1 /* Management of Tx window, Tx resend, ACKs and out-of-sequence reception 2 * 3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 13 14 #include <linux/module.h> 15 #include <linux/circ_buf.h> 16 #include <linux/net.h> 17 #include <linux/skbuff.h> 18 #include <linux/slab.h> 19 #include <linux/udp.h> 20 #include <net/sock.h> 21 #include <net/af_rxrpc.h> 22 #include "ar-internal.h" 23 24 /* 25 * propose an ACK be sent 26 */ 27 void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason, 28 u32 serial, bool immediate) 29 { 30 unsigned long expiry; 31 s8 prior = rxrpc_ack_priority[ack_reason]; 32 33 ASSERTCMP(prior, >, 0); 34 35 _enter("{%d},%s,%%%x,%u", 36 call->debug_id, rxrpc_acks(ack_reason), serial, immediate); 37 38 if (prior < rxrpc_ack_priority[call->ackr_reason]) { 39 if (immediate) 40 goto cancel_timer; 41 return; 42 } 43 44 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial 45 * numbers */ 46 if (prior == rxrpc_ack_priority[call->ackr_reason]) { 47 if (prior <= 4) 48 call->ackr_serial = serial; 49 if (immediate) 50 goto cancel_timer; 51 return; 52 } 53 54 call->ackr_reason = ack_reason; 55 call->ackr_serial = serial; 56 57 switch (ack_reason) { 58 case RXRPC_ACK_DELAY: 59 _debug("run delay timer"); 60 expiry = rxrpc_soft_ack_delay; 61 goto run_timer; 62 63 case RXRPC_ACK_IDLE: 64 if (!immediate) { 65 _debug("run defer timer"); 66 expiry = rxrpc_idle_ack_delay; 67 goto run_timer; 68 } 69 goto cancel_timer; 70 71 case RXRPC_ACK_REQUESTED: 72 expiry = rxrpc_requested_ack_delay; 73 if (!expiry) 74 goto cancel_timer; 75 if (!immediate || serial == 1) { 76 _debug("run defer timer"); 77 goto run_timer; 78 } 79 80 default: 81 _debug("immediate ACK"); 82 goto cancel_timer; 83 } 84 85 run_timer: 86 expiry += jiffies; 87 if (!timer_pending(&call->ack_timer) || 88 time_after(call->ack_timer.expires, expiry)) 89 mod_timer(&call->ack_timer, expiry); 90 return; 91 92 cancel_timer: 93 _debug("cancel timer %%%u", serial); 94 try_to_del_timer_sync(&call->ack_timer); 95 read_lock_bh(&call->state_lock); 96 if (call->state <= RXRPC_CALL_COMPLETE && 97 !test_and_set_bit(RXRPC_CALL_EV_ACK, &call->events)) 98 rxrpc_queue_call(call); 99 read_unlock_bh(&call->state_lock); 100 } 101 102 /* 103 * propose an ACK be sent, locking the call structure 104 */ 105 void rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason, 106 u32 serial, bool immediate) 107 { 108 s8 prior = rxrpc_ack_priority[ack_reason]; 109 110 if (prior > rxrpc_ack_priority[call->ackr_reason]) { 111 spin_lock_bh(&call->lock); 112 __rxrpc_propose_ACK(call, ack_reason, serial, immediate); 113 spin_unlock_bh(&call->lock); 114 } 115 } 116 117 /* 118 * set the resend timer 119 */ 120 static void rxrpc_set_resend(struct rxrpc_call *call, u8 resend, 121 unsigned long resend_at) 122 { 123 read_lock_bh(&call->state_lock); 124 if (call->state >= RXRPC_CALL_COMPLETE) 125 resend = 0; 126 127 if (resend & 1) { 128 _debug("SET RESEND"); 129 set_bit(RXRPC_CALL_EV_RESEND, &call->events); 130 } 131 132 if (resend & 2) { 133 _debug("MODIFY RESEND TIMER"); 134 set_bit(RXRPC_CALL_RUN_RTIMER, &call->flags); 135 mod_timer(&call->resend_timer, resend_at); 136 } else { 137 _debug("KILL RESEND TIMER"); 138 del_timer_sync(&call->resend_timer); 139 clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events); 140 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags); 141 } 142 read_unlock_bh(&call->state_lock); 143 } 144 145 /* 146 * resend packets 147 */ 148 static void rxrpc_resend(struct rxrpc_call *call) 149 { 150 struct rxrpc_wire_header *whdr; 151 struct rxrpc_skb_priv *sp; 152 struct sk_buff *txb; 153 unsigned long *p_txb, resend_at; 154 bool stop; 155 int loop; 156 u8 resend; 157 158 _enter("{%d,%d,%d,%d},", 159 call->acks_hard, call->acks_unacked, 160 atomic_read(&call->sequence), 161 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz)); 162 163 stop = false; 164 resend = 0; 165 resend_at = 0; 166 167 for (loop = call->acks_tail; 168 loop != call->acks_head || stop; 169 loop = (loop + 1) & (call->acks_winsz - 1) 170 ) { 171 p_txb = call->acks_window + loop; 172 smp_read_barrier_depends(); 173 if (*p_txb & 1) 174 continue; 175 176 txb = (struct sk_buff *) *p_txb; 177 sp = rxrpc_skb(txb); 178 179 if (sp->need_resend) { 180 sp->need_resend = false; 181 182 /* each Tx packet has a new serial number */ 183 sp->hdr.serial = atomic_inc_return(&call->conn->serial); 184 185 whdr = (struct rxrpc_wire_header *)txb->head; 186 whdr->serial = htonl(sp->hdr.serial); 187 188 _proto("Tx DATA %%%u { #%d }", 189 sp->hdr.serial, sp->hdr.seq); 190 if (rxrpc_send_data_packet(call->conn, txb) < 0) { 191 stop = true; 192 sp->resend_at = jiffies + 3; 193 } else { 194 sp->resend_at = 195 jiffies + rxrpc_resend_timeout; 196 } 197 } 198 199 if (time_after_eq(jiffies + 1, sp->resend_at)) { 200 sp->need_resend = true; 201 resend |= 1; 202 } else if (resend & 2) { 203 if (time_before(sp->resend_at, resend_at)) 204 resend_at = sp->resend_at; 205 } else { 206 resend_at = sp->resend_at; 207 resend |= 2; 208 } 209 } 210 211 rxrpc_set_resend(call, resend, resend_at); 212 _leave(""); 213 } 214 215 /* 216 * handle resend timer expiry 217 */ 218 static void rxrpc_resend_timer(struct rxrpc_call *call) 219 { 220 struct rxrpc_skb_priv *sp; 221 struct sk_buff *txb; 222 unsigned long *p_txb, resend_at; 223 int loop; 224 u8 resend; 225 226 _enter("%d,%d,%d", 227 call->acks_tail, call->acks_unacked, call->acks_head); 228 229 if (call->state >= RXRPC_CALL_COMPLETE) 230 return; 231 232 resend = 0; 233 resend_at = 0; 234 235 for (loop = call->acks_unacked; 236 loop != call->acks_head; 237 loop = (loop + 1) & (call->acks_winsz - 1) 238 ) { 239 p_txb = call->acks_window + loop; 240 smp_read_barrier_depends(); 241 txb = (struct sk_buff *) (*p_txb & ~1); 242 sp = rxrpc_skb(txb); 243 244 ASSERT(!(*p_txb & 1)); 245 246 if (sp->need_resend) { 247 ; 248 } else if (time_after_eq(jiffies + 1, sp->resend_at)) { 249 sp->need_resend = true; 250 resend |= 1; 251 } else if (resend & 2) { 252 if (time_before(sp->resend_at, resend_at)) 253 resend_at = sp->resend_at; 254 } else { 255 resend_at = sp->resend_at; 256 resend |= 2; 257 } 258 } 259 260 rxrpc_set_resend(call, resend, resend_at); 261 _leave(""); 262 } 263 264 /* 265 * process soft ACKs of our transmitted packets 266 * - these indicate packets the peer has or has not received, but hasn't yet 267 * given to the consumer, and so can still be discarded and re-requested 268 */ 269 static int rxrpc_process_soft_ACKs(struct rxrpc_call *call, 270 struct rxrpc_ackpacket *ack, 271 struct sk_buff *skb) 272 { 273 struct rxrpc_skb_priv *sp; 274 struct sk_buff *txb; 275 unsigned long *p_txb, resend_at; 276 int loop; 277 u8 sacks[RXRPC_MAXACKS], resend; 278 279 _enter("{%d,%d},{%d},", 280 call->acks_hard, 281 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz), 282 ack->nAcks); 283 284 if (skb_copy_bits(skb, 0, sacks, ack->nAcks) < 0) 285 goto protocol_error; 286 287 resend = 0; 288 resend_at = 0; 289 for (loop = 0; loop < ack->nAcks; loop++) { 290 p_txb = call->acks_window; 291 p_txb += (call->acks_tail + loop) & (call->acks_winsz - 1); 292 smp_read_barrier_depends(); 293 txb = (struct sk_buff *) (*p_txb & ~1); 294 sp = rxrpc_skb(txb); 295 296 switch (sacks[loop]) { 297 case RXRPC_ACK_TYPE_ACK: 298 sp->need_resend = false; 299 *p_txb |= 1; 300 break; 301 case RXRPC_ACK_TYPE_NACK: 302 sp->need_resend = true; 303 *p_txb &= ~1; 304 resend = 1; 305 break; 306 default: 307 _debug("Unsupported ACK type %d", sacks[loop]); 308 goto protocol_error; 309 } 310 } 311 312 smp_mb(); 313 call->acks_unacked = (call->acks_tail + loop) & (call->acks_winsz - 1); 314 315 /* anything not explicitly ACK'd is implicitly NACK'd, but may just not 316 * have been received or processed yet by the far end */ 317 for (loop = call->acks_unacked; 318 loop != call->acks_head; 319 loop = (loop + 1) & (call->acks_winsz - 1) 320 ) { 321 p_txb = call->acks_window + loop; 322 smp_read_barrier_depends(); 323 txb = (struct sk_buff *) (*p_txb & ~1); 324 sp = rxrpc_skb(txb); 325 326 if (*p_txb & 1) { 327 /* packet must have been discarded */ 328 sp->need_resend = true; 329 *p_txb &= ~1; 330 resend |= 1; 331 } else if (sp->need_resend) { 332 ; 333 } else if (time_after_eq(jiffies + 1, sp->resend_at)) { 334 sp->need_resend = true; 335 resend |= 1; 336 } else if (resend & 2) { 337 if (time_before(sp->resend_at, resend_at)) 338 resend_at = sp->resend_at; 339 } else { 340 resend_at = sp->resend_at; 341 resend |= 2; 342 } 343 } 344 345 rxrpc_set_resend(call, resend, resend_at); 346 _leave(" = 0"); 347 return 0; 348 349 protocol_error: 350 _leave(" = -EPROTO"); 351 return -EPROTO; 352 } 353 354 /* 355 * discard hard-ACK'd packets from the Tx window 356 */ 357 static void rxrpc_rotate_tx_window(struct rxrpc_call *call, u32 hard) 358 { 359 unsigned long _skb; 360 int tail = call->acks_tail, old_tail; 361 int win = CIRC_CNT(call->acks_head, tail, call->acks_winsz); 362 363 _enter("{%u,%u},%u", call->acks_hard, win, hard); 364 365 ASSERTCMP(hard - call->acks_hard, <=, win); 366 367 while (call->acks_hard < hard) { 368 smp_read_barrier_depends(); 369 _skb = call->acks_window[tail] & ~1; 370 rxrpc_free_skb((struct sk_buff *) _skb); 371 old_tail = tail; 372 tail = (tail + 1) & (call->acks_winsz - 1); 373 call->acks_tail = tail; 374 if (call->acks_unacked == old_tail) 375 call->acks_unacked = tail; 376 call->acks_hard++; 377 } 378 379 wake_up(&call->tx_waitq); 380 } 381 382 /* 383 * clear the Tx window in the event of a failure 384 */ 385 static void rxrpc_clear_tx_window(struct rxrpc_call *call) 386 { 387 rxrpc_rotate_tx_window(call, atomic_read(&call->sequence)); 388 } 389 390 /* 391 * drain the out of sequence received packet queue into the packet Rx queue 392 */ 393 static int rxrpc_drain_rx_oos_queue(struct rxrpc_call *call) 394 { 395 struct rxrpc_skb_priv *sp; 396 struct sk_buff *skb; 397 bool terminal; 398 int ret; 399 400 _enter("{%d,%d}", call->rx_data_post, call->rx_first_oos); 401 402 spin_lock_bh(&call->lock); 403 404 ret = -ECONNRESET; 405 if (test_bit(RXRPC_CALL_RELEASED, &call->flags)) 406 goto socket_unavailable; 407 408 skb = skb_dequeue(&call->rx_oos_queue); 409 if (skb) { 410 sp = rxrpc_skb(skb); 411 412 _debug("drain OOS packet %d [%d]", 413 sp->hdr.seq, call->rx_first_oos); 414 415 if (sp->hdr.seq != call->rx_first_oos) { 416 skb_queue_head(&call->rx_oos_queue, skb); 417 call->rx_first_oos = rxrpc_skb(skb)->hdr.seq; 418 _debug("requeue %p {%u}", skb, call->rx_first_oos); 419 } else { 420 skb->mark = RXRPC_SKB_MARK_DATA; 421 terminal = ((sp->hdr.flags & RXRPC_LAST_PACKET) && 422 !(sp->hdr.flags & RXRPC_CLIENT_INITIATED)); 423 ret = rxrpc_queue_rcv_skb(call, skb, true, terminal); 424 BUG_ON(ret < 0); 425 _debug("drain #%u", call->rx_data_post); 426 call->rx_data_post++; 427 428 /* find out what the next packet is */ 429 skb = skb_peek(&call->rx_oos_queue); 430 if (skb) 431 call->rx_first_oos = rxrpc_skb(skb)->hdr.seq; 432 else 433 call->rx_first_oos = 0; 434 _debug("peek %p {%u}", skb, call->rx_first_oos); 435 } 436 } 437 438 ret = 0; 439 socket_unavailable: 440 spin_unlock_bh(&call->lock); 441 _leave(" = %d", ret); 442 return ret; 443 } 444 445 /* 446 * insert an out of sequence packet into the buffer 447 */ 448 static void rxrpc_insert_oos_packet(struct rxrpc_call *call, 449 struct sk_buff *skb) 450 { 451 struct rxrpc_skb_priv *sp, *psp; 452 struct sk_buff *p; 453 u32 seq; 454 455 sp = rxrpc_skb(skb); 456 seq = sp->hdr.seq; 457 _enter(",,{%u}", seq); 458 459 skb->destructor = rxrpc_packet_destructor; 460 ASSERTCMP(sp->call, ==, NULL); 461 sp->call = call; 462 rxrpc_get_call(call); 463 464 /* insert into the buffer in sequence order */ 465 spin_lock_bh(&call->lock); 466 467 skb_queue_walk(&call->rx_oos_queue, p) { 468 psp = rxrpc_skb(p); 469 if (psp->hdr.seq > seq) { 470 _debug("insert oos #%u before #%u", seq, psp->hdr.seq); 471 skb_insert(p, skb, &call->rx_oos_queue); 472 goto inserted; 473 } 474 } 475 476 _debug("append oos #%u", seq); 477 skb_queue_tail(&call->rx_oos_queue, skb); 478 inserted: 479 480 /* we might now have a new front to the queue */ 481 if (call->rx_first_oos == 0 || seq < call->rx_first_oos) 482 call->rx_first_oos = seq; 483 484 read_lock(&call->state_lock); 485 if (call->state < RXRPC_CALL_COMPLETE && 486 call->rx_data_post == call->rx_first_oos) { 487 _debug("drain rx oos now"); 488 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events); 489 } 490 read_unlock(&call->state_lock); 491 492 spin_unlock_bh(&call->lock); 493 _leave(" [stored #%u]", call->rx_first_oos); 494 } 495 496 /* 497 * clear the Tx window on final ACK reception 498 */ 499 static void rxrpc_zap_tx_window(struct rxrpc_call *call) 500 { 501 struct rxrpc_skb_priv *sp; 502 struct sk_buff *skb; 503 unsigned long _skb, *acks_window; 504 u8 winsz = call->acks_winsz; 505 int tail; 506 507 acks_window = call->acks_window; 508 call->acks_window = NULL; 509 510 while (CIRC_CNT(call->acks_head, call->acks_tail, winsz) > 0) { 511 tail = call->acks_tail; 512 smp_read_barrier_depends(); 513 _skb = acks_window[tail] & ~1; 514 smp_mb(); 515 call->acks_tail = (call->acks_tail + 1) & (winsz - 1); 516 517 skb = (struct sk_buff *) _skb; 518 sp = rxrpc_skb(skb); 519 _debug("+++ clear Tx %u", sp->hdr.seq); 520 rxrpc_free_skb(skb); 521 } 522 523 kfree(acks_window); 524 } 525 526 /* 527 * process the extra information that may be appended to an ACK packet 528 */ 529 static void rxrpc_extract_ackinfo(struct rxrpc_call *call, struct sk_buff *skb, 530 unsigned int latest, int nAcks) 531 { 532 struct rxrpc_ackinfo ackinfo; 533 struct rxrpc_peer *peer; 534 unsigned int mtu; 535 536 if (skb_copy_bits(skb, nAcks + 3, &ackinfo, sizeof(ackinfo)) < 0) { 537 _leave(" [no ackinfo]"); 538 return; 539 } 540 541 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }", 542 latest, 543 ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU), 544 ntohl(ackinfo.rwind), ntohl(ackinfo.jumbo_max)); 545 546 mtu = min(ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU)); 547 548 peer = call->conn->params.peer; 549 if (mtu < peer->maxdata) { 550 spin_lock_bh(&peer->lock); 551 peer->maxdata = mtu; 552 peer->mtu = mtu + peer->hdrsize; 553 spin_unlock_bh(&peer->lock); 554 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata); 555 } 556 } 557 558 /* 559 * process packets in the reception queue 560 */ 561 static int rxrpc_process_rx_queue(struct rxrpc_call *call, 562 u32 *_abort_code) 563 { 564 struct rxrpc_ackpacket ack; 565 struct rxrpc_skb_priv *sp; 566 struct sk_buff *skb; 567 bool post_ACK; 568 int latest; 569 u32 hard, tx; 570 571 _enter(""); 572 573 process_further: 574 skb = skb_dequeue(&call->rx_queue); 575 if (!skb) 576 return -EAGAIN; 577 578 _net("deferred skb %p", skb); 579 580 sp = rxrpc_skb(skb); 581 582 _debug("process %s [st %d]", rxrpc_pkts[sp->hdr.type], call->state); 583 584 post_ACK = false; 585 586 switch (sp->hdr.type) { 587 /* data packets that wind up here have been received out of 588 * order, need security processing or are jumbo packets */ 589 case RXRPC_PACKET_TYPE_DATA: 590 _proto("OOSQ DATA %%%u { #%u }", sp->hdr.serial, sp->hdr.seq); 591 592 /* secured packets must be verified and possibly decrypted */ 593 if (call->conn->security->verify_packet(call, skb, 594 _abort_code) < 0) 595 goto protocol_error; 596 597 rxrpc_insert_oos_packet(call, skb); 598 goto process_further; 599 600 /* partial ACK to process */ 601 case RXRPC_PACKET_TYPE_ACK: 602 if (skb_copy_bits(skb, 0, &ack, sizeof(ack)) < 0) { 603 _debug("extraction failure"); 604 goto protocol_error; 605 } 606 if (!skb_pull(skb, sizeof(ack))) 607 BUG(); 608 609 latest = sp->hdr.serial; 610 hard = ntohl(ack.firstPacket); 611 tx = atomic_read(&call->sequence); 612 613 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }", 614 latest, 615 ntohs(ack.maxSkew), 616 hard, 617 ntohl(ack.previousPacket), 618 ntohl(ack.serial), 619 rxrpc_acks(ack.reason), 620 ack.nAcks); 621 622 rxrpc_extract_ackinfo(call, skb, latest, ack.nAcks); 623 624 if (ack.reason == RXRPC_ACK_PING) { 625 _proto("Rx ACK %%%u PING Request", latest); 626 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE, 627 sp->hdr.serial, true); 628 } 629 630 /* discard any out-of-order or duplicate ACKs */ 631 if (latest - call->acks_latest <= 0) { 632 _debug("discard ACK %d <= %d", 633 latest, call->acks_latest); 634 goto discard; 635 } 636 call->acks_latest = latest; 637 638 if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST && 639 call->state != RXRPC_CALL_CLIENT_AWAIT_REPLY && 640 call->state != RXRPC_CALL_SERVER_SEND_REPLY && 641 call->state != RXRPC_CALL_SERVER_AWAIT_ACK) 642 goto discard; 643 644 _debug("Tx=%d H=%u S=%d", tx, call->acks_hard, call->state); 645 646 if (hard > 0) { 647 if (hard - 1 > tx) { 648 _debug("hard-ACK'd packet %d not transmitted" 649 " (%d top)", 650 hard - 1, tx); 651 goto protocol_error; 652 } 653 654 if ((call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY || 655 call->state == RXRPC_CALL_SERVER_AWAIT_ACK) && 656 hard > tx) { 657 call->acks_hard = tx; 658 goto all_acked; 659 } 660 661 smp_rmb(); 662 rxrpc_rotate_tx_window(call, hard - 1); 663 } 664 665 if (ack.nAcks > 0) { 666 if (hard - 1 + ack.nAcks > tx) { 667 _debug("soft-ACK'd packet %d+%d not" 668 " transmitted (%d top)", 669 hard - 1, ack.nAcks, tx); 670 goto protocol_error; 671 } 672 673 if (rxrpc_process_soft_ACKs(call, &ack, skb) < 0) 674 goto protocol_error; 675 } 676 goto discard; 677 678 /* complete ACK to process */ 679 case RXRPC_PACKET_TYPE_ACKALL: 680 goto all_acked; 681 682 /* abort and busy are handled elsewhere */ 683 case RXRPC_PACKET_TYPE_BUSY: 684 case RXRPC_PACKET_TYPE_ABORT: 685 BUG(); 686 687 /* connection level events - also handled elsewhere */ 688 case RXRPC_PACKET_TYPE_CHALLENGE: 689 case RXRPC_PACKET_TYPE_RESPONSE: 690 case RXRPC_PACKET_TYPE_DEBUG: 691 BUG(); 692 } 693 694 /* if we've had a hard ACK that covers all the packets we've sent, then 695 * that ends that phase of the operation */ 696 all_acked: 697 write_lock_bh(&call->state_lock); 698 _debug("ack all %d", call->state); 699 700 switch (call->state) { 701 case RXRPC_CALL_CLIENT_AWAIT_REPLY: 702 call->state = RXRPC_CALL_CLIENT_RECV_REPLY; 703 break; 704 case RXRPC_CALL_SERVER_AWAIT_ACK: 705 _debug("srv complete"); 706 call->state = RXRPC_CALL_COMPLETE; 707 post_ACK = true; 708 break; 709 case RXRPC_CALL_CLIENT_SEND_REQUEST: 710 case RXRPC_CALL_SERVER_RECV_REQUEST: 711 goto protocol_error_unlock; /* can't occur yet */ 712 default: 713 write_unlock_bh(&call->state_lock); 714 goto discard; /* assume packet left over from earlier phase */ 715 } 716 717 write_unlock_bh(&call->state_lock); 718 719 /* if all the packets we sent are hard-ACK'd, then we can discard 720 * whatever we've got left */ 721 _debug("clear Tx %d", 722 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz)); 723 724 del_timer_sync(&call->resend_timer); 725 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags); 726 clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events); 727 728 if (call->acks_window) 729 rxrpc_zap_tx_window(call); 730 731 if (post_ACK) { 732 /* post the final ACK message for userspace to pick up */ 733 _debug("post ACK"); 734 skb->mark = RXRPC_SKB_MARK_FINAL_ACK; 735 sp->call = call; 736 rxrpc_get_call(call); 737 spin_lock_bh(&call->lock); 738 if (rxrpc_queue_rcv_skb(call, skb, true, true) < 0) 739 BUG(); 740 spin_unlock_bh(&call->lock); 741 goto process_further; 742 } 743 744 discard: 745 rxrpc_free_skb(skb); 746 goto process_further; 747 748 protocol_error_unlock: 749 write_unlock_bh(&call->state_lock); 750 protocol_error: 751 rxrpc_free_skb(skb); 752 _leave(" = -EPROTO"); 753 return -EPROTO; 754 } 755 756 /* 757 * post a message to the socket Rx queue for recvmsg() to pick up 758 */ 759 static int rxrpc_post_message(struct rxrpc_call *call, u32 mark, u32 error, 760 bool fatal) 761 { 762 struct rxrpc_skb_priv *sp; 763 struct sk_buff *skb; 764 int ret; 765 766 _enter("{%d,%lx},%u,%u,%d", 767 call->debug_id, call->flags, mark, error, fatal); 768 769 /* remove timers and things for fatal messages */ 770 if (fatal) { 771 del_timer_sync(&call->resend_timer); 772 del_timer_sync(&call->ack_timer); 773 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags); 774 } 775 776 if (mark != RXRPC_SKB_MARK_NEW_CALL && 777 !test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) { 778 _leave("[no userid]"); 779 return 0; 780 } 781 782 if (!test_bit(RXRPC_CALL_TERMINAL_MSG, &call->flags)) { 783 skb = alloc_skb(0, GFP_NOFS); 784 if (!skb) 785 return -ENOMEM; 786 787 rxrpc_new_skb(skb); 788 789 skb->mark = mark; 790 791 sp = rxrpc_skb(skb); 792 memset(sp, 0, sizeof(*sp)); 793 sp->error = error; 794 sp->call = call; 795 rxrpc_get_call(call); 796 797 spin_lock_bh(&call->lock); 798 ret = rxrpc_queue_rcv_skb(call, skb, true, fatal); 799 spin_unlock_bh(&call->lock); 800 BUG_ON(ret < 0); 801 } 802 803 return 0; 804 } 805 806 /* 807 * handle background processing of incoming call packets and ACK / abort 808 * generation 809 */ 810 void rxrpc_process_call(struct work_struct *work) 811 { 812 struct rxrpc_call *call = 813 container_of(work, struct rxrpc_call, processor); 814 struct rxrpc_wire_header whdr; 815 struct rxrpc_ackpacket ack; 816 struct rxrpc_ackinfo ackinfo; 817 struct msghdr msg; 818 struct kvec iov[5]; 819 enum rxrpc_call_event genbit; 820 unsigned long bits; 821 __be32 data, pad; 822 size_t len; 823 int loop, nbit, ioc, ret, mtu; 824 u32 serial, abort_code = RX_PROTOCOL_ERROR; 825 u8 *acks = NULL; 826 827 //printk("\n--------------------\n"); 828 _enter("{%d,%s,%lx} [%lu]", 829 call->debug_id, rxrpc_call_states[call->state], call->events, 830 (jiffies - call->creation_jif) / (HZ / 10)); 831 832 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY, &call->flags)) { 833 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX"); 834 return; 835 } 836 837 /* there's a good chance we're going to have to send a message, so set 838 * one up in advance */ 839 msg.msg_name = &call->conn->params.peer->srx.transport; 840 msg.msg_namelen = call->conn->params.peer->srx.transport_len; 841 msg.msg_control = NULL; 842 msg.msg_controllen = 0; 843 msg.msg_flags = 0; 844 845 whdr.epoch = htonl(call->conn->proto.epoch); 846 whdr.cid = htonl(call->cid); 847 whdr.callNumber = htonl(call->call_id); 848 whdr.seq = 0; 849 whdr.type = RXRPC_PACKET_TYPE_ACK; 850 whdr.flags = call->conn->out_clientflag; 851 whdr.userStatus = 0; 852 whdr.securityIndex = call->conn->security_ix; 853 whdr._rsvd = 0; 854 whdr.serviceId = htons(call->service_id); 855 856 memset(iov, 0, sizeof(iov)); 857 iov[0].iov_base = &whdr; 858 iov[0].iov_len = sizeof(whdr); 859 860 /* deal with events of a final nature */ 861 if (test_bit(RXRPC_CALL_EV_RCVD_ERROR, &call->events)) { 862 enum rxrpc_skb_mark mark; 863 int error; 864 865 clear_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events); 866 clear_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events); 867 clear_bit(RXRPC_CALL_EV_ABORT, &call->events); 868 869 error = call->error_report; 870 if (error < RXRPC_LOCAL_ERROR_OFFSET) { 871 mark = RXRPC_SKB_MARK_NET_ERROR; 872 _debug("post net error %d", error); 873 } else { 874 mark = RXRPC_SKB_MARK_LOCAL_ERROR; 875 error -= RXRPC_LOCAL_ERROR_OFFSET; 876 _debug("post net local error %d", error); 877 } 878 879 if (rxrpc_post_message(call, mark, error, true) < 0) 880 goto no_mem; 881 clear_bit(RXRPC_CALL_EV_RCVD_ERROR, &call->events); 882 goto kill_ACKs; 883 } 884 885 if (test_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events)) { 886 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE); 887 888 clear_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events); 889 clear_bit(RXRPC_CALL_EV_ABORT, &call->events); 890 891 _debug("post conn abort"); 892 893 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR, 894 call->conn->error, true) < 0) 895 goto no_mem; 896 clear_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events); 897 goto kill_ACKs; 898 } 899 900 if (test_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events)) { 901 whdr.type = RXRPC_PACKET_TYPE_BUSY; 902 genbit = RXRPC_CALL_EV_REJECT_BUSY; 903 goto send_message; 904 } 905 906 if (test_bit(RXRPC_CALL_EV_ABORT, &call->events)) { 907 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE); 908 909 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR, 910 ECONNABORTED, true) < 0) 911 goto no_mem; 912 whdr.type = RXRPC_PACKET_TYPE_ABORT; 913 data = htonl(call->local_abort); 914 iov[1].iov_base = &data; 915 iov[1].iov_len = sizeof(data); 916 genbit = RXRPC_CALL_EV_ABORT; 917 goto send_message; 918 } 919 920 if (test_bit(RXRPC_CALL_EV_ACK_FINAL, &call->events)) { 921 genbit = RXRPC_CALL_EV_ACK_FINAL; 922 923 ack.bufferSpace = htons(8); 924 ack.maxSkew = 0; 925 ack.serial = 0; 926 ack.reason = RXRPC_ACK_IDLE; 927 ack.nAcks = 0; 928 call->ackr_reason = 0; 929 930 spin_lock_bh(&call->lock); 931 ack.serial = htonl(call->ackr_serial); 932 ack.previousPacket = htonl(call->ackr_prev_seq); 933 ack.firstPacket = htonl(call->rx_data_eaten + 1); 934 spin_unlock_bh(&call->lock); 935 936 pad = 0; 937 938 iov[1].iov_base = &ack; 939 iov[1].iov_len = sizeof(ack); 940 iov[2].iov_base = &pad; 941 iov[2].iov_len = 3; 942 iov[3].iov_base = &ackinfo; 943 iov[3].iov_len = sizeof(ackinfo); 944 goto send_ACK; 945 } 946 947 if (call->events & ((1 << RXRPC_CALL_EV_RCVD_BUSY) | 948 (1 << RXRPC_CALL_EV_RCVD_ABORT)) 949 ) { 950 u32 mark; 951 952 if (test_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events)) 953 mark = RXRPC_SKB_MARK_REMOTE_ABORT; 954 else 955 mark = RXRPC_SKB_MARK_BUSY; 956 957 _debug("post abort/busy"); 958 rxrpc_clear_tx_window(call); 959 if (rxrpc_post_message(call, mark, ECONNABORTED, true) < 0) 960 goto no_mem; 961 962 clear_bit(RXRPC_CALL_EV_RCVD_BUSY, &call->events); 963 clear_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events); 964 goto kill_ACKs; 965 } 966 967 if (test_and_clear_bit(RXRPC_CALL_EV_RCVD_ACKALL, &call->events)) { 968 _debug("do implicit ackall"); 969 rxrpc_clear_tx_window(call); 970 } 971 972 if (test_bit(RXRPC_CALL_EV_LIFE_TIMER, &call->events)) { 973 write_lock_bh(&call->state_lock); 974 if (call->state <= RXRPC_CALL_COMPLETE) { 975 call->state = RXRPC_CALL_LOCALLY_ABORTED; 976 call->local_abort = RX_CALL_TIMEOUT; 977 set_bit(RXRPC_CALL_EV_ABORT, &call->events); 978 } 979 write_unlock_bh(&call->state_lock); 980 981 _debug("post timeout"); 982 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR, 983 ETIME, true) < 0) 984 goto no_mem; 985 986 clear_bit(RXRPC_CALL_EV_LIFE_TIMER, &call->events); 987 goto kill_ACKs; 988 } 989 990 /* deal with assorted inbound messages */ 991 if (!skb_queue_empty(&call->rx_queue)) { 992 switch (rxrpc_process_rx_queue(call, &abort_code)) { 993 case 0: 994 case -EAGAIN: 995 break; 996 case -ENOMEM: 997 goto no_mem; 998 case -EKEYEXPIRED: 999 case -EKEYREJECTED: 1000 case -EPROTO: 1001 rxrpc_abort_call(call, abort_code); 1002 goto kill_ACKs; 1003 } 1004 } 1005 1006 /* handle resending */ 1007 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events)) 1008 rxrpc_resend_timer(call); 1009 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND, &call->events)) 1010 rxrpc_resend(call); 1011 1012 /* consider sending an ordinary ACK */ 1013 if (test_bit(RXRPC_CALL_EV_ACK, &call->events)) { 1014 _debug("send ACK: window: %d - %d { %lx }", 1015 call->rx_data_eaten, call->ackr_win_top, 1016 call->ackr_window[0]); 1017 1018 if (call->state > RXRPC_CALL_SERVER_ACK_REQUEST && 1019 call->ackr_reason != RXRPC_ACK_PING_RESPONSE) { 1020 /* ACK by sending reply DATA packet in this state */ 1021 clear_bit(RXRPC_CALL_EV_ACK, &call->events); 1022 goto maybe_reschedule; 1023 } 1024 1025 genbit = RXRPC_CALL_EV_ACK; 1026 1027 acks = kzalloc(call->ackr_win_top - call->rx_data_eaten, 1028 GFP_NOFS); 1029 if (!acks) 1030 goto no_mem; 1031 1032 //hdr.flags = RXRPC_SLOW_START_OK; 1033 ack.bufferSpace = htons(8); 1034 ack.maxSkew = 0; 1035 1036 spin_lock_bh(&call->lock); 1037 ack.reason = call->ackr_reason; 1038 ack.serial = htonl(call->ackr_serial); 1039 ack.previousPacket = htonl(call->ackr_prev_seq); 1040 ack.firstPacket = htonl(call->rx_data_eaten + 1); 1041 1042 ack.nAcks = 0; 1043 for (loop = 0; loop < RXRPC_ACKR_WINDOW_ASZ; loop++) { 1044 nbit = loop * BITS_PER_LONG; 1045 for (bits = call->ackr_window[loop]; bits; bits >>= 1 1046 ) { 1047 _debug("- l=%d n=%d b=%lx", loop, nbit, bits); 1048 if (bits & 1) { 1049 acks[nbit] = RXRPC_ACK_TYPE_ACK; 1050 ack.nAcks = nbit + 1; 1051 } 1052 nbit++; 1053 } 1054 } 1055 call->ackr_reason = 0; 1056 spin_unlock_bh(&call->lock); 1057 1058 pad = 0; 1059 1060 iov[1].iov_base = &ack; 1061 iov[1].iov_len = sizeof(ack); 1062 iov[2].iov_base = acks; 1063 iov[2].iov_len = ack.nAcks; 1064 iov[3].iov_base = &pad; 1065 iov[3].iov_len = 3; 1066 iov[4].iov_base = &ackinfo; 1067 iov[4].iov_len = sizeof(ackinfo); 1068 1069 switch (ack.reason) { 1070 case RXRPC_ACK_REQUESTED: 1071 case RXRPC_ACK_DUPLICATE: 1072 case RXRPC_ACK_OUT_OF_SEQUENCE: 1073 case RXRPC_ACK_EXCEEDS_WINDOW: 1074 case RXRPC_ACK_NOSPACE: 1075 case RXRPC_ACK_PING: 1076 case RXRPC_ACK_PING_RESPONSE: 1077 goto send_ACK_with_skew; 1078 case RXRPC_ACK_DELAY: 1079 case RXRPC_ACK_IDLE: 1080 goto send_ACK; 1081 } 1082 } 1083 1084 /* handle completion of security negotiations on an incoming 1085 * connection */ 1086 if (test_and_clear_bit(RXRPC_CALL_EV_SECURED, &call->events)) { 1087 _debug("secured"); 1088 spin_lock_bh(&call->lock); 1089 1090 if (call->state == RXRPC_CALL_SERVER_SECURING) { 1091 _debug("securing"); 1092 write_lock(&call->socket->call_lock); 1093 if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) && 1094 !test_bit(RXRPC_CALL_EV_RELEASE, &call->events)) { 1095 _debug("not released"); 1096 call->state = RXRPC_CALL_SERVER_ACCEPTING; 1097 list_move_tail(&call->accept_link, 1098 &call->socket->acceptq); 1099 } 1100 write_unlock(&call->socket->call_lock); 1101 read_lock(&call->state_lock); 1102 if (call->state < RXRPC_CALL_COMPLETE) 1103 set_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events); 1104 read_unlock(&call->state_lock); 1105 } 1106 1107 spin_unlock_bh(&call->lock); 1108 if (!test_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events)) 1109 goto maybe_reschedule; 1110 } 1111 1112 /* post a notification of an acceptable connection to the app */ 1113 if (test_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events)) { 1114 _debug("post accept"); 1115 if (rxrpc_post_message(call, RXRPC_SKB_MARK_NEW_CALL, 1116 0, false) < 0) 1117 goto no_mem; 1118 clear_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events); 1119 goto maybe_reschedule; 1120 } 1121 1122 /* handle incoming call acceptance */ 1123 if (test_and_clear_bit(RXRPC_CALL_EV_ACCEPTED, &call->events)) { 1124 _debug("accepted"); 1125 ASSERTCMP(call->rx_data_post, ==, 0); 1126 call->rx_data_post = 1; 1127 read_lock_bh(&call->state_lock); 1128 if (call->state < RXRPC_CALL_COMPLETE) 1129 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events); 1130 read_unlock_bh(&call->state_lock); 1131 } 1132 1133 /* drain the out of sequence received packet queue into the packet Rx 1134 * queue */ 1135 if (test_and_clear_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events)) { 1136 while (call->rx_data_post == call->rx_first_oos) 1137 if (rxrpc_drain_rx_oos_queue(call) < 0) 1138 break; 1139 goto maybe_reschedule; 1140 } 1141 1142 if (test_bit(RXRPC_CALL_EV_RELEASE, &call->events)) { 1143 rxrpc_release_call(call); 1144 clear_bit(RXRPC_CALL_EV_RELEASE, &call->events); 1145 } 1146 1147 /* other events may have been raised since we started checking */ 1148 goto maybe_reschedule; 1149 1150 send_ACK_with_skew: 1151 ack.maxSkew = htons(atomic_read(&call->conn->hi_serial) - 1152 ntohl(ack.serial)); 1153 send_ACK: 1154 mtu = call->conn->params.peer->if_mtu; 1155 mtu -= call->conn->params.peer->hdrsize; 1156 ackinfo.maxMTU = htonl(mtu); 1157 ackinfo.rwind = htonl(rxrpc_rx_window_size); 1158 1159 /* permit the peer to send us jumbo packets if it wants to */ 1160 ackinfo.rxMTU = htonl(rxrpc_rx_mtu); 1161 ackinfo.jumbo_max = htonl(rxrpc_rx_jumbo_max); 1162 1163 serial = atomic_inc_return(&call->conn->serial); 1164 whdr.serial = htonl(serial); 1165 _proto("Tx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }", 1166 serial, 1167 ntohs(ack.maxSkew), 1168 ntohl(ack.firstPacket), 1169 ntohl(ack.previousPacket), 1170 ntohl(ack.serial), 1171 rxrpc_acks(ack.reason), 1172 ack.nAcks); 1173 1174 del_timer_sync(&call->ack_timer); 1175 if (ack.nAcks > 0) 1176 set_bit(RXRPC_CALL_TX_SOFT_ACK, &call->flags); 1177 goto send_message_2; 1178 1179 send_message: 1180 _debug("send message"); 1181 1182 serial = atomic_inc_return(&call->conn->serial); 1183 whdr.serial = htonl(serial); 1184 _proto("Tx %s %%%u", rxrpc_pkts[whdr.type], serial); 1185 send_message_2: 1186 1187 len = iov[0].iov_len; 1188 ioc = 1; 1189 if (iov[4].iov_len) { 1190 ioc = 5; 1191 len += iov[4].iov_len; 1192 len += iov[3].iov_len; 1193 len += iov[2].iov_len; 1194 len += iov[1].iov_len; 1195 } else if (iov[3].iov_len) { 1196 ioc = 4; 1197 len += iov[3].iov_len; 1198 len += iov[2].iov_len; 1199 len += iov[1].iov_len; 1200 } else if (iov[2].iov_len) { 1201 ioc = 3; 1202 len += iov[2].iov_len; 1203 len += iov[1].iov_len; 1204 } else if (iov[1].iov_len) { 1205 ioc = 2; 1206 len += iov[1].iov_len; 1207 } 1208 1209 ret = kernel_sendmsg(call->conn->params.local->socket, 1210 &msg, iov, ioc, len); 1211 if (ret < 0) { 1212 _debug("sendmsg failed: %d", ret); 1213 read_lock_bh(&call->state_lock); 1214 if (call->state < RXRPC_CALL_DEAD) 1215 rxrpc_queue_call(call); 1216 read_unlock_bh(&call->state_lock); 1217 goto error; 1218 } 1219 1220 switch (genbit) { 1221 case RXRPC_CALL_EV_ABORT: 1222 clear_bit(genbit, &call->events); 1223 clear_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events); 1224 goto kill_ACKs; 1225 1226 case RXRPC_CALL_EV_ACK_FINAL: 1227 write_lock_bh(&call->state_lock); 1228 if (call->state == RXRPC_CALL_CLIENT_FINAL_ACK) 1229 call->state = RXRPC_CALL_COMPLETE; 1230 write_unlock_bh(&call->state_lock); 1231 goto kill_ACKs; 1232 1233 default: 1234 clear_bit(genbit, &call->events); 1235 switch (call->state) { 1236 case RXRPC_CALL_CLIENT_AWAIT_REPLY: 1237 case RXRPC_CALL_CLIENT_RECV_REPLY: 1238 case RXRPC_CALL_SERVER_RECV_REQUEST: 1239 case RXRPC_CALL_SERVER_ACK_REQUEST: 1240 _debug("start ACK timer"); 1241 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 1242 call->ackr_serial, false); 1243 default: 1244 break; 1245 } 1246 goto maybe_reschedule; 1247 } 1248 1249 kill_ACKs: 1250 del_timer_sync(&call->ack_timer); 1251 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_FINAL, &call->events)) 1252 rxrpc_put_call(call); 1253 clear_bit(RXRPC_CALL_EV_ACK, &call->events); 1254 1255 maybe_reschedule: 1256 if (call->events || !skb_queue_empty(&call->rx_queue)) { 1257 read_lock_bh(&call->state_lock); 1258 if (call->state < RXRPC_CALL_DEAD) 1259 rxrpc_queue_call(call); 1260 read_unlock_bh(&call->state_lock); 1261 } 1262 1263 /* don't leave aborted connections on the accept queue */ 1264 if (call->state >= RXRPC_CALL_COMPLETE && 1265 !list_empty(&call->accept_link)) { 1266 _debug("X unlinking once-pending call %p { e=%lx f=%lx c=%x }", 1267 call, call->events, call->flags, call->conn->proto.cid); 1268 1269 read_lock_bh(&call->state_lock); 1270 if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) && 1271 !test_and_set_bit(RXRPC_CALL_EV_RELEASE, &call->events)) 1272 rxrpc_queue_call(call); 1273 read_unlock_bh(&call->state_lock); 1274 } 1275 1276 error: 1277 clear_bit(RXRPC_CALL_PROC_BUSY, &call->flags); 1278 kfree(acks); 1279 1280 /* because we don't want two CPUs both processing the work item for one 1281 * call at the same time, we use a flag to note when it's busy; however 1282 * this means there's a race between clearing the flag and setting the 1283 * work pending bit and the work item being processed again */ 1284 if (call->events && !work_pending(&call->processor)) { 1285 _debug("jumpstart %x", call->conn->proto.cid); 1286 rxrpc_queue_call(call); 1287 } 1288 1289 _leave(""); 1290 return; 1291 1292 no_mem: 1293 _debug("out of memory"); 1294 goto maybe_reschedule; 1295 } 1296