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, write to 26 * the Free Software Foundation, 59 Temple Place - Suite 330, 27 * Boston, MA 02111-1307, USA. 28 * 29 * Please send any bug reports or fixes you make to the 30 * email address(es): 31 * lksctp developers <lksctp-developers@lists.sourceforge.net> 32 * 33 * Or submit a bug report through the following website: 34 * http://www.sf.net/projects/lksctp 35 * 36 * Written or modified by: 37 * La Monte H.P. Yarroll <piggy@acm.org> 38 * Karl Knutson <karl@athena.chicago.il.us> 39 * Perry Melange <pmelange@null.cc.uic.edu> 40 * Xingang Guo <xingang.guo@intel.com> 41 * Hui Huang <hui.huang@nokia.com> 42 * Sridhar Samudrala <sri@us.ibm.com> 43 * Jon Grimm <jgrimm@us.ibm.com> 44 * 45 * Any bugs reported given to us we will try to fix... any fixes shared will 46 * be incorporated into the next SCTP release. 47 */ 48 49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 50 51 #include <linux/types.h> 52 #include <linux/list.h> /* For struct list_head */ 53 #include <linux/socket.h> 54 #include <linux/ip.h> 55 #include <linux/slab.h> 56 #include <net/sock.h> /* For skb_set_owner_w */ 57 58 #include <net/sctp/sctp.h> 59 #include <net/sctp/sm.h> 60 61 /* Declare internal functions here. */ 62 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn); 63 static void sctp_check_transmitted(struct sctp_outq *q, 64 struct list_head *transmitted_queue, 65 struct sctp_transport *transport, 66 struct sctp_sackhdr *sack, 67 __u32 *highest_new_tsn); 68 69 static void sctp_mark_missing(struct sctp_outq *q, 70 struct list_head *transmitted_queue, 71 struct sctp_transport *transport, 72 __u32 highest_new_tsn, 73 int count_of_newacks); 74 75 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn); 76 77 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout); 78 79 /* Add data to the front of the queue. */ 80 static inline void sctp_outq_head_data(struct sctp_outq *q, 81 struct sctp_chunk *ch) 82 { 83 list_add(&ch->list, &q->out_chunk_list); 84 q->out_qlen += ch->skb->len; 85 } 86 87 /* Take data from the front of the queue. */ 88 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q) 89 { 90 struct sctp_chunk *ch = NULL; 91 92 if (!list_empty(&q->out_chunk_list)) { 93 struct list_head *entry = q->out_chunk_list.next; 94 95 ch = list_entry(entry, struct sctp_chunk, list); 96 list_del_init(entry); 97 q->out_qlen -= ch->skb->len; 98 } 99 return ch; 100 } 101 /* Add data chunk to the end of the queue. */ 102 static inline void sctp_outq_tail_data(struct sctp_outq *q, 103 struct sctp_chunk *ch) 104 { 105 list_add_tail(&ch->list, &q->out_chunk_list); 106 q->out_qlen += ch->skb->len; 107 } 108 109 /* 110 * SFR-CACC algorithm: 111 * D) If count_of_newacks is greater than or equal to 2 112 * and t was not sent to the current primary then the 113 * sender MUST NOT increment missing report count for t. 114 */ 115 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary, 116 struct sctp_transport *transport, 117 int count_of_newacks) 118 { 119 if (count_of_newacks >=2 && transport != primary) 120 return 1; 121 return 0; 122 } 123 124 /* 125 * SFR-CACC algorithm: 126 * F) If count_of_newacks is less than 2, let d be the 127 * destination to which t was sent. If cacc_saw_newack 128 * is 0 for destination d, then the sender MUST NOT 129 * increment missing report count for t. 130 */ 131 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport, 132 int count_of_newacks) 133 { 134 if (count_of_newacks < 2 && 135 (transport && !transport->cacc.cacc_saw_newack)) 136 return 1; 137 return 0; 138 } 139 140 /* 141 * SFR-CACC algorithm: 142 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD 143 * execute steps C, D, F. 144 * 145 * C has been implemented in sctp_outq_sack 146 */ 147 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary, 148 struct sctp_transport *transport, 149 int count_of_newacks) 150 { 151 if (!primary->cacc.cycling_changeover) { 152 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks)) 153 return 1; 154 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks)) 155 return 1; 156 return 0; 157 } 158 return 0; 159 } 160 161 /* 162 * SFR-CACC algorithm: 163 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less 164 * than next_tsn_at_change of the current primary, then 165 * the sender MUST NOT increment missing report count 166 * for t. 167 */ 168 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn) 169 { 170 if (primary->cacc.cycling_changeover && 171 TSN_lt(tsn, primary->cacc.next_tsn_at_change)) 172 return 1; 173 return 0; 174 } 175 176 /* 177 * SFR-CACC algorithm: 178 * 3) If the missing report count for TSN t is to be 179 * incremented according to [RFC2960] and 180 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set, 181 * then the sender MUST further execute steps 3.1 and 182 * 3.2 to determine if the missing report count for 183 * TSN t SHOULD NOT be incremented. 184 * 185 * 3.3) If 3.1 and 3.2 do not dictate that the missing 186 * report count for t should not be incremented, then 187 * the sender SHOULD increment missing report count for 188 * t (according to [RFC2960] and [SCTP_STEWART_2002]). 189 */ 190 static inline int sctp_cacc_skip(struct sctp_transport *primary, 191 struct sctp_transport *transport, 192 int count_of_newacks, 193 __u32 tsn) 194 { 195 if (primary->cacc.changeover_active && 196 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) || 197 sctp_cacc_skip_3_2(primary, tsn))) 198 return 1; 199 return 0; 200 } 201 202 /* Initialize an existing sctp_outq. This does the boring stuff. 203 * You still need to define handlers if you really want to DO 204 * something with this structure... 205 */ 206 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q) 207 { 208 q->asoc = asoc; 209 INIT_LIST_HEAD(&q->out_chunk_list); 210 INIT_LIST_HEAD(&q->control_chunk_list); 211 INIT_LIST_HEAD(&q->retransmit); 212 INIT_LIST_HEAD(&q->sacked); 213 INIT_LIST_HEAD(&q->abandoned); 214 215 q->fast_rtx = 0; 216 q->outstanding_bytes = 0; 217 q->empty = 1; 218 q->cork = 0; 219 220 q->malloced = 0; 221 q->out_qlen = 0; 222 } 223 224 /* Free the outqueue structure and any related pending chunks. 225 */ 226 void sctp_outq_teardown(struct sctp_outq *q) 227 { 228 struct sctp_transport *transport; 229 struct list_head *lchunk, *temp; 230 struct sctp_chunk *chunk, *tmp; 231 232 /* Throw away unacknowledged chunks. */ 233 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list, 234 transports) { 235 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) { 236 chunk = list_entry(lchunk, struct sctp_chunk, 237 transmitted_list); 238 /* Mark as part of a failed message. */ 239 sctp_chunk_fail(chunk, q->error); 240 sctp_chunk_free(chunk); 241 } 242 } 243 244 /* Throw away chunks that have been gap ACKed. */ 245 list_for_each_safe(lchunk, temp, &q->sacked) { 246 list_del_init(lchunk); 247 chunk = list_entry(lchunk, struct sctp_chunk, 248 transmitted_list); 249 sctp_chunk_fail(chunk, q->error); 250 sctp_chunk_free(chunk); 251 } 252 253 /* Throw away any chunks in the retransmit queue. */ 254 list_for_each_safe(lchunk, temp, &q->retransmit) { 255 list_del_init(lchunk); 256 chunk = list_entry(lchunk, struct sctp_chunk, 257 transmitted_list); 258 sctp_chunk_fail(chunk, q->error); 259 sctp_chunk_free(chunk); 260 } 261 262 /* Throw away any chunks that are in the abandoned queue. */ 263 list_for_each_safe(lchunk, temp, &q->abandoned) { 264 list_del_init(lchunk); 265 chunk = list_entry(lchunk, struct sctp_chunk, 266 transmitted_list); 267 sctp_chunk_fail(chunk, q->error); 268 sctp_chunk_free(chunk); 269 } 270 271 /* Throw away any leftover data chunks. */ 272 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 273 274 /* Mark as send failure. */ 275 sctp_chunk_fail(chunk, q->error); 276 sctp_chunk_free(chunk); 277 } 278 279 q->error = 0; 280 281 /* Throw away any leftover control chunks. */ 282 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 283 list_del_init(&chunk->list); 284 sctp_chunk_free(chunk); 285 } 286 } 287 288 /* Free the outqueue structure and any related pending chunks. */ 289 void sctp_outq_free(struct sctp_outq *q) 290 { 291 /* Throw away leftover chunks. */ 292 sctp_outq_teardown(q); 293 294 /* If we were kmalloc()'d, free the memory. */ 295 if (q->malloced) 296 kfree(q); 297 } 298 299 /* Put a new chunk in an sctp_outq. */ 300 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk) 301 { 302 int error = 0; 303 304 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n", 305 q, chunk, chunk && chunk->chunk_hdr ? 306 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) 307 : "Illegal Chunk"); 308 309 /* If it is data, queue it up, otherwise, send it 310 * immediately. 311 */ 312 if (sctp_chunk_is_data(chunk)) { 313 /* Is it OK to queue data chunks? */ 314 /* From 9. Termination of Association 315 * 316 * When either endpoint performs a shutdown, the 317 * association on each peer will stop accepting new 318 * data from its user and only deliver data in queue 319 * at the time of sending or receiving the SHUTDOWN 320 * chunk. 321 */ 322 switch (q->asoc->state) { 323 case SCTP_STATE_CLOSED: 324 case SCTP_STATE_SHUTDOWN_PENDING: 325 case SCTP_STATE_SHUTDOWN_SENT: 326 case SCTP_STATE_SHUTDOWN_RECEIVED: 327 case SCTP_STATE_SHUTDOWN_ACK_SENT: 328 /* Cannot send after transport endpoint shutdown */ 329 error = -ESHUTDOWN; 330 break; 331 332 default: 333 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n", 334 q, chunk, chunk && chunk->chunk_hdr ? 335 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) 336 : "Illegal Chunk"); 337 338 sctp_outq_tail_data(q, chunk); 339 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED) 340 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS); 341 else 342 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS); 343 q->empty = 0; 344 break; 345 } 346 } else { 347 list_add_tail(&chunk->list, &q->control_chunk_list); 348 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS); 349 } 350 351 if (error < 0) 352 return error; 353 354 if (!q->cork) 355 error = sctp_outq_flush(q, 0); 356 357 return error; 358 } 359 360 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list 361 * and the abandoned list are in ascending order. 362 */ 363 static void sctp_insert_list(struct list_head *head, struct list_head *new) 364 { 365 struct list_head *pos; 366 struct sctp_chunk *nchunk, *lchunk; 367 __u32 ntsn, ltsn; 368 int done = 0; 369 370 nchunk = list_entry(new, struct sctp_chunk, transmitted_list); 371 ntsn = ntohl(nchunk->subh.data_hdr->tsn); 372 373 list_for_each(pos, head) { 374 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list); 375 ltsn = ntohl(lchunk->subh.data_hdr->tsn); 376 if (TSN_lt(ntsn, ltsn)) { 377 list_add(new, pos->prev); 378 done = 1; 379 break; 380 } 381 } 382 if (!done) 383 list_add_tail(new, head); 384 } 385 386 /* Mark all the eligible packets on a transport for retransmission. */ 387 void sctp_retransmit_mark(struct sctp_outq *q, 388 struct sctp_transport *transport, 389 __u8 reason) 390 { 391 struct list_head *lchunk, *ltemp; 392 struct sctp_chunk *chunk; 393 394 /* Walk through the specified transmitted queue. */ 395 list_for_each_safe(lchunk, ltemp, &transport->transmitted) { 396 chunk = list_entry(lchunk, struct sctp_chunk, 397 transmitted_list); 398 399 /* If the chunk is abandoned, move it to abandoned list. */ 400 if (sctp_chunk_abandoned(chunk)) { 401 list_del_init(lchunk); 402 sctp_insert_list(&q->abandoned, lchunk); 403 404 /* If this chunk has not been previousely acked, 405 * stop considering it 'outstanding'. Our peer 406 * will most likely never see it since it will 407 * not be retransmitted 408 */ 409 if (!chunk->tsn_gap_acked) { 410 if (chunk->transport) 411 chunk->transport->flight_size -= 412 sctp_data_size(chunk); 413 q->outstanding_bytes -= sctp_data_size(chunk); 414 q->asoc->peer.rwnd += (sctp_data_size(chunk) + 415 sizeof(struct sk_buff)); 416 } 417 continue; 418 } 419 420 /* If we are doing retransmission due to a timeout or pmtu 421 * discovery, only the chunks that are not yet acked should 422 * be added to the retransmit queue. 423 */ 424 if ((reason == SCTP_RTXR_FAST_RTX && 425 (chunk->fast_retransmit == SCTP_NEED_FRTX)) || 426 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) { 427 /* RFC 2960 6.2.1 Processing a Received SACK 428 * 429 * C) Any time a DATA chunk is marked for 430 * retransmission (via either T3-rtx timer expiration 431 * (Section 6.3.3) or via fast retransmit 432 * (Section 7.2.4)), add the data size of those 433 * chunks to the rwnd. 434 */ 435 q->asoc->peer.rwnd += (sctp_data_size(chunk) + 436 sizeof(struct sk_buff)); 437 q->outstanding_bytes -= sctp_data_size(chunk); 438 if (chunk->transport) 439 transport->flight_size -= sctp_data_size(chunk); 440 441 /* sctpimpguide-05 Section 2.8.2 442 * M5) If a T3-rtx timer expires, the 443 * 'TSN.Missing.Report' of all affected TSNs is set 444 * to 0. 445 */ 446 chunk->tsn_missing_report = 0; 447 448 /* If a chunk that is being used for RTT measurement 449 * has to be retransmitted, we cannot use this chunk 450 * anymore for RTT measurements. Reset rto_pending so 451 * that a new RTT measurement is started when a new 452 * data chunk is sent. 453 */ 454 if (chunk->rtt_in_progress) { 455 chunk->rtt_in_progress = 0; 456 transport->rto_pending = 0; 457 } 458 459 /* Move the chunk to the retransmit queue. The chunks 460 * on the retransmit queue are always kept in order. 461 */ 462 list_del_init(lchunk); 463 sctp_insert_list(&q->retransmit, lchunk); 464 } 465 } 466 467 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, " 468 "cwnd: %d, ssthresh: %d, flight_size: %d, " 469 "pba: %d\n", __func__, 470 transport, reason, 471 transport->cwnd, transport->ssthresh, 472 transport->flight_size, 473 transport->partial_bytes_acked); 474 475 } 476 477 /* Mark all the eligible packets on a transport for retransmission and force 478 * one packet out. 479 */ 480 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport, 481 sctp_retransmit_reason_t reason) 482 { 483 int error = 0; 484 485 switch(reason) { 486 case SCTP_RTXR_T3_RTX: 487 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS); 488 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX); 489 /* Update the retran path if the T3-rtx timer has expired for 490 * the current retran path. 491 */ 492 if (transport == transport->asoc->peer.retran_path) 493 sctp_assoc_update_retran_path(transport->asoc); 494 transport->asoc->rtx_data_chunks += 495 transport->asoc->unack_data; 496 break; 497 case SCTP_RTXR_FAST_RTX: 498 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS); 499 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX); 500 q->fast_rtx = 1; 501 break; 502 case SCTP_RTXR_PMTUD: 503 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS); 504 break; 505 case SCTP_RTXR_T1_RTX: 506 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS); 507 transport->asoc->init_retries++; 508 break; 509 default: 510 BUG(); 511 } 512 513 sctp_retransmit_mark(q, transport, reason); 514 515 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination, 516 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by 517 * following the procedures outlined in C1 - C5. 518 */ 519 if (reason == SCTP_RTXR_T3_RTX) 520 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point); 521 522 /* Flush the queues only on timeout, since fast_rtx is only 523 * triggered during sack processing and the queue 524 * will be flushed at the end. 525 */ 526 if (reason != SCTP_RTXR_FAST_RTX) 527 error = sctp_outq_flush(q, /* rtx_timeout */ 1); 528 529 if (error) 530 q->asoc->base.sk->sk_err = -error; 531 } 532 533 /* 534 * Transmit DATA chunks on the retransmit queue. Upon return from 535 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which 536 * need to be transmitted by the caller. 537 * We assume that pkt->transport has already been set. 538 * 539 * The return value is a normal kernel error return value. 540 */ 541 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt, 542 int rtx_timeout, int *start_timer) 543 { 544 struct list_head *lqueue; 545 struct sctp_transport *transport = pkt->transport; 546 sctp_xmit_t status; 547 struct sctp_chunk *chunk, *chunk1; 548 int fast_rtx; 549 int error = 0; 550 int timer = 0; 551 int done = 0; 552 553 lqueue = &q->retransmit; 554 fast_rtx = q->fast_rtx; 555 556 /* This loop handles time-out retransmissions, fast retransmissions, 557 * and retransmissions due to opening of whindow. 558 * 559 * RFC 2960 6.3.3 Handle T3-rtx Expiration 560 * 561 * E3) Determine how many of the earliest (i.e., lowest TSN) 562 * outstanding DATA chunks for the address for which the 563 * T3-rtx has expired will fit into a single packet, subject 564 * to the MTU constraint for the path corresponding to the 565 * destination transport address to which the retransmission 566 * is being sent (this may be different from the address for 567 * which the timer expires [see Section 6.4]). Call this value 568 * K. Bundle and retransmit those K DATA chunks in a single 569 * packet to the destination endpoint. 570 * 571 * [Just to be painfully clear, if we are retransmitting 572 * because a timeout just happened, we should send only ONE 573 * packet of retransmitted data.] 574 * 575 * For fast retransmissions we also send only ONE packet. However, 576 * if we are just flushing the queue due to open window, we'll 577 * try to send as much as possible. 578 */ 579 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) { 580 /* If the chunk is abandoned, move it to abandoned list. */ 581 if (sctp_chunk_abandoned(chunk)) { 582 list_del_init(&chunk->transmitted_list); 583 sctp_insert_list(&q->abandoned, 584 &chunk->transmitted_list); 585 continue; 586 } 587 588 /* Make sure that Gap Acked TSNs are not retransmitted. A 589 * simple approach is just to move such TSNs out of the 590 * way and into a 'transmitted' queue and skip to the 591 * next chunk. 592 */ 593 if (chunk->tsn_gap_acked) { 594 list_del(&chunk->transmitted_list); 595 list_add_tail(&chunk->transmitted_list, 596 &transport->transmitted); 597 continue; 598 } 599 600 /* If we are doing fast retransmit, ignore non-fast_rtransmit 601 * chunks 602 */ 603 if (fast_rtx && !chunk->fast_retransmit) 604 continue; 605 606 redo: 607 /* Attempt to append this chunk to the packet. */ 608 status = sctp_packet_append_chunk(pkt, chunk); 609 610 switch (status) { 611 case SCTP_XMIT_PMTU_FULL: 612 if (!pkt->has_data && !pkt->has_cookie_echo) { 613 /* If this packet did not contain DATA then 614 * retransmission did not happen, so do it 615 * again. We'll ignore the error here since 616 * control chunks are already freed so there 617 * is nothing we can do. 618 */ 619 sctp_packet_transmit(pkt); 620 goto redo; 621 } 622 623 /* Send this packet. */ 624 error = sctp_packet_transmit(pkt); 625 626 /* If we are retransmitting, we should only 627 * send a single packet. 628 * Otherwise, try appending this chunk again. 629 */ 630 if (rtx_timeout || fast_rtx) 631 done = 1; 632 else 633 goto redo; 634 635 /* Bundle next chunk in the next round. */ 636 break; 637 638 case SCTP_XMIT_RWND_FULL: 639 /* Send this packet. */ 640 error = sctp_packet_transmit(pkt); 641 642 /* Stop sending DATA as there is no more room 643 * at the receiver. 644 */ 645 done = 1; 646 break; 647 648 case SCTP_XMIT_NAGLE_DELAY: 649 /* Send this packet. */ 650 error = sctp_packet_transmit(pkt); 651 652 /* Stop sending DATA because of nagle delay. */ 653 done = 1; 654 break; 655 656 default: 657 /* The append was successful, so add this chunk to 658 * the transmitted list. 659 */ 660 list_del(&chunk->transmitted_list); 661 list_add_tail(&chunk->transmitted_list, 662 &transport->transmitted); 663 664 /* Mark the chunk as ineligible for fast retransmit 665 * after it is retransmitted. 666 */ 667 if (chunk->fast_retransmit == SCTP_NEED_FRTX) 668 chunk->fast_retransmit = SCTP_DONT_FRTX; 669 670 q->empty = 0; 671 break; 672 } 673 674 /* Set the timer if there were no errors */ 675 if (!error && !timer) 676 timer = 1; 677 678 if (done) 679 break; 680 } 681 682 /* If we are here due to a retransmit timeout or a fast 683 * retransmit and if there are any chunks left in the retransmit 684 * queue that could not fit in the PMTU sized packet, they need 685 * to be marked as ineligible for a subsequent fast retransmit. 686 */ 687 if (rtx_timeout || fast_rtx) { 688 list_for_each_entry(chunk1, lqueue, transmitted_list) { 689 if (chunk1->fast_retransmit == SCTP_NEED_FRTX) 690 chunk1->fast_retransmit = SCTP_DONT_FRTX; 691 } 692 } 693 694 *start_timer = timer; 695 696 /* Clear fast retransmit hint */ 697 if (fast_rtx) 698 q->fast_rtx = 0; 699 700 return error; 701 } 702 703 /* Cork the outqueue so queued chunks are really queued. */ 704 int sctp_outq_uncork(struct sctp_outq *q) 705 { 706 int error = 0; 707 if (q->cork) 708 q->cork = 0; 709 error = sctp_outq_flush(q, 0); 710 return error; 711 } 712 713 714 /* 715 * Try to flush an outqueue. 716 * 717 * Description: Send everything in q which we legally can, subject to 718 * congestion limitations. 719 * * Note: This function can be called from multiple contexts so appropriate 720 * locking concerns must be made. Today we use the sock lock to protect 721 * this function. 722 */ 723 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout) 724 { 725 struct sctp_packet *packet; 726 struct sctp_packet singleton; 727 struct sctp_association *asoc = q->asoc; 728 __u16 sport = asoc->base.bind_addr.port; 729 __u16 dport = asoc->peer.port; 730 __u32 vtag = asoc->peer.i.init_tag; 731 struct sctp_transport *transport = NULL; 732 struct sctp_transport *new_transport; 733 struct sctp_chunk *chunk, *tmp; 734 sctp_xmit_t status; 735 int error = 0; 736 int start_timer = 0; 737 int one_packet = 0; 738 739 /* These transports have chunks to send. */ 740 struct list_head transport_list; 741 struct list_head *ltransport; 742 743 INIT_LIST_HEAD(&transport_list); 744 packet = NULL; 745 746 /* 747 * 6.10 Bundling 748 * ... 749 * When bundling control chunks with DATA chunks, an 750 * endpoint MUST place control chunks first in the outbound 751 * SCTP packet. The transmitter MUST transmit DATA chunks 752 * within a SCTP packet in increasing order of TSN. 753 * ... 754 */ 755 756 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) { 757 list_del_init(&chunk->list); 758 759 /* Pick the right transport to use. */ 760 new_transport = chunk->transport; 761 762 if (!new_transport) { 763 /* 764 * If we have a prior transport pointer, see if 765 * the destination address of the chunk 766 * matches the destination address of the 767 * current transport. If not a match, then 768 * try to look up the transport with a given 769 * destination address. We do this because 770 * after processing ASCONFs, we may have new 771 * transports created. 772 */ 773 if (transport && 774 sctp_cmp_addr_exact(&chunk->dest, 775 &transport->ipaddr)) 776 new_transport = transport; 777 else 778 new_transport = sctp_assoc_lookup_paddr(asoc, 779 &chunk->dest); 780 781 /* if we still don't have a new transport, then 782 * use the current active path. 783 */ 784 if (!new_transport) 785 new_transport = asoc->peer.active_path; 786 } else if ((new_transport->state == SCTP_INACTIVE) || 787 (new_transport->state == SCTP_UNCONFIRMED)) { 788 /* If the chunk is Heartbeat or Heartbeat Ack, 789 * send it to chunk->transport, even if it's 790 * inactive. 791 * 792 * 3.3.6 Heartbeat Acknowledgement: 793 * ... 794 * A HEARTBEAT ACK is always sent to the source IP 795 * address of the IP datagram containing the 796 * HEARTBEAT chunk to which this ack is responding. 797 * ... 798 * 799 * ASCONF_ACKs also must be sent to the source. 800 */ 801 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT && 802 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK && 803 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK) 804 new_transport = asoc->peer.active_path; 805 } 806 807 /* Are we switching transports? 808 * Take care of transport locks. 809 */ 810 if (new_transport != transport) { 811 transport = new_transport; 812 if (list_empty(&transport->send_ready)) { 813 list_add_tail(&transport->send_ready, 814 &transport_list); 815 } 816 packet = &transport->packet; 817 sctp_packet_config(packet, vtag, 818 asoc->peer.ecn_capable); 819 } 820 821 switch (chunk->chunk_hdr->type) { 822 /* 823 * 6.10 Bundling 824 * ... 825 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN 826 * COMPLETE with any other chunks. [Send them immediately.] 827 */ 828 case SCTP_CID_INIT: 829 case SCTP_CID_INIT_ACK: 830 case SCTP_CID_SHUTDOWN_COMPLETE: 831 sctp_packet_init(&singleton, transport, sport, dport); 832 sctp_packet_config(&singleton, vtag, 0); 833 sctp_packet_append_chunk(&singleton, chunk); 834 error = sctp_packet_transmit(&singleton); 835 if (error < 0) 836 return error; 837 break; 838 839 case SCTP_CID_ABORT: 840 if (sctp_test_T_bit(chunk)) { 841 packet->vtag = asoc->c.my_vtag; 842 } 843 /* The following chunks are "response" chunks, i.e. 844 * they are generated in response to something we 845 * received. If we are sending these, then we can 846 * send only 1 packet containing these chunks. 847 */ 848 case SCTP_CID_HEARTBEAT_ACK: 849 case SCTP_CID_SHUTDOWN_ACK: 850 case SCTP_CID_COOKIE_ACK: 851 case SCTP_CID_COOKIE_ECHO: 852 case SCTP_CID_ERROR: 853 case SCTP_CID_ECN_CWR: 854 case SCTP_CID_ASCONF_ACK: 855 one_packet = 1; 856 /* Fall through */ 857 858 case SCTP_CID_SACK: 859 case SCTP_CID_HEARTBEAT: 860 case SCTP_CID_SHUTDOWN: 861 case SCTP_CID_ECN_ECNE: 862 case SCTP_CID_ASCONF: 863 case SCTP_CID_FWD_TSN: 864 status = sctp_packet_transmit_chunk(packet, chunk, 865 one_packet); 866 if (status != SCTP_XMIT_OK) { 867 /* put the chunk back */ 868 list_add(&chunk->list, &q->control_chunk_list); 869 } else if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) { 870 /* PR-SCTP C5) If a FORWARD TSN is sent, the 871 * sender MUST assure that at least one T3-rtx 872 * timer is running. 873 */ 874 sctp_transport_reset_timers(transport); 875 } 876 break; 877 878 default: 879 /* We built a chunk with an illegal type! */ 880 BUG(); 881 } 882 } 883 884 /* Is it OK to send data chunks? */ 885 switch (asoc->state) { 886 case SCTP_STATE_COOKIE_ECHOED: 887 /* Only allow bundling when this packet has a COOKIE-ECHO 888 * chunk. 889 */ 890 if (!packet || !packet->has_cookie_echo) 891 break; 892 893 /* fallthru */ 894 case SCTP_STATE_ESTABLISHED: 895 case SCTP_STATE_SHUTDOWN_PENDING: 896 case SCTP_STATE_SHUTDOWN_RECEIVED: 897 /* 898 * RFC 2960 6.1 Transmission of DATA Chunks 899 * 900 * C) When the time comes for the sender to transmit, 901 * before sending new DATA chunks, the sender MUST 902 * first transmit any outstanding DATA chunks which 903 * are marked for retransmission (limited by the 904 * current cwnd). 905 */ 906 if (!list_empty(&q->retransmit)) { 907 if (transport == asoc->peer.retran_path) 908 goto retran; 909 910 /* Switch transports & prepare the packet. */ 911 912 transport = asoc->peer.retran_path; 913 914 if (list_empty(&transport->send_ready)) { 915 list_add_tail(&transport->send_ready, 916 &transport_list); 917 } 918 919 packet = &transport->packet; 920 sctp_packet_config(packet, vtag, 921 asoc->peer.ecn_capable); 922 retran: 923 error = sctp_outq_flush_rtx(q, packet, 924 rtx_timeout, &start_timer); 925 926 if (start_timer) 927 sctp_transport_reset_timers(transport); 928 929 /* This can happen on COOKIE-ECHO resend. Only 930 * one chunk can get bundled with a COOKIE-ECHO. 931 */ 932 if (packet->has_cookie_echo) 933 goto sctp_flush_out; 934 935 /* Don't send new data if there is still data 936 * waiting to retransmit. 937 */ 938 if (!list_empty(&q->retransmit)) 939 goto sctp_flush_out; 940 } 941 942 /* Apply Max.Burst limitation to the current transport in 943 * case it will be used for new data. We are going to 944 * rest it before we return, but we want to apply the limit 945 * to the currently queued data. 946 */ 947 if (transport) 948 sctp_transport_burst_limited(transport); 949 950 /* Finally, transmit new packets. */ 951 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) { 952 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid 953 * stream identifier. 954 */ 955 if (chunk->sinfo.sinfo_stream >= 956 asoc->c.sinit_num_ostreams) { 957 958 /* Mark as failed send. */ 959 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM); 960 sctp_chunk_free(chunk); 961 continue; 962 } 963 964 /* Has this chunk expired? */ 965 if (sctp_chunk_abandoned(chunk)) { 966 sctp_chunk_fail(chunk, 0); 967 sctp_chunk_free(chunk); 968 continue; 969 } 970 971 /* If there is a specified transport, use it. 972 * Otherwise, we want to use the active path. 973 */ 974 new_transport = chunk->transport; 975 if (!new_transport || 976 ((new_transport->state == SCTP_INACTIVE) || 977 (new_transport->state == SCTP_UNCONFIRMED))) 978 new_transport = asoc->peer.active_path; 979 980 /* Change packets if necessary. */ 981 if (new_transport != transport) { 982 transport = new_transport; 983 984 /* Schedule to have this transport's 985 * packet flushed. 986 */ 987 if (list_empty(&transport->send_ready)) { 988 list_add_tail(&transport->send_ready, 989 &transport_list); 990 } 991 992 packet = &transport->packet; 993 sctp_packet_config(packet, vtag, 994 asoc->peer.ecn_capable); 995 /* We've switched transports, so apply the 996 * Burst limit to the new transport. 997 */ 998 sctp_transport_burst_limited(transport); 999 } 1000 1001 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ", 1002 q, chunk, 1003 chunk && chunk->chunk_hdr ? 1004 sctp_cname(SCTP_ST_CHUNK( 1005 chunk->chunk_hdr->type)) 1006 : "Illegal Chunk"); 1007 1008 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head " 1009 "%p skb->users %d.\n", 1010 ntohl(chunk->subh.data_hdr->tsn), 1011 chunk->skb ?chunk->skb->head : NULL, 1012 chunk->skb ? 1013 atomic_read(&chunk->skb->users) : -1); 1014 1015 /* Add the chunk to the packet. */ 1016 status = sctp_packet_transmit_chunk(packet, chunk, 0); 1017 1018 switch (status) { 1019 case SCTP_XMIT_PMTU_FULL: 1020 case SCTP_XMIT_RWND_FULL: 1021 case SCTP_XMIT_NAGLE_DELAY: 1022 /* We could not append this chunk, so put 1023 * the chunk back on the output queue. 1024 */ 1025 SCTP_DEBUG_PRINTK("sctp_outq_flush: could " 1026 "not transmit TSN: 0x%x, status: %d\n", 1027 ntohl(chunk->subh.data_hdr->tsn), 1028 status); 1029 sctp_outq_head_data(q, chunk); 1030 goto sctp_flush_out; 1031 break; 1032 1033 case SCTP_XMIT_OK: 1034 /* The sender is in the SHUTDOWN-PENDING state, 1035 * The sender MAY set the I-bit in the DATA 1036 * chunk header. 1037 */ 1038 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING) 1039 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM; 1040 1041 break; 1042 1043 default: 1044 BUG(); 1045 } 1046 1047 /* BUG: We assume that the sctp_packet_transmit() 1048 * call below will succeed all the time and add the 1049 * chunk to the transmitted list and restart the 1050 * timers. 1051 * It is possible that the call can fail under OOM 1052 * conditions. 1053 * 1054 * Is this really a problem? Won't this behave 1055 * like a lost TSN? 1056 */ 1057 list_add_tail(&chunk->transmitted_list, 1058 &transport->transmitted); 1059 1060 sctp_transport_reset_timers(transport); 1061 1062 q->empty = 0; 1063 1064 /* Only let one DATA chunk get bundled with a 1065 * COOKIE-ECHO chunk. 1066 */ 1067 if (packet->has_cookie_echo) 1068 goto sctp_flush_out; 1069 } 1070 break; 1071 1072 default: 1073 /* Do nothing. */ 1074 break; 1075 } 1076 1077 sctp_flush_out: 1078 1079 /* Before returning, examine all the transports touched in 1080 * this call. Right now, we bluntly force clear all the 1081 * transports. Things might change after we implement Nagle. 1082 * But such an examination is still required. 1083 * 1084 * --xguo 1085 */ 1086 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) { 1087 struct sctp_transport *t = list_entry(ltransport, 1088 struct sctp_transport, 1089 send_ready); 1090 packet = &t->packet; 1091 if (!sctp_packet_empty(packet)) 1092 error = sctp_packet_transmit(packet); 1093 1094 /* Clear the burst limited state, if any */ 1095 sctp_transport_burst_reset(t); 1096 } 1097 1098 return error; 1099 } 1100 1101 /* Update unack_data based on the incoming SACK chunk */ 1102 static void sctp_sack_update_unack_data(struct sctp_association *assoc, 1103 struct sctp_sackhdr *sack) 1104 { 1105 sctp_sack_variable_t *frags; 1106 __u16 unack_data; 1107 int i; 1108 1109 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1; 1110 1111 frags = sack->variable; 1112 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) { 1113 unack_data -= ((ntohs(frags[i].gab.end) - 1114 ntohs(frags[i].gab.start) + 1)); 1115 } 1116 1117 assoc->unack_data = unack_data; 1118 } 1119 1120 /* This is where we REALLY process a SACK. 1121 * 1122 * Process the SACK against the outqueue. Mostly, this just frees 1123 * things off the transmitted queue. 1124 */ 1125 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack) 1126 { 1127 struct sctp_association *asoc = q->asoc; 1128 struct sctp_transport *transport; 1129 struct sctp_chunk *tchunk = NULL; 1130 struct list_head *lchunk, *transport_list, *temp; 1131 sctp_sack_variable_t *frags = sack->variable; 1132 __u32 sack_ctsn, ctsn, tsn; 1133 __u32 highest_tsn, highest_new_tsn; 1134 __u32 sack_a_rwnd; 1135 unsigned outstanding; 1136 struct sctp_transport *primary = asoc->peer.primary_path; 1137 int count_of_newacks = 0; 1138 int gap_ack_blocks; 1139 u8 accum_moved = 0; 1140 1141 /* Grab the association's destination address list. */ 1142 transport_list = &asoc->peer.transport_addr_list; 1143 1144 sack_ctsn = ntohl(sack->cum_tsn_ack); 1145 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks); 1146 /* 1147 * SFR-CACC algorithm: 1148 * On receipt of a SACK the sender SHOULD execute the 1149 * following statements. 1150 * 1151 * 1) If the cumulative ack in the SACK passes next tsn_at_change 1152 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be 1153 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for 1154 * all destinations. 1155 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE 1156 * is set the receiver of the SACK MUST take the following actions: 1157 * 1158 * A) Initialize the cacc_saw_newack to 0 for all destination 1159 * addresses. 1160 * 1161 * Only bother if changeover_active is set. Otherwise, this is 1162 * totally suboptimal to do on every SACK. 1163 */ 1164 if (primary->cacc.changeover_active) { 1165 u8 clear_cycling = 0; 1166 1167 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) { 1168 primary->cacc.changeover_active = 0; 1169 clear_cycling = 1; 1170 } 1171 1172 if (clear_cycling || gap_ack_blocks) { 1173 list_for_each_entry(transport, transport_list, 1174 transports) { 1175 if (clear_cycling) 1176 transport->cacc.cycling_changeover = 0; 1177 if (gap_ack_blocks) 1178 transport->cacc.cacc_saw_newack = 0; 1179 } 1180 } 1181 } 1182 1183 /* Get the highest TSN in the sack. */ 1184 highest_tsn = sack_ctsn; 1185 if (gap_ack_blocks) 1186 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end); 1187 1188 if (TSN_lt(asoc->highest_sacked, highest_tsn)) 1189 asoc->highest_sacked = highest_tsn; 1190 1191 highest_new_tsn = sack_ctsn; 1192 1193 /* Run through the retransmit queue. Credit bytes received 1194 * and free those chunks that we can. 1195 */ 1196 sctp_check_transmitted(q, &q->retransmit, NULL, sack, &highest_new_tsn); 1197 1198 /* Run through the transmitted queue. 1199 * Credit bytes received and free those chunks which we can. 1200 * 1201 * This is a MASSIVE candidate for optimization. 1202 */ 1203 list_for_each_entry(transport, transport_list, transports) { 1204 sctp_check_transmitted(q, &transport->transmitted, 1205 transport, sack, &highest_new_tsn); 1206 /* 1207 * SFR-CACC algorithm: 1208 * C) Let count_of_newacks be the number of 1209 * destinations for which cacc_saw_newack is set. 1210 */ 1211 if (transport->cacc.cacc_saw_newack) 1212 count_of_newacks ++; 1213 } 1214 1215 /* Move the Cumulative TSN Ack Point if appropriate. */ 1216 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) { 1217 asoc->ctsn_ack_point = sack_ctsn; 1218 accum_moved = 1; 1219 } 1220 1221 if (gap_ack_blocks) { 1222 1223 if (asoc->fast_recovery && accum_moved) 1224 highest_new_tsn = highest_tsn; 1225 1226 list_for_each_entry(transport, transport_list, transports) 1227 sctp_mark_missing(q, &transport->transmitted, transport, 1228 highest_new_tsn, count_of_newacks); 1229 } 1230 1231 /* Update unack_data field in the assoc. */ 1232 sctp_sack_update_unack_data(asoc, sack); 1233 1234 ctsn = asoc->ctsn_ack_point; 1235 1236 /* Throw away stuff rotting on the sack queue. */ 1237 list_for_each_safe(lchunk, temp, &q->sacked) { 1238 tchunk = list_entry(lchunk, struct sctp_chunk, 1239 transmitted_list); 1240 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1241 if (TSN_lte(tsn, ctsn)) { 1242 list_del_init(&tchunk->transmitted_list); 1243 sctp_chunk_free(tchunk); 1244 } 1245 } 1246 1247 /* ii) Set rwnd equal to the newly received a_rwnd minus the 1248 * number of bytes still outstanding after processing the 1249 * Cumulative TSN Ack and the Gap Ack Blocks. 1250 */ 1251 1252 sack_a_rwnd = ntohl(sack->a_rwnd); 1253 outstanding = q->outstanding_bytes; 1254 1255 if (outstanding < sack_a_rwnd) 1256 sack_a_rwnd -= outstanding; 1257 else 1258 sack_a_rwnd = 0; 1259 1260 asoc->peer.rwnd = sack_a_rwnd; 1261 1262 sctp_generate_fwdtsn(q, sack_ctsn); 1263 1264 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n", 1265 __func__, sack_ctsn); 1266 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, " 1267 "%p is 0x%x. Adv peer ack point: 0x%x\n", 1268 __func__, asoc, ctsn, asoc->adv_peer_ack_point); 1269 1270 /* See if all chunks are acked. 1271 * Make sure the empty queue handler will get run later. 1272 */ 1273 q->empty = (list_empty(&q->out_chunk_list) && 1274 list_empty(&q->retransmit)); 1275 if (!q->empty) 1276 goto finish; 1277 1278 list_for_each_entry(transport, transport_list, transports) { 1279 q->empty = q->empty && list_empty(&transport->transmitted); 1280 if (!q->empty) 1281 goto finish; 1282 } 1283 1284 SCTP_DEBUG_PRINTK("sack queue is empty.\n"); 1285 finish: 1286 return q->empty; 1287 } 1288 1289 /* Is the outqueue empty? */ 1290 int sctp_outq_is_empty(const struct sctp_outq *q) 1291 { 1292 return q->empty; 1293 } 1294 1295 /******************************************************************** 1296 * 2nd Level Abstractions 1297 ********************************************************************/ 1298 1299 /* Go through a transport's transmitted list or the association's retransmit 1300 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked. 1301 * The retransmit list will not have an associated transport. 1302 * 1303 * I added coherent debug information output. --xguo 1304 * 1305 * Instead of printing 'sacked' or 'kept' for each TSN on the 1306 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5. 1307 * KEPT TSN6-TSN7, etc. 1308 */ 1309 static void sctp_check_transmitted(struct sctp_outq *q, 1310 struct list_head *transmitted_queue, 1311 struct sctp_transport *transport, 1312 struct sctp_sackhdr *sack, 1313 __u32 *highest_new_tsn_in_sack) 1314 { 1315 struct list_head *lchunk; 1316 struct sctp_chunk *tchunk; 1317 struct list_head tlist; 1318 __u32 tsn; 1319 __u32 sack_ctsn; 1320 __u32 rtt; 1321 __u8 restart_timer = 0; 1322 int bytes_acked = 0; 1323 int migrate_bytes = 0; 1324 1325 /* These state variables are for coherent debug output. --xguo */ 1326 1327 #if SCTP_DEBUG 1328 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */ 1329 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */ 1330 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */ 1331 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */ 1332 1333 /* 0 : The last TSN was ACKed. 1334 * 1 : The last TSN was NOT ACKed (i.e. KEPT). 1335 * -1: We need to initialize. 1336 */ 1337 int dbg_prt_state = -1; 1338 #endif /* SCTP_DEBUG */ 1339 1340 sack_ctsn = ntohl(sack->cum_tsn_ack); 1341 1342 INIT_LIST_HEAD(&tlist); 1343 1344 /* The while loop will skip empty transmitted queues. */ 1345 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) { 1346 tchunk = list_entry(lchunk, struct sctp_chunk, 1347 transmitted_list); 1348 1349 if (sctp_chunk_abandoned(tchunk)) { 1350 /* Move the chunk to abandoned list. */ 1351 sctp_insert_list(&q->abandoned, lchunk); 1352 1353 /* If this chunk has not been acked, stop 1354 * considering it as 'outstanding'. 1355 */ 1356 if (!tchunk->tsn_gap_acked) { 1357 if (tchunk->transport) 1358 tchunk->transport->flight_size -= 1359 sctp_data_size(tchunk); 1360 q->outstanding_bytes -= sctp_data_size(tchunk); 1361 } 1362 continue; 1363 } 1364 1365 tsn = ntohl(tchunk->subh.data_hdr->tsn); 1366 if (sctp_acked(sack, tsn)) { 1367 /* If this queue is the retransmit queue, the 1368 * retransmit timer has already reclaimed 1369 * the outstanding bytes for this chunk, so only 1370 * count bytes associated with a transport. 1371 */ 1372 if (transport) { 1373 /* If this chunk is being used for RTT 1374 * measurement, calculate the RTT and update 1375 * the RTO using this value. 1376 * 1377 * 6.3.1 C5) Karn's algorithm: RTT measurements 1378 * MUST NOT be made using packets that were 1379 * retransmitted (and thus for which it is 1380 * ambiguous whether the reply was for the 1381 * first instance of the packet or a later 1382 * instance). 1383 */ 1384 if (!tchunk->tsn_gap_acked && 1385 tchunk->rtt_in_progress) { 1386 tchunk->rtt_in_progress = 0; 1387 rtt = jiffies - tchunk->sent_at; 1388 sctp_transport_update_rto(transport, 1389 rtt); 1390 } 1391 } 1392 1393 /* If the chunk hasn't been marked as ACKED, 1394 * mark it and account bytes_acked if the 1395 * chunk had a valid transport (it will not 1396 * have a transport if ASCONF had deleted it 1397 * while DATA was outstanding). 1398 */ 1399 if (!tchunk->tsn_gap_acked) { 1400 tchunk->tsn_gap_acked = 1; 1401 *highest_new_tsn_in_sack = tsn; 1402 bytes_acked += sctp_data_size(tchunk); 1403 if (!tchunk->transport) 1404 migrate_bytes += sctp_data_size(tchunk); 1405 } 1406 1407 if (TSN_lte(tsn, sack_ctsn)) { 1408 /* RFC 2960 6.3.2 Retransmission Timer Rules 1409 * 1410 * R3) Whenever a SACK is received 1411 * that acknowledges the DATA chunk 1412 * with the earliest outstanding TSN 1413 * for that address, restart T3-rtx 1414 * timer for that address with its 1415 * current RTO. 1416 */ 1417 restart_timer = 1; 1418 1419 if (!tchunk->tsn_gap_acked) { 1420 /* 1421 * SFR-CACC algorithm: 1422 * 2) If the SACK contains gap acks 1423 * and the flag CHANGEOVER_ACTIVE is 1424 * set the receiver of the SACK MUST 1425 * take the following action: 1426 * 1427 * B) For each TSN t being acked that 1428 * has not been acked in any SACK so 1429 * far, set cacc_saw_newack to 1 for 1430 * the destination that the TSN was 1431 * sent to. 1432 */ 1433 if (transport && 1434 sack->num_gap_ack_blocks && 1435 q->asoc->peer.primary_path->cacc. 1436 changeover_active) 1437 transport->cacc.cacc_saw_newack 1438 = 1; 1439 } 1440 1441 list_add_tail(&tchunk->transmitted_list, 1442 &q->sacked); 1443 } else { 1444 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2 1445 * M2) Each time a SACK arrives reporting 1446 * 'Stray DATA chunk(s)' record the highest TSN 1447 * reported as newly acknowledged, call this 1448 * value 'HighestTSNinSack'. A newly 1449 * acknowledged DATA chunk is one not 1450 * previously acknowledged in a SACK. 1451 * 1452 * When the SCTP sender of data receives a SACK 1453 * chunk that acknowledges, for the first time, 1454 * the receipt of a DATA chunk, all the still 1455 * unacknowledged DATA chunks whose TSN is 1456 * older than that newly acknowledged DATA 1457 * chunk, are qualified as 'Stray DATA chunks'. 1458 */ 1459 list_add_tail(lchunk, &tlist); 1460 } 1461 1462 #if SCTP_DEBUG 1463 switch (dbg_prt_state) { 1464 case 0: /* last TSN was ACKed */ 1465 if (dbg_last_ack_tsn + 1 == tsn) { 1466 /* This TSN belongs to the 1467 * current ACK range. 1468 */ 1469 break; 1470 } 1471 1472 if (dbg_last_ack_tsn != dbg_ack_tsn) { 1473 /* Display the end of the 1474 * current range. 1475 */ 1476 SCTP_DEBUG_PRINTK_CONT("-%08x", 1477 dbg_last_ack_tsn); 1478 } 1479 1480 /* Start a new range. */ 1481 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn); 1482 dbg_ack_tsn = tsn; 1483 break; 1484 1485 case 1: /* The last TSN was NOT ACKed. */ 1486 if (dbg_last_kept_tsn != dbg_kept_tsn) { 1487 /* Display the end of current range. */ 1488 SCTP_DEBUG_PRINTK_CONT("-%08x", 1489 dbg_last_kept_tsn); 1490 } 1491 1492 SCTP_DEBUG_PRINTK_CONT("\n"); 1493 1494 /* FALL THROUGH... */ 1495 default: 1496 /* This is the first-ever TSN we examined. */ 1497 /* Start a new range of ACK-ed TSNs. */ 1498 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn); 1499 dbg_prt_state = 0; 1500 dbg_ack_tsn = tsn; 1501 } 1502 1503 dbg_last_ack_tsn = tsn; 1504 #endif /* SCTP_DEBUG */ 1505 1506 } else { 1507 if (tchunk->tsn_gap_acked) { 1508 SCTP_DEBUG_PRINTK("%s: Receiver reneged on " 1509 "data TSN: 0x%x\n", 1510 __func__, 1511 tsn); 1512 tchunk->tsn_gap_acked = 0; 1513 1514 if (tchunk->transport) 1515 bytes_acked -= sctp_data_size(tchunk); 1516 1517 /* RFC 2960 6.3.2 Retransmission Timer Rules 1518 * 1519 * R4) Whenever a SACK is received missing a 1520 * TSN that was previously acknowledged via a 1521 * Gap Ack Block, start T3-rtx for the 1522 * destination address to which the DATA 1523 * chunk was originally 1524 * transmitted if it is not already running. 1525 */ 1526 restart_timer = 1; 1527 } 1528 1529 list_add_tail(lchunk, &tlist); 1530 1531 #if SCTP_DEBUG 1532 /* See the above comments on ACK-ed TSNs. */ 1533 switch (dbg_prt_state) { 1534 case 1: 1535 if (dbg_last_kept_tsn + 1 == tsn) 1536 break; 1537 1538 if (dbg_last_kept_tsn != dbg_kept_tsn) 1539 SCTP_DEBUG_PRINTK_CONT("-%08x", 1540 dbg_last_kept_tsn); 1541 1542 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn); 1543 dbg_kept_tsn = tsn; 1544 break; 1545 1546 case 0: 1547 if (dbg_last_ack_tsn != dbg_ack_tsn) 1548 SCTP_DEBUG_PRINTK_CONT("-%08x", 1549 dbg_last_ack_tsn); 1550 SCTP_DEBUG_PRINTK_CONT("\n"); 1551 1552 /* FALL THROUGH... */ 1553 default: 1554 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn); 1555 dbg_prt_state = 1; 1556 dbg_kept_tsn = tsn; 1557 } 1558 1559 dbg_last_kept_tsn = tsn; 1560 #endif /* SCTP_DEBUG */ 1561 } 1562 } 1563 1564 #if SCTP_DEBUG 1565 /* Finish off the last range, displaying its ending TSN. */ 1566 switch (dbg_prt_state) { 1567 case 0: 1568 if (dbg_last_ack_tsn != dbg_ack_tsn) { 1569 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn); 1570 } else { 1571 SCTP_DEBUG_PRINTK_CONT("\n"); 1572 } 1573 break; 1574 1575 case 1: 1576 if (dbg_last_kept_tsn != dbg_kept_tsn) { 1577 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn); 1578 } else { 1579 SCTP_DEBUG_PRINTK_CONT("\n"); 1580 } 1581 } 1582 #endif /* SCTP_DEBUG */ 1583 if (transport) { 1584 if (bytes_acked) { 1585 struct sctp_association *asoc = transport->asoc; 1586 1587 /* We may have counted DATA that was migrated 1588 * to this transport due to DEL-IP operation. 1589 * Subtract those bytes, since the were never 1590 * send on this transport and shouldn't be 1591 * credited to this transport. 1592 */ 1593 bytes_acked -= migrate_bytes; 1594 1595 /* 8.2. When an outstanding TSN is acknowledged, 1596 * the endpoint shall clear the error counter of 1597 * the destination transport address to which the 1598 * DATA chunk was last sent. 1599 * The association's overall error counter is 1600 * also cleared. 1601 */ 1602 transport->error_count = 0; 1603 transport->asoc->overall_error_count = 0; 1604 1605 /* 1606 * While in SHUTDOWN PENDING, we may have started 1607 * the T5 shutdown guard timer after reaching the 1608 * retransmission limit. Stop that timer as soon 1609 * as the receiver acknowledged any data. 1610 */ 1611 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING && 1612 del_timer(&asoc->timers 1613 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD])) 1614 sctp_association_put(asoc); 1615 1616 /* Mark the destination transport address as 1617 * active if it is not so marked. 1618 */ 1619 if ((transport->state == SCTP_INACTIVE) || 1620 (transport->state == SCTP_UNCONFIRMED)) { 1621 sctp_assoc_control_transport( 1622 transport->asoc, 1623 transport, 1624 SCTP_TRANSPORT_UP, 1625 SCTP_RECEIVED_SACK); 1626 } 1627 1628 sctp_transport_raise_cwnd(transport, sack_ctsn, 1629 bytes_acked); 1630 1631 transport->flight_size -= bytes_acked; 1632 if (transport->flight_size == 0) 1633 transport->partial_bytes_acked = 0; 1634 q->outstanding_bytes -= bytes_acked + migrate_bytes; 1635 } else { 1636 /* RFC 2960 6.1, sctpimpguide-06 2.15.2 1637 * When a sender is doing zero window probing, it 1638 * should not timeout the association if it continues 1639 * to receive new packets from the receiver. The 1640 * reason is that the receiver MAY keep its window 1641 * closed for an indefinite time. 1642 * A sender is doing zero window probing when the 1643 * receiver's advertised window is zero, and there is 1644 * only one data chunk in flight to the receiver. 1645 * 1646 * Allow the association to timeout while in SHUTDOWN 1647 * PENDING or SHUTDOWN RECEIVED in case the receiver 1648 * stays in zero window mode forever. 1649 */ 1650 if (!q->asoc->peer.rwnd && 1651 !list_empty(&tlist) && 1652 (sack_ctsn+2 == q->asoc->next_tsn) && 1653 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) { 1654 SCTP_DEBUG_PRINTK("%s: SACK received for zero " 1655 "window probe: %u\n", 1656 __func__, sack_ctsn); 1657 q->asoc->overall_error_count = 0; 1658 transport->error_count = 0; 1659 } 1660 } 1661 1662 /* RFC 2960 6.3.2 Retransmission Timer Rules 1663 * 1664 * R2) Whenever all outstanding data sent to an address have 1665 * been acknowledged, turn off the T3-rtx timer of that 1666 * address. 1667 */ 1668 if (!transport->flight_size) { 1669 if (timer_pending(&transport->T3_rtx_timer) && 1670 del_timer(&transport->T3_rtx_timer)) { 1671 sctp_transport_put(transport); 1672 } 1673 } else if (restart_timer) { 1674 if (!mod_timer(&transport->T3_rtx_timer, 1675 jiffies + transport->rto)) 1676 sctp_transport_hold(transport); 1677 } 1678 } 1679 1680 list_splice(&tlist, transmitted_queue); 1681 } 1682 1683 /* Mark chunks as missing and consequently may get retransmitted. */ 1684 static void sctp_mark_missing(struct sctp_outq *q, 1685 struct list_head *transmitted_queue, 1686 struct sctp_transport *transport, 1687 __u32 highest_new_tsn_in_sack, 1688 int count_of_newacks) 1689 { 1690 struct sctp_chunk *chunk; 1691 __u32 tsn; 1692 char do_fast_retransmit = 0; 1693 struct sctp_association *asoc = q->asoc; 1694 struct sctp_transport *primary = asoc->peer.primary_path; 1695 1696 list_for_each_entry(chunk, transmitted_queue, transmitted_list) { 1697 1698 tsn = ntohl(chunk->subh.data_hdr->tsn); 1699 1700 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all 1701 * 'Unacknowledged TSN's', if the TSN number of an 1702 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack' 1703 * value, increment the 'TSN.Missing.Report' count on that 1704 * chunk if it has NOT been fast retransmitted or marked for 1705 * fast retransmit already. 1706 */ 1707 if (chunk->fast_retransmit == SCTP_CAN_FRTX && 1708 !chunk->tsn_gap_acked && 1709 TSN_lt(tsn, highest_new_tsn_in_sack)) { 1710 1711 /* SFR-CACC may require us to skip marking 1712 * this chunk as missing. 1713 */ 1714 if (!transport || !sctp_cacc_skip(primary, 1715 chunk->transport, 1716 count_of_newacks, tsn)) { 1717 chunk->tsn_missing_report++; 1718 1719 SCTP_DEBUG_PRINTK( 1720 "%s: TSN 0x%x missing counter: %d\n", 1721 __func__, tsn, 1722 chunk->tsn_missing_report); 1723 } 1724 } 1725 /* 1726 * M4) If any DATA chunk is found to have a 1727 * 'TSN.Missing.Report' 1728 * value larger than or equal to 3, mark that chunk for 1729 * retransmission and start the fast retransmit procedure. 1730 */ 1731 1732 if (chunk->tsn_missing_report >= 3) { 1733 chunk->fast_retransmit = SCTP_NEED_FRTX; 1734 do_fast_retransmit = 1; 1735 } 1736 } 1737 1738 if (transport) { 1739 if (do_fast_retransmit) 1740 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX); 1741 1742 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, " 1743 "ssthresh: %d, flight_size: %d, pba: %d\n", 1744 __func__, transport, transport->cwnd, 1745 transport->ssthresh, transport->flight_size, 1746 transport->partial_bytes_acked); 1747 } 1748 } 1749 1750 /* Is the given TSN acked by this packet? */ 1751 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn) 1752 { 1753 int i; 1754 sctp_sack_variable_t *frags; 1755 __u16 gap; 1756 __u32 ctsn = ntohl(sack->cum_tsn_ack); 1757 1758 if (TSN_lte(tsn, ctsn)) 1759 goto pass; 1760 1761 /* 3.3.4 Selective Acknowledgement (SACK) (3): 1762 * 1763 * Gap Ack Blocks: 1764 * These fields contain the Gap Ack Blocks. They are repeated 1765 * for each Gap Ack Block up to the number of Gap Ack Blocks 1766 * defined in the Number of Gap Ack Blocks field. All DATA 1767 * chunks with TSNs greater than or equal to (Cumulative TSN 1768 * Ack + Gap Ack Block Start) and less than or equal to 1769 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack 1770 * Block are assumed to have been received correctly. 1771 */ 1772 1773 frags = sack->variable; 1774 gap = tsn - ctsn; 1775 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) { 1776 if (TSN_lte(ntohs(frags[i].gab.start), gap) && 1777 TSN_lte(gap, ntohs(frags[i].gab.end))) 1778 goto pass; 1779 } 1780 1781 return 0; 1782 pass: 1783 return 1; 1784 } 1785 1786 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist, 1787 int nskips, __be16 stream) 1788 { 1789 int i; 1790 1791 for (i = 0; i < nskips; i++) { 1792 if (skiplist[i].stream == stream) 1793 return i; 1794 } 1795 return i; 1796 } 1797 1798 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */ 1799 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn) 1800 { 1801 struct sctp_association *asoc = q->asoc; 1802 struct sctp_chunk *ftsn_chunk = NULL; 1803 struct sctp_fwdtsn_skip ftsn_skip_arr[10]; 1804 int nskips = 0; 1805 int skip_pos = 0; 1806 __u32 tsn; 1807 struct sctp_chunk *chunk; 1808 struct list_head *lchunk, *temp; 1809 1810 if (!asoc->peer.prsctp_capable) 1811 return; 1812 1813 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the 1814 * received SACK. 1815 * 1816 * If (Advanced.Peer.Ack.Point < SackCumAck), then update 1817 * Advanced.Peer.Ack.Point to be equal to SackCumAck. 1818 */ 1819 if (TSN_lt(asoc->adv_peer_ack_point, ctsn)) 1820 asoc->adv_peer_ack_point = ctsn; 1821 1822 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point" 1823 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as 1824 * the chunk next in the out-queue space is marked as "abandoned" as 1825 * shown in the following example: 1826 * 1827 * Assuming that a SACK arrived with the Cumulative TSN ACK 102 1828 * and the Advanced.Peer.Ack.Point is updated to this value: 1829 * 1830 * out-queue at the end of ==> out-queue after Adv.Ack.Point 1831 * normal SACK processing local advancement 1832 * ... ... 1833 * Adv.Ack.Pt-> 102 acked 102 acked 1834 * 103 abandoned 103 abandoned 1835 * 104 abandoned Adv.Ack.P-> 104 abandoned 1836 * 105 105 1837 * 106 acked 106 acked 1838 * ... ... 1839 * 1840 * In this example, the data sender successfully advanced the 1841 * "Advanced.Peer.Ack.Point" from 102 to 104 locally. 1842 */ 1843 list_for_each_safe(lchunk, temp, &q->abandoned) { 1844 chunk = list_entry(lchunk, struct sctp_chunk, 1845 transmitted_list); 1846 tsn = ntohl(chunk->subh.data_hdr->tsn); 1847 1848 /* Remove any chunks in the abandoned queue that are acked by 1849 * the ctsn. 1850 */ 1851 if (TSN_lte(tsn, ctsn)) { 1852 list_del_init(lchunk); 1853 sctp_chunk_free(chunk); 1854 } else { 1855 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) { 1856 asoc->adv_peer_ack_point = tsn; 1857 if (chunk->chunk_hdr->flags & 1858 SCTP_DATA_UNORDERED) 1859 continue; 1860 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0], 1861 nskips, 1862 chunk->subh.data_hdr->stream); 1863 ftsn_skip_arr[skip_pos].stream = 1864 chunk->subh.data_hdr->stream; 1865 ftsn_skip_arr[skip_pos].ssn = 1866 chunk->subh.data_hdr->ssn; 1867 if (skip_pos == nskips) 1868 nskips++; 1869 if (nskips == 10) 1870 break; 1871 } else 1872 break; 1873 } 1874 } 1875 1876 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point" 1877 * is greater than the Cumulative TSN ACK carried in the received 1878 * SACK, the data sender MUST send the data receiver a FORWARD TSN 1879 * chunk containing the latest value of the 1880 * "Advanced.Peer.Ack.Point". 1881 * 1882 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD 1883 * list each stream and sequence number in the forwarded TSN. This 1884 * information will enable the receiver to easily find any 1885 * stranded TSN's waiting on stream reorder queues. Each stream 1886 * SHOULD only be reported once; this means that if multiple 1887 * abandoned messages occur in the same stream then only the 1888 * highest abandoned stream sequence number is reported. If the 1889 * total size of the FORWARD TSN does NOT fit in a single MTU then 1890 * the sender of the FORWARD TSN SHOULD lower the 1891 * Advanced.Peer.Ack.Point to the last TSN that will fit in a 1892 * single MTU. 1893 */ 1894 if (asoc->adv_peer_ack_point > ctsn) 1895 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point, 1896 nskips, &ftsn_skip_arr[0]); 1897 1898 if (ftsn_chunk) { 1899 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list); 1900 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS); 1901 } 1902 } 1903