1 /* SCTP kernel implementation 2 * (C) Copyright IBM Corp. 2001, 2004 3 * Copyright (c) 1999-2000 Cisco, Inc. 4 * Copyright (c) 1999-2001 Motorola, Inc. 5 * Copyright (c) 2001-2003 Intel Corp. 6 * 7 * This file is part of the SCTP kernel implementation 8 * 9 * These functions implement the sctp_outq class. The outqueue handles 10 * bundling and queueing of outgoing SCTP chunks. 11 * 12 * This SCTP implementation is free software; 13 * you can redistribute it and/or modify it under the terms of 14 * the GNU General Public License as published by 15 * the Free Software Foundation; either version 2, or (at your option) 16 * any later version. 17 * 18 * This SCTP implementation is distributed in the hope that it 19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied 20 * ************************ 21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. 22 * See the GNU General Public License for more details. 23 * 24 * You should have received a copy of the GNU General Public License 25 * along with GNU CC; see the file COPYING. If not, see 26 * <http://www.gnu.org/licenses/>. 27 * 28 * Please send any bug reports or fixes you make to the 29 * email address(es): 30 * lksctp developers <linux-sctp@vger.kernel.org> 31 * 32 * Written or modified by: 33 * La Monte H.P. Yarroll <piggy@acm.org> 34 * Karl Knutson <karl@athena.chicago.il.us> 35 * Perry Melange <pmelange@null.cc.uic.edu> 36 * Xingang Guo <xingang.guo@intel.com> 37 * Hui Huang <hui.huang@nokia.com> 38 * Sridhar Samudrala <sri@us.ibm.com> 39 * Jon Grimm <jgrimm@us.ibm.com> 40 */ 41 42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 43 44 #include <linux/types.h> 45 #include <linux/list.h> /* For struct list_head */ 46 #include <linux/socket.h> 47 #include <linux/ip.h> 48 #include <linux/slab.h> 49 #include <net/sock.h> /* For skb_set_owner_w */ 50 51 #include <net/sctp/sctp.h> 52 #include <net/sctp/sm.h> 53 #include <net/sctp/stream_sched.h> 54 55 /* Declare internal functions here. */ 56 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 57 static void sctp_check_transmitted(struct sctp_outq *q, 58 struct list_head *transmitted_queue, 59 struct sctp_transport *transport, 60 union sctp_addr *saddr, 61 struct sctp_sackhdr *sack, 62 __u32 *highest_new_tsn); 63 64 static void sctp_mark_missing(struct sctp_outq *q, 65 struct list_head *transmitted_queue, 66 struct sctp_transport *transport, 67 __u32 highest_new_tsn, 68 int count_of_newacks); 69 70 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn); 71 72 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp); 73 74 /* Add data to the front of the queue. */ 75 static inline void sctp_outq_head_data(struct sctp_outq *q, 76 struct sctp_chunk *ch) 77 { 78 struct sctp_stream_out_ext *oute; 79 __u16 stream; 80 81 list_add(&ch->list, &q->out_chunk_list); 82 q->out_qlen += ch->skb->len; 83 84 stream = sctp_chunk_stream_no(ch); 85 oute = q->asoc->stream.out[stream].ext; 86 list_add(&ch->stream_list, &oute->outq); 87 } 88 89 /* Take data from the front of the queue. */ 90 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 91 { 92 return q->sched->dequeue(q); 93 } 94 95 /* Add data chunk to the end of the queue. */ 96 static inline void sctp_outq_tail_data(struct sctp_outq *q, 97 struct sctp_chunk *ch) 98 { 99 struct sctp_stream_out_ext *oute; 100 __u16 stream; 101 102 list_add_tail(&ch->list, &q->out_chunk_list); 103 q->out_qlen += ch->skb->len; 104 105 stream = sctp_chunk_stream_no(ch); 106 oute = q->asoc->stream.out[stream].ext; 107 list_add_tail(&ch->stream_list, &oute->outq); 108 } 109 110 /* 111 * SFR-CACC algorithm: 112 * D) If count_of_newacks is greater than or equal to 2 113 * and t was not sent to the current primary then the 114 * sender MUST NOT increment missing report count for t. 115 */ 116 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 117 struct sctp_transport *transport, 118 int count_of_newacks) 119 { 120 if (count_of_newacks >= 2 && transport != primary) 121 return 1; 122 return 0; 123 } 124 125 /* 126 * SFR-CACC algorithm: 127 * F) If count_of_newacks is less than 2, let d be the 128 * destination to which t was sent. If cacc_saw_newack 129 * is 0 for destination d, then the sender MUST NOT 130 * increment missing report count for t. 131 */ 132 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 133 int count_of_newacks) 134 { 135 if (count_of_newacks < 2 && 136 (transport && !transport->cacc.cacc_saw_newack)) 137 return 1; 138 return 0; 139 } 140 141 /* 142 * SFR-CACC algorithm: 143 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 144 * execute steps C, D, F. 145 * 146 * C has been implemented in sctp_outq_sack 147 */ 148 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 149 struct sctp_transport *transport, 150 int count_of_newacks) 151 { 152 if (!primary->cacc.cycling_changeover) { 153 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 154 return 1; 155 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 156 return 1; 157 return 0; 158 } 159 return 0; 160 } 161 162 /* 163 * SFR-CACC algorithm: 164 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 165 * than next_tsn_at_change of the current primary, then 166 * the sender MUST NOT increment missing report count 167 * for t. 168 */ 169 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 170 { 171 if (primary->cacc.cycling_changeover && 172 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 173 return 1; 174 return 0; 175 } 176 177 /* 178 * SFR-CACC algorithm: 179 * 3) If the missing report count for TSN t is to be 180 * incremented according to [RFC2960] and 181 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 182 * then the sender MUST further execute steps 3.1 and 183 * 3.2 to determine if the missing report count for 184 * TSN t SHOULD NOT be incremented. 185 * 186 * 3.3) If 3.1 and 3.2 do not dictate that the missing 187 * report count for t should not be incremented, then 188 * the sender SHOULD increment missing report count for 189 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 190 */ 191 static inline int sctp_cacc_skip(struct sctp_transport *primary, 192 struct sctp_transport *transport, 193 int count_of_newacks, 194 __u32 tsn) 195 { 196 if (primary->cacc.changeover_active && 197 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 198 sctp_cacc_skip_3_2(primary, tsn))) 199 return 1; 200 return 0; 201 } 202 203 /* Initialize an existing sctp_outq. This does the boring stuff. 204 * You still need to define handlers if you really want to DO 205 * something with this structure... 206 */ 207 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 208 { 209 memset(q, 0, sizeof(struct sctp_outq)); 210 211 q->asoc = asoc; 212 INIT_LIST_HEAD(&q->out_chunk_list); 213 INIT_LIST_HEAD(&q->control_chunk_list); 214 INIT_LIST_HEAD(&q->retransmit); 215 INIT_LIST_HEAD(&q->sacked); 216 INIT_LIST_HEAD(&q->abandoned); 217 sctp_sched_set_sched(asoc, SCTP_SS_FCFS); 218 } 219 220 /* Free the outqueue structure and any related pending chunks. 221 */ 222 static void __sctp_outq_teardown(struct sctp_outq *q) 223 { 224 struct sctp_transport *transport; 225 struct list_head *lchunk, *temp; 226 struct sctp_chunk *chunk, *tmp; 227 228 /* Throw away unacknowledged chunks. */ 229 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 230 transports) { 231 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 232 chunk = list_entry(lchunk, struct sctp_chunk, 233 transmitted_list); 234 /* Mark as part of a failed message. */ 235 sctp_chunk_fail(chunk, q->error); 236 sctp_chunk_free(chunk); 237 } 238 } 239 240 /* Throw away chunks that have been gap ACKed. */ 241 list_for_each_safe(lchunk, temp, &q->sacked) { 242 list_del_init(lchunk); 243 chunk = list_entry(lchunk, struct sctp_chunk, 244 transmitted_list); 245 sctp_chunk_fail(chunk, q->error); 246 sctp_chunk_free(chunk); 247 } 248 249 /* Throw away any chunks in the retransmit queue. */ 250 list_for_each_safe(lchunk, temp, &q->retransmit) { 251 list_del_init(lchunk); 252 chunk = list_entry(lchunk, struct sctp_chunk, 253 transmitted_list); 254 sctp_chunk_fail(chunk, q->error); 255 sctp_chunk_free(chunk); 256 } 257 258 /* Throw away any chunks that are in the abandoned queue. */ 259 list_for_each_safe(lchunk, temp, &q->abandoned) { 260 list_del_init(lchunk); 261 chunk = list_entry(lchunk, struct sctp_chunk, 262 transmitted_list); 263 sctp_chunk_fail(chunk, q->error); 264 sctp_chunk_free(chunk); 265 } 266 267 /* Throw away any leftover data chunks. */ 268 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 269 sctp_sched_dequeue_done(q, chunk); 270 271 /* Mark as send failure. */ 272 sctp_chunk_fail(chunk, q->error); 273 sctp_chunk_free(chunk); 274 } 275 276 /* Throw away any leftover control chunks. */ 277 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 278 list_del_init(&chunk->list); 279 sctp_chunk_free(chunk); 280 } 281 } 282 283 void sctp_outq_teardown(struct sctp_outq *q) 284 { 285 __sctp_outq_teardown(q); 286 sctp_outq_init(q->asoc, q); 287 } 288 289 /* Free the outqueue structure and any related pending chunks. */ 290 void sctp_outq_free(struct sctp_outq *q) 291 { 292 /* Throw away leftover chunks. */ 293 __sctp_outq_teardown(q); 294 } 295 296 /* Put a new chunk in an sctp_outq. */ 297 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp) 298 { 299 struct net *net = sock_net(q->asoc->base.sk); 300 301 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk, 302 chunk && chunk->chunk_hdr ? 303 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 304 "illegal chunk"); 305 306 /* If it is data, queue it up, otherwise, send it 307 * immediately. 308 */ 309 if (sctp_chunk_is_data(chunk)) { 310 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n", 311 __func__, q, chunk, chunk && chunk->chunk_hdr ? 312 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 313 "illegal chunk"); 314 315 sctp_outq_tail_data(q, chunk); 316 if (chunk->asoc->peer.prsctp_capable && 317 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 318 chunk->asoc->sent_cnt_removable++; 319 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 320 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS); 321 else 322 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS); 323 } else { 324 list_add_tail(&chunk->list, &q->control_chunk_list); 325 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS); 326 } 327 328 if (!q->cork) 329 sctp_outq_flush(q, 0, gfp); 330 } 331 332 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 333 * and the abandoned list are in ascending order. 334 */ 335 static void sctp_insert_list(struct list_head *head, struct list_head *new) 336 { 337 struct list_head *pos; 338 struct sctp_chunk *nchunk, *lchunk; 339 __u32 ntsn, ltsn; 340 int done = 0; 341 342 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 343 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 344 345 list_for_each(pos, head) { 346 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 347 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 348 if (TSN_lt(ntsn, ltsn)) { 349 list_add(new, pos->prev); 350 done = 1; 351 break; 352 } 353 } 354 if (!done) 355 list_add_tail(new, head); 356 } 357 358 static int sctp_prsctp_prune_sent(struct sctp_association *asoc, 359 struct sctp_sndrcvinfo *sinfo, 360 struct list_head *queue, int msg_len) 361 { 362 struct sctp_chunk *chk, *temp; 363 364 list_for_each_entry_safe(chk, temp, queue, transmitted_list) { 365 struct sctp_stream_out *streamout; 366 367 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 368 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive) 369 continue; 370 371 list_del_init(&chk->transmitted_list); 372 sctp_insert_list(&asoc->outqueue.abandoned, 373 &chk->transmitted_list); 374 375 streamout = &asoc->stream.out[chk->sinfo.sinfo_stream]; 376 asoc->sent_cnt_removable--; 377 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 378 streamout->ext->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 379 380 if (!chk->tsn_gap_acked) { 381 if (chk->transport) 382 chk->transport->flight_size -= 383 sctp_data_size(chk); 384 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk); 385 } 386 387 msg_len -= SCTP_DATA_SNDSIZE(chk) + 388 sizeof(struct sk_buff) + 389 sizeof(struct sctp_chunk); 390 if (msg_len <= 0) 391 break; 392 } 393 394 return msg_len; 395 } 396 397 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc, 398 struct sctp_sndrcvinfo *sinfo, int msg_len) 399 { 400 struct sctp_outq *q = &asoc->outqueue; 401 struct sctp_chunk *chk, *temp; 402 403 q->sched->unsched_all(&asoc->stream); 404 405 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) { 406 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 407 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive) 408 continue; 409 410 sctp_sched_dequeue_common(q, chk); 411 asoc->sent_cnt_removable--; 412 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 413 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) { 414 struct sctp_stream_out *streamout = 415 &asoc->stream.out[chk->sinfo.sinfo_stream]; 416 417 streamout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 418 } 419 420 msg_len -= SCTP_DATA_SNDSIZE(chk) + 421 sizeof(struct sk_buff) + 422 sizeof(struct sctp_chunk); 423 sctp_chunk_free(chk); 424 if (msg_len <= 0) 425 break; 426 } 427 428 q->sched->sched_all(&asoc->stream); 429 430 return msg_len; 431 } 432 433 /* Abandon the chunks according their priorities */ 434 void sctp_prsctp_prune(struct sctp_association *asoc, 435 struct sctp_sndrcvinfo *sinfo, int msg_len) 436 { 437 struct sctp_transport *transport; 438 439 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable) 440 return; 441 442 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 443 &asoc->outqueue.retransmit, 444 msg_len); 445 if (msg_len <= 0) 446 return; 447 448 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 449 transports) { 450 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 451 &transport->transmitted, 452 msg_len); 453 if (msg_len <= 0) 454 return; 455 } 456 457 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len); 458 } 459 460 /* Mark all the eligible packets on a transport for retransmission. */ 461 void sctp_retransmit_mark(struct sctp_outq *q, 462 struct sctp_transport *transport, 463 __u8 reason) 464 { 465 struct list_head *lchunk, *ltemp; 466 struct sctp_chunk *chunk; 467 468 /* Walk through the specified transmitted queue. */ 469 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 470 chunk = list_entry(lchunk, struct sctp_chunk, 471 transmitted_list); 472 473 /* If the chunk is abandoned, move it to abandoned list. */ 474 if (sctp_chunk_abandoned(chunk)) { 475 list_del_init(lchunk); 476 sctp_insert_list(&q->abandoned, lchunk); 477 478 /* If this chunk has not been previousely acked, 479 * stop considering it 'outstanding'. Our peer 480 * will most likely never see it since it will 481 * not be retransmitted 482 */ 483 if (!chunk->tsn_gap_acked) { 484 if (chunk->transport) 485 chunk->transport->flight_size -= 486 sctp_data_size(chunk); 487 q->outstanding_bytes -= sctp_data_size(chunk); 488 q->asoc->peer.rwnd += sctp_data_size(chunk); 489 } 490 continue; 491 } 492 493 /* If we are doing retransmission due to a timeout or pmtu 494 * discovery, only the chunks that are not yet acked should 495 * be added to the retransmit queue. 496 */ 497 if ((reason == SCTP_RTXR_FAST_RTX && 498 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 499 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 500 /* RFC 2960 6.2.1 Processing a Received SACK 501 * 502 * C) Any time a DATA chunk is marked for 503 * retransmission (via either T3-rtx timer expiration 504 * (Section 6.3.3) or via fast retransmit 505 * (Section 7.2.4)), add the data size of those 506 * chunks to the rwnd. 507 */ 508 q->asoc->peer.rwnd += sctp_data_size(chunk); 509 q->outstanding_bytes -= sctp_data_size(chunk); 510 if (chunk->transport) 511 transport->flight_size -= sctp_data_size(chunk); 512 513 /* sctpimpguide-05 Section 2.8.2 514 * M5) If a T3-rtx timer expires, the 515 * 'TSN.Missing.Report' of all affected TSNs is set 516 * to 0. 517 */ 518 chunk->tsn_missing_report = 0; 519 520 /* If a chunk that is being used for RTT measurement 521 * has to be retransmitted, we cannot use this chunk 522 * anymore for RTT measurements. Reset rto_pending so 523 * that a new RTT measurement is started when a new 524 * data chunk is sent. 525 */ 526 if (chunk->rtt_in_progress) { 527 chunk->rtt_in_progress = 0; 528 transport->rto_pending = 0; 529 } 530 531 /* Move the chunk to the retransmit queue. The chunks 532 * on the retransmit queue are always kept in order. 533 */ 534 list_del_init(lchunk); 535 sctp_insert_list(&q->retransmit, lchunk); 536 } 537 } 538 539 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, " 540 "flight_size:%d, pba:%d\n", __func__, transport, reason, 541 transport->cwnd, transport->ssthresh, transport->flight_size, 542 transport->partial_bytes_acked); 543 } 544 545 /* Mark all the eligible packets on a transport for retransmission and force 546 * one packet out. 547 */ 548 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 549 enum sctp_retransmit_reason reason) 550 { 551 struct net *net = sock_net(q->asoc->base.sk); 552 553 switch (reason) { 554 case SCTP_RTXR_T3_RTX: 555 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 556 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 557 /* Update the retran path if the T3-rtx timer has expired for 558 * the current retran path. 559 */ 560 if (transport == transport->asoc->peer.retran_path) 561 sctp_assoc_update_retran_path(transport->asoc); 562 transport->asoc->rtx_data_chunks += 563 transport->asoc->unack_data; 564 break; 565 case SCTP_RTXR_FAST_RTX: 566 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 567 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 568 q->fast_rtx = 1; 569 break; 570 case SCTP_RTXR_PMTUD: 571 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 572 break; 573 case SCTP_RTXR_T1_RTX: 574 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 575 transport->asoc->init_retries++; 576 break; 577 default: 578 BUG(); 579 } 580 581 sctp_retransmit_mark(q, transport, reason); 582 583 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 584 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 585 * following the procedures outlined in C1 - C5. 586 */ 587 if (reason == SCTP_RTXR_T3_RTX) 588 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point); 589 590 /* Flush the queues only on timeout, since fast_rtx is only 591 * triggered during sack processing and the queue 592 * will be flushed at the end. 593 */ 594 if (reason != SCTP_RTXR_FAST_RTX) 595 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC); 596 } 597 598 /* 599 * Transmit DATA chunks on the retransmit queue. Upon return from 600 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 601 * need to be transmitted by the caller. 602 * We assume that pkt->transport has already been set. 603 * 604 * The return value is a normal kernel error return value. 605 */ 606 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 607 int rtx_timeout, int *start_timer) 608 { 609 struct sctp_transport *transport = pkt->transport; 610 struct sctp_chunk *chunk, *chunk1; 611 struct list_head *lqueue; 612 enum sctp_xmit status; 613 int error = 0; 614 int timer = 0; 615 int done = 0; 616 int fast_rtx; 617 618 lqueue = &q->retransmit; 619 fast_rtx = q->fast_rtx; 620 621 /* This loop handles time-out retransmissions, fast retransmissions, 622 * and retransmissions due to opening of whindow. 623 * 624 * RFC 2960 6.3.3 Handle T3-rtx Expiration 625 * 626 * E3) Determine how many of the earliest (i.e., lowest TSN) 627 * outstanding DATA chunks for the address for which the 628 * T3-rtx has expired will fit into a single packet, subject 629 * to the MTU constraint for the path corresponding to the 630 * destination transport address to which the retransmission 631 * is being sent (this may be different from the address for 632 * which the timer expires [see Section 6.4]). Call this value 633 * K. Bundle and retransmit those K DATA chunks in a single 634 * packet to the destination endpoint. 635 * 636 * [Just to be painfully clear, if we are retransmitting 637 * because a timeout just happened, we should send only ONE 638 * packet of retransmitted data.] 639 * 640 * For fast retransmissions we also send only ONE packet. However, 641 * if we are just flushing the queue due to open window, we'll 642 * try to send as much as possible. 643 */ 644 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 645 /* If the chunk is abandoned, move it to abandoned list. */ 646 if (sctp_chunk_abandoned(chunk)) { 647 list_del_init(&chunk->transmitted_list); 648 sctp_insert_list(&q->abandoned, 649 &chunk->transmitted_list); 650 continue; 651 } 652 653 /* Make sure that Gap Acked TSNs are not retransmitted. A 654 * simple approach is just to move such TSNs out of the 655 * way and into a 'transmitted' queue and skip to the 656 * next chunk. 657 */ 658 if (chunk->tsn_gap_acked) { 659 list_move_tail(&chunk->transmitted_list, 660 &transport->transmitted); 661 continue; 662 } 663 664 /* If we are doing fast retransmit, ignore non-fast_rtransmit 665 * chunks 666 */ 667 if (fast_rtx && !chunk->fast_retransmit) 668 continue; 669 670 redo: 671 /* Attempt to append this chunk to the packet. */ 672 status = sctp_packet_append_chunk(pkt, chunk); 673 674 switch (status) { 675 case SCTP_XMIT_PMTU_FULL: 676 if (!pkt->has_data && !pkt->has_cookie_echo) { 677 /* If this packet did not contain DATA then 678 * retransmission did not happen, so do it 679 * again. We'll ignore the error here since 680 * control chunks are already freed so there 681 * is nothing we can do. 682 */ 683 sctp_packet_transmit(pkt, GFP_ATOMIC); 684 goto redo; 685 } 686 687 /* Send this packet. */ 688 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 689 690 /* If we are retransmitting, we should only 691 * send a single packet. 692 * Otherwise, try appending this chunk again. 693 */ 694 if (rtx_timeout || fast_rtx) 695 done = 1; 696 else 697 goto redo; 698 699 /* Bundle next chunk in the next round. */ 700 break; 701 702 case SCTP_XMIT_RWND_FULL: 703 /* Send this packet. */ 704 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 705 706 /* Stop sending DATA as there is no more room 707 * at the receiver. 708 */ 709 done = 1; 710 break; 711 712 case SCTP_XMIT_DELAY: 713 /* Send this packet. */ 714 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 715 716 /* Stop sending DATA because of nagle delay. */ 717 done = 1; 718 break; 719 720 default: 721 /* The append was successful, so add this chunk to 722 * the transmitted list. 723 */ 724 list_move_tail(&chunk->transmitted_list, 725 &transport->transmitted); 726 727 /* Mark the chunk as ineligible for fast retransmit 728 * after it is retransmitted. 729 */ 730 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 731 chunk->fast_retransmit = SCTP_DONT_FRTX; 732 733 q->asoc->stats.rtxchunks++; 734 break; 735 } 736 737 /* Set the timer if there were no errors */ 738 if (!error && !timer) 739 timer = 1; 740 741 if (done) 742 break; 743 } 744 745 /* If we are here due to a retransmit timeout or a fast 746 * retransmit and if there are any chunks left in the retransmit 747 * queue that could not fit in the PMTU sized packet, they need 748 * to be marked as ineligible for a subsequent fast retransmit. 749 */ 750 if (rtx_timeout || fast_rtx) { 751 list_for_each_entry(chunk1, lqueue, transmitted_list) { 752 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 753 chunk1->fast_retransmit = SCTP_DONT_FRTX; 754 } 755 } 756 757 *start_timer = timer; 758 759 /* Clear fast retransmit hint */ 760 if (fast_rtx) 761 q->fast_rtx = 0; 762 763 return error; 764 } 765 766 /* Cork the outqueue so queued chunks are really queued. */ 767 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp) 768 { 769 if (q->cork) 770 q->cork = 0; 771 772 sctp_outq_flush(q, 0, gfp); 773 } 774 775 776 /* 777 * Try to flush an outqueue. 778 * 779 * Description: Send everything in q which we legally can, subject to 780 * congestion limitations. 781 * * Note: This function can be called from multiple contexts so appropriate 782 * locking concerns must be made. Today we use the sock lock to protect 783 * this function. 784 */ 785 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp) 786 { 787 struct sctp_packet *packet; 788 struct sctp_packet singleton; 789 struct sctp_association *asoc = q->asoc; 790 __u16 sport = asoc->base.bind_addr.port; 791 __u16 dport = asoc->peer.port; 792 __u32 vtag = asoc->peer.i.init_tag; 793 struct sctp_transport *transport = NULL; 794 struct sctp_transport *new_transport; 795 struct sctp_chunk *chunk, *tmp; 796 enum sctp_xmit status; 797 int error = 0; 798 int start_timer = 0; 799 int one_packet = 0; 800 801 /* These transports have chunks to send. */ 802 struct list_head transport_list; 803 struct list_head *ltransport; 804 805 INIT_LIST_HEAD(&transport_list); 806 packet = NULL; 807 808 /* 809 * 6.10 Bundling 810 * ... 811 * When bundling control chunks with DATA chunks, an 812 * endpoint MUST place control chunks first in the outbound 813 * SCTP packet. The transmitter MUST transmit DATA chunks 814 * within a SCTP packet in increasing order of TSN. 815 * ... 816 */ 817 818 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 819 /* RFC 5061, 5.3 820 * F1) This means that until such time as the ASCONF 821 * containing the add is acknowledged, the sender MUST 822 * NOT use the new IP address as a source for ANY SCTP 823 * packet except on carrying an ASCONF Chunk. 824 */ 825 if (asoc->src_out_of_asoc_ok && 826 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 827 continue; 828 829 list_del_init(&chunk->list); 830 831 /* Pick the right transport to use. */ 832 new_transport = chunk->transport; 833 834 if (!new_transport) { 835 /* 836 * If we have a prior transport pointer, see if 837 * the destination address of the chunk 838 * matches the destination address of the 839 * current transport. If not a match, then 840 * try to look up the transport with a given 841 * destination address. We do this because 842 * after processing ASCONFs, we may have new 843 * transports created. 844 */ 845 if (transport && 846 sctp_cmp_addr_exact(&chunk->dest, 847 &transport->ipaddr)) 848 new_transport = transport; 849 else 850 new_transport = sctp_assoc_lookup_paddr(asoc, 851 &chunk->dest); 852 853 /* if we still don't have a new transport, then 854 * use the current active path. 855 */ 856 if (!new_transport) 857 new_transport = asoc->peer.active_path; 858 } else if ((new_transport->state == SCTP_INACTIVE) || 859 (new_transport->state == SCTP_UNCONFIRMED) || 860 (new_transport->state == SCTP_PF)) { 861 /* If the chunk is Heartbeat or Heartbeat Ack, 862 * send it to chunk->transport, even if it's 863 * inactive. 864 * 865 * 3.3.6 Heartbeat Acknowledgement: 866 * ... 867 * A HEARTBEAT ACK is always sent to the source IP 868 * address of the IP datagram containing the 869 * HEARTBEAT chunk to which this ack is responding. 870 * ... 871 * 872 * ASCONF_ACKs also must be sent to the source. 873 */ 874 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT && 875 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK && 876 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK) 877 new_transport = asoc->peer.active_path; 878 } 879 880 /* Are we switching transports? 881 * Take care of transport locks. 882 */ 883 if (new_transport != transport) { 884 transport = new_transport; 885 if (list_empty(&transport->send_ready)) { 886 list_add_tail(&transport->send_ready, 887 &transport_list); 888 } 889 packet = &transport->packet; 890 sctp_packet_config(packet, vtag, 891 asoc->peer.ecn_capable); 892 } 893 894 switch (chunk->chunk_hdr->type) { 895 /* 896 * 6.10 Bundling 897 * ... 898 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 899 * COMPLETE with any other chunks. [Send them immediately.] 900 */ 901 case SCTP_CID_INIT: 902 case SCTP_CID_INIT_ACK: 903 case SCTP_CID_SHUTDOWN_COMPLETE: 904 sctp_packet_init(&singleton, transport, sport, dport); 905 sctp_packet_config(&singleton, vtag, 0); 906 sctp_packet_append_chunk(&singleton, chunk); 907 error = sctp_packet_transmit(&singleton, gfp); 908 if (error < 0) { 909 asoc->base.sk->sk_err = -error; 910 return; 911 } 912 break; 913 914 case SCTP_CID_ABORT: 915 if (sctp_test_T_bit(chunk)) { 916 packet->vtag = asoc->c.my_vtag; 917 } 918 /* The following chunks are "response" chunks, i.e. 919 * they are generated in response to something we 920 * received. If we are sending these, then we can 921 * send only 1 packet containing these chunks. 922 */ 923 case SCTP_CID_HEARTBEAT_ACK: 924 case SCTP_CID_SHUTDOWN_ACK: 925 case SCTP_CID_COOKIE_ACK: 926 case SCTP_CID_COOKIE_ECHO: 927 case SCTP_CID_ERROR: 928 case SCTP_CID_ECN_CWR: 929 case SCTP_CID_ASCONF_ACK: 930 one_packet = 1; 931 /* Fall through */ 932 933 case SCTP_CID_SACK: 934 case SCTP_CID_HEARTBEAT: 935 case SCTP_CID_SHUTDOWN: 936 case SCTP_CID_ECN_ECNE: 937 case SCTP_CID_ASCONF: 938 case SCTP_CID_FWD_TSN: 939 case SCTP_CID_RECONF: 940 status = sctp_packet_transmit_chunk(packet, chunk, 941 one_packet, gfp); 942 if (status != SCTP_XMIT_OK) { 943 /* put the chunk back */ 944 list_add(&chunk->list, &q->control_chunk_list); 945 break; 946 } 947 948 asoc->stats.octrlchunks++; 949 /* PR-SCTP C5) If a FORWARD TSN is sent, the 950 * sender MUST assure that at least one T3-rtx 951 * timer is running. 952 */ 953 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) { 954 sctp_transport_reset_t3_rtx(transport); 955 transport->last_time_sent = jiffies; 956 } 957 958 if (chunk == asoc->strreset_chunk) 959 sctp_transport_reset_reconf_timer(transport); 960 961 break; 962 963 default: 964 /* We built a chunk with an illegal type! */ 965 BUG(); 966 } 967 } 968 969 if (q->asoc->src_out_of_asoc_ok) 970 goto sctp_flush_out; 971 972 /* Is it OK to send data chunks? */ 973 switch (asoc->state) { 974 case SCTP_STATE_COOKIE_ECHOED: 975 /* Only allow bundling when this packet has a COOKIE-ECHO 976 * chunk. 977 */ 978 if (!packet || !packet->has_cookie_echo) 979 break; 980 981 /* fallthru */ 982 case SCTP_STATE_ESTABLISHED: 983 case SCTP_STATE_SHUTDOWN_PENDING: 984 case SCTP_STATE_SHUTDOWN_RECEIVED: 985 /* 986 * RFC 2960 6.1 Transmission of DATA Chunks 987 * 988 * C) When the time comes for the sender to transmit, 989 * before sending new DATA chunks, the sender MUST 990 * first transmit any outstanding DATA chunks which 991 * are marked for retransmission (limited by the 992 * current cwnd). 993 */ 994 if (!list_empty(&q->retransmit)) { 995 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 996 goto sctp_flush_out; 997 if (transport == asoc->peer.retran_path) 998 goto retran; 999 1000 /* Switch transports & prepare the packet. */ 1001 1002 transport = asoc->peer.retran_path; 1003 1004 if (list_empty(&transport->send_ready)) { 1005 list_add_tail(&transport->send_ready, 1006 &transport_list); 1007 } 1008 1009 packet = &transport->packet; 1010 sctp_packet_config(packet, vtag, 1011 asoc->peer.ecn_capable); 1012 retran: 1013 error = sctp_outq_flush_rtx(q, packet, 1014 rtx_timeout, &start_timer); 1015 if (error < 0) 1016 asoc->base.sk->sk_err = -error; 1017 1018 if (start_timer) { 1019 sctp_transport_reset_t3_rtx(transport); 1020 transport->last_time_sent = jiffies; 1021 } 1022 1023 /* This can happen on COOKIE-ECHO resend. Only 1024 * one chunk can get bundled with a COOKIE-ECHO. 1025 */ 1026 if (packet->has_cookie_echo) 1027 goto sctp_flush_out; 1028 1029 /* Don't send new data if there is still data 1030 * waiting to retransmit. 1031 */ 1032 if (!list_empty(&q->retransmit)) 1033 goto sctp_flush_out; 1034 } 1035 1036 /* Apply Max.Burst limitation to the current transport in 1037 * case it will be used for new data. We are going to 1038 * rest it before we return, but we want to apply the limit 1039 * to the currently queued data. 1040 */ 1041 if (transport) 1042 sctp_transport_burst_limited(transport); 1043 1044 /* Finally, transmit new packets. */ 1045 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 1046 __u32 sid = ntohs(chunk->subh.data_hdr->stream); 1047 1048 /* Has this chunk expired? */ 1049 if (sctp_chunk_abandoned(chunk)) { 1050 sctp_sched_dequeue_done(q, chunk); 1051 sctp_chunk_fail(chunk, 0); 1052 sctp_chunk_free(chunk); 1053 continue; 1054 } 1055 1056 if (asoc->stream.out[sid].state == SCTP_STREAM_CLOSED) { 1057 sctp_outq_head_data(q, chunk); 1058 goto sctp_flush_out; 1059 } 1060 1061 /* If there is a specified transport, use it. 1062 * Otherwise, we want to use the active path. 1063 */ 1064 new_transport = chunk->transport; 1065 if (!new_transport || 1066 ((new_transport->state == SCTP_INACTIVE) || 1067 (new_transport->state == SCTP_UNCONFIRMED) || 1068 (new_transport->state == SCTP_PF))) 1069 new_transport = asoc->peer.active_path; 1070 if (new_transport->state == SCTP_UNCONFIRMED) { 1071 WARN_ONCE(1, "Attempt to send packet on unconfirmed path."); 1072 sctp_sched_dequeue_done(q, chunk); 1073 sctp_chunk_fail(chunk, 0); 1074 sctp_chunk_free(chunk); 1075 continue; 1076 } 1077 1078 /* Change packets if necessary. */ 1079 if (new_transport != transport) { 1080 transport = new_transport; 1081 1082 /* Schedule to have this transport's 1083 * packet flushed. 1084 */ 1085 if (list_empty(&transport->send_ready)) { 1086 list_add_tail(&transport->send_ready, 1087 &transport_list); 1088 } 1089 1090 packet = &transport->packet; 1091 sctp_packet_config(packet, vtag, 1092 asoc->peer.ecn_capable); 1093 /* We've switched transports, so apply the 1094 * Burst limit to the new transport. 1095 */ 1096 sctp_transport_burst_limited(transport); 1097 } 1098 1099 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p " 1100 "skb->users:%d\n", 1101 __func__, q, chunk, chunk && chunk->chunk_hdr ? 1102 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1103 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1104 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1105 refcount_read(&chunk->skb->users) : -1); 1106 1107 /* Add the chunk to the packet. */ 1108 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp); 1109 1110 switch (status) { 1111 case SCTP_XMIT_PMTU_FULL: 1112 case SCTP_XMIT_RWND_FULL: 1113 case SCTP_XMIT_DELAY: 1114 /* We could not append this chunk, so put 1115 * the chunk back on the output queue. 1116 */ 1117 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1118 __func__, ntohl(chunk->subh.data_hdr->tsn), 1119 status); 1120 1121 sctp_outq_head_data(q, chunk); 1122 goto sctp_flush_out; 1123 1124 case SCTP_XMIT_OK: 1125 /* The sender is in the SHUTDOWN-PENDING state, 1126 * The sender MAY set the I-bit in the DATA 1127 * chunk header. 1128 */ 1129 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1130 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1131 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1132 asoc->stats.ouodchunks++; 1133 else 1134 asoc->stats.oodchunks++; 1135 1136 /* Only now it's safe to consider this 1137 * chunk as sent, sched-wise. 1138 */ 1139 sctp_sched_dequeue_done(q, chunk); 1140 1141 break; 1142 1143 default: 1144 BUG(); 1145 } 1146 1147 /* BUG: We assume that the sctp_packet_transmit() 1148 * call below will succeed all the time and add the 1149 * chunk to the transmitted list and restart the 1150 * timers. 1151 * It is possible that the call can fail under OOM 1152 * conditions. 1153 * 1154 * Is this really a problem? Won't this behave 1155 * like a lost TSN? 1156 */ 1157 list_add_tail(&chunk->transmitted_list, 1158 &transport->transmitted); 1159 1160 sctp_transport_reset_t3_rtx(transport); 1161 transport->last_time_sent = jiffies; 1162 1163 /* Only let one DATA chunk get bundled with a 1164 * COOKIE-ECHO chunk. 1165 */ 1166 if (packet->has_cookie_echo) 1167 goto sctp_flush_out; 1168 } 1169 break; 1170 1171 default: 1172 /* Do nothing. */ 1173 break; 1174 } 1175 1176 sctp_flush_out: 1177 1178 /* Before returning, examine all the transports touched in 1179 * this call. Right now, we bluntly force clear all the 1180 * transports. Things might change after we implement Nagle. 1181 * But such an examination is still required. 1182 * 1183 * --xguo 1184 */ 1185 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) { 1186 struct sctp_transport *t = list_entry(ltransport, 1187 struct sctp_transport, 1188 send_ready); 1189 packet = &t->packet; 1190 if (!sctp_packet_empty(packet)) { 1191 error = sctp_packet_transmit(packet, gfp); 1192 if (error < 0) 1193 asoc->base.sk->sk_err = -error; 1194 } 1195 1196 /* Clear the burst limited state, if any */ 1197 sctp_transport_burst_reset(t); 1198 } 1199 } 1200 1201 /* Update unack_data based on the incoming SACK chunk */ 1202 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1203 struct sctp_sackhdr *sack) 1204 { 1205 union sctp_sack_variable *frags; 1206 __u16 unack_data; 1207 int i; 1208 1209 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1210 1211 frags = sack->variable; 1212 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1213 unack_data -= ((ntohs(frags[i].gab.end) - 1214 ntohs(frags[i].gab.start) + 1)); 1215 } 1216 1217 assoc->unack_data = unack_data; 1218 } 1219 1220 /* This is where we REALLY process a SACK. 1221 * 1222 * Process the SACK against the outqueue. Mostly, this just frees 1223 * things off the transmitted queue. 1224 */ 1225 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1226 { 1227 struct sctp_association *asoc = q->asoc; 1228 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1229 struct sctp_transport *transport; 1230 struct sctp_chunk *tchunk = NULL; 1231 struct list_head *lchunk, *transport_list, *temp; 1232 union sctp_sack_variable *frags = sack->variable; 1233 __u32 sack_ctsn, ctsn, tsn; 1234 __u32 highest_tsn, highest_new_tsn; 1235 __u32 sack_a_rwnd; 1236 unsigned int outstanding; 1237 struct sctp_transport *primary = asoc->peer.primary_path; 1238 int count_of_newacks = 0; 1239 int gap_ack_blocks; 1240 u8 accum_moved = 0; 1241 1242 /* Grab the association's destination address list. */ 1243 transport_list = &asoc->peer.transport_addr_list; 1244 1245 sack_ctsn = ntohl(sack->cum_tsn_ack); 1246 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1247 asoc->stats.gapcnt += gap_ack_blocks; 1248 /* 1249 * SFR-CACC algorithm: 1250 * On receipt of a SACK the sender SHOULD execute the 1251 * following statements. 1252 * 1253 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1254 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1255 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1256 * all destinations. 1257 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1258 * is set the receiver of the SACK MUST take the following actions: 1259 * 1260 * A) Initialize the cacc_saw_newack to 0 for all destination 1261 * addresses. 1262 * 1263 * Only bother if changeover_active is set. Otherwise, this is 1264 * totally suboptimal to do on every SACK. 1265 */ 1266 if (primary->cacc.changeover_active) { 1267 u8 clear_cycling = 0; 1268 1269 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1270 primary->cacc.changeover_active = 0; 1271 clear_cycling = 1; 1272 } 1273 1274 if (clear_cycling || gap_ack_blocks) { 1275 list_for_each_entry(transport, transport_list, 1276 transports) { 1277 if (clear_cycling) 1278 transport->cacc.cycling_changeover = 0; 1279 if (gap_ack_blocks) 1280 transport->cacc.cacc_saw_newack = 0; 1281 } 1282 } 1283 } 1284 1285 /* Get the highest TSN in the sack. */ 1286 highest_tsn = sack_ctsn; 1287 if (gap_ack_blocks) 1288 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1289 1290 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1291 asoc->highest_sacked = highest_tsn; 1292 1293 highest_new_tsn = sack_ctsn; 1294 1295 /* Run through the retransmit queue. Credit bytes received 1296 * and free those chunks that we can. 1297 */ 1298 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1299 1300 /* Run through the transmitted queue. 1301 * Credit bytes received and free those chunks which we can. 1302 * 1303 * This is a MASSIVE candidate for optimization. 1304 */ 1305 list_for_each_entry(transport, transport_list, transports) { 1306 sctp_check_transmitted(q, &transport->transmitted, 1307 transport, &chunk->source, sack, 1308 &highest_new_tsn); 1309 /* 1310 * SFR-CACC algorithm: 1311 * C) Let count_of_newacks be the number of 1312 * destinations for which cacc_saw_newack is set. 1313 */ 1314 if (transport->cacc.cacc_saw_newack) 1315 count_of_newacks++; 1316 } 1317 1318 /* Move the Cumulative TSN Ack Point if appropriate. */ 1319 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1320 asoc->ctsn_ack_point = sack_ctsn; 1321 accum_moved = 1; 1322 } 1323 1324 if (gap_ack_blocks) { 1325 1326 if (asoc->fast_recovery && accum_moved) 1327 highest_new_tsn = highest_tsn; 1328 1329 list_for_each_entry(transport, transport_list, transports) 1330 sctp_mark_missing(q, &transport->transmitted, transport, 1331 highest_new_tsn, count_of_newacks); 1332 } 1333 1334 /* Update unack_data field in the assoc. */ 1335 sctp_sack_update_unack_data(asoc, sack); 1336 1337 ctsn = asoc->ctsn_ack_point; 1338 1339 /* Throw away stuff rotting on the sack queue. */ 1340 list_for_each_safe(lchunk, temp, &q->sacked) { 1341 tchunk = list_entry(lchunk, struct sctp_chunk, 1342 transmitted_list); 1343 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1344 if (TSN_lte(tsn, ctsn)) { 1345 list_del_init(&tchunk->transmitted_list); 1346 if (asoc->peer.prsctp_capable && 1347 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1348 asoc->sent_cnt_removable--; 1349 sctp_chunk_free(tchunk); 1350 } 1351 } 1352 1353 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1354 * number of bytes still outstanding after processing the 1355 * Cumulative TSN Ack and the Gap Ack Blocks. 1356 */ 1357 1358 sack_a_rwnd = ntohl(sack->a_rwnd); 1359 asoc->peer.zero_window_announced = !sack_a_rwnd; 1360 outstanding = q->outstanding_bytes; 1361 1362 if (outstanding < sack_a_rwnd) 1363 sack_a_rwnd -= outstanding; 1364 else 1365 sack_a_rwnd = 0; 1366 1367 asoc->peer.rwnd = sack_a_rwnd; 1368 1369 sctp_generate_fwdtsn(q, sack_ctsn); 1370 1371 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1372 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1373 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1374 asoc->adv_peer_ack_point); 1375 1376 return sctp_outq_is_empty(q); 1377 } 1378 1379 /* Is the outqueue empty? 1380 * The queue is empty when we have not pending data, no in-flight data 1381 * and nothing pending retransmissions. 1382 */ 1383 int sctp_outq_is_empty(const struct sctp_outq *q) 1384 { 1385 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1386 list_empty(&q->retransmit); 1387 } 1388 1389 /******************************************************************** 1390 * 2nd Level Abstractions 1391 ********************************************************************/ 1392 1393 /* Go through a transport's transmitted list or the association's retransmit 1394 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1395 * The retransmit list will not have an associated transport. 1396 * 1397 * I added coherent debug information output. --xguo 1398 * 1399 * Instead of printing 'sacked' or 'kept' for each TSN on the 1400 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1401 * KEPT TSN6-TSN7, etc. 1402 */ 1403 static void sctp_check_transmitted(struct sctp_outq *q, 1404 struct list_head *transmitted_queue, 1405 struct sctp_transport *transport, 1406 union sctp_addr *saddr, 1407 struct sctp_sackhdr *sack, 1408 __u32 *highest_new_tsn_in_sack) 1409 { 1410 struct list_head *lchunk; 1411 struct sctp_chunk *tchunk; 1412 struct list_head tlist; 1413 __u32 tsn; 1414 __u32 sack_ctsn; 1415 __u32 rtt; 1416 __u8 restart_timer = 0; 1417 int bytes_acked = 0; 1418 int migrate_bytes = 0; 1419 bool forward_progress = false; 1420 1421 sack_ctsn = ntohl(sack->cum_tsn_ack); 1422 1423 INIT_LIST_HEAD(&tlist); 1424 1425 /* The while loop will skip empty transmitted queues. */ 1426 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1427 tchunk = list_entry(lchunk, struct sctp_chunk, 1428 transmitted_list); 1429 1430 if (sctp_chunk_abandoned(tchunk)) { 1431 /* Move the chunk to abandoned list. */ 1432 sctp_insert_list(&q->abandoned, lchunk); 1433 1434 /* If this chunk has not been acked, stop 1435 * considering it as 'outstanding'. 1436 */ 1437 if (!tchunk->tsn_gap_acked) { 1438 if (tchunk->transport) 1439 tchunk->transport->flight_size -= 1440 sctp_data_size(tchunk); 1441 q->outstanding_bytes -= sctp_data_size(tchunk); 1442 } 1443 continue; 1444 } 1445 1446 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1447 if (sctp_acked(sack, tsn)) { 1448 /* If this queue is the retransmit queue, the 1449 * retransmit timer has already reclaimed 1450 * the outstanding bytes for this chunk, so only 1451 * count bytes associated with a transport. 1452 */ 1453 if (transport) { 1454 /* If this chunk is being used for RTT 1455 * measurement, calculate the RTT and update 1456 * the RTO using this value. 1457 * 1458 * 6.3.1 C5) Karn's algorithm: RTT measurements 1459 * MUST NOT be made using packets that were 1460 * retransmitted (and thus for which it is 1461 * ambiguous whether the reply was for the 1462 * first instance of the packet or a later 1463 * instance). 1464 */ 1465 if (!tchunk->tsn_gap_acked && 1466 !sctp_chunk_retransmitted(tchunk) && 1467 tchunk->rtt_in_progress) { 1468 tchunk->rtt_in_progress = 0; 1469 rtt = jiffies - tchunk->sent_at; 1470 sctp_transport_update_rto(transport, 1471 rtt); 1472 } 1473 } 1474 1475 /* If the chunk hasn't been marked as ACKED, 1476 * mark it and account bytes_acked if the 1477 * chunk had a valid transport (it will not 1478 * have a transport if ASCONF had deleted it 1479 * while DATA was outstanding). 1480 */ 1481 if (!tchunk->tsn_gap_acked) { 1482 tchunk->tsn_gap_acked = 1; 1483 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1484 *highest_new_tsn_in_sack = tsn; 1485 bytes_acked += sctp_data_size(tchunk); 1486 if (!tchunk->transport) 1487 migrate_bytes += sctp_data_size(tchunk); 1488 forward_progress = true; 1489 } 1490 1491 if (TSN_lte(tsn, sack_ctsn)) { 1492 /* RFC 2960 6.3.2 Retransmission Timer Rules 1493 * 1494 * R3) Whenever a SACK is received 1495 * that acknowledges the DATA chunk 1496 * with the earliest outstanding TSN 1497 * for that address, restart T3-rtx 1498 * timer for that address with its 1499 * current RTO. 1500 */ 1501 restart_timer = 1; 1502 forward_progress = true; 1503 1504 if (!tchunk->tsn_gap_acked) { 1505 /* 1506 * SFR-CACC algorithm: 1507 * 2) If the SACK contains gap acks 1508 * and the flag CHANGEOVER_ACTIVE is 1509 * set the receiver of the SACK MUST 1510 * take the following action: 1511 * 1512 * B) For each TSN t being acked that 1513 * has not been acked in any SACK so 1514 * far, set cacc_saw_newack to 1 for 1515 * the destination that the TSN was 1516 * sent to. 1517 */ 1518 if (transport && 1519 sack->num_gap_ack_blocks && 1520 q->asoc->peer.primary_path->cacc. 1521 changeover_active) 1522 transport->cacc.cacc_saw_newack 1523 = 1; 1524 } 1525 1526 list_add_tail(&tchunk->transmitted_list, 1527 &q->sacked); 1528 } else { 1529 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1530 * M2) Each time a SACK arrives reporting 1531 * 'Stray DATA chunk(s)' record the highest TSN 1532 * reported as newly acknowledged, call this 1533 * value 'HighestTSNinSack'. A newly 1534 * acknowledged DATA chunk is one not 1535 * previously acknowledged in a SACK. 1536 * 1537 * When the SCTP sender of data receives a SACK 1538 * chunk that acknowledges, for the first time, 1539 * the receipt of a DATA chunk, all the still 1540 * unacknowledged DATA chunks whose TSN is 1541 * older than that newly acknowledged DATA 1542 * chunk, are qualified as 'Stray DATA chunks'. 1543 */ 1544 list_add_tail(lchunk, &tlist); 1545 } 1546 } else { 1547 if (tchunk->tsn_gap_acked) { 1548 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1549 __func__, tsn); 1550 1551 tchunk->tsn_gap_acked = 0; 1552 1553 if (tchunk->transport) 1554 bytes_acked -= sctp_data_size(tchunk); 1555 1556 /* RFC 2960 6.3.2 Retransmission Timer Rules 1557 * 1558 * R4) Whenever a SACK is received missing a 1559 * TSN that was previously acknowledged via a 1560 * Gap Ack Block, start T3-rtx for the 1561 * destination address to which the DATA 1562 * chunk was originally 1563 * transmitted if it is not already running. 1564 */ 1565 restart_timer = 1; 1566 } 1567 1568 list_add_tail(lchunk, &tlist); 1569 } 1570 } 1571 1572 if (transport) { 1573 if (bytes_acked) { 1574 struct sctp_association *asoc = transport->asoc; 1575 1576 /* We may have counted DATA that was migrated 1577 * to this transport due to DEL-IP operation. 1578 * Subtract those bytes, since the were never 1579 * send on this transport and shouldn't be 1580 * credited to this transport. 1581 */ 1582 bytes_acked -= migrate_bytes; 1583 1584 /* 8.2. When an outstanding TSN is acknowledged, 1585 * the endpoint shall clear the error counter of 1586 * the destination transport address to which the 1587 * DATA chunk was last sent. 1588 * The association's overall error counter is 1589 * also cleared. 1590 */ 1591 transport->error_count = 0; 1592 transport->asoc->overall_error_count = 0; 1593 forward_progress = true; 1594 1595 /* 1596 * While in SHUTDOWN PENDING, we may have started 1597 * the T5 shutdown guard timer after reaching the 1598 * retransmission limit. Stop that timer as soon 1599 * as the receiver acknowledged any data. 1600 */ 1601 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1602 del_timer(&asoc->timers 1603 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1604 sctp_association_put(asoc); 1605 1606 /* Mark the destination transport address as 1607 * active if it is not so marked. 1608 */ 1609 if ((transport->state == SCTP_INACTIVE || 1610 transport->state == SCTP_UNCONFIRMED) && 1611 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1612 sctp_assoc_control_transport( 1613 transport->asoc, 1614 transport, 1615 SCTP_TRANSPORT_UP, 1616 SCTP_RECEIVED_SACK); 1617 } 1618 1619 sctp_transport_raise_cwnd(transport, sack_ctsn, 1620 bytes_acked); 1621 1622 transport->flight_size -= bytes_acked; 1623 if (transport->flight_size == 0) 1624 transport->partial_bytes_acked = 0; 1625 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1626 } else { 1627 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1628 * When a sender is doing zero window probing, it 1629 * should not timeout the association if it continues 1630 * to receive new packets from the receiver. The 1631 * reason is that the receiver MAY keep its window 1632 * closed for an indefinite time. 1633 * A sender is doing zero window probing when the 1634 * receiver's advertised window is zero, and there is 1635 * only one data chunk in flight to the receiver. 1636 * 1637 * Allow the association to timeout while in SHUTDOWN 1638 * PENDING or SHUTDOWN RECEIVED in case the receiver 1639 * stays in zero window mode forever. 1640 */ 1641 if (!q->asoc->peer.rwnd && 1642 !list_empty(&tlist) && 1643 (sack_ctsn+2 == q->asoc->next_tsn) && 1644 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1645 pr_debug("%s: sack received for zero window " 1646 "probe:%u\n", __func__, sack_ctsn); 1647 1648 q->asoc->overall_error_count = 0; 1649 transport->error_count = 0; 1650 } 1651 } 1652 1653 /* RFC 2960 6.3.2 Retransmission Timer Rules 1654 * 1655 * R2) Whenever all outstanding data sent to an address have 1656 * been acknowledged, turn off the T3-rtx timer of that 1657 * address. 1658 */ 1659 if (!transport->flight_size) { 1660 if (del_timer(&transport->T3_rtx_timer)) 1661 sctp_transport_put(transport); 1662 } else if (restart_timer) { 1663 if (!mod_timer(&transport->T3_rtx_timer, 1664 jiffies + transport->rto)) 1665 sctp_transport_hold(transport); 1666 } 1667 1668 if (forward_progress) { 1669 if (transport->dst) 1670 sctp_transport_dst_confirm(transport); 1671 } 1672 } 1673 1674 list_splice(&tlist, transmitted_queue); 1675 } 1676 1677 /* Mark chunks as missing and consequently may get retransmitted. */ 1678 static void sctp_mark_missing(struct sctp_outq *q, 1679 struct list_head *transmitted_queue, 1680 struct sctp_transport *transport, 1681 __u32 highest_new_tsn_in_sack, 1682 int count_of_newacks) 1683 { 1684 struct sctp_chunk *chunk; 1685 __u32 tsn; 1686 char do_fast_retransmit = 0; 1687 struct sctp_association *asoc = q->asoc; 1688 struct sctp_transport *primary = asoc->peer.primary_path; 1689 1690 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1691 1692 tsn = ntohl(chunk->subh.data_hdr->tsn); 1693 1694 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1695 * 'Unacknowledged TSN's', if the TSN number of an 1696 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1697 * value, increment the 'TSN.Missing.Report' count on that 1698 * chunk if it has NOT been fast retransmitted or marked for 1699 * fast retransmit already. 1700 */ 1701 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1702 !chunk->tsn_gap_acked && 1703 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1704 1705 /* SFR-CACC may require us to skip marking 1706 * this chunk as missing. 1707 */ 1708 if (!transport || !sctp_cacc_skip(primary, 1709 chunk->transport, 1710 count_of_newacks, tsn)) { 1711 chunk->tsn_missing_report++; 1712 1713 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1714 __func__, tsn, chunk->tsn_missing_report); 1715 } 1716 } 1717 /* 1718 * M4) If any DATA chunk is found to have a 1719 * 'TSN.Missing.Report' 1720 * value larger than or equal to 3, mark that chunk for 1721 * retransmission and start the fast retransmit procedure. 1722 */ 1723 1724 if (chunk->tsn_missing_report >= 3) { 1725 chunk->fast_retransmit = SCTP_NEED_FRTX; 1726 do_fast_retransmit = 1; 1727 } 1728 } 1729 1730 if (transport) { 1731 if (do_fast_retransmit) 1732 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1733 1734 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1735 "flight_size:%d, pba:%d\n", __func__, transport, 1736 transport->cwnd, transport->ssthresh, 1737 transport->flight_size, transport->partial_bytes_acked); 1738 } 1739 } 1740 1741 /* Is the given TSN acked by this packet? */ 1742 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1743 { 1744 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1745 union sctp_sack_variable *frags; 1746 __u16 tsn_offset, blocks; 1747 int i; 1748 1749 if (TSN_lte(tsn, ctsn)) 1750 goto pass; 1751 1752 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1753 * 1754 * Gap Ack Blocks: 1755 * These fields contain the Gap Ack Blocks. They are repeated 1756 * for each Gap Ack Block up to the number of Gap Ack Blocks 1757 * defined in the Number of Gap Ack Blocks field. All DATA 1758 * chunks with TSNs greater than or equal to (Cumulative TSN 1759 * Ack + Gap Ack Block Start) and less than or equal to 1760 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1761 * Block are assumed to have been received correctly. 1762 */ 1763 1764 frags = sack->variable; 1765 blocks = ntohs(sack->num_gap_ack_blocks); 1766 tsn_offset = tsn - ctsn; 1767 for (i = 0; i < blocks; ++i) { 1768 if (tsn_offset >= ntohs(frags[i].gab.start) && 1769 tsn_offset <= ntohs(frags[i].gab.end)) 1770 goto pass; 1771 } 1772 1773 return 0; 1774 pass: 1775 return 1; 1776 } 1777 1778 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1779 int nskips, __be16 stream) 1780 { 1781 int i; 1782 1783 for (i = 0; i < nskips; i++) { 1784 if (skiplist[i].stream == stream) 1785 return i; 1786 } 1787 return i; 1788 } 1789 1790 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1791 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1792 { 1793 struct sctp_association *asoc = q->asoc; 1794 struct sctp_chunk *ftsn_chunk = NULL; 1795 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1796 int nskips = 0; 1797 int skip_pos = 0; 1798 __u32 tsn; 1799 struct sctp_chunk *chunk; 1800 struct list_head *lchunk, *temp; 1801 1802 if (!asoc->peer.prsctp_capable) 1803 return; 1804 1805 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1806 * received SACK. 1807 * 1808 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1809 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1810 */ 1811 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1812 asoc->adv_peer_ack_point = ctsn; 1813 1814 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1815 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1816 * the chunk next in the out-queue space is marked as "abandoned" as 1817 * shown in the following example: 1818 * 1819 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1820 * and the Advanced.Peer.Ack.Point is updated to this value: 1821 * 1822 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1823 * normal SACK processing local advancement 1824 * ... ... 1825 * Adv.Ack.Pt-> 102 acked 102 acked 1826 * 103 abandoned 103 abandoned 1827 * 104 abandoned Adv.Ack.P-> 104 abandoned 1828 * 105 105 1829 * 106 acked 106 acked 1830 * ... ... 1831 * 1832 * In this example, the data sender successfully advanced the 1833 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1834 */ 1835 list_for_each_safe(lchunk, temp, &q->abandoned) { 1836 chunk = list_entry(lchunk, struct sctp_chunk, 1837 transmitted_list); 1838 tsn = ntohl(chunk->subh.data_hdr->tsn); 1839 1840 /* Remove any chunks in the abandoned queue that are acked by 1841 * the ctsn. 1842 */ 1843 if (TSN_lte(tsn, ctsn)) { 1844 list_del_init(lchunk); 1845 sctp_chunk_free(chunk); 1846 } else { 1847 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1848 asoc->adv_peer_ack_point = tsn; 1849 if (chunk->chunk_hdr->flags & 1850 SCTP_DATA_UNORDERED) 1851 continue; 1852 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1853 nskips, 1854 chunk->subh.data_hdr->stream); 1855 ftsn_skip_arr[skip_pos].stream = 1856 chunk->subh.data_hdr->stream; 1857 ftsn_skip_arr[skip_pos].ssn = 1858 chunk->subh.data_hdr->ssn; 1859 if (skip_pos == nskips) 1860 nskips++; 1861 if (nskips == 10) 1862 break; 1863 } else 1864 break; 1865 } 1866 } 1867 1868 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1869 * is greater than the Cumulative TSN ACK carried in the received 1870 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1871 * chunk containing the latest value of the 1872 * "Advanced.Peer.Ack.Point". 1873 * 1874 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1875 * list each stream and sequence number in the forwarded TSN. This 1876 * information will enable the receiver to easily find any 1877 * stranded TSN's waiting on stream reorder queues. Each stream 1878 * SHOULD only be reported once; this means that if multiple 1879 * abandoned messages occur in the same stream then only the 1880 * highest abandoned stream sequence number is reported. If the 1881 * total size of the FORWARD TSN does NOT fit in a single MTU then 1882 * the sender of the FORWARD TSN SHOULD lower the 1883 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1884 * single MTU. 1885 */ 1886 if (asoc->adv_peer_ack_point > ctsn) 1887 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1888 nskips, &ftsn_skip_arr[0]); 1889 1890 if (ftsn_chunk) { 1891 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1892 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS); 1893 } 1894 } 1895