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 \ |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 30 | _ (DROP4, "ip4-drop") \ |
| 31 | _ (DROP6, "ip6-drop") \ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 32 | _ (TCP4_OUTPUT, "tcp4-output") \ |
| 33 | _ (TCP6_OUTPUT, "tcp6-output") |
| 34 | |
| 35 | typedef enum _tcp_established_next |
| 36 | { |
| 37 | #define _(s,n) TCP_ESTABLISHED_NEXT_##s, |
| 38 | foreach_tcp_state_next |
| 39 | #undef _ |
| 40 | TCP_ESTABLISHED_N_NEXT, |
| 41 | } tcp_established_next_t; |
| 42 | |
| 43 | typedef enum _tcp_rcv_process_next |
| 44 | { |
| 45 | #define _(s,n) TCP_RCV_PROCESS_NEXT_##s, |
| 46 | foreach_tcp_state_next |
| 47 | #undef _ |
| 48 | TCP_RCV_PROCESS_N_NEXT, |
| 49 | } tcp_rcv_process_next_t; |
| 50 | |
| 51 | typedef enum _tcp_syn_sent_next |
| 52 | { |
| 53 | #define _(s,n) TCP_SYN_SENT_NEXT_##s, |
| 54 | foreach_tcp_state_next |
| 55 | #undef _ |
| 56 | TCP_SYN_SENT_N_NEXT, |
| 57 | } tcp_syn_sent_next_t; |
| 58 | |
| 59 | typedef enum _tcp_listen_next |
| 60 | { |
| 61 | #define _(s,n) TCP_LISTEN_NEXT_##s, |
| 62 | foreach_tcp_state_next |
| 63 | #undef _ |
| 64 | TCP_LISTEN_N_NEXT, |
| 65 | } tcp_listen_next_t; |
| 66 | |
| 67 | /* Generic, state independent indices */ |
| 68 | typedef enum _tcp_state_next |
| 69 | { |
| 70 | #define _(s,n) TCP_NEXT_##s, |
| 71 | foreach_tcp_state_next |
| 72 | #undef _ |
| 73 | TCP_STATE_N_NEXT, |
| 74 | } tcp_state_next_t; |
| 75 | |
| 76 | #define tcp_next_output(is_ip4) (is_ip4 ? TCP_NEXT_TCP4_OUTPUT \ |
| 77 | : TCP_NEXT_TCP6_OUTPUT) |
| 78 | |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 79 | #define tcp_next_drop(is_ip4) (is_ip4 ? TCP_NEXT_DROP4 \ |
| 80 | : TCP_NEXT_DROP6) |
| 81 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 82 | vlib_node_registration_t tcp4_established_node; |
| 83 | vlib_node_registration_t tcp6_established_node; |
| 84 | |
| 85 | /** |
| 86 | * Validate segment sequence number. As per RFC793: |
| 87 | * |
| 88 | * Segment Receive Test |
| 89 | * Length Window |
| 90 | * ------- ------- ------------------------------------------- |
| 91 | * 0 0 SEG.SEQ = RCV.NXT |
| 92 | * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND |
| 93 | * >0 0 not acceptable |
| 94 | * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND |
| 95 | * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND |
| 96 | * |
| 97 | * This ultimately consists in checking if segment falls within the window. |
| 98 | * The one important difference compared to RFC793 is that we use rcv_las, |
| 99 | * or the rcv_nxt at last ack sent instead of rcv_nxt since that's the |
| 100 | * peer's reference when computing our receive window. |
| 101 | * |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 102 | * This: |
| 103 | * seq_leq (end_seq, tc->rcv_las + tc->rcv_wnd) && seq_geq (seq, tc->rcv_las) |
| 104 | * however, is too strict when we have retransmits. Instead we just check that |
| 105 | * the seq is not beyond the right edge and that the end of the segment is not |
| 106 | * less than the left edge. |
| 107 | * |
| 108 | * N.B. rcv_nxt and rcv_wnd are both updated in this node if acks are sent, so |
| 109 | * use rcv_nxt in the right edge window test instead of rcv_las. |
| 110 | * |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 111 | */ |
| 112 | always_inline u8 |
| 113 | tcp_segment_in_rcv_wnd (tcp_connection_t * tc, u32 seq, u32 end_seq) |
| 114 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 115 | return (seq_geq (end_seq, tc->rcv_las) |
| 116 | && seq_leq (seq, tc->rcv_nxt + tc->rcv_wnd)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 117 | } |
| 118 | |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 119 | /** |
| 120 | * Parse TCP header options. |
| 121 | * |
| 122 | * @param th TCP header |
| 123 | * @param to TCP options data structure to be populated |
| 124 | * @return -1 if parsing failed |
| 125 | */ |
| 126 | int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 127 | tcp_options_parse (tcp_header_t * th, tcp_options_t * to) |
| 128 | { |
| 129 | const u8 *data; |
| 130 | u8 opt_len, opts_len, kind; |
| 131 | int j; |
| 132 | sack_block_t b; |
| 133 | |
| 134 | opts_len = (tcp_doff (th) << 2) - sizeof (tcp_header_t); |
| 135 | data = (const u8 *) (th + 1); |
| 136 | |
| 137 | /* Zero out all flags but those set in SYN */ |
| 138 | to->flags &= (TCP_OPTS_FLAG_SACK_PERMITTED | TCP_OPTS_FLAG_WSCALE); |
| 139 | |
| 140 | for (; opts_len > 0; opts_len -= opt_len, data += opt_len) |
| 141 | { |
| 142 | kind = data[0]; |
| 143 | |
| 144 | /* Get options length */ |
| 145 | if (kind == TCP_OPTION_EOL) |
| 146 | break; |
| 147 | else if (kind == TCP_OPTION_NOOP) |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 148 | { |
| 149 | opt_len = 1; |
| 150 | continue; |
| 151 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 152 | else |
| 153 | { |
| 154 | /* broken options */ |
| 155 | if (opts_len < 2) |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 156 | return -1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 157 | opt_len = data[1]; |
| 158 | |
| 159 | /* weird option length */ |
| 160 | if (opt_len < 2 || opt_len > opts_len) |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 161 | return -1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 162 | } |
| 163 | |
| 164 | /* Parse options */ |
| 165 | switch (kind) |
| 166 | { |
| 167 | case TCP_OPTION_MSS: |
| 168 | if ((opt_len == TCP_OPTION_LEN_MSS) && tcp_syn (th)) |
| 169 | { |
| 170 | to->flags |= TCP_OPTS_FLAG_MSS; |
| 171 | to->mss = clib_net_to_host_u16 (*(u16 *) (data + 2)); |
| 172 | } |
| 173 | break; |
| 174 | case TCP_OPTION_WINDOW_SCALE: |
| 175 | if ((opt_len == TCP_OPTION_LEN_WINDOW_SCALE) && tcp_syn (th)) |
| 176 | { |
| 177 | to->flags |= TCP_OPTS_FLAG_WSCALE; |
| 178 | to->wscale = data[2]; |
| 179 | if (to->wscale > TCP_MAX_WND_SCALE) |
| 180 | { |
| 181 | clib_warning ("Illegal window scaling value: %d", |
| 182 | to->wscale); |
| 183 | to->wscale = TCP_MAX_WND_SCALE; |
| 184 | } |
| 185 | } |
| 186 | break; |
| 187 | case TCP_OPTION_TIMESTAMP: |
| 188 | if (opt_len == TCP_OPTION_LEN_TIMESTAMP) |
| 189 | { |
| 190 | to->flags |= TCP_OPTS_FLAG_TSTAMP; |
| 191 | to->tsval = clib_net_to_host_u32 (*(u32 *) (data + 2)); |
| 192 | to->tsecr = clib_net_to_host_u32 (*(u32 *) (data + 6)); |
| 193 | } |
| 194 | break; |
| 195 | case TCP_OPTION_SACK_PERMITTED: |
| 196 | if (opt_len == TCP_OPTION_LEN_SACK_PERMITTED && tcp_syn (th)) |
| 197 | to->flags |= TCP_OPTS_FLAG_SACK_PERMITTED; |
| 198 | break; |
| 199 | case TCP_OPTION_SACK_BLOCK: |
| 200 | /* If SACK permitted was not advertised or a SYN, break */ |
| 201 | if ((to->flags & TCP_OPTS_FLAG_SACK_PERMITTED) == 0 || tcp_syn (th)) |
| 202 | break; |
| 203 | |
| 204 | /* If too short or not correctly formatted, break */ |
| 205 | if (opt_len < 10 || ((opt_len - 2) % TCP_OPTION_LEN_SACK_BLOCK)) |
| 206 | break; |
| 207 | |
| 208 | to->flags |= TCP_OPTS_FLAG_SACK; |
| 209 | to->n_sack_blocks = (opt_len - 2) / TCP_OPTION_LEN_SACK_BLOCK; |
| 210 | vec_reset_length (to->sacks); |
| 211 | for (j = 0; j < to->n_sack_blocks; j++) |
| 212 | { |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 213 | b.start = clib_net_to_host_u32 (*(u32 *) (data + 2 + 8 * j)); |
| 214 | b.end = clib_net_to_host_u32 (*(u32 *) (data + 6 + 8 * j)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 215 | vec_add1 (to->sacks, b); |
| 216 | } |
| 217 | break; |
| 218 | default: |
| 219 | /* Nothing to see here */ |
| 220 | continue; |
| 221 | } |
| 222 | } |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 223 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 224 | } |
| 225 | |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 226 | /** |
| 227 | * RFC1323: Check against wrapped sequence numbers (PAWS). If we have |
| 228 | * timestamp to echo and it's less than tsval_recent, drop segment |
| 229 | * but still send an ACK in order to retain TCP's mechanism for detecting |
| 230 | * and recovering from half-open connections |
| 231 | * |
| 232 | * Or at least that's what the theory says. It seems that this might not work |
| 233 | * very well with packet reordering and fast retransmit. XXX |
| 234 | */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 235 | always_inline int |
| 236 | tcp_segment_check_paws (tcp_connection_t * tc) |
| 237 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 238 | return tcp_opts_tstamp (&tc->rcv_opts) && tc->tsval_recent |
| 239 | && timestamp_lt (tc->rcv_opts.tsval, tc->tsval_recent); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 240 | } |
| 241 | |
| 242 | /** |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 243 | * Update tsval recent |
| 244 | */ |
| 245 | always_inline void |
| 246 | tcp_update_timestamp (tcp_connection_t * tc, u32 seq, u32 seq_end) |
| 247 | { |
| 248 | /* |
| 249 | * RFC1323: If Last.ACK.sent falls within the range of sequence numbers |
| 250 | * of an incoming segment: |
| 251 | * SEG.SEQ <= Last.ACK.sent < SEG.SEQ + SEG.LEN |
| 252 | * then the TSval from the segment is copied to TS.Recent; |
| 253 | * otherwise, the TSval is ignored. |
| 254 | */ |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 255 | if (tcp_opts_tstamp (&tc->rcv_opts) && seq_leq (seq, tc->rcv_las) |
| 256 | && seq_leq (tc->rcv_las, seq_end)) |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 257 | { |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 258 | ASSERT (timestamp_leq (tc->tsval_recent, tc->rcv_opts.tsval)); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 259 | tc->tsval_recent = tc->rcv_opts.tsval; |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 260 | tc->tsval_recent_age = tcp_time_now (); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | /** |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 265 | * Validate incoming segment as per RFC793 p. 69 and RFC1323 p. 19 |
| 266 | * |
| 267 | * It first verifies if segment has a wrapped sequence number (PAWS) and then |
| 268 | * does the processing associated to the first four steps (ignoring security |
| 269 | * and precedence): sequence number, rst bit and syn bit checks. |
| 270 | * |
| 271 | * @return 0 if segments passes validation. |
| 272 | */ |
| 273 | static int |
| 274 | tcp_segment_validate (vlib_main_t * vm, tcp_connection_t * tc0, |
| 275 | vlib_buffer_t * b0, tcp_header_t * th0, u32 * next0) |
| 276 | { |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 277 | if (PREDICT_FALSE (!tcp_ack (th0) && !tcp_rst (th0) && !tcp_syn (th0))) |
| 278 | return -1; |
| 279 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 280 | if (PREDICT_FALSE (tcp_options_parse (th0, &tc0->rcv_opts))) |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 281 | { |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 282 | clib_warning ("options parse error"); |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 283 | return -1; |
| 284 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 285 | |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 286 | if (tcp_segment_check_paws (tc0)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 287 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 288 | if (CLIB_DEBUG > 2) |
| 289 | { |
| 290 | clib_warning ("paws failed\n%U", format_tcp_connection, tc0, 2); |
| 291 | clib_warning ("seq %u seq_end %u ack %u", |
| 292 | vnet_buffer (b0)->tcp.seq_number - tc0->irs, |
| 293 | vnet_buffer (b0)->tcp.seq_end - tc0->irs, |
| 294 | vnet_buffer (b0)->tcp.ack_number - tc0->iss); |
| 295 | } |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 296 | TCP_EVT_DBG (TCP_EVT_PAWS_FAIL, tc0, vnet_buffer (b0)->tcp.seq_number, |
| 297 | vnet_buffer (b0)->tcp.seq_end); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 298 | |
| 299 | /* If it just so happens that a segment updates tsval_recent for a |
| 300 | * segment over 24 days old, invalidate tsval_recent. */ |
| 301 | if (timestamp_lt (tc0->tsval_recent_age + TCP_PAWS_IDLE, |
| 302 | tcp_time_now ())) |
| 303 | { |
| 304 | /* Age isn't reset until we get a valid tsval (bsd inspired) */ |
| 305 | tc0->tsval_recent = 0; |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 306 | clib_warning ("paws failed - really old segment. REALLY?"); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 307 | } |
| 308 | else |
| 309 | { |
| 310 | /* Drop after ack if not rst */ |
| 311 | if (!tcp_rst (th0)) |
| 312 | { |
| 313 | tcp_make_ack (tc0, b0); |
| 314 | *next0 = tcp_next_output (tc0->c_is_ip4); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 315 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 316 | return -1; |
| 317 | } |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | /* 1st: check sequence number */ |
| 322 | if (!tcp_segment_in_rcv_wnd (tc0, vnet_buffer (b0)->tcp.seq_number, |
| 323 | vnet_buffer (b0)->tcp.seq_end)) |
| 324 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 325 | /* If our window is 0 and the packet is in sequence, let it pass |
| 326 | * through for ack processing. It should be dropped later.*/ |
| 327 | if (tc0->rcv_wnd == 0 |
| 328 | && tc0->rcv_nxt == vnet_buffer (b0)->tcp.seq_number) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 329 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 330 | /* TODO Should segment be tagged? */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 331 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 332 | else |
| 333 | { |
| 334 | /* If not RST, send dup ack */ |
| 335 | if (!tcp_rst (th0)) |
| 336 | { |
| 337 | tcp_make_ack (tc0, b0); |
| 338 | *next0 = tcp_next_output (tc0->c_is_ip4); |
| 339 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc0); |
| 340 | } |
| 341 | return -1; |
| 342 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 343 | } |
| 344 | |
| 345 | /* 2nd: check the RST bit */ |
| 346 | if (tcp_rst (th0)) |
| 347 | { |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 348 | tcp_connection_reset (tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 349 | return -1; |
| 350 | } |
| 351 | |
| 352 | /* 3rd: check security and precedence (skip) */ |
| 353 | |
| 354 | /* 4th: check the SYN bit */ |
| 355 | if (tcp_syn (th0)) |
| 356 | { |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 357 | /* TODO implement RFC 5961 */ |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 358 | if (tc0->state == TCP_STATE_SYN_RCVD) |
| 359 | { |
| 360 | tcp_make_synack (tc0, b0); |
| 361 | TCP_EVT_DBG (TCP_EVT_SYN_RCVD, tc0, 0); |
| 362 | } |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 363 | else |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 364 | { |
| 365 | tcp_make_ack (tc0, b0); |
| 366 | TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, tc0); |
| 367 | } |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 368 | *next0 = tcp_next_output (tc0->c_is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 369 | return -1; |
| 370 | } |
| 371 | |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 372 | /* If segment in window, save timestamp */ |
| 373 | tcp_update_timestamp (tc0, vnet_buffer (b0)->tcp.seq_number, |
| 374 | vnet_buffer (b0)->tcp.seq_end); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | always_inline int |
| 379 | tcp_rcv_ack_is_acceptable (tcp_connection_t * tc0, vlib_buffer_t * tb0) |
| 380 | { |
| 381 | /* SND.UNA =< SEG.ACK =< SND.NXT */ |
| 382 | return (seq_leq (tc0->snd_una, vnet_buffer (tb0)->tcp.ack_number) |
| 383 | && seq_leq (vnet_buffer (tb0)->tcp.ack_number, tc0->snd_nxt)); |
| 384 | } |
| 385 | |
| 386 | /** |
| 387 | * Compute smoothed RTT as per VJ's '88 SIGCOMM and RFC6298 |
| 388 | * |
| 389 | * Note that although the original article, srtt and rttvar are scaled |
| 390 | * to minimize round-off errors, here we don't. Instead, we rely on |
| 391 | * better precision time measurements. |
| 392 | * |
| 393 | * TODO support us rtt resolution |
| 394 | */ |
| 395 | static void |
| 396 | tcp_estimate_rtt (tcp_connection_t * tc, u32 mrtt) |
| 397 | { |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 398 | int err, diff; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 399 | |
| 400 | if (tc->srtt != 0) |
| 401 | { |
| 402 | err = mrtt - tc->srtt; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 403 | |
| 404 | /* XXX Drop in RTT results in RTTVAR increase and bigger RTO. |
| 405 | * The increase should be bound */ |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 406 | tc->srtt = clib_max ((int) tc->srtt + (err >> 3), 1); |
| 407 | diff = (clib_abs (err) - (int) tc->rttvar) >> 2; |
| 408 | tc->rttvar = clib_max ((int) tc->rttvar + diff, 1); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 409 | } |
| 410 | else |
| 411 | { |
| 412 | /* First measurement. */ |
| 413 | tc->srtt = mrtt; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 414 | tc->rttvar = mrtt >> 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 415 | } |
| 416 | } |
| 417 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 418 | void |
| 419 | tcp_update_rto (tcp_connection_t * tc) |
| 420 | { |
| 421 | tc->rto = clib_min (tc->srtt + (tc->rttvar << 2), TCP_RTO_MAX); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 422 | tc->rto = clib_max (tc->rto, TCP_RTO_MIN); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 423 | } |
| 424 | |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 425 | /** |
| 426 | * Update RTT estimate and RTO timer |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 427 | * |
| 428 | * Measure RTT: We have two sources of RTT measurements: TSOPT and ACK |
| 429 | * timing. Middle boxes are known to fiddle with TCP options so we |
| 430 | * should give higher priority to ACK timing. |
| 431 | * |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 432 | * This should be called only if previously sent bytes have been acked. |
| 433 | * |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 434 | * return 1 if valid rtt 0 otherwise |
| 435 | */ |
| 436 | static int |
| 437 | tcp_update_rtt (tcp_connection_t * tc, u32 ack) |
| 438 | { |
| 439 | u32 mrtt = 0; |
| 440 | |
| 441 | /* Karn's rule, part 1. Don't use retransmitted segments to estimate |
| 442 | * RTT because they're ambiguous. */ |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 443 | if (tcp_in_cong_recovery (tc) || tc->sack_sb.sacked_bytes) |
| 444 | goto done; |
| 445 | |
| 446 | if (tc->rtt_ts && seq_geq (ack, tc->rtt_seq)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 447 | { |
| 448 | mrtt = tcp_time_now () - tc->rtt_ts; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 449 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 450 | /* As per RFC7323 TSecr can be used for RTTM only if the segment advances |
| 451 | * snd_una, i.e., the left side of the send window: |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 452 | * seq_lt (tc->snd_una, ack). This is a condition for calling update_rtt */ |
| 453 | else if (tcp_opts_tstamp (&tc->rcv_opts) && tc->rcv_opts.tsecr) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 454 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 455 | mrtt = tcp_time_now () - tc->rcv_opts.tsecr; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 456 | } |
| 457 | |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 458 | /* Ignore dubious measurements */ |
| 459 | if (mrtt == 0 || mrtt > TCP_RTT_MAX) |
| 460 | goto done; |
| 461 | |
| 462 | tcp_estimate_rtt (tc, mrtt); |
| 463 | |
| 464 | done: |
| 465 | |
Florin Coras | 3af90fc | 2017-05-03 21:09:42 -0700 | [diff] [blame] | 466 | /* Allow measuring of a new RTT */ |
| 467 | tc->rtt_ts = 0; |
| 468 | |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 469 | /* If we got here something must've been ACKed so make sure boff is 0, |
| 470 | * even if mrrt is not valid since we update the rto lower */ |
| 471 | tc->rto_boff = 0; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 472 | tcp_update_rto (tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 473 | |
Florin Coras | 3af90fc | 2017-05-03 21:09:42 -0700 | [diff] [blame] | 474 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 475 | } |
| 476 | |
| 477 | /** |
| 478 | * Dequeue bytes that have been acked and while at it update RTT estimates. |
| 479 | */ |
| 480 | static void |
| 481 | tcp_dequeue_acked (tcp_connection_t * tc, u32 ack) |
| 482 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 483 | /* Dequeue the newly ACKed add SACKed bytes */ |
| 484 | stream_session_dequeue_drop (&tc->connection, |
| 485 | tc->bytes_acked + tc->sack_sb.snd_una_adv); |
| 486 | |
| 487 | tcp_validate_txf_size (tc, tc->snd_una_max - tc->snd_una); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 488 | |
| 489 | /* Update rtt and rto */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 490 | tcp_update_rtt (tc, ack); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 491 | |
| 492 | /* If everything has been acked, stop retransmit timer |
| 493 | * otherwise update. */ |
| 494 | tcp_retransmit_timer_update (tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 495 | } |
| 496 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 497 | /** |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 498 | * Check if duplicate ack as per RFC5681 Sec. 2 |
| 499 | */ |
| 500 | static u8 |
| 501 | tcp_ack_is_dupack (tcp_connection_t * tc, vlib_buffer_t * b, u32 prev_snd_wnd, |
| 502 | u32 prev_snd_una) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 503 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 504 | return ((vnet_buffer (b)->tcp.ack_number == prev_snd_una) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 505 | && seq_gt (tc->snd_una_max, tc->snd_una) |
| 506 | && (vnet_buffer (b)->tcp.seq_end == vnet_buffer (b)->tcp.seq_number) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 507 | && (prev_snd_wnd == tc->snd_wnd)); |
| 508 | } |
| 509 | |
| 510 | /** |
| 511 | * Checks if ack is a congestion control event. |
| 512 | */ |
| 513 | static u8 |
| 514 | tcp_ack_is_cc_event (tcp_connection_t * tc, vlib_buffer_t * b, |
| 515 | u32 prev_snd_wnd, u32 prev_snd_una, u8 * is_dack) |
| 516 | { |
| 517 | /* Check if ack is duplicate. Per RFC 6675, ACKs that SACK new data are |
| 518 | * defined to be 'duplicate' */ |
| 519 | *is_dack = tc->sack_sb.last_sacked_bytes |
| 520 | || tcp_ack_is_dupack (tc, b, prev_snd_wnd, prev_snd_una); |
| 521 | |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 522 | return ((*is_dack || tcp_in_cong_recovery (tc)) && !tcp_is_lost_fin (tc)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 523 | } |
| 524 | |
| 525 | void |
| 526 | scoreboard_remove_hole (sack_scoreboard_t * sb, sack_scoreboard_hole_t * hole) |
| 527 | { |
| 528 | sack_scoreboard_hole_t *next, *prev; |
| 529 | |
| 530 | if (hole->next != TCP_INVALID_SACK_HOLE_INDEX) |
| 531 | { |
| 532 | next = pool_elt_at_index (sb->holes, hole->next); |
| 533 | next->prev = hole->prev; |
| 534 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 535 | else |
| 536 | { |
| 537 | sb->tail = hole->prev; |
| 538 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 539 | |
| 540 | if (hole->prev != TCP_INVALID_SACK_HOLE_INDEX) |
| 541 | { |
| 542 | prev = pool_elt_at_index (sb->holes, hole->prev); |
| 543 | prev->next = hole->next; |
| 544 | } |
| 545 | else |
| 546 | { |
| 547 | sb->head = hole->next; |
| 548 | } |
| 549 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 550 | if (scoreboard_hole_index (sb, hole) == sb->cur_rxt_hole) |
| 551 | sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; |
| 552 | |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 553 | /* Poison the entry */ |
| 554 | if (CLIB_DEBUG > 0) |
| 555 | memset (hole, 0xfe, sizeof (*hole)); |
| 556 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 557 | pool_put (sb->holes, hole); |
| 558 | } |
| 559 | |
| 560 | sack_scoreboard_hole_t * |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 561 | scoreboard_insert_hole (sack_scoreboard_t * sb, u32 prev_index, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 562 | u32 start, u32 end) |
| 563 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 564 | sack_scoreboard_hole_t *hole, *next, *prev; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 565 | u32 hole_index; |
| 566 | |
| 567 | pool_get (sb->holes, hole); |
| 568 | memset (hole, 0, sizeof (*hole)); |
| 569 | |
| 570 | hole->start = start; |
| 571 | hole->end = end; |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 572 | hole_index = scoreboard_hole_index (sb, hole); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 573 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 574 | prev = scoreboard_get_hole (sb, prev_index); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 575 | if (prev) |
| 576 | { |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 577 | hole->prev = prev_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 578 | hole->next = prev->next; |
| 579 | |
| 580 | if ((next = scoreboard_next_hole (sb, hole))) |
| 581 | next->prev = hole_index; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 582 | else |
| 583 | sb->tail = hole_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 584 | |
| 585 | prev->next = hole_index; |
| 586 | } |
| 587 | else |
| 588 | { |
| 589 | sb->head = hole_index; |
| 590 | hole->prev = TCP_INVALID_SACK_HOLE_INDEX; |
| 591 | hole->next = TCP_INVALID_SACK_HOLE_INDEX; |
| 592 | } |
| 593 | |
| 594 | return hole; |
| 595 | } |
| 596 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 597 | void |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 598 | scoreboard_update_bytes (tcp_connection_t * tc, sack_scoreboard_t * sb) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 599 | { |
| 600 | sack_scoreboard_hole_t *hole, *prev; |
| 601 | u32 bytes = 0, blks = 0; |
| 602 | |
| 603 | sb->lost_bytes = 0; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 604 | sb->sacked_bytes = 0; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 605 | hole = scoreboard_last_hole (sb); |
| 606 | if (!hole) |
| 607 | return; |
| 608 | |
| 609 | if (seq_gt (sb->high_sacked, hole->end)) |
| 610 | { |
| 611 | bytes = sb->high_sacked - hole->end; |
| 612 | blks = 1; |
| 613 | } |
| 614 | |
| 615 | while ((prev = scoreboard_prev_hole (sb, hole)) |
| 616 | && (bytes < (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss |
| 617 | && blks < TCP_DUPACK_THRESHOLD)) |
| 618 | { |
| 619 | bytes += hole->start - prev->end; |
| 620 | blks++; |
| 621 | hole = prev; |
| 622 | } |
| 623 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 624 | while (hole) |
| 625 | { |
| 626 | sb->lost_bytes += scoreboard_hole_bytes (hole); |
| 627 | hole->is_lost = 1; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 628 | prev = hole; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 629 | hole = scoreboard_prev_hole (sb, hole); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 630 | if (hole) |
| 631 | bytes += prev->start - hole->end; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 632 | } |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 633 | sb->sacked_bytes = bytes; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 634 | } |
| 635 | |
| 636 | /** |
| 637 | * Figure out the next hole to retransmit |
| 638 | * |
| 639 | * Follows logic proposed in RFC6675 Sec. 4, NextSeg() |
| 640 | */ |
| 641 | sack_scoreboard_hole_t * |
| 642 | scoreboard_next_rxt_hole (sack_scoreboard_t * sb, |
| 643 | sack_scoreboard_hole_t * start, |
| 644 | u8 have_sent_1_smss, |
| 645 | u8 * can_rescue, u8 * snd_limited) |
| 646 | { |
| 647 | sack_scoreboard_hole_t *hole = 0; |
| 648 | |
| 649 | hole = start ? start : scoreboard_first_hole (sb); |
| 650 | while (hole && seq_leq (hole->end, sb->high_rxt) && hole->is_lost) |
| 651 | hole = scoreboard_next_hole (sb, hole); |
| 652 | |
| 653 | /* Nothing, return */ |
| 654 | if (!hole) |
| 655 | { |
| 656 | sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; |
| 657 | return 0; |
| 658 | } |
| 659 | |
| 660 | /* Rule (1): if higher than rxt, less than high_sacked and lost */ |
| 661 | if (hole->is_lost && seq_lt (hole->start, sb->high_sacked)) |
| 662 | { |
| 663 | sb->cur_rxt_hole = scoreboard_hole_index (sb, hole); |
| 664 | } |
| 665 | else |
| 666 | { |
| 667 | /* Rule (2): output takes care of transmitting new data */ |
| 668 | if (!have_sent_1_smss) |
| 669 | { |
| 670 | hole = 0; |
| 671 | sb->cur_rxt_hole = TCP_INVALID_SACK_HOLE_INDEX; |
| 672 | } |
| 673 | /* Rule (3): if hole not lost */ |
| 674 | else if (seq_lt (hole->start, sb->high_sacked)) |
| 675 | { |
| 676 | *snd_limited = 1; |
| 677 | sb->cur_rxt_hole = scoreboard_hole_index (sb, hole); |
| 678 | } |
| 679 | /* Rule (4): if hole beyond high_sacked */ |
| 680 | else |
| 681 | { |
| 682 | ASSERT (seq_geq (hole->start, sb->high_sacked)); |
| 683 | *snd_limited = 1; |
| 684 | *can_rescue = 1; |
| 685 | /* HighRxt MUST NOT be updated */ |
| 686 | return 0; |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | if (hole && seq_lt (sb->high_rxt, hole->start)) |
| 691 | sb->high_rxt = hole->start; |
| 692 | |
| 693 | return hole; |
| 694 | } |
| 695 | |
| 696 | void |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 697 | scoreboard_init_high_rxt (sack_scoreboard_t * sb, u32 seq) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 698 | { |
| 699 | sack_scoreboard_hole_t *hole; |
| 700 | hole = scoreboard_first_hole (sb); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 701 | if (hole) |
| 702 | { |
| 703 | seq = seq_gt (seq, hole->start) ? seq : hole->start; |
| 704 | sb->cur_rxt_hole = sb->head; |
| 705 | } |
| 706 | sb->high_rxt = seq; |
| 707 | } |
| 708 | |
| 709 | /** |
| 710 | * Test that scoreboard is sane after recovery |
| 711 | * |
| 712 | * Returns 1 if scoreboard is empty or if first hole beyond |
| 713 | * snd_una. |
| 714 | */ |
| 715 | u8 |
| 716 | tcp_scoreboard_is_sane_post_recovery (tcp_connection_t * tc) |
| 717 | { |
| 718 | sack_scoreboard_hole_t *hole; |
| 719 | hole = scoreboard_first_hole (&tc->sack_sb); |
| 720 | return (!hole || seq_geq (hole->start, tc->snd_una)); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 721 | } |
| 722 | |
| 723 | void |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 724 | tcp_rcv_sacks (tcp_connection_t * tc, u32 ack) |
| 725 | { |
| 726 | sack_scoreboard_t *sb = &tc->sack_sb; |
| 727 | sack_block_t *blk, tmp; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 728 | sack_scoreboard_hole_t *hole, *next_hole, *last_hole; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 729 | u32 blk_index = 0, old_sacked_bytes, hole_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 730 | int i, j; |
| 731 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 732 | sb->last_sacked_bytes = 0; |
| 733 | sb->snd_una_adv = 0; |
| 734 | old_sacked_bytes = sb->sacked_bytes; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 735 | sb->last_bytes_delivered = 0; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 736 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 737 | if (!tcp_opts_sack (&tc->rcv_opts) |
| 738 | && sb->head == TCP_INVALID_SACK_HOLE_INDEX) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 739 | return; |
| 740 | |
| 741 | /* Remove invalid blocks */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 742 | blk = tc->rcv_opts.sacks; |
| 743 | while (blk < vec_end (tc->rcv_opts.sacks)) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 744 | { |
| 745 | if (seq_lt (blk->start, blk->end) |
| 746 | && seq_gt (blk->start, tc->snd_una) |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 747 | && seq_gt (blk->start, ack) && seq_leq (blk->end, tc->snd_una_max)) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 748 | { |
| 749 | blk++; |
| 750 | continue; |
| 751 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 752 | vec_del1 (tc->rcv_opts.sacks, blk - tc->rcv_opts.sacks); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 753 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 754 | |
| 755 | /* Add block for cumulative ack */ |
| 756 | if (seq_gt (ack, tc->snd_una)) |
| 757 | { |
| 758 | tmp.start = tc->snd_una; |
| 759 | tmp.end = ack; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 760 | vec_add1 (tc->rcv_opts.sacks, tmp); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 761 | } |
| 762 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 763 | if (vec_len (tc->rcv_opts.sacks) == 0) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 764 | return; |
| 765 | |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 766 | tcp_scoreboard_trace_add (tc, ack); |
| 767 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 768 | /* Make sure blocks are ordered */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 769 | for (i = 0; i < vec_len (tc->rcv_opts.sacks); i++) |
| 770 | for (j = i + 1; j < vec_len (tc->rcv_opts.sacks); j++) |
| 771 | if (seq_lt (tc->rcv_opts.sacks[j].start, tc->rcv_opts.sacks[i].start)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 772 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 773 | tmp = tc->rcv_opts.sacks[i]; |
| 774 | tc->rcv_opts.sacks[i] = tc->rcv_opts.sacks[j]; |
| 775 | tc->rcv_opts.sacks[j] = tmp; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 776 | } |
| 777 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 778 | if (sb->head == TCP_INVALID_SACK_HOLE_INDEX) |
| 779 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 780 | /* If no holes, insert the first that covers all outstanding bytes */ |
| 781 | last_hole = scoreboard_insert_hole (sb, TCP_INVALID_SACK_HOLE_INDEX, |
| 782 | tc->snd_una, tc->snd_una_max); |
| 783 | sb->tail = scoreboard_hole_index (sb, last_hole); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 784 | tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1]; |
| 785 | sb->high_sacked = tmp.end; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 786 | } |
| 787 | else |
| 788 | { |
| 789 | /* If we have holes but snd_una_max is beyond the last hole, update |
| 790 | * last hole end */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 791 | tmp = tc->rcv_opts.sacks[vec_len (tc->rcv_opts.sacks) - 1]; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 792 | last_hole = scoreboard_last_hole (sb); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 793 | if (seq_gt (tc->snd_una_max, last_hole->end)) |
| 794 | { |
| 795 | if (seq_geq (last_hole->start, sb->high_sacked)) |
| 796 | { |
| 797 | last_hole->end = tc->snd_una_max; |
| 798 | } |
| 799 | /* New hole after high sacked block */ |
| 800 | else if (seq_lt (sb->high_sacked, tc->snd_una_max)) |
| 801 | { |
| 802 | scoreboard_insert_hole (sb, sb->tail, sb->high_sacked, |
| 803 | tc->snd_una_max); |
| 804 | } |
| 805 | } |
| 806 | /* Keep track of max byte sacked for when the last hole |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 807 | * is acked */ |
| 808 | if (seq_gt (tmp.end, sb->high_sacked)) |
| 809 | sb->high_sacked = tmp.end; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 810 | } |
| 811 | |
| 812 | /* Walk the holes with the SACK blocks */ |
| 813 | hole = pool_elt_at_index (sb->holes, sb->head); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 814 | while (hole && blk_index < vec_len (tc->rcv_opts.sacks)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 815 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 816 | blk = &tc->rcv_opts.sacks[blk_index]; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 817 | if (seq_leq (blk->start, hole->start)) |
| 818 | { |
| 819 | /* Block covers hole. Remove hole */ |
| 820 | if (seq_geq (blk->end, hole->end)) |
| 821 | { |
| 822 | next_hole = scoreboard_next_hole (sb, hole); |
| 823 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 824 | /* Byte accounting: snd_una needs to be advanced */ |
| 825 | if (blk->end == ack) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 826 | { |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 827 | if (next_hole) |
Florin Coras | 06d1101 | 2017-05-17 14:21:51 -0700 | [diff] [blame] | 828 | { |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 829 | if (seq_lt (ack, next_hole->start)) |
| 830 | sb->snd_una_adv = next_hole->start - ack; |
| 831 | sb->last_bytes_delivered += |
| 832 | next_hole->start - hole->end; |
Florin Coras | 06d1101 | 2017-05-17 14:21:51 -0700 | [diff] [blame] | 833 | } |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 834 | else |
Florin Coras | 06d1101 | 2017-05-17 14:21:51 -0700 | [diff] [blame] | 835 | { |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 836 | ASSERT (seq_geq (sb->high_sacked, ack)); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 837 | sb->snd_una_adv = sb->high_sacked - ack; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 838 | sb->last_bytes_delivered += sb->high_sacked - hole->end; |
Florin Coras | 06d1101 | 2017-05-17 14:21:51 -0700 | [diff] [blame] | 839 | } |
| 840 | } |
| 841 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 842 | scoreboard_remove_hole (sb, hole); |
| 843 | hole = next_hole; |
| 844 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 845 | /* Partial 'head' overlap */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 846 | else |
| 847 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 848 | if (seq_gt (blk->end, hole->start)) |
| 849 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 850 | hole->start = blk->end; |
| 851 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 852 | blk_index++; |
| 853 | } |
| 854 | } |
| 855 | else |
| 856 | { |
| 857 | /* Hole must be split */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 858 | if (seq_lt (blk->end, hole->end)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 859 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 860 | hole_index = scoreboard_hole_index (sb, hole); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 861 | next_hole = scoreboard_insert_hole (sb, hole_index, blk->end, |
| 862 | hole->end); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 863 | |
| 864 | /* Pool might've moved */ |
| 865 | hole = scoreboard_get_hole (sb, hole_index); |
| 866 | hole->end = blk->start; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 867 | blk_index++; |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 868 | ASSERT (hole->next == scoreboard_hole_index (sb, next_hole)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 869 | } |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 870 | else if (seq_lt (blk->start, hole->end)) |
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 | hole->end = blk->start; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 873 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 874 | hole = scoreboard_next_hole (sb, hole); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 875 | } |
| 876 | } |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 877 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 878 | scoreboard_update_bytes (tc, sb); |
| 879 | sb->last_sacked_bytes = sb->sacked_bytes |
| 880 | - (old_sacked_bytes - sb->last_bytes_delivered); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 881 | ASSERT (sb->last_sacked_bytes <= sb->sacked_bytes); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 882 | ASSERT (sb->sacked_bytes == 0 |
| 883 | || sb->sacked_bytes < tc->snd_una_max - seq_max (tc->snd_una, ack)); |
| 884 | ASSERT (sb->last_sacked_bytes + sb->lost_bytes <= tc->snd_una_max |
| 885 | - seq_max (tc->snd_una, ack)); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 886 | ASSERT (sb->head == TCP_INVALID_SACK_HOLE_INDEX || tcp_in_recovery (tc) |
| 887 | || sb->holes[sb->head].start == ack + sb->snd_una_adv); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 888 | } |
| 889 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 890 | /** |
| 891 | * Try to update snd_wnd based on feedback received from peer. |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 892 | * |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 893 | * If successful, and new window is 'effectively' 0, activate persist |
| 894 | * timer. |
| 895 | */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 896 | static void |
| 897 | tcp_update_snd_wnd (tcp_connection_t * tc, u32 seq, u32 ack, u32 snd_wnd) |
| 898 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 899 | /* If (SND.WL1 < SEG.SEQ or (SND.WL1 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set |
| 900 | * SND.WND <- SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- SEG.ACK */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 901 | if (seq_lt (tc->snd_wl1, seq) |
| 902 | || (tc->snd_wl1 == seq && seq_leq (tc->snd_wl2, ack))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 903 | { |
| 904 | tc->snd_wnd = snd_wnd; |
| 905 | tc->snd_wl1 = seq; |
| 906 | tc->snd_wl2 = ack; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 907 | TCP_EVT_DBG (TCP_EVT_SND_WND, tc); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 908 | |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 909 | if (tc->snd_wnd < tc->snd_mss) |
| 910 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 911 | /* Set persist timer if not set and we just got 0 wnd */ |
| 912 | if (!tcp_timer_is_active (tc, TCP_TIMER_PERSIST) |
| 913 | && !tcp_timer_is_active (tc, TCP_TIMER_RETRANSMIT)) |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 914 | tcp_persist_timer_set (tc); |
| 915 | } |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 916 | else |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 917 | { |
| 918 | tcp_persist_timer_reset (tc); |
| 919 | if (!tcp_in_recovery (tc) && tc->rto_boff > 0) |
| 920 | { |
| 921 | tc->rto_boff = 0; |
| 922 | tcp_update_rto (tc); |
| 923 | } |
| 924 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 925 | } |
| 926 | } |
| 927 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 928 | void |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 929 | tcp_cc_init_congestion (tcp_connection_t * tc) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 930 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 931 | tcp_fastrecovery_on (tc); |
| 932 | tc->snd_congestion = tc->snd_una_max; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 933 | tc->cc_algo->congestion (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 934 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 935 | } |
| 936 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 937 | static void |
| 938 | tcp_cc_recovery_exit (tcp_connection_t * tc) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 939 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 940 | /* Deflate rto */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 941 | tc->rto_boff = 0; |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 942 | tcp_update_rto (tc); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 943 | tc->snd_rxt_ts = 0; |
Florin Coras | 84275e9 | 2017-09-26 12:30:40 -0400 | [diff] [blame] | 944 | tc->snd_nxt = tc->snd_una_max; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 945 | tcp_recovery_off (tc); |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 946 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 947 | } |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 948 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 949 | void |
| 950 | tcp_cc_fastrecovery_exit (tcp_connection_t * tc) |
| 951 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 952 | tc->cc_algo->recovered (tc); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 953 | tc->snd_rxt_bytes = 0; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 954 | tc->rcv_dupacks = 0; |
Florin Coras | 84275e9 | 2017-09-26 12:30:40 -0400 | [diff] [blame] | 955 | tc->snd_nxt = tc->snd_una_max; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 956 | tcp_fastrecovery_off (tc); |
| 957 | tcp_fastrecovery_1_smss_off (tc); |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 958 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 3); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 959 | } |
| 960 | |
| 961 | static void |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 962 | tcp_cc_congestion_undo (tcp_connection_t * tc) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 963 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 964 | tc->cwnd = tc->prev_cwnd; |
| 965 | tc->ssthresh = tc->prev_ssthresh; |
| 966 | tc->snd_nxt = tc->snd_una_max; |
| 967 | tc->rcv_dupacks = 0; |
| 968 | if (tcp_in_recovery (tc)) |
| 969 | tcp_cc_recovery_exit (tc); |
| 970 | ASSERT (tc->rto_boff == 0); |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 971 | TCP_EVT_DBG (TCP_EVT_CC_EVT, tc, 5); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 972 | /* TODO extend for fastrecovery */ |
| 973 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 974 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 975 | static u8 |
| 976 | tcp_cc_is_spurious_retransmit (tcp_connection_t * tc) |
| 977 | { |
Florin Coras | f1762d6 | 2017-09-24 19:43:08 -0400 | [diff] [blame] | 978 | return (tcp_in_recovery (tc) && tc->rto_boff == 1 |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 979 | && tc->snd_rxt_ts |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 980 | && tcp_opts_tstamp (&tc->rcv_opts) |
| 981 | && timestamp_lt (tc->rcv_opts.tsecr, tc->snd_rxt_ts)); |
| 982 | } |
| 983 | |
| 984 | int |
| 985 | tcp_cc_recover (tcp_connection_t * tc) |
| 986 | { |
| 987 | ASSERT (tcp_in_cong_recovery (tc)); |
| 988 | if (tcp_cc_is_spurious_retransmit (tc)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 989 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 990 | tcp_cc_congestion_undo (tc); |
| 991 | return 1; |
| 992 | } |
| 993 | |
| 994 | if (tcp_in_recovery (tc)) |
| 995 | tcp_cc_recovery_exit (tc); |
| 996 | else if (tcp_in_fastrecovery (tc)) |
| 997 | tcp_cc_fastrecovery_exit (tc); |
| 998 | |
| 999 | ASSERT (tc->rto_boff == 0); |
| 1000 | ASSERT (!tcp_in_cong_recovery (tc)); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1001 | ASSERT (tcp_scoreboard_is_sane_post_recovery (tc)); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1002 | return 0; |
| 1003 | } |
| 1004 | |
| 1005 | static void |
| 1006 | tcp_cc_update (tcp_connection_t * tc, vlib_buffer_t * b) |
| 1007 | { |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1008 | ASSERT (!tcp_in_cong_recovery (tc) || tcp_is_lost_fin (tc)); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1009 | |
| 1010 | /* Congestion avoidance */ |
| 1011 | tc->cc_algo->rcv_ack (tc); |
| 1012 | tc->tsecr_last_ack = tc->rcv_opts.tsecr; |
| 1013 | |
| 1014 | /* If a cumulative ack, make sure dupacks is 0 */ |
| 1015 | tc->rcv_dupacks = 0; |
| 1016 | |
| 1017 | /* When dupacks hits the threshold we only enter fast retransmit if |
| 1018 | * cumulative ack covers more than snd_congestion. Should snd_una |
| 1019 | * wrap this test may fail under otherwise valid circumstances. |
| 1020 | * Therefore, proactively update snd_congestion when wrap detected. */ |
| 1021 | if (PREDICT_FALSE |
| 1022 | (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked) |
| 1023 | && seq_gt (tc->snd_congestion, tc->snd_una))) |
| 1024 | tc->snd_congestion = tc->snd_una - 1; |
| 1025 | } |
| 1026 | |
| 1027 | static u8 |
| 1028 | tcp_should_fastrecover_sack (tcp_connection_t * tc) |
| 1029 | { |
| 1030 | return (TCP_DUPACK_THRESHOLD - 1) * tc->snd_mss < tc->sack_sb.sacked_bytes; |
| 1031 | } |
| 1032 | |
| 1033 | static u8 |
| 1034 | tcp_should_fastrecover (tcp_connection_t * tc) |
| 1035 | { |
| 1036 | return (tc->rcv_dupacks == TCP_DUPACK_THRESHOLD |
| 1037 | || tcp_should_fastrecover_sack (tc)); |
| 1038 | } |
| 1039 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1040 | /** |
| 1041 | * One function to rule them all ... and in the darkness bind them |
| 1042 | */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1043 | static void |
| 1044 | tcp_cc_handle_event (tcp_connection_t * tc, u32 is_dack) |
| 1045 | { |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1046 | u32 rxt_delivered; |
| 1047 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1048 | /* |
| 1049 | * Duplicate ACK. Check if we should enter fast recovery, or if already in |
| 1050 | * it account for the bytes that left the network. |
| 1051 | */ |
| 1052 | if (is_dack) |
| 1053 | { |
| 1054 | ASSERT (tc->snd_una != tc->snd_una_max |
| 1055 | || tc->sack_sb.last_sacked_bytes); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1056 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1057 | tc->rcv_dupacks++; |
| 1058 | |
| 1059 | if (tc->rcv_dupacks > TCP_DUPACK_THRESHOLD && !tc->bytes_acked) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1060 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1061 | ASSERT (tcp_in_fastrecovery (tc)); |
| 1062 | /* Pure duplicate ack. If some data got acked, it's handled lower */ |
| 1063 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); |
| 1064 | return; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1065 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1066 | else if (tcp_should_fastrecover (tc)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1067 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1068 | /* Things are already bad */ |
| 1069 | if (tcp_in_cong_recovery (tc)) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1070 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1071 | tc->rcv_dupacks = 0; |
| 1072 | goto partial_ack_test; |
| 1073 | } |
| 1074 | |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1075 | /* If of of the two conditions lower hold, reset dupacks because |
| 1076 | * we're probably after timeout (RFC6582 heuristics). |
| 1077 | * If Cumulative ack does not cover more than congestion threshold, |
| 1078 | * and: |
| 1079 | * 1) The following doesn't hold: The congestion window is greater |
| 1080 | * than SMSS bytes and the difference between highest_ack |
| 1081 | * and prev_highest_ack is at most 4*SMSS bytes |
| 1082 | * 2) Echoed timestamp in the last non-dup ack does not equal the |
| 1083 | * stored timestamp |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1084 | */ |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1085 | if (seq_leq (tc->snd_una, tc->snd_congestion) |
| 1086 | && ((!(tc->cwnd > tc->snd_mss |
| 1087 | && tc->bytes_acked <= 4 * tc->snd_mss)) |
| 1088 | || (tc->rcv_opts.tsecr != tc->tsecr_last_ack))) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1089 | { |
| 1090 | tc->rcv_dupacks = 0; |
| 1091 | return; |
| 1092 | } |
| 1093 | |
| 1094 | tcp_cc_init_congestion (tc); |
| 1095 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); |
| 1096 | |
| 1097 | /* The first segment MUST be retransmitted */ |
| 1098 | tcp_retransmit_first_unacked (tc); |
| 1099 | |
| 1100 | /* Post retransmit update cwnd to ssthresh and account for the |
| 1101 | * three segments that have left the network and should've been |
| 1102 | * buffered at the receiver XXX */ |
| 1103 | tc->cwnd = tc->ssthresh + tc->rcv_dupacks * tc->snd_mss; |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1104 | ASSERT (tc->cwnd >= tc->snd_mss); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1105 | |
| 1106 | /* If cwnd allows, send more data */ |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1107 | if (tcp_opts_sack_permitted (&tc->rcv_opts)) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1108 | { |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1109 | scoreboard_init_high_rxt (&tc->sack_sb, |
| 1110 | tc->snd_una + tc->snd_mss); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1111 | tcp_fast_retransmit_sack (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1112 | } |
| 1113 | else |
| 1114 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1115 | tcp_fast_retransmit_no_sack (tc); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1116 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1117 | |
| 1118 | return; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1119 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1120 | else if (!tc->bytes_acked |
| 1121 | || (tc->bytes_acked && !tcp_in_cong_recovery (tc))) |
| 1122 | { |
| 1123 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_DUPACK); |
| 1124 | return; |
| 1125 | } |
| 1126 | else |
| 1127 | goto partial_ack; |
| 1128 | } |
| 1129 | |
| 1130 | partial_ack_test: |
| 1131 | |
| 1132 | if (!tc->bytes_acked) |
| 1133 | return; |
| 1134 | |
| 1135 | partial_ack: |
| 1136 | /* |
| 1137 | * Legitimate ACK. 1) See if we can exit recovery |
| 1138 | */ |
| 1139 | /* XXX limit this only to first partial ack? */ |
| 1140 | tcp_retransmit_timer_update (tc); |
| 1141 | |
| 1142 | if (seq_geq (tc->snd_una, tc->snd_congestion)) |
| 1143 | { |
| 1144 | /* If spurious return, we've already updated everything */ |
| 1145 | if (tcp_cc_recover (tc)) |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1146 | { |
| 1147 | tc->tsecr_last_ack = tc->rcv_opts.tsecr; |
| 1148 | return; |
| 1149 | } |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1150 | |
| 1151 | tc->snd_nxt = tc->snd_una_max; |
| 1152 | |
| 1153 | /* Treat as congestion avoidance ack */ |
| 1154 | tc->cc_algo->rcv_ack (tc); |
| 1155 | tc->tsecr_last_ack = tc->rcv_opts.tsecr; |
| 1156 | return; |
| 1157 | } |
| 1158 | |
| 1159 | /* |
| 1160 | * Legitimate ACK. 2) If PARTIAL ACK try to retransmit |
| 1161 | */ |
| 1162 | TCP_EVT_DBG (TCP_EVT_CC_PACK, tc); |
| 1163 | |
| 1164 | /* RFC6675: If the incoming ACK is a cumulative acknowledgment, |
| 1165 | * reset dupacks to 0 */ |
| 1166 | tc->rcv_dupacks = 0; |
| 1167 | |
| 1168 | tcp_retransmit_first_unacked (tc); |
| 1169 | |
| 1170 | /* Post RTO timeout don't try anything fancy */ |
| 1171 | if (tcp_in_recovery (tc)) |
| 1172 | return; |
| 1173 | |
| 1174 | /* Remove retransmitted bytes that have been delivered */ |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1175 | ASSERT (tc->bytes_acked + tc->sack_sb.snd_una_adv |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1176 | >= tc->sack_sb.last_bytes_delivered |
| 1177 | || (tc->flags & TCP_CONN_FINSNT)); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1178 | |
| 1179 | if (seq_lt (tc->snd_una, tc->sack_sb.high_rxt)) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1180 | { |
| 1181 | /* If we have sacks and we haven't gotten an ack beyond high_rxt, |
| 1182 | * remove sacked bytes delivered */ |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1183 | rxt_delivered = tc->bytes_acked + tc->sack_sb.snd_una_adv |
| 1184 | - tc->sack_sb.last_bytes_delivered; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1185 | ASSERT (tc->snd_rxt_bytes >= rxt_delivered); |
| 1186 | tc->snd_rxt_bytes -= rxt_delivered; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1187 | } |
| 1188 | else |
| 1189 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1190 | /* Either all retransmitted holes have been acked, or we're |
| 1191 | * "in the blind" and retransmitting segment by segment */ |
| 1192 | tc->snd_rxt_bytes = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1193 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1194 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1195 | tc->cc_algo->rcv_cong_ack (tc, TCP_CC_PARTIALACK); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1196 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1197 | /* |
| 1198 | * Since this was a partial ack, try to retransmit some more data |
| 1199 | */ |
| 1200 | tcp_fast_retransmit (tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1201 | } |
| 1202 | |
| 1203 | void |
| 1204 | tcp_cc_init (tcp_connection_t * tc) |
| 1205 | { |
| 1206 | tc->cc_algo = tcp_cc_algo_get (TCP_CC_NEWRENO); |
| 1207 | tc->cc_algo->init (tc); |
| 1208 | } |
| 1209 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1210 | /** |
| 1211 | * Process incoming ACK |
| 1212 | */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1213 | static int |
| 1214 | tcp_rcv_ack (tcp_connection_t * tc, vlib_buffer_t * b, |
| 1215 | tcp_header_t * th, u32 * next, u32 * error) |
| 1216 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1217 | u32 prev_snd_wnd, prev_snd_una; |
| 1218 | u8 is_dack; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1219 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1220 | TCP_EVT_DBG (TCP_EVT_CC_STAT, tc); |
| 1221 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1222 | /* If the ACK acks something not yet sent (SEG.ACK > SND.NXT) */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1223 | if (PREDICT_FALSE (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_nxt))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1224 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1225 | /* If we have outstanding data and this is within the window, accept it, |
| 1226 | * probably retransmit has timed out. Otherwise ACK segment and then |
| 1227 | * drop it */ |
| 1228 | if (seq_gt (vnet_buffer (b)->tcp.ack_number, tc->snd_una_max)) |
| 1229 | { |
| 1230 | tcp_make_ack (tc, b); |
| 1231 | *next = tcp_next_output (tc->c_is_ip4); |
| 1232 | *error = TCP_ERROR_ACK_INVALID; |
| 1233 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 0, |
| 1234 | vnet_buffer (b)->tcp.ack_number); |
| 1235 | return -1; |
| 1236 | } |
| 1237 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1238 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 2, |
| 1239 | vnet_buffer (b)->tcp.ack_number); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1240 | |
| 1241 | tc->snd_nxt = vnet_buffer (b)->tcp.ack_number; |
| 1242 | *error = TCP_ERROR_ACK_FUTURE; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1243 | } |
| 1244 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1245 | /* If old ACK, probably it's an old dupack */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1246 | if (PREDICT_FALSE (seq_lt (vnet_buffer (b)->tcp.ack_number, tc->snd_una))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1247 | { |
| 1248 | *error = TCP_ERROR_ACK_OLD; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1249 | TCP_EVT_DBG (TCP_EVT_ACK_RCV_ERR, tc, 1, |
| 1250 | vnet_buffer (b)->tcp.ack_number); |
| 1251 | if (tcp_in_fastrecovery (tc) && tc->rcv_dupacks == TCP_DUPACK_THRESHOLD) |
| 1252 | { |
| 1253 | TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1254 | tcp_cc_handle_event (tc, 1); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1255 | } |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 1256 | /* Don't drop yet */ |
| 1257 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1258 | } |
| 1259 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1260 | /* |
| 1261 | * Looks okay, process feedback |
| 1262 | */ |
| 1263 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1264 | if (tcp_opts_sack_permitted (&tc->rcv_opts)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1265 | tcp_rcv_sacks (tc, vnet_buffer (b)->tcp.ack_number); |
| 1266 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1267 | prev_snd_wnd = tc->snd_wnd; |
| 1268 | prev_snd_una = tc->snd_una; |
| 1269 | tcp_update_snd_wnd (tc, vnet_buffer (b)->tcp.seq_number, |
| 1270 | vnet_buffer (b)->tcp.ack_number, |
| 1271 | clib_net_to_host_u16 (th->window) << tc->snd_wscale); |
| 1272 | tc->bytes_acked = vnet_buffer (b)->tcp.ack_number - tc->snd_una; |
| 1273 | tc->snd_una = vnet_buffer (b)->tcp.ack_number + tc->sack_sb.snd_una_adv; |
| 1274 | tcp_validate_txf_size (tc, tc->bytes_acked); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1275 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1276 | if (tc->bytes_acked) |
| 1277 | tcp_dequeue_acked (tc, vnet_buffer (b)->tcp.ack_number); |
| 1278 | |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1279 | TCP_EVT_DBG (TCP_EVT_ACK_RCVD, tc); |
| 1280 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1281 | /* |
| 1282 | * Check if we have congestion event |
| 1283 | */ |
| 1284 | |
| 1285 | if (tcp_ack_is_cc_event (tc, b, prev_snd_wnd, prev_snd_una, &is_dack)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1286 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1287 | tcp_cc_handle_event (tc, is_dack); |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1288 | if (!tcp_in_cong_recovery (tc)) |
| 1289 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1290 | *error = TCP_ERROR_ACK_DUP; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1291 | TCP_EVT_DBG (TCP_EVT_DUPACK_RCVD, tc, 1); |
| 1292 | return vnet_buffer (b)->tcp.data_len ? 0 : -1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1293 | } |
| 1294 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1295 | /* |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1296 | * Update congestion control (slow start/congestion avoidance) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1297 | */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1298 | tcp_cc_update (tc, b); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1299 | |
| 1300 | return 0; |
| 1301 | } |
| 1302 | |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1303 | static u8 |
| 1304 | tcp_sack_vector_is_sane (sack_block_t * sacks) |
| 1305 | { |
| 1306 | int i; |
| 1307 | for (i = 1; i < vec_len (sacks); i++) |
| 1308 | { |
| 1309 | if (sacks[i - 1].end == sacks[i].start) |
| 1310 | return 0; |
| 1311 | } |
| 1312 | return 1; |
| 1313 | } |
| 1314 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1315 | /** |
| 1316 | * Build SACK list as per RFC2018. |
| 1317 | * |
| 1318 | * Makes sure the first block contains the segment that generated the current |
| 1319 | * ACK and the following ones are the ones most recently reported in SACK |
| 1320 | * blocks. |
| 1321 | * |
| 1322 | * @param tc TCP connection for which the SACK list is updated |
| 1323 | * @param start Start sequence number of the newest SACK block |
| 1324 | * @param end End sequence of the newest SACK block |
| 1325 | */ |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1326 | void |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1327 | tcp_update_sack_list (tcp_connection_t * tc, u32 start, u32 end) |
| 1328 | { |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1329 | sack_block_t *new_list = 0, *block = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1330 | int i; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1331 | |
| 1332 | /* If the first segment is ooo add it to the list. Last write might've moved |
| 1333 | * rcv_nxt over the first segment. */ |
| 1334 | if (seq_lt (tc->rcv_nxt, start)) |
| 1335 | { |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1336 | vec_add2 (new_list, block, 1); |
| 1337 | block->start = start; |
| 1338 | block->end = end; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1339 | } |
| 1340 | |
| 1341 | /* Find the blocks still worth keeping. */ |
| 1342 | for (i = 0; i < vec_len (tc->snd_sacks); i++) |
| 1343 | { |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1344 | /* Discard if rcv_nxt advanced beyond current block */ |
| 1345 | if (seq_leq (tc->snd_sacks[i].start, tc->rcv_nxt)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1346 | continue; |
| 1347 | |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1348 | /* Merge or drop if segment overlapped by the new segment */ |
| 1349 | if (block && (seq_geq (tc->snd_sacks[i].end, new_list[0].start) |
| 1350 | && seq_leq (tc->snd_sacks[i].start, new_list[0].end))) |
| 1351 | { |
| 1352 | if (seq_lt (tc->snd_sacks[i].start, new_list[0].start)) |
| 1353 | new_list[0].start = tc->snd_sacks[i].start; |
| 1354 | if (seq_lt (new_list[0].end, tc->snd_sacks[i].end)) |
| 1355 | new_list[0].end = tc->snd_sacks[i].end; |
| 1356 | continue; |
| 1357 | } |
| 1358 | |
| 1359 | /* Save to new SACK list if we have space. */ |
| 1360 | if (vec_len (new_list) < TCP_MAX_SACK_BLOCKS) |
| 1361 | { |
| 1362 | vec_add1 (new_list, tc->snd_sacks[i]); |
| 1363 | } |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1364 | else |
| 1365 | { |
| 1366 | clib_warning ("sack discarded"); |
| 1367 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1368 | } |
| 1369 | |
Florin Coras | 45d3496 | 2017-04-25 00:05:27 -0700 | [diff] [blame] | 1370 | ASSERT (vec_len (new_list) <= TCP_MAX_SACK_BLOCKS); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1371 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1372 | /* Replace old vector with new one */ |
| 1373 | vec_free (tc->snd_sacks); |
| 1374 | tc->snd_sacks = new_list; |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1375 | |
| 1376 | /* Segments should not 'touch' */ |
| 1377 | ASSERT (tcp_sack_vector_is_sane (tc->snd_sacks)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1378 | } |
| 1379 | |
| 1380 | /** Enqueue data for delivery to application */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1381 | always_inline int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1382 | tcp_session_enqueue_data (tcp_connection_t * tc, vlib_buffer_t * b, |
| 1383 | u16 data_len) |
| 1384 | { |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1385 | int written, error = TCP_ERROR_ENQUEUED; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1386 | |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1387 | ASSERT (seq_geq (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt)); |
| 1388 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1389 | /* Pure ACK. Update rcv_nxt and be done. */ |
| 1390 | if (PREDICT_FALSE (data_len == 0)) |
| 1391 | { |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1392 | return TCP_ERROR_PURE_ACK; |
| 1393 | } |
| 1394 | |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1395 | written = session_enqueue_stream_connection (&tc->connection, b, 0, |
| 1396 | 1 /* queue event */ , 1); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1397 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1398 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 0, data_len, written); |
| 1399 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1400 | /* Update rcv_nxt */ |
| 1401 | if (PREDICT_TRUE (written == data_len)) |
| 1402 | { |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1403 | tc->rcv_nxt += written; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1404 | } |
| 1405 | /* If more data written than expected, account for out-of-order bytes. */ |
| 1406 | else if (written > data_len) |
| 1407 | { |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1408 | tc->rcv_nxt += written; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1409 | |
| 1410 | /* Send ACK confirming the update */ |
| 1411 | tc->flags |= TCP_CONN_SNDACK; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1412 | } |
| 1413 | else if (written > 0) |
| 1414 | { |
| 1415 | /* We've written something but FIFO is probably full now */ |
| 1416 | tc->rcv_nxt += written; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1417 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1418 | /* 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] | 1419 | * not be enqueued. Inform peer */ |
| 1420 | tc->flags |= TCP_CONN_SNDACK; |
| 1421 | |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1422 | error = TCP_ERROR_PARTIALLY_ENQUEUED; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1423 | } |
| 1424 | else |
| 1425 | { |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1426 | tc->flags |= TCP_CONN_SNDACK; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1427 | return TCP_ERROR_FIFO_FULL; |
| 1428 | } |
| 1429 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1430 | /* Update SACK list if need be */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1431 | if (tcp_opts_sack_permitted (&tc->rcv_opts)) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1432 | { |
| 1433 | /* Remove SACK blocks that have been delivered */ |
| 1434 | tcp_update_sack_list (tc, tc->rcv_nxt, tc->rcv_nxt); |
| 1435 | } |
| 1436 | |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1437 | return error; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1438 | } |
| 1439 | |
| 1440 | /** Enqueue out-of-order data */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1441 | always_inline int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1442 | tcp_session_enqueue_ooo (tcp_connection_t * tc, vlib_buffer_t * b, |
| 1443 | u16 data_len) |
| 1444 | { |
| 1445 | stream_session_t *s0; |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1446 | int rv, offset; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1447 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1448 | ASSERT (seq_gt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt)); |
| 1449 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1450 | /* Pure ACK. Do nothing */ |
| 1451 | if (PREDICT_FALSE (data_len == 0)) |
| 1452 | { |
| 1453 | return TCP_ERROR_PURE_ACK; |
| 1454 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1455 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1456 | /* Enqueue out-of-order data with relative offset */ |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1457 | rv = session_enqueue_stream_connection (&tc->connection, b, |
| 1458 | vnet_buffer (b)->tcp.seq_number - |
| 1459 | tc->rcv_nxt, 0 /* queue event */ , |
| 1460 | 0); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1461 | |
| 1462 | /* Nothing written */ |
| 1463 | if (rv) |
| 1464 | { |
| 1465 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, 0); |
| 1466 | return TCP_ERROR_FIFO_FULL; |
| 1467 | } |
| 1468 | |
| 1469 | TCP_EVT_DBG (TCP_EVT_INPUT, tc, 1, data_len, data_len); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1470 | |
| 1471 | /* Update SACK list if in use */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 1472 | if (tcp_opts_sack_permitted (&tc->rcv_opts)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1473 | { |
| 1474 | ooo_segment_t *newest; |
| 1475 | u32 start, end; |
| 1476 | |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1477 | s0 = session_get (tc->c_s_index, tc->c_thread_index); |
Florin Coras | f6d68ed | 2017-05-07 19:12:02 -0700 | [diff] [blame] | 1478 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1479 | /* Get the newest segment from the fifo */ |
| 1480 | newest = svm_fifo_newest_ooo_segment (s0->server_rx_fifo); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1481 | if (newest) |
| 1482 | { |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1483 | offset = ooo_segment_offset (s0->server_rx_fifo, newest); |
| 1484 | ASSERT (offset <= vnet_buffer (b)->tcp.seq_number - tc->rcv_nxt); |
| 1485 | start = tc->rcv_nxt + offset; |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1486 | end = start + ooo_segment_length (s0->server_rx_fifo, newest); |
| 1487 | tcp_update_sack_list (tc, start, end); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 1488 | svm_fifo_newest_ooo_segment_reset (s0->server_rx_fifo); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1489 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1490 | } |
| 1491 | |
| 1492 | return TCP_ERROR_ENQUEUED; |
| 1493 | } |
| 1494 | |
| 1495 | /** |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1496 | * Check if ACK could be delayed. If ack can be delayed, it should return |
| 1497 | * true for a full frame. If we're always acking return 0. |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1498 | */ |
| 1499 | always_inline int |
| 1500 | tcp_can_delack (tcp_connection_t * tc) |
| 1501 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1502 | /* Send ack if ... */ |
| 1503 | if (TCP_ALWAYS_ACK |
| 1504 | /* just sent a rcv wnd 0 */ |
| 1505 | || (tc->flags & TCP_CONN_SENT_RCV_WND0) != 0 |
| 1506 | /* constrained to send ack */ |
| 1507 | || (tc->flags & TCP_CONN_SNDACK) != 0 |
| 1508 | /* we're almost out of tx wnd */ |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1509 | || tcp_available_snd_space (tc) < 4 * tc->snd_mss) |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1510 | return 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1511 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1512 | return 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1513 | } |
| 1514 | |
| 1515 | static int |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1516 | tcp_buffer_discard_bytes (vlib_buffer_t * b, u32 n_bytes_to_drop) |
| 1517 | { |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1518 | u32 discard, first = b->current_length; |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1519 | vlib_main_t *vm = vlib_get_main (); |
| 1520 | |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1521 | /* Handle multi-buffer segments */ |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1522 | if (n_bytes_to_drop > b->current_length) |
| 1523 | { |
| 1524 | if (!(b->flags & VLIB_BUFFER_NEXT_PRESENT)) |
| 1525 | return -1; |
| 1526 | do |
| 1527 | { |
| 1528 | discard = clib_min (n_bytes_to_drop, b->current_length); |
| 1529 | vlib_buffer_advance (b, discard); |
| 1530 | b = vlib_get_buffer (vm, b->next_buffer); |
| 1531 | n_bytes_to_drop -= discard; |
| 1532 | } |
| 1533 | while (n_bytes_to_drop); |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1534 | if (n_bytes_to_drop > first) |
| 1535 | b->total_length_not_including_first_buffer -= n_bytes_to_drop - first; |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1536 | } |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 1537 | else |
| 1538 | vlib_buffer_advance (b, n_bytes_to_drop); |
| 1539 | vnet_buffer (b)->tcp.data_len -= n_bytes_to_drop; |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1540 | return 0; |
| 1541 | } |
| 1542 | |
| 1543 | static int |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1544 | tcp_segment_rcv (tcp_main_t * tm, tcp_connection_t * tc, vlib_buffer_t * b, |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1545 | u32 * next0) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1546 | { |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1547 | u32 error = 0, n_bytes_to_drop, n_data_bytes; |
| 1548 | |
| 1549 | vlib_buffer_advance (b, vnet_buffer (b)->tcp.data_offset); |
| 1550 | n_data_bytes = vnet_buffer (b)->tcp.data_len; |
| 1551 | ASSERT (n_data_bytes); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1552 | |
| 1553 | /* Handle out-of-order data */ |
| 1554 | if (PREDICT_FALSE (vnet_buffer (b)->tcp.seq_number != tc->rcv_nxt)) |
| 1555 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1556 | /* Old sequence numbers allowed through because they overlapped |
| 1557 | * the rx window */ |
| 1558 | if (seq_lt (vnet_buffer (b)->tcp.seq_number, tc->rcv_nxt)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1559 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1560 | error = TCP_ERROR_SEGMENT_OLD; |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 1561 | *next0 = tcp_next_drop (tc->c_is_ip4); |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 1562 | |
Dave Barach | 259cdae | 2017-05-15 16:27:05 -0400 | [diff] [blame] | 1563 | /* Completely in the past (possible retransmit) */ |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 1564 | if (seq_leq (vnet_buffer (b)->tcp.seq_end, tc->rcv_nxt)) |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1565 | { |
| 1566 | /* Ack retransmissions since we may not have any data to send */ |
| 1567 | tcp_make_ack (tc, b); |
| 1568 | *next0 = tcp_next_output (tc->c_is_ip4); |
| 1569 | goto done; |
| 1570 | } |
Dave Barach | 259cdae | 2017-05-15 16:27:05 -0400 | [diff] [blame] | 1571 | |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 1572 | /* Chop off the bytes in the past */ |
| 1573 | n_bytes_to_drop = tc->rcv_nxt - vnet_buffer (b)->tcp.seq_number; |
| 1574 | n_data_bytes -= n_bytes_to_drop; |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 1575 | vnet_buffer (b)->tcp.seq_number = tc->rcv_nxt; |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 1576 | if (tcp_buffer_discard_bytes (b, n_bytes_to_drop)) |
| 1577 | goto done; |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 1578 | |
| 1579 | goto in_order; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1580 | } |
| 1581 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1582 | error = tcp_session_enqueue_ooo (tc, b, n_data_bytes); |
| 1583 | |
| 1584 | /* N.B. Should not filter burst of dupacks. Two issues 1) dupacks open |
| 1585 | * cwnd on remote peer when congested 2) acks leaving should have the |
| 1586 | * latest rcv_wnd since the burst may eaten up all of it, so only the |
| 1587 | * old ones could be filtered. |
| 1588 | */ |
| 1589 | |
| 1590 | /* RFC2581: Send DUPACK for fast retransmit */ |
| 1591 | tcp_make_ack (tc, b); |
| 1592 | *next0 = tcp_next_output (tc->c_is_ip4); |
| 1593 | |
| 1594 | /* Mark as DUPACK. We may filter these in output if |
| 1595 | * the burst fills the holes. */ |
| 1596 | if (n_data_bytes) |
| 1597 | vnet_buffer (b)->tcp.flags = TCP_BUF_FLAG_DUPACK; |
| 1598 | |
| 1599 | TCP_EVT_DBG (TCP_EVT_DUPACK_SENT, tc); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1600 | goto done; |
| 1601 | } |
| 1602 | |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 1603 | in_order: |
| 1604 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1605 | /* In order data, enqueue. Fifo figures out by itself if any out-of-order |
| 1606 | * segments can be enqueued after fifo tail offset changes. */ |
| 1607 | error = tcp_session_enqueue_data (tc, b, n_data_bytes); |
| 1608 | |
| 1609 | /* Check if ACK can be delayed */ |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1610 | if (tcp_can_delack (tc)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1611 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1612 | if (!tcp_timer_is_active (tc, TCP_TIMER_DELACK)) |
| 1613 | tcp_timer_set (tc, TCP_TIMER_DELACK, TCP_DELACK_TIME); |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1614 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1615 | } |
| 1616 | |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 1617 | *next0 = tcp_next_output (tc->c_is_ip4); |
| 1618 | tcp_make_ack (tc, b); |
| 1619 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1620 | done: |
| 1621 | return error; |
| 1622 | } |
| 1623 | |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1624 | typedef struct |
| 1625 | { |
| 1626 | tcp_header_t tcp_header; |
| 1627 | tcp_connection_t tcp_connection; |
| 1628 | } tcp_rx_trace_t; |
| 1629 | |
| 1630 | u8 * |
| 1631 | format_tcp_rx_trace (u8 * s, va_list * args) |
| 1632 | { |
| 1633 | CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); |
| 1634 | CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); |
| 1635 | tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); |
Christophe Fontaine | d3c008d | 2017-10-02 18:10:54 +0200 | [diff] [blame] | 1636 | u32 indent = format_get_indent (s); |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1637 | |
| 1638 | s = format (s, "%U\n%U%U", |
| 1639 | format_tcp_header, &t->tcp_header, 128, |
| 1640 | format_white_space, indent, |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 1641 | format_tcp_connection, &t->tcp_connection, 1); |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1642 | |
| 1643 | return s; |
| 1644 | } |
| 1645 | |
| 1646 | u8 * |
| 1647 | format_tcp_rx_trace_short (u8 * s, va_list * args) |
| 1648 | { |
| 1649 | CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); |
| 1650 | CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); |
| 1651 | tcp_rx_trace_t *t = va_arg (*args, tcp_rx_trace_t *); |
| 1652 | |
| 1653 | s = format (s, "%d -> %d (%U)", |
| 1654 | clib_net_to_host_u16 (t->tcp_header.src_port), |
| 1655 | clib_net_to_host_u16 (t->tcp_header.dst_port), format_tcp_state, |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 1656 | t->tcp_connection.state); |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1657 | |
| 1658 | return s; |
| 1659 | } |
| 1660 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 1661 | void |
| 1662 | tcp_set_rx_trace_data (tcp_rx_trace_t * t0, tcp_connection_t * tc0, |
| 1663 | tcp_header_t * th0, vlib_buffer_t * b0, u8 is_ip4) |
| 1664 | { |
| 1665 | if (tc0) |
| 1666 | { |
| 1667 | clib_memcpy (&t0->tcp_connection, tc0, sizeof (t0->tcp_connection)); |
| 1668 | } |
| 1669 | else |
| 1670 | { |
| 1671 | th0 = tcp_buffer_hdr (b0); |
| 1672 | } |
| 1673 | clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header)); |
| 1674 | } |
| 1675 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1676 | always_inline void |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1677 | tcp_node_inc_counter (vlib_main_t * vm, u32 tcp4_node, u32 tcp6_node, |
| 1678 | u8 is_ip4, u8 evt, u8 val) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1679 | { |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1680 | if (PREDICT_TRUE (!val)) |
| 1681 | return; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1682 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1683 | if (is_ip4) |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1684 | vlib_node_increment_counter (vm, tcp4_node, evt, val); |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1685 | else |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1686 | vlib_node_increment_counter (vm, tcp6_node, evt, val); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1687 | } |
| 1688 | |
| 1689 | always_inline uword |
| 1690 | tcp46_established_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1691 | vlib_frame_t * from_frame, int is_ip4) |
| 1692 | { |
| 1693 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 1694 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1695 | tcp_main_t *tm = vnet_get_tcp_main (); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1696 | u8 is_fin = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1697 | |
| 1698 | from = vlib_frame_vector_args (from_frame); |
| 1699 | n_left_from = from_frame->n_vectors; |
| 1700 | |
| 1701 | next_index = node->cached_next_index; |
| 1702 | |
| 1703 | while (n_left_from > 0) |
| 1704 | { |
| 1705 | u32 n_left_to_next; |
| 1706 | |
| 1707 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1708 | while (n_left_from > 0 && n_left_to_next > 0) |
| 1709 | { |
| 1710 | u32 bi0; |
| 1711 | vlib_buffer_t *b0; |
| 1712 | tcp_header_t *th0 = 0; |
| 1713 | tcp_connection_t *tc0; |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 1714 | u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1715 | |
| 1716 | bi0 = from[0]; |
| 1717 | to_next[0] = bi0; |
| 1718 | from += 1; |
| 1719 | to_next += 1; |
| 1720 | n_left_from -= 1; |
| 1721 | n_left_to_next -= 1; |
| 1722 | |
| 1723 | b0 = vlib_get_buffer (vm, bi0); |
| 1724 | tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, |
| 1725 | my_thread_index); |
| 1726 | |
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; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1730 | goto done; |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1731 | } |
| 1732 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 1733 | th0 = tcp_buffer_hdr (b0); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1734 | /* N.B. buffer is rewritten if segment is ooo. Thus, th0 becomes a |
| 1735 | * dangling reference. */ |
| 1736 | is_fin = tcp_is_fin (th0); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1737 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1738 | /* SYNs, FINs and data consume sequence numbers */ |
| 1739 | vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 1740 | + tcp_is_syn (th0) + is_fin + vnet_buffer (b0)->tcp.data_len; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1741 | |
| 1742 | /* TODO header prediction fast path */ |
| 1743 | |
| 1744 | /* 1-4: check SEQ, RST, SYN */ |
| 1745 | if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, th0, &next0))) |
| 1746 | { |
| 1747 | error0 = TCP_ERROR_SEGMENT_INVALID; |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1748 | TCP_EVT_DBG (TCP_EVT_SEG_INVALID, tc0, |
| 1749 | vnet_buffer (b0)->tcp.seq_number, |
| 1750 | vnet_buffer (b0)->tcp.seq_end); |
| 1751 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1752 | } |
| 1753 | |
| 1754 | /* 5: check the ACK field */ |
| 1755 | if (tcp_rcv_ack (tc0, b0, th0, &next0, &error0)) |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1756 | goto done; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1757 | |
| 1758 | /* 6: check the URG bit TODO */ |
| 1759 | |
| 1760 | /* 7: process the segment text */ |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1761 | if (vnet_buffer (b0)->tcp.data_len) |
| 1762 | error0 = tcp_segment_rcv (tm, tc0, b0, &next0); |
Dave Barach | 1f75cfd | 2017-04-14 16:46:44 -0400 | [diff] [blame] | 1763 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1764 | /* 8: check the FIN bit */ |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 1765 | if (PREDICT_FALSE (is_fin)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1766 | { |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 1767 | /* Enter CLOSE-WAIT and notify session. To avoid lingering |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 1768 | * in CLOSE-WAIT, set timer (reuse WAITCLOSE). */ |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 1769 | /* Account for the FIN if nothing else was received */ |
Florin Coras | 6881062 | 2017-07-24 17:40:28 -0700 | [diff] [blame] | 1770 | if (vnet_buffer (b0)->tcp.data_len == 0) |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 1771 | tc0->rcv_nxt += 1; |
| 1772 | tcp_make_ack (tc0, b0); |
| 1773 | next0 = tcp_next_output (tc0->c_is_ip4); |
| 1774 | tc0->state = TCP_STATE_CLOSE_WAIT; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1775 | stream_session_disconnect_notify (&tc0->connection); |
Florin Coras | 6881062 | 2017-07-24 17:40:28 -0700 | [diff] [blame] | 1776 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_CLOSEWAIT_TIME); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 1777 | TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1778 | } |
| 1779 | |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 1780 | done: |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1781 | b0->error = node->errors[error0]; |
| 1782 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 1783 | { |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 1784 | tcp_rx_trace_t *t0 = |
| 1785 | vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 1786 | tcp_set_rx_trace_data (t0, tc0, th0, b0, is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1787 | } |
| 1788 | |
| 1789 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 1790 | n_left_to_next, bi0, next0); |
| 1791 | } |
| 1792 | |
| 1793 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 1794 | } |
| 1795 | |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1796 | errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP, |
| 1797 | my_thread_index); |
| 1798 | tcp_node_inc_counter (vm, is_ip4, tcp4_established_node.index, |
| 1799 | tcp6_established_node.index, |
| 1800 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
Florin Coras | 66b1131 | 2017-07-31 17:18:03 -0700 | [diff] [blame] | 1801 | tcp_flush_frame_to_output (vm, my_thread_index, is_ip4); |
| 1802 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1803 | return from_frame->n_vectors; |
| 1804 | } |
| 1805 | |
| 1806 | static uword |
| 1807 | tcp4_established (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1808 | vlib_frame_t * from_frame) |
| 1809 | { |
| 1810 | return tcp46_established_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 1811 | } |
| 1812 | |
| 1813 | static uword |
| 1814 | tcp6_established (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1815 | vlib_frame_t * from_frame) |
| 1816 | { |
| 1817 | return tcp46_established_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 1818 | } |
| 1819 | |
| 1820 | /* *INDENT-OFF* */ |
| 1821 | VLIB_REGISTER_NODE (tcp4_established_node) = |
| 1822 | { |
| 1823 | .function = tcp4_established, |
| 1824 | .name = "tcp4-established", |
| 1825 | /* Takes a vector of packets. */ |
| 1826 | .vector_size = sizeof (u32), |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 1827 | .n_errors = TCP_N_ERROR, |
| 1828 | .error_strings = tcp_error_strings, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1829 | .n_next_nodes = TCP_ESTABLISHED_N_NEXT, |
| 1830 | .next_nodes = |
| 1831 | { |
| 1832 | #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, |
| 1833 | foreach_tcp_state_next |
| 1834 | #undef _ |
| 1835 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1836 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1837 | }; |
| 1838 | /* *INDENT-ON* */ |
| 1839 | |
| 1840 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_established_node, tcp4_established); |
| 1841 | |
| 1842 | /* *INDENT-OFF* */ |
| 1843 | VLIB_REGISTER_NODE (tcp6_established_node) = |
| 1844 | { |
| 1845 | .function = tcp6_established, |
| 1846 | .name = "tcp6-established", |
| 1847 | /* Takes a vector of packets. */ |
| 1848 | .vector_size = sizeof (u32), |
| 1849 | .n_errors = TCP_N_ERROR, |
| 1850 | .error_strings = tcp_error_strings, |
| 1851 | .n_next_nodes = TCP_ESTABLISHED_N_NEXT, |
| 1852 | .next_nodes = |
| 1853 | { |
| 1854 | #define _(s,n) [TCP_ESTABLISHED_NEXT_##s] = n, |
| 1855 | foreach_tcp_state_next |
| 1856 | #undef _ |
| 1857 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1858 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1859 | }; |
| 1860 | /* *INDENT-ON* */ |
| 1861 | |
| 1862 | |
| 1863 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_established_node, tcp6_established); |
| 1864 | |
| 1865 | vlib_node_registration_t tcp4_syn_sent_node; |
| 1866 | vlib_node_registration_t tcp6_syn_sent_node; |
| 1867 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1868 | static u8 |
| 1869 | tcp_lookup_is_valid (tcp_connection_t * tc, tcp_header_t * hdr) |
| 1870 | { |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1871 | transport_connection_t *tmp = 0; |
| 1872 | u64 handle; |
| 1873 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1874 | if (!tc) |
| 1875 | return 1; |
| 1876 | |
Florin Coras | 7013113 | 2017-11-27 02:43:30 -0800 | [diff] [blame] | 1877 | /* Proxy case */ |
| 1878 | if (tc->c_lcl_port == 0 && tc->state == TCP_STATE_LISTEN) |
| 1879 | return 1; |
| 1880 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1881 | u8 is_valid = (tc->c_lcl_port == hdr->dst_port |
| 1882 | && (tc->state == TCP_STATE_LISTEN |
| 1883 | || tc->c_rmt_port == hdr->src_port)); |
| 1884 | |
| 1885 | if (!is_valid) |
| 1886 | { |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1887 | handle = session_lookup_half_open_handle (&tc->connection); |
| 1888 | tmp = session_lookup_half_open_connection (handle & 0xFFFFFFFF, |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1889 | tc->c_proto, tc->c_is_ip4); |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1890 | |
| 1891 | if (tmp) |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1892 | { |
| 1893 | if (tmp->lcl_port == hdr->dst_port |
| 1894 | && tmp->rmt_port == hdr->src_port) |
| 1895 | { |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1896 | TCP_DBG ("half-open is valid!"); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1897 | } |
| 1898 | } |
| 1899 | } |
| 1900 | return is_valid; |
| 1901 | } |
| 1902 | |
| 1903 | /** |
| 1904 | * Lookup transport connection |
| 1905 | */ |
| 1906 | static tcp_connection_t * |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1907 | tcp_lookup_connection (u32 fib_index, vlib_buffer_t * b, u8 thread_index, |
| 1908 | u8 is_ip4) |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1909 | { |
| 1910 | tcp_header_t *tcp; |
| 1911 | transport_connection_t *tconn; |
| 1912 | tcp_connection_t *tc; |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 1913 | u8 is_filtered = 0; |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1914 | if (is_ip4) |
| 1915 | { |
| 1916 | ip4_header_t *ip4; |
| 1917 | ip4 = vlib_buffer_get_current (b); |
| 1918 | tcp = ip4_next_header (ip4); |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1919 | tconn = session_lookup_connection_wt4 (fib_index, |
| 1920 | &ip4->dst_address, |
| 1921 | &ip4->src_address, |
| 1922 | tcp->dst_port, |
| 1923 | tcp->src_port, |
| 1924 | TRANSPORT_PROTO_TCP, |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 1925 | thread_index, &is_filtered); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1926 | tc = tcp_get_connection_from_transport (tconn); |
| 1927 | ASSERT (tcp_lookup_is_valid (tc, tcp)); |
| 1928 | } |
| 1929 | else |
| 1930 | { |
| 1931 | ip6_header_t *ip6; |
| 1932 | ip6 = vlib_buffer_get_current (b); |
| 1933 | tcp = ip6_next_header (ip6); |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1934 | tconn = session_lookup_connection_wt6 (fib_index, |
| 1935 | &ip6->dst_address, |
| 1936 | &ip6->src_address, |
| 1937 | tcp->dst_port, |
| 1938 | tcp->src_port, |
| 1939 | TRANSPORT_PROTO_TCP, |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 1940 | thread_index, &is_filtered); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1941 | tc = tcp_get_connection_from_transport (tconn); |
| 1942 | ASSERT (tcp_lookup_is_valid (tc, tcp)); |
| 1943 | } |
| 1944 | return tc; |
| 1945 | } |
| 1946 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1947 | always_inline uword |
| 1948 | tcp46_syn_sent_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 1949 | vlib_frame_t * from_frame, int is_ip4) |
| 1950 | { |
| 1951 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 1952 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 1953 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1954 | |
| 1955 | from = vlib_frame_vector_args (from_frame); |
| 1956 | n_left_from = from_frame->n_vectors; |
| 1957 | |
| 1958 | next_index = node->cached_next_index; |
| 1959 | |
| 1960 | while (n_left_from > 0) |
| 1961 | { |
| 1962 | u32 n_left_to_next; |
| 1963 | |
| 1964 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 1965 | |
| 1966 | while (n_left_from > 0 && n_left_to_next > 0) |
| 1967 | { |
| 1968 | u32 bi0, ack0, seq0; |
| 1969 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 1970 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1971 | tcp_header_t *tcp0 = 0; |
| 1972 | tcp_connection_t *tc0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1973 | tcp_connection_t *new_tc0; |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 1974 | u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1975 | |
| 1976 | bi0 = from[0]; |
| 1977 | to_next[0] = bi0; |
| 1978 | from += 1; |
| 1979 | to_next += 1; |
| 1980 | n_left_from -= 1; |
| 1981 | n_left_to_next -= 1; |
| 1982 | |
| 1983 | b0 = vlib_get_buffer (vm, bi0); |
| 1984 | tc0 = |
| 1985 | tcp_half_open_connection_get (vnet_buffer (b0)-> |
| 1986 | tcp.connection_index); |
Dave Barach | d84ba85 | 2017-08-22 17:56:46 -0400 | [diff] [blame] | 1987 | if (PREDICT_FALSE (tc0 == 0)) |
| 1988 | { |
| 1989 | error0 = TCP_ERROR_INVALID_CONNECTION; |
| 1990 | goto drop; |
| 1991 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 1992 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 1993 | /* Half-open completed recently but the connection was't removed |
| 1994 | * yet by the owning thread */ |
| 1995 | if (PREDICT_FALSE (tc0->flags & TCP_CONN_HALF_OPEN_DONE)) |
| 1996 | { |
| 1997 | /* Make sure the connection actually exists */ |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 1998 | ASSERT (tcp_lookup_connection (tc0->c_fib_index, b0, |
| 1999 | my_thread_index, is_ip4)); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2000 | goto drop; |
| 2001 | } |
| 2002 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2003 | ack0 = vnet_buffer (b0)->tcp.ack_number; |
| 2004 | seq0 = vnet_buffer (b0)->tcp.seq_number; |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2005 | tcp0 = tcp_buffer_hdr (b0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2006 | |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2007 | /* Crude check to see if the connection handle does not match |
| 2008 | * the packet. Probably connection just switched to established */ |
| 2009 | if (PREDICT_FALSE (tcp0->dst_port != tc0->c_lcl_port |
| 2010 | || tcp0->src_port != tc0->c_rmt_port)) |
| 2011 | goto drop; |
| 2012 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2013 | if (PREDICT_FALSE |
| 2014 | (!tcp_ack (tcp0) && !tcp_rst (tcp0) && !tcp_syn (tcp0))) |
| 2015 | goto drop; |
| 2016 | |
| 2017 | /* SYNs, FINs and data consume sequence numbers */ |
| 2018 | vnet_buffer (b0)->tcp.seq_end = seq0 + tcp_is_syn (tcp0) |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2019 | + tcp_is_fin (tcp0) + vnet_buffer (b0)->tcp.data_len; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2020 | |
| 2021 | /* |
| 2022 | * 1. check the ACK bit |
| 2023 | */ |
| 2024 | |
| 2025 | /* |
| 2026 | * If the ACK bit is set |
| 2027 | * If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset (unless |
| 2028 | * the RST bit is set, if so drop the segment and return) |
| 2029 | * <SEQ=SEG.ACK><CTL=RST> |
| 2030 | * and discard the segment. Return. |
| 2031 | * If SND.UNA =< SEG.ACK =< SND.NXT then the ACK is acceptable. |
| 2032 | */ |
| 2033 | if (tcp_ack (tcp0)) |
| 2034 | { |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2035 | if (seq_leq (ack0, tc0->iss) || seq_gt (ack0, tc0->snd_nxt)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2036 | { |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2037 | clib_warning ("ack not in rcv wnd"); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2038 | if (!tcp_rst (tcp0)) |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 2039 | tcp_send_reset_w_pkt (tc0, b0, is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2040 | goto drop; |
| 2041 | } |
| 2042 | |
| 2043 | /* Make sure ACK is valid */ |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2044 | if (seq_gt (tc0->snd_una, ack0)) |
| 2045 | { |
| 2046 | clib_warning ("ack invalid"); |
| 2047 | goto drop; |
| 2048 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2049 | } |
| 2050 | |
| 2051 | /* |
| 2052 | * 2. check the RST bit |
| 2053 | */ |
| 2054 | |
| 2055 | if (tcp_rst (tcp0)) |
| 2056 | { |
| 2057 | /* If ACK is acceptable, signal client that peer is not |
| 2058 | * willing to accept connection and drop connection*/ |
| 2059 | if (tcp_ack (tcp0)) |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2060 | tcp_connection_reset (tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2061 | goto drop; |
| 2062 | } |
| 2063 | |
| 2064 | /* |
| 2065 | * 3. check the security and precedence (skipped) |
| 2066 | */ |
| 2067 | |
| 2068 | /* |
| 2069 | * 4. check the SYN bit |
| 2070 | */ |
| 2071 | |
| 2072 | /* No SYN flag. Drop. */ |
| 2073 | if (!tcp_syn (tcp0)) |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2074 | { |
| 2075 | clib_warning ("not synack"); |
| 2076 | goto drop; |
| 2077 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2078 | |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2079 | /* Parse options */ |
| 2080 | if (tcp_options_parse (tcp0, &tc0->rcv_opts)) |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2081 | { |
| 2082 | clib_warning ("options parse fail"); |
| 2083 | goto drop; |
| 2084 | } |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2085 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2086 | /* Valid SYN or SYN-ACK. Move connection from half-open pool to |
| 2087 | * current thread pool. */ |
| 2088 | pool_get (tm->connections[my_thread_index], new_tc0); |
| 2089 | clib_memcpy (new_tc0, tc0, sizeof (*new_tc0)); |
Dave Barach | 259cdae | 2017-05-15 16:27:05 -0400 | [diff] [blame] | 2090 | new_tc0->c_c_index = new_tc0 - tm->connections[my_thread_index]; |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2091 | new_tc0->c_thread_index = my_thread_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2092 | new_tc0->rcv_nxt = vnet_buffer (b0)->tcp.seq_end; |
| 2093 | new_tc0->irs = seq0; |
Florin Coras | 6881062 | 2017-07-24 17:40:28 -0700 | [diff] [blame] | 2094 | new_tc0->timers[TCP_TIMER_ESTABLISH] = TCP_TIMER_HANDLE_INVALID; |
| 2095 | new_tc0->timers[TCP_TIMER_RETRANSMIT_SYN] = |
| 2096 | TCP_TIMER_HANDLE_INVALID; |
| 2097 | |
| 2098 | /* If this is not the owning thread, wait for syn retransmit to |
| 2099 | * expire and cleanup then */ |
| 2100 | if (tcp_half_open_connection_cleanup (tc0)) |
| 2101 | tc0->flags |= TCP_CONN_HALF_OPEN_DONE; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2102 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2103 | if (tcp_opts_tstamp (&new_tc0->rcv_opts)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2104 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2105 | new_tc0->tsval_recent = new_tc0->rcv_opts.tsval; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2106 | new_tc0->tsval_recent_age = tcp_time_now (); |
| 2107 | } |
| 2108 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2109 | if (tcp_opts_wscale (&new_tc0->rcv_opts)) |
| 2110 | new_tc0->snd_wscale = new_tc0->rcv_opts.wscale; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2111 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2112 | /* RFC1323: SYN and SYN-ACK wnd not scaled */ |
| 2113 | new_tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2114 | new_tc0->snd_wl1 = seq0; |
| 2115 | new_tc0->snd_wl2 = ack0; |
| 2116 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2117 | tcp_connection_init_vars (new_tc0); |
| 2118 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2119 | /* SYN-ACK: See if we can switch to ESTABLISHED state */ |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2120 | if (PREDICT_TRUE (tcp_ack (tcp0))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2121 | { |
| 2122 | /* Our SYN is ACKed: we have iss < ack = snd_una */ |
| 2123 | |
| 2124 | /* TODO Dequeue acknowledged segments if we support Fast Open */ |
| 2125 | new_tc0->snd_una = ack0; |
| 2126 | new_tc0->state = TCP_STATE_ESTABLISHED; |
| 2127 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2128 | /* Make sure las is initialized for the wnd computation */ |
| 2129 | new_tc0->rcv_las = new_tc0->rcv_nxt; |
| 2130 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2131 | /* Notify app that we have connection. If session layer can't |
| 2132 | * allocate session send reset */ |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2133 | if (session_stream_connect_notify (&new_tc0->connection, 0)) |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2134 | { |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2135 | clib_warning ("connect notify fail"); |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 2136 | tcp_send_reset_w_pkt (new_tc0, b0, is_ip4); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2137 | tcp_connection_cleanup (new_tc0); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2138 | goto drop; |
| 2139 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2140 | |
| 2141 | /* Make sure after data segment processing ACK is sent */ |
| 2142 | new_tc0->flags |= TCP_CONN_SNDACK; |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 2143 | |
| 2144 | /* Update rtt with the syn-ack sample */ |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 2145 | tcp_update_rtt (new_tc0, vnet_buffer (b0)->tcp.ack_number); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2146 | TCP_EVT_DBG (TCP_EVT_SYNACK_RCVD, new_tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2147 | } |
| 2148 | /* SYN: Simultaneous open. Change state to SYN-RCVD and send SYN-ACK */ |
| 2149 | else |
| 2150 | { |
| 2151 | new_tc0->state = TCP_STATE_SYN_RCVD; |
| 2152 | |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 2153 | /* Notify app that we have connection */ |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2154 | if (session_stream_connect_notify (&new_tc0->connection, 0)) |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2155 | { |
| 2156 | tcp_connection_cleanup (new_tc0); |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 2157 | tcp_send_reset_w_pkt (tc0, b0, is_ip4); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2158 | TCP_EVT_DBG (TCP_EVT_RST_SENT, tc0); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2159 | goto drop; |
| 2160 | } |
| 2161 | |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 2162 | tc0->rtt_ts = 0; |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2163 | tcp_init_snd_vars (tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2164 | tcp_make_synack (new_tc0, b0); |
| 2165 | next0 = tcp_next_output (is_ip4); |
| 2166 | |
| 2167 | goto drop; |
| 2168 | } |
| 2169 | |
| 2170 | /* Read data, if any */ |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2171 | if (PREDICT_FALSE (vnet_buffer (b0)->tcp.data_len)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2172 | { |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2173 | ASSERT (0); |
| 2174 | error0 = tcp_segment_rcv (tm, new_tc0, b0, &next0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2175 | if (error0 == TCP_ERROR_PURE_ACK) |
| 2176 | error0 = TCP_ERROR_SYN_ACKS_RCVD; |
| 2177 | } |
| 2178 | else |
| 2179 | { |
| 2180 | tcp_make_ack (new_tc0, b0); |
| 2181 | next0 = tcp_next_output (new_tc0->c_is_ip4); |
| 2182 | } |
| 2183 | |
| 2184 | drop: |
| 2185 | |
| 2186 | b0->error = error0 ? node->errors[error0] : 0; |
Chris Luke | 879ace3 | 2017-09-26 13:15:16 -0400 | [diff] [blame] | 2187 | if (PREDICT_FALSE |
| 2188 | ((b0->flags & VLIB_BUFFER_IS_TRACED) && tcp0 != 0)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2189 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2190 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 2191 | clib_memcpy (&t0->tcp_header, tcp0, sizeof (t0->tcp_header)); |
| 2192 | clib_memcpy (&t0->tcp_connection, tc0, |
| 2193 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2194 | } |
| 2195 | |
| 2196 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 2197 | n_left_to_next, bi0, next0); |
| 2198 | } |
| 2199 | |
| 2200 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 2201 | } |
| 2202 | |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2203 | errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP, |
| 2204 | my_thread_index); |
| 2205 | tcp_node_inc_counter (vm, is_ip4, tcp4_syn_sent_node.index, |
| 2206 | tcp6_syn_sent_node.index, |
| 2207 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2208 | return from_frame->n_vectors; |
| 2209 | } |
| 2210 | |
| 2211 | static uword |
| 2212 | tcp4_syn_sent (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2213 | vlib_frame_t * from_frame) |
| 2214 | { |
| 2215 | return tcp46_syn_sent_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 2216 | } |
| 2217 | |
| 2218 | static uword |
| 2219 | tcp6_syn_sent_rcv (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2220 | vlib_frame_t * from_frame) |
| 2221 | { |
| 2222 | return tcp46_syn_sent_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 2223 | } |
| 2224 | |
| 2225 | /* *INDENT-OFF* */ |
| 2226 | VLIB_REGISTER_NODE (tcp4_syn_sent_node) = |
| 2227 | { |
| 2228 | .function = tcp4_syn_sent, |
| 2229 | .name = "tcp4-syn-sent", |
| 2230 | /* Takes a vector of packets. */ |
| 2231 | .vector_size = sizeof (u32), |
| 2232 | .n_errors = TCP_N_ERROR, |
| 2233 | .error_strings = tcp_error_strings, |
| 2234 | .n_next_nodes = TCP_SYN_SENT_N_NEXT, |
| 2235 | .next_nodes = |
| 2236 | { |
| 2237 | #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, |
| 2238 | foreach_tcp_state_next |
| 2239 | #undef _ |
| 2240 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2241 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2242 | }; |
| 2243 | /* *INDENT-ON* */ |
| 2244 | |
| 2245 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_syn_sent_node, tcp4_syn_sent); |
| 2246 | |
| 2247 | /* *INDENT-OFF* */ |
| 2248 | VLIB_REGISTER_NODE (tcp6_syn_sent_node) = |
| 2249 | { |
| 2250 | .function = tcp6_syn_sent_rcv, |
| 2251 | .name = "tcp6-syn-sent", |
| 2252 | /* Takes a vector of packets. */ |
| 2253 | .vector_size = sizeof (u32), |
| 2254 | .n_errors = TCP_N_ERROR, |
| 2255 | .error_strings = tcp_error_strings, |
| 2256 | .n_next_nodes = TCP_SYN_SENT_N_NEXT, |
| 2257 | .next_nodes = |
| 2258 | { |
| 2259 | #define _(s,n) [TCP_SYN_SENT_NEXT_##s] = n, |
| 2260 | foreach_tcp_state_next |
| 2261 | #undef _ |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2262 | }, |
| 2263 | .format_trace = format_tcp_rx_trace_short, |
| 2264 | }; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2265 | /* *INDENT-ON* */ |
| 2266 | |
| 2267 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_syn_sent_node, tcp6_syn_sent_rcv); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2268 | |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2269 | vlib_node_registration_t tcp4_rcv_process_node; |
| 2270 | vlib_node_registration_t tcp6_rcv_process_node; |
| 2271 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2272 | /** |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2273 | * Handles reception for all states except LISTEN, SYN-SENT and ESTABLISHED |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2274 | * as per RFC793 p. 64 |
| 2275 | */ |
| 2276 | always_inline uword |
| 2277 | tcp46_rcv_process_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2278 | vlib_frame_t * from_frame, int is_ip4) |
| 2279 | { |
| 2280 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 2281 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 2282 | u32 my_thread_index = vm->thread_index, errors = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2283 | |
| 2284 | from = vlib_frame_vector_args (from_frame); |
| 2285 | n_left_from = from_frame->n_vectors; |
| 2286 | |
| 2287 | next_index = node->cached_next_index; |
| 2288 | |
| 2289 | while (n_left_from > 0) |
| 2290 | { |
| 2291 | u32 n_left_to_next; |
| 2292 | |
| 2293 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 2294 | |
| 2295 | while (n_left_from > 0 && n_left_to_next > 0) |
| 2296 | { |
| 2297 | u32 bi0; |
| 2298 | vlib_buffer_t *b0; |
| 2299 | tcp_header_t *tcp0 = 0; |
| 2300 | tcp_connection_t *tc0; |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2301 | u32 next0 = tcp_next_drop (is_ip4), error0 = TCP_ERROR_ENQUEUED; |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2302 | u8 is_fin0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2303 | |
| 2304 | bi0 = from[0]; |
| 2305 | to_next[0] = bi0; |
| 2306 | from += 1; |
| 2307 | to_next += 1; |
| 2308 | n_left_from -= 1; |
| 2309 | n_left_to_next -= 1; |
| 2310 | |
| 2311 | b0 = vlib_get_buffer (vm, bi0); |
| 2312 | tc0 = tcp_connection_get (vnet_buffer (b0)->tcp.connection_index, |
| 2313 | my_thread_index); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2314 | if (PREDICT_FALSE (tc0 == 0)) |
| 2315 | { |
| 2316 | error0 = TCP_ERROR_INVALID_CONNECTION; |
| 2317 | goto drop; |
| 2318 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2319 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2320 | tcp0 = tcp_buffer_hdr (b0); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2321 | is_fin0 = tcp_is_fin (tcp0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2322 | |
| 2323 | /* SYNs, FINs and data consume sequence numbers */ |
| 2324 | vnet_buffer (b0)->tcp.seq_end = vnet_buffer (b0)->tcp.seq_number |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2325 | + tcp_is_syn (tcp0) + is_fin0 + vnet_buffer (b0)->tcp.data_len; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2326 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2327 | if (CLIB_DEBUG) |
| 2328 | { |
| 2329 | tcp_connection_t *tmp; |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2330 | tmp = |
| 2331 | tcp_lookup_connection (tc0->c_fib_index, b0, my_thread_index, |
| 2332 | is_ip4); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2333 | if (tmp->state != tc0->state) |
| 2334 | { |
| 2335 | clib_warning ("state changed"); |
| 2336 | ASSERT (0); |
| 2337 | goto drop; |
| 2338 | } |
| 2339 | } |
| 2340 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2341 | /* |
| 2342 | * Special treatment for CLOSED |
| 2343 | */ |
| 2344 | switch (tc0->state) |
| 2345 | { |
| 2346 | case TCP_STATE_CLOSED: |
| 2347 | goto drop; |
| 2348 | break; |
| 2349 | } |
| 2350 | |
| 2351 | /* |
| 2352 | * For all other states (except LISTEN) |
| 2353 | */ |
| 2354 | |
| 2355 | /* 1-4: check SEQ, RST, SYN */ |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2356 | if (PREDICT_FALSE (tcp_segment_validate (vm, tc0, b0, tcp0, |
| 2357 | &next0))) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2358 | { |
| 2359 | error0 = TCP_ERROR_SEGMENT_INVALID; |
| 2360 | goto drop; |
| 2361 | } |
| 2362 | |
| 2363 | /* 5: check the ACK field */ |
| 2364 | switch (tc0->state) |
| 2365 | { |
| 2366 | case TCP_STATE_SYN_RCVD: |
| 2367 | /* |
| 2368 | * If the segment acknowledgment is not acceptable, form a |
| 2369 | * reset segment, |
| 2370 | * <SEQ=SEG.ACK><CTL=RST> |
| 2371 | * and send it. |
| 2372 | */ |
| 2373 | if (!tcp_rcv_ack_is_acceptable (tc0, b0)) |
| 2374 | { |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2375 | TCP_DBG ("connection not accepted"); |
Florin Coras | 1f152cd | 2017-08-18 19:28:03 -0700 | [diff] [blame] | 2376 | tcp_send_reset_w_pkt (tc0, b0, is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2377 | goto drop; |
| 2378 | } |
Florin Coras | 3af90fc | 2017-05-03 21:09:42 -0700 | [diff] [blame] | 2379 | |
| 2380 | /* Update rtt and rto */ |
Florin Coras | 3af90fc | 2017-05-03 21:09:42 -0700 | [diff] [blame] | 2381 | tcp_update_rtt (tc0, vnet_buffer (b0)->tcp.ack_number); |
| 2382 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2383 | /* Switch state to ESTABLISHED */ |
| 2384 | tc0->state = TCP_STATE_ESTABLISHED; |
Florin Coras | b384b54 | 2018-01-15 01:08:33 -0800 | [diff] [blame^] | 2385 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2386 | |
| 2387 | /* Initialize session variables */ |
| 2388 | tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2389 | tc0->snd_wnd = clib_net_to_host_u16 (tcp0->window) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2390 | << tc0->rcv_opts.wscale; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2391 | tc0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; |
| 2392 | tc0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2393 | |
Florin Coras | ab0289a | 2017-08-14 11:25:25 -0700 | [diff] [blame] | 2394 | /* Reset SYN-ACK retransmit and SYN_RCV establish timers */ |
Florin Coras | 6792ec0 | 2017-03-13 03:49:51 -0700 | [diff] [blame] | 2395 | tcp_retransmit_timer_reset (tc0); |
Florin Coras | ab0289a | 2017-08-14 11:25:25 -0700 | [diff] [blame] | 2396 | tcp_timer_reset (tc0, TCP_TIMER_ESTABLISH); |
Florin Coras | b384b54 | 2018-01-15 01:08:33 -0800 | [diff] [blame^] | 2397 | |
| 2398 | stream_session_accept_notify (&tc0->connection); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2399 | break; |
| 2400 | case TCP_STATE_ESTABLISHED: |
| 2401 | /* We can get packets in established state here because they |
| 2402 | * were enqueued before state change */ |
| 2403 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2404 | goto drop; |
| 2405 | |
| 2406 | break; |
| 2407 | case TCP_STATE_FIN_WAIT_1: |
| 2408 | /* In addition to the processing for the ESTABLISHED state, if |
| 2409 | * our FIN is now acknowledged then enter FIN-WAIT-2 and |
| 2410 | * continue processing in that state. */ |
| 2411 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2412 | goto drop; |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2413 | |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 2414 | /* Still have to send the FIN */ |
| 2415 | if (tc0->flags & TCP_CONN_FINPNDG) |
| 2416 | { |
| 2417 | /* TX fifo finally drained */ |
| 2418 | if (!stream_session_tx_fifo_max_dequeue (&tc0->connection)) |
| 2419 | tcp_send_fin (tc0); |
| 2420 | } |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2421 | /* If FIN is ACKed */ |
Florin Coras | b2215d6 | 2017-08-01 16:56:58 -0700 | [diff] [blame] | 2422 | else if (tc0->snd_una == tc0->snd_una_max) |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2423 | { |
| 2424 | tc0->state = TCP_STATE_FIN_WAIT_2; |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2425 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
| 2426 | |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2427 | /* Stop all retransmit timers because we have nothing more |
| 2428 | * to send. Enable waitclose though because we're willing to |
| 2429 | * wait for peer's FIN but not indefinitely. */ |
| 2430 | tcp_connection_timers_reset (tc0); |
| 2431 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2432 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2433 | break; |
| 2434 | case TCP_STATE_FIN_WAIT_2: |
| 2435 | /* In addition to the processing for the ESTABLISHED state, if |
| 2436 | * the retransmission queue is empty, the user's CLOSE can be |
| 2437 | * acknowledged ("ok") but do not delete the TCB. */ |
| 2438 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2439 | goto drop; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2440 | break; |
| 2441 | case TCP_STATE_CLOSE_WAIT: |
| 2442 | /* Do the same processing as for the ESTABLISHED state. */ |
| 2443 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2444 | goto drop; |
| 2445 | break; |
| 2446 | case TCP_STATE_CLOSING: |
| 2447 | /* In addition to the processing for the ESTABLISHED state, if |
| 2448 | * the ACK acknowledges our FIN then enter the TIME-WAIT state, |
| 2449 | * otherwise ignore the segment. */ |
| 2450 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2451 | goto drop; |
| 2452 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2453 | tc0->state = TCP_STATE_TIME_WAIT; |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2454 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2455 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2456 | goto drop; |
| 2457 | |
| 2458 | break; |
| 2459 | case TCP_STATE_LAST_ACK: |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2460 | /* The only thing that [should] arrive in this state is an |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2461 | * acknowledgment of our FIN. If our FIN is now acknowledged, |
| 2462 | * delete the TCB, enter the CLOSED state, and return. */ |
| 2463 | |
| 2464 | if (!tcp_rcv_ack_is_acceptable (tc0, b0)) |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2465 | { |
| 2466 | error0 = TCP_ERROR_ACK_INVALID; |
| 2467 | goto drop; |
| 2468 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2469 | |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2470 | tc0->snd_una = vnet_buffer (b0)->tcp.ack_number; |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2471 | /* Apparently our ACK for the peer's FIN was lost */ |
| 2472 | if (is_fin0 && tc0->snd_una != tc0->snd_una_max) |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2473 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2474 | tcp_send_fin (tc0); |
| 2475 | goto drop; |
| 2476 | } |
| 2477 | |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2478 | tc0->state = TCP_STATE_CLOSED; |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2479 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2480 | tcp_connection_timers_reset (tc0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2481 | |
| 2482 | /* Don't delete the connection/session yet. Instead, wait a |
| 2483 | * reasonable amount of time until the pipes are cleared. In |
| 2484 | * particular, this makes sure that we won't have dead sessions |
| 2485 | * when processing events on the tx path */ |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2486 | tcp_timer_set (tc0, TCP_TIMER_WAITCLOSE, TCP_CLEANUP_TIME); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2487 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2488 | goto drop; |
| 2489 | |
| 2490 | break; |
| 2491 | case TCP_STATE_TIME_WAIT: |
| 2492 | /* The only thing that can arrive in this state is a |
| 2493 | * retransmission of the remote FIN. Acknowledge it, and restart |
| 2494 | * the 2 MSL timeout. */ |
| 2495 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2496 | if (tcp_rcv_ack (tc0, b0, tcp0, &next0, &error0)) |
| 2497 | goto drop; |
| 2498 | |
| 2499 | tcp_make_ack (tc0, b0); |
Florin Coras | a096f2d | 2017-09-28 23:49:42 -0400 | [diff] [blame] | 2500 | next0 = tcp_next_output (is_ip4); |
| 2501 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2502 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2503 | goto drop; |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2504 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2505 | break; |
| 2506 | default: |
| 2507 | ASSERT (0); |
| 2508 | } |
| 2509 | |
| 2510 | /* 6: check the URG bit TODO */ |
| 2511 | |
| 2512 | /* 7: process the segment text */ |
| 2513 | switch (tc0->state) |
| 2514 | { |
| 2515 | case TCP_STATE_ESTABLISHED: |
| 2516 | case TCP_STATE_FIN_WAIT_1: |
| 2517 | case TCP_STATE_FIN_WAIT_2: |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2518 | if (vnet_buffer (b0)->tcp.data_len) |
| 2519 | error0 = tcp_segment_rcv (tm, tc0, b0, &next0); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2520 | else if (is_fin0) |
| 2521 | tc0->rcv_nxt += 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2522 | break; |
| 2523 | case TCP_STATE_CLOSE_WAIT: |
| 2524 | case TCP_STATE_CLOSING: |
| 2525 | case TCP_STATE_LAST_ACK: |
| 2526 | case TCP_STATE_TIME_WAIT: |
| 2527 | /* This should not occur, since a FIN has been received from the |
| 2528 | * remote side. Ignore the segment text. */ |
| 2529 | break; |
| 2530 | } |
| 2531 | |
| 2532 | /* 8: check the FIN bit */ |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2533 | if (!is_fin0) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2534 | goto drop; |
| 2535 | |
| 2536 | switch (tc0->state) |
| 2537 | { |
| 2538 | case TCP_STATE_ESTABLISHED: |
| 2539 | case TCP_STATE_SYN_RCVD: |
| 2540 | /* Send FIN-ACK notify app and enter CLOSE-WAIT */ |
| 2541 | tcp_connection_timers_reset (tc0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2542 | tcp_make_fin (tc0, b0); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2543 | tc0->snd_nxt += 1; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2544 | next0 = tcp_next_output (tc0->c_is_ip4); |
| 2545 | stream_session_disconnect_notify (&tc0->connection); |
| 2546 | tc0->state = TCP_STATE_CLOSE_WAIT; |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2547 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2548 | break; |
| 2549 | case TCP_STATE_CLOSE_WAIT: |
| 2550 | case TCP_STATE_CLOSING: |
| 2551 | case TCP_STATE_LAST_ACK: |
| 2552 | /* move along .. */ |
| 2553 | break; |
| 2554 | case TCP_STATE_FIN_WAIT_1: |
Florin Coras | ab0289a | 2017-08-14 11:25:25 -0700 | [diff] [blame] | 2555 | tc0->state = TCP_STATE_CLOSING; |
| 2556 | tcp_make_ack (tc0, b0); |
| 2557 | next0 = tcp_next_output (is_ip4); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2558 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Florin Coras | ab0289a | 2017-08-14 11:25:25 -0700 | [diff] [blame] | 2559 | /* Wait for ACK but not forever */ |
| 2560 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_2MSL_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2561 | break; |
| 2562 | case TCP_STATE_FIN_WAIT_2: |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2563 | /* Got FIN, send ACK! Be more aggressive with resource cleanup */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2564 | tc0->state = TCP_STATE_TIME_WAIT; |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 2565 | tcp_connection_timers_reset (tc0); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2566 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2567 | tcp_make_ack (tc0, b0); |
| 2568 | next0 = tcp_next_output (is_ip4); |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2569 | TCP_EVT_DBG (TCP_EVT_STATE_CHANGE, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2570 | break; |
| 2571 | case TCP_STATE_TIME_WAIT: |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2572 | /* Remain in the TIME-WAIT state. Restart the time-wait |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2573 | * timeout. |
| 2574 | */ |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 2575 | tcp_timer_update (tc0, TCP_TIMER_WAITCLOSE, TCP_TIMEWAIT_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2576 | break; |
| 2577 | } |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 2578 | TCP_EVT_DBG (TCP_EVT_FIN_RCVD, tc0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2579 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2580 | drop: |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2581 | b0->error = error0 ? node->errors[error0] : 0; |
| 2582 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2583 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 2584 | { |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2585 | tcp_rx_trace_t *t0 = |
| 2586 | vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 2587 | tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2588 | } |
| 2589 | |
| 2590 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 2591 | n_left_to_next, bi0, next0); |
| 2592 | } |
| 2593 | |
| 2594 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 2595 | } |
| 2596 | |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2597 | errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_TCP, |
| 2598 | my_thread_index); |
| 2599 | tcp_node_inc_counter (vm, is_ip4, tcp4_rcv_process_node.index, |
| 2600 | tcp6_rcv_process_node.index, |
| 2601 | TCP_ERROR_EVENT_FIFO_FULL, errors); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2602 | |
| 2603 | return from_frame->n_vectors; |
| 2604 | } |
| 2605 | |
| 2606 | static uword |
| 2607 | tcp4_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2608 | vlib_frame_t * from_frame) |
| 2609 | { |
| 2610 | return tcp46_rcv_process_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 2611 | } |
| 2612 | |
| 2613 | static uword |
| 2614 | tcp6_rcv_process (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2615 | vlib_frame_t * from_frame) |
| 2616 | { |
| 2617 | return tcp46_rcv_process_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 2618 | } |
| 2619 | |
| 2620 | /* *INDENT-OFF* */ |
| 2621 | VLIB_REGISTER_NODE (tcp4_rcv_process_node) = |
| 2622 | { |
| 2623 | .function = tcp4_rcv_process, |
| 2624 | .name = "tcp4-rcv-process", |
| 2625 | /* Takes a vector of packets. */ |
| 2626 | .vector_size = sizeof (u32), |
| 2627 | .n_errors = TCP_N_ERROR, |
| 2628 | .error_strings = tcp_error_strings, |
| 2629 | .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, |
| 2630 | .next_nodes = |
| 2631 | { |
| 2632 | #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, |
| 2633 | foreach_tcp_state_next |
| 2634 | #undef _ |
| 2635 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2636 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2637 | }; |
| 2638 | /* *INDENT-ON* */ |
| 2639 | |
| 2640 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_rcv_process_node, tcp4_rcv_process); |
| 2641 | |
| 2642 | /* *INDENT-OFF* */ |
| 2643 | VLIB_REGISTER_NODE (tcp6_rcv_process_node) = |
| 2644 | { |
| 2645 | .function = tcp6_rcv_process, |
| 2646 | .name = "tcp6-rcv-process", |
| 2647 | /* Takes a vector of packets. */ |
| 2648 | .vector_size = sizeof (u32), |
| 2649 | .n_errors = TCP_N_ERROR, |
| 2650 | .error_strings = tcp_error_strings, |
| 2651 | .n_next_nodes = TCP_RCV_PROCESS_N_NEXT, |
| 2652 | .next_nodes = |
| 2653 | { |
| 2654 | #define _(s,n) [TCP_RCV_PROCESS_NEXT_##s] = n, |
| 2655 | foreach_tcp_state_next |
| 2656 | #undef _ |
| 2657 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2658 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2659 | }; |
| 2660 | /* *INDENT-ON* */ |
| 2661 | |
| 2662 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_rcv_process_node, tcp6_rcv_process); |
| 2663 | |
| 2664 | vlib_node_registration_t tcp4_listen_node; |
| 2665 | vlib_node_registration_t tcp6_listen_node; |
| 2666 | |
| 2667 | /** |
| 2668 | * LISTEN state processing as per RFC 793 p. 65 |
| 2669 | */ |
| 2670 | always_inline uword |
| 2671 | tcp46_listen_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2672 | vlib_frame_t * from_frame, int is_ip4) |
| 2673 | { |
| 2674 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 2675 | u32 my_thread_index = vm->thread_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2676 | |
| 2677 | from = vlib_frame_vector_args (from_frame); |
| 2678 | n_left_from = from_frame->n_vectors; |
| 2679 | |
| 2680 | next_index = node->cached_next_index; |
| 2681 | |
| 2682 | while (n_left_from > 0) |
| 2683 | { |
| 2684 | u32 n_left_to_next; |
| 2685 | |
| 2686 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 2687 | |
| 2688 | while (n_left_from > 0 && n_left_to_next > 0) |
| 2689 | { |
| 2690 | u32 bi0; |
| 2691 | vlib_buffer_t *b0; |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2692 | tcp_rx_trace_t *t0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2693 | tcp_header_t *th0 = 0; |
| 2694 | tcp_connection_t *lc0; |
| 2695 | ip4_header_t *ip40; |
| 2696 | ip6_header_t *ip60; |
| 2697 | tcp_connection_t *child0; |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2698 | u32 error0 = TCP_ERROR_SYNS_RCVD, next0 = tcp_next_drop (is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2699 | |
| 2700 | bi0 = from[0]; |
| 2701 | to_next[0] = bi0; |
| 2702 | from += 1; |
| 2703 | to_next += 1; |
| 2704 | n_left_from -= 1; |
| 2705 | n_left_to_next -= 1; |
| 2706 | |
| 2707 | b0 = vlib_get_buffer (vm, bi0); |
| 2708 | lc0 = tcp_listener_get (vnet_buffer (b0)->tcp.connection_index); |
| 2709 | |
| 2710 | if (is_ip4) |
| 2711 | { |
| 2712 | ip40 = vlib_buffer_get_current (b0); |
| 2713 | th0 = ip4_next_header (ip40); |
| 2714 | } |
| 2715 | else |
| 2716 | { |
| 2717 | ip60 = vlib_buffer_get_current (b0); |
| 2718 | th0 = ip6_next_header (ip60); |
| 2719 | } |
| 2720 | |
| 2721 | /* Create child session. For syn-flood protection use filter */ |
| 2722 | |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 2723 | /* 1. first check for an RST: handled in dispatch */ |
| 2724 | /* if (tcp_rst (th0)) |
| 2725 | goto drop; */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2726 | |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 2727 | /* 2. second check for an ACK: handled in dispatch */ |
| 2728 | /* if (tcp_ack (th0)) |
| 2729 | { |
| 2730 | tcp_send_reset (b0, is_ip4); |
| 2731 | goto drop; |
| 2732 | } */ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2733 | |
| 2734 | /* 3. check for a SYN (did that already) */ |
| 2735 | |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2736 | /* Make sure connection wasn't just created */ |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2737 | child0 = |
| 2738 | tcp_lookup_connection (lc0->c_fib_index, b0, my_thread_index, |
| 2739 | is_ip4); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2740 | if (PREDICT_FALSE (child0->state != TCP_STATE_LISTEN)) |
| 2741 | { |
| 2742 | error0 = TCP_ERROR_CREATE_EXISTS; |
| 2743 | goto drop; |
| 2744 | } |
| 2745 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2746 | /* Create child session and send SYN-ACK */ |
Florin Coras | 6534b7a | 2017-07-18 05:38:03 -0400 | [diff] [blame] | 2747 | child0 = tcp_connection_new (my_thread_index); |
Florin Coras | 1c71045 | 2017-10-17 00:03:13 -0700 | [diff] [blame] | 2748 | child0->c_lcl_port = th0->dst_port; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2749 | child0->c_rmt_port = th0->src_port; |
| 2750 | child0->c_is_ip4 = is_ip4; |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 2751 | child0->state = TCP_STATE_SYN_RCVD; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2752 | |
| 2753 | if (is_ip4) |
| 2754 | { |
| 2755 | child0->c_lcl_ip4.as_u32 = ip40->dst_address.as_u32; |
| 2756 | child0->c_rmt_ip4.as_u32 = ip40->src_address.as_u32; |
| 2757 | } |
| 2758 | else |
| 2759 | { |
| 2760 | clib_memcpy (&child0->c_lcl_ip6, &ip60->dst_address, |
| 2761 | sizeof (ip6_address_t)); |
| 2762 | clib_memcpy (&child0->c_rmt_ip6, &ip60->src_address, |
| 2763 | sizeof (ip6_address_t)); |
| 2764 | } |
| 2765 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2766 | if (tcp_options_parse (th0, &child0->rcv_opts)) |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 2767 | { |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2768 | clib_warning ("options parse fail"); |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 2769 | goto drop; |
| 2770 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2771 | |
| 2772 | child0->irs = vnet_buffer (b0)->tcp.seq_number; |
| 2773 | child0->rcv_nxt = vnet_buffer (b0)->tcp.seq_number + 1; |
Florin Coras | 3e350af | 2017-03-30 02:54:28 -0700 | [diff] [blame] | 2774 | child0->rcv_las = child0->rcv_nxt; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2775 | |
| 2776 | /* RFC1323: TSval timestamps sent on {SYN} and {SYN,ACK} |
| 2777 | * segments are used to initialize PAWS. */ |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2778 | if (tcp_opts_tstamp (&child0->rcv_opts)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2779 | { |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2780 | child0->tsval_recent = child0->rcv_opts.tsval; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2781 | child0->tsval_recent_age = tcp_time_now (); |
| 2782 | } |
| 2783 | |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 2784 | if (tcp_opts_wscale (&child0->rcv_opts)) |
| 2785 | child0->snd_wscale = child0->rcv_opts.wscale; |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2786 | |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 2787 | child0->snd_wnd = clib_net_to_host_u16 (th0->window) |
| 2788 | << child0->snd_wscale; |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2789 | child0->snd_wl1 = vnet_buffer (b0)->tcp.seq_number; |
| 2790 | child0->snd_wl2 = vnet_buffer (b0)->tcp.ack_number; |
| 2791 | |
| 2792 | tcp_connection_init_vars (child0); |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2793 | TCP_EVT_DBG (TCP_EVT_SYN_RCVD, child0, 1); |
Florin Coras | e69f495 | 2017-03-07 10:06:24 -0800 | [diff] [blame] | 2794 | |
Florin Coras | 6dfb5ac | 2017-11-09 16:26:03 -0800 | [diff] [blame] | 2795 | if (stream_session_accept (&child0->connection, lc0->c_s_index, |
| 2796 | 0 /* notify */ )) |
| 2797 | { |
| 2798 | clib_warning ("session accept fail"); |
| 2799 | tcp_connection_cleanup (child0); |
| 2800 | error0 = TCP_ERROR_CREATE_SESSION_FAIL; |
| 2801 | goto drop; |
| 2802 | } |
| 2803 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2804 | /* Reuse buffer to make syn-ack and send */ |
| 2805 | tcp_make_synack (child0, b0); |
| 2806 | next0 = tcp_next_output (is_ip4); |
Florin Coras | ab0289a | 2017-08-14 11:25:25 -0700 | [diff] [blame] | 2807 | tcp_timer_set (child0, TCP_TIMER_ESTABLISH, TCP_SYN_RCVD_TIME); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2808 | |
| 2809 | drop: |
| 2810 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 2811 | { |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2812 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
| 2813 | clib_memcpy (&t0->tcp_header, th0, sizeof (t0->tcp_header)); |
| 2814 | clib_memcpy (&t0->tcp_connection, lc0, |
| 2815 | sizeof (t0->tcp_connection)); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2816 | } |
| 2817 | |
Florin Coras | e04c299 | 2017-03-01 08:17:34 -0800 | [diff] [blame] | 2818 | b0->error = node->errors[error0]; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2819 | |
| 2820 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 2821 | n_left_to_next, bi0, next0); |
| 2822 | } |
| 2823 | |
| 2824 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 2825 | } |
| 2826 | return from_frame->n_vectors; |
| 2827 | } |
| 2828 | |
| 2829 | static uword |
| 2830 | tcp4_listen (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2831 | vlib_frame_t * from_frame) |
| 2832 | { |
| 2833 | return tcp46_listen_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 2834 | } |
| 2835 | |
| 2836 | static uword |
| 2837 | tcp6_listen (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2838 | vlib_frame_t * from_frame) |
| 2839 | { |
| 2840 | return tcp46_listen_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 2841 | } |
| 2842 | |
| 2843 | /* *INDENT-OFF* */ |
| 2844 | VLIB_REGISTER_NODE (tcp4_listen_node) = |
| 2845 | { |
| 2846 | .function = tcp4_listen, |
| 2847 | .name = "tcp4-listen", |
| 2848 | /* Takes a vector of packets. */ |
| 2849 | .vector_size = sizeof (u32), |
| 2850 | .n_errors = TCP_N_ERROR, |
| 2851 | .error_strings = tcp_error_strings, |
| 2852 | .n_next_nodes = TCP_LISTEN_N_NEXT, |
| 2853 | .next_nodes = |
| 2854 | { |
| 2855 | #define _(s,n) [TCP_LISTEN_NEXT_##s] = n, |
| 2856 | foreach_tcp_state_next |
| 2857 | #undef _ |
| 2858 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2859 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2860 | }; |
| 2861 | /* *INDENT-ON* */ |
| 2862 | |
| 2863 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_listen_node, tcp4_listen); |
| 2864 | |
| 2865 | /* *INDENT-OFF* */ |
| 2866 | VLIB_REGISTER_NODE (tcp6_listen_node) = |
| 2867 | { |
| 2868 | .function = tcp6_listen, |
| 2869 | .name = "tcp6-listen", |
| 2870 | /* Takes a vector of packets. */ |
| 2871 | .vector_size = sizeof (u32), |
| 2872 | .n_errors = TCP_N_ERROR, |
| 2873 | .error_strings = tcp_error_strings, |
| 2874 | .n_next_nodes = TCP_LISTEN_N_NEXT, |
| 2875 | .next_nodes = |
| 2876 | { |
| 2877 | #define _(s,n) [TCP_LISTEN_NEXT_##s] = n, |
| 2878 | foreach_tcp_state_next |
| 2879 | #undef _ |
| 2880 | }, |
Clement Durand | 6cf260c | 2017-04-13 13:27:04 +0200 | [diff] [blame] | 2881 | .format_trace = format_tcp_rx_trace_short, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2882 | }; |
| 2883 | /* *INDENT-ON* */ |
| 2884 | |
| 2885 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_listen_node, tcp6_listen); |
| 2886 | |
| 2887 | vlib_node_registration_t tcp4_input_node; |
| 2888 | vlib_node_registration_t tcp6_input_node; |
| 2889 | |
| 2890 | typedef enum _tcp_input_next |
| 2891 | { |
| 2892 | TCP_INPUT_NEXT_DROP, |
| 2893 | TCP_INPUT_NEXT_LISTEN, |
| 2894 | TCP_INPUT_NEXT_RCV_PROCESS, |
| 2895 | TCP_INPUT_NEXT_SYN_SENT, |
| 2896 | TCP_INPUT_NEXT_ESTABLISHED, |
| 2897 | TCP_INPUT_NEXT_RESET, |
Pierre Pfister | 7fe51f3 | 2017-09-20 08:48:36 +0200 | [diff] [blame] | 2898 | TCP_INPUT_NEXT_PUNT, |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2899 | TCP_INPUT_N_NEXT |
| 2900 | } tcp_input_next_t; |
| 2901 | |
| 2902 | #define foreach_tcp4_input_next \ |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2903 | _ (DROP, "ip4-drop") \ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2904 | _ (LISTEN, "tcp4-listen") \ |
| 2905 | _ (RCV_PROCESS, "tcp4-rcv-process") \ |
| 2906 | _ (SYN_SENT, "tcp4-syn-sent") \ |
| 2907 | _ (ESTABLISHED, "tcp4-established") \ |
Pierre Pfister | 7fe51f3 | 2017-09-20 08:48:36 +0200 | [diff] [blame] | 2908 | _ (RESET, "tcp4-reset") \ |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2909 | _ (PUNT, "ip4-punt") |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2910 | |
| 2911 | #define foreach_tcp6_input_next \ |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2912 | _ (DROP, "ip6-drop") \ |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2913 | _ (LISTEN, "tcp6-listen") \ |
| 2914 | _ (RCV_PROCESS, "tcp6-rcv-process") \ |
| 2915 | _ (SYN_SENT, "tcp6-syn-sent") \ |
| 2916 | _ (ESTABLISHED, "tcp6-established") \ |
Pierre Pfister | 7fe51f3 | 2017-09-20 08:48:36 +0200 | [diff] [blame] | 2917 | _ (RESET, "tcp6-reset") \ |
Vijayabhaskar Katamreddy | ce07412 | 2017-11-15 13:50:26 -0800 | [diff] [blame] | 2918 | _ (PUNT, "ip6-punt") |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2919 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2920 | #define filter_flags (TCP_FLAG_SYN|TCP_FLAG_ACK|TCP_FLAG_RST|TCP_FLAG_FIN) |
| 2921 | |
| 2922 | always_inline uword |
| 2923 | tcp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 2924 | vlib_frame_t * from_frame, int is_ip4) |
| 2925 | { |
| 2926 | u32 n_left_from, next_index, *from, *to_next; |
Damjan Marion | 586afd7 | 2017-04-05 19:18:20 +0200 | [diff] [blame] | 2927 | u32 my_thread_index = vm->thread_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2928 | tcp_main_t *tm = vnet_get_tcp_main (); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2929 | |
| 2930 | from = vlib_frame_vector_args (from_frame); |
| 2931 | n_left_from = from_frame->n_vectors; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2932 | next_index = node->cached_next_index; |
Florin Coras | 82d3ec8 | 2017-08-14 08:10:42 -0700 | [diff] [blame] | 2933 | tcp_set_time_now (my_thread_index); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2934 | |
| 2935 | while (n_left_from > 0) |
| 2936 | { |
| 2937 | u32 n_left_to_next; |
| 2938 | |
| 2939 | vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); |
| 2940 | |
| 2941 | while (n_left_from > 0 && n_left_to_next > 0) |
| 2942 | { |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2943 | int n_advance_bytes0, n_data_bytes0; |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2944 | u32 bi0, fib_index0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2945 | vlib_buffer_t *b0; |
| 2946 | tcp_header_t *tcp0 = 0; |
| 2947 | tcp_connection_t *tc0; |
Florin Coras | 4eeeaaf | 2017-09-05 14:03:37 -0400 | [diff] [blame] | 2948 | transport_connection_t *tconn; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2949 | ip4_header_t *ip40; |
| 2950 | ip6_header_t *ip60; |
| 2951 | u32 error0 = TCP_ERROR_NO_LISTENER, next0 = TCP_INPUT_NEXT_DROP; |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 2952 | u8 flags0, is_filtered = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2953 | |
| 2954 | bi0 = from[0]; |
| 2955 | to_next[0] = bi0; |
| 2956 | from += 1; |
| 2957 | to_next += 1; |
| 2958 | n_left_from -= 1; |
| 2959 | n_left_to_next -= 1; |
| 2960 | |
| 2961 | b0 = vlib_get_buffer (vm, bi0); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 2962 | vnet_buffer (b0)->tcp.flags = 0; |
Florin Coras | cea194d | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2963 | fib_index0 = vnet_buffer (b0)->ip.fib_index; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2964 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2965 | /* Checksum computed by ipx_local no need to compute again */ |
| 2966 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2967 | if (is_ip4) |
| 2968 | { |
| 2969 | ip40 = vlib_buffer_get_current (b0); |
| 2970 | tcp0 = ip4_next_header (ip40); |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2971 | n_advance_bytes0 = (ip4_header_bytes (ip40) |
| 2972 | + tcp_header_bytes (tcp0)); |
| 2973 | n_data_bytes0 = clib_net_to_host_u16 (ip40->length) |
| 2974 | - n_advance_bytes0; |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2975 | tconn = session_lookup_connection_wt4 (fib_index0, |
| 2976 | &ip40->dst_address, |
| 2977 | &ip40->src_address, |
| 2978 | tcp0->dst_port, |
| 2979 | tcp0->src_port, |
| 2980 | TRANSPORT_PROTO_TCP, |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 2981 | my_thread_index, |
| 2982 | &is_filtered); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 2983 | } |
| 2984 | else |
| 2985 | { |
| 2986 | ip60 = vlib_buffer_get_current (b0); |
| 2987 | tcp0 = ip6_next_header (ip60); |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 2988 | n_advance_bytes0 = tcp_header_bytes (tcp0); |
| 2989 | n_data_bytes0 = clib_net_to_host_u16 (ip60->payload_length) |
| 2990 | - n_advance_bytes0; |
| 2991 | n_advance_bytes0 += sizeof (ip60[0]); |
Florin Coras | 3cbc04b | 2017-10-02 00:18:51 -0700 | [diff] [blame] | 2992 | tconn = session_lookup_connection_wt6 (fib_index0, |
| 2993 | &ip60->dst_address, |
| 2994 | &ip60->src_address, |
| 2995 | tcp0->dst_port, |
| 2996 | tcp0->src_port, |
| 2997 | TRANSPORT_PROTO_TCP, |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 2998 | my_thread_index, |
| 2999 | &is_filtered); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3000 | } |
| 3001 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 3002 | /* Length check */ |
| 3003 | if (PREDICT_FALSE (n_advance_bytes0 < 0)) |
| 3004 | { |
| 3005 | error0 = TCP_ERROR_LENGTH; |
| 3006 | goto done; |
| 3007 | } |
| 3008 | |
Florin Coras | f988e69 | 2017-11-27 04:34:14 -0500 | [diff] [blame] | 3009 | vnet_buffer (b0)->tcp.hdr_offset = (u8 *) tcp0 |
| 3010 | - (u8 *) vlib_buffer_get_current (b0); |
| 3011 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3012 | /* Session exists */ |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 3013 | if (PREDICT_TRUE (0 != tconn)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3014 | { |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 3015 | tc0 = tcp_get_connection_from_transport (tconn); |
| 3016 | ASSERT (tcp_lookup_is_valid (tc0, tcp0)); |
| 3017 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3018 | /* Save connection index */ |
| 3019 | vnet_buffer (b0)->tcp.connection_index = tc0->c_c_index; |
| 3020 | vnet_buffer (b0)->tcp.seq_number = |
| 3021 | clib_net_to_host_u32 (tcp0->seq_number); |
| 3022 | vnet_buffer (b0)->tcp.ack_number = |
| 3023 | clib_net_to_host_u32 (tcp0->ack_number); |
| 3024 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 3025 | vnet_buffer (b0)->tcp.data_offset = n_advance_bytes0; |
| 3026 | vnet_buffer (b0)->tcp.data_len = n_data_bytes0; |
| 3027 | |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3028 | flags0 = tcp0->flags & filter_flags; |
| 3029 | next0 = tm->dispatch_table[tc0->state][flags0].next; |
| 3030 | error0 = tm->dispatch_table[tc0->state][flags0].error; |
| 3031 | |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3032 | if (PREDICT_FALSE (error0 == TCP_ERROR_DISPATCH |
| 3033 | || next0 == TCP_INPUT_NEXT_RESET)) |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3034 | { |
| 3035 | /* Overload tcp flags to store state */ |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3036 | tcp_state_t state0 = tc0->state; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3037 | vnet_buffer (b0)->tcp.flags = tc0->state; |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3038 | |
| 3039 | if (error0 == TCP_ERROR_DISPATCH) |
| 3040 | clib_warning ("disp error state %U flags %U", |
Florin Coras | bb292f4 | 2017-05-19 09:49:19 -0700 | [diff] [blame] | 3041 | format_tcp_state, state0, format_tcp_flags, |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3042 | (int) flags0); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3043 | } |
| 3044 | } |
| 3045 | else |
| 3046 | { |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 3047 | if (is_filtered) |
| 3048 | { |
| 3049 | next0 = TCP_INPUT_NEXT_DROP; |
| 3050 | error0 = TCP_ERROR_FILTERED; |
| 3051 | } |
| 3052 | else if ((is_ip4 && tm->punt_unknown4) || |
| 3053 | (!is_ip4 && tm->punt_unknown6)) |
Pierre Pfister | 7fe51f3 | 2017-09-20 08:48:36 +0200 | [diff] [blame] | 3054 | { |
| 3055 | next0 = TCP_INPUT_NEXT_PUNT; |
| 3056 | error0 = TCP_ERROR_PUNT; |
| 3057 | } |
| 3058 | else |
| 3059 | { |
| 3060 | /* Send reset */ |
| 3061 | next0 = TCP_INPUT_NEXT_RESET; |
| 3062 | error0 = TCP_ERROR_NO_LISTENER; |
| 3063 | } |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 3064 | tc0 = 0; |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3065 | } |
| 3066 | |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 3067 | done: |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3068 | b0->error = error0 ? node->errors[error0] : 0; |
| 3069 | |
| 3070 | if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED)) |
| 3071 | { |
Florin Coras | dff48db | 2017-11-19 18:06:58 -0800 | [diff] [blame] | 3072 | tcp_rx_trace_t *t0; |
| 3073 | t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); |
Florin Coras | 82b13a8 | 2017-04-25 11:58:06 -0700 | [diff] [blame] | 3074 | tcp_set_rx_trace_data (t0, tc0, tcp0, b0, is_ip4); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3075 | } |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3076 | vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, |
| 3077 | n_left_to_next, bi0, next0); |
| 3078 | } |
| 3079 | |
| 3080 | vlib_put_next_frame (vm, node, next_index, n_left_to_next); |
| 3081 | } |
| 3082 | |
| 3083 | return from_frame->n_vectors; |
| 3084 | } |
| 3085 | |
| 3086 | static uword |
| 3087 | tcp4_input (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 3088 | vlib_frame_t * from_frame) |
| 3089 | { |
| 3090 | return tcp46_input_inline (vm, node, from_frame, 1 /* is_ip4 */ ); |
| 3091 | } |
| 3092 | |
| 3093 | static uword |
| 3094 | tcp6_input (vlib_main_t * vm, vlib_node_runtime_t * node, |
| 3095 | vlib_frame_t * from_frame) |
| 3096 | { |
| 3097 | return tcp46_input_inline (vm, node, from_frame, 0 /* is_ip4 */ ); |
| 3098 | } |
| 3099 | |
| 3100 | /* *INDENT-OFF* */ |
| 3101 | VLIB_REGISTER_NODE (tcp4_input_node) = |
| 3102 | { |
| 3103 | .function = tcp4_input, |
| 3104 | .name = "tcp4-input", |
| 3105 | /* Takes a vector of packets. */ |
| 3106 | .vector_size = sizeof (u32), |
| 3107 | .n_errors = TCP_N_ERROR, |
| 3108 | .error_strings = tcp_error_strings, |
| 3109 | .n_next_nodes = TCP_INPUT_N_NEXT, |
| 3110 | .next_nodes = |
| 3111 | { |
| 3112 | #define _(s,n) [TCP_INPUT_NEXT_##s] = n, |
| 3113 | foreach_tcp4_input_next |
| 3114 | #undef _ |
| 3115 | }, |
| 3116 | .format_buffer = format_tcp_header, |
| 3117 | .format_trace = format_tcp_rx_trace, |
| 3118 | }; |
| 3119 | /* *INDENT-ON* */ |
| 3120 | |
| 3121 | VLIB_NODE_FUNCTION_MULTIARCH (tcp4_input_node, tcp4_input); |
| 3122 | |
| 3123 | /* *INDENT-OFF* */ |
| 3124 | VLIB_REGISTER_NODE (tcp6_input_node) = |
| 3125 | { |
| 3126 | .function = tcp6_input, |
| 3127 | .name = "tcp6-input", |
| 3128 | /* Takes a vector of packets. */ |
| 3129 | .vector_size = sizeof (u32), |
| 3130 | .n_errors = TCP_N_ERROR, |
| 3131 | .error_strings = tcp_error_strings, |
| 3132 | .n_next_nodes = TCP_INPUT_N_NEXT, |
| 3133 | .next_nodes = |
| 3134 | { |
| 3135 | #define _(s,n) [TCP_INPUT_NEXT_##s] = n, |
| 3136 | foreach_tcp6_input_next |
| 3137 | #undef _ |
| 3138 | }, |
| 3139 | .format_buffer = format_tcp_header, |
| 3140 | .format_trace = format_tcp_rx_trace, |
| 3141 | }; |
| 3142 | /* *INDENT-ON* */ |
| 3143 | |
| 3144 | VLIB_NODE_FUNCTION_MULTIARCH (tcp6_input_node, tcp6_input); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3145 | |
| 3146 | static void |
| 3147 | tcp_dispatch_table_init (tcp_main_t * tm) |
| 3148 | { |
| 3149 | int i, j; |
| 3150 | for (i = 0; i < ARRAY_LEN (tm->dispatch_table); i++) |
| 3151 | for (j = 0; j < ARRAY_LEN (tm->dispatch_table[i]); j++) |
| 3152 | { |
| 3153 | tm->dispatch_table[i][j].next = TCP_INPUT_NEXT_DROP; |
| 3154 | tm->dispatch_table[i][j].error = TCP_ERROR_DISPATCH; |
| 3155 | } |
| 3156 | |
| 3157 | #define _(t,f,n,e) \ |
| 3158 | do { \ |
| 3159 | tm->dispatch_table[TCP_STATE_##t][f].next = (n); \ |
| 3160 | tm->dispatch_table[TCP_STATE_##t][f].error = (e); \ |
| 3161 | } while (0) |
| 3162 | |
| 3163 | /* SYNs for new connections -> tcp-listen. */ |
| 3164 | _(LISTEN, TCP_FLAG_SYN, TCP_INPUT_NEXT_LISTEN, TCP_ERROR_NONE); |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3165 | _(LISTEN, TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, TCP_ERROR_NONE); |
| 3166 | _(LISTEN, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_NONE); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 3167 | _(LISTEN, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RESET, |
| 3168 | TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3169 | /* ACK for for a SYN-ACK -> tcp-rcv-process. */ |
| 3170 | _(SYN_RCVD, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | c28764f | 2017-04-26 00:08:42 -0700 | [diff] [blame] | 3171 | _(SYN_RCVD, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 3172 | _(SYN_RCVD, TCP_FLAG_SYN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3173 | /* SYN-ACK for a SYN */ |
| 3174 | _(SYN_SENT, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, |
| 3175 | TCP_ERROR_NONE); |
| 3176 | _(SYN_SENT, TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); |
| 3177 | _(SYN_SENT, TCP_FLAG_RST, TCP_INPUT_NEXT_SYN_SENT, TCP_ERROR_NONE); |
| 3178 | _(SYN_SENT, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_SYN_SENT, |
| 3179 | TCP_ERROR_NONE); |
| 3180 | /* ACK for for established connection -> tcp-established. */ |
| 3181 | _(ESTABLISHED, TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
| 3182 | /* FIN for for established connection -> tcp-established. */ |
| 3183 | _(ESTABLISHED, TCP_FLAG_FIN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
| 3184 | _(ESTABLISHED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, |
| 3185 | TCP_ERROR_NONE); |
Florin Coras | d79b41e | 2017-03-04 05:37:52 -0800 | [diff] [blame] | 3186 | _(ESTABLISHED, TCP_FLAG_RST, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3187 | _(ESTABLISHED, TCP_FLAG_RST | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, |
| 3188 | TCP_ERROR_NONE); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 3189 | _(ESTABLISHED, TCP_FLAG_SYN, TCP_INPUT_NEXT_ESTABLISHED, TCP_ERROR_NONE); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 3190 | _(ESTABLISHED, TCP_FLAG_SYN | TCP_FLAG_ACK, TCP_INPUT_NEXT_ESTABLISHED, |
| 3191 | TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3192 | /* ACK or FIN-ACK to our FIN */ |
| 3193 | _(FIN_WAIT_1, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 3194 | _(FIN_WAIT_1, TCP_FLAG_ACK | TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, |
| 3195 | TCP_ERROR_NONE); |
| 3196 | /* FIN in reply to our FIN from the other side */ |
| 3197 | _(FIN_WAIT_1, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 3198 | _(FIN_WAIT_1, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3199 | /* FIN confirming that the peer (app) has closed */ |
| 3200 | _(FIN_WAIT_2, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | f6d68ed | 2017-05-07 19:12:02 -0700 | [diff] [blame] | 3201 | _(FIN_WAIT_2, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3202 | _(FIN_WAIT_2, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, |
| 3203 | TCP_ERROR_NONE); |
Florin Coras | f03a59a | 2017-06-09 21:07:32 -0700 | [diff] [blame] | 3204 | _(CLOSE_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 3205 | _(CLOSE_WAIT, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, |
| 3206 | TCP_ERROR_NONE); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3207 | _(LAST_ACK, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 3208 | _(LAST_ACK, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 3209 | _(LAST_ACK, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, |
| 3210 | TCP_ERROR_NONE); |
Florin Coras | db84e57 | 2017-05-09 18:54:52 -0700 | [diff] [blame] | 3211 | _(LAST_ACK, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | 93992a9 | 2017-05-24 18:03:56 -0700 | [diff] [blame] | 3212 | _(TIME_WAIT, TCP_FLAG_FIN, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
| 3213 | _(TIME_WAIT, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, |
| 3214 | TCP_ERROR_NONE); |
Florin Coras | 3eb5062 | 2017-07-13 01:24:57 -0400 | [diff] [blame] | 3215 | _(TIME_WAIT, TCP_FLAG_RST, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | 5095895 | 2017-08-29 14:50:13 -0700 | [diff] [blame] | 3216 | _(TIME_WAIT, TCP_FLAG_ACK, TCP_INPUT_NEXT_RCV_PROCESS, TCP_ERROR_NONE); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 3217 | _(CLOSED, TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED); |
Florin Coras | dc629cd | 2017-05-09 00:52:37 -0700 | [diff] [blame] | 3218 | _(CLOSED, TCP_FLAG_RST, TCP_INPUT_NEXT_DROP, TCP_ERROR_CONNECTION_CLOSED); |
Florin Coras | 9d06304 | 2017-09-14 03:08:00 -0400 | [diff] [blame] | 3219 | _(CLOSED, TCP_FLAG_FIN | TCP_FLAG_ACK, TCP_INPUT_NEXT_DROP, |
Dave Barach | 2c25a62 | 2017-06-26 11:35:07 -0400 | [diff] [blame] | 3220 | TCP_ERROR_CONNECTION_CLOSED); |
Dave Barach | 68b0fb0 | 2017-02-28 15:15:56 -0500 | [diff] [blame] | 3221 | #undef _ |
| 3222 | } |
| 3223 | |
| 3224 | clib_error_t * |
| 3225 | tcp_input_init (vlib_main_t * vm) |
| 3226 | { |
| 3227 | clib_error_t *error = 0; |
| 3228 | tcp_main_t *tm = vnet_get_tcp_main (); |
| 3229 | |
| 3230 | if ((error = vlib_call_init_function (vm, tcp_init))) |
| 3231 | return error; |
| 3232 | |
| 3233 | /* Initialize dispatch table. */ |
| 3234 | tcp_dispatch_table_init (tm); |
| 3235 | |
| 3236 | return error; |
| 3237 | } |
| 3238 | |
| 3239 | VLIB_INIT_FUNCTION (tcp_input_init); |
| 3240 | |
| 3241 | /* |
| 3242 | * fd.io coding-style-patch-verification: ON |
| 3243 | * |
| 3244 | * Local Variables: |
| 3245 | * eval: (c-set-style "gnu") |
| 3246 | * End: |
| 3247 | */ |