet converted to use multiple FIBS */ struct rtentry *rtalloc1(struct sockaddr *, int, u_long); int rtinit(struct ifaddr *, int, int); /* XXX MRT NEW VERSIONS THAT USE FIBs * For now the protocol indepedent versions are the same as the AF_INET ones * but this will change.. */ void rtalloc_ign_fib(struct route *ro, u_long ignflags, u_int fibnum); struct rtentry *rtalloc1_fib(struct sockaddr *, int, u_long, u_int); int rtioctl_fib(u_long, caddr_t, u_int); int rtrequest_fib(int, struct sockaddr *, struct sockaddr *, struct sockaddr *, int, struct rtentry **, u_int); int rtrequest1_fib(int, struct rt_addrinfo *, struct rtentry **, u_int); int rib_lookup_info(uint32_t, const struct sockaddr *, uint32_t, uint32_t, struct rt_addrinfo *); void rib_free_info(struct rt_addrinfo *info); int rib_add_redirect(u_int fibnum, struct sockaddr *dst, struct sockaddr *gateway, struct sockaddr *author, struct ifnet *ifp, int flags, int expire_sec); /* New API */ void rib_walk(int af, u_int fibnum, rt_walktree_f_t *wa_f, void *arg); #endif #endif Index: head/sys/net/route_var.h =================================================================== --- head/sys/net/route_var.h (revision 360291) +++ head/sys/net/route_var.h (revision 360292) @@ -1,157 +1,157 @@ /*- * Copyright (c) 2015-2016 * Alexander V. Chernikov * * 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 University 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 REGENTS 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 REGENTS 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. * * $FreeBSD$ */ #ifndef _NET_ROUTE_VAR_H_ #define _NET_ROUTE_VAR_H_ struct nh_control; typedef int rnh_preadd_entry_f_t(u_int fibnum, const struct sockaddr *addr, const struct sockaddr *mask, struct nhop_object *nh); struct rib_head { struct radix_head head; rn_matchaddr_f_t *rnh_matchaddr; /* longest match for sockaddr */ rn_addaddr_f_t *rnh_addaddr; /* add based on sockaddr*/ rn_deladdr_f_t *rnh_deladdr; /* remove based on sockaddr */ rn_lookup_f_t *rnh_lookup; /* exact match for sockaddr */ rn_walktree_t *rnh_walktree; /* traverse tree */ rn_walktree_from_t *rnh_walktree_from; /* traverse tree below a */ rn_close_t *rnh_close; /*do something when the last ref drops*/ rnh_preadd_entry_f_t *rnh_preadd; /* hook to alter record prior to insertion */ rt_gen_t rnh_gen; /* generation counter */ int rnh_multipath; /* multipath capable ? */ struct radix_node rnh_nodes[3]; /* empty tree for common case */ struct rmlock rib_lock; /* config/data path lock */ struct radix_mask_head rmhead; /* masks radix head */ struct vnet *rib_vnet; /* vnet pointer */ int rib_family; /* AF of the rtable */ u_int rib_fibnum; /* fib number */ struct callout expire_callout; /* Callout for expiring dynamic routes */ time_t next_expire; /* Next expire run ts */ struct nh_control *nh_control; /* nexthop subsystem data */ }; #define RIB_RLOCK_TRACKER struct rm_priotracker _rib_tracker #define RIB_LOCK_INIT(rh) rm_init(&(rh)->rib_lock, "rib head lock") #define RIB_LOCK_DESTROY(rh) rm_destroy(&(rh)->rib_lock) #define RIB_RLOCK(rh) rm_rlock(&(rh)->rib_lock, &_rib_tracker) #define RIB_RUNLOCK(rh) rm_runlock(&(rh)->rib_lock, &_rib_tracker) #define RIB_WLOCK(rh) rm_wlock(&(rh)->rib_lock) #define RIB_WUNLOCK(rh) rm_wunlock(&(rh)->rib_lock) #define RIB_LOCK_ASSERT(rh) rm_assert(&(rh)->rib_lock, RA_LOCKED) #define RIB_WLOCK_ASSERT(rh) rm_assert(&(rh)->rib_lock, RA_WLOCKED) /* Macro for verifying fields in af-specific 'struct route' structures */ #define CHK_STRUCT_FIELD_GENERIC(_s1, _f1, _s2, _f2) \ _Static_assert(sizeof(((_s1 *)0)->_f1) == sizeof(((_s2 *)0)->_f2), \ "Fields " #_f1 " and " #_f2 " size differs"); \ _Static_assert(__offsetof(_s1, _f1) == __offsetof(_s2, _f2), \ "Fields " #_f1 " and " #_f2 " offset differs"); #define _CHK_ROUTE_FIELD(_route_new, _field) \ CHK_STRUCT_FIELD_GENERIC(struct route, _field, _route_new, _field) #define CHK_STRUCT_ROUTE_FIELDS(_route_new) \ - _CHK_ROUTE_FIELD(_route_new, ro_rt) \ + _CHK_ROUTE_FIELD(_route_new, ro_nh) \ _CHK_ROUTE_FIELD(_route_new, ro_lle) \ _CHK_ROUTE_FIELD(_route_new, ro_prepend)\ _CHK_ROUTE_FIELD(_route_new, ro_plen) \ _CHK_ROUTE_FIELD(_route_new, ro_flags) \ _CHK_ROUTE_FIELD(_route_new, ro_mtu) \ _CHK_ROUTE_FIELD(_route_new, spare) #define CHK_STRUCT_ROUTE_COMPAT(_ro_new, _dst_new) \ CHK_STRUCT_ROUTE_FIELDS(_ro_new); \ _Static_assert(__offsetof(struct route, ro_dst) == __offsetof(_ro_new, _dst_new),\ "ro_dst and " #_dst_new " are at different offset") struct rib_head *rt_tables_get_rnh(int fib, int family); void rt_mpath_init_rnh(struct rib_head *rnh); VNET_PCPUSTAT_DECLARE(struct rtstat, rtstat); #define RTSTAT_ADD(name, val) \ VNET_PCPUSTAT_ADD(struct rtstat, rtstat, name, (val)) #define RTSTAT_INC(name) RTSTAT_ADD(name, 1) /* * With the split between the routing entry and the nexthop, * rt_flags has to be split between these 2 entries. As rtentry * mostly contains prefix data and is thought to be generic enough * so one can transparently change the nexthop pointer w/o requiring * any other rtentry changes, most of rt_flags shifts to the particular nexthop. * / * * RTF_UP: rtentry, as an indication that it is linked. * RTF_HOST: rtentry, nhop. The latter indication is needed for the datapath * RTF_DYNAMIC: nhop, to make rtentry generic. * RTF_MODIFIED: nhop, to make rtentry generic. (legacy) * -- "native" path (nhop) properties: * RTF_GATEWAY, RTF_STATIC, RTF_PROTO1, RTF_PROTO2, RTF_PROTO3, RTF_FIXEDMTU, * RTF_PINNED, RTF_REJECT, RTF_BLACKHOLE, RTF_BROADCAST */ /* Nexthop rt flags mask */ #define NHOP_RT_FLAG_MASK (RTF_GATEWAY | RTF_HOST | RTF_REJECT | RTF_DYNAMIC | \ RTF_MODIFIED | RTF_STATIC | RTF_BLACKHOLE | RTF_PROTO1 | RTF_PROTO2 | \ RTF_PROTO3 | RTF_FIXEDMTU | RTF_PINNED | RTF_BROADCAST) /* rtentry rt flag mask */ #define RTE_RT_FLAG_MASK (RTF_UP | RTF_HOST) /* Nexthop selection */ #define _NH2MP(_nh) ((struct nhgrp_object *)(_nh)) #define _SELECT_NHOP(_nh, _flowid) \ (_NH2MP(_nh))->nhops[(_flowid) % (_NH2MP(_nh))->mp_size] #define _RT_SELECT_NHOP(_nh, _flowid) \ ((!NH_IS_MULTIPATH(_nh)) ? (_nh) : _SELECT_NHOP(_nh, _flowid)) #define RT_SELECT_NHOP(_rt, _flowid) _RT_SELECT_NHOP((_rt)->rt_nhop, _flowid) /* rte<>nhop translation */ static inline uint16_t fib_rte_to_nh_flags(int rt_flags) { uint16_t res; res = (rt_flags & RTF_REJECT) ? NHF_REJECT : 0; res |= (rt_flags & RTF_HOST) ? NHF_HOST : 0; res |= (rt_flags & RTF_BLACKHOLE) ? NHF_BLACKHOLE : 0; res |= (rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) ? NHF_REDIRECT : 0; res |= (rt_flags & RTF_BROADCAST) ? NHF_BROADCAST : 0; res |= (rt_flags & RTF_GATEWAY) ? NHF_GATEWAY : 0; return (res); } void tmproutes_update(struct rib_head *rnh, struct rtentry *rt); void tmproutes_init(struct rib_head *rh); void tmproutes_destroy(struct rib_head *rh); #endif Index: head/sys/netinet/in_fib.c =================================================================== --- head/sys/netinet/in_fib.c (revision 360291) +++ head/sys/netinet/in_fib.c (revision 360292) @@ -1,360 +1,364 @@ /*- * Copyright (c) 2015 * Alexander V. Chernikov * * 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 University 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 REGENTS 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 REGENTS 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 __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_route.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include #include #include #ifdef INET + +/* Verify struct route compatiblity */ +/* Assert 'struct route_in' is compatible with 'struct route' */ +CHK_STRUCT_ROUTE_COMPAT(struct route_in, ro_dst4); static void fib4_rte_to_nh_basic(struct nhop_object *nh, struct in_addr dst, uint32_t flags, struct nhop4_basic *pnh4); static void fib4_rte_to_nh_extended(struct nhop_object *nh, struct in_addr dst, uint32_t flags, struct nhop4_extended *pnh4); #define RNTORT(p) ((struct rtentry *)(p)) static void fib4_rte_to_nh_basic(struct nhop_object *nh, struct in_addr dst, uint32_t flags, struct nhop4_basic *pnh4) { if ((flags & NHR_IFAIF) != 0) pnh4->nh_ifp = nh->nh_ifa->ifa_ifp; else pnh4->nh_ifp = nh->nh_ifp; pnh4->nh_mtu = nh->nh_mtu; if (nh->nh_flags & NHF_GATEWAY) pnh4->nh_addr = nh->gw4_sa.sin_addr; else pnh4->nh_addr = dst; /* Set flags */ pnh4->nh_flags = nh->nh_flags; /* TODO: Handle RTF_BROADCAST here */ } static void fib4_rte_to_nh_extended(struct nhop_object *nh, struct in_addr dst, uint32_t flags, struct nhop4_extended *pnh4) { if ((flags & NHR_IFAIF) != 0) pnh4->nh_ifp = nh->nh_ifa->ifa_ifp; else pnh4->nh_ifp = nh->nh_ifp; pnh4->nh_mtu = nh->nh_mtu; if (nh->nh_flags & NHF_GATEWAY) pnh4->nh_addr = nh->gw4_sa.sin_addr; else pnh4->nh_addr = dst; /* Set flags */ pnh4->nh_flags = nh->nh_flags; pnh4->nh_ia = ifatoia(nh->nh_ifa); pnh4->nh_src = IA_SIN(pnh4->nh_ia)->sin_addr; } /* * Performs IPv4 route table lookup on @dst. Returns 0 on success. * Stores nexthop info provided @pnh4 structure. * Note that * - nh_ifp cannot be safely dereferenced * - nh_ifp represents logical transmit interface (rt_ifp) (e.g. if * looking up address on interface "ix0" pointer to "lo0" interface * will be returned instead of "ix0") * - nh_ifp represents "address" interface if NHR_IFAIF flag is passed * - howewer mtu from "transmit" interface will be returned. */ int fib4_lookup_nh_basic(uint32_t fibnum, struct in_addr dst, uint32_t flags, uint32_t flowid, struct nhop4_basic *pnh4) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct sockaddr_in sin; struct nhop_object *nh; KASSERT((fibnum < rt_numfibs), ("fib4_lookup_nh_basic: bad fibnum")); rh = rt_tables_get_rnh(fibnum, AF_INET); if (rh == NULL) return (ENOENT); /* Prepare lookup key */ memset(&sin, 0, sizeof(sin)); sin.sin_len = sizeof(struct sockaddr_in); sin.sin_addr = dst; RIB_RLOCK(rh); rn = rh->rnh_matchaddr((void *)&sin, &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { nh = RNTORT(rn)->rt_nhop; /* Ensure route & ifp is UP */ if (RT_LINK_IS_UP(nh->nh_ifp)) { fib4_rte_to_nh_basic(nh, dst, flags, pnh4); RIB_RUNLOCK(rh); return (0); } } RIB_RUNLOCK(rh); return (ENOENT); } /* * Performs IPv4 route table lookup on @dst. Returns 0 on success. * Stores extende nexthop info provided @pnh4 structure. * Note that * - nh_ifp cannot be safely dereferenced unless NHR_REF is specified. * - in that case you need to call fib4_free_nh_ext() * - nh_ifp represents logical transmit interface (rt_ifp) (e.g. if * looking up address of interface "ix0" pointer to "lo0" interface * will be returned instead of "ix0") * - nh_ifp represents "address" interface if NHR_IFAIF flag is passed * - howewer mtu from "transmit" interface will be returned. */ int fib4_lookup_nh_ext(uint32_t fibnum, struct in_addr dst, uint32_t flags, uint32_t flowid, struct nhop4_extended *pnh4) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct sockaddr_in sin; struct rtentry *rte; struct nhop_object *nh; KASSERT((fibnum < rt_numfibs), ("fib4_lookup_nh_ext: bad fibnum")); rh = rt_tables_get_rnh(fibnum, AF_INET); if (rh == NULL) return (ENOENT); /* Prepare lookup key */ memset(&sin, 0, sizeof(sin)); sin.sin_len = sizeof(struct sockaddr_in); sin.sin_addr = dst; RIB_RLOCK(rh); rn = rh->rnh_matchaddr((void *)&sin, &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { rte = RNTORT(rn); #ifdef RADIX_MPATH rte = rt_mpath_select(rte, flowid); if (rte == NULL) { RIB_RUNLOCK(rh); return (ENOENT); } #endif nh = rte->rt_nhop; /* Ensure route & ifp is UP */ if (RT_LINK_IS_UP(nh->nh_ifp)) { fib4_rte_to_nh_extended(nh, dst, flags, pnh4); if ((flags & NHR_REF) != 0) { /* TODO: lwref on egress ifp's ? */ } RIB_RUNLOCK(rh); return (0); } } RIB_RUNLOCK(rh); return (ENOENT); } void fib4_free_nh_ext(uint32_t fibnum, struct nhop4_extended *pnh4) { } /* * Looks up path in fib @fibnum specified by @dst. * Returns path nexthop on success. Nexthop is safe to use * within the current network epoch. If longer lifetime is required, * one needs to pass NHR_REF as a flag. This will return referenced * nexthop. */ struct nhop_object * fib4_lookup(uint32_t fibnum, struct in_addr dst, uint32_t scopeid, uint32_t flags, uint32_t flowid) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *rt; struct nhop_object *nh; KASSERT((fibnum < rt_numfibs), ("fib4_lookup: bad fibnum")); rh = rt_tables_get_rnh(fibnum, AF_INET); if (rh == NULL) return (NULL); /* Prepare lookup key */ struct sockaddr_in sin4; memset(&sin4, 0, sizeof(sin4)); sin4.sin_family = AF_INET; sin4.sin_len = sizeof(struct sockaddr_in); sin4.sin_addr = dst; nh = NULL; RIB_RLOCK(rh); rn = rh->rnh_matchaddr((void *)&sin4, &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { rt = RNTORT(rn); #ifdef RADIX_MPATH if (rt_mpath_next(rt) != NULL) rt = rt_mpath_selectrte(rt, flowid); #endif nh = rt->rt_nhop; /* Ensure route & ifp is UP */ if (RT_LINK_IS_UP(nh->nh_ifp)) { if (flags & NHR_REF) nhop_ref_object(nh); RIB_RUNLOCK(rh); return (nh); } } RIB_RUNLOCK(rh); RTSTAT_INC(rts_unreach); return (NULL); } inline static int check_urpf(const struct nhop_object *nh, uint32_t flags, const struct ifnet *src_if) { if (src_if != NULL && nh->nh_aifp == src_if) { return (1); } if (src_if == NULL) { if ((flags & NHR_NODEFAULT) == 0) return (1); else if ((nh->nh_flags & NHF_DEFAULT) == 0) return (1); } return (0); } #ifdef RADIX_MPATH inline static int check_urpf_mpath(struct rtentry *rt, uint32_t flags, const struct ifnet *src_if) { while (rt != NULL) { if (check_urpf(rt->rt_nhop, flags, src_if) != 0) return (1); rt = rt_mpath_next(rt); } return (0); } #endif /* * Performs reverse path forwarding lookup. * If @src_if is non-zero, verifies that at least 1 path goes via * this interface. * If @src_if is zero, verifies that route exist. * if @flags contains NHR_NOTDEFAULT, do not consider default route. * * Returns 1 if route matching conditions is found, 0 otherwise. */ int fib4_check_urpf(uint32_t fibnum, struct in_addr dst, uint32_t scopeid, uint32_t flags, const struct ifnet *src_if) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *rt; int ret; KASSERT((fibnum < rt_numfibs), ("fib4_check_urpf: bad fibnum")); rh = rt_tables_get_rnh(fibnum, AF_INET); if (rh == NULL) return (0); /* Prepare lookup key */ struct sockaddr_in sin4; memset(&sin4, 0, sizeof(sin4)); sin4.sin_len = sizeof(struct sockaddr_in); sin4.sin_addr = dst; RIB_RLOCK(rh); rn = rh->rnh_matchaddr((void *)&sin4, &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { rt = RNTORT(rn); #ifdef RADIX_MPATH ret = check_urpf_mpath(rt, flags, src_if); #else ret = check_urpf(rt->rt_nhop, flags, src_if); #endif RIB_RUNLOCK(rh); return (ret); } RIB_RUNLOCK(rh); return (0); } #endif Index: head/sys/netinet/in_fib.h =================================================================== --- head/sys/netinet/in_fib.h (revision 360291) +++ head/sys/netinet/in_fib.h (revision 360292) @@ -1,66 +1,79 @@ /*- * Copyright (c) 2015 * Alexander V. Chernikov * * 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 University 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 REGENTS 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 REGENTS 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. * * $FreeBSD$ */ #ifndef _NETINET_IN_FIB_H_ #define _NETINET_IN_FIB_H_ +struct route_in { + /* common fields shared among all 'struct route' */ + struct nhop_object *ro_nh; + struct llentry *ro_lle; + char *ro_prepend; + uint16_t ro_plen; + uint16_t ro_flags; + uint16_t ro_mtu; /* saved ro_rt mtu */ + uint16_t spare; + /* custom sockaddr */ + struct sockaddr_in ro_dst4; +}; + /* Basic nexthop info used for uRPF/mtu checks */ struct nhop4_basic { struct ifnet *nh_ifp; /* Logical egress interface */ uint16_t nh_mtu; /* nexthop mtu */ uint16_t nh_flags; /* nhop flags */ struct in_addr nh_addr; /* GW/DST IPv4 address */ }; /* Extended nexthop info used for control protocols */ struct nhop4_extended { struct ifnet *nh_ifp; /* Logical egress interface */ struct in_ifaddr *nh_ia; /* Associated address */ uint16_t nh_mtu; /* nexthop mtu */ uint16_t nh_flags; /* nhop flags */ uint8_t spare[4]; struct in_addr nh_addr; /* GW/DST IPv4 address */ struct in_addr nh_src; /* default source IPv4 address */ uint64_t spare2; }; int fib4_lookup_nh_basic(uint32_t fibnum, struct in_addr dst, uint32_t flags, uint32_t flowid, struct nhop4_basic *pnh4); int fib4_lookup_nh_ext(uint32_t fibnum, struct in_addr dst, uint32_t flags, uint32_t flowid, struct nhop4_extended *pnh4); void fib4_free_nh_ext(uint32_t fibnum, struct nhop4_extended *pnh4); struct nhop_object *fib4_lookup(uint32_t fibnum, struct in_addr dst, uint32_t scopeid, uint32_t flags, uint32_t flowid); int fib4_check_urpf(uint32_t fibnum, struct in_addr dst, uint32_t scopeid, uint32_t flags, const struct ifnet *src_if); #endif Index: head/sys/netinet/ip_input.c =================================================================== --- head/sys/netinet/ip_input.c (revision 360291) +++ head/sys/netinet/ip_input.c (revision 360292) @@ -1,1429 +1,1431 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_bootp.h" #include "opt_ipstealth.h" #include "opt_ipsec.h" #include "opt_route.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #ifdef CTASSERT CTASSERT(sizeof(struct ip) == 20); #endif /* IP reassembly functions are defined in ip_reass.c. */ extern void ipreass_init(void); extern void ipreass_drain(void); extern void ipreass_slowtimo(void); #ifdef VIMAGE extern void ipreass_destroy(void); #endif struct rmlock in_ifaddr_lock; RM_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock"); VNET_DEFINE(int, rsvp_on); VNET_DEFINE(int, ipforwarding); SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipforwarding), 0, "Enable IP forwarding between interfaces"); VNET_DEFINE_STATIC(int, ipsendredirects) = 1; /* XXX */ #define V_ipsendredirects VNET(ipsendredirects) SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsendredirects), 0, "Enable sending IP redirects"); /* * XXX - Setting ip_checkinterface mostly implements the receive side of * the Strong ES model described in RFC 1122, but since the routing table * and transmit implementation do not implement the Strong ES model, * setting this to 1 results in an odd hybrid. * * XXX - ip_checkinterface currently must be disabled if you use ipnat * to translate the destination address to another local interface. * * XXX - ip_checkinterface must be disabled if you add IP aliases * to the loopback interface instead of the interface where the * packets for those addresses are received. */ VNET_DEFINE_STATIC(int, ip_checkinterface); #define V_ip_checkinterface VNET(ip_checkinterface) SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_checkinterface), 0, "Verify packet arrives on correct interface"); VNET_DEFINE(pfil_head_t, inet_pfil_head); /* Packet filter hooks */ static struct netisr_handler ip_nh = { .nh_name = "ip", .nh_handler = ip_input, .nh_proto = NETISR_IP, #ifdef RSS .nh_m2cpuid = rss_soft_m2cpuid_v4, .nh_policy = NETISR_POLICY_CPU, .nh_dispatch = NETISR_DISPATCH_HYBRID, #else .nh_policy = NETISR_POLICY_FLOW, #endif }; #ifdef RSS /* * Directly dispatched frames are currently assumed * to have a flowid already calculated. * * It should likely have something that assert it * actually has valid flow details. */ static struct netisr_handler ip_direct_nh = { .nh_name = "ip_direct", .nh_handler = ip_direct_input, .nh_proto = NETISR_IP_DIRECT, .nh_m2cpuid = rss_soft_m2cpuid_v4, .nh_policy = NETISR_POLICY_CPU, .nh_dispatch = NETISR_DISPATCH_HYBRID, }; #endif extern struct domain inetdomain; extern struct protosw inetsw[]; u_char ip_protox[IPPROTO_MAX]; VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ #ifdef IPCTL_DEFMTU SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, &ip_mtu, 0, "Default MTU"); #endif #ifdef IPSTEALTH VNET_DEFINE(int, ipstealth); SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipstealth), 0, "IP stealth mode, no TTL decrementation on forwarding"); #endif /* * IP statistics are stored in the "array" of counter(9)s. */ VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat); VNET_PCPUSTAT_SYSINIT(ipstat); SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)"); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(ipstat); #endif /* VIMAGE */ /* * Kernel module interface for updating ipstat. The argument is an index * into ipstat treated as an array. */ void kmod_ipstat_inc(int statnum) { counter_u64_add(VNET(ipstat)[statnum], 1); } void kmod_ipstat_dec(int statnum) { counter_u64_add(VNET(ipstat)[statnum], -1); } static int sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&ip_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&ip_nh, qlimit)); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", "Maximum size of the IP input queue"); static int sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) { u_int64_t qdrops_long; int error, qdrops; netisr_getqdrops(&ip_nh, &qdrops_long); qdrops = qdrops_long; error = sysctl_handle_int(oidp, &qdrops, 0, req); if (error || !req->newptr) return (error); if (qdrops != 0) return (EINVAL); netisr_clearqdrops(&ip_nh); return (0); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_netinet_intr_queue_drops, "I", "Number of packets dropped from the IP input queue"); #ifdef RSS static int sysctl_netinet_intr_direct_queue_maxlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&ip_direct_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&ip_direct_nh, qlimit)); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRDQMAXLEN, intr_direct_queue_maxlen, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_netinet_intr_direct_queue_maxlen, "I", "Maximum size of the IP direct input queue"); static int sysctl_netinet_intr_direct_queue_drops(SYSCTL_HANDLER_ARGS) { u_int64_t qdrops_long; int error, qdrops; netisr_getqdrops(&ip_direct_nh, &qdrops_long); qdrops = qdrops_long; error = sysctl_handle_int(oidp, &qdrops, 0, req); if (error || !req->newptr) return (error); if (qdrops != 0) return (EINVAL); netisr_clearqdrops(&ip_direct_nh); return (0); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRDQDROPS, intr_direct_queue_drops, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_netinet_intr_direct_queue_drops, "I", "Number of packets dropped from the IP direct input queue"); #endif /* RSS */ /* * IP initialization: fill in IP protocol switch table. * All protocols not implemented in kernel go to raw IP protocol handler. */ void ip_init(void) { struct pfil_head_args args; struct protosw *pr; int i; CK_STAILQ_INIT(&V_in_ifaddrhead); V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); /* Initialize IP reassembly queue. */ ipreass_init(); /* Initialize packet filter hooks. */ args.pa_version = PFIL_VERSION; args.pa_flags = PFIL_IN | PFIL_OUT; args.pa_type = PFIL_TYPE_IP4; args.pa_headname = PFIL_INET_NAME; V_inet_pfil_head = pfil_head_register(&args); if (hhook_head_register(HHOOK_TYPE_IPSEC_IN, AF_INET, &V_ipsec_hhh_in[HHOOK_IPSEC_INET], HHOOK_WAITOK | HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register input helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_IPSEC_OUT, AF_INET, &V_ipsec_hhh_out[HHOOK_IPSEC_INET], HHOOK_WAITOK | HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register output helper hook\n", __func__); /* Skip initialization of globals for non-default instances. */ #ifdef VIMAGE if (!IS_DEFAULT_VNET(curvnet)) { netisr_register_vnet(&ip_nh); #ifdef RSS netisr_register_vnet(&ip_direct_nh); #endif return; } #endif pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) panic("ip_init: PF_INET not found"); /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ for (i = 0; i < IPPROTO_MAX; i++) ip_protox[i] = pr - inetsw; /* * Cycle through IP protocols and put them into the appropriate place * in ip_protox[]. */ for (pr = inetdomain.dom_protosw; pr < inetdomain.dom_protoswNPROTOSW; pr++) if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { /* Be careful to only index valid IP protocols. */ if (pr->pr_protocol < IPPROTO_MAX) ip_protox[pr->pr_protocol] = pr - inetsw; } netisr_register(&ip_nh); #ifdef RSS netisr_register(&ip_direct_nh); #endif } #ifdef VIMAGE static void ip_destroy(void *unused __unused) { struct ifnet *ifp; int error; #ifdef RSS netisr_unregister_vnet(&ip_direct_nh); #endif netisr_unregister_vnet(&ip_nh); pfil_head_unregister(V_inet_pfil_head); error = hhook_head_deregister(V_ipsec_hhh_in[HHOOK_IPSEC_INET]); if (error != 0) { printf("%s: WARNING: unable to deregister input helper hook " "type HHOOK_TYPE_IPSEC_IN, id HHOOK_IPSEC_INET: " "error %d returned\n", __func__, error); } error = hhook_head_deregister(V_ipsec_hhh_out[HHOOK_IPSEC_INET]); if (error != 0) { printf("%s: WARNING: unable to deregister output helper hook " "type HHOOK_TYPE_IPSEC_OUT, id HHOOK_IPSEC_INET: " "error %d returned\n", __func__, error); } /* Remove the IPv4 addresses from all interfaces. */ in_ifscrub_all(); /* Make sure the IPv4 routes are gone as well. */ IFNET_RLOCK(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) rt_flushifroutes_af(ifp, AF_INET); IFNET_RUNLOCK(); /* Destroy IP reassembly queue. */ ipreass_destroy(); /* Cleanup in_ifaddr hash table; should be empty. */ hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); } VNET_SYSUNINIT(ip, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, ip_destroy, NULL); #endif #ifdef RSS /* * IP direct input routine. * * This is called when reinjecting completed fragments where * all of the previous checking and book-keeping has been done. */ void ip_direct_input(struct mbuf *m) { struct ip *ip; int hlen; ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if (IPSEC_INPUT(ipv4, m, hlen, ip->ip_p) != 0) return; } #endif /* IPSEC */ IPSTAT_INC(ips_delivered); (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); return; } #endif /* * Ip input routine. Checksum and byte swap header. If fragmented * try to reassemble. Process options. Pass to next level. */ void ip_input(struct mbuf *m) { struct rm_priotracker in_ifa_tracker; struct ip *ip = NULL; struct in_ifaddr *ia = NULL; struct ifaddr *ifa; struct ifnet *ifp; int checkif, hlen = 0; uint16_t sum, ip_len; int dchg = 0; /* dest changed after fw */ struct in_addr odst; /* original dst address */ M_ASSERTPKTHDR(m); NET_EPOCH_ASSERT(); if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; /* Set up some basics that will be used later. */ ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; ip_len = ntohs(ip->ip_len); goto ours; } IPSTAT_INC(ips_total); if (m->m_pkthdr.len < sizeof(struct ip)) goto tooshort; if (m->m_len < sizeof (struct ip) && (m = m_pullup(m, sizeof (struct ip))) == NULL) { IPSTAT_INC(ips_toosmall); return; } ip = mtod(m, struct ip *); if (ip->ip_v != IPVERSION) { IPSTAT_INC(ips_badvers); goto bad; } hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) { /* minimum header length */ IPSTAT_INC(ips_badhlen); goto bad; } if (hlen > m->m_len) { if ((m = m_pullup(m, hlen)) == NULL) { IPSTAT_INC(ips_badhlen); return; } ip = mtod(m, struct ip *); } IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL); /* IN_LOOPBACK must not appear on the wire - RFC1122 */ ifp = m->m_pkthdr.rcvif; if (IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) || IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) { if ((ifp->if_flags & IFF_LOOPBACK) == 0) { IPSTAT_INC(ips_badaddr); goto bad; } } if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); } else { if (hlen == sizeof(struct ip)) { sum = in_cksum_hdr(ip); } else { sum = in_cksum(m, hlen); } } if (sum) { IPSTAT_INC(ips_badsum); goto bad; } #ifdef ALTQ if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) /* packet is dropped by traffic conditioner */ return; #endif ip_len = ntohs(ip->ip_len); if (ip_len < hlen) { IPSTAT_INC(ips_badlen); goto bad; } /* * Check that the amount of data in the buffers * is as at least much as the IP header would have us expect. * Trim mbufs if longer than we expect. * Drop packet if shorter than we expect. */ if (m->m_pkthdr.len < ip_len) { tooshort: IPSTAT_INC(ips_tooshort); goto bad; } if (m->m_pkthdr.len > ip_len) { if (m->m_len == m->m_pkthdr.len) { m->m_len = ip_len; m->m_pkthdr.len = ip_len; } else m_adj(m, ip_len - m->m_pkthdr.len); } /* * Try to forward the packet, but if we fail continue. * ip_tryforward() does not generate redirects, so fall * through to normal processing if redirects are required. * ip_tryforward() does inbound and outbound packet firewall * processing. If firewall has decided that destination becomes * our local address, it sets M_FASTFWD_OURS flag. In this * case skip another inbound firewall processing and update * ip pointer. */ if (V_ipforwarding != 0 && V_ipsendredirects == 0 #if defined(IPSEC) || defined(IPSEC_SUPPORT) && (!IPSEC_ENABLED(ipv4) || IPSEC_CAPS(ipv4, m, IPSEC_CAP_OPERABLE) == 0) #endif ) { if ((m = ip_tryforward(m)) == NULL) return; if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; ip = mtod(m, struct ip *); goto ours; } } #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * Bypass packet filtering for packets previously handled by IPsec. */ if (IPSEC_ENABLED(ipv4) && IPSEC_CAPS(ipv4, m, IPSEC_CAP_BYPASS_FILTER) != 0) goto passin; #endif /* * Run through list of hooks for input packets. * * NB: Beware of the destination address changing (e.g. * by NAT rewriting). When this happens, tell * ip_forward to do the right thing. */ /* Jump over all PFIL processing if hooks are not active. */ if (!PFIL_HOOKED_IN(V_inet_pfil_head)) goto passin; odst = ip->ip_dst; if (pfil_run_hooks(V_inet_pfil_head, &m, ifp, PFIL_IN, NULL) != PFIL_PASS) return; if (m == NULL) /* consumed by filter */ return; ip = mtod(m, struct ip *); dchg = (odst.s_addr != ip->ip_dst.s_addr); ifp = m->m_pkthdr.rcvif; if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; goto ours; } if (m->m_flags & M_IP_NEXTHOP) { if (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL) { /* * Directly ship the packet on. This allows * forwarding packets originally destined to us * to some other directly connected host. */ ip_forward(m, 1); return; } } passin: /* * Process options and, if not destined for us, * ship it on. ip_dooptions returns 1 when an * error was detected (causing an icmp message * to be sent and the original packet to be freed). */ if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) return; /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no * matter if it is destined to another node, or whether it is * a multicast one, RSVP wants it! and prevents it from being forwarded * anywhere else. Also checks if the rsvp daemon is running before * grabbing the packet. */ if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) goto ours; /* * Check our list of addresses, to see if the packet is for us. * If we don't have any addresses, assume any unicast packet * we receive might be for us (and let the upper layers deal * with it). */ if (CK_STAILQ_EMPTY(&V_in_ifaddrhead) && (m->m_flags & (M_MCAST|M_BCAST)) == 0) goto ours; /* * Enable a consistency check between the destination address * and the arrival interface for a unicast packet (the RFC 1122 * strong ES model) if IP forwarding is disabled and the packet * is not locally generated and the packet is not subject to * 'ipfw fwd'. * * XXX - Checking also should be disabled if the destination * address is ipnat'ed to a different interface. * * XXX - Checking is incompatible with IP aliases added * to the loopback interface instead of the interface where * the packets are received. * * XXX - This is the case for carp vhost IPs as well so we * insert a workaround. If the packet got here, we already * checked with carp_iamatch() and carp_forus(). */ checkif = V_ip_checkinterface && (V_ipforwarding == 0) && ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && ifp->if_carp == NULL && (dchg == 0); /* * Check for exact addresses in the hash bucket. */ IN_IFADDR_RLOCK(&in_ifa_tracker); LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { /* * If the address matches, verify that the packet * arrived via the correct interface if checking is * enabled. */ if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && (!checkif || ia->ia_ifp == ifp)) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto ours; } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); /* * Check for broadcast addresses. * * Only accept broadcast packets that arrive via the matching * interface. Reception of forwarded directed broadcasts would * be handled via ip_forward() and ether_output() with the loopback * into the stack for SIMPLEX interfaces handled by ether_output(). */ if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = ifatoia(ifa); if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == ip->ip_dst.s_addr) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); goto ours; } #ifdef BOOTP_COMPAT if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); goto ours; } #endif } ia = NULL; } /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { if (V_ip_mrouter) { /* * If we are acting as a multicast router, all * incoming multicast packets are passed to the * kernel-level multicast forwarding function. * The packet is returned (relatively) intact; if * ip_mforward() returns a non-zero value, the packet * must be discarded, else it may be accepted below. */ if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } /* * The process-level routing daemon needs to receive * all multicast IGMP packets, whether or not this * host belongs to their destination groups. */ if (ip->ip_p == IPPROTO_IGMP) goto ours; IPSTAT_INC(ips_forward); } /* * Assume the packet is for us, to avoid prematurely taking * a lock on the in_multi hash. Protocols must perform * their own filtering and update statistics accordingly. */ goto ours; } if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) goto ours; if (ip->ip_dst.s_addr == INADDR_ANY) goto ours; /* * Not for us; forward if possible and desirable. */ if (V_ipforwarding == 0) { IPSTAT_INC(ips_cantforward); m_freem(m); } else { ip_forward(m, dchg); } return; ours: #ifdef IPSTEALTH /* * IPSTEALTH: Process non-routing options only * if the packet is destined for us. */ if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) return; #endif /* IPSTEALTH */ /* * Attempt reassembly; if it succeeds, proceed. * ip_reass() will return a different mbuf. */ if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { /* XXXGL: shouldn't we save & set m_flags? */ m = ip_reass(m); if (m == NULL) return; ip = mtod(m, struct ip *); /* Get the header length of the reassembled packet */ hlen = ip->ip_hl << 2; } #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if (IPSEC_INPUT(ipv4, m, hlen, ip->ip_p) != 0) return; } #endif /* IPSEC */ /* * Switch out to protocol's input routine. */ IPSTAT_INC(ips_delivered); (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); return; bad: m_freem(m); } /* * IP timer processing; * if a timer expires on a reassembly * queue, discard it. */ void ip_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); ipreass_slowtimo(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } void ip_drain(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); ipreass_drain(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * The protocol to be inserted into ip_protox[] must be already registered * in inetsw[], either statically or through pf_proto_register(). */ int ipproto_register(short ipproto) { struct protosw *pr; /* Sanity checks. */ if (ipproto <= 0 || ipproto >= IPPROTO_MAX) return (EPROTONOSUPPORT); /* * The protocol slot must not be occupied by another protocol * already. An index pointing to IPPROTO_RAW is unused. */ pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) return (EPFNOSUPPORT); if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ return (EEXIST); /* Find the protocol position in inetsw[] and set the index. */ for (pr = inetdomain.dom_protosw; pr < inetdomain.dom_protoswNPROTOSW; pr++) { if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol && pr->pr_protocol == ipproto) { ip_protox[pr->pr_protocol] = pr - inetsw; return (0); } } return (EPROTONOSUPPORT); } int ipproto_unregister(short ipproto) { struct protosw *pr; /* Sanity checks. */ if (ipproto <= 0 || ipproto >= IPPROTO_MAX) return (EPROTONOSUPPORT); /* Check if the protocol was indeed registered. */ pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) return (EPFNOSUPPORT); if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ return (ENOENT); /* Reset the protocol slot to IPPROTO_RAW. */ ip_protox[ipproto] = pr - inetsw; return (0); } u_char inetctlerrmap[PRC_NCMDS] = { 0, 0, 0, 0, 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, EMSGSIZE, EHOSTUNREACH, 0, 0, 0, 0, EHOSTUNREACH, 0, ENOPROTOOPT, ECONNREFUSED }; /* * Forward a packet. If some error occurs return the sender * an icmp packet. Note we can't always generate a meaningful * icmp message because icmp doesn't have a large enough repertoire * of codes and types. * * If not forwarding, just drop the packet. This could be confusing * if ipforwarding was zero but some routing protocol was advancing * us as a gateway to somewhere. However, we must let the routing * protocol deal with that. * * The srcrt parameter indicates whether the packet is being forwarded * via a source route. */ void ip_forward(struct mbuf *m, int srcrt) { struct ip *ip = mtod(m, struct ip *); struct in_ifaddr *ia; struct mbuf *mcopy; struct sockaddr_in *sin; struct in_addr dest; struct route ro; int error, type = 0, code = 0, mtu = 0; NET_EPOCH_ASSERT(); if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } if ( #ifdef IPSTEALTH V_ipstealth == 0 && #endif ip->ip_ttl <= IPTTLDEC) { icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0); return; } bzero(&ro, sizeof(ro)); sin = (struct sockaddr_in *)&ro.ro_dst; sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = ip->ip_dst; #ifdef RADIX_MPATH rtalloc_mpath_fib(&ro, ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), M_GETFIB(m)); #else - in_rtalloc_ign(&ro, 0, M_GETFIB(m)); + ro.ro_nh = fib4_lookup(M_GETFIB(m), ip->ip_dst, 0, NHR_REF, + m->m_pkthdr.flowid); #endif - if (ro.ro_rt != NULL) { - ia = ifatoia(ro.ro_rt->rt_ifa); + if (ro.ro_nh != NULL) { + ia = ifatoia(ro.ro_nh->nh_ifa); } else ia = NULL; /* * Save the IP header and at most 8 bytes of the payload, * in case we need to generate an ICMP message to the src. * * XXX this can be optimized a lot by saving the data in a local * buffer on the stack (72 bytes at most), and only allocating the * mbuf if really necessary. The vast majority of the packets * are forwarded without having to send an ICMP back (either * because unnecessary, or because rate limited), so we are * really we are wasting a lot of work here. * * We don't use m_copym() because it might return a reference * to a shared cluster. Both this function and ip_output() * assume exclusive access to the IP header in `m', so any * data in a cluster may change before we reach icmp_error(). */ mcopy = m_gethdr(M_NOWAIT, m->m_type); if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) { /* * It's probably ok if the pkthdr dup fails (because * the deep copy of the tag chain failed), but for now * be conservative and just discard the copy since * code below may some day want the tags. */ m_free(mcopy); mcopy = NULL; } if (mcopy != NULL) { mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy)); mcopy->m_pkthdr.len = mcopy->m_len; m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); } #ifdef IPSTEALTH if (V_ipstealth == 0) #endif ip->ip_ttl -= IPTTLDEC; #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if ((error = IPSEC_FORWARD(ipv4, m)) != 0) { /* mbuf consumed by IPsec */ m_freem(mcopy); if (error != EINPROGRESS) IPSTAT_INC(ips_cantforward); return; } /* No IPsec processing required */ } #endif /* IPSEC */ /* * If forwarding packet using same interface that it came in on, * perhaps should send a redirect to sender to shortcut a hop. * Only send redirect if source is sending directly to us, * and if packet was not source routed (or has any options). * Also, don't send redirect if forwarding using a default route * or a route modified by a redirect. */ dest.s_addr = 0; if (!srcrt && V_ipsendredirects && ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { - struct rtentry *rt; + struct nhop_object *nh; - rt = ro.ro_rt; + nh = ro.ro_nh; - if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && - satosin(rt_key(rt))->sin_addr.s_addr != 0) { -#define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) + if (nh != NULL && ((nh->nh_flags & (NHF_REDIRECT|NHF_DEFAULT)) == 0)) { + struct in_ifaddr *nh_ia = (struct in_ifaddr *)(nh->nh_ifa); u_long src = ntohl(ip->ip_src.s_addr); - if (RTA(rt) && - (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { - if (rt->rt_flags & RTF_GATEWAY) - dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; + if (nh_ia != NULL && + (src & nh_ia->ia_subnetmask) == nh_ia->ia_subnet) { + if (nh->nh_flags & NHF_GATEWAY) + dest.s_addr = nh->gw4_sa.sin_addr.s_addr; else dest.s_addr = ip->ip_dst.s_addr; /* Router requirements says to only send host redirects */ type = ICMP_REDIRECT; code = ICMP_REDIRECT_HOST; } } } error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); - if (error == EMSGSIZE && ro.ro_rt) - mtu = ro.ro_rt->rt_mtu; - RO_RTFREE(&ro); + if (error == EMSGSIZE && ro.ro_nh) + mtu = ro.ro_nh->nh_mtu; + RO_NHFREE(&ro); if (error) IPSTAT_INC(ips_cantforward); else { IPSTAT_INC(ips_forward); if (type) IPSTAT_INC(ips_redirectsent); else { if (mcopy) m_freem(mcopy); return; } } if (mcopy == NULL) return; switch (error) { case 0: /* forwarded, but need redirect */ /* type, code set above */ break; case ENETUNREACH: case EHOSTUNREACH: case ENETDOWN: case EHOSTDOWN: default: type = ICMP_UNREACH; code = ICMP_UNREACH_HOST; break; case EMSGSIZE: type = ICMP_UNREACH; code = ICMP_UNREACH_NEEDFRAG; /* * If the MTU was set before make sure we are below the * interface MTU. * If the MTU wasn't set before use the interface mtu or * fall back to the next smaller mtu step compared to the * current packet size. */ if (mtu != 0) { if (ia != NULL) mtu = min(mtu, ia->ia_ifp->if_mtu); } else { if (ia != NULL) mtu = ia->ia_ifp->if_mtu; else mtu = ip_next_mtu(ntohs(ip->ip_len), 0); } IPSTAT_INC(ips_cantfrag); break; case ENOBUFS: case EACCES: /* ipfw denied packet */ m_freem(mcopy); return; } icmp_error(mcopy, type, code, dest.s_addr, mtu); } #define CHECK_SO_CT(sp, ct) \ (((sp->so_options & SO_TIMESTAMP) && (sp->so_ts_clock == ct)) ? 1 : 0) void ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, struct mbuf *m) { bool stamped; stamped = false; if ((inp->inp_socket->so_options & SO_BINTIME) || CHECK_SO_CT(inp->inp_socket, SO_TS_BINTIME)) { struct bintime boottimebin, bt; struct timespec ts1; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts1); timespec2bintime(&ts1, &bt); getboottimebin(&boottimebin); bintime_add(&bt, &boottimebin); } else { bintime(&bt); } *mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt), SCM_BINTIME, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } if (CHECK_SO_CT(inp->inp_socket, SO_TS_REALTIME_MICRO)) { struct bintime boottimebin, bt1; struct timespec ts1; struct timeval tv; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts1); timespec2bintime(&ts1, &bt1); getboottimebin(&boottimebin); bintime_add(&bt1, &boottimebin); bintime2timeval(&bt1, &tv); } else { microtime(&tv); } *mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv), SCM_TIMESTAMP, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } else if (CHECK_SO_CT(inp->inp_socket, SO_TS_REALTIME)) { struct bintime boottimebin; struct timespec ts, ts1; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts); getboottimebin(&boottimebin); bintime2timespec(&boottimebin, &ts1); timespecadd(&ts, &ts1, &ts); } else { nanotime(&ts); } *mp = sbcreatecontrol((caddr_t)&ts, sizeof(ts), SCM_REALTIME, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } else if (CHECK_SO_CT(inp->inp_socket, SO_TS_MONOTONIC)) { struct timespec ts; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) mbuf_tstmp2timespec(m, &ts); else nanouptime(&ts); *mp = sbcreatecontrol((caddr_t)&ts, sizeof(ts), SCM_MONOTONIC, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } if (stamped && (m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { struct sock_timestamp_info sti; bzero(&sti, sizeof(sti)); sti.st_info_flags = ST_INFO_HW; if ((m->m_flags & M_TSTMP_HPREC) != 0) sti.st_info_flags |= ST_INFO_HW_HPREC; *mp = sbcreatecontrol((caddr_t)&sti, sizeof(sti), SCM_TIME_INFO, SOL_SOCKET); if (*mp != NULL) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVDSTADDR) { *mp = sbcreatecontrol((caddr_t)&ip->ip_dst, sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVTTL) { *mp = sbcreatecontrol((caddr_t)&ip->ip_ttl, sizeof(u_char), IP_RECVTTL, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #ifdef notyet /* XXX * Moving these out of udp_input() made them even more broken * than they already were. */ /* options were tossed already */ if (inp->inp_flags & INP_RECVOPTS) { *mp = sbcreatecontrol((caddr_t)opts_deleted_above, sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } /* ip_srcroute doesn't do what we want here, need to fix */ if (inp->inp_flags & INP_RECVRETOPTS) { *mp = sbcreatecontrol((caddr_t)ip_srcroute(m), sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #endif if (inp->inp_flags & INP_RECVIF) { struct ifnet *ifp; struct sdlbuf { struct sockaddr_dl sdl; u_char pad[32]; } sdlbuf; struct sockaddr_dl *sdp; struct sockaddr_dl *sdl2 = &sdlbuf.sdl; if ((ifp = m->m_pkthdr.rcvif) && ifp->if_index && ifp->if_index <= V_if_index) { sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; /* * Change our mind and don't try copy. */ if (sdp->sdl_family != AF_LINK || sdp->sdl_len > sizeof(sdlbuf)) { goto makedummy; } bcopy(sdp, sdl2, sdp->sdl_len); } else { makedummy: sdl2->sdl_len = offsetof(struct sockaddr_dl, sdl_data[0]); sdl2->sdl_family = AF_LINK; sdl2->sdl_index = 0; sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; } *mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len, IP_RECVIF, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVTOS) { *mp = sbcreatecontrol((caddr_t)&ip->ip_tos, sizeof(u_char), IP_RECVTOS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags2 & INP_RECVFLOWID) { uint32_t flowid, flow_type; flowid = m->m_pkthdr.flowid; flow_type = M_HASHTYPE_GET(m); /* * XXX should handle the failure of one or the * other - don't populate both? */ *mp = sbcreatecontrol((caddr_t) &flowid, sizeof(uint32_t), IP_FLOWID, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; *mp = sbcreatecontrol((caddr_t) &flow_type, sizeof(uint32_t), IP_FLOWTYPE, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #ifdef RSS if (inp->inp_flags2 & INP_RECVRSSBUCKETID) { uint32_t flowid, flow_type; uint32_t rss_bucketid; flowid = m->m_pkthdr.flowid; flow_type = M_HASHTYPE_GET(m); if (rss_hash2bucket(flowid, flow_type, &rss_bucketid) == 0) { *mp = sbcreatecontrol((caddr_t) &rss_bucketid, sizeof(uint32_t), IP_RSSBUCKETID, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } } #endif } /* * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on * locking. This code remains in ip_input.c as ip_mroute.c is optionally * compiled. */ VNET_DEFINE_STATIC(int, ip_rsvp_on); VNET_DEFINE(struct socket *, ip_rsvpd); #define V_ip_rsvp_on VNET(ip_rsvp_on) int ip_rsvp_init(struct socket *so) { if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_RSVP) return EOPNOTSUPP; if (V_ip_rsvpd != NULL) return EADDRINUSE; V_ip_rsvpd = so; /* * This may seem silly, but we need to be sure we don't over-increment * the RSVP counter, in case something slips up. */ if (!V_ip_rsvp_on) { V_ip_rsvp_on = 1; V_rsvp_on++; } return 0; } int ip_rsvp_done(void) { V_ip_rsvpd = NULL; /* * This may seem silly, but we need to be sure we don't over-decrement * the RSVP counter, in case something slips up. */ if (V_ip_rsvp_on) { V_ip_rsvp_on = 0; V_rsvp_on--; } return 0; } int rsvp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m; m = *mp; *mp = NULL; if (rsvp_input_p) { /* call the real one if loaded */ *mp = m; rsvp_input_p(mp, offp, proto); return (IPPROTO_DONE); } /* Can still get packets with rsvp_on = 0 if there is a local member * of the group to which the RSVP packet is addressed. But in this * case we want to throw the packet away. */ if (!V_rsvp_on) { m_freem(m); return (IPPROTO_DONE); } if (V_ip_rsvpd != NULL) { *mp = m; rip_input(mp, offp, proto); return (IPPROTO_DONE); } /* Drop the packet */ m_freem(m); return (IPPROTO_DONE); } Index: head/sys/netinet/ip_output.c =================================================================== --- head/sys/netinet/ip_output.c (revision 360291) +++ head/sys/netinet/ip_output.c (revision 360292) @@ -1,1577 +1,1590 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)ip_output.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_mbuf_stress_test.h" #include "opt_mpath.h" #include "opt_ratelimit.h" #include "opt_route.h" #include "opt_rss.h" #include "opt_sctp.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #ifdef RADIX_MPATH #include #endif #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #ifdef SCTP #include #include #endif #include #include #include #ifdef MBUF_STRESS_TEST static int mbuf_frag_size = 0; SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); #endif static void ip_mloopback(struct ifnet *, const struct mbuf *, int); extern int in_mcast_loop; extern struct protosw inetsw[]; static inline int ip_output_pfil(struct mbuf **mp, struct ifnet *ifp, int flags, struct inpcb *inp, struct sockaddr_in *dst, int *fibnum, int *error) { struct m_tag *fwd_tag = NULL; struct mbuf *m; struct in_addr odst; struct ip *ip; int pflags = PFIL_OUT; if (flags & IP_FORWARDING) pflags |= PFIL_FWD; m = *mp; ip = mtod(m, struct ip *); /* Run through list of hooks for output packets. */ odst.s_addr = ip->ip_dst.s_addr; switch (pfil_run_hooks(V_inet_pfil_head, mp, ifp, pflags, inp)) { case PFIL_DROPPED: *error = EACCES; /* FALLTHROUGH */ case PFIL_CONSUMED: return 1; /* Finished */ case PFIL_PASS: *error = 0; } m = *mp; ip = mtod(m, struct ip *); /* See if destination IP address was changed by packet filter. */ if (odst.s_addr != ip->ip_dst.s_addr) { m->m_flags |= M_SKIP_FIREWALL; /* If destination is now ourself drop to ip_input(). */ if (in_localip(ip->ip_dst)) { m->m_flags |= M_FASTFWD_OURS; if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID; #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif *error = netisr_queue(NETISR_IP, m); return 1; /* Finished */ } bzero(dst, sizeof(*dst)); dst->sin_family = AF_INET; dst->sin_len = sizeof(*dst); dst->sin_addr = ip->ip_dst; return -1; /* Reloop */ } /* See if fib was changed by packet filter. */ if ((*fibnum) != M_GETFIB(m)) { m->m_flags |= M_SKIP_FIREWALL; *fibnum = M_GETFIB(m); return -1; /* Reloop for FIB change */ } /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ if (m->m_flags & M_FASTFWD_OURS) { if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID; *error = netisr_queue(NETISR_IP, m); return 1; /* Finished */ } /* Or forward to some other address? */ if ((m->m_flags & M_IP_NEXTHOP) && ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) { bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); m->m_flags |= M_SKIP_FIREWALL; m->m_flags &= ~M_IP_NEXTHOP; m_tag_delete(m, fwd_tag); return -1; /* Reloop for CHANGE of dst */ } return 0; } static int ip_output_send(struct inpcb *inp, struct ifnet *ifp, struct mbuf *m, const struct sockaddr_in *gw, struct route *ro, bool stamp_tag) { #ifdef KERN_TLS struct ktls_session *tls = NULL; #endif struct m_snd_tag *mst; int error; MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); mst = NULL; #ifdef KERN_TLS /* * If this is an unencrypted TLS record, save a reference to * the record. This local reference is used to call * ktls_output_eagain after the mbuf has been freed (thus * dropping the mbuf's reference) in if_output. */ if (m->m_next != NULL && mbuf_has_tls_session(m->m_next)) { tls = ktls_hold(m->m_next->m_ext_pgs.tls); mst = tls->snd_tag; /* * If a TLS session doesn't have a valid tag, it must * have had an earlier ifp mismatch, so drop this * packet. */ if (mst == NULL) { error = EAGAIN; goto done; } } #endif #ifdef RATELIMIT if (inp != NULL && mst == NULL) { if ((inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) != 0 || (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp)) in_pcboutput_txrtlmt(inp, ifp, m); if (inp->inp_snd_tag != NULL) mst = inp->inp_snd_tag; } #endif if (stamp_tag && mst != NULL) { KASSERT(m->m_pkthdr.rcvif == NULL, ("trying to add a send tag to a forwarded packet")); if (mst->ifp != ifp) { error = EAGAIN; goto done; } /* stamp send tag on mbuf */ m->m_pkthdr.snd_tag = m_snd_tag_ref(mst); m->m_pkthdr.csum_flags |= CSUM_SND_TAG; } error = (*ifp->if_output)(ifp, m, (const struct sockaddr *)gw, ro); done: /* Check for route change invalidating send tags. */ #ifdef KERN_TLS if (tls != NULL) { if (error == EAGAIN) error = ktls_output_eagain(inp, tls); ktls_free(tls); } #endif #ifdef RATELIMIT if (error == EAGAIN) in_pcboutput_eagain(inp); #endif return (error); } +/* rte<>ro_flags translation */ +static inline void +rt_update_ro_flags(struct route *ro) +{ + int nh_flags = ro->ro_nh->nh_flags; + + ro->ro_flags &= ~ (RT_REJECT|RT_BLACKHOLE|RT_HAS_GW); + + ro->ro_flags |= (nh_flags & NHF_REJECT) ? RT_REJECT : 0; + ro->ro_flags |= (nh_flags & NHF_BLACKHOLE) ? RT_BLACKHOLE : 0; + ro->ro_flags |= (nh_flags & NHF_GATEWAY) ? RT_HAS_GW : 0; +} + /* * IP output. The packet in mbuf chain m contains a skeletal IP * header (with len, off, ttl, proto, tos, src, dst). * The mbuf chain containing the packet will be freed. * The mbuf opt, if present, will not be freed. * If route ro is present and has ro_rt initialized, route lookup would be * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL, * then result of route lookup is stored in ro->ro_rt. * * In the IP forwarding case, the packet will arrive with options already * inserted, so must have a NULL opt pointer. */ int ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, struct ip_moptions *imo, struct inpcb *inp) { struct rm_priotracker in_ifa_tracker; struct ip *ip; struct ifnet *ifp = NULL; /* keep compiler happy */ struct mbuf *m0; int hlen = sizeof (struct ip); int mtu; int error = 0; struct sockaddr_in *dst, sin; const struct sockaddr_in *gw; struct in_ifaddr *ia; struct in_addr src; int isbroadcast; uint16_t ip_len, ip_off; uint32_t fibnum; #if defined(IPSEC) || defined(IPSEC_SUPPORT) int no_route_but_check_spd = 0; #endif M_ASSERTPKTHDR(m); NET_EPOCH_ASSERT(); if (inp != NULL) { INP_LOCK_ASSERT(inp); M_SETFIB(m, inp->inp_inc.inc_fibnum); if ((flags & IP_NODEFAULTFLOWID) == 0) { m->m_pkthdr.flowid = inp->inp_flowid; M_HASHTYPE_SET(m, inp->inp_flowtype); } #ifdef NUMA m->m_pkthdr.numa_domain = inp->inp_numa_domain; #endif } if (opt) { int len = 0; m = ip_insertoptions(m, opt, &len); if (len != 0) hlen = len; /* ip->ip_hl is updated above */ } ip = mtod(m, struct ip *); ip_len = ntohs(ip->ip_len); ip_off = ntohs(ip->ip_off); if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { ip->ip_v = IPVERSION; ip->ip_hl = hlen >> 2; ip_fillid(ip); } else { /* Header already set, fetch hlen from there */ hlen = ip->ip_hl << 2; } if ((flags & IP_FORWARDING) == 0) IPSTAT_INC(ips_localout); /* * dst/gw handling: * * gw is readonly but can point either to dst OR rt_gateway, * therefore we need restore gw if we're redoing lookup. */ fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m); if (ro != NULL) dst = (struct sockaddr_in *)&ro->ro_dst; else dst = &sin; - if (ro == NULL || ro->ro_rt == NULL) { + if (ro == NULL || ro->ro_nh == NULL) { bzero(dst, sizeof(*dst)); dst->sin_family = AF_INET; dst->sin_len = sizeof(*dst); dst->sin_addr = ip->ip_dst; } gw = dst; again: /* * Validate route against routing table additions; * a better/more specific route might have been added. */ - if (inp != NULL && ro != NULL && ro->ro_rt != NULL) - RT_VALIDATE(ro, &inp->inp_rt_cookie, fibnum); + if (inp != NULL && ro != NULL && ro->ro_nh != NULL) + NH_VALIDATE(ro, &inp->inp_rt_cookie, fibnum); /* * If there is a cached route, * check that it is to the same destination * and is still up. If not, free it and try again. * The address family should also be checked in case of sharing the * cache with IPv6. * Also check whether routing cache needs invalidation. */ - if (ro != NULL && ro->ro_rt != NULL && - ((ro->ro_rt->rt_flags & RTF_UP) == 0 || - ro->ro_rt->rt_ifp == NULL || !RT_LINK_IS_UP(ro->ro_rt->rt_ifp) || + if (ro != NULL && ro->ro_nh != NULL && + ((!NH_IS_VALID(ro->ro_nh)) || !RT_LINK_IS_UP(ro->ro_nh->nh_ifp) || dst->sin_family != AF_INET || dst->sin_addr.s_addr != ip->ip_dst.s_addr)) RO_INVALIDATE_CACHE(ro); ia = NULL; /* * If routing to interface only, short circuit routing lookup. * The use of an all-ones broadcast address implies this; an * interface is specified by the broadcast address of an interface, * or the destination address of a ptp interface. */ if (flags & IP_SENDONES) { if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst), M_GETFIB(m)))) == NULL && (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst), M_GETFIB(m)))) == NULL) { IPSTAT_INC(ips_noroute); error = ENETUNREACH; goto bad; } ip->ip_dst.s_addr = INADDR_BROADCAST; dst->sin_addr = ip->ip_dst; ifp = ia->ia_ifp; mtu = ifp->if_mtu; ip->ip_ttl = 1; isbroadcast = 1; src = IA_SIN(ia)->sin_addr; } else if (flags & IP_ROUTETOIF) { if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst), M_GETFIB(m)))) == NULL && (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0, M_GETFIB(m)))) == NULL) { IPSTAT_INC(ips_noroute); error = ENETUNREACH; goto bad; } ifp = ia->ia_ifp; mtu = ifp->if_mtu; ip->ip_ttl = 1; isbroadcast = ifp->if_flags & IFF_BROADCAST ? in_ifaddr_broadcast(dst->sin_addr, ia) : 0; src = IA_SIN(ia)->sin_addr; } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && imo != NULL && imo->imo_multicast_ifp != NULL) { /* * Bypass the normal routing lookup for multicast * packets if the interface is specified. */ ifp = imo->imo_multicast_ifp; mtu = ifp->if_mtu; IFP_TO_IA(ifp, ia, &in_ifa_tracker); isbroadcast = 0; /* fool gcc */ /* Interface may have no addresses. */ if (ia != NULL) src = IA_SIN(ia)->sin_addr; else src.s_addr = INADDR_ANY; } else if (ro != NULL) { - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { /* * We want to do any cloning requested by the link * layer, as this is probably required in all cases * for correct operation (as it is for ARP). */ #ifdef RADIX_MPATH rtalloc_mpath_fib(ro, ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), fibnum); #else - in_rtalloc_ign(ro, 0, fibnum); + ro->ro_nh = fib4_lookup(fibnum, dst->sin_addr, 0, + NHR_REF, m->m_pkthdr.flowid); #endif - if (ro->ro_rt == NULL || - (ro->ro_rt->rt_flags & RTF_UP) == 0 || - ro->ro_rt->rt_ifp == NULL || - !RT_LINK_IS_UP(ro->ro_rt->rt_ifp)) { + if (ro->ro_nh == NULL || (!NH_IS_VALID(ro->ro_nh)) || + !RT_LINK_IS_UP(ro->ro_nh->nh_ifp)) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * There is no route for this packet, but it is * possible that a matching SPD entry exists. */ no_route_but_check_spd = 1; mtu = 0; /* Silence GCC warning. */ goto sendit; #endif IPSTAT_INC(ips_noroute); error = EHOSTUNREACH; goto bad; } } - ia = ifatoia(ro->ro_rt->rt_ifa); - ifp = ro->ro_rt->rt_ifp; - counter_u64_add(ro->ro_rt->rt_pksent, 1); + ia = ifatoia(ro->ro_nh->nh_ifa); + ifp = ro->ro_nh->nh_ifp; + counter_u64_add(ro->ro_nh->nh_pksent, 1); rt_update_ro_flags(ro); - if (ro->ro_rt->rt_flags & RTF_GATEWAY) - gw = (struct sockaddr_in *)ro->ro_rt->rt_gateway; - if (ro->ro_rt->rt_flags & RTF_HOST) - isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); + if (ro->ro_nh->nh_flags & NHF_GATEWAY) + gw = &ro->ro_nh->gw4_sa; + if (ro->ro_nh->nh_flags & NHF_HOST) + isbroadcast = (ro->ro_nh->nh_flags & NHF_BROADCAST); else if (ifp->if_flags & IFF_BROADCAST) isbroadcast = in_ifaddr_broadcast(gw->sin_addr, ia); else isbroadcast = 0; - if (ro->ro_rt->rt_flags & RTF_HOST) - mtu = ro->ro_rt->rt_mtu; + if (ro->ro_nh->nh_flags & NHF_HOST) + mtu = ro->ro_nh->nh_mtu; else mtu = ifp->if_mtu; src = IA_SIN(ia)->sin_addr; } else { struct nhop4_extended nh; bzero(&nh, sizeof(nh)); if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh) != 0) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * There is no route for this packet, but it is * possible that a matching SPD entry exists. */ no_route_but_check_spd = 1; mtu = 0; /* Silence GCC warning. */ goto sendit; #endif IPSTAT_INC(ips_noroute); error = EHOSTUNREACH; goto bad; } ifp = nh.nh_ifp; mtu = nh.nh_mtu; /* * We are rewriting here dst to be gw actually, contradicting * comment at the beginning of the function. However, in this * case we are always dealing with on stack dst. * In case if pfil(9) sends us back to beginning of the * function, the dst would be rewritten by ip_output_pfil(). */ MPASS(dst == &sin); dst->sin_addr = nh.nh_addr; ia = nh.nh_ia; src = nh.nh_src; isbroadcast = (((nh.nh_flags & (NHF_HOST | NHF_BROADCAST)) == (NHF_HOST | NHF_BROADCAST)) || ((ifp->if_flags & IFF_BROADCAST) && in_ifaddr_broadcast(dst->sin_addr, ia))); } /* Catch a possible divide by zero later. */ - KASSERT(mtu > 0, ("%s: mtu %d <= 0, ro=%p (rt_flags=0x%08x) ifp=%p", + KASSERT(mtu > 0, ("%s: mtu %d <= 0, ro=%p (nh_flags=0x%08x) ifp=%p", __func__, mtu, ro, - (ro != NULL && ro->ro_rt != NULL) ? ro->ro_rt->rt_flags : 0, ifp)); + (ro != NULL && ro->ro_nh != NULL) ? ro->ro_nh->nh_flags : 0, ifp)); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { m->m_flags |= M_MCAST; /* * IP destination address is multicast. Make sure "gw" * still points to the address in "ro". (It may have been * changed to point to a gateway address, above.) */ gw = dst; /* * See if the caller provided any multicast options */ if (imo != NULL) { ip->ip_ttl = imo->imo_multicast_ttl; if (imo->imo_multicast_vif != -1) ip->ip_src.s_addr = ip_mcast_src ? ip_mcast_src(imo->imo_multicast_vif) : INADDR_ANY; } else ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; /* * Confirm that the outgoing interface supports multicast. */ if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { if ((ifp->if_flags & IFF_MULTICAST) == 0) { IPSTAT_INC(ips_noroute); error = ENETUNREACH; goto bad; } } /* * If source address not specified yet, use address * of outgoing interface. */ if (ip->ip_src.s_addr == INADDR_ANY) ip->ip_src = src; if ((imo == NULL && in_mcast_loop) || (imo && imo->imo_multicast_loop)) { /* * Loop back multicast datagram if not expressly * forbidden to do so, even if we are not a member * of the group; ip_input() will filter it later, * thus deferring a hash lookup and mutex acquisition * at the expense of a cheap copy using m_copym(). */ ip_mloopback(ifp, m, hlen); } else { /* * If we are acting as a multicast router, perform * multicast forwarding as if the packet had just * arrived on the interface to which we are about * to send. The multicast forwarding function * recursively calls this function, using the * IP_FORWARDING flag to prevent infinite recursion. * * Multicasts that are looped back by ip_mloopback(), * above, will be forwarded by the ip_input() routine, * if necessary. */ if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { /* * If rsvp daemon is not running, do not * set ip_moptions. This ensures that the packet * is multicast and not just sent down one link * as prescribed by rsvpd. */ if (!V_rsvp_on) imo = NULL; if (ip_mforward && ip_mforward(ip, ifp, m, imo) != 0) { m_freem(m); goto done; } } } /* * Multicasts with a time-to-live of zero may be looped- * back, above, but must not be transmitted on a network. * Also, multicasts addressed to the loopback interface * are not sent -- the above call to ip_mloopback() will * loop back a copy. ip_input() will drop the copy if * this host does not belong to the destination group on * the loopback interface. */ if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { m_freem(m); goto done; } goto sendit; } /* * If the source address is not specified yet, use the address * of the outoing interface. */ if (ip->ip_src.s_addr == INADDR_ANY) ip->ip_src = src; /* * Look for broadcast address and * verify user is allowed to send * such a packet. */ if (isbroadcast) { if ((ifp->if_flags & IFF_BROADCAST) == 0) { error = EADDRNOTAVAIL; goto bad; } if ((flags & IP_ALLOWBROADCAST) == 0) { error = EACCES; goto bad; } /* don't allow broadcast messages to be fragmented */ if (ip_len > mtu) { error = EMSGSIZE; goto bad; } m->m_flags |= M_BCAST; } else { m->m_flags &= ~M_BCAST; } sendit: #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if ((error = IPSEC_OUTPUT(ipv4, m, inp)) != 0) { if (error == EINPROGRESS) error = 0; goto done; } } /* * Check if there was a route for this packet; return error if not. */ if (no_route_but_check_spd) { IPSTAT_INC(ips_noroute); error = EHOSTUNREACH; goto bad; } /* Update variables that are affected by ipsec4_output(). */ ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; #endif /* IPSEC */ /* Jump over all PFIL processing if hooks are not active. */ if (PFIL_HOOKED_OUT(V_inet_pfil_head)) { switch (ip_output_pfil(&m, ifp, flags, inp, dst, &fibnum, &error)) { case 1: /* Finished */ goto done; case 0: /* Continue normally */ ip = mtod(m, struct ip *); break; case -1: /* Need to try again */ /* Reset everything for a new round */ if (ro != NULL) { - RO_RTFREE(ro); + RO_NHFREE(ro); ro->ro_prepend = NULL; } gw = dst; ip = mtod(m, struct ip *); goto again; } } /* IN_LOOPBACK must not appear on the wire - RFC1122. */ if (IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) || IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) { if ((ifp->if_flags & IFF_LOOPBACK) == 0) { IPSTAT_INC(ips_badaddr); error = EADDRNOTAVAIL; goto bad; } } m->m_pkthdr.csum_flags |= CSUM_IP; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { m = mb_unmapped_to_ext(m); if (m == NULL) { IPSTAT_INC(ips_odropped); error = ENOBUFS; goto bad; } in_delayed_cksum(m); m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } else if ((ifp->if_capenable & IFCAP_NOMAP) == 0) { m = mb_unmapped_to_ext(m); if (m == NULL) { IPSTAT_INC(ips_odropped); error = ENOBUFS; goto bad; } } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { m = mb_unmapped_to_ext(m); if (m == NULL) { IPSTAT_INC(ips_odropped); error = ENOBUFS; goto bad; } sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); m->m_pkthdr.csum_flags &= ~CSUM_SCTP; } #endif /* * If small enough for interface, or the interface will take * care of the fragmentation for us, we can just send directly. */ if (ip_len <= mtu || (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { ip->ip_sum = 0; if (m->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { ip->ip_sum = in_cksum(m, hlen); m->m_pkthdr.csum_flags &= ~CSUM_IP; } /* * Record statistics for this interface address. * With CSUM_TSO the byte/packet count will be slightly * incorrect because we count the IP+TCP headers only * once instead of for every generated packet. */ if (!(flags & IP_FORWARDING) && ia) { if (m->m_pkthdr.csum_flags & CSUM_TSO) counter_u64_add(ia->ia_ifa.ifa_opackets, m->m_pkthdr.len / m->m_pkthdr.tso_segsz); else counter_u64_add(ia->ia_ifa.ifa_opackets, 1); counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len); } #ifdef MBUF_STRESS_TEST if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) m = m_fragment(m, M_NOWAIT, mbuf_frag_size); #endif /* * Reset layer specific mbuf flags * to avoid confusing lower layers. */ m_clrprotoflags(m); IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL); error = ip_output_send(inp, ifp, m, gw, ro, (flags & IP_NO_SND_TAG_RL) ? false : true); goto done; } /* Balk when DF bit is set or the interface didn't support TSO. */ if ((ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { error = EMSGSIZE; IPSTAT_INC(ips_cantfrag); goto bad; } /* * Too large for interface; fragment if possible. If successful, * on return, m will point to a list of packets to be sent. */ error = ip_fragment(ip, &m, mtu, ifp->if_hwassist); if (error) goto bad; for (; m; m = m0) { m0 = m->m_nextpkt; m->m_nextpkt = 0; if (error == 0) { /* Record statistics for this interface address. */ if (ia != NULL) { counter_u64_add(ia->ia_ifa.ifa_opackets, 1); counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len); } /* * Reset layer specific mbuf flags * to avoid confusing upper layers. */ m_clrprotoflags(m); IP_PROBE(send, NULL, NULL, mtod(m, struct ip *), ifp, mtod(m, struct ip *), NULL); error = ip_output_send(inp, ifp, m, gw, ro, true); } else m_freem(m); } if (error == 0) IPSTAT_INC(ips_fragmented); done: return (error); bad: m_freem(m); goto done; } /* * Create a chain of fragments which fit the given mtu. m_frag points to the * mbuf to be fragmented; on return it points to the chain with the fragments. * Return 0 if no error. If error, m_frag may contain a partially built * chain of fragments that should be freed by the caller. * * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) */ int ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, u_long if_hwassist_flags) { int error = 0; int hlen = ip->ip_hl << 2; int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ int off; struct mbuf *m0 = *m_frag; /* the original packet */ int firstlen; struct mbuf **mnext; int nfrags; uint16_t ip_len, ip_off; ip_len = ntohs(ip->ip_len); ip_off = ntohs(ip->ip_off); if (ip_off & IP_DF) { /* Fragmentation not allowed */ IPSTAT_INC(ips_cantfrag); return EMSGSIZE; } /* * Must be able to put at least 8 bytes per fragment. */ if (len < 8) return EMSGSIZE; /* * If the interface will not calculate checksums on * fragmented packets, then do it here. */ if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { m0 = mb_unmapped_to_ext(m0); if (m0 == NULL) { error = ENOBUFS; IPSTAT_INC(ips_odropped); goto done; } in_delayed_cksum(m0); m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } #ifdef SCTP if (m0->m_pkthdr.csum_flags & CSUM_SCTP) { m0 = mb_unmapped_to_ext(m0); if (m0 == NULL) { error = ENOBUFS; IPSTAT_INC(ips_odropped); goto done; } sctp_delayed_cksum(m0, hlen); m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; } #endif if (len > PAGE_SIZE) { /* * Fragment large datagrams such that each segment * contains a multiple of PAGE_SIZE amount of data, * plus headers. This enables a receiver to perform * page-flipping zero-copy optimizations. * * XXX When does this help given that sender and receiver * could have different page sizes, and also mtu could * be less than the receiver's page size ? */ int newlen; off = MIN(mtu, m0->m_pkthdr.len); /* * firstlen (off - hlen) must be aligned on an * 8-byte boundary */ if (off < hlen) goto smart_frag_failure; off = ((off - hlen) & ~7) + hlen; newlen = (~PAGE_MASK) & mtu; if ((newlen + sizeof (struct ip)) > mtu) { /* we failed, go back the default */ smart_frag_failure: newlen = len; off = hlen + len; } len = newlen; } else { off = hlen + len; } firstlen = off - hlen; mnext = &m0->m_nextpkt; /* pointer to next packet */ /* * Loop through length of segment after first fragment, * make new header and copy data of each part and link onto chain. * Here, m0 is the original packet, m is the fragment being created. * The fragments are linked off the m_nextpkt of the original * packet, which after processing serves as the first fragment. */ for (nfrags = 1; off < ip_len; off += len, nfrags++) { struct ip *mhip; /* ip header on the fragment */ struct mbuf *m; int mhlen = sizeof (struct ip); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; IPSTAT_INC(ips_odropped); goto done; } /* * Make sure the complete packet header gets copied * from the originating mbuf to the newly created * mbuf. This also ensures that existing firewall * classification(s), VLAN tags and so on get copied * to the resulting fragmented packet(s): */ if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) { m_free(m); error = ENOBUFS; IPSTAT_INC(ips_odropped); goto done; } /* * In the first mbuf, leave room for the link header, then * copy the original IP header including options. The payload * goes into an additional mbuf chain returned by m_copym(). */ m->m_data += max_linkhdr; mhip = mtod(m, struct ip *); *mhip = *ip; if (hlen > sizeof (struct ip)) { mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); mhip->ip_v = IPVERSION; mhip->ip_hl = mhlen >> 2; } m->m_len = mhlen; /* XXX do we need to add ip_off below ? */ mhip->ip_off = ((off - hlen) >> 3) + ip_off; if (off + len >= ip_len) len = ip_len - off; else mhip->ip_off |= IP_MF; mhip->ip_len = htons((u_short)(len + mhlen)); m->m_next = m_copym(m0, off, len, M_NOWAIT); if (m->m_next == NULL) { /* copy failed */ m_free(m); error = ENOBUFS; /* ??? */ IPSTAT_INC(ips_odropped); goto done; } m->m_pkthdr.len = mhlen + len; #ifdef MAC mac_netinet_fragment(m0, m); #endif mhip->ip_off = htons(mhip->ip_off); mhip->ip_sum = 0; if (m->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) { mhip->ip_sum = in_cksum(m, mhlen); m->m_pkthdr.csum_flags &= ~CSUM_IP; } *mnext = m; mnext = &m->m_nextpkt; } IPSTAT_ADD(ips_ofragments, nfrags); /* * Update first fragment by trimming what's been copied out * and updating header. */ m_adj(m0, hlen + firstlen - ip_len); m0->m_pkthdr.len = hlen + firstlen; ip->ip_len = htons((u_short)m0->m_pkthdr.len); ip->ip_off = htons(ip_off | IP_MF); ip->ip_sum = 0; if (m0->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) { ip->ip_sum = in_cksum(m0, hlen); m0->m_pkthdr.csum_flags &= ~CSUM_IP; } done: *m_frag = m0; return error; } void in_delayed_cksum(struct mbuf *m) { struct ip *ip; struct udphdr *uh; uint16_t cklen, csum, offset; ip = mtod(m, struct ip *); offset = ip->ip_hl << 2 ; if (m->m_pkthdr.csum_flags & CSUM_UDP) { /* if udp header is not in the first mbuf copy udplen */ if (offset + sizeof(struct udphdr) > m->m_len) { m_copydata(m, offset + offsetof(struct udphdr, uh_ulen), sizeof(cklen), (caddr_t)&cklen); cklen = ntohs(cklen); } else { uh = (struct udphdr *)mtodo(m, offset); cklen = ntohs(uh->uh_ulen); } csum = in_cksum_skip(m, cklen + offset, offset); if (csum == 0) csum = 0xffff; } else { cklen = ntohs(ip->ip_len); csum = in_cksum_skip(m, cklen, offset); } offset += m->m_pkthdr.csum_data; /* checksum offset */ if (offset + sizeof(csum) > m->m_len) m_copyback(m, offset, sizeof(csum), (caddr_t)&csum); else *(u_short *)mtodo(m, offset) = csum; } /* * IP socket option processing. */ int ip_ctloutput(struct socket *so, struct sockopt *sopt) { struct inpcb *inp = sotoinpcb(so); int error, optval; #ifdef RSS uint32_t rss_bucket; int retval; #endif error = optval = 0; if (sopt->sopt_level != IPPROTO_IP) { error = EINVAL; if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_dir == SOPT_SET) { switch (sopt->sopt_name) { case SO_REUSEADDR: INP_WLOCK(inp); if ((so->so_options & SO_REUSEADDR) != 0) inp->inp_flags2 |= INP_REUSEADDR; else inp->inp_flags2 &= ~INP_REUSEADDR; INP_WUNLOCK(inp); error = 0; break; case SO_REUSEPORT: INP_WLOCK(inp); if ((so->so_options & SO_REUSEPORT) != 0) inp->inp_flags2 |= INP_REUSEPORT; else inp->inp_flags2 &= ~INP_REUSEPORT; INP_WUNLOCK(inp); error = 0; break; case SO_REUSEPORT_LB: INP_WLOCK(inp); if ((so->so_options & SO_REUSEPORT_LB) != 0) inp->inp_flags2 |= INP_REUSEPORT_LB; else inp->inp_flags2 &= ~INP_REUSEPORT_LB; INP_WUNLOCK(inp); error = 0; break; case SO_SETFIB: INP_WLOCK(inp); inp->inp_inc.inc_fibnum = so->so_fibnum; INP_WUNLOCK(inp); error = 0; break; case SO_MAX_PACING_RATE: #ifdef RATELIMIT INP_WLOCK(inp); inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; INP_WUNLOCK(inp); error = 0; #else error = EOPNOTSUPP; #endif break; default: break; } } return (error); } switch (sopt->sopt_dir) { case SOPT_SET: switch (sopt->sopt_name) { case IP_OPTIONS: #ifdef notyet case IP_RETOPTS: #endif { struct mbuf *m; if (sopt->sopt_valsize > MLEN) { error = EMSGSIZE; break; } m = m_get(sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; break; } m->m_len = sopt->sopt_valsize; error = sooptcopyin(sopt, mtod(m, char *), m->m_len, m->m_len); if (error) { m_free(m); break; } INP_WLOCK(inp); error = ip_pcbopts(inp, sopt->sopt_name, m); INP_WUNLOCK(inp); return (error); } case IP_BINDANY: if (sopt->sopt_td != NULL) { error = priv_check(sopt->sopt_td, PRIV_NETINET_BINDANY); if (error) break; } /* FALLTHROUGH */ case IP_BINDMULTI: #ifdef RSS case IP_RSS_LISTEN_BUCKET: #endif case IP_TOS: case IP_TTL: case IP_MINTTL: case IP_RECVOPTS: case IP_RECVRETOPTS: case IP_ORIGDSTADDR: case IP_RECVDSTADDR: case IP_RECVTTL: case IP_RECVIF: case IP_ONESBCAST: case IP_DONTFRAG: case IP_RECVTOS: case IP_RECVFLOWID: #ifdef RSS case IP_RECVRSSBUCKETID: #endif error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; switch (sopt->sopt_name) { case IP_TOS: inp->inp_ip_tos = optval; break; case IP_TTL: inp->inp_ip_ttl = optval; break; case IP_MINTTL: if (optval >= 0 && optval <= MAXTTL) inp->inp_ip_minttl = optval; else error = EINVAL; break; #define OPTSET(bit) do { \ INP_WLOCK(inp); \ if (optval) \ inp->inp_flags |= bit; \ else \ inp->inp_flags &= ~bit; \ INP_WUNLOCK(inp); \ } while (0) #define OPTSET2(bit, val) do { \ INP_WLOCK(inp); \ if (val) \ inp->inp_flags2 |= bit; \ else \ inp->inp_flags2 &= ~bit; \ INP_WUNLOCK(inp); \ } while (0) case IP_RECVOPTS: OPTSET(INP_RECVOPTS); break; case IP_RECVRETOPTS: OPTSET(INP_RECVRETOPTS); break; case IP_RECVDSTADDR: OPTSET(INP_RECVDSTADDR); break; case IP_ORIGDSTADDR: OPTSET2(INP_ORIGDSTADDR, optval); break; case IP_RECVTTL: OPTSET(INP_RECVTTL); break; case IP_RECVIF: OPTSET(INP_RECVIF); break; case IP_ONESBCAST: OPTSET(INP_ONESBCAST); break; case IP_DONTFRAG: OPTSET(INP_DONTFRAG); break; case IP_BINDANY: OPTSET(INP_BINDANY); break; case IP_RECVTOS: OPTSET(INP_RECVTOS); break; case IP_BINDMULTI: OPTSET2(INP_BINDMULTI, optval); break; case IP_RECVFLOWID: OPTSET2(INP_RECVFLOWID, optval); break; #ifdef RSS case IP_RSS_LISTEN_BUCKET: if ((optval >= 0) && (optval < rss_getnumbuckets())) { inp->inp_rss_listen_bucket = optval; OPTSET2(INP_RSS_BUCKET_SET, 1); } else { error = EINVAL; } break; case IP_RECVRSSBUCKETID: OPTSET2(INP_RECVRSSBUCKETID, optval); break; #endif } break; #undef OPTSET #undef OPTSET2 /* * Multicast socket options are processed by the in_mcast * module. */ case IP_MULTICAST_IF: case IP_MULTICAST_VIF: case IP_MULTICAST_TTL: case IP_MULTICAST_LOOP: case IP_ADD_MEMBERSHIP: case IP_DROP_MEMBERSHIP: case IP_ADD_SOURCE_MEMBERSHIP: case IP_DROP_SOURCE_MEMBERSHIP: case IP_BLOCK_SOURCE: case IP_UNBLOCK_SOURCE: case IP_MSFILTER: case MCAST_JOIN_GROUP: case MCAST_LEAVE_GROUP: case MCAST_JOIN_SOURCE_GROUP: case MCAST_LEAVE_SOURCE_GROUP: case MCAST_BLOCK_SOURCE: case MCAST_UNBLOCK_SOURCE: error = inp_setmoptions(inp, sopt); break; case IP_PORTRANGE: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; INP_WLOCK(inp); switch (optval) { case IP_PORTRANGE_DEFAULT: inp->inp_flags &= ~(INP_LOWPORT); inp->inp_flags &= ~(INP_HIGHPORT); break; case IP_PORTRANGE_HIGH: inp->inp_flags &= ~(INP_LOWPORT); inp->inp_flags |= INP_HIGHPORT; break; case IP_PORTRANGE_LOW: inp->inp_flags &= ~(INP_HIGHPORT); inp->inp_flags |= INP_LOWPORT; break; default: error = EINVAL; break; } INP_WUNLOCK(inp); break; #if defined(IPSEC) || defined(IPSEC_SUPPORT) case IP_IPSEC_POLICY: if (IPSEC_ENABLED(ipv4)) { error = IPSEC_PCBCTL(ipv4, inp, sopt); break; } /* FALLTHROUGH */ #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; case SOPT_GET: switch (sopt->sopt_name) { case IP_OPTIONS: case IP_RETOPTS: INP_RLOCK(inp); if (inp->inp_options) { struct mbuf *options; options = m_copym(inp->inp_options, 0, M_COPYALL, M_NOWAIT); INP_RUNLOCK(inp); if (options != NULL) { error = sooptcopyout(sopt, mtod(options, char *), options->m_len); m_freem(options); } else error = ENOMEM; } else { INP_RUNLOCK(inp); sopt->sopt_valsize = 0; } break; case IP_TOS: case IP_TTL: case IP_MINTTL: case IP_RECVOPTS: case IP_RECVRETOPTS: case IP_ORIGDSTADDR: case IP_RECVDSTADDR: case IP_RECVTTL: case IP_RECVIF: case IP_PORTRANGE: case IP_ONESBCAST: case IP_DONTFRAG: case IP_BINDANY: case IP_RECVTOS: case IP_BINDMULTI: case IP_FLOWID: case IP_FLOWTYPE: case IP_RECVFLOWID: #ifdef RSS case IP_RSSBUCKETID: case IP_RECVRSSBUCKETID: #endif switch (sopt->sopt_name) { case IP_TOS: optval = inp->inp_ip_tos; break; case IP_TTL: optval = inp->inp_ip_ttl; break; case IP_MINTTL: optval = inp->inp_ip_minttl; break; #define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) #define OPTBIT2(bit) (inp->inp_flags2 & bit ? 1 : 0) case IP_RECVOPTS: optval = OPTBIT(INP_RECVOPTS); break; case IP_RECVRETOPTS: optval = OPTBIT(INP_RECVRETOPTS); break; case IP_RECVDSTADDR: optval = OPTBIT(INP_RECVDSTADDR); break; case IP_ORIGDSTADDR: optval = OPTBIT2(INP_ORIGDSTADDR); break; case IP_RECVTTL: optval = OPTBIT(INP_RECVTTL); break; case IP_RECVIF: optval = OPTBIT(INP_RECVIF); break; case IP_PORTRANGE: if (inp->inp_flags & INP_HIGHPORT) optval = IP_PORTRANGE_HIGH; else if (inp->inp_flags & INP_LOWPORT) optval = IP_PORTRANGE_LOW; else optval = 0; break; case IP_ONESBCAST: optval = OPTBIT(INP_ONESBCAST); break; case IP_DONTFRAG: optval = OPTBIT(INP_DONTFRAG); break; case IP_BINDANY: optval = OPTBIT(INP_BINDANY); break; case IP_RECVTOS: optval = OPTBIT(INP_RECVTOS); break; case IP_FLOWID: optval = inp->inp_flowid; break; case IP_FLOWTYPE: optval = inp->inp_flowtype; break; case IP_RECVFLOWID: optval = OPTBIT2(INP_RECVFLOWID); break; #ifdef RSS case IP_RSSBUCKETID: retval = rss_hash2bucket(inp->inp_flowid, inp->inp_flowtype, &rss_bucket); if (retval == 0) optval = rss_bucket; else error = EINVAL; break; case IP_RECVRSSBUCKETID: optval = OPTBIT2(INP_RECVRSSBUCKETID); break; #endif case IP_BINDMULTI: optval = OPTBIT2(INP_BINDMULTI); break; } error = sooptcopyout(sopt, &optval, sizeof optval); break; /* * Multicast socket options are processed by the in_mcast * module. */ case IP_MULTICAST_IF: case IP_MULTICAST_VIF: case IP_MULTICAST_TTL: case IP_MULTICAST_LOOP: case IP_MSFILTER: error = inp_getmoptions(inp, sopt); break; #if defined(IPSEC) || defined(IPSEC_SUPPORT) case IP_IPSEC_POLICY: if (IPSEC_ENABLED(ipv4)) { error = IPSEC_PCBCTL(ipv4, inp, sopt); break; } /* FALLTHROUGH */ #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; } return (error); } /* * Routine called from ip_output() to loop back a copy of an IP multicast * packet to the input queue of a specified interface. Note that this * calls the output routine of the loopback "driver", but with an interface * pointer that might NOT be a loopback interface -- evil, but easier than * replicating that code here. */ static void ip_mloopback(struct ifnet *ifp, const struct mbuf *m, int hlen) { struct ip *ip; struct mbuf *copym; /* * Make a deep copy of the packet because we're going to * modify the pack in order to generate checksums. */ copym = m_dup(m, M_NOWAIT); if (copym != NULL && (!M_WRITABLE(copym) || copym->m_len < hlen)) copym = m_pullup(copym, hlen); if (copym != NULL) { /* If needed, compute the checksum and mark it as valid. */ if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { in_delayed_cksum(copym); copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; copym->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; copym->m_pkthdr.csum_data = 0xffff; } /* * We don't bother to fragment if the IP length is greater * than the interface's MTU. Can this possibly matter? */ ip = mtod(copym, struct ip *); ip->ip_sum = 0; ip->ip_sum = in_cksum(copym, hlen); if_simloop(ifp, copym, AF_INET, 0); } } Index: head/sys/netinet/sctp_asconf.c =================================================================== --- head/sys/netinet/sctp_asconf.c (revision 360291) +++ head/sys/netinet/sctp_asconf.c (revision 360292) @@ -1,3506 +1,3494 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include /* * debug flags: * SCTP_DEBUG_ASCONF1: protocol info, general info and errors * SCTP_DEBUG_ASCONF2: detailed info */ /* * RFC 5061 * * An ASCONF parameter queue exists per asoc which holds the pending address * operations. Lists are updated upon receipt of ASCONF-ACK. * * A restricted_addrs list exists per assoc to hold local addresses that are * not (yet) usable by the assoc as a source address. These addresses are * either pending an ASCONF operation (and exist on the ASCONF parameter * queue), or they are permanently restricted (the peer has returned an * ERROR indication to an ASCONF(ADD), or the peer does not support ASCONF). * * Deleted addresses are always immediately removed from the lists as they will * (shortly) no longer exist in the kernel. We send ASCONFs as a courtesy, * only if allowed. */ /* * ASCONF parameter processing. * response_required: set if a reply is required (eg. SUCCESS_REPORT). * returns a mbuf to an "error" response parameter or NULL/"success" if ok. * FIX: allocating this many mbufs on the fly is pretty inefficient... */ static struct mbuf * sctp_asconf_success_response(uint32_t id) { struct mbuf *m_reply = NULL; struct sctp_asconf_paramhdr *aph; m_reply = sctp_get_mbuf_for_msg(sizeof(struct sctp_asconf_paramhdr), 0, M_NOWAIT, 1, MT_DATA); if (m_reply == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_success_response: couldn't get mbuf!\n"); return (NULL); } aph = mtod(m_reply, struct sctp_asconf_paramhdr *); aph->correlation_id = id; aph->ph.param_type = htons(SCTP_SUCCESS_REPORT); aph->ph.param_length = sizeof(struct sctp_asconf_paramhdr); SCTP_BUF_LEN(m_reply) = aph->ph.param_length; aph->ph.param_length = htons(aph->ph.param_length); return (m_reply); } static struct mbuf * sctp_asconf_error_response(uint32_t id, uint16_t cause, uint8_t *error_tlv, uint16_t tlv_length) { struct mbuf *m_reply = NULL; struct sctp_asconf_paramhdr *aph; struct sctp_error_cause *error; uint32_t buf_len; uint16_t i, param_length, cause_length, padding_length; uint8_t *tlv; if (error_tlv == NULL) { tlv_length = 0; } cause_length = sizeof(struct sctp_error_cause) + tlv_length; param_length = sizeof(struct sctp_asconf_paramhdr) + cause_length; padding_length = tlv_length % 4; if (padding_length != 0) { padding_length = 4 - padding_length; } buf_len = param_length + padding_length; if (buf_len > MLEN) { SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_error_response: tlv_length (%xh) too big\n", tlv_length); return (NULL); } m_reply = sctp_get_mbuf_for_msg(buf_len, 0, M_NOWAIT, 1, MT_DATA); if (m_reply == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_error_response: couldn't get mbuf!\n"); return (NULL); } aph = mtod(m_reply, struct sctp_asconf_paramhdr *); aph->ph.param_type = htons(SCTP_ERROR_CAUSE_IND); aph->ph.param_length = htons(param_length); aph->correlation_id = id; error = (struct sctp_error_cause *)(aph + 1); error->code = htons(cause); error->length = htons(cause_length); if (error_tlv != NULL) { tlv = (uint8_t *)(error + 1); memcpy(tlv, error_tlv, tlv_length); for (i = 0; i < padding_length; i++) { tlv[tlv_length + i] = 0; } } SCTP_BUF_LEN(m_reply) = buf_len; return (m_reply); } static struct mbuf * sctp_process_asconf_add_ip(struct sockaddr *src, struct sctp_asconf_paramhdr *aph, struct sctp_tcb *stcb, int send_hb, int response_required) { struct sctp_nets *net; struct mbuf *m_reply = NULL; union sctp_sockstore store; struct sctp_paramhdr *ph; uint16_t param_type, aparam_length; #if defined(INET) || defined(INET6) uint16_t param_length; #endif struct sockaddr *sa; int zero_address = 0; int bad_address = 0; #ifdef INET struct sockaddr_in *sin; struct sctp_ipv4addr_param *v4addr; #endif #ifdef INET6 struct sockaddr_in6 *sin6; struct sctp_ipv6addr_param *v6addr; #endif aparam_length = ntohs(aph->ph.param_length); if (aparam_length < sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_paramhdr)) { return (NULL); } ph = (struct sctp_paramhdr *)(aph + 1); param_type = ntohs(ph->param_type); #if defined(INET) || defined(INET6) param_length = ntohs(ph->param_length); if (param_length + sizeof(struct sctp_asconf_paramhdr) != aparam_length) { return (NULL); } #endif sa = &store.sa; switch (param_type) { #ifdef INET case SCTP_IPV4_ADDRESS: if (param_length != sizeof(struct sctp_ipv4addr_param)) { /* invalid param size */ return (NULL); } v4addr = (struct sctp_ipv4addr_param *)ph; sin = &store.sin; memset(sin, 0, sizeof(*sin)); sin->sin_family = AF_INET; sin->sin_len = sizeof(struct sockaddr_in); sin->sin_port = stcb->rport; sin->sin_addr.s_addr = v4addr->addr; if ((sin->sin_addr.s_addr == INADDR_BROADCAST) || IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { bad_address = 1; } if (sin->sin_addr.s_addr == INADDR_ANY) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_add_ip: adding "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif #ifdef INET6 case SCTP_IPV6_ADDRESS: if (param_length != sizeof(struct sctp_ipv6addr_param)) { /* invalid param size */ return (NULL); } v6addr = (struct sctp_ipv6addr_param *)ph; sin6 = &store.sin6; memset(sin6, 0, sizeof(*sin6)); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(struct sockaddr_in6); sin6->sin6_port = stcb->rport; memcpy((caddr_t)&sin6->sin6_addr, v6addr->addr, sizeof(struct in6_addr)); if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { bad_address = 1; } if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_add_ip: adding "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif default: m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_INVALID_PARAM, (uint8_t *)aph, aparam_length); return (m_reply); } /* end switch */ /* if 0.0.0.0/::0, add the source address instead */ if (zero_address && SCTP_BASE_SYSCTL(sctp_nat_friendly)) { sa = src; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_add_ip: using source addr "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, src); } net = NULL; /* add the address */ if (bad_address) { m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_INVALID_PARAM, (uint8_t *)aph, aparam_length); } else if (sctp_add_remote_addr(stcb, sa, &net, stcb->asoc.port, SCTP_DONOT_SETSCOPE, SCTP_ADDR_DYNAMIC_ADDED) != 0) { SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_add_ip: error adding address\n"); m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_RESOURCE_SHORTAGE, (uint8_t *)aph, aparam_length); } else { if (response_required) { m_reply = sctp_asconf_success_response(aph->correlation_id); } if (net != NULL) { /* notify upper layer */ sctp_ulp_notify(SCTP_NOTIFY_ASCONF_ADD_IP, stcb, 0, sa, SCTP_SO_NOT_LOCKED); sctp_timer_start(SCTP_TIMER_TYPE_PATHMTURAISE, stcb->sctp_ep, stcb, net); sctp_timer_start(SCTP_TIMER_TYPE_HEARTBEAT, stcb->sctp_ep, stcb, net); if (send_hb) { sctp_send_hb(stcb, net, SCTP_SO_NOT_LOCKED); } } } return (m_reply); } static int sctp_asconf_del_remote_addrs_except(struct sctp_tcb *stcb, struct sockaddr *src) { struct sctp_nets *src_net, *net, *nnet; /* make sure the source address exists as a destination net */ src_net = sctp_findnet(stcb, src); if (src_net == NULL) { /* not found */ return (-1); } /* delete all destination addresses except the source */ TAILQ_FOREACH_SAFE(net, &stcb->asoc.nets, sctp_next, nnet) { if (net != src_net) { /* delete this address */ SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_del_remote_addrs_except: deleting "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, (struct sockaddr *)&net->ro._l_addr); /* notify upper layer */ sctp_ulp_notify(SCTP_NOTIFY_ASCONF_DELETE_IP, stcb, 0, (struct sockaddr *)&net->ro._l_addr, SCTP_SO_NOT_LOCKED); sctp_remove_net(stcb, net); } } return (0); } static struct mbuf * sctp_process_asconf_delete_ip(struct sockaddr *src, struct sctp_asconf_paramhdr *aph, struct sctp_tcb *stcb, int response_required) { struct mbuf *m_reply = NULL; union sctp_sockstore store; struct sctp_paramhdr *ph; uint16_t param_type, aparam_length; #if defined(INET) || defined(INET6) uint16_t param_length; #endif struct sockaddr *sa; int zero_address = 0; int result; #ifdef INET struct sockaddr_in *sin; struct sctp_ipv4addr_param *v4addr; #endif #ifdef INET6 struct sockaddr_in6 *sin6; struct sctp_ipv6addr_param *v6addr; #endif aparam_length = ntohs(aph->ph.param_length); if (aparam_length < sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_paramhdr)) { return (NULL); } ph = (struct sctp_paramhdr *)(aph + 1); param_type = ntohs(ph->param_type); #if defined(INET) || defined(INET6) param_length = ntohs(ph->param_length); if (param_length + sizeof(struct sctp_asconf_paramhdr) != aparam_length) { return (NULL); } #endif sa = &store.sa; switch (param_type) { #ifdef INET case SCTP_IPV4_ADDRESS: if (param_length != sizeof(struct sctp_ipv4addr_param)) { /* invalid param size */ return (NULL); } v4addr = (struct sctp_ipv4addr_param *)ph; sin = &store.sin; memset(sin, 0, sizeof(*sin)); sin->sin_family = AF_INET; sin->sin_len = sizeof(struct sockaddr_in); sin->sin_port = stcb->rport; sin->sin_addr.s_addr = v4addr->addr; if (sin->sin_addr.s_addr == INADDR_ANY) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_delete_ip: deleting "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif #ifdef INET6 case SCTP_IPV6_ADDRESS: if (param_length != sizeof(struct sctp_ipv6addr_param)) { /* invalid param size */ return (NULL); } v6addr = (struct sctp_ipv6addr_param *)ph; sin6 = &store.sin6; memset(sin6, 0, sizeof(*sin6)); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(struct sockaddr_in6); sin6->sin6_port = stcb->rport; memcpy(&sin6->sin6_addr, v6addr->addr, sizeof(struct in6_addr)); if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_delete_ip: deleting "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif default: m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_UNRESOLVABLE_ADDR, (uint8_t *)aph, aparam_length); return (m_reply); } /* make sure the source address is not being deleted */ if (sctp_cmpaddr(sa, src)) { /* trying to delete the source address! */ SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_delete_ip: tried to delete source addr\n"); m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_DELETING_SRC_ADDR, (uint8_t *)aph, aparam_length); return (m_reply); } /* if deleting 0.0.0.0/::0, delete all addresses except src addr */ if (zero_address && SCTP_BASE_SYSCTL(sctp_nat_friendly)) { result = sctp_asconf_del_remote_addrs_except(stcb, src); if (result) { /* src address did not exist? */ SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_delete_ip: src addr does not exist?\n"); /* what error to reply with?? */ m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_REQUEST_REFUSED, (uint8_t *)aph, aparam_length); } else if (response_required) { m_reply = sctp_asconf_success_response(aph->correlation_id); } return (m_reply); } /* delete the address */ result = sctp_del_remote_addr(stcb, sa); /* * note if result == -2, the address doesn't exist in the asoc but * since it's being deleted anyways, we just ack the delete -- but * this probably means something has already gone awry */ if (result == -1) { /* only one address in the asoc */ SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_delete_ip: tried to delete last IP addr!\n"); m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_DELETING_LAST_ADDR, (uint8_t *)aph, aparam_length); } else { if (response_required) { m_reply = sctp_asconf_success_response(aph->correlation_id); } /* notify upper layer */ sctp_ulp_notify(SCTP_NOTIFY_ASCONF_DELETE_IP, stcb, 0, sa, SCTP_SO_NOT_LOCKED); } return (m_reply); } static struct mbuf * sctp_process_asconf_set_primary(struct sockaddr *src, struct sctp_asconf_paramhdr *aph, struct sctp_tcb *stcb, int response_required) { struct mbuf *m_reply = NULL; union sctp_sockstore store; struct sctp_paramhdr *ph; uint16_t param_type, aparam_length; #if defined(INET) || defined(INET6) uint16_t param_length; #endif struct sockaddr *sa; int zero_address = 0; #ifdef INET struct sockaddr_in *sin; struct sctp_ipv4addr_param *v4addr; #endif #ifdef INET6 struct sockaddr_in6 *sin6; struct sctp_ipv6addr_param *v6addr; #endif aparam_length = ntohs(aph->ph.param_length); if (aparam_length < sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_paramhdr)) { return (NULL); } ph = (struct sctp_paramhdr *)(aph + 1); param_type = ntohs(ph->param_type); #if defined(INET) || defined(INET6) param_length = ntohs(ph->param_length); if (param_length + sizeof(struct sctp_asconf_paramhdr) != aparam_length) { return (NULL); } #endif sa = &store.sa; switch (param_type) { #ifdef INET case SCTP_IPV4_ADDRESS: if (param_length != sizeof(struct sctp_ipv4addr_param)) { /* invalid param size */ return (NULL); } v4addr = (struct sctp_ipv4addr_param *)ph; sin = &store.sin; memset(sin, 0, sizeof(*sin)); sin->sin_family = AF_INET; sin->sin_len = sizeof(struct sockaddr_in); sin->sin_addr.s_addr = v4addr->addr; if (sin->sin_addr.s_addr == INADDR_ANY) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_set_primary: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif #ifdef INET6 case SCTP_IPV6_ADDRESS: if (param_length != sizeof(struct sctp_ipv6addr_param)) { /* invalid param size */ return (NULL); } v6addr = (struct sctp_ipv6addr_param *)ph; sin6 = &store.sin6; memset(sin6, 0, sizeof(*sin6)); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(struct sockaddr_in6); memcpy((caddr_t)&sin6->sin6_addr, v6addr->addr, sizeof(struct in6_addr)); if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) zero_address = 1; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_set_primary: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); break; #endif default: m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_UNRESOLVABLE_ADDR, (uint8_t *)aph, aparam_length); return (m_reply); } /* if 0.0.0.0/::0, use the source address instead */ if (zero_address && SCTP_BASE_SYSCTL(sctp_nat_friendly)) { sa = src; SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_set_primary: using source addr "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, src); } /* set the primary address */ if (sctp_set_primary_addr(stcb, sa, NULL) == 0) { SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_set_primary: primary address set\n"); /* notify upper layer */ sctp_ulp_notify(SCTP_NOTIFY_ASCONF_SET_PRIMARY, stcb, 0, sa, SCTP_SO_NOT_LOCKED); if ((stcb->asoc.primary_destination->dest_state & SCTP_ADDR_REACHABLE) && (!(stcb->asoc.primary_destination->dest_state & SCTP_ADDR_PF)) && (stcb->asoc.alternate)) { sctp_free_remote_addr(stcb->asoc.alternate); stcb->asoc.alternate = NULL; } if (response_required) { m_reply = sctp_asconf_success_response(aph->correlation_id); } /* * Mobility adaptation. Ideally, when the reception of SET * PRIMARY with DELETE IP ADDRESS of the previous primary * destination, unacknowledged DATA are retransmitted * immediately to the new primary destination for seamless * handover. If the destination is UNCONFIRMED and marked to * REQ_PRIM, The retransmission occur when reception of the * HEARTBEAT-ACK. (See sctp_handle_heartbeat_ack in * sctp_input.c) Also, when change of the primary * destination, it is better that all subsequent new DATA * containing already queued DATA are transmitted to the new * primary destination. (by micchie) */ if ((sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE) || sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) && sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_PRIM_DELETED) && (stcb->asoc.primary_destination->dest_state & SCTP_ADDR_UNCONFIRMED) == 0) { sctp_timer_stop(SCTP_TIMER_TYPE_PRIM_DELETED, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_ASCONF + SCTP_LOC_1); if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) { sctp_assoc_immediate_retrans(stcb, stcb->asoc.primary_destination); } if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE)) { sctp_move_chunks_from_net(stcb, stcb->asoc.deleted_primary); } sctp_delete_prim_timer(stcb->sctp_ep, stcb); } } else { /* couldn't set the requested primary address! */ SCTPDBG(SCTP_DEBUG_ASCONF1, "process_asconf_set_primary: set primary failed!\n"); /* must have been an invalid address, so report */ m_reply = sctp_asconf_error_response(aph->correlation_id, SCTP_CAUSE_UNRESOLVABLE_ADDR, (uint8_t *)aph, aparam_length); } return (m_reply); } /* * handles an ASCONF chunk. * if all parameters are processed ok, send a plain (empty) ASCONF-ACK */ void sctp_handle_asconf(struct mbuf *m, unsigned int offset, struct sockaddr *src, struct sctp_asconf_chunk *cp, struct sctp_tcb *stcb, int first) { struct sctp_association *asoc; uint32_t serial_num; struct mbuf *n, *m_ack, *m_result, *m_tail; struct sctp_asconf_ack_chunk *ack_cp; struct sctp_asconf_paramhdr *aph; struct sctp_ipv6addr_param *p_addr; unsigned int asconf_limit, cnt; int error = 0; /* did an error occur? */ /* asconf param buffer */ uint8_t aparam_buf[SCTP_PARAM_BUFFER_SIZE]; struct sctp_asconf_ack *ack, *ack_next; /* verify minimum length */ if (ntohs(cp->ch.chunk_length) < sizeof(struct sctp_asconf_chunk)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: chunk too small = %xh\n", ntohs(cp->ch.chunk_length)); return; } asoc = &stcb->asoc; serial_num = ntohl(cp->serial_number); if (SCTP_TSN_GE(asoc->asconf_seq_in, serial_num)) { /* got a duplicate ASCONF */ SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: got duplicate serial number = %xh\n", serial_num); return; } else if (serial_num != (asoc->asconf_seq_in + 1)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: incorrect serial number = %xh (expected next = %xh)\n", serial_num, asoc->asconf_seq_in + 1); return; } /* it's the expected "next" sequence number, so process it */ asoc->asconf_seq_in = serial_num; /* update sequence */ /* get length of all the param's in the ASCONF */ asconf_limit = offset + ntohs(cp->ch.chunk_length); SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: asconf_limit=%u, sequence=%xh\n", asconf_limit, serial_num); if (first) { /* delete old cache */ SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: Now processing first ASCONF. Try to delete old cache\n"); TAILQ_FOREACH_SAFE(ack, &asoc->asconf_ack_sent, next, ack_next) { if (ack->serial_number == serial_num) break; SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: delete old(%u) < first(%u)\n", ack->serial_number, serial_num); TAILQ_REMOVE(&asoc->asconf_ack_sent, ack, next); if (ack->data != NULL) { sctp_m_freem(ack->data); } SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asconf_ack), ack); } } m_ack = sctp_get_mbuf_for_msg(sizeof(struct sctp_asconf_ack_chunk), 0, M_NOWAIT, 1, MT_DATA); if (m_ack == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: couldn't get mbuf!\n"); return; } m_tail = m_ack; /* current reply chain's tail */ /* fill in ASCONF-ACK header */ ack_cp = mtod(m_ack, struct sctp_asconf_ack_chunk *); ack_cp->ch.chunk_type = SCTP_ASCONF_ACK; ack_cp->ch.chunk_flags = 0; ack_cp->serial_number = htonl(serial_num); /* set initial lengths (eg. just an ASCONF-ACK), ntohx at the end! */ SCTP_BUF_LEN(m_ack) = sizeof(struct sctp_asconf_ack_chunk); ack_cp->ch.chunk_length = sizeof(struct sctp_asconf_ack_chunk); /* skip the lookup address parameter */ offset += sizeof(struct sctp_asconf_chunk); p_addr = (struct sctp_ipv6addr_param *)sctp_m_getptr(m, offset, sizeof(struct sctp_paramhdr), (uint8_t *)&aparam_buf); if (p_addr == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: couldn't get lookup addr!\n"); /* respond with a missing/invalid mandatory parameter error */ sctp_m_freem(m_ack); return; } /* skip lookup addr */ offset += SCTP_SIZE32(ntohs(p_addr->ph.param_length)); /* get pointer to first asconf param in ASCONF */ aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(m, offset, sizeof(struct sctp_asconf_paramhdr), (uint8_t *)&aparam_buf); if (aph == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "Empty ASCONF received?\n"); goto send_reply; } /* process through all parameters */ cnt = 0; while (aph != NULL) { unsigned int param_length, param_type; param_type = ntohs(aph->ph.param_type); param_length = ntohs(aph->ph.param_length); if (offset + param_length > asconf_limit) { /* parameter goes beyond end of chunk! */ sctp_m_freem(m_ack); return; } m_result = NULL; if (param_length > sizeof(aparam_buf)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: param length (%u) larger than buffer size!\n", param_length); sctp_m_freem(m_ack); return; } if (param_length <= sizeof(struct sctp_paramhdr)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: param length (%u) too short\n", param_length); sctp_m_freem(m_ack); return; } /* get the entire parameter */ aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(m, offset, param_length, aparam_buf); if (aph == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: couldn't get entire param\n"); sctp_m_freem(m_ack); return; } switch (param_type) { case SCTP_ADD_IP_ADDRESS: m_result = sctp_process_asconf_add_ip(src, aph, stcb, (cnt < SCTP_BASE_SYSCTL(sctp_hb_maxburst)), error); cnt++; break; case SCTP_DEL_IP_ADDRESS: m_result = sctp_process_asconf_delete_ip(src, aph, stcb, error); break; case SCTP_ERROR_CAUSE_IND: /* not valid in an ASCONF chunk */ break; case SCTP_SET_PRIM_ADDR: m_result = sctp_process_asconf_set_primary(src, aph, stcb, error); break; case SCTP_NAT_VTAGS: SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: sees a NAT VTAG state parameter\n"); break; case SCTP_SUCCESS_REPORT: /* not valid in an ASCONF chunk */ break; case SCTP_ULP_ADAPTATION: /* FIX */ break; default: if ((param_type & 0x8000) == 0) { /* Been told to STOP at this param */ asconf_limit = offset; /* * FIX FIX - We need to call * sctp_arethere_unrecognized_parameters() * to get a operr and send it for any * param's with the 0x4000 bit set OR do it * here ourselves... note we still must STOP * if the 0x8000 bit is clear. */ } /* unknown/invalid param type */ break; } /* switch */ /* add any (error) result to the reply mbuf chain */ if (m_result != NULL) { SCTP_BUF_NEXT(m_tail) = m_result; m_tail = m_result; ack_cp->ch.chunk_length += SCTP_BUF_LEN(m_result); /* set flag to force success reports */ error = 1; } offset += SCTP_SIZE32(param_length); /* update remaining ASCONF message length to process */ if (offset >= asconf_limit) { /* no more data in the mbuf chain */ break; } /* get pointer to next asconf param */ aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(m, offset, sizeof(struct sctp_asconf_paramhdr), (uint8_t *)&aparam_buf); if (aph == NULL) { /* can't get an asconf paramhdr */ SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: can't get asconf param hdr!\n"); /* FIX ME - add error here... */ } } send_reply: ack_cp->ch.chunk_length = htons(ack_cp->ch.chunk_length); /* save the ASCONF-ACK reply */ ack = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_asconf_ack), struct sctp_asconf_ack); if (ack == NULL) { sctp_m_freem(m_ack); return; } ack->serial_number = serial_num; ack->last_sent_to = NULL; ack->data = m_ack; ack->len = 0; for (n = m_ack; n != NULL; n = SCTP_BUF_NEXT(n)) { ack->len += SCTP_BUF_LEN(n); } TAILQ_INSERT_TAIL(&stcb->asoc.asconf_ack_sent, ack, next); /* see if last_control_chunk_from is set properly (use IP src addr) */ if (stcb->asoc.last_control_chunk_from == NULL) { /* * this could happen if the source address was just newly * added */ SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: looking up net for IP source address\n"); SCTPDBG(SCTP_DEBUG_ASCONF1, "Looking for IP source: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, src); /* look up the from address */ stcb->asoc.last_control_chunk_from = sctp_findnet(stcb, src); #ifdef SCTP_DEBUG if (stcb->asoc.last_control_chunk_from == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf: IP source address not found?!\n"); } #endif } } /* * does the address match? returns 0 if not, 1 if so */ static uint32_t sctp_asconf_addr_match(struct sctp_asconf_addr *aa, struct sockaddr *sa) { switch (sa->sa_family) { #ifdef INET6 case AF_INET6: { /* XXX scopeid */ struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa; if ((aa->ap.addrp.ph.param_type == SCTP_IPV6_ADDRESS) && (memcmp(&aa->ap.addrp.addr, &sin6->sin6_addr, sizeof(struct in6_addr)) == 0)) { return (1); } break; } #endif #ifdef INET case AF_INET: { struct sockaddr_in *sin = (struct sockaddr_in *)sa; if ((aa->ap.addrp.ph.param_type == SCTP_IPV4_ADDRESS) && (memcmp(&aa->ap.addrp.addr, &sin->sin_addr, sizeof(struct in_addr)) == 0)) { return (1); } break; } #endif default: break; } return (0); } /* * does the address match? returns 0 if not, 1 if so */ static uint32_t sctp_addr_match(struct sctp_paramhdr *ph, struct sockaddr *sa) { #if defined(INET) || defined(INET6) uint16_t param_type, param_length; param_type = ntohs(ph->param_type); param_length = ntohs(ph->param_length); #endif switch (sa->sa_family) { #ifdef INET6 case AF_INET6: { /* XXX scopeid */ struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa; struct sctp_ipv6addr_param *v6addr; v6addr = (struct sctp_ipv6addr_param *)ph; if ((param_type == SCTP_IPV6_ADDRESS) && (param_length == sizeof(struct sctp_ipv6addr_param)) && (memcmp(&v6addr->addr, &sin6->sin6_addr, sizeof(struct in6_addr)) == 0)) { return (1); } break; } #endif #ifdef INET case AF_INET: { struct sockaddr_in *sin = (struct sockaddr_in *)sa; struct sctp_ipv4addr_param *v4addr; v4addr = (struct sctp_ipv4addr_param *)ph; if ((param_type == SCTP_IPV4_ADDRESS) && (param_length == sizeof(struct sctp_ipv4addr_param)) && (memcmp(&v4addr->addr, &sin->sin_addr, sizeof(struct in_addr)) == 0)) { return (1); } break; } #endif default: break; } return (0); } /* * Cleanup for non-responded/OP ERR'd ASCONF */ void sctp_asconf_cleanup(struct sctp_tcb *stcb) { /* * clear out any existing asconfs going out */ sctp_timer_stop(SCTP_TIMER_TYPE_ASCONF, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_ASCONF + SCTP_LOC_2); stcb->asoc.asconf_seq_out_acked = stcb->asoc.asconf_seq_out; /* remove the old ASCONF on our outbound queue */ sctp_toss_old_asconf(stcb); } /* * cleanup any cached source addresses that may be topologically * incorrect after a new address has been added to this interface. */ static void sctp_asconf_nets_cleanup(struct sctp_tcb *stcb, struct sctp_ifn *ifn) { struct sctp_nets *net; /* * Ideally, we want to only clear cached routes and source addresses * that are topologically incorrect. But since there is no easy way * to know whether the newly added address on the ifn would cause a * routing change (i.e. a new egress interface would be chosen) * without doing a new routing lookup and source address selection, * we will (for now) just flush any cached route using a different * ifn (and cached source addrs) and let output re-choose them * during the next send on that net. */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { /* * clear any cached route (and cached source address) if the * route's interface is NOT the same as the address change. * If it's the same interface, just clear the cached source * address. */ if (SCTP_ROUTE_HAS_VALID_IFN(&net->ro) && ((ifn == NULL) || (SCTP_GET_IF_INDEX_FROM_ROUTE(&net->ro) != ifn->ifn_index))) { /* clear any cached route */ - RTFREE(net->ro.ro_rt); - net->ro.ro_rt = NULL; + RO_NHFREE(&net->ro); } /* clear any cached source address */ if (net->src_addr_selected) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } } } void sctp_assoc_immediate_retrans(struct sctp_tcb *stcb, struct sctp_nets *dstnet) { int error; if (dstnet->dest_state & SCTP_ADDR_UNCONFIRMED) { return; } if (stcb->asoc.deleted_primary == NULL) { return; } if (!TAILQ_EMPTY(&stcb->asoc.sent_queue)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "assoc_immediate_retrans: Deleted primary is "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, &stcb->asoc.deleted_primary->ro._l_addr.sa); SCTPDBG(SCTP_DEBUG_ASCONF1, "Current Primary is "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, &stcb->asoc.primary_destination->ro._l_addr.sa); sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep, stcb, stcb->asoc.deleted_primary, SCTP_FROM_SCTP_ASCONF + SCTP_LOC_3); stcb->asoc.num_send_timers_up--; if (stcb->asoc.num_send_timers_up < 0) { stcb->asoc.num_send_timers_up = 0; } SCTP_TCB_LOCK_ASSERT(stcb); error = sctp_t3rxt_timer(stcb->sctp_ep, stcb, stcb->asoc.deleted_primary); if (error) { SCTP_INP_DECR_REF(stcb->sctp_ep); return; } SCTP_TCB_LOCK_ASSERT(stcb); #ifdef SCTP_AUDITING_ENABLED sctp_auditing(4, stcb->sctp_ep, stcb, stcb->asoc.deleted_primary); #endif sctp_chunk_output(stcb->sctp_ep, stcb, SCTP_OUTPUT_FROM_T3, SCTP_SO_NOT_LOCKED); if ((stcb->asoc.num_send_timers_up == 0) && (stcb->asoc.sent_queue_cnt > 0)) { struct sctp_tmit_chunk *chk; chk = TAILQ_FIRST(&stcb->asoc.sent_queue); sctp_timer_start(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep, stcb, chk->whoTo); } } return; } static int sctp_asconf_queue_mgmt(struct sctp_tcb *, struct sctp_ifa *, uint16_t); void sctp_net_immediate_retrans(struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_tmit_chunk *chk; SCTPDBG(SCTP_DEBUG_ASCONF1, "net_immediate_retrans: RTO is %d\n", net->RTO); sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_ASCONF + SCTP_LOC_4); stcb->asoc.cc_functions.sctp_set_initial_cc_param(stcb, net); net->error_count = 0; TAILQ_FOREACH(chk, &stcb->asoc.sent_queue, sctp_next) { if (chk->whoTo == net) { if (chk->sent < SCTP_DATAGRAM_RESEND) { chk->sent = SCTP_DATAGRAM_RESEND; sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); sctp_flight_size_decrease(chk); sctp_total_flight_decrease(stcb, chk); net->marked_retrans++; stcb->asoc.marked_retrans++; } } } if (net->marked_retrans) { sctp_chunk_output(stcb->sctp_ep, stcb, SCTP_OUTPUT_FROM_T3, SCTP_SO_NOT_LOCKED); } } static void sctp_path_check_and_react(struct sctp_tcb *stcb, struct sctp_ifa *newifa) { struct sctp_nets *net; int addrnum, changed; /* * If number of local valid addresses is 1, the valid address is * probably newly added address. Several valid addresses in this * association. A source address may not be changed. Additionally, * they can be configured on a same interface as "alias" addresses. * (by micchie) */ addrnum = sctp_local_addr_count(stcb); SCTPDBG(SCTP_DEBUG_ASCONF1, "p_check_react(): %d local addresses\n", addrnum); if (addrnum == 1) { TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { /* clear any cached route and source address */ - if (net->ro.ro_rt) { - RTFREE(net->ro.ro_rt); - net->ro.ro_rt = NULL; - } + RO_NHFREE(&net->ro); if (net->src_addr_selected) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } /* Retransmit unacknowledged DATA chunks immediately */ if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) { sctp_net_immediate_retrans(stcb, net); } /* also, SET PRIMARY is maybe already sent */ } return; } /* Multiple local addresses exsist in the association. */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { /* clear any cached route and source address */ - if (net->ro.ro_rt) { - RTFREE(net->ro.ro_rt); - net->ro.ro_rt = NULL; - } + RO_NHFREE(&net->ro); if (net->src_addr_selected) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } /* * Check if the nexthop is corresponding to the new address. * If the new address is corresponding to the current * nexthop, the path will be changed. If the new address is * NOT corresponding to the current nexthop, the path will * not be changed. */ SCTP_RTALLOC((sctp_route_t *)&net->ro, stcb->sctp_ep->def_vrf_id, stcb->sctp_ep->fibnum); - if (net->ro.ro_rt == NULL) + if (net->ro.ro_nh == NULL) continue; changed = 0; switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: if (sctp_v4src_match_nexthop(newifa, (sctp_route_t *)&net->ro)) { changed = 1; } break; #endif #ifdef INET6 case AF_INET6: if (sctp_v6src_match_nexthop( &newifa->address.sin6, (sctp_route_t *)&net->ro)) { changed = 1; } break; #endif default: break; } /* * if the newly added address does not relate routing * information, we skip. */ if (changed == 0) continue; /* Retransmit unacknowledged DATA chunks immediately */ if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) { sctp_net_immediate_retrans(stcb, net); } /* Send SET PRIMARY for this new address */ if (net == stcb->asoc.primary_destination) { (void)sctp_asconf_queue_mgmt(stcb, newifa, SCTP_SET_PRIM_ADDR); } } } /* * process an ADD/DELETE IP ack from peer. * addr: corresponding sctp_ifa to the address being added/deleted. * type: SCTP_ADD_IP_ADDRESS or SCTP_DEL_IP_ADDRESS. * flag: 1=success, 0=failure. */ static void sctp_asconf_addr_mgmt_ack(struct sctp_tcb *stcb, struct sctp_ifa *addr, uint32_t flag) { /* * do the necessary asoc list work- if we get a failure indication, * leave the address on the assoc's restricted list. If we get a * success indication, remove the address from the restricted list. */ /* * Note: this will only occur for ADD_IP_ADDRESS, since * DEL_IP_ADDRESS is never actually added to the list... */ if (flag) { /* success case, so remove from the restricted list */ sctp_del_local_addr_restricted(stcb, addr); if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE) || sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) { sctp_path_check_and_react(stcb, addr); return; } /* clear any cached/topologically incorrect source addresses */ sctp_asconf_nets_cleanup(stcb, addr->ifn_p); } /* else, leave it on the list */ } /* * add an asconf add/delete/set primary IP address parameter to the queue. * type = SCTP_ADD_IP_ADDRESS, SCTP_DEL_IP_ADDRESS, SCTP_SET_PRIM_ADDR. * returns 0 if queued, -1 if not queued/removed. * NOTE: if adding, but a delete for the same address is already scheduled * (and not yet sent out), simply remove it from queue. Same for deleting * an address already scheduled for add. If a duplicate operation is found, * ignore the new one. */ static int sctp_asconf_queue_mgmt(struct sctp_tcb *stcb, struct sctp_ifa *ifa, uint16_t type) { struct sctp_asconf_addr *aa, *aa_next; /* make sure the request isn't already in the queue */ TAILQ_FOREACH_SAFE(aa, &stcb->asoc.asconf_queue, next, aa_next) { /* address match? */ if (sctp_asconf_addr_match(aa, &ifa->address.sa) == 0) continue; /* * is the request already in queue but not sent? pass the * request already sent in order to resolve the following * case: 1. arrival of ADD, then sent 2. arrival of DEL. we * can't remove the ADD request already sent 3. arrival of * ADD */ if (aa->ap.aph.ph.param_type == type && aa->sent == 0) { return (-1); } /* is the negative request already in queue, and not sent */ if ((aa->sent == 0) && (type == SCTP_ADD_IP_ADDRESS) && (aa->ap.aph.ph.param_type == SCTP_DEL_IP_ADDRESS)) { /* add requested, delete already queued */ TAILQ_REMOVE(&stcb->asoc.asconf_queue, aa, next); /* remove the ifa from the restricted list */ sctp_del_local_addr_restricted(stcb, ifa); /* free the asconf param */ SCTP_FREE(aa, SCTP_M_ASC_ADDR); SCTPDBG(SCTP_DEBUG_ASCONF2, "asconf_queue_mgmt: add removes queued entry\n"); return (-1); } if ((aa->sent == 0) && (type == SCTP_DEL_IP_ADDRESS) && (aa->ap.aph.ph.param_type == SCTP_ADD_IP_ADDRESS)) { /* delete requested, add already queued */ TAILQ_REMOVE(&stcb->asoc.asconf_queue, aa, next); /* remove the aa->ifa from the restricted list */ sctp_del_local_addr_restricted(stcb, aa->ifa); /* free the asconf param */ SCTP_FREE(aa, SCTP_M_ASC_ADDR); SCTPDBG(SCTP_DEBUG_ASCONF2, "asconf_queue_mgmt: delete removes queued entry\n"); return (-1); } } /* for each aa */ /* adding new request to the queue */ SCTP_MALLOC(aa, struct sctp_asconf_addr *, sizeof(*aa), SCTP_M_ASC_ADDR); if (aa == NULL) { /* didn't get memory */ SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_queue_mgmt: failed to get memory!\n"); return (-1); } aa->special_del = 0; /* fill in asconf address parameter fields */ /* top level elements are "networked" during send */ aa->ap.aph.ph.param_type = type; aa->ifa = ifa; atomic_add_int(&ifa->refcount, 1); /* correlation_id filled in during send routine later... */ switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; sin6 = &ifa->address.sin6; aa->ap.addrp.ph.param_type = SCTP_IPV6_ADDRESS; aa->ap.addrp.ph.param_length = (sizeof(struct sctp_ipv6addr_param)); aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_ipv6addr_param); memcpy(&aa->ap.addrp.addr, &sin6->sin6_addr, sizeof(struct in6_addr)); break; } #endif #ifdef INET case AF_INET: { struct sockaddr_in *sin; sin = &ifa->address.sin; aa->ap.addrp.ph.param_type = SCTP_IPV4_ADDRESS; aa->ap.addrp.ph.param_length = (sizeof(struct sctp_ipv4addr_param)); aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_ipv4addr_param); memcpy(&aa->ap.addrp.addr, &sin->sin_addr, sizeof(struct in_addr)); break; } #endif default: /* invalid family! */ SCTP_FREE(aa, SCTP_M_ASC_ADDR); sctp_free_ifa(ifa); return (-1); } aa->sent = 0; /* clear sent flag */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); #ifdef SCTP_DEBUG if (SCTP_BASE_SYSCTL(sctp_debug_on) & SCTP_DEBUG_ASCONF2) { if (type == SCTP_ADD_IP_ADDRESS) { SCTP_PRINTF("asconf_queue_mgmt: inserted asconf ADD_IP_ADDRESS: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF2, &ifa->address.sa); } else if (type == SCTP_DEL_IP_ADDRESS) { SCTP_PRINTF("asconf_queue_mgmt: appended asconf DEL_IP_ADDRESS: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF2, &ifa->address.sa); } else { SCTP_PRINTF("asconf_queue_mgmt: appended asconf SET_PRIM_ADDR: "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF2, &ifa->address.sa); } } #endif return (0); } /* * add an asconf operation for the given ifa and type. * type = SCTP_ADD_IP_ADDRESS, SCTP_DEL_IP_ADDRESS, SCTP_SET_PRIM_ADDR. * returns 0 if completed, -1 if not completed, 1 if immediate send is * advisable. */ static int sctp_asconf_queue_add(struct sctp_tcb *stcb, struct sctp_ifa *ifa, uint16_t type) { uint32_t status; int pending_delete_queued = 0; int last; /* see if peer supports ASCONF */ if (stcb->asoc.asconf_supported == 0) { return (-1); } /* * if this is deleting the last address from the assoc, mark it as * pending. */ if ((type == SCTP_DEL_IP_ADDRESS) && !stcb->asoc.asconf_del_pending) { if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { last = (sctp_local_addr_count(stcb) == 0); } else { last = (sctp_local_addr_count(stcb) == 1); } if (last) { /* set the pending delete info only */ stcb->asoc.asconf_del_pending = 1; stcb->asoc.asconf_addr_del_pending = ifa; atomic_add_int(&ifa->refcount, 1); SCTPDBG(SCTP_DEBUG_ASCONF2, "asconf_queue_add: mark delete last address pending\n"); return (-1); } } /* queue an asconf parameter */ status = sctp_asconf_queue_mgmt(stcb, ifa, type); /* * if this is an add, and there is a delete also pending (i.e. the * last local address is being changed), queue the pending delete * too. */ if ((type == SCTP_ADD_IP_ADDRESS) && stcb->asoc.asconf_del_pending && (status == 0)) { /* queue in the pending delete */ if (sctp_asconf_queue_mgmt(stcb, stcb->asoc.asconf_addr_del_pending, SCTP_DEL_IP_ADDRESS) == 0) { SCTPDBG(SCTP_DEBUG_ASCONF2, "asconf_queue_add: queuing pending delete\n"); pending_delete_queued = 1; /* clear out the pending delete info */ stcb->asoc.asconf_del_pending = 0; sctp_free_ifa(stcb->asoc.asconf_addr_del_pending); stcb->asoc.asconf_addr_del_pending = NULL; } } if (pending_delete_queued) { struct sctp_nets *net; /* * since we know that the only/last address is now being * changed in this case, reset the cwnd/rto on all nets to * start as a new address and path. Also clear the error * counts to give the assoc the best chance to complete the * address change. */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { stcb->asoc.cc_functions.sctp_set_initial_cc_param(stcb, net); net->RTO = 0; net->error_count = 0; } stcb->asoc.overall_error_count = 0; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) { sctp_misc_ints(SCTP_THRESHOLD_CLEAR, stcb->asoc.overall_error_count, 0, SCTP_FROM_SCTP_ASCONF, __LINE__); } /* queue in an advisory set primary too */ (void)sctp_asconf_queue_mgmt(stcb, ifa, SCTP_SET_PRIM_ADDR); /* let caller know we should send this out immediately */ status = 1; } return (status); } /*- * add an asconf delete IP address parameter to the queue by sockaddr and * possibly with no sctp_ifa available. This is only called by the routine * that checks the addresses in an INIT-ACK against the current address list. * returns 0 if completed, non-zero if not completed. * NOTE: if an add is already scheduled (and not yet sent out), simply * remove it from queue. If a duplicate operation is found, ignore the * new one. */ static int sctp_asconf_queue_sa_delete(struct sctp_tcb *stcb, struct sockaddr *sa) { struct sctp_ifa *ifa; struct sctp_asconf_addr *aa, *aa_next; if (stcb == NULL) { return (-1); } /* see if peer supports ASCONF */ if (stcb->asoc.asconf_supported == 0) { return (-1); } /* make sure the request isn't already in the queue */ TAILQ_FOREACH_SAFE(aa, &stcb->asoc.asconf_queue, next, aa_next) { /* address match? */ if (sctp_asconf_addr_match(aa, sa) == 0) continue; /* is the request already in queue (sent or not) */ if (aa->ap.aph.ph.param_type == SCTP_DEL_IP_ADDRESS) { return (-1); } /* is the negative request already in queue, and not sent */ if (aa->sent == 1) continue; if (aa->ap.aph.ph.param_type == SCTP_ADD_IP_ADDRESS) { /* add already queued, so remove existing entry */ TAILQ_REMOVE(&stcb->asoc.asconf_queue, aa, next); sctp_del_local_addr_restricted(stcb, aa->ifa); /* free the entry */ SCTP_FREE(aa, SCTP_M_ASC_ADDR); return (-1); } } /* for each aa */ /* find any existing ifa-- NOTE ifa CAN be allowed to be NULL */ ifa = sctp_find_ifa_by_addr(sa, stcb->asoc.vrf_id, SCTP_ADDR_NOT_LOCKED); /* adding new request to the queue */ SCTP_MALLOC(aa, struct sctp_asconf_addr *, sizeof(*aa), SCTP_M_ASC_ADDR); if (aa == NULL) { /* didn't get memory */ SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_queue_sa_delete: failed to get memory!\n"); return (-1); } aa->special_del = 0; /* fill in asconf address parameter fields */ /* top level elements are "networked" during send */ aa->ap.aph.ph.param_type = SCTP_DEL_IP_ADDRESS; aa->ifa = ifa; if (ifa) atomic_add_int(&ifa->refcount, 1); /* correlation_id filled in during send routine later... */ switch (sa->sa_family) { #ifdef INET6 case AF_INET6: { /* IPv6 address */ struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)sa; aa->ap.addrp.ph.param_type = SCTP_IPV6_ADDRESS; aa->ap.addrp.ph.param_length = (sizeof(struct sctp_ipv6addr_param)); aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_ipv6addr_param); memcpy(&aa->ap.addrp.addr, &sin6->sin6_addr, sizeof(struct in6_addr)); break; } #endif #ifdef INET case AF_INET: { /* IPv4 address */ struct sockaddr_in *sin = (struct sockaddr_in *)sa; aa->ap.addrp.ph.param_type = SCTP_IPV4_ADDRESS; aa->ap.addrp.ph.param_length = (sizeof(struct sctp_ipv4addr_param)); aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_paramhdr) + sizeof(struct sctp_ipv4addr_param); memcpy(&aa->ap.addrp.addr, &sin->sin_addr, sizeof(struct in_addr)); break; } #endif default: /* invalid family! */ SCTP_FREE(aa, SCTP_M_ASC_ADDR); if (ifa) sctp_free_ifa(ifa); return (-1); } aa->sent = 0; /* clear sent flag */ /* delete goes to the back of the queue */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); /* sa_ignore MEMLEAK {memory is put on the tailq} */ return (0); } /* * find a specific asconf param on our "sent" queue */ static struct sctp_asconf_addr * sctp_asconf_find_param(struct sctp_tcb *stcb, uint32_t correlation_id) { struct sctp_asconf_addr *aa; TAILQ_FOREACH(aa, &stcb->asoc.asconf_queue, next) { if (aa->ap.aph.correlation_id == correlation_id && aa->sent == 1) { /* found it */ return (aa); } } /* didn't find it */ return (NULL); } /* * process an SCTP_ERROR_CAUSE_IND for a ASCONF-ACK parameter and do * notifications based on the error response */ static void sctp_asconf_process_error(struct sctp_tcb *stcb SCTP_UNUSED, struct sctp_asconf_paramhdr *aph) { struct sctp_error_cause *eh; struct sctp_paramhdr *ph; uint16_t param_type; uint16_t error_code; eh = (struct sctp_error_cause *)(aph + 1); ph = (struct sctp_paramhdr *)(eh + 1); /* validate lengths */ if (htons(eh->length) + sizeof(struct sctp_error_cause) > htons(aph->ph.param_length)) { /* invalid error cause length */ SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_process_error: cause element too long\n"); return; } if (htons(ph->param_length) + sizeof(struct sctp_paramhdr) > htons(eh->length)) { /* invalid included TLV length */ SCTPDBG(SCTP_DEBUG_ASCONF1, "asconf_process_error: included TLV too long\n"); return; } /* which error code ? */ error_code = ntohs(eh->code); param_type = ntohs(aph->ph.param_type); /* FIX: this should go back up the REMOTE_ERROR ULP notify */ switch (error_code) { case SCTP_CAUSE_RESOURCE_SHORTAGE: /* we allow ourselves to "try again" for this error */ break; default: /* peer can't handle it... */ switch (param_type) { case SCTP_ADD_IP_ADDRESS: case SCTP_DEL_IP_ADDRESS: case SCTP_SET_PRIM_ADDR: break; default: break; } } } /* * process an asconf queue param. * aparam: parameter to process, will be removed from the queue. * flag: 1=success case, 0=failure case */ static void sctp_asconf_process_param_ack(struct sctp_tcb *stcb, struct sctp_asconf_addr *aparam, uint32_t flag) { uint16_t param_type; /* process this param */ param_type = aparam->ap.aph.ph.param_type; switch (param_type) { case SCTP_ADD_IP_ADDRESS: SCTPDBG(SCTP_DEBUG_ASCONF1, "process_param_ack: added IP address\n"); sctp_asconf_addr_mgmt_ack(stcb, aparam->ifa, flag); break; case SCTP_DEL_IP_ADDRESS: SCTPDBG(SCTP_DEBUG_ASCONF1, "process_param_ack: deleted IP address\n"); /* nothing really to do... lists already updated */ break; case SCTP_SET_PRIM_ADDR: SCTPDBG(SCTP_DEBUG_ASCONF1, "process_param_ack: set primary IP address\n"); /* nothing to do... peer may start using this addr */ break; default: /* should NEVER happen */ break; } /* remove the param and free it */ TAILQ_REMOVE(&stcb->asoc.asconf_queue, aparam, next); if (aparam->ifa) sctp_free_ifa(aparam->ifa); SCTP_FREE(aparam, SCTP_M_ASC_ADDR); } /* * cleanup from a bad asconf ack parameter */ static void sctp_asconf_ack_clear(struct sctp_tcb *stcb SCTP_UNUSED) { /* assume peer doesn't really know how to do asconfs */ /* XXX we could free the pending queue here */ } void sctp_handle_asconf_ack(struct mbuf *m, int offset, struct sctp_asconf_ack_chunk *cp, struct sctp_tcb *stcb, struct sctp_nets *net, int *abort_no_unlock) { struct sctp_association *asoc; uint32_t serial_num; uint16_t ack_length; struct sctp_asconf_paramhdr *aph; struct sctp_asconf_addr *aa, *aa_next; uint32_t last_error_id = 0; /* last error correlation id */ uint32_t id; struct sctp_asconf_addr *ap; /* asconf param buffer */ uint8_t aparam_buf[SCTP_PARAM_BUFFER_SIZE]; /* verify minimum length */ if (ntohs(cp->ch.chunk_length) < sizeof(struct sctp_asconf_ack_chunk)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf_ack: chunk too small = %xh\n", ntohs(cp->ch.chunk_length)); return; } asoc = &stcb->asoc; serial_num = ntohl(cp->serial_number); /* * NOTE: we may want to handle this differently- currently, we will * abort when we get an ack for the expected serial number + 1 (eg. * we didn't send it), process an ack normally if it is the expected * serial number, and re-send the previous ack for *ALL* other * serial numbers */ /* * if the serial number is the next expected, but I didn't send it, * abort the asoc, since someone probably just hijacked us... */ if (serial_num == (asoc->asconf_seq_out + 1)) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf_ack: got unexpected next serial number! Aborting asoc!\n"); snprintf(msg, sizeof(msg), "Never sent serial number %8.8x", serial_num); op_err = sctp_generate_cause(SCTP_CAUSE_PROTOCOL_VIOLATION, msg); sctp_abort_an_association(stcb->sctp_ep, stcb, op_err, SCTP_SO_NOT_LOCKED); *abort_no_unlock = 1; return; } if (serial_num != asoc->asconf_seq_out_acked + 1) { /* got a duplicate/unexpected ASCONF-ACK */ SCTPDBG(SCTP_DEBUG_ASCONF1, "handle_asconf_ack: got duplicate/unexpected serial number = %xh (expected = %xh)\n", serial_num, asoc->asconf_seq_out_acked + 1); return; } if (serial_num == asoc->asconf_seq_out - 1) { /* stop our timer */ sctp_timer_stop(SCTP_TIMER_TYPE_ASCONF, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_ASCONF + SCTP_LOC_5); } /* process the ASCONF-ACK contents */ ack_length = ntohs(cp->ch.chunk_length) - sizeof(struct sctp_asconf_ack_chunk); offset += sizeof(struct sctp_asconf_ack_chunk); /* process through all parameters */ while (ack_length >= sizeof(struct sctp_asconf_paramhdr)) { unsigned int param_length, param_type; /* get pointer to next asconf parameter */ aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(m, offset, sizeof(struct sctp_asconf_paramhdr), aparam_buf); if (aph == NULL) { /* can't get an asconf paramhdr */ sctp_asconf_ack_clear(stcb); return; } param_type = ntohs(aph->ph.param_type); param_length = ntohs(aph->ph.param_length); if (param_length > ack_length) { sctp_asconf_ack_clear(stcb); return; } if (param_length < sizeof(struct sctp_paramhdr)) { sctp_asconf_ack_clear(stcb); return; } /* get the complete parameter... */ if (param_length > sizeof(aparam_buf)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "param length (%u) larger than buffer size!\n", param_length); sctp_asconf_ack_clear(stcb); return; } aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(m, offset, param_length, aparam_buf); if (aph == NULL) { sctp_asconf_ack_clear(stcb); return; } /* correlation_id is transparent to peer, no ntohl needed */ id = aph->correlation_id; switch (param_type) { case SCTP_ERROR_CAUSE_IND: last_error_id = id; /* find the corresponding asconf param in our queue */ ap = sctp_asconf_find_param(stcb, id); if (ap == NULL) { /* hmm... can't find this in our queue! */ break; } /* process the parameter, failed flag */ sctp_asconf_process_param_ack(stcb, ap, 0); /* process the error response */ sctp_asconf_process_error(stcb, aph); break; case SCTP_SUCCESS_REPORT: /* find the corresponding asconf param in our queue */ ap = sctp_asconf_find_param(stcb, id); if (ap == NULL) { /* hmm... can't find this in our queue! */ break; } /* process the parameter, success flag */ sctp_asconf_process_param_ack(stcb, ap, 1); break; default: break; } /* switch */ /* update remaining ASCONF-ACK message length to process */ ack_length -= SCTP_SIZE32(param_length); if (ack_length <= 0) { /* no more data in the mbuf chain */ break; } offset += SCTP_SIZE32(param_length); } /* while */ /* * if there are any "sent" params still on the queue, these are * implicitly "success", or "failed" (if we got an error back) ... * so process these appropriately * * we assume that the correlation_id's are monotonically increasing * beginning from 1 and that we don't have *that* many outstanding * at any given time */ if (last_error_id == 0) last_error_id--; /* set to "max" value */ TAILQ_FOREACH_SAFE(aa, &stcb->asoc.asconf_queue, next, aa_next) { if (aa->sent == 1) { /* * implicitly successful or failed if correlation_id * < last_error_id, then success else, failure */ if (aa->ap.aph.correlation_id < last_error_id) sctp_asconf_process_param_ack(stcb, aa, 1); else sctp_asconf_process_param_ack(stcb, aa, 0); } else { /* * since we always process in order (FIFO queue) if * we reach one that hasn't been sent, the rest * should not have been sent either. so, we're * done... */ break; } } /* update the next sequence number to use */ asoc->asconf_seq_out_acked++; /* remove the old ASCONF on our outbound queue */ sctp_toss_old_asconf(stcb); if (!TAILQ_EMPTY(&stcb->asoc.asconf_queue)) { #ifdef SCTP_TIMER_BASED_ASCONF /* we have more params, so restart our timer */ sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, stcb->sctp_ep, stcb, net); #else /* we have more params, so send out more */ sctp_send_asconf(stcb, net, SCTP_ADDR_NOT_LOCKED); #endif } } #ifdef INET6 static uint32_t sctp_is_scopeid_in_nets(struct sctp_tcb *stcb, struct sockaddr *sa) { struct sockaddr_in6 *sin6, *net6; struct sctp_nets *net; if (sa->sa_family != AF_INET6) { /* wrong family */ return (0); } sin6 = (struct sockaddr_in6 *)sa; if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) == 0) { /* not link local address */ return (0); } /* hunt through our destination nets list for this scope_id */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { if (((struct sockaddr *)(&net->ro._l_addr))->sa_family != AF_INET6) continue; net6 = (struct sockaddr_in6 *)&net->ro._l_addr; if (IN6_IS_ADDR_LINKLOCAL(&net6->sin6_addr) == 0) continue; if (sctp_is_same_scope(sin6, net6)) { /* found one */ return (1); } } /* didn't find one */ return (0); } #endif /* * address management functions */ static void sctp_addr_mgmt_assoc(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_ifa *ifa, uint16_t type, int addr_locked) { int status; if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) == 0 || sctp_is_feature_off(inp, SCTP_PCB_FLAGS_DO_ASCONF)) { /* subset bound, no ASCONF allowed case, so ignore */ return; } /* * note: we know this is not the subset bound, no ASCONF case eg. * this is boundall or subset bound w/ASCONF allowed */ /* first, make sure that the address is IPv4 or IPv6 and not jailed */ switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: if (prison_check_ip6(inp->ip_inp.inp.inp_cred, &ifa->address.sin6.sin6_addr) != 0) { return; } break; #endif #ifdef INET case AF_INET: if (prison_check_ip4(inp->ip_inp.inp.inp_cred, &ifa->address.sin.sin_addr) != 0) { return; } break; #endif default: return; } #ifdef INET6 /* make sure we're "allowed" to add this type of addr */ if (ifa->address.sa.sa_family == AF_INET6) { /* invalid if we're not a v6 endpoint */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) == 0) return; /* is the v6 addr really valid ? */ if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { return; } } #endif /* put this address on the "pending/do not use yet" list */ sctp_add_local_addr_restricted(stcb, ifa); /* * check address scope if address is out of scope, don't queue * anything... note: this would leave the address on both inp and * asoc lists */ switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; sin6 = &ifa->address.sin6; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* we skip unspecifed addresses */ return; } if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) { if (stcb->asoc.scope.local_scope == 0) { return; } /* is it the right link local scope? */ if (sctp_is_scopeid_in_nets(stcb, &ifa->address.sa) == 0) { return; } } if (stcb->asoc.scope.site_scope == 0 && IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr)) { return; } break; } #endif #ifdef INET case AF_INET: { struct sockaddr_in *sin; /* invalid if we are a v6 only endpoint */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) return; sin = &ifa->address.sin; if (sin->sin_addr.s_addr == 0) { /* we skip unspecifed addresses */ return; } if (stcb->asoc.scope.ipv4_local_scope == 0 && IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) { return; } break; } #endif default: /* else, not AF_INET or AF_INET6, so skip */ return; } /* queue an asconf for this address add/delete */ if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_DO_ASCONF)) { /* does the peer do asconf? */ if (stcb->asoc.asconf_supported) { /* queue an asconf for this addr */ status = sctp_asconf_queue_add(stcb, ifa, type); /* * if queued ok, and in the open state, send out the * ASCONF. If in the non-open state, these will be * sent when the state goes open. */ if (status == 0 && ((SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED))) { #ifdef SCTP_TIMER_BASED_ASCONF sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, inp, stcb, stcb->asoc.primary_destination); #else sctp_send_asconf(stcb, NULL, addr_locked); #endif } } } } int sctp_asconf_iterator_ep(struct sctp_inpcb *inp, void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_asconf_iterator *asc; struct sctp_ifa *ifa; struct sctp_laddr *l; int cnt_invalid = 0; asc = (struct sctp_asconf_iterator *)ptr; LIST_FOREACH(l, &asc->list_of_work, sctp_nxt_addr) { ifa = l->ifa; switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: /* invalid if we're not a v6 endpoint */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) == 0) { cnt_invalid++; if (asc->cnt == cnt_invalid) return (1); } break; #endif #ifdef INET case AF_INET: { /* invalid if we are a v6 only endpoint */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) { cnt_invalid++; if (asc->cnt == cnt_invalid) return (1); } break; } #endif default: /* invalid address family */ cnt_invalid++; if (asc->cnt == cnt_invalid) return (1); } } return (0); } static int sctp_asconf_iterator_ep_end(struct sctp_inpcb *inp, void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_ifa *ifa; struct sctp_asconf_iterator *asc; struct sctp_laddr *laddr, *nladdr, *l; /* Only for specific case not bound all */ asc = (struct sctp_asconf_iterator *)ptr; LIST_FOREACH(l, &asc->list_of_work, sctp_nxt_addr) { ifa = l->ifa; if (l->action == SCTP_ADD_IP_ADDRESS) { LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == ifa) { laddr->action = 0; break; } } } else if (l->action == SCTP_DEL_IP_ADDRESS) { LIST_FOREACH_SAFE(laddr, &inp->sctp_addr_list, sctp_nxt_addr, nladdr) { /* remove only after all guys are done */ if (laddr->ifa == ifa) { sctp_del_local_addr_ep(inp, ifa); } } } } return (0); } void sctp_asconf_iterator_stcb(struct sctp_inpcb *inp, struct sctp_tcb *stcb, void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_asconf_iterator *asc; struct sctp_ifa *ifa; struct sctp_laddr *l; int cnt_invalid = 0; int type, status; int num_queued = 0; asc = (struct sctp_asconf_iterator *)ptr; LIST_FOREACH(l, &asc->list_of_work, sctp_nxt_addr) { ifa = l->ifa; type = l->action; /* address's vrf_id must be the vrf_id of the assoc */ if (ifa->vrf_id != stcb->asoc.vrf_id) { continue; } /* Same checks again for assoc */ switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: { /* invalid if we're not a v6 endpoint */ struct sockaddr_in6 *sin6; if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) == 0) { cnt_invalid++; if (asc->cnt == cnt_invalid) return; else continue; } sin6 = &ifa->address.sin6; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* we skip unspecifed addresses */ continue; } if (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { continue; } if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) { if (stcb->asoc.scope.local_scope == 0) { continue; } /* is it the right link local scope? */ if (sctp_is_scopeid_in_nets(stcb, &ifa->address.sa) == 0) { continue; } } break; } #endif #ifdef INET case AF_INET: { /* invalid if we are a v6 only endpoint */ struct sockaddr_in *sin; /* invalid if we are a v6 only endpoint */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) continue; sin = &ifa->address.sin; if (sin->sin_addr.s_addr == 0) { /* we skip unspecifed addresses */ continue; } if (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { continue; } if (stcb->asoc.scope.ipv4_local_scope == 0 && IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) { continue; } if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) { cnt_invalid++; if (asc->cnt == cnt_invalid) return; else continue; } break; } #endif default: /* invalid address family */ cnt_invalid++; if (asc->cnt == cnt_invalid) return; else continue; break; } if (type == SCTP_ADD_IP_ADDRESS) { /* prevent this address from being used as a source */ sctp_add_local_addr_restricted(stcb, ifa); } else if (type == SCTP_DEL_IP_ADDRESS) { struct sctp_nets *net; TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { - sctp_rtentry_t *rt; /* delete this address if cached */ if (net->ro._s_addr == ifa) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; - rt = net->ro.ro_rt; - if (rt) { - RTFREE(rt); - net->ro.ro_rt = NULL; - } + RO_NHFREE(&net->ro); /* * Now we deleted our src address, * should we not also now reset the * cwnd/rto to start as if its a new * address? */ stcb->asoc.cc_functions.sctp_set_initial_cc_param(stcb, net); net->RTO = 0; } } } else if (type == SCTP_SET_PRIM_ADDR) { if ((stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) == 0) { /* must validate the ifa is in the ep */ if (sctp_is_addr_in_ep(stcb->sctp_ep, ifa) == 0) { continue; } } else { /* Need to check scopes for this guy */ if (sctp_is_address_in_scope(ifa, &stcb->asoc.scope, 0) == 0) { continue; } } } /* queue an asconf for this address add/delete */ if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_DO_ASCONF) && stcb->asoc.asconf_supported == 1) { /* queue an asconf for this addr */ status = sctp_asconf_queue_add(stcb, ifa, type); /* * if queued ok, and in the open state, update the * count of queued params. If in the non-open * state, these get sent when the assoc goes open. */ if ((SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED)) { if (status >= 0) { num_queued++; } } } } /* * If we have queued params in the open state, send out an ASCONF. */ if (num_queued > 0) { sctp_send_asconf(stcb, NULL, SCTP_ADDR_NOT_LOCKED); } } void sctp_asconf_iterator_end(void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_asconf_iterator *asc; struct sctp_ifa *ifa; struct sctp_laddr *l, *nl; asc = (struct sctp_asconf_iterator *)ptr; LIST_FOREACH_SAFE(l, &asc->list_of_work, sctp_nxt_addr, nl) { ifa = l->ifa; if (l->action == SCTP_ADD_IP_ADDRESS) { /* Clear the defer use flag */ ifa->localifa_flags &= ~SCTP_ADDR_DEFER_USE; } sctp_free_ifa(ifa); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_laddr), l); SCTP_DECR_LADDR_COUNT(); } SCTP_FREE(asc, SCTP_M_ASC_IT); } /* * sa is the sockaddr to ask the peer to set primary to. * returns: 0 = completed, -1 = error */ int32_t sctp_set_primary_ip_address_sa(struct sctp_tcb *stcb, struct sockaddr *sa) { uint32_t vrf_id; struct sctp_ifa *ifa; /* find the ifa for the desired set primary */ vrf_id = stcb->asoc.vrf_id; ifa = sctp_find_ifa_by_addr(sa, vrf_id, SCTP_ADDR_NOT_LOCKED); if (ifa == NULL) { /* Invalid address */ return (-1); } /* queue an ASCONF:SET_PRIM_ADDR to be sent */ if (!sctp_asconf_queue_add(stcb, ifa, SCTP_SET_PRIM_ADDR)) { /* set primary queuing succeeded */ SCTPDBG(SCTP_DEBUG_ASCONF1, "set_primary_ip_address_sa: queued on tcb=%p, ", (void *)stcb); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); if ((SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED)) { #ifdef SCTP_TIMER_BASED_ASCONF sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, stcb->sctp_ep, stcb, stcb->asoc.primary_destination); #else sctp_send_asconf(stcb, NULL, SCTP_ADDR_NOT_LOCKED); #endif } } else { SCTPDBG(SCTP_DEBUG_ASCONF1, "set_primary_ip_address_sa: failed to add to queue on tcb=%p, ", (void *)stcb); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, sa); return (-1); } return (0); } int sctp_is_addr_pending(struct sctp_tcb *stcb, struct sctp_ifa *sctp_ifa) { struct sctp_tmit_chunk *chk, *nchk; unsigned int offset, asconf_limit; struct sctp_asconf_chunk *acp; struct sctp_asconf_paramhdr *aph; uint8_t aparam_buf[SCTP_PARAM_BUFFER_SIZE]; struct sctp_paramhdr *ph; int add_cnt, del_cnt; uint16_t last_param_type; add_cnt = del_cnt = 0; last_param_type = 0; TAILQ_FOREACH_SAFE(chk, &stcb->asoc.asconf_send_queue, sctp_next, nchk) { if (chk->data == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: No mbuf data?\n"); continue; } offset = 0; acp = mtod(chk->data, struct sctp_asconf_chunk *); offset += sizeof(struct sctp_asconf_chunk); asconf_limit = ntohs(acp->ch.chunk_length); ph = (struct sctp_paramhdr *)sctp_m_getptr(chk->data, offset, sizeof(struct sctp_paramhdr), aparam_buf); if (ph == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: couldn't get lookup addr!\n"); continue; } offset += ntohs(ph->param_length); aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(chk->data, offset, sizeof(struct sctp_asconf_paramhdr), aparam_buf); if (aph == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: Empty ASCONF will be sent?\n"); continue; } while (aph != NULL) { unsigned int param_length, param_type; param_type = ntohs(aph->ph.param_type); param_length = ntohs(aph->ph.param_length); if (offset + param_length > asconf_limit) { /* parameter goes beyond end of chunk! */ break; } if (param_length > sizeof(aparam_buf)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: param length (%u) larger than buffer size!\n", param_length); break; } if (param_length <= sizeof(struct sctp_paramhdr)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: param length(%u) too short\n", param_length); break; } aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(chk->data, offset, param_length, aparam_buf); if (aph == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "is_addr_pending: couldn't get entire param\n"); break; } ph = (struct sctp_paramhdr *)(aph + 1); if (sctp_addr_match(ph, &sctp_ifa->address.sa) != 0) { switch (param_type) { case SCTP_ADD_IP_ADDRESS: add_cnt++; break; case SCTP_DEL_IP_ADDRESS: del_cnt++; break; default: break; } last_param_type = param_type; } offset += SCTP_SIZE32(param_length); if (offset >= asconf_limit) { /* no more data in the mbuf chain */ break; } /* get pointer to next asconf param */ aph = (struct sctp_asconf_paramhdr *)sctp_m_getptr(chk->data, offset, sizeof(struct sctp_asconf_paramhdr), aparam_buf); } } /* * we want to find the sequences which consist of ADD -> DEL -> ADD * or DEL -> ADD */ if (add_cnt > del_cnt || (add_cnt == del_cnt && last_param_type == SCTP_ADD_IP_ADDRESS)) { return (1); } return (0); } static struct sockaddr * sctp_find_valid_localaddr(struct sctp_tcb *stcb, int addr_locked) { struct sctp_vrf *vrf = NULL; struct sctp_ifn *sctp_ifn; struct sctp_ifa *sctp_ifa; if (addr_locked == SCTP_ADDR_NOT_LOCKED) SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(stcb->asoc.vrf_id); if (vrf == NULL) { if (addr_locked == SCTP_ADDR_NOT_LOCKED) SCTP_IPI_ADDR_RUNLOCK(); return (NULL); } LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { if (stcb->asoc.scope.loopback_scope == 0 && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { /* Skip if loopback_scope not set */ continue; } LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { switch (sctp_ifa->address.sa.sa_family) { #ifdef INET case AF_INET: if (stcb->asoc.scope.ipv4_addr_legal) { struct sockaddr_in *sin; sin = &sctp_ifa->address.sin; if (sin->sin_addr.s_addr == 0) { /* skip unspecifed addresses */ continue; } if (prison_check_ip4(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { continue; } if (stcb->asoc.scope.ipv4_local_scope == 0 && IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) continue; if (sctp_is_addr_restricted(stcb, sctp_ifa) && (!sctp_is_addr_pending(stcb, sctp_ifa))) continue; /* * found a valid local v4 address to * use */ if (addr_locked == SCTP_ADDR_NOT_LOCKED) SCTP_IPI_ADDR_RUNLOCK(); return (&sctp_ifa->address.sa); } break; #endif #ifdef INET6 case AF_INET6: if (stcb->asoc.scope.ipv6_addr_legal) { struct sockaddr_in6 *sin6; if (sctp_ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { continue; } sin6 = &sctp_ifa->address.sin6; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* * we skip unspecifed * addresses */ continue; } if (prison_check_ip6(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { continue; } if (stcb->asoc.scope.local_scope == 0 && IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) continue; if (stcb->asoc.scope.site_scope == 0 && IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr)) continue; if (sctp_is_addr_restricted(stcb, sctp_ifa) && (!sctp_is_addr_pending(stcb, sctp_ifa))) continue; /* * found a valid local v6 address to * use */ if (addr_locked == SCTP_ADDR_NOT_LOCKED) SCTP_IPI_ADDR_RUNLOCK(); return (&sctp_ifa->address.sa); } break; #endif default: break; } } } /* no valid addresses found */ if (addr_locked == SCTP_ADDR_NOT_LOCKED) SCTP_IPI_ADDR_RUNLOCK(); return (NULL); } static struct sockaddr * sctp_find_valid_localaddr_ep(struct sctp_tcb *stcb) { struct sctp_laddr *laddr; LIST_FOREACH(laddr, &stcb->sctp_ep->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { continue; } /* is the address restricted ? */ if (sctp_is_addr_restricted(stcb, laddr->ifa) && (!sctp_is_addr_pending(stcb, laddr->ifa))) continue; /* found a valid local address to use */ return (&laddr->ifa->address.sa); } /* no valid addresses found */ return (NULL); } /* * builds an ASCONF chunk from queued ASCONF params. * returns NULL on error (no mbuf, no ASCONF params queued, etc). */ struct mbuf * sctp_compose_asconf(struct sctp_tcb *stcb, int *retlen, int addr_locked) { struct mbuf *m_asconf, *m_asconf_chk; struct sctp_asconf_addr *aa; struct sctp_asconf_chunk *acp; struct sctp_asconf_paramhdr *aph; struct sctp_asconf_addr_param *aap; uint32_t p_length; uint32_t correlation_id = 1; /* 0 is reserved... */ caddr_t ptr, lookup_ptr; uint8_t lookup_used = 0; /* are there any asconf params to send? */ TAILQ_FOREACH(aa, &stcb->asoc.asconf_queue, next) { if (aa->sent == 0) break; } if (aa == NULL) return (NULL); /* * get a chunk header mbuf and a cluster for the asconf params since * it's simpler to fill in the asconf chunk header lookup address on * the fly */ m_asconf_chk = sctp_get_mbuf_for_msg(sizeof(struct sctp_asconf_chunk), 0, M_NOWAIT, 1, MT_DATA); if (m_asconf_chk == NULL) { /* no mbuf's */ SCTPDBG(SCTP_DEBUG_ASCONF1, "compose_asconf: couldn't get chunk mbuf!\n"); return (NULL); } m_asconf = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (m_asconf == NULL) { /* no mbuf's */ SCTPDBG(SCTP_DEBUG_ASCONF1, "compose_asconf: couldn't get mbuf!\n"); sctp_m_freem(m_asconf_chk); return (NULL); } SCTP_BUF_LEN(m_asconf_chk) = sizeof(struct sctp_asconf_chunk); SCTP_BUF_LEN(m_asconf) = 0; acp = mtod(m_asconf_chk, struct sctp_asconf_chunk *); memset(acp, 0, sizeof(struct sctp_asconf_chunk)); /* save pointers to lookup address and asconf params */ lookup_ptr = (caddr_t)(acp + 1); /* after the header */ ptr = mtod(m_asconf, caddr_t); /* beginning of cluster */ /* fill in chunk header info */ acp->ch.chunk_type = SCTP_ASCONF; acp->ch.chunk_flags = 0; acp->serial_number = htonl(stcb->asoc.asconf_seq_out); stcb->asoc.asconf_seq_out++; /* add parameters... up to smallest MTU allowed */ TAILQ_FOREACH(aa, &stcb->asoc.asconf_queue, next) { if (aa->sent) continue; /* get the parameter length */ p_length = SCTP_SIZE32(aa->ap.aph.ph.param_length); /* will it fit in current chunk? */ if ((SCTP_BUF_LEN(m_asconf) + p_length > stcb->asoc.smallest_mtu) || (SCTP_BUF_LEN(m_asconf) + p_length > MCLBYTES)) { /* won't fit, so we're done with this chunk */ break; } /* assign (and store) a correlation id */ aa->ap.aph.correlation_id = correlation_id++; /* * fill in address if we're doing a delete this is a simple * way for us to fill in the correlation address, which * should only be used by the peer if we're deleting our * source address and adding a new address (e.g. renumbering * case) */ if (lookup_used == 0 && (aa->special_del == 0) && aa->ap.aph.ph.param_type == SCTP_DEL_IP_ADDRESS) { struct sctp_ipv6addr_param *lookup; uint16_t p_size, addr_size; lookup = (struct sctp_ipv6addr_param *)lookup_ptr; lookup->ph.param_type = htons(aa->ap.addrp.ph.param_type); if (aa->ap.addrp.ph.param_type == SCTP_IPV6_ADDRESS) { /* copy IPv6 address */ p_size = sizeof(struct sctp_ipv6addr_param); addr_size = sizeof(struct in6_addr); } else { /* copy IPv4 address */ p_size = sizeof(struct sctp_ipv4addr_param); addr_size = sizeof(struct in_addr); } lookup->ph.param_length = htons(SCTP_SIZE32(p_size)); memcpy(lookup->addr, &aa->ap.addrp.addr, addr_size); SCTP_BUF_LEN(m_asconf_chk) += SCTP_SIZE32(p_size); lookup_used = 1; } /* copy into current space */ memcpy(ptr, &aa->ap, p_length); /* network elements and update lengths */ aph = (struct sctp_asconf_paramhdr *)ptr; aap = (struct sctp_asconf_addr_param *)ptr; /* correlation_id is transparent to peer, no htonl needed */ aph->ph.param_type = htons(aph->ph.param_type); aph->ph.param_length = htons(aph->ph.param_length); aap->addrp.ph.param_type = htons(aap->addrp.ph.param_type); aap->addrp.ph.param_length = htons(aap->addrp.ph.param_length); SCTP_BUF_LEN(m_asconf) += SCTP_SIZE32(p_length); ptr += SCTP_SIZE32(p_length); /* * these params are removed off the pending list upon * getting an ASCONF-ACK back from the peer, just set flag */ aa->sent = 1; } /* check to see if the lookup addr has been populated yet */ if (lookup_used == 0) { /* NOTE: if the address param is optional, can skip this... */ /* add any valid (existing) address... */ struct sctp_ipv6addr_param *lookup; uint16_t p_size, addr_size; struct sockaddr *found_addr; caddr_t addr_ptr; if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) found_addr = sctp_find_valid_localaddr(stcb, addr_locked); else found_addr = sctp_find_valid_localaddr_ep(stcb); lookup = (struct sctp_ipv6addr_param *)lookup_ptr; if (found_addr != NULL) { switch (found_addr->sa_family) { #ifdef INET6 case AF_INET6: /* copy IPv6 address */ lookup->ph.param_type = htons(SCTP_IPV6_ADDRESS); p_size = sizeof(struct sctp_ipv6addr_param); addr_size = sizeof(struct in6_addr); addr_ptr = (caddr_t)&((struct sockaddr_in6 *) found_addr)->sin6_addr; break; #endif #ifdef INET case AF_INET: /* copy IPv4 address */ lookup->ph.param_type = htons(SCTP_IPV4_ADDRESS); p_size = sizeof(struct sctp_ipv4addr_param); addr_size = sizeof(struct in_addr); addr_ptr = (caddr_t)&((struct sockaddr_in *) found_addr)->sin_addr; break; #endif default: p_size = 0; addr_size = 0; addr_ptr = NULL; break; } lookup->ph.param_length = htons(SCTP_SIZE32(p_size)); memcpy(lookup->addr, addr_ptr, addr_size); SCTP_BUF_LEN(m_asconf_chk) += SCTP_SIZE32(p_size); } else { /* uh oh... don't have any address?? */ SCTPDBG(SCTP_DEBUG_ASCONF1, "compose_asconf: no lookup addr!\n"); /* XXX for now, we send a IPv4 address of 0.0.0.0 */ lookup->ph.param_type = htons(SCTP_IPV4_ADDRESS); lookup->ph.param_length = htons(SCTP_SIZE32(sizeof(struct sctp_ipv4addr_param))); memset(lookup->addr, 0, sizeof(struct in_addr)); SCTP_BUF_LEN(m_asconf_chk) += SCTP_SIZE32(sizeof(struct sctp_ipv4addr_param)); } } /* chain it all together */ SCTP_BUF_NEXT(m_asconf_chk) = m_asconf; *retlen = SCTP_BUF_LEN(m_asconf_chk) + SCTP_BUF_LEN(m_asconf); acp->ch.chunk_length = htons(*retlen); return (m_asconf_chk); } /* * section to handle address changes before an association is up eg. changes * during INIT/INIT-ACK/COOKIE-ECHO handshake */ /* * processes the (local) addresses in the INIT-ACK chunk */ static void sctp_process_initack_addresses(struct sctp_tcb *stcb, struct mbuf *m, unsigned int offset, unsigned int length) { struct sctp_paramhdr tmp_param, *ph; uint16_t plen, ptype; struct sctp_ifa *sctp_ifa; union sctp_sockstore store; #ifdef INET6 struct sctp_ipv6addr_param addr6_store; #endif #ifdef INET struct sctp_ipv4addr_param addr4_store; #endif SCTPDBG(SCTP_DEBUG_ASCONF2, "processing init-ack addresses\n"); if (stcb == NULL) /* Un-needed check for SA */ return; /* convert to upper bound */ length += offset; if ((offset + sizeof(struct sctp_paramhdr)) > length) { return; } /* go through the addresses in the init-ack */ ph = (struct sctp_paramhdr *) sctp_m_getptr(m, offset, sizeof(struct sctp_paramhdr), (uint8_t *)&tmp_param); while (ph != NULL) { ptype = ntohs(ph->param_type); plen = ntohs(ph->param_length); switch (ptype) { #ifdef INET6 case SCTP_IPV6_ADDRESS: { struct sctp_ipv6addr_param *a6p; /* get the entire IPv6 address param */ a6p = (struct sctp_ipv6addr_param *) sctp_m_getptr(m, offset, sizeof(struct sctp_ipv6addr_param), (uint8_t *)&addr6_store); if (plen != sizeof(struct sctp_ipv6addr_param) || a6p == NULL) { return; } memset(&store, 0, sizeof(union sctp_sockstore)); store.sin6.sin6_family = AF_INET6; store.sin6.sin6_len = sizeof(struct sockaddr_in6); store.sin6.sin6_port = stcb->rport; memcpy(&store.sin6.sin6_addr, a6p->addr, sizeof(struct in6_addr)); break; } #endif #ifdef INET case SCTP_IPV4_ADDRESS: { struct sctp_ipv4addr_param *a4p; /* get the entire IPv4 address param */ a4p = (struct sctp_ipv4addr_param *)sctp_m_getptr(m, offset, sizeof(struct sctp_ipv4addr_param), (uint8_t *)&addr4_store); if (plen != sizeof(struct sctp_ipv4addr_param) || a4p == NULL) { return; } memset(&store, 0, sizeof(union sctp_sockstore)); store.sin.sin_family = AF_INET; store.sin.sin_len = sizeof(struct sockaddr_in); store.sin.sin_port = stcb->rport; store.sin.sin_addr.s_addr = a4p->addr; break; } #endif default: goto next_addr; } /* see if this address really (still) exists */ sctp_ifa = sctp_find_ifa_by_addr(&store.sa, stcb->asoc.vrf_id, SCTP_ADDR_NOT_LOCKED); if (sctp_ifa == NULL) { /* address doesn't exist anymore */ int status; /* are ASCONFs allowed ? */ if ((sctp_is_feature_on(stcb->sctp_ep, SCTP_PCB_FLAGS_DO_ASCONF)) && stcb->asoc.asconf_supported) { /* queue an ASCONF DEL_IP_ADDRESS */ status = sctp_asconf_queue_sa_delete(stcb, &store.sa); /* * if queued ok, and in correct state, send * out the ASCONF. */ if (status == 0 && SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) { #ifdef SCTP_TIMER_BASED_ASCONF sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, stcb->sctp_ep, stcb, stcb->asoc.primary_destination); #else sctp_send_asconf(stcb, NULL, SCTP_ADDR_NOT_LOCKED); #endif } } } next_addr: /* * Sanity check: Make sure the length isn't 0, otherwise * we'll be stuck in this loop for a long time... */ if (SCTP_SIZE32(plen) == 0) { SCTP_PRINTF("process_initack_addrs: bad len (%d) type=%xh\n", plen, ptype); return; } /* get next parameter */ offset += SCTP_SIZE32(plen); if ((offset + sizeof(struct sctp_paramhdr)) > length) return; ph = (struct sctp_paramhdr *)sctp_m_getptr(m, offset, sizeof(struct sctp_paramhdr), (uint8_t *)&tmp_param); } /* while */ } /* FIX ME: need to verify return result for v6 address type if v6 disabled */ /* * checks to see if a specific address is in the initack address list returns * 1 if found, 0 if not */ static uint32_t sctp_addr_in_initack(struct mbuf *m, uint32_t offset, uint32_t length, struct sockaddr *sa) { struct sctp_paramhdr tmp_param, *ph; uint16_t plen, ptype; #ifdef INET struct sockaddr_in *sin; struct sctp_ipv4addr_param *a4p; struct sctp_ipv6addr_param addr4_store; #endif #ifdef INET6 struct sockaddr_in6 *sin6; struct sctp_ipv6addr_param *a6p; struct sctp_ipv6addr_param addr6_store; struct sockaddr_in6 sin6_tmp; #endif switch (sa->sa_family) { #ifdef INET case AF_INET: break; #endif #ifdef INET6 case AF_INET6: break; #endif default: return (0); } SCTPDBG(SCTP_DEBUG_ASCONF2, "find_initack_addr: starting search for "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF2, sa); /* convert to upper bound */ length += offset; if ((offset + sizeof(struct sctp_paramhdr)) > length) { SCTPDBG(SCTP_DEBUG_ASCONF1, "find_initack_addr: invalid offset?\n"); return (0); } /* go through the addresses in the init-ack */ ph = (struct sctp_paramhdr *)sctp_m_getptr(m, offset, sizeof(struct sctp_paramhdr), (uint8_t *)&tmp_param); while (ph != NULL) { ptype = ntohs(ph->param_type); plen = ntohs(ph->param_length); switch (ptype) { #ifdef INET6 case SCTP_IPV6_ADDRESS: if (sa->sa_family == AF_INET6) { /* get the entire IPv6 address param */ if (plen != sizeof(struct sctp_ipv6addr_param)) { break; } /* get the entire IPv6 address param */ a6p = (struct sctp_ipv6addr_param *) sctp_m_getptr(m, offset, sizeof(struct sctp_ipv6addr_param), (uint8_t *)&addr6_store); if (a6p == NULL) { return (0); } sin6 = (struct sockaddr_in6 *)sa; if (IN6_IS_SCOPE_LINKLOCAL(&sin6->sin6_addr)) { /* create a copy and clear scope */ memcpy(&sin6_tmp, sin6, sizeof(struct sockaddr_in6)); sin6 = &sin6_tmp; in6_clearscope(&sin6->sin6_addr); } if (memcmp(&sin6->sin6_addr, a6p->addr, sizeof(struct in6_addr)) == 0) { /* found it */ return (1); } } break; #endif /* INET6 */ #ifdef INET case SCTP_IPV4_ADDRESS: if (sa->sa_family == AF_INET) { if (plen != sizeof(struct sctp_ipv4addr_param)) { break; } /* get the entire IPv4 address param */ a4p = (struct sctp_ipv4addr_param *) sctp_m_getptr(m, offset, sizeof(struct sctp_ipv4addr_param), (uint8_t *)&addr4_store); if (a4p == NULL) { return (0); } sin = (struct sockaddr_in *)sa; if (sin->sin_addr.s_addr == a4p->addr) { /* found it */ return (1); } } break; #endif default: break; } /* get next parameter */ offset += SCTP_SIZE32(plen); if (offset + sizeof(struct sctp_paramhdr) > length) { return (0); } ph = (struct sctp_paramhdr *) sctp_m_getptr(m, offset, sizeof(struct sctp_paramhdr), (uint8_t *)&tmp_param); } /* while */ /* not found! */ return (0); } /* * makes sure that the current endpoint local addr list is consistent with * the new association (eg. subset bound, asconf allowed) adds addresses as * necessary */ static void sctp_check_address_list_ep(struct sctp_tcb *stcb, struct mbuf *m, int offset, int length, struct sockaddr *init_addr) { struct sctp_laddr *laddr; /* go through the endpoint list */ LIST_FOREACH(laddr, &stcb->sctp_ep->sctp_addr_list, sctp_nxt_addr) { /* be paranoid and validate the laddr */ if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "check_addr_list_ep: laddr->ifa is NULL"); continue; } if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "check_addr_list_ep: laddr->ifa->ifa_addr is NULL"); continue; } /* do i have it implicitly? */ if (sctp_cmpaddr(&laddr->ifa->address.sa, init_addr)) { continue; } /* check to see if in the init-ack */ if (!sctp_addr_in_initack(m, offset, length, &laddr->ifa->address.sa)) { /* try to add it */ sctp_addr_mgmt_assoc(stcb->sctp_ep, stcb, laddr->ifa, SCTP_ADD_IP_ADDRESS, SCTP_ADDR_NOT_LOCKED); } } } /* * makes sure that the current kernel address list is consistent with the new * association (with all addrs bound) adds addresses as necessary */ static void sctp_check_address_list_all(struct sctp_tcb *stcb, struct mbuf *m, int offset, int length, struct sockaddr *init_addr, uint16_t local_scope, uint16_t site_scope, uint16_t ipv4_scope, uint16_t loopback_scope) { struct sctp_vrf *vrf = NULL; struct sctp_ifn *sctp_ifn; struct sctp_ifa *sctp_ifa; uint32_t vrf_id; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif if (stcb) { vrf_id = stcb->asoc.vrf_id; } else { return; } SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { SCTP_IPI_ADDR_RUNLOCK(); return; } /* go through all our known interfaces */ LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { if (loopback_scope == 0 && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { /* skip loopback interface */ continue; } /* go through each interface address */ LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { /* do i have it implicitly? */ if (sctp_cmpaddr(&sctp_ifa->address.sa, init_addr)) { continue; } switch (sctp_ifa->address.sa.sa_family) { #ifdef INET case AF_INET: sin = &sctp_ifa->address.sin; if (prison_check_ip4(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { continue; } if ((ipv4_scope == 0) && (IN4_ISPRIVATE_ADDRESS(&sin->sin_addr))) { /* private address not in scope */ continue; } break; #endif #ifdef INET6 case AF_INET6: sin6 = &sctp_ifa->address.sin6; if (prison_check_ip6(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { continue; } if ((local_scope == 0) && (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))) { continue; } if ((site_scope == 0) && (IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr))) { continue; } break; #endif default: break; } /* check to see if in the init-ack */ if (!sctp_addr_in_initack(m, offset, length, &sctp_ifa->address.sa)) { /* try to add it */ sctp_addr_mgmt_assoc(stcb->sctp_ep, stcb, sctp_ifa, SCTP_ADD_IP_ADDRESS, SCTP_ADDR_LOCKED); } } /* end foreach ifa */ } /* end foreach ifn */ SCTP_IPI_ADDR_RUNLOCK(); } /* * validates an init-ack chunk (from a cookie-echo) with current addresses * adds addresses from the init-ack into our local address list, if needed * queues asconf adds/deletes addresses as needed and makes appropriate list * changes for source address selection m, offset: points to the start of the * address list in an init-ack chunk length: total length of the address * params only init_addr: address where my INIT-ACK was sent from */ void sctp_check_address_list(struct sctp_tcb *stcb, struct mbuf *m, int offset, int length, struct sockaddr *init_addr, uint16_t local_scope, uint16_t site_scope, uint16_t ipv4_scope, uint16_t loopback_scope) { /* process the local addresses in the initack */ sctp_process_initack_addresses(stcb, m, offset, length); if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { /* bound all case */ sctp_check_address_list_all(stcb, m, offset, length, init_addr, local_scope, site_scope, ipv4_scope, loopback_scope); } else { /* subset bound case */ if (sctp_is_feature_on(stcb->sctp_ep, SCTP_PCB_FLAGS_DO_ASCONF)) { /* asconf's allowed */ sctp_check_address_list_ep(stcb, m, offset, length, init_addr); } /* else, no asconfs allowed, so what we sent is what we get */ } } /* * sctp_bindx() support */ uint32_t sctp_addr_mgmt_ep_sa(struct sctp_inpcb *inp, struct sockaddr *sa, uint32_t type, uint32_t vrf_id, struct sctp_ifa *sctp_ifap) { struct sctp_ifa *ifa; struct sctp_laddr *laddr, *nladdr; if (sa->sa_len == 0) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_ASCONF, EINVAL); return (EINVAL); } if (sctp_ifap) { ifa = sctp_ifap; } else if (type == SCTP_ADD_IP_ADDRESS) { /* For an add the address MUST be on the system */ ifa = sctp_find_ifa_by_addr(sa, vrf_id, SCTP_ADDR_NOT_LOCKED); } else if (type == SCTP_DEL_IP_ADDRESS) { /* For a delete we need to find it in the inp */ ifa = sctp_find_ifa_in_ep(inp, sa, SCTP_ADDR_NOT_LOCKED); } else { ifa = NULL; } if (ifa != NULL) { if (type == SCTP_ADD_IP_ADDRESS) { sctp_add_local_addr_ep(inp, ifa, type); } else if (type == SCTP_DEL_IP_ADDRESS) { if (inp->laddr_count < 2) { /* can't delete the last local address */ SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_ASCONF, EINVAL); return (EINVAL); } LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (ifa == laddr->ifa) { /* Mark in the delete */ laddr->action = type; } } } if (LIST_EMPTY(&inp->sctp_asoc_list)) { /* * There is no need to start the iterator if the inp * has no associations. */ if (type == SCTP_DEL_IP_ADDRESS) { LIST_FOREACH_SAFE(laddr, &inp->sctp_addr_list, sctp_nxt_addr, nladdr) { if (laddr->ifa == ifa) { sctp_del_local_addr_ep(inp, ifa); } } } } else { struct sctp_asconf_iterator *asc; struct sctp_laddr *wi; int ret; SCTP_MALLOC(asc, struct sctp_asconf_iterator *, sizeof(struct sctp_asconf_iterator), SCTP_M_ASC_IT); if (asc == NULL) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_ASCONF, ENOMEM); return (ENOMEM); } wi = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_laddr), struct sctp_laddr); if (wi == NULL) { SCTP_FREE(asc, SCTP_M_ASC_IT); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_ASCONF, ENOMEM); return (ENOMEM); } LIST_INIT(&asc->list_of_work); asc->cnt = 1; SCTP_INCR_LADDR_COUNT(); wi->ifa = ifa; wi->action = type; atomic_add_int(&ifa->refcount, 1); LIST_INSERT_HEAD(&asc->list_of_work, wi, sctp_nxt_addr); ret = sctp_initiate_iterator(sctp_asconf_iterator_ep, sctp_asconf_iterator_stcb, sctp_asconf_iterator_ep_end, SCTP_PCB_ANY_FLAGS, SCTP_PCB_ANY_FEATURES, SCTP_ASOC_ANY_STATE, (void *)asc, 0, sctp_asconf_iterator_end, inp, 0); if (ret) { SCTP_PRINTF("Failed to initiate iterator for addr_mgmt_ep_sa\n"); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_ASCONF, EFAULT); sctp_asconf_iterator_end(asc, 0); return (EFAULT); } } return (0); } else { /* invalid address! */ SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_ASCONF, EADDRNOTAVAIL); return (EADDRNOTAVAIL); } } void sctp_asconf_send_nat_state_update(struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_asconf_addr *aa; struct sctp_ifa *sctp_ifap; struct sctp_asconf_tag_param *vtag; #ifdef INET struct sockaddr_in *to; #endif #ifdef INET6 struct sockaddr_in6 *to6; #endif if (net == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: Missing net\n"); return; } if (stcb == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: Missing stcb\n"); return; } /* * Need to have in the asconf: - vtagparam(my_vtag/peer_vtag) - * add(0.0.0.0) - del(0.0.0.0) - Any global addresses add(addr) */ SCTP_MALLOC(aa, struct sctp_asconf_addr *, sizeof(*aa), SCTP_M_ASC_ADDR); if (aa == NULL) { /* didn't get memory */ SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: failed to get memory!\n"); return; } aa->special_del = 0; /* fill in asconf address parameter fields */ /* top level elements are "networked" during send */ aa->ifa = NULL; aa->sent = 0; /* clear sent flag */ vtag = (struct sctp_asconf_tag_param *)&aa->ap.aph; vtag->aph.ph.param_type = SCTP_NAT_VTAGS; vtag->aph.ph.param_length = sizeof(struct sctp_asconf_tag_param); vtag->local_vtag = htonl(stcb->asoc.my_vtag); vtag->remote_vtag = htonl(stcb->asoc.peer_vtag); TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); SCTP_MALLOC(aa, struct sctp_asconf_addr *, sizeof(*aa), SCTP_M_ASC_ADDR); if (aa == NULL) { /* didn't get memory */ SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: failed to get memory!\n"); return; } memset(aa, 0, sizeof(struct sctp_asconf_addr)); /* fill in asconf address parameter fields */ /* ADD(0.0.0.0) */ switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: aa->ap.aph.ph.param_type = SCTP_ADD_IP_ADDRESS; aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_addrv4_param); aa->ap.addrp.ph.param_type = SCTP_IPV4_ADDRESS; aa->ap.addrp.ph.param_length = sizeof(struct sctp_ipv4addr_param); /* No need to add an address, we are using 0.0.0.0 */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); break; #endif #ifdef INET6 case AF_INET6: aa->ap.aph.ph.param_type = SCTP_ADD_IP_ADDRESS; aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_addr_param); aa->ap.addrp.ph.param_type = SCTP_IPV6_ADDRESS; aa->ap.addrp.ph.param_length = sizeof(struct sctp_ipv6addr_param); /* No need to add an address, we are using 0.0.0.0 */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); break; #endif default: SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: unknown address family\n"); SCTP_FREE(aa, SCTP_M_ASC_ADDR); return; } SCTP_MALLOC(aa, struct sctp_asconf_addr *, sizeof(*aa), SCTP_M_ASC_ADDR); if (aa == NULL) { /* didn't get memory */ SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: failed to get memory!\n"); return; } memset(aa, 0, sizeof(struct sctp_asconf_addr)); /* fill in asconf address parameter fields */ /* ADD(0.0.0.0) */ switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: aa->ap.aph.ph.param_type = SCTP_ADD_IP_ADDRESS; aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_addrv4_param); aa->ap.addrp.ph.param_type = SCTP_IPV4_ADDRESS; aa->ap.addrp.ph.param_length = sizeof(struct sctp_ipv4addr_param); /* No need to add an address, we are using 0.0.0.0 */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); break; #endif #ifdef INET6 case AF_INET6: aa->ap.aph.ph.param_type = SCTP_DEL_IP_ADDRESS; aa->ap.aph.ph.param_length = sizeof(struct sctp_asconf_addr_param); aa->ap.addrp.ph.param_type = SCTP_IPV6_ADDRESS; aa->ap.addrp.ph.param_length = sizeof(struct sctp_ipv6addr_param); /* No need to add an address, we are using 0.0.0.0 */ TAILQ_INSERT_TAIL(&stcb->asoc.asconf_queue, aa, next); break; #endif default: SCTPDBG(SCTP_DEBUG_ASCONF1, "sctp_asconf_send_nat_state_update: unknown address family\n"); SCTP_FREE(aa, SCTP_M_ASC_ADDR); return; } /* Now we must hunt the addresses and add all global addresses */ if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { struct sctp_vrf *vrf = NULL; struct sctp_ifn *sctp_ifnp; uint32_t vrf_id; vrf_id = stcb->sctp_ep->def_vrf_id; vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { goto skip_rest; } SCTP_IPI_ADDR_RLOCK(); LIST_FOREACH(sctp_ifnp, &vrf->ifnlist, next_ifn) { LIST_FOREACH(sctp_ifap, &sctp_ifnp->ifalist, next_ifa) { switch (sctp_ifap->address.sa.sa_family) { #ifdef INET case AF_INET: to = &sctp_ifap->address.sin; if (prison_check_ip4(stcb->sctp_ep->ip_inp.inp.inp_cred, &to->sin_addr) != 0) { continue; } if (IN4_ISPRIVATE_ADDRESS(&to->sin_addr)) { continue; } if (IN4_ISLOOPBACK_ADDRESS(&to->sin_addr)) { continue; } break; #endif #ifdef INET6 case AF_INET6: to6 = &sctp_ifap->address.sin6; if (prison_check_ip6(stcb->sctp_ep->ip_inp.inp.inp_cred, &to6->sin6_addr) != 0) { continue; } if (IN6_IS_ADDR_LOOPBACK(&to6->sin6_addr)) { continue; } if (IN6_IS_ADDR_LINKLOCAL(&to6->sin6_addr)) { continue; } break; #endif default: continue; } sctp_asconf_queue_mgmt(stcb, sctp_ifap, SCTP_ADD_IP_ADDRESS); } } SCTP_IPI_ADDR_RUNLOCK(); } else { struct sctp_laddr *laddr; LIST_FOREACH(laddr, &stcb->sctp_ep->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { continue; } if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) /* * Address being deleted by the system, dont * list. */ continue; if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* * Address being deleted on this ep don't * list. */ continue; } sctp_ifap = laddr->ifa; switch (sctp_ifap->address.sa.sa_family) { #ifdef INET case AF_INET: to = &sctp_ifap->address.sin; if (IN4_ISPRIVATE_ADDRESS(&to->sin_addr)) { continue; } if (IN4_ISLOOPBACK_ADDRESS(&to->sin_addr)) { continue; } break; #endif #ifdef INET6 case AF_INET6: to6 = &sctp_ifap->address.sin6; if (IN6_IS_ADDR_LOOPBACK(&to6->sin6_addr)) { continue; } if (IN6_IS_ADDR_LINKLOCAL(&to6->sin6_addr)) { continue; } break; #endif default: continue; } sctp_asconf_queue_mgmt(stcb, sctp_ifap, SCTP_ADD_IP_ADDRESS); } } skip_rest: /* Now we must send the asconf into the queue */ sctp_send_asconf(stcb, net, SCTP_ADDR_NOT_LOCKED); } Index: head/sys/netinet/sctp_os_bsd.h =================================================================== --- head/sys/netinet/sctp_os_bsd.h (revision 360291) +++ head/sys/netinet/sctp_os_bsd.h (revision 360292) @@ -1,481 +1,488 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2006-2007, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #ifndef _NETINET_SCTP_OS_BSD_H_ #define _NETINET_SCTP_OS_BSD_H_ /* * includes */ #include "opt_inet6.h" #include "opt_inet.h" #include "opt_sctp.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include +#include #include #include #include #include #include #ifdef INET6 #include #include +#include #include #include #include #include #include #endif /* INET6 */ #include #include #include /* Declare all the malloc names for all the various mallocs */ MALLOC_DECLARE(SCTP_M_MAP); MALLOC_DECLARE(SCTP_M_STRMI); MALLOC_DECLARE(SCTP_M_STRMO); MALLOC_DECLARE(SCTP_M_ASC_ADDR); MALLOC_DECLARE(SCTP_M_ASC_IT); MALLOC_DECLARE(SCTP_M_AUTH_CL); MALLOC_DECLARE(SCTP_M_AUTH_KY); MALLOC_DECLARE(SCTP_M_AUTH_HL); MALLOC_DECLARE(SCTP_M_AUTH_IF); MALLOC_DECLARE(SCTP_M_STRESET); MALLOC_DECLARE(SCTP_M_CMSG); MALLOC_DECLARE(SCTP_M_COPYAL); MALLOC_DECLARE(SCTP_M_VRF); MALLOC_DECLARE(SCTP_M_IFA); MALLOC_DECLARE(SCTP_M_IFN); MALLOC_DECLARE(SCTP_M_TIMW); MALLOC_DECLARE(SCTP_M_MVRF); MALLOC_DECLARE(SCTP_M_ITER); MALLOC_DECLARE(SCTP_M_SOCKOPT); MALLOC_DECLARE(SCTP_M_MCORE); #if defined(SCTP_LOCAL_TRACE_BUF) #define SCTP_GET_CYCLECOUNT get_cyclecount() #define SCTP_CTR6 sctp_log_trace #else #define SCTP_CTR6 CTR6 #endif /* * Macros to expand out globals defined by various modules * to either a real global or a virtualized instance of one, * depending on whether VIMAGE is defined. */ /* then define the macro(s) that hook into the vimage macros */ #define MODULE_GLOBAL(__SYMBOL) V_##__SYMBOL #define V_system_base_info VNET(system_base_info) #define SCTP_BASE_INFO(__m) V_system_base_info.sctppcbinfo.__m #define SCTP_BASE_STATS V_system_base_info.sctpstat #define SCTP_BASE_STAT(__m) V_system_base_info.sctpstat.__m #define SCTP_BASE_SYSCTL(__m) V_system_base_info.sctpsysctl.__m #define SCTP_BASE_VAR(__m) V_system_base_info.__m #define SCTP_PRINTF(params...) printf(params) #if defined(SCTP_DEBUG) #define SCTPDBG(level, params...) \ { \ do { \ if (SCTP_BASE_SYSCTL(sctp_debug_on) & level ) { \ SCTP_PRINTF(params); \ } \ } while (0); \ } #define SCTPDBG_ADDR(level, addr) \ { \ do { \ if (SCTP_BASE_SYSCTL(sctp_debug_on) & level ) { \ sctp_print_address(addr); \ } \ } while (0); \ } #else #define SCTPDBG(level, params...) #define SCTPDBG_ADDR(level, addr) #endif #ifdef SCTP_LTRACE_CHUNKS #define SCTP_LTRACE_CHK(a, b, c, d) if(SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LTRACE_CHUNK_ENABLE) SCTP_CTR6(KTR_SUBSYS, "SCTP:%d[%d]:%x-%x-%x-%x", SCTP_LOG_CHUNK_PROC, 0, a, b, c, d) #else #define SCTP_LTRACE_CHK(a, b, c, d) #endif #ifdef SCTP_LTRACE_ERRORS #define SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, file, err) \ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LTRACE_ERROR_ENABLE) \ SCTP_PRINTF("mbuf:%p inp:%p stcb:%p net:%p file:%x line:%d error:%d\n", \ m, inp, stcb, net, file, __LINE__, err); #define SCTP_LTRACE_ERR_RET(inp, stcb, net, file, err) \ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LTRACE_ERROR_ENABLE) \ SCTP_PRINTF("inp:%p stcb:%p net:%p file:%x line:%d error:%d\n", \ inp, stcb, net, file, __LINE__, err); #else #define SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, file, err) #define SCTP_LTRACE_ERR_RET(inp, stcb, net, file, err) #endif /* * Local address and interface list handling */ #define SCTP_MAX_VRF_ID 0 #define SCTP_SIZE_OF_VRF_HASH 3 #define SCTP_IFNAMSIZ IFNAMSIZ #define SCTP_DEFAULT_VRFID 0 #define SCTP_VRF_ADDR_HASH_SIZE 16 #define SCTP_VRF_IFN_HASH_SIZE 3 #define SCTP_INIT_VRF_TABLEID(vrf) #define SCTP_IFN_IS_IFT_LOOP(ifn) ((ifn)->ifn_type == IFT_LOOP) -#define SCTP_ROUTE_IS_REAL_LOOP(ro) ((ro)->ro_rt && (ro)->ro_rt->rt_ifa && (ro)->ro_rt->rt_ifa->ifa_ifp && (ro)->ro_rt->rt_ifa->ifa_ifp->if_type == IFT_LOOP) +#define SCTP_ROUTE_IS_REAL_LOOP(ro) ((ro)->ro_nh && (ro)->ro_nh->nh_ifa && (ro)->ro_nh->nh_ifa->ifa_ifp && (ro)->ro_nh->nh_ifa->ifa_ifp->if_type == IFT_LOOP) /* * Access to IFN's to help with src-addr-selection */ /* This could return VOID if the index works but for BSD we provide both. */ -#define SCTP_GET_IFN_VOID_FROM_ROUTE(ro) (void *)ro->ro_rt->rt_ifp -#define SCTP_GET_IF_INDEX_FROM_ROUTE(ro) (ro)->ro_rt->rt_ifp->if_index -#define SCTP_ROUTE_HAS_VALID_IFN(ro) ((ro)->ro_rt && (ro)->ro_rt->rt_ifp) +#define SCTP_GET_IFN_VOID_FROM_ROUTE(ro) (void *)ro->ro_nh->nh_ifp +#define SCTP_GET_IF_INDEX_FROM_ROUTE(ro) (ro)->ro_nh->nh_ifp->if_index +#define SCTP_ROUTE_HAS_VALID_IFN(ro) ((ro)->ro_nh && (ro)->ro_nh->nh_ifp) /* * general memory allocation */ #define SCTP_MALLOC(var, type, size, name) \ do { \ var = (type)malloc(size, name, M_NOWAIT); \ } while (0) #define SCTP_FREE(var, type) free(var, type) #define SCTP_MALLOC_SONAME(var, type, size) \ do { \ var = (type)malloc(size, M_SONAME, M_WAITOK | M_ZERO); \ } while (0) #define SCTP_FREE_SONAME(var) free(var, M_SONAME) #define SCTP_PROCESS_STRUCT struct proc * /* * zone allocation functions */ #include /* SCTP_ZONE_INIT: initialize the zone */ typedef struct uma_zone *sctp_zone_t; #define SCTP_ZONE_INIT(zone, name, size, number) { \ zone = uma_zcreate(name, size, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,\ 0); \ uma_zone_set_max(zone, number); \ } #define SCTP_ZONE_DESTROY(zone) uma_zdestroy(zone) /* SCTP_ZONE_GET: allocate element from the zone */ #define SCTP_ZONE_GET(zone, type) \ (type *)uma_zalloc(zone, M_NOWAIT); /* SCTP_ZONE_FREE: free element from the zone */ #define SCTP_ZONE_FREE(zone, element) \ uma_zfree(zone, element); #define SCTP_HASH_INIT(size, hashmark) hashinit_flags(size, M_PCB, hashmark, HASH_NOWAIT) #define SCTP_HASH_FREE(table, hashmark) hashdestroy(table, M_PCB, hashmark) #define SCTP_M_COPYM m_copym /* * timers */ #include typedef struct callout sctp_os_timer_t; #define SCTP_OS_TIMER_INIT(tmr) callout_init(tmr, 1) #define SCTP_OS_TIMER_START callout_reset #define SCTP_OS_TIMER_STOP callout_stop #define SCTP_OS_TIMER_STOP_DRAIN callout_drain #define SCTP_OS_TIMER_PENDING callout_pending #define SCTP_OS_TIMER_ACTIVE callout_active #define SCTP_OS_TIMER_DEACTIVATE callout_deactivate #define sctp_get_tick_count() (ticks) #define SCTP_UNUSED __attribute__((unused)) /* * Functions */ /* Mbuf manipulation and access macros */ #define SCTP_BUF_LEN(m) (m->m_len) #define SCTP_BUF_NEXT(m) (m->m_next) #define SCTP_BUF_NEXT_PKT(m) (m->m_nextpkt) #define SCTP_BUF_RESV_UF(m, size) m->m_data += size #define SCTP_BUF_AT(m, size) m->m_data + size #define SCTP_BUF_IS_EXTENDED(m) (m->m_flags & M_EXT) #define SCTP_BUF_SIZE M_SIZE #define SCTP_BUF_TYPE(m) (m->m_type) #define SCTP_BUF_RECVIF(m) (m->m_pkthdr.rcvif) #define SCTP_BUF_PREPEND M_PREPEND #define SCTP_ALIGN_TO_END(m, len) M_ALIGN(m, len) /* We make it so if you have up to 4 threads * writing based on the default size of * the packet log 65 k, that would be * 4 16k packets before we would hit * a problem. */ #define SCTP_PKTLOG_WRITERS_NEED_LOCK 3 /*************************/ /* MTU */ /*************************/ #define SCTP_GATHER_MTU_FROM_IFN_INFO(ifn, ifn_index, af) ((struct ifnet *)ifn)->if_mtu -#define SCTP_GATHER_MTU_FROM_ROUTE(sctp_ifa, sa, rt) ((uint32_t)((rt != NULL) ? rt->rt_mtu : 0)) +#define SCTP_GATHER_MTU_FROM_ROUTE(sctp_ifa, sa, nh) ((uint32_t)((nh != NULL) ? nh->nh_mtu : 0)) #define SCTP_GATHER_MTU_FROM_INTFC(sctp_ifn) ((sctp_ifn->ifn_p != NULL) ? ((struct ifnet *)(sctp_ifn->ifn_p))->if_mtu : 0) -#define SCTP_SET_MTU_OF_ROUTE(sa, rt, mtu) do { \ - if (rt != NULL) \ - rt->rt_mtu = mtu; \ - } while(0) +/* XXX: Setting MTU from the protocol in this way is simply incorrect */ +#define SCTP_SET_MTU_OF_ROUTE(sa, rt, mtu) /* (de-)register interface event notifications */ #define SCTP_REGISTER_INTERFACE(ifhandle, af) #define SCTP_DEREGISTER_INTERFACE(ifhandle, af) /*************************/ /* These are for logging */ /*************************/ /* return the base ext data pointer */ #define SCTP_BUF_EXTEND_BASE(m) (m->m_ext.ext_buf) /* return the refcnt of the data pointer */ #define SCTP_BUF_EXTEND_REFCNT(m) (*m->m_ext.ext_cnt) /* return any buffer related flags, this is * used beyond logging for apple only. */ #define SCTP_BUF_GET_FLAGS(m) (m->m_flags) /* For BSD this just accesses the M_PKTHDR length * so it operates on an mbuf with hdr flag. Other * O/S's may have separate packet header and mbuf * chain pointers.. thus the macro. */ #define SCTP_HEADER_TO_CHAIN(m) (m) #define SCTP_DETACH_HEADER_FROM_CHAIN(m) #define SCTP_HEADER_LEN(m) ((m)->m_pkthdr.len) #define SCTP_GET_HEADER_FOR_OUTPUT(o_pak) 0 #define SCTP_RELEASE_HEADER(m) #define SCTP_RELEASE_PKT(m) sctp_m_freem(m) #define SCTP_ENABLE_UDP_CSUM(m) do { \ m->m_pkthdr.csum_flags = CSUM_UDP; \ m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); \ } while (0) #define SCTP_GET_PKT_VRFID(m, vrf_id) ((vrf_id = SCTP_DEFAULT_VRFID) != SCTP_DEFAULT_VRFID) /* Attach the chain of data into the sendable packet. */ #define SCTP_ATTACH_CHAIN(pak, m, packet_length) do { \ pak = m; \ pak->m_pkthdr.len = packet_length; \ } while(0) /* Other m_pkthdr type things */ #define SCTP_IS_IT_BROADCAST(dst, m) ((m->m_flags & M_PKTHDR) ? in_broadcast(dst, m->m_pkthdr.rcvif) : 0) #define SCTP_IS_IT_LOOPBACK(m) ((m->m_flags & M_PKTHDR) && ((m->m_pkthdr.rcvif == NULL) || (m->m_pkthdr.rcvif->if_type == IFT_LOOP))) /* This converts any input packet header * into the chain of data holders, for BSD * its a NOP. */ /* get the v6 hop limit */ -#define SCTP_GET_HLIM(inp, ro) in6_selecthlim(&inp->ip_inp.inp, (ro ? (ro->ro_rt ? (ro->ro_rt->rt_ifp) : (NULL)) : (NULL))); +#define SCTP_GET_HLIM(inp, ro) in6_selecthlim(&inp->ip_inp.inp, (ro ? (ro->ro_nh ? (ro->ro_nh->nh_ifp) : (NULL)) : (NULL))); /* is the endpoint v6only? */ #define SCTP_IPV6_V6ONLY(sctp_inpcb) ((sctp_inpcb)->ip_inp.inp.inp_flags & IN6P_IPV6_V6ONLY) /* is the socket non-blocking? */ #define SCTP_SO_IS_NBIO(so) ((so)->so_state & SS_NBIO) #define SCTP_SET_SO_NBIO(so) ((so)->so_state |= SS_NBIO) #define SCTP_CLEAR_SO_NBIO(so) ((so)->so_state &= ~SS_NBIO) /* get the socket type */ #define SCTP_SO_TYPE(so) ((so)->so_type) /* Use a macro for renaming sb_cc to sb_acc. * Initially sb_ccc was used, but this broke select() when used * with SCTP sockets. */ #define sb_cc sb_acc /* reserve sb space for a socket */ #define SCTP_SORESERVE(so, send, recv) soreserve(so, send, recv) /* wakeup a socket */ #define SCTP_SOWAKEUP(so) wakeup(&(so)->so_timeo) /* clear the socket buffer state */ #define SCTP_SB_CLEAR(sb) \ (sb).sb_cc = 0; \ (sb).sb_mb = NULL; \ (sb).sb_mbcnt = 0; #define SCTP_SB_LIMIT_RCV(so) (SOLISTENING(so) ? so->sol_sbrcv_hiwat : so->so_rcv.sb_hiwat) #define SCTP_SB_LIMIT_SND(so) (SOLISTENING(so) ? so->sol_sbsnd_hiwat : so->so_snd.sb_hiwat) /* * routes, output, etc. */ typedef struct route sctp_route_t; -typedef struct rtentry sctp_rtentry_t; #define SCTP_RTALLOC(ro, vrf_id, fibnum) \ - rtalloc_ign_fib((struct route *)ro, 0UL, fibnum) +{ \ + if ((ro)->ro_nh == NULL) { \ + if ((ro)->ro_dst.sa_family == AF_INET) \ + (ro)->ro_nh = fib4_lookup(fibnum, ((struct sockaddr_in *)&(ro)->ro_dst)->sin_addr, NHR_REF, 0, 0); \ + if ((ro)->ro_dst.sa_family == AF_INET6) \ + (ro)->ro_nh = fib6_lookup(fibnum, &((struct sockaddr_in6 *)&(ro)->ro_dst)->sin6_addr, NHR_REF, 0, 0); \ + } \ +} /* * SCTP protocol specific mbuf flags. */ #define M_NOTIFICATION M_PROTO1 /* SCTP notification */ /* * IP output routines */ #define SCTP_IP_OUTPUT(result, o_pak, ro, stcb, vrf_id) \ { \ int o_flgs = IP_RAWOUTPUT; \ struct sctp_tcb *local_stcb = stcb; \ if (local_stcb && \ local_stcb->sctp_ep && \ local_stcb->sctp_ep->sctp_socket) \ o_flgs |= local_stcb->sctp_ep->sctp_socket->so_options & SO_DONTROUTE; \ m_clrprotoflags(o_pak); \ result = ip_output(o_pak, NULL, ro, o_flgs, 0, NULL); \ } #define SCTP_IP6_OUTPUT(result, o_pak, ro, ifp, stcb, vrf_id) \ { \ struct sctp_tcb *local_stcb = stcb; \ m_clrprotoflags(o_pak); \ if (local_stcb && local_stcb->sctp_ep) \ result = ip6_output(o_pak, \ ((struct inpcb *)(local_stcb->sctp_ep))->in6p_outputopts, \ (ro), 0, 0, ifp, NULL); \ else \ result = ip6_output(o_pak, NULL, (ro), 0, 0, ifp, NULL); \ } struct mbuf * sctp_get_mbuf_for_msg(unsigned int space_needed, int want_header, int how, int allonebuf, int type); /* * SCTP AUTH */ #define SCTP_READ_RANDOM(buf, len) arc4rand(buf, len, 0) /* map standard crypto API names */ #define SCTP_SHA1_CTX SHA1_CTX #define SCTP_SHA1_INIT SHA1Init #define SCTP_SHA1_UPDATE SHA1Update #define SCTP_SHA1_FINAL(x,y) SHA1Final((caddr_t)x, y) #define SCTP_SHA256_CTX SHA256_CTX #define SCTP_SHA256_INIT SHA256_Init #define SCTP_SHA256_UPDATE SHA256_Update #define SCTP_SHA256_FINAL(x,y) SHA256_Final((caddr_t)x, y) #define SCTP_DECREMENT_AND_CHECK_REFCOUNT(addr) (atomic_fetchadd_int(addr, -1) == 1) #if defined(INVARIANTS) #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \ { \ int32_t oldval; \ oldval = atomic_fetchadd_int(addr, -val); \ if (oldval < val) { \ panic("Counter goes negative"); \ } \ } #else #define SCTP_SAVE_ATOMIC_DECREMENT(addr, val) \ { \ int32_t oldval; \ oldval = atomic_fetchadd_int(addr, -val); \ if (oldval < val) { \ *addr = 0; \ } \ } #endif #define SCTP_IS_LISTENING(inp) ((inp->sctp_flags & SCTP_PCB_FLAGS_ACCEPTING) != 0) #endif Index: head/sys/netinet/sctp_output.c =================================================================== --- head/sys/netinet/sctp_output.c (revision 360291) +++ head/sys/netinet/sctp_output.c (revision 360292) @@ -1,13917 +1,13911 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2008, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #endif #include #include #define SCTP_MAX_GAPS_INARRAY 4 struct sack_track { uint8_t right_edge; /* mergable on the right edge */ uint8_t left_edge; /* mergable on the left edge */ uint8_t num_entries; uint8_t spare; struct sctp_gap_ack_block gaps[SCTP_MAX_GAPS_INARRAY]; }; const struct sack_track sack_array[256] = { {0, 0, 0, 0, /* 0x00 */ {{0, 0}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x01 */ {{0, 0}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x02 */ {{1, 1}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x03 */ {{0, 1}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x04 */ {{2, 2}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x05 */ {{0, 0}, {2, 2}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x06 */ {{1, 2}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x07 */ {{0, 2}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x08 */ {{3, 3}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x09 */ {{0, 0}, {3, 3}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x0a */ {{1, 1}, {3, 3}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x0b */ {{0, 1}, {3, 3}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x0c */ {{2, 3}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x0d */ {{0, 0}, {2, 3}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x0e */ {{1, 3}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x0f */ {{0, 3}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x10 */ {{4, 4}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x11 */ {{0, 0}, {4, 4}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x12 */ {{1, 1}, {4, 4}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x13 */ {{0, 1}, {4, 4}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x14 */ {{2, 2}, {4, 4}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x15 */ {{0, 0}, {2, 2}, {4, 4}, {0, 0} } }, {0, 0, 2, 0, /* 0x16 */ {{1, 2}, {4, 4}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x17 */ {{0, 2}, {4, 4}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x18 */ {{3, 4}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x19 */ {{0, 0}, {3, 4}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x1a */ {{1, 1}, {3, 4}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x1b */ {{0, 1}, {3, 4}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x1c */ {{2, 4}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x1d */ {{0, 0}, {2, 4}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x1e */ {{1, 4}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x1f */ {{0, 4}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x20 */ {{5, 5}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x21 */ {{0, 0}, {5, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x22 */ {{1, 1}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x23 */ {{0, 1}, {5, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x24 */ {{2, 2}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x25 */ {{0, 0}, {2, 2}, {5, 5}, {0, 0} } }, {0, 0, 2, 0, /* 0x26 */ {{1, 2}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x27 */ {{0, 2}, {5, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x28 */ {{3, 3}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x29 */ {{0, 0}, {3, 3}, {5, 5}, {0, 0} } }, {0, 0, 3, 0, /* 0x2a */ {{1, 1}, {3, 3}, {5, 5}, {0, 0} } }, {1, 0, 3, 0, /* 0x2b */ {{0, 1}, {3, 3}, {5, 5}, {0, 0} } }, {0, 0, 2, 0, /* 0x2c */ {{2, 3}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x2d */ {{0, 0}, {2, 3}, {5, 5}, {0, 0} } }, {0, 0, 2, 0, /* 0x2e */ {{1, 3}, {5, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x2f */ {{0, 3}, {5, 5}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x30 */ {{4, 5}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x31 */ {{0, 0}, {4, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x32 */ {{1, 1}, {4, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x33 */ {{0, 1}, {4, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x34 */ {{2, 2}, {4, 5}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x35 */ {{0, 0}, {2, 2}, {4, 5}, {0, 0} } }, {0, 0, 2, 0, /* 0x36 */ {{1, 2}, {4, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x37 */ {{0, 2}, {4, 5}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x38 */ {{3, 5}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x39 */ {{0, 0}, {3, 5}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x3a */ {{1, 1}, {3, 5}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x3b */ {{0, 1}, {3, 5}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x3c */ {{2, 5}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x3d */ {{0, 0}, {2, 5}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x3e */ {{1, 5}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x3f */ {{0, 5}, {0, 0}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x40 */ {{6, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x41 */ {{0, 0}, {6, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x42 */ {{1, 1}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x43 */ {{0, 1}, {6, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x44 */ {{2, 2}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x45 */ {{0, 0}, {2, 2}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x46 */ {{1, 2}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x47 */ {{0, 2}, {6, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x48 */ {{3, 3}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x49 */ {{0, 0}, {3, 3}, {6, 6}, {0, 0} } }, {0, 0, 3, 0, /* 0x4a */ {{1, 1}, {3, 3}, {6, 6}, {0, 0} } }, {1, 0, 3, 0, /* 0x4b */ {{0, 1}, {3, 3}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x4c */ {{2, 3}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x4d */ {{0, 0}, {2, 3}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x4e */ {{1, 3}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x4f */ {{0, 3}, {6, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x50 */ {{4, 4}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x51 */ {{0, 0}, {4, 4}, {6, 6}, {0, 0} } }, {0, 0, 3, 0, /* 0x52 */ {{1, 1}, {4, 4}, {6, 6}, {0, 0} } }, {1, 0, 3, 0, /* 0x53 */ {{0, 1}, {4, 4}, {6, 6}, {0, 0} } }, {0, 0, 3, 0, /* 0x54 */ {{2, 2}, {4, 4}, {6, 6}, {0, 0} } }, {1, 0, 4, 0, /* 0x55 */ {{0, 0}, {2, 2}, {4, 4}, {6, 6} } }, {0, 0, 3, 0, /* 0x56 */ {{1, 2}, {4, 4}, {6, 6}, {0, 0} } }, {1, 0, 3, 0, /* 0x57 */ {{0, 2}, {4, 4}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x58 */ {{3, 4}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x59 */ {{0, 0}, {3, 4}, {6, 6}, {0, 0} } }, {0, 0, 3, 0, /* 0x5a */ {{1, 1}, {3, 4}, {6, 6}, {0, 0} } }, {1, 0, 3, 0, /* 0x5b */ {{0, 1}, {3, 4}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x5c */ {{2, 4}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x5d */ {{0, 0}, {2, 4}, {6, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x5e */ {{1, 4}, {6, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x5f */ {{0, 4}, {6, 6}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x60 */ {{5, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x61 */ {{0, 0}, {5, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x62 */ {{1, 1}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x63 */ {{0, 1}, {5, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x64 */ {{2, 2}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x65 */ {{0, 0}, {2, 2}, {5, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x66 */ {{1, 2}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x67 */ {{0, 2}, {5, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x68 */ {{3, 3}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x69 */ {{0, 0}, {3, 3}, {5, 6}, {0, 0} } }, {0, 0, 3, 0, /* 0x6a */ {{1, 1}, {3, 3}, {5, 6}, {0, 0} } }, {1, 0, 3, 0, /* 0x6b */ {{0, 1}, {3, 3}, {5, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x6c */ {{2, 3}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x6d */ {{0, 0}, {2, 3}, {5, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x6e */ {{1, 3}, {5, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x6f */ {{0, 3}, {5, 6}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x70 */ {{4, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x71 */ {{0, 0}, {4, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x72 */ {{1, 1}, {4, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x73 */ {{0, 1}, {4, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x74 */ {{2, 2}, {4, 6}, {0, 0}, {0, 0} } }, {1, 0, 3, 0, /* 0x75 */ {{0, 0}, {2, 2}, {4, 6}, {0, 0} } }, {0, 0, 2, 0, /* 0x76 */ {{1, 2}, {4, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x77 */ {{0, 2}, {4, 6}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x78 */ {{3, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x79 */ {{0, 0}, {3, 6}, {0, 0}, {0, 0} } }, {0, 0, 2, 0, /* 0x7a */ {{1, 1}, {3, 6}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x7b */ {{0, 1}, {3, 6}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x7c */ {{2, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 2, 0, /* 0x7d */ {{0, 0}, {2, 6}, {0, 0}, {0, 0} } }, {0, 0, 1, 0, /* 0x7e */ {{1, 6}, {0, 0}, {0, 0}, {0, 0} } }, {1, 0, 1, 0, /* 0x7f */ {{0, 6}, {0, 0}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0x80 */ {{7, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0x81 */ {{0, 0}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0x82 */ {{1, 1}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0x83 */ {{0, 1}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0x84 */ {{2, 2}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x85 */ {{0, 0}, {2, 2}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x86 */ {{1, 2}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0x87 */ {{0, 2}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0x88 */ {{3, 3}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x89 */ {{0, 0}, {3, 3}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0x8a */ {{1, 1}, {3, 3}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0x8b */ {{0, 1}, {3, 3}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x8c */ {{2, 3}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x8d */ {{0, 0}, {2, 3}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x8e */ {{1, 3}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0x8f */ {{0, 3}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0x90 */ {{4, 4}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x91 */ {{0, 0}, {4, 4}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0x92 */ {{1, 1}, {4, 4}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0x93 */ {{0, 1}, {4, 4}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0x94 */ {{2, 2}, {4, 4}, {7, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0x95 */ {{0, 0}, {2, 2}, {4, 4}, {7, 7} } }, {0, 1, 3, 0, /* 0x96 */ {{1, 2}, {4, 4}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0x97 */ {{0, 2}, {4, 4}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x98 */ {{3, 4}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x99 */ {{0, 0}, {3, 4}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0x9a */ {{1, 1}, {3, 4}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0x9b */ {{0, 1}, {3, 4}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x9c */ {{2, 4}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0x9d */ {{0, 0}, {2, 4}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0x9e */ {{1, 4}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0x9f */ {{0, 4}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xa0 */ {{5, 5}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xa1 */ {{0, 0}, {5, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xa2 */ {{1, 1}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xa3 */ {{0, 1}, {5, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xa4 */ {{2, 2}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0xa5 */ {{0, 0}, {2, 2}, {5, 5}, {7, 7} } }, {0, 1, 3, 0, /* 0xa6 */ {{1, 2}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xa7 */ {{0, 2}, {5, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xa8 */ {{3, 3}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0xa9 */ {{0, 0}, {3, 3}, {5, 5}, {7, 7} } }, {0, 1, 4, 0, /* 0xaa */ {{1, 1}, {3, 3}, {5, 5}, {7, 7} } }, {1, 1, 4, 0, /* 0xab */ {{0, 1}, {3, 3}, {5, 5}, {7, 7} } }, {0, 1, 3, 0, /* 0xac */ {{2, 3}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0xad */ {{0, 0}, {2, 3}, {5, 5}, {7, 7} } }, {0, 1, 3, 0, /* 0xae */ {{1, 3}, {5, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xaf */ {{0, 3}, {5, 5}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xb0 */ {{4, 5}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xb1 */ {{0, 0}, {4, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xb2 */ {{1, 1}, {4, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xb3 */ {{0, 1}, {4, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xb4 */ {{2, 2}, {4, 5}, {7, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0xb5 */ {{0, 0}, {2, 2}, {4, 5}, {7, 7} } }, {0, 1, 3, 0, /* 0xb6 */ {{1, 2}, {4, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xb7 */ {{0, 2}, {4, 5}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xb8 */ {{3, 5}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xb9 */ {{0, 0}, {3, 5}, {7, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xba */ {{1, 1}, {3, 5}, {7, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xbb */ {{0, 1}, {3, 5}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xbc */ {{2, 5}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xbd */ {{0, 0}, {2, 5}, {7, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xbe */ {{1, 5}, {7, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xbf */ {{0, 5}, {7, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xc0 */ {{6, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xc1 */ {{0, 0}, {6, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xc2 */ {{1, 1}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xc3 */ {{0, 1}, {6, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xc4 */ {{2, 2}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xc5 */ {{0, 0}, {2, 2}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xc6 */ {{1, 2}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xc7 */ {{0, 2}, {6, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xc8 */ {{3, 3}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xc9 */ {{0, 0}, {3, 3}, {6, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xca */ {{1, 1}, {3, 3}, {6, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xcb */ {{0, 1}, {3, 3}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xcc */ {{2, 3}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xcd */ {{0, 0}, {2, 3}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xce */ {{1, 3}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xcf */ {{0, 3}, {6, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xd0 */ {{4, 4}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xd1 */ {{0, 0}, {4, 4}, {6, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xd2 */ {{1, 1}, {4, 4}, {6, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xd3 */ {{0, 1}, {4, 4}, {6, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xd4 */ {{2, 2}, {4, 4}, {6, 7}, {0, 0} } }, {1, 1, 4, 0, /* 0xd5 */ {{0, 0}, {2, 2}, {4, 4}, {6, 7} } }, {0, 1, 3, 0, /* 0xd6 */ {{1, 2}, {4, 4}, {6, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xd7 */ {{0, 2}, {4, 4}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xd8 */ {{3, 4}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xd9 */ {{0, 0}, {3, 4}, {6, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xda */ {{1, 1}, {3, 4}, {6, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xdb */ {{0, 1}, {3, 4}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xdc */ {{2, 4}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xdd */ {{0, 0}, {2, 4}, {6, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xde */ {{1, 4}, {6, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xdf */ {{0, 4}, {6, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xe0 */ {{5, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xe1 */ {{0, 0}, {5, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xe2 */ {{1, 1}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xe3 */ {{0, 1}, {5, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xe4 */ {{2, 2}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xe5 */ {{0, 0}, {2, 2}, {5, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xe6 */ {{1, 2}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xe7 */ {{0, 2}, {5, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xe8 */ {{3, 3}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xe9 */ {{0, 0}, {3, 3}, {5, 7}, {0, 0} } }, {0, 1, 3, 0, /* 0xea */ {{1, 1}, {3, 3}, {5, 7}, {0, 0} } }, {1, 1, 3, 0, /* 0xeb */ {{0, 1}, {3, 3}, {5, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xec */ {{2, 3}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xed */ {{0, 0}, {2, 3}, {5, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xee */ {{1, 3}, {5, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xef */ {{0, 3}, {5, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xf0 */ {{4, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xf1 */ {{0, 0}, {4, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xf2 */ {{1, 1}, {4, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xf3 */ {{0, 1}, {4, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xf4 */ {{2, 2}, {4, 7}, {0, 0}, {0, 0} } }, {1, 1, 3, 0, /* 0xf5 */ {{0, 0}, {2, 2}, {4, 7}, {0, 0} } }, {0, 1, 2, 0, /* 0xf6 */ {{1, 2}, {4, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xf7 */ {{0, 2}, {4, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xf8 */ {{3, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xf9 */ {{0, 0}, {3, 7}, {0, 0}, {0, 0} } }, {0, 1, 2, 0, /* 0xfa */ {{1, 1}, {3, 7}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xfb */ {{0, 1}, {3, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xfc */ {{2, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 2, 0, /* 0xfd */ {{0, 0}, {2, 7}, {0, 0}, {0, 0} } }, {0, 1, 1, 0, /* 0xfe */ {{1, 7}, {0, 0}, {0, 0}, {0, 0} } }, {1, 1, 1, 0, /* 0xff */ {{0, 7}, {0, 0}, {0, 0}, {0, 0} } } }; int sctp_is_address_in_scope(struct sctp_ifa *ifa, struct sctp_scoping *scope, int do_update) { if ((scope->loopback_scope == 0) && (ifa->ifn_p) && SCTP_IFN_IS_IFT_LOOP(ifa->ifn_p)) { /* * skip loopback if not in scope * */ return (0); } switch (ifa->address.sa.sa_family) { #ifdef INET case AF_INET: if (scope->ipv4_addr_legal) { struct sockaddr_in *sin; sin = &ifa->address.sin; if (sin->sin_addr.s_addr == 0) { /* not in scope , unspecified */ return (0); } if ((scope->ipv4_local_scope == 0) && (IN4_ISPRIVATE_ADDRESS(&sin->sin_addr))) { /* private address not in scope */ return (0); } } else { return (0); } break; #endif #ifdef INET6 case AF_INET6: if (scope->ipv6_addr_legal) { struct sockaddr_in6 *sin6; /* * Must update the flags, bummer, which means any * IFA locks must now be applied HERE <-> */ if (do_update) { sctp_gather_internal_ifa_flags(ifa); } if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { return (0); } /* ok to use deprecated addresses? */ sin6 = &ifa->address.sin6; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* skip unspecifed addresses */ return (0); } if ( /* (local_scope == 0) && */ (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))) { return (0); } if ((scope->site_scope == 0) && (IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr))) { return (0); } } else { return (0); } break; #endif default: return (0); } return (1); } static struct mbuf * sctp_add_addr_to_mbuf(struct mbuf *m, struct sctp_ifa *ifa, uint16_t *len) { #if defined(INET) || defined(INET6) struct sctp_paramhdr *paramh; struct mbuf *mret; uint16_t plen; #endif switch (ifa->address.sa.sa_family) { #ifdef INET case AF_INET: plen = (uint16_t)sizeof(struct sctp_ipv4addr_param); break; #endif #ifdef INET6 case AF_INET6: plen = (uint16_t)sizeof(struct sctp_ipv6addr_param); break; #endif default: return (m); } #if defined(INET) || defined(INET6) if (M_TRAILINGSPACE(m) >= plen) { /* easy side we just drop it on the end */ paramh = (struct sctp_paramhdr *)(SCTP_BUF_AT(m, SCTP_BUF_LEN(m))); mret = m; } else { /* Need more space */ mret = m; while (SCTP_BUF_NEXT(mret) != NULL) { mret = SCTP_BUF_NEXT(mret); } SCTP_BUF_NEXT(mret) = sctp_get_mbuf_for_msg(plen, 0, M_NOWAIT, 1, MT_DATA); if (SCTP_BUF_NEXT(mret) == NULL) { /* We are hosed, can't add more addresses */ return (m); } mret = SCTP_BUF_NEXT(mret); paramh = mtod(mret, struct sctp_paramhdr *); } /* now add the parameter */ switch (ifa->address.sa.sa_family) { #ifdef INET case AF_INET: { struct sctp_ipv4addr_param *ipv4p; struct sockaddr_in *sin; sin = &ifa->address.sin; ipv4p = (struct sctp_ipv4addr_param *)paramh; paramh->param_type = htons(SCTP_IPV4_ADDRESS); paramh->param_length = htons(plen); ipv4p->addr = sin->sin_addr.s_addr; SCTP_BUF_LEN(mret) += plen; break; } #endif #ifdef INET6 case AF_INET6: { struct sctp_ipv6addr_param *ipv6p; struct sockaddr_in6 *sin6; sin6 = &ifa->address.sin6; ipv6p = (struct sctp_ipv6addr_param *)paramh; paramh->param_type = htons(SCTP_IPV6_ADDRESS); paramh->param_length = htons(plen); memcpy(ipv6p->addr, &sin6->sin6_addr, sizeof(ipv6p->addr)); /* clear embedded scope in the address */ in6_clearscope((struct in6_addr *)ipv6p->addr); SCTP_BUF_LEN(mret) += plen; break; } #endif default: return (m); } if (len != NULL) { *len += plen; } return (mret); #endif } struct mbuf * sctp_add_addresses_to_i_ia(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_scoping *scope, struct mbuf *m_at, int cnt_inits_to, uint16_t *padding_len, uint16_t *chunk_len) { struct sctp_vrf *vrf = NULL; int cnt, limit_out = 0, total_count; uint32_t vrf_id; vrf_id = inp->def_vrf_id; SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { SCTP_IPI_ADDR_RUNLOCK(); return (m_at); } if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { struct sctp_ifa *sctp_ifap; struct sctp_ifn *sctp_ifnp; cnt = cnt_inits_to; if (vrf->total_ifa_count > SCTP_COUNT_LIMIT) { limit_out = 1; cnt = SCTP_ADDRESS_LIMIT; goto skip_count; } LIST_FOREACH(sctp_ifnp, &vrf->ifnlist, next_ifn) { if ((scope->loopback_scope == 0) && SCTP_IFN_IS_IFT_LOOP(sctp_ifnp)) { /* * Skip loopback devices if loopback_scope * not set */ continue; } LIST_FOREACH(sctp_ifap, &sctp_ifnp->ifalist, next_ifa) { #ifdef INET if ((sctp_ifap->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifap->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifap->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifap->address.sin6.sin6_addr) != 0)) { continue; } #endif if (sctp_is_addr_restricted(stcb, sctp_ifap)) { continue; } if (sctp_is_address_in_scope(sctp_ifap, scope, 1) == 0) { continue; } cnt++; if (cnt > SCTP_ADDRESS_LIMIT) { break; } } if (cnt > SCTP_ADDRESS_LIMIT) { break; } } skip_count: if (cnt > 1) { total_count = 0; LIST_FOREACH(sctp_ifnp, &vrf->ifnlist, next_ifn) { cnt = 0; if ((scope->loopback_scope == 0) && SCTP_IFN_IS_IFT_LOOP(sctp_ifnp)) { /* * Skip loopback devices if * loopback_scope not set */ continue; } LIST_FOREACH(sctp_ifap, &sctp_ifnp->ifalist, next_ifa) { #ifdef INET if ((sctp_ifap->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifap->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifap->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifap->address.sin6.sin6_addr) != 0)) { continue; } #endif if (sctp_is_addr_restricted(stcb, sctp_ifap)) { continue; } if (sctp_is_address_in_scope(sctp_ifap, scope, 0) == 0) { continue; } if ((chunk_len != NULL) && (padding_len != NULL) && (*padding_len > 0)) { memset(mtod(m_at, caddr_t)+*chunk_len, 0, *padding_len); SCTP_BUF_LEN(m_at) += *padding_len; *chunk_len += *padding_len; *padding_len = 0; } m_at = sctp_add_addr_to_mbuf(m_at, sctp_ifap, chunk_len); if (limit_out) { cnt++; total_count++; if (cnt >= 2) { /* * two from each * address */ break; } if (total_count > SCTP_ADDRESS_LIMIT) { /* No more addresses */ break; } } } } } } else { struct sctp_laddr *laddr; cnt = cnt_inits_to; /* First, how many ? */ LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { continue; } if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) /* * Address being deleted by the system, dont * list. */ continue; if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* * Address being deleted on this ep don't * list. */ continue; } if (sctp_is_address_in_scope(laddr->ifa, scope, 1) == 0) { continue; } cnt++; } /* * To get through a NAT we only list addresses if we have * more than one. That way if you just bind a single address * we let the source of the init dictate our address. */ if (cnt > 1) { cnt = cnt_inits_to; LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { continue; } if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) { continue; } if (sctp_is_address_in_scope(laddr->ifa, scope, 0) == 0) { continue; } if ((chunk_len != NULL) && (padding_len != NULL) && (*padding_len > 0)) { memset(mtod(m_at, caddr_t)+*chunk_len, 0, *padding_len); SCTP_BUF_LEN(m_at) += *padding_len; *chunk_len += *padding_len; *padding_len = 0; } m_at = sctp_add_addr_to_mbuf(m_at, laddr->ifa, chunk_len); cnt++; if (cnt >= SCTP_ADDRESS_LIMIT) { break; } } } } SCTP_IPI_ADDR_RUNLOCK(); return (m_at); } static struct sctp_ifa * sctp_is_ifa_addr_preferred(struct sctp_ifa *ifa, uint8_t dest_is_loop, uint8_t dest_is_priv, sa_family_t fam) { uint8_t dest_is_global = 0; /* dest_is_priv is true if destination is a private address */ /* dest_is_loop is true if destination is a loopback addresses */ /** * Here we determine if its a preferred address. A preferred address * means it is the same scope or higher scope then the destination. * L = loopback, P = private, G = global * ----------------------------------------- * src | dest | result * ---------------------------------------- * L | L | yes * ----------------------------------------- * P | L | yes-v4 no-v6 * ----------------------------------------- * G | L | yes-v4 no-v6 * ----------------------------------------- * L | P | no * ----------------------------------------- * P | P | yes * ----------------------------------------- * G | P | no * ----------------------------------------- * L | G | no * ----------------------------------------- * P | G | no * ----------------------------------------- * G | G | yes * ----------------------------------------- */ if (ifa->address.sa.sa_family != fam) { /* forget mis-matched family */ return (NULL); } if ((dest_is_priv == 0) && (dest_is_loop == 0)) { dest_is_global = 1; } SCTPDBG(SCTP_DEBUG_OUTPUT2, "Is destination preferred:"); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &ifa->address.sa); /* Ok the address may be ok */ #ifdef INET6 if (fam == AF_INET6) { /* ok to use deprecated addresses? no lets not! */ if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:1\n"); return (NULL); } if (ifa->src_is_priv && !ifa->src_is_loop) { if (dest_is_loop) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:2\n"); return (NULL); } } if (ifa->src_is_glob) { if (dest_is_loop) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:3\n"); return (NULL); } } } #endif /* * Now that we know what is what, implement or table this could in * theory be done slicker (it used to be), but this is * straightforward and easier to validate :-) */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "src_loop:%d src_priv:%d src_glob:%d\n", ifa->src_is_loop, ifa->src_is_priv, ifa->src_is_glob); SCTPDBG(SCTP_DEBUG_OUTPUT3, "dest_loop:%d dest_priv:%d dest_glob:%d\n", dest_is_loop, dest_is_priv, dest_is_global); if ((ifa->src_is_loop) && (dest_is_priv)) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:4\n"); return (NULL); } if ((ifa->src_is_glob) && (dest_is_priv)) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:5\n"); return (NULL); } if ((ifa->src_is_loop) && (dest_is_global)) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:6\n"); return (NULL); } if ((ifa->src_is_priv) && (dest_is_global)) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "NO:7\n"); return (NULL); } SCTPDBG(SCTP_DEBUG_OUTPUT3, "YES\n"); /* its a preferred address */ return (ifa); } static struct sctp_ifa * sctp_is_ifa_addr_acceptable(struct sctp_ifa *ifa, uint8_t dest_is_loop, uint8_t dest_is_priv, sa_family_t fam) { uint8_t dest_is_global = 0; /** * Here we determine if its a acceptable address. A acceptable * address means it is the same scope or higher scope but we can * allow for NAT which means its ok to have a global dest and a * private src. * * L = loopback, P = private, G = global * ----------------------------------------- * src | dest | result * ----------------------------------------- * L | L | yes * ----------------------------------------- * P | L | yes-v4 no-v6 * ----------------------------------------- * G | L | yes * ----------------------------------------- * L | P | no * ----------------------------------------- * P | P | yes * ----------------------------------------- * G | P | yes - May not work * ----------------------------------------- * L | G | no * ----------------------------------------- * P | G | yes - May not work * ----------------------------------------- * G | G | yes * ----------------------------------------- */ if (ifa->address.sa.sa_family != fam) { /* forget non matching family */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "ifa_fam:%d fam:%d\n", ifa->address.sa.sa_family, fam); return (NULL); } /* Ok the address may be ok */ SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT3, &ifa->address.sa); SCTPDBG(SCTP_DEBUG_OUTPUT3, "dst_is_loop:%d dest_is_priv:%d\n", dest_is_loop, dest_is_priv); if ((dest_is_loop == 0) && (dest_is_priv == 0)) { dest_is_global = 1; } #ifdef INET6 if (fam == AF_INET6) { /* ok to use deprecated addresses? */ if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { return (NULL); } if (ifa->src_is_priv) { /* Special case, linklocal to loop */ if (dest_is_loop) return (NULL); } } #endif /* * Now that we know what is what, implement our table. This could in * theory be done slicker (it used to be), but this is * straightforward and easier to validate :-) */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "ifa->src_is_loop:%d dest_is_priv:%d\n", ifa->src_is_loop, dest_is_priv); if ((ifa->src_is_loop == 1) && (dest_is_priv)) { return (NULL); } SCTPDBG(SCTP_DEBUG_OUTPUT3, "ifa->src_is_loop:%d dest_is_glob:%d\n", ifa->src_is_loop, dest_is_global); if ((ifa->src_is_loop == 1) && (dest_is_global)) { return (NULL); } SCTPDBG(SCTP_DEBUG_OUTPUT3, "address is acceptable\n"); /* its an acceptable address */ return (ifa); } int sctp_is_addr_restricted(struct sctp_tcb *stcb, struct sctp_ifa *ifa) { struct sctp_laddr *laddr; if (stcb == NULL) { /* There are no restrictions, no TCB :-) */ return (0); } LIST_FOREACH(laddr, &stcb->asoc.sctp_restricted_addrs, sctp_nxt_addr) { if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "%s: NULL ifa\n", __func__); continue; } if (laddr->ifa == ifa) { /* Yes it is on the list */ return (1); } } return (0); } int sctp_is_addr_in_ep(struct sctp_inpcb *inp, struct sctp_ifa *ifa) { struct sctp_laddr *laddr; if (ifa == NULL) return (0); LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "%s: NULL ifa\n", __func__); continue; } if ((laddr->ifa == ifa) && laddr->action == 0) /* same pointer */ return (1); } return (0); } static struct sctp_ifa * sctp_choose_boundspecific_inp(struct sctp_inpcb *inp, sctp_route_t *ro, uint32_t vrf_id, int non_asoc_addr_ok, uint8_t dest_is_priv, uint8_t dest_is_loop, sa_family_t fam) { struct sctp_laddr *laddr, *starting_point; void *ifn; int resettotop = 0; struct sctp_ifn *sctp_ifn; struct sctp_ifa *sctp_ifa, *sifa; struct sctp_vrf *vrf; uint32_t ifn_index; vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) return (NULL); ifn = SCTP_GET_IFN_VOID_FROM_ROUTE(ro); ifn_index = SCTP_GET_IF_INDEX_FROM_ROUTE(ro); sctp_ifn = sctp_find_ifn(ifn, ifn_index); /* * first question, is the ifn we will emit on in our list, if so, we * want such an address. Note that we first looked for a preferred * address. */ if (sctp_ifn) { /* is a preferred one on the interface we route out? */ LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; sifa = sctp_is_ifa_addr_preferred(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (sctp_is_addr_in_ep(inp, sifa)) { atomic_add_int(&sifa->refcount, 1); return (sifa); } } } /* * ok, now we now need to find one on the list of the addresses. We * can't get one on the emitting interface so let's find first a * preferred one. If not that an acceptable one otherwise... we * return NULL. */ starting_point = inp->next_addr_touse; once_again: if (inp->next_addr_touse == NULL) { inp->next_addr_touse = LIST_FIRST(&inp->sctp_addr_list); resettotop = 1; } for (laddr = inp->next_addr_touse; laddr; laddr = LIST_NEXT(laddr, sctp_nxt_addr)) { if (laddr->ifa == NULL) { /* address has been removed */ continue; } if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* address is being deleted */ continue; } sifa = sctp_is_ifa_addr_preferred(laddr->ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; atomic_add_int(&sifa->refcount, 1); return (sifa); } if (resettotop == 0) { inp->next_addr_touse = NULL; goto once_again; } inp->next_addr_touse = starting_point; resettotop = 0; once_again_too: if (inp->next_addr_touse == NULL) { inp->next_addr_touse = LIST_FIRST(&inp->sctp_addr_list); resettotop = 1; } /* ok, what about an acceptable address in the inp */ for (laddr = inp->next_addr_touse; laddr; laddr = LIST_NEXT(laddr, sctp_nxt_addr)) { if (laddr->ifa == NULL) { /* address has been removed */ continue; } if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* address is being deleted */ continue; } sifa = sctp_is_ifa_addr_acceptable(laddr->ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; atomic_add_int(&sifa->refcount, 1); return (sifa); } if (resettotop == 0) { inp->next_addr_touse = NULL; goto once_again_too; } /* * no address bound can be a source for the destination we are in * trouble */ return (NULL); } static struct sctp_ifa * sctp_choose_boundspecific_stcb(struct sctp_inpcb *inp, struct sctp_tcb *stcb, sctp_route_t *ro, uint32_t vrf_id, uint8_t dest_is_priv, uint8_t dest_is_loop, int non_asoc_addr_ok, sa_family_t fam) { struct sctp_laddr *laddr, *starting_point; void *ifn; struct sctp_ifn *sctp_ifn; struct sctp_ifa *sctp_ifa, *sifa; uint8_t start_at_beginning = 0; struct sctp_vrf *vrf; uint32_t ifn_index; /* * first question, is the ifn we will emit on in our list, if so, we * want that one. */ vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) return (NULL); ifn = SCTP_GET_IFN_VOID_FROM_ROUTE(ro); ifn_index = SCTP_GET_IF_INDEX_FROM_ROUTE(ro); sctp_ifn = sctp_find_ifn(ifn, ifn_index); /* * first question, is the ifn we will emit on in our list? If so, * we want that one. First we look for a preferred. Second, we go * for an acceptable. */ if (sctp_ifn) { /* first try for a preferred address on the ep */ LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; if (sctp_is_addr_in_ep(inp, sctp_ifa)) { sifa = sctp_is_ifa_addr_preferred(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* on the no-no list */ continue; } atomic_add_int(&sifa->refcount, 1); return (sifa); } } /* next try for an acceptable address on the ep */ LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; if (sctp_is_addr_in_ep(inp, sctp_ifa)) { sifa = sctp_is_ifa_addr_acceptable(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* on the no-no list */ continue; } atomic_add_int(&sifa->refcount, 1); return (sifa); } } } /* * if we can't find one like that then we must look at all addresses * bound to pick one at first preferable then secondly acceptable. */ starting_point = stcb->asoc.last_used_address; sctp_from_the_top: if (stcb->asoc.last_used_address == NULL) { start_at_beginning = 1; stcb->asoc.last_used_address = LIST_FIRST(&inp->sctp_addr_list); } /* search beginning with the last used address */ for (laddr = stcb->asoc.last_used_address; laddr; laddr = LIST_NEXT(laddr, sctp_nxt_addr)) { if (laddr->ifa == NULL) { /* address has been removed */ continue; } if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* address is being deleted */ continue; } sifa = sctp_is_ifa_addr_preferred(laddr->ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* on the no-no list */ continue; } stcb->asoc.last_used_address = laddr; atomic_add_int(&sifa->refcount, 1); return (sifa); } if (start_at_beginning == 0) { stcb->asoc.last_used_address = NULL; goto sctp_from_the_top; } /* now try for any higher scope than the destination */ stcb->asoc.last_used_address = starting_point; start_at_beginning = 0; sctp_from_the_top2: if (stcb->asoc.last_used_address == NULL) { start_at_beginning = 1; stcb->asoc.last_used_address = LIST_FIRST(&inp->sctp_addr_list); } /* search beginning with the last used address */ for (laddr = stcb->asoc.last_used_address; laddr; laddr = LIST_NEXT(laddr, sctp_nxt_addr)) { if (laddr->ifa == NULL) { /* address has been removed */ continue; } if (laddr->action == SCTP_DEL_IP_ADDRESS) { /* address is being deleted */ continue; } sifa = sctp_is_ifa_addr_acceptable(laddr->ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* on the no-no list */ continue; } stcb->asoc.last_used_address = laddr; atomic_add_int(&sifa->refcount, 1); return (sifa); } if (start_at_beginning == 0) { stcb->asoc.last_used_address = NULL; goto sctp_from_the_top2; } return (NULL); } static struct sctp_ifa * sctp_select_nth_preferred_addr_from_ifn_boundall(struct sctp_ifn *ifn, struct sctp_inpcb *inp, struct sctp_tcb *stcb, int non_asoc_addr_ok, uint8_t dest_is_loop, uint8_t dest_is_priv, int addr_wanted, sa_family_t fam, sctp_route_t *ro ) { struct sctp_ifa *ifa, *sifa; int num_eligible_addr = 0; #ifdef INET6 struct sockaddr_in6 sin6, lsa6; if (fam == AF_INET6) { memcpy(&sin6, &ro->ro_dst, sizeof(struct sockaddr_in6)); (void)sa6_recoverscope(&sin6); } #endif /* INET6 */ LIST_FOREACH(ifa, &ifn->ifalist, next_ifa) { #ifdef INET if ((ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; sifa = sctp_is_ifa_addr_preferred(ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; #ifdef INET6 if (fam == AF_INET6 && dest_is_loop && sifa->src_is_loop && sifa->src_is_priv) { /* * don't allow fe80::1 to be a src on loop ::1, we * don't list it to the peer so we will get an * abort. */ continue; } if (fam == AF_INET6 && IN6_IS_ADDR_LINKLOCAL(&sifa->address.sin6.sin6_addr) && IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr)) { /* * link-local <-> link-local must belong to the same * scope. */ memcpy(&lsa6, &sifa->address.sin6, sizeof(struct sockaddr_in6)); (void)sa6_recoverscope(&lsa6); if (sin6.sin6_scope_id != lsa6.sin6_scope_id) { continue; } } #endif /* INET6 */ /* * Check if the IPv6 address matches to next-hop. In the * mobile case, old IPv6 address may be not deleted from the * interface. Then, the interface has previous and new * addresses. We should use one corresponding to the * next-hop. (by micchie) */ #ifdef INET6 if (stcb && fam == AF_INET6 && sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE)) { if (sctp_v6src_match_nexthop(&sifa->address.sin6, ro) == 0) { continue; } } #endif #ifdef INET /* Avoid topologically incorrect IPv4 address */ if (stcb && fam == AF_INET && sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE)) { if (sctp_v4src_match_nexthop(sifa, ro) == 0) { continue; } } #endif if (stcb) { if (sctp_is_address_in_scope(ifa, &stcb->asoc.scope, 0) == 0) { continue; } if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* * It is restricted for some reason.. * probably not yet added. */ continue; } } if (num_eligible_addr >= addr_wanted) { return (sifa); } num_eligible_addr++; } return (NULL); } static int sctp_count_num_preferred_boundall(struct sctp_ifn *ifn, struct sctp_inpcb *inp, struct sctp_tcb *stcb, int non_asoc_addr_ok, uint8_t dest_is_loop, uint8_t dest_is_priv, sa_family_t fam) { struct sctp_ifa *ifa, *sifa; int num_eligible_addr = 0; LIST_FOREACH(ifa, &ifn->ifalist, next_ifa) { #ifdef INET if ((ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((ifa->address.sa.sa_family == AF_INET6) && (stcb != NULL) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) { continue; } sifa = sctp_is_ifa_addr_preferred(ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) { continue; } if (stcb) { if (sctp_is_address_in_scope(ifa, &stcb->asoc.scope, 0) == 0) { continue; } if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* * It is restricted for some reason.. * probably not yet added. */ continue; } } num_eligible_addr++; } return (num_eligible_addr); } static struct sctp_ifa * sctp_choose_boundall(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net, sctp_route_t *ro, uint32_t vrf_id, uint8_t dest_is_priv, uint8_t dest_is_loop, int non_asoc_addr_ok, sa_family_t fam) { int cur_addr_num = 0, num_preferred = 0; void *ifn; struct sctp_ifn *sctp_ifn, *looked_at = NULL, *emit_ifn; struct sctp_ifa *sctp_ifa, *sifa; uint32_t ifn_index; struct sctp_vrf *vrf; #ifdef INET int retried = 0; #endif /*- * For boundall we can use any address in the association. * If non_asoc_addr_ok is set we can use any address (at least in * theory). So we look for preferred addresses first. If we find one, * we use it. Otherwise we next try to get an address on the * interface, which we should be able to do (unless non_asoc_addr_ok * is false and we are routed out that way). In these cases where we * can't use the address of the interface we go through all the * ifn's looking for an address we can use and fill that in. Punting * means we send back address 0, which will probably cause problems * actually since then IP will fill in the address of the route ifn, * which means we probably already rejected it.. i.e. here comes an * abort :-<. */ vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) return (NULL); ifn = SCTP_GET_IFN_VOID_FROM_ROUTE(ro); ifn_index = SCTP_GET_IF_INDEX_FROM_ROUTE(ro); SCTPDBG(SCTP_DEBUG_OUTPUT2, "ifn from route:%p ifn_index:%d\n", ifn, ifn_index); emit_ifn = looked_at = sctp_ifn = sctp_find_ifn(ifn, ifn_index); if (sctp_ifn == NULL) { /* ?? We don't have this guy ?? */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "No ifn emit interface?\n"); goto bound_all_plan_b; } SCTPDBG(SCTP_DEBUG_OUTPUT2, "ifn_index:%d name:%s is emit interface\n", ifn_index, sctp_ifn->ifn_name); if (net) { cur_addr_num = net->indx_of_eligible_next_to_use; } num_preferred = sctp_count_num_preferred_boundall(sctp_ifn, inp, stcb, non_asoc_addr_ok, dest_is_loop, dest_is_priv, fam); SCTPDBG(SCTP_DEBUG_OUTPUT2, "Found %d preferred source addresses for intf:%s\n", num_preferred, sctp_ifn->ifn_name); if (num_preferred == 0) { /* * no eligible addresses, we must use some other interface * address if we can find one. */ goto bound_all_plan_b; } /* * Ok we have num_eligible_addr set with how many we can use, this * may vary from call to call due to addresses being deprecated * etc.. */ if (cur_addr_num >= num_preferred) { cur_addr_num = 0; } /* * select the nth address from the list (where cur_addr_num is the * nth) and 0 is the first one, 1 is the second one etc... */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "cur_addr_num:%d\n", cur_addr_num); sctp_ifa = sctp_select_nth_preferred_addr_from_ifn_boundall(sctp_ifn, inp, stcb, non_asoc_addr_ok, dest_is_loop, dest_is_priv, cur_addr_num, fam, ro); /* if sctp_ifa is NULL something changed??, fall to plan b. */ if (sctp_ifa) { atomic_add_int(&sctp_ifa->refcount, 1); if (net) { /* save off where the next one we will want */ net->indx_of_eligible_next_to_use = cur_addr_num + 1; } return (sctp_ifa); } /* * plan_b: Look at all interfaces and find a preferred address. If * no preferred fall through to plan_c. */ bound_all_plan_b: SCTPDBG(SCTP_DEBUG_OUTPUT2, "Trying Plan B\n"); LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "Examine interface %s\n", sctp_ifn->ifn_name); if (dest_is_loop == 0 && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { /* wrong base scope */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "skip\n"); continue; } if ((sctp_ifn == looked_at) && looked_at) { /* already looked at this guy */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "already seen\n"); continue; } num_preferred = sctp_count_num_preferred_boundall(sctp_ifn, inp, stcb, non_asoc_addr_ok, dest_is_loop, dest_is_priv, fam); SCTPDBG(SCTP_DEBUG_OUTPUT2, "Found ifn:%p %d preferred source addresses\n", ifn, num_preferred); if (num_preferred == 0) { /* None on this interface. */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "No preferred -- skipping to next\n"); continue; } SCTPDBG(SCTP_DEBUG_OUTPUT2, "num preferred:%d on interface:%p cur_addr_num:%d\n", num_preferred, (void *)sctp_ifn, cur_addr_num); /* * Ok we have num_eligible_addr set with how many we can * use, this may vary from call to call due to addresses * being deprecated etc.. */ if (cur_addr_num >= num_preferred) { cur_addr_num = 0; } sifa = sctp_select_nth_preferred_addr_from_ifn_boundall(sctp_ifn, inp, stcb, non_asoc_addr_ok, dest_is_loop, dest_is_priv, cur_addr_num, fam, ro); if (sifa == NULL) continue; if (net) { net->indx_of_eligible_next_to_use = cur_addr_num + 1; SCTPDBG(SCTP_DEBUG_OUTPUT2, "we selected %d\n", cur_addr_num); SCTPDBG(SCTP_DEBUG_OUTPUT2, "Source:"); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &sifa->address.sa); SCTPDBG(SCTP_DEBUG_OUTPUT2, "Dest:"); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &net->ro._l_addr.sa); } atomic_add_int(&sifa->refcount, 1); return (sifa); } #ifdef INET again_with_private_addresses_allowed: #endif /* plan_c: do we have an acceptable address on the emit interface */ sifa = NULL; SCTPDBG(SCTP_DEBUG_OUTPUT2, "Trying Plan C: find acceptable on interface\n"); if (emit_ifn == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "Jump to Plan D - no emit_ifn\n"); goto plan_d; } LIST_FOREACH(sctp_ifa, &emit_ifn->ifalist, next_ifa) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "ifa:%p\n", (void *)sctp_ifa); #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "Jailed\n"); continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "Jailed\n"); continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "Defer\n"); continue; } sifa = sctp_is_ifa_addr_acceptable(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "IFA not acceptable\n"); continue; } if (stcb) { if (sctp_is_address_in_scope(sifa, &stcb->asoc.scope, 0) == 0) { SCTPDBG(SCTP_DEBUG_OUTPUT2, "NOT in scope\n"); sifa = NULL; continue; } if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* * It is restricted for some reason.. * probably not yet added. */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "Its restricted\n"); sifa = NULL; continue; } } atomic_add_int(&sifa->refcount, 1); goto out; } plan_d: /* * plan_d: We are in trouble. No preferred address on the emit * interface. And not even a preferred address on all interfaces. Go * out and see if we can find an acceptable address somewhere * amongst all interfaces. */ SCTPDBG(SCTP_DEBUG_OUTPUT2, "Trying Plan D looked_at is %p\n", (void *)looked_at); LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { if (dest_is_loop == 0 && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { /* wrong base scope */ continue; } LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; sifa = sctp_is_ifa_addr_acceptable(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (sifa == NULL) continue; if (stcb) { if (sctp_is_address_in_scope(sifa, &stcb->asoc.scope, 0) == 0) { sifa = NULL; continue; } if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, sifa)) && (!sctp_is_addr_pending(stcb, sifa)))) { /* * It is restricted for some * reason.. probably not yet added. */ sifa = NULL; continue; } } goto out; } } #ifdef INET if (stcb) { if ((retried == 0) && (stcb->asoc.scope.ipv4_local_scope == 0)) { stcb->asoc.scope.ipv4_local_scope = 1; retried = 1; goto again_with_private_addresses_allowed; } else if (retried == 1) { stcb->asoc.scope.ipv4_local_scope = 0; } } #endif out: #ifdef INET if (sifa) { if (retried == 1) { LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { if (dest_is_loop == 0 && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { /* wrong base scope */ continue; } LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { struct sctp_ifa *tmp_sifa; #ifdef INET if ((sctp_ifa->address.sa.sa_family == AF_INET) && (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin.sin_addr) != 0)) { continue; } #endif #ifdef INET6 if ((sctp_ifa->address.sa.sa_family == AF_INET6) && (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sctp_ifa->address.sin6.sin6_addr) != 0)) { continue; } #endif if ((sctp_ifa->localifa_flags & SCTP_ADDR_DEFER_USE) && (non_asoc_addr_ok == 0)) continue; tmp_sifa = sctp_is_ifa_addr_acceptable(sctp_ifa, dest_is_loop, dest_is_priv, fam); if (tmp_sifa == NULL) { continue; } if (tmp_sifa == sifa) { continue; } if (stcb) { if (sctp_is_address_in_scope(tmp_sifa, &stcb->asoc.scope, 0) == 0) { continue; } if (((non_asoc_addr_ok == 0) && (sctp_is_addr_restricted(stcb, tmp_sifa))) || (non_asoc_addr_ok && (sctp_is_addr_restricted(stcb, tmp_sifa)) && (!sctp_is_addr_pending(stcb, tmp_sifa)))) { /* * It is restricted * for some reason.. * probably not yet * added. */ continue; } } if ((tmp_sifa->address.sin.sin_family == AF_INET) && (IN4_ISPRIVATE_ADDRESS(&(tmp_sifa->address.sin.sin_addr)))) { sctp_add_local_addr_restricted(stcb, tmp_sifa); } } } } atomic_add_int(&sifa->refcount, 1); } #endif return (sifa); } /* tcb may be NULL */ struct sctp_ifa * sctp_source_address_selection(struct sctp_inpcb *inp, struct sctp_tcb *stcb, sctp_route_t *ro, struct sctp_nets *net, int non_asoc_addr_ok, uint32_t vrf_id) { struct sctp_ifa *answer; uint8_t dest_is_priv, dest_is_loop; sa_family_t fam; #ifdef INET struct sockaddr_in *to = (struct sockaddr_in *)&ro->ro_dst; #endif #ifdef INET6 struct sockaddr_in6 *to6 = (struct sockaddr_in6 *)&ro->ro_dst; #endif /** * Rules: * - Find the route if needed, cache if I can. * - Look at interface address in route, Is it in the bound list. If so we * have the best source. * - If not we must rotate amongst the addresses. * * Cavets and issues * * Do we need to pay attention to scope. We can have a private address * or a global address we are sourcing or sending to. So if we draw * it out * zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz * For V4 * ------------------------------------------ * source * dest * result * ----------------------------------------- * Private * Global * NAT * ----------------------------------------- * Private * Private * No problem * ----------------------------------------- * Global * Private * Huh, How will this work? * ----------------------------------------- * Global * Global * No Problem *------------------------------------------ * zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz * For V6 *------------------------------------------ * source * dest * result * ----------------------------------------- * Linklocal * Global * * ----------------------------------------- * Linklocal * Linklocal * No problem * ----------------------------------------- * Global * Linklocal * Huh, How will this work? * ----------------------------------------- * Global * Global * No Problem *------------------------------------------ * zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz * * And then we add to that what happens if there are multiple addresses * assigned to an interface. Remember the ifa on a ifn is a linked * list of addresses. So one interface can have more than one IP * address. What happens if we have both a private and a global * address? Do we then use context of destination to sort out which * one is best? And what about NAT's sending P->G may get you a NAT * translation, or should you select the G thats on the interface in * preference. * * Decisions: * * - count the number of addresses on the interface. * - if it is one, no problem except case . * For we will assume a NAT out there. * - if there are more than one, then we need to worry about scope P * or G. We should prefer G -> G and P -> P if possible. * Then as a secondary fall back to mixed types G->P being a last * ditch one. * - The above all works for bound all, but bound specific we need to * use the same concept but instead only consider the bound * addresses. If the bound set is NOT assigned to the interface then * we must use rotation amongst the bound addresses.. */ - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { /* * Need a route to cache. */ SCTP_RTALLOC(ro, vrf_id, inp->fibnum); } - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { return (NULL); } fam = ro->ro_dst.sa_family; dest_is_priv = dest_is_loop = 0; /* Setup our scopes for the destination */ switch (fam) { #ifdef INET case AF_INET: /* Scope based on outbound address */ if (IN4_ISLOOPBACK_ADDRESS(&to->sin_addr)) { dest_is_loop = 1; if (net != NULL) { /* mark it as local */ net->addr_is_local = 1; } } else if ((IN4_ISPRIVATE_ADDRESS(&to->sin_addr))) { dest_is_priv = 1; } break; #endif #ifdef INET6 case AF_INET6: /* Scope based on outbound address */ if (IN6_IS_ADDR_LOOPBACK(&to6->sin6_addr) || SCTP_ROUTE_IS_REAL_LOOP(ro)) { /* * If the address is a loopback address, which * consists of "::1" OR "fe80::1%lo0", we are * loopback scope. But we don't use dest_is_priv * (link local addresses). */ dest_is_loop = 1; if (net != NULL) { /* mark it as local */ net->addr_is_local = 1; } } else if (IN6_IS_ADDR_LINKLOCAL(&to6->sin6_addr)) { dest_is_priv = 1; } break; #endif } SCTPDBG(SCTP_DEBUG_OUTPUT2, "Select source addr for:"); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, (struct sockaddr *)&ro->ro_dst); SCTP_IPI_ADDR_RLOCK(); if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { /* * Bound all case */ answer = sctp_choose_boundall(inp, stcb, net, ro, vrf_id, dest_is_priv, dest_is_loop, non_asoc_addr_ok, fam); SCTP_IPI_ADDR_RUNLOCK(); return (answer); } /* * Subset bound case */ if (stcb) { answer = sctp_choose_boundspecific_stcb(inp, stcb, ro, vrf_id, dest_is_priv, dest_is_loop, non_asoc_addr_ok, fam); } else { answer = sctp_choose_boundspecific_inp(inp, ro, vrf_id, non_asoc_addr_ok, dest_is_priv, dest_is_loop, fam); } SCTP_IPI_ADDR_RUNLOCK(); return (answer); } static int sctp_find_cmsg(int c_type, void *data, struct mbuf *control, size_t cpsize) { struct cmsghdr cmh; struct sctp_sndinfo sndinfo; struct sctp_prinfo prinfo; struct sctp_authinfo authinfo; int tot_len, rem_len, cmsg_data_len, cmsg_data_off, off; int found; /* * Independent of how many mbufs, find the c_type inside the control * structure and copy out the data. */ found = 0; tot_len = SCTP_BUF_LEN(control); for (off = 0; off < tot_len; off += CMSG_ALIGN(cmh.cmsg_len)) { rem_len = tot_len - off; if (rem_len < (int)CMSG_ALIGN(sizeof(cmh))) { /* There is not enough room for one more. */ return (found); } m_copydata(control, off, sizeof(cmh), (caddr_t)&cmh); if (cmh.cmsg_len < CMSG_ALIGN(sizeof(cmh))) { /* We dont't have a complete CMSG header. */ return (found); } if ((cmh.cmsg_len > INT_MAX) || ((int)cmh.cmsg_len > rem_len)) { /* We don't have the complete CMSG. */ return (found); } cmsg_data_len = (int)cmh.cmsg_len - CMSG_ALIGN(sizeof(cmh)); cmsg_data_off = off + CMSG_ALIGN(sizeof(cmh)); if ((cmh.cmsg_level == IPPROTO_SCTP) && ((c_type == cmh.cmsg_type) || ((c_type == SCTP_SNDRCV) && ((cmh.cmsg_type == SCTP_SNDINFO) || (cmh.cmsg_type == SCTP_PRINFO) || (cmh.cmsg_type == SCTP_AUTHINFO))))) { if (c_type == cmh.cmsg_type) { if (cpsize > INT_MAX) { return (found); } if (cmsg_data_len < (int)cpsize) { return (found); } /* It is exactly what we want. Copy it out. */ m_copydata(control, cmsg_data_off, (int)cpsize, (caddr_t)data); return (1); } else { struct sctp_sndrcvinfo *sndrcvinfo; sndrcvinfo = (struct sctp_sndrcvinfo *)data; if (found == 0) { if (cpsize < sizeof(struct sctp_sndrcvinfo)) { return (found); } memset(sndrcvinfo, 0, sizeof(struct sctp_sndrcvinfo)); } switch (cmh.cmsg_type) { case SCTP_SNDINFO: if (cmsg_data_len < (int)sizeof(struct sctp_sndinfo)) { return (found); } m_copydata(control, cmsg_data_off, sizeof(struct sctp_sndinfo), (caddr_t)&sndinfo); sndrcvinfo->sinfo_stream = sndinfo.snd_sid; sndrcvinfo->sinfo_flags = sndinfo.snd_flags; sndrcvinfo->sinfo_ppid = sndinfo.snd_ppid; sndrcvinfo->sinfo_context = sndinfo.snd_context; sndrcvinfo->sinfo_assoc_id = sndinfo.snd_assoc_id; break; case SCTP_PRINFO: if (cmsg_data_len < (int)sizeof(struct sctp_prinfo)) { return (found); } m_copydata(control, cmsg_data_off, sizeof(struct sctp_prinfo), (caddr_t)&prinfo); if (prinfo.pr_policy != SCTP_PR_SCTP_NONE) { sndrcvinfo->sinfo_timetolive = prinfo.pr_value; } else { sndrcvinfo->sinfo_timetolive = 0; } sndrcvinfo->sinfo_flags |= prinfo.pr_policy; break; case SCTP_AUTHINFO: if (cmsg_data_len < (int)sizeof(struct sctp_authinfo)) { return (found); } m_copydata(control, cmsg_data_off, sizeof(struct sctp_authinfo), (caddr_t)&authinfo); sndrcvinfo->sinfo_keynumber_valid = 1; sndrcvinfo->sinfo_keynumber = authinfo.auth_keynumber; break; default: return (found); } found = 1; } } } return (found); } static int sctp_process_cmsgs_for_init(struct sctp_tcb *stcb, struct mbuf *control, int *error) { struct cmsghdr cmh; struct sctp_initmsg initmsg; #ifdef INET struct sockaddr_in sin; #endif #ifdef INET6 struct sockaddr_in6 sin6; #endif int tot_len, rem_len, cmsg_data_len, cmsg_data_off, off; tot_len = SCTP_BUF_LEN(control); for (off = 0; off < tot_len; off += CMSG_ALIGN(cmh.cmsg_len)) { rem_len = tot_len - off; if (rem_len < (int)CMSG_ALIGN(sizeof(cmh))) { /* There is not enough room for one more. */ *error = EINVAL; return (1); } m_copydata(control, off, sizeof(cmh), (caddr_t)&cmh); if (cmh.cmsg_len < CMSG_ALIGN(sizeof(cmh))) { /* We dont't have a complete CMSG header. */ *error = EINVAL; return (1); } if ((cmh.cmsg_len > INT_MAX) || ((int)cmh.cmsg_len > rem_len)) { /* We don't have the complete CMSG. */ *error = EINVAL; return (1); } cmsg_data_len = (int)cmh.cmsg_len - CMSG_ALIGN(sizeof(cmh)); cmsg_data_off = off + CMSG_ALIGN(sizeof(cmh)); if (cmh.cmsg_level == IPPROTO_SCTP) { switch (cmh.cmsg_type) { case SCTP_INIT: if (cmsg_data_len < (int)sizeof(struct sctp_initmsg)) { *error = EINVAL; return (1); } m_copydata(control, cmsg_data_off, sizeof(struct sctp_initmsg), (caddr_t)&initmsg); if (initmsg.sinit_max_attempts) stcb->asoc.max_init_times = initmsg.sinit_max_attempts; if (initmsg.sinit_num_ostreams) stcb->asoc.pre_open_streams = initmsg.sinit_num_ostreams; if (initmsg.sinit_max_instreams) stcb->asoc.max_inbound_streams = initmsg.sinit_max_instreams; if (initmsg.sinit_max_init_timeo) stcb->asoc.initial_init_rto_max = initmsg.sinit_max_init_timeo; if (stcb->asoc.streamoutcnt < stcb->asoc.pre_open_streams) { struct sctp_stream_out *tmp_str; unsigned int i; #if defined(SCTP_DETAILED_STR_STATS) int j; #endif /* Default is NOT correct */ SCTPDBG(SCTP_DEBUG_OUTPUT1, "Ok, default:%d pre_open:%d\n", stcb->asoc.streamoutcnt, stcb->asoc.pre_open_streams); SCTP_TCB_UNLOCK(stcb); SCTP_MALLOC(tmp_str, struct sctp_stream_out *, (stcb->asoc.pre_open_streams * sizeof(struct sctp_stream_out)), SCTP_M_STRMO); SCTP_TCB_LOCK(stcb); if (tmp_str != NULL) { SCTP_FREE(stcb->asoc.strmout, SCTP_M_STRMO); stcb->asoc.strmout = tmp_str; stcb->asoc.strm_realoutsize = stcb->asoc.streamoutcnt = stcb->asoc.pre_open_streams; } else { stcb->asoc.pre_open_streams = stcb->asoc.streamoutcnt; } for (i = 0; i < stcb->asoc.streamoutcnt; i++) { TAILQ_INIT(&stcb->asoc.strmout[i].outqueue); stcb->asoc.strmout[i].chunks_on_queues = 0; stcb->asoc.strmout[i].next_mid_ordered = 0; stcb->asoc.strmout[i].next_mid_unordered = 0; #if defined(SCTP_DETAILED_STR_STATS) for (j = 0; j < SCTP_PR_SCTP_MAX + 1; j++) { stcb->asoc.strmout[i].abandoned_sent[j] = 0; stcb->asoc.strmout[i].abandoned_unsent[j] = 0; } #else stcb->asoc.strmout[i].abandoned_sent[0] = 0; stcb->asoc.strmout[i].abandoned_unsent[0] = 0; #endif stcb->asoc.strmout[i].sid = i; stcb->asoc.strmout[i].last_msg_incomplete = 0; stcb->asoc.strmout[i].state = SCTP_STREAM_OPENING; stcb->asoc.ss_functions.sctp_ss_init_stream(stcb, &stcb->asoc.strmout[i], NULL); } } break; #ifdef INET case SCTP_DSTADDRV4: if (cmsg_data_len < (int)sizeof(struct in_addr)) { *error = EINVAL; return (1); } memset(&sin, 0, sizeof(struct sockaddr_in)); sin.sin_family = AF_INET; sin.sin_len = sizeof(struct sockaddr_in); sin.sin_port = stcb->rport; m_copydata(control, cmsg_data_off, sizeof(struct in_addr), (caddr_t)&sin.sin_addr); if ((sin.sin_addr.s_addr == INADDR_ANY) || (sin.sin_addr.s_addr == INADDR_BROADCAST) || IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { *error = EINVAL; return (1); } if (sctp_add_remote_addr(stcb, (struct sockaddr *)&sin, NULL, stcb->asoc.port, SCTP_DONOT_SETSCOPE, SCTP_ADDR_IS_CONFIRMED)) { *error = ENOBUFS; return (1); } break; #endif #ifdef INET6 case SCTP_DSTADDRV6: if (cmsg_data_len < (int)sizeof(struct in6_addr)) { *error = EINVAL; return (1); } memset(&sin6, 0, sizeof(struct sockaddr_in6)); sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_port = stcb->rport; m_copydata(control, cmsg_data_off, sizeof(struct in6_addr), (caddr_t)&sin6.sin6_addr); if (IN6_IS_ADDR_UNSPECIFIED(&sin6.sin6_addr) || IN6_IS_ADDR_MULTICAST(&sin6.sin6_addr)) { *error = EINVAL; return (1); } #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&sin6.sin6_addr)) { in6_sin6_2_sin(&sin, &sin6); if ((sin.sin_addr.s_addr == INADDR_ANY) || (sin.sin_addr.s_addr == INADDR_BROADCAST) || IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { *error = EINVAL; return (1); } if (sctp_add_remote_addr(stcb, (struct sockaddr *)&sin, NULL, stcb->asoc.port, SCTP_DONOT_SETSCOPE, SCTP_ADDR_IS_CONFIRMED)) { *error = ENOBUFS; return (1); } } else #endif if (sctp_add_remote_addr(stcb, (struct sockaddr *)&sin6, NULL, stcb->asoc.port, SCTP_DONOT_SETSCOPE, SCTP_ADDR_IS_CONFIRMED)) { *error = ENOBUFS; return (1); } break; #endif default: break; } } } return (0); } #if defined(INET) || defined(INET6) static struct sctp_tcb * sctp_findassociation_cmsgs(struct sctp_inpcb **inp_p, uint16_t port, struct mbuf *control, struct sctp_nets **net_p, int *error) { struct cmsghdr cmh; struct sctp_tcb *stcb; struct sockaddr *addr; #ifdef INET struct sockaddr_in sin; #endif #ifdef INET6 struct sockaddr_in6 sin6; #endif int tot_len, rem_len, cmsg_data_len, cmsg_data_off, off; tot_len = SCTP_BUF_LEN(control); for (off = 0; off < tot_len; off += CMSG_ALIGN(cmh.cmsg_len)) { rem_len = tot_len - off; if (rem_len < (int)CMSG_ALIGN(sizeof(cmh))) { /* There is not enough room for one more. */ *error = EINVAL; return (NULL); } m_copydata(control, off, sizeof(cmh), (caddr_t)&cmh); if (cmh.cmsg_len < CMSG_ALIGN(sizeof(cmh))) { /* We dont't have a complete CMSG header. */ *error = EINVAL; return (NULL); } if ((cmh.cmsg_len > INT_MAX) || ((int)cmh.cmsg_len > rem_len)) { /* We don't have the complete CMSG. */ *error = EINVAL; return (NULL); } cmsg_data_len = (int)cmh.cmsg_len - CMSG_ALIGN(sizeof(cmh)); cmsg_data_off = off + CMSG_ALIGN(sizeof(cmh)); if (cmh.cmsg_level == IPPROTO_SCTP) { switch (cmh.cmsg_type) { #ifdef INET case SCTP_DSTADDRV4: if (cmsg_data_len < (int)sizeof(struct in_addr)) { *error = EINVAL; return (NULL); } memset(&sin, 0, sizeof(struct sockaddr_in)); sin.sin_family = AF_INET; sin.sin_len = sizeof(struct sockaddr_in); sin.sin_port = port; m_copydata(control, cmsg_data_off, sizeof(struct in_addr), (caddr_t)&sin.sin_addr); addr = (struct sockaddr *)&sin; break; #endif #ifdef INET6 case SCTP_DSTADDRV6: if (cmsg_data_len < (int)sizeof(struct in6_addr)) { *error = EINVAL; return (NULL); } memset(&sin6, 0, sizeof(struct sockaddr_in6)); sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_port = port; m_copydata(control, cmsg_data_off, sizeof(struct in6_addr), (caddr_t)&sin6.sin6_addr); #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&sin6.sin6_addr)) { in6_sin6_2_sin(&sin, &sin6); addr = (struct sockaddr *)&sin; } else #endif addr = (struct sockaddr *)&sin6; break; #endif default: addr = NULL; break; } if (addr) { stcb = sctp_findassociation_ep_addr(inp_p, addr, net_p, NULL, NULL); if (stcb != NULL) { return (stcb); } } } } return (NULL); } #endif static struct mbuf * sctp_add_cookie(struct mbuf *init, int init_offset, struct mbuf *initack, int initack_offset, struct sctp_state_cookie *stc_in, uint8_t **signature) { struct mbuf *copy_init, *copy_initack, *m_at, *sig, *mret; struct sctp_state_cookie *stc; struct sctp_paramhdr *ph; uint8_t *foo; int sig_offset; uint16_t cookie_sz; mret = sctp_get_mbuf_for_msg((sizeof(struct sctp_state_cookie) + sizeof(struct sctp_paramhdr)), 0, M_NOWAIT, 1, MT_DATA); if (mret == NULL) { return (NULL); } copy_init = SCTP_M_COPYM(init, init_offset, M_COPYALL, M_NOWAIT); if (copy_init == NULL) { sctp_m_freem(mret); return (NULL); } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(copy_init, SCTP_MBUF_ICOPY); } #endif copy_initack = SCTP_M_COPYM(initack, initack_offset, M_COPYALL, M_NOWAIT); if (copy_initack == NULL) { sctp_m_freem(mret); sctp_m_freem(copy_init); return (NULL); } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(copy_initack, SCTP_MBUF_ICOPY); } #endif /* easy side we just drop it on the end */ ph = mtod(mret, struct sctp_paramhdr *); SCTP_BUF_LEN(mret) = sizeof(struct sctp_state_cookie) + sizeof(struct sctp_paramhdr); stc = (struct sctp_state_cookie *)((caddr_t)ph + sizeof(struct sctp_paramhdr)); ph->param_type = htons(SCTP_STATE_COOKIE); ph->param_length = 0; /* fill in at the end */ /* Fill in the stc cookie data */ memcpy(stc, stc_in, sizeof(struct sctp_state_cookie)); /* tack the INIT and then the INIT-ACK onto the chain */ cookie_sz = 0; for (m_at = mret; m_at; m_at = SCTP_BUF_NEXT(m_at)) { cookie_sz += SCTP_BUF_LEN(m_at); if (SCTP_BUF_NEXT(m_at) == NULL) { SCTP_BUF_NEXT(m_at) = copy_init; break; } } for (m_at = copy_init; m_at; m_at = SCTP_BUF_NEXT(m_at)) { cookie_sz += SCTP_BUF_LEN(m_at); if (SCTP_BUF_NEXT(m_at) == NULL) { SCTP_BUF_NEXT(m_at) = copy_initack; break; } } for (m_at = copy_initack; m_at; m_at = SCTP_BUF_NEXT(m_at)) { cookie_sz += SCTP_BUF_LEN(m_at); if (SCTP_BUF_NEXT(m_at) == NULL) { break; } } sig = sctp_get_mbuf_for_msg(SCTP_SECRET_SIZE, 0, M_NOWAIT, 1, MT_DATA); if (sig == NULL) { /* no space, so free the entire chain */ sctp_m_freem(mret); return (NULL); } SCTP_BUF_LEN(sig) = 0; SCTP_BUF_NEXT(m_at) = sig; sig_offset = 0; foo = (uint8_t *)(mtod(sig, caddr_t)+sig_offset); memset(foo, 0, SCTP_SIGNATURE_SIZE); *signature = foo; SCTP_BUF_LEN(sig) += SCTP_SIGNATURE_SIZE; cookie_sz += SCTP_SIGNATURE_SIZE; ph->param_length = htons(cookie_sz); return (mret); } static uint8_t sctp_get_ect(struct sctp_tcb *stcb) { if ((stcb != NULL) && (stcb->asoc.ecn_supported == 1)) { return (SCTP_ECT0_BIT); } else { return (0); } } #if defined(INET) || defined(INET6) static void sctp_handle_no_route(struct sctp_tcb *stcb, struct sctp_nets *net, int so_locked) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "dropped packet - no valid source addr\n"); if (net) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Destination was "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT1, &net->ro._l_addr.sa); if (net->dest_state & SCTP_ADDR_CONFIRMED) { if ((net->dest_state & SCTP_ADDR_REACHABLE) && stcb) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "no route takes interface %p down\n", (void *)net); sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_DOWN, stcb, 0, (void *)net, so_locked); net->dest_state &= ~SCTP_ADDR_REACHABLE; net->dest_state &= ~SCTP_ADDR_PF; } } if (stcb) { if (net == stcb->asoc.primary_destination) { /* need a new primary */ struct sctp_nets *alt; alt = sctp_find_alternate_net(stcb, net, 0); if (alt != net) { if (stcb->asoc.alternate) { sctp_free_remote_addr(stcb->asoc.alternate); } stcb->asoc.alternate = alt; atomic_add_int(&stcb->asoc.alternate->ref_count, 1); if (net->ro._s_addr) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; } net->src_addr_selected = 0; } } } } } #endif static int sctp_lowlevel_chunk_output(struct sctp_inpcb *inp, struct sctp_tcb *stcb, /* may be NULL */ struct sctp_nets *net, struct sockaddr *to, struct mbuf *m, uint32_t auth_offset, struct sctp_auth_chunk *auth, uint16_t auth_keyid, int nofragment_flag, int ecn_ok, int out_of_asoc_ok, uint16_t src_port, uint16_t dest_port, uint32_t v_tag, uint16_t port, union sctp_sockstore *over_addr, uint8_t mflowtype, uint32_t mflowid, #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) int so_locked SCTP_UNUSED #else int so_locked #endif ) { /* nofragment_flag to tell if IP_DF should be set (IPv4 only) */ /** * Given a mbuf chain (via SCTP_BUF_NEXT()) that holds a packet header * WITH an SCTPHDR but no IP header, endpoint inp and sa structure: * - fill in the HMAC digest of any AUTH chunk in the packet. * - calculate and fill in the SCTP checksum. * - prepend an IP address header. * - if boundall use INADDR_ANY. * - if boundspecific do source address selection. * - set fragmentation option for ipV4. * - On return from IP output, check/adjust mtu size of output * interface and smallest_mtu size as well. */ /* Will need ifdefs around this */ struct mbuf *newm; struct sctphdr *sctphdr; int packet_length; int ret; #if defined(INET) || defined(INET6) uint32_t vrf_id; #endif #if defined(INET) || defined(INET6) struct mbuf *o_pak; sctp_route_t *ro = NULL; struct udphdr *udp = NULL; #endif uint8_t tos_value; #if defined(__APPLE__) || defined(SCTP_SO_LOCK_TESTING) struct socket *so = NULL; #endif if ((net) && (net->dest_state & SCTP_ADDR_OUT_OF_SCOPE)) { SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EFAULT); sctp_m_freem(m); return (EFAULT); } #if defined(INET) || defined(INET6) if (stcb) { vrf_id = stcb->asoc.vrf_id; } else { vrf_id = inp->def_vrf_id; } #endif /* fill in the HMAC digest for any AUTH chunk in the packet */ if ((auth != NULL) && (stcb != NULL)) { sctp_fill_hmac_digest_m(m, auth_offset, auth, stcb, auth_keyid); } if (net) { tos_value = net->dscp; } else if (stcb) { tos_value = stcb->asoc.default_dscp; } else { tos_value = inp->sctp_ep.default_dscp; } switch (to->sa_family) { #ifdef INET case AF_INET: { struct ip *ip = NULL; sctp_route_t iproute; int len; len = SCTP_MIN_V4_OVERHEAD; if (port) { len += sizeof(struct udphdr); } newm = sctp_get_mbuf_for_msg(len, 1, M_NOWAIT, 1, MT_DATA); if (newm == NULL) { sctp_m_freem(m); SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_ALIGN_TO_END(newm, len); SCTP_BUF_LEN(newm) = len; SCTP_BUF_NEXT(newm) = m; m = newm; if (net != NULL) { m->m_pkthdr.flowid = net->flowid; M_HASHTYPE_SET(m, net->flowtype); } else { m->m_pkthdr.flowid = mflowid; M_HASHTYPE_SET(m, mflowtype); } packet_length = sctp_calculate_len(m); ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = (sizeof(struct ip) >> 2); if (tos_value == 0) { /* * This means especially, that it is not set * at the SCTP layer. So use the value from * the IP layer. */ tos_value = inp->ip_inp.inp.inp_ip_tos; } tos_value &= 0xfc; if (ecn_ok) { tos_value |= sctp_get_ect(stcb); } if ((nofragment_flag) && (port == 0)) { ip->ip_off = htons(IP_DF); } else { ip->ip_off = htons(0); } /* FreeBSD has a function for ip_id's */ ip_fillid(ip); ip->ip_ttl = inp->ip_inp.inp.inp_ip_ttl; ip->ip_len = htons(packet_length); ip->ip_tos = tos_value; if (port) { ip->ip_p = IPPROTO_UDP; } else { ip->ip_p = IPPROTO_SCTP; } ip->ip_sum = 0; if (net == NULL) { ro = &iproute; memset(&iproute, 0, sizeof(iproute)); memcpy(&ro->ro_dst, to, to->sa_len); } else { ro = (sctp_route_t *)&net->ro; } /* Now the address selection part */ ip->ip_dst.s_addr = ((struct sockaddr_in *)to)->sin_addr.s_addr; /* call the routine to select the src address */ if (net && out_of_asoc_ok == 0) { if (net->ro._s_addr && (net->ro._s_addr->localifa_flags & (SCTP_BEING_DELETED | SCTP_ADDR_IFA_UNUSEABLE))) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; - if (ro->ro_rt) { - RTFREE(ro->ro_rt); - ro->ro_rt = NULL; - } + RO_NHFREE(ro); } if (net->src_addr_selected == 0) { /* Cache the source address */ net->ro._s_addr = sctp_source_address_selection(inp, stcb, ro, net, 0, vrf_id); net->src_addr_selected = 1; } if (net->ro._s_addr == NULL) { /* No route to host */ net->src_addr_selected = 0; sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } ip->ip_src = net->ro._s_addr->address.sin.sin_addr; } else { if (over_addr == NULL) { struct sctp_ifa *_lsrc; _lsrc = sctp_source_address_selection(inp, stcb, ro, net, out_of_asoc_ok, vrf_id); if (_lsrc == NULL) { sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } ip->ip_src = _lsrc->address.sin.sin_addr; sctp_free_ifa(_lsrc); } else { ip->ip_src = over_addr->sin.sin_addr; SCTP_RTALLOC(ro, vrf_id, inp->fibnum); } } if (port) { if (htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)) == 0) { sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } udp = (struct udphdr *)((caddr_t)ip + sizeof(struct ip)); udp->uh_sport = htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)); udp->uh_dport = port; udp->uh_ulen = htons((uint16_t)(packet_length - sizeof(struct ip))); if (V_udp_cksum) { udp->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, udp->uh_ulen + htons(IPPROTO_UDP)); } else { udp->uh_sum = 0; } sctphdr = (struct sctphdr *)((caddr_t)udp + sizeof(struct udphdr)); } else { sctphdr = (struct sctphdr *)((caddr_t)ip + sizeof(struct ip)); } sctphdr->src_port = src_port; sctphdr->dest_port = dest_port; sctphdr->v_tag = v_tag; sctphdr->checksum = 0; /* * If source address selection fails and we find no * route then the ip_output should fail as well with * a NO_ROUTE_TO_HOST type error. We probably should * catch that somewhere and abort the association * right away (assuming this is an INIT being sent). */ - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { /* * src addr selection failed to find a route * (or valid source addr), so we can't get * there from here (yet)! */ sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } if (ro != &iproute) { memcpy(&iproute, ro, sizeof(*ro)); } SCTPDBG(SCTP_DEBUG_OUTPUT3, "Calling ipv4 output routine from low level src addr:%x\n", (uint32_t)(ntohl(ip->ip_src.s_addr))); SCTPDBG(SCTP_DEBUG_OUTPUT3, "Destination is %x\n", (uint32_t)(ntohl(ip->ip_dst.s_addr))); SCTPDBG(SCTP_DEBUG_OUTPUT3, "RTP route is %p through\n", - (void *)ro->ro_rt); + (void *)ro->ro_nh); if (SCTP_GET_HEADER_FOR_OUTPUT(o_pak)) { /* failed to prepend data, give up */ SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); sctp_m_freem(m); return (ENOMEM); } SCTP_ATTACH_CHAIN(o_pak, m, packet_length); if (port) { sctphdr->checksum = sctp_calculate_cksum(m, sizeof(struct ip) + sizeof(struct udphdr)); SCTP_STAT_INCR(sctps_sendswcrc); if (V_udp_cksum) { SCTP_ENABLE_UDP_CSUM(o_pak); } } else { m->m_pkthdr.csum_flags = CSUM_SCTP; m->m_pkthdr.csum_data = offsetof(struct sctphdr, checksum); SCTP_STAT_INCR(sctps_sendhwcrc); } #ifdef SCTP_PACKET_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LAST_PACKET_TRACING) sctp_packet_log(o_pak); #endif /* send it out. table id is taken from stcb */ #if defined(__APPLE__) || defined(SCTP_SO_LOCK_TESTING) if ((SCTP_BASE_SYSCTL(sctp_output_unlocked)) && (so_locked)) { so = SCTP_INP_SO(inp); SCTP_SOCKET_UNLOCK(so, 0); } #endif SCTP_PROBE5(send, NULL, stcb, ip, stcb, sctphdr); SCTP_IP_OUTPUT(ret, o_pak, ro, stcb, vrf_id); #if defined(__APPLE__) || defined(SCTP_SO_LOCK_TESTING) if ((SCTP_BASE_SYSCTL(sctp_output_unlocked)) && (so_locked)) { atomic_add_int(&stcb->asoc.refcnt, 1); SCTP_TCB_UNLOCK(stcb); SCTP_SOCKET_LOCK(so, 0); SCTP_TCB_LOCK(stcb); atomic_subtract_int(&stcb->asoc.refcnt, 1); } #endif if (port) { UDPSTAT_INC(udps_opackets); } SCTP_STAT_INCR(sctps_sendpackets); SCTP_STAT_INCR_COUNTER64(sctps_outpackets); if (ret) SCTP_STAT_INCR(sctps_senderrors); SCTPDBG(SCTP_DEBUG_OUTPUT3, "IP output returns %d\n", ret); if (net == NULL) { /* free tempy routes */ - RO_RTFREE(ro); + RO_NHFREE(ro); } else { - if ((ro->ro_rt != NULL) && (net->ro._s_addr) && + if ((ro->ro_nh != NULL) && (net->ro._s_addr) && ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0)) { uint32_t mtu; - mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_rt); + mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_nh); if (mtu > 0) { if (net->port) { mtu -= sizeof(struct udphdr); } if (mtu < net->mtu) { if ((stcb != NULL) && (stcb->asoc.smallest_mtu > mtu)) { sctp_mtu_size_reset(inp, &stcb->asoc, mtu); } net->mtu = mtu; } } - } else if (ro->ro_rt == NULL) { + } else if (ro->ro_nh == NULL) { /* route was freed */ if (net->ro._s_addr && net->src_addr_selected) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; } net->src_addr_selected = 0; } } return (ret); } #endif #ifdef INET6 case AF_INET6: { uint32_t flowlabel, flowinfo; struct ip6_hdr *ip6h; struct route_in6 ip6route; struct ifnet *ifp; struct sockaddr_in6 *sin6, tmp, *lsa6, lsa6_tmp; int prev_scope = 0; struct sockaddr_in6 lsa6_storage; int error; u_short prev_port = 0; int len; if (net) { flowlabel = net->flowlabel; } else if (stcb) { flowlabel = stcb->asoc.default_flowlabel; } else { flowlabel = inp->sctp_ep.default_flowlabel; } if (flowlabel == 0) { /* * This means especially, that it is not set * at the SCTP layer. So use the value from * the IP layer. */ flowlabel = ntohl(((struct inpcb *)inp)->inp_flow); } flowlabel &= 0x000fffff; len = SCTP_MIN_OVERHEAD; if (port) { len += sizeof(struct udphdr); } newm = sctp_get_mbuf_for_msg(len, 1, M_NOWAIT, 1, MT_DATA); if (newm == NULL) { sctp_m_freem(m); SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_ALIGN_TO_END(newm, len); SCTP_BUF_LEN(newm) = len; SCTP_BUF_NEXT(newm) = m; m = newm; if (net != NULL) { m->m_pkthdr.flowid = net->flowid; M_HASHTYPE_SET(m, net->flowtype); } else { m->m_pkthdr.flowid = mflowid; M_HASHTYPE_SET(m, mflowtype); } packet_length = sctp_calculate_len(m); ip6h = mtod(m, struct ip6_hdr *); /* protect *sin6 from overwrite */ sin6 = (struct sockaddr_in6 *)to; tmp = *sin6; sin6 = &tmp; /* KAME hack: embed scopeid */ if (sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)) != 0) { SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); sctp_m_freem(m); return (EINVAL); } if (net == NULL) { memset(&ip6route, 0, sizeof(ip6route)); ro = (sctp_route_t *)&ip6route; memcpy(&ro->ro_dst, sin6, sin6->sin6_len); } else { ro = (sctp_route_t *)&net->ro; } /* * We assume here that inp_flow is in host byte * order within the TCB! */ if (tos_value == 0) { /* * This means especially, that it is not set * at the SCTP layer. So use the value from * the IP layer. */ tos_value = (ntohl(((struct inpcb *)inp)->inp_flow) >> 20) & 0xff; } tos_value &= 0xfc; if (ecn_ok) { tos_value |= sctp_get_ect(stcb); } flowinfo = 0x06; flowinfo <<= 8; flowinfo |= tos_value; flowinfo <<= 20; flowinfo |= flowlabel; ip6h->ip6_flow = htonl(flowinfo); if (port) { ip6h->ip6_nxt = IPPROTO_UDP; } else { ip6h->ip6_nxt = IPPROTO_SCTP; } ip6h->ip6_plen = htons((uint16_t)(packet_length - sizeof(struct ip6_hdr))); ip6h->ip6_dst = sin6->sin6_addr; /* * Add SRC address selection here: we can only reuse * to a limited degree the kame src-addr-sel, since * we can try their selection but it may not be * bound. */ memset(&lsa6_tmp, 0, sizeof(lsa6_tmp)); lsa6_tmp.sin6_family = AF_INET6; lsa6_tmp.sin6_len = sizeof(lsa6_tmp); lsa6 = &lsa6_tmp; if (net && out_of_asoc_ok == 0) { if (net->ro._s_addr && (net->ro._s_addr->localifa_flags & (SCTP_BEING_DELETED | SCTP_ADDR_IFA_UNUSEABLE))) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; - if (ro->ro_rt) { - RTFREE(ro->ro_rt); - ro->ro_rt = NULL; - } + RO_NHFREE(ro); } if (net->src_addr_selected == 0) { sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; /* KAME hack: embed scopeid */ if (sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)) != 0) { SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); sctp_m_freem(m); return (EINVAL); } /* Cache the source address */ net->ro._s_addr = sctp_source_address_selection(inp, stcb, ro, net, 0, vrf_id); (void)sa6_recoverscope(sin6); net->src_addr_selected = 1; } if (net->ro._s_addr == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "V6:No route to host\n"); net->src_addr_selected = 0; sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } lsa6->sin6_addr = net->ro._s_addr->address.sin6.sin6_addr; } else { sin6 = (struct sockaddr_in6 *)&ro->ro_dst; /* KAME hack: embed scopeid */ if (sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)) != 0) { SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); sctp_m_freem(m); return (EINVAL); } if (over_addr == NULL) { struct sctp_ifa *_lsrc; _lsrc = sctp_source_address_selection(inp, stcb, ro, net, out_of_asoc_ok, vrf_id); if (_lsrc == NULL) { sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } lsa6->sin6_addr = _lsrc->address.sin6.sin6_addr; sctp_free_ifa(_lsrc); } else { lsa6->sin6_addr = over_addr->sin6.sin6_addr; SCTP_RTALLOC(ro, vrf_id, inp->fibnum); } (void)sa6_recoverscope(sin6); } lsa6->sin6_port = inp->sctp_lport; - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { /* * src addr selection failed to find a route * (or valid source addr), so we can't get * there from here! */ sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } /* * XXX: sa6 may not have a valid sin6_scope_id in * the non-SCOPEDROUTING case. */ memset(&lsa6_storage, 0, sizeof(lsa6_storage)); lsa6_storage.sin6_family = AF_INET6; lsa6_storage.sin6_len = sizeof(lsa6_storage); lsa6_storage.sin6_addr = lsa6->sin6_addr; if ((error = sa6_recoverscope(&lsa6_storage)) != 0) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "recover scope fails error %d\n", error); sctp_m_freem(m); return (error); } /* XXX */ lsa6_storage.sin6_addr = lsa6->sin6_addr; lsa6_storage.sin6_port = inp->sctp_lport; lsa6 = &lsa6_storage; ip6h->ip6_src = lsa6->sin6_addr; if (port) { if (htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)) == 0) { sctp_handle_no_route(stcb, net, so_locked); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EHOSTUNREACH); sctp_m_freem(m); return (EHOSTUNREACH); } udp = (struct udphdr *)((caddr_t)ip6h + sizeof(struct ip6_hdr)); udp->uh_sport = htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)); udp->uh_dport = port; udp->uh_ulen = htons((uint16_t)(packet_length - sizeof(struct ip6_hdr))); udp->uh_sum = 0; sctphdr = (struct sctphdr *)((caddr_t)udp + sizeof(struct udphdr)); } else { sctphdr = (struct sctphdr *)((caddr_t)ip6h + sizeof(struct ip6_hdr)); } sctphdr->src_port = src_port; sctphdr->dest_port = dest_port; sctphdr->v_tag = v_tag; sctphdr->checksum = 0; /* * We set the hop limit now since there is a good * chance that our ro pointer is now filled */ ip6h->ip6_hlim = SCTP_GET_HLIM(inp, ro); ifp = SCTP_GET_IFN_VOID_FROM_ROUTE(ro); #ifdef SCTP_DEBUG /* Copy to be sure something bad is not happening */ sin6->sin6_addr = ip6h->ip6_dst; lsa6->sin6_addr = ip6h->ip6_src; #endif SCTPDBG(SCTP_DEBUG_OUTPUT3, "Calling ipv6 output routine from low level\n"); SCTPDBG(SCTP_DEBUG_OUTPUT3, "src: "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT3, (struct sockaddr *)lsa6); SCTPDBG(SCTP_DEBUG_OUTPUT3, "dst: "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT3, (struct sockaddr *)sin6); if (net) { sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; /* * preserve the port and scope for link * local send */ prev_scope = sin6->sin6_scope_id; prev_port = sin6->sin6_port; } if (SCTP_GET_HEADER_FOR_OUTPUT(o_pak)) { /* failed to prepend data, give up */ sctp_m_freem(m); SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_ATTACH_CHAIN(o_pak, m, packet_length); if (port) { sctphdr->checksum = sctp_calculate_cksum(m, sizeof(struct ip6_hdr) + sizeof(struct udphdr)); SCTP_STAT_INCR(sctps_sendswcrc); if ((udp->uh_sum = in6_cksum(o_pak, IPPROTO_UDP, sizeof(struct ip6_hdr), packet_length - sizeof(struct ip6_hdr))) == 0) { udp->uh_sum = 0xffff; } } else { m->m_pkthdr.csum_flags = CSUM_SCTP_IPV6; m->m_pkthdr.csum_data = offsetof(struct sctphdr, checksum); SCTP_STAT_INCR(sctps_sendhwcrc); } /* send it out. table id is taken from stcb */ #if defined(__APPLE__) || defined(SCTP_SO_LOCK_TESTING) if ((SCTP_BASE_SYSCTL(sctp_output_unlocked)) && (so_locked)) { so = SCTP_INP_SO(inp); SCTP_SOCKET_UNLOCK(so, 0); } #endif #ifdef SCTP_PACKET_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LAST_PACKET_TRACING) sctp_packet_log(o_pak); #endif SCTP_PROBE5(send, NULL, stcb, ip6h, stcb, sctphdr); SCTP_IP6_OUTPUT(ret, o_pak, (struct route_in6 *)ro, &ifp, stcb, vrf_id); #if defined(__APPLE__) || defined(SCTP_SO_LOCK_TESTING) if ((SCTP_BASE_SYSCTL(sctp_output_unlocked)) && (so_locked)) { atomic_add_int(&stcb->asoc.refcnt, 1); SCTP_TCB_UNLOCK(stcb); SCTP_SOCKET_LOCK(so, 0); SCTP_TCB_LOCK(stcb); atomic_subtract_int(&stcb->asoc.refcnt, 1); } #endif if (net) { /* for link local this must be done */ sin6->sin6_scope_id = prev_scope; sin6->sin6_port = prev_port; } SCTPDBG(SCTP_DEBUG_OUTPUT3, "return from send is %d\n", ret); if (port) { UDPSTAT_INC(udps_opackets); } SCTP_STAT_INCR(sctps_sendpackets); SCTP_STAT_INCR_COUNTER64(sctps_outpackets); if (ret) { SCTP_STAT_INCR(sctps_senderrors); } if (net == NULL) { /* Now if we had a temp route free it */ - RO_RTFREE(ro); + RO_NHFREE(ro); } else { /* * PMTU check versus smallest asoc MTU goes * here */ - if (ro->ro_rt == NULL) { + if (ro->ro_nh == NULL) { /* Route was freed */ if (net->ro._s_addr && net->src_addr_selected) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; } net->src_addr_selected = 0; } - if ((ro->ro_rt != NULL) && (net->ro._s_addr) && + if ((ro->ro_nh != NULL) && (net->ro._s_addr) && ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0)) { uint32_t mtu; - mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_rt); + mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_nh); if (mtu > 0) { if (net->port) { mtu -= sizeof(struct udphdr); } if (mtu < net->mtu) { if ((stcb != NULL) && (stcb->asoc.smallest_mtu > mtu)) { sctp_mtu_size_reset(inp, &stcb->asoc, mtu); } net->mtu = mtu; } } } else if (ifp) { if (ND_IFINFO(ifp)->linkmtu && (stcb->asoc.smallest_mtu > ND_IFINFO(ifp)->linkmtu)) { sctp_mtu_size_reset(inp, &stcb->asoc, ND_IFINFO(ifp)->linkmtu); } } } return (ret); } #endif default: SCTPDBG(SCTP_DEBUG_OUTPUT1, "Unknown protocol (TSNH) type %d\n", ((struct sockaddr *)to)->sa_family); sctp_m_freem(m); SCTP_LTRACE_ERR_RET_PKT(m, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EFAULT); return (EFAULT); } } void sctp_send_initiate(struct sctp_inpcb *inp, struct sctp_tcb *stcb, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { struct mbuf *m, *m_last; struct sctp_nets *net; struct sctp_init_chunk *init; struct sctp_supported_addr_param *sup_addr; struct sctp_adaptation_layer_indication *ali; struct sctp_supported_chunk_types_param *pr_supported; struct sctp_paramhdr *ph; int cnt_inits_to = 0; int error; uint16_t num_ext, chunk_len, padding_len, parameter_len; /* INIT's always go to the primary (and usually ONLY address) */ net = stcb->asoc.primary_destination; if (net == NULL) { net = TAILQ_FIRST(&stcb->asoc.nets); if (net == NULL) { /* TSNH */ return; } /* we confirm any address we send an INIT to */ net->dest_state &= ~SCTP_ADDR_UNCONFIRMED; (void)sctp_set_primary_addr(stcb, NULL, net); } else { /* we confirm any address we send an INIT to */ net->dest_state &= ~SCTP_ADDR_UNCONFIRMED; } SCTPDBG(SCTP_DEBUG_OUTPUT4, "Sending INIT\n"); #ifdef INET6 if (net->ro._l_addr.sa.sa_family == AF_INET6) { /* * special hook, if we are sending to link local it will not * show up in our private address count. */ if (IN6_IS_ADDR_LINKLOCAL(&net->ro._l_addr.sin6.sin6_addr)) cnt_inits_to = 1; } #endif if (SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) { /* This case should not happen */ SCTPDBG(SCTP_DEBUG_OUTPUT4, "Sending INIT - failed timer?\n"); return; } /* start the INIT timer */ sctp_timer_start(SCTP_TIMER_TYPE_INIT, inp, stcb, net); m = sctp_get_mbuf_for_msg(MCLBYTES, 1, M_NOWAIT, 1, MT_DATA); if (m == NULL) { /* No memory, INIT timer will re-attempt. */ SCTPDBG(SCTP_DEBUG_OUTPUT4, "Sending INIT - mbuf?\n"); return; } chunk_len = (uint16_t)sizeof(struct sctp_init_chunk); padding_len = 0; /* Now lets put the chunk header in place */ init = mtod(m, struct sctp_init_chunk *); /* now the chunk header */ init->ch.chunk_type = SCTP_INITIATION; init->ch.chunk_flags = 0; /* fill in later from mbuf we build */ init->ch.chunk_length = 0; /* place in my tag */ init->init.initiate_tag = htonl(stcb->asoc.my_vtag); /* set up some of the credits. */ init->init.a_rwnd = htonl(max(inp->sctp_socket ? SCTP_SB_LIMIT_RCV(inp->sctp_socket) : 0, SCTP_MINIMAL_RWND)); init->init.num_outbound_streams = htons(stcb->asoc.pre_open_streams); init->init.num_inbound_streams = htons(stcb->asoc.max_inbound_streams); init->init.initial_tsn = htonl(stcb->asoc.init_seq_number); /* Adaptation layer indication parameter */ if (inp->sctp_ep.adaptation_layer_indicator_provided) { parameter_len = (uint16_t)sizeof(struct sctp_adaptation_layer_indication); ali = (struct sctp_adaptation_layer_indication *)(mtod(m, caddr_t)+chunk_len); ali->ph.param_type = htons(SCTP_ULP_ADAPTATION); ali->ph.param_length = htons(parameter_len); ali->indication = htonl(inp->sctp_ep.adaptation_layer_indicator); chunk_len += parameter_len; } /* ECN parameter */ if (stcb->asoc.ecn_supported == 1) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_ECN_CAPABLE); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* PR-SCTP supported parameter */ if (stcb->asoc.prsctp_supported == 1) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_PRSCTP_SUPPORTED); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* Add NAT friendly parameter. */ if (SCTP_BASE_SYSCTL(sctp_inits_include_nat_friendly)) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_HAS_NAT_SUPPORT); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* And now tell the peer which extensions we support */ num_ext = 0; pr_supported = (struct sctp_supported_chunk_types_param *)(mtod(m, caddr_t)+chunk_len); if (stcb->asoc.prsctp_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_FORWARD_CUM_TSN; if (stcb->asoc.idata_supported) { pr_supported->chunk_types[num_ext++] = SCTP_IFORWARD_CUM_TSN; } } if (stcb->asoc.auth_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_AUTHENTICATION; } if (stcb->asoc.asconf_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_ASCONF; pr_supported->chunk_types[num_ext++] = SCTP_ASCONF_ACK; } if (stcb->asoc.reconfig_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_STREAM_RESET; } if (stcb->asoc.idata_supported) { pr_supported->chunk_types[num_ext++] = SCTP_IDATA; } if (stcb->asoc.nrsack_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_NR_SELECTIVE_ACK; } if (stcb->asoc.pktdrop_supported == 1) { pr_supported->chunk_types[num_ext++] = SCTP_PACKET_DROPPED; } if (num_ext > 0) { parameter_len = (uint16_t)sizeof(struct sctp_supported_chunk_types_param) + num_ext; pr_supported->ph.param_type = htons(SCTP_SUPPORTED_CHUNK_EXT); pr_supported->ph.param_length = htons(parameter_len); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } /* add authentication parameters */ if (stcb->asoc.auth_supported) { /* attach RANDOM parameter, if available */ if (stcb->asoc.authinfo.random != NULL) { struct sctp_auth_random *randp; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } randp = (struct sctp_auth_random *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)sizeof(struct sctp_auth_random) + stcb->asoc.authinfo.random_len; /* random key already contains the header */ memcpy(randp, stcb->asoc.authinfo.random->key, parameter_len); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } /* add HMAC_ALGO parameter */ if (stcb->asoc.local_hmacs != NULL) { struct sctp_auth_hmac_algo *hmacs; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } hmacs = (struct sctp_auth_hmac_algo *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)(sizeof(struct sctp_auth_hmac_algo) + stcb->asoc.local_hmacs->num_algo * sizeof(uint16_t)); hmacs->ph.param_type = htons(SCTP_HMAC_LIST); hmacs->ph.param_length = htons(parameter_len); sctp_serialize_hmaclist(stcb->asoc.local_hmacs, (uint8_t *)hmacs->hmac_ids); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } /* add CHUNKS parameter */ if (stcb->asoc.local_auth_chunks != NULL) { struct sctp_auth_chunk_list *chunks; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } chunks = (struct sctp_auth_chunk_list *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)(sizeof(struct sctp_auth_chunk_list) + sctp_auth_get_chklist_size(stcb->asoc.local_auth_chunks)); chunks->ph.param_type = htons(SCTP_CHUNK_LIST); chunks->ph.param_length = htons(parameter_len); sctp_serialize_auth_chunks(stcb->asoc.local_auth_chunks, chunks->chunk_types); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } } /* now any cookie time extensions */ if (stcb->asoc.cookie_preserve_req) { struct sctp_cookie_perserve_param *cookie_preserve; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } parameter_len = (uint16_t)sizeof(struct sctp_cookie_perserve_param); cookie_preserve = (struct sctp_cookie_perserve_param *)(mtod(m, caddr_t)+chunk_len); cookie_preserve->ph.param_type = htons(SCTP_COOKIE_PRESERVE); cookie_preserve->ph.param_length = htons(parameter_len); cookie_preserve->time = htonl(stcb->asoc.cookie_preserve_req); stcb->asoc.cookie_preserve_req = 0; chunk_len += parameter_len; } if (stcb->asoc.scope.ipv4_addr_legal || stcb->asoc.scope.ipv6_addr_legal) { uint8_t i; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); if (stcb->asoc.scope.ipv4_addr_legal) { parameter_len += (uint16_t)sizeof(uint16_t); } if (stcb->asoc.scope.ipv6_addr_legal) { parameter_len += (uint16_t)sizeof(uint16_t); } sup_addr = (struct sctp_supported_addr_param *)(mtod(m, caddr_t)+chunk_len); sup_addr->ph.param_type = htons(SCTP_SUPPORTED_ADDRTYPE); sup_addr->ph.param_length = htons(parameter_len); i = 0; if (stcb->asoc.scope.ipv4_addr_legal) { sup_addr->addr_type[i++] = htons(SCTP_IPV4_ADDRESS); } if (stcb->asoc.scope.ipv6_addr_legal) { sup_addr->addr_type[i++] = htons(SCTP_IPV6_ADDRESS); } padding_len = 4 - 2 * i; chunk_len += parameter_len; } SCTP_BUF_LEN(m) = chunk_len; /* now the addresses */ /* * To optimize this we could put the scoping stuff into a structure * and remove the individual uint8's from the assoc structure. Then * we could just sifa in the address within the stcb. But for now * this is a quick hack to get the address stuff teased apart. */ m_last = sctp_add_addresses_to_i_ia(inp, stcb, &stcb->asoc.scope, m, cnt_inits_to, &padding_len, &chunk_len); init->ch.chunk_length = htons(chunk_len); if (padding_len > 0) { if (sctp_add_pad_tombuf(m_last, padding_len) == NULL) { sctp_m_freem(m); return; } } SCTPDBG(SCTP_DEBUG_OUTPUT4, "Sending INIT - calls lowlevel_output\n"); if ((error = sctp_lowlevel_chunk_output(inp, stcb, net, (struct sockaddr *)&net->ro._l_addr, m, 0, NULL, 0, 0, 0, 0, inp->sctp_lport, stcb->rport, htonl(0), net->port, NULL, 0, 0, so_locked))) { SCTPDBG(SCTP_DEBUG_OUTPUT4, "Gak send error %d\n", error); if (error == ENOBUFS) { stcb->asoc.ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } } else { stcb->asoc.ifp_had_enobuf = 0; } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); (void)SCTP_GETTIME_TIMEVAL(&net->last_sent_time); } struct mbuf * sctp_arethere_unrecognized_parameters(struct mbuf *in_initpkt, int param_offset, int *abort_processing, struct sctp_chunkhdr *cp, int *nat_friendly, int *cookie_found) { /* * Given a mbuf containing an INIT or INIT-ACK with the param_offset * being equal to the beginning of the params i.e. (iphlen + * sizeof(struct sctp_init_msg) parse through the parameters to the * end of the mbuf verifying that all parameters are known. * * For unknown parameters build and return a mbuf with * UNRECOGNIZED_PARAMETER errors. If the flags indicate to stop * processing this chunk stop, and set *abort_processing to 1. * * By having param_offset be pre-set to where parameters begin it is * hoped that this routine may be reused in the future by new * features. */ struct sctp_paramhdr *phdr, params; struct mbuf *mat, *m_tmp, *op_err, *op_err_last; int at, limit, pad_needed; uint16_t ptype, plen, padded_size; *abort_processing = 0; if (cookie_found != NULL) { *cookie_found = 0; } mat = in_initpkt; limit = ntohs(cp->chunk_length) - sizeof(struct sctp_init_chunk); at = param_offset; op_err = NULL; op_err_last = NULL; pad_needed = 0; SCTPDBG(SCTP_DEBUG_OUTPUT1, "Check for unrecognized param's\n"); phdr = sctp_get_next_param(mat, at, ¶ms, sizeof(params)); while ((phdr != NULL) && ((size_t)limit >= sizeof(struct sctp_paramhdr))) { ptype = ntohs(phdr->param_type); plen = ntohs(phdr->param_length); if ((plen > limit) || (plen < sizeof(struct sctp_paramhdr))) { /* wacked parameter */ SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error %d\n", plen); goto invalid_size; } limit -= SCTP_SIZE32(plen); /*- * All parameters for all chunks that we know/understand are * listed here. We process them other places and make * appropriate stop actions per the upper bits. However this * is the generic routine processor's can call to get back * an operr.. to either incorporate (init-ack) or send. */ padded_size = SCTP_SIZE32(plen); switch (ptype) { /* Param's with variable size */ case SCTP_HEARTBEAT_INFO: case SCTP_UNRECOG_PARAM: case SCTP_ERROR_CAUSE_IND: /* ok skip fwd */ at += padded_size; break; case SCTP_STATE_COOKIE: if (cookie_found != NULL) { *cookie_found = 1; } at += padded_size; break; /* Param's with variable size within a range */ case SCTP_CHUNK_LIST: case SCTP_SUPPORTED_CHUNK_EXT: if (padded_size > (sizeof(struct sctp_supported_chunk_types_param) + (sizeof(uint8_t) * SCTP_MAX_SUPPORTED_EXT))) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error chklist %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_SUPPORTED_ADDRTYPE: if (padded_size > SCTP_MAX_ADDR_PARAMS_SIZE) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error supaddrtype %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_RANDOM: if (padded_size > (sizeof(struct sctp_auth_random) + SCTP_RANDOM_MAX_SIZE)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error random %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_SET_PRIM_ADDR: case SCTP_DEL_IP_ADDRESS: case SCTP_ADD_IP_ADDRESS: if ((padded_size != sizeof(struct sctp_asconf_addrv4_param)) && (padded_size != sizeof(struct sctp_asconf_addr_param))) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error setprim %d\n", plen); goto invalid_size; } at += padded_size; break; /* Param's with a fixed size */ case SCTP_IPV4_ADDRESS: if (padded_size != sizeof(struct sctp_ipv4addr_param)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error ipv4 addr %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_IPV6_ADDRESS: if (padded_size != sizeof(struct sctp_ipv6addr_param)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error ipv6 addr %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_COOKIE_PRESERVE: if (padded_size != sizeof(struct sctp_cookie_perserve_param)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error cookie-preserve %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_HAS_NAT_SUPPORT: *nat_friendly = 1; /* fall through */ case SCTP_PRSCTP_SUPPORTED: if (padded_size != sizeof(struct sctp_paramhdr)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error prsctp/nat support %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_ECN_CAPABLE: if (padded_size != sizeof(struct sctp_paramhdr)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error ecn %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_ULP_ADAPTATION: if (padded_size != sizeof(struct sctp_adaptation_layer_indication)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error adapatation %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_SUCCESS_REPORT: if (padded_size != sizeof(struct sctp_asconf_paramhdr)) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Invalid size - error success %d\n", plen); goto invalid_size; } at += padded_size; break; case SCTP_HOSTNAME_ADDRESS: { /* Hostname parameters are deprecated. */ struct sctp_gen_error_cause *cause; int l_len; SCTPDBG(SCTP_DEBUG_OUTPUT1, "Can't handle hostname addresses.. abort processing\n"); *abort_processing = 1; sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; #ifdef INET6 l_len = SCTP_MIN_OVERHEAD; #else l_len = SCTP_MIN_V4_OVERHEAD; #endif l_len += sizeof(struct sctp_chunkhdr); l_len += sizeof(struct sctp_gen_error_cause); op_err = sctp_get_mbuf_for_msg(l_len, 0, M_NOWAIT, 1, MT_DATA); if (op_err != NULL) { /* * Pre-reserve space for IP, SCTP, * and chunk header. */ #ifdef INET6 SCTP_BUF_RESV_UF(op_err, sizeof(struct ip6_hdr)); #else SCTP_BUF_RESV_UF(op_err, sizeof(struct ip)); #endif SCTP_BUF_RESV_UF(op_err, sizeof(struct sctphdr)); SCTP_BUF_RESV_UF(op_err, sizeof(struct sctp_chunkhdr)); SCTP_BUF_LEN(op_err) = sizeof(struct sctp_gen_error_cause); cause = mtod(op_err, struct sctp_gen_error_cause *); cause->code = htons(SCTP_CAUSE_UNRESOLVABLE_ADDR); cause->length = htons((uint16_t)(sizeof(struct sctp_gen_error_cause) + plen)); SCTP_BUF_NEXT(op_err) = SCTP_M_COPYM(mat, at, plen, M_NOWAIT); if (SCTP_BUF_NEXT(op_err) == NULL) { sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; } } return (op_err); break; } default: /* * we do not recognize the parameter figure out what * we do. */ SCTPDBG(SCTP_DEBUG_OUTPUT1, "Hit default param %x\n", ptype); if ((ptype & 0x4000) == 0x4000) { /* Report bit is set?? */ SCTPDBG(SCTP_DEBUG_OUTPUT1, "report op err\n"); if (op_err == NULL) { int l_len; /* Ok need to try to get an mbuf */ #ifdef INET6 l_len = SCTP_MIN_OVERHEAD; #else l_len = SCTP_MIN_V4_OVERHEAD; #endif l_len += sizeof(struct sctp_chunkhdr); l_len += sizeof(struct sctp_paramhdr); op_err = sctp_get_mbuf_for_msg(l_len, 0, M_NOWAIT, 1, MT_DATA); if (op_err) { SCTP_BUF_LEN(op_err) = 0; #ifdef INET6 SCTP_BUF_RESV_UF(op_err, sizeof(struct ip6_hdr)); #else SCTP_BUF_RESV_UF(op_err, sizeof(struct ip)); #endif SCTP_BUF_RESV_UF(op_err, sizeof(struct sctphdr)); SCTP_BUF_RESV_UF(op_err, sizeof(struct sctp_chunkhdr)); op_err_last = op_err; } } if (op_err != NULL) { /* If we have space */ struct sctp_paramhdr *param; if (pad_needed > 0) { op_err_last = sctp_add_pad_tombuf(op_err_last, pad_needed); } if (op_err_last == NULL) { sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; goto more_processing; } if (M_TRAILINGSPACE(op_err_last) < (int)sizeof(struct sctp_paramhdr)) { m_tmp = sctp_get_mbuf_for_msg(sizeof(struct sctp_paramhdr), 0, M_NOWAIT, 1, MT_DATA); if (m_tmp == NULL) { sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; goto more_processing; } SCTP_BUF_LEN(m_tmp) = 0; SCTP_BUF_NEXT(m_tmp) = NULL; SCTP_BUF_NEXT(op_err_last) = m_tmp; op_err_last = m_tmp; } param = (struct sctp_paramhdr *)(mtod(op_err_last, caddr_t)+SCTP_BUF_LEN(op_err_last)); param->param_type = htons(SCTP_UNRECOG_PARAM); param->param_length = htons((uint16_t)sizeof(struct sctp_paramhdr) + plen); SCTP_BUF_LEN(op_err_last) += sizeof(struct sctp_paramhdr); SCTP_BUF_NEXT(op_err_last) = SCTP_M_COPYM(mat, at, plen, M_NOWAIT); if (SCTP_BUF_NEXT(op_err_last) == NULL) { sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; goto more_processing; } else { while (SCTP_BUF_NEXT(op_err_last) != NULL) { op_err_last = SCTP_BUF_NEXT(op_err_last); } } if (plen % 4 != 0) { pad_needed = 4 - (plen % 4); } else { pad_needed = 0; } } } more_processing: if ((ptype & 0x8000) == 0x0000) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "stop proc\n"); return (op_err); } else { /* skip this chunk and continue processing */ SCTPDBG(SCTP_DEBUG_OUTPUT1, "move on\n"); at += SCTP_SIZE32(plen); } break; } phdr = sctp_get_next_param(mat, at, ¶ms, sizeof(params)); } return (op_err); invalid_size: SCTPDBG(SCTP_DEBUG_OUTPUT1, "abort flag set\n"); *abort_processing = 1; sctp_m_freem(op_err); op_err = NULL; op_err_last = NULL; if (phdr != NULL) { struct sctp_paramhdr *param; int l_len; #ifdef INET6 l_len = SCTP_MIN_OVERHEAD; #else l_len = SCTP_MIN_V4_OVERHEAD; #endif l_len += sizeof(struct sctp_chunkhdr); l_len += (2 * sizeof(struct sctp_paramhdr)); op_err = sctp_get_mbuf_for_msg(l_len, 0, M_NOWAIT, 1, MT_DATA); if (op_err) { SCTP_BUF_LEN(op_err) = 0; #ifdef INET6 SCTP_BUF_RESV_UF(op_err, sizeof(struct ip6_hdr)); #else SCTP_BUF_RESV_UF(op_err, sizeof(struct ip)); #endif SCTP_BUF_RESV_UF(op_err, sizeof(struct sctphdr)); SCTP_BUF_RESV_UF(op_err, sizeof(struct sctp_chunkhdr)); SCTP_BUF_LEN(op_err) = 2 * sizeof(struct sctp_paramhdr); param = mtod(op_err, struct sctp_paramhdr *); param->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION); param->param_length = htons(2 * sizeof(struct sctp_paramhdr)); param++; param->param_type = htons(ptype); param->param_length = htons(plen); } } return (op_err); } static int sctp_are_there_new_addresses(struct sctp_association *asoc, struct mbuf *in_initpkt, int offset, struct sockaddr *src) { /* * Given a INIT packet, look through the packet to verify that there * are NO new addresses. As we go through the parameters add reports * of any un-understood parameters that require an error. Also we * must return (1) to drop the packet if we see a un-understood * parameter that tells us to drop the chunk. */ struct sockaddr *sa_touse; struct sockaddr *sa; struct sctp_paramhdr *phdr, params; uint16_t ptype, plen; uint8_t fnd; struct sctp_nets *net; int check_src; #ifdef INET struct sockaddr_in sin4, *sa4; #endif #ifdef INET6 struct sockaddr_in6 sin6, *sa6; #endif #ifdef INET memset(&sin4, 0, sizeof(sin4)); sin4.sin_family = AF_INET; sin4.sin_len = sizeof(sin4); #endif #ifdef INET6 memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(sin6); #endif /* First what about the src address of the pkt ? */ check_src = 0; switch (src->sa_family) { #ifdef INET case AF_INET: if (asoc->scope.ipv4_addr_legal) { check_src = 1; } break; #endif #ifdef INET6 case AF_INET6: if (asoc->scope.ipv6_addr_legal) { check_src = 1; } break; #endif default: /* TSNH */ break; } if (check_src) { fnd = 0; TAILQ_FOREACH(net, &asoc->nets, sctp_next) { sa = (struct sockaddr *)&net->ro._l_addr; if (sa->sa_family == src->sa_family) { #ifdef INET if (sa->sa_family == AF_INET) { struct sockaddr_in *src4; sa4 = (struct sockaddr_in *)sa; src4 = (struct sockaddr_in *)src; if (sa4->sin_addr.s_addr == src4->sin_addr.s_addr) { fnd = 1; break; } } #endif #ifdef INET6 if (sa->sa_family == AF_INET6) { struct sockaddr_in6 *src6; sa6 = (struct sockaddr_in6 *)sa; src6 = (struct sockaddr_in6 *)src; if (SCTP6_ARE_ADDR_EQUAL(sa6, src6)) { fnd = 1; break; } } #endif } } if (fnd == 0) { /* New address added! no need to look further. */ return (1); } } /* Ok so far lets munge through the rest of the packet */ offset += sizeof(struct sctp_init_chunk); phdr = sctp_get_next_param(in_initpkt, offset, ¶ms, sizeof(params)); while (phdr) { sa_touse = NULL; ptype = ntohs(phdr->param_type); plen = ntohs(phdr->param_length); switch (ptype) { #ifdef INET case SCTP_IPV4_ADDRESS: { struct sctp_ipv4addr_param *p4, p4_buf; if (plen != sizeof(struct sctp_ipv4addr_param)) { return (1); } phdr = sctp_get_next_param(in_initpkt, offset, (struct sctp_paramhdr *)&p4_buf, sizeof(p4_buf)); if (phdr == NULL) { return (1); } if (asoc->scope.ipv4_addr_legal) { p4 = (struct sctp_ipv4addr_param *)phdr; sin4.sin_addr.s_addr = p4->addr; sa_touse = (struct sockaddr *)&sin4; } break; } #endif #ifdef INET6 case SCTP_IPV6_ADDRESS: { struct sctp_ipv6addr_param *p6, p6_buf; if (plen != sizeof(struct sctp_ipv6addr_param)) { return (1); } phdr = sctp_get_next_param(in_initpkt, offset, (struct sctp_paramhdr *)&p6_buf, sizeof(p6_buf)); if (phdr == NULL) { return (1); } if (asoc->scope.ipv6_addr_legal) { p6 = (struct sctp_ipv6addr_param *)phdr; memcpy((caddr_t)&sin6.sin6_addr, p6->addr, sizeof(p6->addr)); sa_touse = (struct sockaddr *)&sin6; } break; } #endif default: sa_touse = NULL; break; } if (sa_touse) { /* ok, sa_touse points to one to check */ fnd = 0; TAILQ_FOREACH(net, &asoc->nets, sctp_next) { sa = (struct sockaddr *)&net->ro._l_addr; if (sa->sa_family != sa_touse->sa_family) { continue; } #ifdef INET if (sa->sa_family == AF_INET) { sa4 = (struct sockaddr_in *)sa; if (sa4->sin_addr.s_addr == sin4.sin_addr.s_addr) { fnd = 1; break; } } #endif #ifdef INET6 if (sa->sa_family == AF_INET6) { sa6 = (struct sockaddr_in6 *)sa; if (SCTP6_ARE_ADDR_EQUAL( sa6, &sin6)) { fnd = 1; break; } } #endif } if (!fnd) { /* New addr added! no need to look further */ return (1); } } offset += SCTP_SIZE32(plen); phdr = sctp_get_next_param(in_initpkt, offset, ¶ms, sizeof(params)); } return (0); } /* * Given a MBUF chain that was sent into us containing an INIT. Build a * INIT-ACK with COOKIE and send back. We assume that the in_initpkt has done * a pullup to include IPv6/4header, SCTP header and initial part of INIT * message (i.e. the struct sctp_init_msg). */ void sctp_send_initiate_ack(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *src_net, struct mbuf *init_pkt, int iphlen, int offset, struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, struct sctp_init_chunk *init_chk, uint8_t mflowtype, uint32_t mflowid, uint32_t vrf_id, uint16_t port) { struct sctp_association *asoc; struct mbuf *m, *m_tmp, *m_last, *m_cookie, *op_err; struct sctp_init_ack_chunk *initack; struct sctp_adaptation_layer_indication *ali; struct sctp_supported_chunk_types_param *pr_supported; struct sctp_paramhdr *ph; union sctp_sockstore *over_addr; struct sctp_scoping scp; struct timeval now; #ifdef INET struct sockaddr_in *dst4 = (struct sockaddr_in *)dst; struct sockaddr_in *src4 = (struct sockaddr_in *)src; struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *dst6 = (struct sockaddr_in6 *)dst; struct sockaddr_in6 *src6 = (struct sockaddr_in6 *)src; struct sockaddr_in6 *sin6; #endif struct sockaddr *to; struct sctp_state_cookie stc; struct sctp_nets *net = NULL; uint8_t *signature = NULL; int cnt_inits_to = 0; uint16_t his_limit, i_want; int abort_flag; int nat_friendly = 0; int error; struct socket *so; uint16_t num_ext, chunk_len, padding_len, parameter_len; if (stcb) { asoc = &stcb->asoc; } else { asoc = NULL; } if ((asoc != NULL) && (SCTP_GET_STATE(stcb) != SCTP_STATE_COOKIE_WAIT)) { if (sctp_are_there_new_addresses(asoc, init_pkt, offset, src)) { /* * new addresses, out of here in non-cookie-wait * states * * Send an ABORT, without the new address error * cause. This looks no different than if no * listener was present. */ op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), "Address added"); sctp_send_abort(init_pkt, iphlen, src, dst, sh, 0, op_err, mflowtype, mflowid, inp->fibnum, vrf_id, port); return; } if (src_net != NULL && (src_net->port != port)) { /* * change of remote encapsulation port, out of here * in non-cookie-wait states * * Send an ABORT, without an specific error cause. * This looks no different than if no listener was * present. */ op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), "Remote encapsulation port changed"); sctp_send_abort(init_pkt, iphlen, src, dst, sh, 0, op_err, mflowtype, mflowid, inp->fibnum, vrf_id, port); return; } } abort_flag = 0; op_err = sctp_arethere_unrecognized_parameters(init_pkt, (offset + sizeof(struct sctp_init_chunk)), &abort_flag, (struct sctp_chunkhdr *)init_chk, &nat_friendly, NULL); if (abort_flag) { do_a_abort: if (op_err == NULL) { char msg[SCTP_DIAG_INFO_LEN]; snprintf(msg, sizeof(msg), "%s:%d at %s", __FILE__, __LINE__, __func__); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); } sctp_send_abort(init_pkt, iphlen, src, dst, sh, init_chk->init.initiate_tag, op_err, mflowtype, mflowid, inp->fibnum, vrf_id, port); return; } m = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (m == NULL) { /* No memory, INIT timer will re-attempt. */ sctp_m_freem(op_err); return; } chunk_len = (uint16_t)sizeof(struct sctp_init_ack_chunk); padding_len = 0; /* * We might not overwrite the identification[] completely and on * some platforms time_entered will contain some padding. Therefore * zero out the cookie to avoid putting uninitialized memory on the * wire. */ memset(&stc, 0, sizeof(struct sctp_state_cookie)); /* the time I built cookie */ (void)SCTP_GETTIME_TIMEVAL(&now); stc.time_entered.tv_sec = now.tv_sec; stc.time_entered.tv_usec = now.tv_usec; /* populate any tie tags */ if (asoc != NULL) { /* unlock before tag selections */ stc.tie_tag_my_vtag = asoc->my_vtag_nonce; stc.tie_tag_peer_vtag = asoc->peer_vtag_nonce; stc.cookie_life = asoc->cookie_life; net = asoc->primary_destination; } else { stc.tie_tag_my_vtag = 0; stc.tie_tag_peer_vtag = 0; /* life I will award this cookie */ stc.cookie_life = inp->sctp_ep.def_cookie_life; } /* copy in the ports for later check */ stc.myport = sh->dest_port; stc.peerport = sh->src_port; /* * If we wanted to honor cookie life extensions, we would add to * stc.cookie_life. For now we should NOT honor any extension */ stc.site_scope = stc.local_scope = stc.loopback_scope = 0; if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) { stc.ipv6_addr_legal = 1; if (SCTP_IPV6_V6ONLY(inp)) { stc.ipv4_addr_legal = 0; } else { stc.ipv4_addr_legal = 1; } } else { stc.ipv6_addr_legal = 0; stc.ipv4_addr_legal = 1; } stc.ipv4_scope = 0; if (net == NULL) { to = src; switch (dst->sa_family) { #ifdef INET case AF_INET: { /* lookup address */ stc.address[0] = src4->sin_addr.s_addr; stc.address[1] = 0; stc.address[2] = 0; stc.address[3] = 0; stc.addr_type = SCTP_IPV4_ADDRESS; /* local from address */ stc.laddress[0] = dst4->sin_addr.s_addr; stc.laddress[1] = 0; stc.laddress[2] = 0; stc.laddress[3] = 0; stc.laddr_type = SCTP_IPV4_ADDRESS; /* scope_id is only for v6 */ stc.scope_id = 0; if ((IN4_ISPRIVATE_ADDRESS(&src4->sin_addr)) || (IN4_ISPRIVATE_ADDRESS(&dst4->sin_addr))) { stc.ipv4_scope = 1; } /* Must use the address in this case */ if (sctp_is_address_on_local_host(src, vrf_id)) { stc.loopback_scope = 1; stc.ipv4_scope = 1; stc.site_scope = 1; stc.local_scope = 0; } break; } #endif #ifdef INET6 case AF_INET6: { stc.addr_type = SCTP_IPV6_ADDRESS; memcpy(&stc.address, &src6->sin6_addr, sizeof(struct in6_addr)); stc.scope_id = ntohs(in6_getscope(&src6->sin6_addr)); if (sctp_is_address_on_local_host(src, vrf_id)) { stc.loopback_scope = 1; stc.local_scope = 0; stc.site_scope = 1; stc.ipv4_scope = 1; } else if (IN6_IS_ADDR_LINKLOCAL(&src6->sin6_addr) || IN6_IS_ADDR_LINKLOCAL(&dst6->sin6_addr)) { /* * If the new destination or source * is a LINK_LOCAL we must have * common both site and local scope. * Don't set local scope though * since we must depend on the * source to be added implicitly. We * cannot assure just because we * share one link that all links are * common. */ stc.local_scope = 0; stc.site_scope = 1; stc.ipv4_scope = 1; /* * we start counting for the private * address stuff at 1. since the * link local we source from won't * show up in our scoped count. */ cnt_inits_to = 1; /* * pull out the scope_id from * incoming pkt */ } else if (IN6_IS_ADDR_SITELOCAL(&src6->sin6_addr) || IN6_IS_ADDR_SITELOCAL(&dst6->sin6_addr)) { /* * If the new destination or source * is SITE_LOCAL then we must have * site scope in common. */ stc.site_scope = 1; } memcpy(&stc.laddress, &dst6->sin6_addr, sizeof(struct in6_addr)); stc.laddr_type = SCTP_IPV6_ADDRESS; break; } #endif default: /* TSNH */ goto do_a_abort; break; } } else { /* set the scope per the existing tcb */ #ifdef INET6 struct sctp_nets *lnet; #endif stc.loopback_scope = asoc->scope.loopback_scope; stc.ipv4_scope = asoc->scope.ipv4_local_scope; stc.site_scope = asoc->scope.site_scope; stc.local_scope = asoc->scope.local_scope; #ifdef INET6 /* Why do we not consider IPv4 LL addresses? */ TAILQ_FOREACH(lnet, &asoc->nets, sctp_next) { if (lnet->ro._l_addr.sin6.sin6_family == AF_INET6) { if (IN6_IS_ADDR_LINKLOCAL(&lnet->ro._l_addr.sin6.sin6_addr)) { /* * if we have a LL address, start * counting at 1. */ cnt_inits_to = 1; } } } #endif /* use the net pointer */ to = (struct sockaddr *)&net->ro._l_addr; switch (to->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)to; stc.address[0] = sin->sin_addr.s_addr; stc.address[1] = 0; stc.address[2] = 0; stc.address[3] = 0; stc.addr_type = SCTP_IPV4_ADDRESS; if (net->src_addr_selected == 0) { /* * strange case here, the INIT should have * did the selection. */ net->ro._s_addr = sctp_source_address_selection(inp, stcb, (sctp_route_t *)&net->ro, net, 0, vrf_id); if (net->ro._s_addr == NULL) { sctp_m_freem(op_err); sctp_m_freem(m); return; } net->src_addr_selected = 1; } stc.laddress[0] = net->ro._s_addr->address.sin.sin_addr.s_addr; stc.laddress[1] = 0; stc.laddress[2] = 0; stc.laddress[3] = 0; stc.laddr_type = SCTP_IPV4_ADDRESS; /* scope_id is only for v6 */ stc.scope_id = 0; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)to; memcpy(&stc.address, &sin6->sin6_addr, sizeof(struct in6_addr)); stc.addr_type = SCTP_IPV6_ADDRESS; stc.scope_id = sin6->sin6_scope_id; if (net->src_addr_selected == 0) { /* * strange case here, the INIT should have * done the selection. */ net->ro._s_addr = sctp_source_address_selection(inp, stcb, (sctp_route_t *)&net->ro, net, 0, vrf_id); if (net->ro._s_addr == NULL) { sctp_m_freem(op_err); sctp_m_freem(m); return; } net->src_addr_selected = 1; } memcpy(&stc.laddress, &net->ro._s_addr->address.sin6.sin6_addr, sizeof(struct in6_addr)); stc.laddr_type = SCTP_IPV6_ADDRESS; break; #endif } } /* Now lets put the SCTP header in place */ initack = mtod(m, struct sctp_init_ack_chunk *); /* Save it off for quick ref */ stc.peers_vtag = ntohl(init_chk->init.initiate_tag); /* who are we */ memcpy(stc.identification, SCTP_VERSION_STRING, min(strlen(SCTP_VERSION_STRING), sizeof(stc.identification))); memset(stc.reserved, 0, SCTP_RESERVE_SPACE); /* now the chunk header */ initack->ch.chunk_type = SCTP_INITIATION_ACK; initack->ch.chunk_flags = 0; /* fill in later from mbuf we build */ initack->ch.chunk_length = 0; /* place in my tag */ if ((asoc != NULL) && ((SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_WAIT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_INUSE) || (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED))) { /* re-use the v-tags and init-seq here */ initack->init.initiate_tag = htonl(asoc->my_vtag); initack->init.initial_tsn = htonl(asoc->init_seq_number); } else { uint32_t vtag, itsn; if (asoc) { atomic_add_int(&asoc->refcnt, 1); SCTP_TCB_UNLOCK(stcb); new_tag: vtag = sctp_select_a_tag(inp, inp->sctp_lport, sh->src_port, 1); if ((asoc->peer_supports_nat) && (vtag == asoc->my_vtag)) { /* * Got a duplicate vtag on some guy behind a * nat make sure we don't use it. */ goto new_tag; } initack->init.initiate_tag = htonl(vtag); /* get a TSN to use too */ itsn = sctp_select_initial_TSN(&inp->sctp_ep); initack->init.initial_tsn = htonl(itsn); SCTP_TCB_LOCK(stcb); atomic_add_int(&asoc->refcnt, -1); } else { SCTP_INP_INCR_REF(inp); SCTP_INP_RUNLOCK(inp); vtag = sctp_select_a_tag(inp, inp->sctp_lport, sh->src_port, 1); initack->init.initiate_tag = htonl(vtag); /* get a TSN to use too */ initack->init.initial_tsn = htonl(sctp_select_initial_TSN(&inp->sctp_ep)); SCTP_INP_RLOCK(inp); SCTP_INP_DECR_REF(inp); } } /* save away my tag to */ stc.my_vtag = initack->init.initiate_tag; /* set up some of the credits. */ so = inp->sctp_socket; if (so == NULL) { /* memory problem */ sctp_m_freem(op_err); sctp_m_freem(m); return; } else { initack->init.a_rwnd = htonl(max(SCTP_SB_LIMIT_RCV(so), SCTP_MINIMAL_RWND)); } /* set what I want */ his_limit = ntohs(init_chk->init.num_inbound_streams); /* choose what I want */ if (asoc != NULL) { if (asoc->streamoutcnt > asoc->pre_open_streams) { i_want = asoc->streamoutcnt; } else { i_want = asoc->pre_open_streams; } } else { i_want = inp->sctp_ep.pre_open_stream_count; } if (his_limit < i_want) { /* I Want more :< */ initack->init.num_outbound_streams = init_chk->init.num_inbound_streams; } else { /* I can have what I want :> */ initack->init.num_outbound_streams = htons(i_want); } /* tell him his limit. */ initack->init.num_inbound_streams = htons(inp->sctp_ep.max_open_streams_intome); /* adaptation layer indication parameter */ if (inp->sctp_ep.adaptation_layer_indicator_provided) { parameter_len = (uint16_t)sizeof(struct sctp_adaptation_layer_indication); ali = (struct sctp_adaptation_layer_indication *)(mtod(m, caddr_t)+chunk_len); ali->ph.param_type = htons(SCTP_ULP_ADAPTATION); ali->ph.param_length = htons(parameter_len); ali->indication = htonl(inp->sctp_ep.adaptation_layer_indicator); chunk_len += parameter_len; } /* ECN parameter */ if (((asoc != NULL) && (asoc->ecn_supported == 1)) || ((asoc == NULL) && (inp->ecn_supported == 1))) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_ECN_CAPABLE); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* PR-SCTP supported parameter */ if (((asoc != NULL) && (asoc->prsctp_supported == 1)) || ((asoc == NULL) && (inp->prsctp_supported == 1))) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_PRSCTP_SUPPORTED); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* Add NAT friendly parameter */ if (nat_friendly) { parameter_len = (uint16_t)sizeof(struct sctp_paramhdr); ph = (struct sctp_paramhdr *)(mtod(m, caddr_t)+chunk_len); ph->param_type = htons(SCTP_HAS_NAT_SUPPORT); ph->param_length = htons(parameter_len); chunk_len += parameter_len; } /* And now tell the peer which extensions we support */ num_ext = 0; pr_supported = (struct sctp_supported_chunk_types_param *)(mtod(m, caddr_t)+chunk_len); if (((asoc != NULL) && (asoc->prsctp_supported == 1)) || ((asoc == NULL) && (inp->prsctp_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_FORWARD_CUM_TSN; if (((asoc != NULL) && (asoc->idata_supported == 1)) || ((asoc == NULL) && (inp->idata_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_IFORWARD_CUM_TSN; } } if (((asoc != NULL) && (asoc->auth_supported == 1)) || ((asoc == NULL) && (inp->auth_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_AUTHENTICATION; } if (((asoc != NULL) && (asoc->asconf_supported == 1)) || ((asoc == NULL) && (inp->asconf_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_ASCONF; pr_supported->chunk_types[num_ext++] = SCTP_ASCONF_ACK; } if (((asoc != NULL) && (asoc->reconfig_supported == 1)) || ((asoc == NULL) && (inp->reconfig_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_STREAM_RESET; } if (((asoc != NULL) && (asoc->idata_supported == 1)) || ((asoc == NULL) && (inp->idata_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_IDATA; } if (((asoc != NULL) && (asoc->nrsack_supported == 1)) || ((asoc == NULL) && (inp->nrsack_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_NR_SELECTIVE_ACK; } if (((asoc != NULL) && (asoc->pktdrop_supported == 1)) || ((asoc == NULL) && (inp->pktdrop_supported == 1))) { pr_supported->chunk_types[num_ext++] = SCTP_PACKET_DROPPED; } if (num_ext > 0) { parameter_len = (uint16_t)sizeof(struct sctp_supported_chunk_types_param) + num_ext; pr_supported->ph.param_type = htons(SCTP_SUPPORTED_CHUNK_EXT); pr_supported->ph.param_length = htons(parameter_len); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } /* add authentication parameters */ if (((asoc != NULL) && (asoc->auth_supported == 1)) || ((asoc == NULL) && (inp->auth_supported == 1))) { struct sctp_auth_random *randp; struct sctp_auth_hmac_algo *hmacs; struct sctp_auth_chunk_list *chunks; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } /* generate and add RANDOM parameter */ randp = (struct sctp_auth_random *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)sizeof(struct sctp_auth_random) + SCTP_AUTH_RANDOM_SIZE_DEFAULT; randp->ph.param_type = htons(SCTP_RANDOM); randp->ph.param_length = htons(parameter_len); SCTP_READ_RANDOM(randp->random_data, SCTP_AUTH_RANDOM_SIZE_DEFAULT); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } /* add HMAC_ALGO parameter */ hmacs = (struct sctp_auth_hmac_algo *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)sizeof(struct sctp_auth_hmac_algo) + sctp_serialize_hmaclist(inp->sctp_ep.local_hmacs, (uint8_t *)hmacs->hmac_ids); hmacs->ph.param_type = htons(SCTP_HMAC_LIST); hmacs->ph.param_length = htons(parameter_len); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; padding_len = 0; } /* add CHUNKS parameter */ chunks = (struct sctp_auth_chunk_list *)(mtod(m, caddr_t)+chunk_len); parameter_len = (uint16_t)sizeof(struct sctp_auth_chunk_list) + sctp_serialize_auth_chunks(inp->sctp_ep.local_auth_chunks, chunks->chunk_types); chunks->ph.param_type = htons(SCTP_CHUNK_LIST); chunks->ph.param_length = htons(parameter_len); padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; } SCTP_BUF_LEN(m) = chunk_len; m_last = m; /* now the addresses */ /* * To optimize this we could put the scoping stuff into a structure * and remove the individual uint8's from the stc structure. Then we * could just sifa in the address within the stc.. but for now this * is a quick hack to get the address stuff teased apart. */ scp.ipv4_addr_legal = stc.ipv4_addr_legal; scp.ipv6_addr_legal = stc.ipv6_addr_legal; scp.loopback_scope = stc.loopback_scope; scp.ipv4_local_scope = stc.ipv4_scope; scp.local_scope = stc.local_scope; scp.site_scope = stc.site_scope; m_last = sctp_add_addresses_to_i_ia(inp, stcb, &scp, m_last, cnt_inits_to, &padding_len, &chunk_len); /* padding_len can only be positive, if no addresses have been added */ if (padding_len > 0) { memset(mtod(m, caddr_t)+chunk_len, 0, padding_len); chunk_len += padding_len; SCTP_BUF_LEN(m) += padding_len; padding_len = 0; } /* tack on the operational error if present */ if (op_err) { parameter_len = 0; for (m_tmp = op_err; m_tmp != NULL; m_tmp = SCTP_BUF_NEXT(m_tmp)) { parameter_len += SCTP_BUF_LEN(m_tmp); } padding_len = SCTP_SIZE32(parameter_len) - parameter_len; SCTP_BUF_NEXT(m_last) = op_err; while (SCTP_BUF_NEXT(m_last) != NULL) { m_last = SCTP_BUF_NEXT(m_last); } chunk_len += parameter_len; } if (padding_len > 0) { m_last = sctp_add_pad_tombuf(m_last, padding_len); if (m_last == NULL) { /* Houston we have a problem, no space */ sctp_m_freem(m); return; } chunk_len += padding_len; padding_len = 0; } /* Now we must build a cookie */ m_cookie = sctp_add_cookie(init_pkt, offset, m, 0, &stc, &signature); if (m_cookie == NULL) { /* memory problem */ sctp_m_freem(m); return; } /* Now append the cookie to the end and update the space/size */ SCTP_BUF_NEXT(m_last) = m_cookie; parameter_len = 0; for (m_tmp = m_cookie; m_tmp != NULL; m_tmp = SCTP_BUF_NEXT(m_tmp)) { parameter_len += SCTP_BUF_LEN(m_tmp); if (SCTP_BUF_NEXT(m_tmp) == NULL) { m_last = m_tmp; } } padding_len = SCTP_SIZE32(parameter_len) - parameter_len; chunk_len += parameter_len; /* * Place in the size, but we don't include the last pad (if any) in * the INIT-ACK. */ initack->ch.chunk_length = htons(chunk_len); /* * Time to sign the cookie, we don't sign over the cookie signature * though thus we set trailer. */ (void)sctp_hmac_m(SCTP_HMAC, (uint8_t *)inp->sctp_ep.secret_key[(int)(inp->sctp_ep.current_secret_number)], SCTP_SECRET_SIZE, m_cookie, sizeof(struct sctp_paramhdr), (uint8_t *)signature, SCTP_SIGNATURE_SIZE); /* * We sifa 0 here to NOT set IP_DF if its IPv4, we ignore the return * here since the timer will drive a retranmission. */ if (padding_len > 0) { if (sctp_add_pad_tombuf(m_last, padding_len) == NULL) { sctp_m_freem(m); return; } } if (stc.loopback_scope) { over_addr = (union sctp_sockstore *)dst; } else { over_addr = NULL; } if ((error = sctp_lowlevel_chunk_output(inp, NULL, NULL, to, m, 0, NULL, 0, 0, 0, 0, inp->sctp_lport, sh->src_port, init_chk->init.initiate_tag, port, over_addr, mflowtype, mflowid, SCTP_SO_NOT_LOCKED))) { SCTPDBG(SCTP_DEBUG_OUTPUT4, "Gak send error %d\n", error); if (error == ENOBUFS) { if (asoc != NULL) { asoc->ifp_had_enobuf = 1; } SCTP_STAT_INCR(sctps_lowlevelerr); } } else { if (asoc != NULL) { asoc->ifp_had_enobuf = 0; } } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } static void sctp_prune_prsctp(struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_sndrcvinfo *srcv, int dataout) { int freed_spc = 0; struct sctp_tmit_chunk *chk, *nchk; SCTP_TCB_LOCK_ASSERT(stcb); if ((asoc->prsctp_supported) && (asoc->sent_queue_cnt_removeable > 0)) { TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) { /* * Look for chunks marked with the PR_SCTP flag AND * the buffer space flag. If the one being sent is * equal or greater priority then purge the old one * and free some space. */ if (PR_SCTP_BUF_ENABLED(chk->flags)) { /* * This one is PR-SCTP AND buffer space * limited type */ if (chk->rec.data.timetodrop.tv_sec >= (long)srcv->sinfo_timetolive) { /* * Lower numbers equates to higher * priority so if the one we are * looking at has a larger or equal * priority we want to drop the data * and NOT retransmit it. */ if (chk->data) { /* * We release the book_size * if the mbuf is here */ int ret_spc; uint8_t sent; if (chk->sent > SCTP_DATAGRAM_UNSENT) sent = 1; else sent = 0; ret_spc = sctp_release_pr_sctp_chunk(stcb, chk, sent, SCTP_SO_LOCKED); freed_spc += ret_spc; if (freed_spc >= dataout) { return; } } /* if chunk was present */ } /* if of sufficient priority */ } /* if chunk has enabled */ } /* tailqforeach */ TAILQ_FOREACH_SAFE(chk, &asoc->send_queue, sctp_next, nchk) { /* Here we must move to the sent queue and mark */ if (PR_SCTP_BUF_ENABLED(chk->flags)) { if (chk->rec.data.timetodrop.tv_sec >= (long)srcv->sinfo_timetolive) { if (chk->data) { /* * We release the book_size * if the mbuf is here */ int ret_spc; ret_spc = sctp_release_pr_sctp_chunk(stcb, chk, 0, SCTP_SO_LOCKED); freed_spc += ret_spc; if (freed_spc >= dataout) { return; } } /* end if chk->data */ } /* end if right class */ } /* end if chk pr-sctp */ } /* tailqforeachsafe (chk) */ } /* if enabled in asoc */ } int sctp_get_frag_point(struct sctp_tcb *stcb, struct sctp_association *asoc) { int siz, ovh; /* * For endpoints that have both v6 and v4 addresses we must reserve * room for the ipv6 header, for those that are only dealing with V4 * we use a larger frag point. */ if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) { ovh = SCTP_MIN_OVERHEAD; } else { ovh = SCTP_MIN_V4_OVERHEAD; } ovh += SCTP_DATA_CHUNK_OVERHEAD(stcb); if (stcb->asoc.sctp_frag_point > asoc->smallest_mtu) siz = asoc->smallest_mtu - ovh; else siz = (stcb->asoc.sctp_frag_point - ovh); /* * if (siz > (MCLBYTES-sizeof(struct sctp_data_chunk))) { */ /* A data chunk MUST fit in a cluster */ /* siz = (MCLBYTES - sizeof(struct sctp_data_chunk)); */ /* } */ /* adjust for an AUTH chunk if DATA requires auth */ if (sctp_auth_is_required_chunk(SCTP_DATA, stcb->asoc.peer_auth_chunks)) siz -= sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); if (siz % 4) { /* make it an even word boundary please */ siz -= (siz % 4); } return (siz); } static void sctp_set_prsctp_policy(struct sctp_stream_queue_pending *sp) { /* * We assume that the user wants PR_SCTP_TTL if the user provides a * positive lifetime but does not specify any PR_SCTP policy. */ if (PR_SCTP_ENABLED(sp->sinfo_flags)) { sp->act_flags |= PR_SCTP_POLICY(sp->sinfo_flags); } else if (sp->timetolive > 0) { sp->sinfo_flags |= SCTP_PR_SCTP_TTL; sp->act_flags |= PR_SCTP_POLICY(sp->sinfo_flags); } else { return; } switch (PR_SCTP_POLICY(sp->sinfo_flags)) { case CHUNK_FLAGS_PR_SCTP_BUF: /* * Time to live is a priority stored in tv_sec when doing * the buffer drop thing. */ sp->ts.tv_sec = sp->timetolive; sp->ts.tv_usec = 0; break; case CHUNK_FLAGS_PR_SCTP_TTL: { struct timeval tv; (void)SCTP_GETTIME_TIMEVAL(&sp->ts); tv.tv_sec = sp->timetolive / 1000; tv.tv_usec = (sp->timetolive * 1000) % 1000000; /* * TODO sctp_constants.h needs alternative time * macros when _KERNEL is undefined. */ timevaladd(&sp->ts, &tv); } break; case CHUNK_FLAGS_PR_SCTP_RTX: /* * Time to live is a the number or retransmissions stored in * tv_sec. */ sp->ts.tv_sec = sp->timetolive; sp->ts.tv_usec = 0; break; default: SCTPDBG(SCTP_DEBUG_USRREQ1, "Unknown PR_SCTP policy %u.\n", PR_SCTP_POLICY(sp->sinfo_flags)); break; } } static int sctp_msg_append(struct sctp_tcb *stcb, struct sctp_nets *net, struct mbuf *m, struct sctp_sndrcvinfo *srcv, int hold_stcb_lock) { int error = 0; struct mbuf *at; struct sctp_stream_queue_pending *sp = NULL; struct sctp_stream_out *strm; /* * Given an mbuf chain, put it into the association send queue and * place it on the wheel */ if (srcv->sinfo_stream >= stcb->asoc.streamoutcnt) { /* Invalid stream number */ SCTP_LTRACE_ERR_RET_PKT(m, NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_now; } if ((stcb->asoc.stream_locked) && (stcb->asoc.stream_locked_on != srcv->sinfo_stream)) { SCTP_LTRACE_ERR_RET_PKT(m, NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_now; } strm = &stcb->asoc.strmout[srcv->sinfo_stream]; /* Now can we send this? */ if ((SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_SENT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_ACK_SENT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED) || (stcb->asoc.state & SCTP_STATE_SHUTDOWN_PENDING)) { /* got data while shutting down */ SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ECONNRESET); error = ECONNRESET; goto out_now; } sctp_alloc_a_strmoq(stcb, sp); if (sp == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); error = ENOMEM; goto out_now; } sp->sinfo_flags = srcv->sinfo_flags; sp->timetolive = srcv->sinfo_timetolive; sp->ppid = srcv->sinfo_ppid; sp->context = srcv->sinfo_context; sp->fsn = 0; if (sp->sinfo_flags & SCTP_ADDR_OVER) { sp->net = net; atomic_add_int(&sp->net->ref_count, 1); } else { sp->net = NULL; } (void)SCTP_GETTIME_TIMEVAL(&sp->ts); sp->sid = srcv->sinfo_stream; sp->msg_is_complete = 1; sp->sender_all_done = 1; sp->some_taken = 0; sp->data = m; sp->tail_mbuf = NULL; sctp_set_prsctp_policy(sp); /* * We could in theory (for sendall) sifa the length in, but we would * still have to hunt through the chain since we need to setup the * tail_mbuf */ sp->length = 0; for (at = m; at; at = SCTP_BUF_NEXT(at)) { if (SCTP_BUF_NEXT(at) == NULL) sp->tail_mbuf = at; sp->length += SCTP_BUF_LEN(at); } if (srcv->sinfo_keynumber_valid) { sp->auth_keyid = srcv->sinfo_keynumber; } else { sp->auth_keyid = stcb->asoc.authinfo.active_keyid; } if (sctp_auth_is_required_chunk(SCTP_DATA, stcb->asoc.peer_auth_chunks)) { sctp_auth_key_acquire(stcb, sp->auth_keyid); sp->holds_key_ref = 1; } if (hold_stcb_lock == 0) { SCTP_TCB_SEND_LOCK(stcb); } sctp_snd_sb_alloc(stcb, sp->length); atomic_add_int(&stcb->asoc.stream_queue_cnt, 1); TAILQ_INSERT_TAIL(&strm->outqueue, sp, next); stcb->asoc.ss_functions.sctp_ss_add_to_stream(stcb, &stcb->asoc, strm, sp, 1); m = NULL; if (hold_stcb_lock == 0) { SCTP_TCB_SEND_UNLOCK(stcb); } out_now: if (m) { sctp_m_freem(m); } return (error); } static struct mbuf * sctp_copy_mbufchain(struct mbuf *clonechain, struct mbuf *outchain, struct mbuf **endofchain, int can_take_mbuf, int sizeofcpy, uint8_t copy_by_ref) { struct mbuf *m; struct mbuf *appendchain; caddr_t cp; int len; if (endofchain == NULL) { /* error */ error_out: if (outchain) sctp_m_freem(outchain); return (NULL); } if (can_take_mbuf) { appendchain = clonechain; } else { if (!copy_by_ref && (sizeofcpy <= (int)((((SCTP_BASE_SYSCTL(sctp_mbuf_threshold_count) - 1) * MLEN) + MHLEN))) ) { /* Its not in a cluster */ if (*endofchain == NULL) { /* lets get a mbuf cluster */ if (outchain == NULL) { /* This is the general case */ new_mbuf: outchain = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_HEADER); if (outchain == NULL) { goto error_out; } SCTP_BUF_LEN(outchain) = 0; *endofchain = outchain; /* get the prepend space */ SCTP_BUF_RESV_UF(outchain, (SCTP_FIRST_MBUF_RESV + 4)); } else { /* * We really should not get a NULL * in endofchain */ /* find end */ m = outchain; while (m) { if (SCTP_BUF_NEXT(m) == NULL) { *endofchain = m; break; } m = SCTP_BUF_NEXT(m); } /* sanity */ if (*endofchain == NULL) { /* * huh, TSNH XXX maybe we * should panic */ sctp_m_freem(outchain); goto new_mbuf; } } /* get the new end of length */ len = (int)M_TRAILINGSPACE(*endofchain); } else { /* how much is left at the end? */ len = (int)M_TRAILINGSPACE(*endofchain); } /* Find the end of the data, for appending */ cp = (mtod((*endofchain), caddr_t)+SCTP_BUF_LEN((*endofchain))); /* Now lets copy it out */ if (len >= sizeofcpy) { /* It all fits, copy it in */ m_copydata(clonechain, 0, sizeofcpy, cp); SCTP_BUF_LEN((*endofchain)) += sizeofcpy; } else { /* fill up the end of the chain */ if (len > 0) { m_copydata(clonechain, 0, len, cp); SCTP_BUF_LEN((*endofchain)) += len; /* now we need another one */ sizeofcpy -= len; } m = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_HEADER); if (m == NULL) { /* We failed */ goto error_out; } SCTP_BUF_NEXT((*endofchain)) = m; *endofchain = m; cp = mtod((*endofchain), caddr_t); m_copydata(clonechain, len, sizeofcpy, cp); SCTP_BUF_LEN((*endofchain)) += sizeofcpy; } return (outchain); } else { /* copy the old fashion way */ appendchain = SCTP_M_COPYM(clonechain, 0, M_COPYALL, M_NOWAIT); #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(appendchain, SCTP_MBUF_ICOPY); } #endif } } if (appendchain == NULL) { /* error */ if (outchain) sctp_m_freem(outchain); return (NULL); } if (outchain) { /* tack on to the end */ if (*endofchain != NULL) { SCTP_BUF_NEXT(((*endofchain))) = appendchain; } else { m = outchain; while (m) { if (SCTP_BUF_NEXT(m) == NULL) { SCTP_BUF_NEXT(m) = appendchain; break; } m = SCTP_BUF_NEXT(m); } } /* * save off the end and update the end-chain position */ m = appendchain; while (m) { if (SCTP_BUF_NEXT(m) == NULL) { *endofchain = m; break; } m = SCTP_BUF_NEXT(m); } return (outchain); } else { /* save off the end and update the end-chain position */ m = appendchain; while (m) { if (SCTP_BUF_NEXT(m) == NULL) { *endofchain = m; break; } m = SCTP_BUF_NEXT(m); } return (appendchain); } } static int sctp_med_chunk_output(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_association *asoc, int *num_out, int *reason_code, int control_only, int from_where, struct timeval *now, int *now_filled, int frag_point, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ); static void sctp_sendall_iterator(struct sctp_inpcb *inp, struct sctp_tcb *stcb, void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_copy_all *ca; struct mbuf *m; int ret = 0; int added_control = 0; int un_sent, do_chunk_output = 1; struct sctp_association *asoc; struct sctp_nets *net; ca = (struct sctp_copy_all *)ptr; if (ca->m == NULL) { return; } if (ca->inp != inp) { /* TSNH */ return; } if (ca->sndlen > 0) { m = SCTP_M_COPYM(ca->m, 0, M_COPYALL, M_NOWAIT); if (m == NULL) { /* can't copy so we are done */ ca->cnt_failed++; return; } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(m, SCTP_MBUF_ICOPY); } #endif } else { m = NULL; } SCTP_TCB_LOCK_ASSERT(stcb); if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } if (ca->sndrcv.sinfo_flags & SCTP_ABORT) { /* Abort this assoc with m as the user defined reason */ if (m != NULL) { SCTP_BUF_PREPEND(m, sizeof(struct sctp_paramhdr), M_NOWAIT); } else { m = sctp_get_mbuf_for_msg(sizeof(struct sctp_paramhdr), 0, M_NOWAIT, 1, MT_DATA); SCTP_BUF_LEN(m) = sizeof(struct sctp_paramhdr); } if (m != NULL) { struct sctp_paramhdr *ph; ph = mtod(m, struct sctp_paramhdr *); ph->param_type = htons(SCTP_CAUSE_USER_INITIATED_ABT); ph->param_length = htons((uint16_t)(sizeof(struct sctp_paramhdr) + ca->sndlen)); } /* * We add one here to keep the assoc from dis-appearing on * us. */ atomic_add_int(&stcb->asoc.refcnt, 1); sctp_abort_an_association(inp, stcb, m, SCTP_SO_NOT_LOCKED); /* * sctp_abort_an_association calls sctp_free_asoc() free * association will NOT free it since we incremented the * refcnt .. we do this to prevent it being freed and things * getting tricky since we could end up (from free_asoc) * calling inpcb_free which would get a recursive lock call * to the iterator lock.. But as a consequence of that the * stcb will return to us un-locked.. since free_asoc * returns with either no TCB or the TCB unlocked, we must * relock.. to unlock in the iterator timer :-0 */ SCTP_TCB_LOCK(stcb); atomic_add_int(&stcb->asoc.refcnt, -1); goto no_chunk_output; } else { if (m) { ret = sctp_msg_append(stcb, net, m, &ca->sndrcv, 1); } asoc = &stcb->asoc; if (ca->sndrcv.sinfo_flags & SCTP_EOF) { /* shutdown this assoc */ if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue) && sctp_is_there_unsent_data(stcb, SCTP_SO_NOT_LOCKED) == 0) { if ((*asoc->ss_functions.sctp_ss_is_user_msgs_incomplete) (stcb, asoc)) { goto abort_anyway; } /* * there is nothing queued to send, so I'm * done... */ if ((SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_SENT) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_RECEIVED) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_ACK_SENT)) { /* * only send SHUTDOWN the first time * through */ if (SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } SCTP_SET_STATE(stcb, SCTP_STATE_SHUTDOWN_SENT); sctp_stop_timers_for_shutdown(stcb); sctp_send_shutdown(stcb, net); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN, stcb->sctp_ep, stcb, net); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, stcb->sctp_ep, stcb, NULL); added_control = 1; do_chunk_output = 0; } } else { /* * we still got (or just got) data to send, * so set SHUTDOWN_PENDING */ /* * XXX sockets draft says that SCTP_EOF * should be sent with no data. currently, * we will allow user data to be sent first * and move to SHUTDOWN-PENDING */ if ((SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_SENT) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_RECEIVED) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_ACK_SENT)) { if ((*asoc->ss_functions.sctp_ss_is_user_msgs_incomplete) (stcb, asoc)) { SCTP_ADD_SUBSTATE(stcb, SCTP_STATE_PARTIAL_MSG_LEFT); } SCTP_ADD_SUBSTATE(stcb, SCTP_STATE_SHUTDOWN_PENDING); if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue) && (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT)) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; abort_anyway: snprintf(msg, sizeof(msg), "%s:%d at %s", __FILE__, __LINE__, __func__); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); atomic_add_int(&stcb->asoc.refcnt, 1); sctp_abort_an_association(stcb->sctp_ep, stcb, op_err, SCTP_SO_NOT_LOCKED); atomic_add_int(&stcb->asoc.refcnt, -1); goto no_chunk_output; } sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, stcb->sctp_ep, stcb, NULL); } } } } un_sent = ((stcb->asoc.total_output_queue_size - stcb->asoc.total_flight) + (stcb->asoc.stream_queue_cnt * SCTP_DATA_CHUNK_OVERHEAD(stcb))); if ((sctp_is_feature_off(inp, SCTP_PCB_FLAGS_NODELAY)) && (stcb->asoc.total_flight > 0) && (un_sent < (int)(stcb->asoc.smallest_mtu - SCTP_MIN_OVERHEAD))) { do_chunk_output = 0; } if (do_chunk_output) sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_NOT_LOCKED); else if (added_control) { int num_out, reason, now_filled = 0; struct timeval now; int frag_point; frag_point = sctp_get_frag_point(stcb, &stcb->asoc); (void)sctp_med_chunk_output(inp, stcb, &stcb->asoc, &num_out, &reason, 1, 1, &now, &now_filled, frag_point, SCTP_SO_NOT_LOCKED); } no_chunk_output: if (ret) { ca->cnt_failed++; } else { ca->cnt_sent++; } } static void sctp_sendall_completes(void *ptr, uint32_t val SCTP_UNUSED) { struct sctp_copy_all *ca; ca = (struct sctp_copy_all *)ptr; /* * Do a notify here? Kacheong suggests that the notify be done at * the send time.. so you would push up a notification if any send * failed. Don't know if this is feasible since the only failures we * have is "memory" related and if you cannot get an mbuf to send * the data you surely can't get an mbuf to send up to notify the * user you can't send the data :-> */ /* now free everything */ if (ca->inp) { /* Lets clear the flag to allow others to run. */ ca->inp->sctp_flags &= ~SCTP_PCB_FLAGS_SND_ITERATOR_UP; } sctp_m_freem(ca->m); SCTP_FREE(ca, SCTP_M_COPYAL); } static struct mbuf * sctp_copy_out_all(struct uio *uio, ssize_t len) { struct mbuf *ret, *at; ssize_t left, willcpy, cancpy, error; ret = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_WAITOK, 1, MT_DATA); if (ret == NULL) { /* TSNH */ return (NULL); } left = len; SCTP_BUF_LEN(ret) = 0; /* save space for the data chunk header */ cancpy = (int)M_TRAILINGSPACE(ret); willcpy = min(cancpy, left); at = ret; while (left > 0) { /* Align data to the end */ error = uiomove(mtod(at, caddr_t), (int)willcpy, uio); if (error) { err_out_now: sctp_m_freem(at); return (NULL); } SCTP_BUF_LEN(at) = (int)willcpy; SCTP_BUF_NEXT_PKT(at) = SCTP_BUF_NEXT(at) = 0; left -= willcpy; if (left > 0) { SCTP_BUF_NEXT(at) = sctp_get_mbuf_for_msg((unsigned int)left, 0, M_WAITOK, 1, MT_DATA); if (SCTP_BUF_NEXT(at) == NULL) { goto err_out_now; } at = SCTP_BUF_NEXT(at); SCTP_BUF_LEN(at) = 0; cancpy = (int)M_TRAILINGSPACE(at); willcpy = min(cancpy, left); } } return (ret); } static int sctp_sendall(struct sctp_inpcb *inp, struct uio *uio, struct mbuf *m, struct sctp_sndrcvinfo *srcv) { int ret; struct sctp_copy_all *ca; if (inp->sctp_flags & SCTP_PCB_FLAGS_SND_ITERATOR_UP) { /* There is another. */ return (EBUSY); } if (uio->uio_resid > (ssize_t)SCTP_BASE_SYSCTL(sctp_sendall_limit)) { /* You must not be larger than the limit! */ return (EMSGSIZE); } SCTP_MALLOC(ca, struct sctp_copy_all *, sizeof(struct sctp_copy_all), SCTP_M_COPYAL); if (ca == NULL) { sctp_m_freem(m); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } memset(ca, 0, sizeof(struct sctp_copy_all)); ca->inp = inp; if (srcv) { memcpy(&ca->sndrcv, srcv, sizeof(struct sctp_nonpad_sndrcvinfo)); } /* * take off the sendall flag, it would be bad if we failed to do * this :-0 */ ca->sndrcv.sinfo_flags &= ~SCTP_SENDALL; /* get length and mbuf chain */ if (uio) { ca->sndlen = uio->uio_resid; ca->m = sctp_copy_out_all(uio, ca->sndlen); if (ca->m == NULL) { SCTP_FREE(ca, SCTP_M_COPYAL); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } } else { /* Gather the length of the send */ struct mbuf *mat; ca->sndlen = 0; for (mat = m; mat; mat = SCTP_BUF_NEXT(mat)) { ca->sndlen += SCTP_BUF_LEN(mat); } } inp->sctp_flags |= SCTP_PCB_FLAGS_SND_ITERATOR_UP; ret = sctp_initiate_iterator(NULL, sctp_sendall_iterator, NULL, SCTP_PCB_ANY_FLAGS, SCTP_PCB_ANY_FEATURES, SCTP_ASOC_ANY_STATE, (void *)ca, 0, sctp_sendall_completes, inp, 1); if (ret) { inp->sctp_flags &= ~SCTP_PCB_FLAGS_SND_ITERATOR_UP; SCTP_FREE(ca, SCTP_M_COPYAL); SCTP_LTRACE_ERR_RET_PKT(m, inp, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EFAULT); return (EFAULT); } return (0); } void sctp_toss_old_cookies(struct sctp_tcb *stcb, struct sctp_association *asoc) { struct sctp_tmit_chunk *chk, *nchk; TAILQ_FOREACH_SAFE(chk, &asoc->control_send_queue, sctp_next, nchk) { if (chk->rec.chunk_id.id == SCTP_COOKIE_ECHO) { TAILQ_REMOVE(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt--; if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); } } } void sctp_toss_old_asconf(struct sctp_tcb *stcb) { struct sctp_association *asoc; struct sctp_tmit_chunk *chk, *nchk; struct sctp_asconf_chunk *acp; asoc = &stcb->asoc; TAILQ_FOREACH_SAFE(chk, &asoc->asconf_send_queue, sctp_next, nchk) { /* find SCTP_ASCONF chunk in queue */ if (chk->rec.chunk_id.id == SCTP_ASCONF) { if (chk->data) { acp = mtod(chk->data, struct sctp_asconf_chunk *); if (SCTP_TSN_GT(ntohl(acp->serial_number), asoc->asconf_seq_out_acked)) { /* Not Acked yet */ break; } } TAILQ_REMOVE(&asoc->asconf_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt--; if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); } } } static void sctp_clean_up_datalist(struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_tmit_chunk **data_list, int bundle_at, struct sctp_nets *net) { int i; struct sctp_tmit_chunk *tp1; for (i = 0; i < bundle_at; i++) { /* off of the send queue */ TAILQ_REMOVE(&asoc->send_queue, data_list[i], sctp_next); asoc->send_queue_cnt--; if (i > 0) { /* * Any chunk NOT 0 you zap the time chunk 0 gets * zapped or set based on if a RTO measurment is * needed. */ data_list[i]->do_rtt = 0; } /* record time */ data_list[i]->sent_rcv_time = net->last_sent_time; data_list[i]->rec.data.cwnd_at_send = net->cwnd; data_list[i]->rec.data.fast_retran_tsn = data_list[i]->rec.data.tsn; if (data_list[i]->whoTo == NULL) { data_list[i]->whoTo = net; atomic_add_int(&net->ref_count, 1); } /* on to the sent queue */ tp1 = TAILQ_LAST(&asoc->sent_queue, sctpchunk_listhead); if ((tp1) && SCTP_TSN_GT(tp1->rec.data.tsn, data_list[i]->rec.data.tsn)) { struct sctp_tmit_chunk *tpp; /* need to move back */ back_up_more: tpp = TAILQ_PREV(tp1, sctpchunk_listhead, sctp_next); if (tpp == NULL) { TAILQ_INSERT_BEFORE(tp1, data_list[i], sctp_next); goto all_done; } tp1 = tpp; if (SCTP_TSN_GT(tp1->rec.data.tsn, data_list[i]->rec.data.tsn)) { goto back_up_more; } TAILQ_INSERT_AFTER(&asoc->sent_queue, tp1, data_list[i], sctp_next); } else { TAILQ_INSERT_TAIL(&asoc->sent_queue, data_list[i], sctp_next); } all_done: /* This does not lower until the cum-ack passes it */ asoc->sent_queue_cnt++; if ((asoc->peers_rwnd <= 0) && (asoc->total_flight == 0) && (bundle_at == 1)) { /* Mark the chunk as being a window probe */ SCTP_STAT_INCR(sctps_windowprobed); } #ifdef SCTP_AUDITING_ENABLED sctp_audit_log(0xC2, 3); #endif data_list[i]->sent = SCTP_DATAGRAM_SENT; data_list[i]->snd_count = 1; data_list[i]->rec.data.chunk_was_revoked = 0; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) { sctp_misc_ints(SCTP_FLIGHT_LOG_UP, data_list[i]->whoTo->flight_size, data_list[i]->book_size, (uint32_t)(uintptr_t)data_list[i]->whoTo, data_list[i]->rec.data.tsn); } sctp_flight_size_increase(data_list[i]); sctp_total_flight_increase(stcb, data_list[i]); if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) { sctp_log_rwnd(SCTP_DECREASE_PEER_RWND, asoc->peers_rwnd, data_list[i]->send_size, SCTP_BASE_SYSCTL(sctp_peer_chunk_oh)); } asoc->peers_rwnd = sctp_sbspace_sub(asoc->peers_rwnd, (uint32_t)(data_list[i]->send_size + SCTP_BASE_SYSCTL(sctp_peer_chunk_oh))); if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) { /* SWS sender side engages */ asoc->peers_rwnd = 0; } } if (asoc->cc_functions.sctp_cwnd_update_packet_transmitted) { (*asoc->cc_functions.sctp_cwnd_update_packet_transmitted) (stcb, net); } } static void sctp_clean_up_ctl(struct sctp_tcb *stcb, struct sctp_association *asoc, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { struct sctp_tmit_chunk *chk, *nchk; TAILQ_FOREACH_SAFE(chk, &asoc->control_send_queue, sctp_next, nchk) { if ((chk->rec.chunk_id.id == SCTP_SELECTIVE_ACK) || (chk->rec.chunk_id.id == SCTP_NR_SELECTIVE_ACK) || /* EY */ (chk->rec.chunk_id.id == SCTP_HEARTBEAT_REQUEST) || (chk->rec.chunk_id.id == SCTP_HEARTBEAT_ACK) || (chk->rec.chunk_id.id == SCTP_FORWARD_CUM_TSN) || (chk->rec.chunk_id.id == SCTP_SHUTDOWN) || (chk->rec.chunk_id.id == SCTP_SHUTDOWN_ACK) || (chk->rec.chunk_id.id == SCTP_OPERATION_ERROR) || (chk->rec.chunk_id.id == SCTP_PACKET_DROPPED) || (chk->rec.chunk_id.id == SCTP_COOKIE_ACK) || (chk->rec.chunk_id.id == SCTP_ECN_CWR) || (chk->rec.chunk_id.id == SCTP_ASCONF_ACK)) { /* Stray chunks must be cleaned up */ clean_up_anyway: TAILQ_REMOVE(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt--; if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } if (chk->rec.chunk_id.id == SCTP_FORWARD_CUM_TSN) { asoc->fwd_tsn_cnt--; } sctp_free_a_chunk(stcb, chk, so_locked); } else if (chk->rec.chunk_id.id == SCTP_STREAM_RESET) { /* special handling, we must look into the param */ if (chk != asoc->str_reset) { goto clean_up_anyway; } } } } static uint32_t sctp_can_we_split_this(struct sctp_tcb *stcb, uint32_t length, uint32_t space_left, uint32_t frag_point, int eeor_on) { /* * Make a decision on if I should split a msg into multiple parts. * This is only asked of incomplete messages. */ if (eeor_on) { /* * If we are doing EEOR we need to always send it if its the * entire thing, since it might be all the guy is putting in * the hopper. */ if (space_left >= length) { /*- * If we have data outstanding, * we get another chance when the sack * arrives to transmit - wait for more data */ if (stcb->asoc.total_flight == 0) { /* * If nothing is in flight, we zero the * packet counter. */ return (length); } return (0); } else { /* You can fill the rest */ return (space_left); } } /*- * For those strange folk that make the send buffer * smaller than our fragmentation point, we can't * get a full msg in so we have to allow splitting. */ if (SCTP_SB_LIMIT_SND(stcb->sctp_socket) < frag_point) { return (length); } if ((length <= space_left) || ((length - space_left) < SCTP_BASE_SYSCTL(sctp_min_residual))) { /* Sub-optimial residual don't split in non-eeor mode. */ return (0); } /* * If we reach here length is larger than the space_left. Do we wish * to split it for the sake of packet putting together? */ if (space_left >= min(SCTP_BASE_SYSCTL(sctp_min_split_point), frag_point)) { /* Its ok to split it */ return (min(space_left, frag_point)); } /* Nope, can't split */ return (0); } static uint32_t sctp_move_to_outqueue(struct sctp_tcb *stcb, struct sctp_stream_out *strq, uint32_t space_left, uint32_t frag_point, int *giveup, int eeor_mode, int *bail, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { /* Move from the stream to the send_queue keeping track of the total */ struct sctp_association *asoc; struct sctp_stream_queue_pending *sp; struct sctp_tmit_chunk *chk; struct sctp_data_chunk *dchkh = NULL; struct sctp_idata_chunk *ndchkh = NULL; uint32_t to_move, length; int leading; uint8_t rcv_flags = 0; uint8_t some_taken; uint8_t send_lock_up = 0; SCTP_TCB_LOCK_ASSERT(stcb); asoc = &stcb->asoc; one_more_time: /* sa_ignore FREED_MEMORY */ sp = TAILQ_FIRST(&strq->outqueue); if (sp == NULL) { if (send_lock_up == 0) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } sp = TAILQ_FIRST(&strq->outqueue); if (sp) { goto one_more_time; } if ((sctp_is_feature_on(stcb->sctp_ep, SCTP_PCB_FLAGS_EXPLICIT_EOR) == 0) && (stcb->asoc.idata_supported == 0) && (strq->last_msg_incomplete)) { SCTP_PRINTF("Huh? Stream:%d lm_in_c=%d but queue is NULL\n", strq->sid, strq->last_msg_incomplete); strq->last_msg_incomplete = 0; } to_move = 0; if (send_lock_up) { SCTP_TCB_SEND_UNLOCK(stcb); send_lock_up = 0; } goto out_of; } if ((sp->msg_is_complete) && (sp->length == 0)) { if (sp->sender_all_done) { /* * We are doing deferred cleanup. Last time through * when we took all the data the sender_all_done was * not set. */ if ((sp->put_last_out == 0) && (sp->discard_rest == 0)) { SCTP_PRINTF("Gak, put out entire msg with NO end!-1\n"); SCTP_PRINTF("sender_done:%d len:%d msg_comp:%d put_last_out:%d send_lock:%d\n", sp->sender_all_done, sp->length, sp->msg_is_complete, sp->put_last_out, send_lock_up); } if ((TAILQ_NEXT(sp, next) == NULL) && (send_lock_up == 0)) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } atomic_subtract_int(&asoc->stream_queue_cnt, 1); TAILQ_REMOVE(&strq->outqueue, sp, next); stcb->asoc.ss_functions.sctp_ss_remove_from_stream(stcb, asoc, strq, sp, send_lock_up); if ((strq->state == SCTP_STREAM_RESET_PENDING) && (strq->chunks_on_queues == 0) && TAILQ_EMPTY(&strq->outqueue)) { stcb->asoc.trigger_reset = 1; } if (sp->net) { sctp_free_remote_addr(sp->net); sp->net = NULL; } if (sp->data) { sctp_m_freem(sp->data); sp->data = NULL; } sctp_free_a_strmoq(stcb, sp, so_locked); /* we can't be locked to it */ if (send_lock_up) { SCTP_TCB_SEND_UNLOCK(stcb); send_lock_up = 0; } /* back to get the next msg */ goto one_more_time; } else { /* * sender just finished this but still holds a * reference */ *giveup = 1; to_move = 0; goto out_of; } } else { /* is there some to get */ if (sp->length == 0) { /* no */ *giveup = 1; to_move = 0; goto out_of; } else if (sp->discard_rest) { if (send_lock_up == 0) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } /* Whack down the size */ atomic_subtract_int(&stcb->asoc.total_output_queue_size, sp->length); if ((stcb->sctp_socket != NULL) && ((stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE) || (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL))) { atomic_subtract_int(&stcb->sctp_socket->so_snd.sb_cc, sp->length); } if (sp->data) { sctp_m_freem(sp->data); sp->data = NULL; sp->tail_mbuf = NULL; } sp->length = 0; sp->some_taken = 1; *giveup = 1; to_move = 0; goto out_of; } } some_taken = sp->some_taken; re_look: length = sp->length; if (sp->msg_is_complete) { /* The message is complete */ to_move = min(length, frag_point); if (to_move == length) { /* All of it fits in the MTU */ if (sp->some_taken) { rcv_flags |= SCTP_DATA_LAST_FRAG; } else { rcv_flags |= SCTP_DATA_NOT_FRAG; } sp->put_last_out = 1; if (sp->sinfo_flags & SCTP_SACK_IMMEDIATELY) { rcv_flags |= SCTP_DATA_SACK_IMMEDIATELY; } } else { /* Not all of it fits, we fragment */ if (sp->some_taken == 0) { rcv_flags |= SCTP_DATA_FIRST_FRAG; } sp->some_taken = 1; } } else { to_move = sctp_can_we_split_this(stcb, length, space_left, frag_point, eeor_mode); if (to_move) { /*- * We use a snapshot of length in case it * is expanding during the compare. */ uint32_t llen; llen = length; if (to_move >= llen) { to_move = llen; if (send_lock_up == 0) { /*- * We are taking all of an incomplete msg * thus we need a send lock. */ SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; if (sp->msg_is_complete) { /* * the sender finished the * msg */ goto re_look; } } } if (sp->some_taken == 0) { rcv_flags |= SCTP_DATA_FIRST_FRAG; sp->some_taken = 1; } } else { /* Nothing to take. */ *giveup = 1; to_move = 0; goto out_of; } } /* If we reach here, we can copy out a chunk */ sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* No chunk memory */ *giveup = 1; to_move = 0; goto out_of; } /* * Setup for unordered if needed by looking at the user sent info * flags. */ if (sp->sinfo_flags & SCTP_UNORDERED) { rcv_flags |= SCTP_DATA_UNORDERED; } if (SCTP_BASE_SYSCTL(sctp_enable_sack_immediately) && (sp->sinfo_flags & SCTP_EOF) == SCTP_EOF) { rcv_flags |= SCTP_DATA_SACK_IMMEDIATELY; } /* clear out the chunk before setting up */ memset(chk, 0, sizeof(*chk)); chk->rec.data.rcv_flags = rcv_flags; if (to_move >= length) { /* we think we can steal the whole thing */ if ((sp->sender_all_done == 0) && (send_lock_up == 0)) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } if (to_move < sp->length) { /* bail, it changed */ goto dont_do_it; } chk->data = sp->data; chk->last_mbuf = sp->tail_mbuf; /* register the stealing */ sp->data = sp->tail_mbuf = NULL; } else { struct mbuf *m; dont_do_it: chk->data = SCTP_M_COPYM(sp->data, 0, to_move, M_NOWAIT); chk->last_mbuf = NULL; if (chk->data == NULL) { sp->some_taken = some_taken; sctp_free_a_chunk(stcb, chk, so_locked); *bail = 1; to_move = 0; goto out_of; } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(chk->data, SCTP_MBUF_ICOPY); } #endif /* Pull off the data */ m_adj(sp->data, to_move); /* Now lets work our way down and compact it */ m = sp->data; while (m && (SCTP_BUF_LEN(m) == 0)) { sp->data = SCTP_BUF_NEXT(m); SCTP_BUF_NEXT(m) = NULL; if (sp->tail_mbuf == m) { /*- * Freeing tail? TSNH since * we supposedly were taking less * than the sp->length. */ #ifdef INVARIANTS panic("Huh, freing tail? - TSNH"); #else SCTP_PRINTF("Huh, freeing tail? - TSNH\n"); sp->tail_mbuf = sp->data = NULL; sp->length = 0; #endif } sctp_m_free(m); m = sp->data; } } if (SCTP_BUF_IS_EXTENDED(chk->data)) { chk->copy_by_ref = 1; } else { chk->copy_by_ref = 0; } /* * get last_mbuf and counts of mb usage This is ugly but hopefully * its only one mbuf. */ if (chk->last_mbuf == NULL) { chk->last_mbuf = chk->data; while (SCTP_BUF_NEXT(chk->last_mbuf) != NULL) { chk->last_mbuf = SCTP_BUF_NEXT(chk->last_mbuf); } } if (to_move > length) { /*- This should not happen either * since we always lower to_move to the size * of sp->length if its larger. */ #ifdef INVARIANTS panic("Huh, how can to_move be larger?"); #else SCTP_PRINTF("Huh, how can to_move be larger?\n"); sp->length = 0; #endif } else { atomic_subtract_int(&sp->length, to_move); } leading = SCTP_DATA_CHUNK_OVERHEAD(stcb); if (M_LEADINGSPACE(chk->data) < leading) { /* Not enough room for a chunk header, get some */ struct mbuf *m; m = sctp_get_mbuf_for_msg(1, 0, M_NOWAIT, 1, MT_DATA); if (m == NULL) { /* * we're in trouble here. _PREPEND below will free * all the data if there is no leading space, so we * must put the data back and restore. */ if (send_lock_up == 0) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } if (sp->data == NULL) { /* unsteal the data */ sp->data = chk->data; sp->tail_mbuf = chk->last_mbuf; } else { struct mbuf *m_tmp; /* reassemble the data */ m_tmp = sp->data; sp->data = chk->data; SCTP_BUF_NEXT(chk->last_mbuf) = m_tmp; } sp->some_taken = some_taken; atomic_add_int(&sp->length, to_move); chk->data = NULL; *bail = 1; sctp_free_a_chunk(stcb, chk, so_locked); to_move = 0; goto out_of; } else { SCTP_BUF_LEN(m) = 0; SCTP_BUF_NEXT(m) = chk->data; chk->data = m; M_ALIGN(chk->data, 4); } } SCTP_BUF_PREPEND(chk->data, SCTP_DATA_CHUNK_OVERHEAD(stcb), M_NOWAIT); if (chk->data == NULL) { /* HELP, TSNH since we assured it would not above? */ #ifdef INVARIANTS panic("prepend failes HELP?"); #else SCTP_PRINTF("prepend fails HELP?\n"); sctp_free_a_chunk(stcb, chk, so_locked); #endif *bail = 1; to_move = 0; goto out_of; } sctp_snd_sb_alloc(stcb, SCTP_DATA_CHUNK_OVERHEAD(stcb)); chk->book_size = chk->send_size = (uint16_t)(to_move + SCTP_DATA_CHUNK_OVERHEAD(stcb)); chk->book_size_scale = 0; chk->sent = SCTP_DATAGRAM_UNSENT; chk->flags = 0; chk->asoc = &stcb->asoc; chk->pad_inplace = 0; chk->no_fr_allowed = 0; if (stcb->asoc.idata_supported == 0) { if (rcv_flags & SCTP_DATA_UNORDERED) { /* Just use 0. The receiver ignores the values. */ chk->rec.data.mid = 0; } else { chk->rec.data.mid = strq->next_mid_ordered; if (rcv_flags & SCTP_DATA_LAST_FRAG) { strq->next_mid_ordered++; } } } else { if (rcv_flags & SCTP_DATA_UNORDERED) { chk->rec.data.mid = strq->next_mid_unordered; if (rcv_flags & SCTP_DATA_LAST_FRAG) { strq->next_mid_unordered++; } } else { chk->rec.data.mid = strq->next_mid_ordered; if (rcv_flags & SCTP_DATA_LAST_FRAG) { strq->next_mid_ordered++; } } } chk->rec.data.sid = sp->sid; chk->rec.data.ppid = sp->ppid; chk->rec.data.context = sp->context; chk->rec.data.doing_fast_retransmit = 0; chk->rec.data.timetodrop = sp->ts; chk->flags = sp->act_flags; if (sp->net) { chk->whoTo = sp->net; atomic_add_int(&chk->whoTo->ref_count, 1); } else chk->whoTo = NULL; if (sp->holds_key_ref) { chk->auth_keyid = sp->auth_keyid; sctp_auth_key_acquire(stcb, chk->auth_keyid); chk->holds_key_ref = 1; } chk->rec.data.tsn = atomic_fetchadd_int(&asoc->sending_seq, 1); if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_AT_SEND_2_OUTQ) { sctp_misc_ints(SCTP_STRMOUT_LOG_SEND, (uint32_t)(uintptr_t)stcb, sp->length, (uint32_t)((chk->rec.data.sid << 16) | (0x0000ffff & chk->rec.data.mid)), chk->rec.data.tsn); } if (stcb->asoc.idata_supported == 0) { dchkh = mtod(chk->data, struct sctp_data_chunk *); } else { ndchkh = mtod(chk->data, struct sctp_idata_chunk *); } /* * Put the rest of the things in place now. Size was done earlier in * previous loop prior to padding. */ #ifdef SCTP_ASOCLOG_OF_TSNS SCTP_TCB_LOCK_ASSERT(stcb); if (asoc->tsn_out_at >= SCTP_TSN_LOG_SIZE) { asoc->tsn_out_at = 0; asoc->tsn_out_wrapped = 1; } asoc->out_tsnlog[asoc->tsn_out_at].tsn = chk->rec.data.tsn; asoc->out_tsnlog[asoc->tsn_out_at].strm = chk->rec.data.sid; asoc->out_tsnlog[asoc->tsn_out_at].seq = chk->rec.data.mid; asoc->out_tsnlog[asoc->tsn_out_at].sz = chk->send_size; asoc->out_tsnlog[asoc->tsn_out_at].flgs = chk->rec.data.rcv_flags; asoc->out_tsnlog[asoc->tsn_out_at].stcb = (void *)stcb; asoc->out_tsnlog[asoc->tsn_out_at].in_pos = asoc->tsn_out_at; asoc->out_tsnlog[asoc->tsn_out_at].in_out = 2; asoc->tsn_out_at++; #endif if (stcb->asoc.idata_supported == 0) { dchkh->ch.chunk_type = SCTP_DATA; dchkh->ch.chunk_flags = chk->rec.data.rcv_flags; dchkh->dp.tsn = htonl(chk->rec.data.tsn); dchkh->dp.sid = htons(strq->sid); dchkh->dp.ssn = htons((uint16_t)chk->rec.data.mid); dchkh->dp.ppid = chk->rec.data.ppid; dchkh->ch.chunk_length = htons(chk->send_size); } else { ndchkh->ch.chunk_type = SCTP_IDATA; ndchkh->ch.chunk_flags = chk->rec.data.rcv_flags; ndchkh->dp.tsn = htonl(chk->rec.data.tsn); ndchkh->dp.sid = htons(strq->sid); ndchkh->dp.reserved = htons(0); ndchkh->dp.mid = htonl(chk->rec.data.mid); if (sp->fsn == 0) ndchkh->dp.ppid_fsn.ppid = chk->rec.data.ppid; else ndchkh->dp.ppid_fsn.fsn = htonl(sp->fsn); sp->fsn++; ndchkh->ch.chunk_length = htons(chk->send_size); } /* Now advance the chk->send_size by the actual pad needed. */ if (chk->send_size < SCTP_SIZE32(chk->book_size)) { /* need a pad */ struct mbuf *lm; int pads; pads = SCTP_SIZE32(chk->book_size) - chk->send_size; lm = sctp_pad_lastmbuf(chk->data, pads, chk->last_mbuf); if (lm != NULL) { chk->last_mbuf = lm; chk->pad_inplace = 1; } chk->send_size += pads; } if (PR_SCTP_ENABLED(chk->flags)) { asoc->pr_sctp_cnt++; } if (sp->msg_is_complete && (sp->length == 0) && (sp->sender_all_done)) { /* All done pull and kill the message */ if (sp->put_last_out == 0) { SCTP_PRINTF("Gak, put out entire msg with NO end!-2\n"); SCTP_PRINTF("sender_done:%d len:%d msg_comp:%d put_last_out:%d send_lock:%d\n", sp->sender_all_done, sp->length, sp->msg_is_complete, sp->put_last_out, send_lock_up); } if ((send_lock_up == 0) && (TAILQ_NEXT(sp, next) == NULL)) { SCTP_TCB_SEND_LOCK(stcb); send_lock_up = 1; } atomic_subtract_int(&asoc->stream_queue_cnt, 1); TAILQ_REMOVE(&strq->outqueue, sp, next); stcb->asoc.ss_functions.sctp_ss_remove_from_stream(stcb, asoc, strq, sp, send_lock_up); if ((strq->state == SCTP_STREAM_RESET_PENDING) && (strq->chunks_on_queues == 0) && TAILQ_EMPTY(&strq->outqueue)) { stcb->asoc.trigger_reset = 1; } if (sp->net) { sctp_free_remote_addr(sp->net); sp->net = NULL; } if (sp->data) { sctp_m_freem(sp->data); sp->data = NULL; } sctp_free_a_strmoq(stcb, sp, so_locked); } asoc->chunks_on_out_queue++; strq->chunks_on_queues++; TAILQ_INSERT_TAIL(&asoc->send_queue, chk, sctp_next); asoc->send_queue_cnt++; out_of: if (send_lock_up) { SCTP_TCB_SEND_UNLOCK(stcb); } return (to_move); } static void sctp_fill_outqueue(struct sctp_tcb *stcb, struct sctp_nets *net, int frag_point, int eeor_mode, int *quit_now, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { struct sctp_association *asoc; struct sctp_stream_out *strq; uint32_t space_left, moved, total_moved; int bail, giveup; SCTP_TCB_LOCK_ASSERT(stcb); asoc = &stcb->asoc; total_moved = 0; switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: space_left = net->mtu - SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: space_left = net->mtu - SCTP_MIN_OVERHEAD; break; #endif default: /* TSNH */ space_left = net->mtu; break; } /* Need an allowance for the data chunk header too */ space_left -= SCTP_DATA_CHUNK_OVERHEAD(stcb); /* must make even word boundary */ space_left &= 0xfffffffc; strq = stcb->asoc.ss_functions.sctp_ss_select_stream(stcb, net, asoc); giveup = 0; bail = 0; while ((space_left > 0) && (strq != NULL)) { moved = sctp_move_to_outqueue(stcb, strq, space_left, frag_point, &giveup, eeor_mode, &bail, so_locked); stcb->asoc.ss_functions.sctp_ss_scheduled(stcb, net, asoc, strq, moved); if ((giveup != 0) || (bail != 0)) { break; } strq = stcb->asoc.ss_functions.sctp_ss_select_stream(stcb, net, asoc); total_moved += moved; space_left -= moved; if (space_left >= SCTP_DATA_CHUNK_OVERHEAD(stcb)) { space_left -= SCTP_DATA_CHUNK_OVERHEAD(stcb); } else { space_left = 0; } space_left &= 0xfffffffc; } if (bail != 0) *quit_now = 1; stcb->asoc.ss_functions.sctp_ss_packet_done(stcb, net, asoc); if (total_moved == 0) { if ((stcb->asoc.sctp_cmt_on_off == 0) && (net == stcb->asoc.primary_destination)) { /* ran dry for primary network net */ SCTP_STAT_INCR(sctps_primary_randry); } else if (stcb->asoc.sctp_cmt_on_off > 0) { /* ran dry with CMT on */ SCTP_STAT_INCR(sctps_cmt_randry); } } } void sctp_fix_ecn_echo(struct sctp_association *asoc) { struct sctp_tmit_chunk *chk; TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if (chk->rec.chunk_id.id == SCTP_ECN_ECHO) { chk->sent = SCTP_DATAGRAM_UNSENT; } } } void sctp_move_chunks_from_net(struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_association *asoc; struct sctp_tmit_chunk *chk; struct sctp_stream_queue_pending *sp; unsigned int i; if (net == NULL) { return; } asoc = &stcb->asoc; for (i = 0; i < stcb->asoc.streamoutcnt; i++) { TAILQ_FOREACH(sp, &stcb->asoc.strmout[i].outqueue, next) { if (sp->net == net) { sctp_free_remote_addr(sp->net); sp->net = NULL; } } } TAILQ_FOREACH(chk, &asoc->send_queue, sctp_next) { if (chk->whoTo == net) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = NULL; } } } int sctp_med_chunk_output(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_association *asoc, int *num_out, int *reason_code, int control_only, int from_where, struct timeval *now, int *now_filled, int frag_point, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { /** * Ok this is the generic chunk service queue. we must do the * following: * - Service the stream queue that is next, moving any * message (note I must get a complete message i.e. FIRST/MIDDLE and * LAST to the out queue in one pass) and assigning TSN's. This * only applys though if the peer does not support NDATA. For NDATA * chunks its ok to not send the entire message ;-) * - Check to see if the cwnd/rwnd allows any output, if so we go ahead and * fomulate and send the low level chunks. Making sure to combine * any control in the control chunk queue also. */ struct sctp_nets *net, *start_at, *sack_goes_to = NULL, *old_start_at = NULL; struct mbuf *outchain, *endoutchain; struct sctp_tmit_chunk *chk, *nchk; /* temp arrays for unlinking */ struct sctp_tmit_chunk *data_list[SCTP_MAX_DATA_BUNDLING]; int no_fragmentflg, error; unsigned int max_rwnd_per_dest, max_send_per_dest; int one_chunk, hbflag, skip_data_for_this_net; int asconf, cookie, no_out_cnt; int bundle_at, ctl_cnt, no_data_chunks, eeor_mode; unsigned int mtu, r_mtu, omtu, mx_mtu, to_out; int tsns_sent = 0; uint32_t auth_offset; struct sctp_auth_chunk *auth; uint16_t auth_keyid; int override_ok = 1; int skip_fill_up = 0; int data_auth_reqd = 0; /* * JRS 5/14/07 - Add flag for whether a heartbeat is sent to the * destination. */ int quit_now = 0; *num_out = 0; *reason_code = 0; auth_keyid = stcb->asoc.authinfo.active_keyid; if ((asoc->state & SCTP_STATE_SHUTDOWN_PENDING) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED) || (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXPLICIT_EOR))) { eeor_mode = 1; } else { eeor_mode = 0; } ctl_cnt = no_out_cnt = asconf = cookie = 0; /* * First lets prime the pump. For each destination, if there is room * in the flight size, attempt to pull an MTU's worth out of the * stream queues into the general send_queue */ #ifdef SCTP_AUDITING_ENABLED sctp_audit_log(0xC2, 2); #endif SCTP_TCB_LOCK_ASSERT(stcb); hbflag = 0; if (control_only) no_data_chunks = 1; else no_data_chunks = 0; /* Nothing to possible to send? */ if ((TAILQ_EMPTY(&asoc->control_send_queue) || (asoc->ctrl_queue_cnt == stcb->asoc.ecn_echo_cnt_onq)) && TAILQ_EMPTY(&asoc->asconf_send_queue) && TAILQ_EMPTY(&asoc->send_queue) && sctp_is_there_unsent_data(stcb, so_locked) == 0) { nothing_to_send: *reason_code = 9; return (0); } if (asoc->peers_rwnd == 0) { /* No room in peers rwnd */ *reason_code = 1; if (asoc->total_flight > 0) { /* we are allowed one chunk in flight */ no_data_chunks = 1; } } if (stcb->asoc.ecn_echo_cnt_onq) { /* Record where a sack goes, if any */ if (no_data_chunks && (asoc->ctrl_queue_cnt == stcb->asoc.ecn_echo_cnt_onq)) { /* Nothing but ECNe to send - we don't do that */ goto nothing_to_send; } TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if ((chk->rec.chunk_id.id == SCTP_SELECTIVE_ACK) || (chk->rec.chunk_id.id == SCTP_NR_SELECTIVE_ACK)) { sack_goes_to = chk->whoTo; break; } } } max_rwnd_per_dest = ((asoc->peers_rwnd + asoc->total_flight) / asoc->numnets); if (stcb->sctp_socket) max_send_per_dest = SCTP_SB_LIMIT_SND(stcb->sctp_socket) / asoc->numnets; else max_send_per_dest = 0; if (no_data_chunks == 0) { /* How many non-directed chunks are there? */ TAILQ_FOREACH(chk, &asoc->send_queue, sctp_next) { if (chk->whoTo == NULL) { /* * We already have non-directed chunks on * the queue, no need to do a fill-up. */ skip_fill_up = 1; break; } } } if ((no_data_chunks == 0) && (skip_fill_up == 0) && (!stcb->asoc.ss_functions.sctp_ss_is_empty(stcb, asoc))) { TAILQ_FOREACH(net, &asoc->nets, sctp_next) { /* * This for loop we are in takes in each net, if * its's got space in cwnd and has data sent to it * (when CMT is off) then it calls * sctp_fill_outqueue for the net. This gets data on * the send queue for that network. * * In sctp_fill_outqueue TSN's are assigned and data * is copied out of the stream buffers. Note mostly * copy by reference (we hope). */ net->window_probe = 0; if ((net != stcb->asoc.alternate) && ((net->dest_state & SCTP_ADDR_PF) || (!(net->dest_state & SCTP_ADDR_REACHABLE)) || (net->dest_state & SCTP_ADDR_UNCONFIRMED))) { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, 1, SCTP_CWND_LOG_FILL_OUTQ_CALLED); } continue; } if ((stcb->asoc.cc_functions.sctp_cwnd_new_transmission_begins) && (net->flight_size == 0)) { (*stcb->asoc.cc_functions.sctp_cwnd_new_transmission_begins) (stcb, net); } if (net->flight_size >= net->cwnd) { /* skip this network, no room - can't fill */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, 3, SCTP_CWND_LOG_FILL_OUTQ_CALLED); } continue; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, 4, SCTP_CWND_LOG_FILL_OUTQ_CALLED); } sctp_fill_outqueue(stcb, net, frag_point, eeor_mode, &quit_now, so_locked); if (quit_now) { /* memory alloc failure */ no_data_chunks = 1; break; } } } /* now service each destination and send out what we can for it */ /* Nothing to send? */ if (TAILQ_EMPTY(&asoc->control_send_queue) && TAILQ_EMPTY(&asoc->asconf_send_queue) && TAILQ_EMPTY(&asoc->send_queue)) { *reason_code = 8; return (0); } if (asoc->sctp_cmt_on_off > 0) { /* get the last start point */ start_at = asoc->last_net_cmt_send_started; if (start_at == NULL) { /* null so to beginning */ start_at = TAILQ_FIRST(&asoc->nets); } else { start_at = TAILQ_NEXT(asoc->last_net_cmt_send_started, sctp_next); if (start_at == NULL) { start_at = TAILQ_FIRST(&asoc->nets); } } asoc->last_net_cmt_send_started = start_at; } else { start_at = TAILQ_FIRST(&asoc->nets); } TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if (chk->whoTo == NULL) { if (asoc->alternate) { chk->whoTo = asoc->alternate; } else { chk->whoTo = asoc->primary_destination; } atomic_add_int(&chk->whoTo->ref_count, 1); } } old_start_at = NULL; again_one_more_time: for (net = start_at; net != NULL; net = TAILQ_NEXT(net, sctp_next)) { /* how much can we send? */ /* SCTPDBG("Examine for sending net:%x\n", (uint32_t)net); */ if (old_start_at && (old_start_at == net)) { /* through list ocmpletely. */ break; } tsns_sent = 0xa; if (TAILQ_EMPTY(&asoc->control_send_queue) && TAILQ_EMPTY(&asoc->asconf_send_queue) && (net->flight_size >= net->cwnd)) { /* * Nothing on control or asconf and flight is full, * we can skip even in the CMT case. */ continue; } bundle_at = 0; endoutchain = outchain = NULL; auth = NULL; auth_offset = 0; no_fragmentflg = 1; one_chunk = 0; if (net->dest_state & SCTP_ADDR_UNCONFIRMED) { skip_data_for_this_net = 1; } else { skip_data_for_this_net = 0; } switch (((struct sockaddr *)&net->ro._l_addr)->sa_family) { #ifdef INET case AF_INET: mtu = net->mtu - SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: mtu = net->mtu - SCTP_MIN_OVERHEAD; break; #endif default: /* TSNH */ mtu = net->mtu; break; } mx_mtu = mtu; to_out = 0; if (mtu > asoc->peers_rwnd) { if (asoc->total_flight > 0) { /* We have a packet in flight somewhere */ r_mtu = asoc->peers_rwnd; } else { /* We are always allowed to send one MTU out */ one_chunk = 1; r_mtu = mtu; } } else { r_mtu = mtu; } error = 0; /************************/ /* ASCONF transmission */ /************************/ /* Now first lets go through the asconf queue */ TAILQ_FOREACH_SAFE(chk, &asoc->asconf_send_queue, sctp_next, nchk) { if (chk->rec.chunk_id.id != SCTP_ASCONF) { continue; } if (chk->whoTo == NULL) { if (asoc->alternate == NULL) { if (asoc->primary_destination != net) { break; } } else { if (asoc->alternate != net) { break; } } } else { if (chk->whoTo != net) { break; } } if (chk->data == NULL) { break; } if (chk->sent != SCTP_DATAGRAM_UNSENT && chk->sent != SCTP_DATAGRAM_RESEND) { break; } /* * if no AUTH is yet included and this chunk * requires it, make sure to account for it. We * don't apply the size until the AUTH chunk is * actually added below in case there is no room for * this chunk. NOTE: we overload the use of "omtu" * here */ if ((auth == NULL) && sctp_auth_is_required_chunk(chk->rec.chunk_id.id, stcb->asoc.peer_auth_chunks)) { omtu = sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); } else omtu = 0; /* Here we do NOT factor the r_mtu */ if ((chk->send_size < (int)(mtu - omtu)) || (chk->flags & CHUNK_FLAGS_FRAGMENT_OK)) { /* * We probably should glom the mbuf chain * from the chk->data for control but the * problem is it becomes yet one more level * of tracking to do if for some reason * output fails. Then I have got to * reconstruct the merged control chain.. el * yucko.. for now we take the easy way and * do the copy */ /* * Add an AUTH chunk, if chunk requires it * save the offset into the chain for AUTH */ if ((auth == NULL) && (sctp_auth_is_required_chunk(chk->rec.chunk_id.id, stcb->asoc.peer_auth_chunks))) { outchain = sctp_add_auth_chunk(outchain, &endoutchain, &auth, &auth_offset, stcb, chk->rec.chunk_id.id); SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } outchain = sctp_copy_mbufchain(chk->data, outchain, &endoutchain, (int)chk->rec.chunk_id.can_take_data, chk->send_size, chk->copy_by_ref); if (outchain == NULL) { *reason_code = 8; SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); /* update our MTU size */ if (mtu > (chk->send_size + omtu)) mtu -= (chk->send_size + omtu); else mtu = 0; to_out += (chk->send_size + omtu); /* Do clear IP_DF ? */ if (chk->flags & CHUNK_FLAGS_FRAGMENT_OK) { no_fragmentflg = 0; } if (chk->rec.chunk_id.can_take_data) chk->data = NULL; /* * set hb flag since we can use these for * RTO */ hbflag = 1; asconf = 1; /* * should sysctl this: don't bundle data * with ASCONF since it requires AUTH */ no_data_chunks = 1; chk->sent = SCTP_DATAGRAM_SENT; if (chk->whoTo == NULL) { chk->whoTo = net; atomic_add_int(&net->ref_count, 1); } chk->snd_count++; if (mtu == 0) { /* * Ok we are out of room but we can * output without effecting the * flight size since this little guy * is a control only packet. */ sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, inp, stcb, net); /* * do NOT clear the asconf flag as * it is used to do appropriate * source address selection. */ if (*now_filled == 0) { (void)SCTP_GETTIME_TIMEVAL(now); *now_filled = 1; } net->last_sent_time = *now; hbflag = 0; if ((error = sctp_lowlevel_chunk_output(inp, stcb, net, (struct sockaddr *)&net->ro._l_addr, outchain, auth_offset, auth, stcb->asoc.authinfo.active_keyid, no_fragmentflg, 0, asconf, inp->sctp_lport, stcb->rport, htonl(stcb->asoc.peer_vtag), net->port, NULL, 0, 0, so_locked))) { /* * error, we could not * output */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (from_where == 0) { SCTP_STAT_INCR(sctps_lowlevelerrusr); } if (error == ENOBUFS) { asoc->ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } /* error, could not output */ if (error == EHOSTUNREACH) { /* * Destination went * unreachable * during this send */ sctp_move_chunks_from_net(stcb, net); } *reason_code = 7; break; } else { asoc->ifp_had_enobuf = 0; } /* * increase the number we sent, if a * cookie is sent we don't tell them * any was sent out. */ outchain = endoutchain = NULL; auth = NULL; auth_offset = 0; if (!no_out_cnt) *num_out += ctl_cnt; /* recalc a clean slate and setup */ switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: mtu = net->mtu - SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: mtu = net->mtu - SCTP_MIN_OVERHEAD; break; #endif default: /* TSNH */ mtu = net->mtu; break; } to_out = 0; no_fragmentflg = 1; } } } if (error != 0) { /* try next net */ continue; } /************************/ /* Control transmission */ /************************/ /* Now first lets go through the control queue */ TAILQ_FOREACH_SAFE(chk, &asoc->control_send_queue, sctp_next, nchk) { if ((sack_goes_to) && (chk->rec.chunk_id.id == SCTP_ECN_ECHO) && (chk->whoTo != sack_goes_to)) { /* * if we have a sack in queue, and we are * looking at an ecn echo that is NOT queued * to where the sack is going.. */ if (chk->whoTo == net) { /* * Don't transmit it to where its * going (current net) */ continue; } else if (sack_goes_to == net) { /* * But do transmit it to this * address */ goto skip_net_check; } } if (chk->whoTo == NULL) { if (asoc->alternate == NULL) { if (asoc->primary_destination != net) { continue; } } else { if (asoc->alternate != net) { continue; } } } else { if (chk->whoTo != net) { continue; } } skip_net_check: if (chk->data == NULL) { continue; } if (chk->sent != SCTP_DATAGRAM_UNSENT) { /* * It must be unsent. Cookies and ASCONF's * hang around but there timers will force * when marked for resend. */ continue; } /* * if no AUTH is yet included and this chunk * requires it, make sure to account for it. We * don't apply the size until the AUTH chunk is * actually added below in case there is no room for * this chunk. NOTE: we overload the use of "omtu" * here */ if ((auth == NULL) && sctp_auth_is_required_chunk(chk->rec.chunk_id.id, stcb->asoc.peer_auth_chunks)) { omtu = sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); } else omtu = 0; /* Here we do NOT factor the r_mtu */ if ((chk->send_size <= (int)(mtu - omtu)) || (chk->flags & CHUNK_FLAGS_FRAGMENT_OK)) { /* * We probably should glom the mbuf chain * from the chk->data for control but the * problem is it becomes yet one more level * of tracking to do if for some reason * output fails. Then I have got to * reconstruct the merged control chain.. el * yucko.. for now we take the easy way and * do the copy */ /* * Add an AUTH chunk, if chunk requires it * save the offset into the chain for AUTH */ if ((auth == NULL) && (sctp_auth_is_required_chunk(chk->rec.chunk_id.id, stcb->asoc.peer_auth_chunks))) { outchain = sctp_add_auth_chunk(outchain, &endoutchain, &auth, &auth_offset, stcb, chk->rec.chunk_id.id); SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } outchain = sctp_copy_mbufchain(chk->data, outchain, &endoutchain, (int)chk->rec.chunk_id.can_take_data, chk->send_size, chk->copy_by_ref); if (outchain == NULL) { *reason_code = 8; SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); /* update our MTU size */ if (mtu > (chk->send_size + omtu)) mtu -= (chk->send_size + omtu); else mtu = 0; to_out += (chk->send_size + omtu); /* Do clear IP_DF ? */ if (chk->flags & CHUNK_FLAGS_FRAGMENT_OK) { no_fragmentflg = 0; } if (chk->rec.chunk_id.can_take_data) chk->data = NULL; /* Mark things to be removed, if needed */ if ((chk->rec.chunk_id.id == SCTP_SELECTIVE_ACK) || (chk->rec.chunk_id.id == SCTP_NR_SELECTIVE_ACK) || /* EY */ (chk->rec.chunk_id.id == SCTP_HEARTBEAT_REQUEST) || (chk->rec.chunk_id.id == SCTP_HEARTBEAT_ACK) || (chk->rec.chunk_id.id == SCTP_SHUTDOWN) || (chk->rec.chunk_id.id == SCTP_SHUTDOWN_ACK) || (chk->rec.chunk_id.id == SCTP_OPERATION_ERROR) || (chk->rec.chunk_id.id == SCTP_COOKIE_ACK) || (chk->rec.chunk_id.id == SCTP_ECN_CWR) || (chk->rec.chunk_id.id == SCTP_PACKET_DROPPED) || (chk->rec.chunk_id.id == SCTP_ASCONF_ACK)) { if (chk->rec.chunk_id.id == SCTP_HEARTBEAT_REQUEST) { hbflag = 1; } /* remove these chunks at the end */ if ((chk->rec.chunk_id.id == SCTP_SELECTIVE_ACK) || (chk->rec.chunk_id.id == SCTP_NR_SELECTIVE_ACK)) { /* turn off the timer */ if (SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) { sctp_timer_stop(SCTP_TIMER_TYPE_RECV, inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_1); } } ctl_cnt++; } else { /* * Other chunks, since they have * timers running (i.e. COOKIE) we * just "trust" that it gets sent or * retransmitted. */ ctl_cnt++; if (chk->rec.chunk_id.id == SCTP_COOKIE_ECHO) { cookie = 1; no_out_cnt = 1; } else if (chk->rec.chunk_id.id == SCTP_ECN_ECHO) { /* * Increment ecne send count * here this means we may be * over-zealous in our * counting if the send * fails, but its the best * place to do it (we used * to do it in the queue of * the chunk, but that did * not tell how many times * it was sent. */ SCTP_STAT_INCR(sctps_sendecne); } chk->sent = SCTP_DATAGRAM_SENT; if (chk->whoTo == NULL) { chk->whoTo = net; atomic_add_int(&net->ref_count, 1); } chk->snd_count++; } if (mtu == 0) { /* * Ok we are out of room but we can * output without effecting the * flight size since this little guy * is a control only packet. */ if (asconf) { sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, inp, stcb, net); /* * do NOT clear the asconf * flag as it is used to do * appropriate source * address selection. */ } if (cookie) { sctp_timer_start(SCTP_TIMER_TYPE_COOKIE, inp, stcb, net); cookie = 0; } /* Only HB or ASCONF advances time */ if (hbflag) { if (*now_filled == 0) { (void)SCTP_GETTIME_TIMEVAL(now); *now_filled = 1; } net->last_sent_time = *now; hbflag = 0; } if ((error = sctp_lowlevel_chunk_output(inp, stcb, net, (struct sockaddr *)&net->ro._l_addr, outchain, auth_offset, auth, stcb->asoc.authinfo.active_keyid, no_fragmentflg, 0, asconf, inp->sctp_lport, stcb->rport, htonl(stcb->asoc.peer_vtag), net->port, NULL, 0, 0, so_locked))) { /* * error, we could not * output */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (from_where == 0) { SCTP_STAT_INCR(sctps_lowlevelerrusr); } if (error == ENOBUFS) { asoc->ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } if (error == EHOSTUNREACH) { /* * Destination went * unreachable * during this send */ sctp_move_chunks_from_net(stcb, net); } *reason_code = 7; break; } else { asoc->ifp_had_enobuf = 0; } /* * increase the number we sent, if a * cookie is sent we don't tell them * any was sent out. */ outchain = endoutchain = NULL; auth = NULL; auth_offset = 0; if (!no_out_cnt) *num_out += ctl_cnt; /* recalc a clean slate and setup */ switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: mtu = net->mtu - SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: mtu = net->mtu - SCTP_MIN_OVERHEAD; break; #endif default: /* TSNH */ mtu = net->mtu; break; } to_out = 0; no_fragmentflg = 1; } } } if (error != 0) { /* try next net */ continue; } /* JRI: if dest is in PF state, do not send data to it */ if ((asoc->sctp_cmt_on_off > 0) && (net != stcb->asoc.alternate) && (net->dest_state & SCTP_ADDR_PF)) { goto no_data_fill; } if (net->flight_size >= net->cwnd) { goto no_data_fill; } if ((asoc->sctp_cmt_on_off > 0) && (SCTP_BASE_SYSCTL(sctp_buffer_splitting) & SCTP_RECV_BUFFER_SPLITTING) && (net->flight_size > max_rwnd_per_dest)) { goto no_data_fill; } /* * We need a specific accounting for the usage of the send * buffer. We also need to check the number of messages per * net. For now, this is better than nothing and it disabled * by default... */ if ((asoc->sctp_cmt_on_off > 0) && (SCTP_BASE_SYSCTL(sctp_buffer_splitting) & SCTP_SEND_BUFFER_SPLITTING) && (max_send_per_dest > 0) && (net->flight_size > max_send_per_dest)) { goto no_data_fill; } /*********************/ /* Data transmission */ /*********************/ /* * if AUTH for DATA is required and no AUTH has been added * yet, account for this in the mtu now... if no data can be * bundled, this adjustment won't matter anyways since the * packet will be going out... */ data_auth_reqd = sctp_auth_is_required_chunk(SCTP_DATA, stcb->asoc.peer_auth_chunks); if (data_auth_reqd && (auth == NULL)) { mtu -= sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); } /* now lets add any data within the MTU constraints */ switch (((struct sockaddr *)&net->ro._l_addr)->sa_family) { #ifdef INET case AF_INET: if (net->mtu > SCTP_MIN_V4_OVERHEAD) omtu = net->mtu - SCTP_MIN_V4_OVERHEAD; else omtu = 0; break; #endif #ifdef INET6 case AF_INET6: if (net->mtu > SCTP_MIN_OVERHEAD) omtu = net->mtu - SCTP_MIN_OVERHEAD; else omtu = 0; break; #endif default: /* TSNH */ omtu = 0; break; } if ((((SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED)) && (skip_data_for_this_net == 0)) || (cookie)) { TAILQ_FOREACH_SAFE(chk, &asoc->send_queue, sctp_next, nchk) { if (no_data_chunks) { /* let only control go out */ *reason_code = 1; break; } if (net->flight_size >= net->cwnd) { /* skip this net, no room for data */ *reason_code = 2; break; } if ((chk->whoTo != NULL) && (chk->whoTo != net)) { /* Don't send the chunk on this net */ continue; } if (asoc->sctp_cmt_on_off == 0) { if ((asoc->alternate) && (asoc->alternate != net) && (chk->whoTo == NULL)) { continue; } else if ((net != asoc->primary_destination) && (asoc->alternate == NULL) && (chk->whoTo == NULL)) { continue; } } if ((chk->send_size > omtu) && ((chk->flags & CHUNK_FLAGS_FRAGMENT_OK) == 0)) { /*- * strange, we have a chunk that is * to big for its destination and * yet no fragment ok flag. * Something went wrong when the * PMTU changed...we did not mark * this chunk for some reason?? I * will fix it here by letting IP * fragment it for now and printing * a warning. This really should not * happen ... */ SCTP_PRINTF("Warning chunk of %d bytes > mtu:%d and yet PMTU disc missed\n", chk->send_size, mtu); chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; } if (SCTP_BASE_SYSCTL(sctp_enable_sack_immediately) && (asoc->state & SCTP_STATE_SHUTDOWN_PENDING)) { struct sctp_data_chunk *dchkh; dchkh = mtod(chk->data, struct sctp_data_chunk *); dchkh->ch.chunk_flags |= SCTP_DATA_SACK_IMMEDIATELY; } if (((chk->send_size <= mtu) && (chk->send_size <= r_mtu)) || ((chk->flags & CHUNK_FLAGS_FRAGMENT_OK) && (chk->send_size <= asoc->peers_rwnd))) { /* ok we will add this one */ /* * Add an AUTH chunk, if chunk * requires it, save the offset into * the chain for AUTH */ if (data_auth_reqd) { if (auth == NULL) { outchain = sctp_add_auth_chunk(outchain, &endoutchain, &auth, &auth_offset, stcb, SCTP_DATA); auth_keyid = chk->auth_keyid; override_ok = 0; SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } else if (override_ok) { /* * use this data's * keyid */ auth_keyid = chk->auth_keyid; override_ok = 0; } else if (auth_keyid != chk->auth_keyid) { /* * different keyid, * so done bundling */ break; } } outchain = sctp_copy_mbufchain(chk->data, outchain, &endoutchain, 0, chk->send_size, chk->copy_by_ref); if (outchain == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "No memory?\n"); if (!SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) { sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, net); } *reason_code = 3; SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } /* upate our MTU size */ /* Do clear IP_DF ? */ if (chk->flags & CHUNK_FLAGS_FRAGMENT_OK) { no_fragmentflg = 0; } /* unsigned subtraction of mtu */ if (mtu > chk->send_size) mtu -= chk->send_size; else mtu = 0; /* unsigned subtraction of r_mtu */ if (r_mtu > chk->send_size) r_mtu -= chk->send_size; else r_mtu = 0; to_out += chk->send_size; if ((to_out > mx_mtu) && no_fragmentflg) { #ifdef INVARIANTS panic("Exceeding mtu of %d out size is %d", mx_mtu, to_out); #else SCTP_PRINTF("Exceeding mtu of %d out size is %d\n", mx_mtu, to_out); #endif } chk->window_probe = 0; data_list[bundle_at++] = chk; if (bundle_at >= SCTP_MAX_DATA_BUNDLING) { break; } if (chk->sent == SCTP_DATAGRAM_UNSENT) { if ((chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED) == 0) { SCTP_STAT_INCR_COUNTER64(sctps_outorderchunks); } else { SCTP_STAT_INCR_COUNTER64(sctps_outunorderchunks); } if (((chk->rec.data.rcv_flags & SCTP_DATA_LAST_FRAG) == SCTP_DATA_LAST_FRAG) && ((chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) == 0)) /* * Count number of * user msg's that * were fragmented * we do this by * counting when we * see a LAST * fragment only. */ SCTP_STAT_INCR_COUNTER64(sctps_fragusrmsgs); } if ((mtu == 0) || (r_mtu == 0) || (one_chunk)) { if ((one_chunk) && (stcb->asoc.total_flight == 0)) { data_list[0]->window_probe = 1; net->window_probe = 1; } break; } } else { /* * Must be sent in order of the * TSN's (on a network) */ break; } } /* for (chunk gather loop for this net) */ } /* if asoc.state OPEN */ no_data_fill: /* Is there something to send for this destination? */ if (outchain) { /* We may need to start a control timer or two */ if (asconf) { sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, inp, stcb, net); /* * do NOT clear the asconf flag as it is * used to do appropriate source address * selection. */ } if (cookie) { sctp_timer_start(SCTP_TIMER_TYPE_COOKIE, inp, stcb, net); cookie = 0; } /* must start a send timer if data is being sent */ if (bundle_at && (!SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer))) { /* * no timer running on this destination * restart it. */ sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, net); } if (bundle_at || hbflag) { /* For data/asconf and hb set time */ if (*now_filled == 0) { (void)SCTP_GETTIME_TIMEVAL(now); *now_filled = 1; } net->last_sent_time = *now; } /* Now send it, if there is anything to send :> */ if ((error = sctp_lowlevel_chunk_output(inp, stcb, net, (struct sockaddr *)&net->ro._l_addr, outchain, auth_offset, auth, auth_keyid, no_fragmentflg, bundle_at, asconf, inp->sctp_lport, stcb->rport, htonl(stcb->asoc.peer_vtag), net->port, NULL, 0, 0, so_locked))) { /* error, we could not output */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (from_where == 0) { SCTP_STAT_INCR(sctps_lowlevelerrusr); } if (error == ENOBUFS) { asoc->ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } if (error == EHOSTUNREACH) { /* * Destination went unreachable * during this send */ sctp_move_chunks_from_net(stcb, net); } *reason_code = 6; /*- * I add this line to be paranoid. As far as * I can tell the continue, takes us back to * the top of the for, but just to make sure * I will reset these again here. */ ctl_cnt = bundle_at = 0; continue; /* This takes us back to the * for() for the nets. */ } else { asoc->ifp_had_enobuf = 0; } endoutchain = NULL; auth = NULL; auth_offset = 0; if (!no_out_cnt) { *num_out += (ctl_cnt + bundle_at); } if (bundle_at) { /* setup for a RTO measurement */ tsns_sent = data_list[0]->rec.data.tsn; /* fill time if not already filled */ if (*now_filled == 0) { (void)SCTP_GETTIME_TIMEVAL(&asoc->time_last_sent); *now_filled = 1; *now = asoc->time_last_sent; } else { asoc->time_last_sent = *now; } if (net->rto_needed) { data_list[0]->do_rtt = 1; net->rto_needed = 0; } SCTP_STAT_INCR_BY(sctps_senddata, bundle_at); sctp_clean_up_datalist(stcb, asoc, data_list, bundle_at, net); } if (one_chunk) { break; } } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, tsns_sent, SCTP_CWND_LOG_FROM_SEND); } } if (old_start_at == NULL) { old_start_at = start_at; start_at = TAILQ_FIRST(&asoc->nets); if (old_start_at) goto again_one_more_time; } /* * At the end there should be no NON timed chunks hanging on this * queue. */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, *num_out, SCTP_CWND_LOG_FROM_SEND); } if ((*num_out == 0) && (*reason_code == 0)) { *reason_code = 4; } else { *reason_code = 5; } sctp_clean_up_ctl(stcb, asoc, so_locked); return (0); } void sctp_queue_op_err(struct sctp_tcb *stcb, struct mbuf *op_err) { /*- * Prepend a OPERATIONAL_ERROR chunk header and put on the end of * the control chunk queue. */ struct sctp_chunkhdr *hdr; struct sctp_tmit_chunk *chk; struct mbuf *mat, *last_mbuf; uint32_t chunk_length; uint16_t padding_length; SCTP_TCB_LOCK_ASSERT(stcb); SCTP_BUF_PREPEND(op_err, sizeof(struct sctp_chunkhdr), M_NOWAIT); if (op_err == NULL) { return; } last_mbuf = NULL; chunk_length = 0; for (mat = op_err; mat != NULL; mat = SCTP_BUF_NEXT(mat)) { chunk_length += SCTP_BUF_LEN(mat); if (SCTP_BUF_NEXT(mat) == NULL) { last_mbuf = mat; } } if (chunk_length > SCTP_MAX_CHUNK_LENGTH) { sctp_m_freem(op_err); return; } padding_length = chunk_length % 4; if (padding_length != 0) { padding_length = 4 - padding_length; } if (padding_length != 0) { if (sctp_add_pad_tombuf(last_mbuf, padding_length) == NULL) { sctp_m_freem(op_err); return; } } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(op_err); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_OPERATION_ERROR; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->send_size = (uint16_t)chunk_length; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = op_err; chk->whoTo = NULL; hdr = mtod(op_err, struct sctp_chunkhdr *); hdr->chunk_type = SCTP_OPERATION_ERROR; hdr->chunk_flags = 0; hdr->chunk_length = htons(chk->send_size); TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; } int sctp_send_cookie_echo(struct mbuf *m, int offset, int limit, struct sctp_tcb *stcb, struct sctp_nets *net) { /*- * pull out the cookie and put it at the front of the control chunk * queue. */ int at; struct mbuf *cookie; struct sctp_paramhdr param, *phdr; struct sctp_chunkhdr *hdr; struct sctp_tmit_chunk *chk; uint16_t ptype, plen; SCTP_TCB_LOCK_ASSERT(stcb); /* First find the cookie in the param area */ cookie = NULL; at = offset + sizeof(struct sctp_init_chunk); for (;;) { phdr = sctp_get_next_param(m, at, ¶m, sizeof(param)); if (phdr == NULL) { return (-3); } ptype = ntohs(phdr->param_type); plen = ntohs(phdr->param_length); if (plen < sizeof(struct sctp_paramhdr)) { return (-6); } if (ptype == SCTP_STATE_COOKIE) { int pad; /* found the cookie */ if (at + plen > limit) { return (-7); } cookie = SCTP_M_COPYM(m, at, plen, M_NOWAIT); if (cookie == NULL) { /* No memory */ return (-2); } if ((pad = (plen % 4)) > 0) { pad = 4 - pad; } if (pad > 0) { if (sctp_pad_lastmbuf(cookie, pad, NULL) == NULL) { return (-8); } } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(cookie, SCTP_MBUF_ICOPY); } #endif break; } at += SCTP_SIZE32(plen); } /* ok, we got the cookie lets change it into a cookie echo chunk */ /* first the change from param to cookie */ hdr = mtod(cookie, struct sctp_chunkhdr *); hdr->chunk_type = SCTP_COOKIE_ECHO; hdr->chunk_flags = 0; /* get the chunk stuff now and place it in the FRONT of the queue */ sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(cookie); return (-5); } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_COOKIE_ECHO; chk->rec.chunk_id.can_take_data = 0; chk->flags = CHUNK_FLAGS_FRAGMENT_OK; chk->send_size = SCTP_SIZE32(plen); chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = cookie; chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); TAILQ_INSERT_HEAD(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; return (0); } void sctp_send_heartbeat_ack(struct sctp_tcb *stcb, struct mbuf *m, int offset, int chk_length, struct sctp_nets *net) { /* * take a HB request and make it into a HB ack and send it. */ struct mbuf *outchain; struct sctp_chunkhdr *chdr; struct sctp_tmit_chunk *chk; if (net == NULL) /* must have a net pointer */ return; outchain = SCTP_M_COPYM(m, offset, chk_length, M_NOWAIT); if (outchain == NULL) { /* gak out of memory */ return; } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(outchain, SCTP_MBUF_ICOPY); } #endif chdr = mtod(outchain, struct sctp_chunkhdr *); chdr->chunk_type = SCTP_HEARTBEAT_ACK; chdr->chunk_flags = 0; if (chk_length % 4 != 0) { sctp_pad_lastmbuf(outchain, 4 - (chk_length % 4), NULL); } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(outchain); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_HEARTBEAT_ACK; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->send_size = chk_length; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = outchain; chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; } void sctp_send_cookie_ack(struct sctp_tcb *stcb) { /* formulate and queue a cookie-ack back to sender */ struct mbuf *cookie_ack; struct sctp_chunkhdr *hdr; struct sctp_tmit_chunk *chk; SCTP_TCB_LOCK_ASSERT(stcb); cookie_ack = sctp_get_mbuf_for_msg(sizeof(struct sctp_chunkhdr), 0, M_NOWAIT, 1, MT_HEADER); if (cookie_ack == NULL) { /* no mbuf's */ return; } SCTP_BUF_RESV_UF(cookie_ack, SCTP_MIN_OVERHEAD); sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(cookie_ack); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_COOKIE_ACK; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->send_size = sizeof(struct sctp_chunkhdr); chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = cookie_ack; if (chk->asoc->last_control_chunk_from != NULL) { chk->whoTo = chk->asoc->last_control_chunk_from; atomic_add_int(&chk->whoTo->ref_count, 1); } else { chk->whoTo = NULL; } hdr = mtod(cookie_ack, struct sctp_chunkhdr *); hdr->chunk_type = SCTP_COOKIE_ACK; hdr->chunk_flags = 0; hdr->chunk_length = htons(chk->send_size); SCTP_BUF_LEN(cookie_ack) = chk->send_size; TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; return; } void sctp_send_shutdown_ack(struct sctp_tcb *stcb, struct sctp_nets *net) { /* formulate and queue a SHUTDOWN-ACK back to the sender */ struct mbuf *m_shutdown_ack; struct sctp_shutdown_ack_chunk *ack_cp; struct sctp_tmit_chunk *chk; m_shutdown_ack = sctp_get_mbuf_for_msg(sizeof(struct sctp_shutdown_ack_chunk), 0, M_NOWAIT, 1, MT_HEADER); if (m_shutdown_ack == NULL) { /* no mbuf's */ return; } SCTP_BUF_RESV_UF(m_shutdown_ack, SCTP_MIN_OVERHEAD); sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(m_shutdown_ack); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_SHUTDOWN_ACK; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->send_size = sizeof(struct sctp_chunkhdr); chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = m_shutdown_ack; chk->whoTo = net; if (chk->whoTo) { atomic_add_int(&chk->whoTo->ref_count, 1); } ack_cp = mtod(m_shutdown_ack, struct sctp_shutdown_ack_chunk *); ack_cp->ch.chunk_type = SCTP_SHUTDOWN_ACK; ack_cp->ch.chunk_flags = 0; ack_cp->ch.chunk_length = htons(chk->send_size); SCTP_BUF_LEN(m_shutdown_ack) = chk->send_size; TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; return; } void sctp_send_shutdown(struct sctp_tcb *stcb, struct sctp_nets *net) { /* formulate and queue a SHUTDOWN to the sender */ struct mbuf *m_shutdown; struct sctp_shutdown_chunk *shutdown_cp; struct sctp_tmit_chunk *chk; TAILQ_FOREACH(chk, &stcb->asoc.control_send_queue, sctp_next) { if (chk->rec.chunk_id.id == SCTP_SHUTDOWN) { /* We already have a SHUTDOWN queued. Reuse it. */ if (chk->whoTo) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = NULL; } break; } } if (chk == NULL) { m_shutdown = sctp_get_mbuf_for_msg(sizeof(struct sctp_shutdown_chunk), 0, M_NOWAIT, 1, MT_HEADER); if (m_shutdown == NULL) { /* no mbuf's */ return; } SCTP_BUF_RESV_UF(m_shutdown, SCTP_MIN_OVERHEAD); sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(m_shutdown); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_SHUTDOWN; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->send_size = sizeof(struct sctp_shutdown_chunk); chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->data = m_shutdown; chk->whoTo = net; if (chk->whoTo) { atomic_add_int(&chk->whoTo->ref_count, 1); } shutdown_cp = mtod(m_shutdown, struct sctp_shutdown_chunk *); shutdown_cp->ch.chunk_type = SCTP_SHUTDOWN; shutdown_cp->ch.chunk_flags = 0; shutdown_cp->ch.chunk_length = htons(chk->send_size); shutdown_cp->cumulative_tsn_ack = htonl(stcb->asoc.cumulative_tsn); SCTP_BUF_LEN(m_shutdown) = chk->send_size; TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; } else { TAILQ_REMOVE(&stcb->asoc.control_send_queue, chk, sctp_next); chk->whoTo = net; if (chk->whoTo) { atomic_add_int(&chk->whoTo->ref_count, 1); } shutdown_cp = mtod(chk->data, struct sctp_shutdown_chunk *); shutdown_cp->cumulative_tsn_ack = htonl(stcb->asoc.cumulative_tsn); TAILQ_INSERT_TAIL(&stcb->asoc.control_send_queue, chk, sctp_next); } return; } void sctp_send_asconf(struct sctp_tcb *stcb, struct sctp_nets *net, int addr_locked) { /* * formulate and queue an ASCONF to the peer. ASCONF parameters * should be queued on the assoc queue. */ struct sctp_tmit_chunk *chk; struct mbuf *m_asconf; int len; SCTP_TCB_LOCK_ASSERT(stcb); if ((!TAILQ_EMPTY(&stcb->asoc.asconf_send_queue)) && (!sctp_is_feature_on(stcb->sctp_ep, SCTP_PCB_FLAGS_MULTIPLE_ASCONFS))) { /* can't send a new one if there is one in flight already */ return; } /* compose an ASCONF chunk, maximum length is PMTU */ m_asconf = sctp_compose_asconf(stcb, &len, addr_locked); if (m_asconf == NULL) { return; } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ sctp_m_freem(m_asconf); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_ASCONF; chk->rec.chunk_id.can_take_data = 0; chk->flags = CHUNK_FLAGS_FRAGMENT_OK; chk->data = m_asconf; chk->send_size = len; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; chk->whoTo = net; if (chk->whoTo) { atomic_add_int(&chk->whoTo->ref_count, 1); } TAILQ_INSERT_TAIL(&chk->asoc->asconf_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; return; } void sctp_send_asconf_ack(struct sctp_tcb *stcb) { /* * formulate and queue a asconf-ack back to sender. the asconf-ack * must be stored in the tcb. */ struct sctp_tmit_chunk *chk; struct sctp_asconf_ack *ack, *latest_ack; struct mbuf *m_ack; struct sctp_nets *net = NULL; SCTP_TCB_LOCK_ASSERT(stcb); /* Get the latest ASCONF-ACK */ latest_ack = TAILQ_LAST(&stcb->asoc.asconf_ack_sent, sctp_asconf_ackhead); if (latest_ack == NULL) { return; } if (latest_ack->last_sent_to != NULL && latest_ack->last_sent_to == stcb->asoc.last_control_chunk_from) { /* we're doing a retransmission */ net = sctp_find_alternate_net(stcb, stcb->asoc.last_control_chunk_from, 0); if (net == NULL) { /* no alternate */ if (stcb->asoc.last_control_chunk_from == NULL) { if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } } else { net = stcb->asoc.last_control_chunk_from; } } } else { /* normal case */ if (stcb->asoc.last_control_chunk_from == NULL) { if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } } else { net = stcb->asoc.last_control_chunk_from; } } latest_ack->last_sent_to = net; TAILQ_FOREACH(ack, &stcb->asoc.asconf_ack_sent, next) { if (ack->data == NULL) { continue; } /* copy the asconf_ack */ m_ack = SCTP_M_COPYM(ack->data, 0, M_COPYALL, M_NOWAIT); if (m_ack == NULL) { /* couldn't copy it */ return; } #ifdef SCTP_MBUF_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) { sctp_log_mbc(m_ack, SCTP_MBUF_ICOPY); } #endif sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { /* no memory */ if (m_ack) sctp_m_freem(m_ack); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_ASCONF_ACK; chk->rec.chunk_id.can_take_data = 1; chk->flags = CHUNK_FLAGS_FRAGMENT_OK; chk->whoTo = net; if (chk->whoTo) { atomic_add_int(&chk->whoTo->ref_count, 1); } chk->data = m_ack; chk->send_size = ack->len; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->asoc = &stcb->asoc; TAILQ_INSERT_TAIL(&chk->asoc->control_send_queue, chk, sctp_next); chk->asoc->ctrl_queue_cnt++; } return; } static int sctp_chunk_retransmission(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_association *asoc, int *cnt_out, struct timeval *now, int *now_filled, int *fr_done, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { /*- * send out one MTU of retransmission. If fast_retransmit is * happening we ignore the cwnd. Otherwise we obey the cwnd and * rwnd. For a Cookie or Asconf in the control chunk queue we * retransmit them by themselves. * * For data chunks we will pick out the lowest TSN's in the sent_queue * marked for resend and bundle them all together (up to a MTU of * destination). The address to send to should have been * selected/changed where the retransmission was marked (i.e. in FR * or t3-timeout routines). */ struct sctp_tmit_chunk *data_list[SCTP_MAX_DATA_BUNDLING]; struct sctp_tmit_chunk *chk, *fwd; struct mbuf *m, *endofchain; struct sctp_nets *net = NULL; uint32_t tsns_sent = 0; int no_fragmentflg, bundle_at, cnt_thru; unsigned int mtu; int error, i, one_chunk, fwd_tsn, ctl_cnt, tmr_started; struct sctp_auth_chunk *auth = NULL; uint32_t auth_offset = 0; uint16_t auth_keyid; int override_ok = 1; int data_auth_reqd = 0; uint32_t dmtu = 0; SCTP_TCB_LOCK_ASSERT(stcb); tmr_started = ctl_cnt = bundle_at = error = 0; no_fragmentflg = 1; fwd_tsn = 0; *cnt_out = 0; fwd = NULL; endofchain = m = NULL; auth_keyid = stcb->asoc.authinfo.active_keyid; #ifdef SCTP_AUDITING_ENABLED sctp_audit_log(0xC3, 1); #endif if ((TAILQ_EMPTY(&asoc->sent_queue)) && (TAILQ_EMPTY(&asoc->control_send_queue))) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "SCTP hits empty queue with cnt set to %d?\n", asoc->sent_queue_retran_cnt); asoc->sent_queue_cnt = 0; asoc->sent_queue_cnt_removeable = 0; /* send back 0/0 so we enter normal transmission */ *cnt_out = 0; return (0); } TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if ((chk->rec.chunk_id.id == SCTP_COOKIE_ECHO) || (chk->rec.chunk_id.id == SCTP_STREAM_RESET) || (chk->rec.chunk_id.id == SCTP_FORWARD_CUM_TSN)) { if (chk->sent != SCTP_DATAGRAM_RESEND) { continue; } if (chk->rec.chunk_id.id == SCTP_STREAM_RESET) { if (chk != asoc->str_reset) { /* * not eligible for retran if its * not ours */ continue; } } ctl_cnt++; if (chk->rec.chunk_id.id == SCTP_FORWARD_CUM_TSN) { fwd_tsn = 1; } /* * Add an AUTH chunk, if chunk requires it save the * offset into the chain for AUTH */ if ((auth == NULL) && (sctp_auth_is_required_chunk(chk->rec.chunk_id.id, stcb->asoc.peer_auth_chunks))) { m = sctp_add_auth_chunk(m, &endofchain, &auth, &auth_offset, stcb, chk->rec.chunk_id.id); SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } m = sctp_copy_mbufchain(chk->data, m, &endofchain, 0, chk->send_size, chk->copy_by_ref); break; } } one_chunk = 0; cnt_thru = 0; /* do we have control chunks to retransmit? */ if (m != NULL) { /* Start a timer no matter if we succeed or fail */ if (chk->rec.chunk_id.id == SCTP_COOKIE_ECHO) { sctp_timer_start(SCTP_TIMER_TYPE_COOKIE, inp, stcb, chk->whoTo); } else if (chk->rec.chunk_id.id == SCTP_ASCONF) sctp_timer_start(SCTP_TIMER_TYPE_ASCONF, inp, stcb, chk->whoTo); chk->snd_count++; /* update our count */ if ((error = sctp_lowlevel_chunk_output(inp, stcb, chk->whoTo, (struct sockaddr *)&chk->whoTo->ro._l_addr, m, auth_offset, auth, stcb->asoc.authinfo.active_keyid, no_fragmentflg, 0, 0, inp->sctp_lport, stcb->rport, htonl(stcb->asoc.peer_vtag), chk->whoTo->port, NULL, 0, 0, so_locked))) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (error == ENOBUFS) { asoc->ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } return (error); } else { asoc->ifp_had_enobuf = 0; } endofchain = NULL; auth = NULL; auth_offset = 0; /* * We don't want to mark the net->sent time here since this * we use this for HB and retrans cannot measure RTT */ /* (void)SCTP_GETTIME_TIMEVAL(&chk->whoTo->last_sent_time); */ *cnt_out += 1; chk->sent = SCTP_DATAGRAM_SENT; sctp_ucount_decr(stcb->asoc.sent_queue_retran_cnt); if (fwd_tsn == 0) { return (0); } else { /* Clean up the fwd-tsn list */ sctp_clean_up_ctl(stcb, asoc, so_locked); return (0); } } /* * Ok, it is just data retransmission we need to do or that and a * fwd-tsn with it all. */ if (TAILQ_EMPTY(&asoc->sent_queue)) { return (SCTP_RETRAN_DONE); } if ((SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED) || (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_WAIT)) { /* not yet open, resend the cookie and that is it */ return (1); } #ifdef SCTP_AUDITING_ENABLED sctp_auditing(20, inp, stcb, NULL); #endif data_auth_reqd = sctp_auth_is_required_chunk(SCTP_DATA, stcb->asoc.peer_auth_chunks); TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) { if (chk->sent != SCTP_DATAGRAM_RESEND) { /* No, not sent to this net or not ready for rtx */ continue; } if (chk->data == NULL) { SCTP_PRINTF("TSN:%x chk->snd_count:%d chk->sent:%d can't retran - no data\n", chk->rec.data.tsn, chk->snd_count, chk->sent); continue; } if ((SCTP_BASE_SYSCTL(sctp_max_retran_chunk)) && (chk->snd_count >= SCTP_BASE_SYSCTL(sctp_max_retran_chunk))) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; snprintf(msg, sizeof(msg), "TSN %8.8x retransmitted %d times, giving up", chk->rec.data.tsn, chk->snd_count); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); atomic_add_int(&stcb->asoc.refcnt, 1); sctp_abort_an_association(stcb->sctp_ep, stcb, op_err, so_locked); SCTP_TCB_LOCK(stcb); atomic_subtract_int(&stcb->asoc.refcnt, 1); return (SCTP_RETRAN_EXIT); } /* pick up the net */ net = chk->whoTo; switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: mtu = net->mtu - SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: mtu = net->mtu - SCTP_MIN_OVERHEAD; break; #endif default: /* TSNH */ mtu = net->mtu; break; } if ((asoc->peers_rwnd < mtu) && (asoc->total_flight > 0)) { /* No room in peers rwnd */ uint32_t tsn; tsn = asoc->last_acked_seq + 1; if (tsn == chk->rec.data.tsn) { /* * we make a special exception for this * case. The peer has no rwnd but is missing * the lowest chunk.. which is probably what * is holding up the rwnd. */ goto one_chunk_around; } return (1); } one_chunk_around: if (asoc->peers_rwnd < mtu) { one_chunk = 1; if ((asoc->peers_rwnd == 0) && (asoc->total_flight == 0)) { chk->window_probe = 1; chk->whoTo->window_probe = 1; } } #ifdef SCTP_AUDITING_ENABLED sctp_audit_log(0xC3, 2); #endif bundle_at = 0; m = NULL; net->fast_retran_ip = 0; if (chk->rec.data.doing_fast_retransmit == 0) { /* * if no FR in progress skip destination that have * flight_size > cwnd. */ if (net->flight_size >= net->cwnd) { continue; } } else { /* * Mark the destination net to have FR recovery * limits put on it. */ *fr_done = 1; net->fast_retran_ip = 1; } /* * if no AUTH is yet included and this chunk requires it, * make sure to account for it. We don't apply the size * until the AUTH chunk is actually added below in case * there is no room for this chunk. */ if (data_auth_reqd && (auth == NULL)) { dmtu = sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); } else dmtu = 0; if ((chk->send_size <= (mtu - dmtu)) || (chk->flags & CHUNK_FLAGS_FRAGMENT_OK)) { /* ok we will add this one */ if (data_auth_reqd) { if (auth == NULL) { m = sctp_add_auth_chunk(m, &endofchain, &auth, &auth_offset, stcb, SCTP_DATA); auth_keyid = chk->auth_keyid; override_ok = 0; SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } else if (override_ok) { auth_keyid = chk->auth_keyid; override_ok = 0; } else if (chk->auth_keyid != auth_keyid) { /* different keyid, so done bundling */ break; } } m = sctp_copy_mbufchain(chk->data, m, &endofchain, 0, chk->send_size, chk->copy_by_ref); if (m == NULL) { SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } /* Do clear IP_DF ? */ if (chk->flags & CHUNK_FLAGS_FRAGMENT_OK) { no_fragmentflg = 0; } /* upate our MTU size */ if (mtu > (chk->send_size + dmtu)) mtu -= (chk->send_size + dmtu); else mtu = 0; data_list[bundle_at++] = chk; if (one_chunk && (asoc->total_flight <= 0)) { SCTP_STAT_INCR(sctps_windowprobed); } } if (one_chunk == 0) { /* * now are there anymore forward from chk to pick * up? */ for (fwd = TAILQ_NEXT(chk, sctp_next); fwd != NULL; fwd = TAILQ_NEXT(fwd, sctp_next)) { if (fwd->sent != SCTP_DATAGRAM_RESEND) { /* Nope, not for retran */ continue; } if (fwd->whoTo != net) { /* Nope, not the net in question */ continue; } if (data_auth_reqd && (auth == NULL)) { dmtu = sctp_get_auth_chunk_len(stcb->asoc.peer_hmac_id); } else dmtu = 0; if (fwd->send_size <= (mtu - dmtu)) { if (data_auth_reqd) { if (auth == NULL) { m = sctp_add_auth_chunk(m, &endofchain, &auth, &auth_offset, stcb, SCTP_DATA); auth_keyid = fwd->auth_keyid; override_ok = 0; SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } else if (override_ok) { auth_keyid = fwd->auth_keyid; override_ok = 0; } else if (fwd->auth_keyid != auth_keyid) { /* * different keyid, * so done bundling */ break; } } m = sctp_copy_mbufchain(fwd->data, m, &endofchain, 0, fwd->send_size, fwd->copy_by_ref); if (m == NULL) { SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } /* Do clear IP_DF ? */ if (fwd->flags & CHUNK_FLAGS_FRAGMENT_OK) { no_fragmentflg = 0; } /* upate our MTU size */ if (mtu > (fwd->send_size + dmtu)) mtu -= (fwd->send_size + dmtu); else mtu = 0; data_list[bundle_at++] = fwd; if (bundle_at >= SCTP_MAX_DATA_BUNDLING) { break; } } else { /* can't fit so we are done */ break; } } } /* Is there something to send for this destination? */ if (m) { /* * No matter if we fail/or succeed we should start a * timer. A failure is like a lost IP packet :-) */ if (!SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) { /* * no timer running on this destination * restart it. */ sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, net); tmr_started = 1; } /* Now lets send it, if there is anything to send :> */ if ((error = sctp_lowlevel_chunk_output(inp, stcb, net, (struct sockaddr *)&net->ro._l_addr, m, auth_offset, auth, auth_keyid, no_fragmentflg, 0, 0, inp->sctp_lport, stcb->rport, htonl(stcb->asoc.peer_vtag), net->port, NULL, 0, 0, so_locked))) { /* error, we could not output */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (error == ENOBUFS) { asoc->ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } return (error); } else { asoc->ifp_had_enobuf = 0; } endofchain = NULL; auth = NULL; auth_offset = 0; /* For HB's */ /* * We don't want to mark the net->sent time here * since this we use this for HB and retrans cannot * measure RTT */ /* (void)SCTP_GETTIME_TIMEVAL(&net->last_sent_time); */ /* For auto-close */ cnt_thru++; if (*now_filled == 0) { (void)SCTP_GETTIME_TIMEVAL(&asoc->time_last_sent); *now = asoc->time_last_sent; *now_filled = 1; } else { asoc->time_last_sent = *now; } *cnt_out += bundle_at; #ifdef SCTP_AUDITING_ENABLED sctp_audit_log(0xC4, bundle_at); #endif if (bundle_at) { tsns_sent = data_list[0]->rec.data.tsn; } for (i = 0; i < bundle_at; i++) { SCTP_STAT_INCR(sctps_sendretransdata); data_list[i]->sent = SCTP_DATAGRAM_SENT; /* * When we have a revoked data, and we * retransmit it, then we clear the revoked * flag since this flag dictates if we * subtracted from the fs */ if (data_list[i]->rec.data.chunk_was_revoked) { /* Deflate the cwnd */ data_list[i]->whoTo->cwnd -= data_list[i]->book_size; data_list[i]->rec.data.chunk_was_revoked = 0; } data_list[i]->snd_count++; sctp_ucount_decr(asoc->sent_queue_retran_cnt); /* record the time */ data_list[i]->sent_rcv_time = asoc->time_last_sent; if (data_list[i]->book_size_scale) { /* * need to double the book size on * this one */ data_list[i]->book_size_scale = 0; /* * Since we double the booksize, we * must also double the output queue * size, since this get shrunk when * we free by this amount. */ atomic_add_int(&((asoc)->total_output_queue_size), data_list[i]->book_size); data_list[i]->book_size *= 2; } else { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) { sctp_log_rwnd(SCTP_DECREASE_PEER_RWND, asoc->peers_rwnd, data_list[i]->send_size, SCTP_BASE_SYSCTL(sctp_peer_chunk_oh)); } asoc->peers_rwnd = sctp_sbspace_sub(asoc->peers_rwnd, (uint32_t)(data_list[i]->send_size + SCTP_BASE_SYSCTL(sctp_peer_chunk_oh))); } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) { sctp_misc_ints(SCTP_FLIGHT_LOG_UP_RSND, data_list[i]->whoTo->flight_size, data_list[i]->book_size, (uint32_t)(uintptr_t)data_list[i]->whoTo, data_list[i]->rec.data.tsn); } sctp_flight_size_increase(data_list[i]); sctp_total_flight_increase(stcb, data_list[i]); if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) { /* SWS sender side engages */ asoc->peers_rwnd = 0; } if ((i == 0) && (data_list[i]->rec.data.doing_fast_retransmit)) { SCTP_STAT_INCR(sctps_sendfastretrans); if ((data_list[i] == TAILQ_FIRST(&asoc->sent_queue)) && (tmr_started == 0)) { /*- * ok we just fast-retrans'd * the lowest TSN, i.e the * first on the list. In * this case we want to give * some more time to get a * SACK back without a * t3-expiring. */ sctp_timer_stop(SCTP_TIMER_TYPE_SEND, inp, stcb, net, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_2); sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, net); } } } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, net, tsns_sent, SCTP_CWND_LOG_FROM_RESEND); } #ifdef SCTP_AUDITING_ENABLED sctp_auditing(21, inp, stcb, NULL); #endif } else { /* None will fit */ return (1); } if (asoc->sent_queue_retran_cnt <= 0) { /* all done we have no more to retran */ asoc->sent_queue_retran_cnt = 0; break; } if (one_chunk) { /* No more room in rwnd */ return (1); } /* stop the for loop here. we sent out a packet */ break; } return (0); } static void sctp_timer_validation(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_association *asoc) { struct sctp_nets *net; /* Validate that a timer is running somewhere */ TAILQ_FOREACH(net, &asoc->nets, sctp_next) { if (SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) { /* Here is a timer */ return; } } SCTP_TCB_LOCK_ASSERT(stcb); /* Gak, we did not have a timer somewhere */ SCTPDBG(SCTP_DEBUG_OUTPUT3, "Deadlock avoided starting timer on a dest at retran\n"); if (asoc->alternate) { sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, asoc->alternate); } else { sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, asoc->primary_destination); } return; } void sctp_chunk_output(struct sctp_inpcb *inp, struct sctp_tcb *stcb, int from_where, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { /*- * Ok this is the generic chunk service queue. we must do the * following: * - See if there are retransmits pending, if so we must * do these first. * - Service the stream queue that is next, moving any * message (note I must get a complete message i.e. * FIRST/MIDDLE and LAST to the out queue in one pass) and assigning * TSN's * - Check to see if the cwnd/rwnd allows any output, if so we * go ahead and fomulate and send the low level chunks. Making sure * to combine any control in the control chunk queue also. */ struct sctp_association *asoc; struct sctp_nets *net; int error = 0, num_out, tot_out = 0, ret = 0, reason_code; unsigned int burst_cnt = 0; struct timeval now; int now_filled = 0; int nagle_on; int frag_point = sctp_get_frag_point(stcb, &stcb->asoc); int un_sent = 0; int fr_done; unsigned int tot_frs = 0; asoc = &stcb->asoc; do_it_again: /* The Nagle algorithm is only applied when handling a send call. */ if (from_where == SCTP_OUTPUT_FROM_USR_SEND) { if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_NODELAY)) { nagle_on = 0; } else { nagle_on = 1; } } else { nagle_on = 0; } SCTP_TCB_LOCK_ASSERT(stcb); un_sent = (stcb->asoc.total_output_queue_size - stcb->asoc.total_flight); if ((un_sent <= 0) && (TAILQ_EMPTY(&asoc->control_send_queue)) && (TAILQ_EMPTY(&asoc->asconf_send_queue)) && (asoc->sent_queue_retran_cnt == 0) && (asoc->trigger_reset == 0)) { /* Nothing to do unless there is something to be sent left */ return; } /* * Do we have something to send, data or control AND a sack timer * running, if so piggy-back the sack. */ if (SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) { sctp_send_sack(stcb, so_locked); sctp_timer_stop(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_3); } while (asoc->sent_queue_retran_cnt) { /*- * Ok, it is retransmission time only, we send out only ONE * packet with a single call off to the retran code. */ if (from_where == SCTP_OUTPUT_FROM_COOKIE_ACK) { /*- * Special hook for handling cookiess discarded * by peer that carried data. Send cookie-ack only * and then the next call with get the retran's. */ (void)sctp_med_chunk_output(inp, stcb, asoc, &num_out, &reason_code, 1, from_where, &now, &now_filled, frag_point, so_locked); return; } else if (from_where != SCTP_OUTPUT_FROM_HB_TMR) { /* if its not from a HB then do it */ fr_done = 0; ret = sctp_chunk_retransmission(inp, stcb, asoc, &num_out, &now, &now_filled, &fr_done, so_locked); if (fr_done) { tot_frs++; } } else { /* * its from any other place, we don't allow retran * output (only control) */ ret = 1; } if (ret > 0) { /* Can't send anymore */ /*- * now lets push out control by calling med-level * output once. this assures that we WILL send HB's * if queued too. */ (void)sctp_med_chunk_output(inp, stcb, asoc, &num_out, &reason_code, 1, from_where, &now, &now_filled, frag_point, so_locked); #ifdef SCTP_AUDITING_ENABLED sctp_auditing(8, inp, stcb, NULL); #endif sctp_timer_validation(inp, stcb, asoc); return; } if (ret < 0) { /*- * The count was off.. retran is not happening so do * the normal retransmission. */ #ifdef SCTP_AUDITING_ENABLED sctp_auditing(9, inp, stcb, NULL); #endif if (ret == SCTP_RETRAN_EXIT) { return; } break; } if (from_where == SCTP_OUTPUT_FROM_T3) { /* Only one transmission allowed out of a timeout */ #ifdef SCTP_AUDITING_ENABLED sctp_auditing(10, inp, stcb, NULL); #endif /* Push out any control */ (void)sctp_med_chunk_output(inp, stcb, asoc, &num_out, &reason_code, 1, from_where, &now, &now_filled, frag_point, so_locked); return; } if ((asoc->fr_max_burst > 0) && (tot_frs >= asoc->fr_max_burst)) { /* Hit FR burst limit */ return; } if ((num_out == 0) && (ret == 0)) { /* No more retrans to send */ break; } } #ifdef SCTP_AUDITING_ENABLED sctp_auditing(12, inp, stcb, NULL); #endif /* Check for bad destinations, if they exist move chunks around. */ TAILQ_FOREACH(net, &asoc->nets, sctp_next) { if (!(net->dest_state & SCTP_ADDR_REACHABLE)) { /*- * if possible move things off of this address we * still may send below due to the dormant state but * we try to find an alternate address to send to * and if we have one we move all queued data on the * out wheel to this alternate address. */ if (net->ref_count > 1) sctp_move_chunks_from_net(stcb, net); } else { /*- * if ((asoc->sat_network) || (net->addr_is_local)) * { burst_limit = asoc->max_burst * * SCTP_SAT_NETWORK_BURST_INCR; } */ if (asoc->max_burst > 0) { if (SCTP_BASE_SYSCTL(sctp_use_cwnd_based_maxburst)) { if ((net->flight_size + (asoc->max_burst * net->mtu)) < net->cwnd) { /* * JRS - Use the congestion * control given in the * congestion control module */ asoc->cc_functions.sctp_cwnd_update_after_output(stcb, net, asoc->max_burst); if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_MAXBURST_ENABLE) { sctp_log_maxburst(stcb, net, 0, asoc->max_burst, SCTP_MAX_BURST_APPLIED); } SCTP_STAT_INCR(sctps_maxburstqueued); } net->fast_retran_ip = 0; } else { if (net->flight_size == 0) { /* * Should be decaying the * cwnd here */ ; } } } } } burst_cnt = 0; do { error = sctp_med_chunk_output(inp, stcb, asoc, &num_out, &reason_code, 0, from_where, &now, &now_filled, frag_point, so_locked); if (error) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "Error %d was returned from med-c-op\n", error); if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_MAXBURST_ENABLE) { sctp_log_maxburst(stcb, asoc->primary_destination, error, burst_cnt, SCTP_MAX_BURST_ERROR_STOP); } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, NULL, error, SCTP_SEND_NOW_COMPLETES); sctp_log_cwnd(stcb, NULL, 0xdeadbeef, SCTP_SEND_NOW_COMPLETES); } break; } SCTPDBG(SCTP_DEBUG_OUTPUT3, "m-c-o put out %d\n", num_out); tot_out += num_out; burst_cnt++; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, NULL, num_out, SCTP_SEND_NOW_COMPLETES); if (num_out == 0) { sctp_log_cwnd(stcb, NULL, reason_code, SCTP_SEND_NOW_COMPLETES); } } if (nagle_on) { /* * When the Nagle algorithm is used, look at how * much is unsent, then if its smaller than an MTU * and we have data in flight we stop, except if we * are handling a fragmented user message. */ un_sent = stcb->asoc.total_output_queue_size - stcb->asoc.total_flight; if ((un_sent < (int)(stcb->asoc.smallest_mtu - SCTP_MIN_OVERHEAD)) && (stcb->asoc.total_flight > 0)) { /* && sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXPLICIT_EOR))) {*/ break; } } if (TAILQ_EMPTY(&asoc->control_send_queue) && TAILQ_EMPTY(&asoc->send_queue) && sctp_is_there_unsent_data(stcb, so_locked) == 0) { /* Nothing left to send */ break; } if ((stcb->asoc.total_output_queue_size - stcb->asoc.total_flight) <= 0) { /* Nothing left to send */ break; } } while (num_out && ((asoc->max_burst == 0) || SCTP_BASE_SYSCTL(sctp_use_cwnd_based_maxburst) || (burst_cnt < asoc->max_burst))); if (SCTP_BASE_SYSCTL(sctp_use_cwnd_based_maxburst) == 0) { if ((asoc->max_burst > 0) && (burst_cnt >= asoc->max_burst)) { SCTP_STAT_INCR(sctps_maxburstqueued); asoc->burst_limit_applied = 1; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_MAXBURST_ENABLE) { sctp_log_maxburst(stcb, asoc->primary_destination, 0, burst_cnt, SCTP_MAX_BURST_APPLIED); } } else { asoc->burst_limit_applied = 0; } } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { sctp_log_cwnd(stcb, NULL, tot_out, SCTP_SEND_NOW_COMPLETES); } SCTPDBG(SCTP_DEBUG_OUTPUT1, "Ok, we have put out %d chunks\n", tot_out); /*- * Now we need to clean up the control chunk chain if a ECNE is on * it. It must be marked as UNSENT again so next call will continue * to send it until such time that we get a CWR, to remove it. */ if (stcb->asoc.ecn_echo_cnt_onq) sctp_fix_ecn_echo(asoc); if (stcb->asoc.trigger_reset) { if (sctp_send_stream_reset_out_if_possible(stcb, so_locked) == 0) { goto do_it_again; } } return; } int sctp_output( struct sctp_inpcb *inp, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *p, int flags) { if (inp == NULL) { SCTP_LTRACE_ERR_RET_PKT(m, inp, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } if (inp->sctp_socket == NULL) { SCTP_LTRACE_ERR_RET_PKT(m, inp, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } return (sctp_sosend(inp->sctp_socket, addr, (struct uio *)NULL, m, control, flags, p )); } void send_forward_tsn(struct sctp_tcb *stcb, struct sctp_association *asoc) { struct sctp_tmit_chunk *chk, *at, *tp1, *last; struct sctp_forward_tsn_chunk *fwdtsn; struct sctp_strseq *strseq; struct sctp_strseq_mid *strseq_m; uint32_t advance_peer_ack_point; unsigned int cnt_of_space, i, ovh; unsigned int space_needed; unsigned int cnt_of_skipped = 0; SCTP_TCB_LOCK_ASSERT(stcb); TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if (chk->rec.chunk_id.id == SCTP_FORWARD_CUM_TSN) { /* mark it to unsent */ chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; /* Do we correct its output location? */ if (chk->whoTo) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = NULL; } goto sctp_fill_in_rest; } } /* Ok if we reach here we must build one */ sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { return; } asoc->fwd_tsn_cnt++; chk->copy_by_ref = 0; /* * We don't do the old thing here since this is used not for on-wire * but to tell if we are sending a fwd-tsn by the stack during * output. And if its a IFORWARD or a FORWARD it is a fwd-tsn. */ chk->rec.chunk_id.id = SCTP_FORWARD_CUM_TSN; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->asoc = asoc; chk->whoTo = NULL; chk->data = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; TAILQ_INSERT_TAIL(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; sctp_fill_in_rest: /*- * Here we go through and fill out the part that deals with * stream/seq of the ones we skip. */ SCTP_BUF_LEN(chk->data) = 0; TAILQ_FOREACH(at, &asoc->sent_queue, sctp_next) { if ((at->sent != SCTP_FORWARD_TSN_SKIP) && (at->sent != SCTP_DATAGRAM_NR_ACKED)) { /* no more to look at */ break; } if (!asoc->idata_supported && (at->rec.data.rcv_flags & SCTP_DATA_UNORDERED)) { /* We don't report these */ continue; } cnt_of_skipped++; } if (asoc->idata_supported) { space_needed = (sizeof(struct sctp_forward_tsn_chunk) + (cnt_of_skipped * sizeof(struct sctp_strseq_mid))); } else { space_needed = (sizeof(struct sctp_forward_tsn_chunk) + (cnt_of_skipped * sizeof(struct sctp_strseq))); } cnt_of_space = (unsigned int)M_TRAILINGSPACE(chk->data); if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) { ovh = SCTP_MIN_OVERHEAD; } else { ovh = SCTP_MIN_V4_OVERHEAD; } if (cnt_of_space > (asoc->smallest_mtu - ovh)) { /* trim to a mtu size */ cnt_of_space = asoc->smallest_mtu - ovh; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_TRY_ADVANCE) { sctp_misc_ints(SCTP_FWD_TSN_CHECK, 0xff, 0, cnt_of_skipped, asoc->advanced_peer_ack_point); } advance_peer_ack_point = asoc->advanced_peer_ack_point; if (cnt_of_space < space_needed) { /*- * ok we must trim down the chunk by lowering the * advance peer ack point. */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_TRY_ADVANCE) { sctp_misc_ints(SCTP_FWD_TSN_CHECK, 0xff, 0xff, cnt_of_space, space_needed); } cnt_of_skipped = cnt_of_space - sizeof(struct sctp_forward_tsn_chunk); if (asoc->idata_supported) { cnt_of_skipped /= sizeof(struct sctp_strseq_mid); } else { cnt_of_skipped /= sizeof(struct sctp_strseq); } /*- * Go through and find the TSN that will be the one * we report. */ at = TAILQ_FIRST(&asoc->sent_queue); if (at != NULL) { for (i = 0; i < cnt_of_skipped; i++) { tp1 = TAILQ_NEXT(at, sctp_next); if (tp1 == NULL) { break; } at = tp1; } } if (at && SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_TRY_ADVANCE) { sctp_misc_ints(SCTP_FWD_TSN_CHECK, 0xff, cnt_of_skipped, at->rec.data.tsn, asoc->advanced_peer_ack_point); } last = at; /*- * last now points to last one I can report, update * peer ack point */ if (last) { advance_peer_ack_point = last->rec.data.tsn; } if (asoc->idata_supported) { space_needed = sizeof(struct sctp_forward_tsn_chunk) + cnt_of_skipped * sizeof(struct sctp_strseq_mid); } else { space_needed = sizeof(struct sctp_forward_tsn_chunk) + cnt_of_skipped * sizeof(struct sctp_strseq); } } chk->send_size = space_needed; /* Setup the chunk */ fwdtsn = mtod(chk->data, struct sctp_forward_tsn_chunk *); fwdtsn->ch.chunk_length = htons(chk->send_size); fwdtsn->ch.chunk_flags = 0; if (asoc->idata_supported) { fwdtsn->ch.chunk_type = SCTP_IFORWARD_CUM_TSN; } else { fwdtsn->ch.chunk_type = SCTP_FORWARD_CUM_TSN; } fwdtsn->new_cumulative_tsn = htonl(advance_peer_ack_point); SCTP_BUF_LEN(chk->data) = chk->send_size; fwdtsn++; /*- * Move pointer to after the fwdtsn and transfer to the * strseq pointer. */ if (asoc->idata_supported) { strseq_m = (struct sctp_strseq_mid *)fwdtsn; strseq = NULL; } else { strseq = (struct sctp_strseq *)fwdtsn; strseq_m = NULL; } /*- * Now populate the strseq list. This is done blindly * without pulling out duplicate stream info. This is * inefficent but won't harm the process since the peer will * look at these in sequence and will thus release anything. * It could mean we exceed the PMTU and chop off some that * we could have included.. but this is unlikely (aka 1432/4 * would mean 300+ stream seq's would have to be reported in * one FWD-TSN. With a bit of work we can later FIX this to * optimize and pull out duplicates.. but it does add more * overhead. So for now... not! */ i = 0; TAILQ_FOREACH(at, &asoc->sent_queue, sctp_next) { if (i >= cnt_of_skipped) { break; } if (!asoc->idata_supported && (at->rec.data.rcv_flags & SCTP_DATA_UNORDERED)) { /* We don't report these */ continue; } if (at->rec.data.tsn == advance_peer_ack_point) { at->rec.data.fwd_tsn_cnt = 0; } if (asoc->idata_supported) { strseq_m->sid = htons(at->rec.data.sid); if (at->rec.data.rcv_flags & SCTP_DATA_UNORDERED) { strseq_m->flags = htons(PR_SCTP_UNORDERED_FLAG); } else { strseq_m->flags = 0; } strseq_m->mid = htonl(at->rec.data.mid); strseq_m++; } else { strseq->sid = htons(at->rec.data.sid); strseq->ssn = htons((uint16_t)at->rec.data.mid); strseq++; } i++; } return; } void sctp_send_sack(struct sctp_tcb *stcb, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { /*- * Queue up a SACK or NR-SACK in the control queue. * We must first check to see if a SACK or NR-SACK is * somehow on the control queue. * If so, we will take and and remove the old one. */ struct sctp_association *asoc; struct sctp_tmit_chunk *chk, *a_chk; struct sctp_sack_chunk *sack; struct sctp_nr_sack_chunk *nr_sack; struct sctp_gap_ack_block *gap_descriptor; const struct sack_track *selector; int mergeable = 0; int offset; caddr_t limit; uint32_t *dup; int limit_reached = 0; unsigned int i, siz, j; unsigned int num_gap_blocks = 0, num_nr_gap_blocks = 0, space; int num_dups = 0; int space_req; uint32_t highest_tsn; uint8_t flags; uint8_t type; uint8_t tsn_map; if (stcb->asoc.nrsack_supported == 1) { type = SCTP_NR_SELECTIVE_ACK; } else { type = SCTP_SELECTIVE_ACK; } a_chk = NULL; asoc = &stcb->asoc; SCTP_TCB_LOCK_ASSERT(stcb); if (asoc->last_data_chunk_from == NULL) { /* Hmm we never received anything */ return; } sctp_slide_mapping_arrays(stcb); sctp_set_rwnd(stcb, asoc); TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if (chk->rec.chunk_id.id == type) { /* Hmm, found a sack already on queue, remove it */ TAILQ_REMOVE(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt--; a_chk = chk; if (a_chk->data) { sctp_m_freem(a_chk->data); a_chk->data = NULL; } if (a_chk->whoTo) { sctp_free_remote_addr(a_chk->whoTo); a_chk->whoTo = NULL; } break; } } if (a_chk == NULL) { sctp_alloc_a_chunk(stcb, a_chk); if (a_chk == NULL) { /* No memory so we drop the idea, and set a timer */ if (stcb->asoc.delayed_ack) { sctp_timer_stop(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_4); sctp_timer_start(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL); } else { stcb->asoc.send_sack = 1; } return; } a_chk->copy_by_ref = 0; a_chk->rec.chunk_id.id = type; a_chk->rec.chunk_id.can_take_data = 1; } /* Clear our pkt counts */ asoc->data_pkts_seen = 0; a_chk->flags = 0; a_chk->asoc = asoc; a_chk->snd_count = 0; a_chk->send_size = 0; /* fill in later */ a_chk->sent = SCTP_DATAGRAM_UNSENT; a_chk->whoTo = NULL; if (!(asoc->last_data_chunk_from->dest_state & SCTP_ADDR_REACHABLE)) { /*- * Ok, the destination for the SACK is unreachable, lets see if * we can select an alternate to asoc->last_data_chunk_from */ a_chk->whoTo = sctp_find_alternate_net(stcb, asoc->last_data_chunk_from, 0); if (a_chk->whoTo == NULL) { /* Nope, no alternate */ a_chk->whoTo = asoc->last_data_chunk_from; } } else { a_chk->whoTo = asoc->last_data_chunk_from; } if (a_chk->whoTo) { atomic_add_int(&a_chk->whoTo->ref_count, 1); } if (SCTP_TSN_GT(asoc->highest_tsn_inside_map, asoc->highest_tsn_inside_nr_map)) { highest_tsn = asoc->highest_tsn_inside_map; } else { highest_tsn = asoc->highest_tsn_inside_nr_map; } if (highest_tsn == asoc->cumulative_tsn) { /* no gaps */ if (type == SCTP_SELECTIVE_ACK) { space_req = sizeof(struct sctp_sack_chunk); } else { space_req = sizeof(struct sctp_nr_sack_chunk); } } else { /* gaps get a cluster */ space_req = MCLBYTES; } /* Ok now lets formulate a MBUF with our sack */ a_chk->data = sctp_get_mbuf_for_msg(space_req, 0, M_NOWAIT, 1, MT_DATA); if ((a_chk->data == NULL) || (a_chk->whoTo == NULL)) { /* rats, no mbuf memory */ if (a_chk->data) { /* was a problem with the destination */ sctp_m_freem(a_chk->data); a_chk->data = NULL; } sctp_free_a_chunk(stcb, a_chk, so_locked); /* sa_ignore NO_NULL_CHK */ if (stcb->asoc.delayed_ack) { sctp_timer_stop(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_5); sctp_timer_start(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL); } else { stcb->asoc.send_sack = 1; } return; } /* ok, lets go through and fill it in */ SCTP_BUF_RESV_UF(a_chk->data, SCTP_MIN_OVERHEAD); space = (unsigned int)M_TRAILINGSPACE(a_chk->data); if (space > (a_chk->whoTo->mtu - SCTP_MIN_OVERHEAD)) { space = (a_chk->whoTo->mtu - SCTP_MIN_OVERHEAD); } limit = mtod(a_chk->data, caddr_t); limit += space; flags = 0; if ((asoc->sctp_cmt_on_off > 0) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) { /*- * CMT DAC algorithm: If 2 (i.e., 0x10) packets have been * received, then set high bit to 1, else 0. Reset * pkts_rcvd. */ flags |= (asoc->cmt_dac_pkts_rcvd << 6); asoc->cmt_dac_pkts_rcvd = 0; } #ifdef SCTP_ASOCLOG_OF_TSNS stcb->asoc.cumack_logsnt[stcb->asoc.cumack_log_atsnt] = asoc->cumulative_tsn; stcb->asoc.cumack_log_atsnt++; if (stcb->asoc.cumack_log_atsnt >= SCTP_TSN_LOG_SIZE) { stcb->asoc.cumack_log_atsnt = 0; } #endif /* reset the readers interpretation */ stcb->freed_by_sorcv_sincelast = 0; if (type == SCTP_SELECTIVE_ACK) { sack = mtod(a_chk->data, struct sctp_sack_chunk *); nr_sack = NULL; gap_descriptor = (struct sctp_gap_ack_block *)((caddr_t)sack + sizeof(struct sctp_sack_chunk)); if (highest_tsn > asoc->mapping_array_base_tsn) { siz = (((highest_tsn - asoc->mapping_array_base_tsn) + 1) + 7) / 8; } else { siz = (((MAX_TSN - highest_tsn) + 1) + highest_tsn + 7) / 8; } } else { sack = NULL; nr_sack = mtod(a_chk->data, struct sctp_nr_sack_chunk *); gap_descriptor = (struct sctp_gap_ack_block *)((caddr_t)nr_sack + sizeof(struct sctp_nr_sack_chunk)); if (asoc->highest_tsn_inside_map > asoc->mapping_array_base_tsn) { siz = (((asoc->highest_tsn_inside_map - asoc->mapping_array_base_tsn) + 1) + 7) / 8; } else { siz = (((MAX_TSN - asoc->mapping_array_base_tsn) + 1) + asoc->highest_tsn_inside_map + 7) / 8; } } if (SCTP_TSN_GT(asoc->mapping_array_base_tsn, asoc->cumulative_tsn)) { offset = 1; } else { offset = asoc->mapping_array_base_tsn - asoc->cumulative_tsn; } if (((type == SCTP_SELECTIVE_ACK) && SCTP_TSN_GT(highest_tsn, asoc->cumulative_tsn)) || ((type == SCTP_NR_SELECTIVE_ACK) && SCTP_TSN_GT(asoc->highest_tsn_inside_map, asoc->cumulative_tsn))) { /* we have a gap .. maybe */ for (i = 0; i < siz; i++) { tsn_map = asoc->mapping_array[i]; if (type == SCTP_SELECTIVE_ACK) { tsn_map |= asoc->nr_mapping_array[i]; } if (i == 0) { /* * Clear all bits corresponding to TSNs * smaller or equal to the cumulative TSN. */ tsn_map &= (~0U << (1 - offset)); } selector = &sack_array[tsn_map]; if (mergeable && selector->right_edge) { /* * Backup, left and right edges were ok to * merge. */ num_gap_blocks--; gap_descriptor--; } if (selector->num_entries == 0) mergeable = 0; else { for (j = 0; j < selector->num_entries; j++) { if (mergeable && selector->right_edge) { /* * do a merge by NOT setting * the left side */ mergeable = 0; } else { /* * no merge, set the left * side */ mergeable = 0; gap_descriptor->start = htons((selector->gaps[j].start + offset)); } gap_descriptor->end = htons((selector->gaps[j].end + offset)); num_gap_blocks++; gap_descriptor++; if (((caddr_t)gap_descriptor + sizeof(struct sctp_gap_ack_block)) > limit) { /* no more room */ limit_reached = 1; break; } } if (selector->left_edge) { mergeable = 1; } } if (limit_reached) { /* Reached the limit stop */ break; } offset += 8; } } if ((type == SCTP_NR_SELECTIVE_ACK) && (limit_reached == 0)) { mergeable = 0; if (asoc->highest_tsn_inside_nr_map > asoc->mapping_array_base_tsn) { siz = (((asoc->highest_tsn_inside_nr_map - asoc->mapping_array_base_tsn) + 1) + 7) / 8; } else { siz = (((MAX_TSN - asoc->mapping_array_base_tsn) + 1) + asoc->highest_tsn_inside_nr_map + 7) / 8; } if (SCTP_TSN_GT(asoc->mapping_array_base_tsn, asoc->cumulative_tsn)) { offset = 1; } else { offset = asoc->mapping_array_base_tsn - asoc->cumulative_tsn; } if (SCTP_TSN_GT(asoc->highest_tsn_inside_nr_map, asoc->cumulative_tsn)) { /* we have a gap .. maybe */ for (i = 0; i < siz; i++) { tsn_map = asoc->nr_mapping_array[i]; if (i == 0) { /* * Clear all bits corresponding to * TSNs smaller or equal to the * cumulative TSN. */ tsn_map &= (~0U << (1 - offset)); } selector = &sack_array[tsn_map]; if (mergeable && selector->right_edge) { /* * Backup, left and right edges were * ok to merge. */ num_nr_gap_blocks--; gap_descriptor--; } if (selector->num_entries == 0) mergeable = 0; else { for (j = 0; j < selector->num_entries; j++) { if (mergeable && selector->right_edge) { /* * do a merge by NOT * setting the left * side */ mergeable = 0; } else { /* * no merge, set the * left side */ mergeable = 0; gap_descriptor->start = htons((selector->gaps[j].start + offset)); } gap_descriptor->end = htons((selector->gaps[j].end + offset)); num_nr_gap_blocks++; gap_descriptor++; if (((caddr_t)gap_descriptor + sizeof(struct sctp_gap_ack_block)) > limit) { /* no more room */ limit_reached = 1; break; } } if (selector->left_edge) { mergeable = 1; } } if (limit_reached) { /* Reached the limit stop */ break; } offset += 8; } } } /* now we must add any dups we are going to report. */ if ((limit_reached == 0) && (asoc->numduptsns)) { dup = (uint32_t *)gap_descriptor; for (i = 0; i < asoc->numduptsns; i++) { *dup = htonl(asoc->dup_tsns[i]); dup++; num_dups++; if (((caddr_t)dup + sizeof(uint32_t)) > limit) { /* no more room */ break; } } asoc->numduptsns = 0; } /* * now that the chunk is prepared queue it to the control chunk * queue. */ if (type == SCTP_SELECTIVE_ACK) { a_chk->send_size = (uint16_t)(sizeof(struct sctp_sack_chunk) + (num_gap_blocks + num_nr_gap_blocks) * sizeof(struct sctp_gap_ack_block) + num_dups * sizeof(int32_t)); SCTP_BUF_LEN(a_chk->data) = a_chk->send_size; sack->sack.cum_tsn_ack = htonl(asoc->cumulative_tsn); sack->sack.a_rwnd = htonl(asoc->my_rwnd); sack->sack.num_gap_ack_blks = htons(num_gap_blocks); sack->sack.num_dup_tsns = htons(num_dups); sack->ch.chunk_type = type; sack->ch.chunk_flags = flags; sack->ch.chunk_length = htons(a_chk->send_size); } else { a_chk->send_size = (uint16_t)(sizeof(struct sctp_nr_sack_chunk) + (num_gap_blocks + num_nr_gap_blocks) * sizeof(struct sctp_gap_ack_block) + num_dups * sizeof(int32_t)); SCTP_BUF_LEN(a_chk->data) = a_chk->send_size; nr_sack->nr_sack.cum_tsn_ack = htonl(asoc->cumulative_tsn); nr_sack->nr_sack.a_rwnd = htonl(asoc->my_rwnd); nr_sack->nr_sack.num_gap_ack_blks = htons(num_gap_blocks); nr_sack->nr_sack.num_nr_gap_ack_blks = htons(num_nr_gap_blocks); nr_sack->nr_sack.num_dup_tsns = htons(num_dups); nr_sack->nr_sack.reserved = 0; nr_sack->ch.chunk_type = type; nr_sack->ch.chunk_flags = flags; nr_sack->ch.chunk_length = htons(a_chk->send_size); } TAILQ_INSERT_TAIL(&asoc->control_send_queue, a_chk, sctp_next); asoc->my_last_reported_rwnd = asoc->my_rwnd; asoc->ctrl_queue_cnt++; asoc->send_sack = 0; SCTP_STAT_INCR(sctps_sendsacks); return; } void sctp_send_abort_tcb(struct sctp_tcb *stcb, struct mbuf *operr, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { struct mbuf *m_abort, *m, *m_last; struct mbuf *m_out, *m_end = NULL; struct sctp_abort_chunk *abort; struct sctp_auth_chunk *auth = NULL; struct sctp_nets *net; uint32_t vtag; uint32_t auth_offset = 0; int error; uint16_t cause_len, chunk_len, padding_len; SCTP_TCB_LOCK_ASSERT(stcb); /*- * Add an AUTH chunk, if chunk requires it and save the offset into * the chain for AUTH */ if (sctp_auth_is_required_chunk(SCTP_ABORT_ASSOCIATION, stcb->asoc.peer_auth_chunks)) { m_out = sctp_add_auth_chunk(NULL, &m_end, &auth, &auth_offset, stcb, SCTP_ABORT_ASSOCIATION); SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } else { m_out = NULL; } m_abort = sctp_get_mbuf_for_msg(sizeof(struct sctp_abort_chunk), 0, M_NOWAIT, 1, MT_HEADER); if (m_abort == NULL) { if (m_out) { sctp_m_freem(m_out); } if (operr) { sctp_m_freem(operr); } return; } /* link in any error */ SCTP_BUF_NEXT(m_abort) = operr; cause_len = 0; m_last = NULL; for (m = operr; m; m = SCTP_BUF_NEXT(m)) { cause_len += (uint16_t)SCTP_BUF_LEN(m); if (SCTP_BUF_NEXT(m) == NULL) { m_last = m; } } SCTP_BUF_LEN(m_abort) = sizeof(struct sctp_abort_chunk); chunk_len = (uint16_t)sizeof(struct sctp_abort_chunk) + cause_len; padding_len = SCTP_SIZE32(chunk_len) - chunk_len; if (m_out == NULL) { /* NO Auth chunk prepended, so reserve space in front */ SCTP_BUF_RESV_UF(m_abort, SCTP_MIN_OVERHEAD); m_out = m_abort; } else { /* Put AUTH chunk at the front of the chain */ SCTP_BUF_NEXT(m_end) = m_abort; } if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } /* Fill in the ABORT chunk header. */ abort = mtod(m_abort, struct sctp_abort_chunk *); abort->ch.chunk_type = SCTP_ABORT_ASSOCIATION; if (stcb->asoc.peer_vtag == 0) { /* This happens iff the assoc is in COOKIE-WAIT state. */ vtag = stcb->asoc.my_vtag; abort->ch.chunk_flags = SCTP_HAD_NO_TCB; } else { vtag = stcb->asoc.peer_vtag; abort->ch.chunk_flags = 0; } abort->ch.chunk_length = htons(chunk_len); /* Add padding, if necessary. */ if (padding_len > 0) { if ((m_last == NULL) || (sctp_add_pad_tombuf(m_last, padding_len) == NULL)) { sctp_m_freem(m_out); return; } } if ((error = sctp_lowlevel_chunk_output(stcb->sctp_ep, stcb, net, (struct sockaddr *)&net->ro._l_addr, m_out, auth_offset, auth, stcb->asoc.authinfo.active_keyid, 1, 0, 0, stcb->sctp_ep->sctp_lport, stcb->rport, htonl(vtag), stcb->asoc.primary_destination->port, NULL, 0, 0, so_locked))) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (error == ENOBUFS) { stcb->asoc.ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } } else { stcb->asoc.ifp_had_enobuf = 0; } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); } void sctp_send_shutdown_complete(struct sctp_tcb *stcb, struct sctp_nets *net, int reflect_vtag) { /* formulate and SEND a SHUTDOWN-COMPLETE */ struct mbuf *m_shutdown_comp; struct sctp_shutdown_complete_chunk *shutdown_complete; uint32_t vtag; int error; uint8_t flags; m_shutdown_comp = sctp_get_mbuf_for_msg(sizeof(struct sctp_chunkhdr), 0, M_NOWAIT, 1, MT_HEADER); if (m_shutdown_comp == NULL) { /* no mbuf's */ return; } if (reflect_vtag) { flags = SCTP_HAD_NO_TCB; vtag = stcb->asoc.my_vtag; } else { flags = 0; vtag = stcb->asoc.peer_vtag; } shutdown_complete = mtod(m_shutdown_comp, struct sctp_shutdown_complete_chunk *); shutdown_complete->ch.chunk_type = SCTP_SHUTDOWN_COMPLETE; shutdown_complete->ch.chunk_flags = flags; shutdown_complete->ch.chunk_length = htons(sizeof(struct sctp_shutdown_complete_chunk)); SCTP_BUF_LEN(m_shutdown_comp) = sizeof(struct sctp_shutdown_complete_chunk); if ((error = sctp_lowlevel_chunk_output(stcb->sctp_ep, stcb, net, (struct sockaddr *)&net->ro._l_addr, m_shutdown_comp, 0, NULL, 0, 1, 0, 0, stcb->sctp_ep->sctp_lport, stcb->rport, htonl(vtag), net->port, NULL, 0, 0, SCTP_SO_NOT_LOCKED))) { SCTPDBG(SCTP_DEBUG_OUTPUT3, "Gak send error %d\n", error); if (error == ENOBUFS) { stcb->asoc.ifp_had_enobuf = 1; SCTP_STAT_INCR(sctps_lowlevelerr); } } else { stcb->asoc.ifp_had_enobuf = 0; } SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); return; } static void sctp_send_resp_msg(struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, uint32_t vtag, uint8_t type, struct mbuf *cause, uint8_t mflowtype, uint32_t mflowid, uint16_t fibnum, uint32_t vrf_id, uint16_t port) { struct mbuf *o_pak; struct mbuf *mout; struct sctphdr *shout; struct sctp_chunkhdr *ch; #if defined(INET) || defined(INET6) struct udphdr *udp; #endif int ret, len, cause_len, padding_len; #ifdef INET struct sockaddr_in *src_sin, *dst_sin; struct ip *ip; #endif #ifdef INET6 struct sockaddr_in6 *src_sin6, *dst_sin6; struct ip6_hdr *ip6; #endif /* Compute the length of the cause and add final padding. */ cause_len = 0; if (cause != NULL) { struct mbuf *m_at, *m_last = NULL; for (m_at = cause; m_at; m_at = SCTP_BUF_NEXT(m_at)) { if (SCTP_BUF_NEXT(m_at) == NULL) m_last = m_at; cause_len += SCTP_BUF_LEN(m_at); } padding_len = cause_len % 4; if (padding_len != 0) { padding_len = 4 - padding_len; } if (padding_len != 0) { if (sctp_add_pad_tombuf(m_last, padding_len) == NULL) { sctp_m_freem(cause); return; } } } else { padding_len = 0; } /* Get an mbuf for the header. */ len = sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr); switch (dst->sa_family) { #ifdef INET case AF_INET: len += sizeof(struct ip); break; #endif #ifdef INET6 case AF_INET6: len += sizeof(struct ip6_hdr); break; #endif default: break; } #if defined(INET) || defined(INET6) if (port) { len += sizeof(struct udphdr); } #endif mout = sctp_get_mbuf_for_msg(len + max_linkhdr, 1, M_NOWAIT, 1, MT_DATA); if (mout == NULL) { if (cause) { sctp_m_freem(cause); } return; } SCTP_BUF_RESV_UF(mout, max_linkhdr); SCTP_BUF_LEN(mout) = len; SCTP_BUF_NEXT(mout) = cause; M_SETFIB(mout, fibnum); mout->m_pkthdr.flowid = mflowid; M_HASHTYPE_SET(mout, mflowtype); #ifdef INET ip = NULL; #endif #ifdef INET6 ip6 = NULL; #endif switch (dst->sa_family) { #ifdef INET case AF_INET: src_sin = (struct sockaddr_in *)src; dst_sin = (struct sockaddr_in *)dst; ip = mtod(mout, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = (sizeof(struct ip) >> 2); ip->ip_tos = 0; ip->ip_off = htons(IP_DF); ip_fillid(ip); ip->ip_ttl = MODULE_GLOBAL(ip_defttl); if (port) { ip->ip_p = IPPROTO_UDP; } else { ip->ip_p = IPPROTO_SCTP; } ip->ip_src.s_addr = dst_sin->sin_addr.s_addr; ip->ip_dst.s_addr = src_sin->sin_addr.s_addr; ip->ip_sum = 0; len = sizeof(struct ip); shout = (struct sctphdr *)((caddr_t)ip + len); break; #endif #ifdef INET6 case AF_INET6: src_sin6 = (struct sockaddr_in6 *)src; dst_sin6 = (struct sockaddr_in6 *)dst; ip6 = mtod(mout, struct ip6_hdr *); ip6->ip6_flow = htonl(0x60000000); if (V_ip6_auto_flowlabel) { ip6->ip6_flow |= (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); } ip6->ip6_hlim = MODULE_GLOBAL(ip6_defhlim); if (port) { ip6->ip6_nxt = IPPROTO_UDP; } else { ip6->ip6_nxt = IPPROTO_SCTP; } ip6->ip6_src = dst_sin6->sin6_addr; ip6->ip6_dst = src_sin6->sin6_addr; len = sizeof(struct ip6_hdr); shout = (struct sctphdr *)((caddr_t)ip6 + len); break; #endif default: len = 0; shout = mtod(mout, struct sctphdr *); break; } #if defined(INET) || defined(INET6) if (port) { if (htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)) == 0) { sctp_m_freem(mout); return; } udp = (struct udphdr *)shout; udp->uh_sport = htons(SCTP_BASE_SYSCTL(sctp_udp_tunneling_port)); udp->uh_dport = port; udp->uh_sum = 0; udp->uh_ulen = htons((uint16_t)(sizeof(struct udphdr) + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + cause_len + padding_len)); len += sizeof(struct udphdr); shout = (struct sctphdr *)((caddr_t)shout + sizeof(struct udphdr)); } else { udp = NULL; } #endif shout->src_port = sh->dest_port; shout->dest_port = sh->src_port; shout->checksum = 0; if (vtag) { shout->v_tag = htonl(vtag); } else { shout->v_tag = sh->v_tag; } len += sizeof(struct sctphdr); ch = (struct sctp_chunkhdr *)((caddr_t)shout + sizeof(struct sctphdr)); ch->chunk_type = type; if (vtag) { ch->chunk_flags = 0; } else { ch->chunk_flags = SCTP_HAD_NO_TCB; } ch->chunk_length = htons((uint16_t)(sizeof(struct sctp_chunkhdr) + cause_len)); len += sizeof(struct sctp_chunkhdr); len += cause_len + padding_len; if (SCTP_GET_HEADER_FOR_OUTPUT(o_pak)) { sctp_m_freem(mout); return; } SCTP_ATTACH_CHAIN(o_pak, mout, len); switch (dst->sa_family) { #ifdef INET case AF_INET: if (port) { if (V_udp_cksum) { udp->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, udp->uh_ulen + htons(IPPROTO_UDP)); } else { udp->uh_sum = 0; } } ip->ip_len = htons(len); if (port) { shout->checksum = sctp_calculate_cksum(mout, sizeof(struct ip) + sizeof(struct udphdr)); SCTP_STAT_INCR(sctps_sendswcrc); if (V_udp_cksum) { SCTP_ENABLE_UDP_CSUM(o_pak); } } else { mout->m_pkthdr.csum_flags = CSUM_SCTP; mout->m_pkthdr.csum_data = offsetof(struct sctphdr, checksum); SCTP_STAT_INCR(sctps_sendhwcrc); } #ifdef SCTP_PACKET_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LAST_PACKET_TRACING) { sctp_packet_log(o_pak); } #endif SCTP_PROBE5(send, NULL, NULL, ip, NULL, shout); SCTP_IP_OUTPUT(ret, o_pak, NULL, NULL, vrf_id); break; #endif #ifdef INET6 case AF_INET6: ip6->ip6_plen = htons((uint16_t)(len - sizeof(struct ip6_hdr))); if (port) { shout->checksum = sctp_calculate_cksum(mout, sizeof(struct ip6_hdr) + sizeof(struct udphdr)); SCTP_STAT_INCR(sctps_sendswcrc); if ((udp->uh_sum = in6_cksum(o_pak, IPPROTO_UDP, sizeof(struct ip6_hdr), len - sizeof(struct ip6_hdr))) == 0) { udp->uh_sum = 0xffff; } } else { mout->m_pkthdr.csum_flags = CSUM_SCTP_IPV6; mout->m_pkthdr.csum_data = offsetof(struct sctphdr, checksum); SCTP_STAT_INCR(sctps_sendhwcrc); } #ifdef SCTP_PACKET_LOGGING if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LAST_PACKET_TRACING) { sctp_packet_log(o_pak); } #endif SCTP_PROBE5(send, NULL, NULL, ip6, NULL, shout); SCTP_IP6_OUTPUT(ret, o_pak, NULL, NULL, NULL, vrf_id); break; #endif default: SCTPDBG(SCTP_DEBUG_OUTPUT1, "Unknown protocol (TSNH) type %d\n", dst->sa_family); sctp_m_freem(mout); SCTP_LTRACE_ERR_RET_PKT(mout, NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EFAULT); return; } SCTPDBG(SCTP_DEBUG_OUTPUT3, "return from send is %d\n", ret); if (port) { UDPSTAT_INC(udps_opackets); } SCTP_STAT_INCR(sctps_sendpackets); SCTP_STAT_INCR_COUNTER64(sctps_outpackets); SCTP_STAT_INCR_COUNTER64(sctps_outcontrolchunks); if (ret) { SCTP_STAT_INCR(sctps_senderrors); } return; } void sctp_send_shutdown_complete2(struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, uint8_t mflowtype, uint32_t mflowid, uint16_t fibnum, uint32_t vrf_id, uint16_t port) { sctp_send_resp_msg(src, dst, sh, 0, SCTP_SHUTDOWN_COMPLETE, NULL, mflowtype, mflowid, fibnum, vrf_id, port); } void sctp_send_hb(struct sctp_tcb *stcb, struct sctp_nets *net, int so_locked #if !defined(__APPLE__) && !defined(SCTP_SO_LOCK_TESTING) SCTP_UNUSED #endif ) { struct sctp_tmit_chunk *chk; struct sctp_heartbeat_chunk *hb; struct timeval now; SCTP_TCB_LOCK_ASSERT(stcb); if (net == NULL) { return; } (void)SCTP_GETTIME_TIMEVAL(&now); switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: break; #endif #ifdef INET6 case AF_INET6: break; #endif default: return; } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { SCTPDBG(SCTP_DEBUG_OUTPUT4, "Gak, can't get a chunk for hb\n"); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_HEARTBEAT_REQUEST; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->asoc = &stcb->asoc; chk->send_size = sizeof(struct sctp_heartbeat_chunk); chk->data = sctp_get_mbuf_for_msg(chk->send_size, 0, M_NOWAIT, 1, MT_HEADER); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, so_locked); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); SCTP_BUF_LEN(chk->data) = chk->send_size; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); /* Now we have a mbuf that we can fill in with the details */ hb = mtod(chk->data, struct sctp_heartbeat_chunk *); memset(hb, 0, sizeof(struct sctp_heartbeat_chunk)); /* fill out chunk header */ hb->ch.chunk_type = SCTP_HEARTBEAT_REQUEST; hb->ch.chunk_flags = 0; hb->ch.chunk_length = htons(chk->send_size); /* Fill out hb parameter */ hb->heartbeat.hb_info.ph.param_type = htons(SCTP_HEARTBEAT_INFO); hb->heartbeat.hb_info.ph.param_length = htons(sizeof(struct sctp_heartbeat_info_param)); hb->heartbeat.hb_info.time_value_1 = now.tv_sec; hb->heartbeat.hb_info.time_value_2 = now.tv_usec; /* Did our user request this one, put it in */ hb->heartbeat.hb_info.addr_family = (uint8_t)net->ro._l_addr.sa.sa_family; hb->heartbeat.hb_info.addr_len = net->ro._l_addr.sa.sa_len; if (net->dest_state & SCTP_ADDR_UNCONFIRMED) { /* * we only take from the entropy pool if the address is not * confirmed. */ net->heartbeat_random1 = hb->heartbeat.hb_info.random_value1 = sctp_select_initial_TSN(&stcb->sctp_ep->sctp_ep); net->heartbeat_random2 = hb->heartbeat.hb_info.random_value2 = sctp_select_initial_TSN(&stcb->sctp_ep->sctp_ep); } else { net->heartbeat_random1 = hb->heartbeat.hb_info.random_value1 = 0; net->heartbeat_random2 = hb->heartbeat.hb_info.random_value2 = 0; } switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: memcpy(hb->heartbeat.hb_info.address, &net->ro._l_addr.sin.sin_addr, sizeof(net->ro._l_addr.sin.sin_addr)); break; #endif #ifdef INET6 case AF_INET6: memcpy(hb->heartbeat.hb_info.address, &net->ro._l_addr.sin6.sin6_addr, sizeof(net->ro._l_addr.sin6.sin6_addr)); break; #endif default: if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } sctp_free_a_chunk(stcb, chk, so_locked); return; break; } net->hb_responded = 0; TAILQ_INSERT_TAIL(&stcb->asoc.control_send_queue, chk, sctp_next); stcb->asoc.ctrl_queue_cnt++; SCTP_STAT_INCR(sctps_sendheartbeat); return; } void sctp_send_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net, uint32_t high_tsn) { struct sctp_association *asoc; struct sctp_ecne_chunk *ecne; struct sctp_tmit_chunk *chk; if (net == NULL) { return; } asoc = &stcb->asoc; SCTP_TCB_LOCK_ASSERT(stcb); TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if ((chk->rec.chunk_id.id == SCTP_ECN_ECHO) && (net == chk->whoTo)) { /* found a previous ECN_ECHO update it if needed */ uint32_t cnt, ctsn; ecne = mtod(chk->data, struct sctp_ecne_chunk *); ctsn = ntohl(ecne->tsn); if (SCTP_TSN_GT(high_tsn, ctsn)) { ecne->tsn = htonl(high_tsn); SCTP_STAT_INCR(sctps_queue_upd_ecne); } cnt = ntohl(ecne->num_pkts_since_cwr); cnt++; ecne->num_pkts_since_cwr = htonl(cnt); return; } } /* nope could not find one to update so we must build one */ sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { return; } SCTP_STAT_INCR(sctps_queue_upd_ecne); chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_ECN_ECHO; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->asoc = &stcb->asoc; chk->send_size = sizeof(struct sctp_ecne_chunk); chk->data = sctp_get_mbuf_for_msg(chk->send_size, 0, M_NOWAIT, 1, MT_HEADER); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); SCTP_BUF_LEN(chk->data) = chk->send_size; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); stcb->asoc.ecn_echo_cnt_onq++; ecne = mtod(chk->data, struct sctp_ecne_chunk *); ecne->ch.chunk_type = SCTP_ECN_ECHO; ecne->ch.chunk_flags = 0; ecne->ch.chunk_length = htons(sizeof(struct sctp_ecne_chunk)); ecne->tsn = htonl(high_tsn); ecne->num_pkts_since_cwr = htonl(1); TAILQ_INSERT_HEAD(&stcb->asoc.control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; } void sctp_send_packet_dropped(struct sctp_tcb *stcb, struct sctp_nets *net, struct mbuf *m, int len, int iphlen, int bad_crc) { struct sctp_association *asoc; struct sctp_pktdrop_chunk *drp; struct sctp_tmit_chunk *chk; uint8_t *datap; int was_trunc = 0; int fullsz = 0; long spc; int offset; struct sctp_chunkhdr *ch, chunk_buf; unsigned int chk_length; if (!stcb) { return; } asoc = &stcb->asoc; SCTP_TCB_LOCK_ASSERT(stcb); if (asoc->pktdrop_supported == 0) { /*- * peer must declare support before I send one. */ return; } if (stcb->sctp_socket == NULL) { return; } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_PACKET_DROPPED; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; len -= iphlen; chk->send_size = len; /* Validate that we do not have an ABORT in here. */ offset = iphlen + sizeof(struct sctphdr); ch = (struct sctp_chunkhdr *)sctp_m_getptr(m, offset, sizeof(*ch), (uint8_t *)&chunk_buf); while (ch != NULL) { chk_length = ntohs(ch->chunk_length); if (chk_length < sizeof(*ch)) { /* break to abort land */ break; } switch (ch->chunk_type) { case SCTP_PACKET_DROPPED: case SCTP_ABORT_ASSOCIATION: case SCTP_INITIATION_ACK: /** * We don't respond with an PKT-DROP to an ABORT * or PKT-DROP. We also do not respond to an * INIT-ACK, because we can't know if the initiation * tag is correct or not. */ sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); return; default: break; } offset += SCTP_SIZE32(chk_length); ch = (struct sctp_chunkhdr *)sctp_m_getptr(m, offset, sizeof(*ch), (uint8_t *)&chunk_buf); } if ((len + SCTP_MAX_OVERHEAD + sizeof(struct sctp_pktdrop_chunk)) > min(stcb->asoc.smallest_mtu, MCLBYTES)) { /* * only send 1 mtu worth, trim off the excess on the end. */ fullsz = len; len = min(stcb->asoc.smallest_mtu, MCLBYTES) - SCTP_MAX_OVERHEAD; was_trunc = 1; } chk->asoc = &stcb->asoc; chk->data = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (chk->data == NULL) { jump_out: sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); drp = mtod(chk->data, struct sctp_pktdrop_chunk *); if (drp == NULL) { sctp_m_freem(chk->data); chk->data = NULL; goto jump_out; } chk->book_size = SCTP_SIZE32((chk->send_size + sizeof(struct sctp_pktdrop_chunk) + sizeof(struct sctphdr) + SCTP_MED_OVERHEAD)); chk->book_size_scale = 0; if (was_trunc) { drp->ch.chunk_flags = SCTP_PACKET_TRUNCATED; drp->trunc_len = htons(fullsz); /* * Len is already adjusted to size minus overhead above take * out the pkt_drop chunk itself from it. */ chk->send_size = (uint16_t)(len - sizeof(struct sctp_pktdrop_chunk)); len = chk->send_size; } else { /* no truncation needed */ drp->ch.chunk_flags = 0; drp->trunc_len = htons(0); } if (bad_crc) { drp->ch.chunk_flags |= SCTP_BADCRC; } chk->send_size += sizeof(struct sctp_pktdrop_chunk); SCTP_BUF_LEN(chk->data) = chk->send_size; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; if (net) { /* we should hit here */ chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); } else { chk->whoTo = NULL; } drp->ch.chunk_type = SCTP_PACKET_DROPPED; drp->ch.chunk_length = htons(chk->send_size); spc = SCTP_SB_LIMIT_RCV(stcb->sctp_socket); if (spc < 0) { spc = 0; } drp->bottle_bw = htonl(spc); if (asoc->my_rwnd) { drp->current_onq = htonl(asoc->size_on_reasm_queue + asoc->size_on_all_streams + asoc->my_rwnd_control_len + stcb->sctp_socket->so_rcv.sb_cc); } else { /*- * If my rwnd is 0, possibly from mbuf depletion as well as * space used, tell the peer there is NO space aka onq == bw */ drp->current_onq = htonl(spc); } drp->reserved = 0; datap = drp->data; m_copydata(m, iphlen, len, (caddr_t)datap); TAILQ_INSERT_TAIL(&stcb->asoc.control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; } void sctp_send_cwr(struct sctp_tcb *stcb, struct sctp_nets *net, uint32_t high_tsn, uint8_t override) { struct sctp_association *asoc; struct sctp_cwr_chunk *cwr; struct sctp_tmit_chunk *chk; SCTP_TCB_LOCK_ASSERT(stcb); if (net == NULL) { return; } asoc = &stcb->asoc; TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if ((chk->rec.chunk_id.id == SCTP_ECN_CWR) && (net == chk->whoTo)) { /* * found a previous CWR queued to same destination * update it if needed */ uint32_t ctsn; cwr = mtod(chk->data, struct sctp_cwr_chunk *); ctsn = ntohl(cwr->tsn); if (SCTP_TSN_GT(high_tsn, ctsn)) { cwr->tsn = htonl(high_tsn); } if (override & SCTP_CWR_REDUCE_OVERRIDE) { /* Make sure override is carried */ cwr->ch.chunk_flags |= SCTP_CWR_REDUCE_OVERRIDE; } return; } } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_ECN_CWR; chk->rec.chunk_id.can_take_data = 1; chk->flags = 0; chk->asoc = &stcb->asoc; chk->send_size = sizeof(struct sctp_cwr_chunk); chk->data = sctp_get_mbuf_for_msg(chk->send_size, 0, M_NOWAIT, 1, MT_HEADER); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); SCTP_BUF_LEN(chk->data) = chk->send_size; chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; chk->whoTo = net; atomic_add_int(&chk->whoTo->ref_count, 1); cwr = mtod(chk->data, struct sctp_cwr_chunk *); cwr->ch.chunk_type = SCTP_ECN_CWR; cwr->ch.chunk_flags = override; cwr->ch.chunk_length = htons(sizeof(struct sctp_cwr_chunk)); cwr->tsn = htonl(high_tsn); TAILQ_INSERT_TAIL(&stcb->asoc.control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; } static int sctp_add_stream_reset_out(struct sctp_tcb *stcb, struct sctp_tmit_chunk *chk, uint32_t seq, uint32_t resp_seq, uint32_t last_sent) { uint16_t len, old_len, i; struct sctp_stream_reset_out_request *req_out; struct sctp_chunkhdr *ch; int at; int number_entries = 0; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ req_out = (struct sctp_stream_reset_out_request *)((caddr_t)ch + len); /* now how long will this param be? */ for (i = 0; i < stcb->asoc.streamoutcnt; i++) { if ((stcb->asoc.strmout[i].state == SCTP_STREAM_RESET_PENDING) && (stcb->asoc.strmout[i].chunks_on_queues == 0) && TAILQ_EMPTY(&stcb->asoc.strmout[i].outqueue)) { number_entries++; } } if (number_entries == 0) { return (0); } if (number_entries == stcb->asoc.streamoutcnt) { number_entries = 0; } if (number_entries > SCTP_MAX_STREAMS_AT_ONCE_RESET) { number_entries = SCTP_MAX_STREAMS_AT_ONCE_RESET; } len = (uint16_t)(sizeof(struct sctp_stream_reset_out_request) + (sizeof(uint16_t) * number_entries)); req_out->ph.param_type = htons(SCTP_STR_RESET_OUT_REQUEST); req_out->ph.param_length = htons(len); req_out->request_seq = htonl(seq); req_out->response_seq = htonl(resp_seq); req_out->send_reset_at_tsn = htonl(last_sent); at = 0; if (number_entries) { for (i = 0; i < stcb->asoc.streamoutcnt; i++) { if ((stcb->asoc.strmout[i].state == SCTP_STREAM_RESET_PENDING) && (stcb->asoc.strmout[i].chunks_on_queues == 0) && TAILQ_EMPTY(&stcb->asoc.strmout[i].outqueue)) { req_out->list_of_streams[at] = htons(i); at++; stcb->asoc.strmout[i].state = SCTP_STREAM_RESET_IN_FLIGHT; if (at >= number_entries) { break; } } } } else { for (i = 0; i < stcb->asoc.streamoutcnt; i++) { stcb->asoc.strmout[i].state = SCTP_STREAM_RESET_IN_FLIGHT; } } if (SCTP_SIZE32(len) > len) { /*- * Need to worry about the pad we may end up adding to the * end. This is easy since the struct is either aligned to 4 * bytes or 2 bytes off. */ req_out->list_of_streams[number_entries] = 0; } /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->book_size = len + old_len; chk->book_size_scale = 0; chk->send_size = SCTP_SIZE32(chk->book_size); SCTP_BUF_LEN(chk->data) = chk->send_size; return (1); } static void sctp_add_stream_reset_in(struct sctp_tmit_chunk *chk, int number_entries, uint16_t *list, uint32_t seq) { uint16_t len, old_len, i; struct sctp_stream_reset_in_request *req_in; struct sctp_chunkhdr *ch; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ req_in = (struct sctp_stream_reset_in_request *)((caddr_t)ch + len); /* now how long will this param be? */ len = (uint16_t)(sizeof(struct sctp_stream_reset_in_request) + (sizeof(uint16_t) * number_entries)); req_in->ph.param_type = htons(SCTP_STR_RESET_IN_REQUEST); req_in->ph.param_length = htons(len); req_in->request_seq = htonl(seq); if (number_entries) { for (i = 0; i < number_entries; i++) { req_in->list_of_streams[i] = htons(list[i]); } } if (SCTP_SIZE32(len) > len) { /*- * Need to worry about the pad we may end up adding to the * end. This is easy since the struct is either aligned to 4 * bytes or 2 bytes off. */ req_in->list_of_streams[number_entries] = 0; } /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->book_size = len + old_len; chk->book_size_scale = 0; chk->send_size = SCTP_SIZE32(chk->book_size); SCTP_BUF_LEN(chk->data) = chk->send_size; return; } static void sctp_add_stream_reset_tsn(struct sctp_tmit_chunk *chk, uint32_t seq) { uint16_t len, old_len; struct sctp_stream_reset_tsn_request *req_tsn; struct sctp_chunkhdr *ch; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ req_tsn = (struct sctp_stream_reset_tsn_request *)((caddr_t)ch + len); /* now how long will this param be? */ len = sizeof(struct sctp_stream_reset_tsn_request); req_tsn->ph.param_type = htons(SCTP_STR_RESET_TSN_REQUEST); req_tsn->ph.param_length = htons(len); req_tsn->request_seq = htonl(seq); /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->send_size = len + old_len; chk->book_size = SCTP_SIZE32(chk->send_size); chk->book_size_scale = 0; SCTP_BUF_LEN(chk->data) = SCTP_SIZE32(chk->send_size); return; } void sctp_add_stream_reset_result(struct sctp_tmit_chunk *chk, uint32_t resp_seq, uint32_t result) { uint16_t len, old_len; struct sctp_stream_reset_response *resp; struct sctp_chunkhdr *ch; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ resp = (struct sctp_stream_reset_response *)((caddr_t)ch + len); /* now how long will this param be? */ len = sizeof(struct sctp_stream_reset_response); resp->ph.param_type = htons(SCTP_STR_RESET_RESPONSE); resp->ph.param_length = htons(len); resp->response_seq = htonl(resp_seq); resp->result = ntohl(result); /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->book_size = len + old_len; chk->book_size_scale = 0; chk->send_size = SCTP_SIZE32(chk->book_size); SCTP_BUF_LEN(chk->data) = chk->send_size; return; } void sctp_send_deferred_reset_response(struct sctp_tcb *stcb, struct sctp_stream_reset_list *ent, int response) { struct sctp_association *asoc; struct sctp_tmit_chunk *chk; struct sctp_chunkhdr *ch; asoc = &stcb->asoc; /* * Reset our last reset action to the new one IP -> response * (PERFORMED probably). This assures that if we fail to send, a * retran from the peer will get the new response. */ asoc->last_reset_action[0] = response; if (asoc->stream_reset_outstanding) { return; } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return; } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_STREAM_RESET; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->asoc = &stcb->asoc; chk->book_size = sizeof(struct sctp_chunkhdr); chk->send_size = SCTP_SIZE32(chk->book_size); chk->book_size_scale = 0; chk->data = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, SCTP_SO_LOCKED); SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return; } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); /* setup chunk parameters */ chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; if (stcb->asoc.alternate) { chk->whoTo = stcb->asoc.alternate; } else { chk->whoTo = stcb->asoc.primary_destination; } ch = mtod(chk->data, struct sctp_chunkhdr *); ch->chunk_type = SCTP_STREAM_RESET; ch->chunk_flags = 0; ch->chunk_length = htons(chk->book_size); atomic_add_int(&chk->whoTo->ref_count, 1); SCTP_BUF_LEN(chk->data) = chk->send_size; sctp_add_stream_reset_result(chk, ent->seq, response); /* insert the chunk for sending */ TAILQ_INSERT_TAIL(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; } void sctp_add_stream_reset_result_tsn(struct sctp_tmit_chunk *chk, uint32_t resp_seq, uint32_t result, uint32_t send_una, uint32_t recv_next) { uint16_t len, old_len; struct sctp_stream_reset_response_tsn *resp; struct sctp_chunkhdr *ch; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ resp = (struct sctp_stream_reset_response_tsn *)((caddr_t)ch + len); /* now how long will this param be? */ len = sizeof(struct sctp_stream_reset_response_tsn); resp->ph.param_type = htons(SCTP_STR_RESET_RESPONSE); resp->ph.param_length = htons(len); resp->response_seq = htonl(resp_seq); resp->result = htonl(result); resp->senders_next_tsn = htonl(send_una); resp->receivers_next_tsn = htonl(recv_next); /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->book_size = len + old_len; chk->send_size = SCTP_SIZE32(chk->book_size); chk->book_size_scale = 0; SCTP_BUF_LEN(chk->data) = chk->send_size; return; } static void sctp_add_an_out_stream(struct sctp_tmit_chunk *chk, uint32_t seq, uint16_t adding) { uint16_t len, old_len; struct sctp_chunkhdr *ch; struct sctp_stream_reset_add_strm *addstr; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ addstr = (struct sctp_stream_reset_add_strm *)((caddr_t)ch + len); /* now how long will this param be? */ len = sizeof(struct sctp_stream_reset_add_strm); /* Fill it out. */ addstr->ph.param_type = htons(SCTP_STR_RESET_ADD_OUT_STREAMS); addstr->ph.param_length = htons(len); addstr->request_seq = htonl(seq); addstr->number_of_streams = htons(adding); addstr->reserved = 0; /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->send_size = len + old_len; chk->book_size = SCTP_SIZE32(chk->send_size); chk->book_size_scale = 0; SCTP_BUF_LEN(chk->data) = SCTP_SIZE32(chk->send_size); return; } static void sctp_add_an_in_stream(struct sctp_tmit_chunk *chk, uint32_t seq, uint16_t adding) { uint16_t len, old_len; struct sctp_chunkhdr *ch; struct sctp_stream_reset_add_strm *addstr; ch = mtod(chk->data, struct sctp_chunkhdr *); old_len = len = SCTP_SIZE32(ntohs(ch->chunk_length)); /* get to new offset for the param. */ addstr = (struct sctp_stream_reset_add_strm *)((caddr_t)ch + len); /* now how long will this param be? */ len = sizeof(struct sctp_stream_reset_add_strm); /* Fill it out. */ addstr->ph.param_type = htons(SCTP_STR_RESET_ADD_IN_STREAMS); addstr->ph.param_length = htons(len); addstr->request_seq = htonl(seq); addstr->number_of_streams = htons(adding); addstr->reserved = 0; /* now fix the chunk length */ ch->chunk_length = htons(len + old_len); chk->send_size = len + old_len; chk->book_size = SCTP_SIZE32(chk->send_size); chk->book_size_scale = 0; SCTP_BUF_LEN(chk->data) = SCTP_SIZE32(chk->send_size); return; } int sctp_send_stream_reset_out_if_possible(struct sctp_tcb *stcb, int so_locked) { struct sctp_association *asoc; struct sctp_tmit_chunk *chk; struct sctp_chunkhdr *ch; uint32_t seq; asoc = &stcb->asoc; asoc->trigger_reset = 0; if (asoc->stream_reset_outstanding) { return (EALREADY); } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_STREAM_RESET; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->asoc = &stcb->asoc; chk->book_size = sizeof(struct sctp_chunkhdr); chk->send_size = SCTP_SIZE32(chk->book_size); chk->book_size_scale = 0; chk->data = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, so_locked); SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); /* setup chunk parameters */ chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; if (stcb->asoc.alternate) { chk->whoTo = stcb->asoc.alternate; } else { chk->whoTo = stcb->asoc.primary_destination; } ch = mtod(chk->data, struct sctp_chunkhdr *); ch->chunk_type = SCTP_STREAM_RESET; ch->chunk_flags = 0; ch->chunk_length = htons(chk->book_size); atomic_add_int(&chk->whoTo->ref_count, 1); SCTP_BUF_LEN(chk->data) = chk->send_size; seq = stcb->asoc.str_reset_seq_out; if (sctp_add_stream_reset_out(stcb, chk, seq, (stcb->asoc.str_reset_seq_in - 1), (stcb->asoc.sending_seq - 1))) { seq++; asoc->stream_reset_outstanding++; } else { m_freem(chk->data); chk->data = NULL; sctp_free_a_chunk(stcb, chk, so_locked); return (ENOENT); } asoc->str_reset = chk; /* insert the chunk for sending */ TAILQ_INSERT_TAIL(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; if (stcb->asoc.send_sack) { sctp_send_sack(stcb, so_locked); } sctp_timer_start(SCTP_TIMER_TYPE_STRRESET, stcb->sctp_ep, stcb, chk->whoTo); return (0); } int sctp_send_str_reset_req(struct sctp_tcb *stcb, uint16_t number_entries, uint16_t *list, uint8_t send_in_req, uint8_t send_tsn_req, uint8_t add_stream, uint16_t adding_o, uint16_t adding_i, uint8_t peer_asked) { struct sctp_association *asoc; struct sctp_tmit_chunk *chk; struct sctp_chunkhdr *ch; int can_send_out_req = 0; uint32_t seq; asoc = &stcb->asoc; if (asoc->stream_reset_outstanding) { /*- * Already one pending, must get ACK back to clear the flag. */ SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EBUSY); return (EBUSY); } if ((send_in_req == 0) && (send_tsn_req == 0) && (add_stream == 0)) { /* nothing to do */ SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } if (send_tsn_req && send_in_req) { /* error, can't do that */ SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } else if (send_in_req) { can_send_out_req = 1; } if (number_entries > (MCLBYTES - SCTP_MIN_OVERHEAD - sizeof(struct sctp_chunkhdr) - sizeof(struct sctp_stream_reset_out_request)) / sizeof(uint16_t)) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } sctp_alloc_a_chunk(stcb, chk); if (chk == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } chk->copy_by_ref = 0; chk->rec.chunk_id.id = SCTP_STREAM_RESET; chk->rec.chunk_id.can_take_data = 0; chk->flags = 0; chk->asoc = &stcb->asoc; chk->book_size = sizeof(struct sctp_chunkhdr); chk->send_size = SCTP_SIZE32(chk->book_size); chk->book_size_scale = 0; chk->data = sctp_get_mbuf_for_msg(MCLBYTES, 0, M_NOWAIT, 1, MT_DATA); if (chk->data == NULL) { sctp_free_a_chunk(stcb, chk, SCTP_SO_LOCKED); SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ENOMEM); return (ENOMEM); } SCTP_BUF_RESV_UF(chk->data, SCTP_MIN_OVERHEAD); /* setup chunk parameters */ chk->sent = SCTP_DATAGRAM_UNSENT; chk->snd_count = 0; if (stcb->asoc.alternate) { chk->whoTo = stcb->asoc.alternate; } else { chk->whoTo = stcb->asoc.primary_destination; } atomic_add_int(&chk->whoTo->ref_count, 1); ch = mtod(chk->data, struct sctp_chunkhdr *); ch->chunk_type = SCTP_STREAM_RESET; ch->chunk_flags = 0; ch->chunk_length = htons(chk->book_size); SCTP_BUF_LEN(chk->data) = chk->send_size; seq = stcb->asoc.str_reset_seq_out; if (can_send_out_req) { int ret; ret = sctp_add_stream_reset_out(stcb, chk, seq, (stcb->asoc.str_reset_seq_in - 1), (stcb->asoc.sending_seq - 1)); if (ret) { seq++; asoc->stream_reset_outstanding++; } } if ((add_stream & 1) && ((stcb->asoc.strm_realoutsize - stcb->asoc.streamoutcnt) < adding_o)) { /* Need to allocate more */ struct sctp_stream_out *oldstream; struct sctp_stream_queue_pending *sp, *nsp; int i; #if defined(SCTP_DETAILED_STR_STATS) int j; #endif oldstream = stcb->asoc.strmout; /* get some more */ SCTP_MALLOC(stcb->asoc.strmout, struct sctp_stream_out *, (stcb->asoc.streamoutcnt + adding_o) * sizeof(struct sctp_stream_out), SCTP_M_STRMO); if (stcb->asoc.strmout == NULL) { uint8_t x; stcb->asoc.strmout = oldstream; /* Turn off the bit */ x = add_stream & 0xfe; add_stream = x; goto skip_stuff; } /* * Ok now we proceed with copying the old out stuff and * initializing the new stuff. */ SCTP_TCB_SEND_LOCK(stcb); stcb->asoc.ss_functions.sctp_ss_clear(stcb, &stcb->asoc, 0, 1); for (i = 0; i < stcb->asoc.streamoutcnt; i++) { TAILQ_INIT(&stcb->asoc.strmout[i].outqueue); stcb->asoc.strmout[i].chunks_on_queues = oldstream[i].chunks_on_queues; stcb->asoc.strmout[i].next_mid_ordered = oldstream[i].next_mid_ordered; stcb->asoc.strmout[i].next_mid_unordered = oldstream[i].next_mid_unordered; stcb->asoc.strmout[i].last_msg_incomplete = oldstream[i].last_msg_incomplete; stcb->asoc.strmout[i].sid = i; stcb->asoc.strmout[i].state = oldstream[i].state; /* FIX ME FIX ME */ /* * This should be a SS_COPY operation FIX ME STREAM * SCHEDULER EXPERT */ stcb->asoc.ss_functions.sctp_ss_init_stream(stcb, &stcb->asoc.strmout[i], &oldstream[i]); /* now anything on those queues? */ TAILQ_FOREACH_SAFE(sp, &oldstream[i].outqueue, next, nsp) { TAILQ_REMOVE(&oldstream[i].outqueue, sp, next); TAILQ_INSERT_TAIL(&stcb->asoc.strmout[i].outqueue, sp, next); } } /* now the new streams */ stcb->asoc.ss_functions.sctp_ss_init(stcb, &stcb->asoc, 1); for (i = stcb->asoc.streamoutcnt; i < (stcb->asoc.streamoutcnt + adding_o); i++) { TAILQ_INIT(&stcb->asoc.strmout[i].outqueue); stcb->asoc.strmout[i].chunks_on_queues = 0; #if defined(SCTP_DETAILED_STR_STATS) for (j = 0; j < SCTP_PR_SCTP_MAX + 1; j++) { stcb->asoc.strmout[i].abandoned_sent[j] = 0; stcb->asoc.strmout[i].abandoned_unsent[j] = 0; } #else stcb->asoc.strmout[i].abandoned_sent[0] = 0; stcb->asoc.strmout[i].abandoned_unsent[0] = 0; #endif stcb->asoc.strmout[i].next_mid_ordered = 0; stcb->asoc.strmout[i].next_mid_unordered = 0; stcb->asoc.strmout[i].sid = i; stcb->asoc.strmout[i].last_msg_incomplete = 0; stcb->asoc.ss_functions.sctp_ss_init_stream(stcb, &stcb->asoc.strmout[i], NULL); stcb->asoc.strmout[i].state = SCTP_STREAM_CLOSED; } stcb->asoc.strm_realoutsize = stcb->asoc.streamoutcnt + adding_o; SCTP_FREE(oldstream, SCTP_M_STRMO); SCTP_TCB_SEND_UNLOCK(stcb); } skip_stuff: if ((add_stream & 1) && (adding_o > 0)) { asoc->strm_pending_add_size = adding_o; asoc->peer_req_out = peer_asked; sctp_add_an_out_stream(chk, seq, adding_o); seq++; asoc->stream_reset_outstanding++; } if ((add_stream & 2) && (adding_i > 0)) { sctp_add_an_in_stream(chk, seq, adding_i); seq++; asoc->stream_reset_outstanding++; } if (send_in_req) { sctp_add_stream_reset_in(chk, number_entries, list, seq); seq++; asoc->stream_reset_outstanding++; } if (send_tsn_req) { sctp_add_stream_reset_tsn(chk, seq); asoc->stream_reset_outstanding++; } asoc->str_reset = chk; /* insert the chunk for sending */ TAILQ_INSERT_TAIL(&asoc->control_send_queue, chk, sctp_next); asoc->ctrl_queue_cnt++; if (stcb->asoc.send_sack) { sctp_send_sack(stcb, SCTP_SO_LOCKED); } sctp_timer_start(SCTP_TIMER_TYPE_STRRESET, stcb->sctp_ep, stcb, chk->whoTo); return (0); } void sctp_send_abort(struct mbuf *m, int iphlen, struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, uint32_t vtag, struct mbuf *cause, uint8_t mflowtype, uint32_t mflowid, uint16_t fibnum, uint32_t vrf_id, uint16_t port) { /* Don't respond to an ABORT with an ABORT. */ if (sctp_is_there_an_abort_here(m, iphlen, &vtag)) { if (cause) sctp_m_freem(cause); return; } sctp_send_resp_msg(src, dst, sh, vtag, SCTP_ABORT_ASSOCIATION, cause, mflowtype, mflowid, fibnum, vrf_id, port); return; } void sctp_send_operr_to(struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, uint32_t vtag, struct mbuf *cause, uint8_t mflowtype, uint32_t mflowid, uint16_t fibnum, uint32_t vrf_id, uint16_t port) { sctp_send_resp_msg(src, dst, sh, vtag, SCTP_OPERATION_ERROR, cause, mflowtype, mflowid, fibnum, vrf_id, port); return; } static struct mbuf * sctp_copy_resume(struct uio *uio, int max_send_len, int user_marks_eor, int *error, uint32_t *sndout, struct mbuf **new_tail) { struct mbuf *m; m = m_uiotombuf(uio, M_WAITOK, max_send_len, 0, (M_PKTHDR | (user_marks_eor ? M_EOR : 0))); if (m == NULL) { SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, ENOBUFS); *error = ENOBUFS; } else { *sndout = m_length(m, NULL); *new_tail = m_last(m); } return (m); } static int sctp_copy_one(struct sctp_stream_queue_pending *sp, struct uio *uio, int resv_upfront) { sp->data = m_uiotombuf(uio, M_WAITOK, sp->length, resv_upfront, 0); if (sp->data == NULL) { SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, ENOBUFS); return (ENOBUFS); } sp->tail_mbuf = m_last(sp->data); return (0); } static struct sctp_stream_queue_pending * sctp_copy_it_in(struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_sndrcvinfo *srcv, struct uio *uio, struct sctp_nets *net, ssize_t max_send_len, int user_marks_eor, int *error) { /*- * This routine must be very careful in its work. Protocol * processing is up and running so care must be taken to spl...() * when you need to do something that may effect the stcb/asoc. The * sb is locked however. When data is copied the protocol processing * should be enabled since this is a slower operation... */ struct sctp_stream_queue_pending *sp = NULL; int resv_in_first; *error = 0; /* Now can we send this? */ if ((SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_SENT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_ACK_SENT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED) || (asoc->state & SCTP_STATE_SHUTDOWN_PENDING)) { /* got data while shutting down */ SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ECONNRESET); *error = ECONNRESET; goto out_now; } sctp_alloc_a_strmoq(stcb, sp); if (sp == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, ENOMEM); *error = ENOMEM; goto out_now; } sp->act_flags = 0; sp->sender_all_done = 0; sp->sinfo_flags = srcv->sinfo_flags; sp->timetolive = srcv->sinfo_timetolive; sp->ppid = srcv->sinfo_ppid; sp->context = srcv->sinfo_context; sp->fsn = 0; (void)SCTP_GETTIME_TIMEVAL(&sp->ts); sp->sid = srcv->sinfo_stream; sp->length = (uint32_t)min(uio->uio_resid, max_send_len); if ((sp->length == (uint32_t)uio->uio_resid) && ((user_marks_eor == 0) || (srcv->sinfo_flags & SCTP_EOF) || (user_marks_eor && (srcv->sinfo_flags & SCTP_EOR)))) { sp->msg_is_complete = 1; } else { sp->msg_is_complete = 0; } sp->sender_all_done = 0; sp->some_taken = 0; sp->put_last_out = 0; resv_in_first = SCTP_DATA_CHUNK_OVERHEAD(stcb); sp->data = sp->tail_mbuf = NULL; if (sp->length == 0) { goto skip_copy; } if (srcv->sinfo_keynumber_valid) { sp->auth_keyid = srcv->sinfo_keynumber; } else { sp->auth_keyid = stcb->asoc.authinfo.active_keyid; } if (sctp_auth_is_required_chunk(SCTP_DATA, stcb->asoc.peer_auth_chunks)) { sctp_auth_key_acquire(stcb, sp->auth_keyid); sp->holds_key_ref = 1; } *error = sctp_copy_one(sp, uio, resv_in_first); skip_copy: if (*error) { sctp_free_a_strmoq(stcb, sp, SCTP_SO_LOCKED); sp = NULL; } else { if (sp->sinfo_flags & SCTP_ADDR_OVER) { sp->net = net; atomic_add_int(&sp->net->ref_count, 1); } else { sp->net = NULL; } sctp_set_prsctp_policy(sp); } out_now: return (sp); } int sctp_sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *p ) { int error, use_sndinfo = 0; struct sctp_sndrcvinfo sndrcvninfo; struct sockaddr *addr_to_use; #if defined(INET) && defined(INET6) struct sockaddr_in sin; #endif if (control) { /* process cmsg snd/rcv info (maybe a assoc-id) */ if (sctp_find_cmsg(SCTP_SNDRCV, (void *)&sndrcvninfo, control, sizeof(sndrcvninfo))) { /* got one */ use_sndinfo = 1; } } addr_to_use = addr; #if defined(INET) && defined(INET6) if ((addr) && (addr->sa_family == AF_INET6)) { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)addr; if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) { in6_sin6_2_sin(&sin, sin6); addr_to_use = (struct sockaddr *)&sin; } } #endif error = sctp_lower_sosend(so, addr_to_use, uio, top, control, flags, use_sndinfo ? &sndrcvninfo : NULL ,p ); return (error); } int sctp_lower_sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *i_pak, struct mbuf *control, int flags, struct sctp_sndrcvinfo *srcv , struct thread *p ) { struct epoch_tracker et; ssize_t sndlen = 0, max_len, local_add_more; int error, len; struct mbuf *top = NULL; int queue_only = 0, queue_only_for_init = 0; int free_cnt_applied = 0; int un_sent; int now_filled = 0; unsigned int inqueue_bytes = 0; struct sctp_block_entry be; struct sctp_inpcb *inp; struct sctp_tcb *stcb = NULL; struct timeval now; struct sctp_nets *net; struct sctp_association *asoc; struct sctp_inpcb *t_inp; int user_marks_eor; int create_lock_applied = 0; int nagle_applies = 0; int some_on_control = 0; int got_all_of_the_send = 0; int hold_tcblock = 0; int non_blocking = 0; ssize_t local_soresv = 0; uint16_t port; uint16_t sinfo_flags; sctp_assoc_t sinfo_assoc_id; error = 0; net = NULL; stcb = NULL; asoc = NULL; t_inp = inp = (struct sctp_inpcb *)so->so_pcb; if (inp == NULL) { SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; if (i_pak) { SCTP_RELEASE_PKT(i_pak); } return (error); } if ((uio == NULL) && (i_pak == NULL)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } user_marks_eor = sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXPLICIT_EOR); atomic_add_int(&inp->total_sends, 1); if (uio) { if (uio->uio_resid < 0) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); return (EINVAL); } sndlen = uio->uio_resid; } else { top = SCTP_HEADER_TO_CHAIN(i_pak); sndlen = SCTP_HEADER_LEN(i_pak); } SCTPDBG(SCTP_DEBUG_OUTPUT1, "Send called addr:%p send length %zu\n", (void *)addr, sndlen); if ((inp->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE) && SCTP_IS_LISTENING(inp)) { /* The listener can NOT send */ SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, ENOTCONN); error = ENOTCONN; goto out_unlocked; } /** * Pre-screen address, if one is given the sin-len * must be set correctly! */ if (addr) { union sctp_sockstore *raddr = (union sctp_sockstore *)addr; switch (raddr->sa.sa_family) { #ifdef INET case AF_INET: if (raddr->sin.sin_len != sizeof(struct sockaddr_in)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } port = raddr->sin.sin_port; break; #endif #ifdef INET6 case AF_INET6: if (raddr->sin6.sin6_len != sizeof(struct sockaddr_in6)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } port = raddr->sin6.sin6_port; break; #endif default: SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EAFNOSUPPORT); error = EAFNOSUPPORT; goto out_unlocked; } } else port = 0; if (srcv) { sinfo_flags = srcv->sinfo_flags; sinfo_assoc_id = srcv->sinfo_assoc_id; if (INVALID_SINFO_FLAG(sinfo_flags) || PR_SCTP_INVALID_POLICY(sinfo_flags)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } if (srcv->sinfo_flags) SCTP_STAT_INCR(sctps_sends_with_flags); } else { sinfo_flags = inp->def_send.sinfo_flags; sinfo_assoc_id = inp->def_send.sinfo_assoc_id; } if (flags & MSG_EOR) { sinfo_flags |= SCTP_EOR; } if (flags & MSG_EOF) { sinfo_flags |= SCTP_EOF; } if (sinfo_flags & SCTP_SENDALL) { /* its a sendall */ error = sctp_sendall(inp, uio, top, srcv); top = NULL; goto out_unlocked; } if ((sinfo_flags & SCTP_ADDR_OVER) && (addr == NULL)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } /* now we must find the assoc */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_CONNECTED) || (inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL)) { SCTP_INP_RLOCK(inp); stcb = LIST_FIRST(&inp->sctp_asoc_list); if (stcb) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } SCTP_INP_RUNLOCK(inp); } else if (sinfo_assoc_id) { stcb = sctp_findassociation_ep_asocid(inp, sinfo_assoc_id, 1); if (stcb != NULL) { hold_tcblock = 1; } } else if (addr) { /*- * Since we did not use findep we must * increment it, and if we don't find a tcb * decrement it. */ SCTP_INP_WLOCK(inp); SCTP_INP_INCR_REF(inp); SCTP_INP_WUNLOCK(inp); stcb = sctp_findassociation_ep_addr(&t_inp, addr, &net, NULL, NULL); if (stcb == NULL) { SCTP_INP_WLOCK(inp); SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); } else { hold_tcblock = 1; } } if ((stcb == NULL) && (addr)) { /* Possible implicit send? */ SCTP_ASOC_CREATE_LOCK(inp); create_lock_applied = 1; if ((inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE)) { /* Should I really unlock ? */ SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } if (((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) == 0) && (addr->sa_family == AF_INET6)) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } SCTP_INP_WLOCK(inp); SCTP_INP_INCR_REF(inp); SCTP_INP_WUNLOCK(inp); /* With the lock applied look again */ stcb = sctp_findassociation_ep_addr(&t_inp, addr, &net, NULL, NULL); #if defined(INET) || defined(INET6) if ((stcb == NULL) && (control != NULL) && (port > 0)) { stcb = sctp_findassociation_cmsgs(&t_inp, port, control, &net, &error); } #endif if (stcb == NULL) { SCTP_INP_WLOCK(inp); SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); } else { hold_tcblock = 1; } if (error) { goto out_unlocked; } if (t_inp != inp) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, ENOTCONN); error = ENOTCONN; goto out_unlocked; } } if (stcb == NULL) { if (addr == NULL) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, ENOENT); error = ENOENT; goto out_unlocked; } else { /* We must go ahead and start the INIT process */ uint32_t vrf_id; if ((sinfo_flags & SCTP_ABORT) || ((sinfo_flags & SCTP_EOF) && (sndlen == 0))) { /*- * User asks to abort a non-existant assoc, * or EOF a non-existant assoc with no data */ SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, ENOENT); error = ENOENT; goto out_unlocked; } /* get an asoc/stcb struct */ vrf_id = inp->def_vrf_id; #ifdef INVARIANTS if (create_lock_applied == 0) { panic("Error, should hold create lock and I don't?"); } #endif stcb = sctp_aloc_assoc(inp, addr, &error, 0, vrf_id, inp->sctp_ep.pre_open_stream_count, inp->sctp_ep.port, p, SCTP_INITIALIZE_AUTH_PARAMS); if (stcb == NULL) { /* Error is setup for us in the call */ goto out_unlocked; } if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE) { stcb->sctp_ep->sctp_flags |= SCTP_PCB_FLAGS_CONNECTED; /* * Set the connected flag so we can queue * data */ soisconnecting(so); } hold_tcblock = 1; if (create_lock_applied) { SCTP_ASOC_CREATE_UNLOCK(inp); create_lock_applied = 0; } else { SCTP_PRINTF("Huh-3? create lock should have been on??\n"); } /* * Turn on queue only flag to prevent data from * being sent */ queue_only = 1; asoc = &stcb->asoc; SCTP_SET_STATE(stcb, SCTP_STATE_COOKIE_WAIT); (void)SCTP_GETTIME_TIMEVAL(&asoc->time_entered); if (control) { if (sctp_process_cmsgs_for_init(stcb, control, &error)) { sctp_free_assoc(inp, stcb, SCTP_PCBFREE_FORCE, SCTP_FROM_SCTP_OUTPUT + SCTP_LOC_6); hold_tcblock = 0; stcb = NULL; goto out_unlocked; } } /* out with the INIT */ queue_only_for_init = 1; /*- * we may want to dig in after this call and adjust the MTU * value. It defaulted to 1500 (constant) but the ro * structure may now have an update and thus we may need to * change it BEFORE we append the message. */ } } else asoc = &stcb->asoc; if (srcv == NULL) { srcv = (struct sctp_sndrcvinfo *)&asoc->def_send; sinfo_flags = srcv->sinfo_flags; if (flags & MSG_EOR) { sinfo_flags |= SCTP_EOR; } if (flags & MSG_EOF) { sinfo_flags |= SCTP_EOF; } } if (sinfo_flags & SCTP_ADDR_OVER) { if (addr) net = sctp_findnet(stcb, addr); else net = NULL; if ((net == NULL) || ((port != 0) && (port != stcb->rport))) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } } else { if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } } atomic_add_int(&stcb->total_sends, 1); /* Keep the stcb from being freed under our feet */ atomic_add_int(&asoc->refcnt, 1); free_cnt_applied = 1; if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_NO_FRAGMENT)) { if (sndlen > (ssize_t)asoc->smallest_mtu) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EMSGSIZE); error = EMSGSIZE; goto out_unlocked; } } if (SCTP_SO_IS_NBIO(so) || (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0 ) { non_blocking = 1; } /* would we block? */ if (non_blocking) { ssize_t amount; if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); if (user_marks_eor == 0) { amount = sndlen; } else { amount = 1; } if ((SCTP_SB_LIMIT_SND(so) < (amount + inqueue_bytes + stcb->asoc.sb_send_resv)) || (stcb->asoc.chunks_on_out_queue >= SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue))) { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EWOULDBLOCK); if (sndlen > (ssize_t)SCTP_SB_LIMIT_SND(so)) error = EMSGSIZE; else error = EWOULDBLOCK; goto out_unlocked; } stcb->asoc.sb_send_resv += (uint32_t)sndlen; SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } else { atomic_add_int(&stcb->asoc.sb_send_resv, sndlen); } local_soresv = sndlen; if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ECONNRESET); error = ECONNRESET; goto out_unlocked; } if (create_lock_applied) { SCTP_ASOC_CREATE_UNLOCK(inp); create_lock_applied = 0; } /* Is the stream no. valid? */ if (srcv->sinfo_stream >= asoc->streamoutcnt) { /* Invalid stream number */ SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } if ((asoc->strmout[srcv->sinfo_stream].state != SCTP_STREAM_OPEN) && (asoc->strmout[srcv->sinfo_stream].state != SCTP_STREAM_OPENING)) { /* * Can't queue any data while stream reset is underway. */ if (asoc->strmout[srcv->sinfo_stream].state > SCTP_STREAM_OPEN) { error = EAGAIN; } else { error = EINVAL; } SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, error); goto out_unlocked; } if ((SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_WAIT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED)) { queue_only = 1; } /* we are now done with all control */ if (control) { sctp_m_freem(control); control = NULL; } if ((SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_SENT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_ACK_SENT) || (asoc->state & SCTP_STATE_SHUTDOWN_PENDING)) { if (sinfo_flags & SCTP_ABORT) { ; } else { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ECONNRESET); error = ECONNRESET; goto out_unlocked; } } /* Ok, we will attempt a msgsnd :> */ if (p) { p->td_ru.ru_msgsnd++; } /* Are we aborting? */ if (sinfo_flags & SCTP_ABORT) { struct mbuf *mm; ssize_t tot_demand, tot_out = 0, max_out; SCTP_STAT_INCR(sctps_sends_with_abort); if ((SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_WAIT) || (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED)) { /* It has to be up before we abort */ /* how big is the user initiated abort? */ SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out; } if (hold_tcblock) { SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } if (top) { struct mbuf *cntm = NULL; mm = sctp_get_mbuf_for_msg(sizeof(struct sctp_paramhdr), 0, M_WAITOK, 1, MT_DATA); if (sndlen != 0) { for (cntm = top; cntm; cntm = SCTP_BUF_NEXT(cntm)) { tot_out += SCTP_BUF_LEN(cntm); } } } else { /* Must fit in a MTU */ tot_out = sndlen; tot_demand = (tot_out + sizeof(struct sctp_paramhdr)); if (tot_demand > SCTP_DEFAULT_ADD_MORE) { /* To big */ SCTP_LTRACE_ERR_RET(NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, EMSGSIZE); error = EMSGSIZE; goto out; } mm = sctp_get_mbuf_for_msg((unsigned int)tot_demand, 0, M_WAITOK, 1, MT_DATA); } if (mm == NULL) { SCTP_LTRACE_ERR_RET(NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, ENOMEM); error = ENOMEM; goto out; } max_out = asoc->smallest_mtu - sizeof(struct sctp_paramhdr); max_out -= sizeof(struct sctp_abort_msg); if (tot_out > max_out) { tot_out = max_out; } if (mm) { struct sctp_paramhdr *ph; /* now move forward the data pointer */ ph = mtod(mm, struct sctp_paramhdr *); ph->param_type = htons(SCTP_CAUSE_USER_INITIATED_ABT); ph->param_length = htons((uint16_t)(sizeof(struct sctp_paramhdr) + tot_out)); ph++; SCTP_BUF_LEN(mm) = (int)(tot_out + sizeof(struct sctp_paramhdr)); if (top == NULL) { error = uiomove((caddr_t)ph, (int)tot_out, uio); if (error) { /*- * Here if we can't get his data we * still abort we just don't get to * send the users note :-0 */ sctp_m_freem(mm); mm = NULL; } } else { if (sndlen != 0) { SCTP_BUF_NEXT(mm) = top; } } } if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); } atomic_add_int(&stcb->asoc.refcnt, -1); free_cnt_applied = 0; /* release this lock, otherwise we hang on ourselves */ NET_EPOCH_ENTER(et); sctp_abort_an_association(stcb->sctp_ep, stcb, mm, SCTP_SO_LOCKED); NET_EPOCH_EXIT(et); /* now relock the stcb so everything is sane */ hold_tcblock = 0; stcb = NULL; /* * In this case top is already chained to mm avoid double * free, since we free it below if top != NULL and driver * would free it after sending the packet out */ if (sndlen != 0) { top = NULL; } goto out_unlocked; } /* Calculate the maximum we can send */ inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); if (SCTP_SB_LIMIT_SND(so) > inqueue_bytes) { max_len = SCTP_SB_LIMIT_SND(so) - inqueue_bytes; } else { max_len = 0; } if (hold_tcblock) { SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } if (asoc->strmout == NULL) { /* huh? software error */ SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EFAULT); error = EFAULT; goto out_unlocked; } /* Unless E_EOR mode is on, we must make a send FIT in one call. */ if ((user_marks_eor == 0) && (sndlen > (ssize_t)SCTP_SB_LIMIT_SND(stcb->sctp_socket))) { /* It will NEVER fit */ SCTP_LTRACE_ERR_RET(NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, EMSGSIZE); error = EMSGSIZE; goto out_unlocked; } if ((uio == NULL) && user_marks_eor) { /*- * We do not support eeor mode for * sending with mbuf chains (like sendfile). */ SCTP_LTRACE_ERR_RET(NULL, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out_unlocked; } if (user_marks_eor) { local_add_more = (ssize_t)min(SCTP_SB_LIMIT_SND(so), SCTP_BASE_SYSCTL(sctp_add_more_threshold)); } else { /*- * For non-eeor the whole message must fit in * the socket send buffer. */ local_add_more = sndlen; } len = 0; if (non_blocking) { goto skip_preblock; } if (((max_len <= local_add_more) && ((ssize_t)SCTP_SB_LIMIT_SND(so) >= local_add_more)) || (max_len == 0) || ((stcb->asoc.chunks_on_out_queue + stcb->asoc.stream_queue_cnt) >= SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue))) { /* No room right now ! */ SOCKBUF_LOCK(&so->so_snd); inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); while ((SCTP_SB_LIMIT_SND(so) < (inqueue_bytes + local_add_more)) || ((stcb->asoc.stream_queue_cnt + stcb->asoc.chunks_on_out_queue) >= SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue))) { SCTPDBG(SCTP_DEBUG_OUTPUT1, "pre_block limit:%u <(inq:%d + %zd) || (%d+%d > %d)\n", (unsigned int)SCTP_SB_LIMIT_SND(so), inqueue_bytes, local_add_more, stcb->asoc.stream_queue_cnt, stcb->asoc.chunks_on_out_queue, SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue)); if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_log_block(SCTP_BLOCK_LOG_INTO_BLKA, asoc, sndlen); } be.error = 0; stcb->block_entry = &be; error = sbwait(&so->so_snd); stcb->block_entry = NULL; if (error || so->so_error || be.error) { if (error == 0) { if (so->so_error) error = so->so_error; if (be.error) { error = be.error; } } SOCKBUF_UNLOCK(&so->so_snd); goto out_unlocked; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_log_block(SCTP_BLOCK_LOG_OUTOF_BLK, asoc, stcb->asoc.total_output_queue_size); } if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SOCKBUF_UNLOCK(&so->so_snd); goto out_unlocked; } inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); } if (SCTP_SB_LIMIT_SND(so) > inqueue_bytes) { max_len = SCTP_SB_LIMIT_SND(so) - inqueue_bytes; } else { max_len = 0; } SOCKBUF_UNLOCK(&so->so_snd); } skip_preblock: if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { goto out_unlocked; } /* * sndlen covers for mbuf case uio_resid covers for the non-mbuf * case NOTE: uio will be null when top/mbuf is passed */ if (sndlen == 0) { if (sinfo_flags & SCTP_EOF) { got_all_of_the_send = 1; goto dataless_eof; } else { SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out; } } if (top == NULL) { struct sctp_stream_queue_pending *sp; struct sctp_stream_out *strm; uint32_t sndout; SCTP_TCB_SEND_LOCK(stcb); if ((asoc->stream_locked) && (asoc->stream_locked_on != srcv->sinfo_stream)) { SCTP_TCB_SEND_UNLOCK(stcb); SCTP_LTRACE_ERR_RET(inp, stcb, net, SCTP_FROM_SCTP_OUTPUT, EINVAL); error = EINVAL; goto out; } SCTP_TCB_SEND_UNLOCK(stcb); strm = &stcb->asoc.strmout[srcv->sinfo_stream]; if (strm->last_msg_incomplete == 0) { do_a_copy_in: sp = sctp_copy_it_in(stcb, asoc, srcv, uio, net, max_len, user_marks_eor, &error); if (error) { goto out; } SCTP_TCB_SEND_LOCK(stcb); if (sp->msg_is_complete) { strm->last_msg_incomplete = 0; asoc->stream_locked = 0; } else { /* * Just got locked to this guy in case of an * interrupt. */ strm->last_msg_incomplete = 1; if (stcb->asoc.idata_supported == 0) { asoc->stream_locked = 1; asoc->stream_locked_on = srcv->sinfo_stream; } sp->sender_all_done = 0; } sctp_snd_sb_alloc(stcb, sp->length); atomic_add_int(&asoc->stream_queue_cnt, 1); if (sinfo_flags & SCTP_UNORDERED) { SCTP_STAT_INCR(sctps_sends_with_unord); } TAILQ_INSERT_TAIL(&strm->outqueue, sp, next); stcb->asoc.ss_functions.sctp_ss_add_to_stream(stcb, asoc, strm, sp, 1); SCTP_TCB_SEND_UNLOCK(stcb); } else { SCTP_TCB_SEND_LOCK(stcb); sp = TAILQ_LAST(&strm->outqueue, sctp_streamhead); SCTP_TCB_SEND_UNLOCK(stcb); if (sp == NULL) { /* ???? Huh ??? last msg is gone */ #ifdef INVARIANTS panic("Warning: Last msg marked incomplete, yet nothing left?"); #else SCTP_PRINTF("Warning: Last msg marked incomplete, yet nothing left?\n"); strm->last_msg_incomplete = 0; #endif goto do_a_copy_in; } } while (uio->uio_resid > 0) { /* How much room do we have? */ struct mbuf *new_tail, *mm; inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); if (SCTP_SB_LIMIT_SND(so) > inqueue_bytes) max_len = SCTP_SB_LIMIT_SND(so) - inqueue_bytes; else max_len = 0; if ((max_len > (ssize_t)SCTP_BASE_SYSCTL(sctp_add_more_threshold)) || (max_len && (SCTP_SB_LIMIT_SND(so) < SCTP_BASE_SYSCTL(sctp_add_more_threshold))) || (uio->uio_resid && (uio->uio_resid <= max_len))) { sndout = 0; new_tail = NULL; if (hold_tcblock) { SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } mm = sctp_copy_resume(uio, (int)max_len, user_marks_eor, &error, &sndout, &new_tail); if ((mm == NULL) || error) { if (mm) { sctp_m_freem(mm); } goto out; } /* Update the mbuf and count */ SCTP_TCB_SEND_LOCK(stcb); if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { /* * we need to get out. Peer probably * aborted. */ sctp_m_freem(mm); if (stcb->asoc.state & SCTP_PCB_FLAGS_WAS_ABORTED) { SCTP_LTRACE_ERR_RET(NULL, stcb, NULL, SCTP_FROM_SCTP_OUTPUT, ECONNRESET); error = ECONNRESET; } SCTP_TCB_SEND_UNLOCK(stcb); goto out; } if (sp->tail_mbuf) { /* tack it to the end */ SCTP_BUF_NEXT(sp->tail_mbuf) = mm; sp->tail_mbuf = new_tail; } else { /* A stolen mbuf */ sp->data = mm; sp->tail_mbuf = new_tail; } sctp_snd_sb_alloc(stcb, sndout); atomic_add_int(&sp->length, sndout); len += sndout; if (sinfo_flags & SCTP_SACK_IMMEDIATELY) { sp->sinfo_flags |= SCTP_SACK_IMMEDIATELY; } /* Did we reach EOR? */ if ((uio->uio_resid == 0) && ((user_marks_eor == 0) || (sinfo_flags & SCTP_EOF) || (user_marks_eor && (sinfo_flags & SCTP_EOR)))) { sp->msg_is_complete = 1; } else { sp->msg_is_complete = 0; } SCTP_TCB_SEND_UNLOCK(stcb); } if (uio->uio_resid == 0) { /* got it all? */ continue; } /* PR-SCTP? */ if ((asoc->prsctp_supported) && (asoc->sent_queue_cnt_removeable > 0)) { /* * This is ugly but we must assure locking * order */ if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } sctp_prune_prsctp(stcb, asoc, srcv, (int)sndlen); inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); if (SCTP_SB_LIMIT_SND(so) > inqueue_bytes) max_len = SCTP_SB_LIMIT_SND(so) - inqueue_bytes; else max_len = 0; if (max_len > 0) { continue; } SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } /* wait for space now */ if (non_blocking) { /* Non-blocking io in place out */ goto skip_out_eof; } /* What about the INIT, send it maybe */ if (queue_only_for_init) { if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } if (SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) { /* a collision took us forward? */ queue_only = 0; } else { NET_EPOCH_ENTER(et); sctp_send_initiate(inp, stcb, SCTP_SO_LOCKED); NET_EPOCH_EXIT(et); SCTP_SET_STATE(stcb, SCTP_STATE_COOKIE_WAIT); queue_only = 1; } } if ((net->flight_size > net->cwnd) && (asoc->sctp_cmt_on_off == 0)) { SCTP_STAT_INCR(sctps_send_cwnd_avoid); queue_only = 1; } else if (asoc->ifp_had_enobuf) { SCTP_STAT_INCR(sctps_ifnomemqueued); if (net->flight_size > (2 * net->mtu)) { queue_only = 1; } asoc->ifp_had_enobuf = 0; } un_sent = stcb->asoc.total_output_queue_size - stcb->asoc.total_flight; if ((sctp_is_feature_off(inp, SCTP_PCB_FLAGS_NODELAY)) && (stcb->asoc.total_flight > 0) && (stcb->asoc.stream_queue_cnt < SCTP_MAX_DATA_BUNDLING) && (un_sent < (int)(stcb->asoc.smallest_mtu - SCTP_MIN_OVERHEAD))) { /*- * Ok, Nagle is set on and we have data outstanding. * Don't send anything and let SACKs drive out the * data unless we have a "full" segment to send. */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_NAGLE_LOGGING_ENABLE) { sctp_log_nagle_event(stcb, SCTP_NAGLE_APPLIED); } SCTP_STAT_INCR(sctps_naglequeued); nagle_applies = 1; } else { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_NAGLE_LOGGING_ENABLE) { if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_NODELAY)) sctp_log_nagle_event(stcb, SCTP_NAGLE_SKIPPED); } SCTP_STAT_INCR(sctps_naglesent); nagle_applies = 0; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_misc_ints(SCTP_CWNDLOG_PRESEND, queue_only_for_init, queue_only, nagle_applies, un_sent); sctp_misc_ints(SCTP_CWNDLOG_PRESEND, stcb->asoc.total_output_queue_size, stcb->asoc.total_flight, stcb->asoc.chunks_on_out_queue, stcb->asoc.total_flight_count); } if (queue_only_for_init) queue_only_for_init = 0; if ((queue_only == 0) && (nagle_applies == 0)) { /*- * need to start chunk output * before blocking.. note that if * a lock is already applied, then * the input via the net is happening * and I don't need to start output :-D */ NET_EPOCH_ENTER(et); if (hold_tcblock == 0) { if (SCTP_TCB_TRYLOCK(stcb)) { hold_tcblock = 1; sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_LOCKED); } } else { sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_LOCKED); } NET_EPOCH_EXIT(et); } if (hold_tcblock == 1) { SCTP_TCB_UNLOCK(stcb); hold_tcblock = 0; } SOCKBUF_LOCK(&so->so_snd); /*- * This is a bit strange, but I think it will * work. The total_output_queue_size is locked and * protected by the TCB_LOCK, which we just released. * There is a race that can occur between releasing it * above, and me getting the socket lock, where sacks * come in but we have not put the SB_WAIT on the * so_snd buffer to get the wakeup. After the LOCK * is applied the sack_processing will also need to * LOCK the so->so_snd to do the actual sowwakeup(). So * once we have the socket buffer lock if we recheck the * size we KNOW we will get to sleep safely with the * wakeup flag in place. */ inqueue_bytes = stcb->asoc.total_output_queue_size - (stcb->asoc.chunks_on_out_queue * SCTP_DATA_CHUNK_OVERHEAD(stcb)); if (SCTP_SB_LIMIT_SND(so) <= (inqueue_bytes + min(SCTP_BASE_SYSCTL(sctp_add_more_threshold), SCTP_SB_LIMIT_SND(so)))) { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_log_block(SCTP_BLOCK_LOG_INTO_BLK, asoc, uio->uio_resid); } be.error = 0; stcb->block_entry = &be; error = sbwait(&so->so_snd); stcb->block_entry = NULL; if (error || so->so_error || be.error) { if (error == 0) { if (so->so_error) error = so->so_error; if (be.error) { error = be.error; } } SOCKBUF_UNLOCK(&so->so_snd); goto out_unlocked; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_log_block(SCTP_BLOCK_LOG_OUTOF_BLK, asoc, stcb->asoc.total_output_queue_size); } } SOCKBUF_UNLOCK(&so->so_snd); if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { goto out_unlocked; } } SCTP_TCB_SEND_LOCK(stcb); if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_TCB_SEND_UNLOCK(stcb); goto out_unlocked; } if (sp) { if (sp->msg_is_complete == 0) { strm->last_msg_incomplete = 1; if (stcb->asoc.idata_supported == 0) { asoc->stream_locked = 1; asoc->stream_locked_on = srcv->sinfo_stream; } } else { sp->sender_all_done = 1; strm->last_msg_incomplete = 0; asoc->stream_locked = 0; } } else { SCTP_PRINTF("Huh no sp TSNH?\n"); strm->last_msg_incomplete = 0; asoc->stream_locked = 0; } SCTP_TCB_SEND_UNLOCK(stcb); if (uio->uio_resid == 0) { got_all_of_the_send = 1; } } else { /* We send in a 0, since we do NOT have any locks */ error = sctp_msg_append(stcb, net, top, srcv, 0); top = NULL; if (sinfo_flags & SCTP_EOF) { /* * This should only happen for Panda for the mbuf * send case, which does NOT yet support EEOR mode. * Thus, we can just set this flag to do the proper * EOF handling. */ got_all_of_the_send = 1; } } if (error) { goto out; } dataless_eof: /* EOF thing ? */ if ((sinfo_flags & SCTP_EOF) && (got_all_of_the_send == 1)) { SCTP_STAT_INCR(sctps_sends_with_eof); error = 0; if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue) && sctp_is_there_unsent_data(stcb, SCTP_SO_LOCKED) == 0) { if ((*asoc->ss_functions.sctp_ss_is_user_msgs_incomplete) (stcb, asoc)) { goto abort_anyway; } /* there is nothing queued to send, so I'm done... */ if ((SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_SENT) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_RECEIVED) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_ACK_SENT)) { struct sctp_nets *netp; /* only send SHUTDOWN the first time through */ if (SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } SCTP_SET_STATE(stcb, SCTP_STATE_SHUTDOWN_SENT); sctp_stop_timers_for_shutdown(stcb); if (stcb->asoc.alternate) { netp = stcb->asoc.alternate; } else { netp = stcb->asoc.primary_destination; } sctp_send_shutdown(stcb, netp); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN, stcb->sctp_ep, stcb, netp); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, stcb->sctp_ep, stcb, NULL); } } else { /*- * we still got (or just got) data to send, so set * SHUTDOWN_PENDING */ /*- * XXX sockets draft says that SCTP_EOF should be * sent with no data. currently, we will allow user * data to be sent first and move to * SHUTDOWN-PENDING */ if ((SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_SENT) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_RECEIVED) && (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_ACK_SENT)) { if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } if ((*asoc->ss_functions.sctp_ss_is_user_msgs_incomplete) (stcb, asoc)) { SCTP_ADD_SUBSTATE(stcb, SCTP_STATE_PARTIAL_MSG_LEFT); } SCTP_ADD_SUBSTATE(stcb, SCTP_STATE_SHUTDOWN_PENDING); if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue) && (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT)) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; abort_anyway: if (free_cnt_applied) { atomic_add_int(&stcb->asoc.refcnt, -1); free_cnt_applied = 0; } snprintf(msg, sizeof(msg), "%s:%d at %s", __FILE__, __LINE__, __func__); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); NET_EPOCH_ENTER(et); sctp_abort_an_association(stcb->sctp_ep, stcb, op_err, SCTP_SO_LOCKED); NET_EPOCH_EXIT(et); /* * now relock the stcb so everything * is sane */ hold_tcblock = 0; stcb = NULL; goto out; } sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, stcb->sctp_ep, stcb, NULL); sctp_feature_off(inp, SCTP_PCB_FLAGS_NODELAY); } } } skip_out_eof: if (!TAILQ_EMPTY(&stcb->asoc.control_send_queue)) { some_on_control = 1; } if (queue_only_for_init) { if (hold_tcblock == 0) { SCTP_TCB_LOCK(stcb); hold_tcblock = 1; } if (SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) { /* a collision took us forward? */ queue_only = 0; } else { NET_EPOCH_ENTER(et); sctp_send_initiate(inp, stcb, SCTP_SO_LOCKED); NET_EPOCH_EXIT(et); SCTP_SET_STATE(stcb, SCTP_STATE_COOKIE_WAIT); queue_only = 1; } } if ((net->flight_size > net->cwnd) && (stcb->asoc.sctp_cmt_on_off == 0)) { SCTP_STAT_INCR(sctps_send_cwnd_avoid); queue_only = 1; } else if (asoc->ifp_had_enobuf) { SCTP_STAT_INCR(sctps_ifnomemqueued); if (net->flight_size > (2 * net->mtu)) { queue_only = 1; } asoc->ifp_had_enobuf = 0; } un_sent = stcb->asoc.total_output_queue_size - stcb->asoc.total_flight; if ((sctp_is_feature_off(inp, SCTP_PCB_FLAGS_NODELAY)) && (stcb->asoc.total_flight > 0) && (stcb->asoc.stream_queue_cnt < SCTP_MAX_DATA_BUNDLING) && (un_sent < (int)(stcb->asoc.smallest_mtu - SCTP_MIN_OVERHEAD))) { /*- * Ok, Nagle is set on and we have data outstanding. * Don't send anything and let SACKs drive out the * data unless wen have a "full" segment to send. */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_NAGLE_LOGGING_ENABLE) { sctp_log_nagle_event(stcb, SCTP_NAGLE_APPLIED); } SCTP_STAT_INCR(sctps_naglequeued); nagle_applies = 1; } else { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_NAGLE_LOGGING_ENABLE) { if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_NODELAY)) sctp_log_nagle_event(stcb, SCTP_NAGLE_SKIPPED); } SCTP_STAT_INCR(sctps_naglesent); nagle_applies = 0; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_BLK_LOGGING_ENABLE) { sctp_misc_ints(SCTP_CWNDLOG_PRESEND, queue_only_for_init, queue_only, nagle_applies, un_sent); sctp_misc_ints(SCTP_CWNDLOG_PRESEND, stcb->asoc.total_output_queue_size, stcb->asoc.total_flight, stcb->asoc.chunks_on_out_queue, stcb->asoc.total_flight_count); } NET_EPOCH_ENTER(et); if ((queue_only == 0) && (nagle_applies == 0) && (stcb->asoc.peers_rwnd && un_sent)) { /* we can attempt to send too. */ if (hold_tcblock == 0) { /* * If there is activity recv'ing sacks no need to * send */ if (SCTP_TCB_TRYLOCK(stcb)) { sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_LOCKED); hold_tcblock = 1; } } else { sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_LOCKED); } } else if ((queue_only == 0) && (stcb->asoc.peers_rwnd == 0) && (stcb->asoc.total_flight == 0)) { /* We get to have a probe outstanding */ if (hold_tcblock == 0) { hold_tcblock = 1; SCTP_TCB_LOCK(stcb); } sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_USR_SEND, SCTP_SO_LOCKED); } else if (some_on_control) { int num_out, reason, frag_point; /* Here we do control only */ if (hold_tcblock == 0) { hold_tcblock = 1; SCTP_TCB_LOCK(stcb); } frag_point = sctp_get_frag_point(stcb, &stcb->asoc); (void)sctp_med_chunk_output(inp, stcb, &stcb->asoc, &num_out, &reason, 1, 1, &now, &now_filled, frag_point, SCTP_SO_LOCKED); } NET_EPOCH_EXIT(et); SCTPDBG(SCTP_DEBUG_OUTPUT1, "USR Send complete qo:%d prw:%d unsent:%d tf:%d cooq:%d toqs:%d err:%d\n", queue_only, stcb->asoc.peers_rwnd, un_sent, stcb->asoc.total_flight, stcb->asoc.chunks_on_out_queue, stcb->asoc.total_output_queue_size, error); out: out_unlocked: if (local_soresv && stcb) { atomic_subtract_int(&stcb->asoc.sb_send_resv, sndlen); } if (create_lock_applied) { SCTP_ASOC_CREATE_UNLOCK(inp); } if ((stcb) && hold_tcblock) { SCTP_TCB_UNLOCK(stcb); } if (stcb && free_cnt_applied) { atomic_add_int(&stcb->asoc.refcnt, -1); } #ifdef INVARIANTS if (stcb) { if (mtx_owned(&stcb->tcb_mtx)) { panic("Leaving with tcb mtx owned?"); } if (mtx_owned(&stcb->tcb_send_mtx)) { panic("Leaving with tcb send mtx owned?"); } } #endif if (top) { sctp_m_freem(top); } if (control) { sctp_m_freem(control); } return (error); } /* * generate an AUTHentication chunk, if required */ struct mbuf * sctp_add_auth_chunk(struct mbuf *m, struct mbuf **m_end, struct sctp_auth_chunk **auth_ret, uint32_t *offset, struct sctp_tcb *stcb, uint8_t chunk) { struct mbuf *m_auth; struct sctp_auth_chunk *auth; int chunk_len; struct mbuf *cn; if ((m_end == NULL) || (auth_ret == NULL) || (offset == NULL) || (stcb == NULL)) return (m); if (stcb->asoc.auth_supported == 0) { return (m); } /* does the requested chunk require auth? */ if (!sctp_auth_is_required_chunk(chunk, stcb->asoc.peer_auth_chunks)) { return (m); } m_auth = sctp_get_mbuf_for_msg(sizeof(*auth), 0, M_NOWAIT, 1, MT_HEADER); if (m_auth == NULL) { /* no mbuf's */ return (m); } /* reserve some space if this will be the first mbuf */ if (m == NULL) SCTP_BUF_RESV_UF(m_auth, SCTP_MIN_OVERHEAD); /* fill in the AUTH chunk details */ auth = mtod(m_auth, struct sctp_auth_chunk *); memset(auth, 0, sizeof(*auth)); auth->ch.chunk_type = SCTP_AUTHENTICATION; auth->ch.chunk_flags = 0; chunk_len = sizeof(*auth) + sctp_get_hmac_digest_len(stcb->asoc.peer_hmac_id); auth->ch.chunk_length = htons(chunk_len); auth->hmac_id = htons(stcb->asoc.peer_hmac_id); /* key id and hmac digest will be computed and filled in upon send */ /* save the offset where the auth was inserted into the chain */ *offset = 0; for (cn = m; cn; cn = SCTP_BUF_NEXT(cn)) { *offset += SCTP_BUF_LEN(cn); } /* update length and return pointer to the auth chunk */ SCTP_BUF_LEN(m_auth) = chunk_len; m = sctp_copy_mbufchain(m_auth, m, m_end, 1, chunk_len, 0); if (auth_ret != NULL) *auth_ret = auth; return (m); } #ifdef INET6 int sctp_v6src_match_nexthop(struct sockaddr_in6 *src6, sctp_route_t *ro) { struct nd_prefix *pfx = NULL; struct nd_pfxrouter *pfxrtr = NULL; struct sockaddr_in6 gw6; - if (ro == NULL || ro->ro_rt == NULL || src6->sin6_family != AF_INET6) + if (ro == NULL || ro->ro_nh == NULL || src6->sin6_family != AF_INET6) return (0); /* get prefix entry of address */ ND6_RLOCK(); LIST_FOREACH(pfx, &MODULE_GLOBAL(nd_prefix), ndpr_entry) { if (pfx->ndpr_stateflags & NDPRF_DETACHED) continue; if (IN6_ARE_MASKED_ADDR_EQUAL(&pfx->ndpr_prefix.sin6_addr, &src6->sin6_addr, &pfx->ndpr_mask)) break; } /* no prefix entry in the prefix list */ if (pfx == NULL) { ND6_RUNLOCK(); SCTPDBG(SCTP_DEBUG_OUTPUT2, "No prefix entry for "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, (struct sockaddr *)src6); return (0); } SCTPDBG(SCTP_DEBUG_OUTPUT2, "v6src_match_nexthop(), Prefix entry is "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, (struct sockaddr *)src6); /* search installed gateway from prefix entry */ LIST_FOREACH(pfxrtr, &pfx->ndpr_advrtrs, pfr_entry) { memset(&gw6, 0, sizeof(struct sockaddr_in6)); gw6.sin6_family = AF_INET6; gw6.sin6_len = sizeof(struct sockaddr_in6); memcpy(&gw6.sin6_addr, &pfxrtr->router->rtaddr, sizeof(struct in6_addr)); SCTPDBG(SCTP_DEBUG_OUTPUT2, "prefix router is "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, (struct sockaddr *)&gw6); SCTPDBG(SCTP_DEBUG_OUTPUT2, "installed router is "); - SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, ro->ro_rt->rt_gateway); - if (sctp_cmpaddr((struct sockaddr *)&gw6, ro->ro_rt->rt_gateway)) { + SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &ro->ro_nh->gw_sa); + if (sctp_cmpaddr((struct sockaddr *)&gw6, &ro->ro_nh->gw_sa)) { ND6_RUNLOCK(); SCTPDBG(SCTP_DEBUG_OUTPUT2, "pfxrouter is installed\n"); return (1); } } ND6_RUNLOCK(); SCTPDBG(SCTP_DEBUG_OUTPUT2, "pfxrouter is not installed\n"); return (0); } #endif int sctp_v4src_match_nexthop(struct sctp_ifa *sifa, sctp_route_t *ro) { #ifdef INET struct sockaddr_in *sin, *mask; struct ifaddr *ifa; struct in_addr srcnetaddr, gwnetaddr; - if (ro == NULL || ro->ro_rt == NULL || + if (ro == NULL || ro->ro_nh == NULL || sifa->address.sa.sa_family != AF_INET) { return (0); } ifa = (struct ifaddr *)sifa->ifa; mask = (struct sockaddr_in *)(ifa->ifa_netmask); sin = &sifa->address.sin; srcnetaddr.s_addr = (sin->sin_addr.s_addr & mask->sin_addr.s_addr); SCTPDBG(SCTP_DEBUG_OUTPUT1, "match_nexthop4: src address is "); SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &sifa->address.sa); SCTPDBG(SCTP_DEBUG_OUTPUT1, "network address is %x\n", srcnetaddr.s_addr); - sin = (struct sockaddr_in *)ro->ro_rt->rt_gateway; + sin = &ro->ro_nh->gw4_sa; gwnetaddr.s_addr = (sin->sin_addr.s_addr & mask->sin_addr.s_addr); SCTPDBG(SCTP_DEBUG_OUTPUT1, "match_nexthop4: nexthop is "); - SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, ro->ro_rt->rt_gateway); + SCTPDBG_ADDR(SCTP_DEBUG_OUTPUT2, &ro->ro_nh->gw4_sa); SCTPDBG(SCTP_DEBUG_OUTPUT1, "network address is %x\n", gwnetaddr.s_addr); if (srcnetaddr.s_addr == gwnetaddr.s_addr) { return (1); } #endif return (0); } Index: head/sys/netinet/sctp_pcb.c =================================================================== --- head/sys/netinet/sctp_pcb.c (revision 360291) +++ head/sys/netinet/sctp_pcb.c (revision 360292) @@ -1,7140 +1,7127 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2008, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #endif #ifdef INET6 #include #endif #include #include #include VNET_DEFINE(struct sctp_base_info, system_base_info); /* FIX: we don't handle multiple link local scopes */ /* "scopeless" replacement IN6_ARE_ADDR_EQUAL */ #ifdef INET6 int SCTP6_ARE_ADDR_EQUAL(struct sockaddr_in6 *a, struct sockaddr_in6 *b) { struct sockaddr_in6 tmp_a, tmp_b; memcpy(&tmp_a, a, sizeof(struct sockaddr_in6)); if (sa6_embedscope(&tmp_a, MODULE_GLOBAL(ip6_use_defzone)) != 0) { return (0); } memcpy(&tmp_b, b, sizeof(struct sockaddr_in6)); if (sa6_embedscope(&tmp_b, MODULE_GLOBAL(ip6_use_defzone)) != 0) { return (0); } return (IN6_ARE_ADDR_EQUAL(&tmp_a.sin6_addr, &tmp_b.sin6_addr)); } #endif void sctp_fill_pcbinfo(struct sctp_pcbinfo *spcb) { /* * We really don't need to lock this, but I will just because it * does not hurt. */ SCTP_INP_INFO_RLOCK(); spcb->ep_count = SCTP_BASE_INFO(ipi_count_ep); spcb->asoc_count = SCTP_BASE_INFO(ipi_count_asoc); spcb->laddr_count = SCTP_BASE_INFO(ipi_count_laddr); spcb->raddr_count = SCTP_BASE_INFO(ipi_count_raddr); spcb->chk_count = SCTP_BASE_INFO(ipi_count_chunk); spcb->readq_count = SCTP_BASE_INFO(ipi_count_readq); spcb->stream_oque = SCTP_BASE_INFO(ipi_count_strmoq); spcb->free_chunks = SCTP_BASE_INFO(ipi_free_chunks); SCTP_INP_INFO_RUNLOCK(); } /*- * Addresses are added to VRF's (Virtual Router's). For BSD we * have only the default VRF 0. We maintain a hash list of * VRF's. Each VRF has its own list of sctp_ifn's. Each of * these has a list of addresses. When we add a new address * to a VRF we lookup the ifn/ifn_index, if the ifn does * not exist we create it and add it to the list of IFN's * within the VRF. Once we have the sctp_ifn, we add the * address to the list. So we look something like: * * hash-vrf-table * vrf-> ifn-> ifn -> ifn * vrf | * ... +--ifa-> ifa -> ifa * vrf * * We keep these separate lists since the SCTP subsystem will * point to these from its source address selection nets structure. * When an address is deleted it does not happen right away on * the SCTP side, it gets scheduled. What we do when a * delete happens is immediately remove the address from * the master list and decrement the refcount. As our * addip iterator works through and frees the src address * selection pointing to the sctp_ifa, eventually the refcount * will reach 0 and we will delete it. Note that it is assumed * that any locking on system level ifn/ifa is done at the * caller of these functions and these routines will only * lock the SCTP structures as they add or delete things. * * Other notes on VRF concepts. * - An endpoint can be in multiple VRF's * - An association lives within a VRF and only one VRF. * - Any incoming packet we can deduce the VRF for by * looking at the mbuf/pak inbound (for BSD its VRF=0 :D) * - Any downward send call or connect call must supply the * VRF via ancillary data or via some sort of set default * VRF socket option call (again for BSD no brainer since * the VRF is always 0). * - An endpoint may add multiple VRF's to it. * - Listening sockets can accept associations in any * of the VRF's they are in but the assoc will end up * in only one VRF (gotten from the packet or connect/send). * */ struct sctp_vrf * sctp_allocate_vrf(int vrf_id) { struct sctp_vrf *vrf = NULL; struct sctp_vrflist *bucket; /* First allocate the VRF structure */ vrf = sctp_find_vrf(vrf_id); if (vrf) { /* Already allocated */ return (vrf); } SCTP_MALLOC(vrf, struct sctp_vrf *, sizeof(struct sctp_vrf), SCTP_M_VRF); if (vrf == NULL) { /* No memory */ #ifdef INVARIANTS panic("No memory for VRF:%d", vrf_id); #endif return (NULL); } /* setup the VRF */ memset(vrf, 0, sizeof(struct sctp_vrf)); vrf->vrf_id = vrf_id; LIST_INIT(&vrf->ifnlist); vrf->total_ifa_count = 0; vrf->refcount = 0; /* now also setup table ids */ SCTP_INIT_VRF_TABLEID(vrf); /* Init the HASH of addresses */ vrf->vrf_addr_hash = SCTP_HASH_INIT(SCTP_VRF_ADDR_HASH_SIZE, &vrf->vrf_addr_hashmark); if (vrf->vrf_addr_hash == NULL) { /* No memory */ #ifdef INVARIANTS panic("No memory for VRF:%d", vrf_id); #endif SCTP_FREE(vrf, SCTP_M_VRF); return (NULL); } /* Add it to the hash table */ bucket = &SCTP_BASE_INFO(sctp_vrfhash)[(vrf_id & SCTP_BASE_INFO(hashvrfmark))]; LIST_INSERT_HEAD(bucket, vrf, next_vrf); atomic_add_int(&SCTP_BASE_INFO(ipi_count_vrfs), 1); return (vrf); } struct sctp_ifn * sctp_find_ifn(void *ifn, uint32_t ifn_index) { struct sctp_ifn *sctp_ifnp; struct sctp_ifnlist *hash_ifn_head; /* * We assume the lock is held for the addresses if that's wrong * problems could occur :-) */ hash_ifn_head = &SCTP_BASE_INFO(vrf_ifn_hash)[(ifn_index & SCTP_BASE_INFO(vrf_ifn_hashmark))]; LIST_FOREACH(sctp_ifnp, hash_ifn_head, next_bucket) { if (sctp_ifnp->ifn_index == ifn_index) { return (sctp_ifnp); } if (sctp_ifnp->ifn_p && ifn && (sctp_ifnp->ifn_p == ifn)) { return (sctp_ifnp); } } return (NULL); } struct sctp_vrf * sctp_find_vrf(uint32_t vrf_id) { struct sctp_vrflist *bucket; struct sctp_vrf *liste; bucket = &SCTP_BASE_INFO(sctp_vrfhash)[(vrf_id & SCTP_BASE_INFO(hashvrfmark))]; LIST_FOREACH(liste, bucket, next_vrf) { if (vrf_id == liste->vrf_id) { return (liste); } } return (NULL); } void sctp_free_vrf(struct sctp_vrf *vrf) { if (SCTP_DECREMENT_AND_CHECK_REFCOUNT(&vrf->refcount)) { if (vrf->vrf_addr_hash) { SCTP_HASH_FREE(vrf->vrf_addr_hash, vrf->vrf_addr_hashmark); vrf->vrf_addr_hash = NULL; } /* We zero'd the count */ LIST_REMOVE(vrf, next_vrf); SCTP_FREE(vrf, SCTP_M_VRF); atomic_subtract_int(&SCTP_BASE_INFO(ipi_count_vrfs), 1); } } void sctp_free_ifn(struct sctp_ifn *sctp_ifnp) { if (SCTP_DECREMENT_AND_CHECK_REFCOUNT(&sctp_ifnp->refcount)) { /* We zero'd the count */ if (sctp_ifnp->vrf) { sctp_free_vrf(sctp_ifnp->vrf); } SCTP_FREE(sctp_ifnp, SCTP_M_IFN); atomic_subtract_int(&SCTP_BASE_INFO(ipi_count_ifns), 1); } } void sctp_update_ifn_mtu(uint32_t ifn_index, uint32_t mtu) { struct sctp_ifn *sctp_ifnp; sctp_ifnp = sctp_find_ifn((void *)NULL, ifn_index); if (sctp_ifnp != NULL) { sctp_ifnp->ifn_mtu = mtu; } } void sctp_free_ifa(struct sctp_ifa *sctp_ifap) { if (SCTP_DECREMENT_AND_CHECK_REFCOUNT(&sctp_ifap->refcount)) { /* We zero'd the count */ if (sctp_ifap->ifn_p) { sctp_free_ifn(sctp_ifap->ifn_p); } SCTP_FREE(sctp_ifap, SCTP_M_IFA); atomic_subtract_int(&SCTP_BASE_INFO(ipi_count_ifas), 1); } } static void sctp_delete_ifn(struct sctp_ifn *sctp_ifnp, int hold_addr_lock) { struct sctp_ifn *found; found = sctp_find_ifn(sctp_ifnp->ifn_p, sctp_ifnp->ifn_index); if (found == NULL) { /* Not in the list.. sorry */ return; } if (hold_addr_lock == 0) SCTP_IPI_ADDR_WLOCK(); LIST_REMOVE(sctp_ifnp, next_bucket); LIST_REMOVE(sctp_ifnp, next_ifn); SCTP_DEREGISTER_INTERFACE(sctp_ifnp->ifn_index, sctp_ifnp->registered_af); if (hold_addr_lock == 0) SCTP_IPI_ADDR_WUNLOCK(); /* Take away the reference, and possibly free it */ sctp_free_ifn(sctp_ifnp); } void sctp_mark_ifa_addr_down(uint32_t vrf_id, struct sockaddr *addr, const char *if_name, uint32_t ifn_index) { struct sctp_vrf *vrf; struct sctp_ifa *sctp_ifap; SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "Can't find vrf_id 0x%x\n", vrf_id); goto out; } sctp_ifap = sctp_find_ifa_by_addr(addr, vrf->vrf_id, SCTP_ADDR_LOCKED); if (sctp_ifap == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "Can't find sctp_ifap for address\n"); goto out; } if (sctp_ifap->ifn_p == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "IFA has no IFN - can't mark unusable\n"); goto out; } if (if_name) { if (strncmp(if_name, sctp_ifap->ifn_p->ifn_name, SCTP_IFNAMSIZ) != 0) { SCTPDBG(SCTP_DEBUG_PCB4, "IFN %s of IFA not the same as %s\n", sctp_ifap->ifn_p->ifn_name, if_name); goto out; } } else { if (sctp_ifap->ifn_p->ifn_index != ifn_index) { SCTPDBG(SCTP_DEBUG_PCB4, "IFA owned by ifn_index:%d down command for ifn_index:%d - ignored\n", sctp_ifap->ifn_p->ifn_index, ifn_index); goto out; } } sctp_ifap->localifa_flags &= (~SCTP_ADDR_VALID); sctp_ifap->localifa_flags |= SCTP_ADDR_IFA_UNUSEABLE; out: SCTP_IPI_ADDR_RUNLOCK(); } void sctp_mark_ifa_addr_up(uint32_t vrf_id, struct sockaddr *addr, const char *if_name, uint32_t ifn_index) { struct sctp_vrf *vrf; struct sctp_ifa *sctp_ifap; SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "Can't find vrf_id 0x%x\n", vrf_id); goto out; } sctp_ifap = sctp_find_ifa_by_addr(addr, vrf->vrf_id, SCTP_ADDR_LOCKED); if (sctp_ifap == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "Can't find sctp_ifap for address\n"); goto out; } if (sctp_ifap->ifn_p == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "IFA has no IFN - can't mark unusable\n"); goto out; } if (if_name) { if (strncmp(if_name, sctp_ifap->ifn_p->ifn_name, SCTP_IFNAMSIZ) != 0) { SCTPDBG(SCTP_DEBUG_PCB4, "IFN %s of IFA not the same as %s\n", sctp_ifap->ifn_p->ifn_name, if_name); goto out; } } else { if (sctp_ifap->ifn_p->ifn_index != ifn_index) { SCTPDBG(SCTP_DEBUG_PCB4, "IFA owned by ifn_index:%d down command for ifn_index:%d - ignored\n", sctp_ifap->ifn_p->ifn_index, ifn_index); goto out; } } sctp_ifap->localifa_flags &= (~SCTP_ADDR_IFA_UNUSEABLE); sctp_ifap->localifa_flags |= SCTP_ADDR_VALID; out: SCTP_IPI_ADDR_RUNLOCK(); } /*- * Add an ifa to an ifn. * Register the interface as necessary. * NOTE: ADDR write lock MUST be held. */ static void sctp_add_ifa_to_ifn(struct sctp_ifn *sctp_ifnp, struct sctp_ifa *sctp_ifap) { int ifa_af; LIST_INSERT_HEAD(&sctp_ifnp->ifalist, sctp_ifap, next_ifa); sctp_ifap->ifn_p = sctp_ifnp; atomic_add_int(&sctp_ifap->ifn_p->refcount, 1); /* update address counts */ sctp_ifnp->ifa_count++; ifa_af = sctp_ifap->address.sa.sa_family; switch (ifa_af) { #ifdef INET case AF_INET: sctp_ifnp->num_v4++; break; #endif #ifdef INET6 case AF_INET6: sctp_ifnp->num_v6++; break; #endif default: break; } if (sctp_ifnp->ifa_count == 1) { /* register the new interface */ SCTP_REGISTER_INTERFACE(sctp_ifnp->ifn_index, ifa_af); sctp_ifnp->registered_af = ifa_af; } } /*- * Remove an ifa from its ifn. * If no more addresses exist, remove the ifn too. Otherwise, re-register * the interface based on the remaining address families left. * NOTE: ADDR write lock MUST be held. */ static void sctp_remove_ifa_from_ifn(struct sctp_ifa *sctp_ifap) { LIST_REMOVE(sctp_ifap, next_ifa); if (sctp_ifap->ifn_p) { /* update address counts */ sctp_ifap->ifn_p->ifa_count--; switch (sctp_ifap->address.sa.sa_family) { #ifdef INET case AF_INET: sctp_ifap->ifn_p->num_v4--; break; #endif #ifdef INET6 case AF_INET6: sctp_ifap->ifn_p->num_v6--; break; #endif default: break; } if (LIST_EMPTY(&sctp_ifap->ifn_p->ifalist)) { /* remove the ifn, possibly freeing it */ sctp_delete_ifn(sctp_ifap->ifn_p, SCTP_ADDR_LOCKED); } else { /* re-register address family type, if needed */ if ((sctp_ifap->ifn_p->num_v6 == 0) && (sctp_ifap->ifn_p->registered_af == AF_INET6)) { SCTP_DEREGISTER_INTERFACE(sctp_ifap->ifn_p->ifn_index, AF_INET6); SCTP_REGISTER_INTERFACE(sctp_ifap->ifn_p->ifn_index, AF_INET); sctp_ifap->ifn_p->registered_af = AF_INET; } else if ((sctp_ifap->ifn_p->num_v4 == 0) && (sctp_ifap->ifn_p->registered_af == AF_INET)) { SCTP_DEREGISTER_INTERFACE(sctp_ifap->ifn_p->ifn_index, AF_INET); SCTP_REGISTER_INTERFACE(sctp_ifap->ifn_p->ifn_index, AF_INET6); sctp_ifap->ifn_p->registered_af = AF_INET6; } /* free the ifn refcount */ sctp_free_ifn(sctp_ifap->ifn_p); } sctp_ifap->ifn_p = NULL; } } struct sctp_ifa * sctp_add_addr_to_vrf(uint32_t vrf_id, void *ifn, uint32_t ifn_index, uint32_t ifn_type, const char *if_name, void *ifa, struct sockaddr *addr, uint32_t ifa_flags, int dynamic_add) { struct sctp_vrf *vrf; struct sctp_ifn *sctp_ifnp = NULL; struct sctp_ifa *sctp_ifap = NULL; struct sctp_ifalist *hash_addr_head; struct sctp_ifnlist *hash_ifn_head; uint32_t hash_of_addr; int new_ifn_af = 0; #ifdef SCTP_DEBUG SCTPDBG(SCTP_DEBUG_PCB4, "vrf_id 0x%x: adding address: ", vrf_id); SCTPDBG_ADDR(SCTP_DEBUG_PCB4, addr); #endif SCTP_IPI_ADDR_WLOCK(); sctp_ifnp = sctp_find_ifn(ifn, ifn_index); if (sctp_ifnp) { vrf = sctp_ifnp->vrf; } else { vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { vrf = sctp_allocate_vrf(vrf_id); if (vrf == NULL) { SCTP_IPI_ADDR_WUNLOCK(); return (NULL); } } } if (sctp_ifnp == NULL) { /* * build one and add it, can't hold lock until after malloc * done though. */ SCTP_IPI_ADDR_WUNLOCK(); SCTP_MALLOC(sctp_ifnp, struct sctp_ifn *, sizeof(struct sctp_ifn), SCTP_M_IFN); if (sctp_ifnp == NULL) { #ifdef INVARIANTS panic("No memory for IFN"); #endif return (NULL); } memset(sctp_ifnp, 0, sizeof(struct sctp_ifn)); sctp_ifnp->ifn_index = ifn_index; sctp_ifnp->ifn_p = ifn; sctp_ifnp->ifn_type = ifn_type; sctp_ifnp->refcount = 0; sctp_ifnp->vrf = vrf; atomic_add_int(&vrf->refcount, 1); sctp_ifnp->ifn_mtu = SCTP_GATHER_MTU_FROM_IFN_INFO(ifn, ifn_index, addr->sa_family); if (if_name != NULL) { snprintf(sctp_ifnp->ifn_name, SCTP_IFNAMSIZ, "%s", if_name); } else { snprintf(sctp_ifnp->ifn_name, SCTP_IFNAMSIZ, "%s", "unknown"); } hash_ifn_head = &SCTP_BASE_INFO(vrf_ifn_hash)[(ifn_index & SCTP_BASE_INFO(vrf_ifn_hashmark))]; LIST_INIT(&sctp_ifnp->ifalist); SCTP_IPI_ADDR_WLOCK(); LIST_INSERT_HEAD(hash_ifn_head, sctp_ifnp, next_bucket); LIST_INSERT_HEAD(&vrf->ifnlist, sctp_ifnp, next_ifn); atomic_add_int(&SCTP_BASE_INFO(ipi_count_ifns), 1); new_ifn_af = 1; } sctp_ifap = sctp_find_ifa_by_addr(addr, vrf->vrf_id, SCTP_ADDR_LOCKED); if (sctp_ifap) { /* Hmm, it already exists? */ if ((sctp_ifap->ifn_p) && (sctp_ifap->ifn_p->ifn_index == ifn_index)) { SCTPDBG(SCTP_DEBUG_PCB4, "Using existing ifn %s (0x%x) for ifa %p\n", sctp_ifap->ifn_p->ifn_name, ifn_index, (void *)sctp_ifap); if (new_ifn_af) { /* Remove the created one that we don't want */ sctp_delete_ifn(sctp_ifnp, SCTP_ADDR_LOCKED); } if (sctp_ifap->localifa_flags & SCTP_BEING_DELETED) { /* easy to solve, just switch back to active */ SCTPDBG(SCTP_DEBUG_PCB4, "Clearing deleted ifa flag\n"); sctp_ifap->localifa_flags = SCTP_ADDR_VALID; sctp_ifap->ifn_p = sctp_ifnp; atomic_add_int(&sctp_ifap->ifn_p->refcount, 1); } exit_stage_left: SCTP_IPI_ADDR_WUNLOCK(); return (sctp_ifap); } else { if (sctp_ifap->ifn_p) { /* * The last IFN gets the address, remove the * old one */ SCTPDBG(SCTP_DEBUG_PCB4, "Moving ifa %p from %s (0x%x) to %s (0x%x)\n", (void *)sctp_ifap, sctp_ifap->ifn_p->ifn_name, sctp_ifap->ifn_p->ifn_index, if_name, ifn_index); /* remove the address from the old ifn */ sctp_remove_ifa_from_ifn(sctp_ifap); /* move the address over to the new ifn */ sctp_add_ifa_to_ifn(sctp_ifnp, sctp_ifap); goto exit_stage_left; } else { /* repair ifnp which was NULL ? */ sctp_ifap->localifa_flags = SCTP_ADDR_VALID; SCTPDBG(SCTP_DEBUG_PCB4, "Repairing ifn %p for ifa %p\n", (void *)sctp_ifnp, (void *)sctp_ifap); sctp_add_ifa_to_ifn(sctp_ifnp, sctp_ifap); } goto exit_stage_left; } } SCTP_IPI_ADDR_WUNLOCK(); SCTP_MALLOC(sctp_ifap, struct sctp_ifa *, sizeof(struct sctp_ifa), SCTP_M_IFA); if (sctp_ifap == NULL) { #ifdef INVARIANTS panic("No memory for IFA"); #endif return (NULL); } memset(sctp_ifap, 0, sizeof(struct sctp_ifa)); sctp_ifap->ifn_p = sctp_ifnp; atomic_add_int(&sctp_ifnp->refcount, 1); sctp_ifap->vrf_id = vrf_id; sctp_ifap->ifa = ifa; memcpy(&sctp_ifap->address, addr, addr->sa_len); sctp_ifap->localifa_flags = SCTP_ADDR_VALID | SCTP_ADDR_DEFER_USE; sctp_ifap->flags = ifa_flags; /* Set scope */ switch (sctp_ifap->address.sa.sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; sin = &sctp_ifap->address.sin; if (SCTP_IFN_IS_IFT_LOOP(sctp_ifap->ifn_p) || (IN4_ISLOOPBACK_ADDRESS(&sin->sin_addr))) { sctp_ifap->src_is_loop = 1; } if ((IN4_ISPRIVATE_ADDRESS(&sin->sin_addr))) { sctp_ifap->src_is_priv = 1; } sctp_ifnp->num_v4++; if (new_ifn_af) new_ifn_af = AF_INET; break; } #endif #ifdef INET6 case AF_INET6: { /* ok to use deprecated addresses? */ struct sockaddr_in6 *sin6; sin6 = &sctp_ifap->address.sin6; if (SCTP_IFN_IS_IFT_LOOP(sctp_ifap->ifn_p) || (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr))) { sctp_ifap->src_is_loop = 1; } if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) { sctp_ifap->src_is_priv = 1; } sctp_ifnp->num_v6++; if (new_ifn_af) new_ifn_af = AF_INET6; break; } #endif default: new_ifn_af = 0; break; } hash_of_addr = sctp_get_ifa_hash_val(&sctp_ifap->address.sa); if ((sctp_ifap->src_is_priv == 0) && (sctp_ifap->src_is_loop == 0)) { sctp_ifap->src_is_glob = 1; } SCTP_IPI_ADDR_WLOCK(); hash_addr_head = &vrf->vrf_addr_hash[(hash_of_addr & vrf->vrf_addr_hashmark)]; LIST_INSERT_HEAD(hash_addr_head, sctp_ifap, next_bucket); sctp_ifap->refcount = 1; LIST_INSERT_HEAD(&sctp_ifnp->ifalist, sctp_ifap, next_ifa); sctp_ifnp->ifa_count++; vrf->total_ifa_count++; atomic_add_int(&SCTP_BASE_INFO(ipi_count_ifas), 1); if (new_ifn_af) { SCTP_REGISTER_INTERFACE(ifn_index, new_ifn_af); sctp_ifnp->registered_af = new_ifn_af; } SCTP_IPI_ADDR_WUNLOCK(); if (dynamic_add) { /* * Bump up the refcount so that when the timer completes it * will drop back down. */ struct sctp_laddr *wi; atomic_add_int(&sctp_ifap->refcount, 1); wi = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_laddr), struct sctp_laddr); if (wi == NULL) { /* * Gak, what can we do? We have lost an address * change can you say HOSED? */ SCTPDBG(SCTP_DEBUG_PCB4, "Lost an address change?\n"); /* Opps, must decrement the count */ sctp_del_addr_from_vrf(vrf_id, addr, ifn_index, if_name); return (NULL); } SCTP_INCR_LADDR_COUNT(); memset(wi, 0, sizeof(*wi)); (void)SCTP_GETTIME_TIMEVAL(&wi->start_time); wi->ifa = sctp_ifap; wi->action = SCTP_ADD_IP_ADDRESS; SCTP_WQ_ADDR_LOCK(); LIST_INSERT_HEAD(&SCTP_BASE_INFO(addr_wq), wi, sctp_nxt_addr); sctp_timer_start(SCTP_TIMER_TYPE_ADDR_WQ, (struct sctp_inpcb *)NULL, (struct sctp_tcb *)NULL, (struct sctp_nets *)NULL); SCTP_WQ_ADDR_UNLOCK(); } else { /* it's ready for use */ sctp_ifap->localifa_flags &= ~SCTP_ADDR_DEFER_USE; } return (sctp_ifap); } void sctp_del_addr_from_vrf(uint32_t vrf_id, struct sockaddr *addr, uint32_t ifn_index, const char *if_name) { struct sctp_vrf *vrf; struct sctp_ifa *sctp_ifap = NULL; SCTP_IPI_ADDR_WLOCK(); vrf = sctp_find_vrf(vrf_id); if (vrf == NULL) { SCTPDBG(SCTP_DEBUG_PCB4, "Can't find vrf_id 0x%x\n", vrf_id); goto out_now; } #ifdef SCTP_DEBUG SCTPDBG(SCTP_DEBUG_PCB4, "vrf_id 0x%x: deleting address:", vrf_id); SCTPDBG_ADDR(SCTP_DEBUG_PCB4, addr); #endif sctp_ifap = sctp_find_ifa_by_addr(addr, vrf->vrf_id, SCTP_ADDR_LOCKED); if (sctp_ifap) { /* Validate the delete */ if (sctp_ifap->ifn_p) { int valid = 0; /*- * The name has priority over the ifn_index * if its given. We do this especially for * panda who might recycle indexes fast. */ if (if_name) { if (strncmp(if_name, sctp_ifap->ifn_p->ifn_name, SCTP_IFNAMSIZ) == 0) { /* They match its a correct delete */ valid = 1; } } if (!valid) { /* last ditch check ifn_index */ if (ifn_index == sctp_ifap->ifn_p->ifn_index) { valid = 1; } } if (!valid) { SCTPDBG(SCTP_DEBUG_PCB4, "ifn:%d ifname:%s does not match addresses\n", ifn_index, ((if_name == NULL) ? "NULL" : if_name)); SCTPDBG(SCTP_DEBUG_PCB4, "ifn:%d ifname:%s - ignoring delete\n", sctp_ifap->ifn_p->ifn_index, sctp_ifap->ifn_p->ifn_name); SCTP_IPI_ADDR_WUNLOCK(); return; } } SCTPDBG(SCTP_DEBUG_PCB4, "Deleting ifa %p\n", (void *)sctp_ifap); sctp_ifap->localifa_flags &= SCTP_ADDR_VALID; /* * We don't set the flag. This means that the structure will * hang around in EP's that have bound specific to it until * they close. This gives us TCP like behavior if someone * removes an address (or for that matter adds it right * back). */ /* sctp_ifap->localifa_flags |= SCTP_BEING_DELETED; */ vrf->total_ifa_count--; LIST_REMOVE(sctp_ifap, next_bucket); sctp_remove_ifa_from_ifn(sctp_ifap); } #ifdef SCTP_DEBUG else { SCTPDBG(SCTP_DEBUG_PCB4, "Del Addr-ifn:%d Could not find address:", ifn_index); SCTPDBG_ADDR(SCTP_DEBUG_PCB1, addr); } #endif out_now: SCTP_IPI_ADDR_WUNLOCK(); if (sctp_ifap) { struct sctp_laddr *wi; wi = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_laddr), struct sctp_laddr); if (wi == NULL) { /* * Gak, what can we do? We have lost an address * change can you say HOSED? */ SCTPDBG(SCTP_DEBUG_PCB4, "Lost an address change?\n"); /* Oops, must decrement the count */ sctp_free_ifa(sctp_ifap); return; } SCTP_INCR_LADDR_COUNT(); memset(wi, 0, sizeof(*wi)); (void)SCTP_GETTIME_TIMEVAL(&wi->start_time); wi->ifa = sctp_ifap; wi->action = SCTP_DEL_IP_ADDRESS; SCTP_WQ_ADDR_LOCK(); /* * Should this really be a tailq? As it is we will process * the newest first :-0 */ LIST_INSERT_HEAD(&SCTP_BASE_INFO(addr_wq), wi, sctp_nxt_addr); sctp_timer_start(SCTP_TIMER_TYPE_ADDR_WQ, (struct sctp_inpcb *)NULL, (struct sctp_tcb *)NULL, (struct sctp_nets *)NULL); SCTP_WQ_ADDR_UNLOCK(); } return; } static int sctp_does_stcb_own_this_addr(struct sctp_tcb *stcb, struct sockaddr *to) { int loopback_scope; #if defined(INET) int ipv4_local_scope, ipv4_addr_legal; #endif #if defined(INET6) int local_scope, site_scope, ipv6_addr_legal; #endif struct sctp_vrf *vrf; struct sctp_ifn *sctp_ifn; struct sctp_ifa *sctp_ifa; loopback_scope = stcb->asoc.scope.loopback_scope; #if defined(INET) ipv4_local_scope = stcb->asoc.scope.ipv4_local_scope; ipv4_addr_legal = stcb->asoc.scope.ipv4_addr_legal; #endif #if defined(INET6) local_scope = stcb->asoc.scope.local_scope; site_scope = stcb->asoc.scope.site_scope; ipv6_addr_legal = stcb->asoc.scope.ipv6_addr_legal; #endif SCTP_IPI_ADDR_RLOCK(); vrf = sctp_find_vrf(stcb->asoc.vrf_id); if (vrf == NULL) { /* no vrf, no addresses */ SCTP_IPI_ADDR_RUNLOCK(); return (0); } if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { LIST_FOREACH(sctp_ifn, &vrf->ifnlist, next_ifn) { if ((loopback_scope == 0) && SCTP_IFN_IS_IFT_LOOP(sctp_ifn)) { continue; } LIST_FOREACH(sctp_ifa, &sctp_ifn->ifalist, next_ifa) { if (sctp_is_addr_restricted(stcb, sctp_ifa) && (!sctp_is_addr_pending(stcb, sctp_ifa))) { /* * We allow pending addresses, where * we have sent an asconf-add to be * considered valid. */ continue; } if (sctp_ifa->address.sa.sa_family != to->sa_family) { continue; } switch (sctp_ifa->address.sa.sa_family) { #ifdef INET case AF_INET: if (ipv4_addr_legal) { struct sockaddr_in *sin, *rsin; sin = &sctp_ifa->address.sin; rsin = (struct sockaddr_in *)to; if ((ipv4_local_scope == 0) && IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) { continue; } if (prison_check_ip4(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { continue; } if (sin->sin_addr.s_addr == rsin->sin_addr.s_addr) { SCTP_IPI_ADDR_RUNLOCK(); return (1); } } break; #endif #ifdef INET6 case AF_INET6: if (ipv6_addr_legal) { struct sockaddr_in6 *sin6, *rsin6; sin6 = &sctp_ifa->address.sin6; rsin6 = (struct sockaddr_in6 *)to; if (prison_check_ip6(stcb->sctp_ep->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { continue; } if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) { if (local_scope == 0) continue; if (sin6->sin6_scope_id == 0) { if (sa6_recoverscope(sin6) != 0) continue; } } if ((site_scope == 0) && (IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr))) { continue; } if (SCTP6_ARE_ADDR_EQUAL(sin6, rsin6)) { SCTP_IPI_ADDR_RUNLOCK(); return (1); } } break; #endif default: /* TSNH */ break; } } } } else { struct sctp_laddr *laddr; LIST_FOREACH(laddr, &stcb->sctp_ep->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) { SCTPDBG(SCTP_DEBUG_PCB1, "ifa being deleted\n"); continue; } if (sctp_is_addr_restricted(stcb, laddr->ifa) && (!sctp_is_addr_pending(stcb, laddr->ifa))) { /* * We allow pending addresses, where we have * sent an asconf-add to be considered * valid. */ continue; } if (laddr->ifa->address.sa.sa_family != to->sa_family) { continue; } switch (to->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin, *rsin; sin = &laddr->ifa->address.sin; rsin = (struct sockaddr_in *)to; if (sin->sin_addr.s_addr == rsin->sin_addr.s_addr) { SCTP_IPI_ADDR_RUNLOCK(); return (1); } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6, *rsin6; sin6 = &laddr->ifa->address.sin6; rsin6 = (struct sockaddr_in6 *)to; if (SCTP6_ARE_ADDR_EQUAL(sin6, rsin6)) { SCTP_IPI_ADDR_RUNLOCK(); return (1); } break; } #endif default: /* TSNH */ break; } } } SCTP_IPI_ADDR_RUNLOCK(); return (0); } static struct sctp_tcb * sctp_tcb_special_locate(struct sctp_inpcb **inp_p, struct sockaddr *from, struct sockaddr *to, struct sctp_nets **netp, uint32_t vrf_id) { /**** ASSUMES THE CALLER holds the INP_INFO_RLOCK */ /* * If we support the TCP model, then we must now dig through to see * if we can find our endpoint in the list of tcp ep's. */ uint16_t lport, rport; struct sctppcbhead *ephead; struct sctp_inpcb *inp; struct sctp_laddr *laddr; struct sctp_tcb *stcb; struct sctp_nets *net; if ((to == NULL) || (from == NULL)) { return (NULL); } switch (to->sa_family) { #ifdef INET case AF_INET: if (from->sa_family == AF_INET) { lport = ((struct sockaddr_in *)to)->sin_port; rport = ((struct sockaddr_in *)from)->sin_port; } else { return (NULL); } break; #endif #ifdef INET6 case AF_INET6: if (from->sa_family == AF_INET6) { lport = ((struct sockaddr_in6 *)to)->sin6_port; rport = ((struct sockaddr_in6 *)from)->sin6_port; } else { return (NULL); } break; #endif default: return (NULL); } ephead = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport | rport), SCTP_BASE_INFO(hashtcpmark))]; /* * Ok now for each of the guys in this bucket we must look and see: * - Does the remote port match. - Does there single association's * addresses match this address (to). If so we update p_ep to point * to this ep and return the tcb from it. */ LIST_FOREACH(inp, ephead, sctp_hash) { SCTP_INP_RLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { SCTP_INP_RUNLOCK(inp); continue; } if (lport != inp->sctp_lport) { SCTP_INP_RUNLOCK(inp); continue; } switch (to->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; sin = (struct sockaddr_in *)to; if (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { SCTP_INP_RUNLOCK(inp); continue; } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)to; if (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { SCTP_INP_RUNLOCK(inp); continue; } break; } #endif default: SCTP_INP_RUNLOCK(inp); continue; } if (inp->def_vrf_id != vrf_id) { SCTP_INP_RUNLOCK(inp); continue; } /* check to see if the ep has one of the addresses */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) == 0) { /* We are NOT bound all, so look further */ int match = 0; LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_PCB1, "%s: NULL ifa\n", __func__); continue; } if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) { SCTPDBG(SCTP_DEBUG_PCB1, "ifa being deleted\n"); continue; } if (laddr->ifa->address.sa.sa_family == to->sa_family) { /* see if it matches */ #ifdef INET if (from->sa_family == AF_INET) { struct sockaddr_in *intf_addr, *sin; intf_addr = &laddr->ifa->address.sin; sin = (struct sockaddr_in *)to; if (sin->sin_addr.s_addr == intf_addr->sin_addr.s_addr) { match = 1; break; } } #endif #ifdef INET6 if (from->sa_family == AF_INET6) { struct sockaddr_in6 *intf_addr6; struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *) to; intf_addr6 = &laddr->ifa->address.sin6; if (SCTP6_ARE_ADDR_EQUAL(sin6, intf_addr6)) { match = 1; break; } } #endif } } if (match == 0) { /* This endpoint does not have this address */ SCTP_INP_RUNLOCK(inp); continue; } } /* * Ok if we hit here the ep has the address, does it hold * the tcb? */ /* XXX: Why don't we TAILQ_FOREACH through sctp_asoc_list? */ stcb = LIST_FIRST(&inp->sctp_asoc_list); if (stcb == NULL) { SCTP_INP_RUNLOCK(inp); continue; } SCTP_TCB_LOCK(stcb); if (!sctp_does_stcb_own_this_addr(stcb, to)) { SCTP_TCB_UNLOCK(stcb); SCTP_INP_RUNLOCK(inp); continue; } if (stcb->rport != rport) { /* remote port does not match. */ SCTP_TCB_UNLOCK(stcb); SCTP_INP_RUNLOCK(inp); continue; } if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_TCB_UNLOCK(stcb); SCTP_INP_RUNLOCK(inp); continue; } if (!sctp_does_stcb_own_this_addr(stcb, to)) { SCTP_TCB_UNLOCK(stcb); SCTP_INP_RUNLOCK(inp); continue; } /* Does this TCB have a matching address? */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { if (net->ro._l_addr.sa.sa_family != from->sa_family) { /* not the same family, can't be a match */ continue; } switch (from->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin, *rsin; sin = (struct sockaddr_in *)&net->ro._l_addr; rsin = (struct sockaddr_in *)from; if (sin->sin_addr.s_addr == rsin->sin_addr.s_addr) { /* found it */ if (netp != NULL) { *netp = net; } /* * Update the endpoint * pointer */ *inp_p = inp; SCTP_INP_RUNLOCK(inp); return (stcb); } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6, *rsin6; sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; rsin6 = (struct sockaddr_in6 *)from; if (SCTP6_ARE_ADDR_EQUAL(sin6, rsin6)) { /* found it */ if (netp != NULL) { *netp = net; } /* * Update the endpoint * pointer */ *inp_p = inp; SCTP_INP_RUNLOCK(inp); return (stcb); } break; } #endif default: /* TSNH */ break; } } SCTP_TCB_UNLOCK(stcb); SCTP_INP_RUNLOCK(inp); } return (NULL); } /* * rules for use * * 1) If I return a NULL you must decrement any INP ref cnt. 2) If I find an * stcb, both will be locked (locked_tcb and stcb) but decrement will be done * (if locked == NULL). 3) Decrement happens on return ONLY if locked == * NULL. */ struct sctp_tcb * sctp_findassociation_ep_addr(struct sctp_inpcb **inp_p, struct sockaddr *remote, struct sctp_nets **netp, struct sockaddr *local, struct sctp_tcb *locked_tcb) { struct sctpasochead *head; struct sctp_inpcb *inp; struct sctp_tcb *stcb = NULL; struct sctp_nets *net; uint16_t rport; inp = *inp_p; switch (remote->sa_family) { #ifdef INET case AF_INET: rport = (((struct sockaddr_in *)remote)->sin_port); break; #endif #ifdef INET6 case AF_INET6: rport = (((struct sockaddr_in6 *)remote)->sin6_port); break; #endif default: return (NULL); } if (locked_tcb) { /* * UN-lock so we can do proper locking here this occurs when * called from load_addresses_from_init. */ atomic_add_int(&locked_tcb->asoc.refcnt, 1); SCTP_TCB_UNLOCK(locked_tcb); } SCTP_INP_INFO_RLOCK(); if ((inp->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE) || (inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL)) { /*- * Now either this guy is our listener or it's the * connector. If it is the one that issued the connect, then * it's only chance is to be the first TCB in the list. If * it is the acceptor, then do the special_lookup to hash * and find the real inp. */ if ((inp->sctp_socket) && SCTP_IS_LISTENING(inp)) { /* to is peer addr, from is my addr */ stcb = sctp_tcb_special_locate(inp_p, remote, local, netp, inp->def_vrf_id); if ((stcb != NULL) && (locked_tcb == NULL)) { /* we have a locked tcb, lower refcount */ SCTP_INP_DECR_REF(inp); } if ((locked_tcb != NULL) && (locked_tcb != stcb)) { SCTP_INP_RLOCK(locked_tcb->sctp_ep); SCTP_TCB_LOCK(locked_tcb); atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); SCTP_INP_RUNLOCK(locked_tcb->sctp_ep); } SCTP_INP_INFO_RUNLOCK(); return (stcb); } else { SCTP_INP_WLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { goto null_return; } stcb = LIST_FIRST(&inp->sctp_asoc_list); if (stcb == NULL) { goto null_return; } SCTP_TCB_LOCK(stcb); if (stcb->rport != rport) { /* remote port does not match. */ SCTP_TCB_UNLOCK(stcb); goto null_return; } if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_TCB_UNLOCK(stcb); goto null_return; } if (local && !sctp_does_stcb_own_this_addr(stcb, local)) { SCTP_TCB_UNLOCK(stcb); goto null_return; } /* now look at the list of remote addresses */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { #ifdef INVARIANTS if (net == (TAILQ_NEXT(net, sctp_next))) { panic("Corrupt net list"); } #endif if (net->ro._l_addr.sa.sa_family != remote->sa_family) { /* not the same family */ continue; } switch (remote->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin, *rsin; sin = (struct sockaddr_in *) &net->ro._l_addr; rsin = (struct sockaddr_in *)remote; if (sin->sin_addr.s_addr == rsin->sin_addr.s_addr) { /* found it */ if (netp != NULL) { *netp = net; } if (locked_tcb == NULL) { SCTP_INP_DECR_REF(inp); } else if (locked_tcb != stcb) { SCTP_TCB_LOCK(locked_tcb); } if (locked_tcb) { atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); } SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_RUNLOCK(); return (stcb); } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6, *rsin6; sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; rsin6 = (struct sockaddr_in6 *)remote; if (SCTP6_ARE_ADDR_EQUAL(sin6, rsin6)) { /* found it */ if (netp != NULL) { *netp = net; } if (locked_tcb == NULL) { SCTP_INP_DECR_REF(inp); } else if (locked_tcb != stcb) { SCTP_TCB_LOCK(locked_tcb); } if (locked_tcb) { atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); } SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_RUNLOCK(); return (stcb); } break; } #endif default: /* TSNH */ break; } } SCTP_TCB_UNLOCK(stcb); } } else { SCTP_INP_WLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { goto null_return; } head = &inp->sctp_tcbhash[SCTP_PCBHASH_ALLADDR(rport, inp->sctp_hashmark)]; LIST_FOREACH(stcb, head, sctp_tcbhash) { if (stcb->rport != rport) { /* remote port does not match */ continue; } SCTP_TCB_LOCK(stcb); if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_TCB_UNLOCK(stcb); continue; } if (local && !sctp_does_stcb_own_this_addr(stcb, local)) { SCTP_TCB_UNLOCK(stcb); continue; } /* now look at the list of remote addresses */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { #ifdef INVARIANTS if (net == (TAILQ_NEXT(net, sctp_next))) { panic("Corrupt net list"); } #endif if (net->ro._l_addr.sa.sa_family != remote->sa_family) { /* not the same family */ continue; } switch (remote->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin, *rsin; sin = (struct sockaddr_in *) &net->ro._l_addr; rsin = (struct sockaddr_in *)remote; if (sin->sin_addr.s_addr == rsin->sin_addr.s_addr) { /* found it */ if (netp != NULL) { *netp = net; } if (locked_tcb == NULL) { SCTP_INP_DECR_REF(inp); } else if (locked_tcb != stcb) { SCTP_TCB_LOCK(locked_tcb); } if (locked_tcb) { atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); } SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_RUNLOCK(); return (stcb); } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6, *rsin6; sin6 = (struct sockaddr_in6 *) &net->ro._l_addr; rsin6 = (struct sockaddr_in6 *)remote; if (SCTP6_ARE_ADDR_EQUAL(sin6, rsin6)) { /* found it */ if (netp != NULL) { *netp = net; } if (locked_tcb == NULL) { SCTP_INP_DECR_REF(inp); } else if (locked_tcb != stcb) { SCTP_TCB_LOCK(locked_tcb); } if (locked_tcb) { atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); } SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_RUNLOCK(); return (stcb); } break; } #endif default: /* TSNH */ break; } } SCTP_TCB_UNLOCK(stcb); } } null_return: /* clean up for returning null */ if (locked_tcb) { SCTP_TCB_LOCK(locked_tcb); atomic_subtract_int(&locked_tcb->asoc.refcnt, 1); } SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_RUNLOCK(); /* not found */ return (NULL); } /* * Find an association for a specific endpoint using the association id given * out in the COMM_UP notification */ struct sctp_tcb * sctp_findasoc_ep_asocid_locked(struct sctp_inpcb *inp, sctp_assoc_t asoc_id, int want_lock) { /* * Use my the assoc_id to find a endpoint */ struct sctpasochead *head; struct sctp_tcb *stcb; uint32_t id; if (inp == NULL) { SCTP_PRINTF("TSNH ep_associd\n"); return (NULL); } if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { SCTP_PRINTF("TSNH ep_associd0\n"); return (NULL); } id = (uint32_t)asoc_id; head = &inp->sctp_asocidhash[SCTP_PCBHASH_ASOC(id, inp->hashasocidmark)]; if (head == NULL) { /* invalid id TSNH */ SCTP_PRINTF("TSNH ep_associd1\n"); return (NULL); } LIST_FOREACH(stcb, head, sctp_tcbasocidhash) { if (stcb->asoc.assoc_id == id) { if (inp != stcb->sctp_ep) { /* * some other guy has the same id active (id * collision ??). */ SCTP_PRINTF("TSNH ep_associd2\n"); continue; } if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { continue; } if (want_lock) { SCTP_TCB_LOCK(stcb); } return (stcb); } } return (NULL); } struct sctp_tcb * sctp_findassociation_ep_asocid(struct sctp_inpcb *inp, sctp_assoc_t asoc_id, int want_lock) { struct sctp_tcb *stcb; SCTP_INP_RLOCK(inp); stcb = sctp_findasoc_ep_asocid_locked(inp, asoc_id, want_lock); SCTP_INP_RUNLOCK(inp); return (stcb); } /* * Endpoint probe expects that the INP_INFO is locked. */ static struct sctp_inpcb * sctp_endpoint_probe(struct sockaddr *nam, struct sctppcbhead *head, uint16_t lport, uint32_t vrf_id) { struct sctp_inpcb *inp; struct sctp_laddr *laddr; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; struct sockaddr_in6 *intf_addr6; #endif int fnd; #ifdef INET sin = NULL; #endif #ifdef INET6 sin6 = NULL; #endif switch (nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nam; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nam; break; #endif default: /* unsupported family */ return (NULL); } if (head == NULL) return (NULL); LIST_FOREACH(inp, head, sctp_hash) { SCTP_INP_RLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { SCTP_INP_RUNLOCK(inp); continue; } if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) && (inp->sctp_lport == lport)) { /* got it */ switch (nam->sa_family) { #ifdef INET case AF_INET: if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) { /* * IPv4 on a IPv6 socket with ONLY * IPv6 set */ SCTP_INP_RUNLOCK(inp); continue; } if (prison_check_ip4(inp->ip_inp.inp.inp_cred, &sin->sin_addr) != 0) { SCTP_INP_RUNLOCK(inp); continue; } break; #endif #ifdef INET6 case AF_INET6: /* * A V6 address and the endpoint is NOT * bound V6 */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) == 0) { SCTP_INP_RUNLOCK(inp); continue; } if (prison_check_ip6(inp->ip_inp.inp.inp_cred, &sin6->sin6_addr) != 0) { SCTP_INP_RUNLOCK(inp); continue; } break; #endif default: break; } /* does a VRF id match? */ fnd = 0; if (inp->def_vrf_id == vrf_id) fnd = 1; SCTP_INP_RUNLOCK(inp); if (!fnd) continue; return (inp); } SCTP_INP_RUNLOCK(inp); } switch (nam->sa_family) { #ifdef INET case AF_INET: if (sin->sin_addr.s_addr == INADDR_ANY) { /* Can't hunt for one that has no address specified */ return (NULL); } break; #endif #ifdef INET6 case AF_INET6: if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* Can't hunt for one that has no address specified */ return (NULL); } break; #endif default: break; } /* * ok, not bound to all so see if we can find a EP bound to this * address. */ LIST_FOREACH(inp, head, sctp_hash) { SCTP_INP_RLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { SCTP_INP_RUNLOCK(inp); continue; } if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL)) { SCTP_INP_RUNLOCK(inp); continue; } /* * Ok this could be a likely candidate, look at all of its * addresses */ if (inp->sctp_lport != lport) { SCTP_INP_RUNLOCK(inp); continue; } /* does a VRF id match? */ fnd = 0; if (inp->def_vrf_id == vrf_id) fnd = 1; if (!fnd) { SCTP_INP_RUNLOCK(inp); continue; } LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_PCB1, "%s: NULL ifa\n", __func__); continue; } SCTPDBG(SCTP_DEBUG_PCB1, "Ok laddr->ifa:%p is possible, ", (void *)laddr->ifa); if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) { SCTPDBG(SCTP_DEBUG_PCB1, "Huh IFA being deleted\n"); continue; } if (laddr->ifa->address.sa.sa_family == nam->sa_family) { /* possible, see if it matches */ switch (nam->sa_family) { #ifdef INET case AF_INET: if (sin->sin_addr.s_addr == laddr->ifa->address.sin.sin_addr.s_addr) { SCTP_INP_RUNLOCK(inp); return (inp); } break; #endif #ifdef INET6 case AF_INET6: intf_addr6 = &laddr->ifa->address.sin6; if (SCTP6_ARE_ADDR_EQUAL(sin6, intf_addr6)) { SCTP_INP_RUNLOCK(inp); return (inp); } break; #endif } } } SCTP_INP_RUNLOCK(inp); } return (NULL); } static struct sctp_inpcb * sctp_isport_inuse(struct sctp_inpcb *inp, uint16_t lport, uint32_t vrf_id) { struct sctppcbhead *head; struct sctp_inpcb *t_inp; int fnd; head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashmark))]; LIST_FOREACH(t_inp, head, sctp_hash) { if (t_inp->sctp_lport != lport) { continue; } /* is it in the VRF in question */ fnd = 0; if (t_inp->def_vrf_id == vrf_id) fnd = 1; if (!fnd) continue; /* This one is in use. */ /* check the v6/v4 binding issue */ if ((t_inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(t_inp)) { if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) { /* collision in V6 space */ return (t_inp); } else { /* inp is BOUND_V4 no conflict */ continue; } } else if (t_inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) { /* t_inp is bound v4 and v6, conflict always */ return (t_inp); } else { /* t_inp is bound only V4 */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_BOUND_V6) && SCTP_IPV6_V6ONLY(inp)) { /* no conflict */ continue; } /* else fall through to conflict */ } return (t_inp); } return (NULL); } int sctp_swap_inpcb_for_listen(struct sctp_inpcb *inp) { /* For 1-2-1 with port reuse */ struct sctppcbhead *head; struct sctp_inpcb *tinp, *ninp; if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_PORTREUSE)) { /* only works with port reuse on */ return (-1); } if ((inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL) == 0) { return (0); } SCTP_INP_RUNLOCK(inp); SCTP_INP_INFO_WLOCK(); head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(inp->sctp_lport, SCTP_BASE_INFO(hashmark))]; /* Kick out all non-listeners to the TCP hash */ LIST_FOREACH_SAFE(tinp, head, sctp_hash, ninp) { if (tinp->sctp_lport != inp->sctp_lport) { continue; } if (tinp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { continue; } if (tinp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) { continue; } if (SCTP_IS_LISTENING(tinp)) { continue; } SCTP_INP_WLOCK(tinp); LIST_REMOVE(tinp, sctp_hash); head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR(tinp->sctp_lport, SCTP_BASE_INFO(hashtcpmark))]; tinp->sctp_flags |= SCTP_PCB_FLAGS_IN_TCPPOOL; LIST_INSERT_HEAD(head, tinp, sctp_hash); SCTP_INP_WUNLOCK(tinp); } SCTP_INP_WLOCK(inp); /* Pull from where he was */ LIST_REMOVE(inp, sctp_hash); inp->sctp_flags &= ~SCTP_PCB_FLAGS_IN_TCPPOOL; head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(inp->sctp_lport, SCTP_BASE_INFO(hashmark))]; LIST_INSERT_HEAD(head, inp, sctp_hash); SCTP_INP_WUNLOCK(inp); SCTP_INP_RLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return (0); } struct sctp_inpcb * sctp_pcb_findep(struct sockaddr *nam, int find_tcp_pool, int have_lock, uint32_t vrf_id) { /* * First we check the hash table to see if someone has this port * bound with just the port. */ struct sctp_inpcb *inp; struct sctppcbhead *head; int lport; unsigned int i; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif switch (nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nam; lport = sin->sin_port; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nam; lport = sin6->sin6_port; break; #endif default: return (NULL); } /* * I could cheat here and just cast to one of the types but we will * do it right. It also provides the check against an Unsupported * type too. */ /* Find the head of the ALLADDR chain */ if (have_lock == 0) { SCTP_INP_INFO_RLOCK(); } head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashmark))]; inp = sctp_endpoint_probe(nam, head, lport, vrf_id); /* * If the TCP model exists it could be that the main listening * endpoint is gone but there still exists a connected socket for * this guy. If so we can return the first one that we find. This * may NOT be the correct one so the caller should be wary on the * returned INP. Currently the only caller that sets find_tcp_pool * is in bindx where we are verifying that a user CAN bind the * address. He either has bound it already, or someone else has, or * its open to bind, so this is good enough. */ if (inp == NULL && find_tcp_pool) { for (i = 0; i < SCTP_BASE_INFO(hashtcpmark) + 1; i++) { head = &SCTP_BASE_INFO(sctp_tcpephash)[i]; inp = sctp_endpoint_probe(nam, head, lport, vrf_id); if (inp) { break; } } } if (inp) { SCTP_INP_INCR_REF(inp); } if (have_lock == 0) { SCTP_INP_INFO_RUNLOCK(); } return (inp); } /* * Find an association for an endpoint with the pointer to whom you want to * send to and the endpoint pointer. The address can be IPv4 or IPv6. We may * need to change the *to to some other struct like a mbuf... */ struct sctp_tcb * sctp_findassociation_addr_sa(struct sockaddr *from, struct sockaddr *to, struct sctp_inpcb **inp_p, struct sctp_nets **netp, int find_tcp_pool, uint32_t vrf_id) { struct sctp_inpcb *inp = NULL; struct sctp_tcb *stcb; SCTP_INP_INFO_RLOCK(); if (find_tcp_pool) { if (inp_p != NULL) { stcb = sctp_tcb_special_locate(inp_p, from, to, netp, vrf_id); } else { stcb = sctp_tcb_special_locate(&inp, from, to, netp, vrf_id); } if (stcb != NULL) { SCTP_INP_INFO_RUNLOCK(); return (stcb); } } inp = sctp_pcb_findep(to, 0, 1, vrf_id); if (inp_p != NULL) { *inp_p = inp; } SCTP_INP_INFO_RUNLOCK(); if (inp == NULL) { return (NULL); } /* * ok, we have an endpoint, now lets find the assoc for it (if any) * we now place the source address or from in the to of the find * endpoint call. Since in reality this chain is used from the * inbound packet side. */ if (inp_p != NULL) { stcb = sctp_findassociation_ep_addr(inp_p, from, netp, to, NULL); } else { stcb = sctp_findassociation_ep_addr(&inp, from, netp, to, NULL); } return (stcb); } /* * This routine will grub through the mbuf that is a INIT or INIT-ACK and * find all addresses that the sender has specified in any address list. Each * address will be used to lookup the TCB and see if one exits. */ static struct sctp_tcb * sctp_findassociation_special_addr(struct mbuf *m, int offset, struct sctphdr *sh, struct sctp_inpcb **inp_p, struct sctp_nets **netp, struct sockaddr *dst) { struct sctp_paramhdr *phdr, param_buf; #if defined(INET) || defined(INET6) struct sctp_tcb *stcb; uint16_t ptype; #endif uint16_t plen; #ifdef INET struct sockaddr_in sin4; #endif #ifdef INET6 struct sockaddr_in6 sin6; #endif #ifdef INET memset(&sin4, 0, sizeof(sin4)); sin4.sin_len = sizeof(sin4); sin4.sin_family = AF_INET; sin4.sin_port = sh->src_port; #endif #ifdef INET6 memset(&sin6, 0, sizeof(sin6)); sin6.sin6_len = sizeof(sin6); sin6.sin6_family = AF_INET6; sin6.sin6_port = sh->src_port; #endif offset += sizeof(struct sctp_init_chunk); phdr = sctp_get_next_param(m, offset, ¶m_buf, sizeof(param_buf)); while (phdr != NULL) { /* now we must see if we want the parameter */ #if defined(INET) || defined(INET6) ptype = ntohs(phdr->param_type); #endif plen = ntohs(phdr->param_length); if (plen == 0) { break; } #ifdef INET if (ptype == SCTP_IPV4_ADDRESS && plen == sizeof(struct sctp_ipv4addr_param)) { /* Get the rest of the address */ struct sctp_ipv4addr_param ip4_param, *p4; phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&ip4_param, sizeof(ip4_param)); if (phdr == NULL) { return (NULL); } p4 = (struct sctp_ipv4addr_param *)phdr; memcpy(&sin4.sin_addr, &p4->addr, sizeof(p4->addr)); /* look it up */ stcb = sctp_findassociation_ep_addr(inp_p, (struct sockaddr *)&sin4, netp, dst, NULL); if (stcb != NULL) { return (stcb); } } #endif #ifdef INET6 if (ptype == SCTP_IPV6_ADDRESS && plen == sizeof(struct sctp_ipv6addr_param)) { /* Get the rest of the address */ struct sctp_ipv6addr_param ip6_param, *p6; phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&ip6_param, sizeof(ip6_param)); if (phdr == NULL) { return (NULL); } p6 = (struct sctp_ipv6addr_param *)phdr; memcpy(&sin6.sin6_addr, &p6->addr, sizeof(p6->addr)); /* look it up */ stcb = sctp_findassociation_ep_addr(inp_p, (struct sockaddr *)&sin6, netp, dst, NULL); if (stcb != NULL) { return (stcb); } } #endif offset += SCTP_SIZE32(plen); phdr = sctp_get_next_param(m, offset, ¶m_buf, sizeof(param_buf)); } return (NULL); } static struct sctp_tcb * sctp_findassoc_by_vtag(struct sockaddr *from, struct sockaddr *to, uint32_t vtag, struct sctp_inpcb **inp_p, struct sctp_nets **netp, uint16_t rport, uint16_t lport, int skip_src_check, uint32_t vrf_id, uint32_t remote_tag) { /* * Use my vtag to hash. If we find it we then verify the source addr * is in the assoc. If all goes well we save a bit on rec of a * packet. */ struct sctpasochead *head; struct sctp_nets *net; struct sctp_tcb *stcb; SCTP_INP_INFO_RLOCK(); head = &SCTP_BASE_INFO(sctp_asochash)[SCTP_PCBHASH_ASOC(vtag, SCTP_BASE_INFO(hashasocmark))]; LIST_FOREACH(stcb, head, sctp_asocs) { SCTP_INP_RLOCK(stcb->sctp_ep); if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { SCTP_INP_RUNLOCK(stcb->sctp_ep); continue; } if (stcb->sctp_ep->def_vrf_id != vrf_id) { SCTP_INP_RUNLOCK(stcb->sctp_ep); continue; } SCTP_TCB_LOCK(stcb); SCTP_INP_RUNLOCK(stcb->sctp_ep); if (stcb->asoc.my_vtag == vtag) { /* candidate */ if (stcb->rport != rport) { SCTP_TCB_UNLOCK(stcb); continue; } if (stcb->sctp_ep->sctp_lport != lport) { SCTP_TCB_UNLOCK(stcb); continue; } if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { SCTP_TCB_UNLOCK(stcb); continue; } /* RRS:Need toaddr check here */ if (sctp_does_stcb_own_this_addr(stcb, to) == 0) { /* Endpoint does not own this address */ SCTP_TCB_UNLOCK(stcb); continue; } if (remote_tag) { /* * If we have both vtags that's all we match * on */ if (stcb->asoc.peer_vtag == remote_tag) { /* * If both tags match we consider it * conclusive and check NO * source/destination addresses */ goto conclusive; } } if (skip_src_check) { conclusive: if (from) { *netp = sctp_findnet(stcb, from); } else { *netp = NULL; /* unknown */ } if (inp_p) *inp_p = stcb->sctp_ep; SCTP_INP_INFO_RUNLOCK(); return (stcb); } net = sctp_findnet(stcb, from); if (net) { /* yep its him. */ *netp = net; SCTP_STAT_INCR(sctps_vtagexpress); *inp_p = stcb->sctp_ep; SCTP_INP_INFO_RUNLOCK(); return (stcb); } else { /* * not him, this should only happen in rare * cases so I peg it. */ SCTP_STAT_INCR(sctps_vtagbogus); } } SCTP_TCB_UNLOCK(stcb); } SCTP_INP_INFO_RUNLOCK(); return (NULL); } /* * Find an association with the pointer to the inbound IP packet. This can be * a IPv4 or IPv6 packet. */ struct sctp_tcb * sctp_findassociation_addr(struct mbuf *m, int offset, struct sockaddr *src, struct sockaddr *dst, struct sctphdr *sh, struct sctp_chunkhdr *ch, struct sctp_inpcb **inp_p, struct sctp_nets **netp, uint32_t vrf_id) { struct sctp_tcb *stcb; struct sctp_inpcb *inp; if (sh->v_tag) { /* we only go down this path if vtag is non-zero */ stcb = sctp_findassoc_by_vtag(src, dst, ntohl(sh->v_tag), inp_p, netp, sh->src_port, sh->dest_port, 0, vrf_id, 0); if (stcb) { return (stcb); } } if (inp_p) { stcb = sctp_findassociation_addr_sa(src, dst, inp_p, netp, 1, vrf_id); inp = *inp_p; } else { stcb = sctp_findassociation_addr_sa(src, dst, &inp, netp, 1, vrf_id); } SCTPDBG(SCTP_DEBUG_PCB1, "stcb:%p inp:%p\n", (void *)stcb, (void *)inp); if (stcb == NULL && inp) { /* Found a EP but not this address */ if ((ch->chunk_type == SCTP_INITIATION) || (ch->chunk_type == SCTP_INITIATION_ACK)) { /*- * special hook, we do NOT return linp or an * association that is linked to an existing * association that is under the TCP pool (i.e. no * listener exists). The endpoint finding routine * will always find a listener before examining the * TCP pool. */ if (inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL) { if (inp_p) { *inp_p = NULL; } return (NULL); } stcb = sctp_findassociation_special_addr(m, offset, sh, &inp, netp, dst); if (inp_p != NULL) { *inp_p = inp; } } } SCTPDBG(SCTP_DEBUG_PCB1, "stcb is %p\n", (void *)stcb); return (stcb); } /* * lookup an association by an ASCONF lookup address. * if the lookup address is 0.0.0.0 or ::0, use the vtag to do the lookup */ struct sctp_tcb * sctp_findassociation_ep_asconf(struct mbuf *m, int offset, struct sockaddr *dst, struct sctphdr *sh, struct sctp_inpcb **inp_p, struct sctp_nets **netp, uint32_t vrf_id) { struct sctp_tcb *stcb; union sctp_sockstore remote_store; struct sctp_paramhdr param_buf, *phdr; int ptype; int zero_address = 0; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif memset(&remote_store, 0, sizeof(remote_store)); phdr = sctp_get_next_param(m, offset + sizeof(struct sctp_asconf_chunk), ¶m_buf, sizeof(struct sctp_paramhdr)); if (phdr == NULL) { SCTPDBG(SCTP_DEBUG_INPUT3, "%s: failed to get asconf lookup addr\n", __func__); return NULL; } ptype = (int)((uint32_t)ntohs(phdr->param_type)); /* get the correlation address */ switch (ptype) { #ifdef INET6 case SCTP_IPV6_ADDRESS: { /* ipv6 address param */ struct sctp_ipv6addr_param *p6, p6_buf; if (ntohs(phdr->param_length) != sizeof(struct sctp_ipv6addr_param)) { return NULL; } p6 = (struct sctp_ipv6addr_param *)sctp_get_next_param(m, offset + sizeof(struct sctp_asconf_chunk), &p6_buf.ph, sizeof(p6_buf)); if (p6 == NULL) { SCTPDBG(SCTP_DEBUG_INPUT3, "%s: failed to get asconf v6 lookup addr\n", __func__); return (NULL); } sin6 = &remote_store.sin6; sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(*sin6); sin6->sin6_port = sh->src_port; memcpy(&sin6->sin6_addr, &p6->addr, sizeof(struct in6_addr)); if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) zero_address = 1; break; } #endif #ifdef INET case SCTP_IPV4_ADDRESS: { /* ipv4 address param */ struct sctp_ipv4addr_param *p4, p4_buf; if (ntohs(phdr->param_length) != sizeof(struct sctp_ipv4addr_param)) { return NULL; } p4 = (struct sctp_ipv4addr_param *)sctp_get_next_param(m, offset + sizeof(struct sctp_asconf_chunk), &p4_buf.ph, sizeof(p4_buf)); if (p4 == NULL) { SCTPDBG(SCTP_DEBUG_INPUT3, "%s: failed to get asconf v4 lookup addr\n", __func__); return (NULL); } sin = &remote_store.sin; sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_port = sh->src_port; memcpy(&sin->sin_addr, &p4->addr, sizeof(struct in_addr)); if (sin->sin_addr.s_addr == INADDR_ANY) zero_address = 1; break; } #endif default: /* invalid address param type */ return NULL; } if (zero_address) { stcb = sctp_findassoc_by_vtag(NULL, dst, ntohl(sh->v_tag), inp_p, netp, sh->src_port, sh->dest_port, 1, vrf_id, 0); if (stcb != NULL) { SCTP_INP_DECR_REF(*inp_p); } } else { stcb = sctp_findassociation_ep_addr(inp_p, &remote_store.sa, netp, dst, NULL); } return (stcb); } /* * allocate a sctp_inpcb and setup a temporary binding to a port/all * addresses. This way if we don't get a bind we by default pick a ephemeral * port with all addresses bound. */ int sctp_inpcb_alloc(struct socket *so, uint32_t vrf_id) { /* * we get called when a new endpoint starts up. We need to allocate * the sctp_inpcb structure from the zone and init it. Mark it as * unbound and find a port that we can use as an ephemeral with * INADDR_ANY. If the user binds later no problem we can then add in * the specific addresses. And setup the default parameters for the * EP. */ int i, error; struct sctp_inpcb *inp; struct sctp_pcb *m; struct timeval time; sctp_sharedkey_t *null_key; error = 0; SCTP_INP_INFO_WLOCK(); inp = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_ep), struct sctp_inpcb); if (inp == NULL) { SCTP_PRINTF("Out of SCTP-INPCB structures - no resources\n"); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOBUFS); return (ENOBUFS); } /* zap it */ memset(inp, 0, sizeof(*inp)); /* bump generations */ /* setup socket pointers */ inp->sctp_socket = so; inp->ip_inp.inp.inp_socket = so; inp->ip_inp.inp.inp_cred = crhold(so->so_cred); #ifdef INET6 if (INP_SOCKAF(so) == AF_INET6) { if (MODULE_GLOBAL(ip6_auto_flowlabel)) { inp->ip_inp.inp.inp_flags |= IN6P_AUTOFLOWLABEL; } if (MODULE_GLOBAL(ip6_v6only)) { inp->ip_inp.inp.inp_flags |= IN6P_IPV6_V6ONLY; } } #endif inp->sctp_associd_counter = 1; inp->partial_delivery_point = SCTP_SB_LIMIT_RCV(so) >> SCTP_PARTIAL_DELIVERY_SHIFT; inp->sctp_frag_point = SCTP_DEFAULT_MAXSEGMENT; inp->max_cwnd = 0; inp->sctp_cmt_on_off = SCTP_BASE_SYSCTL(sctp_cmt_on_off); inp->ecn_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_ecn_enable); inp->prsctp_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_pr_enable); inp->auth_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_auth_enable); inp->asconf_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_asconf_enable); inp->reconfig_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_reconfig_enable); inp->nrsack_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_nrsack_enable); inp->pktdrop_supported = (uint8_t)SCTP_BASE_SYSCTL(sctp_pktdrop_enable); inp->idata_supported = 0; inp->fibnum = so->so_fibnum; /* init the small hash table we use to track asocid <-> tcb */ inp->sctp_asocidhash = SCTP_HASH_INIT(SCTP_STACK_VTAG_HASH_SIZE, &inp->hashasocidmark); if (inp->sctp_asocidhash == NULL) { crfree(inp->ip_inp.inp.inp_cred); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_ep), inp); SCTP_INP_INFO_WUNLOCK(); return (ENOBUFS); } SCTP_INCR_EP_COUNT(); inp->ip_inp.inp.inp_ip_ttl = MODULE_GLOBAL(ip_defttl); SCTP_INP_INFO_WUNLOCK(); so->so_pcb = (caddr_t)inp; if (SCTP_SO_TYPE(so) == SOCK_SEQPACKET) { /* UDP style socket */ inp->sctp_flags = (SCTP_PCB_FLAGS_UDPTYPE | SCTP_PCB_FLAGS_UNBOUND); /* Be sure it is NON-BLOCKING IO for UDP */ /* SCTP_SET_SO_NBIO(so); */ } else if (SCTP_SO_TYPE(so) == SOCK_STREAM) { /* TCP style socket */ inp->sctp_flags = (SCTP_PCB_FLAGS_TCPTYPE | SCTP_PCB_FLAGS_UNBOUND); /* Be sure we have blocking IO by default */ SOCK_LOCK(so); SCTP_CLEAR_SO_NBIO(so); SOCK_UNLOCK(so); } else { /* * unsupported socket type (RAW, etc)- in case we missed it * in protosw */ SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EOPNOTSUPP); so->so_pcb = NULL; crfree(inp->ip_inp.inp.inp_cred); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_ep), inp); return (EOPNOTSUPP); } if (SCTP_BASE_SYSCTL(sctp_default_frag_interleave) == SCTP_FRAG_LEVEL_1) { sctp_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE); sctp_feature_off(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS); } else if (SCTP_BASE_SYSCTL(sctp_default_frag_interleave) == SCTP_FRAG_LEVEL_2) { sctp_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE); sctp_feature_on(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS); } else if (SCTP_BASE_SYSCTL(sctp_default_frag_interleave) == SCTP_FRAG_LEVEL_0) { sctp_feature_off(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE); sctp_feature_off(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS); } inp->sctp_tcbhash = SCTP_HASH_INIT(SCTP_BASE_SYSCTL(sctp_pcbtblsize), &inp->sctp_hashmark); if (inp->sctp_tcbhash == NULL) { SCTP_PRINTF("Out of SCTP-INPCB->hashinit - no resources\n"); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOBUFS); so->so_pcb = NULL; crfree(inp->ip_inp.inp.inp_cred); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_ep), inp); return (ENOBUFS); } inp->def_vrf_id = vrf_id; SCTP_INP_INFO_WLOCK(); SCTP_INP_LOCK_INIT(inp); INP_LOCK_INIT(&inp->ip_inp.inp, "inp", "sctpinp"); SCTP_INP_READ_INIT(inp); SCTP_ASOC_CREATE_LOCK_INIT(inp); /* lock the new ep */ SCTP_INP_WLOCK(inp); /* add it to the info area */ LIST_INSERT_HEAD(&SCTP_BASE_INFO(listhead), inp, sctp_list); SCTP_INP_INFO_WUNLOCK(); TAILQ_INIT(&inp->read_queue); LIST_INIT(&inp->sctp_addr_list); LIST_INIT(&inp->sctp_asoc_list); #ifdef SCTP_TRACK_FREED_ASOCS /* TEMP CODE */ LIST_INIT(&inp->sctp_asoc_free_list); #endif /* Init the timer structure for signature change */ SCTP_OS_TIMER_INIT(&inp->sctp_ep.signature_change.timer); inp->sctp_ep.signature_change.type = SCTP_TIMER_TYPE_NEWCOOKIE; /* now init the actual endpoint default data */ m = &inp->sctp_ep; /* setup the base timeout information */ m->sctp_timeoutticks[SCTP_TIMER_SEND] = sctp_secs_to_ticks(SCTP_SEND_SEC); /* needed ? */ m->sctp_timeoutticks[SCTP_TIMER_INIT] = sctp_secs_to_ticks(SCTP_INIT_SEC); /* needed ? */ m->sctp_timeoutticks[SCTP_TIMER_RECV] = sctp_msecs_to_ticks(SCTP_BASE_SYSCTL(sctp_delayed_sack_time_default)); m->sctp_timeoutticks[SCTP_TIMER_HEARTBEAT] = sctp_msecs_to_ticks(SCTP_BASE_SYSCTL(sctp_heartbeat_interval_default)); m->sctp_timeoutticks[SCTP_TIMER_PMTU] = sctp_secs_to_ticks(SCTP_BASE_SYSCTL(sctp_pmtu_raise_time_default)); m->sctp_timeoutticks[SCTP_TIMER_MAXSHUTDOWN] = sctp_secs_to_ticks(SCTP_BASE_SYSCTL(sctp_shutdown_guard_time_default)); m->sctp_timeoutticks[SCTP_TIMER_SIGNATURE] = sctp_secs_to_ticks(SCTP_BASE_SYSCTL(sctp_secret_lifetime_default)); /* all max/min max are in ms */ m->sctp_maxrto = SCTP_BASE_SYSCTL(sctp_rto_max_default); m->sctp_minrto = SCTP_BASE_SYSCTL(sctp_rto_min_default); m->initial_rto = SCTP_BASE_SYSCTL(sctp_rto_initial_default); m->initial_init_rto_max = SCTP_BASE_SYSCTL(sctp_init_rto_max_default); m->sctp_sack_freq = SCTP_BASE_SYSCTL(sctp_sack_freq_default); m->max_init_times = SCTP_BASE_SYSCTL(sctp_init_rtx_max_default); m->max_send_times = SCTP_BASE_SYSCTL(sctp_assoc_rtx_max_default); m->def_net_failure = SCTP_BASE_SYSCTL(sctp_path_rtx_max_default); m->def_net_pf_threshold = SCTP_BASE_SYSCTL(sctp_path_pf_threshold); m->sctp_sws_sender = SCTP_SWS_SENDER_DEF; m->sctp_sws_receiver = SCTP_SWS_RECEIVER_DEF; m->max_burst = SCTP_BASE_SYSCTL(sctp_max_burst_default); m->fr_max_burst = SCTP_BASE_SYSCTL(sctp_fr_max_burst_default); m->sctp_default_cc_module = SCTP_BASE_SYSCTL(sctp_default_cc_module); m->sctp_default_ss_module = SCTP_BASE_SYSCTL(sctp_default_ss_module); m->max_open_streams_intome = SCTP_BASE_SYSCTL(sctp_nr_incoming_streams_default); /* number of streams to pre-open on a association */ m->pre_open_stream_count = SCTP_BASE_SYSCTL(sctp_nr_outgoing_streams_default); m->default_mtu = 0; /* Add adaptation cookie */ m->adaptation_layer_indicator = 0; m->adaptation_layer_indicator_provided = 0; /* seed random number generator */ m->random_counter = 1; m->store_at = SCTP_SIGNATURE_SIZE; SCTP_READ_RANDOM(m->random_numbers, sizeof(m->random_numbers)); sctp_fill_random_store(m); /* Minimum cookie size */ m->size_of_a_cookie = (sizeof(struct sctp_init_msg) * 2) + sizeof(struct sctp_state_cookie); m->size_of_a_cookie += SCTP_SIGNATURE_SIZE; /* Setup the initial secret */ (void)SCTP_GETTIME_TIMEVAL(&time); m->time_of_secret_change = time.tv_sec; for (i = 0; i < SCTP_NUMBER_OF_SECRETS; i++) { m->secret_key[0][i] = sctp_select_initial_TSN(m); } sctp_timer_start(SCTP_TIMER_TYPE_NEWCOOKIE, inp, NULL, NULL); /* How long is a cookie good for ? */ m->def_cookie_life = sctp_msecs_to_ticks(SCTP_BASE_SYSCTL(sctp_valid_cookie_life_default)); /* * Initialize authentication parameters */ m->local_hmacs = sctp_default_supported_hmaclist(); m->local_auth_chunks = sctp_alloc_chunklist(); if (inp->asconf_supported) { sctp_auth_add_chunk(SCTP_ASCONF, m->local_auth_chunks); sctp_auth_add_chunk(SCTP_ASCONF_ACK, m->local_auth_chunks); } m->default_dscp = 0; #ifdef INET6 m->default_flowlabel = 0; #endif m->port = 0; /* encapsulation disabled by default */ LIST_INIT(&m->shared_keys); /* add default NULL key as key id 0 */ null_key = sctp_alloc_sharedkey(); sctp_insert_sharedkey(&m->shared_keys, null_key); SCTP_INP_WUNLOCK(inp); #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 12); #endif return (error); } void sctp_move_pcb_and_assoc(struct sctp_inpcb *old_inp, struct sctp_inpcb *new_inp, struct sctp_tcb *stcb) { struct sctp_nets *net; uint16_t lport, rport; struct sctppcbhead *head; struct sctp_laddr *laddr, *oladdr; atomic_add_int(&stcb->asoc.refcnt, 1); SCTP_TCB_UNLOCK(stcb); SCTP_INP_INFO_WLOCK(); SCTP_INP_WLOCK(old_inp); SCTP_INP_WLOCK(new_inp); SCTP_TCB_LOCK(stcb); atomic_subtract_int(&stcb->asoc.refcnt, 1); new_inp->sctp_ep.time_of_secret_change = old_inp->sctp_ep.time_of_secret_change; memcpy(new_inp->sctp_ep.secret_key, old_inp->sctp_ep.secret_key, sizeof(old_inp->sctp_ep.secret_key)); new_inp->sctp_ep.current_secret_number = old_inp->sctp_ep.current_secret_number; new_inp->sctp_ep.last_secret_number = old_inp->sctp_ep.last_secret_number; new_inp->sctp_ep.size_of_a_cookie = old_inp->sctp_ep.size_of_a_cookie; /* make it so new data pours into the new socket */ stcb->sctp_socket = new_inp->sctp_socket; stcb->sctp_ep = new_inp; /* Copy the port across */ lport = new_inp->sctp_lport = old_inp->sctp_lport; rport = stcb->rport; /* Pull the tcb from the old association */ LIST_REMOVE(stcb, sctp_tcbhash); LIST_REMOVE(stcb, sctp_tcblist); if (stcb->asoc.in_asocid_hash) { LIST_REMOVE(stcb, sctp_tcbasocidhash); } /* Now insert the new_inp into the TCP connected hash */ head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport | rport), SCTP_BASE_INFO(hashtcpmark))]; LIST_INSERT_HEAD(head, new_inp, sctp_hash); /* Its safe to access */ new_inp->sctp_flags &= ~SCTP_PCB_FLAGS_UNBOUND; /* Now move the tcb into the endpoint list */ LIST_INSERT_HEAD(&new_inp->sctp_asoc_list, stcb, sctp_tcblist); /* * Question, do we even need to worry about the ep-hash since we * only have one connection? Probably not :> so lets get rid of it * and not suck up any kernel memory in that. */ if (stcb->asoc.in_asocid_hash) { struct sctpasochead *lhd; lhd = &new_inp->sctp_asocidhash[SCTP_PCBHASH_ASOC(stcb->asoc.assoc_id, new_inp->hashasocidmark)]; LIST_INSERT_HEAD(lhd, stcb, sctp_tcbasocidhash); } /* Ok. Let's restart timer. */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { sctp_timer_start(SCTP_TIMER_TYPE_PATHMTURAISE, new_inp, stcb, net); } SCTP_INP_INFO_WUNLOCK(); if (new_inp->sctp_tcbhash != NULL) { SCTP_HASH_FREE(new_inp->sctp_tcbhash, new_inp->sctp_hashmark); new_inp->sctp_tcbhash = NULL; } if ((new_inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) == 0) { /* Subset bound, so copy in the laddr list from the old_inp */ LIST_FOREACH(oladdr, &old_inp->sctp_addr_list, sctp_nxt_addr) { laddr = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_laddr), struct sctp_laddr); if (laddr == NULL) { /* * Gak, what can we do? This assoc is really * HOSED. We probably should send an abort * here. */ SCTPDBG(SCTP_DEBUG_PCB1, "Association hosed in TCP model, out of laddr memory\n"); continue; } SCTP_INCR_LADDR_COUNT(); memset(laddr, 0, sizeof(*laddr)); (void)SCTP_GETTIME_TIMEVAL(&laddr->start_time); laddr->ifa = oladdr->ifa; atomic_add_int(&laddr->ifa->refcount, 1); LIST_INSERT_HEAD(&new_inp->sctp_addr_list, laddr, sctp_nxt_addr); new_inp->laddr_count++; if (oladdr == stcb->asoc.last_used_address) { stcb->asoc.last_used_address = laddr; } } } /* * Now any running timers need to be adjusted since we really don't * care if they are running or not just blast in the new_inp into * all of them. */ stcb->asoc.dack_timer.ep = (void *)new_inp; stcb->asoc.asconf_timer.ep = (void *)new_inp; stcb->asoc.strreset_timer.ep = (void *)new_inp; stcb->asoc.shut_guard_timer.ep = (void *)new_inp; stcb->asoc.autoclose_timer.ep = (void *)new_inp; stcb->asoc.delete_prim_timer.ep = (void *)new_inp; /* now what about the nets? */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { net->pmtu_timer.ep = (void *)new_inp; net->hb_timer.ep = (void *)new_inp; net->rxt_timer.ep = (void *)new_inp; } SCTP_INP_WUNLOCK(new_inp); SCTP_INP_WUNLOCK(old_inp); } /* * insert an laddr entry with the given ifa for the desired list */ static int sctp_insert_laddr(struct sctpladdr *list, struct sctp_ifa *ifa, uint32_t act) { struct sctp_laddr *laddr; laddr = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_laddr), struct sctp_laddr); if (laddr == NULL) { /* out of memory? */ SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } SCTP_INCR_LADDR_COUNT(); memset(laddr, 0, sizeof(*laddr)); (void)SCTP_GETTIME_TIMEVAL(&laddr->start_time); laddr->ifa = ifa; laddr->action = act; atomic_add_int(&ifa->refcount, 1); /* insert it */ LIST_INSERT_HEAD(list, laddr, sctp_nxt_addr); return (0); } /* * Remove an laddr entry from the local address list (on an assoc) */ static void sctp_remove_laddr(struct sctp_laddr *laddr) { /* remove from the list */ LIST_REMOVE(laddr, sctp_nxt_addr); sctp_free_ifa(laddr->ifa); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_laddr), laddr); SCTP_DECR_LADDR_COUNT(); } /* sctp_ifap is used to bypass normal local address validation checks */ int sctp_inpcb_bind(struct socket *so, struct sockaddr *addr, struct sctp_ifa *sctp_ifap, struct thread *p) { /* bind a ep to a socket address */ struct sctppcbhead *head; struct sctp_inpcb *inp, *inp_tmp; struct inpcb *ip_inp; int port_reuse_active = 0; int bindall; uint16_t lport; int error; uint32_t vrf_id; lport = 0; bindall = 1; inp = (struct sctp_inpcb *)so->so_pcb; ip_inp = (struct inpcb *)so->so_pcb; #ifdef SCTP_DEBUG if (addr) { SCTPDBG(SCTP_DEBUG_PCB1, "Bind called port: %d\n", ntohs(((struct sockaddr_in *)addr)->sin_port)); SCTPDBG(SCTP_DEBUG_PCB1, "Addr: "); SCTPDBG_ADDR(SCTP_DEBUG_PCB1, addr); } #endif if ((inp->sctp_flags & SCTP_PCB_FLAGS_UNBOUND) == 0) { /* already did a bind, subsequent binds NOT allowed ! */ SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } #ifdef INVARIANTS if (p == NULL) panic("null proc/thread"); #endif if (addr != NULL) { switch (addr->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; /* IPV6_V6ONLY socket? */ if (SCTP_IPV6_V6ONLY(inp)) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } if (addr->sa_len != sizeof(*sin)) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } sin = (struct sockaddr_in *)addr; lport = sin->sin_port; /* * For LOOPBACK the prison_local_ip4() call * will transmute the ip address to the * proper value. */ if (p && (error = prison_local_ip4(p->td_ucred, &sin->sin_addr)) != 0) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, error); return (error); } if (sin->sin_addr.s_addr != INADDR_ANY) { bindall = 0; } break; } #endif #ifdef INET6 case AF_INET6: { /* * Only for pure IPv6 Address. (No IPv4 * Mapped!) */ struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)addr; if (addr->sa_len != sizeof(*sin6)) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } lport = sin6->sin6_port; /* * For LOOPBACK the prison_local_ip6() call * will transmute the ipv6 address to the * proper value. */ if (p && (error = prison_local_ip6(p->td_ucred, &sin6->sin6_addr, (SCTP_IPV6_V6ONLY(inp) != 0))) != 0) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, error); return (error); } if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { bindall = 0; /* KAME hack: embed scopeid */ if (sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)) != 0) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } } /* this must be cleared for ifa_ifwithaddr() */ sin6->sin6_scope_id = 0; break; } #endif default: SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EAFNOSUPPORT); return (EAFNOSUPPORT); } } SCTP_INP_INFO_WLOCK(); SCTP_INP_WLOCK(inp); /* Setup a vrf_id to be the default for the non-bind-all case. */ vrf_id = inp->def_vrf_id; /* increase our count due to the unlock we do */ SCTP_INP_INCR_REF(inp); if (lport) { /* * Did the caller specify a port? if so we must see if an ep * already has this one bound. */ /* got to be root to get at low ports */ if (ntohs(lport) < IPPORT_RESERVED) { if ((p != NULL) && ((error = priv_check(p, PRIV_NETINET_RESERVEDPORT) ) != 0)) { SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return (error); } } SCTP_INP_WUNLOCK(inp); if (bindall) { vrf_id = inp->def_vrf_id; inp_tmp = sctp_pcb_findep(addr, 0, 1, vrf_id); if (inp_tmp != NULL) { /* * lock guy returned and lower count note * that we are not bound so inp_tmp should * NEVER be inp. And it is this inp * (inp_tmp) that gets the reference bump, * so we must lower it. */ SCTP_INP_DECR_REF(inp_tmp); /* unlock info */ if ((sctp_is_feature_on(inp, SCTP_PCB_FLAGS_PORTREUSE)) && (sctp_is_feature_on(inp_tmp, SCTP_PCB_FLAGS_PORTREUSE))) { /* * Ok, must be one-2-one and * allowing port re-use */ port_reuse_active = 1; goto continue_anyway; } SCTP_INP_DECR_REF(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EADDRINUSE); return (EADDRINUSE); } } else { inp_tmp = sctp_pcb_findep(addr, 0, 1, vrf_id); if (inp_tmp != NULL) { /* * lock guy returned and lower count note * that we are not bound so inp_tmp should * NEVER be inp. And it is this inp * (inp_tmp) that gets the reference bump, * so we must lower it. */ SCTP_INP_DECR_REF(inp_tmp); /* unlock info */ if ((sctp_is_feature_on(inp, SCTP_PCB_FLAGS_PORTREUSE)) && (sctp_is_feature_on(inp_tmp, SCTP_PCB_FLAGS_PORTREUSE))) { /* * Ok, must be one-2-one and * allowing port re-use */ port_reuse_active = 1; goto continue_anyway; } SCTP_INP_DECR_REF(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EADDRINUSE); return (EADDRINUSE); } } continue_anyway: SCTP_INP_WLOCK(inp); if (bindall) { /* verify that no lport is not used by a singleton */ if ((port_reuse_active == 0) && (inp_tmp = sctp_isport_inuse(inp, lport, vrf_id))) { /* Sorry someone already has this one bound */ if ((sctp_is_feature_on(inp, SCTP_PCB_FLAGS_PORTREUSE)) && (sctp_is_feature_on(inp_tmp, SCTP_PCB_FLAGS_PORTREUSE))) { port_reuse_active = 1; } else { SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EADDRINUSE); return (EADDRINUSE); } } } } else { uint16_t first, last, candidate; uint16_t count; int done; if (ip_inp->inp_flags & INP_HIGHPORT) { first = MODULE_GLOBAL(ipport_hifirstauto); last = MODULE_GLOBAL(ipport_hilastauto); } else if (ip_inp->inp_flags & INP_LOWPORT) { if (p && (error = priv_check(p, PRIV_NETINET_RESERVEDPORT) )) { SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, error); return (error); } first = MODULE_GLOBAL(ipport_lowfirstauto); last = MODULE_GLOBAL(ipport_lowlastauto); } else { first = MODULE_GLOBAL(ipport_firstauto); last = MODULE_GLOBAL(ipport_lastauto); } if (first > last) { uint16_t temp; temp = first; first = last; last = temp; } count = last - first + 1; /* number of candidates */ candidate = first + sctp_select_initial_TSN(&inp->sctp_ep) % (count); done = 0; while (!done) { if (sctp_isport_inuse(inp, htons(candidate), inp->def_vrf_id) == NULL) { done = 1; } if (!done) { if (--count == 0) { SCTP_INP_DECR_REF(inp); SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EADDRINUSE); return (EADDRINUSE); } if (candidate == last) candidate = first; else candidate = candidate + 1; } } lport = htons(candidate); } SCTP_INP_DECR_REF(inp); if (inp->sctp_flags & (SCTP_PCB_FLAGS_SOCKET_GONE | SCTP_PCB_FLAGS_SOCKET_ALLGONE)) { /* * this really should not happen. The guy did a non-blocking * bind and then did a close at the same time. */ SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } /* ok we look clear to give out this port, so lets setup the binding */ if (bindall) { /* binding to all addresses, so just set in the proper flags */ inp->sctp_flags |= SCTP_PCB_FLAGS_BOUNDALL; /* set the automatic addr changes from kernel flag */ if (SCTP_BASE_SYSCTL(sctp_auto_asconf) == 0) { sctp_feature_off(inp, SCTP_PCB_FLAGS_DO_ASCONF); sctp_feature_off(inp, SCTP_PCB_FLAGS_AUTO_ASCONF); } else { sctp_feature_on(inp, SCTP_PCB_FLAGS_DO_ASCONF); sctp_feature_on(inp, SCTP_PCB_FLAGS_AUTO_ASCONF); } if (SCTP_BASE_SYSCTL(sctp_multiple_asconfs) == 0) { sctp_feature_off(inp, SCTP_PCB_FLAGS_MULTIPLE_ASCONFS); } else { sctp_feature_on(inp, SCTP_PCB_FLAGS_MULTIPLE_ASCONFS); } /* * set the automatic mobility_base from kernel flag (by * micchie) */ if (SCTP_BASE_SYSCTL(sctp_mobility_base) == 0) { sctp_mobility_feature_off(inp, SCTP_MOBILITY_BASE); sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); } else { sctp_mobility_feature_on(inp, SCTP_MOBILITY_BASE); sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); } /* * set the automatic mobility_fasthandoff from kernel flag * (by micchie) */ if (SCTP_BASE_SYSCTL(sctp_mobility_fasthandoff) == 0) { sctp_mobility_feature_off(inp, SCTP_MOBILITY_FASTHANDOFF); sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); } else { sctp_mobility_feature_on(inp, SCTP_MOBILITY_FASTHANDOFF); sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); } } else { /* * bind specific, make sure flags is off and add a new * address structure to the sctp_addr_list inside the ep * structure. * * We will need to allocate one and insert it at the head. * The socketopt call can just insert new addresses in there * as well. It will also have to do the embed scope kame * hack too (before adding). */ struct sctp_ifa *ifa; union sctp_sockstore store; memset(&store, 0, sizeof(store)); switch (addr->sa_family) { #ifdef INET case AF_INET: memcpy(&store.sin, addr, sizeof(struct sockaddr_in)); store.sin.sin_port = 0; break; #endif #ifdef INET6 case AF_INET6: memcpy(&store.sin6, addr, sizeof(struct sockaddr_in6)); store.sin6.sin6_port = 0; break; #endif default: break; } /* * first find the interface with the bound address need to * zero out the port to find the address! yuck! can't do * this earlier since need port for sctp_pcb_findep() */ if (sctp_ifap != NULL) { ifa = sctp_ifap; } else { /* * Note for BSD we hit here always other O/S's will * pass things in via the sctp_ifap argument * (Panda). */ ifa = sctp_find_ifa_by_addr(&store.sa, vrf_id, SCTP_ADDR_NOT_LOCKED); } if (ifa == NULL) { /* Can't find an interface with that address */ SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EADDRNOTAVAIL); return (EADDRNOTAVAIL); } #ifdef INET6 if (addr->sa_family == AF_INET6) { /* GAK, more FIXME IFA lock? */ if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { /* Can't bind a non-existent addr. */ SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); return (EINVAL); } } #endif /* we're not bound all */ inp->sctp_flags &= ~SCTP_PCB_FLAGS_BOUNDALL; /* allow bindx() to send ASCONF's for binding changes */ sctp_feature_on(inp, SCTP_PCB_FLAGS_DO_ASCONF); /* clear automatic addr changes from kernel flag */ sctp_feature_off(inp, SCTP_PCB_FLAGS_AUTO_ASCONF); /* add this address to the endpoint list */ error = sctp_insert_laddr(&inp->sctp_addr_list, ifa, 0); if (error != 0) { SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return (error); } inp->laddr_count++; } /* find the bucket */ if (port_reuse_active) { /* Put it into tcp 1-2-1 hash */ head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashtcpmark))]; inp->sctp_flags |= SCTP_PCB_FLAGS_IN_TCPPOOL; } else { head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashmark))]; } /* put it in the bucket */ LIST_INSERT_HEAD(head, inp, sctp_hash); SCTPDBG(SCTP_DEBUG_PCB1, "Main hash to bind at head:%p, bound port:%d - in tcp_pool=%d\n", (void *)head, ntohs(lport), port_reuse_active); /* set in the port */ inp->sctp_lport = lport; /* turn off just the unbound flag */ inp->sctp_flags &= ~SCTP_PCB_FLAGS_UNBOUND; SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return (0); } static void sctp_iterator_inp_being_freed(struct sctp_inpcb *inp) { struct sctp_iterator *it, *nit; /* * We enter with the only the ITERATOR_LOCK in place and a write * lock on the inp_info stuff. */ it = sctp_it_ctl.cur_it; if (it && (it->vn != curvnet)) { /* Its not looking at our VNET */ return; } if (it && (it->inp == inp)) { /* * This is tricky and we hold the iterator lock, but when it * returns and gets the lock (when we release it) the * iterator will try to operate on inp. We need to stop that * from happening. But of course the iterator has a * reference on the stcb and inp. We can mark it and it will * stop. * * If its a single iterator situation, we set the end * iterator flag. Otherwise we set the iterator to go to the * next inp. * */ if (it->iterator_flags & SCTP_ITERATOR_DO_SINGLE_INP) { sctp_it_ctl.iterator_flags |= SCTP_ITERATOR_STOP_CUR_IT; } else { sctp_it_ctl.iterator_flags |= SCTP_ITERATOR_STOP_CUR_INP; } } /* * Now go through and remove any single reference to our inp that * may be still pending on the list */ SCTP_IPI_ITERATOR_WQ_LOCK(); TAILQ_FOREACH_SAFE(it, &sctp_it_ctl.iteratorhead, sctp_nxt_itr, nit) { if (it->vn != curvnet) { continue; } if (it->inp == inp) { /* This one points to me is it inp specific? */ if (it->iterator_flags & SCTP_ITERATOR_DO_SINGLE_INP) { /* Remove and free this one */ TAILQ_REMOVE(&sctp_it_ctl.iteratorhead, it, sctp_nxt_itr); if (it->function_atend != NULL) { (*it->function_atend) (it->pointer, it->val); } SCTP_FREE(it, SCTP_M_ITER); } else { it->inp = LIST_NEXT(it->inp, sctp_list); if (it->inp) { SCTP_INP_INCR_REF(it->inp); } } /* * When its put in the refcnt is incremented so decr * it */ SCTP_INP_DECR_REF(inp); } } SCTP_IPI_ITERATOR_WQ_UNLOCK(); } /* release sctp_inpcb unbind the port */ void sctp_inpcb_free(struct sctp_inpcb *inp, int immediate, int from) { /* * Here we free a endpoint. We must find it (if it is in the Hash * table) and remove it from there. Then we must also find it in the * overall list and remove it from there. After all removals are * complete then any timer has to be stopped. Then start the actual * freeing. a) Any local lists. b) Any associations. c) The hash of * all associations. d) finally the ep itself. */ struct sctp_tcb *asoc, *nasoc; struct sctp_laddr *laddr, *nladdr; struct inpcb *ip_pcb; struct socket *so; int being_refed = 0; struct sctp_queued_to_read *sq, *nsq; int cnt; sctp_sharedkey_t *shared_key, *nshared_key; #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 0); #endif SCTP_ITERATOR_LOCK(); /* mark any iterators on the list or being processed */ sctp_iterator_inp_being_freed(inp); SCTP_ITERATOR_UNLOCK(); so = inp->sctp_socket; if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { /* been here before.. eeks.. get out of here */ SCTP_PRINTF("This conflict in free SHOULD not be happening! from %d, imm %d\n", from, immediate); #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 1); #endif return; } SCTP_ASOC_CREATE_LOCK(inp); SCTP_INP_INFO_WLOCK(); SCTP_INP_WLOCK(inp); if (from == SCTP_CALLED_AFTER_CMPSET_OFCLOSE) { inp->sctp_flags &= ~SCTP_PCB_FLAGS_CLOSE_IP; /* socket is gone, so no more wakeups allowed */ inp->sctp_flags |= SCTP_PCB_FLAGS_DONT_WAKE; inp->sctp_flags &= ~SCTP_PCB_FLAGS_WAKEINPUT; inp->sctp_flags &= ~SCTP_PCB_FLAGS_WAKEOUTPUT; } /* First time through we have the socket lock, after that no more. */ sctp_timer_stop(SCTP_TIMER_TYPE_NEWCOOKIE, inp, NULL, NULL, SCTP_FROM_SCTP_PCB + SCTP_LOC_1); if (inp->control) { sctp_m_freem(inp->control); inp->control = NULL; } if (inp->pkt) { sctp_m_freem(inp->pkt); inp->pkt = NULL; } ip_pcb = &inp->ip_inp.inp; /* we could just cast the main pointer * here but I will be nice :> (i.e. * ip_pcb = ep;) */ if (immediate == SCTP_FREE_SHOULD_USE_GRACEFUL_CLOSE) { int cnt_in_sd; cnt_in_sd = 0; LIST_FOREACH_SAFE(asoc, &inp->sctp_asoc_list, sctp_tcblist, nasoc) { SCTP_TCB_LOCK(asoc); if (asoc->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { /* Skip guys being freed */ cnt_in_sd++; if (asoc->asoc.state & SCTP_STATE_IN_ACCEPT_QUEUE) { /* * Special case - we did not start a * kill timer on the asoc due to it * was not closed. So go ahead and * start it now. */ SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_IN_ACCEPT_QUEUE); sctp_timer_start(SCTP_TIMER_TYPE_ASOCKILL, inp, asoc, NULL); } SCTP_TCB_UNLOCK(asoc); continue; } if (((SCTP_GET_STATE(asoc) == SCTP_STATE_COOKIE_WAIT) || (SCTP_GET_STATE(asoc) == SCTP_STATE_COOKIE_ECHOED)) && (asoc->asoc.total_output_queue_size == 0)) { /* * If we have data in queue, we don't want * to just free since the app may have done, * send()/close or connect/send/close. And * it wants the data to get across first. */ /* Just abandon things in the front states */ if (sctp_free_assoc(inp, asoc, SCTP_PCBFREE_NOFORCE, SCTP_FROM_SCTP_PCB + SCTP_LOC_2) == 0) { cnt_in_sd++; } continue; } /* Disconnect the socket please */ asoc->sctp_socket = NULL; SCTP_ADD_SUBSTATE(asoc, SCTP_STATE_CLOSED_SOCKET); if ((asoc->asoc.size_on_reasm_queue > 0) || (asoc->asoc.control_pdapi) || (asoc->asoc.size_on_all_streams > 0) || (so && (so->so_rcv.sb_cc > 0))) { /* Left with Data unread */ struct mbuf *op_err; op_err = sctp_generate_cause(SCTP_CAUSE_USER_INITIATED_ABT, ""); asoc->sctp_ep->last_abort_code = SCTP_FROM_SCTP_PCB + SCTP_LOC_3; sctp_send_abort_tcb(asoc, op_err, SCTP_SO_LOCKED); SCTP_STAT_INCR_COUNTER32(sctps_aborted); if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } if (sctp_free_assoc(inp, asoc, SCTP_PCBFREE_NOFORCE, SCTP_FROM_SCTP_PCB + SCTP_LOC_4) == 0) { cnt_in_sd++; } continue; } else if (TAILQ_EMPTY(&asoc->asoc.send_queue) && TAILQ_EMPTY(&asoc->asoc.sent_queue) && (asoc->asoc.stream_queue_cnt == 0)) { if ((*asoc->asoc.ss_functions.sctp_ss_is_user_msgs_incomplete) (asoc, &asoc->asoc)) { goto abort_anyway; } if ((SCTP_GET_STATE(asoc) != SCTP_STATE_SHUTDOWN_SENT) && (SCTP_GET_STATE(asoc) != SCTP_STATE_SHUTDOWN_ACK_SENT)) { struct sctp_nets *netp; /* * there is nothing queued to send, * so I send shutdown */ if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } SCTP_SET_STATE(asoc, SCTP_STATE_SHUTDOWN_SENT); sctp_stop_timers_for_shutdown(asoc); if (asoc->asoc.alternate) { netp = asoc->asoc.alternate; } else { netp = asoc->asoc.primary_destination; } sctp_send_shutdown(asoc, netp); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN, asoc->sctp_ep, asoc, netp); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, asoc->sctp_ep, asoc, NULL); sctp_chunk_output(inp, asoc, SCTP_OUTPUT_FROM_SHUT_TMR, SCTP_SO_LOCKED); } } else { /* mark into shutdown pending */ SCTP_ADD_SUBSTATE(asoc, SCTP_STATE_SHUTDOWN_PENDING); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, asoc->sctp_ep, asoc, NULL); if ((*asoc->asoc.ss_functions.sctp_ss_is_user_msgs_incomplete) (asoc, &asoc->asoc)) { SCTP_ADD_SUBSTATE(asoc, SCTP_STATE_PARTIAL_MSG_LEFT); } if (TAILQ_EMPTY(&asoc->asoc.send_queue) && TAILQ_EMPTY(&asoc->asoc.sent_queue) && (asoc->asoc.state & SCTP_STATE_PARTIAL_MSG_LEFT)) { struct mbuf *op_err; abort_anyway: op_err = sctp_generate_cause(SCTP_CAUSE_USER_INITIATED_ABT, ""); asoc->sctp_ep->last_abort_code = SCTP_FROM_SCTP_PCB + SCTP_LOC_5; sctp_send_abort_tcb(asoc, op_err, SCTP_SO_LOCKED); SCTP_STAT_INCR_COUNTER32(sctps_aborted); if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } if (sctp_free_assoc(inp, asoc, SCTP_PCBFREE_NOFORCE, SCTP_FROM_SCTP_PCB + SCTP_LOC_6) == 0) { cnt_in_sd++; } continue; } else { sctp_chunk_output(inp, asoc, SCTP_OUTPUT_FROM_CLOSING, SCTP_SO_LOCKED); } } cnt_in_sd++; SCTP_TCB_UNLOCK(asoc); } /* now is there some left in our SHUTDOWN state? */ if (cnt_in_sd) { #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 2); #endif inp->sctp_socket = NULL; SCTP_INP_WUNLOCK(inp); SCTP_ASOC_CREATE_UNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return; } } inp->sctp_socket = NULL; if ((inp->sctp_flags & SCTP_PCB_FLAGS_UNBOUND) != SCTP_PCB_FLAGS_UNBOUND) { /* * ok, this guy has been bound. It's port is somewhere in * the SCTP_BASE_INFO(hash table). Remove it! */ LIST_REMOVE(inp, sctp_hash); inp->sctp_flags |= SCTP_PCB_FLAGS_UNBOUND; } /* * If there is a timer running to kill us, forget it, since it may * have a contest on the INP lock.. which would cause us to die ... */ cnt = 0; LIST_FOREACH_SAFE(asoc, &inp->sctp_asoc_list, sctp_tcblist, nasoc) { SCTP_TCB_LOCK(asoc); if (asoc->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { if (asoc->asoc.state & SCTP_STATE_IN_ACCEPT_QUEUE) { SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_IN_ACCEPT_QUEUE); sctp_timer_start(SCTP_TIMER_TYPE_ASOCKILL, inp, asoc, NULL); } cnt++; SCTP_TCB_UNLOCK(asoc); continue; } /* Free associations that are NOT killing us */ if ((SCTP_GET_STATE(asoc) != SCTP_STATE_COOKIE_WAIT) && ((asoc->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) == 0)) { struct mbuf *op_err; op_err = sctp_generate_cause(SCTP_CAUSE_USER_INITIATED_ABT, ""); asoc->sctp_ep->last_abort_code = SCTP_FROM_SCTP_PCB + SCTP_LOC_7; sctp_send_abort_tcb(asoc, op_err, SCTP_SO_LOCKED); SCTP_STAT_INCR_COUNTER32(sctps_aborted); } else if (asoc->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) { cnt++; SCTP_TCB_UNLOCK(asoc); continue; } if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } if (sctp_free_assoc(inp, asoc, SCTP_PCBFREE_FORCE, SCTP_FROM_SCTP_PCB + SCTP_LOC_8) == 0) { cnt++; } } if (cnt) { /* Ok we have someone out there that will kill us */ #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 3); #endif SCTP_INP_WUNLOCK(inp); SCTP_ASOC_CREATE_UNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return; } if (SCTP_INP_LOCK_CONTENDED(inp)) being_refed++; if (SCTP_INP_READ_CONTENDED(inp)) being_refed++; if (SCTP_ASOC_CREATE_LOCK_CONTENDED(inp)) being_refed++; if ((inp->refcount) || (being_refed) || (inp->sctp_flags & SCTP_PCB_FLAGS_CLOSE_IP)) { #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 4); #endif sctp_timer_start(SCTP_TIMER_TYPE_INPKILL, inp, NULL, NULL); SCTP_INP_WUNLOCK(inp); SCTP_ASOC_CREATE_UNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); return; } inp->sctp_ep.signature_change.type = 0; inp->sctp_flags |= SCTP_PCB_FLAGS_SOCKET_ALLGONE; /* * Remove it from the list .. last thing we need a lock for. */ LIST_REMOVE(inp, sctp_list); SCTP_INP_WUNLOCK(inp); SCTP_ASOC_CREATE_UNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); /* * Now we release all locks. Since this INP cannot be found anymore * except possibly by the kill timer that might be running. We call * the drain function here. It should hit the case were it sees the * ACTIVE flag cleared and exit out freeing us to proceed and * destroy everything. */ if (from != SCTP_CALLED_FROM_INPKILL_TIMER) { (void)SCTP_OS_TIMER_STOP_DRAIN(&inp->sctp_ep.signature_change.timer); } else { /* Probably un-needed */ (void)SCTP_OS_TIMER_STOP(&inp->sctp_ep.signature_change.timer); } #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 5); #endif if ((inp->sctp_asocidhash) != NULL) { SCTP_HASH_FREE(inp->sctp_asocidhash, inp->hashasocidmark); inp->sctp_asocidhash = NULL; } /* sa_ignore FREED_MEMORY */ TAILQ_FOREACH_SAFE(sq, &inp->read_queue, next, nsq) { /* Its only abandoned if it had data left */ if (sq->length) SCTP_STAT_INCR(sctps_left_abandon); TAILQ_REMOVE(&inp->read_queue, sq, next); sctp_free_remote_addr(sq->whoFrom); if (so) so->so_rcv.sb_cc -= sq->length; if (sq->data) { sctp_m_freem(sq->data); sq->data = NULL; } /* * no need to free the net count, since at this point all * assoc's are gone. */ sctp_free_a_readq(NULL, sq); } /* Now the sctp_pcb things */ /* * free each asoc if it is not already closed/free. we can't use the * macro here since le_next will get freed as part of the * sctp_free_assoc() call. */ if (ip_pcb->inp_options) { (void)sctp_m_free(ip_pcb->inp_options); ip_pcb->inp_options = 0; } #ifdef INET6 if (ip_pcb->inp_vflag & INP_IPV6) { ip6_freepcbopts(ip_pcb->in6p_outputopts); } #endif /* INET6 */ ip_pcb->inp_vflag = 0; /* free up authentication fields */ if (inp->sctp_ep.local_auth_chunks != NULL) sctp_free_chunklist(inp->sctp_ep.local_auth_chunks); if (inp->sctp_ep.local_hmacs != NULL) sctp_free_hmaclist(inp->sctp_ep.local_hmacs); LIST_FOREACH_SAFE(shared_key, &inp->sctp_ep.shared_keys, next, nshared_key) { LIST_REMOVE(shared_key, next); sctp_free_sharedkey(shared_key); /* sa_ignore FREED_MEMORY */ } /* * if we have an address list the following will free the list of * ifaddr's that are set into this ep. Again macro limitations here, * since the LIST_FOREACH could be a bad idea. */ LIST_FOREACH_SAFE(laddr, &inp->sctp_addr_list, sctp_nxt_addr, nladdr) { sctp_remove_laddr(laddr); } #ifdef SCTP_TRACK_FREED_ASOCS /* TEMP CODE */ LIST_FOREACH_SAFE(asoc, &inp->sctp_asoc_free_list, sctp_tcblist, nasoc) { LIST_REMOVE(asoc, sctp_tcblist); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), asoc); SCTP_DECR_ASOC_COUNT(); } /* *** END TEMP CODE *** */ #endif /* Now lets see about freeing the EP hash table. */ if (inp->sctp_tcbhash != NULL) { SCTP_HASH_FREE(inp->sctp_tcbhash, inp->sctp_hashmark); inp->sctp_tcbhash = NULL; } /* Now we must put the ep memory back into the zone pool */ crfree(inp->ip_inp.inp.inp_cred); INP_LOCK_DESTROY(&inp->ip_inp.inp); SCTP_INP_LOCK_DESTROY(inp); SCTP_INP_READ_DESTROY(inp); SCTP_ASOC_CREATE_LOCK_DESTROY(inp); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_ep), inp); SCTP_DECR_EP_COUNT(); } struct sctp_nets * sctp_findnet(struct sctp_tcb *stcb, struct sockaddr *addr) { struct sctp_nets *net; /* locate the address */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { if (sctp_cmpaddr(addr, (struct sockaddr *)&net->ro._l_addr)) return (net); } return (NULL); } int sctp_is_address_on_local_host(struct sockaddr *addr, uint32_t vrf_id) { struct sctp_ifa *sctp_ifa; sctp_ifa = sctp_find_ifa_by_addr(addr, vrf_id, SCTP_ADDR_NOT_LOCKED); if (sctp_ifa) { return (1); } else { return (0); } } /* * add's a remote endpoint address, done with the INIT/INIT-ACK as well as * when a ASCONF arrives that adds it. It will also initialize all the cwnd * stats of stuff. */ int sctp_add_remote_addr(struct sctp_tcb *stcb, struct sockaddr *newaddr, struct sctp_nets **netp, uint16_t port, int set_scope, int from) { /* * The following is redundant to the same lines in the * sctp_aloc_assoc() but is needed since others call the add address * function */ struct sctp_nets *net, *netfirst; int addr_inscope; SCTPDBG(SCTP_DEBUG_PCB1, "Adding an address (from:%d) to the peer: ", from); SCTPDBG_ADDR(SCTP_DEBUG_PCB1, newaddr); netfirst = sctp_findnet(stcb, newaddr); if (netfirst) { /* * Lie and return ok, we don't want to make the association * go away for this behavior. It will happen in the TCP * model in a connected socket. It does not reach the hash * table until after the association is built so it can't be * found. Mark as reachable, since the initial creation will * have been cleared and the NOT_IN_ASSOC flag will have * been added... and we don't want to end up removing it * back out. */ if (netfirst->dest_state & SCTP_ADDR_UNCONFIRMED) { netfirst->dest_state = (SCTP_ADDR_REACHABLE | SCTP_ADDR_UNCONFIRMED); } else { netfirst->dest_state = SCTP_ADDR_REACHABLE; } return (0); } addr_inscope = 1; switch (newaddr->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; sin = (struct sockaddr_in *)newaddr; if (sin->sin_addr.s_addr == 0) { /* Invalid address */ return (-1); } /* zero out the zero area */ memset(&sin->sin_zero, 0, sizeof(sin->sin_zero)); /* assure len is set */ sin->sin_len = sizeof(struct sockaddr_in); if (set_scope) { if (IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) { stcb->asoc.scope.ipv4_local_scope = 1; } } else { /* Validate the address is in scope */ if ((IN4_ISPRIVATE_ADDRESS(&sin->sin_addr)) && (stcb->asoc.scope.ipv4_local_scope == 0)) { addr_inscope = 0; } } break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)newaddr; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* Invalid address */ return (-1); } /* assure len is set */ sin6->sin6_len = sizeof(struct sockaddr_in6); if (set_scope) { if (sctp_is_address_on_local_host(newaddr, stcb->asoc.vrf_id)) { stcb->asoc.scope.loopback_scope = 1; stcb->asoc.scope.local_scope = 0; stcb->asoc.scope.ipv4_local_scope = 1; stcb->asoc.scope.site_scope = 1; } else if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr)) { /* * If the new destination is a * LINK_LOCAL we must have common * site scope. Don't set the local * scope since we may not share all * links, only loopback can do this. * Links on the local network would * also be on our private network * for v4 too. */ stcb->asoc.scope.ipv4_local_scope = 1; stcb->asoc.scope.site_scope = 1; } else if (IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr)) { /* * If the new destination is * SITE_LOCAL then we must have site * scope in common. */ stcb->asoc.scope.site_scope = 1; } } else { /* Validate the address is in scope */ if (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr) && (stcb->asoc.scope.loopback_scope == 0)) { addr_inscope = 0; } else if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) && (stcb->asoc.scope.local_scope == 0)) { addr_inscope = 0; } else if (IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr) && (stcb->asoc.scope.site_scope == 0)) { addr_inscope = 0; } } break; } #endif default: /* not supported family type */ return (-1); } net = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_net), struct sctp_nets); if (net == NULL) { return (-1); } SCTP_INCR_RADDR_COUNT(); memset(net, 0, sizeof(struct sctp_nets)); (void)SCTP_GETTIME_TIMEVAL(&net->start_time); memcpy(&net->ro._l_addr, newaddr, newaddr->sa_len); switch (newaddr->sa_family) { #ifdef INET case AF_INET: ((struct sockaddr_in *)&net->ro._l_addr)->sin_port = stcb->rport; break; #endif #ifdef INET6 case AF_INET6: ((struct sockaddr_in6 *)&net->ro._l_addr)->sin6_port = stcb->rport; break; #endif default: break; } net->addr_is_local = sctp_is_address_on_local_host(newaddr, stcb->asoc.vrf_id); if (net->addr_is_local && ((set_scope || (from == SCTP_ADDR_IS_CONFIRMED)))) { stcb->asoc.scope.loopback_scope = 1; stcb->asoc.scope.ipv4_local_scope = 1; stcb->asoc.scope.local_scope = 0; stcb->asoc.scope.site_scope = 1; addr_inscope = 1; } net->failure_threshold = stcb->asoc.def_net_failure; net->pf_threshold = stcb->asoc.def_net_pf_threshold; if (addr_inscope == 0) { net->dest_state = (SCTP_ADDR_REACHABLE | SCTP_ADDR_OUT_OF_SCOPE); } else { if (from == SCTP_ADDR_IS_CONFIRMED) /* SCTP_ADDR_IS_CONFIRMED is passed by connect_x */ net->dest_state = SCTP_ADDR_REACHABLE; else net->dest_state = SCTP_ADDR_REACHABLE | SCTP_ADDR_UNCONFIRMED; } /* * We set this to 0, the timer code knows that this means its an * initial value */ net->rto_needed = 1; net->RTO = 0; net->RTO_measured = 0; stcb->asoc.numnets++; net->ref_count = 1; net->cwr_window_tsn = net->last_cwr_tsn = stcb->asoc.sending_seq - 1; net->port = port; net->dscp = stcb->asoc.default_dscp; #ifdef INET6 net->flowlabel = stcb->asoc.default_flowlabel; #endif if (sctp_stcb_is_feature_on(stcb->sctp_ep, stcb, SCTP_PCB_FLAGS_DONOT_HEARTBEAT)) { net->dest_state |= SCTP_ADDR_NOHB; } else { net->dest_state &= ~SCTP_ADDR_NOHB; } if (sctp_stcb_is_feature_on(stcb->sctp_ep, stcb, SCTP_PCB_FLAGS_DO_NOT_PMTUD)) { net->dest_state |= SCTP_ADDR_NO_PMTUD; } else { net->dest_state &= ~SCTP_ADDR_NO_PMTUD; } net->heart_beat_delay = stcb->asoc.heart_beat_delay; /* Init the timer structure */ SCTP_OS_TIMER_INIT(&net->rxt_timer.timer); SCTP_OS_TIMER_INIT(&net->pmtu_timer.timer); SCTP_OS_TIMER_INIT(&net->hb_timer.timer); /* Now generate a route for this guy */ #ifdef INET6 /* KAME hack: embed scopeid */ if (newaddr->sa_family == AF_INET6) { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; (void)sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)); sin6->sin6_scope_id = 0; } #endif SCTP_RTALLOC((sctp_route_t *)&net->ro, stcb->asoc.vrf_id, stcb->sctp_ep->fibnum); net->src_addr_selected = 0; if (SCTP_ROUTE_HAS_VALID_IFN(&net->ro)) { /* Get source address */ net->ro._s_addr = sctp_source_address_selection(stcb->sctp_ep, stcb, (sctp_route_t *)&net->ro, net, 0, stcb->asoc.vrf_id); if (stcb->asoc.default_mtu > 0) { net->mtu = stcb->asoc.default_mtu; switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: net->mtu += SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: net->mtu += SCTP_MIN_OVERHEAD; break; #endif default: break; } #if defined(INET) || defined(INET6) if (net->port) { net->mtu += (uint32_t)sizeof(struct udphdr); } #endif } else if (net->ro._s_addr != NULL) { uint32_t imtu, rmtu, hcmtu; net->src_addr_selected = 1; /* Now get the interface MTU */ if (net->ro._s_addr->ifn_p != NULL) { imtu = SCTP_GATHER_MTU_FROM_INTFC(net->ro._s_addr->ifn_p); } else { imtu = 0; } - rmtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, net->ro.ro_rt); + rmtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, net->ro.ro_nh); hcmtu = sctp_hc_get_mtu(&net->ro._l_addr, stcb->sctp_ep->fibnum); net->mtu = sctp_min_mtu(hcmtu, rmtu, imtu); - if (rmtu == 0) { - /* - * Start things off to match mtu of - * interface please. - */ - SCTP_SET_MTU_OF_ROUTE(&net->ro._l_addr.sa, - net->ro.ro_rt, net->mtu); - } } } if (net->mtu == 0) { if (stcb->asoc.default_mtu > 0) { net->mtu = stcb->asoc.default_mtu; switch (net->ro._l_addr.sa.sa_family) { #ifdef INET case AF_INET: net->mtu += SCTP_MIN_V4_OVERHEAD; break; #endif #ifdef INET6 case AF_INET6: net->mtu += SCTP_MIN_OVERHEAD; break; #endif default: break; } #if defined(INET) || defined(INET6) if (net->port) { net->mtu += (uint32_t)sizeof(struct udphdr); } #endif } else { switch (newaddr->sa_family) { #ifdef INET case AF_INET: net->mtu = SCTP_DEFAULT_MTU; break; #endif #ifdef INET6 case AF_INET6: net->mtu = 1280; break; #endif default: break; } } } #if defined(INET) || defined(INET6) if (net->port) { net->mtu -= (uint32_t)sizeof(struct udphdr); } #endif if (from == SCTP_ALLOC_ASOC) { stcb->asoc.smallest_mtu = net->mtu; } if (stcb->asoc.smallest_mtu > net->mtu) { sctp_pathmtu_adjustment(stcb, net->mtu); } #ifdef INET6 if (newaddr->sa_family == AF_INET6) { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; (void)sa6_recoverscope(sin6); } #endif /* JRS - Use the congestion control given in the CC module */ if (stcb->asoc.cc_functions.sctp_set_initial_cc_param != NULL) (*stcb->asoc.cc_functions.sctp_set_initial_cc_param) (stcb, net); /* * CMT: CUC algo - set find_pseudo_cumack to TRUE (1) at beginning * of assoc (2005/06/27, iyengar@cis.udel.edu) */ net->find_pseudo_cumack = 1; net->find_rtx_pseudo_cumack = 1; /* Choose an initial flowid. */ net->flowid = stcb->asoc.my_vtag ^ ntohs(stcb->rport) ^ ntohs(stcb->sctp_ep->sctp_lport); net->flowtype = M_HASHTYPE_OPAQUE_HASH; if (netp) { *netp = net; } netfirst = TAILQ_FIRST(&stcb->asoc.nets); - if (net->ro.ro_rt == NULL) { + if (net->ro.ro_nh == NULL) { /* Since we have no route put it at the back */ TAILQ_INSERT_TAIL(&stcb->asoc.nets, net, sctp_next); } else if (netfirst == NULL) { /* We are the first one in the pool. */ TAILQ_INSERT_HEAD(&stcb->asoc.nets, net, sctp_next); - } else if (netfirst->ro.ro_rt == NULL) { + } else if (netfirst->ro.ro_nh == NULL) { /* * First one has NO route. Place this one ahead of the first * one. */ TAILQ_INSERT_HEAD(&stcb->asoc.nets, net, sctp_next); - } else if (net->ro.ro_rt->rt_ifp != netfirst->ro.ro_rt->rt_ifp) { + } else if (net->ro.ro_nh->nh_ifp != netfirst->ro.ro_nh->nh_ifp) { /* * This one has a different interface than the one at the * top of the list. Place it ahead. */ TAILQ_INSERT_HEAD(&stcb->asoc.nets, net, sctp_next); } else { /* * Ok we have the same interface as the first one. Move * forward until we find either a) one with a NULL route... * insert ahead of that b) one with a different ifp.. insert * after that. c) end of the list.. insert at the tail. */ struct sctp_nets *netlook; do { netlook = TAILQ_NEXT(netfirst, sctp_next); if (netlook == NULL) { /* End of the list */ TAILQ_INSERT_TAIL(&stcb->asoc.nets, net, sctp_next); break; - } else if (netlook->ro.ro_rt == NULL) { + } else if (netlook->ro.ro_nh == NULL) { /* next one has NO route */ TAILQ_INSERT_BEFORE(netfirst, net, sctp_next); break; - } else if (netlook->ro.ro_rt->rt_ifp != net->ro.ro_rt->rt_ifp) { + } else if (netlook->ro.ro_nh->nh_ifp != net->ro.ro_nh->nh_ifp) { TAILQ_INSERT_AFTER(&stcb->asoc.nets, netlook, net, sctp_next); break; } /* Shift forward */ netfirst = netlook; } while (netlook != NULL); } /* got to have a primary set */ if (stcb->asoc.primary_destination == 0) { stcb->asoc.primary_destination = net; - } else if ((stcb->asoc.primary_destination->ro.ro_rt == NULL) && - (net->ro.ro_rt) && + } else if ((stcb->asoc.primary_destination->ro.ro_nh == NULL) && + (net->ro.ro_nh) && ((net->dest_state & SCTP_ADDR_UNCONFIRMED) == 0)) { /* No route to current primary adopt new primary */ stcb->asoc.primary_destination = net; } /* Validate primary is first */ net = TAILQ_FIRST(&stcb->asoc.nets); if ((net != stcb->asoc.primary_destination) && (stcb->asoc.primary_destination)) { /* * first one on the list is NOT the primary sctp_cmpaddr() * is much more efficient if the primary is the first on the * list, make it so. */ TAILQ_REMOVE(&stcb->asoc.nets, stcb->asoc.primary_destination, sctp_next); TAILQ_INSERT_HEAD(&stcb->asoc.nets, stcb->asoc.primary_destination, sctp_next); } return (0); } static uint32_t sctp_aloc_a_assoc_id(struct sctp_inpcb *inp, struct sctp_tcb *stcb) { uint32_t id; struct sctpasochead *head; struct sctp_tcb *lstcb; try_again: if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { /* TSNH */ return (0); } /* * We don't allow assoc id to be one of SCTP_FUTURE_ASSOC, * SCTP_CURRENT_ASSOC and SCTP_ALL_ASSOC. */ if (inp->sctp_associd_counter <= SCTP_ALL_ASSOC) { inp->sctp_associd_counter = SCTP_ALL_ASSOC + 1; } id = inp->sctp_associd_counter; inp->sctp_associd_counter++; lstcb = sctp_findasoc_ep_asocid_locked(inp, (sctp_assoc_t)id, 0); if (lstcb) { goto try_again; } head = &inp->sctp_asocidhash[SCTP_PCBHASH_ASOC(id, inp->hashasocidmark)]; LIST_INSERT_HEAD(head, stcb, sctp_tcbasocidhash); stcb->asoc.in_asocid_hash = 1; return (id); } /* * allocate an association and add it to the endpoint. The caller must be * careful to add all additional addresses once they are know right away or * else the assoc will be may experience a blackout scenario. */ struct sctp_tcb * sctp_aloc_assoc(struct sctp_inpcb *inp, struct sockaddr *firstaddr, int *error, uint32_t override_tag, uint32_t vrf_id, uint16_t o_streams, uint16_t port, struct thread *p, int initialize_auth_params) { /* note the p argument is only valid in unbound sockets */ struct sctp_tcb *stcb; struct sctp_association *asoc; struct sctpasochead *head; uint16_t rport; int err; /* * Assumption made here: Caller has done a * sctp_findassociation_ep_addr(ep, addr's); to make sure the * address does not exist already. */ if (SCTP_BASE_INFO(ipi_count_asoc) >= SCTP_MAX_NUM_OF_ASOC) { /* Hit max assoc, sorry no more */ SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOBUFS); *error = ENOBUFS; return (NULL); } if (firstaddr == NULL) { SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } SCTP_INP_RLOCK(inp); if ((inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL) && ((sctp_is_feature_off(inp, SCTP_PCB_FLAGS_PORTREUSE)) || (inp->sctp_flags & SCTP_PCB_FLAGS_CONNECTED))) { /* * If its in the TCP pool, its NOT allowed to create an * association. The parent listener needs to call * sctp_aloc_assoc.. or the one-2-many socket. If a peeled * off, or connected one does this.. its an error. */ SCTP_INP_RUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } if ((inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL) || (inp->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE)) { if ((inp->sctp_flags & SCTP_PCB_FLAGS_WAS_CONNECTED) || (inp->sctp_flags & SCTP_PCB_FLAGS_WAS_ABORTED)) { SCTP_INP_RUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } } SCTPDBG(SCTP_DEBUG_PCB3, "Allocate an association for peer:"); #ifdef SCTP_DEBUG if (firstaddr) { SCTPDBG_ADDR(SCTP_DEBUG_PCB3, firstaddr); switch (firstaddr->sa_family) { #ifdef INET case AF_INET: SCTPDBG(SCTP_DEBUG_PCB3, "Port:%d\n", ntohs(((struct sockaddr_in *)firstaddr)->sin_port)); break; #endif #ifdef INET6 case AF_INET6: SCTPDBG(SCTP_DEBUG_PCB3, "Port:%d\n", ntohs(((struct sockaddr_in6 *)firstaddr)->sin6_port)); break; #endif default: break; } } else { SCTPDBG(SCTP_DEBUG_PCB3, "None\n"); } #endif /* SCTP_DEBUG */ switch (firstaddr->sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; sin = (struct sockaddr_in *)firstaddr; if ((ntohs(sin->sin_port) == 0) || (sin->sin_addr.s_addr == INADDR_ANY) || (sin->sin_addr.s_addr == INADDR_BROADCAST) || IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { /* Invalid address */ SCTP_INP_RUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } rport = sin->sin_port; break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)firstaddr; if ((ntohs(sin6->sin6_port) == 0) || IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) || IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { /* Invalid address */ SCTP_INP_RUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } rport = sin6->sin6_port; break; } #endif default: /* not supported family type */ SCTP_INP_RUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } SCTP_INP_RUNLOCK(inp); if (inp->sctp_flags & SCTP_PCB_FLAGS_UNBOUND) { /* * If you have not performed a bind, then we need to do the * ephemeral bind for you. */ if ((err = sctp_inpcb_bind(inp->sctp_socket, (struct sockaddr *)NULL, (struct sctp_ifa *)NULL, p ))) { /* bind error, probably perm */ *error = err; return (NULL); } } stcb = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_asoc), struct sctp_tcb); if (stcb == NULL) { /* out of memory? */ SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOMEM); *error = ENOMEM; return (NULL); } SCTP_INCR_ASOC_COUNT(); memset(stcb, 0, sizeof(*stcb)); asoc = &stcb->asoc; SCTP_TCB_LOCK_INIT(stcb); SCTP_TCB_SEND_LOCK_INIT(stcb); stcb->rport = rport; /* setup back pointer's */ stcb->sctp_ep = inp; stcb->sctp_socket = inp->sctp_socket; if ((err = sctp_init_asoc(inp, stcb, override_tag, vrf_id, o_streams))) { /* failed */ SCTP_TCB_LOCK_DESTROY(stcb); SCTP_TCB_SEND_LOCK_DESTROY(stcb); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), stcb); SCTP_DECR_ASOC_COUNT(); *error = err; return (NULL); } /* and the port */ SCTP_INP_INFO_WLOCK(); SCTP_INP_WLOCK(inp); if (inp->sctp_flags & (SCTP_PCB_FLAGS_SOCKET_GONE | SCTP_PCB_FLAGS_SOCKET_ALLGONE)) { /* inpcb freed while alloc going on */ SCTP_TCB_LOCK_DESTROY(stcb); SCTP_TCB_SEND_LOCK_DESTROY(stcb); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), stcb); SCTP_INP_WUNLOCK(inp); SCTP_INP_INFO_WUNLOCK(); SCTP_DECR_ASOC_COUNT(); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, EINVAL); *error = EINVAL; return (NULL); } SCTP_TCB_LOCK(stcb); asoc->assoc_id = sctp_aloc_a_assoc_id(inp, stcb); /* now that my_vtag is set, add it to the hash */ head = &SCTP_BASE_INFO(sctp_asochash)[SCTP_PCBHASH_ASOC(stcb->asoc.my_vtag, SCTP_BASE_INFO(hashasocmark))]; /* put it in the bucket in the vtag hash of assoc's for the system */ LIST_INSERT_HEAD(head, stcb, sctp_asocs); SCTP_INP_INFO_WUNLOCK(); if ((err = sctp_add_remote_addr(stcb, firstaddr, NULL, port, SCTP_DO_SETSCOPE, SCTP_ALLOC_ASOC))) { /* failure.. memory error? */ if (asoc->strmout) { SCTP_FREE(asoc->strmout, SCTP_M_STRMO); asoc->strmout = NULL; } if (asoc->mapping_array) { SCTP_FREE(asoc->mapping_array, SCTP_M_MAP); asoc->mapping_array = NULL; } if (asoc->nr_mapping_array) { SCTP_FREE(asoc->nr_mapping_array, SCTP_M_MAP); asoc->nr_mapping_array = NULL; } SCTP_DECR_ASOC_COUNT(); SCTP_TCB_UNLOCK(stcb); SCTP_TCB_LOCK_DESTROY(stcb); SCTP_TCB_SEND_LOCK_DESTROY(stcb); LIST_REMOVE(stcb, sctp_tcbasocidhash); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), stcb); SCTP_INP_WUNLOCK(inp); SCTP_LTRACE_ERR_RET(inp, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOBUFS); *error = ENOBUFS; return (NULL); } /* Init all the timers */ SCTP_OS_TIMER_INIT(&asoc->dack_timer.timer); SCTP_OS_TIMER_INIT(&asoc->strreset_timer.timer); SCTP_OS_TIMER_INIT(&asoc->asconf_timer.timer); SCTP_OS_TIMER_INIT(&asoc->shut_guard_timer.timer); SCTP_OS_TIMER_INIT(&asoc->autoclose_timer.timer); SCTP_OS_TIMER_INIT(&asoc->delete_prim_timer.timer); LIST_INSERT_HEAD(&inp->sctp_asoc_list, stcb, sctp_tcblist); /* now file the port under the hash as well */ if (inp->sctp_tcbhash != NULL) { head = &inp->sctp_tcbhash[SCTP_PCBHASH_ALLADDR(stcb->rport, inp->sctp_hashmark)]; LIST_INSERT_HEAD(head, stcb, sctp_tcbhash); } if (initialize_auth_params == SCTP_INITIALIZE_AUTH_PARAMS) { sctp_initialize_auth_params(inp, stcb); } SCTP_INP_WUNLOCK(inp); SCTPDBG(SCTP_DEBUG_PCB1, "Association %p now allocated\n", (void *)stcb); return (stcb); } void sctp_remove_net(struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_inpcb *inp; struct sctp_association *asoc; inp = stcb->sctp_ep; asoc = &stcb->asoc; asoc->numnets--; TAILQ_REMOVE(&asoc->nets, net, sctp_next); if (net == asoc->primary_destination) { /* Reset primary */ struct sctp_nets *lnet; lnet = TAILQ_FIRST(&asoc->nets); /* * Mobility adaptation Ideally, if deleted destination is * the primary, it becomes a fast retransmission trigger by * the subsequent SET PRIMARY. (by micchie) */ if (sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_BASE) || sctp_is_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_FASTHANDOFF)) { SCTPDBG(SCTP_DEBUG_ASCONF1, "remove_net: primary dst is deleting\n"); if (asoc->deleted_primary != NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "remove_net: deleted primary may be already stored\n"); goto out; } asoc->deleted_primary = net; atomic_add_int(&net->ref_count, 1); memset(&net->lastsa, 0, sizeof(net->lastsa)); memset(&net->lastsv, 0, sizeof(net->lastsv)); sctp_mobility_feature_on(stcb->sctp_ep, SCTP_MOBILITY_PRIM_DELETED); sctp_timer_start(SCTP_TIMER_TYPE_PRIM_DELETED, stcb->sctp_ep, stcb, NULL); } out: /* Try to find a confirmed primary */ asoc->primary_destination = sctp_find_alternate_net(stcb, lnet, 0); } if (net == asoc->last_data_chunk_from) { /* Reset primary */ asoc->last_data_chunk_from = TAILQ_FIRST(&asoc->nets); } if (net == asoc->last_control_chunk_from) { /* Clear net */ asoc->last_control_chunk_from = NULL; } if (net == stcb->asoc.alternate) { sctp_free_remote_addr(stcb->asoc.alternate); stcb->asoc.alternate = NULL; } sctp_timer_stop(SCTP_TIMER_TYPE_PATHMTURAISE, inp, stcb, net, SCTP_FROM_SCTP_PCB + SCTP_LOC_9); sctp_timer_stop(SCTP_TIMER_TYPE_HEARTBEAT, inp, stcb, net, SCTP_FROM_SCTP_PCB + SCTP_LOC_10); net->dest_state |= SCTP_ADDR_BEING_DELETED; sctp_free_remote_addr(net); } /* * remove a remote endpoint address from an association, it will fail if the * address does not exist. */ int sctp_del_remote_addr(struct sctp_tcb *stcb, struct sockaddr *remaddr) { /* * Here we need to remove a remote address. This is quite simple, we * first find it in the list of address for the association * (tasoc->asoc.nets) and then if it is there, we do a LIST_REMOVE * on that item. Note we do not allow it to be removed if there are * no other addresses. */ struct sctp_association *asoc; struct sctp_nets *net, *nnet; asoc = &stcb->asoc; /* locate the address */ TAILQ_FOREACH_SAFE(net, &asoc->nets, sctp_next, nnet) { if (net->ro._l_addr.sa.sa_family != remaddr->sa_family) { continue; } if (sctp_cmpaddr((struct sockaddr *)&net->ro._l_addr, remaddr)) { /* we found the guy */ if (asoc->numnets < 2) { /* Must have at LEAST two remote addresses */ return (-1); } else { sctp_remove_net(stcb, net); return (0); } } } /* not found. */ return (-2); } void sctp_delete_from_timewait(uint32_t tag, uint16_t lport, uint16_t rport) { struct sctpvtaghead *chain; struct sctp_tagblock *twait_block; int found = 0; int i; chain = &SCTP_BASE_INFO(vtag_timewait)[(tag % SCTP_STACK_VTAG_HASH_SIZE)]; LIST_FOREACH(twait_block, chain, sctp_nxt_tagblock) { for (i = 0; i < SCTP_NUMBER_IN_VTAG_BLOCK; i++) { if ((twait_block->vtag_block[i].v_tag == tag) && (twait_block->vtag_block[i].lport == lport) && (twait_block->vtag_block[i].rport == rport)) { twait_block->vtag_block[i].tv_sec_at_expire = 0; twait_block->vtag_block[i].v_tag = 0; twait_block->vtag_block[i].lport = 0; twait_block->vtag_block[i].rport = 0; found = 1; break; } } if (found) break; } } int sctp_is_in_timewait(uint32_t tag, uint16_t lport, uint16_t rport) { struct sctpvtaghead *chain; struct sctp_tagblock *twait_block; int found = 0; int i; SCTP_INP_INFO_WLOCK(); chain = &SCTP_BASE_INFO(vtag_timewait)[(tag % SCTP_STACK_VTAG_HASH_SIZE)]; LIST_FOREACH(twait_block, chain, sctp_nxt_tagblock) { for (i = 0; i < SCTP_NUMBER_IN_VTAG_BLOCK; i++) { if ((twait_block->vtag_block[i].v_tag == tag) && (twait_block->vtag_block[i].lport == lport) && (twait_block->vtag_block[i].rport == rport)) { found = 1; break; } } if (found) break; } SCTP_INP_INFO_WUNLOCK(); return (found); } void sctp_add_vtag_to_timewait(uint32_t tag, uint32_t time, uint16_t lport, uint16_t rport) { struct sctpvtaghead *chain; struct sctp_tagblock *twait_block; struct timeval now; int set, i; if (time == 0) { /* Its disabled */ return; } (void)SCTP_GETTIME_TIMEVAL(&now); chain = &SCTP_BASE_INFO(vtag_timewait)[(tag % SCTP_STACK_VTAG_HASH_SIZE)]; set = 0; LIST_FOREACH(twait_block, chain, sctp_nxt_tagblock) { /* Block(s) present, lets find space, and expire on the fly */ for (i = 0; i < SCTP_NUMBER_IN_VTAG_BLOCK; i++) { if ((twait_block->vtag_block[i].v_tag == 0) && !set) { twait_block->vtag_block[i].tv_sec_at_expire = now.tv_sec + time; twait_block->vtag_block[i].v_tag = tag; twait_block->vtag_block[i].lport = lport; twait_block->vtag_block[i].rport = rport; set = 1; } else if ((twait_block->vtag_block[i].v_tag) && ((long)twait_block->vtag_block[i].tv_sec_at_expire < now.tv_sec)) { /* Audit expires this guy */ twait_block->vtag_block[i].tv_sec_at_expire = 0; twait_block->vtag_block[i].v_tag = 0; twait_block->vtag_block[i].lport = 0; twait_block->vtag_block[i].rport = 0; if (set == 0) { /* Reuse it for my new tag */ twait_block->vtag_block[i].tv_sec_at_expire = now.tv_sec + time; twait_block->vtag_block[i].v_tag = tag; twait_block->vtag_block[i].lport = lport; twait_block->vtag_block[i].rport = rport; set = 1; } } } if (set) { /* * We only do up to the block where we can place our * tag for audits */ break; } } /* Need to add a new block to chain */ if (!set) { SCTP_MALLOC(twait_block, struct sctp_tagblock *, sizeof(struct sctp_tagblock), SCTP_M_TIMW); if (twait_block == NULL) { return; } memset(twait_block, 0, sizeof(struct sctp_tagblock)); LIST_INSERT_HEAD(chain, twait_block, sctp_nxt_tagblock); twait_block->vtag_block[0].tv_sec_at_expire = now.tv_sec + time; twait_block->vtag_block[0].v_tag = tag; twait_block->vtag_block[0].lport = lport; twait_block->vtag_block[0].rport = rport; } } void sctp_clean_up_stream(struct sctp_tcb *stcb, struct sctp_readhead *rh) { struct sctp_tmit_chunk *chk, *nchk; struct sctp_queued_to_read *control, *ncontrol; TAILQ_FOREACH_SAFE(control, rh, next_instrm, ncontrol) { TAILQ_REMOVE(rh, control, next_instrm); control->on_strm_q = 0; if (control->on_read_q == 0) { sctp_free_remote_addr(control->whoFrom); if (control->data) { sctp_m_freem(control->data); control->data = NULL; } } /* Reassembly free? */ TAILQ_FOREACH_SAFE(chk, &control->reasm, sctp_next, nchk) { TAILQ_REMOVE(&control->reasm, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); sctp_free_remote_addr(chk->whoTo); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); /* sa_ignore FREED_MEMORY */ } /* * We don't free the address here since all the net's were * freed above. */ if (control->on_read_q == 0) { sctp_free_a_readq(stcb, control); } } } /*- * Free the association after un-hashing the remote port. This * function ALWAYS returns holding NO LOCK on the stcb. It DOES * expect that the input to this function IS a locked TCB. * It will return 0, if it did NOT destroy the association (instead * it unlocks it. It will return NON-zero if it either destroyed the * association OR the association is already destroyed. */ int sctp_free_assoc(struct sctp_inpcb *inp, struct sctp_tcb *stcb, int from_inpcbfree, int from_location) { int i; struct sctp_association *asoc; struct sctp_nets *net, *nnet; struct sctp_laddr *laddr, *naddr; struct sctp_tmit_chunk *chk, *nchk; struct sctp_asconf_addr *aparam, *naparam; struct sctp_asconf_ack *aack, *naack; struct sctp_stream_reset_list *strrst, *nstrrst; struct sctp_queued_to_read *sq, *nsq; struct sctp_stream_queue_pending *sp, *nsp; sctp_sharedkey_t *shared_key, *nshared_key; struct socket *so; /* first, lets purge the entry from the hash table. */ #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, stcb, 6); #endif if (stcb->asoc.state == 0) { #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 7); #endif /* there is no asoc, really TSNH :-0 */ return (1); } if (stcb->asoc.alternate) { sctp_free_remote_addr(stcb->asoc.alternate); stcb->asoc.alternate = NULL; } /* TEMP CODE */ if (stcb->freed_from_where == 0) { /* Only record the first place free happened from */ stcb->freed_from_where = from_location; } /* TEMP CODE */ asoc = &stcb->asoc; if ((inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE)) /* nothing around */ so = NULL; else so = inp->sctp_socket; /* * We used timer based freeing if a reader or writer is in the way. * So we first check if we are actually being called from a timer, * if so we abort early if a reader or writer is still in the way. */ if ((stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) && (from_inpcbfree == SCTP_NORMAL_PROC)) { /* * is it the timer driving us? if so are the reader/writers * gone? */ if (stcb->asoc.refcnt) { /* nope, reader or writer in the way */ sctp_timer_start(SCTP_TIMER_TYPE_ASOCKILL, inp, stcb, NULL); /* no asoc destroyed */ SCTP_TCB_UNLOCK(stcb); #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, stcb, 8); #endif return (0); } } /* Now clean up any other timers */ sctp_stop_association_timers(stcb, false); /* Now the read queue needs to be cleaned up (only once) */ if ((stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) == 0) { SCTP_ADD_SUBSTATE(stcb, SCTP_STATE_ABOUT_TO_BE_FREED); SCTP_INP_READ_LOCK(inp); TAILQ_FOREACH(sq, &inp->read_queue, next) { if (sq->stcb == stcb) { sq->do_not_ref_stcb = 1; sq->sinfo_cumtsn = stcb->asoc.cumulative_tsn; /* * If there is no end, there never will be * now. */ if (sq->end_added == 0) { /* Held for PD-API clear that. */ sq->pdapi_aborted = 1; sq->held_length = 0; if (sctp_stcb_is_feature_on(inp, stcb, SCTP_PCB_FLAGS_PDAPIEVNT) && (so != NULL)) { /* * Need to add a PD-API * aborted indication. * Setting the control_pdapi * assures that it will be * added right after this * msg. */ uint32_t strseq; stcb->asoc.control_pdapi = sq; strseq = (sq->sinfo_stream << 16) | (sq->mid & 0x0000ffff); sctp_ulp_notify(SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION, stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)&strseq, SCTP_SO_LOCKED); stcb->asoc.control_pdapi = NULL; } } /* Add an end to wake them */ sq->end_added = 1; } } SCTP_INP_READ_UNLOCK(inp); if (stcb->block_entry) { SCTP_LTRACE_ERR_RET(inp, stcb, NULL, SCTP_FROM_SCTP_PCB, ECONNRESET); stcb->block_entry->error = ECONNRESET; stcb->block_entry = NULL; } } if ((stcb->asoc.refcnt) || (stcb->asoc.state & SCTP_STATE_IN_ACCEPT_QUEUE)) { /* * Someone holds a reference OR the socket is unaccepted * yet. */ if ((stcb->asoc.refcnt) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE)) { SCTP_CLEAR_SUBSTATE(stcb, SCTP_STATE_IN_ACCEPT_QUEUE); sctp_timer_start(SCTP_TIMER_TYPE_ASOCKILL, inp, stcb, NULL); } SCTP_TCB_UNLOCK(stcb); if ((inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE)) /* nothing around */ so = NULL; if (so) { /* Wake any reader/writers */ sctp_sorwakeup(inp, so); sctp_sowwakeup(inp, so); } #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, stcb, 9); #endif /* no asoc destroyed */ return (0); } #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, stcb, 10); #endif /* * When I reach here, no others want to kill the assoc yet.. and I * own the lock. Now its possible an abort comes in when I do the * lock exchange below to grab all the locks to do the final take * out. to prevent this we increment the count, which will start a * timer and blow out above thus assuring us that we hold exclusive * killing of the asoc. Note that after getting back the TCB lock we * will go ahead and increment the counter back up and stop any * timer a passing stranger may have started :-S */ if (from_inpcbfree == SCTP_NORMAL_PROC) { atomic_add_int(&stcb->asoc.refcnt, 1); SCTP_TCB_UNLOCK(stcb); SCTP_INP_INFO_WLOCK(); SCTP_INP_WLOCK(inp); SCTP_TCB_LOCK(stcb); } /* Double check the GONE flag */ if ((inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) || (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE)) /* nothing around */ so = NULL; if ((inp->sctp_flags & SCTP_PCB_FLAGS_TCPTYPE) || (inp->sctp_flags & SCTP_PCB_FLAGS_IN_TCPPOOL)) { /* * For TCP type we need special handling when we are * connected. We also include the peel'ed off ones to. */ if (inp->sctp_flags & SCTP_PCB_FLAGS_CONNECTED) { inp->sctp_flags &= ~SCTP_PCB_FLAGS_CONNECTED; inp->sctp_flags |= SCTP_PCB_FLAGS_WAS_CONNECTED; if (so) { SOCKBUF_LOCK(&so->so_rcv); so->so_state &= ~(SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING | SS_ISCONNECTED); so->so_state |= SS_ISDISCONNECTED; socantrcvmore_locked(so); socantsendmore(so); sctp_sowwakeup(inp, so); sctp_sorwakeup(inp, so); SCTP_SOWAKEUP(so); } } } /* * Make it invalid too, that way if its about to run it will abort * and return. */ /* re-increment the lock */ if (from_inpcbfree == SCTP_NORMAL_PROC) { atomic_add_int(&stcb->asoc.refcnt, -1); } if (stcb->asoc.refcnt) { SCTP_CLEAR_SUBSTATE(stcb, SCTP_STATE_IN_ACCEPT_QUEUE); sctp_timer_start(SCTP_TIMER_TYPE_ASOCKILL, inp, stcb, NULL); if (from_inpcbfree == SCTP_NORMAL_PROC) { SCTP_INP_INFO_WUNLOCK(); SCTP_INP_WUNLOCK(inp); } SCTP_TCB_UNLOCK(stcb); return (0); } asoc->state = 0; if (inp->sctp_tcbhash) { LIST_REMOVE(stcb, sctp_tcbhash); } if (stcb->asoc.in_asocid_hash) { LIST_REMOVE(stcb, sctp_tcbasocidhash); } /* Now lets remove it from the list of ALL associations in the EP */ LIST_REMOVE(stcb, sctp_tcblist); if (from_inpcbfree == SCTP_NORMAL_PROC) { SCTP_INP_INCR_REF(inp); SCTP_INP_WUNLOCK(inp); } /* pull from vtag hash */ LIST_REMOVE(stcb, sctp_asocs); sctp_add_vtag_to_timewait(asoc->my_vtag, SCTP_BASE_SYSCTL(sctp_vtag_time_wait), inp->sctp_lport, stcb->rport); /* * Now restop the timers to be sure this is paranoia at is finest! */ sctp_stop_association_timers(stcb, true); /* * The chunk lists and such SHOULD be empty but we check them just * in case. */ /* anything on the wheel needs to be removed */ SCTP_TCB_SEND_LOCK(stcb); for (i = 0; i < asoc->streamoutcnt; i++) { struct sctp_stream_out *outs; outs = &asoc->strmout[i]; /* now clean up any chunks here */ TAILQ_FOREACH_SAFE(sp, &outs->outqueue, next, nsp) { atomic_subtract_int(&asoc->stream_queue_cnt, 1); TAILQ_REMOVE(&outs->outqueue, sp, next); stcb->asoc.ss_functions.sctp_ss_remove_from_stream(stcb, asoc, outs, sp, 1); sctp_free_spbufspace(stcb, asoc, sp); if (sp->data) { if (so) { /* Still an open socket - report */ sctp_ulp_notify(SCTP_NOTIFY_SPECIAL_SP_FAIL, stcb, 0, (void *)sp, SCTP_SO_LOCKED); } if (sp->data) { sctp_m_freem(sp->data); sp->data = NULL; sp->tail_mbuf = NULL; sp->length = 0; } } if (sp->net) { sctp_free_remote_addr(sp->net); sp->net = NULL; } sctp_free_a_strmoq(stcb, sp, SCTP_SO_LOCKED); } } SCTP_TCB_SEND_UNLOCK(stcb); /* sa_ignore FREED_MEMORY */ TAILQ_FOREACH_SAFE(strrst, &asoc->resetHead, next_resp, nstrrst) { TAILQ_REMOVE(&asoc->resetHead, strrst, next_resp); SCTP_FREE(strrst, SCTP_M_STRESET); } TAILQ_FOREACH_SAFE(sq, &asoc->pending_reply_queue, next, nsq) { TAILQ_REMOVE(&asoc->pending_reply_queue, sq, next); if (sq->data) { sctp_m_freem(sq->data); sq->data = NULL; } sctp_free_remote_addr(sq->whoFrom); sq->whoFrom = NULL; sq->stcb = NULL; /* Free the ctl entry */ sctp_free_a_readq(stcb, sq); /* sa_ignore FREED_MEMORY */ } TAILQ_FOREACH_SAFE(chk, &asoc->free_chunks, sctp_next, nchk) { TAILQ_REMOVE(&asoc->free_chunks, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); atomic_subtract_int(&SCTP_BASE_INFO(ipi_free_chunks), 1); asoc->free_chunk_cnt--; /* sa_ignore FREED_MEMORY */ } /* pending send queue SHOULD be empty */ TAILQ_FOREACH_SAFE(chk, &asoc->send_queue, sctp_next, nchk) { if (asoc->strmout[chk->rec.data.sid].chunks_on_queues > 0) { asoc->strmout[chk->rec.data.sid].chunks_on_queues--; #ifdef INVARIANTS } else { panic("No chunks on the queues for sid %u.", chk->rec.data.sid); #endif } TAILQ_REMOVE(&asoc->send_queue, chk, sctp_next); if (chk->data) { if (so) { /* Still a socket? */ sctp_ulp_notify(SCTP_NOTIFY_UNSENT_DG_FAIL, stcb, 0, chk, SCTP_SO_LOCKED); } if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); if (chk->whoTo) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = NULL; } SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); /* sa_ignore FREED_MEMORY */ } /* sent queue SHOULD be empty */ TAILQ_FOREACH_SAFE(chk, &asoc->sent_queue, sctp_next, nchk) { if (chk->sent != SCTP_DATAGRAM_NR_ACKED) { if (asoc->strmout[chk->rec.data.sid].chunks_on_queues > 0) { asoc->strmout[chk->rec.data.sid].chunks_on_queues--; #ifdef INVARIANTS } else { panic("No chunks on the queues for sid %u.", chk->rec.data.sid); #endif } } TAILQ_REMOVE(&asoc->sent_queue, chk, sctp_next); if (chk->data) { if (so) { /* Still a socket? */ sctp_ulp_notify(SCTP_NOTIFY_SENT_DG_FAIL, stcb, 0, chk, SCTP_SO_LOCKED); } if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); sctp_free_remote_addr(chk->whoTo); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); /* sa_ignore FREED_MEMORY */ } #ifdef INVARIANTS for (i = 0; i < stcb->asoc.streamoutcnt; i++) { if (stcb->asoc.strmout[i].chunks_on_queues > 0) { panic("%u chunks left for stream %u.", stcb->asoc.strmout[i].chunks_on_queues, i); } } #endif /* control queue MAY not be empty */ TAILQ_FOREACH_SAFE(chk, &asoc->control_send_queue, sctp_next, nchk) { TAILQ_REMOVE(&asoc->control_send_queue, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); sctp_free_remote_addr(chk->whoTo); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); /* sa_ignore FREED_MEMORY */ } /* ASCONF queue MAY not be empty */ TAILQ_FOREACH_SAFE(chk, &asoc->asconf_send_queue, sctp_next, nchk) { TAILQ_REMOVE(&asoc->asconf_send_queue, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } if (chk->holds_key_ref) sctp_auth_key_release(stcb, chk->auth_keyid, SCTP_SO_LOCKED); sctp_free_remote_addr(chk->whoTo); SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), chk); SCTP_DECR_CHK_COUNT(); /* sa_ignore FREED_MEMORY */ } if (asoc->mapping_array) { SCTP_FREE(asoc->mapping_array, SCTP_M_MAP); asoc->mapping_array = NULL; } if (asoc->nr_mapping_array) { SCTP_FREE(asoc->nr_mapping_array, SCTP_M_MAP); asoc->nr_mapping_array = NULL; } /* the stream outs */ if (asoc->strmout) { SCTP_FREE(asoc->strmout, SCTP_M_STRMO); asoc->strmout = NULL; } asoc->strm_realoutsize = asoc->streamoutcnt = 0; if (asoc->strmin) { for (i = 0; i < asoc->streamincnt; i++) { sctp_clean_up_stream(stcb, &asoc->strmin[i].inqueue); sctp_clean_up_stream(stcb, &asoc->strmin[i].uno_inqueue); } SCTP_FREE(asoc->strmin, SCTP_M_STRMI); asoc->strmin = NULL; } asoc->streamincnt = 0; TAILQ_FOREACH_SAFE(net, &asoc->nets, sctp_next, nnet) { #ifdef INVARIANTS if (SCTP_BASE_INFO(ipi_count_raddr) == 0) { panic("no net's left alloc'ed, or list points to itself"); } #endif TAILQ_REMOVE(&asoc->nets, net, sctp_next); sctp_free_remote_addr(net); } LIST_FOREACH_SAFE(laddr, &asoc->sctp_restricted_addrs, sctp_nxt_addr, naddr) { /* sa_ignore FREED_MEMORY */ sctp_remove_laddr(laddr); } /* pending asconf (address) parameters */ TAILQ_FOREACH_SAFE(aparam, &asoc->asconf_queue, next, naparam) { /* sa_ignore FREED_MEMORY */ TAILQ_REMOVE(&asoc->asconf_queue, aparam, next); SCTP_FREE(aparam, SCTP_M_ASC_ADDR); } TAILQ_FOREACH_SAFE(aack, &asoc->asconf_ack_sent, next, naack) { /* sa_ignore FREED_MEMORY */ TAILQ_REMOVE(&asoc->asconf_ack_sent, aack, next); if (aack->data != NULL) { sctp_m_freem(aack->data); } SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asconf_ack), aack); } /* clean up auth stuff */ if (asoc->local_hmacs) sctp_free_hmaclist(asoc->local_hmacs); if (asoc->peer_hmacs) sctp_free_hmaclist(asoc->peer_hmacs); if (asoc->local_auth_chunks) sctp_free_chunklist(asoc->local_auth_chunks); if (asoc->peer_auth_chunks) sctp_free_chunklist(asoc->peer_auth_chunks); sctp_free_authinfo(&asoc->authinfo); LIST_FOREACH_SAFE(shared_key, &asoc->shared_keys, next, nshared_key) { LIST_REMOVE(shared_key, next); sctp_free_sharedkey(shared_key); /* sa_ignore FREED_MEMORY */ } /* Insert new items here :> */ /* Get rid of LOCK */ SCTP_TCB_UNLOCK(stcb); SCTP_TCB_LOCK_DESTROY(stcb); SCTP_TCB_SEND_LOCK_DESTROY(stcb); if (from_inpcbfree == SCTP_NORMAL_PROC) { SCTP_INP_INFO_WUNLOCK(); SCTP_INP_RLOCK(inp); } #ifdef SCTP_TRACK_FREED_ASOCS if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) { /* now clean up the tasoc itself */ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), stcb); SCTP_DECR_ASOC_COUNT(); } else { LIST_INSERT_HEAD(&inp->sctp_asoc_free_list, stcb, sctp_tcblist); } #else SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_asoc), stcb); SCTP_DECR_ASOC_COUNT(); #endif if (from_inpcbfree == SCTP_NORMAL_PROC) { if (inp->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) { /* * If its NOT the inp_free calling us AND sctp_close * as been called, we call back... */ SCTP_INP_RUNLOCK(inp); /* * This will start the kill timer (if we are the * last one) since we hold an increment yet. But * this is the only safe way to do this since * otherwise if the socket closes at the same time * we are here we might collide in the cleanup. */ sctp_inpcb_free(inp, SCTP_FREE_SHOULD_USE_GRACEFUL_CLOSE, SCTP_CALLED_DIRECTLY_NOCMPSET); SCTP_INP_DECR_REF(inp); goto out_of; } else { /* The socket is still open. */ SCTP_INP_DECR_REF(inp); } } if (from_inpcbfree == SCTP_NORMAL_PROC) { SCTP_INP_RUNLOCK(inp); } out_of: /* destroyed the asoc */ #ifdef SCTP_LOG_CLOSING sctp_log_closing(inp, NULL, 11); #endif return (1); } /* * determine if a destination is "reachable" based upon the addresses bound * to the current endpoint (e.g. only v4 or v6 currently bound) */ /* * FIX: if we allow assoc-level bindx(), then this needs to be fixed to use * assoc level v4/v6 flags, as the assoc *may* not have the same address * types bound as its endpoint */ int sctp_destination_is_reachable(struct sctp_tcb *stcb, struct sockaddr *destaddr) { struct sctp_inpcb *inp; int answer; /* * No locks here, the TCB, in all cases is already locked and an * assoc is up. There is either a INP lock by the caller applied (in * asconf case when deleting an address) or NOT in the HB case, * however if HB then the INP increment is up and the INP will not * be removed (on top of the fact that we have a TCB lock). So we * only want to read the sctp_flags, which is either bound-all or * not.. no protection needed since once an assoc is up you can't be * changing your binding. */ inp = stcb->sctp_ep; if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { /* if bound all, destination is not restricted */ /* * RRS: Question during lock work: Is this correct? If you * are bound-all you still might need to obey the V4--V6 * flags??? IMO this bound-all stuff needs to be removed! */ return (1); } /* NOTE: all "scope" checks are done when local addresses are added */ switch (destaddr->sa_family) { #ifdef INET6 case AF_INET6: answer = inp->ip_inp.inp.inp_vflag & INP_IPV6; break; #endif #ifdef INET case AF_INET: answer = inp->ip_inp.inp.inp_vflag & INP_IPV4; break; #endif default: /* invalid family, so it's unreachable */ answer = 0; break; } return (answer); } /* * update the inp_vflags on an endpoint */ static void sctp_update_ep_vflag(struct sctp_inpcb *inp) { struct sctp_laddr *laddr; /* first clear the flag */ inp->ip_inp.inp.inp_vflag = 0; /* set the flag based on addresses on the ep list */ LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == NULL) { SCTPDBG(SCTP_DEBUG_PCB1, "%s: NULL ifa\n", __func__); continue; } if (laddr->ifa->localifa_flags & SCTP_BEING_DELETED) { continue; } switch (laddr->ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: inp->ip_inp.inp.inp_vflag |= INP_IPV6; break; #endif #ifdef INET case AF_INET: inp->ip_inp.inp.inp_vflag |= INP_IPV4; break; #endif default: break; } } } /* * Add the address to the endpoint local address list There is nothing to be * done if we are bound to all addresses */ void sctp_add_local_addr_ep(struct sctp_inpcb *inp, struct sctp_ifa *ifa, uint32_t action) { struct sctp_laddr *laddr; struct sctp_tcb *stcb; int fnd, error = 0; fnd = 0; if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { /* You are already bound to all. You have it already */ return; } #ifdef INET6 if (ifa->address.sa.sa_family == AF_INET6) { if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { /* Can't bind a non-useable addr. */ return; } } #endif /* first, is it already present? */ LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == ifa) { fnd = 1; break; } } if (fnd == 0) { /* Not in the ep list */ error = sctp_insert_laddr(&inp->sctp_addr_list, ifa, action); if (error != 0) return; inp->laddr_count++; /* update inp_vflag flags */ switch (ifa->address.sa.sa_family) { #ifdef INET6 case AF_INET6: inp->ip_inp.inp.inp_vflag |= INP_IPV6; break; #endif #ifdef INET case AF_INET: inp->ip_inp.inp.inp_vflag |= INP_IPV4; break; #endif default: break; } LIST_FOREACH(stcb, &inp->sctp_asoc_list, sctp_tcblist) { sctp_add_local_addr_restricted(stcb, ifa); } } return; } /* * select a new (hopefully reachable) destination net (should only be used * when we deleted an ep addr that is the only usable source address to reach * the destination net) */ static void sctp_select_primary_destination(struct sctp_tcb *stcb) { struct sctp_nets *net; TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { /* for now, we'll just pick the first reachable one we find */ if (net->dest_state & SCTP_ADDR_UNCONFIRMED) continue; if (sctp_destination_is_reachable(stcb, (struct sockaddr *)&net->ro._l_addr)) { /* found a reachable destination */ stcb->asoc.primary_destination = net; } } /* I can't there from here! ...we're gonna die shortly... */ } /* * Delete the address from the endpoint local address list. There is nothing * to be done if we are bound to all addresses */ void sctp_del_local_addr_ep(struct sctp_inpcb *inp, struct sctp_ifa *ifa) { struct sctp_laddr *laddr; int fnd; fnd = 0; if (inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) { /* You are already bound to all. You have it already */ return; } LIST_FOREACH(laddr, &inp->sctp_addr_list, sctp_nxt_addr) { if (laddr->ifa == ifa) { fnd = 1; break; } } if (fnd && (inp->laddr_count < 2)) { /* can't delete unless there are at LEAST 2 addresses */ return; } if (fnd) { /* * clean up any use of this address go through our * associations and clear any last_used_address that match * this one for each assoc, see if a new primary_destination * is needed */ struct sctp_tcb *stcb; /* clean up "next_addr_touse" */ if (inp->next_addr_touse == laddr) /* delete this address */ inp->next_addr_touse = NULL; /* clean up "last_used_address" */ LIST_FOREACH(stcb, &inp->sctp_asoc_list, sctp_tcblist) { struct sctp_nets *net; SCTP_TCB_LOCK(stcb); if (stcb->asoc.last_used_address == laddr) /* delete this address */ stcb->asoc.last_used_address = NULL; /* * Now spin through all the nets and purge any ref * to laddr */ TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { if (net->ro._s_addr == laddr->ifa) { /* Yep, purge src address selected */ - sctp_rtentry_t *rt; /* delete this address if cached */ - rt = net->ro.ro_rt; - if (rt != NULL) { - RTFREE(rt); - net->ro.ro_rt = NULL; - } + RO_NHFREE(&net->ro); sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } } SCTP_TCB_UNLOCK(stcb); } /* for each tcb */ /* remove it from the ep list */ sctp_remove_laddr(laddr); inp->laddr_count--; /* update inp_vflag flags */ sctp_update_ep_vflag(inp); } return; } /* * Add the address to the TCB local address restricted list. * This is a "pending" address list (eg. addresses waiting for an * ASCONF-ACK response) and cannot be used as a valid source address. */ void sctp_add_local_addr_restricted(struct sctp_tcb *stcb, struct sctp_ifa *ifa) { struct sctp_laddr *laddr; struct sctpladdr *list; /* * Assumes TCB is locked.. and possibly the INP. May need to * confirm/fix that if we need it and is not the case. */ list = &stcb->asoc.sctp_restricted_addrs; #ifdef INET6 if (ifa->address.sa.sa_family == AF_INET6) { if (ifa->localifa_flags & SCTP_ADDR_IFA_UNUSEABLE) { /* Can't bind a non-existent addr. */ return; } } #endif /* does the address already exist? */ LIST_FOREACH(laddr, list, sctp_nxt_addr) { if (laddr->ifa == ifa) { return; } } /* add to the list */ (void)sctp_insert_laddr(list, ifa, 0); return; } /* * Remove a local address from the TCB local address restricted list */ void sctp_del_local_addr_restricted(struct sctp_tcb *stcb, struct sctp_ifa *ifa) { struct sctp_inpcb *inp; struct sctp_laddr *laddr; /* * This is called by asconf work. It is assumed that a) The TCB is * locked and b) The INP is locked. This is true in as much as I can * trace through the entry asconf code where I did these locks. * Again, the ASCONF code is a bit different in that it does lock * the INP during its work often times. This must be since we don't * want other proc's looking up things while what they are looking * up is changing :-D */ inp = stcb->sctp_ep; /* if subset bound and don't allow ASCONF's, can't delete last */ if (((inp->sctp_flags & SCTP_PCB_FLAGS_BOUNDALL) == 0) && sctp_is_feature_off(inp, SCTP_PCB_FLAGS_DO_ASCONF)) { if (stcb->sctp_ep->laddr_count < 2) { /* can't delete last address */ return; } } LIST_FOREACH(laddr, &stcb->asoc.sctp_restricted_addrs, sctp_nxt_addr) { /* remove the address if it exists */ if (laddr->ifa == NULL) continue; if (laddr->ifa == ifa) { sctp_remove_laddr(laddr); return; } } /* address not found! */ return; } /* * Temporarily remove for __APPLE__ until we use the Tiger equivalents */ /* sysctl */ static int sctp_max_number_of_assoc = SCTP_MAX_NUM_OF_ASOC; static int sctp_scale_up_for_address = SCTP_SCALE_FOR_ADDR; #if defined(__FreeBSD__) && defined(SCTP_MCORE_INPUT) && defined(SMP) struct sctp_mcore_ctrl *sctp_mcore_workers = NULL; int *sctp_cpuarry = NULL; void sctp_queue_to_mcore(struct mbuf *m, int off, int cpu_to_use) { /* Queue a packet to a processor for the specified core */ struct sctp_mcore_queue *qent; struct sctp_mcore_ctrl *wkq; int need_wake = 0; if (sctp_mcore_workers == NULL) { /* Something went way bad during setup */ sctp_input_with_port(m, off, 0); return; } SCTP_MALLOC(qent, struct sctp_mcore_queue *, (sizeof(struct sctp_mcore_queue)), SCTP_M_MCORE); if (qent == NULL) { /* This is trouble */ sctp_input_with_port(m, off, 0); return; } qent->vn = curvnet; qent->m = m; qent->off = off; qent->v6 = 0; wkq = &sctp_mcore_workers[cpu_to_use]; SCTP_MCORE_QLOCK(wkq); TAILQ_INSERT_TAIL(&wkq->que, qent, next); if (wkq->running == 0) { need_wake = 1; } SCTP_MCORE_QUNLOCK(wkq); if (need_wake) { wakeup(&wkq->running); } } static void sctp_mcore_thread(void *arg) { struct sctp_mcore_ctrl *wkq; struct sctp_mcore_queue *qent; wkq = (struct sctp_mcore_ctrl *)arg; struct mbuf *m; int off, v6; /* Wait for first tickle */ SCTP_MCORE_LOCK(wkq); wkq->running = 0; msleep(&wkq->running, &wkq->core_mtx, 0, "wait for pkt", 0); SCTP_MCORE_UNLOCK(wkq); /* Bind to our cpu */ thread_lock(curthread); sched_bind(curthread, wkq->cpuid); thread_unlock(curthread); /* Now lets start working */ SCTP_MCORE_LOCK(wkq); /* Now grab lock and go */ for (;;) { SCTP_MCORE_QLOCK(wkq); skip_sleep: wkq->running = 1; qent = TAILQ_FIRST(&wkq->que); if (qent) { TAILQ_REMOVE(&wkq->que, qent, next); SCTP_MCORE_QUNLOCK(wkq); CURVNET_SET(qent->vn); m = qent->m; off = qent->off; v6 = qent->v6; SCTP_FREE(qent, SCTP_M_MCORE); if (v6 == 0) { sctp_input_with_port(m, off, 0); } else { SCTP_PRINTF("V6 not yet supported\n"); sctp_m_freem(m); } CURVNET_RESTORE(); SCTP_MCORE_QLOCK(wkq); } wkq->running = 0; if (!TAILQ_EMPTY(&wkq->que)) { goto skip_sleep; } SCTP_MCORE_QUNLOCK(wkq); msleep(&wkq->running, &wkq->core_mtx, 0, "wait for pkt", 0); } } static void sctp_startup_mcore_threads(void) { int i, cpu; if (mp_ncpus == 1) return; if (sctp_mcore_workers != NULL) { /* * Already been here in some previous vnet? */ return; } SCTP_MALLOC(sctp_mcore_workers, struct sctp_mcore_ctrl *, ((mp_maxid + 1) * sizeof(struct sctp_mcore_ctrl)), SCTP_M_MCORE); if (sctp_mcore_workers == NULL) { /* TSNH I hope */ return; } memset(sctp_mcore_workers, 0, ((mp_maxid + 1) * sizeof(struct sctp_mcore_ctrl))); /* Init the structures */ for (i = 0; i <= mp_maxid; i++) { TAILQ_INIT(&sctp_mcore_workers[i].que); SCTP_MCORE_LOCK_INIT(&sctp_mcore_workers[i]); SCTP_MCORE_QLOCK_INIT(&sctp_mcore_workers[i]); sctp_mcore_workers[i].cpuid = i; } if (sctp_cpuarry == NULL) { SCTP_MALLOC(sctp_cpuarry, int *, (mp_ncpus * sizeof(int)), SCTP_M_MCORE); i = 0; CPU_FOREACH(cpu) { sctp_cpuarry[i] = cpu; i++; } } /* Now start them all */ CPU_FOREACH(cpu) { (void)kproc_create(sctp_mcore_thread, (void *)&sctp_mcore_workers[cpu], &sctp_mcore_workers[cpu].thread_proc, RFPROC, SCTP_KTHREAD_PAGES, SCTP_MCORE_NAME); } } #endif void sctp_pcb_init(void) { /* * SCTP initialization for the PCB structures should be called by * the sctp_init() function. */ int i; struct timeval tv; if (SCTP_BASE_VAR(sctp_pcb_initialized) != 0) { /* error I was called twice */ return; } SCTP_BASE_VAR(sctp_pcb_initialized) = 1; #if defined(SCTP_LOCAL_TRACE_BUF) memset(&SCTP_BASE_SYSCTL(sctp_log), 0, sizeof(struct sctp_log)); #endif #if defined(__FreeBSD__) && defined(SMP) && defined(SCTP_USE_PERCPU_STAT) SCTP_MALLOC(SCTP_BASE_STATS, struct sctpstat *, ((mp_maxid + 1) * sizeof(struct sctpstat)), SCTP_M_MCORE); #endif (void)SCTP_GETTIME_TIMEVAL(&tv); #if defined(__FreeBSD__) && defined(SMP) && defined(SCTP_USE_PERCPU_STAT) memset(SCTP_BASE_STATS, 0, sizeof(struct sctpstat) * (mp_maxid + 1)); SCTP_BASE_STATS[PCPU_GET(cpuid)].sctps_discontinuitytime.tv_sec = (uint32_t)tv.tv_sec; SCTP_BASE_STATS[PCPU_GET(cpuid)].sctps_discontinuitytime.tv_usec = (uint32_t)tv.tv_usec; #else memset(&SCTP_BASE_STATS, 0, sizeof(struct sctpstat)); SCTP_BASE_STAT(sctps_discontinuitytime).tv_sec = (uint32_t)tv.tv_sec; SCTP_BASE_STAT(sctps_discontinuitytime).tv_usec = (uint32_t)tv.tv_usec; #endif /* init the empty list of (All) Endpoints */ LIST_INIT(&SCTP_BASE_INFO(listhead)); /* init the hash table of endpoints */ TUNABLE_INT_FETCH("net.inet.sctp.tcbhashsize", &SCTP_BASE_SYSCTL(sctp_hashtblsize)); TUNABLE_INT_FETCH("net.inet.sctp.pcbhashsize", &SCTP_BASE_SYSCTL(sctp_pcbtblsize)); TUNABLE_INT_FETCH("net.inet.sctp.chunkscale", &SCTP_BASE_SYSCTL(sctp_chunkscale)); SCTP_BASE_INFO(sctp_asochash) = SCTP_HASH_INIT((SCTP_BASE_SYSCTL(sctp_hashtblsize) * 31), &SCTP_BASE_INFO(hashasocmark)); SCTP_BASE_INFO(sctp_ephash) = SCTP_HASH_INIT(SCTP_BASE_SYSCTL(sctp_hashtblsize), &SCTP_BASE_INFO(hashmark)); SCTP_BASE_INFO(sctp_tcpephash) = SCTP_HASH_INIT(SCTP_BASE_SYSCTL(sctp_hashtblsize), &SCTP_BASE_INFO(hashtcpmark)); SCTP_BASE_INFO(hashtblsize) = SCTP_BASE_SYSCTL(sctp_hashtblsize); SCTP_BASE_INFO(sctp_vrfhash) = SCTP_HASH_INIT(SCTP_SIZE_OF_VRF_HASH, &SCTP_BASE_INFO(hashvrfmark)); SCTP_BASE_INFO(vrf_ifn_hash) = SCTP_HASH_INIT(SCTP_VRF_IFN_HASH_SIZE, &SCTP_BASE_INFO(vrf_ifn_hashmark)); /* init the zones */ /* * FIX ME: Should check for NULL returns, but if it does fail we are * doomed to panic anyways... add later maybe. */ SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_ep), "sctp_ep", sizeof(struct sctp_inpcb), maxsockets); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_asoc), "sctp_asoc", sizeof(struct sctp_tcb), sctp_max_number_of_assoc); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_laddr), "sctp_laddr", sizeof(struct sctp_laddr), (sctp_max_number_of_assoc * sctp_scale_up_for_address)); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_net), "sctp_raddr", sizeof(struct sctp_nets), (sctp_max_number_of_assoc * sctp_scale_up_for_address)); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_chunk), "sctp_chunk", sizeof(struct sctp_tmit_chunk), (sctp_max_number_of_assoc * SCTP_BASE_SYSCTL(sctp_chunkscale))); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_readq), "sctp_readq", sizeof(struct sctp_queued_to_read), (sctp_max_number_of_assoc * SCTP_BASE_SYSCTL(sctp_chunkscale))); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_strmoq), "sctp_stream_msg_out", sizeof(struct sctp_stream_queue_pending), (sctp_max_number_of_assoc * SCTP_BASE_SYSCTL(sctp_chunkscale))); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_asconf), "sctp_asconf", sizeof(struct sctp_asconf), (sctp_max_number_of_assoc * SCTP_BASE_SYSCTL(sctp_chunkscale))); SCTP_ZONE_INIT(SCTP_BASE_INFO(ipi_zone_asconf_ack), "sctp_asconf_ack", sizeof(struct sctp_asconf_ack), (sctp_max_number_of_assoc * SCTP_BASE_SYSCTL(sctp_chunkscale))); /* Master Lock INIT for info structure */ SCTP_INP_INFO_LOCK_INIT(); SCTP_STATLOG_INIT_LOCK(); SCTP_IPI_COUNT_INIT(); SCTP_IPI_ADDR_INIT(); #ifdef SCTP_PACKET_LOGGING SCTP_IP_PKTLOG_INIT(); #endif LIST_INIT(&SCTP_BASE_INFO(addr_wq)); SCTP_WQ_ADDR_INIT(); /* not sure if we need all the counts */ SCTP_BASE_INFO(ipi_count_ep) = 0; /* assoc/tcb zone info */ SCTP_BASE_INFO(ipi_count_asoc) = 0; /* local addrlist zone info */ SCTP_BASE_INFO(ipi_count_laddr) = 0; /* remote addrlist zone info */ SCTP_BASE_INFO(ipi_count_raddr) = 0; /* chunk info */ SCTP_BASE_INFO(ipi_count_chunk) = 0; /* socket queue zone info */ SCTP_BASE_INFO(ipi_count_readq) = 0; /* stream out queue cont */ SCTP_BASE_INFO(ipi_count_strmoq) = 0; SCTP_BASE_INFO(ipi_free_strmoq) = 0; SCTP_BASE_INFO(ipi_free_chunks) = 0; SCTP_OS_TIMER_INIT(&SCTP_BASE_INFO(addr_wq_timer.timer)); /* Init the TIMEWAIT list */ for (i = 0; i < SCTP_STACK_VTAG_HASH_SIZE; i++) { LIST_INIT(&SCTP_BASE_INFO(vtag_timewait)[i]); } sctp_startup_iterator(); #if defined(__FreeBSD__) && defined(SCTP_MCORE_INPUT) && defined(SMP) sctp_startup_mcore_threads(); #endif /* * INIT the default VRF which for BSD is the only one, other O/S's * may have more. But initially they must start with one and then * add the VRF's as addresses are added. */ sctp_init_vrf_list(SCTP_DEFAULT_VRF); } /* * Assumes that the SCTP_BASE_INFO() lock is NOT held. */ void sctp_pcb_finish(void) { struct sctp_vrflist *vrf_bucket; struct sctp_vrf *vrf, *nvrf; struct sctp_ifn *ifn, *nifn; struct sctp_ifa *ifa, *nifa; struct sctpvtaghead *chain; struct sctp_tagblock *twait_block, *prev_twait_block; struct sctp_laddr *wi, *nwi; int i; struct sctp_iterator *it, *nit; if (SCTP_BASE_VAR(sctp_pcb_initialized) == 0) { SCTP_PRINTF("%s: race condition on teardown.\n", __func__); return; } SCTP_BASE_VAR(sctp_pcb_initialized) = 0; /* * In FreeBSD the iterator thread never exits but we do clean up. * The only way FreeBSD reaches here is if we have VRF's but we * still add the ifdef to make it compile on old versions. */ retry: SCTP_IPI_ITERATOR_WQ_LOCK(); /* * sctp_iterator_worker() might be working on an it entry without * holding the lock. We won't find it on the list either and * continue and free/destroy it. While holding the lock, spin, to * avoid the race condition as sctp_iterator_worker() will have to * wait to re-aquire the lock. */ if (sctp_it_ctl.iterator_running != 0 || sctp_it_ctl.cur_it != NULL) { SCTP_IPI_ITERATOR_WQ_UNLOCK(); SCTP_PRINTF("%s: Iterator running while we held the lock. Retry. " "cur_it=%p\n", __func__, sctp_it_ctl.cur_it); DELAY(10); goto retry; } TAILQ_FOREACH_SAFE(it, &sctp_it_ctl.iteratorhead, sctp_nxt_itr, nit) { if (it->vn != curvnet) { continue; } TAILQ_REMOVE(&sctp_it_ctl.iteratorhead, it, sctp_nxt_itr); if (it->function_atend != NULL) { (*it->function_atend) (it->pointer, it->val); } SCTP_FREE(it, SCTP_M_ITER); } SCTP_IPI_ITERATOR_WQ_UNLOCK(); SCTP_ITERATOR_LOCK(); if ((sctp_it_ctl.cur_it) && (sctp_it_ctl.cur_it->vn == curvnet)) { sctp_it_ctl.iterator_flags |= SCTP_ITERATOR_STOP_CUR_IT; } SCTP_ITERATOR_UNLOCK(); SCTP_OS_TIMER_STOP_DRAIN(&SCTP_BASE_INFO(addr_wq_timer.timer)); SCTP_WQ_ADDR_LOCK(); LIST_FOREACH_SAFE(wi, &SCTP_BASE_INFO(addr_wq), sctp_nxt_addr, nwi) { LIST_REMOVE(wi, sctp_nxt_addr); SCTP_DECR_LADDR_COUNT(); if (wi->action == SCTP_DEL_IP_ADDRESS) { SCTP_FREE(wi->ifa, SCTP_M_IFA); } SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_laddr), wi); } SCTP_WQ_ADDR_UNLOCK(); /* * free the vrf/ifn/ifa lists and hashes (be sure address monitor is * destroyed first). */ vrf_bucket = &SCTP_BASE_INFO(sctp_vrfhash)[(SCTP_DEFAULT_VRFID & SCTP_BASE_INFO(hashvrfmark))]; LIST_FOREACH_SAFE(vrf, vrf_bucket, next_vrf, nvrf) { LIST_FOREACH_SAFE(ifn, &vrf->ifnlist, next_ifn, nifn) { LIST_FOREACH_SAFE(ifa, &ifn->ifalist, next_ifa, nifa) { /* free the ifa */ LIST_REMOVE(ifa, next_bucket); LIST_REMOVE(ifa, next_ifa); SCTP_FREE(ifa, SCTP_M_IFA); } /* free the ifn */ LIST_REMOVE(ifn, next_bucket); LIST_REMOVE(ifn, next_ifn); SCTP_FREE(ifn, SCTP_M_IFN); } SCTP_HASH_FREE(vrf->vrf_addr_hash, vrf->vrf_addr_hashmark); /* free the vrf */ LIST_REMOVE(vrf, next_vrf); SCTP_FREE(vrf, SCTP_M_VRF); } /* free the vrf hashes */ SCTP_HASH_FREE(SCTP_BASE_INFO(sctp_vrfhash), SCTP_BASE_INFO(hashvrfmark)); SCTP_HASH_FREE(SCTP_BASE_INFO(vrf_ifn_hash), SCTP_BASE_INFO(vrf_ifn_hashmark)); /* * free the TIMEWAIT list elements malloc'd in the function * sctp_add_vtag_to_timewait()... */ for (i = 0; i < SCTP_STACK_VTAG_HASH_SIZE; i++) { chain = &SCTP_BASE_INFO(vtag_timewait)[i]; if (!LIST_EMPTY(chain)) { prev_twait_block = NULL; LIST_FOREACH(twait_block, chain, sctp_nxt_tagblock) { if (prev_twait_block) { SCTP_FREE(prev_twait_block, SCTP_M_TIMW); } prev_twait_block = twait_block; } SCTP_FREE(prev_twait_block, SCTP_M_TIMW); } } /* free the locks and mutexes */ #ifdef SCTP_PACKET_LOGGING SCTP_IP_PKTLOG_DESTROY(); #endif SCTP_IPI_ADDR_DESTROY(); SCTP_STATLOG_DESTROY(); SCTP_INP_INFO_LOCK_DESTROY(); SCTP_WQ_ADDR_DESTROY(); /* Get rid of other stuff too. */ if (SCTP_BASE_INFO(sctp_asochash) != NULL) SCTP_HASH_FREE(SCTP_BASE_INFO(sctp_asochash), SCTP_BASE_INFO(hashasocmark)); if (SCTP_BASE_INFO(sctp_ephash) != NULL) SCTP_HASH_FREE(SCTP_BASE_INFO(sctp_ephash), SCTP_BASE_INFO(hashmark)); if (SCTP_BASE_INFO(sctp_tcpephash) != NULL) SCTP_HASH_FREE(SCTP_BASE_INFO(sctp_tcpephash), SCTP_BASE_INFO(hashtcpmark)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_ep)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_asoc)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_laddr)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_net)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_chunk)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_readq)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_strmoq)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_asconf)); SCTP_ZONE_DESTROY(SCTP_BASE_INFO(ipi_zone_asconf_ack)); #if defined(__FreeBSD__) && defined(SMP) && defined(SCTP_USE_PERCPU_STAT) SCTP_FREE(SCTP_BASE_STATS, SCTP_M_MCORE); #endif } int sctp_load_addresses_from_init(struct sctp_tcb *stcb, struct mbuf *m, int offset, int limit, struct sockaddr *src, struct sockaddr *dst, struct sockaddr *altsa, uint16_t port) { /* * grub through the INIT pulling addresses and loading them to the * nets structure in the asoc. The from address in the mbuf should * also be loaded (if it is not already). This routine can be called * with either INIT or INIT-ACK's as long as the m points to the IP * packet and the offset points to the beginning of the parameters. */ struct sctp_inpcb *inp; struct sctp_nets *net, *nnet, *net_tmp; struct sctp_paramhdr *phdr, param_buf; struct sctp_tcb *stcb_tmp; uint16_t ptype, plen; struct sockaddr *sa; uint8_t random_store[SCTP_PARAM_BUFFER_SIZE]; struct sctp_auth_random *p_random = NULL; uint16_t random_len = 0; uint8_t hmacs_store[SCTP_PARAM_BUFFER_SIZE]; struct sctp_auth_hmac_algo *hmacs = NULL; uint16_t hmacs_len = 0; uint8_t saw_asconf = 0; uint8_t saw_asconf_ack = 0; uint8_t chunks_store[SCTP_PARAM_BUFFER_SIZE]; struct sctp_auth_chunk_list *chunks = NULL; uint16_t num_chunks = 0; sctp_key_t *new_key; uint32_t keylen; int got_random = 0, got_hmacs = 0, got_chklist = 0; uint8_t peer_supports_ecn; uint8_t peer_supports_prsctp; uint8_t peer_supports_auth; uint8_t peer_supports_asconf; uint8_t peer_supports_asconf_ack; uint8_t peer_supports_reconfig; uint8_t peer_supports_nrsack; uint8_t peer_supports_pktdrop; uint8_t peer_supports_idata; #ifdef INET struct sockaddr_in sin; #endif #ifdef INET6 struct sockaddr_in6 sin6; #endif /* First get the destination address setup too. */ #ifdef INET memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); sin.sin_port = stcb->rport; #endif #ifdef INET6 memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_port = stcb->rport; #endif if (altsa) { sa = altsa; } else { sa = src; } peer_supports_idata = 0; peer_supports_ecn = 0; peer_supports_prsctp = 0; peer_supports_auth = 0; peer_supports_asconf = 0; peer_supports_reconfig = 0; peer_supports_nrsack = 0; peer_supports_pktdrop = 0; TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) { /* mark all addresses that we have currently on the list */ net->dest_state |= SCTP_ADDR_NOT_IN_ASSOC; } /* does the source address already exist? if so skip it */ inp = stcb->sctp_ep; atomic_add_int(&stcb->asoc.refcnt, 1); stcb_tmp = sctp_findassociation_ep_addr(&inp, sa, &net_tmp, dst, stcb); atomic_add_int(&stcb->asoc.refcnt, -1); if ((stcb_tmp == NULL && inp == stcb->sctp_ep) || inp == NULL) { /* we must add the source address */ /* no scope set here since we have a tcb already. */ switch (sa->sa_family) { #ifdef INET case AF_INET: if (stcb->asoc.scope.ipv4_addr_legal) { if (sctp_add_remote_addr(stcb, sa, NULL, port, SCTP_DONOT_SETSCOPE, SCTP_LOAD_ADDR_2)) { return (-1); } } break; #endif #ifdef INET6 case AF_INET6: if (stcb->asoc.scope.ipv6_addr_legal) { if (sctp_add_remote_addr(stcb, sa, NULL, port, SCTP_DONOT_SETSCOPE, SCTP_LOAD_ADDR_3)) { return (-2); } } break; #endif default: break; } } else { if (net_tmp != NULL && stcb_tmp == stcb) { net_tmp->dest_state &= ~SCTP_ADDR_NOT_IN_ASSOC; } else if (stcb_tmp != stcb) { /* It belongs to another association? */ if (stcb_tmp) SCTP_TCB_UNLOCK(stcb_tmp); return (-3); } } if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-4); } /* now we must go through each of the params. */ phdr = sctp_get_next_param(m, offset, ¶m_buf, sizeof(param_buf)); while (phdr) { ptype = ntohs(phdr->param_type); plen = ntohs(phdr->param_length); /* * SCTP_PRINTF("ptype => %0x, plen => %d\n", * (uint32_t)ptype, (int)plen); */ if (offset + plen > limit) { break; } if (plen < sizeof(struct sctp_paramhdr)) { break; } #ifdef INET if (ptype == SCTP_IPV4_ADDRESS) { if (stcb->asoc.scope.ipv4_addr_legal) { struct sctp_ipv4addr_param *p4, p4_buf; /* ok get the v4 address and check/add */ phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&p4_buf, sizeof(p4_buf)); if (plen != sizeof(struct sctp_ipv4addr_param) || phdr == NULL) { return (-5); } p4 = (struct sctp_ipv4addr_param *)phdr; sin.sin_addr.s_addr = p4->addr; if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) { /* Skip multi-cast addresses */ goto next_param; } if ((sin.sin_addr.s_addr == INADDR_BROADCAST) || (sin.sin_addr.s_addr == INADDR_ANY)) { goto next_param; } sa = (struct sockaddr *)&sin; inp = stcb->sctp_ep; atomic_add_int(&stcb->asoc.refcnt, 1); stcb_tmp = sctp_findassociation_ep_addr(&inp, sa, &net, dst, stcb); atomic_add_int(&stcb->asoc.refcnt, -1); if ((stcb_tmp == NULL && inp == stcb->sctp_ep) || inp == NULL) { /* we must add the source address */ /* * no scope set since we have a tcb * already */ /* * we must validate the state again * here */ add_it_now: if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-7); } if (sctp_add_remote_addr(stcb, sa, NULL, port, SCTP_DONOT_SETSCOPE, SCTP_LOAD_ADDR_4)) { return (-8); } } else if (stcb_tmp == stcb) { if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-10); } if (net != NULL) { /* clear flag */ net->dest_state &= ~SCTP_ADDR_NOT_IN_ASSOC; } } else { /* * strange, address is in another * assoc? straighten out locks. */ if (stcb_tmp) { if (SCTP_GET_STATE(stcb_tmp) == SCTP_STATE_COOKIE_WAIT) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; /* * in setup state we * abort this guy */ snprintf(msg, sizeof(msg), "%s:%d at %s", __FILE__, __LINE__, __func__); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); sctp_abort_an_association(stcb_tmp->sctp_ep, stcb_tmp, op_err, SCTP_SO_NOT_LOCKED); goto add_it_now; } SCTP_TCB_UNLOCK(stcb_tmp); } if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-12); } return (-13); } } } else #endif #ifdef INET6 if (ptype == SCTP_IPV6_ADDRESS) { if (stcb->asoc.scope.ipv6_addr_legal) { /* ok get the v6 address and check/add */ struct sctp_ipv6addr_param *p6, p6_buf; phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&p6_buf, sizeof(p6_buf)); if (plen != sizeof(struct sctp_ipv6addr_param) || phdr == NULL) { return (-14); } p6 = (struct sctp_ipv6addr_param *)phdr; memcpy((caddr_t)&sin6.sin6_addr, p6->addr, sizeof(p6->addr)); if (IN6_IS_ADDR_MULTICAST(&sin6.sin6_addr)) { /* Skip multi-cast addresses */ goto next_param; } if (IN6_IS_ADDR_LINKLOCAL(&sin6.sin6_addr)) { /* * Link local make no sense without * scope */ goto next_param; } sa = (struct sockaddr *)&sin6; inp = stcb->sctp_ep; atomic_add_int(&stcb->asoc.refcnt, 1); stcb_tmp = sctp_findassociation_ep_addr(&inp, sa, &net, dst, stcb); atomic_add_int(&stcb->asoc.refcnt, -1); if (stcb_tmp == NULL && (inp == stcb->sctp_ep || inp == NULL)) { /* * we must validate the state again * here */ add_it_now6: if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-16); } /* * we must add the address, no scope * set */ if (sctp_add_remote_addr(stcb, sa, NULL, port, SCTP_DONOT_SETSCOPE, SCTP_LOAD_ADDR_5)) { return (-17); } } else if (stcb_tmp == stcb) { /* * we must validate the state again * here */ if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-19); } if (net != NULL) { /* clear flag */ net->dest_state &= ~SCTP_ADDR_NOT_IN_ASSOC; } } else { /* * strange, address is in another * assoc? straighten out locks. */ if (stcb_tmp) { if (SCTP_GET_STATE(stcb_tmp) == SCTP_STATE_COOKIE_WAIT) { struct mbuf *op_err; char msg[SCTP_DIAG_INFO_LEN]; /* * in setup state we * abort this guy */ snprintf(msg, sizeof(msg), "%s:%d at %s", __FILE__, __LINE__, __func__); op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), msg); sctp_abort_an_association(stcb_tmp->sctp_ep, stcb_tmp, op_err, SCTP_SO_NOT_LOCKED); goto add_it_now6; } SCTP_TCB_UNLOCK(stcb_tmp); } if (stcb->asoc.state == 0) { /* the assoc was freed? */ return (-21); } return (-22); } } } else #endif if (ptype == SCTP_ECN_CAPABLE) { peer_supports_ecn = 1; } else if (ptype == SCTP_ULP_ADAPTATION) { if (stcb->asoc.state != SCTP_STATE_OPEN) { struct sctp_adaptation_layer_indication ai, *aip; phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&ai, sizeof(ai)); aip = (struct sctp_adaptation_layer_indication *)phdr; if (aip) { stcb->asoc.peers_adaptation = ntohl(aip->indication); stcb->asoc.adaptation_needed = 1; } } } else if (ptype == SCTP_SET_PRIM_ADDR) { struct sctp_asconf_addr_param lstore, *fee; int lptype; struct sockaddr *lsa = NULL; #ifdef INET struct sctp_asconf_addrv4_param *fii; #endif if (stcb->asoc.asconf_supported == 0) { return (-100); } if (plen > sizeof(lstore)) { return (-23); } if (plen < sizeof(struct sctp_asconf_addrv4_param)) { return (-101); } phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&lstore, plen); if (phdr == NULL) { return (-24); } fee = (struct sctp_asconf_addr_param *)phdr; lptype = ntohs(fee->addrp.ph.param_type); switch (lptype) { #ifdef INET case SCTP_IPV4_ADDRESS: if (plen != sizeof(struct sctp_asconf_addrv4_param)) { SCTP_PRINTF("Sizeof setprim in init/init ack not %d but %d - ignored\n", (int)sizeof(struct sctp_asconf_addrv4_param), plen); } else { fii = (struct sctp_asconf_addrv4_param *)fee; sin.sin_addr.s_addr = fii->addrp.addr; lsa = (struct sockaddr *)&sin; } break; #endif #ifdef INET6 case SCTP_IPV6_ADDRESS: if (plen != sizeof(struct sctp_asconf_addr_param)) { SCTP_PRINTF("Sizeof setprim (v6) in init/init ack not %d but %d - ignored\n", (int)sizeof(struct sctp_asconf_addr_param), plen); } else { memcpy(sin6.sin6_addr.s6_addr, fee->addrp.addr, sizeof(fee->addrp.addr)); lsa = (struct sockaddr *)&sin6; } break; #endif default: break; } if (lsa) { (void)sctp_set_primary_addr(stcb, sa, NULL); } } else if (ptype == SCTP_HAS_NAT_SUPPORT) { stcb->asoc.peer_supports_nat = 1; } else if (ptype == SCTP_PRSCTP_SUPPORTED) { /* Peer supports pr-sctp */ peer_supports_prsctp = 1; } else if (ptype == SCTP_SUPPORTED_CHUNK_EXT) { /* A supported extension chunk */ struct sctp_supported_chunk_types_param *pr_supported; uint8_t local_store[SCTP_PARAM_BUFFER_SIZE]; int num_ent, i; if (plen > sizeof(local_store)) { return (-35); } phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)&local_store, plen); if (phdr == NULL) { return (-25); } pr_supported = (struct sctp_supported_chunk_types_param *)phdr; num_ent = plen - sizeof(struct sctp_paramhdr); for (i = 0; i < num_ent; i++) { switch (pr_supported->chunk_types[i]) { case SCTP_ASCONF: peer_supports_asconf = 1; break; case SCTP_ASCONF_ACK: peer_supports_asconf_ack = 1; break; case SCTP_FORWARD_CUM_TSN: peer_supports_prsctp = 1; break; case SCTP_PACKET_DROPPED: peer_supports_pktdrop = 1; break; case SCTP_NR_SELECTIVE_ACK: peer_supports_nrsack = 1; break; case SCTP_STREAM_RESET: peer_supports_reconfig = 1; break; case SCTP_AUTHENTICATION: peer_supports_auth = 1; break; case SCTP_IDATA: peer_supports_idata = 1; break; default: /* one I have not learned yet */ break; } } } else if (ptype == SCTP_RANDOM) { if (plen > sizeof(random_store)) break; if (got_random) { /* already processed a RANDOM */ goto next_param; } phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)random_store, plen); if (phdr == NULL) return (-26); p_random = (struct sctp_auth_random *)phdr; random_len = plen - sizeof(*p_random); /* enforce the random length */ if (random_len != SCTP_AUTH_RANDOM_SIZE_REQUIRED) { SCTPDBG(SCTP_DEBUG_AUTH1, "SCTP: invalid RANDOM len\n"); return (-27); } got_random = 1; } else if (ptype == SCTP_HMAC_LIST) { uint16_t num_hmacs; uint16_t i; if (plen > sizeof(hmacs_store)) break; if (got_hmacs) { /* already processed a HMAC list */ goto next_param; } phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)hmacs_store, plen); if (phdr == NULL) return (-28); hmacs = (struct sctp_auth_hmac_algo *)phdr; hmacs_len = plen - sizeof(*hmacs); num_hmacs = hmacs_len / sizeof(hmacs->hmac_ids[0]); /* validate the hmac list */ if (sctp_verify_hmac_param(hmacs, num_hmacs)) { return (-29); } if (stcb->asoc.peer_hmacs != NULL) sctp_free_hmaclist(stcb->asoc.peer_hmacs); stcb->asoc.peer_hmacs = sctp_alloc_hmaclist(num_hmacs); if (stcb->asoc.peer_hmacs != NULL) { for (i = 0; i < num_hmacs; i++) { (void)sctp_auth_add_hmacid(stcb->asoc.peer_hmacs, ntohs(hmacs->hmac_ids[i])); } } got_hmacs = 1; } else if (ptype == SCTP_CHUNK_LIST) { int i; if (plen > sizeof(chunks_store)) break; if (got_chklist) { /* already processed a Chunks list */ goto next_param; } phdr = sctp_get_next_param(m, offset, (struct sctp_paramhdr *)chunks_store, plen); if (phdr == NULL) return (-30); chunks = (struct sctp_auth_chunk_list *)phdr; num_chunks = plen - sizeof(*chunks); if (stcb->asoc.peer_auth_chunks != NULL) sctp_clear_chunklist(stcb->asoc.peer_auth_chunks); else stcb->asoc.peer_auth_chunks = sctp_alloc_chunklist(); for (i = 0; i < num_chunks; i++) { (void)sctp_auth_add_chunk(chunks->chunk_types[i], stcb->asoc.peer_auth_chunks); /* record asconf/asconf-ack if listed */ if (chunks->chunk_types[i] == SCTP_ASCONF) saw_asconf = 1; if (chunks->chunk_types[i] == SCTP_ASCONF_ACK) saw_asconf_ack = 1; } got_chklist = 1; } else if ((ptype == SCTP_HEARTBEAT_INFO) || (ptype == SCTP_STATE_COOKIE) || (ptype == SCTP_UNRECOG_PARAM) || (ptype == SCTP_COOKIE_PRESERVE) || (ptype == SCTP_SUPPORTED_ADDRTYPE) || (ptype == SCTP_ADD_IP_ADDRESS) || (ptype == SCTP_DEL_IP_ADDRESS) || (ptype == SCTP_ERROR_CAUSE_IND) || (ptype == SCTP_SUCCESS_REPORT)) { /* don't care */ ; } else { if ((ptype & 0x8000) == 0x0000) { /* * must stop processing the rest of the * param's. Any report bits were handled * with the call to * sctp_arethere_unrecognized_parameters() * when the INIT or INIT-ACK was first seen. */ break; } } next_param: offset += SCTP_SIZE32(plen); if (offset >= limit) { break; } phdr = sctp_get_next_param(m, offset, ¶m_buf, sizeof(param_buf)); } /* Now check to see if we need to purge any addresses */ TAILQ_FOREACH_SAFE(net, &stcb->asoc.nets, sctp_next, nnet) { if ((net->dest_state & SCTP_ADDR_NOT_IN_ASSOC) == SCTP_ADDR_NOT_IN_ASSOC) { /* This address has been removed from the asoc */ /* remove and free it */ stcb->asoc.numnets--; TAILQ_REMOVE(&stcb->asoc.nets, net, sctp_next); sctp_free_remote_addr(net); if (net == stcb->asoc.primary_destination) { stcb->asoc.primary_destination = NULL; sctp_select_primary_destination(stcb); } } } if ((stcb->asoc.ecn_supported == 1) && (peer_supports_ecn == 0)) { stcb->asoc.ecn_supported = 0; } if ((stcb->asoc.prsctp_supported == 1) && (peer_supports_prsctp == 0)) { stcb->asoc.prsctp_supported = 0; } if ((stcb->asoc.auth_supported == 1) && ((peer_supports_auth == 0) || (got_random == 0) || (got_hmacs == 0))) { stcb->asoc.auth_supported = 0; } if ((stcb->asoc.asconf_supported == 1) && ((peer_supports_asconf == 0) || (peer_supports_asconf_ack == 0) || (stcb->asoc.auth_supported == 0) || (saw_asconf == 0) || (saw_asconf_ack == 0))) { stcb->asoc.asconf_supported = 0; } if ((stcb->asoc.reconfig_supported == 1) && (peer_supports_reconfig == 0)) { stcb->asoc.reconfig_supported = 0; } if ((stcb->asoc.idata_supported == 1) && (peer_supports_idata == 0)) { stcb->asoc.idata_supported = 0; } if ((stcb->asoc.nrsack_supported == 1) && (peer_supports_nrsack == 0)) { stcb->asoc.nrsack_supported = 0; } if ((stcb->asoc.pktdrop_supported == 1) && (peer_supports_pktdrop == 0)) { stcb->asoc.pktdrop_supported = 0; } /* validate authentication required parameters */ if ((peer_supports_auth == 0) && (got_chklist == 1)) { /* peer does not support auth but sent a chunks list? */ return (-31); } if ((peer_supports_asconf == 1) && (peer_supports_auth == 0)) { /* peer supports asconf but not auth? */ return (-32); } else if ((peer_supports_asconf == 1) && (peer_supports_auth == 1) && ((saw_asconf == 0) || (saw_asconf_ack == 0))) { return (-33); } /* concatenate the full random key */ keylen = sizeof(*p_random) + random_len + sizeof(*hmacs) + hmacs_len; if (chunks != NULL) { keylen += sizeof(*chunks) + num_chunks; } new_key = sctp_alloc_key(keylen); if (new_key != NULL) { /* copy in the RANDOM */ if (p_random != NULL) { keylen = sizeof(*p_random) + random_len; memcpy(new_key->key, p_random, keylen); } else { keylen = 0; } /* append in the AUTH chunks */ if (chunks != NULL) { memcpy(new_key->key + keylen, chunks, sizeof(*chunks) + num_chunks); keylen += sizeof(*chunks) + num_chunks; } /* append in the HMACs */ if (hmacs != NULL) { memcpy(new_key->key + keylen, hmacs, sizeof(*hmacs) + hmacs_len); } } else { /* failed to get memory for the key */ return (-34); } if (stcb->asoc.authinfo.peer_random != NULL) sctp_free_key(stcb->asoc.authinfo.peer_random); stcb->asoc.authinfo.peer_random = new_key; sctp_clear_cachedkeys(stcb, stcb->asoc.authinfo.assoc_keyid); sctp_clear_cachedkeys(stcb, stcb->asoc.authinfo.recv_keyid); return (0); } int sctp_set_primary_addr(struct sctp_tcb *stcb, struct sockaddr *sa, struct sctp_nets *net) { /* make sure the requested primary address exists in the assoc */ if (net == NULL && sa) net = sctp_findnet(stcb, sa); if (net == NULL) { /* didn't find the requested primary address! */ return (-1); } else { /* set the primary address */ if (net->dest_state & SCTP_ADDR_UNCONFIRMED) { /* Must be confirmed, so queue to set */ net->dest_state |= SCTP_ADDR_REQ_PRIMARY; return (0); } stcb->asoc.primary_destination = net; if (!(net->dest_state & SCTP_ADDR_PF) && (stcb->asoc.alternate)) { sctp_free_remote_addr(stcb->asoc.alternate); stcb->asoc.alternate = NULL; } net = TAILQ_FIRST(&stcb->asoc.nets); if (net != stcb->asoc.primary_destination) { /* * first one on the list is NOT the primary * sctp_cmpaddr() is much more efficient if the * primary is the first on the list, make it so. */ TAILQ_REMOVE(&stcb->asoc.nets, stcb->asoc.primary_destination, sctp_next); TAILQ_INSERT_HEAD(&stcb->asoc.nets, stcb->asoc.primary_destination, sctp_next); } return (0); } } int sctp_is_vtag_good(uint32_t tag, uint16_t lport, uint16_t rport, struct timeval *now) { /* * This function serves two purposes. It will see if a TAG can be * re-used and return 1 for yes it is ok and 0 for don't use that * tag. A secondary function it will do is purge out old tags that * can be removed. */ struct sctpvtaghead *chain; struct sctp_tagblock *twait_block; struct sctpasochead *head; struct sctp_tcb *stcb; int i; SCTP_INP_INFO_RLOCK(); head = &SCTP_BASE_INFO(sctp_asochash)[SCTP_PCBHASH_ASOC(tag, SCTP_BASE_INFO(hashasocmark))]; LIST_FOREACH(stcb, head, sctp_asocs) { /* * We choose not to lock anything here. TCB's can't be * removed since we have the read lock, so they can't be * freed on us, same thing for the INP. I may be wrong with * this assumption, but we will go with it for now :-) */ if (stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) { continue; } if (stcb->asoc.my_vtag == tag) { /* candidate */ if (stcb->rport != rport) { continue; } if (stcb->sctp_ep->sctp_lport != lport) { continue; } /* Its a used tag set */ SCTP_INP_INFO_RUNLOCK(); return (0); } } chain = &SCTP_BASE_INFO(vtag_timewait)[(tag % SCTP_STACK_VTAG_HASH_SIZE)]; /* Now what about timed wait ? */ LIST_FOREACH(twait_block, chain, sctp_nxt_tagblock) { /* * Block(s) are present, lets see if we have this tag in the * list */ for (i = 0; i < SCTP_NUMBER_IN_VTAG_BLOCK; i++) { if (twait_block->vtag_block[i].v_tag == 0) { /* not used */ continue; } else if ((long)twait_block->vtag_block[i].tv_sec_at_expire < now->tv_sec) { /* Audit expires this guy */ twait_block->vtag_block[i].tv_sec_at_expire = 0; twait_block->vtag_block[i].v_tag = 0; twait_block->vtag_block[i].lport = 0; twait_block->vtag_block[i].rport = 0; } else if ((twait_block->vtag_block[i].v_tag == tag) && (twait_block->vtag_block[i].lport == lport) && (twait_block->vtag_block[i].rport == rport)) { /* Bad tag, sorry :< */ SCTP_INP_INFO_RUNLOCK(); return (0); } } } SCTP_INP_INFO_RUNLOCK(); return (1); } static void sctp_drain_mbufs(struct sctp_tcb *stcb) { /* * We must hunt this association for MBUF's past the cumack (i.e. * out of order data that we can renege on). */ struct sctp_association *asoc; struct sctp_tmit_chunk *chk, *nchk; uint32_t cumulative_tsn_p1; struct sctp_queued_to_read *control, *ncontrol; int cnt, strmat; uint32_t gap, i; int fnd = 0; /* We look for anything larger than the cum-ack + 1 */ asoc = &stcb->asoc; if (asoc->cumulative_tsn == asoc->highest_tsn_inside_map) { /* none we can reneg on. */ return; } SCTP_STAT_INCR(sctps_protocol_drains_done); cumulative_tsn_p1 = asoc->cumulative_tsn + 1; cnt = 0; /* Ok that was fun, now we will drain all the inbound streams? */ for (strmat = 0; strmat < asoc->streamincnt; strmat++) { TAILQ_FOREACH_SAFE(control, &asoc->strmin[strmat].inqueue, next_instrm, ncontrol) { #ifdef INVARIANTS if (control->on_strm_q != SCTP_ON_ORDERED) { panic("Huh control: %p on_q: %d -- not ordered?", control, control->on_strm_q); } #endif if (SCTP_TSN_GT(control->sinfo_tsn, cumulative_tsn_p1)) { /* Yep it is above cum-ack */ cnt++; SCTP_CALC_TSN_TO_GAP(gap, control->sinfo_tsn, asoc->mapping_array_base_tsn); KASSERT(control->length > 0, ("control has zero length")); if (asoc->size_on_all_streams >= control->length) { asoc->size_on_all_streams -= control->length; } else { #ifdef INVARIANTS panic("size_on_all_streams = %u smaller than control length %u", asoc->size_on_all_streams, control->length); #else asoc->size_on_all_streams = 0; #endif } sctp_ucount_decr(asoc->cnt_on_all_streams); SCTP_UNSET_TSN_PRESENT(asoc->mapping_array, gap); if (control->on_read_q) { TAILQ_REMOVE(&stcb->sctp_ep->read_queue, control, next); control->on_read_q = 0; } TAILQ_REMOVE(&asoc->strmin[strmat].inqueue, control, next_instrm); control->on_strm_q = 0; if (control->data) { sctp_m_freem(control->data); control->data = NULL; } sctp_free_remote_addr(control->whoFrom); /* Now its reasm? */ TAILQ_FOREACH_SAFE(chk, &control->reasm, sctp_next, nchk) { cnt++; SCTP_CALC_TSN_TO_GAP(gap, chk->rec.data.tsn, asoc->mapping_array_base_tsn); KASSERT(chk->send_size > 0, ("chunk has zero length")); if (asoc->size_on_reasm_queue >= chk->send_size) { asoc->size_on_reasm_queue -= chk->send_size; } else { #ifdef INVARIANTS panic("size_on_reasm_queue = %u smaller than chunk length %u", asoc->size_on_reasm_queue, chk->send_size); #else asoc->size_on_reasm_queue = 0; #endif } sctp_ucount_decr(asoc->cnt_on_reasm_queue); SCTP_UNSET_TSN_PRESENT(asoc->mapping_array, gap); TAILQ_REMOVE(&control->reasm, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); } sctp_free_a_readq(stcb, control); } } TAILQ_FOREACH_SAFE(control, &asoc->strmin[strmat].uno_inqueue, next_instrm, ncontrol) { #ifdef INVARIANTS if (control->on_strm_q != SCTP_ON_UNORDERED) { panic("Huh control: %p on_q: %d -- not unordered?", control, control->on_strm_q); } #endif if (SCTP_TSN_GT(control->sinfo_tsn, cumulative_tsn_p1)) { /* Yep it is above cum-ack */ cnt++; SCTP_CALC_TSN_TO_GAP(gap, control->sinfo_tsn, asoc->mapping_array_base_tsn); KASSERT(control->length > 0, ("control has zero length")); if (asoc->size_on_all_streams >= control->length) { asoc->size_on_all_streams -= control->length; } else { #ifdef INVARIANTS panic("size_on_all_streams = %u smaller than control length %u", asoc->size_on_all_streams, control->length); #else asoc->size_on_all_streams = 0; #endif } sctp_ucount_decr(asoc->cnt_on_all_streams); SCTP_UNSET_TSN_PRESENT(asoc->mapping_array, gap); if (control->on_read_q) { TAILQ_REMOVE(&stcb->sctp_ep->read_queue, control, next); control->on_read_q = 0; } TAILQ_REMOVE(&asoc->strmin[strmat].uno_inqueue, control, next_instrm); control->on_strm_q = 0; if (control->data) { sctp_m_freem(control->data); control->data = NULL; } sctp_free_remote_addr(control->whoFrom); /* Now its reasm? */ TAILQ_FOREACH_SAFE(chk, &control->reasm, sctp_next, nchk) { cnt++; SCTP_CALC_TSN_TO_GAP(gap, chk->rec.data.tsn, asoc->mapping_array_base_tsn); KASSERT(chk->send_size > 0, ("chunk has zero length")); if (asoc->size_on_reasm_queue >= chk->send_size) { asoc->size_on_reasm_queue -= chk->send_size; } else { #ifdef INVARIANTS panic("size_on_reasm_queue = %u smaller than chunk length %u", asoc->size_on_reasm_queue, chk->send_size); #else asoc->size_on_reasm_queue = 0; #endif } sctp_ucount_decr(asoc->cnt_on_reasm_queue); SCTP_UNSET_TSN_PRESENT(asoc->mapping_array, gap); TAILQ_REMOVE(&control->reasm, chk, sctp_next); if (chk->data) { sctp_m_freem(chk->data); chk->data = NULL; } sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); } sctp_free_a_readq(stcb, control); } } } if (cnt) { /* We must back down to see what the new highest is */ for (i = asoc->highest_tsn_inside_map; SCTP_TSN_GE(i, asoc->mapping_array_base_tsn); i--) { SCTP_CALC_TSN_TO_GAP(gap, i, asoc->mapping_array_base_tsn); if (SCTP_IS_TSN_PRESENT(asoc->mapping_array, gap)) { asoc->highest_tsn_inside_map = i; fnd = 1; break; } } if (!fnd) { asoc->highest_tsn_inside_map = asoc->mapping_array_base_tsn - 1; } /* * Question, should we go through the delivery queue? The * only reason things are on here is the app not reading OR * a p-d-api up. An attacker COULD send enough in to * initiate the PD-API and then send a bunch of stuff to * other streams... these would wind up on the delivery * queue.. and then we would not get to them. But in order * to do this I then have to back-track and un-deliver * sequence numbers in streams.. el-yucko. I think for now * we will NOT look at the delivery queue and leave it to be * something to consider later. An alternative would be to * abort the P-D-API with a notification and then deliver * the data.... Or another method might be to keep track of * how many times the situation occurs and if we see a * possible attack underway just abort the association. */ #ifdef SCTP_DEBUG SCTPDBG(SCTP_DEBUG_PCB1, "Freed %d chunks from reneg harvest\n", cnt); #endif /* * Now do we need to find a new * asoc->highest_tsn_inside_map? */ asoc->last_revoke_count = cnt; sctp_timer_stop(SCTP_TIMER_TYPE_RECV, stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_PCB + SCTP_LOC_11); /* sa_ignore NO_NULL_CHK */ sctp_send_sack(stcb, SCTP_SO_NOT_LOCKED); sctp_chunk_output(stcb->sctp_ep, stcb, SCTP_OUTPUT_FROM_DRAIN, SCTP_SO_NOT_LOCKED); } /* * Another issue, in un-setting the TSN's in the mapping array we * DID NOT adjust the highest_tsn marker. This will cause one of * two things to occur. It may cause us to do extra work in checking * for our mapping array movement. More importantly it may cause us * to SACK every datagram. This may not be a bad thing though since * we will recover once we get our cum-ack above and all this stuff * we dumped recovered. */ } void sctp_drain() { /* * We must walk the PCB lists for ALL associations here. The system * is LOW on MBUF's and needs help. This is where reneging will * occur. We really hope this does NOT happen! */ VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct sctp_inpcb *inp; struct sctp_tcb *stcb; SCTP_STAT_INCR(sctps_protocol_drain_calls); if (SCTP_BASE_SYSCTL(sctp_do_drain) == 0) { #ifdef VIMAGE continue; #else return; #endif } SCTP_INP_INFO_RLOCK(); LIST_FOREACH(inp, &SCTP_BASE_INFO(listhead), sctp_list) { /* For each endpoint */ SCTP_INP_RLOCK(inp); LIST_FOREACH(stcb, &inp->sctp_asoc_list, sctp_tcblist) { /* For each association */ SCTP_TCB_LOCK(stcb); sctp_drain_mbufs(stcb); SCTP_TCB_UNLOCK(stcb); } SCTP_INP_RUNLOCK(inp); } SCTP_INP_INFO_RUNLOCK(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * start a new iterator * iterates through all endpoints and associations based on the pcb_state * flags and asoc_state. "af" (mandatory) is executed for all matching * assocs and "ef" (optional) is executed when the iterator completes. * "inpf" (optional) is executed for each new endpoint as it is being * iterated through. inpe (optional) is called when the inp completes * its way through all the stcbs. */ int sctp_initiate_iterator(inp_func inpf, asoc_func af, inp_func inpe, uint32_t pcb_state, uint32_t pcb_features, uint32_t asoc_state, void *argp, uint32_t argi, end_func ef, struct sctp_inpcb *s_inp, uint8_t chunk_output_off) { struct sctp_iterator *it = NULL; if (af == NULL) { return (-1); } if (SCTP_BASE_VAR(sctp_pcb_initialized) == 0) { SCTP_PRINTF("%s: abort on initialize being %d\n", __func__, SCTP_BASE_VAR(sctp_pcb_initialized)); return (-1); } SCTP_MALLOC(it, struct sctp_iterator *, sizeof(struct sctp_iterator), SCTP_M_ITER); if (it == NULL) { SCTP_LTRACE_ERR_RET(NULL, NULL, NULL, SCTP_FROM_SCTP_PCB, ENOMEM); return (-1); } memset(it, 0, sizeof(*it)); it->function_assoc = af; it->function_inp = inpf; if (inpf) it->done_current_ep = 0; else it->done_current_ep = 1; it->function_atend = ef; it->pointer = argp; it->val = argi; it->pcb_flags = pcb_state; it->pcb_features = pcb_features; it->asoc_state = asoc_state; it->function_inp_end = inpe; it->no_chunk_output = chunk_output_off; it->vn = curvnet; if (s_inp) { /* Assume lock is held here */ it->inp = s_inp; SCTP_INP_INCR_REF(it->inp); it->iterator_flags = SCTP_ITERATOR_DO_SINGLE_INP; } else { SCTP_INP_INFO_RLOCK(); it->inp = LIST_FIRST(&SCTP_BASE_INFO(listhead)); if (it->inp) { SCTP_INP_INCR_REF(it->inp); } SCTP_INP_INFO_RUNLOCK(); it->iterator_flags = SCTP_ITERATOR_DO_ALL_INP; } SCTP_IPI_ITERATOR_WQ_LOCK(); if (SCTP_BASE_VAR(sctp_pcb_initialized) == 0) { SCTP_IPI_ITERATOR_WQ_UNLOCK(); SCTP_PRINTF("%s: rollback on initialize being %d it=%p\n", __func__, SCTP_BASE_VAR(sctp_pcb_initialized), it); SCTP_FREE(it, SCTP_M_ITER); return (-1); } TAILQ_INSERT_TAIL(&sctp_it_ctl.iteratorhead, it, sctp_nxt_itr); if (sctp_it_ctl.iterator_running == 0) { sctp_wakeup_iterator(); } SCTP_IPI_ITERATOR_WQ_UNLOCK(); /* sa_ignore MEMLEAK {memory is put on the tailq for the iterator} */ return (0); } Index: head/sys/netinet/sctp_structs.h =================================================================== --- head/sys/netinet/sctp_structs.h (revision 360291) +++ head/sys/netinet/sctp_structs.h (revision 360292) @@ -1,1250 +1,1250 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2008, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #ifndef _NETINET_SCTP_STRUCTS_H_ #define _NETINET_SCTP_STRUCTS_H_ #include #include #include struct sctp_timer { sctp_os_timer_t timer; int type; /* * Depending on the timer type these will be setup and cast with the * appropriate entity. */ void *ep; void *tcb; void *net; void *vnet; /* for sanity checking */ void *self; uint32_t ticks; uint32_t stopped_from; }; struct sctp_foo_stuff { struct sctp_inpcb *inp; uint32_t lineno; uint32_t ticks; int updown; }; /* * This is the information we track on each interface that we know about from * the distant end. */ TAILQ_HEAD(sctpnetlisthead, sctp_nets); struct sctp_stream_reset_list { TAILQ_ENTRY(sctp_stream_reset_list) next_resp; uint32_t seq; uint32_t tsn; uint32_t number_entries; uint16_t list_of_streams[]; }; TAILQ_HEAD(sctp_resethead, sctp_stream_reset_list); /* * Users of the iterator need to malloc a iterator with a call to * sctp_initiate_iterator(inp_func, assoc_func, inp_func, pcb_flags, pcb_features, * asoc_state, void-ptr-arg, uint32-arg, end_func, inp); * * Use the following two defines if you don't care what pcb flags are on the EP * and/or you don't care what state the association is in. * * Note that if you specify an INP as the last argument then ONLY each * association of that single INP will be executed upon. Note that the pcb * flags STILL apply so if the inp you specify has different pcb_flags then * what you put in pcb_flags nothing will happen. use SCTP_PCB_ANY_FLAGS to * assure the inp you specify gets treated. */ #define SCTP_PCB_ANY_FLAGS 0x00000000 #define SCTP_PCB_ANY_FEATURES 0x00000000 #define SCTP_ASOC_ANY_STATE 0x00000000 typedef void (*asoc_func) (struct sctp_inpcb *, struct sctp_tcb *, void *ptr, uint32_t val); typedef int (*inp_func) (struct sctp_inpcb *, void *ptr, uint32_t val); typedef void (*end_func) (void *ptr, uint32_t val); #if defined(__FreeBSD__) && defined(SCTP_MCORE_INPUT) && defined(SMP) /* whats on the mcore control struct */ struct sctp_mcore_queue { TAILQ_ENTRY(sctp_mcore_queue) next; struct vnet *vn; struct mbuf *m; int off; int v6; }; TAILQ_HEAD(sctp_mcore_qhead, sctp_mcore_queue); struct sctp_mcore_ctrl { SCTP_PROCESS_STRUCT thread_proc; struct sctp_mcore_qhead que; struct mtx core_mtx; struct mtx que_mtx; int running; int cpuid; }; #endif struct sctp_iterator { TAILQ_ENTRY(sctp_iterator) sctp_nxt_itr; struct vnet *vn; struct sctp_timer tmr; struct sctp_inpcb *inp; /* current endpoint */ struct sctp_tcb *stcb; /* current* assoc */ struct sctp_inpcb *next_inp; /* special hook to skip to */ asoc_func function_assoc; /* per assoc function */ inp_func function_inp; /* per endpoint function */ inp_func function_inp_end; /* end INP function */ end_func function_atend; /* iterator completion function */ void *pointer; /* pointer for apply func to use */ uint32_t val; /* value for apply func to use */ uint32_t pcb_flags; /* endpoint flags being checked */ uint32_t pcb_features; /* endpoint features being checked */ uint32_t asoc_state; /* assoc state being checked */ uint32_t iterator_flags; uint8_t no_chunk_output; uint8_t done_current_ep; }; /* iterator_flags values */ #define SCTP_ITERATOR_DO_ALL_INP 0x00000001 #define SCTP_ITERATOR_DO_SINGLE_INP 0x00000002 TAILQ_HEAD(sctpiterators, sctp_iterator); struct sctp_copy_all { struct sctp_inpcb *inp; /* ep */ struct mbuf *m; struct sctp_sndrcvinfo sndrcv; ssize_t sndlen; int cnt_sent; int cnt_failed; }; struct sctp_asconf_iterator { struct sctpladdr list_of_work; int cnt; }; struct iterator_control { struct mtx ipi_iterator_wq_mtx; struct mtx it_mtx; SCTP_PROCESS_STRUCT thread_proc; struct sctpiterators iteratorhead; struct sctp_iterator *cur_it; uint32_t iterator_running; uint32_t iterator_flags; }; #define SCTP_ITERATOR_STOP_CUR_IT 0x00000004 #define SCTP_ITERATOR_STOP_CUR_INP 0x00000008 struct sctp_net_route { - sctp_rtentry_t *ro_rt; + struct nhop_object *ro_nh; struct llentry *ro_lle; char *ro_prepend; uint16_t ro_plen; uint16_t ro_flags; uint16_t ro_mtu; uint16_t spare; union sctp_sockstore _l_addr; /* remote peer addr */ struct sctp_ifa *_s_addr; /* our selected src addr */ }; struct htcp { uint16_t alpha; /* Fixed point arith, << 7 */ uint8_t beta; /* Fixed point arith, << 7 */ uint8_t modeswitch; /* Delay modeswitch until we had at least one * congestion event */ uint32_t last_cong; /* Time since last congestion event end */ uint32_t undo_last_cong; uint16_t bytes_acked; uint32_t bytecount; uint32_t minRTT; uint32_t maxRTT; uint32_t undo_maxRTT; uint32_t undo_old_maxB; /* Bandwidth estimation */ uint32_t minB; uint32_t maxB; uint32_t old_maxB; uint32_t Bi; uint32_t lasttime; }; struct rtcc_cc { struct timeval tls; /* The time we started the sending */ uint64_t lbw; /* Our last estimated bw */ uint64_t lbw_rtt; /* RTT at bw estimate */ uint64_t bw_bytes; /* The total bytes since this sending began */ uint64_t bw_tot_time; /* The total time since sending began */ uint64_t new_tot_time; /* temp holding the new value */ uint64_t bw_bytes_at_last_rttc; /* What bw_bytes was at last rtt calc */ uint32_t cwnd_at_bw_set; /* Cwnd at last bw saved - lbw */ uint32_t vol_reduce; /* cnt of voluntary reductions */ uint16_t steady_step; /* The number required to be in steady state */ uint16_t step_cnt; /* The current number */ uint8_t ret_from_eq; /* When all things are equal what do I return * 0/1 - 1 no cc advance */ uint8_t use_dccc_ecn; /* Flag to enable DCCC ECN */ uint8_t tls_needs_set; /* Flag to indicate we need to set tls 0 or 1 * means set at send 2 not */ uint8_t last_step_state; /* Last state if steady state stepdown * is on */ uint8_t rtt_set_this_sack; /* Flag saying this sack had RTT calc * on it */ uint8_t last_inst_ind; /* Last saved inst indication */ }; struct sctp_nets { TAILQ_ENTRY(sctp_nets) sctp_next; /* next link */ /* * Things on the top half may be able to be split into a common * structure shared by all. */ struct sctp_timer pmtu_timer; struct sctp_timer hb_timer; /* * The following two in combination equate to a route entry for v6 * or v4. */ struct sctp_net_route ro; /* mtu discovered so far */ uint32_t mtu; uint32_t ssthresh; /* not sure about this one for split */ uint32_t last_cwr_tsn; uint32_t cwr_window_tsn; uint32_t ecn_ce_pkt_cnt; uint32_t lost_cnt; /* smoothed average things for RTT and RTO itself */ int lastsa; int lastsv; uint64_t rtt; /* last measured rtt value in us */ uint32_t RTO; /* This is used for SHUTDOWN/SHUTDOWN-ACK/SEND or INIT timers */ struct sctp_timer rxt_timer; /* last time in seconds I sent to it */ struct timeval last_sent_time; union cc_control_data { struct htcp htcp_ca; /* JRS - struct used in HTCP algorithm */ struct rtcc_cc rtcc; /* rtcc module cc stuff */ } cc_mod; int ref_count; /* Congestion stats per destination */ /* * flight size variables and such, sorry Vern, I could not avoid * this if I wanted performance :> */ uint32_t flight_size; uint32_t cwnd; /* actual cwnd */ uint32_t prev_cwnd; /* cwnd before any processing */ uint32_t ecn_prev_cwnd; /* ECN prev cwnd at first ecn_echo seen in new * window */ uint32_t partial_bytes_acked; /* in CA tracks when to incr a MTU */ /* tracking variables to avoid the aloc/free in sack processing */ unsigned int net_ack; unsigned int net_ack2; /* * JRS - 5/8/07 - Variable to track last time a destination was * active for CMT PF */ uint32_t last_active; /* * CMT variables (iyengar@cis.udel.edu) */ uint32_t this_sack_highest_newack; /* tracks highest TSN newly * acked for a given dest in * the current SACK. Used in * SFR and HTNA algos */ uint32_t pseudo_cumack; /* CMT CUC algorithm. Maintains next expected * pseudo-cumack for this destination */ uint32_t rtx_pseudo_cumack; /* CMT CUC algorithm. Maintains next * expected pseudo-cumack for this * destination */ /* CMT fast recovery variables */ uint32_t fast_recovery_tsn; uint32_t heartbeat_random1; uint32_t heartbeat_random2; #ifdef INET6 uint32_t flowlabel; #endif uint8_t dscp; struct timeval start_time; /* time when this net was created */ uint32_t marked_retrans; /* number or DATA chunks marked for * timer based retransmissions */ uint32_t marked_fastretrans; uint32_t heart_beat_delay; /* Heart Beat delay in ms */ /* if this guy is ok or not ... status */ uint16_t dest_state; /* number of timeouts to consider the destination unreachable */ uint16_t failure_threshold; /* number of timeouts to consider the destination potentially failed */ uint16_t pf_threshold; /* error stats on the destination */ uint16_t error_count; /* UDP port number in case of UDP tunneling */ uint16_t port; uint8_t fast_retran_loss_recovery; uint8_t will_exit_fast_recovery; /* Flags that probably can be combined into dest_state */ uint8_t fast_retran_ip; /* fast retransmit in progress */ uint8_t hb_responded; uint8_t saw_newack; /* CMT's SFR algorithm flag */ uint8_t src_addr_selected; /* if we split we move */ uint8_t indx_of_eligible_next_to_use; uint8_t addr_is_local; /* its a local address (if known) could move * in split */ /* * CMT variables (iyengar@cis.udel.edu) */ uint8_t find_pseudo_cumack; /* CMT CUC algorithm. Flag used to * find a new pseudocumack. This flag * is set after a new pseudo-cumack * has been received and indicates * that the sender should find the * next pseudo-cumack expected for * this destination */ uint8_t find_rtx_pseudo_cumack; /* CMT CUCv2 algorithm. Flag used to * find a new rtx-pseudocumack. This * flag is set after a new * rtx-pseudo-cumack has been received * and indicates that the sender * should find the next * rtx-pseudo-cumack expected for this * destination */ uint8_t new_pseudo_cumack; /* CMT CUC algorithm. Flag used to * indicate if a new pseudo-cumack or * rtx-pseudo-cumack has been received */ uint8_t window_probe; /* Doing a window probe? */ uint8_t RTO_measured; /* Have we done the first measure */ uint8_t last_hs_used; /* index into the last HS table entry we used */ uint8_t lan_type; uint8_t rto_needed; uint32_t flowid; uint8_t flowtype; }; struct sctp_data_chunkrec { uint32_t tsn; /* the TSN of this transmit */ uint32_t mid; /* the message identifier of this transmit */ uint16_t sid; /* the stream number of this guy */ uint32_t ppid; uint32_t context; /* from send */ uint32_t cwnd_at_send; /* * part of the Highest sacked algorithm to be able to stroke counts * on ones that are FR'd. */ uint32_t fast_retran_tsn; /* sending_seq at the time of FR */ struct timeval timetodrop; /* time we drop it from queue */ uint32_t fsn; /* Fragment Sequence Number */ uint8_t doing_fast_retransmit; uint8_t rcv_flags; /* flags pulled from data chunk on inbound for * outbound holds sending flags for PR-SCTP. */ uint8_t state_flags; uint8_t chunk_was_revoked; uint8_t fwd_tsn_cnt; }; TAILQ_HEAD(sctpchunk_listhead, sctp_tmit_chunk); /* The lower byte is used to enumerate PR_SCTP policies */ #define CHUNK_FLAGS_PR_SCTP_TTL SCTP_PR_SCTP_TTL #define CHUNK_FLAGS_PR_SCTP_BUF SCTP_PR_SCTP_BUF #define CHUNK_FLAGS_PR_SCTP_RTX SCTP_PR_SCTP_RTX /* The upper byte is used as a bit mask */ #define CHUNK_FLAGS_FRAGMENT_OK 0x0100 struct chk_id { uint8_t id; uint8_t can_take_data; }; struct sctp_tmit_chunk { union { struct sctp_data_chunkrec data; struct chk_id chunk_id; } rec; struct sctp_association *asoc; /* bp to asoc this belongs to */ struct timeval sent_rcv_time; /* filled in if RTT being calculated */ struct mbuf *data; /* pointer to mbuf chain of data */ struct mbuf *last_mbuf; /* pointer to last mbuf in chain */ struct sctp_nets *whoTo; TAILQ_ENTRY(sctp_tmit_chunk) sctp_next; /* next link */ int32_t sent; /* the send status */ uint16_t snd_count; /* number of times I sent */ uint16_t flags; /* flags, such as FRAGMENT_OK */ uint16_t send_size; uint16_t book_size; uint16_t mbcnt; uint16_t auth_keyid; uint8_t holds_key_ref; /* flag if auth keyid refcount is held */ uint8_t pad_inplace; uint8_t do_rtt; uint8_t book_size_scale; uint8_t no_fr_allowed; uint8_t copy_by_ref; uint8_t window_probe; }; struct sctp_queued_to_read { /* sinfo structure Pluse more */ uint16_t sinfo_stream; /* off the wire */ uint16_t sinfo_flags; /* SCTP_UNORDERED from wire use SCTP_EOF for * EOR */ uint32_t sinfo_ppid; /* off the wire */ uint32_t sinfo_context; /* pick this up from assoc def context? */ uint32_t sinfo_timetolive; /* not used by kernel */ uint32_t sinfo_tsn; /* Use this in reassembly as first TSN */ uint32_t sinfo_cumtsn; /* Use this in reassembly as last TSN */ sctp_assoc_t sinfo_assoc_id; /* our assoc id */ /* Non sinfo stuff */ uint32_t mid; /* Fragment Index */ uint32_t length; /* length of data */ uint32_t held_length; /* length held in sb */ uint32_t top_fsn; /* Highest FSN in queue */ uint32_t fsn_included; /* Highest FSN in *data portion */ struct sctp_nets *whoFrom; /* where it came from */ struct mbuf *data; /* front of the mbuf chain of data with * PKT_HDR */ struct mbuf *tail_mbuf; /* used for multi-part data */ struct mbuf *aux_data; /* used to hold/cache control if o/s does not * take it from us */ struct sctp_tcb *stcb; /* assoc, used for window update */ TAILQ_ENTRY(sctp_queued_to_read) next; TAILQ_ENTRY(sctp_queued_to_read) next_instrm; struct sctpchunk_listhead reasm; uint16_t port_from; uint16_t spec_flags; /* Flags to hold the notification field */ uint8_t do_not_ref_stcb; uint8_t end_added; uint8_t pdapi_aborted; uint8_t pdapi_started; uint8_t some_taken; uint8_t last_frag_seen; uint8_t first_frag_seen; uint8_t on_read_q; uint8_t on_strm_q; }; #define SCTP_ON_ORDERED 1 #define SCTP_ON_UNORDERED 2 /* This data structure will be on the outbound * stream queues. Data will be pulled off from * the front of the mbuf data and chunk-ified * by the output routines. We will custom * fit every chunk we pull to the send/sent * queue to make up the next full packet * if we can. An entry cannot be removed * from the stream_out queue until * the msg_is_complete flag is set. This * means at times data/tail_mbuf MIGHT * be NULL.. If that occurs it happens * for one of two reasons. Either the user * is blocked on a send() call and has not * awoken to copy more data down... OR * the user is in the explict MSG_EOR mode * and wrote some data, but has not completed * sending. */ struct sctp_stream_queue_pending { struct mbuf *data; struct mbuf *tail_mbuf; struct timeval ts; struct sctp_nets *net; TAILQ_ENTRY(sctp_stream_queue_pending) next; TAILQ_ENTRY(sctp_stream_queue_pending) ss_next; uint32_t fsn; uint32_t length; uint32_t timetolive; uint32_t ppid; uint32_t context; uint16_t sinfo_flags; uint16_t sid; uint16_t act_flags; uint16_t auth_keyid; uint8_t holds_key_ref; uint8_t msg_is_complete; uint8_t some_taken; uint8_t sender_all_done; uint8_t put_last_out; uint8_t discard_rest; }; /* * this struct contains info that is used to track inbound stream data and * help with ordering. */ TAILQ_HEAD(sctpwheelunrel_listhead, sctp_stream_in); struct sctp_stream_in { struct sctp_readhead inqueue; struct sctp_readhead uno_inqueue; uint32_t last_mid_delivered; /* used for re-order */ uint16_t sid; uint8_t delivery_started; uint8_t pd_api_started; }; TAILQ_HEAD(sctpwheel_listhead, sctp_stream_out); TAILQ_HEAD(sctplist_listhead, sctp_stream_queue_pending); /* Round-robin schedulers */ struct ss_rr { /* next link in wheel */ TAILQ_ENTRY(sctp_stream_out) next_spoke; }; /* Priority scheduler */ struct ss_prio { /* next link in wheel */ TAILQ_ENTRY(sctp_stream_out) next_spoke; /* priority id */ uint16_t priority; }; /* Fair Bandwidth scheduler */ struct ss_fb { /* next link in wheel */ TAILQ_ENTRY(sctp_stream_out) next_spoke; /* stores message size */ int32_t rounds; }; /* * This union holds all data necessary for * different stream schedulers. */ struct scheduling_data { struct sctp_stream_out *locked_on_sending; /* circular looking for output selection */ struct sctp_stream_out *last_out_stream; union { struct sctpwheel_listhead wheel; struct sctplist_listhead list; } out; }; /* * This union holds all parameters per stream * necessary for different stream schedulers. */ union scheduling_parameters { struct ss_rr rr; struct ss_prio prio; struct ss_fb fb; }; /* States for outgoing streams */ #define SCTP_STREAM_CLOSED 0x00 #define SCTP_STREAM_OPENING 0x01 #define SCTP_STREAM_OPEN 0x02 #define SCTP_STREAM_RESET_PENDING 0x03 #define SCTP_STREAM_RESET_IN_FLIGHT 0x04 #define SCTP_MAX_STREAMS_AT_ONCE_RESET 200 /* This struct is used to track the traffic on outbound streams */ struct sctp_stream_out { struct sctp_streamhead outqueue; union scheduling_parameters ss_params; uint32_t chunks_on_queues; /* send queue and sent queue */ #if defined(SCTP_DETAILED_STR_STATS) uint32_t abandoned_unsent[SCTP_PR_SCTP_MAX + 1]; uint32_t abandoned_sent[SCTP_PR_SCTP_MAX + 1]; #else /* Only the aggregation */ uint32_t abandoned_unsent[1]; uint32_t abandoned_sent[1]; #endif /* * For associations using DATA chunks, the lower 16-bit of * next_mid_ordered are used as the next SSN. */ uint32_t next_mid_ordered; uint32_t next_mid_unordered; uint16_t sid; uint8_t last_msg_incomplete; uint8_t state; }; /* used to keep track of the addresses yet to try to add/delete */ TAILQ_HEAD(sctp_asconf_addrhead, sctp_asconf_addr); struct sctp_asconf_addr { TAILQ_ENTRY(sctp_asconf_addr) next; struct sctp_asconf_addr_param ap; struct sctp_ifa *ifa; /* save the ifa for add/del ip */ uint8_t sent; /* has this been sent yet? */ uint8_t special_del; /* not to be used in lookup */ }; struct sctp_scoping { uint8_t ipv4_addr_legal; uint8_t ipv6_addr_legal; uint8_t loopback_scope; uint8_t ipv4_local_scope; uint8_t local_scope; uint8_t site_scope; }; #define SCTP_TSN_LOG_SIZE 40 struct sctp_tsn_log { void *stcb; uint32_t tsn; uint32_t seq; uint16_t strm; uint16_t sz; uint16_t flgs; uint16_t in_pos; uint16_t in_out; uint16_t resv; }; #define SCTP_FS_SPEC_LOG_SIZE 200 struct sctp_fs_spec_log { uint32_t sent; uint32_t total_flight; uint32_t tsn; uint16_t book; uint8_t incr; uint8_t decr; }; /* This struct is here to cut out the compatiabilty * pad that bulks up both the inp and stcb. The non * pad portion MUST stay in complete sync with * sctp_sndrcvinfo... i.e. if sinfo_xxxx is added * this must be done here too. */ struct sctp_nonpad_sndrcvinfo { uint16_t sinfo_stream; uint16_t sinfo_ssn; uint16_t sinfo_flags; uint32_t sinfo_ppid; uint32_t sinfo_context; uint32_t sinfo_timetolive; uint32_t sinfo_tsn; uint32_t sinfo_cumtsn; sctp_assoc_t sinfo_assoc_id; uint16_t sinfo_keynumber; uint16_t sinfo_keynumber_valid; }; /* * JRS - Structure to hold function pointers to the functions responsible * for congestion control. */ struct sctp_cc_functions { void (*sctp_set_initial_cc_param) (struct sctp_tcb *stcb, struct sctp_nets *net); void (*sctp_cwnd_update_after_sack) (struct sctp_tcb *stcb, struct sctp_association *asoc, int accum_moved, int reneged_all, int will_exit); void (*sctp_cwnd_update_exit_pf) (struct sctp_tcb *stcb, struct sctp_nets *net); void (*sctp_cwnd_update_after_fr) (struct sctp_tcb *stcb, struct sctp_association *asoc); void (*sctp_cwnd_update_after_timeout) (struct sctp_tcb *stcb, struct sctp_nets *net); void (*sctp_cwnd_update_after_ecn_echo) (struct sctp_tcb *stcb, struct sctp_nets *net, int in_window, int num_pkt_lost); void (*sctp_cwnd_update_after_packet_dropped) (struct sctp_tcb *stcb, struct sctp_nets *net, struct sctp_pktdrop_chunk *cp, uint32_t *bottle_bw, uint32_t *on_queue); void (*sctp_cwnd_update_after_output) (struct sctp_tcb *stcb, struct sctp_nets *net, int burst_limit); void (*sctp_cwnd_update_packet_transmitted) (struct sctp_tcb *stcb, struct sctp_nets *net); void (*sctp_cwnd_update_tsn_acknowledged) (struct sctp_nets *net, struct sctp_tmit_chunk *); void (*sctp_cwnd_new_transmission_begins) (struct sctp_tcb *stcb, struct sctp_nets *net); void (*sctp_cwnd_prepare_net_for_sack) (struct sctp_tcb *stcb, struct sctp_nets *net); int (*sctp_cwnd_socket_option) (struct sctp_tcb *stcb, int set, struct sctp_cc_option *); void (*sctp_rtt_calculated) (struct sctp_tcb *, struct sctp_nets *, struct timeval *); }; /* * RS - Structure to hold function pointers to the functions responsible * for stream scheduling. */ struct sctp_ss_functions { void (*sctp_ss_init) (struct sctp_tcb *stcb, struct sctp_association *asoc, int holds_lock); void (*sctp_ss_clear) (struct sctp_tcb *stcb, struct sctp_association *asoc, int clear_values, int holds_lock); void (*sctp_ss_init_stream) (struct sctp_tcb *stcb, struct sctp_stream_out *strq, struct sctp_stream_out *with_strq); void (*sctp_ss_add_to_stream) (struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_stream_out *strq, struct sctp_stream_queue_pending *sp, int holds_lock); int (*sctp_ss_is_empty) (struct sctp_tcb *stcb, struct sctp_association *asoc); void (*sctp_ss_remove_from_stream) (struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_stream_out *strq, struct sctp_stream_queue_pending *sp, int holds_lock); struct sctp_stream_out *(*sctp_ss_select_stream) (struct sctp_tcb *stcb, struct sctp_nets *net, struct sctp_association *asoc); void (*sctp_ss_scheduled) (struct sctp_tcb *stcb, struct sctp_nets *net, struct sctp_association *asoc, struct sctp_stream_out *strq, int moved_how_much); void (*sctp_ss_packet_done) (struct sctp_tcb *stcb, struct sctp_nets *net, struct sctp_association *asoc); int (*sctp_ss_get_value) (struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_stream_out *strq, uint16_t *value); int (*sctp_ss_set_value) (struct sctp_tcb *stcb, struct sctp_association *asoc, struct sctp_stream_out *strq, uint16_t value); int (*sctp_ss_is_user_msgs_incomplete) (struct sctp_tcb *stcb, struct sctp_association *asoc); }; /* used to save ASCONF chunks for retransmission */ TAILQ_HEAD(sctp_asconf_head, sctp_asconf); struct sctp_asconf { TAILQ_ENTRY(sctp_asconf) next; uint32_t serial_number; uint16_t snd_count; struct mbuf *data; uint16_t len; }; /* used to save ASCONF-ACK chunks for retransmission */ TAILQ_HEAD(sctp_asconf_ackhead, sctp_asconf_ack); struct sctp_asconf_ack { TAILQ_ENTRY(sctp_asconf_ack) next; uint32_t serial_number; struct sctp_nets *last_sent_to; struct mbuf *data; uint16_t len; }; /* * Here we have information about each individual association that we track. * We probably in production would be more dynamic. But for ease of * implementation we will have a fixed array that we hunt for in a linear * fashion. */ struct sctp_association { /* association state */ int state; /* queue of pending addrs to add/delete */ struct sctp_asconf_addrhead asconf_queue; struct timeval time_entered; /* time we entered state */ struct timeval time_last_rcvd; struct timeval time_last_sent; struct timeval time_last_sat_advance; struct sctp_nonpad_sndrcvinfo def_send; /* timers and such */ struct sctp_timer dack_timer; /* Delayed ack timer */ struct sctp_timer asconf_timer; /* asconf */ struct sctp_timer strreset_timer; /* stream reset */ struct sctp_timer shut_guard_timer; /* shutdown guard */ struct sctp_timer autoclose_timer; /* automatic close timer */ struct sctp_timer delete_prim_timer; /* deleting primary dst */ /* list of restricted local addresses */ struct sctpladdr sctp_restricted_addrs; /* last local address pending deletion (waiting for an address add) */ struct sctp_ifa *asconf_addr_del_pending; /* Deleted primary destination (used to stop timer) */ struct sctp_nets *deleted_primary; struct sctpnetlisthead nets; /* remote address list */ /* Free chunk list */ struct sctpchunk_listhead free_chunks; /* Control chunk queue */ struct sctpchunk_listhead control_send_queue; /* ASCONF chunk queue */ struct sctpchunk_listhead asconf_send_queue; /* * Once a TSN hits the wire it is moved to the sent_queue. We * maintain two counts here (don't know if any but retran_cnt is * needed). The idea is that the sent_queue_retran_cnt reflects how * many chunks have been marked for retranmission by either T3-rxt * or FR. */ struct sctpchunk_listhead sent_queue; struct sctpchunk_listhead send_queue; /* Scheduling queues */ struct scheduling_data ss_data; /* If an iterator is looking at me, this is it */ struct sctp_iterator *stcb_starting_point_for_iterator; /* ASCONF save the last ASCONF-ACK so we can resend it if necessary */ struct sctp_asconf_ackhead asconf_ack_sent; /* * pointer to last stream reset queued to control queue by us with * requests. */ struct sctp_tmit_chunk *str_reset; /* * if Source Address Selection happening, this will rotate through * the link list. */ struct sctp_laddr *last_used_address; /* stream arrays */ struct sctp_stream_in *strmin; struct sctp_stream_out *strmout; uint8_t *mapping_array; /* primary destination to use */ struct sctp_nets *primary_destination; struct sctp_nets *alternate; /* If primary is down or PF */ /* For CMT */ struct sctp_nets *last_net_cmt_send_started; /* last place I got a data chunk from */ struct sctp_nets *last_data_chunk_from; /* last place I got a control from */ struct sctp_nets *last_control_chunk_from; /* * wait to the point the cum-ack passes req->send_reset_at_tsn for * any req on the list. */ struct sctp_resethead resetHead; /* queue of chunks waiting to be sent into the local stack */ struct sctp_readhead pending_reply_queue; /* JRS - the congestion control functions are in this struct */ struct sctp_cc_functions cc_functions; /* * JRS - value to store the currently loaded congestion control * module */ uint32_t congestion_control_module; /* RS - the stream scheduling functions are in this struct */ struct sctp_ss_functions ss_functions; /* RS - value to store the currently loaded stream scheduling module */ uint32_t stream_scheduling_module; uint32_t vrf_id; uint32_t cookie_preserve_req; /* ASCONF next seq I am sending out, inits at init-tsn */ uint32_t asconf_seq_out; uint32_t asconf_seq_out_acked; /* ASCONF last received ASCONF from peer, starts at peer's TSN-1 */ uint32_t asconf_seq_in; /* next seq I am sending in str reset messages */ uint32_t str_reset_seq_out; /* next seq I am expecting in str reset messages */ uint32_t str_reset_seq_in; /* various verification tag information */ uint32_t my_vtag; /* The tag to be used. if assoc is re-initited * by remote end, and I have unlocked this * will be regenerated to a new random value. */ uint32_t peer_vtag; /* The peers last tag */ uint32_t my_vtag_nonce; uint32_t peer_vtag_nonce; uint32_t assoc_id; /* This is the SCTP fragmentation threshold */ uint32_t smallest_mtu; /* * Special hook for Fast retransmit, allows us to track the highest * TSN that is NEW in this SACK if gap ack blocks are present. */ uint32_t this_sack_highest_gap; /* * The highest consecutive TSN that has been acked by peer on my * sends */ uint32_t last_acked_seq; /* The next TSN that I will use in sending. */ uint32_t sending_seq; /* Original seq number I used ??questionable to keep?? */ uint32_t init_seq_number; /* The Advanced Peer Ack Point, as required by the PR-SCTP */ /* (A1 in Section 4.2) */ uint32_t advanced_peer_ack_point; /* * The highest consequetive TSN at the bottom of the mapping array * (for his sends). */ uint32_t cumulative_tsn; /* * Used to track the mapping array and its offset bits. This MAY be * lower then cumulative_tsn. */ uint32_t mapping_array_base_tsn; /* * used to track highest TSN we have received and is listed in the * mapping array. */ uint32_t highest_tsn_inside_map; /* EY - new NR variables used for nr_sack based on mapping_array */ uint8_t *nr_mapping_array; uint32_t highest_tsn_inside_nr_map; uint32_t fast_recovery_tsn; uint32_t sat_t3_recovery_tsn; uint32_t tsn_last_delivered; /* * For the pd-api we should re-write this a bit more efficient. We * could have multiple sctp_queued_to_read's that we are building at * once. Now we only do this when we get ready to deliver to the * socket buffer. Note that we depend on the fact that the struct is * "stuck" on the read queue until we finish all the pd-api. */ struct sctp_queued_to_read *control_pdapi; uint32_t tsn_of_pdapi_last_delivered; uint32_t pdapi_ppid; uint32_t context; uint32_t last_reset_action[SCTP_MAX_RESET_PARAMS]; uint32_t last_sending_seq[SCTP_MAX_RESET_PARAMS]; uint32_t last_base_tsnsent[SCTP_MAX_RESET_PARAMS]; #ifdef SCTP_ASOCLOG_OF_TSNS /* * special log - This adds considerable size to the asoc, but * provides a log that you can use to detect problems via kgdb. */ struct sctp_tsn_log in_tsnlog[SCTP_TSN_LOG_SIZE]; struct sctp_tsn_log out_tsnlog[SCTP_TSN_LOG_SIZE]; uint32_t cumack_log[SCTP_TSN_LOG_SIZE]; uint32_t cumack_logsnt[SCTP_TSN_LOG_SIZE]; uint16_t tsn_in_at; uint16_t tsn_out_at; uint16_t tsn_in_wrapped; uint16_t tsn_out_wrapped; uint16_t cumack_log_at; uint16_t cumack_log_atsnt; #endif /* SCTP_ASOCLOG_OF_TSNS */ #ifdef SCTP_FS_SPEC_LOG struct sctp_fs_spec_log fslog[SCTP_FS_SPEC_LOG_SIZE]; uint16_t fs_index; #endif /* * window state information and smallest MTU that I use to bound * segmentation */ uint32_t peers_rwnd; uint32_t my_rwnd; uint32_t my_last_reported_rwnd; uint32_t sctp_frag_point; uint32_t total_output_queue_size; uint32_t sb_cc; /* shadow of sb_cc */ uint32_t sb_send_resv; /* amount reserved on a send */ uint32_t my_rwnd_control_len; /* shadow of sb_mbcnt used for rwnd * control */ #ifdef INET6 uint32_t default_flowlabel; #endif uint32_t pr_sctp_cnt; int ctrl_queue_cnt; /* could be removed REM - NO IT CAN'T!! RRS */ /* * All outbound datagrams queue into this list from the individual * stream queue. Here they get assigned a TSN and then await * sending. The stream seq comes when it is first put in the * individual str queue */ unsigned int stream_queue_cnt; unsigned int send_queue_cnt; unsigned int sent_queue_cnt; unsigned int sent_queue_cnt_removeable; /* * Number on sent queue that are marked for retran until this value * is 0 we only send one packet of retran'ed data. */ unsigned int sent_queue_retran_cnt; unsigned int size_on_reasm_queue; unsigned int cnt_on_reasm_queue; unsigned int fwd_tsn_cnt; /* amount of data (bytes) currently in flight (on all destinations) */ unsigned int total_flight; /* Total book size in flight */ unsigned int total_flight_count; /* count of chunks used with * book total */ /* count of destinaton nets and list of destination nets */ unsigned int numnets; /* Total error count on this association */ unsigned int overall_error_count; unsigned int cnt_msg_on_sb; /* All stream count of chunks for delivery */ unsigned int size_on_all_streams; unsigned int cnt_on_all_streams; /* Heart Beat delay in ms */ uint32_t heart_beat_delay; /* autoclose */ uint32_t sctp_autoclose_ticks; /* how many preopen streams we have */ unsigned int pre_open_streams; /* How many streams I support coming into me */ unsigned int max_inbound_streams; /* the cookie life I award for any cookie, in seconds */ uint32_t cookie_life; /* time to delay acks for */ unsigned int delayed_ack; unsigned int old_delayed_ack; unsigned int sack_freq; unsigned int data_pkts_seen; unsigned int numduptsns; int dup_tsns[SCTP_MAX_DUP_TSNS]; uint32_t initial_init_rto_max; /* initial RTO for INIT's */ uint32_t initial_rto; /* initial send RTO */ uint32_t minrto; /* per assoc RTO-MIN */ uint32_t maxrto; /* per assoc RTO-MAX */ /* authentication fields */ sctp_auth_chklist_t *local_auth_chunks; sctp_auth_chklist_t *peer_auth_chunks; sctp_hmaclist_t *local_hmacs; /* local HMACs supported */ sctp_hmaclist_t *peer_hmacs; /* peer HMACs supported */ struct sctp_keyhead shared_keys; /* assoc's shared keys */ sctp_authinfo_t authinfo; /* randoms, cached keys */ /* * refcnt to block freeing when a sender or receiver is off coping * user data in. */ uint32_t refcnt; uint32_t chunks_on_out_queue; /* total chunks floating around, * locked by send socket buffer */ uint32_t peers_adaptation; uint32_t default_mtu; uint16_t peer_hmac_id; /* peer HMAC id to send */ /* * Being that we have no bag to collect stale cookies, and that we * really would not want to anyway.. we will count them in this * counter. We of course feed them to the pigeons right away (I have * always thought of pigeons as flying rats). */ uint16_t stale_cookie_count; /* * For the partial delivery API, if up, invoked this is what last * TSN I delivered */ uint16_t str_of_pdapi; uint16_t ssn_of_pdapi; /* counts of actual built streams. Allocation may be more however */ /* could re-arrange to optimize space here. */ uint16_t streamincnt; uint16_t streamoutcnt; uint16_t strm_realoutsize; uint16_t strm_pending_add_size; /* my maximum number of retrans of INIT and SEND */ /* copied from SCTP but should be individually setable */ uint16_t max_init_times; uint16_t max_send_times; uint16_t def_net_failure; uint16_t def_net_pf_threshold; /* * lock flag: 0 is ok to send, 1+ (duals as a retran count) is * awaiting ACK */ uint16_t mapping_array_size; uint16_t last_strm_seq_delivered; uint16_t last_strm_no_delivered; uint16_t last_revoke_count; int16_t num_send_timers_up; uint16_t stream_locked_on; uint16_t ecn_echo_cnt_onq; uint16_t free_chunk_cnt; uint8_t stream_locked; uint8_t authenticated; /* packet authenticated ok */ /* * This flag indicates that a SACK need to be sent. Initially this * is 1 to send the first sACK immediately. */ uint8_t send_sack; /* max burst of new packets into the network */ uint32_t max_burst; /* max burst of fast retransmit packets */ uint32_t fr_max_burst; uint8_t sat_network; /* RTT is in range of sat net or greater */ uint8_t sat_network_lockout; /* lockout code */ uint8_t burst_limit_applied; /* Burst limit in effect at last send? */ /* flag goes on when we are doing a partial delivery api */ uint8_t hb_random_values[4]; uint8_t fragmented_delivery_inprogress; uint8_t fragment_flags; uint8_t last_flags_delivered; uint8_t hb_ect_randombit; uint8_t hb_random_idx; uint8_t default_dscp; uint8_t asconf_del_pending; /* asconf delete last addr pending */ uint8_t trigger_reset; /* * This value, plus all other ack'd but above cum-ack is added * together to cross check against the bit that we have yet to * define (probably in the SACK). When the cum-ack is updated, this * sum is updated as well. */ /* Flags whether an extension is supported or not */ uint8_t ecn_supported; uint8_t prsctp_supported; uint8_t auth_supported; uint8_t asconf_supported; uint8_t reconfig_supported; uint8_t nrsack_supported; uint8_t pktdrop_supported; uint8_t idata_supported; /* Did the peer make the stream config (add out) request */ uint8_t peer_req_out; uint8_t local_strreset_support; uint8_t peer_supports_nat; struct sctp_scoping scope; /* flags to handle send alternate net tracking */ uint8_t used_alt_asconfack; uint8_t fast_retran_loss_recovery; uint8_t sat_t3_loss_recovery; uint8_t dropped_special_cnt; uint8_t seen_a_sack_this_pkt; uint8_t stream_reset_outstanding; uint8_t stream_reset_out_is_outstanding; uint8_t delayed_connection; uint8_t ifp_had_enobuf; uint8_t saw_sack_with_frags; uint8_t saw_sack_with_nr_frags; uint8_t in_asocid_hash; uint8_t assoc_up_sent; uint8_t adaptation_needed; uint8_t adaptation_sent; /* CMT variables */ uint8_t cmt_dac_pkts_rcvd; uint8_t sctp_cmt_on_off; uint8_t iam_blocking; uint8_t cookie_how[8]; /* JRS 5/21/07 - CMT PF variable */ uint8_t sctp_cmt_pf; uint8_t use_precise_time; uint64_t sctp_features; uint32_t max_cwnd; uint16_t port; /* remote UDP encapsulation port */ /* * The mapping array is used to track out of order sequences above * last_acked_seq. 0 indicates packet missing 1 indicates packet * rec'd. We slide it up every time we raise last_acked_seq and 0 * trailing locactions out. If I get a TSN above the array * mappingArraySz, I discard the datagram and let retransmit happen. */ uint32_t marked_retrans; uint32_t timoinit; uint32_t timodata; uint32_t timosack; uint32_t timoshutdown; uint32_t timoheartbeat; uint32_t timocookie; uint32_t timoshutdownack; struct timeval start_time; struct timeval discontinuity_time; uint64_t abandoned_unsent[SCTP_PR_SCTP_MAX + 1]; uint64_t abandoned_sent[SCTP_PR_SCTP_MAX + 1]; }; #endif Index: head/sys/netinet/sctp_timer.c =================================================================== --- head/sys/netinet/sctp_timer.c (revision 360291) +++ head/sys/netinet/sctp_timer.c (revision 360292) @@ -1,1596 +1,1593 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #define _IP_VHL #include #include #ifdef INET6 #endif #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #endif void sctp_audit_retranmission_queue(struct sctp_association *asoc) { struct sctp_tmit_chunk *chk; SCTPDBG(SCTP_DEBUG_TIMER4, "Audit invoked on send queue cnt:%d onqueue:%d\n", asoc->sent_queue_retran_cnt, asoc->sent_queue_cnt); asoc->sent_queue_retran_cnt = 0; asoc->sent_queue_cnt = 0; TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) { if (chk->sent == SCTP_DATAGRAM_RESEND) { sctp_ucount_incr(asoc->sent_queue_retran_cnt); } asoc->sent_queue_cnt++; } TAILQ_FOREACH(chk, &asoc->control_send_queue, sctp_next) { if (chk->sent == SCTP_DATAGRAM_RESEND) { sctp_ucount_incr(asoc->sent_queue_retran_cnt); } } TAILQ_FOREACH(chk, &asoc->asconf_send_queue, sctp_next) { if (chk->sent == SCTP_DATAGRAM_RESEND) { sctp_ucount_incr(asoc->sent_queue_retran_cnt); } } SCTPDBG(SCTP_DEBUG_TIMER4, "Audit completes retran:%d onqueue:%d\n", asoc->sent_queue_retran_cnt, asoc->sent_queue_cnt); } static int sctp_threshold_management(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net, uint16_t threshold) { if (net) { net->error_count++; SCTPDBG(SCTP_DEBUG_TIMER4, "Error count for %p now %d thresh:%d\n", (void *)net, net->error_count, net->failure_threshold); if (net->error_count > net->failure_threshold) { /* We had a threshold failure */ if (net->dest_state & SCTP_ADDR_REACHABLE) { net->dest_state &= ~SCTP_ADDR_REACHABLE; net->dest_state &= ~SCTP_ADDR_REQ_PRIMARY; net->dest_state &= ~SCTP_ADDR_PF; sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_DOWN, stcb, 0, (void *)net, SCTP_SO_NOT_LOCKED); } } else if ((net->pf_threshold < net->failure_threshold) && (net->error_count > net->pf_threshold)) { if (!(net->dest_state & SCTP_ADDR_PF)) { net->dest_state |= SCTP_ADDR_PF; net->last_active = sctp_get_tick_count(); sctp_send_hb(stcb, net, SCTP_SO_NOT_LOCKED); sctp_timer_stop(SCTP_TIMER_TYPE_HEARTBEAT, inp, stcb, net, SCTP_FROM_SCTP_TIMER + SCTP_LOC_1); sctp_timer_start(SCTP_TIMER_TYPE_HEARTBEAT, inp, stcb, net); } } } if (stcb == NULL) return (0); if (net) { if ((net->dest_state & SCTP_ADDR_UNCONFIRMED) == 0) { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) { sctp_misc_ints(SCTP_THRESHOLD_INCR, stcb->asoc.overall_error_count, (stcb->asoc.overall_error_count + 1), SCTP_FROM_SCTP_TIMER, __LINE__); } stcb->asoc.overall_error_count++; } } else { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) { sctp_misc_ints(SCTP_THRESHOLD_INCR, stcb->asoc.overall_error_count, (stcb->asoc.overall_error_count + 1), SCTP_FROM_SCTP_TIMER, __LINE__); } stcb->asoc.overall_error_count++; } SCTPDBG(SCTP_DEBUG_TIMER4, "Overall error count for %p now %d thresh:%u state:%x\n", (void *)&stcb->asoc, stcb->asoc.overall_error_count, (uint32_t)threshold, ((net == NULL) ? (uint32_t)0 : (uint32_t)net->dest_state)); /* * We specifically do not do >= to give the assoc one more change * before we fail it. */ if (stcb->asoc.overall_error_count > threshold) { /* Abort notification sends a ULP notify */ struct mbuf *op_err; op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), "Association error counter exceeded"); inp->last_abort_code = SCTP_FROM_SCTP_TIMER + SCTP_LOC_2; sctp_abort_an_association(inp, stcb, op_err, SCTP_SO_NOT_LOCKED); return (1); } return (0); } /* * sctp_find_alternate_net() returns a non-NULL pointer as long * the argument net is non-NULL. */ struct sctp_nets * sctp_find_alternate_net(struct sctp_tcb *stcb, struct sctp_nets *net, int mode) { /* Find and return an alternate network if possible */ struct sctp_nets *alt, *mnet, *min_errors_net = NULL, *max_cwnd_net = NULL; int once; /* JRS 5/14/07 - Initialize min_errors to an impossible value. */ int min_errors = -1; uint32_t max_cwnd = 0; if (stcb->asoc.numnets == 1) { /* No others but net */ return (TAILQ_FIRST(&stcb->asoc.nets)); } /* * JRS 5/14/07 - If mode is set to 2, use the CMT PF find alternate * net algorithm. This algorithm chooses the active destination (not * in PF state) with the largest cwnd value. If all destinations are * in PF state, unreachable, or unconfirmed, choose the desination * that is in PF state with the lowest error count. In case of a * tie, choose the destination that was most recently active. */ if (mode == 2) { TAILQ_FOREACH(mnet, &stcb->asoc.nets, sctp_next) { /* * JRS 5/14/07 - If the destination is unreachable * or unconfirmed, skip it. */ if (((mnet->dest_state & SCTP_ADDR_REACHABLE) != SCTP_ADDR_REACHABLE) || (mnet->dest_state & SCTP_ADDR_UNCONFIRMED)) { continue; } /* * JRS 5/14/07 - If the destination is reachable * but in PF state, compare the error count of the * destination to the minimum error count seen thus * far. Store the destination with the lower error * count. If the error counts are equal, store the * destination that was most recently active. */ if (mnet->dest_state & SCTP_ADDR_PF) { /* * JRS 5/14/07 - If the destination under * consideration is the current destination, * work as if the error count is one higher. * The actual error count will not be * incremented until later in the t3 * handler. */ if (mnet == net) { if (min_errors == -1) { min_errors = mnet->error_count + 1; min_errors_net = mnet; } else if (mnet->error_count + 1 < min_errors) { min_errors = mnet->error_count + 1; min_errors_net = mnet; } else if (mnet->error_count + 1 == min_errors && mnet->last_active > min_errors_net->last_active) { min_errors_net = mnet; min_errors = mnet->error_count + 1; } continue; } else { if (min_errors == -1) { min_errors = mnet->error_count; min_errors_net = mnet; } else if (mnet->error_count < min_errors) { min_errors = mnet->error_count; min_errors_net = mnet; } else if (mnet->error_count == min_errors && mnet->last_active > min_errors_net->last_active) { min_errors_net = mnet; min_errors = mnet->error_count; } continue; } } /* * JRS 5/14/07 - If the destination is reachable and * not in PF state, compare the cwnd of the * destination to the highest cwnd seen thus far. * Store the destination with the higher cwnd value. * If the cwnd values are equal, randomly choose one * of the two destinations. */ if (max_cwnd < mnet->cwnd) { max_cwnd_net = mnet; max_cwnd = mnet->cwnd; } else if (max_cwnd == mnet->cwnd) { uint32_t rndval; uint8_t this_random; if (stcb->asoc.hb_random_idx > 3) { rndval = sctp_select_initial_TSN(&stcb->sctp_ep->sctp_ep); memcpy(stcb->asoc.hb_random_values, &rndval, sizeof(stcb->asoc.hb_random_values)); this_random = stcb->asoc.hb_random_values[0]; stcb->asoc.hb_random_idx++; stcb->asoc.hb_ect_randombit = 0; } else { this_random = stcb->asoc.hb_random_values[stcb->asoc.hb_random_idx]; stcb->asoc.hb_random_idx++; stcb->asoc.hb_ect_randombit = 0; } if (this_random % 2 == 1) { max_cwnd_net = mnet; max_cwnd = mnet->cwnd; /* Useless? */ } } } if (max_cwnd_net == NULL) { if (min_errors_net == NULL) { return (net); } return (min_errors_net); } else { return (max_cwnd_net); } } /* JRS 5/14/07 - If mode is set to 1, use the * CMT policy for choosing an alternate net. */ else if (mode == 1) { TAILQ_FOREACH(mnet, &stcb->asoc.nets, sctp_next) { if (((mnet->dest_state & SCTP_ADDR_REACHABLE) != SCTP_ADDR_REACHABLE) || (mnet->dest_state & SCTP_ADDR_UNCONFIRMED)) { /* * will skip ones that are not-reachable or * unconfirmed */ continue; } if (max_cwnd < mnet->cwnd) { max_cwnd_net = mnet; max_cwnd = mnet->cwnd; } else if (max_cwnd == mnet->cwnd) { uint32_t rndval; uint8_t this_random; if (stcb->asoc.hb_random_idx > 3) { rndval = sctp_select_initial_TSN(&stcb->sctp_ep->sctp_ep); memcpy(stcb->asoc.hb_random_values, &rndval, sizeof(stcb->asoc.hb_random_values)); this_random = stcb->asoc.hb_random_values[0]; stcb->asoc.hb_random_idx = 0; stcb->asoc.hb_ect_randombit = 0; } else { this_random = stcb->asoc.hb_random_values[stcb->asoc.hb_random_idx]; stcb->asoc.hb_random_idx++; stcb->asoc.hb_ect_randombit = 0; } if (this_random % 2) { max_cwnd_net = mnet; max_cwnd = mnet->cwnd; } } } if (max_cwnd_net) { return (max_cwnd_net); } } mnet = net; once = 0; if (mnet == NULL) { mnet = TAILQ_FIRST(&stcb->asoc.nets); if (mnet == NULL) { return (NULL); } } for (;;) { alt = TAILQ_NEXT(mnet, sctp_next); if (alt == NULL) { once++; if (once > 1) { break; } alt = TAILQ_FIRST(&stcb->asoc.nets); if (alt == NULL) { return (NULL); } } - if (alt->ro.ro_rt == NULL) { + if (alt->ro.ro_nh == NULL) { if (alt->ro._s_addr) { sctp_free_ifa(alt->ro._s_addr); alt->ro._s_addr = NULL; } alt->src_addr_selected = 0; } if (((alt->dest_state & SCTP_ADDR_REACHABLE) == SCTP_ADDR_REACHABLE) && - (alt->ro.ro_rt != NULL) && + (alt->ro.ro_nh != NULL) && (!(alt->dest_state & SCTP_ADDR_UNCONFIRMED))) { /* Found a reachable address */ break; } mnet = alt; } if (alt == NULL) { /* Case where NO insv network exists (dormant state) */ /* we rotate destinations */ once = 0; mnet = net; for (;;) { if (mnet == NULL) { return (TAILQ_FIRST(&stcb->asoc.nets)); } alt = TAILQ_NEXT(mnet, sctp_next); if (alt == NULL) { once++; if (once > 1) { break; } alt = TAILQ_FIRST(&stcb->asoc.nets); if (alt == NULL) { break; } } if ((!(alt->dest_state & SCTP_ADDR_UNCONFIRMED)) && (alt != net)) { /* Found an alternate address */ break; } mnet = alt; } } if (alt == NULL) { return (net); } return (alt); } static void sctp_backoff_on_timeout(struct sctp_tcb *stcb, struct sctp_nets *net, int win_probe, int num_marked, int num_abandoned) { if (net->RTO == 0) { if (net->RTO_measured) { net->RTO = stcb->asoc.minrto; } else { net->RTO = stcb->asoc.initial_rto; } } net->RTO <<= 1; if (net->RTO > stcb->asoc.maxrto) { net->RTO = stcb->asoc.maxrto; } if ((win_probe == 0) && (num_marked || num_abandoned)) { /* We don't apply penalty to window probe scenarios */ /* JRS - Use the congestion control given in the CC module */ stcb->asoc.cc_functions.sctp_cwnd_update_after_timeout(stcb, net); } } #ifndef INVARIANTS static void sctp_recover_sent_list(struct sctp_tcb *stcb) { struct sctp_tmit_chunk *chk, *nchk; struct sctp_association *asoc; asoc = &stcb->asoc; TAILQ_FOREACH_SAFE(chk, &asoc->sent_queue, sctp_next, nchk) { if (SCTP_TSN_GE(asoc->last_acked_seq, chk->rec.data.tsn)) { SCTP_PRINTF("Found chk:%p tsn:%x <= last_acked_seq:%x\n", (void *)chk, chk->rec.data.tsn, asoc->last_acked_seq); if (chk->sent != SCTP_DATAGRAM_NR_ACKED) { if (asoc->strmout[chk->rec.data.sid].chunks_on_queues > 0) { asoc->strmout[chk->rec.data.sid].chunks_on_queues--; } } if ((asoc->strmout[chk->rec.data.sid].chunks_on_queues == 0) && (asoc->strmout[chk->rec.data.sid].state == SCTP_STREAM_RESET_PENDING) && TAILQ_EMPTY(&asoc->strmout[chk->rec.data.sid].outqueue)) { asoc->trigger_reset = 1; } TAILQ_REMOVE(&asoc->sent_queue, chk, sctp_next); if (PR_SCTP_ENABLED(chk->flags)) { if (asoc->pr_sctp_cnt != 0) asoc->pr_sctp_cnt--; } if (chk->data) { /* sa_ignore NO_NULL_CHK */ sctp_free_bufspace(stcb, asoc, chk, 1); sctp_m_freem(chk->data); chk->data = NULL; if (asoc->prsctp_supported && PR_SCTP_BUF_ENABLED(chk->flags)) { asoc->sent_queue_cnt_removeable--; } } asoc->sent_queue_cnt--; sctp_free_a_chunk(stcb, chk, SCTP_SO_NOT_LOCKED); } } SCTP_PRINTF("after recover order is as follows\n"); TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) { SCTP_PRINTF("chk:%p TSN:%x\n", (void *)chk, chk->rec.data.tsn); } } #endif static int sctp_mark_all_for_resend(struct sctp_tcb *stcb, struct sctp_nets *net, struct sctp_nets *alt, int window_probe, int *num_marked, int *num_abandoned) { /* * Mark all chunks (well not all) that were sent to *net for * retransmission. Move them to alt for there destination as well... * We only mark chunks that have been outstanding long enough to * have received feed-back. */ struct sctp_tmit_chunk *chk, *nchk; struct sctp_nets *lnets; struct timeval now, min_wait, tv; int cur_rto; int cnt_abandoned; int audit_tf, num_mk, fir; unsigned int cnt_mk; uint32_t orig_flight, orig_tf; uint32_t tsnlast, tsnfirst; int recovery_cnt = 0; /* none in flight now */ audit_tf = 0; fir = 0; /* * figure out how long a data chunk must be pending before we can * mark it .. */ (void)SCTP_GETTIME_TIMEVAL(&now); /* get cur rto in micro-seconds */ cur_rto = (net->lastsa >> SCTP_RTT_SHIFT) + net->lastsv; cur_rto *= 1000; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(cur_rto, stcb->asoc.peers_rwnd, window_probe, SCTP_FR_T3_MARK_TIME); sctp_log_fr(net->flight_size, 0, 0, SCTP_FR_CWND_REPORT); sctp_log_fr(net->flight_size, net->cwnd, stcb->asoc.total_flight, SCTP_FR_CWND_REPORT); } tv.tv_sec = cur_rto / 1000000; tv.tv_usec = cur_rto % 1000000; min_wait = now; timevalsub(&min_wait, &tv); if (min_wait.tv_sec < 0 || min_wait.tv_usec < 0) { /* * if we hit here, we don't have enough seconds on the clock * to account for the RTO. We just let the lower seconds be * the bounds and don't worry about it. This may mean we * will mark a lot more than we should. */ min_wait.tv_sec = min_wait.tv_usec = 0; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(cur_rto, now.tv_sec, now.tv_usec, SCTP_FR_T3_MARK_TIME); sctp_log_fr(0, min_wait.tv_sec, min_wait.tv_usec, SCTP_FR_T3_MARK_TIME); } /* * Our rwnd will be incorrect here since we are not adding back the * cnt * mbuf but we will fix that down below. */ orig_flight = net->flight_size; orig_tf = stcb->asoc.total_flight; net->fast_retran_ip = 0; /* Now on to each chunk */ cnt_abandoned = 0; num_mk = cnt_mk = 0; tsnfirst = tsnlast = 0; #ifndef INVARIANTS start_again: #endif TAILQ_FOREACH_SAFE(chk, &stcb->asoc.sent_queue, sctp_next, nchk) { if (SCTP_TSN_GE(stcb->asoc.last_acked_seq, chk->rec.data.tsn)) { /* Strange case our list got out of order? */ SCTP_PRINTF("Our list is out of order? last_acked:%x chk:%x\n", (unsigned int)stcb->asoc.last_acked_seq, (unsigned int)chk->rec.data.tsn); recovery_cnt++; #ifdef INVARIANTS panic("last acked >= chk on sent-Q"); #else SCTP_PRINTF("Recover attempts a restart cnt:%d\n", recovery_cnt); sctp_recover_sent_list(stcb); if (recovery_cnt < 10) { goto start_again; } else { SCTP_PRINTF("Recovery fails %d times??\n", recovery_cnt); } #endif } if ((chk->whoTo == net) && (chk->sent < SCTP_DATAGRAM_ACKED)) { /* * found one to mark: If it is less than * DATAGRAM_ACKED it MUST not be a skipped or marked * TSN but instead one that is either already set * for retransmission OR one that needs * retransmission. */ /* validate its been outstanding long enough */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(chk->rec.data.tsn, chk->sent_rcv_time.tv_sec, chk->sent_rcv_time.tv_usec, SCTP_FR_T3_MARK_TIME); } if ((chk->sent_rcv_time.tv_sec > min_wait.tv_sec) && (window_probe == 0)) { /* * we have reached a chunk that was sent * some seconds past our min.. forget it we * will find no more to send. */ if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(0, chk->sent_rcv_time.tv_sec, chk->sent_rcv_time.tv_usec, SCTP_FR_T3_STOPPED); } continue; } else if ((chk->sent_rcv_time.tv_sec == min_wait.tv_sec) && (window_probe == 0)) { /* * we must look at the micro seconds to * know. */ if (chk->sent_rcv_time.tv_usec >= min_wait.tv_usec) { /* * ok it was sent after our boundary * time. */ continue; } } if (stcb->asoc.prsctp_supported && PR_SCTP_TTL_ENABLED(chk->flags)) { /* Is it expired? */ if (timevalcmp(&now, &chk->rec.data.timetodrop, >)) { /* Yes so drop it */ if (chk->data) { (void)sctp_release_pr_sctp_chunk(stcb, chk, 1, SCTP_SO_NOT_LOCKED); cnt_abandoned++; } continue; } } if (stcb->asoc.prsctp_supported && PR_SCTP_RTX_ENABLED(chk->flags)) { /* Has it been retransmitted tv_sec times? */ if (chk->snd_count > chk->rec.data.timetodrop.tv_sec) { if (chk->data) { (void)sctp_release_pr_sctp_chunk(stcb, chk, 1, SCTP_SO_NOT_LOCKED); cnt_abandoned++; } continue; } } if (chk->sent < SCTP_DATAGRAM_RESEND) { sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); num_mk++; if (fir == 0) { fir = 1; tsnfirst = chk->rec.data.tsn; } tsnlast = chk->rec.data.tsn; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(chk->rec.data.tsn, chk->snd_count, 0, SCTP_FR_T3_MARKED); } if (chk->rec.data.chunk_was_revoked) { /* deflate the cwnd */ chk->whoTo->cwnd -= chk->book_size; chk->rec.data.chunk_was_revoked = 0; } net->marked_retrans++; stcb->asoc.marked_retrans++; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) { sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_RSND_TO, chk->whoTo->flight_size, chk->book_size, (uint32_t)(uintptr_t)chk->whoTo, chk->rec.data.tsn); } sctp_flight_size_decrease(chk); sctp_total_flight_decrease(stcb, chk); stcb->asoc.peers_rwnd += chk->send_size; stcb->asoc.peers_rwnd += SCTP_BASE_SYSCTL(sctp_peer_chunk_oh); } chk->sent = SCTP_DATAGRAM_RESEND; chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; SCTP_STAT_INCR(sctps_markedretrans); /* reset the TSN for striking and other FR stuff */ chk->rec.data.doing_fast_retransmit = 0; /* Clear any time so NO RTT is being done */ if (chk->do_rtt) { if (chk->whoTo->rto_needed == 0) { chk->whoTo->rto_needed = 1; } } chk->do_rtt = 0; if (alt != net) { sctp_free_remote_addr(chk->whoTo); chk->no_fr_allowed = 1; chk->whoTo = alt; atomic_add_int(&alt->ref_count, 1); } else { chk->no_fr_allowed = 0; if (TAILQ_EMPTY(&stcb->asoc.send_queue)) { chk->rec.data.fast_retran_tsn = stcb->asoc.sending_seq; } else { chk->rec.data.fast_retran_tsn = (TAILQ_FIRST(&stcb->asoc.send_queue))->rec.data.tsn; } } /* * CMT: Do not allow FRs on retransmitted TSNs. */ if (stcb->asoc.sctp_cmt_on_off > 0) { chk->no_fr_allowed = 1; } #ifdef THIS_SHOULD_NOT_BE_DONE } else if (chk->sent == SCTP_DATAGRAM_ACKED) { /* remember highest acked one */ could_be_sent = chk; #endif } if (chk->sent == SCTP_DATAGRAM_RESEND) { cnt_mk++; } } if ((orig_flight - net->flight_size) != (orig_tf - stcb->asoc.total_flight)) { /* we did not subtract the same things? */ audit_tf = 1; } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(tsnfirst, tsnlast, num_mk, SCTP_FR_T3_TIMEOUT); } #ifdef SCTP_DEBUG if (num_mk) { SCTPDBG(SCTP_DEBUG_TIMER1, "LAST TSN marked was %x\n", tsnlast); SCTPDBG(SCTP_DEBUG_TIMER1, "Num marked for retransmission was %d peer-rwd:%u\n", num_mk, stcb->asoc.peers_rwnd); } #endif *num_marked = num_mk; *num_abandoned = cnt_abandoned; /* * Now check for a ECN Echo that may be stranded And include the * cnt_mk'd to have all resends in the control queue. */ TAILQ_FOREACH(chk, &stcb->asoc.control_send_queue, sctp_next) { if (chk->sent == SCTP_DATAGRAM_RESEND) { cnt_mk++; } if ((chk->whoTo == net) && (chk->rec.chunk_id.id == SCTP_ECN_ECHO)) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = alt; if (chk->sent != SCTP_DATAGRAM_RESEND) { chk->sent = SCTP_DATAGRAM_RESEND; chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); cnt_mk++; } atomic_add_int(&alt->ref_count, 1); } } #ifdef THIS_SHOULD_NOT_BE_DONE if ((stcb->asoc.sent_queue_retran_cnt == 0) && (could_be_sent)) { /* fix it so we retransmit the highest acked anyway */ sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); cnt_mk++; could_be_sent->sent = SCTP_DATAGRAM_RESEND; } #endif if (stcb->asoc.sent_queue_retran_cnt != cnt_mk) { #ifdef INVARIANTS SCTP_PRINTF("Local Audit says there are %d for retran asoc cnt:%d we marked:%d this time\n", cnt_mk, stcb->asoc.sent_queue_retran_cnt, num_mk); #endif #ifndef SCTP_AUDITING_ENABLED stcb->asoc.sent_queue_retran_cnt = cnt_mk; #endif } if (audit_tf) { SCTPDBG(SCTP_DEBUG_TIMER4, "Audit total flight due to negative value net:%p\n", (void *)net); stcb->asoc.total_flight = 0; stcb->asoc.total_flight_count = 0; /* Clear all networks flight size */ TAILQ_FOREACH(lnets, &stcb->asoc.nets, sctp_next) { lnets->flight_size = 0; SCTPDBG(SCTP_DEBUG_TIMER4, "Net:%p c-f cwnd:%d ssthresh:%d\n", (void *)lnets, lnets->cwnd, lnets->ssthresh); } TAILQ_FOREACH(chk, &stcb->asoc.sent_queue, sctp_next) { if (chk->sent < SCTP_DATAGRAM_RESEND) { if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) { sctp_misc_ints(SCTP_FLIGHT_LOG_UP, chk->whoTo->flight_size, chk->book_size, (uint32_t)(uintptr_t)chk->whoTo, chk->rec.data.tsn); } sctp_flight_size_increase(chk); sctp_total_flight_increase(stcb, chk); } } } /* We return 1 if we only have a window probe outstanding */ return (0); } int sctp_t3rxt_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_nets *alt; int win_probe, num_mk, num_abandoned; if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) { sctp_log_fr(0, 0, 0, SCTP_FR_T3_TIMEOUT); } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) { struct sctp_nets *lnet; TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) { if (net == lnet) { sctp_log_cwnd(stcb, lnet, 1, SCTP_CWND_LOG_FROM_T3); } else { sctp_log_cwnd(stcb, lnet, 0, SCTP_CWND_LOG_FROM_T3); } } } /* Find an alternate and mark those for retransmission */ if ((stcb->asoc.peers_rwnd == 0) && (stcb->asoc.total_flight < net->mtu)) { SCTP_STAT_INCR(sctps_timowindowprobe); win_probe = 1; } else { win_probe = 0; } if (win_probe == 0) { /* We don't do normal threshold management on window probes */ if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Association was destroyed */ return (1); } else { if (net != stcb->asoc.primary_destination) { /* send a immediate HB if our RTO is stale */ struct timeval now; unsigned int ms_goneby; (void)SCTP_GETTIME_TIMEVAL(&now); if (net->last_sent_time.tv_sec) { ms_goneby = (now.tv_sec - net->last_sent_time.tv_sec) * 1000; } else { ms_goneby = 0; } if ((net->dest_state & SCTP_ADDR_PF) == 0) { if ((ms_goneby > net->RTO) || (net->RTO == 0)) { /* * no recent feed back in an * RTO or more, request a * RTT update */ sctp_send_hb(stcb, net, SCTP_SO_NOT_LOCKED); } } } } } else { /* * For a window probe we don't penalize the net's but only * the association. This may fail it if SACKs are not coming * back. If sack's are coming with rwnd locked at 0, we will * continue to hold things waiting for rwnd to raise */ if (sctp_threshold_management(inp, stcb, NULL, stcb->asoc.max_send_times)) { /* Association was destroyed */ return (1); } } if (stcb->asoc.sctp_cmt_on_off > 0) { if (net->pf_threshold < net->failure_threshold) { alt = sctp_find_alternate_net(stcb, net, 2); } else { /* * CMT: Using RTX_SSTHRESH policy for CMT. If CMT is * being used, then pick dest with largest ssthresh * for any retransmission. */ alt = sctp_find_alternate_net(stcb, net, 1); /* * CUCv2: If a different dest is picked for the * retransmission, then new (rtx-)pseudo_cumack * needs to be tracked for orig dest. Let CUCv2 * track new (rtx-) pseudo-cumack always. */ net->find_pseudo_cumack = 1; net->find_rtx_pseudo_cumack = 1; } } else { alt = sctp_find_alternate_net(stcb, net, 0); } num_mk = 0; num_abandoned = 0; (void)sctp_mark_all_for_resend(stcb, net, alt, win_probe, &num_mk, &num_abandoned); /* FR Loss recovery just ended with the T3. */ stcb->asoc.fast_retran_loss_recovery = 0; /* CMT FR loss recovery ended with the T3 */ net->fast_retran_loss_recovery = 0; if ((stcb->asoc.cc_functions.sctp_cwnd_new_transmission_begins) && (net->flight_size == 0)) { (*stcb->asoc.cc_functions.sctp_cwnd_new_transmission_begins) (stcb, net); } /* * setup the sat loss recovery that prevents satellite cwnd advance. */ stcb->asoc.sat_t3_loss_recovery = 1; stcb->asoc.sat_t3_recovery_tsn = stcb->asoc.sending_seq; /* Backoff the timer and cwnd */ sctp_backoff_on_timeout(stcb, net, win_probe, num_mk, num_abandoned); if ((!(net->dest_state & SCTP_ADDR_REACHABLE)) || (net->dest_state & SCTP_ADDR_PF)) { /* Move all pending over too */ sctp_move_chunks_from_net(stcb, net); /* * Get the address that failed, to force a new src address * selecton and a route allocation. */ if (net->ro._s_addr) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; } net->src_addr_selected = 0; /* Force a route allocation too */ - if (net->ro.ro_rt) { - RTFREE(net->ro.ro_rt); - net->ro.ro_rt = NULL; - } + RO_NHFREE(&net->ro); /* Was it our primary? */ if ((stcb->asoc.primary_destination == net) && (alt != net)) { /* * Yes, note it as such and find an alternate note: * this means HB code must use this to resent the * primary if it goes active AND if someone does a * change-primary then this flag must be cleared * from any net structures. */ if (stcb->asoc.alternate) { sctp_free_remote_addr(stcb->asoc.alternate); } stcb->asoc.alternate = alt; atomic_add_int(&stcb->asoc.alternate->ref_count, 1); } } /* * Special case for cookie-echo'ed case, we don't do output but must * await the COOKIE-ACK before retransmission */ if (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED) { /* * Here we just reset the timer and start again since we * have not established the asoc */ sctp_timer_start(SCTP_TIMER_TYPE_SEND, inp, stcb, net); return (0); } if (stcb->asoc.prsctp_supported) { struct sctp_tmit_chunk *lchk; lchk = sctp_try_advance_peer_ack_point(stcb, &stcb->asoc); /* C3. See if we need to send a Fwd-TSN */ if (SCTP_TSN_GT(stcb->asoc.advanced_peer_ack_point, stcb->asoc.last_acked_seq)) { send_forward_tsn(stcb, &stcb->asoc); if (lchk) { /* Assure a timer is up */ sctp_timer_start(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep, stcb, lchk->whoTo); } } } if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) { sctp_log_cwnd(stcb, net, net->cwnd, SCTP_CWND_LOG_FROM_RTX); } return (0); } int sctp_t1init_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { /* bump the thresholds */ if (stcb->asoc.delayed_connection) { /* * special hook for delayed connection. The library did NOT * complete the rest of its sends. */ stcb->asoc.delayed_connection = 0; sctp_send_initiate(inp, stcb, SCTP_SO_NOT_LOCKED); return (0); } if (SCTP_GET_STATE(stcb) != SCTP_STATE_COOKIE_WAIT) { return (0); } if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_init_times)) { /* Association was destroyed */ return (1); } stcb->asoc.dropped_special_cnt = 0; sctp_backoff_on_timeout(stcb, stcb->asoc.primary_destination, 1, 0, 0); if (stcb->asoc.initial_init_rto_max < net->RTO) { net->RTO = stcb->asoc.initial_init_rto_max; } if (stcb->asoc.numnets > 1) { /* If we have more than one addr use it */ struct sctp_nets *alt; alt = sctp_find_alternate_net(stcb, stcb->asoc.primary_destination, 0); if (alt != stcb->asoc.primary_destination) { sctp_move_chunks_from_net(stcb, stcb->asoc.primary_destination); stcb->asoc.primary_destination = alt; } } /* Send out a new init */ sctp_send_initiate(inp, stcb, SCTP_SO_NOT_LOCKED); return (0); } /* * For cookie and asconf we actually need to find and mark for resend, then * increment the resend counter (after all the threshold management stuff of * course). */ int sctp_cookie_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net SCTP_UNUSED) { struct sctp_nets *alt; struct sctp_tmit_chunk *cookie; /* first before all else we must find the cookie */ TAILQ_FOREACH(cookie, &stcb->asoc.control_send_queue, sctp_next) { if (cookie->rec.chunk_id.id == SCTP_COOKIE_ECHO) { break; } } if (cookie == NULL) { if (SCTP_GET_STATE(stcb) == SCTP_STATE_COOKIE_ECHOED) { /* FOOBAR! */ struct mbuf *op_err; op_err = sctp_generate_cause(SCTP_BASE_SYSCTL(sctp_diag_info_code), "Cookie timer expired, but no cookie"); inp->last_abort_code = SCTP_FROM_SCTP_TIMER + SCTP_LOC_3; sctp_abort_an_association(inp, stcb, op_err, SCTP_SO_NOT_LOCKED); } else { #ifdef INVARIANTS panic("Cookie timer expires in wrong state?"); #else SCTP_PRINTF("Strange in state %d not cookie-echoed yet c-e timer expires?\n", SCTP_GET_STATE(stcb)); return (0); #endif } return (0); } /* Ok we found the cookie, threshold management next */ if (sctp_threshold_management(inp, stcb, cookie->whoTo, stcb->asoc.max_init_times)) { /* Assoc is over */ return (1); } /* * Cleared threshold management, now lets backoff the address and * select an alternate */ stcb->asoc.dropped_special_cnt = 0; sctp_backoff_on_timeout(stcb, cookie->whoTo, 1, 0, 0); alt = sctp_find_alternate_net(stcb, cookie->whoTo, 0); if (alt != cookie->whoTo) { sctp_free_remote_addr(cookie->whoTo); cookie->whoTo = alt; atomic_add_int(&alt->ref_count, 1); } /* Now mark the retran info */ if (cookie->sent != SCTP_DATAGRAM_RESEND) { sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); } cookie->sent = SCTP_DATAGRAM_RESEND; cookie->flags |= CHUNK_FLAGS_FRAGMENT_OK; /* * Now call the output routine to kick out the cookie again, Note we * don't mark any chunks for retran so that FR will need to kick in * to move these (or a send timer). */ return (0); } int sctp_strreset_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb) { struct sctp_nets *alt, *net; struct sctp_tmit_chunk *strrst = NULL, *chk = NULL; if (stcb->asoc.stream_reset_outstanding == 0) { return (0); } /* find the existing STRRESET, we use the seq number we sent out on */ (void)sctp_find_stream_reset(stcb, stcb->asoc.str_reset_seq_out, &strrst); if (strrst == NULL) { return (0); } net = strrst->whoTo; /* do threshold management */ if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Assoc is over */ return (1); } /* * Cleared threshold management, now lets backoff the address and * select an alternate */ sctp_backoff_on_timeout(stcb, net, 1, 0, 0); alt = sctp_find_alternate_net(stcb, net, 0); strrst->whoTo = alt; atomic_add_int(&alt->ref_count, 1); /* See if a ECN Echo is also stranded */ TAILQ_FOREACH(chk, &stcb->asoc.control_send_queue, sctp_next) { if ((chk->whoTo == net) && (chk->rec.chunk_id.id == SCTP_ECN_ECHO)) { sctp_free_remote_addr(chk->whoTo); if (chk->sent != SCTP_DATAGRAM_RESEND) { chk->sent = SCTP_DATAGRAM_RESEND; chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); } chk->whoTo = alt; atomic_add_int(&alt->ref_count, 1); } } if (!(net->dest_state & SCTP_ADDR_REACHABLE)) { /* * If the address went un-reachable, we need to move to * alternates for ALL chk's in queue */ sctp_move_chunks_from_net(stcb, net); } sctp_free_remote_addr(net); /* mark the retran info */ if (strrst->sent != SCTP_DATAGRAM_RESEND) sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); strrst->sent = SCTP_DATAGRAM_RESEND; strrst->flags |= CHUNK_FLAGS_FRAGMENT_OK; /* restart the timer */ sctp_timer_start(SCTP_TIMER_TYPE_STRRESET, inp, stcb, alt); return (0); } int sctp_asconf_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_nets *alt; struct sctp_tmit_chunk *asconf, *chk; /* is this a first send, or a retransmission? */ if (TAILQ_EMPTY(&stcb->asoc.asconf_send_queue)) { /* compose a new ASCONF chunk and send it */ sctp_send_asconf(stcb, net, SCTP_ADDR_NOT_LOCKED); } else { /* * Retransmission of the existing ASCONF is needed */ /* find the existing ASCONF */ asconf = TAILQ_FIRST(&stcb->asoc.asconf_send_queue); if (asconf == NULL) { return (0); } net = asconf->whoTo; /* do threshold management */ if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Assoc is over */ return (1); } if (asconf->snd_count > stcb->asoc.max_send_times) { /* * Something is rotten: our peer is not responding * to ASCONFs but apparently is to other chunks. * i.e. it is not properly handling the chunk type * upper bits. Mark this peer as ASCONF incapable * and cleanup. */ SCTPDBG(SCTP_DEBUG_TIMER1, "asconf_timer: Peer has not responded to our repeated ASCONFs\n"); sctp_asconf_cleanup(stcb); return (0); } /* * cleared threshold management, so now backoff the net and * select an alternate */ sctp_backoff_on_timeout(stcb, net, 1, 0, 0); alt = sctp_find_alternate_net(stcb, net, 0); if (asconf->whoTo != alt) { asconf->whoTo = alt; atomic_add_int(&alt->ref_count, 1); } /* See if an ECN Echo is also stranded */ TAILQ_FOREACH(chk, &stcb->asoc.control_send_queue, sctp_next) { if ((chk->whoTo == net) && (chk->rec.chunk_id.id == SCTP_ECN_ECHO)) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = alt; if (chk->sent != SCTP_DATAGRAM_RESEND) { chk->sent = SCTP_DATAGRAM_RESEND; chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); } atomic_add_int(&alt->ref_count, 1); } } TAILQ_FOREACH(chk, &stcb->asoc.asconf_send_queue, sctp_next) { if (chk->whoTo != alt) { sctp_free_remote_addr(chk->whoTo); chk->whoTo = alt; atomic_add_int(&alt->ref_count, 1); } if (asconf->sent != SCTP_DATAGRAM_RESEND && chk->sent != SCTP_DATAGRAM_UNSENT) sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); chk->sent = SCTP_DATAGRAM_RESEND; chk->flags |= CHUNK_FLAGS_FRAGMENT_OK; } if (!(net->dest_state & SCTP_ADDR_REACHABLE)) { /* * If the address went un-reachable, we need to move * to the alternate for ALL chunks in queue */ sctp_move_chunks_from_net(stcb, net); } sctp_free_remote_addr(net); /* mark the retran info */ if (asconf->sent != SCTP_DATAGRAM_RESEND) sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt); asconf->sent = SCTP_DATAGRAM_RESEND; asconf->flags |= CHUNK_FLAGS_FRAGMENT_OK; /* send another ASCONF if any and we can do */ sctp_send_asconf(stcb, alt, SCTP_ADDR_NOT_LOCKED); } return (0); } /* Mobility adaptation */ void sctp_delete_prim_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb) { if (stcb->asoc.deleted_primary == NULL) { SCTPDBG(SCTP_DEBUG_ASCONF1, "delete_prim_timer: deleted_primary is not stored...\n"); sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); return; } SCTPDBG(SCTP_DEBUG_ASCONF1, "delete_prim_timer: finished to keep deleted primary "); SCTPDBG_ADDR(SCTP_DEBUG_ASCONF1, &stcb->asoc.deleted_primary->ro._l_addr.sa); sctp_free_remote_addr(stcb->asoc.deleted_primary); stcb->asoc.deleted_primary = NULL; sctp_mobility_feature_off(inp, SCTP_MOBILITY_PRIM_DELETED); return; } /* * For the shutdown and shutdown-ack, we do not keep one around on the * control queue. This means we must generate a new one and call the general * chunk output routine, AFTER having done threshold management. * It is assumed that net is non-NULL. */ int sctp_shutdown_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_nets *alt; /* first threshold management */ if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Assoc is over */ return (1); } sctp_backoff_on_timeout(stcb, net, 1, 0, 0); /* second select an alternative */ alt = sctp_find_alternate_net(stcb, net, 0); /* third generate a shutdown into the queue for out net */ sctp_send_shutdown(stcb, alt); /* fourth restart timer */ sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN, inp, stcb, alt); return (0); } int sctp_shutdownack_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { struct sctp_nets *alt; /* first threshold management */ if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Assoc is over */ return (1); } sctp_backoff_on_timeout(stcb, net, 1, 0, 0); /* second select an alternative */ alt = sctp_find_alternate_net(stcb, net, 0); /* third generate a shutdown into the queue for out net */ sctp_send_shutdown_ack(stcb, alt); /* fourth restart timer */ sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNACK, inp, stcb, alt); return (0); } static void sctp_audit_stream_queues_for_size(struct sctp_inpcb *inp, struct sctp_tcb *stcb) { struct sctp_stream_queue_pending *sp; unsigned int i, chks_in_queue = 0; int being_filled = 0; /* * This function is ONLY called when the send/sent queues are empty. */ if ((stcb == NULL) || (inp == NULL)) return; if (stcb->asoc.sent_queue_retran_cnt) { SCTP_PRINTF("Hmm, sent_queue_retran_cnt is non-zero %d\n", stcb->asoc.sent_queue_retran_cnt); stcb->asoc.sent_queue_retran_cnt = 0; } if (stcb->asoc.ss_functions.sctp_ss_is_empty(stcb, &stcb->asoc)) { /* No stream scheduler information, initialize scheduler */ stcb->asoc.ss_functions.sctp_ss_init(stcb, &stcb->asoc, 0); if (!stcb->asoc.ss_functions.sctp_ss_is_empty(stcb, &stcb->asoc)) { /* yep, we lost a stream or two */ SCTP_PRINTF("Found additional streams NOT managed by scheduler, corrected\n"); } else { /* no streams lost */ stcb->asoc.total_output_queue_size = 0; } } /* Check to see if some data queued, if so report it */ for (i = 0; i < stcb->asoc.streamoutcnt; i++) { if (!TAILQ_EMPTY(&stcb->asoc.strmout[i].outqueue)) { TAILQ_FOREACH(sp, &stcb->asoc.strmout[i].outqueue, next) { if (sp->msg_is_complete) being_filled++; chks_in_queue++; } } } if (chks_in_queue != stcb->asoc.stream_queue_cnt) { SCTP_PRINTF("Hmm, stream queue cnt at %d I counted %d in stream out wheel\n", stcb->asoc.stream_queue_cnt, chks_in_queue); } if (chks_in_queue) { /* call the output queue function */ sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_T3, SCTP_SO_NOT_LOCKED); if ((TAILQ_EMPTY(&stcb->asoc.send_queue)) && (TAILQ_EMPTY(&stcb->asoc.sent_queue))) { /* * Probably should go in and make it go back through * and add fragments allowed */ if (being_filled == 0) { SCTP_PRINTF("Still nothing moved %d chunks are stuck\n", chks_in_queue); } } } else { SCTP_PRINTF("Found no chunks on any queue tot:%lu\n", (u_long)stcb->asoc.total_output_queue_size); stcb->asoc.total_output_queue_size = 0; } } int sctp_heartbeat_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { uint8_t net_was_pf; if (net->dest_state & SCTP_ADDR_PF) { net_was_pf = 1; } else { net_was_pf = 0; } if (net->hb_responded == 0) { if (net->ro._s_addr) { /* * Invalidate the src address if we did not get a * response last time. */ sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } sctp_backoff_on_timeout(stcb, net, 1, 0, 0); if (sctp_threshold_management(inp, stcb, net, stcb->asoc.max_send_times)) { /* Assoc is over */ return (1); } } /* Zero PBA, if it needs it */ if (net->partial_bytes_acked) { net->partial_bytes_acked = 0; } if ((stcb->asoc.total_output_queue_size > 0) && (TAILQ_EMPTY(&stcb->asoc.send_queue)) && (TAILQ_EMPTY(&stcb->asoc.sent_queue))) { sctp_audit_stream_queues_for_size(inp, stcb); } if (!(net->dest_state & SCTP_ADDR_NOHB) && !((net_was_pf == 0) && (net->dest_state & SCTP_ADDR_PF))) { /* * when move to PF during threshold mangement, a HB has been * queued in that routine */ uint32_t ms_gone_by; if ((net->last_sent_time.tv_sec > 0) || (net->last_sent_time.tv_usec > 0)) { struct timeval diff; SCTP_GETTIME_TIMEVAL(&diff); timevalsub(&diff, &net->last_sent_time); ms_gone_by = (uint32_t)(diff.tv_sec * 1000) + (uint32_t)(diff.tv_usec / 1000); } else { ms_gone_by = 0xffffffff; } if ((ms_gone_by >= net->heart_beat_delay) || (net->dest_state & SCTP_ADDR_PF)) { sctp_send_hb(stcb, net, SCTP_SO_NOT_LOCKED); } } return (0); } void sctp_pathmtu_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb, struct sctp_nets *net) { uint32_t next_mtu, mtu; next_mtu = sctp_get_next_mtu(net->mtu); if ((next_mtu > net->mtu) && (net->port == 0)) { if ((net->src_addr_selected == 0) || (net->ro._s_addr == NULL) || (net->ro._s_addr->localifa_flags & SCTP_BEING_DELETED)) { if ((net->ro._s_addr != NULL) && (net->ro._s_addr->localifa_flags & SCTP_BEING_DELETED)) { sctp_free_ifa(net->ro._s_addr); net->ro._s_addr = NULL; net->src_addr_selected = 0; } else if (net->ro._s_addr == NULL) { #if defined(INET6) && defined(SCTP_EMBEDDED_V6_SCOPE) if (net->ro._l_addr.sa.sa_family == AF_INET6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; /* KAME hack: embed scopeid */ (void)sa6_embedscope(sin6, MODULE_GLOBAL(ip6_use_defzone)); } #endif net->ro._s_addr = sctp_source_address_selection(inp, stcb, (sctp_route_t *)&net->ro, net, 0, stcb->asoc.vrf_id); #if defined(INET6) && defined(SCTP_EMBEDDED_V6_SCOPE) if (net->ro._l_addr.sa.sa_family == AF_INET6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&net->ro._l_addr; (void)sa6_recoverscope(sin6); } #endif /* INET6 */ } if (net->ro._s_addr) net->src_addr_selected = 1; } if (net->ro._s_addr) { - mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._s_addr.sa, net->ro.ro_rt); + mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._s_addr.sa, net->ro.ro_nh); #if defined(INET) || defined(INET6) if (net->port) { mtu -= sizeof(struct udphdr); } #endif if (mtu > next_mtu) { net->mtu = next_mtu; } else { net->mtu = mtu; } } } /* restart the timer */ sctp_timer_start(SCTP_TIMER_TYPE_PATHMTURAISE, inp, stcb, net); } void sctp_autoclose_timer(struct sctp_inpcb *inp, struct sctp_tcb *stcb) { struct timeval tn, *tim_touse; struct sctp_association *asoc; uint32_t ticks_gone_by; (void)SCTP_GETTIME_TIMEVAL(&tn); if (stcb->asoc.sctp_autoclose_ticks > 0 && sctp_is_feature_on(inp, SCTP_PCB_FLAGS_AUTOCLOSE)) { /* Auto close is on */ asoc = &stcb->asoc; /* pick the time to use */ if (asoc->time_last_rcvd.tv_sec > asoc->time_last_sent.tv_sec) { tim_touse = &asoc->time_last_rcvd; } else { tim_touse = &asoc->time_last_sent; } /* Now has long enough transpired to autoclose? */ ticks_gone_by = sctp_secs_to_ticks((uint32_t)(tn.tv_sec - tim_touse->tv_sec)); if (ticks_gone_by >= asoc->sctp_autoclose_ticks) { /* * autoclose time has hit, call the output routine, * which should do nothing just to be SURE we don't * have hanging data. We can then safely check the * queues and know that we are clear to send * shutdown */ sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_AUTOCLOSE_TMR, SCTP_SO_NOT_LOCKED); /* Are we clean? */ if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue)) { /* * there is nothing queued to send, so I'm * done... */ if (SCTP_GET_STATE(stcb) != SCTP_STATE_SHUTDOWN_SENT) { /* only send SHUTDOWN 1st time thru */ struct sctp_nets *net; if ((SCTP_GET_STATE(stcb) == SCTP_STATE_OPEN) || (SCTP_GET_STATE(stcb) == SCTP_STATE_SHUTDOWN_RECEIVED)) { SCTP_STAT_DECR_GAUGE32(sctps_currestab); } SCTP_SET_STATE(stcb, SCTP_STATE_SHUTDOWN_SENT); sctp_stop_timers_for_shutdown(stcb); if (stcb->asoc.alternate) { net = stcb->asoc.alternate; } else { net = stcb->asoc.primary_destination; } sctp_send_shutdown(stcb, net); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN, stcb->sctp_ep, stcb, net); sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD, stcb->sctp_ep, stcb, NULL); } } } else { /* * No auto close at this time, reset t-o to check * later */ uint32_t tmp; /* fool the timer startup to use the time left */ tmp = asoc->sctp_autoclose_ticks; asoc->sctp_autoclose_ticks -= ticks_gone_by; sctp_timer_start(SCTP_TIMER_TYPE_AUTOCLOSE, inp, stcb, NULL); /* restore the real tick value */ asoc->sctp_autoclose_ticks = tmp; } } } Index: head/sys/netinet/sctp_var.h =================================================================== --- head/sys/netinet/sctp_var.h (revision 360291) +++ head/sys/netinet/sctp_var.h (revision 360292) @@ -1,360 +1,357 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001-2008, by Cisco Systems, Inc. All rights reserved. * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved. * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * a) Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * b) 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. * * c) Neither the name of Cisco Systems, Inc. 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 OWNER 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 __FBSDID("$FreeBSD$"); #ifndef _NETINET_SCTP_VAR_H_ #define _NETINET_SCTP_VAR_H_ #include #if defined(_KERNEL) || defined(__Userspace__) extern struct pr_usrreqs sctp_usrreqs; #define sctp_feature_on(inp, feature) (inp->sctp_features |= feature) #define sctp_feature_off(inp, feature) (inp->sctp_features &= ~feature) #define sctp_is_feature_on(inp, feature) ((inp->sctp_features & feature) == feature) #define sctp_is_feature_off(inp, feature) ((inp->sctp_features & feature) == 0) #define sctp_stcb_feature_on(inp, stcb, feature) {\ if (stcb) { \ stcb->asoc.sctp_features |= feature; \ } else if (inp) { \ inp->sctp_features |= feature; \ } \ } #define sctp_stcb_feature_off(inp, stcb, feature) {\ if (stcb) { \ stcb->asoc.sctp_features &= ~feature; \ } else if (inp) { \ inp->sctp_features &= ~feature; \ } \ } #define sctp_stcb_is_feature_on(inp, stcb, feature) \ (((stcb != NULL) && \ ((stcb->asoc.sctp_features & feature) == feature)) || \ ((stcb == NULL) && (inp != NULL) && \ ((inp->sctp_features & feature) == feature))) #define sctp_stcb_is_feature_off(inp, stcb, feature) \ (((stcb != NULL) && \ ((stcb->asoc.sctp_features & feature) == 0)) || \ ((stcb == NULL) && (inp != NULL) && \ ((inp->sctp_features & feature) == 0)) || \ ((stcb == NULL) && (inp == NULL))) /* managing mobility_feature in inpcb (by micchie) */ #define sctp_mobility_feature_on(inp, feature) (inp->sctp_mobility_features |= feature) #define sctp_mobility_feature_off(inp, feature) (inp->sctp_mobility_features &= ~feature) #define sctp_is_mobility_feature_on(inp, feature) (inp->sctp_mobility_features & feature) #define sctp_is_mobility_feature_off(inp, feature) ((inp->sctp_mobility_features & feature) == 0) #define sctp_maxspace(sb) (max((sb)->sb_hiwat,SCTP_MINIMAL_RWND)) #define sctp_sbspace(asoc, sb) ((long) ((sctp_maxspace(sb) > (asoc)->sb_cc) ? (sctp_maxspace(sb) - (asoc)->sb_cc) : 0)) #define sctp_sbspace_failedmsgs(sb) ((long) ((sctp_maxspace(sb) > (sb)->sb_cc) ? (sctp_maxspace(sb) - (sb)->sb_cc) : 0)) #define sctp_sbspace_sub(a,b) (((a) > (b)) ? ((a) - (b)) : 0) /* * I tried to cache the readq entries at one point. But the reality * is that it did not add any performance since this meant we had to * lock the STCB on read. And at that point once you have to do an * extra lock, it really does not matter if the lock is in the ZONE * stuff or in our code. Note that this same problem would occur with * an mbuf cache as well so it is not really worth doing, at least * right now :-D */ #ifdef INVARIANTS #define sctp_free_a_readq(_stcb, _readq) { \ if ((_readq)->on_strm_q) \ panic("On strm q stcb:%p readq:%p", (_stcb), (_readq)); \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_readq), (_readq)); \ SCTP_DECR_READQ_COUNT(); \ } #else #define sctp_free_a_readq(_stcb, _readq) { \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_readq), (_readq)); \ SCTP_DECR_READQ_COUNT(); \ } #endif #define sctp_alloc_a_readq(_stcb, _readq) { \ (_readq) = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_readq), struct sctp_queued_to_read); \ if ((_readq)) { \ SCTP_INCR_READQ_COUNT(); \ } \ } #define sctp_free_a_strmoq(_stcb, _strmoq, _so_locked) { \ if ((_strmoq)->holds_key_ref) { \ sctp_auth_key_release(stcb, sp->auth_keyid, _so_locked); \ (_strmoq)->holds_key_ref = 0; \ } \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_strmoq), (_strmoq)); \ SCTP_DECR_STRMOQ_COUNT(); \ } #define sctp_alloc_a_strmoq(_stcb, _strmoq) { \ (_strmoq) = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_strmoq), struct sctp_stream_queue_pending); \ if ((_strmoq)) { \ memset(_strmoq, 0, sizeof(struct sctp_stream_queue_pending)); \ SCTP_INCR_STRMOQ_COUNT(); \ (_strmoq)->holds_key_ref = 0; \ } \ } #define sctp_free_a_chunk(_stcb, _chk, _so_locked) { \ if ((_chk)->holds_key_ref) {\ sctp_auth_key_release((_stcb), (_chk)->auth_keyid, _so_locked); \ (_chk)->holds_key_ref = 0; \ } \ if (_stcb) { \ SCTP_TCB_LOCK_ASSERT((_stcb)); \ if ((_chk)->whoTo) { \ sctp_free_remote_addr((_chk)->whoTo); \ (_chk)->whoTo = NULL; \ } \ if (((_stcb)->asoc.free_chunk_cnt > SCTP_BASE_SYSCTL(sctp_asoc_free_resc_limit)) || \ (SCTP_BASE_INFO(ipi_free_chunks) > SCTP_BASE_SYSCTL(sctp_system_free_resc_limit))) { \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), (_chk)); \ SCTP_DECR_CHK_COUNT(); \ } else { \ TAILQ_INSERT_TAIL(&(_stcb)->asoc.free_chunks, (_chk), sctp_next); \ (_stcb)->asoc.free_chunk_cnt++; \ atomic_add_int(&SCTP_BASE_INFO(ipi_free_chunks), 1); \ } \ } else { \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_chunk), (_chk)); \ SCTP_DECR_CHK_COUNT(); \ } \ } #define sctp_alloc_a_chunk(_stcb, _chk) { \ if (TAILQ_EMPTY(&(_stcb)->asoc.free_chunks)) { \ (_chk) = SCTP_ZONE_GET(SCTP_BASE_INFO(ipi_zone_chunk), struct sctp_tmit_chunk); \ if ((_chk)) { \ SCTP_INCR_CHK_COUNT(); \ (_chk)->whoTo = NULL; \ (_chk)->holds_key_ref = 0; \ } \ } else { \ (_chk) = TAILQ_FIRST(&(_stcb)->asoc.free_chunks); \ TAILQ_REMOVE(&(_stcb)->asoc.free_chunks, (_chk), sctp_next); \ atomic_subtract_int(&SCTP_BASE_INFO(ipi_free_chunks), 1); \ (_chk)->holds_key_ref = 0; \ SCTP_STAT_INCR(sctps_cached_chk); \ (_stcb)->asoc.free_chunk_cnt--; \ } \ } #define sctp_free_remote_addr(__net) { \ if ((__net)) { \ if (SCTP_DECREMENT_AND_CHECK_REFCOUNT(&(__net)->ref_count)) { \ (void)SCTP_OS_TIMER_STOP(&(__net)->rxt_timer.timer); \ - if ((__net)->ro.ro_rt) { \ - RTFREE((__net)->ro.ro_rt); \ - (__net)->ro.ro_rt = NULL; \ - } \ + RO_NHFREE(&(__net)->ro); \ if ((__net)->src_addr_selected) { \ sctp_free_ifa((__net)->ro._s_addr); \ (__net)->ro._s_addr = NULL; \ } \ (__net)->src_addr_selected = 0; \ (__net)->dest_state &= ~SCTP_ADDR_REACHABLE; \ SCTP_ZONE_FREE(SCTP_BASE_INFO(ipi_zone_net), (__net)); \ SCTP_DECR_RADDR_COUNT(); \ } \ } \ } #define sctp_sbfree(ctl, stcb, sb, m) { \ SCTP_SAVE_ATOMIC_DECREMENT(&(sb)->sb_cc, SCTP_BUF_LEN((m))); \ SCTP_SAVE_ATOMIC_DECREMENT(&(sb)->sb_mbcnt, MSIZE); \ if (((ctl)->do_not_ref_stcb == 0) && stcb) {\ SCTP_SAVE_ATOMIC_DECREMENT(&(stcb)->asoc.sb_cc, SCTP_BUF_LEN((m))); \ SCTP_SAVE_ATOMIC_DECREMENT(&(stcb)->asoc.my_rwnd_control_len, MSIZE); \ } \ if (SCTP_BUF_TYPE(m) != MT_DATA && SCTP_BUF_TYPE(m) != MT_HEADER && \ SCTP_BUF_TYPE(m) != MT_OOBDATA) \ atomic_subtract_int(&(sb)->sb_ctl,SCTP_BUF_LEN((m))); \ } #define sctp_sballoc(stcb, sb, m) { \ atomic_add_int(&(sb)->sb_cc,SCTP_BUF_LEN((m))); \ atomic_add_int(&(sb)->sb_mbcnt, MSIZE); \ if (stcb) { \ atomic_add_int(&(stcb)->asoc.sb_cc, SCTP_BUF_LEN((m))); \ atomic_add_int(&(stcb)->asoc.my_rwnd_control_len, MSIZE); \ } \ if (SCTP_BUF_TYPE(m) != MT_DATA && SCTP_BUF_TYPE(m) != MT_HEADER && \ SCTP_BUF_TYPE(m) != MT_OOBDATA) \ atomic_add_int(&(sb)->sb_ctl,SCTP_BUF_LEN((m))); \ } #define sctp_ucount_incr(val) { \ val++; \ } #define sctp_ucount_decr(val) { \ if (val > 0) { \ val--; \ } else { \ val = 0; \ } \ } #define sctp_mbuf_crush(data) do { \ struct mbuf *_m; \ _m = (data); \ while (_m && (SCTP_BUF_LEN(_m) == 0)) { \ (data) = SCTP_BUF_NEXT(_m); \ SCTP_BUF_NEXT(_m) = NULL; \ sctp_m_free(_m); \ _m = (data); \ } \ } while (0) #define sctp_flight_size_decrease(tp1) do { \ if (tp1->whoTo->flight_size >= tp1->book_size) \ tp1->whoTo->flight_size -= tp1->book_size; \ else \ tp1->whoTo->flight_size = 0; \ } while (0) #define sctp_flight_size_increase(tp1) do { \ (tp1)->whoTo->flight_size += (tp1)->book_size; \ } while (0) #ifdef SCTP_FS_SPEC_LOG #define sctp_total_flight_decrease(stcb, tp1) do { \ if (stcb->asoc.fs_index > SCTP_FS_SPEC_LOG_SIZE) \ stcb->asoc.fs_index = 0;\ stcb->asoc.fslog[stcb->asoc.fs_index].total_flight = stcb->asoc.total_flight; \ stcb->asoc.fslog[stcb->asoc.fs_index].tsn = tp1->rec.data.tsn; \ stcb->asoc.fslog[stcb->asoc.fs_index].book = tp1->book_size; \ stcb->asoc.fslog[stcb->asoc.fs_index].sent = tp1->sent; \ stcb->asoc.fslog[stcb->asoc.fs_index].incr = 0; \ stcb->asoc.fslog[stcb->asoc.fs_index].decr = 1; \ stcb->asoc.fs_index++; \ tp1->window_probe = 0; \ if (stcb->asoc.total_flight >= tp1->book_size) { \ stcb->asoc.total_flight -= tp1->book_size; \ if (stcb->asoc.total_flight_count > 0) \ stcb->asoc.total_flight_count--; \ } else { \ stcb->asoc.total_flight = 0; \ stcb->asoc.total_flight_count = 0; \ } \ } while (0) #define sctp_total_flight_increase(stcb, tp1) do { \ if (stcb->asoc.fs_index > SCTP_FS_SPEC_LOG_SIZE) \ stcb->asoc.fs_index = 0;\ stcb->asoc.fslog[stcb->asoc.fs_index].total_flight = stcb->asoc.total_flight; \ stcb->asoc.fslog[stcb->asoc.fs_index].tsn = tp1->rec.data.tsn; \ stcb->asoc.fslog[stcb->asoc.fs_index].book = tp1->book_size; \ stcb->asoc.fslog[stcb->asoc.fs_index].sent = tp1->sent; \ stcb->asoc.fslog[stcb->asoc.fs_index].incr = 1; \ stcb->asoc.fslog[stcb->asoc.fs_index].decr = 0; \ stcb->asoc.fs_index++; \ (stcb)->asoc.total_flight_count++; \ (stcb)->asoc.total_flight += (tp1)->book_size; \ } while (0) #else #define sctp_total_flight_decrease(stcb, tp1) do { \ tp1->window_probe = 0; \ if (stcb->asoc.total_flight >= tp1->book_size) { \ stcb->asoc.total_flight -= tp1->book_size; \ if (stcb->asoc.total_flight_count > 0) \ stcb->asoc.total_flight_count--; \ } else { \ stcb->asoc.total_flight = 0; \ stcb->asoc.total_flight_count = 0; \ } \ } while (0) #define sctp_total_flight_increase(stcb, tp1) do { \ (stcb)->asoc.total_flight_count++; \ (stcb)->asoc.total_flight += (tp1)->book_size; \ } while (0) #endif #define SCTP_PF_ENABLED(_net) (_net->pf_threshold < _net->failure_threshold) #define SCTP_NET_IS_PF(_net) (_net->pf_threshold < _net->error_count) struct sctp_nets; struct sctp_inpcb; struct sctp_tcb; struct sctphdr; void sctp_close(struct socket *so); int sctp_disconnect(struct socket *so); void sctp_ctlinput(int, struct sockaddr *, void *); int sctp_ctloutput(struct socket *, struct sockopt *); #ifdef INET void sctp_input_with_port(struct mbuf *, int, uint16_t); int sctp_input(struct mbuf **, int *, int); #endif void sctp_pathmtu_adjustment(struct sctp_tcb *, uint16_t); void sctp_drain(void); void sctp_init(void); void sctp_notify(struct sctp_inpcb *, struct sctp_tcb *, struct sctp_nets *, uint8_t, uint8_t, uint16_t, uint32_t); int sctp_flush(struct socket *, int); int sctp_shutdown(struct socket *); int sctp_bindx(struct socket *, int, struct sockaddr_storage *, int, int, struct proc *); /* can't use sctp_assoc_t here */ int sctp_peeloff(struct socket *, struct socket *, int, caddr_t, int *); int sctp_ingetaddr(struct socket *, struct sockaddr **); int sctp_peeraddr(struct socket *, struct sockaddr **); int sctp_listen(struct socket *, int, struct thread *); int sctp_accept(struct socket *, struct sockaddr **); #endif /* _KERNEL */ #endif /* !_NETINET_SCTP_VAR_H_ */ Index: head/sys/netinet/tcp_output.c =================================================================== --- head/sys/netinet/tcp_output.c (revision 360291) +++ head/sys/netinet/tcp_output.c (revision 360292) @@ -1,2104 +1,2105 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_tcpdebug.h" #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef KERN_TLS #include #endif #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #endif #include #define TCPOUTFLAGS #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include VNET_DEFINE(int, path_mtu_discovery) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(path_mtu_discovery), 1, "Enable Path MTU Discovery"); VNET_DEFINE(int, tcp_do_tso) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_tso), 0, "Enable TCP Segmentation Offload"); VNET_DEFINE(int, tcp_sendspace) = 1024*32; #define V_tcp_sendspace VNET(tcp_sendspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size"); VNET_DEFINE(int, tcp_do_autosndbuf) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autosndbuf), 0, "Enable automatic send buffer sizing"); VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autosndbuf_inc), 0, "Incrementor step size of automatic send buffer"); VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autosndbuf_max), 0, "Max size of automatic send buffer"); VNET_DEFINE(int, tcp_sendbuf_auto_lowat) = 0; #define V_tcp_sendbuf_auto_lowat VNET(tcp_sendbuf_auto_lowat) SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto_lowat, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sendbuf_auto_lowat), 0, "Modify threshold for auto send buffer growth to account for SO_SNDLOWAT"); /* * Make sure that either retransmit or persist timer is set for SYN, FIN and * non-ACK. */ #define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags) \ KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\ tcp_timer_active((tp), TT_REXMT) || \ tcp_timer_active((tp), TT_PERSIST), \ ("neither rexmt nor persist timer is set")) static void inline cc_after_idle(struct tcpcb *tp); #ifdef TCP_HHOOK /* * Wrapper for the TCP established output helper hook. */ void hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t len, int tso) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_data.len = len; hhook_data.tso = tso; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data, tp->osd); } } #endif /* * CC wrapper hook functions */ static void inline cc_after_idle(struct tcpcb *tp) { INP_WLOCK_ASSERT(tp->t_inpcb); if (CC_ALGO(tp)->after_idle != NULL) CC_ALGO(tp)->after_idle(tp->ccv); } /* * Tcp output routine: figure out what should be sent and send it. */ int tcp_output(struct tcpcb *tp) { struct socket *so = tp->t_inpcb->inp_socket; int32_t len; uint32_t recwin, sendwin; int off, flags, error = 0; /* Keep compiler happy */ u_int if_hw_tsomaxsegcount = 0; u_int if_hw_tsomaxsegsize = 0; struct mbuf *m; struct ip *ip = NULL; #ifdef TCPDEBUG struct ipovly *ipov = NULL; #endif struct tcphdr *th; u_char opt[TCP_MAXOLEN]; unsigned ipoptlen, optlen, hdrlen; #if defined(IPSEC) || defined(IPSEC_SUPPORT) unsigned ipsec_optlen = 0; #endif int idle, sendalot, curticks; int sack_rxmit, sack_bytes_rxmt; struct sackhole *p; int tso, mtu; struct tcpopt to; unsigned int wanted_cookie = 0; unsigned int dont_sendalot = 0; #if 0 int maxburst = TCP_MAXBURST; #endif #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6; isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; #endif #ifdef KERN_TLS const bool hw_tls = (so->so_snd.sb_flags & SB_TLS_IFNET) != 0; #else const bool hw_tls = false; #endif NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(tp->t_inpcb); #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) return (tcp_offload_output(tp)); #endif /* * For TFO connections in SYN_SENT or SYN_RECEIVED, * only allow the initial SYN or SYN|ACK and those sent * by the retransmit timer. */ if (IS_FASTOPEN(tp->t_flags) && ((tp->t_state == TCPS_SYN_SENT) || (tp->t_state == TCPS_SYN_RECEIVED)) && SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN or SYN|ACK sent */ (tp->snd_nxt != tp->snd_una)) /* not a retransmit */ return (0); /* * Determine length of data that should be transmitted, * and flags that will be used. * If there is some data or critical controls (SYN, RST) * to send, then transmit; otherwise, investigate further. */ idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur) cc_after_idle(tp); tp->t_flags &= ~TF_LASTIDLE; if (idle) { if (tp->t_flags & TF_MORETOCOME) { tp->t_flags |= TF_LASTIDLE; idle = 0; } } again: /* * If we've recently taken a timeout, snd_max will be greater than * snd_nxt. There may be SACK information that allows us to avoid * resending already delivered data. Adjust snd_nxt accordingly. */ if ((tp->t_flags & TF_SACK_PERMIT) && SEQ_LT(tp->snd_nxt, tp->snd_max)) tcp_sack_adjust(tp); sendalot = 0; tso = 0; mtu = 0; off = tp->snd_nxt - tp->snd_una; sendwin = min(tp->snd_wnd, tp->snd_cwnd); flags = tcp_outflags[tp->t_state]; /* * Send any SACK-generated retransmissions. If we're explicitly trying * to send out new data (when sendalot is 1), bypass this function. * If we retransmit in fast recovery mode, decrement snd_cwnd, since * we're replacing a (future) new transmission with a retransmission * now, and we previously incremented snd_cwnd in tcp_input(). */ /* * Still in sack recovery , reset rxmit flag to zero. */ sack_rxmit = 0; sack_bytes_rxmt = 0; len = 0; p = NULL; if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) && (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { uint32_t cwin; cwin = imax(min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt, 0); /* Do not retransmit SACK segments beyond snd_recover */ if (SEQ_GT(p->end, tp->snd_recover)) { /* * (At least) part of sack hole extends beyond * snd_recover. Check to see if we can rexmit data * for this hole. */ if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { /* * Can't rexmit any more data for this hole. * That data will be rexmitted in the next * sack recovery episode, when snd_recover * moves past p->rxmit. */ p = NULL; goto after_sack_rexmit; } else /* Can rexmit part of the current hole */ len = ((int32_t)ulmin(cwin, tp->snd_recover - p->rxmit)); } else len = ((int32_t)ulmin(cwin, p->end - p->rxmit)); off = p->rxmit - tp->snd_una; KASSERT(off >= 0,("%s: sack block to the left of una : %d", __func__, off)); if (len > 0) { sack_rxmit = 1; sendalot = 1; TCPSTAT_INC(tcps_sack_rexmits); TCPSTAT_ADD(tcps_sack_rexmit_bytes, min(len, tp->t_maxseg)); } } after_sack_rexmit: /* * Get standard flags, and add SYN or FIN if requested by 'hidden' * state flags. */ if (tp->t_flags & TF_NEEDFIN) flags |= TH_FIN; if (tp->t_flags & TF_NEEDSYN) flags |= TH_SYN; SOCKBUF_LOCK(&so->so_snd); /* * If in persist timeout with window of 0, send 1 byte. * Otherwise, if window is small but nonzero * and timer expired, we will send what we can * and go to transmit state. */ if (tp->t_flags & TF_FORCEDATA) { if (sendwin == 0) { /* * If we still have some data to send, then * clear the FIN bit. Usually this would * happen below when it realizes that we * aren't sending all the data. However, * if we have exactly 1 byte of unsent data, * then it won't clear the FIN bit below, * and if we are in persist state, we wind * up sending the packet without recording * that we sent the FIN bit. * * We can't just blindly clear the FIN bit, * because if we don't have any more data * to send then the probe will be the FIN * itself. */ if (off < sbused(&so->so_snd)) flags &= ~TH_FIN; sendwin = 1; } else { tcp_timer_activate(tp, TT_PERSIST, 0); tp->t_rxtshift = 0; } } /* * If snd_nxt == snd_max and we have transmitted a FIN, the * offset will be > 0 even if so_snd.sb_cc is 0, resulting in * a negative length. This can also occur when TCP opens up * its congestion window while receiving additional duplicate * acks after fast-retransmit because TCP will reset snd_nxt * to snd_max after the fast-retransmit. * * In the normal retransmit-FIN-only case, however, snd_nxt will * be set to snd_una, the offset will be 0, and the length may * wind up 0. * * If sack_rxmit is true we are retransmitting from the scoreboard * in which case len is already set. */ if (sack_rxmit == 0) { if (sack_bytes_rxmt == 0) len = ((int32_t)min(sbavail(&so->so_snd), sendwin) - off); else { int32_t cwin; /* * We are inside of a SACK recovery episode and are * sending new data, having retransmitted all the * data possible in the scoreboard. */ len = ((int32_t)min(sbavail(&so->so_snd), tp->snd_wnd) - off); /* * Don't remove this (len > 0) check ! * We explicitly check for len > 0 here (although it * isn't really necessary), to work around a gcc * optimization issue - to force gcc to compute * len above. Without this check, the computation * of len is bungled by the optimizer. */ if (len > 0) { cwin = tp->snd_cwnd - (tp->snd_nxt - tp->snd_recover) - sack_bytes_rxmt; if (cwin < 0) cwin = 0; len = imin(len, cwin); } } } /* * Lop off SYN bit if it has already been sent. However, if this * is SYN-SENT state and if segment contains data and if we don't * know that foreign host supports TAO, suppress sending segment. */ if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { if (tp->t_state != TCPS_SYN_RECEIVED) flags &= ~TH_SYN; /* * When sending additional segments following a TFO SYN|ACK, * do not include the SYN bit. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_state == TCPS_SYN_RECEIVED)) flags &= ~TH_SYN; off--, len++; } /* * Be careful not to send data and/or FIN on SYN segments. * This measure is needed to prevent interoperability problems * with not fully conformant TCP implementations. */ if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { len = 0; flags &= ~TH_FIN; } /* * On TFO sockets, ensure no data is sent in the following cases: * * - When retransmitting SYN|ACK on a passively-created socket * * - When retransmitting SYN on an actively created socket * * - When sending a zero-length cookie (cookie request) on an * actively created socket * * - When the socket is in the CLOSED state (RST is being sent) */ if (IS_FASTOPEN(tp->t_flags) && (((flags & TH_SYN) && (tp->t_rxtshift > 0)) || ((tp->t_state == TCPS_SYN_SENT) && (tp->t_tfo_client_cookie_len == 0)) || (flags & TH_RST))) len = 0; if (len <= 0) { /* * If FIN has been sent but not acked, * but we haven't been called to retransmit, * len will be < 0. Otherwise, window shrank * after we sent into it. If window shrank to 0, * cancel pending retransmit, pull snd_nxt back * to (closed) window, and set the persist timer * if it isn't already going. If the window didn't * close completely, just wait for an ACK. * * We also do a general check here to ensure that * we will set the persist timer when we have data * to send, but a 0-byte window. This makes sure * the persist timer is set even if the packet * hits one of the "goto send" lines below. */ len = 0; if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && (off < (int) sbavail(&so->so_snd))) { tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rxtshift = 0; tp->snd_nxt = tp->snd_una; if (!tcp_timer_active(tp, TT_PERSIST)) tcp_setpersist(tp); } } /* len will be >= 0 after this point. */ KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); tcp_sndbuf_autoscale(tp, so, sendwin); /* * Decide if we can use TCP Segmentation Offloading (if supported by * hardware). * * TSO may only be used if we are in a pure bulk sending state. The * presence of TCP-MD5, SACK retransmits, SACK advertizements and * IP options prevent using TSO. With TSO the TCP header is the same * (except for the sequence number) for all generated packets. This * makes it impossible to transmit any options which vary per generated * segment or packet. * * IPv4 handling has a clear separation of ip options and ip header * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does * the right thing below to provide length of just ip options and thus * checking for ipoptlen is enough to decide if ip options are present. */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * Pre-calculate here as we save another lookup into the darknesses * of IPsec that way and can actually decide if TSO is ok. */ #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6)) ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb); #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4)) ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb); #endif /* INET */ #endif /* IPSEC */ #ifdef INET6 if (isipv6) ipoptlen = ip6_optlen(tp->t_inpcb); else #endif if (tp->t_inpcb->inp_options) ipoptlen = tp->t_inpcb->inp_options->m_len - offsetof(struct ipoption, ipopt_list); else ipoptlen = 0; #if defined(IPSEC) || defined(IPSEC_SUPPORT) ipoptlen += ipsec_optlen; #endif if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg && ((tp->t_flags & TF_SIGNATURE) == 0) && tp->rcv_numsacks == 0 && sack_rxmit == 0 && ipoptlen == 0 && !(flags & TH_SYN)) tso = 1; if (sack_rxmit) { if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd))) flags &= ~TH_FIN; } else { if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + sbused(&so->so_snd))) flags &= ~TH_FIN; } recwin = lmin(lmax(sbspace(&so->so_rcv), 0), (long)TCP_MAXWIN << tp->rcv_scale); /* * Sender silly window avoidance. We transmit under the following * conditions when len is non-zero: * * - We have a full segment (or more with TSO) * - This is the last buffer in a write()/send() and we are * either idle or running NODELAY * - we've timed out (e.g. persist timer) * - we have more then 1/2 the maximum send window's worth of * data (receiver may be limited the window size) * - we need to retransmit */ if (len) { if (len >= tp->t_maxseg) goto send; /* * NOTE! on localhost connections an 'ack' from the remote * end may occur synchronously with the output and cause * us to flush a buffer queued with moretocome. XXX * * note: the len + off check is almost certainly unnecessary. */ if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ (idle || (tp->t_flags & TF_NODELAY)) && (uint32_t)len + (uint32_t)off >= sbavail(&so->so_snd) && (tp->t_flags & TF_NOPUSH) == 0) { goto send; } if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ goto send; if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) goto send; if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ goto send; if (sack_rxmit) goto send; } /* * Sending of standalone window updates. * * Window updates are important when we close our window due to a * full socket buffer and are opening it again after the application * reads data from it. Once the window has opened again and the * remote end starts to send again the ACK clock takes over and * provides the most current window information. * * We must avoid the silly window syndrome whereas every read * from the receive buffer, no matter how small, causes a window * update to be sent. We also should avoid sending a flurry of * window updates when the socket buffer had queued a lot of data * and the application is doing small reads. * * Prevent a flurry of pointless window updates by only sending * an update when we can increase the advertized window by more * than 1/4th of the socket buffer capacity. When the buffer is * getting full or is very small be more aggressive and send an * update whenever we can increase by two mss sized segments. * In all other situations the ACK's to new incoming data will * carry further window increases. * * Don't send an independent window update if a delayed * ACK is pending (it will get piggy-backed on it) or the * remote side already has done a half-close and won't send * more data. Skip this if the connection is in T/TCP * half-open state. */ if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && !(tp->t_flags & TF_DELACK) && !TCPS_HAVERCVDFIN(tp->t_state)) { /* * "adv" is the amount we could increase the window, * taking into account that we are limited by * TCP_MAXWIN << tp->rcv_scale. */ int32_t adv; int oldwin; adv = recwin; if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { oldwin = (tp->rcv_adv - tp->rcv_nxt); adv -= oldwin; } else oldwin = 0; /* * If the new window size ends up being the same as or less * than the old size when it is scaled, then don't force * a window update. */ if (oldwin >> tp->rcv_scale >= (adv + oldwin) >> tp->rcv_scale) goto dontupdate; if (adv >= (int32_t)(2 * tp->t_maxseg) && (adv >= (int32_t)(so->so_rcv.sb_hiwat / 4) || recwin <= (so->so_rcv.sb_hiwat / 8) || so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg || adv >= TCP_MAXWIN << tp->rcv_scale)) goto send; if (2 * adv >= (int32_t)so->so_rcv.sb_hiwat) goto send; } dontupdate: /* * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW * is also a catch-all for the retransmit timer timeout case. */ if (tp->t_flags & TF_ACKNOW) goto send; if ((flags & TH_RST) || ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) goto send; if (SEQ_GT(tp->snd_up, tp->snd_una)) goto send; /* * If our state indicates that FIN should be sent * and we have not yet done so, then we need to send. */ if (flags & TH_FIN && ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) goto send; /* * In SACK, it is possible for tcp_output to fail to send a segment * after the retransmission timer has been turned off. Make sure * that the retransmission timer is set. */ if ((tp->t_flags & TF_SACK_PERMIT) && SEQ_GT(tp->snd_max, tp->snd_una) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); goto just_return; } /* * TCP window updates are not reliable, rather a polling protocol * using ``persist'' packets is used to insure receipt of window * updates. The three ``states'' for the output side are: * idle not doing retransmits or persists * persisting to move a small or zero window * (re)transmitting and thereby not persisting * * tcp_timer_active(tp, TT_PERSIST) * is true when we are in persist state. * (tp->t_flags & TF_FORCEDATA) * is set when we are called to send a persist packet. * tcp_timer_active(tp, TT_REXMT) * is set when we are retransmitting * The output side is idle when both timers are zero. * * If send window is too small, there is data to transmit, and no * retransmit or persist is pending, then go to persist state. * If nothing happens soon, send when timer expires: * if window is nonzero, transmit what we can, * otherwise force out a byte. */ if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { tp->t_rxtshift = 0; tcp_setpersist(tp); } /* * No reason to send a segment, just return. */ just_return: SOCKBUF_UNLOCK(&so->so_snd); return (0); send: SOCKBUF_LOCK_ASSERT(&so->so_snd); if (len > 0) { if (len >= tp->t_maxseg) tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; else tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; } /* * Before ESTABLISHED, force sending of initial options * unless TCP set not to do any options. * NOTE: we assume that the IP/TCP header plus TCP options * always fit in a single mbuf, leaving room for a maximum * link header, i.e. * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES */ optlen = 0; #ifdef INET6 if (isipv6) hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); else #endif hdrlen = sizeof (struct tcpiphdr); /* * Compute options for segment. * We only have to care about SYN and established connection * segments. Options for SYN-ACK segments are handled in TCP * syncache. */ to.to_flags = 0; if ((tp->t_flags & TF_NOOPT) == 0) { /* Maximum segment size. */ if (flags & TH_SYN) { tp->snd_nxt = tp->iss; to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc); to.to_flags |= TOF_MSS; /* * On SYN or SYN|ACK transmits on TFO connections, * only include the TFO option if it is not a * retransmit, as the presence of the TFO option may * have caused the original SYN or SYN|ACK to have * been dropped by a middlebox. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_rxtshift == 0)) { if (tp->t_state == TCPS_SYN_RECEIVED) { to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; to.to_tfo_cookie = (u_int8_t *)&tp->t_tfo_cookie.server; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; } else if (tp->t_state == TCPS_SYN_SENT) { to.to_tfo_len = tp->t_tfo_client_cookie_len; to.to_tfo_cookie = tp->t_tfo_cookie.client; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; /* * If we wind up having more data to * send with the SYN than can fit in * one segment, don't send any more * until the SYN|ACK comes back from * the other end. */ dont_sendalot = 1; } } } /* Window scaling. */ if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { to.to_wscale = tp->request_r_scale; to.to_flags |= TOF_SCALE; } /* Timestamps. */ if ((tp->t_flags & TF_RCVD_TSTMP) || ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { curticks = tcp_ts_getticks(); to.to_tsval = curticks + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; if (tp->t_rxtshift == 1) tp->t_badrxtwin = curticks; } /* Set receive buffer autosizing timestamp. */ if (tp->rfbuf_ts == 0 && (so->so_rcv.sb_flags & SB_AUTOSIZE)) tp->rfbuf_ts = tcp_ts_getticks(); /* Selective ACK's. */ if (tp->t_flags & TF_SACK_PERMIT) { if (flags & TH_SYN) to.to_flags |= TOF_SACKPERM; else if (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { to.to_flags |= TOF_SACK; to.to_nsacks = tp->rcv_numsacks; to.to_sacks = (u_char *)tp->sackblks; } } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ /* * Check that TCP_MD5SIG is enabled in tcpcb to * account the size needed to set this TCP option. */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* TCP_SIGNATURE */ /* Processing the options. */ hdrlen += optlen = tcp_addoptions(&to, opt); /* * If we wanted a TFO option to be added, but it was unable * to fit, ensure no data is sent. */ if (IS_FASTOPEN(tp->t_flags) && wanted_cookie && !(to.to_flags & TOF_FASTOPEN)) len = 0; } /* * Adjust data length if insertion of options will * bump the packet length beyond the t_maxseg length. * Clear the FIN bit because we cut off the tail of * the segment. */ if (len + optlen + ipoptlen > tp->t_maxseg) { flags &= ~TH_FIN; if (tso) { u_int if_hw_tsomax; u_int moff; int max_len; /* extract TSO information */ if_hw_tsomax = tp->t_tsomax; if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; /* * Limit a TSO burst to prevent it from * overflowing or exceeding the maximum length * allowed by the network interface: */ KASSERT(ipoptlen == 0, ("%s: TSO can't do IP options", __func__)); /* * Check if we should limit by maximum payload * length: */ if (if_hw_tsomax != 0) { /* compute maximum TSO length */ max_len = (if_hw_tsomax - hdrlen - max_linkhdr); if (max_len <= 0) { len = 0; } else if (len > max_len) { sendalot = 1; len = max_len; } } /* * Prevent the last segment from being * fractional unless the send sockbuf can be * emptied: */ max_len = (tp->t_maxseg - optlen); if (((uint32_t)off + (uint32_t)len) < sbavail(&so->so_snd)) { moff = len % max_len; if (moff != 0) { len -= moff; sendalot = 1; } } /* * In case there are too many small fragments * don't use TSO: */ if (len <= max_len) { len = max_len; sendalot = 1; tso = 0; } /* * Send the FIN in a separate segment * after the bulk sending is done. * We don't trust the TSO implementations * to clear the FIN flag on all but the * last segment. */ if (tp->t_flags & TF_NEEDFIN) sendalot = 1; } else { if (optlen + ipoptlen >= tp->t_maxseg) { /* * Since we don't have enough space to put * the IP header chain and the TCP header in * one packet as required by RFC 7112, don't * send it. Also ensure that at least one * byte of the payload can be put into the * TCP segment. */ SOCKBUF_UNLOCK(&so->so_snd); error = EMSGSIZE; sack_rxmit = 0; goto out; } len = tp->t_maxseg - optlen - ipoptlen; sendalot = 1; if (dont_sendalot) sendalot = 0; } } else tso = 0; KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, ("%s: len > IP_MAXPACKET", __func__)); /*#ifdef DIAGNOSTIC*/ #ifdef INET6 if (max_linkhdr + hdrlen > MCLBYTES) #else if (max_linkhdr + hdrlen > MHLEN) #endif panic("tcphdr too big"); /*#endif*/ /* * This KASSERT is here to catch edge cases at a well defined place. * Before, those had triggered (random) panic conditions further down. */ KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); /* * Grab a header mbuf, attaching a copy of data to * be transmitted, and initialize the header from * the template for sends on this connection. */ if (len) { struct mbuf *mb; struct sockbuf *msb; u_int moff; if ((tp->t_flags & TF_FORCEDATA) && len == 1) { TCPSTAT_INC(tcps_sndprobe); #ifdef STATS if (SEQ_LT(tp->snd_nxt, tp->snd_max)) stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); else stats_voi_update_abs_u64(tp->t_stats, VOI_TCP_TXPB, len); #endif /* STATS */ } else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { tp->t_sndrexmitpack++; TCPSTAT_INC(tcps_sndrexmitpack); TCPSTAT_ADD(tcps_sndrexmitbyte, len); #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); #endif /* STATS */ } else { TCPSTAT_INC(tcps_sndpack); TCPSTAT_ADD(tcps_sndbyte, len); #ifdef STATS stats_voi_update_abs_u64(tp->t_stats, VOI_TCP_TXPB, len); #endif /* STATS */ } #ifdef INET6 if (MHLEN < hdrlen + max_linkhdr) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else #endif m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOBUFS; sack_rxmit = 0; goto out; } m->m_data += max_linkhdr; m->m_len = hdrlen; /* * Start the m_copy functions from the closest mbuf * to the offset in the socket buffer chain. */ mb = sbsndptr_noadv(&so->so_snd, off, &moff); if (len <= MHLEN - hdrlen - max_linkhdr && !hw_tls) { m_copydata(mb, moff, len, mtod(m, caddr_t) + hdrlen); if (SEQ_LT(tp->snd_nxt, tp->snd_max)) sbsndptr_adv(&so->so_snd, mb, len); m->m_len += len; } else { if (SEQ_LT(tp->snd_nxt, tp->snd_max)) msb = NULL; else msb = &so->so_snd; m->m_next = tcp_m_copym(mb, moff, &len, if_hw_tsomaxsegcount, if_hw_tsomaxsegsize, msb, hw_tls); if (len <= (tp->t_maxseg - optlen)) { /* * Must have ran out of mbufs for the copy * shorten it to no longer need tso. Lets * not put on sendalot since we are low on * mbufs. */ tso = 0; } if (m->m_next == NULL) { SOCKBUF_UNLOCK(&so->so_snd); (void) m_free(m); error = ENOBUFS; sack_rxmit = 0; goto out; } } /* * If we're sending everything we've got, set PUSH. * (This will keep happy those implementations which only * give data to the user when a buffer fills or * a PUSH comes in.) */ if (((uint32_t)off + (uint32_t)len == sbused(&so->so_snd)) && !(flags & TH_SYN)) flags |= TH_PUSH; SOCKBUF_UNLOCK(&so->so_snd); } else { SOCKBUF_UNLOCK(&so->so_snd); if (tp->t_flags & TF_ACKNOW) TCPSTAT_INC(tcps_sndacks); else if (flags & (TH_SYN|TH_FIN|TH_RST)) TCPSTAT_INC(tcps_sndctrl); else if (SEQ_GT(tp->snd_up, tp->snd_una)) TCPSTAT_INC(tcps_sndurg); else TCPSTAT_INC(tcps_sndwinup); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; sack_rxmit = 0; goto out; } #ifdef INET6 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && MHLEN >= hdrlen) { M_ALIGN(m, hdrlen); } else #endif m->m_data += max_linkhdr; m->m_len = hdrlen; } SOCKBUF_UNLOCK_ASSERT(&so->so_snd); m->m_pkthdr.rcvif = (struct ifnet *)0; #ifdef MAC mac_inpcb_create_mbuf(tp->t_inpcb, m); #endif #ifdef INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)(ip6 + 1); tcpip_fillheaders(tp->t_inpcb, ip6, th); } else #endif /* INET6 */ { ip = mtod(m, struct ip *); #ifdef TCPDEBUG ipov = (struct ipovly *)ip; #endif th = (struct tcphdr *)(ip + 1); tcpip_fillheaders(tp->t_inpcb, ip, th); } /* * Fill in fields, remembering maximum advertised * window for use in delaying messages about window sizes. * If resending a FIN, be sure not to use a new sequence number. */ if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && tp->snd_nxt == tp->snd_max) tp->snd_nxt--; /* * If we are starting a connection, send ECN setup * SYN packet. If we are on a retransmit, we may * resend those bits a number of times as per * RFC 3168. */ if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) { if (tp->t_rxtshift >= 1) { if (tp->t_rxtshift <= V_tcp_ecn_maxretries) flags |= TH_ECE|TH_CWR; } else flags |= TH_ECE|TH_CWR; } if (tp->t_state == TCPS_ESTABLISHED && (tp->t_flags2 & TF2_ECN_PERMIT)) { /* * If the peer has ECN, mark data packets with * ECN capable transmission (ECT). * Ignore pure ack packets, retransmissions and window probes. */ if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && (sack_rxmit == 0) && !((tp->t_flags & TF_FORCEDATA) && len == 1)) { #ifdef INET6 if (isipv6) ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20); else #endif ip->ip_tos |= IPTOS_ECN_ECT0; TCPSTAT_INC(tcps_ecn_ect0); } /* * Reply with proper ECN notifications. */ if (tp->t_flags2 & TF2_ECN_SND_CWR) { flags |= TH_CWR; tp->t_flags2 &= ~TF2_ECN_SND_CWR; } if (tp->t_flags2 & TF2_ECN_SND_ECE) flags |= TH_ECE; } /* * If we are doing retransmissions, then snd_nxt will * not reflect the first unsent octet. For ACK only * packets, we do not want the sequence number of the * retransmitted packet, we want the sequence number * of the next unsent octet. So, if there is no data * (and no SYN or FIN), use snd_max instead of snd_nxt * when filling in ti_seq. But if we are in persist * state, snd_max might reflect one byte beyond the * right edge of the window, so use snd_nxt in that * case, since we know we aren't doing a retransmission. * (retransmit and persist are mutually exclusive...) */ if (sack_rxmit == 0) { if (len || (flags & (TH_SYN|TH_FIN)) || tcp_timer_active(tp, TT_PERSIST)) th->th_seq = htonl(tp->snd_nxt); else th->th_seq = htonl(tp->snd_max); } else { th->th_seq = htonl(p->rxmit); p->rxmit += len; tp->sackhint.sack_bytes_rexmit += len; } th->th_ack = htonl(tp->rcv_nxt); if (optlen) { bcopy(opt, th + 1, optlen); th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; } th->th_flags = flags; /* * Calculate receive window. Don't shrink window, * but avoid silly window syndrome. * If a RST segment is sent, advertise a window of zero. */ if (flags & TH_RST) { recwin = 0; } else { if (recwin < (so->so_rcv.sb_hiwat / 4) && recwin < tp->t_maxseg) recwin = 0; if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && recwin < (tp->rcv_adv - tp->rcv_nxt)) recwin = (tp->rcv_adv - tp->rcv_nxt); } /* * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. The * case is handled in syncache. */ if (flags & TH_SYN) th->th_win = htons((u_short) (min(sbspace(&so->so_rcv), TCP_MAXWIN))); else th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); /* * Adjust the RXWIN0SENT flag - indicate that we have advertised * a 0 window. This may cause the remote transmitter to stall. This * flag tells soreceive() to disable delayed acknowledgements when * draining the buffer. This can occur if the receiver is attempting * to read more data than can be buffered prior to transmitting on * the connection. */ if (th->th_win == 0) { tp->t_sndzerowin++; tp->t_flags |= TF_RXWIN0SENT; } else tp->t_flags &= ~TF_RXWIN0SENT; if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); th->th_flags |= TH_URG; } else /* * If no urgent pointer to send, then we pull * the urgent pointer to the left edge of the send window * so that it doesn't drift into the send window on sequence * number wraparound. */ tp->snd_up = tp->snd_una; /* drag it along */ /* * Put TCP length in extended header, and then * checksum extended header and data. */ m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { /* * Calculate MD5 signature and put it into the place * determined before. * NOTE: since TCP options buffer doesn't point into * mbuf's data, calculate offset and use it. */ if (!TCPMD5_ENABLED() || (error = TCPMD5_OUTPUT(m, th, (u_char *)(th + 1) + (to.to_signature - opt))) != 0) { /* * Do not send segment if the calculation of MD5 * digest has failed. */ m_freem(m); goto out; } } #endif #ifdef INET6 if (isipv6) { /* * There is no need to fill in ip6_plen right now. * It will be filled later by ip6_output. */ m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + optlen + len, IPPROTO_TCP, 0); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); /* IP version must be set here for ipv4/ipv6 checking later */ KASSERT(ip->ip_v == IPVERSION, ("%s: IP version incorrect: %d", __func__, ip->ip_v)); } #endif /* * Enable TSO and specify the size of the segments. * The TCP pseudo header checksum is always provided. */ if (tso) { KASSERT(len > tp->t_maxseg - optlen, ("%s: len <= tso_segsz", __func__)); m->m_pkthdr.csum_flags |= CSUM_TSO; m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen; } KASSERT(len + hdrlen == m_length(m, NULL), ("%s: mbuf chain shorter than expected: %d + %u != %u", __func__, len, hdrlen, m_length(m, NULL))); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */ hhook_run_tcp_est_out(tp, th, &to, len, tso); #endif #ifdef TCPDEBUG /* * Trace. */ if (so->so_options & SO_DEBUG) { u_short save = 0; #ifdef INET6 if (!isipv6) #endif { save = ipov->ih_len; ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */); } tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); #ifdef INET6 if (!isipv6) #endif ipov->ih_len = save; } #endif /* TCPDEBUG */ TCP_PROBE3(debug__output, tp, th, m); /* We're getting ready to send; log now. */ TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_OUT, ERRNO_UNK, len, NULL, false); /* * Fill in IP length and desired time to live and * send to IP level. There should be a better way * to handle ttl and tos; we could keep them in * the template, but need a way to checksum without them. */ /* * m->m_pkthdr.len should have been set before checksum calculation, * because in6_cksum() need it. */ #ifdef INET6 if (isipv6) { /* * we separately set hoplimit for every segment, since the * user might want to change the value via setsockopt. * Also, desired default hop limit might be changed via * Neighbor Discovery. */ ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL); /* * Set the packet size here for the benefit of DTrace probes. * ip6_output() will set it properly; it's supposed to include * the option header lengths as well. */ ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) tp->t_flags2 |= TF2_PLPMTU_PMTUD; else tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); TCP_PROBE5(send, NULL, tp, ip6, tp, th); #ifdef TCPPCAP /* Save packet, if requested. */ tcp_pcap_add(th, m, &(tp->t_outpkts)); #endif /* TODO: IPv6 IP6TOS_ECT bit on */ error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &tp->t_inpcb->inp_route6, ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), NULL, NULL, tp->t_inpcb); - if (error == EMSGSIZE && tp->t_inpcb->inp_route6.ro_rt != NULL) - mtu = tp->t_inpcb->inp_route6.ro_rt->rt_mtu; + if (error == EMSGSIZE && tp->t_inpcb->inp_route6.ro_nh != NULL) + mtu = tp->t_inpcb->inp_route6.ro_nh->nh_mtu; } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(m->m_pkthdr.len); #ifdef INET6 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO) ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL); #endif /* INET6 */ /* * If we do path MTU discovery, then we set DF on every packet. * This might not be the best thing to do according to RFC3390 * Section 2. However the tcp hostcache migitates the problem * so it affects only the first tcp connection with a host. * * NB: Don't set DF on small MTU/MSS to have a safe fallback. */ if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { ip->ip_off |= htons(IP_DF); tp->t_flags2 |= TF2_PLPMTU_PMTUD; } else { tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; } if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); TCP_PROBE5(send, NULL, tp, ip, tp, th); #ifdef TCPPCAP /* Save packet, if requested. */ tcp_pcap_add(th, m, &(tp->t_outpkts)); #endif error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route, ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, tp->t_inpcb); - if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL) - mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu; + if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_nh != NULL) + mtu = tp->t_inpcb->inp_route.ro_nh->nh_mtu; } #endif /* INET */ out: /* * In transmit state, time the transmission and arrange for * the retransmit. In persist state, just set snd_max. */ if ((tp->t_flags & TF_FORCEDATA) == 0 || !tcp_timer_active(tp, TT_PERSIST)) { tcp_seq startseq = tp->snd_nxt; /* * Advance snd_nxt over sequence space of this segment. */ if (flags & (TH_SYN|TH_FIN)) { if (flags & TH_SYN) tp->snd_nxt++; if (flags & TH_FIN) { tp->snd_nxt++; tp->t_flags |= TF_SENTFIN; } } if (sack_rxmit) goto timer; tp->snd_nxt += len; if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { tp->snd_max = tp->snd_nxt; /* * Time this transmission if not a retransmission and * not currently timing anything. */ if (tp->t_rtttime == 0) { tp->t_rtttime = ticks; tp->t_rtseq = startseq; TCPSTAT_INC(tcps_segstimed); } #ifdef STATS if (!(tp->t_flags & TF_GPUTINPROG) && len) { tp->t_flags |= TF_GPUTINPROG; tp->gput_seq = startseq; tp->gput_ack = startseq + ulmin(sbavail(&so->so_snd) - off, sendwin); tp->gput_ts = tcp_ts_getticks(); } #endif /* STATS */ } /* * Set retransmit timer if not currently set, * and not doing a pure ack or a keep-alive probe. * Initial value for retransmit timer is smoothed * round-trip time + 2 * round-trip time variance. * Initialize shift counter which is used for backoff * of retransmit time. */ timer: if (!tcp_timer_active(tp, TT_REXMT) && ((sack_rxmit && tp->snd_nxt != tp->snd_max) || (tp->snd_nxt != tp->snd_una))) { if (tcp_timer_active(tp, TT_PERSIST)) { tcp_timer_activate(tp, TT_PERSIST, 0); tp->t_rxtshift = 0; } tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); } else if (len == 0 && sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { /* * Avoid a situation where we do not set persist timer * after a zero window condition. For example: * 1) A -> B: packet with enough data to fill the window * 2) B -> A: ACK for #1 + new data (0 window * advertisement) * 3) A -> B: ACK for #2, 0 len packet * * In this case, A will not activate the persist timer, * because it chose to send a packet. Unless tcp_output * is called for some other reason (delayed ack timer, * another input packet from B, socket syscall), A will * not send zero window probes. * * So, if you send a 0-length packet, but there is data * in the socket buffer, and neither the rexmt or * persist timer is already set, then activate the * persist timer. */ tp->t_rxtshift = 0; tcp_setpersist(tp); } } else { /* * Persist case, update snd_max but since we are in * persist mode (no window) we do not update snd_nxt. */ int xlen = len; if (flags & TH_SYN) ++xlen; if (flags & TH_FIN) { ++xlen; tp->t_flags |= TF_SENTFIN; } if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) tp->snd_max = tp->snd_nxt + xlen; } if ((error == 0) && (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0)) { /* Clean up any DSACK's sent */ tcp_clean_dsack_blocks(tp); } if (error) { /* Record the error. */ TCP_LOG_EVENT(tp, NULL, &so->so_rcv, &so->so_snd, TCP_LOG_OUT, error, 0, NULL, false); /* * We know that the packet was lost, so back out the * sequence number advance, if any. * * If the error is EPERM the packet got blocked by the * local firewall. Normally we should terminate the * connection but the blocking may have been spurious * due to a firewall reconfiguration cycle. So we treat * it like a packet loss and let the retransmit timer and * timeouts do their work over time. * XXX: It is a POLA question whether calling tcp_drop right * away would be the really correct behavior instead. */ if (((tp->t_flags & TF_FORCEDATA) == 0 || !tcp_timer_active(tp, TT_PERSIST)) && ((flags & TH_SYN) == 0) && (error != EPERM)) { if (sack_rxmit) { p->rxmit -= len; tp->sackhint.sack_bytes_rexmit -= len; KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, ("sackhint bytes rtx >= 0")); } else tp->snd_nxt -= len; } SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ switch (error) { case EACCES: case EPERM: tp->t_softerror = error; return (error); case ENOBUFS: TCP_XMIT_TIMER_ASSERT(tp, len, flags); tp->snd_cwnd = tp->t_maxseg; return (0); case EMSGSIZE: /* * For some reason the interface we used initially * to send segments changed to another or lowered * its MTU. * If TSO was active we either got an interface * without TSO capabilits or TSO was turned off. * If we obtained mtu from ip_output() then update * it and try again. */ if (tso) tp->t_flags &= ~TF_TSO; if (mtu != 0) { tcp_mss_update(tp, -1, mtu, NULL, NULL); goto again; } return (error); case EHOSTDOWN: case EHOSTUNREACH: case ENETDOWN: case ENETUNREACH: if (TCPS_HAVERCVDSYN(tp->t_state)) { tp->t_softerror = error; return (0); } /* FALLTHROUGH */ default: return (error); } } TCPSTAT_INC(tcps_sndtotal); /* * Data sent (as far as we can tell). * If this advertises a larger window than any other segment, * then remember the size of the advertised window. * Any pending ACK has now been sent. */ if (SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) tp->rcv_adv = tp->rcv_nxt + recwin; tp->last_ack_sent = tp->rcv_nxt; tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); if (tcp_timer_active(tp, TT_DELACK)) tcp_timer_activate(tp, TT_DELACK, 0); #if 0 /* * This completely breaks TCP if newreno is turned on. What happens * is that if delayed-acks are turned on on the receiver, this code * on the transmitter effectively destroys the TCP window, forcing * it to four packets (1.5Kx4 = 6K window). */ if (sendalot && --maxburst) goto again; #endif if (sendalot) goto again; return (0); } void tcp_setpersist(struct tcpcb *tp) { int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; int tt; tp->t_flags &= ~TF_PREVVALID; if (tcp_timer_active(tp, TT_REXMT)) panic("tcp_setpersist: retransmit pending"); /* * Start/restart persistence timer. */ TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], tcp_persmin, tcp_persmax); tcp_timer_activate(tp, TT_PERSIST, tt); if (tp->t_rxtshift < TCP_MAXRXTSHIFT) tp->t_rxtshift++; } /* * Insert TCP options according to the supplied parameters to the place * optp in a consistent way. Can handle unaligned destinations. * * The order of the option processing is crucial for optimal packing and * alignment for the scarce option space. * * The optimal order for a SYN/SYN-ACK segment is: * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. * * The SACK options should be last. SACK blocks consume 8*n+2 bytes. * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). * At minimum we need 10 bytes (to generate 1 SACK block). If both * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, * we only have 10 bytes for SACK options (40 - (12 + 18)). */ int tcp_addoptions(struct tcpopt *to, u_char *optp) { u_int32_t mask, optlen = 0; for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { if ((to->to_flags & mask) != mask) continue; if (optlen == TCP_MAXOLEN) break; switch (to->to_flags & mask) { case TOF_MSS: while (optlen % 4) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG) continue; optlen += TCPOLEN_MAXSEG; *optp++ = TCPOPT_MAXSEG; *optp++ = TCPOLEN_MAXSEG; to->to_mss = htons(to->to_mss); bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); optp += sizeof(to->to_mss); break; case TOF_SCALE: while (!optlen || optlen % 2 != 1) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW) continue; optlen += TCPOLEN_WINDOW; *optp++ = TCPOPT_WINDOW; *optp++ = TCPOLEN_WINDOW; *optp++ = to->to_wscale; break; case TOF_SACKPERM: while (optlen % 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED) continue; optlen += TCPOLEN_SACK_PERMITTED; *optp++ = TCPOPT_SACK_PERMITTED; *optp++ = TCPOLEN_SACK_PERMITTED; break; case TOF_TS: while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP) continue; optlen += TCPOLEN_TIMESTAMP; *optp++ = TCPOPT_TIMESTAMP; *optp++ = TCPOLEN_TIMESTAMP; to->to_tsval = htonl(to->to_tsval); to->to_tsecr = htonl(to->to_tsecr); bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); optp += sizeof(to->to_tsval); bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); optp += sizeof(to->to_tsecr); break; case TOF_SIGNATURE: { int siglen = TCPOLEN_SIGNATURE - 2; while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) { to->to_flags &= ~TOF_SIGNATURE; continue; } optlen += TCPOLEN_SIGNATURE; *optp++ = TCPOPT_SIGNATURE; *optp++ = TCPOLEN_SIGNATURE; to->to_signature = optp; while (siglen--) *optp++ = 0; break; } case TOF_SACK: { int sackblks = 0; struct sackblk *sack = (struct sackblk *)to->to_sacks; tcp_seq sack_seq; while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK) continue; optlen += TCPOLEN_SACKHDR; *optp++ = TCPOPT_SACK; sackblks = min(to->to_nsacks, (TCP_MAXOLEN - optlen) / TCPOLEN_SACK); *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; while (sackblks--) { sack_seq = htonl(sack->start); bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); optp += sizeof(sack_seq); sack_seq = htonl(sack->end); bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); optp += sizeof(sack_seq); optlen += TCPOLEN_SACK; sack++; } TCPSTAT_INC(tcps_sack_send_blocks); break; } case TOF_FASTOPEN: { int total_len; /* XXX is there any point to aligning this option? */ total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len; if (TCP_MAXOLEN - optlen < total_len) { to->to_flags &= ~TOF_FASTOPEN; continue; } *optp++ = TCPOPT_FAST_OPEN; *optp++ = total_len; if (to->to_tfo_len > 0) { bcopy(to->to_tfo_cookie, optp, to->to_tfo_len); optp += to->to_tfo_len; } optlen += total_len; break; } default: panic("%s: unknown TCP option type", __func__); break; } } /* Terminate and pad TCP options to a 4 byte boundary. */ if (optlen % 4) { optlen += TCPOLEN_EOL; *optp++ = TCPOPT_EOL; } /* * According to RFC 793 (STD0007): * "The content of the header beyond the End-of-Option option * must be header padding (i.e., zero)." * and later: "The padding is composed of zeros." */ while (optlen % 4) { optlen += TCPOLEN_PAD; *optp++ = TCPOPT_PAD; } KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__)); return (optlen); } /* * This is a copy of m_copym(), taking the TSO segment size/limit * constraints into account, and advancing the sndptr as it goes. */ struct mbuf * tcp_m_copym(struct mbuf *m, int32_t off0, int32_t *plen, int32_t seglimit, int32_t segsize, struct sockbuf *sb, bool hw_tls) { #ifdef KERN_TLS struct ktls_session *tls, *ntls; struct mbuf *start; #endif struct mbuf *n, **np; struct mbuf *top; int32_t off = off0; int32_t len = *plen; int32_t fragsize; int32_t len_cp = 0; int32_t *pkthdrlen; uint32_t mlen, frags; bool copyhdr; KASSERT(off >= 0, ("tcp_m_copym, negative off %d", off)); KASSERT(len >= 0, ("tcp_m_copym, negative len %d", len)); if (off == 0 && m->m_flags & M_PKTHDR) copyhdr = true; else copyhdr = false; while (off > 0) { KASSERT(m != NULL, ("tcp_m_copym, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; if ((sb) && (m == sb->sb_sndptr)) { sb->sb_sndptroff += m->m_len; sb->sb_sndptr = m->m_next; } m = m->m_next; } np = ⊤ top = NULL; pkthdrlen = NULL; #ifdef KERN_TLS if (hw_tls && (m->m_flags & M_NOMAP)) tls = m->m_ext_pgs.tls; else tls = NULL; start = m; #endif while (len > 0) { if (m == NULL) { KASSERT(len == M_COPYALL, ("tcp_m_copym, length > size of mbuf chain")); *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } #ifdef KERN_TLS if (hw_tls) { if (m->m_flags & M_NOMAP) ntls = m->m_ext_pgs.tls; else ntls = NULL; /* * Avoid mixing TLS records with handshake * data or TLS records from different * sessions. */ if (tls != ntls) { MPASS(m != start); *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } /* * Don't end a send in the middle of a TLS * record if it spans multiple TLS records. */ if (tls != NULL && (m != start) && len < m->m_len) { *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } } #endif mlen = min(len, m->m_len - off); if (seglimit) { /* * For M_NOMAP mbufs, add 3 segments * + 1 in case we are crossing page boundaries * + 2 in case the TLS hdr/trailer are used * It is cheaper to just add the segments * than it is to take the cache miss to look * at the mbuf ext_pgs state in detail. */ if (m->m_flags & M_NOMAP) { fragsize = min(segsize, PAGE_SIZE); frags = 3; } else { fragsize = segsize; frags = 0; } /* Break if we really can't fit anymore. */ if ((frags + 1) >= seglimit) { *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } /* * Reduce size if you can't copy the whole * mbuf. If we can't copy the whole mbuf, also * adjust len so the loop will end after this * mbuf. */ if ((frags + howmany(mlen, fragsize)) >= seglimit) { mlen = (seglimit - frags - 1) * fragsize; len = mlen; *plen = len_cp + len; if (pkthdrlen != NULL) *pkthdrlen = *plen; } frags += howmany(mlen, fragsize); if (frags == 0) frags++; seglimit -= frags; KASSERT(seglimit > 0, ("%s: seglimit went too low", __func__)); } if (copyhdr) n = m_gethdr(M_NOWAIT, m->m_type); else n = m_get(M_NOWAIT, m->m_type); *np = n; if (n == NULL) goto nospace; if (copyhdr) { if (!m_dup_pkthdr(n, m, M_NOWAIT)) goto nospace; if (len == M_COPYALL) n->m_pkthdr.len -= off0; else n->m_pkthdr.len = len; pkthdrlen = &n->m_pkthdr.len; copyhdr = false; } n->m_len = mlen; len_cp += n->m_len; if (m->m_flags & M_EXT) { n->m_data = m->m_data + off; mb_dupcl(n, m); } else bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), (u_int)n->m_len); if (sb && (sb->sb_sndptr == m) && ((n->m_len + off) >= m->m_len) && m->m_next) { sb->sb_sndptroff += m->m_len; sb->sb_sndptr = m->m_next; } off = 0; if (len != M_COPYALL) { len -= n->m_len; } m = m->m_next; np = &n->m_next; } return (top); nospace: m_freem(top); return (NULL); } void tcp_sndbuf_autoscale(struct tcpcb *tp, struct socket *so, uint32_t sendwin) { /* * Automatic sizing of send socket buffer. Often the send buffer * size is not optimally adjusted to the actual network conditions * at hand (delay bandwidth product). Setting the buffer size too * small limits throughput on links with high bandwidth and high * delay (eg. trans-continental/oceanic links). Setting the * buffer size too big consumes too much real kernel memory, * especially with many connections on busy servers. * * The criteria to step up the send buffer one notch are: * 1. receive window of remote host is larger than send buffer * (with a fudge factor of 5/4th); * 2. send buffer is filled to 7/8th with data (so we actually * have data to make use of it); * 3. send buffer fill has not hit maximal automatic size; * 4. our send window (slow start and cogestion controlled) is * larger than sent but unacknowledged data in send buffer. * * The remote host receive window scaling factor may limit the * growing of the send buffer before it reaches its allowed * maximum. * * It scales directly with slow start or congestion window * and does at most one step per received ACK. This fast * scaling has the drawback of growing the send buffer beyond * what is strictly necessary to make full use of a given * delay*bandwidth product. However testing has shown this not * to be much of an problem. At worst we are trading wasting * of available bandwidth (the non-use of it) for wasting some * socket buffer memory. * * TODO: Shrink send buffer during idle periods together * with congestion window. Requires another timer. Has to * wait for upcoming tcp timer rewrite. * * XXXGL: should there be used sbused() or sbavail()? */ if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { int lowat; lowat = V_tcp_sendbuf_auto_lowat ? so->so_snd.sb_lowat : 0; if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat - lowat && sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) - lowat && sbused(&so->so_snd) < V_tcp_autosndbuf_max && sendwin >= (sbused(&so->so_snd) - (tp->snd_nxt - tp->snd_una))) { if (!sbreserve_locked(&so->so_snd, min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc, V_tcp_autosndbuf_max), so, curthread)) so->so_snd.sb_flags &= ~SB_AUTOSIZE; } } } Index: head/sys/netinet/tcp_stacks/bbr.c =================================================================== --- head/sys/netinet/tcp_stacks/bbr.c (revision 360291) +++ head/sys/netinet/tcp_stacks/bbr.c (revision 360292) @@ -1,15205 +1,15205 @@ /*- * Copyright (c) 2016-9 * Netflix Inc. * All rights reserved. * * 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. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * */ /** * Author: Randall Stewart * This work is based on the ACM Queue paper * BBR - Congestion Based Congestion Control * and also numerous discussions with Neal, Yuchung and Van. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include "opt_ratelimit.h" #include "opt_kern_tls.h" #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #include #include #include #include #ifdef KERN_TLS #include #endif #include #include #ifdef STATS #include #include #include /* Must come after qmath.h and tree.h */ #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #define TCPOUTFLAGS #include #include #include #include #include #include #include #include #include #include #include #ifdef TCPDEBUG #include #endif /* TCPDEBUG */ #ifdef TCP_OFFLOAD #include #endif #ifdef INET6 #include #endif #include #include #include #include #include #if defined(IPSEC) || defined(IPSEC_SUPPORT) #include #include #endif /* IPSEC */ #include #include #include #ifdef MAC #include #endif #include "sack_filter.h" #include "tcp_bbr.h" #include "rack_bbr_common.h" uma_zone_t bbr_zone; uma_zone_t bbr_pcb_zone; struct sysctl_ctx_list bbr_sysctl_ctx; struct sysctl_oid *bbr_sysctl_root; #define TCPT_RANGESET_NOSLOP(tv, value, tvmin, tvmax) do { \ (tv) = (value); \ if ((u_long)(tv) < (u_long)(tvmin)) \ (tv) = (tvmin); \ if ((u_long)(tv) > (u_long)(tvmax)) \ (tv) = (tvmax); \ } while(0) /*#define BBR_INVARIANT 1*/ /* * initial window */ static uint32_t bbr_def_init_win = 10; static int32_t bbr_persist_min = 250000; /* 250ms */ static int32_t bbr_persist_max = 1000000; /* 1 Second */ static int32_t bbr_cwnd_may_shrink = 0; static int32_t bbr_cwndtarget_rtt_touse = BBR_RTT_PROP; static int32_t bbr_num_pktepo_for_del_limit = BBR_NUM_RTTS_FOR_DEL_LIMIT; static int32_t bbr_hardware_pacing_limit = 8000; static int32_t bbr_quanta = 3; /* How much extra quanta do we get? */ static int32_t bbr_no_retran = 0; static int32_t bbr_error_base_paceout = 10000; /* usec to pace */ static int32_t bbr_max_net_error_cnt = 10; /* Should the following be dynamic too -- loss wise */ static int32_t bbr_rtt_gain_thresh = 0; /* Measurement controls */ static int32_t bbr_use_google_algo = 1; static int32_t bbr_ts_limiting = 1; static int32_t bbr_ts_can_raise = 0; static int32_t bbr_do_red = 600; static int32_t bbr_red_scale = 20000; static int32_t bbr_red_mul = 1; static int32_t bbr_red_div = 2; static int32_t bbr_red_growth_restrict = 1; static int32_t bbr_target_is_bbunit = 0; static int32_t bbr_drop_limit = 0; /* * How much gain do we need to see to * stay in startup? */ static int32_t bbr_marks_rxt_sack_passed = 0; static int32_t bbr_start_exit = 25; static int32_t bbr_low_start_exit = 25; /* When we are in reduced gain */ static int32_t bbr_startup_loss_thresh = 2000; /* 20.00% loss */ static int32_t bbr_hptsi_max_mul = 1; /* These two mul/div assure a min pacing */ static int32_t bbr_hptsi_max_div = 2; /* time, 0 means turned off. We need this * if we go back ever to where the pacer * has priority over timers. */ static int32_t bbr_policer_call_from_rack_to = 0; static int32_t bbr_policer_detection_enabled = 1; static int32_t bbr_min_measurements_req = 1; /* We need at least 2 * measurments before we are * "good" note that 2 == 1. * This is because we use a > * comparison. This means if * min_measure was 0, it takes * num-measures > min(0) and * you get 1 measurement and * you are good. Set to 1, you * have to have two * measurements (this is done * to prevent it from being ok * to have no measurements). */ static int32_t bbr_no_pacing_until = 4; static int32_t bbr_min_usec_delta = 20000; /* 20,000 usecs */ static int32_t bbr_min_peer_delta = 20; /* 20 units */ static int32_t bbr_delta_percent = 150; /* 15.0 % */ static int32_t bbr_target_cwnd_mult_limit = 8; /* * bbr_cwnd_min_val is the number of * segments we hold to in the RTT probe * state typically 4. */ static int32_t bbr_cwnd_min_val = BBR_PROBERTT_NUM_MSS; static int32_t bbr_cwnd_min_val_hs = BBR_HIGHSPEED_NUM_MSS; static int32_t bbr_gain_to_target = 1; static int32_t bbr_gain_gets_extra_too = 1; /* * bbr_high_gain is the 2/ln(2) value we need * to double the sending rate in startup. This * is used for both cwnd and hptsi gain's. */ static int32_t bbr_high_gain = BBR_UNIT * 2885 / 1000 + 1; static int32_t bbr_startup_lower = BBR_UNIT * 1500 / 1000 + 1; static int32_t bbr_use_lower_gain_in_startup = 1; /* thresholds for reduction on drain in sub-states/drain */ static int32_t bbr_drain_rtt = BBR_SRTT; static int32_t bbr_drain_floor = 88; static int32_t google_allow_early_out = 1; static int32_t google_consider_lost = 1; static int32_t bbr_drain_drop_mul = 4; static int32_t bbr_drain_drop_div = 5; static int32_t bbr_rand_ot = 50; static int32_t bbr_can_force_probertt = 0; static int32_t bbr_can_adjust_probertt = 1; static int32_t bbr_probertt_sets_rtt = 0; static int32_t bbr_can_use_ts_for_rtt = 1; static int32_t bbr_is_ratio = 0; static int32_t bbr_sub_drain_app_limit = 1; static int32_t bbr_prtt_slam_cwnd = 1; static int32_t bbr_sub_drain_slam_cwnd = 1; static int32_t bbr_slam_cwnd_in_main_drain = 1; static int32_t bbr_filter_len_sec = 6; /* How long does the rttProp filter * hold */ static uint32_t bbr_rtt_probe_limit = (USECS_IN_SECOND * 4); /* * bbr_drain_gain is the reverse of the high_gain * designed to drain back out the standing queue * that is formed in startup by causing a larger * hptsi gain and thus drainging the packets * in flight. */ static int32_t bbr_drain_gain = BBR_UNIT * 1000 / 2885; static int32_t bbr_rttprobe_gain = 192; /* * The cwnd_gain is the default cwnd gain applied when * calculating a target cwnd. Note that the cwnd is * a secondary factor in the way BBR works (see the * paper and think about it, it will take some time). * Basically the hptsi_gain spreads the packets out * so you never get more than BDP to the peer even * if the cwnd is high. In our implemenation that * means in non-recovery/retransmission scenarios * cwnd will never be reached by the flight-size. */ static int32_t bbr_cwnd_gain = BBR_UNIT * 2; static int32_t bbr_tlp_type_to_use = BBR_SRTT; static int32_t bbr_delack_time = 100000; /* 100ms in useconds */ static int32_t bbr_sack_not_required = 0; /* set to one to allow non-sack to use bbr */ static int32_t bbr_initial_bw_bps = 62500; /* 500kbps in bytes ps */ static int32_t bbr_ignore_data_after_close = 1; static int16_t bbr_hptsi_gain[] = { (BBR_UNIT *5 / 4), (BBR_UNIT * 3 / 4), BBR_UNIT, BBR_UNIT, BBR_UNIT, BBR_UNIT, BBR_UNIT, BBR_UNIT }; int32_t bbr_use_rack_resend_cheat = 1; int32_t bbr_sends_full_iwnd = 1; #define BBR_HPTSI_GAIN_MAX 8 /* * The BBR module incorporates a number of * TCP ideas that have been put out into the IETF * over the last few years: * - Yuchung Cheng's RACK TCP (for which its named) that * will stop us using the number of dup acks and instead * use time as the gage of when we retransmit. * - Reorder Detection of RFC4737 and the Tail-Loss probe draft * of Dukkipati et.al. * - Van Jacobson's et.al BBR. * * RACK depends on SACK, so if an endpoint arrives that * cannot do SACK the state machine below will shuttle the * connection back to using the "default" TCP stack that is * in FreeBSD. * * To implement BBR and RACK the original TCP stack was first decomposed * into a functional state machine with individual states * for each of the possible TCP connection states. The do_segement * functions role in life is to mandate the connection supports SACK * initially and then assure that the RACK state matches the conenction * state before calling the states do_segment function. Data processing * of inbound segments also now happens in the hpts_do_segment in general * with only one exception. This is so we can keep the connection on * a single CPU. * * Each state is simplified due to the fact that the original do_segment * has been decomposed and we *know* what state we are in (no * switches on the state) and all tests for SACK are gone. This * greatly simplifies what each state does. * * TCP output is also over-written with a new version since it * must maintain the new rack scoreboard and has had hptsi * integrated as a requirment. Still todo is to eliminate the * use of the callout_() system and use the hpts for all * timers as well. */ static uint32_t bbr_rtt_probe_time = 200000; /* 200ms in micro seconds */ static uint32_t bbr_rtt_probe_cwndtarg = 4; /* How many mss's outstanding */ static const int32_t bbr_min_req_free = 2; /* The min we must have on the * free list */ static int32_t bbr_tlp_thresh = 1; static int32_t bbr_reorder_thresh = 2; static int32_t bbr_reorder_fade = 60000000; /* 0 - never fade, def * 60,000,000 - 60 seconds */ static int32_t bbr_pkt_delay = 1000; static int32_t bbr_min_to = 1000; /* Number of usec's minimum timeout */ static int32_t bbr_incr_timers = 1; static int32_t bbr_tlp_min = 10000; /* 10ms in usecs */ static int32_t bbr_delayed_ack_time = 200000; /* 200ms in usecs */ static int32_t bbr_exit_startup_at_loss = 1; /* * bbr_lt_bw_ratio is 1/8th * bbr_lt_bw_diff is < 4 Kbit/sec */ static uint64_t bbr_lt_bw_diff = 4000 / 8; /* In bytes per second */ static uint64_t bbr_lt_bw_ratio = 8; /* For 1/8th */ static uint32_t bbr_lt_bw_max_rtts = 48; /* How many rtt's do we use * the lt_bw for */ static uint32_t bbr_lt_intvl_min_rtts = 4; /* Min num of RTT's to measure * lt_bw */ static int32_t bbr_lt_intvl_fp = 0; /* False positive epoch diff */ static int32_t bbr_lt_loss_thresh = 196; /* Lost vs delivered % */ static int32_t bbr_lt_fd_thresh = 100; /* false detection % */ static int32_t bbr_verbose_logging = 0; /* * Currently regular tcp has a rto_min of 30ms * the backoff goes 12 times so that ends up * being a total of 122.850 seconds before a * connection is killed. */ static int32_t bbr_rto_min_ms = 30; /* 30ms same as main freebsd */ static int32_t bbr_rto_max_sec = 4; /* 4 seconds */ /****************************************************/ /* DEFAULT TSO SIZING (cpu performance impacting) */ /****************************************************/ /* What amount is our formula using to get TSO size */ static int32_t bbr_hptsi_per_second = 1000; /* * For hptsi under bbr_cross_over connections what is delay * target 7ms (in usec) combined with a seg_max of 2 * gets us close to identical google behavior in * TSO size selection (possibly more 1MSS sends). */ static int32_t bbr_hptsi_segments_delay_tar = 7000; /* Does pacing delay include overhead's in its time calculations? */ static int32_t bbr_include_enet_oh = 0; static int32_t bbr_include_ip_oh = 1; static int32_t bbr_include_tcp_oh = 1; static int32_t bbr_google_discount = 10; /* Do we use (nf mode) pkt-epoch to drive us or rttProp? */ static int32_t bbr_state_is_pkt_epoch = 0; static int32_t bbr_state_drain_2_tar = 1; /* What is the max the 0 - bbr_cross_over MBPS TSO target * can reach using our delay target. Note that this * value becomes the floor for the cross over * algorithm. */ static int32_t bbr_hptsi_segments_max = 2; static int32_t bbr_hptsi_segments_floor = 1; static int32_t bbr_hptsi_utter_max = 0; /* What is the min the 0 - bbr_cross-over MBPS TSO target can be */ static int32_t bbr_hptsi_bytes_min = 1460; static int32_t bbr_all_get_min = 0; /* Cross over point from algo-a to algo-b */ static uint32_t bbr_cross_over = TWENTY_THREE_MBPS; /* Do we deal with our restart state? */ static int32_t bbr_uses_idle_restart = 0; static int32_t bbr_idle_restart_threshold = 100000; /* 100ms in useconds */ /* Do we allow hardware pacing? */ static int32_t bbr_allow_hdwr_pacing = 0; static int32_t bbr_hdwr_pace_adjust = 2; /* multipler when we calc the tso size */ static int32_t bbr_hdwr_pace_floor = 1; static int32_t bbr_hdwr_pacing_delay_cnt = 10; /****************************************************/ static int32_t bbr_resends_use_tso = 0; static int32_t bbr_tlp_max_resend = 2; static int32_t bbr_sack_block_limit = 128; #define BBR_MAX_STAT 19 counter_u64_t bbr_state_time[BBR_MAX_STAT]; counter_u64_t bbr_state_lost[BBR_MAX_STAT]; counter_u64_t bbr_state_resend[BBR_MAX_STAT]; counter_u64_t bbr_stat_arry[BBR_STAT_SIZE]; counter_u64_t bbr_opts_arry[BBR_OPTS_SIZE]; counter_u64_t bbr_out_size[TCP_MSS_ACCT_SIZE]; counter_u64_t bbr_flows_whdwr_pacing; counter_u64_t bbr_flows_nohdwr_pacing; counter_u64_t bbr_nohdwr_pacing_enobuf; counter_u64_t bbr_hdwr_pacing_enobuf; static inline uint64_t bbr_get_bw(struct tcp_bbr *bbr); /* * Static defintions we need for forward declarations. */ static uint32_t bbr_get_pacing_length(struct tcp_bbr *bbr, uint16_t gain, uint32_t useconds_time, uint64_t bw); static uint32_t bbr_get_a_state_target(struct tcp_bbr *bbr, uint32_t gain); static void bbr_set_state(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t win); static void bbr_set_probebw_gains(struct tcp_bbr *bbr, uint32_t cts, uint32_t losses); static void bbr_substate_change(struct tcp_bbr *bbr, uint32_t cts, int line, int dolog); static uint32_t bbr_get_target_cwnd(struct tcp_bbr *bbr, uint64_t bw, uint32_t gain); static void bbr_state_change(struct tcp_bbr *bbr, uint32_t cts, int32_t epoch, int32_t pkt_epoch, uint32_t losses); static uint32_t bbr_calc_thresh_rack(struct tcp_bbr *bbr, uint32_t srtt, uint32_t cts, struct bbr_sendmap *rsm); static uint32_t bbr_initial_cwnd(struct tcp_bbr *bbr, struct tcpcb *tp); static uint32_t bbr_calc_thresh_tlp(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t srtt, uint32_t cts); static void bbr_exit_persist(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, int32_t line); static void bbr_set_state_target(struct tcp_bbr *bbr, int line); static void bbr_enter_probe_rtt(struct tcp_bbr *bbr, uint32_t cts, int32_t line); static void bbr_log_progress_event(struct tcp_bbr *bbr, struct tcpcb *tp, uint32_t tick, int event, int line); static void tcp_bbr_tso_size_check(struct tcp_bbr *bbr, uint32_t cts); static void bbr_setup_red_bw(struct tcp_bbr *bbr, uint32_t cts); static void bbr_log_rtt_shrinks(struct tcp_bbr *bbr, uint32_t cts, uint32_t applied, uint32_t rtt, uint32_t line, uint8_t is_start, uint16_t set); static struct bbr_sendmap * bbr_find_lowest_rsm(struct tcp_bbr *bbr); static __inline uint32_t bbr_get_rtt(struct tcp_bbr *bbr, int32_t rtt_type); static void bbr_log_to_start(struct tcp_bbr *bbr, uint32_t cts, uint32_t to, int32_t slot, uint8_t which); static void bbr_log_timer_var(struct tcp_bbr *bbr, int mode, uint32_t cts, uint32_t time_since_sent, uint32_t srtt, uint32_t thresh, uint32_t to); static void bbr_log_hpts_diag(struct tcp_bbr *bbr, uint32_t cts, struct hpts_diag *diag); static void bbr_log_type_bbrsnd(struct tcp_bbr *bbr, uint32_t len, uint32_t slot, uint32_t del_by, uint32_t cts, uint32_t sloton, uint32_t prev_delay); static void bbr_enter_persist(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, int32_t line); static void bbr_stop_all_timers(struct tcpcb *tp); static void bbr_exit_probe_rtt(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts); static void bbr_check_probe_rtt_limits(struct tcp_bbr *bbr, uint32_t cts); static void bbr_timer_cancel(struct tcp_bbr *bbr, int32_t line, uint32_t cts); static void bbr_log_pacing_delay_calc(struct tcp_bbr *bbr, uint16_t gain, uint32_t len, uint32_t cts, uint32_t usecs, uint64_t bw, uint32_t override, int mod); static inline uint8_t bbr_state_val(struct tcp_bbr *bbr) { return(bbr->rc_bbr_substate); } static inline uint32_t get_min_cwnd(struct tcp_bbr *bbr) { int mss; mss = min((bbr->rc_tp->t_maxseg - bbr->rc_last_options), bbr->r_ctl.rc_pace_max_segs); if (bbr_get_rtt(bbr, BBR_RTT_PROP) < BBR_HIGH_SPEED) return (bbr_cwnd_min_val_hs * mss); else return (bbr_cwnd_min_val * mss); } static uint32_t bbr_get_persists_timer_val(struct tcpcb *tp, struct tcp_bbr *bbr) { uint64_t srtt, var; uint64_t ret_val; bbr->r_ctl.rc_hpts_flags |= PACE_TMR_PERSIT; if (tp->t_srtt == 0) { srtt = (uint64_t)BBR_INITIAL_RTO; var = 0; } else { srtt = ((uint64_t)TICKS_2_USEC(tp->t_srtt) >> TCP_RTT_SHIFT); var = ((uint64_t)TICKS_2_USEC(tp->t_rttvar) >> TCP_RTT_SHIFT); } TCPT_RANGESET_NOSLOP(ret_val, ((srtt + var) * tcp_backoff[tp->t_rxtshift]), bbr_persist_min, bbr_persist_max); return ((uint32_t)ret_val); } static uint32_t bbr_timer_start(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { /* * Start the FR timer, we do this based on getting the first one in * the rc_tmap. Note that if its NULL we must stop the timer. in all * events we need to stop the running timer (if its running) before * starting the new one. */ uint32_t thresh, exp, to, srtt, time_since_sent, tstmp_touse; int32_t idx; int32_t is_tlp_timer = 0; struct bbr_sendmap *rsm; if (bbr->rc_all_timers_stopped) { /* All timers have been stopped none are to run */ return (0); } if (bbr->rc_in_persist) { /* We can't start any timer in persists */ return (bbr_get_persists_timer_val(tp, bbr)); } rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if ((rsm == NULL) || ((tp->t_flags & TF_SACK_PERMIT) == 0) || (tp->t_state < TCPS_ESTABLISHED)) { /* Nothing on the send map */ activate_rxt: if (SEQ_LT(tp->snd_una, tp->snd_max) || sbavail(&(tp->t_inpcb->inp_socket->so_snd))) { uint64_t tov; time_since_sent = 0; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if (rsm) { idx = rsm->r_rtr_cnt - 1; if (TSTMP_GEQ(rsm->r_tim_lastsent[idx], bbr->r_ctl.rc_tlp_rxt_last_time)) tstmp_touse = rsm->r_tim_lastsent[idx]; else tstmp_touse = bbr->r_ctl.rc_tlp_rxt_last_time; if (TSTMP_GT(tstmp_touse, cts)) time_since_sent = cts - tstmp_touse; } bbr->r_ctl.rc_hpts_flags |= PACE_TMR_RXT; if (tp->t_srtt == 0) tov = BBR_INITIAL_RTO; else tov = ((uint64_t)(TICKS_2_USEC(tp->t_srtt) + ((uint64_t)TICKS_2_USEC(tp->t_rttvar) * (uint64_t)4)) >> TCP_RTT_SHIFT); if (tp->t_rxtshift) tov *= tcp_backoff[tp->t_rxtshift]; if (tov > time_since_sent) tov -= time_since_sent; else tov = bbr->r_ctl.rc_min_to; TCPT_RANGESET_NOSLOP(to, tov, (bbr->r_ctl.rc_min_rto_ms * MS_IN_USEC), (bbr->rc_max_rto_sec * USECS_IN_SECOND)); bbr_log_timer_var(bbr, 2, cts, 0, srtt, 0, to); return (to); } return (0); } if (rsm->r_flags & BBR_ACKED) { rsm = bbr_find_lowest_rsm(bbr); if (rsm == NULL) { /* No lowest? */ goto activate_rxt; } } /* Convert from ms to usecs */ if (rsm->r_flags & BBR_SACK_PASSED) { if ((tp->t_flags & TF_SENTFIN) && ((tp->snd_max - tp->snd_una) == 1) && (rsm->r_flags & BBR_HAS_FIN)) { /* * We don't start a bbr rack timer if all we have is * a FIN outstanding. */ goto activate_rxt; } srtt = bbr_get_rtt(bbr, BBR_RTT_RACK); thresh = bbr_calc_thresh_rack(bbr, srtt, cts, rsm); idx = rsm->r_rtr_cnt - 1; exp = rsm->r_tim_lastsent[idx] + thresh; if (SEQ_GEQ(exp, cts)) { to = exp - cts; if (to < bbr->r_ctl.rc_min_to) { to = bbr->r_ctl.rc_min_to; } } else { to = bbr->r_ctl.rc_min_to; } } else { /* Ok we need to do a TLP not RACK */ if (bbr->rc_tlp_in_progress != 0) { /* * The previous send was a TLP. */ goto activate_rxt; } rsm = TAILQ_LAST_FAST(&bbr->r_ctl.rc_tmap, bbr_sendmap, r_tnext); if (rsm == NULL) { /* We found no rsm to TLP with. */ goto activate_rxt; } if (rsm->r_flags & BBR_HAS_FIN) { /* If its a FIN we don't do TLP */ rsm = NULL; goto activate_rxt; } time_since_sent = 0; idx = rsm->r_rtr_cnt - 1; if (TSTMP_GEQ(rsm->r_tim_lastsent[idx], bbr->r_ctl.rc_tlp_rxt_last_time)) tstmp_touse = rsm->r_tim_lastsent[idx]; else tstmp_touse = bbr->r_ctl.rc_tlp_rxt_last_time; if (TSTMP_GT(tstmp_touse, cts)) time_since_sent = cts - tstmp_touse; is_tlp_timer = 1; srtt = bbr_get_rtt(bbr, bbr_tlp_type_to_use); thresh = bbr_calc_thresh_tlp(tp, bbr, rsm, srtt, cts); if (thresh > time_since_sent) to = thresh - time_since_sent; else to = bbr->r_ctl.rc_min_to; if (to > (((uint32_t)bbr->rc_max_rto_sec) * USECS_IN_SECOND)) { /* * If the TLP time works out to larger than the max * RTO lets not do TLP.. just RTO. */ goto activate_rxt; } if ((bbr->rc_tlp_rtx_out == 1) && (rsm->r_start == bbr->r_ctl.rc_last_tlp_seq)) { /* * Second retransmit of the same TLP * lets not. */ bbr->rc_tlp_rtx_out = 0; goto activate_rxt; } if (rsm->r_start != bbr->r_ctl.rc_last_tlp_seq) { /* * The tail is no longer the last one I did a probe * on */ bbr->r_ctl.rc_tlp_seg_send_cnt = 0; bbr->r_ctl.rc_last_tlp_seq = rsm->r_start; } } if (is_tlp_timer == 0) { BBR_STAT_INC(bbr_to_arm_rack); bbr->r_ctl.rc_hpts_flags |= PACE_TMR_RACK; } else { bbr_log_timer_var(bbr, 1, cts, time_since_sent, srtt, thresh, to); if (bbr->r_ctl.rc_tlp_seg_send_cnt > bbr_tlp_max_resend) { /* * We have exceeded how many times we can retran the * current TLP timer, switch to the RTO timer. */ goto activate_rxt; } else { BBR_STAT_INC(bbr_to_arm_tlp); bbr->r_ctl.rc_hpts_flags |= PACE_TMR_TLP; } } return (to); } static inline int32_t bbr_minseg(struct tcp_bbr *bbr) { return (bbr->r_ctl.rc_pace_min_segs - bbr->rc_last_options); } static void bbr_start_hpts_timer(struct tcp_bbr *bbr, struct tcpcb *tp, uint32_t cts, int32_t frm, int32_t slot, uint32_t tot_len) { struct inpcb *inp; struct hpts_diag diag; uint32_t delayed_ack = 0; uint32_t left = 0; uint32_t hpts_timeout; uint8_t stopped; int32_t delay_calc = 0; uint32_t prev_delay = 0; inp = tp->t_inpcb; if (inp->inp_in_hpts) { /* A previous call is already set up */ return; } if ((tp->t_state == TCPS_CLOSED) || (tp->t_state == TCPS_LISTEN)) { return; } stopped = bbr->rc_tmr_stopped; if (stopped && TSTMP_GT(bbr->r_ctl.rc_timer_exp, cts)) { left = bbr->r_ctl.rc_timer_exp - cts; } bbr->r_ctl.rc_hpts_flags = 0; bbr->r_ctl.rc_timer_exp = 0; prev_delay = bbr->r_ctl.rc_last_delay_val; if (bbr->r_ctl.rc_last_delay_val && (slot == 0)) { /* * If a previous pacer delay was in place we * are not coming from the output side (where * we calculate a delay, more likely a timer). */ slot = bbr->r_ctl.rc_last_delay_val; if (TSTMP_GT(cts, bbr->rc_pacer_started)) { /* Compensate for time passed */ delay_calc = cts - bbr->rc_pacer_started; if (delay_calc <= slot) slot -= delay_calc; } } /* Do we have early to make up for by pushing out the pacing time? */ if (bbr->r_agg_early_set) { bbr_log_pacing_delay_calc(bbr, 0, bbr->r_ctl.rc_agg_early, cts, slot, 0, bbr->r_agg_early_set, 2); slot += bbr->r_ctl.rc_agg_early; bbr->r_ctl.rc_agg_early = 0; bbr->r_agg_early_set = 0; } /* Are we running a total debt that needs to be compensated for? */ if (bbr->r_ctl.rc_hptsi_agg_delay) { if (slot > bbr->r_ctl.rc_hptsi_agg_delay) { /* We nuke the delay */ slot -= bbr->r_ctl.rc_hptsi_agg_delay; bbr->r_ctl.rc_hptsi_agg_delay = 0; } else { /* We nuke some of the delay, put in a minimal 100usecs */ bbr->r_ctl.rc_hptsi_agg_delay -= slot; bbr->r_ctl.rc_last_delay_val = slot = 100; } } bbr->r_ctl.rc_last_delay_val = slot; hpts_timeout = bbr_timer_start(tp, bbr, cts); if (tp->t_flags & TF_DELACK) { if (bbr->rc_in_persist == 0) { delayed_ack = bbr_delack_time; } else { /* * We are in persists and have * gotten a new data element. */ if (hpts_timeout > bbr_delack_time) { /* * Lets make the persists timer (which acks) * be the smaller of hpts_timeout and bbr_delack_time. */ hpts_timeout = bbr_delack_time; } } } if (delayed_ack && ((hpts_timeout == 0) || (delayed_ack < hpts_timeout))) { /* We need a Delayed ack timer */ bbr->r_ctl.rc_hpts_flags = PACE_TMR_DELACK; hpts_timeout = delayed_ack; } if (slot) { /* Mark that we have a pacing timer up */ BBR_STAT_INC(bbr_paced_segments); bbr->r_ctl.rc_hpts_flags |= PACE_PKT_OUTPUT; } /* * If no timers are going to run and we will fall off thfe hptsi * wheel, we resort to a keep-alive timer if its configured. */ if ((hpts_timeout == 0) && (slot == 0)) { if ((V_tcp_always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) && (tp->t_state <= TCPS_CLOSING)) { /* * Ok we have no timer (persists, rack, tlp, rxt or * del-ack), we don't have segments being paced. So * all that is left is the keepalive timer. */ if (TCPS_HAVEESTABLISHED(tp->t_state)) { hpts_timeout = TICKS_2_USEC(TP_KEEPIDLE(tp)); } else { hpts_timeout = TICKS_2_USEC(TP_KEEPINIT(tp)); } bbr->r_ctl.rc_hpts_flags |= PACE_TMR_KEEP; } } if (left && (stopped & (PACE_TMR_KEEP | PACE_TMR_DELACK)) == (bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK)) { /* * RACK, TLP, persists and RXT timers all are restartable * based on actions input .. i.e we received a packet (ack * or sack) and that changes things (rw, or snd_una etc). * Thus we can restart them with a new value. For * keep-alive, delayed_ack we keep track of what was left * and restart the timer with a smaller value. */ if (left < hpts_timeout) hpts_timeout = left; } if (bbr->r_ctl.rc_incr_tmrs && slot && (bbr->r_ctl.rc_hpts_flags & (PACE_TMR_TLP|PACE_TMR_RXT))) { /* * If configured to do so, and the timer is either * the TLP or RXT timer, we need to increase the timeout * by the pacing time. Consider the bottleneck at my * machine as an example, we are sending something * to start a TLP on. The last packet won't be emitted * fully until the pacing time (the bottleneck will hold * the data in place). Once the packet is emitted that * is when we want to start waiting for the TLP. This * is most evident with hardware pacing (where the nic * is holding the packet(s) before emitting). But it * can also show up in the network so we do it for all * cases. Technically we would take off one packet from * this extra delay but this is easier and being more * conservative is probably better. */ hpts_timeout += slot; } if (hpts_timeout) { /* * Hack alert for now we can't time-out over 2147 seconds (a * bit more than 35min) */ if (hpts_timeout > 0x7ffffffe) hpts_timeout = 0x7ffffffe; bbr->r_ctl.rc_timer_exp = cts + hpts_timeout; } else bbr->r_ctl.rc_timer_exp = 0; if ((slot) && (bbr->rc_use_google || bbr->output_error_seen || (slot <= hpts_timeout)) ) { /* * Tell LRO that it can queue packets while * we pace. */ bbr->rc_inp->inp_flags2 |= INP_MBUF_QUEUE_READY; if ((bbr->r_ctl.rc_hpts_flags & PACE_TMR_RACK) && (bbr->rc_cwnd_limited == 0)) { /* * If we are not cwnd limited and we * are running a rack timer we put on * the do not disturbe even for sack. */ inp->inp_flags2 |= INP_DONT_SACK_QUEUE; } else inp->inp_flags2 &= ~INP_DONT_SACK_QUEUE; bbr->rc_pacer_started = cts; (void)tcp_hpts_insert_diag(tp->t_inpcb, HPTS_USEC_TO_SLOTS(slot), __LINE__, &diag); bbr->rc_timer_first = 0; bbr->bbr_timer_src = frm; bbr_log_to_start(bbr, cts, hpts_timeout, slot, 1); bbr_log_hpts_diag(bbr, cts, &diag); } else if (hpts_timeout) { (void)tcp_hpts_insert_diag(tp->t_inpcb, HPTS_USEC_TO_SLOTS(hpts_timeout), __LINE__, &diag); /* * We add the flag here as well if the slot is set, * since hpts will call in to clear the queue first before * calling the output routine (which does our timers). * We don't want to set the flag if its just a timer * else the arrival of data might (that causes us * to send more) might get delayed. Imagine being * on a keep-alive timer and a request comes in for * more data. */ if (slot) bbr->rc_pacer_started = cts; if ((bbr->r_ctl.rc_hpts_flags & PACE_TMR_RACK) && (bbr->rc_cwnd_limited == 0)) { /* * For a rack timer, don't wake us even * if a sack arrives as long as we are * not cwnd limited. */ bbr->rc_inp->inp_flags2 |= INP_MBUF_QUEUE_READY; inp->inp_flags2 |= INP_DONT_SACK_QUEUE; } else { /* All other timers wake us up */ bbr->rc_inp->inp_flags2 &= ~INP_MBUF_QUEUE_READY; inp->inp_flags2 &= ~INP_DONT_SACK_QUEUE; } bbr->bbr_timer_src = frm; bbr_log_to_start(bbr, cts, hpts_timeout, slot, 0); bbr_log_hpts_diag(bbr, cts, &diag); bbr->rc_timer_first = 1; } bbr->rc_tmr_stopped = 0; bbr_log_type_bbrsnd(bbr, tot_len, slot, delay_calc, cts, frm, prev_delay); } static void bbr_timer_audit(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, struct sockbuf *sb) { /* * We received an ack, and then did not call send or were bounced * out due to the hpts was running. Now a timer is up as well, is it * the right timer? */ struct inpcb *inp; struct bbr_sendmap *rsm; uint32_t hpts_timeout; int tmr_up; tmr_up = bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK; if (bbr->rc_in_persist && (tmr_up == PACE_TMR_PERSIT)) return; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if (((rsm == NULL) || (tp->t_state < TCPS_ESTABLISHED)) && (tmr_up == PACE_TMR_RXT)) { /* Should be an RXT */ return; } inp = bbr->rc_inp; if (rsm == NULL) { /* Nothing outstanding? */ if (tp->t_flags & TF_DELACK) { if (tmr_up == PACE_TMR_DELACK) /* * We are supposed to have delayed ack up * and we do */ return; } else if (sbavail(&inp->inp_socket->so_snd) && (tmr_up == PACE_TMR_RXT)) { /* * if we hit enobufs then we would expect the * possiblity of nothing outstanding and the RXT up * (and the hptsi timer). */ return; } else if (((V_tcp_always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) && (tp->t_state <= TCPS_CLOSING)) && (tmr_up == PACE_TMR_KEEP) && (tp->snd_max == tp->snd_una)) { /* We should have keep alive up and we do */ return; } } if (rsm && (rsm->r_flags & BBR_SACK_PASSED)) { if ((tp->t_flags & TF_SENTFIN) && ((tp->snd_max - tp->snd_una) == 1) && (rsm->r_flags & BBR_HAS_FIN)) { /* needs to be a RXT */ if (tmr_up == PACE_TMR_RXT) return; else goto wrong_timer; } else if (tmr_up == PACE_TMR_RACK) return; else goto wrong_timer; } else if (rsm && (tmr_up == PACE_TMR_RACK)) { /* Rack timer has priority if we have data out */ return; } else if (SEQ_GT(tp->snd_max, tp->snd_una) && ((tmr_up == PACE_TMR_TLP) || (tmr_up == PACE_TMR_RXT))) { /* * Either a TLP or RXT is fine if no sack-passed is in place * and data is outstanding. */ return; } else if (tmr_up == PACE_TMR_DELACK) { /* * If the delayed ack was going to go off before the * rtx/tlp/rack timer were going to expire, then that would * be the timer in control. Note we don't check the time * here trusting the code is correct. */ return; } if (SEQ_GT(tp->snd_max, tp->snd_una) && ((tmr_up == PACE_TMR_RXT) || (tmr_up == PACE_TMR_TLP) || (tmr_up == PACE_TMR_RACK))) { /* * We have outstanding data and * we *do* have a RACK, TLP or RXT * timer running. We won't restart * anything here since thats probably ok we * will get called with some timer here shortly. */ return; } /* * Ok the timer originally started is not what we want now. We will * force the hpts to be stopped if any, and restart with the slot * set to what was in the saved slot. */ wrong_timer: if ((bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) == 0) { if (inp->inp_in_hpts) tcp_hpts_remove(inp, HPTS_REMOVE_OUTPUT); bbr_timer_cancel(bbr, __LINE__, cts); bbr_start_hpts_timer(bbr, tp, cts, 1, bbr->r_ctl.rc_last_delay_val, 0); } else { /* * Output is hptsi so we just need to switch the type of * timer. We don't bother with keep-alive, since when we * jump through the output, it will start the keep-alive if * nothing is sent. * * We only need a delayed-ack added and or the hpts_timeout. */ hpts_timeout = bbr_timer_start(tp, bbr, cts); if (tp->t_flags & TF_DELACK) { if (hpts_timeout == 0) { hpts_timeout = bbr_delack_time; bbr->r_ctl.rc_hpts_flags = PACE_TMR_DELACK; } else if (hpts_timeout > bbr_delack_time) { hpts_timeout = bbr_delack_time; bbr->r_ctl.rc_hpts_flags = PACE_TMR_DELACK; } } if (hpts_timeout) { if (hpts_timeout > 0x7ffffffe) hpts_timeout = 0x7ffffffe; bbr->r_ctl.rc_timer_exp = cts + hpts_timeout; } } } int32_t bbr_clear_lost = 0; /* * Considers the two time values now (cts) and earlier. * If cts is smaller than earlier, we could have * had a sequence wrap (our counter wraps every * 70 min or so) or it could be just clock skew * getting us two differnt time values. Clock skew * will show up within 10ms or so. So in such * a case (where cts is behind earlier time by * less than 10ms) we return 0. Otherwise we * return the true difference between them. */ static inline uint32_t bbr_calc_time(uint32_t cts, uint32_t earlier_time) { /* * Given two timestamps, the current time stamp cts, and some other * time-stamp taken in theory earlier return the difference. The * trick is here sometimes locking will get the other timestamp * after the cts. If this occurs we need to return 0. */ if (TSTMP_GEQ(cts, earlier_time)) return (cts - earlier_time); /* * cts is behind earlier_time if its less than 10ms consider it 0. * If its more than 10ms difference then we had a time wrap. Else * its just the normal locking foo. I wonder if we should not go to * 64bit TS and get rid of this issue. */ if (TSTMP_GEQ((cts + 10000), earlier_time)) return (0); /* * Ok the time must have wrapped. So we need to answer a large * amount of time, which the normal subtraction should do. */ return (cts - earlier_time); } static int sysctl_bbr_clear_lost(SYSCTL_HANDLER_ARGS) { uint32_t stat; int32_t error; error = SYSCTL_OUT(req, &bbr_clear_lost, sizeof(uint32_t)); if (error || req->newptr == NULL) return error; error = SYSCTL_IN(req, &stat, sizeof(uint32_t)); if (error) return (error); if (stat == 1) { #ifdef BBR_INVARIANTS printf("Clearing BBR lost counters\n"); #endif COUNTER_ARRAY_ZERO(bbr_state_lost, BBR_MAX_STAT); COUNTER_ARRAY_ZERO(bbr_state_time, BBR_MAX_STAT); COUNTER_ARRAY_ZERO(bbr_state_resend, BBR_MAX_STAT); } else if (stat == 2) { #ifdef BBR_INVARIANTS printf("Clearing BBR option counters\n"); #endif COUNTER_ARRAY_ZERO(bbr_opts_arry, BBR_OPTS_SIZE); } else if (stat == 3) { #ifdef BBR_INVARIANTS printf("Clearing BBR stats counters\n"); #endif COUNTER_ARRAY_ZERO(bbr_stat_arry, BBR_STAT_SIZE); } else if (stat == 4) { #ifdef BBR_INVARIANTS printf("Clearing BBR out-size counters\n"); #endif COUNTER_ARRAY_ZERO(bbr_out_size, TCP_MSS_ACCT_SIZE); } bbr_clear_lost = 0; return (0); } static void bbr_init_sysctls(void) { struct sysctl_oid *bbr_probertt; struct sysctl_oid *bbr_hptsi; struct sysctl_oid *bbr_measure; struct sysctl_oid *bbr_cwnd; struct sysctl_oid *bbr_timeout; struct sysctl_oid *bbr_states; struct sysctl_oid *bbr_startup; struct sysctl_oid *bbr_policer; /* Probe rtt controls */ bbr_probertt = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "probertt", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "gain", CTLFLAG_RW, &bbr_rttprobe_gain, 192, "What is the filter gain drop in probe_rtt (0=disable)?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "cwnd", CTLFLAG_RW, &bbr_rtt_probe_cwndtarg, 4, "How many mss's are outstanding during probe-rtt"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "int", CTLFLAG_RW, &bbr_rtt_probe_limit, 4000000, "If RTT has not shrank in this many micro-seconds enter probe-rtt"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "mintime", CTLFLAG_RW, &bbr_rtt_probe_time, 200000, "How many microseconds in probe-rtt"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "filter_len_sec", CTLFLAG_RW, &bbr_filter_len_sec, 6, "How long in seconds does the rttProp filter run?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "drain_rtt", CTLFLAG_RW, &bbr_drain_rtt, BBR_SRTT, "What is the drain rtt to use in probeRTT (rtt_prop=0, rtt_rack=1, rtt_pkt=2, rtt_srtt=3?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "can_force", CTLFLAG_RW, &bbr_can_force_probertt, 0, "If we keep setting new low rtt's but delay going in probe-rtt can we force in??"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "enter_sets_force", CTLFLAG_RW, &bbr_probertt_sets_rtt, 0, "In NF mode, do we imitate google_mode and set the rttProp on entry to probe-rtt?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "can_adjust", CTLFLAG_RW, &bbr_can_adjust_probertt, 1, "Can we dynamically adjust the probe-rtt limits and times?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "is_ratio", CTLFLAG_RW, &bbr_is_ratio, 0, "is the limit to filter a ratio?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "use_cwnd", CTLFLAG_RW, &bbr_prtt_slam_cwnd, 0, "Should we set/recover cwnd?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_probertt), OID_AUTO, "can_use_ts", CTLFLAG_RW, &bbr_can_use_ts_for_rtt, 1, "Can we use the ms timestamp if available for retransmistted rtt calculations?"); /* Pacing controls */ bbr_hptsi = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "pacing", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "hw_pacing", CTLFLAG_RW, &bbr_allow_hdwr_pacing, 1, "Do we allow hardware pacing?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "hw_pacing_limit", CTLFLAG_RW, &bbr_hardware_pacing_limit, 4000, "Do we have a limited number of connections for pacing chelsio (0=no limit)?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "hw_pacing_adj", CTLFLAG_RW, &bbr_hdwr_pace_adjust, 2, "Multiplier to calculated tso size?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "hw_pacing_floor", CTLFLAG_RW, &bbr_hdwr_pace_floor, 1, "Do we invoke the hardware pacing floor?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "hw_pacing_delay_cnt", CTLFLAG_RW, &bbr_hdwr_pacing_delay_cnt, 10, "How many packets must be sent after hdwr pacing is enabled"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "bw_cross", CTLFLAG_RW, &bbr_cross_over, 3000000, "What is the point where we cross over to linux like TSO size set"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "seg_deltarg", CTLFLAG_RW, &bbr_hptsi_segments_delay_tar, 7000, "What is the worse case delay target for hptsi < 48Mbp connections"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "enet_oh", CTLFLAG_RW, &bbr_include_enet_oh, 0, "Do we include the ethernet overhead in calculating pacing delay?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "ip_oh", CTLFLAG_RW, &bbr_include_ip_oh, 1, "Do we include the IP overhead in calculating pacing delay?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "tcp_oh", CTLFLAG_RW, &bbr_include_tcp_oh, 0, "Do we include the TCP overhead in calculating pacing delay?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "google_discount", CTLFLAG_RW, &bbr_google_discount, 10, "What is the default google discount percentage wise for pacing (11 = 1.1%%)?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "all_get_min", CTLFLAG_RW, &bbr_all_get_min, 0, "If you are less than a MSS do you just get the min?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "tso_min", CTLFLAG_RW, &bbr_hptsi_bytes_min, 1460, "For 0 -> 24Mbps what is floor number of segments for TSO"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "seg_tso_max", CTLFLAG_RW, &bbr_hptsi_segments_max, 6, "For 0 -> 24Mbps what is top number of segments for TSO"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "seg_floor", CTLFLAG_RW, &bbr_hptsi_segments_floor, 1, "Minimum TSO size we will fall too in segments"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "utter_max", CTLFLAG_RW, &bbr_hptsi_utter_max, 0, "The absolute maximum that any pacing (outside of hardware) can be"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "seg_divisor", CTLFLAG_RW, &bbr_hptsi_per_second, 100, "What is the divisor in our hptsi TSO calculation 512Mbps < X > 24Mbps "); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "srtt_mul", CTLFLAG_RW, &bbr_hptsi_max_mul, 1, "The multiplier for pace len max"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_hptsi), OID_AUTO, "srtt_div", CTLFLAG_RW, &bbr_hptsi_max_div, 2, "The divisor for pace len max"); /* Measurement controls */ bbr_measure = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "measure", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Measurement controls"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "min_i_bw", CTLFLAG_RW, &bbr_initial_bw_bps, 62500, "Minimum initial b/w in bytes per second"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "no_sack_needed", CTLFLAG_RW, &bbr_sack_not_required, 0, "Do we allow bbr to run on connections not supporting SACK?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "use_google", CTLFLAG_RW, &bbr_use_google_algo, 0, "Use has close to google V1.0 has possible?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "ts_limiting", CTLFLAG_RW, &bbr_ts_limiting, 1, "Do we attempt to use the peers timestamp to limit b/w caculations?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "ts_can_raise", CTLFLAG_RW, &bbr_ts_can_raise, 0, "Can we raise the b/w via timestamp b/w calculation?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "ts_delta", CTLFLAG_RW, &bbr_min_usec_delta, 20000, "How long in usec between ts of our sends in ts validation code?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "ts_peer_delta", CTLFLAG_RW, &bbr_min_peer_delta, 20, "What min numerical value should be between the peer deltas?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "ts_delta_percent", CTLFLAG_RW, &bbr_delta_percent, 150, "What percentage (150 = 15.0) do we allow variance for?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "min_measure_good_bw", CTLFLAG_RW, &bbr_min_measurements_req, 1, "What is the minimum measurment count we need before we switch to our b/w estimate"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "min_measure_before_pace", CTLFLAG_RW, &bbr_no_pacing_until, 4, "How many pkt-epoch's (0 is off) do we need before pacing is on?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "quanta", CTLFLAG_RW, &bbr_quanta, 2, "Extra quanta to add when calculating the target (ID section 4.2.3.2)."); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_measure), OID_AUTO, "noretran", CTLFLAG_RW, &bbr_no_retran, 0, "Should google mode not use retransmission measurements for the b/w estimation?"); /* State controls */ bbr_states = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "states", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "State controls"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "idle_restart", CTLFLAG_RW, &bbr_uses_idle_restart, 0, "Do we use a new special idle_restart state to ramp back up quickly?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "idle_restart_threshold", CTLFLAG_RW, &bbr_idle_restart_threshold, 100000, "How long must we be idle before we restart??"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "use_pkt_epoch", CTLFLAG_RW, &bbr_state_is_pkt_epoch, 0, "Do we use a pkt-epoch for substate if 0 rttProp?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "startup_rtt_gain", CTLFLAG_RW, &bbr_rtt_gain_thresh, 0, "What increase in RTT triggers us to stop ignoring no-loss and possibly exit startup?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "drain_floor", CTLFLAG_RW, &bbr_drain_floor, 88, "What is the lowest we can drain (pg) too?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "drain_2_target", CTLFLAG_RW, &bbr_state_drain_2_tar, 1, "Do we drain to target in drain substate?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "gain_2_target", CTLFLAG_RW, &bbr_gain_to_target, 1, "Does probe bw gain to target??"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "gain_extra_time", CTLFLAG_RW, &bbr_gain_gets_extra_too, 1, "Does probe bw gain get the extra time too?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "ld_div", CTLFLAG_RW, &bbr_drain_drop_div, 5, "Long drain drop divider?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "ld_mul", CTLFLAG_RW, &bbr_drain_drop_mul, 4, "Long drain drop multiplier?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "rand_ot_disc", CTLFLAG_RW, &bbr_rand_ot, 50, "Random discount of the ot?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "dr_filter_life", CTLFLAG_RW, &bbr_num_pktepo_for_del_limit, BBR_NUM_RTTS_FOR_DEL_LIMIT, "How many packet-epochs does the b/w delivery rate last?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "subdrain_applimited", CTLFLAG_RW, &bbr_sub_drain_app_limit, 0, "Does our sub-state drain invoke app limited if its long?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "use_cwnd_subdrain", CTLFLAG_RW, &bbr_sub_drain_slam_cwnd, 0, "Should we set/recover cwnd for sub-state drain?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "use_cwnd_maindrain", CTLFLAG_RW, &bbr_slam_cwnd_in_main_drain, 0, "Should we set/recover cwnd for main-state drain?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "google_gets_earlyout", CTLFLAG_RW, &google_allow_early_out, 1, "Should we allow google probe-bw/drain to exit early at flight target?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_states), OID_AUTO, "google_exit_loss", CTLFLAG_RW, &google_consider_lost, 1, "Should we have losses exit gain of probebw in google mode??"); /* Startup controls */ bbr_startup = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "startup", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Startup controls"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "cheat_iwnd", CTLFLAG_RW, &bbr_sends_full_iwnd, 1, "Do we not pace but burst out initial windows has our TSO size?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "loss_threshold", CTLFLAG_RW, &bbr_startup_loss_thresh, 2000, "In startup what is the loss threshold in a pe that will exit us from startup?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "use_lowerpg", CTLFLAG_RW, &bbr_use_lower_gain_in_startup, 1, "Should we use a lower hptsi gain if we see loss in startup?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "gain", CTLFLAG_RW, &bbr_start_exit, 25, "What gain percent do we need to see to stay in startup??"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "low_gain", CTLFLAG_RW, &bbr_low_start_exit, 15, "What gain percent do we need to see to stay in the lower gain startup??"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_startup), OID_AUTO, "loss_exit", CTLFLAG_RW, &bbr_exit_startup_at_loss, 1, "Should we exit startup at loss in an epoch if we are not gaining?"); /* CWND controls */ bbr_cwnd = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "cwnd", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Cwnd controls"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "tar_rtt", CTLFLAG_RW, &bbr_cwndtarget_rtt_touse, 0, "Target cwnd rtt measurment to use (0=rtt_prop, 1=rtt_rack, 2=pkt_rtt, 3=srtt)?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "may_shrink", CTLFLAG_RW, &bbr_cwnd_may_shrink, 0, "Can the cwnd shrink if it would grow to more than the target?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "max_target_limit", CTLFLAG_RW, &bbr_target_cwnd_mult_limit, 8, "Do we limit the cwnd to some multiple of the cwnd target if cwnd can't shrink 0=no?"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "highspeed_min", CTLFLAG_RW, &bbr_cwnd_min_val_hs, BBR_HIGHSPEED_NUM_MSS, "What is the high-speed min cwnd (rttProp under 1ms)"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "lowspeed_min", CTLFLAG_RW, &bbr_cwnd_min_val, BBR_PROBERTT_NUM_MSS, "What is the min cwnd (rttProp > 1ms)"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "initwin", CTLFLAG_RW, &bbr_def_init_win, 10, "What is the BBR initial window, if 0 use tcp version"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "do_loss_red", CTLFLAG_RW, &bbr_do_red, 600, "Do we reduce the b/w at exit from recovery based on ratio of prop/srtt (800=80.0, 0=off)?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "red_scale", CTLFLAG_RW, &bbr_red_scale, 20000, "What RTT do we scale with?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "red_growslow", CTLFLAG_RW, &bbr_red_growth_restrict, 1, "Do we restrict cwnd growth for whats in flight?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "red_div", CTLFLAG_RW, &bbr_red_div, 2, "If we reduce whats the divisor?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "red_mul", CTLFLAG_RW, &bbr_red_mul, 1, "If we reduce whats the mulitiplier?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "target_is_unit", CTLFLAG_RW, &bbr_target_is_bbunit, 0, "Is the state target the pacing_gain or BBR_UNIT?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_cwnd), OID_AUTO, "drop_limit", CTLFLAG_RW, &bbr_drop_limit, 0, "Number of segments limit for drop (0=use min_cwnd w/flight)?"); /* Timeout controls */ bbr_timeout = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "timeout", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Time out controls"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "delack", CTLFLAG_RW, &bbr_delack_time, 100000, "BBR's delayed ack time"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "tlp_uses", CTLFLAG_RW, &bbr_tlp_type_to_use, 3, "RTT that TLP uses in its calculations, 0=rttProp, 1=Rack_rtt, 2=pkt_rtt and 3=srtt"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "persmin", CTLFLAG_RW, &bbr_persist_min, 250000, "What is the minimum time in microseconds between persists"); SYSCTL_ADD_U32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "persmax", CTLFLAG_RW, &bbr_persist_max, 1000000, "What is the largest delay in microseconds between persists"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "tlp_minto", CTLFLAG_RW, &bbr_tlp_min, 10000, "TLP Min timeout in usecs"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "tlp_dack_time", CTLFLAG_RW, &bbr_delayed_ack_time, 200000, "TLP delayed ack compensation value"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "minrto", CTLFLAG_RW, &bbr_rto_min_ms, 30, "Minimum RTO in ms"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "maxrto", CTLFLAG_RW, &bbr_rto_max_sec, 4, "Maxiumum RTO in seconds -- should be at least as large as min_rto"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "tlp_retry", CTLFLAG_RW, &bbr_tlp_max_resend, 2, "How many times does TLP retry a single segment or multiple with no ACK"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "minto", CTLFLAG_RW, &bbr_min_to, 1000, "Minimum rack timeout in useconds"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "pktdelay", CTLFLAG_RW, &bbr_pkt_delay, 1000, "Extra RACK time (in useconds) besides reordering thresh"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "incr_tmrs", CTLFLAG_RW, &bbr_incr_timers, 1, "Increase the RXT/TLP timer by the pacing time used?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_timeout), OID_AUTO, "rxtmark_sackpassed", CTLFLAG_RW, &bbr_marks_rxt_sack_passed, 0, "Mark sack passed on all those not ack'd when a RXT hits?"); /* Policer controls */ bbr_policer = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "policer", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Policer controls"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "detect_enable", CTLFLAG_RW, &bbr_policer_detection_enabled, 1, "Is policer detection enabled??"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "min_pes", CTLFLAG_RW, &bbr_lt_intvl_min_rtts, 4, "Minimum number of PE's?"); SYSCTL_ADD_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "bwdiff", CTLFLAG_RW, &bbr_lt_bw_diff, (4000/8), "Minimal bw diff?"); SYSCTL_ADD_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "bwratio", CTLFLAG_RW, &bbr_lt_bw_ratio, 8, "Minimal bw diff?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "from_rack_rxt", CTLFLAG_RW, &bbr_policer_call_from_rack_to, 0, "Do we call the policer detection code from a rack-timeout?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "false_postive", CTLFLAG_RW, &bbr_lt_intvl_fp, 0, "What packet epoch do we do false-postive detection at (0=no)?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "loss_thresh", CTLFLAG_RW, &bbr_lt_loss_thresh, 196, "Loss threshold 196 = 19.6%?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_policer), OID_AUTO, "false_postive_thresh", CTLFLAG_RW, &bbr_lt_fd_thresh, 100, "What percentage is the false detection threshold (150=15.0)?"); /* All the rest */ SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "cheat_rxt", CTLFLAG_RW, &bbr_use_rack_resend_cheat, 0, "Do we burst 1ms between sends on retransmissions (like rack)?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "error_paceout", CTLFLAG_RW, &bbr_error_base_paceout, 10000, "When we hit an error what is the min to pace out in usec's?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "kill_paceout", CTLFLAG_RW, &bbr_max_net_error_cnt, 10, "When we hit this many errors in a row, kill the session?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "data_after_close", CTLFLAG_RW, &bbr_ignore_data_after_close, 1, "Do we hold off sending a RST until all pending data is ack'd"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "resend_use_tso", CTLFLAG_RW, &bbr_resends_use_tso, 0, "Can resends use TSO?"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "sblklimit", CTLFLAG_RW, &bbr_sack_block_limit, 128, "When do we start ignoring small sack blocks"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "bb_verbose", CTLFLAG_RW, &bbr_verbose_logging, 0, "Should BBR black box logging be verbose"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "reorder_thresh", CTLFLAG_RW, &bbr_reorder_thresh, 2, "What factor for rack will be added when seeing reordering (shift right)"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "reorder_fade", CTLFLAG_RW, &bbr_reorder_fade, 0, "Does reorder detection fade, if so how many ms (0 means never)"); SYSCTL_ADD_S32(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "rtt_tlp_thresh", CTLFLAG_RW, &bbr_tlp_thresh, 1, "what divisor for TLP rtt/retran will be added (1=rtt, 2=1/2 rtt etc)"); /* Stats and counters */ /* The pacing counters for hdwr/software can't be in the array */ bbr_nohdwr_pacing_enobuf = counter_u64_alloc(M_WAITOK); bbr_hdwr_pacing_enobuf = counter_u64_alloc(M_WAITOK); SYSCTL_ADD_COUNTER_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "enob_hdwr_pacing", CTLFLAG_RD, &bbr_hdwr_pacing_enobuf, "Total number of enobufs for hardware paced flows"); SYSCTL_ADD_COUNTER_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "enob_no_hdwr_pacing", CTLFLAG_RD, &bbr_nohdwr_pacing_enobuf, "Total number of enobufs for non-hardware paced flows"); bbr_flows_whdwr_pacing = counter_u64_alloc(M_WAITOK); SYSCTL_ADD_COUNTER_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "hdwr_pacing", CTLFLAG_RD, &bbr_flows_whdwr_pacing, "Total number of hardware paced flows"); bbr_flows_nohdwr_pacing = counter_u64_alloc(M_WAITOK); SYSCTL_ADD_COUNTER_U64(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "software_pacing", CTLFLAG_RD, &bbr_flows_nohdwr_pacing, "Total number of software paced flows"); COUNTER_ARRAY_ALLOC(bbr_stat_arry, BBR_STAT_SIZE, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "stats", CTLFLAG_RD, bbr_stat_arry, BBR_STAT_SIZE, "BBR Stats"); COUNTER_ARRAY_ALLOC(bbr_opts_arry, BBR_OPTS_SIZE, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "opts", CTLFLAG_RD, bbr_opts_arry, BBR_OPTS_SIZE, "BBR Option Stats"); COUNTER_ARRAY_ALLOC(bbr_state_lost, BBR_MAX_STAT, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "lost", CTLFLAG_RD, bbr_state_lost, BBR_MAX_STAT, "Stats of when losses occur"); COUNTER_ARRAY_ALLOC(bbr_state_resend, BBR_MAX_STAT, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "stateresend", CTLFLAG_RD, bbr_state_resend, BBR_MAX_STAT, "Stats of what states resend"); COUNTER_ARRAY_ALLOC(bbr_state_time, BBR_MAX_STAT, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "statetime", CTLFLAG_RD, bbr_state_time, BBR_MAX_STAT, "Stats of time spent in the states"); COUNTER_ARRAY_ALLOC(bbr_out_size, TCP_MSS_ACCT_SIZE, M_WAITOK); SYSCTL_ADD_COUNTER_U64_ARRAY(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "outsize", CTLFLAG_RD, bbr_out_size, TCP_MSS_ACCT_SIZE, "Size of output calls"); SYSCTL_ADD_PROC(&bbr_sysctl_ctx, SYSCTL_CHILDREN(bbr_sysctl_root), OID_AUTO, "clrlost", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, &bbr_clear_lost, 0, sysctl_bbr_clear_lost, "IU", "Clear lost counters"); } static inline int32_t bbr_progress_timeout_check(struct tcp_bbr *bbr) { if (bbr->rc_tp->t_maxunacktime && bbr->rc_tp->t_acktime && TSTMP_GT(ticks, bbr->rc_tp->t_acktime)) { if ((((uint32_t)ticks - bbr->rc_tp->t_acktime)) >= bbr->rc_tp->t_maxunacktime) { /* * There is an assumption here that the caller will * drop the connection, so we increment the * statistics. */ bbr_log_progress_event(bbr, bbr->rc_tp, ticks, PROGRESS_DROP, __LINE__); BBR_STAT_INC(bbr_progress_drops); #ifdef NETFLIX_STATS KMOD_TCPSTAT_INC(tcps_progdrops); #endif return (1); } } return (0); } static void bbr_counter_destroy(void) { COUNTER_ARRAY_FREE(bbr_stat_arry, BBR_STAT_SIZE); COUNTER_ARRAY_FREE(bbr_opts_arry, BBR_OPTS_SIZE); COUNTER_ARRAY_FREE(bbr_out_size, TCP_MSS_ACCT_SIZE); COUNTER_ARRAY_FREE(bbr_state_lost, BBR_MAX_STAT); COUNTER_ARRAY_FREE(bbr_state_time, BBR_MAX_STAT); COUNTER_ARRAY_FREE(bbr_state_resend, BBR_MAX_STAT); counter_u64_free(bbr_flows_whdwr_pacing); counter_u64_free(bbr_flows_nohdwr_pacing); } static __inline void bbr_fill_in_logging_data(struct tcp_bbr *bbr, struct tcp_log_bbr *l, uint32_t cts) { memset(l, 0, sizeof(union tcp_log_stackspecific)); l->cur_del_rate = bbr->r_ctl.rc_bbr_cur_del_rate; l->delRate = get_filter_value(&bbr->r_ctl.rc_delrate); l->rttProp = get_filter_value_small(&bbr->r_ctl.rc_rttprop); l->bw_inuse = bbr_get_bw(bbr); l->inflight = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); l->applimited = bbr->r_ctl.r_app_limited_until; l->delivered = bbr->r_ctl.rc_delivered; l->timeStamp = cts; l->lost = bbr->r_ctl.rc_lost; l->bbr_state = bbr->rc_bbr_state; l->bbr_substate = bbr_state_val(bbr); l->epoch = bbr->r_ctl.rc_rtt_epoch; l->lt_epoch = bbr->r_ctl.rc_lt_epoch; l->pacing_gain = bbr->r_ctl.rc_bbr_hptsi_gain; l->cwnd_gain = bbr->r_ctl.rc_bbr_cwnd_gain; l->inhpts = bbr->rc_inp->inp_in_hpts; l->ininput = bbr->rc_inp->inp_in_input; l->use_lt_bw = bbr->rc_lt_use_bw; l->pkts_out = bbr->r_ctl.rc_flight_at_input; l->pkt_epoch = bbr->r_ctl.rc_pkt_epoch; } static void bbr_log_type_bw_reduce(struct tcp_bbr *bbr, int reason) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = 0; log.u_bbr.flex2 = 0; log.u_bbr.flex5 = 0; log.u_bbr.flex3 = 0; log.u_bbr.flex4 = bbr->r_ctl.rc_pkt_epoch_loss_rate; log.u_bbr.flex7 = reason; log.u_bbr.flex6 = bbr->r_ctl.rc_bbr_enters_probertt; log.u_bbr.flex8 = 0; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BW_RED_EV, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_rwnd_collapse(struct tcp_bbr *bbr, int seq, int mode, uint32_t count) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = seq; log.u_bbr.flex2 = count; log.u_bbr.flex8 = mode; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_LOWGAIN, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_just_return(struct tcp_bbr *bbr, uint32_t cts, uint32_t tlen, uint8_t hpts_calling, uint8_t reason, uint32_t p_maxseg, int len) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = p_maxseg; log.u_bbr.flex2 = bbr->r_ctl.rc_hpts_flags; log.u_bbr.flex3 = bbr->r_ctl.rc_timer_exp; log.u_bbr.flex4 = reason; log.u_bbr.flex5 = bbr->rc_in_persist; log.u_bbr.flex6 = bbr->r_ctl.rc_last_delay_val; log.u_bbr.flex7 = p_maxseg; log.u_bbr.flex8 = bbr->rc_in_persist; log.u_bbr.pkts_out = 0; log.u_bbr.applimited = len; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_JUSTRET, 0, tlen, &log, false, &bbr->rc_tv); } } static void bbr_log_type_enter_rec(struct tcp_bbr *bbr, uint32_t seq) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = seq; log.u_bbr.flex2 = bbr->r_ctl.rc_cwnd_on_ent; log.u_bbr.flex3 = bbr->r_ctl.rc_recovery_start; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_ENTREC, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_msgsize_fail(struct tcp_bbr *bbr, struct tcpcb *tp, uint32_t len, uint32_t maxseg, uint32_t mtu, int32_t csum_flags, int32_t tso, uint32_t cts) { if (tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = tso; log.u_bbr.flex2 = maxseg; log.u_bbr.flex3 = mtu; log.u_bbr.flex4 = csum_flags; TCP_LOG_EVENTP(tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_MSGSIZE, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_flowend(struct tcp_bbr *bbr) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; struct sockbuf *r, *s; struct timeval tv; if (bbr->rc_inp->inp_socket) { r = &bbr->rc_inp->inp_socket->so_rcv; s = &bbr->rc_inp->inp_socket->so_snd; } else { r = s = NULL; } bbr_fill_in_logging_data(bbr, &log.u_bbr, tcp_get_usecs(&tv)); TCP_LOG_EVENTP(bbr->rc_tp, NULL, r, s, TCP_LOG_FLOWEND, 0, 0, &log, false, &tv); } } static void bbr_log_pkt_epoch(struct tcp_bbr *bbr, uint32_t cts, uint32_t line, uint32_t lost, uint32_t del) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = lost; log.u_bbr.flex2 = del; log.u_bbr.flex3 = bbr->r_ctl.rc_bbr_lastbtlbw; log.u_bbr.flex4 = bbr->r_ctl.rc_pkt_epoch_rtt; log.u_bbr.flex5 = bbr->r_ctl.rc_bbr_last_startup_epoch; log.u_bbr.flex6 = bbr->r_ctl.rc_lost_at_startup; log.u_bbr.flex7 = line; log.u_bbr.flex8 = 0; log.u_bbr.inflight = bbr->r_ctl.r_measurement_count; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_PKT_EPOCH, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_time_epoch(struct tcp_bbr *bbr, uint32_t cts, uint32_t line, uint32_t epoch_time) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->r_ctl.rc_lost; log.u_bbr.flex2 = bbr->rc_inp->inp_socket->so_snd.sb_lowat; log.u_bbr.flex3 = bbr->rc_inp->inp_socket->so_snd.sb_hiwat; log.u_bbr.flex7 = line; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TIME_EPOCH, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_set_of_state_target(struct tcp_bbr *bbr, uint32_t new_tar, int line, int meth) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex2 = new_tar; log.u_bbr.flex3 = line; log.u_bbr.flex4 = bbr->r_ctl.rc_pace_max_segs; log.u_bbr.flex5 = bbr_quanta; log.u_bbr.flex6 = bbr->r_ctl.rc_pace_min_segs; log.u_bbr.flex7 = bbr->rc_last_options; log.u_bbr.flex8 = meth; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_STATE_TARGET, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_statechange(struct tcp_bbr *bbr, uint32_t cts, int32_t line) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = line; log.u_bbr.flex2 = bbr->r_ctl.rc_rtt_shrinks; log.u_bbr.flex3 = bbr->r_ctl.rc_probertt_int; if (bbr_state_is_pkt_epoch) log.u_bbr.flex4 = bbr_get_rtt(bbr, BBR_RTT_PKTRTT); else log.u_bbr.flex4 = bbr_get_rtt(bbr, BBR_RTT_PROP); log.u_bbr.flex5 = bbr->r_ctl.rc_bbr_last_startup_epoch; log.u_bbr.flex6 = bbr->r_ctl.rc_lost_at_startup; log.u_bbr.flex7 = (bbr->r_ctl.rc_target_at_state/1000); log.u_bbr.lt_epoch = bbr->r_ctl.rc_level_state_extra; log.u_bbr.pkts_out = bbr->r_ctl.rc_target_at_state; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_STATE, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_rtt_shrinks(struct tcp_bbr *bbr, uint32_t cts, uint32_t applied, uint32_t rtt, uint32_t line, uint8_t reas, uint16_t cond) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = line; log.u_bbr.flex2 = bbr->r_ctl.rc_rtt_shrinks; log.u_bbr.flex3 = bbr->r_ctl.last_in_probertt; log.u_bbr.flex4 = applied; log.u_bbr.flex5 = rtt; log.u_bbr.flex6 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex7 = cond; log.u_bbr.flex8 = reas; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_RTT_SHRINKS, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_exit_rec(struct tcp_bbr *bbr) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = bbr->r_ctl.rc_recovery_start; log.u_bbr.flex2 = bbr->r_ctl.rc_cwnd_on_ent; log.u_bbr.flex5 = bbr->r_ctl.rc_target_at_state; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_EXITREC, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_cwndupd(struct tcp_bbr *bbr, uint32_t bytes_this_ack, uint32_t chg, uint32_t prev_acked, int32_t meth, uint32_t target, uint32_t th_ack, int32_t line) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = line; log.u_bbr.flex2 = prev_acked; log.u_bbr.flex3 = bytes_this_ack; log.u_bbr.flex4 = chg; log.u_bbr.flex5 = th_ack; log.u_bbr.flex6 = target; log.u_bbr.flex8 = meth; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_CWND, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_rtt_sample(struct tcp_bbr *bbr, uint32_t rtt, uint32_t tsin) { /* * Log the rtt sample we are applying to the srtt algorithm in * useconds. */ if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = rtt; log.u_bbr.flex2 = bbr->r_ctl.rc_bbr_state_time; log.u_bbr.flex3 = bbr->r_ctl.rc_ack_hdwr_delay; log.u_bbr.flex4 = bbr->rc_tp->ts_offset; log.u_bbr.flex5 = bbr->r_ctl.rc_target_at_state; log.u_bbr.pkts_out = tcp_tv_to_mssectick(&bbr->rc_tv); log.u_bbr.flex6 = tsin; log.u_bbr.flex7 = 0; log.u_bbr.flex8 = bbr->rc_ack_was_delayed; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, TCP_LOG_RTT, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_pesist(struct tcp_bbr *bbr, uint32_t cts, uint32_t time_in, int32_t line, uint8_t enter_exit) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = time_in; log.u_bbr.flex2 = line; log.u_bbr.flex8 = enter_exit; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_PERSIST, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_ack_clear(struct tcp_bbr *bbr, uint32_t cts) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->rc_tp->ts_recent_age; log.u_bbr.flex2 = bbr->r_ctl.rc_rtt_shrinks; log.u_bbr.flex3 = bbr->r_ctl.rc_probertt_int; log.u_bbr.flex4 = bbr->r_ctl.rc_went_idle_time; log.u_bbr.flex5 = bbr->r_ctl.rc_target_at_state; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_ACKCLEAR, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_ack_event(struct tcp_bbr *bbr, struct tcphdr *th, struct tcpopt *to, uint32_t tlen, uint16_t nsegs, uint32_t cts, int32_t nxt_pkt, struct mbuf *m) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; struct timeval tv; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = nsegs; log.u_bbr.flex2 = bbr->r_ctl.rc_lost_bytes; if (m) { struct timespec ts; log.u_bbr.flex3 = m->m_flags; if (m->m_flags & M_TSTMP) { mbuf_tstmp2timespec(m, &ts); tv.tv_sec = ts.tv_sec; tv.tv_usec = ts.tv_nsec / 1000; log.u_bbr.lt_epoch = tcp_tv_to_usectick(&tv); } else { log.u_bbr.lt_epoch = 0; } if (m->m_flags & M_TSTMP_LRO) { tv.tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000; tv.tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000; log.u_bbr.flex5 = tcp_tv_to_usectick(&tv); } else { /* No arrival timestamp */ log.u_bbr.flex5 = 0; } log.u_bbr.pkts_out = tcp_get_usecs(&tv); } else { log.u_bbr.flex3 = 0; log.u_bbr.flex5 = 0; log.u_bbr.flex6 = 0; log.u_bbr.pkts_out = 0; } log.u_bbr.flex4 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex7 = bbr->r_wanted_output; log.u_bbr.flex8 = bbr->rc_in_persist; TCP_LOG_EVENTP(bbr->rc_tp, th, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, TCP_LOG_IN, 0, tlen, &log, true, &bbr->rc_tv); } } static void bbr_log_doseg_done(struct tcp_bbr *bbr, uint32_t cts, int32_t nxt_pkt, int32_t did_out) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = did_out; log.u_bbr.flex2 = nxt_pkt; log.u_bbr.flex3 = bbr->r_ctl.rc_last_delay_val; log.u_bbr.flex4 = bbr->r_ctl.rc_hpts_flags; log.u_bbr.flex5 = bbr->r_ctl.rc_timer_exp; log.u_bbr.flex6 = bbr->r_ctl.rc_lost_bytes; log.u_bbr.flex7 = bbr->r_wanted_output; log.u_bbr.flex8 = bbr->rc_in_persist; log.u_bbr.pkts_out = bbr->r_ctl.highest_hdwr_delay; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_DOSEG_DONE, 0, 0, &log, true, &bbr->rc_tv); } } static void bbr_log_enobuf_jmp(struct tcp_bbr *bbr, uint32_t len, uint32_t cts, int32_t line, uint32_t o_len, uint32_t segcnt, uint32_t segsiz) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = line; log.u_bbr.flex2 = o_len; log.u_bbr.flex3 = segcnt; log.u_bbr.flex4 = segsiz; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_ENOBUF_JMP, ENOBUFS, len, &log, true, &bbr->rc_tv); } } static void bbr_log_to_processing(struct tcp_bbr *bbr, uint32_t cts, int32_t ret, int32_t timers, uint8_t hpts_calling) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = timers; log.u_bbr.flex2 = ret; log.u_bbr.flex3 = bbr->r_ctl.rc_timer_exp; log.u_bbr.flex4 = bbr->r_ctl.rc_hpts_flags; log.u_bbr.flex5 = cts; log.u_bbr.flex6 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex8 = hpts_calling; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TO_PROCESS, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_to_event(struct tcp_bbr *bbr, uint32_t cts, int32_t to_num) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; uint64_t ar; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->bbr_timer_src; log.u_bbr.flex2 = 0; log.u_bbr.flex3 = bbr->r_ctl.rc_hpts_flags; ar = (uint64_t)(bbr->r_ctl.rc_resend); ar >>= 32; ar &= 0x00000000ffffffff; log.u_bbr.flex4 = (uint32_t)ar; ar = (uint64_t)bbr->r_ctl.rc_resend; ar &= 0x00000000ffffffff; log.u_bbr.flex5 = (uint32_t)ar; log.u_bbr.flex6 = TICKS_2_USEC(bbr->rc_tp->t_rxtcur); log.u_bbr.flex8 = to_num; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_RTO, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_startup_event(struct tcp_bbr *bbr, uint32_t cts, uint32_t flex1, uint32_t flex2, uint32_t flex3, uint8_t reason) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = flex1; log.u_bbr.flex2 = flex2; log.u_bbr.flex3 = flex3; log.u_bbr.flex4 = 0; log.u_bbr.flex5 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex6 = bbr->r_ctl.rc_lost_at_startup; log.u_bbr.flex8 = reason; log.u_bbr.cur_del_rate = bbr->r_ctl.rc_bbr_lastbtlbw; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_REDUCE, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_hpts_diag(struct tcp_bbr *bbr, uint32_t cts, struct hpts_diag *diag) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = diag->p_nxt_slot; log.u_bbr.flex2 = diag->p_cur_slot; log.u_bbr.flex3 = diag->slot_req; log.u_bbr.flex4 = diag->inp_hptsslot; log.u_bbr.flex5 = diag->slot_remaining; log.u_bbr.flex6 = diag->need_new_to; log.u_bbr.flex7 = diag->p_hpts_active; log.u_bbr.flex8 = diag->p_on_min_sleep; /* Hijack other fields as needed */ log.u_bbr.epoch = diag->have_slept; log.u_bbr.lt_epoch = diag->yet_to_sleep; log.u_bbr.pkts_out = diag->co_ret; log.u_bbr.applimited = diag->hpts_sleep_time; log.u_bbr.delivered = diag->p_prev_slot; log.u_bbr.inflight = diag->p_runningtick; log.u_bbr.bw_inuse = diag->wheel_tick; log.u_bbr.rttProp = diag->wheel_cts; log.u_bbr.delRate = diag->maxticks; log.u_bbr.cur_del_rate = diag->p_curtick; log.u_bbr.cur_del_rate <<= 32; log.u_bbr.cur_del_rate |= diag->p_lasttick; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_HPTSDIAG, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_timer_var(struct tcp_bbr *bbr, int mode, uint32_t cts, uint32_t time_since_sent, uint32_t srtt, uint32_t thresh, uint32_t to) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->rc_tp->t_rttvar; log.u_bbr.flex2 = time_since_sent; log.u_bbr.flex3 = srtt; log.u_bbr.flex4 = thresh; log.u_bbr.flex5 = to; log.u_bbr.flex6 = bbr->rc_tp->t_srtt; log.u_bbr.flex8 = mode; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TIMERPREP, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_pacing_delay_calc(struct tcp_bbr *bbr, uint16_t gain, uint32_t len, uint32_t cts, uint32_t usecs, uint64_t bw, uint32_t override, int mod) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = usecs; log.u_bbr.flex2 = len; log.u_bbr.flex3 = (uint32_t)((bw >> 32) & 0x00000000ffffffff); log.u_bbr.flex4 = (uint32_t)(bw & 0x00000000ffffffff); if (override) log.u_bbr.flex5 = (1 << 2); else log.u_bbr.flex5 = 0; log.u_bbr.flex6 = override; log.u_bbr.flex7 = gain; log.u_bbr.flex8 = mod; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_HPTSI_CALC, 0, len, &log, false, &bbr->rc_tv); } } static void bbr_log_to_start(struct tcp_bbr *bbr, uint32_t cts, uint32_t to, int32_t slot, uint8_t which) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->bbr_timer_src; log.u_bbr.flex2 = to; log.u_bbr.flex3 = bbr->r_ctl.rc_hpts_flags; log.u_bbr.flex4 = slot; log.u_bbr.flex5 = bbr->rc_inp->inp_hptsslot; log.u_bbr.flex6 = TICKS_2_USEC(bbr->rc_tp->t_rxtcur); log.u_bbr.pkts_out = bbr->rc_inp->inp_flags2; log.u_bbr.flex8 = which; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TIMERSTAR, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_thresh_choice(struct tcp_bbr *bbr, uint32_t cts, uint32_t thresh, uint32_t lro, uint32_t srtt, struct bbr_sendmap *rsm, uint8_t frm) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = thresh; log.u_bbr.flex2 = lro; log.u_bbr.flex3 = bbr->r_ctl.rc_reorder_ts; log.u_bbr.flex4 = rsm->r_tim_lastsent[(rsm->r_rtr_cnt - 1)]; log.u_bbr.flex5 = TICKS_2_USEC(bbr->rc_tp->t_rxtcur); log.u_bbr.flex6 = srtt; log.u_bbr.flex7 = bbr->r_ctl.rc_reorder_shift; log.u_bbr.flex8 = frm; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_THRESH_CALC, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_to_cancel(struct tcp_bbr *bbr, int32_t line, uint32_t cts, uint8_t hpts_removed) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = line; log.u_bbr.flex2 = bbr->bbr_timer_src; log.u_bbr.flex3 = bbr->r_ctl.rc_hpts_flags; log.u_bbr.flex4 = bbr->rc_in_persist; log.u_bbr.flex5 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex6 = TICKS_2_USEC(bbr->rc_tp->t_rxtcur); log.u_bbr.flex8 = hpts_removed; log.u_bbr.pkts_out = bbr->rc_pacer_started; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TIMERCANC, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_tstmp_validation(struct tcp_bbr *bbr, uint64_t peer_delta, uint64_t delta) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = bbr->r_ctl.bbr_peer_tsratio; log.u_bbr.flex2 = (peer_delta >> 32); log.u_bbr.flex3 = (peer_delta & 0x00000000ffffffff); log.u_bbr.flex4 = (delta >> 32); log.u_bbr.flex5 = (delta & 0x00000000ffffffff); log.u_bbr.flex7 = bbr->rc_ts_clock_set; log.u_bbr.flex8 = bbr->rc_ts_cant_be_used; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_TSTMP_VAL, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_tsosize(struct tcp_bbr *bbr, uint32_t cts, uint32_t tsosz, uint32_t tls, uint32_t old_val, uint32_t maxseg, int hdwr) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = tsosz; log.u_bbr.flex2 = tls; log.u_bbr.flex3 = tcp_min_hptsi_time; log.u_bbr.flex4 = bbr->r_ctl.bbr_hptsi_bytes_min; log.u_bbr.flex5 = old_val; log.u_bbr.flex6 = maxseg; log.u_bbr.flex7 = bbr->rc_no_pacing; log.u_bbr.flex7 <<= 1; log.u_bbr.flex7 |= bbr->rc_past_init_win; if (hdwr) log.u_bbr.flex8 = 0x80 | bbr->rc_use_google; else log.u_bbr.flex8 = bbr->rc_use_google; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BBRTSO, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_rsmclear(struct tcp_bbr *bbr, uint32_t cts, struct bbr_sendmap *rsm, uint32_t flags, uint32_t line) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = line; log.u_bbr.flex2 = rsm->r_start; log.u_bbr.flex3 = rsm->r_end; log.u_bbr.flex4 = rsm->r_delivered; log.u_bbr.flex5 = rsm->r_rtr_cnt; log.u_bbr.flex6 = rsm->r_dupack; log.u_bbr.flex7 = rsm->r_tim_lastsent[0]; log.u_bbr.flex8 = rsm->r_flags; /* Hijack the pkts_out fids */ log.u_bbr.applimited = flags; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_RSM_CLEARED, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_bbrupd(struct tcp_bbr *bbr, uint8_t flex8, uint32_t cts, uint32_t flex3, uint32_t flex2, uint32_t flex5, uint32_t flex6, uint32_t pkts_out, int flex7, uint32_t flex4, uint32_t flex1) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = flex1; log.u_bbr.flex2 = flex2; log.u_bbr.flex3 = flex3; log.u_bbr.flex4 = flex4; log.u_bbr.flex5 = flex5; log.u_bbr.flex6 = flex6; log.u_bbr.flex7 = flex7; /* Hijack the pkts_out fids */ log.u_bbr.pkts_out = pkts_out; log.u_bbr.flex8 = flex8; if (bbr->rc_ack_was_delayed) log.u_bbr.epoch = bbr->r_ctl.rc_ack_hdwr_delay; else log.u_bbr.epoch = 0; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BBRUPD, 0, flex2, &log, false, &bbr->rc_tv); } } static void bbr_log_type_ltbw(struct tcp_bbr *bbr, uint32_t cts, int32_t reason, uint32_t newbw, uint32_t obw, uint32_t diff, uint32_t tim) { if (/*bbr_verbose_logging && */(bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = reason; log.u_bbr.flex2 = newbw; log.u_bbr.flex3 = obw; log.u_bbr.flex4 = diff; log.u_bbr.flex5 = bbr->r_ctl.rc_lt_lost; log.u_bbr.flex6 = bbr->r_ctl.rc_lt_del; log.u_bbr.flex7 = bbr->rc_lt_is_sampling; log.u_bbr.pkts_out = tim; log.u_bbr.bw_inuse = bbr->r_ctl.rc_lt_bw; if (bbr->rc_lt_use_bw == 0) log.u_bbr.epoch = bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_lt_epoch; else log.u_bbr.epoch = bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_lt_epoch_use; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BWSAMP, 0, 0, &log, false, &bbr->rc_tv); } } static inline void bbr_log_progress_event(struct tcp_bbr *bbr, struct tcpcb *tp, uint32_t tick, int event, int line) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); log.u_bbr.flex1 = line; log.u_bbr.flex2 = tick; log.u_bbr.flex3 = tp->t_maxunacktime; log.u_bbr.flex4 = tp->t_acktime; log.u_bbr.flex8 = event; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_PROGRESS, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_type_log_hdwr_pacing(struct tcp_bbr *bbr, const struct ifnet *ifp, uint64_t rate, uint64_t hw_rate, int line, uint32_t cts, int error) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = ((hw_rate >> 32) & 0x00000000ffffffff); log.u_bbr.flex2 = (hw_rate & 0x00000000ffffffff); log.u_bbr.flex3 = (((uint64_t)ifp >> 32) & 0x00000000ffffffff); log.u_bbr.flex4 = ((uint64_t)ifp & 0x00000000ffffffff); log.u_bbr.bw_inuse = rate; log.u_bbr.flex5 = line; log.u_bbr.flex6 = error; log.u_bbr.flex8 = bbr->skip_gain; log.u_bbr.flex8 <<= 1; log.u_bbr.flex8 |= bbr->gain_is_limited; log.u_bbr.flex8 <<= 1; log.u_bbr.flex8 |= bbr->bbr_hdrw_pacing; log.u_bbr.pkts_out = bbr->rc_tp->t_maxseg; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_HDWR_PACE, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_type_bbrsnd(struct tcp_bbr *bbr, uint32_t len, uint32_t slot, uint32_t del_by, uint32_t cts, uint32_t line, uint32_t prev_delay) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = slot; log.u_bbr.flex2 = del_by; log.u_bbr.flex3 = prev_delay; log.u_bbr.flex4 = line; log.u_bbr.flex5 = bbr->r_ctl.rc_last_delay_val; log.u_bbr.flex6 = bbr->r_ctl.rc_hptsi_agg_delay; log.u_bbr.flex7 = (0x0000ffff & bbr->r_ctl.rc_hpts_flags); log.u_bbr.flex8 = bbr->rc_in_persist; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BBRSND, 0, len, &log, false, &bbr->rc_tv); } } static void bbr_log_type_bbrrttprop(struct tcp_bbr *bbr, uint32_t t, uint32_t end, uint32_t tsconv, uint32_t cts, int32_t match, uint32_t seq, uint8_t flags) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->r_ctl.rc_delivered; log.u_bbr.flex2 = 0; log.u_bbr.flex3 = bbr->r_ctl.rc_lowest_rtt; log.u_bbr.flex4 = end; log.u_bbr.flex5 = seq; log.u_bbr.flex6 = t; log.u_bbr.flex7 = match; log.u_bbr.flex8 = flags; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_BBRRTT, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_exit_gain(struct tcp_bbr *bbr, uint32_t cts, int32_t entry_method) { if (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->r_ctl.rc_target_at_state; log.u_bbr.flex2 = (bbr->rc_tp->t_maxseg - bbr->rc_last_options); log.u_bbr.flex3 = bbr->r_ctl.gain_epoch; log.u_bbr.flex4 = bbr->r_ctl.rc_pace_max_segs; log.u_bbr.flex5 = bbr->r_ctl.rc_pace_min_segs; log.u_bbr.flex6 = bbr->r_ctl.rc_bbr_state_atflight; log.u_bbr.flex7 = 0; log.u_bbr.flex8 = entry_method; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_EXIT_GAIN, 0, 0, &log, false, &bbr->rc_tv); } } static void bbr_log_settings_change(struct tcp_bbr *bbr, int settings_desired) { if (bbr_verbose_logging && (bbr->rc_tp->t_logstate != TCP_LOG_STATE_OFF)) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, bbr->r_ctl.rc_rcvtime); /* R-HU */ log.u_bbr.flex1 = 0; log.u_bbr.flex2 = 0; log.u_bbr.flex3 = 0; log.u_bbr.flex4 = 0; log.u_bbr.flex7 = 0; log.u_bbr.flex8 = settings_desired; TCP_LOG_EVENTP(bbr->rc_tp, NULL, &bbr->rc_inp->inp_socket->so_rcv, &bbr->rc_inp->inp_socket->so_snd, BBR_LOG_SETTINGS_CHG, 0, 0, &log, false, &bbr->rc_tv); } } /* * Returns the bw from the our filter. */ static inline uint64_t bbr_get_full_bw(struct tcp_bbr *bbr) { uint64_t bw; bw = get_filter_value(&bbr->r_ctl.rc_delrate); return (bw); } static inline void bbr_set_pktepoch(struct tcp_bbr *bbr, uint32_t cts, int32_t line) { uint64_t calclr; uint32_t lost, del; if (bbr->r_ctl.rc_lost > bbr->r_ctl.rc_lost_at_pktepoch) lost = bbr->r_ctl.rc_lost - bbr->r_ctl.rc_lost_at_pktepoch; else lost = 0; del = bbr->r_ctl.rc_delivered - bbr->r_ctl.rc_pkt_epoch_del; if (lost == 0) { calclr = 0; } else if (del) { calclr = lost; calclr *= (uint64_t)1000; calclr /= (uint64_t)del; } else { /* Nothing delivered? 100.0% loss */ calclr = 1000; } bbr->r_ctl.rc_pkt_epoch_loss_rate = (uint32_t)calclr; if (IN_RECOVERY(bbr->rc_tp->t_flags)) bbr->r_ctl.recovery_lr += (uint32_t)calclr; bbr->r_ctl.rc_pkt_epoch++; if (bbr->rc_no_pacing && (bbr->r_ctl.rc_pkt_epoch >= bbr->no_pacing_until)) { bbr->rc_no_pacing = 0; tcp_bbr_tso_size_check(bbr, cts); } bbr->r_ctl.rc_pkt_epoch_rtt = bbr_calc_time(cts, bbr->r_ctl.rc_pkt_epoch_time); bbr->r_ctl.rc_pkt_epoch_time = cts; /* What was our loss rate */ bbr_log_pkt_epoch(bbr, cts, line, lost, del); bbr->r_ctl.rc_pkt_epoch_del = bbr->r_ctl.rc_delivered; bbr->r_ctl.rc_lost_at_pktepoch = bbr->r_ctl.rc_lost; } static inline void bbr_set_epoch(struct tcp_bbr *bbr, uint32_t cts, int32_t line) { uint32_t epoch_time; /* Tick the RTT clock */ bbr->r_ctl.rc_rtt_epoch++; epoch_time = cts - bbr->r_ctl.rc_rcv_epoch_start; bbr_log_time_epoch(bbr, cts, line, epoch_time); bbr->r_ctl.rc_rcv_epoch_start = cts; } static inline void bbr_isit_a_pkt_epoch(struct tcp_bbr *bbr, uint32_t cts, struct bbr_sendmap *rsm, int32_t line, int32_t cum_acked) { if (SEQ_GEQ(rsm->r_delivered, bbr->r_ctl.rc_pkt_epoch_del)) { bbr->rc_is_pkt_epoch_now = 1; } } /* * Returns the bw from either the b/w filter * or from the lt_bw (if the connection is being * policed). */ static inline uint64_t __bbr_get_bw(struct tcp_bbr *bbr) { uint64_t bw, min_bw; uint64_t rtt; int gm_measure_cnt = 1; /* * For startup we make, like google, a * minimum b/w. This is generated from the * IW and the rttProp. We do fall back to srtt * if for some reason (initial handshake) we don't * have a rttProp. We, in the worst case, fall back * to the configured min_bw (rc_initial_hptsi_bw). */ if (bbr->rc_bbr_state == BBR_STATE_STARTUP) { /* Attempt first to use rttProp */ rtt = (uint64_t)get_filter_value_small(&bbr->r_ctl.rc_rttprop); if (rtt && (rtt < 0xffffffff)) { measure: min_bw = (uint64_t)(bbr_initial_cwnd(bbr, bbr->rc_tp)) * ((uint64_t)1000000); min_bw /= rtt; if (min_bw < bbr->r_ctl.rc_initial_hptsi_bw) { min_bw = bbr->r_ctl.rc_initial_hptsi_bw; } } else if (bbr->rc_tp->t_srtt != 0) { /* No rttProp, use srtt? */ rtt = bbr_get_rtt(bbr, BBR_SRTT); goto measure; } else { min_bw = bbr->r_ctl.rc_initial_hptsi_bw; } } else min_bw = 0; if ((bbr->rc_past_init_win == 0) && (bbr->r_ctl.rc_delivered > bbr_initial_cwnd(bbr, bbr->rc_tp))) bbr->rc_past_init_win = 1; if ((bbr->rc_use_google) && (bbr->r_ctl.r_measurement_count >= 1)) gm_measure_cnt = 0; if (gm_measure_cnt && ((bbr->r_ctl.r_measurement_count < bbr_min_measurements_req) || (bbr->rc_past_init_win == 0))) { /* For google we use our guess rate until we get 1 measurement */ use_initial_window: rtt = (uint64_t)get_filter_value_small(&bbr->r_ctl.rc_rttprop); if (rtt && (rtt < 0xffffffff)) { /* * We have an RTT measurment. Use that in * combination with our initial window to calculate * a b/w. */ bw = (uint64_t)(bbr_initial_cwnd(bbr, bbr->rc_tp)) * ((uint64_t)1000000); bw /= rtt; if (bw < bbr->r_ctl.rc_initial_hptsi_bw) { bw = bbr->r_ctl.rc_initial_hptsi_bw; } } else { /* Drop back to the 40 and punt to a default */ bw = bbr->r_ctl.rc_initial_hptsi_bw; } if (bw < 1) /* Probably should panic */ bw = 1; if (bw > min_bw) return (bw); else return (min_bw); } if (bbr->rc_lt_use_bw) bw = bbr->r_ctl.rc_lt_bw; else if (bbr->r_recovery_bw && (bbr->rc_use_google == 0)) bw = bbr->r_ctl.red_bw; else bw = get_filter_value(&bbr->r_ctl.rc_delrate); if (bbr->rc_tp->t_peakrate_thr && (bbr->rc_use_google == 0)) { /* * Enforce user set rate limit, keep in mind that * t_peakrate_thr is in B/s already */ bw = uqmin((uint64_t)bbr->rc_tp->t_peakrate_thr, bw); } if (bw == 0) { /* We should not be at 0, go to the initial window then */ goto use_initial_window; } if (bw < 1) /* Probably should panic */ bw = 1; if (bw < min_bw) bw = min_bw; return (bw); } static inline uint64_t bbr_get_bw(struct tcp_bbr *bbr) { uint64_t bw; bw = __bbr_get_bw(bbr); return (bw); } static inline void bbr_reset_lt_bw_interval(struct tcp_bbr *bbr, uint32_t cts) { bbr->r_ctl.rc_lt_epoch = bbr->r_ctl.rc_pkt_epoch; bbr->r_ctl.rc_lt_time = bbr->r_ctl.rc_del_time; bbr->r_ctl.rc_lt_del = bbr->r_ctl.rc_delivered; bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } static inline void bbr_reset_lt_bw_sampling(struct tcp_bbr *bbr, uint32_t cts) { bbr->rc_lt_is_sampling = 0; bbr->rc_lt_use_bw = 0; bbr->r_ctl.rc_lt_bw = 0; bbr_reset_lt_bw_interval(bbr, cts); } static inline void bbr_lt_bw_samp_done(struct tcp_bbr *bbr, uint64_t bw, uint32_t cts, uint32_t timin) { uint64_t diff; /* Do we have a previous sample? */ if (bbr->r_ctl.rc_lt_bw) { /* Get the diff in bytes per second */ if (bbr->r_ctl.rc_lt_bw > bw) diff = bbr->r_ctl.rc_lt_bw - bw; else diff = bw - bbr->r_ctl.rc_lt_bw; if ((diff <= bbr_lt_bw_diff) || (diff <= (bbr->r_ctl.rc_lt_bw / bbr_lt_bw_ratio))) { /* Consider us policed */ uint32_t saved_bw; saved_bw = (uint32_t)bbr->r_ctl.rc_lt_bw; bbr->r_ctl.rc_lt_bw = (bw + bbr->r_ctl.rc_lt_bw) / 2; /* average of two */ bbr->rc_lt_use_bw = 1; bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; /* * Use pkt based epoch for measuring length of * policer up */ bbr->r_ctl.rc_lt_epoch_use = bbr->r_ctl.rc_pkt_epoch; /* * reason 4 is we need to start consider being * policed */ bbr_log_type_ltbw(bbr, cts, 4, (uint32_t)bw, saved_bw, (uint32_t)diff, timin); return; } } bbr->r_ctl.rc_lt_bw = bw; bbr_reset_lt_bw_interval(bbr, cts); bbr_log_type_ltbw(bbr, cts, 5, 0, (uint32_t)bw, 0, timin); } /* * RRS: Copied from user space! * Calculate a uniformly distributed random number less than upper_bound * avoiding "modulo bias". * * Uniformity is achieved by generating new random numbers until the one * returned is outside the range [0, 2**32 % upper_bound). This * guarantees the selected random number will be inside * [2**32 % upper_bound, 2**32) which maps back to [0, upper_bound) * after reduction modulo upper_bound. */ static uint32_t arc4random_uniform(uint32_t upper_bound) { uint32_t r, min; if (upper_bound < 2) return 0; /* 2**32 % x == (2**32 - x) % x */ min = -upper_bound % upper_bound; /* * This could theoretically loop forever but each retry has * p > 0.5 (worst case, usually far better) of selecting a * number inside the range we need, so it should rarely need * to re-roll. */ for (;;) { r = arc4random(); if (r >= min) break; } return r % upper_bound; } static void bbr_randomize_extra_state_time(struct tcp_bbr *bbr) { uint32_t ran, deduct; ran = arc4random_uniform(bbr_rand_ot); if (ran) { deduct = bbr->r_ctl.rc_level_state_extra / ran; bbr->r_ctl.rc_level_state_extra -= deduct; } } /* * Return randomly the starting state * to use in probebw. */ static uint8_t bbr_pick_probebw_substate(struct tcp_bbr *bbr, uint32_t cts) { uint32_t ran; uint8_t ret_val; /* Initialize the offset to 0 */ bbr->r_ctl.rc_exta_time_gd = 0; bbr->rc_hit_state_1 = 0; bbr->r_ctl.rc_level_state_extra = 0; ran = arc4random_uniform((BBR_SUBSTATE_COUNT-1)); /* * The math works funny here :) the return value is used to set the * substate and then the state change is called which increments by * one. So if we return 1 (DRAIN) we will increment to 2 (LEVEL1) when * we fully enter the state. Note that the (8 - 1 - ran) assures that * we return 1 - 7, so we dont return 0 and end up starting in * state 1 (DRAIN). */ ret_val = BBR_SUBSTATE_COUNT - 1 - ran; /* Set an epoch */ if ((cts - bbr->r_ctl.rc_rcv_epoch_start) >= bbr_get_rtt(bbr, BBR_RTT_PROP)) bbr_set_epoch(bbr, cts, __LINE__); bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; return (ret_val); } static void bbr_lt_bw_sampling(struct tcp_bbr *bbr, uint32_t cts, int32_t loss_detected) { uint32_t diff, d_time; uint64_t del_time, bw, lost, delivered; if (bbr->r_use_policer == 0) return; if (bbr->rc_lt_use_bw) { /* We are using lt bw do we stop yet? */ diff = bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_lt_epoch_use; if (diff > bbr_lt_bw_max_rtts) { /* Reset it all */ reset_all: bbr_reset_lt_bw_sampling(bbr, cts); if (bbr->rc_filled_pipe) { bbr_set_epoch(bbr, cts, __LINE__); bbr->rc_bbr_substate = bbr_pick_probebw_substate(bbr, cts); bbr_substate_change(bbr, cts, __LINE__, 0); bbr->rc_bbr_state = BBR_STATE_PROBE_BW; bbr_log_type_statechange(bbr, cts, __LINE__); } else { /* * This should not happen really * unless we remove the startup/drain * restrictions above. */ bbr->rc_bbr_state = BBR_STATE_STARTUP; bbr_set_epoch(bbr, cts, __LINE__); bbr->r_ctl.rc_bbr_state_time = cts; bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.rc_startup_pg; bbr->r_ctl.rc_bbr_cwnd_gain = bbr->r_ctl.rc_startup_pg; bbr_set_state_target(bbr, __LINE__); bbr_log_type_statechange(bbr, cts, __LINE__); } /* reason 0 is to stop using lt-bw */ bbr_log_type_ltbw(bbr, cts, 0, 0, 0, 0, 0); return; } if (bbr_lt_intvl_fp == 0) { /* Not doing false-postive detection */ return; } /* False positive detection */ if (diff == bbr_lt_intvl_fp) { /* At bbr_lt_intvl_fp we record the lost */ bbr->r_ctl.rc_lt_del = bbr->r_ctl.rc_delivered; bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } else if (diff > (bbr_lt_intvl_min_rtts + bbr_lt_intvl_fp)) { /* Now is our loss rate still high? */ lost = bbr->r_ctl.rc_lost - bbr->r_ctl.rc_lt_lost; delivered = bbr->r_ctl.rc_delivered - bbr->r_ctl.rc_lt_del; if ((delivered == 0) || (((lost * 1000)/delivered) < bbr_lt_fd_thresh)) { /* No still below our threshold */ bbr_log_type_ltbw(bbr, cts, 7, lost, delivered, 0, 0); } else { /* Yikes its still high, it must be a false positive */ bbr_log_type_ltbw(bbr, cts, 8, lost, delivered, 0, 0); goto reset_all; } } return; } /* * Wait for the first loss before sampling, to let the policer * exhaust its tokens and estimate the steady-state rate allowed by * the policer. Starting samples earlier includes bursts that * over-estimate the bw. */ if (bbr->rc_lt_is_sampling == 0) { /* reason 1 is to begin doing the sampling */ if (loss_detected == 0) return; bbr_reset_lt_bw_interval(bbr, cts); bbr->rc_lt_is_sampling = 1; bbr_log_type_ltbw(bbr, cts, 1, 0, 0, 0, 0); return; } /* Now how long were we delivering long term last> */ if (TSTMP_GEQ(bbr->r_ctl.rc_del_time, bbr->r_ctl.rc_lt_time)) d_time = bbr->r_ctl.rc_del_time - bbr->r_ctl.rc_lt_time; else d_time = 0; /* To avoid underestimates, reset sampling if we run out of data. */ if (bbr->r_ctl.r_app_limited_until) { /* Can not measure in app-limited state */ bbr_reset_lt_bw_sampling(bbr, cts); /* reason 2 is to reset sampling due to app limits */ bbr_log_type_ltbw(bbr, cts, 2, 0, 0, 0, d_time); return; } diff = bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_lt_epoch; if (diff < bbr_lt_intvl_min_rtts) { /* * need more samples (we don't * start on a round like linux so * we need 1 more). */ /* 6 is not_enough time or no-loss */ bbr_log_type_ltbw(bbr, cts, 6, 0, 0, 0, d_time); return; } if (diff > (4 * bbr_lt_intvl_min_rtts)) { /* * For now if we wait too long, reset all sampling. We need * to do some research here, its possible that we should * base this on how much loss as occurred.. something like * if its under 10% (or some thresh) reset all otherwise * don't. Thats for phase II I guess. */ bbr_reset_lt_bw_sampling(bbr, cts); /* reason 3 is to reset sampling due too long of sampling */ bbr_log_type_ltbw(bbr, cts, 3, 0, 0, 0, d_time); return; } /* * End sampling interval when a packet is lost, so we estimate the * policer tokens were exhausted. Stopping the sampling before the * tokens are exhausted under-estimates the policed rate. */ if (loss_detected == 0) { /* 6 is not_enough time or no-loss */ bbr_log_type_ltbw(bbr, cts, 6, 0, 0, 0, d_time); return; } /* Calculate packets lost and delivered in sampling interval. */ lost = bbr->r_ctl.rc_lost - bbr->r_ctl.rc_lt_lost; delivered = bbr->r_ctl.rc_delivered - bbr->r_ctl.rc_lt_del; if ((delivered == 0) || (((lost * 1000)/delivered) < bbr_lt_loss_thresh)) { bbr_log_type_ltbw(bbr, cts, 6, lost, delivered, 0, d_time); return; } if (d_time < 1000) { /* Not enough time. wait */ /* 6 is not_enough time or no-loss */ bbr_log_type_ltbw(bbr, cts, 6, 0, 0, 0, d_time); return; } if (d_time >= (0xffffffff / USECS_IN_MSEC)) { /* Too long */ bbr_reset_lt_bw_sampling(bbr, cts); /* reason 3 is to reset sampling due too long of sampling */ bbr_log_type_ltbw(bbr, cts, 3, 0, 0, 0, d_time); return; } del_time = d_time; bw = delivered; bw *= (uint64_t)USECS_IN_SECOND; bw /= del_time; bbr_lt_bw_samp_done(bbr, bw, cts, d_time); } /* * Allocate a sendmap from our zone. */ static struct bbr_sendmap * bbr_alloc(struct tcp_bbr *bbr) { struct bbr_sendmap *rsm; BBR_STAT_INC(bbr_to_alloc); rsm = uma_zalloc(bbr_zone, (M_NOWAIT | M_ZERO)); if (rsm) { bbr->r_ctl.rc_num_maps_alloced++; return (rsm); } if (bbr->r_ctl.rc_free_cnt) { BBR_STAT_INC(bbr_to_alloc_emerg); rsm = TAILQ_FIRST(&bbr->r_ctl.rc_free); TAILQ_REMOVE(&bbr->r_ctl.rc_free, rsm, r_next); bbr->r_ctl.rc_free_cnt--; return (rsm); } BBR_STAT_INC(bbr_to_alloc_failed); return (NULL); } static struct bbr_sendmap * bbr_alloc_full_limit(struct tcp_bbr *bbr) { if ((V_tcp_map_entries_limit > 0) && (bbr->r_ctl.rc_num_maps_alloced >= V_tcp_map_entries_limit)) { BBR_STAT_INC(bbr_alloc_limited); if (!bbr->alloc_limit_reported) { bbr->alloc_limit_reported = 1; BBR_STAT_INC(bbr_alloc_limited_conns); } return (NULL); } return (bbr_alloc(bbr)); } /* wrapper to allocate a sendmap entry, subject to a specific limit */ static struct bbr_sendmap * bbr_alloc_limit(struct tcp_bbr *bbr, uint8_t limit_type) { struct bbr_sendmap *rsm; if (limit_type) { /* currently there is only one limit type */ if (V_tcp_map_split_limit > 0 && bbr->r_ctl.rc_num_split_allocs >= V_tcp_map_split_limit) { BBR_STAT_INC(bbr_split_limited); if (!bbr->alloc_limit_reported) { bbr->alloc_limit_reported = 1; BBR_STAT_INC(bbr_alloc_limited_conns); } return (NULL); } } /* allocate and mark in the limit type, if set */ rsm = bbr_alloc(bbr); if (rsm != NULL && limit_type) { rsm->r_limit_type = limit_type; bbr->r_ctl.rc_num_split_allocs++; } return (rsm); } static void bbr_free(struct tcp_bbr *bbr, struct bbr_sendmap *rsm) { if (rsm->r_limit_type) { /* currently there is only one limit type */ bbr->r_ctl.rc_num_split_allocs--; } if (rsm->r_is_smallmap) bbr->r_ctl.rc_num_small_maps_alloced--; if (bbr->r_ctl.rc_tlp_send == rsm) bbr->r_ctl.rc_tlp_send = NULL; if (bbr->r_ctl.rc_resend == rsm) { bbr->r_ctl.rc_resend = NULL; } if (bbr->r_ctl.rc_next == rsm) bbr->r_ctl.rc_next = NULL; if (bbr->r_ctl.rc_sacklast == rsm) bbr->r_ctl.rc_sacklast = NULL; if (bbr->r_ctl.rc_free_cnt < bbr_min_req_free) { memset(rsm, 0, sizeof(struct bbr_sendmap)); TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_free, rsm, r_next); rsm->r_limit_type = 0; bbr->r_ctl.rc_free_cnt++; return; } bbr->r_ctl.rc_num_maps_alloced--; uma_zfree(bbr_zone, rsm); } /* * Returns the BDP. */ static uint64_t bbr_get_bw_delay_prod(uint64_t rtt, uint64_t bw) { /* * Calculate the bytes in flight needed given the bw (in bytes per * second) and the specifyed rtt in useconds. We need to put out the * returned value per RTT to match that rate. Gain will normaly * raise it up from there. * * This should not overflow as long as the bandwidth is below 1 * TByte per second (bw < 10**12 = 2**40) and the rtt is smaller * than 1000 seconds (rtt < 10**3 * 10**6 = 10**9 = 2**30). */ uint64_t usec_per_sec; usec_per_sec = USECS_IN_SECOND; return ((rtt * bw) / usec_per_sec); } /* * Return the initial cwnd. */ static uint32_t bbr_initial_cwnd(struct tcp_bbr *bbr, struct tcpcb *tp) { uint32_t i_cwnd; if (bbr->rc_init_win) { i_cwnd = bbr->rc_init_win * tp->t_maxseg; } else if (V_tcp_initcwnd_segments) i_cwnd = min((V_tcp_initcwnd_segments * tp->t_maxseg), max(2 * tp->t_maxseg, 14600)); else if (V_tcp_do_rfc3390) i_cwnd = min(4 * tp->t_maxseg, max(2 * tp->t_maxseg, 4380)); else { /* Per RFC5681 Section 3.1 */ if (tp->t_maxseg > 2190) i_cwnd = 2 * tp->t_maxseg; else if (tp->t_maxseg > 1095) i_cwnd = 3 * tp->t_maxseg; else i_cwnd = 4 * tp->t_maxseg; } return (i_cwnd); } /* * Given a specified gain, return the target * cwnd based on that gain. */ static uint32_t bbr_get_raw_target_cwnd(struct tcp_bbr *bbr, uint32_t gain, uint64_t bw) { uint64_t bdp, rtt; uint32_t cwnd; if ((get_filter_value_small(&bbr->r_ctl.rc_rttprop) == 0xffffffff) || (bbr_get_full_bw(bbr) == 0)) { /* No measurements yet */ return (bbr_initial_cwnd(bbr, bbr->rc_tp)); } /* * Get bytes per RTT needed (rttProp is normally in * bbr_cwndtarget_rtt_touse) */ rtt = bbr_get_rtt(bbr, bbr_cwndtarget_rtt_touse); /* Get the bdp from the two values */ bdp = bbr_get_bw_delay_prod(rtt, bw); /* Now apply the gain */ cwnd = (uint32_t)(((bdp * ((uint64_t)gain)) + (uint64_t)(BBR_UNIT - 1)) / ((uint64_t)BBR_UNIT)); return (cwnd); } static uint32_t bbr_get_target_cwnd(struct tcp_bbr *bbr, uint64_t bw, uint32_t gain) { uint32_t cwnd, mss; mss = min((bbr->rc_tp->t_maxseg - bbr->rc_last_options), bbr->r_ctl.rc_pace_max_segs); /* Get the base cwnd with gain rounded to a mss */ cwnd = roundup(bbr_get_raw_target_cwnd(bbr, bw, gain), mss); /* * Add in N (2 default since we do not have a * fq layer to trap packets in) quanta's per the I-D * section 4.2.3.2 quanta adjust. */ cwnd += (bbr_quanta * bbr->r_ctl.rc_pace_max_segs); if (bbr->rc_use_google) { if((bbr->rc_bbr_state == BBR_STATE_PROBE_BW) && (bbr_state_val(bbr) == BBR_SUB_GAIN)) { /* * The linux implementation adds * an extra 2 x mss in gain cycle which * is documented no-where except in the code. * so we add more for Neal undocumented feature */ cwnd += 2 * mss; } if ((cwnd / mss) & 0x1) { /* Round up for odd num mss */ cwnd += mss; } } /* Are we below the min cwnd? */ if (cwnd < get_min_cwnd(bbr)) return (get_min_cwnd(bbr)); return (cwnd); } static uint16_t bbr_gain_adjust(struct tcp_bbr *bbr, uint16_t gain) { if (gain < 1) gain = 1; return (gain); } static uint32_t bbr_get_header_oh(struct tcp_bbr *bbr) { int seg_oh; seg_oh = 0; if (bbr->r_ctl.rc_inc_tcp_oh) { /* Do we include TCP overhead? */ seg_oh = (bbr->rc_last_options + sizeof(struct tcphdr)); } if (bbr->r_ctl.rc_inc_ip_oh) { /* Do we include IP overhead? */ #ifdef INET6 if (bbr->r_is_v6) seg_oh += sizeof(struct ip6_hdr); else #endif #ifdef INET seg_oh += sizeof(struct ip); #endif } if (bbr->r_ctl.rc_inc_enet_oh) { /* Do we include the ethernet overhead? */ seg_oh += sizeof(struct ether_header); } return(seg_oh); } static uint32_t bbr_get_pacing_length(struct tcp_bbr *bbr, uint16_t gain, uint32_t useconds_time, uint64_t bw) { uint64_t divor, res, tim; if (useconds_time == 0) return (0); gain = bbr_gain_adjust(bbr, gain); divor = (uint64_t)USECS_IN_SECOND * (uint64_t)BBR_UNIT; tim = useconds_time; res = (tim * bw * gain) / divor; if (res == 0) res = 1; return ((uint32_t)res); } /* * Given a gain and a length return the delay in useconds that * should be used to evenly space out packets * on the connection (based on the gain factor). */ static uint32_t bbr_get_pacing_delay(struct tcp_bbr *bbr, uint16_t gain, int32_t len, uint32_t cts, int nolog) { uint64_t bw, lentim, res; uint32_t usecs, srtt, over = 0; uint32_t seg_oh, num_segs, maxseg; if (len == 0) return (0); maxseg = bbr->rc_tp->t_maxseg - bbr->rc_last_options; num_segs = (len + maxseg - 1) / maxseg; if (bbr->rc_use_google == 0) { seg_oh = bbr_get_header_oh(bbr); len += (num_segs * seg_oh); } gain = bbr_gain_adjust(bbr, gain); bw = bbr_get_bw(bbr); if (bbr->rc_use_google) { uint64_t cbw; /* * Reduce the b/w by the google discount * factor 10 = 1%. */ cbw = bw * (uint64_t)(1000 - bbr->r_ctl.bbr_google_discount); cbw /= (uint64_t)1000; /* We don't apply a discount if it results in 0 */ if (cbw > 0) bw = cbw; } lentim = ((uint64_t)len * (uint64_t)USECS_IN_SECOND * (uint64_t)BBR_UNIT); res = lentim / ((uint64_t)gain * bw); if (res == 0) res = 1; usecs = (uint32_t)res; srtt = bbr_get_rtt(bbr, BBR_SRTT); if (bbr_hptsi_max_mul && bbr_hptsi_max_div && (bbr->rc_use_google == 0) && (usecs > ((srtt * bbr_hptsi_max_mul) / bbr_hptsi_max_div))) { /* * We cannot let the delay be more than 1/2 the srtt time. * Otherwise we cannot pace out or send properly. */ over = usecs = (srtt * bbr_hptsi_max_mul) / bbr_hptsi_max_div; BBR_STAT_INC(bbr_hpts_min_time); } if (!nolog) bbr_log_pacing_delay_calc(bbr, gain, len, cts, usecs, bw, over, 1); return (usecs); } static void bbr_ack_received(struct tcpcb *tp, struct tcp_bbr *bbr, struct tcphdr *th, uint32_t bytes_this_ack, uint32_t sack_changed, uint32_t prev_acked, int32_t line, uint32_t losses) { INP_WLOCK_ASSERT(tp->t_inpcb); uint64_t bw; uint32_t cwnd, target_cwnd, saved_bytes, maxseg; int32_t meth; #ifdef STATS if ((tp->t_flags & TF_GPUTINPROG) && SEQ_GEQ(th->th_ack, tp->gput_ack)) { /* * Strech acks and compressed acks will cause this to * oscillate but we are doing it the same way as the main * stack so it will be compariable (though possibly not * ideal). */ int32_t cgput; int64_t gput, time_stamp; gput = (int64_t) (th->th_ack - tp->gput_seq) * 8; time_stamp = max(1, ((bbr->r_ctl.rc_rcvtime - tp->gput_ts) / 1000)); cgput = gput / time_stamp; stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, cgput); if (tp->t_stats_gput_prev > 0) stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_GPUT_ND, ((gput - tp->t_stats_gput_prev) * 100) / tp->t_stats_gput_prev); tp->t_flags &= ~TF_GPUTINPROG; tp->t_stats_gput_prev = cgput; } #endif if ((bbr->rc_bbr_state == BBR_STATE_PROBE_RTT) && ((bbr->r_ctl.bbr_rttprobe_gain_val == 0) || bbr->rc_use_google)) { /* We don't change anything in probe-rtt */ return; } maxseg = tp->t_maxseg - bbr->rc_last_options; saved_bytes = bytes_this_ack; bytes_this_ack += sack_changed; if (bytes_this_ack > prev_acked) { bytes_this_ack -= prev_acked; /* * A byte ack'd gives us a full mss * to be like linux i.e. they count packets. */ if ((bytes_this_ack < maxseg) && bbr->rc_use_google) bytes_this_ack = maxseg; } else { /* Unlikely */ bytes_this_ack = 0; } cwnd = tp->snd_cwnd; bw = get_filter_value(&bbr->r_ctl.rc_delrate); if (bw) target_cwnd = bbr_get_target_cwnd(bbr, bw, (uint32_t)bbr->r_ctl.rc_bbr_cwnd_gain); else target_cwnd = bbr_initial_cwnd(bbr, bbr->rc_tp); if (IN_RECOVERY(tp->t_flags) && (bbr->bbr_prev_in_rec == 0)) { /* * We are entering recovery and * thus packet conservation. */ bbr->pkt_conservation = 1; bbr->r_ctl.rc_recovery_start = bbr->r_ctl.rc_rcvtime; cwnd = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + bytes_this_ack; } if (IN_RECOVERY(tp->t_flags)) { uint32_t flight; bbr->bbr_prev_in_rec = 1; if (cwnd > losses) { cwnd -= losses; if (cwnd < maxseg) cwnd = maxseg; } else cwnd = maxseg; flight = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); bbr_log_type_cwndupd(bbr, flight, 0, losses, 10, 0, 0, line); if (bbr->pkt_conservation) { uint32_t time_in; if (TSTMP_GEQ(bbr->r_ctl.rc_rcvtime, bbr->r_ctl.rc_recovery_start)) time_in = bbr->r_ctl.rc_rcvtime - bbr->r_ctl.rc_recovery_start; else time_in = 0; if (time_in >= bbr_get_rtt(bbr, BBR_RTT_PROP)) { /* Clear packet conservation after an rttProp */ bbr->pkt_conservation = 0; } else { if ((flight + bytes_this_ack) > cwnd) cwnd = flight + bytes_this_ack; if (cwnd < get_min_cwnd(bbr)) cwnd = get_min_cwnd(bbr); tp->snd_cwnd = cwnd; bbr_log_type_cwndupd(bbr, saved_bytes, sack_changed, prev_acked, 1, target_cwnd, th->th_ack, line); return; } } } else bbr->bbr_prev_in_rec = 0; if ((bbr->rc_use_google == 0) && bbr->r_ctl.restrict_growth) { bbr->r_ctl.restrict_growth--; if (bytes_this_ack > maxseg) bytes_this_ack = maxseg; } if (bbr->rc_filled_pipe) { /* * Here we have exited startup and filled the pipe. We will * thus allow the cwnd to shrink to the target. We hit here * mostly. */ uint32_t s_cwnd; meth = 2; s_cwnd = min((cwnd + bytes_this_ack), target_cwnd); if (s_cwnd > cwnd) cwnd = s_cwnd; else if (bbr_cwnd_may_shrink || bbr->rc_use_google || bbr->rc_no_pacing) cwnd = s_cwnd; } else { /* * Here we are still in startup, we increase cwnd by what * has been acked. */ if ((cwnd < target_cwnd) || (bbr->rc_past_init_win == 0)) { meth = 3; cwnd += bytes_this_ack; } else { /* * Method 4 means we are at target so no gain in * startup and past the initial window. */ meth = 4; } } tp->snd_cwnd = max(cwnd, get_min_cwnd(bbr)); bbr_log_type_cwndupd(bbr, saved_bytes, sack_changed, prev_acked, meth, target_cwnd, th->th_ack, line); } static void tcp_bbr_partialack(struct tcpcb *tp) { struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) <= tp->snd_cwnd) { bbr->r_wanted_output = 1; } } static void bbr_post_recovery(struct tcpcb *tp) { struct tcp_bbr *bbr; uint32_t flight; INP_WLOCK_ASSERT(tp->t_inpcb); bbr = (struct tcp_bbr *)tp->t_fb_ptr; /* * Here we just exit recovery. */ EXIT_RECOVERY(tp->t_flags); /* Lock in our b/w reduction for the specified number of pkt-epochs */ bbr->r_recovery_bw = 0; tp->snd_recover = tp->snd_una; tcp_bbr_tso_size_check(bbr, bbr->r_ctl.rc_rcvtime); bbr->pkt_conservation = 0; if (bbr->rc_use_google == 0) { /* * For non-google mode lets * go ahead and make sure we clear * the recovery state so if we * bounce back in to recovery we * will do PC. */ bbr->bbr_prev_in_rec = 0; } bbr_log_type_exit_rec(bbr); if (bbr->rc_bbr_state != BBR_STATE_PROBE_RTT) { tp->snd_cwnd = max(tp->snd_cwnd, bbr->r_ctl.rc_cwnd_on_ent); bbr_log_type_cwndupd(bbr, 0, 0, 0, 15, 0, 0, __LINE__); } else { /* For probe-rtt case lets fix up its saved_cwnd */ if (bbr->r_ctl.rc_saved_cwnd < bbr->r_ctl.rc_cwnd_on_ent) { bbr->r_ctl.rc_saved_cwnd = bbr->r_ctl.rc_cwnd_on_ent; bbr_log_type_cwndupd(bbr, 0, 0, 0, 16, 0, 0, __LINE__); } } flight = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if ((bbr->rc_use_google == 0) && bbr_do_red) { uint64_t val, lr2use; uint32_t maxseg, newcwnd, acks_inflight, ratio, cwnd; uint32_t *cwnd_p; if (bbr_get_rtt(bbr, BBR_SRTT)) { val = ((uint64_t)bbr_get_rtt(bbr, BBR_RTT_PROP) * (uint64_t)1000); val /= bbr_get_rtt(bbr, BBR_SRTT); ratio = (uint32_t)val; } else ratio = 1000; bbr_log_type_cwndupd(bbr, bbr_red_mul, bbr_red_div, bbr->r_ctl.recovery_lr, 21, ratio, bbr->r_ctl.rc_red_cwnd_pe, __LINE__); if ((ratio < bbr_do_red) || (bbr_do_red == 0)) goto done; if (((bbr->rc_bbr_state == BBR_STATE_PROBE_RTT) && bbr_prtt_slam_cwnd) || (bbr_sub_drain_slam_cwnd && (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) && bbr->rc_hit_state_1 && (bbr_state_val(bbr) == BBR_SUB_DRAIN)) || ((bbr->rc_bbr_state == BBR_STATE_DRAIN) && bbr_slam_cwnd_in_main_drain)) { /* * Here we must poke at the saved cwnd * as well as the cwnd. */ cwnd = bbr->r_ctl.rc_saved_cwnd; cwnd_p = &bbr->r_ctl.rc_saved_cwnd; } else { cwnd = tp->snd_cwnd; cwnd_p = &tp->snd_cwnd; } maxseg = tp->t_maxseg - bbr->rc_last_options; /* Add the overall lr with the recovery lr */ if (bbr->r_ctl.rc_lost == 0) lr2use = 0; else if (bbr->r_ctl.rc_delivered == 0) lr2use = 1000; else { lr2use = bbr->r_ctl.rc_lost * 1000; lr2use /= bbr->r_ctl.rc_delivered; } lr2use += bbr->r_ctl.recovery_lr; acks_inflight = (flight / (maxseg * 2)); if (bbr_red_scale) { lr2use *= bbr_get_rtt(bbr, BBR_SRTT); lr2use /= bbr_red_scale; if ((bbr_red_growth_restrict) && ((bbr_get_rtt(bbr, BBR_SRTT)/bbr_red_scale) > 1)) bbr->r_ctl.restrict_growth += acks_inflight; } if (lr2use) { val = (uint64_t)cwnd * lr2use; val /= 1000; if (cwnd > val) newcwnd = roundup((cwnd - val), maxseg); else newcwnd = maxseg; } else { val = (uint64_t)cwnd * (uint64_t)bbr_red_mul; val /= (uint64_t)bbr_red_div; newcwnd = roundup((uint32_t)val, maxseg); } /* with standard delayed acks how many acks can I expect? */ if (bbr_drop_limit == 0) { /* * Anticpate how much we will * raise the cwnd based on the acks. */ if ((newcwnd + (acks_inflight * maxseg)) < get_min_cwnd(bbr)) { /* We do enforce the min (with the acks) */ newcwnd = (get_min_cwnd(bbr) - acks_inflight); } } else { /* * A strict drop limit of N is is inplace */ if (newcwnd < (bbr_drop_limit * maxseg)) { newcwnd = bbr_drop_limit * maxseg; } } /* For the next N acks do we restrict the growth */ *cwnd_p = newcwnd; if (tp->snd_cwnd > newcwnd) tp->snd_cwnd = newcwnd; bbr_log_type_cwndupd(bbr, bbr_red_mul, bbr_red_div, val, 22, (uint32_t)lr2use, bbr_get_rtt(bbr, BBR_SRTT), __LINE__); bbr->r_ctl.rc_red_cwnd_pe = bbr->r_ctl.rc_pkt_epoch; } done: bbr->r_ctl.recovery_lr = 0; if (flight <= tp->snd_cwnd) { bbr->r_wanted_output = 1; } tcp_bbr_tso_size_check(bbr, bbr->r_ctl.rc_rcvtime); } static void bbr_setup_red_bw(struct tcp_bbr *bbr, uint32_t cts) { bbr->r_ctl.red_bw = get_filter_value(&bbr->r_ctl.rc_delrate); /* Limit the drop in b/w to 1/2 our current filter. */ if (bbr->r_ctl.red_bw > bbr->r_ctl.rc_bbr_cur_del_rate) bbr->r_ctl.red_bw = bbr->r_ctl.rc_bbr_cur_del_rate; if (bbr->r_ctl.red_bw < (get_filter_value(&bbr->r_ctl.rc_delrate) / 2)) bbr->r_ctl.red_bw = get_filter_value(&bbr->r_ctl.rc_delrate) / 2; tcp_bbr_tso_size_check(bbr, cts); } static void bbr_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type, struct bbr_sendmap *rsm) { struct tcp_bbr *bbr; INP_WLOCK_ASSERT(tp->t_inpcb); bbr = (struct tcp_bbr *)tp->t_fb_ptr; switch (type) { case CC_NDUPACK: if (!IN_RECOVERY(tp->t_flags)) { tp->snd_recover = tp->snd_max; /* Start a new epoch */ bbr_set_pktepoch(bbr, bbr->r_ctl.rc_rcvtime, __LINE__); if (bbr->rc_lt_is_sampling || bbr->rc_lt_use_bw) { /* * Move forward the lt epoch * so it won't count the truncated * epoch. */ bbr->r_ctl.rc_lt_epoch++; } if (bbr->rc_bbr_state == BBR_STATE_STARTUP) { /* * Just like the policer detection code * if we are in startup we must push * forward the last startup epoch * to hide the truncated PE. */ bbr->r_ctl.rc_bbr_last_startup_epoch++; } bbr->r_ctl.rc_cwnd_on_ent = tp->snd_cwnd; ENTER_RECOVERY(tp->t_flags); bbr->rc_tlp_rtx_out = 0; bbr->r_ctl.recovery_lr = bbr->r_ctl.rc_pkt_epoch_loss_rate; tcp_bbr_tso_size_check(bbr, bbr->r_ctl.rc_rcvtime); if (bbr->rc_inp->inp_in_hpts && ((bbr->r_ctl.rc_hpts_flags & PACE_TMR_RACK) == 0)) { /* * When we enter recovery, we need to restart * any timers. This may mean we gain an agg * early, which will be made up for at the last * rxt out. */ bbr->rc_timer_first = 1; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); } /* * Calculate a new cwnd based on to the current * delivery rate with no gain. We get the bdp * without gaining it up like we normally would and * we use the last cur_del_rate. */ if ((bbr->rc_use_google == 0) && (bbr->r_ctl.bbr_rttprobe_gain_val || (bbr->rc_bbr_state != BBR_STATE_PROBE_RTT))) { tp->snd_cwnd = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + (tp->t_maxseg - bbr->rc_last_options); if (tp->snd_cwnd < get_min_cwnd(bbr)) { /* We always gate to min cwnd */ tp->snd_cwnd = get_min_cwnd(bbr); } bbr_log_type_cwndupd(bbr, 0, 0, 0, 14, 0, 0, __LINE__); } bbr_log_type_enter_rec(bbr, rsm->r_start); } break; case CC_RTO_ERR: KMOD_TCPSTAT_INC(tcps_sndrexmitbad); /* RTO was unnecessary, so reset everything. */ bbr_reset_lt_bw_sampling(bbr, bbr->r_ctl.rc_rcvtime); if (bbr->rc_bbr_state != BBR_STATE_PROBE_RTT) { tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; tp->snd_recover = tp->snd_recover_prev; tp->snd_cwnd = max(tp->snd_cwnd, bbr->r_ctl.rc_cwnd_on_ent); bbr_log_type_cwndupd(bbr, 0, 0, 0, 13, 0, 0, __LINE__); } tp->t_badrxtwin = 0; break; } } /* * Indicate whether this ack should be delayed. We can delay the ack if * following conditions are met: * - There is no delayed ack timer in progress. * - Our last ack wasn't a 0-sized window. We never want to delay * the ack that opens up a 0-sized window. * - LRO wasn't used for this segment. We make sure by checking that the * segment size is not larger than the MSS. * - Delayed acks are enabled or this is a half-synchronized T/TCP * connection. * - The data being acked is less than a full segment (a stretch ack * of more than a segment we should ack. * - nsegs is 1 (if its more than that we received more than 1 ack). */ #define DELAY_ACK(tp, bbr, nsegs) \ (((tp->t_flags & TF_RXWIN0SENT) == 0) && \ ((bbr->bbr_segs_rcvd + nsegs) < tp->t_delayed_ack) && \ (tp->t_delayed_ack || (tp->t_flags & TF_NEEDSYN))) /* * Return the lowest RSM in the map of * packets still in flight that is not acked. * This should normally find on the first one * since we remove packets from the send * map after they are marked ACKED. */ static struct bbr_sendmap * bbr_find_lowest_rsm(struct tcp_bbr *bbr) { struct bbr_sendmap *rsm; /* * Walk the time-order transmitted list looking for an rsm that is * not acked. This will be the one that was sent the longest time * ago that is still outstanding. */ TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_tmap, r_tnext) { if (rsm->r_flags & BBR_ACKED) { continue; } goto finish; } finish: return (rsm); } static struct bbr_sendmap * bbr_find_high_nonack(struct tcp_bbr *bbr, struct bbr_sendmap *rsm) { struct bbr_sendmap *prsm; /* * Walk the sequence order list backward until we hit and arrive at * the highest seq not acked. In theory when this is called it * should be the last segment (which it was not). */ prsm = rsm; TAILQ_FOREACH_REVERSE_FROM(prsm, &bbr->r_ctl.rc_map, bbr_head, r_next) { if (prsm->r_flags & (BBR_ACKED | BBR_HAS_FIN)) { continue; } return (prsm); } return (NULL); } /* * Returns to the caller the number of microseconds that * the packet can be outstanding before we think we * should have had an ack returned. */ static uint32_t bbr_calc_thresh_rack(struct tcp_bbr *bbr, uint32_t srtt, uint32_t cts, struct bbr_sendmap *rsm) { /* * lro is the flag we use to determine if we have seen reordering. * If it gets set we have seen reordering. The reorder logic either * works in one of two ways: * * If reorder-fade is configured, then we track the last time we saw * re-ordering occur. If we reach the point where enough time as * passed we no longer consider reordering has occuring. * * Or if reorder-face is 0, then once we see reordering we consider * the connection to alway be subject to reordering and just set lro * to 1. * * In the end if lro is non-zero we add the extra time for * reordering in. */ int32_t lro; uint32_t thresh, t_rxtcur; if (srtt == 0) srtt = 1; if (bbr->r_ctl.rc_reorder_ts) { if (bbr->r_ctl.rc_reorder_fade) { if (SEQ_GEQ(cts, bbr->r_ctl.rc_reorder_ts)) { lro = cts - bbr->r_ctl.rc_reorder_ts; if (lro == 0) { /* * No time as passed since the last * reorder, mark it as reordering. */ lro = 1; } } else { /* Negative time? */ lro = 0; } if (lro > bbr->r_ctl.rc_reorder_fade) { /* Turn off reordering seen too */ bbr->r_ctl.rc_reorder_ts = 0; lro = 0; } } else { /* Reodering does not fade */ lro = 1; } } else { lro = 0; } thresh = srtt + bbr->r_ctl.rc_pkt_delay; if (lro) { /* It must be set, if not you get 1/4 rtt */ if (bbr->r_ctl.rc_reorder_shift) thresh += (srtt >> bbr->r_ctl.rc_reorder_shift); else thresh += (srtt >> 2); } else { thresh += 1000; } /* We don't let the rack timeout be above a RTO */ if ((bbr->rc_tp)->t_srtt == 0) t_rxtcur = BBR_INITIAL_RTO; else t_rxtcur = TICKS_2_USEC(bbr->rc_tp->t_rxtcur); if (thresh > t_rxtcur) { thresh = t_rxtcur; } /* And we don't want it above the RTO max either */ if (thresh > (((uint32_t)bbr->rc_max_rto_sec) * USECS_IN_SECOND)) { thresh = (((uint32_t)bbr->rc_max_rto_sec) * USECS_IN_SECOND); } bbr_log_thresh_choice(bbr, cts, thresh, lro, srtt, rsm, BBR_TO_FRM_RACK); return (thresh); } /* * Return to the caller the amount of time in mico-seconds * that should be used for the TLP timer from the last * send time of this packet. */ static uint32_t bbr_calc_thresh_tlp(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t srtt, uint32_t cts) { uint32_t thresh, len, maxseg, t_rxtcur; struct bbr_sendmap *prsm; if (srtt == 0) srtt = 1; if (bbr->rc_tlp_threshold) thresh = srtt + (srtt / bbr->rc_tlp_threshold); else thresh = (srtt * 2); maxseg = tp->t_maxseg - bbr->rc_last_options; /* Get the previous sent packet, if any */ len = rsm->r_end - rsm->r_start; /* 2.1 behavior */ prsm = TAILQ_PREV(rsm, bbr_head, r_tnext); if (prsm && (len <= maxseg)) { /* * Two packets outstanding, thresh should be (2*srtt) + * possible inter-packet delay (if any). */ uint32_t inter_gap = 0; int idx, nidx; idx = rsm->r_rtr_cnt - 1; nidx = prsm->r_rtr_cnt - 1; if (TSTMP_GEQ(rsm->r_tim_lastsent[nidx], prsm->r_tim_lastsent[idx])) { /* Yes it was sent later (or at the same time) */ inter_gap = rsm->r_tim_lastsent[idx] - prsm->r_tim_lastsent[nidx]; } thresh += inter_gap; } else if (len <= maxseg) { /* * Possibly compensate for delayed-ack. */ uint32_t alt_thresh; alt_thresh = srtt + (srtt / 2) + bbr_delayed_ack_time; if (alt_thresh > thresh) thresh = alt_thresh; } /* Not above the current RTO */ if (tp->t_srtt == 0) t_rxtcur = BBR_INITIAL_RTO; else t_rxtcur = TICKS_2_USEC(tp->t_rxtcur); bbr_log_thresh_choice(bbr, cts, thresh, t_rxtcur, srtt, rsm, BBR_TO_FRM_TLP); /* Not above an RTO */ if (thresh > t_rxtcur) { thresh = t_rxtcur; } /* Not above a RTO max */ if (thresh > (((uint32_t)bbr->rc_max_rto_sec) * USECS_IN_SECOND)) { thresh = (((uint32_t)bbr->rc_max_rto_sec) * USECS_IN_SECOND); } /* And now apply the user TLP min */ if (thresh < bbr_tlp_min) { thresh = bbr_tlp_min; } return (thresh); } /* * Return one of three RTTs to use (in microseconds). */ static __inline uint32_t bbr_get_rtt(struct tcp_bbr *bbr, int32_t rtt_type) { uint32_t f_rtt; uint32_t srtt; f_rtt = get_filter_value_small(&bbr->r_ctl.rc_rttprop); if (get_filter_value_small(&bbr->r_ctl.rc_rttprop) == 0xffffffff) { /* We have no rtt at all */ if (bbr->rc_tp->t_srtt == 0) f_rtt = BBR_INITIAL_RTO; else f_rtt = (TICKS_2_USEC(bbr->rc_tp->t_srtt) >> TCP_RTT_SHIFT); /* * Since we don't know how good the rtt is apply a * delayed-ack min */ if (f_rtt < bbr_delayed_ack_time) { f_rtt = bbr_delayed_ack_time; } } /* Take the filter version or last measured pkt-rtt */ if (rtt_type == BBR_RTT_PROP) { srtt = f_rtt; } else if (rtt_type == BBR_RTT_PKTRTT) { if (bbr->r_ctl.rc_pkt_epoch_rtt) { srtt = bbr->r_ctl.rc_pkt_epoch_rtt; } else { /* No pkt rtt yet */ srtt = f_rtt; } } else if (rtt_type == BBR_RTT_RACK) { srtt = bbr->r_ctl.rc_last_rtt; /* We need to add in any internal delay for our timer */ if (bbr->rc_ack_was_delayed) srtt += bbr->r_ctl.rc_ack_hdwr_delay; } else if (rtt_type == BBR_SRTT) { srtt = (TICKS_2_USEC(bbr->rc_tp->t_srtt) >> TCP_RTT_SHIFT); } else { /* TSNH */ srtt = f_rtt; #ifdef BBR_INVARIANTS panic("Unknown rtt request type %d", rtt_type); #endif } return (srtt); } static int bbr_is_lost(struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t cts) { uint32_t thresh; thresh = bbr_calc_thresh_rack(bbr, bbr_get_rtt(bbr, BBR_RTT_RACK), cts, rsm); if ((cts - rsm->r_tim_lastsent[(rsm->r_rtr_cnt - 1)]) >= thresh) { /* It is lost (past time) */ return (1); } return (0); } /* * Return a sendmap if we need to retransmit something. */ static struct bbr_sendmap * bbr_check_recovery_mode(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { /* * Check to see that we don't need to fall into recovery. We will * need to do so if our oldest transmit is past the time we should * have had an ack. */ struct bbr_sendmap *rsm; int32_t idx; if (TAILQ_EMPTY(&bbr->r_ctl.rc_map)) { /* Nothing outstanding that we know of */ return (NULL); } rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if (rsm == NULL) { /* Nothing in the transmit map */ return (NULL); } if (tp->t_flags & TF_SENTFIN) { /* Fin restricted, don't find anything once a fin is sent */ return (NULL); } if (rsm->r_flags & BBR_ACKED) { /* * Ok the first one is acked (this really should not happen * since we remove the from the tmap once they are acked) */ rsm = bbr_find_lowest_rsm(bbr); if (rsm == NULL) return (NULL); } idx = rsm->r_rtr_cnt - 1; if (SEQ_LEQ(cts, rsm->r_tim_lastsent[idx])) { /* Send timestamp is the same or less? can't be ready */ return (NULL); } /* Get our RTT time */ if (bbr_is_lost(bbr, rsm, cts) && ((rsm->r_dupack >= DUP_ACK_THRESHOLD) || (rsm->r_flags & BBR_SACK_PASSED))) { if ((rsm->r_flags & BBR_MARKED_LOST) == 0) { rsm->r_flags |= BBR_MARKED_LOST; bbr->r_ctl.rc_lost += rsm->r_end - rsm->r_start; bbr->r_ctl.rc_lost_bytes += rsm->r_end - rsm->r_start; } bbr_cong_signal(tp, NULL, CC_NDUPACK, rsm); #ifdef BBR_INVARIANTS if ((rsm->r_end - rsm->r_start) == 0) panic("tp:%p bbr:%p rsm:%p length is 0?", tp, bbr, rsm); #endif return (rsm); } return (NULL); } /* * RACK Timer, here we simply do logging and house keeping. * the normal bbr_output_wtime() function will call the * appropriate thing to check if we need to do a RACK retransmit. * We return 1, saying don't proceed with bbr_output_wtime only * when all timers have been stopped (destroyed PCB?). */ static int bbr_timeout_rack(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { /* * This timer simply provides an internal trigger to send out data. * The check_recovery_mode call will see if there are needed * retransmissions, if so we will enter fast-recovery. The output * call may or may not do the same thing depending on sysctl * settings. */ uint32_t lost; if (bbr->rc_all_timers_stopped) { return (1); } if (TSTMP_LT(cts, bbr->r_ctl.rc_timer_exp)) { /* Its not time yet */ return (0); } BBR_STAT_INC(bbr_to_tot); lost = bbr->r_ctl.rc_lost; if (bbr->r_state && (bbr->r_state != tp->t_state)) bbr_set_state(tp, bbr, 0); bbr_log_to_event(bbr, cts, BBR_TO_FRM_RACK); if (bbr->r_ctl.rc_resend == NULL) { /* Lets do the check here */ bbr->r_ctl.rc_resend = bbr_check_recovery_mode(tp, bbr, cts); } if (bbr_policer_call_from_rack_to) bbr_lt_bw_sampling(bbr, cts, (bbr->r_ctl.rc_lost > lost)); bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_RACK; return (0); } static __inline void bbr_clone_rsm(struct tcp_bbr *bbr, struct bbr_sendmap *nrsm, struct bbr_sendmap *rsm, uint32_t start) { int idx; nrsm->r_start = start; nrsm->r_end = rsm->r_end; nrsm->r_rtr_cnt = rsm->r_rtr_cnt; nrsm->r_flags = rsm->r_flags; /* We don't transfer forward the SYN flag */ nrsm->r_flags &= ~BBR_HAS_SYN; /* We move forward the FIN flag, not that this should happen */ rsm->r_flags &= ~BBR_HAS_FIN; nrsm->r_dupack = rsm->r_dupack; nrsm->r_rtr_bytes = 0; nrsm->r_is_gain = rsm->r_is_gain; nrsm->r_is_drain = rsm->r_is_drain; nrsm->r_delivered = rsm->r_delivered; nrsm->r_ts_valid = rsm->r_ts_valid; nrsm->r_del_ack_ts = rsm->r_del_ack_ts; nrsm->r_del_time = rsm->r_del_time; nrsm->r_app_limited = rsm->r_app_limited; nrsm->r_first_sent_time = rsm->r_first_sent_time; nrsm->r_flight_at_send = rsm->r_flight_at_send; /* We split a piece the lower section looses any just_ret flag. */ nrsm->r_bbr_state = rsm->r_bbr_state; for (idx = 0; idx < nrsm->r_rtr_cnt; idx++) { nrsm->r_tim_lastsent[idx] = rsm->r_tim_lastsent[idx]; } rsm->r_end = nrsm->r_start; idx = min((bbr->rc_tp->t_maxseg - bbr->rc_last_options), bbr->r_ctl.rc_pace_max_segs); idx /= 8; /* Check if we got too small */ if ((rsm->r_is_smallmap == 0) && ((rsm->r_end - rsm->r_start) <= idx)) { bbr->r_ctl.rc_num_small_maps_alloced++; rsm->r_is_smallmap = 1; } /* Check the new one as well */ if ((nrsm->r_end - nrsm->r_start) <= idx) { bbr->r_ctl.rc_num_small_maps_alloced++; nrsm->r_is_smallmap = 1; } } static int bbr_sack_mergable(struct bbr_sendmap *at, uint32_t start, uint32_t end) { /* * Given a sack block defined by * start and end, and a current postion * at. Return 1 if either side of at * would show that the block is mergable * to that side. A block to be mergable * must have overlap with the start/end * and be in the SACK'd state. */ struct bbr_sendmap *l_rsm; struct bbr_sendmap *r_rsm; /* first get the either side blocks */ l_rsm = TAILQ_PREV(at, bbr_head, r_next); r_rsm = TAILQ_NEXT(at, r_next); if (l_rsm && (l_rsm->r_flags & BBR_ACKED)) { /* Potentially mergeable */ if ((l_rsm->r_end == start) || (SEQ_LT(start, l_rsm->r_end) && SEQ_GT(end, l_rsm->r_end))) { /* * map blk |------| * sack blk |------| * * map blk |------| * sack blk |------| */ return (1); } } if (r_rsm && (r_rsm->r_flags & BBR_ACKED)) { /* Potentially mergeable */ if ((r_rsm->r_start == end) || (SEQ_LT(start, r_rsm->r_start) && SEQ_GT(end, r_rsm->r_start))) { /* * map blk |---------| * sack blk |----| * * map blk |---------| * sack blk |-------| */ return (1); } } return (0); } static struct bbr_sendmap * bbr_merge_rsm(struct tcp_bbr *bbr, struct bbr_sendmap *l_rsm, struct bbr_sendmap *r_rsm) { /* * We are merging two ack'd RSM's, * the l_rsm is on the left (lower seq * values) and the r_rsm is on the right * (higher seq value). The simplest way * to merge these is to move the right * one into the left. I don't think there * is any reason we need to try to find * the oldest (or last oldest retransmitted). */ l_rsm->r_end = r_rsm->r_end; if (l_rsm->r_dupack < r_rsm->r_dupack) l_rsm->r_dupack = r_rsm->r_dupack; if (r_rsm->r_rtr_bytes) l_rsm->r_rtr_bytes += r_rsm->r_rtr_bytes; if (r_rsm->r_in_tmap) { /* This really should not happen */ TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, r_rsm, r_tnext); } if (r_rsm->r_app_limited) l_rsm->r_app_limited = r_rsm->r_app_limited; /* Now the flags */ if (r_rsm->r_flags & BBR_HAS_FIN) l_rsm->r_flags |= BBR_HAS_FIN; if (r_rsm->r_flags & BBR_TLP) l_rsm->r_flags |= BBR_TLP; if (r_rsm->r_flags & BBR_RWND_COLLAPSED) l_rsm->r_flags |= BBR_RWND_COLLAPSED; if (r_rsm->r_flags & BBR_MARKED_LOST) { /* This really should not happen */ bbr->r_ctl.rc_lost_bytes -= r_rsm->r_end - r_rsm->r_start; } TAILQ_REMOVE(&bbr->r_ctl.rc_map, r_rsm, r_next); if ((r_rsm->r_limit_type == 0) && (l_rsm->r_limit_type != 0)) { /* Transfer the split limit to the map we free */ r_rsm->r_limit_type = l_rsm->r_limit_type; l_rsm->r_limit_type = 0; } bbr_free(bbr, r_rsm); return(l_rsm); } /* * TLP Timer, here we simply setup what segment we want to * have the TLP expire on, the normal bbr_output_wtime() will then * send it out. * * We return 1, saying don't proceed with bbr_output_wtime only * when all timers have been stopped (destroyed PCB?). */ static int bbr_timeout_tlp(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { /* * Tail Loss Probe. */ struct bbr_sendmap *rsm = NULL; struct socket *so; uint32_t amm; uint32_t out, avail; uint32_t maxseg; int collapsed_win = 0; if (bbr->rc_all_timers_stopped) { return (1); } if (TSTMP_LT(cts, bbr->r_ctl.rc_timer_exp)) { /* Its not time yet */ return (0); } if (bbr_progress_timeout_check(bbr)) { tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); return (1); } /* Did we somehow get into persists? */ if (bbr->rc_in_persist) { return (0); } if (bbr->r_state && (bbr->r_state != tp->t_state)) bbr_set_state(tp, bbr, 0); BBR_STAT_INC(bbr_tlp_tot); maxseg = tp->t_maxseg - bbr->rc_last_options; #ifdef KERN_TLS if (bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) { /* * For hardware TLS we do *not* want to send * new data. */ goto need_retran; } #endif /* * A TLP timer has expired. We have been idle for 2 rtts. So we now * need to figure out how to force a full MSS segment out. */ so = tp->t_inpcb->inp_socket; avail = sbavail(&so->so_snd); out = ctf_outstanding(tp); if (out > tp->snd_wnd) { /* special case, we need a retransmission */ collapsed_win = 1; goto need_retran; } if (avail > out) { /* New data is available */ amm = avail - out; if (amm > maxseg) { amm = maxseg; } else if ((amm < maxseg) && ((tp->t_flags & TF_NODELAY) == 0)) { /* not enough to fill a MTU and no-delay is off */ goto need_retran; } /* Set the send-new override */ if ((out + amm) <= tp->snd_wnd) { bbr->rc_tlp_new_data = 1; } else { goto need_retran; } bbr->r_ctl.rc_tlp_seg_send_cnt = 0; bbr->r_ctl.rc_last_tlp_seq = tp->snd_max; bbr->r_ctl.rc_tlp_send = NULL; /* cap any slots */ BBR_STAT_INC(bbr_tlp_newdata); goto send; } need_retran: /* * Ok we need to arrange the last un-acked segment to be re-sent, or * optionally the first un-acked segment. */ if (collapsed_win == 0) { rsm = TAILQ_LAST_FAST(&bbr->r_ctl.rc_map, bbr_sendmap, r_next); if (rsm && (BBR_ACKED | BBR_HAS_FIN)) { rsm = bbr_find_high_nonack(bbr, rsm); } if (rsm == NULL) { goto restore; } } else { /* * We must find the last segment * that was acceptable by the client. */ TAILQ_FOREACH_REVERSE(rsm, &bbr->r_ctl.rc_map, bbr_head, r_next) { if ((rsm->r_flags & BBR_RWND_COLLAPSED) == 0) { /* Found one */ break; } } if (rsm == NULL) { /* None? if so send the first */ rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); if (rsm == NULL) goto restore; } } if ((rsm->r_end - rsm->r_start) > maxseg) { /* * We need to split this the last segment in two. */ struct bbr_sendmap *nrsm; nrsm = bbr_alloc_full_limit(bbr); if (nrsm == NULL) { /* * We can't get memory to split, we can either just * not split it. Or retransmit the whole piece, lets * do the large send (BTLP :-) ). */ goto go_for_it; } bbr_clone_rsm(bbr, nrsm, rsm, (rsm->r_end - maxseg)); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } rsm->r_flags &= (~BBR_HAS_FIN); rsm = nrsm; } go_for_it: bbr->r_ctl.rc_tlp_send = rsm; bbr->rc_tlp_rtx_out = 1; if (rsm->r_start == bbr->r_ctl.rc_last_tlp_seq) { bbr->r_ctl.rc_tlp_seg_send_cnt++; tp->t_rxtshift++; } else { bbr->r_ctl.rc_last_tlp_seq = rsm->r_start; bbr->r_ctl.rc_tlp_seg_send_cnt = 1; } send: if (bbr->r_ctl.rc_tlp_seg_send_cnt > bbr_tlp_max_resend) { /* * Can't [re]/transmit a segment we have retranmitted the * max times. We need the retransmit timer to take over. */ restore: bbr->rc_tlp_new_data = 0; bbr->r_ctl.rc_tlp_send = NULL; if (rsm) rsm->r_flags &= ~BBR_TLP; BBR_STAT_INC(bbr_tlp_retran_fail); return (0); } else if (rsm) { rsm->r_flags |= BBR_TLP; } if (rsm && (rsm->r_start == bbr->r_ctl.rc_last_tlp_seq) && (bbr->r_ctl.rc_tlp_seg_send_cnt > bbr_tlp_max_resend)) { /* * We have retransmitted to many times for TLP. Switch to * the regular RTO timer */ goto restore; } bbr_log_to_event(bbr, cts, BBR_TO_FRM_TLP); bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_TLP; return (0); } /* * Delayed ack Timer, here we simply need to setup the * ACK_NOW flag and remove the DELACK flag. From there * the output routine will send the ack out. * * We only return 1, saying don't proceed, if all timers * are stopped (destroyed PCB?). */ static int bbr_timeout_delack(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { if (bbr->rc_all_timers_stopped) { return (1); } bbr_log_to_event(bbr, cts, BBR_TO_FRM_DELACK); tp->t_flags &= ~TF_DELACK; tp->t_flags |= TF_ACKNOW; KMOD_TCPSTAT_INC(tcps_delack); bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_DELACK; return (0); } /* * Persists timer, here we simply need to setup the * FORCE-DATA flag the output routine will send * the one byte send. * * We only return 1, saying don't proceed, if all timers * are stopped (destroyed PCB?). */ static int bbr_timeout_persist(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { struct tcptemp *t_template; int32_t retval = 1; if (bbr->rc_all_timers_stopped) { return (1); } if (bbr->rc_in_persist == 0) return (0); KASSERT(tp->t_inpcb != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp)); /* * Persistence timer into zero window. Force a byte to be output, if * possible. */ bbr_log_to_event(bbr, cts, BBR_TO_FRM_PERSIST); bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_PERSIT; KMOD_TCPSTAT_INC(tcps_persisttimeo); /* * Have we exceeded the user specified progress time? */ if (bbr_progress_timeout_check(bbr)) { tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); goto out; } /* * Hack: if the peer is dead/unreachable, we do not time out if the * window is closed. After a full backoff, drop the connection if * the idle time (no responses to probes) reaches the maximum * backoff that we would use if retransmitting. */ if (tp->t_rxtshift == TCP_MAXRXTSHIFT && (ticks - tp->t_rcvtime >= tcp_maxpersistidle || ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { KMOD_TCPSTAT_INC(tcps_persistdrop); tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); goto out; } if ((sbavail(&bbr->rc_inp->inp_socket->so_snd) == 0) && tp->snd_una == tp->snd_max) { bbr_exit_persist(tp, bbr, cts, __LINE__); retval = 0; goto out; } /* * If the user has closed the socket then drop a persisting * connection after a much reduced timeout. */ if (tp->t_state > TCPS_CLOSE_WAIT && (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) { KMOD_TCPSTAT_INC(tcps_persistdrop); tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); goto out; } t_template = tcpip_maketemplate(bbr->rc_inp); if (t_template) { tcp_respond(tp, t_template->tt_ipgen, &t_template->tt_t, (struct mbuf *)NULL, tp->rcv_nxt, tp->snd_una - 1, 0); /* This sends an ack */ if (tp->t_flags & TF_DELACK) tp->t_flags &= ~TF_DELACK; free(t_template, M_TEMP); } if (tp->t_rxtshift < TCP_MAXRXTSHIFT) tp->t_rxtshift++; bbr_start_hpts_timer(bbr, tp, cts, 3, 0, 0); out: return (retval); } /* * If a keepalive goes off, we had no other timers * happening. We always return 1 here since this * routine either drops the connection or sends * out a segment with respond. */ static int bbr_timeout_keepalive(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { struct tcptemp *t_template; struct inpcb *inp; if (bbr->rc_all_timers_stopped) { return (1); } bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_KEEP; inp = tp->t_inpcb; bbr_log_to_event(bbr, cts, BBR_TO_FRM_KEEP); /* * Keep-alive timer went off; send something or drop connection if * idle for too long. */ KMOD_TCPSTAT_INC(tcps_keeptimeo); if (tp->t_state < TCPS_ESTABLISHED) goto dropit; if ((V_tcp_always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) && tp->t_state <= TCPS_CLOSING) { if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp)) goto dropit; /* * Send a packet designed to force a response if the peer is * up and reachable: either an ACK if the connection is * still alive, or an RST if the peer has closed the * connection due to timeout or reboot. Using sequence * number tp->snd_una-1 causes the transmitted zero-length * segment to lie outside the receive window; by the * protocol spec, this requires the correspondent TCP to * respond. */ KMOD_TCPSTAT_INC(tcps_keepprobe); t_template = tcpip_maketemplate(inp); if (t_template) { tcp_respond(tp, t_template->tt_ipgen, &t_template->tt_t, (struct mbuf *)NULL, tp->rcv_nxt, tp->snd_una - 1, 0); free(t_template, M_TEMP); } } bbr_start_hpts_timer(bbr, tp, cts, 4, 0, 0); return (1); dropit: KMOD_TCPSTAT_INC(tcps_keepdrops); tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); return (1); } /* * Retransmit helper function, clear up all the ack * flags and take care of important book keeping. */ static void bbr_remxt_tmr(struct tcpcb *tp) { /* * The retransmit timer went off, all sack'd blocks must be * un-acked. */ struct bbr_sendmap *rsm, *trsm = NULL; struct tcp_bbr *bbr; uint32_t cts, lost; bbr = (struct tcp_bbr *)tp->t_fb_ptr; cts = tcp_get_usecs(&bbr->rc_tv); lost = bbr->r_ctl.rc_lost; if (bbr->r_state && (bbr->r_state != tp->t_state)) bbr_set_state(tp, bbr, 0); TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_map, r_next) { if (rsm->r_flags & BBR_ACKED) { uint32_t old_flags; rsm->r_dupack = 0; if (rsm->r_in_tmap == 0) { /* We must re-add it back to the tlist */ if (trsm == NULL) { TAILQ_INSERT_HEAD(&bbr->r_ctl.rc_tmap, rsm, r_tnext); } else { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, trsm, rsm, r_tnext); } rsm->r_in_tmap = 1; } old_flags = rsm->r_flags; rsm->r_flags |= BBR_RXT_CLEARED; rsm->r_flags &= ~(BBR_ACKED | BBR_SACK_PASSED | BBR_WAS_SACKPASS); bbr_log_type_rsmclear(bbr, cts, rsm, old_flags, __LINE__); } else { if ((rsm->r_flags & BBR_MARKED_LOST) == 0) { bbr->r_ctl.rc_lost += rsm->r_end - rsm->r_start; bbr->r_ctl.rc_lost_bytes += rsm->r_end - rsm->r_start; } if (bbr_marks_rxt_sack_passed) { /* * With this option, we will rack out * in 1ms increments the rest of the packets. */ rsm->r_flags |= BBR_SACK_PASSED | BBR_MARKED_LOST; rsm->r_flags &= ~BBR_WAS_SACKPASS; } else { /* * With this option we only mark them lost * and remove all sack'd markings. We will run * another RXT or a TLP. This will cause * us to eventually send more based on what * ack's come in. */ rsm->r_flags |= BBR_MARKED_LOST; rsm->r_flags &= ~BBR_WAS_SACKPASS; rsm->r_flags &= ~BBR_SACK_PASSED; } } trsm = rsm; } bbr->r_ctl.rc_resend = TAILQ_FIRST(&bbr->r_ctl.rc_map); /* Clear the count (we just un-acked them) */ bbr_log_to_event(bbr, cts, BBR_TO_FRM_TMR); bbr->rc_tlp_new_data = 0; bbr->r_ctl.rc_tlp_seg_send_cnt = 0; /* zap the behindness on a rxt */ bbr->r_ctl.rc_hptsi_agg_delay = 0; bbr->r_agg_early_set = 0; bbr->r_ctl.rc_agg_early = 0; bbr->rc_tlp_rtx_out = 0; bbr->r_ctl.rc_sacked = 0; bbr->r_ctl.rc_sacklast = NULL; bbr->r_timer_override = 1; bbr_lt_bw_sampling(bbr, cts, (bbr->r_ctl.rc_lost > lost)); } /* * Re-transmit timeout! If we drop the PCB we will return 1, otherwise * we will setup to retransmit the lowest seq number outstanding. */ static int bbr_timeout_rxt(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { int32_t rexmt; int32_t retval = 0; bool isipv6; bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_RXT; if (bbr->rc_all_timers_stopped) { return (1); } if (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->snd_una == tp->snd_max)) { /* Nothing outstanding .. nothing to do */ return (0); } /* * Retransmission timer went off. Message has not been acked within * retransmit interval. Back off to a longer retransmit interval * and retransmit one segment. */ if (bbr_progress_timeout_check(bbr)) { retval = 1; tcp_set_inp_to_drop(bbr->rc_inp, ETIMEDOUT); goto out; } bbr_remxt_tmr(tp); if ((bbr->r_ctl.rc_resend == NULL) || ((bbr->r_ctl.rc_resend->r_flags & BBR_RWND_COLLAPSED) == 0)) { /* * If the rwnd collapsed on * the one we are retransmitting * it does not count against the * rxt count. */ tp->t_rxtshift++; } if (tp->t_rxtshift > TCP_MAXRXTSHIFT) { tp->t_rxtshift = TCP_MAXRXTSHIFT; KMOD_TCPSTAT_INC(tcps_timeoutdrop); retval = 1; tcp_set_inp_to_drop(bbr->rc_inp, (tp->t_softerror ? (uint16_t) tp->t_softerror : ETIMEDOUT)); goto out; } if (tp->t_state == TCPS_SYN_SENT) { /* * If the SYN was retransmitted, indicate CWND to be limited * to 1 segment in cc_conn_init(). */ tp->snd_cwnd = 1; } else if (tp->t_rxtshift == 1) { /* * first retransmit; record ssthresh and cwnd so they can be * recovered if this turns out to be a "bad" retransmit. A * retransmit is considered "bad" if an ACK for this segment * is received within RTT/2 interval; the assumption here is * that the ACK was already in flight. See "On Estimating * End-to-End Network Path Properties" by Allman and Paxson * for more details. */ tp->snd_cwnd = tp->t_maxseg - bbr->rc_last_options; if (!IN_RECOVERY(tp->t_flags)) { tp->snd_cwnd_prev = tp->snd_cwnd; tp->snd_ssthresh_prev = tp->snd_ssthresh; tp->snd_recover_prev = tp->snd_recover; tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); tp->t_flags |= TF_PREVVALID; } else { tp->t_flags &= ~TF_PREVVALID; } tp->snd_cwnd = tp->t_maxseg - bbr->rc_last_options; } else { tp->snd_cwnd = tp->t_maxseg - bbr->rc_last_options; tp->t_flags &= ~TF_PREVVALID; } KMOD_TCPSTAT_INC(tcps_rexmttimeo); if ((tp->t_state == TCPS_SYN_SENT) || (tp->t_state == TCPS_SYN_RECEIVED)) rexmt = USEC_2_TICKS(BBR_INITIAL_RTO) * tcp_backoff[tp->t_rxtshift]; else rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; TCPT_RANGESET(tp->t_rxtcur, rexmt, MSEC_2_TICKS(bbr->r_ctl.rc_min_rto_ms), MSEC_2_TICKS(((uint32_t)bbr->rc_max_rto_sec) * 1000)); /* * We enter the path for PLMTUD if connection is established or, if * connection is FIN_WAIT_1 status, reason for the last is that if * amount of data we send is very small, we could send it in couple * of packets and process straight to FIN. In that case we won't * catch ESTABLISHED state. */ #ifdef INET6 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? true : false; #else isipv6 = false; #endif if (((V_tcp_pmtud_blackhole_detect == 1) || (V_tcp_pmtud_blackhole_detect == 2 && !isipv6) || (V_tcp_pmtud_blackhole_detect == 3 && isipv6)) && ((tp->t_state == TCPS_ESTABLISHED) || (tp->t_state == TCPS_FIN_WAIT_1))) { /* * Idea here is that at each stage of mtu probe (usually, * 1448 -> 1188 -> 524) should be given 2 chances to recover * before further clamping down. 'tp->t_rxtshift % 2 == 0' * should take care of that. */ if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD | TF2_PLPMTU_MAXSEGSNT)) == (TF2_PLPMTU_PMTUD | TF2_PLPMTU_MAXSEGSNT)) && (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 && tp->t_rxtshift % 2 == 0)) { /* * Enter Path MTU Black-hole Detection mechanism: - * Disable Path MTU Discovery (IP "DF" bit). - * Reduce MTU to lower value than what we negotiated * with peer. */ if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) { /* * Record that we may have found a black * hole. */ tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE; /* Keep track of previous MSS. */ tp->t_pmtud_saved_maxseg = tp->t_maxseg; } /* * Reduce the MSS to blackhole value or to the * default in an attempt to retransmit. */ #ifdef INET6 isipv6 = bbr->r_is_v6; if (isipv6 && tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) { /* Use the sysctl tuneable blackhole MSS. */ tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss; KMOD_TCPSTAT_INC(tcps_pmtud_blackhole_activated); } else if (isipv6) { /* Use the default MSS. */ tp->t_maxseg = V_tcp_v6mssdflt; /* * Disable Path MTU Discovery when we switch * to minmss. */ tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; KMOD_TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) { /* Use the sysctl tuneable blackhole MSS. */ tp->t_maxseg = V_tcp_pmtud_blackhole_mss; KMOD_TCPSTAT_INC(tcps_pmtud_blackhole_activated); } else { /* Use the default MSS. */ tp->t_maxseg = V_tcp_mssdflt; /* * Disable Path MTU Discovery when we switch * to minmss. */ tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; KMOD_TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss); } #endif } else { /* * If further retransmissions are still unsuccessful * with a lowered MTU, maybe this isn't a blackhole * and we restore the previous MSS and blackhole * detection flags. The limit '6' is determined by * giving each probe stage (1448, 1188, 524) 2 * chances to recover. */ if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) && (tp->t_rxtshift >= 6)) { tp->t_flags2 |= TF2_PLPMTU_PMTUD; tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE; tp->t_maxseg = tp->t_pmtud_saved_maxseg; KMOD_TCPSTAT_INC(tcps_pmtud_blackhole_failed); } } } /* * Disable RFC1323 and SACK if we haven't got any response to our * third SYN to work-around some broken terminal servers (most of * which have hopefully been retired) that have bad VJ header * compression code which trashes TCP segments containing * unknown-to-them TCP options. */ if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3)) tp->t_flags &= ~(TF_REQ_SCALE | TF_REQ_TSTMP | TF_SACK_PERMIT); /* * If we backed off this far, our srtt estimate is probably bogus. * Clobber it so we'll take the next rtt measurement as our srtt; * move the current srtt into rttvar to keep the current retransmit * times until then. */ if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { #ifdef INET6 if (bbr->r_is_v6) in6_losing(tp->t_inpcb); else #endif in_losing(tp->t_inpcb); tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); tp->t_srtt = 0; } sack_filter_clear(&bbr->r_ctl.bbr_sf, tp->snd_una); tp->snd_recover = tp->snd_max; tp->t_flags |= TF_ACKNOW; tp->t_rtttime = 0; out: return (retval); } static int bbr_process_timers(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, uint8_t hpts_calling) { int32_t ret = 0; int32_t timers = (bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK); if (timers == 0) { return (0); } if (tp->t_state == TCPS_LISTEN) { /* no timers on listen sockets */ if (bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) return (0); return (1); } if (TSTMP_LT(cts, bbr->r_ctl.rc_timer_exp)) { uint32_t left; if (bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) { ret = -1; bbr_log_to_processing(bbr, cts, ret, 0, hpts_calling); return (0); } if (hpts_calling == 0) { ret = -2; bbr_log_to_processing(bbr, cts, ret, 0, hpts_calling); return (0); } /* * Ok our timer went off early and we are not paced false * alarm, go back to sleep. */ left = bbr->r_ctl.rc_timer_exp - cts; ret = -3; bbr_log_to_processing(bbr, cts, ret, left, hpts_calling); tcp_hpts_insert(tp->t_inpcb, HPTS_USEC_TO_SLOTS(left)); return (1); } bbr->rc_tmr_stopped = 0; bbr->r_ctl.rc_hpts_flags &= ~PACE_TMR_MASK; if (timers & PACE_TMR_DELACK) { ret = bbr_timeout_delack(tp, bbr, cts); } else if (timers & PACE_TMR_PERSIT) { ret = bbr_timeout_persist(tp, bbr, cts); } else if (timers & PACE_TMR_RACK) { bbr->r_ctl.rc_tlp_rxt_last_time = cts; ret = bbr_timeout_rack(tp, bbr, cts); } else if (timers & PACE_TMR_TLP) { bbr->r_ctl.rc_tlp_rxt_last_time = cts; ret = bbr_timeout_tlp(tp, bbr, cts); } else if (timers & PACE_TMR_RXT) { bbr->r_ctl.rc_tlp_rxt_last_time = cts; ret = bbr_timeout_rxt(tp, bbr, cts); } else if (timers & PACE_TMR_KEEP) { ret = bbr_timeout_keepalive(tp, bbr, cts); } bbr_log_to_processing(bbr, cts, ret, timers, hpts_calling); return (ret); } static void bbr_timer_cancel(struct tcp_bbr *bbr, int32_t line, uint32_t cts) { if (bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK) { uint8_t hpts_removed = 0; if (bbr->rc_inp->inp_in_hpts && (bbr->rc_timer_first == 1)) { /* * If we are canceling timer's when we have the * timer ahead of the output being paced. We also * must remove ourselves from the hpts. */ hpts_removed = 1; tcp_hpts_remove(bbr->rc_inp, HPTS_REMOVE_OUTPUT); if (bbr->r_ctl.rc_last_delay_val) { /* Update the last hptsi delay too */ uint32_t time_since_send; if (TSTMP_GT(cts, bbr->rc_pacer_started)) time_since_send = cts - bbr->rc_pacer_started; else time_since_send = 0; if (bbr->r_ctl.rc_last_delay_val > time_since_send) { /* Cut down our slot time */ bbr->r_ctl.rc_last_delay_val -= time_since_send; } else { bbr->r_ctl.rc_last_delay_val = 0; } bbr->rc_pacer_started = cts; } } bbr->rc_timer_first = 0; bbr_log_to_cancel(bbr, line, cts, hpts_removed); bbr->rc_tmr_stopped = bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK; bbr->r_ctl.rc_hpts_flags &= ~(PACE_TMR_MASK); } } static void bbr_timer_stop(struct tcpcb *tp, uint32_t timer_type) { struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; bbr->rc_all_timers_stopped = 1; return; } /* * stop all timers always returning 0. */ static int bbr_stopall(struct tcpcb *tp) { return (0); } static void bbr_timer_activate(struct tcpcb *tp, uint32_t timer_type, uint32_t delta) { return; } /* * return true if a bbr timer (rack or tlp) is active. */ static int bbr_timer_active(struct tcpcb *tp, uint32_t timer_type) { return (0); } static uint32_t bbr_get_earliest_send_outstanding(struct tcp_bbr *bbr, struct bbr_sendmap *u_rsm, uint32_t cts) { struct bbr_sendmap *rsm; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if ((rsm == NULL) || (u_rsm == rsm)) return (cts); return(rsm->r_tim_lastsent[(rsm->r_rtr_cnt-1)]); } static void bbr_update_rsm(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t cts, uint32_t pacing_time) { int32_t idx; rsm->r_rtr_cnt++; rsm->r_dupack = 0; if (rsm->r_rtr_cnt > BBR_NUM_OF_RETRANS) { rsm->r_rtr_cnt = BBR_NUM_OF_RETRANS; rsm->r_flags |= BBR_OVERMAX; } if (rsm->r_flags & BBR_RWND_COLLAPSED) { /* Take off the collapsed flag at rxt */ rsm->r_flags &= ~BBR_RWND_COLLAPSED; } if (rsm->r_flags & BBR_MARKED_LOST) { /* We have retransmitted, its no longer lost */ rsm->r_flags &= ~BBR_MARKED_LOST; bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; } if (rsm->r_flags & BBR_RXT_CLEARED) { /* * We hit a RXT timer on it and * we cleared the "acked" flag. * We now have it going back into * flight, we can remove the cleared * flag and possibly do accounting on * this piece. */ rsm->r_flags &= ~BBR_RXT_CLEARED; } if ((rsm->r_rtr_cnt > 1) && ((rsm->r_flags & BBR_TLP) == 0)) { bbr->r_ctl.rc_holes_rxt += (rsm->r_end - rsm->r_start); rsm->r_rtr_bytes += (rsm->r_end - rsm->r_start); } idx = rsm->r_rtr_cnt - 1; rsm->r_tim_lastsent[idx] = cts; rsm->r_pacing_delay = pacing_time; rsm->r_delivered = bbr->r_ctl.rc_delivered; rsm->r_ts_valid = bbr->rc_ts_valid; if (bbr->rc_ts_valid) rsm->r_del_ack_ts = bbr->r_ctl.last_inbound_ts; if (bbr->r_ctl.r_app_limited_until) rsm->r_app_limited = 1; else rsm->r_app_limited = 0; if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) rsm->r_bbr_state = bbr_state_val(bbr); else rsm->r_bbr_state = 8; if (rsm->r_flags & BBR_ACKED) { /* Problably MTU discovery messing with us */ uint32_t old_flags; old_flags = rsm->r_flags; rsm->r_flags &= ~BBR_ACKED; bbr_log_type_rsmclear(bbr, cts, rsm, old_flags, __LINE__); bbr->r_ctl.rc_sacked -= (rsm->r_end - rsm->r_start); if (bbr->r_ctl.rc_sacked == 0) bbr->r_ctl.rc_sacklast = NULL; } if (rsm->r_in_tmap) { TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, rsm, r_tnext); } TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 1; if (rsm->r_flags & BBR_SACK_PASSED) { /* We have retransmitted due to the SACK pass */ rsm->r_flags &= ~BBR_SACK_PASSED; rsm->r_flags |= BBR_WAS_SACKPASS; } rsm->r_first_sent_time = bbr_get_earliest_send_outstanding(bbr, rsm, cts); rsm->r_flight_at_send = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); bbr->r_ctl.rc_next = TAILQ_NEXT(rsm, r_next); if (bbr->r_ctl.rc_bbr_hptsi_gain > BBR_UNIT) { rsm->r_is_gain = 1; rsm->r_is_drain = 0; } else if (bbr->r_ctl.rc_bbr_hptsi_gain < BBR_UNIT) { rsm->r_is_drain = 1; rsm->r_is_gain = 0; } else { rsm->r_is_drain = 0; rsm->r_is_gain = 0; } rsm->r_del_time = bbr->r_ctl.rc_del_time; /* TEMP GOOGLE CODE */ } /* * Returns 0, or the sequence where we stopped * updating. We also update the lenp to be the amount * of data left. */ static uint32_t bbr_update_entry(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t cts, int32_t *lenp, uint32_t pacing_time) { /* * We (re-)transmitted starting at rsm->r_start for some length * (possibly less than r_end. */ struct bbr_sendmap *nrsm; uint32_t c_end; int32_t len; len = *lenp; c_end = rsm->r_start + len; if (SEQ_GEQ(c_end, rsm->r_end)) { /* * We retransmitted the whole piece or more than the whole * slopping into the next rsm. */ bbr_update_rsm(tp, bbr, rsm, cts, pacing_time); if (c_end == rsm->r_end) { *lenp = 0; return (0); } else { int32_t act_len; /* Hangs over the end return whats left */ act_len = rsm->r_end - rsm->r_start; *lenp = (len - act_len); return (rsm->r_end); } /* We don't get out of this block. */ } /* * Here we retransmitted less than the whole thing which means we * have to split this into what was transmitted and what was not. */ nrsm = bbr_alloc_full_limit(bbr); if (nrsm == NULL) { *lenp = 0; return (0); } /* * So here we are going to take the original rsm and make it what we * retransmitted. nrsm will be the tail portion we did not * retransmit. For example say the chunk was 1, 11 (10 bytes). And * we retransmitted 5 bytes i.e. 1, 5. The original piece shrinks to * 1, 6 and the new piece will be 6, 11. */ bbr_clone_rsm(bbr, nrsm, rsm, c_end); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); nrsm->r_dupack = 0; if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } rsm->r_flags &= (~BBR_HAS_FIN); bbr_update_rsm(tp, bbr, rsm, cts, pacing_time); *lenp = 0; return (0); } static uint64_t bbr_get_hardware_rate(struct tcp_bbr *bbr) { uint64_t bw; bw = bbr_get_bw(bbr); bw *= (uint64_t)bbr_hptsi_gain[BBR_SUB_GAIN]; bw /= (uint64_t)BBR_UNIT; return(bw); } static void bbr_setup_less_of_rate(struct tcp_bbr *bbr, uint32_t cts, uint64_t act_rate, uint64_t rate_wanted) { /* * We could not get a full gains worth * of rate. */ if (get_filter_value(&bbr->r_ctl.rc_delrate) >= act_rate) { /* we can't even get the real rate */ uint64_t red; bbr->skip_gain = 1; bbr->gain_is_limited = 0; red = get_filter_value(&bbr->r_ctl.rc_delrate) - act_rate; if (red) filter_reduce_by(&bbr->r_ctl.rc_delrate, red, cts); } else { /* We can use a lower gain */ bbr->skip_gain = 0; bbr->gain_is_limited = 1; } } static void bbr_update_hardware_pacing_rate(struct tcp_bbr *bbr, uint32_t cts) { const struct tcp_hwrate_limit_table *nrte; int error, rate = -1; if (bbr->r_ctl.crte == NULL) return; - if ((bbr->rc_inp->inp_route.ro_rt == NULL) || - (bbr->rc_inp->inp_route.ro_rt->rt_ifp == NULL)) { + if ((bbr->rc_inp->inp_route.ro_nh == NULL) || + (bbr->rc_inp->inp_route.ro_nh->nh_ifp == NULL)) { /* Lost our routes? */ /* Clear the way for a re-attempt */ bbr->bbr_attempt_hdwr_pace = 0; lost_rate: bbr->gain_is_limited = 0; bbr->skip_gain = 0; bbr->bbr_hdrw_pacing = 0; counter_u64_add(bbr_flows_whdwr_pacing, -1); counter_u64_add(bbr_flows_nohdwr_pacing, 1); tcp_bbr_tso_size_check(bbr, cts); return; } rate = bbr_get_hardware_rate(bbr); nrte = tcp_chg_pacing_rate(bbr->r_ctl.crte, bbr->rc_tp, - bbr->rc_inp->inp_route.ro_rt->rt_ifp, + bbr->rc_inp->inp_route.ro_nh->nh_ifp, rate, (RS_PACING_GEQ|RS_PACING_SUB_OK), &error); if (nrte == NULL) { goto lost_rate; } if (nrte != bbr->r_ctl.crte) { bbr->r_ctl.crte = nrte; if (error == 0) { BBR_STAT_INC(bbr_hdwr_rl_mod_ok); if (bbr->r_ctl.crte->rate < rate) { /* We have a problem */ bbr_setup_less_of_rate(bbr, cts, bbr->r_ctl.crte->rate, rate); } else { /* We are good */ bbr->gain_is_limited = 0; bbr->skip_gain = 0; } } else { /* A failure should release the tag */ BBR_STAT_INC(bbr_hdwr_rl_mod_fail); bbr->gain_is_limited = 0; bbr->skip_gain = 0; bbr->bbr_hdrw_pacing = 0; } bbr_type_log_hdwr_pacing(bbr, bbr->r_ctl.crte->ptbl->rs_ifp, rate, ((bbr->r_ctl.crte == NULL) ? 0 : bbr->r_ctl.crte->rate), __LINE__, cts, error); } } static void bbr_adjust_for_hw_pacing(struct tcp_bbr *bbr, uint32_t cts) { /* * If we have hardware pacing support * we need to factor that in for our * TSO size. */ const struct tcp_hwrate_limit_table *rlp; uint32_t cur_delay, seg_sz, maxseg, new_tso, delta, hdwr_delay; if ((bbr->bbr_hdrw_pacing == 0) || (IN_RECOVERY(bbr->rc_tp->t_flags)) || (bbr->r_ctl.crte == NULL)) return; if (bbr->hw_pacing_set == 0) { /* Not yet by the hdwr pacing count delay */ return; } if (bbr_hdwr_pace_adjust == 0) { /* No adjustment */ return; } rlp = bbr->r_ctl.crte; if (bbr->rc_tp->t_maxseg > bbr->rc_last_options) maxseg = bbr->rc_tp->t_maxseg - bbr->rc_last_options; else maxseg = BBR_MIN_SEG - bbr->rc_last_options; /* * So lets first get the * time we will take between * TSO sized sends currently without * hardware help. */ cur_delay = bbr_get_pacing_delay(bbr, BBR_UNIT, bbr->r_ctl.rc_pace_max_segs, cts, 1); hdwr_delay = bbr->r_ctl.rc_pace_max_segs / maxseg; hdwr_delay *= rlp->time_between; if (cur_delay > hdwr_delay) delta = cur_delay - hdwr_delay; else delta = 0; bbr_log_type_tsosize(bbr, cts, delta, cur_delay, hdwr_delay, (bbr->r_ctl.rc_pace_max_segs / maxseg), 1); if (delta && (delta < (max(rlp->time_between, bbr->r_ctl.bbr_hptsi_segments_delay_tar)))) { /* * Now lets divide by the pacing * time between each segment the * hardware sends rounding up and * derive a bytes from that. We multiply * that by bbr_hdwr_pace_adjust to get * more bang for our buck. * * The goal is to have the software pacer * waiting no more than an additional * pacing delay if we can (without the * compensation i.e. x bbr_hdwr_pace_adjust). */ seg_sz = max(((cur_delay + rlp->time_between)/rlp->time_between), (bbr->r_ctl.rc_pace_max_segs/maxseg)); seg_sz *= bbr_hdwr_pace_adjust; if (bbr_hdwr_pace_floor && (seg_sz < bbr->r_ctl.crte->ptbl->rs_min_seg)) { /* Currently hardware paces * out rs_min_seg segments at a time. * We need to make sure we always send at least * a full burst of bbr_hdwr_pace_floor down. */ seg_sz = bbr->r_ctl.crte->ptbl->rs_min_seg; } seg_sz *= maxseg; } else if (delta == 0) { /* * The highest pacing rate is * above our b/w gained. This means * we probably are going quite fast at * the hardware highest rate. Lets just multiply * the calculated TSO size by the * multiplier factor (its probably * 4 segments in the default config for * mlx). */ seg_sz = bbr->r_ctl.rc_pace_max_segs * bbr_hdwr_pace_adjust; if (bbr_hdwr_pace_floor && (seg_sz < bbr->r_ctl.crte->ptbl->rs_min_seg)) { /* Currently hardware paces * out rs_min_seg segments at a time. * We need to make sure we always send at least * a full burst of bbr_hdwr_pace_floor down. */ seg_sz = bbr->r_ctl.crte->ptbl->rs_min_seg; } } else { /* * The pacing time difference is so * big that the hardware will * pace out more rapidly then we * really want and then we * will have a long delay. Lets just keep * the same TSO size so its as if * we were not using hdwr pacing (we * just gain a bit of spacing from the * hardware if seg_sz > 1). */ seg_sz = bbr->r_ctl.rc_pace_max_segs; } if (seg_sz > bbr->r_ctl.rc_pace_max_segs) new_tso = seg_sz; else new_tso = bbr->r_ctl.rc_pace_max_segs; if (new_tso >= (PACE_MAX_IP_BYTES-maxseg)) new_tso = PACE_MAX_IP_BYTES - maxseg; if (new_tso != bbr->r_ctl.rc_pace_max_segs) { bbr_log_type_tsosize(bbr, cts, new_tso, 0, bbr->r_ctl.rc_pace_max_segs, maxseg, 0); bbr->r_ctl.rc_pace_max_segs = new_tso; } } static void tcp_bbr_tso_size_check(struct tcp_bbr *bbr, uint32_t cts) { uint64_t bw; uint32_t old_tso = 0, new_tso; uint32_t maxseg, bytes; uint32_t tls_seg=0; /* * Google/linux uses the following algorithm to determine * the TSO size based on the b/w of the link (from Neal Cardwell email 9/27/18): * * bytes = bw_in_bytes_per_second / 1000 * bytes = min(bytes, 64k) * tso_segs = bytes / MSS * if (bw < 1.2Mbs) * min_tso_segs = 1 * else * min_tso_segs = 2 * tso_segs = max(tso_segs, min_tso_segs) * * * Note apply a device specific limit (we apply this in the * tcp_m_copym). * Note that before the initial measurement is made google bursts out * a full iwnd just like new-reno/cubic. * * We do not use this algorithm. Instead we * use a two phased approach: * * if ( bw <= per-tcb-cross-over) * goal_tso = calculate how much with this bw we * can send in goal-time seconds. * if (goal_tso > mss) * seg = goal_tso / mss * tso = seg * mss * else * tso = mss * if (tso > per-tcb-max) * tso = per-tcb-max * else if ( bw > 512Mbps) * tso = max-tso (64k/mss) * else * goal_tso = bw / per-tcb-divsor * seg = (goal_tso + mss-1)/mss * tso = seg * mss * * if (tso < per-tcb-floor) * tso = per-tcb-floor * if (tso > per-tcb-utter_max) * tso = per-tcb-utter_max * * Note the default per-tcb-divisor is 1000 (same as google). * the goal cross over is 30Mbps however. To recreate googles * algorithm you need to set: * * cross-over = 23,168,000 bps * goal-time = 18000 * per-tcb-max = 2 * per-tcb-divisor = 1000 * per-tcb-floor = 1 * * This will get you "google bbr" behavior with respect to tso size. * * Note we do set anything TSO size until we are past the initial * window. Before that we gnerally use either a single MSS * or we use the full IW size (so we burst a IW at a time) * Also note that Hardware-TLS is special and does alternate * things to minimize PCI Bus Bandwidth use. */ if (bbr->rc_tp->t_maxseg > bbr->rc_last_options) { maxseg = bbr->rc_tp->t_maxseg - bbr->rc_last_options; } else { maxseg = BBR_MIN_SEG - bbr->rc_last_options; } #ifdef KERN_TLS if (bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) { tls_seg = ctf_get_opt_tls_size(bbr->rc_inp->inp_socket, bbr->rc_tp->snd_wnd); bbr->r_ctl.rc_pace_min_segs = (tls_seg + bbr->rc_last_options); } #endif old_tso = bbr->r_ctl.rc_pace_max_segs; if (bbr->rc_past_init_win == 0) { /* * Not enough data has been acknowledged to make a * judgement unless we are hardware TLS. Set up * the initial TSO based on if we are sending a * full IW at once or not. */ if (bbr->rc_use_google) bbr->r_ctl.rc_pace_max_segs = ((bbr->rc_tp->t_maxseg - bbr->rc_last_options) * 2); else if (bbr->bbr_init_win_cheat) bbr->r_ctl.rc_pace_max_segs = bbr_initial_cwnd(bbr, bbr->rc_tp); else bbr->r_ctl.rc_pace_max_segs = bbr->rc_tp->t_maxseg - bbr->rc_last_options; if (bbr->r_ctl.rc_pace_min_segs != bbr->rc_tp->t_maxseg) bbr->r_ctl.rc_pace_min_segs = bbr->rc_tp->t_maxseg; #ifdef KERN_TLS if ((bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) && tls_seg) { /* * For hardware TLS we set our min to the tls_seg size. */ bbr->r_ctl.rc_pace_max_segs = tls_seg; bbr->r_ctl.rc_pace_min_segs = tls_seg + bbr->rc_last_options; } #endif if (bbr->r_ctl.rc_pace_max_segs == 0) { bbr->r_ctl.rc_pace_max_segs = maxseg; } bbr_log_type_tsosize(bbr, cts, bbr->r_ctl.rc_pace_max_segs, tls_seg, old_tso, maxseg, 0); #ifdef KERN_TLS if ((bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) == 0) #endif bbr_adjust_for_hw_pacing(bbr, cts); return; } /** * Now lets set the TSO goal based on our delivery rate in * bytes per second. Note we only do this if * we have acked at least the initial cwnd worth of data. */ bw = bbr_get_bw(bbr); if (IN_RECOVERY(bbr->rc_tp->t_flags) && (bbr->rc_use_google == 0)) { /* We clamp to one MSS in recovery */ new_tso = maxseg; } else if (bbr->rc_use_google) { int min_tso_segs; /* Google considers the gain too */ if (bbr->r_ctl.rc_bbr_hptsi_gain != BBR_UNIT) { bw *= bbr->r_ctl.rc_bbr_hptsi_gain; bw /= BBR_UNIT; } bytes = bw / 1024; if (bytes > (64 * 1024)) bytes = 64 * 1024; new_tso = bytes / maxseg; if (bw < ONE_POINT_TWO_MEG) min_tso_segs = 1; else min_tso_segs = 2; if (new_tso < min_tso_segs) new_tso = min_tso_segs; new_tso *= maxseg; } else if (bbr->rc_no_pacing) { new_tso = (PACE_MAX_IP_BYTES / maxseg) * maxseg; } else if (bw <= bbr->r_ctl.bbr_cross_over) { /* * Calculate the worse case b/w TSO if we are inserting no * more than a delay_target number of TSO's. */ uint32_t tso_len, min_tso; tso_len = bbr_get_pacing_length(bbr, BBR_UNIT, bbr->r_ctl.bbr_hptsi_segments_delay_tar, bw); if (tso_len > maxseg) { new_tso = tso_len / maxseg; if (new_tso > bbr->r_ctl.bbr_hptsi_segments_max) new_tso = bbr->r_ctl.bbr_hptsi_segments_max; new_tso *= maxseg; } else { /* * less than a full sized frame yikes.. long rtt or * low bw? */ min_tso = bbr_minseg(bbr); if ((tso_len > min_tso) && (bbr_all_get_min == 0)) new_tso = rounddown(tso_len, min_tso); else new_tso = min_tso; } } else if (bw > FIVETWELVE_MBPS) { /* * This guy is so fast b/w wise that we can TSO as large as * possible of segments that the NIC will allow. */ new_tso = rounddown(PACE_MAX_IP_BYTES, maxseg); } else { /* * This formula is based on attempting to send a segment or * more every bbr_hptsi_per_second. The default is 1000 * which means you are targeting what you can send every 1ms * based on the peers bw. * * If the number drops to say 500, then you are looking more * at 2ms and you will raise how much we send in a single * TSO thus saving CPU (less bbr_output_wtime() calls). The * trade off of course is you will send more at once and * thus tend to clump up the sends into larger "bursts" * building a queue. */ bw /= bbr->r_ctl.bbr_hptsi_per_second; new_tso = roundup(bw, (uint64_t)maxseg); /* * Gate the floor to match what our lower than 48Mbps * algorithm does. The ceiling (bbr_hptsi_segments_max) thus * becomes the floor for this calculation. */ if (new_tso < (bbr->r_ctl.bbr_hptsi_segments_max * maxseg)) new_tso = (bbr->r_ctl.bbr_hptsi_segments_max * maxseg); } if (bbr->r_ctl.bbr_hptsi_segments_floor && (new_tso < (maxseg * bbr->r_ctl.bbr_hptsi_segments_floor))) new_tso = maxseg * bbr->r_ctl.bbr_hptsi_segments_floor; if (new_tso > PACE_MAX_IP_BYTES) new_tso = rounddown(PACE_MAX_IP_BYTES, maxseg); /* Enforce an utter maximum if we are not HW-TLS */ #ifdef KERN_TLS if ((bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) == 0) #endif if (bbr->r_ctl.bbr_utter_max && (new_tso > (bbr->r_ctl.bbr_utter_max * maxseg))) { new_tso = bbr->r_ctl.bbr_utter_max * maxseg; } #ifdef KERN_TLS if (tls_seg) { /* * Lets move the output size * up to 1 or more TLS record sizes. */ uint32_t temp; temp = roundup(new_tso, tls_seg); new_tso = temp; /* Back down if needed to under a full frame */ while (new_tso > PACE_MAX_IP_BYTES) new_tso -= tls_seg; } #endif if (old_tso != new_tso) { /* Only log changes */ bbr_log_type_tsosize(bbr, cts, new_tso, tls_seg, old_tso, maxseg, 0); bbr->r_ctl.rc_pace_max_segs = new_tso; } #ifdef KERN_TLS if ((bbr->rc_inp->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) && tls_seg) { bbr->r_ctl.rc_pace_min_segs = tls_seg + bbr->rc_last_options; } else #endif /* We have hardware pacing and not hardware TLS! */ bbr_adjust_for_hw_pacing(bbr, cts); } static void bbr_log_output(struct tcp_bbr *bbr, struct tcpcb *tp, struct tcpopt *to, int32_t len, uint32_t seq_out, uint8_t th_flags, int32_t err, uint32_t cts, struct mbuf *mb, int32_t * abandon, struct bbr_sendmap *hintrsm, uint32_t delay_calc, struct sockbuf *sb) { struct bbr_sendmap *rsm, *nrsm; register uint32_t snd_max, snd_una; uint32_t pacing_time; /* * Add to the RACK log of packets in flight or retransmitted. If * there is a TS option we will use the TS echoed, if not we will * grab a TS. * * Retransmissions will increment the count and move the ts to its * proper place. Note that if options do not include TS's then we * won't be able to effectively use the ACK for an RTT on a retran. * * Notes about r_start and r_end. Lets consider a send starting at * sequence 1 for 10 bytes. In such an example the r_start would be * 1 (starting sequence) but the r_end would be r_start+len i.e. 11. * This means that r_end is actually the first sequence for the next * slot (11). * */ INP_WLOCK_ASSERT(tp->t_inpcb); if (err) { /* * We don't log errors -- we could but snd_max does not * advance in this case either. */ return; } if (th_flags & TH_RST) { /* * We don't log resets and we return immediately from * sending */ *abandon = 1; return; } snd_una = tp->snd_una; if (th_flags & (TH_SYN | TH_FIN) && (hintrsm == NULL)) { /* * The call to bbr_log_output is made before bumping * snd_max. This means we can record one extra byte on a SYN * or FIN if seq_out is adding more on and a FIN is present * (and we are not resending). */ if (th_flags & TH_SYN) len++; if (th_flags & TH_FIN) len++; } if (SEQ_LEQ((seq_out + len), snd_una)) { /* Are sending an old segment to induce an ack (keep-alive)? */ return; } if (SEQ_LT(seq_out, snd_una)) { /* huh? should we panic? */ uint32_t end; end = seq_out + len; seq_out = snd_una; len = end - seq_out; } snd_max = tp->snd_max; if (len == 0) { /* We don't log zero window probes */ return; } pacing_time = bbr_get_pacing_delay(bbr, bbr->r_ctl.rc_bbr_hptsi_gain, len, cts, 1); /* First question is it a retransmission? */ if (seq_out == snd_max) { again: rsm = bbr_alloc(bbr); if (rsm == NULL) { return; } rsm->r_flags = 0; if (th_flags & TH_SYN) rsm->r_flags |= BBR_HAS_SYN; if (th_flags & TH_FIN) rsm->r_flags |= BBR_HAS_FIN; rsm->r_tim_lastsent[0] = cts; rsm->r_rtr_cnt = 1; rsm->r_rtr_bytes = 0; rsm->r_start = seq_out; rsm->r_end = rsm->r_start + len; rsm->r_dupack = 0; rsm->r_delivered = bbr->r_ctl.rc_delivered; rsm->r_pacing_delay = pacing_time; rsm->r_ts_valid = bbr->rc_ts_valid; if (bbr->rc_ts_valid) rsm->r_del_ack_ts = bbr->r_ctl.last_inbound_ts; rsm->r_del_time = bbr->r_ctl.rc_del_time; if (bbr->r_ctl.r_app_limited_until) rsm->r_app_limited = 1; else rsm->r_app_limited = 0; rsm->r_first_sent_time = bbr_get_earliest_send_outstanding(bbr, rsm, cts); rsm->r_flight_at_send = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); /* * Here we must also add in this rsm since snd_max * is updated after we return from a new send. */ rsm->r_flight_at_send += len; TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_map, rsm, r_next); TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 1; if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) rsm->r_bbr_state = bbr_state_val(bbr); else rsm->r_bbr_state = 8; if (bbr->r_ctl.rc_bbr_hptsi_gain > BBR_UNIT) { rsm->r_is_gain = 1; rsm->r_is_drain = 0; } else if (bbr->r_ctl.rc_bbr_hptsi_gain < BBR_UNIT) { rsm->r_is_drain = 1; rsm->r_is_gain = 0; } else { rsm->r_is_drain = 0; rsm->r_is_gain = 0; } return; } /* * If we reach here its a retransmission and we need to find it. */ more: if (hintrsm && (hintrsm->r_start == seq_out)) { rsm = hintrsm; hintrsm = NULL; } else if (bbr->r_ctl.rc_next) { /* We have a hint from a previous run */ rsm = bbr->r_ctl.rc_next; } else { /* No hints sorry */ rsm = NULL; } if ((rsm) && (rsm->r_start == seq_out)) { /* * We used rc_next or hintrsm to retransmit, hopefully the * likely case. */ seq_out = bbr_update_entry(tp, bbr, rsm, cts, &len, pacing_time); if (len == 0) { return; } else { goto more; } } /* Ok it was not the last pointer go through it the hard way. */ TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_map, r_next) { if (rsm->r_start == seq_out) { seq_out = bbr_update_entry(tp, bbr, rsm, cts, &len, pacing_time); bbr->r_ctl.rc_next = TAILQ_NEXT(rsm, r_next); if (len == 0) { return; } else { continue; } } if (SEQ_GEQ(seq_out, rsm->r_start) && SEQ_LT(seq_out, rsm->r_end)) { /* Transmitted within this piece */ /* * Ok we must split off the front and then let the * update do the rest */ nrsm = bbr_alloc_full_limit(bbr); if (nrsm == NULL) { bbr_update_rsm(tp, bbr, rsm, cts, pacing_time); return; } /* * copy rsm to nrsm and then trim the front of rsm * to not include this part. */ bbr_clone_rsm(bbr, nrsm, rsm, seq_out); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } rsm->r_flags &= (~BBR_HAS_FIN); seq_out = bbr_update_entry(tp, bbr, nrsm, cts, &len, pacing_time); if (len == 0) { return; } } } /* * Hmm not found in map did they retransmit both old and on into the * new? */ if (seq_out == tp->snd_max) { goto again; } else if (SEQ_LT(seq_out, tp->snd_max)) { #ifdef BBR_INVARIANTS printf("seq_out:%u len:%d snd_una:%u snd_max:%u -- but rsm not found?\n", seq_out, len, tp->snd_una, tp->snd_max); printf("Starting Dump of all rack entries\n"); TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_map, r_next) { printf("rsm:%p start:%u end:%u\n", rsm, rsm->r_start, rsm->r_end); } printf("Dump complete\n"); panic("seq_out not found rack:%p tp:%p", bbr, tp); #endif } else { #ifdef BBR_INVARIANTS /* * Hmm beyond sndmax? (only if we are using the new rtt-pack * flag) */ panic("seq_out:%u(%d) is beyond snd_max:%u tp:%p", seq_out, len, tp->snd_max, tp); #endif } } static void bbr_collapse_rtt(struct tcpcb *tp, struct tcp_bbr *bbr, int32_t rtt) { /* * Collapse timeout back the cum-ack moved. */ tp->t_rxtshift = 0; tp->t_softerror = 0; } static void tcp_bbr_xmit_timer(struct tcp_bbr *bbr, uint32_t rtt_usecs, uint32_t rsm_send_time, uint32_t r_start, uint32_t tsin) { bbr->rtt_valid = 1; bbr->r_ctl.cur_rtt = rtt_usecs; bbr->r_ctl.ts_in = tsin; if (rsm_send_time) bbr->r_ctl.cur_rtt_send_time = rsm_send_time; } static void bbr_make_timestamp_determination(struct tcp_bbr *bbr) { /** * We have in our bbr control: * 1) The timestamp we started observing cum-acks (bbr->r_ctl.bbr_ts_check_tstmp). * 2) Our timestamp indicating when we sent that packet (bbr->r_ctl.rsm->bbr_ts_check_our_cts). * 3) The current timestamp that just came in (bbr->r_ctl.last_inbound_ts) * 4) The time that the packet that generated that ack was sent (bbr->r_ctl.cur_rtt_send_time) * * Now we can calculate the time between the sends by doing: * * delta = bbr->r_ctl.cur_rtt_send_time - bbr->r_ctl.bbr_ts_check_our_cts * * And the peer's time between receiving them by doing: * * peer_delta = bbr->r_ctl.last_inbound_ts - bbr->r_ctl.bbr_ts_check_tstmp * * We want to figure out if the timestamp values are in msec, 10msec or usec. * We also may find that we can't use the timestamps if say we see * that the peer_delta indicates that though we may have taken 10ms to * pace out the data, it only saw 1ms between the two packets. This would * indicate that somewhere on the path is a batching entity that is giving * out time-slices of the actual b/w. This would mean we could not use * reliably the peers timestamps. * * We expect delta > peer_delta initially. Until we figure out the * timestamp difference which we will store in bbr->r_ctl.bbr_peer_tsratio. * If we place 1000 there then its a ms vs our usec. If we place 10000 there * then its 10ms vs our usec. If the peer is running a usec clock we would * put a 1 there. If the value is faster then ours, we will disable the * use of timestamps (though we could revist this later if we find it to be not * just an isolated one or two flows)). * * To detect the batching middle boxes we will come up with our compensation and * if with it in place, we find the peer is drastically off (by some margin) in * the smaller direction, then we will assume the worst case and disable use of timestamps. * */ uint64_t delta, peer_delta, delta_up; delta = bbr->r_ctl.cur_rtt_send_time - bbr->r_ctl.bbr_ts_check_our_cts; if (delta < bbr_min_usec_delta) { /* * Have not seen a min amount of time * between our send times so we can * make a determination of the timestamp * yet. */ return; } peer_delta = bbr->r_ctl.last_inbound_ts - bbr->r_ctl.bbr_ts_check_tstmp; if (peer_delta < bbr_min_peer_delta) { /* * We may have enough in the form of * our delta but the peers number * has not changed that much. It could * be its clock ratio is such that * we need more data (10ms tick) or * there may be other compression scenarios * going on. In any event we need the * spread to be larger. */ return; } /* Ok lets first see which way our delta is going */ if (peer_delta > delta) { /* Very unlikely, the peer without * compensation shows that it saw * the two sends arrive further apart * then we saw then in micro-seconds. */ if (peer_delta < (delta + ((delta * (uint64_t)1000)/ (uint64_t)bbr_delta_percent))) { /* well it looks like the peer is a micro-second clock. */ bbr->rc_ts_clock_set = 1; bbr->r_ctl.bbr_peer_tsratio = 1; } else { bbr->rc_ts_cant_be_used = 1; bbr->rc_ts_clock_set = 1; } return; } /* Ok we know that the peer_delta is smaller than our send distance */ bbr->rc_ts_clock_set = 1; /* First question is it within the percentage that they are using usec time? */ delta_up = (peer_delta * 1000) / (uint64_t)bbr_delta_percent; if ((peer_delta + delta_up) >= delta) { /* Its a usec clock */ bbr->r_ctl.bbr_peer_tsratio = 1; bbr_log_tstmp_validation(bbr, peer_delta, delta); return; } /* Ok if not usec, what about 10usec (though unlikely)? */ delta_up = (peer_delta * 1000 * 10) / (uint64_t)bbr_delta_percent; if (((peer_delta * 10) + delta_up) >= delta) { bbr->r_ctl.bbr_peer_tsratio = 10; bbr_log_tstmp_validation(bbr, peer_delta, delta); return; } /* And what about 100usec (though again unlikely)? */ delta_up = (peer_delta * 1000 * 100) / (uint64_t)bbr_delta_percent; if (((peer_delta * 100) + delta_up) >= delta) { bbr->r_ctl.bbr_peer_tsratio = 100; bbr_log_tstmp_validation(bbr, peer_delta, delta); return; } /* And how about 1 msec (the most likely one)? */ delta_up = (peer_delta * 1000 * 1000) / (uint64_t)bbr_delta_percent; if (((peer_delta * 1000) + delta_up) >= delta) { bbr->r_ctl.bbr_peer_tsratio = 1000; bbr_log_tstmp_validation(bbr, peer_delta, delta); return; } /* Ok if not msec could it be 10 msec? */ delta_up = (peer_delta * 1000 * 10000) / (uint64_t)bbr_delta_percent; if (((peer_delta * 10000) + delta_up) >= delta) { bbr->r_ctl.bbr_peer_tsratio = 10000; return; } /* If we fall down here the clock tick so slowly we can't use it */ bbr->rc_ts_cant_be_used = 1; bbr->r_ctl.bbr_peer_tsratio = 0; bbr_log_tstmp_validation(bbr, peer_delta, delta); } /* * Collect new round-trip time estimate * and update averages and current timeout. */ static void tcp_bbr_xmit_timer_commit(struct tcp_bbr *bbr, struct tcpcb *tp, uint32_t cts) { int32_t delta; uint32_t rtt, tsin; int32_t rtt_ticks; if (bbr->rtt_valid == 0) /* No valid sample */ return; rtt = bbr->r_ctl.cur_rtt; tsin = bbr->r_ctl.ts_in; if (bbr->rc_prtt_set_ts) { /* * We are to force feed the rttProp filter due * to an entry into PROBE_RTT. This assures * that the times are sync'd between when we * go into PROBE_RTT and the filter expiration. * * Google does not use a true filter, so they do * this implicitly since they only keep one value * and when they enter probe-rtt they update the * value to the newest rtt. */ uint32_t rtt_prop; bbr->rc_prtt_set_ts = 0; rtt_prop = get_filter_value_small(&bbr->r_ctl.rc_rttprop); if (rtt > rtt_prop) filter_increase_by_small(&bbr->r_ctl.rc_rttprop, (rtt - rtt_prop), cts); else apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, cts); } if (bbr->rc_ack_was_delayed) rtt += bbr->r_ctl.rc_ack_hdwr_delay; if (rtt < bbr->r_ctl.rc_lowest_rtt) bbr->r_ctl.rc_lowest_rtt = rtt; bbr_log_rtt_sample(bbr, rtt, tsin); if (bbr->r_init_rtt) { /* * The initial rtt is not-trusted, nuke it and lets get * our first valid measurement in. */ bbr->r_init_rtt = 0; tp->t_srtt = 0; } if ((bbr->rc_ts_clock_set == 0) && bbr->rc_ts_valid) { /* * So we have not yet figured out * what the peers TSTMP value is * in (most likely ms). We need a * series of cum-ack's to determine * this reliably. */ if (bbr->rc_ack_is_cumack) { if (bbr->rc_ts_data_set) { /* Lets attempt to determine the timestamp granularity. */ bbr_make_timestamp_determination(bbr); } else { bbr->rc_ts_data_set = 1; bbr->r_ctl.bbr_ts_check_tstmp = bbr->r_ctl.last_inbound_ts; bbr->r_ctl.bbr_ts_check_our_cts = bbr->r_ctl.cur_rtt_send_time; } } else { /* * We have to have consecutive acks * reset any "filled" state to none. */ bbr->rc_ts_data_set = 0; } } /* Round it up */ rtt_ticks = USEC_2_TICKS((rtt + (USECS_IN_MSEC - 1))); if (rtt_ticks == 0) rtt_ticks = 1; if (tp->t_srtt != 0) { /* * srtt is stored as fixed point with 5 bits after the * binary point (i.e., scaled by 8). The following magic is * equivalent to the smoothing algorithm in rfc793 with an * alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed point). * Adjust rtt to origin 0. */ delta = ((rtt_ticks - 1) << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); tp->t_srtt += delta; if (tp->t_srtt <= 0) tp->t_srtt = 1; /* * We accumulate a smoothed rtt variance (actually, a * smoothed mean difference), then set the retransmit timer * to smoothed rtt + 4 times the smoothed variance. rttvar * is stored as fixed point with 4 bits after the binary * point (scaled by 16). The following is equivalent to * rfc793 smoothing with an alpha of .75 (rttvar = * rttvar*3/4 + |delta| / 4). This replaces rfc793's * wired-in beta. */ if (delta < 0) delta = -delta; delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); tp->t_rttvar += delta; if (tp->t_rttvar <= 0) tp->t_rttvar = 1; if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } else { /* * No rtt measurement yet - use the unsmoothed rtt. Set the * variance to half the rtt (so our first retransmit happens * at 3*rtt). */ tp->t_srtt = rtt_ticks << TCP_RTT_SHIFT; tp->t_rttvar = rtt_ticks << (TCP_RTTVAR_SHIFT - 1); tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } KMOD_TCPSTAT_INC(tcps_rttupdated); tp->t_rttupdated++; #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, imax(0, rtt_ticks)); #endif /* * the retransmit should happen at rtt + 4 * rttvar. Because of the * way we do the smoothing, srtt and rttvar will each average +1/2 * tick of bias. When we compute the retransmit timer, we want 1/2 * tick of rounding and 1 extra tick because of +-1/2 tick * uncertainty in the firing of the timer. The bias will give us * exactly the 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below the minimum * feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), max(MSEC_2_TICKS(bbr->r_ctl.rc_min_rto_ms), rtt_ticks + 2), MSEC_2_TICKS(((uint32_t)bbr->rc_max_rto_sec) * 1000)); /* * We received an ack for a packet that wasn't retransmitted; it is * probably safe to discard any error indications we've received * recently. This isn't quite right, but close enough for now (a * route might have failed after we sent a segment, and the return * path might not be symmetrical). */ tp->t_softerror = 0; rtt = (TICKS_2_USEC(bbr->rc_tp->t_srtt) >> TCP_RTT_SHIFT); if (bbr->r_ctl.bbr_smallest_srtt_this_state > rtt) bbr->r_ctl.bbr_smallest_srtt_this_state = rtt; } static void bbr_earlier_retran(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t t, uint32_t cts, int ack_type) { /* * For this RSM, we acknowledged the data from a previous * transmission, not the last one we made. This means we did a false * retransmit. */ if (rsm->r_flags & BBR_HAS_FIN) { /* * The sending of the FIN often is multiple sent when we * have everything outstanding ack'd. We ignore this case * since its over now. */ return; } if (rsm->r_flags & BBR_TLP) { /* * We expect TLP's to have this occur often */ bbr->rc_tlp_rtx_out = 0; return; } if (ack_type != BBR_CUM_ACKED) { /* * If it was not a cum-ack we * don't really know for sure since * the timestamp could be from some * other transmission. */ return; } if (rsm->r_flags & BBR_WAS_SACKPASS) { /* * We retransmitted based on a sack and the earlier * retransmission ack'd it - re-ordering is occuring. */ BBR_STAT_INC(bbr_reorder_seen); bbr->r_ctl.rc_reorder_ts = cts; } /* Back down the loss count */ if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost -= rsm->r_end - rsm->r_start; bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; rsm->r_flags &= ~BBR_MARKED_LOST; if (SEQ_GT(bbr->r_ctl.rc_lt_lost, bbr->r_ctl.rc_lost)) /* LT sampling also needs adjustment */ bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } /***** RRS HERE ************************/ /* Do we need to do this??? */ /* bbr_reset_lt_bw_sampling(bbr, cts); */ /***** RRS HERE ************************/ BBR_STAT_INC(bbr_badfr); BBR_STAT_ADD(bbr_badfr_bytes, (rsm->r_end - rsm->r_start)); } static void bbr_set_reduced_rtt(struct tcp_bbr *bbr, uint32_t cts, uint32_t line) { bbr->r_ctl.rc_rtt_shrinks = cts; if (bbr_can_force_probertt && (TSTMP_GT(cts, bbr->r_ctl.last_in_probertt)) && ((cts - bbr->r_ctl.last_in_probertt) > bbr->r_ctl.rc_probertt_int)) { /* * We should enter probe-rtt its been too long * since we have been there. */ bbr_enter_probe_rtt(bbr, cts, __LINE__); } else bbr_check_probe_rtt_limits(bbr, cts); } static void tcp_bbr_commit_bw(struct tcp_bbr *bbr, uint32_t cts) { uint64_t orig_bw; if (bbr->r_ctl.rc_bbr_cur_del_rate == 0) { /* We never apply a zero measurment */ bbr_log_type_bbrupd(bbr, 20, cts, 0, 0, 0, 0, 0, 0, 0, 0); return; } if (bbr->r_ctl.r_measurement_count < 0xffffffff) bbr->r_ctl.r_measurement_count++; orig_bw = get_filter_value(&bbr->r_ctl.rc_delrate); apply_filter_max(&bbr->r_ctl.rc_delrate, bbr->r_ctl.rc_bbr_cur_del_rate, bbr->r_ctl.rc_pkt_epoch); bbr_log_type_bbrupd(bbr, 21, cts, (uint32_t)orig_bw, (uint32_t)get_filter_value(&bbr->r_ctl.rc_delrate), 0, 0, 0, 0, 0, 0); if (orig_bw && (orig_bw != get_filter_value(&bbr->r_ctl.rc_delrate))) { if (bbr->bbr_hdrw_pacing) { /* * Apply a new rate to the hardware * possibly. */ bbr_update_hardware_pacing_rate(bbr, cts); } bbr_set_state_target(bbr, __LINE__); tcp_bbr_tso_size_check(bbr, cts); if (bbr->r_recovery_bw) { bbr_setup_red_bw(bbr, cts); bbr_log_type_bw_reduce(bbr, BBR_RED_BW_USELRBW); } } else if ((orig_bw == 0) && get_filter_value(&bbr->r_ctl.rc_delrate)) tcp_bbr_tso_size_check(bbr, cts); } static void bbr_nf_measurement(struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t rtt, uint32_t cts) { if (bbr->rc_in_persist == 0) { /* We log only when not in persist */ /* Translate to a Bytes Per Second */ uint64_t tim, bw, ts_diff, ts_bw; uint32_t upper, lower, delivered; if (TSTMP_GT(bbr->r_ctl.rc_del_time, rsm->r_del_time)) tim = (uint64_t)(bbr->r_ctl.rc_del_time - rsm->r_del_time); else tim = 1; /* * Now that we have processed the tim (skipping the sample * or possibly updating the time, go ahead and * calculate the cdr. */ delivered = (bbr->r_ctl.rc_delivered - rsm->r_delivered); bw = (uint64_t)delivered; bw *= (uint64_t)USECS_IN_SECOND; bw /= tim; if (bw == 0) { /* We must have a calculatable amount */ return; } upper = (bw >> 32) & 0x00000000ffffffff; lower = bw & 0x00000000ffffffff; /* * If we are using this b/w shove it in now so we * can see in the trace viewer if it gets over-ridden. */ if (rsm->r_ts_valid && bbr->rc_ts_valid && bbr->rc_ts_clock_set && (bbr->rc_ts_cant_be_used == 0) && bbr->rc_use_ts_limit) { ts_diff = max((bbr->r_ctl.last_inbound_ts - rsm->r_del_ack_ts), 1); ts_diff *= bbr->r_ctl.bbr_peer_tsratio; if ((delivered == 0) || (rtt < 1000)) { /* Can't use the ts */ bbr_log_type_bbrupd(bbr, 61, cts, ts_diff, bbr->r_ctl.last_inbound_ts, rsm->r_del_ack_ts, 0, 0, 0, 0, delivered); } else { ts_bw = (uint64_t)delivered; ts_bw *= (uint64_t)USECS_IN_SECOND; ts_bw /= ts_diff; bbr_log_type_bbrupd(bbr, 62, cts, (ts_bw >> 32), (ts_bw & 0xffffffff), 0, 0, 0, 0, ts_diff, delivered); if ((bbr->ts_can_raise) && (ts_bw > bw)) { bbr_log_type_bbrupd(bbr, 8, cts, delivered, ts_diff, (bw >> 32), (bw & 0x00000000ffffffff), 0, 0, 0, 0); bw = ts_bw; } else if (ts_bw && (ts_bw < bw)) { bbr_log_type_bbrupd(bbr, 7, cts, delivered, ts_diff, (bw >> 32), (bw & 0x00000000ffffffff), 0, 0, 0, 0); bw = ts_bw; } } } if (rsm->r_first_sent_time && TSTMP_GT(rsm->r_tim_lastsent[(rsm->r_rtr_cnt -1)],rsm->r_first_sent_time)) { uint64_t sbw, sti; /* * We use what was in flight at the time of our * send and the size of this send to figure * out what we have been sending at (amount). * For the time we take from the time of * the send of the first send outstanding * until this send plus this sends pacing * time. This gives us a good calculation * as to the rate we have been sending at. */ sbw = (uint64_t)(rsm->r_flight_at_send); sbw *= (uint64_t)USECS_IN_SECOND; sti = rsm->r_tim_lastsent[(rsm->r_rtr_cnt -1)] - rsm->r_first_sent_time; sti += rsm->r_pacing_delay; sbw /= sti; if (sbw < bw) { bbr_log_type_bbrupd(bbr, 6, cts, delivered, (uint32_t)sti, (bw >> 32), (uint32_t)bw, rsm->r_first_sent_time, 0, (sbw >> 32), (uint32_t)sbw); bw = sbw; } } /* Use the google algorithm for b/w measurements */ bbr->r_ctl.rc_bbr_cur_del_rate = bw; if ((rsm->r_app_limited == 0) || (bw > get_filter_value(&bbr->r_ctl.rc_delrate))) { tcp_bbr_commit_bw(bbr, cts); bbr_log_type_bbrupd(bbr, 10, cts, (uint32_t)tim, delivered, 0, 0, 0, 0, bbr->r_ctl.rc_del_time, rsm->r_del_time); } } } static void bbr_google_measurement(struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t rtt, uint32_t cts) { if (bbr->rc_in_persist == 0) { /* We log only when not in persist */ /* Translate to a Bytes Per Second */ uint64_t tim, bw; uint32_t upper, lower, delivered; int no_apply = 0; if (TSTMP_GT(bbr->r_ctl.rc_del_time, rsm->r_del_time)) tim = (uint64_t)(bbr->r_ctl.rc_del_time - rsm->r_del_time); else tim = 1; /* * Now that we have processed the tim (skipping the sample * or possibly updating the time, go ahead and * calculate the cdr. */ delivered = (bbr->r_ctl.rc_delivered - rsm->r_delivered); bw = (uint64_t)delivered; bw *= (uint64_t)USECS_IN_SECOND; bw /= tim; if (tim < bbr->r_ctl.rc_lowest_rtt) { bbr_log_type_bbrupd(bbr, 99, cts, (uint32_t)tim, delivered, tim, bbr->r_ctl.rc_lowest_rtt, 0, 0, 0, 0); no_apply = 1; } upper = (bw >> 32) & 0x00000000ffffffff; lower = bw & 0x00000000ffffffff; /* * If we are using this b/w shove it in now so we * can see in the trace viewer if it gets over-ridden. */ bbr->r_ctl.rc_bbr_cur_del_rate = bw; /* Gate by the sending rate */ if (rsm->r_first_sent_time && TSTMP_GT(rsm->r_tim_lastsent[(rsm->r_rtr_cnt -1)],rsm->r_first_sent_time)) { uint64_t sbw, sti; /* * We use what was in flight at the time of our * send and the size of this send to figure * out what we have been sending at (amount). * For the time we take from the time of * the send of the first send outstanding * until this send plus this sends pacing * time. This gives us a good calculation * as to the rate we have been sending at. */ sbw = (uint64_t)(rsm->r_flight_at_send); sbw *= (uint64_t)USECS_IN_SECOND; sti = rsm->r_tim_lastsent[(rsm->r_rtr_cnt -1)] - rsm->r_first_sent_time; sti += rsm->r_pacing_delay; sbw /= sti; if (sbw < bw) { bbr_log_type_bbrupd(bbr, 6, cts, delivered, (uint32_t)sti, (bw >> 32), (uint32_t)bw, rsm->r_first_sent_time, 0, (sbw >> 32), (uint32_t)sbw); bw = sbw; } if ((sti > tim) && (sti < bbr->r_ctl.rc_lowest_rtt)) { bbr_log_type_bbrupd(bbr, 99, cts, (uint32_t)tim, delivered, (uint32_t)sti, bbr->r_ctl.rc_lowest_rtt, 0, 0, 0, 0); no_apply = 1; } else no_apply = 0; } bbr->r_ctl.rc_bbr_cur_del_rate = bw; if ((no_apply == 0) && ((rsm->r_app_limited == 0) || (bw > get_filter_value(&bbr->r_ctl.rc_delrate)))) { tcp_bbr_commit_bw(bbr, cts); bbr_log_type_bbrupd(bbr, 10, cts, (uint32_t)tim, delivered, 0, 0, 0, 0, bbr->r_ctl.rc_del_time, rsm->r_del_time); } } } static void bbr_update_bbr_info(struct tcp_bbr *bbr, struct bbr_sendmap *rsm, uint32_t rtt, uint32_t cts, uint32_t tsin, uint32_t uts, int32_t match, uint32_t rsm_send_time, int32_t ack_type, struct tcpopt *to) { uint64_t old_rttprop; /* Update our delivery time and amount */ bbr->r_ctl.rc_delivered += (rsm->r_end - rsm->r_start); bbr->r_ctl.rc_del_time = cts; if (rtt == 0) { /* * 0 means its a retransmit, for now we don't use these for * the rest of BBR. */ return; } if ((bbr->rc_use_google == 0) && (match != BBR_RTT_BY_EXACTMATCH) && (match != BBR_RTT_BY_TIMESTAMP)){ /* * We get a lot of rtt updates, lets not pay attention to * any that are not an exact match. That way we don't have * to worry about timestamps and the whole nonsense of * unsure if its a retransmission etc (if we ever had the * timestamp fixed to always have the last thing sent this * would not be a issue). */ return; } if ((bbr_no_retran && bbr->rc_use_google) && (match != BBR_RTT_BY_EXACTMATCH) && (match != BBR_RTT_BY_TIMESTAMP)){ /* * We only do measurements in google mode * with bbr_no_retran on for sure things. */ return; } /* Only update srtt if we know by exact match */ tcp_bbr_xmit_timer(bbr, rtt, rsm_send_time, rsm->r_start, tsin); if (ack_type == BBR_CUM_ACKED) bbr->rc_ack_is_cumack = 1; else bbr->rc_ack_is_cumack = 0; old_rttprop = bbr_get_rtt(bbr, BBR_RTT_PROP); /* * Note the following code differs to the original * BBR spec. It calls for <= not <. However after a * long discussion in email with Neal, he acknowledged * that it should be < than so that we will have flows * going into probe-rtt (we were seeing cases where that * did not happen and caused ugly things to occur). We * have added this agreed upon fix to our code base. */ if (rtt < old_rttprop) { /* Update when we last saw a rtt drop */ bbr_log_rtt_shrinks(bbr, cts, 0, rtt, __LINE__, BBR_RTTS_NEWRTT, 0); bbr_set_reduced_rtt(bbr, cts, __LINE__); } bbr_log_type_bbrrttprop(bbr, rtt, (rsm ? rsm->r_end : 0), uts, cts, match, rsm->r_start, rsm->r_flags); apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, cts); if (old_rttprop != bbr_get_rtt(bbr, BBR_RTT_PROP)) { /* * The RTT-prop moved, reset the target (may be a * nop for some states). */ bbr_set_state_target(bbr, __LINE__); if (bbr->rc_bbr_state == BBR_STATE_PROBE_RTT) bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_NEW_TARGET, 0); else if (old_rttprop < bbr_get_rtt(bbr, BBR_RTT_PROP)) /* It went up */ bbr_check_probe_rtt_limits(bbr, cts); } if ((bbr->rc_use_google == 0) && (match == BBR_RTT_BY_TIMESTAMP)) { /* * We don't do b/w update with * these since they are not really * reliable. */ return; } if (bbr->r_ctl.r_app_limited_until && (bbr->r_ctl.rc_delivered >= bbr->r_ctl.r_app_limited_until)) { /* We are no longer app-limited */ bbr->r_ctl.r_app_limited_until = 0; } if (bbr->rc_use_google) { bbr_google_measurement(bbr, rsm, rtt, cts); } else { bbr_nf_measurement(bbr, rsm, rtt, cts); } } /* * Convert a timestamp that the main stack * uses (milliseconds) into one that bbr uses * (microseconds). Return that converted timestamp. */ static uint32_t bbr_ts_convert(uint32_t cts) { uint32_t sec, msec; sec = cts / MS_IN_USEC; msec = cts - (MS_IN_USEC * sec); return ((sec * USECS_IN_SECOND) + (msec * MS_IN_USEC)); } /* * Return 0 if we did not update the RTT time, return * 1 if we did. */ static int bbr_update_rtt(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, struct tcpopt *to, uint32_t cts, int32_t ack_type, uint32_t th_ack) { int32_t i; uint32_t t, uts = 0; if ((rsm->r_flags & BBR_ACKED) || (rsm->r_flags & BBR_WAS_RENEGED) || (rsm->r_flags & BBR_RXT_CLEARED)) { /* Already done */ return (0); } if (rsm->r_rtr_cnt == 1) { /* * Only one transmit. Hopefully the normal case. */ if (TSTMP_GT(cts, rsm->r_tim_lastsent[0])) t = cts - rsm->r_tim_lastsent[0]; else t = 1; if ((int)t <= 0) t = 1; bbr->r_ctl.rc_last_rtt = t; bbr_update_bbr_info(bbr, rsm, t, cts, to->to_tsecr, 0, BBR_RTT_BY_EXACTMATCH, rsm->r_tim_lastsent[0], ack_type, to); return (1); } /* Convert to usecs */ if ((bbr_can_use_ts_for_rtt == 1) && (bbr->rc_use_google == 1) && (ack_type == BBR_CUM_ACKED) && (to->to_flags & TOF_TS) && (to->to_tsecr != 0)) { t = tcp_tv_to_mssectick(&bbr->rc_tv) - to->to_tsecr; if (t < 1) t = 1; t *= MS_IN_USEC; bbr_update_bbr_info(bbr, rsm, t, cts, to->to_tsecr, 0, BBR_RTT_BY_TIMESTAMP, rsm->r_tim_lastsent[(rsm->r_rtr_cnt-1)], ack_type, to); return (1); } uts = bbr_ts_convert(to->to_tsecr); if ((to->to_flags & TOF_TS) && (to->to_tsecr != 0) && (ack_type == BBR_CUM_ACKED) && ((rsm->r_flags & BBR_OVERMAX) == 0)) { /* * Now which timestamp does it match? In this block the ACK * may be coming from a previous transmission. */ uint32_t fudge; fudge = BBR_TIMER_FUDGE; for (i = 0; i < rsm->r_rtr_cnt; i++) { if ((SEQ_GEQ(uts, (rsm->r_tim_lastsent[i] - fudge))) && (SEQ_LEQ(uts, (rsm->r_tim_lastsent[i] + fudge)))) { if (TSTMP_GT(cts, rsm->r_tim_lastsent[i])) t = cts - rsm->r_tim_lastsent[i]; else t = 1; if ((int)t <= 0) t = 1; bbr->r_ctl.rc_last_rtt = t; bbr_update_bbr_info(bbr, rsm, t, cts, to->to_tsecr, uts, BBR_RTT_BY_TSMATCHING, rsm->r_tim_lastsent[i], ack_type, to); if ((i + 1) < rsm->r_rtr_cnt) { /* Likely */ bbr_earlier_retran(tp, bbr, rsm, t, cts, ack_type); } else if (rsm->r_flags & BBR_TLP) { bbr->rc_tlp_rtx_out = 0; } return (1); } } /* Fall through if we can't find a matching timestamp */ } /* * Ok its a SACK block that we retransmitted. or a windows * machine without timestamps. We can tell nothing from the * time-stamp since its not there or the time the peer last * recieved a segment that moved forward its cum-ack point. * * Lets look at the last retransmit and see what we can tell * (with BBR for space we only keep 2 note we have to keep * at least 2 so the map can not be condensed more). */ i = rsm->r_rtr_cnt - 1; if (TSTMP_GT(cts, rsm->r_tim_lastsent[i])) t = cts - rsm->r_tim_lastsent[i]; else goto not_sure; if (t < bbr->r_ctl.rc_lowest_rtt) { /* * We retransmitted and the ack came back in less * than the smallest rtt we have observed in the * windowed rtt. We most likey did an improper * retransmit as outlined in 4.2 Step 3 point 2 in * the rack-draft. * * Use the prior transmission to update all the * information as long as there is only one prior * transmission. */ if ((rsm->r_flags & BBR_OVERMAX) == 0) { #ifdef BBR_INVARIANTS if (rsm->r_rtr_cnt == 1) panic("rsm:%p bbr:%p rsm has overmax and only 1 retranmit flags:%x?", rsm, bbr, rsm->r_flags); #endif i = rsm->r_rtr_cnt - 2; if (TSTMP_GT(cts, rsm->r_tim_lastsent[i])) t = cts - rsm->r_tim_lastsent[i]; else t = 1; bbr_update_bbr_info(bbr, rsm, t, cts, to->to_tsecr, uts, BBR_RTT_BY_EARLIER_RET, rsm->r_tim_lastsent[i], ack_type, to); bbr_earlier_retran(tp, bbr, rsm, t, cts, ack_type); } else { /* * Too many prior transmissions, just * updated BBR delivered */ not_sure: bbr_update_bbr_info(bbr, rsm, 0, cts, to->to_tsecr, uts, BBR_RTT_BY_SOME_RETRAN, 0, ack_type, to); } } else { /* * We retransmitted it and the retransmit did the * job. */ if (rsm->r_flags & BBR_TLP) bbr->rc_tlp_rtx_out = 0; if ((rsm->r_flags & BBR_OVERMAX) == 0) bbr_update_bbr_info(bbr, rsm, t, cts, to->to_tsecr, uts, BBR_RTT_BY_THIS_RETRAN, 0, ack_type, to); else bbr_update_bbr_info(bbr, rsm, 0, cts, to->to_tsecr, uts, BBR_RTT_BY_SOME_RETRAN, 0, ack_type, to); return (1); } return (0); } /* * Mark the SACK_PASSED flag on all entries prior to rsm send wise. */ static void bbr_log_sack_passed(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm) { struct bbr_sendmap *nrsm; nrsm = rsm; TAILQ_FOREACH_REVERSE_FROM(nrsm, &bbr->r_ctl.rc_tmap, bbr_head, r_tnext) { if (nrsm == rsm) { /* Skip orginal segment he is acked */ continue; } if (nrsm->r_flags & BBR_ACKED) { /* Skip ack'd segments */ continue; } if (nrsm->r_flags & BBR_SACK_PASSED) { /* * We found one that is already marked * passed, we have been here before and * so all others below this are marked. */ break; } BBR_STAT_INC(bbr_sack_passed); nrsm->r_flags |= BBR_SACK_PASSED; if (((nrsm->r_flags & BBR_MARKED_LOST) == 0) && bbr_is_lost(bbr, nrsm, bbr->r_ctl.rc_rcvtime)) { bbr->r_ctl.rc_lost += nrsm->r_end - nrsm->r_start; bbr->r_ctl.rc_lost_bytes += nrsm->r_end - nrsm->r_start; nrsm->r_flags |= BBR_MARKED_LOST; } nrsm->r_flags &= ~BBR_WAS_SACKPASS; } } /* * Returns the number of bytes that were * newly ack'd by sack blocks. */ static uint32_t bbr_proc_sack_blk(struct tcpcb *tp, struct tcp_bbr *bbr, struct sackblk *sack, struct tcpopt *to, struct bbr_sendmap **prsm, uint32_t cts) { int32_t times = 0; uint32_t start, end, maxseg, changed = 0; struct bbr_sendmap *rsm, *nrsm; int32_t used_ref = 1; uint8_t went_back = 0, went_fwd = 0; maxseg = tp->t_maxseg - bbr->rc_last_options; start = sack->start; end = sack->end; rsm = *prsm; if (rsm == NULL) used_ref = 0; /* Do we locate the block behind where we last were? */ if (rsm && SEQ_LT(start, rsm->r_start)) { went_back = 1; TAILQ_FOREACH_REVERSE_FROM(rsm, &bbr->r_ctl.rc_map, bbr_head, r_next) { if (SEQ_GEQ(start, rsm->r_start) && SEQ_LT(start, rsm->r_end)) { goto do_rest_ofb; } } } start_at_beginning: went_fwd = 1; /* * Ok lets locate the block where this guy is fwd from rsm (if its * set) */ TAILQ_FOREACH_FROM(rsm, &bbr->r_ctl.rc_map, r_next) { if (SEQ_GEQ(start, rsm->r_start) && SEQ_LT(start, rsm->r_end)) { break; } } do_rest_ofb: if (rsm == NULL) { /* * This happens when we get duplicate sack blocks with the * same end. For example SACK 4: 100 SACK 3: 100 The sort * will not change there location so we would just start at * the end of the first one and get lost. */ if (tp->t_flags & TF_SENTFIN) { /* * Check to see if we have not logged the FIN that * went out. */ nrsm = TAILQ_LAST_FAST(&bbr->r_ctl.rc_map, bbr_sendmap, r_next); if (nrsm && (nrsm->r_end + 1) == tp->snd_max) { /* * Ok we did not get the FIN logged. */ nrsm->r_end++; rsm = nrsm; goto do_rest_ofb; } } if (times == 1) { #ifdef BBR_INVARIANTS panic("tp:%p bbr:%p sack:%p to:%p prsm:%p", tp, bbr, sack, to, prsm); #else goto out; #endif } times++; BBR_STAT_INC(bbr_sack_proc_restart); rsm = NULL; goto start_at_beginning; } /* Ok we have an ACK for some piece of rsm */ if (rsm->r_start != start) { /* * Need to split this in two pieces the before and after. */ if (bbr_sack_mergable(rsm, start, end)) nrsm = bbr_alloc_full_limit(bbr); else nrsm = bbr_alloc_limit(bbr, BBR_LIMIT_TYPE_SPLIT); if (nrsm == NULL) { /* We could not allocate ignore the sack */ struct sackblk blk; blk.start = start; blk.end = end; sack_filter_reject(&bbr->r_ctl.bbr_sf, &blk); goto out; } bbr_clone_rsm(bbr, nrsm, rsm, start); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } rsm->r_flags &= (~BBR_HAS_FIN); rsm = nrsm; } if (SEQ_GEQ(end, rsm->r_end)) { /* * The end of this block is either beyond this guy or right * at this guy. */ if ((rsm->r_flags & BBR_ACKED) == 0) { bbr_update_rtt(tp, bbr, rsm, to, cts, BBR_SACKED, 0); changed += (rsm->r_end - rsm->r_start); bbr->r_ctl.rc_sacked += (rsm->r_end - rsm->r_start); bbr_log_sack_passed(tp, bbr, rsm); if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; } /* Is Reordering occuring? */ if (rsm->r_flags & BBR_SACK_PASSED) { BBR_STAT_INC(bbr_reorder_seen); bbr->r_ctl.rc_reorder_ts = cts; if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost -= rsm->r_end - rsm->r_start; if (SEQ_GT(bbr->r_ctl.rc_lt_lost, bbr->r_ctl.rc_lost)) /* LT sampling also needs adjustment */ bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } } rsm->r_flags |= BBR_ACKED; rsm->r_flags &= ~(BBR_TLP|BBR_WAS_RENEGED|BBR_RXT_CLEARED|BBR_MARKED_LOST); if (rsm->r_in_tmap) { TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 0; } } bbr_isit_a_pkt_epoch(bbr, cts, rsm, __LINE__, BBR_SACKED); if (end == rsm->r_end) { /* This block only - done */ goto out; } /* There is more not coverend by this rsm move on */ start = rsm->r_end; nrsm = TAILQ_NEXT(rsm, r_next); rsm = nrsm; times = 0; goto do_rest_ofb; } if (rsm->r_flags & BBR_ACKED) { /* Been here done that */ goto out; } /* Ok we need to split off this one at the tail */ if (bbr_sack_mergable(rsm, start, end)) nrsm = bbr_alloc_full_limit(bbr); else nrsm = bbr_alloc_limit(bbr, BBR_LIMIT_TYPE_SPLIT); if (nrsm == NULL) { /* failed XXXrrs what can we do but loose the sack info? */ struct sackblk blk; blk.start = start; blk.end = end; sack_filter_reject(&bbr->r_ctl.bbr_sf, &blk); goto out; } /* Clone it */ bbr_clone_rsm(bbr, nrsm, rsm, end); /* The sack block does not cover this guy fully */ rsm->r_flags &= (~BBR_HAS_FIN); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } nrsm->r_dupack = 0; bbr_update_rtt(tp, bbr, rsm, to, cts, BBR_SACKED, 0); bbr_isit_a_pkt_epoch(bbr, cts, rsm, __LINE__, BBR_SACKED); changed += (rsm->r_end - rsm->r_start); bbr->r_ctl.rc_sacked += (rsm->r_end - rsm->r_start); bbr_log_sack_passed(tp, bbr, rsm); /* Is Reordering occuring? */ if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; } if (rsm->r_flags & BBR_SACK_PASSED) { BBR_STAT_INC(bbr_reorder_seen); bbr->r_ctl.rc_reorder_ts = cts; if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost -= rsm->r_end - rsm->r_start; if (SEQ_GT(bbr->r_ctl.rc_lt_lost, bbr->r_ctl.rc_lost)) /* LT sampling also needs adjustment */ bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } } rsm->r_flags &= ~(BBR_TLP|BBR_WAS_RENEGED|BBR_RXT_CLEARED|BBR_MARKED_LOST); rsm->r_flags |= BBR_ACKED; if (rsm->r_in_tmap) { TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 0; } out: if (rsm && (rsm->r_flags & BBR_ACKED)) { /* * Now can we merge this newly acked * block with either the previous or * next block? */ nrsm = TAILQ_NEXT(rsm, r_next); if (nrsm && (nrsm->r_flags & BBR_ACKED)) { /* yep this and next can be merged */ rsm = bbr_merge_rsm(bbr, rsm, nrsm); } /* Now what about the previous? */ nrsm = TAILQ_PREV(rsm, bbr_head, r_next); if (nrsm && (nrsm->r_flags & BBR_ACKED)) { /* yep the previous and this can be merged */ rsm = bbr_merge_rsm(bbr, nrsm, rsm); } } if (used_ref == 0) { BBR_STAT_INC(bbr_sack_proc_all); } else { BBR_STAT_INC(bbr_sack_proc_short); } if (went_fwd && went_back) { BBR_STAT_INC(bbr_sack_search_both); } else if (went_fwd) { BBR_STAT_INC(bbr_sack_search_fwd); } else if (went_back) { BBR_STAT_INC(bbr_sack_search_back); } /* Save off where the next seq is */ if (rsm) bbr->r_ctl.rc_sacklast = TAILQ_NEXT(rsm, r_next); else bbr->r_ctl.rc_sacklast = NULL; *prsm = rsm; return (changed); } static void inline bbr_peer_reneges(struct tcp_bbr *bbr, struct bbr_sendmap *rsm, tcp_seq th_ack) { struct bbr_sendmap *tmap; BBR_STAT_INC(bbr_reneges_seen); tmap = NULL; while (rsm && (rsm->r_flags & BBR_ACKED)) { /* Its no longer sacked, mark it so */ uint32_t oflags; bbr->r_ctl.rc_sacked -= (rsm->r_end - rsm->r_start); #ifdef BBR_INVARIANTS if (rsm->r_in_tmap) { panic("bbr:%p rsm:%p flags:0x%x in tmap?", bbr, rsm, rsm->r_flags); } #endif oflags = rsm->r_flags; if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost -= rsm->r_end - rsm->r_start; bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; if (SEQ_GT(bbr->r_ctl.rc_lt_lost, bbr->r_ctl.rc_lost)) /* LT sampling also needs adjustment */ bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } rsm->r_flags &= ~(BBR_ACKED | BBR_SACK_PASSED | BBR_WAS_SACKPASS | BBR_MARKED_LOST); rsm->r_flags |= BBR_WAS_RENEGED; rsm->r_flags |= BBR_RXT_CLEARED; bbr_log_type_rsmclear(bbr, bbr->r_ctl.rc_rcvtime, rsm, oflags, __LINE__); /* Rebuild it into our tmap */ if (tmap == NULL) { TAILQ_INSERT_HEAD(&bbr->r_ctl.rc_tmap, rsm, r_tnext); tmap = rsm; } else { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, tmap, rsm, r_tnext); tmap = rsm; } tmap->r_in_tmap = 1; /* * XXXrrs Delivered? Should we do anything here? * * Of course we don't on a rxt timeout so maybe its ok that * we don't? * * For now lets not. */ rsm = TAILQ_NEXT(rsm, r_next); } /* * Now lets possibly clear the sack filter so we start recognizing * sacks that cover this area. */ sack_filter_clear(&bbr->r_ctl.bbr_sf, th_ack); } static void bbr_log_syn(struct tcpcb *tp, struct tcpopt *to) { struct tcp_bbr *bbr; struct bbr_sendmap *rsm; uint32_t cts; bbr = (struct tcp_bbr *)tp->t_fb_ptr; cts = bbr->r_ctl.rc_rcvtime; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); if (rsm && (rsm->r_flags & BBR_HAS_SYN)) { if ((rsm->r_end - rsm->r_start) <= 1) { /* Log out the SYN completely */ bbr->r_ctl.rc_holes_rxt -= rsm->r_rtr_bytes; rsm->r_rtr_bytes = 0; TAILQ_REMOVE(&bbr->r_ctl.rc_map, rsm, r_next); if (rsm->r_in_tmap) { TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 0; } if (bbr->r_ctl.rc_next == rsm) { /* scoot along the marker */ bbr->r_ctl.rc_next = TAILQ_FIRST(&bbr->r_ctl.rc_map); } if (to != NULL) bbr_update_rtt(tp, bbr, rsm, to, cts, BBR_CUM_ACKED, 0); bbr_free(bbr, rsm); } else { /* There is more (Fast open)? strip out SYN. */ rsm->r_flags &= ~BBR_HAS_SYN; rsm->r_start++; } } } /* * Returns the number of bytes that were * acknowledged by SACK blocks. */ static uint32_t bbr_log_ack(struct tcpcb *tp, struct tcpopt *to, struct tcphdr *th, uint32_t *prev_acked) { uint32_t changed, last_seq, entered_recovery = 0; struct tcp_bbr *bbr; struct bbr_sendmap *rsm; struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1]; register uint32_t th_ack; int32_t i, j, k, new_sb, num_sack_blks = 0; uint32_t cts, acked, ack_point, sack_changed = 0; uint32_t p_maxseg, maxseg, p_acked = 0; INP_WLOCK_ASSERT(tp->t_inpcb); if (th->th_flags & TH_RST) { /* We don't log resets */ return (0); } bbr = (struct tcp_bbr *)tp->t_fb_ptr; cts = bbr->r_ctl.rc_rcvtime; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); changed = 0; maxseg = tp->t_maxseg - bbr->rc_last_options; p_maxseg = min(bbr->r_ctl.rc_pace_max_segs, maxseg); th_ack = th->th_ack; if (SEQ_GT(th_ack, tp->snd_una)) { acked = th_ack - tp->snd_una; bbr_log_progress_event(bbr, tp, ticks, PROGRESS_UPDATE, __LINE__); bbr->rc_tp->t_acktime = ticks; } else acked = 0; if (SEQ_LEQ(th_ack, tp->snd_una)) { /* Only sent here for sack processing */ goto proc_sack; } if (rsm && SEQ_GT(th_ack, rsm->r_start)) { changed = th_ack - rsm->r_start; } else if ((rsm == NULL) && ((th_ack - 1) == tp->iss)) { /* * For the SYN incoming case we will not have called * tcp_output for the sending of the SYN, so there will be * no map. All other cases should probably be a panic. */ if ((to->to_flags & TOF_TS) && (to->to_tsecr != 0)) { /* * We have a timestamp that can be used to generate * an initial RTT. */ uint32_t ts, now, rtt; ts = bbr_ts_convert(to->to_tsecr); now = bbr_ts_convert(tcp_tv_to_mssectick(&bbr->rc_tv)); rtt = now - ts; if (rtt < 1) rtt = 1; bbr_log_type_bbrrttprop(bbr, rtt, tp->iss, 0, cts, BBR_RTT_BY_TIMESTAMP, tp->iss, 0); apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, cts); changed = 1; bbr->r_wanted_output = 1; goto out; } goto proc_sack; } else if (rsm == NULL) { goto out; } if (changed) { /* * The ACK point is advancing to th_ack, we must drop off * the packets in the rack log and calculate any eligble * RTT's. */ bbr->r_wanted_output = 1; more: if (rsm == NULL) { if (tp->t_flags & TF_SENTFIN) { /* if we send a FIN we will not hav a map */ goto proc_sack; } #ifdef BBR_INVARIANTS panic("No rack map tp:%p for th:%p state:%d bbr:%p snd_una:%u snd_max:%u chg:%d\n", tp, th, tp->t_state, bbr, tp->snd_una, tp->snd_max, changed); #endif goto proc_sack; } } if (SEQ_LT(th_ack, rsm->r_start)) { /* Huh map is missing this */ #ifdef BBR_INVARIANTS printf("Rack map starts at r_start:%u for th_ack:%u huh? ts:%d rs:%d bbr:%p\n", rsm->r_start, th_ack, tp->t_state, bbr->r_state, bbr); panic("th-ack is bad bbr:%p tp:%p", bbr, tp); #endif goto proc_sack; } else if (th_ack == rsm->r_start) { /* None here to ack */ goto proc_sack; } /* * Clear the dup ack counter, it will * either be freed or if there is some * remaining we need to start it at zero. */ rsm->r_dupack = 0; /* Now do we consume the whole thing? */ if (SEQ_GEQ(th_ack, rsm->r_end)) { /* Its all consumed. */ uint32_t left; if (rsm->r_flags & BBR_ACKED) { /* * It was acked on the scoreboard -- remove it from * total */ p_acked += (rsm->r_end - rsm->r_start); bbr->r_ctl.rc_sacked -= (rsm->r_end - rsm->r_start); if (bbr->r_ctl.rc_sacked == 0) bbr->r_ctl.rc_sacklast = NULL; } else { bbr_update_rtt(tp, bbr, rsm, to, cts, BBR_CUM_ACKED, th_ack); if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost_bytes -= rsm->r_end - rsm->r_start; } if (rsm->r_flags & BBR_SACK_PASSED) { /* * There are acked segments ACKED on the * scoreboard further up. We are seeing * reordering. */ BBR_STAT_INC(bbr_reorder_seen); bbr->r_ctl.rc_reorder_ts = cts; if (rsm->r_flags & BBR_MARKED_LOST) { bbr->r_ctl.rc_lost -= rsm->r_end - rsm->r_start; if (SEQ_GT(bbr->r_ctl.rc_lt_lost, bbr->r_ctl.rc_lost)) /* LT sampling also needs adjustment */ bbr->r_ctl.rc_lt_lost = bbr->r_ctl.rc_lost; } } rsm->r_flags &= ~BBR_MARKED_LOST; } bbr->r_ctl.rc_holes_rxt -= rsm->r_rtr_bytes; rsm->r_rtr_bytes = 0; TAILQ_REMOVE(&bbr->r_ctl.rc_map, rsm, r_next); if (rsm->r_in_tmap) { TAILQ_REMOVE(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 0; } if (bbr->r_ctl.rc_next == rsm) { /* scoot along the marker */ bbr->r_ctl.rc_next = TAILQ_FIRST(&bbr->r_ctl.rc_map); } bbr_isit_a_pkt_epoch(bbr, cts, rsm, __LINE__, BBR_CUM_ACKED); /* Adjust the packet counts */ left = th_ack - rsm->r_end; /* Free back to zone */ bbr_free(bbr, rsm); if (left) { rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); goto more; } goto proc_sack; } if (rsm->r_flags & BBR_ACKED) { /* * It was acked on the scoreboard -- remove it from total * for the part being cum-acked. */ p_acked += (rsm->r_end - rsm->r_start); bbr->r_ctl.rc_sacked -= (th_ack - rsm->r_start); if (bbr->r_ctl.rc_sacked == 0) bbr->r_ctl.rc_sacklast = NULL; } else { /* * It was acked up to th_ack point for the first time */ struct bbr_sendmap lrsm; memcpy(&lrsm, rsm, sizeof(struct bbr_sendmap)); lrsm.r_end = th_ack; bbr_update_rtt(tp, bbr, &lrsm, to, cts, BBR_CUM_ACKED, th_ack); } if ((rsm->r_flags & BBR_MARKED_LOST) && ((rsm->r_flags & BBR_ACKED) == 0)) { /* * It was marked lost and partly ack'd now * for the first time. We lower the rc_lost_bytes * and still leave it MARKED. */ bbr->r_ctl.rc_lost_bytes -= th_ack - rsm->r_start; } bbr_isit_a_pkt_epoch(bbr, cts, rsm, __LINE__, BBR_CUM_ACKED); bbr->r_ctl.rc_holes_rxt -= rsm->r_rtr_bytes; rsm->r_rtr_bytes = 0; /* adjust packet count */ rsm->r_start = th_ack; proc_sack: /* Check for reneging */ rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); if (rsm && (rsm->r_flags & BBR_ACKED) && (th_ack == rsm->r_start)) { /* * The peer has moved snd_una up to the edge of this send, * i.e. one that it had previously acked. The only way that * can be true if the peer threw away data (space issues) * that it had previously sacked (else it would have given * us snd_una up to (rsm->r_end). We need to undo the acked * markings here. * * Note we have to look to make sure th_ack is our * rsm->r_start in case we get an old ack where th_ack is * behind snd_una. */ bbr_peer_reneges(bbr, rsm, th->th_ack); } if ((to->to_flags & TOF_SACK) == 0) { /* We are done nothing left to log */ goto out; } rsm = TAILQ_LAST_FAST(&bbr->r_ctl.rc_map, bbr_sendmap, r_next); if (rsm) { last_seq = rsm->r_end; } else { last_seq = tp->snd_max; } /* Sack block processing */ if (SEQ_GT(th_ack, tp->snd_una)) ack_point = th_ack; else ack_point = tp->snd_una; for (i = 0; i < to->to_nsacks; i++) { bcopy((to->to_sacks + i * TCPOLEN_SACK), &sack, sizeof(sack)); sack.start = ntohl(sack.start); sack.end = ntohl(sack.end); if (SEQ_GT(sack.end, sack.start) && SEQ_GT(sack.start, ack_point) && SEQ_LT(sack.start, tp->snd_max) && SEQ_GT(sack.end, ack_point) && SEQ_LEQ(sack.end, tp->snd_max)) { if ((bbr->r_ctl.rc_num_small_maps_alloced > bbr_sack_block_limit) && (SEQ_LT(sack.end, last_seq)) && ((sack.end - sack.start) < (p_maxseg / 8))) { /* * Not the last piece and its smaller than * 1/8th of a p_maxseg. We ignore this. */ BBR_STAT_INC(bbr_runt_sacks); continue; } sack_blocks[num_sack_blks] = sack; num_sack_blks++; #ifdef NETFLIX_STATS } else if (SEQ_LEQ(sack.start, th_ack) && SEQ_LEQ(sack.end, th_ack)) { /* * Its a D-SACK block. */ tcp_record_dsack(sack.start, sack.end); #endif } } if (num_sack_blks == 0) goto out; /* * Sort the SACK blocks so we can update the rack scoreboard with * just one pass. */ new_sb = sack_filter_blks(&bbr->r_ctl.bbr_sf, sack_blocks, num_sack_blks, th->th_ack); ctf_log_sack_filter(bbr->rc_tp, new_sb, sack_blocks); BBR_STAT_ADD(bbr_sack_blocks, num_sack_blks); BBR_STAT_ADD(bbr_sack_blocks_skip, (num_sack_blks - new_sb)); num_sack_blks = new_sb; if (num_sack_blks < 2) { goto do_sack_work; } /* Sort the sacks */ for (i = 0; i < num_sack_blks; i++) { for (j = i + 1; j < num_sack_blks; j++) { if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { sack = sack_blocks[i]; sack_blocks[i] = sack_blocks[j]; sack_blocks[j] = sack; } } } /* * Now are any of the sack block ends the same (yes some * implememtations send these)? */ again: if (num_sack_blks > 1) { for (i = 0; i < num_sack_blks; i++) { for (j = i + 1; j < num_sack_blks; j++) { if (sack_blocks[i].end == sack_blocks[j].end) { /* * Ok these two have the same end we * want the smallest end and then * throw away the larger and start * again. */ if (SEQ_LT(sack_blocks[j].start, sack_blocks[i].start)) { /* * The second block covers * more area use that */ sack_blocks[i].start = sack_blocks[j].start; } /* * Now collapse out the dup-sack and * lower the count */ for (k = (j + 1); k < num_sack_blks; k++) { sack_blocks[j].start = sack_blocks[k].start; sack_blocks[j].end = sack_blocks[k].end; j++; } num_sack_blks--; goto again; } } } } do_sack_work: rsm = bbr->r_ctl.rc_sacklast; for (i = 0; i < num_sack_blks; i++) { acked = bbr_proc_sack_blk(tp, bbr, &sack_blocks[i], to, &rsm, cts); if (acked) { bbr->r_wanted_output = 1; changed += acked; sack_changed += acked; } } out: *prev_acked = p_acked; if ((sack_changed) && (!IN_RECOVERY(tp->t_flags))) { /* * Ok we have a high probability that we need to go in to * recovery since we have data sack'd */ struct bbr_sendmap *rsm; rsm = bbr_check_recovery_mode(tp, bbr, cts); if (rsm) { /* Enter recovery */ entered_recovery = 1; bbr->r_wanted_output = 1; /* * When we enter recovery we need to assure we send * one packet. */ if (bbr->r_ctl.rc_resend == NULL) { bbr->r_ctl.rc_resend = rsm; } } } if (IN_RECOVERY(tp->t_flags) && (entered_recovery == 0)) { /* * See if we need to rack-retransmit anything if so set it * up as the thing to resend assuming something else is not * already in that position. */ if (bbr->r_ctl.rc_resend == NULL) { bbr->r_ctl.rc_resend = bbr_check_recovery_mode(tp, bbr, cts); } } /* * We return the amount that changed via sack, this is used by the * ack-received code to augment what was changed between th_ack <-> * snd_una. */ return (sack_changed); } static void bbr_strike_dupack(struct tcp_bbr *bbr) { struct bbr_sendmap *rsm; rsm = TAILQ_FIRST(&bbr->r_ctl.rc_tmap); if (rsm && (rsm->r_dupack < 0xff)) { rsm->r_dupack++; if (rsm->r_dupack >= DUP_ACK_THRESHOLD) bbr->r_wanted_output = 1; } } /* * Return value of 1, we do not need to call bbr_process_data(). * return value of 0, bbr_process_data can be called. * For ret_val if its 0 the TCB is locked and valid, if its non-zero * its unlocked and probably unsafe to touch the TCB. */ static int bbr_process_ack(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, uint32_t tiwin, int32_t tlen, int32_t * ofia, int32_t thflags, int32_t * ret_val) { int32_t ourfinisacked = 0; int32_t acked_amount; uint16_t nsegs; int32_t acked; uint32_t lost, sack_changed = 0; struct mbuf *mfree; struct tcp_bbr *bbr; uint32_t prev_acked = 0; bbr = (struct tcp_bbr *)tp->t_fb_ptr; lost = bbr->r_ctl.rc_lost; nsegs = max(1, m->m_pkthdr.lro_nsegs); if (SEQ_GT(th->th_ack, tp->snd_max)) { ctf_do_dropafterack(m, tp, th, thflags, tlen, ret_val); bbr->r_wanted_output = 1; return (1); } if (SEQ_GEQ(th->th_ack, tp->snd_una) || to->to_nsacks) { /* Process the ack */ if (bbr->rc_in_persist) tp->t_rxtshift = 0; if ((th->th_ack == tp->snd_una) && (tiwin == tp->snd_wnd)) bbr_strike_dupack(bbr); sack_changed = bbr_log_ack(tp, to, th, &prev_acked); } bbr_lt_bw_sampling(bbr, bbr->r_ctl.rc_rcvtime, (bbr->r_ctl.rc_lost > lost)); if (__predict_false(SEQ_LEQ(th->th_ack, tp->snd_una))) { /* * Old ack, behind the last one rcv'd or a duplicate ack * with SACK info. */ if (th->th_ack == tp->snd_una) { bbr_ack_received(tp, bbr, th, 0, sack_changed, prev_acked, __LINE__, 0); if (bbr->r_state == TCPS_SYN_SENT) { /* * Special case on where we sent SYN. When * the SYN-ACK is processed in syn_sent * state it bumps the snd_una. This causes * us to hit here even though we did ack 1 * byte. * * Go through the nothing left case so we * send data. */ goto nothing_left; } } return (0); } /* * If we reach this point, ACK is not a duplicate, i.e., it ACKs * something we sent. */ if (tp->t_flags & TF_NEEDSYN) { /* * T/TCP: Connection was half-synchronized, and our SYN has * been ACK'd (so connection is now fully synchronized). Go * to non-starred state, increment snd_una for ACK of SYN, * and check if we can do window scaling. */ tp->t_flags &= ~TF_NEEDSYN; tp->snd_una++; /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == (TF_RCVD_SCALE | TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; /* Send window already scaled. */ } } INP_WLOCK_ASSERT(tp->t_inpcb); acked = BYTES_THIS_ACK(tp, th); KMOD_TCPSTAT_ADD(tcps_rcvackpack, (int)nsegs); KMOD_TCPSTAT_ADD(tcps_rcvackbyte, acked); /* * If we just performed our first retransmit, and the ACK arrives * within our recovery window, then it was a mistake to do the * retransmit in the first place. Recover our original cwnd and * ssthresh, and proceed to transmit where we left off. */ if (tp->t_flags & TF_PREVVALID) { tp->t_flags &= ~TF_PREVVALID; if (tp->t_rxtshift == 1 && (int)(ticks - tp->t_badrxtwin) < 0) bbr_cong_signal(tp, th, CC_RTO_ERR, NULL); } SOCKBUF_LOCK(&so->so_snd); acked_amount = min(acked, (int)sbavail(&so->so_snd)); tp->snd_wnd -= acked_amount; mfree = sbcut_locked(&so->so_snd, acked_amount); /* NB: sowwakeup_locked() does an implicit unlock. */ sowwakeup_locked(so); m_freem(mfree); if (SEQ_GT(th->th_ack, tp->snd_una)) { bbr_collapse_rtt(tp, bbr, TCP_REXMTVAL(tp)); } tp->snd_una = th->th_ack; bbr_ack_received(tp, bbr, th, acked, sack_changed, prev_acked, __LINE__, (bbr->r_ctl.rc_lost - lost)); if (IN_RECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover) && (SEQ_LT(th->th_ack, tp->snd_max))) { tcp_bbr_partialack(tp); } else { bbr_post_recovery(tp); } } if (SEQ_GT(tp->snd_una, tp->snd_recover)) { tp->snd_recover = tp->snd_una; } if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { tp->snd_nxt = tp->snd_max; } if (tp->snd_una == tp->snd_max) { /* Nothing left outstanding */ nothing_left: bbr_log_progress_event(bbr, tp, ticks, PROGRESS_CLEAR, __LINE__); if (sbavail(&tp->t_inpcb->inp_socket->so_snd) == 0) bbr->rc_tp->t_acktime = 0; if ((sbused(&so->so_snd) == 0) && (tp->t_flags & TF_SENTFIN)) { ourfinisacked = 1; } bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); if (bbr->rc_in_persist == 0) { bbr->r_ctl.rc_went_idle_time = bbr->r_ctl.rc_rcvtime; } sack_filter_clear(&bbr->r_ctl.bbr_sf, tp->snd_una); bbr_log_ack_clear(bbr, bbr->r_ctl.rc_rcvtime); /* * We invalidate the last ack here since we * don't want to transfer forward the time * for our sum's calculations. */ if ((tp->t_state >= TCPS_FIN_WAIT_1) && (sbavail(&so->so_snd) == 0) && (tp->t_flags2 & TF2_DROP_AF_DATA)) { /* * The socket was gone and the peer sent data, time * to reset him. */ *ret_val = 1; tp = tcp_close(tp); ctf_do_dropwithreset(m, tp, th, BANDLIM_UNLIMITED, tlen); BBR_STAT_INC(bbr_dropped_af_data); return (1); } /* Set need output so persist might get set */ bbr->r_wanted_output = 1; } if (ofia) *ofia = ourfinisacked; return (0); } static void bbr_enter_persist(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, int32_t line) { if (bbr->rc_in_persist == 0) { bbr_timer_cancel(bbr, __LINE__, cts); bbr->r_ctl.rc_last_delay_val = 0; tp->t_rxtshift = 0; bbr->rc_in_persist = 1; bbr->r_ctl.rc_went_idle_time = cts; /* We should be capped when rw went to 0 but just in case */ bbr_log_type_pesist(bbr, cts, 0, line, 1); /* Time freezes for the state, so do the accounting now */ if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { uint32_t time_in; time_in = cts - bbr->r_ctl.rc_bbr_state_time; if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) { int32_t idx; idx = bbr_state_val(bbr); counter_u64_add(bbr_state_time[(idx + 5)], time_in); } else { counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } } bbr->r_ctl.rc_bbr_state_time = cts; } } static void bbr_restart_after_idle(struct tcp_bbr *bbr, uint32_t cts, uint32_t idle_time) { /* * Note that if idle time does not exceed our * threshold, we do nothing continuing the state * transitions we were last walking through. */ if (idle_time >= bbr_idle_restart_threshold) { if (bbr->rc_use_idle_restart) { bbr->rc_bbr_state = BBR_STATE_IDLE_EXIT; /* * Set our target using BBR_UNIT, so * we increase at a dramatic rate but * we stop when we get the pipe * full again for our current b/w estimate. */ bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; bbr->r_ctl.rc_bbr_cwnd_gain = BBR_UNIT; bbr_set_state_target(bbr, __LINE__); /* Now setup our gains to ramp up */ bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.rc_startup_pg; bbr->r_ctl.rc_bbr_cwnd_gain = bbr->r_ctl.rc_startup_pg; bbr_log_type_statechange(bbr, cts, __LINE__); } else { bbr_substate_change(bbr, cts, __LINE__, 1); } } } static void bbr_exit_persist(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts, int32_t line) { uint32_t idle_time; if (bbr->rc_in_persist == 0) return; idle_time = bbr_calc_time(cts, bbr->r_ctl.rc_went_idle_time); bbr->rc_in_persist = 0; bbr->rc_hit_state_1 = 0; tp->t_flags &= ~TF_FORCEDATA; bbr->r_ctl.rc_del_time = cts; /* * We invalidate the last ack here since we * don't want to transfer forward the time * for our sum's calculations. */ if (bbr->rc_inp->inp_in_hpts) { tcp_hpts_remove(bbr->rc_inp, HPTS_REMOVE_OUTPUT); bbr->rc_timer_first = 0; bbr->r_ctl.rc_hpts_flags = 0; bbr->r_ctl.rc_last_delay_val = 0; bbr->r_ctl.rc_hptsi_agg_delay = 0; bbr->r_agg_early_set = 0; bbr->r_ctl.rc_agg_early = 0; } bbr_log_type_pesist(bbr, cts, idle_time, line, 0); if (idle_time >= bbr_rtt_probe_time) { /* * This qualifies as a RTT_PROBE session since we drop the * data outstanding to nothing and waited more than * bbr_rtt_probe_time. */ bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_PERSIST, 0); bbr->r_ctl.last_in_probertt = bbr->r_ctl.rc_rtt_shrinks = cts; } tp->t_rxtshift = 0; /* * If in probeBW and we have persisted more than an RTT lets do * special handling. */ /* Force a time based epoch */ bbr_set_epoch(bbr, cts, __LINE__); /* * Setup the lost so we don't count anything against the guy * we have been stuck with during persists. */ bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; /* Time un-freezes for the state */ bbr->r_ctl.rc_bbr_state_time = cts; if ((bbr->rc_bbr_state == BBR_STATE_PROBE_BW) || (bbr->rc_bbr_state == BBR_STATE_PROBE_RTT)) { /* * If we are going back to probe-bw * or probe_rtt, we may need to possibly * do a fast restart. */ bbr_restart_after_idle(bbr, cts, idle_time); } } static void bbr_collapsed_window(struct tcp_bbr *bbr) { /* * Now we must walk the * send map and divide the * ones left stranded. These * guys can't cause us to abort * the connection and are really * "unsent". However if a buggy * client actually did keep some * of the data i.e. collapsed the win * and refused to ack and then opened * the win and acked that data. We would * get into an ack war, the simplier * method then of just pretending we * did not send those segments something * won't work. */ struct bbr_sendmap *rsm, *nrsm; tcp_seq max_seq; uint32_t maxseg; int can_split = 0; int fnd = 0; maxseg = bbr->rc_tp->t_maxseg - bbr->rc_last_options; max_seq = bbr->rc_tp->snd_una + bbr->rc_tp->snd_wnd; bbr_log_type_rwnd_collapse(bbr, max_seq, 1, 0); TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_map, r_next) { /* Find the first seq past or at maxseq */ if (rsm->r_flags & BBR_RWND_COLLAPSED) rsm->r_flags &= ~BBR_RWND_COLLAPSED; if (SEQ_GEQ(max_seq, rsm->r_start) && SEQ_GEQ(rsm->r_end, max_seq)) { fnd = 1; break; } } bbr->rc_has_collapsed = 0; if (!fnd) { /* Nothing to do strange */ return; } /* * Now can we split? * * We don't want to split if splitting * would generate too many small segments * less we let an attacker fragment our * send_map and leave us out of memory. */ if ((max_seq != rsm->r_start) && (max_seq != rsm->r_end)){ /* can we split? */ int res1, res2; res1 = max_seq - rsm->r_start; res2 = rsm->r_end - max_seq; if ((res1 >= (maxseg/8)) && (res2 >= (maxseg/8))) { /* No small pieces here */ can_split = 1; } else if (bbr->r_ctl.rc_num_small_maps_alloced < bbr_sack_block_limit) { /* We are under the limit */ can_split = 1; } } /* Ok do we need to split this rsm? */ if (max_seq == rsm->r_start) { /* It's this guy no split required */ nrsm = rsm; } else if (max_seq == rsm->r_end) { /* It's the next one no split required. */ nrsm = TAILQ_NEXT(rsm, r_next); if (nrsm == NULL) { /* Huh? */ return; } } else if (can_split && SEQ_LT(max_seq, rsm->r_end)) { /* yep we need to split it */ nrsm = bbr_alloc_limit(bbr, BBR_LIMIT_TYPE_SPLIT); if (nrsm == NULL) { /* failed XXXrrs what can we do mark the whole? */ nrsm = rsm; goto no_split; } /* Clone it */ bbr_log_type_rwnd_collapse(bbr, max_seq, 3, 0); bbr_clone_rsm(bbr, nrsm, rsm, max_seq); TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_map, rsm, nrsm, r_next); if (rsm->r_in_tmap) { TAILQ_INSERT_AFTER(&bbr->r_ctl.rc_tmap, rsm, nrsm, r_tnext); nrsm->r_in_tmap = 1; } } else { /* * Split not allowed just start here just * use this guy. */ nrsm = rsm; } no_split: BBR_STAT_INC(bbr_collapsed_win); /* reuse fnd as a count */ fnd = 0; TAILQ_FOREACH_FROM(nrsm, &bbr->r_ctl.rc_map, r_next) { nrsm->r_flags |= BBR_RWND_COLLAPSED; fnd++; bbr->rc_has_collapsed = 1; } bbr_log_type_rwnd_collapse(bbr, max_seq, 4, fnd); } static void bbr_un_collapse_window(struct tcp_bbr *bbr) { struct bbr_sendmap *rsm; int cleared = 0; TAILQ_FOREACH_REVERSE(rsm, &bbr->r_ctl.rc_map, bbr_head, r_next) { if (rsm->r_flags & BBR_RWND_COLLAPSED) { /* Clear the flag */ rsm->r_flags &= ~BBR_RWND_COLLAPSED; cleared++; } else break; } bbr_log_type_rwnd_collapse(bbr, (bbr->rc_tp->snd_una + bbr->rc_tp->snd_wnd), 0, cleared); bbr->rc_has_collapsed = 0; } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_process_data(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { /* * Update window information. Don't look at window if no ACK: TAC's * send garbage on first SYN. */ uint16_t nsegs; int32_t tfo_syn; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; INP_WLOCK_ASSERT(tp->t_inpcb); nsegs = max(1, m->m_pkthdr.lro_nsegs); if ((thflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) KMOD_TCPSTAT_INC(tcps_rcvwinupd); tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; bbr->r_wanted_output = 1; } else if (thflags & TH_ACK) { if ((tp->snd_wl2 == th->th_ack) && (tiwin < tp->snd_wnd)) { tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; } } if (tp->snd_wnd < ctf_outstanding(tp)) /* The peer collapsed its window on us */ bbr_collapsed_window(bbr); else if (bbr->rc_has_collapsed) bbr_un_collapse_window(bbr); /* Was persist timer active and now we have window space? */ if ((bbr->rc_in_persist != 0) && (tp->snd_wnd >= min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr)))) { /* * Make the rate persist at end of persist mode if idle long * enough */ bbr_exit_persist(tp, bbr, bbr->r_ctl.rc_rcvtime, __LINE__); /* Make sure we output to start the timer */ bbr->r_wanted_output = 1; } /* Do we need to enter persist? */ if ((bbr->rc_in_persist == 0) && (tp->snd_wnd < min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr))) && TCPS_HAVEESTABLISHED(tp->t_state) && (tp->snd_max == tp->snd_una) && sbavail(&tp->t_inpcb->inp_socket->so_snd) && (sbavail(&tp->t_inpcb->inp_socket->so_snd) > tp->snd_wnd)) { /* No send window.. we must enter persist */ bbr_enter_persist(tp, bbr, bbr->r_ctl.rc_rcvtime, __LINE__); } if (tp->t_flags2 & TF2_DROP_AF_DATA) { m_freem(m); return (0); } /* * Process segments with URG. */ if ((thflags & TH_URG) && th->th_urp && TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* * This is a kludge, but if we receive and accept random * urgent pointers, we'll crash in soreceive. It's hard to * imagine someone actually wanting to send this much urgent * data. */ SOCKBUF_LOCK(&so->so_rcv); if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { th->th_urp = 0; /* XXX */ thflags &= ~TH_URG; /* XXX */ SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, then * mark the data stream. This should not happen in * CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since a * FIN has been received from the remote side. In these * states we ignore the URG. * * According to RFC961 (Assigned Protocols), the urgent * pointer points to the last octet of urgent data. We * continue, however, to consider it to indicate the first * octet of data past the urgent section as the original * spec states (in one of two places). */ if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = sbavail(&so->so_rcv) + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) so->so_rcv.sb_state |= SBS_RCVATMARK; sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); } SOCKBUF_UNLOCK(&so->so_rcv); /* * Remove out of band data so doesn't get presented to user. * This can happen independent of advancing the URG pointer, * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ if (th->th_urp <= (uint32_t)tlen && !(so->so_options & SO_OOBINLINE)) { /* hdr drop is delayed */ tcp_pulloutofband(so, th, m, drop_hdrlen); } } else { /* * If no out of band data is expected, pull receive urgent * pointer along with the receive window. */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; } dodata: /* XXX */ INP_WLOCK_ASSERT(tp->t_inpcb); /* * Process the segment text, merging it into the TCP sequencing * queue, and arranging for acknowledgment of receipt if necessary. * This process logically involves adjusting tp->rcv_wnd as data is * presented to the user (this happens in tcp_usrreq.c, case * PRU_RCVD). If a FIN has already been received on this connection * then we just ignore the text. */ tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && IS_FASTOPEN(tp->t_flags)); if ((tlen || (thflags & TH_FIN) || tfo_syn) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { tcp_seq save_start = th->th_seq; tcp_seq save_rnxt = tp->rcv_nxt; int save_tlen = tlen; m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly * queue with control block tp. Set thflags to whether * reassembly now includes a segment with FIN. This handles * the common case inline (segment is the next to be * received on an established connection, and the queue is * empty), avoiding linkage into and removal from the queue * and repetition of various conversions. Set DELACK for * segments received in order, but ack immediately when * segments are out of order (so fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && SEGQ_EMPTY(tp) && (TCPS_HAVEESTABLISHED(tp->t_state) || tfo_syn)) { #ifdef NETFLIX_SB_LIMITS u_int mcnt, appended; if (so->so_rcv.sb_shlim) { mcnt = m_memcnt(m); appended = 0; if (counter_fo_get(so->so_rcv.sb_shlim, mcnt, CFO_NOSLEEP, NULL) == false) { counter_u64_add(tcp_sb_shlim_fails, 1); m_freem(m); return (0); } } #endif if (DELAY_ACK(tp, bbr, nsegs) || tfo_syn) { bbr->bbr_segs_rcvd += max(1, nsegs); tp->t_flags |= TF_DELACK; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); } else { bbr->r_wanted_output = 1; tp->t_flags |= TF_ACKNOW; } tp->rcv_nxt += tlen; thflags = th->th_flags & TH_FIN; KMOD_TCPSTAT_ADD(tcps_rcvpack, (int)nsegs); KMOD_TCPSTAT_ADD(tcps_rcvbyte, tlen); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) m_freem(m); else #ifdef NETFLIX_SB_LIMITS appended = #endif sbappendstream_locked(&so->so_rcv, m, 0); /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); #ifdef NETFLIX_SB_LIMITS if (so->so_rcv.sb_shlim && appended != mcnt) counter_fo_release(so->so_rcv.sb_shlim, mcnt - appended); #endif } else { /* * XXX: Due to the header drop above "th" is * theoretically invalid by now. Fortunately * m_adj() doesn't actually frees any mbufs when * trimming from the head. */ tcp_seq temp = save_start; thflags = tcp_reass(tp, th, &temp, &tlen, m); tp->t_flags |= TF_ACKNOW; } if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) { if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { /* * DSACK actually handled in the fastpath * above. */ tcp_update_sack_list(tp, save_start, save_start + save_tlen); } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { if ((tp->rcv_numsacks >= 1) && (tp->sackblks[0].end == save_start)) { /* * Partial overlap, recorded at todrop * above. */ tcp_update_sack_list(tp, tp->sackblks[0].start, tp->sackblks[0].end); } else { tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } } else if (tlen >= save_tlen) { /* Update of sackblks. */ tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } else if (tlen > 0) { tcp_update_dsack_list(tp, save_start, save_start + tlen); } } } else { m_freem(m); thflags &= ~TH_FIN; } /* * If FIN is received ACK the FIN and let the user know that the * connection is closing. */ if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { socantrcvmore(so); /* * If connection is half-synchronized (ie NEEDSYN * flag on) then delay ACK, so it may be piggybacked * when SYN is sent. Otherwise, since we received a * FIN then no more input can be expected, send ACK * now. */ if (tp->t_flags & TF_NEEDSYN) { tp->t_flags |= TF_DELACK; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); } else { tp->t_flags |= TF_ACKNOW; } tp->rcv_nxt++; } switch (tp->t_state) { /* * In SYN_RECEIVED and ESTABLISHED STATES enter the * CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tcp_state_change(tp, TCPS_CLOSE_WAIT); break; /* * If still in FIN_WAIT_1 STATE FIN has not been * acked so enter the CLOSING state. */ case TCPS_FIN_WAIT_1: tcp_state_change(tp, TCPS_CLOSING); break; /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the * other standard timers. */ case TCPS_FIN_WAIT_2: bbr->rc_timer_first = 1; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); INP_WLOCK_ASSERT(tp->t_inpcb); tcp_twstart(tp); return (1); } } /* * Return any desired output. */ if ((tp->t_flags & TF_ACKNOW) || (sbavail(&so->so_snd) > ctf_outstanding(tp))) { bbr->r_wanted_output = 1; } INP_WLOCK_ASSERT(tp->t_inpcb); return (0); } /* * Here nothing is really faster, its just that we * have broken out the fast-data path also just like * the fast-ack. Return 1 if we processed the packet * return 0 if you need to take the "slow-path". */ static int bbr_do_fastnewdata(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t nxt_pkt) { uint16_t nsegs; int32_t newsize = 0; /* automatic sockbuf scaling */ struct tcp_bbr *bbr; #ifdef NETFLIX_SB_LIMITS u_int mcnt, appended; #endif #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif /* On the hpts and we would have called output */ bbr = (struct tcp_bbr *)tp->t_fb_ptr; /* * If last ACK falls within this segment's sequence numbers, record * the timestamp. NOTE that the test is modified according to the * latest proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if (bbr->r_ctl.rc_resend != NULL) { return (0); } if (tiwin && tiwin != tp->snd_wnd) { return (0); } if (__predict_false((tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)))) { return (0); } if (__predict_false((to->to_flags & TOF_TS) && (TSTMP_LT(to->to_tsval, tp->ts_recent)))) { return (0); } if (__predict_false((th->th_ack != tp->snd_una))) { return (0); } if (__predict_false(tlen > sbspace(&so->so_rcv))) { return (0); } if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * This is a pure, in-sequence data packet with nothing on the * reassembly queue and we have enough buffer space to take it. */ nsegs = max(1, m->m_pkthdr.lro_nsegs); #ifdef NETFLIX_SB_LIMITS if (so->so_rcv.sb_shlim) { mcnt = m_memcnt(m); appended = 0; if (counter_fo_get(so->so_rcv.sb_shlim, mcnt, CFO_NOSLEEP, NULL) == false) { counter_u64_add(tcp_sb_shlim_fails, 1); m_freem(m); return (1); } } #endif /* Clean receiver SACK report if present */ if (tp->rcv_numsacks) tcp_clean_sackreport(tp); KMOD_TCPSTAT_INC(tcps_preddat); tp->rcv_nxt += tlen; /* * Pull snd_wl1 up to prevent seq wrap relative to th_seq. */ tp->snd_wl1 = th->th_seq; /* * Pull rcv_up up to prevent seq wrap relative to rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; KMOD_TCPSTAT_ADD(tcps_rcvpack, (int)nsegs); KMOD_TCPSTAT_ADD(tcps_rcvbyte, tlen); #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif newsize = tcp_autorcvbuf(m, th, so, tp, tlen); /* Add data to socket buffer. */ SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { m_freem(m); } else { /* * Set new socket buffer size. Give up when limit is * reached. */ if (newsize) if (!sbreserve_locked(&so->so_rcv, newsize, so, NULL)) so->so_rcv.sb_flags &= ~SB_AUTOSIZE; m_adj(m, drop_hdrlen); /* delayed header drop */ #ifdef NETFLIX_SB_LIMITS appended = #endif sbappendstream_locked(&so->so_rcv, m, 0); ctf_calc_rwin(so, tp); } /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); #ifdef NETFLIX_SB_LIMITS if (so->so_rcv.sb_shlim && mcnt != appended) counter_fo_release(so->so_rcv.sb_shlim, mcnt - appended); #endif if (DELAY_ACK(tp, bbr, nsegs)) { bbr->bbr_segs_rcvd += max(1, nsegs); tp->t_flags |= TF_DELACK; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); } else { bbr->r_wanted_output = 1; tp->t_flags |= TF_ACKNOW; } return (1); } /* * This subfunction is used to try to highly optimize the * fast path. We again allow window updates that are * in sequence to remain in the fast-path. We also add * in the __predict's to attempt to help the compiler. * Note that if we return a 0, then we can *not* process * it and the caller should push the packet into the * slow-path. If we return 1, then all is well and * the packet is fully processed. */ static int bbr_fastack(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t nxt_pkt) { int32_t acked; uint16_t nsegs; uint32_t sack_changed; #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif uint32_t prev_acked = 0; struct tcp_bbr *bbr; if (__predict_false(SEQ_LEQ(th->th_ack, tp->snd_una))) { /* Old ack, behind (or duplicate to) the last one rcv'd */ return (0); } if (__predict_false(SEQ_GT(th->th_ack, tp->snd_max))) { /* Above what we have sent? */ return (0); } if (__predict_false(tiwin == 0)) { /* zero window */ return (0); } if (__predict_false(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN))) { /* We need a SYN or a FIN, unlikely.. */ return (0); } if ((to->to_flags & TOF_TS) && __predict_false(TSTMP_LT(to->to_tsval, tp->ts_recent))) { /* Timestamp is behind .. old ack with seq wrap? */ return (0); } if (__predict_false(IN_RECOVERY(tp->t_flags))) { /* Still recovering */ return (0); } bbr = (struct tcp_bbr *)tp->t_fb_ptr; if (__predict_false(bbr->r_ctl.rc_resend != NULL)) { /* We are retransmitting */ return (0); } if (__predict_false(bbr->rc_in_persist != 0)) { /* In persist mode */ return (0); } if (bbr->r_ctl.rc_sacked) { /* We have sack holes on our scoreboard */ return (0); } /* Ok if we reach here, we can process a fast-ack */ nsegs = max(1, m->m_pkthdr.lro_nsegs); sack_changed = bbr_log_ack(tp, to, th, &prev_acked); /* * We never detect loss in fast ack [we can't * have a sack and can't be in recovery so * we always pass 0 (nothing detected)]. */ bbr_lt_bw_sampling(bbr, bbr->r_ctl.rc_rcvtime, 0); /* Did the window get updated? */ if (tiwin != tp->snd_wnd) { tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; } /* Do we need to exit persists? */ if ((bbr->rc_in_persist != 0) && (tp->snd_wnd >= min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr)))) { bbr_exit_persist(tp, bbr, bbr->r_ctl.rc_rcvtime, __LINE__); bbr->r_wanted_output = 1; } /* Do we need to enter persists? */ if ((bbr->rc_in_persist == 0) && (tp->snd_wnd < min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr))) && TCPS_HAVEESTABLISHED(tp->t_state) && (tp->snd_max == tp->snd_una) && sbavail(&tp->t_inpcb->inp_socket->so_snd) && (sbavail(&tp->t_inpcb->inp_socket->so_snd) > tp->snd_wnd)) { /* No send window.. we must enter persist */ bbr_enter_persist(tp, bbr, bbr->r_ctl.rc_rcvtime, __LINE__); } /* * If last ACK falls within this segment's sequence numbers, record * the timestamp. NOTE that the test is modified according to the * latest proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = bbr->r_ctl.rc_rcvtime; tp->ts_recent = to->to_tsval; } /* * This is a pure ack for outstanding data. */ KMOD_TCPSTAT_INC(tcps_predack); /* * "bad retransmit" recovery. */ if (tp->t_flags & TF_PREVVALID) { tp->t_flags &= ~TF_PREVVALID; if (tp->t_rxtshift == 1 && (int)(ticks - tp->t_badrxtwin) < 0) bbr_cong_signal(tp, th, CC_RTO_ERR, NULL); } /* * Recalculate the transmit timer / rtt. * * Some boxes send broken timestamp replies during the SYN+ACK * phase, ignore timestamps of 0 or we could calculate a huge RTT * and blow up the retransmit timer. */ acked = BYTES_THIS_ACK(tp, th); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, to); #endif KMOD_TCPSTAT_ADD(tcps_rcvackpack, (int)nsegs); KMOD_TCPSTAT_ADD(tcps_rcvackbyte, acked); sbdrop(&so->so_snd, acked); if (SEQ_GT(th->th_ack, tp->snd_una)) bbr_collapse_rtt(tp, bbr, TCP_REXMTVAL(tp)); tp->snd_una = th->th_ack; if (tp->snd_wnd < ctf_outstanding(tp)) /* The peer collapsed its window on us */ bbr_collapsed_window(bbr); else if (bbr->rc_has_collapsed) bbr_un_collapse_window(bbr); if (SEQ_GT(tp->snd_una, tp->snd_recover)) { tp->snd_recover = tp->snd_una; } bbr_ack_received(tp, bbr, th, acked, sack_changed, prev_acked, __LINE__, 0); /* * Pull snd_wl2 up to prevent seq wrap relative to th_ack. */ tp->snd_wl2 = th->th_ack; m_freem(m); /* * If all outstanding data are acked, stop retransmit timer, * otherwise restart timer using current (possibly backed-off) * value. If process is waiting for space, wakeup/selwakeup/signal. * If data are ready to send, let tcp_output decide between more * output or persist. */ #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif /* Wake up the socket if we have room to write more */ sowwakeup(so); if (tp->snd_una == tp->snd_max) { /* Nothing left outstanding */ bbr_log_progress_event(bbr, tp, ticks, PROGRESS_CLEAR, __LINE__); if (sbavail(&tp->t_inpcb->inp_socket->so_snd) == 0) bbr->rc_tp->t_acktime = 0; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); if (bbr->rc_in_persist == 0) { bbr->r_ctl.rc_went_idle_time = bbr->r_ctl.rc_rcvtime; } sack_filter_clear(&bbr->r_ctl.bbr_sf, tp->snd_una); bbr_log_ack_clear(bbr, bbr->r_ctl.rc_rcvtime); /* * We invalidate the last ack here since we * don't want to transfer forward the time * for our sum's calculations. */ bbr->r_wanted_output = 1; } if (sbavail(&so->so_snd)) { bbr->r_wanted_output = 1; } return (1); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_syn_sent(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t todrop; int32_t ourfinisacked = 0; struct tcp_bbr *bbr; int32_t ret_val = 0; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); /* * If the state is SYN_SENT: if seg contains an ACK, but not for our * SYN, drop the input. if seg contains a RST, then drop the * connection. if seg does not contain SYN, then drop it. Otherwise * this is an acceptable SYN segment initialize tp->rcv_nxt and * tp->irs if seg contains ack then advance tp->snd_una. BRR does * not support ECN so we will not say we are capable. if SYN has * been acked change to ESTABLISHED else SYN_RCVD state arrange for * segment to be acked (eventually) continue processing rest of * data/controls, beginning with URG */ if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } if ((thflags & (TH_ACK | TH_RST)) == (TH_ACK | TH_RST)) { TCP_PROBE5(connect__refused, NULL, tp, mtod(m, const char *), tp, th); tp = tcp_drop(tp, ECONNREFUSED); ctf_do_drop(m, tp); return (1); } if (thflags & TH_RST) { ctf_do_drop(m, tp); return (1); } if (!(thflags & TH_SYN)) { ctf_do_drop(m, tp); return (1); } tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { int tfo_partial = 0; KMOD_TCPSTAT_INC(tcps_connects); soisconnected(so); #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif /* Do window scaling on this connection? */ if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == (TF_RCVD_SCALE | TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); /* * If not all the data that was sent in the TFO SYN * has been acked, resend the remainder right away. */ if (IS_FASTOPEN(tp->t_flags) && (tp->snd_una != tp->snd_max)) { tp->snd_nxt = th->th_ack; tfo_partial = 1; } /* * If there's data, delay ACK; if there's also a FIN ACKNOW * will be turned on later. */ if (DELAY_ACK(tp, bbr, 1) && tlen != 0 && (tfo_partial == 0)) { bbr->bbr_segs_rcvd += 1; tp->t_flags |= TF_DELACK; bbr_timer_cancel(bbr, __LINE__, bbr->r_ctl.rc_rcvtime); } else { bbr->r_wanted_output = 1; tp->t_flags |= TF_ACKNOW; } if (SEQ_GT(th->th_ack, tp->iss)) { /* * The SYN is acked * handle it specially. */ bbr_log_syn(tp, to); } if (SEQ_GT(th->th_ack, tp->snd_una)) { /* * We advance snd_una for the * fast open case. If th_ack is * acknowledging data beyond * snd_una we can't just call * ack-processing since the * data stream in our send-map * will start at snd_una + 1 (one * beyond the SYN). If its just * equal we don't need to do that * and there is no send_map. */ tp->snd_una++; } /* * Received in SYN_SENT[*] state. Transitions: * SYN_SENT --> ESTABLISHED SYN_SENT* --> FIN_WAIT_1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(connect__established, NULL, tp, mtod(m, const char *), tp, th); cc_conn_init(tp); } } else { /* * Received initial SYN in SYN-SENT[*] state => simultaneous * open. If segment contains CC option and there is a * cached CC, apply TAO test. If it succeeds, connection is * * half-synchronized. Otherwise, do 3-way handshake: * SYN-SENT -> SYN-RECEIVED SYN-SENT* -> SYN-RECEIVED* If * there was no CC option, clear cached CC value. */ tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); tcp_state_change(tp, TCPS_SYN_RECEIVED); } INP_WLOCK_ASSERT(tp->t_inpcb); /* * Advance th->th_seq to correspond to first data byte. If data, * trim to stay within window, dropping FIN if necessary. */ th->th_seq++; if (tlen > tp->rcv_wnd) { todrop = tlen - tp->rcv_wnd; m_adj(m, -todrop); tlen = tp->rcv_wnd; thflags &= ~TH_FIN; KMOD_TCPSTAT_INC(tcps_rcvpackafterwin); KMOD_TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); } tp->snd_wl1 = th->th_seq - 1; tp->rcv_up = th->th_seq; /* * Client side of transaction: already sent SYN and data. If the * remote host used T/TCP to validate the SYN, our data will be * ACK'd; if so, enter normal data segment processing in the middle * of step 5, ack processing. Otherwise, goto step 6. */ if (thflags & TH_ACK) { if ((to->to_flags & TOF_TS) != 0) { uint32_t t, rtt; t = tcp_tv_to_mssectick(&bbr->rc_tv); if (TSTMP_GEQ(t, to->to_tsecr)) { rtt = t - to->to_tsecr; if (rtt == 0) { rtt = 1; } rtt *= MS_IN_USEC; tcp_bbr_xmit_timer(bbr, rtt, 0, 0, 0); apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, bbr->r_ctl.rc_rcvtime); } } if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) return (ret_val); /* We may have changed to FIN_WAIT_1 above */ if (tp->t_state == TCPS_FIN_WAIT_1) { /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now * acknowledged then enter FIN_WAIT_2. */ if (ourfinisacked) { /* * If we can't receive any more data, then * closing user can proceed. Starting the * timer is contrary to the specification, * but if we don't get a FIN we'll hang * forever. * * XXXjl: we should release the tp also, and * use a compressed state. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_syn_recv(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t ourfinisacked = 0; int32_t ret_val; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max))) { ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } if (IS_FASTOPEN(tp->t_flags)) { /* * When a TFO connection is in SYN_RECEIVED, the only valid * packets are the initial SYN, a retransmit/copy of the * initial SYN (possibly with a subset of the original * data), a valid ACK, a FIN, or a RST. */ if ((thflags & (TH_SYN | TH_ACK)) == (TH_SYN | TH_ACK)) { ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } else if (thflags & TH_SYN) { /* non-initial SYN is ignored */ if ((bbr->r_ctl.rc_hpts_flags & PACE_TMR_RXT) || (bbr->r_ctl.rc_hpts_flags & PACE_TMR_TLP) || (bbr->r_ctl.rc_hpts_flags & PACE_TMR_RACK)) { ctf_do_drop(m, NULL); return (0); } } else if (!(thflags & (TH_ACK | TH_FIN | TH_RST))) { ctf_do_drop(m, NULL); return (0); } } if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive * window. Check the sequence number versus IRS since we know the * sequence numbers haven't wrapped. This is a partial fix for the * "LAND" DoS attack. */ if (SEQ_LT(th->th_seq, tp->irs)) { ctf_do_dropwithreset(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } tp->snd_wnd = tiwin; /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (IS_FASTOPEN(tp->t_flags)) { cc_conn_init(tp); } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } KMOD_TCPSTAT_INC(tcps_connects); soisconnected(so); /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == (TF_RCVD_SCALE | TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } /* * ok for the first time in lets see if we can use the ts to figure * out what the initial RTT was. */ if ((to->to_flags & TOF_TS) != 0) { uint32_t t, rtt; t = tcp_tv_to_mssectick(&bbr->rc_tv); if (TSTMP_GEQ(t, to->to_tsecr)) { rtt = t - to->to_tsecr; if (rtt == 0) { rtt = 1; } rtt *= MS_IN_USEC; tcp_bbr_xmit_timer(bbr, rtt, 0, 0, 0); apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, bbr->r_ctl.rc_rcvtime); } } /* Drop off any SYN in the send map (probably not there) */ if (thflags & TH_ACK) bbr_log_syn(tp, to); if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; /* * Account for the ACK of our SYN prior to regular * ACK processing below. */ tp->snd_una++; } /* * Make transitions: SYN-RECEIVED -> ESTABLISHED SYN-RECEIVED* -> * FIN-WAIT-1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(accept__established, NULL, tp, mtod(m, const char *), tp, th); /* * TFO connections call cc_conn_init() during SYN * processing. Calling it again here for such connections * is not harmless as it would undo the snd_cwnd reduction * that occurs when a TFO SYN|ACK is retransmitted. */ if (!IS_FASTOPEN(tp->t_flags)) cc_conn_init(tp); } /* * If segment contains data or ACK, will call tcp_reass() later; if * not, do so now to pass queued data to user. */ if (tlen == 0 && (thflags & TH_FIN) == 0) (void)tcp_reass(tp, (struct tcphdr *)0, NULL, 0, (struct mbuf *)0); tp->snd_wl1 = th->th_seq - 1; if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) { return (ret_val); } if (tp->t_state == TCPS_FIN_WAIT_1) { /* We could have went to FIN_WAIT_1 (or EST) above */ /* * In FIN_WAIT_1 STATE in addition to the processing for the * ESTABLISHED state if our FIN is now acknowledged then * enter FIN_WAIT_2. */ if (ourfinisacked) { /* * If we can't receive any more data, then closing * user can proceed. Starting the timer is contrary * to the specification, but if we don't get a FIN * we'll hang forever. * * XXXjl: we should release the tp also, and use a * compressed state. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_established(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { struct tcp_bbr *bbr; int32_t ret_val; /* * Header prediction: check for the two common cases of a * uni-directional data xfer. If the packet has no control flags, * is in-sequence, the window didn't change and we're not * retransmitting, it's a candidate. If the length is zero and the * ack moved forward, we're the sender side of the xfer. Just free * the data acked & wake any higher level process that was blocked * waiting for space. If the length is non-zero and the ack didn't * move, we're the receiver side. If we're getting packets in-order * (the reassembly queue is empty), add the data toc The socket * buffer and note that we need a delayed ack. Make sure that the * hidden state-flags are also off. Since we check for * TCPS_ESTABLISHED first, it can only be TH_NEEDSYN. */ bbr = (struct tcp_bbr *)tp->t_fb_ptr; if (bbr->r_ctl.rc_delivered < (4 * tp->t_maxseg)) { /* * If we have delived under 4 segments increase the initial * window if raised by the peer. We use this to determine * dynamic and static rwnd's at the end of a connection. */ bbr->r_ctl.rc_init_rwnd = max(tiwin, tp->snd_wnd); } if (__predict_true(((to->to_flags & TOF_SACK) == 0)) && __predict_true((thflags & (TH_SYN | TH_FIN | TH_RST | TH_URG | TH_ACK)) == TH_ACK) && __predict_true(SEGQ_EMPTY(tp)) && __predict_true(th->th_seq == tp->rcv_nxt)) { if (tlen == 0) { if (bbr_fastack(m, th, so, tp, to, drop_hdrlen, tlen, tiwin, nxt_pkt)) { return (0); } } else { if (bbr_do_fastnewdata(m, th, so, tp, to, drop_hdrlen, tlen, tiwin, nxt_pkt)) { return (0); } } } ctf_calc_rwin(so, tp); if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * Ack processing. */ if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, NULL, thflags, &ret_val)) { return (ret_val); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } /* State changes only happen in bbr_process_data() */ return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_close_wait(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { struct tcp_bbr *bbr; int32_t ret_val; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * Ack processing. */ if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, NULL, thflags, &ret_val)) { return (ret_val); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } static int bbr_check_data_after_close(struct mbuf *m, struct tcp_bbr *bbr, struct tcpcb *tp, int32_t * tlen, struct tcphdr *th, struct socket *so) { if (bbr->rc_allow_data_af_clo == 0) { close_now: tp = tcp_close(tp); KMOD_TCPSTAT_INC(tcps_rcvafterclose); ctf_do_dropwithreset(m, tp, th, BANDLIM_UNLIMITED, (*tlen)); return (1); } if (sbavail(&so->so_snd) == 0) goto close_now; /* Ok we allow data that is ignored and a followup reset */ tp->rcv_nxt = th->th_seq + *tlen; tp->t_flags2 |= TF2_DROP_AF_DATA; bbr->r_wanted_output = 1; *tlen = 0; return (0); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_fin_wait_1(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t ourfinisacked = 0; int32_t ret_val; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If new data are received on a connection after the user processes * are gone, then RST the other end. */ if ((so->so_state & SS_NOFDREF) && tlen) { /* * We call a new function now so we might continue and setup * to reset at all data being ack'd. */ if (bbr_check_data_after_close(m, bbr, tp, &tlen, th, so)) return (1); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * Ack processing. */ if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) { return (ret_val); } if (ourfinisacked) { /* * If we can't receive any more data, then closing user can * proceed. Starting the timer is contrary to the * specification, but if we don't get a FIN we'll hang * forever. * * XXXjl: we should release the tp also, and use a * compressed state. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_closing(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t ourfinisacked = 0; int32_t ret_val; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If new data are received on a connection after the user processes * are gone, then RST the other end. */ if ((so->so_state & SS_NOFDREF) && tlen) { /* * We call a new function now so we might continue and setup * to reset at all data being ack'd. */ if (bbr_check_data_after_close(m, bbr, tp, &tlen, th, so)) return (1); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * Ack processing. */ if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) { return (ret_val); } if (ourfinisacked) { tcp_twstart(tp); m_freem(m); return (1); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_lastack(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t ourfinisacked = 0; int32_t ret_val; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If new data are received on a connection after the user processes * are gone, then RST the other end. */ if ((so->so_state & SS_NOFDREF) && tlen) { /* * We call a new function now so we might continue and setup * to reset at all data being ack'd. */ if (bbr_check_data_after_close(m, bbr, tp, &tlen, th, so)) return (1); } /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * case TCPS_LAST_ACK: Ack processing. */ if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) { return (ret_val); } if (ourfinisacked) { tp = tcp_close(tp); ctf_do_drop(m, tp); return (1); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } /* * Return value of 1, the TCB is unlocked and most * likely gone, return value of 0, the TCB is still * locked. */ static int bbr_do_fin_wait_2(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, struct tcpopt *to, int32_t drop_hdrlen, int32_t tlen, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) { int32_t ourfinisacked = 0; int32_t ret_val; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; ctf_calc_rwin(so, tp); /* Reset receive buffer auto scaling when not in bulk receive mode. */ if ((thflags & TH_RST) || (tp->t_fin_is_rst && (thflags & TH_FIN))) return (ctf_process_rst(m, th, so, tp)); /* * RFC5961 Section 4.2 Send challenge ACK for any SYN in * synchronized state. */ if (thflags & TH_SYN) { ctf_challenge_ack(m, th, tp, &ret_val); return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); /* * RFC 1323 PAWS: If we have a timestamp reply on this segment and * it's less than ts_recent, drop it. */ if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to->to_tsval, tp->ts_recent)) { if (ctf_ts_check(m, th, tp, tlen, thflags, &ret_val)) return (ret_val); } INP_WLOCK_ASSERT(tp->t_inpcb); if (ctf_drop_checks(to, m, th, tp, &tlen, &thflags, &drop_hdrlen, &ret_val)) { return (ret_val); } /* * If new data are received on a connection after the user processes * are gone, then we may RST the other end depending on the outcome * of bbr_check_data_after_close. */ if ((so->so_state & SS_NOFDREF) && tlen) { /* * We call a new function now so we might continue and setup * to reset at all data being ack'd. */ if (bbr_check_data_after_close(m, bbr, tp, &tlen, th, so)) return (1); } INP_WLOCK_ASSERT(tp->t_inpcb); /* * If last ACK falls within this segment's sequence numbers, record * its timestamp. NOTE: 1) That the test incorporates suggestions * from the latest proposal of the tcplw@cray.com list (Braden * 1993/04/26). 2) That updating only on newer timestamps interferes * with our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. 3) That we * modify the segment boundary check to be Last.ACK.Sent <= SEG.SEQ * + SEG.Len instead of RFC1323's Last.ACK.Sent < SEG.SEQ + * SEG.Len, This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated Vol. 2 * p.869. In such cases, we can still calculate the RTT correctly * when RCV.NXT == Last.ACK.Sent. */ INP_WLOCK_ASSERT(tp->t_inpcb); if ((to->to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN | TH_FIN)) != 0))) { tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); tp->ts_recent = to->to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * is on (half-synchronized state), then queue data for later * processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_flags & TF_NEEDSYN) { return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } else if (tp->t_flags & TF_ACKNOW) { ctf_do_dropafterack(m, tp, th, thflags, tlen, &ret_val); bbr->r_wanted_output = 1; return (ret_val); } else { ctf_do_drop(m, NULL); return (0); } } /* * Ack processing. */ INP_WLOCK_ASSERT(tp->t_inpcb); if (bbr_process_ack(m, th, so, tp, to, tiwin, tlen, &ourfinisacked, thflags, &ret_val)) { return (ret_val); } if (sbavail(&so->so_snd)) { if (bbr_progress_timeout_check(bbr)) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } } INP_WLOCK_ASSERT(tp->t_inpcb); return (bbr_process_data(m, th, so, tp, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt)); } static void bbr_stop_all_timers(struct tcpcb *tp) { struct tcp_bbr *bbr; /* * Assure no timers are running. */ if (tcp_timer_active(tp, TT_PERSIST)) { /* We enter in persists, set the flag appropriately */ bbr = (struct tcp_bbr *)tp->t_fb_ptr; bbr->rc_in_persist = 1; } tcp_timer_suspend(tp, TT_PERSIST); tcp_timer_suspend(tp, TT_REXMT); tcp_timer_suspend(tp, TT_KEEP); tcp_timer_suspend(tp, TT_DELACK); } static void bbr_google_mode_on(struct tcp_bbr *bbr) { bbr->rc_use_google = 1; bbr->rc_no_pacing = 0; bbr->r_ctl.bbr_google_discount = bbr_google_discount; bbr->r_use_policer = bbr_policer_detection_enabled; bbr->r_ctl.rc_probertt_int = (USECS_IN_SECOND * 10); bbr->bbr_use_rack_cheat = 0; bbr->r_ctl.rc_incr_tmrs = 0; bbr->r_ctl.rc_inc_tcp_oh = 0; bbr->r_ctl.rc_inc_ip_oh = 0; bbr->r_ctl.rc_inc_enet_oh = 0; reset_time(&bbr->r_ctl.rc_delrate, BBR_NUM_RTTS_FOR_GOOG_DEL_LIMIT); reset_time_small(&bbr->r_ctl.rc_rttprop, (11 * USECS_IN_SECOND)); tcp_bbr_tso_size_check(bbr, tcp_get_usecs(&bbr->rc_tv)); } static void bbr_google_mode_off(struct tcp_bbr *bbr) { bbr->rc_use_google = 0; bbr->r_ctl.bbr_google_discount = 0; bbr->no_pacing_until = bbr_no_pacing_until; bbr->r_use_policer = 0; if (bbr->no_pacing_until) bbr->rc_no_pacing = 1; else bbr->rc_no_pacing = 0; if (bbr_use_rack_resend_cheat) bbr->bbr_use_rack_cheat = 1; else bbr->bbr_use_rack_cheat = 0; if (bbr_incr_timers) bbr->r_ctl.rc_incr_tmrs = 1; else bbr->r_ctl.rc_incr_tmrs = 0; if (bbr_include_tcp_oh) bbr->r_ctl.rc_inc_tcp_oh = 1; else bbr->r_ctl.rc_inc_tcp_oh = 0; if (bbr_include_ip_oh) bbr->r_ctl.rc_inc_ip_oh = 1; else bbr->r_ctl.rc_inc_ip_oh = 0; if (bbr_include_enet_oh) bbr->r_ctl.rc_inc_enet_oh = 1; else bbr->r_ctl.rc_inc_enet_oh = 0; bbr->r_ctl.rc_probertt_int = bbr_rtt_probe_limit; reset_time(&bbr->r_ctl.rc_delrate, bbr_num_pktepo_for_del_limit); reset_time_small(&bbr->r_ctl.rc_rttprop, (bbr_filter_len_sec * USECS_IN_SECOND)); tcp_bbr_tso_size_check(bbr, tcp_get_usecs(&bbr->rc_tv)); } /* * Return 0 on success, non-zero on failure * which indicates the error (usually no memory). */ static int bbr_init(struct tcpcb *tp) { struct tcp_bbr *bbr = NULL; struct inpcb *inp; uint32_t cts; tp->t_fb_ptr = uma_zalloc(bbr_pcb_zone, (M_NOWAIT | M_ZERO)); if (tp->t_fb_ptr == NULL) { /* * We need to allocate memory but cant. The INP and INP_INFO * locks and they are recusive (happens during setup. So a * scheme to drop the locks fails :( * */ return (ENOMEM); } bbr = (struct tcp_bbr *)tp->t_fb_ptr; bbr->rtt_valid = 0; inp = tp->t_inpcb; inp->inp_flags2 |= INP_CANNOT_DO_ECN; inp->inp_flags2 |= INP_SUPPORTS_MBUFQ; TAILQ_INIT(&bbr->r_ctl.rc_map); TAILQ_INIT(&bbr->r_ctl.rc_free); TAILQ_INIT(&bbr->r_ctl.rc_tmap); bbr->rc_tp = tp; if (tp->t_inpcb) { bbr->rc_inp = tp->t_inpcb; } cts = tcp_get_usecs(&bbr->rc_tv); tp->t_acktime = 0; bbr->rc_allow_data_af_clo = bbr_ignore_data_after_close; bbr->r_ctl.rc_reorder_fade = bbr_reorder_fade; bbr->rc_tlp_threshold = bbr_tlp_thresh; bbr->r_ctl.rc_reorder_shift = bbr_reorder_thresh; bbr->r_ctl.rc_pkt_delay = bbr_pkt_delay; bbr->r_ctl.rc_min_to = bbr_min_to; bbr->rc_bbr_state = BBR_STATE_STARTUP; bbr->r_ctl.bbr_lost_at_state = 0; bbr->r_ctl.rc_lost_at_startup = 0; bbr->rc_all_timers_stopped = 0; bbr->r_ctl.rc_bbr_lastbtlbw = 0; bbr->r_ctl.rc_pkt_epoch_del = 0; bbr->r_ctl.rc_pkt_epoch = 0; bbr->r_ctl.rc_lowest_rtt = 0xffffffff; bbr->r_ctl.rc_bbr_hptsi_gain = bbr_high_gain; bbr->r_ctl.rc_bbr_cwnd_gain = bbr_high_gain; bbr->r_ctl.rc_went_idle_time = cts; bbr->rc_pacer_started = cts; bbr->r_ctl.rc_pkt_epoch_time = cts; bbr->r_ctl.rc_rcvtime = cts; bbr->r_ctl.rc_bbr_state_time = cts; bbr->r_ctl.rc_del_time = cts; bbr->r_ctl.rc_tlp_rxt_last_time = cts; bbr->r_ctl.last_in_probertt = cts; bbr->skip_gain = 0; bbr->gain_is_limited = 0; bbr->no_pacing_until = bbr_no_pacing_until; if (bbr->no_pacing_until) bbr->rc_no_pacing = 1; if (bbr_use_google_algo) { bbr->rc_no_pacing = 0; bbr->rc_use_google = 1; bbr->r_ctl.bbr_google_discount = bbr_google_discount; bbr->r_use_policer = bbr_policer_detection_enabled; } else { bbr->rc_use_google = 0; bbr->r_ctl.bbr_google_discount = 0; bbr->r_use_policer = 0; } if (bbr_ts_limiting) bbr->rc_use_ts_limit = 1; else bbr->rc_use_ts_limit = 0; if (bbr_ts_can_raise) bbr->ts_can_raise = 1; else bbr->ts_can_raise = 0; if (V_tcp_delack_enabled == 1) tp->t_delayed_ack = 2; else if (V_tcp_delack_enabled == 0) tp->t_delayed_ack = 0; else if (V_tcp_delack_enabled < 100) tp->t_delayed_ack = V_tcp_delack_enabled; else tp->t_delayed_ack = 2; if (bbr->rc_use_google == 0) bbr->r_ctl.rc_probertt_int = bbr_rtt_probe_limit; else bbr->r_ctl.rc_probertt_int = (USECS_IN_SECOND * 10); bbr->r_ctl.rc_min_rto_ms = bbr_rto_min_ms; bbr->rc_max_rto_sec = bbr_rto_max_sec; bbr->rc_init_win = bbr_def_init_win; if (tp->t_flags & TF_REQ_TSTMP) bbr->rc_last_options = TCP_TS_OVERHEAD; bbr->r_ctl.rc_pace_max_segs = tp->t_maxseg - bbr->rc_last_options; bbr->r_ctl.rc_high_rwnd = tp->snd_wnd; bbr->r_init_rtt = 1; counter_u64_add(bbr_flows_nohdwr_pacing, 1); if (bbr_allow_hdwr_pacing) bbr->bbr_hdw_pace_ena = 1; else bbr->bbr_hdw_pace_ena = 0; if (bbr_sends_full_iwnd) bbr->bbr_init_win_cheat = 1; else bbr->bbr_init_win_cheat = 0; bbr->r_ctl.bbr_utter_max = bbr_hptsi_utter_max; bbr->r_ctl.rc_drain_pg = bbr_drain_gain; bbr->r_ctl.rc_startup_pg = bbr_high_gain; bbr->rc_loss_exit = bbr_exit_startup_at_loss; bbr->r_ctl.bbr_rttprobe_gain_val = bbr_rttprobe_gain; bbr->r_ctl.bbr_hptsi_per_second = bbr_hptsi_per_second; bbr->r_ctl.bbr_hptsi_segments_delay_tar = bbr_hptsi_segments_delay_tar; bbr->r_ctl.bbr_hptsi_segments_max = bbr_hptsi_segments_max; bbr->r_ctl.bbr_hptsi_segments_floor = bbr_hptsi_segments_floor; bbr->r_ctl.bbr_hptsi_bytes_min = bbr_hptsi_bytes_min; bbr->r_ctl.bbr_cross_over = bbr_cross_over; bbr->r_ctl.rc_rtt_shrinks = cts; if (bbr->rc_use_google) { setup_time_filter(&bbr->r_ctl.rc_delrate, FILTER_TYPE_MAX, BBR_NUM_RTTS_FOR_GOOG_DEL_LIMIT); setup_time_filter_small(&bbr->r_ctl.rc_rttprop, FILTER_TYPE_MIN, (11 * USECS_IN_SECOND)); } else { setup_time_filter(&bbr->r_ctl.rc_delrate, FILTER_TYPE_MAX, bbr_num_pktepo_for_del_limit); setup_time_filter_small(&bbr->r_ctl.rc_rttprop, FILTER_TYPE_MIN, (bbr_filter_len_sec * USECS_IN_SECOND)); } bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_INIT, 0); if (bbr_uses_idle_restart) bbr->rc_use_idle_restart = 1; else bbr->rc_use_idle_restart = 0; bbr->r_ctl.rc_bbr_cur_del_rate = 0; bbr->r_ctl.rc_initial_hptsi_bw = bbr_initial_bw_bps; if (bbr_resends_use_tso) bbr->rc_resends_use_tso = 1; #ifdef NETFLIX_PEAKRATE tp->t_peakrate_thr = tp->t_maxpeakrate; #endif if (tp->snd_una != tp->snd_max) { /* Create a send map for the current outstanding data */ struct bbr_sendmap *rsm; rsm = bbr_alloc(bbr); if (rsm == NULL) { uma_zfree(bbr_pcb_zone, tp->t_fb_ptr); tp->t_fb_ptr = NULL; return (ENOMEM); } rsm->r_flags = BBR_OVERMAX; rsm->r_tim_lastsent[0] = cts; rsm->r_rtr_cnt = 1; rsm->r_rtr_bytes = 0; rsm->r_start = tp->snd_una; rsm->r_end = tp->snd_max; rsm->r_dupack = 0; rsm->r_delivered = bbr->r_ctl.rc_delivered; rsm->r_ts_valid = 0; rsm->r_del_ack_ts = tp->ts_recent; rsm->r_del_time = cts; if (bbr->r_ctl.r_app_limited_until) rsm->r_app_limited = 1; else rsm->r_app_limited = 0; TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_map, rsm, r_next); TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_tmap, rsm, r_tnext); rsm->r_in_tmap = 1; if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) rsm->r_bbr_state = bbr_state_val(bbr); else rsm->r_bbr_state = 8; } if (bbr_use_rack_resend_cheat && (bbr->rc_use_google == 0)) bbr->bbr_use_rack_cheat = 1; if (bbr_incr_timers && (bbr->rc_use_google == 0)) bbr->r_ctl.rc_incr_tmrs = 1; if (bbr_include_tcp_oh && (bbr->rc_use_google == 0)) bbr->r_ctl.rc_inc_tcp_oh = 1; if (bbr_include_ip_oh && (bbr->rc_use_google == 0)) bbr->r_ctl.rc_inc_ip_oh = 1; if (bbr_include_enet_oh && (bbr->rc_use_google == 0)) bbr->r_ctl.rc_inc_enet_oh = 1; bbr_log_type_statechange(bbr, cts, __LINE__); if (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_srtt)) { uint32_t rtt; rtt = (TICKS_2_USEC(tp->t_srtt) >> TCP_RTT_SHIFT); apply_filter_min_small(&bbr->r_ctl.rc_rttprop, rtt, cts); } /* announce the settings and state */ bbr_log_settings_change(bbr, BBR_RECOVERY_LOWRTT); tcp_bbr_tso_size_check(bbr, cts); /* * Now call the generic function to start a timer. This will place * the TCB on the hptsi wheel if a timer is needed with appropriate * flags. */ bbr_stop_all_timers(tp); bbr_start_hpts_timer(bbr, tp, cts, 5, 0, 0); return (0); } /* * Return 0 if we can accept the connection. Return * non-zero if we can't handle the connection. A EAGAIN * means you need to wait until the connection is up. * a EADDRNOTAVAIL means we can never handle the connection * (no SACK). */ static int bbr_handoff_ok(struct tcpcb *tp) { if ((tp->t_state == TCPS_CLOSED) || (tp->t_state == TCPS_LISTEN)) { /* Sure no problem though it may not stick */ return (0); } if ((tp->t_state == TCPS_SYN_SENT) || (tp->t_state == TCPS_SYN_RECEIVED)) { /* * We really don't know you have to get to ESTAB or beyond * to tell. */ return (EAGAIN); } if ((tp->t_flags & TF_SACK_PERMIT) || bbr_sack_not_required) { return (0); } /* * If we reach here we don't do SACK on this connection so we can * never do rack. */ return (EINVAL); } static void bbr_fini(struct tcpcb *tp, int32_t tcb_is_purged) { if (tp->t_fb_ptr) { uint32_t calc; struct tcp_bbr *bbr; struct bbr_sendmap *rsm; bbr = (struct tcp_bbr *)tp->t_fb_ptr; if (bbr->r_ctl.crte) tcp_rel_pacing_rate(bbr->r_ctl.crte, bbr->rc_tp); bbr_log_flowend(bbr); bbr->rc_tp = NULL; if (tp->t_inpcb) { /* Backout any flags2 we applied */ tp->t_inpcb->inp_flags2 &= ~INP_CANNOT_DO_ECN; tp->t_inpcb->inp_flags2 &= ~INP_SUPPORTS_MBUFQ; tp->t_inpcb->inp_flags2 &= ~INP_MBUF_QUEUE_READY; } if (bbr->bbr_hdrw_pacing) counter_u64_add(bbr_flows_whdwr_pacing, -1); else counter_u64_add(bbr_flows_nohdwr_pacing, -1); rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); while (rsm) { TAILQ_REMOVE(&bbr->r_ctl.rc_map, rsm, r_next); uma_zfree(bbr_zone, rsm); rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); } rsm = TAILQ_FIRST(&bbr->r_ctl.rc_free); while (rsm) { TAILQ_REMOVE(&bbr->r_ctl.rc_free, rsm, r_next); uma_zfree(bbr_zone, rsm); rsm = TAILQ_FIRST(&bbr->r_ctl.rc_free); } calc = bbr->r_ctl.rc_high_rwnd - bbr->r_ctl.rc_init_rwnd; if (calc > (bbr->r_ctl.rc_init_rwnd / 10)) BBR_STAT_INC(bbr_dynamic_rwnd); else BBR_STAT_INC(bbr_static_rwnd); bbr->r_ctl.rc_free_cnt = 0; uma_zfree(bbr_pcb_zone, tp->t_fb_ptr); tp->t_fb_ptr = NULL; } /* Make sure snd_nxt is correctly set */ tp->snd_nxt = tp->snd_max; } static void bbr_set_state(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t win) { switch (tp->t_state) { case TCPS_SYN_SENT: bbr->r_state = TCPS_SYN_SENT; bbr->r_substate = bbr_do_syn_sent; break; case TCPS_SYN_RECEIVED: bbr->r_state = TCPS_SYN_RECEIVED; bbr->r_substate = bbr_do_syn_recv; break; case TCPS_ESTABLISHED: bbr->r_ctl.rc_init_rwnd = max(win, bbr->rc_tp->snd_wnd); bbr->r_state = TCPS_ESTABLISHED; bbr->r_substate = bbr_do_established; break; case TCPS_CLOSE_WAIT: bbr->r_state = TCPS_CLOSE_WAIT; bbr->r_substate = bbr_do_close_wait; break; case TCPS_FIN_WAIT_1: bbr->r_state = TCPS_FIN_WAIT_1; bbr->r_substate = bbr_do_fin_wait_1; break; case TCPS_CLOSING: bbr->r_state = TCPS_CLOSING; bbr->r_substate = bbr_do_closing; break; case TCPS_LAST_ACK: bbr->r_state = TCPS_LAST_ACK; bbr->r_substate = bbr_do_lastack; break; case TCPS_FIN_WAIT_2: bbr->r_state = TCPS_FIN_WAIT_2; bbr->r_substate = bbr_do_fin_wait_2; break; case TCPS_LISTEN: case TCPS_CLOSED: case TCPS_TIME_WAIT: default: break; }; } static void bbr_substate_change(struct tcp_bbr *bbr, uint32_t cts, int32_t line, int dolog) { /* * Now what state are we going into now? Is there adjustments * needed? */ int32_t old_state, old_gain; old_state = bbr_state_val(bbr); old_gain = bbr->r_ctl.rc_bbr_hptsi_gain; if (bbr_state_val(bbr) == BBR_SUB_LEVEL1) { /* Save the lowest srtt we saw in our end of the sub-state */ bbr->rc_hit_state_1 = 0; if (bbr->r_ctl.bbr_smallest_srtt_this_state != 0xffffffff) bbr->r_ctl.bbr_smallest_srtt_state2 = bbr->r_ctl.bbr_smallest_srtt_this_state; } bbr->rc_bbr_substate++; if (bbr->rc_bbr_substate >= BBR_SUBSTATE_COUNT) { /* Cycle back to first state-> gain */ bbr->rc_bbr_substate = 0; } if (bbr_state_val(bbr) == BBR_SUB_GAIN) { /* * We enter the gain(5/4) cycle (possibly less if * shallow buffer detection is enabled) */ if (bbr->skip_gain) { /* * Hardware pacing has set our rate to * the max and limited our b/w just * do level i.e. no gain. */ bbr->r_ctl.rc_bbr_hptsi_gain = bbr_hptsi_gain[BBR_SUB_LEVEL1]; } else if (bbr->gain_is_limited && bbr->bbr_hdrw_pacing && bbr->r_ctl.crte) { /* * We can't gain above the hardware pacing * rate which is less than our rate + the gain * calculate the gain needed to reach the hardware * pacing rate.. */ uint64_t bw, rate, gain_calc; bw = bbr_get_bw(bbr); rate = bbr->r_ctl.crte->rate; if ((rate > bw) && (((bw * (uint64_t)bbr_hptsi_gain[BBR_SUB_GAIN]) / (uint64_t)BBR_UNIT) > rate)) { gain_calc = (rate * BBR_UNIT) / bw; if (gain_calc < BBR_UNIT) gain_calc = BBR_UNIT; bbr->r_ctl.rc_bbr_hptsi_gain = (uint16_t)gain_calc; } else { bbr->r_ctl.rc_bbr_hptsi_gain = bbr_hptsi_gain[BBR_SUB_GAIN]; } } else bbr->r_ctl.rc_bbr_hptsi_gain = bbr_hptsi_gain[BBR_SUB_GAIN]; if ((bbr->rc_use_google == 0) && (bbr_gain_to_target == 0)) { bbr->r_ctl.rc_bbr_state_atflight = cts; } else bbr->r_ctl.rc_bbr_state_atflight = 0; } else if (bbr_state_val(bbr) == BBR_SUB_DRAIN) { bbr->rc_hit_state_1 = 1; bbr->r_ctl.rc_exta_time_gd = 0; bbr->r_ctl.flightsize_at_drain = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if (bbr_state_drain_2_tar) { bbr->r_ctl.rc_bbr_state_atflight = 0; } else bbr->r_ctl.rc_bbr_state_atflight = cts; bbr->r_ctl.rc_bbr_hptsi_gain = bbr_hptsi_gain[BBR_SUB_DRAIN]; } else { /* All other cycles hit here 2-7 */ if ((old_state == BBR_SUB_DRAIN) && bbr->rc_hit_state_1) { if (bbr_sub_drain_slam_cwnd && (bbr->rc_use_google == 0) && (bbr->rc_tp->snd_cwnd < bbr->r_ctl.rc_saved_cwnd)) { bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_saved_cwnd; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } if ((cts - bbr->r_ctl.rc_bbr_state_time) > bbr_get_rtt(bbr, BBR_RTT_PROP)) bbr->r_ctl.rc_exta_time_gd += ((cts - bbr->r_ctl.rc_bbr_state_time) - bbr_get_rtt(bbr, BBR_RTT_PROP)); else bbr->r_ctl.rc_exta_time_gd = 0; if (bbr->r_ctl.rc_exta_time_gd) { bbr->r_ctl.rc_level_state_extra = bbr->r_ctl.rc_exta_time_gd; /* Now chop up the time for each state (div by 7) */ bbr->r_ctl.rc_level_state_extra /= 7; if (bbr_rand_ot && bbr->r_ctl.rc_level_state_extra) { /* Add a randomization */ bbr_randomize_extra_state_time(bbr); } } } bbr->r_ctl.rc_bbr_state_atflight = max(1, cts); bbr->r_ctl.rc_bbr_hptsi_gain = bbr_hptsi_gain[bbr_state_val(bbr)]; } if (bbr->rc_use_google) { bbr->r_ctl.rc_bbr_state_atflight = max(1, cts); } bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; bbr->r_ctl.rc_bbr_cwnd_gain = bbr_cwnd_gain; if (dolog) bbr_log_type_statechange(bbr, cts, line); if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { uint32_t time_in; time_in = cts - bbr->r_ctl.rc_bbr_state_time; if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) { counter_u64_add(bbr_state_time[(old_state + 5)], time_in); } else { counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } } bbr->r_ctl.bbr_smallest_srtt_this_state = 0xffffffff; bbr_set_state_target(bbr, __LINE__); if (bbr_sub_drain_slam_cwnd && (bbr->rc_use_google == 0) && (bbr_state_val(bbr) == BBR_SUB_DRAIN)) { /* Slam down the cwnd */ bbr->r_ctl.rc_saved_cwnd = bbr->rc_tp->snd_cwnd; bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; if (bbr_sub_drain_app_limit) { /* Go app limited if we are on a long drain */ bbr->r_ctl.r_app_limited_until = (bbr->r_ctl.rc_delivered + ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes))); } bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } if (bbr->rc_lt_use_bw) { /* In policed mode we clamp pacing_gain to BBR_UNIT */ bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; } /* Google changes TSO size every cycle */ if (bbr->rc_use_google) tcp_bbr_tso_size_check(bbr, cts); bbr->r_ctl.gain_epoch = cts; bbr->r_ctl.rc_bbr_state_time = cts; bbr->r_ctl.substate_pe = bbr->r_ctl.rc_pkt_epoch; } static void bbr_set_probebw_google_gains(struct tcp_bbr *bbr, uint32_t cts, uint32_t losses) { if ((bbr_state_val(bbr) == BBR_SUB_DRAIN) && (google_allow_early_out == 1) && (bbr->r_ctl.rc_flight_at_input <= bbr->r_ctl.rc_target_at_state)) { /* We have reached out target flight size possibly early */ goto change_state; } if (TSTMP_LT(cts, bbr->r_ctl.rc_bbr_state_time)) { return; } if ((cts - bbr->r_ctl.rc_bbr_state_time) < bbr_get_rtt(bbr, BBR_RTT_PROP)) { /* * Must be a rttProp movement forward before * we can change states. */ return; } if (bbr_state_val(bbr) == BBR_SUB_GAIN) { /* * The needed time has passed but for * the gain cycle extra rules apply: * 1) If we have seen loss, we exit * 2) If we have not reached the target * we stay in GAIN (gain-to-target). */ if (google_consider_lost && losses) goto change_state; if (bbr->r_ctl.rc_target_at_state > bbr->r_ctl.rc_flight_at_input) { return; } } change_state: /* For gain we must reach our target, all others last 1 rttProp */ bbr_substate_change(bbr, cts, __LINE__, 1); } static void bbr_set_probebw_gains(struct tcp_bbr *bbr, uint32_t cts, uint32_t losses) { uint32_t flight, bbr_cur_cycle_time; if (bbr->rc_use_google) { bbr_set_probebw_google_gains(bbr, cts, losses); return; } if (cts == 0) { /* * Never alow cts to be 0 we * do this so we can judge if * we have set a timestamp. */ cts = 1; } if (bbr_state_is_pkt_epoch) bbr_cur_cycle_time = bbr_get_rtt(bbr, BBR_RTT_PKTRTT); else bbr_cur_cycle_time = bbr_get_rtt(bbr, BBR_RTT_PROP); if (bbr->r_ctl.rc_bbr_state_atflight == 0) { if (bbr_state_val(bbr) == BBR_SUB_DRAIN) { flight = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if (bbr_sub_drain_slam_cwnd && bbr->rc_hit_state_1) { /* Keep it slam down */ if (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_target_at_state) { bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } if (bbr_sub_drain_app_limit) { /* Go app limited if we are on a long drain */ bbr->r_ctl.r_app_limited_until = (bbr->r_ctl.rc_delivered + flight); } } if (TSTMP_GT(cts, bbr->r_ctl.gain_epoch) && (((cts - bbr->r_ctl.gain_epoch) > bbr_get_rtt(bbr, BBR_RTT_PROP)) || (flight >= bbr->r_ctl.flightsize_at_drain))) { /* * Still here after the same time as * the gain. We need to drain harder * for the next srtt. Reduce by a set amount * the gain drop is capped at DRAIN states * value (88). */ bbr->r_ctl.flightsize_at_drain = flight; if (bbr_drain_drop_mul && bbr_drain_drop_div && (bbr_drain_drop_mul < bbr_drain_drop_div)) { /* Use your specific drop value (def 4/5 = 20%) */ bbr->r_ctl.rc_bbr_hptsi_gain *= bbr_drain_drop_mul; bbr->r_ctl.rc_bbr_hptsi_gain /= bbr_drain_drop_div; } else { /* You get drop of 20% */ bbr->r_ctl.rc_bbr_hptsi_gain *= 4; bbr->r_ctl.rc_bbr_hptsi_gain /= 5; } if (bbr->r_ctl.rc_bbr_hptsi_gain <= bbr_drain_floor) { /* Reduce our gain again to the bottom */ bbr->r_ctl.rc_bbr_hptsi_gain = max(bbr_drain_floor, 1); } bbr_log_exit_gain(bbr, cts, 4); /* * Extend out so we wait another * epoch before dropping again. */ bbr->r_ctl.gain_epoch = cts; } if (flight <= bbr->r_ctl.rc_target_at_state) { if (bbr_sub_drain_slam_cwnd && (bbr->rc_use_google == 0) && (bbr->rc_tp->snd_cwnd < bbr->r_ctl.rc_saved_cwnd)) { bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_saved_cwnd; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } bbr->r_ctl.rc_bbr_state_atflight = max(cts, 1); bbr_log_exit_gain(bbr, cts, 3); } } else { /* Its a gain */ if (bbr->r_ctl.rc_lost > bbr->r_ctl.bbr_lost_at_state) { bbr->r_ctl.rc_bbr_state_atflight = max(cts, 1); goto change_state; } if ((ctf_outstanding(bbr->rc_tp) >= bbr->r_ctl.rc_target_at_state) || ((ctf_outstanding(bbr->rc_tp) + bbr->rc_tp->t_maxseg - 1) >= bbr->rc_tp->snd_wnd)) { bbr->r_ctl.rc_bbr_state_atflight = max(cts, 1); bbr_log_exit_gain(bbr, cts, 2); } } /** * We fall through and return always one of two things has * occured. * 1) We are still not at target * * 2) We reached the target and set rc_bbr_state_atflight * which means we no longer hit this block * next time we are called. */ return; } change_state: if (TSTMP_LT(cts, bbr->r_ctl.rc_bbr_state_time)) return; if ((cts - bbr->r_ctl.rc_bbr_state_time) < bbr_cur_cycle_time) { /* Less than a full time-period has passed */ return; } if (bbr->r_ctl.rc_level_state_extra && (bbr_state_val(bbr) > BBR_SUB_DRAIN) && ((cts - bbr->r_ctl.rc_bbr_state_time) < (bbr_cur_cycle_time + bbr->r_ctl.rc_level_state_extra))) { /* Less than a full time-period + extra has passed */ return; } if (bbr_gain_gets_extra_too && bbr->r_ctl.rc_level_state_extra && (bbr_state_val(bbr) == BBR_SUB_GAIN) && ((cts - bbr->r_ctl.rc_bbr_state_time) < (bbr_cur_cycle_time + bbr->r_ctl.rc_level_state_extra))) { /* Less than a full time-period + extra has passed */ return; } bbr_substate_change(bbr, cts, __LINE__, 1); } static uint32_t bbr_get_a_state_target(struct tcp_bbr *bbr, uint32_t gain) { uint32_t mss, tar; if (bbr->rc_use_google) { /* Google just uses the cwnd target */ tar = bbr_get_target_cwnd(bbr, bbr_get_bw(bbr), gain); } else { mss = min((bbr->rc_tp->t_maxseg - bbr->rc_last_options), bbr->r_ctl.rc_pace_max_segs); /* Get the base cwnd with gain rounded to a mss */ tar = roundup(bbr_get_raw_target_cwnd(bbr, bbr_get_bw(bbr), gain), mss); /* Make sure it is within our min */ if (tar < get_min_cwnd(bbr)) return (get_min_cwnd(bbr)); } return (tar); } static void bbr_set_state_target(struct tcp_bbr *bbr, int line) { uint32_t tar, meth; if ((bbr->rc_bbr_state == BBR_STATE_PROBE_RTT) && ((bbr->r_ctl.bbr_rttprobe_gain_val == 0) || bbr->rc_use_google)) { /* Special case using old probe-rtt method */ tar = bbr_rtt_probe_cwndtarg * (bbr->rc_tp->t_maxseg - bbr->rc_last_options); meth = 1; } else { /* Non-probe-rtt case and reduced probe-rtt */ if ((bbr->rc_bbr_state == BBR_STATE_PROBE_BW) && (bbr->r_ctl.rc_bbr_hptsi_gain > BBR_UNIT)) { /* For gain cycle we use the hptsi gain */ tar = bbr_get_a_state_target(bbr, bbr->r_ctl.rc_bbr_hptsi_gain); meth = 2; } else if ((bbr_target_is_bbunit) || bbr->rc_use_google) { /* * If configured, or for google all other states * get BBR_UNIT. */ tar = bbr_get_a_state_target(bbr, BBR_UNIT); meth = 3; } else { /* * Or we set a target based on the pacing gain * for non-google mode and default (non-configured). * Note we don't set a target goal below drain (192). */ if (bbr->r_ctl.rc_bbr_hptsi_gain < bbr_hptsi_gain[BBR_SUB_DRAIN]) { tar = bbr_get_a_state_target(bbr, bbr_hptsi_gain[BBR_SUB_DRAIN]); meth = 4; } else { tar = bbr_get_a_state_target(bbr, bbr->r_ctl.rc_bbr_hptsi_gain); meth = 5; } } } bbr_log_set_of_state_target(bbr, tar, line, meth); bbr->r_ctl.rc_target_at_state = tar; } static void bbr_enter_probe_rtt(struct tcp_bbr *bbr, uint32_t cts, int32_t line) { /* Change to probe_rtt */ uint32_t time_in; bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; bbr->r_ctl.flightsize_at_drain = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); bbr->r_ctl.r_app_limited_until = (bbr->r_ctl.flightsize_at_drain + bbr->r_ctl.rc_delivered); /* Setup so we force feed the filter */ if (bbr->rc_use_google || bbr_probertt_sets_rtt) bbr->rc_prtt_set_ts = 1; if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { time_in = cts - bbr->r_ctl.rc_bbr_state_time; counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_ENTERPROBE, 0); bbr->r_ctl.rc_rtt_shrinks = cts; bbr->r_ctl.last_in_probertt = cts; bbr->r_ctl.rc_probertt_srttchktim = cts; bbr->r_ctl.rc_bbr_state_time = cts; bbr->rc_bbr_state = BBR_STATE_PROBE_RTT; /* We need to force the filter to update */ if ((bbr_sub_drain_slam_cwnd) && bbr->rc_hit_state_1 && (bbr->rc_use_google == 0) && (bbr_state_val(bbr) == BBR_SUB_DRAIN)) { if (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_saved_cwnd) bbr->r_ctl.rc_saved_cwnd = bbr->rc_tp->snd_cwnd; } else bbr->r_ctl.rc_saved_cwnd = bbr->rc_tp->snd_cwnd; /* Update the lost */ bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; if ((bbr->r_ctl.bbr_rttprobe_gain_val == 0) || bbr->rc_use_google){ /* Set to the non-configurable default of 4 (PROBE_RTT_MIN) */ bbr->rc_tp->snd_cwnd = bbr_rtt_probe_cwndtarg * (bbr->rc_tp->t_maxseg - bbr->rc_last_options); bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; bbr->r_ctl.rc_bbr_cwnd_gain = BBR_UNIT; bbr_log_set_of_state_target(bbr, bbr->rc_tp->snd_cwnd, __LINE__, 6); bbr->r_ctl.rc_target_at_state = bbr->rc_tp->snd_cwnd; } else { /* * We bring it down slowly by using a hptsi gain that is * probably 75%. This will slowly float down our outstanding * without tampering with the cwnd. */ bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.bbr_rttprobe_gain_val; bbr->r_ctl.rc_bbr_cwnd_gain = BBR_UNIT; bbr_set_state_target(bbr, __LINE__); if (bbr_prtt_slam_cwnd && (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_target_at_state)) { bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } } if (ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) <= bbr->r_ctl.rc_target_at_state) { /* We are at target */ bbr->r_ctl.rc_bbr_enters_probertt = cts; } else { /* We need to come down to reach target before our time begins */ bbr->r_ctl.rc_bbr_enters_probertt = 0; } bbr->r_ctl.rc_pe_of_prtt = bbr->r_ctl.rc_pkt_epoch; BBR_STAT_INC(bbr_enter_probertt); bbr_log_exit_gain(bbr, cts, 0); bbr_log_type_statechange(bbr, cts, line); } static void bbr_check_probe_rtt_limits(struct tcp_bbr *bbr, uint32_t cts) { /* * Sanity check on probe-rtt intervals. * In crazy situations where we are competing * against new-reno flows with huge buffers * our rtt-prop interval could come to dominate * things if we can't get through a full set * of cycles, we need to adjust it. */ if (bbr_can_adjust_probertt && (bbr->rc_use_google == 0)) { uint16_t val = 0; uint32_t cur_rttp, fval, newval, baseval; /* Are we to small and go into probe-rtt to often? */ baseval = (bbr_get_rtt(bbr, BBR_RTT_PROP) * (BBR_SUBSTATE_COUNT + 1)); cur_rttp = roundup(baseval, USECS_IN_SECOND); fval = bbr_filter_len_sec * USECS_IN_SECOND; if (bbr_is_ratio == 0) { if (fval > bbr_rtt_probe_limit) newval = cur_rttp + (fval - bbr_rtt_probe_limit); else newval = cur_rttp; } else { int mul; mul = fval / bbr_rtt_probe_limit; newval = cur_rttp * mul; } if (cur_rttp > bbr->r_ctl.rc_probertt_int) { bbr->r_ctl.rc_probertt_int = cur_rttp; reset_time_small(&bbr->r_ctl.rc_rttprop, newval); val = 1; } else { /* * No adjustments were made * do we need to shrink it? */ if (bbr->r_ctl.rc_probertt_int > bbr_rtt_probe_limit) { if (cur_rttp <= bbr_rtt_probe_limit) { /* * Things have calmed down lets * shrink all the way to default */ bbr->r_ctl.rc_probertt_int = bbr_rtt_probe_limit; reset_time_small(&bbr->r_ctl.rc_rttprop, (bbr_filter_len_sec * USECS_IN_SECOND)); cur_rttp = bbr_rtt_probe_limit; newval = (bbr_filter_len_sec * USECS_IN_SECOND); val = 2; } else { /* * Well does some adjustment make sense? */ if (cur_rttp < bbr->r_ctl.rc_probertt_int) { /* We can reduce interval time some */ bbr->r_ctl.rc_probertt_int = cur_rttp; reset_time_small(&bbr->r_ctl.rc_rttprop, newval); val = 3; } } } } if (val) bbr_log_rtt_shrinks(bbr, cts, cur_rttp, newval, __LINE__, BBR_RTTS_RESETS_VALUES, val); } } static void bbr_exit_probe_rtt(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t cts) { /* Exit probe-rtt */ if (tp->snd_cwnd < bbr->r_ctl.rc_saved_cwnd) { tp->snd_cwnd = bbr->r_ctl.rc_saved_cwnd; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } bbr_log_exit_gain(bbr, cts, 1); bbr->rc_hit_state_1 = 0; bbr->r_ctl.rc_rtt_shrinks = cts; bbr->r_ctl.last_in_probertt = cts; bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_RTTPROBE, 0); bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; bbr->r_ctl.r_app_limited_until = (ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + bbr->r_ctl.rc_delivered); if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { uint32_t time_in; time_in = cts - bbr->r_ctl.rc_bbr_state_time; counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } if (bbr->rc_filled_pipe) { /* Switch to probe_bw */ bbr->rc_bbr_state = BBR_STATE_PROBE_BW; bbr->rc_bbr_substate = bbr_pick_probebw_substate(bbr, cts); bbr->r_ctl.rc_bbr_cwnd_gain = bbr_cwnd_gain; bbr_substate_change(bbr, cts, __LINE__, 0); bbr_log_type_statechange(bbr, cts, __LINE__); } else { /* Back to startup */ bbr->rc_bbr_state = BBR_STATE_STARTUP; bbr->r_ctl.rc_bbr_state_time = cts; /* * We don't want to give a complete free 3 * measurements until we exit, so we use * the number of pe's we were in probe-rtt * to add to the startup_epoch. That way * we will still retain the old state. */ bbr->r_ctl.rc_bbr_last_startup_epoch += (bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_pe_of_prtt); bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; /* Make sure to use the lower pg when shifting back in */ if (bbr->r_ctl.rc_lost && bbr_use_lower_gain_in_startup && (bbr->rc_use_google == 0)) bbr->r_ctl.rc_bbr_hptsi_gain = bbr_startup_lower; else bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.rc_startup_pg; bbr->r_ctl.rc_bbr_cwnd_gain = bbr->r_ctl.rc_startup_pg; /* Probably not needed but set it anyway */ bbr_set_state_target(bbr, __LINE__); bbr_log_type_statechange(bbr, cts, __LINE__); bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 0); } bbr_check_probe_rtt_limits(bbr, cts); } static int32_t inline bbr_should_enter_probe_rtt(struct tcp_bbr *bbr, uint32_t cts) { if ((bbr->rc_past_init_win == 1) && (bbr->rc_in_persist == 0) && (bbr_calc_time(cts, bbr->r_ctl.rc_rtt_shrinks) >= bbr->r_ctl.rc_probertt_int)) { return (1); } if (bbr_can_force_probertt && (bbr->rc_in_persist == 0) && (TSTMP_GT(cts, bbr->r_ctl.last_in_probertt)) && ((cts - bbr->r_ctl.last_in_probertt) > bbr->r_ctl.rc_probertt_int)) { return (1); } return (0); } static int32_t bbr_google_startup(struct tcp_bbr *bbr, uint32_t cts, int32_t pkt_epoch) { uint64_t btlbw, gain; if (pkt_epoch == 0) { /* * Need to be on a pkt-epoch to continue. */ return (0); } btlbw = bbr_get_full_bw(bbr); gain = ((bbr->r_ctl.rc_bbr_lastbtlbw * (uint64_t)bbr_start_exit) / (uint64_t)100) + bbr->r_ctl.rc_bbr_lastbtlbw; if (btlbw >= gain) { bbr->r_ctl.rc_bbr_last_startup_epoch = bbr->r_ctl.rc_pkt_epoch; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 3); bbr->r_ctl.rc_bbr_lastbtlbw = btlbw; } if ((bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_bbr_last_startup_epoch) >= BBR_STARTUP_EPOCHS) return (1); bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 8); return(0); } static int32_t inline bbr_state_startup(struct tcp_bbr *bbr, uint32_t cts, int32_t epoch, int32_t pkt_epoch) { /* Have we gained 25% in the last 3 packet based epoch's? */ uint64_t btlbw, gain; int do_exit; int delta, rtt_gain; if ((bbr->rc_tp->snd_una == bbr->rc_tp->snd_max) && (bbr_calc_time(cts, bbr->r_ctl.rc_went_idle_time) >= bbr_rtt_probe_time)) { /* * This qualifies as a RTT_PROBE session since we drop the * data outstanding to nothing and waited more than * bbr_rtt_probe_time. */ bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_WASIDLE, 0); bbr_set_reduced_rtt(bbr, cts, __LINE__); } if (bbr_should_enter_probe_rtt(bbr, cts)) { bbr_enter_probe_rtt(bbr, cts, __LINE__); return (0); } if (bbr->rc_use_google) return (bbr_google_startup(bbr, cts, pkt_epoch)); if ((bbr->r_ctl.rc_lost > bbr->r_ctl.rc_lost_at_startup) && (bbr_use_lower_gain_in_startup)) { /* Drop to a lower gain 1.5 x since we saw loss */ bbr->r_ctl.rc_bbr_hptsi_gain = bbr_startup_lower; } if (pkt_epoch == 0) { /* * Need to be on a pkt-epoch to continue. */ return (0); } if (bbr_rtt_gain_thresh) { /* * Do we allow a flow to stay * in startup with no loss and no * gain in rtt over a set threshold? */ if (bbr->r_ctl.rc_pkt_epoch_rtt && bbr->r_ctl.startup_last_srtt && (bbr->r_ctl.rc_pkt_epoch_rtt > bbr->r_ctl.startup_last_srtt)) { delta = bbr->r_ctl.rc_pkt_epoch_rtt - bbr->r_ctl.startup_last_srtt; rtt_gain = (delta * 100) / bbr->r_ctl.startup_last_srtt; } else rtt_gain = 0; if ((bbr->r_ctl.startup_last_srtt == 0) || (bbr->r_ctl.rc_pkt_epoch_rtt < bbr->r_ctl.startup_last_srtt)) /* First time or new lower value */ bbr->r_ctl.startup_last_srtt = bbr->r_ctl.rc_pkt_epoch_rtt; if ((bbr->r_ctl.rc_lost == 0) && (rtt_gain < bbr_rtt_gain_thresh)) { /* * No loss, and we are under * our gain threhold for * increasing RTT. */ if (bbr->r_ctl.rc_bbr_last_startup_epoch < bbr->r_ctl.rc_pkt_epoch) bbr->r_ctl.rc_bbr_last_startup_epoch++; bbr_log_startup_event(bbr, cts, rtt_gain, delta, bbr->r_ctl.startup_last_srtt, 10); return (0); } } if ((bbr->r_ctl.r_measurement_count == bbr->r_ctl.last_startup_measure) && (bbr->r_ctl.rc_lost_at_startup == bbr->r_ctl.rc_lost) && (!IN_RECOVERY(bbr->rc_tp->t_flags))) { /* * We only assess if we have a new measurment when * we have no loss and are not in recovery. * Drag up by one our last_startup epoch so we will hold * the number of non-gain we have already accumulated. */ if (bbr->r_ctl.rc_bbr_last_startup_epoch < bbr->r_ctl.rc_pkt_epoch) bbr->r_ctl.rc_bbr_last_startup_epoch++; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 9); return (0); } /* Case where we reduced the lost (bad retransmit) */ if (bbr->r_ctl.rc_lost_at_startup > bbr->r_ctl.rc_lost) bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; bbr->r_ctl.last_startup_measure = bbr->r_ctl.r_measurement_count; btlbw = bbr_get_full_bw(bbr); if (bbr->r_ctl.rc_bbr_hptsi_gain == bbr_startup_lower) gain = ((bbr->r_ctl.rc_bbr_lastbtlbw * (uint64_t)bbr_low_start_exit) / (uint64_t)100) + bbr->r_ctl.rc_bbr_lastbtlbw; else gain = ((bbr->r_ctl.rc_bbr_lastbtlbw * (uint64_t)bbr_start_exit) / (uint64_t)100) + bbr->r_ctl.rc_bbr_lastbtlbw; do_exit = 0; if (btlbw > bbr->r_ctl.rc_bbr_lastbtlbw) bbr->r_ctl.rc_bbr_lastbtlbw = btlbw; if (btlbw >= gain) { bbr->r_ctl.rc_bbr_last_startup_epoch = bbr->r_ctl.rc_pkt_epoch; /* Update the lost so we won't exit in next set of tests */ bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 3); } if ((bbr->rc_loss_exit && (bbr->r_ctl.rc_lost > bbr->r_ctl.rc_lost_at_startup) && (bbr->r_ctl.rc_pkt_epoch_loss_rate > bbr_startup_loss_thresh)) && ((bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_bbr_last_startup_epoch) >= BBR_STARTUP_EPOCHS)) { /* * If we had no gain, we had loss and that loss was above * our threshould, the rwnd is not constrained, and we have * had at least 3 packet epochs exit. Note that this is * switched off by sysctl. Google does not do this by the * way. */ if ((ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + (2 * max(bbr->r_ctl.rc_pace_max_segs, bbr->rc_tp->t_maxseg))) <= bbr->rc_tp->snd_wnd) { do_exit = 1; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 4); } else { /* Just record an updated loss value */ bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 5); } } else bbr->r_ctl.rc_lost_at_startup = bbr->r_ctl.rc_lost; if (((bbr->r_ctl.rc_pkt_epoch - bbr->r_ctl.rc_bbr_last_startup_epoch) >= BBR_STARTUP_EPOCHS) || do_exit) { /* Return 1 to exit the startup state. */ return (1); } /* Stay in startup */ bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 8); return (0); } static void bbr_state_change(struct tcp_bbr *bbr, uint32_t cts, int32_t epoch, int32_t pkt_epoch, uint32_t losses) { /* * A tick occured in the rtt epoch do we need to do anything? */ #ifdef BBR_INVARIANTS if ((bbr->rc_bbr_state != BBR_STATE_STARTUP) && (bbr->rc_bbr_state != BBR_STATE_DRAIN) && (bbr->rc_bbr_state != BBR_STATE_PROBE_RTT) && (bbr->rc_bbr_state != BBR_STATE_IDLE_EXIT) && (bbr->rc_bbr_state != BBR_STATE_PROBE_BW)) { /* Debug code? */ panic("Unknown BBR state %d?\n", bbr->rc_bbr_state); } #endif if (bbr->rc_bbr_state == BBR_STATE_STARTUP) { /* Do we exit the startup state? */ if (bbr_state_startup(bbr, cts, epoch, pkt_epoch)) { uint32_t time_in; bbr_log_startup_event(bbr, cts, bbr->r_ctl.rc_bbr_last_startup_epoch, bbr->r_ctl.rc_lost_at_startup, bbr_start_exit, 6); bbr->rc_filled_pipe = 1; bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { time_in = cts - bbr->r_ctl.rc_bbr_state_time; counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } else time_in = 0; if (bbr->rc_no_pacing) bbr->rc_no_pacing = 0; bbr->r_ctl.rc_bbr_state_time = cts; bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.rc_drain_pg; bbr->rc_bbr_state = BBR_STATE_DRAIN; bbr_set_state_target(bbr, __LINE__); if ((bbr->rc_use_google == 0) && bbr_slam_cwnd_in_main_drain) { /* Here we don't have to worry about probe-rtt */ bbr->r_ctl.rc_saved_cwnd = bbr->rc_tp->snd_cwnd; bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } bbr->r_ctl.rc_bbr_cwnd_gain = bbr_high_gain; bbr_log_type_statechange(bbr, cts, __LINE__); if (ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) <= bbr->r_ctl.rc_target_at_state) { /* * Switch to probe_bw if we are already * there */ bbr->rc_bbr_substate = bbr_pick_probebw_substate(bbr, cts); bbr_substate_change(bbr, cts, __LINE__, 0); bbr->rc_bbr_state = BBR_STATE_PROBE_BW; bbr_log_type_statechange(bbr, cts, __LINE__); } } } else if (bbr->rc_bbr_state == BBR_STATE_IDLE_EXIT) { uint32_t inflight; struct tcpcb *tp; tp = bbr->rc_tp; inflight = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if (inflight >= bbr->r_ctl.rc_target_at_state) { /* We have reached a flight of the cwnd target */ bbr->rc_bbr_state = BBR_STATE_PROBE_BW; bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; bbr->r_ctl.rc_bbr_cwnd_gain = BBR_UNIT; bbr_set_state_target(bbr, __LINE__); /* * Rig it so we don't do anything crazy and * start fresh with a new randomization. */ bbr->r_ctl.bbr_smallest_srtt_this_state = 0xffffffff; bbr->rc_bbr_substate = BBR_SUB_LEVEL6; bbr_substate_change(bbr, cts, __LINE__, 1); } } else if (bbr->rc_bbr_state == BBR_STATE_DRAIN) { /* Has in-flight reached the bdp (or less)? */ uint32_t inflight; struct tcpcb *tp; tp = bbr->rc_tp; inflight = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if ((bbr->rc_use_google == 0) && bbr_slam_cwnd_in_main_drain && (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_target_at_state)) { /* * Here we don't have to worry about probe-rtt * re-slam it, but keep it slammed down. */ bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } if (inflight <= bbr->r_ctl.rc_target_at_state) { /* We have drained */ bbr->rc_bbr_state = BBR_STATE_PROBE_BW; bbr->r_ctl.bbr_lost_at_state = bbr->r_ctl.rc_lost; if (SEQ_GT(cts, bbr->r_ctl.rc_bbr_state_time)) { uint32_t time_in; time_in = cts - bbr->r_ctl.rc_bbr_state_time; counter_u64_add(bbr_state_time[bbr->rc_bbr_state], time_in); } if ((bbr->rc_use_google == 0) && bbr_slam_cwnd_in_main_drain && (tp->snd_cwnd < bbr->r_ctl.rc_saved_cwnd)) { /* Restore the cwnd */ tp->snd_cwnd = bbr->r_ctl.rc_saved_cwnd; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } /* Setup probe-rtt has being done now RRS-HERE */ bbr->r_ctl.rc_rtt_shrinks = cts; bbr->r_ctl.last_in_probertt = cts; bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_LEAVE_DRAIN, 0); /* Randomly pick a sub-state */ bbr->rc_bbr_substate = bbr_pick_probebw_substate(bbr, cts); bbr_substate_change(bbr, cts, __LINE__, 0); bbr_log_type_statechange(bbr, cts, __LINE__); } } else if (bbr->rc_bbr_state == BBR_STATE_PROBE_RTT) { uint32_t flight; flight = ctf_flight_size(bbr->rc_tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); bbr->r_ctl.r_app_limited_until = (flight + bbr->r_ctl.rc_delivered); if (((bbr->r_ctl.bbr_rttprobe_gain_val == 0) || bbr->rc_use_google) && (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_target_at_state)) { /* * We must keep cwnd at the desired MSS. */ bbr->rc_tp->snd_cwnd = bbr_rtt_probe_cwndtarg * (bbr->rc_tp->t_maxseg - bbr->rc_last_options); bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } else if ((bbr_prtt_slam_cwnd) && (bbr->rc_tp->snd_cwnd > bbr->r_ctl.rc_target_at_state)) { /* Re-slam it */ bbr->rc_tp->snd_cwnd = bbr->r_ctl.rc_target_at_state; bbr_log_type_cwndupd(bbr, 0, 0, 0, 12, 0, 0, __LINE__); } if (bbr->r_ctl.rc_bbr_enters_probertt == 0) { /* Has outstanding reached our target? */ if (flight <= bbr->r_ctl.rc_target_at_state) { bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_REACHTAR, 0); bbr->r_ctl.rc_bbr_enters_probertt = cts; /* If time is exactly 0, be 1usec off */ if (bbr->r_ctl.rc_bbr_enters_probertt == 0) bbr->r_ctl.rc_bbr_enters_probertt = 1; if (bbr->rc_use_google == 0) { /* * Restore any lowering that as occured to * reach here */ if (bbr->r_ctl.bbr_rttprobe_gain_val) bbr->r_ctl.rc_bbr_hptsi_gain = bbr->r_ctl.bbr_rttprobe_gain_val; else bbr->r_ctl.rc_bbr_hptsi_gain = BBR_UNIT; } } if ((bbr->r_ctl.rc_bbr_enters_probertt == 0) && (bbr->rc_use_google == 0) && bbr->r_ctl.bbr_rttprobe_gain_val && (((cts - bbr->r_ctl.rc_probertt_srttchktim) > bbr_get_rtt(bbr, bbr_drain_rtt)) || (flight >= bbr->r_ctl.flightsize_at_drain))) { /* * We have doddled with our current hptsi * gain an srtt and have still not made it * to target, or we have increased our flight. * Lets reduce the gain by xx% * flooring the reduce at DRAIN (based on * mul/div) */ int red; bbr->r_ctl.flightsize_at_drain = flight; bbr->r_ctl.rc_probertt_srttchktim = cts; red = max((bbr->r_ctl.bbr_rttprobe_gain_val / 10), 1); if ((bbr->r_ctl.rc_bbr_hptsi_gain - red) > max(bbr_drain_floor, 1)) { /* Reduce our gain again */ bbr->r_ctl.rc_bbr_hptsi_gain -= red; bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_SHRINK_PG, 0); } else if (bbr->r_ctl.rc_bbr_hptsi_gain > max(bbr_drain_floor, 1)) { /* one more chance before we give up */ bbr->r_ctl.rc_bbr_hptsi_gain = max(bbr_drain_floor, 1); bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_SHRINK_PG_FINAL, 0); } else { /* At the very bottom */ bbr->r_ctl.rc_bbr_hptsi_gain = max((bbr_drain_floor-1), 1); } } } if (bbr->r_ctl.rc_bbr_enters_probertt && (TSTMP_GT(cts, bbr->r_ctl.rc_bbr_enters_probertt)) && ((cts - bbr->r_ctl.rc_bbr_enters_probertt) >= bbr_rtt_probe_time)) { /* Time to exit probe RTT normally */ bbr_exit_probe_rtt(bbr->rc_tp, bbr, cts); } } else if (bbr->rc_bbr_state == BBR_STATE_PROBE_BW) { if ((bbr->rc_tp->snd_una == bbr->rc_tp->snd_max) && (bbr_calc_time(cts, bbr->r_ctl.rc_went_idle_time) >= bbr_rtt_probe_time)) { /* * This qualifies as a RTT_PROBE session since we * drop the data outstanding to nothing and waited * more than bbr_rtt_probe_time. */ bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_WASIDLE, 0); bbr_set_reduced_rtt(bbr, cts, __LINE__); } if (bbr_should_enter_probe_rtt(bbr, cts)) { bbr_enter_probe_rtt(bbr, cts, __LINE__); } else { bbr_set_probebw_gains(bbr, cts, losses); } } } static void bbr_check_bbr_for_state(struct tcp_bbr *bbr, uint32_t cts, int32_t line, uint32_t losses) { int32_t epoch = 0; if ((cts - bbr->r_ctl.rc_rcv_epoch_start) >= bbr_get_rtt(bbr, BBR_RTT_PROP)) { bbr_set_epoch(bbr, cts, line); /* At each epoch doe lt bw sampling */ epoch = 1; } bbr_state_change(bbr, cts, epoch, bbr->rc_is_pkt_epoch_now, losses); } static int bbr_do_segment_nounlock(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int32_t drop_hdrlen, int32_t tlen, uint8_t iptos, int32_t nxt_pkt, struct timeval *tv) { int32_t thflags, retval; uint32_t cts, lcts; uint32_t tiwin; struct tcpopt to; struct tcp_bbr *bbr; struct bbr_sendmap *rsm; struct timeval ltv; int32_t did_out = 0; int32_t in_recovery; uint16_t nsegs; int32_t prev_state; uint32_t lost; nsegs = max(1, m->m_pkthdr.lro_nsegs); bbr = (struct tcp_bbr *)tp->t_fb_ptr; /* add in our stats */ kern_prefetch(bbr, &prev_state); prev_state = 0; thflags = th->th_flags; /* * If this is either a state-changing packet or current state isn't * established, we require a write lock on tcbinfo. Otherwise, we * allow the tcbinfo to be in either alocked or unlocked, as the * caller may have unnecessarily acquired a write lock due to a * race. */ INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", __func__)); tp->t_rcvtime = ticks; /* * Unscale the window into a 32-bit value. For the SYN_SENT state * the scale is zero. */ tiwin = th->th_win << tp->snd_scale; #ifdef STATS stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); #endif /* * Parse options on any incoming segment. */ tcp_dooptions(&to, (u_char *)(th + 1), (th->th_off << 2) - sizeof(struct tcphdr), (thflags & TH_SYN) ? TO_SYN : 0); if (m->m_flags & M_TSTMP) { /* Prefer the hardware timestamp if present */ struct timespec ts; mbuf_tstmp2timespec(m, &ts); bbr->rc_tv.tv_sec = ts.tv_sec; bbr->rc_tv.tv_usec = ts.tv_nsec / 1000; bbr->r_ctl.rc_rcvtime = cts = tcp_tv_to_usectick(&bbr->rc_tv); } else if (m->m_flags & M_TSTMP_LRO) { /* Next the arrival timestamp */ struct timespec ts; mbuf_tstmp2timespec(m, &ts); bbr->rc_tv.tv_sec = ts.tv_sec; bbr->rc_tv.tv_usec = ts.tv_nsec / 1000; bbr->r_ctl.rc_rcvtime = cts = tcp_tv_to_usectick(&bbr->rc_tv); } else { /* * Ok just get the current time. */ bbr->r_ctl.rc_rcvtime = lcts = cts = tcp_get_usecs(&bbr->rc_tv); } /* * If echoed timestamp is later than the current time, fall back to * non RFC1323 RTT calculation. Normalize timestamp if syncookies * were used when this connection was established. */ if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { to.to_tsecr -= tp->ts_offset; if (TSTMP_GT(to.to_tsecr, tcp_tv_to_mssectick(&bbr->rc_tv))) to.to_tsecr = 0; } /* * If its the first time in we need to take care of options and * verify we can do SACK for rack! */ if (bbr->r_state == 0) { /* * Process options only when we get SYN/ACK back. The SYN * case for incoming connections is handled in tcp_syncache. * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. XXX * this is traditional behavior, may need to be cleaned up. */ if (bbr->rc_inp == NULL) { bbr->rc_inp = tp->t_inpcb; } /* * We need to init rc_inp here since its not init'd when * bbr_init is called */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) { tp->t_flags |= TF_RCVD_SCALE; tp->snd_scale = to.to_wscale; } /* * Initial send window. It will be updated with the * next incoming segment to the scaled value. */ tp->snd_wnd = th->th_win; if (to.to_flags & TOF_TS) { tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to.to_tsval; tp->ts_recent_age = tcp_tv_to_mssectick(&bbr->rc_tv); } if (to.to_flags & TOF_MSS) tcp_mss(tp, to.to_mss); if ((tp->t_flags & TF_SACK_PERMIT) && (to.to_flags & TOF_SACKPERM) == 0) tp->t_flags &= ~TF_SACK_PERMIT; if (IS_FASTOPEN(tp->t_flags)) { if (to.to_flags & TOF_FASTOPEN) { uint16_t mss; if (to.to_flags & TOF_MSS) mss = to.to_mss; else if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) mss = TCP6_MSS; else mss = TCP_MSS; tcp_fastopen_update_cache(tp, mss, to.to_tfo_len, to.to_tfo_cookie); } else tcp_fastopen_disable_path(tp); } } /* * At this point we are at the initial call. Here we decide * if we are doing RACK or not. We do this by seeing if * TF_SACK_PERMIT is set, if not rack is *not* possible and * we switch to the default code. */ if ((tp->t_flags & TF_SACK_PERMIT) == 0) { /* Bail */ tcp_switch_back_to_default(tp); (*tp->t_fb->tfb_tcp_do_segment) (m, th, so, tp, drop_hdrlen, tlen, iptos); return (1); } /* Set the flag */ bbr->r_is_v6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; tcp_set_hpts(tp->t_inpcb); sack_filter_clear(&bbr->r_ctl.bbr_sf, th->th_ack); } if (thflags & TH_ACK) { /* Track ack types */ if (to.to_flags & TOF_SACK) BBR_STAT_INC(bbr_acks_with_sacks); else BBR_STAT_INC(bbr_plain_acks); } /* * This is the one exception case where we set the rack state * always. All other times (timers etc) we must have a rack-state * set (so we assure we have done the checks above for SACK). */ if (bbr->r_state != tp->t_state) bbr_set_state(tp, bbr, tiwin); if (SEQ_GT(th->th_ack, tp->snd_una) && (rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map)) != NULL) kern_prefetch(rsm, &prev_state); prev_state = bbr->r_state; bbr->rc_ack_was_delayed = 0; lost = bbr->r_ctl.rc_lost; bbr->rc_is_pkt_epoch_now = 0; if (m->m_flags & (M_TSTMP|M_TSTMP_LRO)) { /* Get the real time into lcts and figure the real delay */ lcts = tcp_get_usecs(<v); if (TSTMP_GT(lcts, cts)) { bbr->r_ctl.rc_ack_hdwr_delay = lcts - cts; bbr->rc_ack_was_delayed = 1; if (TSTMP_GT(bbr->r_ctl.rc_ack_hdwr_delay, bbr->r_ctl.highest_hdwr_delay)) bbr->r_ctl.highest_hdwr_delay = bbr->r_ctl.rc_ack_hdwr_delay; } else { bbr->r_ctl.rc_ack_hdwr_delay = 0; bbr->rc_ack_was_delayed = 0; } } else { bbr->r_ctl.rc_ack_hdwr_delay = 0; bbr->rc_ack_was_delayed = 0; } bbr_log_ack_event(bbr, th, &to, tlen, nsegs, cts, nxt_pkt, m); if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { retval = 0; m_freem(m); goto done_with_input; } /* * If a segment with the ACK-bit set arrives in the SYN-SENT state * check SEQ.ACK first as described on page 66 of RFC 793, section 3.9. */ if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { ctf_do_dropwithreset_conn(m, tp, th, BANDLIM_RST_OPENPORT, tlen); return (1); } in_recovery = IN_RECOVERY(tp->t_flags); if (tiwin > bbr->r_ctl.rc_high_rwnd) bbr->r_ctl.rc_high_rwnd = tiwin; #ifdef BBR_INVARIANTS if ((tp->t_inpcb->inp_flags & INP_DROPPED) || (tp->t_inpcb->inp_flags2 & INP_FREED)) { panic("tp:%p bbr:%p given a dropped inp:%p", tp, bbr, tp->t_inpcb); } #endif bbr->r_ctl.rc_flight_at_input = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); bbr->rtt_valid = 0; if (to.to_flags & TOF_TS) { bbr->rc_ts_valid = 1; bbr->r_ctl.last_inbound_ts = to.to_tsval; } else { bbr->rc_ts_valid = 0; bbr->r_ctl.last_inbound_ts = 0; } retval = (*bbr->r_substate) (m, th, so, tp, &to, drop_hdrlen, tlen, tiwin, thflags, nxt_pkt); #ifdef BBR_INVARIANTS if ((retval == 0) && (tp->t_inpcb == NULL)) { panic("retval:%d tp:%p t_inpcb:NULL state:%d", retval, tp, prev_state); } #endif if (nxt_pkt == 0) BBR_STAT_INC(bbr_rlock_left_ret0); else BBR_STAT_INC(bbr_rlock_left_ret1); if (retval == 0) { /* * If retval is 1 the tcb is unlocked and most likely the tp * is gone. */ INP_WLOCK_ASSERT(tp->t_inpcb); tcp_bbr_xmit_timer_commit(bbr, tp, cts); if (bbr->rc_is_pkt_epoch_now) bbr_set_pktepoch(bbr, cts, __LINE__); bbr_check_bbr_for_state(bbr, cts, __LINE__, (bbr->r_ctl.rc_lost - lost)); if (nxt_pkt == 0) { if (bbr->r_wanted_output != 0) { bbr->rc_output_starts_timer = 0; did_out = 1; (void)tp->t_fb->tfb_tcp_output(tp); } else bbr_start_hpts_timer(bbr, tp, cts, 6, 0, 0); } if ((nxt_pkt == 0) && ((bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK) == 0) && (SEQ_GT(tp->snd_max, tp->snd_una) || (tp->t_flags & TF_DELACK) || ((V_tcp_always_keepalive || bbr->rc_inp->inp_socket->so_options & SO_KEEPALIVE) && (tp->t_state <= TCPS_CLOSING)))) { /* * We could not send (probably in the hpts but * stopped the timer)? */ if ((tp->snd_max == tp->snd_una) && ((tp->t_flags & TF_DELACK) == 0) && (bbr->rc_inp->inp_in_hpts) && (bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT)) { /* * keep alive not needed if we are hptsi * output yet */ ; } else { if (bbr->rc_inp->inp_in_hpts) { tcp_hpts_remove(bbr->rc_inp, HPTS_REMOVE_OUTPUT); if ((bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) && (TSTMP_GT(lcts, bbr->rc_pacer_started))) { uint32_t del; del = lcts - bbr->rc_pacer_started; if (bbr->r_ctl.rc_last_delay_val > del) { BBR_STAT_INC(bbr_force_timer_start); bbr->r_ctl.rc_last_delay_val -= del; bbr->rc_pacer_started = lcts; } else { /* We are late */ bbr->r_ctl.rc_last_delay_val = 0; BBR_STAT_INC(bbr_force_output); (void)tp->t_fb->tfb_tcp_output(tp); } } } bbr_start_hpts_timer(bbr, tp, cts, 8, bbr->r_ctl.rc_last_delay_val, 0); } } else if ((bbr->rc_output_starts_timer == 0) && (nxt_pkt == 0)) { /* Do we have the correct timer running? */ bbr_timer_audit(tp, bbr, lcts, &so->so_snd); } /* Do we have a new state */ if (bbr->r_state != tp->t_state) bbr_set_state(tp, bbr, tiwin); done_with_input: bbr_log_doseg_done(bbr, cts, nxt_pkt, did_out); if (did_out) bbr->r_wanted_output = 0; #ifdef BBR_INVARIANTS if (tp->t_inpcb == NULL) { panic("OP:%d retval:%d tp:%p t_inpcb:NULL state:%d", did_out, retval, tp, prev_state); } #endif } return (retval); } static void bbr_log_type_hrdwtso(struct tcpcb *tp, struct tcp_bbr *bbr, int len, int mod, int what_we_can_send) { if (tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; struct timeval tv; uint32_t cts; cts = tcp_get_usecs(&tv); bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); log.u_bbr.flex1 = bbr->r_ctl.rc_pace_min_segs; log.u_bbr.flex2 = what_we_can_send; log.u_bbr.flex3 = bbr->r_ctl.rc_pace_max_segs; log.u_bbr.flex4 = len; log.u_bbr.flex5 = 0; log.u_bbr.flex7 = mod; log.u_bbr.flex8 = 1; TCP_LOG_EVENTP(tp, NULL, &tp->t_inpcb->inp_socket->so_rcv, &tp->t_inpcb->inp_socket->so_snd, TCP_HDWR_TLS, 0, 0, &log, false, &tv); } } static void bbr_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int32_t drop_hdrlen, int32_t tlen, uint8_t iptos) { struct timeval tv; int retval; /* First lets see if we have old packets */ if (tp->t_in_pkt) { if (ctf_do_queued_segments(so, tp, 1)) { m_freem(m); return; } } if (m->m_flags & M_TSTMP_LRO) { tv.tv_sec = m->m_pkthdr.rcv_tstmp /1000000000; tv.tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000)/1000; } else { /* Should not be should we kassert instead? */ tcp_get_usecs(&tv); } retval = bbr_do_segment_nounlock(m, th, so, tp, drop_hdrlen, tlen, iptos, 0, &tv); if (retval == 0) INP_WUNLOCK(tp->t_inpcb); } /* * Return how much data can be sent without violating the * cwnd or rwnd. */ static inline uint32_t bbr_what_can_we_send(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t sendwin, uint32_t avail, int32_t sb_offset, uint32_t cts) { uint32_t len; if (ctf_outstanding(tp) >= tp->snd_wnd) { /* We never want to go over our peers rcv-window */ len = 0; } else { uint32_t flight; flight = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)); if (flight >= sendwin) { /* * We have in flight what we are allowed by cwnd (if * it was rwnd blocking it would have hit above out * >= tp->snd_wnd). */ return (0); } len = sendwin - flight; if ((len + ctf_outstanding(tp)) > tp->snd_wnd) { /* We would send too much (beyond the rwnd) */ len = tp->snd_wnd - ctf_outstanding(tp); } if ((len + sb_offset) > avail) { /* * We don't have that much in the SB, how much is * there? */ len = avail - sb_offset; } } return (len); } static inline void bbr_do_error_accounting(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, int32_t len, int32_t error) { #ifdef NETFLIX_STATS KMOD_TCPSTAT_INC(tcps_sndpack_error); KMOD_TCPSTAT_ADD(tcps_sndbyte_error, len); #endif } static inline void bbr_do_send_accounting(struct tcpcb *tp, struct tcp_bbr *bbr, struct bbr_sendmap *rsm, int32_t len, int32_t error) { if (error) { bbr_do_error_accounting(tp, bbr, rsm, len, error); return; } if ((tp->t_flags & TF_FORCEDATA) && len == 1) { /* Window probe */ KMOD_TCPSTAT_INC(tcps_sndprobe); #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); #endif } else if (rsm) { if (rsm->r_flags & BBR_TLP) { /* * TLP should not count in retran count, but in its * own bin */ #ifdef NETFLIX_STATS tp->t_sndtlppack++; tp->t_sndtlpbyte += len; KMOD_TCPSTAT_INC(tcps_tlpresends); KMOD_TCPSTAT_ADD(tcps_tlpresend_bytes, len); #endif } else { /* Retransmit */ tp->t_sndrexmitpack++; KMOD_TCPSTAT_INC(tcps_sndrexmitpack); KMOD_TCPSTAT_ADD(tcps_sndrexmitbyte, len); #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); #endif } /* * Logs in 0 - 8, 8 is all non probe_bw states 0-7 is * sub-state */ counter_u64_add(bbr_state_lost[rsm->r_bbr_state], len); if (bbr->rc_bbr_state != BBR_STATE_PROBE_BW) { /* Non probe_bw log in 1, 2, or 4. */ counter_u64_add(bbr_state_resend[bbr->rc_bbr_state], len); } else { /* * Log our probe state 3, and log also 5-13 to show * us the recovery sub-state for the send. This * means that 3 == (5+6+7+8+9+10+11+12+13) */ counter_u64_add(bbr_state_resend[BBR_STATE_PROBE_BW], len); counter_u64_add(bbr_state_resend[(bbr_state_val(bbr) + 5)], len); } /* Place in both 16's the totals of retransmitted */ counter_u64_add(bbr_state_lost[16], len); counter_u64_add(bbr_state_resend[16], len); /* Place in 17's the total sent */ counter_u64_add(bbr_state_resend[17], len); counter_u64_add(bbr_state_lost[17], len); } else { /* New sends */ KMOD_TCPSTAT_INC(tcps_sndpack); KMOD_TCPSTAT_ADD(tcps_sndbyte, len); /* Place in 17's the total sent */ counter_u64_add(bbr_state_resend[17], len); counter_u64_add(bbr_state_lost[17], len); #ifdef STATS stats_voi_update_abs_u64(tp->t_stats, VOI_TCP_TXPB, len); #endif } } static void bbr_cwnd_limiting(struct tcpcb *tp, struct tcp_bbr *bbr, uint32_t in_level) { if (bbr->rc_filled_pipe && bbr_target_cwnd_mult_limit && (bbr->rc_use_google == 0)) { /* * Limit the cwnd to not be above N x the target plus whats * is outstanding. The target is based on the current b/w * estimate. */ uint32_t target; target = bbr_get_target_cwnd(bbr, bbr_get_bw(bbr), BBR_UNIT); target += ctf_outstanding(tp); target *= bbr_target_cwnd_mult_limit; if (tp->snd_cwnd > target) tp->snd_cwnd = target; bbr_log_type_cwndupd(bbr, 0, 0, 0, 10, 0, 0, __LINE__); } } static int bbr_window_update_needed(struct tcpcb *tp, struct socket *so, uint32_t recwin, int32_t maxseg) { /* * "adv" is the amount we could increase the window, taking into * account that we are limited by TCP_MAXWIN << tp->rcv_scale. */ uint32_t adv; int32_t oldwin; adv = min(recwin, TCP_MAXWIN << tp->rcv_scale); if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { oldwin = (tp->rcv_adv - tp->rcv_nxt); adv -= oldwin; } else oldwin = 0; /* * If the new window size ends up being the same as the old size * when it is scaled, then don't force a window update. */ if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale) return (0); if (adv >= (2 * maxseg) && (adv >= (so->so_rcv.sb_hiwat / 4) || recwin <= (so->so_rcv.sb_hiwat / 8) || so->so_rcv.sb_hiwat <= 8 * maxseg)) { return (1); } if (2 * adv >= (int32_t) so->so_rcv.sb_hiwat) return (1); return (0); } /* * Return 0 on success and a errno on failure to send. * Note that a 0 return may not mean we sent anything * if the TCB was on the hpts. A non-zero return * does indicate the error we got from ip[6]_output. */ static int bbr_output_wtime(struct tcpcb *tp, const struct timeval *tv) { struct socket *so; int32_t len; uint32_t cts; uint32_t recwin, sendwin; int32_t sb_offset; int32_t flags, abandon, error = 0; struct tcp_log_buffer *lgb = NULL; struct mbuf *m; struct mbuf *mb; uint32_t if_hw_tsomaxsegcount = 0; uint32_t if_hw_tsomaxsegsize = 0; uint32_t if_hw_tsomax = 0; struct ip *ip = NULL; #ifdef TCPDEBUG struct ipovly *ipov = NULL; #endif struct tcp_bbr *bbr; struct tcphdr *th; #ifdef NETFLIX_TCPOUDP struct udphdr *udp = NULL; #endif u_char opt[TCP_MAXOLEN]; unsigned ipoptlen, optlen, hdrlen; #ifdef NETFLIX_TCPOUDP unsigned ulen; #endif uint32_t bbr_seq; uint32_t delay_calc=0; uint8_t doing_tlp = 0; uint8_t local_options; #ifdef BBR_INVARIANTS uint8_t doing_retran_from = 0; uint8_t picked_up_retran = 0; #endif uint8_t wanted_cookie = 0; uint8_t more_to_rxt=0; int32_t prefetch_so_done = 0; int32_t prefetch_rsm = 0; uint32_t what_we_can = 0; uint32_t tot_len = 0; uint32_t rtr_cnt = 0; uint32_t maxseg, pace_max_segs, p_maxseg; int32_t csum_flags; int32_t hw_tls; #if defined(IPSEC) || defined(IPSEC_SUPPORT) unsigned ipsec_optlen = 0; #endif volatile int32_t sack_rxmit; struct bbr_sendmap *rsm = NULL; int32_t tso, mtu; int force_tso = 0; struct tcpopt to; int32_t slot = 0; struct inpcb *inp; struct sockbuf *sb; uint32_t hpts_calling; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int32_t isipv6; #endif uint8_t app_limited = BBR_JR_SENT_DATA; uint8_t filled_all = 0; bbr = (struct tcp_bbr *)tp->t_fb_ptr; /* We take a cache hit here */ memcpy(&bbr->rc_tv, tv, sizeof(struct timeval)); cts = tcp_tv_to_usectick(&bbr->rc_tv); inp = bbr->rc_inp; so = inp->inp_socket; sb = &so->so_snd; #ifdef KERN_TLS if (sb->sb_flags & SB_TLS_IFNET) hw_tls = 1; else #endif hw_tls = 0; kern_prefetch(sb, &maxseg); maxseg = tp->t_maxseg - bbr->rc_last_options; if (bbr_minseg(bbr) < maxseg) { tcp_bbr_tso_size_check(bbr, cts); } /* Remove any flags that indicate we are pacing on the inp */ pace_max_segs = bbr->r_ctl.rc_pace_max_segs; p_maxseg = min(maxseg, pace_max_segs); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) return (tcp_offload_output(tp)); #endif #ifdef INET6 if (bbr->r_state) { /* Use the cache line loaded if possible */ isipv6 = bbr->r_is_v6; } else { isipv6 = (inp->inp_vflag & INP_IPV6) != 0; } #endif if (((bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) == 0) && inp->inp_in_hpts) { /* * We are on the hpts for some timer but not hptsi output. * Possibly remove from the hpts so we can send/recv etc. */ if ((tp->t_flags & TF_ACKNOW) == 0) { /* * No immediate demand right now to send an ack, but * the user may have read, making room for new data * (a window update). If so we may want to cancel * whatever timer is running (KEEP/DEL-ACK?) and * continue to send out a window update. Or we may * have gotten more data into the socket buffer to * send. */ recwin = min(max(sbspace(&so->so_rcv), 0), TCP_MAXWIN << tp->rcv_scale); if ((bbr_window_update_needed(tp, so, recwin, maxseg) == 0) && ((sbavail(sb) + ((tcp_outflags[tp->t_state] & TH_FIN) ? 1 : 0)) <= (tp->snd_max - tp->snd_una))) { /* * Nothing new to send and no window update * is needed to send. Lets just return and * let the timer-run off. */ return (0); } } tcp_hpts_remove(inp, HPTS_REMOVE_OUTPUT); bbr_timer_cancel(bbr, __LINE__, cts); } if (bbr->r_ctl.rc_last_delay_val) { /* Calculate a rough delay for early escape to sending */ if (SEQ_GT(cts, bbr->rc_pacer_started)) delay_calc = cts - bbr->rc_pacer_started; if (delay_calc >= bbr->r_ctl.rc_last_delay_val) delay_calc -= bbr->r_ctl.rc_last_delay_val; else delay_calc = 0; } /* Mark that we have called bbr_output(). */ if ((bbr->r_timer_override) || (tp->t_flags & TF_FORCEDATA) || (tp->t_state < TCPS_ESTABLISHED)) { /* Timeouts or early states are exempt */ if (inp->inp_in_hpts) tcp_hpts_remove(inp, HPTS_REMOVE_OUTPUT); } else if (inp->inp_in_hpts) { if ((bbr->r_ctl.rc_last_delay_val) && (bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT) && delay_calc) { /* * We were being paced for output and the delay has * already exceeded when we were supposed to be * called, lets go ahead and pull out of the hpts * and call output. */ counter_u64_add(bbr_out_size[TCP_MSS_ACCT_LATE], 1); bbr->r_ctl.rc_last_delay_val = 0; tcp_hpts_remove(inp, HPTS_REMOVE_OUTPUT); } else if (tp->t_state == TCPS_CLOSED) { bbr->r_ctl.rc_last_delay_val = 0; tcp_hpts_remove(inp, HPTS_REMOVE_OUTPUT); } else { /* * On the hpts, you shall not pass! even if ACKNOW * is on, we will when the hpts fires, unless of * course we are overdue. */ counter_u64_add(bbr_out_size[TCP_MSS_ACCT_INPACE], 1); return (0); } } bbr->rc_cwnd_limited = 0; if (bbr->r_ctl.rc_last_delay_val) { /* recalculate the real delay and deal with over/under */ if (SEQ_GT(cts, bbr->rc_pacer_started)) delay_calc = cts - bbr->rc_pacer_started; else delay_calc = 0; if (delay_calc >= bbr->r_ctl.rc_last_delay_val) /* Setup the delay which will be added in */ delay_calc -= bbr->r_ctl.rc_last_delay_val; else { /* * We are early setup to adjust * our slot time. */ uint64_t merged_val; bbr->r_ctl.rc_agg_early += (bbr->r_ctl.rc_last_delay_val - delay_calc); bbr->r_agg_early_set = 1; if (bbr->r_ctl.rc_hptsi_agg_delay) { if (bbr->r_ctl.rc_hptsi_agg_delay >= bbr->r_ctl.rc_agg_early) { /* Nope our previous late cancels out the early */ bbr->r_ctl.rc_hptsi_agg_delay -= bbr->r_ctl.rc_agg_early; bbr->r_agg_early_set = 0; bbr->r_ctl.rc_agg_early = 0; } else { bbr->r_ctl.rc_agg_early -= bbr->r_ctl.rc_hptsi_agg_delay; bbr->r_ctl.rc_hptsi_agg_delay = 0; } } merged_val = bbr->rc_pacer_started; merged_val <<= 32; merged_val |= bbr->r_ctl.rc_last_delay_val; bbr_log_pacing_delay_calc(bbr, inp->inp_hpts_calls, bbr->r_ctl.rc_agg_early, cts, delay_calc, merged_val, bbr->r_agg_early_set, 3); bbr->r_ctl.rc_last_delay_val = 0; BBR_STAT_INC(bbr_early); delay_calc = 0; } } else { /* We were not delayed due to hptsi */ if (bbr->r_agg_early_set) bbr->r_ctl.rc_agg_early = 0; bbr->r_agg_early_set = 0; delay_calc = 0; } if (delay_calc) { /* * We had a hptsi delay which means we are falling behind on * sending at the expected rate. Calculate an extra amount * of data we can send, if any, to put us back on track. */ if ((bbr->r_ctl.rc_hptsi_agg_delay + delay_calc) < bbr->r_ctl.rc_hptsi_agg_delay) bbr->r_ctl.rc_hptsi_agg_delay = 0xffffffff; else bbr->r_ctl.rc_hptsi_agg_delay += delay_calc; } sendwin = min(tp->snd_wnd, tp->snd_cwnd); if ((tp->snd_una == tp->snd_max) && (bbr->rc_bbr_state != BBR_STATE_IDLE_EXIT) && (sbavail(sb))) { /* * Ok we have been idle with nothing outstanding * we possibly need to start fresh with either a new * suite of states or a fast-ramp up. */ bbr_restart_after_idle(bbr, cts, bbr_calc_time(cts, bbr->r_ctl.rc_went_idle_time)); } /* * Now was there a hptsi delay where we are behind? We only count * being behind if: a) We are not in recovery. b) There was a delay. * c) We had room to send something. * */ hpts_calling = inp->inp_hpts_calls; inp->inp_hpts_calls = 0; if (bbr->r_ctl.rc_hpts_flags & PACE_TMR_MASK) { if (bbr_process_timers(tp, bbr, cts, hpts_calling)) { counter_u64_add(bbr_out_size[TCP_MSS_ACCT_ATIMER], 1); return (0); } } bbr->rc_inp->inp_flags2 &= ~INP_MBUF_QUEUE_READY; if (hpts_calling && (bbr->r_ctl.rc_hpts_flags & PACE_PKT_OUTPUT)) { bbr->r_ctl.rc_last_delay_val = 0; } bbr->r_timer_override = 0; bbr->r_wanted_output = 0; /* * For TFO connections in SYN_RECEIVED, only allow the initial * SYN|ACK and those sent by the retransmit timer. */ if (IS_FASTOPEN(tp->t_flags) && ((tp->t_state == TCPS_SYN_RECEIVED) || (tp->t_state == TCPS_SYN_SENT)) && SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN or SYN|ACK sent */ (tp->t_rxtshift == 0)) { /* not a retransmit */ return (0); } /* * Before sending anything check for a state update. For hpts * calling without input this is important. If its input calling * then this was already done. */ if (bbr->rc_use_google == 0) bbr_check_bbr_for_state(bbr, cts, __LINE__, 0); again: /* * If we've recently taken a timeout, snd_max will be greater than * snd_max. BBR in general does not pay much attention to snd_nxt * for historic reasons the persist timer still uses it. This means * we have to look at it. All retransmissions that are not persits * use the rsm that needs to be sent so snd_nxt is ignored. At the * end of this routine we pull snd_nxt always up to snd_max. */ doing_tlp = 0; #ifdef BBR_INVARIANTS doing_retran_from = picked_up_retran = 0; #endif error = 0; tso = 0; slot = 0; mtu = 0; sendwin = min(tp->snd_wnd, tp->snd_cwnd); sb_offset = tp->snd_max - tp->snd_una; flags = tcp_outflags[tp->t_state]; sack_rxmit = 0; len = 0; rsm = NULL; if (flags & TH_RST) { SOCKBUF_LOCK(sb); goto send; } recheck_resend: while (bbr->r_ctl.rc_free_cnt < bbr_min_req_free) { /* We need to always have one in reserve */ rsm = bbr_alloc(bbr); if (rsm == NULL) { error = ENOMEM; /* Lie to get on the hpts */ tot_len = tp->t_maxseg; if (hpts_calling) /* Retry in a ms */ slot = 1001; goto just_return_nolock; } TAILQ_INSERT_TAIL(&bbr->r_ctl.rc_free, rsm, r_next); bbr->r_ctl.rc_free_cnt++; rsm = NULL; } /* What do we send, a resend? */ if (bbr->r_ctl.rc_resend == NULL) { /* Check for rack timeout */ bbr->r_ctl.rc_resend = bbr_check_recovery_mode(tp, bbr, cts); if (bbr->r_ctl.rc_resend) { #ifdef BBR_INVARIANTS picked_up_retran = 1; #endif bbr_cong_signal(tp, NULL, CC_NDUPACK, bbr->r_ctl.rc_resend); } } if (bbr->r_ctl.rc_resend) { rsm = bbr->r_ctl.rc_resend; #ifdef BBR_INVARIANTS doing_retran_from = 1; #endif /* Remove any TLP flags its a RACK or T-O */ rsm->r_flags &= ~BBR_TLP; bbr->r_ctl.rc_resend = NULL; if (SEQ_LT(rsm->r_start, tp->snd_una)) { #ifdef BBR_INVARIANTS panic("Huh, tp:%p bbr:%p rsm:%p start:%u < snd_una:%u\n", tp, bbr, rsm, rsm->r_start, tp->snd_una); goto recheck_resend; #else /* TSNH */ rsm = NULL; goto recheck_resend; #endif } rtr_cnt++; if (rsm->r_flags & BBR_HAS_SYN) { /* Only retransmit a SYN by itself */ len = 0; if ((flags & TH_SYN) == 0) { /* Huh something is wrong */ rsm->r_start++; if (rsm->r_start == rsm->r_end) { /* Clean it up, somehow we missed the ack? */ bbr_log_syn(tp, NULL); } else { /* TFO with data? */ rsm->r_flags &= ~BBR_HAS_SYN; len = rsm->r_end - rsm->r_start; } } else { /* Retransmitting SYN */ rsm = NULL; SOCKBUF_LOCK(sb); goto send; } } else len = rsm->r_end - rsm->r_start; if ((bbr->rc_resends_use_tso == 0) && #ifdef KERN_TLS ((sb->sb_flags & SB_TLS_IFNET) == 0) && #endif (len > maxseg)) { len = maxseg; more_to_rxt = 1; } sb_offset = rsm->r_start - tp->snd_una; if (len > 0) { sack_rxmit = 1; KMOD_TCPSTAT_INC(tcps_sack_rexmits); KMOD_TCPSTAT_ADD(tcps_sack_rexmit_bytes, min(len, maxseg)); } else { /* I dont think this can happen */ rsm = NULL; goto recheck_resend; } BBR_STAT_INC(bbr_resends_set); } else if (bbr->r_ctl.rc_tlp_send) { /* * Tail loss probe */ doing_tlp = 1; rsm = bbr->r_ctl.rc_tlp_send; bbr->r_ctl.rc_tlp_send = NULL; sack_rxmit = 1; len = rsm->r_end - rsm->r_start; rtr_cnt++; if ((bbr->rc_resends_use_tso == 0) && (len > maxseg)) len = maxseg; if (SEQ_GT(tp->snd_una, rsm->r_start)) { #ifdef BBR_INVARIANTS panic("tp:%p bbc:%p snd_una:%u rsm:%p r_start:%u", tp, bbr, tp->snd_una, rsm, rsm->r_start); #else /* TSNH */ rsm = NULL; goto recheck_resend; #endif } sb_offset = rsm->r_start - tp->snd_una; BBR_STAT_INC(bbr_tlp_set); } /* * Enforce a connection sendmap count limit if set * as long as we are not retransmiting. */ if ((rsm == NULL) && (V_tcp_map_entries_limit > 0) && (bbr->r_ctl.rc_num_maps_alloced >= V_tcp_map_entries_limit)) { BBR_STAT_INC(bbr_alloc_limited); if (!bbr->alloc_limit_reported) { bbr->alloc_limit_reported = 1; BBR_STAT_INC(bbr_alloc_limited_conns); } goto just_return_nolock; } #ifdef BBR_INVARIANTS if (rsm && SEQ_LT(rsm->r_start, tp->snd_una)) { panic("tp:%p bbr:%p rsm:%p sb_offset:%u len:%u", tp, bbr, rsm, sb_offset, len); } #endif /* * Get standard flags, and add SYN or FIN if requested by 'hidden' * state flags. */ if (tp->t_flags & TF_NEEDFIN && (rsm == NULL)) flags |= TH_FIN; if (tp->t_flags & TF_NEEDSYN) flags |= TH_SYN; if (rsm && (rsm->r_flags & BBR_HAS_FIN)) { /* we are retransmitting the fin */ len--; if (len) { /* * When retransmitting data do *not* include the * FIN. This could happen from a TLP probe if we * allowed data with a FIN. */ flags &= ~TH_FIN; } } else if (rsm) { if (flags & TH_FIN) flags &= ~TH_FIN; } if ((sack_rxmit == 0) && (prefetch_rsm == 0)) { void *end_rsm; end_rsm = TAILQ_LAST_FAST(&bbr->r_ctl.rc_tmap, bbr_sendmap, r_tnext); if (end_rsm) kern_prefetch(end_rsm, &prefetch_rsm); prefetch_rsm = 1; } SOCKBUF_LOCK(sb); /* * If in persist timeout with window of 0, send 1 byte. Otherwise, * if window is small but nonzero and time TF_SENTFIN expired, we * will send what we can and go to transmit state. */ if (tp->t_flags & TF_FORCEDATA) { if ((sendwin == 0) || (sendwin <= (tp->snd_max - tp->snd_una))) { /* * If we still have some data to send, then clear * the FIN bit. Usually this would happen below * when it realizes that we aren't sending all the * data. However, if we have exactly 1 byte of * unsent data, then it won't clear the FIN bit * below, and if we are in persist state, we wind up * sending the packet without recording that we sent * the FIN bit. * * We can't just blindly clear the FIN bit, because * if we don't have any more data to send then the * probe will be the FIN itself. */ if (sb_offset < sbused(sb)) flags &= ~TH_FIN; sendwin = 1; } else { if ((bbr->rc_in_persist != 0) && (tp->snd_wnd >= min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr)))) { /* Exit persists if there is space */ bbr_exit_persist(tp, bbr, cts, __LINE__); } if (rsm == NULL) { /* * If we are dropping persist mode then we * need to correct sb_offset if not a * retransmit. */ sb_offset = tp->snd_max - tp->snd_una; } } } /* * If snd_nxt == snd_max and we have transmitted a FIN, the * sb_offset will be > 0 even if so_snd.sb_cc is 0, resulting in a * negative length. This can also occur when TCP opens up its * congestion window while receiving additional duplicate acks after * fast-retransmit because TCP will reset snd_nxt to snd_max after * the fast-retransmit. * * In the normal retransmit-FIN-only case, however, snd_nxt will be * set to snd_una, the sb_offset will be 0, and the length may wind * up 0. * * If sack_rxmit is true we are retransmitting from the scoreboard * in which case len is already set. */ if (sack_rxmit == 0) { uint32_t avail; avail = sbavail(sb); if (SEQ_GT(tp->snd_max, tp->snd_una)) sb_offset = tp->snd_max - tp->snd_una; else sb_offset = 0; if (bbr->rc_tlp_new_data) { /* TLP is forcing out new data */ uint32_t tlplen; doing_tlp = 1; tlplen = maxseg; if (tlplen > (uint32_t)(avail - sb_offset)) { tlplen = (uint32_t)(avail - sb_offset); } if (tlplen > tp->snd_wnd) { len = tp->snd_wnd; } else { len = tlplen; } bbr->rc_tlp_new_data = 0; } else { what_we_can = len = bbr_what_can_we_send(tp, bbr, sendwin, avail, sb_offset, cts); if ((len < p_maxseg) && (bbr->rc_in_persist == 0) && (ctf_outstanding(tp) >= (2 * p_maxseg)) && ((avail - sb_offset) >= p_maxseg)) { /* * We are not completing whats in the socket * buffer (i.e. there is at least a segment * waiting to send) and we have 2 or more * segments outstanding. There is no sense * of sending a little piece. Lets defer and * and wait until we can send a whole * segment. */ len = 0; } if ((tp->t_flags & TF_FORCEDATA) && (bbr->rc_in_persist)) { /* * We are in persists, figure out if * a retransmit is available (maybe the previous * persists we sent) or if we have to send new * data. */ rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); if (rsm) { len = rsm->r_end - rsm->r_start; if (rsm->r_flags & BBR_HAS_FIN) len--; if ((bbr->rc_resends_use_tso == 0) && (len > maxseg)) len = maxseg; if (len > 1) BBR_STAT_INC(bbr_persist_reneg); /* * XXXrrs we could force the len to * 1 byte here to cause the chunk to * split apart.. but that would then * mean we always retransmit it as * one byte even after the window * opens. */ sack_rxmit = 1; sb_offset = rsm->r_start - tp->snd_una; } else { /* * First time through in persists or peer * acked our one byte. Though we do have * to have something in the sb. */ len = 1; sb_offset = 0; if (avail == 0) len = 0; } } } } if (prefetch_so_done == 0) { kern_prefetch(so, &prefetch_so_done); prefetch_so_done = 1; } /* * Lop off SYN bit if it has already been sent. However, if this is * SYN-SENT state and if segment contains data and if we don't know * that foreign host supports TAO, suppress sending segment. */ if ((flags & TH_SYN) && (rsm == NULL) && SEQ_GT(tp->snd_max, tp->snd_una)) { if (tp->t_state != TCPS_SYN_RECEIVED) flags &= ~TH_SYN; /* * When sending additional segments following a TFO SYN|ACK, * do not include the SYN bit. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_state == TCPS_SYN_RECEIVED)) flags &= ~TH_SYN; sb_offset--, len++; if (sbavail(sb) == 0) len = 0; } else if ((flags & TH_SYN) && rsm) { /* * Subtract one from the len for the SYN being * retransmitted. */ len--; } /* * Be careful not to send data and/or FIN on SYN segments. This * measure is needed to prevent interoperability problems with not * fully conformant TCP implementations. */ if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { len = 0; flags &= ~TH_FIN; } /* * On TFO sockets, ensure no data is sent in the following cases: * * - When retransmitting SYN|ACK on a passively-created socket * - When retransmitting SYN on an actively created socket * - When sending a zero-length cookie (cookie request) on an * actively created socket * - When the socket is in the CLOSED state (RST is being sent) */ if (IS_FASTOPEN(tp->t_flags) && (((flags & TH_SYN) && (tp->t_rxtshift > 0)) || ((tp->t_state == TCPS_SYN_SENT) && (tp->t_tfo_client_cookie_len == 0)) || (flags & TH_RST))) { len = 0; sack_rxmit = 0; rsm = NULL; } /* Without fast-open there should never be data sent on a SYN */ if ((flags & TH_SYN) && (!IS_FASTOPEN(tp->t_flags))) len = 0; if (len <= 0) { /* * If FIN has been sent but not acked, but we haven't been * called to retransmit, len will be < 0. Otherwise, window * shrank after we sent into it. If window shrank to 0, * cancel pending retransmit, pull snd_nxt back to (closed) * window, and set the persist timer if it isn't already * going. If the window didn't close completely, just wait * for an ACK. * * We also do a general check here to ensure that we will * set the persist timer when we have data to send, but a * 0-byte window. This makes sure the persist timer is set * even if the packet hits one of the "goto send" lines * below. */ len = 0; if ((tp->snd_wnd == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && (tp->snd_una == tp->snd_max) && (sb_offset < (int)sbavail(sb))) { /* * Not enough room in the rwnd to send * a paced segment out. */ bbr_enter_persist(tp, bbr, cts, __LINE__); } } else if ((rsm == NULL) && (doing_tlp == 0) && (len < bbr->r_ctl.rc_pace_max_segs)) { /* * We are not sending a full segment for * some reason. Should we not send anything (think * sws or persists)? */ if ((tp->snd_wnd < min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr))) && (TCPS_HAVEESTABLISHED(tp->t_state)) && (len < (int)(sbavail(sb) - sb_offset))) { /* * Here the rwnd is less than * the pacing size, this is not a retransmit, * we are established and * the send is not the last in the socket buffer * lets not send, and possibly enter persists. */ len = 0; if (tp->snd_max == tp->snd_una) bbr_enter_persist(tp, bbr, cts, __LINE__); } else if ((tp->snd_cwnd >= bbr->r_ctl.rc_pace_max_segs) && (ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) > (2 * maxseg)) && (len < (int)(sbavail(sb) - sb_offset)) && (len < bbr_minseg(bbr))) { /* * Here we are not retransmitting, and * the cwnd is not so small that we could * not send at least a min size (rxt timer * not having gone off), We have 2 segments or * more already in flight, its not the tail end * of the socket buffer and the cwnd is blocking * us from sending out minimum pacing segment size. * Lets not send anything. */ bbr->rc_cwnd_limited = 1; len = 0; } else if (((tp->snd_wnd - ctf_outstanding(tp)) < min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr))) && (ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) > (2 * maxseg)) && (len < (int)(sbavail(sb) - sb_offset)) && (TCPS_HAVEESTABLISHED(tp->t_state))) { /* * Here we have a send window but we have * filled it up and we can't send another pacing segment. * We also have in flight more than 2 segments * and we are not completing the sb i.e. we allow * the last bytes of the sb to go out even if * its not a full pacing segment. */ len = 0; } } /* len will be >= 0 after this point. */ KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); tcp_sndbuf_autoscale(tp, so, sendwin); /* * */ if (bbr->rc_in_persist && len && (rsm == NULL) && (len < min((bbr->r_ctl.rc_high_rwnd/2), bbr->r_ctl.rc_pace_max_segs))) { /* * We are in persist, not doing a retransmit and don't have enough space * yet to send a full TSO. So is it at the end of the sb * if so we need to send else nuke to 0 and don't send. */ int sbleft; if (sbavail(sb) > sb_offset) sbleft = sbavail(sb) - sb_offset; else sbleft = 0; if (sbleft >= min((bbr->r_ctl.rc_high_rwnd/2), bbr->r_ctl.rc_pace_max_segs)) { /* not at end of sb lets not send */ len = 0; } } /* * Decide if we can use TCP Segmentation Offloading (if supported by * hardware). * * TSO may only be used if we are in a pure bulk sending state. The * presence of TCP-MD5, SACK retransmits, SACK advertizements and IP * options prevent using TSO. With TSO the TCP header is the same * (except for the sequence number) for all generated packets. This * makes it impossible to transmit any options which vary per * generated segment or packet. * * IPv4 handling has a clear separation of ip options and ip header * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() * does the right thing below to provide length of just ip options * and thus checking for ipoptlen is enough to decide if ip options * are present. */ #ifdef INET6 if (isipv6) ipoptlen = ip6_optlen(inp); else #endif if (inp->inp_options) ipoptlen = inp->inp_options->m_len - offsetof(struct ipoption, ipopt_list); else ipoptlen = 0; #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * Pre-calculate here as we save another lookup into the darknesses * of IPsec that way and can actually decide if TSO is ok. */ #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6)) ipsec_optlen = IPSEC_HDRSIZE(ipv6, inp); #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4)) ipsec_optlen = IPSEC_HDRSIZE(ipv4, inp); #endif /* INET */ #endif /* IPSEC */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) ipoptlen += ipsec_optlen; #endif if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && (len > maxseg) && (tp->t_port == 0) && ((tp->t_flags & TF_SIGNATURE) == 0) && tp->rcv_numsacks == 0 && ipoptlen == 0) tso = 1; recwin = min(max(sbspace(&so->so_rcv), 0), TCP_MAXWIN << tp->rcv_scale); /* * Sender silly window avoidance. We transmit under the following * conditions when len is non-zero: * * - We have a full segment (or more with TSO) - This is the last * buffer in a write()/send() and we are either idle or running * NODELAY - we've timed out (e.g. persist timer) - we have more * then 1/2 the maximum send window's worth of data (receiver may be * limited the window size) - we need to retransmit */ if (rsm) goto send; if (len) { if (sack_rxmit) goto send; if (len >= p_maxseg) goto send; /* * NOTE! on localhost connections an 'ack' from the remote * end may occur synchronously with the output and cause us * to flush a buffer queued with moretocome. XXX * */ if (((tp->t_flags & TF_MORETOCOME) == 0) && /* normal case */ ((tp->t_flags & TF_NODELAY) || ((uint32_t)len + (uint32_t)sb_offset) >= sbavail(&so->so_snd)) && (tp->t_flags & TF_NOPUSH) == 0) { goto send; } if ((tp->snd_una == tp->snd_max) && len) { /* Nothing outstanding */ goto send; } if (tp->t_flags & TF_FORCEDATA) { /* typ. timeout case */ goto send; } if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) { goto send; } } /* * Sending of standalone window updates. * * Window updates are important when we close our window due to a * full socket buffer and are opening it again after the application * reads data from it. Once the window has opened again and the * remote end starts to send again the ACK clock takes over and * provides the most current window information. * * We must avoid the silly window syndrome whereas every read from * the receive buffer, no matter how small, causes a window update * to be sent. We also should avoid sending a flurry of window * updates when the socket buffer had queued a lot of data and the * application is doing small reads. * * Prevent a flurry of pointless window updates by only sending an * update when we can increase the advertized window by more than * 1/4th of the socket buffer capacity. When the buffer is getting * full or is very small be more aggressive and send an update * whenever we can increase by two mss sized segments. In all other * situations the ACK's to new incoming data will carry further * window increases. * * Don't send an independent window update if a delayed ACK is * pending (it will get piggy-backed on it) or the remote side * already has done a half-close and won't send more data. Skip * this if the connection is in T/TCP half-open state. */ if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && !(tp->t_flags & TF_DELACK) && !TCPS_HAVERCVDFIN(tp->t_state)) { /* Check to see if we should do a window update */ if (bbr_window_update_needed(tp, so, recwin, maxseg)) goto send; } /* * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW * is also a catch-all for the retransmit timer timeout case. */ if (tp->t_flags & TF_ACKNOW) { goto send; } if (((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) { goto send; } if (SEQ_GT(tp->snd_up, tp->snd_una)) { goto send; } /* * If our state indicates that FIN should be sent and we have not * yet done so, then we need to send. */ if (flags & TH_FIN && ((tp->t_flags & TF_SENTFIN) == 0)) { goto send; } /* * No reason to send a segment, just return. */ just_return: SOCKBUF_UNLOCK(sb); just_return_nolock: if (tot_len) slot = bbr_get_pacing_delay(bbr, bbr->r_ctl.rc_bbr_hptsi_gain, tot_len, cts, 0); if (bbr->rc_no_pacing) slot = 0; if (tot_len == 0) { if ((ctf_outstanding(tp) + min((bbr->r_ctl.rc_high_rwnd/2), bbr_minseg(bbr))) >= tp->snd_wnd) { BBR_STAT_INC(bbr_rwnd_limited); app_limited = BBR_JR_RWND_LIMITED; bbr_cwnd_limiting(tp, bbr, ctf_outstanding(tp)); if ((bbr->rc_in_persist == 0) && TCPS_HAVEESTABLISHED(tp->t_state) && (tp->snd_max == tp->snd_una) && sbavail(&tp->t_inpcb->inp_socket->so_snd)) { /* No send window.. we must enter persist */ bbr_enter_persist(tp, bbr, bbr->r_ctl.rc_rcvtime, __LINE__); } } else if (ctf_outstanding(tp) >= sbavail(sb)) { BBR_STAT_INC(bbr_app_limited); app_limited = BBR_JR_APP_LIMITED; bbr_cwnd_limiting(tp, bbr, ctf_outstanding(tp)); } else if ((ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + p_maxseg) >= tp->snd_cwnd) { BBR_STAT_INC(bbr_cwnd_limited); app_limited = BBR_JR_CWND_LIMITED; bbr_cwnd_limiting(tp, bbr, ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes))); bbr->rc_cwnd_limited = 1; } else { BBR_STAT_INC(bbr_app_limited); app_limited = BBR_JR_APP_LIMITED; bbr_cwnd_limiting(tp, bbr, ctf_outstanding(tp)); } bbr->r_ctl.rc_hptsi_agg_delay = 0; bbr->r_agg_early_set = 0; bbr->r_ctl.rc_agg_early = 0; bbr->r_ctl.rc_last_delay_val = 0; } else if (bbr->rc_use_google == 0) bbr_check_bbr_for_state(bbr, cts, __LINE__, 0); /* Are we app limited? */ if ((app_limited == BBR_JR_APP_LIMITED) || (app_limited == BBR_JR_RWND_LIMITED)) { /** * We are application limited. */ bbr->r_ctl.r_app_limited_until = (ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + bbr->r_ctl.rc_delivered); } if (tot_len == 0) counter_u64_add(bbr_out_size[TCP_MSS_ACCT_JUSTRET], 1); tp->t_flags &= ~TF_FORCEDATA; /* Dont update the time if we did not send */ bbr->r_ctl.rc_last_delay_val = 0; bbr->rc_output_starts_timer = 1; bbr_start_hpts_timer(bbr, tp, cts, 9, slot, tot_len); bbr_log_type_just_return(bbr, cts, tot_len, hpts_calling, app_limited, p_maxseg, len); if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { /* Make sure snd_nxt is drug up */ tp->snd_nxt = tp->snd_max; } return (error); send: if (doing_tlp == 0) { /* * Data not a TLP, and its not the rxt firing. If it is the * rxt firing, we want to leave the tlp_in_progress flag on * so we don't send another TLP. It has to be a rack timer * or normal send (response to acked data) to clear the tlp * in progress flag. */ bbr->rc_tlp_in_progress = 0; bbr->rc_tlp_rtx_out = 0; } else { /* * Its a TLP. */ bbr->rc_tlp_in_progress = 1; } bbr_timer_cancel(bbr, __LINE__, cts); if (rsm == NULL) { if (sbused(sb) > 0) { /* * This is sub-optimal. We only send a stand alone * FIN on its own segment. */ if (flags & TH_FIN) { flags &= ~TH_FIN; if ((len == 0) && ((tp->t_flags & TF_ACKNOW) == 0)) { /* Lets not send this */ slot = 0; goto just_return; } } } } else { /* * We do *not* send a FIN on a retransmit if it has data. * The if clause here where len > 1 should never come true. */ if ((len > 0) && (((rsm->r_flags & BBR_HAS_FIN) == 0) && (flags & TH_FIN))) { flags &= ~TH_FIN; len--; } } SOCKBUF_LOCK_ASSERT(sb); if (len > 0) { if ((tp->snd_una == tp->snd_max) && (bbr_calc_time(cts, bbr->r_ctl.rc_went_idle_time) >= bbr_rtt_probe_time)) { /* * This qualifies as a RTT_PROBE session since we * drop the data outstanding to nothing and waited * more than bbr_rtt_probe_time. */ bbr_log_rtt_shrinks(bbr, cts, 0, 0, __LINE__, BBR_RTTS_WASIDLE, 0); bbr_set_reduced_rtt(bbr, cts, __LINE__); } if (len >= maxseg) tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; else tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; } /* * Before ESTABLISHED, force sending of initial options unless TCP * set not to do any options. NOTE: we assume that the IP/TCP header * plus TCP options always fit in a single mbuf, leaving room for a * maximum link header, i.e. max_linkhdr + sizeof (struct tcpiphdr) * + optlen <= MCLBYTES */ optlen = 0; #ifdef INET6 if (isipv6) hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); else #endif hdrlen = sizeof(struct tcpiphdr); /* * Compute options for segment. We only have to care about SYN and * established connection segments. Options for SYN-ACK segments * are handled in TCP syncache. */ to.to_flags = 0; local_options = 0; if ((tp->t_flags & TF_NOOPT) == 0) { /* Maximum segment size. */ if (flags & TH_SYN) { to.to_mss = tcp_mssopt(&inp->inp_inc); #ifdef NETFLIX_TCPOUDP if (tp->t_port) to.to_mss -= V_tcp_udp_tunneling_overhead; #endif to.to_flags |= TOF_MSS; /* * On SYN or SYN|ACK transmits on TFO connections, * only include the TFO option if it is not a * retransmit, as the presence of the TFO option may * have caused the original SYN or SYN|ACK to have * been dropped by a middlebox. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_rxtshift == 0)) { if (tp->t_state == TCPS_SYN_RECEIVED) { to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; to.to_tfo_cookie = (u_int8_t *)&tp->t_tfo_cookie.server; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; } else if (tp->t_state == TCPS_SYN_SENT) { to.to_tfo_len = tp->t_tfo_client_cookie_len; to.to_tfo_cookie = tp->t_tfo_cookie.client; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; } } } /* Window scaling. */ if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { to.to_wscale = tp->request_r_scale; to.to_flags |= TOF_SCALE; } /* Timestamps. */ if ((tp->t_flags & TF_RCVD_TSTMP) || ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { to.to_tsval = tcp_tv_to_mssectick(&bbr->rc_tv) + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; local_options += TCPOLEN_TIMESTAMP + 2; } /* Set receive buffer autosizing timestamp. */ if (tp->rfbuf_ts == 0 && (so->so_rcv.sb_flags & SB_AUTOSIZE)) tp->rfbuf_ts = tcp_tv_to_mssectick(&bbr->rc_tv); /* Selective ACK's. */ if (flags & TH_SYN) to.to_flags |= TOF_SACKPERM; else if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->rcv_numsacks > 0) { to.to_flags |= TOF_SACK; to.to_nsacks = tp->rcv_numsacks; to.to_sacks = (u_char *)tp->sackblks; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* TCP_SIGNATURE */ /* Processing the options. */ hdrlen += (optlen = tcp_addoptions(&to, opt)); /* * If we wanted a TFO option to be added, but it was unable * to fit, ensure no data is sent. */ if (IS_FASTOPEN(tp->t_flags) && wanted_cookie && !(to.to_flags & TOF_FASTOPEN)) len = 0; } #ifdef NETFLIX_TCPOUDP if (tp->t_port) { if (V_tcp_udp_tunneling_port == 0) { /* The port was removed?? */ SOCKBUF_UNLOCK(&so->so_snd); return (EHOSTUNREACH); } hdrlen += sizeof(struct udphdr); } #endif #ifdef INET6 if (isipv6) ipoptlen = ip6_optlen(tp->t_inpcb); else #endif if (tp->t_inpcb->inp_options) ipoptlen = tp->t_inpcb->inp_options->m_len - offsetof(struct ipoption, ipopt_list); else ipoptlen = 0; ipoptlen = 0; #if defined(IPSEC) || defined(IPSEC_SUPPORT) ipoptlen += ipsec_optlen; #endif if (bbr->rc_last_options != local_options) { /* * Cache the options length this generally does not change * on a connection. We use this to calculate TSO. */ bbr->rc_last_options = local_options; } maxseg = tp->t_maxseg - (ipoptlen + optlen); p_maxseg = min(maxseg, pace_max_segs); /* * Adjust data length if insertion of options will bump the packet * length beyond the t_maxseg length. Clear the FIN bit because we * cut off the tail of the segment. */ #ifdef KERN_TLS /* force TSO for so TLS offload can get mss */ if (sb->sb_flags & SB_TLS_IFNET) { force_tso = 1; } #endif if (len > maxseg) { if (len != 0 && (flags & TH_FIN)) { flags &= ~TH_FIN; } if (tso) { uint32_t moff; int32_t max_len; /* extract TSO information */ if_hw_tsomax = tp->t_tsomax; if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; KASSERT(ipoptlen == 0, ("%s: TSO can't do IP options", __func__)); /* * Check if we should limit by maximum payload * length: */ if (if_hw_tsomax != 0) { /* compute maximum TSO length */ max_len = (if_hw_tsomax - hdrlen - max_linkhdr); if (max_len <= 0) { len = 0; } else if (len > max_len) { len = max_len; } } /* * Prevent the last segment from being fractional * unless the send sockbuf can be emptied: */ if (((sb_offset + len) < sbavail(sb)) && (hw_tls == 0)) { moff = len % (uint32_t)maxseg; if (moff != 0) { len -= moff; } } /* * In case there are too many small fragments don't * use TSO: */ if (len <= maxseg) { len = maxseg; tso = 0; } } else { /* Not doing TSO */ if (optlen + ipoptlen >= tp->t_maxseg) { /* * Since we don't have enough space to put * the IP header chain and the TCP header in * one packet as required by RFC 7112, don't * send it. Also ensure that at least one * byte of the payload can be put into the * TCP segment. */ SOCKBUF_UNLOCK(&so->so_snd); error = EMSGSIZE; sack_rxmit = 0; goto out; } len = maxseg; } } else { /* Not doing TSO */ if_hw_tsomaxsegcount = 0; tso = 0; } KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, ("%s: len > IP_MAXPACKET", __func__)); #ifdef DIAGNOSTIC #ifdef INET6 if (max_linkhdr + hdrlen > MCLBYTES) #else if (max_linkhdr + hdrlen > MHLEN) #endif panic("tcphdr too big"); #endif /* * This KASSERT is here to catch edge cases at a well defined place. * Before, those had triggered (random) panic conditions further * down. */ #ifdef BBR_INVARIANTS if (sack_rxmit) { if (SEQ_LT(rsm->r_start, tp->snd_una)) { panic("RSM:%p TP:%p bbr:%p start:%u is < snd_una:%u", rsm, tp, bbr, rsm->r_start, tp->snd_una); } } #endif KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); if ((len == 0) && (flags & TH_FIN) && (sbused(sb))) { /* * We have outstanding data, don't send a fin by itself!. */ slot = 0; goto just_return; } /* * Grab a header mbuf, attaching a copy of data to be transmitted, * and initialize the header from the template for sends on this * connection. */ if (len) { uint32_t moff; uint32_t orig_len; /* * We place a limit on sending with hptsi. */ if ((rsm == NULL) && len > pace_max_segs) len = pace_max_segs; if (len <= maxseg) tso = 0; #ifdef INET6 if (MHLEN < hdrlen + max_linkhdr) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else #endif m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { BBR_STAT_INC(bbr_failed_mbuf_aloc); bbr_log_enobuf_jmp(bbr, len, cts, __LINE__, len, 0, 0); SOCKBUF_UNLOCK(sb); error = ENOBUFS; sack_rxmit = 0; goto out; } m->m_data += max_linkhdr; m->m_len = hdrlen; /* * Start the m_copy functions from the closest mbuf to the * sb_offset in the socket buffer chain. */ if ((sb_offset > sbavail(sb)) || ((len + sb_offset) > sbavail(sb))) { #ifdef BBR_INVARIANTS if ((len + sb_offset) > (sbavail(sb) + ((flags & (TH_FIN | TH_SYN)) ? 1 : 0))) panic("tp:%p bbr:%p len:%u sb_offset:%u sbavail:%u rsm:%p %u:%u:%u", tp, bbr, len, sb_offset, sbavail(sb), rsm, doing_retran_from, picked_up_retran, doing_tlp); #endif /* * In this messed up situation we have two choices, * a) pretend the send worked, and just start timers * and what not (not good since that may lead us * back here a lot). b) Send the lowest segment * in the map. c) Drop the connection. Lets do * which if it continues to happen will lead to * via timeouts. */ BBR_STAT_INC(bbr_offset_recovery); rsm = TAILQ_FIRST(&bbr->r_ctl.rc_map); sb_offset = 0; if (rsm == NULL) { sack_rxmit = 0; len = sbavail(sb); } else { sack_rxmit = 1; if (rsm->r_start != tp->snd_una) { /* * Things are really messed up, * is the only thing to do. */ BBR_STAT_INC(bbr_offset_drop); tcp_set_inp_to_drop(inp, EFAULT); return (0); } len = rsm->r_end - rsm->r_start; } if (len > sbavail(sb)) len = sbavail(sb); if (len > maxseg) len = maxseg; } mb = sbsndptr_noadv(sb, sb_offset, &moff); if (len <= MHLEN - hdrlen - max_linkhdr && !hw_tls) { m_copydata(mb, moff, (int)len, mtod(m, caddr_t)+hdrlen); if (rsm == NULL) sbsndptr_adv(sb, mb, len); m->m_len += len; } else { struct sockbuf *msb; if (rsm) msb = NULL; else msb = sb; #ifdef BBR_INVARIANTS if ((len + moff) > (sbavail(sb) + ((flags & (TH_FIN | TH_SYN)) ? 1 : 0))) { if (rsm) { panic("tp:%p bbr:%p len:%u moff:%u sbavail:%u rsm:%p snd_una:%u rsm_start:%u flg:%x %u:%u:%u sr:%d ", tp, bbr, len, moff, sbavail(sb), rsm, tp->snd_una, rsm->r_flags, rsm->r_start, doing_retran_from, picked_up_retran, doing_tlp, sack_rxmit); } else { panic("tp:%p bbr:%p len:%u moff:%u sbavail:%u sb_offset:%u snd_una:%u", tp, bbr, len, moff, sbavail(sb), sb_offset, tp->snd_una); } } #endif orig_len = len; m->m_next = tcp_m_copym( #ifdef NETFLIX_COPY_ARGS tp, #endif mb, moff, &len, if_hw_tsomaxsegcount, if_hw_tsomaxsegsize, msb, ((rsm == NULL) ? hw_tls : 0) #ifdef NETFLIX_COPY_ARGS , &filled_all #endif ); if (len <= maxseg && !force_tso) { /* * Must have ran out of mbufs for the copy * shorten it to no longer need tso. Lets * not put on sendalot since we are low on * mbufs. */ tso = 0; } if (m->m_next == NULL) { SOCKBUF_UNLOCK(sb); (void)m_free(m); error = ENOBUFS; sack_rxmit = 0; goto out; } } #ifdef BBR_INVARIANTS if (tso && len < maxseg) { panic("tp:%p tso on, but len:%d < maxseg:%d", tp, len, maxseg); } if (tso && if_hw_tsomaxsegcount) { int32_t seg_cnt = 0; struct mbuf *foo; foo = m; while (foo) { seg_cnt++; foo = foo->m_next; } if (seg_cnt > if_hw_tsomaxsegcount) { panic("seg_cnt:%d > max:%d", seg_cnt, if_hw_tsomaxsegcount); } } #endif /* * If we're sending everything we've got, set PUSH. (This * will keep happy those implementations which only give * data to the user when a buffer fills or a PUSH comes in.) */ if (sb_offset + len == sbused(sb) && sbused(sb) && !(flags & TH_SYN)) { flags |= TH_PUSH; } SOCKBUF_UNLOCK(sb); } else { SOCKBUF_UNLOCK(sb); if (tp->t_flags & TF_ACKNOW) KMOD_TCPSTAT_INC(tcps_sndacks); else if (flags & (TH_SYN | TH_FIN | TH_RST)) KMOD_TCPSTAT_INC(tcps_sndctrl); else if (SEQ_GT(tp->snd_up, tp->snd_una)) KMOD_TCPSTAT_INC(tcps_sndurg); else KMOD_TCPSTAT_INC(tcps_sndwinup); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { BBR_STAT_INC(bbr_failed_mbuf_aloc); bbr_log_enobuf_jmp(bbr, len, cts, __LINE__, len, 0, 0); error = ENOBUFS; /* Fudge the send time since we could not send */ sack_rxmit = 0; goto out; } #ifdef INET6 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && MHLEN >= hdrlen) { M_ALIGN(m, hdrlen); } else #endif m->m_data += max_linkhdr; m->m_len = hdrlen; } SOCKBUF_UNLOCK_ASSERT(sb); m->m_pkthdr.rcvif = (struct ifnet *)0; #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif #ifdef INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); #ifdef NETFLIX_TCPOUDP if (tp->t_port) { udp = (struct udphdr *)((caddr_t)ip6 + ipoptlen + sizeof(struct ip6_hdr)); udp->uh_sport = htons(V_tcp_udp_tunneling_port); udp->uh_dport = tp->t_port; ulen = hdrlen + len - sizeof(struct ip6_hdr); udp->uh_ulen = htons(ulen); th = (struct tcphdr *)(udp + 1); } else { #endif th = (struct tcphdr *)(ip6 + 1); #ifdef NETFLIX_TCPOUDP } #endif tcpip_fillheaders(inp, #ifdef NETFLIX_TCPOUDP tp->t_port, #endif ip6, th); } else #endif /* INET6 */ { ip = mtod(m, struct ip *); #ifdef TCPDEBUG ipov = (struct ipovly *)ip; #endif #ifdef NETFLIX_TCPOUDP if (tp->t_port) { udp = (struct udphdr *)((caddr_t)ip + ipoptlen + sizeof(struct ip)); udp->uh_sport = htons(V_tcp_udp_tunneling_port); udp->uh_dport = tp->t_port; ulen = hdrlen + len - sizeof(struct ip); udp->uh_ulen = htons(ulen); th = (struct tcphdr *)(udp + 1); } else #endif th = (struct tcphdr *)(ip + 1); tcpip_fillheaders(inp, #ifdef NETFLIX_TCPOUDP tp->t_port, #endif ip, th); } /* * If we are doing retransmissions, then snd_nxt will not reflect * the first unsent octet. For ACK only packets, we do not want the * sequence number of the retransmitted packet, we want the sequence * number of the next unsent octet. So, if there is no data (and no * SYN or FIN), use snd_max instead of snd_nxt when filling in * ti_seq. But if we are in persist state, snd_max might reflect * one byte beyond the right edge of the window, so use snd_nxt in * that case, since we know we aren't doing a retransmission. * (retransmit and persist are mutually exclusive...) */ if (sack_rxmit == 0) { if (len && ((flags & (TH_FIN | TH_SYN | TH_RST)) == 0)) { /* New data (including new persists) */ th->th_seq = htonl(tp->snd_max); bbr_seq = tp->snd_max; } else if (flags & TH_SYN) { /* Syn's always send from iss */ th->th_seq = htonl(tp->iss); bbr_seq = tp->iss; } else if (flags & TH_FIN) { if (flags & TH_FIN && tp->t_flags & TF_SENTFIN) { /* * If we sent the fin already its 1 minus * snd_max */ th->th_seq = (htonl(tp->snd_max - 1)); bbr_seq = (tp->snd_max - 1); } else { /* First time FIN use snd_max */ th->th_seq = htonl(tp->snd_max); bbr_seq = tp->snd_max; } } else if (flags & TH_RST) { /* * For a Reset send the last cum ack in sequence * (this like any other choice may still generate a * challenge ack, if a ack-update packet is in * flight). */ th->th_seq = htonl(tp->snd_una); bbr_seq = tp->snd_una; } else { /* * len == 0 and not persist we use snd_max, sending * an ack unless we have sent the fin then its 1 * minus. */ /* * XXXRRS Question if we are in persists and we have * nothing outstanding to send and we have not sent * a FIN, we will send an ACK. In such a case it * might be better to send (tp->snd_una - 1) which * would force the peer to ack. */ if (tp->t_flags & TF_SENTFIN) { th->th_seq = htonl(tp->snd_max - 1); bbr_seq = (tp->snd_max - 1); } else { th->th_seq = htonl(tp->snd_max); bbr_seq = tp->snd_max; } } } else { /* All retransmits use the rsm to guide the send */ th->th_seq = htonl(rsm->r_start); bbr_seq = rsm->r_start; } th->th_ack = htonl(tp->rcv_nxt); if (optlen) { bcopy(opt, th + 1, optlen); th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; } th->th_flags = flags; /* * Calculate receive window. Don't shrink window, but avoid silly * window syndrome. */ if ((flags & TH_RST) || ((recwin < (so->so_rcv.sb_hiwat / 4) && recwin < maxseg))) recwin = 0; if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && recwin < (tp->rcv_adv - tp->rcv_nxt)) recwin = (tp->rcv_adv - tp->rcv_nxt); if (recwin > TCP_MAXWIN << tp->rcv_scale) recwin = TCP_MAXWIN << tp->rcv_scale; /* * According to RFC1323 the window field in a SYN (i.e., a or * ) segment itself is never scaled. The case is * handled in syncache. */ if (flags & TH_SYN) th->th_win = htons((u_short) (min(sbspace(&so->so_rcv), TCP_MAXWIN))); else th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); /* * Adjust the RXWIN0SENT flag - indicate that we have advertised a 0 * window. This may cause the remote transmitter to stall. This * flag tells soreceive() to disable delayed acknowledgements when * draining the buffer. This can occur if the receiver is * attempting to read more data than can be buffered prior to * transmitting on the connection. */ if (th->th_win == 0) { tp->t_sndzerowin++; tp->t_flags |= TF_RXWIN0SENT; } else tp->t_flags &= ~TF_RXWIN0SENT; if (SEQ_GT(tp->snd_up, tp->snd_max)) { th->th_urp = htons((u_short)(tp->snd_up - tp->snd_max)); th->th_flags |= TH_URG; } else /* * If no urgent pointer to send, then we pull the urgent * pointer to the left edge of the send window so that it * doesn't drift into the send window on sequence number * wraparound. */ tp->snd_up = tp->snd_una; /* drag it along */ #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { /* * Calculate MD5 signature and put it into the place * determined before. NOTE: since TCP options buffer doesn't * point into mbuf's data, calculate offset and use it. */ if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, (u_char *)(th + 1) + (to.to_signature - opt)) != 0) { /* * Do not send segment if the calculation of MD5 * digest has failed. */ goto out; } } #endif /* * Put TCP length in extended header, and then checksum extended * header and data. */ m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ #ifdef INET6 if (isipv6) { /* * ip6_plen is not need to be filled now, and will be filled * in ip6_output. */ #ifdef NETFLIX_TCPOUDP if (tp->t_port) { m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); udp->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); th->th_sum = htons(0); UDPSTAT_INC(udps_opackets); } else { #endif csum_flags = m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + optlen + len, IPPROTO_TCP, 0); #ifdef NETFLIX_TCPOUDP } #endif } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { #ifdef NETFLIX_TCPOUDP if (tp->t_port) { m->m_pkthdr.csum_flags = CSUM_UDP; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); udp->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP)); th->th_sum = htons(0); UDPSTAT_INC(udps_opackets); } else { #endif csum_flags = m->m_pkthdr.csum_flags = CSUM_TCP; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); #ifdef NETFLIX_TCPOUDP } #endif /* IP version must be set here for ipv4/ipv6 checking later */ KASSERT(ip->ip_v == IPVERSION, ("%s: IP version incorrect: %d", __func__, ip->ip_v)); } #endif /* * Enable TSO and specify the size of the segments. The TCP pseudo * header checksum is always provided. XXX: Fixme: This is currently * not the case for IPv6. */ if (tso || force_tso) { KASSERT(force_tso || len > maxseg, ("%s: len:%d <= tso_segsz:%d", __func__, len, maxseg)); m->m_pkthdr.csum_flags |= CSUM_TSO; csum_flags |= CSUM_TSO; m->m_pkthdr.tso_segsz = maxseg; } KASSERT(len + hdrlen == m_length(m, NULL), ("%s: mbuf chain different than expected: %d + %u != %u", __func__, len, hdrlen, m_length(m, NULL))); #ifdef TCP_HHOOK /* Run HHOOK_TC_ESTABLISHED_OUT helper hooks. */ hhook_run_tcp_est_out(tp, th, &to, len, tso); #endif #ifdef TCPDEBUG /* * Trace. */ if (so->so_options & SO_DEBUG) { u_short save = 0; #ifdef INET6 if (!isipv6) #endif { save = ipov->ih_len; ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + * (th->th_off << 2) */ ); } tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0); #ifdef INET6 if (!isipv6) #endif ipov->ih_len = save; } #endif /* TCPDEBUG */ /* Log to the black box */ if (tp->t_logstate != TCP_LOG_STATE_OFF) { union tcp_log_stackspecific log; bbr_fill_in_logging_data(bbr, &log.u_bbr, cts); /* Record info on type of transmission */ log.u_bbr.flex1 = bbr->r_ctl.rc_hptsi_agg_delay; log.u_bbr.flex2 = (bbr->r_recovery_bw << 3); log.u_bbr.flex3 = maxseg; log.u_bbr.flex4 = delay_calc; /* Encode filled_all into the upper flex5 bit */ log.u_bbr.flex5 = bbr->rc_past_init_win; log.u_bbr.flex5 <<= 1; log.u_bbr.flex5 |= bbr->rc_no_pacing; log.u_bbr.flex5 <<= 29; if (filled_all) log.u_bbr.flex5 |= 0x80000000; log.u_bbr.flex5 |= tp->t_maxseg; log.u_bbr.flex6 = bbr->r_ctl.rc_pace_max_segs; log.u_bbr.flex7 = (bbr->rc_bbr_state << 8) | bbr_state_val(bbr); /* lets poke in the low and the high here for debugging */ log.u_bbr.pkts_out = bbr->rc_tp->t_maxseg; if (rsm || sack_rxmit) { if (doing_tlp) log.u_bbr.flex8 = 2; else log.u_bbr.flex8 = 1; } else { log.u_bbr.flex8 = 0; } lgb = tcp_log_event_(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_OUT, ERRNO_UNK, len, &log, false, NULL, NULL, 0, tv); } else { lgb = NULL; } /* * Fill in IP length and desired time to live and send to IP level. * There should be a better way to handle ttl and tos; we could keep * them in the template, but need a way to checksum without them. */ /* * m->m_pkthdr.len should have been set before cksum calcuration, * because in6_cksum() need it. */ #ifdef INET6 if (isipv6) { /* * we separately set hoplimit for every segment, since the * user might want to change the value via setsockopt. Also, * desired default hop limit might be changed via Neighbor * Discovery. */ ip6->ip6_hlim = in6_selecthlim(inp, NULL); /* * Set the packet size here for the benefit of DTrace * probes. ip6_output() will set it properly; it's supposed * to include the option header lengths as well. */ ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); if (V_path_mtu_discovery && maxseg > V_tcp_minmss) tp->t_flags2 |= TF2_PLPMTU_PMTUD; else tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); TCP_PROBE5(send, NULL, tp, ip6, tp, th); /* TODO: IPv6 IP6TOS_ECT bit on */ error = ip6_output(m, inp->in6p_outputopts, &inp->inp_route6, ((rsm || sack_rxmit) ? IP_NO_SND_TAG_RL : 0), NULL, NULL, inp); - if (error == EMSGSIZE && inp->inp_route6.ro_rt != NULL) - mtu = inp->inp_route6.ro_rt->rt_mtu; + if (error == EMSGSIZE && inp->inp_route6.ro_nh != NULL) + mtu = inp->inp_route6.ro_nh->nh_mtu; } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(m->m_pkthdr.len); #ifdef INET6 if (isipv6) ip->ip_ttl = in6_selecthlim(inp, NULL); #endif /* INET6 */ /* * If we do path MTU discovery, then we set DF on every * packet. This might not be the best thing to do according * to RFC3390 Section 2. However the tcp hostcache migitates * the problem so it affects only the first tcp connection * with a host. * * NB: Don't set DF on small MTU/MSS to have a safe * fallback. */ if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { tp->t_flags2 |= TF2_PLPMTU_PMTUD; if (tp->t_port == 0 || len < V_tcp_minmss) { ip->ip_off |= htons(IP_DF); } } else { tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; } if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); TCP_PROBE5(send, NULL, tp, ip, tp, th); error = ip_output(m, inp->inp_options, &inp->inp_route, ((rsm || sack_rxmit) ? IP_NO_SND_TAG_RL : 0), 0, inp); - if (error == EMSGSIZE && inp->inp_route.ro_rt != NULL) - mtu = inp->inp_route.ro_rt->rt_mtu; + if (error == EMSGSIZE && inp->inp_route.ro_nh != NULL) + mtu = inp->inp_route.ro_nh->nh_mtu; } #endif /* INET */ out: if (lgb) { lgb->tlb_errno = error; lgb = NULL; } /* * In transmit state, time the transmission and arrange for the * retransmit. In persist state, just set snd_max. */ if (error == 0) { if (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) tcp_clean_dsack_blocks(tp); /* We sent an ack clear the bbr_segs_rcvd count */ bbr->output_error_seen = 0; bbr->oerror_cnt = 0; bbr->bbr_segs_rcvd = 0; if (len == 0) counter_u64_add(bbr_out_size[TCP_MSS_ACCT_SNDACK], 1); else if (hw_tls) { if (filled_all || (len >= bbr->r_ctl.rc_pace_max_segs)) BBR_STAT_INC(bbr_meets_tso_thresh); else { if (doing_tlp) { BBR_STAT_INC(bbr_miss_tlp); bbr_log_type_hrdwtso(tp, bbr, len, 1, what_we_can); } else if (rsm) { BBR_STAT_INC(bbr_miss_retran); bbr_log_type_hrdwtso(tp, bbr, len, 2, what_we_can); } else if ((ctf_outstanding(tp) + bbr->r_ctl.rc_pace_max_segs) > sbavail(sb)) { BBR_STAT_INC(bbr_miss_tso_app); bbr_log_type_hrdwtso(tp, bbr, len, 3, what_we_can); } else if ((ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) + bbr->r_ctl.rc_pace_max_segs) > tp->snd_cwnd) { BBR_STAT_INC(bbr_miss_tso_cwnd); bbr_log_type_hrdwtso(tp, bbr, len, 4, what_we_can); } else if ((ctf_outstanding(tp) + bbr->r_ctl.rc_pace_max_segs) > tp->snd_wnd) { BBR_STAT_INC(bbr_miss_tso_rwnd); bbr_log_type_hrdwtso(tp, bbr, len, 5, what_we_can); } else { BBR_STAT_INC(bbr_miss_unknown); bbr_log_type_hrdwtso(tp, bbr, len, 6, what_we_can); } } } /* Do accounting for new sends */ if ((len > 0) && (rsm == NULL)) { int idx; if (tp->snd_una == tp->snd_max) { /* * Special case to match google, when * nothing is in flight the delivered * time does get updated to the current * time (see tcp_rate_bsd.c). */ bbr->r_ctl.rc_del_time = cts; } if (len >= maxseg) { idx = (len / maxseg) + 3; if (idx >= TCP_MSS_ACCT_ATIMER) counter_u64_add(bbr_out_size[(TCP_MSS_ACCT_ATIMER - 1)], 1); else counter_u64_add(bbr_out_size[idx], 1); } else { /* smaller than a MSS */ idx = len / (bbr_hptsi_bytes_min - bbr->rc_last_options); if (idx >= TCP_MSS_SMALL_MAX_SIZE_DIV) idx = (TCP_MSS_SMALL_MAX_SIZE_DIV - 1); counter_u64_add(bbr_out_size[(idx + TCP_MSS_SMALL_SIZE_OFF)], 1); } } } abandon = 0; /* * We must do the send accounting before we log the output, * otherwise the state of the rsm could change and we account to the * wrong bucket. */ if (len > 0) { bbr_do_send_accounting(tp, bbr, rsm, len, error); if (error == 0) { if (tp->snd_una == tp->snd_max) bbr->r_ctl.rc_tlp_rxt_last_time = cts; } } bbr_log_output(bbr, tp, &to, len, bbr_seq, (uint8_t) flags, error, cts, mb, &abandon, rsm, 0, sb); if (abandon) { /* * If bbr_log_output destroys the TCB or sees a TH_RST being * sent we should hit this condition. */ return (0); } if (((tp->t_flags & TF_FORCEDATA) == 0) || (bbr->rc_in_persist == 0)) { /* * Advance snd_nxt over sequence space of this segment. */ if (error) /* We don't log or do anything with errors */ goto skip_upd; if (tp->snd_una == tp->snd_max && (len || (flags & (TH_SYN | TH_FIN)))) { /* * Update the time we just added data since none was * outstanding. */ bbr_log_progress_event(bbr, tp, ticks, PROGRESS_START, __LINE__); bbr->rc_tp->t_acktime = ticks; } if (flags & (TH_SYN | TH_FIN) && (rsm == NULL)) { if (flags & TH_SYN) { tp->snd_max++; } if ((flags & TH_FIN) && ((tp->t_flags & TF_SENTFIN) == 0)) { tp->snd_max++; tp->t_flags |= TF_SENTFIN; } } if (sack_rxmit == 0) tp->snd_max += len; skip_upd: if ((error == 0) && len) tot_len += len; } else { /* Persists case */ int32_t xlen = len; if (error) goto nomore; if (flags & TH_SYN) ++xlen; if ((flags & TH_FIN) && ((tp->t_flags & TF_SENTFIN) == 0)) { ++xlen; tp->t_flags |= TF_SENTFIN; } if (xlen && (tp->snd_una == tp->snd_max)) { /* * Update the time we just added data since none was * outstanding. */ bbr_log_progress_event(bbr, tp, ticks, PROGRESS_START, __LINE__); bbr->rc_tp->t_acktime = ticks; } if (sack_rxmit == 0) tp->snd_max += xlen; tot_len += (len + optlen + ipoptlen); } nomore: if (error) { /* * Failures do not advance the seq counter above. For the * case of ENOBUFS we will fall out and become ack-clocked. * capping the cwnd at the current flight. * Everything else will just have to retransmit with the timer * (no pacer). */ SOCKBUF_UNLOCK_ASSERT(sb); BBR_STAT_INC(bbr_saw_oerr); /* Clear all delay/early tracks */ bbr->r_ctl.rc_hptsi_agg_delay = 0; bbr->r_ctl.rc_agg_early = 0; bbr->r_agg_early_set = 0; bbr->output_error_seen = 1; if (bbr->oerror_cnt < 0xf) bbr->oerror_cnt++; if (bbr_max_net_error_cnt && (bbr->oerror_cnt >= bbr_max_net_error_cnt)) { /* drop the session */ tcp_set_inp_to_drop(inp, ENETDOWN); } switch (error) { case ENOBUFS: /* * Make this guy have to get ack's to send * more but lets make sure we don't * slam him below a T-O (1MSS). */ if (bbr->rc_bbr_state != BBR_STATE_PROBE_RTT) { tp->snd_cwnd = ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes)) - maxseg; if (tp->snd_cwnd < maxseg) tp->snd_cwnd = maxseg; } slot = (bbr_error_base_paceout + 1) << bbr->oerror_cnt; BBR_STAT_INC(bbr_saw_enobuf); if (bbr->bbr_hdrw_pacing) counter_u64_add(bbr_hdwr_pacing_enobuf, 1); else counter_u64_add(bbr_nohdwr_pacing_enobuf, 1); /* * Here even in the enobuf's case we want to do our * state update. The reason being we may have been * called by the input function. If so we have had * things change. */ error = 0; goto enobufs; case EMSGSIZE: /* * For some reason the interface we used initially * to send segments changed to another or lowered * its MTU. If TSO was active we either got an * interface without TSO capabilits or TSO was * turned off. If we obtained mtu from ip_output() * then update it and try again. */ /* Turn on tracing (or try to) */ { int old_maxseg; old_maxseg = tp->t_maxseg; BBR_STAT_INC(bbr_saw_emsgsiz); bbr_log_msgsize_fail(bbr, tp, len, maxseg, mtu, csum_flags, tso, cts); if (mtu != 0) tcp_mss_update(tp, -1, mtu, NULL, NULL); if (old_maxseg <= tp->t_maxseg) { /* Huh it did not shrink? */ tp->t_maxseg = old_maxseg - 40; bbr_log_msgsize_fail(bbr, tp, len, maxseg, mtu, 0, tso, cts); } tp->t_flags &= ~TF_FORCEDATA; /* * Nuke all other things that can interfere * with slot */ if ((tot_len + len) && (len >= tp->t_maxseg)) { slot = bbr_get_pacing_delay(bbr, bbr->r_ctl.rc_bbr_hptsi_gain, (tot_len + len), cts, 0); if (slot < bbr_error_base_paceout) slot = (bbr_error_base_paceout + 2) << bbr->oerror_cnt; } else slot = (bbr_error_base_paceout + 2) << bbr->oerror_cnt; bbr->rc_output_starts_timer = 1; bbr_start_hpts_timer(bbr, tp, cts, 10, slot, tot_len); return (error); } case EPERM: tp->t_softerror = error; /* Fall through */ case EHOSTDOWN: case EHOSTUNREACH: case ENETDOWN: case ENETUNREACH: if (TCPS_HAVERCVDSYN(tp->t_state)) { tp->t_softerror = error; } /* FALLTHROUGH */ default: tp->t_flags &= ~TF_FORCEDATA; slot = (bbr_error_base_paceout + 3) << bbr->oerror_cnt; bbr->rc_output_starts_timer = 1; bbr_start_hpts_timer(bbr, tp, cts, 11, slot, 0); return (error); } #ifdef STATS } else if (((tp->t_flags & TF_GPUTINPROG) == 0) && len && (rsm == NULL) && (bbr->rc_in_persist == 0)) { tp->gput_seq = bbr_seq; tp->gput_ack = bbr_seq + min(sbavail(&so->so_snd) - sb_offset, sendwin); tp->gput_ts = cts; tp->t_flags |= TF_GPUTINPROG; #endif } KMOD_TCPSTAT_INC(tcps_sndtotal); if ((bbr->bbr_hdw_pace_ena) && (bbr->bbr_attempt_hdwr_pace == 0) && (bbr->rc_past_init_win) && (bbr->rc_bbr_state != BBR_STATE_STARTUP) && (get_filter_value(&bbr->r_ctl.rc_delrate)) && - (inp->inp_route.ro_rt && - inp->inp_route.ro_rt->rt_ifp)) { + (inp->inp_route.ro_nh && + inp->inp_route.ro_nh->nh_ifp)) { /* * We are past the initial window and * have at least one measurement so we * could use hardware pacing if its available. * We have an interface and we have not attempted * to setup hardware pacing, lets try to now. */ uint64_t rate_wanted; int err = 0; rate_wanted = bbr_get_hardware_rate(bbr); bbr->bbr_attempt_hdwr_pace = 1; bbr->r_ctl.crte = tcp_set_pacing_rate(bbr->rc_tp, - inp->inp_route.ro_rt->rt_ifp, + inp->inp_route.ro_nh->nh_ifp, rate_wanted, (RS_PACING_GEQ|RS_PACING_SUB_OK), &err); if (bbr->r_ctl.crte) { bbr_type_log_hdwr_pacing(bbr, bbr->r_ctl.crte->ptbl->rs_ifp, rate_wanted, bbr->r_ctl.crte->rate, __LINE__, cts, err); BBR_STAT_INC(bbr_hdwr_rl_add_ok); counter_u64_add(bbr_flows_nohdwr_pacing, -1); counter_u64_add(bbr_flows_whdwr_pacing, 1); bbr->bbr_hdrw_pacing = 1; /* Now what is our gain status? */ if (bbr->r_ctl.crte->rate < rate_wanted) { /* We have a problem */ bbr_setup_less_of_rate(bbr, cts, bbr->r_ctl.crte->rate, rate_wanted); } else { /* We are good */ bbr->gain_is_limited = 0; bbr->skip_gain = 0; } tcp_bbr_tso_size_check(bbr, cts); } else { bbr_type_log_hdwr_pacing(bbr, - inp->inp_route.ro_rt->rt_ifp, + inp->inp_route.ro_nh->nh_ifp, rate_wanted, 0, __LINE__, cts, err); BBR_STAT_INC(bbr_hdwr_rl_add_fail); } } if (bbr->bbr_hdrw_pacing) { /* * Worry about cases where the route * changes or something happened that we * lost our hardware pacing possibly during * the last ip_output call. */ if (inp->inp_snd_tag == NULL) { /* A change during ip output disabled hw pacing? */ bbr->bbr_hdrw_pacing = 0; - } else if ((inp->inp_route.ro_rt == NULL) || - (inp->inp_route.ro_rt->rt_ifp != inp->inp_snd_tag->ifp)) { + } else if ((inp->inp_route.ro_nh == NULL) || + (inp->inp_route.ro_nh->nh_ifp != inp->inp_snd_tag->ifp)) { /* * We had an interface or route change, * detach from the current hdwr pacing * and setup to re-attempt next go * round. */ bbr->bbr_hdrw_pacing = 0; bbr->bbr_attempt_hdwr_pace = 0; tcp_rel_pacing_rate(bbr->r_ctl.crte, bbr->rc_tp); tcp_bbr_tso_size_check(bbr, cts); } } /* * Data sent (as far as we can tell). If this advertises a larger * window than any other segment, then remember the size of the * advertised window. Any pending ACK has now been sent. */ if (SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) tp->rcv_adv = tp->rcv_nxt + recwin; tp->last_ack_sent = tp->rcv_nxt; if ((error == 0) && (bbr->r_ctl.rc_pace_max_segs > tp->t_maxseg) && (doing_tlp == 0) && (tso == 0) && (hw_tls == 0) && (len > 0) && ((flags & TH_RST) == 0) && (IN_RECOVERY(tp->t_flags) == 0) && (bbr->rc_in_persist == 0) && ((tp->t_flags & TF_FORCEDATA) == 0) && (tot_len < bbr->r_ctl.rc_pace_max_segs)) { /* * For non-tso we need to goto again until we have sent out * enough data to match what we are hptsi out every hptsi * interval. */ if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { /* Make sure snd_nxt is drug up */ tp->snd_nxt = tp->snd_max; } if (rsm != NULL) { rsm = NULL; goto skip_again; } rsm = NULL; sack_rxmit = 0; tp->t_flags &= ~(TF_ACKNOW | TF_DELACK | TF_FORCEDATA); goto again; } skip_again: if (((flags & (TH_RST | TH_SYN | TH_FIN)) == 0) && tot_len) { /* * Calculate/Re-Calculate the hptsi slot in usecs based on * what we have sent so far */ slot = bbr_get_pacing_delay(bbr, bbr->r_ctl.rc_bbr_hptsi_gain, tot_len, cts, 0); if (bbr->rc_no_pacing) slot = 0; } tp->t_flags &= ~(TF_ACKNOW | TF_DELACK | TF_FORCEDATA); enobufs: if (bbr->rc_use_google == 0) bbr_check_bbr_for_state(bbr, cts, __LINE__, 0); bbr_cwnd_limiting(tp, bbr, ctf_flight_size(tp, (bbr->r_ctl.rc_sacked + bbr->r_ctl.rc_lost_bytes))); bbr->rc_output_starts_timer = 1; if (bbr->bbr_use_rack_cheat && (more_to_rxt || ((bbr->r_ctl.rc_resend = bbr_check_recovery_mode(tp, bbr, cts)) != NULL))) { /* Rack cheats and shotguns out all rxt's 1ms apart */ if (slot > 1000) slot = 1000; } if (bbr->bbr_hdrw_pacing && (bbr->hw_pacing_set == 0)) { /* * We don't change the tso size until some number of sends * to give the hardware commands time to get down * to the interface. */ bbr->r_ctl.bbr_hdwr_cnt_noset_snt++; if (bbr->r_ctl.bbr_hdwr_cnt_noset_snt >= bbr_hdwr_pacing_delay_cnt) { bbr->hw_pacing_set = 1; tcp_bbr_tso_size_check(bbr, cts); } } bbr_start_hpts_timer(bbr, tp, cts, 12, slot, tot_len); if (SEQ_LT(tp->snd_nxt, tp->snd_max)) { /* Make sure snd_nxt is drug up */ tp->snd_nxt = tp->snd_max; } return (error); } /* * See bbr_output_wtime() for return values. */ static int bbr_output(struct tcpcb *tp) { int32_t ret; struct timeval tv; struct tcp_bbr *bbr; NET_EPOCH_ASSERT(); bbr = (struct tcp_bbr *)tp->t_fb_ptr; INP_WLOCK_ASSERT(tp->t_inpcb); (void)tcp_get_usecs(&tv); ret = bbr_output_wtime(tp, &tv); return (ret); } static void bbr_mtu_chg(struct tcpcb *tp) { struct tcp_bbr *bbr; struct bbr_sendmap *rsm, *frsm = NULL; uint32_t maxseg; /* * The MTU has changed. a) Clear the sack filter. b) Mark everything * over the current size as SACK_PASS so a retransmit will occur. */ bbr = (struct tcp_bbr *)tp->t_fb_ptr; maxseg = tp->t_maxseg - bbr->rc_last_options; sack_filter_clear(&bbr->r_ctl.bbr_sf, tp->snd_una); TAILQ_FOREACH(rsm, &bbr->r_ctl.rc_map, r_next) { /* Don't mess with ones acked (by sack?) */ if (rsm->r_flags & BBR_ACKED) continue; if ((rsm->r_end - rsm->r_start) > maxseg) { /* * We mark sack-passed on all the previous large * sends we did. This will force them to retransmit. */ rsm->r_flags |= BBR_SACK_PASSED; if (((rsm->r_flags & BBR_MARKED_LOST) == 0) && bbr_is_lost(bbr, rsm, bbr->r_ctl.rc_rcvtime)) { bbr->r_ctl.rc_lost_bytes += rsm->r_end - rsm->r_start; bbr->r_ctl.rc_lost += rsm->r_end - rsm->r_start; rsm->r_flags |= BBR_MARKED_LOST; } if (frsm == NULL) frsm = rsm; } } if (frsm) { bbr->r_ctl.rc_resend = frsm; } } /* * bbr_ctloutput() must drop the inpcb lock before performing copyin on * socket option arguments. When it re-acquires the lock after the copy, it * has to revalidate that the connection is still valid for the socket * option. */ static int bbr_set_sockopt(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp, struct tcp_bbr *bbr) { int32_t error = 0, optval; switch (sopt->sopt_name) { case TCP_RACK_PACE_MAX_SEG: case TCP_RACK_MIN_TO: case TCP_RACK_REORD_THRESH: case TCP_RACK_REORD_FADE: case TCP_RACK_TLP_THRESH: case TCP_RACK_PKT_DELAY: case TCP_BBR_ALGORITHM: case TCP_BBR_TSLIMITS: case TCP_BBR_IWINTSO: case TCP_BBR_RECFORCE: case TCP_BBR_STARTUP_PG: case TCP_BBR_DRAIN_PG: case TCP_BBR_RWND_IS_APP: case TCP_BBR_PROBE_RTT_INT: case TCP_BBR_PROBE_RTT_GAIN: case TCP_BBR_PROBE_RTT_LEN: case TCP_BBR_STARTUP_LOSS_EXIT: case TCP_BBR_USEDEL_RATE: case TCP_BBR_MIN_RTO: case TCP_BBR_MAX_RTO: case TCP_BBR_PACE_PER_SEC: case TCP_DELACK: case TCP_BBR_PACE_DEL_TAR: case TCP_BBR_SEND_IWND_IN_TSO: case TCP_BBR_EXTRA_STATE: case TCP_BBR_UTTER_MAX_TSO: case TCP_BBR_MIN_TOPACEOUT: case TCP_BBR_FLOOR_MIN_TSO: case TCP_BBR_TSTMP_RAISES: case TCP_BBR_POLICER_DETECT: case TCP_BBR_USE_RACK_CHEAT: case TCP_DATA_AFTER_CLOSE: case TCP_BBR_HDWR_PACE: case TCP_BBR_PACE_SEG_MAX: case TCP_BBR_PACE_SEG_MIN: case TCP_BBR_PACE_CROSS: case TCP_BBR_PACE_OH: #ifdef NETFLIX_PEAKRATE case TCP_MAXPEAKRATE: #endif case TCP_BBR_TMR_PACE_OH: case TCP_BBR_RACK_RTT_USE: case TCP_BBR_RETRAN_WTSO: break; default: return (tcp_default_ctloutput(so, sopt, inp, tp)); break; } INP_WUNLOCK(inp); error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) return (error); INP_WLOCK(inp); if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_WUNLOCK(inp); return (ECONNRESET); } tp = intotcpcb(inp); bbr = (struct tcp_bbr *)tp->t_fb_ptr; switch (sopt->sopt_name) { case TCP_BBR_PACE_PER_SEC: BBR_OPTS_INC(tcp_bbr_pace_per_sec); bbr->r_ctl.bbr_hptsi_per_second = optval; break; case TCP_BBR_PACE_DEL_TAR: BBR_OPTS_INC(tcp_bbr_pace_del_tar); bbr->r_ctl.bbr_hptsi_segments_delay_tar = optval; break; case TCP_BBR_PACE_SEG_MAX: BBR_OPTS_INC(tcp_bbr_pace_seg_max); bbr->r_ctl.bbr_hptsi_segments_max = optval; break; case TCP_BBR_PACE_SEG_MIN: BBR_OPTS_INC(tcp_bbr_pace_seg_min); bbr->r_ctl.bbr_hptsi_bytes_min = optval; break; case TCP_BBR_PACE_CROSS: BBR_OPTS_INC(tcp_bbr_pace_cross); bbr->r_ctl.bbr_cross_over = optval; break; case TCP_BBR_ALGORITHM: BBR_OPTS_INC(tcp_bbr_algorithm); if (optval && (bbr->rc_use_google == 0)) { /* Turn on the google mode */ bbr_google_mode_on(bbr); if ((optval > 3) && (optval < 500)) { /* * Must be at least greater than .3% * and must be less than 50.0%. */ bbr->r_ctl.bbr_google_discount = optval; } } else if ((optval == 0) && (bbr->rc_use_google == 1)) { /* Turn off the google mode */ bbr_google_mode_off(bbr); } break; case TCP_BBR_TSLIMITS: BBR_OPTS_INC(tcp_bbr_tslimits); if (optval == 1) bbr->rc_use_ts_limit = 1; else if (optval == 0) bbr->rc_use_ts_limit = 0; else error = EINVAL; break; case TCP_BBR_IWINTSO: BBR_OPTS_INC(tcp_bbr_iwintso); if ((optval >= 0) && (optval < 128)) { uint32_t twin; bbr->rc_init_win = optval; twin = bbr_initial_cwnd(bbr, tp); if ((bbr->rc_past_init_win == 0) && (twin > tp->snd_cwnd)) tp->snd_cwnd = twin; else error = EBUSY; } else error = EINVAL; break; case TCP_BBR_STARTUP_PG: BBR_OPTS_INC(tcp_bbr_startup_pg); if ((optval > 0) && (optval < BBR_MAX_GAIN_VALUE)) { bbr->r_ctl.rc_startup_pg = optval; if (bbr->rc_bbr_state == BBR_STATE_STARTUP) { bbr->r_ctl.rc_bbr_hptsi_gain = optval; } } else error = EINVAL; break; case TCP_BBR_DRAIN_PG: BBR_OPTS_INC(tcp_bbr_drain_pg); if ((optval > 0) && (optval < BBR_MAX_GAIN_VALUE)) bbr->r_ctl.rc_drain_pg = optval; else error = EINVAL; break; case TCP_BBR_PROBE_RTT_LEN: BBR_OPTS_INC(tcp_bbr_probertt_len); if (optval <= 1) reset_time_small(&bbr->r_ctl.rc_rttprop, (optval * USECS_IN_SECOND)); else error = EINVAL; break; case TCP_BBR_PROBE_RTT_GAIN: BBR_OPTS_INC(tcp_bbr_probertt_gain); if (optval <= BBR_UNIT) bbr->r_ctl.bbr_rttprobe_gain_val = optval; else error = EINVAL; break; case TCP_BBR_PROBE_RTT_INT: BBR_OPTS_INC(tcp_bbr_probe_rtt_int); if (optval > 1000) bbr->r_ctl.rc_probertt_int = optval; else error = EINVAL; break; case TCP_BBR_MIN_TOPACEOUT: BBR_OPTS_INC(tcp_bbr_topaceout); if (optval == 0) { bbr->no_pacing_until = 0; bbr->rc_no_pacing = 0; } else if (optval <= 0x00ff) { bbr->no_pacing_until = optval; if ((bbr->r_ctl.rc_pkt_epoch < bbr->no_pacing_until) && (bbr->rc_bbr_state == BBR_STATE_STARTUP)){ /* Turn on no pacing */ bbr->rc_no_pacing = 1; } } else error = EINVAL; break; case TCP_BBR_STARTUP_LOSS_EXIT: BBR_OPTS_INC(tcp_bbr_startup_loss_exit); bbr->rc_loss_exit = optval; break; case TCP_BBR_USEDEL_RATE: error = EINVAL; break; case TCP_BBR_MIN_RTO: BBR_OPTS_INC(tcp_bbr_min_rto); bbr->r_ctl.rc_min_rto_ms = optval; break; case TCP_BBR_MAX_RTO: BBR_OPTS_INC(tcp_bbr_max_rto); bbr->rc_max_rto_sec = optval; break; case TCP_RACK_MIN_TO: /* Minimum time between rack t-o's in ms */ BBR_OPTS_INC(tcp_rack_min_to); bbr->r_ctl.rc_min_to = optval; break; case TCP_RACK_REORD_THRESH: /* RACK reorder threshold (shift amount) */ BBR_OPTS_INC(tcp_rack_reord_thresh); if ((optval > 0) && (optval < 31)) bbr->r_ctl.rc_reorder_shift = optval; else error = EINVAL; break; case TCP_RACK_REORD_FADE: /* Does reordering fade after ms time */ BBR_OPTS_INC(tcp_rack_reord_fade); bbr->r_ctl.rc_reorder_fade = optval; break; case TCP_RACK_TLP_THRESH: /* RACK TLP theshold i.e. srtt+(srtt/N) */ BBR_OPTS_INC(tcp_rack_tlp_thresh); if (optval) bbr->rc_tlp_threshold = optval; else error = EINVAL; break; case TCP_BBR_USE_RACK_CHEAT: BBR_OPTS_INC(tcp_use_rackcheat); if (bbr->rc_use_google) { error = EINVAL; break; } BBR_OPTS_INC(tcp_rack_cheat); if (optval) bbr->bbr_use_rack_cheat = 1; else bbr->bbr_use_rack_cheat = 0; break; case TCP_BBR_FLOOR_MIN_TSO: BBR_OPTS_INC(tcp_utter_max_tso); if ((optval >= 0) && (optval < 40)) bbr->r_ctl.bbr_hptsi_segments_floor = optval; else error = EINVAL; break; case TCP_BBR_UTTER_MAX_TSO: BBR_OPTS_INC(tcp_utter_max_tso); if ((optval >= 0) && (optval < 0xffff)) bbr->r_ctl.bbr_utter_max = optval; else error = EINVAL; break; case TCP_BBR_EXTRA_STATE: BBR_OPTS_INC(tcp_extra_state); if (optval) bbr->rc_use_idle_restart = 1; else bbr->rc_use_idle_restart = 0; break; case TCP_BBR_SEND_IWND_IN_TSO: BBR_OPTS_INC(tcp_iwnd_tso); if (optval) { bbr->bbr_init_win_cheat = 1; if (bbr->rc_past_init_win == 0) { uint32_t cts; cts = tcp_get_usecs(&bbr->rc_tv); tcp_bbr_tso_size_check(bbr, cts); } } else bbr->bbr_init_win_cheat = 0; break; case TCP_BBR_HDWR_PACE: BBR_OPTS_INC(tcp_hdwr_pacing); if (optval){ bbr->bbr_hdw_pace_ena = 1; bbr->bbr_attempt_hdwr_pace = 0; } else { bbr->bbr_hdw_pace_ena = 0; #ifdef RATELIMIT if (bbr->bbr_hdrw_pacing) { bbr->bbr_hdrw_pacing = 0; in_pcbdetach_txrtlmt(bbr->rc_inp); } #endif } break; case TCP_DELACK: BBR_OPTS_INC(tcp_delack); if (optval < 100) { if (optval == 0) /* off */ tp->t_delayed_ack = 0; else if (optval == 1) /* on which is 2 */ tp->t_delayed_ack = 2; else /* higher than 2 and less than 100 */ tp->t_delayed_ack = optval; if (tp->t_flags & TF_DELACK) { tp->t_flags &= ~TF_DELACK; tp->t_flags |= TF_ACKNOW; bbr_output(tp); } } else error = EINVAL; break; case TCP_RACK_PKT_DELAY: /* RACK added ms i.e. rack-rtt + reord + N */ BBR_OPTS_INC(tcp_rack_pkt_delay); bbr->r_ctl.rc_pkt_delay = optval; break; #ifdef NETFLIX_PEAKRATE case TCP_MAXPEAKRATE: BBR_OPTS_INC(tcp_maxpeak); error = tcp_set_maxpeakrate(tp, optval); if (!error) tp->t_peakrate_thr = tp->t_maxpeakrate; break; #endif case TCP_BBR_RETRAN_WTSO: BBR_OPTS_INC(tcp_retran_wtso); if (optval) bbr->rc_resends_use_tso = 1; else bbr->rc_resends_use_tso = 0; break; case TCP_DATA_AFTER_CLOSE: BBR_OPTS_INC(tcp_data_ac); if (optval) bbr->rc_allow_data_af_clo = 1; else bbr->rc_allow_data_af_clo = 0; break; case TCP_BBR_POLICER_DETECT: BBR_OPTS_INC(tcp_policer_det); if (bbr->rc_use_google == 0) error = EINVAL; else if (optval) bbr->r_use_policer = 1; else bbr->r_use_policer = 0; break; case TCP_BBR_TSTMP_RAISES: BBR_OPTS_INC(tcp_ts_raises); if (optval) bbr->ts_can_raise = 1; else bbr->ts_can_raise = 0; break; case TCP_BBR_TMR_PACE_OH: BBR_OPTS_INC(tcp_pacing_oh_tmr); if (bbr->rc_use_google) { error = EINVAL; } else { if (optval) bbr->r_ctl.rc_incr_tmrs = 1; else bbr->r_ctl.rc_incr_tmrs = 0; } break; case TCP_BBR_PACE_OH: BBR_OPTS_INC(tcp_pacing_oh); if (bbr->rc_use_google) { error = EINVAL; } else { if (optval > (BBR_INCL_TCP_OH| BBR_INCL_IP_OH| BBR_INCL_ENET_OH)) { error = EINVAL; break; } if (optval & BBR_INCL_TCP_OH) bbr->r_ctl.rc_inc_tcp_oh = 1; else bbr->r_ctl.rc_inc_tcp_oh = 0; if (optval & BBR_INCL_IP_OH) bbr->r_ctl.rc_inc_ip_oh = 1; else bbr->r_ctl.rc_inc_ip_oh = 0; if (optval & BBR_INCL_ENET_OH) bbr->r_ctl.rc_inc_enet_oh = 1; else bbr->r_ctl.rc_inc_enet_oh = 0; } break; default: return (tcp_default_ctloutput(so, sopt, inp, tp)); break; } #ifdef NETFLIX_STATS tcp_log_socket_option(tp, sopt->sopt_name, optval, error); #endif INP_WUNLOCK(inp); return (error); } /* * return 0 on success, error-num on failure */ static int bbr_get_sockopt(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp, struct tcp_bbr *bbr) { int32_t error, optval; /* * Because all our options are either boolean or an int, we can just * pull everything into optval and then unlock and copy. If we ever * add a option that is not a int, then this will have quite an * impact to this routine. */ switch (sopt->sopt_name) { case TCP_BBR_PACE_PER_SEC: optval = bbr->r_ctl.bbr_hptsi_per_second; break; case TCP_BBR_PACE_DEL_TAR: optval = bbr->r_ctl.bbr_hptsi_segments_delay_tar; break; case TCP_BBR_PACE_SEG_MAX: optval = bbr->r_ctl.bbr_hptsi_segments_max; break; case TCP_BBR_MIN_TOPACEOUT: optval = bbr->no_pacing_until; break; case TCP_BBR_PACE_SEG_MIN: optval = bbr->r_ctl.bbr_hptsi_bytes_min; break; case TCP_BBR_PACE_CROSS: optval = bbr->r_ctl.bbr_cross_over; break; case TCP_BBR_ALGORITHM: optval = bbr->rc_use_google; break; case TCP_BBR_TSLIMITS: optval = bbr->rc_use_ts_limit; break; case TCP_BBR_IWINTSO: optval = bbr->rc_init_win; break; case TCP_BBR_STARTUP_PG: optval = bbr->r_ctl.rc_startup_pg; break; case TCP_BBR_DRAIN_PG: optval = bbr->r_ctl.rc_drain_pg; break; case TCP_BBR_PROBE_RTT_INT: optval = bbr->r_ctl.rc_probertt_int; break; case TCP_BBR_PROBE_RTT_LEN: optval = (bbr->r_ctl.rc_rttprop.cur_time_limit / USECS_IN_SECOND); break; case TCP_BBR_PROBE_RTT_GAIN: optval = bbr->r_ctl.bbr_rttprobe_gain_val; break; case TCP_BBR_STARTUP_LOSS_EXIT: optval = bbr->rc_loss_exit; break; case TCP_BBR_USEDEL_RATE: error = EINVAL; break; case TCP_BBR_MIN_RTO: optval = bbr->r_ctl.rc_min_rto_ms; break; case TCP_BBR_MAX_RTO: optval = bbr->rc_max_rto_sec; break; case TCP_RACK_PACE_MAX_SEG: /* Max segments in a pace */ optval = bbr->r_ctl.rc_pace_max_segs; break; case TCP_RACK_MIN_TO: /* Minimum time between rack t-o's in ms */ optval = bbr->r_ctl.rc_min_to; break; case TCP_RACK_REORD_THRESH: /* RACK reorder threshold (shift amount) */ optval = bbr->r_ctl.rc_reorder_shift; break; case TCP_RACK_REORD_FADE: /* Does reordering fade after ms time */ optval = bbr->r_ctl.rc_reorder_fade; break; case TCP_BBR_USE_RACK_CHEAT: /* Do we use the rack cheat for rxt */ optval = bbr->bbr_use_rack_cheat; break; case TCP_BBR_FLOOR_MIN_TSO: optval = bbr->r_ctl.bbr_hptsi_segments_floor; break; case TCP_BBR_UTTER_MAX_TSO: optval = bbr->r_ctl.bbr_utter_max; break; case TCP_BBR_SEND_IWND_IN_TSO: /* Do we send TSO size segments initially */ optval = bbr->bbr_init_win_cheat; break; case TCP_BBR_EXTRA_STATE: optval = bbr->rc_use_idle_restart; break; case TCP_RACK_TLP_THRESH: /* RACK TLP theshold i.e. srtt+(srtt/N) */ optval = bbr->rc_tlp_threshold; break; case TCP_RACK_PKT_DELAY: /* RACK added ms i.e. rack-rtt + reord + N */ optval = bbr->r_ctl.rc_pkt_delay; break; case TCP_BBR_RETRAN_WTSO: optval = bbr->rc_resends_use_tso; break; case TCP_DATA_AFTER_CLOSE: optval = bbr->rc_allow_data_af_clo; break; case TCP_DELACK: optval = tp->t_delayed_ack; break; case TCP_BBR_HDWR_PACE: optval = bbr->bbr_hdw_pace_ena; break; case TCP_BBR_POLICER_DETECT: optval = bbr->r_use_policer; break; case TCP_BBR_TSTMP_RAISES: optval = bbr->ts_can_raise; break; case TCP_BBR_TMR_PACE_OH: optval = bbr->r_ctl.rc_incr_tmrs; break; case TCP_BBR_PACE_OH: optval = 0; if (bbr->r_ctl.rc_inc_tcp_oh) optval |= BBR_INCL_TCP_OH; if (bbr->r_ctl.rc_inc_ip_oh) optval |= BBR_INCL_IP_OH; if (bbr->r_ctl.rc_inc_enet_oh) optval |= BBR_INCL_ENET_OH; break; default: return (tcp_default_ctloutput(so, sopt, inp, tp)); break; } INP_WUNLOCK(inp); error = sooptcopyout(sopt, &optval, sizeof optval); return (error); } /* * return 0 on success, error-num on failure */ static int bbr_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp) { int32_t error = EINVAL; struct tcp_bbr *bbr; bbr = (struct tcp_bbr *)tp->t_fb_ptr; if (bbr == NULL) { /* Huh? */ goto out; } if (sopt->sopt_dir == SOPT_SET) { return (bbr_set_sockopt(so, sopt, inp, tp, bbr)); } else if (sopt->sopt_dir == SOPT_GET) { return (bbr_get_sockopt(so, sopt, inp, tp, bbr)); } out: INP_WUNLOCK(inp); return (error); } struct tcp_function_block __tcp_bbr = { .tfb_tcp_block_name = __XSTRING(STACKNAME), .tfb_tcp_output = bbr_output, .tfb_do_queued_segments = ctf_do_queued_segments, .tfb_do_segment_nounlock = bbr_do_segment_nounlock, .tfb_tcp_do_segment = bbr_do_segment, .tfb_tcp_ctloutput = bbr_ctloutput, .tfb_tcp_fb_init = bbr_init, .tfb_tcp_fb_fini = bbr_fini, .tfb_tcp_timer_stop_all = bbr_stopall, .tfb_tcp_timer_activate = bbr_timer_activate, .tfb_tcp_timer_active = bbr_timer_active, .tfb_tcp_timer_stop = bbr_timer_stop, .tfb_tcp_rexmit_tmr = bbr_remxt_tmr, .tfb_tcp_handoff_ok = bbr_handoff_ok, .tfb_tcp_mtu_chg = bbr_mtu_chg }; static const char *bbr_stack_names[] = { __XSTRING(STACKNAME), #ifdef STACKALIAS __XSTRING(STACKALIAS), #endif }; static bool bbr_mod_inited = false; static int tcp_addbbr(module_t mod, int32_t type, void *data) { int32_t err = 0; int num_stacks; switch (type) { case MOD_LOAD: printf("Attempting to load " __XSTRING(MODNAME) "\n"); bbr_zone = uma_zcreate(__XSTRING(MODNAME) "_map", sizeof(struct bbr_sendmap), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); bbr_pcb_zone = uma_zcreate(__XSTRING(MODNAME) "_pcb", sizeof(struct tcp_bbr), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); sysctl_ctx_init(&bbr_sysctl_ctx); bbr_sysctl_root = SYSCTL_ADD_NODE(&bbr_sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp), OID_AUTO, #ifdef STACKALIAS __XSTRING(STACKALIAS), #else __XSTRING(STACKNAME), #endif CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); if (bbr_sysctl_root == NULL) { printf("Failed to add sysctl node\n"); err = EFAULT; goto free_uma; } bbr_init_sysctls(); num_stacks = nitems(bbr_stack_names); err = register_tcp_functions_as_names(&__tcp_bbr, M_WAITOK, bbr_stack_names, &num_stacks); if (err) { printf("Failed to register %s stack name for " "%s module\n", bbr_stack_names[num_stacks], __XSTRING(MODNAME)); sysctl_ctx_free(&bbr_sysctl_ctx); free_uma: uma_zdestroy(bbr_zone); uma_zdestroy(bbr_pcb_zone); bbr_counter_destroy(); printf("Failed to register " __XSTRING(MODNAME) " module err:%d\n", err); return (err); } tcp_lro_reg_mbufq(); bbr_mod_inited = true; printf(__XSTRING(MODNAME) " is now available\n"); break; case MOD_QUIESCE: err = deregister_tcp_functions(&__tcp_bbr, true, false); break; case MOD_UNLOAD: err = deregister_tcp_functions(&__tcp_bbr, false, true); if (err == EBUSY) break; if (bbr_mod_inited) { uma_zdestroy(bbr_zone); uma_zdestroy(bbr_pcb_zone); sysctl_ctx_free(&bbr_sysctl_ctx); bbr_counter_destroy(); printf(__XSTRING(MODNAME) " is now no longer available\n"); bbr_mod_inited = false; } tcp_lro_dereg_mbufq(); err = 0; break; default: return (EOPNOTSUPP); } return (err); } static moduledata_t tcp_bbr = { .name = __XSTRING(MODNAME), .evhand = tcp_addbbr, .priv = 0 }; MODULE_VERSION(MODNAME, 1); DECLARE_MODULE(MODNAME, tcp_bbr, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); MODULE_DEPEND(MODNAME, tcphpts, 1, 1, 1); Index: head/sys/netinet/tcp_subr.c =================================================================== --- head/sys/netinet/tcp_subr.c (revision 360291) +++ head/sys/netinet/tcp_subr.c (revision 360292) @@ -1,3466 +1,3467 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_tcpdebug.h" #include #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef TCP_HHOOK #include #endif #ifdef KERN_TLS #include #endif #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #include #include #endif #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include #include VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; #ifdef INET6 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; #endif #ifdef NETFLIX_EXP_DETECTION /* Sack attack detection thresholds and such */ SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Sack Attack detection thresholds"); int32_t tcp_force_detection = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, CTLFLAG_RW, &tcp_force_detection, 0, "Do we force detection even if the INP has it off?"); int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, CTLFLAG_RW, &tcp_sack_to_ack_thresh, 700, "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, CTLFLAG_RW, &tcp_sack_to_move_thresh, 600, "Percentage of sack moves we must see above (10.1 percent is 101)"); int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, CTLFLAG_RW, &tcp_restoral_thresh, 550, "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); int32_t tcp_sad_decay_val = 800; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, CTLFLAG_RW, &tcp_sad_decay_val, 800, "The decay percentage (10.1 percent equals 101 )"); int32_t tcp_map_minimum = 500; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, CTLFLAG_RW, &tcp_map_minimum, 500, "Number of Map enteries before we start detection"); int32_t tcp_attack_on_turns_on_logging = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, CTLFLAG_RW, &tcp_attack_on_turns_on_logging, 0, "When we have a positive hit on attack, do we turn on logging?"); int32_t tcp_sad_pacing_interval = 2000; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, CTLFLAG_RW, &tcp_sad_pacing_interval, 2000, "What is the minimum pacing interval for a classified attacker?"); int32_t tcp_sad_low_pps = 100; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, CTLFLAG_RW, &tcp_sad_low_pps, 100, "What is the input pps that below which we do not decay?"); #endif struct rwlock tcp_function_lock; static int sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", "Default TCP Maximum Segment Size"); #ifdef INET6 static int sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_v6mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_v6mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", "Default TCP Maximum Segment Size for IPv6"); #endif /* INET6 */ /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds * of packets instead of one. The effect scales with the available * bandwidth and quickly saturates the CPU and network interface * with packet generation and sending. Set to zero to disable MINMSS * checking. This setting prevents us from sending too small packets. */ VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_minmss), 0, "Minimum TCP Maximum Segment Size"); VNET_DEFINE(int, tcp_do_rfc1323) = 1; SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc1323), 0, "Enable rfc1323 (high performance TCP) extensions"); VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ts_offset_per_conn), 0, "Initialize TCP timestamps per connection instead of per host pair"); static int tcp_log_debug = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); static int tcp_tcbhashsize; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); static int do_tcpdrain = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; #define V_icmp_may_rst VNET(icmp_may_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_may_rst), 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_isn_reseed_interval), 0, "Seconds between reseeding of ISN secret"); static int tcp_soreceive_stream; SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); VNET_DEFINE(uma_zone_t, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ static int sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_tcp_map_entries_limit; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { /* only allow "0" and value > minimum */ if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) error = EINVAL; else V_tcp_map_entries_limit = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(tcp_map_entries_limit), 0, &sysctl_net_inet_tcp_map_limit_check, "IU", "Total sendmap entries limit"); VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_map_split_limit), 0, "Total sendmap split entries limit"); #ifdef TCP_HHOOK VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); #endif #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); #define V_ts_offset_secret VNET(ts_offset_secret) static int tcp_default_fb_init(struct tcpcb *tp); static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); static int tcp_default_handoff_ok(struct tcpcb *tp); static struct inpcb *tcp_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); static void tcp_mtudisc(struct inpcb *, int); static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr); static struct tcp_function_block tcp_def_funcblk = { .tfb_tcp_block_name = "freebsd", .tfb_tcp_output = tcp_output, .tfb_tcp_do_segment = tcp_do_segment, .tfb_tcp_ctloutput = tcp_default_ctloutput, .tfb_tcp_handoff_ok = tcp_default_handoff_ok, .tfb_tcp_fb_init = tcp_default_fb_init, .tfb_tcp_fb_fini = tcp_default_fb_fini, }; static int tcp_fb_cnt = 0; struct tcp_funchead t_functions; static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; static struct tcp_function_block * find_tcp_functions_locked(struct tcp_function_set *fs) { struct tcp_function *f; struct tcp_function_block *blk=NULL; TAILQ_FOREACH(f, &t_functions, tf_next) { if (strcmp(f->tf_name, fs->function_set_name) == 0) { blk = f->tf_fb; break; } } return(blk); } static struct tcp_function_block * find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) { struct tcp_function_block *rblk=NULL; struct tcp_function *f; TAILQ_FOREACH(f, &t_functions, tf_next) { if (f->tf_fb == blk) { rblk = blk; if (s) { *s = f; } break; } } return (rblk); } struct tcp_function_block * find_and_ref_tcp_functions(struct tcp_function_set *fs) { struct tcp_function_block *blk; rw_rlock(&tcp_function_lock); blk = find_tcp_functions_locked(fs); if (blk) refcount_acquire(&blk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(blk); } struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *blk) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = find_tcp_fb_locked(blk, NULL); if (rblk) refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(rblk); } static struct tcp_function_block * find_and_ref_tcp_default_fb(void) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = tcp_func_set_ptr; refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return (rblk); } void tcp_switch_back_to_default(struct tcpcb *tp) { struct tcp_function_block *tfb; KASSERT(tp->t_fb != &tcp_def_funcblk, ("%s: called by the built-in default stack", __func__)); /* * Release the old stack. This function will either find a new one * or panic. */ if (tp->t_fb->tfb_tcp_fb_fini != NULL) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); /* * Now, we'll find a new function block to use. * Start by trying the current user-selected * default, unless this stack is the user-selected * default. */ tfb = find_and_ref_tcp_default_fb(); if (tfb == tp->t_fb) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Does the stack accept this connection? */ if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Try to use that stack. */ if (tfb != NULL) { /* Initialize the new stack. If it succeeds, we are done. */ tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init == NULL || (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) return; /* * Initialization failed. Release the reference count on * the stack. */ refcount_release(&tfb->tfb_refcnt); } /* * If that wasn't feasible, use the built-in default * stack which is not allowed to reject anyone. */ tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); if (tfb == NULL) { /* there always should be a default */ panic("Can't refer to tcp_def_funcblk"); } if (tfb->tfb_tcp_handoff_ok != NULL) { if ((*tfb->tfb_tcp_handoff_ok) (tp)) { /* The default stack cannot say no */ panic("Default stack rejects a new session?"); } } tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init != NULL && (*tp->t_fb->tfb_tcp_fb_init)(tp)) { /* The default stack cannot fail */ panic("Default stack initialization failed"); } } static int sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) { int error=ENOENT; struct tcp_function_set fs; struct tcp_function_block *blk; memset(&fs, 0, sizeof(fs)); rw_rlock(&tcp_function_lock); blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); if (blk) { /* Found him */ strcpy(fs.function_set_name, blk->tfb_tcp_block_name); fs.pcbcnt = blk->tfb_refcnt; } rw_runlock(&tcp_function_lock); error = sysctl_handle_string(oidp, fs.function_set_name, sizeof(fs.function_set_name), req); /* Check for error or no change */ if (error != 0 || req->newptr == NULL) return(error); rw_wlock(&tcp_function_lock); blk = find_tcp_functions_locked(&fs); if ((blk == NULL) || (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { error = ENOENT; goto done; } tcp_func_set_ptr = blk; done: rw_wunlock(&tcp_function_lock); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_net_inet_default_tcp_functions, "A", "Set/get the default TCP functions"); static int sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) { int error, cnt, linesz; struct tcp_function *f; char *buffer, *cp; size_t bufsz, outsz; bool alias; cnt = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { cnt++; } rw_runlock(&tcp_function_lock); bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; buffer = malloc(bufsz, M_TEMP, M_WAITOK); error = 0; cp = buffer; linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', "Alias", "PCB count"); cp += linesz; bufsz -= linesz; outsz = linesz; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", f->tf_fb->tfb_tcp_block_name, (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', alias ? f->tf_name : "-", f->tf_fb->tfb_refcnt); if (linesz >= bufsz) { error = EOVERFLOW; break; } cp += linesz; bufsz -= linesz; outsz += linesz; } rw_runlock(&tcp_function_lock); if (error == 0) error = sysctl_handle_string(oidp, buffer, outsz + 1, req); free(buffer, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_net_inet_list_available, "A", "list available TCP Function sets"); /* * Exports one (struct tcp_function_info) for each alias/name. */ static int sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) { int cnt, error; struct tcp_function *f; struct tcp_function_info tfi; /* * We don't allow writes. */ if (req->newptr != NULL) return (EINVAL); /* * Wire the old buffer so we can directly copy the functions to * user space without dropping the lock. */ if (req->oldptr != NULL) { error = sysctl_wire_old_buffer(req, 0); if (error) return (error); } /* * Walk the list and copy out matching entries. If INVARIANTS * is compiled in, also walk the list to verify the length of * the list matches what we have recorded. */ rw_rlock(&tcp_function_lock); cnt = 0; #ifndef INVARIANTS if (req->oldptr == NULL) { cnt = tcp_fb_cnt; goto skip_loop; } #endif TAILQ_FOREACH(f, &t_functions, tf_next) { #ifdef INVARIANTS cnt++; #endif if (req->oldptr != NULL) { bzero(&tfi, sizeof(tfi)); tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; tfi.tfi_id = f->tf_fb->tfb_id; (void)strlcpy(tfi.tfi_alias, f->tf_name, sizeof(tfi.tfi_alias)); (void)strlcpy(tfi.tfi_name, f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); /* * Don't stop on error, as that is the * mechanism we use to accumulate length * information if the buffer was too short. */ } } KASSERT(cnt == tcp_fb_cnt, ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); #ifndef INVARIANTS skip_loop: #endif rw_runlock(&tcp_function_lock); if (req->oldptr == NULL) error = SYSCTL_OUT(req, NULL, (cnt + 1) * sizeof(struct tcp_function_info)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", "List TCP function block name-to-ID mappings"); /* * tfb_tcp_handoff_ok() function for the default stack. * Note that we'll basically try to take all comers. */ static int tcp_default_handoff_ok(struct tcpcb *tp) { return (0); } /* * tfb_tcp_fb_init() function for the default stack. * * This handles making sure we have appropriate timers set if you are * transitioning a socket that has some amount of setup done. * * The init() fuction from the default can *never* return non-zero i.e. * it is required to always succeed since it is the stack of last resort! */ static int tcp_default_fb_init(struct tcpcb *tp) { struct socket *so; INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, ("%s: connection %p in unexpected state %d", __func__, tp, tp->t_state)); /* * Nothing to do for ESTABLISHED or LISTEN states. And, we don't * know what to do for unexpected states (which includes TIME_WAIT). */ if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) return (0); /* * Make sure some kind of transmission timer is set if there is * outstanding data. */ so = tp->t_inpcb->inp_socket; if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || tcp_timer_active(tp, TT_PERSIST))) { /* * If the session has established and it looks like it should * be in the persist state, set the persist timer. Otherwise, * set the retransmit timer. */ if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && (int32_t)(tp->snd_nxt - tp->snd_una) < (int32_t)sbavail(&so->so_snd)) tcp_setpersist(tp); else tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); } /* All non-embryonic sessions get a keepalive timer. */ if (!tcp_timer_active(tp, TT_KEEP)) tcp_timer_activate(tp, TT_KEEP, TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : TP_KEEPINIT(tp)); return (0); } /* * tfb_tcp_fb_fini() function for the default stack. * * This changes state as necessary (or prudent) to prepare for another stack * to assume responsibility for the connection. */ static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) { INP_WLOCK_ASSERT(tp->t_inpcb); return; } /* * Target size of TCP PCB hash tables. Must be a power of two. * * Note that this can be overridden by the kernel environment * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE #define TCBHASHSIZE 0 #endif /* * XXX * Callouts should be moved into struct tcp directly. They are currently * separate because the tcpcb structure is exported to userland for sysctl * parsing purposes, which do not know about callouts. */ struct tcpcb_mem { struct tcpcb tcb; struct tcp_timer tt; struct cc_var ccv; #ifdef TCP_HHOOK struct osd osd; #endif }; VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); #define V_tcpcb_zone VNET(tcpcb_zone) MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); static struct mtx isn_mtx; #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) #define ISN_LOCK() mtx_lock(&isn_mtx) #define ISN_UNLOCK() mtx_unlock(&isn_mtx) /* * TCP initialization. */ static void tcp_zone_change(void *tag) { uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_tcpcb_zone, maxsockets); tcp_tw_zone_change(); } static int tcp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp = mem; INP_LOCK_INIT(inp, "inp", "tcpinp"); return (0); } /* * Take a value and get the next power of 2 that doesn't overflow. * Used to size the tcp_inpcb hash buckets. */ static int maketcp_hashsize(int size) { int hashsize; /* * auto tune. * get the next power of 2 higher than maxsockets. */ hashsize = 1 << fls(size); /* catch overflow, and just go one power of 2 smaller */ if (hashsize < size) { hashsize = 1 << (fls(size) - 1); } return (hashsize); } static volatile int next_tcp_stack_id = 1; /* * Register a TCP function block with the name provided in the names * array. (Note that this function does NOT automatically register * blk->tfb_tcp_block_name as a stack name. Therefore, you should * explicitly include blk->tfb_tcp_block_name in the list of names if * you wish to register the stack with that name.) * * Either all name registrations will succeed or all will fail. If * a name registration fails, the function will update the num_names * argument to point to the array index of the name that encountered * the failure. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, const char *names[], int *num_names) { struct tcp_function *n; struct tcp_function_set fs; int error, i; KASSERT(names != NULL && *num_names > 0, ("%s: Called with 0-length name list", __func__)); KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); KASSERT(rw_initialized(&tcp_function_lock), ("%s: called too early", __func__)); if ((blk->tfb_tcp_output == NULL) || (blk->tfb_tcp_do_segment == NULL) || (blk->tfb_tcp_ctloutput == NULL) || (strlen(blk->tfb_tcp_block_name) == 0)) { /* * These functions are required and you * need a name. */ *num_names = 0; return (EINVAL); } if (blk->tfb_tcp_timer_stop_all || blk->tfb_tcp_timer_activate || blk->tfb_tcp_timer_active || blk->tfb_tcp_timer_stop) { /* * If you define one timer function you * must have them all. */ if ((blk->tfb_tcp_timer_stop_all == NULL) || (blk->tfb_tcp_timer_activate == NULL) || (blk->tfb_tcp_timer_active == NULL) || (blk->tfb_tcp_timer_stop == NULL)) { *num_names = 0; return (EINVAL); } } if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { *num_names = 0; return (EINVAL); } refcount_init(&blk->tfb_refcnt, 0); blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); for (i = 0; i < *num_names; i++) { n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); if (n == NULL) { error = ENOMEM; goto cleanup; } n->tf_fb = blk; (void)strlcpy(fs.function_set_name, names[i], sizeof(fs.function_set_name)); rw_wlock(&tcp_function_lock); if (find_tcp_functions_locked(&fs) != NULL) { /* Duplicate name space not allowed */ rw_wunlock(&tcp_function_lock); free(n, M_TCPFUNCTIONS); error = EALREADY; goto cleanup; } (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); TAILQ_INSERT_TAIL(&t_functions, n, tf_next); tcp_fb_cnt++; rw_wunlock(&tcp_function_lock); } return(0); cleanup: /* * Deregister the names we just added. Because registration failed * for names[i], we don't need to deregister that name. */ *num_names = i; rw_wlock(&tcp_function_lock); while (--i >= 0) { TAILQ_FOREACH(n, &t_functions, tf_next) { if (!strncmp(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX)) { TAILQ_REMOVE(&t_functions, n, tf_next); tcp_fb_cnt--; n->tf_fb = NULL; free(n, M_TCPFUNCTIONS); break; } } } rw_wunlock(&tcp_function_lock); return (error); } /* * Register a TCP function block using the name provided in the name * argument. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, int wait) { const char *name_list[1]; int num_names, rv; num_names = 1; if (name != NULL) name_list[0] = name; else name_list[0] = blk->tfb_tcp_block_name; rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); return (rv); } /* * Register a TCP function block using the name defined in * blk->tfb_tcp_block_name. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions(struct tcp_function_block *blk, int wait) { return (register_tcp_functions_as_name(blk, NULL, wait)); } /* * Deregister all names associated with a function block. This * functionally removes the function block from use within the system. * * When called with a true quiesce argument, mark the function block * as being removed so no more stacks will use it and determine * whether the removal would succeed. * * When called with a false quiesce argument, actually attempt the * removal. * * When called with a force argument, attempt to switch all TCBs to * use the default stack instead of returning EBUSY. * * Returns 0 on success (or if the removal would succeed, or an error * code on failure. */ int deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, bool force) { struct tcp_function *f; if (blk == &tcp_def_funcblk) { /* You can't un-register the default */ return (EPERM); } rw_wlock(&tcp_function_lock); if (blk == tcp_func_set_ptr) { /* You can't free the current default */ rw_wunlock(&tcp_function_lock); return (EBUSY); } /* Mark the block so no more stacks can use it. */ blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; /* * If TCBs are still attached to the stack, attempt to switch them * to the default stack. */ if (force && blk->tfb_refcnt) { struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); rw_wunlock(&tcp_function_lock); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inp); if (inp->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inp); continue; } tp = intotcpcb(inp); if (tp == NULL || tp->t_fb != blk) { INP_WUNLOCK(inp); continue; } tcp_switch_back_to_default(tp); INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); rw_wlock(&tcp_function_lock); } if (blk->tfb_refcnt) { /* TCBs still attached. */ rw_wunlock(&tcp_function_lock); return (EBUSY); } if (quiesce) { /* Skip removal. */ rw_wunlock(&tcp_function_lock); return (0); } /* Remove any function names that map to this function block. */ while (find_tcp_fb_locked(blk, &f) != NULL) { TAILQ_REMOVE(&t_functions, f, tf_next); tcp_fb_cnt--; f->tf_fb = NULL; free(f, M_TCPFUNCTIONS); } rw_wunlock(&tcp_function_lock); return (0); } void tcp_init(void) { const char *tcbhash_tuneable; int hashsize; tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; #ifdef TCP_HHOOK if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); #endif #ifdef STATS if (tcp_stats_init()) printf("%s: WARNING: unable to initialise TCP stats\n", __func__); #endif hashsize = TCBHASHSIZE; TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); if (hashsize == 0) { /* * Auto tune the hash size based on maxsockets. * A perfect hash would have a 1:1 mapping * (hashsize = maxsockets) however it's been * suggested that O(2) average is better. */ hashsize = maketcp_hashsize(maxsockets / 4); /* * Our historical default is 512, * do not autotune lower than this. */ if (hashsize < 512) hashsize = 512; if (bootverbose && IS_DEFAULT_VNET(curvnet)) printf("%s: %s auto tuned to %d\n", __func__, tcbhash_tuneable, hashsize); } /* * We require a hashsize to be a power of two. * Previously if it was not a power of two we would just reset it * back to 512, which could be a nasty surprise if you did not notice * the error message. * Instead what we do is clip it to the closest power of two lower * than the specified hash value. */ if (!powerof2(hashsize)) { int oldhashsize = hashsize; hashsize = maketcp_hashsize(hashsize); /* prevent absurdly low value */ if (hashsize < 16) hashsize = 16; printf("%s: WARNING: TCB hash size not a power of 2, " "clipped from %d to %d.\n", __func__, oldhashsize, hashsize); } in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); /* * These have to be type stable for the benefit of the timers. */ V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_tcpcb_zone, maxsockets); uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); tcp_tw_init(); syncache_init(); tcp_hc_init(); TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); tcp_fastopen_init(); /* Skip initialization of globals for non-default instances. */ if (!IS_DEFAULT_VNET(curvnet)) return; tcp_reass_global_init(); /* XXX virtualize those bellow? */ tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; tcp_rexmit_initial = TCPTV_RTOBASE; if (tcp_rexmit_initial < 1) tcp_rexmit_initial = 1; tcp_rexmit_min = TCPTV_MIN; if (tcp_rexmit_min < 1) tcp_rexmit_min = 1; tcp_persmin = TCPTV_PERSMIN; tcp_persmax = TCPTV_PERSMAX; tcp_rexmit_slop = TCPTV_CPU_VAR; tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; tcp_tcbhashsize = hashsize; /* Setup the tcp function block list */ TAILQ_INIT(&t_functions); rw_init(&tcp_function_lock, "tcp_func_lock"); register_tcp_functions(&tcp_def_funcblk, M_WAITOK); #ifdef TCP_BLACKBOX /* Initialize the TCP logging data. */ tcp_log_init(); #endif arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); if (tcp_soreceive_stream) { #ifdef INET tcp_usrreqs.pru_soreceive = soreceive_stream; #endif #ifdef INET6 tcp6_usrreqs.pru_soreceive = soreceive_stream; #endif /* INET6 */ } #ifdef INET6 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ if (max_protohdr < TCP_MINPROTOHDR) max_protohdr = TCP_MINPROTOHDR; if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) panic("tcp_init"); #undef TCP_MINPROTOHDR ISN_LOCK_INIT(); EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, SHUTDOWN_PRI_DEFAULT); EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, EVENTHANDLER_PRI_ANY); tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); #ifdef TCPPCAP tcp_pcap_init(); #endif } #ifdef VIMAGE static void tcp_destroy(void *unused __unused) { int n; #ifdef TCP_HHOOK int error; #endif /* * All our processes are gone, all our sockets should be cleaned * up, which means, we should be past the tcp_discardcb() calls. * Sleep to let all tcpcb timers really disappear and cleanup. */ for (;;) { INP_LIST_RLOCK(&V_tcbinfo); n = V_tcbinfo.ipi_count; INP_LIST_RUNLOCK(&V_tcbinfo); if (n == 0) break; pause("tcpdes", hz / 10); } tcp_hc_destroy(); syncache_destroy(); tcp_tw_destroy(); in_pcbinfo_destroy(&V_tcbinfo); /* tcp_discardcb() clears the sack_holes up. */ uma_zdestroy(V_sack_hole_zone); uma_zdestroy(V_tcpcb_zone); /* * Cannot free the zone until all tcpcbs are released as we attach * the allocations to them. */ tcp_fastopen_destroy(); #ifdef TCP_HHOOK error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); } error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); } #endif } VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); #endif void tcp_fini(void *xtp) { } /* * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. * tcp_template used to store this data in mbufs, but we now recopy it out * of the tcpcb each time to conserve mbufs. */ void tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) { struct tcphdr *th = (struct tcphdr *)tcp_ptr; INP_WLOCK_ASSERT(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { struct ip *ip; ip = (struct ip *)ip_ptr; ip->ip_v = IPVERSION; ip->ip_hl = 5; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = 0; ip->ip_id = 0; ip->ip_off = 0; ip->ip_ttl = inp->inp_ip_ttl; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } #endif /* INET */ th->th_sport = inp->inp_lport; th->th_dport = inp->inp_fport; th->th_seq = 0; th->th_ack = 0; th->th_x2 = 0; th->th_off = 5; th->th_flags = 0; th->th_win = 0; th->th_urp = 0; th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ } /* * Create template to be used to send tcp packets on a connection. * Allocates an mbuf and fills in a skeletal tcp/ip header. The only * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * tcpip_maketemplate(struct inpcb *inp) { struct tcptemp *t; t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); if (t == NULL) return (NULL); tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); return (t); } /* * Send a single message to the TCP at address specified by * the given TCP/IP header. If m == NULL, then we make a copy * of the tcpiphdr at th and send directly to the addressed host. * This is used to force keep alive messages out using the TCP * template for a connection. If flags are given then we send * a message back to the TCP which originated the segment th, * and discard the mbuf containing it and any other attached mbufs. * * In any case the ack and sequence number of the transmitted * segment are as specified by the parameters. * * NOTE: If m != NULL, then th must point to *inside* the mbuf. */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) { struct tcpopt to; struct inpcb *inp; struct ip *ip; struct mbuf *optm; struct tcphdr *nth; u_char *optp; #ifdef INET6 struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ int optlen, tlen, win; bool incl_opts; KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); NET_EPOCH_ASSERT(); #ifdef INET6 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); ip6 = ipgen; #endif /* INET6 */ ip = ipgen; if (tp != NULL) { inp = tp->t_inpcb; KASSERT(inp != NULL, ("tcp control block w/o inpcb")); INP_WLOCK_ASSERT(inp); } else inp = NULL; incl_opts = false; win = 0; if (tp != NULL) { if (!(flags & TH_RST)) { win = sbspace(&inp->inp_socket->so_rcv); if (win > TCP_MAXWIN << tp->rcv_scale) win = TCP_MAXWIN << tp->rcv_scale; } if ((tp->t_flags & TF_NOOPT) == 0) incl_opts = true; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; m->m_data += max_linkhdr; #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); flags = TH_ACK; } else if (!M_WRITABLE(m)) { struct mbuf *n; /* Can't reuse 'm', allocate a new mbuf. */ n = m_gethdr(M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return; } if (!m_dup_pkthdr(n, m, M_NOWAIT)) { m_freem(m); m_freem(n); return; } n->m_data += max_linkhdr; /* m_len is set later */ #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(n, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(n, struct ip6_hdr *); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); ip = mtod(n, struct ip *); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); xchg(nth->th_dport, nth->th_sport, uint16_t); th = nth; m_freem(m); m = n; } else { /* * reuse the mbuf. * XXX MRT We inherit the FIB, which is lucky. */ m_freem(m->m_next); m->m_next = NULL; m->m_data = (caddr_t)ipgen; /* m_len is set later */ #ifdef INET6 if (isipv6) { xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } if (th != nth) { /* * this is usually a case when an extension header * exists between the IPv6 header and the * TCP header. */ nth->th_sport = th->th_sport; nth->th_dport = th->th_dport; } xchg(nth->th_dport, nth->th_sport, uint16_t); #undef xchg } tlen = 0; #ifdef INET6 if (isipv6) tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tlen = sizeof (struct tcpiphdr); #endif #ifdef INVARIANTS m->m_len = 0; KASSERT(M_TRAILINGSPACE(m) >= tlen, ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", m, tlen, (long)M_TRAILINGSPACE(m))); #endif m->m_len = tlen; to.to_flags = 0; if (incl_opts) { /* Make sure we have room. */ if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { m->m_next = m_get(M_NOWAIT, MT_DATA); if (m->m_next) { optp = mtod(m->m_next, u_char *); optm = m->m_next; } else incl_opts = false; } else { optp = (u_char *) (nth + 1); optm = m; } } if (incl_opts) { /* Timestamps. */ if (tp->t_flags & TF_RCVD_TSTMP) { to.to_tsval = tcp_ts_getticks() + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* Add the options. */ tlen += optlen = tcp_addoptions(&to, optp); /* Update m_len in the correct mbuf. */ optm->m_len += optlen; } else optlen = 0; #ifdef INET6 if (isipv6) { ip6->ip6_flow = 0; ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(tlen - sizeof(*ip6)); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(tlen); ip->ip_ttl = V_ip_defttl; if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); } #endif m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef MAC if (inp != NULL) { /* * Packet is associated with a socket, so allow the * label of the response to reflect the socket label. */ INP_WLOCK_ASSERT(inp); mac_inpcb_create_mbuf(inp, m); } else { /* * Packet is not associated with a socket, so possibly * update the label in place. */ mac_netinet_tcp_reply(m); } #endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; nth->th_flags = flags; if (tp != NULL) nth->th_win = htons((u_short) (win >> tp->rcv_scale)); else nth->th_win = htons((u_short)win); nth->th_urp = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { m_freem(m); return; } } #endif m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #ifdef INET6 if (isipv6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; nth->th_sum = in6_cksum_pseudo(ip6, tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : NULL, NULL); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); } #endif /* INET */ #ifdef TCPDEBUG if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); #endif TCP_PROBE3(debug__output, tp, th, m); if (flags & TH_RST) TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); #ifdef INET6 if (isipv6) { TCP_PROBE5(send, NULL, tp, ip6, tp, nth); (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { TCP_PROBE5(send, NULL, tp, ip, tp, nth); (void)ip_output(m, NULL, NULL, 0, NULL, inp); } #endif } /* * Create a new TCP control block, making an * empty reassembly queue and hooking it to the argument * protocol control block. The `inp' parameter must have * come from the zone allocator set up in tcp_init(). */ struct tcpcb * tcp_newtcpcb(struct inpcb *inp) { struct tcpcb_mem *tm; struct tcpcb *tp; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) return (NULL); tp = &tm->tcb; /* Initialise cc_var struct for this tcpcb. */ tp->ccv = &tm->ccv; tp->ccv->type = IPPROTO_TCP; tp->ccv->ccvc.tcp = tp; rw_rlock(&tcp_function_lock); tp->t_fb = tcp_func_set_ptr; refcount_acquire(&tp->t_fb->tfb_refcnt); rw_runlock(&tcp_function_lock); /* * Use the current system default CC algorithm. */ CC_LIST_RLOCK(); KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); CC_ALGO(tp) = CC_DEFAULT(); CC_LIST_RUNLOCK(); if (CC_ALGO(tp)->cb_init != NULL) if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #ifdef TCP_HHOOK tp->osd = &tm->osd; if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #endif #ifdef VIMAGE tp->t_vnet = inp->inp_vnet; #endif tp->t_timers = &tm->tt; TAILQ_INIT(&tp->t_segq); tp->t_maxseg = #ifdef INET6 isipv6 ? V_tcp_v6mssdflt : #endif /* INET6 */ V_tcp_mssdflt; /* Set up our timeouts. */ callout_init(&tp->t_timers->tt_rexmt, 1); callout_init(&tp->t_timers->tt_persist, 1); callout_init(&tp->t_timers->tt_keep, 1); callout_init(&tp->t_timers->tt_2msl, 1); callout_init(&tp->t_timers->tt_delack, 1); if (V_tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); if (V_tcp_do_sack) tp->t_flags |= TF_SACK_PERMIT; TAILQ_INIT(&tp->snd_holes); /* * The tcpcb will hold a reference on its inpcb until tcp_discardcb() * is called. */ in_pcbref(inp); /* Reference for tcpcb */ tp->t_inpcb = inp; /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_rexmit_min; tp->t_rxtcur = tcp_rexmit_initial; tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = ticks; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ inp->inp_ip_ttl = V_ip_defttl; inp->inp_ppcb = tp; #ifdef TCPPCAP /* * Init the TCP PCAP queues. */ tcp_pcap_tcpcb_init(tp); #endif #ifdef TCP_BLACKBOX /* Initialize the per-TCPCB log data. */ tcp_log_tcpcbinit(tp); #endif if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } #ifdef STATS if (V_tcp_perconn_stats_enable == 1) tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); #endif return (tp); /* XXX */ } /* * Switch the congestion control algorithm back to NewReno for any active * control blocks using an algorithm which is about to go away. * This ensures the CC framework can allow the unload to proceed without leaving * any dangling pointers which would trigger a panic. * Returning non-zero would inform the CC framework that something went wrong * and it would be unsafe to allow the unload to proceed. However, there is no * way for this to occur with this implementation so we always return zero. */ int tcp_ccalgounload(struct cc_algo *unload_algo) { struct cc_algo *tmpalgo; struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); /* * Check all active control blocks across all network stacks and change * any that are using "unload_algo" back to NewReno. If "unload_algo" * requires cleanup code to be run, call it. */ VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); /* * New connections already part way through being initialised * with the CC algo we're removing will not race with this code * because the INP_INFO_WLOCK is held during initialisation. We * therefore don't enter the loop below until the connection * list has stabilised. */ CK_LIST_FOREACH(inp, &V_tcb, inp_list) { INP_WLOCK(inp); /* Important to skip tcptw structs. */ if (!(inp->inp_flags & INP_TIMEWAIT) && (tp = intotcpcb(inp)) != NULL) { /* * By holding INP_WLOCK here, we are assured * that the connection is not currently * executing inside the CC module's functions * i.e. it is safe to make the switch back to * NewReno. */ if (CC_ALGO(tp) == unload_algo) { tmpalgo = CC_ALGO(tp); if (tmpalgo->cb_destroy != NULL) tmpalgo->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; /* * NewReno may allocate memory on * demand for certain stateful * configuration as needed, but is * coded to never fail on memory * allocation failure so it is a safe * fallback. */ CC_ALGO(tp) = &newreno_cc_algo; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } /* * Drop a TCP connection, reporting * the specified error. If connection is synchronized, * then send a RST to peer. */ struct tcpcb * tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; NET_EPOCH_ASSERT(); INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); if (TCPS_HAVERCVDSYN(tp->t_state)) { tcp_state_change(tp, TCPS_CLOSED); (void) tp->t_fb->tfb_tcp_output(tp); TCPSTAT_INC(tcps_drops); } else TCPSTAT_INC(tcps_conndrops); if (errno == ETIMEDOUT && tp->t_softerror) errno = tp->t_softerror; so->so_error = errno; return (tcp_close(tp)); } void tcp_discardcb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ int released __unused; INP_WLOCK_ASSERT(inp); /* * Make sure that all of our timers are stopped before we delete the * PCB. * * If stopping a timer fails, we schedule a discard function in same * callout, and the last discard function called will take care of * deleting the tcpcb. */ tp->t_timers->tt_draincnt = 0; tcp_timer_stop(tp, TT_REXMT); tcp_timer_stop(tp, TT_PERSIST); tcp_timer_stop(tp, TT_KEEP); tcp_timer_stop(tp, TT_2MSL); tcp_timer_stop(tp, TT_DELACK); if (tp->t_fb->tfb_tcp_timer_stop_all) { /* * Call the stop-all function of the methods, * this function should call the tcp_timer_stop() * method with each of the function specific timeouts. * That stop will be called via the tfb_tcp_timer_stop() * which should use the async drain function of the * callout system (see tcp_var.h). */ tp->t_fb->tfb_tcp_timer_stop_all(tp); } /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. * 'Enough' is arbitrarily defined as 4 rtt samples. * 4 samples is enough for the srtt filter to converge * to within enough % of the correct value; fewer samples * and we could save a bogus rtt. The danger is not high * as tcp quickly recovers from everything. * XXX: Works very well but needs some more statistics! */ if (tp->t_rttupdated >= 4) { struct hc_metrics_lite metrics; uint32_t ssthresh; bzero(&metrics, sizeof(metrics)); /* * Update the ssthresh always when the conditions below * are satisfied. This gives us better new start value * for the congestion avoidance for new connections. * ssthresh is only set if packet loss occurred on a session. * * XXXRW: 'so' may be NULL here, and/or socket buffer may be * being torn down. Ideally this code would not use 'so'. */ ssthresh = tp->snd_ssthresh; if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { /* * convert the limit from user data bytes to * packets then to packet data bytes. */ ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; if (ssthresh < 2) ssthresh = 2; ssthresh *= (tp->t_maxseg + #ifdef INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : #endif sizeof (struct tcpiphdr) #ifdef INET6 ) #endif ); } else ssthresh = 0; metrics.rmx_ssthresh = ssthresh; metrics.rmx_rtt = tp->t_srtt; metrics.rmx_rttvar = tp->t_rttvar; metrics.rmx_cwnd = tp->snd_cwnd; metrics.rmx_sendpipe = 0; metrics.rmx_recvpipe = 0; tcp_hc_update(&inp->inp_inc, &metrics); } /* free the reassembly queue, if any */ tcp_reass_flush(tp); #ifdef TCP_OFFLOAD /* Disconnect offload device, if any. */ if (tp->t_flags & TF_TOE) tcp_offload_detach(tp); #endif tcp_free_sackholes(tp); #ifdef TCPPCAP /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tp->t_inpkts)); tcp_pcap_drain(&(tp->t_outpkts)); #endif /* Allow the CC algorithm to clean up after itself. */ if (CC_ALGO(tp)->cb_destroy != NULL) CC_ALGO(tp)->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; #ifdef TCP_HHOOK khelp_destroy_osd(tp->osd); #endif #ifdef STATS stats_blob_destroy(tp->t_stats); #endif CC_ALGO(tp) = NULL; inp->inp_ppcb = NULL; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); released = in_pcbrele_wlocked(inp); KASSERT(!released, ("%s: inp %p should not have been released " "here", __func__, inp)); } } void tcp_timer_discard(void *ptp) { struct inpcb *inp; struct tcpcb *tp; struct epoch_tracker et; tp = (struct tcpcb *)ptp; CURVNET_SET(tp->t_vnet); NET_EPOCH_ENTER(et); inp = tp->t_inpcb; KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp)); INP_WLOCK(inp); KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, ("%s: tcpcb has to be stopped here", __func__)); tp->t_timers->tt_draincnt--; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on this tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); if (in_pcbrele_wlocked(inp)) { NET_EPOCH_EXIT(et); CURVNET_RESTORE(); return; } } INP_WUNLOCK(inp); NET_EPOCH_EXIT(et); CURVNET_RESTORE(); } /* * Attempt to close a TCP control block, marking it as dropped, and freeing * the socket if we hold the only reference. */ struct tcpcb * tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_state == TCPS_LISTEN) tcp_offload_listen_stop(tp); #endif /* * This releases the TFO pending counter resource for TFO listen * sockets as well as passively-created TFO sockets that transition * from SYN_RECEIVED to CLOSED. */ if (tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } in_pcbdrop(inp); TCPSTAT_INC(tcps_closed); if (tp->t_state != TCPS_CLOSED) tcp_state_change(tp, TCPS_CLOSED); KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); so = inp->inp_socket; soisdisconnected(so); if (inp->inp_flags & INP_SOCKREF) { KASSERT(so->so_state & SS_PROTOREF, ("tcp_close: !SS_PROTOREF")); inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); SOCK_LOCK(so); so->so_state &= ~SS_PROTOREF; sofree(so); return (NULL); } return (tp); } void tcp_drain(void) { VNET_ITERATOR_DECL(vnet_iter); if (!do_tcpdrain) return; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct inpcb *inpb; struct tcpcb *tcpb; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... * XXX: The "Net/3" implementation doesn't imply that the TCP * reassembly queue should be flushed, but in a situation * where we're really low on mbufs, this is potentially * useful. */ INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inpb); if (inpb->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inpb); continue; } if ((tcpb = intotcpcb(inpb)) != NULL) { tcp_reass_flush(tcpb); tcp_clean_sackreport(tcpb); #ifdef TCP_BLACKBOX tcp_log_drain(tcpb); #endif #ifdef TCPPCAP if (tcp_pcap_aggressive_free) { /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tcpb->t_inpkts)); tcp_pcap_drain(&(tcpb->t_outpkts)); } #endif } INP_WUNLOCK(inpb); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Notify a tcp user of an asynchronous error; * store error as soft error, but wake up user * (for now, won't do anything until can select for soft error). * * Do not wake up user since there currently is no mechanism for * reporting soft errors (yet - a kqueue filter may be added). */ static struct inpcb * tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, * and receives a second error, give up now. This is better * than waiting a long time to establish a connection that * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && (error == EHOSTUNREACH || error == ENETUNREACH || error == EHOSTDOWN)) { - if (inp->inp_route.ro_rt) { - RTFREE(inp->inp_route.ro_rt); - inp->inp_route.ro_rt = (struct rtentry *)NULL; + if (inp->inp_route.ro_nh) { + NH_FREE(inp->inp_route.ro_nh); + inp->inp_route.ro_nh = (struct nhop_object *)NULL; } return (inp); } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) { tp = tcp_drop(tp, error); if (tp != NULL) return (inp); else return (NULL); } else { tp->t_softerror = error; return (inp); } #if 0 wakeup( &so->so_timeo); sorwakeup(so); sowwakeup(so); #endif } static int tcp_pcblist(SYSCTL_HANDLER_ARGS) { struct epoch_tracker et; struct inpcb *inp; struct xinpgen xig; int error; if (req->newptr != NULL) return (EPERM); if (req->oldptr == NULL) { int n; n = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); return (0); } if ((error = sysctl_wire_old_buffer(req, 0)) != 0) return (error); bzero(&xig, sizeof(xig)); xig.xig_len = sizeof xig; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); error = syncache_pcblist(req); if (error) return (error); NET_EPOCH_ENTER(et); for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); inp != NULL; inp = CK_LIST_NEXT(inp, inp_list)) { INP_RLOCK(inp); if (inp->inp_gencnt <= xig.xig_gen) { int crerr; /* * XXX: This use of cr_cansee(), introduced with * TCP state changes, is not quite right, but for * now, better than nothing. */ if (inp->inp_flags & INP_TIMEWAIT) { if (intotw(inp) != NULL) crerr = cr_cansee(req->td->td_ucred, intotw(inp)->tw_cred); else crerr = EINVAL; /* Skip this inp. */ } else crerr = cr_canseeinpcb(req->td->td_ucred, inp); if (crerr == 0) { struct xtcpcb xt; tcp_inptoxtp(inp, &xt); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xt, sizeof xt); if (error) break; else continue; } } INP_RUNLOCK(inp); } NET_EPOCH_EXIT(et); if (!error) { /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); error = SYSCTL_OUT(req, &xig, sizeof xig); } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); #ifdef INET static int tcp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct epoch_tracker et; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); NET_EPOCH_ENTER(et); inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 0, 0, tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); #endif /* INET */ #ifdef INET6 static int tcp6_getcred(SYSCTL_HANDLER_ARGS) { struct epoch_tracker et; struct xucred xuc; struct sockaddr_in6 addrs[2]; struct inpcb *inp; int error; #ifdef INET int mapped = 0; #endif error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { return (error); } if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) mapped = 1; else #endif return (EINVAL); } NET_EPOCH_ENTER(et); #ifdef INET if (mapped == 1) inp = in_pcblookup(&V_tcbinfo, *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], addrs[1].sin6_port, *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); else #endif inp = in6_pcblookup(&V_tcbinfo, &addrs[1].sin6_addr, addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 0, 0, tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); #endif /* INET6 */ #ifdef INET void tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { struct ip *ip = vip; struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct icmp *icp; struct in_conninfo inc; tcp_seq icmp_tcp_seq; int mtu; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip == NULL) { in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); return; } icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } icmp_tcp_seq = th->th_seq; if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohs(icp->icmp_nextmtu); /* * If no alternative MTU was * proposed, try the next smaller * one. */ if (!mtu) mtu = ip_next_mtu( ntohs(ip->ip_len), 1); if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) mtu = V_tcp_minmss + sizeof(struct tcpiphdr); /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof(struct tcpiphdr)) { bzero(&inc, sizeof(inc)); inc.inc_faddr = faddr; inc.inc_fibnum = inp->inp_inc.inc_fibnum; tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); } } else inp = (*notify)(inp, inetctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fport = th->th_dport; inc.inc_lport = th->th_sport; inc.inc_faddr = faddr; inc.inc_laddr = ip->ip_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); } #endif /* INET */ #ifdef INET6 void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { struct in6_addr *dst; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; struct inpcb *inp; struct tcpcb *tp; struct icmp6_hdr *icmp6; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; struct in_conninfo inc; struct tcp_ports { uint16_t th_sport; uint16_t th_dport; } t_ports; tcp_seq icmp_tcp_seq; unsigned int mtu; unsigned int off; if (sa->sa_family != AF_INET6 || sa->sa_len != sizeof(struct sockaddr_in6)) return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; icmp6 = ip6cp->ip6c_icmp6; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; dst = ip6cp->ip6c_finaldst; } else { m = NULL; ip6 = NULL; off = 0; /* fool gcc */ sa6_src = &sa6_any; dst = NULL; } if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip6 = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) return; if (ip6 == NULL) { in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); return; } /* Check if we can safely get the ports from the tcp hdr */ if (m == NULL || (m->m_pkthdr.len < (int32_t) (off + sizeof(struct tcp_ports)))) { return; } bzero(&t_ports, sizeof(struct tcp_ports)); m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } off += sizeof(struct tcp_ports); if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { goto out; } m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohl(icmp6->icmp6_mtu); /* * If no alternative MTU was * proposed, or the proposed * MTU was too small, set to * the min. */ if (mtu < IPV6_MMTU) mtu = IPV6_MMTU - 8; bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL)) goto out; /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof (struct tcphdr) + sizeof (struct ip6_hdr)) { tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); ICMP6STAT_INC(icp6s_pmtuchg); } } else inp = (*notify)(inp, inet6ctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc_fport = t_ports.th_dport; inc.inc_lport = t_ports.th_sport; inc.inc6_faddr = *dst; inc.inc6_laddr = ip6->ip6_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); } #endif /* INET6 */ static uint32_t tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) { SIPHASH_CTX ctx; uint32_t hash[2]; KASSERT(len >= SIPHASH_KEY_LENGTH, ("%s: keylen %u too short ", __func__, len)); SipHash24_Init(&ctx); SipHash_SetKey(&ctx, (uint8_t *)key); SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); break; #endif } SipHash_Final((uint8_t *)hash, &ctx); return (hash[0] ^ hash[1]); } uint32_t tcp_new_ts_offset(struct in_conninfo *inc) { struct in_conninfo inc_store, *local_inc; if (!V_tcp_ts_offset_per_conn) { memcpy(&inc_store, inc, sizeof(struct in_conninfo)); inc_store.inc_lport = 0; inc_store.inc_fport = 0; local_inc = &inc_store; } else { local_inc = inc; } return (tcp_keyed_hash(local_inc, V_ts_offset_secret, sizeof(V_ts_offset_secret))); } /* * Following is where TCP initial sequence number generation occurs. * * There are two places where we must use initial sequence numbers: * 1. In SYN-ACK packets. * 2. In SYN packets. * * All ISNs for SYN-ACK packets are generated by the syncache. See * tcp_syncache.c for details. * * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling * depends on this property. In addition, these ISNs should be * unguessable so as to prevent connection hijacking. To satisfy * the requirements of this situation, the algorithm outlined in * RFC 1948 is used, with only small modifications. * * Implementation details: * * Time is based off the system timer, and is corrected so that it * increases by one megabyte per second. This allows for proper * recycling on high speed LANs while still leaving over an hour * before rollover. * * As reading the *exact* system time is too expensive to be done * whenever setting up a TCP connection, we increment the time * offset in two ways. First, a small random positive increment * is added to isn_offset for each connection that is set up. * Second, the function tcp_isn_tick fires once per clock tick * and increments isn_offset as necessary so that sequence numbers * are incremented at approximately ISN_BYTES_PER_SECOND. The * random positive increments serve only to ensure that the same * exact sequence number is never sent out twice (as could otherwise * happen when a port is recycled in less than the system tick * interval.) * * net.inet.tcp.isn_reseed_interval controls the number of seconds * between seeding of isn_secret. This is normally set to zero, * as reseeding should not be necessary. * * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, * isn_offset_old, and isn_ctx is performed using the ISN lock. In * general, this means holding an exclusive (write) lock. */ #define ISN_BYTES_PER_SECOND 1048576 #define ISN_STATIC_INCREMENT 4096 #define ISN_RANDOM_INCREMENT (4096 - 1) #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); VNET_DEFINE_STATIC(int, isn_last); VNET_DEFINE_STATIC(int, isn_last_reseed); VNET_DEFINE_STATIC(u_int32_t, isn_offset); VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); #define V_isn_secret VNET(isn_secret) #define V_isn_last VNET(isn_last) #define V_isn_last_reseed VNET(isn_last_reseed) #define V_isn_offset VNET(isn_offset) #define V_isn_offset_old VNET(isn_offset_old) tcp_seq tcp_new_isn(struct in_conninfo *inc) { tcp_seq new_isn; u_int32_t projected_offset; ISN_LOCK(); /* Seed if this is the first use, reseed if requested. */ if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) < (u_int)ticks))) { arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); V_isn_last_reseed = ticks; } /* Compute the hash and return the ISN. */ new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, sizeof(V_isn_secret)); V_isn_offset += ISN_STATIC_INCREMENT + (arc4random() & ISN_RANDOM_INCREMENT); if (ticks != V_isn_last) { projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); if (SEQ_GT(projected_offset, V_isn_offset)) V_isn_offset = projected_offset; V_isn_offset_old = V_isn_offset; V_isn_last = ticks; } new_isn += V_isn_offset; ISN_UNLOCK(); return (new_isn); } /* * When a specific ICMP unreachable message is received and the * connection state is SYN-SENT, drop the connection. This behavior * is controlled by the icmp_may_rst sysctl. */ struct inpcb * tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp; NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); if (tp->t_state != TCPS_SYN_SENT) return (inp); if (IS_FASTOPEN(tp->t_flags)) tcp_fastopen_disable_path(tp); tp = tcp_drop(tp, errno); if (tp != NULL) return (inp); else return (NULL); } /* * When `need fragmentation' ICMP is received, update our idea of the MSS * based on the new value. Also nudge TCP to send something, since we * know the packet we just sent was dropped. * This duplicates some code in the tcp_mss() function in tcp_input.c. */ static struct inpcb * tcp_mtudisc_notify(struct inpcb *inp, int error) { tcp_mtudisc(inp, -1); return (inp); } static void tcp_mtudisc(struct inpcb *inp, int mtuoffer) { struct tcpcb *tp; struct socket *so; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return; tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); /* If the mss is larger than the socket buffer, decrease the mss. */ if (so->so_snd.sb_hiwat < tp->t_maxseg) tp->t_maxseg = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); TCPSTAT_INC(tcps_mturesent); tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; tcp_free_sackholes(tp); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_SACK_PERMIT) EXIT_FASTRECOVERY(tp->t_flags); tp->t_fb->tfb_tcp_output(tp); } #ifdef INET /* * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated, then return 0. This routine * is called by TCP routines that access the rmx structure and by * tcp_mss_update to get the peer/interface MTU. */ uint32_t tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop4_extended nh4; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); if (inc->inc_faddr.s_addr != INADDR_ANY) { if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, NHR_REF, 0, &nh4) != 0) return (0); ifp = nh4.nh_ifp; maxmtu = nh4.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO4 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib4_free_nh_ext(inc->inc_fibnum, &nh4); } return (maxmtu); } #endif /* INET */ #ifdef INET6 uint32_t tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop6_extended nh6; struct in6_addr dst6; uint32_t scopeid; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); if (inc->inc_flags & INC_IPV6MINMTU) return (IPV6_MMTU); if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 0, &nh6) != 0) return (0); ifp = nh6.nh_ifp; maxmtu = nh6.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO6 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib6_free_nh_ext(inc->inc_fibnum, &nh6); } return (maxmtu); } #endif /* INET6 */ /* * Calculate effective SMSS per RFC5681 definition for a given TCP * connection at its current state, taking into account SACK and etc. */ u_int tcp_maxseg(const struct tcpcb *tp) { u_int optlen; if (tp->t_flags & TF_NOOPT) return (tp->t_maxseg); /* * Here we have a simplified code from tcp_addoptions(), * without a proper loop, and having most of paddings hardcoded. * We might make mistakes with padding here in some edge cases, * but this is harmless, since result of tcp_maxseg() is used * only in cwnd and ssthresh estimations. */ #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) if (TCPS_HAVEESTABLISHED(tp->t_state)) { if (tp->t_flags & TF_RCVD_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { optlen += TCPOLEN_SACKHDR; optlen += tp->rcv_numsacks * TCPOLEN_SACK; optlen = PAD(optlen); } } else { if (tp->t_flags & TF_REQ_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = PAD(TCPOLEN_MAXSEG); if (tp->t_flags & TF_REQ_SCALE) optlen += PAD(TCPOLEN_WINDOW); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if (tp->t_flags & TF_SACK_PERMIT) optlen += PAD(TCPOLEN_SACK_PERMITTED); } #undef PAD optlen = min(optlen, TCP_MAXOLEN); return (tp->t_maxseg - optlen); } static int sysctl_drop(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct tcpcb *tp; struct tcptw *tw; struct sockaddr_in *fin, *lin; struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } if (inp != NULL) { if (inp->inp_flags & INP_TIMEWAIT) { /* * XXXRW: There currently exists a state where an * inpcb is present, but its timewait state has been * discarded. For now, don't allow dropping of this * type of inpcb. */ tw = intotw(inp); if (tw != NULL) tcp_twclose(tw, 0); else INP_WUNLOCK(inp); } else if (!(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { tp = intotcpcb(inp); tp = tcp_drop(tp, ECONNABORTED); if (tp != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } else error = ESRCH; NET_EPOCH_EXIT(et); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", "Drop TCP connection"); #ifdef KERN_TLS static int sysctl_switch_tls(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct sockaddr_in *fin, *lin; struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } NET_EPOCH_EXIT(et); if (inp != NULL) { if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || inp->inp_socket == NULL) { error = ECONNRESET; INP_WUNLOCK(inp); } else { struct socket *so; so = inp->inp_socket; soref(so); error = ktls_set_tx_mode(so, arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); INP_WUNLOCK(inp); SOCK_LOCK(so); sorele(so); } } else error = ESRCH; return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS"); SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS"); #endif /* * Generate a standardized TCP log line for use throughout the * tcp subsystem. Memory allocation is done with M_NOWAIT to * allow use in the interrupt context. * * NB: The caller MUST free(s, M_TCPLOG) the returned string. * NB: The function may return NULL if memory allocation failed. * * Due to header inclusion and ordering limitations the struct ip * and ip6_hdr pointers have to be passed as void pointers. */ char * tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (V_tcp_log_in_vain == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } char * tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_debug == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { char *s, *sp; size_t size; struct ip *ip; #ifdef INET6 const struct ip6_hdr *ip6; ip6 = (const struct ip6_hdr *)ip6hdr; #endif /* INET6 */ ip = (struct ip *)ip4hdr; /* * The log line looks like this: * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2" */ size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + sizeof(PRINT_TH_FLAGS) + 1 + #ifdef INET6 2 * INET6_ADDRSTRLEN; #else 2 * INET_ADDRSTRLEN; #endif /* INET6 */ s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); if (s == NULL) return (NULL); strcat(s, "TCP: ["); sp = s + strlen(s); if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { inet_ntoa_r(inc->inc_faddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); inet_ntoa_r(inc->inc_laddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); #ifdef INET6 } else if (inc) { ip6_sprintf(sp, &inc->inc6_faddr); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); ip6_sprintf(sp, &inc->inc6_laddr); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); } else if (ip6 && th) { ip6_sprintf(sp, &ip6->ip6_src); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); ip6_sprintf(sp, &ip6->ip6_dst); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET6 */ #ifdef INET } else if (ip && th) { inet_ntoa_r(ip->ip_src, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); inet_ntoa_r(ip->ip_dst, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET */ } else { free(s, M_TCPLOG); return (NULL); } sp = s + strlen(s); if (th) sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); if (*(s + size - 1) != '\0') panic("%s: string too long", __func__); return (s); } /* * A subroutine which makes it easy to track TCP state changes with DTrace. * This function shouldn't be called for t_state initializations that don't * correspond to actual TCP state transitions. */ void tcp_state_change(struct tcpcb *tp, int newstate) { #if defined(KDTRACE_HOOKS) int pstate = tp->t_state; #endif TCPSTATES_DEC(tp->t_state); TCPSTATES_INC(newstate); tp->t_state = newstate; TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); } /* * Create an external-format (``xtcpcb'') structure using the information in * the kernel-format tcpcb structure pointed to by tp. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) { struct tcpcb *tp = intotcpcb(inp); sbintime_t now; bzero(xt, sizeof(*xt)); if (inp->inp_flags & INP_TIMEWAIT) { xt->t_state = TCPS_TIME_WAIT; } else { xt->t_state = tp->t_state; xt->t_logstate = tp->t_logstate; xt->t_flags = tp->t_flags; xt->t_sndzerowin = tp->t_sndzerowin; xt->t_sndrexmitpack = tp->t_sndrexmitpack; xt->t_rcvoopack = tp->t_rcvoopack; now = getsbinuptime(); #define COPYTIMER(ttt) do { \ if (callout_active(&tp->t_timers->ttt)) \ xt->ttt = (tp->t_timers->ttt.c_time - now) / \ SBT_1MS; \ else \ xt->ttt = 0; \ } while (0) COPYTIMER(tt_delack); COPYTIMER(tt_rexmt); COPYTIMER(tt_persist); COPYTIMER(tt_keep); COPYTIMER(tt_2msl); #undef COPYTIMER xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, TCP_FUNCTION_NAME_LEN_MAX); #ifdef TCP_BLACKBOX (void)tcp_log_get_id(tp, xt->xt_logid); #endif } xt->xt_len = sizeof(struct xtcpcb); in_pcbtoxinpcb(inp, &xt->xt_inp); if (inp->inp_socket == NULL) xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; } Index: head/sys/netinet/udp_usrreq.c =================================================================== --- head/sys/netinet/udp_usrreq.c (revision 360291) +++ head/sys/netinet/udp_usrreq.c (revision 360292) @@ -1,1733 +1,1734 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. * Copyright (c) 2008 Robert N. M. Watson * Copyright (c) 2010-2011 Juniper Networks, Inc. * Copyright (c) 2014 Kevin Lo * All rights reserved. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include /* * UDP and UDP-Lite protocols implementation. * Per RFC 768, August, 1980. * Per RFC 3828, July, 2004. */ /* * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums * removes the only data integrity mechanism for packets and malformed * packets that would otherwise be discarded due to bad checksums, and may * cause problems (especially for NFS data blocks). */ VNET_DEFINE(int, udp_cksum) = 1; SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(udp_cksum), 0, "compute udp checksum"); VNET_DEFINE(int, udp_log_in_vain) = 0; SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets"); VNET_DEFINE(int, udp_blackhole) = 0; SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(udp_blackhole), 0, "Do not send port unreachables for refused connects"); u_long udp_sendspace = 9216; /* really max datagram size */ SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); u_long udp_recvspace = 40 * (1024 + #ifdef INET6 sizeof(struct sockaddr_in6) #else sizeof(struct sockaddr_in) #endif ); /* 40 1K datagrams */ SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ VNET_DEFINE(struct inpcbinfo, udbinfo); VNET_DEFINE(struct inpcbhead, ulitecb); VNET_DEFINE(struct inpcbinfo, ulitecbinfo); VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone); #define V_udpcb_zone VNET(udpcb_zone) #ifndef UDBHASHSIZE #define UDBHASHSIZE 128 #endif VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ VNET_PCPUSTAT_SYSINIT(udpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(udpstat); #endif /* VIMAGE */ #ifdef INET static void udp_detach(struct socket *so); static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, struct mbuf *, struct thread *); #endif static void udp_zone_change(void *tag) { uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_udpcb_zone, maxsockets); } static int udp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp; inp = mem; INP_LOCK_INIT(inp, "inp", "udpinp"); return (0); } static int udplite_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp; inp = mem; INP_LOCK_INIT(inp, "inp", "udpliteinp"); return (0); } void udp_init(void) { /* * For now default to 2-tuple UDP hashing - until the fragment * reassembly code can also update the flowid. * * Once we can calculate the flowid that way and re-establish * a 4-tuple, flip this to 4-tuple. */ in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE); V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_udpcb_zone, maxsockets); uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, EVENTHANDLER_PRI_ANY); } void udplite_init(void) { in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, IPI_HASHFIELDS_2TUPLE); } /* * Kernel module interface for updating udpstat. The argument is an index * into udpstat treated as an array of u_long. While this encodes the * general layout of udpstat into the caller, it doesn't encode its location, * so that future changes to add, for example, per-CPU stats support won't * cause binary compatibility problems for kernel modules. */ void kmod_udpstat_inc(int statnum) { counter_u64_add(VNET(udpstat)[statnum], 1); } int udp_newudpcb(struct inpcb *inp) { struct udpcb *up; up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); if (up == NULL) return (ENOBUFS); inp->inp_ppcb = up; return (0); } void udp_discardcb(struct udpcb *up) { uma_zfree(V_udpcb_zone, up); } #ifdef VIMAGE static void udp_destroy(void *unused __unused) { in_pcbinfo_destroy(&V_udbinfo); uma_zdestroy(V_udpcb_zone); } VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); static void udplite_destroy(void *unused __unused) { in_pcbinfo_destroy(&V_ulitecbinfo); } VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, NULL); #endif #ifdef INET /* * Subroutine of udp_input(), which appends the provided mbuf chain to the * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that * contains the source address. If the socket ends up being an IPv6 socket, * udp_append() will convert to a sockaddr_in6 before passing the address * into the socket code. * * In the normal case udp_append() will return 0, indicating that you * must unlock the inp. However if a tunneling protocol is in place we increment * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we * then decrement the reference count. If the inp_rele returns 1, indicating the * inp is gone, we return that to the caller to tell them *not* to unlock * the inp. In the case of multi-cast this will cause the distribution * to stop (though most tunneling protocols known currently do *not* use * multicast). */ static int udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, struct sockaddr_in *udp_in) { struct sockaddr *append_sa; struct socket *so; struct mbuf *tmpopts, *opts = NULL; #ifdef INET6 struct sockaddr_in6 udp_in6; #endif struct udpcb *up; INP_LOCK_ASSERT(inp); /* * Engage the tunneling protocol. */ up = intoudpcb(inp); if (up->u_tun_func != NULL) { in_pcbref(inp); INP_RUNLOCK(inp); (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], up->u_tun_ctx); INP_RLOCK(inp); return (in_pcbrele_rlocked(inp)); } off += sizeof(struct udphdr); #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* Check AH/ESP integrity. */ if (IPSEC_ENABLED(ipv4) && IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { m_freem(n); return (0); } if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ if (IPSEC_ENABLED(ipv4) && UDPENCAP_INPUT(n, off, AF_INET) != 0) return (0); /* Consumed. */ } #endif /* IPSEC */ #ifdef MAC if (mac_inpcb_check_deliver(inp, n) != 0) { m_freem(n); return (0); } #endif /* MAC */ if (inp->inp_flags & INP_CONTROLOPTS || inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { #ifdef INET6 if (inp->inp_vflag & INP_IPV6) (void)ip6_savecontrol_v4(inp, n, &opts, NULL); else #endif /* INET6 */ ip_savecontrol(inp, &opts, ip, n); } if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); if (tmpopts) { if (opts) { tmpopts->m_next = opts; opts = tmpopts; } else opts = tmpopts; } } #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { bzero(&udp_in6, sizeof(udp_in6)); udp_in6.sin6_len = sizeof(udp_in6); udp_in6.sin6_family = AF_INET6; in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); append_sa = (struct sockaddr *)&udp_in6; } else #endif /* INET6 */ append_sa = (struct sockaddr *)&udp_in[0]; m_adj(n, off); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_rcv); if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { SOCKBUF_UNLOCK(&so->so_rcv); m_freem(n); if (opts) m_freem(opts); UDPSTAT_INC(udps_fullsock); } else sorwakeup_locked(so); return (0); } int udp_input(struct mbuf **mp, int *offp, int proto) { struct ip *ip; struct udphdr *uh; struct ifnet *ifp; struct inpcb *inp; uint16_t len, ip_len; struct inpcbinfo *pcbinfo; struct ip save_ip; struct sockaddr_in udp_in[2]; struct mbuf *m; struct m_tag *fwd_tag; int cscov_partial, iphlen; m = *mp; iphlen = *offp; ifp = m->m_pkthdr.rcvif; *mp = NULL; UDPSTAT_INC(udps_ipackets); /* * Strip IP options, if any; should skip this, make available to * user, and use on returned packets, but we don't yet have a way to * check the checksum with options still present. */ if (iphlen > sizeof (struct ip)) { ip_stripoptions(m); iphlen = sizeof(struct ip); } /* * Get IP and UDP header together in first mbuf. */ if (m->m_len < iphlen + sizeof(struct udphdr)) { if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { UDPSTAT_INC(udps_hdrops); return (IPPROTO_DONE); } } ip = mtod(m, struct ip *); uh = (struct udphdr *)((caddr_t)ip + iphlen); cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; /* * Destination port of 0 is illegal, based on RFC768. */ if (uh->uh_dport == 0) goto badunlocked; /* * Construct sockaddr format source address. Stuff source address * and datagram in user buffer. */ bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); udp_in[0].sin_len = sizeof(struct sockaddr_in); udp_in[0].sin_family = AF_INET; udp_in[0].sin_port = uh->uh_sport; udp_in[0].sin_addr = ip->ip_src; udp_in[1].sin_len = sizeof(struct sockaddr_in); udp_in[1].sin_family = AF_INET; udp_in[1].sin_port = uh->uh_dport; udp_in[1].sin_addr = ip->ip_dst; /* * Make mbuf data length reflect UDP length. If not enough data to * reflect UDP length, drop. */ len = ntohs((u_short)uh->uh_ulen); ip_len = ntohs(ip->ip_len) - iphlen; if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { /* Zero means checksum over the complete packet. */ if (len == 0) len = ip_len; cscov_partial = 0; } if (ip_len != len) { if (len > ip_len || len < sizeof(struct udphdr)) { UDPSTAT_INC(udps_badlen); goto badunlocked; } if (proto == IPPROTO_UDP) m_adj(m, len - ip_len); } /* * Save a copy of the IP header in case we want restore it for * sending an ICMP error message in response. */ if (!V_udp_blackhole) save_ip = *ip; else memset(&save_ip, 0, sizeof(save_ip)); /* * Checksum extended UDP header and data. */ if (uh->uh_sum) { u_short uh_sum; if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && !cscov_partial) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) uh_sum = m->m_pkthdr.csum_data; else uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + proto)); uh_sum ^= 0xffff; } else { char b[9]; bcopy(((struct ipovly *)ip)->ih_x1, b, 9); bzero(((struct ipovly *)ip)->ih_x1, 9); ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? uh->uh_ulen : htons(ip_len); uh_sum = in_cksum(m, len + sizeof (struct ip)); bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); } if (uh_sum) { UDPSTAT_INC(udps_badsum); m_freem(m); return (IPPROTO_DONE); } } else { if (proto == IPPROTO_UDP) { UDPSTAT_INC(udps_nosum); } else { /* UDPLite requires a checksum */ /* XXX: What is the right UDPLite MIB counter here? */ m_freem(m); return (IPPROTO_DONE); } } pcbinfo = udp_get_inpcbinfo(proto); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || in_broadcast(ip->ip_dst, ifp)) { struct inpcb *last; struct inpcbhead *pcblist; NET_EPOCH_ASSERT(); pcblist = udp_get_pcblist(proto); last = NULL; CK_LIST_FOREACH(inp, pcblist, inp_list) { if (inp->inp_lport != uh->uh_dport) continue; #ifdef INET6 if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_laddr.s_addr != INADDR_ANY && inp->inp_laddr.s_addr != ip->ip_dst.s_addr) continue; if (inp->inp_faddr.s_addr != INADDR_ANY && inp->inp_faddr.s_addr != ip->ip_src.s_addr) continue; if (inp->inp_fport != 0 && inp->inp_fport != uh->uh_sport) continue; INP_RLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_RUNLOCK(inp); continue; } /* * XXXRW: Because we weren't holding either the inpcb * or the hash lock when we checked for a match * before, we should probably recheck now that the * inpcb lock is held. */ /* * Handle socket delivery policy for any-source * and source-specific multicast. [RFC3678] */ if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { struct ip_moptions *imo; struct sockaddr_in group; int blocked; imo = inp->inp_moptions; if (imo == NULL) { INP_RUNLOCK(inp); continue; } bzero(&group, sizeof(struct sockaddr_in)); group.sin_len = sizeof(struct sockaddr_in); group.sin_family = AF_INET; group.sin_addr = ip->ip_dst; blocked = imo_multi_filter(imo, ifp, (struct sockaddr *)&group, (struct sockaddr *)&udp_in[0]); if (blocked != MCAST_PASS) { if (blocked == MCAST_NOTGMEMBER) IPSTAT_INC(ips_notmember); if (blocked == MCAST_NOTSMEMBER || blocked == MCAST_MUTED) UDPSTAT_INC(udps_filtermcast); INP_RUNLOCK(inp); continue; } } if (last != NULL) { struct mbuf *n; if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { if (proto == IPPROTO_UDPLITE) UDPLITE_PROBE(receive, NULL, last, ip, last, uh); else UDP_PROBE(receive, NULL, last, ip, last, uh); if (udp_append(last, ip, n, iphlen, udp_in)) { goto inp_lost; } } INP_RUNLOCK(last); } last = inp; /* * Don't look for additional matches if this one does * not have either the SO_REUSEPORT or SO_REUSEADDR * socket options set. This heuristic avoids * searching through all pcbs in the common case of a * non-shared port. It assumes that an application * will never clear these options after setting them. */ if ((last->inp_socket->so_options & (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) break; } if (last == NULL) { /* * No matching pcb found; discard datagram. (No need * to send an ICMP Port Unreachable for a broadcast * or multicast datgram.) */ UDPSTAT_INC(udps_noportbcast); if (inp) INP_RUNLOCK(inp); goto badunlocked; } if (proto == IPPROTO_UDPLITE) UDPLITE_PROBE(receive, NULL, last, ip, last, uh); else UDP_PROBE(receive, NULL, last, ip, last, uh); if (udp_append(last, ip, m, iphlen, udp_in) == 0) INP_RUNLOCK(last); inp_lost: return (IPPROTO_DONE); } /* * Locate pcb for datagram. */ /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ((m->m_flags & M_IP_NEXTHOP) && (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(pcbinfo, ip->ip_src, uh->uh_sport, next_hop->sin_addr, next_hop->sin_port ? htons(next_hop->sin_port) : uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, ifp); } /* Remove the tag from the packet. We don't need it anymore. */ m_tag_delete(m, fwd_tag); m->m_flags &= ~M_IP_NEXTHOP; } else inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, ifp, m); if (inp == NULL) { if (V_udp_log_in_vain) { char src[INET_ADDRSTRLEN]; char dst[INET_ADDRSTRLEN]; log(LOG_INFO, "Connection attempt to UDP %s:%d from %s:%d\n", inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); } if (proto == IPPROTO_UDPLITE) UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); else UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); UDPSTAT_INC(udps_noport); if (m->m_flags & (M_BCAST | M_MCAST)) { UDPSTAT_INC(udps_noportbcast); goto badunlocked; } if (V_udp_blackhole) goto badunlocked; if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) goto badunlocked; *ip = save_ip; icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); return (IPPROTO_DONE); } /* * Check the minimum TTL for socket. */ INP_RLOCK_ASSERT(inp); if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { if (proto == IPPROTO_UDPLITE) UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); else UDP_PROBE(receive, NULL, inp, ip, inp, uh); INP_RUNLOCK(inp); m_freem(m); return (IPPROTO_DONE); } if (cscov_partial) { struct udpcb *up; up = intoudpcb(inp); if (up->u_rxcslen == 0 || up->u_rxcslen > len) { INP_RUNLOCK(inp); m_freem(m); return (IPPROTO_DONE); } } if (proto == IPPROTO_UDPLITE) UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); else UDP_PROBE(receive, NULL, inp, ip, inp, uh); if (udp_append(inp, ip, m, iphlen, udp_in) == 0) INP_RUNLOCK(inp); return (IPPROTO_DONE); badunlocked: m_freem(m); return (IPPROTO_DONE); } #endif /* INET */ /* * Notify a udp user of an asynchronous error; just wake up so that they can * collect error status. */ struct inpcb * udp_notify(struct inpcb *inp, int errno) { INP_WLOCK_ASSERT(inp); if ((errno == EHOSTUNREACH || errno == ENETUNREACH || - errno == EHOSTDOWN) && inp->inp_route.ro_rt) { - RTFREE(inp->inp_route.ro_rt); - inp->inp_route.ro_rt = (struct rtentry *)NULL; + errno == EHOSTDOWN) && inp->inp_route.ro_nh) { + NH_FREE(inp->inp_route.ro_nh); + inp->inp_route.ro_nh = (struct nhop_object *)NULL; } inp->inp_socket->so_error = errno; sorwakeup(inp->inp_socket); sowwakeup(inp->inp_socket); return (inp); } #ifdef INET static void udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, struct inpcbinfo *pcbinfo) { struct ip *ip = vip; struct udphdr *uh; struct in_addr faddr; struct inpcb *inp; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); return; } /* * Hostdead is ugly because it goes linearly through all PCBs. * * XXX: We never get this from ICMP, otherwise it makes an excellent * DoS attack on machines with many connections. */ if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip != NULL) { uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL) { INP_WLOCK_ASSERT(inp); if (inp->inp_socket != NULL) { udp_notify(inp, inetctlerrmap[cmd]); } INP_WUNLOCK(inp); } else { inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, ip->ip_src, uh->uh_sport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { struct udpcb *up; void *ctx; udp_tun_icmp_t func; up = intoudpcb(inp); ctx = up->u_tun_ctx; func = up->u_icmp_func; INP_RUNLOCK(inp); if (func != NULL) (*func)(cmd, sa, vip, ctx); } } } else in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], udp_notify); } void udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); } void udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) { return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); } #endif /* INET */ static int udp_pcblist(SYSCTL_HANDLER_ARGS) { struct xinpgen xig; struct epoch_tracker et; struct inpcb *inp; int error; if (req->newptr != 0) return (EPERM); if (req->oldptr == 0) { int n; n = V_udbinfo.ipi_count; n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); return (0); } if ((error = sysctl_wire_old_buffer(req, 0)) != 0) return (error); bzero(&xig, sizeof(xig)); xig.xig_len = sizeof xig; xig.xig_count = V_udbinfo.ipi_count; xig.xig_gen = V_udbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); NET_EPOCH_ENTER(et); for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead); inp != NULL; inp = CK_LIST_NEXT(inp, inp_list)) { INP_RLOCK(inp); if (inp->inp_gencnt <= xig.xig_gen && cr_canseeinpcb(req->td->td_ucred, inp) == 0) { struct xinpcb xi; in_pcbtoxinpcb(inp, &xi); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xi, sizeof xi); if (error) break; } else INP_RUNLOCK(inp); } NET_EPOCH_EXIT(et); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ xig.xig_gen = V_udbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_udbinfo.ipi_count; error = SYSCTL_OUT(req, &xig, sizeof xig); } return (error); } SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, udp_pcblist, "S,xinpcb", "List of active UDP sockets"); #ifdef INET static int udp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct epoch_tracker et; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); NET_EPOCH_ENTER(et); inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { INP_RLOCK_ASSERT(inp); if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 0, 0, udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); #endif /* INET */ int udp_ctloutput(struct socket *so, struct sockopt *sopt) { struct inpcb *inp; struct udpcb *up; int isudplite, error, optval; error = 0; isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; inp = sotoinpcb(so); KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); INP_WLOCK(inp); if (sopt->sopt_level != so->so_proto->pr_protocol) { #ifdef INET6 if (INP_CHECK_SOCKAF(so, AF_INET6)) { INP_WUNLOCK(inp); error = ip6_ctloutput(so, sopt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { INP_WUNLOCK(inp); error = ip_ctloutput(so, sopt); } #endif return (error); } switch (sopt->sopt_dir) { case SOPT_SET: switch (sopt->sopt_name) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET case UDP_ENCAP: if (!IPSEC_ENABLED(ipv4)) { INP_WUNLOCK(inp); return (ENOPROTOOPT); } error = UDPENCAP_PCBCTL(inp, sopt); break; #endif /* INET */ #endif /* IPSEC */ case UDPLITE_SEND_CSCOV: case UDPLITE_RECV_CSCOV: if (!isudplite) { INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } INP_WUNLOCK(inp); error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error != 0) break; inp = sotoinpcb(so); KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); INP_WLOCK(inp); up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); if ((optval != 0 && optval < 8) || (optval > 65535)) { INP_WUNLOCK(inp); error = EINVAL; break; } if (sopt->sopt_name == UDPLITE_SEND_CSCOV) up->u_txcslen = optval; else up->u_rxcslen = optval; INP_WUNLOCK(inp); break; default: INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } break; case SOPT_GET: switch (sopt->sopt_name) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET case UDP_ENCAP: if (!IPSEC_ENABLED(ipv4)) { INP_WUNLOCK(inp); return (ENOPROTOOPT); } error = UDPENCAP_PCBCTL(inp, sopt); break; #endif /* INET */ #endif /* IPSEC */ case UDPLITE_SEND_CSCOV: case UDPLITE_RECV_CSCOV: if (!isudplite) { INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); if (sopt->sopt_name == UDPLITE_SEND_CSCOV) optval = up->u_txcslen; else optval = up->u_rxcslen; INP_WUNLOCK(inp); error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } break; } return (error); } #ifdef INET static int udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { struct udpiphdr *ui; int len = m->m_pkthdr.len; struct in_addr faddr, laddr; struct cmsghdr *cm; struct inpcbinfo *pcbinfo; struct sockaddr_in *sin, src; struct epoch_tracker et; int cscov_partial = 0; int error = 0; int ipflags; u_short fport, lport; u_char tos; uint8_t pr; uint16_t cscov = 0; uint32_t flowid = 0; uint8_t flowtype = M_HASHTYPE_NONE; if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { if (control) m_freem(control); m_freem(m); return (EMSGSIZE); } src.sin_family = 0; sin = (struct sockaddr_in *)addr; /* * udp_output() may need to temporarily bind or connect the current * inpcb. As such, we don't know up front whether we will need the * pcbinfo lock or not. Do any work to decide what is needed up * front before acquiring any locks. * * We will need network epoch in either case, to safely lookup into * pcb hash. */ if (sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) INP_WLOCK(inp); else INP_RLOCK(inp); NET_EPOCH_ENTER(et); tos = inp->inp_ip_tos; if (control != NULL) { /* * XXX: Currently, we assume all the optional information is * stored in a single mbuf. */ if (control->m_next) { m_freem(control); error = EINVAL; goto release; } for (; control->m_len > 0; control->m_data += CMSG_ALIGN(cm->cmsg_len), control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { cm = mtod(control, struct cmsghdr *); if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 || cm->cmsg_len > control->m_len) { error = EINVAL; break; } if (cm->cmsg_level != IPPROTO_IP) continue; switch (cm->cmsg_type) { case IP_SENDSRCADDR: if (cm->cmsg_len != CMSG_LEN(sizeof(struct in_addr))) { error = EINVAL; break; } bzero(&src, sizeof(src)); src.sin_family = AF_INET; src.sin_len = sizeof(src); src.sin_port = inp->inp_lport; src.sin_addr = *(struct in_addr *)CMSG_DATA(cm); break; case IP_TOS: if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { error = EINVAL; break; } tos = *(u_char *)CMSG_DATA(cm); break; case IP_FLOWID: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } flowid = *(uint32_t *) CMSG_DATA(cm); break; case IP_FLOWTYPE: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } flowtype = *(uint32_t *) CMSG_DATA(cm); break; #ifdef RSS case IP_RSSBUCKETID: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } /* This is just a placeholder for now */ break; #endif /* RSS */ default: error = ENOPROTOOPT; break; } if (error) break; } m_freem(control); } if (error) goto release; pr = inp->inp_socket->so_proto->pr_protocol; pcbinfo = udp_get_inpcbinfo(pr); /* * If the IP_SENDSRCADDR control message was specified, override the * source address for this datagram. Its use is invalidated if the * address thus specified is incomplete or clobbers other inpcbs. */ laddr = inp->inp_laddr; lport = inp->inp_lport; if (src.sin_family == AF_INET) { INP_HASH_LOCK_ASSERT(pcbinfo); if ((lport == 0) || (laddr.s_addr == INADDR_ANY && src.sin_addr.s_addr == INADDR_ANY)) { error = EINVAL; goto release; } error = in_pcbbind_setup(inp, (struct sockaddr *)&src, &laddr.s_addr, &lport, td->td_ucred); if (error) goto release; } /* * If a UDP socket has been connected, then a local address/port will * have been selected and bound. * * If a UDP socket has not been connected to, then an explicit * destination address must be used, in which case a local * address/port may not have been selected and bound. */ if (sin != NULL) { INP_LOCK_ASSERT(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { error = EISCONN; goto release; } /* * Jail may rewrite the destination address, so let it do * that before we use it. */ error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); if (error) goto release; /* * If a local address or port hasn't yet been selected, or if * the destination address needs to be rewritten due to using * a special INADDR_ constant, invoke in_pcbconnect_setup() * to do the heavy lifting. Once a port is selected, we * commit the binding back to the socket; we also commit the * binding of the address if in jail. * * If we already have a valid binding and we're not * requesting a destination address rewrite, use a fast path. */ if (inp->inp_laddr.s_addr == INADDR_ANY || inp->inp_lport == 0 || sin->sin_addr.s_addr == INADDR_ANY || sin->sin_addr.s_addr == INADDR_BROADCAST) { INP_HASH_LOCK_ASSERT(pcbinfo); error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport, &faddr.s_addr, &fport, NULL, td->td_ucred); if (error) goto release; /* * XXXRW: Why not commit the port if the address is * !INADDR_ANY? */ /* Commit the local port if newly assigned. */ if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { INP_WLOCK_ASSERT(inp); /* * Remember addr if jailed, to prevent * rebinding. */ if (prison_flag(td->td_ucred, PR_IP4)) inp->inp_laddr = laddr; inp->inp_lport = lport; INP_HASH_WLOCK(pcbinfo); error = in_pcbinshash(inp); INP_HASH_WUNLOCK(pcbinfo); if (error != 0) { inp->inp_lport = 0; error = EAGAIN; goto release; } inp->inp_flags |= INP_ANONPORT; } } else { faddr = sin->sin_addr; fport = sin->sin_port; } } else { INP_LOCK_ASSERT(inp); faddr = inp->inp_faddr; fport = inp->inp_fport; if (faddr.s_addr == INADDR_ANY) { error = ENOTCONN; goto release; } } /* * Calculate data length and get a mbuf for UDP, IP, and possible * link-layer headers. Immediate slide the data pointer back forward * since we won't use that space at this layer. */ M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); if (m == NULL) { error = ENOBUFS; goto release; } m->m_data += max_linkhdr; m->m_len -= max_linkhdr; m->m_pkthdr.len -= max_linkhdr; /* * Fill in mbuf with extended UDP header and addresses and length put * into network format. */ ui = mtod(m, struct udpiphdr *); bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ ui->ui_v = IPVERSION << 4; ui->ui_pr = pr; ui->ui_src = laddr; ui->ui_dst = faddr; ui->ui_sport = lport; ui->ui_dport = fport; ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); if (pr == IPPROTO_UDPLITE) { struct udpcb *up; uint16_t plen; up = intoudpcb(inp); cscov = up->u_txcslen; plen = (u_short)len + sizeof(struct udphdr); if (cscov >= plen) cscov = 0; ui->ui_len = htons(plen); ui->ui_ulen = htons(cscov); /* * For UDP-Lite, checksum coverage length of zero means * the entire UDPLite packet is covered by the checksum. */ cscov_partial = (cscov == 0) ? 0 : 1; } /* * Set the Don't Fragment bit in the IP header. */ if (inp->inp_flags & INP_DONTFRAG) { struct ip *ip; ip = (struct ip *)&ui->ui_i; ip->ip_off |= htons(IP_DF); } ipflags = 0; if (inp->inp_socket->so_options & SO_DONTROUTE) ipflags |= IP_ROUTETOIF; if (inp->inp_socket->so_options & SO_BROADCAST) ipflags |= IP_ALLOWBROADCAST; if (inp->inp_flags & INP_ONESBCAST) ipflags |= IP_SENDONES; #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif /* * Set up checksum and output datagram. */ ui->ui_sum = 0; if (pr == IPPROTO_UDPLITE) { if (inp->inp_flags & INP_ONESBCAST) faddr.s_addr = INADDR_BROADCAST; if (cscov_partial) { if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) ui->ui_sum = 0xffff; } else { if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) ui->ui_sum = 0xffff; } } else if (V_udp_cksum) { if (inp->inp_flags & INP_ONESBCAST) faddr.s_addr = INADDR_BROADCAST; ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, htons((u_short)len + sizeof(struct udphdr) + pr)); m->m_pkthdr.csum_flags = CSUM_UDP; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); } ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ ((struct ip *)ui)->ip_tos = tos; /* XXX */ UDPSTAT_INC(udps_opackets); /* * Setup flowid / RSS information for outbound socket. * * Once the UDP code decides to set a flowid some other way, * this allows the flowid to be overridden by userland. */ if (flowtype != M_HASHTYPE_NONE) { m->m_pkthdr.flowid = flowid; M_HASHTYPE_SET(m, flowtype); } #ifdef RSS else { uint32_t hash_val, hash_type; /* * Calculate an appropriate RSS hash for UDP and * UDP Lite. * * The called function will take care of figuring out * whether a 2-tuple or 4-tuple hash is required based * on the currently configured scheme. * * Later later on connected socket values should be * cached in the inpcb and reused, rather than constantly * re-calculating it. * * UDP Lite is a different protocol number and will * likely end up being hashed as a 2-tuple until * RSS / NICs grow UDP Lite protocol awareness. */ if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, pr, &hash_val, &hash_type) == 0) { m->m_pkthdr.flowid = hash_val; M_HASHTYPE_SET(m, hash_type); } } /* * Don't override with the inp cached flowid value. * * Depending upon the kind of send being done, the inp * flowid/flowtype values may actually not be appropriate * for this particular socket send. * * We should either leave the flowid at zero (which is what is * currently done) or set it to some software generated * hash value based on the packet contents. */ ipflags |= IP_NODEFAULTFLOWID; #endif /* RSS */ if (pr == IPPROTO_UDPLITE) UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); else UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); error = ip_output(m, inp->inp_options, INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags, inp->inp_moptions, inp); INP_UNLOCK(inp); NET_EPOCH_EXIT(et); return (error); release: INP_UNLOCK(inp); NET_EPOCH_EXIT(et); m_freem(m); return (error); } static void udp_abort(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_abort: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); soisdisconnected(so); } INP_WUNLOCK(inp); } static int udp_attach(struct socket *so, int proto, struct thread *td) { static uint32_t udp_flowid; struct inpcb *inp; struct inpcbinfo *pcbinfo; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp == NULL, ("udp_attach: inp != NULL")); error = soreserve(so, udp_sendspace, udp_recvspace); if (error) return (error); INP_INFO_WLOCK(pcbinfo); error = in_pcballoc(so, pcbinfo); if (error) { INP_INFO_WUNLOCK(pcbinfo); return (error); } inp = sotoinpcb(so); inp->inp_vflag |= INP_IPV4; inp->inp_ip_ttl = V_ip_defttl; inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); inp->inp_flowtype = M_HASHTYPE_OPAQUE; error = udp_newudpcb(inp); if (error) { in_pcbdetach(inp); in_pcbfree(inp); INP_INFO_WUNLOCK(pcbinfo); return (error); } INP_WUNLOCK(inp); INP_INFO_WUNLOCK(pcbinfo); return (0); } #endif /* INET */ int udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) { struct inpcb *inp; struct udpcb *up; KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram")); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); INP_WLOCK(inp); up = intoudpcb(inp); if ((up->u_tun_func != NULL) || (up->u_icmp_func != NULL)) { INP_WUNLOCK(inp); return (EBUSY); } up->u_tun_func = f; up->u_icmp_func = i; up->u_tun_ctx = ctx; INP_WUNLOCK(inp); return (0); } #ifdef INET static int udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { struct inpcb *inp; struct inpcbinfo *pcbinfo; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_bind: inp == NULL")); INP_WLOCK(inp); INP_HASH_WLOCK(pcbinfo); error = in_pcbbind(inp, nam, td->td_ucred); INP_HASH_WUNLOCK(pcbinfo); INP_WUNLOCK(inp); return (error); } static void udp_close(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_close: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); soisdisconnected(so); } INP_WUNLOCK(inp); } static int udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { struct epoch_tracker et; struct inpcb *inp; struct inpcbinfo *pcbinfo; struct sockaddr_in *sin; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_connect: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_WUNLOCK(inp); return (EISCONN); } sin = (struct sockaddr_in *)nam; error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); if (error != 0) { INP_WUNLOCK(inp); return (error); } NET_EPOCH_ENTER(et); INP_HASH_WLOCK(pcbinfo); error = in_pcbconnect(inp, nam, td->td_ucred); INP_HASH_WUNLOCK(pcbinfo); NET_EPOCH_EXIT(et); if (error == 0) soisconnected(so); INP_WUNLOCK(inp); return (error); } static void udp_detach(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; struct udpcb *up; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_detach: inp == NULL")); KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, ("udp_detach: not disconnected")); INP_INFO_WLOCK(pcbinfo); INP_WLOCK(inp); up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); inp->inp_ppcb = NULL; in_pcbdetach(inp); in_pcbfree(inp); INP_INFO_WUNLOCK(pcbinfo); udp_discardcb(up); } static int udp_disconnect(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr == INADDR_ANY) { INP_WUNLOCK(inp); return (ENOTCONN); } INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; /* XXX */ SOCK_UNLOCK(so); INP_WUNLOCK(inp); return (0); } static int udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_send: inp == NULL")); return (udp_output(inp, m, addr, control, td)); } #endif /* INET */ int udp_shutdown(struct socket *so) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); INP_WLOCK(inp); socantsendmore(so); INP_WUNLOCK(inp); return (0); } #ifdef INET struct pr_usrreqs udp_usrreqs = { .pru_abort = udp_abort, .pru_attach = udp_attach, .pru_bind = udp_bind, .pru_connect = udp_connect, .pru_control = in_control, .pru_detach = udp_detach, .pru_disconnect = udp_disconnect, .pru_peeraddr = in_getpeeraddr, .pru_send = udp_send, .pru_soreceive = soreceive_dgram, .pru_sosend = sosend_dgram, .pru_shutdown = udp_shutdown, .pru_sockaddr = in_getsockaddr, .pru_sosetlabel = in_pcbsosetlabel, .pru_close = udp_close, }; #endif /* INET */ Index: head/sys/netinet6/in6.h =================================================================== --- head/sys/netinet6/in6.h (revision 360291) +++ head/sys/netinet6/in6.h (revision 360292) @@ -1,750 +1,751 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: in6.h,v 1.89 2001/05/27 13:28:35 itojun Exp $ */ /*- * Copyright (c) 1982, 1986, 1990, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)in.h 8.3 (Berkeley) 1/3/94 * $FreeBSD$ */ #ifndef __KAME_NETINET_IN_H_INCLUDED_ #error "do not include netinet6/in6.h directly, include netinet/in.h. see RFC2553" #endif #ifndef _NETINET6_IN6_H_ #define _NETINET6_IN6_H_ /* * Identification of the network protocol stack * for *BSD-current/release: http://www.kame.net/dev/cvsweb.cgi/kame/COVERAGE * has the table of implementation/integration differences. */ #define __KAME__ #define __KAME_VERSION "FreeBSD" /* * IPv6 port allocation rules should mirror the IPv4 rules and are controlled * by the net.inet.ip.portrange sysctl tree. The following defines exist * for compatibility with userland applications that need them. */ #if __BSD_VISIBLE #define IPV6PORT_RESERVED 1024 #define IPV6PORT_ANONMIN 49152 #define IPV6PORT_ANONMAX 65535 #define IPV6PORT_RESERVEDMIN 600 #define IPV6PORT_RESERVEDMAX (IPV6PORT_RESERVED-1) #endif /* * IPv6 address */ struct in6_addr { union { uint8_t __u6_addr8[16]; uint16_t __u6_addr16[8]; uint32_t __u6_addr32[4]; } __u6_addr; /* 128-bit IP6 address */ }; #define s6_addr __u6_addr.__u6_addr8 #ifdef _KERNEL /* XXX nonstandard */ #define s6_addr8 __u6_addr.__u6_addr8 #define s6_addr16 __u6_addr.__u6_addr16 #define s6_addr32 __u6_addr.__u6_addr32 #endif #define INET6_ADDRSTRLEN 46 /* * XXX missing POSIX.1-2001 macro IPPROTO_IPV6. */ /* * Socket address for IPv6 */ #if __BSD_VISIBLE #define SIN6_LEN #endif struct sockaddr_in6 { uint8_t sin6_len; /* length of this struct */ sa_family_t sin6_family; /* AF_INET6 */ in_port_t sin6_port; /* Transport layer port # */ uint32_t sin6_flowinfo; /* IP6 flow information */ struct in6_addr sin6_addr; /* IP6 address */ uint32_t sin6_scope_id; /* scope zone index */ }; /* * Local definition for masks */ #ifdef _KERNEL /* XXX nonstandard */ #define IN6MASK0 {{{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }}} #define IN6MASK32 {{{ 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }}} #define IN6MASK64 {{{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }}} #define IN6MASK96 {{{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, \ 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00 }}} #define IN6MASK128 {{{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, \ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }}} #endif #ifdef _KERNEL extern const struct sockaddr_in6 sa6_any; extern const struct in6_addr in6mask0; extern const struct in6_addr in6mask32; extern const struct in6_addr in6mask64; extern const struct in6_addr in6mask96; extern const struct in6_addr in6mask128; #endif /* _KERNEL */ /* * Macros started with IPV6_ADDR is KAME local */ #ifdef _KERNEL /* XXX nonstandard */ #if _BYTE_ORDER == _BIG_ENDIAN #define IPV6_ADDR_INT32_ONE 1 #define IPV6_ADDR_INT32_TWO 2 #define IPV6_ADDR_INT32_MNL 0xff010000 #define IPV6_ADDR_INT32_MLL 0xff020000 #define IPV6_ADDR_INT32_SMP 0x0000ffff #define IPV6_ADDR_INT16_ULL 0xfe80 #define IPV6_ADDR_INT16_USL 0xfec0 #define IPV6_ADDR_INT16_MLL 0xff02 #elif _BYTE_ORDER == _LITTLE_ENDIAN #define IPV6_ADDR_INT32_ONE 0x01000000 #define IPV6_ADDR_INT32_TWO 0x02000000 #define IPV6_ADDR_INT32_MNL 0x000001ff #define IPV6_ADDR_INT32_MLL 0x000002ff #define IPV6_ADDR_INT32_SMP 0xffff0000 #define IPV6_ADDR_INT16_ULL 0x80fe #define IPV6_ADDR_INT16_USL 0xc0fe #define IPV6_ADDR_INT16_MLL 0x02ff #endif #endif /* * Definition of some useful macros to handle IP6 addresses */ #if __BSD_VISIBLE #define IN6ADDR_ANY_INIT \ {{{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }}} #define IN6ADDR_LOOPBACK_INIT \ {{{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01 }}} #define IN6ADDR_NODELOCAL_ALLNODES_INIT \ {{{ 0xff, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01 }}} #define IN6ADDR_INTFACELOCAL_ALLNODES_INIT \ {{{ 0xff, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01 }}} #define IN6ADDR_LINKLOCAL_ALLNODES_INIT \ {{{ 0xff, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01 }}} #define IN6ADDR_LINKLOCAL_ALLROUTERS_INIT \ {{{ 0xff, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02 }}} #define IN6ADDR_LINKLOCAL_ALLV2ROUTERS_INIT \ {{{ 0xff, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, \ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x16 }}} #endif extern const struct in6_addr in6addr_any; extern const struct in6_addr in6addr_loopback; #if __BSD_VISIBLE extern const struct in6_addr in6addr_nodelocal_allnodes; extern const struct in6_addr in6addr_linklocal_allnodes; extern const struct in6_addr in6addr_linklocal_allrouters; extern const struct in6_addr in6addr_linklocal_allv2routers; #endif /* * Equality * NOTE: Some of kernel programming environment (for example, openbsd/sparc) * does not supply memcmp(). For userland memcmp() is preferred as it is * in ANSI standard. */ #ifdef _KERNEL #define IN6_ARE_ADDR_EQUAL(a, b) \ (bcmp(&(a)->s6_addr[0], &(b)->s6_addr[0], sizeof(struct in6_addr)) == 0) #else #if __BSD_VISIBLE #define IN6_ARE_ADDR_EQUAL(a, b) \ (memcmp(&(a)->s6_addr[0], &(b)->s6_addr[0], sizeof(struct in6_addr)) == 0) #endif #endif /* * Unspecified */ #define IN6_IS_ADDR_UNSPECIFIED(a) \ ((a)->__u6_addr.__u6_addr32[0] == 0 && \ (a)->__u6_addr.__u6_addr32[1] == 0 && \ (a)->__u6_addr.__u6_addr32[2] == 0 && \ (a)->__u6_addr.__u6_addr32[3] == 0) /* * Loopback */ #define IN6_IS_ADDR_LOOPBACK(a) \ ((a)->__u6_addr.__u6_addr32[0] == 0 && \ (a)->__u6_addr.__u6_addr32[1] == 0 && \ (a)->__u6_addr.__u6_addr32[2] == 0 && \ (a)->__u6_addr.__u6_addr32[3] == ntohl(1)) /* * IPv4 compatible */ #define IN6_IS_ADDR_V4COMPAT(a) \ ((a)->__u6_addr.__u6_addr32[0] == 0 && \ (a)->__u6_addr.__u6_addr32[1] == 0 && \ (a)->__u6_addr.__u6_addr32[2] == 0 && \ (a)->__u6_addr.__u6_addr32[3] != 0 && \ (a)->__u6_addr.__u6_addr32[3] != ntohl(1)) /* * Mapped */ #define IN6_IS_ADDR_V4MAPPED(a) \ ((a)->__u6_addr.__u6_addr32[0] == 0 && \ (a)->__u6_addr.__u6_addr32[1] == 0 && \ (a)->__u6_addr.__u6_addr32[2] == ntohl(0x0000ffff)) /* * KAME Scope Values */ #ifdef _KERNEL /* XXX nonstandard */ #define IPV6_ADDR_SCOPE_NODELOCAL 0x01 #define IPV6_ADDR_SCOPE_INTFACELOCAL 0x01 #define IPV6_ADDR_SCOPE_LINKLOCAL 0x02 #define IPV6_ADDR_SCOPE_SITELOCAL 0x05 #define IPV6_ADDR_SCOPE_ORGLOCAL 0x08 /* just used in this file */ #define IPV6_ADDR_SCOPE_GLOBAL 0x0e #else #define __IPV6_ADDR_SCOPE_NODELOCAL 0x01 #define __IPV6_ADDR_SCOPE_INTFACELOCAL 0x01 #define __IPV6_ADDR_SCOPE_LINKLOCAL 0x02 #define __IPV6_ADDR_SCOPE_SITELOCAL 0x05 #define __IPV6_ADDR_SCOPE_ORGLOCAL 0x08 /* just used in this file */ #define __IPV6_ADDR_SCOPE_GLOBAL 0x0e #endif /* * Unicast Scope * Note that we must check topmost 10 bits only, not 16 bits (see RFC2373). */ #define IN6_IS_ADDR_LINKLOCAL(a) \ (((a)->s6_addr[0] == 0xfe) && (((a)->s6_addr[1] & 0xc0) == 0x80)) #define IN6_IS_ADDR_SITELOCAL(a) \ (((a)->s6_addr[0] == 0xfe) && (((a)->s6_addr[1] & 0xc0) == 0xc0)) /* * Multicast */ #define IN6_IS_ADDR_MULTICAST(a) ((a)->s6_addr[0] == 0xff) #ifdef _KERNEL /* XXX nonstandard */ #define IPV6_ADDR_MC_SCOPE(a) ((a)->s6_addr[1] & 0x0f) #else #define __IPV6_ADDR_MC_SCOPE(a) ((a)->s6_addr[1] & 0x0f) #endif /* * Multicast Scope */ #ifdef _KERNEL /* refers nonstandard items */ #define IN6_IS_ADDR_MC_NODELOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_NODELOCAL)) #define IN6_IS_ADDR_MC_INTFACELOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_INTFACELOCAL)) #define IN6_IS_ADDR_MC_LINKLOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_LINKLOCAL)) #define IN6_IS_ADDR_MC_SITELOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_SITELOCAL)) #define IN6_IS_ADDR_MC_ORGLOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_ORGLOCAL)) #define IN6_IS_ADDR_MC_GLOBAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (IPV6_ADDR_MC_SCOPE(a) == IPV6_ADDR_SCOPE_GLOBAL)) #else #define IN6_IS_ADDR_MC_NODELOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (__IPV6_ADDR_MC_SCOPE(a) == __IPV6_ADDR_SCOPE_NODELOCAL)) #define IN6_IS_ADDR_MC_LINKLOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (__IPV6_ADDR_MC_SCOPE(a) == __IPV6_ADDR_SCOPE_LINKLOCAL)) #define IN6_IS_ADDR_MC_SITELOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (__IPV6_ADDR_MC_SCOPE(a) == __IPV6_ADDR_SCOPE_SITELOCAL)) #define IN6_IS_ADDR_MC_ORGLOCAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (__IPV6_ADDR_MC_SCOPE(a) == __IPV6_ADDR_SCOPE_ORGLOCAL)) #define IN6_IS_ADDR_MC_GLOBAL(a) \ (IN6_IS_ADDR_MULTICAST(a) && \ (__IPV6_ADDR_MC_SCOPE(a) == __IPV6_ADDR_SCOPE_GLOBAL)) #endif #ifdef _KERNEL /* nonstandard */ /* * KAME Scope */ #define IN6_IS_SCOPE_LINKLOCAL(a) \ ((IN6_IS_ADDR_LINKLOCAL(a)) || \ (IN6_IS_ADDR_MC_LINKLOCAL(a))) #define IN6_IS_SCOPE_EMBED(a) \ ((IN6_IS_ADDR_LINKLOCAL(a)) || \ (IN6_IS_ADDR_MC_LINKLOCAL(a)) || \ (IN6_IS_ADDR_MC_INTFACELOCAL(a))) #define IFA6_IS_DEPRECATED(a) \ ((a)->ia6_lifetime.ia6t_pltime != ND6_INFINITE_LIFETIME && \ (u_int32_t)((time_uptime - (a)->ia6_updatetime)) > \ (a)->ia6_lifetime.ia6t_pltime) #define IFA6_IS_INVALID(a) \ ((a)->ia6_lifetime.ia6t_vltime != ND6_INFINITE_LIFETIME && \ (u_int32_t)((time_uptime - (a)->ia6_updatetime)) > \ (a)->ia6_lifetime.ia6t_vltime) #endif /* _KERNEL */ /* * IP6 route structure */ #if __BSD_VISIBLE +struct nhop_object; struct route_in6 { - struct rtentry *ro_rt; + struct nhop_object *ro_nh; struct llentry *ro_lle; /* * ro_prepend and ro_plen are only used for bpf to pass in a * preformed header. They are not cacheable. */ char *ro_prepend; uint16_t ro_plen; uint16_t ro_flags; uint16_t ro_mtu; /* saved ro_rt mtu */ uint16_t spare; struct sockaddr_in6 ro_dst; }; #endif #ifdef _KERNEL #define MTAG_ABI_IPV6 1444287380 /* IPv6 ABI */ #define IPV6_TAG_DIRECT 0 /* direct-dispatch IPv6 */ #endif /* _KERNEL */ /* * Options for use with [gs]etsockopt at the IPV6 level. * First word of comment is data type; bool is stored in int. */ /* no hdrincl */ #if 0 /* the followings are relic in IPv4 and hence are disabled */ #define IPV6_OPTIONS 1 /* buf/ip6_opts; set/get IP6 options */ #define IPV6_RECVOPTS 5 /* bool; receive all IP6 opts w/dgram */ #define IPV6_RECVRETOPTS 6 /* bool; receive IP6 opts for response */ #define IPV6_RECVDSTADDR 7 /* bool; receive IP6 dst addr w/dgram */ #define IPV6_RETOPTS 8 /* ip6_opts; set/get IP6 options */ #endif #define IPV6_SOCKOPT_RESERVED1 3 /* reserved for future use */ #define IPV6_UNICAST_HOPS 4 /* int; IP6 hops */ #define IPV6_MULTICAST_IF 9 /* u_int; set/get IP6 multicast i/f */ #define IPV6_MULTICAST_HOPS 10 /* int; set/get IP6 multicast hops */ #define IPV6_MULTICAST_LOOP 11 /* u_int; set/get IP6 multicast loopback */ #define IPV6_JOIN_GROUP 12 /* ipv6_mreq; join a group membership */ #define IPV6_LEAVE_GROUP 13 /* ipv6_mreq; leave a group membership */ #define IPV6_PORTRANGE 14 /* int; range to choose for unspec port */ #define ICMP6_FILTER 18 /* icmp6_filter; icmp6 filter */ /* RFC2292 options */ #ifdef _KERNEL #define IPV6_2292PKTINFO 19 /* bool; send/recv if, src/dst addr */ #define IPV6_2292HOPLIMIT 20 /* bool; hop limit */ #define IPV6_2292NEXTHOP 21 /* bool; next hop addr */ #define IPV6_2292HOPOPTS 22 /* bool; hop-by-hop option */ #define IPV6_2292DSTOPTS 23 /* bool; destinaion option */ #define IPV6_2292RTHDR 24 /* bool; routing header */ #define IPV6_2292PKTOPTIONS 25 /* buf/cmsghdr; set/get IPv6 options */ #endif #define IPV6_CHECKSUM 26 /* int; checksum offset for raw socket */ #define IPV6_V6ONLY 27 /* bool; make AF_INET6 sockets v6 only */ #ifndef _KERNEL #define IPV6_BINDV6ONLY IPV6_V6ONLY #endif #define IPV6_IPSEC_POLICY 28 /* struct; get/set security policy */ /* 29; unused; was IPV6_FAITH */ #if 1 /* IPV6FIREWALL */ #define IPV6_FW_ADD 30 /* add a firewall rule to chain */ #define IPV6_FW_DEL 31 /* delete a firewall rule from chain */ #define IPV6_FW_FLUSH 32 /* flush firewall rule chain */ #define IPV6_FW_ZERO 33 /* clear single/all firewall counter(s) */ #define IPV6_FW_GET 34 /* get entire firewall rule chain */ #endif /* new socket options introduced in RFC3542 */ #define IPV6_RTHDRDSTOPTS 35 /* ip6_dest; send dst option before rthdr */ #define IPV6_RECVPKTINFO 36 /* bool; recv if, dst addr */ #define IPV6_RECVHOPLIMIT 37 /* bool; recv hop limit */ #define IPV6_RECVRTHDR 38 /* bool; recv routing header */ #define IPV6_RECVHOPOPTS 39 /* bool; recv hop-by-hop option */ #define IPV6_RECVDSTOPTS 40 /* bool; recv dst option after rthdr */ #ifdef _KERNEL #define IPV6_RECVRTHDRDSTOPTS 41 /* bool; recv dst option before rthdr */ #endif #define IPV6_USE_MIN_MTU 42 /* bool; send packets at the minimum MTU */ #define IPV6_RECVPATHMTU 43 /* bool; notify an according MTU */ #define IPV6_PATHMTU 44 /* mtuinfo; get the current path MTU (sopt), 4 bytes int; MTU notification (cmsg) */ #if 0 /*obsoleted during 2292bis -> 3542*/ #define IPV6_REACHCONF 45 /* no data; ND reachability confirm (cmsg only/not in of RFC3542) */ #endif /* more new socket options introduced in RFC3542 */ #define IPV6_PKTINFO 46 /* in6_pktinfo; send if, src addr */ #define IPV6_HOPLIMIT 47 /* int; send hop limit */ #define IPV6_NEXTHOP 48 /* sockaddr; next hop addr */ #define IPV6_HOPOPTS 49 /* ip6_hbh; send hop-by-hop option */ #define IPV6_DSTOPTS 50 /* ip6_dest; send dst option befor rthdr */ #define IPV6_RTHDR 51 /* ip6_rthdr; send routing header */ #if 0 #define IPV6_PKTOPTIONS 52 /* buf/cmsghdr; set/get IPv6 options */ /* obsoleted by RFC3542 */ #endif #define IPV6_RECVTCLASS 57 /* bool; recv traffic class values */ #define IPV6_AUTOFLOWLABEL 59 /* bool; attach flowlabel automagically */ #define IPV6_TCLASS 61 /* int; send traffic class value */ #define IPV6_DONTFRAG 62 /* bool; disable IPv6 fragmentation */ #define IPV6_PREFER_TEMPADDR 63 /* int; prefer temporary addresses as * the source address. */ #define IPV6_BINDANY 64 /* bool: allow bind to any address */ #define IPV6_BINDMULTI 65 /* bool; allow multibind to same addr/port */ #define IPV6_RSS_LISTEN_BUCKET 66 /* int; set RSS listen bucket */ #define IPV6_FLOWID 67 /* int; flowid of given socket */ #define IPV6_FLOWTYPE 68 /* int; flowtype of given socket */ #define IPV6_RSSBUCKETID 69 /* int; RSS bucket ID of given socket */ #define IPV6_RECVFLOWID 70 /* bool; receive IP6 flowid/flowtype w/ datagram */ #define IPV6_RECVRSSBUCKETID 71 /* bool; receive IP6 RSS bucket id w/ datagram */ #define IPV6_ORIGDSTADDR 72 /* bool: allow getting dstaddr /port info */ #define IPV6_RECVORIGDSTADDR IPV6_ORIGDSTADDR /* * The following option is private; do not use it from user applications. * It is deliberately defined to the same value as IP_MSFILTER. */ #define IPV6_MSFILTER 74 /* struct __msfilterreq; * set/get multicast source filter list. */ /* to define items, should talk with KAME guys first, for *BSD compatibility */ #define IPV6_RTHDR_LOOSE 0 /* this hop need not be a neighbor. XXX old spec */ #define IPV6_RTHDR_STRICT 1 /* this hop must be a neighbor. XXX old spec */ #define IPV6_RTHDR_TYPE_0 0 /* IPv6 routing header type 0 */ /* * Defaults and limits for options */ #define IPV6_DEFAULT_MULTICAST_HOPS 1 /* normally limit m'casts to 1 hop */ #define IPV6_DEFAULT_MULTICAST_LOOP 1 /* normally hear sends if a member */ /* * Limit for IPv6 multicast memberships */ #define IPV6_MAX_MEMBERSHIPS 4095 /* * Default resource limits for IPv6 multicast source filtering. * These may be modified by sysctl. */ #define IPV6_MAX_GROUP_SRC_FILTER 512 /* sources per group */ #define IPV6_MAX_SOCK_SRC_FILTER 128 /* sources per socket/group */ /* * Argument structure for IPV6_JOIN_GROUP and IPV6_LEAVE_GROUP. */ struct ipv6_mreq { struct in6_addr ipv6mr_multiaddr; unsigned int ipv6mr_interface; }; /* * IPV6_PKTINFO: Packet information(RFC2292 sec 5) */ struct in6_pktinfo { struct in6_addr ipi6_addr; /* src/dst IPv6 address */ unsigned int ipi6_ifindex; /* send/recv interface index */ }; /* * Control structure for IPV6_RECVPATHMTU socket option. */ struct ip6_mtuinfo { struct sockaddr_in6 ip6m_addr; /* or sockaddr_storage? */ uint32_t ip6m_mtu; }; /* * Argument for IPV6_PORTRANGE: * - which range to search when port is unspecified at bind() or connect() */ #define IPV6_PORTRANGE_DEFAULT 0 /* default range */ #define IPV6_PORTRANGE_HIGH 1 /* "high" - request firewall bypass */ #define IPV6_PORTRANGE_LOW 2 /* "low" - vouchsafe security */ #if __BSD_VISIBLE /* * Definitions for inet6 sysctl operations. * * Third level is protocol number. * Fourth level is desired variable within that protocol. */ #define IPV6PROTO_MAXID (IPPROTO_PIM + 1) /* don't list to IPV6PROTO_MAX */ /* * Names for IP sysctl objects */ #define IPV6CTL_FORWARDING 1 /* act as router */ #define IPV6CTL_SENDREDIRECTS 2 /* may send redirects when forwarding*/ #define IPV6CTL_DEFHLIM 3 /* default Hop-Limit */ #ifdef notyet #define IPV6CTL_DEFMTU 4 /* default MTU */ #endif #define IPV6CTL_FORWSRCRT 5 /* forward source-routed dgrams */ #define IPV6CTL_STATS 6 /* stats */ #define IPV6CTL_MRTSTATS 7 /* multicast forwarding stats */ #define IPV6CTL_MRTPROTO 8 /* multicast routing protocol */ #define IPV6CTL_MAXFRAGPACKETS 9 /* max packets reassembly queue */ #define IPV6CTL_SOURCECHECK 10 /* verify source route and intf */ #define IPV6CTL_SOURCECHECK_LOGINT 11 /* minimume logging interval */ #define IPV6CTL_ACCEPT_RTADV 12 /* 13; unused; was: IPV6CTL_KEEPFAITH */ #define IPV6CTL_LOG_INTERVAL 14 #define IPV6CTL_HDRNESTLIMIT 15 #define IPV6CTL_DAD_COUNT 16 #define IPV6CTL_AUTO_FLOWLABEL 17 #define IPV6CTL_DEFMCASTHLIM 18 #define IPV6CTL_GIF_HLIM 19 /* default HLIM for gif encap packet */ #define IPV6CTL_KAME_VERSION 20 #define IPV6CTL_USE_DEPRECATED 21 /* use deprecated addr (RFC2462 5.5.4) */ #define IPV6CTL_RR_PRUNE 22 /* walk timer for router renumbering */ #if 0 /* obsolete */ #define IPV6CTL_MAPPED_ADDR 23 #endif #define IPV6CTL_V6ONLY 24 /* IPV6CTL_RTEXPIRE 25 deprecated */ /* IPV6CTL_RTMINEXPIRE 26 deprecated */ /* IPV6CTL_RTMAXCACHE 27 deprecated */ #define IPV6CTL_USETEMPADDR 32 /* use temporary addresses (RFC3041) */ #define IPV6CTL_TEMPPLTIME 33 /* preferred lifetime for tmpaddrs */ #define IPV6CTL_TEMPVLTIME 34 /* valid lifetime for tmpaddrs */ #define IPV6CTL_AUTO_LINKLOCAL 35 /* automatic link-local addr assign */ #define IPV6CTL_RIP6STATS 36 /* raw_ip6 stats */ #define IPV6CTL_PREFER_TEMPADDR 37 /* prefer temporary addr as src */ #define IPV6CTL_ADDRCTLPOLICY 38 /* get/set address selection policy */ #define IPV6CTL_USE_DEFAULTZONE 39 /* use default scope zone */ #define IPV6CTL_MAXFRAGS 41 /* max fragments */ #if 0 #define IPV6CTL_IFQ 42 /* ip6intrq node */ #define IPV6CTL_ISATAPRTR 43 /* isatap router */ #endif #define IPV6CTL_MCAST_PMTU 44 /* enable pMTU discovery for multicast? */ /* New entries should be added here from current IPV6CTL_MAXID value. */ /* to define items, should talk with KAME guys first, for *BSD compatibility */ #define IPV6CTL_STEALTH 45 #define ICMPV6CTL_ND6_ONLINKNSRFC4861 47 #define IPV6CTL_NO_RADR 48 /* No defroute from RA */ #define IPV6CTL_NORBIT_RAIF 49 /* Disable R-bit in NA on RA * receiving IF. */ #define IPV6CTL_RFC6204W3 50 /* Accept defroute even when forwarding enabled */ #define IPV6CTL_INTRQMAXLEN 51 /* max length of IPv6 netisr queue */ #define IPV6CTL_INTRDQMAXLEN 52 /* max length of direct IPv6 netisr * queue */ #define IPV6CTL_MAXFRAGSPERPACKET 53 /* Max fragments per packet */ #define IPV6CTL_MAXFRAGBUCKETSIZE 54 /* Max reassembly queues per bucket */ #define IPV6CTL_MAXID 55 #endif /* __BSD_VISIBLE */ /* * Since both netinet/ and netinet6/ call into netipsec/ and netpfil/, * the protocol specific mbuf flags are shared between them. */ #define M_FASTFWD_OURS M_PROTO1 /* changed dst to local */ #define M_IP6_NEXTHOP M_PROTO2 /* explicit ip nexthop */ #define M_IP_NEXTHOP M_PROTO2 /* explicit ip nexthop */ #define M_SKIP_FIREWALL M_PROTO3 /* skip firewall processing */ #define M_AUTHIPHDR M_PROTO4 #define M_DECRYPTED M_PROTO5 #define M_LOOP M_PROTO6 #define M_AUTHIPDGM M_PROTO7 #define M_RTALERT_MLD M_PROTO8 #define M_FRAGMENTED M_PROTO9 /* contained fragment header */ #ifdef _KERNEL struct cmsghdr; struct ip6_hdr; int in6_cksum_pseudo(struct ip6_hdr *, uint32_t, uint8_t, uint16_t); int in6_cksum(struct mbuf *, u_int8_t, u_int32_t, u_int32_t); int in6_cksum_partial(struct mbuf *, u_int8_t, u_int32_t, u_int32_t, u_int32_t); int in6_localaddr(struct in6_addr *); int in6_localip(struct in6_addr *); int in6_ifhasaddr(struct ifnet *, struct in6_addr *); int in6_addrscope(const struct in6_addr *); char *ip6_sprintf(char *, const struct in6_addr *); struct in6_ifaddr *in6_ifawithifp(struct ifnet *, struct in6_addr *); extern void in6_if_up(struct ifnet *); struct sockaddr; extern u_char ip6_protox[]; void in6_sin6_2_sin(struct sockaddr_in *sin, struct sockaddr_in6 *sin6); void in6_sin_2_v4mapsin6(struct sockaddr_in *sin, struct sockaddr_in6 *sin6); void in6_sin6_2_sin_in_sock(struct sockaddr *nam); void in6_sin_2_v4mapsin6_in_sock(struct sockaddr **nam); extern void addrsel_policy_init(void); #define satosin6(sa) ((struct sockaddr_in6 *)(sa)) #define sin6tosa(sin6) ((struct sockaddr *)(sin6)) #define ifatoia6(ifa) ((struct in6_ifaddr *)(ifa)) #endif /* _KERNEL */ #ifndef _SIZE_T_DECLARED typedef __size_t size_t; #define _SIZE_T_DECLARED #endif #ifndef _SOCKLEN_T_DECLARED typedef __socklen_t socklen_t; #define _SOCKLEN_T_DECLARED #endif #if __BSD_VISIBLE __BEGIN_DECLS struct cmsghdr; extern int inet6_option_space(int); extern int inet6_option_init(void *, struct cmsghdr **, int); extern int inet6_option_append(struct cmsghdr *, const uint8_t *, int, int); extern uint8_t *inet6_option_alloc(struct cmsghdr *, int, int, int); extern int inet6_option_next(const struct cmsghdr *, uint8_t **); extern int inet6_option_find(const struct cmsghdr *, uint8_t **, int); extern size_t inet6_rthdr_space(int, int); extern struct cmsghdr *inet6_rthdr_init(void *, int); extern int inet6_rthdr_add(struct cmsghdr *, const struct in6_addr *, unsigned int); extern int inet6_rthdr_lasthop(struct cmsghdr *, unsigned int); #if 0 /* not implemented yet */ extern int inet6_rthdr_reverse(const struct cmsghdr *, struct cmsghdr *); #endif extern int inet6_rthdr_segments(const struct cmsghdr *); extern struct in6_addr *inet6_rthdr_getaddr(struct cmsghdr *, int); extern int inet6_rthdr_getflags(const struct cmsghdr *, int); extern int inet6_opt_init(void *, socklen_t); extern int inet6_opt_append(void *, socklen_t, int, uint8_t, socklen_t, uint8_t, void **); extern int inet6_opt_finish(void *, socklen_t, int); extern int inet6_opt_set_val(void *, int, void *, socklen_t); extern int inet6_opt_next(void *, socklen_t, int, uint8_t *, socklen_t *, void **); extern int inet6_opt_find(void *, socklen_t, int, uint8_t, socklen_t *, void **); extern int inet6_opt_get_val(void *, int, void *, socklen_t); extern socklen_t inet6_rth_space(int, int); extern void *inet6_rth_init(void *, socklen_t, int, int); extern int inet6_rth_add(void *, const struct in6_addr *); extern int inet6_rth_reverse(const void *, void *); extern int inet6_rth_segments(const void *); extern struct in6_addr *inet6_rth_getaddr(const void *, int); __END_DECLS #endif /* __BSD_VISIBLE */ #endif /* !_NETINET6_IN6_H_ */ Index: head/sys/netinet6/in6_pcb.c =================================================================== --- head/sys/netinet6/in6_pcb.c (revision 360291) +++ head/sys/netinet6/in6_pcb.c (revision 360292) @@ -1,1380 +1,1382 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: in6_pcb.c,v 1.31 2001/05/21 05:45:10 jinmei Exp $ */ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)in_pcb.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_pcbgroup.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include +#include #include static struct inpcb *in6_pcblookup_hash_locked(struct inpcbinfo *, struct in6_addr *, u_int, struct in6_addr *, u_int, int, struct ifnet *); int in6_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { struct socket *so = inp->inp_socket; struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)NULL; struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; u_short lport = 0; int error, lookupflags = 0; int reuseport = (so->so_options & SO_REUSEPORT); /* * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here * so that we don't have to add to the (already messy) code below. */ int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); if (CK_STAILQ_EMPTY(&V_in6_ifaddrhead)) /* XXX broken! */ return (EADDRNOTAVAIL); if (inp->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) return (EINVAL); if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) lookupflags = INPLOOKUP_WILDCARD; if (nam == NULL) { if ((error = prison_local_ip6(cred, &inp->in6p_laddr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0) return (error); } else { sin6 = (struct sockaddr_in6 *)nam; if (nam->sa_len != sizeof(*sin6)) return (EINVAL); /* * family check. */ if (nam->sa_family != AF_INET6) return (EAFNOSUPPORT); if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) return(error); if ((error = prison_local_ip6(cred, &sin6->sin6_addr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0) return (error); lport = sin6->sin6_port; if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { /* * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; * allow compepte duplication of binding if * SO_REUSEPORT is set, or if SO_REUSEADDR is set * and a multicast address is bound on both * new and duplicated sockets. */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) reuseport = SO_REUSEADDR|SO_REUSEPORT; /* * XXX: How to deal with SO_REUSEPORT_LB here? * Treat same as SO_REUSEPORT for now. */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; } else if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { struct epoch_tracker et; struct ifaddr *ifa; sin6->sin6_port = 0; /* yech... */ NET_EPOCH_ENTER(et); if ((ifa = ifa_ifwithaddr((struct sockaddr *)sin6)) == NULL && (inp->inp_flags & INP_BINDANY) == 0) { NET_EPOCH_EXIT(et); return (EADDRNOTAVAIL); } /* * XXX: bind to an anycast address might accidentally * cause sending a packet with anycast source address. * We should allow to bind to a deprecated address, since * the application dares to use it. */ if (ifa != NULL && ((struct in6_ifaddr *)ifa)->ia6_flags & (IN6_IFF_ANYCAST|IN6_IFF_NOTREADY|IN6_IFF_DETACHED)) { NET_EPOCH_EXIT(et); return (EADDRNOTAVAIL); } NET_EPOCH_EXIT(et); } if (lport) { struct inpcb *t; struct tcptw *tw; /* GROSS */ if (ntohs(lport) <= V_ipport_reservedhigh && ntohs(lport) >= V_ipport_reservedlow && priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) return (EACCES); if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr) && priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { t = in6_pcblookup_local(pcbinfo, &sin6->sin6_addr, lport, INPLOOKUP_WILDCARD, cred); if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && ((t->inp_flags & INP_TIMEWAIT) == 0) && (so->so_type != SOCK_STREAM || IN6_IS_ADDR_UNSPECIFIED(&t->in6p_faddr)) && (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) || !IN6_IS_ADDR_UNSPECIFIED(&t->in6p_laddr) || (t->inp_flags2 & INP_REUSEPORT) || (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && (inp->inp_cred->cr_uid != t->inp_cred->cr_uid)) return (EADDRINUSE); /* * If the socket is a BINDMULTI socket, then * the credentials need to match and the * original socket also has to have been bound * with BINDMULTI. */ if (t && (! in_pcbbind_check_bindmulti(inp, t))) return (EADDRINUSE); #ifdef INET if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 && IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { struct sockaddr_in sin; in6_sin6_2_sin(&sin, sin6); t = in_pcblookup_local(pcbinfo, sin.sin_addr, lport, INPLOOKUP_WILDCARD, cred); if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && ((t->inp_flags & INP_TIMEWAIT) == 0) && (so->so_type != SOCK_STREAM || ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && (inp->inp_cred->cr_uid != t->inp_cred->cr_uid)) return (EADDRINUSE); if (t && (! in_pcbbind_check_bindmulti(inp, t))) return (EADDRINUSE); } #endif } t = in6_pcblookup_local(pcbinfo, &sin6->sin6_addr, lport, lookupflags, cred); if (t && (t->inp_flags & INP_TIMEWAIT)) { /* * XXXRW: If an incpb has had its timewait * state recycled, we treat the address as * being in use (for now). This is better * than a panic, but not desirable. */ tw = intotw(t); if (tw == NULL || ((reuseport & tw->tw_so_options) == 0 && (reuseport_lb & tw->tw_so_options) == 0)) return (EADDRINUSE); } else if (t && (reuseport & inp_so_options(t)) == 0 && (reuseport_lb & inp_so_options(t)) == 0) { return (EADDRINUSE); } #ifdef INET if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 && IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { struct sockaddr_in sin; in6_sin6_2_sin(&sin, sin6); t = in_pcblookup_local(pcbinfo, sin.sin_addr, lport, lookupflags, cred); if (t && t->inp_flags & INP_TIMEWAIT) { tw = intotw(t); if (tw == NULL) return (EADDRINUSE); if ((reuseport & tw->tw_so_options) == 0 && (reuseport_lb & tw->tw_so_options) == 0 && (ntohl(t->inp_laddr.s_addr) != INADDR_ANY || ((inp->inp_vflag & INP_IPV6PROTO) == (t->inp_vflag & INP_IPV6PROTO)))) return (EADDRINUSE); } else if (t && (reuseport & inp_so_options(t)) == 0 && (reuseport_lb & inp_so_options(t)) == 0 && (ntohl(t->inp_laddr.s_addr) != INADDR_ANY || (t->inp_vflag & INP_IPV6PROTO) != 0)) { return (EADDRINUSE); } } #endif } inp->in6p_laddr = sin6->sin6_addr; } if (lport == 0) { if ((error = in6_pcbsetport(&inp->in6p_laddr, inp, cred)) != 0) { /* Undo an address bind that may have occurred. */ inp->in6p_laddr = in6addr_any; return (error); } } else { inp->inp_lport = lport; if (in_pcbinshash(inp) != 0) { inp->in6p_laddr = in6addr_any; inp->inp_lport = 0; return (EAGAIN); } } return (0); } /* * Transform old in6_pcbconnect() into an inner subroutine for new * in6_pcbconnect(): Do some validity-checking on the remote * address (in mbuf 'nam') and then determine local host address * (i.e., which interface) to use to access that remote host. * * This preserves definition of in6_pcbconnect(), while supporting a * slightly different version for T/TCP. (This is more than * a bit of a kludge, but cleaning up the internal interfaces would * have forced minor changes in every protocol). */ static int in6_pcbladdr(struct inpcb *inp, struct sockaddr *nam, struct in6_addr *plocal_addr6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; int error = 0; int scope_ambiguous = 0; struct in6_addr in6a; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); /* XXXRW: why? */ if (nam->sa_len != sizeof (*sin6)) return (EINVAL); if (sin6->sin6_family != AF_INET6) return (EAFNOSUPPORT); if (sin6->sin6_port == 0) return (EADDRNOTAVAIL); if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone) scope_ambiguous = 1; if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) return(error); if (!CK_STAILQ_EMPTY(&V_in6_ifaddrhead)) { /* * If the destination address is UNSPECIFIED addr, * use the loopback addr, e.g ::1. */ if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) sin6->sin6_addr = in6addr_loopback; } if ((error = prison_remote_ip6(inp->inp_cred, &sin6->sin6_addr)) != 0) return (error); error = in6_selectsrc_socket(sin6, inp->in6p_outputopts, inp, inp->inp_cred, scope_ambiguous, &in6a, NULL); if (error) return (error); /* * Do not update this earlier, in case we return with an error. * * XXX: this in6_selectsrc_socket result might replace the bound local * address with the address specified by setsockopt(IPV6_PKTINFO). * Is it the intended behavior? */ *plocal_addr6 = in6a; /* * Don't do pcblookup call here; return interface in * plocal_addr6 * and exit to caller, that will do the lookup. */ return (0); } /* * Outer subroutine: * Connect from a socket to a specified address. * Both address and port must be specified in argument sin. * If don't have a local address for this socket yet, * then pick one. */ int in6_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred, struct mbuf *m, bool rehash) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; struct in6_addr addr6; int error; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); /* * Call inner routine, to assign local interface address. * in6_pcbladdr() may automatically fill in sin6_scope_id. */ if ((error = in6_pcbladdr(inp, nam, &addr6)) != 0) return (error); if (in6_pcblookup_hash_locked(pcbinfo, &sin6->sin6_addr, sin6->sin6_port, IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ? &addr6 : &inp->in6p_laddr, inp->inp_lport, 0, NULL) != NULL) { return (EADDRINUSE); } if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (inp->inp_lport == 0) { KASSERT(rehash == true, ("Rehashing required for unbound inps")); error = in6_pcbbind(inp, (struct sockaddr *)0, cred); if (error) return (error); } inp->in6p_laddr = addr6; } inp->in6p_faddr = sin6->sin6_addr; inp->inp_fport = sin6->sin6_port; /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; if (inp->inp_flags & IN6P_AUTOFLOWLABEL) inp->inp_flow |= (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); if (rehash) { in_pcbrehash_mbuf(inp, m); } else { in_pcbinshash_mbuf(inp, m); } return (0); } int in6_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { return (in6_pcbconnect_mbuf(inp, nam, cred, NULL, true)); } void in6_pcbdisconnect(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); bzero((caddr_t)&inp->in6p_faddr, sizeof(inp->in6p_faddr)); inp->inp_fport = 0; /* clear flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; in_pcbrehash(inp); } struct sockaddr * in6_sockaddr(in_port_t port, struct in6_addr *addr_p) { struct sockaddr_in6 *sin6; sin6 = malloc(sizeof *sin6, M_SONAME, M_WAITOK); bzero(sin6, sizeof *sin6); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(*sin6); sin6->sin6_port = port; sin6->sin6_addr = *addr_p; (void)sa6_recoverscope(sin6); /* XXX: should catch errors */ return (struct sockaddr *)sin6; } struct sockaddr * in6_v4mapsin6_sockaddr(in_port_t port, struct in_addr *addr_p) { struct sockaddr_in sin; struct sockaddr_in6 *sin6_p; bzero(&sin, sizeof sin); sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); sin.sin_port = port; sin.sin_addr = *addr_p; sin6_p = malloc(sizeof *sin6_p, M_SONAME, M_WAITOK); in6_sin_2_v4mapsin6(&sin, sin6_p); return (struct sockaddr *)sin6_p; } int in6_getsockaddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in6_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in6_getsockaddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_lport; addr = inp->in6p_laddr; INP_RUNLOCK(inp); *nam = in6_sockaddr(port, &addr); return 0; } int in6_getpeeraddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in6_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in6_getpeeraddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_fport; addr = inp->in6p_faddr; INP_RUNLOCK(inp); *nam = in6_sockaddr(port, &addr); return 0; } int in6_mapped_sockaddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; int error; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in6_mapped_sockaddr: inp == NULL")); #ifdef INET if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) { error = in_getsockaddr(so, nam); if (error == 0) in6_sin_2_v4mapsin6_in_sock(nam); } else #endif { /* scope issues will be handled in in6_getsockaddr(). */ error = in6_getsockaddr(so, nam); } return error; } int in6_mapped_peeraddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; int error; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in6_mapped_peeraddr: inp == NULL")); #ifdef INET if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) { error = in_getpeeraddr(so, nam); if (error == 0) in6_sin_2_v4mapsin6_in_sock(nam); } else #endif /* scope issues will be handled in in6_getpeeraddr(). */ error = in6_getpeeraddr(so, nam); return error; } /* * Pass some notification to all connections of a protocol * associated with address dst. The local address and/or port numbers * may be specified to limit the search. The "usual action" will be * taken, depending on the ctlinput cmd. The caller must filter any * cmds that are uninteresting (e.g., no error in the map). * Call the protocol specific routine (if any) to report * any errors for each matching socket. */ void in6_pcbnotify(struct inpcbinfo *pcbinfo, struct sockaddr *dst, u_int fport_arg, const struct sockaddr *src, u_int lport_arg, int cmd, void *cmdarg, struct inpcb *(*notify)(struct inpcb *, int)) { struct inpcb *inp, *inp_temp; struct sockaddr_in6 sa6_src, *sa6_dst; u_short fport = fport_arg, lport = lport_arg; u_int32_t flowinfo; int errno; if ((unsigned)cmd >= PRC_NCMDS || dst->sa_family != AF_INET6) return; sa6_dst = (struct sockaddr_in6 *)dst; if (IN6_IS_ADDR_UNSPECIFIED(&sa6_dst->sin6_addr)) return; /* * note that src can be NULL when we get notify by local fragmentation. */ sa6_src = (src == NULL) ? sa6_any : *(const struct sockaddr_in6 *)src; flowinfo = sa6_src.sin6_flowinfo; /* * Redirects go to all references to the destination, * and use in6_rtchange to invalidate the route cache. * Dead host indications: also use in6_rtchange to invalidate * the cache, and deliver the error to all the sockets. * Otherwise, if we have knowledge of the local port and address, * deliver only to that socket. */ if (PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) { fport = 0; lport = 0; bzero((caddr_t)&sa6_src.sin6_addr, sizeof(sa6_src.sin6_addr)); if (cmd != PRC_HOSTDEAD) notify = in6_rtchange; } errno = inet6ctlerrmap[cmd]; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { INP_WLOCK(inp); if ((inp->inp_vflag & INP_IPV6) == 0) { INP_WUNLOCK(inp); continue; } /* * If the error designates a new path MTU for a destination * and the application (associated with this socket) wanted to * know the value, notify. * XXX: should we avoid to notify the value to TCP sockets? */ if (cmd == PRC_MSGSIZE && cmdarg != NULL) ip6_notify_pmtu(inp, (struct sockaddr_in6 *)dst, *(u_int32_t *)cmdarg); /* * Detect if we should notify the error. If no source and * destination ports are specifed, but non-zero flowinfo and * local address match, notify the error. This is the case * when the error is delivered with an encrypted buffer * by ESP. Otherwise, just compare addresses and ports * as usual. */ if (lport == 0 && fport == 0 && flowinfo && inp->inp_socket != NULL && flowinfo == (inp->inp_flow & IPV6_FLOWLABEL_MASK) && IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, &sa6_src.sin6_addr)) goto do_notify; else if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, &sa6_dst->sin6_addr) || inp->inp_socket == 0 || (lport && inp->inp_lport != lport) || (!IN6_IS_ADDR_UNSPECIFIED(&sa6_src.sin6_addr) && !IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, &sa6_src.sin6_addr)) || (fport && inp->inp_fport != fport)) { INP_WUNLOCK(inp); continue; } do_notify: if (notify) { if ((*notify)(inp, errno)) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); } /* * Lookup a PCB based on the local address and port. Caller must hold the * hash lock. No inpcb locks or references are acquired. */ struct inpcb * in6_pcblookup_local(struct inpcbinfo *pcbinfo, struct in6_addr *laddr, u_short lport, int lookupflags, struct ucred *cred) { struct inpcb *inp; int matchwild = 3, wildcard; KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); INP_HASH_LOCK_ASSERT(pcbinfo); if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { struct inpcbhead *head; /* * Look for an unconnected (wildcard foreign addr) PCB that * matches the local address and port we're looking for. */ head = &pcbinfo->ipi_hashbase[INP_PCBHASH( INP6_PCBHASHKEY(&in6addr_any), lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) && IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) && inp->inp_lport == lport) { /* Found. */ if (cred == NULL || prison_equal_ip6(cred->cr_prison, inp->inp_cred->cr_prison)) return (inp); } } /* * Not found. */ return (NULL); } else { struct inpcbporthead *porthash; struct inpcbport *phd; struct inpcb *match = NULL; /* * Best fit PCB lookup. * * First see if this local port is in use by looking on the * port hash list. */ porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, pcbinfo->ipi_porthashmask)]; CK_LIST_FOREACH(phd, porthash, phd_hash) { if (phd->phd_port == lport) break; } if (phd != NULL) { /* * Port is in use by one or more PCBs. Look for best * fit. */ CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { wildcard = 0; if (cred != NULL && !prison_equal_ip6(cred->cr_prison, inp->inp_cred->cr_prison)) continue; /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) wildcard++; if (!IN6_IS_ADDR_UNSPECIFIED( &inp->in6p_laddr)) { if (IN6_IS_ADDR_UNSPECIFIED(laddr)) wildcard++; else if (!IN6_ARE_ADDR_EQUAL( &inp->in6p_laddr, laddr)) continue; } else { if (!IN6_IS_ADDR_UNSPECIFIED(laddr)) wildcard++; } if (wildcard < matchwild) { match = inp; matchwild = wildcard; if (matchwild == 0) break; } } } return (match); } } void in6_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) { struct inpcb *inp; struct in6_multi *inm; struct in6_mfilter *imf; struct ip6_moptions *im6o; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { INP_WLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_WUNLOCK(inp); continue; } im6o = inp->in6p_moptions; if ((inp->inp_vflag & INP_IPV6) && im6o != NULL) { /* * Unselect the outgoing ifp for multicast if it * is being detached. */ if (im6o->im6o_multicast_ifp == ifp) im6o->im6o_multicast_ifp = NULL; /* * Drop multicast group membership if we joined * through the interface being detached. */ restart: IP6_MFILTER_FOREACH(imf, &im6o->im6o_head) { if ((inm = imf->im6f_in6m) == NULL) continue; if (inm->in6m_ifp != ifp) continue; ip6_mfilter_remove(&im6o->im6o_head, imf); IN6_MULTI_LOCK_ASSERT(); in6_leavegroup_locked(inm, NULL); ip6_mfilter_free(imf); goto restart; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); } /* * Check for alternatives when higher level complains * about service problems. For now, invalidate cached * routing information. If the route was created dynamically * (by a redirect), time to try a default gateway again. */ void in6_losing(struct inpcb *inp) { RO_INVALIDATE_CACHE(&inp->inp_route6); } /* * After a routing change, flush old routing * and allocate a (hopefully) better one. */ struct inpcb * in6_rtchange(struct inpcb *inp, int errno __unused) { RO_INVALIDATE_CACHE(&inp->inp_route6); return inp; } static struct inpcb * in6_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, const struct in6_addr *laddr, uint16_t lport, const struct in6_addr *faddr, uint16_t fport, int lookupflags) { struct inpcb *local_wild; const struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; uint32_t idx; INP_HASH_LOCK_ASSERT(pcbinfo); hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; /* * Order of socket selection: * 1. non-wild. * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). * * NOTE: * - Load balanced group does not contain jailed sockets. * - Load balanced does not contain IPv4 mapped INET6 wild sockets. */ local_wild = NULL; CK_LIST_FOREACH(grp, hdr, il_list) { #ifdef INET if (!(grp->il_vflag & INP_IPV6)) continue; #endif if (grp->il_lport != lport) continue; idx = INP_PCBLBGROUP_PKTHASH(INP6_PCBHASHKEY(faddr), lport, fport) % grp->il_inpcnt; if (IN6_ARE_ADDR_EQUAL(&grp->il6_laddr, laddr)) return (grp->il_inp[idx]); if (IN6_IS_ADDR_UNSPECIFIED(&grp->il6_laddr) && (lookupflags & INPLOOKUP_WILDCARD) != 0) local_wild = grp->il_inp[idx]; } return (local_wild); } #ifdef PCBGROUP /* * Lookup PCB in hash list, using pcbgroup tables. */ static struct inpcb * in6_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, struct in6_addr *faddr, u_int fport_arg, struct in6_addr *laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp) { struct inpcbhead *head; struct inpcb *inp, *tmpinp; u_short fport = fport_arg, lport = lport_arg; bool locked; /* * First look for an exact match. */ tmpinp = NULL; INP_GROUP_LOCK(pcbgroup); head = &pcbgroup->ipg_hashbase[INP_PCBHASH( INP6_PCBHASHKEY(faddr), lport, fport, pcbgroup->ipg_hashmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, faddr) && IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) && inp->inp_fport == fport && inp->inp_lport == lport) { /* * XXX We should be able to directly return * the inp here, without any checks. * Well unless both bound with SO_REUSEPORT? */ if (prison_flag(inp->inp_cred, PR_IP6)) goto found; if (tmpinp == NULL) tmpinp = inp; } } if (tmpinp != NULL) { inp = tmpinp; goto found; } /* * Then look for a wildcard match in the pcbgroup. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; struct inpcb *jail_wild = NULL; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbgroup->ipg_hashbase[ INP_PCBHASH(INADDR_ANY, lport, 0, pcbgroup->ipg_hashmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) || inp->inp_lport != lport) { continue; } injail = prison_flag(inp->inp_cred, PR_IP6); if (injail) { if (prison_check_ip6(inp->inp_cred, laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) { if (injail) goto found; else local_exact = inp; } else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; if (inp != NULL) goto found; } /* * Then look for a wildcard match, if requested. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; struct inpcb *jail_wild = NULL; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbinfo->ipi_wildbase[INP_PCBHASH( INP6_PCBHASHKEY(&in6addr_any), lport, 0, pcbinfo->ipi_wildmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) || inp->inp_lport != lport) { continue; } injail = prison_flag(inp->inp_cred, PR_IP6); if (injail) { if (prison_check_ip6(inp->inp_cred, laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) { if (injail) goto found; else local_exact = inp; } else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; if (inp != NULL) goto found; } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ INP_GROUP_UNLOCK(pcbgroup); return (NULL); found: if (lookupflags & INPLOOKUP_WLOCKPCB) locked = INP_TRY_WLOCK(inp); else if (lookupflags & INPLOOKUP_RLOCKPCB) locked = INP_TRY_RLOCK(inp); else panic("%s: locking buf", __func__); if (!locked) in_pcbref(inp); INP_GROUP_UNLOCK(pcbgroup); if (!locked) { if (lookupflags & INPLOOKUP_WLOCKPCB) { INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) return (NULL); } else { INP_RLOCK(inp); if (in_pcbrele_rlocked(inp)) return (NULL); } } #ifdef INVARIANTS if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WLOCK_ASSERT(inp); else INP_RLOCK_ASSERT(inp); #endif return (inp); } #endif /* PCBGROUP */ /* * Lookup PCB in hash list. */ static struct inpcb * in6_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport_arg, struct in6_addr *laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp) { struct inpcbhead *head; struct inpcb *inp, *tmpinp; u_short fport = fport_arg, lport = lport_arg; KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); INP_HASH_LOCK_ASSERT(pcbinfo); /* * First look for an exact match. */ tmpinp = NULL; head = &pcbinfo->ipi_hashbase[INP_PCBHASH( INP6_PCBHASHKEY(faddr), lport, fport, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, faddr) && IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) && inp->inp_fport == fport && inp->inp_lport == lport) { /* * XXX We should be able to directly return * the inp here, without any checks. * Well unless both bound with SO_REUSEPORT? */ if (prison_flag(inp->inp_cred, PR_IP6)) return (inp); if (tmpinp == NULL) tmpinp = inp; } } if (tmpinp != NULL) return (tmpinp); /* * Then look in lb group (for wildcard match). */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { inp = in6_pcblookup_lbgroup(pcbinfo, laddr, lport, faddr, fport, lookupflags); if (inp != NULL) return (inp); } /* * Then look for a wildcard match, if requested. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; struct inpcb *jail_wild = NULL; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbinfo->ipi_hashbase[INP_PCBHASH( INP6_PCBHASHKEY(&in6addr_any), lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV6) == 0) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) || inp->inp_lport != lport) { continue; } injail = prison_flag(inp->inp_cred, PR_IP6); if (injail) { if (prison_check_ip6(inp->inp_cred, laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) { if (injail) return (inp); else local_exact = inp; } else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ if (jail_wild != NULL) return (jail_wild); if (local_exact != NULL) return (local_exact); if (local_wild != NULL) return (local_wild); } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ /* * Not found. */ return (NULL); } /* * Lookup PCB in hash list, using pcbinfo tables. This variation locks the * hash list lock, and will return the inpcb locked (i.e., requires * INPLOOKUP_LOCKPCB). */ static struct inpcb * in6_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport, struct in6_addr *laddr, u_int lport, int lookupflags, struct ifnet *ifp) { struct inpcb *inp; inp = in6_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); if (inp != NULL) { if (lookupflags & INPLOOKUP_WLOCKPCB) { INP_WLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_WUNLOCK(inp); inp = NULL; } } else if (lookupflags & INPLOOKUP_RLOCKPCB) { INP_RLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_RUNLOCK(inp); inp = NULL; } } else panic("%s: locking bug", __func__); #ifdef INVARIANTS if (inp != NULL) { if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WLOCK_ASSERT(inp); else INP_RLOCK_ASSERT(inp); } #endif } return (inp); } /* * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf * from which a pre-calculated hash value may be extracted. * * Possibly more of this logic should be in in6_pcbgroup.c. */ struct inpcb * in6_pcblookup(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport, struct in6_addr *laddr, u_int lport, int lookupflags, struct ifnet *ifp) { #if defined(PCBGROUP) && !defined(RSS) struct inpcbgroup *pcbgroup; #endif KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, ("%s: LOCKPCB not set", __func__)); /* * When not using RSS, use connection groups in preference to the * reservation table when looking up 4-tuples. When using RSS, just * use the reservation table, due to the cost of the Toeplitz hash * in software. * * XXXRW: This policy belongs in the pcbgroup code, as in principle * we could be doing RSS with a non-Toeplitz hash that is affordable * in software. */ #if defined(PCBGROUP) && !defined(RSS) if (in_pcbgroup_enabled(pcbinfo)) { pcbgroup = in6_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); } #endif return (in6_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } struct inpcb * in6_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport, struct in6_addr *laddr, u_int lport, int lookupflags, struct ifnet *ifp, struct mbuf *m) { #ifdef PCBGROUP struct inpcbgroup *pcbgroup; #endif KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, ("%s: LOCKPCB not set", __func__)); #ifdef PCBGROUP /* * If we can use a hardware-generated hash to look up the connection * group, use that connection group to find the inpcb. Otherwise * fall back on a software hash -- or the reservation table if we're * using RSS. * * XXXRW: As above, that policy belongs in the pcbgroup code. */ if (in_pcbgroup_enabled(pcbinfo) && M_HASHTYPE_TEST(m, M_HASHTYPE_NONE) == 0) { pcbgroup = in6_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), m->m_pkthdr.flowid); if (pcbgroup != NULL) return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #ifndef RSS pcbgroup = in6_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #endif } #endif return (in6_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } void init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m, int srcordst) { struct ip6_hdr *ip; ip = mtod(m, struct ip6_hdr *); bzero(sin6, sizeof(*sin6)); sin6->sin6_len = sizeof(*sin6); sin6->sin6_family = AF_INET6; sin6->sin6_addr = srcordst ? ip->ip6_dst : ip->ip6_src; (void)sa6_recoverscope(sin6); /* XXX: should catch errors... */ return; } Index: head/sys/netinet6/in6_src.c =================================================================== --- head/sys/netinet6/in6_src.c (revision 360291) +++ head/sys/netinet6/in6_src.c (revision 360292) @@ -1,1231 +1,1232 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: in6_src.c,v 1.132 2003/08/26 04:42:27 keiichi Exp $ */ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)in_pcb.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #ifdef RADIX_MPATH #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static struct mtx addrsel_lock; #define ADDRSEL_LOCK_INIT() mtx_init(&addrsel_lock, "addrsel_lock", NULL, MTX_DEF) #define ADDRSEL_LOCK() mtx_lock(&addrsel_lock) #define ADDRSEL_UNLOCK() mtx_unlock(&addrsel_lock) #define ADDRSEL_LOCK_ASSERT() mtx_assert(&addrsel_lock, MA_OWNED) static struct sx addrsel_sxlock; #define ADDRSEL_SXLOCK_INIT() sx_init(&addrsel_sxlock, "addrsel_sxlock") #define ADDRSEL_SLOCK() sx_slock(&addrsel_sxlock) #define ADDRSEL_SUNLOCK() sx_sunlock(&addrsel_sxlock) #define ADDRSEL_XLOCK() sx_xlock(&addrsel_sxlock) #define ADDRSEL_XUNLOCK() sx_xunlock(&addrsel_sxlock) #define ADDR_LABEL_NOTAPP (-1) VNET_DEFINE_STATIC(struct in6_addrpolicy, defaultaddrpolicy); #define V_defaultaddrpolicy VNET(defaultaddrpolicy) VNET_DEFINE(int, ip6_prefer_tempaddr) = 0; static int selectroute(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct route_in6 *, struct ifnet **, - struct rtentry **, int, u_int); + struct nhop_object **, int, u_int, uint32_t); static int in6_selectif(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct ifnet **, struct ifnet *, u_int); static int in6_selectsrc(uint32_t, struct sockaddr_in6 *, struct ip6_pktopts *, struct inpcb *, struct ucred *, struct ifnet **, struct in6_addr *); static struct in6_addrpolicy *lookup_addrsel_policy(struct sockaddr_in6 *); static void init_policy_queue(void); static int add_addrsel_policyent(struct in6_addrpolicy *); static int delete_addrsel_policyent(struct in6_addrpolicy *); static int walk_addrsel_policy(int (*)(struct in6_addrpolicy *, void *), void *); static int dump_addrsel_policyent(struct in6_addrpolicy *, void *); static struct in6_addrpolicy *match_addrsel_policy(struct sockaddr_in6 *); /* * Return an IPv6 address, which is the most appropriate for a given * destination and user specified options. * If necessary, this function lookups the routing table and returns * an entry to the caller for later use. */ #define REPLACE(r) do {\ IP6STAT_INC(ip6s_sources_rule[(r)]); \ /* { \ char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \ printf("in6_selectsrc: replace %s with %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \ } */ \ goto replace; \ } while(0) #define NEXT(r) do {\ /* { \ char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \ printf("in6_selectsrc: keep %s against %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \ } */ \ goto next; /* XXX: we can't use 'continue' here */ \ } while(0) #define BREAK(r) do { \ IP6STAT_INC(ip6s_sources_rule[(r)]); \ goto out; /* XXX: we can't use 'break' here */ \ } while(0) static int in6_selectsrc(uint32_t fibnum, struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred, struct ifnet **ifpp, struct in6_addr *srcp) { struct rm_priotracker in6_ifa_tracker; struct in6_addr dst, tmp; struct ifnet *ifp = NULL, *oifp = NULL; struct in6_ifaddr *ia = NULL, *ia_best = NULL; struct in6_pktinfo *pi = NULL; int dst_scope = -1, best_scope = -1, best_matchlen = -1; struct in6_addrpolicy *dst_policy = NULL, *best_policy = NULL; u_int32_t odstzone; int prefer_tempaddr; int error; struct ip6_moptions *mopts; KASSERT(srcp != NULL, ("%s: srcp is NULL", __func__)); dst = dstsock->sin6_addr; /* make a copy for local operation */ if (ifpp) { /* * Save a possibly passed in ifp for in6_selectsrc. Only * neighbor discovery code should use this feature, where * we may know the interface but not the FIB number holding * the connected subnet in case someone deleted it from the * default FIB and we need to check the interface. */ if (*ifpp != NULL) oifp = *ifpp; *ifpp = NULL; } if (inp != NULL) { INP_LOCK_ASSERT(inp); mopts = inp->in6p_moptions; } else { mopts = NULL; } /* * If the source address is explicitly specified by the caller, * check if the requested source address is indeed a unicast address * assigned to the node, and can be used as the packet's source * address. If everything is okay, use the address as source. */ if (opts && (pi = opts->ip6po_pktinfo) && !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) { /* get the outgoing interface */ if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp, fibnum)) != 0) return (error); /* * determine the appropriate zone id of the source based on * the zone of the destination and the outgoing interface. * If the specified address is ambiguous wrt the scope zone, * the interface must be specified; otherwise, ifa_ifwithaddr() * will fail matching the address. */ tmp = pi->ipi6_addr; if (ifp) { error = in6_setscope(&tmp, ifp, &odstzone); if (error) return (error); } if (cred != NULL && (error = prison_local_ip6(cred, &tmp, (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) return (error); /* * If IPV6_BINDANY socket option is set, we allow to specify * non local addresses as source address in IPV6_PKTINFO * ancillary data. */ if ((inp->inp_flags & INP_BINDANY) == 0) { ia = in6ifa_ifwithaddr(&tmp, 0 /* XXX */); if (ia == NULL || (ia->ia6_flags & (IN6_IFF_ANYCAST | IN6_IFF_NOTREADY))) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (EADDRNOTAVAIL); } bcopy(&ia->ia_addr.sin6_addr, srcp, sizeof(*srcp)); ifa_free(&ia->ia_ifa); } else bcopy(&tmp, srcp, sizeof(*srcp)); pi->ipi6_addr = tmp; /* XXX: this overrides pi */ if (ifpp) *ifpp = ifp; return (0); } /* * Otherwise, if the socket has already bound the source, just use it. */ if (inp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (cred != NULL && (error = prison_local_ip6(cred, &inp->in6p_laddr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0) return (error); bcopy(&inp->in6p_laddr, srcp, sizeof(*srcp)); return (0); } /* * Bypass source address selection and use the primary jail IP * if requested. */ if (cred != NULL && !prison_saddrsel_ip6(cred, srcp)) return (0); /* * If the address is not specified, choose the best one based on * the outgoing interface and the destination address. */ /* get the outgoing interface */ if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp, (inp != NULL) ? inp->inp_inc.inc_fibnum : fibnum)) != 0) return (error); #ifdef DIAGNOSTIC if (ifp == NULL) /* this should not happen */ panic("in6_selectsrc: NULL ifp"); #endif error = in6_setscope(&dst, ifp, &odstzone); if (error) return (error); IN6_IFADDR_RLOCK(&in6_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { int new_scope = -1, new_matchlen = -1; struct in6_addrpolicy *new_policy = NULL; u_int32_t srczone, osrczone, dstzone; struct in6_addr src; struct ifnet *ifp1 = ia->ia_ifp; /* * We'll never take an address that breaks the scope zone * of the destination. We also skip an address if its zone * does not contain the outgoing interface. * XXX: we should probably use sin6_scope_id here. */ if (in6_setscope(&dst, ifp1, &dstzone) || odstzone != dstzone) { continue; } src = ia->ia_addr.sin6_addr; if (in6_setscope(&src, ifp, &osrczone) || in6_setscope(&src, ifp1, &srczone) || osrczone != srczone) { continue; } /* avoid unusable addresses */ if ((ia->ia6_flags & (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) { continue; } if (!V_ip6_use_deprecated && IFA6_IS_DEPRECATED(ia)) continue; /* If jailed only take addresses of the jail into account. */ if (cred != NULL && prison_check_ip6(cred, &ia->ia_addr.sin6_addr) != 0) continue; /* Rule 1: Prefer same address */ if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) { ia_best = ia; BREAK(1); /* there should be no better candidate */ } if (ia_best == NULL) REPLACE(0); /* Rule 2: Prefer appropriate scope */ if (dst_scope < 0) dst_scope = in6_addrscope(&dst); new_scope = in6_addrscope(&ia->ia_addr.sin6_addr); if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) { if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0) REPLACE(2); NEXT(2); } else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) { if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0) NEXT(2); REPLACE(2); } /* * Rule 3: Avoid deprecated addresses. Note that the case of * !ip6_use_deprecated is already rejected above. */ if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia)) NEXT(3); if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia)) REPLACE(3); /* Rule 4: Prefer home addresses */ /* * XXX: This is a TODO. We should probably merge the MIP6 * case above. */ /* Rule 5: Prefer outgoing interface */ if (!(ND_IFINFO(ifp)->flags & ND6_IFF_NO_PREFER_IFACE)) { if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp) NEXT(5); if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp) REPLACE(5); } /* * Rule 6: Prefer matching label * Note that best_policy should be non-NULL here. */ if (dst_policy == NULL) dst_policy = lookup_addrsel_policy(dstsock); if (dst_policy->label != ADDR_LABEL_NOTAPP) { new_policy = lookup_addrsel_policy(&ia->ia_addr); if (dst_policy->label == best_policy->label && dst_policy->label != new_policy->label) NEXT(6); if (dst_policy->label != best_policy->label && dst_policy->label == new_policy->label) REPLACE(6); } /* * Rule 7: Prefer public addresses. * We allow users to reverse the logic by configuring * a sysctl variable, so that privacy conscious users can * always prefer temporary addresses. */ if (opts == NULL || opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) { prefer_tempaddr = V_ip6_prefer_tempaddr; } else if (opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_NOTPREFER) { prefer_tempaddr = 0; } else prefer_tempaddr = 1; if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) && (ia->ia6_flags & IN6_IFF_TEMPORARY)) { if (prefer_tempaddr) REPLACE(7); else NEXT(7); } if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) && !(ia->ia6_flags & IN6_IFF_TEMPORARY)) { if (prefer_tempaddr) NEXT(7); else REPLACE(7); } /* * Rule 8: prefer addresses on alive interfaces. * This is a KAME specific rule. */ if ((ia_best->ia_ifp->if_flags & IFF_UP) && !(ia->ia_ifp->if_flags & IFF_UP)) NEXT(8); if (!(ia_best->ia_ifp->if_flags & IFF_UP) && (ia->ia_ifp->if_flags & IFF_UP)) REPLACE(8); /* * Rule 9: prefer address with better virtual status. */ if (ifa_preferred(&ia_best->ia_ifa, &ia->ia_ifa)) REPLACE(9); if (ifa_preferred(&ia->ia_ifa, &ia_best->ia_ifa)) NEXT(9); /* * Rule 10: prefer address with `prefer_source' flag. */ if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0 && (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0) REPLACE(10); if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0 && (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0) NEXT(10); /* * Rule 14: Use longest matching prefix. * Note: in the address selection draft, this rule is * documented as "Rule 8". However, since it is also * documented that this rule can be overridden, we assign * a large number so that it is easy to assign smaller numbers * to more preferred rules. */ new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst); if (best_matchlen < new_matchlen) REPLACE(14); if (new_matchlen < best_matchlen) NEXT(14); /* Rule 15 is reserved. */ /* * Last resort: just keep the current candidate. * Or, do we need more rules? */ continue; replace: ia_best = ia; best_scope = (new_scope >= 0 ? new_scope : in6_addrscope(&ia_best->ia_addr.sin6_addr)); best_policy = (new_policy ? new_policy : lookup_addrsel_policy(&ia_best->ia_addr)); best_matchlen = (new_matchlen >= 0 ? new_matchlen : in6_matchlen(&ia_best->ia_addr.sin6_addr, &dst)); next: continue; out: break; } if ((ia = ia_best) == NULL) { IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); IP6STAT_INC(ip6s_sources_none); return (EADDRNOTAVAIL); } /* * At this point at least one of the addresses belonged to the jail * but it could still be, that we want to further restrict it, e.g. * theoratically IN6_IS_ADDR_LOOPBACK. * It must not be IN6_IS_ADDR_UNSPECIFIED anymore. * prison_local_ip6() will fix an IN6_IS_ADDR_LOOPBACK but should * let all others previously selected pass. * Use tmp to not change ::1 on lo0 to the primary jail address. */ tmp = ia->ia_addr.sin6_addr; if (cred != NULL && prison_local_ip6(cred, &tmp, (inp != NULL && (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) { IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); IP6STAT_INC(ip6s_sources_none); return (EADDRNOTAVAIL); } if (ifpp) *ifpp = ifp; bcopy(&tmp, srcp, sizeof(*srcp)); if (ia->ia_ifp == ifp) IP6STAT_INC(ip6s_sources_sameif[best_scope]); else IP6STAT_INC(ip6s_sources_otherif[best_scope]); if (dst_scope == best_scope) IP6STAT_INC(ip6s_sources_samescope[best_scope]); else IP6STAT_INC(ip6s_sources_otherscope[best_scope]); if (IFA6_IS_DEPRECATED(ia)) IP6STAT_INC(ip6s_sources_deprecated[best_scope]); IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); return (0); } /* * Select source address based on @inp, @dstsock and @opts. * Stores selected address to @srcp. If @scope_ambiguous is set, * embed scope from selected outgoing interface. If @hlim pointer * is provided, stores calculated hop limit there. * Returns 0 on success. */ int in6_selectsrc_socket(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred, int scope_ambiguous, struct in6_addr *srcp, int *hlim) { struct ifnet *retifp; uint32_t fibnum; int error; fibnum = inp->inp_inc.inc_fibnum; retifp = NULL; error = in6_selectsrc(fibnum, dstsock, opts, inp, cred, &retifp, srcp); if (error != 0) return (error); if (hlim != NULL) *hlim = in6_selecthlim(inp, retifp); if (retifp == NULL || scope_ambiguous == 0) return (0); /* * Application should provide a proper zone ID or the use of * default zone IDs should be enabled. Unfortunately, some * applications do not behave as it should, so we need a * workaround. Even if an appropriate ID is not determined * (when it's required), if we can determine the outgoing * interface. determine the zone ID based on the interface. */ error = in6_setscope(&dstsock->sin6_addr, retifp, NULL); return (error); } /* * Select source address based on @fibnum, @dst and @scopeid. * Stores selected address to @srcp. * Returns 0 on success. * * Used by non-socket based consumers (ND code mostly) */ int in6_selectsrc_addr(uint32_t fibnum, const struct in6_addr *dst, uint32_t scopeid, struct ifnet *ifp, struct in6_addr *srcp, int *hlim) { struct ifnet *retifp; struct sockaddr_in6 dst_sa; int error; retifp = ifp; bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = *dst; dst_sa.sin6_scope_id = scopeid; sa6_embedscope(&dst_sa, 0); error = in6_selectsrc(fibnum, &dst_sa, NULL, NULL, NULL, &retifp, srcp); if (hlim != NULL) *hlim = in6_selecthlim(NULL, retifp); return (error); } /* * clone - meaningful only for bsdi and freebsd */ static int selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct route_in6 *ro, - struct ifnet **retifp, struct rtentry **retrt, int norouteok, u_int fibnum) + struct ifnet **retifp, struct nhop_object **retnh, int norouteok, + u_int fibnum, uint32_t flowid) { int error = 0; struct ifnet *ifp = NULL; - struct rtentry *rt = NULL; + struct nhop_object *nh = NULL; struct sockaddr_in6 *sin6_next; struct in6_pktinfo *pi = NULL; struct in6_addr *dst = &dstsock->sin6_addr; uint32_t zoneid; #if 0 char ip6buf[INET6_ADDRSTRLEN]; if (dstsock->sin6_addr.s6_addr32[0] == 0 && dstsock->sin6_addr.s6_addr32[1] == 0 && !IN6_IS_ADDR_LOOPBACK(&dstsock->sin6_addr)) { printf("%s: strange destination %s\n", __func__, ip6_sprintf(ip6buf, &dstsock->sin6_addr)); } else { printf("%s: destination = %s%%%d\n", __func__, ip6_sprintf(ip6buf, &dstsock->sin6_addr), dstsock->sin6_scope_id); /* for debug */ } #endif /* If the caller specify the outgoing interface explicitly, use it. */ if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) { /* XXX boundary check is assumed to be already done. */ ifp = ifnet_byindex(pi->ipi6_ifindex); if (ifp != NULL && - (norouteok || retrt == NULL || + (norouteok || retnh == NULL || IN6_IS_ADDR_MULTICAST(dst))) { /* * we do not have to check or get the route for * multicast. */ goto done; } else goto getroute; } /* * If the destination address is a multicast address and the outgoing * interface for the address is specified by the caller, use it. */ if (IN6_IS_ADDR_MULTICAST(dst) && mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) { goto done; /* we do not need a route for multicast. */ } /* * If destination address is LLA or link- or node-local multicast, * use it's embedded scope zone id to determine outgoing interface. */ if (IN6_IS_ADDR_MC_LINKLOCAL(dst) || IN6_IS_ADDR_MC_NODELOCAL(dst)) { zoneid = ntohs(in6_getscope(dst)); if (zoneid > 0) { ifp = in6_getlinkifnet(zoneid); goto done; } } getroute: /* * If the next hop address for the packet is specified by the caller, * use it as the gateway. */ if (opts && opts->ip6po_nexthop) { struct route_in6 *ron; sin6_next = satosin6(opts->ip6po_nexthop); if (IN6_IS_ADDR_LINKLOCAL(&sin6_next->sin6_addr)) { /* * Next hop is LLA, thus it should be neighbor. * Determine outgoing interface by zone index. */ zoneid = ntohs(in6_getscope(&sin6_next->sin6_addr)); if (zoneid > 0) { ifp = in6_getlinkifnet(zoneid); goto done; } } ron = &opts->ip6po_nextroute; /* Use a cached route if it exists and is valid. */ - if (ron->ro_rt != NULL && ( - (ron->ro_rt->rt_flags & RTF_UP) == 0 || + if (ron->ro_nh != NULL && ( + !NH_IS_VALID(ron->ro_nh) || ron->ro_dst.sin6_family != AF_INET6 || !IN6_ARE_ADDR_EQUAL(&ron->ro_dst.sin6_addr, &sin6_next->sin6_addr))) - RO_RTFREE(ron); - if (ron->ro_rt == NULL) { + RO_NHFREE(ron); + if (ron->ro_nh == NULL) { ron->ro_dst = *sin6_next; - in6_rtalloc(ron, fibnum); /* multi path case? */ + /* + * sin6_next is not link-local OR scopeid is 0, + * no need to clear scope + */ + ron->ro_nh = fib6_lookup(fibnum, + &sin6_next->sin6_addr, 0, NHR_REF, flowid); } /* * The node identified by that address must be a * neighbor of the sending host. */ - if (ron->ro_rt == NULL || - (ron->ro_rt->rt_flags & RTF_GATEWAY) != 0) + if (ron->ro_nh == NULL || + (ron->ro_nh->nh_flags & NHF_GATEWAY) != 0) error = EHOSTUNREACH; else { - rt = ron->ro_rt; - ifp = rt->rt_ifp; + nh = ron->ro_nh; + ifp = nh->nh_ifp; } goto done; } /* * Use a cached route if it exists and is valid, else try to allocate * a new one. Note that we should check the address family of the * cached destination, in case of sharing the cache with IPv4. */ if (ro) { - if (ro->ro_rt && - (!(ro->ro_rt->rt_flags & RTF_UP) || + if (ro->ro_nh && + (!NH_IS_VALID(ro->ro_nh) || ((struct sockaddr *)(&ro->ro_dst))->sa_family != AF_INET6 || !IN6_ARE_ADDR_EQUAL(&satosin6(&ro->ro_dst)->sin6_addr, dst))) { - RTFREE(ro->ro_rt); - ro->ro_rt = (struct rtentry *)NULL; + RO_NHFREE(ro); } - if (ro->ro_rt == (struct rtentry *)NULL) { + if (ro->ro_nh == (struct nhop_object *)NULL) { struct sockaddr_in6 *sa6; /* No route yet, so try to acquire one */ bzero(&ro->ro_dst, sizeof(struct sockaddr_in6)); sa6 = (struct sockaddr_in6 *)&ro->ro_dst; *sa6 = *dstsock; sa6->sin6_scope_id = 0; + /* + * Currently dst has scopeid embedded iff it is LL. + * New routing API accepts scopeid as a separate argument. + * Convert dst before/after doing lookup + */ + uint32_t scopeid = 0; + if (IN6_IS_SCOPE_LINKLOCAL(&sa6->sin6_addr)) { + /* Unwrap in6_getscope() and in6_clearscope() */ + scopeid = ntohs(sa6->sin6_addr.s6_addr16[1]); + sa6->sin6_addr.s6_addr16[1] = 0; + + } + #ifdef RADIX_MPATH rtalloc_mpath_fib((struct route *)ro, ntohl(sa6->sin6_addr.s6_addr32[3]), fibnum); #else - ro->ro_rt = in6_rtalloc1((struct sockaddr *) - &ro->ro_dst, 0, 0UL, fibnum); - if (ro->ro_rt) - RT_UNLOCK(ro->ro_rt); + ro->ro_nh = fib6_lookup(fibnum, + &sa6->sin6_addr, scopeid, NHR_REF, flowid); #endif + if (IN6_IS_SCOPE_LINKLOCAL(&sa6->sin6_addr)) + sa6->sin6_addr.s6_addr16[1] = htons(scopeid); } /* * do not care about the result if we have the nexthop * explicitly specified. */ if (opts && opts->ip6po_nexthop) goto done; - if (ro->ro_rt) { - ifp = ro->ro_rt->rt_ifp; - - if (ifp == NULL) { /* can this really happen? */ - RTFREE(ro->ro_rt); - ro->ro_rt = NULL; - } - } - if (ro->ro_rt == NULL) + if (ro->ro_nh) + ifp = ro->ro_nh->nh_ifp; + else error = EHOSTUNREACH; - rt = ro->ro_rt; + nh = ro->ro_nh; /* * Check if the outgoing interface conflicts with * the interface specified by ipi6_ifindex (if specified). * Note that loopback interface is always okay. * (this may happen when we are sending a packet to one of * our own addresses.) */ if (ifp && opts && opts->ip6po_pktinfo && opts->ip6po_pktinfo->ipi6_ifindex) { if (!(ifp->if_flags & IFF_LOOPBACK) && ifp->if_index != opts->ip6po_pktinfo->ipi6_ifindex) { error = EHOSTUNREACH; goto done; } } } done: - if (ifp == NULL && rt == NULL) { + if (ifp == NULL && nh == NULL) { /* * This can happen if the caller did not pass a cached route * nor any other hints. We treat this case an error. */ error = EHOSTUNREACH; } if (error == EHOSTUNREACH) IP6STAT_INC(ip6s_noroute); if (retifp != NULL) { - *retifp = ifp; - - /* - * Adjust the "outgoing" interface. If we're going to loop - * the packet back to ourselves, the ifp would be the loopback - * interface. However, we'd rather know the interface associated - * to the destination address (which should probably be one of - * our own addresses.) - */ - if (rt) { - if ((rt->rt_ifp->if_flags & IFF_LOOPBACK) && - (rt->rt_gateway->sa_family == AF_LINK)) - *retifp = - ifnet_byindex(((struct sockaddr_dl *) - rt->rt_gateway)->sdl_index); - } + if (nh != NULL) + *retifp = nh->nh_aifp; + else + *retifp = ifp; } - if (retrt != NULL) - *retrt = rt; /* rt may be NULL */ + if (retnh != NULL) + *retnh = nh; /* nh may be NULL */ return (error); } static int in6_selectif(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct ifnet **retifp, struct ifnet *oifp, u_int fibnum) { int error; struct route_in6 sro; - struct rtentry *rt = NULL; - int rt_flags; + struct nhop_object *nh = NULL; + uint16_t nh_flags; KASSERT(retifp != NULL, ("%s: retifp is NULL", __func__)); bzero(&sro, sizeof(sro)); - rt_flags = 0; + nh_flags = 0; - error = selectroute(dstsock, opts, mopts, &sro, retifp, &rt, 1, fibnum); + error = selectroute(dstsock, opts, mopts, &sro, retifp, &nh, 1, fibnum, 0); - if (rt) - rt_flags = rt->rt_flags; - if (rt && rt == sro.ro_rt) - RTFREE(rt); + if (nh != NULL) + nh_flags = nh->nh_flags; + if (nh != NULL && nh == sro.ro_nh) + NH_FREE(nh); if (error != 0) { /* Help ND. See oifp comment in in6_selectsrc(). */ if (oifp != NULL && fibnum == RT_DEFAULT_FIB) { *retifp = oifp; error = 0; } return (error); } /* * do not use a rejected or black hole route. * XXX: this check should be done in the L2 output routine. * However, if we skipped this check here, we'd see the following * scenario: * - install a rejected route for a scoped address prefix * (like fe80::/10) * - send a packet to a destination that matches the scoped prefix, * with ambiguity about the scope zone. * - pick the outgoing interface from the route, and disambiguate the * scope zone with the interface. * - ip6_output() would try to get another route with the "new" * destination, which may be valid. * - we'd see no error on output. * Although this may not be very harmful, it should still be confusing. * We thus reject the case here. */ - if (rt_flags & (RTF_REJECT | RTF_BLACKHOLE)) { - error = (rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); + if (nh_flags & (NHF_REJECT | NHF_BLACKHOLE)) { + error = (nh_flags & NHF_HOST ? EHOSTUNREACH : ENETUNREACH); return (error); } return (0); } /* Public wrapper function to selectroute(). */ int in6_selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct route_in6 *ro, - struct ifnet **retifp, struct rtentry **retrt, u_int fibnum) + struct ifnet **retifp, struct nhop_object **retnh, u_int fibnum, uint32_t flowid) { return (selectroute(dstsock, opts, mopts, ro, retifp, - retrt, 0, fibnum)); + retnh, 0, fibnum, flowid)); } /* * Default hop limit selection. The precedence is as follows: * 1. Hoplimit value specified via ioctl. * 2. (If the outgoing interface is detected) the current * hop limit of the interface specified by router advertisement. * 3. The system default hoplimit. */ int in6_selecthlim(struct inpcb *inp, struct ifnet *ifp) { if (inp && inp->in6p_hops >= 0) return (inp->in6p_hops); else if (ifp) return (ND_IFINFO(ifp)->chlim); else if (inp && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { struct nhop6_basic nh6; struct in6_addr dst; uint32_t fibnum, scopeid; int hlim; fibnum = inp->inp_inc.inc_fibnum; in6_splitscope(&inp->in6p_faddr, &dst, &scopeid); if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6)==0){ hlim = ND_IFINFO(nh6.nh_ifp)->chlim; return (hlim); } } return (V_ip6_defhlim); } /* * XXX: this is borrowed from in6_pcbbind(). If possible, we should * share this function by all *bsd*... */ int in6_pcbsetport(struct in6_addr *laddr, struct inpcb *inp, struct ucred *cred) { struct socket *so = inp->inp_socket; u_int16_t lport = 0; int error, lookupflags = 0; #ifdef INVARIANTS struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; #endif INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); error = prison_local_ip6(cred, laddr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)); if (error) return(error); /* XXX: this is redundant when called from in6_pcbbind */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) lookupflags = INPLOOKUP_WILDCARD; inp->inp_flags |= INP_ANONPORT; error = in_pcb_lport(inp, NULL, &lport, cred, lookupflags); if (error != 0) return (error); inp->inp_lport = lport; if (in_pcbinshash(inp) != 0) { inp->in6p_laddr = in6addr_any; inp->inp_lport = 0; return (EAGAIN); } return (0); } void addrsel_policy_init(void) { init_policy_queue(); /* initialize the "last resort" policy */ bzero(&V_defaultaddrpolicy, sizeof(V_defaultaddrpolicy)); V_defaultaddrpolicy.label = ADDR_LABEL_NOTAPP; if (!IS_DEFAULT_VNET(curvnet)) return; ADDRSEL_LOCK_INIT(); ADDRSEL_SXLOCK_INIT(); } static struct in6_addrpolicy * lookup_addrsel_policy(struct sockaddr_in6 *key) { struct in6_addrpolicy *match = NULL; ADDRSEL_LOCK(); match = match_addrsel_policy(key); if (match == NULL) match = &V_defaultaddrpolicy; else match->use++; ADDRSEL_UNLOCK(); return (match); } /* * Subroutines to manage the address selection policy table via sysctl. */ struct walkarg { struct sysctl_req *w_req; }; static int in6_src_sysctl(SYSCTL_HANDLER_ARGS); SYSCTL_DECL(_net_inet6_ip6); static SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy, CTLFLAG_RD | CTLFLAG_MPSAFE, in6_src_sysctl, ""); static int in6_src_sysctl(SYSCTL_HANDLER_ARGS) { struct walkarg w; if (req->newptr) return EPERM; bzero(&w, sizeof(w)); w.w_req = req; return (walk_addrsel_policy(dump_addrsel_policyent, &w)); } int in6_src_ioctl(u_long cmd, caddr_t data) { struct in6_addrpolicy ent0; if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY) return (EOPNOTSUPP); /* check for safety */ ent0 = *(struct in6_addrpolicy *)data; if (ent0.label == ADDR_LABEL_NOTAPP) return (EINVAL); /* check if the prefix mask is consecutive. */ if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0) return (EINVAL); /* clear trailing garbages (if any) of the prefix address. */ IN6_MASK_ADDR(&ent0.addr.sin6_addr, &ent0.addrmask.sin6_addr); ent0.use = 0; switch (cmd) { case SIOCAADDRCTL_POLICY: return (add_addrsel_policyent(&ent0)); case SIOCDADDRCTL_POLICY: return (delete_addrsel_policyent(&ent0)); } return (0); /* XXX: compromise compilers */ } /* * The followings are implementation of the policy table using a * simple tail queue. * XXX such details should be hidden. * XXX implementation using binary tree should be more efficient. */ struct addrsel_policyent { TAILQ_ENTRY(addrsel_policyent) ape_entry; struct in6_addrpolicy ape_policy; }; TAILQ_HEAD(addrsel_policyhead, addrsel_policyent); VNET_DEFINE_STATIC(struct addrsel_policyhead, addrsel_policytab); #define V_addrsel_policytab VNET(addrsel_policytab) static void init_policy_queue(void) { TAILQ_INIT(&V_addrsel_policytab); } static int add_addrsel_policyent(struct in6_addrpolicy *newpolicy) { struct addrsel_policyent *new, *pol; new = malloc(sizeof(*new), M_IFADDR, M_WAITOK); ADDRSEL_XLOCK(); ADDRSEL_LOCK(); /* duplication check */ TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr, &pol->ape_policy.addr.sin6_addr) && IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr, &pol->ape_policy.addrmask.sin6_addr)) { ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); free(new, M_IFADDR); return (EEXIST); /* or override it? */ } } bzero(new, sizeof(*new)); /* XXX: should validate entry */ new->ape_policy = *newpolicy; TAILQ_INSERT_TAIL(&V_addrsel_policytab, new, ape_entry); ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); return (0); } static int delete_addrsel_policyent(struct in6_addrpolicy *key) { struct addrsel_policyent *pol; ADDRSEL_XLOCK(); ADDRSEL_LOCK(); /* search for the entry in the table */ TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr, &pol->ape_policy.addr.sin6_addr) && IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr, &pol->ape_policy.addrmask.sin6_addr)) { break; } } if (pol == NULL) { ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); return (ESRCH); } TAILQ_REMOVE(&V_addrsel_policytab, pol, ape_entry); ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); free(pol, M_IFADDR); return (0); } static int walk_addrsel_policy(int (*callback)(struct in6_addrpolicy *, void *), void *w) { struct addrsel_policyent *pol; int error = 0; ADDRSEL_SLOCK(); TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if ((error = (*callback)(&pol->ape_policy, w)) != 0) { ADDRSEL_SUNLOCK(); return (error); } } ADDRSEL_SUNLOCK(); return (error); } static int dump_addrsel_policyent(struct in6_addrpolicy *pol, void *arg) { int error = 0; struct walkarg *w = arg; error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol)); return (error); } static struct in6_addrpolicy * match_addrsel_policy(struct sockaddr_in6 *key) { struct addrsel_policyent *pent; struct in6_addrpolicy *bestpol = NULL, *pol; int matchlen, bestmatchlen = -1; u_char *mp, *ep, *k, *p, m; TAILQ_FOREACH(pent, &V_addrsel_policytab, ape_entry) { matchlen = 0; pol = &pent->ape_policy; mp = (u_char *)&pol->addrmask.sin6_addr; ep = mp + 16; /* XXX: scope field? */ k = (u_char *)&key->sin6_addr; p = (u_char *)&pol->addr.sin6_addr; for (; mp < ep && *mp; mp++, k++, p++) { m = *mp; if ((*k & m) != *p) goto next; /* not match */ if (m == 0xff) /* short cut for a typical case */ matchlen += 8; else { while (m >= 0x80) { matchlen++; m <<= 1; } } } /* matched. check if this is better than the current best. */ if (bestpol == NULL || matchlen > bestmatchlen) { bestpol = pol; bestmatchlen = matchlen; } next: continue; } return (bestpol); } Index: head/sys/netinet6/ip6_output.c =================================================================== --- head/sys/netinet6/ip6_output.c (revision 360291) +++ head/sys/netinet6/ip6_output.c (revision 360292) @@ -1,3330 +1,3328 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: ip6_output.c,v 1.279 2002/01/26 06:12:30 jinmei Exp $ */ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)ip_output.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_ratelimit.h" #include "opt_route.h" #include "opt_rss.h" #include "opt_sctp.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SCTP #include #include #endif #include #include extern int in6_mcast_loop; struct ip6_exthdrs { struct mbuf *ip6e_ip6; struct mbuf *ip6e_hbh; struct mbuf *ip6e_dest1; struct mbuf *ip6e_rthdr; struct mbuf *ip6e_dest2; }; static MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); static int ip6_pcbopt(int, u_char *, int, struct ip6_pktopts **, struct ucred *, int); static int ip6_pcbopts(struct ip6_pktopts **, struct mbuf *, struct socket *, struct sockopt *); static int ip6_getpcbopt(struct inpcb *, int, struct sockopt *); static int ip6_setpktopt(int, u_char *, int, struct ip6_pktopts *, struct ucred *, int, int, int); static int ip6_copyexthdr(struct mbuf **, caddr_t, int); static int ip6_insertfraghdr(struct mbuf *, struct mbuf *, int, struct ip6_frag **); static int ip6_insert_jumboopt(struct ip6_exthdrs *, u_int32_t); static int ip6_splithdr(struct mbuf *, struct ip6_exthdrs *); static int ip6_getpmtu(struct route_in6 *, int, struct ifnet *, const struct in6_addr *, u_long *, int *, u_int, u_int); static int ip6_calcmtu(struct ifnet *, const struct in6_addr *, u_long, u_long *, int *, u_int); static int ip6_getpmtu_ctl(u_int, const struct in6_addr *, u_long *); static int copypktopts(struct ip6_pktopts *, struct ip6_pktopts *, int); /* * Make an extension header from option data. hp is the source, * mp is the destination, and _ol is the optlen. */ #define MAKE_EXTHDR(hp, mp, _ol) \ do { \ if (hp) { \ struct ip6_ext *eh = (struct ip6_ext *)(hp); \ error = ip6_copyexthdr((mp), (caddr_t)(hp), \ ((eh)->ip6e_len + 1) << 3); \ if (error) \ goto freehdrs; \ (_ol) += (*(mp))->m_len; \ } \ } while (/*CONSTCOND*/ 0) /* * Form a chain of extension headers. * m is the extension header mbuf * mp is the previous mbuf in the chain * p is the next header * i is the type of option. */ #define MAKE_CHAIN(m, mp, p, i)\ do {\ if (m) {\ if (!hdrsplit) \ panic("%s:%d: assumption failed: "\ "hdr not split: hdrsplit %d exthdrs %p",\ __func__, __LINE__, hdrsplit, &exthdrs);\ *mtod((m), u_char *) = *(p);\ *(p) = (i);\ p = mtod((m), u_char *);\ (m)->m_next = (mp)->m_next;\ (mp)->m_next = (m);\ (mp) = (m);\ }\ } while (/*CONSTCOND*/ 0) void in6_delayed_cksum(struct mbuf *m, uint32_t plen, u_short offset) { u_short csum; csum = in_cksum_skip(m, offset + plen, offset); if (m->m_pkthdr.csum_flags & CSUM_UDP_IPV6 && csum == 0) csum = 0xffff; offset += m->m_pkthdr.csum_data; /* checksum offset */ if (offset + sizeof(csum) > m->m_len) m_copyback(m, offset, sizeof(csum), (caddr_t)&csum); else *(u_short *)mtodo(m, offset) = csum; } static int ip6_output_delayed_csum(struct mbuf *m, struct ifnet *ifp, int csum_flags, int plen, int optlen, bool frag) { KASSERT((plen >= optlen), ("%s:%d: plen %d < optlen %d, m %p, ifp %p " "csum_flags %#x frag %d\n", __func__, __LINE__, plen, optlen, m, ifp, csum_flags, frag)); if ((csum_flags & CSUM_DELAY_DATA_IPV6) || #ifdef SCTP (csum_flags & CSUM_SCTP_IPV6) || #endif (!frag && (ifp->if_capenable & IFCAP_NOMAP) == 0)) { m = mb_unmapped_to_ext(m); if (m == NULL) { if (frag) in6_ifstat_inc(ifp, ifs6_out_fragfail); else IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } if (csum_flags & CSUM_DELAY_DATA_IPV6) { in6_delayed_cksum(m, plen - optlen, sizeof(struct ip6_hdr) + optlen); m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; } #ifdef SCTP if (csum_flags & CSUM_SCTP_IPV6) { sctp_delayed_cksum(m, sizeof(struct ip6_hdr) + optlen); m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; } #endif } return (0); } int ip6_fragment(struct ifnet *ifp, struct mbuf *m0, int hlen, u_char nextproto, int fraglen , uint32_t id) { struct mbuf *m, **mnext, *m_frgpart; struct ip6_hdr *ip6, *mhip6; struct ip6_frag *ip6f; int off; int error; int tlen = m0->m_pkthdr.len; KASSERT((fraglen % 8 == 0), ("Fragment length must be a multiple of 8")); m = m0; ip6 = mtod(m, struct ip6_hdr *); mnext = &m->m_nextpkt; for (off = hlen; off < tlen; off += fraglen) { m = m_gethdr(M_NOWAIT, MT_DATA); if (!m) { IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } /* * Make sure the complete packet header gets copied * from the originating mbuf to the newly created * mbuf. This also ensures that existing firewall * classification(s), VLAN tags and so on get copied * to the resulting fragmented packet(s): */ if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) { m_free(m); IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } *mnext = m; mnext = &m->m_nextpkt; m->m_data += max_linkhdr; mhip6 = mtod(m, struct ip6_hdr *); *mhip6 = *ip6; m->m_len = sizeof(*mhip6); error = ip6_insertfraghdr(m0, m, hlen, &ip6f); if (error) { IP6STAT_INC(ip6s_odropped); return (error); } ip6f->ip6f_offlg = htons((u_short)((off - hlen) & ~7)); if (off + fraglen >= tlen) fraglen = tlen - off; else ip6f->ip6f_offlg |= IP6F_MORE_FRAG; mhip6->ip6_plen = htons((u_short)(fraglen + hlen + sizeof(*ip6f) - sizeof(struct ip6_hdr))); if ((m_frgpart = m_copym(m0, off, fraglen, M_NOWAIT)) == NULL) { IP6STAT_INC(ip6s_odropped); return (ENOBUFS); } m_cat(m, m_frgpart); m->m_pkthdr.len = fraglen + hlen + sizeof(*ip6f); ip6f->ip6f_reserved = 0; ip6f->ip6f_ident = id; ip6f->ip6f_nxt = nextproto; IP6STAT_INC(ip6s_ofragments); in6_ifstat_inc(ifp, ifs6_out_fragcreat); } return (0); } static int ip6_output_send(struct inpcb *inp, struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, struct sockaddr_in6 *dst, struct route_in6 *ro) { #ifdef KERN_TLS struct ktls_session *tls = NULL; #endif struct m_snd_tag *mst; int error; MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); mst = NULL; #ifdef KERN_TLS /* * If this is an unencrypted TLS record, save a reference to * the record. This local reference is used to call * ktls_output_eagain after the mbuf has been freed (thus * dropping the mbuf's reference) in if_output. */ if (m->m_next != NULL && mbuf_has_tls_session(m->m_next)) { tls = ktls_hold(m->m_next->m_ext_pgs.tls); mst = tls->snd_tag; /* * If a TLS session doesn't have a valid tag, it must * have had an earlier ifp mismatch, so drop this * packet. */ if (mst == NULL) { error = EAGAIN; goto done; } } #endif #ifdef RATELIMIT if (inp != NULL && mst == NULL) { if ((inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) != 0 || (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp)) in_pcboutput_txrtlmt(inp, ifp, m); if (inp->inp_snd_tag != NULL) mst = inp->inp_snd_tag; } #endif if (mst != NULL) { KASSERT(m->m_pkthdr.rcvif == NULL, ("trying to add a send tag to a forwarded packet")); if (mst->ifp != ifp) { error = EAGAIN; goto done; } /* stamp send tag on mbuf */ m->m_pkthdr.snd_tag = m_snd_tag_ref(mst); m->m_pkthdr.csum_flags |= CSUM_SND_TAG; } error = nd6_output_ifp(ifp, origifp, m, dst, (struct route *)ro); done: /* Check for route change invalidating send tags. */ #ifdef KERN_TLS if (tls != NULL) { if (error == EAGAIN) error = ktls_output_eagain(inp, tls); ktls_free(tls); } #endif #ifdef RATELIMIT if (error == EAGAIN) in_pcboutput_eagain(inp); #endif return (error); } /* * IP6 output. * The packet in mbuf chain m contains a skeletal IP6 header (with pri, len, * nxt, hlim, src, dst). * This function may modify ver and hlim only. * The mbuf chain containing the packet will be freed. * The mbuf opt, if present, will not be freed. - * If route_in6 ro is present and has ro_rt initialized, route lookup would be - * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL, - * then result of route lookup is stored in ro->ro_rt. + * If route_in6 ro is present and has ro_nh initialized, route lookup would be + * skipped and ro->ro_nh would be used. If ro is present but ro->ro_nh is NULL, + * then result of route lookup is stored in ro->ro_nh. * * Type of "mtu": rt_mtu is u_long, ifnet.ifr_mtu is int, and nd_ifinfo.linkmtu * is uint32_t. So we use u_long to hold largest one, which is rt_mtu. * * ifpp - XXX: just for statistics */ /* * XXX TODO: no flowid is assigned for outbound flows? */ int ip6_output(struct mbuf *m0, struct ip6_pktopts *opt, struct route_in6 *ro, int flags, struct ip6_moptions *im6o, struct ifnet **ifpp, struct inpcb *inp) { struct ip6_hdr *ip6; struct ifnet *ifp, *origifp; struct mbuf *m = m0; struct mbuf *mprev; struct route_in6 *ro_pmtu; - struct rtentry *rt; + struct nhop_object *nh; struct sockaddr_in6 *dst, sin6, src_sa, dst_sa; struct in6_addr odst; u_char *nexthdrp; int tlen, len; int error = 0; struct in6_ifaddr *ia = NULL; u_long mtu; int alwaysfrag, dontfrag; u_int32_t optlen, plen = 0, unfragpartlen; struct ip6_exthdrs exthdrs; struct in6_addr src0, dst0; u_int32_t zone; bool hdrsplit; int sw_csum, tso; int needfiblookup; uint32_t fibnum; struct m_tag *fwd_tag = NULL; uint32_t id; NET_EPOCH_ASSERT(); if (inp != NULL) { INP_LOCK_ASSERT(inp); M_SETFIB(m, inp->inp_inc.inc_fibnum); if ((flags & IP_NODEFAULTFLOWID) == 0) { /* Unconditionally set flowid. */ m->m_pkthdr.flowid = inp->inp_flowid; M_HASHTYPE_SET(m, inp->inp_flowtype); } #ifdef NUMA m->m_pkthdr.numa_domain = inp->inp_numa_domain; #endif } #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * IPSec checking which handles several cases. * FAST IPSEC: We re-injected the packet. * XXX: need scope argument. */ if (IPSEC_ENABLED(ipv6)) { if ((error = IPSEC_OUTPUT(ipv6, m, inp)) != 0) { if (error == EINPROGRESS) error = 0; goto done; } } #endif /* IPSEC */ /* Source address validation. */ ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && (flags & IPV6_UNSPECSRC) == 0) { error = EOPNOTSUPP; IP6STAT_INC(ip6s_badscope); goto bad; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { error = EOPNOTSUPP; IP6STAT_INC(ip6s_badscope); goto bad; } /* * If we are given packet options to add extension headers prepare them. * Calculate the total length of the extension header chain. * Keep the length of the unfragmentable part for fragmentation. */ bzero(&exthdrs, sizeof(exthdrs)); optlen = 0; unfragpartlen = sizeof(struct ip6_hdr); if (opt) { /* Hop-by-Hop options header. */ MAKE_EXTHDR(opt->ip6po_hbh, &exthdrs.ip6e_hbh, optlen); /* Destination options header (1st part). */ if (opt->ip6po_rthdr) { #ifndef RTHDR_SUPPORT_IMPLEMENTED /* * If there is a routing header, discard the packet * right away here. RH0/1 are obsolete and we do not * currently support RH2/3/4. * People trying to use RH253/254 may want to disable * this check. * The moment we do support any routing header (again) * this block should check the routing type more * selectively. */ error = EINVAL; goto bad; #endif /* * Destination options header (1st part). * This only makes sense with a routing header. * See Section 9.2 of RFC 3542. * Disabling this part just for MIP6 convenience is * a bad idea. We need to think carefully about a * way to make the advanced API coexist with MIP6 * options, which might automatically be inserted in * the kernel. */ MAKE_EXTHDR(opt->ip6po_dest1, &exthdrs.ip6e_dest1, optlen); } /* Routing header. */ MAKE_EXTHDR(opt->ip6po_rthdr, &exthdrs.ip6e_rthdr, optlen); unfragpartlen += optlen; /* * NOTE: we don't add AH/ESP length here (done in * ip6_ipsec_output()). */ /* Destination options header (2nd part). */ MAKE_EXTHDR(opt->ip6po_dest2, &exthdrs.ip6e_dest2, optlen); } /* * If there is at least one extension header, * separate IP6 header from the payload. */ hdrsplit = false; if (optlen) { if ((error = ip6_splithdr(m, &exthdrs)) != 0) { m = NULL; goto freehdrs; } m = exthdrs.ip6e_ip6; ip6 = mtod(m, struct ip6_hdr *); hdrsplit = true; } /* Adjust mbuf packet header length. */ m->m_pkthdr.len += optlen; plen = m->m_pkthdr.len - sizeof(*ip6); /* If this is a jumbo payload, insert a jumbo payload option. */ if (plen > IPV6_MAXPACKET) { if (!hdrsplit) { if ((error = ip6_splithdr(m, &exthdrs)) != 0) { m = NULL; goto freehdrs; } m = exthdrs.ip6e_ip6; ip6 = mtod(m, struct ip6_hdr *); hdrsplit = true; } if ((error = ip6_insert_jumboopt(&exthdrs, plen)) != 0) goto freehdrs; ip6->ip6_plen = 0; } else ip6->ip6_plen = htons(plen); nexthdrp = &ip6->ip6_nxt; if (optlen) { /* * Concatenate headers and fill in next header fields. * Here we have, on "m" * IPv6 payload * and we insert headers accordingly. * Finally, we should be getting: * IPv6 hbh dest1 rthdr ah* [esp* dest2 payload]. * * During the header composing process "m" points to IPv6 * header. "mprev" points to an extension header prior to esp. */ mprev = m; /* * We treat dest2 specially. This makes IPsec processing * much easier. The goal here is to make mprev point the * mbuf prior to dest2. * * Result: IPv6 dest2 payload. * m and mprev will point to IPv6 header. */ if (exthdrs.ip6e_dest2) { if (!hdrsplit) panic("%s:%d: assumption failed: " "hdr not split: hdrsplit %d exthdrs %p", __func__, __LINE__, hdrsplit, &exthdrs); exthdrs.ip6e_dest2->m_next = m->m_next; m->m_next = exthdrs.ip6e_dest2; *mtod(exthdrs.ip6e_dest2, u_char *) = ip6->ip6_nxt; ip6->ip6_nxt = IPPROTO_DSTOPTS; } /* * Result: IPv6 hbh dest1 rthdr dest2 payload. * m will point to IPv6 header. mprev will point to the * extension header prior to dest2 (rthdr in the above case). */ MAKE_CHAIN(exthdrs.ip6e_hbh, mprev, nexthdrp, IPPROTO_HOPOPTS); MAKE_CHAIN(exthdrs.ip6e_dest1, mprev, nexthdrp, IPPROTO_DSTOPTS); MAKE_CHAIN(exthdrs.ip6e_rthdr, mprev, nexthdrp, IPPROTO_ROUTING); } IP6STAT_INC(ip6s_localout); /* Route packet. */ ro_pmtu = ro; if (opt && opt->ip6po_rthdr) ro = &opt->ip6po_route; if (ro != NULL) dst = (struct sockaddr_in6 *)&ro->ro_dst; else dst = &sin6; fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m); again: /* * If specified, try to fill in the traffic class field. * Do not override if a non-zero value is already set. * We check the diffserv field and the ECN field separately. */ if (opt && opt->ip6po_tclass >= 0) { int mask = 0; if ((ip6->ip6_flow & htonl(0xfc << 20)) == 0) mask |= 0xfc; if ((ip6->ip6_flow & htonl(0x03 << 20)) == 0) mask |= 0x03; if (mask != 0) ip6->ip6_flow |= htonl((opt->ip6po_tclass & mask) << 20); } /* Fill in or override the hop limit field, if necessary. */ if (opt && opt->ip6po_hlim != -1) ip6->ip6_hlim = opt->ip6po_hlim & 0xff; else if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { if (im6o != NULL) ip6->ip6_hlim = im6o->im6o_multicast_hlim; else ip6->ip6_hlim = V_ip6_defmcasthlim; } - if (ro == NULL || ro->ro_rt == NULL) { + if (ro == NULL || ro->ro_nh == NULL) { bzero(dst, sizeof(*dst)); dst->sin6_family = AF_INET6; dst->sin6_len = sizeof(*dst); dst->sin6_addr = ip6->ip6_dst; } /* * Validate route against routing table changes. * Make sure that the address family is set in route. */ - rt = NULL; + nh = NULL; ifp = NULL; mtu = 0; if (ro != NULL) { - if (ro->ro_rt != NULL && inp != NULL) { + if (ro->ro_nh != NULL && inp != NULL) { ro->ro_dst.sin6_family = AF_INET6; /* XXX KASSERT? */ - RT_VALIDATE((struct route *)ro, &inp->inp_rt_cookie, + NH_VALIDATE((struct route *)ro, &inp->inp_rt_cookie, fibnum); } - if (ro->ro_rt != NULL && fwd_tag == NULL && - ((ro->ro_rt->rt_flags & RTF_UP) == 0 || - ro->ro_rt->rt_ifp == NULL || - !RT_LINK_IS_UP(ro->ro_rt->rt_ifp) || + if (ro->ro_nh != NULL && fwd_tag == NULL && + (!NH_IS_VALID(ro->ro_nh) || ro->ro_dst.sin6_family != AF_INET6 || !IN6_ARE_ADDR_EQUAL(&ro->ro_dst.sin6_addr, &ip6->ip6_dst))) RO_INVALIDATE_CACHE(ro); - if (ro->ro_rt != NULL && fwd_tag == NULL && - (ro->ro_rt->rt_flags & RTF_UP) && + if (ro->ro_nh != NULL && fwd_tag == NULL && ro->ro_dst.sin6_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&ro->ro_dst.sin6_addr, &ip6->ip6_dst)) { - rt = ro->ro_rt; - ifp = ro->ro_rt->rt_ifp; + nh = ro->ro_nh; + ifp = nh->nh_ifp; } else { if (ro->ro_lle) LLE_FREE(ro->ro_lle); /* zeros ro_lle */ ro->ro_lle = NULL; if (fwd_tag == NULL) { bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = ip6->ip6_dst; } error = in6_selectroute(&dst_sa, opt, im6o, ro, &ifp, - &rt, fibnum); + &nh, fibnum, m->m_pkthdr.flowid); if (error != 0) { IP6STAT_INC(ip6s_noroute); if (ifp != NULL) in6_ifstat_inc(ifp, ifs6_out_discard); goto bad; } if (ifp != NULL) mtu = ifp->if_mtu; } - if (rt == NULL) { + if (nh == NULL) { /* - * If in6_selectroute() does not return a route entry + * If in6_selectroute() does not return a nexthop * dst may not have been updated. */ *dst = dst_sa; /* XXX */ } else { - if (rt->rt_flags & RTF_HOST) - mtu = rt->rt_mtu; - ia = (struct in6_ifaddr *)(rt->rt_ifa); - counter_u64_add(rt->rt_pksent, 1); + if (nh->nh_flags & NHF_HOST) + mtu = nh->nh_mtu; + ia = (struct in6_ifaddr *)(nh->nh_ifa); + counter_u64_add(nh->nh_pksent, 1); } } else { struct nhop6_extended nh6; struct in6_addr kdst; uint32_t scopeid; if (fwd_tag == NULL) { bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = ip6->ip6_dst; } if (IN6_IS_ADDR_MULTICAST(&dst_sa.sin6_addr) && im6o != NULL && (ifp = im6o->im6o_multicast_ifp) != NULL) { /* We do not need a route lookup. */ *dst = dst_sa; /* XXX */ goto nonh6lookup; } in6_splitscope(&dst_sa.sin6_addr, &kdst, &scopeid); if (IN6_IS_ADDR_MC_LINKLOCAL(&dst_sa.sin6_addr) || IN6_IS_ADDR_MC_NODELOCAL(&dst_sa.sin6_addr)) { if (scopeid > 0) { ifp = in6_getlinkifnet(scopeid); *dst = dst_sa; /* XXX */ goto nonh6lookup; } } error = fib6_lookup_nh_ext(fibnum, &kdst, scopeid, NHR_REF, 0, &nh6); if (error != 0) { IP6STAT_INC(ip6s_noroute); /* No ifp in6_ifstat_inc(ifp, ifs6_out_discard); */ error = EHOSTUNREACH;; goto bad; } ifp = nh6.nh_ifp; mtu = nh6.nh_mtu; dst->sin6_addr = nh6.nh_addr; ia = nh6.nh_ia; fib6_free_nh_ext(fibnum, &nh6); nonh6lookup: ; } - /* Then rt (for unicast) and ifp must be non-NULL valid values. */ + /* Then nh (for unicast) and ifp must be non-NULL valid values. */ if ((flags & IPV6_FORWARDING) == 0) { /* XXX: the FORWARDING flag can be set for mrouting. */ in6_ifstat_inc(ifp, ifs6_out_request); } /* Setup data structures for scope ID checks. */ src0 = ip6->ip6_src; bzero(&src_sa, sizeof(src_sa)); src_sa.sin6_family = AF_INET6; src_sa.sin6_len = sizeof(src_sa); src_sa.sin6_addr = ip6->ip6_src; dst0 = ip6->ip6_dst; /* Re-initialize to be sure. */ bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = ip6->ip6_dst; /* Check for valid scope ID. */ if (in6_setscope(&src0, ifp, &zone) == 0 && sa6_recoverscope(&src_sa) == 0 && zone == src_sa.sin6_scope_id && in6_setscope(&dst0, ifp, &zone) == 0 && sa6_recoverscope(&dst_sa) == 0 && zone == dst_sa.sin6_scope_id) { /* * The outgoing interface is in the zone of the source * and destination addresses. * * Because the loopback interface cannot receive * packets with a different scope ID than its own, * there is a trick to pretend the outgoing packet * was received by the real network interface, by * setting "origifp" different from "ifp". This is * only allowed when "ifp" is a loopback network * interface. Refer to code in nd6_output_ifp() for * more details. */ origifp = ifp; /* * We should use ia_ifp to support the case of sending * packets to an address of our own. */ if (ia != NULL && ia->ia_ifp) ifp = ia->ia_ifp; } else if ((ifp->if_flags & IFF_LOOPBACK) == 0 || sa6_recoverscope(&src_sa) != 0 || sa6_recoverscope(&dst_sa) != 0 || dst_sa.sin6_scope_id == 0 || (src_sa.sin6_scope_id != 0 && src_sa.sin6_scope_id != dst_sa.sin6_scope_id) || (origifp = ifnet_byindex(dst_sa.sin6_scope_id)) == NULL) { /* * If the destination network interface is not a * loopback interface, or the destination network * address has no scope ID, or the source address has * a scope ID set which is different from the * destination address one, or there is no network * interface representing this scope ID, the address * pair is considered invalid. */ IP6STAT_INC(ip6s_badscope); in6_ifstat_inc(ifp, ifs6_out_discard); if (error == 0) error = EHOSTUNREACH; /* XXX */ goto bad; } /* All scope ID checks are successful. */ - if (rt && !IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { - if (opt && opt->ip6po_nextroute.ro_rt) { + if (nh && !IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { + if (opt && opt->ip6po_nextroute.ro_nh) { /* * The nexthop is explicitly specified by the * application. We assume the next hop is an IPv6 * address. */ dst = (struct sockaddr_in6 *)opt->ip6po_nexthop; } - else if ((rt->rt_flags & RTF_GATEWAY)) - dst = (struct sockaddr_in6 *)rt->rt_gateway; + else if ((nh->nh_flags & NHF_GATEWAY)) + dst = &nh->gw6_sa; } if (!IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { m->m_flags &= ~(M_BCAST | M_MCAST); /* Just in case. */ } else { m->m_flags = (m->m_flags & ~M_BCAST) | M_MCAST; in6_ifstat_inc(ifp, ifs6_out_mcast); /* Confirm that the outgoing interface supports multicast. */ if (!(ifp->if_flags & IFF_MULTICAST)) { IP6STAT_INC(ip6s_noroute); in6_ifstat_inc(ifp, ifs6_out_discard); error = ENETUNREACH; goto bad; } if ((im6o == NULL && in6_mcast_loop) || (im6o && im6o->im6o_multicast_loop)) { /* * Loop back multicast datagram if not expressly * forbidden to do so, even if we have not joined * the address; protocols will filter it later, * thus deferring a hash lookup and lock acquisition * at the expense of an m_copym(). */ ip6_mloopback(ifp, m); } else { /* * If we are acting as a multicast router, perform * multicast forwarding as if the packet had just * arrived on the interface to which we are about * to send. The multicast forwarding function * recursively calls this function, using the * IPV6_FORWARDING flag to prevent infinite recursion. * * Multicasts that are looped back by ip6_mloopback(), * above, will be forwarded by the ip6_input() routine, * if necessary. */ if (V_ip6_mrouter && (flags & IPV6_FORWARDING) == 0) { /* * XXX: ip6_mforward expects that rcvif is NULL * when it is called from the originating path. * However, it may not always be the case. */ m->m_pkthdr.rcvif = NULL; if (ip6_mforward(ip6, ifp, m) != 0) { m_freem(m); goto done; } } } /* * Multicasts with a hoplimit of zero may be looped back, * above, but must not be transmitted on a network. * Also, multicasts addressed to the loopback interface * are not sent -- the above call to ip6_mloopback() will * loop back a copy if this host actually belongs to the * destination group on the loopback interface. */ if (ip6->ip6_hlim == 0 || (ifp->if_flags & IFF_LOOPBACK) || IN6_IS_ADDR_MC_INTFACELOCAL(&ip6->ip6_dst)) { m_freem(m); goto done; } } /* * Fill the outgoing inteface to tell the upper layer * to increment per-interface statistics. */ if (ifpp) *ifpp = ifp; /* Determine path MTU. */ if ((error = ip6_getpmtu(ro_pmtu, ro != ro_pmtu, ifp, &ip6->ip6_dst, &mtu, &alwaysfrag, fibnum, *nexthdrp)) != 0) goto bad; KASSERT(mtu > 0, ("%s:%d: mtu %ld, ro_pmtu %p ro %p ifp %p " "alwaysfrag %d fibnum %u\n", __func__, __LINE__, mtu, ro_pmtu, ro, ifp, alwaysfrag, fibnum)); /* * The caller of this function may specify to use the minimum MTU * in some cases. * An advanced API option (IPV6_USE_MIN_MTU) can also override MTU * setting. The logic is a bit complicated; by default, unicast * packets will follow path MTU while multicast packets will be sent at * the minimum MTU. If IP6PO_MINMTU_ALL is specified, all packets * including unicast ones will be sent at the minimum MTU. Multicast * packets will always be sent at the minimum MTU unless * IP6PO_MINMTU_DISABLE is explicitly specified. * See RFC 3542 for more details. */ if (mtu > IPV6_MMTU) { if ((flags & IPV6_MINMTU)) mtu = IPV6_MMTU; else if (opt && opt->ip6po_minmtu == IP6PO_MINMTU_ALL) mtu = IPV6_MMTU; else if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) && (opt == NULL || opt->ip6po_minmtu != IP6PO_MINMTU_DISABLE)) { mtu = IPV6_MMTU; } } /* * Clear embedded scope identifiers if necessary. * in6_clearscope() will touch the addresses only when necessary. */ in6_clearscope(&ip6->ip6_src); in6_clearscope(&ip6->ip6_dst); /* * If the outgoing packet contains a hop-by-hop options header, * it must be examined and processed even by the source node. * (RFC 2460, section 4.) */ if (exthdrs.ip6e_hbh) { struct ip6_hbh *hbh = mtod(exthdrs.ip6e_hbh, struct ip6_hbh *); u_int32_t dummy; /* XXX unused */ u_int32_t plen = 0; /* XXX: ip6_process will check the value */ #ifdef DIAGNOSTIC if ((hbh->ip6h_len + 1) << 3 > exthdrs.ip6e_hbh->m_len) panic("ip6e_hbh is not contiguous"); #endif /* * XXX: if we have to send an ICMPv6 error to the sender, * we need the M_LOOP flag since icmp6_error() expects * the IPv6 and the hop-by-hop options header are * contiguous unless the flag is set. */ m->m_flags |= M_LOOP; m->m_pkthdr.rcvif = ifp; if (ip6_process_hopopts(m, (u_int8_t *)(hbh + 1), ((hbh->ip6h_len + 1) << 3) - sizeof(struct ip6_hbh), &dummy, &plen) < 0) { /* m was already freed at this point. */ error = EINVAL;/* better error? */ goto done; } m->m_flags &= ~M_LOOP; /* XXX */ m->m_pkthdr.rcvif = NULL; } /* Jump over all PFIL processing if hooks are not active. */ if (!PFIL_HOOKED_OUT(V_inet6_pfil_head)) goto passout; odst = ip6->ip6_dst; /* Run through list of hooks for output packets. */ switch (pfil_run_hooks(V_inet6_pfil_head, &m, ifp, PFIL_OUT, inp)) { case PFIL_PASS: ip6 = mtod(m, struct ip6_hdr *); break; case PFIL_DROPPED: error = EACCES; /* FALLTHROUGH */ case PFIL_CONSUMED: goto done; } needfiblookup = 0; /* See if destination IP address was changed by packet filter. */ if (!IN6_ARE_ADDR_EQUAL(&odst, &ip6->ip6_dst)) { m->m_flags |= M_SKIP_FIREWALL; /* If destination is now ourself drop to ip6_input(). */ if (in6_localip(&ip6->ip6_dst)) { m->m_flags |= M_FASTFWD_OURS; if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif error = netisr_queue(NETISR_IPV6, m); goto done; } else { if (ro != NULL) RO_INVALIDATE_CACHE(ro); needfiblookup = 1; /* Redo the routing table lookup. */ } } /* See if fib was changed by packet filter. */ if (fibnum != M_GETFIB(m)) { m->m_flags |= M_SKIP_FIREWALL; fibnum = M_GETFIB(m); if (ro != NULL) RO_INVALIDATE_CACHE(ro); needfiblookup = 1; } if (needfiblookup) goto again; /* See if local, if yes, send it to netisr. */ if (m->m_flags & M_FASTFWD_OURS) { if (m->m_pkthdr.rcvif == NULL) m->m_pkthdr.rcvif = V_loif; if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; #endif error = netisr_queue(NETISR_IPV6, m); goto done; } /* Or forward to some other address? */ if ((m->m_flags & M_IP6_NEXTHOP) && (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { if (ro != NULL) dst = (struct sockaddr_in6 *)&ro->ro_dst; else dst = &sin6; bcopy((fwd_tag+1), &dst_sa, sizeof(struct sockaddr_in6)); m->m_flags |= M_SKIP_FIREWALL; m->m_flags &= ~M_IP6_NEXTHOP; m_tag_delete(m, fwd_tag); goto again; } passout: /* * Send the packet to the outgoing interface. * If necessary, do IPv6 fragmentation before sending. * * The logic here is rather complex: * 1: normal case (dontfrag == 0, alwaysfrag == 0) * 1-a: send as is if tlen <= path mtu * 1-b: fragment if tlen > path mtu * * 2: if user asks us not to fragment (dontfrag == 1) * 2-a: send as is if tlen <= interface mtu * 2-b: error if tlen > interface mtu * * 3: if we always need to attach fragment header (alwaysfrag == 1) * always fragment * * 4: if dontfrag == 1 && alwaysfrag == 1 * error, as we cannot handle this conflicting request. */ sw_csum = m->m_pkthdr.csum_flags; if (!hdrsplit) { tso = ((sw_csum & ifp->if_hwassist & CSUM_TSO) != 0) ? 1 : 0; sw_csum &= ~ifp->if_hwassist; } else tso = 0; /* * If we added extension headers, we will not do TSO and calculate the * checksums ourselves for now. * XXX-BZ Need a framework to know when the NIC can handle it, even * with ext. hdrs. */ error = ip6_output_delayed_csum(m, ifp, sw_csum, plen, optlen, false); if (error != 0) goto bad; /* XXX-BZ m->m_pkthdr.csum_flags &= ~ifp->if_hwassist; */ tlen = m->m_pkthdr.len; if ((opt && (opt->ip6po_flags & IP6PO_DONTFRAG)) || tso) dontfrag = 1; else dontfrag = 0; if (dontfrag && alwaysfrag) { /* Case 4. */ /* Conflicting request - can't transmit. */ error = EMSGSIZE; goto bad; } if (dontfrag && tlen > IN6_LINKMTU(ifp) && !tso) { /* Case 2-b. */ /* * Even if the DONTFRAG option is specified, we cannot send the * packet when the data length is larger than the MTU of the * outgoing interface. * Notify the error by sending IPV6_PATHMTU ancillary data if * application wanted to know the MTU value. Also return an * error code (this is not described in the API spec). */ if (inp != NULL) ip6_notify_pmtu(inp, &dst_sa, (u_int32_t)mtu); error = EMSGSIZE; goto bad; } /* Transmit packet without fragmentation. */ if (dontfrag || (!alwaysfrag && tlen <= mtu)) { /* Cases 1-a and 2-a. */ struct in6_ifaddr *ia6; ip6 = mtod(m, struct ip6_hdr *); ia6 = in6_ifawithifp(ifp, &ip6->ip6_src); if (ia6) { /* Record statistics for this interface address. */ counter_u64_add(ia6->ia_ifa.ifa_opackets, 1); counter_u64_add(ia6->ia_ifa.ifa_obytes, m->m_pkthdr.len); ifa_free(&ia6->ia_ifa); } error = ip6_output_send(inp, ifp, origifp, m, dst, ro); goto done; } /* Try to fragment the packet. Cases 1-b and 3. */ if (mtu < IPV6_MMTU) { /* Path MTU cannot be less than IPV6_MMTU. */ error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } else if (ip6->ip6_plen == 0) { /* Jumbo payload cannot be fragmented. */ error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } else { u_char nextproto; /* * Too large for the destination or interface; * fragment if possible. * Must be able to put at least 8 bytes per fragment. */ if (mtu > IPV6_MAXPACKET) mtu = IPV6_MAXPACKET; len = (mtu - unfragpartlen - sizeof(struct ip6_frag)) & ~7; if (len < 8) { error = EMSGSIZE; in6_ifstat_inc(ifp, ifs6_out_fragfail); goto bad; } /* * If the interface will not calculate checksums on * fragmented packets, then do it here. * XXX-BZ handle the hw offloading case. Need flags. */ error = ip6_output_delayed_csum(m, ifp, m->m_pkthdr.csum_flags, plen, optlen, true); if (error != 0) goto bad; /* * Change the next header field of the last header in the * unfragmentable part. */ if (exthdrs.ip6e_rthdr) { nextproto = *mtod(exthdrs.ip6e_rthdr, u_char *); *mtod(exthdrs.ip6e_rthdr, u_char *) = IPPROTO_FRAGMENT; } else if (exthdrs.ip6e_dest1) { nextproto = *mtod(exthdrs.ip6e_dest1, u_char *); *mtod(exthdrs.ip6e_dest1, u_char *) = IPPROTO_FRAGMENT; } else if (exthdrs.ip6e_hbh) { nextproto = *mtod(exthdrs.ip6e_hbh, u_char *); *mtod(exthdrs.ip6e_hbh, u_char *) = IPPROTO_FRAGMENT; } else { ip6 = mtod(m, struct ip6_hdr *); nextproto = ip6->ip6_nxt; ip6->ip6_nxt = IPPROTO_FRAGMENT; } /* * Loop through length of segment after first fragment, * make new header and copy data of each part and link onto * chain. */ m0 = m; id = htonl(ip6_randomid()); error = ip6_fragment(ifp, m, unfragpartlen, nextproto,len, id); if (error != 0) goto sendorfree; in6_ifstat_inc(ifp, ifs6_out_fragok); } /* Remove leading garbage. */ sendorfree: m = m0->m_nextpkt; m0->m_nextpkt = 0; m_freem(m0); for (; m; m = m0) { m0 = m->m_nextpkt; m->m_nextpkt = 0; if (error == 0) { /* Record statistics for this interface address. */ if (ia) { counter_u64_add(ia->ia_ifa.ifa_opackets, 1); counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len); } error = ip6_output_send(inp, ifp, origifp, m, dst, ro); } else m_freem(m); } if (error == 0) IP6STAT_INC(ip6s_fragmented); done: return (error); freehdrs: m_freem(exthdrs.ip6e_hbh); /* m_freem() checks if mbuf is NULL. */ m_freem(exthdrs.ip6e_dest1); m_freem(exthdrs.ip6e_rthdr); m_freem(exthdrs.ip6e_dest2); /* FALLTHROUGH */ bad: if (m) m_freem(m); goto done; } static int ip6_copyexthdr(struct mbuf **mp, caddr_t hdr, int hlen) { struct mbuf *m; if (hlen > MCLBYTES) return (ENOBUFS); /* XXX */ if (hlen > MLEN) m = m_getcl(M_NOWAIT, MT_DATA, 0); else m = m_get(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); m->m_len = hlen; if (hdr) bcopy(hdr, mtod(m, caddr_t), hlen); *mp = m; return (0); } /* * Insert jumbo payload option. */ static int ip6_insert_jumboopt(struct ip6_exthdrs *exthdrs, u_int32_t plen) { struct mbuf *mopt; u_char *optbuf; u_int32_t v; #define JUMBOOPTLEN 8 /* length of jumbo payload option and padding */ /* * If there is no hop-by-hop options header, allocate new one. * If there is one but it doesn't have enough space to store the * jumbo payload option, allocate a cluster to store the whole options. * Otherwise, use it to store the options. */ if (exthdrs->ip6e_hbh == NULL) { mopt = m_get(M_NOWAIT, MT_DATA); if (mopt == NULL) return (ENOBUFS); mopt->m_len = JUMBOOPTLEN; optbuf = mtod(mopt, u_char *); optbuf[1] = 0; /* = ((JUMBOOPTLEN) >> 3) - 1 */ exthdrs->ip6e_hbh = mopt; } else { struct ip6_hbh *hbh; mopt = exthdrs->ip6e_hbh; if (M_TRAILINGSPACE(mopt) < JUMBOOPTLEN) { /* * XXX assumption: * - exthdrs->ip6e_hbh is not referenced from places * other than exthdrs. * - exthdrs->ip6e_hbh is not an mbuf chain. */ int oldoptlen = mopt->m_len; struct mbuf *n; /* * XXX: give up if the whole (new) hbh header does * not fit even in an mbuf cluster. */ if (oldoptlen + JUMBOOPTLEN > MCLBYTES) return (ENOBUFS); /* * As a consequence, we must always prepare a cluster * at this point. */ n = m_getcl(M_NOWAIT, MT_DATA, 0); if (n == NULL) return (ENOBUFS); n->m_len = oldoptlen + JUMBOOPTLEN; bcopy(mtod(mopt, caddr_t), mtod(n, caddr_t), oldoptlen); optbuf = mtod(n, caddr_t) + oldoptlen; m_freem(mopt); mopt = exthdrs->ip6e_hbh = n; } else { optbuf = mtod(mopt, u_char *) + mopt->m_len; mopt->m_len += JUMBOOPTLEN; } optbuf[0] = IP6OPT_PADN; optbuf[1] = 1; /* * Adjust the header length according to the pad and * the jumbo payload option. */ hbh = mtod(mopt, struct ip6_hbh *); hbh->ip6h_len += (JUMBOOPTLEN >> 3); } /* fill in the option. */ optbuf[2] = IP6OPT_JUMBO; optbuf[3] = 4; v = (u_int32_t)htonl(plen + JUMBOOPTLEN); bcopy(&v, &optbuf[4], sizeof(u_int32_t)); /* finally, adjust the packet header length */ exthdrs->ip6e_ip6->m_pkthdr.len += JUMBOOPTLEN; return (0); #undef JUMBOOPTLEN } /* * Insert fragment header and copy unfragmentable header portions. */ static int ip6_insertfraghdr(struct mbuf *m0, struct mbuf *m, int hlen, struct ip6_frag **frghdrp) { struct mbuf *n, *mlast; if (hlen > sizeof(struct ip6_hdr)) { n = m_copym(m0, sizeof(struct ip6_hdr), hlen - sizeof(struct ip6_hdr), M_NOWAIT); if (n == NULL) return (ENOBUFS); m->m_next = n; } else n = m; /* Search for the last mbuf of unfragmentable part. */ for (mlast = n; mlast->m_next; mlast = mlast->m_next) ; if (M_WRITABLE(mlast) && M_TRAILINGSPACE(mlast) >= sizeof(struct ip6_frag)) { /* use the trailing space of the last mbuf for the fragment hdr */ *frghdrp = (struct ip6_frag *)(mtod(mlast, caddr_t) + mlast->m_len); mlast->m_len += sizeof(struct ip6_frag); m->m_pkthdr.len += sizeof(struct ip6_frag); } else { /* allocate a new mbuf for the fragment header */ struct mbuf *mfrg; mfrg = m_get(M_NOWAIT, MT_DATA); if (mfrg == NULL) return (ENOBUFS); mfrg->m_len = sizeof(struct ip6_frag); *frghdrp = mtod(mfrg, struct ip6_frag *); mlast->m_next = mfrg; } return (0); } /* * Calculates IPv6 path mtu for destination @dst. * Resulting MTU is stored in @mtup. * * Returns 0 on success. */ static int ip6_getpmtu_ctl(u_int fibnum, const struct in6_addr *dst, u_long *mtup) { struct nhop6_extended nh6; struct in6_addr kdst; uint32_t scopeid; struct ifnet *ifp; u_long mtu; int error; in6_splitscope(dst, &kdst, &scopeid); if (fib6_lookup_nh_ext(fibnum, &kdst, scopeid, NHR_REF, 0, &nh6) != 0) return (EHOSTUNREACH); ifp = nh6.nh_ifp; mtu = nh6.nh_mtu; error = ip6_calcmtu(ifp, dst, mtu, mtup, NULL, 0); fib6_free_nh_ext(fibnum, &nh6); return (error); } /* * Calculates IPv6 path MTU for @dst based on transmit @ifp, * and cached data in @ro_pmtu. * MTU from (successful) route lookup is saved (along with dst) * inside @ro_pmtu to avoid subsequent route lookups after packet * filter processing. * * Stores mtu and always-frag value into @mtup and @alwaysfragp. * Returns 0 on success. */ static int ip6_getpmtu(struct route_in6 *ro_pmtu, int do_lookup, struct ifnet *ifp, const struct in6_addr *dst, u_long *mtup, int *alwaysfragp, u_int fibnum, u_int proto) { struct nhop6_basic nh6; struct in6_addr kdst; uint32_t scopeid; struct sockaddr_in6 *sa6_dst, sin6; u_long mtu; mtu = 0; if (ro_pmtu == NULL || do_lookup) { /* * Here ro_pmtu has final destination address, while * ro might represent immediate destination. * Use ro_pmtu destination since mtu might differ. */ if (ro_pmtu != NULL) { sa6_dst = (struct sockaddr_in6 *)&ro_pmtu->ro_dst; if (!IN6_ARE_ADDR_EQUAL(&sa6_dst->sin6_addr, dst)) ro_pmtu->ro_mtu = 0; } else sa6_dst = &sin6; if (ro_pmtu == NULL || ro_pmtu->ro_mtu == 0) { bzero(sa6_dst, sizeof(*sa6_dst)); sa6_dst->sin6_family = AF_INET6; sa6_dst->sin6_len = sizeof(struct sockaddr_in6); sa6_dst->sin6_addr = *dst; in6_splitscope(dst, &kdst, &scopeid); if (fib6_lookup_nh_basic(fibnum, &kdst, scopeid, 0, 0, &nh6) == 0) { mtu = nh6.nh_mtu; if (ro_pmtu != NULL) ro_pmtu->ro_mtu = mtu; } } else mtu = ro_pmtu->ro_mtu; } - if (ro_pmtu != NULL && ro_pmtu->ro_rt != NULL) - mtu = ro_pmtu->ro_rt->rt_mtu; + if (ro_pmtu != NULL && ro_pmtu->ro_nh != NULL) + mtu = ro_pmtu->ro_nh->nh_mtu; return (ip6_calcmtu(ifp, dst, mtu, mtup, alwaysfragp, proto)); } /* * Calculate MTU based on transmit @ifp, route mtu @rt_mtu and * hostcache data for @dst. * Stores mtu and always-frag value into @mtup and @alwaysfragp. * * Returns 0 on success. */ static int ip6_calcmtu(struct ifnet *ifp, const struct in6_addr *dst, u_long rt_mtu, u_long *mtup, int *alwaysfragp, u_int proto) { u_long mtu = 0; int alwaysfrag = 0; int error = 0; if (rt_mtu > 0) { u_int32_t ifmtu; struct in_conninfo inc; bzero(&inc, sizeof(inc)); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; ifmtu = IN6_LINKMTU(ifp); /* TCP is known to react to pmtu changes so skip hc */ if (proto != IPPROTO_TCP) mtu = tcp_hc_getmtu(&inc); if (mtu) mtu = min(mtu, rt_mtu); else mtu = rt_mtu; if (mtu == 0) mtu = ifmtu; else if (mtu < IPV6_MMTU) { /* * RFC2460 section 5, last paragraph: * if we record ICMPv6 too big message with * mtu < IPV6_MMTU, transmit packets sized IPV6_MMTU * or smaller, with framgent header attached. * (fragment header is needed regardless from the * packet size, for translators to identify packets) */ alwaysfrag = 1; mtu = IPV6_MMTU; } } else if (ifp) { mtu = IN6_LINKMTU(ifp); } else error = EHOSTUNREACH; /* XXX */ *mtup = mtu; if (alwaysfragp) *alwaysfragp = alwaysfrag; return (error); } /* * IP6 socket option processing. */ int ip6_ctloutput(struct socket *so, struct sockopt *sopt) { int optdatalen, uproto; void *optdata; struct inpcb *inp = sotoinpcb(so); int error, optval; int level, op, optname; int optlen; struct thread *td; #ifdef RSS uint32_t rss_bucket; int retval; #endif /* * Don't use more than a quarter of mbuf clusters. N.B.: * nmbclusters is an int, but nmbclusters * MCLBYTES may overflow * on LP64 architectures, so cast to u_long to avoid undefined * behavior. ILP32 architectures cannot have nmbclusters * large enough to overflow for other reasons. */ #define IPV6_PKTOPTIONS_MBUF_LIMIT ((u_long)nmbclusters * MCLBYTES / 4) level = sopt->sopt_level; op = sopt->sopt_dir; optname = sopt->sopt_name; optlen = sopt->sopt_valsize; td = sopt->sopt_td; error = 0; optval = 0; uproto = (int)so->so_proto->pr_protocol; if (level != IPPROTO_IPV6) { error = EINVAL; if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_dir == SOPT_SET) { switch (sopt->sopt_name) { case SO_REUSEADDR: INP_WLOCK(inp); if ((so->so_options & SO_REUSEADDR) != 0) inp->inp_flags2 |= INP_REUSEADDR; else inp->inp_flags2 &= ~INP_REUSEADDR; INP_WUNLOCK(inp); error = 0; break; case SO_REUSEPORT: INP_WLOCK(inp); if ((so->so_options & SO_REUSEPORT) != 0) inp->inp_flags2 |= INP_REUSEPORT; else inp->inp_flags2 &= ~INP_REUSEPORT; INP_WUNLOCK(inp); error = 0; break; case SO_REUSEPORT_LB: INP_WLOCK(inp); if ((so->so_options & SO_REUSEPORT_LB) != 0) inp->inp_flags2 |= INP_REUSEPORT_LB; else inp->inp_flags2 &= ~INP_REUSEPORT_LB; INP_WUNLOCK(inp); error = 0; break; case SO_SETFIB: INP_WLOCK(inp); inp->inp_inc.inc_fibnum = so->so_fibnum; INP_WUNLOCK(inp); error = 0; break; case SO_MAX_PACING_RATE: #ifdef RATELIMIT INP_WLOCK(inp); inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; INP_WUNLOCK(inp); error = 0; #else error = EOPNOTSUPP; #endif break; default: break; } } } else { /* level == IPPROTO_IPV6 */ switch (op) { case SOPT_SET: switch (optname) { case IPV6_2292PKTOPTIONS: #ifdef IPV6_PKTOPTIONS case IPV6_PKTOPTIONS: #endif { struct mbuf *m; if (optlen > IPV6_PKTOPTIONS_MBUF_LIMIT) { printf("ip6_ctloutput: mbuf limit hit\n"); error = ENOBUFS; break; } error = soopt_getm(sopt, &m); /* XXX */ if (error != 0) break; error = soopt_mcopyin(sopt, m); /* XXX */ if (error != 0) break; INP_WLOCK(inp); error = ip6_pcbopts(&inp->in6p_outputopts, m, so, sopt); INP_WUNLOCK(inp); m_freem(m); /* XXX */ break; } /* * Use of some Hop-by-Hop options or some * Destination options, might require special * privilege. That is, normal applications * (without special privilege) might be forbidden * from setting certain options in outgoing packets, * and might never see certain options in received * packets. [RFC 2292 Section 6] * KAME specific note: * KAME prevents non-privileged users from sending or * receiving ANY hbh/dst options in order to avoid * overhead of parsing options in the kernel. */ case IPV6_RECVHOPOPTS: case IPV6_RECVDSTOPTS: case IPV6_RECVRTHDRDSTOPTS: if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) break; } /* FALLTHROUGH */ case IPV6_UNICAST_HOPS: case IPV6_HOPLIMIT: case IPV6_RECVPKTINFO: case IPV6_RECVHOPLIMIT: case IPV6_RECVRTHDR: case IPV6_RECVPATHMTU: case IPV6_RECVTCLASS: case IPV6_RECVFLOWID: #ifdef RSS case IPV6_RECVRSSBUCKETID: #endif case IPV6_V6ONLY: case IPV6_AUTOFLOWLABEL: case IPV6_ORIGDSTADDR: case IPV6_BINDANY: case IPV6_BINDMULTI: #ifdef RSS case IPV6_RSS_LISTEN_BUCKET: #endif if (optname == IPV6_BINDANY && td != NULL) { error = priv_check(td, PRIV_NETINET_BINDANY); if (error) break; } if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; switch (optname) { case IPV6_UNICAST_HOPS: if (optval < -1 || optval >= 256) error = EINVAL; else { /* -1 = kernel default */ inp->in6p_hops = optval; if ((inp->inp_vflag & INP_IPV4) != 0) inp->inp_ip_ttl = optval; } break; #define OPTSET(bit) \ do { \ INP_WLOCK(inp); \ if (optval) \ inp->inp_flags |= (bit); \ else \ inp->inp_flags &= ~(bit); \ INP_WUNLOCK(inp); \ } while (/*CONSTCOND*/ 0) #define OPTSET2292(bit) \ do { \ INP_WLOCK(inp); \ inp->inp_flags |= IN6P_RFC2292; \ if (optval) \ inp->inp_flags |= (bit); \ else \ inp->inp_flags &= ~(bit); \ INP_WUNLOCK(inp); \ } while (/*CONSTCOND*/ 0) #define OPTBIT(bit) (inp->inp_flags & (bit) ? 1 : 0) #define OPTSET2_N(bit, val) do { \ if (val) \ inp->inp_flags2 |= bit; \ else \ inp->inp_flags2 &= ~bit; \ } while (0) #define OPTSET2(bit, val) do { \ INP_WLOCK(inp); \ OPTSET2_N(bit, val); \ INP_WUNLOCK(inp); \ } while (0) #define OPTBIT2(bit) (inp->inp_flags2 & (bit) ? 1 : 0) #define OPTSET2292_EXCLUSIVE(bit) \ do { \ INP_WLOCK(inp); \ if (OPTBIT(IN6P_RFC2292)) { \ error = EINVAL; \ } else { \ if (optval) \ inp->inp_flags |= (bit); \ else \ inp->inp_flags &= ~(bit); \ } \ INP_WUNLOCK(inp); \ } while (/*CONSTCOND*/ 0) case IPV6_RECVPKTINFO: OPTSET2292_EXCLUSIVE(IN6P_PKTINFO); break; case IPV6_HOPLIMIT: { struct ip6_pktopts **optp; /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } INP_WLOCK(inp); if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_WUNLOCK(inp); return (ECONNRESET); } optp = &inp->in6p_outputopts; error = ip6_pcbopt(IPV6_HOPLIMIT, (u_char *)&optval, sizeof(optval), optp, (td != NULL) ? td->td_ucred : NULL, uproto); INP_WUNLOCK(inp); break; } case IPV6_RECVHOPLIMIT: OPTSET2292_EXCLUSIVE(IN6P_HOPLIMIT); break; case IPV6_RECVHOPOPTS: OPTSET2292_EXCLUSIVE(IN6P_HOPOPTS); break; case IPV6_RECVDSTOPTS: OPTSET2292_EXCLUSIVE(IN6P_DSTOPTS); break; case IPV6_RECVRTHDRDSTOPTS: OPTSET2292_EXCLUSIVE(IN6P_RTHDRDSTOPTS); break; case IPV6_RECVRTHDR: OPTSET2292_EXCLUSIVE(IN6P_RTHDR); break; case IPV6_RECVPATHMTU: /* * We ignore this option for TCP * sockets. * (RFC3542 leaves this case * unspecified.) */ if (uproto != IPPROTO_TCP) OPTSET(IN6P_MTU); break; case IPV6_RECVFLOWID: OPTSET2(INP_RECVFLOWID, optval); break; #ifdef RSS case IPV6_RECVRSSBUCKETID: OPTSET2(INP_RECVRSSBUCKETID, optval); break; #endif case IPV6_V6ONLY: INP_WLOCK(inp); if (inp->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { /* * The socket is already bound. */ INP_WUNLOCK(inp); error = EINVAL; break; } if (optval) { inp->inp_flags |= IN6P_IPV6_V6ONLY; inp->inp_vflag &= ~INP_IPV4; } else { inp->inp_flags &= ~IN6P_IPV6_V6ONLY; inp->inp_vflag |= INP_IPV4; } INP_WUNLOCK(inp); break; case IPV6_RECVTCLASS: /* cannot mix with RFC2292 XXX */ OPTSET2292_EXCLUSIVE(IN6P_TCLASS); break; case IPV6_AUTOFLOWLABEL: OPTSET(IN6P_AUTOFLOWLABEL); break; case IPV6_ORIGDSTADDR: OPTSET2(INP_ORIGDSTADDR, optval); break; case IPV6_BINDANY: OPTSET(INP_BINDANY); break; case IPV6_BINDMULTI: OPTSET2(INP_BINDMULTI, optval); break; #ifdef RSS case IPV6_RSS_LISTEN_BUCKET: if ((optval >= 0) && (optval < rss_getnumbuckets())) { INP_WLOCK(inp); inp->inp_rss_listen_bucket = optval; OPTSET2_N(INP_RSS_BUCKET_SET, 1); INP_WUNLOCK(inp); } else { error = EINVAL; } break; #endif } break; case IPV6_TCLASS: case IPV6_DONTFRAG: case IPV6_USE_MIN_MTU: case IPV6_PREFER_TEMPADDR: if (optlen != sizeof(optval)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; { struct ip6_pktopts **optp; INP_WLOCK(inp); if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_WUNLOCK(inp); return (ECONNRESET); } optp = &inp->in6p_outputopts; error = ip6_pcbopt(optname, (u_char *)&optval, sizeof(optval), optp, (td != NULL) ? td->td_ucred : NULL, uproto); INP_WUNLOCK(inp); break; } case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292HOPOPTS: case IPV6_2292DSTOPTS: case IPV6_2292RTHDR: /* RFC 2292 */ if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; switch (optname) { case IPV6_2292PKTINFO: OPTSET2292(IN6P_PKTINFO); break; case IPV6_2292HOPLIMIT: OPTSET2292(IN6P_HOPLIMIT); break; case IPV6_2292HOPOPTS: /* * Check super-user privilege. * See comments for IPV6_RECVHOPOPTS. */ if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) return (error); } OPTSET2292(IN6P_HOPOPTS); break; case IPV6_2292DSTOPTS: if (td != NULL) { error = priv_check(td, PRIV_NETINET_SETHDROPTS); if (error) return (error); } OPTSET2292(IN6P_DSTOPTS|IN6P_RTHDRDSTOPTS); /* XXX */ break; case IPV6_2292RTHDR: OPTSET2292(IN6P_RTHDR); break; } break; case IPV6_PKTINFO: case IPV6_HOPOPTS: case IPV6_RTHDR: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_NEXTHOP: { /* new advanced API (RFC3542) */ u_char *optbuf; u_char optbuf_storage[MCLBYTES]; int optlen; struct ip6_pktopts **optp; /* cannot mix with RFC2292 */ if (OPTBIT(IN6P_RFC2292)) { error = EINVAL; break; } /* * We only ensure valsize is not too large * here. Further validation will be done * later. */ error = sooptcopyin(sopt, optbuf_storage, sizeof(optbuf_storage), 0); if (error) break; optlen = sopt->sopt_valsize; optbuf = optbuf_storage; INP_WLOCK(inp); if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { INP_WUNLOCK(inp); return (ECONNRESET); } optp = &inp->in6p_outputopts; error = ip6_pcbopt(optname, optbuf, optlen, optp, (td != NULL) ? td->td_ucred : NULL, uproto); INP_WUNLOCK(inp); break; } #undef OPTSET case IPV6_MULTICAST_IF: case IPV6_MULTICAST_HOPS: case IPV6_MULTICAST_LOOP: case IPV6_JOIN_GROUP: case IPV6_LEAVE_GROUP: case IPV6_MSFILTER: case MCAST_BLOCK_SOURCE: case MCAST_UNBLOCK_SOURCE: case MCAST_JOIN_GROUP: case MCAST_LEAVE_GROUP: case MCAST_JOIN_SOURCE_GROUP: case MCAST_LEAVE_SOURCE_GROUP: error = ip6_setmoptions(inp, sopt); break; case IPV6_PORTRANGE: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; INP_WLOCK(inp); switch (optval) { case IPV6_PORTRANGE_DEFAULT: inp->inp_flags &= ~(INP_LOWPORT); inp->inp_flags &= ~(INP_HIGHPORT); break; case IPV6_PORTRANGE_HIGH: inp->inp_flags &= ~(INP_LOWPORT); inp->inp_flags |= INP_HIGHPORT; break; case IPV6_PORTRANGE_LOW: inp->inp_flags &= ~(INP_HIGHPORT); inp->inp_flags |= INP_LOWPORT; break; default: error = EINVAL; break; } INP_WUNLOCK(inp); break; #if defined(IPSEC) || defined(IPSEC_SUPPORT) case IPV6_IPSEC_POLICY: if (IPSEC_ENABLED(ipv6)) { error = IPSEC_PCBCTL(ipv6, inp, sopt); break; } /* FALLTHROUGH */ #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; case SOPT_GET: switch (optname) { case IPV6_2292PKTOPTIONS: #ifdef IPV6_PKTOPTIONS case IPV6_PKTOPTIONS: #endif /* * RFC3542 (effectively) deprecated the * semantics of the 2292-style pktoptions. * Since it was not reliable in nature (i.e., * applications had to expect the lack of some * information after all), it would make sense * to simplify this part by always returning * empty data. */ sopt->sopt_valsize = 0; break; case IPV6_RECVHOPOPTS: case IPV6_RECVDSTOPTS: case IPV6_RECVRTHDRDSTOPTS: case IPV6_UNICAST_HOPS: case IPV6_RECVPKTINFO: case IPV6_RECVHOPLIMIT: case IPV6_RECVRTHDR: case IPV6_RECVPATHMTU: case IPV6_V6ONLY: case IPV6_PORTRANGE: case IPV6_RECVTCLASS: case IPV6_AUTOFLOWLABEL: case IPV6_BINDANY: case IPV6_FLOWID: case IPV6_FLOWTYPE: case IPV6_RECVFLOWID: #ifdef RSS case IPV6_RSSBUCKETID: case IPV6_RECVRSSBUCKETID: #endif case IPV6_BINDMULTI: switch (optname) { case IPV6_RECVHOPOPTS: optval = OPTBIT(IN6P_HOPOPTS); break; case IPV6_RECVDSTOPTS: optval = OPTBIT(IN6P_DSTOPTS); break; case IPV6_RECVRTHDRDSTOPTS: optval = OPTBIT(IN6P_RTHDRDSTOPTS); break; case IPV6_UNICAST_HOPS: optval = inp->in6p_hops; break; case IPV6_RECVPKTINFO: optval = OPTBIT(IN6P_PKTINFO); break; case IPV6_RECVHOPLIMIT: optval = OPTBIT(IN6P_HOPLIMIT); break; case IPV6_RECVRTHDR: optval = OPTBIT(IN6P_RTHDR); break; case IPV6_RECVPATHMTU: optval = OPTBIT(IN6P_MTU); break; case IPV6_V6ONLY: optval = OPTBIT(IN6P_IPV6_V6ONLY); break; case IPV6_PORTRANGE: { int flags; flags = inp->inp_flags; if (flags & INP_HIGHPORT) optval = IPV6_PORTRANGE_HIGH; else if (flags & INP_LOWPORT) optval = IPV6_PORTRANGE_LOW; else optval = 0; break; } case IPV6_RECVTCLASS: optval = OPTBIT(IN6P_TCLASS); break; case IPV6_AUTOFLOWLABEL: optval = OPTBIT(IN6P_AUTOFLOWLABEL); break; case IPV6_ORIGDSTADDR: optval = OPTBIT2(INP_ORIGDSTADDR); break; case IPV6_BINDANY: optval = OPTBIT(INP_BINDANY); break; case IPV6_FLOWID: optval = inp->inp_flowid; break; case IPV6_FLOWTYPE: optval = inp->inp_flowtype; break; case IPV6_RECVFLOWID: optval = OPTBIT2(INP_RECVFLOWID); break; #ifdef RSS case IPV6_RSSBUCKETID: retval = rss_hash2bucket(inp->inp_flowid, inp->inp_flowtype, &rss_bucket); if (retval == 0) optval = rss_bucket; else error = EINVAL; break; case IPV6_RECVRSSBUCKETID: optval = OPTBIT2(INP_RECVRSSBUCKETID); break; #endif case IPV6_BINDMULTI: optval = OPTBIT2(INP_BINDMULTI); break; } if (error) break; error = sooptcopyout(sopt, &optval, sizeof optval); break; case IPV6_PATHMTU: { u_long pmtu = 0; struct ip6_mtuinfo mtuinfo; struct in6_addr addr; if (!(so->so_state & SS_ISCONNECTED)) return (ENOTCONN); /* * XXX: we dot not consider the case of source * routing, or optional information to specify * the outgoing interface. * Copy faddr out of inp to avoid holding lock * on inp during route lookup. */ INP_RLOCK(inp); bcopy(&inp->in6p_faddr, &addr, sizeof(addr)); INP_RUNLOCK(inp); error = ip6_getpmtu_ctl(so->so_fibnum, &addr, &pmtu); if (error) break; if (pmtu > IPV6_MAXPACKET) pmtu = IPV6_MAXPACKET; bzero(&mtuinfo, sizeof(mtuinfo)); mtuinfo.ip6m_mtu = (u_int32_t)pmtu; optdata = (void *)&mtuinfo; optdatalen = sizeof(mtuinfo); error = sooptcopyout(sopt, optdata, optdatalen); break; } case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292HOPOPTS: case IPV6_2292RTHDR: case IPV6_2292DSTOPTS: switch (optname) { case IPV6_2292PKTINFO: optval = OPTBIT(IN6P_PKTINFO); break; case IPV6_2292HOPLIMIT: optval = OPTBIT(IN6P_HOPLIMIT); break; case IPV6_2292HOPOPTS: optval = OPTBIT(IN6P_HOPOPTS); break; case IPV6_2292RTHDR: optval = OPTBIT(IN6P_RTHDR); break; case IPV6_2292DSTOPTS: optval = OPTBIT(IN6P_DSTOPTS|IN6P_RTHDRDSTOPTS); break; } error = sooptcopyout(sopt, &optval, sizeof optval); break; case IPV6_PKTINFO: case IPV6_HOPOPTS: case IPV6_RTHDR: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_NEXTHOP: case IPV6_TCLASS: case IPV6_DONTFRAG: case IPV6_USE_MIN_MTU: case IPV6_PREFER_TEMPADDR: error = ip6_getpcbopt(inp, optname, sopt); break; case IPV6_MULTICAST_IF: case IPV6_MULTICAST_HOPS: case IPV6_MULTICAST_LOOP: case IPV6_MSFILTER: error = ip6_getmoptions(inp, sopt); break; #if defined(IPSEC) || defined(IPSEC_SUPPORT) case IPV6_IPSEC_POLICY: if (IPSEC_ENABLED(ipv6)) { error = IPSEC_PCBCTL(ipv6, inp, sopt); break; } /* FALLTHROUGH */ #endif /* IPSEC */ default: error = ENOPROTOOPT; break; } break; } } return (error); } int ip6_raw_ctloutput(struct socket *so, struct sockopt *sopt) { int error = 0, optval, optlen; const int icmp6off = offsetof(struct icmp6_hdr, icmp6_cksum); struct inpcb *inp = sotoinpcb(so); int level, op, optname; level = sopt->sopt_level; op = sopt->sopt_dir; optname = sopt->sopt_name; optlen = sopt->sopt_valsize; if (level != IPPROTO_IPV6) { return (EINVAL); } switch (optname) { case IPV6_CHECKSUM: /* * For ICMPv6 sockets, no modification allowed for checksum * offset, permit "no change" values to help existing apps. * * RFC3542 says: "An attempt to set IPV6_CHECKSUM * for an ICMPv6 socket will fail." * The current behavior does not meet RFC3542. */ switch (op) { case SOPT_SET: if (optlen != sizeof(int)) { error = EINVAL; break; } error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; if (optval < -1 || (optval % 2) != 0) { /* * The API assumes non-negative even offset * values or -1 as a special value. */ error = EINVAL; } else if (so->so_proto->pr_protocol == IPPROTO_ICMPV6) { if (optval != icmp6off) error = EINVAL; } else inp->in6p_cksum = optval; break; case SOPT_GET: if (so->so_proto->pr_protocol == IPPROTO_ICMPV6) optval = icmp6off; else optval = inp->in6p_cksum; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EINVAL; break; } break; default: error = ENOPROTOOPT; break; } return (error); } /* * Set up IP6 options in pcb for insertion in output packets or * specifying behavior of outgoing packets. */ static int ip6_pcbopts(struct ip6_pktopts **pktopt, struct mbuf *m, struct socket *so, struct sockopt *sopt) { struct ip6_pktopts *opt = *pktopt; int error = 0; struct thread *td = sopt->sopt_td; /* turn off any old options. */ if (opt) { #ifdef DIAGNOSTIC if (opt->ip6po_pktinfo || opt->ip6po_nexthop || opt->ip6po_hbh || opt->ip6po_dest1 || opt->ip6po_dest2 || opt->ip6po_rhinfo.ip6po_rhi_rthdr) printf("ip6_pcbopts: all specified options are cleared.\n"); #endif ip6_clearpktopts(opt, -1); } else { opt = malloc(sizeof(*opt), M_IP6OPT, M_NOWAIT); if (opt == NULL) return (ENOMEM); } *pktopt = NULL; if (!m || m->m_len == 0) { /* * Only turning off any previous options, regardless of * whether the opt is just created or given. */ free(opt, M_IP6OPT); return (0); } /* set options specified by user. */ if ((error = ip6_setpktopts(m, opt, NULL, (td != NULL) ? td->td_ucred : NULL, so->so_proto->pr_protocol)) != 0) { ip6_clearpktopts(opt, -1); /* XXX: discard all options */ free(opt, M_IP6OPT); return (error); } *pktopt = opt; return (0); } /* * initialize ip6_pktopts. beware that there are non-zero default values in * the struct. */ void ip6_initpktopts(struct ip6_pktopts *opt) { bzero(opt, sizeof(*opt)); opt->ip6po_hlim = -1; /* -1 means default hop limit */ opt->ip6po_tclass = -1; /* -1 means default traffic class */ opt->ip6po_minmtu = IP6PO_MINMTU_MCASTONLY; opt->ip6po_prefer_tempaddr = IP6PO_TEMPADDR_SYSTEM; } static int ip6_pcbopt(int optname, u_char *buf, int len, struct ip6_pktopts **pktopt, struct ucred *cred, int uproto) { struct ip6_pktopts *opt; if (*pktopt == NULL) { *pktopt = malloc(sizeof(struct ip6_pktopts), M_IP6OPT, M_NOWAIT); if (*pktopt == NULL) return (ENOBUFS); ip6_initpktopts(*pktopt); } opt = *pktopt; return (ip6_setpktopt(optname, buf, len, opt, cred, 1, 0, uproto)); } #define GET_PKTOPT_VAR(field, lenexpr) do { \ if (pktopt && pktopt->field) { \ INP_RUNLOCK(inp); \ optdata = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK); \ malloc_optdata = true; \ INP_RLOCK(inp); \ if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \ INP_RUNLOCK(inp); \ free(optdata, M_TEMP); \ return (ECONNRESET); \ } \ pktopt = inp->in6p_outputopts; \ if (pktopt && pktopt->field) { \ optdatalen = min(lenexpr, sopt->sopt_valsize); \ bcopy(&pktopt->field, optdata, optdatalen); \ } else { \ free(optdata, M_TEMP); \ optdata = NULL; \ malloc_optdata = false; \ } \ } \ } while(0) #define GET_PKTOPT_EXT_HDR(field) GET_PKTOPT_VAR(field, \ (((struct ip6_ext *)pktopt->field)->ip6e_len + 1) << 3) #define GET_PKTOPT_SOCKADDR(field) GET_PKTOPT_VAR(field, \ pktopt->field->sa_len) static int ip6_getpcbopt(struct inpcb *inp, int optname, struct sockopt *sopt) { void *optdata = NULL; bool malloc_optdata = false; int optdatalen = 0; int error = 0; struct in6_pktinfo null_pktinfo; int deftclass = 0, on; int defminmtu = IP6PO_MINMTU_MCASTONLY; int defpreftemp = IP6PO_TEMPADDR_SYSTEM; struct ip6_pktopts *pktopt; INP_RLOCK(inp); pktopt = inp->in6p_outputopts; switch (optname) { case IPV6_PKTINFO: optdata = (void *)&null_pktinfo; if (pktopt && pktopt->ip6po_pktinfo) { bcopy(pktopt->ip6po_pktinfo, &null_pktinfo, sizeof(null_pktinfo)); in6_clearscope(&null_pktinfo.ipi6_addr); } else { /* XXX: we don't have to do this every time... */ bzero(&null_pktinfo, sizeof(null_pktinfo)); } optdatalen = sizeof(struct in6_pktinfo); break; case IPV6_TCLASS: if (pktopt && pktopt->ip6po_tclass >= 0) deftclass = pktopt->ip6po_tclass; optdata = (void *)&deftclass; optdatalen = sizeof(int); break; case IPV6_HOPOPTS: GET_PKTOPT_EXT_HDR(ip6po_hbh); break; case IPV6_RTHDR: GET_PKTOPT_EXT_HDR(ip6po_rthdr); break; case IPV6_RTHDRDSTOPTS: GET_PKTOPT_EXT_HDR(ip6po_dest1); break; case IPV6_DSTOPTS: GET_PKTOPT_EXT_HDR(ip6po_dest2); break; case IPV6_NEXTHOP: GET_PKTOPT_SOCKADDR(ip6po_nexthop); break; case IPV6_USE_MIN_MTU: if (pktopt) defminmtu = pktopt->ip6po_minmtu; optdata = (void *)&defminmtu; optdatalen = sizeof(int); break; case IPV6_DONTFRAG: if (pktopt && ((pktopt->ip6po_flags) & IP6PO_DONTFRAG)) on = 1; else on = 0; optdata = (void *)&on; optdatalen = sizeof(on); break; case IPV6_PREFER_TEMPADDR: if (pktopt) defpreftemp = pktopt->ip6po_prefer_tempaddr; optdata = (void *)&defpreftemp; optdatalen = sizeof(int); break; default: /* should not happen */ #ifdef DIAGNOSTIC panic("ip6_getpcbopt: unexpected option\n"); #endif INP_RUNLOCK(inp); return (ENOPROTOOPT); } INP_RUNLOCK(inp); error = sooptcopyout(sopt, optdata, optdatalen); if (malloc_optdata) free(optdata, M_TEMP); return (error); } void ip6_clearpktopts(struct ip6_pktopts *pktopt, int optname) { if (pktopt == NULL) return; if (optname == -1 || optname == IPV6_PKTINFO) { if (pktopt->ip6po_pktinfo) free(pktopt->ip6po_pktinfo, M_IP6OPT); pktopt->ip6po_pktinfo = NULL; } if (optname == -1 || optname == IPV6_HOPLIMIT) pktopt->ip6po_hlim = -1; if (optname == -1 || optname == IPV6_TCLASS) pktopt->ip6po_tclass = -1; if (optname == -1 || optname == IPV6_NEXTHOP) { - if (pktopt->ip6po_nextroute.ro_rt) { - RTFREE(pktopt->ip6po_nextroute.ro_rt); - pktopt->ip6po_nextroute.ro_rt = NULL; + if (pktopt->ip6po_nextroute.ro_nh) { + NH_FREE(pktopt->ip6po_nextroute.ro_nh); + pktopt->ip6po_nextroute.ro_nh = NULL; } if (pktopt->ip6po_nexthop) free(pktopt->ip6po_nexthop, M_IP6OPT); pktopt->ip6po_nexthop = NULL; } if (optname == -1 || optname == IPV6_HOPOPTS) { if (pktopt->ip6po_hbh) free(pktopt->ip6po_hbh, M_IP6OPT); pktopt->ip6po_hbh = NULL; } if (optname == -1 || optname == IPV6_RTHDRDSTOPTS) { if (pktopt->ip6po_dest1) free(pktopt->ip6po_dest1, M_IP6OPT); pktopt->ip6po_dest1 = NULL; } if (optname == -1 || optname == IPV6_RTHDR) { if (pktopt->ip6po_rhinfo.ip6po_rhi_rthdr) free(pktopt->ip6po_rhinfo.ip6po_rhi_rthdr, M_IP6OPT); pktopt->ip6po_rhinfo.ip6po_rhi_rthdr = NULL; - if (pktopt->ip6po_route.ro_rt) { - RTFREE(pktopt->ip6po_route.ro_rt); - pktopt->ip6po_route.ro_rt = NULL; + if (pktopt->ip6po_route.ro_nh) { + NH_FREE(pktopt->ip6po_route.ro_nh); + pktopt->ip6po_route.ro_nh = NULL; } } if (optname == -1 || optname == IPV6_DSTOPTS) { if (pktopt->ip6po_dest2) free(pktopt->ip6po_dest2, M_IP6OPT); pktopt->ip6po_dest2 = NULL; } } #define PKTOPT_EXTHDRCPY(type) \ do {\ if (src->type) {\ int hlen = (((struct ip6_ext *)src->type)->ip6e_len + 1) << 3;\ dst->type = malloc(hlen, M_IP6OPT, canwait);\ if (dst->type == NULL)\ goto bad;\ bcopy(src->type, dst->type, hlen);\ }\ } while (/*CONSTCOND*/ 0) static int copypktopts(struct ip6_pktopts *dst, struct ip6_pktopts *src, int canwait) { if (dst == NULL || src == NULL) { printf("ip6_clearpktopts: invalid argument\n"); return (EINVAL); } dst->ip6po_hlim = src->ip6po_hlim; dst->ip6po_tclass = src->ip6po_tclass; dst->ip6po_flags = src->ip6po_flags; dst->ip6po_minmtu = src->ip6po_minmtu; dst->ip6po_prefer_tempaddr = src->ip6po_prefer_tempaddr; if (src->ip6po_pktinfo) { dst->ip6po_pktinfo = malloc(sizeof(*dst->ip6po_pktinfo), M_IP6OPT, canwait); if (dst->ip6po_pktinfo == NULL) goto bad; *dst->ip6po_pktinfo = *src->ip6po_pktinfo; } if (src->ip6po_nexthop) { dst->ip6po_nexthop = malloc(src->ip6po_nexthop->sa_len, M_IP6OPT, canwait); if (dst->ip6po_nexthop == NULL) goto bad; bcopy(src->ip6po_nexthop, dst->ip6po_nexthop, src->ip6po_nexthop->sa_len); } PKTOPT_EXTHDRCPY(ip6po_hbh); PKTOPT_EXTHDRCPY(ip6po_dest1); PKTOPT_EXTHDRCPY(ip6po_dest2); PKTOPT_EXTHDRCPY(ip6po_rthdr); /* not copy the cached route */ return (0); bad: ip6_clearpktopts(dst, -1); return (ENOBUFS); } #undef PKTOPT_EXTHDRCPY struct ip6_pktopts * ip6_copypktopts(struct ip6_pktopts *src, int canwait) { int error; struct ip6_pktopts *dst; dst = malloc(sizeof(*dst), M_IP6OPT, canwait); if (dst == NULL) return (NULL); ip6_initpktopts(dst); if ((error = copypktopts(dst, src, canwait)) != 0) { free(dst, M_IP6OPT); return (NULL); } return (dst); } void ip6_freepcbopts(struct ip6_pktopts *pktopt) { if (pktopt == NULL) return; ip6_clearpktopts(pktopt, -1); free(pktopt, M_IP6OPT); } /* * Set IPv6 outgoing packet options based on advanced API. */ int ip6_setpktopts(struct mbuf *control, struct ip6_pktopts *opt, struct ip6_pktopts *stickyopt, struct ucred *cred, int uproto) { struct cmsghdr *cm = NULL; if (control == NULL || opt == NULL) return (EINVAL); ip6_initpktopts(opt); if (stickyopt) { int error; /* * If stickyopt is provided, make a local copy of the options * for this particular packet, then override them by ancillary * objects. * XXX: copypktopts() does not copy the cached route to a next * hop (if any). This is not very good in terms of efficiency, * but we can allow this since this option should be rarely * used. */ if ((error = copypktopts(opt, stickyopt, M_NOWAIT)) != 0) return (error); } /* * XXX: Currently, we assume all the optional information is stored * in a single mbuf. */ if (control->m_next) return (EINVAL); for (; control->m_len > 0; control->m_data += CMSG_ALIGN(cm->cmsg_len), control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { int error; if (control->m_len < CMSG_LEN(0)) return (EINVAL); cm = mtod(control, struct cmsghdr *); if (cm->cmsg_len == 0 || cm->cmsg_len > control->m_len) return (EINVAL); if (cm->cmsg_level != IPPROTO_IPV6) continue; error = ip6_setpktopt(cm->cmsg_type, CMSG_DATA(cm), cm->cmsg_len - CMSG_LEN(0), opt, cred, 0, 1, uproto); if (error) return (error); } return (0); } /* * Set a particular packet option, as a sticky option or an ancillary data * item. "len" can be 0 only when it's a sticky option. * We have 4 cases of combination of "sticky" and "cmsg": * "sticky=0, cmsg=0": impossible * "sticky=0, cmsg=1": RFC2292 or RFC3542 ancillary data * "sticky=1, cmsg=0": RFC3542 socket option * "sticky=1, cmsg=1": RFC2292 socket option */ static int ip6_setpktopt(int optname, u_char *buf, int len, struct ip6_pktopts *opt, struct ucred *cred, int sticky, int cmsg, int uproto) { int minmtupolicy, preftemp; int error; if (!sticky && !cmsg) { #ifdef DIAGNOSTIC printf("ip6_setpktopt: impossible case\n"); #endif return (EINVAL); } /* * IPV6_2292xxx is for backward compatibility to RFC2292, and should * not be specified in the context of RFC3542. Conversely, * RFC3542 types should not be specified in the context of RFC2292. */ if (!cmsg) { switch (optname) { case IPV6_2292PKTINFO: case IPV6_2292HOPLIMIT: case IPV6_2292NEXTHOP: case IPV6_2292HOPOPTS: case IPV6_2292DSTOPTS: case IPV6_2292RTHDR: case IPV6_2292PKTOPTIONS: return (ENOPROTOOPT); } } if (sticky && cmsg) { switch (optname) { case IPV6_PKTINFO: case IPV6_HOPLIMIT: case IPV6_NEXTHOP: case IPV6_HOPOPTS: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: case IPV6_RTHDR: case IPV6_USE_MIN_MTU: case IPV6_DONTFRAG: case IPV6_TCLASS: case IPV6_PREFER_TEMPADDR: /* XXX: not an RFC3542 option */ return (ENOPROTOOPT); } } switch (optname) { case IPV6_2292PKTINFO: case IPV6_PKTINFO: { struct ifnet *ifp = NULL; struct in6_pktinfo *pktinfo; if (len != sizeof(struct in6_pktinfo)) return (EINVAL); pktinfo = (struct in6_pktinfo *)buf; /* * An application can clear any sticky IPV6_PKTINFO option by * doing a "regular" setsockopt with ipi6_addr being * in6addr_any and ipi6_ifindex being zero. * [RFC 3542, Section 6] */ if (optname == IPV6_PKTINFO && opt->ip6po_pktinfo && pktinfo->ipi6_ifindex == 0 && IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { ip6_clearpktopts(opt, optname); break; } if (uproto == IPPROTO_TCP && optname == IPV6_PKTINFO && sticky && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { return (EINVAL); } if (IN6_IS_ADDR_MULTICAST(&pktinfo->ipi6_addr)) return (EINVAL); /* validate the interface index if specified. */ if (pktinfo->ipi6_ifindex > V_if_index) return (ENXIO); if (pktinfo->ipi6_ifindex) { ifp = ifnet_byindex(pktinfo->ipi6_ifindex); if (ifp == NULL) return (ENXIO); } if (ifp != NULL && (ifp->if_afdata[AF_INET6] == NULL || (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)) return (ENETDOWN); if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { struct in6_ifaddr *ia; in6_setscope(&pktinfo->ipi6_addr, ifp, NULL); ia = in6ifa_ifpwithaddr(ifp, &pktinfo->ipi6_addr); if (ia == NULL) return (EADDRNOTAVAIL); ifa_free(&ia->ia_ifa); } /* * We store the address anyway, and let in6_selectsrc() * validate the specified address. This is because ipi6_addr * may not have enough information about its scope zone, and * we may need additional information (such as outgoing * interface or the scope zone of a destination address) to * disambiguate the scope. * XXX: the delay of the validation may confuse the * application when it is used as a sticky option. */ if (opt->ip6po_pktinfo == NULL) { opt->ip6po_pktinfo = malloc(sizeof(*pktinfo), M_IP6OPT, M_NOWAIT); if (opt->ip6po_pktinfo == NULL) return (ENOBUFS); } bcopy(pktinfo, opt->ip6po_pktinfo, sizeof(*pktinfo)); break; } case IPV6_2292HOPLIMIT: case IPV6_HOPLIMIT: { int *hlimp; /* * RFC 3542 deprecated the usage of sticky IPV6_HOPLIMIT * to simplify the ordering among hoplimit options. */ if (optname == IPV6_HOPLIMIT && sticky) return (ENOPROTOOPT); if (len != sizeof(int)) return (EINVAL); hlimp = (int *)buf; if (*hlimp < -1 || *hlimp > 255) return (EINVAL); opt->ip6po_hlim = *hlimp; break; } case IPV6_TCLASS: { int tclass; if (len != sizeof(int)) return (EINVAL); tclass = *(int *)buf; if (tclass < -1 || tclass > 255) return (EINVAL); opt->ip6po_tclass = tclass; break; } case IPV6_2292NEXTHOP: case IPV6_NEXTHOP: if (cred != NULL) { error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS); if (error) return (error); } if (len == 0) { /* just remove the option */ ip6_clearpktopts(opt, IPV6_NEXTHOP); break; } /* check if cmsg_len is large enough for sa_len */ if (len < sizeof(struct sockaddr) || len < *buf) return (EINVAL); switch (((struct sockaddr *)buf)->sa_family) { case AF_INET6: { struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)buf; int error; if (sa6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) || IN6_IS_ADDR_MULTICAST(&sa6->sin6_addr)) { return (EINVAL); } if ((error = sa6_embedscope(sa6, V_ip6_use_defzone)) != 0) { return (error); } break; } case AF_LINK: /* should eventually be supported */ default: return (EAFNOSUPPORT); } /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, IPV6_NEXTHOP); opt->ip6po_nexthop = malloc(*buf, M_IP6OPT, M_NOWAIT); if (opt->ip6po_nexthop == NULL) return (ENOBUFS); bcopy(buf, opt->ip6po_nexthop, *buf); break; case IPV6_2292HOPOPTS: case IPV6_HOPOPTS: { struct ip6_hbh *hbh; int hbhlen; /* * XXX: We don't allow a non-privileged user to set ANY HbH * options, since per-option restriction has too much * overhead. */ if (cred != NULL) { error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS); if (error) return (error); } if (len == 0) { ip6_clearpktopts(opt, IPV6_HOPOPTS); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_hbh)) return (EINVAL); hbh = (struct ip6_hbh *)buf; hbhlen = (hbh->ip6h_len + 1) << 3; if (len != hbhlen) return (EINVAL); /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, IPV6_HOPOPTS); opt->ip6po_hbh = malloc(hbhlen, M_IP6OPT, M_NOWAIT); if (opt->ip6po_hbh == NULL) return (ENOBUFS); bcopy(hbh, opt->ip6po_hbh, hbhlen); break; } case IPV6_2292DSTOPTS: case IPV6_DSTOPTS: case IPV6_RTHDRDSTOPTS: { struct ip6_dest *dest, **newdest = NULL; int destlen; if (cred != NULL) { /* XXX: see the comment for IPV6_HOPOPTS */ error = priv_check_cred(cred, PRIV_NETINET_SETHDROPTS); if (error) return (error); } if (len == 0) { ip6_clearpktopts(opt, optname); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_dest)) return (EINVAL); dest = (struct ip6_dest *)buf; destlen = (dest->ip6d_len + 1) << 3; if (len != destlen) return (EINVAL); /* * Determine the position that the destination options header * should be inserted; before or after the routing header. */ switch (optname) { case IPV6_2292DSTOPTS: /* * The old advacned API is ambiguous on this point. * Our approach is to determine the position based * according to the existence of a routing header. * Note, however, that this depends on the order of the * extension headers in the ancillary data; the 1st * part of the destination options header must appear * before the routing header in the ancillary data, * too. * RFC3542 solved the ambiguity by introducing * separate ancillary data or option types. */ if (opt->ip6po_rthdr == NULL) newdest = &opt->ip6po_dest1; else newdest = &opt->ip6po_dest2; break; case IPV6_RTHDRDSTOPTS: newdest = &opt->ip6po_dest1; break; case IPV6_DSTOPTS: newdest = &opt->ip6po_dest2; break; } /* turn off the previous option, then set the new option. */ ip6_clearpktopts(opt, optname); *newdest = malloc(destlen, M_IP6OPT, M_NOWAIT); if (*newdest == NULL) return (ENOBUFS); bcopy(dest, *newdest, destlen); break; } case IPV6_2292RTHDR: case IPV6_RTHDR: { struct ip6_rthdr *rth; int rthlen; if (len == 0) { ip6_clearpktopts(opt, IPV6_RTHDR); break; /* just remove the option */ } /* message length validation */ if (len < sizeof(struct ip6_rthdr)) return (EINVAL); rth = (struct ip6_rthdr *)buf; rthlen = (rth->ip6r_len + 1) << 3; if (len != rthlen) return (EINVAL); switch (rth->ip6r_type) { case IPV6_RTHDR_TYPE_0: if (rth->ip6r_len == 0) /* must contain one addr */ return (EINVAL); if (rth->ip6r_len % 2) /* length must be even */ return (EINVAL); if (rth->ip6r_len / 2 != rth->ip6r_segleft) return (EINVAL); break; default: return (EINVAL); /* not supported */ } /* turn off the previous option */ ip6_clearpktopts(opt, IPV6_RTHDR); opt->ip6po_rthdr = malloc(rthlen, M_IP6OPT, M_NOWAIT); if (opt->ip6po_rthdr == NULL) return (ENOBUFS); bcopy(rth, opt->ip6po_rthdr, rthlen); break; } case IPV6_USE_MIN_MTU: if (len != sizeof(int)) return (EINVAL); minmtupolicy = *(int *)buf; if (minmtupolicy != IP6PO_MINMTU_MCASTONLY && minmtupolicy != IP6PO_MINMTU_DISABLE && minmtupolicy != IP6PO_MINMTU_ALL) { return (EINVAL); } opt->ip6po_minmtu = minmtupolicy; break; case IPV6_DONTFRAG: if (len != sizeof(int)) return (EINVAL); if (uproto == IPPROTO_TCP || *(int *)buf == 0) { /* * we ignore this option for TCP sockets. * (RFC3542 leaves this case unspecified.) */ opt->ip6po_flags &= ~IP6PO_DONTFRAG; } else opt->ip6po_flags |= IP6PO_DONTFRAG; break; case IPV6_PREFER_TEMPADDR: if (len != sizeof(int)) return (EINVAL); preftemp = *(int *)buf; if (preftemp != IP6PO_TEMPADDR_SYSTEM && preftemp != IP6PO_TEMPADDR_NOTPREFER && preftemp != IP6PO_TEMPADDR_PREFER) { return (EINVAL); } opt->ip6po_prefer_tempaddr = preftemp; break; default: return (ENOPROTOOPT); } /* end of switch */ return (0); } /* * Routine called from ip6_output() to loop back a copy of an IP6 multicast * packet to the input queue of a specified interface. Note that this * calls the output routine of the loopback "driver", but with an interface * pointer that might NOT be &loif -- easier than replicating that code here. */ void ip6_mloopback(struct ifnet *ifp, struct mbuf *m) { struct mbuf *copym; struct ip6_hdr *ip6; copym = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (copym == NULL) return; /* * Make sure to deep-copy IPv6 header portion in case the data * is in an mbuf cluster, so that we can safely override the IPv6 * header portion later. */ if (!M_WRITABLE(copym) || copym->m_len < sizeof(struct ip6_hdr)) { copym = m_pullup(copym, sizeof(struct ip6_hdr)); if (copym == NULL) return; } ip6 = mtod(copym, struct ip6_hdr *); /* * clear embedded scope identifiers if necessary. * in6_clearscope will touch the addresses only when necessary. */ in6_clearscope(&ip6->ip6_src); in6_clearscope(&ip6->ip6_dst); if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { copym->m_pkthdr.csum_flags |= CSUM_DATA_VALID_IPV6 | CSUM_PSEUDO_HDR; copym->m_pkthdr.csum_data = 0xffff; } if_simloop(ifp, copym, AF_INET6, 0); } /* * Chop IPv6 header off from the payload. */ static int ip6_splithdr(struct mbuf *m, struct ip6_exthdrs *exthdrs) { struct mbuf *mh; struct ip6_hdr *ip6; ip6 = mtod(m, struct ip6_hdr *); if (m->m_len > sizeof(*ip6)) { mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { m_freem(m); return ENOBUFS; } m_move_pkthdr(mh, m); M_ALIGN(mh, sizeof(*ip6)); m->m_len -= sizeof(*ip6); m->m_data += sizeof(*ip6); mh->m_next = m; m = mh; m->m_len = sizeof(*ip6); bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(*ip6)); } exthdrs->ip6e_ip6 = m; return 0; } /* * Compute IPv6 extension header length. */ int ip6_optlen(struct inpcb *inp) { int len; if (!inp->in6p_outputopts) return 0; len = 0; #define elen(x) \ (((struct ip6_ext *)(x)) ? (((struct ip6_ext *)(x))->ip6e_len + 1) << 3 : 0) len += elen(inp->in6p_outputopts->ip6po_hbh); if (inp->in6p_outputopts->ip6po_rthdr) /* dest1 is valid with rthdr only */ len += elen(inp->in6p_outputopts->ip6po_dest1); len += elen(inp->in6p_outputopts->ip6po_rthdr); len += elen(inp->in6p_outputopts->ip6po_dest2); return len; #undef elen } Index: head/sys/netinet6/ip6_var.h =================================================================== --- head/sys/netinet6/ip6_var.h (revision 360291) +++ head/sys/netinet6/ip6_var.h (revision 360292) @@ -1,425 +1,425 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * 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 project 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 PROJECT 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 PROJECT 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. * * $KAME: ip6_var.h,v 1.62 2001/05/03 14:51:48 itojun Exp $ */ /*- * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * 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 University 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 REGENTS 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 REGENTS 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. * * @(#)ip_var.h 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #ifndef _NETINET6_IP6_VAR_H_ #define _NETINET6_IP6_VAR_H_ #include #ifdef _KERNEL struct ip6asfrag; /* frag6.c */ TAILQ_HEAD(ip6fraghead, ip6asfrag); /* * IP6 reassembly queue structure. Each fragment * being reassembled is attached to one of these structures. */ struct ip6q { struct ip6fraghead ip6q_frags; u_int32_t ip6q_ident; u_int8_t ip6q_nxt; u_int8_t ip6q_ecn; u_int8_t ip6q_ttl; struct in6_addr ip6q_src, ip6q_dst; TAILQ_ENTRY(ip6q) ip6q_tq; int ip6q_unfrglen; /* len of unfragmentable part */ int ip6q_nfrag; /* # of fragments */ struct label *ip6q_label; }; #endif /* _KERNEL */ /* * IP6 reinjecting structure. */ struct ip6_direct_ctx { uint32_t ip6dc_nxt; /* next header to process */ uint32_t ip6dc_off; /* offset to next header */ }; #if defined(_NETINET6_IN6_VAR_H_) && defined(_KERNEL) /* * Structure attached to inpcb.in6p_moptions and * passed to ip6_output when IPv6 multicast options are in use. * This structure is lazy-allocated. */ struct ip6_moptions { struct ifnet *im6o_multicast_ifp; /* ifp for outgoing multicasts */ u_char im6o_multicast_hlim; /* hoplimit for outgoing multicasts */ u_char im6o_multicast_loop; /* 1 >= hear sends if a member */ struct ip6_mfilter_head im6o_head; /* group membership list */ }; #else struct ip6_moptions; #endif /* * Control options for outgoing packets */ /* Routing header related info */ struct ip6po_rhinfo { struct ip6_rthdr *ip6po_rhi_rthdr; /* Routing header */ struct route_in6 ip6po_rhi_route; /* Route to the 1st hop */ }; #define ip6po_rthdr ip6po_rhinfo.ip6po_rhi_rthdr #define ip6po_route ip6po_rhinfo.ip6po_rhi_route /* Nexthop related info */ struct ip6po_nhinfo { struct sockaddr *ip6po_nhi_nexthop; struct route_in6 ip6po_nhi_route; /* Route to the nexthop */ }; #define ip6po_nexthop ip6po_nhinfo.ip6po_nhi_nexthop #define ip6po_nextroute ip6po_nhinfo.ip6po_nhi_route struct ip6_pktopts { struct mbuf *ip6po_m; /* Pointer to mbuf storing the data */ int ip6po_hlim; /* Hoplimit for outgoing packets */ /* Outgoing IF/address information */ struct in6_pktinfo *ip6po_pktinfo; /* Next-hop address information */ struct ip6po_nhinfo ip6po_nhinfo; struct ip6_hbh *ip6po_hbh; /* Hop-by-Hop options header */ /* Destination options header (before a routing header) */ struct ip6_dest *ip6po_dest1; /* Routing header related info. */ struct ip6po_rhinfo ip6po_rhinfo; /* Destination options header (after a routing header) */ struct ip6_dest *ip6po_dest2; int ip6po_tclass; /* traffic class */ int ip6po_minmtu; /* fragment vs PMTU discovery policy */ #define IP6PO_MINMTU_MCASTONLY -1 /* default; send at min MTU for multicast*/ #define IP6PO_MINMTU_DISABLE 0 /* always perform pmtu disc */ #define IP6PO_MINMTU_ALL 1 /* always send at min MTU */ int ip6po_prefer_tempaddr; /* whether temporary addresses are preferred as source address */ #define IP6PO_TEMPADDR_SYSTEM -1 /* follow the system default */ #define IP6PO_TEMPADDR_NOTPREFER 0 /* not prefer temporary address */ #define IP6PO_TEMPADDR_PREFER 1 /* prefer temporary address */ int ip6po_flags; #if 0 /* parameters in this block is obsolete. do not reuse the values. */ #define IP6PO_REACHCONF 0x01 /* upper-layer reachability confirmation. */ #define IP6PO_MINMTU 0x02 /* use minimum MTU (IPV6_USE_MIN_MTU) */ #endif #define IP6PO_DONTFRAG 0x04 /* disable fragmentation (IPV6_DONTFRAG) */ #define IP6PO_USECOA 0x08 /* use care of address */ }; /* * Control options for incoming packets */ struct ip6stat { uint64_t ip6s_total; /* total packets received */ uint64_t ip6s_tooshort; /* packet too short */ uint64_t ip6s_toosmall; /* not enough data */ uint64_t ip6s_fragments; /* fragments received */ uint64_t ip6s_fragdropped; /* frags dropped(dups, out of space) */ uint64_t ip6s_fragtimeout; /* fragments timed out */ uint64_t ip6s_fragoverflow; /* fragments that exceeded limit */ uint64_t ip6s_forward; /* packets forwarded */ uint64_t ip6s_cantforward; /* packets rcvd for unreachable dest */ uint64_t ip6s_redirectsent; /* packets forwarded on same net */ uint64_t ip6s_delivered; /* datagrams delivered to upper level*/ uint64_t ip6s_localout; /* total ip packets generated here */ uint64_t ip6s_odropped; /* lost packets due to nobufs, etc. */ uint64_t ip6s_reassembled; /* total packets reassembled ok */ uint64_t ip6s_atomicfrags; /* atomic fragments */ uint64_t ip6s_fragmented; /* datagrams successfully fragmented */ uint64_t ip6s_ofragments; /* output fragments created */ uint64_t ip6s_cantfrag; /* don't fragment flag was set, etc. */ uint64_t ip6s_badoptions; /* error in option processing */ uint64_t ip6s_noroute; /* packets discarded due to no route */ uint64_t ip6s_badvers; /* ip6 version != 6 */ uint64_t ip6s_rawout; /* total raw ip packets generated */ uint64_t ip6s_badscope; /* scope error */ uint64_t ip6s_notmember; /* don't join this multicast group */ #define IP6S_HDRCNT 256 /* headers count */ uint64_t ip6s_nxthist[IP6S_HDRCNT]; /* next header history */ uint64_t ip6s_m1; /* one mbuf */ #define IP6S_M2MMAX 32 uint64_t ip6s_m2m[IP6S_M2MMAX]; /* two or more mbuf */ uint64_t ip6s_mext1; /* one ext mbuf */ uint64_t ip6s_mext2m; /* two or more ext mbuf */ uint64_t ip6s_exthdrtoolong; /* ext hdr are not contiguous */ uint64_t ip6s_nogif; /* no match gif found */ uint64_t ip6s_toomanyhdr; /* discarded due to too many headers */ /* * statistics for improvement of the source address selection * algorithm: * XXX: hardcoded 16 = # of ip6 multicast scope types + 1 */ #define IP6S_RULESMAX 16 #define IP6S_SCOPECNT 16 /* number of times that address selection fails */ uint64_t ip6s_sources_none; /* number of times that an address on the outgoing I/F is chosen */ uint64_t ip6s_sources_sameif[IP6S_SCOPECNT]; /* number of times that an address on a non-outgoing I/F is chosen */ uint64_t ip6s_sources_otherif[IP6S_SCOPECNT]; /* * number of times that an address that has the same scope * from the destination is chosen. */ uint64_t ip6s_sources_samescope[IP6S_SCOPECNT]; /* * number of times that an address that has a different scope * from the destination is chosen. */ uint64_t ip6s_sources_otherscope[IP6S_SCOPECNT]; /* number of times that a deprecated address is chosen */ uint64_t ip6s_sources_deprecated[IP6S_SCOPECNT]; /* number of times that each rule of source selection is applied. */ uint64_t ip6s_sources_rule[IP6S_RULESMAX]; }; #ifdef _KERNEL #include VNET_PCPUSTAT_DECLARE(struct ip6stat, ip6stat); #define IP6STAT_ADD(name, val) \ VNET_PCPUSTAT_ADD(struct ip6stat, ip6stat, name, (val)) #define IP6STAT_SUB(name, val) IP6STAT_ADD(name, -(val)) #define IP6STAT_INC(name) IP6STAT_ADD(name, 1) #define IP6STAT_DEC(name) IP6STAT_SUB(name, 1) #endif #ifdef _KERNEL /* flags passed to ip6_output as last parameter */ #define IPV6_UNSPECSRC 0x01 /* allow :: as the source address */ #define IPV6_FORWARDING 0x02 /* most of IPv6 header exists */ #define IPV6_MINMTU 0x04 /* use minimum MTU (IPV6_USE_MIN_MTU) */ #ifdef __NO_STRICT_ALIGNMENT #define IP6_HDR_ALIGNED_P(ip) 1 #else #define IP6_HDR_ALIGNED_P(ip) ((((intptr_t) (ip)) & 3) == 0) #endif VNET_DECLARE(int, ip6_defhlim); /* default hop limit */ VNET_DECLARE(int, ip6_defmcasthlim); /* default multicast hop limit */ VNET_DECLARE(int, ip6_forwarding); /* act as router? */ VNET_DECLARE(int, ip6_use_deprecated); /* allow deprecated addr as source */ VNET_DECLARE(int, ip6_rr_prune); /* router renumbering prefix * walk list every 5 sec. */ VNET_DECLARE(int, ip6_mcast_pmtu); /* enable pMTU discovery for multicast? */ VNET_DECLARE(int, ip6_v6only); #define V_ip6_defhlim VNET(ip6_defhlim) #define V_ip6_defmcasthlim VNET(ip6_defmcasthlim) #define V_ip6_forwarding VNET(ip6_forwarding) #define V_ip6_use_deprecated VNET(ip6_use_deprecated) #define V_ip6_rr_prune VNET(ip6_rr_prune) #define V_ip6_mcast_pmtu VNET(ip6_mcast_pmtu) #define V_ip6_v6only VNET(ip6_v6only) VNET_DECLARE(struct socket *, ip6_mrouter); /* multicast routing daemon */ VNET_DECLARE(int, ip6_sendredirects); /* send IP redirects when forwarding? */ VNET_DECLARE(int, ip6_accept_rtadv); /* Acts as a host not a router */ VNET_DECLARE(int, ip6_no_radr); /* No defroute from RA */ VNET_DECLARE(int, ip6_norbit_raif); /* Disable R-bit in NA on RA * receiving IF. */ VNET_DECLARE(int, ip6_rfc6204w3); /* Accept defroute from RA even when forwarding enabled */ VNET_DECLARE(int, ip6_log_interval); VNET_DECLARE(time_t, ip6_log_time); VNET_DECLARE(int, ip6_hdrnestlimit); /* upper limit of # of extension * headers */ VNET_DECLARE(int, ip6_dad_count); /* DupAddrDetectionTransmits */ #define V_ip6_mrouter VNET(ip6_mrouter) #define V_ip6_sendredirects VNET(ip6_sendredirects) #define V_ip6_accept_rtadv VNET(ip6_accept_rtadv) #define V_ip6_no_radr VNET(ip6_no_radr) #define V_ip6_norbit_raif VNET(ip6_norbit_raif) #define V_ip6_rfc6204w3 VNET(ip6_rfc6204w3) #define V_ip6_log_interval VNET(ip6_log_interval) #define V_ip6_log_time VNET(ip6_log_time) #define V_ip6_hdrnestlimit VNET(ip6_hdrnestlimit) #define V_ip6_dad_count VNET(ip6_dad_count) VNET_DECLARE(int, ip6_auto_flowlabel); VNET_DECLARE(int, ip6_auto_linklocal); #define V_ip6_auto_flowlabel VNET(ip6_auto_flowlabel) #define V_ip6_auto_linklocal VNET(ip6_auto_linklocal) VNET_DECLARE(int, ip6_use_tempaddr); /* Whether to use temporary addresses */ VNET_DECLARE(int, ip6_prefer_tempaddr); /* Whether to prefer temporary * addresses in the source address * selection */ #define V_ip6_use_tempaddr VNET(ip6_use_tempaddr) #define V_ip6_prefer_tempaddr VNET(ip6_prefer_tempaddr) VNET_DECLARE(int, ip6_use_defzone); /* Whether to use the default scope * zone when unspecified */ #define V_ip6_use_defzone VNET(ip6_use_defzone) VNET_DECLARE(struct pfil_head *, inet6_pfil_head); #define V_inet6_pfil_head VNET(inet6_pfil_head) #define PFIL_INET6_NAME "inet6" #ifdef IPSTEALTH VNET_DECLARE(int, ip6stealth); #define V_ip6stealth VNET(ip6stealth) #endif #ifdef EXPERIMENTAL VNET_DECLARE(int, nd6_ignore_ipv6_only_ra); #define V_nd6_ignore_ipv6_only_ra VNET(nd6_ignore_ipv6_only_ra) #endif extern struct pr_usrreqs rip6_usrreqs; struct sockopt; struct inpcb; int icmp6_ctloutput(struct socket *, struct sockopt *sopt); struct in6_ifaddr; void ip6_init(void); int ip6proto_register(short); int ip6proto_unregister(short); void ip6_input(struct mbuf *); void ip6_direct_input(struct mbuf *); void ip6_freepcbopts(struct ip6_pktopts *); int ip6_unknown_opt(u_int8_t *, struct mbuf *, int); int ip6_get_prevhdr(const struct mbuf *, int); int ip6_nexthdr(const struct mbuf *, int, int, int *); int ip6_lasthdr(const struct mbuf *, int, int, int *); extern int (*ip6_mforward)(struct ip6_hdr *, struct ifnet *, struct mbuf *); int ip6_process_hopopts(struct mbuf *, u_int8_t *, int, u_int32_t *, u_int32_t *); struct mbuf **ip6_savecontrol_v4(struct inpcb *, struct mbuf *, struct mbuf **, int *); void ip6_savecontrol(struct inpcb *, struct mbuf *, struct mbuf **); void ip6_notify_pmtu(struct inpcb *, struct sockaddr_in6 *, u_int32_t); int ip6_sysctl(int *, u_int, void *, size_t *, void *, size_t); void ip6_forward(struct mbuf *, int); void ip6_mloopback(struct ifnet *, struct mbuf *); int ip6_output(struct mbuf *, struct ip6_pktopts *, struct route_in6 *, int, struct ip6_moptions *, struct ifnet **, struct inpcb *); int ip6_ctloutput(struct socket *, struct sockopt *); int ip6_raw_ctloutput(struct socket *, struct sockopt *); void ip6_initpktopts(struct ip6_pktopts *); int ip6_setpktopts(struct mbuf *, struct ip6_pktopts *, struct ip6_pktopts *, struct ucred *, int); void ip6_clearpktopts(struct ip6_pktopts *, int); struct ip6_pktopts *ip6_copypktopts(struct ip6_pktopts *, int); int ip6_optlen(struct inpcb *); int ip6_deletefraghdr(struct mbuf *, int, int); int ip6_fragment(struct ifnet *, struct mbuf *, int, u_char, int, uint32_t); int route6_input(struct mbuf **, int *, int); void frag6_init(void); void frag6_destroy(void); int frag6_input(struct mbuf **, int *, int); void frag6_slowtimo(void); void frag6_drain(void); void rip6_init(void); int rip6_input(struct mbuf **, int *, int); void rip6_ctlinput(int, struct sockaddr *, void *); int rip6_ctloutput(struct socket *, struct sockopt *); int rip6_output(struct mbuf *, struct socket *, ...); int rip6_usrreq(struct socket *, int, struct mbuf *, struct mbuf *, struct mbuf *, struct thread *); int dest6_input(struct mbuf **, int *, int); int none_input(struct mbuf **, int *, int); int in6_selectsrc_socket(struct sockaddr_in6 *, struct ip6_pktopts *, struct inpcb *, struct ucred *, int, struct in6_addr *, int *); int in6_selectsrc_addr(uint32_t, const struct in6_addr *, uint32_t, struct ifnet *, struct in6_addr *, int *); int in6_selectroute(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct route_in6 *, struct ifnet **, - struct rtentry **, u_int); + struct nhop_object **, u_int, uint32_t); u_int32_t ip6_randomid(void); u_int32_t ip6_randomflowlabel(void); void in6_delayed_cksum(struct mbuf *m, uint32_t plen, u_short offset); #endif /* _KERNEL */ #endif /* !_NETINET6_IP6_VAR_H_ */