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 status = sctp_packet_transmit_chunk(packet, chunk, 919 one_packet, gfp); 920 if (status != SCTP_XMIT_OK) { 921 /* put the chunk back */ 922 list_add(&chunk->list, &q->control_chunk_list); 923 } else { 924 asoc->stats.octrlchunks++; 925 /* PR-SCTP C5) If a FORWARD TSN is sent, the 926 * sender MUST assure that at least one T3-rtx 927 * timer is running. 928 */ 929 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) { 930 sctp_transport_reset_t3_rtx(transport); 931 transport->last_time_sent = jiffies; 932 } 933 } 934 break; 935 936 default: 937 /* We built a chunk with an illegal type! */ 938 BUG(); 939 } 940 } 941 942 if (q->asoc->src_out_of_asoc_ok) 943 goto sctp_flush_out; 944 945 /* Is it OK to send data chunks? */ 946 switch (asoc->state) { 947 case SCTP_STATE_COOKIE_ECHOED: 948 /* Only allow bundling when this packet has a COOKIE-ECHO 949 * chunk. 950 */ 951 if (!packet || !packet->has_cookie_echo) 952 break; 953 954 /* fallthru */ 955 case SCTP_STATE_ESTABLISHED: 956 case SCTP_STATE_SHUTDOWN_PENDING: 957 case SCTP_STATE_SHUTDOWN_RECEIVED: 958 /* 959 * RFC 2960 6.1 Transmission of DATA Chunks 960 * 961 * C) When the time comes for the sender to transmit, 962 * before sending new DATA chunks, the sender MUST 963 * first transmit any outstanding DATA chunks which 964 * are marked for retransmission (limited by the 965 * current cwnd). 966 */ 967 if (!list_empty(&q->retransmit)) { 968 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED) 969 goto sctp_flush_out; 970 if (transport == asoc->peer.retran_path) 971 goto retran; 972 973 /* Switch transports & prepare the packet. */ 974 975 transport = asoc->peer.retran_path; 976 977 if (list_empty(&transport->send_ready)) { 978 list_add_tail(&transport->send_ready, 979 &transport_list); 980 } 981 982 packet = &transport->packet; 983 sctp_packet_config(packet, vtag, 984 asoc->peer.ecn_capable); 985 retran: 986 error = sctp_outq_flush_rtx(q, packet, 987 rtx_timeout, &start_timer); 988 if (error < 0) 989 asoc->base.sk->sk_err = -error; 990 991 if (start_timer) { 992 sctp_transport_reset_t3_rtx(transport); 993 transport->last_time_sent = jiffies; 994 } 995 996 /* This can happen on COOKIE-ECHO resend. Only 997 * one chunk can get bundled with a COOKIE-ECHO. 998 */ 999 if (packet->has_cookie_echo) 1000 goto sctp_flush_out; 1001 1002 /* Don't send new data if there is still data 1003 * waiting to retransmit. 1004 */ 1005 if (!list_empty(&q->retransmit)) 1006 goto sctp_flush_out; 1007 } 1008 1009 /* Apply Max.Burst limitation to the current transport in 1010 * case it will be used for new data. We are going to 1011 * rest it before we return, but we want to apply the limit 1012 * to the currently queued data. 1013 */ 1014 if (transport) 1015 sctp_transport_burst_limited(transport); 1016 1017 /* Finally, transmit new packets. */ 1018 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 1019 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid 1020 * stream identifier. 1021 */ 1022 if (chunk->sinfo.sinfo_stream >= 1023 asoc->c.sinit_num_ostreams) { 1024 1025 /* Mark as failed send. */ 1026 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM); 1027 if (asoc->peer.prsctp_capable && 1028 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1029 asoc->sent_cnt_removable--; 1030 sctp_chunk_free(chunk); 1031 continue; 1032 } 1033 1034 /* Has this chunk expired? */ 1035 if (sctp_chunk_abandoned(chunk)) { 1036 sctp_chunk_fail(chunk, 0); 1037 sctp_chunk_free(chunk); 1038 continue; 1039 } 1040 1041 /* If there is a specified transport, use it. 1042 * Otherwise, we want to use the active path. 1043 */ 1044 new_transport = chunk->transport; 1045 if (!new_transport || 1046 ((new_transport->state == SCTP_INACTIVE) || 1047 (new_transport->state == SCTP_UNCONFIRMED) || 1048 (new_transport->state == SCTP_PF))) 1049 new_transport = asoc->peer.active_path; 1050 if (new_transport->state == SCTP_UNCONFIRMED) { 1051 WARN_ONCE(1, "Atempt to send packet on unconfirmed path."); 1052 sctp_chunk_fail(chunk, 0); 1053 sctp_chunk_free(chunk); 1054 continue; 1055 } 1056 1057 /* Change packets if necessary. */ 1058 if (new_transport != transport) { 1059 transport = new_transport; 1060 1061 /* Schedule to have this transport's 1062 * packet flushed. 1063 */ 1064 if (list_empty(&transport->send_ready)) { 1065 list_add_tail(&transport->send_ready, 1066 &transport_list); 1067 } 1068 1069 packet = &transport->packet; 1070 sctp_packet_config(packet, vtag, 1071 asoc->peer.ecn_capable); 1072 /* We've switched transports, so apply the 1073 * Burst limit to the new transport. 1074 */ 1075 sctp_transport_burst_limited(transport); 1076 } 1077 1078 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p " 1079 "skb->users:%d\n", 1080 __func__, q, chunk, chunk && chunk->chunk_hdr ? 1081 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) : 1082 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn), 1083 chunk->skb ? chunk->skb->head : NULL, chunk->skb ? 1084 atomic_read(&chunk->skb->users) : -1); 1085 1086 /* Add the chunk to the packet. */ 1087 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp); 1088 1089 switch (status) { 1090 case SCTP_XMIT_PMTU_FULL: 1091 case SCTP_XMIT_RWND_FULL: 1092 case SCTP_XMIT_DELAY: 1093 /* We could not append this chunk, so put 1094 * the chunk back on the output queue. 1095 */ 1096 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n", 1097 __func__, ntohl(chunk->subh.data_hdr->tsn), 1098 status); 1099 1100 sctp_outq_head_data(q, chunk); 1101 goto sctp_flush_out; 1102 1103 case SCTP_XMIT_OK: 1104 /* The sender is in the SHUTDOWN-PENDING state, 1105 * The sender MAY set the I-bit in the DATA 1106 * chunk header. 1107 */ 1108 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1109 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1110 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 1111 asoc->stats.ouodchunks++; 1112 else 1113 asoc->stats.oodchunks++; 1114 1115 break; 1116 1117 default: 1118 BUG(); 1119 } 1120 1121 /* BUG: We assume that the sctp_packet_transmit() 1122 * call below will succeed all the time and add the 1123 * chunk to the transmitted list and restart the 1124 * timers. 1125 * It is possible that the call can fail under OOM 1126 * conditions. 1127 * 1128 * Is this really a problem? Won't this behave 1129 * like a lost TSN? 1130 */ 1131 list_add_tail(&chunk->transmitted_list, 1132 &transport->transmitted); 1133 1134 sctp_transport_reset_t3_rtx(transport); 1135 transport->last_time_sent = jiffies; 1136 1137 /* Only let one DATA chunk get bundled with a 1138 * COOKIE-ECHO chunk. 1139 */ 1140 if (packet->has_cookie_echo) 1141 goto sctp_flush_out; 1142 } 1143 break; 1144 1145 default: 1146 /* Do nothing. */ 1147 break; 1148 } 1149 1150 sctp_flush_out: 1151 1152 /* Before returning, examine all the transports touched in 1153 * this call. Right now, we bluntly force clear all the 1154 * transports. Things might change after we implement Nagle. 1155 * But such an examination is still required. 1156 * 1157 * --xguo 1158 */ 1159 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) { 1160 struct sctp_transport *t = list_entry(ltransport, 1161 struct sctp_transport, 1162 send_ready); 1163 packet = &t->packet; 1164 if (!sctp_packet_empty(packet)) { 1165 error = sctp_packet_transmit(packet, gfp); 1166 if (error < 0) 1167 asoc->base.sk->sk_err = -error; 1168 } 1169 1170 /* Clear the burst limited state, if any */ 1171 sctp_transport_burst_reset(t); 1172 } 1173 } 1174 1175 /* Update unack_data based on the incoming SACK chunk */ 1176 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1177 struct sctp_sackhdr *sack) 1178 { 1179 sctp_sack_variable_t *frags; 1180 __u16 unack_data; 1181 int i; 1182 1183 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1184 1185 frags = sack->variable; 1186 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1187 unack_data -= ((ntohs(frags[i].gab.end) - 1188 ntohs(frags[i].gab.start) + 1)); 1189 } 1190 1191 assoc->unack_data = unack_data; 1192 } 1193 1194 /* This is where we REALLY process a SACK. 1195 * 1196 * Process the SACK against the outqueue. Mostly, this just frees 1197 * things off the transmitted queue. 1198 */ 1199 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk) 1200 { 1201 struct sctp_association *asoc = q->asoc; 1202 struct sctp_sackhdr *sack = chunk->subh.sack_hdr; 1203 struct sctp_transport *transport; 1204 struct sctp_chunk *tchunk = NULL; 1205 struct list_head *lchunk, *transport_list, *temp; 1206 sctp_sack_variable_t *frags = sack->variable; 1207 __u32 sack_ctsn, ctsn, tsn; 1208 __u32 highest_tsn, highest_new_tsn; 1209 __u32 sack_a_rwnd; 1210 unsigned int outstanding; 1211 struct sctp_transport *primary = asoc->peer.primary_path; 1212 int count_of_newacks = 0; 1213 int gap_ack_blocks; 1214 u8 accum_moved = 0; 1215 1216 /* Grab the association's destination address list. */ 1217 transport_list = &asoc->peer.transport_addr_list; 1218 1219 sack_ctsn = ntohl(sack->cum_tsn_ack); 1220 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1221 asoc->stats.gapcnt += gap_ack_blocks; 1222 /* 1223 * SFR-CACC algorithm: 1224 * On receipt of a SACK the sender SHOULD execute the 1225 * following statements. 1226 * 1227 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1228 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1229 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1230 * all destinations. 1231 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1232 * is set the receiver of the SACK MUST take the following actions: 1233 * 1234 * A) Initialize the cacc_saw_newack to 0 for all destination 1235 * addresses. 1236 * 1237 * Only bother if changeover_active is set. Otherwise, this is 1238 * totally suboptimal to do on every SACK. 1239 */ 1240 if (primary->cacc.changeover_active) { 1241 u8 clear_cycling = 0; 1242 1243 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1244 primary->cacc.changeover_active = 0; 1245 clear_cycling = 1; 1246 } 1247 1248 if (clear_cycling || gap_ack_blocks) { 1249 list_for_each_entry(transport, transport_list, 1250 transports) { 1251 if (clear_cycling) 1252 transport->cacc.cycling_changeover = 0; 1253 if (gap_ack_blocks) 1254 transport->cacc.cacc_saw_newack = 0; 1255 } 1256 } 1257 } 1258 1259 /* Get the highest TSN in the sack. */ 1260 highest_tsn = sack_ctsn; 1261 if (gap_ack_blocks) 1262 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1263 1264 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1265 asoc->highest_sacked = highest_tsn; 1266 1267 highest_new_tsn = sack_ctsn; 1268 1269 /* Run through the retransmit queue. Credit bytes received 1270 * and free those chunks that we can. 1271 */ 1272 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn); 1273 1274 /* Run through the transmitted queue. 1275 * Credit bytes received and free those chunks which we can. 1276 * 1277 * This is a MASSIVE candidate for optimization. 1278 */ 1279 list_for_each_entry(transport, transport_list, transports) { 1280 sctp_check_transmitted(q, &transport->transmitted, 1281 transport, &chunk->source, sack, 1282 &highest_new_tsn); 1283 /* 1284 * SFR-CACC algorithm: 1285 * C) Let count_of_newacks be the number of 1286 * destinations for which cacc_saw_newack is set. 1287 */ 1288 if (transport->cacc.cacc_saw_newack) 1289 count_of_newacks++; 1290 } 1291 1292 /* Move the Cumulative TSN Ack Point if appropriate. */ 1293 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1294 asoc->ctsn_ack_point = sack_ctsn; 1295 accum_moved = 1; 1296 } 1297 1298 if (gap_ack_blocks) { 1299 1300 if (asoc->fast_recovery && accum_moved) 1301 highest_new_tsn = highest_tsn; 1302 1303 list_for_each_entry(transport, transport_list, transports) 1304 sctp_mark_missing(q, &transport->transmitted, transport, 1305 highest_new_tsn, count_of_newacks); 1306 } 1307 1308 /* Update unack_data field in the assoc. */ 1309 sctp_sack_update_unack_data(asoc, sack); 1310 1311 ctsn = asoc->ctsn_ack_point; 1312 1313 /* Throw away stuff rotting on the sack queue. */ 1314 list_for_each_safe(lchunk, temp, &q->sacked) { 1315 tchunk = list_entry(lchunk, struct sctp_chunk, 1316 transmitted_list); 1317 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1318 if (TSN_lte(tsn, ctsn)) { 1319 list_del_init(&tchunk->transmitted_list); 1320 if (asoc->peer.prsctp_capable && 1321 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags)) 1322 asoc->sent_cnt_removable--; 1323 sctp_chunk_free(tchunk); 1324 } 1325 } 1326 1327 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1328 * number of bytes still outstanding after processing the 1329 * Cumulative TSN Ack and the Gap Ack Blocks. 1330 */ 1331 1332 sack_a_rwnd = ntohl(sack->a_rwnd); 1333 asoc->peer.zero_window_announced = !sack_a_rwnd; 1334 outstanding = q->outstanding_bytes; 1335 1336 if (outstanding < sack_a_rwnd) 1337 sack_a_rwnd -= outstanding; 1338 else 1339 sack_a_rwnd = 0; 1340 1341 asoc->peer.rwnd = sack_a_rwnd; 1342 1343 sctp_generate_fwdtsn(q, sack_ctsn); 1344 1345 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn); 1346 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, " 1347 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn, 1348 asoc->adv_peer_ack_point); 1349 1350 return sctp_outq_is_empty(q); 1351 } 1352 1353 /* Is the outqueue empty? 1354 * The queue is empty when we have not pending data, no in-flight data 1355 * and nothing pending retransmissions. 1356 */ 1357 int sctp_outq_is_empty(const struct sctp_outq *q) 1358 { 1359 return q->out_qlen == 0 && q->outstanding_bytes == 0 && 1360 list_empty(&q->retransmit); 1361 } 1362 1363 /******************************************************************** 1364 * 2nd Level Abstractions 1365 ********************************************************************/ 1366 1367 /* Go through a transport's transmitted list or the association's retransmit 1368 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1369 * The retransmit list will not have an associated transport. 1370 * 1371 * I added coherent debug information output. --xguo 1372 * 1373 * Instead of printing 'sacked' or 'kept' for each TSN on the 1374 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1375 * KEPT TSN6-TSN7, etc. 1376 */ 1377 static void sctp_check_transmitted(struct sctp_outq *q, 1378 struct list_head *transmitted_queue, 1379 struct sctp_transport *transport, 1380 union sctp_addr *saddr, 1381 struct sctp_sackhdr *sack, 1382 __u32 *highest_new_tsn_in_sack) 1383 { 1384 struct list_head *lchunk; 1385 struct sctp_chunk *tchunk; 1386 struct list_head tlist; 1387 __u32 tsn; 1388 __u32 sack_ctsn; 1389 __u32 rtt; 1390 __u8 restart_timer = 0; 1391 int bytes_acked = 0; 1392 int migrate_bytes = 0; 1393 bool forward_progress = false; 1394 1395 sack_ctsn = ntohl(sack->cum_tsn_ack); 1396 1397 INIT_LIST_HEAD(&tlist); 1398 1399 /* The while loop will skip empty transmitted queues. */ 1400 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1401 tchunk = list_entry(lchunk, struct sctp_chunk, 1402 transmitted_list); 1403 1404 if (sctp_chunk_abandoned(tchunk)) { 1405 /* Move the chunk to abandoned list. */ 1406 sctp_insert_list(&q->abandoned, lchunk); 1407 1408 /* If this chunk has not been acked, stop 1409 * considering it as 'outstanding'. 1410 */ 1411 if (!tchunk->tsn_gap_acked) { 1412 if (tchunk->transport) 1413 tchunk->transport->flight_size -= 1414 sctp_data_size(tchunk); 1415 q->outstanding_bytes -= sctp_data_size(tchunk); 1416 } 1417 continue; 1418 } 1419 1420 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1421 if (sctp_acked(sack, tsn)) { 1422 /* If this queue is the retransmit queue, the 1423 * retransmit timer has already reclaimed 1424 * the outstanding bytes for this chunk, so only 1425 * count bytes associated with a transport. 1426 */ 1427 if (transport) { 1428 /* If this chunk is being used for RTT 1429 * measurement, calculate the RTT and update 1430 * the RTO using this value. 1431 * 1432 * 6.3.1 C5) Karn's algorithm: RTT measurements 1433 * MUST NOT be made using packets that were 1434 * retransmitted (and thus for which it is 1435 * ambiguous whether the reply was for the 1436 * first instance of the packet or a later 1437 * instance). 1438 */ 1439 if (!tchunk->tsn_gap_acked && 1440 !sctp_chunk_retransmitted(tchunk) && 1441 tchunk->rtt_in_progress) { 1442 tchunk->rtt_in_progress = 0; 1443 rtt = jiffies - tchunk->sent_at; 1444 sctp_transport_update_rto(transport, 1445 rtt); 1446 } 1447 } 1448 1449 /* If the chunk hasn't been marked as ACKED, 1450 * mark it and account bytes_acked if the 1451 * chunk had a valid transport (it will not 1452 * have a transport if ASCONF had deleted it 1453 * while DATA was outstanding). 1454 */ 1455 if (!tchunk->tsn_gap_acked) { 1456 tchunk->tsn_gap_acked = 1; 1457 if (TSN_lt(*highest_new_tsn_in_sack, tsn)) 1458 *highest_new_tsn_in_sack = tsn; 1459 bytes_acked += sctp_data_size(tchunk); 1460 if (!tchunk->transport) 1461 migrate_bytes += sctp_data_size(tchunk); 1462 forward_progress = true; 1463 } 1464 1465 if (TSN_lte(tsn, sack_ctsn)) { 1466 /* RFC 2960 6.3.2 Retransmission Timer Rules 1467 * 1468 * R3) Whenever a SACK is received 1469 * that acknowledges the DATA chunk 1470 * with the earliest outstanding TSN 1471 * for that address, restart T3-rtx 1472 * timer for that address with its 1473 * current RTO. 1474 */ 1475 restart_timer = 1; 1476 forward_progress = true; 1477 1478 if (!tchunk->tsn_gap_acked) { 1479 /* 1480 * SFR-CACC algorithm: 1481 * 2) If the SACK contains gap acks 1482 * and the flag CHANGEOVER_ACTIVE is 1483 * set the receiver of the SACK MUST 1484 * take the following action: 1485 * 1486 * B) For each TSN t being acked that 1487 * has not been acked in any SACK so 1488 * far, set cacc_saw_newack to 1 for 1489 * the destination that the TSN was 1490 * sent to. 1491 */ 1492 if (transport && 1493 sack->num_gap_ack_blocks && 1494 q->asoc->peer.primary_path->cacc. 1495 changeover_active) 1496 transport->cacc.cacc_saw_newack 1497 = 1; 1498 } 1499 1500 list_add_tail(&tchunk->transmitted_list, 1501 &q->sacked); 1502 } else { 1503 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1504 * M2) Each time a SACK arrives reporting 1505 * 'Stray DATA chunk(s)' record the highest TSN 1506 * reported as newly acknowledged, call this 1507 * value 'HighestTSNinSack'. A newly 1508 * acknowledged DATA chunk is one not 1509 * previously acknowledged in a SACK. 1510 * 1511 * When the SCTP sender of data receives a SACK 1512 * chunk that acknowledges, for the first time, 1513 * the receipt of a DATA chunk, all the still 1514 * unacknowledged DATA chunks whose TSN is 1515 * older than that newly acknowledged DATA 1516 * chunk, are qualified as 'Stray DATA chunks'. 1517 */ 1518 list_add_tail(lchunk, &tlist); 1519 } 1520 } else { 1521 if (tchunk->tsn_gap_acked) { 1522 pr_debug("%s: receiver reneged on data TSN:0x%x\n", 1523 __func__, tsn); 1524 1525 tchunk->tsn_gap_acked = 0; 1526 1527 if (tchunk->transport) 1528 bytes_acked -= sctp_data_size(tchunk); 1529 1530 /* RFC 2960 6.3.2 Retransmission Timer Rules 1531 * 1532 * R4) Whenever a SACK is received missing a 1533 * TSN that was previously acknowledged via a 1534 * Gap Ack Block, start T3-rtx for the 1535 * destination address to which the DATA 1536 * chunk was originally 1537 * transmitted if it is not already running. 1538 */ 1539 restart_timer = 1; 1540 } 1541 1542 list_add_tail(lchunk, &tlist); 1543 } 1544 } 1545 1546 if (transport) { 1547 if (bytes_acked) { 1548 struct sctp_association *asoc = transport->asoc; 1549 1550 /* We may have counted DATA that was migrated 1551 * to this transport due to DEL-IP operation. 1552 * Subtract those bytes, since the were never 1553 * send on this transport and shouldn't be 1554 * credited to this transport. 1555 */ 1556 bytes_acked -= migrate_bytes; 1557 1558 /* 8.2. When an outstanding TSN is acknowledged, 1559 * the endpoint shall clear the error counter of 1560 * the destination transport address to which the 1561 * DATA chunk was last sent. 1562 * The association's overall error counter is 1563 * also cleared. 1564 */ 1565 transport->error_count = 0; 1566 transport->asoc->overall_error_count = 0; 1567 forward_progress = true; 1568 1569 /* 1570 * While in SHUTDOWN PENDING, we may have started 1571 * the T5 shutdown guard timer after reaching the 1572 * retransmission limit. Stop that timer as soon 1573 * as the receiver acknowledged any data. 1574 */ 1575 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1576 del_timer(&asoc->timers 1577 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1578 sctp_association_put(asoc); 1579 1580 /* Mark the destination transport address as 1581 * active if it is not so marked. 1582 */ 1583 if ((transport->state == SCTP_INACTIVE || 1584 transport->state == SCTP_UNCONFIRMED) && 1585 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) { 1586 sctp_assoc_control_transport( 1587 transport->asoc, 1588 transport, 1589 SCTP_TRANSPORT_UP, 1590 SCTP_RECEIVED_SACK); 1591 } 1592 1593 sctp_transport_raise_cwnd(transport, sack_ctsn, 1594 bytes_acked); 1595 1596 transport->flight_size -= bytes_acked; 1597 if (transport->flight_size == 0) 1598 transport->partial_bytes_acked = 0; 1599 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1600 } else { 1601 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1602 * When a sender is doing zero window probing, it 1603 * should not timeout the association if it continues 1604 * to receive new packets from the receiver. The 1605 * reason is that the receiver MAY keep its window 1606 * closed for an indefinite time. 1607 * A sender is doing zero window probing when the 1608 * receiver's advertised window is zero, and there is 1609 * only one data chunk in flight to the receiver. 1610 * 1611 * Allow the association to timeout while in SHUTDOWN 1612 * PENDING or SHUTDOWN RECEIVED in case the receiver 1613 * stays in zero window mode forever. 1614 */ 1615 if (!q->asoc->peer.rwnd && 1616 !list_empty(&tlist) && 1617 (sack_ctsn+2 == q->asoc->next_tsn) && 1618 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1619 pr_debug("%s: sack received for zero window " 1620 "probe:%u\n", __func__, sack_ctsn); 1621 1622 q->asoc->overall_error_count = 0; 1623 transport->error_count = 0; 1624 } 1625 } 1626 1627 /* RFC 2960 6.3.2 Retransmission Timer Rules 1628 * 1629 * R2) Whenever all outstanding data sent to an address have 1630 * been acknowledged, turn off the T3-rtx timer of that 1631 * address. 1632 */ 1633 if (!transport->flight_size) { 1634 if (del_timer(&transport->T3_rtx_timer)) 1635 sctp_transport_put(transport); 1636 } else if (restart_timer) { 1637 if (!mod_timer(&transport->T3_rtx_timer, 1638 jiffies + transport->rto)) 1639 sctp_transport_hold(transport); 1640 } 1641 1642 if (forward_progress) { 1643 if (transport->dst) 1644 dst_confirm(transport->dst); 1645 } 1646 } 1647 1648 list_splice(&tlist, transmitted_queue); 1649 } 1650 1651 /* Mark chunks as missing and consequently may get retransmitted. */ 1652 static void sctp_mark_missing(struct sctp_outq *q, 1653 struct list_head *transmitted_queue, 1654 struct sctp_transport *transport, 1655 __u32 highest_new_tsn_in_sack, 1656 int count_of_newacks) 1657 { 1658 struct sctp_chunk *chunk; 1659 __u32 tsn; 1660 char do_fast_retransmit = 0; 1661 struct sctp_association *asoc = q->asoc; 1662 struct sctp_transport *primary = asoc->peer.primary_path; 1663 1664 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1665 1666 tsn = ntohl(chunk->subh.data_hdr->tsn); 1667 1668 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1669 * 'Unacknowledged TSN's', if the TSN number of an 1670 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1671 * value, increment the 'TSN.Missing.Report' count on that 1672 * chunk if it has NOT been fast retransmitted or marked for 1673 * fast retransmit already. 1674 */ 1675 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1676 !chunk->tsn_gap_acked && 1677 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1678 1679 /* SFR-CACC may require us to skip marking 1680 * this chunk as missing. 1681 */ 1682 if (!transport || !sctp_cacc_skip(primary, 1683 chunk->transport, 1684 count_of_newacks, tsn)) { 1685 chunk->tsn_missing_report++; 1686 1687 pr_debug("%s: tsn:0x%x missing counter:%d\n", 1688 __func__, tsn, chunk->tsn_missing_report); 1689 } 1690 } 1691 /* 1692 * M4) If any DATA chunk is found to have a 1693 * 'TSN.Missing.Report' 1694 * value larger than or equal to 3, mark that chunk for 1695 * retransmission and start the fast retransmit procedure. 1696 */ 1697 1698 if (chunk->tsn_missing_report >= 3) { 1699 chunk->fast_retransmit = SCTP_NEED_FRTX; 1700 do_fast_retransmit = 1; 1701 } 1702 } 1703 1704 if (transport) { 1705 if (do_fast_retransmit) 1706 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1707 1708 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, " 1709 "flight_size:%d, pba:%d\n", __func__, transport, 1710 transport->cwnd, transport->ssthresh, 1711 transport->flight_size, transport->partial_bytes_acked); 1712 } 1713 } 1714 1715 /* Is the given TSN acked by this packet? */ 1716 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1717 { 1718 int i; 1719 sctp_sack_variable_t *frags; 1720 __u16 tsn_offset, blocks; 1721 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1722 1723 if (TSN_lte(tsn, ctsn)) 1724 goto pass; 1725 1726 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1727 * 1728 * Gap Ack Blocks: 1729 * These fields contain the Gap Ack Blocks. They are repeated 1730 * for each Gap Ack Block up to the number of Gap Ack Blocks 1731 * defined in the Number of Gap Ack Blocks field. All DATA 1732 * chunks with TSNs greater than or equal to (Cumulative TSN 1733 * Ack + Gap Ack Block Start) and less than or equal to 1734 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1735 * Block are assumed to have been received correctly. 1736 */ 1737 1738 frags = sack->variable; 1739 blocks = ntohs(sack->num_gap_ack_blocks); 1740 tsn_offset = tsn - ctsn; 1741 for (i = 0; i < blocks; ++i) { 1742 if (tsn_offset >= ntohs(frags[i].gab.start) && 1743 tsn_offset <= ntohs(frags[i].gab.end)) 1744 goto pass; 1745 } 1746 1747 return 0; 1748 pass: 1749 return 1; 1750 } 1751 1752 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1753 int nskips, __be16 stream) 1754 { 1755 int i; 1756 1757 for (i = 0; i < nskips; i++) { 1758 if (skiplist[i].stream == stream) 1759 return i; 1760 } 1761 return i; 1762 } 1763 1764 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1765 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1766 { 1767 struct sctp_association *asoc = q->asoc; 1768 struct sctp_chunk *ftsn_chunk = NULL; 1769 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1770 int nskips = 0; 1771 int skip_pos = 0; 1772 __u32 tsn; 1773 struct sctp_chunk *chunk; 1774 struct list_head *lchunk, *temp; 1775 1776 if (!asoc->peer.prsctp_capable) 1777 return; 1778 1779 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1780 * received SACK. 1781 * 1782 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1783 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1784 */ 1785 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1786 asoc->adv_peer_ack_point = ctsn; 1787 1788 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1789 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1790 * the chunk next in the out-queue space is marked as "abandoned" as 1791 * shown in the following example: 1792 * 1793 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1794 * and the Advanced.Peer.Ack.Point is updated to this value: 1795 * 1796 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1797 * normal SACK processing local advancement 1798 * ... ... 1799 * Adv.Ack.Pt-> 102 acked 102 acked 1800 * 103 abandoned 103 abandoned 1801 * 104 abandoned Adv.Ack.P-> 104 abandoned 1802 * 105 105 1803 * 106 acked 106 acked 1804 * ... ... 1805 * 1806 * In this example, the data sender successfully advanced the 1807 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1808 */ 1809 list_for_each_safe(lchunk, temp, &q->abandoned) { 1810 chunk = list_entry(lchunk, struct sctp_chunk, 1811 transmitted_list); 1812 tsn = ntohl(chunk->subh.data_hdr->tsn); 1813 1814 /* Remove any chunks in the abandoned queue that are acked by 1815 * the ctsn. 1816 */ 1817 if (TSN_lte(tsn, ctsn)) { 1818 list_del_init(lchunk); 1819 sctp_chunk_free(chunk); 1820 } else { 1821 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1822 asoc->adv_peer_ack_point = tsn; 1823 if (chunk->chunk_hdr->flags & 1824 SCTP_DATA_UNORDERED) 1825 continue; 1826 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1827 nskips, 1828 chunk->subh.data_hdr->stream); 1829 ftsn_skip_arr[skip_pos].stream = 1830 chunk->subh.data_hdr->stream; 1831 ftsn_skip_arr[skip_pos].ssn = 1832 chunk->subh.data_hdr->ssn; 1833 if (skip_pos == nskips) 1834 nskips++; 1835 if (nskips == 10) 1836 break; 1837 } else 1838 break; 1839 } 1840 } 1841 1842 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1843 * is greater than the Cumulative TSN ACK carried in the received 1844 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1845 * chunk containing the latest value of the 1846 * "Advanced.Peer.Ack.Point". 1847 * 1848 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1849 * list each stream and sequence number in the forwarded TSN. This 1850 * information will enable the receiver to easily find any 1851 * stranded TSN's waiting on stream reorder queues. Each stream 1852 * SHOULD only be reported once; this means that if multiple 1853 * abandoned messages occur in the same stream then only the 1854 * highest abandoned stream sequence number is reported. If the 1855 * total size of the FORWARD TSN does NOT fit in a single MTU then 1856 * the sender of the FORWARD TSN SHOULD lower the 1857 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1858 * single MTU. 1859 */ 1860 if (asoc->adv_peer_ack_point > ctsn) 1861 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1862 nskips, &ftsn_skip_arr[0]); 1863 1864 if (ftsn_chunk) { 1865 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1866 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS); 1867 } 1868 } 1869