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 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, 0, false, true, 302 rxrpc_propose_ack_client_tx_end); 303 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply); 304 } else { 305 trace_rxrpc_transmit(call, rxrpc_transmit_end); 306 } 307 _leave(" = ok"); 308 return true; 309 310 bad_state: 311 write_unlock(&call->state_lock); 312 kdebug("end_tx %s", rxrpc_call_states[call->state]); 313 rxrpc_proto_abort(abort_why, call, call->tx_top); 314 return false; 315 } 316 317 /* 318 * Begin the reply reception phase of a call. 319 */ 320 static bool rxrpc_receiving_reply(struct rxrpc_call *call) 321 { 322 struct rxrpc_ack_summary summary = { 0 }; 323 rxrpc_seq_t top = READ_ONCE(call->tx_top); 324 325 if (call->ackr_reason) { 326 spin_lock_bh(&call->lock); 327 call->ackr_reason = 0; 328 call->resend_at = call->expire_at; 329 call->ack_at = call->expire_at; 330 spin_unlock_bh(&call->lock); 331 rxrpc_set_timer(call, rxrpc_timer_init_for_reply, 332 ktime_get_real()); 333 } 334 335 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) 336 rxrpc_rotate_tx_window(call, top, &summary); 337 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) { 338 rxrpc_proto_abort("TXL", call, top); 339 return false; 340 } 341 if (!rxrpc_end_tx_phase(call, true, "ETD")) 342 return false; 343 call->tx_phase = false; 344 return true; 345 } 346 347 /* 348 * Scan a jumbo packet to validate its structure and to work out how many 349 * subpackets it contains. 350 * 351 * A jumbo packet is a collection of consecutive packets glued together with 352 * little headers between that indicate how to change the initial header for 353 * each subpacket. 354 * 355 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but 356 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any 357 * size. 358 */ 359 static bool rxrpc_validate_jumbo(struct sk_buff *skb) 360 { 361 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 362 unsigned int offset = sizeof(struct rxrpc_wire_header); 363 unsigned int len = skb->len; 364 int nr_jumbo = 1; 365 u8 flags = sp->hdr.flags; 366 367 do { 368 nr_jumbo++; 369 if (len - offset < RXRPC_JUMBO_SUBPKTLEN) 370 goto protocol_error; 371 if (flags & RXRPC_LAST_PACKET) 372 goto protocol_error; 373 offset += RXRPC_JUMBO_DATALEN; 374 if (skb_copy_bits(skb, offset, &flags, 1) < 0) 375 goto protocol_error; 376 offset += sizeof(struct rxrpc_jumbo_header); 377 } while (flags & RXRPC_JUMBO_PACKET); 378 379 sp->nr_jumbo = nr_jumbo; 380 return true; 381 382 protocol_error: 383 return false; 384 } 385 386 /* 387 * Handle reception of a duplicate packet. 388 * 389 * We have to take care to avoid an attack here whereby we're given a series of 390 * jumbograms, each with a sequence number one before the preceding one and 391 * filled up to maximum UDP size. If they never send us the first packet in 392 * the sequence, they can cause us to have to hold on to around 2MiB of kernel 393 * space until the call times out. 394 * 395 * We limit the space usage by only accepting three duplicate jumbo packets per 396 * call. After that, we tell the other side we're no longer accepting jumbos 397 * (that information is encoded in the ACK packet). 398 */ 399 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq, 400 u8 annotation, bool *_jumbo_bad) 401 { 402 /* Discard normal packets that are duplicates. */ 403 if (annotation == 0) 404 return; 405 406 /* Skip jumbo subpackets that are duplicates. When we've had three or 407 * more partially duplicate jumbo packets, we refuse to take any more 408 * jumbos for this call. 409 */ 410 if (!*_jumbo_bad) { 411 call->nr_jumbo_bad++; 412 *_jumbo_bad = true; 413 } 414 } 415 416 /* 417 * Process a DATA packet, adding the packet to the Rx ring. 418 */ 419 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb, 420 u16 skew) 421 { 422 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 423 enum rxrpc_call_state state; 424 unsigned int offset = sizeof(struct rxrpc_wire_header); 425 unsigned int ix; 426 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0; 427 rxrpc_seq_t seq = sp->hdr.seq, hard_ack; 428 bool immediate_ack = false, jumbo_bad = false, queued; 429 u16 len; 430 u8 ack = 0, flags, annotation = 0; 431 432 _enter("{%u,%u},{%u,%u}", 433 call->rx_hard_ack, call->rx_top, skb->len, seq); 434 435 _proto("Rx DATA %%%u { #%u f=%02x }", 436 sp->hdr.serial, seq, sp->hdr.flags); 437 438 state = READ_ONCE(call->state); 439 if (state >= RXRPC_CALL_COMPLETE) 440 return; 441 442 /* Received data implicitly ACKs all of the request packets we sent 443 * when we're acting as a client. 444 */ 445 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST || 446 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) && 447 !rxrpc_receiving_reply(call)) 448 return; 449 450 call->ackr_prev_seq = seq; 451 452 hard_ack = READ_ONCE(call->rx_hard_ack); 453 if (after(seq, hard_ack + call->rx_winsize)) { 454 ack = RXRPC_ACK_EXCEEDS_WINDOW; 455 ack_serial = serial; 456 goto ack; 457 } 458 459 flags = sp->hdr.flags; 460 if (flags & RXRPC_JUMBO_PACKET) { 461 if (call->nr_jumbo_bad > 3) { 462 ack = RXRPC_ACK_NOSPACE; 463 ack_serial = serial; 464 goto ack; 465 } 466 annotation = 1; 467 } 468 469 next_subpacket: 470 queued = false; 471 ix = seq & RXRPC_RXTX_BUFF_MASK; 472 len = skb->len; 473 if (flags & RXRPC_JUMBO_PACKET) 474 len = RXRPC_JUMBO_DATALEN; 475 476 if (flags & RXRPC_LAST_PACKET) { 477 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) && 478 seq != call->rx_top) 479 return rxrpc_proto_abort("LSN", call, seq); 480 } else { 481 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) && 482 after_eq(seq, call->rx_top)) 483 return rxrpc_proto_abort("LSA", call, seq); 484 } 485 486 trace_rxrpc_rx_data(call, seq, serial, flags, annotation); 487 if (before_eq(seq, hard_ack)) { 488 ack = RXRPC_ACK_DUPLICATE; 489 ack_serial = serial; 490 goto skip; 491 } 492 493 if (flags & RXRPC_REQUEST_ACK && !ack) { 494 ack = RXRPC_ACK_REQUESTED; 495 ack_serial = serial; 496 } 497 498 if (call->rxtx_buffer[ix]) { 499 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad); 500 if (ack != RXRPC_ACK_DUPLICATE) { 501 ack = RXRPC_ACK_DUPLICATE; 502 ack_serial = serial; 503 } 504 immediate_ack = true; 505 goto skip; 506 } 507 508 /* Queue the packet. We use a couple of memory barriers here as need 509 * to make sure that rx_top is perceived to be set after the buffer 510 * pointer and that the buffer pointer is set after the annotation and 511 * the skb data. 512 * 513 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window() 514 * and also rxrpc_fill_out_ack(). 515 */ 516 rxrpc_get_skb(skb, rxrpc_skb_rx_got); 517 call->rxtx_annotations[ix] = annotation; 518 smp_wmb(); 519 call->rxtx_buffer[ix] = skb; 520 if (after(seq, call->rx_top)) { 521 smp_store_release(&call->rx_top, seq); 522 } else if (before(seq, call->rx_top)) { 523 /* Send an immediate ACK if we fill in a hole */ 524 if (!ack) { 525 ack = RXRPC_ACK_DELAY; 526 ack_serial = serial; 527 } 528 immediate_ack = true; 529 } 530 if (flags & RXRPC_LAST_PACKET) { 531 set_bit(RXRPC_CALL_RX_LAST, &call->flags); 532 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq); 533 } else { 534 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq); 535 } 536 queued = true; 537 538 if (after_eq(seq, call->rx_expect_next)) { 539 if (after(seq, call->rx_expect_next)) { 540 _net("OOS %u > %u", seq, call->rx_expect_next); 541 ack = RXRPC_ACK_OUT_OF_SEQUENCE; 542 ack_serial = serial; 543 } 544 call->rx_expect_next = seq + 1; 545 } 546 547 skip: 548 offset += len; 549 if (flags & RXRPC_JUMBO_PACKET) { 550 if (skb_copy_bits(skb, offset, &flags, 1) < 0) 551 return rxrpc_proto_abort("XJF", call, seq); 552 offset += sizeof(struct rxrpc_jumbo_header); 553 seq++; 554 serial++; 555 annotation++; 556 if (flags & RXRPC_JUMBO_PACKET) 557 annotation |= RXRPC_RX_ANNO_JLAST; 558 if (after(seq, hard_ack + call->rx_winsize)) { 559 ack = RXRPC_ACK_EXCEEDS_WINDOW; 560 ack_serial = serial; 561 if (!jumbo_bad) { 562 call->nr_jumbo_bad++; 563 jumbo_bad = true; 564 } 565 goto ack; 566 } 567 568 _proto("Rx DATA Jumbo %%%u", serial); 569 goto next_subpacket; 570 } 571 572 if (queued && flags & RXRPC_LAST_PACKET && !ack) { 573 ack = RXRPC_ACK_DELAY; 574 ack_serial = serial; 575 } 576 577 ack: 578 if (ack) 579 rxrpc_propose_ACK(call, ack, skew, ack_serial, 580 immediate_ack, true, 581 rxrpc_propose_ack_input_data); 582 583 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1) 584 rxrpc_notify_socket(call); 585 _leave(" [queued]"); 586 } 587 588 /* 589 * Process a requested ACK. 590 */ 591 static void rxrpc_input_requested_ack(struct rxrpc_call *call, 592 ktime_t resp_time, 593 rxrpc_serial_t orig_serial, 594 rxrpc_serial_t ack_serial) 595 { 596 struct rxrpc_skb_priv *sp; 597 struct sk_buff *skb; 598 ktime_t sent_at; 599 int ix; 600 601 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) { 602 skb = call->rxtx_buffer[ix]; 603 if (!skb) 604 continue; 605 606 sp = rxrpc_skb(skb); 607 if (sp->hdr.serial != orig_serial) 608 continue; 609 smp_rmb(); 610 sent_at = skb->tstamp; 611 goto found; 612 } 613 return; 614 615 found: 616 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack, 617 orig_serial, ack_serial, sent_at, resp_time); 618 } 619 620 /* 621 * Process a ping response. 622 */ 623 static void rxrpc_input_ping_response(struct rxrpc_call *call, 624 ktime_t resp_time, 625 rxrpc_serial_t orig_serial, 626 rxrpc_serial_t ack_serial) 627 { 628 rxrpc_serial_t ping_serial; 629 ktime_t ping_time; 630 631 ping_time = call->ping_time; 632 smp_rmb(); 633 ping_serial = call->ping_serial; 634 635 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) || 636 before(orig_serial, ping_serial)) 637 return; 638 clear_bit(RXRPC_CALL_PINGING, &call->flags); 639 if (after(orig_serial, ping_serial)) 640 return; 641 642 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response, 643 orig_serial, ack_serial, ping_time, resp_time); 644 } 645 646 /* 647 * Process the extra information that may be appended to an ACK packet 648 */ 649 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb, 650 struct rxrpc_ackinfo *ackinfo) 651 { 652 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 653 struct rxrpc_peer *peer; 654 unsigned int mtu; 655 bool wake = false; 656 u32 rwind = ntohl(ackinfo->rwind); 657 658 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }", 659 sp->hdr.serial, 660 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU), 661 rwind, ntohl(ackinfo->jumbo_max)); 662 663 if (call->tx_winsize != rwind) { 664 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1) 665 rwind = RXRPC_RXTX_BUFF_SIZE - 1; 666 if (rwind > call->tx_winsize) 667 wake = true; 668 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, 669 ntohl(ackinfo->rwind), wake); 670 call->tx_winsize = rwind; 671 } 672 673 if (call->cong_ssthresh > rwind) 674 call->cong_ssthresh = rwind; 675 676 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU)); 677 678 peer = call->peer; 679 if (mtu < peer->maxdata) { 680 spin_lock_bh(&peer->lock); 681 peer->maxdata = mtu; 682 peer->mtu = mtu + peer->hdrsize; 683 spin_unlock_bh(&peer->lock); 684 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata); 685 } 686 687 if (wake) 688 wake_up(&call->waitq); 689 } 690 691 /* 692 * Process individual soft ACKs. 693 * 694 * Each ACK in the array corresponds to one packet and can be either an ACK or 695 * a NAK. If we get find an explicitly NAK'd packet we resend immediately; 696 * packets that lie beyond the end of the ACK list are scheduled for resend by 697 * the timer on the basis that the peer might just not have processed them at 698 * the time the ACK was sent. 699 */ 700 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks, 701 rxrpc_seq_t seq, int nr_acks, 702 struct rxrpc_ack_summary *summary) 703 { 704 int ix; 705 u8 annotation, anno_type; 706 707 for (; nr_acks > 0; nr_acks--, seq++) { 708 ix = seq & RXRPC_RXTX_BUFF_MASK; 709 annotation = call->rxtx_annotations[ix]; 710 anno_type = annotation & RXRPC_TX_ANNO_MASK; 711 annotation &= ~RXRPC_TX_ANNO_MASK; 712 switch (*acks++) { 713 case RXRPC_ACK_TYPE_ACK: 714 summary->nr_acks++; 715 if (anno_type == RXRPC_TX_ANNO_ACK) 716 continue; 717 summary->nr_new_acks++; 718 call->rxtx_annotations[ix] = 719 RXRPC_TX_ANNO_ACK | annotation; 720 break; 721 case RXRPC_ACK_TYPE_NACK: 722 if (!summary->nr_nacks && 723 call->acks_lowest_nak != seq) { 724 call->acks_lowest_nak = seq; 725 summary->new_low_nack = true; 726 } 727 summary->nr_nacks++; 728 if (anno_type == RXRPC_TX_ANNO_NAK) 729 continue; 730 summary->nr_new_nacks++; 731 if (anno_type == RXRPC_TX_ANNO_RETRANS) 732 continue; 733 call->rxtx_annotations[ix] = 734 RXRPC_TX_ANNO_NAK | annotation; 735 break; 736 default: 737 return rxrpc_proto_abort("SFT", call, 0); 738 } 739 } 740 } 741 742 /* 743 * Process an ACK packet. 744 * 745 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet 746 * in the ACK array. Anything before that is hard-ACK'd and may be discarded. 747 * 748 * A hard-ACK means that a packet has been processed and may be discarded; a 749 * soft-ACK means that the packet may be discarded and retransmission 750 * requested. A phase is complete when all packets are hard-ACK'd. 751 */ 752 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb, 753 u16 skew) 754 { 755 struct rxrpc_ack_summary summary = { 0 }; 756 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 757 union { 758 struct rxrpc_ackpacket ack; 759 struct rxrpc_ackinfo info; 760 u8 acks[RXRPC_MAXACKS]; 761 } buf; 762 rxrpc_serial_t acked_serial; 763 rxrpc_seq_t first_soft_ack, hard_ack; 764 int nr_acks, offset, ioffset; 765 766 _enter(""); 767 768 offset = sizeof(struct rxrpc_wire_header); 769 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) { 770 _debug("extraction failure"); 771 return rxrpc_proto_abort("XAK", call, 0); 772 } 773 offset += sizeof(buf.ack); 774 775 acked_serial = ntohl(buf.ack.serial); 776 first_soft_ack = ntohl(buf.ack.firstPacket); 777 hard_ack = first_soft_ack - 1; 778 nr_acks = buf.ack.nAcks; 779 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ? 780 buf.ack.reason : RXRPC_ACK__INVALID); 781 782 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial, 783 first_soft_ack, ntohl(buf.ack.previousPacket), 784 summary.ack_reason, nr_acks); 785 786 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE) 787 rxrpc_input_ping_response(call, skb->tstamp, acked_serial, 788 sp->hdr.serial); 789 if (buf.ack.reason == RXRPC_ACK_REQUESTED) 790 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial, 791 sp->hdr.serial); 792 793 if (buf.ack.reason == RXRPC_ACK_PING) { 794 _proto("Rx ACK %%%u PING Request", sp->hdr.serial); 795 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE, 796 skew, sp->hdr.serial, true, true, 797 rxrpc_propose_ack_respond_to_ping); 798 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) { 799 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED, 800 skew, sp->hdr.serial, true, true, 801 rxrpc_propose_ack_respond_to_ack); 802 } 803 804 ioffset = offset + nr_acks + 3; 805 if (skb->len >= ioffset + sizeof(buf.info)) { 806 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0) 807 return rxrpc_proto_abort("XAI", call, 0); 808 rxrpc_input_ackinfo(call, skb, &buf.info); 809 } 810 811 if (first_soft_ack == 0) 812 return rxrpc_proto_abort("AK0", call, 0); 813 814 /* Ignore ACKs unless we are or have just been transmitting. */ 815 switch (READ_ONCE(call->state)) { 816 case RXRPC_CALL_CLIENT_SEND_REQUEST: 817 case RXRPC_CALL_CLIENT_AWAIT_REPLY: 818 case RXRPC_CALL_SERVER_SEND_REPLY: 819 case RXRPC_CALL_SERVER_AWAIT_ACK: 820 break; 821 default: 822 return; 823 } 824 825 /* Discard any out-of-order or duplicate ACKs. */ 826 if (before_eq(sp->hdr.serial, call->acks_latest)) { 827 _debug("discard ACK %d <= %d", 828 sp->hdr.serial, call->acks_latest); 829 return; 830 } 831 call->acks_latest_ts = skb->tstamp; 832 call->acks_latest = sp->hdr.serial; 833 834 if (before(hard_ack, call->tx_hard_ack) || 835 after(hard_ack, call->tx_top)) 836 return rxrpc_proto_abort("AKW", call, 0); 837 if (nr_acks > call->tx_top - hard_ack) 838 return rxrpc_proto_abort("AKN", call, 0); 839 840 if (after(hard_ack, call->tx_hard_ack)) 841 rxrpc_rotate_tx_window(call, hard_ack, &summary); 842 843 if (nr_acks > 0) { 844 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) 845 return rxrpc_proto_abort("XSA", call, 0); 846 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks, 847 &summary); 848 } 849 850 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) { 851 rxrpc_end_tx_phase(call, false, "ETA"); 852 return; 853 } 854 855 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] & 856 RXRPC_TX_ANNO_LAST && 857 summary.nr_acks == call->tx_top - hard_ack && 858 rxrpc_is_client_call(call)) 859 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial, 860 false, true, 861 rxrpc_propose_ack_ping_for_lost_reply); 862 863 return rxrpc_congestion_management(call, skb, &summary, acked_serial); 864 } 865 866 /* 867 * Process an ACKALL packet. 868 */ 869 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb) 870 { 871 struct rxrpc_ack_summary summary = { 0 }; 872 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 873 874 _proto("Rx ACKALL %%%u", sp->hdr.serial); 875 876 rxrpc_rotate_tx_window(call, call->tx_top, &summary); 877 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) 878 rxrpc_end_tx_phase(call, false, "ETL"); 879 } 880 881 /* 882 * Process an ABORT packet directed at a call. 883 */ 884 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb) 885 { 886 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 887 __be32 wtmp; 888 u32 abort_code = RX_CALL_DEAD; 889 890 _enter(""); 891 892 if (skb->len >= 4 && 893 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header), 894 &wtmp, sizeof(wtmp)) >= 0) 895 abort_code = ntohl(wtmp); 896 897 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code); 898 899 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code); 900 901 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED, 902 abort_code, -ECONNABORTED)) 903 rxrpc_notify_socket(call); 904 } 905 906 /* 907 * Process an incoming call packet. 908 */ 909 static void rxrpc_input_call_packet(struct rxrpc_call *call, 910 struct sk_buff *skb, u16 skew) 911 { 912 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 913 914 _enter("%p,%p", call, skb); 915 916 switch (sp->hdr.type) { 917 case RXRPC_PACKET_TYPE_DATA: 918 rxrpc_input_data(call, skb, skew); 919 break; 920 921 case RXRPC_PACKET_TYPE_ACK: 922 rxrpc_input_ack(call, skb, skew); 923 break; 924 925 case RXRPC_PACKET_TYPE_BUSY: 926 _proto("Rx BUSY %%%u", sp->hdr.serial); 927 928 /* Just ignore BUSY packets from the server; the retry and 929 * lifespan timers will take care of business. BUSY packets 930 * from the client don't make sense. 931 */ 932 break; 933 934 case RXRPC_PACKET_TYPE_ABORT: 935 rxrpc_input_abort(call, skb); 936 break; 937 938 case RXRPC_PACKET_TYPE_ACKALL: 939 rxrpc_input_ackall(call, skb); 940 break; 941 942 default: 943 break; 944 } 945 946 _leave(""); 947 } 948 949 /* 950 * Handle a new call on a channel implicitly completing the preceding call on 951 * that channel. 952 * 953 * TODO: If callNumber > call_id + 1, renegotiate security. 954 */ 955 static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn, 956 struct rxrpc_call *call) 957 { 958 switch (READ_ONCE(call->state)) { 959 case RXRPC_CALL_SERVER_AWAIT_ACK: 960 rxrpc_call_completed(call); 961 break; 962 case RXRPC_CALL_COMPLETE: 963 break; 964 default: 965 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) { 966 set_bit(RXRPC_CALL_EV_ABORT, &call->events); 967 rxrpc_queue_call(call); 968 } 969 break; 970 } 971 972 trace_rxrpc_improper_term(call); 973 __rxrpc_disconnect_call(conn, call); 974 rxrpc_notify_socket(call); 975 } 976 977 /* 978 * post connection-level events to the connection 979 * - this includes challenges, responses, some aborts and call terminal packet 980 * retransmission. 981 */ 982 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn, 983 struct sk_buff *skb) 984 { 985 _enter("%p,%p", conn, skb); 986 987 skb_queue_tail(&conn->rx_queue, skb); 988 rxrpc_queue_conn(conn); 989 } 990 991 /* 992 * post endpoint-level events to the local endpoint 993 * - this includes debug and version messages 994 */ 995 static void rxrpc_post_packet_to_local(struct rxrpc_local *local, 996 struct sk_buff *skb) 997 { 998 _enter("%p,%p", local, skb); 999 1000 skb_queue_tail(&local->event_queue, skb); 1001 rxrpc_queue_local(local); 1002 } 1003 1004 /* 1005 * put a packet up for transport-level abort 1006 */ 1007 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb) 1008 { 1009 CHECK_SLAB_OKAY(&local->usage); 1010 1011 skb_queue_tail(&local->reject_queue, skb); 1012 rxrpc_queue_local(local); 1013 } 1014 1015 /* 1016 * Extract the wire header from a packet and translate the byte order. 1017 */ 1018 static noinline 1019 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb) 1020 { 1021 struct rxrpc_wire_header whdr; 1022 1023 /* dig out the RxRPC connection details */ 1024 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) { 1025 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial, 1026 tracepoint_string("bad_hdr")); 1027 return -EBADMSG; 1028 } 1029 1030 memset(sp, 0, sizeof(*sp)); 1031 sp->hdr.epoch = ntohl(whdr.epoch); 1032 sp->hdr.cid = ntohl(whdr.cid); 1033 sp->hdr.callNumber = ntohl(whdr.callNumber); 1034 sp->hdr.seq = ntohl(whdr.seq); 1035 sp->hdr.serial = ntohl(whdr.serial); 1036 sp->hdr.flags = whdr.flags; 1037 sp->hdr.type = whdr.type; 1038 sp->hdr.userStatus = whdr.userStatus; 1039 sp->hdr.securityIndex = whdr.securityIndex; 1040 sp->hdr._rsvd = ntohs(whdr._rsvd); 1041 sp->hdr.serviceId = ntohs(whdr.serviceId); 1042 return 0; 1043 } 1044 1045 /* 1046 * handle data received on the local endpoint 1047 * - may be called in interrupt context 1048 * 1049 * The socket is locked by the caller and this prevents the socket from being 1050 * shut down and the local endpoint from going away, thus sk_user_data will not 1051 * be cleared until this function returns. 1052 */ 1053 void rxrpc_data_ready(struct sock *udp_sk) 1054 { 1055 struct rxrpc_connection *conn; 1056 struct rxrpc_channel *chan; 1057 struct rxrpc_call *call; 1058 struct rxrpc_skb_priv *sp; 1059 struct rxrpc_local *local = udp_sk->sk_user_data; 1060 struct sk_buff *skb; 1061 unsigned int channel; 1062 int ret, skew; 1063 1064 _enter("%p", udp_sk); 1065 1066 ASSERT(!irqs_disabled()); 1067 1068 skb = skb_recv_udp(udp_sk, 0, 1, &ret); 1069 if (!skb) { 1070 if (ret == -EAGAIN) 1071 return; 1072 _debug("UDP socket error %d", ret); 1073 return; 1074 } 1075 1076 rxrpc_new_skb(skb, rxrpc_skb_rx_received); 1077 1078 _net("recv skb %p", skb); 1079 1080 /* we'll probably need to checksum it (didn't call sock_recvmsg) */ 1081 if (skb_checksum_complete(skb)) { 1082 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 1083 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0); 1084 _leave(" [CSUM failed]"); 1085 return; 1086 } 1087 1088 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0); 1089 1090 /* The UDP protocol already released all skb resources; 1091 * we are free to add our own data there. 1092 */ 1093 sp = rxrpc_skb(skb); 1094 1095 /* dig out the RxRPC connection details */ 1096 if (rxrpc_extract_header(sp, skb) < 0) 1097 goto bad_message; 1098 1099 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) { 1100 static int lose; 1101 if ((lose++ & 7) == 7) { 1102 trace_rxrpc_rx_lose(sp); 1103 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost); 1104 return; 1105 } 1106 } 1107 1108 trace_rxrpc_rx_packet(sp); 1109 1110 _net("Rx RxRPC %s ep=%x call=%x:%x", 1111 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient", 1112 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber); 1113 1114 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES || 1115 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) { 1116 _proto("Rx Bad Packet Type %u", sp->hdr.type); 1117 goto bad_message; 1118 } 1119 1120 switch (sp->hdr.type) { 1121 case RXRPC_PACKET_TYPE_VERSION: 1122 rxrpc_post_packet_to_local(local, skb); 1123 goto out; 1124 1125 case RXRPC_PACKET_TYPE_BUSY: 1126 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED) 1127 goto discard; 1128 1129 case RXRPC_PACKET_TYPE_DATA: 1130 if (sp->hdr.callNumber == 0) 1131 goto bad_message; 1132 if (sp->hdr.flags & RXRPC_JUMBO_PACKET && 1133 !rxrpc_validate_jumbo(skb)) 1134 goto bad_message; 1135 break; 1136 } 1137 1138 rcu_read_lock(); 1139 1140 conn = rxrpc_find_connection_rcu(local, skb); 1141 if (conn) { 1142 if (sp->hdr.securityIndex != conn->security_ix) 1143 goto wrong_security; 1144 1145 if (sp->hdr.serviceId != conn->service_id) { 1146 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) || 1147 conn->service_id != conn->params.service_id) 1148 goto reupgrade; 1149 conn->service_id = sp->hdr.serviceId; 1150 } 1151 1152 if (sp->hdr.callNumber == 0) { 1153 /* Connection-level packet */ 1154 _debug("CONN %p {%d}", conn, conn->debug_id); 1155 rxrpc_post_packet_to_conn(conn, skb); 1156 goto out_unlock; 1157 } 1158 1159 /* Note the serial number skew here */ 1160 skew = (int)sp->hdr.serial - (int)conn->hi_serial; 1161 if (skew >= 0) { 1162 if (skew > 0) 1163 conn->hi_serial = sp->hdr.serial; 1164 } else { 1165 skew = -skew; 1166 skew = min(skew, 65535); 1167 } 1168 1169 /* Call-bound packets are routed by connection channel. */ 1170 channel = sp->hdr.cid & RXRPC_CHANNELMASK; 1171 chan = &conn->channels[channel]; 1172 1173 /* Ignore really old calls */ 1174 if (sp->hdr.callNumber < chan->last_call) 1175 goto discard_unlock; 1176 1177 if (sp->hdr.callNumber == chan->last_call) { 1178 /* For the previous service call, if completed successfully, we 1179 * discard all further packets. 1180 */ 1181 if (rxrpc_conn_is_service(conn) && 1182 (chan->last_type == RXRPC_PACKET_TYPE_ACK || 1183 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)) 1184 goto discard_unlock; 1185 1186 /* But otherwise we need to retransmit the final packet from 1187 * data cached in the connection record. 1188 */ 1189 rxrpc_post_packet_to_conn(conn, skb); 1190 goto out_unlock; 1191 } 1192 1193 call = rcu_dereference(chan->call); 1194 1195 if (sp->hdr.callNumber > chan->call_id) { 1196 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) { 1197 rcu_read_unlock(); 1198 goto reject_packet; 1199 } 1200 if (call) 1201 rxrpc_input_implicit_end_call(conn, call); 1202 call = NULL; 1203 } 1204 1205 if (call && sp->hdr.serviceId != call->service_id) 1206 call->service_id = sp->hdr.serviceId; 1207 } else { 1208 skew = 0; 1209 call = NULL; 1210 } 1211 1212 if (!call || atomic_read(&call->usage) == 0) { 1213 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) || 1214 sp->hdr.callNumber == 0 || 1215 sp->hdr.type != RXRPC_PACKET_TYPE_DATA) 1216 goto bad_message_unlock; 1217 if (sp->hdr.seq != 1) 1218 goto discard_unlock; 1219 call = rxrpc_new_incoming_call(local, conn, skb); 1220 if (!call) { 1221 rcu_read_unlock(); 1222 goto reject_packet; 1223 } 1224 rxrpc_send_ping(call, skb, skew); 1225 mutex_unlock(&call->user_mutex); 1226 } 1227 1228 rxrpc_input_call_packet(call, skb, skew); 1229 goto discard_unlock; 1230 1231 discard_unlock: 1232 rcu_read_unlock(); 1233 discard: 1234 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 1235 out: 1236 trace_rxrpc_rx_done(0, 0); 1237 return; 1238 1239 out_unlock: 1240 rcu_read_unlock(); 1241 goto out; 1242 1243 wrong_security: 1244 rcu_read_unlock(); 1245 trace_rxrpc_abort("SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1246 RXKADINCONSISTENCY, EBADMSG); 1247 skb->priority = RXKADINCONSISTENCY; 1248 goto post_abort; 1249 1250 reupgrade: 1251 rcu_read_unlock(); 1252 trace_rxrpc_abort("UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1253 RX_PROTOCOL_ERROR, EBADMSG); 1254 goto protocol_error; 1255 1256 bad_message_unlock: 1257 rcu_read_unlock(); 1258 bad_message: 1259 trace_rxrpc_abort("BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq, 1260 RX_PROTOCOL_ERROR, EBADMSG); 1261 protocol_error: 1262 skb->priority = RX_PROTOCOL_ERROR; 1263 post_abort: 1264 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT; 1265 reject_packet: 1266 trace_rxrpc_rx_done(skb->mark, skb->priority); 1267 rxrpc_reject_packet(local, skb); 1268 _leave(" [badmsg]"); 1269 } 1270