From 68b0fb0c620c7451ef1a6380c43c39de6614db51 Mon Sep 17 00:00:00 2001 From: Dave Barach Date: Tue, 28 Feb 2017 15:15:56 -0500 Subject: VPP-598: tcp stack initial commit Change-Id: I49e5ce0aae6e4ff634024387ceaf7dbc432a0351 Signed-off-by: Dave Barach Signed-off-by: Florin Coras --- src/vnet/tcp/tcp.c | 708 ++++++++++++ src/vnet/tcp/tcp.h | 624 +++++++++++ src/vnet/tcp/tcp_error.def | 35 + src/vnet/tcp/tcp_format.c | 136 +++ src/vnet/tcp/tcp_input.c | 2316 ++++++++++++++++++++++++++++++++++++++++ src/vnet/tcp/tcp_newreno.c | 93 ++ src/vnet/tcp/tcp_output.c | 1412 ++++++++++++++++++++++++ src/vnet/tcp/tcp_packet.h | 184 ++++ src/vnet/tcp/tcp_pg.c | 236 ++++ src/vnet/tcp/tcp_syn_filter4.c | 542 ++++++++++ src/vnet/tcp/tcp_timer.h | 29 + 11 files changed, 6315 insertions(+) create mode 100644 src/vnet/tcp/tcp.c create mode 100644 src/vnet/tcp/tcp.h create mode 100644 src/vnet/tcp/tcp_error.def create mode 100644 src/vnet/tcp/tcp_format.c create mode 100644 src/vnet/tcp/tcp_input.c create mode 100644 src/vnet/tcp/tcp_newreno.c create mode 100644 src/vnet/tcp/tcp_output.c create mode 100644 src/vnet/tcp/tcp_packet.h create mode 100644 src/vnet/tcp/tcp_pg.c create mode 100644 src/vnet/tcp/tcp_syn_filter4.c create mode 100644 src/vnet/tcp/tcp_timer.h (limited to 'src/vnet/tcp') diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c new file mode 100644 index 00000000000..0f9b7097b42 --- /dev/null +++ b/src/vnet/tcp/tcp.c @@ -0,0 +1,708 @@ +/* + * 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. + */ + +#include +#include +#include +#include + +tcp_main_t tcp_main; + +static u32 +tcp_connection_bind (vlib_main_t * vm, u32 session_index, ip46_address_t * ip, + u16 port_host_byte_order, u8 is_ip4) +{ + tcp_main_t *tm = &tcp_main; + tcp_connection_t *listener; + + pool_get (tm->listener_pool, listener); + memset (listener, 0, sizeof (*listener)); + + listener->c_c_index = listener - tm->listener_pool; + listener->c_lcl_port = clib_host_to_net_u16 (port_host_byte_order); + + if (is_ip4) + listener->c_lcl_ip4.as_u32 = ip->ip4.as_u32; + else + clib_memcpy (&listener->c_lcl_ip6, &ip->ip6, sizeof (ip6_address_t)); + + listener->c_s_index = session_index; + listener->c_proto = SESSION_TYPE_IP4_TCP; + listener->state = TCP_STATE_LISTEN; + listener->c_is_ip4 = 1; + + return listener->c_c_index; +} + +u32 +tcp_session_bind_ip4 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_host_byte_order) +{ + return tcp_connection_bind (vm, session_index, ip, port_host_byte_order, 1); +} + +u32 +tcp_session_bind_ip6 (vlib_main_t * vm, u32 session_index, + ip46_address_t * ip, u16 port_host_byte_order) +{ + return tcp_connection_bind (vm, session_index, ip, port_host_byte_order, 0); + +} + +static void +tcp_session_unbind (u32 listener_index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + pool_put_index (tm->listener_pool, listener_index); +} + +u32 +tcp_session_unbind_ip4 (vlib_main_t * vm, u32 listener_index) +{ + tcp_session_unbind (listener_index); + return 0; +} + +u32 +tcp_session_unbind_ip6 (vlib_main_t * vm, u32 listener_index) +{ + tcp_session_unbind (listener_index); + return 0; +} + +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; +} + +/** + * Cleans up connection state. + * + * No notifications. + */ +void +tcp_connection_cleanup (tcp_connection_t * tc) +{ + tcp_main_t *tm = &tcp_main; + u32 tepi; + transport_endpoint_t *tep; + + /* Cleanup local endpoint if this was an active connect */ + tepi = transport_endpoint_lookup (&tm->local_endpoints_table, &tc->c_lcl_ip, + tc->c_lcl_port); + + /*XXX lock */ + if (tepi != TRANSPORT_ENDPOINT_INVALID_INDEX) + { + tep = pool_elt_at_index (tm->local_endpoints, tepi); + transport_endpoint_table_del (&tm->local_endpoints_table, tep); + pool_put (tm->local_endpoints, tep); + } + + /* Make sure all timers are cleared */ + tcp_connection_timers_reset (tc); + + /* Check if half-open */ + if (tc->state == TCP_STATE_SYN_SENT) + pool_put (tm->half_open_connections, tc); + else + pool_put (tm->connections[tc->c_thread_index], 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) +{ + stream_session_delete_notify (&tc->connection); + tcp_connection_cleanup (tc); +} + +/** + * Begin connection closing procedure. + * + * If at the end the connection is not in CLOSED state, it is not removed. + * Instead, we rely on on TCP to advance through state machine to either + * 1) LAST_ACK (passive close) whereby when the last ACK is received + * tcp_connection_del is called. This notifies session of the delete and + * calls cleanup. + * 2) TIME_WAIT (active close) whereby after 2MSL the 2MSL timer triggers + * and cleanup is called. + */ +void +tcp_connection_close (tcp_connection_t * tc) +{ + /* Send FIN if needed */ + if (tc->state == TCP_STATE_ESTABLISHED || tc->state == TCP_STATE_SYN_RCVD + || tc->state == TCP_STATE_CLOSE_WAIT) + tcp_send_fin (tc); + + /* Switch state */ + if (tc->state == TCP_STATE_ESTABLISHED || tc->state == TCP_STATE_SYN_RCVD) + tc->state = TCP_STATE_FIN_WAIT_1; + else if (tc->state == TCP_STATE_SYN_SENT) + tc->state = TCP_STATE_CLOSED; + else if (tc->state == TCP_STATE_CLOSE_WAIT) + tc->state = TCP_STATE_LAST_ACK; + + /* Half-close connections are not supported XXX */ + + if (tc->state == TCP_STATE_CLOSED) + tcp_connection_del (tc); +} + +void +tcp_session_close (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc; + tc = tcp_connection_get (conn_index, thread_index); + tcp_connection_close (tc); +} + +void +tcp_session_cleanup (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc; + tc = tcp_connection_get (conn_index, thread_index); + tcp_connection_cleanup (tc); +} + +void * +ip_interface_get_first_ip (u32 sw_if_index, u8 is_ip4) +{ + ip_lookup_main_t *lm4 = &ip4_main.lookup_main; + ip_lookup_main_t *lm6 = &ip6_main.lookup_main; + ip_interface_address_t *ia = 0; + + if (is_ip4) + { + /* *INDENT-OFF* */ + foreach_ip_interface_address (lm4, ia, sw_if_index, 1 /* unnumbered */ , + ({ + return ip_interface_address_get_address (lm4, ia); + })); + /* *INDENT-ON* */ + } + else + { + /* *INDENT-OFF* */ + foreach_ip_interface_address (lm6, ia, sw_if_index, 1 /* unnumbered */ , + ({ + return ip_interface_address_get_address (lm6, ia); + })); + /* *INDENT-ON* */ + } + + return 0; +} + +/** + * Allocate local port and add if successful add entry to local endpoint + * table to mark the pair as used. + */ +u16 +tcp_allocate_local_port (tcp_main_t * tm, ip46_address_t * ip) +{ + u8 unique = 0; + transport_endpoint_t *tep; + u32 time_now, tei; + u16 min = 1024, max = 65535, tries; /* XXX configurable ? */ + + tries = max - min; + time_now = tcp_time_now (); + + /* Start at random point or max */ + pool_get (tm->local_endpoints, tep); + clib_memcpy (&tep->ip, ip, sizeof (*ip)); + tep->port = random_u32 (&time_now) << 16; + tep->port = tep->port < min ? max : tep->port; + + /* Search for first free slot */ + while (tries) + { + tei = transport_endpoint_lookup (&tm->local_endpoints_table, &tep->ip, + tep->port); + if (tei == TRANSPORT_ENDPOINT_INVALID_INDEX) + { + unique = 1; + break; + } + + tep->port--; + + if (tep->port < min) + tep->port = max; + + tries--; + } + + if (unique) + { + transport_endpoint_table_add (&tm->local_endpoints_table, tep, + tep - tm->local_endpoints); + + return tep->port; + } + + /* Failed */ + pool_put (tm->local_endpoints, tep); + return -1; +} + +/** + * Initialize all connection timers as invalid + */ +void +tcp_connection_timers_init (tcp_connection_t * tc) +{ + int i; + + /* Set all to invalid */ + for (i = 0; i < TCP_N_TIMERS; i++) + { + tc->timers[i] = TCP_TIMER_HANDLE_INVALID; + } + + tc->rto = TCP_RTO_INIT; +} + +/** + * Stop all connection timers + */ +void +tcp_connection_timers_reset (tcp_connection_t * tc) +{ + int i; + for (i = 0; i < TCP_N_TIMERS; i++) + { + tcp_timer_reset (tc, i); + } +} + +/** Initialize tcp connection variables + * + * Should be called after having received a msg from the peer, i.e., a SYN or + * a SYNACK, such that connection options have already been exchanged. */ +void +tcp_connection_init_vars (tcp_connection_t * tc) +{ + tcp_connection_timers_init (tc); + tcp_set_snd_mss (tc); + tc->sack_sb.head = TCP_INVALID_SACK_HOLE_INDEX; + tcp_cc_init (tc); +} + +int +tcp_connection_open (ip46_address_t * rmt_addr, u16 rmt_port, u8 is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + tcp_connection_t *tc; + fib_prefix_t prefix; + u32 fei, sw_if_index; + ip46_address_t lcl_addr; + u16 lcl_port; + + /* + * Find the local address and allocate port + */ + memset (&lcl_addr, 0, sizeof (lcl_addr)); + + /* Find a FIB path to the destination */ + clib_memcpy (&prefix.fp_addr, rmt_addr, sizeof (*rmt_addr)); + prefix.fp_proto = is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6; + prefix.fp_len = is_ip4 ? 32 : 128; + + fei = fib_table_lookup (0, &prefix); + + /* Couldn't find route to destination. Bail out. */ + if (fei == FIB_NODE_INDEX_INVALID) + return -1; + + sw_if_index = fib_entry_get_resolving_interface (fei); + + if (sw_if_index == (u32) ~ 0) + return -1; + + if (is_ip4) + { + ip4_address_t *ip4; + ip4 = ip_interface_get_first_ip (sw_if_index, 1); + lcl_addr.ip4.as_u32 = ip4->as_u32; + } + else + { + ip6_address_t *ip6; + ip6 = ip_interface_get_first_ip (sw_if_index, 0); + clib_memcpy (&lcl_addr.ip6, ip6, sizeof (*ip6)); + } + + /* Allocate source port */ + lcl_port = tcp_allocate_local_port (tm, &lcl_addr); + if (lcl_port < 1) + return -1; + + /* + * Create connection and send SYN + */ + + pool_get (tm->half_open_connections, tc); + memset (tc, 0, sizeof (*tc)); + + clib_memcpy (&tc->c_rmt_ip, rmt_addr, sizeof (ip46_address_t)); + clib_memcpy (&tc->c_lcl_ip, &lcl_addr, sizeof (ip46_address_t)); + tc->c_rmt_port = clib_host_to_net_u16 (rmt_port); + tc->c_lcl_port = clib_host_to_net_u16 (lcl_port); + tc->c_c_index = tc - tm->half_open_connections; + tc->c_is_ip4 = is_ip4; + + /* The other connection vars will be initialized after SYN ACK */ + tcp_connection_timers_init (tc); + + tcp_send_syn (tc); + + tc->state = TCP_STATE_SYN_SENT; + + return tc->c_c_index; +} + +int +tcp_session_open_ip4 (ip46_address_t * addr, u16 port) +{ + return tcp_connection_open (addr, port, 1); +} + +int +tcp_session_open_ip6 (ip46_address_t * addr, u16 port) +{ + return tcp_connection_open (addr, port, 0); +} + +u8 * +format_tcp_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + tcp_connection_t *tc; + + tc = tcp_connection_get (tci, thread_index); + + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + + return s; +} + +u8 * +format_tcp_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + tcp_connection_t *tc = tcp_connection_get (tci, thread_index); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_listener_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_listener_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_listener_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_half_open_session_ip4 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_half_open_connection_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip4_address, + &tc->c_lcl_ip4, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip4_address, &tc->c_rmt_ip4, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +u8 * +format_tcp_half_open_session_ip6 (u8 * s, va_list * args) +{ + u32 tci = va_arg (*args, u32); + tcp_connection_t *tc = tcp_half_open_connection_get (tci); + s = format (s, "[%s] %U:%d->%U:%d", "tcp", format_ip6_address, + &tc->c_lcl_ip6, clib_net_to_host_u16 (tc->c_lcl_port), + format_ip6_address, &tc->c_rmt_ip6, + clib_net_to_host_u16 (tc->c_rmt_port)); + return s; +} + +transport_connection_t * +tcp_session_get_transport (u32 conn_index, u32 thread_index) +{ + tcp_connection_t *tc = tcp_connection_get (conn_index, thread_index); + return &tc->connection; +} + +transport_connection_t * +tcp_half_open_session_get_transport (u32 conn_index) +{ + tcp_connection_t *tc = tcp_half_open_connection_get (conn_index); + return &tc->connection; +} + +u16 +tcp_session_send_mss (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return tc->snd_mss; +} + +u32 +tcp_session_send_space (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return tcp_available_snd_space (tc); +} + +u32 +tcp_session_rx_fifo_offset (transport_connection_t * trans_conn) +{ + tcp_connection_t *tc = (tcp_connection_t *) trans_conn; + return (tc->snd_una_max - tc->snd_una); +} + +/* *INDENT-OFF* */ +const static transport_proto_vft_t tcp4_proto = { + .bind = tcp_session_bind_ip4, + .unbind = tcp_session_unbind_ip4, + .push_header = tcp_push_header, + .get_connection = tcp_session_get_transport, + .get_listener = tcp_session_get_listener, + .get_half_open = tcp_half_open_session_get_transport, + .open = tcp_session_open_ip4, + .close = tcp_session_close, + .cleanup = tcp_session_cleanup, + .send_mss = tcp_session_send_mss, + .send_space = tcp_session_send_space, + .rx_fifo_offset = tcp_session_rx_fifo_offset, + .format_connection = format_tcp_session_ip4, + .format_listener = format_tcp_listener_session_ip4, + .format_half_open = format_tcp_half_open_session_ip4 +}; + +const static transport_proto_vft_t tcp6_proto = { + .bind = tcp_session_bind_ip6, + .unbind = tcp_session_unbind_ip6, + .push_header = tcp_push_header, + .get_connection = tcp_session_get_transport, + .get_listener = tcp_session_get_listener, + .get_half_open = tcp_half_open_session_get_transport, + .open = tcp_session_open_ip6, + .close = tcp_session_close, + .cleanup = tcp_session_cleanup, + .send_mss = tcp_session_send_mss, + .send_space = tcp_session_send_space, + .rx_fifo_offset = tcp_session_rx_fifo_offset, + .format_connection = format_tcp_session_ip6, + .format_listener = format_tcp_listener_session_ip6, + .format_half_open = format_tcp_half_open_session_ip6 +}; +/* *INDENT-ON* */ + +void +tcp_timer_keep_handler (u32 conn_index) +{ + u32 cpu_index = os_get_cpu_number (); + tcp_connection_t *tc; + + tc = tcp_connection_get (conn_index, cpu_index); + tc->timers[TCP_TIMER_KEEP] = TCP_TIMER_HANDLE_INVALID; + + tcp_connection_close (tc); +} + +void +tcp_timer_establish_handler (u32 conn_index) +{ + tcp_connection_t *tc; + u8 sst; + + tc = tcp_half_open_connection_get (conn_index); + tc->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID; + + ASSERT (tc->state == TCP_STATE_SYN_SENT); + + sst = tc->c_is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + stream_session_connect_notify (&tc->connection, sst, 1 /* fail */ ); + + tcp_connection_cleanup (tc); +} + +void +tcp_timer_2msl_handler (u32 conn_index) +{ + u32 cpu_index = os_get_cpu_number (); + tcp_connection_t *tc; + + tc = tcp_connection_get (conn_index, cpu_index); + tc->timers[TCP_TIMER_2MSL] = TCP_TIMER_HANDLE_INVALID; + + tcp_connection_del (tc); +} + +/* *INDENT-OFF* */ +static timer_expiration_handler *timer_expiration_handlers[TCP_N_TIMERS] = +{ + tcp_timer_retransmit_handler, + tcp_timer_delack_handler, + 0, + tcp_timer_keep_handler, + tcp_timer_2msl_handler, + tcp_timer_retransmit_syn_handler, + tcp_timer_establish_handler +}; +/* *INDENT-ON* */ + +static void +tcp_expired_timers_dispatch (u32 * expired_timers) +{ + int i; + u32 connection_index, timer_id; + + for (i = 0; i < vec_len (expired_timers); i++) + { + /* Get session index and timer id */ + connection_index = expired_timers[i] & 0x0FFFFFFF; + timer_id = expired_timers[i] >> 28; + + /* Handle expiration */ + (*timer_expiration_handlers[timer_id]) (connection_index); + } +} + +void +tcp_initialize_timer_wheels (tcp_main_t * tm) +{ + tw_timer_wheel_16t_2w_512sl_t *tw; + vec_foreach (tw, tm->timer_wheels) + { + tw_timer_wheel_init_16t_2w_512sl (tw, tcp_expired_timers_dispatch, + 100e-3 /* timer period 100ms */ , ~0); + tw->last_run_time = vlib_time_now (tm->vlib_main); + } +} + +clib_error_t * +tcp_init (vlib_main_t * vm) +{ + ip_main_t *im = &ip_main; + ip_protocol_info_t *pi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_thread_main_t *vtm = vlib_get_thread_main (); + clib_error_t *error = 0; + u32 num_threads; + + tm->vlib_main = vm; + tm->vnet_main = vnet_get_main (); + + 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 + */ + + /* Register with IP */ + 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; + + ip4_register_protocol (IP_PROTOCOL_TCP, tcp4_input_node.index); + + /* Register as transport with URI */ + session_register_transport (SESSION_TYPE_IP4_TCP, &tcp4_proto); + session_register_transport (SESSION_TYPE_IP6_TCP, &tcp6_proto); + + /* + * Initialize data structures + */ + + num_threads = 1 /* main thread */ + vtm->n_threads; + vec_validate (tm->connections, num_threads - 1); + + /* Initialize per worker thread tx buffers (used for control messages) */ + vec_validate (tm->tx_buffers, num_threads - 1); + + /* Initialize timer wheels */ + vec_validate (tm->timer_wheels, num_threads - 1); + tcp_initialize_timer_wheels (tm); + + vec_validate (tm->delack_connections, num_threads - 1); + + /* Initialize clocks per tick for TCP timestamp. Used to compute + * monotonically increasing timestamps. */ + tm->tstamp_ticks_per_clock = vm->clib_time.seconds_per_clock + / TCP_TSTAMP_RESOLUTION; + + clib_bihash_init_24_8 (&tm->local_endpoints_table, "local endpoint table", + 200000 /* $$$$ config parameter nbuckets */ , + (64 << 20) /*$$$ config parameter table size */ ); + + return error; +} + +VLIB_INIT_FUNCTION (tcp_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h new file mode 100644 index 00000000000..22f00a63273 --- /dev/null +++ b/src/vnet/tcp/tcp.h @@ -0,0 +1,624 @@ +/* + * 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. + */ + +#ifndef _vnet_tcp_h_ +#define _vnet_tcp_h_ + +#include +#include +#include +#include +#include +#include + +#define TCP_TICK 10e-3 /**< TCP tick period (s) */ +#define THZ 1/TCP_TICK /**< TCP tick frequency */ +#define TCP_TSTAMP_RESOLUTION TCP_TICK /**< Time stamp resolution */ +#define TCP_PAWS_IDLE 24 * 24 * 60 * 60 * THZ /**< 24 days */ +#define TCP_MAX_OPTION_SPACE 40 + +#define TCP_DUPACK_THRESHOLD 3 +#define TCP_DEFAULT_RX_FIFO_SIZE 64 << 10 + +/** TCP FSM state definitions as per RFC793. */ +#define foreach_tcp_fsm_state \ + _(CLOSED, "CLOSED") \ + _(LISTEN, "LISTEN") \ + _(SYN_SENT, "SYN_SENT") \ + _(SYN_RCVD, "SYN_RCVD") \ + _(ESTABLISHED, "ESTABLISHED") \ + _(CLOSE_WAIT, "CLOSE_WAIT") \ + _(FIN_WAIT_1, "FIN_WAIT_1") \ + _(LAST_ACK, "LAST_ACK") \ + _(CLOSING, "CLOSING") \ + _(FIN_WAIT_2, "FIN_WAIT_2") \ + _(TIME_WAIT, "TIME_WAIT") + +typedef enum _tcp_state +{ +#define _(sym, str) TCP_STATE_##sym, + foreach_tcp_fsm_state +#undef _ + TCP_N_STATES +} tcp_state_t; + +format_function_t format_tcp_state; + +/** TCP timers */ +#define foreach_tcp_timer \ + _(RETRANSMIT, "RETRANSMIT") \ + _(DELACK, "DELAYED ACK") \ + _(PERSIST, "PERSIST") \ + _(KEEP, "KEEP") \ + _(2MSL, "2MSL") \ + _(RETRANSMIT_SYN, "RETRANSMIT_SYN") \ + _(ESTABLISH, "ESTABLISH") + +typedef enum _tcp_timers +{ +#define _(sym, str) TCP_TIMER_##sym, + foreach_tcp_timer +#undef _ + TCP_N_TIMERS +} tcp_timers_e; + +typedef void (timer_expiration_handler) (u32 index); + +extern timer_expiration_handler tcp_timer_delack_handler; +extern timer_expiration_handler tcp_timer_retransmit_handler; +extern timer_expiration_handler tcp_timer_retransmit_syn_handler; + +#define TCP_TIMER_HANDLE_INVALID ((u32) ~0) + +/* Timer delays as multiples of 100ms */ +#define TCP_TO_TIMER_TICK TCP_TICK*10 /* Period for converting from TCP + * ticks to timer units */ +#define TCP_DELACK_TIME 1 /* 0.1s */ +#define TCP_ESTABLISH_TIME 750 /* 75s */ +#define TCP_2MSL_TIME 300 /* 30s */ + +#define TCP_RTO_MAX 60 * THZ /* Min max RTO (60s) as per RFC6298 */ +#define TCP_RTT_MAX 30 * THZ /* 30s (probably too much) */ +#define TCP_RTO_SYN_RETRIES 3 /* SYN retries without doubling RTO */ +#define TCP_RTO_INIT 1 * THZ /* Initial retransmit timer */ + +void tcp_update_time (f64 now, u32 thread_index); + +/** TCP connection flags */ +#define foreach_tcp_connection_flag \ + _(DELACK, "Delay ACK") \ + _(SNDACK, "Send ACK") \ + _(BURSTACK, "Burst ACK set") \ + _(SENT_RCV_WND0, "Sent 0 receive window") \ + _(RECOVERY, "Recovery on") \ + _(FAST_RECOVERY, "Fast Recovery on") + +typedef enum _tcp_connection_flag_bits +{ +#define _(sym, str) TCP_CONN_##sym##_BIT, + foreach_tcp_connection_flag +#undef _ + TCP_CONN_N_FLAG_BITS +} tcp_connection_flag_bits_e; + +typedef enum _tcp_connection_flag +{ +#define _(sym, str) TCP_CONN_##sym = 1 << TCP_CONN_##sym##_BIT, + foreach_tcp_connection_flag +#undef _ + TCP_CONN_N_FLAGS +} tcp_connection_flags_e; + +/** TCP buffer flags */ +#define foreach_tcp_buf_flag \ + _ (ACK) /**< Sending ACK. */ \ + _ (DUPACK) /**< Sending DUPACK. */ \ + +enum +{ +#define _(f) TCP_BUF_BIT_##f, + foreach_tcp_buf_flag +#undef _ + TCP_N_BUF_BITS, +}; + +enum +{ +#define _(f) TCP_BUF_FLAG_##f = 1 << TCP_BUF_BIT_##f, + foreach_tcp_buf_flag +#undef _ +}; + +#define TCP_MAX_SACK_BLOCKS 5 /**< Max number of SACK blocks stored */ +#define TCP_INVALID_SACK_HOLE_INDEX ((u32)~0) + +typedef struct _sack_scoreboard_hole +{ + u32 next; /**< Index for next entry in linked list */ + u32 prev; /**< Index for previous entry in linked list */ + u32 start; /**< Start sequence number */ + u32 end; /**< End sequence number */ +} sack_scoreboard_hole_t; + +typedef struct _sack_scoreboard +{ + sack_scoreboard_hole_t *holes; /**< Pool of holes */ + u32 head; /**< Index to first entry */ + u32 sacked_bytes; /**< Number of bytes sacked in sb */ +} sack_scoreboard_t; + +typedef enum _tcp_cc_algorithm_type +{ + TCP_CC_NEWRENO, +} tcp_cc_algorithm_type_e; + +typedef struct _tcp_cc_algorithm tcp_cc_algorithm_t; + +typedef enum _tcp_cc_ack_t +{ + TCP_CC_ACK, + TCP_CC_DUPACK, + TCP_CC_PARTIALACK +} tcp_cc_ack_t; + +typedef struct _tcp_connection +{ + transport_connection_t connection; /**< Common transport data. First! */ + + u8 state; /**< TCP state as per tcp_state_t */ + u16 flags; /**< Connection flags (see tcp_conn_flags_e) */ + u32 timers[TCP_N_TIMERS]; /**< Timer handles into timer wheel */ + + /* TODO RFC4898 */ + + /** Send sequence variables RFC793 */ + u32 snd_una; /**< oldest unacknowledged sequence number */ + u32 snd_una_max; /**< newest unacknowledged sequence number + 1*/ + u32 snd_wnd; /**< send window */ + u32 snd_wl1; /**< seq number used for last snd.wnd update */ + u32 snd_wl2; /**< ack number used for last snd.wnd update */ + u32 snd_nxt; /**< next seq number to be sent */ + + /** Receive sequence variables RFC793 */ + u32 rcv_nxt; /**< next sequence number expected */ + u32 rcv_wnd; /**< receive window we expect */ + + u32 rcv_las; /**< rcv_nxt at last ack sent/rcv_wnd update */ + u32 iss; /**< initial sent sequence */ + u32 irs; /**< initial remote sequence */ + + /* Options */ + tcp_options_t opt; /**< TCP connection options parsed */ + u8 rcv_wscale; /**< Window scale to advertise to peer */ + u8 snd_wscale; /**< Window scale to use when sending */ + u32 tsval_recent; /**< Last timestamp received */ + u32 tsval_recent_age; /**< When last updated tstamp_recent*/ + + sack_block_t *snd_sacks; /**< Vector of SACKs to send. XXX Fixed size? */ + sack_scoreboard_t sack_sb; /**< SACK "scoreboard" that tracks holes */ + + u8 rcv_dupacks; /**< Number of DUPACKs received */ + u8 snt_dupacks; /**< Number of DUPACKs sent in a burst */ + + /* Congestion control */ + u32 cwnd; /**< Congestion window */ + u32 ssthresh; /**< Slow-start threshold */ + u32 prev_ssthresh; /**< ssthresh before congestion */ + u32 bytes_acked; /**< Bytes acknowledged by current segment */ + u32 rtx_bytes; /**< Retransmitted bytes */ + u32 tsecr_last_ack; /**< Timestamp echoed to us in last health ACK */ + tcp_cc_algorithm_t *cc_algo; /**< Congestion control algorithm */ + + /* RTT and RTO */ + u32 rto; /**< Retransmission timeout */ + u32 rto_boff; /**< Index for RTO backoff */ + u32 srtt; /**< Smoothed RTT */ + u32 rttvar; /**< Smoothed mean RTT difference. Approximates variance */ + u32 rtt_ts; /**< Timestamp for tracked ACK */ + u32 rtt_seq; /**< Sequence number for tracked ACK */ + + u16 snd_mss; /**< Send MSS */ +} tcp_connection_t; + +struct _tcp_cc_algorithm +{ + void (*rcv_ack) (tcp_connection_t * tc); + void (*rcv_cong_ack) (tcp_connection_t * tc, tcp_cc_ack_t ack); + void (*congestion) (tcp_connection_t * tc); + void (*recovered) (tcp_connection_t * tc); + void (*init) (tcp_connection_t * tc); +}; + +#define tcp_fastrecovery_on(tc) (tc)->flags |= TCP_CONN_FAST_RECOVERY +#define tcp_fastrecovery_off(tc) (tc)->flags &= ~TCP_CONN_FAST_RECOVERY +#define tcp_in_fastrecovery(tc) ((tc)->flags & TCP_CONN_FAST_RECOVERY) +#define tcp_in_recovery(tc) ((tc)->flags & (TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY)) +#define tcp_recovery_off(tc) ((tc)->flags &= ~(TCP_CONN_FAST_RECOVERY | TCP_CONN_RECOVERY)) +#define tcp_in_slowstart(tc) (tc->cwnd < tc->ssthresh) + +typedef enum +{ + TCP_IP4, + TCP_IP6, + TCP_N_AF, +} tcp_af_t; + +typedef enum _tcp_error +{ +#define tcp_error(n,s) TCP_ERROR_##n, +#include +#undef tcp_error + TCP_N_ERROR, +} tcp_error_t; + +typedef struct _tcp_lookup_dispatch +{ + u8 next, error; +} tcp_lookup_dispatch_t; + +typedef struct _tcp_main +{ + /* Per-worker thread tcp connection pools */ + tcp_connection_t **connections; + + /* Pool of listeners. */ + tcp_connection_t *listener_pool; + + /** Dispatch table by state and flags */ + tcp_lookup_dispatch_t dispatch_table[TCP_N_STATES][64]; + + u8 log2_tstamp_clocks_per_tick; + f64 tstamp_ticks_per_clock; + + /** per-worker tx buffer free lists */ + u32 **tx_buffers; + + /* Per worker-thread timer wheel for connections timers */ + tw_timer_wheel_16t_2w_512sl_t *timer_wheels; + + /* Convenience per worker-thread vector of connections to DELACK */ + u32 **delack_connections; + + /* Pool of half-open connections on which we've sent a SYN */ + tcp_connection_t *half_open_connections; + + /* Pool of local TCP endpoints */ + transport_endpoint_t *local_endpoints; + + /* Local endpoints lookup table */ + transport_endpoint_table_t local_endpoints_table; + + /* Congestion control algorithms registered */ + tcp_cc_algorithm_t *cc_algos; + + /* convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + ip4_main_t *ip4_main; + ip6_main_t *ip6_main; +} tcp_main_t; + +extern tcp_main_t tcp_main; +extern vlib_node_registration_t tcp4_input_node; +extern vlib_node_registration_t tcp6_input_node; +extern vlib_node_registration_t tcp4_output_node; +extern vlib_node_registration_t tcp6_output_node; + +always_inline tcp_main_t * +vnet_get_tcp_main () +{ + return &tcp_main; +} + +always_inline tcp_connection_t * +tcp_connection_get (u32 conn_index, u32 thread_index) +{ + return pool_elt_at_index (tcp_main.connections[thread_index], conn_index); +} + +always_inline tcp_connection_t * +tcp_connection_get_if_valid (u32 conn_index, u32 thread_index) +{ + if (tcp_main.connections[thread_index] == 0) + return 0; + if (pool_is_free_index (tcp_main.connections[thread_index], conn_index)) + return 0; + return pool_elt_at_index (tcp_main.connections[thread_index], conn_index); +} + +void tcp_connection_close (tcp_connection_t * tc); +void tcp_connection_cleanup (tcp_connection_t * tc); +void tcp_connection_del (tcp_connection_t * tc); + +always_inline tcp_connection_t * +tcp_listener_get (u32 tli) +{ + return pool_elt_at_index (tcp_main.listener_pool, tli); +} + +always_inline tcp_connection_t * +tcp_half_open_connection_get (u32 conn_index) +{ + return pool_elt_at_index (tcp_main.half_open_connections, conn_index); +} + +void tcp_make_ack (tcp_connection_t * ts, vlib_buffer_t * b); +void tcp_make_finack (tcp_connection_t * tc, vlib_buffer_t * b); +void tcp_make_synack (tcp_connection_t * ts, vlib_buffer_t * b); +void tcp_send_reset (vlib_buffer_t * pkt, u8 is_ip4); +void tcp_send_syn (tcp_connection_t * tc); +void tcp_send_fin (tcp_connection_t * tc); +void tcp_set_snd_mss (tcp_connection_t * tc); + +always_inline u32 +tcp_end_seq (tcp_header_t * th, u32 len) +{ + return th->seq_number + tcp_is_syn (th) + tcp_is_fin (th) + len; +} + +/* Modulo arithmetic for TCP sequence numbers */ +#define seq_lt(_s1, _s2) ((i32)((_s1)-(_s2)) < 0) +#define seq_leq(_s1, _s2) ((i32)((_s1)-(_s2)) <= 0) +#define seq_gt(_s1, _s2) ((i32)((_s1)-(_s2)) > 0) +#define seq_geq(_s1, _s2) ((i32)((_s1)-(_s2)) >= 0) + +/* Modulo arithmetic for timestamps */ +#define timestamp_lt(_t1, _t2) ((i32)((_t1)-(_t2)) < 0) +#define timestamp_leq(_t1, _t2) ((i32)((_t1)-(_t2)) <= 0) + +always_inline u32 +tcp_flight_size (const tcp_connection_t * tc) +{ + return tc->snd_una_max - tc->snd_una - tc->sack_sb.sacked_bytes + + tc->rtx_bytes; +} + +/** + * Initial cwnd as per RFC5681 + */ +always_inline u32 +tcp_initial_cwnd (const tcp_connection_t * tc) +{ + if (tc->snd_mss > 2190) + return 2 * tc->snd_mss; + else if (tc->snd_mss > 1095) + return 3 * tc->snd_mss; + else + return 4 * tc->snd_mss; +} + +always_inline u32 +tcp_loss_wnd (const tcp_connection_t * tc) +{ + return tc->snd_mss; +} + +always_inline u32 +tcp_available_wnd (const tcp_connection_t * tc) +{ + return clib_min (tc->cwnd, tc->snd_wnd); +} + +always_inline u32 +tcp_available_snd_space (const tcp_connection_t * tc) +{ + u32 available_wnd = tcp_available_wnd (tc); + u32 flight_size = tcp_flight_size (tc); + + if (available_wnd <= flight_size) + return 0; + + return available_wnd - flight_size; +} + +void tcp_retransmit_first_unacked (tcp_connection_t * tc); + +void tcp_fast_retransmit (tcp_connection_t * tc); + +always_inline u32 +tcp_time_now (void) +{ + return clib_cpu_time_now () * tcp_main.tstamp_ticks_per_clock; +} + +u32 tcp_push_header (transport_connection_t * tconn, vlib_buffer_t * b); + +u32 +tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, + u32 max_bytes); + +void tcp_connection_timers_init (tcp_connection_t * tc); +void tcp_connection_timers_reset (tcp_connection_t * tc); + +void tcp_connection_init_vars (tcp_connection_t * tc); + +always_inline void +tcp_connection_force_ack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + /* Reset flags, make sure ack is sent */ + tc->flags = TCP_CONN_SNDACK; + vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK; +} + +always_inline void +tcp_timer_set (tcp_connection_t * tc, u8 timer_id, u32 interval) +{ + tc->timers[timer_id] + = tw_timer_start_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->c_c_index, timer_id, interval); +} + +always_inline void +tcp_retransmit_timer_set (tcp_main_t * tm, tcp_connection_t * tc) +{ + /* XXX Switch to faster TW */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT, + clib_max (tc->rto * TCP_TO_TIMER_TICK, 1)); +} + +always_inline void +tcp_timer_reset (tcp_connection_t * tc, u8 timer_id) +{ + if (tc->timers[timer_id] == TCP_TIMER_HANDLE_INVALID) + return; + + tw_timer_stop_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->timers[timer_id]); + tc->timers[timer_id] = TCP_TIMER_HANDLE_INVALID; +} + +always_inline void +tcp_timer_update (tcp_connection_t * tc, u8 timer_id, u32 interval) +{ + if (tc->timers[timer_id] != TCP_TIMER_HANDLE_INVALID) + tw_timer_stop_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->timers[timer_id]); + tc->timers[timer_id] = + tw_timer_start_16t_2w_512sl (&tcp_main.timer_wheels[tc->c_thread_index], + tc->c_c_index, timer_id, interval); +} + +always_inline u8 +tcp_timer_is_active (tcp_connection_t * tc, tcp_timers_e timer) +{ + return tc->timers[timer] != TCP_TIMER_HANDLE_INVALID; +} + +void +scoreboard_remove_hole (sack_scoreboard_t * sb, + sack_scoreboard_hole_t * hole); + +always_inline sack_scoreboard_hole_t * +scoreboard_next_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) +{ + if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) + return pool_elt_at_index (sb->holes, hole->next); + return 0; +} + +always_inline sack_scoreboard_hole_t * +scoreboard_first_hole (sack_scoreboard_t * sb) +{ + if (sb->head != TCP_INVALID_SACK_HOLE_INDEX) + return pool_elt_at_index (sb->holes, sb->head); + return 0; +} + +always_inline void +scoreboard_clear (sack_scoreboard_t * sb) +{ + sack_scoreboard_hole_t *hole = scoreboard_first_hole (sb); + while ((hole = scoreboard_first_hole (sb))) + { + scoreboard_remove_hole (sb, hole); + } +} + +always_inline u32 +scoreboard_hole_bytes (sack_scoreboard_hole_t * hole) +{ + return hole->end - hole->start; +} + +always_inline 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; +} + +always_inline 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]; +} + +void tcp_cc_init (tcp_connection_t * tc); + +/** + * Push TCP header to buffer + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param sp_net - source port net order + * @param dp_net - destination port net order + * @param seq - sequence number net order + * @param ack - ack number net order + * @param tcp_hdr_opts_len - header and options length in bytes + * @param flags - header flags + * @param wnd - window size + * + * @return - pointer to start of TCP header + */ +always_inline void * +vlib_buffer_push_tcp_net_order (vlib_buffer_t * b, u16 sp, u16 dp, u32 seq, + u32 ack, u8 tcp_hdr_opts_len, u8 flags, + u16 wnd) +{ + tcp_header_t *th; + + th = vlib_buffer_push_uninit (b, tcp_hdr_opts_len); + + th->src_port = sp; + th->dst_port = dp; + th->seq_number = seq; + th->ack_number = ack; + th->data_offset_and_reserved = (tcp_hdr_opts_len >> 2) << 4; + th->flags = flags; + th->window = wnd; + th->checksum = 0; + th->urgent_pointer = 0; + return th; +} + +/** + * Push TCP header to buffer + * + * @param vm - vlib_main + * @param b - buffer to write the header to + * @param sp_net - source port net order + * @param dp_net - destination port net order + * @param seq - sequence number host order + * @param ack - ack number host order + * @param tcp_hdr_opts_len - header and options length in bytes + * @param flags - header flags + * @param wnd - window size + * + * @return - pointer to start of TCP header + */ +always_inline void * +vlib_buffer_push_tcp (vlib_buffer_t * b, u16 sp_net, u16 dp_net, u32 seq, + u32 ack, u8 tcp_hdr_opts_len, u8 flags, u16 wnd) +{ + return vlib_buffer_push_tcp_net_order (b, sp_net, dp_net, + clib_host_to_net_u32 (seq), + clib_host_to_net_u32 (ack), + tcp_hdr_opts_len, flags, + clib_host_to_net_u16 (wnd)); +} + +#endif /* _vnet_tcp_h_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_error.def b/src/vnet/tcp/tcp_error.def new file mode 100644 index 00000000000..cff5ec13a70 --- /dev/null +++ b/src/vnet/tcp/tcp_error.def @@ -0,0 +1,35 @@ +/* + * 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. + */ + +tcp_error (NONE, "no error") +tcp_error (NO_LISTENER, "no listener for dst port") +tcp_error (LOOKUP_DROPS, "lookup drops") +tcp_error (DISPATCH, "Dispatch error") +tcp_error (ENQUEUED, "Packets pushed into rx fifo") +tcp_error (PURE_ACK, "Pure acks") +tcp_error (SYNS_RCVD, "SYNs received") +tcp_error (SYN_ACKS_RCVD, "SYN-ACKs received") +tcp_error (NOT_READY, "Session not ready for packets") +tcp_error (FIFO_FULL, "Packets dropped for lack of rx fifo space") +tcp_error (EVENT_FIFO_FULL, "Events not sent for lack of event fifo space") +tcp_error (API_QUEUE_FULL, "Sessions not created for lack of API queue space") +tcp_error (CREATE_SESSION_FAIL, "Sessions couldn't be allocated") +tcp_error (SEGMENT_INVALID, "Invalid segment") +tcp_error (ACK_INVALID, "Invalid ACK") +tcp_error (ACK_DUP, "Duplicate ACK") +tcp_error (ACK_OLD, "Old ACK") +tcp_error (PKTS_SENT, "Packets sent") +tcp_error (FILTERED_DUPACKS, "Filtered duplicate ACKs") +tcp_error (RST_SENT, "Resets sent") \ No newline at end of file diff --git a/src/vnet/tcp/tcp_format.c b/src/vnet/tcp/tcp_format.c new file mode 100644 index 00000000000..7136741d212 --- /dev/null +++ b/src/vnet/tcp/tcp_format.c @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2015 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. + */ +/* + * tcp/tcp_format.c: tcp formatting + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include +#include + +static u8 * +format_tcp_flags (u8 * s, va_list * args) +{ + int flags = va_arg (*args, int); + +#define _(f) if (flags & TCP_FLAG_##f) s = format (s, "%s, ", #f); + foreach_tcp_flag +#undef _ + return s; +} + +/* Format TCP header. */ +u8 * +format_tcp_header (u8 * s, va_list * args) +{ + tcp_header_t *tcp = va_arg (*args, tcp_header_t *); + u32 max_header_bytes = va_arg (*args, u32); + u32 header_bytes; + uword indent; + + /* Nothing to do. */ + if (max_header_bytes < sizeof (tcp[0])) + return format (s, "TCP header truncated"); + + indent = format_get_indent (s); + indent += 2; + header_bytes = tcp_header_bytes (tcp); + + s = format (s, "TCP: %d -> %d", clib_net_to_host_u16 (tcp->src), + clib_net_to_host_u16 (tcp->dst)); + + s = format (s, "\n%Useq. 0x%08x ack 0x%08x", format_white_space, indent, + clib_net_to_host_u32 (tcp->seq_number), + clib_net_to_host_u32 (tcp->ack_number)); + + s = format (s, "\n%Uflags %U, tcp header: %d bytes", format_white_space, + indent, format_tcp_flags, tcp->flags, header_bytes); + + s = format (s, "\n%Uwindow %d, checksum 0x%04x", format_white_space, indent, + clib_net_to_host_u16 (tcp->window), + clib_net_to_host_u16 (tcp->checksum)); + + +#if 0 + /* Format TCP options. */ + { + u8 *o; + u8 *option_start = (void *) (tcp + 1); + u8 *option_end = (void *) tcp + header_bytes; + + for (o = option_start; o < option_end;) + { + u32 length = o[1]; + switch (o[0]) + { + case TCP_OPTION_END: + length = 1; + o = option_end; + break; + + case TCP_OPTION_NOOP: + length = 1; + break; + + } + } + } +#endif + + /* Recurse into next protocol layer. */ + if (max_header_bytes != 0 && header_bytes < max_header_bytes) + { + ip_main_t *im = &ip_main; + tcp_udp_port_info_t *pi; + + pi = ip_get_tcp_udp_port_info (im, tcp->dst); + + if (pi && pi->format_header) + s = format (s, "\n%U%U", format_white_space, indent - 2, + pi->format_header, + /* next protocol header */ (void *) tcp + header_bytes, + max_header_bytes - header_bytes); + } + + return s; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c new file mode 100644 index 00000000000..daa0683b48e --- /dev/null +++ b/src/vnet/tcp/tcp_input.c @@ -0,0 +1,2316 @@ +/* + * 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. + */ + +#include +#include +#include +#include +#include + +static char *tcp_error_strings[] = { +#define tcp_error(n,s) s, +#include +#undef tcp_error +}; + +/* All TCP nodes have the same outgoing arcs */ +#define foreach_tcp_state_next \ + _ (DROP, "error-drop") \ + _ (TCP4_OUTPUT, "tcp4-output") \ + _ (TCP6_OUTPUT, "tcp6-output") + +typedef enum _tcp_established_next +{ +#define _(s,n) TCP_ESTABLISHED_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_ESTABLISHED_N_NEXT, +} tcp_established_next_t; + +typedef enum _tcp_rcv_process_next +{ +#define _(s,n) TCP_RCV_PROCESS_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_RCV_PROCESS_N_NEXT, +} tcp_rcv_process_next_t; + +typedef enum _tcp_syn_sent_next +{ +#define _(s,n) TCP_SYN_SENT_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_SYN_SENT_N_NEXT, +} tcp_syn_sent_next_t; + +typedef enum _tcp_listen_next +{ +#define _(s,n) TCP_LISTEN_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_LISTEN_N_NEXT, +} tcp_listen_next_t; + +/* Generic, state independent indices */ +typedef enum _tcp_state_next +{ +#define _(s,n) TCP_NEXT_##s, + foreach_tcp_state_next +#undef _ + TCP_STATE_N_NEXT, +} tcp_state_next_t; + +#define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \ + : TCP_NEXT_TCP6_OUTPUT) + +vlib_node_registration_t tcp4_established_node; +vlib_node_registration_t tcp6_established_node; + +/** + * Validate segment sequence number. As per RFC793: + * + * Segment Receive Test + * Length Window + * ------- ------- ------------------------------------------- + * 0 0 SEG.SEQ = RCV.NXT + * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND + * >0 0 not acceptable + * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND + * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND + * + * This ultimately consists in checking if segment falls within the window. + * The one important difference compared to RFC793 is that we use rcv_las, + * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the + * peer's reference when computing our receive window. + * + * This accepts only segments within the window. + */ +always_inline u8 +tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) +{ + return seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) + && seq_geq (seq, tc->rcv_nxt); +} + +void +tcp_options_parse (tcp_header_t * th, tcp_options_t * to) +{ + const u8 *data; + u8 opt_len, opts_len, kind; + int j; + sack_block_t b; + + opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t); + data = (const u8 *) (th + 1); + + /* Zero out all flags but those set in SYN */ + to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE); + + for (; opts_len > 0; opts_len -= opt_len, data += opt_len) + { + kind = data[0]; + + /* Get options length */ + if (kind == TCP_OPTION_EOL) + break; + else if (kind == TCP_OPTION_NOOP) + opt_len = 1; + else + { + /* broken options */ + if (opts_len < 2) + break; + opt_len = data[1]; + + /* weird option length */ + if (opt_len < 2 || opt_len > opts_len) + break; + } + + /* Parse options */ + switch (kind) + { + case TCP_OPTION_MSS: + if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th)) + { + to->flags |= TCP_OPTS_FLAG_MSS; + to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2)); + } + break; + case TCP_OPTION_WINDOW_SCALE: + if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th)) + { + to->flags |= TCP_OPTS_FLAG_WSCALE; + to->wscale = data[2]; + if (to->wscale > TCP_MAX_WND_SCALE) + { + clib_warning ("Illegal window scaling value: %d", + to->wscale); + to->wscale = TCP_MAX_WND_SCALE; + } + } + break; + case TCP_OPTION_TIMESTAMP: + if (opt_len == TCP_OPTION_LEN_TIMESTAMP) + { + to->flags |= TCP_OPTS_FLAG_TSTAMP; + to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2)); + to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6)); + } + break; + case TCP_OPTION_SACK_PERMITTED: + if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th)) + to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + break; + case TCP_OPTION_SACK_BLOCK: + /* If SACK permitted was not advertised or a SYN, break */ + if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th)) + break; + + /* If too short or not correctly formatted, break */ + if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK)) + break; + + to->flags |= TCP_OPTS_FLAG_SACK; + to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK; + vec_reset_length (to->sacks); + for (j = 0; j < to->n_sack_blocks; j++) + { + b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 4 * j)); + b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 4 * j)); + vec_add1 (to->sacks, b); + } + break; + default: + /* Nothing to see here */ + continue; + } + } +} + +always_inline int +tcp_segment_check_paws (tcp_connection_t * tc) +{ + /* XXX normally test for timestamp should be lt instead of leq, but for + * local testing this is not enough */ + return tcp_opts_tstamp (&tc->opt) && tc->tsval_recent + && timestamp_lt (tc->opt.tsval, tc->tsval_recent); +} + +/** + * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19 + * + * It first verifies if segment has a wrapped sequence number (PAWS) and then + * does the processing associated to the first four steps (ignoring security + * and precedence): sequence number, rst bit and syn bit checks. + * + * @return 0 if segments passes validation. + */ +static int +tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, + vlib_buffer_t * b0, tcp_header_t * th0, u32 * next0) +{ + u8 paws_failed; + + if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0))) + return -1; + + tcp_options_parse (th0, &tc0->opt); + + /* RFC1323: Check against wrapped sequence numbers (PAWS). If we have + * timestamp to echo and it's less than tsval_recent, drop segment + * but still send an ACK in order to retain TCP's mechanism for detecting + * and recovering from half-open connections */ + paws_failed = tcp_segment_check_paws (tc0); + if (paws_failed) + { + clib_warning ("paws failed"); + + /* If it just so happens that a segment updates tsval_recent for a + * segment over 24 days old, invalidate tsval_recent. */ + if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE, + tcp_time_now ())) + { + /* Age isn't reset until we get a valid tsval (bsd inspired) */ + tc0->tsval_recent = 0; + } + else + { + /* Drop after ack if not rst */ + if (!tcp_rst (th0)) + { + tcp_make_ack (tc0, b0); + *next0 = tcp_next_output (tc0->c_is_ip4); + return -1; + } + } + } + + /* 1st: check sequence number */ + if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number, + vnet_buffer (b0)->tcp.seq_end)) + { + if (!tcp_rst (th0)) + { + /* Send dup ack */ + tcp_make_ack (tc0, b0); + *next0 = tcp_next_output (tc0->c_is_ip4); + } + return -1; + } + + /* 2nd: check the RST bit */ + if (tcp_rst (th0)) + { + /* Notify session that connection has been reset. Switch + * state to closed and await for session to do the cleanup. */ + stream_session_reset_notify (&tc0->connection); + tc0->state = TCP_STATE_CLOSED; + return -1; + } + + /* 3rd: check security and precedence (skip) */ + + /* 4th: check the SYN bit */ + if (tcp_syn (th0)) + { + tcp_send_reset (b0, tc0->c_is_ip4); + return -1; + } + + /* If PAWS passed and segment in window, save timestamp */ + if (!paws_failed) + { + tc0->tsval_recent = tc0->opt.tsval; + tc0->tsval_recent_age = tcp_time_now (); + } + + return 0; +} + +always_inline int +tcp_rcv_ack_is_acceptable (tcp_connection_t * tc0, vlib_buffer_t * tb0) +{ + /* SND.UNA =< SEG.ACK =< SND.NXT */ + return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number) + && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt)); +} + +/** + * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298 + * + * Note that although the original article, srtt and rttvar are scaled + * to minimize round-off errors, here we don't. Instead, we rely on + * better precision time measurements. + * + * TODO support us rtt resolution + */ +static void +tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) +{ + int err; + + if (tc->srtt != 0) + { + err = mrtt - tc->srtt; + tc->srtt += err >> 3; + + /* XXX Drop in RTT results in RTTVAR increase and bigger RTO. + * The increase should be bound */ + tc->rttvar += (clib_abs (err) - tc->rttvar) >> 2; + } + else + { + /* First measurement. */ + tc->srtt = mrtt; + tc->rttvar = mrtt << 1; + } +} + +/** Update RTT estimate and RTO timer + * + * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK + * timing. Middle boxes are known to fiddle with TCP options so we + * should give higher priority to ACK timing. + * + * return 1 if valid rtt 0 otherwise + */ +static int +tcp_update_rtt (tcp_connection_t * tc, u32 ack) +{ + u32 mrtt = 0; + + /* Karn's rule, part 1. Don't use retransmitted segments to estimate + * RTT because they're ambiguous. */ + if (tc->rtt_seq && seq_gt (ack, tc->rtt_seq) && !tc->rto_boff) + { + mrtt = tcp_time_now () - tc->rtt_ts; + tc->rtt_seq = 0; + } + + /* As per RFC7323 TSecr can be used for RTTM only if the segment advances + * snd_una, i.e., the left side of the send window: + * seq_lt (tc->snd_una, ack). Note: last condition could be dropped, we don't + * try to update rtt for dupacks */ + else if (tcp_opts_tstamp (&tc->opt) && tc->opt.tsecr && tc->bytes_acked) + { + mrtt = tcp_time_now () - tc->opt.tsecr; + } + + /* Ignore dubious measurements */ + if (mrtt == 0 || mrtt > TCP_RTT_MAX) + return 0; + + tcp_estimate_rtt (tc, mrtt); + + tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); + + return 1; +} + +/** + * Dequeue bytes that have been acked and while at it update RTT estimates. + */ +static void +tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) +{ + /* Dequeue the newly ACKed bytes */ + stream_session_dequeue_drop (&tc->connection, tc->bytes_acked); + + /* Update rtt and rto */ + if (tcp_update_rtt (tc, ack)) + { + /* Good ACK received and valid RTT, make sure retransmit backoff is 0 */ + tc->rto_boff = 0; + } +} + +/** Check if dupack as per RFC5681 Sec. 2 */ +always_inline u8 +tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 new_snd_wnd) +{ + return ((vnet_buffer (b)->tcp.ack_number == tc->snd_una) + && seq_gt (tc->snd_una_max, tc->snd_una) + && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number) + && (new_snd_wnd == tc->snd_wnd)); +} + +void +scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) +{ + sack_scoreboard_hole_t *next, *prev; + + if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) + { + next = pool_elt_at_index (sb->holes, hole->next); + next->prev = hole->prev; + } + + if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) + { + prev = pool_elt_at_index (sb->holes, hole->prev); + prev->next = hole->next; + } + else + { + sb->head = hole->next; + } + + pool_put (sb->holes, hole); +} + +sack_scoreboard_hole_t * +scoreboard_insert_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * prev, + u32 start, u32 end) +{ + sack_scoreboard_hole_t *hole, *next; + u32 hole_index; + + pool_get (sb->holes, hole); + memset (hole, 0, sizeof (*hole)); + + hole->start = start; + hole->end = end; + hole_index = hole - sb->holes; + + if (prev) + { + hole->prev = prev - sb->holes; + hole->next = prev->next; + + if ((next = scoreboard_next_hole (sb, hole))) + next->prev = hole_index; + + prev->next = hole_index; + } + else + { + sb->head = hole_index; + hole->prev = TCP_INVALID_SACK_HOLE_INDEX; + hole->next = TCP_INVALID_SACK_HOLE_INDEX; + } + + return hole; +} + +static void +tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) +{ + sack_scoreboard_t *sb = &tc->sack_sb; + sack_block_t *blk, tmp; + sack_scoreboard_hole_t *hole, *next_hole; + u32 blk_index = 0; + int i, j; + + if (!tcp_opts_sack (tc) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) + return; + + /* Remove invalid blocks */ + vec_foreach (blk, tc->opt.sacks) + { + if (seq_lt (blk->start, blk->end) + && seq_gt (blk->start, tc->snd_una) + && seq_gt (blk->start, ack) && seq_lt (blk->end, tc->snd_nxt)) + continue; + + vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); + } + + /* Add block for cumulative ack */ + if (seq_gt (ack, tc->snd_una)) + { + tmp.start = tc->snd_una; + tmp.end = ack; + vec_add1 (tc->opt.sacks, tmp); + } + + if (vec_len (tc->opt.sacks) == 0) + return; + + /* Make sure blocks are ordered */ + for (i = 0; i < vec_len (tc->opt.sacks); i++) + for (j = i; j < vec_len (tc->opt.sacks); j++) + if (seq_lt (tc->opt.sacks[j].start, tc->opt.sacks[i].start)) + { + tmp = tc->opt.sacks[i]; + tc->opt.sacks[i] = tc->opt.sacks[j]; + tc->opt.sacks[j] = tmp; + } + + /* If no holes, insert the first that covers all outstanding bytes */ + if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) + { + scoreboard_insert_hole (sb, 0, tc->snd_una, tc->snd_una_max); + } + + /* Walk the holes with the SACK blocks */ + hole = pool_elt_at_index (sb->holes, sb->head); + while (hole && blk_index < vec_len (tc->opt.sacks)) + { + blk = &tc->opt.sacks[blk_index]; + + if (seq_leq (blk->start, hole->start)) + { + /* Block covers hole. Remove hole */ + if (seq_geq (blk->end, hole->end)) + { + next_hole = scoreboard_next_hole (sb, hole); + + /* Byte accounting */ + if (seq_lt (hole->end, ack)) + { + /* Bytes lost because snd wnd left edge advances */ + if (seq_lt (next_hole->start, ack)) + sb->sacked_bytes -= next_hole->start - hole->end; + else + sb->sacked_bytes -= ack - hole->end; + } + else + { + sb->sacked_bytes += scoreboard_hole_bytes (hole); + } + + scoreboard_remove_hole (sb, hole); + hole = next_hole; + } + /* Partial overlap */ + else + { + sb->sacked_bytes += blk->end - hole->start; + hole->start = blk->end; + blk_index++; + } + } + else + { + /* Hole must be split */ + if (seq_leq (blk->end, hole->end)) + { + sb->sacked_bytes += blk->end - blk->start; + scoreboard_insert_hole (sb, hole, blk->end, hole->end); + hole->end = blk->start - 1; + blk_index++; + } + else + { + sb->sacked_bytes += hole->end - blk->start + 1; + hole->end = blk->start - 1; + hole = scoreboard_next_hole (sb, hole); + } + } + } +} + +/** Update snd_wnd + * + * If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set + * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ +static void +tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) +{ + if (tc->snd_wl1 < seq || (tc->snd_wl1 == seq && tc->snd_wl2 <= ack)) + { + tc->snd_wnd = snd_wnd; + tc->snd_wl1 = seq; + tc->snd_wl2 = ack; + } +} + +static void +tcp_cc_congestion (tcp_connection_t * tc) +{ + tc->cc_algo->congestion (tc); +} + +static void +tcp_cc_recover (tcp_connection_t * tc) +{ + if (tcp_in_fastrecovery (tc)) + { + tc->cc_algo->recovered (tc); + tcp_recovery_off (tc); + } + else if (tcp_in_recovery (tc)) + { + tcp_recovery_off (tc); + tc->cwnd = tcp_loss_wnd (tc); + } +} + +static void +tcp_cc_rcv_ack (tcp_connection_t * tc) +{ + u8 partial_ack; + + if (tcp_in_recovery (tc)) + { + partial_ack = seq_lt (tc->snd_una, tc->snd_una_max); + if (!partial_ack) + { + /* Clear retransmitted bytes. */ + tc->rtx_bytes = 0; + tcp_cc_recover (tc); + } + else + { + /* Clear retransmitted bytes. XXX should we clear all? */ + tc->rtx_bytes = 0; + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); + + /* Retransmit first unacked segment */ + tcp_retransmit_first_unacked (tc); + } + } + else + { + tc->cc_algo->rcv_ack (tc); + } + + tc->rcv_dupacks = 0; + tc->tsecr_last_ack = tc->opt.tsecr; +} + +static void +tcp_cc_rcv_dupack (tcp_connection_t * tc, u32 ack) +{ + ASSERT (tc->snd_una == ack); + + tc->rcv_dupacks++; + if (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) + { + /* RFC6582 NewReno heuristic to avoid multiple fast retransmits */ + if (tc->opt.tsecr != tc->tsecr_last_ack) + { + tc->rcv_dupacks = 0; + return; + } + + tcp_fastrecovery_on (tc); + + /* Handle congestion and dupack */ + tcp_cc_congestion (tc); + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + + tcp_fast_retransmit (tc); + + /* Post retransmit update cwnd to ssthresh and account for the + * three segments that have left the network and should've been + * buffered at the receiver */ + tc->cwnd = tc->ssthresh + TCP_DUPACK_THRESHOLD * tc->snd_mss; + } + else if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD) + { + ASSERT (tcp_in_fastrecovery (tc)); + + tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); + } +} + +void +tcp_cc_init (tcp_connection_t * tc) +{ + tc->cc_algo = tcp_cc_algo_get (TCP_CC_NEWRENO); + tc->cc_algo->init (tc); +} + +static int +tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, + tcp_header_t * th, u32 * next, u32 * error) +{ + u32 new_snd_wnd; + + /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) then send an + * ACK, drop the segment, and return */ + if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)) + { + tcp_make_ack (tc, b); + *next = tcp_next_output (tc->c_is_ip4); + *error = TCP_ERROR_ACK_INVALID; + return -1; + } + + /* If old ACK, discard */ + if (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)) + { + *error = TCP_ERROR_ACK_OLD; + return -1; + } + + if (tcp_opts_sack_permitted (&tc->opt)) + tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); + + new_snd_wnd = clib_net_to_host_u32 (th->window) << tc->snd_wscale; + + if (tcp_ack_is_dupack (tc, b, new_snd_wnd)) + { + tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); + *error = TCP_ERROR_ACK_DUP; + return -1; + } + + /* Valid ACK */ + tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; + tc->snd_una = vnet_buffer (b)->tcp.ack_number; + + /* Dequeue ACKed packet and update RTT */ + tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); + + tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, + vnet_buffer (b)->tcp.ack_number, new_snd_wnd); + + /* Updates congestion control (slow start/congestion avoidance) */ + tcp_cc_rcv_ack (tc); + + /* If everything has been acked, stop retransmit timer + * otherwise update */ + if (tc->snd_una == tc->snd_una_max) + tcp_timer_reset (tc, TCP_TIMER_RETRANSMIT); + else + tcp_timer_update (tc, TCP_TIMER_RETRANSMIT, tc->rto); + + return 0; +} + +/** + * Build SACK list as per RFC2018. + * + * Makes sure the first block contains the segment that generated the current + * ACK and the following ones are the ones most recently reported in SACK + * blocks. + * + * @param tc TCP connection for which the SACK list is updated + * @param start Start sequence number of the newest SACK block + * @param end End sequence of the newest SACK block + */ +static void +tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) +{ + sack_block_t *new_list = 0, block; + u32 n_elts; + int i; + u8 new_head = 0; + + /* If the first segment is ooo add it to the list. Last write might've moved + * rcv_nxt over the first segment. */ + if (seq_lt (tc->rcv_nxt, start)) + { + block.start = start; + block.end = end; + vec_add1 (new_list, block); + new_head = 1; + } + + /* Find the blocks still worth keeping. */ + for (i = 0; i < vec_len (tc->snd_sacks); i++) + { + /* Discard if: + * 1) rcv_nxt advanced beyond current block OR + * 2) Segment overlapped by the first segment, i.e., it has been merged + * into it.*/ + if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt) + || seq_leq (tc->snd_sacks[i].start, end)) + continue; + + /* Save subsequent segments to new SACK list. */ + n_elts = clib_min (vec_len (tc->snd_sacks) - i, + TCP_MAX_SACK_BLOCKS - new_head); + vec_insert_elts (new_list, &tc->snd_sacks[i], n_elts, new_head); + break; + } + + /* Replace old vector with new one */ + vec_free (tc->snd_sacks); + tc->snd_sacks = new_list; +} + +/** Enqueue data for delivery to application */ +always_inline u32 +tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, + u16 data_len) +{ + int written; + + /* Pure ACK. Update rcv_nxt and be done. */ + if (PREDICT_FALSE (data_len == 0)) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; + return TCP_ERROR_PURE_ACK; + } + + written = stream_session_enqueue_data (&tc->connection, + vlib_buffer_get_current (b), + data_len, 1 /* queue event */ ); + + /* Update rcv_nxt */ + if (PREDICT_TRUE (written == data_len)) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; + } + /* If more data written than expected, account for out-of-order bytes. */ + else if (written > data_len) + { + tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end + written - data_len; + + /* Send ACK confirming the update */ + tc->flags |= TCP_CONN_SNDACK; + + /* Update SACK list if need be */ + if (tcp_opts_sack_permitted (&tc->opt)) + { + /* Remove SACK blocks that have been delivered */ + tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); + } + } + else + { + ASSERT (0); + return TCP_ERROR_FIFO_FULL; + } + + return TCP_ERROR_ENQUEUED; +} + +/** Enqueue out-of-order data */ +always_inline u32 +tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, + u16 data_len) +{ + stream_session_t *s0; + u32 offset, seq; + + s0 = stream_session_get (tc->c_s_index, tc->c_thread_index); + seq = vnet_buffer (b)->tcp.seq_number; + offset = seq - tc->rcv_nxt; + + if (svm_fifo_enqueue_with_offset (s0->server_rx_fifo, s0->pid, offset, + data_len, vlib_buffer_get_current (b))) + return TCP_ERROR_FIFO_FULL; + + /* Update SACK list if in use */ + if (tcp_opts_sack_permitted (&tc->opt)) + { + ooo_segment_t *newest; + u32 start, end; + + /* Get the newest segment from the fifo */ + newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo); + start = tc->rcv_nxt + ooo_segment_offset (s0->server_rx_fifo, newest); + end = tc->rcv_nxt + ooo_segment_end_offset (s0->server_rx_fifo, newest); + + tcp_update_sack_list (tc, start, end); + } + + return TCP_ERROR_ENQUEUED; +} + +/** + * Check if ACK could be delayed. DELACK timer is set only after frame is + * processed so this can return true for a full bursts of packets. + */ +always_inline int +tcp_can_delack (tcp_connection_t * tc) +{ + /* If there's no DELACK timer set and the last window sent wasn't 0 we + * can safely delay. */ + if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK) + && (tc->flags & TCP_CONN_SENT_RCV_WND0) == 0 + && (tc->flags & TCP_CONN_SNDACK) == 0) + return 1; + + return 0; +} + +static int +tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, + u16 n_data_bytes, u32 * next0) +{ + u32 error = 0; + + /* Handle out-of-order data */ + if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt)) + { + error = tcp_session_enqueue_ooo (tc, b, n_data_bytes); + + /* Don't send more than 3 dupacks per burst + * XXX decide if this is good */ + if (tc->snt_dupacks < 3) + { + /* RFC2581: Send DUPACK for fast retransmit */ + tcp_make_ack (tc, b); + *next0 = tcp_next_output (tc->c_is_ip4); + + /* Mark as DUPACK. We may filter these in output if + * the burst fills the holes. */ + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; + + tc->snt_dupacks++; + } + + goto done; + } + + /* In order data, enqueue. Fifo figures out by itself if any out-of-order + * segments can be enqueued after fifo tail offset changes. */ + error = tcp_session_enqueue_data (tc, b, n_data_bytes); + + /* Check if ACK can be delayed */ + if (tcp_can_delack (tc)) + { + /* Nothing to do for pure ACKs */ + if (n_data_bytes == 0) + goto done; + + /* If connection has not been previously marked for delay ack + * add it to the list and flag it */ + if (!tc->flags & TCP_CONN_DELACK) + { + vec_add1 (tm->delack_connections[tc->c_thread_index], + tc->c_c_index); + tc->flags |= TCP_CONN_DELACK; + } + } + else + { + /* Check if a packet has already been enqueued to output for burst. + * If yes, then drop this one, otherwise, let it pass through to + * output */ + if ((tc->flags & TCP_CONN_BURSTACK) == 0) + { + *next0 = tcp_next_output (tc->c_is_ip4); + tcp_make_ack (tc, b); + error = TCP_ERROR_ENQUEUED; + + /* TODO: maybe add counter to ensure N acks will be sent/burst */ + tc->flags |= TCP_CONN_BURSTACK; + } + } + +done: + return error; +} + +void +delack_timers_init (tcp_main_t * tm, u32 thread_index) +{ + tcp_connection_t *tc; + u32 i, *conns; + tw_timer_wheel_16t_2w_512sl_t *tw; + + tw = &tm->timer_wheels[thread_index]; + conns = tm->delack_connections[thread_index]; + for (i = 0; i < vec_len (conns); i++) + { + tc = pool_elt_at_index (tm->connections[thread_index], conns[i]); + ASSERT (0 != tc); + + tc->timers[TCP_TIMER_DELACK] + = tw_timer_start_16t_2w_512sl (tw, conns[i], + TCP_TIMER_DELACK, TCP_DELACK_TIME); + } + vec_reset_length (tm->delack_connections[thread_index]); +} + +always_inline uword +tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + tcp_main_t *tm = vnet_get_tcp_main (); + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *th0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + u32 next0 = TCP_ESTABLISHED_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (th0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (th0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number + + tcp_is_syn (th0) + tcp_is_fin (th0) + n_data_bytes0; + + /* TODO header prediction fast path */ + + /* 1-4: check SEQ, RST, SYN */ + if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0))) + { + error0 = TCP_ERROR_SEGMENT_INVALID; + goto drop; + } + + /* 5: check the ACK field */ + if (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)) + { + goto drop; + } + + /* 6: check the URG bit TODO */ + + /* 7: process the segment text */ + vlib_buffer_advance (b0, n_advance_bytes0); + error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); + + /* 8: check the FIN bit */ + if (tcp_fin (th0)) + { + /* Send ACK and enter CLOSE-WAIT */ + tcp_make_ack (tc0, b0); + tcp_connection_force_ack (tc0, b0); + next0 = tcp_next_output (tc0->c_is_ip4); + tc0->state = TCP_STATE_CLOSE_WAIT; + stream_session_disconnect_notify (&tc0->connection); + } + + drop: + b0->error = node->errors[error0]; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + delack_timers_init (tm, my_thread_index); + + return from_frame->n_vectors; +} + +static uword +tcp4_established (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_established (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_established_node) = +{ + .function = tcp4_established, + .name = "tcp4-established", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR,.error_strings = tcp_error_strings, + .n_next_nodes = TCP_ESTABLISHED_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_established_node, tcp4_established); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_established_node) = +{ + .function = tcp6_established, + .name = "tcp6-established", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_ESTABLISHED_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_established_node, tcp6_established); + +vlib_node_registration_t tcp4_syn_sent_node; +vlib_node_registration_t tcp6_syn_sent_node; + +always_inline uword +tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0, ack0, seq0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + tcp_connection_t *new_tc0; + u32 next0 = TCP_SYN_SENT_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = + tcp_half_open_connection_get (vnet_buffer (b0)-> + tcp.connection_index); + + ack0 = vnet_buffer (b0)->tcp.ack_number; + seq0 = vnet_buffer (b0)->tcp.seq_number; + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (tcp0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (tcp0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + if (PREDICT_FALSE + (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0))) + goto drop; + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0) + + tcp_is_fin (tcp0) + n_data_bytes0; + + /* + * 1. check the ACK bit + */ + + /* + * If the ACK bit is set + * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless + * the RST bit is set, if so drop the segment and return) + * + * and discard the segment. Return. + * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable. + */ + if (tcp_ack (tcp0)) + { + if (ack0 <= tc0->iss || ack0 > tc0->snd_nxt) + { + if (!tcp_rst (tcp0)) + tcp_send_reset (b0, is_ip4); + + goto drop; + } + + /* Make sure ACK is valid */ + if (tc0->snd_una > ack0) + goto drop; + } + + /* + * 2. check the RST bit + */ + + if (tcp_rst (tcp0)) + { + /* If ACK is acceptable, signal client that peer is not + * willing to accept connection and drop connection*/ + if (tcp_ack (tcp0)) + { + stream_session_connect_notify (&tc0->connection, sst, + 1 /* fail */ ); + tcp_connection_cleanup (tc0); + } + goto drop; + } + + /* + * 3. check the security and precedence (skipped) + */ + + /* + * 4. check the SYN bit + */ + + /* No SYN flag. Drop. */ + if (!tcp_syn (tcp0)) + goto drop; + + /* Stop connection establishment and retransmit timers */ + tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH); + tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT_SYN); + + /* Valid SYN or SYN-ACK. Move connection from half-open pool to + * current thread pool. */ + pool_get (tm->connections[my_thread_index], new_tc0); + clib_memcpy (new_tc0, tc0, sizeof (*new_tc0)); + + new_tc0->c_thread_index = my_thread_index; + + /* Cleanup half-open connection XXX lock */ + pool_put (tm->half_open_connections, tc0); + + new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end; + new_tc0->irs = seq0; + + /* Parse options */ + tcp_options_parse (tcp0, &new_tc0->opt); + tcp_connection_init_vars (new_tc0); + + if (tcp_opts_tstamp (&new_tc0->opt)) + { + new_tc0->tsval_recent = new_tc0->opt.tsval; + new_tc0->tsval_recent_age = tcp_time_now (); + } + + if (tcp_opts_wscale (&new_tc0->opt)) + new_tc0->snd_wscale = new_tc0->opt.wscale; + + new_tc0->snd_wnd = clib_net_to_host_u32 (tcp0->window) + << new_tc0->snd_wscale; + new_tc0->snd_wl1 = seq0; + new_tc0->snd_wl2 = ack0; + + /* SYN-ACK: See if we can switch to ESTABLISHED state */ + if (tcp_ack (tcp0)) + { + /* Our SYN is ACKed: we have iss < ack = snd_una */ + + /* TODO Dequeue acknowledged segments if we support Fast Open */ + new_tc0->snd_una = ack0; + new_tc0->state = TCP_STATE_ESTABLISHED; + + /* Notify app that we have connection */ + stream_session_connect_notify (&new_tc0->connection, sst, 0); + + /* Make sure after data segment processing ACK is sent */ + new_tc0->flags |= TCP_CONN_SNDACK; + } + /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */ + else + { + new_tc0->state = TCP_STATE_SYN_RCVD; + + /* Notify app that we have connection XXX */ + stream_session_connect_notify (&new_tc0->connection, sst, 0); + + tcp_make_synack (new_tc0, b0); + next0 = tcp_next_output (is_ip4); + + goto drop; + } + + /* Read data, if any */ + if (n_data_bytes0) + { + error0 = + tcp_segment_rcv (tm, new_tc0, b0, n_data_bytes0, &next0); + if (error0 == TCP_ERROR_PURE_ACK) + error0 = TCP_ERROR_SYN_ACKS_RCVD; + } + else + { + tcp_make_ack (new_tc0, b0); + next0 = tcp_next_output (new_tc0->c_is_ip4); + } + + drop: + + b0->error = error0 ? node->errors[error0] : 0; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_syn_sent (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_syn_sent_rcv (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_syn_sent_node) = +{ + .function = tcp4_syn_sent, + .name = "tcp4-syn-sent", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_SYN_SENT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_syn_sent_node, tcp4_syn_sent); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_syn_sent_node) = +{ + .function = tcp6_syn_sent_rcv, + .name = "tcp6-syn-sent", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_SYN_SENT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + } +,}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_syn_sent_node, tcp6_syn_sent_rcv); +/** + * Handles reception for all states except LISTEN, SYN-SEND and ESTABLISHED + * as per RFC793 p. 64 + */ +always_inline uword +tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index, errors = 0; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 n_advance_bytes0, n_data_bytes0; + u32 next0 = TCP_RCV_PROCESS_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + + /* Checksum computed by ipx_local no need to compute again */ + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + n_advance_bytes0 = (ip4_header_bytes (ip40) + + tcp_header_bytes (tcp0)); + n_data_bytes0 = clib_net_to_host_u16 (ip40->length) + - n_advance_bytes0; + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + n_advance_bytes0 = tcp_header_bytes (tcp0); + n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) + - n_advance_bytes0; + n_advance_bytes0 += sizeof (ip60[0]); + } + + /* SYNs, FINs and data consume sequence numbers */ + vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number + + tcp_is_syn (tcp0) + tcp_is_fin (tcp0) + n_data_bytes0; + + /* + * Special treatment for CLOSED + */ + switch (tc0->state) + { + case TCP_STATE_CLOSED: + goto drop; + break; + } + + /* + * For all other states (except LISTEN) + */ + + /* 1-4: check SEQ, RST, SYN */ + if (PREDICT_FALSE + (tcp_segment_validate (vm, tc0, b0, tcp0, &next0))) + { + error0 = TCP_ERROR_SEGMENT_INVALID; + goto drop; + } + + /* 5: check the ACK field */ + switch (tc0->state) + { + case TCP_STATE_SYN_RCVD: + /* + * If the segment acknowledgment is not acceptable, form a + * reset segment, + * + * and send it. + */ + if (!tcp_rcv_ack_is_acceptable (tc0, b0)) + { + tcp_send_reset (b0, is_ip4); + goto drop; + } + /* Switch state to ESTABLISHED */ + tc0->state = TCP_STATE_ESTABLISHED; + + /* Initialize session variables */ + tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; + tc0->snd_wnd = clib_net_to_host_u32 (tcp0->window) + << tc0->opt.wscale; + tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; + tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; + + /* Shoulder tap the server */ + stream_session_accept_notify (&tc0->connection); + + tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT_SYN); + break; + case TCP_STATE_ESTABLISHED: + /* We can get packets in established state here because they + * were enqueued before state change */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + + break; + case TCP_STATE_FIN_WAIT_1: + /* In addition to the processing for the ESTABLISHED state, if + * our FIN is now acknowledged then enter FIN-WAIT-2 and + * continue processing in that state. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + tc0->state = TCP_STATE_FIN_WAIT_2; + /* Stop all timers, 2MSL will be set lower */ + tcp_connection_timers_reset (tc0); + break; + case TCP_STATE_FIN_WAIT_2: + /* In addition to the processing for the ESTABLISHED state, if + * the retransmission queue is empty, the user's CLOSE can be + * acknowledged ("ok") but do not delete the TCB. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + /* check if rtx queue is empty and ack CLOSE TODO */ + break; + case TCP_STATE_CLOSE_WAIT: + /* Do the same processing as for the ESTABLISHED state. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + break; + case TCP_STATE_CLOSING: + /* In addition to the processing for the ESTABLISHED state, if + * the ACK acknowledges our FIN then enter the TIME-WAIT state, + * otherwise ignore the segment. */ + if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) + goto drop; + + /* XXX test that send queue empty */ + tc0->state = TCP_STATE_TIME_WAIT; + goto drop; + + break; + case TCP_STATE_LAST_ACK: + /* The only thing that can arrive in this state is an + * acknowledgment of our FIN. If our FIN is now acknowledged, + * delete the TCB, enter the CLOSED state, and return. */ + + if (!tcp_rcv_ack_is_acceptable (tc0, b0)) + goto drop; + + tcp_connection_del (tc0); + goto drop; + + break; + case TCP_STATE_TIME_WAIT: + /* The only thing that can arrive in this state is a + * retransmission of the remote FIN. Acknowledge it, and restart + * the 2 MSL timeout. */ + + /* TODO */ + goto drop; + break; + default: + ASSERT (0); + } + + /* 6: check the URG bit TODO */ + + /* 7: process the segment text */ + switch (tc0->state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_FIN_WAIT_1: + case TCP_STATE_FIN_WAIT_2: + error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); + break; + case TCP_STATE_CLOSE_WAIT: + case TCP_STATE_CLOSING: + case TCP_STATE_LAST_ACK: + case TCP_STATE_TIME_WAIT: + /* This should not occur, since a FIN has been received from the + * remote side. Ignore the segment text. */ + break; + } + + /* 8: check the FIN bit */ + if (!tcp_fin (tcp0)) + goto drop; + + switch (tc0->state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_SYN_RCVD: + /* Send FIN-ACK notify app and enter CLOSE-WAIT */ + tcp_connection_timers_reset (tc0); + tcp_make_finack (tc0, b0); + next0 = tcp_next_output (tc0->c_is_ip4); + stream_session_disconnect_notify (&tc0->connection); + tc0->state = TCP_STATE_CLOSE_WAIT; + break; + case TCP_STATE_CLOSE_WAIT: + case TCP_STATE_CLOSING: + case TCP_STATE_LAST_ACK: + /* move along .. */ + break; + case TCP_STATE_FIN_WAIT_1: + tc0->state = TCP_STATE_TIME_WAIT; + tcp_connection_timers_reset (tc0); + tcp_timer_set (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + break; + case TCP_STATE_FIN_WAIT_2: + /* Got FIN, send ACK! */ + tc0->state = TCP_STATE_TIME_WAIT; + tcp_timer_set (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + tcp_make_ack (tc0, b0); + next0 = tcp_next_output (is_ip4); + break; + case TCP_STATE_TIME_WAIT: + /* Remain in the TIME-WAIT state. Restart the 2 MSL time-wait + * timeout. + */ + tcp_timer_update (tc0, TCP_TIMER_2MSL, TCP_2MSL_TIME); + break; + } + + b0->error = error0 ? node->errors[error0] : 0; + + drop: + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + errors = session_manager_flush_enqueue_events (my_thread_index); + if (errors) + { + if (is_ip4) + vlib_node_increment_counter (vm, tcp4_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + else + vlib_node_increment_counter (vm, tcp6_established_node.index, + TCP_ERROR_EVENT_FIFO_FULL, errors); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_rcv_process_node) = +{ + .function = tcp4_rcv_process, + .name = "tcp4-rcv-process", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_rcv_process_node, tcp4_rcv_process); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_rcv_process_node) = +{ + .function = tcp6_rcv_process, + .name = "tcp6-rcv-process", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_rcv_process_node, tcp6_rcv_process); + +vlib_node_registration_t tcp4_listen_node; +vlib_node_registration_t tcp6_listen_node; + +/** + * LISTEN state processing as per RFC 793 p. 65 + */ +always_inline uword +tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + tcp_main_t *tm = vnet_get_tcp_main (); + u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *th0 = 0; + tcp_connection_t *lc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + tcp_connection_t *child0; + u32 error0 = TCP_ERROR_SYNS_RCVD, next0 = TCP_LISTEN_NEXT_DROP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index); + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ip40); + } + else + { + ip60 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ip60); + } + + /* Create child session. For syn-flood protection use filter */ + + /* 1. first check for an RST */ + if (tcp_rst (th0)) + goto drop; + + /* 2. second check for an ACK */ + if (tcp_ack (th0)) + { + tcp_send_reset (b0, is_ip4); + goto drop; + } + + /* 3. check for a SYN (did that already) */ + + /* Create child session and send SYN-ACK */ + pool_get (tm->connections[my_thread_index], child0); + memset (child0, 0, sizeof (*child0)); + + child0->c_c_index = child0 - tm->connections[my_thread_index]; + child0->c_lcl_port = lc0->c_lcl_port; + child0->c_rmt_port = th0->src_port; + child0->c_is_ip4 = is_ip4; + child0->c_thread_index = my_thread_index; + + if (is_ip4) + { + child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32; + child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32; + } + else + { + clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address, + sizeof (ip6_address_t)); + clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address, + sizeof (ip6_address_t)); + } + + if (stream_session_accept (&child0->connection, lc0->c_s_index, sst, + 0 /* notify */ )) + { + error0 = TCP_ERROR_CREATE_SESSION_FAIL; + goto drop; + } + + tcp_options_parse (th0, &child0->opt); + tcp_connection_init_vars (child0); + + child0->irs = vnet_buffer (b0)->tcp.seq_number; + child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1; + child0->state = TCP_STATE_SYN_RCVD; + + /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK} + * segments are used to initialize PAWS. */ + if (tcp_opts_tstamp (&child0->opt)) + { + child0->tsval_recent = child0->opt.tsval; + child0->tsval_recent_age = tcp_time_now (); + } + + /* Reuse buffer to make syn-ack and send */ + tcp_make_synack (child0, b0); + next0 = tcp_next_output (is_ip4); + + drop: + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + b0->error = error0 ? node->errors[error0] : 0; + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + return from_frame->n_vectors; +} + +static uword +tcp4_listen (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_listen (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_listen_node) = +{ + .function = tcp4_listen, + .name = "tcp4-listen", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_LISTEN_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_LISTEN_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_listen_node, tcp4_listen); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_listen_node) = +{ + .function = tcp6_listen, + .name = "tcp6-listen", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_LISTEN_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_LISTEN_NEXT_##s] = n, + foreach_tcp_state_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_listen_node, tcp6_listen); + +vlib_node_registration_t tcp4_input_node; +vlib_node_registration_t tcp6_input_node; + +typedef enum _tcp_input_next +{ + TCP_INPUT_NEXT_DROP, + TCP_INPUT_NEXT_LISTEN, + TCP_INPUT_NEXT_RCV_PROCESS, + TCP_INPUT_NEXT_SYN_SENT, + TCP_INPUT_NEXT_ESTABLISHED, + TCP_INPUT_NEXT_RESET, + TCP_INPUT_N_NEXT +} tcp_input_next_t; + +#define foreach_tcp4_input_next \ + _ (DROP, "error-drop") \ + _ (LISTEN, "tcp4-listen") \ + _ (RCV_PROCESS, "tcp4-rcv-process") \ + _ (SYN_SENT, "tcp4-syn-sent") \ + _ (ESTABLISHED, "tcp4-established") \ + _ (RESET, "tcp4-reset") + +#define foreach_tcp6_input_next \ + _ (DROP, "error-drop") \ + _ (LISTEN, "tcp6-listen") \ + _ (RCV_PROCESS, "tcp6-rcv-process") \ + _ (SYN_SENT, "tcp6-syn-sent") \ + _ (ESTABLISHED, "tcp6-established") \ + _ (RESET, "tcp6-reset") + +typedef struct +{ + u16 src_port; + u16 dst_port; + u8 state; +} tcp_rx_trace_t; + +const char *tcp_fsm_states[] = { +#define _(sym, str) str, + foreach_tcp_fsm_state +#undef _ +}; + +u8 * +format_tcp_state (u8 * s, va_list * args) +{ + tcp_state_t *state = va_arg (*args, tcp_state_t *); + + if (state[0] < TCP_N_STATES) + s = format (s, "%s", tcp_fsm_states[state[0]]); + else + s = format (s, "UNKNOWN"); + + return s; +} + +u8 * +format_tcp_rx_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); + + s = format (s, "TCP: src-port %d dst-port %U%s\n", + clib_net_to_host_u16 (t->src_port), + clib_net_to_host_u16 (t->dst_port), format_tcp_state, t->state); + + return s; +} + +#define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN) + +always_inline uword +tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + tcp_main_t *tm = vnet_get_tcp_main (); + session_manager_main_t *ssm = vnet_get_session_manager_main (); + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_header_t *tcp0 = 0; + tcp_connection_t *tc0; + ip4_header_t *ip40; + ip6_header_t *ip60; + u32 error0 = TCP_ERROR_NO_LISTENER, next0 = TCP_INPUT_NEXT_DROP; + u8 flags0; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + if (is_ip4) + { + ip40 = vlib_buffer_get_current (b0); + tcp0 = ip4_next_header (ip40); + + /* lookup session */ + tc0 = + (tcp_connection_t *) stream_session_lookup_transport4 (ssm, + &ip40->dst_address, + &ip40->src_address, + tcp0->dst_port, + tcp0->src_port, + SESSION_TYPE_IP4_TCP, + my_thread_index); + } + else + { + ip60 = vlib_buffer_get_current (b0); + tcp0 = ip6_next_header (ip60); + tc0 = + (tcp_connection_t *) stream_session_lookup_transport6 (ssm, + &ip60->src_address, + &ip60->dst_address, + tcp0->src_port, + tcp0->dst_port, + SESSION_TYPE_IP6_TCP, + my_thread_index); + } + + /* Session exists */ + if (PREDICT_TRUE (0 != tc0)) + { + /* Save connection index */ + vnet_buffer (b0)->tcp.connection_index = tc0->c_c_index; + vnet_buffer (b0)->tcp.seq_number = + clib_net_to_host_u32 (tcp0->seq_number); + vnet_buffer (b0)->tcp.ack_number = + clib_net_to_host_u32 (tcp0->ack_number); + + flags0 = tcp0->flags & filter_flags; + next0 = tm->dispatch_table[tc0->state][flags0].next; + error0 = tm->dispatch_table[tc0->state][flags0].error; + + if (PREDICT_FALSE (error0 == TCP_ERROR_DISPATCH)) + { + /* Overload tcp flags to store state */ + vnet_buffer (b0)->tcp.flags = tc0->state; + } + } + else + { + /* Send reset */ + next0 = TCP_INPUT_NEXT_RESET; + error0 = TCP_ERROR_NO_LISTENER; + vnet_buffer (b0)->tcp.flags = 0; + } + + b0->error = error0 ? node->errors[error0] : 0; + + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_input (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_input (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_input_node) = +{ + .function = tcp4_input, + .name = "tcp4-input", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_INPUT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_INPUT_NEXT_##s] = n, + foreach_tcp4_input_next +#undef _ + }, + .format_buffer = format_tcp_header, + .format_trace = format_tcp_rx_trace, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_input_node, tcp4_input); + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_input_node) = +{ + .function = tcp6_input, + .name = "tcp6-input", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_INPUT_N_NEXT, + .next_nodes = + { +#define _(s,n) [TCP_INPUT_NEXT_##s] = n, + foreach_tcp6_input_next +#undef _ + }, + .format_buffer = format_tcp_header, + .format_trace = format_tcp_rx_trace, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_input_node, tcp6_input); +void +tcp_update_time (f64 now, u32 thread_index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + tw_timer_expire_timers_16t_2w_512sl (&tm->timer_wheels[thread_index], now); +} + +static void +tcp_dispatch_table_init (tcp_main_t * tm) +{ + int i, j; + for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++) + for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++) + { + tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP; + tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH; + } + +#define _(t,f,n,e) \ +do { \ + tm->dispatch_table[TCP_STATE_##t][f].next = (n); \ + tm->dispatch_table[TCP_STATE_##t][f].error = (e); \ +} while (0) + + /* SYNs for new connections -> tcp-listen. */ + _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE); + /* ACK for for a SYN-ACK -> tcp-rcv-process. */ + _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + /* SYN-ACK for a SYN */ + _(SYN_SENT, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, + TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); + _(SYN_SENT, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, + TCP_ERROR_NONE); + /* ACK for for established connection -> tcp-established. */ + _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); + /* FIN for for established connection -> tcp-established. */ + _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); + _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, + TCP_ERROR_NONE); + /* ACK or FIN-ACK to our FIN */ + _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(FIN_WAIT_1, TCP_FLAG_ACK | TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); + /* FIN in reply to our FIN from the other side */ + _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + /* FIN confirming that the peer (app) has closed */ + _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); + _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, + TCP_ERROR_NONE); + _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); +#undef _ +} + +clib_error_t * +tcp_input_init (vlib_main_t * vm) +{ + clib_error_t *error = 0; + tcp_main_t *tm = vnet_get_tcp_main (); + + if ((error = vlib_call_init_function (vm, tcp_init))) + return error; + + /* Initialize dispatch table. */ + tcp_dispatch_table_init (tm); + + return error; +} + +VLIB_INIT_FUNCTION (tcp_input_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_newreno.c b/src/vnet/tcp/tcp_newreno.c new file mode 100644 index 00000000000..856dffe4af9 --- /dev/null +++ b/src/vnet/tcp/tcp_newreno.c @@ -0,0 +1,93 @@ +/* + * Copyright (c) 2017 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. + */ + +#include + +void +newreno_congestion (tcp_connection_t * tc) +{ + tc->prev_ssthresh = tc->ssthresh; + tc->ssthresh = clib_max (tcp_flight_size (tc) / 2, 2 * tc->snd_mss); +} + +void +newreno_recovered (tcp_connection_t * tc) +{ + tc->cwnd = tc->ssthresh; +} + +void +newreno_rcv_ack (tcp_connection_t * tc) +{ + if (tcp_in_slowstart (tc)) + { + tc->cwnd += clib_min (tc->snd_mss, tc->bytes_acked); + } + else + { + /* Round up to 1 if needed */ + tc->cwnd += clib_max (tc->snd_mss * tc->snd_mss / tc->cwnd, 1); + } +} + +void +newreno_rcv_cong_ack (tcp_connection_t * tc, tcp_cc_ack_t ack_type) +{ + if (ack_type == TCP_CC_DUPACK) + { + tc->cwnd += tc->snd_mss; + } + else if (ack_type == TCP_CC_PARTIALACK) + { + tc->cwnd -= tc->bytes_acked; + if (tc->bytes_acked > tc->snd_mss) + tc->bytes_acked += tc->snd_mss; + } +} + +void +newreno_conn_init (tcp_connection_t * tc) +{ + tc->ssthresh = tc->snd_wnd; + tc->cwnd = tcp_initial_cwnd (tc); +} + +const static tcp_cc_algorithm_t tcp_newreno = { + .congestion = newreno_congestion, + .recovered = newreno_recovered, + .rcv_ack = newreno_rcv_ack, + .rcv_cong_ack = newreno_rcv_cong_ack, + .init = newreno_conn_init +}; + +clib_error_t * +newreno_init (vlib_main_t * vm) +{ + clib_error_t *error = 0; + + tcp_cc_algo_register (TCP_CC_NEWRENO, &tcp_newreno); + + return error; +} + +VLIB_INIT_FUNCTION (newreno_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_output.c b/src/vnet/tcp/tcp_output.c new file mode 100644 index 00000000000..dbcf1f74975 --- /dev/null +++ b/src/vnet/tcp/tcp_output.c @@ -0,0 +1,1412 @@ +/* + * 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. + */ + +#include +#include + +vlib_node_registration_t tcp4_output_node; +vlib_node_registration_t tcp6_output_node; + +typedef enum _tcp_output_nect +{ + TCP_OUTPUT_NEXT_DROP, + TCP_OUTPUT_NEXT_IP_LOOKUP, + TCP_OUTPUT_N_NEXT +} tcp_output_next_t; + +#define foreach_tcp4_output_next \ + _ (DROP, "error-drop") \ + _ (IP_LOOKUP, "ip4-lookup") + +#define foreach_tcp6_output_next \ + _ (DROP, "error-drop") \ + _ (IP_LOOKUP, "ip6-lookup") + +static char *tcp_error_strings[] = { +#define tcp_error(n,s) s, +#include +#undef tcp_error +}; + +typedef struct +{ + u16 src_port; + u16 dst_port; + u8 state; +} tcp_tx_trace_t; + +u16 dummy_mtu = 400; + +u8 * +format_tcp_tx_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + + s = format (s, "TBD\n"); + + return s; +} + +void +tcp_set_snd_mss (tcp_connection_t * tc) +{ + u16 snd_mss; + + /* TODO find our iface MTU */ + snd_mss = dummy_mtu; + + /* TODO cache mss and consider PMTU discovery */ + snd_mss = tc->opt.mss < snd_mss ? tc->opt.mss : snd_mss; + + tc->snd_mss = snd_mss; + + if (tc->snd_mss == 0) + { + clib_warning ("snd mss is 0"); + tc->snd_mss = dummy_mtu; + } +} + +static u8 +tcp_window_compute_scale (u32 available_space) +{ + u8 wnd_scale = 0; + while (wnd_scale < TCP_MAX_WND_SCALE + && (available_space >> wnd_scale) > TCP_WND_MAX) + wnd_scale++; + return wnd_scale; +} + +/** + * Compute initial window and scale factor. As per RFC1323, window field in + * SYN and SYN-ACK segments is never scaled. + */ +u32 +tcp_initial_window_to_advertise (tcp_connection_t * tc) +{ + u32 available_space; + + /* Initial wnd for SYN. Fifos are not allocated yet. + * Use some predefined value */ + if (tc->state != TCP_STATE_SYN_RCVD) + { + return TCP_DEFAULT_RX_FIFO_SIZE; + } + + available_space = stream_session_max_enqueue (&tc->connection); + tc->rcv_wscale = tcp_window_compute_scale (available_space); + tc->rcv_wnd = clib_min (available_space, TCP_WND_MAX << tc->rcv_wscale); + + return clib_min (tc->rcv_wnd, TCP_WND_MAX); +} + +/** + * Compute and return window to advertise, scaled as per RFC1323 + */ +u32 +tcp_window_to_advertise (tcp_connection_t * tc, tcp_state_t state) +{ + u32 available_space, wnd, scaled_space; + + if (state != TCP_STATE_ESTABLISHED) + return tcp_initial_window_to_advertise (tc); + + available_space = stream_session_max_enqueue (&tc->connection); + scaled_space = available_space >> tc->rcv_wscale; + + /* Need to update scale */ + if (PREDICT_FALSE ((scaled_space == 0 && available_space != 0)) + || (scaled_space >= TCP_WND_MAX)) + tc->rcv_wscale = tcp_window_compute_scale (available_space); + + wnd = clib_min (available_space, TCP_WND_MAX << tc->rcv_wscale); + tc->rcv_wnd = wnd; + + return wnd >> tc->rcv_wscale; +} + +/** + * Write TCP options to segment. + */ +u32 +tcp_options_write (u8 * data, tcp_options_t * opts) +{ + u32 opts_len = 0; + u32 buf, seq_len = 4; + + if (tcp_opts_mss (opts)) + { + *data++ = TCP_OPTION_MSS; + *data++ = TCP_OPTION_LEN_MSS; + buf = clib_host_to_net_u16 (opts->mss); + clib_memcpy (data, &buf, sizeof (opts->mss)); + data += sizeof (opts->mss); + opts_len += TCP_OPTION_LEN_MSS; + } + + if (tcp_opts_wscale (opts)) + { + *data++ = TCP_OPTION_WINDOW_SCALE; + *data++ = TCP_OPTION_LEN_WINDOW_SCALE; + *data++ = opts->wscale; + opts_len += TCP_OPTION_LEN_WINDOW_SCALE; + } + + if (tcp_opts_sack_permitted (opts)) + { + *data++ = TCP_OPTION_SACK_PERMITTED; + *data++ = TCP_OPTION_LEN_SACK_PERMITTED; + opts_len += TCP_OPTION_LEN_SACK_PERMITTED; + } + + if (tcp_opts_tstamp (opts)) + { + *data++ = TCP_OPTION_TIMESTAMP; + *data++ = TCP_OPTION_LEN_TIMESTAMP; + buf = clib_host_to_net_u32 (opts->tsval); + clib_memcpy (data, &buf, sizeof (opts->tsval)); + data += sizeof (opts->tsval); + buf = clib_host_to_net_u32 (opts->tsecr); + clib_memcpy (data, &buf, sizeof (opts->tsecr)); + data += sizeof (opts->tsecr); + opts_len += TCP_OPTION_LEN_TIMESTAMP; + } + + if (tcp_opts_sack (opts)) + { + int i; + u32 n_sack_blocks = clib_min (vec_len (opts->sacks), + TCP_OPTS_MAX_SACK_BLOCKS); + + if (n_sack_blocks != 0) + { + *data++ = TCP_OPTION_SACK_BLOCK; + *data++ = 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK; + for (i = 0; i < n_sack_blocks; i++) + { + buf = clib_host_to_net_u32 (opts->sacks[i].start); + clib_memcpy (data, &buf, seq_len); + data += seq_len; + buf = clib_host_to_net_u32 (opts->sacks[i].end); + clib_memcpy (data, &buf, seq_len); + data += seq_len; + } + opts_len += 2 + n_sack_blocks * TCP_OPTION_LEN_SACK_BLOCK; + } + } + + /* Terminate TCP options */ + if (opts_len % 4) + { + *data++ = TCP_OPTION_EOL; + opts_len += TCP_OPTION_LEN_EOL; + } + + /* Pad with zeroes to a u32 boundary */ + while (opts_len % 4) + { + *data++ = TCP_OPTION_NOOP; + opts_len += TCP_OPTION_LEN_NOOP; + } + return opts_len; +} + +always_inline int +tcp_make_syn_options (tcp_options_t * opts, u32 initial_wnd) +{ + u8 len = 0; + + opts->flags |= TCP_OPTS_FLAG_MSS; + opts->mss = dummy_mtu; /*XXX discover that */ + len += TCP_OPTION_LEN_MSS; + + opts->flags |= TCP_OPTS_FLAG_WSCALE; + opts->wscale = tcp_window_compute_scale (initial_wnd); + len += TCP_OPTION_LEN_WINDOW_SCALE; + + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = 0; + len += TCP_OPTION_LEN_TIMESTAMP; + + opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + len += TCP_OPTION_LEN_SACK_PERMITTED; + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_synack_options (tcp_connection_t * tc, tcp_options_t * opts) +{ + u8 len = 0; + + opts->flags |= TCP_OPTS_FLAG_MSS; + opts->mss = dummy_mtu; /*XXX discover that */ + len += TCP_OPTION_LEN_MSS; + + if (tcp_opts_wscale (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_WSCALE; + opts->wscale = tc->rcv_wscale; + len += TCP_OPTION_LEN_WINDOW_SCALE; + } + + if (tcp_opts_tstamp (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = tc->tsval_recent; + len += TCP_OPTION_LEN_TIMESTAMP; + } + + if (tcp_opts_sack_permitted (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; + len += TCP_OPTION_LEN_SACK_PERMITTED; + } + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_established_options (tcp_connection_t * tc, tcp_options_t * opts) +{ + u8 len = 0; + + opts->flags = 0; + + if (tcp_opts_tstamp (&tc->opt)) + { + opts->flags |= TCP_OPTS_FLAG_TSTAMP; + opts->tsval = tcp_time_now (); + opts->tsecr = tc->tsval_recent; + len += TCP_OPTION_LEN_TIMESTAMP; + } + if (tcp_opts_sack_permitted (&tc->opt)) + { + if (vec_len (tc->snd_sacks)) + { + opts->flags |= TCP_OPTS_FLAG_SACK; + opts->sacks = tc->snd_sacks; + opts->n_sack_blocks = vec_len (tc->snd_sacks); + len += 2 + TCP_OPTION_LEN_SACK_BLOCK * opts->n_sack_blocks; + } + } + + /* Align to needed boundary */ + len += (TCP_OPTS_ALIGN - len % TCP_OPTS_ALIGN) % TCP_OPTS_ALIGN; + return len; +} + +always_inline int +tcp_make_options (tcp_connection_t * tc, tcp_options_t * opts, + tcp_state_t state) +{ + switch (state) + { + case TCP_STATE_ESTABLISHED: + case TCP_STATE_FIN_WAIT_1: + return tcp_make_established_options (tc, opts); + case TCP_STATE_SYN_RCVD: + return tcp_make_synack_options (tc, opts); + case TCP_STATE_SYN_SENT: + return tcp_make_syn_options (opts, + tcp_initial_window_to_advertise (tc)); + default: + clib_warning ("Not handled!"); + return 0; + } +} + +#define tcp_get_free_buffer_index(tm, bidx) \ +do { \ + u32 *my_tx_buffers, n_free_buffers; \ + u32 cpu_index = tm->vlib_main->cpu_index; \ + my_tx_buffers = tm->tx_buffers[cpu_index]; \ + if (PREDICT_FALSE(vec_len (my_tx_buffers) == 0)) \ + { \ + n_free_buffers = 32; /* TODO config or macro */ \ + vec_validate (my_tx_buffers, n_free_buffers - 1); \ + _vec_len(my_tx_buffers) = vlib_buffer_alloc_from_free_list ( \ + tm->vlib_main, my_tx_buffers, n_free_buffers, \ + VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX); \ + tm->tx_buffers[cpu_index] = my_tx_buffers; \ + } \ + /* buffer shortage */ \ + if (PREDICT_FALSE (vec_len (my_tx_buffers) == 0)) \ + return; \ + *bidx = my_tx_buffers[_vec_len (my_tx_buffers)-1]; \ + _vec_len (my_tx_buffers) -= 1; \ +} while (0) + +always_inline void +tcp_reuse_buffer (vlib_main_t * vm, vlib_buffer_t * b) +{ + vlib_buffer_t *it = b; + do + { + it->current_data = 0; + it->current_length = 0; + it->total_length_not_including_first_buffer = 0; + } + while ((it->flags & VLIB_BUFFER_NEXT_PRESENT) + && (it = vlib_get_buffer (vm, it->next_buffer))); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); +} + +/** + * Prepare ACK + */ +void +tcp_make_ack_i (tcp_connection_t * tc, vlib_buffer_t * b, tcp_state_t state, + u8 flags) +{ + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + u8 tcp_opts_len, tcp_hdr_opts_len; + tcp_header_t *th; + u16 wnd; + + wnd = tcp_window_to_advertise (tc, state); + + /* Make and write options */ + tcp_opts_len = tcp_make_established_options (tc, snd_opts); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt, + tc->rcv_nxt, tcp_hdr_opts_len, flags, wnd); + + tcp_options_write ((u8 *) (th + 1), snd_opts); + + /* Mark as ACK */ + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; +} + +/** + * Convert buffer to ACK + */ +void +tcp_make_ack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_reuse_buffer (vm, b); + tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK); + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK; +} + +/** + * Convert buffer to FIN-ACK + */ +void +tcp_make_finack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_reuse_buffer (vm, b); + tcp_make_ack_i (tc, b, TCP_STATE_ESTABLISHED, TCP_FLAG_ACK | TCP_FLAG_FIN); + + /* Reset flags, make sure ack is sent */ + tc->flags = TCP_CONN_SNDACK; + vnet_buffer (b)->tcp.flags &= ~TCP_BUF_FLAG_DUPACK; + + tc->snd_nxt += 1; +} + +/** + * Convert buffer to SYN-ACK + */ +void +tcp_make_synack (tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + u8 tcp_opts_len, tcp_hdr_opts_len; + tcp_header_t *th; + u16 initial_wnd; + u32 time_now; + + memset (snd_opts, 0, sizeof (*snd_opts)); + + tcp_reuse_buffer (vm, b); + + /* Set random initial sequence */ + time_now = tcp_time_now (); + + tc->iss = random_u32 (&time_now); + tc->snd_una = tc->iss; + tc->snd_nxt = tc->iss + 1; + tc->snd_una_max = tc->snd_nxt; + + initial_wnd = tcp_initial_window_to_advertise (tc); + + /* Make and write options */ + tcp_opts_len = tcp_make_synack_options (tc, snd_opts); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss, + tc->rcv_nxt, tcp_hdr_opts_len, + TCP_FLAG_SYN | TCP_FLAG_ACK, initial_wnd); + + tcp_options_write ((u8 *) (th + 1), snd_opts); + + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; + vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_ACK; + + /* Init retransmit timer */ + tcp_retransmit_timer_set (tm, tc); +} + +always_inline void +tcp_enqueue_to_ip_lookup (vlib_main_t * vm, vlib_buffer_t * b, u32 bi, + u8 is_ip4) +{ + u32 *to_next, next_index; + vlib_frame_t *f; + + b->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + b->error = 0; + + /* Default FIB for now */ + vnet_buffer (b)->sw_if_index[VLIB_TX] = 0; + + /* Send to IP lookup */ + next_index = is_ip4 ? ip4_lookup_node.index : ip6_lookup_node.index; + f = vlib_get_frame_to_node (vm, next_index); + + /* Enqueue the packet */ + to_next = vlib_frame_vector_args (f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node (vm, next_index, f); +} + +int +tcp_make_reset_in_place (vlib_main_t * vm, vlib_buffer_t * b0, + tcp_state_t state, u32 my_thread_index, u8 is_ip4) +{ + u8 tcp_hdr_len = sizeof (tcp_header_t); + ip4_header_t *ih4; + ip6_header_t *ih6; + tcp_header_t *th0; + ip4_address_t src_ip40; + ip6_address_t src_ip60; + u16 src_port0; + u32 tmp; + + /* Find IP and TCP headers */ + if (is_ip4) + { + ih4 = vlib_buffer_get_current (b0); + th0 = ip4_next_header (ih4); + } + else + { + ih6 = vlib_buffer_get_current (b0); + th0 = ip6_next_header (ih6); + } + + /* Swap src and dst ip */ + if (is_ip4) + { + ASSERT ((ih4->ip_version_and_header_length & 0xF0) == 0x40); + src_ip40.as_u32 = ih4->src_address.as_u32; + ih4->src_address.as_u32 = ih4->dst_address.as_u32; + ih4->dst_address.as_u32 = src_ip40.as_u32; + + /* Chop the end of the pkt */ + b0->current_length += ip4_header_bytes (ih4) + tcp_hdr_len; + } + else + { + ASSERT ((ih6->ip_version_traffic_class_and_flow_label & 0xF0) == 0x60); + clib_memcpy (&src_ip60, &ih6->src_address, sizeof (ip6_address_t)); + clib_memcpy (&ih6->src_address, &ih6->dst_address, + sizeof (ip6_address_t)); + clib_memcpy (&ih6->dst_address, &src_ip60, sizeof (ip6_address_t)); + + /* Chop the end of the pkt */ + b0->current_length += sizeof (ip6_header_t) + tcp_hdr_len; + } + + /* Try to determine what/why we're actually resetting and swap + * src and dst ports */ + if (state == TCP_STATE_CLOSED) + { + if (!tcp_syn (th0)) + return -1; + + tmp = clib_net_to_host_u32 (th0->seq_number); + + /* Got a SYN for no listener. */ + th0->flags = TCP_FLAG_RST | TCP_FLAG_ACK; + th0->ack_number = clib_host_to_net_u32 (tmp + 1); + th0->seq_number = 0; + + } + else if (state >= TCP_STATE_SYN_SENT) + { + th0->flags = TCP_FLAG_RST | TCP_FLAG_ACK; + th0->seq_number = th0->ack_number; + th0->ack_number = 0; + } + + src_port0 = th0->src_port; + th0->src_port = th0->dst_port; + th0->dst_port = src_port0; + th0->window = 0; + th0->data_offset_and_reserved = (tcp_hdr_len >> 2) << 4; + th0->urgent_pointer = 0; + + /* Compute checksum */ + if (is_ip4) + { + th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih4); + } + else + { + int bogus = ~0; + th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih6, &bogus); + ASSERT (!bogus); + } + + return 0; +} + +/** + * Send reset without reusing existing buffer + */ +void +tcp_send_reset (vlib_buffer_t * pkt, u8 is_ip4) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u8 tcp_hdr_len, flags = 0; + tcp_header_t *th, *pkt_th; + u32 seq, ack; + ip4_header_t *ih4, *pkt_ih4; + ip6_header_t *ih6, *pkt_ih6; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + /* Make and write options */ + tcp_hdr_len = sizeof (tcp_header_t); + + if (is_ip4) + { + pkt_ih4 = vlib_buffer_get_current (pkt); + pkt_th = ip4_next_header (pkt_ih4); + } + else + { + pkt_ih6 = vlib_buffer_get_current (pkt); + pkt_th = ip6_next_header (pkt_ih6); + } + + if (tcp_ack (pkt_th)) + { + flags = TCP_FLAG_RST; + seq = pkt_th->ack_number; + ack = 0; + } + else + { + flags = TCP_FLAG_RST | TCP_FLAG_ACK; + seq = 0; + ack = clib_host_to_net_u32 (vnet_buffer (pkt)->tcp.seq_end); + } + + th = vlib_buffer_push_tcp_net_order (b, pkt_th->dst_port, pkt_th->src_port, + seq, ack, tcp_hdr_len, flags, 0); + + /* Swap src and dst ip */ + if (is_ip4) + { + ASSERT ((pkt_ih4->ip_version_and_header_length & 0xF0) == 0x40); + ih4 = vlib_buffer_push_ip4 (vm, b, &pkt_ih4->dst_address, + &pkt_ih4->src_address, IP_PROTOCOL_TCP); + th->checksum = ip4_tcp_udp_compute_checksum (vm, b, ih4); + } + else + { + int bogus = ~0; + pkt_ih6 = (ip6_header_t *) (pkt_th - 1); + ASSERT ((pkt_ih6->ip_version_traffic_class_and_flow_label & 0xF0) == + 0x60); + ih6 = + vlib_buffer_push_ip6 (vm, b, &pkt_ih6->dst_address, + &pkt_ih6->src_address, IP_PROTOCOL_TCP); + th->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b, ih6, &bogus); + ASSERT (!bogus); + } + + tcp_enqueue_to_ip_lookup (vm, b, bi, is_ip4); +} + +void +tcp_push_ip_hdr (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b) +{ + tcp_header_t *th = vlib_buffer_get_current (b); + + if (tc->c_is_ip4) + { + ip4_header_t *ih; + ih = vlib_buffer_push_ip4 (tm->vlib_main, b, &tc->c_lcl_ip4, + &tc->c_rmt_ip4, IP_PROTOCOL_TCP); + th->checksum = ip4_tcp_udp_compute_checksum (tm->vlib_main, b, ih); + } + else + { + ip6_header_t *ih; + int bogus = ~0; + + ih = vlib_buffer_push_ip6 (tm->vlib_main, b, &tc->c_lcl_ip6, + &tc->c_rmt_ip6, IP_PROTOCOL_TCP); + th->checksum = ip6_tcp_udp_icmp_compute_checksum (tm->vlib_main, b, ih, + &bogus); + ASSERT (!bogus); + } +} + +/** + * Send SYN + * + * Builds a SYN packet for a half-open connection and sends it to ipx_lookup. + * The packet is not forwarded through tcpx_output to avoid doing lookups + * in the half_open pool. + */ +void +tcp_send_syn (tcp_connection_t * tc) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u8 tcp_hdr_opts_len, tcp_opts_len; + tcp_header_t *th; + u32 time_now; + u16 initial_wnd; + tcp_options_t snd_opts; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + /* Set random initial sequence */ + time_now = tcp_time_now (); + + tc->iss = random_u32 (&time_now); + tc->snd_una = tc->iss; + tc->snd_una_max = tc->snd_nxt = tc->iss + 1; + + initial_wnd = tcp_initial_window_to_advertise (tc); + + /* Make and write options */ + memset (&snd_opts, 0, sizeof (snd_opts)); + tcp_opts_len = tcp_make_syn_options (&snd_opts, initial_wnd); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->iss, + tc->rcv_nxt, tcp_hdr_opts_len, TCP_FLAG_SYN, + initial_wnd); + + tcp_options_write ((u8 *) (th + 1), &snd_opts); + + /* Measure RTT with this */ + tc->rtt_ts = tcp_time_now (); + tc->rtt_seq = tc->snd_nxt; + + /* Start retransmit trimer */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN, tc->rto * TCP_TO_TIMER_TICK); + tc->rto_boff = 0; + + /* Set the connection establishment timer */ + tcp_timer_set (tc, TCP_TIMER_ESTABLISH, TCP_ESTABLISH_TIME); + + tcp_push_ip_hdr (tm, tc, b); + tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4); +} + +always_inline void +tcp_enqueue_to_output (vlib_main_t * vm, vlib_buffer_t * b, u32 bi, u8 is_ip4) +{ + u32 *to_next, next_index; + vlib_frame_t *f; + + b->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + b->error = 0; + + /* Decide where to send the packet */ + next_index = is_ip4 ? tcp4_output_node.index : tcp6_output_node.index; + f = vlib_get_frame_to_node (vm, next_index); + + /* Enqueue the packet */ + to_next = vlib_frame_vector_args (f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node (vm, next_index, f); +} + +/** + * Send FIN + */ +void +tcp_send_fin (tcp_connection_t * tc) +{ + vlib_buffer_t *b; + u32 bi; + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Leave enough space for headers */ + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + + tcp_make_finack (tc, b); + + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); +} + +always_inline u8 +tcp_make_state_flags (tcp_state_t next_state) +{ + switch (next_state) + { + case TCP_STATE_ESTABLISHED: + return TCP_FLAG_ACK; + case TCP_STATE_SYN_RCVD: + return TCP_FLAG_SYN | TCP_FLAG_ACK; + case TCP_STATE_SYN_SENT: + return TCP_FLAG_SYN; + case TCP_STATE_LAST_ACK: + case TCP_STATE_FIN_WAIT_1: + return TCP_FLAG_FIN; + default: + clib_warning ("Shouldn't be here!"); + } + return 0; +} + +/** + * Push TCP header and update connection variables + */ +static void +tcp_push_hdr_i (tcp_connection_t * tc, vlib_buffer_t * b, + tcp_state_t next_state) +{ + u32 advertise_wnd, data_len; + u8 tcp_opts_len, tcp_hdr_opts_len, opts_write_len, flags; + tcp_options_t _snd_opts, *snd_opts = &_snd_opts; + tcp_header_t *th; + + data_len = b->current_length; + vnet_buffer (b)->tcp.flags = 0; + + /* Make and write options */ + memset (snd_opts, 0, sizeof (*snd_opts)); + tcp_opts_len = tcp_make_options (tc, snd_opts, next_state); + tcp_hdr_opts_len = tcp_opts_len + sizeof (tcp_header_t); + + /* Get rcv window to advertise */ + advertise_wnd = tcp_window_to_advertise (tc, next_state); + flags = tcp_make_state_flags (next_state); + + /* Push header and options */ + th = vlib_buffer_push_tcp (b, tc->c_lcl_port, tc->c_rmt_port, tc->snd_nxt, + tc->rcv_nxt, tcp_hdr_opts_len, flags, + advertise_wnd); + + opts_write_len = tcp_options_write ((u8 *) (th + 1), snd_opts); + + ASSERT (opts_write_len == tcp_opts_len); + + /* Tag the buffer with the connection index */ + vnet_buffer (b)->tcp.connection_index = tc->c_c_index; + + tc->snd_nxt += data_len; +} + +/* Send delayed ACK when timer expires */ +void +tcp_timer_delack_handler (u32 index) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 thread_index = os_get_cpu_number (); + tcp_connection_t *tc; + vlib_buffer_t *b; + u32 bi; + + tc = tcp_connection_get (index, thread_index); + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + /* Fill in the ACK */ + tcp_make_ack (tc, b); + + tc->timers[TCP_TIMER_DELACK] = TCP_TIMER_HANDLE_INVALID; + tc->flags &= ~TCP_CONN_DELACK; + + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); +} + +/** Build a retransmit segment + * + * @return the number of bytes in the segment or 0 if there's nothing to + * retransmit + * */ +u32 +tcp_prepare_retransmit_segment (tcp_connection_t * tc, vlib_buffer_t * b, + u32 max_bytes) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 n_bytes, offset = 0; + sack_scoreboard_hole_t *hole; + u32 hole_size; + + tcp_reuse_buffer (vm, b); + + ASSERT (tc->state == TCP_STATE_ESTABLISHED); + ASSERT (max_bytes != 0); + + if (tcp_opts_sack_permitted (&tc->opt)) + { + /* XXX get first hole not retransmitted yet */ + hole = scoreboard_first_hole (&tc->sack_sb); + if (!hole) + return 0; + + offset = hole->start - tc->snd_una; + hole_size = hole->end - hole->start; + + ASSERT (hole_size); + + if (hole_size < max_bytes) + max_bytes = hole_size; + } + else + { + if (seq_geq (tc->snd_nxt, tc->snd_una_max)) + return 0; + } + + n_bytes = stream_session_peek_bytes (&tc->connection, + vlib_buffer_get_current (b), offset, + max_bytes); + ASSERT (n_bytes != 0); + + tc->snd_nxt += n_bytes; + tcp_push_hdr_i (tc, b, tc->state); + + return n_bytes; +} + +static void +tcp_timer_retransmit_handler_i (u32 index, u8 is_syn) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + vlib_main_t *vm = tm->vlib_main; + u32 thread_index = os_get_cpu_number (); + tcp_connection_t *tc; + vlib_buffer_t *b; + u32 bi, max_bytes, snd_space; + + if (is_syn) + { + tc = tcp_half_open_connection_get (index); + } + else + { + tc = tcp_connection_get (index, thread_index); + } + + /* Make sure timer handle is set to invalid */ + tc->timers[TCP_TIMER_RETRANSMIT] = TCP_TIMER_HANDLE_INVALID; + + /* Increment RTO backoff (also equal to number of retries) */ + tc->rto_boff += 1; + + /* Go back to first un-acked byte */ + tc->snd_nxt = tc->snd_una; + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (vm, bi); + + if (tc->state == TCP_STATE_ESTABLISHED) + { + tcp_fastrecovery_off (tc); + + /* Exponential backoff */ + tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX); + + /* Figure out what and how many bytes we can send */ + snd_space = tcp_available_snd_space (tc); + max_bytes = clib_min (tc->snd_mss, snd_space); + tcp_prepare_retransmit_segment (tc, b, max_bytes); + + tc->rtx_bytes += max_bytes; + + /* No fancy recovery for now! */ + scoreboard_clear (&tc->sack_sb); + } + else + { + /* Retransmit for SYN/SYNACK */ + ASSERT (tc->state == TCP_STATE_SYN_RCVD + || tc->state == TCP_STATE_SYN_SENT); + + /* Try without increasing RTO a number of times. If this fails, + * start growing RTO exponentially */ + if (tc->rto_boff > TCP_RTO_SYN_RETRIES) + tc->rto = clib_min (tc->rto << 1, TCP_RTO_MAX); + + vlib_buffer_make_headroom (b, MAX_HDRS_LEN); + tcp_push_hdr_i (tc, b, tc->state); + } + + if (!is_syn) + { + tcp_enqueue_to_output (vm, b, bi, tc->c_is_ip4); + + /* Re-enable retransmit timer */ + tcp_retransmit_timer_set (tm, tc); + } + else + { + ASSERT (tc->state == TCP_STATE_SYN_SENT); + + /* This goes straight to ipx_lookup */ + tcp_push_ip_hdr (tm, tc, b); + tcp_enqueue_to_ip_lookup (vm, b, bi, tc->c_is_ip4); + + /* Re-enable retransmit timer */ + tcp_timer_set (tc, TCP_TIMER_RETRANSMIT_SYN, + tc->rto * TCP_TO_TIMER_TICK); + } +} + +void +tcp_timer_retransmit_handler (u32 index) +{ + tcp_timer_retransmit_handler_i (index, 0); +} + +void +tcp_timer_retransmit_syn_handler (u32 index) +{ + tcp_timer_retransmit_handler_i (index, 1); +} + +/** + * Retansmit first unacked segment */ +void +tcp_retransmit_first_unacked (tcp_connection_t * tc) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 snd_nxt = tc->snd_nxt; + vlib_buffer_t *b; + u32 bi; + + tc->snd_nxt = tc->snd_una; + + /* Get buffer */ + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (tm->vlib_main, bi); + + tcp_prepare_retransmit_segment (tc, b, tc->snd_mss); + tcp_enqueue_to_output (tm->vlib_main, b, bi, tc->c_is_ip4); + + tc->snd_nxt = snd_nxt; + tc->rtx_bytes += tc->snd_mss; +} + +void +tcp_fast_retransmit (tcp_connection_t * tc) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 snd_space, max_bytes, n_bytes, bi; + vlib_buffer_t *b; + + ASSERT (tcp_in_fastrecovery (tc)); + + clib_warning ("fast retransmit!"); + + /* Start resending from first un-acked segment */ + tc->snd_nxt = tc->snd_una; + + snd_space = tcp_available_snd_space (tc); + + while (snd_space) + { + tcp_get_free_buffer_index (tm, &bi); + b = vlib_get_buffer (tm->vlib_main, bi); + + max_bytes = clib_min (tc->snd_mss, snd_space); + n_bytes = tcp_prepare_retransmit_segment (tc, b, max_bytes); + + /* Nothing left to retransmit */ + if (n_bytes == 0) + return; + + tcp_enqueue_to_output (tm->vlib_main, b, bi, tc->c_is_ip4); + + snd_space -= n_bytes; + } + + /* If window allows, send new data */ + tc->snd_nxt = tc->snd_una_max; +} + +always_inline u32 +tcp_session_has_ooo_data (tcp_connection_t * tc) +{ + stream_session_t *s = + stream_session_get (tc->c_s_index, tc->c_thread_index); + return svm_fifo_has_ooo_data (s->server_rx_fifo); +} + +always_inline uword +tcp46_output_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip4) +{ + tcp_main_t *tm = vnet_get_tcp_main (); + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + tcp_connection_t *tc0; + tcp_header_t *th0; + u32 error0 = TCP_ERROR_PKTS_SENT, next0 = TCP_OUTPUT_NEXT_IP_LOOKUP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, + my_thread_index); + th0 = vlib_buffer_get_current (b0); + + if (is_ip4) + { + ip4_header_t *ih0; + ih0 = vlib_buffer_push_ip4 (vm, b0, &tc0->c_lcl_ip4, + &tc0->c_rmt_ip4, IP_PROTOCOL_TCP); + th0->checksum = ip4_tcp_udp_compute_checksum (vm, b0, ih0); + } + else + { + ip6_header_t *ih0; + int bogus = ~0; + + ih0 = vlib_buffer_push_ip6 (vm, b0, &tc0->c_lcl_ip6, + &tc0->c_rmt_ip6, IP_PROTOCOL_TCP); + th0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0, ih0, + &bogus); + ASSERT (!bogus); + } + + /* Filter out DUPACKs if there are no OOO segments left */ + if (PREDICT_FALSE + (vnet_buffer (b0)->tcp.flags & TCP_BUF_FLAG_DUPACK)) + { + tc0->snt_dupacks--; + ASSERT (tc0->snt_dupacks >= 0); + if (!tcp_session_has_ooo_data (tc0)) + { + error0 = TCP_ERROR_FILTERED_DUPACKS; + next0 = TCP_OUTPUT_NEXT_DROP; + goto done; + } + } + + /* Retransmitted SYNs do reach this but it should be harmless */ + tc0->rcv_las = tc0->rcv_nxt; + + /* Stop DELACK timer and fix flags */ + tc0->flags &= + ~(TCP_CONN_SNDACK | TCP_CONN_DELACK | TCP_CONN_BURSTACK); + if (tcp_timer_is_active (tc0, TCP_TIMER_DELACK)) + { + tcp_timer_reset (tc0, TCP_TIMER_DELACK); + } + + /* If not retransmitting + * 1) update snd_una_max (SYN, SYNACK, new data, FIN) + * 2) If we're not tracking an ACK, start tracking */ + if (seq_lt (tc0->snd_una_max, tc0->snd_nxt)) + { + tc0->snd_una_max = tc0->snd_nxt; + if (tc0->rtt_ts == 0) + { + tc0->rtt_ts = tcp_time_now (); + tc0->rtt_seq = tc0->snd_nxt; + } + } + + /* Set the retransmit timer if not set already and not + * doing a pure ACK */ + if (!tcp_timer_is_active (tc0, TCP_TIMER_RETRANSMIT) + && tc0->snd_nxt != tc0->snd_una) + { + tcp_retransmit_timer_set (tm, tc0); + tc0->rto_boff = 0; + } + + /* set fib index to default and lookup node */ + /* XXX network virtualization (vrf/vni) */ + vnet_buffer (b0)->sw_if_index[VLIB_RX] = 0; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0; + + b0->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + + done: + b0->error = error0 != 0 ? node->errors[error0] : 0; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return from_frame->n_vectors; +} + +static uword +tcp4_output (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_output_inline (vm, node, from_frame, 1 /* is_ip4 */ ); +} + +static uword +tcp6_output (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_output_inline (vm, node, from_frame, 0 /* is_ip4 */ ); +} + +VLIB_REGISTER_NODE (tcp4_output_node) = +{ + .function = tcp4_output,.name = "tcp4-output", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32),.n_errors = TCP_N_ERROR,.error_strings = + tcp_error_strings,.n_next_nodes = TCP_OUTPUT_N_NEXT,.next_nodes = + { +#define _(s,n) [TCP_OUTPUT_NEXT_##s] = n, + foreach_tcp4_output_next +#undef _ + } +,.format_buffer = format_tcp_header,.format_trace = format_tcp_tx_trace,}; + +VLIB_NODE_FUNCTION_MULTIARCH (tcp4_output_node, tcp4_output) +VLIB_REGISTER_NODE (tcp6_output_node) = +{ + .function = tcp6_output,.name = "tcp6-output", + /* Takes a vector of packets. */ + .vector_size = sizeof (u32),.n_errors = TCP_N_ERROR,.error_strings = + tcp_error_strings,.n_next_nodes = TCP_OUTPUT_N_NEXT,.next_nodes = + { +#define _(s,n) [TCP_OUTPUT_NEXT_##s] = n, + foreach_tcp6_output_next +#undef _ + } +,.format_buffer = format_tcp_header,.format_trace = format_tcp_tx_trace,}; + +VLIB_NODE_FUNCTION_MULTIARCH (tcp6_output_node, tcp6_output) u32 +tcp_push_header (transport_connection_t * tconn, vlib_buffer_t * b) +{ + tcp_connection_t *tc; + + tc = (tcp_connection_t *) tconn; + tcp_push_hdr_i (tc, b, TCP_STATE_ESTABLISHED); + return 0; +} + +typedef enum _tcp_reset_next +{ + TCP_RESET_NEXT_DROP, + TCP_RESET_NEXT_IP_LOOKUP, + TCP_RESET_N_NEXT +} tcp_reset_next_t; + +#define foreach_tcp4_reset_next \ + _(DROP, "error-drop") \ + _(IP_LOOKUP, "ip4-lookup") + +#define foreach_tcp6_reset_next \ + _(DROP, "error-drop") \ + _(IP_LOOKUP, "ip6-lookup") + +static uword +tcp46_send_reset_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame, u8 is_ip4) +{ + u32 n_left_from, next_index, *from, *to_next; + u32 my_thread_index = vm->cpu_index; + + from = vlib_frame_vector_args (from_frame); + n_left_from = from_frame->n_vectors; + + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 error0 = TCP_ERROR_RST_SENT, next0 = TCP_RESET_NEXT_IP_LOOKUP; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + if (tcp_make_reset_in_place (vm, b0, vnet_buffer (b0)->tcp.flags, + my_thread_index, is_ip4)) + { + error0 = TCP_ERROR_LOOKUP_DROPS; + next0 = TCP_RESET_NEXT_DROP; + goto done; + } + + /* Prepare to send to IP lookup */ + vnet_buffer (b0)->sw_if_index[VLIB_TX] = 0; + next0 = TCP_RESET_NEXT_IP_LOOKUP; + + done: + b0->error = error0 != 0 ? node->errors[error0] : 0; + b0->flags |= VNET_BUFFER_LOCALLY_ORIGINATED; + if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) + { + + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, bi0, next0); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + return from_frame->n_vectors; +} + +static uword +tcp4_send_reset (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_send_reset_inline (vm, node, from_frame, 1); +} + +static uword +tcp6_send_reset (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return tcp46_send_reset_inline (vm, node, from_frame, 0); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp4_reset_node) = { + .function = tcp4_send_reset, + .name = "tcp4-reset", + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RESET_N_NEXT, + .next_nodes = { +#define _(s,n) [TCP_RESET_NEXT_##s] = n, + foreach_tcp4_reset_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (tcp6_reset_node) = { + .function = tcp6_send_reset, + .name = "tcp6-reset", + .vector_size = sizeof (u32), + .n_errors = TCP_N_ERROR, + .error_strings = tcp_error_strings, + .n_next_nodes = TCP_RESET_N_NEXT, + .next_nodes = { +#define _(s,n) [TCP_RESET_NEXT_##s] = n, + foreach_tcp6_reset_next +#undef _ + }, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_packet.h b/src/vnet/tcp/tcp_packet.h new file mode 100644 index 00000000000..866c5fd6046 --- /dev/null +++ b/src/vnet/tcp/tcp_packet.h @@ -0,0 +1,184 @@ +/* + * 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. + */ + +#ifndef included_tcp_packet_h +#define included_tcp_packet_h + +#include + +/* TCP flags bit 0 first. */ +#define foreach_tcp_flag \ + _ (FIN) /**< No more data from sender. */ \ + _ (SYN) /**< Synchronize sequence numbers. */ \ + _ (RST) /**< Reset the connection. */ \ + _ (PSH) /**< Push function. */ \ + _ (ACK) /**< Ack field significant. */ \ + _ (URG) /**< Urgent pointer field significant. */ \ + _ (ECE) /**< ECN-echo. Receiver got CE packet */ \ + _ (CWR) /**< Sender reduced congestion window */ + +enum +{ +#define _(f) TCP_FLAG_BIT_##f, + foreach_tcp_flag +#undef _ + TCP_N_FLAG_BITS, +}; + +enum +{ +#define _(f) TCP_FLAG_##f = 1 << TCP_FLAG_BIT_##f, + foreach_tcp_flag +#undef _ +}; + +typedef struct _tcp_header +{ + union + { + struct + { + u16 src_port; /**< Source port. */ + u16 dst_port; /**< Destination port. */ + }; + struct + { + u16 src, dst; + }; + }; + + u32 seq_number; /**< Sequence number of the first data octet in this + * segment, except when SYN is present. If SYN + * is present the seq number is is the ISN and the + * first data octet is ISN+1 */ + u32 ack_number; /**< Acknowledgement number if ACK is set. It contains + * the value of the next sequence number the sender + * of the segment is expecting to receive. */ + u8 data_offset_and_reserved; + u8 flags; /**< Flags: see the macro above */ + u16 window; /**< Number of bytes sender is willing to receive. */ + + u16 checksum; /**< Checksum of TCP pseudo header and data. */ + u16 urgent_pointer; /**< Seq number of the byte after the urgent data. */ +} __attribute__ ((packed)) tcp_header_t; + +/* Flag tests that return 0 or !0 */ +#define tcp_doff(_th) ((_th)->data_offset_and_reserved >> 4) +#define tcp_fin(_th) ((_th)->flags & TCP_FLAG_FIN) +#define tcp_syn(_th) ((_th)->flags & TCP_FLAG_SYN) +#define tcp_rst(_th) ((_th)->flags & TCP_FLAG_RST) +#define tcp_psh(_th) ((_th)->flags & TCP_FLAG_PSH) +#define tcp_ack(_th) ((_th)->flags & TCP_FLAG_ACK) +#define tcp_urg(_th) ((_th)->flags & TCP_FLAG_URG) +#define tcp_ece(_th) ((_th)->flags & TCP_FLAG_ECE) +#define tcp_cwr(_th) ((_th)->flags & TCP_FLAG_CWR) + +/* Flag tests that return 0 or 1 */ +#define tcp_is_syn(_th) !!((_th)->flags & TCP_FLAG_SYN) +#define tcp_is_fin(_th) !!((_th)->flags & TCP_FLAG_FIN) + +always_inline int +tcp_header_bytes (tcp_header_t * t) +{ + return tcp_doff (t) * sizeof (u32); +} + +/* + * TCP options. + */ + +typedef enum tcp_option_type +{ + TCP_OPTION_EOL = 0, /**< End of options. */ + TCP_OPTION_NOOP = 1, /**< No operation. */ + TCP_OPTION_MSS = 2, /**< Limit MSS. */ + TCP_OPTION_WINDOW_SCALE = 3, /**< Window scale. */ + TCP_OPTION_SACK_PERMITTED = 4, /**< Selective Ack permitted. */ + TCP_OPTION_SACK_BLOCK = 5, /**< Selective Ack block. */ + TCP_OPTION_TIMESTAMP = 8, /**< Timestamps. */ + TCP_OPTION_UTO = 28, /**< User timeout. */ + TCP_OPTION_AO = 29, /**< Authentication Option. */ +} tcp_option_type_t; + +#define foreach_tcp_options_flag \ + _ (MSS) /**< MSS advertised in SYN */ \ + _ (TSTAMP) /**< Timestamp capability advertised in SYN */ \ + _ (WSCALE) /**< Wnd scale capability advertised in SYN */ \ + _ (SACK_PERMITTED) /**< SACK capability advertised in SYN */ \ + _ (SACK) /**< SACK present */ + +enum +{ +#define _(f) TCP_OPTS_FLAG_BIT_##f, + foreach_tcp_options_flag +#undef _ + TCP_OPTIONS_N_FLAG_BITS, +}; + +enum +{ +#define _(f) TCP_OPTS_FLAG_##f = 1 << TCP_OPTS_FLAG_BIT_##f, + foreach_tcp_options_flag +#undef _ +}; + +typedef struct _sack_block +{ + u32 start; /**< Start sequence number */ + u32 end; /**< End sequence number */ +} sack_block_t; + +typedef struct +{ + u8 flags; /** Option flags, see above */ + + /* Received options */ + u16 mss; /**< Maximum segment size advertised by peer */ + u8 wscale; /**< Window scale advertised by peer */ + u32 tsval; /**< Peer's timestamp value */ + u32 tsecr; /**< Echoed/reflected time stamp */ + sack_block_t *sacks; /**< SACK blocks received */ + u8 n_sack_blocks; /**< Number of SACKs blocks */ +} tcp_options_t; + +/* Flag tests that return 0 or !0 */ +#define tcp_opts_mss(_to) ((_to)->flags & TCP_OPTS_FLAG_MSS) +#define tcp_opts_tstamp(_to) ((_to)->flags & TCP_OPTS_FLAG_TSTAMP) +#define tcp_opts_wscale(_to) ((_to)->flags & TCP_OPTS_FLAG_WSCALE) +#define tcp_opts_sack(_to) ((_to)->flags & TCP_OPTS_FLAG_SACK) +#define tcp_opts_sack_permitted(_to) ((_to)->flags & TCP_OPTS_FLAG_SACK_PERMITTED) + +/* TCP option lengths */ +#define TCP_OPTION_LEN_EOL 1 +#define TCP_OPTION_LEN_NOOP 1 +#define TCP_OPTION_LEN_MSS 4 +#define TCP_OPTION_LEN_WINDOW_SCALE 3 +#define TCP_OPTION_LEN_SACK_PERMITTED 2 +#define TCP_OPTION_LEN_TIMESTAMP 10 +#define TCP_OPTION_LEN_SACK_BLOCK 8 + +#define TCP_WND_MAX 65535U +#define TCP_MAX_WND_SCALE 14 /* See RFC 1323 */ +#define TCP_OPTS_ALIGN 4 +#define TCP_OPTS_MAX_SACK_BLOCKS 3 +#endif /* included_tcp_packet_h */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_pg.c b/src/vnet/tcp/tcp_pg.c new file mode 100644 index 00000000000..dc3240496df --- /dev/null +++ b/src/vnet/tcp/tcp_pg.c @@ -0,0 +1,236 @@ +/* + * 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. + */ +/* + * ip/tcp_pg: TCP packet-generator interface + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include +#include + +/* TCP flags bit 0 first. */ +#define foreach_tcp_flag \ + _ (FIN) \ + _ (SYN) \ + _ (RST) \ + _ (PSH) \ + _ (ACK) \ + _ (URG) \ + _ (ECE) \ + _ (CWR) + +static void +tcp_pg_edit_function (pg_main_t * pg, + pg_stream_t * s, + pg_edit_group_t * g, + u32 * packets, + u32 n_packets) +{ + vlib_main_t * vm = vlib_get_main(); + u32 ip_offset, tcp_offset; + + tcp_offset = g->start_byte_offset; + ip_offset = (g-1)->start_byte_offset; + + while (n_packets >= 1) + { + vlib_buffer_t * p0; + ip4_header_t * ip0; + tcp_header_t * tcp0; + ip_csum_t sum0; + u32 tcp_len0; + + p0 = vlib_get_buffer (vm, packets[0]); + n_packets -= 1; + packets += 1; + + ASSERT (p0->current_data == 0); + ip0 = (void *) (p0->data + ip_offset); + tcp0 = (void *) (p0->data + tcp_offset); + tcp_len0 = clib_net_to_host_u16 (ip0->length) - sizeof (ip0[0]); + + /* Initialize checksum with header. */ + if (BITS (sum0) == 32) + { + sum0 = clib_mem_unaligned (&ip0->src_address, u32); + sum0 = ip_csum_with_carry (sum0, clib_mem_unaligned (&ip0->dst_address, u32)); + } + else + sum0 = clib_mem_unaligned (&ip0->src_address, u64); + + sum0 = ip_csum_with_carry + (sum0, clib_host_to_net_u32 (tcp_len0 + (ip0->protocol << 16))); + + /* Invalidate possibly old checksum. */ + tcp0->checksum = 0; + + sum0 = ip_incremental_checksum_buffer (vm, p0, tcp_offset, tcp_len0, sum0); + + tcp0->checksum = ~ ip_csum_fold (sum0); + } +} + +typedef struct { + pg_edit_t src, dst; + pg_edit_t seq_number, ack_number; + pg_edit_t data_offset_and_reserved; +#define _(f) pg_edit_t f##_flag; + foreach_tcp_flag +#undef _ + pg_edit_t window; + pg_edit_t checksum; + pg_edit_t urgent_pointer; +} pg_tcp_header_t; + +static inline void +pg_tcp_header_init (pg_tcp_header_t * p) +{ + /* Initialize fields that are not bit fields in the IP header. */ +#define _(f) pg_edit_init (&p->f, tcp_header_t, f); + _ (src); + _ (dst); + _ (seq_number); + _ (ack_number); + _ (window); + _ (checksum); + _ (urgent_pointer); +#undef _ + + /* Initialize bit fields. */ +#define _(f) \ + pg_edit_init_bitfield (&p->f##_flag, tcp_header_t, \ + flags, \ + TCP_FLAG_BIT_##f, 1); + + foreach_tcp_flag +#undef _ + + pg_edit_init_bitfield (&p->data_offset_and_reserved, tcp_header_t, + data_offset_and_reserved, + 4, 4); +} + +uword +unformat_pg_tcp_header (unformat_input_t * input, va_list * args) +{ + pg_stream_t * s = va_arg (*args, pg_stream_t *); + pg_tcp_header_t * p; + u32 group_index; + + p = pg_create_edit_group (s, sizeof (p[0]), sizeof (tcp_header_t), + &group_index); + pg_tcp_header_init (p); + + /* Defaults. */ + pg_edit_set_fixed (&p->seq_number, 0); + pg_edit_set_fixed (&p->ack_number, 0); + + pg_edit_set_fixed (&p->data_offset_and_reserved, + sizeof (tcp_header_t) / sizeof (u32)); + + pg_edit_set_fixed (&p->window, 4096); + pg_edit_set_fixed (&p->urgent_pointer, 0); + +#define _(f) pg_edit_set_fixed (&p->f##_flag, 0); + foreach_tcp_flag +#undef _ + + p->checksum.type = PG_EDIT_UNSPECIFIED; + + if (! unformat (input, "TCP: %U -> %U", + unformat_pg_edit, + unformat_tcp_udp_port, &p->src, + unformat_pg_edit, + unformat_tcp_udp_port, &p->dst)) + goto error; + + /* Parse options. */ + while (1) + { + if (unformat (input, "window %U", + unformat_pg_edit, + unformat_pg_number, &p->window)) + ; + + else if (unformat (input, "checksum %U", + unformat_pg_edit, + unformat_pg_number, &p->checksum)) + ; + + /* Flags. */ +#define _(f) else if (unformat (input, #f)) pg_edit_set_fixed (&p->f##_flag, 1); + foreach_tcp_flag +#undef _ + + /* Can't parse input: try next protocol level. */ + else + break; + } + + { + ip_main_t * im = &ip_main; + u16 dst_port; + tcp_udp_port_info_t * pi; + + pi = 0; + if (p->dst.type == PG_EDIT_FIXED) + { + dst_port = pg_edit_get_value (&p->dst, PG_EDIT_LO); + pi = ip_get_tcp_udp_port_info (im, dst_port); + } + + if (pi && pi->unformat_pg_edit + && unformat_user (input, pi->unformat_pg_edit, s)) + ; + + else if (! unformat_user (input, unformat_pg_payload, s)) + goto error; + + if (p->checksum.type == PG_EDIT_UNSPECIFIED) + { + pg_edit_group_t * g = pg_stream_get_group (s, group_index); + g->edit_function = tcp_pg_edit_function; + g->edit_function_opaque = 0; + } + + return 1; + } + + error: + /* Free up any edits we may have added. */ + pg_free_edit_group (s); + return 0; +} + diff --git a/src/vnet/tcp/tcp_syn_filter4.c b/src/vnet/tcp/tcp_syn_filter4.c new file mode 100644 index 00000000000..c7605a30c98 --- /dev/null +++ b/src/vnet/tcp/tcp_syn_filter4.c @@ -0,0 +1,542 @@ +/* + * 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. + */ + +#include +#include +#include +#include +#include +#include +#include + +typedef struct +{ + f64 next_reset; + f64 reset_interval; + u8 *syn_counts; +} syn_filter4_runtime_t; + +typedef struct +{ + u32 next_index; + int not_a_syn; + u8 filter_value; +} syn_filter4_trace_t; + +/* packet trace format function */ +static u8 * +format_syn_filter4_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + syn_filter4_trace_t *t = va_arg (*args, syn_filter4_trace_t *); + + s = format (s, "SYN_FILTER4: next index %d, %s", + t->next_index, t->not_a_syn ? "not a syn" : "syn"); + if (t->not_a_syn == 0) + s = format (s, ", filter value %d\n", t->filter_value); + else + s = format (s, "\n"); + return s; +} + +static vlib_node_registration_t syn_filter4_node; + +#define foreach_syn_filter_error \ +_(THROTTLED, "TCP SYN packet throttle drops") \ +_(OK, "TCP SYN packets passed") + +typedef enum +{ +#define _(sym,str) SYN_FILTER_ERROR_##sym, + foreach_syn_filter_error +#undef _ + SYN_FILTER_N_ERROR, +} syn_filter_error_t; + +static char *syn_filter4_error_strings[] = { +#define _(sym,string) string, + foreach_syn_filter_error +#undef _ +}; + +typedef enum +{ + SYN_FILTER_NEXT_DROP, + SYN_FILTER_N_NEXT, +} syn_filter_next_t; + +extern vnet_feature_arc_registration_t vnet_feat_arc_ip4_local; + +static uword +syn_filter4_node_fn (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + u32 n_left_from, *from, *to_next; + syn_filter_next_t next_index; + u32 ok_syn_packets = 0; + vnet_feature_main_t *fm = &feature_main; + u8 arc_index = vnet_feat_arc_ip4_local.feature_arc_index; + vnet_feature_config_main_t *cm = &fm->feature_config_mains[arc_index]; + syn_filter4_runtime_t *rt = (syn_filter4_runtime_t *) node->runtime_data; + f64 now = vlib_time_now (vm); + /* Shut up spurious gcc warnings. */ + u8 *c0 = 0, *c1 = 0, *c2 = 0, *c3 = 0; + + from = vlib_frame_vector_args (frame); + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + if (now > rt->next_reset) + { + memset (rt->syn_counts, 0, vec_len (rt->syn_counts)); + rt->next_reset = now + rt->reset_interval; + } + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from >= 8 && n_left_to_next >= 4) + { + u32 bi0, bi1, bi2, bi3; + vlib_buffer_t *b0, *b1, *b2, *b3; + u32 next0, next1, next2, next3; + ip4_header_t *ip0, *ip1, *ip2, *ip3; + tcp_header_t *tcp0, *tcp1, *tcp2, *tcp3; + u32 not_a_syn0 = 1, not_a_syn1 = 1, not_a_syn2 = 1, not_a_syn3 = 1; + u64 hash0, hash1, hash2, hash3; + + /* Prefetch next iteration. */ + { + vlib_buffer_t *p4, *p5, *p6, *p7; + + p4 = vlib_get_buffer (vm, from[4]); + p5 = vlib_get_buffer (vm, from[5]); + p6 = vlib_get_buffer (vm, from[6]); + p7 = vlib_get_buffer (vm, from[7]); + + vlib_prefetch_buffer_header (p4, LOAD); + vlib_prefetch_buffer_header (p5, LOAD); + vlib_prefetch_buffer_header (p6, LOAD); + vlib_prefetch_buffer_header (p7, LOAD); + + CLIB_PREFETCH (p4->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p5->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p6->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p7->data, CLIB_CACHE_LINE_BYTES, STORE); + } + + /* speculatively enqueue b0 and b1 to the current next frame */ + to_next[0] = bi0 = from[0]; + to_next[1] = bi1 = from[1]; + to_next[2] = bi2 = from[2]; + to_next[3] = bi3 = from[3]; + from += 4; + to_next += 4; + n_left_from -= 4; + n_left_to_next -= 4; + + b0 = vlib_get_buffer (vm, bi0); + b1 = vlib_get_buffer (vm, bi1); + b2 = vlib_get_buffer (vm, bi2); + b3 = vlib_get_buffer (vm, bi3); + + vnet_get_config_data + (&cm->config_main, &b0->current_config_index, + &next0, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b1->current_config_index, + &next1, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b2->current_config_index, + &next2, 0 /* sizeof (c0[0]) */ ); + vnet_get_config_data + (&cm->config_main, &b3->current_config_index, + &next3, 0 /* sizeof (c0[0]) */ ); + + /* Not TCP? */ + ip0 = vlib_buffer_get_current (b0); + if (ip0->protocol != IP_PROTOCOL_TCP) + goto trace00; + + tcp0 = ip4_next_header (ip0); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp0->flags & 0x2))) + goto trace00; + + not_a_syn0 = 0; + hash0 = clib_xxhash ((u64) ip0->src_address.as_u32); + c0 = &rt->syn_counts[hash0 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c0 >= 0x80)) + { + next0 = SYN_FILTER_NEXT_DROP; + b0->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace00; + } + *c0 += 1; + ok_syn_packets++; + + trace00: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->not_a_syn = not_a_syn0; + t->next_index = next0; + t->filter_value = not_a_syn0 ? 0 : *c0; + } + + /* Not TCP? */ + ip1 = vlib_buffer_get_current (b1); + if (ip1->protocol != IP_PROTOCOL_TCP) + goto trace01; + + tcp1 = ip4_next_header (ip1); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp1->flags & 0x2))) + goto trace01; + + not_a_syn1 = 0; + hash1 = clib_xxhash ((u64) ip1->src_address.as_u32); + c1 = &rt->syn_counts[hash1 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c1 >= 0x80)) + { + next1 = SYN_FILTER_NEXT_DROP; + b1->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace01; + } + *c1 += 1; + ok_syn_packets++; + + trace01: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b1->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b1, sizeof (*t)); + t->not_a_syn = not_a_syn1; + t->next_index = next1; + t->filter_value = not_a_syn1 ? 0 : *c1; + } + + /* Not TCP? */ + ip2 = vlib_buffer_get_current (b2); + if (ip2->protocol != IP_PROTOCOL_TCP) + goto trace02; + + tcp2 = ip4_next_header (ip2); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp2->flags & 0x2))) + goto trace02; + + not_a_syn2 = 0; + hash2 = clib_xxhash ((u64) ip2->src_address.as_u32); + c2 = &rt->syn_counts[hash2 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c2 >= 0x80)) + { + next2 = SYN_FILTER_NEXT_DROP; + b2->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace02; + } + *c2 += 1; + ok_syn_packets++; + + trace02: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b2->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b2, sizeof (*t)); + t->not_a_syn = not_a_syn2; + t->next_index = next2; + t->filter_value = not_a_syn2 ? 0 : *c2; + } + + /* Not TCP? */ + ip3 = vlib_buffer_get_current (b3); + if (ip3->protocol != IP_PROTOCOL_TCP) + goto trace03; + + tcp3 = ip4_next_header (ip3); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp3->flags & 0x2))) + goto trace03; + + not_a_syn3 = 0; + hash3 = clib_xxhash ((u64) ip3->src_address.as_u32); + c3 = &rt->syn_counts[hash3 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c3 >= 0x80)) + { + next3 = SYN_FILTER_NEXT_DROP; + b3->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace03; + } + *c3 += 1; + ok_syn_packets++; + + trace03: + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b3->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b3, sizeof (*t)); + t->not_a_syn = not_a_syn3; + t->next_index = next3; + t->filter_value = not_a_syn3 ? 0 : *c3; + } + vlib_validate_buffer_enqueue_x4 (vm, node, next_index, + to_next, n_left_to_next, + bi0, bi1, bi2, bi3, + next0, next1, next2, next3); + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t *b0; + u32 next0; + ip4_header_t *ip0; + tcp_header_t *tcp0; + u32 not_a_syn0 = 1; + u32 hash0; + u8 *c0; + + /* speculatively enqueue b0 to the current next frame */ + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + + vnet_get_config_data + (&cm->config_main, &b0->current_config_index, + &next0, 0 /* sizeof (c0[0]) */ ); + + /* Not TCP? */ + ip0 = vlib_buffer_get_current (b0); + if (ip0->protocol != IP_PROTOCOL_TCP) + goto trace0; + + tcp0 = ip4_next_header (ip0); + /* + * Not a SYN? + * $$$$ hack: the TCP bitfield flags seem not to compile + * correct code. + */ + if (PREDICT_TRUE (!(tcp0->flags & 0x2))) + goto trace0; + + not_a_syn0 = 0; + hash0 = clib_xxhash ((u64) ip0->src_address.as_u32); + c0 = &rt->syn_counts[hash0 & (_vec_len (rt->syn_counts) - 1)]; + if (PREDICT_FALSE (*c0 >= 0x80)) + { + next0 = SYN_FILTER_NEXT_DROP; + b0->error = node->errors[SYN_FILTER_ERROR_THROTTLED]; + goto trace0; + } + *c0 += 1; + ok_syn_packets++; + + trace0: + + if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + syn_filter4_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->not_a_syn = not_a_syn0; + t->next_index = next0; + t->filter_value = not_a_syn0 ? 0 : *c0; + } + + /* verify speculative enqueue, maybe switch current next frame */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + vlib_node_increment_counter (vm, syn_filter4_node.index, + SYN_FILTER_ERROR_OK, ok_syn_packets); + return frame->n_vectors; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (syn_filter4_node, static) = +{ + .function = syn_filter4_node_fn, + .name = "syn-filter-4", + .vector_size = sizeof (u32), + .format_trace = format_syn_filter4_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .runtime_data_bytes = sizeof (syn_filter4_runtime_t), + .n_errors = ARRAY_LEN(syn_filter4_error_strings), + .error_strings = syn_filter4_error_strings, + + .n_next_nodes = SYN_FILTER_N_NEXT, + + /* edit / add dispositions here */ + .next_nodes = { + [SYN_FILTER_NEXT_DROP] = "error-drop", + }, +}; +/* *INDENT-ON* */ + +VLIB_NODE_FUNCTION_MULTIARCH (syn_filter4_node, syn_filter4_node_fn); + +/* *INDENT-OFF* */ +VNET_FEATURE_INIT (syn_filter_4, static) = +{ + .arc_name = "ip4-local", + .node_name = "syn-filter-4", + .runs_before = VNET_FEATURES("ip4-local-end-of-arc"), +}; +/* *INDENT-ON* */ + +int +syn_filter_enable_disable (u32 sw_if_index, int enable_disable) +{ + vnet_main_t *vnm = vnet_get_main (); + vnet_sw_interface_t *sw; + int rv = 0; + + /* Utterly wrong? */ + if (pool_is_free_index (vnm->interface_main.sw_interfaces, sw_if_index)) + return VNET_API_ERROR_INVALID_SW_IF_INDEX; + + /* Not a physical port? */ + sw = vnet_get_sw_interface (vnm, sw_if_index); + if (sw->type != VNET_SW_INTERFACE_TYPE_HARDWARE) + return VNET_API_ERROR_INVALID_SW_IF_INDEX; + + if (enable_disable) + { + vlib_main_t *vm = vlib_get_main (); + syn_filter4_runtime_t *rt; + + rt = vlib_node_get_runtime_data (vm, syn_filter4_node.index); + vec_validate (rt->syn_counts, 1023); + /* + * Given perfect disperson / optimal hashing results: + * Allow 128k (successful) syns/sec. 1024, buckets each of which + * absorb 128 syns before filtering. Reset table once a second. + * Reality bites, lets try resetting once every 100ms. + */ + rt->reset_interval = 0.1; /* reset interval in seconds */ + } + + rv = vnet_feature_enable_disable ("ip4-local", "syn-filter-4", + sw_if_index, enable_disable, 0, 0); + + return rv; +} + +static clib_error_t * +syn_filter_enable_disable_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + vnet_main_t *vnm = vnet_get_main (); + u32 sw_if_index = ~0; + int enable_disable = 1; + int rv; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "disable")) + enable_disable = 0; + else if (unformat (input, "%U", unformat_vnet_sw_interface, + vnm, &sw_if_index)) + ; + else + break; + } + + if (sw_if_index == ~0) + return clib_error_return (0, "Please specify an interface..."); + + rv = syn_filter_enable_disable (sw_if_index, enable_disable); + + switch (rv) + { + case 0: + break; + + case VNET_API_ERROR_INVALID_SW_IF_INDEX: + return clib_error_return + (0, "Invalid interface, only works on physical ports"); + break; + + case VNET_API_ERROR_UNIMPLEMENTED: + return clib_error_return (0, + "Device driver doesn't support redirection"); + break; + + case VNET_API_ERROR_INVALID_VALUE: + return clib_error_return (0, "feature arc not found"); + + case VNET_API_ERROR_INVALID_VALUE_2: + return clib_error_return (0, "feature node not found"); + + default: + return clib_error_return (0, "syn_filter_enable_disable returned %d", + rv); + } + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (sr_content_command, static) = +{ + .path = "ip syn filter", + .short_help = "ip syn filter [disable]", + .function = syn_filter_enable_disable_command_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/tcp/tcp_timer.h b/src/vnet/tcp/tcp_timer.h new file mode 100644 index 00000000000..fa25268c1bf --- /dev/null +++ b/src/vnet/tcp/tcp_timer.h @@ -0,0 +1,29 @@ +/* + * 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. + */ +#ifndef __included_tcp_timer_h__ +#define __included_tcp_timer_h__ + +#include +#include + +#endif /* __included_tcp_timer_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ -- cgit 1.2.3-korg