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 54 /* Declare internal functions here. */ 55 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 56 static void sctp_check_transmitted(struct sctp_outq *q, 57 struct list_head *transmitted_queue, 58 struct sctp_transport *transport, 59 union sctp_addr *saddr, 60 struct sctp_sackhdr *sack, 61 __u32 *highest_new_tsn); 62 63 static void sctp_mark_missing(struct sctp_outq *q, 64 struct list_head *transmitted_queue, 65 struct sctp_transport *transport, 66 __u32 highest_new_tsn, 67 int count_of_newacks); 68 69 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn); 70 71 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp); 72 73 /* Add data to the front of the queue. */ 74 static inline void sctp_outq_head_data(struct sctp_outq *q, 75 struct sctp_chunk *ch) 76 { 77 list_add(&ch->list, &q->out_chunk_list); 78 q->out_qlen += ch->skb->len; 79 } 80 81 /* Take data from the front of the queue. */ 82 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 83 { 84 struct sctp_chunk *ch = NULL; 85 86 if (!list_empty(&q->out_chunk_list)) { 87 struct list_head *entry = q->out_chunk_list.next; 88 89 ch = list_entry(entry, struct sctp_chunk, list); 90 list_del_init(entry); 91 q->out_qlen -= ch->skb->len; 92 } 93 return ch; 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 list_add_tail(&ch->list, &q->out_chunk_list); 100 q->out_qlen += ch->skb->len; 101 } 102 103 /* 104 * SFR-CACC algorithm: 105 * D) If count_of_newacks is greater than or equal to 2 106 * and t was not sent to the current primary then the 107 * sender MUST NOT increment missing report count for t. 108 */ 109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 110 struct sctp_transport *transport, 111 int count_of_newacks) 112 { 113 if (count_of_newacks >= 2 && transport != primary) 114 return 1; 115 return 0; 116 } 117 118 /* 119 * SFR-CACC algorithm: 120 * F) If count_of_newacks is less than 2, let d be the 121 * destination to which t was sent. If cacc_saw_newack 122 * is 0 for destination d, then the sender MUST NOT 123 * increment missing report count for t. 124 */ 125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 126 int count_of_newacks) 127 { 128 if (count_of_newacks < 2 && 129 (transport && !transport->cacc.cacc_saw_newack)) 130 return 1; 131 return 0; 132 } 133 134 /* 135 * SFR-CACC algorithm: 136 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 137 * execute steps C, D, F. 138 * 139 * C has been implemented in sctp_outq_sack 140 */ 141 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 142 struct sctp_transport *transport, 143 int count_of_newacks) 144 { 145 if (!primary->cacc.cycling_changeover) { 146 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 147 return 1; 148 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 149 return 1; 150 return 0; 151 } 152 return 0; 153 } 154 155 /* 156 * SFR-CACC algorithm: 157 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 158 * than next_tsn_at_change of the current primary, then 159 * the sender MUST NOT increment missing report count 160 * for t. 161 */ 162 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 163 { 164 if (primary->cacc.cycling_changeover && 165 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 166 return 1; 167 return 0; 168 } 169 170 /* 171 * SFR-CACC algorithm: 172 * 3) If the missing report count for TSN t is to be 173 * incremented according to [RFC2960] and 174 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 175 * then the sender MUST further execute steps 3.1 and 176 * 3.2 to determine if the missing report count for 177 * TSN t SHOULD NOT be incremented. 178 * 179 * 3.3) If 3.1 and 3.2 do not dictate that the missing 180 * report count for t should not be incremented, then 181 * the sender SHOULD increment missing report count for 182 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 183 */ 184 static inline int sctp_cacc_skip(struct sctp_transport *primary, 185 struct sctp_transport *transport, 186 int count_of_newacks, 187 __u32 tsn) 188 { 189 if (primary->cacc.changeover_active && 190 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 191 sctp_cacc_skip_3_2(primary, tsn))) 192 return 1; 193 return 0; 194 } 195 196 /* Initialize an existing sctp_outq. This does the boring stuff. 197 * You still need to define handlers if you really want to DO 198 * something with this structure... 199 */ 200 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 201 { 202 memset(q, 0, sizeof(struct sctp_outq)); 203 204 q->asoc = asoc; 205 INIT_LIST_HEAD(&q->out_chunk_list); 206 INIT_LIST_HEAD(&q->control_chunk_list); 207 INIT_LIST_HEAD(&q->retransmit); 208 INIT_LIST_HEAD(&q->sacked); 209 INIT_LIST_HEAD(&q->abandoned); 210 } 211 212 /* Free the outqueue structure and any related pending chunks. 213 */ 214 static void __sctp_outq_teardown(struct sctp_outq *q) 215 { 216 struct sctp_transport *transport; 217 struct list_head *lchunk, *temp; 218 struct sctp_chunk *chunk, *tmp; 219 220 /* Throw away unacknowledged chunks. */ 221 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 222 transports) { 223 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 224 chunk = list_entry(lchunk, struct sctp_chunk, 225 transmitted_list); 226 /* Mark as part of a failed message. */ 227 sctp_chunk_fail(chunk, q->error); 228 sctp_chunk_free(chunk); 229 } 230 } 231 232 /* Throw away chunks that have been gap ACKed. */ 233 list_for_each_safe(lchunk, temp, &q->sacked) { 234 list_del_init(lchunk); 235 chunk = list_entry(lchunk, struct sctp_chunk, 236 transmitted_list); 237 sctp_chunk_fail(chunk, q->error); 238 sctp_chunk_free(chunk); 239 } 240 241 /* Throw away any chunks in the retransmit queue. */ 242 list_for_each_safe(lchunk, temp, &q->retransmit) { 243 list_del_init(lchunk); 244 chunk = list_entry(lchunk, struct sctp_chunk, 245 transmitted_list); 246 sctp_chunk_fail(chunk, q->error); 247 sctp_chunk_free(chunk); 248 } 249 250 /* Throw away any chunks that are in the abandoned queue. */ 251 list_for_each_safe(lchunk, temp, &q->abandoned) { 252 list_del_init(lchunk); 253 chunk = list_entry(lchunk, struct sctp_chunk, 254 transmitted_list); 255 sctp_chunk_fail(chunk, q->error); 256 sctp_chunk_free(chunk); 257 } 258 259 /* Throw away any leftover data chunks. */ 260 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 261 262 /* Mark as send failure. */ 263 sctp_chunk_fail(chunk, q->error); 264 sctp_chunk_free(chunk); 265 } 266 267 /* Throw away any leftover control chunks. */ 268 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 269 list_del_init(&chunk->list); 270 sctp_chunk_free(chunk); 271 } 272 } 273 274 void sctp_outq_teardown(struct sctp_outq *q) 275 { 276 __sctp_outq_teardown(q); 277 sctp_outq_init(q->asoc, q); 278 } 279 280 /* Free the outqueue structure and any related pending chunks. */ 281 void sctp_outq_free(struct sctp_outq *q) 282 { 283 /* Throw away leftover chunks. */ 284 __sctp_outq_teardown(q); 285 } 286 287 /* Put a new chunk in an sctp_outq. */ 288 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp) 289 { 290 struct net *net = sock_net(q->asoc->base.sk); 291 int error = 0; 292 293 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk, 294 chunk && chunk->chunk_hdr ? 295 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 296 "illegal chunk"); 297 298 /* If it is data, queue it up, otherwise, send it 299 * immediately. 300 */ 301 if (sctp_chunk_is_data(chunk)) { 302 /* Is it OK to queue data chunks? */ 303 /* From 9. Termination of Association 304 * 305 * When either endpoint performs a shutdown, the 306 * association on each peer will stop accepting new 307 * data from its user and only deliver data in queue 308 * at the time of sending or receiving the SHUTDOWN 309 * chunk. 310 */ 311 switch (q->asoc->state) { 312 case SCTP_STATE_CLOSED: 313 case SCTP_STATE_SHUTDOWN_PENDING: 314 case SCTP_STATE_SHUTDOWN_SENT: 315 case SCTP_STATE_SHUTDOWN_RECEIVED: 316 case SCTP_STATE_SHUTDOWN_ACK_SENT: 317 /* Cannot send after transport endpoint shutdown */ 318 error = -ESHUTDOWN; 319 break; 320 321 default: 322 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n", 323 __func__, q, chunk, chunk && chunk->chunk_hdr ? 324 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 325 "illegal chunk"); 326 327 sctp_chunk_hold(chunk); 328 sctp_outq_tail_data(q, chunk); 329 if (chunk->asoc->prsctp_enable && 330 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 331 chunk->asoc->sent_cnt_removable++; 332 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 333 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS); 334 else 335 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS); 336 break; 337 } 338 } else { 339 list_add_tail(&chunk->list, &q->control_chunk_list); 340 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS); 341 } 342 343 if (error < 0) 344 return error; 345 346 if (!q->cork) 347 error = sctp_outq_flush(q, 0, gfp); 348 349 return error; 350 } 351 352 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 353 * and the abandoned list are in ascending order. 354 */ 355 static void sctp_insert_list(struct list_head *head, struct list_head *new) 356 { 357 struct list_head *pos; 358 struct sctp_chunk *nchunk, *lchunk; 359 __u32 ntsn, ltsn; 360 int done = 0; 361 362 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 363 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 364 365 list_for_each(pos, head) { 366 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 367 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 368 if (TSN_lt(ntsn, ltsn)) { 369 list_add(new, pos->prev); 370 done = 1; 371 break; 372 } 373 } 374 if (!done) 375 list_add_tail(new, head); 376 } 377 378 static int sctp_prsctp_prune_sent(struct sctp_association *asoc, 379 struct sctp_sndrcvinfo *sinfo, 380 struct list_head *queue, int msg_len) 381 { 382 struct sctp_chunk *chk, *temp; 383 384 list_for_each_entry_safe(chk, temp, queue, transmitted_list) { 385 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 386 chk->prsctp_param <= sinfo->sinfo_timetolive) 387 continue; 388 389 list_del_init(&chk->transmitted_list); 390 sctp_insert_list(&asoc->outqueue.abandoned, 391 &chk->transmitted_list); 392 393 asoc->sent_cnt_removable--; 394 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++; 395 396 if (!chk->tsn_gap_acked) { 397 if (chk->transport) 398 chk->transport->flight_size -= 399 sctp_data_size(chk); 400 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk); 401 } 402 403 msg_len -= SCTP_DATA_SNDSIZE(chk) + 404 sizeof(struct sk_buff) + 405 sizeof(struct sctp_chunk); 406 if (msg_len <= 0) 407 break; 408 } 409 410 return msg_len; 411 } 412 413 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc, 414 struct sctp_sndrcvinfo *sinfo, 415 struct list_head *queue, int msg_len) 416 { 417 struct sctp_chunk *chk, *temp; 418 419 list_for_each_entry_safe(chk, temp, queue, list) { 420 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) || 421 chk->prsctp_param <= sinfo->sinfo_timetolive) 422 continue; 423 424 list_del_init(&chk->list); 425 asoc->sent_cnt_removable--; 426 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++; 427 428 msg_len -= SCTP_DATA_SNDSIZE(chk) + 429 sizeof(struct sk_buff) + 430 sizeof(struct sctp_chunk); 431 sctp_chunk_free(chk); 432 if (msg_len <= 0) 433 break; 434 } 435 436 return msg_len; 437 } 438 439 /* Abandon the chunks according their priorities */ 440 void sctp_prsctp_prune(struct sctp_association *asoc, 441 struct sctp_sndrcvinfo *sinfo, int msg_len) 442 { 443 struct sctp_transport *transport; 444 445 if (!asoc->prsctp_enable || !asoc->sent_cnt_removable) 446 return; 447 448 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 449 &asoc->outqueue.retransmit, 450 msg_len); 451 if (msg_len <= 0) 452 return; 453 454 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 455 transports) { 456 msg_len = sctp_prsctp_prune_sent(asoc, sinfo, 457 &transport->transmitted, 458 msg_len); 459 if (msg_len <= 0) 460 return; 461 } 462 463 sctp_prsctp_prune_unsent(asoc, sinfo, 464 &asoc->outqueue.out_chunk_list, 465 msg_len); 466 } 467 468 /* Mark all the eligible packets on a transport for retransmission. */ 469 void sctp_retransmit_mark(struct sctp_outq *q, 470 struct sctp_transport *transport, 471 __u8 reason) 472 { 473 struct list_head *lchunk, *ltemp; 474 struct sctp_chunk *chunk; 475 476 /* Walk through the specified transmitted queue. */ 477 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 478 chunk = list_entry(lchunk, struct sctp_chunk, 479 transmitted_list); 480 481 /* If the chunk is abandoned, move it to abandoned list. */ 482 if (sctp_chunk_abandoned(chunk)) { 483 list_del_init(lchunk); 484 sctp_insert_list(&q->abandoned, lchunk); 485 486 /* If this chunk has not been previousely acked, 487 * stop considering it 'outstanding'. Our peer 488 * will most likely never see it since it will 489 * not be retransmitted 490 */ 491 if (!chunk->tsn_gap_acked) { 492 if (chunk->transport) 493 chunk->transport->flight_size -= 494 sctp_data_size(chunk); 495 q->outstanding_bytes -= sctp_data_size(chunk); 496 q->asoc->peer.rwnd += sctp_data_size(chunk); 497 } 498 continue; 499 } 500 501 /* If we are doing retransmission due to a timeout or pmtu 502 * discovery, only the chunks that are not yet acked should 503 * be added to the retransmit queue. 504 */ 505 if ((reason == SCTP_RTXR_FAST_RTX && 506 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 507 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 508 /* RFC 2960 6.2.1 Processing a Received SACK 509 * 510 * C) Any time a DATA chunk is marked for 511 * retransmission (via either T3-rtx timer expiration 512 * (Section 6.3.3) or via fast retransmit 513 * (Section 7.2.4)), add the data size of those 514 * chunks to the rwnd. 515 */ 516 q->asoc->peer.rwnd += sctp_data_size(chunk); 517 q->outstanding_bytes -= sctp_data_size(chunk); 518 if (chunk->transport) 519 transport->flight_size -= sctp_data_size(chunk); 520 521 /* sctpimpguide-05 Section 2.8.2 522 * M5) If a T3-rtx timer expires, the 523 * 'TSN.Missing.Report' of all affected TSNs is set 524 * to 0. 525 */ 526 chunk->tsn_missing_report = 0; 527 528 /* If a chunk that is being used for RTT measurement 529 * has to be retransmitted, we cannot use this chunk 530 * anymore for RTT measurements. Reset rto_pending so 531 * that a new RTT measurement is started when a new 532 * data chunk is sent. 533 */ 534 if (chunk->rtt_in_progress) { 535 chunk->rtt_in_progress = 0; 536 transport->rto_pending = 0; 537 } 538 539 chunk->resent = 1; 540 541 /* Move the chunk to the retransmit queue. The chunks 542 * on the retransmit queue are always kept in order. 543 */ 544 list_del_init(lchunk); 545 sctp_insert_list(&q->retransmit, lchunk); 546 } 547 } 548 549 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, " 550 "flight_size:%d, pba:%d\n", __func__, transport, reason, 551 transport->cwnd, transport->ssthresh, transport->flight_size, 552 transport->partial_bytes_acked); 553 } 554 555 /* Mark all the eligible packets on a transport for retransmission and force 556 * one packet out. 557 */ 558 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 559 sctp_retransmit_reason_t reason) 560 { 561 struct net *net = sock_net(q->asoc->base.sk); 562 int error = 0; 563 564 switch (reason) { 565 case SCTP_RTXR_T3_RTX: 566 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS); 567 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 568 /* Update the retran path if the T3-rtx timer has expired for 569 * the current retran path. 570 */ 571 if (transport == transport->asoc->peer.retran_path) 572 sctp_assoc_update_retran_path(transport->asoc); 573 transport->asoc->rtx_data_chunks += 574 transport->asoc->unack_data; 575 break; 576 case SCTP_RTXR_FAST_RTX: 577 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS); 578 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 579 q->fast_rtx = 1; 580 break; 581 case SCTP_RTXR_PMTUD: 582 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS); 583 break; 584 case SCTP_RTXR_T1_RTX: 585 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS); 586 transport->asoc->init_retries++; 587 break; 588 default: 589 BUG(); 590 } 591 592 sctp_retransmit_mark(q, transport, reason); 593 594 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 595 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 596 * following the procedures outlined in C1 - C5. 597 */ 598 if (reason == SCTP_RTXR_T3_RTX) 599 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point); 600 601 /* Flush the queues only on timeout, since fast_rtx is only 602 * triggered during sack processing and the queue 603 * will be flushed at the end. 604 */ 605 if (reason != SCTP_RTXR_FAST_RTX) 606 error = sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC); 607 608 if (error) 609 q->asoc->base.sk->sk_err = -error; 610 } 611 612 /* 613 * Transmit DATA chunks on the retransmit queue. Upon return from 614 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 615 * need to be transmitted by the caller. 616 * We assume that pkt->transport has already been set. 617 * 618 * The return value is a normal kernel error return value. 619 */ 620 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 621 int rtx_timeout, int *start_timer) 622 { 623 struct list_head *lqueue; 624 struct sctp_transport *transport = pkt->transport; 625 sctp_xmit_t status; 626 struct sctp_chunk *chunk, *chunk1; 627 int fast_rtx; 628 int error = 0; 629 int timer = 0; 630 int done = 0; 631 632 lqueue = &q->retransmit; 633 fast_rtx = q->fast_rtx; 634 635 /* This loop handles time-out retransmissions, fast retransmissions, 636 * and retransmissions due to opening of whindow. 637 * 638 * RFC 2960 6.3.3 Handle T3-rtx Expiration 639 * 640 * E3) Determine how many of the earliest (i.e., lowest TSN) 641 * outstanding DATA chunks for the address for which the 642 * T3-rtx has expired will fit into a single packet, subject 643 * to the MTU constraint for the path corresponding to the 644 * destination transport address to which the retransmission 645 * is being sent (this may be different from the address for 646 * which the timer expires [see Section 6.4]). Call this value 647 * K. Bundle and retransmit those K DATA chunks in a single 648 * packet to the destination endpoint. 649 * 650 * [Just to be painfully clear, if we are retransmitting 651 * because a timeout just happened, we should send only ONE 652 * packet of retransmitted data.] 653 * 654 * For fast retransmissions we also send only ONE packet. However, 655 * if we are just flushing the queue due to open window, we'll 656 * try to send as much as possible. 657 */ 658 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 659 /* If the chunk is abandoned, move it to abandoned list. */ 660 if (sctp_chunk_abandoned(chunk)) { 661 list_del_init(&chunk->transmitted_list); 662 sctp_insert_list(&q->abandoned, 663 &chunk->transmitted_list); 664 continue; 665 } 666 667 /* Make sure that Gap Acked TSNs are not retransmitted. A 668 * simple approach is just to move such TSNs out of the 669 * way and into a 'transmitted' queue and skip to the 670 * next chunk. 671 */ 672 if (chunk->tsn_gap_acked) { 673 list_move_tail(&chunk->transmitted_list, 674 &transport->transmitted); 675 continue; 676 } 677 678 /* If we are doing fast retransmit, ignore non-fast_rtransmit 679 * chunks 680 */ 681 if (fast_rtx && !chunk->fast_retransmit) 682 continue; 683 684 redo: 685 /* Attempt to append this chunk to the packet. */ 686 status = sctp_packet_append_chunk(pkt, chunk); 687 688 switch (status) { 689 case SCTP_XMIT_PMTU_FULL: 690 if (!pkt->has_data && !pkt->has_cookie_echo) { 691 /* If this packet did not contain DATA then 692 * retransmission did not happen, so do it 693 * again. We'll ignore the error here since 694 * control chunks are already freed so there 695 * is nothing we can do. 696 */ 697 sctp_packet_transmit(pkt, GFP_ATOMIC); 698 goto redo; 699 } 700 701 /* Send this packet. */ 702 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 703 704 /* If we are retransmitting, we should only 705 * send a single packet. 706 * Otherwise, try appending this chunk again. 707 */ 708 if (rtx_timeout || fast_rtx) 709 done = 1; 710 else 711 goto redo; 712 713 /* Bundle next chunk in the next round. */ 714 break; 715 716 case SCTP_XMIT_RWND_FULL: 717 /* Send this packet. */ 718 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 719 720 /* Stop sending DATA as there is no more room 721 * at the receiver. 722 */ 723 done = 1; 724 break; 725 726 case SCTP_XMIT_DELAY: 727 /* Send this packet. */ 728 error = sctp_packet_transmit(pkt, GFP_ATOMIC); 729 730 /* Stop sending DATA because of nagle delay. */ 731 done = 1; 732 break; 733 734 default: 735 /* The append was successful, so add this chunk to 736 * the transmitted list. 737 */ 738 list_move_tail(&chunk->transmitted_list, 739 &transport->transmitted); 740 741 /* Mark the chunk as ineligible for fast retransmit 742 * after it is retransmitted. 743 */ 744 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 745 chunk->fast_retransmit = SCTP_DONT_FRTX; 746 747 q->asoc->stats.rtxchunks++; 748 break; 749 } 750 751 /* Set the timer if there were no errors */ 752 if (!error && !timer) 753 timer = 1; 754 755 if (done) 756 break; 757 } 758 759 /* If we are here due to a retransmit timeout or a fast 760 * retransmit and if there are any chunks left in the retransmit 761 * queue that could not fit in the PMTU sized packet, they need 762 * to be marked as ineligible for a subsequent fast retransmit. 763 */ 764 if (rtx_timeout || fast_rtx) { 765 list_for_each_entry(chunk1, lqueue, transmitted_list) { 766 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 767 chunk1->fast_retransmit = SCTP_DONT_FRTX; 768 } 769 } 770 771 *start_timer = timer; 772 773 /* Clear fast retransmit hint */ 774 if (fast_rtx) 775 q->fast_rtx = 0; 776 777 return error; 778 } 779 780 /* Cork the outqueue so queued chunks are really queued. */ 781 int sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp) 782 { 783 if (q->cork) 784 q->cork = 0; 785 786 return sctp_outq_flush(q, 0, gfp); 787 } 788 789 790 /* 791 * Try to flush an outqueue. 792 * 793 * Description: Send everything in q which we legally can, subject to 794 * congestion limitations. 795 * * Note: This function can be called from multiple contexts so appropriate 796 * locking concerns must be made. Today we use the sock lock to protect 797 * this function. 798 */ 799 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp) 800 { 801 struct sctp_packet *packet; 802 struct sctp_packet singleton; 803 struct sctp_association *asoc = q->asoc; 804 __u16 sport = asoc->base.bind_addr.port; 805 __u16 dport = asoc->peer.port; 806 __u32 vtag = asoc->peer.i.init_tag; 807 struct sctp_transport *transport = NULL; 808 struct sctp_transport *new_transport; 809 struct sctp_chunk *chunk, *tmp; 810 sctp_xmit_t status; 811 int error = 0; 812 int start_timer = 0; 813 int one_packet = 0; 814 815 /* These transports have chunks to send. */ 816 struct list_head transport_list; 817 struct list_head *ltransport; 818 819 INIT_LIST_HEAD(&transport_list); 820 packet = NULL; 821 822 /* 823 * 6.10 Bundling 824 * ... 825 * When bundling control chunks with DATA chunks, an 826 * endpoint MUST place control chunks first in the outbound 827 * SCTP packet. The transmitter MUST transmit DATA chunks 828 * within a SCTP packet in increasing order of TSN. 829 * ... 830 */ 831 832 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 833 /* RFC 5061, 5.3 834 * F1) This means that until such time as the ASCONF 835 * containing the add is acknowledged, the sender MUST 836 * NOT use the new IP address as a source for ANY SCTP 837 * packet except on carrying an ASCONF Chunk. 838 */ 839 if (asoc->src_out_of_asoc_ok && 840 chunk->chunk_hdr->type != SCTP_CID_ASCONF) 841 continue; 842 843 list_del_init(&chunk->list); 844 845 /* Pick the right transport to use. */ 846 new_transport = chunk->transport; 847 848 if (!new_transport) { 849 /* 850 * If we have a prior transport pointer, see if 851 * the destination address of the chunk 852 * matches the destination address of the 853 * current transport. If not a match, then 854 * try to look up the transport with a given 855 * destination address. We do this because 856 * after processing ASCONFs, we may have new 857 * transports created. 858 */ 859 if (transport && 860 sctp_cmp_addr_exact(&chunk->dest, 861 &transport->ipaddr)) 862 new_transport = transport; 863 else 864 new_transport = sctp_assoc_lookup_paddr(asoc, 865 &chunk->dest); 866 867 /* if we still don't have a new transport, then 868 * use the current active path. 869 */ 870 if (!new_transport) 871 new_transport = asoc->peer.active_path; 872 } else if ((new_transport->state == SCTP_INACTIVE) || 873 (new_transport->state == SCTP_UNCONFIRMED) || 874 (new_transport->state == SCTP_PF)) { 875 /* If the chunk is Heartbeat or Heartbeat Ack, 876 * send it to chunk->transport, even if it's 877 * inactive. 878 * 879 * 3.3.6 Heartbeat Acknowledgement: 880 * ... 881 * A HEARTBEAT ACK is always sent to the source IP 882 * address of the IP datagram containing the 883 * HEARTBEAT chunk to which this ack is responding. 884 * ... 885 * 886 * ASCONF_ACKs also must be sent to the source. 887 */ 888 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT && 889 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK && 890 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK) 891 new_transport = asoc->peer.active_path; 892 } 893 894 /* Are we switching transports? 895 * Take care of transport locks. 896 */ 897 if (new_transport != transport) { 898 transport = new_transport; 899 if (list_empty(&transport->send_ready)) { 900 list_add_tail(&transport->send_ready, 901 &transport_list); 902 } 903 packet = &transport->packet; 904 sctp_packet_config(packet, vtag, 905 asoc->peer.ecn_capable); 906 } 907 908 switch (chunk->chunk_hdr->type) { 909 /* 910 * 6.10 Bundling 911 * ... 912 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 913 * COMPLETE with any other chunks. [Send them immediately.] 914 */ 915 case SCTP_CID_INIT: 916 case SCTP_CID_INIT_ACK: 917 case SCTP_CID_SHUTDOWN_COMPLETE: 918 sctp_packet_init(&singleton, transport, sport, dport); 919 sctp_packet_config(&singleton, vtag, 0); 920 sctp_packet_append_chunk(&singleton, chunk); 921 error = sctp_packet_transmit(&singleton, gfp); 922 if (error < 0) 923 return error; 924 break; 925 926 case SCTP_CID_ABORT: 927 if (sctp_test_T_bit(chunk)) { 928 packet->vtag = asoc->c.my_vtag; 929 } 930 /* The following chunks are "response" chunks, i.e. 931 * they are generated in response to something we 932 * received. If we are sending these, then we can 933 * send only 1 packet containing these chunks. 934 */ 935 case SCTP_CID_HEARTBEAT_ACK: 936 case SCTP_CID_SHUTDOWN_ACK: 937 case SCTP_CID_COOKIE_ACK: 938 case SCTP_CID_COOKIE_ECHO: 939 case SCTP_CID_ERROR: 940 case SCTP_CID_ECN_CWR: 941 case SCTP_CID_ASCONF_ACK: 942 one_packet = 1; 943 /* Fall through */ 944 945 case SCTP_CID_SACK: 946 case SCTP_CID_HEARTBEAT: 947 case SCTP_CID_SHUTDOWN: 948 case SCTP_CID_ECN_ECNE: 949 case SCTP_CID_ASCONF: 950 case SCTP_CID_FWD_TSN: 951 status = sctp_packet_transmit_chunk(packet, chunk, 952 one_packet, gfp); 953 if (status != SCTP_XMIT_OK) { 954 /* put the chunk back */ 955 list_add(&chunk->list, &q->control_chunk_list); 956 } else { 957 asoc->stats.octrlchunks++; 958 /* PR-SCTP C5) If a FORWARD TSN is sent, the 959 * sender MUST assure that at least one T3-rtx 960 * timer is running. 961 */ 962 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) { 963 sctp_transport_reset_t3_rtx(transport); 964 transport->last_time_sent = jiffies; 965 } 966 } 967 break; 968 969 default: 970 /* We built a chunk with an illegal type! */ 971 BUG(); 972 } 973 } 974 975 if (q->asoc->src_out_of_asoc_ok) 976 goto sctp_flush_out; 977 978 /* Is it OK to send data chunks? */ 979 switch (asoc->state) { 980 case SCTP_STATE_COOKIE_ECHOED: 981 /* Only allow bundling when this packet has a COOKIE-ECHO 982 * chunk. 983 */ 984 if (!packet || !packet->has_cookie_echo) 985 break; 986 987 /* fallthru */ 988 case SCTP_STATE_ESTABLISHED: 989 case SCTP_STATE_SHUTDOWN_PENDING: 990 case SCTP_STATE_SHUTDOWN_RECEIVED: 991 /* 992 * RFC 2960 6.1 Transmission of DATA Chunks 993 * 994 * C) When the time comes for the sender to transmit, 995 * before sending new DATA chunks, the sender MUST 996 * first transmit any outstanding DATA chunks which 997 * are marked for retransmission (limited by the 998 * current cwnd). 999 */ 1000 if (!list_empty(&q->retransmit)) { 1001 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 1002 goto sctp_flush_out; 1003 if (transport == asoc->peer.retran_path) 1004 goto retran; 1005 1006 /* Switch transports & prepare the packet. */ 1007 1008 transport = asoc->peer.retran_path; 1009 1010 if (list_empty(&transport->send_ready)) { 1011 list_add_tail(&transport->send_ready, 1012 &transport_list); 1013 } 1014 1015 packet = &transport->packet; 1016 sctp_packet_config(packet, vtag, 1017 asoc->peer.ecn_capable); 1018 retran: 1019 error = sctp_outq_flush_rtx(q, packet, 1020 rtx_timeout, &start_timer); 1021 1022 if (start_timer) { 1023 sctp_transport_reset_t3_rtx(transport); 1024 transport->last_time_sent = jiffies; 1025 } 1026 1027 /* This can happen on COOKIE-ECHO resend. Only 1028 * one chunk can get bundled with a COOKIE-ECHO. 1029 */ 1030 if (packet->has_cookie_echo) 1031 goto sctp_flush_out; 1032 1033 /* Don't send new data if there is still data 1034 * waiting to retransmit. 1035 */ 1036 if (!list_empty(&q->retransmit)) 1037 goto sctp_flush_out; 1038 } 1039 1040 /* Apply Max.Burst limitation to the current transport in 1041 * case it will be used for new data. We are going to 1042 * rest it before we return, but we want to apply the limit 1043 * to the currently queued data. 1044 */ 1045 if (transport) 1046 sctp_transport_burst_limited(transport); 1047 1048 /* Finally, transmit new packets. */ 1049 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 1050 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid 1051 * stream identifier. 1052 */ 1053 if (chunk->sinfo.sinfo_stream >= 1054 asoc->c.sinit_num_ostreams) { 1055 1056 /* Mark as failed send. */ 1057 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM); 1058 if (asoc->prsctp_enable && 1059 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1060 asoc->sent_cnt_removable--; 1061 sctp_chunk_free(chunk); 1062 continue; 1063 } 1064 1065 /* Has this chunk expired? */ 1066 if (sctp_chunk_abandoned(chunk)) { 1067 sctp_chunk_fail(chunk, 0); 1068 sctp_chunk_free(chunk); 1069 continue; 1070 } 1071 1072 /* If there is a specified transport, use it. 1073 * Otherwise, we want to use the active path. 1074 */ 1075 new_transport = chunk->transport; 1076 if (!new_transport || 1077 ((new_transport->state == SCTP_INACTIVE) || 1078 (new_transport->state == SCTP_UNCONFIRMED) || 1079 (new_transport->state == SCTP_PF))) 1080 new_transport = asoc->peer.active_path; 1081 if (new_transport->state == SCTP_UNCONFIRMED) { 1082 WARN_ONCE(1, "Atempt to send packet on unconfirmed path."); 1083 sctp_chunk_fail(chunk, 0); 1084 sctp_chunk_free(chunk); 1085 continue; 1086 } 1087 1088 /* Change packets if necessary. */ 1089 if (new_transport != transport) { 1090 transport = new_transport; 1091 1092 /* Schedule to have this transport's 1093 * packet flushed. 1094 */ 1095 if (list_empty(&transport->send_ready)) { 1096 list_add_tail(&transport->send_ready, 1097 &transport_list); 1098 } 1099 1100 packet = &transport->packet; 1101 sctp_packet_config(packet, vtag, 1102 asoc->peer.ecn_capable); 1103 /* We've switched transports, so apply the 1104 * Burst limit to the new transport. 1105 */ 1106 sctp_transport_burst_limited(transport); 1107 } 1108 1109 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p " 1110 "skb->users:%d\n", 1111 __func__, q, chunk, chunk && chunk->chunk_hdr ? 1112 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1113 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1114 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1115 atomic_read(&chunk->skb->users) : -1); 1116 1117 /* Add the chunk to the packet. */ 1118 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp); 1119 1120 switch (status) { 1121 case SCTP_XMIT_PMTU_FULL: 1122 case SCTP_XMIT_RWND_FULL: 1123 case SCTP_XMIT_DELAY: 1124 /* We could not append this chunk, so put 1125 * the chunk back on the output queue. 1126 */ 1127 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1128 __func__, ntohl(chunk->subh.data_hdr->tsn), 1129 status); 1130 1131 sctp_outq_head_data(q, chunk); 1132 goto sctp_flush_out; 1133 1134 case SCTP_XMIT_OK: 1135 /* The sender is in the SHUTDOWN-PENDING state, 1136 * The sender MAY set the I-bit in the DATA 1137 * chunk header. 1138 */ 1139 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1140 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1141 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1142 asoc->stats.ouodchunks++; 1143 else 1144 asoc->stats.oodchunks++; 1145 1146 break; 1147 1148 default: 1149 BUG(); 1150 } 1151 1152 /* BUG: We assume that the sctp_packet_transmit() 1153 * call below will succeed all the time and add the 1154 * chunk to the transmitted list and restart the 1155 * timers. 1156 * It is possible that the call can fail under OOM 1157 * conditions. 1158 * 1159 * Is this really a problem? Won't this behave 1160 * like a lost TSN? 1161 */ 1162 list_add_tail(&chunk->transmitted_list, 1163 &transport->transmitted); 1164 1165 sctp_transport_reset_t3_rtx(transport); 1166 transport->last_time_sent = jiffies; 1167 1168 /* Only let one DATA chunk get bundled with a 1169 * COOKIE-ECHO chunk. 1170 */ 1171 if (packet->has_cookie_echo) 1172 goto sctp_flush_out; 1173 } 1174 break; 1175 1176 default: 1177 /* Do nothing. */ 1178 break; 1179 } 1180 1181 sctp_flush_out: 1182 1183 /* Before returning, examine all the transports touched in 1184 * this call. Right now, we bluntly force clear all the 1185 * transports. Things might change after we implement Nagle. 1186 * But such an examination is still required. 1187 * 1188 * --xguo 1189 */ 1190 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) { 1191 struct sctp_transport *t = list_entry(ltransport, 1192 struct sctp_transport, 1193 send_ready); 1194 packet = &t->packet; 1195 if (!sctp_packet_empty(packet)) 1196 error = sctp_packet_transmit(packet, gfp); 1197 1198 /* Clear the burst limited state, if any */ 1199 sctp_transport_burst_reset(t); 1200 } 1201 1202 return error; 1203 } 1204 1205 /* Update unack_data based on the incoming SACK chunk */ 1206 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1207 struct sctp_sackhdr *sack) 1208 { 1209 sctp_sack_variable_t *frags; 1210 __u16 unack_data; 1211 int i; 1212 1213 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1214 1215 frags = sack->variable; 1216 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1217 unack_data -= ((ntohs(frags[i].gab.end) - 1218 ntohs(frags[i].gab.start) + 1)); 1219 } 1220 1221 assoc->unack_data = unack_data; 1222 } 1223 1224 /* This is where we REALLY process a SACK. 1225 * 1226 * Process the SACK against the outqueue. Mostly, this just frees 1227 * things off the transmitted queue. 1228 */ 1229 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1230 { 1231 struct sctp_association *asoc = q->asoc; 1232 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1233 struct sctp_transport *transport; 1234 struct sctp_chunk *tchunk = NULL; 1235 struct list_head *lchunk, *transport_list, *temp; 1236 sctp_sack_variable_t *frags = sack->variable; 1237 __u32 sack_ctsn, ctsn, tsn; 1238 __u32 highest_tsn, highest_new_tsn; 1239 __u32 sack_a_rwnd; 1240 unsigned int outstanding; 1241 struct sctp_transport *primary = asoc->peer.primary_path; 1242 int count_of_newacks = 0; 1243 int gap_ack_blocks; 1244 u8 accum_moved = 0; 1245 1246 /* Grab the association's destination address list. */ 1247 transport_list = &asoc->peer.transport_addr_list; 1248 1249 sack_ctsn = ntohl(sack->cum_tsn_ack); 1250 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1251 asoc->stats.gapcnt += gap_ack_blocks; 1252 /* 1253 * SFR-CACC algorithm: 1254 * On receipt of a SACK the sender SHOULD execute the 1255 * following statements. 1256 * 1257 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1258 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1259 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1260 * all destinations. 1261 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1262 * is set the receiver of the SACK MUST take the following actions: 1263 * 1264 * A) Initialize the cacc_saw_newack to 0 for all destination 1265 * addresses. 1266 * 1267 * Only bother if changeover_active is set. Otherwise, this is 1268 * totally suboptimal to do on every SACK. 1269 */ 1270 if (primary->cacc.changeover_active) { 1271 u8 clear_cycling = 0; 1272 1273 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1274 primary->cacc.changeover_active = 0; 1275 clear_cycling = 1; 1276 } 1277 1278 if (clear_cycling || gap_ack_blocks) { 1279 list_for_each_entry(transport, transport_list, 1280 transports) { 1281 if (clear_cycling) 1282 transport->cacc.cycling_changeover = 0; 1283 if (gap_ack_blocks) 1284 transport->cacc.cacc_saw_newack = 0; 1285 } 1286 } 1287 } 1288 1289 /* Get the highest TSN in the sack. */ 1290 highest_tsn = sack_ctsn; 1291 if (gap_ack_blocks) 1292 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1293 1294 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1295 asoc->highest_sacked = highest_tsn; 1296 1297 highest_new_tsn = sack_ctsn; 1298 1299 /* Run through the retransmit queue. Credit bytes received 1300 * and free those chunks that we can. 1301 */ 1302 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1303 1304 /* Run through the transmitted queue. 1305 * Credit bytes received and free those chunks which we can. 1306 * 1307 * This is a MASSIVE candidate for optimization. 1308 */ 1309 list_for_each_entry(transport, transport_list, transports) { 1310 sctp_check_transmitted(q, &transport->transmitted, 1311 transport, &chunk->source, sack, 1312 &highest_new_tsn); 1313 /* 1314 * SFR-CACC algorithm: 1315 * C) Let count_of_newacks be the number of 1316 * destinations for which cacc_saw_newack is set. 1317 */ 1318 if (transport->cacc.cacc_saw_newack) 1319 count_of_newacks++; 1320 } 1321 1322 /* Move the Cumulative TSN Ack Point if appropriate. */ 1323 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1324 asoc->ctsn_ack_point = sack_ctsn; 1325 accum_moved = 1; 1326 } 1327 1328 if (gap_ack_blocks) { 1329 1330 if (asoc->fast_recovery && accum_moved) 1331 highest_new_tsn = highest_tsn; 1332 1333 list_for_each_entry(transport, transport_list, transports) 1334 sctp_mark_missing(q, &transport->transmitted, transport, 1335 highest_new_tsn, count_of_newacks); 1336 } 1337 1338 /* Update unack_data field in the assoc. */ 1339 sctp_sack_update_unack_data(asoc, sack); 1340 1341 ctsn = asoc->ctsn_ack_point; 1342 1343 /* Throw away stuff rotting on the sack queue. */ 1344 list_for_each_safe(lchunk, temp, &q->sacked) { 1345 tchunk = list_entry(lchunk, struct sctp_chunk, 1346 transmitted_list); 1347 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1348 if (TSN_lte(tsn, ctsn)) { 1349 list_del_init(&tchunk->transmitted_list); 1350 if (asoc->prsctp_enable && 1351 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1352 asoc->sent_cnt_removable--; 1353 sctp_chunk_free(tchunk); 1354 } 1355 } 1356 1357 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1358 * number of bytes still outstanding after processing the 1359 * Cumulative TSN Ack and the Gap Ack Blocks. 1360 */ 1361 1362 sack_a_rwnd = ntohl(sack->a_rwnd); 1363 asoc->peer.zero_window_announced = !sack_a_rwnd; 1364 outstanding = q->outstanding_bytes; 1365 1366 if (outstanding < sack_a_rwnd) 1367 sack_a_rwnd -= outstanding; 1368 else 1369 sack_a_rwnd = 0; 1370 1371 asoc->peer.rwnd = sack_a_rwnd; 1372 1373 sctp_generate_fwdtsn(q, sack_ctsn); 1374 1375 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1376 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1377 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1378 asoc->adv_peer_ack_point); 1379 1380 return sctp_outq_is_empty(q); 1381 } 1382 1383 /* Is the outqueue empty? 1384 * The queue is empty when we have not pending data, no in-flight data 1385 * and nothing pending retransmissions. 1386 */ 1387 int sctp_outq_is_empty(const struct sctp_outq *q) 1388 { 1389 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1390 list_empty(&q->retransmit); 1391 } 1392 1393 /******************************************************************** 1394 * 2nd Level Abstractions 1395 ********************************************************************/ 1396 1397 /* Go through a transport's transmitted list or the association's retransmit 1398 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1399 * The retransmit list will not have an associated transport. 1400 * 1401 * I added coherent debug information output. --xguo 1402 * 1403 * Instead of printing 'sacked' or 'kept' for each TSN on the 1404 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1405 * KEPT TSN6-TSN7, etc. 1406 */ 1407 static void sctp_check_transmitted(struct sctp_outq *q, 1408 struct list_head *transmitted_queue, 1409 struct sctp_transport *transport, 1410 union sctp_addr *saddr, 1411 struct sctp_sackhdr *sack, 1412 __u32 *highest_new_tsn_in_sack) 1413 { 1414 struct list_head *lchunk; 1415 struct sctp_chunk *tchunk; 1416 struct list_head tlist; 1417 __u32 tsn; 1418 __u32 sack_ctsn; 1419 __u32 rtt; 1420 __u8 restart_timer = 0; 1421 int bytes_acked = 0; 1422 int migrate_bytes = 0; 1423 bool forward_progress = false; 1424 1425 sack_ctsn = ntohl(sack->cum_tsn_ack); 1426 1427 INIT_LIST_HEAD(&tlist); 1428 1429 /* The while loop will skip empty transmitted queues. */ 1430 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1431 tchunk = list_entry(lchunk, struct sctp_chunk, 1432 transmitted_list); 1433 1434 if (sctp_chunk_abandoned(tchunk)) { 1435 /* Move the chunk to abandoned list. */ 1436 sctp_insert_list(&q->abandoned, lchunk); 1437 1438 /* If this chunk has not been acked, stop 1439 * considering it as 'outstanding'. 1440 */ 1441 if (!tchunk->tsn_gap_acked) { 1442 if (tchunk->transport) 1443 tchunk->transport->flight_size -= 1444 sctp_data_size(tchunk); 1445 q->outstanding_bytes -= sctp_data_size(tchunk); 1446 } 1447 continue; 1448 } 1449 1450 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1451 if (sctp_acked(sack, tsn)) { 1452 /* If this queue is the retransmit queue, the 1453 * retransmit timer has already reclaimed 1454 * the outstanding bytes for this chunk, so only 1455 * count bytes associated with a transport. 1456 */ 1457 if (transport) { 1458 /* If this chunk is being used for RTT 1459 * measurement, calculate the RTT and update 1460 * the RTO using this value. 1461 * 1462 * 6.3.1 C5) Karn's algorithm: RTT measurements 1463 * MUST NOT be made using packets that were 1464 * retransmitted (and thus for which it is 1465 * ambiguous whether the reply was for the 1466 * first instance of the packet or a later 1467 * instance). 1468 */ 1469 if (!tchunk->tsn_gap_acked && 1470 !tchunk->resent && 1471 tchunk->rtt_in_progress) { 1472 tchunk->rtt_in_progress = 0; 1473 rtt = jiffies - tchunk->sent_at; 1474 sctp_transport_update_rto(transport, 1475 rtt); 1476 } 1477 } 1478 1479 /* If the chunk hasn't been marked as ACKED, 1480 * mark it and account bytes_acked if the 1481 * chunk had a valid transport (it will not 1482 * have a transport if ASCONF had deleted it 1483 * while DATA was outstanding). 1484 */ 1485 if (!tchunk->tsn_gap_acked) { 1486 tchunk->tsn_gap_acked = 1; 1487 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1488 *highest_new_tsn_in_sack = tsn; 1489 bytes_acked += sctp_data_size(tchunk); 1490 if (!tchunk->transport) 1491 migrate_bytes += sctp_data_size(tchunk); 1492 forward_progress = true; 1493 } 1494 1495 if (TSN_lte(tsn, sack_ctsn)) { 1496 /* RFC 2960 6.3.2 Retransmission Timer Rules 1497 * 1498 * R3) Whenever a SACK is received 1499 * that acknowledges the DATA chunk 1500 * with the earliest outstanding TSN 1501 * for that address, restart T3-rtx 1502 * timer for that address with its 1503 * current RTO. 1504 */ 1505 restart_timer = 1; 1506 forward_progress = true; 1507 1508 if (!tchunk->tsn_gap_acked) { 1509 /* 1510 * SFR-CACC algorithm: 1511 * 2) If the SACK contains gap acks 1512 * and the flag CHANGEOVER_ACTIVE is 1513 * set the receiver of the SACK MUST 1514 * take the following action: 1515 * 1516 * B) For each TSN t being acked that 1517 * has not been acked in any SACK so 1518 * far, set cacc_saw_newack to 1 for 1519 * the destination that the TSN was 1520 * sent to. 1521 */ 1522 if (transport && 1523 sack->num_gap_ack_blocks && 1524 q->asoc->peer.primary_path->cacc. 1525 changeover_active) 1526 transport->cacc.cacc_saw_newack 1527 = 1; 1528 } 1529 1530 list_add_tail(&tchunk->transmitted_list, 1531 &q->sacked); 1532 } else { 1533 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1534 * M2) Each time a SACK arrives reporting 1535 * 'Stray DATA chunk(s)' record the highest TSN 1536 * reported as newly acknowledged, call this 1537 * value 'HighestTSNinSack'. A newly 1538 * acknowledged DATA chunk is one not 1539 * previously acknowledged in a SACK. 1540 * 1541 * When the SCTP sender of data receives a SACK 1542 * chunk that acknowledges, for the first time, 1543 * the receipt of a DATA chunk, all the still 1544 * unacknowledged DATA chunks whose TSN is 1545 * older than that newly acknowledged DATA 1546 * chunk, are qualified as 'Stray DATA chunks'. 1547 */ 1548 list_add_tail(lchunk, &tlist); 1549 } 1550 } else { 1551 if (tchunk->tsn_gap_acked) { 1552 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1553 __func__, tsn); 1554 1555 tchunk->tsn_gap_acked = 0; 1556 1557 if (tchunk->transport) 1558 bytes_acked -= sctp_data_size(tchunk); 1559 1560 /* RFC 2960 6.3.2 Retransmission Timer Rules 1561 * 1562 * R4) Whenever a SACK is received missing a 1563 * TSN that was previously acknowledged via a 1564 * Gap Ack Block, start T3-rtx for the 1565 * destination address to which the DATA 1566 * chunk was originally 1567 * transmitted if it is not already running. 1568 */ 1569 restart_timer = 1; 1570 } 1571 1572 list_add_tail(lchunk, &tlist); 1573 } 1574 } 1575 1576 if (transport) { 1577 if (bytes_acked) { 1578 struct sctp_association *asoc = transport->asoc; 1579 1580 /* We may have counted DATA that was migrated 1581 * to this transport due to DEL-IP operation. 1582 * Subtract those bytes, since the were never 1583 * send on this transport and shouldn't be 1584 * credited to this transport. 1585 */ 1586 bytes_acked -= migrate_bytes; 1587 1588 /* 8.2. When an outstanding TSN is acknowledged, 1589 * the endpoint shall clear the error counter of 1590 * the destination transport address to which the 1591 * DATA chunk was last sent. 1592 * The association's overall error counter is 1593 * also cleared. 1594 */ 1595 transport->error_count = 0; 1596 transport->asoc->overall_error_count = 0; 1597 forward_progress = true; 1598 1599 /* 1600 * While in SHUTDOWN PENDING, we may have started 1601 * the T5 shutdown guard timer after reaching the 1602 * retransmission limit. Stop that timer as soon 1603 * as the receiver acknowledged any data. 1604 */ 1605 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1606 del_timer(&asoc->timers 1607 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1608 sctp_association_put(asoc); 1609 1610 /* Mark the destination transport address as 1611 * active if it is not so marked. 1612 */ 1613 if ((transport->state == SCTP_INACTIVE || 1614 transport->state == SCTP_UNCONFIRMED) && 1615 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1616 sctp_assoc_control_transport( 1617 transport->asoc, 1618 transport, 1619 SCTP_TRANSPORT_UP, 1620 SCTP_RECEIVED_SACK); 1621 } 1622 1623 sctp_transport_raise_cwnd(transport, sack_ctsn, 1624 bytes_acked); 1625 1626 transport->flight_size -= bytes_acked; 1627 if (transport->flight_size == 0) 1628 transport->partial_bytes_acked = 0; 1629 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1630 } else { 1631 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1632 * When a sender is doing zero window probing, it 1633 * should not timeout the association if it continues 1634 * to receive new packets from the receiver. The 1635 * reason is that the receiver MAY keep its window 1636 * closed for an indefinite time. 1637 * A sender is doing zero window probing when the 1638 * receiver's advertised window is zero, and there is 1639 * only one data chunk in flight to the receiver. 1640 * 1641 * Allow the association to timeout while in SHUTDOWN 1642 * PENDING or SHUTDOWN RECEIVED in case the receiver 1643 * stays in zero window mode forever. 1644 */ 1645 if (!q->asoc->peer.rwnd && 1646 !list_empty(&tlist) && 1647 (sack_ctsn+2 == q->asoc->next_tsn) && 1648 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1649 pr_debug("%s: sack received for zero window " 1650 "probe:%u\n", __func__, sack_ctsn); 1651 1652 q->asoc->overall_error_count = 0; 1653 transport->error_count = 0; 1654 } 1655 } 1656 1657 /* RFC 2960 6.3.2 Retransmission Timer Rules 1658 * 1659 * R2) Whenever all outstanding data sent to an address have 1660 * been acknowledged, turn off the T3-rtx timer of that 1661 * address. 1662 */ 1663 if (!transport->flight_size) { 1664 if (del_timer(&transport->T3_rtx_timer)) 1665 sctp_transport_put(transport); 1666 } else if (restart_timer) { 1667 if (!mod_timer(&transport->T3_rtx_timer, 1668 jiffies + transport->rto)) 1669 sctp_transport_hold(transport); 1670 } 1671 1672 if (forward_progress) { 1673 if (transport->dst) 1674 dst_confirm(transport->dst); 1675 } 1676 } 1677 1678 list_splice(&tlist, transmitted_queue); 1679 } 1680 1681 /* Mark chunks as missing and consequently may get retransmitted. */ 1682 static void sctp_mark_missing(struct sctp_outq *q, 1683 struct list_head *transmitted_queue, 1684 struct sctp_transport *transport, 1685 __u32 highest_new_tsn_in_sack, 1686 int count_of_newacks) 1687 { 1688 struct sctp_chunk *chunk; 1689 __u32 tsn; 1690 char do_fast_retransmit = 0; 1691 struct sctp_association *asoc = q->asoc; 1692 struct sctp_transport *primary = asoc->peer.primary_path; 1693 1694 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1695 1696 tsn = ntohl(chunk->subh.data_hdr->tsn); 1697 1698 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1699 * 'Unacknowledged TSN's', if the TSN number of an 1700 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1701 * value, increment the 'TSN.Missing.Report' count on that 1702 * chunk if it has NOT been fast retransmitted or marked for 1703 * fast retransmit already. 1704 */ 1705 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1706 !chunk->tsn_gap_acked && 1707 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1708 1709 /* SFR-CACC may require us to skip marking 1710 * this chunk as missing. 1711 */ 1712 if (!transport || !sctp_cacc_skip(primary, 1713 chunk->transport, 1714 count_of_newacks, tsn)) { 1715 chunk->tsn_missing_report++; 1716 1717 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1718 __func__, tsn, chunk->tsn_missing_report); 1719 } 1720 } 1721 /* 1722 * M4) If any DATA chunk is found to have a 1723 * 'TSN.Missing.Report' 1724 * value larger than or equal to 3, mark that chunk for 1725 * retransmission and start the fast retransmit procedure. 1726 */ 1727 1728 if (chunk->tsn_missing_report >= 3) { 1729 chunk->fast_retransmit = SCTP_NEED_FRTX; 1730 do_fast_retransmit = 1; 1731 } 1732 } 1733 1734 if (transport) { 1735 if (do_fast_retransmit) 1736 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1737 1738 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1739 "flight_size:%d, pba:%d\n", __func__, transport, 1740 transport->cwnd, transport->ssthresh, 1741 transport->flight_size, transport->partial_bytes_acked); 1742 } 1743 } 1744 1745 /* Is the given TSN acked by this packet? */ 1746 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1747 { 1748 int i; 1749 sctp_sack_variable_t *frags; 1750 __u16 gap; 1751 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1752 1753 if (TSN_lte(tsn, ctsn)) 1754 goto pass; 1755 1756 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1757 * 1758 * Gap Ack Blocks: 1759 * These fields contain the Gap Ack Blocks. They are repeated 1760 * for each Gap Ack Block up to the number of Gap Ack Blocks 1761 * defined in the Number of Gap Ack Blocks field. All DATA 1762 * chunks with TSNs greater than or equal to (Cumulative TSN 1763 * Ack + Gap Ack Block Start) and less than or equal to 1764 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1765 * Block are assumed to have been received correctly. 1766 */ 1767 1768 frags = sack->variable; 1769 gap = tsn - ctsn; 1770 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) { 1771 if (TSN_lte(ntohs(frags[i].gab.start), gap) && 1772 TSN_lte(gap, ntohs(frags[i].gab.end))) 1773 goto pass; 1774 } 1775 1776 return 0; 1777 pass: 1778 return 1; 1779 } 1780 1781 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1782 int nskips, __be16 stream) 1783 { 1784 int i; 1785 1786 for (i = 0; i < nskips; i++) { 1787 if (skiplist[i].stream == stream) 1788 return i; 1789 } 1790 return i; 1791 } 1792 1793 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1794 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1795 { 1796 struct sctp_association *asoc = q->asoc; 1797 struct sctp_chunk *ftsn_chunk = NULL; 1798 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1799 int nskips = 0; 1800 int skip_pos = 0; 1801 __u32 tsn; 1802 struct sctp_chunk *chunk; 1803 struct list_head *lchunk, *temp; 1804 1805 if (!asoc->peer.prsctp_capable) 1806 return; 1807 1808 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1809 * received SACK. 1810 * 1811 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1812 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1813 */ 1814 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1815 asoc->adv_peer_ack_point = ctsn; 1816 1817 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1818 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1819 * the chunk next in the out-queue space is marked as "abandoned" as 1820 * shown in the following example: 1821 * 1822 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1823 * and the Advanced.Peer.Ack.Point is updated to this value: 1824 * 1825 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1826 * normal SACK processing local advancement 1827 * ... ... 1828 * Adv.Ack.Pt-> 102 acked 102 acked 1829 * 103 abandoned 103 abandoned 1830 * 104 abandoned Adv.Ack.P-> 104 abandoned 1831 * 105 105 1832 * 106 acked 106 acked 1833 * ... ... 1834 * 1835 * In this example, the data sender successfully advanced the 1836 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1837 */ 1838 list_for_each_safe(lchunk, temp, &q->abandoned) { 1839 chunk = list_entry(lchunk, struct sctp_chunk, 1840 transmitted_list); 1841 tsn = ntohl(chunk->subh.data_hdr->tsn); 1842 1843 /* Remove any chunks in the abandoned queue that are acked by 1844 * the ctsn. 1845 */ 1846 if (TSN_lte(tsn, ctsn)) { 1847 list_del_init(lchunk); 1848 sctp_chunk_free(chunk); 1849 } else { 1850 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1851 asoc->adv_peer_ack_point = tsn; 1852 if (chunk->chunk_hdr->flags & 1853 SCTP_DATA_UNORDERED) 1854 continue; 1855 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1856 nskips, 1857 chunk->subh.data_hdr->stream); 1858 ftsn_skip_arr[skip_pos].stream = 1859 chunk->subh.data_hdr->stream; 1860 ftsn_skip_arr[skip_pos].ssn = 1861 chunk->subh.data_hdr->ssn; 1862 if (skip_pos == nskips) 1863 nskips++; 1864 if (nskips == 10) 1865 break; 1866 } else 1867 break; 1868 } 1869 } 1870 1871 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1872 * is greater than the Cumulative TSN ACK carried in the received 1873 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1874 * chunk containing the latest value of the 1875 * "Advanced.Peer.Ack.Point". 1876 * 1877 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1878 * list each stream and sequence number in the forwarded TSN. This 1879 * information will enable the receiver to easily find any 1880 * stranded TSN's waiting on stream reorder queues. Each stream 1881 * SHOULD only be reported once; this means that if multiple 1882 * abandoned messages occur in the same stream then only the 1883 * highest abandoned stream sequence number is reported. If the 1884 * total size of the FORWARD TSN does NOT fit in a single MTU then 1885 * the sender of the FORWARD TSN SHOULD lower the 1886 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1887 * single MTU. 1888 */ 1889 if (asoc->adv_peer_ack_point > ctsn) 1890 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1891 nskips, &ftsn_skip_arr[0]); 1892 1893 if (ftsn_chunk) { 1894 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1895 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS); 1896 } 1897 } 1898