/* * Copyright (c) 2016-2019 Cisco and/or its affiliates. * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ /** * @file * @brief TCP host stack utilities */ #include #include #include #include #include #include tcp_main_t tcp_main; typedef struct { fib_protocol_t nh_proto; vnet_link_t link_type; ip46_address_t ip; u32 sw_if_index; u8 is_add; } tcp_add_del_adj_args_t; static void tcp_add_del_adj_cb (tcp_add_del_adj_args_t * args) { u32 ai; if (args->is_add) { adj_nbr_add_or_lock (args->nh_proto, args->link_type, &args->ip, args->sw_if_index); } else { ai = adj_nbr_find (FIB_PROTOCOL_IP6, VNET_LINK_IP6, &args->ip, args->sw_if_index); if (ai != ADJ_INDEX_INVALID) adj_unlock (ai); } } static void tcp_add_del_adjacency (tcp_connection_t * tc, u8 is_add) { tcp_add_del_adj_args_t args = { .nh_proto = FIB_PROTOCOL_IP6, .link_type = VNET_LINK_IP6, .ip = tc->c_rmt_ip, .sw_if_index = tc->sw_if_index, .is_add = is_add }; vlib_rpc_call_main_thread (tcp_add_del_adj_cb, (u8 *) & args, sizeof (args)); } static void tcp_cc_init (tcp_connection_t * tc) { tc->cc_algo->init (tc); } static void tcp_cc_cleanup (tcp_connection_t * tc) { if (tc->cc_algo->cleanup) tc->cc_algo->cleanup (tc); } void tcp_cc_algo_register (tcp_cc_algorithm_type_e type, const tcp_cc_algorithm_t * vft) { tcp_main_t *tm = vnet_get_tcp_main (); vec_validate (tm->cc_algos, type); tm->cc_algos[type] = *vft; hash_set_mem (tm->cc_algo_by_name, vft->name, type); } tcp_cc_algorithm_t * tcp_cc_algo_get (tcp_cc_algorithm_type_e type) { tcp_main_t *tm = vnet_get_tcp_main (); return &tm->cc_algos[type]; } tcp_cc_algorithm_type_e tcp_cc_algo_new_type (const tcp_cc_algorithm_t * vft) { tcp_main_t *tm = vnet_get_tcp_main (); tcp_cc_algo_register (++tm->cc_last_type, vft); return tm->cc_last_type; } static u32 tcp_connection_bind (u32 session_index, transport_endpoint_t * lcl) { tcp_main_t *tm = &tcp_main; tcp_connection_t *listener; void *iface_ip; pool_get (tm->listener_pool, listener); clib_memset (listener, 0, sizeof (*listener)); listener->c_c_index = listener - tm->listener_pool; listener->c_lcl_port = lcl->port; /* If we are provided a sw_if_index, bind using one of its ips */ if (ip_is_zero (&lcl->ip, 1) && lcl->sw_if_index != ENDPOINT_INVALID_INDEX) { if ((iface_ip = ip_interface_get_first_ip (lcl->sw_if_index, lcl->is_ip4))) ip_set (&lcl->ip, iface_ip, lcl->is_ip4); } ip_copy (&listener->c_lcl_ip, &lcl->ip, lcl->is_ip4); listener->c_is_ip4 = lcl->is_ip4; listener->c_proto = TRANSPORT_PROTO_TCP; listener->c_s_index = session_index; listener->c_fib_index = lcl->fib_index; listener->state = TCP_STATE_LISTEN; listener->cc_algo = tcp_cc_algo_get (tcp_cfg.cc_algo); tcp_connection_timers_init (listener); TCP_EVT (TCP_EVT_BIND, listener); return listener->c_c_index; } static u32 tcp_session_bind (u32 session_index, transport_endpoint_t * tep) { return tcp_connection_bind (session_index, tep); } static void tcp_connection_unbind (u32 listener_index) { tcp_main_t *tm = vnet_get_tcp_main (); tcp_connection_t *tc; tc = pool_elt_at_index (tm->listener_pool, listener_index); TCP_EVT (TCP_EVT_UNBIND, tc); /* Poison the entry */ if (CLIB_DEBUG > 0) clib_memset (tc, 0xFA, sizeof (*tc)); pool_put_index (tm->listener_pool, listener_index); } static u32 tcp_session_unbind (u32 listener_index) { tcp_connection_unbind (listener_index); return 0; } static transport_connection_t * tcp_session_get_listener (u32 listener_index) { tcp_main_t *tm = vnet_get_tcp_main (); tcp_connection_t *tc; tc = pool_elt_at_index (tm->listener_pool, listener_index); return &tc->connection; } /** * Cleanup half-open connection * */ static void tcp_half_open_connection_free (tcp_connection_t * tc) { tcp_main_t *tm = vnet_get_tcp_main (); clib_spinlock_lock_if_init (&tm->half_open_lock); if (CLIB_DEBUG) clib_memset (tc, 0xFA, sizeof (*tc)); pool_put (tm->half_open_connections, tc); clib_spinlock_unlock_if_init (&tm->half_open_lock); } /** * Try to cleanup half-open connection * * If called from a thread that doesn't own tc, the call won't have any * effect. * * @param tc - connection to be cleaned up * @return non-zero if cleanup failed. */ int tcp_half_open_connection_cleanup (tcp_connection_t * tc) { tcp_worker_ctx_t *wrk; /* Make sure this is the owning thread */ if (tc->c_thread_index != vlib_get_thread_index ()) return 1; session_half_open_delete_notify (TRANSPORT_PROTO_TCP, tc->c_s_ho_handle); wrk = tcp_get_worker (tc->c_thread_index); tcp_timer_reset (&wrk->timer_wheel, tc, TCP_TIMER_RETRANSMIT_SYN); tcp_half_open_connection_free (tc); return 0; } static tcp_connection_t * tcp_half_open_connection_new (void) { tcp_main_t *tm = vnet_get_tcp_main (); tcp_connection_t *tc = 0; ASSERT (vlib_get_thread_index () == 0); pool_get (tm->half_open_connections, tc); clib_memset (tc, 0, sizeof (*tc)); tc->c_c_index = tc - tm->half_open_connections; return tc; } /** * Cleans up connection state. * * No notifications. */ void tcp_connection_cleanup (tcp_connection_t * tc) { TCP_EVT (TCP_EVT_DELETE, tc); /* Cleanup local endpoint if this was an active connect */ if (!(tc->cfg_flags & TCP_CFG_F_NO_ENDPOINT)) transport_endpoint_cleanup (TRANSPORT_PROTO_TCP, &tc->c_lcl_ip, tc->c_lcl_port); /* Check if connection is not yet fully established */ if (tc->state == TCP_STATE_SYN_SENT) { /* Try to remove the half-open connection. If this is not the owning * thread, tc won't be removed. Retransmit or establish timers will * eventually expire and call again cleanup on the right thread. */ if (tcp_half_open_connection_cleanup (tc)) tc->flags |= TCP_CONN_HALF_OPEN_DONE; } else { /* Make sure all timers are cleared */ tcp_connection_timers_reset (tc); if (!tc->c_is_ip4 && ip6_address_is_link_local_unicast (&tc->c_rmt_ip6)) tcp_add_del_adjacency (tc, 0); tcp_cc_cleanup (tc); vec_free (tc->snd_sacks); vec_free (tc->snd_sacks_fl); vec_free (tc->rcv_opts.sacks); pool_free (tc->sack_sb.holes); if (tc->cfg_flags & TCP_CFG_F_RATE_SAMPLE) tcp_bt_cleanup (tc); tcp_connection_free (tc); } } /** * Connection removal. * * This should be called only once connection enters CLOSED state. Note * that it notifies the session of the removal event, so if the goal is to * just remove the connection, call tcp_connection_cleanup instead. */ void tcp_connection_del (tcp_connection_t * tc) { session_transport_delete_notify (&tc->connection); tcp_connection_cleanup (tc); } tcp_connection_t * tcp_connection_alloc (u8 thread_index) { tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index); tcp_connection_t *tc; pool_get (wrk->connections, tc); clib_memset (tc, 0, sizeof (*tc)); tc->c_c_index = tc - wrk->connections; tc->c_thread_index = thread_index; return tc; } tcp_connection_t * t
/*
 * Copyright (c) 2016 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
/**
 * @brief
 * The data-path object representing puntping the packet
 */

#include <vnet/dpo/dpo.h>

static dpo_id_t punt_dpos[DPO_PROTO_NUM];

const dpo_id_t *
punt_dpo_get (dpo_proto_t proto)
{
    dpo_set(&punt_dpos[proto], DPO_PUNT, proto, 1);

    return (&punt_dpos[proto]);
}

int
dpo_is_punt (const dpo_id_t *dpo)
{
    return (dpo->dpoi_type == DPO_PUNT);
}

static void
punt_dpo_lock (dpo_id_t *dpo)
{
    /*
     * not maintaining a lock count on the punt
     * more trouble than it's worth.
     * There always needs to be one around. no point it managaing its lifetime
     */
}
static void
punt_dpo_unlock (dpo_id_t *dpo)
{
}

static u8*
format_punt_dpo (u8 *s, va_list *ap)
{
    CLIB_UNUSED(index_t index) = va_arg(*ap, index_t);
    CLIB_UNUSED(u32 indent) = va_arg(*ap, u32);

    return (format(s, "dpo-punt"));
}

const static dpo_vft_t punt_vft = {
    .dv_lock   = punt_dpo_lock,
    .dv_unlock = punt_dpo_unlock,
    .dv_format = format_punt_dpo,
};

/**
 * @brief The per-protocol VLIB graph nodes that are assigned to a punt
 *        object.
 *
 * this means that these graph nodes are ones from which a punt is the
 * parent object in the DPO-graph.
 */
const static char* const punt_ip4_nodes[] =
{
    "ip4-punt",
    NULL,
};
const static char* const punt_ip6_nodes[] =
{
    "ip6-punt",
    NULL,
};
const static char* const punt_mpls_nodes[] =
{
    "mpls-punt",
    NULL,
};
const static char* const * const punt_nodes[DPO_PROTO_NUM] =
{
    [DPO_PROTO_IP4]  = punt_ip4_nodes,
    [DPO_PROTO_IP6]  = punt_ip6_nodes,
    [DPO_PROTO_MPLS] = punt_mpls_nodes,
};

void
punt_dpo_module_init (void)
{
    dpo_register(DPO_PUNT, &punt_vft, punt_nodes);
}
; /* This still works if fast retransmit is on */ sp->tx_offset = tc->snd_nxt - tc->snd_una; sp->flags = sp->snd_space ? 0 : TRANSPORT_SND_F_DESCHED; return 0; } static void tcp_timer_waitclose_handler (tcp_connection_t * tc) { tcp_worker_ctx_t *wrk = tcp_get_worker (tc->c_thread_index); switch (tc->state) { case TCP_STATE_CLOSE_WAIT: tcp_connection_timers_reset (tc); /* App never returned with a close */ if (!(tc->flags & TCP_CONN_FINPNDG)) { tcp_connection_set_state (tc, TCP_STATE_CLOSED); session_transport_closed_notify (&tc->connection); tcp_program_cleanup (wrk, tc); tcp_worker_stats_inc (wrk, to_closewait, 1); break; } /* Send FIN either way and switch to LAST_ACK. */ tcp_cong_recovery_off (tc); /* Make sure we don't try to send unsent data */ tc->snd_nxt = tc->snd_una; tcp_send_fin (tc); tcp_connection_set_state (tc, TCP_STATE_LAST_ACK); session_transport_closed_notify (&tc->connection); /* Make sure we don't wait in LAST ACK forever */ tcp_timer_set (&wrk->timer_wheel, tc, TCP_TIMER_WAITCLOSE, tcp_cfg.lastack_time); tcp_worker_stats_inc (wrk, to_closewait2, 1); /* Don't delete the connection yet */ break; case TCP_STATE_FIN_WAIT_1: tcp_connection_timers_reset (tc); if (tc->flags & TCP_CONN_FINPNDG) { /* If FIN pending, we haven't sent everything, but we did try. * Notify session layer that transport is closed. */ tcp_connection_set_state (tc, TCP_STATE_CLOSED); tcp_send_reset (tc); tcp_program_cleanup (wrk, tc); } else { /* We've sent the fin but no progress. Close the connection and * to make sure everything is flushed, setup a cleanup timer */ tcp_connection_set_state (tc, TCP_STATE_CLOSED); tcp_program_cleanup (wrk, tc); } session_transport_closed_notify (&tc->connection); tcp_worker_stats_inc (wrk, to_finwait1, 1); break; case TCP_STATE_LAST_ACK: tcp_connection_timers_reset (tc); tcp_connection_set_state (tc, TCP_STATE_CLOSED); session_transport_closed_notify (&tc->connection); tcp_program_cleanup (wrk, tc); tcp_worker_stats_inc (wrk, to_lastack, 1); break; case TCP_STATE_CLOSING: tcp_connection_timers_reset (tc); tcp_connection_set_state (tc, TCP_STATE_CLOSED); session_transport_closed_notify (&tc->connection); tcp_program_cleanup (wrk, tc); tcp_worker_stats_inc (wrk, to_closing, 1); break; case TCP_STATE_FIN_WAIT_2: tcp_send_reset (tc); tcp_connection_timers_reset (tc); tcp_connection_set_state (tc, TCP_STATE_CLOSED); session_transport_closed_notify (&tc->connection); tcp_program_cleanup (wrk, tc); tcp_worker_stats_inc (wrk, to_finwait2, 1); break; case TCP_STATE_TIME_WAIT: tcp_connection_set_state (tc, TCP_STATE_CLOSED); tcp_program_cleanup (wrk, tc); break; default: clib_warning ("waitclose in state: %U", format_tcp_state, tc->state); break; } } /* *INDENT-OFF* */ static timer_expiration_handler *timer_expiration_handlers[TCP_N_TIMERS] = { tcp_timer_retransmit_handler, tcp_timer_persist_handler, tcp_timer_waitclose_handler, tcp_timer_retransmit_syn_handler, }; /* *INDENT-ON* */ static void tcp_dispatch_pending_timers (tcp_worker_ctx_t * wrk) { u32 n_timers, connection_index, timer_id, thread_index, timer_handle; tcp_connection_t *tc; int i; if (!(n_timers = clib_fifo_elts (wrk->pending_timers))) return; thread_index = wrk->vm->thread_index; for (i = 0; i < clib_min (n_timers, wrk->max_timers_per_loop); i++) { clib_fifo_sub1 (wrk->pending_timers, timer_handle); connection_index = timer_handle & 0x0FFFFFFF; timer_id = timer_handle >> 28; if (PREDICT_TRUE (timer_id != TCP_TIMER_RETRANSMIT_SYN)) tc = tcp_connection_get (connection_index, thread_index); else tc = tcp_half_open_connection_get (connection_index); if (PREDICT_FALSE (!tc)) continue; /* Skip if the timer is not pending. Probably it was reset while * wating for dispatch */ if (PREDICT_FALSE (!(tc->pending_timers & (1 << timer_id)))) continue; tc->pending_timers &= ~(1 << timer_id); /* Skip timer if it was rearmed while pending dispatch */ if (PREDICT_FALSE (tc->timers[timer_id] != TCP_TIMER_HANDLE_INVALID)) continue; (*timer_expiration_handlers[timer_id]) (tc); } if (thread_index == 0 && clib_fifo_elts (wrk->pending_timers)) session_queue_run_on_main_thread (wrk->vm); } static void tcp_handle_cleanups (tcp_worker_ctx_t * wrk, clib_time_type_t now) { u32 thread_index = wrk->vm->thread_index; tcp_cleanup_req_t *req; tcp_connection_t *tc; while (clib_fifo_elts (wrk->pending_cleanups)) { req = clib_fifo_head (wrk->pending_cleanups); if (req->free_time > now) break; clib_fifo_sub2 (wrk->pending_cleanups, req); tc = tcp_connection_get (req->connection_index, thread_index); if (PREDICT_FALSE (!tc)) continue; session_transport_delete_notify (&tc->connection); tcp_connection_cleanup (tc); } } static void tcp_update_time (f64 now, u8 thread_index) { tcp_worker_ctx_t *wrk = tcp_get_worker (thread_index); tcp_set_time_now (wrk); tcp_handle_cleanups (wrk, now); tw_timer_expire_timers_16t_2w_512sl (&wrk->timer_wheel, now); tcp_dispatch_pending_timers (wrk); } static void tcp_session_flush_data (transport_connection_t * tconn) { tcp_connection_t *tc = (tcp_connection_t *) tconn; if (tc->flags & TCP_CONN_PSH_PENDING) return; tc->flags |= TCP_CONN_PSH_PENDING; tc->psh_seq = tc->snd_una + transport_max_tx_dequeue (tconn) - 1; } /* *INDENT-OFF* */ const static transport_proto_vft_t tcp_proto = { .enable = vnet_tcp_enable_disable, .start_listen = tcp_session_bind, .stop_listen = tcp_session_unbind, .push_header = tcp_session_push_header, .get_connection = tcp_session_get_transport, .get_listener = tcp_session_get_listener, .get_half_open = tcp_half_open_session_get_transport, .connect = tcp_session_open, .close = tcp_session_close, .cleanup = tcp_session_cleanup, .cleanup_ho = tcp_session_cleanup_ho, .reset = tcp_session_reset, .send_params = tcp_session_send_params, .update_time = tcp_update_time, .flush_data = tcp_session_flush_data, .custom_tx = tcp_session_custom_tx, .format_connection = format_tcp_session, .format_listener = format_tcp_listener_session, .format_half_open = format_tcp_half_open_session, .transport_options = { .name = "tcp", .short_name = "T", .tx_type = TRANSPORT_TX_PEEK, .service_type = TRANSPORT_SERVICE_VC, }, }; /* *INDENT-ON* */ void tcp_connection_tx_pacer_update (tcp_connection_t * tc) { if (!transport_connection_is_tx_paced (&tc->connection)) return; f64 srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us); transport_connection_tx_pacer_update (&tc->connection, tcp_cc_get_pacing_rate (tc), srtt * CLIB_US_TIME_FREQ); } void tcp_connection_tx_pacer_reset (tcp_connection_t * tc, u32 window, u32 start_bucket) { f64 srtt = clib_min ((f64) tc->srtt * TCP_TICK, tc->mrtt_us); transport_connection_tx_pacer_reset (&tc->connection, tcp_cc_get_pacing_rate (tc), start_bucket, srtt * CLIB_US_TIME_FREQ); } void tcp_reschedule (tcp_connection_t * tc) { if (tcp_in_cong_recovery (tc) || tcp_snd_space_inline (tc)) transport_connection_reschedule (&tc->connection); } static void tcp_expired_timers_dispatch (u32 * expired_timers) { u32 thread_index = vlib_get_thread_index (), n_left, max_per_loop; u32 connection_index, timer_id, n_expired, max_loops; tcp_worker_ctx_t *wrk; tcp_connection_t *tc; int i; wrk = tcp_get_worker (thread_index); n_expired = vec_len (expired_timers); tcp_worker_stats_inc (wrk, timer_expirations, n_expired); n_left = clib_fifo_elts (wrk->pending_timers); /* * Invalidate all timer handles before dispatching. This avoids dangling * index references to timer wheel pool entries that have been freed. */ for (i = 0; i < n_expired; i++) { connection_index = expired_timers[i] & 0x0FFFFFFF; timer_id = expired_timers[i] >> 28; if (timer_id != TCP_TIMER_RETRANSMIT_SYN) tc = tcp_connection_get (connection_index, thread_index); else tc = tcp_half_open_connection_get (connection_index); TCP_EVT (TCP_EVT_TIMER_POP, connection_index, timer_id); tc->timers[timer_id] = TCP_TIMER_HANDLE_INVALID; tc->pending_timers |= (1 << timer_id); } clib_fifo_add (wrk->pending_timers, expired_timers, n_expired); max_loops = clib_max (1, 0.5 * TCP_TIMER_TICK * wrk->vm->loops_per_second); max_per_loop = clib_max ((n_left + n_expired) / max_loops, 10); max_per_loop = clib_min (max_per_loop, VLIB_FRAME_SIZE); wrk->max_timers_per_loop = clib_max (n_left ? wrk->max_timers_per_loop : 0, max_per_loop); if (thread_index == 0) session_queue_run_on_main_thread (wrk->vm); } static void tcp_initialize_timer_wheels (tcp_main_t * tm) { vlib_main_t *vm = vlib_get_main (); tw_timer_wheel_16t_2w_512sl_t *tw; /* *INDENT-OFF* */ foreach_vlib_main (({ tw = &tm->wrk_ctx[ii].timer_wheel; tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch, TCP_TIMER_TICK, ~0); tw->last_run_time = vlib_time_now (vm); })); /* *INDENT-ON* */ } static void tcp_initialize_iss_seed (tcp_main_t * tm) { u32 default_seed = random_default_seed (); u64 time_now = clib_cpu_time_now (); tm->iss_seed.first = (u64) random_u32 (&default_seed) << 32; tm->iss_seed.second = random_u64 (&time_now); } static clib_error_t * tcp_main_enable (vlib_main_t * vm) { vlib_thread_main_t *vtm = vlib_get_thread_main (); u32 num_threads, n_workers, prealloc_conn_per_wrk; tcp_connection_t *tc __attribute__ ((unused)); tcp_main_t *tm = vnet_get_tcp_main (); tcp_worker_ctx_t *wrk; clib_error_t *error = 0; int thread; if ((error = vlib_call_init_function (vm, ip_main_init))) return error; if ((error = vlib_call_init_function (vm, ip4_lookup_init))) return error; if ((error = vlib_call_init_function (vm, ip6_lookup_init))) return error; /* * Registrations */ ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index); ip6_register_protocol (IP_PROTOCOL_TCP, tcp6_input_node.index); /* * Initialize data structures */ num_threads = 1 /* main thread */ + vtm->n_threads; vec_validate (tm->wrk_ctx, num_threads - 1); n_workers = num_threads == 1 ? 1 : vtm->n_threads; prealloc_conn_per_wrk = tcp_cfg.preallocated_connections / n_workers; wrk = &tm->wrk_ctx[0]; wrk->tco_next_node[0] = vlib_node_get_next (vm, session_queue_node.index, tcp4_output_node.index); wrk->tco_next_node[1] = vlib_node_get_next (vm, session_queue_node.index, tcp6_output_node.index); for (thread = 0; thread < num_threads; thread++) { wrk = &tm->wrk_ctx[thread]; vec_validate (wrk->pending_deq_acked, 255); vec_validate (wrk->pending_disconnects, 255); vec_validate (wrk->pending_resets, 255); vec_reset_length (wrk->pending_deq_acked); vec_reset_length (wrk->pending_disconnects); vec_reset_length (wrk->pending_resets); wrk->vm = vlib_mains[thread]; wrk->max_timers_per_loop = 10; if (thread > 0) { wrk->tco_next_node[0] = tm->wrk_ctx[0].tco_next_node[0]; wrk->tco_next_node[1] = tm->wrk_ctx[0].tco_next_node[1]; } /* * Preallocate connections. Assume that thread 0 won't * use preallocated threads when running multi-core */ if ((thread > 0 || num_threads == 1) && prealloc_conn_per_wrk) pool_init_fixed (wrk->connections, prealloc_conn_per_wrk); } /* * Use a preallocated half-open connection pool? */ if (tcp_cfg.preallocated_half_open_connections) pool_init_fixed (tm->half_open_connections, tcp_cfg.preallocated_half_open_connections); if (num_threads > 1) { clib_spinlock_init (&tm->half_open_lock); } tcp_initialize_timer_wheels (tm); tcp_initialize_iss_seed (tm); tm->bytes_per_buffer = vlib_buffer_get_default_data_size (vm); tm->cc_last_type = TCP_CC_LAST; tm->ipl_next_node[0] = vlib_node_get_next (vm, session_queue_node.index, ip4_lookup_node.index); tm->ipl_next_node[1] = vlib_node_get_next (vm, session_queue_node.index, ip6_lookup_node.index); return error; } clib_error_t * vnet_tcp_enable_disable (vlib_main_t * vm, u8 is_en) { if (is_en) { if (tcp_main.is_enabled) return 0; return tcp_main_enable (vm); } else { tcp_main.is_enabled = 0; } return 0; } void tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add) { tcp_main_t *tm = &tcp_main; if (is_ip4) tm->punt_unknown4 = is_add; else tm->punt_unknown6 = is_add; } /** * Initialize default values for tcp parameters */ static void tcp_configuration_init (void) { /* Initial wnd for SYN. Fifos are not allocated at that point so use some * predefined value. For SYN-ACK we still want the scale to be computed in * the same way */ tcp_cfg.max_rx_fifo = 32 << 20; tcp_cfg.min_rx_fifo = 4 << 10; tcp_cfg.default_mtu = 1500; tcp_cfg.initial_cwnd_multiplier = 0; tcp_cfg.enable_tx_pacing = 1; tcp_cfg.allow_tso = 0; tcp_cfg.csum_offload = 1; tcp_cfg.cc_algo = TCP_CC_CUBIC; tcp_cfg.rwnd_min_update_ack = 1; /* Time constants defined as timer tick (100ms) multiples */ tcp_cfg.delack_time = 1; /* 0.1s */ tcp_cfg.closewait_time = 20; /* 2s */ tcp_cfg.timewait_time = 100; /* 10s */ tcp_cfg.finwait1_time = 600; /* 60s */ tcp_cfg.lastack_time = 300; /* 30s */ tcp_cfg.finwait2_time = 300; /* 30s */ tcp_cfg.closing_time = 300; /* 30s */ tcp_cfg.cleanup_time = 0.1; /* 100ms */ } static clib_error_t * tcp_init (vlib_main_t * vm) { tcp_main_t *tm = vnet_get_tcp_main (); ip_main_t *im = &ip_main; ip_protocol_info_t *pi; /* Session layer, and by implication tcp, are disabled by default */ tm->is_enabled = 0; /* Register with IP for header parsing */ pi = ip_get_protocol_info (im, IP_PROTOCOL_TCP); if (pi == 0) return clib_error_return (0, "TCP protocol info AWOL"); pi->format_header = format_tcp_header; pi->unformat_pg_edit = unformat_pg_tcp_header; /* Register as transport with session layer */ transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto, FIB_PROTOCOL_IP4, tcp4_output_node.index); transport_register_protocol (TRANSPORT_PROTO_TCP, &tcp_proto, FIB_PROTOCOL_IP6, tcp6_output_node.index); tcp_configuration_init (); tm->cc_algo_by_name = hash_create_string (0, sizeof (uword)); return 0; } VLIB_INIT_FUNCTION (tcp_init); /* * fd.io coding-style-patch-verification: ON * * Local Variables: * eval: (c-set-style "gnu") * End: */