Index: head/sys/kern/uipc_ktls.c =================================================================== --- head/sys/kern/uipc_ktls.c (revision 358807) +++ head/sys/kern/uipc_ktls.c (revision 358808) @@ -1,1583 +1,1584 @@ /*- * 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 #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: case CRYPTO_AES_128_NIST_GMAC: case CRYPTO_AES_192_NIST_GMAC: case CRYPTO_AES_256_NIST_GMAC: 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 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) { INP_RUNLOCK(inp); return (ENXIO); } ifp = rt->rt_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.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.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->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; } 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.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.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->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->hdr, pgs->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->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; 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); uma_zfree(zone_extpgs, p); } } } } Index: head/sys/net/ieee8023ad_lacp.c =================================================================== --- head/sys/net/ieee8023ad_lacp.c (revision 358807) +++ head/sys/net/ieee8023ad_lacp.c (revision 358808) @@ -1,2221 +1,2204 @@ /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c)2005 YAMAMOTO Takashi, * Copyright (c)2008 Andrew Thompson * 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 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 AUTHOR 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_kern_tls.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include /* hz */ #include /* for net/if.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * actor system priority and port priority. * XXX should be configurable. */ #define LACP_SYSTEM_PRIO 0x8000 #define LACP_PORT_PRIO 0x8000 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] = { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 }; static const struct tlv_template lacp_info_tlv_template[] = { { LACP_TYPE_ACTORINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, { LACP_TYPE_PARTNERINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, { LACP_TYPE_COLLECTORINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) }, { 0, 0 }, }; static const struct tlv_template marker_info_tlv_template[] = { { MARKER_TYPE_INFO, sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, { 0, 0 }, }; static const struct tlv_template marker_response_tlv_template[] = { { MARKER_TYPE_RESPONSE, sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, { 0, 0 }, }; typedef void (*lacp_timer_func_t)(struct lacp_port *); static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *); static void lacp_fill_markerinfo(struct lacp_port *, struct lacp_markerinfo *); static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *); static void lacp_suppress_distributing(struct lacp_softc *, struct lacp_aggregator *); static void lacp_transit_expire(void *); static void lacp_update_portmap(struct lacp_softc *); static void lacp_select_active_aggregator(struct lacp_softc *); static uint16_t lacp_compose_key(struct lacp_port *); static int tlv_check(const void *, size_t, const struct tlvhdr *, const struct tlv_template *, boolean_t); static void lacp_tick(void *); static void lacp_fill_aggregator_id(struct lacp_aggregator *, const struct lacp_port *); static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *, const struct lacp_peerinfo *); static int lacp_aggregator_is_compatible(const struct lacp_aggregator *, const struct lacp_port *); static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *, const struct lacp_peerinfo *); static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *, struct lacp_port *); static void lacp_aggregator_addref(struct lacp_softc *, struct lacp_aggregator *); static void lacp_aggregator_delref(struct lacp_softc *, struct lacp_aggregator *); /* receive machine */ static int lacp_pdu_input(struct lacp_port *, struct mbuf *); static int lacp_marker_input(struct lacp_port *, struct mbuf *); static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_timer(struct lacp_port *); static void lacp_sm_rx_set_expired(struct lacp_port *); static void lacp_sm_rx_update_ntt(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_record_pdu(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_update_selected(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_record_default(struct lacp_port *); static void lacp_sm_rx_update_default_selected(struct lacp_port *); static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *, const struct lacp_peerinfo *); /* mux machine */ static void lacp_sm_mux(struct lacp_port *); static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state); static void lacp_sm_mux_timer(struct lacp_port *); /* periodic transmit machine */ static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t); static void lacp_sm_ptx_tx_schedule(struct lacp_port *); static void lacp_sm_ptx_timer(struct lacp_port *); /* transmit machine */ static void lacp_sm_tx(struct lacp_port *); static void lacp_sm_assert_ntt(struct lacp_port *); static void lacp_run_timers(struct lacp_port *); static int lacp_compare_peerinfo(const struct lacp_peerinfo *, const struct lacp_peerinfo *); static int lacp_compare_systemid(const struct lacp_systemid *, const struct lacp_systemid *); static void lacp_port_enable(struct lacp_port *); static void lacp_port_disable(struct lacp_port *); static void lacp_select(struct lacp_port *); static void lacp_unselect(struct lacp_port *); static void lacp_disable_collecting(struct lacp_port *); static void lacp_enable_collecting(struct lacp_port *); static void lacp_disable_distributing(struct lacp_port *); static void lacp_enable_distributing(struct lacp_port *); static int lacp_xmit_lacpdu(struct lacp_port *); static int lacp_xmit_marker(struct lacp_port *); /* Debugging */ static void lacp_dump_lacpdu(const struct lacpdu *); static const char *lacp_format_partner(const struct lacp_peerinfo *, char *, size_t); static const char *lacp_format_lagid(const struct lacp_peerinfo *, const struct lacp_peerinfo *, char *, size_t); static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *, char *, size_t); static const char *lacp_format_state(uint8_t, char *, size_t); static const char *lacp_format_mac(const uint8_t *, char *, size_t); static const char *lacp_format_systemid(const struct lacp_systemid *, char *, size_t); static const char *lacp_format_portid(const struct lacp_portid *, char *, size_t); static void lacp_dprintf(const struct lacp_port *, const char *, ...) __attribute__((__format__(__printf__, 2, 3))); VNET_DEFINE_STATIC(int, lacp_debug); #define V_lacp_debug VNET(lacp_debug) SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "ieee802.3ad"); SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)"); VNET_DEFINE_STATIC(int, lacp_default_strict_mode) = 1; SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_default_strict_mode), 0, "LACP strict protocol compliance default"); #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; } #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); } #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; } /* * partner administration variables. * XXX should be configurable. */ static const struct lacp_peerinfo lacp_partner_admin_optimistic = { .lip_systemid = { .lsi_prio = 0xffff }, .lip_portid = { .lpi_prio = 0xffff }, .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION | LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING, }; static const struct lacp_peerinfo lacp_partner_admin_strict = { .lip_systemid = { .lsi_prio = 0xffff }, .lip_portid = { .lpi_prio = 0xffff }, .lip_state = 0, }; static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = { [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer, [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer, [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer, }; struct mbuf * lacp_input(struct lagg_port *lgp, struct mbuf *m) { struct lacp_port *lp = LACP_PORT(lgp); uint8_t subtype; if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) { m_freem(m); return (NULL); } m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype); switch (subtype) { case SLOWPROTOCOLS_SUBTYPE_LACP: lacp_pdu_input(lp, m); return (NULL); case SLOWPROTOCOLS_SUBTYPE_MARKER: lacp_marker_input(lp, m); return (NULL); } /* Not a subtype we are interested in */ return (m); } /* * lacp_pdu_input: process lacpdu */ static int lacp_pdu_input(struct lacp_port *lp, struct mbuf *m) { struct lacp_softc *lsc = lp->lp_lsc; struct lacpdu *du; int error = 0; if (m->m_pkthdr.len != sizeof(*du)) { goto bad; } if ((m->m_flags & M_MCAST) == 0) { goto bad; } if (m->m_len < sizeof(*du)) { m = m_pullup(m, sizeof(*du)); if (m == NULL) { return (ENOMEM); } } du = mtod(m, struct lacpdu *); if (memcmp(&du->ldu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { goto bad; } /* * ignore the version for compatibility with * the future protocol revisions. */ #if 0 if (du->ldu_sph.sph_version != 1) { goto bad; } #endif /* * ignore tlv types for compatibility with * the future protocol revisions. */ if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor, lacp_info_tlv_template, FALSE)) { goto bad; } if (V_lacp_debug > 0) { lacp_dprintf(lp, "lacpdu receive\n"); lacp_dump_lacpdu(du); } if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) { LACP_TPRINTF((lp, "Dropping RX PDU\n")); goto bad; } LACP_LOCK(lsc); lacp_sm_rx(lp, du); LACP_UNLOCK(lsc); m_freem(m); return (error); bad: m_freem(m); return (EINVAL); } static void lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO); memcpy(&info->lip_systemid.lsi_mac, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO); info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index); info->lip_state = lp->lp_state; } static void lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info) { struct ifnet *ifp = lp->lp_ifp; /* Fill in the port index and system id (encoded as the MAC) */ info->mi_rq_port = htons(ifp->if_index); memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN); info->mi_rq_xid = htonl(0); } static int lacp_xmit_lacpdu(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct mbuf *m; struct lacpdu *du; int error; LACP_LOCK_ASSERT(lp->lp_lsc); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { return (ENOMEM); } m->m_len = m->m_pkthdr.len = sizeof(*du); du = mtod(m, struct lacpdu *); memset(du, 0, sizeof(*du)); memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW); du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP; du->ldu_sph.sph_version = 1; TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor)); du->ldu_actor = lp->lp_actor; TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO, sizeof(du->ldu_partner)); du->ldu_partner = lp->lp_partner; TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO, sizeof(du->ldu_collector)); du->ldu_collector.lci_maxdelay = 0; if (V_lacp_debug > 0) { lacp_dprintf(lp, "lacpdu transmit\n"); lacp_dump_lacpdu(du); } m->m_flags |= M_MCAST; /* * XXX should use higher priority queue. * otherwise network congestion can break aggregation. */ error = lagg_enqueue(lp->lp_ifp, m); return (error); } static int lacp_xmit_marker(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct mbuf *m; struct markerdu *mdu; int error; LACP_LOCK_ASSERT(lp->lp_lsc); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { return (ENOMEM); } m->m_len = m->m_pkthdr.len = sizeof(*mdu); mdu = mtod(m, struct markerdu *); memset(mdu, 0, sizeof(*mdu)); memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW); mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER; mdu->mdu_sph.sph_version = 1; /* Bump the transaction id and copy over the marker info */ lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1); TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info)); mdu->mdu_info = lp->lp_marker; LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n", ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", ntohl(mdu->mdu_info.mi_rq_xid))); m->m_flags |= M_MCAST; error = lagg_enqueue(lp->lp_ifp, m); return (error); } void lacp_linkstate(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; struct ifnet *ifp = lgp->lp_ifp; struct ifmediareq ifmr; int error = 0; u_int media; uint8_t old_state; uint16_t old_key; bzero((char *)&ifmr, sizeof(ifmr)); error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr); if (error != 0) { bzero((char *)&ifmr, sizeof(ifmr)); error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr); } if (error != 0) return; LACP_LOCK(lsc); media = ifmr.ifm_active; LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, " "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER, (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP)); old_state = lp->lp_state; old_key = lp->lp_key; lp->lp_media = media; /* * If the port is not an active full duplex Ethernet link then it can * not be aggregated. */ if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 || ifp->if_link_state != LINK_STATE_UP) { lacp_port_disable(lp); } else { lacp_port_enable(lp); } lp->lp_key = lacp_compose_key(lp); if (old_state != lp->lp_state || old_key != lp->lp_key) { LACP_DPRINTF((lp, "-> UNSELECTED\n")); lp->lp_selected = LACP_UNSELECTED; } LACP_UNLOCK(lsc); } static void lacp_tick(void *arg) { struct lacp_softc *lsc = arg; struct lacp_port *lp; LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) continue; CURVNET_SET(lp->lp_ifp->if_vnet); lacp_run_timers(lp); lacp_select(lp); lacp_sm_mux(lp); lacp_sm_tx(lp); lacp_sm_ptx_tx_schedule(lp); CURVNET_RESTORE(); } callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); } int lacp_port_create(struct lagg_port *lgp) { struct lagg_softc *sc = lgp->lp_softc; struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_port *lp; struct ifnet *ifp = lgp->lp_ifp; struct sockaddr_dl sdl; struct ifmultiaddr *rifma = NULL; int error; link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); sdl.sdl_alen = ETHER_ADDR_LEN; bcopy(ðermulticastaddr_slowprotocols, LLADDR(&sdl), ETHER_ADDR_LEN); error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); if (error) { printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifp->if_xname); return (error); } lp = malloc(sizeof(struct lacp_port), M_DEVBUF, M_NOWAIT|M_ZERO); if (lp == NULL) return (ENOMEM); LACP_LOCK(lsc); lgp->lp_psc = lp; lp->lp_ifp = ifp; lp->lp_lagg = lgp; lp->lp_lsc = lsc; lp->lp_ifma = rifma; LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next); lacp_fill_actorinfo(lp, &lp->lp_actor); lacp_fill_markerinfo(lp, &lp->lp_marker); lp->lp_state = LACP_STATE_ACTIVITY; lp->lp_aggregator = NULL; lacp_sm_rx_set_expired(lp); LACP_UNLOCK(lsc); lacp_linkstate(lgp); return (0); } void lacp_port_destroy(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; int i; LACP_LOCK(lsc); for (i = 0; i < LACP_NTIMER; i++) { LACP_TIMER_DISARM(lp, i); } lacp_disable_collecting(lp); lacp_disable_distributing(lp); lacp_unselect(lp); LIST_REMOVE(lp, lp_next); LACP_UNLOCK(lsc); /* The address may have already been removed by if_purgemaddrs() */ if (!lgp->lp_detaching) if_delmulti_ifma(lp->lp_ifma); free(lp, M_DEVBUF); } void lacp_req(struct lagg_softc *sc, void *data) { struct lacp_opreq *req = (struct lacp_opreq *)data; struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_aggregator *la; bzero(req, sizeof(struct lacp_opreq)); /* * If the LACP softc is NULL, return with the opreq structure full of * zeros. It is normal for the softc to be NULL while the lagg is * being destroyed. */ if (NULL == lsc) return; la = lsc->lsc_active_aggregator; LACP_LOCK(lsc); if (la != NULL) { req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio); memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->actor_key = ntohs(la->la_actor.lip_key); req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio); req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno); req->actor_state = la->la_actor.lip_state; req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio); memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->partner_key = ntohs(la->la_partner.lip_key); req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio); req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno); req->partner_state = la->la_partner.lip_state; } LACP_UNLOCK(lsc); } void lacp_portreq(struct lagg_port *lgp, void *data) { struct lacp_opreq *req = (struct lacp_opreq *)data; struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; LACP_LOCK(lsc); req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio); memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->actor_key = ntohs(lp->lp_actor.lip_key); req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio); req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno); req->actor_state = lp->lp_actor.lip_state; req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio); memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->partner_key = ntohs(lp->lp_partner.lip_key); req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio); req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno); req->partner_state = lp->lp_partner.lip_state; LACP_UNLOCK(lsc); } static void lacp_disable_collecting(struct lacp_port *lp) { LACP_DPRINTF((lp, "collecting disabled\n")); lp->lp_state &= ~LACP_STATE_COLLECTING; } static void lacp_enable_collecting(struct lacp_port *lp) { LACP_DPRINTF((lp, "collecting enabled\n")); lp->lp_state |= LACP_STATE_COLLECTING; } static void lacp_disable_distributing(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; struct lacp_softc *lsc = lp->lp_lsc; struct lagg_softc *sc = lsc->lsc_softc; char buf[LACP_LAGIDSTR_MAX+1]; LACP_LOCK_ASSERT(lsc); if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) { return; } KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports")); KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports)); KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid")); LACP_DPRINTF((lp, "disable distributing on aggregator %s, " "nports %d -> %d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), la->la_nports, la->la_nports - 1)); TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q); la->la_nports--; sc->sc_active = la->la_nports; if (lsc->lsc_active_aggregator == la) { lacp_suppress_distributing(lsc, la); lacp_select_active_aggregator(lsc); /* regenerate the port map, the active aggregator has changed */ lacp_update_portmap(lsc); } lp->lp_state &= ~LACP_STATE_DISTRIBUTING; if_link_state_change(sc->sc_ifp, sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); } static void lacp_enable_distributing(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; struct lacp_softc *lsc = lp->lp_lsc; struct lagg_softc *sc = lsc->lsc_softc; char buf[LACP_LAGIDSTR_MAX+1]; LACP_LOCK_ASSERT(lsc); if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) { return; } LACP_DPRINTF((lp, "enable distributing on aggregator %s, " "nports %d -> %d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), la->la_nports, la->la_nports + 1)); KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid")); TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q); la->la_nports++; sc->sc_active = la->la_nports; lp->lp_state |= LACP_STATE_DISTRIBUTING; if (lsc->lsc_active_aggregator == la) { lacp_suppress_distributing(lsc, la); lacp_update_portmap(lsc); } else /* try to become the active aggregator */ lacp_select_active_aggregator(lsc); if_link_state_change(sc->sc_ifp, sc->sc_active ? LINK_STATE_UP : LINK_STATE_DOWN); } static void lacp_transit_expire(void *vp) { struct lacp_softc *lsc = vp; LACP_LOCK_ASSERT(lsc); CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet); LACP_TRACE(NULL); CURVNET_RESTORE(); lsc->lsc_suppress_distributing = FALSE; } void lacp_attach(struct lagg_softc *sc) { struct lacp_softc *lsc; lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO); sc->sc_psc = lsc; lsc->lsc_softc = sc; lsc->lsc_hashkey = m_ether_tcpip_hash_init(); lsc->lsc_active_aggregator = NULL; lsc->lsc_strict_mode = VNET(lacp_default_strict_mode); LACP_LOCK_INIT(lsc); TAILQ_INIT(&lsc->lsc_aggregators); LIST_INIT(&lsc->lsc_ports); callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0); callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0); /* if the lagg is already up then do the same */ if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) lacp_init(sc); } void lacp_detach(void *psc) { struct lacp_softc *lsc = (struct lacp_softc *)psc; KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators), ("aggregators still active")); KASSERT(lsc->lsc_active_aggregator == NULL, ("aggregator still attached")); callout_drain(&lsc->lsc_transit_callout); callout_drain(&lsc->lsc_callout); LACP_LOCK_DESTROY(lsc); free(lsc, M_DEVBUF); } void lacp_init(struct lagg_softc *sc) { struct lacp_softc *lsc = LACP_SOFTC(sc); LACP_LOCK(lsc); callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); LACP_UNLOCK(lsc); } void lacp_stop(struct lagg_softc *sc) { struct lacp_softc *lsc = LACP_SOFTC(sc); LACP_LOCK(lsc); callout_stop(&lsc->lsc_transit_callout); callout_stop(&lsc->lsc_callout); LACP_UNLOCK(lsc); } struct lagg_port * -lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) +lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t hash, uint8_t numa_domain) { struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_portmap *pm; struct lacp_port *lp; struct lacp_port **map; - uint32_t hash; int count; if (__predict_false(lsc->lsc_suppress_distributing)) { LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); return (NULL); } pm = &lsc->lsc_pmap[lsc->lsc_activemap]; if (pm->pm_count == 0) { LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); return (NULL); } #ifdef NUMA if ((sc->sc_opts & LAGG_OPT_USE_NUMA) && - pm->pm_num_dom > 1 && m->m_pkthdr.numa_domain < MAXMEMDOM) { - count = pm->pm_numa[m->m_pkthdr.numa_domain].count; + pm->pm_num_dom > 1 && numa_domain < MAXMEMDOM) { + count = pm->pm_numa[numa_domain].count; if (count > 0) { - map = pm->pm_numa[m->m_pkthdr.numa_domain].map; + map = pm->pm_numa[numa_domain].map; } else { /* No ports on this domain; use global hash. */ map = pm->pm_map; count = pm->pm_count; } } else #endif { map = pm->pm_map; count = pm->pm_count; } - if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && - M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) - hash = m->m_pkthdr.flowid >> sc->flowid_shift; - else - hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey); hash %= count; lp = map[hash]; KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0, ("aggregated port is not distributing")); return (lp->lp_lagg); } -#if defined(RATELIMIT) || defined(KERN_TLS) struct lagg_port * -lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t flowid) +lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) { struct lacp_softc *lsc = LACP_SOFTC(sc); - struct lacp_portmap *pm; - struct lacp_port *lp; uint32_t hash; + uint8_t numa_domain; - if (__predict_false(lsc->lsc_suppress_distributing)) { - LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); - return (NULL); - } + if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && + M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) + hash = m->m_pkthdr.flowid >> sc->flowid_shift; + else + hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey); - pm = &lsc->lsc_pmap[lsc->lsc_activemap]; - if (pm->pm_count == 0) { - LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); - return (NULL); - } - - hash = flowid >> sc->flowid_shift; - hash %= pm->pm_count; - lp = pm->pm_map[hash]; - - return (lp->lp_lagg); + numa_domain = m->m_pkthdr.numa_domain; + return (lacp_select_tx_port_by_hash(sc, hash, numa_domain)); } -#endif /* * lacp_suppress_distributing: drop transmit packets for a while * to preserve packet ordering. */ static void lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la) { struct lacp_port *lp; if (lsc->lsc_active_aggregator != la) { return; } LACP_TRACE(NULL); lsc->lsc_suppress_distributing = TRUE; /* send a marker frame down each port to verify the queues are empty */ LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { lp->lp_flags |= LACP_PORT_MARK; lacp_xmit_marker(lp); } /* set a timeout for the marker frames */ callout_reset(&lsc->lsc_transit_callout, LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc); } static int lacp_compare_peerinfo(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b) { return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state))); } static int lacp_compare_systemid(const struct lacp_systemid *a, const struct lacp_systemid *b) { return (memcmp(a, b, sizeof(*a))); } #if 0 /* unused */ static int lacp_compare_portid(const struct lacp_portid *a, const struct lacp_portid *b) { return (memcmp(a, b, sizeof(*a))); } #endif static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *la) { struct lacp_port *lp; uint64_t speed; lp = TAILQ_FIRST(&la->la_ports); if (lp == NULL) { return (0); } speed = ifmedia_baudrate(lp->lp_media); speed *= la->la_nports; if (speed == 0) { LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n", lp->lp_media, la->la_nports)); } return (speed); } /* * lacp_select_active_aggregator: select an aggregator to be used to transmit * packets from lagg(4) interface. */ static void lacp_select_active_aggregator(struct lacp_softc *lsc) { struct lacp_aggregator *la; struct lacp_aggregator *best_la = NULL; uint64_t best_speed = 0; char buf[LACP_LAGIDSTR_MAX+1]; LACP_TRACE(NULL); TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { uint64_t speed; if (la->la_nports == 0) { continue; } speed = lacp_aggregator_bandwidth(la); LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), speed, la->la_nports)); /* * This aggregator is chosen if the partner has a better * system priority or, the total aggregated speed is higher * or, it is already the chosen aggregator */ if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) < LACP_SYS_PRI(best_la->la_partner)) || speed > best_speed || (speed == best_speed && la == lsc->lsc_active_aggregator)) { best_la = la; best_speed = speed; } } KASSERT(best_la == NULL || best_la->la_nports > 0, ("invalid aggregator refcnt")); KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports), ("invalid aggregator list")); if (lsc->lsc_active_aggregator != best_la) { LACP_DPRINTF((NULL, "active aggregator changed\n")); LACP_DPRINTF((NULL, "old %s\n", lacp_format_lagid_aggregator(lsc->lsc_active_aggregator, buf, sizeof(buf)))); } else { LACP_DPRINTF((NULL, "active aggregator not changed\n")); } LACP_DPRINTF((NULL, "new %s\n", lacp_format_lagid_aggregator(best_la, buf, sizeof(buf)))); if (lsc->lsc_active_aggregator != best_la) { lsc->lsc_active_aggregator = best_la; lacp_update_portmap(lsc); if (best_la) { lacp_suppress_distributing(lsc, best_la); } } } /* * Updated the inactive portmap array with the new list of ports and * make it live. */ static void lacp_update_portmap(struct lacp_softc *lsc) { struct lagg_softc *sc = lsc->lsc_softc; struct lacp_aggregator *la; struct lacp_portmap *p; struct lacp_port *lp; uint64_t speed; u_int newmap; int i; #ifdef NUMA int count; uint8_t domain; #endif newmap = lsc->lsc_activemap == 0 ? 1 : 0; p = &lsc->lsc_pmap[newmap]; la = lsc->lsc_active_aggregator; speed = 0; bzero(p, sizeof(struct lacp_portmap)); if (la != NULL && la->la_nports > 0) { p->pm_count = la->la_nports; i = 0; TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) { p->pm_map[i++] = lp; #ifdef NUMA domain = lp->lp_ifp->if_numa_domain; if (domain >= MAXMEMDOM) continue; count = p->pm_numa[domain].count; p->pm_numa[domain].map[count] = lp; p->pm_numa[domain].count++; #endif } KASSERT(i == p->pm_count, ("Invalid port count")); #ifdef NUMA for (i = 0; i < MAXMEMDOM; i++) { if (p->pm_numa[i].count != 0) p->pm_num_dom++; } #endif speed = lacp_aggregator_bandwidth(la); } sc->sc_ifp->if_baudrate = speed; /* switch the active portmap over */ atomic_store_rel_int(&lsc->lsc_activemap, newmap); LACP_DPRINTF((NULL, "Set table %d with %d ports\n", lsc->lsc_activemap, lsc->lsc_pmap[lsc->lsc_activemap].pm_count)); } static uint16_t lacp_compose_key(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; u_int media = lp->lp_media; uint16_t key; if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) { /* * non-aggregatable links should have unique keys. * * XXX this isn't really unique as if_index is 16 bit. */ /* bit 0..14: (some bits of) if_index of this port */ key = lp->lp_ifp->if_index; /* bit 15: 1 */ key |= 0x8000; } else { u_int subtype = IFM_SUBTYPE(media); KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type")); KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface")); /* bit 0..4: IFM_SUBTYPE modulo speed */ switch (subtype) { case IFM_10_T: case IFM_10_2: case IFM_10_5: case IFM_10_STP: case IFM_10_FL: key = IFM_10_T; break; case IFM_100_TX: case IFM_100_FX: case IFM_100_T4: case IFM_100_VG: case IFM_100_T2: case IFM_100_T: case IFM_100_SGMII: key = IFM_100_TX; break; case IFM_1000_SX: case IFM_1000_LX: case IFM_1000_CX: case IFM_1000_T: case IFM_1000_KX: case IFM_1000_SGMII: case IFM_1000_CX_SGMII: key = IFM_1000_SX; break; case IFM_10G_LR: case IFM_10G_SR: case IFM_10G_CX4: case IFM_10G_TWINAX: case IFM_10G_TWINAX_LONG: case IFM_10G_LRM: case IFM_10G_T: case IFM_10G_KX4: case IFM_10G_KR: case IFM_10G_CR1: case IFM_10G_ER: case IFM_10G_SFI: case IFM_10G_AOC: key = IFM_10G_LR; break; case IFM_20G_KR2: key = IFM_20G_KR2; break; case IFM_2500_KX: case IFM_2500_T: case IFM_2500_X: key = IFM_2500_KX; break; case IFM_5000_T: case IFM_5000_KR: case IFM_5000_KR_S: case IFM_5000_KR1: key = IFM_5000_T; break; case IFM_50G_PCIE: case IFM_50G_CR2: case IFM_50G_KR2: case IFM_50G_KR4: case IFM_50G_SR2: case IFM_50G_LR2: case IFM_50G_LAUI2_AC: case IFM_50G_LAUI2: case IFM_50G_AUI2_AC: case IFM_50G_AUI2: case IFM_50G_CP: case IFM_50G_SR: case IFM_50G_LR: case IFM_50G_FR: case IFM_50G_KR_PAM4: case IFM_50G_AUI1_AC: case IFM_50G_AUI1: key = IFM_50G_PCIE; break; case IFM_56G_R4: key = IFM_56G_R4; break; case IFM_25G_PCIE: case IFM_25G_CR: case IFM_25G_KR: case IFM_25G_SR: case IFM_25G_LR: case IFM_25G_ACC: case IFM_25G_AOC: case IFM_25G_T: case IFM_25G_CR_S: case IFM_25G_CR1: case IFM_25G_KR_S: case IFM_25G_AUI: case IFM_25G_KR1: key = IFM_25G_PCIE; break; case IFM_40G_CR4: case IFM_40G_SR4: case IFM_40G_LR4: case IFM_40G_XLPPI: case IFM_40G_KR4: case IFM_40G_XLAUI: case IFM_40G_XLAUI_AC: case IFM_40G_ER4: key = IFM_40G_CR4; break; case IFM_100G_CR4: case IFM_100G_SR4: case IFM_100G_KR4: case IFM_100G_LR4: case IFM_100G_CAUI4_AC: case IFM_100G_CAUI4: case IFM_100G_AUI4_AC: case IFM_100G_AUI4: case IFM_100G_CR_PAM4: case IFM_100G_KR_PAM4: case IFM_100G_CP2: case IFM_100G_SR2: case IFM_100G_DR: case IFM_100G_KR2_PAM4: case IFM_100G_CAUI2_AC: case IFM_100G_CAUI2: case IFM_100G_AUI2_AC: case IFM_100G_AUI2: key = IFM_100G_CR4; break; case IFM_200G_CR4_PAM4: case IFM_200G_SR4: case IFM_200G_FR4: case IFM_200G_LR4: case IFM_200G_DR4: case IFM_200G_KR4_PAM4: case IFM_200G_AUI4_AC: case IFM_200G_AUI4: case IFM_200G_AUI8_AC: case IFM_200G_AUI8: key = IFM_200G_CR4_PAM4; break; case IFM_400G_FR8: case IFM_400G_LR8: case IFM_400G_DR4: case IFM_400G_AUI8_AC: case IFM_400G_AUI8: key = IFM_400G_FR8; break; default: key = subtype; break; } /* bit 5..14: (some bits of) if_index of lagg device */ key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5); /* bit 15: 0 */ } return (htons(key)); } static void lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la) { char buf[LACP_LAGIDSTR_MAX+1]; LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_refcnt, la->la_refcnt + 1)); KASSERT(la->la_refcnt > 0, ("refcount <= 0")); la->la_refcnt++; KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount")); } static void lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la) { char buf[LACP_LAGIDSTR_MAX+1]; LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_refcnt, la->la_refcnt - 1)); KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt")); la->la_refcnt--; if (la->la_refcnt > 0) { return; } KASSERT(la->la_refcnt == 0, ("refcount not zero")); KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active")); TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q); free(la, M_DEVBUF); } /* * lacp_aggregator_get: allocate an aggregator. */ static struct lacp_aggregator * lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp) { struct lacp_aggregator *la; la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT); if (la) { la->la_refcnt = 1; la->la_nports = 0; TAILQ_INIT(&la->la_ports); la->la_pending = 0; TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q); } return (la); } /* * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port. */ static void lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp) { lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner); lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor); la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION; } static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr, const struct lacp_peerinfo *lpi_port) { memset(lpi_aggr, 0, sizeof(*lpi_aggr)); lpi_aggr->lip_systemid = lpi_port->lip_systemid; lpi_aggr->lip_key = lpi_port->lip_key; } /* * lacp_aggregator_is_compatible: check if a port can join to an aggregator. */ static int lacp_aggregator_is_compatible(const struct lacp_aggregator *la, const struct lacp_port *lp) { if (!(lp->lp_state & LACP_STATE_AGGREGATION) || !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) { return (0); } if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) { return (0); } if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) { return (0); } if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) { return (0); } return (1); } static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b) { if (memcmp(&a->lip_systemid, &b->lip_systemid, sizeof(a->lip_systemid))) { return (0); } if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) { return (0); } return (1); } static void lacp_port_enable(struct lacp_port *lp) { lp->lp_state |= LACP_STATE_AGGREGATION; } static void lacp_port_disable(struct lacp_port *lp) { lacp_set_mux(lp, LACP_MUX_DETACHED); lp->lp_state &= ~LACP_STATE_AGGREGATION; lp->lp_selected = LACP_UNSELECTED; lacp_sm_rx_record_default(lp); lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION; lp->lp_state &= ~LACP_STATE_EXPIRED; } /* * lacp_select: select an aggregator. create one if necessary. */ static void lacp_select(struct lacp_port *lp) { struct lacp_softc *lsc = lp->lp_lsc; struct lacp_aggregator *la; char buf[LACP_LAGIDSTR_MAX+1]; if (lp->lp_aggregator) { return; } /* If we haven't heard from our peer, skip this step. */ if (lp->lp_state & LACP_STATE_DEFAULTED) return; KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); LACP_DPRINTF((lp, "port lagid=%s\n", lacp_format_lagid(&lp->lp_actor, &lp->lp_partner, buf, sizeof(buf)))); TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { if (lacp_aggregator_is_compatible(la, lp)) { break; } } if (la == NULL) { la = lacp_aggregator_get(lsc, lp); if (la == NULL) { LACP_DPRINTF((lp, "aggregator creation failed\n")); /* * will retry on the next tick. */ return; } lacp_fill_aggregator_id(la, lp); LACP_DPRINTF((lp, "aggregator created\n")); } else { LACP_DPRINTF((lp, "compatible aggregator found\n")); if (la->la_refcnt == LACP_MAX_PORTS) return; lacp_aggregator_addref(lsc, la); } LACP_DPRINTF((lp, "aggregator lagid=%s\n", lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)))); lp->lp_aggregator = la; lp->lp_selected = LACP_SELECTED; } /* * lacp_unselect: finish unselect/detach process. */ static void lacp_unselect(struct lacp_port *lp) { struct lacp_softc *lsc = lp->lp_lsc; struct lacp_aggregator *la = lp->lp_aggregator; KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); if (la == NULL) { return; } lp->lp_aggregator = NULL; lacp_aggregator_delref(lsc, la); } /* mux machine */ static void lacp_sm_mux(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; enum lacp_mux_state new_state; boolean_t p_sync = (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0; boolean_t p_collecting = (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0; enum lacp_selected selected = lp->lp_selected; struct lacp_aggregator *la; if (V_lacp_debug > 1) lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, " "p_sync= 0x%x, p_collecting= 0x%x\n", __func__, lp->lp_mux_state, selected, p_sync, p_collecting); re_eval: la = lp->lp_aggregator; KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL, ("MUX not detached")); new_state = lp->lp_mux_state; switch (lp->lp_mux_state) { case LACP_MUX_DETACHED: if (selected != LACP_UNSELECTED) { new_state = LACP_MUX_WAITING; } break; case LACP_MUX_WAITING: KASSERT(la->la_pending > 0 || !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); if (selected == LACP_SELECTED && la->la_pending == 0) { new_state = LACP_MUX_ATTACHED; } else if (selected == LACP_UNSELECTED) { new_state = LACP_MUX_DETACHED; } break; case LACP_MUX_ATTACHED: if (selected == LACP_SELECTED && p_sync) { new_state = LACP_MUX_COLLECTING; } else if (selected != LACP_SELECTED) { new_state = LACP_MUX_DETACHED; } break; case LACP_MUX_COLLECTING: if (selected == LACP_SELECTED && p_sync && p_collecting) { new_state = LACP_MUX_DISTRIBUTING; } else if (selected != LACP_SELECTED || !p_sync) { new_state = LACP_MUX_ATTACHED; } break; case LACP_MUX_DISTRIBUTING: if (selected != LACP_SELECTED || !p_sync || !p_collecting) { new_state = LACP_MUX_COLLECTING; lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n"); sc->sc_flapping++; } break; default: panic("%s: unknown state", __func__); } if (lp->lp_mux_state == new_state) { return; } lacp_set_mux(lp, new_state); goto re_eval; } static void lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state) { struct lacp_aggregator *la = lp->lp_aggregator; if (lp->lp_mux_state == new_state) { return; } switch (new_state) { case LACP_MUX_DETACHED: lp->lp_state &= ~LACP_STATE_SYNC; lacp_disable_distributing(lp); lacp_disable_collecting(lp); lacp_sm_assert_ntt(lp); /* cancel timer */ if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) { KASSERT(la->la_pending > 0, ("timer_wait_while not active")); la->la_pending--; } LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE); lacp_unselect(lp); break; case LACP_MUX_WAITING: LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE, LACP_AGGREGATE_WAIT_TIME); la->la_pending++; break; case LACP_MUX_ATTACHED: lp->lp_state |= LACP_STATE_SYNC; lacp_disable_collecting(lp); lacp_sm_assert_ntt(lp); break; case LACP_MUX_COLLECTING: lacp_enable_collecting(lp); lacp_disable_distributing(lp); lacp_sm_assert_ntt(lp); break; case LACP_MUX_DISTRIBUTING: lacp_enable_distributing(lp); break; default: panic("%s: unknown state", __func__); } LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state)); lp->lp_mux_state = new_state; } static void lacp_sm_mux_timer(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; char buf[LACP_LAGIDSTR_MAX+1]; KASSERT(la->la_pending > 0, ("no pending event")); LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_pending, la->la_pending - 1)); la->la_pending--; } /* periodic transmit machine */ static void lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate) { if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state, LACP_STATE_TIMEOUT)) { return; } LACP_DPRINTF((lp, "partner timeout changed\n")); /* * FAST_PERIODIC -> SLOW_PERIODIC * or * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC * * let lacp_sm_ptx_tx_schedule to update timeout. */ LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); /* * if timeout has been shortened, assert NTT. */ if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) { lacp_sm_assert_ntt(lp); } } static void lacp_sm_ptx_tx_schedule(struct lacp_port *lp) { int timeout; if (!(lp->lp_state & LACP_STATE_ACTIVITY) && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) { /* * NO_PERIODIC */ LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); return; } if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) { return; } timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ? LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME; LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout); } static void lacp_sm_ptx_timer(struct lacp_port *lp) { lacp_sm_assert_ntt(lp); } static void lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du) { int timeout; /* * check LACP_DISABLED first */ if (!(lp->lp_state & LACP_STATE_AGGREGATION)) { return; } /* * check loopback condition. */ if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid, &lp->lp_actor.lip_systemid)) { return; } /* * EXPIRED, DEFAULTED, CURRENT -> CURRENT */ lacp_sm_rx_update_selected(lp, du); lacp_sm_rx_update_ntt(lp, du); lacp_sm_rx_record_pdu(lp, du); timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ? LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME; LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout); lp->lp_state &= ~LACP_STATE_EXPIRED; /* * kick transmit machine without waiting the next tick. */ lacp_sm_tx(lp); } static void lacp_sm_rx_set_expired(struct lacp_port *lp) { lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT; LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME); lp->lp_state |= LACP_STATE_EXPIRED; } static void lacp_sm_rx_timer(struct lacp_port *lp) { if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) { /* CURRENT -> EXPIRED */ LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__)); lacp_sm_rx_set_expired(lp); } else { /* EXPIRED -> DEFAULTED */ LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__)); lacp_sm_rx_update_default_selected(lp); lacp_sm_rx_record_default(lp); lp->lp_state &= ~LACP_STATE_EXPIRED; } } static void lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du) { boolean_t active; uint8_t oldpstate; char buf[LACP_STATESTR_MAX+1]; LACP_TRACE(lp); oldpstate = lp->lp_partner.lip_state; active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY) || ((lp->lp_state & LACP_STATE_ACTIVITY) && (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY)); lp->lp_partner = du->ldu_actor; if (active && ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, LACP_STATE_AGGREGATION) && !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner)) || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) { /* * XXX Maintain legacy behavior of leaving the * LACP_STATE_SYNC bit unchanged from the partner's * advertisement if lsc_strict_mode is false. * TODO: We should re-examine the concept of the "strict mode" * to ensure it makes sense to maintain a non-strict mode. */ if (lp->lp_lsc->lsc_strict_mode) lp->lp_partner.lip_state |= LACP_STATE_SYNC; } else { lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; } lp->lp_state &= ~LACP_STATE_DEFAULTED; if (oldpstate != lp->lp_partner.lip_state) { LACP_DPRINTF((lp, "old pstate %s\n", lacp_format_state(oldpstate, buf, sizeof(buf)))); LACP_DPRINTF((lp, "new pstate %s\n", lacp_format_state(lp->lp_partner.lip_state, buf, sizeof(buf)))); } lacp_sm_ptx_update_timeout(lp, oldpstate); } static void lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du) { LACP_TRACE(lp); if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) || !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) { LACP_DPRINTF((lp, "%s: assert ntt\n", __func__)); lacp_sm_assert_ntt(lp); } } static void lacp_sm_rx_record_default(struct lacp_port *lp) { uint8_t oldpstate; LACP_TRACE(lp); oldpstate = lp->lp_partner.lip_state; if (lp->lp_lsc->lsc_strict_mode) lp->lp_partner = lacp_partner_admin_strict; else lp->lp_partner = lacp_partner_admin_optimistic; lp->lp_state |= LACP_STATE_DEFAULTED; lacp_sm_ptx_update_timeout(lp, oldpstate); } static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp, const struct lacp_peerinfo *info) { LACP_TRACE(lp); if (lacp_compare_peerinfo(&lp->lp_partner, info) || !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state, LACP_STATE_AGGREGATION)) { lp->lp_selected = LACP_UNSELECTED; /* mux machine will clean up lp->lp_aggregator */ } } static void lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du) { LACP_TRACE(lp); lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor); } static void lacp_sm_rx_update_default_selected(struct lacp_port *lp) { LACP_TRACE(lp); if (lp->lp_lsc->lsc_strict_mode) lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin_strict); else lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin_optimistic); } /* transmit machine */ static void lacp_sm_tx(struct lacp_port *lp) { int error = 0; if (!(lp->lp_state & LACP_STATE_AGGREGATION) #if 1 || (!(lp->lp_state & LACP_STATE_ACTIVITY) && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) #endif ) { lp->lp_flags &= ~LACP_PORT_NTT; } if (!(lp->lp_flags & LACP_PORT_NTT)) { return; } /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */ if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent, (3 / LACP_FAST_PERIODIC_TIME)) == 0) { LACP_DPRINTF((lp, "rate limited pdu\n")); return; } if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) { error = lacp_xmit_lacpdu(lp); } else { LACP_TPRINTF((lp, "Dropping TX PDU\n")); } if (error == 0) { lp->lp_flags &= ~LACP_PORT_NTT; } else { LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n", error)); } } static void lacp_sm_assert_ntt(struct lacp_port *lp) { lp->lp_flags |= LACP_PORT_NTT; } static void lacp_run_timers(struct lacp_port *lp) { int i; for (i = 0; i < LACP_NTIMER; i++) { KASSERT(lp->lp_timer[i] >= 0, ("invalid timer value %d", lp->lp_timer[i])); if (lp->lp_timer[i] == 0) { continue; } else if (--lp->lp_timer[i] <= 0) { if (lacp_timer_funcs[i]) { (*lacp_timer_funcs[i])(lp); } } } } int lacp_marker_input(struct lacp_port *lp, struct mbuf *m) { struct lacp_softc *lsc = lp->lp_lsc; struct lagg_port *lgp = lp->lp_lagg; struct lacp_port *lp2; struct markerdu *mdu; int error = 0; int pending = 0; if (m->m_pkthdr.len != sizeof(*mdu)) { goto bad; } if ((m->m_flags & M_MCAST) == 0) { goto bad; } if (m->m_len < sizeof(*mdu)) { m = m_pullup(m, sizeof(*mdu)); if (m == NULL) { return (ENOMEM); } } mdu = mtod(m, struct markerdu *); if (memcmp(&mdu->mdu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { goto bad; } if (mdu->mdu_sph.sph_version != 1) { goto bad; } switch (mdu->mdu_tlv.tlv_type) { case MARKER_TYPE_INFO: if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, marker_info_tlv_template, TRUE)) { goto bad; } mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE; memcpy(&mdu->mdu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); error = lagg_enqueue(lp->lp_ifp, m); break; case MARKER_TYPE_RESPONSE: if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, marker_response_tlv_template, TRUE)) { goto bad; } LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n", ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", ntohl(mdu->mdu_info.mi_rq_xid))); /* Verify that it is the last marker we sent out */ if (memcmp(&mdu->mdu_info, &lp->lp_marker, sizeof(struct lacp_markerinfo))) goto bad; LACP_LOCK(lsc); lp->lp_flags &= ~LACP_PORT_MARK; if (lsc->lsc_suppress_distributing) { /* Check if any ports are waiting for a response */ LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) { if (lp2->lp_flags & LACP_PORT_MARK) { pending = 1; break; } } if (pending == 0) { /* All interface queues are clear */ LACP_DPRINTF((NULL, "queue flush complete\n")); lsc->lsc_suppress_distributing = FALSE; } } LACP_UNLOCK(lsc); m_freem(m); break; default: goto bad; } return (error); bad: LACP_DPRINTF((lp, "bad marker frame\n")); m_freem(m); return (EINVAL); } static int tlv_check(const void *p, size_t size, const struct tlvhdr *tlv, const struct tlv_template *tmpl, boolean_t check_type) { while (/* CONSTCOND */ 1) { if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) { return (EINVAL); } if ((check_type && tlv->tlv_type != tmpl->tmpl_type) || tlv->tlv_length != tmpl->tmpl_length) { return (EINVAL); } if (tmpl->tmpl_type == 0) { break; } tlv = (const struct tlvhdr *) ((const char *)tlv + tlv->tlv_length); tmpl++; } return (0); } /* Debugging */ const char * lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen) { snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X", (int)mac[0], (int)mac[1], (int)mac[2], (int)mac[3], (int)mac[4], (int)mac[5]); return (buf); } const char * lacp_format_systemid(const struct lacp_systemid *sysid, char *buf, size_t buflen) { char macbuf[LACP_MACSTR_MAX+1]; snprintf(buf, buflen, "%04X,%s", ntohs(sysid->lsi_prio), lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf))); return (buf); } const char * lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen) { snprintf(buf, buflen, "%04X,%04X", ntohs(portid->lpi_prio), ntohs(portid->lpi_portno)); return (buf); } const char * lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen) { char sysid[LACP_SYSTEMIDSTR_MAX+1]; char portid[LACP_PORTIDSTR_MAX+1]; snprintf(buf, buflen, "(%s,%04X,%s)", lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)), ntohs(peer->lip_key), lacp_format_portid(&peer->lip_portid, portid, sizeof(portid))); return (buf); } const char * lacp_format_lagid(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b, char *buf, size_t buflen) { char astr[LACP_PARTNERSTR_MAX+1]; char bstr[LACP_PARTNERSTR_MAX+1]; #if 0 /* * there's a convention to display small numbered peer * in the left. */ if (lacp_compare_peerinfo(a, b) > 0) { const struct lacp_peerinfo *t; t = a; a = b; b = t; } #endif snprintf(buf, buflen, "[%s,%s]", lacp_format_partner(a, astr, sizeof(astr)), lacp_format_partner(b, bstr, sizeof(bstr))); return (buf); } const char * lacp_format_lagid_aggregator(const struct lacp_aggregator *la, char *buf, size_t buflen) { if (la == NULL) { return ("(none)"); } return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen)); } const char * lacp_format_state(uint8_t state, char *buf, size_t buflen) { snprintf(buf, buflen, "%b", state, LACP_STATE_BITS); return (buf); } static void lacp_dump_lacpdu(const struct lacpdu *du) { char buf[LACP_PARTNERSTR_MAX+1]; char buf2[LACP_STATESTR_MAX+1]; printf("actor=%s\n", lacp_format_partner(&du->ldu_actor, buf, sizeof(buf))); printf("actor.state=%s\n", lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2))); printf("partner=%s\n", lacp_format_partner(&du->ldu_partner, buf, sizeof(buf))); printf("partner.state=%s\n", lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2))); printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay)); } static void lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...) { va_list va; if (lp) { printf("%s: ", lp->lp_ifp->if_xname); } va_start(va, fmt); vprintf(fmt, va); va_end(va); } Index: head/sys/net/ieee8023ad_lacp.h =================================================================== --- head/sys/net/ieee8023ad_lacp.h (revision 358807) +++ head/sys/net/ieee8023ad_lacp.h (revision 358808) @@ -1,353 +1,351 @@ /* $NetBSD: ieee8023ad_impl.h,v 1.2 2005/12/10 23:21:39 elad Exp $ */ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c)2005 YAMAMOTO Takashi, * 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 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 AUTHOR 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. * * $FreeBSD$ */ /* * IEEE802.3ad LACP * * implementation details. */ #define LACP_TIMER_CURRENT_WHILE 0 #define LACP_TIMER_PERIODIC 1 #define LACP_TIMER_WAIT_WHILE 2 #define LACP_NTIMER 3 #define LACP_TIMER_ARM(port, timer, val) \ (port)->lp_timer[(timer)] = (val) #define LACP_TIMER_DISARM(port, timer) \ (port)->lp_timer[(timer)] = 0 #define LACP_TIMER_ISARMED(port, timer) \ ((port)->lp_timer[(timer)] > 0) /* * IEEE802.3ad LACP * * protocol definitions. */ #define LACP_STATE_ACTIVITY (1<<0) #define LACP_STATE_TIMEOUT (1<<1) #define LACP_STATE_AGGREGATION (1<<2) #define LACP_STATE_SYNC (1<<3) #define LACP_STATE_COLLECTING (1<<4) #define LACP_STATE_DISTRIBUTING (1<<5) #define LACP_STATE_DEFAULTED (1<<6) #define LACP_STATE_EXPIRED (1<<7) #define LACP_PORT_NTT 0x00000001 #define LACP_PORT_MARK 0x00000002 #define LACP_STATE_BITS \ "\020" \ "\001ACTIVITY" \ "\002TIMEOUT" \ "\003AGGREGATION" \ "\004SYNC" \ "\005COLLECTING" \ "\006DISTRIBUTING" \ "\007DEFAULTED" \ "\010EXPIRED" #ifdef _KERNEL /* * IEEE802.3 slow protocols * * protocol (on-wire) definitions. * * XXX should be elsewhere. */ #define SLOWPROTOCOLS_SUBTYPE_LACP 1 #define SLOWPROTOCOLS_SUBTYPE_MARKER 2 struct slowprothdr { uint8_t sph_subtype; uint8_t sph_version; } __packed; /* * TLV on-wire structure. */ struct tlvhdr { uint8_t tlv_type; uint8_t tlv_length; /* uint8_t tlv_value[]; */ } __packed; /* * ... and our implementation. */ #define TLV_SET(tlv, type, length) \ do { \ (tlv)->tlv_type = (type); \ (tlv)->tlv_length = sizeof(*tlv) + (length); \ } while (/*CONSTCOND*/0) struct tlv_template { uint8_t tmpl_type; uint8_t tmpl_length; }; struct lacp_systemid { uint16_t lsi_prio; uint8_t lsi_mac[6]; } __packed; struct lacp_portid { uint16_t lpi_prio; uint16_t lpi_portno; } __packed; struct lacp_peerinfo { struct lacp_systemid lip_systemid; uint16_t lip_key; struct lacp_portid lip_portid; uint8_t lip_state; uint8_t lip_resv[3]; } __packed; struct lacp_collectorinfo { uint16_t lci_maxdelay; uint8_t lci_resv[12]; } __packed; struct lacpdu { struct ether_header ldu_eh; struct slowprothdr ldu_sph; struct tlvhdr ldu_tlv_actor; struct lacp_peerinfo ldu_actor; struct tlvhdr ldu_tlv_partner; struct lacp_peerinfo ldu_partner; struct tlvhdr ldu_tlv_collector; struct lacp_collectorinfo ldu_collector; struct tlvhdr ldu_tlv_term; uint8_t ldu_resv[50]; } __packed; /* * IEEE802.3ad marker protocol * * protocol (on-wire) definitions. */ struct lacp_markerinfo { uint16_t mi_rq_port; uint8_t mi_rq_system[ETHER_ADDR_LEN]; uint32_t mi_rq_xid; uint8_t mi_pad[2]; } __packed; struct markerdu { struct ether_header mdu_eh; struct slowprothdr mdu_sph; struct tlvhdr mdu_tlv; struct lacp_markerinfo mdu_info; struct tlvhdr mdu_tlv_term; uint8_t mdu_resv[90]; } __packed; #define MARKER_TYPE_INFO 0x01 #define MARKER_TYPE_RESPONSE 0x02 enum lacp_selected { LACP_UNSELECTED, LACP_STANDBY, /* not used in this implementation */ LACP_SELECTED, }; enum lacp_mux_state { LACP_MUX_DETACHED, LACP_MUX_WAITING, LACP_MUX_ATTACHED, LACP_MUX_COLLECTING, LACP_MUX_DISTRIBUTING, }; #define LACP_MAX_PORTS 32 struct lacp_numa { int count; struct lacp_port *map[LACP_MAX_PORTS]; }; struct lacp_portmap { int pm_count; int pm_num_dom; struct lacp_numa pm_numa[MAXMEMDOM]; struct lacp_port *pm_map[LACP_MAX_PORTS]; }; struct lacp_port { TAILQ_ENTRY(lacp_port) lp_dist_q; LIST_ENTRY(lacp_port) lp_next; struct lacp_softc *lp_lsc; struct lagg_port *lp_lagg; struct ifnet *lp_ifp; struct lacp_peerinfo lp_partner; struct lacp_peerinfo lp_actor; struct lacp_markerinfo lp_marker; #define lp_state lp_actor.lip_state #define lp_key lp_actor.lip_key #define lp_systemid lp_actor.lip_systemid struct timeval lp_last_lacpdu; int lp_lacpdu_sent; enum lacp_mux_state lp_mux_state; enum lacp_selected lp_selected; int lp_flags; u_int lp_media; /* XXX redundant */ int lp_timer[LACP_NTIMER]; struct ifmultiaddr *lp_ifma; struct lacp_aggregator *lp_aggregator; }; struct lacp_aggregator { TAILQ_ENTRY(lacp_aggregator) la_q; int la_refcnt; /* num of ports which selected us */ int la_nports; /* num of distributing ports */ TAILQ_HEAD(, lacp_port) la_ports; /* distributing ports */ struct lacp_peerinfo la_partner; struct lacp_peerinfo la_actor; int la_pending; /* number of ports in wait_while */ }; struct lacp_softc { struct lagg_softc *lsc_softc; struct mtx lsc_mtx; struct lacp_aggregator *lsc_active_aggregator; TAILQ_HEAD(, lacp_aggregator) lsc_aggregators; boolean_t lsc_suppress_distributing; struct callout lsc_transit_callout; struct callout lsc_callout; LIST_HEAD(, lacp_port) lsc_ports; struct lacp_portmap lsc_pmap[2]; volatile u_int lsc_activemap; u_int32_t lsc_hashkey; struct { u_int32_t lsc_rx_test; u_int32_t lsc_tx_test; } lsc_debug; u_int32_t lsc_strict_mode; boolean_t lsc_fast_timeout; /* if set, fast timeout */ }; #define LACP_TYPE_ACTORINFO 1 #define LACP_TYPE_PARTNERINFO 2 #define LACP_TYPE_COLLECTORINFO 3 /* timeout values (in sec) */ #define LACP_FAST_PERIODIC_TIME (1) #define LACP_SLOW_PERIODIC_TIME (30) #define LACP_SHORT_TIMEOUT_TIME (3 * LACP_FAST_PERIODIC_TIME) #define LACP_LONG_TIMEOUT_TIME (3 * LACP_SLOW_PERIODIC_TIME) #define LACP_CHURN_DETECTION_TIME (60) #define LACP_AGGREGATE_WAIT_TIME (2) #define LACP_TRANSIT_DELAY 3000 /* in msec */ #define LACP_STATE_EQ(s1, s2, mask) \ ((((s1) ^ (s2)) & (mask)) == 0) #define LACP_SYS_PRI(peer) (peer).lip_systemid.lsi_prio #define LACP_PORT(_lp) ((struct lacp_port *)(_lp)->lp_psc) #define LACP_SOFTC(_sc) ((struct lacp_softc *)(_sc)->sc_psc) #define LACP_LOCK_INIT(_lsc) mtx_init(&(_lsc)->lsc_mtx, \ "lacp mtx", NULL, MTX_DEF) #define LACP_LOCK_DESTROY(_lsc) mtx_destroy(&(_lsc)->lsc_mtx) #define LACP_LOCK(_lsc) mtx_lock(&(_lsc)->lsc_mtx) #define LACP_UNLOCK(_lsc) mtx_unlock(&(_lsc)->lsc_mtx) #define LACP_LOCK_ASSERT(_lsc) mtx_assert(&(_lsc)->lsc_mtx, MA_OWNED) struct mbuf *lacp_input(struct lagg_port *, struct mbuf *); struct lagg_port *lacp_select_tx_port(struct lagg_softc *, struct mbuf *); -#if defined(RATELIMIT) || defined(KERN_TLS) -struct lagg_port *lacp_select_tx_port_by_hash(struct lagg_softc *, uint32_t); -#endif +struct lagg_port *lacp_select_tx_port_by_hash(struct lagg_softc *, uint32_t, uint8_t); void lacp_attach(struct lagg_softc *); void lacp_detach(void *); void lacp_init(struct lagg_softc *); void lacp_stop(struct lagg_softc *); int lacp_port_create(struct lagg_port *); void lacp_port_destroy(struct lagg_port *); void lacp_linkstate(struct lagg_port *); void lacp_req(struct lagg_softc *, void *); void lacp_portreq(struct lagg_port *, void *); static __inline int lacp_isactive(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; struct lacp_aggregator *la = lp->lp_aggregator; /* This port is joined to the active aggregator */ if (la != NULL && la == lsc->lsc_active_aggregator) return (1); return (0); } static __inline int lacp_iscollecting(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); return ((lp->lp_state & LACP_STATE_COLLECTING) != 0); } static __inline int lacp_isdistributing(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); return ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0); } /* following constants don't include terminating NUL */ #define LACP_MACSTR_MAX (2*6 + 5) #define LACP_SYSTEMPRIOSTR_MAX (4) #define LACP_SYSTEMIDSTR_MAX (LACP_SYSTEMPRIOSTR_MAX + 1 + LACP_MACSTR_MAX) #define LACP_PORTPRIOSTR_MAX (4) #define LACP_PORTNOSTR_MAX (4) #define LACP_PORTIDSTR_MAX (LACP_PORTPRIOSTR_MAX + 1 + LACP_PORTNOSTR_MAX) #define LACP_KEYSTR_MAX (4) #define LACP_PARTNERSTR_MAX \ (1 + LACP_SYSTEMIDSTR_MAX + 1 + LACP_KEYSTR_MAX + 1 \ + LACP_PORTIDSTR_MAX + 1) #define LACP_LAGIDSTR_MAX \ (1 + LACP_PARTNERSTR_MAX + 1 + LACP_PARTNERSTR_MAX + 1) #define LACP_STATESTR_MAX (255) /* XXX */ #endif /* _KERNEL */ Index: head/sys/net/if_lagg.c =================================================================== --- head/sys/net/if_lagg.c (revision 358807) +++ head/sys/net/if_lagg.c (revision 358808) @@ -1,2423 +1,2426 @@ /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ /* * Copyright (c) 2005, 2006 Reyk Floeter * Copyright (c) 2007 Andrew Thompson * Copyright (c) 2014, 2016 Marcelo Araujo * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_kern_tls.h" #include "opt_ratelimit.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 #if defined(INET) || defined(INET6) #include #include #endif #ifdef INET #include #include #endif #ifdef INET6 #include #include #include #endif #include #include #include #ifdef INET6 /* * XXX: declare here to avoid to include many inet6 related files.. * should be more generalized? */ extern void nd6_setmtu(struct ifnet *); #endif #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et) #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et) #define LAGG_RLOCK_ASSERT() NET_EPOCH_ASSERT() #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt)) #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx") #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx) #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx) #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx) #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED) #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED) /* Special flags we should propagate to the lagg ports. */ static struct { int flag; int (*func)(struct ifnet *, int); } lagg_pflags[] = { {IFF_PROMISC, ifpromisc}, {IFF_ALLMULTI, if_allmulti}, {0, NULL} }; struct lagg_snd_tag { struct m_snd_tag com; struct m_snd_tag *tag; }; VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ #define V_lagg_list VNET(lagg_list) VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx); #define V_lagg_list_mtx VNET(lagg_list_mtx) #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ "if_lagg list", NULL, MTX_DEF) #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) eventhandler_tag lagg_detach_cookie = NULL; static int lagg_clone_create(struct if_clone *, int, caddr_t); static void lagg_clone_destroy(struct ifnet *); VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner); #define V_lagg_cloner VNET(lagg_cloner) static const char laggname[] = "lagg"; static MALLOC_DEFINE(M_LAGG, laggname, "802.3AD Link Aggregation Interface"); static void lagg_capabilities(struct lagg_softc *); static int lagg_port_create(struct lagg_softc *, struct ifnet *); static int lagg_port_destroy(struct lagg_port *, int); static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); static void lagg_linkstate(struct lagg_softc *); static void lagg_port_state(struct ifnet *, int); static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); static int lagg_port_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); #ifdef LAGG_PORT_STACKING static int lagg_port_checkstacking(struct lagg_softc *); #endif static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); static void lagg_init(void *); static void lagg_stop(struct lagg_softc *); static int lagg_ioctl(struct ifnet *, u_long, caddr_t); #if defined(KERN_TLS) || defined(RATELIMIT) static int lagg_snd_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *, struct m_snd_tag **); static int lagg_snd_tag_modify(struct m_snd_tag *, union if_snd_tag_modify_params *); static int lagg_snd_tag_query(struct m_snd_tag *, union if_snd_tag_query_params *); static void lagg_snd_tag_free(struct m_snd_tag *); static void lagg_ratelimit_query(struct ifnet *, struct if_ratelimit_query_results *); #endif static int lagg_setmulti(struct lagg_port *); static int lagg_clrmulti(struct lagg_port *); static int lagg_setcaps(struct lagg_port *, int cap); static int lagg_setflag(struct lagg_port *, int, int, int (*func)(struct ifnet *, int)); static int lagg_setflags(struct lagg_port *, int status); static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); static int lagg_transmit(struct ifnet *, struct mbuf *); static void lagg_qflush(struct ifnet *); static int lagg_media_change(struct ifnet *); static void lagg_media_status(struct ifnet *, struct ifmediareq *); static struct lagg_port *lagg_link_active(struct lagg_softc *, struct lagg_port *); /* Simple round robin */ static void lagg_rr_attach(struct lagg_softc *); static int lagg_rr_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* Active failover */ static int lagg_fail_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* Loadbalancing */ static void lagg_lb_attach(struct lagg_softc *); static void lagg_lb_detach(struct lagg_softc *); static int lagg_lb_port_create(struct lagg_port *); static void lagg_lb_port_destroy(struct lagg_port *); static int lagg_lb_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); /* Broadcast */ static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* 802.3ad LACP */ static void lagg_lacp_attach(struct lagg_softc *); static void lagg_lacp_detach(struct lagg_softc *); static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); static void lagg_lacp_lladdr(struct lagg_softc *); /* lagg protocol table */ static const struct lagg_proto { lagg_proto pr_num; void (*pr_attach)(struct lagg_softc *); void (*pr_detach)(struct lagg_softc *); int (*pr_start)(struct lagg_softc *, struct mbuf *); struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, struct mbuf *); int (*pr_addport)(struct lagg_port *); void (*pr_delport)(struct lagg_port *); void (*pr_linkstate)(struct lagg_port *); void (*pr_init)(struct lagg_softc *); void (*pr_stop)(struct lagg_softc *); void (*pr_lladdr)(struct lagg_softc *); void (*pr_request)(struct lagg_softc *, void *); void (*pr_portreq)(struct lagg_port *, void *); } lagg_protos[] = { { .pr_num = LAGG_PROTO_NONE }, { .pr_num = LAGG_PROTO_ROUNDROBIN, .pr_attach = lagg_rr_attach, .pr_start = lagg_rr_start, .pr_input = lagg_rr_input, }, { .pr_num = LAGG_PROTO_FAILOVER, .pr_start = lagg_fail_start, .pr_input = lagg_fail_input, }, { .pr_num = LAGG_PROTO_LOADBALANCE, .pr_attach = lagg_lb_attach, .pr_detach = lagg_lb_detach, .pr_start = lagg_lb_start, .pr_input = lagg_lb_input, .pr_addport = lagg_lb_port_create, .pr_delport = lagg_lb_port_destroy, }, { .pr_num = LAGG_PROTO_LACP, .pr_attach = lagg_lacp_attach, .pr_detach = lagg_lacp_detach, .pr_start = lagg_lacp_start, .pr_input = lagg_lacp_input, .pr_addport = lacp_port_create, .pr_delport = lacp_port_destroy, .pr_linkstate = lacp_linkstate, .pr_init = lacp_init, .pr_stop = lacp_stop, .pr_lladdr = lagg_lacp_lladdr, .pr_request = lacp_req, .pr_portreq = lacp_portreq, }, { .pr_num = LAGG_PROTO_BROADCAST, .pr_start = lagg_bcast_start, .pr_input = lagg_bcast_input, }, }; SYSCTL_DECL(_net_link); SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Link Aggregation"); /* Allow input on any failover links */ VNET_DEFINE_STATIC(int, lagg_failover_rx_all); #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(lagg_failover_rx_all), 0, "Accept input from any interface in a failover lagg"); /* Default value for using flowid */ VNET_DEFINE_STATIC(int, def_use_flowid) = 0; #define V_def_use_flowid VNET(def_use_flowid) SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, &VNET_NAME(def_use_flowid), 0, "Default setting for using flow id for load sharing"); /* Default value for using numa */ VNET_DEFINE_STATIC(int, def_use_numa) = 1; #define V_def_use_numa VNET(def_use_numa) SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_numa, CTLFLAG_RWTUN, &VNET_NAME(def_use_numa), 0, "Use numa to steer flows"); /* Default value for flowid shift */ VNET_DEFINE_STATIC(int, def_flowid_shift) = 16; #define V_def_flowid_shift VNET(def_flowid_shift) SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, &VNET_NAME(def_flowid_shift), 0, "Default setting for flowid shift for load sharing"); static void vnet_lagg_init(const void *unused __unused) { LAGG_LIST_LOCK_INIT(); SLIST_INIT(&V_lagg_list); V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, lagg_clone_destroy, 0); } VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_lagg_init, NULL); static void vnet_lagg_uninit(const void *unused __unused) { if_clone_detach(V_lagg_cloner); LAGG_LIST_LOCK_DESTROY(); } VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_lagg_uninit, NULL); static int lagg_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: lagg_input_p = lagg_input; lagg_linkstate_p = lagg_port_state; lagg_detach_cookie = EVENTHANDLER_REGISTER( ifnet_departure_event, lagg_port_ifdetach, NULL, EVENTHANDLER_PRI_ANY); break; case MOD_UNLOAD: EVENTHANDLER_DEREGISTER(ifnet_departure_event, lagg_detach_cookie); lagg_input_p = NULL; lagg_linkstate_p = NULL; break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t lagg_mod = { "if_lagg", lagg_modevent, 0 }; DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_lagg, 1); static void lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) { LAGG_XLOCK_ASSERT(sc); KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", __func__, sc)); if (sc->sc_ifflags & IFF_DEBUG) if_printf(sc->sc_ifp, "using proto %u\n", pr); if (lagg_protos[pr].pr_attach != NULL) lagg_protos[pr].pr_attach(sc); sc->sc_proto = pr; } static void lagg_proto_detach(struct lagg_softc *sc) { lagg_proto pr; LAGG_XLOCK_ASSERT(sc); pr = sc->sc_proto; sc->sc_proto = LAGG_PROTO_NONE; if (lagg_protos[pr].pr_detach != NULL) lagg_protos[pr].pr_detach(sc); } static int lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) { return (lagg_protos[sc->sc_proto].pr_start(sc, m)); } static struct mbuf * lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); } static int lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_addport == NULL) return (0); else return (lagg_protos[sc->sc_proto].pr_addport(lp)); } static void lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_delport != NULL) lagg_protos[sc->sc_proto].pr_delport(lp); } static void lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) lagg_protos[sc->sc_proto].pr_linkstate(lp); } static void lagg_proto_init(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_init != NULL) lagg_protos[sc->sc_proto].pr_init(sc); } static void lagg_proto_stop(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_stop != NULL) lagg_protos[sc->sc_proto].pr_stop(sc); } static void lagg_proto_lladdr(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) lagg_protos[sc->sc_proto].pr_lladdr(sc); } static void lagg_proto_request(struct lagg_softc *sc, void *v) { if (lagg_protos[sc->sc_proto].pr_request != NULL) lagg_protos[sc->sc_proto].pr_request(sc, v); } static void lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) { if (lagg_protos[sc->sc_proto].pr_portreq != NULL) lagg_protos[sc->sc_proto].pr_portreq(lp, v); } /* * This routine is run via an vlan * config EVENT */ static void lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) { struct lagg_softc *sc = ifp->if_softc; struct lagg_port *lp; if (ifp->if_softc != arg) /* Not our event */ return; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); LAGG_RUNLOCK(); } /* * This routine is run via an vlan * unconfig EVENT */ static void lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) { struct lagg_softc *sc = ifp->if_softc; struct lagg_port *lp; if (ifp->if_softc != arg) /* Not our event */ return; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); LAGG_RUNLOCK(); } static int lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct lagg_softc *sc; struct ifnet *ifp; static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ sc = malloc(sizeof(*sc), M_LAGG, M_WAITOK|M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { free(sc, M_LAGG); return (ENOSPC); } LAGG_SX_INIT(sc); LAGG_XLOCK(sc); if (V_def_use_flowid) sc->sc_opts |= LAGG_OPT_USE_FLOWID; if (V_def_use_numa) sc->sc_opts |= LAGG_OPT_USE_NUMA; sc->flowid_shift = V_def_flowid_shift; /* Hash all layers by default */ sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); CK_SLIST_INIT(&sc->sc_ports); /* Initialise pseudo media types */ ifmedia_init(&sc->sc_media, 0, lagg_media_change, lagg_media_status); ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); if_initname(ifp, laggname, unit); ifp->if_softc = sc; ifp->if_transmit = lagg_transmit; ifp->if_qflush = lagg_qflush; ifp->if_init = lagg_init; ifp->if_ioctl = lagg_ioctl; ifp->if_get_counter = lagg_get_counter; ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; #if defined(KERN_TLS) || defined(RATELIMIT) ifp->if_snd_tag_alloc = lagg_snd_tag_alloc; ifp->if_snd_tag_modify = lagg_snd_tag_modify; ifp->if_snd_tag_query = lagg_snd_tag_query; ifp->if_snd_tag_free = lagg_snd_tag_free; ifp->if_ratelimit_query = lagg_ratelimit_query; #endif ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; /* * Attach as an ordinary ethernet device, children will be attached * as special device IFT_IEEE8023ADLAG. */ ether_ifattach(ifp, eaddr); sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); /* Insert into the global list of laggs */ LAGG_LIST_LOCK(); SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); LAGG_LIST_UNLOCK(); LAGG_XUNLOCK(sc); return (0); } static void lagg_clone_destroy(struct ifnet *ifp) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_port *lp; LAGG_XLOCK(sc); sc->sc_destroying = 1; lagg_stop(sc); ifp->if_flags &= ~IFF_UP; EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); /* Shutdown and remove lagg ports */ while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL) lagg_port_destroy(lp, 1); /* Unhook the aggregation protocol */ lagg_proto_detach(sc); LAGG_XUNLOCK(sc); ifmedia_removeall(&sc->sc_media); ether_ifdetach(ifp); if_free(ifp); LAGG_LIST_LOCK(); SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); LAGG_LIST_UNLOCK(); LAGG_SX_DESTROY(sc); free(sc, M_LAGG); } static void lagg_capabilities(struct lagg_softc *sc) { struct lagg_port *lp; int cap, ena, pena; uint64_t hwa; struct ifnet_hw_tsomax hw_tsomax; LAGG_XLOCK_ASSERT(sc); /* Get common enabled capabilities for the lagg ports */ ena = ~0; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) ena &= lp->lp_ifp->if_capenable; ena = (ena == ~0 ? 0 : ena); /* * Apply common enabled capabilities back to the lagg ports. * May require several iterations if they are dependent. */ do { pena = ena; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_setcaps(lp, ena); ena &= lp->lp_ifp->if_capenable; } } while (pena != ena); /* Get other capabilities from the lagg ports */ cap = ~0; hwa = ~(uint64_t)0; memset(&hw_tsomax, 0, sizeof(hw_tsomax)); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { cap &= lp->lp_ifp->if_capabilities; hwa &= lp->lp_ifp->if_hwassist; if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); } cap = (cap == ~0 ? 0 : cap); hwa = (hwa == ~(uint64_t)0 ? 0 : hwa); if (sc->sc_ifp->if_capabilities != cap || sc->sc_ifp->if_capenable != ena || sc->sc_ifp->if_hwassist != hwa || if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { sc->sc_ifp->if_capabilities = cap; sc->sc_ifp->if_capenable = ena; sc->sc_ifp->if_hwassist = hwa; getmicrotime(&sc->sc_ifp->if_lastchange); if (sc->sc_ifflags & IFF_DEBUG) if_printf(sc->sc_ifp, "capabilities 0x%08x enabled 0x%08x\n", cap, ena); } } static int lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) { struct lagg_softc *sc_ptr; struct lagg_port *lp, *tlp; struct ifreq ifr; int error, i, oldmtu; uint64_t *pval; LAGG_XLOCK_ASSERT(sc); if (sc->sc_ifp == ifp) { if_printf(sc->sc_ifp, "cannot add a lagg to itself as a port\n"); return (EINVAL); } /* Limit the maximal number of lagg ports */ if (sc->sc_count >= LAGG_MAX_PORTS) return (ENOSPC); /* Check if port has already been associated to a lagg */ if (ifp->if_lagg != NULL) { /* Port is already in the current lagg? */ lp = (struct lagg_port *)ifp->if_lagg; if (lp->lp_softc == sc) return (EEXIST); return (EBUSY); } /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) return (EPROTONOSUPPORT); /* Allow the first Ethernet member to define the MTU */ oldmtu = -1; if (CK_SLIST_EMPTY(&sc->sc_ports)) { sc->sc_ifp->if_mtu = ifp->if_mtu; } else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { if (ifp->if_ioctl == NULL) { if_printf(sc->sc_ifp, "cannot change MTU for %s\n", ifp->if_xname); return (EINVAL); } oldmtu = ifp->if_mtu; strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)); ifr.ifr_mtu = sc->sc_ifp->if_mtu; error = (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); if (error != 0) { if_printf(sc->sc_ifp, "invalid MTU for %s\n", ifp->if_xname); return (error); } ifr.ifr_mtu = oldmtu; } lp = malloc(sizeof(struct lagg_port), M_LAGG, M_WAITOK|M_ZERO); lp->lp_softc = sc; /* Check if port is a stacked lagg */ LAGG_LIST_LOCK(); SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { if (ifp == sc_ptr->sc_ifp) { LAGG_LIST_UNLOCK(); free(lp, M_LAGG); if (oldmtu != -1) (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); return (EINVAL); /* XXX disable stacking for the moment, its untested */ #ifdef LAGG_PORT_STACKING lp->lp_flags |= LAGG_PORT_STACK; if (lagg_port_checkstacking(sc_ptr) >= LAGG_MAX_STACKING) { LAGG_LIST_UNLOCK(); free(lp, M_LAGG); if (oldmtu != -1) (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); return (E2BIG); } #endif } } LAGG_LIST_UNLOCK(); if_ref(ifp); lp->lp_ifp = ifp; bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); lp->lp_ifcapenable = ifp->if_capenable; if (CK_SLIST_EMPTY(&sc->sc_ports)) { bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); lagg_proto_lladdr(sc); EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } else { if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); } lagg_setflags(lp, 1); if (CK_SLIST_EMPTY(&sc->sc_ports)) sc->sc_primary = lp; /* Change the interface type */ lp->lp_iftype = ifp->if_type; ifp->if_type = IFT_IEEE8023ADLAG; ifp->if_lagg = lp; lp->lp_ioctl = ifp->if_ioctl; ifp->if_ioctl = lagg_port_ioctl; lp->lp_output = ifp->if_output; ifp->if_output = lagg_port_output; /* Read port counters */ pval = lp->port_counters.val; for (i = 0; i < IFCOUNTERS; i++, pval++) *pval = ifp->if_get_counter(ifp, i); /* * Insert into the list of ports. * Keep ports sorted by if_index. It is handy, when configuration * is predictable and `ifconfig laggN create ...` command * will lead to the same result each time. */ CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { if (tlp->lp_ifp->if_index < ifp->if_index && ( CK_SLIST_NEXT(tlp, lp_entries) == NULL || ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index > ifp->if_index)) break; } if (tlp != NULL) CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries); else CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); sc->sc_count++; lagg_setmulti(lp); if ((error = lagg_proto_addport(sc, lp)) != 0) { /* Remove the port, without calling pr_delport. */ lagg_port_destroy(lp, 0); if (oldmtu != -1) (*ifp->if_ioctl)(ifp, SIOCSIFMTU, (caddr_t)&ifr); return (error); } /* Update lagg capabilities */ lagg_capabilities(sc); lagg_linkstate(sc); return (0); } #ifdef LAGG_PORT_STACKING static int lagg_port_checkstacking(struct lagg_softc *sc) { struct lagg_softc *sc_ptr; struct lagg_port *lp; int m = 0; LAGG_SXLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_flags & LAGG_PORT_STACK) { sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; m = MAX(m, lagg_port_checkstacking(sc_ptr)); } } return (m + 1); } #endif static void lagg_port_destroy_cb(epoch_context_t ec) { struct lagg_port *lp; struct ifnet *ifp; lp = __containerof(ec, struct lagg_port, lp_epoch_ctx); ifp = lp->lp_ifp; if_rele(ifp); free(lp, M_LAGG); } static int lagg_port_destroy(struct lagg_port *lp, int rundelport) { struct lagg_softc *sc = lp->lp_softc; struct lagg_port *lp_ptr, *lp0; struct ifnet *ifp = lp->lp_ifp; uint64_t *pval, vdiff; int i; LAGG_XLOCK_ASSERT(sc); if (rundelport) lagg_proto_delport(sc, lp); if (lp->lp_detaching == 0) lagg_clrmulti(lp); /* Restore interface */ ifp->if_type = lp->lp_iftype; ifp->if_ioctl = lp->lp_ioctl; ifp->if_output = lp->lp_output; ifp->if_lagg = NULL; /* Update detached port counters */ pval = lp->port_counters.val; for (i = 0; i < IFCOUNTERS; i++, pval++) { vdiff = ifp->if_get_counter(ifp, i) - *pval; sc->detached_counters.val[i] += vdiff; } /* Finally, remove the port from the lagg */ CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); sc->sc_count--; /* Update the primary interface */ if (lp == sc->sc_primary) { uint8_t lladdr[ETHER_ADDR_LEN]; if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL) bzero(&lladdr, ETHER_ADDR_LEN); else bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN); sc->sc_primary = lp0; if (sc->sc_destroying == 0) { bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); lagg_proto_lladdr(sc); EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } /* * Update lladdr for each port (new primary needs update * as well, to switch from old lladdr to its 'real' one) */ CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN); } if (lp->lp_ifflags) if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); if (lp->lp_detaching == 0) { lagg_setflags(lp, 0); lagg_setcaps(lp, lp->lp_ifcapenable); if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN); } /* * free port and release it's ifnet reference after a grace period has * elapsed. */ NET_EPOCH_CALL(lagg_port_destroy_cb, &lp->lp_epoch_ctx); /* Update lagg capabilities */ lagg_capabilities(sc); lagg_linkstate(sc); return (0); } static int lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct lagg_reqport *rp = (struct lagg_reqport *)data; struct lagg_softc *sc; struct lagg_port *lp = NULL; int error = 0; /* Should be checked by the caller */ if (ifp->if_type != IFT_IEEE8023ADLAG || (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) goto fallback; switch (cmd) { case SIOCGLAGGPORT: if (rp->rp_portname[0] == '\0' || ifunit(rp->rp_portname) != ifp) { error = EINVAL; break; } LAGG_RLOCK(); if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_RUNLOCK(); break; } lagg_port2req(lp, rp); LAGG_RUNLOCK(); break; case SIOCSIFCAP: if (lp->lp_ioctl == NULL) { error = EINVAL; break; } error = (*lp->lp_ioctl)(ifp, cmd, data); if (error) break; /* Update lagg interface capabilities */ LAGG_XLOCK(sc); lagg_capabilities(sc); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(sc->sc_ifp); break; case SIOCSIFMTU: /* Do not allow the MTU to be changed once joined */ error = EINVAL; break; default: goto fallback; } return (error); fallback: if (lp != NULL && lp->lp_ioctl != NULL) return ((*lp->lp_ioctl)(ifp, cmd, data)); return (EINVAL); } /* * Requests counter @cnt data. * * Counter value is calculated the following way: * 1) for each port, sum difference between current and "initial" measurements. * 2) add lagg logical interface counters. * 3) add data from detached_counters array. * * We also do the following things on ports attach/detach: * 1) On port attach we store all counters it has into port_counter array. * 2) On port detach we add the different between "initial" and * current counters data to detached_counters array. */ static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt) { struct lagg_softc *sc; struct lagg_port *lp; struct ifnet *lpifp; uint64_t newval, oldval, vsum; /* Revise this when we've got non-generic counters. */ KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); sc = (struct lagg_softc *)ifp->if_softc; vsum = 0; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { /* Saved attached value */ oldval = lp->port_counters.val[cnt]; /* current value */ lpifp = lp->lp_ifp; newval = lpifp->if_get_counter(lpifp, cnt); /* Calculate diff and save new */ vsum += newval - oldval; } LAGG_RUNLOCK(); /* * Add counter data which might be added by upper * layer protocols operating on logical interface. */ vsum += if_get_counter_default(ifp, cnt); /* * Add counter data from detached ports counters */ vsum += sc->detached_counters.val[cnt]; return (vsum); } /* * For direct output to child ports. */ static int lagg_port_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { struct lagg_port *lp = ifp->if_lagg; switch (dst->sa_family) { case pseudo_AF_HDRCMPLT: case AF_UNSPEC: return ((*lp->lp_output)(ifp, m, dst, ro)); } /* drop any other frames */ m_freem(m); return (ENETDOWN); } static void lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) { struct lagg_port *lp; struct lagg_softc *sc; if ((lp = ifp->if_lagg) == NULL) return; /* If the ifnet is just being renamed, don't do anything. */ if (ifp->if_flags & IFF_RENAMING) return; sc = lp->lp_softc; LAGG_XLOCK(sc); lp->lp_detaching = 1; lagg_port_destroy(lp, 1); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(sc->sc_ifp); } static void lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) { struct lagg_softc *sc = lp->lp_softc; strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); rp->rp_prio = lp->lp_prio; rp->rp_flags = lp->lp_flags; lagg_proto_portreq(sc, lp, &rp->rp_psc); /* Add protocol specific flags */ switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: if (lp == sc->sc_primary) rp->rp_flags |= LAGG_PORT_MASTER; if (lp == lagg_link_active(sc, sc->sc_primary)) rp->rp_flags |= LAGG_PORT_ACTIVE; break; case LAGG_PROTO_ROUNDROBIN: case LAGG_PROTO_LOADBALANCE: case LAGG_PROTO_BROADCAST: if (LAGG_PORTACTIVE(lp)) rp->rp_flags |= LAGG_PORT_ACTIVE; break; case LAGG_PROTO_LACP: /* LACP has a different definition of active */ if (lacp_isactive(lp)) rp->rp_flags |= LAGG_PORT_ACTIVE; if (lacp_iscollecting(lp)) rp->rp_flags |= LAGG_PORT_COLLECTING; if (lacp_isdistributing(lp)) rp->rp_flags |= LAGG_PORT_DISTRIBUTING; break; } } static void lagg_init(void *xsc) { struct lagg_softc *sc = (struct lagg_softc *)xsc; struct ifnet *ifp = sc->sc_ifp; struct lagg_port *lp; LAGG_XLOCK(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { LAGG_XUNLOCK(sc); return; } ifp->if_drv_flags |= IFF_DRV_RUNNING; /* * Update the port lladdrs if needed. * This might be if_setlladdr() notification * that lladdr has been changed. */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp), ETHER_ADDR_LEN) != 0) if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN); } lagg_proto_init(sc); LAGG_XUNLOCK(sc); } static void lagg_stop(struct lagg_softc *sc) { struct ifnet *ifp = sc->sc_ifp; LAGG_XLOCK_ASSERT(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; lagg_proto_stop(sc); } static int lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_reqall *ra = (struct lagg_reqall *)data; struct lagg_reqopts *ro = (struct lagg_reqopts *)data; struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; struct lagg_reqflags *rf = (struct lagg_reqflags *)data; struct ifreq *ifr = (struct ifreq *)data; struct lagg_port *lp; struct ifnet *tpif; struct thread *td = curthread; char *buf, *outbuf; int count, buflen, len, error = 0, oldmtu; bzero(&rpbuf, sizeof(rpbuf)); switch (cmd) { case SIOCGLAGG: LAGG_XLOCK(sc); buflen = sc->sc_count * sizeof(struct lagg_reqport); outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); ra->ra_proto = sc->sc_proto; lagg_proto_request(sc, &ra->ra_psc); count = 0; buf = outbuf; len = min(ra->ra_size, buflen); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (len < sizeof(rpbuf)) break; lagg_port2req(lp, &rpbuf); memcpy(buf, &rpbuf, sizeof(rpbuf)); count++; buf += sizeof(rpbuf); len -= sizeof(rpbuf); } LAGG_XUNLOCK(sc); ra->ra_ports = count; ra->ra_size = count * sizeof(rpbuf); error = copyout(outbuf, ra->ra_port, ra->ra_size); free(outbuf, M_TEMP); break; case SIOCSLAGG: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (ra->ra_proto >= LAGG_PROTO_MAX) { error = EPROTONOSUPPORT; break; } LAGG_XLOCK(sc); lagg_proto_detach(sc); LAGG_UNLOCK_ASSERT(); lagg_proto_attach(sc, ra->ra_proto); LAGG_XUNLOCK(sc); break; case SIOCGLAGGOPTS: LAGG_XLOCK(sc); ro->ro_opts = sc->sc_opts; if (sc->sc_proto == LAGG_PROTO_LACP) { struct lacp_softc *lsc; lsc = (struct lacp_softc *)sc->sc_psc; if (lsc->lsc_debug.lsc_tx_test != 0) ro->ro_opts |= LAGG_OPT_LACP_TXTEST; if (lsc->lsc_debug.lsc_rx_test != 0) ro->ro_opts |= LAGG_OPT_LACP_RXTEST; if (lsc->lsc_strict_mode != 0) ro->ro_opts |= LAGG_OPT_LACP_STRICT; if (lsc->lsc_fast_timeout != 0) ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; ro->ro_active = sc->sc_active; } else { ro->ro_active = 0; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) ro->ro_active += LAGG_PORTACTIVE(lp); } ro->ro_bkt = sc->sc_stride; ro->ro_flapping = sc->sc_flapping; ro->ro_flowid_shift = sc->flowid_shift; LAGG_XUNLOCK(sc); break; case SIOCSLAGGOPTS: error = priv_check(td, PRIV_NET_LAGG); if (error) break; /* * The stride option was added without defining a corresponding * LAGG_OPT flag, so handle a non-zero value before checking * anything else to preserve compatibility. */ LAGG_XLOCK(sc); if (ro->ro_opts == 0 && ro->ro_bkt != 0) { if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN) { LAGG_XUNLOCK(sc); error = EINVAL; break; } sc->sc_stride = ro->ro_bkt; } if (ro->ro_opts == 0) { LAGG_XUNLOCK(sc); break; } /* * Set options. LACP options are stored in sc->sc_psc, * not in sc_opts. */ int valid, lacp; switch (ro->ro_opts) { case LAGG_OPT_USE_FLOWID: case -LAGG_OPT_USE_FLOWID: case LAGG_OPT_USE_NUMA: case -LAGG_OPT_USE_NUMA: case LAGG_OPT_FLOWIDSHIFT: case LAGG_OPT_RR_LIMIT: valid = 1; lacp = 0; break; case LAGG_OPT_LACP_TXTEST: case -LAGG_OPT_LACP_TXTEST: case LAGG_OPT_LACP_RXTEST: case -LAGG_OPT_LACP_RXTEST: case LAGG_OPT_LACP_STRICT: case -LAGG_OPT_LACP_STRICT: case LAGG_OPT_LACP_TIMEOUT: case -LAGG_OPT_LACP_TIMEOUT: valid = lacp = 1; break; default: valid = lacp = 0; break; } if (valid == 0 || (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { /* Invalid combination of options specified. */ error = EINVAL; LAGG_XUNLOCK(sc); break; /* Return from SIOCSLAGGOPTS. */ } /* * Store new options into sc->sc_opts except for * FLOWIDSHIFT, RR and LACP options. */ if (lacp == 0) { if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) sc->flowid_shift = ro->ro_flowid_shift; else if (ro->ro_opts == LAGG_OPT_RR_LIMIT) { if (sc->sc_proto != LAGG_PROTO_ROUNDROBIN || ro->ro_bkt == 0) { error = EINVAL; LAGG_XUNLOCK(sc); break; } sc->sc_stride = ro->ro_bkt; } else if (ro->ro_opts > 0) sc->sc_opts |= ro->ro_opts; else sc->sc_opts &= ~ro->ro_opts; } else { struct lacp_softc *lsc; struct lacp_port *lp; lsc = (struct lacp_softc *)sc->sc_psc; switch (ro->ro_opts) { case LAGG_OPT_LACP_TXTEST: lsc->lsc_debug.lsc_tx_test = 1; break; case -LAGG_OPT_LACP_TXTEST: lsc->lsc_debug.lsc_tx_test = 0; break; case LAGG_OPT_LACP_RXTEST: lsc->lsc_debug.lsc_rx_test = 1; break; case -LAGG_OPT_LACP_RXTEST: lsc->lsc_debug.lsc_rx_test = 0; break; case LAGG_OPT_LACP_STRICT: lsc->lsc_strict_mode = 1; break; case -LAGG_OPT_LACP_STRICT: lsc->lsc_strict_mode = 0; break; case LAGG_OPT_LACP_TIMEOUT: LACP_LOCK(lsc); LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) lp->lp_state |= LACP_STATE_TIMEOUT; LACP_UNLOCK(lsc); lsc->lsc_fast_timeout = 1; break; case -LAGG_OPT_LACP_TIMEOUT: LACP_LOCK(lsc); LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) lp->lp_state &= ~LACP_STATE_TIMEOUT; LACP_UNLOCK(lsc); lsc->lsc_fast_timeout = 0; break; } } LAGG_XUNLOCK(sc); break; case SIOCGLAGGFLAGS: rf->rf_flags = 0; LAGG_XLOCK(sc); if (sc->sc_flags & MBUF_HASHFLAG_L2) rf->rf_flags |= LAGG_F_HASHL2; if (sc->sc_flags & MBUF_HASHFLAG_L3) rf->rf_flags |= LAGG_F_HASHL3; if (sc->sc_flags & MBUF_HASHFLAG_L4) rf->rf_flags |= LAGG_F_HASHL4; LAGG_XUNLOCK(sc); break; case SIOCSLAGGHASH: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { error = EINVAL; break; } LAGG_XLOCK(sc); sc->sc_flags = 0; if (rf->rf_flags & LAGG_F_HASHL2) sc->sc_flags |= MBUF_HASHFLAG_L2; if (rf->rf_flags & LAGG_F_HASHL3) sc->sc_flags |= MBUF_HASHFLAG_L3; if (rf->rf_flags & LAGG_F_HASHL4) sc->sc_flags |= MBUF_HASHFLAG_L4; LAGG_XUNLOCK(sc); break; case SIOCGLAGGPORT: if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } LAGG_RLOCK(); if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_RUNLOCK(); if_rele(tpif); break; } lagg_port2req(lp, rp); LAGG_RUNLOCK(); if_rele(tpif); break; case SIOCSLAGGPORT: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } #ifdef INET6 /* * A laggport interface should not have inet6 address * because two interfaces with a valid link-local * scope zone must not be merged in any form. This * restriction is needed to prevent violation of * link-local scope zone. Attempts to add a laggport * interface which has inet6 addresses triggers * removal of all inet6 addresses on the member * interface. */ if (in6ifa_llaonifp(tpif)) { in6_ifdetach(tpif); if_printf(sc->sc_ifp, "IPv6 addresses on %s have been removed " "before adding it as a member to prevent " "IPv6 address scope violation.\n", tpif->if_xname); } #endif oldmtu = ifp->if_mtu; LAGG_XLOCK(sc); error = lagg_port_create(sc, tpif); LAGG_XUNLOCK(sc); if_rele(tpif); /* * LAGG MTU may change during addition of the first port. * If it did, do network layer specific procedure. */ if (ifp->if_mtu != oldmtu) { #ifdef INET6 nd6_setmtu(ifp); #endif rt_updatemtu(ifp); } VLAN_CAPABILITIES(ifp); break; case SIOCSLAGGDELPORT: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } LAGG_XLOCK(sc); if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_XUNLOCK(sc); if_rele(tpif); break; } error = lagg_port_destroy(lp, 1); LAGG_XUNLOCK(sc); if_rele(tpif); VLAN_CAPABILITIES(ifp); break; case SIOCSIFFLAGS: /* Set flags on ports too */ LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_setflags(lp, 1); } if (!(ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked down and it is running, * then stop and disable it. */ lagg_stop(sc); LAGG_XUNLOCK(sc); } else if ((ifp->if_flags & IFF_UP) && !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked up and it is stopped, then * start it. */ LAGG_XUNLOCK(sc); (*ifp->if_init)(sc); } else LAGG_XUNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_clrmulti(lp); lagg_setmulti(lp); } LAGG_XUNLOCK(sc); error = 0; break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); break; case SIOCSIFCAP: LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ioctl != NULL) (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); } lagg_capabilities(sc); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(ifp); error = 0; break; case SIOCSIFMTU: LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ioctl != NULL) error = (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); else error = EINVAL; if (error != 0) { if_printf(ifp, "failed to change MTU to %d on port %s, " "reverting all ports to original MTU (%d)\n", ifr->ifr_mtu, lp->lp_ifp->if_xname, ifp->if_mtu); break; } } if (error == 0) { ifp->if_mtu = ifr->ifr_mtu; } else { /* set every port back to the original MTU */ ifr->ifr_mtu = ifp->if_mtu; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ioctl != NULL) (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); } } LAGG_XUNLOCK(sc); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } #if defined(KERN_TLS) || defined(RATELIMIT) static inline struct lagg_snd_tag * mst_to_lst(struct m_snd_tag *mst) { return (__containerof(mst, struct lagg_snd_tag, com)); } /* * Look up the port used by a specific flow. This only works for lagg * protocols with deterministic port mappings (e.g. not roundrobin). * In addition protocols which use a hash to map flows to ports must * be configured to use the mbuf flowid rather than hashing packet * contents. */ static struct lagg_port * -lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype) +lookup_snd_tag_port(struct ifnet *ifp, uint32_t flowid, uint32_t flowtype, + uint8_t numa_domain) { struct lagg_softc *sc; struct lagg_port *lp; struct lagg_lb *lb; - uint32_t p; + uint32_t hash, p; sc = ifp->if_softc; switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: return (lagg_link_active(sc, sc->sc_primary)); case LAGG_PROTO_LOADBALANCE: if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || flowtype == M_HASHTYPE_NONE) return (NULL); p = flowid >> sc->flowid_shift; p %= sc->sc_count; lb = (struct lagg_lb *)sc->sc_psc; lp = lb->lb_ports[p]; return (lagg_link_active(sc, lp)); case LAGG_PROTO_LACP: if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || flowtype == M_HASHTYPE_NONE) return (NULL); - return (lacp_select_tx_port_by_hash(sc, flowid)); + hash = flowid >> sc->flowid_shift; + return (lacp_select_tx_port_by_hash(sc, hash, numa_domain)); default: return (NULL); } } static int lagg_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { struct lagg_snd_tag *lst; struct lagg_softc *sc; struct lagg_port *lp; struct ifnet *lp_ifp; int error; sc = ifp->if_softc; LAGG_RLOCK(); - lp = lookup_snd_tag_port(ifp, params->hdr.flowid, params->hdr.flowtype); + lp = lookup_snd_tag_port(ifp, params->hdr.flowid, + params->hdr.flowtype, params->hdr.numa_domain); if (lp == NULL) { LAGG_RUNLOCK(); return (EOPNOTSUPP); } if (lp->lp_ifp == NULL || lp->lp_ifp->if_snd_tag_alloc == NULL) { LAGG_RUNLOCK(); return (EOPNOTSUPP); } lp_ifp = lp->lp_ifp; if_ref(lp_ifp); LAGG_RUNLOCK(); lst = malloc(sizeof(*lst), M_LAGG, M_NOWAIT); if (lst == NULL) { if_rele(lp_ifp); return (ENOMEM); } error = lp_ifp->if_snd_tag_alloc(lp_ifp, params, &lst->tag); if_rele(lp_ifp); if (error) { free(lst, M_LAGG); return (error); } m_snd_tag_init(&lst->com, ifp); *ppmt = &lst->com; return (0); } static int lagg_snd_tag_modify(struct m_snd_tag *mst, union if_snd_tag_modify_params *params) { struct lagg_snd_tag *lst; lst = mst_to_lst(mst); return (lst->tag->ifp->if_snd_tag_modify(lst->tag, params)); } static int lagg_snd_tag_query(struct m_snd_tag *mst, union if_snd_tag_query_params *params) { struct lagg_snd_tag *lst; lst = mst_to_lst(mst); return (lst->tag->ifp->if_snd_tag_query(lst->tag, params)); } static void lagg_snd_tag_free(struct m_snd_tag *mst) { struct lagg_snd_tag *lst; lst = mst_to_lst(mst); m_snd_tag_rele(lst->tag); free(lst, M_LAGG); } static void lagg_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q) { /* * For lagg, we have an indirect * interface. The caller needs to * get a ratelimit tag on the actual * interface the flow will go on. */ q->rate_table = NULL; q->flags = RT_IS_INDIRECT; q->max_flows = 0; q->number_of_rates = 0; } #endif static int lagg_setmulti(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; struct ifnet *ifp = lp->lp_ifp; struct ifnet *scifp = sc->sc_ifp; struct lagg_mc *mc; struct ifmultiaddr *ifma; int error; IF_ADDR_WLOCK(scifp); CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; mc = malloc(sizeof(struct lagg_mc), M_LAGG, M_NOWAIT); if (mc == NULL) { IF_ADDR_WUNLOCK(scifp); return (ENOMEM); } bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len); mc->mc_addr.sdl_index = ifp->if_index; mc->mc_ifma = NULL; SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); } IF_ADDR_WUNLOCK(scifp); SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { error = if_addmulti(ifp, (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); if (error) return (error); } return (0); } static int lagg_clrmulti(struct lagg_port *lp) { struct lagg_mc *mc; LAGG_XLOCK_ASSERT(lp->lp_softc); while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); if (mc->mc_ifma && lp->lp_detaching == 0) if_delmulti_ifma(mc->mc_ifma); free(mc, M_LAGG); } return (0); } static int lagg_setcaps(struct lagg_port *lp, int cap) { struct ifreq ifr; if (lp->lp_ifp->if_capenable == cap) return (0); if (lp->lp_ioctl == NULL) return (ENXIO); ifr.ifr_reqcap = cap; return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); } /* Handle a ref counted flag that should be set on the lagg port as well */ static int lagg_setflag(struct lagg_port *lp, int flag, int status, int (*func)(struct ifnet *, int)) { struct lagg_softc *sc = lp->lp_softc; struct ifnet *scifp = sc->sc_ifp; struct ifnet *ifp = lp->lp_ifp; int error; LAGG_XLOCK_ASSERT(sc); status = status ? (scifp->if_flags & flag) : 0; /* Now "status" contains the flag value or 0 */ /* * See if recorded ports status is different from what * we want it to be. If it is, flip it. We record ports * status in lp_ifflags so that we won't clear ports flag * we haven't set. In fact, we don't clear or set ports * flags directly, but get or release references to them. * That's why we can be sure that recorded flags still are * in accord with actual ports flags. */ if (status != (lp->lp_ifflags & flag)) { error = (*func)(ifp, status); if (error) return (error); lp->lp_ifflags &= ~flag; lp->lp_ifflags |= status; } return (0); } /* * Handle IFF_* flags that require certain changes on the lagg port * if "status" is true, update ports flags respective to the lagg * if "status" is false, forcedly clear the flags set on port. */ static int lagg_setflags(struct lagg_port *lp, int status) { int error, i; for (i = 0; lagg_pflags[i].flag; i++) { error = lagg_setflag(lp, lagg_pflags[i].flag, status, lagg_pflags[i].func); if (error) return (error); } return (0); } static int lagg_transmit(struct ifnet *ifp, struct mbuf *m) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; int error; #if defined(KERN_TLS) || defined(RATELIMIT) if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) MPASS(m->m_pkthdr.snd_tag->ifp == ifp); #endif LAGG_RLOCK(); /* We need a Tx algorithm and at least one port */ if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { LAGG_RUNLOCK(); m_freem(m); if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (ENXIO); } ETHER_BPF_MTAP(ifp, m); error = lagg_proto_start(sc, m); LAGG_RUNLOCK(); if (error != 0) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (error); } /* * The ifp->if_qflush entry point for lagg(4) is no-op. */ static void lagg_qflush(struct ifnet *ifp __unused) { } static struct mbuf * lagg_input(struct ifnet *ifp, struct mbuf *m) { struct lagg_port *lp = ifp->if_lagg; struct lagg_softc *sc = lp->lp_softc; struct ifnet *scifp = sc->sc_ifp; LAGG_RLOCK(); if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || lp->lp_detaching != 0 || sc->sc_proto == LAGG_PROTO_NONE) { LAGG_RUNLOCK(); m_freem(m); return (NULL); } ETHER_BPF_MTAP(scifp, m); m = lagg_proto_input(sc, lp, m); if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { m_freem(m); m = NULL; } LAGG_RUNLOCK(); return (m); } static int lagg_media_change(struct ifnet *ifp) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; if (sc->sc_ifflags & IFF_DEBUG) printf("%s\n", __func__); /* Ignore */ return (0); } static void lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_port *lp; imr->ifm_status = IFM_AVALID; imr->ifm_active = IFM_ETHER | IFM_AUTO; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (LAGG_PORTACTIVE(lp)) imr->ifm_status |= IFM_ACTIVE; } LAGG_RUNLOCK(); } static void lagg_linkstate(struct lagg_softc *sc) { struct lagg_port *lp; int new_link = LINK_STATE_DOWN; uint64_t speed; LAGG_XLOCK_ASSERT(sc); /* LACP handles link state itself */ if (sc->sc_proto == LAGG_PROTO_LACP) return; /* Our link is considered up if at least one of our ports is active */ LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { new_link = LINK_STATE_UP; break; } } LAGG_RUNLOCK(); if_link_state_change(sc->sc_ifp, new_link); /* Update if_baudrate to reflect the max possible speed */ switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? sc->sc_primary->lp_ifp->if_baudrate : 0; break; case LAGG_PROTO_ROUNDROBIN: case LAGG_PROTO_LOADBALANCE: case LAGG_PROTO_BROADCAST: speed = 0; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) speed += lp->lp_ifp->if_baudrate; LAGG_RUNLOCK(); sc->sc_ifp->if_baudrate = speed; break; case LAGG_PROTO_LACP: /* LACP updates if_baudrate itself */ break; } } static void lagg_port_state(struct ifnet *ifp, int state) { struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; struct lagg_softc *sc = NULL; if (lp != NULL) sc = lp->lp_softc; if (sc == NULL) return; LAGG_XLOCK(sc); lagg_linkstate(sc); lagg_proto_linkstate(sc, lp); LAGG_XUNLOCK(sc); } struct lagg_port * lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) { struct lagg_port *lp_next, *rval = NULL; /* * Search a port which reports an active link state. */ #ifdef INVARIANTS /* * This is called with either LAGG_RLOCK() held or * LAGG_XLOCK(sc) held. */ if (!in_epoch(net_epoch_preempt)) LAGG_XLOCK_ASSERT(sc); #endif if (lp == NULL) goto search; if (LAGG_PORTACTIVE(lp)) { rval = lp; goto found; } if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL && LAGG_PORTACTIVE(lp_next)) { rval = lp_next; goto found; } search: CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { if (LAGG_PORTACTIVE(lp_next)) { return (lp_next); } } found: return (rval); } int lagg_enqueue(struct ifnet *ifp, struct mbuf *m) { #if defined(KERN_TLS) || defined(RATELIMIT) if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { struct lagg_snd_tag *lst; struct m_snd_tag *mst; mst = m->m_pkthdr.snd_tag; lst = mst_to_lst(mst); if (lst->tag->ifp != ifp) { m_freem(m); return (EAGAIN); } m->m_pkthdr.snd_tag = m_snd_tag_ref(lst->tag); m_snd_tag_rele(mst); } #endif return (ifp->if_transmit)(ifp, m); } /* * Simple round robin aggregation */ static void lagg_rr_attach(struct lagg_softc *sc) { sc->sc_seq = 0; sc->sc_stride = 1; } static int lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; uint32_t p; p = atomic_fetchadd_32(&sc->sc_seq, 1); p /= sc->sc_stride; p %= sc->sc_count; lp = CK_SLIST_FIRST(&sc->sc_ports); while (p--) lp = CK_SLIST_NEXT(lp, lp_entries); /* * Check the port's link state. This will return the next active * port if the link is down or the port is NULL. */ if ((lp = lagg_link_active(sc, lp)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * Broadcast mode */ static int lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) { int active_ports = 0; int errors = 0; int ret; struct lagg_port *lp, *last = NULL; struct mbuf *m0; LAGG_RLOCK_ASSERT(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (!LAGG_PORTACTIVE(lp)) continue; active_ports++; if (last != NULL) { m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (m0 == NULL) { ret = ENOBUFS; errors++; break; } ret = lagg_enqueue(last->lp_ifp, m0); if (ret != 0) errors++; } last = lp; } if (last == NULL) { m_freem(m); return (ENOENT); } if ((last = lagg_link_active(sc, last)) == NULL) { m_freem(m); return (ENETDOWN); } ret = lagg_enqueue(last->lp_ifp, m); if (ret != 0) errors++; if (errors == 0) return (ret); return (0); } static struct mbuf* lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * Active failover */ static int lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; /* Use the master port if active or the next available port */ if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; struct lagg_port *tmp_tp; if (lp == sc->sc_primary || V_lagg_failover_rx_all) { m->m_pkthdr.rcvif = ifp; return (m); } if (!LAGG_PORTACTIVE(sc->sc_primary)) { tmp_tp = lagg_link_active(sc, sc->sc_primary); /* * If tmp_tp is null, we've received a packet when all * our links are down. Weird, but process it anyways. */ if ((tmp_tp == NULL || tmp_tp == lp)) { m->m_pkthdr.rcvif = ifp; return (m); } } m_freem(m); return (NULL); } /* * Loadbalancing */ static void lagg_lb_attach(struct lagg_softc *sc) { struct lagg_port *lp; struct lagg_lb *lb; LAGG_XLOCK_ASSERT(sc); lb = malloc(sizeof(struct lagg_lb), M_LAGG, M_WAITOK | M_ZERO); lb->lb_key = m_ether_tcpip_hash_init(); sc->sc_psc = lb; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lagg_lb_port_create(lp); } static void lagg_lb_detach(struct lagg_softc *sc) { struct lagg_lb *lb; lb = (struct lagg_lb *)sc->sc_psc; if (lb != NULL) free(lb, M_LAGG); } static int lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) { struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; struct lagg_port *lp_next; int i = 0, rv; rv = 0; bzero(&lb->lb_ports, sizeof(lb->lb_ports)); LAGG_XLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { if (lp_next == lp) continue; if (i >= LAGG_MAX_PORTS) { rv = EINVAL; break; } if (sc->sc_ifflags & IFF_DEBUG) printf("%s: port %s at index %d\n", sc->sc_ifname, lp_next->lp_ifp->if_xname, i); lb->lb_ports[i++] = lp_next; } return (rv); } static int lagg_lb_port_create(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; return (lagg_lb_porttable(sc, NULL)); } static void lagg_lb_port_destroy(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; lagg_lb_porttable(sc, lp); } static int lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; struct lagg_port *lp = NULL; uint32_t p = 0; if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) p = m->m_pkthdr.flowid >> sc->flowid_shift; else p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); p %= sc->sc_count; lp = lb->lb_ports[p]; /* * Check the port's link state. This will return the next active * port if the link is down or the port is NULL. */ if ((lp = lagg_link_active(sc, lp)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * 802.3ad LACP */ static void lagg_lacp_attach(struct lagg_softc *sc) { struct lagg_port *lp; lacp_attach(sc); LAGG_XLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_create(lp); } static void lagg_lacp_detach(struct lagg_softc *sc) { struct lagg_port *lp; void *psc; LAGG_XLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_destroy(lp); psc = sc->sc_psc; sc->sc_psc = NULL; lacp_detach(psc); } static void lagg_lacp_lladdr(struct lagg_softc *sc) { struct lagg_port *lp; LAGG_SXLOCK_ASSERT(sc); /* purge all the lacp ports */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_destroy(lp); /* add them back in */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_create(lp); } static int lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; lp = lacp_select_tx_port(sc, m); if (lp == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; struct ether_header *eh; u_short etype; eh = mtod(m, struct ether_header *); etype = ntohs(eh->ether_type); /* Tap off LACP control messages */ if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { m = lacp_input(lp, m); if (m == NULL) return (NULL); } /* * If the port is not collecting or not in the active aggregator then * free and return. */ if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { m_freem(m); return (NULL); } m->m_pkthdr.rcvif = ifp; return (m); } Index: head/sys/net/if_var.h =================================================================== --- head/sys/net/if_var.h (revision 358807) +++ head/sys/net/if_var.h (revision 358808) @@ -1,804 +1,805 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 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. * * From: @(#)if.h 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #ifndef _NET_IF_VAR_H_ #define _NET_IF_VAR_H_ /* * Structures defining a network interface, providing a packet * transport mechanism (ala level 0 of the PUP protocols). * * Each interface accepts output datagrams of a specified maximum * length, and provides higher level routines with input datagrams * received from its medium. * * Output occurs when the routine if_output is called, with three parameters: * (*ifp->if_output)(ifp, m, dst, rt) * Here m is the mbuf chain to be sent and dst is the destination address. * The output routine encapsulates the supplied datagram if necessary, * and then transmits it on its medium. * * On input, each interface unwraps the data received by it, and either * places it on the input queue of an internetwork datagram routine * and posts the associated software interrupt, or passes the datagram to a raw * packet input routine. * * Routines exist for locating interfaces by their addresses * or for locating an interface on a certain network, as well as more general * routing and gateway routines maintaining information used to locate * interfaces. These routines live in the files if.c and route.c */ struct rtentry; /* ifa_rtrequest */ struct rt_addrinfo; /* ifa_rtrequest */ struct socket; struct carp_if; struct carp_softc; struct ifvlantrunk; struct route; /* if_output */ struct vnet; struct ifmedia; struct netmap_adapter; struct debugnet_methods; #ifdef _KERNEL #include #include /* ifqueue only? */ #include #include #endif /* _KERNEL */ #include #include #include #include /* XXX */ #include /* struct ifqueue */ #include /* XXX */ #include /* XXX */ #include /* if_link_task */ #define IF_DUNIT_NONE -1 #include CK_STAILQ_HEAD(ifnethead, ifnet); /* we use TAILQs so that the order of */ CK_STAILQ_HEAD(ifaddrhead, ifaddr); /* instantiation is preserved in the list */ CK_STAILQ_HEAD(ifmultihead, ifmultiaddr); CK_STAILQ_HEAD(ifgrouphead, ifg_group); #ifdef _KERNEL VNET_DECLARE(struct pfil_head *, link_pfil_head); #define V_link_pfil_head VNET(link_pfil_head) #define PFIL_ETHER_NAME "ethernet" #define HHOOK_IPSEC_INET 0 #define HHOOK_IPSEC_INET6 1 #define HHOOK_IPSEC_COUNT 2 VNET_DECLARE(struct hhook_head *, ipsec_hhh_in[HHOOK_IPSEC_COUNT]); VNET_DECLARE(struct hhook_head *, ipsec_hhh_out[HHOOK_IPSEC_COUNT]); #define V_ipsec_hhh_in VNET(ipsec_hhh_in) #define V_ipsec_hhh_out VNET(ipsec_hhh_out) #endif /* _KERNEL */ typedef enum { IFCOUNTER_IPACKETS = 0, IFCOUNTER_IERRORS, IFCOUNTER_OPACKETS, IFCOUNTER_OERRORS, IFCOUNTER_COLLISIONS, IFCOUNTER_IBYTES, IFCOUNTER_OBYTES, IFCOUNTER_IMCASTS, IFCOUNTER_OMCASTS, IFCOUNTER_IQDROPS, IFCOUNTER_OQDROPS, IFCOUNTER_NOPROTO, IFCOUNTERS /* Array size. */ } ift_counter; typedef struct ifnet * if_t; typedef void (*if_start_fn_t)(if_t); typedef int (*if_ioctl_fn_t)(if_t, u_long, caddr_t); typedef void (*if_init_fn_t)(void *); typedef void (*if_qflush_fn_t)(if_t); typedef int (*if_transmit_fn_t)(if_t, struct mbuf *); typedef uint64_t (*if_get_counter_t)(if_t, ift_counter); struct ifnet_hw_tsomax { u_int tsomaxbytes; /* TSO total burst length limit in bytes */ u_int tsomaxsegcount; /* TSO maximum segment count */ u_int tsomaxsegsize; /* TSO maximum segment size in bytes */ }; /* Interface encap request types */ typedef enum { IFENCAP_LL = 1 /* pre-calculate link-layer header */ } ife_type; /* * The structure below allows to request various pre-calculated L2/L3 headers * for different media. Requests varies by type (rtype field). * * IFENCAP_LL type: pre-calculates link header based on address family * and destination lladdr. * * Input data fields: * buf: pointer to destination buffer * bufsize: buffer size * flags: IFENCAP_FLAG_BROADCAST if destination is broadcast * family: address family defined by AF_ constant. * lladdr: pointer to link-layer address * lladdr_len: length of link-layer address * hdata: pointer to L3 header (optional, used for ARP requests). * Output data fields: * buf: encap data is stored here * bufsize: resulting encap length is stored here * lladdr_off: offset of link-layer address from encap hdr start * hdata: L3 header may be altered if necessary */ struct if_encap_req { u_char *buf; /* Destination buffer (w) */ size_t bufsize; /* size of provided buffer (r) */ ife_type rtype; /* request type (r) */ uint32_t flags; /* Request flags (r) */ int family; /* Address family AF_* (r) */ int lladdr_off; /* offset from header start (w) */ int lladdr_len; /* lladdr length (r) */ char *lladdr; /* link-level address pointer (r) */ char *hdata; /* Upper layer header data (rw) */ }; #define IFENCAP_FLAG_BROADCAST 0x02 /* Destination is broadcast */ /* * Network interface send tag support. The storage of "struct * m_snd_tag" comes from the network driver and it is free to allocate * as much additional space as it wants for its own use. */ struct ktls_session; struct m_snd_tag; #define IF_SND_TAG_TYPE_RATE_LIMIT 0 #define IF_SND_TAG_TYPE_UNLIMITED 1 #define IF_SND_TAG_TYPE_TLS 2 #define IF_SND_TAG_TYPE_MAX 3 struct if_snd_tag_alloc_header { uint32_t type; /* send tag type, see IF_SND_TAG_XXX */ uint32_t flowid; /* mbuf hash value */ uint32_t flowtype; /* mbuf hash type */ + uint8_t numa_domain; /* numa domain of associated inp */ }; struct if_snd_tag_alloc_rate_limit { struct if_snd_tag_alloc_header hdr; uint64_t max_rate; /* in bytes/s */ uint32_t flags; /* M_NOWAIT or M_WAITOK */ uint32_t reserved; /* alignment */ }; struct if_snd_tag_alloc_tls { struct if_snd_tag_alloc_header hdr; struct inpcb *inp; const struct ktls_session *tls; }; struct if_snd_tag_rate_limit_params { uint64_t max_rate; /* in bytes/s */ uint32_t queue_level; /* 0 (empty) .. 65535 (full) */ #define IF_SND_QUEUE_LEVEL_MIN 0 #define IF_SND_QUEUE_LEVEL_MAX 65535 uint32_t flags; /* M_NOWAIT or M_WAITOK */ }; union if_snd_tag_alloc_params { struct if_snd_tag_alloc_header hdr; struct if_snd_tag_alloc_rate_limit rate_limit; struct if_snd_tag_alloc_rate_limit unlimited; struct if_snd_tag_alloc_tls tls; }; union if_snd_tag_modify_params { struct if_snd_tag_rate_limit_params rate_limit; struct if_snd_tag_rate_limit_params unlimited; }; union if_snd_tag_query_params { struct if_snd_tag_rate_limit_params rate_limit; struct if_snd_tag_rate_limit_params unlimited; }; /* Query return flags */ #define RT_NOSUPPORT 0x00000000 /* Not supported */ #define RT_IS_INDIRECT 0x00000001 /* * Interface like a lagg, select * the actual interface for * capabilities. */ #define RT_IS_SELECTABLE 0x00000002 /* * No rate table, you select * rates and the first * number_of_rates are created. */ #define RT_IS_FIXED_TABLE 0x00000004 /* A fixed table is attached */ #define RT_IS_UNUSABLE 0x00000008 /* It is not usable for this */ #define RT_IS_SETUP_REQ 0x00000010 /* The interface setup must be called before use */ struct if_ratelimit_query_results { const uint64_t *rate_table; /* Pointer to table if present */ uint32_t flags; /* Flags indicating results */ uint32_t max_flows; /* Max flows using, 0=unlimited */ uint32_t number_of_rates; /* How many unique rates can be created */ uint32_t min_segment_burst; /* The amount the adapter bursts at each send */ }; typedef int (if_snd_tag_alloc_t)(struct ifnet *, union if_snd_tag_alloc_params *, struct m_snd_tag **); typedef int (if_snd_tag_modify_t)(struct m_snd_tag *, union if_snd_tag_modify_params *); typedef int (if_snd_tag_query_t)(struct m_snd_tag *, union if_snd_tag_query_params *); typedef void (if_snd_tag_free_t)(struct m_snd_tag *); typedef void (if_ratelimit_query_t)(struct ifnet *, struct if_ratelimit_query_results *); typedef int (if_ratelimit_setup_t)(struct ifnet *, uint64_t, uint32_t); /* * Structure defining a network interface. */ struct ifnet { /* General book keeping of interface lists. */ CK_STAILQ_ENTRY(ifnet) if_link; /* all struct ifnets are chained (CK_) */ LIST_ENTRY(ifnet) if_clones; /* interfaces of a cloner */ CK_STAILQ_HEAD(, ifg_list) if_groups; /* linked list of groups per if (CK_) */ /* protected by if_addr_lock */ u_char if_alloctype; /* if_type at time of allocation */ uint8_t if_numa_domain; /* NUMA domain of device */ /* Driver and protocol specific information that remains stable. */ void *if_softc; /* pointer to driver state */ void *if_llsoftc; /* link layer softc */ void *if_l2com; /* pointer to protocol bits */ const char *if_dname; /* driver name */ int if_dunit; /* unit or IF_DUNIT_NONE */ u_short if_index; /* numeric abbreviation for this if */ short if_index_reserved; /* spare space to grow if_index */ char if_xname[IFNAMSIZ]; /* external name (name + unit) */ char *if_description; /* interface description */ /* Variable fields that are touched by the stack and drivers. */ int if_flags; /* up/down, broadcast, etc. */ int if_drv_flags; /* driver-managed status flags */ int if_capabilities; /* interface features & capabilities */ int if_capenable; /* enabled features & capabilities */ void *if_linkmib; /* link-type-specific MIB data */ size_t if_linkmiblen; /* length of above data */ u_int if_refcount; /* reference count */ /* These fields are shared with struct if_data. */ uint8_t if_type; /* ethernet, tokenring, etc */ uint8_t if_addrlen; /* media address length */ uint8_t if_hdrlen; /* media header length */ uint8_t if_link_state; /* current link state */ uint32_t if_mtu; /* maximum transmission unit */ uint32_t if_metric; /* routing metric (external only) */ uint64_t if_baudrate; /* linespeed */ uint64_t if_hwassist; /* HW offload capabilities, see IFCAP */ time_t if_epoch; /* uptime at attach or stat reset */ struct timeval if_lastchange; /* time of last administrative change */ struct ifaltq if_snd; /* output queue (includes altq) */ struct task if_linktask; /* task for link change events */ struct task if_addmultitask; /* task for SIOCADDMULTI */ /* Addresses of different protocol families assigned to this if. */ struct mtx if_addr_lock; /* lock to protect address lists */ /* * if_addrhead is the list of all addresses associated to * an interface. * Some code in the kernel assumes that first element * of the list has type AF_LINK, and contains sockaddr_dl * addresses which store the link-level address and the name * of the interface. * However, access to the AF_LINK address through this * field is deprecated. Use if_addr or ifaddr_byindex() instead. */ struct ifaddrhead if_addrhead; /* linked list of addresses per if */ struct ifmultihead if_multiaddrs; /* multicast addresses configured */ int if_amcount; /* number of all-multicast requests */ struct ifaddr *if_addr; /* pointer to link-level address */ void *if_hw_addr; /* hardware link-level address */ const u_int8_t *if_broadcastaddr; /* linklevel broadcast bytestring */ struct mtx if_afdata_lock; void *if_afdata[AF_MAX]; int if_afdata_initialized; /* Additional features hung off the interface. */ u_int if_fib; /* interface FIB */ struct vnet *if_vnet; /* pointer to network stack instance */ struct vnet *if_home_vnet; /* where this ifnet originates from */ struct ifvlantrunk *if_vlantrunk; /* pointer to 802.1q data */ struct bpf_if *if_bpf; /* packet filter structure */ int if_pcount; /* number of promiscuous listeners */ void *if_bridge; /* bridge glue */ void *if_lagg; /* lagg glue */ void *if_pf_kif; /* pf glue */ struct carp_if *if_carp; /* carp interface structure */ struct label *if_label; /* interface MAC label */ struct netmap_adapter *if_netmap; /* netmap(4) softc */ /* Various procedures of the layer2 encapsulation and drivers. */ int (*if_output) /* output routine (enqueue) */ (struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); void (*if_input) /* input routine (from h/w driver) */ (struct ifnet *, struct mbuf *); struct mbuf *(*if_bridge_input)(struct ifnet *, struct mbuf *); int (*if_bridge_output)(struct ifnet *, struct mbuf *, struct sockaddr *, struct rtentry *); void (*if_bridge_linkstate)(struct ifnet *ifp); if_start_fn_t if_start; /* initiate output routine */ if_ioctl_fn_t if_ioctl; /* ioctl routine */ if_init_fn_t if_init; /* Init routine */ int (*if_resolvemulti) /* validate/resolve multicast */ (struct ifnet *, struct sockaddr **, struct sockaddr *); if_qflush_fn_t if_qflush; /* flush any queue */ if_transmit_fn_t if_transmit; /* initiate output routine */ void (*if_reassign) /* reassign to vnet routine */ (struct ifnet *, struct vnet *, char *); if_get_counter_t if_get_counter; /* get counter values */ int (*if_requestencap) /* make link header from request */ (struct ifnet *, struct if_encap_req *); /* Statistics. */ counter_u64_t if_counters[IFCOUNTERS]; /* Stuff that's only temporary and doesn't belong here. */ /* * Network adapter TSO limits: * =========================== * * If the "if_hw_tsomax" field is zero the maximum segment * length limit does not apply. If the "if_hw_tsomaxsegcount" * or the "if_hw_tsomaxsegsize" field is zero the TSO segment * count limit does not apply. If all three fields are zero, * there is no TSO limit. * * NOTE: The TSO limits should reflect the values used in the * BUSDMA tag a network adapter is using to load a mbuf chain * for transmission. The TCP/IP network stack will subtract * space for all linklevel and protocol level headers and * ensure that the full mbuf chain passed to the network * adapter fits within the given limits. */ u_int if_hw_tsomax; /* TSO maximum size in bytes */ u_int if_hw_tsomaxsegcount; /* TSO maximum segment count */ u_int if_hw_tsomaxsegsize; /* TSO maximum segment size in bytes */ /* * Network adapter send tag support: */ if_snd_tag_alloc_t *if_snd_tag_alloc; if_snd_tag_modify_t *if_snd_tag_modify; if_snd_tag_query_t *if_snd_tag_query; if_snd_tag_free_t *if_snd_tag_free; if_ratelimit_query_t *if_ratelimit_query; if_ratelimit_setup_t *if_ratelimit_setup; /* Ethernet PCP */ uint8_t if_pcp; /* * Debugnet (Netdump) hooks to be called while in db/panic. */ struct debugnet_methods *if_debugnet_methods; struct epoch_context if_epoch_ctx; /* * Spare fields to be added before branching a stable branch, so * that structure can be enhanced without changing the kernel * binary interface. */ int if_ispare[4]; /* general use */ }; /* for compatibility with other BSDs */ #define if_name(ifp) ((ifp)->if_xname) #define IF_NODOM 255 /* * Locks for address lists on the network interface. */ #define IF_ADDR_LOCK_INIT(if) mtx_init(&(if)->if_addr_lock, "if_addr_lock", NULL, MTX_DEF) #define IF_ADDR_LOCK_DESTROY(if) mtx_destroy(&(if)->if_addr_lock) #define IF_ADDR_WLOCK(if) mtx_lock(&(if)->if_addr_lock) #define IF_ADDR_WUNLOCK(if) mtx_unlock(&(if)->if_addr_lock) #define IF_ADDR_LOCK_ASSERT(if) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(if)->if_addr_lock)) #define IF_ADDR_WLOCK_ASSERT(if) mtx_assert(&(if)->if_addr_lock, MA_OWNED) #ifdef _KERNEL /* interface link layer address change event */ typedef void (*iflladdr_event_handler_t)(void *, struct ifnet *); EVENTHANDLER_DECLARE(iflladdr_event, iflladdr_event_handler_t); /* interface address change event */ typedef void (*ifaddr_event_handler_t)(void *, struct ifnet *); EVENTHANDLER_DECLARE(ifaddr_event, ifaddr_event_handler_t); typedef void (*ifaddr_event_ext_handler_t)(void *, struct ifnet *, struct ifaddr *, int); EVENTHANDLER_DECLARE(ifaddr_event_ext, ifaddr_event_ext_handler_t); #define IFADDR_EVENT_ADD 0 #define IFADDR_EVENT_DEL 1 /* new interface arrival event */ typedef void (*ifnet_arrival_event_handler_t)(void *, struct ifnet *); EVENTHANDLER_DECLARE(ifnet_arrival_event, ifnet_arrival_event_handler_t); /* interface departure event */ typedef void (*ifnet_departure_event_handler_t)(void *, struct ifnet *); EVENTHANDLER_DECLARE(ifnet_departure_event, ifnet_departure_event_handler_t); /* Interface link state change event */ typedef void (*ifnet_link_event_handler_t)(void *, struct ifnet *, int); EVENTHANDLER_DECLARE(ifnet_link_event, ifnet_link_event_handler_t); /* Interface up/down event */ #define IFNET_EVENT_UP 0 #define IFNET_EVENT_DOWN 1 #define IFNET_EVENT_PCP 2 /* priority code point, PCP */ typedef void (*ifnet_event_fn)(void *, struct ifnet *ifp, int event); EVENTHANDLER_DECLARE(ifnet_event, ifnet_event_fn); /* * interface groups */ struct ifg_group { char ifg_group[IFNAMSIZ]; u_int ifg_refcnt; void *ifg_pf_kif; CK_STAILQ_HEAD(, ifg_member) ifg_members; /* (CK_) */ CK_STAILQ_ENTRY(ifg_group) ifg_next; /* (CK_) */ }; struct ifg_member { CK_STAILQ_ENTRY(ifg_member) ifgm_next; /* (CK_) */ struct ifnet *ifgm_ifp; }; struct ifg_list { struct ifg_group *ifgl_group; CK_STAILQ_ENTRY(ifg_list) ifgl_next; /* (CK_) */ }; #ifdef _SYS_EVENTHANDLER_H_ /* group attach event */ typedef void (*group_attach_event_handler_t)(void *, struct ifg_group *); EVENTHANDLER_DECLARE(group_attach_event, group_attach_event_handler_t); /* group detach event */ typedef void (*group_detach_event_handler_t)(void *, struct ifg_group *); EVENTHANDLER_DECLARE(group_detach_event, group_detach_event_handler_t); /* group change event */ typedef void (*group_change_event_handler_t)(void *, const char *); EVENTHANDLER_DECLARE(group_change_event, group_change_event_handler_t); #endif /* _SYS_EVENTHANDLER_H_ */ #define IF_AFDATA_LOCK_INIT(ifp) \ mtx_init(&(ifp)->if_afdata_lock, "if_afdata", NULL, MTX_DEF) #define IF_AFDATA_WLOCK(ifp) mtx_lock(&(ifp)->if_afdata_lock) #define IF_AFDATA_WUNLOCK(ifp) mtx_unlock(&(ifp)->if_afdata_lock) #define IF_AFDATA_LOCK(ifp) IF_AFDATA_WLOCK(ifp) #define IF_AFDATA_UNLOCK(ifp) IF_AFDATA_WUNLOCK(ifp) #define IF_AFDATA_TRYLOCK(ifp) mtx_trylock(&(ifp)->if_afdata_lock) #define IF_AFDATA_DESTROY(ifp) mtx_destroy(&(ifp)->if_afdata_lock) #define IF_AFDATA_LOCK_ASSERT(ifp) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ifp)->if_afdata_lock)) #define IF_AFDATA_WLOCK_ASSERT(ifp) mtx_assert(&(ifp)->if_afdata_lock, MA_OWNED) #define IF_AFDATA_UNLOCK_ASSERT(ifp) mtx_assert(&(ifp)->if_afdata_lock, MA_NOTOWNED) /* * 72 was chosen below because it is the size of a TCP/IP * header (40) + the minimum mss (32). */ #define IF_MINMTU 72 #define IF_MAXMTU 65535 #define TOEDEV(ifp) ((ifp)->if_llsoftc) /* * The ifaddr structure contains information about one address * of an interface. They are maintained by the different address families, * are allocated and attached when an address is set, and are linked * together so all addresses for an interface can be located. * * NOTE: a 'struct ifaddr' is always at the beginning of a larger * chunk of malloc'ed memory, where we store the three addresses * (ifa_addr, ifa_dstaddr and ifa_netmask) referenced here. */ struct ifaddr { struct sockaddr *ifa_addr; /* address of interface */ struct sockaddr *ifa_dstaddr; /* other end of p-to-p link */ #define ifa_broadaddr ifa_dstaddr /* broadcast address interface */ struct sockaddr *ifa_netmask; /* used to determine subnet */ struct ifnet *ifa_ifp; /* back-pointer to interface */ struct carp_softc *ifa_carp; /* pointer to CARP data */ CK_STAILQ_ENTRY(ifaddr) ifa_link; /* queue macro glue */ void (*ifa_rtrequest) /* check or clean routes (+ or -)'d */ (int, struct rtentry *, struct rt_addrinfo *); u_short ifa_flags; /* mostly rt_flags for cloning */ #define IFA_ROUTE RTF_UP /* route installed */ #define IFA_RTSELF RTF_HOST /* loopback route to self installed */ u_int ifa_refcnt; /* references to this structure */ counter_u64_t ifa_ipackets; counter_u64_t ifa_opackets; counter_u64_t ifa_ibytes; counter_u64_t ifa_obytes; struct epoch_context ifa_epoch_ctx; }; struct ifaddr * ifa_alloc(size_t size, int flags); void ifa_free(struct ifaddr *ifa); void ifa_ref(struct ifaddr *ifa); /* * Multicast address structure. This is analogous to the ifaddr * structure except that it keeps track of multicast addresses. */ #define IFMA_F_ENQUEUED 0x1 struct ifmultiaddr { CK_STAILQ_ENTRY(ifmultiaddr) ifma_link; /* queue macro glue */ struct sockaddr *ifma_addr; /* address this membership is for */ struct sockaddr *ifma_lladdr; /* link-layer translation, if any */ struct ifnet *ifma_ifp; /* back-pointer to interface */ u_int ifma_refcount; /* reference count */ int ifma_flags; void *ifma_protospec; /* protocol-specific state, if any */ struct ifmultiaddr *ifma_llifma; /* pointer to ifma for ifma_lladdr */ struct epoch_context ifma_epoch_ctx; }; extern struct rwlock ifnet_rwlock; extern struct sx ifnet_sxlock; #define IFNET_WLOCK() do { \ sx_xlock(&ifnet_sxlock); \ rw_wlock(&ifnet_rwlock); \ } while (0) #define IFNET_WUNLOCK() do { \ rw_wunlock(&ifnet_rwlock); \ sx_xunlock(&ifnet_sxlock); \ } while (0) /* * To assert the ifnet lock, you must know not only whether it's for read or * write, but also whether it was acquired with sleep support or not. */ #define IFNET_RLOCK_ASSERT() sx_assert(&ifnet_sxlock, SA_SLOCKED) #define IFNET_WLOCK_ASSERT() do { \ sx_assert(&ifnet_sxlock, SA_XLOCKED); \ rw_assert(&ifnet_rwlock, RA_WLOCKED); \ } while (0) #define IFNET_RLOCK() sx_slock(&ifnet_sxlock) #define IFNET_RUNLOCK() sx_sunlock(&ifnet_sxlock) /* * Look up an ifnet given its index; the _ref variant also acquires a * reference that must be freed using if_rele(). It is almost always a bug * to call ifnet_byindex() instead of ifnet_byindex_ref(). */ struct ifnet *ifnet_byindex(u_short idx); struct ifnet *ifnet_byindex_ref(u_short idx); /* * Given the index, ifaddr_byindex() returns the one and only * link-level ifaddr for the interface. You are not supposed to use * it to traverse the list of addresses associated to the interface. */ struct ifaddr *ifaddr_byindex(u_short idx); VNET_DECLARE(struct ifnethead, ifnet); VNET_DECLARE(struct ifgrouphead, ifg_head); VNET_DECLARE(int, if_index); VNET_DECLARE(struct ifnet *, loif); /* first loopback interface */ #define V_ifnet VNET(ifnet) #define V_ifg_head VNET(ifg_head) #define V_if_index VNET(if_index) #define V_loif VNET(loif) #ifdef MCAST_VERBOSE #define MCDPRINTF printf #else #define MCDPRINTF(...) #endif int if_addgroup(struct ifnet *, const char *); int if_delgroup(struct ifnet *, const char *); int if_addmulti(struct ifnet *, struct sockaddr *, struct ifmultiaddr **); int if_allmulti(struct ifnet *, int); struct ifnet* if_alloc(u_char); struct ifnet* if_alloc_dev(u_char, device_t dev); struct ifnet* if_alloc_domain(u_char, int numa_domain); void if_attach(struct ifnet *); void if_dead(struct ifnet *); int if_delmulti(struct ifnet *, struct sockaddr *); void if_delmulti_ifma(struct ifmultiaddr *); void if_delmulti_ifma_flags(struct ifmultiaddr *, int flags); void if_detach(struct ifnet *); void if_purgeaddrs(struct ifnet *); void if_delallmulti(struct ifnet *); void if_down(struct ifnet *); struct ifmultiaddr * if_findmulti(struct ifnet *, const struct sockaddr *); void if_freemulti(struct ifmultiaddr *ifma); void if_free(struct ifnet *); void if_initname(struct ifnet *, const char *, int); void if_link_state_change(struct ifnet *, int); int if_printf(struct ifnet *, const char *, ...) __printflike(2, 3); void if_ref(struct ifnet *); void if_rele(struct ifnet *); int if_setlladdr(struct ifnet *, const u_char *, int); int if_tunnel_check_nesting(struct ifnet *, struct mbuf *, uint32_t, int); void if_up(struct ifnet *); int ifioctl(struct socket *, u_long, caddr_t, struct thread *); int ifpromisc(struct ifnet *, int); struct ifnet *ifunit(const char *); struct ifnet *ifunit_ref(const char *); int ifa_add_loopback_route(struct ifaddr *, struct sockaddr *); int ifa_del_loopback_route(struct ifaddr *, struct sockaddr *); int ifa_switch_loopback_route(struct ifaddr *, struct sockaddr *); struct ifaddr *ifa_ifwithaddr(const struct sockaddr *); int ifa_ifwithaddr_check(const struct sockaddr *); struct ifaddr *ifa_ifwithbroadaddr(const struct sockaddr *, int); struct ifaddr *ifa_ifwithdstaddr(const struct sockaddr *, int); struct ifaddr *ifa_ifwithnet(const struct sockaddr *, int, int); struct ifaddr *ifa_ifwithroute(int, const struct sockaddr *, struct sockaddr *, u_int); struct ifaddr *ifaof_ifpforaddr(const struct sockaddr *, struct ifnet *); int ifa_preferred(struct ifaddr *, struct ifaddr *); int if_simloop(struct ifnet *ifp, struct mbuf *m, int af, int hlen); typedef void *if_com_alloc_t(u_char type, struct ifnet *ifp); typedef void if_com_free_t(void *com, u_char type); void if_register_com_alloc(u_char type, if_com_alloc_t *a, if_com_free_t *f); void if_deregister_com_alloc(u_char type); void if_data_copy(struct ifnet *, struct if_data *); uint64_t if_get_counter_default(struct ifnet *, ift_counter); void if_inc_counter(struct ifnet *, ift_counter, int64_t); #define IF_LLADDR(ifp) \ LLADDR((struct sockaddr_dl *)((ifp)->if_addr->ifa_addr)) uint64_t if_setbaudrate(if_t ifp, uint64_t baudrate); uint64_t if_getbaudrate(if_t ifp); int if_setcapabilities(if_t ifp, int capabilities); int if_setcapabilitiesbit(if_t ifp, int setbit, int clearbit); int if_getcapabilities(if_t ifp); int if_togglecapenable(if_t ifp, int togglecap); int if_setcapenable(if_t ifp, int capenable); int if_setcapenablebit(if_t ifp, int setcap, int clearcap); int if_getcapenable(if_t ifp); const char *if_getdname(if_t ifp); int if_setdev(if_t ifp, void *dev); int if_setdrvflagbits(if_t ifp, int if_setflags, int clear_flags); int if_getdrvflags(if_t ifp); int if_setdrvflags(if_t ifp, int flags); int if_clearhwassist(if_t ifp); int if_sethwassistbits(if_t ifp, int toset, int toclear); int if_sethwassist(if_t ifp, int hwassist_bit); int if_gethwassist(if_t ifp); int if_setsoftc(if_t ifp, void *softc); void *if_getsoftc(if_t ifp); int if_setflags(if_t ifp, int flags); int if_gethwaddr(if_t ifp, struct ifreq *); int if_setmtu(if_t ifp, int mtu); int if_getmtu(if_t ifp); int if_getmtu_family(if_t ifp, int family); int if_setflagbits(if_t ifp, int set, int clear); int if_getflags(if_t ifp); int if_sendq_empty(if_t ifp); int if_setsendqready(if_t ifp); int if_setsendqlen(if_t ifp, int tx_desc_count); int if_sethwtsomax(if_t ifp, u_int if_hw_tsomax); int if_sethwtsomaxsegcount(if_t ifp, u_int if_hw_tsomaxsegcount); int if_sethwtsomaxsegsize(if_t ifp, u_int if_hw_tsomaxsegsize); u_int if_gethwtsomax(if_t ifp); u_int if_gethwtsomaxsegcount(if_t ifp); u_int if_gethwtsomaxsegsize(if_t ifp); int if_input(if_t ifp, struct mbuf* sendmp); int if_sendq_prepend(if_t ifp, struct mbuf *m); struct mbuf *if_dequeue(if_t ifp); int if_setifheaderlen(if_t ifp, int len); void if_setrcvif(struct mbuf *m, if_t ifp); void if_setvtag(struct mbuf *m, u_int16_t tag); u_int16_t if_getvtag(struct mbuf *m); int if_vlantrunkinuse(if_t ifp); caddr_t if_getlladdr(if_t ifp); void *if_gethandle(u_char); void if_bpfmtap(if_t ifp, struct mbuf *m); void if_etherbpfmtap(if_t ifp, struct mbuf *m); void if_vlancap(if_t ifp); /* * Traversing through interface address lists. */ struct sockaddr_dl; typedef u_int iflladdr_cb_t(void *, struct sockaddr_dl *, u_int); u_int if_foreach_lladdr(if_t, iflladdr_cb_t, void *); u_int if_foreach_llmaddr(if_t, iflladdr_cb_t, void *); u_int if_lladdr_count(if_t); u_int if_llmaddr_count(if_t); int if_getamcount(if_t ifp); struct ifaddr * if_getifaddr(if_t ifp); /* Functions */ void if_setinitfn(if_t ifp, void (*)(void *)); void if_setioctlfn(if_t ifp, int (*)(if_t, u_long, caddr_t)); void if_setstartfn(if_t ifp, void (*)(if_t)); void if_settransmitfn(if_t ifp, if_transmit_fn_t); void if_setqflushfn(if_t ifp, if_qflush_fn_t); void if_setgetcounterfn(if_t ifp, if_get_counter_t); /* Revisit the below. These are inline functions originally */ int drbr_inuse_drv(if_t ifp, struct buf_ring *br); struct mbuf* drbr_dequeue_drv(if_t ifp, struct buf_ring *br); int drbr_needs_enqueue_drv(if_t ifp, struct buf_ring *br); int drbr_enqueue_drv(if_t ifp, struct buf_ring *br, struct mbuf *m); /* TSO */ void if_hw_tsomax_common(if_t ifp, struct ifnet_hw_tsomax *); int if_hw_tsomax_update(if_t ifp, struct ifnet_hw_tsomax *); /* accessors for struct ifreq */ void *ifr_data_get_ptr(void *ifrp); void *ifr_buffer_get_buffer(void *data); size_t ifr_buffer_get_length(void *data); int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *); #ifdef DEVICE_POLLING enum poll_cmd { POLL_ONLY, POLL_AND_CHECK_STATUS }; typedef int poll_handler_t(if_t ifp, enum poll_cmd cmd, int count); int ether_poll_register(poll_handler_t *h, if_t ifp); int ether_poll_deregister(if_t ifp); #endif /* DEVICE_POLLING */ #endif /* _KERNEL */ #include /* XXXAO: temporary unconditional include */ #endif /* !_NET_IF_VAR_H_ */ Index: head/sys/netinet/in_pcb.c =================================================================== --- head/sys/netinet/in_pcb.c (revision 358807) +++ head/sys/netinet/in_pcb.c (revision 358808) @@ -1,3474 +1,3475 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1991, 1993, 1995 * The Regents of the University of California. * Copyright (c) 2007-2009 Robert N. M. Watson * 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 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 8.4 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_ipsec.h" #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ratelimit.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 #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #ifdef INET #include #endif #include #include #ifdef TCPHPTS #include #endif #include #include #ifdef INET6 #include #include #include #include #endif /* INET6 */ #endif #include #include #define INPCBLBGROUP_SIZMIN 8 #define INPCBLBGROUP_SIZMAX 256 static struct callout ipport_tick_callout; /* * These configure the range of local port addresses assigned to * "unspecified" outgoing connections/packets/whatever. */ VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ /* * Reserved ports accessible only to root. There are significant * security considerations that must be accounted for when changing these, * but the security benefits can be great. Please be careful. */ VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ VNET_DEFINE(int, ipport_reservedlow); /* Variables dealing with random ephemeral port allocation. */ VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ VNET_DEFINE(int, ipport_tcpallocs); VNET_DEFINE_STATIC(int, ipport_tcplastcount); #define V_ipport_tcplastcount VNET(ipport_tcplastcount) static void in_pcbremlists(struct inpcb *inp); #ifdef INET static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp); #define RANGECHK(var, min, max) \ if ((var) < (min)) { (var) = (min); } \ else if ((var) > (max)) { (var) = (max); } static int sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) { int error; error = sysctl_handle_int(oidp, arg1, arg2, req); if (error == 0) { RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); } return (error); } #undef RANGECHK static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "IP Ports"); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " "allocations before switching to a sequental one"); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomtime), 0, "Minimum time to keep sequental port " "allocation before switching to a random one"); #ifdef RATELIMIT counter_u64_t rate_limit_active; counter_u64_t rate_limit_alloc_fail; counter_u64_t rate_limit_set_ok; static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "IP Rate Limiting"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, &rate_limit_active, "Active rate limited connections"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, &rate_limit_alloc_fail, "Rate limited connection failures"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, &rate_limit_set_ok, "Rate limited setting succeeded"); #endif /* RATELIMIT */ #endif /* INET */ /* * in_pcb.c: manage the Protocol Control Blocks. * * NOTE: It is assumed that most of these functions will be called with * the pcbinfo lock held, and often, the inpcb lock held, as these utility * functions often modify hash chains or addresses in pcbs. */ static struct inpcblbgroup * in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, uint16_t port, const union in_dependaddr *addr, int size) { struct inpcblbgroup *grp; size_t bytes; bytes = __offsetof(struct inpcblbgroup, il_inp[size]); grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); if (!grp) return (NULL); grp->il_vflag = vflag; grp->il_lport = port; grp->il_dependladdr = *addr; grp->il_inpsiz = size; CK_LIST_INSERT_HEAD(hdr, grp, il_list); return (grp); } static void in_pcblbgroup_free_deferred(epoch_context_t ctx) { struct inpcblbgroup *grp; grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); free(grp, M_PCB); } static void in_pcblbgroup_free(struct inpcblbgroup *grp) { CK_LIST_REMOVE(grp, il_list); NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); } static struct inpcblbgroup * in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, struct inpcblbgroup *old_grp, int size) { struct inpcblbgroup *grp; int i; grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, old_grp->il_lport, &old_grp->il_dependladdr, size); if (grp == NULL) return (NULL); KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, ("invalid new local group size %d and old local group count %d", grp->il_inpsiz, old_grp->il_inpcnt)); for (i = 0; i < old_grp->il_inpcnt; ++i) grp->il_inp[i] = old_grp->il_inp[i]; grp->il_inpcnt = old_grp->il_inpcnt; in_pcblbgroup_free(old_grp); return (grp); } /* * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] * and shrink group if possible. */ static void in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, int i) { struct inpcblbgroup *grp, *new_grp; grp = *grpp; for (; i + 1 < grp->il_inpcnt; ++i) grp->il_inp[i] = grp->il_inp[i + 1]; grp->il_inpcnt--; if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && grp->il_inpcnt <= grp->il_inpsiz / 4) { /* Shrink this group. */ new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); if (new_grp != NULL) *grpp = new_grp; } } /* * Add PCB to load balance group for SO_REUSEPORT_LB option. */ static int in_pcbinslbgrouphash(struct inpcb *inp) { const static struct timeval interval = { 60, 0 }; static struct timeval lastprint; struct inpcbinfo *pcbinfo; struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; uint32_t idx; pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); /* * Don't allow jailed socket to join local group. */ if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred)) return (0); #ifdef INET6 /* * Don't allow IPv4 mapped INET6 wild socket. */ if ((inp->inp_vflag & INP_IPV4) && inp->inp_laddr.s_addr == INADDR_ANY && INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { return (0); } #endif idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; CK_LIST_FOREACH(grp, hdr, il_list) { if (grp->il_vflag == inp->inp_vflag && grp->il_lport == inp->inp_lport && memcmp(&grp->il_dependladdr, &inp->inp_inc.inc_ie.ie_dependladdr, sizeof(grp->il_dependladdr)) == 0) break; } if (grp == NULL) { /* Create new load balance group. */ grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, INPCBLBGROUP_SIZMIN); if (grp == NULL) return (ENOBUFS); } else if (grp->il_inpcnt == grp->il_inpsiz) { if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { if (ratecheck(&lastprint, &interval)) printf("lb group port %d, limit reached\n", ntohs(grp->il_lport)); return (0); } /* Expand this local group. */ grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); if (grp == NULL) return (ENOBUFS); } KASSERT(grp->il_inpcnt < grp->il_inpsiz, ("invalid local group size %d and count %d", grp->il_inpsiz, grp->il_inpcnt)); grp->il_inp[grp->il_inpcnt] = inp; grp->il_inpcnt++; return (0); } /* * Remove PCB from load balance group. */ static void in_pcbremlbgrouphash(struct inpcb *inp) { struct inpcbinfo *pcbinfo; struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; int i; pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; CK_LIST_FOREACH(grp, hdr, il_list) { for (i = 0; i < grp->il_inpcnt; ++i) { if (grp->il_inp[i] != inp) continue; if (grp->il_inpcnt == 1) { /* We are the last, free this local group. */ in_pcblbgroup_free(grp); } else { /* Pull up inpcbs, shrink group if possible. */ in_pcblbgroup_reorder(hdr, &grp, i); } return; } } } /* * Different protocols initialize their inpcbs differently - giving * different name to the lock. But they all are disposed the same. */ static void inpcb_fini(void *mem, int size) { struct inpcb *inp = mem; INP_LOCK_DESTROY(inp); } /* * Initialize an inpcbinfo -- we should be able to reduce the number of * arguments in time. */ void in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) { porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); INP_INFO_LOCK_INIT(pcbinfo, name); INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); #ifdef VIMAGE pcbinfo->ipi_vnet = curvnet; #endif pcbinfo->ipi_listhead = listhead; CK_LIST_INIT(pcbinfo->ipi_listhead); pcbinfo->ipi_count = 0; pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, &pcbinfo->ipi_hashmask); pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, &pcbinfo->ipi_porthashmask); pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, &pcbinfo->ipi_lbgrouphashmask); #ifdef PCBGROUP in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); #endif pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); uma_zone_set_warning(pcbinfo->ipi_zone, "kern.ipc.maxsockets limit reached"); } /* * Destroy an inpcbinfo. */ void in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) { KASSERT(pcbinfo->ipi_count == 0, ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, pcbinfo->ipi_porthashmask); hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, pcbinfo->ipi_lbgrouphashmask); #ifdef PCBGROUP in_pcbgroup_destroy(pcbinfo); #endif uma_zdestroy(pcbinfo->ipi_zone); INP_LIST_LOCK_DESTROY(pcbinfo); INP_HASH_LOCK_DESTROY(pcbinfo); INP_INFO_LOCK_DESTROY(pcbinfo); } /* * Allocate a PCB and associate it with the socket. * On success return with the PCB locked. */ int in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) { struct inpcb *inp; int error; error = 0; inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); if (inp == NULL) return (ENOBUFS); bzero(&inp->inp_start_zero, inp_zero_size); #ifdef NUMA inp->inp_numa_domain = M_NODOM; #endif inp->inp_pcbinfo = pcbinfo; inp->inp_socket = so; inp->inp_cred = crhold(so->so_cred); inp->inp_inc.inc_fibnum = so->so_fibnum; #ifdef MAC error = mac_inpcb_init(inp, M_NOWAIT); if (error != 0) goto out; mac_inpcb_create(so, inp); #endif #if defined(IPSEC) || defined(IPSEC_SUPPORT) error = ipsec_init_pcbpolicy(inp); if (error != 0) { #ifdef MAC mac_inpcb_destroy(inp); #endif goto out; } #endif /*IPSEC*/ #ifdef INET6 if (INP_SOCKAF(so) == AF_INET6) { inp->inp_vflag |= INP_IPV6PROTO; if (V_ip6_v6only) inp->inp_flags |= IN6P_IPV6_V6ONLY; } #endif INP_WLOCK(inp); INP_LIST_WLOCK(pcbinfo); CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); pcbinfo->ipi_count++; so->so_pcb = (caddr_t)inp; #ifdef INET6 if (V_ip6_auto_flowlabel) inp->inp_flags |= IN6P_AUTOFLOWLABEL; #endif inp->inp_gencnt = ++pcbinfo->ipi_gencnt; refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ /* * Routes in inpcb's can cache L2 as well; they are guaranteed * to be cleaned up. */ inp->inp_route.ro_flags = RT_LLE_CACHE; INP_LIST_WUNLOCK(pcbinfo); #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) out: if (error != 0) { crfree(inp->inp_cred); uma_zfree(pcbinfo->ipi_zone, inp); } #endif return (error); } #ifdef INET int in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { int anonport, error; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) return (EINVAL); anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, &inp->inp_lport, cred); if (error) return (error); if (in_pcbinshash(inp) != 0) { inp->inp_laddr.s_addr = INADDR_ANY; inp->inp_lport = 0; return (EAGAIN); } if (anonport) inp->inp_flags |= INP_ANONPORT; return (0); } #endif /* * Select a local port (number) to use. */ #if defined(INET) || defined(INET6) int in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, struct ucred *cred, int lookupflags) { struct inpcbinfo *pcbinfo; struct inpcb *tmpinp; unsigned short *lastport; int count, dorandom, error; u_short aux, first, last, lport; #ifdef INET struct in_addr laddr; #endif pcbinfo = inp->inp_pcbinfo; /* * Because no actual state changes occur here, a global write lock on * the pcbinfo isn't required. */ INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(pcbinfo); if (inp->inp_flags & INP_HIGHPORT) { first = V_ipport_hifirstauto; /* sysctl */ last = V_ipport_hilastauto; lastport = &pcbinfo->ipi_lasthi; } else if (inp->inp_flags & INP_LOWPORT) { error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); if (error) return (error); first = V_ipport_lowfirstauto; /* 1023 */ last = V_ipport_lowlastauto; /* 600 */ lastport = &pcbinfo->ipi_lastlow; } else { first = V_ipport_firstauto; /* sysctl */ last = V_ipport_lastauto; lastport = &pcbinfo->ipi_lastport; } /* * For UDP(-Lite), use random port allocation as long as the user * allows it. For TCP (and as of yet unknown) connections, * use random port allocation only if the user allows it AND * ipport_tick() allows it. */ if (V_ipport_randomized && (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || pcbinfo == &V_ulitecbinfo)) dorandom = 1; else dorandom = 0; /* * It makes no sense to do random port allocation if * we have the only port available. */ if (first == last) dorandom = 0; /* Make sure to not include UDP(-Lite) packets in the count. */ if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) V_ipport_tcpallocs++; /* * Instead of having two loops further down counting up or down * make sure that first is always <= last and go with only one * code path implementing all logic. */ if (first > last) { aux = first; first = last; last = aux; } #ifdef INET /* Make the compiler happy. */ laddr.s_addr = 0; if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", __func__, inp)); laddr = *laddrp; } #endif tmpinp = NULL; /* Make compiler happy. */ lport = *lportp; if (dorandom) *lastport = first + (arc4random() % (last - first)); count = last - first; do { if (count-- < 0) /* completely used? */ return (EADDRNOTAVAIL); ++*lastport; if (*lastport < first || *lastport > last) *lastport = first; lport = htons(*lastport); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) tmpinp = in6_pcblookup_local(pcbinfo, &inp->in6p_laddr, lport, lookupflags, cred); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tmpinp = in_pcblookup_local(pcbinfo, laddr, lport, lookupflags, cred); #endif } while (tmpinp != NULL); #ifdef INET if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) laddrp->s_addr = laddr.s_addr; #endif *lportp = lport; return (0); } /* * Return cached socket options. */ int inp_so_options(const struct inpcb *inp) { int so_options; so_options = 0; if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) so_options |= SO_REUSEPORT_LB; if ((inp->inp_flags2 & INP_REUSEPORT) != 0) so_options |= SO_REUSEPORT; if ((inp->inp_flags2 & INP_REUSEADDR) != 0) so_options |= SO_REUSEADDR; return (so_options); } #endif /* INET || INET6 */ /* * Check if a new BINDMULTI socket is allowed to be created. * * ni points to the new inp. * oi points to the exisitng inp. * * This checks whether the existing inp also has BINDMULTI and * whether the credentials match. */ int in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) { /* Check permissions match */ if ((ni->inp_flags2 & INP_BINDMULTI) && (ni->inp_cred->cr_uid != oi->inp_cred->cr_uid)) return (0); /* Check the existing inp has BINDMULTI set */ if ((ni->inp_flags2 & INP_BINDMULTI) && ((oi->inp_flags2 & INP_BINDMULTI) == 0)) return (0); /* * We're okay - either INP_BINDMULTI isn't set on ni, or * it is and it matches the checks. */ return (1); } #ifdef INET /* * Set up a bind operation on a PCB, performing port allocation * as required, but do not actually modify the PCB. Callers can * either complete the bind by setting inp_laddr/inp_lport and * calling in_pcbinshash(), or they can just use the resulting * port and address to authorise the sending of a once-off packet. * * On error, the values of *laddrp and *lportp are not changed. */ int in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, u_short *lportp, struct ucred *cred) { struct socket *so = inp->inp_socket; struct sockaddr_in *sin; struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct in_addr laddr; u_short lport = 0; int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); int error; /* * 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); /* * No state changes, so read locks are sufficient here. */ INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(pcbinfo); if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ return (EADDRNOTAVAIL); laddr.s_addr = *laddrp; if (nam != NULL && laddr.s_addr != INADDR_ANY) return (EINVAL); if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) lookupflags = INPLOOKUP_WILDCARD; if (nam == NULL) { if ((error = prison_local_ip4(cred, &laddr)) != 0) return (error); } else { sin = (struct sockaddr_in *)nam; if (nam->sa_len != sizeof (*sin)) return (EINVAL); #ifdef notdef /* * We should check the family, but old programs * incorrectly fail to initialize it. */ if (sin->sin_family != AF_INET) return (EAFNOSUPPORT); #endif error = prison_local_ip4(cred, &sin->sin_addr); if (error) return (error); if (sin->sin_port != *lportp) { /* Don't allow the port to change. */ if (*lportp != 0) return (EINVAL); lport = sin->sin_port; } /* NB: lport is left as 0 if the port isn't being changed. */ if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { /* * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; * allow complete 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 (sin->sin_addr.s_addr != INADDR_ANY) { sin->sin_port = 0; /* yech... */ bzero(&sin->sin_zero, sizeof(sin->sin_zero)); /* * Is the address a local IP address? * If INP_BINDANY is set, then the socket may be bound * to any endpoint address, local or not. */ if ((inp->inp_flags & INP_BINDANY) == 0 && ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) return (EADDRNOTAVAIL); } laddr = sin->sin_addr; 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 (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { t = in_pcblookup_local(pcbinfo, sin->sin_addr, lport, INPLOOKUP_WILDCARD, cred); /* * XXX * This entire block sorely needs a rewrite. */ 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) && (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || ntohl(t->inp_laddr.s_addr) != INADDR_ANY || (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); } t = in_pcblookup_local(pcbinfo, sin->sin_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 && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && (reuseport & inp_so_options(t)) == 0 && (reuseport_lb & inp_so_options(t)) == 0) { #ifdef INET6 if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || ntohl(t->inp_laddr.s_addr) != INADDR_ANY || (inp->inp_vflag & INP_IPV6PROTO) == 0 || (t->inp_vflag & INP_IPV6PROTO) == 0) #endif return (EADDRINUSE); if (t && (! in_pcbbind_check_bindmulti(inp, t))) return (EADDRINUSE); } } } if (*lportp != 0) lport = *lportp; if (lport == 0) { error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); if (error != 0) return (error); } *laddrp = laddr.s_addr; *lportp = lport; return (0); } /* * 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 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred, struct mbuf *m, bool rehash) { u_short lport, fport; in_addr_t laddr, faddr; int anonport, error; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); lport = inp->inp_lport; laddr = inp->inp_laddr.s_addr; anonport = (lport == 0); error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, NULL, cred); if (error) return (error); /* Do the initial binding of the local address if required. */ if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { KASSERT(rehash == true, ("Rehashing required for unbound inps")); inp->inp_lport = lport; inp->inp_laddr.s_addr = laddr; if (in_pcbinshash(inp) != 0) { inp->inp_laddr.s_addr = INADDR_ANY; inp->inp_lport = 0; return (EAGAIN); } } /* Commit the remaining changes. */ inp->inp_lport = lport; inp->inp_laddr.s_addr = laddr; inp->inp_faddr.s_addr = faddr; inp->inp_fport = fport; if (rehash) { in_pcbrehash_mbuf(inp, m); } else { in_pcbinshash_mbuf(inp, m); } if (anonport) inp->inp_flags |= INP_ANONPORT; return (0); } int in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true)); } /* * Do proper source address selection on an unbound socket in case * of connect. Take jails into account as well. */ int in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, struct ucred *cred) { struct ifaddr *ifa; struct sockaddr *sa; struct sockaddr_in *sin; struct route sro; int error; NET_EPOCH_ASSERT(); KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); /* * Bypass source address selection and use the primary jail IP * if requested. */ if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) return (0); error = 0; bzero(&sro, sizeof(sro)); sin = (struct sockaddr_in *)&sro.ro_dst; sin->sin_family = AF_INET; sin->sin_len = sizeof(struct sockaddr_in); sin->sin_addr.s_addr = faddr->s_addr; /* * If route is known our src addr is taken from the i/f, * else punt. * * Find out route to destination. */ if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); /* * If we found a route, use the address corresponding to * the outgoing interface. * * Otherwise assume faddr is reachable on a directly connected * network and try to find a corresponding interface to take * the source address from. */ if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { struct in_ifaddr *ia; struct ifnet *ifp; ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, inp->inp_socket->so_fibnum)); if (ia == NULL) { ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, inp->inp_socket->so_fibnum)); } if (ia == NULL) { error = ENETUNREACH; goto done; } if (cred == NULL || !prison_flag(cred, PR_IP4)) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } ifp = ia->ia_ifp; ia = NULL; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } /* * If the outgoing interface on the route found is not * a loopback interface, use the address from that interface. * In case of jails do those three steps: * 1. check if the interface address belongs to the jail. If so use it. * 2. check if we have any address on the outgoing interface * belonging to this jail. If so use it. * 3. as a last resort return the 'default' jail address. */ if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { struct in_ifaddr *ia; struct ifnet *ifp; /* If not jailed, use the default returned. */ if (cred == NULL || !prison_flag(cred, PR_IP4)) { ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* Jailed. */ /* 1. Check if the iface address belongs to the jail. */ sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* * 2. Check if we have any address on the outgoing interface * belonging to this jail. */ ia = NULL; ifp = sro.ro_rt->rt_ifp; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } /* * The outgoing interface is marked with 'loopback net', so a route * to ourselves is here. * Try to find the interface of the destination address and then * take the address from there. That interface is not necessarily * a loopback interface. * In case of jails, check that it is an address of the jail * and if we cannot find, fall back to the 'default' jail address. */ if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { struct sockaddr_in sain; struct in_ifaddr *ia; bzero(&sain, sizeof(struct sockaddr_in)); sain.sin_family = AF_INET; sain.sin_len = sizeof(struct sockaddr_in); sain.sin_addr.s_addr = faddr->s_addr; ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); if (cred == NULL || !prison_flag(cred, PR_IP4)) { if (ia == NULL) { error = ENETUNREACH; goto done; } laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* Jailed. */ if (ia != NULL) { struct ifnet *ifp; ifp = ia->ia_ifp; ia = NULL; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } done: if (sro.ro_rt != NULL) RTFREE(sro.ro_rt); return (error); } /* * Set up for a connect from a socket to the specified address. * On entry, *laddrp and *lportp should contain the current local * address and port for the PCB; these are updated to the values * that should be placed in inp_laddr and inp_lport to complete * the connect. * * On success, *faddrp and *fportp will be set to the remote address * and port. These are not updated in the error case. * * If the operation fails because the connection already exists, * *oinpp will be set to the PCB of that connection so that the * caller can decide to override it. In all other cases, *oinpp * is set to NULL. */ int in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, struct inpcb **oinpp, struct ucred *cred) { struct rm_priotracker in_ifa_tracker; struct sockaddr_in *sin = (struct sockaddr_in *)nam; struct in_ifaddr *ia; struct inpcb *oinp; struct in_addr laddr, faddr; u_short lport, fport; int error; /* * Because a global state change doesn't actually occur here, a read * lock is sufficient. */ NET_EPOCH_ASSERT(); INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); if (oinpp != NULL) *oinpp = NULL; if (nam->sa_len != sizeof (*sin)) return (EINVAL); if (sin->sin_family != AF_INET) return (EAFNOSUPPORT); if (sin->sin_port == 0) return (EADDRNOTAVAIL); laddr.s_addr = *laddrp; lport = *lportp; faddr = sin->sin_addr; fport = sin->sin_port; if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { /* * If the destination address is INADDR_ANY, * use the primary local address. * If the supplied address is INADDR_BROADCAST, * and the primary interface supports broadcast, * choose the broadcast address for that interface. */ if (faddr.s_addr == INADDR_ANY) { IN_IFADDR_RLOCK(&in_ifa_tracker); faddr = IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; IN_IFADDR_RUNLOCK(&in_ifa_tracker); if (cred != NULL && (error = prison_get_ip4(cred, &faddr)) != 0) return (error); } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { IN_IFADDR_RLOCK(&in_ifa_tracker); if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & IFF_BROADCAST) faddr = satosin(&CK_STAILQ_FIRST( &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; IN_IFADDR_RUNLOCK(&in_ifa_tracker); } } if (laddr.s_addr == INADDR_ANY) { error = in_pcbladdr(inp, &faddr, &laddr, cred); /* * If the destination address is multicast and an outgoing * interface has been set as a multicast option, prefer the * address of that interface as our source address. */ if (IN_MULTICAST(ntohl(faddr.s_addr)) && inp->inp_moptions != NULL) { struct ip_moptions *imo; struct ifnet *ifp; imo = inp->inp_moptions; if (imo->imo_multicast_ifp != NULL) { ifp = imo->imo_multicast_ifp; IN_IFADDR_RLOCK(&in_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { if ((ia->ia_ifp == ifp) && (cred == NULL || prison_check_ip4(cred, &ia->ia_addr.sin_addr) == 0)) break; } if (ia == NULL) error = EADDRNOTAVAIL; else { laddr = ia->ia_addr.sin_addr; error = 0; } IN_IFADDR_RUNLOCK(&in_ifa_tracker); } } if (error) return (error); } oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, laddr, lport, 0, NULL); if (oinp != NULL) { if (oinpp != NULL) *oinpp = oinp; return (EADDRINUSE); } if (lport == 0) { error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, cred); if (error) return (error); } *laddrp = laddr.s_addr; *lportp = lport; *faddrp = faddr.s_addr; *fportp = fport; return (0); } void in_pcbdisconnect(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); inp->inp_faddr.s_addr = INADDR_ANY; inp->inp_fport = 0; in_pcbrehash(inp); } #endif /* INET */ /* * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. * For most protocols, this will be invoked immediately prior to calling * in_pcbfree(). However, with TCP the inpcb may significantly outlive the * socket, in which case in_pcbfree() is deferred. */ void in_pcbdetach(struct inpcb *inp) { KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); #ifdef RATELIMIT if (inp->inp_snd_tag != NULL) in_pcbdetach_txrtlmt(inp); #endif inp->inp_socket->so_pcb = NULL; inp->inp_socket = NULL; } /* * in_pcbref() bumps the reference count on an inpcb in order to maintain * stability of an inpcb pointer despite the inpcb lock being released. This * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, * but where the inpcb lock may already held, or when acquiring a reference * via a pcbgroup. * * in_pcbref() should be used only to provide brief memory stability, and * must always be followed by a call to INP_WLOCK() and in_pcbrele() to * garbage collect the inpcb if it has been in_pcbfree()'d from another * context. Until in_pcbrele() has returned that the inpcb is still valid, * lock and rele are the *only* safe operations that may be performed on the * inpcb. * * While the inpcb will not be freed, releasing the inpcb lock means that the * connection's state may change, so the caller should be careful to * revalidate any cached state on reacquiring the lock. Drop the reference * using in_pcbrele(). */ void in_pcbref(struct inpcb *inp) { KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); refcount_acquire(&inp->inp_refcount); } /* * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we * return a flag indicating whether or not the inpcb remains valid. If it is * valid, we return with the inpcb lock held. * * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a * reference on an inpcb. Historically more work was done here (actually, in * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely * about memory stability (and continued use of the write lock). */ int in_pcbrele_rlocked(struct inpcb *inp) { struct inpcbinfo *pcbinfo; KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); INP_RLOCK_ASSERT(inp); if (refcount_release(&inp->inp_refcount) == 0) { /* * If the inpcb has been freed, let the caller know, even if * this isn't the last reference. */ if (inp->inp_flags2 & INP_FREED) { INP_RUNLOCK(inp); return (1); } return (0); } KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); #ifdef TCPHPTS if (inp->inp_in_hpts || inp->inp_in_input) { struct tcp_hpts_entry *hpts; /* * We should not be on the hpts at * this point in any form. we must * get the lock to be sure. */ hpts = tcp_hpts_lock(inp); if (inp->inp_in_hpts) panic("Hpts:%p inp:%p at free still on hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if (inp->inp_in_input) panic("Hpts:%p inp:%p at free still on input hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); } #endif INP_RUNLOCK(inp); pcbinfo = inp->inp_pcbinfo; uma_zfree(pcbinfo->ipi_zone, inp); return (1); } int in_pcbrele_wlocked(struct inpcb *inp) { struct inpcbinfo *pcbinfo; KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); INP_WLOCK_ASSERT(inp); if (refcount_release(&inp->inp_refcount) == 0) { /* * If the inpcb has been freed, let the caller know, even if * this isn't the last reference. */ if (inp->inp_flags2 & INP_FREED) { INP_WUNLOCK(inp); return (1); } return (0); } KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); #ifdef TCPHPTS if (inp->inp_in_hpts || inp->inp_in_input) { struct tcp_hpts_entry *hpts; /* * We should not be on the hpts at * this point in any form. we must * get the lock to be sure. */ hpts = tcp_hpts_lock(inp); if (inp->inp_in_hpts) panic("Hpts:%p inp:%p at free still on hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if (inp->inp_in_input) panic("Hpts:%p inp:%p at free still on input hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); } #endif INP_WUNLOCK(inp); pcbinfo = inp->inp_pcbinfo; uma_zfree(pcbinfo->ipi_zone, inp); return (1); } /* * Temporary wrapper. */ int in_pcbrele(struct inpcb *inp) { return (in_pcbrele_wlocked(inp)); } void in_pcblist_rele_rlocked(epoch_context_t ctx) { struct in_pcblist *il; struct inpcb *inp; struct inpcbinfo *pcbinfo; int i, n; il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); pcbinfo = il->il_pcbinfo; n = il->il_count; INP_INFO_WLOCK(pcbinfo); for (i = 0; i < n; i++) { inp = il->il_inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); free(il, M_TEMP); } static void inpcbport_free(epoch_context_t ctx) { struct inpcbport *phd; phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); free(phd, M_PCB); } static void in_pcbfree_deferred(epoch_context_t ctx) { struct inpcb *inp; int released __unused; inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); INP_WLOCK(inp); CURVNET_SET(inp->inp_vnet); #ifdef INET struct ip_moptions *imo = inp->inp_moptions; inp->inp_moptions = NULL; #endif /* XXXRW: Do as much as possible here. */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (inp->inp_sp != NULL) ipsec_delete_pcbpolicy(inp); #endif #ifdef INET6 struct ip6_moptions *im6o = NULL; if (inp->inp_vflag & INP_IPV6PROTO) { ip6_freepcbopts(inp->in6p_outputopts); im6o = inp->in6p_moptions; inp->in6p_moptions = NULL; } #endif if (inp->inp_options) (void)m_free(inp->inp_options); inp->inp_vflag = 0; crfree(inp->inp_cred); #ifdef MAC mac_inpcb_destroy(inp); #endif released = in_pcbrele_wlocked(inp); MPASS(released); #ifdef INET6 ip6_freemoptions(im6o); #endif #ifdef INET inp_freemoptions(imo); #endif CURVNET_RESTORE(); } /* * Unconditionally schedule an inpcb to be freed by decrementing its * reference count, which should occur only after the inpcb has been detached * from its socket. If another thread holds a temporary reference (acquired * using in_pcbref()) then the free is deferred until that reference is * released using in_pcbrele(), but the inpcb is still unlocked. Almost all * work, including removal from global lists, is done in this context, where * the pcbinfo lock is held. */ void in_pcbfree(struct inpcb *inp) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); KASSERT((inp->inp_flags2 & INP_FREED) == 0, ("%s: called twice for pcb %p", __func__, inp)); if (inp->inp_flags2 & INP_FREED) { INP_WUNLOCK(inp); return; } INP_WLOCK_ASSERT(inp); INP_LIST_WLOCK(pcbinfo); in_pcbremlists(inp); INP_LIST_WUNLOCK(pcbinfo); RO_INVALIDATE_CACHE(&inp->inp_route); /* mark as destruction in progress */ inp->inp_flags2 |= INP_FREED; INP_WUNLOCK(inp); NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx); } /* * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and * port reservation, and preventing it from being returned by inpcb lookups. * * It is used by TCP to mark an inpcb as unused and avoid future packet * delivery or event notification when a socket remains open but TCP has * closed. This might occur as a result of a shutdown()-initiated TCP close * or a RST on the wire, and allows the port binding to be reused while still * maintaining the invariant that so_pcb always points to a valid inpcb until * in_pcbdetach(). * * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by * in_pcbnotifyall() and in_pcbpurgeif0()? */ void in_pcbdrop(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); #ifdef INVARIANTS if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) MPASS(inp->inp_refcount > 1); #endif /* * XXXRW: Possibly we should protect the setting of INP_DROPPED with * the hash lock...? */ inp->inp_flags |= INP_DROPPED; if (inp->inp_flags & INP_INHASHLIST) { struct inpcbport *phd = inp->inp_phd; INP_HASH_WLOCK(inp->inp_pcbinfo); in_pcbremlbgrouphash(inp); CK_LIST_REMOVE(inp, inp_hash); CK_LIST_REMOVE(inp, inp_portlist); if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { CK_LIST_REMOVE(phd, phd_hash); NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); } INP_HASH_WUNLOCK(inp->inp_pcbinfo); inp->inp_flags &= ~INP_INHASHLIST; #ifdef PCBGROUP in_pcbgroup_remove(inp); #endif } } #ifdef INET /* * Common routines to return the socket addresses associated with inpcbs. */ struct sockaddr * in_sockaddr(in_port_t port, struct in_addr *addr_p) { struct sockaddr_in *sin; sin = malloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = *addr_p; sin->sin_port = port; return (struct sockaddr *)sin; } int in_getsockaddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_lport; addr = inp->inp_laddr; INP_RUNLOCK(inp); *nam = in_sockaddr(port, &addr); return 0; } int in_getpeeraddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_fport; addr = inp->inp_faddr; INP_RUNLOCK(inp); *nam = in_sockaddr(port, &addr); return 0; } void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, struct inpcb *(*notify)(struct inpcb *, int)) { struct inpcb *inp, *inp_temp; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { INP_WLOCK(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV4) == 0) { INP_WUNLOCK(inp); continue; } #endif if (inp->inp_faddr.s_addr != faddr.s_addr || inp->inp_socket == NULL) { INP_WUNLOCK(inp); continue; } if ((*notify)(inp, errno)) INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); } void in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) { struct inpcb *inp; struct in_multi *inm; struct in_mfilter *imf; struct ip_moptions *imo; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { INP_WLOCK(inp); imo = inp->inp_moptions; if ((inp->inp_vflag & INP_IPV4) && imo != NULL) { /* * Unselect the outgoing interface if it is being * detached. */ if (imo->imo_multicast_ifp == ifp) imo->imo_multicast_ifp = NULL; /* * Drop multicast group membership if we joined * through the interface being detached. * * XXX This can all be deferred to an epoch_call */ restart: IP_MFILTER_FOREACH(imf, &imo->imo_head) { if ((inm = imf->imf_inm) == NULL) continue; if (inm->inm_ifp != ifp) continue; ip_mfilter_remove(&imo->imo_head, imf); IN_MULTI_LOCK_ASSERT(); in_leavegroup_locked(inm, NULL); ip_mfilter_free(imf); goto restart; } } 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. */ #define INP_LOOKUP_MAPPED_PCB_COST 3 struct inpcb * in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, u_short lport, int lookupflags, struct ucred *cred) { struct inpcb *inp; #ifdef INET6 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; #else int matchwild = 3; #endif int 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(INADDR_ANY, lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == INADDR_ANY && inp->inp_laddr.s_addr == laddr.s_addr && inp->inp_lport == lport) { /* * Found? */ if (cred == NULL || prison_equal_ip4(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_ip4(inp->inp_cred->cr_prison, cred->cr_prison)) continue; #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; /* * We never select the PCB that has * INP_IPV6 flag and is bound to :: if * we have another PCB which is bound * to 0.0.0.0. If a PCB has the * INP_IPV6 flag, then we set its cost * higher than IPv4 only PCBs. * * Note that the case only happens * when a socket is bound to ::, under * the condition that the use of the * mapped address is allowed. */ if ((inp->inp_vflag & INP_IPV6) != 0) wildcard += INP_LOOKUP_MAPPED_PCB_COST; #endif if (inp->inp_faddr.s_addr != INADDR_ANY) wildcard++; if (inp->inp_laddr.s_addr != INADDR_ANY) { if (laddr.s_addr == INADDR_ANY) wildcard++; else if (inp->inp_laddr.s_addr != laddr.s_addr) continue; } else { if (laddr.s_addr != INADDR_ANY) wildcard++; } if (wildcard < matchwild) { match = inp; matchwild = wildcard; if (matchwild == 0) break; } } } return (match); } } #undef INP_LOOKUP_MAPPED_PCB_COST static struct inpcb * in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, const struct in_addr *laddr, uint16_t lport, const struct in_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 group does not contain IPv4 mapped INET6 wild sockets */ local_wild = NULL; CK_LIST_FOREACH(grp, hdr, il_list) { #ifdef INET6 if (!(grp->il_vflag & INP_IPV4)) continue; #endif if (grp->il_lport != lport) continue; idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % grp->il_inpcnt; if (grp->il_laddr.s_addr == laddr->s_addr) return (grp->il_inp[idx]); if (grp->il_laddr.s_addr == INADDR_ANY && (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 * in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, struct in_addr faddr, u_int fport_arg, struct in_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(faddr.s_addr, lport, fport, pcbgroup->ipg_hashmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == faddr.s_addr && inp->inp_laddr.s_addr == laddr.s_addr && 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_IP4)) goto found; if (tmpinp == NULL) tmpinp = inp; } } if (tmpinp != NULL) { inp = tmpinp; goto found; } #ifdef RSS /* * For incoming connections, we may wish to do a wildcard * match for an RSS-local socket. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif struct inpcb *jail_wild = NULL; struct inpcbhead *head; 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) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) goto found; else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; #ifdef INET6 if (inp == NULL) inp = local_wild_mapped; #endif if (inp != NULL) goto found; } #endif /* * Then look for a wildcard match, if requested. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif struct inpcb *jail_wild = NULL; struct inpcbhead *head; 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(INADDR_ANY, lport, 0, pcbinfo->ipi_wildmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) goto found; else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; #ifdef INET6 if (inp == NULL) inp = local_wild_mapped; #endif if (inp != NULL) goto found; } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 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 bug", __func__); if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); return (NULL); } else 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, using pcbinfo tables. This variation assumes * that the caller has locked the hash list, and will not perform any further * locking or reference operations on either the hash list or the connection. */ static struct inpcb * in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport_arg, struct in_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(faddr.s_addr, lport, fport, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == faddr.s_addr && inp->inp_laddr.s_addr == laddr.s_addr && 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_IP4)) 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 = in_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; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif 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(INADDR_ANY, lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) return (inp); else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif 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); #ifdef INET6 if (local_wild_mapped != NULL) return (local_wild_mapped); #endif } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 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 * in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) { struct inpcb *inp; inp = in_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 in_pcbgroup.c. */ struct inpcb * in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_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 = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); } #endif return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } struct inpcb * in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_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))) { pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), m->m_pkthdr.flowid); if (pcbgroup != NULL) return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #ifndef RSS pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #endif } #endif return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } #endif /* INET */ /* * Insert PCB onto various hash lists. */ static int in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m) { struct inpcbhead *pcbhash; struct inpcbporthead *pcbporthash; struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct inpcbport *phd; u_int32_t hashkey_faddr; int so_options; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, ("in_pcbinshash: INP_INHASHLIST")); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); else #endif hashkey_faddr = inp->inp_faddr.s_addr; pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; pcbporthash = &pcbinfo->ipi_porthashbase[ INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; /* * Add entry to load balance group. * Only do this if SO_REUSEPORT_LB is set. */ so_options = inp_so_options(inp); if (so_options & SO_REUSEPORT_LB) { int ret = in_pcbinslbgrouphash(inp); if (ret) { /* pcb lb group malloc fail (ret=ENOBUFS). */ return (ret); } } /* * Go through port list and look for a head for this lport. */ CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { if (phd->phd_port == inp->inp_lport) break; } /* * If none exists, malloc one and tack it on. */ if (phd == NULL) { phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); if (phd == NULL) { return (ENOBUFS); /* XXX */ } bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); phd->phd_port = inp->inp_lport; CK_LIST_INIT(&phd->phd_pcblist); CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); } inp->inp_phd = phd; CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); inp->inp_flags |= INP_INHASHLIST; #ifdef PCBGROUP if (m != NULL) { in_pcbgroup_update_mbuf(inp, m); } else { in_pcbgroup_update(inp); } #endif return (0); } int in_pcbinshash(struct inpcb *inp) { return (in_pcbinshash_internal(inp, NULL)); } int in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m) { return (in_pcbinshash_internal(inp, m)); } /* * Move PCB to the proper hash bucket when { faddr, fport } have been * changed. NOTE: This does not handle the case of the lport changing (the * hashed port list would have to be updated as well), so the lport must * not change after in_pcbinshash() has been called. */ void in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct inpcbhead *head; u_int32_t hashkey_faddr; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); KASSERT(inp->inp_flags & INP_INHASHLIST, ("in_pcbrehash: !INP_INHASHLIST")); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); else #endif hashkey_faddr = inp->inp_faddr.s_addr; head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; CK_LIST_REMOVE(inp, inp_hash); CK_LIST_INSERT_HEAD(head, inp, inp_hash); #ifdef PCBGROUP if (m != NULL) in_pcbgroup_update_mbuf(inp, m); else in_pcbgroup_update(inp); #endif } void in_pcbrehash(struct inpcb *inp) { in_pcbrehash_mbuf(inp, NULL); } /* * Remove PCB from various lists. */ static void in_pcbremlists(struct inpcb *inp) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_LIST_WLOCK_ASSERT(pcbinfo); inp->inp_gencnt = ++pcbinfo->ipi_gencnt; if (inp->inp_flags & INP_INHASHLIST) { struct inpcbport *phd = inp->inp_phd; INP_HASH_WLOCK(pcbinfo); /* XXX: Only do if SO_REUSEPORT_LB set? */ in_pcbremlbgrouphash(inp); CK_LIST_REMOVE(inp, inp_hash); CK_LIST_REMOVE(inp, inp_portlist); if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { CK_LIST_REMOVE(phd, phd_hash); NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); } INP_HASH_WUNLOCK(pcbinfo); inp->inp_flags &= ~INP_INHASHLIST; } CK_LIST_REMOVE(inp, inp_list); pcbinfo->ipi_count--; #ifdef PCBGROUP in_pcbgroup_remove(inp); #endif } /* * 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 in_losing(struct inpcb *inp) { RO_INVALIDATE_CACHE(&inp->inp_route); return; } /* * A set label operation has occurred at the socket layer, propagate the * label change into the in_pcb for the socket. */ void in_pcbsosetlabel(struct socket *so) { #ifdef MAC struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); INP_WLOCK(inp); SOCK_LOCK(so); mac_inpcb_sosetlabel(so, inp); SOCK_UNLOCK(so); INP_WUNLOCK(inp); #endif } /* * ipport_tick runs once per second, determining if random port allocation * should be continued. If more than ipport_randomcps ports have been * allocated in the last second, then we return to sequential port * allocation. We return to random allocation only once we drop below * ipport_randomcps for at least ipport_randomtime seconds. */ static void ipport_tick(void *xtp) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ if (V_ipport_tcpallocs <= V_ipport_tcplastcount + V_ipport_randomcps) { if (V_ipport_stoprandom > 0) V_ipport_stoprandom--; } else V_ipport_stoprandom = V_ipport_randomtime; V_ipport_tcplastcount = V_ipport_tcpallocs; CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); } static void ip_fini(void *xtp) { callout_stop(&ipport_tick_callout); } /* * The ipport_callout should start running at about the time we attach the * inet or inet6 domains. */ static void ipport_tick_init(const void *unused __unused) { /* Start ipport_tick. */ callout_init(&ipport_tick_callout, 1); callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, SHUTDOWN_PRI_DEFAULT); } SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, ipport_tick_init, NULL); void inp_wlock(struct inpcb *inp) { INP_WLOCK(inp); } void inp_wunlock(struct inpcb *inp) { INP_WUNLOCK(inp); } void inp_rlock(struct inpcb *inp) { INP_RLOCK(inp); } void inp_runlock(struct inpcb *inp) { INP_RUNLOCK(inp); } #ifdef INVARIANT_SUPPORT void inp_lock_assert(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); } void inp_unlock_assert(struct inpcb *inp) { INP_UNLOCK_ASSERT(inp); } #endif void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) { struct inpcb *inp; INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inp); func(inp, arg); INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); } struct socket * inp_inpcbtosocket(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); return (inp->inp_socket); } struct tcpcb * inp_inpcbtotcpcb(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); return ((struct tcpcb *)inp->inp_ppcb); } int inp_ip_tos_get(const struct inpcb *inp) { return (inp->inp_ip_tos); } void inp_ip_tos_set(struct inpcb *inp, int val) { inp->inp_ip_tos = val; } void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, uint32_t *faddr, uint16_t *fp) { INP_LOCK_ASSERT(inp); *laddr = inp->inp_laddr.s_addr; *faddr = inp->inp_faddr.s_addr; *lp = inp->inp_lport; *fp = inp->inp_fport; } struct inpcb * so_sotoinpcb(struct socket *so) { return (sotoinpcb(so)); } struct tcpcb * so_sototcpcb(struct socket *so) { return (sototcpcb(so)); } /* * Create an external-format (``xinpcb'') structure using the information in * the kernel-format in_pcb structure pointed to by inp. 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 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) { bzero(xi, sizeof(*xi)); xi->xi_len = sizeof(struct xinpcb); if (inp->inp_socket) sotoxsocket(inp->inp_socket, &xi->xi_socket); bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); xi->inp_gencnt = inp->inp_gencnt; xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; xi->inp_flow = inp->inp_flow; xi->inp_flowid = inp->inp_flowid; xi->inp_flowtype = inp->inp_flowtype; xi->inp_flags = inp->inp_flags; xi->inp_flags2 = inp->inp_flags2; xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; xi->in6p_cksum = inp->in6p_cksum; xi->in6p_hops = inp->in6p_hops; xi->inp_ip_tos = inp->inp_ip_tos; xi->inp_vflag = inp->inp_vflag; xi->inp_ip_ttl = inp->inp_ip_ttl; xi->inp_ip_p = inp->inp_ip_p; xi->inp_ip_minttl = inp->inp_ip_minttl; } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) { char faddr_str[48], laddr_str[48]; db_print_indent(indent); db_printf("%s at %p\n", name, inc); indent += 2; #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { /* IPv6. */ ip6_sprintf(laddr_str, &inc->inc6_laddr); ip6_sprintf(faddr_str, &inc->inc6_faddr); } else #endif { /* IPv4. */ inet_ntoa_r(inc->inc_laddr, laddr_str); inet_ntoa_r(inc->inc_faddr, faddr_str); } db_print_indent(indent); db_printf("inc_laddr %s inc_lport %u\n", laddr_str, ntohs(inc->inc_lport)); db_print_indent(indent); db_printf("inc_faddr %s inc_fport %u\n", faddr_str, ntohs(inc->inc_fport)); } static void db_print_inpflags(int inp_flags) { int comma; comma = 0; if (inp_flags & INP_RECVOPTS) { db_printf("%sINP_RECVOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVRETOPTS) { db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVDSTADDR) { db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ORIGDSTADDR) { db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_HDRINCL) { db_printf("%sINP_HDRINCL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_HIGHPORT) { db_printf("%sINP_HIGHPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_LOWPORT) { db_printf("%sINP_LOWPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ANONPORT) { db_printf("%sINP_ANONPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVIF) { db_printf("%sINP_RECVIF", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_MTUDISC) { db_printf("%sINP_MTUDISC", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVTTL) { db_printf("%sINP_RECVTTL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_DONTFRAG) { db_printf("%sINP_DONTFRAG", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVTOS) { db_printf("%sINP_RECVTOS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_IPV6_V6ONLY) { db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_PKTINFO) { db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_HOPLIMIT) { db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_HOPOPTS) { db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_DSTOPTS) { db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RTHDR) { db_printf("%sIN6P_RTHDR", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RTHDRDSTOPTS) { db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_TCLASS) { db_printf("%sIN6P_TCLASS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_AUTOFLOWLABEL) { db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_TIMEWAIT) { db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ONESBCAST) { db_printf("%sINP_ONESBCAST", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_DROPPED) { db_printf("%sINP_DROPPED", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_SOCKREF) { db_printf("%sINP_SOCKREF", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RFC2292) { db_printf("%sIN6P_RFC2292", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_MTU) { db_printf("IN6P_MTU%s", comma ? ", " : ""); comma = 1; } } static void db_print_inpvflag(u_char inp_vflag) { int comma; comma = 0; if (inp_vflag & INP_IPV4) { db_printf("%sINP_IPV4", comma ? ", " : ""); comma = 1; } if (inp_vflag & INP_IPV6) { db_printf("%sINP_IPV6", comma ? ", " : ""); comma = 1; } if (inp_vflag & INP_IPV6PROTO) { db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); comma = 1; } } static void db_print_inpcb(struct inpcb *inp, const char *name, int indent) { db_print_indent(indent); db_printf("%s at %p\n", name, inp); indent += 2; db_print_indent(indent); db_printf("inp_flow: 0x%x\n", inp->inp_flow); db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); db_print_indent(indent); db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); db_print_indent(indent); db_printf("inp_label: %p inp_flags: 0x%x (", inp->inp_label, inp->inp_flags); db_print_inpflags(inp->inp_flags); db_printf(")\n"); db_print_indent(indent); db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, inp->inp_vflag); db_print_inpvflag(inp->inp_vflag); db_printf(")\n"); db_print_indent(indent); db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); db_print_indent(indent); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { db_printf("in6p_options: %p in6p_outputopts: %p " "in6p_moptions: %p\n", inp->in6p_options, inp->in6p_outputopts, inp->in6p_moptions); db_printf("in6p_icmp6filt: %p in6p_cksum %d " "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, inp->in6p_hops); } else #endif { db_printf("inp_ip_tos: %d inp_ip_options: %p " "inp_ip_moptions: %p\n", inp->inp_ip_tos, inp->inp_options, inp->inp_moptions); } db_print_indent(indent); db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, (uintmax_t)inp->inp_gencnt); } DB_SHOW_COMMAND(inpcb, db_show_inpcb) { struct inpcb *inp; if (!have_addr) { db_printf("usage: show inpcb \n"); return; } inp = (struct inpcb *)addr; db_print_inpcb(inp, "inpcb", 0); } #endif /* DDB */ #ifdef RATELIMIT /* * Modify TX rate limit based on the existing "inp->inp_snd_tag", * if any. */ int in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) { union if_snd_tag_modify_params params = { .rate_limit.max_rate = max_pacing_rate, .rate_limit.flags = M_NOWAIT, }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_modify == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_modify(mst, ¶ms); } return (error); } /* * Query existing TX rate limit based on the existing * "inp->inp_snd_tag", if any. */ int in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) { union if_snd_tag_query_params params = { }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_query == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_query(mst, ¶ms); if (error == 0 && p_max_pacing_rate != NULL) *p_max_pacing_rate = params.rate_limit.max_rate; } return (error); } /* * Query existing TX queue level based on the existing * "inp->inp_snd_tag", if any. */ int in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) { union if_snd_tag_query_params params = { }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_query == NULL) return (EOPNOTSUPP); error = ifp->if_snd_tag_query(mst, ¶ms); if (error == 0 && p_txqueue_level != NULL) *p_txqueue_level = params.rate_limit.queue_level; return (error); } /* * Allocate a new TX rate limit send tag from the network interface * given by the "ifp" argument and save it in "inp->inp_snd_tag": */ int in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) { union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = (max_pacing_rate == -1U) ? IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = flowid, .rate_limit.hdr.flowtype = flowtype, + .rate_limit.hdr.numa_domain = inp->inp_numa_domain, .rate_limit.max_rate = max_pacing_rate, .rate_limit.flags = M_NOWAIT, }; int error; INP_WLOCK_ASSERT(inp); if (*st != NULL) return (EINVAL); if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); #ifdef INET if (error == 0) { counter_u64_add(rate_limit_set_ok, 1); counter_u64_add(rate_limit_active, 1); } else counter_u64_add(rate_limit_alloc_fail, 1); #endif } return (error); } void in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst) { if (ifp == NULL) return; /* * If the device was detached while we still had reference(s) * on the ifp, we assume if_snd_tag_free() was replaced with * stubs. */ ifp->if_snd_tag_free(mst); /* release reference count on network interface */ if_rele(ifp); #ifdef INET counter_u64_add(rate_limit_active, -1); #endif } /* * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", * if any: */ void in_pcbdetach_txrtlmt(struct inpcb *inp) { struct m_snd_tag *mst; INP_WLOCK_ASSERT(inp); mst = inp->inp_snd_tag; inp->inp_snd_tag = NULL; if (mst == NULL) return; m_snd_tag_rele(mst); } int in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) { int error; /* * If the existing send tag is for the wrong interface due to * a route change, first drop the existing tag. Set the * CHANGED flag so that we will keep trying to allocate a new * tag if we fail to allocate one this time. */ if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { in_pcbdetach_txrtlmt(inp); inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; } /* * NOTE: When attaching to a network interface a reference is * made to ensure the network interface doesn't go away until * all ratelimit connections are gone. The network interface * pointers compared below represent valid network interfaces, * except when comparing towards NULL. */ if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { error = 0; } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { if (inp->inp_snd_tag != NULL) in_pcbdetach_txrtlmt(inp); error = 0; } else if (inp->inp_snd_tag == NULL) { /* * In order to utilize packet pacing with RSS, we need * to wait until there is a valid RSS hash before we * can proceed: */ if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { error = EAGAIN; } else { error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); } } else { error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); } if (error == 0 || error == EOPNOTSUPP) inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; return (error); } /* * This function should be called when the INP_RATE_LIMIT_CHANGED flag * is set in the fast path and will attach/detach/modify the TX rate * limit send tag based on the socket's so_max_pacing_rate value. */ void in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) { struct socket *socket; uint32_t max_pacing_rate; bool did_upgrade; int error; if (inp == NULL) return; socket = inp->inp_socket; if (socket == NULL) return; if (!INP_WLOCKED(inp)) { /* * NOTE: If the write locking fails, we need to bail * out and use the non-ratelimited ring for the * transmit until there is a new chance to get the * write lock. */ if (!INP_TRY_UPGRADE(inp)) return; did_upgrade = 1; } else { did_upgrade = 0; } /* * NOTE: The so_max_pacing_rate value is read unlocked, * because atomic updates are not required since the variable * is checked at every mbuf we send. It is assumed that the * variable read itself will be atomic. */ max_pacing_rate = socket->so_max_pacing_rate; error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); if (did_upgrade) INP_DOWNGRADE(inp); } /* * Track route changes for TX rate limiting. */ void in_pcboutput_eagain(struct inpcb *inp) { bool did_upgrade; if (inp == NULL) return; if (inp->inp_snd_tag == NULL) return; if (!INP_WLOCKED(inp)) { /* * NOTE: If the write locking fails, we need to bail * out and use the non-ratelimited ring for the * transmit until there is a new chance to get the * write lock. */ if (!INP_TRY_UPGRADE(inp)) return; did_upgrade = 1; } else { did_upgrade = 0; } /* detach rate limiting */ in_pcbdetach_txrtlmt(inp); /* make sure new mbuf send tag allocation is made */ inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; if (did_upgrade) INP_DOWNGRADE(inp); } #ifdef INET static void rl_init(void *st) { rate_limit_active = counter_u64_alloc(M_WAITOK); rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); rate_limit_set_ok = counter_u64_alloc(M_WAITOK); } SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); #endif #endif /* RATELIMIT */ Index: head/sys/netinet/tcp_ratelimit.c =================================================================== --- head/sys/netinet/tcp_ratelimit.c (revision 358807) +++ head/sys/netinet/tcp_ratelimit.c (revision 358808) @@ -1,1502 +1,1503 @@ /*- * * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2018-2019 * 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 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #ifdef INET6 #include #endif #include #ifndef USECS_IN_SECOND #define USECS_IN_SECOND 1000000 #endif /* * For the purposes of each send, what is the size * of an ethernet frame. */ MALLOC_DEFINE(M_TCPPACE, "tcp_hwpace", "TCP Hardware pacing memory"); #ifdef RATELIMIT /* * The following preferred table will seem weird to * the casual viewer. Why do we not have any rates below * 1Mbps? Why do we have a rate at 1.44Mbps called common? * Why do the rates cluster in the 1-100Mbps range more * than others? Why does the table jump around at the beginnign * and then be more consistently raising? * * Let me try to answer those questions. A lot of * this is dependant on the hardware. We have three basic * supporters of rate limiting * * Chelsio - Supporting 16 configurable rates. * Mlx - c4 supporting 13 fixed rates. * Mlx - c5 & c6 supporting 127 configurable rates. * * The c4 is why we have a common rate that is available * in all rate tables. This is a selected rate from the * c4 table and we assure its available in all ratelimit * tables. This way the tcp_ratelimit code has an assured * rate it should always be able to get. This answers a * couple of the questions above. * * So what about the rest, well the table is built to * try to get the most out of a joint hardware/software * pacing system. The software pacer will always pick * a rate higher than the b/w that it is estimating * * on the path. This is done for two reasons. * a) So we can discover more b/w * and * b) So we can send a block of MSS's down and then * have the software timer go off after the previous * send is completely out of the hardware. * * But when we do we don't want to have the delay * between the last packet sent by the hardware be * excessively long (to reach our desired rate). * * So let me give an example for clarity. * * Lets assume that the tcp stack sees that 29,110,000 bps is * what the bw of the path is. The stack would select the * rate 31Mbps. 31Mbps means that each send that is done * by the hardware will cause a 390 micro-second gap between * the packets sent at that rate. For 29,110,000 bps we * would need 416 micro-seconds gap between each send. * * Note that are calculating a complete time for pacing * which includes the ethernet, IP and TCP overhead. So * a full 1514 bytes is used for the above calculations. * My testing has shown that both cards are also using this * as their basis i.e. full payload size of the ethernet frame. * The TCP stack caller needs to be aware of this and make the * appropriate overhead calculations be included in its choices. * * Now, continuing our example, we pick a MSS size based on the * delta between the two rates (416 - 390) divided into the rate * we really wish to send at rounded up. That results in a MSS * send of 17 mss's at once. The hardware then will * run out of data in a single 17MSS send in 6,630 micro-seconds. * * On the other hand the software pacer will send more data * in 7,072 micro-seconds. This means that we will refill * the hardware 52 microseconds after it would have sent * next if it had not ran out of data. This is a win since we are * only sending every 7ms or so and yet all the packets are spaced on * the wire with 94% of what they should be and only * the last packet is delayed extra to make up for the * difference. * * Note that the above formula has two important caveat. * If we are above (b/w wise) over 100Mbps we double the result * of the MSS calculation. The second caveat is if we are 500Mbps * or more we just send the maximum MSS at once i.e. 45MSS. At * the higher b/w's even the cards have limits to what times (timer granularity) * they can insert between packets and start to send more than one * packet at a time on the wire. * */ #define COMMON_RATE 180500 const uint64_t desired_rates[] = { 122500, /* 1Mbps - rate 1 */ 180500, /* 1.44Mpbs - rate 2 common rate */ 375000, /* 3Mbps - rate 3 */ 625000, /* 5Mbps - rate 4 */ 875000, /* 7Mbps - rate 5 */ 1125000, /* 9Mbps - rate 6 */ 1375000, /* 11Mbps - rate 7 */ 1625000, /* 13Mbps - rate 8 */ 2625000, /* 21Mbps - rate 9 */ 3875000, /* 31Mbps - rate 10 */ 5125000, /* 41Meg - rate 11 */ 12500000, /* 100Mbps - rate 12 */ 25000000, /* 200Mbps - rate 13 */ 50000000, /* 400Mbps - rate 14 */ 63750000, /* 51Mbps - rate 15 */ 100000000, /* 800Mbps - rate 16 */ 1875000, /* 15Mbps - rate 17 */ 2125000, /* 17Mbps - rate 18 */ 2375000, /* 19Mbps - rate 19 */ 2875000, /* 23Mbps - rate 20 */ 3125000, /* 25Mbps - rate 21 */ 3375000, /* 27Mbps - rate 22 */ 3625000, /* 29Mbps - rate 23 */ 4125000, /* 33Mbps - rate 24 */ 4375000, /* 35Mbps - rate 25 */ 4625000, /* 37Mbps - rate 26 */ 4875000, /* 39Mbps - rate 27 */ 5375000, /* 43Mbps - rate 28 */ 5625000, /* 45Mbps - rate 29 */ 5875000, /* 47Mbps - rate 30 */ 6125000, /* 49Mbps - rate 31 */ 6625000, /* 53Mbps - rate 32 */ 6875000, /* 55Mbps - rate 33 */ 7125000, /* 57Mbps - rate 34 */ 7375000, /* 59Mbps - rate 35 */ 7625000, /* 61Mbps - rate 36 */ 7875000, /* 63Mbps - rate 37 */ 8125000, /* 65Mbps - rate 38 */ 8375000, /* 67Mbps - rate 39 */ 8625000, /* 69Mbps - rate 40 */ 8875000, /* 71Mbps - rate 41 */ 9125000, /* 73Mbps - rate 42 */ 9375000, /* 75Mbps - rate 43 */ 9625000, /* 77Mbps - rate 44 */ 9875000, /* 79Mbps - rate 45 */ 10125000, /* 81Mbps - rate 46 */ 10375000, /* 83Mbps - rate 47 */ 10625000, /* 85Mbps - rate 48 */ 10875000, /* 87Mbps - rate 49 */ 11125000, /* 89Mbps - rate 50 */ 11375000, /* 91Mbps - rate 51 */ 11625000, /* 93Mbps - rate 52 */ 11875000, /* 95Mbps - rate 53 */ 13125000, /* 105Mbps - rate 54 */ 13750000, /* 110Mbps - rate 55 */ 14375000, /* 115Mbps - rate 56 */ 15000000, /* 120Mbps - rate 57 */ 15625000, /* 125Mbps - rate 58 */ 16250000, /* 130Mbps - rate 59 */ 16875000, /* 135Mbps - rate 60 */ 17500000, /* 140Mbps - rate 61 */ 18125000, /* 145Mbps - rate 62 */ 18750000, /* 150Mbps - rate 64 */ 20000000, /* 160Mbps - rate 65 */ 21250000, /* 170Mbps - rate 66 */ 22500000, /* 180Mbps - rate 67 */ 23750000, /* 190Mbps - rate 68 */ 26250000, /* 210Mbps - rate 69 */ 27500000, /* 220Mbps - rate 70 */ 28750000, /* 230Mbps - rate 71 */ 30000000, /* 240Mbps - rate 72 */ 31250000, /* 250Mbps - rate 73 */ 34375000, /* 275Mbps - rate 74 */ 37500000, /* 300Mbps - rate 75 */ 40625000, /* 325Mbps - rate 76 */ 43750000, /* 350Mbps - rate 77 */ 46875000, /* 375Mbps - rate 78 */ 53125000, /* 425Mbps - rate 79 */ 56250000, /* 450Mbps - rate 80 */ 59375000, /* 475Mbps - rate 81 */ 62500000, /* 500Mbps - rate 82 */ 68750000, /* 550Mbps - rate 83 */ 75000000, /* 600Mbps - rate 84 */ 81250000, /* 650Mbps - rate 85 */ 87500000, /* 700Mbps - rate 86 */ 93750000, /* 750Mbps - rate 87 */ 106250000, /* 850Mbps - rate 88 */ 112500000, /* 900Mbps - rate 89 */ 125000000, /* 1Gbps - rate 90 */ 156250000, /* 1.25Gps - rate 91 */ 187500000, /* 1.5Gps - rate 92 */ 218750000, /* 1.75Gps - rate 93 */ 250000000, /* 2Gbps - rate 94 */ 281250000, /* 2.25Gps - rate 95 */ 312500000, /* 2.5Gbps - rate 96 */ 343750000, /* 2.75Gbps - rate 97 */ 375000000, /* 3Gbps - rate 98 */ 500000000, /* 4Gbps - rate 99 */ 625000000, /* 5Gbps - rate 100 */ 750000000, /* 6Gbps - rate 101 */ 875000000, /* 7Gbps - rate 102 */ 1000000000, /* 8Gbps - rate 103 */ 1125000000, /* 9Gbps - rate 104 */ 1250000000, /* 10Gbps - rate 105 */ 1875000000, /* 15Gbps - rate 106 */ 2500000000 /* 20Gbps - rate 107 */ }; #define MAX_HDWR_RATES (sizeof(desired_rates)/sizeof(uint64_t)) #define RS_ORDERED_COUNT 16 /* * Number that are in order * at the beginning of the table, * over this a sort is required. */ #define RS_NEXT_ORDER_GROUP 16 /* * The point in our table where * we come fill in a second ordered * group (index wise means -1). */ #define ALL_HARDWARE_RATES 1004 /* * 1Meg - 1Gig in 1 Meg steps * plus 100, 200k and 500k and * 10Gig */ #define RS_ONE_MEGABIT_PERSEC 1000000 #define RS_ONE_GIGABIT_PERSEC 1000000000 #define RS_TEN_GIGABIT_PERSEC 10000000000 static struct head_tcp_rate_set int_rs; static struct mtx rs_mtx; uint32_t rs_number_alive; uint32_t rs_number_dead; SYSCTL_NODE(_net_inet_tcp, OID_AUTO, rl, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP Ratelimit stats"); SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, alive, CTLFLAG_RW, &rs_number_alive, 0, "Number of interfaces initialized for ratelimiting"); SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, dead, CTLFLAG_RW, &rs_number_dead, 0, "Number of interfaces departing from ratelimiting"); static void rl_add_syctl_entries(struct sysctl_oid *rl_sysctl_root, struct tcp_rate_set *rs) { /* * Add sysctl entries for thus interface. */ if (rs->rs_flags & RS_INTF_NO_SUP) { SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "disable", CTLFLAG_RD, &rs->rs_disable, 0, "Disable this interface from new hdwr limiting?"); } else { SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "disable", CTLFLAG_RW, &rs->rs_disable, 0, "Disable this interface from new hdwr limiting?"); } SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "minseg", CTLFLAG_RW, &rs->rs_min_seg, 0, "What is the minimum we need to send on this interface?"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flow_limit", CTLFLAG_RW, &rs->rs_flow_limit, 0, "What is the limit for number of flows (0=unlimited)?"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "highest", CTLFLAG_RD, &rs->rs_highest_valid, 0, "Highest valid rate"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "lowest", CTLFLAG_RD, &rs->rs_lowest_valid, 0, "Lowest valid rate"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flags", CTLFLAG_RD, &rs->rs_flags, 0, "What lags are on the entry?"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "numrates", CTLFLAG_RD, &rs->rs_rate_cnt, 0, "How many rates re there?"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flows_using", CTLFLAG_RD, &rs->rs_flows_using, 0, "How many flows are using this interface now?"); #ifdef DETAILED_RATELIMIT_SYSCTL if (rs->rs_rlt && rs->rs_rate_cnt > 0) { /* Lets display the rates */ int i; struct sysctl_oid *rl_rates; struct sysctl_oid *rl_rate_num; char rate_num[16]; rl_rates = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "rate", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Ratelist"); for( i = 0; i < rs->rs_rate_cnt; i++) { sprintf(rate_num, "%d", i); rl_rate_num = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rates), OID_AUTO, rate_num, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Individual Rate"); SYSCTL_ADD_U32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "flags", CTLFLAG_RD, &rs->rs_rlt[i].flags, 0, "Flags on this rate"); SYSCTL_ADD_U32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "pacetime", CTLFLAG_RD, &rs->rs_rlt[i].time_between, 0, "Time hardware inserts between 1500 byte sends"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "rate", CTLFLAG_RD, &rs->rs_rlt[i].rate, 0, "Rate in bytes per second"); } } #endif } static void rs_destroy(epoch_context_t ctx) { struct tcp_rate_set *rs; bool do_free_rs; rs = __containerof(ctx, struct tcp_rate_set, rs_epoch_ctx); mtx_lock(&rs_mtx); rs->rs_flags &= ~RS_FUNERAL_SCHD; /* * In theory its possible (but unlikely) * that while the delete was occuring * and we were applying the DEAD flag * someone slipped in and found the * interface in a lookup. While we * decided rs_flows_using were 0 and * scheduling the epoch_call, the other * thread incremented rs_flow_using. This * is because users have a pointer and * we only use the rs_flows_using in an * atomic fashion, i.e. the other entities * are not protected. To assure this did * not occur, we check rs_flows_using here * before deleting. */ do_free_rs = (rs->rs_flows_using == 0); rs_number_dead--; mtx_unlock(&rs_mtx); if (do_free_rs) { sysctl_ctx_free(&rs->sysctl_ctx); free(rs->rs_rlt, M_TCPPACE); free(rs, M_TCPPACE); } } static void rs_defer_destroy(struct tcp_rate_set *rs) { mtx_assert(&rs_mtx, MA_OWNED); /* Check if already pending. */ if (rs->rs_flags & RS_FUNERAL_SCHD) return; rs_number_dead++; /* Set flag to only defer once. */ rs->rs_flags |= RS_FUNERAL_SCHD; NET_EPOCH_CALL(rs_destroy, &rs->rs_epoch_ctx); } #ifdef INET extern counter_u64_t rate_limit_set_ok; extern counter_u64_t rate_limit_active; extern counter_u64_t rate_limit_alloc_fail; #endif static int rl_attach_txrtlmt(struct ifnet *ifp, uint32_t flowtype, int flowid, uint64_t cfg_rate, struct m_snd_tag **tag) { int error; union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = flowid, .rate_limit.hdr.flowtype = flowtype, .rate_limit.max_rate = cfg_rate, .rate_limit.flags = M_NOWAIT, }; if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_alloc(ifp, ¶ms, tag); #ifdef INET if (error == 0) { if_ref((*tag)->ifp); counter_u64_add(rate_limit_set_ok, 1); counter_u64_add(rate_limit_active, 1); } else counter_u64_add(rate_limit_alloc_fail, 1); #endif } return (error); } static void populate_canned_table(struct tcp_rate_set *rs, const uint64_t *rate_table_act) { /* * The internal table is "special", it * is two seperate ordered tables that * must be merged. We get here when the * adapter specifies a number of rates that * covers both ranges in the table in some * form. */ int i, at_low, at_high; uint8_t low_disabled = 0, high_disabled = 0; for(i = 0, at_low = 0, at_high = RS_NEXT_ORDER_GROUP; i < rs->rs_rate_cnt; i++) { rs->rs_rlt[i].flags = 0; rs->rs_rlt[i].time_between = 0; if ((low_disabled == 0) && (high_disabled || (rate_table_act[at_low] < rate_table_act[at_high]))) { rs->rs_rlt[i].rate = rate_table_act[at_low]; at_low++; if (at_low == RS_NEXT_ORDER_GROUP) low_disabled = 1; } else if (high_disabled == 0) { rs->rs_rlt[i].rate = rate_table_act[at_high]; at_high++; if (at_high == MAX_HDWR_RATES) high_disabled = 1; } } } static struct tcp_rate_set * rt_setup_new_rs(struct ifnet *ifp, int *error) { struct tcp_rate_set *rs; const uint64_t *rate_table_act; uint64_t lentim, res; size_t sz; uint32_t hash_type; int i; struct if_ratelimit_query_results rl; struct sysctl_oid *rl_sysctl_root; /* * We expect to enter with the * mutex locked. */ if (ifp->if_ratelimit_query == NULL) { /* * We can do nothing if we cannot * get a query back from the driver. */ printf("Warning:No query functions for %s:%d-- failed\n", ifp->if_dname, ifp->if_dunit); return (NULL); } rs = malloc(sizeof(struct tcp_rate_set), M_TCPPACE, M_NOWAIT | M_ZERO); if (rs == NULL) { if (error) *error = ENOMEM; printf("Warning:No memory for malloc of tcp_rate_set\n"); return (NULL); } memset(&rl, 0, sizeof(rl)); rl.flags = RT_NOSUPPORT; ifp->if_ratelimit_query(ifp, &rl); if (rl.flags & RT_IS_UNUSABLE) { /* * The interface does not really support * the rate-limiting. */ memset(rs, 0, sizeof(struct tcp_rate_set)); rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_flags = RS_INTF_NO_SUP; rs->rs_disable = 1; rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } else if ((rl.flags & RT_IS_INDIRECT) == RT_IS_INDIRECT) { memset(rs, 0, sizeof(struct tcp_rate_set)); rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_flags = RS_IS_DEFF; rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } else if ((rl.flags & RT_IS_FIXED_TABLE) == RT_IS_FIXED_TABLE) { /* Mellanox C4 likely */ rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_rate_cnt = rl.number_of_rates; rs->rs_min_seg = rl.min_segment_burst; rs->rs_highest_valid = 0; rs->rs_flow_limit = rl.max_flows; rs->rs_flags = RS_IS_INTF | RS_NO_PRE; rs->rs_disable = 0; rate_table_act = rl.rate_table; } else if ((rl.flags & RT_IS_SELECTABLE) == RT_IS_SELECTABLE) { /* Chelsio, C5 and C6 of Mellanox? */ rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_rate_cnt = rl.number_of_rates; rs->rs_min_seg = rl.min_segment_burst; rs->rs_disable = 0; rs->rs_flow_limit = rl.max_flows; rate_table_act = desired_rates; if ((rs->rs_rate_cnt > MAX_HDWR_RATES) && (rs->rs_rate_cnt < ALL_HARDWARE_RATES)) { /* * Our desired table is not big * enough, do what we can. */ rs->rs_rate_cnt = MAX_HDWR_RATES; } if (rs->rs_rate_cnt <= RS_ORDERED_COUNT) rs->rs_flags = RS_IS_INTF; else rs->rs_flags = RS_IS_INTF | RS_INT_TBL; if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) rs->rs_rate_cnt = ALL_HARDWARE_RATES; } else { free(rs, M_TCPPACE); return (NULL); } sz = sizeof(struct tcp_hwrate_limit_table) * rs->rs_rate_cnt; rs->rs_rlt = malloc(sz, M_TCPPACE, M_NOWAIT); if (rs->rs_rlt == NULL) { if (error) *error = ENOMEM; bail: free(rs, M_TCPPACE); return (NULL); } if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) { /* * The interface supports all * the rates we could possibly want. */ uint64_t rat; rs->rs_rlt[0].rate = 12500; /* 100k */ rs->rs_rlt[1].rate = 25000; /* 200k */ rs->rs_rlt[2].rate = 62500; /* 500k */ /* Note 125000 == 1Megabit * populate 1Meg - 1000meg. */ for(i = 3, rat = 125000; i< (ALL_HARDWARE_RATES-1); i++) { rs->rs_rlt[i].rate = rat; rat += 125000; } rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate = 1250000000; } else if (rs->rs_flags & RS_INT_TBL) { /* We populate this in a special way */ populate_canned_table(rs, rate_table_act); } else { /* * Just copy in the rates from * the table, it is in order. */ for (i=0; irs_rate_cnt; i++) { rs->rs_rlt[i].rate = rate_table_act[i]; rs->rs_rlt[i].time_between = 0; rs->rs_rlt[i].flags = 0; } } for (i = (rs->rs_rate_cnt - 1); i >= 0; i--) { /* * We go backwards through the list so that if we can't get * a rate and fail to init one, we have at least a chance of * getting the highest one. */ rs->rs_rlt[i].ptbl = rs; rs->rs_rlt[i].tag = NULL; /* * Calculate the time between. */ lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND; res = lentim / rs->rs_rlt[i].rate; if (res > 0) rs->rs_rlt[i].time_between = res; else rs->rs_rlt[i].time_between = 1; if (rs->rs_flags & RS_NO_PRE) { rs->rs_rlt[i].flags = HDWRPACE_INITED; rs->rs_lowest_valid = i; } else { int err; if ((rl.flags & RT_IS_SETUP_REQ) && (ifp->if_ratelimit_query)) { err = ifp->if_ratelimit_setup(ifp, rs->rs_rlt[i].rate, i); if (err) goto handle_err; } #ifdef RSS hash_type = M_HASHTYPE_RSS_TCP_IPV4; #else hash_type = M_HASHTYPE_OPAQUE_HASH; #endif err = rl_attach_txrtlmt(ifp, hash_type, (i + 1), rs->rs_rlt[i].rate, &rs->rs_rlt[i].tag); if (err) { handle_err: if (i == (rs->rs_rate_cnt - 1)) { /* * Huh - first rate and we can't get * it? */ free(rs->rs_rlt, M_TCPPACE); if (error) *error = err; goto bail; } else { if (error) *error = err; } break; } else { rs->rs_rlt[i].flags = HDWRPACE_INITED | HDWRPACE_TAGPRESENT; rs->rs_lowest_valid = i; } } } /* Did we get at least 1 rate? */ if (rs->rs_rlt[(rs->rs_rate_cnt - 1)].flags & HDWRPACE_INITED) rs->rs_highest_valid = rs->rs_rate_cnt - 1; else { free(rs->rs_rlt, M_TCPPACE); goto bail; } rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } static const struct tcp_hwrate_limit_table * tcp_int_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags) { struct tcp_hwrate_limit_table *arte = NULL, *rte = NULL; uint64_t mbits_per_sec, ind_calc; int i; mbits_per_sec = (bytes_per_sec * 8); if (flags & RS_PACING_LT) { if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ /* * Smaller than 1Meg, only * 3 entries can match it. */ for(i = rs->rs_lowest_valid; i < 3; i++) { if (bytes_per_sec <= rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } else if (rs->rs_rlt[i].flags & HDWRPACE_INITED) { arte = &rs->rs_rlt[i]; } } goto done; } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ /* * Larger than 1G (the majority of * our table. */ if (mbits_per_sec < RS_TEN_GIGABIT_PERSEC) rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; else arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; goto done; } /* * If we reach here its in our table (between 1Meg - 1000Meg), * just take the rounded down mbits per second, and add * 1Megabit to it, from this we can calculate * the index in the table. */ ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) != mbits_per_sec) ind_calc++; /* our table is offset by 3, we add 2 */ ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if ((ind_calc >= rs->rs_lowest_valid) && (ind_calc <= rs->rs_highest_valid)) rte = &rs->rs_rlt[ind_calc]; } else if (flags & RS_PACING_EXACT_MATCH) { if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ for(i = rs->rs_lowest_valid; i < 3; i++) { if (bytes_per_sec == rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } } } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { /* > 1Gbps only one rate */ if (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) { /* Its 10G wow */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } } else { /* Ok it must be a exact meg (its between 1G and 1Meg) */ ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { /* its an exact Mbps */ ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) rte = &rs->rs_rlt[ind_calc]; } } } else { /* we want greater than the requested rate */ if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ arte = &rs->rs_rlt[3]; /* set alternate to 1Meg */ for (i=2; i>=rs->rs_lowest_valid; i--) { if (bytes_per_sec < rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } else if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; break; } else { arte = &rs->rs_rlt[i]; /* new alternate */ } } } else if (mbits_per_sec > RS_ONE_GIGABIT_PERSEC) { if ((bytes_per_sec < rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ /* Our top rate is larger than the request */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } else if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { /* It matches our top rate */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } else if (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED) { /* The top rate is an alternative */ arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } } else { /* Its in our range 1Meg - 1Gig */ if (flags & RS_PACING_GEQ) { ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = (ALL_HARDWARE_RATES-1); } rte = &rs->rs_rlt[ind_calc]; } goto done; } ind_calc = (mbits_per_sec + (RS_ONE_MEGABIT_PERSEC-1))/RS_ONE_MEGABIT_PERSEC; ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) rte = &rs->rs_rlt[ind_calc]; } } done: if ((rte == NULL) && (arte != NULL) && (flags & RS_PACING_SUB_OK)) { /* We can use the substitute */ rte = arte; } return (rte); } static const struct tcp_hwrate_limit_table * tcp_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags) { /** * Hunt the rate table with the restrictions in flags and find a * suitable rate if possible. * RS_PACING_EXACT_MATCH - look for an exact match to rate. * RS_PACING_GT - must be greater than. * RS_PACING_GEQ - must be greater than or equal. * RS_PACING_LT - must be less than. * RS_PACING_SUB_OK - If we don't meet criteria a * substitute is ok. */ int i, matched; struct tcp_hwrate_limit_table *rte = NULL; if ((rs->rs_flags & RS_INT_TBL) && (rs->rs_rate_cnt >= ALL_HARDWARE_RATES)) { /* * Here we don't want to paw thru * a big table, we have everything * from 1Meg - 1000Meg in 1Meg increments. * Use an alternate method to "lookup". */ return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags)); } if ((flags & RS_PACING_LT) || (flags & RS_PACING_EXACT_MATCH)) { /* * For exact and less than we go forward through the table. * This way when we find one larger we stop (exact was a * toss up). */ for (i = rs->rs_lowest_valid, matched = 0; i <= rs->rs_highest_valid; i++) { if ((flags & RS_PACING_EXACT_MATCH) && (bytes_per_sec == rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; matched = 1; break; } else if ((flags & RS_PACING_LT) && (bytes_per_sec <= rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; matched = 1; break; } if (bytes_per_sec > rs->rs_rlt[i].rate) break; } if ((matched == 0) && (flags & RS_PACING_LT) && (flags & RS_PACING_SUB_OK)) { /* Kick in a substitute (the lowest) */ rte = &rs->rs_rlt[rs->rs_lowest_valid]; } } else { /* * Here we go backward through the table so that we can find * the one greater in theory faster (but its probably a * wash). */ for (i = rs->rs_highest_valid, matched = 0; i >= rs->rs_lowest_valid; i--) { if (rs->rs_rlt[i].rate > bytes_per_sec) { /* A possible candidate */ rte = &rs->rs_rlt[i]; } if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[i].rate)) { /* An exact match and we want equal */ matched = 1; rte = &rs->rs_rlt[i]; break; } else if (rte) { /* * Found one that is larger than but don't * stop, there may be a more closer match. */ matched = 1; } if (rs->rs_rlt[i].rate < bytes_per_sec) { /* * We found a table entry that is smaller, * stop there will be none greater or equal. */ break; } } if ((matched == 0) && (flags & RS_PACING_SUB_OK)) { /* Kick in a substitute (the highest) */ rte = &rs->rs_rlt[rs->rs_highest_valid]; } } return (rte); } static struct ifnet * rt_find_real_interface(struct ifnet *ifp, struct inpcb *inp, int *error) { struct ifnet *tifp; struct m_snd_tag *tag; union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = 1, + .rate_limit.hdr.numa_domain = inp->inp_numa_domain, .rate_limit.max_rate = COMMON_RATE, .rate_limit.flags = M_NOWAIT, }; int err; #ifdef RSS params.rate_limit.hdr.flowtype = ((inp->inp_vflag & INP_IPV6) ? M_HASHTYPE_RSS_TCP_IPV6 : M_HASHTYPE_RSS_TCP_IPV4); #else params.rate_limit.hdr.flowtype = M_HASHTYPE_OPAQUE_HASH; #endif tag = NULL; if (ifp->if_snd_tag_alloc) { if (error) *error = ENODEV; return (NULL); } err = ifp->if_snd_tag_alloc(ifp, ¶ms, &tag); if (err) { /* Failed to setup a tag? */ if (error) *error = err; return (NULL); } tifp = tag->ifp; tifp->if_snd_tag_free(tag); return (tifp); } static const struct tcp_hwrate_limit_table * rt_setup_rate(struct inpcb *inp, struct ifnet *ifp, uint64_t bytes_per_sec, uint32_t flags, int *error) { /* First lets find the interface if it exists */ const struct tcp_hwrate_limit_table *rte; struct tcp_rate_set *rs; struct epoch_tracker et; int err; NET_EPOCH_ENTER(et); use_real_interface: CK_LIST_FOREACH(rs, &int_rs, next) { /* * Note we don't look with the lock since we either see a * new entry or will get one when we try to add it. */ if (rs->rs_flags & RS_IS_DEAD) { /* The dead are not looked at */ continue; } if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { /* Ok we found it */ break; } } if ((rs == NULL) || (rs->rs_flags & RS_INTF_NO_SUP) || (rs->rs_flags & RS_IS_DEAD)) { /* * This means we got a packet *before* * the IF-UP was processed below, * while or after we already received an interface * departed event. In either case we really don't * want to do anything with pacing, in * the departing case the packet is not * going to go very far. The new case * might be arguable, but its impossible * to tell from the departing case. */ if (rs->rs_disable && error) *error = ENODEV; NET_EPOCH_EXIT(et); return (NULL); } if ((rs == NULL) || (rs->rs_disable != 0)) { if (rs->rs_disable && error) *error = ENOSPC; NET_EPOCH_EXIT(et); return (NULL); } if (rs->rs_flags & RS_IS_DEFF) { /* We need to find the real interface */ struct ifnet *tifp; tifp = rt_find_real_interface(ifp, inp, error); if (tifp == NULL) { if (rs->rs_disable && error) *error = ENOTSUP; NET_EPOCH_EXIT(et); return (NULL); } goto use_real_interface; } if (rs->rs_flow_limit && ((rs->rs_flows_using + 1) > rs->rs_flow_limit)) { if (error) *error = ENOSPC; NET_EPOCH_EXIT(et); return (NULL); } rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); if (rte) { err = in_pcbattach_txrtlmt(inp, rs->rs_ifp, inp->inp_flowtype, inp->inp_flowid, rte->rate, &inp->inp_snd_tag); if (err) { /* Failed to attach */ if (error) *error = err; rte = NULL; } } if (rte) { /* * We use an atomic here for accounting so we don't have to * use locks when freeing. */ atomic_add_64(&rs->rs_flows_using, 1); } NET_EPOCH_EXIT(et); return (rte); } static void tcp_rl_ifnet_link(void *arg __unused, struct ifnet *ifp, int link_state) { int error; struct tcp_rate_set *rs; if (((ifp->if_capabilities & IFCAP_TXRTLMT) == 0) || (link_state != LINK_STATE_UP)) { /* * We only care on an interface going up that is rate-limit * capable. */ return; } mtx_lock(&rs_mtx); CK_LIST_FOREACH(rs, &int_rs, next) { if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { /* We already have initialized this guy */ mtx_unlock(&rs_mtx); return; } } mtx_unlock(&rs_mtx); rt_setup_new_rs(ifp, &error); } static void tcp_rl_ifnet_departure(void *arg __unused, struct ifnet *ifp) { struct tcp_rate_set *rs, *nrs; struct ifnet *tifp; int i; mtx_lock(&rs_mtx); CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { CK_LIST_REMOVE(rs, next); rs_number_alive--; rs->rs_flags |= RS_IS_DEAD; for (i = 0; i < rs->rs_rate_cnt; i++) { if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { tifp = rs->rs_rlt[i].tag->ifp; in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); rs->rs_rlt[i].tag = NULL; } rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; } if (rs->rs_flows_using == 0) rs_defer_destroy(rs); break; } } mtx_unlock(&rs_mtx); } static void tcp_rl_shutdown(void *arg __unused, int howto __unused) { struct tcp_rate_set *rs, *nrs; struct ifnet *tifp; int i; mtx_lock(&rs_mtx); CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { CK_LIST_REMOVE(rs, next); rs_number_alive--; rs->rs_flags |= RS_IS_DEAD; for (i = 0; i < rs->rs_rate_cnt; i++) { if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { tifp = rs->rs_rlt[i].tag->ifp; in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); rs->rs_rlt[i].tag = NULL; } rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; } if (rs->rs_flows_using == 0) rs_defer_destroy(rs); } mtx_unlock(&rs_mtx); } const struct tcp_hwrate_limit_table * tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, uint64_t bytes_per_sec, int flags, int *error) { const struct tcp_hwrate_limit_table *rte; if (tp->t_inpcb->inp_snd_tag == NULL) { /* * We are setting up a rate for the first time. */ if ((ifp->if_capabilities & IFCAP_TXRTLMT) == 0) { /* Not supported by the egress */ if (error) *error = ENODEV; return (NULL); } #ifdef KERN_TLS if (tp->t_inpcb->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) { /* * We currently can't do both TLS and hardware * pacing */ if (error) *error = EINVAL; return (NULL); } #endif rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error); } else { /* * We are modifying a rate, wrong interface? */ if (error) *error = EINVAL; rte = NULL; } *error = 0; return (rte); } const struct tcp_hwrate_limit_table * tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, struct tcpcb *tp, struct ifnet *ifp, uint64_t bytes_per_sec, int flags, int *error) { const struct tcp_hwrate_limit_table *nrte; const struct tcp_rate_set *rs; int is_indirect = 0; int err; if ((tp->t_inpcb->inp_snd_tag == NULL) || (crte == NULL)) { /* Wrong interface */ if (error) *error = EINVAL; return (NULL); } rs = crte->ptbl; if ((rs->rs_flags & RS_IS_DEAD) || (crte->flags & HDWRPACE_IFPDEPARTED)) { /* Release the rate, and try anew */ re_rate: tcp_rel_pacing_rate(crte, tp); nrte = tcp_set_pacing_rate(tp, ifp, bytes_per_sec, flags, error); return (nrte); } if ((rs->rs_flags & RT_IS_INDIRECT ) == RT_IS_INDIRECT) is_indirect = 1; else is_indirect = 0; if ((is_indirect == 0) && ((ifp != rs->rs_ifp) || (ifp->if_dunit != rs->rs_if_dunit))) { /* * Something changed, the user is not pointing to the same * ifp? Maybe a route updated on this guy? */ goto re_rate; } else if (is_indirect) { /* * For indirect we have to dig in and find the real interface. */ struct ifnet *rifp; rifp = rt_find_real_interface(ifp, tp->t_inpcb, error); if (rifp == NULL) { /* Can't find it? */ goto re_rate; } if ((rifp != rs->rs_ifp) || (ifp->if_dunit != rs->rs_if_dunit)) { goto re_rate; } } nrte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); if (nrte == crte) { /* No change */ if (error) *error = 0; return (crte); } if (nrte == NULL) { /* Release the old rate */ tcp_rel_pacing_rate(crte, tp); return (NULL); } /* Change rates to our new entry */ err = in_pcbmodify_txrtlmt(tp->t_inpcb, nrte->rate); if (err) { if (error) *error = err; return (NULL); } if (error) *error = 0; return (nrte); } void tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte, struct tcpcb *tp) { const struct tcp_rate_set *crs; struct tcp_rate_set *rs; uint64_t pre; crs = crte->ptbl; /* * Now we must break the const * in order to release our refcount. */ rs = __DECONST(struct tcp_rate_set *, crs); pre = atomic_fetchadd_64(&rs->rs_flows_using, -1); if (pre == 1) { mtx_lock(&rs_mtx); /* * Is it dead? */ if (rs->rs_flags & RS_IS_DEAD) rs_defer_destroy(rs); mtx_unlock(&rs_mtx); } in_pcbdetach_txrtlmt(tp->t_inpcb); } #define ONE_POINT_TWO_MEG 150000 /* 1.2 megabits in bytes */ #define ONE_HUNDRED_MBPS 12500000 /* 100Mbps in bytes per second */ #define FIVE_HUNDRED_MBPS 62500000 /* 500Mbps in bytes per second */ #define MAX_MSS_SENT 43 /* 43 mss = 43 x 1500 = 64,500 bytes */ uint32_t tcp_get_pacing_burst_size (uint64_t bw, uint32_t segsiz, int can_use_1mss, const struct tcp_hwrate_limit_table *te, int *err) { /* * We use the google formula to calculate the * TSO size. I.E. * bw < 24Meg * tso = 2mss * else * tso = min(bw/1000, 64k) * * Note for these calculations we ignore the * packet overhead (enet hdr, ip hdr and tcp hdr). */ uint64_t lentim, res, bytes; uint32_t new_tso, min_tso_segs; bytes = bw / 1000; if (bytes > (64 * 1000)) bytes = 64 * 1000; /* Round up */ new_tso = (bytes + segsiz - 1) / segsiz; if (can_use_1mss && (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; if (new_tso > MAX_MSS_SENT) new_tso = MAX_MSS_SENT; new_tso *= segsiz; /* * If we are not doing hardware pacing * then we are done. */ if (te == NULL) { if (err) *err = 0; return(new_tso); } /* * For hardware pacing we look at the * rate you are sending at and compare * that to the rate you have in hardware. * * If the hardware rate is slower than your * software rate then you are in error and * we will build a queue in our hardware whic * is probably not desired, in such a case * just return the non-hardware TSO size. * * If the rate in hardware is faster (which * it should be) then look at how long it * takes to send one ethernet segment size at * your b/w and compare that to the time it * takes to send at the rate you had selected. * * If your time is greater (which we hope it is) * we get the delta between the two, and then * divide that into your pacing time. This tells * us how many MSS you can send down at once (rounded up). * * Note we also double this value if the b/w is over * 100Mbps. If its over 500meg we just set you to the * max (43 segments). */ if (te->rate > FIVE_HUNDRED_MBPS) return (segsiz * MAX_MSS_SENT); if (te->rate == bw) { /* We are pacing at exactly the hdwr rate */ return (segsiz * MAX_MSS_SENT); } lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND; res = lentim / bw; if (res > te->time_between) { uint32_t delta, segs; delta = res - te->time_between; segs = (res + delta - 1)/delta; if (te->rate > ONE_HUNDRED_MBPS) segs *= 2; if (segs < min_tso_segs) segs = min_tso_segs; if (segs > MAX_MSS_SENT) segs = MAX_MSS_SENT; segs *= segsiz; if (err) *err = 0; if (segs < new_tso) { /* unexpected ? */ return(new_tso); } else { return (segs); } } else { /* * Your time is smaller which means * we will grow a queue on our * hardware. Send back the non-hardware * rate. */ if (err) *err = -1; return (new_tso); } } static eventhandler_tag rl_ifnet_departs; static eventhandler_tag rl_ifnet_arrives; static eventhandler_tag rl_shutdown_start; static void tcp_rs_init(void *st __unused) { CK_LIST_INIT(&int_rs); rs_number_alive = 0; rs_number_dead = 0;; mtx_init(&rs_mtx, "tcp_rs_mtx", "rsmtx", MTX_DEF); rl_ifnet_departs = EVENTHANDLER_REGISTER(ifnet_departure_event, tcp_rl_ifnet_departure, NULL, EVENTHANDLER_PRI_ANY); rl_ifnet_arrives = EVENTHANDLER_REGISTER(ifnet_link_event, tcp_rl_ifnet_link, NULL, EVENTHANDLER_PRI_ANY); rl_shutdown_start = EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_rl_shutdown, NULL, SHUTDOWN_PRI_FIRST); printf("TCP_ratelimit: Is now initialized\n"); } SYSINIT(tcp_rl_init, SI_SUB_SMP + 1, SI_ORDER_ANY, tcp_rs_init, NULL); #endif