Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016 Cisco and/or its affiliates. |
| 3 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | * you may not use this file except in compliance with the License. |
| 5 | * You may obtain a copy of the License at: |
| 6 | * |
| 7 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | * |
| 9 | * Unless required by applicable law or agreed to in writing, software |
| 10 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | * See the License for the specific language governing permissions and |
| 13 | * limitations under the License. |
| 14 | */ |
| 15 | |
| 16 | #include <vppinfra/sparse_vec.h> |
| 17 | #include <vnet/tcp/tcp_packet.h> |
| 18 | #include <vnet/tcp/tcp.h> |
| 19 | #include <vnet/session/session.h> |
| 20 | #include <math.h> |
| 21 | |
| 22 | static char *tcp_error_strings[] = { |
| 23 | #define tcp_error(n,s) s, |
| 24 | #include <vnet/tcp/tcp_error.def> |
| 25 | #undef tcp_error |
| 26 | }; |
| 27 | |
| 28 | /* All TCP nodes have the same outgoing arcs */ |
| 29 | #define foreach_tcp_state_next \ |
| 30 | _ (DROP, "error-drop") \ |
| 31 | _ (TCP4_OUTPUT, "tcp4-output") \ |
| 32 | _ (TCP6_OUTPUT, "tcp6-output") |
| 33 | |
| 34 | typedef enum _tcp_established_next |
| 35 | { |
| 36 | #define _(s,n) TCP_ESTABLISHED_NEXT_##s, |
| 37 | foreach_tcp_state_next |
| 38 | #undef _ |
| 39 | TCP_ESTABLISHED_N_NEXT, |
| 40 | } tcp_established_next_t; |
| 41 | |
| 42 | typedef enum _tcp_rcv_process_next |
| 43 | { |
| 44 | #define _(s,n) TCP_RCV_PROCESS_NEXT_##s, |
| 45 | foreach_tcp_state_next |
| 46 | #undef _ |
| 47 | TCP_RCV_PROCESS_N_NEXT, |
| 48 | } tcp_rcv_process_next_t; |
| 49 | |
| 50 | typedef enum _tcp_syn_sent_next |
| 51 | { |
| 52 | #define _(s,n) TCP_SYN_SENT_NEXT_##s, |
| 53 | foreach_tcp_state_next |
| 54 | #undef _ |
| 55 | TCP_SYN_SENT_N_NEXT, |
| 56 | } tcp_syn_sent_next_t; |
| 57 | |
| 58 | typedef enum _tcp_listen_next |
| 59 | { |
| 60 | #define _(s,n) TCP_LISTEN_NEXT_##s, |
| 61 | foreach_tcp_state_next |
| 62 | #undef _ |
| 63 | TCP_LISTEN_N_NEXT, |
| 64 | } tcp_listen_next_t; |
| 65 | |
| 66 | /* Generic, state independent indices */ |
| 67 | typedef enum _tcp_state_next |
| 68 | { |
| 69 | #define _(s,n) TCP_NEXT_##s, |
| 70 | foreach_tcp_state_next |
| 71 | #undef _ |
| 72 | TCP_STATE_N_NEXT, |
| 73 | } tcp_state_next_t; |
| 74 | |
| 75 | #define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \ |
| 76 | : TCP_NEXT_TCP6_OUTPUT) |
| 77 | |
| 78 | vlib_node_registration_t tcp4_established_node; |
| 79 | vlib_node_registration_t tcp6_established_node; |
| 80 | |
| 81 | /** |
| 82 | * Validate segment sequence number. As per RFC793: |
| 83 | * |
| 84 | * Segment Receive Test |
| 85 | * Length Window |
| 86 | * ------- ------- ------------------------------------------- |
| 87 | * 0 0 SEG.SEQ = RCV.NXT |
| 88 | * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND |
| 89 | * >0 0 not acceptable |
| 90 | * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND |
| 91 | * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND |
| 92 | * |
| 93 | * This ultimately consists in checking if segment falls within the window. |
| 94 | * The one important difference compared to RFC793 is that we use rcv_las, |
| 95 | * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the |
| 96 | * peer's reference when computing our receive window. |
| 97 | * |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 98 | * This: |
| 99 | * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las) |
| 100 | * however, is too strict when we have retransmits. Instead we just check that |
| 101 | * the seq is not beyond the right edge and that the end of the segment is not |
| 102 | * less than the left edge. |
| 103 | * |
| 104 | * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so |
| 105 | * use rcv_nxt in the right edge window test instead of rcv_las. |
| 106 | * |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 107 | */ |
| 108 | always_inline u8 |
| 109 | tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) |
| 110 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 111 | return (seq_geq (end_seq, tc->rcv_las) |
| 112 | && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 113 | } |
| 114 | |
| 115 | void |
| 116 | tcp_options_parse (tcp_header_t * th, tcp_options_t * to) |
| 117 | { |
| 118 | const u8 *data; |
| 119 | u8 opt_len, opts_len, kind; |
| 120 | int j; |
| 121 | sack_block_t b; |
| 122 | |
| 123 | opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t); |
| 124 | data = (const u8 *) (th + 1); |
| 125 | |
| 126 | /* Zero out all flags but those set in SYN */ |
| 127 | to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE); |
| 128 | |
| 129 | for (; opts_len > 0; opts_len -= opt_len, data += opt_len) |
| 130 | { |
| 131 | kind = data[0]; |
| 132 | |
| 133 | /* Get options length */ |
| 134 | if (kind == TCP_OPTION_EOL) |
| 135 | break; |
| 136 | else if (kind == TCP_OPTION_NOOP) |
| 137 | opt_len = 1; |
| 138 | else |
| 139 | { |
| 140 | /* broken options */ |
| 141 | if (opts_len < 2) |
| 142 | break; |
| 143 | opt_len = data[1]; |
| 144 | |
| 145 | /* weird option length */ |
| 146 | if (opt_len < 2 || opt_len > opts_len) |
| 147 | break; |
| 148 | } |
| 149 | |
| 150 | /* Parse options */ |
| 151 | switch (kind) |
| 152 | { |
| 153 | case TCP_OPTION_MSS: |
| 154 | if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th)) |
| 155 | { |
| 156 | to->flags |= TCP_OPTS_FLAG_MSS; |
| 157 | to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2)); |
| 158 | } |
| 159 | break; |
| 160 | case TCP_OPTION_WINDOW_SCALE: |
| 161 | if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th)) |
| 162 | { |
| 163 | to->flags |= TCP_OPTS_FLAG_WSCALE; |
| 164 | to->wscale = data[2]; |
| 165 | if (to->wscale > TCP_MAX_WND_SCALE) |
| 166 | { |
| 167 | clib_warning ("Illegal window scaling value: %d", |
| 168 | to->wscale); |
| 169 | to->wscale = TCP_MAX_WND_SCALE; |
| 170 | } |
| 171 | } |
| 172 | break; |
| 173 | case TCP_OPTION_TIMESTAMP: |
| 174 | if (opt_len == TCP_OPTION_LEN_TIMESTAMP) |
| 175 | { |
| 176 | to->flags |= TCP_OPTS_FLAG_TSTAMP; |
| 177 | to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2)); |
| 178 | to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6)); |
| 179 | } |
| 180 | break; |
| 181 | case TCP_OPTION_SACK_PERMITTED: |
| 182 | if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th)) |
| 183 | to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; |
| 184 | break; |
| 185 | case TCP_OPTION_SACK_BLOCK: |
| 186 | /* If SACK permitted was not advertised or a SYN, break */ |
| 187 | if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th)) |
| 188 | break; |
| 189 | |
| 190 | /* If too short or not correctly formatted, break */ |
| 191 | if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK)) |
| 192 | break; |
| 193 | |
| 194 | to->flags |= TCP_OPTS_FLAG_SACK; |
| 195 | to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK; |
| 196 | vec_reset_length (to->sacks); |
| 197 | for (j = 0; j < to->n_sack_blocks; j++) |
| 198 | { |
| 199 | b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 4 * j)); |
| 200 | b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 4 * j)); |
| 201 | vec_add1 (to->sacks, b); |
| 202 | } |
| 203 | break; |
| 204 | default: |
| 205 | /* Nothing to see here */ |
| 206 | continue; |
| 207 | } |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | always_inline int |
| 212 | tcp_segment_check_paws (tcp_connection_t * tc) |
| 213 | { |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 214 | return tcp_opts_tstamp (&tc->opt) && tc->tsval_recent |
| 215 | && timestamp_lt (tc->opt.tsval, tc->tsval_recent); |
| 216 | } |
| 217 | |
| 218 | /** |
| 219 | * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19 |
| 220 | * |
| 221 | * It first verifies if segment has a wrapped sequence number (PAWS) and then |
| 222 | * does the processing associated to the first four steps (ignoring security |
| 223 | * and precedence): sequence number, rst bit and syn bit checks. |
| 224 | * |
| 225 | * @return 0 if segments passes validation. |
| 226 | */ |
| 227 | static int |
| 228 | tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, |
| 229 | vlib_buffer_t * b0, tcp_header_t * th0, u32 * next0) |
| 230 | { |
| 231 | u8 paws_failed; |
| 232 | |
| 233 | if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0))) |
| 234 | return -1; |
| 235 | |
| 236 | tcp_options_parse (th0, &tc0->opt); |
| 237 | |
| 238 | /* RFC1323: Check against wrapped sequence numbers (PAWS). If we have |
| 239 | * timestamp to echo and it's less than tsval_recent, drop segment |
| 240 | * but still send an ACK in order to retain TCP's mechanism for detecting |
| 241 | * and recovering from half-open connections */ |
| 242 | paws_failed = tcp_segment_check_paws (tc0); |
| 243 | if (paws_failed) |
| 244 | { |
| 245 | clib_warning ("paws failed"); |
| 246 | |
| 247 | /* If it just so happens that a segment updates tsval_recent for a |
| 248 | * segment over 24 days old, invalidate tsval_recent. */ |
| 249 | if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE, |
| 250 | tcp_time_now ())) |
| 251 | { |
| 252 | /* Age isn't reset until we get a valid tsval (bsd inspired) */ |
| 253 | tc0->tsval_recent = 0; |
| 254 | } |
| 255 | else |
| 256 | { |
| 257 | /* Drop after ack if not rst */ |
| 258 | if (!tcp_rst (th0)) |
| 259 | { |
| 260 | tcp_make_ack (tc0, b0); |
| 261 | *next0 = tcp_next_output (tc0->c_is_ip4); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 262 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 263 | return -1; |
| 264 | } |
| 265 | } |
| 266 | } |
| 267 | |
| 268 | /* 1st: check sequence number */ |
| 269 | if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number, |
| 270 | vnet_buffer (b0)->tcp.seq_end)) |
| 271 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 272 | /* If our window is 0 and the packet is in sequence, let it pass |
| 273 | * through for ack processing. It should be dropped later.*/ |
| 274 | if (tc0->rcv_wnd == 0 |
| 275 | && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 276 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 277 | /* TODO Should segment be tagged? */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 278 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 279 | else |
| 280 | { |
| 281 | /* If not RST, send dup ack */ |
| 282 | if (!tcp_rst (th0)) |
| 283 | { |
| 284 | tcp_make_ack (tc0, b0); |
| 285 | *next0 = tcp_next_output (tc0->c_is_ip4); |
| 286 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); |
| 287 | } |
| 288 | return -1; |
| 289 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 290 | } |
| 291 | |
| 292 | /* 2nd: check the RST bit */ |
| 293 | if (tcp_rst (th0)) |
| 294 | { |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 295 | tcp_connection_reset (tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 296 | return -1; |
| 297 | } |
| 298 | |
| 299 | /* 3rd: check security and precedence (skip) */ |
| 300 | |
| 301 | /* 4th: check the SYN bit */ |
| 302 | if (tcp_syn (th0)) |
| 303 | { |
| 304 | tcp_send_reset (b0, tc0->c_is_ip4); |
| 305 | return -1; |
| 306 | } |
| 307 | |
| 308 | /* If PAWS passed and segment in window, save timestamp */ |
| 309 | if (!paws_failed) |
| 310 | { |
| 311 | tc0->tsval_recent = tc0->opt.tsval; |
| 312 | tc0->tsval_recent_age = tcp_time_now (); |
| 313 | } |
| 314 | |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | always_inline int |
| 319 | tcp_rcv_ack_is_acceptable (tcp_connection_t * tc0, vlib_buffer_t * tb0) |
| 320 | { |
| 321 | /* SND.UNA =< SEG.ACK =< SND.NXT */ |
| 322 | return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number) |
| 323 | && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt)); |
| 324 | } |
| 325 | |
| 326 | /** |
| 327 | * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298 |
| 328 | * |
| 329 | * Note that although the original article, srtt and rttvar are scaled |
| 330 | * to minimize round-off errors, here we don't. Instead, we rely on |
| 331 | * better precision time measurements. |
| 332 | * |
| 333 | * TODO support us rtt resolution |
| 334 | */ |
| 335 | static void |
| 336 | tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) |
| 337 | { |
| 338 | int err; |
| 339 | |
| 340 | if (tc->srtt != 0) |
| 341 | { |
| 342 | err = mrtt - tc->srtt; |
| 343 | tc->srtt += err >> 3; |
| 344 | |
| 345 | /* XXX Drop in RTT results in RTTVAR increase and bigger RTO. |
| 346 | * The increase should be bound */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 347 | tc->rttvar += ((int) clib_abs (err) - (int) tc->rttvar) >> 2; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 348 | } |
| 349 | else |
| 350 | { |
| 351 | /* First measurement. */ |
| 352 | tc->srtt = mrtt; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 353 | tc->rttvar = mrtt >> 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 354 | } |
| 355 | } |
| 356 | |
| 357 | /** Update RTT estimate and RTO timer |
| 358 | * |
| 359 | * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK |
| 360 | * timing. Middle boxes are known to fiddle with TCP options so we |
| 361 | * should give higher priority to ACK timing. |
| 362 | * |
| 363 | * return 1 if valid rtt 0 otherwise |
| 364 | */ |
| 365 | static int |
| 366 | tcp_update_rtt (tcp_connection_t * tc, u32 ack) |
| 367 | { |
| 368 | u32 mrtt = 0; |
| 369 | |
| 370 | /* Karn's rule, part 1. Don't use retransmitted segments to estimate |
| 371 | * RTT because they're ambiguous. */ |
| 372 | if (tc->rtt_seq && seq_gt (ack, tc->rtt_seq) && !tc->rto_boff) |
| 373 | { |
| 374 | mrtt = tcp_time_now () - tc->rtt_ts; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 375 | } |
| 376 | |
| 377 | /* As per RFC7323 TSecr can be used for RTTM only if the segment advances |
| 378 | * snd_una, i.e., the left side of the send window: |
| 379 | * seq_lt (tc->snd_una, ack). Note: last condition could be dropped, we don't |
| 380 | * try to update rtt for dupacks */ |
| 381 | else if (tcp_opts_tstamp (&tc->opt) && tc->opt.tsecr && tc->bytes_acked) |
| 382 | { |
| 383 | mrtt = tcp_time_now () - tc->opt.tsecr; |
| 384 | } |
| 385 | |
| 386 | /* Ignore dubious measurements */ |
| 387 | if (mrtt == 0 || mrtt > TCP_RTT_MAX) |
| 388 | return 0; |
| 389 | |
| 390 | tcp_estimate_rtt (tc, mrtt); |
| 391 | |
| 392 | tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); |
| 393 | |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 394 | /* Allow measuring of RTT and make sure boff is 0 */ |
| 395 | tc->rtt_seq = 0; |
| 396 | tc->rto_boff = 0; |
| 397 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 398 | return 1; |
| 399 | } |
| 400 | |
| 401 | /** |
| 402 | * Dequeue bytes that have been acked and while at it update RTT estimates. |
| 403 | */ |
| 404 | static void |
| 405 | tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) |
| 406 | { |
| 407 | /* Dequeue the newly ACKed bytes */ |
| 408 | stream_session_dequeue_drop (&tc->connection, tc->bytes_acked); |
| 409 | |
| 410 | /* Update rtt and rto */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 411 | tcp_update_rtt (tc, ack); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 412 | } |
| 413 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 414 | /** |
| 415 | * Check if dupack as per RFC5681 Sec. 2 |
| 416 | * |
| 417 | * This works only if called before updating snd_wnd. |
| 418 | * */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 419 | always_inline u8 |
| 420 | tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 new_snd_wnd) |
| 421 | { |
| 422 | return ((vnet_buffer (b)->tcp.ack_number == tc->snd_una) |
| 423 | && seq_gt (tc->snd_una_max, tc->snd_una) |
| 424 | && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number) |
| 425 | && (new_snd_wnd == tc->snd_wnd)); |
| 426 | } |
| 427 | |
| 428 | void |
| 429 | scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) |
| 430 | { |
| 431 | sack_scoreboard_hole_t *next, *prev; |
| 432 | |
| 433 | if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) |
| 434 | { |
| 435 | next = pool_elt_at_index (sb->holes, hole->next); |
| 436 | next->prev = hole->prev; |
| 437 | } |
| 438 | |
| 439 | if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) |
| 440 | { |
| 441 | prev = pool_elt_at_index (sb->holes, hole->prev); |
| 442 | prev->next = hole->next; |
| 443 | } |
| 444 | else |
| 445 | { |
| 446 | sb->head = hole->next; |
| 447 | } |
| 448 | |
| 449 | pool_put (sb->holes, hole); |
| 450 | } |
| 451 | |
| 452 | sack_scoreboard_hole_t * |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 453 | scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 454 | u32 start, u32 end) |
| 455 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 456 | sack_scoreboard_hole_t *hole, *next, *prev; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 457 | u32 hole_index; |
| 458 | |
| 459 | pool_get (sb->holes, hole); |
| 460 | memset (hole, 0, sizeof (*hole)); |
| 461 | |
| 462 | hole->start = start; |
| 463 | hole->end = end; |
| 464 | hole_index = hole - sb->holes; |
| 465 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 466 | prev = scoreboard_get_hole (sb, prev_index); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 467 | if (prev) |
| 468 | { |
| 469 | hole->prev = prev - sb->holes; |
| 470 | hole->next = prev->next; |
| 471 | |
| 472 | if ((next = scoreboard_next_hole (sb, hole))) |
| 473 | next->prev = hole_index; |
| 474 | |
| 475 | prev->next = hole_index; |
| 476 | } |
| 477 | else |
| 478 | { |
| 479 | sb->head = hole_index; |
| 480 | hole->prev = TCP_INVALID_SACK_HOLE_INDEX; |
| 481 | hole->next = TCP_INVALID_SACK_HOLE_INDEX; |
| 482 | } |
| 483 | |
| 484 | return hole; |
| 485 | } |
| 486 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 487 | void |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 488 | tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) |
| 489 | { |
| 490 | sack_scoreboard_t *sb = &tc->sack_sb; |
| 491 | sack_block_t *blk, tmp; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 492 | sack_scoreboard_hole_t *hole, *next_hole, *last_hole, *new_hole; |
| 493 | u32 blk_index = 0, old_sacked_bytes, hole_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 494 | int i, j; |
| 495 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 496 | sb->last_sacked_bytes = 0; |
| 497 | sb->snd_una_adv = 0; |
| 498 | old_sacked_bytes = sb->sacked_bytes; |
| 499 | |
| 500 | if (!tcp_opts_sack (&tc->opt) && sb->head == TCP_INVALID_SACK_HOLE_INDEX) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 501 | return; |
| 502 | |
| 503 | /* Remove invalid blocks */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 504 | blk = tc->opt.sacks; |
| 505 | while (blk < vec_end (tc->opt.sacks)) |
| 506 | { |
| 507 | if (seq_lt (blk->start, blk->end) |
| 508 | && seq_gt (blk->start, tc->snd_una) |
| 509 | && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_nxt)) |
| 510 | { |
| 511 | blk++; |
| 512 | continue; |
| 513 | } |
| 514 | vec_del1 (tc->opt.sacks, blk - tc->opt.sacks); |
| 515 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 516 | |
| 517 | /* Add block for cumulative ack */ |
| 518 | if (seq_gt (ack, tc->snd_una)) |
| 519 | { |
| 520 | tmp.start = tc->snd_una; |
| 521 | tmp.end = ack; |
| 522 | vec_add1 (tc->opt.sacks, tmp); |
| 523 | } |
| 524 | |
| 525 | if (vec_len (tc->opt.sacks) == 0) |
| 526 | return; |
| 527 | |
| 528 | /* Make sure blocks are ordered */ |
| 529 | for (i = 0; i < vec_len (tc->opt.sacks); i++) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 530 | for (j = i + 1; j < vec_len (tc->opt.sacks); j++) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 531 | if (seq_lt (tc->opt.sacks[j].start, tc->opt.sacks[i].start)) |
| 532 | { |
| 533 | tmp = tc->opt.sacks[i]; |
| 534 | tc->opt.sacks[i] = tc->opt.sacks[j]; |
| 535 | tc->opt.sacks[j] = tmp; |
| 536 | } |
| 537 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 538 | if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) |
| 539 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 540 | /* If no holes, insert the first that covers all outstanding bytes */ |
| 541 | last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, |
| 542 | tc->snd_una, tc->snd_una_max); |
| 543 | sb->tail = scoreboard_hole_index (sb, last_hole); |
| 544 | } |
| 545 | else |
| 546 | { |
| 547 | /* If we have holes but snd_una_max is beyond the last hole, update |
| 548 | * last hole end */ |
| 549 | tmp = tc->opt.sacks[vec_len (tc->opt.sacks) - 1]; |
| 550 | last_hole = scoreboard_last_hole (sb); |
| 551 | if (seq_gt (tc->snd_una_max, sb->max_byte_sacked) |
| 552 | && seq_gt (tc->snd_una_max, last_hole->end)) |
| 553 | last_hole->end = tc->snd_una_max; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 554 | } |
| 555 | |
| 556 | /* Walk the holes with the SACK blocks */ |
| 557 | hole = pool_elt_at_index (sb->holes, sb->head); |
| 558 | while (hole && blk_index < vec_len (tc->opt.sacks)) |
| 559 | { |
| 560 | blk = &tc->opt.sacks[blk_index]; |
| 561 | |
| 562 | if (seq_leq (blk->start, hole->start)) |
| 563 | { |
| 564 | /* Block covers hole. Remove hole */ |
| 565 | if (seq_geq (blk->end, hole->end)) |
| 566 | { |
| 567 | next_hole = scoreboard_next_hole (sb, hole); |
| 568 | |
| 569 | /* Byte accounting */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 570 | if (seq_leq (hole->end, ack)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 571 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 572 | /* Bytes lost because snd_wnd left edge advances */ |
| 573 | if (next_hole && seq_leq (next_hole->start, ack)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 574 | sb->sacked_bytes -= next_hole->start - hole->end; |
| 575 | else |
| 576 | sb->sacked_bytes -= ack - hole->end; |
| 577 | } |
| 578 | else |
| 579 | { |
| 580 | sb->sacked_bytes += scoreboard_hole_bytes (hole); |
| 581 | } |
| 582 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 583 | /* snd_una needs to be advanced */ |
| 584 | if (seq_geq (ack, hole->end)) |
| 585 | { |
| 586 | if (next_hole && seq_lt (ack, next_hole->start)) |
| 587 | sb->snd_una_adv = next_hole->start - ack; |
| 588 | else |
| 589 | sb->snd_una_adv = sb->max_byte_sacked - ack; |
| 590 | |
| 591 | /* all these can be delivered */ |
| 592 | sb->sacked_bytes -= sb->snd_una_adv; |
| 593 | } |
| 594 | |
| 595 | /* About to remove last hole */ |
| 596 | if (hole == last_hole) |
| 597 | { |
| 598 | sb->tail = hole->prev; |
| 599 | last_hole = scoreboard_last_hole (sb); |
| 600 | /* keep track of max byte sacked in case the last hole |
| 601 | * is acked */ |
| 602 | if (seq_gt (hole->end, sb->max_byte_sacked)) |
| 603 | sb->max_byte_sacked = hole->end; |
| 604 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 605 | scoreboard_remove_hole (sb, hole); |
| 606 | hole = next_hole; |
| 607 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 608 | /* Partial 'head' overlap */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 609 | else |
| 610 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 611 | if (seq_gt (blk->end, hole->start)) |
| 612 | { |
| 613 | sb->sacked_bytes += blk->end - hole->start; |
| 614 | hole->start = blk->end; |
| 615 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 616 | blk_index++; |
| 617 | } |
| 618 | } |
| 619 | else |
| 620 | { |
| 621 | /* Hole must be split */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 622 | if (seq_lt (blk->end, hole->end)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 623 | { |
| 624 | sb->sacked_bytes += blk->end - blk->start; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 625 | hole_index = scoreboard_hole_index (sb, hole); |
| 626 | new_hole = scoreboard_insert_hole (sb, hole_index, blk->end, |
| 627 | hole->end); |
| 628 | |
| 629 | /* Pool might've moved */ |
| 630 | hole = scoreboard_get_hole (sb, hole_index); |
| 631 | hole->end = blk->start; |
| 632 | |
| 633 | /* New or split of tail */ |
| 634 | if ((last_hole->end == new_hole->end) |
| 635 | || seq_lt (last_hole->end, new_hole->start)) |
| 636 | { |
| 637 | last_hole = new_hole; |
| 638 | sb->tail = scoreboard_hole_index (sb, new_hole); |
| 639 | } |
| 640 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 641 | blk_index++; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 642 | hole = scoreboard_next_hole (sb, hole); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 643 | } |
| 644 | else |
| 645 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 646 | sb->sacked_bytes += hole->end - blk->start; |
| 647 | hole->end = blk->start; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 648 | hole = scoreboard_next_hole (sb, hole); |
| 649 | } |
| 650 | } |
| 651 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 652 | |
| 653 | sb->last_sacked_bytes = sb->sacked_bytes + sb->snd_una_adv |
| 654 | - old_sacked_bytes; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 655 | } |
| 656 | |
| 657 | /** Update snd_wnd |
| 658 | * |
| 659 | * If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set |
| 660 | * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ |
| 661 | static void |
| 662 | tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) |
| 663 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 664 | if (seq_lt (tc->snd_wl1, seq) |
| 665 | || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 666 | { |
| 667 | tc->snd_wnd = snd_wnd; |
| 668 | tc->snd_wl1 = seq; |
| 669 | tc->snd_wl2 = ack; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 670 | TCP_EVT_DBG (TCP_EVT_SND_WND, tc); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 671 | |
| 672 | /* Set probe timer if we just got 0 wnd */ |
| 673 | if (tc->snd_wnd < tc->snd_mss |
| 674 | && !tcp_timer_is_active (tc, TCP_TIMER_PERSIST)) |
| 675 | tcp_persist_timer_set (tc); |
| 676 | else |
| 677 | tcp_persist_timer_reset (tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 678 | } |
| 679 | } |
| 680 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 681 | void |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 682 | tcp_cc_congestion (tcp_connection_t * tc) |
| 683 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 684 | tc->snd_congestion = tc->snd_nxt; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 685 | tc->cc_algo->congestion (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 686 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 687 | } |
| 688 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 689 | void |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 690 | tcp_cc_recover (tcp_connection_t * tc) |
| 691 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 692 | /* TODO: check if time to recover was small. It might be that RTO popped |
| 693 | * too soon. |
| 694 | */ |
| 695 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 696 | tc->cc_algo->recovered (tc); |
| 697 | |
| 698 | tc->rtx_bytes = 0; |
| 699 | tc->rcv_dupacks = 0; |
| 700 | tc->snd_nxt = tc->snd_una; |
| 701 | |
| 702 | tc->cc_algo->rcv_ack (tc); |
| 703 | tc->tsecr_last_ack = tc->opt.tsecr; |
| 704 | |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 705 | tcp_cong_recovery_off (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 706 | |
| 707 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 708 | } |
| 709 | |
| 710 | static void |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 711 | tcp_cc_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 712 | { |
| 713 | u8 partial_ack; |
| 714 | |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 715 | if (tcp_in_cong_recovery (tc)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 716 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 717 | partial_ack = seq_lt (tc->snd_una, tc->snd_congestion); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 718 | if (!partial_ack) |
| 719 | { |
| 720 | /* Clear retransmitted bytes. */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 721 | tcp_cc_recover (tc); |
| 722 | } |
| 723 | else |
| 724 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 725 | TCP_EVT_DBG (TCP_EVT_CC_PACK, tc); |
| 726 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 727 | /* Clear retransmitted bytes. XXX should we clear all? */ |
| 728 | tc->rtx_bytes = 0; |
| 729 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); |
| 730 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 731 | /* In case snd_nxt is still in the past and output tries to |
| 732 | * shove some new bytes */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 733 | tc->snd_nxt = tc->snd_una_max; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 734 | |
| 735 | /* XXX need proper RFC6675 support */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 736 | if (tc->sack_sb.last_sacked_bytes && !tcp_in_recovery (tc)) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 737 | { |
| 738 | tcp_fast_retransmit (tc); |
| 739 | } |
| 740 | else |
| 741 | { |
| 742 | /* Retransmit first unacked segment */ |
| 743 | tcp_retransmit_first_unacked (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 744 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 745 | } |
| 746 | } |
| 747 | else |
| 748 | { |
| 749 | tc->cc_algo->rcv_ack (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 750 | tc->tsecr_last_ack = tc->opt.tsecr; |
| 751 | tc->rcv_dupacks = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 752 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 753 | } |
| 754 | |
| 755 | static void |
| 756 | tcp_cc_rcv_dupack (tcp_connection_t * tc, u32 ack) |
| 757 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 758 | // ASSERT (seq_geq(tc->snd_una, ack)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 759 | |
| 760 | tc->rcv_dupacks++; |
| 761 | if (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) |
| 762 | { |
| 763 | /* RFC6582 NewReno heuristic to avoid multiple fast retransmits */ |
| 764 | if (tc->opt.tsecr != tc->tsecr_last_ack) |
| 765 | { |
| 766 | tc->rcv_dupacks = 0; |
| 767 | return; |
| 768 | } |
| 769 | |
| 770 | tcp_fastrecovery_on (tc); |
| 771 | |
| 772 | /* Handle congestion and dupack */ |
| 773 | tcp_cc_congestion (tc); |
| 774 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); |
| 775 | |
| 776 | tcp_fast_retransmit (tc); |
| 777 | |
| 778 | /* Post retransmit update cwnd to ssthresh and account for the |
| 779 | * three segments that have left the network and should've been |
| 780 | * buffered at the receiver */ |
| 781 | tc->cwnd = tc->ssthresh + TCP_DUPACK_THRESHOLD * tc->snd_mss; |
| 782 | } |
| 783 | else if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD) |
| 784 | { |
| 785 | ASSERT (tcp_in_fastrecovery (tc)); |
| 786 | |
| 787 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | void |
| 792 | tcp_cc_init (tcp_connection_t * tc) |
| 793 | { |
| 794 | tc->cc_algo = tcp_cc_algo_get (TCP_CC_NEWRENO); |
| 795 | tc->cc_algo->init (tc); |
| 796 | } |
| 797 | |
| 798 | static int |
| 799 | tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, |
| 800 | tcp_header_t * th, u32 * next, u32 * error) |
| 801 | { |
| 802 | u32 new_snd_wnd; |
| 803 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 804 | /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 805 | if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt)) |
| 806 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 807 | /* If we have outstanding data and this is within the window, accept it, |
| 808 | * probably retransmit has timed out. Otherwise ACK segment and then |
| 809 | * drop it */ |
| 810 | if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max)) |
| 811 | { |
| 812 | tcp_make_ack (tc, b); |
| 813 | *next = tcp_next_output (tc->c_is_ip4); |
| 814 | *error = TCP_ERROR_ACK_INVALID; |
| 815 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0, |
| 816 | vnet_buffer (b)->tcp.ack_number); |
| 817 | return -1; |
| 818 | } |
| 819 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 820 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2, |
| 821 | vnet_buffer (b)->tcp.ack_number); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 822 | |
| 823 | tc->snd_nxt = vnet_buffer (b)->tcp.ack_number; |
| 824 | *error = TCP_ERROR_ACK_FUTURE; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 825 | } |
| 826 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 827 | /* If old ACK, probably it's an old dupack */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 828 | if (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una)) |
| 829 | { |
| 830 | *error = TCP_ERROR_ACK_OLD; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 831 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1, |
| 832 | vnet_buffer (b)->tcp.ack_number); |
| 833 | if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) |
| 834 | { |
| 835 | TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc); |
| 836 | tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); |
| 837 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 838 | return -1; |
| 839 | } |
| 840 | |
| 841 | if (tcp_opts_sack_permitted (&tc->opt)) |
| 842 | tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); |
| 843 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 844 | new_snd_wnd = clib_net_to_host_u16 (th->window) << tc->snd_wscale; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 845 | |
| 846 | if (tcp_ack_is_dupack (tc, b, new_snd_wnd)) |
| 847 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 848 | TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 849 | tcp_cc_rcv_dupack (tc, vnet_buffer (b)->tcp.ack_number); |
| 850 | *error = TCP_ERROR_ACK_DUP; |
| 851 | return -1; |
| 852 | } |
| 853 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 854 | /* |
| 855 | * Valid ACK |
| 856 | */ |
| 857 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 858 | tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 859 | tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 860 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 861 | /* Dequeue ACKed data and update RTT */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 862 | tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 863 | tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, |
| 864 | vnet_buffer (b)->tcp.ack_number, new_snd_wnd); |
| 865 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 866 | /* If some of our sent bytes have been acked, update cc and retransmit |
| 867 | * timer. */ |
| 868 | if (tc->bytes_acked) |
| 869 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 870 | TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 871 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 872 | /* Updates congestion control (slow start/congestion avoidance) */ |
| 873 | tcp_cc_rcv_ack (tc, b); |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 874 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 875 | /* If everything has been acked, stop retransmit timer |
| 876 | * otherwise update */ |
| 877 | if (tc->snd_una == tc->snd_una_max) |
| 878 | tcp_retransmit_timer_reset (tc); |
| 879 | else |
| 880 | tcp_retransmit_timer_update (tc); |
| 881 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 882 | |
| 883 | return 0; |
| 884 | } |
| 885 | |
| 886 | /** |
| 887 | * Build SACK list as per RFC2018. |
| 888 | * |
| 889 | * Makes sure the first block contains the segment that generated the current |
| 890 | * ACK and the following ones are the ones most recently reported in SACK |
| 891 | * blocks. |
| 892 | * |
| 893 | * @param tc TCP connection for which the SACK list is updated |
| 894 | * @param start Start sequence number of the newest SACK block |
| 895 | * @param end End sequence of the newest SACK block |
| 896 | */ |
| 897 | static void |
| 898 | tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) |
| 899 | { |
| 900 | sack_block_t *new_list = 0, block; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 901 | int i; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 902 | |
| 903 | /* If the first segment is ooo add it to the list. Last write might've moved |
| 904 | * rcv_nxt over the first segment. */ |
| 905 | if (seq_lt (tc->rcv_nxt, start)) |
| 906 | { |
| 907 | block.start = start; |
| 908 | block.end = end; |
| 909 | vec_add1 (new_list, block); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 910 | } |
| 911 | |
| 912 | /* Find the blocks still worth keeping. */ |
| 913 | for (i = 0; i < vec_len (tc->snd_sacks); i++) |
| 914 | { |
| 915 | /* Discard if: |
| 916 | * 1) rcv_nxt advanced beyond current block OR |
| 917 | * 2) Segment overlapped by the first segment, i.e., it has been merged |
| 918 | * into it.*/ |
| 919 | if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt) |
| 920 | || seq_leq (tc->snd_sacks[i].start, end)) |
| 921 | continue; |
| 922 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 923 | /* Save to new SACK list. */ |
| 924 | vec_add1 (new_list, tc->snd_sacks[i]); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 925 | } |
| 926 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 927 | ASSERT (vec_len (new_list) < TCP_MAX_SACK_BLOCKS); |
| 928 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 929 | /* Replace old vector with new one */ |
| 930 | vec_free (tc->snd_sacks); |
| 931 | tc->snd_sacks = new_list; |
| 932 | } |
| 933 | |
| 934 | /** Enqueue data for delivery to application */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 935 | always_inline int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 936 | tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, |
| 937 | u16 data_len) |
| 938 | { |
| 939 | int written; |
| 940 | |
| 941 | /* Pure ACK. Update rcv_nxt and be done. */ |
| 942 | if (PREDICT_FALSE (data_len == 0)) |
| 943 | { |
| 944 | tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; |
| 945 | return TCP_ERROR_PURE_ACK; |
| 946 | } |
| 947 | |
| 948 | written = stream_session_enqueue_data (&tc->connection, |
| 949 | vlib_buffer_get_current (b), |
| 950 | data_len, 1 /* queue event */ ); |
| 951 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 952 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written); |
| 953 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 954 | /* Update rcv_nxt */ |
| 955 | if (PREDICT_TRUE (written == data_len)) |
| 956 | { |
| 957 | tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end; |
| 958 | } |
| 959 | /* If more data written than expected, account for out-of-order bytes. */ |
| 960 | else if (written > data_len) |
| 961 | { |
| 962 | tc->rcv_nxt = vnet_buffer (b)->tcp.seq_end + written - data_len; |
| 963 | |
| 964 | /* Send ACK confirming the update */ |
| 965 | tc->flags |= TCP_CONN_SNDACK; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 966 | } |
| 967 | else if (written > 0) |
| 968 | { |
| 969 | /* We've written something but FIFO is probably full now */ |
| 970 | tc->rcv_nxt += written; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 971 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 972 | /* Depending on how fast the app is, all remaining buffers in burst will |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 973 | * not be enqueued. Inform peer */ |
| 974 | tc->flags |= TCP_CONN_SNDACK; |
| 975 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 976 | return TCP_ERROR_PARTIALLY_ENQUEUED; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 977 | } |
| 978 | else |
| 979 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 980 | tc->flags |= TCP_CONN_SNDACK; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 981 | return TCP_ERROR_FIFO_FULL; |
| 982 | } |
| 983 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 984 | /* Update SACK list if need be */ |
| 985 | if (tcp_opts_sack_permitted (&tc->opt)) |
| 986 | { |
| 987 | /* Remove SACK blocks that have been delivered */ |
| 988 | tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); |
| 989 | } |
| 990 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 991 | return TCP_ERROR_ENQUEUED; |
| 992 | } |
| 993 | |
| 994 | /** Enqueue out-of-order data */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 995 | always_inline int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 996 | tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, |
| 997 | u16 data_len) |
| 998 | { |
| 999 | stream_session_t *s0; |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1000 | u32 offset; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1001 | int rv; |
| 1002 | |
| 1003 | /* Pure ACK. Do nothing */ |
| 1004 | if (PREDICT_FALSE (data_len == 0)) |
| 1005 | { |
| 1006 | return TCP_ERROR_PURE_ACK; |
| 1007 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1008 | |
| 1009 | s0 = stream_session_get (tc->c_s_index, tc->c_thread_index); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1010 | offset = vnet_buffer (b)->tcp.seq_number - tc->irs; |
| 1011 | |
| 1012 | clib_warning ("ooo: offset %d len %d", offset, data_len); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1013 | |
Florin Coras | a546481 | 2017-04-19 13:00:05 -0700 | [diff] [blame^] | 1014 | rv = svm_fifo_enqueue_with_offset (s0->server_rx_fifo, offset, data_len, |
| 1015 | vlib_buffer_get_current (b)); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1016 | |
| 1017 | /* Nothing written */ |
| 1018 | if (rv) |
| 1019 | { |
| 1020 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0); |
| 1021 | return TCP_ERROR_FIFO_FULL; |
| 1022 | } |
| 1023 | |
| 1024 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1025 | |
| 1026 | /* Update SACK list if in use */ |
| 1027 | if (tcp_opts_sack_permitted (&tc->opt)) |
| 1028 | { |
| 1029 | ooo_segment_t *newest; |
| 1030 | u32 start, end; |
| 1031 | |
| 1032 | /* Get the newest segment from the fifo */ |
| 1033 | newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1034 | start = ooo_segment_offset (s0->server_rx_fifo, newest); |
| 1035 | end = ooo_segment_end_offset (s0->server_rx_fifo, newest); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1036 | |
| 1037 | tcp_update_sack_list (tc, start, end); |
| 1038 | } |
| 1039 | |
| 1040 | return TCP_ERROR_ENQUEUED; |
| 1041 | } |
| 1042 | |
| 1043 | /** |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1044 | * Check if ACK could be delayed. If ack can be delayed, it should return |
| 1045 | * true for a full frame. If we're always acking return 0. |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1046 | */ |
| 1047 | always_inline int |
| 1048 | tcp_can_delack (tcp_connection_t * tc) |
| 1049 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1050 | /* Send ack if ... */ |
| 1051 | if (TCP_ALWAYS_ACK |
| 1052 | /* just sent a rcv wnd 0 */ |
| 1053 | || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0 |
| 1054 | /* constrained to send ack */ |
| 1055 | || (tc->flags & TCP_CONN_SNDACK) != 0 |
| 1056 | /* we're almost out of tx wnd */ |
| 1057 | || tcp_available_snd_space (tc) < 2 * tc->snd_mss) |
| 1058 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1059 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1060 | return 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1061 | } |
| 1062 | |
| 1063 | static int |
| 1064 | tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, |
| 1065 | u16 n_data_bytes, u32 * next0) |
| 1066 | { |
| 1067 | u32 error = 0; |
| 1068 | |
| 1069 | /* Handle out-of-order data */ |
| 1070 | if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt)) |
| 1071 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1072 | /* Old sequence numbers allowed through because they overlapped |
| 1073 | * the rx window */ |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1074 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1075 | if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1076 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1077 | error = TCP_ERROR_SEGMENT_OLD; |
| 1078 | *next0 = TCP_NEXT_DROP; |
| 1079 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1080 | } |
| 1081 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1082 | error = tcp_session_enqueue_ooo (tc, b, n_data_bytes); |
| 1083 | |
| 1084 | /* N.B. Should not filter burst of dupacks. Two issues 1) dupacks open |
| 1085 | * cwnd on remote peer when congested 2) acks leaving should have the |
| 1086 | * latest rcv_wnd since the burst may eaten up all of it, so only the |
| 1087 | * old ones could be filtered. |
| 1088 | */ |
| 1089 | |
| 1090 | /* RFC2581: Send DUPACK for fast retransmit */ |
| 1091 | tcp_make_ack (tc, b); |
| 1092 | *next0 = tcp_next_output (tc->c_is_ip4); |
| 1093 | |
| 1094 | /* Mark as DUPACK. We may filter these in output if |
| 1095 | * the burst fills the holes. */ |
| 1096 | if (n_data_bytes) |
| 1097 | vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; |
| 1098 | |
| 1099 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1100 | goto done; |
| 1101 | } |
| 1102 | |
| 1103 | /* In order data, enqueue. Fifo figures out by itself if any out-of-order |
| 1104 | * segments can be enqueued after fifo tail offset changes. */ |
| 1105 | error = tcp_session_enqueue_data (tc, b, n_data_bytes); |
| 1106 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1107 | if (n_data_bytes == 0) |
| 1108 | { |
| 1109 | *next0 = TCP_NEXT_DROP; |
| 1110 | goto done; |
| 1111 | } |
| 1112 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1113 | /* Check if ACK can be delayed */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1114 | if (tcp_can_delack (tc)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1115 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1116 | if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK)) |
| 1117 | tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1118 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1119 | } |
| 1120 | |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1121 | *next0 = tcp_next_output (tc->c_is_ip4); |
| 1122 | tcp_make_ack (tc, b); |
| 1123 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1124 | done: |
| 1125 | return error; |
| 1126 | } |
| 1127 | |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1128 | typedef struct |
| 1129 | { |
| 1130 | tcp_header_t tcp_header; |
| 1131 | tcp_connection_t tcp_connection; |
| 1132 | } tcp_rx_trace_t; |
| 1133 | |
| 1134 | u8 * |
| 1135 | format_tcp_rx_trace (u8 * s, va_list * args) |
| 1136 | { |
| 1137 | CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); |
| 1138 | CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); |
| 1139 | tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); |
| 1140 | uword indent = format_get_indent (s); |
| 1141 | |
| 1142 | s = format (s, "%U\n%U%U", |
| 1143 | format_tcp_header, &t->tcp_header, 128, |
| 1144 | format_white_space, indent, |
| 1145 | format_tcp_connection_verbose, &t->tcp_connection); |
| 1146 | |
| 1147 | return s; |
| 1148 | } |
| 1149 | |
| 1150 | u8 * |
| 1151 | format_tcp_rx_trace_short (u8 * s, va_list * args) |
| 1152 | { |
| 1153 | CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); |
| 1154 | CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); |
| 1155 | tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); |
| 1156 | |
| 1157 | s = format (s, "%d -> %d (%U)", |
| 1158 | clib_net_to_host_u16 (t->tcp_header.src_port), |
| 1159 | clib_net_to_host_u16 (t->tcp_header.dst_port), format_tcp_state, |
| 1160 | &t->tcp_connection.state); |
| 1161 | |
| 1162 | return s; |
| 1163 | } |
| 1164 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1165 | always_inline void |
| 1166 | tcp_established_inc_counter (vlib_main_t * vm, u8 is_ip4, u8 evt, u8 val) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1167 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1168 | if (PREDICT_TRUE (!val)) |
| 1169 | return; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1170 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1171 | if (is_ip4) |
| 1172 | vlib_node_increment_counter (vm, tcp4_established_node.index, evt, val); |
| 1173 | else |
| 1174 | vlib_node_increment_counter (vm, tcp6_established_node.index, evt, val); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1175 | } |
| 1176 | |
| 1177 | always_inline uword |
| 1178 | tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1179 | vlib_frame_t * from_frame, int is_ip4) |
| 1180 | { |
| 1181 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 1182 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1183 | tcp_main_t *tm = vnet_get_tcp_main (); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1184 | u8 is_fin = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1185 | |
| 1186 | from = vlib_frame_vector_args (from_frame); |
| 1187 | n_left_from = from_frame->n_vectors; |
| 1188 | |
| 1189 | next_index = node->cached_next_index; |
| 1190 | |
| 1191 | while (n_left_from > 0) |
| 1192 | { |
| 1193 | u32 n_left_to_next; |
| 1194 | |
| 1195 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 1196 | |
| 1197 | while (n_left_from > 0 && n_left_to_next > 0) |
| 1198 | { |
| 1199 | u32 bi0; |
| 1200 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1201 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1202 | tcp_header_t *th0 = 0; |
| 1203 | tcp_connection_t *tc0; |
| 1204 | ip4_header_t *ip40; |
| 1205 | ip6_header_t *ip60; |
| 1206 | u32 n_advance_bytes0, n_data_bytes0; |
| 1207 | u32 next0 = TCP_ESTABLISHED_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; |
| 1208 | |
| 1209 | bi0 = from[0]; |
| 1210 | to_next[0] = bi0; |
| 1211 | from += 1; |
| 1212 | to_next += 1; |
| 1213 | n_left_from -= 1; |
| 1214 | n_left_to_next -= 1; |
| 1215 | |
| 1216 | b0 = vlib_get_buffer (vm, bi0); |
| 1217 | tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, |
| 1218 | my_thread_index); |
| 1219 | |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1220 | if (PREDICT_FALSE (tc0 == 0)) |
| 1221 | { |
| 1222 | error0 = TCP_ERROR_INVALID_CONNECTION; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1223 | goto done; |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1224 | } |
| 1225 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1226 | /* Checksum computed by ipx_local no need to compute again */ |
| 1227 | |
| 1228 | if (is_ip4) |
| 1229 | { |
| 1230 | ip40 = vlib_buffer_get_current (b0); |
| 1231 | th0 = ip4_next_header (ip40); |
| 1232 | n_advance_bytes0 = (ip4_header_bytes (ip40) |
| 1233 | + tcp_header_bytes (th0)); |
| 1234 | n_data_bytes0 = clib_net_to_host_u16 (ip40->length) |
| 1235 | - n_advance_bytes0; |
| 1236 | } |
| 1237 | else |
| 1238 | { |
| 1239 | ip60 = vlib_buffer_get_current (b0); |
| 1240 | th0 = ip6_next_header (ip60); |
| 1241 | n_advance_bytes0 = tcp_header_bytes (th0); |
| 1242 | n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) |
| 1243 | - n_advance_bytes0; |
| 1244 | n_advance_bytes0 += sizeof (ip60[0]); |
| 1245 | } |
| 1246 | |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1247 | is_fin = (th0->flags & TCP_FLAG_FIN) != 0; |
| 1248 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1249 | /* SYNs, FINs and data consume sequence numbers */ |
| 1250 | vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1251 | + tcp_is_syn (th0) + is_fin + n_data_bytes0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1252 | |
| 1253 | /* TODO header prediction fast path */ |
| 1254 | |
| 1255 | /* 1-4: check SEQ, RST, SYN */ |
| 1256 | if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0))) |
| 1257 | { |
| 1258 | error0 = TCP_ERROR_SEGMENT_INVALID; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1259 | TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, |
| 1260 | vnet_buffer (b0)->tcp.seq_number, |
| 1261 | vnet_buffer (b0)->tcp.seq_end); |
| 1262 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1263 | } |
| 1264 | |
| 1265 | /* 5: check the ACK field */ |
| 1266 | if (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)) |
| 1267 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1268 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1269 | } |
| 1270 | |
| 1271 | /* 6: check the URG bit TODO */ |
| 1272 | |
| 1273 | /* 7: process the segment text */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1274 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1275 | vlib_buffer_advance (b0, n_advance_bytes0); |
| 1276 | error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); |
| 1277 | |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1278 | /* N.B. buffer is rewritten if segment is ooo. Thus, th0 becomes a |
| 1279 | * dangling reference. */ |
| 1280 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1281 | /* 8: check the FIN bit */ |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1282 | if (is_fin) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1283 | { |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1284 | /* Enter CLOSE-WAIT and notify session. Don't send ACK, instead |
| 1285 | * wait for session to call close. To avoid lingering |
| 1286 | * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1287 | tc0->state = TCP_STATE_CLOSE_WAIT; |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1288 | TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1289 | stream_session_disconnect_notify (&tc0->connection); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1290 | tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1291 | } |
| 1292 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1293 | done: |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1294 | b0->error = node->errors[error0]; |
| 1295 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 1296 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1297 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 1298 | clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header)); |
| 1299 | clib_memcpy (&t0->tcp_connection, tc0, |
| 1300 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1301 | } |
| 1302 | |
| 1303 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 1304 | n_left_to_next, bi0, next0); |
| 1305 | } |
| 1306 | |
| 1307 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 1308 | } |
| 1309 | |
| 1310 | errors = session_manager_flush_enqueue_events (my_thread_index); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1311 | tcp_established_inc_counter (vm, is_ip4, TCP_ERROR_EVENT_FIFO_FULL, errors); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1312 | |
| 1313 | return from_frame->n_vectors; |
| 1314 | } |
| 1315 | |
| 1316 | static uword |
| 1317 | tcp4_established (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1318 | vlib_frame_t * from_frame) |
| 1319 | { |
| 1320 | return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 1321 | } |
| 1322 | |
| 1323 | static uword |
| 1324 | tcp6_established (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1325 | vlib_frame_t * from_frame) |
| 1326 | { |
| 1327 | return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 1328 | } |
| 1329 | |
| 1330 | /* *INDENT-OFF* */ |
| 1331 | VLIB_REGISTER_NODE (tcp4_established_node) = |
| 1332 | { |
| 1333 | .function = tcp4_established, |
| 1334 | .name = "tcp4-established", |
| 1335 | /* Takes a vector of packets. */ |
| 1336 | .vector_size = sizeof (u32), |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1337 | .n_errors = TCP_N_ERROR, |
| 1338 | .error_strings = tcp_error_strings, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1339 | .n_next_nodes = TCP_ESTABLISHED_N_NEXT, |
| 1340 | .next_nodes = |
| 1341 | { |
| 1342 | #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, |
| 1343 | foreach_tcp_state_next |
| 1344 | #undef _ |
| 1345 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1346 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1347 | }; |
| 1348 | /* *INDENT-ON* */ |
| 1349 | |
| 1350 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_established_node, tcp4_established); |
| 1351 | |
| 1352 | /* *INDENT-OFF* */ |
| 1353 | VLIB_REGISTER_NODE (tcp6_established_node) = |
| 1354 | { |
| 1355 | .function = tcp6_established, |
| 1356 | .name = "tcp6-established", |
| 1357 | /* Takes a vector of packets. */ |
| 1358 | .vector_size = sizeof (u32), |
| 1359 | .n_errors = TCP_N_ERROR, |
| 1360 | .error_strings = tcp_error_strings, |
| 1361 | .n_next_nodes = TCP_ESTABLISHED_N_NEXT, |
| 1362 | .next_nodes = |
| 1363 | { |
| 1364 | #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, |
| 1365 | foreach_tcp_state_next |
| 1366 | #undef _ |
| 1367 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1368 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1369 | }; |
| 1370 | /* *INDENT-ON* */ |
| 1371 | |
| 1372 | |
| 1373 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_established_node, tcp6_established); |
| 1374 | |
| 1375 | vlib_node_registration_t tcp4_syn_sent_node; |
| 1376 | vlib_node_registration_t tcp6_syn_sent_node; |
| 1377 | |
| 1378 | always_inline uword |
| 1379 | tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1380 | vlib_frame_t * from_frame, int is_ip4) |
| 1381 | { |
| 1382 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 1383 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 1384 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1385 | u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; |
| 1386 | |
| 1387 | from = vlib_frame_vector_args (from_frame); |
| 1388 | n_left_from = from_frame->n_vectors; |
| 1389 | |
| 1390 | next_index = node->cached_next_index; |
| 1391 | |
| 1392 | while (n_left_from > 0) |
| 1393 | { |
| 1394 | u32 n_left_to_next; |
| 1395 | |
| 1396 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 1397 | |
| 1398 | while (n_left_from > 0 && n_left_to_next > 0) |
| 1399 | { |
| 1400 | u32 bi0, ack0, seq0; |
| 1401 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1402 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1403 | tcp_header_t *tcp0 = 0; |
| 1404 | tcp_connection_t *tc0; |
| 1405 | ip4_header_t *ip40; |
| 1406 | ip6_header_t *ip60; |
| 1407 | u32 n_advance_bytes0, n_data_bytes0; |
| 1408 | tcp_connection_t *new_tc0; |
| 1409 | u32 next0 = TCP_SYN_SENT_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; |
| 1410 | |
| 1411 | bi0 = from[0]; |
| 1412 | to_next[0] = bi0; |
| 1413 | from += 1; |
| 1414 | to_next += 1; |
| 1415 | n_left_from -= 1; |
| 1416 | n_left_to_next -= 1; |
| 1417 | |
| 1418 | b0 = vlib_get_buffer (vm, bi0); |
| 1419 | tc0 = |
| 1420 | tcp_half_open_connection_get (vnet_buffer (b0)-> |
| 1421 | tcp.connection_index); |
| 1422 | |
| 1423 | ack0 = vnet_buffer (b0)->tcp.ack_number; |
| 1424 | seq0 = vnet_buffer (b0)->tcp.seq_number; |
| 1425 | |
| 1426 | /* Checksum computed by ipx_local no need to compute again */ |
| 1427 | |
| 1428 | if (is_ip4) |
| 1429 | { |
| 1430 | ip40 = vlib_buffer_get_current (b0); |
| 1431 | tcp0 = ip4_next_header (ip40); |
| 1432 | n_advance_bytes0 = (ip4_header_bytes (ip40) |
| 1433 | + tcp_header_bytes (tcp0)); |
| 1434 | n_data_bytes0 = clib_net_to_host_u16 (ip40->length) |
| 1435 | - n_advance_bytes0; |
| 1436 | } |
| 1437 | else |
| 1438 | { |
| 1439 | ip60 = vlib_buffer_get_current (b0); |
| 1440 | tcp0 = ip6_next_header (ip60); |
| 1441 | n_advance_bytes0 = tcp_header_bytes (tcp0); |
| 1442 | n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) |
| 1443 | - n_advance_bytes0; |
| 1444 | n_advance_bytes0 += sizeof (ip60[0]); |
| 1445 | } |
| 1446 | |
| 1447 | if (PREDICT_FALSE |
| 1448 | (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0))) |
| 1449 | goto drop; |
| 1450 | |
| 1451 | /* SYNs, FINs and data consume sequence numbers */ |
| 1452 | vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0) |
| 1453 | + tcp_is_fin (tcp0) + n_data_bytes0; |
| 1454 | |
| 1455 | /* |
| 1456 | * 1. check the ACK bit |
| 1457 | */ |
| 1458 | |
| 1459 | /* |
| 1460 | * If the ACK bit is set |
| 1461 | * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless |
| 1462 | * the RST bit is set, if so drop the segment and return) |
| 1463 | * <SEQ=SEG.ACK><CTL=RST> |
| 1464 | * and discard the segment. Return. |
| 1465 | * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable. |
| 1466 | */ |
| 1467 | if (tcp_ack (tcp0)) |
| 1468 | { |
| 1469 | if (ack0 <= tc0->iss || ack0 > tc0->snd_nxt) |
| 1470 | { |
| 1471 | if (!tcp_rst (tcp0)) |
| 1472 | tcp_send_reset (b0, is_ip4); |
| 1473 | |
| 1474 | goto drop; |
| 1475 | } |
| 1476 | |
| 1477 | /* Make sure ACK is valid */ |
| 1478 | if (tc0->snd_una > ack0) |
| 1479 | goto drop; |
| 1480 | } |
| 1481 | |
| 1482 | /* |
| 1483 | * 2. check the RST bit |
| 1484 | */ |
| 1485 | |
| 1486 | if (tcp_rst (tcp0)) |
| 1487 | { |
| 1488 | /* If ACK is acceptable, signal client that peer is not |
| 1489 | * willing to accept connection and drop connection*/ |
| 1490 | if (tcp_ack (tcp0)) |
| 1491 | { |
| 1492 | stream_session_connect_notify (&tc0->connection, sst, |
| 1493 | 1 /* fail */ ); |
| 1494 | tcp_connection_cleanup (tc0); |
| 1495 | } |
| 1496 | goto drop; |
| 1497 | } |
| 1498 | |
| 1499 | /* |
| 1500 | * 3. check the security and precedence (skipped) |
| 1501 | */ |
| 1502 | |
| 1503 | /* |
| 1504 | * 4. check the SYN bit |
| 1505 | */ |
| 1506 | |
| 1507 | /* No SYN flag. Drop. */ |
| 1508 | if (!tcp_syn (tcp0)) |
| 1509 | goto drop; |
| 1510 | |
| 1511 | /* Stop connection establishment and retransmit timers */ |
| 1512 | tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH); |
| 1513 | tcp_timer_reset (tc0, TCP_TIMER_RETRANSMIT_SYN); |
| 1514 | |
| 1515 | /* Valid SYN or SYN-ACK. Move connection from half-open pool to |
| 1516 | * current thread pool. */ |
| 1517 | pool_get (tm->connections[my_thread_index], new_tc0); |
| 1518 | clib_memcpy (new_tc0, tc0, sizeof (*new_tc0)); |
| 1519 | |
| 1520 | new_tc0->c_thread_index = my_thread_index; |
| 1521 | |
| 1522 | /* Cleanup half-open connection XXX lock */ |
| 1523 | pool_put (tm->half_open_connections, tc0); |
| 1524 | |
| 1525 | new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end; |
| 1526 | new_tc0->irs = seq0; |
| 1527 | |
| 1528 | /* Parse options */ |
| 1529 | tcp_options_parse (tcp0, &new_tc0->opt); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1530 | |
| 1531 | if (tcp_opts_tstamp (&new_tc0->opt)) |
| 1532 | { |
| 1533 | new_tc0->tsval_recent = new_tc0->opt.tsval; |
| 1534 | new_tc0->tsval_recent_age = tcp_time_now (); |
| 1535 | } |
| 1536 | |
| 1537 | if (tcp_opts_wscale (&new_tc0->opt)) |
| 1538 | new_tc0->snd_wscale = new_tc0->opt.wscale; |
| 1539 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 1540 | /* No scaling */ |
| 1541 | new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1542 | new_tc0->snd_wl1 = seq0; |
| 1543 | new_tc0->snd_wl2 = ack0; |
| 1544 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 1545 | tcp_connection_init_vars (new_tc0); |
| 1546 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1547 | /* SYN-ACK: See if we can switch to ESTABLISHED state */ |
| 1548 | if (tcp_ack (tcp0)) |
| 1549 | { |
| 1550 | /* Our SYN is ACKed: we have iss < ack = snd_una */ |
| 1551 | |
| 1552 | /* TODO Dequeue acknowledged segments if we support Fast Open */ |
| 1553 | new_tc0->snd_una = ack0; |
| 1554 | new_tc0->state = TCP_STATE_ESTABLISHED; |
| 1555 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 1556 | /* Make sure las is initialized for the wnd computation */ |
| 1557 | new_tc0->rcv_las = new_tc0->rcv_nxt; |
| 1558 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1559 | /* Notify app that we have connection */ |
| 1560 | stream_session_connect_notify (&new_tc0->connection, sst, 0); |
| 1561 | |
| 1562 | /* Make sure after data segment processing ACK is sent */ |
| 1563 | new_tc0->flags |= TCP_CONN_SNDACK; |
| 1564 | } |
| 1565 | /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */ |
| 1566 | else |
| 1567 | { |
| 1568 | new_tc0->state = TCP_STATE_SYN_RCVD; |
| 1569 | |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1570 | /* Notify app that we have connection */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1571 | stream_session_connect_notify (&new_tc0->connection, sst, 0); |
| 1572 | |
| 1573 | tcp_make_synack (new_tc0, b0); |
| 1574 | next0 = tcp_next_output (is_ip4); |
| 1575 | |
| 1576 | goto drop; |
| 1577 | } |
| 1578 | |
| 1579 | /* Read data, if any */ |
| 1580 | if (n_data_bytes0) |
| 1581 | { |
| 1582 | error0 = |
| 1583 | tcp_segment_rcv (tm, new_tc0, b0, n_data_bytes0, &next0); |
| 1584 | if (error0 == TCP_ERROR_PURE_ACK) |
| 1585 | error0 = TCP_ERROR_SYN_ACKS_RCVD; |
| 1586 | } |
| 1587 | else |
| 1588 | { |
| 1589 | tcp_make_ack (new_tc0, b0); |
| 1590 | next0 = tcp_next_output (new_tc0->c_is_ip4); |
| 1591 | } |
| 1592 | |
| 1593 | drop: |
| 1594 | |
| 1595 | b0->error = error0 ? node->errors[error0] : 0; |
| 1596 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 1597 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1598 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 1599 | clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header)); |
| 1600 | clib_memcpy (&t0->tcp_connection, tc0, |
| 1601 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1602 | } |
| 1603 | |
| 1604 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 1605 | n_left_to_next, bi0, next0); |
| 1606 | } |
| 1607 | |
| 1608 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 1609 | } |
| 1610 | |
| 1611 | errors = session_manager_flush_enqueue_events (my_thread_index); |
| 1612 | if (errors) |
| 1613 | { |
| 1614 | if (is_ip4) |
| 1615 | vlib_node_increment_counter (vm, tcp4_established_node.index, |
| 1616 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
| 1617 | else |
| 1618 | vlib_node_increment_counter (vm, tcp6_established_node.index, |
| 1619 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
| 1620 | } |
| 1621 | |
| 1622 | return from_frame->n_vectors; |
| 1623 | } |
| 1624 | |
| 1625 | static uword |
| 1626 | tcp4_syn_sent (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1627 | vlib_frame_t * from_frame) |
| 1628 | { |
| 1629 | return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 1630 | } |
| 1631 | |
| 1632 | static uword |
| 1633 | tcp6_syn_sent_rcv (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1634 | vlib_frame_t * from_frame) |
| 1635 | { |
| 1636 | return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 1637 | } |
| 1638 | |
| 1639 | /* *INDENT-OFF* */ |
| 1640 | VLIB_REGISTER_NODE (tcp4_syn_sent_node) = |
| 1641 | { |
| 1642 | .function = tcp4_syn_sent, |
| 1643 | .name = "tcp4-syn-sent", |
| 1644 | /* Takes a vector of packets. */ |
| 1645 | .vector_size = sizeof (u32), |
| 1646 | .n_errors = TCP_N_ERROR, |
| 1647 | .error_strings = tcp_error_strings, |
| 1648 | .n_next_nodes = TCP_SYN_SENT_N_NEXT, |
| 1649 | .next_nodes = |
| 1650 | { |
| 1651 | #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, |
| 1652 | foreach_tcp_state_next |
| 1653 | #undef _ |
| 1654 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1655 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1656 | }; |
| 1657 | /* *INDENT-ON* */ |
| 1658 | |
| 1659 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_syn_sent_node, tcp4_syn_sent); |
| 1660 | |
| 1661 | /* *INDENT-OFF* */ |
| 1662 | VLIB_REGISTER_NODE (tcp6_syn_sent_node) = |
| 1663 | { |
| 1664 | .function = tcp6_syn_sent_rcv, |
| 1665 | .name = "tcp6-syn-sent", |
| 1666 | /* Takes a vector of packets. */ |
| 1667 | .vector_size = sizeof (u32), |
| 1668 | .n_errors = TCP_N_ERROR, |
| 1669 | .error_strings = tcp_error_strings, |
| 1670 | .n_next_nodes = TCP_SYN_SENT_N_NEXT, |
| 1671 | .next_nodes = |
| 1672 | { |
| 1673 | #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, |
| 1674 | foreach_tcp_state_next |
| 1675 | #undef _ |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1676 | }, |
| 1677 | .format_trace = format_tcp_rx_trace_short, |
| 1678 | }; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1679 | /* *INDENT-ON* */ |
| 1680 | |
| 1681 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_syn_sent_node, tcp6_syn_sent_rcv); |
| 1682 | /** |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1683 | * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1684 | * as per RFC793 p. 64 |
| 1685 | */ |
| 1686 | always_inline uword |
| 1687 | tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1688 | vlib_frame_t * from_frame, int is_ip4) |
| 1689 | { |
| 1690 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 1691 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 1692 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1693 | |
| 1694 | from = vlib_frame_vector_args (from_frame); |
| 1695 | n_left_from = from_frame->n_vectors; |
| 1696 | |
| 1697 | next_index = node->cached_next_index; |
| 1698 | |
| 1699 | while (n_left_from > 0) |
| 1700 | { |
| 1701 | u32 n_left_to_next; |
| 1702 | |
| 1703 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 1704 | |
| 1705 | while (n_left_from > 0 && n_left_to_next > 0) |
| 1706 | { |
| 1707 | u32 bi0; |
| 1708 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1709 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1710 | tcp_header_t *tcp0 = 0; |
| 1711 | tcp_connection_t *tc0; |
| 1712 | ip4_header_t *ip40; |
| 1713 | ip6_header_t *ip60; |
| 1714 | u32 n_advance_bytes0, n_data_bytes0; |
| 1715 | u32 next0 = TCP_RCV_PROCESS_NEXT_DROP, error0 = TCP_ERROR_ENQUEUED; |
| 1716 | |
| 1717 | bi0 = from[0]; |
| 1718 | to_next[0] = bi0; |
| 1719 | from += 1; |
| 1720 | to_next += 1; |
| 1721 | n_left_from -= 1; |
| 1722 | n_left_to_next -= 1; |
| 1723 | |
| 1724 | b0 = vlib_get_buffer (vm, bi0); |
| 1725 | tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, |
| 1726 | my_thread_index); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1727 | if (PREDICT_FALSE (tc0 == 0)) |
| 1728 | { |
| 1729 | error0 = TCP_ERROR_INVALID_CONNECTION; |
| 1730 | goto drop; |
| 1731 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1732 | |
| 1733 | /* Checksum computed by ipx_local no need to compute again */ |
| 1734 | |
| 1735 | if (is_ip4) |
| 1736 | { |
| 1737 | ip40 = vlib_buffer_get_current (b0); |
| 1738 | tcp0 = ip4_next_header (ip40); |
| 1739 | n_advance_bytes0 = (ip4_header_bytes (ip40) |
| 1740 | + tcp_header_bytes (tcp0)); |
| 1741 | n_data_bytes0 = clib_net_to_host_u16 (ip40->length) |
| 1742 | - n_advance_bytes0; |
| 1743 | } |
| 1744 | else |
| 1745 | { |
| 1746 | ip60 = vlib_buffer_get_current (b0); |
| 1747 | tcp0 = ip6_next_header (ip60); |
| 1748 | n_advance_bytes0 = tcp_header_bytes (tcp0); |
| 1749 | n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) |
| 1750 | - n_advance_bytes0; |
| 1751 | n_advance_bytes0 += sizeof (ip60[0]); |
| 1752 | } |
| 1753 | |
| 1754 | /* SYNs, FINs and data consume sequence numbers */ |
| 1755 | vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number |
| 1756 | + tcp_is_syn (tcp0) + tcp_is_fin (tcp0) + n_data_bytes0; |
| 1757 | |
| 1758 | /* |
| 1759 | * Special treatment for CLOSED |
| 1760 | */ |
| 1761 | switch (tc0->state) |
| 1762 | { |
| 1763 | case TCP_STATE_CLOSED: |
| 1764 | goto drop; |
| 1765 | break; |
| 1766 | } |
| 1767 | |
| 1768 | /* |
| 1769 | * For all other states (except LISTEN) |
| 1770 | */ |
| 1771 | |
| 1772 | /* 1-4: check SEQ, RST, SYN */ |
| 1773 | if (PREDICT_FALSE |
| 1774 | (tcp_segment_validate (vm, tc0, b0, tcp0, &next0))) |
| 1775 | { |
| 1776 | error0 = TCP_ERROR_SEGMENT_INVALID; |
| 1777 | goto drop; |
| 1778 | } |
| 1779 | |
| 1780 | /* 5: check the ACK field */ |
| 1781 | switch (tc0->state) |
| 1782 | { |
| 1783 | case TCP_STATE_SYN_RCVD: |
| 1784 | /* |
| 1785 | * If the segment acknowledgment is not acceptable, form a |
| 1786 | * reset segment, |
| 1787 | * <SEQ=SEG.ACK><CTL=RST> |
| 1788 | * and send it. |
| 1789 | */ |
| 1790 | if (!tcp_rcv_ack_is_acceptable (tc0, b0)) |
| 1791 | { |
| 1792 | tcp_send_reset (b0, is_ip4); |
| 1793 | goto drop; |
| 1794 | } |
| 1795 | /* Switch state to ESTABLISHED */ |
| 1796 | tc0->state = TCP_STATE_ESTABLISHED; |
| 1797 | |
| 1798 | /* Initialize session variables */ |
| 1799 | tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 1800 | tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1801 | << tc0->opt.wscale; |
| 1802 | tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; |
| 1803 | tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; |
| 1804 | |
| 1805 | /* Shoulder tap the server */ |
| 1806 | stream_session_accept_notify (&tc0->connection); |
| 1807 | |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1808 | /* Reset SYN-ACK retransmit timer */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1809 | tcp_retransmit_timer_reset (tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1810 | break; |
| 1811 | case TCP_STATE_ESTABLISHED: |
| 1812 | /* We can get packets in established state here because they |
| 1813 | * were enqueued before state change */ |
| 1814 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 1815 | goto drop; |
| 1816 | |
| 1817 | break; |
| 1818 | case TCP_STATE_FIN_WAIT_1: |
| 1819 | /* In addition to the processing for the ESTABLISHED state, if |
| 1820 | * our FIN is now acknowledged then enter FIN-WAIT-2 and |
| 1821 | * continue processing in that state. */ |
| 1822 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 1823 | goto drop; |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1824 | |
| 1825 | /* If FIN is ACKed */ |
| 1826 | if (tc0->snd_una == tc0->snd_una_max) |
| 1827 | { |
| 1828 | tc0->state = TCP_STATE_FIN_WAIT_2; |
| 1829 | /* Stop all timers, 2MSL will be set lower */ |
| 1830 | tcp_connection_timers_reset (tc0); |
| 1831 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1832 | break; |
| 1833 | case TCP_STATE_FIN_WAIT_2: |
| 1834 | /* In addition to the processing for the ESTABLISHED state, if |
| 1835 | * the retransmission queue is empty, the user's CLOSE can be |
| 1836 | * acknowledged ("ok") but do not delete the TCB. */ |
| 1837 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 1838 | goto drop; |
| 1839 | /* check if rtx queue is empty and ack CLOSE TODO */ |
| 1840 | break; |
| 1841 | case TCP_STATE_CLOSE_WAIT: |
| 1842 | /* Do the same processing as for the ESTABLISHED state. */ |
| 1843 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 1844 | goto drop; |
| 1845 | break; |
| 1846 | case TCP_STATE_CLOSING: |
| 1847 | /* In addition to the processing for the ESTABLISHED state, if |
| 1848 | * the ACK acknowledges our FIN then enter the TIME-WAIT state, |
| 1849 | * otherwise ignore the segment. */ |
| 1850 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 1851 | goto drop; |
| 1852 | |
| 1853 | /* XXX test that send queue empty */ |
| 1854 | tc0->state = TCP_STATE_TIME_WAIT; |
| 1855 | goto drop; |
| 1856 | |
| 1857 | break; |
| 1858 | case TCP_STATE_LAST_ACK: |
| 1859 | /* The only thing that can arrive in this state is an |
| 1860 | * acknowledgment of our FIN. If our FIN is now acknowledged, |
| 1861 | * delete the TCB, enter the CLOSED state, and return. */ |
| 1862 | |
| 1863 | if (!tcp_rcv_ack_is_acceptable (tc0, b0)) |
| 1864 | goto drop; |
| 1865 | |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1866 | tc0->state = TCP_STATE_CLOSED; |
| 1867 | |
| 1868 | /* Don't delete the connection/session yet. Instead, wait a |
| 1869 | * reasonable amount of time until the pipes are cleared. In |
| 1870 | * particular, this makes sure that we won't have dead sessions |
| 1871 | * when processing events on the tx path */ |
| 1872 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME); |
| 1873 | |
| 1874 | /* Stop retransmit */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1875 | tcp_retransmit_timer_reset (tc0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1876 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1877 | goto drop; |
| 1878 | |
| 1879 | break; |
| 1880 | case TCP_STATE_TIME_WAIT: |
| 1881 | /* The only thing that can arrive in this state is a |
| 1882 | * retransmission of the remote FIN. Acknowledge it, and restart |
| 1883 | * the 2 MSL timeout. */ |
| 1884 | |
| 1885 | /* TODO */ |
| 1886 | goto drop; |
| 1887 | break; |
| 1888 | default: |
| 1889 | ASSERT (0); |
| 1890 | } |
| 1891 | |
| 1892 | /* 6: check the URG bit TODO */ |
| 1893 | |
| 1894 | /* 7: process the segment text */ |
| 1895 | switch (tc0->state) |
| 1896 | { |
| 1897 | case TCP_STATE_ESTABLISHED: |
| 1898 | case TCP_STATE_FIN_WAIT_1: |
| 1899 | case TCP_STATE_FIN_WAIT_2: |
Florin Coras | 6cf30ad | 2017-04-04 23:08:23 -0700 | [diff] [blame] | 1900 | vlib_buffer_advance (b0, n_advance_bytes0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1901 | error0 = tcp_segment_rcv (tm, tc0, b0, n_data_bytes0, &next0); |
| 1902 | break; |
| 1903 | case TCP_STATE_CLOSE_WAIT: |
| 1904 | case TCP_STATE_CLOSING: |
| 1905 | case TCP_STATE_LAST_ACK: |
| 1906 | case TCP_STATE_TIME_WAIT: |
| 1907 | /* This should not occur, since a FIN has been received from the |
| 1908 | * remote side. Ignore the segment text. */ |
| 1909 | break; |
| 1910 | } |
| 1911 | |
| 1912 | /* 8: check the FIN bit */ |
| 1913 | if (!tcp_fin (tcp0)) |
| 1914 | goto drop; |
| 1915 | |
| 1916 | switch (tc0->state) |
| 1917 | { |
| 1918 | case TCP_STATE_ESTABLISHED: |
| 1919 | case TCP_STATE_SYN_RCVD: |
| 1920 | /* Send FIN-ACK notify app and enter CLOSE-WAIT */ |
| 1921 | tcp_connection_timers_reset (tc0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1922 | tcp_make_fin (tc0, b0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1923 | next0 = tcp_next_output (tc0->c_is_ip4); |
| 1924 | stream_session_disconnect_notify (&tc0->connection); |
| 1925 | tc0->state = TCP_STATE_CLOSE_WAIT; |
| 1926 | break; |
| 1927 | case TCP_STATE_CLOSE_WAIT: |
| 1928 | case TCP_STATE_CLOSING: |
| 1929 | case TCP_STATE_LAST_ACK: |
| 1930 | /* move along .. */ |
| 1931 | break; |
| 1932 | case TCP_STATE_FIN_WAIT_1: |
| 1933 | tc0->state = TCP_STATE_TIME_WAIT; |
| 1934 | tcp_connection_timers_reset (tc0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1935 | tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1936 | break; |
| 1937 | case TCP_STATE_FIN_WAIT_2: |
| 1938 | /* Got FIN, send ACK! */ |
| 1939 | tc0->state = TCP_STATE_TIME_WAIT; |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1940 | tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1941 | tcp_make_ack (tc0, b0); |
| 1942 | next0 = tcp_next_output (is_ip4); |
| 1943 | break; |
| 1944 | case TCP_STATE_TIME_WAIT: |
| 1945 | /* Remain in the TIME-WAIT state. Restart the 2 MSL time-wait |
| 1946 | * timeout. |
| 1947 | */ |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1948 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1949 | break; |
| 1950 | } |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1951 | TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1952 | |
| 1953 | b0->error = error0 ? node->errors[error0] : 0; |
| 1954 | |
| 1955 | drop: |
| 1956 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 1957 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1958 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 1959 | clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header)); |
| 1960 | clib_memcpy (&t0->tcp_connection, tc0, |
| 1961 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1962 | } |
| 1963 | |
| 1964 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 1965 | n_left_to_next, bi0, next0); |
| 1966 | } |
| 1967 | |
| 1968 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 1969 | } |
| 1970 | |
| 1971 | errors = session_manager_flush_enqueue_events (my_thread_index); |
| 1972 | if (errors) |
| 1973 | { |
| 1974 | if (is_ip4) |
| 1975 | vlib_node_increment_counter (vm, tcp4_established_node.index, |
| 1976 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
| 1977 | else |
| 1978 | vlib_node_increment_counter (vm, tcp6_established_node.index, |
| 1979 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
| 1980 | } |
| 1981 | |
| 1982 | return from_frame->n_vectors; |
| 1983 | } |
| 1984 | |
| 1985 | static uword |
| 1986 | tcp4_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1987 | vlib_frame_t * from_frame) |
| 1988 | { |
| 1989 | return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 1990 | } |
| 1991 | |
| 1992 | static uword |
| 1993 | tcp6_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1994 | vlib_frame_t * from_frame) |
| 1995 | { |
| 1996 | return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 1997 | } |
| 1998 | |
| 1999 | /* *INDENT-OFF* */ |
| 2000 | VLIB_REGISTER_NODE (tcp4_rcv_process_node) = |
| 2001 | { |
| 2002 | .function = tcp4_rcv_process, |
| 2003 | .name = "tcp4-rcv-process", |
| 2004 | /* Takes a vector of packets. */ |
| 2005 | .vector_size = sizeof (u32), |
| 2006 | .n_errors = TCP_N_ERROR, |
| 2007 | .error_strings = tcp_error_strings, |
| 2008 | .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, |
| 2009 | .next_nodes = |
| 2010 | { |
| 2011 | #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, |
| 2012 | foreach_tcp_state_next |
| 2013 | #undef _ |
| 2014 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2015 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2016 | }; |
| 2017 | /* *INDENT-ON* */ |
| 2018 | |
| 2019 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_rcv_process_node, tcp4_rcv_process); |
| 2020 | |
| 2021 | /* *INDENT-OFF* */ |
| 2022 | VLIB_REGISTER_NODE (tcp6_rcv_process_node) = |
| 2023 | { |
| 2024 | .function = tcp6_rcv_process, |
| 2025 | .name = "tcp6-rcv-process", |
| 2026 | /* Takes a vector of packets. */ |
| 2027 | .vector_size = sizeof (u32), |
| 2028 | .n_errors = TCP_N_ERROR, |
| 2029 | .error_strings = tcp_error_strings, |
| 2030 | .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, |
| 2031 | .next_nodes = |
| 2032 | { |
| 2033 | #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, |
| 2034 | foreach_tcp_state_next |
| 2035 | #undef _ |
| 2036 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2037 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2038 | }; |
| 2039 | /* *INDENT-ON* */ |
| 2040 | |
| 2041 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_rcv_process_node, tcp6_rcv_process); |
| 2042 | |
| 2043 | vlib_node_registration_t tcp4_listen_node; |
| 2044 | vlib_node_registration_t tcp6_listen_node; |
| 2045 | |
| 2046 | /** |
| 2047 | * LISTEN state processing as per RFC 793 p. 65 |
| 2048 | */ |
| 2049 | always_inline uword |
| 2050 | tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2051 | vlib_frame_t * from_frame, int is_ip4) |
| 2052 | { |
| 2053 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 2054 | u32 my_thread_index = vm->thread_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2055 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 2056 | u8 sst = is_ip4 ? SESSION_TYPE_IP4_TCP : SESSION_TYPE_IP6_TCP; |
| 2057 | |
| 2058 | from = vlib_frame_vector_args (from_frame); |
| 2059 | n_left_from = from_frame->n_vectors; |
| 2060 | |
| 2061 | next_index = node->cached_next_index; |
| 2062 | |
| 2063 | while (n_left_from > 0) |
| 2064 | { |
| 2065 | u32 n_left_to_next; |
| 2066 | |
| 2067 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 2068 | |
| 2069 | while (n_left_from > 0 && n_left_to_next > 0) |
| 2070 | { |
| 2071 | u32 bi0; |
| 2072 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2073 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2074 | tcp_header_t *th0 = 0; |
| 2075 | tcp_connection_t *lc0; |
| 2076 | ip4_header_t *ip40; |
| 2077 | ip6_header_t *ip60; |
| 2078 | tcp_connection_t *child0; |
| 2079 | u32 error0 = TCP_ERROR_SYNS_RCVD, next0 = TCP_LISTEN_NEXT_DROP; |
| 2080 | |
| 2081 | bi0 = from[0]; |
| 2082 | to_next[0] = bi0; |
| 2083 | from += 1; |
| 2084 | to_next += 1; |
| 2085 | n_left_from -= 1; |
| 2086 | n_left_to_next -= 1; |
| 2087 | |
| 2088 | b0 = vlib_get_buffer (vm, bi0); |
| 2089 | lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index); |
| 2090 | |
| 2091 | if (is_ip4) |
| 2092 | { |
| 2093 | ip40 = vlib_buffer_get_current (b0); |
| 2094 | th0 = ip4_next_header (ip40); |
| 2095 | } |
| 2096 | else |
| 2097 | { |
| 2098 | ip60 = vlib_buffer_get_current (b0); |
| 2099 | th0 = ip6_next_header (ip60); |
| 2100 | } |
| 2101 | |
| 2102 | /* Create child session. For syn-flood protection use filter */ |
| 2103 | |
| 2104 | /* 1. first check for an RST */ |
| 2105 | if (tcp_rst (th0)) |
| 2106 | goto drop; |
| 2107 | |
| 2108 | /* 2. second check for an ACK */ |
| 2109 | if (tcp_ack (th0)) |
| 2110 | { |
| 2111 | tcp_send_reset (b0, is_ip4); |
| 2112 | goto drop; |
| 2113 | } |
| 2114 | |
| 2115 | /* 3. check for a SYN (did that already) */ |
| 2116 | |
| 2117 | /* Create child session and send SYN-ACK */ |
| 2118 | pool_get (tm->connections[my_thread_index], child0); |
| 2119 | memset (child0, 0, sizeof (*child0)); |
| 2120 | |
| 2121 | child0->c_c_index = child0 - tm->connections[my_thread_index]; |
| 2122 | child0->c_lcl_port = lc0->c_lcl_port; |
| 2123 | child0->c_rmt_port = th0->src_port; |
| 2124 | child0->c_is_ip4 = is_ip4; |
| 2125 | child0->c_thread_index = my_thread_index; |
| 2126 | |
| 2127 | if (is_ip4) |
| 2128 | { |
| 2129 | child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32; |
| 2130 | child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32; |
| 2131 | } |
| 2132 | else |
| 2133 | { |
| 2134 | clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address, |
| 2135 | sizeof (ip6_address_t)); |
| 2136 | clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address, |
| 2137 | sizeof (ip6_address_t)); |
| 2138 | } |
| 2139 | |
| 2140 | if (stream_session_accept (&child0->connection, lc0->c_s_index, sst, |
| 2141 | 0 /* notify */ )) |
| 2142 | { |
| 2143 | error0 = TCP_ERROR_CREATE_SESSION_FAIL; |
| 2144 | goto drop; |
| 2145 | } |
| 2146 | |
| 2147 | tcp_options_parse (th0, &child0->opt); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2148 | |
| 2149 | child0->irs = vnet_buffer (b0)->tcp.seq_number; |
| 2150 | child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1; |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 2151 | child0->rcv_las = child0->rcv_nxt; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2152 | child0->state = TCP_STATE_SYN_RCVD; |
| 2153 | |
| 2154 | /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK} |
| 2155 | * segments are used to initialize PAWS. */ |
| 2156 | if (tcp_opts_tstamp (&child0->opt)) |
| 2157 | { |
| 2158 | child0->tsval_recent = child0->opt.tsval; |
| 2159 | child0->tsval_recent_age = tcp_time_now (); |
| 2160 | } |
| 2161 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2162 | if (tcp_opts_wscale (&child0->opt)) |
| 2163 | child0->snd_wscale = child0->opt.wscale; |
| 2164 | |
| 2165 | /* No scaling */ |
| 2166 | child0->snd_wnd = clib_net_to_host_u16 (th0->window); |
| 2167 | child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; |
| 2168 | child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; |
| 2169 | |
| 2170 | tcp_connection_init_vars (child0); |
| 2171 | |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 2172 | TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0); |
| 2173 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2174 | /* Reuse buffer to make syn-ack and send */ |
| 2175 | tcp_make_synack (child0, b0); |
| 2176 | next0 = tcp_next_output (is_ip4); |
| 2177 | |
| 2178 | drop: |
| 2179 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 2180 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2181 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 2182 | clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header)); |
| 2183 | clib_memcpy (&t0->tcp_connection, lc0, |
| 2184 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2185 | } |
| 2186 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2187 | b0->error = node->errors[error0]; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2188 | |
| 2189 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 2190 | n_left_to_next, bi0, next0); |
| 2191 | } |
| 2192 | |
| 2193 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 2194 | } |
| 2195 | return from_frame->n_vectors; |
| 2196 | } |
| 2197 | |
| 2198 | static uword |
| 2199 | tcp4_listen (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2200 | vlib_frame_t * from_frame) |
| 2201 | { |
| 2202 | return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 2203 | } |
| 2204 | |
| 2205 | static uword |
| 2206 | tcp6_listen (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2207 | vlib_frame_t * from_frame) |
| 2208 | { |
| 2209 | return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 2210 | } |
| 2211 | |
| 2212 | /* *INDENT-OFF* */ |
| 2213 | VLIB_REGISTER_NODE (tcp4_listen_node) = |
| 2214 | { |
| 2215 | .function = tcp4_listen, |
| 2216 | .name = "tcp4-listen", |
| 2217 | /* Takes a vector of packets. */ |
| 2218 | .vector_size = sizeof (u32), |
| 2219 | .n_errors = TCP_N_ERROR, |
| 2220 | .error_strings = tcp_error_strings, |
| 2221 | .n_next_nodes = TCP_LISTEN_N_NEXT, |
| 2222 | .next_nodes = |
| 2223 | { |
| 2224 | #define _(s,n) [TCP_LISTEN_NEXT_##s] = n, |
| 2225 | foreach_tcp_state_next |
| 2226 | #undef _ |
| 2227 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2228 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2229 | }; |
| 2230 | /* *INDENT-ON* */ |
| 2231 | |
| 2232 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_listen_node, tcp4_listen); |
| 2233 | |
| 2234 | /* *INDENT-OFF* */ |
| 2235 | VLIB_REGISTER_NODE (tcp6_listen_node) = |
| 2236 | { |
| 2237 | .function = tcp6_listen, |
| 2238 | .name = "tcp6-listen", |
| 2239 | /* Takes a vector of packets. */ |
| 2240 | .vector_size = sizeof (u32), |
| 2241 | .n_errors = TCP_N_ERROR, |
| 2242 | .error_strings = tcp_error_strings, |
| 2243 | .n_next_nodes = TCP_LISTEN_N_NEXT, |
| 2244 | .next_nodes = |
| 2245 | { |
| 2246 | #define _(s,n) [TCP_LISTEN_NEXT_##s] = n, |
| 2247 | foreach_tcp_state_next |
| 2248 | #undef _ |
| 2249 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2250 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2251 | }; |
| 2252 | /* *INDENT-ON* */ |
| 2253 | |
| 2254 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_listen_node, tcp6_listen); |
| 2255 | |
| 2256 | vlib_node_registration_t tcp4_input_node; |
| 2257 | vlib_node_registration_t tcp6_input_node; |
| 2258 | |
| 2259 | typedef enum _tcp_input_next |
| 2260 | { |
| 2261 | TCP_INPUT_NEXT_DROP, |
| 2262 | TCP_INPUT_NEXT_LISTEN, |
| 2263 | TCP_INPUT_NEXT_RCV_PROCESS, |
| 2264 | TCP_INPUT_NEXT_SYN_SENT, |
| 2265 | TCP_INPUT_NEXT_ESTABLISHED, |
| 2266 | TCP_INPUT_NEXT_RESET, |
| 2267 | TCP_INPUT_N_NEXT |
| 2268 | } tcp_input_next_t; |
| 2269 | |
| 2270 | #define foreach_tcp4_input_next \ |
| 2271 | _ (DROP, "error-drop") \ |
| 2272 | _ (LISTEN, "tcp4-listen") \ |
| 2273 | _ (RCV_PROCESS, "tcp4-rcv-process") \ |
| 2274 | _ (SYN_SENT, "tcp4-syn-sent") \ |
| 2275 | _ (ESTABLISHED, "tcp4-established") \ |
| 2276 | _ (RESET, "tcp4-reset") |
| 2277 | |
| 2278 | #define foreach_tcp6_input_next \ |
| 2279 | _ (DROP, "error-drop") \ |
| 2280 | _ (LISTEN, "tcp6-listen") \ |
| 2281 | _ (RCV_PROCESS, "tcp6-rcv-process") \ |
| 2282 | _ (SYN_SENT, "tcp6-syn-sent") \ |
| 2283 | _ (ESTABLISHED, "tcp6-established") \ |
| 2284 | _ (RESET, "tcp6-reset") |
| 2285 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2286 | #define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN) |
| 2287 | |
| 2288 | always_inline uword |
| 2289 | tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2290 | vlib_frame_t * from_frame, int is_ip4) |
| 2291 | { |
| 2292 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 2293 | u32 my_thread_index = vm->thread_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2294 | tcp_main_t *tm = vnet_get_tcp_main (); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2295 | |
| 2296 | from = vlib_frame_vector_args (from_frame); |
| 2297 | n_left_from = from_frame->n_vectors; |
| 2298 | |
| 2299 | next_index = node->cached_next_index; |
| 2300 | |
| 2301 | while (n_left_from > 0) |
| 2302 | { |
| 2303 | u32 n_left_to_next; |
| 2304 | |
| 2305 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 2306 | |
| 2307 | while (n_left_from > 0 && n_left_to_next > 0) |
| 2308 | { |
| 2309 | u32 bi0; |
| 2310 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2311 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2312 | tcp_header_t *tcp0 = 0; |
| 2313 | tcp_connection_t *tc0; |
| 2314 | ip4_header_t *ip40; |
| 2315 | ip6_header_t *ip60; |
| 2316 | u32 error0 = TCP_ERROR_NO_LISTENER, next0 = TCP_INPUT_NEXT_DROP; |
| 2317 | u8 flags0; |
| 2318 | |
| 2319 | bi0 = from[0]; |
| 2320 | to_next[0] = bi0; |
| 2321 | from += 1; |
| 2322 | to_next += 1; |
| 2323 | n_left_from -= 1; |
| 2324 | n_left_to_next -= 1; |
| 2325 | |
| 2326 | b0 = vlib_get_buffer (vm, bi0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2327 | vnet_buffer (b0)->tcp.flags = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2328 | |
| 2329 | if (is_ip4) |
| 2330 | { |
| 2331 | ip40 = vlib_buffer_get_current (b0); |
| 2332 | tcp0 = ip4_next_header (ip40); |
| 2333 | |
| 2334 | /* lookup session */ |
| 2335 | tc0 = |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2336 | (tcp_connection_t *) |
| 2337 | stream_session_lookup_transport4 (&ip40->dst_address, |
| 2338 | &ip40->src_address, |
| 2339 | tcp0->dst_port, |
| 2340 | tcp0->src_port, |
| 2341 | SESSION_TYPE_IP4_TCP, |
| 2342 | my_thread_index); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2343 | } |
| 2344 | else |
| 2345 | { |
| 2346 | ip60 = vlib_buffer_get_current (b0); |
| 2347 | tcp0 = ip6_next_header (ip60); |
| 2348 | tc0 = |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2349 | (tcp_connection_t *) |
| 2350 | stream_session_lookup_transport6 (&ip60->src_address, |
| 2351 | &ip60->dst_address, |
| 2352 | tcp0->src_port, |
| 2353 | tcp0->dst_port, |
| 2354 | SESSION_TYPE_IP6_TCP, |
| 2355 | my_thread_index); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2356 | } |
| 2357 | |
| 2358 | /* Session exists */ |
| 2359 | if (PREDICT_TRUE (0 != tc0)) |
| 2360 | { |
| 2361 | /* Save connection index */ |
| 2362 | vnet_buffer (b0)->tcp.connection_index = tc0->c_c_index; |
| 2363 | vnet_buffer (b0)->tcp.seq_number = |
| 2364 | clib_net_to_host_u32 (tcp0->seq_number); |
| 2365 | vnet_buffer (b0)->tcp.ack_number = |
| 2366 | clib_net_to_host_u32 (tcp0->ack_number); |
| 2367 | |
| 2368 | flags0 = tcp0->flags & filter_flags; |
| 2369 | next0 = tm->dispatch_table[tc0->state][flags0].next; |
| 2370 | error0 = tm->dispatch_table[tc0->state][flags0].error; |
| 2371 | |
| 2372 | if (PREDICT_FALSE (error0 == TCP_ERROR_DISPATCH)) |
| 2373 | { |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 2374 | tcp_state_t state0 = tc0->state; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2375 | /* Overload tcp flags to store state */ |
| 2376 | vnet_buffer (b0)->tcp.flags = tc0->state; |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 2377 | clib_warning ("disp error state %U flags %U", |
| 2378 | format_tcp_state, &state0, |
| 2379 | format_tcp_flags, flags0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2380 | } |
| 2381 | } |
| 2382 | else |
| 2383 | { |
| 2384 | /* Send reset */ |
| 2385 | next0 = TCP_INPUT_NEXT_RESET; |
| 2386 | error0 = TCP_ERROR_NO_LISTENER; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2387 | } |
| 2388 | |
| 2389 | b0->error = error0 ? node->errors[error0] : 0; |
| 2390 | |
| 2391 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 2392 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2393 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 2394 | clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header)); |
Florin Coras | a546481 | 2017-04-19 13:00:05 -0700 | [diff] [blame^] | 2395 | if (tc0) |
| 2396 | clib_memcpy (&t0->tcp_connection, tc0, sizeof (*tc0)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2397 | } |
| 2398 | |
| 2399 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 2400 | n_left_to_next, bi0, next0); |
| 2401 | } |
| 2402 | |
| 2403 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 2404 | } |
| 2405 | |
| 2406 | return from_frame->n_vectors; |
| 2407 | } |
| 2408 | |
| 2409 | static uword |
| 2410 | tcp4_input (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2411 | vlib_frame_t * from_frame) |
| 2412 | { |
| 2413 | return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 2414 | } |
| 2415 | |
| 2416 | static uword |
| 2417 | tcp6_input (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2418 | vlib_frame_t * from_frame) |
| 2419 | { |
| 2420 | return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 2421 | } |
| 2422 | |
| 2423 | /* *INDENT-OFF* */ |
| 2424 | VLIB_REGISTER_NODE (tcp4_input_node) = |
| 2425 | { |
| 2426 | .function = tcp4_input, |
| 2427 | .name = "tcp4-input", |
| 2428 | /* Takes a vector of packets. */ |
| 2429 | .vector_size = sizeof (u32), |
| 2430 | .n_errors = TCP_N_ERROR, |
| 2431 | .error_strings = tcp_error_strings, |
| 2432 | .n_next_nodes = TCP_INPUT_N_NEXT, |
| 2433 | .next_nodes = |
| 2434 | { |
| 2435 | #define _(s,n) [TCP_INPUT_NEXT_##s] = n, |
| 2436 | foreach_tcp4_input_next |
| 2437 | #undef _ |
| 2438 | }, |
| 2439 | .format_buffer = format_tcp_header, |
| 2440 | .format_trace = format_tcp_rx_trace, |
| 2441 | }; |
| 2442 | /* *INDENT-ON* */ |
| 2443 | |
| 2444 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_input_node, tcp4_input); |
| 2445 | |
| 2446 | /* *INDENT-OFF* */ |
| 2447 | VLIB_REGISTER_NODE (tcp6_input_node) = |
| 2448 | { |
| 2449 | .function = tcp6_input, |
| 2450 | .name = "tcp6-input", |
| 2451 | /* Takes a vector of packets. */ |
| 2452 | .vector_size = sizeof (u32), |
| 2453 | .n_errors = TCP_N_ERROR, |
| 2454 | .error_strings = tcp_error_strings, |
| 2455 | .n_next_nodes = TCP_INPUT_N_NEXT, |
| 2456 | .next_nodes = |
| 2457 | { |
| 2458 | #define _(s,n) [TCP_INPUT_NEXT_##s] = n, |
| 2459 | foreach_tcp6_input_next |
| 2460 | #undef _ |
| 2461 | }, |
| 2462 | .format_buffer = format_tcp_header, |
| 2463 | .format_trace = format_tcp_rx_trace, |
| 2464 | }; |
| 2465 | /* *INDENT-ON* */ |
| 2466 | |
| 2467 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_input_node, tcp6_input); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2468 | |
| 2469 | static void |
| 2470 | tcp_dispatch_table_init (tcp_main_t * tm) |
| 2471 | { |
| 2472 | int i, j; |
| 2473 | for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++) |
| 2474 | for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++) |
| 2475 | { |
| 2476 | tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP; |
| 2477 | tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH; |
| 2478 | } |
| 2479 | |
| 2480 | #define _(t,f,n,e) \ |
| 2481 | do { \ |
| 2482 | tm->dispatch_table[TCP_STATE_##t][f].next = (n); \ |
| 2483 | tm->dispatch_table[TCP_STATE_##t][f].error = (e); \ |
| 2484 | } while (0) |
| 2485 | |
| 2486 | /* SYNs for new connections -> tcp-listen. */ |
| 2487 | _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE); |
| 2488 | /* ACK for for a SYN-ACK -> tcp-rcv-process. */ |
| 2489 | _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 2490 | /* SYN-ACK for a SYN */ |
| 2491 | _(SYN_SENT, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, |
| 2492 | TCP_ERROR_NONE); |
| 2493 | _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); |
| 2494 | _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); |
| 2495 | _(SYN_SENT, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, |
| 2496 | TCP_ERROR_NONE); |
| 2497 | /* ACK for for established connection -> tcp-established. */ |
| 2498 | _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
| 2499 | /* FIN for for established connection -> tcp-established. */ |
| 2500 | _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
| 2501 | _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, |
| 2502 | TCP_ERROR_NONE); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2503 | _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2504 | /* ACK or FIN-ACK to our FIN */ |
| 2505 | _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 2506 | _(FIN_WAIT_1, TCP_FLAG_ACK | TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, |
| 2507 | TCP_ERROR_NONE); |
| 2508 | /* FIN in reply to our FIN from the other side */ |
| 2509 | _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 2510 | /* FIN confirming that the peer (app) has closed */ |
| 2511 | _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 2512 | _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, |
| 2513 | TCP_ERROR_NONE); |
| 2514 | _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 2515 | #undef _ |
| 2516 | } |
| 2517 | |
| 2518 | clib_error_t * |
| 2519 | tcp_input_init (vlib_main_t * vm) |
| 2520 | { |
| 2521 | clib_error_t *error = 0; |
| 2522 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 2523 | |
| 2524 | if ((error = vlib_call_init_function (vm, tcp_init))) |
| 2525 | return error; |
| 2526 | |
| 2527 | /* Initialize dispatch table. */ |
| 2528 | tcp_dispatch_table_init (tm); |
| 2529 | |
| 2530 | return error; |
| 2531 | } |
| 2532 | |
| 2533 | VLIB_INIT_FUNCTION (tcp_input_init); |
| 2534 | |
| 2535 | /* |
| 2536 | * fd.io coding-style-patch-verification: ON |
| 2537 | * |
| 2538 | * Local Variables: |
| 2539 | * eval: (c-set-style "gnu") |
| 2540 | * End: |
| 2541 | */ |