1 /* RxRPC packet reception 2 * 3 * Copyright (C) 2007, 2016 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/net.h> 16 #include <linux/skbuff.h> 17 #include <linux/errqueue.h> 18 #include <linux/udp.h> 19 #include <linux/in.h> 20 #include <linux/in6.h> 21 #include <linux/icmp.h> 22 #include <linux/gfp.h> 23 #include <net/sock.h> 24 #include <net/af_rxrpc.h> 25 #include <net/ip.h> 26 #include <net/udp.h> 27 #include <net/net_namespace.h> 28 #include "ar-internal.h" 29 30 static void rxrpc_proto_abort(const char *why, 31 struct rxrpc_call *call, rxrpc_seq_t seq) 32 { 33 if (rxrpc_abort_call(why, call, seq, RX_PROTOCOL_ERROR, -EBADMSG)) { 34 set_bit(RXRPC_CALL_EV_ABORT, &call->events); 35 rxrpc_queue_call(call); 36 } 37 } 38 39 /* 40 * Do TCP-style congestion management [RFC 5681]. 41 */ 42 static void rxrpc_congestion_management(struct rxrpc_call *call, 43 struct sk_buff *skb, 44 struct rxrpc_ack_summary *summary, 45 rxrpc_serial_t acked_serial) 46 { 47 enum rxrpc_congest_change change = rxrpc_cong_no_change; 48 unsigned int cumulative_acks = call->cong_cumul_acks; 49 unsigned int cwnd = call->cong_cwnd; 50 bool resend = false; 51 52 summary->flight_size = 53 (call->tx_top - call->tx_hard_ack) - summary->nr_acks; 54 55 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) { 56 summary->retrans_timeo = true; 57 call->cong_ssthresh = max_t(unsigned int, 58 summary->flight_size / 2, 2); 59 cwnd = 1; 60 if (cwnd >= call->cong_ssthresh && 61 call->cong_mode == RXRPC_CALL_SLOW_START) { 62 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE; 63 call->cong_tstamp = skb->tstamp; 64 cumulative_acks = 0; 65 } 66 } 67 68 cumulative_acks += summary->nr_new_acks; 69 cumulative_acks += summary->nr_rot_new_acks; 70 if (cumulative_acks > 255) 71 cumulative_acks = 255; 72 73 summary->mode = call->cong_mode; 74 summary->cwnd = call->cong_cwnd; 75 summary->ssthresh = call->cong_ssthresh; 76 summary->cumulative_acks = cumulative_acks; 77 summary->dup_acks = call->cong_dup_acks; 78 79 switch (call->cong_mode) { 80 case RXRPC_CALL_SLOW_START: 81 if (summary->nr_nacks > 0) 82 goto packet_loss_detected; 83 if (summary->cumulative_acks > 0) 84 cwnd += 1; 85 if (cwnd >= call->cong_ssthresh) { 86 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE; 87 call->cong_tstamp = skb->tstamp; 88 } 89 goto out; 90 91 case RXRPC_CALL_CONGEST_AVOIDANCE: 92 if (summary->nr_nacks > 0) 93 goto packet_loss_detected; 94 95 /* We analyse the number of packets that get ACK'd per RTT 96 * period and increase the window if we managed to fill it. 97 */ 98 if (call->peer->rtt_usage == 0) 99 goto out; 100 if (ktime_before(skb->tstamp, 101 ktime_add_ns(call->cong_tstamp, 102 call->peer->rtt))) 103 goto out_no_clear_ca; 104 change = rxrpc_cong_rtt_window_end; 105 call->cong_tstamp = skb->tstamp; 106 if (cumulative_acks >= cwnd) 107 cwnd++; 108 goto out; 109 110 case RXRPC_CALL_PACKET_LOSS: 111 if (summary->nr_nacks == 0) 112 goto resume_normality; 113 114 if (summary->new_low_nack) { 115 change = rxrpc_cong_new_low_nack; 116 call->cong_dup_acks = 1; 117 if (call->cong_extra > 1) 118 call->cong_extra = 1; 119 goto send_extra_data; 120 } 121 122 call->cong_dup_acks++; 123 if (call->cong_dup_acks < 3) 124 goto send_extra_data; 125 126 change = rxrpc_cong_begin_retransmission; 127 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT; 128 call->cong_ssthresh = max_t(unsigned int, 129 summary->flight_size / 2, 2); 130 cwnd = call->cong_ssthresh + 3; 131 call->cong_extra = 0; 132 call->cong_dup_acks = 0; 133 resend = true; 134 goto out; 135 136 case RXRPC_CALL_FAST_RETRANSMIT: 137 if (!summary->new_low_nack) { 138 if (summary->nr_new_acks == 0) 139 cwnd += 1; 140 call->cong_dup_acks++; 141 if (call->cong_dup_acks == 2) { 142 change = rxrpc_cong_retransmit_again; 143 call->cong_dup_acks = 0; 144 resend = true; 145 } 146 } else { 147 change = rxrpc_cong_progress; 148 cwnd = call->cong_ssthresh; 149 if (summary->nr_nacks == 0) 150 goto resume_normality; 151 } 152 goto out; 153 154 default: 155 BUG(); 156 goto out; 157 } 158 159 resume_normality: 160 change = rxrpc_cong_cleared_nacks; 161 call->cong_dup_acks = 0; 162 call->cong_extra = 0; 163 call->cong_tstamp = skb->tstamp; 164 if (cwnd < call->cong_ssthresh) 165 call->cong_mode = RXRPC_CALL_SLOW_START; 166 else 167 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE; 168 out: 169 cumulative_acks = 0; 170 out_no_clear_ca: 171 if (cwnd >= RXRPC_RXTX_BUFF_SIZE - 1) 172 cwnd = RXRPC_RXTX_BUFF_SIZE - 1; 173 call->cong_cwnd = cwnd; 174 call->cong_cumul_acks = cumulative_acks; 175 trace_rxrpc_congest(call, summary, acked_serial, change); 176 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events)) 177 rxrpc_queue_call(call); 178 return; 179 180 packet_loss_detected: 181 change = rxrpc_cong_saw_nack; 182 call->cong_mode = RXRPC_CALL_PACKET_LOSS; 183 call->cong_dup_acks = 0; 184 goto send_extra_data; 185 186 send_extra_data: 187 /* Send some previously unsent DATA if we have some to advance the ACK 188 * state. 189 */ 190 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] & 191 RXRPC_TX_ANNO_LAST || 192 summary->nr_acks != call->tx_top - call->tx_hard_ack) { 193 call->cong_extra++; 194 wake_up(&call->waitq); 195 } 196 goto out_no_clear_ca; 197 } 198 199 /* 200 * Ping the other end to fill our RTT cache and to retrieve the rwind 201 * and MTU parameters. 202 */ 203 static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb, 204 int skew) 205 { 206 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 207 ktime_t now = skb->tstamp; 208 209 if (call->peer->rtt_usage < 3 || 210 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now)) 211 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial, 212 true, true, 213 rxrpc_propose_ack_ping_for_params); 214 } 215 216 /* 217 * Apply a hard ACK by advancing the Tx window. 218 */ 219 static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to, 220 struct rxrpc_ack_summary *summary) 221 { 222 struct sk_buff *skb, *list = NULL; 223 int ix; 224 u8 annotation; 225 226 if (call->acks_lowest_nak == call->tx_hard_ack) { 227 call->acks_lowest_nak = to; 228 } else if (before_eq(call->acks_lowest_nak, to)) { 229 summary->new_low_nack = true; 230 call->acks_lowest_nak = to; 231 } 232 233 spin_lock(&call->lock); 234 235 while (before(call->tx_hard_ack, to)) { 236 call->tx_hard_ack++; 237 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK; 238 skb = call->rxtx_buffer[ix]; 239 annotation = call->rxtx_annotations[ix]; 240 rxrpc_see_skb(skb, rxrpc_skb_tx_rotated); 241 call->rxtx_buffer[ix] = NULL; 242 call->rxtx_annotations[ix] = 0; 243 skb->next = list; 244 list = skb; 245 246 if (annotation & RXRPC_TX_ANNO_LAST) 247 set_bit(RXRPC_CALL_TX_LAST, &call->flags); 248 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK) 249 summary->nr_rot_new_acks++; 250 } 251 252 spin_unlock(&call->lock); 253 254 trace_rxrpc_transmit(call, (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ? 255 rxrpc_transmit_rotate_last : 256 rxrpc_transmit_rotate)); 257 wake_up(&call->waitq); 258 259 while (list) { 260 skb = list; 261 list = skb->next; 262 skb->next = NULL; 263 rxrpc_free_skb(skb, rxrpc_skb_tx_freed); 264 } 265 } 266 267 /* 268 * End the transmission phase of a call. 269 * 270 * This occurs when we get an ACKALL packet, the first DATA packet of a reply, 271 * or a final ACK packet. 272 */ 273 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun, 274 const char *abort_why) 275 { 276 277 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags)); 278 279 write_lock(&call->state_lock); 280 281 switch (call->state) { 282 case RXRPC_CALL_CLIENT_SEND_REQUEST: 283 case RXRPC_CALL_CLIENT_AWAIT_REPLY: 284 if (reply_begun) 285 call->state = RXRPC_CALL_CLIENT_RECV_REPLY; 286 else 287 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY; 288 break; 289 290 case RXRPC_CALL_SERVER_AWAIT_ACK: 291 __rxrpc_call_completed(call); 292 rxrpc_notify_socket(call); 293 break; 294 295 default: 296 goto bad_state; 297 } 298 299 write_unlock(&call->state_lock); 300 if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) { 301 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply); 302 } else { 303 trace_rxrpc_transmit(call, rxrpc_transmit_end); 304 } 305 _leave(" = ok"); 306 return true; 307 308 bad_state: 309 write_unlock(&call->state_lock); 310 kdebug("end_tx %s", rxrpc_call_states[call->state]); 311 rxrpc_proto_abort(abort_why, call, call->tx_top); 312 return false; 313 } 314 315 /* 316 * Begin the reply reception phase of a call. 317 */ 318 static bool rxrpc_receiving_reply(struct rxrpc_call *call) 319 { 320 struct rxrpc_ack_summary summary = { 0 }; 321 rxrpc_seq_t top = READ_ONCE(call->tx_top); 322 323 if (call->ackr_reason) { 324 spin_lock_bh(&call->lock); 325 call->ackr_reason = 0; 326 call->resend_at = call->expire_at; 327 call->ack_at = call->expire_at; 328 spin_unlock_bh(&call->lock); 329 rxrpc_set_timer(call, rxrpc_timer_init_for_reply, 330 ktime_get_real()); 331 } 332 333 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) 334 rxrpc_rotate_tx_window(call, top, &summary); 335 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) { 336 rxrpc_proto_abort("TXL", call, top); 337 return false; 338 } 339 if (!rxrpc_end_tx_phase(call, true, "ETD")) 340 return false; 341 call->tx_phase = false; 342 return true; 343 } 344 345 /* 346 * Scan a jumbo packet to validate its structure and to work out how many 347 * subpackets it contains. 348 * 349 * A jumbo packet is a collection of consecutive packets glued together with 350 * little headers between that indicate how to change the initial header for 351 * each subpacket. 352 * 353 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but 354 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any 355 * size. 356 */ 357 static bool rxrpc_validate_jumbo(struct sk_buff *skb) 358 { 359 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 360 unsigned int offset = sizeof(struct rxrpc_wire_header); 361 unsigned int len = skb->len; 362 int nr_jumbo = 1; 363 u8 flags = sp->hdr.flags; 364 365 do { 366 nr_jumbo++; 367 if (len - offset < RXRPC_JUMBO_SUBPKTLEN) 368 goto protocol_error; 369 if (flags & RXRPC_LAST_PACKET) 370 goto protocol_error; 371 offset += RXRPC_JUMBO_DATALEN; 372 if (skb_copy_bits(skb, offset, &flags, 1) < 0) 373 goto protocol_error; 374 offset += sizeof(struct rxrpc_jumbo_header); 375 } while (flags & RXRPC_JUMBO_PACKET); 376 377 sp->nr_jumbo = nr_jumbo; 378 return true; 379 380 protocol_error: 381 return false; 382 } 383 384 /* 385 * Handle reception of a duplicate packet. 386 * 387 * We have to take care to avoid an attack here whereby we're given a series of 388 * jumbograms, each with a sequence number one before the preceding one and 389 * filled up to maximum UDP size. If they never send us the first packet in 390 * the sequence, they can cause us to have to hold on to around 2MiB of kernel 391 * space until the call times out. 392 * 393 * We limit the space usage by only accepting three duplicate jumbo packets per 394 * call. After that, we tell the other side we're no longer accepting jumbos 395 * (that information is encoded in the ACK packet). 396 */ 397 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq, 398 u8 annotation, bool *_jumbo_bad) 399 { 400 /* Discard normal packets that are duplicates. */ 401 if (annotation == 0) 402 return; 403 404 /* Skip jumbo subpackets that are duplicates. When we've had three or 405 * more partially duplicate jumbo packets, we refuse to take any more 406 * jumbos for this call. 407 */ 408 if (!*_jumbo_bad) { 409 call->nr_jumbo_bad++; 410 *_jumbo_bad = true; 411 } 412 } 413 414 /* 415 * Process a DATA packet, adding the packet to the Rx ring. 416 */ 417 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb, 418 u16 skew) 419 { 420 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 421 enum rxrpc_call_state state; 422 unsigned int offset = sizeof(struct rxrpc_wire_header); 423 unsigned int ix; 424 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0; 425 rxrpc_seq_t seq = sp->hdr.seq, hard_ack; 426 bool immediate_ack = false, jumbo_bad = false, queued; 427 u16 len; 428 u8 ack = 0, flags, annotation = 0; 429 430 _enter("{%u,%u},{%u,%u}", 431 call->rx_hard_ack, call->rx_top, skb->len, seq); 432 433 _proto("Rx DATA %%%u { #%u f=%02x }", 434 sp->hdr.serial, seq, sp->hdr.flags); 435 436 state = READ_ONCE(call->state); 437 if (state >= RXRPC_CALL_COMPLETE) 438 return; 439 440 /* Received data implicitly ACKs all of the request packets we sent 441 * when we're acting as a client. 442 */ 443 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST || 444 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) && 445 !rxrpc_receiving_reply(call)) 446 return; 447 448 call->ackr_prev_seq = seq; 449 450 hard_ack = READ_ONCE(call->rx_hard_ack); 451 if (after(seq, hard_ack + call->rx_winsize)) { 452 ack = RXRPC_ACK_EXCEEDS_WINDOW; 453 ack_serial = serial; 454 goto ack; 455 } 456 457 flags = sp->hdr.flags; 458 if (flags & RXRPC_JUMBO_PACKET) { 459 if (call->nr_jumbo_bad > 3) { 460 ack = RXRPC_ACK_NOSPACE; 461 ack_serial = serial; 462 goto ack; 463 } 464 annotation = 1; 465 } 466 467 next_subpacket: 468 queued = false; 469 ix = seq & RXRPC_RXTX_BUFF_MASK; 470 len = skb->len; 471 if (flags & RXRPC_JUMBO_PACKET) 472 len = RXRPC_JUMBO_DATALEN; 473 474 if (flags & RXRPC_LAST_PACKET) { 475 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) && 476 seq != call->rx_top) 477 return rxrpc_proto_abort("LSN", call, seq); 478 } else { 479 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) && 480 after_eq(seq, call->rx_top)) 481 return rxrpc_proto_abort("LSA", call, seq); 482 } 483 484 trace_rxrpc_rx_data(call, seq, serial, flags, annotation); 485 if (before_eq(seq, hard_ack)) { 486 ack = RXRPC_ACK_DUPLICATE; 487 ack_serial = serial; 488 goto skip; 489 } 490 491 if (flags & RXRPC_REQUEST_ACK && !ack) { 492 ack = RXRPC_ACK_REQUESTED; 493 ack_serial = serial; 494 } 495 496 if (call->rxtx_buffer[ix]) { 497 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad); 498 if (ack != RXRPC_ACK_DUPLICATE) { 499 ack = RXRPC_ACK_DUPLICATE; 500 ack_serial = serial; 501 } 502 immediate_ack = true; 503 goto skip; 504 } 505 506 /* Queue the packet. We use a couple of memory barriers here as need 507 * to make sure that rx_top is perceived to be set after the buffer 508 * pointer and that the buffer pointer is set after the annotation and 509 * the skb data. 510 * 511 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window() 512 * and also rxrpc_fill_out_ack(). 513 */ 514 rxrpc_get_skb(skb, rxrpc_skb_rx_got); 515 call->rxtx_annotations[ix] = annotation; 516 smp_wmb(); 517 call->rxtx_buffer[ix] = skb; 518 if (after(seq, call->rx_top)) { 519 smp_store_release(&call->rx_top, seq); 520 } else if (before(seq, call->rx_top)) { 521 /* Send an immediate ACK if we fill in a hole */ 522 if (!ack) { 523 ack = RXRPC_ACK_DELAY; 524 ack_serial = serial; 525 } 526 immediate_ack = true; 527 } 528 if (flags & RXRPC_LAST_PACKET) { 529 set_bit(RXRPC_CALL_RX_LAST, &call->flags); 530 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq); 531 } else { 532 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq); 533 } 534 queued = true; 535 536 if (after_eq(seq, call->rx_expect_next)) { 537 if (after(seq, call->rx_expect_next)) { 538 _net("OOS %u > %u", seq, call->rx_expect_next); 539 ack = RXRPC_ACK_OUT_OF_SEQUENCE; 540 ack_serial = serial; 541 } 542 call->rx_expect_next = seq + 1; 543 } 544 545 skip: 546 offset += len; 547 if (flags & RXRPC_JUMBO_PACKET) { 548 if (skb_copy_bits(skb, offset, &flags, 1) < 0) 549 return rxrpc_proto_abort("XJF", call, seq); 550 offset += sizeof(struct rxrpc_jumbo_header); 551 seq++; 552 serial++; 553 annotation++; 554 if (flags & RXRPC_JUMBO_PACKET) 555 annotation |= RXRPC_RX_ANNO_JLAST; 556 if (after(seq, hard_ack + call->rx_winsize)) { 557 ack = RXRPC_ACK_EXCEEDS_WINDOW; 558 ack_serial = serial; 559 if (!jumbo_bad) { 560 call->nr_jumbo_bad++; 561 jumbo_bad = true; 562 } 563 goto ack; 564 } 565 566 _proto("Rx DATA Jumbo %%%u", serial); 567 goto next_subpacket; 568 } 569 570 if (queued && flags & RXRPC_LAST_PACKET && !ack) { 571 ack = RXRPC_ACK_DELAY; 572 ack_serial = serial; 573 } 574 575 ack: 576 if (ack) 577 rxrpc_propose_ACK(call, ack, skew, ack_serial, 578 immediate_ack, true, 579 rxrpc_propose_ack_input_data); 580 581 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1) 582 rxrpc_notify_socket(call); 583 _leave(" [queued]"); 584 } 585 586 /* 587 * Process a requested ACK. 588 */ 589 static void rxrpc_input_requested_ack(struct rxrpc_call *call, 590 ktime_t resp_time, 591 rxrpc_serial_t orig_serial, 592 rxrpc_serial_t ack_serial) 593 { 594 struct rxrpc_skb_priv *sp; 595 struct sk_buff *skb; 596 ktime_t sent_at; 597 int ix; 598 599 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) { 600 skb = call->rxtx_buffer[ix]; 601 if (!skb) 602 continue; 603 604 sp = rxrpc_skb(skb); 605 if (sp->hdr.serial != orig_serial) 606 continue; 607 smp_rmb(); 608 sent_at = skb->tstamp; 609 goto found; 610 } 611 return; 612 613 found: 614 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack, 615 orig_serial, ack_serial, sent_at, resp_time); 616 } 617 618 /* 619 * Process a ping response. 620 */ 621 static void rxrpc_input_ping_response(struct rxrpc_call *call, 622 ktime_t resp_time, 623 rxrpc_serial_t orig_serial, 624 rxrpc_serial_t ack_serial) 625 { 626 rxrpc_serial_t ping_serial; 627 ktime_t ping_time; 628 629 ping_time = call->ping_time; 630 smp_rmb(); 631 ping_serial = call->ping_serial; 632 633 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) || 634 before(orig_serial, ping_serial)) 635 return; 636 clear_bit(RXRPC_CALL_PINGING, &call->flags); 637 if (after(orig_serial, ping_serial)) 638 return; 639 640 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response, 641 orig_serial, ack_serial, ping_time, resp_time); 642 } 643 644 /* 645 * Process the extra information that may be appended to an ACK packet 646 */ 647 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb, 648 struct rxrpc_ackinfo *ackinfo) 649 { 650 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 651 struct rxrpc_peer *peer; 652 unsigned int mtu; 653 bool wake = false; 654 u32 rwind = ntohl(ackinfo->rwind); 655 656 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }", 657 sp->hdr.serial, 658 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU), 659 rwind, ntohl(ackinfo->jumbo_max)); 660 661 if (call->tx_winsize != rwind) { 662 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1) 663 rwind = RXRPC_RXTX_BUFF_SIZE - 1; 664 if (rwind > call->tx_winsize) 665 wake = true; 666 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, 667 ntohl(ackinfo->rwind), wake); 668 call->tx_winsize = rwind; 669 } 670 671 if (call->cong_ssthresh > rwind) 672 call->cong_ssthresh = rwind; 673 674 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU)); 675 676 peer = call->peer; 677 if (mtu < peer->maxdata) { 678 spin_lock_bh(&peer->lock); 679 peer->maxdata = mtu; 680 peer->mtu = mtu + peer->hdrsize; 681 spin_unlock_bh(&peer->lock); 682 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata); 683 } 684 685 if (wake) 686 wake_up(&call->waitq); 687 } 688 689 /* 690 * Process individual soft ACKs. 691 * 692 * Each ACK in the array corresponds to one packet and can be either an ACK or 693 * a NAK. If we get find an explicitly NAK'd packet we resend immediately; 694 * packets that lie beyond the end of the ACK list are scheduled for resend by 695 * the timer on the basis that the peer might just not have processed them at 696 * the time the ACK was sent. 697 */ 698 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks, 699 rxrpc_seq_t seq, int nr_acks, 700 struct rxrpc_ack_summary *summary) 701 { 702 int ix; 703 u8 annotation, anno_type; 704 705 for (; nr_acks > 0; nr_acks--, seq++) { 706 ix = seq & RXRPC_RXTX_BUFF_MASK; 707 annotation = call->rxtx_annotations[ix]; 708 anno_type = annotation & RXRPC_TX_ANNO_MASK; 709 annotation &= ~RXRPC_TX_ANNO_MASK; 710 switch (*acks++) { 711 case RXRPC_ACK_TYPE_ACK: 712 summary->nr_acks++; 713 if (anno_type == RXRPC_TX_ANNO_ACK) 714 continue; 715 summary->nr_new_acks++; 716 call->rxtx_annotations[ix] = 717 RXRPC_TX_ANNO_ACK | annotation; 718 break; 719 case RXRPC_ACK_TYPE_NACK: 720 if (!summary->nr_nacks && 721 call->acks_lowest_nak != seq) { 722 call->acks_lowest_nak = seq; 723 summary->new_low_nack = true; 724 } 725 summary->nr_nacks++; 726 if (anno_type == RXRPC_TX_ANNO_NAK) 727 continue; 728 summary->nr_new_nacks++; 729 if (anno_type == RXRPC_TX_ANNO_RETRANS) 730 continue; 731 call->rxtx_annotations[ix] = 732 RXRPC_TX_ANNO_NAK | annotation; 733 break; 734 default: 735 return rxrpc_proto_abort("SFT", call, 0); 736 } 737 } 738 } 739 740 /* 741 * Process an ACK packet. 742 * 743 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet 744 * in the ACK array. Anything before that is hard-ACK'd and may be discarded. 745 * 746 * A hard-ACK means that a packet has been processed and may be discarded; a 747 * soft-ACK means that the packet may be discarded and retransmission 748 * requested. A phase is complete when all packets are hard-ACK'd. 749 */ 750 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb, 751 u16 skew) 752 { 753 struct rxrpc_ack_summary summary = { 0 }; 754 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 755 union { 756 struct rxrpc_ackpacket ack; 757 struct rxrpc_ackinfo info; 758 u8 acks[RXRPC_MAXACKS]; 759 } buf; 760 rxrpc_serial_t acked_serial; 761 rxrpc_seq_t first_soft_ack, hard_ack; 762 int nr_acks, offset, ioffset; 763 764 _enter(""); 765 766 offset = sizeof(struct rxrpc_wire_header); 767 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) { 768 _debug("extraction failure"); 769 return rxrpc_proto_abort("XAK", call, 0); 770 } 771 offset += sizeof(buf.ack); 772 773 acked_serial = ntohl(buf.ack.serial); 774 first_soft_ack = ntohl(buf.ack.firstPacket); 775 hard_ack = first_soft_ack - 1; 776 nr_acks = buf.ack.nAcks; 777 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ? 778 buf.ack.reason : RXRPC_ACK__INVALID); 779 780 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial, 781 first_soft_ack, ntohl(buf.ack.previousPacket), 782 summary.ack_reason, nr_acks); 783 784 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE) 785 rxrpc_input_ping_response(call, skb->tstamp, acked_serial, 786 sp->hdr.serial); 787 if (buf.ack.reason == RXRPC_ACK_REQUESTED) 788 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial, 789 sp->hdr.serial); 790 791 if (buf.ack.reason == RXRPC_ACK_PING) { 792 _proto("Rx ACK %%%u PING Request", sp->hdr.serial); 793 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE, 794 skew, sp->hdr.serial, true, true, 795 rxrpc_propose_ack_respond_to_ping); 796 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) { 797 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED, 798 skew, sp->hdr.serial, true, true, 799 rxrpc_propose_ack_respond_to_ack); 800 } 801 802 ioffset = offset + nr_acks + 3; 803 if (skb->len >= ioffset + sizeof(buf.info)) { 804 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0) 805 return rxrpc_proto_abort("XAI", call, 0); 806 rxrpc_input_ackinfo(call, skb, &buf.info); 807 } 808 809 if (first_soft_ack == 0) 810 return rxrpc_proto_abort("AK0", call, 0); 811 812 /* Ignore ACKs unless we are or have just been transmitting. */ 813 switch (READ_ONCE(call->state)) { 814 case RXRPC_CALL_CLIENT_SEND_REQUEST: 815 case RXRPC_CALL_CLIENT_AWAIT_REPLY: 816 case RXRPC_CALL_SERVER_SEND_REPLY: 817 case RXRPC_CALL_SERVER_AWAIT_ACK: 818 break; 819 default: 820 return; 821 } 822 823 /* Discard any out-of-order or duplicate ACKs. */ 824 if (before_eq(sp->hdr.serial, call->acks_latest)) { 825 _debug("discard ACK %d <= %d", 826 sp->hdr.serial, call->acks_latest); 827 return; 828 } 829 call->acks_latest_ts = skb->tstamp; 830 call->acks_latest = sp->hdr.serial; 831 832 if (before(hard_ack, call->tx_hard_ack) || 833 after(hard_ack, call->tx_top)) 834 return rxrpc_proto_abort("AKW", call, 0); 835 if (nr_acks > call->tx_top - hard_ack) 836 return rxrpc_proto_abort("AKN", call, 0); 837 838 if (after(hard_ack, call->tx_hard_ack)) 839 rxrpc_rotate_tx_window(call, hard_ack, &summary); 840 841 if (nr_acks > 0) { 842 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) 843 return rxrpc_proto_abort("XSA", call, 0); 844 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks, 845 &summary); 846 } 847 848 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) { 849 rxrpc_end_tx_phase(call, false, "ETA"); 850 return; 851 } 852 853 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] & 854 RXRPC_TX_ANNO_LAST && 855 summary.nr_acks == call->tx_top - hard_ack && 856 rxrpc_is_client_call(call)) 857 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial, 858 false, true, 859 rxrpc_propose_ack_ping_for_lost_reply); 860 861 return rxrpc_congestion_management(call, skb, &summary, acked_serial); 862 } 863 864 /* 865 * Process an ACKALL packet. 866 */ 867 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb) 868 { 869 struct rxrpc_ack_summary summary = { 0 }; 870 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 871 872 _proto("Rx ACKALL %%%u", sp->hdr.serial); 873 874 rxrpc_rotate_tx_window(call, call->tx_top, &summary); 875 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) 876 rxrpc_end_tx_phase(call, false, "ETL"); 877 } 878 879 /* 880 * Process an ABORT packet directed at a call. 881 */ 882 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb) 883 { 884 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 885 __be32 wtmp; 886 u32 abort_code = RX_CALL_DEAD; 887 888 _enter(""); 889 890 if (skb->len >= 4 && 891 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header), 892 &wtmp, sizeof(wtmp)) >= 0) 893 abort_code = ntohl(wtmp); 894 895 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code); 896 897 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code); 898 899 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED, 900 abort_code, -ECONNABORTED)) 901 rxrpc_notify_socket(call); 902 } 903 904 /* 905 * Process an incoming call packet. 906 */ 907 static void rxrpc_input_call_packet(struct rxrpc_call *call, 908 struct sk_buff *skb, u16 skew) 909 { 910 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 911 912 _enter("%p,%p", call, skb); 913 914 switch (sp->hdr.type) { 915 case RXRPC_PACKET_TYPE_DATA: 916 rxrpc_input_data(call, skb, skew); 917 break; 918 919 case RXRPC_PACKET_TYPE_ACK: 920 rxrpc_input_ack(call, skb, skew); 921 break; 922 923 case RXRPC_PACKET_TYPE_BUSY: 924 _proto("Rx BUSY %%%u", sp->hdr.serial); 925 926 /* Just ignore BUSY packets from the server; the retry and 927 * lifespan timers will take care of business. BUSY packets 928 * from the client don't make sense. 929 */ 930 break; 931 932 case RXRPC_PACKET_TYPE_ABORT: 933 rxrpc_input_abort(call, skb); 934 break; 935 936 case RXRPC_PACKET_TYPE_ACKALL: 937 rxrpc_input_ackall(call, skb); 938 break; 939 940 default: 941 break; 942 } 943 944 _leave(""); 945 } 946 947 /* 948 * Handle a new call on a channel implicitly completing the preceding call on 949 * that channel. 950 * 951 * TODO: If callNumber > call_id + 1, renegotiate security. 952 */ 953 static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn, 954 struct rxrpc_call *call) 955 { 956 switch (READ_ONCE(call->state)) { 957 case RXRPC_CALL_SERVER_AWAIT_ACK: 958 rxrpc_call_completed(call); 959 break; 960 case RXRPC_CALL_COMPLETE: 961 break; 962 default: 963 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) { 964 set_bit(RXRPC_CALL_EV_ABORT, &call->events); 965 rxrpc_queue_call(call); 966 } 967 break; 968 } 969 970 trace_rxrpc_improper_term(call); 971 __rxrpc_disconnect_call(conn, call); 972 rxrpc_notify_socket(call); 973 } 974 975 /* 976 * post connection-level events to the connection 977 * - this includes challenges, responses, some aborts and call terminal packet 978 * retransmission. 979 */ 980 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn, 981 struct sk_buff *skb) 982 { 983 _enter("%p,%p", conn, skb); 984 985 skb_queue_tail(&conn->rx_queue, skb); 986 rxrpc_queue_conn(conn); 987 } 988 989 /* 990 * post endpoint-level events to the local endpoint 991 * - this includes debug and version messages 992 */ 993 static void rxrpc_post_packet_to_local(struct rxrpc_local *local, 994 struct sk_buff *skb) 995 { 996 _enter("%p,%p", local, skb); 997 998 skb_queue_tail(&local->event_queue, skb); 999 rxrpc_queue_local(local); 1000 } 1001 1002 /* 1003 * put a packet up for transport-level abort 1004 */ 1005 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb) 1006 { 1007 CHECK_SLAB_OKAY(&local->usage); 1008 1009 skb_queue_tail(&local->reject_queue, skb); 1010 rxrpc_queue_local(local); 1011 } 1012 1013 /* 1014 * Extract the wire header from a packet and translate the byte order. 1015 */ 1016 static noinline 1017 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb) 1018 { 1019 struct rxrpc_wire_header whdr; 1020 1021 /* dig out the RxRPC connection details */ 1022 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) { 1023 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial, 1024 tracepoint_string("bad_hdr")); 1025 return -EBADMSG; 1026 } 1027 1028 memset(sp, 0, sizeof(*sp)); 1029 sp->hdr.epoch = ntohl(whdr.epoch); 1030 sp->hdr.cid = ntohl(whdr.cid); 1031 sp->hdr.callNumber = ntohl(whdr.callNumber); 1032 sp->hdr.seq = ntohl(whdr.seq); 1033 sp->hdr.serial = ntohl(whdr.serial); 1034 sp->hdr.flags = whdr.flags; 1035 sp->hdr.type = whdr.type; 1036 sp->hdr.userStatus = whdr.userStatus; 1037 sp->hdr.securityIndex = whdr.securityIndex; 1038 sp->hdr._rsvd = ntohs(whdr._rsvd); 1039 sp->hdr.serviceId = ntohs(whdr.serviceId); 1040 return 0; 1041 } 1042 1043 /* 1044 * handle data received on the local endpoint 1045 * - may be called in interrupt context 1046 * 1047 * The socket is locked by the caller and this prevents the socket from being 1048 * shut down and the local endpoint from going away, thus sk_user_data will not 1049 * be cleared until this function returns. 1050 */ 1051 void rxrpc_data_ready(struct sock *udp_sk) 1052 { 1053 struct rxrpc_connection *conn; 1054 struct rxrpc_channel *chan; 1055 struct rxrpc_call *call; 1056 struct rxrpc_skb_priv *sp; 1057 struct rxrpc_local *local = udp_sk->sk_user_data; 1058 struct sk_buff *skb; 1059 unsigned int channel; 1060 int ret, skew; 1061 1062 _enter("%p", udp_sk); 1063 1064 ASSERT(!irqs_disabled()); 1065 1066 skb = skb_recv_udp(udp_sk, 0, 1, &ret); 1067 if (!skb) { 1068 if (ret == -EAGAIN) 1069 return; 1070 _debug("UDP socket error %d", ret); 1071 return; 1072 } 1073 1074 rxrpc_new_skb(skb, rxrpc_skb_rx_received); 1075 1076 _net("recv skb %p", skb); 1077 1078 /* we'll probably need to checksum it (didn't call sock_recvmsg) */ 1079 if (skb_checksum_complete(skb)) { 1080 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 1081 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0); 1082 _leave(" [CSUM failed]"); 1083 return; 1084 } 1085 1086 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0); 1087 1088 /* The UDP protocol already released all skb resources; 1089 * we are free to add our own data there. 1090 */ 1091 sp = rxrpc_skb(skb); 1092 1093 /* dig out the RxRPC connection details */ 1094 if (rxrpc_extract_header(sp, skb) < 0) 1095 goto bad_message; 1096 1097 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) { 1098 static int lose; 1099 if ((lose++ & 7) == 7) { 1100 trace_rxrpc_rx_lose(sp); 1101 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost); 1102 return; 1103 } 1104 } 1105 1106 trace_rxrpc_rx_packet(sp); 1107 1108 _net("Rx RxRPC %s ep=%x call=%x:%x", 1109 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient", 1110 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber); 1111 1112 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES || 1113 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) { 1114 _proto("Rx Bad Packet Type %u", sp->hdr.type); 1115 goto bad_message; 1116 } 1117 1118 switch (sp->hdr.type) { 1119 case RXRPC_PACKET_TYPE_VERSION: 1120 rxrpc_post_packet_to_local(local, skb); 1121 goto out; 1122 1123 case RXRPC_PACKET_TYPE_BUSY: 1124 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED) 1125 goto discard; 1126 /* Fall through */ 1127 1128 case RXRPC_PACKET_TYPE_DATA: 1129 if (sp->hdr.callNumber == 0) 1130 goto bad_message; 1131 if (sp->hdr.flags & RXRPC_JUMBO_PACKET && 1132 !rxrpc_validate_jumbo(skb)) 1133 goto bad_message; 1134 break; 1135 } 1136 1137 rcu_read_lock(); 1138 1139 conn = rxrpc_find_connection_rcu(local, skb); 1140 if (conn) { 1141 if (sp->hdr.securityIndex != conn->security_ix) 1142 goto wrong_security; 1143 1144 if (sp->hdr.serviceId != conn->service_id) { 1145 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) || 1146 conn->service_id != conn->params.service_id) 1147 goto reupgrade; 1148 conn->service_id = sp->hdr.serviceId; 1149 } 1150 1151 if (sp->hdr.callNumber == 0) { 1152 /* Connection-level packet */ 1153 _debug("CONN %p {%d}", conn, conn->debug_id); 1154 rxrpc_post_packet_to_conn(conn, skb); 1155 goto out_unlock; 1156 } 1157 1158 /* Note the serial number skew here */ 1159 skew = (int)sp->hdr.serial - (int)conn->hi_serial; 1160 if (skew >= 0) { 1161 if (skew > 0) 1162 conn->hi_serial = sp->hdr.serial; 1163 } else { 1164 skew = -skew; 1165 skew = min(skew, 65535); 1166 } 1167 1168 /* Call-bound packets are routed by connection channel. */ 1169 channel = sp->hdr.cid & RXRPC_CHANNELMASK; 1170 chan = &conn->channels[channel]; 1171 1172 /* Ignore really old calls */ 1173 if (sp->hdr.callNumber < chan->last_call) 1174 goto discard_unlock; 1175 1176 if (sp->hdr.callNumber == chan->last_call) { 1177 /* For the previous service call, if completed successfully, we 1178 * discard all further packets. 1179 */ 1180 if (rxrpc_conn_is_service(conn) && 1181 (chan->last_type == RXRPC_PACKET_TYPE_ACK || 1182 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)) 1183 goto discard_unlock; 1184 1185 /* But otherwise we need to retransmit the final packet from 1186 * data cached in the connection record. 1187 */ 1188 rxrpc_post_packet_to_conn(conn, skb); 1189 goto out_unlock; 1190 } 1191 1192 call = rcu_dereference(chan->call); 1193 1194 if (sp->hdr.callNumber > chan->call_id) { 1195 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) { 1196 rcu_read_unlock(); 1197 goto reject_packet; 1198 } 1199 if (call) 1200 rxrpc_input_implicit_end_call(conn, call); 1201 call = NULL; 1202 } 1203 1204 if (call && sp->hdr.serviceId != call->service_id) 1205 call->service_id = sp->hdr.serviceId; 1206 } else { 1207 skew = 0; 1208 call = NULL; 1209 } 1210 1211 if (!call || atomic_read(&call->usage) == 0) { 1212 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) || 1213 sp->hdr.callNumber == 0 || 1214 sp->hdr.type != RXRPC_PACKET_TYPE_DATA) 1215 goto bad_message_unlock; 1216 if (sp->hdr.seq != 1) 1217 goto discard_unlock; 1218 call = rxrpc_new_incoming_call(local, conn, skb); 1219 if (!call) { 1220 rcu_read_unlock(); 1221 goto reject_packet; 1222 } 1223 rxrpc_send_ping(call, skb, skew); 1224 mutex_unlock(&call->user_mutex); 1225 } 1226 1227 rxrpc_input_call_packet(call, skb, skew); 1228 goto discard_unlock; 1229 1230 discard_unlock: 1231 rcu_read_unlock(); 1232 discard: 1233 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 1234 out: 1235 trace_rxrpc_rx_done(0, 0); 1236 return; 1237 1238 out_unlock: 1239 rcu_read_unlock(); 1240 goto out; 1241 1242 wrong_security: 1243 rcu_read_unlock(); 1244 trace_rxrpc_abort("SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1245 RXKADINCONSISTENCY, EBADMSG); 1246 skb->priority = RXKADINCONSISTENCY; 1247 goto post_abort; 1248 1249 reupgrade: 1250 rcu_read_unlock(); 1251 trace_rxrpc_abort("UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1252 RX_PROTOCOL_ERROR, EBADMSG); 1253 goto protocol_error; 1254 1255 bad_message_unlock: 1256 rcu_read_unlock(); 1257 bad_message: 1258 trace_rxrpc_abort("BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1259 RX_PROTOCOL_ERROR, EBADMSG); 1260 protocol_error: 1261 skb->priority = RX_PROTOCOL_ERROR; 1262 post_abort: 1263 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT; 1264 reject_packet: 1265 trace_rxrpc_rx_done(skb->mark, skb->priority); 1266 rxrpc_reject_packet(local, skb); 1267 _leave(" [badmsg]"); 1268 } 1269