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