Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* RxRPC packet reception
3 *
4 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/module.h>
11#include <linux/net.h>
12#include <linux/skbuff.h>
13#include <linux/errqueue.h>
14#include <linux/udp.h>
15#include <linux/in.h>
16#include <linux/in6.h>
17#include <linux/icmp.h>
18#include <linux/gfp.h>
19#include <net/sock.h>
20#include <net/af_rxrpc.h>
21#include <net/ip.h>
22#include <net/udp.h>
23#include <net/net_namespace.h>
24#include "ar-internal.h"
25
26static void rxrpc_proto_abort(const char *why,
27 struct rxrpc_call *call, rxrpc_seq_t seq)
28{
29 if (rxrpc_abort_call(why, call, seq, RX_PROTOCOL_ERROR, -EBADMSG)) {
30 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
31 rxrpc_queue_call(call);
32 }
33}
34
35/*
36 * Do TCP-style congestion management [RFC 5681].
37 */
38static void rxrpc_congestion_management(struct rxrpc_call *call,
39 struct sk_buff *skb,
40 struct rxrpc_ack_summary *summary,
41 rxrpc_serial_t acked_serial)
42{
43 enum rxrpc_congest_change change = rxrpc_cong_no_change;
44 unsigned int cumulative_acks = call->cong_cumul_acks;
45 unsigned int cwnd = call->cong_cwnd;
46 bool resend = false;
47
48 summary->flight_size =
49 (call->tx_top - call->tx_hard_ack) - summary->nr_acks;
50
51 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
52 summary->retrans_timeo = true;
53 call->cong_ssthresh = max_t(unsigned int,
54 summary->flight_size / 2, 2);
55 cwnd = 1;
56 if (cwnd >= call->cong_ssthresh &&
57 call->cong_mode == RXRPC_CALL_SLOW_START) {
58 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
59 call->cong_tstamp = skb->tstamp;
60 cumulative_acks = 0;
61 }
62 }
63
64 cumulative_acks += summary->nr_new_acks;
65 cumulative_acks += summary->nr_rot_new_acks;
66 if (cumulative_acks > 255)
67 cumulative_acks = 255;
68
69 summary->mode = call->cong_mode;
70 summary->cwnd = call->cong_cwnd;
71 summary->ssthresh = call->cong_ssthresh;
72 summary->cumulative_acks = cumulative_acks;
73 summary->dup_acks = call->cong_dup_acks;
74
75 switch (call->cong_mode) {
76 case RXRPC_CALL_SLOW_START:
77 if (summary->nr_nacks > 0)
78 goto packet_loss_detected;
79 if (summary->cumulative_acks > 0)
80 cwnd += 1;
81 if (cwnd >= call->cong_ssthresh) {
82 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
83 call->cong_tstamp = skb->tstamp;
84 }
85 goto out;
86
87 case RXRPC_CALL_CONGEST_AVOIDANCE:
88 if (summary->nr_nacks > 0)
89 goto packet_loss_detected;
90
91 /* We analyse the number of packets that get ACK'd per RTT
92 * period and increase the window if we managed to fill it.
93 */
94 if (call->peer->rtt_count == 0)
95 goto out;
96 if (ktime_before(skb->tstamp,
97 ktime_add_us(call->cong_tstamp,
98 call->peer->srtt_us >> 3)))
99 goto out_no_clear_ca;
100 change = rxrpc_cong_rtt_window_end;
101 call->cong_tstamp = skb->tstamp;
102 if (cumulative_acks >= cwnd)
103 cwnd++;
104 goto out;
105
106 case RXRPC_CALL_PACKET_LOSS:
107 if (summary->nr_nacks == 0)
108 goto resume_normality;
109
110 if (summary->new_low_nack) {
111 change = rxrpc_cong_new_low_nack;
112 call->cong_dup_acks = 1;
113 if (call->cong_extra > 1)
114 call->cong_extra = 1;
115 goto send_extra_data;
116 }
117
118 call->cong_dup_acks++;
119 if (call->cong_dup_acks < 3)
120 goto send_extra_data;
121
122 change = rxrpc_cong_begin_retransmission;
123 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
124 call->cong_ssthresh = max_t(unsigned int,
125 summary->flight_size / 2, 2);
126 cwnd = call->cong_ssthresh + 3;
127 call->cong_extra = 0;
128 call->cong_dup_acks = 0;
129 resend = true;
130 goto out;
131
132 case RXRPC_CALL_FAST_RETRANSMIT:
133 if (!summary->new_low_nack) {
134 if (summary->nr_new_acks == 0)
135 cwnd += 1;
136 call->cong_dup_acks++;
137 if (call->cong_dup_acks == 2) {
138 change = rxrpc_cong_retransmit_again;
139 call->cong_dup_acks = 0;
140 resend = true;
141 }
142 } else {
143 change = rxrpc_cong_progress;
144 cwnd = call->cong_ssthresh;
145 if (summary->nr_nacks == 0)
146 goto resume_normality;
147 }
148 goto out;
149
150 default:
151 BUG();
152 goto out;
153 }
154
155resume_normality:
156 change = rxrpc_cong_cleared_nacks;
157 call->cong_dup_acks = 0;
158 call->cong_extra = 0;
159 call->cong_tstamp = skb->tstamp;
160 if (cwnd < call->cong_ssthresh)
161 call->cong_mode = RXRPC_CALL_SLOW_START;
162 else
163 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
164out:
165 cumulative_acks = 0;
166out_no_clear_ca:
167 if (cwnd >= RXRPC_RXTX_BUFF_SIZE - 1)
168 cwnd = RXRPC_RXTX_BUFF_SIZE - 1;
169 call->cong_cwnd = cwnd;
170 call->cong_cumul_acks = cumulative_acks;
171 trace_rxrpc_congest(call, summary, acked_serial, change);
172 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
173 rxrpc_queue_call(call);
174 return;
175
176packet_loss_detected:
177 change = rxrpc_cong_saw_nack;
178 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
179 call->cong_dup_acks = 0;
180 goto send_extra_data;
181
182send_extra_data:
183 /* Send some previously unsent DATA if we have some to advance the ACK
184 * state.
185 */
186 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
187 RXRPC_TX_ANNO_LAST ||
188 summary->nr_acks != call->tx_top - call->tx_hard_ack) {
189 call->cong_extra++;
190 wake_up(&call->waitq);
191 }
192 goto out_no_clear_ca;
193}
194
195/*
196 * Apply a hard ACK by advancing the Tx window.
197 */
198static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
199 struct rxrpc_ack_summary *summary)
200{
201 struct sk_buff *skb, *list = NULL;
202 bool rot_last = false;
203 int ix;
204 u8 annotation;
205
206 if (call->acks_lowest_nak == call->tx_hard_ack) {
207 call->acks_lowest_nak = to;
208 } else if (before_eq(call->acks_lowest_nak, to)) {
209 summary->new_low_nack = true;
210 call->acks_lowest_nak = to;
211 }
212
213 spin_lock(&call->lock);
214
215 while (before(call->tx_hard_ack, to)) {
216 call->tx_hard_ack++;
217 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
218 skb = call->rxtx_buffer[ix];
219 annotation = call->rxtx_annotations[ix];
220 rxrpc_see_skb(skb, rxrpc_skb_rotated);
221 call->rxtx_buffer[ix] = NULL;
222 call->rxtx_annotations[ix] = 0;
223 skb->next = list;
224 list = skb;
225
226 if (annotation & RXRPC_TX_ANNO_LAST) {
227 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
228 rot_last = true;
229 }
230 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
231 summary->nr_rot_new_acks++;
232 }
233
234 spin_unlock(&call->lock);
235
236 trace_rxrpc_transmit(call, (rot_last ?
237 rxrpc_transmit_rotate_last :
238 rxrpc_transmit_rotate));
239 wake_up(&call->waitq);
240
241 while (list) {
242 skb = list;
243 list = skb->next;
244 skb_mark_not_on_list(skb);
245 rxrpc_free_skb(skb, rxrpc_skb_freed);
246 }
247
248 return rot_last;
249}
250
251/*
252 * End the transmission phase of a call.
253 *
254 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
255 * or a final ACK packet.
256 */
257static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
258 const char *abort_why)
259{
260 unsigned int state;
261
262 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
263
264 write_lock(&call->state_lock);
265
266 state = call->state;
267 switch (state) {
268 case RXRPC_CALL_CLIENT_SEND_REQUEST:
269 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
270 if (reply_begun)
271 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
272 else
273 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
274 break;
275
276 case RXRPC_CALL_SERVER_AWAIT_ACK:
277 __rxrpc_call_completed(call);
278 state = call->state;
279 break;
280
281 default:
282 goto bad_state;
283 }
284
285 write_unlock(&call->state_lock);
286 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
287 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
288 else
289 trace_rxrpc_transmit(call, rxrpc_transmit_end);
290 _leave(" = ok");
291 return true;
292
293bad_state:
294 write_unlock(&call->state_lock);
295 kdebug("end_tx %s", rxrpc_call_states[call->state]);
296 rxrpc_proto_abort(abort_why, call, call->tx_top);
297 return false;
298}
299
300/*
301 * Begin the reply reception phase of a call.
302 */
303static bool rxrpc_receiving_reply(struct rxrpc_call *call)
304{
305 struct rxrpc_ack_summary summary = { 0 };
306 unsigned long now, timo;
307 rxrpc_seq_t top = READ_ONCE(call->tx_top);
308
309 if (call->ackr_reason) {
310 spin_lock_bh(&call->lock);
311 call->ackr_reason = 0;
312 spin_unlock_bh(&call->lock);
313 now = jiffies;
314 timo = now + MAX_JIFFY_OFFSET;
315 WRITE_ONCE(call->resend_at, timo);
316 WRITE_ONCE(call->ack_at, timo);
317 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
318 }
319
320 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
321 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
322 rxrpc_proto_abort("TXL", call, top);
323 return false;
324 }
325 }
326 if (!rxrpc_end_tx_phase(call, true, "ETD"))
327 return false;
328 call->tx_phase = false;
329 return true;
330}
331
332/*
333 * Scan a data packet to validate its structure and to work out how many
334 * subpackets it contains.
335 *
336 * A jumbo packet is a collection of consecutive packets glued together with
337 * little headers between that indicate how to change the initial header for
338 * each subpacket.
339 *
340 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
341 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
342 * size.
343 */
344static bool rxrpc_validate_data(struct sk_buff *skb)
345{
346 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
347 unsigned int offset = sizeof(struct rxrpc_wire_header);
348 unsigned int len = skb->len;
349 u8 flags = sp->hdr.flags;
350
351 for (;;) {
352 if (flags & RXRPC_REQUEST_ACK)
353 __set_bit(sp->nr_subpackets, sp->rx_req_ack);
354 sp->nr_subpackets++;
355
356 if (!(flags & RXRPC_JUMBO_PACKET))
357 break;
358
359 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
360 goto protocol_error;
361 if (flags & RXRPC_LAST_PACKET)
362 goto protocol_error;
363 offset += RXRPC_JUMBO_DATALEN;
364 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
365 goto protocol_error;
366 offset += sizeof(struct rxrpc_jumbo_header);
367 }
368
369 if (flags & RXRPC_LAST_PACKET)
370 sp->rx_flags |= RXRPC_SKB_INCL_LAST;
371 return true;
372
373protocol_error:
374 return false;
375}
376
377/*
378 * Handle reception of a duplicate packet.
379 *
380 * We have to take care to avoid an attack here whereby we're given a series of
381 * jumbograms, each with a sequence number one before the preceding one and
382 * filled up to maximum UDP size. If they never send us the first packet in
383 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
384 * space until the call times out.
385 *
386 * We limit the space usage by only accepting three duplicate jumbo packets per
387 * call. After that, we tell the other side we're no longer accepting jumbos
388 * (that information is encoded in the ACK packet).
389 */
390static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
391 bool is_jumbo, bool *_jumbo_bad)
392{
393 /* Discard normal packets that are duplicates. */
394 if (is_jumbo)
395 return;
396
397 /* Skip jumbo subpackets that are duplicates. When we've had three or
398 * more partially duplicate jumbo packets, we refuse to take any more
399 * jumbos for this call.
400 */
401 if (!*_jumbo_bad) {
402 call->nr_jumbo_bad++;
403 *_jumbo_bad = true;
404 }
405}
406
407/*
408 * Process a DATA packet, adding the packet to the Rx ring. The caller's
409 * packet ref must be passed on or discarded.
410 */
411static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
412{
413 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
414 enum rxrpc_call_state state;
415 unsigned int j, nr_subpackets;
416 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
417 rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
418 bool immediate_ack = false, jumbo_bad = false;
419 u8 ack = 0;
420
421 _enter("{%u,%u},{%u,%u}",
422 call->rx_hard_ack, call->rx_top, skb->len, seq0);
423
424 _proto("Rx DATA %%%u { #%u f=%02x n=%u }",
425 sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
426
427 state = READ_ONCE(call->state);
428 if (state >= RXRPC_CALL_COMPLETE) {
429 rxrpc_free_skb(skb, rxrpc_skb_freed);
430 return;
431 }
432
433 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
434 unsigned long timo = READ_ONCE(call->next_req_timo);
435 unsigned long now, expect_req_by;
436
437 if (timo) {
438 now = jiffies;
439 expect_req_by = now + timo;
440 WRITE_ONCE(call->expect_req_by, expect_req_by);
441 rxrpc_reduce_call_timer(call, expect_req_by, now,
442 rxrpc_timer_set_for_idle);
443 }
444 }
445
446 spin_lock(&call->input_lock);
447
448 /* Received data implicitly ACKs all of the request packets we sent
449 * when we're acting as a client.
450 */
451 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
452 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
453 !rxrpc_receiving_reply(call))
454 goto unlock;
455
456 call->ackr_prev_seq = seq0;
457 hard_ack = READ_ONCE(call->rx_hard_ack);
458
459 nr_subpackets = sp->nr_subpackets;
460 if (nr_subpackets > 1) {
461 if (call->nr_jumbo_bad > 3) {
462 ack = RXRPC_ACK_NOSPACE;
463 ack_serial = serial;
464 goto ack;
465 }
466 }
467
468 for (j = 0; j < nr_subpackets; j++) {
469 rxrpc_serial_t serial = sp->hdr.serial + j;
470 rxrpc_seq_t seq = seq0 + j;
471 unsigned int ix = seq & RXRPC_RXTX_BUFF_MASK;
472 bool terminal = (j == nr_subpackets - 1);
473 bool last = terminal && (sp->rx_flags & RXRPC_SKB_INCL_LAST);
474 u8 flags, annotation = j;
475
476 _proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
477 j, serial, seq, terminal, last);
478
479 if (last) {
480 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
481 seq != call->rx_top) {
482 rxrpc_proto_abort("LSN", call, seq);
483 goto unlock;
484 }
485 } else {
486 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
487 after_eq(seq, call->rx_top)) {
488 rxrpc_proto_abort("LSA", call, seq);
489 goto unlock;
490 }
491 }
492
493 flags = 0;
494 if (last)
495 flags |= RXRPC_LAST_PACKET;
496 if (!terminal)
497 flags |= RXRPC_JUMBO_PACKET;
498 if (test_bit(j, sp->rx_req_ack))
499 flags |= RXRPC_REQUEST_ACK;
500 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
501
502 if (before_eq(seq, hard_ack)) {
503 ack = RXRPC_ACK_DUPLICATE;
504 ack_serial = serial;
505 continue;
506 }
507
508 if (call->rxtx_buffer[ix]) {
509 rxrpc_input_dup_data(call, seq, nr_subpackets > 1,
510 &jumbo_bad);
511 if (ack != RXRPC_ACK_DUPLICATE) {
512 ack = RXRPC_ACK_DUPLICATE;
513 ack_serial = serial;
514 }
515 immediate_ack = true;
516 continue;
517 }
518
519 if (after(seq, hard_ack + call->rx_winsize)) {
520 ack = RXRPC_ACK_EXCEEDS_WINDOW;
521 ack_serial = serial;
522 if (flags & RXRPC_JUMBO_PACKET) {
523 if (!jumbo_bad) {
524 call->nr_jumbo_bad++;
525 jumbo_bad = true;
526 }
527 }
528
529 goto ack;
530 }
531
532 if (flags & RXRPC_REQUEST_ACK && !ack) {
533 ack = RXRPC_ACK_REQUESTED;
534 ack_serial = serial;
535 }
536
537 /* Queue the packet. We use a couple of memory barriers here as need
538 * to make sure that rx_top is perceived to be set after the buffer
539 * pointer and that the buffer pointer is set after the annotation and
540 * the skb data.
541 *
542 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
543 * and also rxrpc_fill_out_ack().
544 */
545 if (!terminal)
546 rxrpc_get_skb(skb, rxrpc_skb_got);
547 call->rxtx_annotations[ix] = annotation;
548 smp_wmb();
549 call->rxtx_buffer[ix] = skb;
550 if (after(seq, call->rx_top)) {
551 smp_store_release(&call->rx_top, seq);
552 } else if (before(seq, call->rx_top)) {
553 /* Send an immediate ACK if we fill in a hole */
554 if (!ack) {
555 ack = RXRPC_ACK_DELAY;
556 ack_serial = serial;
557 }
558 immediate_ack = true;
559 }
560
561 if (terminal) {
562 /* From this point on, we're not allowed to touch the
563 * packet any longer as its ref now belongs to the Rx
564 * ring.
565 */
566 skb = NULL;
567 sp = NULL;
568 }
569
570 if (last) {
571 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
572 if (!ack) {
573 ack = RXRPC_ACK_DELAY;
574 ack_serial = serial;
575 }
576 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
577 } else {
578 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
579 }
580
581 if (after_eq(seq, call->rx_expect_next)) {
582 if (after(seq, call->rx_expect_next)) {
583 _net("OOS %u > %u", seq, call->rx_expect_next);
584 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
585 ack_serial = serial;
586 }
587 call->rx_expect_next = seq + 1;
588 }
589 }
590
591ack:
592 if (ack)
593 rxrpc_propose_ACK(call, ack, ack_serial,
594 immediate_ack, true,
595 rxrpc_propose_ack_input_data);
596 else
597 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
598 false, true,
599 rxrpc_propose_ack_input_data);
600
601 trace_rxrpc_notify_socket(call->debug_id, serial);
602 rxrpc_notify_socket(call);
603
604unlock:
605 spin_unlock(&call->input_lock);
606 rxrpc_free_skb(skb, rxrpc_skb_freed);
607 _leave(" [queued]");
608}
609
610/*
611 * See if there's a cached RTT probe to complete.
612 */
613static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
614 ktime_t resp_time,
615 rxrpc_serial_t acked_serial,
616 rxrpc_serial_t ack_serial,
617 enum rxrpc_rtt_rx_trace type)
618{
619 rxrpc_serial_t orig_serial;
620 unsigned long avail;
621 ktime_t sent_at;
622 bool matched = false;
623 int i;
624
625 avail = READ_ONCE(call->rtt_avail);
626 smp_rmb(); /* Read avail bits before accessing data. */
627
628 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
629 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
630 continue;
631
632 sent_at = call->rtt_sent_at[i];
633 orig_serial = call->rtt_serial[i];
634
635 if (orig_serial == acked_serial) {
636 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
637 smp_mb(); /* Read data before setting avail bit */
638 set_bit(i, &call->rtt_avail);
639 if (type != rxrpc_rtt_rx_cancel)
640 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
641 sent_at, resp_time);
642 else
643 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
644 orig_serial, acked_serial, 0, 0);
645 matched = true;
646 }
647
648 /* If a later serial is being acked, then mark this slot as
649 * being available.
650 */
651 if (after(acked_serial, orig_serial)) {
652 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
653 orig_serial, acked_serial, 0, 0);
654 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
655 smp_wmb();
656 set_bit(i, &call->rtt_avail);
657 }
658 }
659
660 if (!matched)
661 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
662}
663
664/*
665 * Process the response to a ping that we sent to find out if we lost an ACK.
666 *
667 * If we got back a ping response that indicates a lower tx_top than what we
668 * had at the time of the ping transmission, we adjudge all the DATA packets
669 * sent between the response tx_top and the ping-time tx_top to have been lost.
670 */
671static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
672{
673 rxrpc_seq_t top, bottom, seq;
674 bool resend = false;
675
676 spin_lock_bh(&call->lock);
677
678 bottom = call->tx_hard_ack + 1;
679 top = call->acks_lost_top;
680 if (before(bottom, top)) {
681 for (seq = bottom; before_eq(seq, top); seq++) {
682 int ix = seq & RXRPC_RXTX_BUFF_MASK;
683 u8 annotation = call->rxtx_annotations[ix];
684 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
685
686 if (anno_type != RXRPC_TX_ANNO_UNACK)
687 continue;
688 annotation &= ~RXRPC_TX_ANNO_MASK;
689 annotation |= RXRPC_TX_ANNO_RETRANS;
690 call->rxtx_annotations[ix] = annotation;
691 resend = true;
692 }
693 }
694
695 spin_unlock_bh(&call->lock);
696
697 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
698 rxrpc_queue_call(call);
699}
700
701/*
702 * Process a ping response.
703 */
704static void rxrpc_input_ping_response(struct rxrpc_call *call,
705 ktime_t resp_time,
706 rxrpc_serial_t acked_serial,
707 rxrpc_serial_t ack_serial)
708{
709 if (acked_serial == call->acks_lost_ping)
710 rxrpc_input_check_for_lost_ack(call);
711}
712
713/*
714 * Process the extra information that may be appended to an ACK packet
715 */
716static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
717 struct rxrpc_ackinfo *ackinfo)
718{
719 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
720 struct rxrpc_peer *peer;
721 unsigned int mtu;
722 bool wake = false;
723 u32 rwind = ntohl(ackinfo->rwind);
724
725 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
726 sp->hdr.serial,
727 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
728 rwind, ntohl(ackinfo->jumbo_max));
729
730 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
731 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
732 if (call->tx_winsize != rwind) {
733 if (rwind > call->tx_winsize)
734 wake = true;
735 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
736 call->tx_winsize = rwind;
737 }
738
739 if (call->cong_ssthresh > rwind)
740 call->cong_ssthresh = rwind;
741
742 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
743
744 peer = call->peer;
745 if (mtu < peer->maxdata) {
746 spin_lock_bh(&peer->lock);
747 peer->maxdata = mtu;
748 peer->mtu = mtu + peer->hdrsize;
749 spin_unlock_bh(&peer->lock);
750 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
751 }
752
753 if (wake)
754 wake_up(&call->waitq);
755}
756
757/*
758 * Process individual soft ACKs.
759 *
760 * Each ACK in the array corresponds to one packet and can be either an ACK or
761 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
762 * packets that lie beyond the end of the ACK list are scheduled for resend by
763 * the timer on the basis that the peer might just not have processed them at
764 * the time the ACK was sent.
765 */
766static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
767 rxrpc_seq_t seq, int nr_acks,
768 struct rxrpc_ack_summary *summary)
769{
770 int ix;
771 u8 annotation, anno_type;
772
773 for (; nr_acks > 0; nr_acks--, seq++) {
774 ix = seq & RXRPC_RXTX_BUFF_MASK;
775 annotation = call->rxtx_annotations[ix];
776 anno_type = annotation & RXRPC_TX_ANNO_MASK;
777 annotation &= ~RXRPC_TX_ANNO_MASK;
778 switch (*acks++) {
779 case RXRPC_ACK_TYPE_ACK:
780 summary->nr_acks++;
781 if (anno_type == RXRPC_TX_ANNO_ACK)
782 continue;
783 summary->nr_new_acks++;
784 call->rxtx_annotations[ix] =
785 RXRPC_TX_ANNO_ACK | annotation;
786 break;
787 case RXRPC_ACK_TYPE_NACK:
788 if (!summary->nr_nacks &&
789 call->acks_lowest_nak != seq) {
790 call->acks_lowest_nak = seq;
791 summary->new_low_nack = true;
792 }
793 summary->nr_nacks++;
794 if (anno_type == RXRPC_TX_ANNO_NAK)
795 continue;
796 summary->nr_new_nacks++;
797 if (anno_type == RXRPC_TX_ANNO_RETRANS)
798 continue;
799 call->rxtx_annotations[ix] =
800 RXRPC_TX_ANNO_NAK | annotation;
801 break;
802 default:
803 return rxrpc_proto_abort("SFT", call, 0);
804 }
805 }
806}
807
808/*
809 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
810 * with respect to the ack state conveyed by preceding ACKs.
811 */
812static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
813 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
814{
815 rxrpc_seq_t base = READ_ONCE(call->ackr_first_seq);
816
817 if (after(first_pkt, base))
818 return true; /* The window advanced */
819
820 if (before(first_pkt, base))
821 return false; /* firstPacket regressed */
822
823 if (after_eq(prev_pkt, call->ackr_prev_seq))
824 return true; /* previousPacket hasn't regressed. */
825
826 /* Some rx implementations put a serial number in previousPacket. */
827 if (after_eq(prev_pkt, base + call->tx_winsize))
828 return false;
829 return true;
830}
831
832/*
833 * Process an ACK packet.
834 *
835 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
836 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
837 *
838 * A hard-ACK means that a packet has been processed and may be discarded; a
839 * soft-ACK means that the packet may be discarded and retransmission
840 * requested. A phase is complete when all packets are hard-ACK'd.
841 */
842static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
843{
844 struct rxrpc_ack_summary summary = { 0 };
845 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
846 union {
847 struct rxrpc_ackpacket ack;
848 struct rxrpc_ackinfo info;
849 u8 acks[RXRPC_MAXACKS];
850 } buf;
851 rxrpc_serial_t ack_serial, acked_serial;
852 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
853 int nr_acks, offset, ioffset;
854
855 _enter("");
856
857 offset = sizeof(struct rxrpc_wire_header);
858 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
859 _debug("extraction failure");
860 return rxrpc_proto_abort("XAK", call, 0);
861 }
862 offset += sizeof(buf.ack);
863
864 ack_serial = sp->hdr.serial;
865 acked_serial = ntohl(buf.ack.serial);
866 first_soft_ack = ntohl(buf.ack.firstPacket);
867 prev_pkt = ntohl(buf.ack.previousPacket);
868 hard_ack = first_soft_ack - 1;
869 nr_acks = buf.ack.nAcks;
870 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
871 buf.ack.reason : RXRPC_ACK__INVALID);
872
873 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
874 first_soft_ack, prev_pkt,
875 summary.ack_reason, nr_acks);
876
877 switch (buf.ack.reason) {
878 case RXRPC_ACK_PING_RESPONSE:
879 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
880 ack_serial);
881 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
882 rxrpc_rtt_rx_ping_response);
883 break;
884 case RXRPC_ACK_REQUESTED:
885 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
886 rxrpc_rtt_rx_requested_ack);
887 break;
888 default:
889 if (acked_serial != 0)
890 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
891 rxrpc_rtt_rx_cancel);
892 break;
893 }
894
895 if (buf.ack.reason == RXRPC_ACK_PING) {
896 _proto("Rx ACK %%%u PING Request", ack_serial);
897 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
898 ack_serial, true, true,
899 rxrpc_propose_ack_respond_to_ping);
900 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
901 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
902 ack_serial, true, true,
903 rxrpc_propose_ack_respond_to_ack);
904 }
905
906 /* Discard any out-of-order or duplicate ACKs (outside lock). */
907 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
908 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
909 first_soft_ack, call->ackr_first_seq,
910 prev_pkt, call->ackr_prev_seq);
911 return;
912 }
913
914 buf.info.rxMTU = 0;
915 ioffset = offset + nr_acks + 3;
916 if (skb->len >= ioffset + sizeof(buf.info) &&
917 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
918 return rxrpc_proto_abort("XAI", call, 0);
919
920 spin_lock(&call->input_lock);
921
922 /* Discard any out-of-order or duplicate ACKs (inside lock). */
923 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
924 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
925 first_soft_ack, call->ackr_first_seq,
926 prev_pkt, call->ackr_prev_seq);
927 goto out;
928 }
929 call->acks_latest_ts = skb->tstamp;
930
931 call->ackr_first_seq = first_soft_ack;
932 call->ackr_prev_seq = prev_pkt;
933
934 /* Parse rwind and mtu sizes if provided. */
935 if (buf.info.rxMTU)
936 rxrpc_input_ackinfo(call, skb, &buf.info);
937
938 if (first_soft_ack == 0) {
939 rxrpc_proto_abort("AK0", call, 0);
940 goto out;
941 }
942
943 /* Ignore ACKs unless we are or have just been transmitting. */
944 switch (READ_ONCE(call->state)) {
945 case RXRPC_CALL_CLIENT_SEND_REQUEST:
946 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
947 case RXRPC_CALL_SERVER_SEND_REPLY:
948 case RXRPC_CALL_SERVER_AWAIT_ACK:
949 break;
950 default:
951 goto out;
952 }
953
954 if (before(hard_ack, call->tx_hard_ack) ||
955 after(hard_ack, call->tx_top)) {
956 rxrpc_proto_abort("AKW", call, 0);
957 goto out;
958 }
959 if (nr_acks > call->tx_top - hard_ack) {
960 rxrpc_proto_abort("AKN", call, 0);
961 goto out;
962 }
963
964 if (after(hard_ack, call->tx_hard_ack)) {
965 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
966 rxrpc_end_tx_phase(call, false, "ETA");
967 goto out;
968 }
969 }
970
971 if (nr_acks > 0) {
972 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
973 rxrpc_proto_abort("XSA", call, 0);
974 goto out;
975 }
976 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
977 &summary);
978 }
979
980 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
981 RXRPC_TX_ANNO_LAST &&
982 summary.nr_acks == call->tx_top - hard_ack &&
983 rxrpc_is_client_call(call))
984 rxrpc_propose_ACK(call, RXRPC_ACK_PING, ack_serial,
985 false, true,
986 rxrpc_propose_ack_ping_for_lost_reply);
987
988 rxrpc_congestion_management(call, skb, &summary, acked_serial);
989out:
990 spin_unlock(&call->input_lock);
991}
992
993/*
994 * Process an ACKALL packet.
995 */
996static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
997{
998 struct rxrpc_ack_summary summary = { 0 };
999 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1000
1001 _proto("Rx ACKALL %%%u", sp->hdr.serial);
1002
1003 spin_lock(&call->input_lock);
1004
1005 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1006 rxrpc_end_tx_phase(call, false, "ETL");
1007
1008 spin_unlock(&call->input_lock);
1009}
1010
1011/*
1012 * Process an ABORT packet directed at a call.
1013 */
1014static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1015{
1016 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1017 __be32 wtmp;
1018 u32 abort_code = RX_CALL_DEAD;
1019
1020 _enter("");
1021
1022 if (skb->len >= 4 &&
1023 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
1024 &wtmp, sizeof(wtmp)) >= 0)
1025 abort_code = ntohl(wtmp);
1026
1027 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1028
1029 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1030
1031 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1032 abort_code, -ECONNABORTED);
1033}
1034
1035/*
1036 * Process an incoming call packet.
1037 */
1038static void rxrpc_input_call_packet(struct rxrpc_call *call,
1039 struct sk_buff *skb)
1040{
1041 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1042 unsigned long timo;
1043
1044 _enter("%p,%p", call, skb);
1045
1046 timo = READ_ONCE(call->next_rx_timo);
1047 if (timo) {
1048 unsigned long now = jiffies, expect_rx_by;
1049
1050 expect_rx_by = now + timo;
1051 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1052 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1053 rxrpc_timer_set_for_normal);
1054 }
1055
1056 switch (sp->hdr.type) {
1057 case RXRPC_PACKET_TYPE_DATA:
1058 rxrpc_input_data(call, skb);
1059 goto no_free;
1060
1061 case RXRPC_PACKET_TYPE_ACK:
1062 rxrpc_input_ack(call, skb);
1063 break;
1064
1065 case RXRPC_PACKET_TYPE_BUSY:
1066 _proto("Rx BUSY %%%u", sp->hdr.serial);
1067
1068 /* Just ignore BUSY packets from the server; the retry and
1069 * lifespan timers will take care of business. BUSY packets
1070 * from the client don't make sense.
1071 */
1072 break;
1073
1074 case RXRPC_PACKET_TYPE_ABORT:
1075 rxrpc_input_abort(call, skb);
1076 break;
1077
1078 case RXRPC_PACKET_TYPE_ACKALL:
1079 rxrpc_input_ackall(call, skb);
1080 break;
1081
1082 default:
1083 break;
1084 }
1085
1086 rxrpc_free_skb(skb, rxrpc_skb_freed);
1087no_free:
1088 _leave("");
1089}
1090
1091/*
1092 * Handle a new service call on a channel implicitly completing the preceding
1093 * call on that channel. This does not apply to client conns.
1094 *
1095 * TODO: If callNumber > call_id + 1, renegotiate security.
1096 */
1097static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1098 struct rxrpc_connection *conn,
1099 struct rxrpc_call *call)
1100{
1101 switch (READ_ONCE(call->state)) {
1102 case RXRPC_CALL_SERVER_AWAIT_ACK:
1103 rxrpc_call_completed(call);
1104 fallthrough;
1105 case RXRPC_CALL_COMPLETE:
1106 break;
1107 default:
1108 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1109 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1110 rxrpc_queue_call(call);
1111 }
1112 trace_rxrpc_improper_term(call);
1113 break;
1114 }
1115
1116 spin_lock(&rx->incoming_lock);
1117 __rxrpc_disconnect_call(conn, call);
1118 spin_unlock(&rx->incoming_lock);
1119}
1120
1121/*
1122 * post connection-level events to the connection
1123 * - this includes challenges, responses, some aborts and call terminal packet
1124 * retransmission.
1125 */
1126static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1127 struct sk_buff *skb)
1128{
1129 _enter("%p,%p", conn, skb);
1130
1131 skb_queue_tail(&conn->rx_queue, skb);
1132 rxrpc_queue_conn(conn);
1133}
1134
1135/*
1136 * post endpoint-level events to the local endpoint
1137 * - this includes debug and version messages
1138 */
1139static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1140 struct sk_buff *skb)
1141{
1142 _enter("%p,%p", local, skb);
1143
1144 if (rxrpc_get_local_maybe(local)) {
1145 skb_queue_tail(&local->event_queue, skb);
1146 rxrpc_queue_local(local);
1147 } else {
1148 rxrpc_free_skb(skb, rxrpc_skb_freed);
1149 }
1150}
1151
1152/*
1153 * put a packet up for transport-level abort
1154 */
1155static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1156{
1157 CHECK_SLAB_OKAY(&local->usage);
1158
1159 if (rxrpc_get_local_maybe(local)) {
1160 skb_queue_tail(&local->reject_queue, skb);
1161 rxrpc_queue_local(local);
1162 } else {
1163 rxrpc_free_skb(skb, rxrpc_skb_freed);
1164 }
1165}
1166
1167/*
1168 * Extract the wire header from a packet and translate the byte order.
1169 */
1170static noinline
1171int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1172{
1173 struct rxrpc_wire_header whdr;
1174
1175 /* dig out the RxRPC connection details */
1176 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1177 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1178 tracepoint_string("bad_hdr"));
1179 return -EBADMSG;
1180 }
1181
1182 memset(sp, 0, sizeof(*sp));
1183 sp->hdr.epoch = ntohl(whdr.epoch);
1184 sp->hdr.cid = ntohl(whdr.cid);
1185 sp->hdr.callNumber = ntohl(whdr.callNumber);
1186 sp->hdr.seq = ntohl(whdr.seq);
1187 sp->hdr.serial = ntohl(whdr.serial);
1188 sp->hdr.flags = whdr.flags;
1189 sp->hdr.type = whdr.type;
1190 sp->hdr.userStatus = whdr.userStatus;
1191 sp->hdr.securityIndex = whdr.securityIndex;
1192 sp->hdr._rsvd = ntohs(whdr._rsvd);
1193 sp->hdr.serviceId = ntohs(whdr.serviceId);
1194 return 0;
1195}
1196
1197/*
1198 * handle data received on the local endpoint
1199 * - may be called in interrupt context
1200 *
1201 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1202 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1203 * being cleared in the middle of processing this function.
1204 *
1205 * Called with the RCU read lock held from the IP layer via UDP.
1206 */
1207int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1208{
1209 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1210 struct rxrpc_connection *conn;
1211 struct rxrpc_channel *chan;
1212 struct rxrpc_call *call = NULL;
1213 struct rxrpc_skb_priv *sp;
1214 struct rxrpc_peer *peer = NULL;
1215 struct rxrpc_sock *rx = NULL;
1216 unsigned int channel;
1217
1218 _enter("%p", udp_sk);
1219
1220 if (unlikely(!local)) {
1221 kfree_skb(skb);
1222 return 0;
1223 }
1224 if (skb->tstamp == 0)
1225 skb->tstamp = ktime_get_real();
1226
1227 rxrpc_new_skb(skb, rxrpc_skb_received);
1228
1229 skb_pull(skb, sizeof(struct udphdr));
1230
1231 /* The UDP protocol already released all skb resources;
1232 * we are free to add our own data there.
1233 */
1234 sp = rxrpc_skb(skb);
1235
1236 /* dig out the RxRPC connection details */
1237 if (rxrpc_extract_header(sp, skb) < 0)
1238 goto bad_message;
1239
1240 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1241 static int lose;
1242 if ((lose++ & 7) == 7) {
1243 trace_rxrpc_rx_lose(sp);
1244 rxrpc_free_skb(skb, rxrpc_skb_lost);
1245 return 0;
1246 }
1247 }
1248
1249 if (skb->tstamp == 0)
1250 skb->tstamp = ktime_get_real();
1251 trace_rxrpc_rx_packet(sp);
1252
1253 switch (sp->hdr.type) {
1254 case RXRPC_PACKET_TYPE_VERSION:
1255 if (rxrpc_to_client(sp))
1256 goto discard;
1257 rxrpc_post_packet_to_local(local, skb);
1258 goto out;
1259
1260 case RXRPC_PACKET_TYPE_BUSY:
1261 if (rxrpc_to_server(sp))
1262 goto discard;
1263 fallthrough;
1264 case RXRPC_PACKET_TYPE_ACK:
1265 case RXRPC_PACKET_TYPE_ACKALL:
1266 if (sp->hdr.callNumber == 0)
1267 goto bad_message;
1268 fallthrough;
1269 case RXRPC_PACKET_TYPE_ABORT:
1270 break;
1271
1272 case RXRPC_PACKET_TYPE_DATA:
1273 if (sp->hdr.callNumber == 0 ||
1274 sp->hdr.seq == 0)
1275 goto bad_message;
1276 if (!rxrpc_validate_data(skb))
1277 goto bad_message;
1278
1279 /* Unshare the packet so that it can be modified for in-place
1280 * decryption.
1281 */
1282 if (sp->hdr.securityIndex != 0) {
1283 struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1284 if (!nskb) {
1285 rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1286 goto out;
1287 }
1288
1289 if (nskb != skb) {
1290 rxrpc_eaten_skb(skb, rxrpc_skb_received);
1291 skb = nskb;
1292 rxrpc_new_skb(skb, rxrpc_skb_unshared);
1293 sp = rxrpc_skb(skb);
1294 }
1295 }
1296 break;
1297
1298 case RXRPC_PACKET_TYPE_CHALLENGE:
1299 if (rxrpc_to_server(sp))
1300 goto discard;
1301 break;
1302 case RXRPC_PACKET_TYPE_RESPONSE:
1303 if (rxrpc_to_client(sp))
1304 goto discard;
1305 break;
1306
1307 /* Packet types 9-11 should just be ignored. */
1308 case RXRPC_PACKET_TYPE_PARAMS:
1309 case RXRPC_PACKET_TYPE_10:
1310 case RXRPC_PACKET_TYPE_11:
1311 goto discard;
1312
1313 default:
1314 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1315 goto bad_message;
1316 }
1317
1318 if (sp->hdr.serviceId == 0)
1319 goto bad_message;
1320
1321 if (rxrpc_to_server(sp)) {
1322 /* Weed out packets to services we're not offering. Packets
1323 * that would begin a call are explicitly rejected and the rest
1324 * are just discarded.
1325 */
1326 rx = rcu_dereference(local->service);
1327 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1328 sp->hdr.serviceId != rx->second_service)) {
1329 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1330 sp->hdr.seq == 1)
1331 goto unsupported_service;
1332 goto discard;
1333 }
1334 }
1335
1336 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1337 if (conn) {
1338 if (sp->hdr.securityIndex != conn->security_ix)
1339 goto wrong_security;
1340
1341 if (sp->hdr.serviceId != conn->service_id) {
1342 int old_id;
1343
1344 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1345 goto reupgrade;
1346 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1347 sp->hdr.serviceId);
1348
1349 if (old_id != conn->params.service_id &&
1350 old_id != sp->hdr.serviceId)
1351 goto reupgrade;
1352 }
1353
1354 if (sp->hdr.callNumber == 0) {
1355 /* Connection-level packet */
1356 _debug("CONN %p {%d}", conn, conn->debug_id);
1357 rxrpc_post_packet_to_conn(conn, skb);
1358 goto out;
1359 }
1360
1361 if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1362 conn->hi_serial = sp->hdr.serial;
1363
1364 /* Call-bound packets are routed by connection channel. */
1365 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1366 chan = &conn->channels[channel];
1367
1368 /* Ignore really old calls */
1369 if (sp->hdr.callNumber < chan->last_call)
1370 goto discard;
1371
1372 if (sp->hdr.callNumber == chan->last_call) {
1373 if (chan->call ||
1374 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1375 goto discard;
1376
1377 /* For the previous service call, if completed
1378 * successfully, we discard all further packets.
1379 */
1380 if (rxrpc_conn_is_service(conn) &&
1381 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1382 goto discard;
1383
1384 /* But otherwise we need to retransmit the final packet
1385 * from data cached in the connection record.
1386 */
1387 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1388 trace_rxrpc_rx_data(chan->call_debug_id,
1389 sp->hdr.seq,
1390 sp->hdr.serial,
1391 sp->hdr.flags, 0);
1392 rxrpc_post_packet_to_conn(conn, skb);
1393 goto out;
1394 }
1395
1396 call = rcu_dereference(chan->call);
1397
1398 if (sp->hdr.callNumber > chan->call_id) {
1399 if (rxrpc_to_client(sp))
1400 goto reject_packet;
1401 if (call)
1402 rxrpc_input_implicit_end_call(rx, conn, call);
1403 call = NULL;
1404 }
1405
1406 if (call) {
1407 if (sp->hdr.serviceId != call->service_id)
1408 call->service_id = sp->hdr.serviceId;
1409 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1410 call->rx_serial = sp->hdr.serial;
1411 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1412 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1413 }
1414 }
1415
1416 if (!call || atomic_read(&call->usage) == 0) {
1417 if (rxrpc_to_client(sp) ||
1418 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1419 goto bad_message;
1420 if (sp->hdr.seq != 1)
1421 goto discard;
1422 call = rxrpc_new_incoming_call(local, rx, skb);
1423 if (!call)
1424 goto reject_packet;
1425 }
1426
1427 /* Process a call packet; this either discards or passes on the ref
1428 * elsewhere.
1429 */
1430 rxrpc_input_call_packet(call, skb);
1431 goto out;
1432
1433discard:
1434 rxrpc_free_skb(skb, rxrpc_skb_freed);
1435out:
1436 trace_rxrpc_rx_done(0, 0);
1437 return 0;
1438
1439wrong_security:
1440 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1441 RXKADINCONSISTENCY, EBADMSG);
1442 skb->priority = RXKADINCONSISTENCY;
1443 goto post_abort;
1444
1445unsupported_service:
1446 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1447 RX_INVALID_OPERATION, EOPNOTSUPP);
1448 skb->priority = RX_INVALID_OPERATION;
1449 goto post_abort;
1450
1451reupgrade:
1452 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1453 RX_PROTOCOL_ERROR, EBADMSG);
1454 goto protocol_error;
1455
1456bad_message:
1457 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1458 RX_PROTOCOL_ERROR, EBADMSG);
1459protocol_error:
1460 skb->priority = RX_PROTOCOL_ERROR;
1461post_abort:
1462 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1463reject_packet:
1464 trace_rxrpc_rx_done(skb->mark, skb->priority);
1465 rxrpc_reject_packet(local, skb);
1466 _leave(" [badmsg]");
1467 return 0;
1468}
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
30static 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 */
42static 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
159resume_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;
168out:
169 cumulative_acks = 0;
170out_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
180packet_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
186send_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 */
203static 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 */
219static 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 */
273static 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
308bad_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 */
318static bool rxrpc_receiving_reply(struct rxrpc_call *call)
319{
320 struct rxrpc_ack_summary summary = { 0 };
321 unsigned long now, timo;
322 rxrpc_seq_t top = READ_ONCE(call->tx_top);
323
324 if (call->ackr_reason) {
325 spin_lock_bh(&call->lock);
326 call->ackr_reason = 0;
327 spin_unlock_bh(&call->lock);
328 now = jiffies;
329 timo = now + MAX_JIFFY_OFFSET;
330 WRITE_ONCE(call->resend_at, timo);
331 WRITE_ONCE(call->ack_at, timo);
332 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
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 */
359static 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
382protocol_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 */
399static 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 */
419static 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 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
443 unsigned long timo = READ_ONCE(call->next_req_timo);
444 unsigned long now, expect_req_by;
445
446 if (timo) {
447 now = jiffies;
448 expect_req_by = now + timo;
449 WRITE_ONCE(call->expect_req_by, expect_req_by);
450 rxrpc_reduce_call_timer(call, expect_req_by, now,
451 rxrpc_timer_set_for_idle);
452 }
453 }
454
455 /* Received data implicitly ACKs all of the request packets we sent
456 * when we're acting as a client.
457 */
458 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
459 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
460 !rxrpc_receiving_reply(call))
461 return;
462
463 call->ackr_prev_seq = seq;
464
465 hard_ack = READ_ONCE(call->rx_hard_ack);
466 if (after(seq, hard_ack + call->rx_winsize)) {
467 ack = RXRPC_ACK_EXCEEDS_WINDOW;
468 ack_serial = serial;
469 goto ack;
470 }
471
472 flags = sp->hdr.flags;
473 if (flags & RXRPC_JUMBO_PACKET) {
474 if (call->nr_jumbo_bad > 3) {
475 ack = RXRPC_ACK_NOSPACE;
476 ack_serial = serial;
477 goto ack;
478 }
479 annotation = 1;
480 }
481
482next_subpacket:
483 queued = false;
484 ix = seq & RXRPC_RXTX_BUFF_MASK;
485 len = skb->len;
486 if (flags & RXRPC_JUMBO_PACKET)
487 len = RXRPC_JUMBO_DATALEN;
488
489 if (flags & RXRPC_LAST_PACKET) {
490 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
491 seq != call->rx_top)
492 return rxrpc_proto_abort("LSN", call, seq);
493 } else {
494 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
495 after_eq(seq, call->rx_top))
496 return rxrpc_proto_abort("LSA", call, seq);
497 }
498
499 trace_rxrpc_rx_data(call, seq, serial, flags, annotation);
500 if (before_eq(seq, hard_ack)) {
501 ack = RXRPC_ACK_DUPLICATE;
502 ack_serial = serial;
503 goto skip;
504 }
505
506 if (flags & RXRPC_REQUEST_ACK && !ack) {
507 ack = RXRPC_ACK_REQUESTED;
508 ack_serial = serial;
509 }
510
511 if (call->rxtx_buffer[ix]) {
512 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
513 if (ack != RXRPC_ACK_DUPLICATE) {
514 ack = RXRPC_ACK_DUPLICATE;
515 ack_serial = serial;
516 }
517 immediate_ack = true;
518 goto skip;
519 }
520
521 /* Queue the packet. We use a couple of memory barriers here as need
522 * to make sure that rx_top is perceived to be set after the buffer
523 * pointer and that the buffer pointer is set after the annotation and
524 * the skb data.
525 *
526 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
527 * and also rxrpc_fill_out_ack().
528 */
529 rxrpc_get_skb(skb, rxrpc_skb_rx_got);
530 call->rxtx_annotations[ix] = annotation;
531 smp_wmb();
532 call->rxtx_buffer[ix] = skb;
533 if (after(seq, call->rx_top)) {
534 smp_store_release(&call->rx_top, seq);
535 } else if (before(seq, call->rx_top)) {
536 /* Send an immediate ACK if we fill in a hole */
537 if (!ack) {
538 ack = RXRPC_ACK_DELAY;
539 ack_serial = serial;
540 }
541 immediate_ack = true;
542 }
543 if (flags & RXRPC_LAST_PACKET) {
544 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
545 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
546 } else {
547 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
548 }
549 queued = true;
550
551 if (after_eq(seq, call->rx_expect_next)) {
552 if (after(seq, call->rx_expect_next)) {
553 _net("OOS %u > %u", seq, call->rx_expect_next);
554 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
555 ack_serial = serial;
556 }
557 call->rx_expect_next = seq + 1;
558 }
559
560skip:
561 offset += len;
562 if (flags & RXRPC_JUMBO_PACKET) {
563 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
564 return rxrpc_proto_abort("XJF", call, seq);
565 offset += sizeof(struct rxrpc_jumbo_header);
566 seq++;
567 serial++;
568 annotation++;
569 if (flags & RXRPC_JUMBO_PACKET)
570 annotation |= RXRPC_RX_ANNO_JLAST;
571 if (after(seq, hard_ack + call->rx_winsize)) {
572 ack = RXRPC_ACK_EXCEEDS_WINDOW;
573 ack_serial = serial;
574 if (!jumbo_bad) {
575 call->nr_jumbo_bad++;
576 jumbo_bad = true;
577 }
578 goto ack;
579 }
580
581 _proto("Rx DATA Jumbo %%%u", serial);
582 goto next_subpacket;
583 }
584
585 if (queued && flags & RXRPC_LAST_PACKET && !ack) {
586 ack = RXRPC_ACK_DELAY;
587 ack_serial = serial;
588 }
589
590ack:
591 if (ack)
592 rxrpc_propose_ACK(call, ack, skew, ack_serial,
593 immediate_ack, true,
594 rxrpc_propose_ack_input_data);
595
596 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1)
597 rxrpc_notify_socket(call);
598 _leave(" [queued]");
599}
600
601/*
602 * Process a requested ACK.
603 */
604static void rxrpc_input_requested_ack(struct rxrpc_call *call,
605 ktime_t resp_time,
606 rxrpc_serial_t orig_serial,
607 rxrpc_serial_t ack_serial)
608{
609 struct rxrpc_skb_priv *sp;
610 struct sk_buff *skb;
611 ktime_t sent_at;
612 int ix;
613
614 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
615 skb = call->rxtx_buffer[ix];
616 if (!skb)
617 continue;
618
619 sp = rxrpc_skb(skb);
620 if (sp->hdr.serial != orig_serial)
621 continue;
622 smp_rmb();
623 sent_at = skb->tstamp;
624 goto found;
625 }
626 return;
627
628found:
629 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
630 orig_serial, ack_serial, sent_at, resp_time);
631}
632
633/*
634 * Process the response to a ping that we sent to find out if we lost an ACK.
635 *
636 * If we got back a ping response that indicates a lower tx_top than what we
637 * had at the time of the ping transmission, we adjudge all the DATA packets
638 * sent between the response tx_top and the ping-time tx_top to have been lost.
639 */
640static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
641{
642 rxrpc_seq_t top, bottom, seq;
643 bool resend = false;
644
645 spin_lock_bh(&call->lock);
646
647 bottom = call->tx_hard_ack + 1;
648 top = call->acks_lost_top;
649 if (before(bottom, top)) {
650 for (seq = bottom; before_eq(seq, top); seq++) {
651 int ix = seq & RXRPC_RXTX_BUFF_MASK;
652 u8 annotation = call->rxtx_annotations[ix];
653 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
654
655 if (anno_type != RXRPC_TX_ANNO_UNACK)
656 continue;
657 annotation &= ~RXRPC_TX_ANNO_MASK;
658 annotation |= RXRPC_TX_ANNO_RETRANS;
659 call->rxtx_annotations[ix] = annotation;
660 resend = true;
661 }
662 }
663
664 spin_unlock_bh(&call->lock);
665
666 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
667 rxrpc_queue_call(call);
668}
669
670/*
671 * Process a ping response.
672 */
673static void rxrpc_input_ping_response(struct rxrpc_call *call,
674 ktime_t resp_time,
675 rxrpc_serial_t orig_serial,
676 rxrpc_serial_t ack_serial)
677{
678 rxrpc_serial_t ping_serial;
679 ktime_t ping_time;
680
681 ping_time = call->ping_time;
682 smp_rmb();
683 ping_serial = call->ping_serial;
684
685 if (orig_serial == call->acks_lost_ping)
686 rxrpc_input_check_for_lost_ack(call);
687
688 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) ||
689 before(orig_serial, ping_serial))
690 return;
691 clear_bit(RXRPC_CALL_PINGING, &call->flags);
692 if (after(orig_serial, ping_serial))
693 return;
694
695 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
696 orig_serial, ack_serial, ping_time, resp_time);
697}
698
699/*
700 * Process the extra information that may be appended to an ACK packet
701 */
702static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
703 struct rxrpc_ackinfo *ackinfo)
704{
705 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
706 struct rxrpc_peer *peer;
707 unsigned int mtu;
708 bool wake = false;
709 u32 rwind = ntohl(ackinfo->rwind);
710
711 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
712 sp->hdr.serial,
713 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
714 rwind, ntohl(ackinfo->jumbo_max));
715
716 if (call->tx_winsize != rwind) {
717 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
718 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
719 if (rwind > call->tx_winsize)
720 wake = true;
721 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
722 ntohl(ackinfo->rwind), wake);
723 call->tx_winsize = rwind;
724 }
725
726 if (call->cong_ssthresh > rwind)
727 call->cong_ssthresh = rwind;
728
729 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
730
731 peer = call->peer;
732 if (mtu < peer->maxdata) {
733 spin_lock_bh(&peer->lock);
734 peer->maxdata = mtu;
735 peer->mtu = mtu + peer->hdrsize;
736 spin_unlock_bh(&peer->lock);
737 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
738 }
739
740 if (wake)
741 wake_up(&call->waitq);
742}
743
744/*
745 * Process individual soft ACKs.
746 *
747 * Each ACK in the array corresponds to one packet and can be either an ACK or
748 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
749 * packets that lie beyond the end of the ACK list are scheduled for resend by
750 * the timer on the basis that the peer might just not have processed them at
751 * the time the ACK was sent.
752 */
753static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
754 rxrpc_seq_t seq, int nr_acks,
755 struct rxrpc_ack_summary *summary)
756{
757 int ix;
758 u8 annotation, anno_type;
759
760 for (; nr_acks > 0; nr_acks--, seq++) {
761 ix = seq & RXRPC_RXTX_BUFF_MASK;
762 annotation = call->rxtx_annotations[ix];
763 anno_type = annotation & RXRPC_TX_ANNO_MASK;
764 annotation &= ~RXRPC_TX_ANNO_MASK;
765 switch (*acks++) {
766 case RXRPC_ACK_TYPE_ACK:
767 summary->nr_acks++;
768 if (anno_type == RXRPC_TX_ANNO_ACK)
769 continue;
770 summary->nr_new_acks++;
771 call->rxtx_annotations[ix] =
772 RXRPC_TX_ANNO_ACK | annotation;
773 break;
774 case RXRPC_ACK_TYPE_NACK:
775 if (!summary->nr_nacks &&
776 call->acks_lowest_nak != seq) {
777 call->acks_lowest_nak = seq;
778 summary->new_low_nack = true;
779 }
780 summary->nr_nacks++;
781 if (anno_type == RXRPC_TX_ANNO_NAK)
782 continue;
783 summary->nr_new_nacks++;
784 if (anno_type == RXRPC_TX_ANNO_RETRANS)
785 continue;
786 call->rxtx_annotations[ix] =
787 RXRPC_TX_ANNO_NAK | annotation;
788 break;
789 default:
790 return rxrpc_proto_abort("SFT", call, 0);
791 }
792 }
793}
794
795/*
796 * Process an ACK packet.
797 *
798 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
799 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
800 *
801 * A hard-ACK means that a packet has been processed and may be discarded; a
802 * soft-ACK means that the packet may be discarded and retransmission
803 * requested. A phase is complete when all packets are hard-ACK'd.
804 */
805static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
806 u16 skew)
807{
808 struct rxrpc_ack_summary summary = { 0 };
809 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
810 union {
811 struct rxrpc_ackpacket ack;
812 struct rxrpc_ackinfo info;
813 u8 acks[RXRPC_MAXACKS];
814 } buf;
815 rxrpc_serial_t acked_serial;
816 rxrpc_seq_t first_soft_ack, hard_ack;
817 int nr_acks, offset, ioffset;
818
819 _enter("");
820
821 offset = sizeof(struct rxrpc_wire_header);
822 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
823 _debug("extraction failure");
824 return rxrpc_proto_abort("XAK", call, 0);
825 }
826 offset += sizeof(buf.ack);
827
828 acked_serial = ntohl(buf.ack.serial);
829 first_soft_ack = ntohl(buf.ack.firstPacket);
830 hard_ack = first_soft_ack - 1;
831 nr_acks = buf.ack.nAcks;
832 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
833 buf.ack.reason : RXRPC_ACK__INVALID);
834
835 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
836 first_soft_ack, ntohl(buf.ack.previousPacket),
837 summary.ack_reason, nr_acks);
838
839 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
840 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
841 sp->hdr.serial);
842 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
843 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
844 sp->hdr.serial);
845
846 if (buf.ack.reason == RXRPC_ACK_PING) {
847 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
848 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
849 skew, sp->hdr.serial, true, true,
850 rxrpc_propose_ack_respond_to_ping);
851 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
852 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
853 skew, sp->hdr.serial, true, true,
854 rxrpc_propose_ack_respond_to_ack);
855 }
856
857 ioffset = offset + nr_acks + 3;
858 if (skb->len >= ioffset + sizeof(buf.info)) {
859 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
860 return rxrpc_proto_abort("XAI", call, 0);
861 rxrpc_input_ackinfo(call, skb, &buf.info);
862 }
863
864 if (first_soft_ack == 0)
865 return rxrpc_proto_abort("AK0", call, 0);
866
867 /* Ignore ACKs unless we are or have just been transmitting. */
868 switch (READ_ONCE(call->state)) {
869 case RXRPC_CALL_CLIENT_SEND_REQUEST:
870 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
871 case RXRPC_CALL_SERVER_SEND_REPLY:
872 case RXRPC_CALL_SERVER_AWAIT_ACK:
873 break;
874 default:
875 return;
876 }
877
878 /* Discard any out-of-order or duplicate ACKs. */
879 if (before_eq(sp->hdr.serial, call->acks_latest)) {
880 _debug("discard ACK %d <= %d",
881 sp->hdr.serial, call->acks_latest);
882 return;
883 }
884 call->acks_latest_ts = skb->tstamp;
885 call->acks_latest = sp->hdr.serial;
886
887 if (before(hard_ack, call->tx_hard_ack) ||
888 after(hard_ack, call->tx_top))
889 return rxrpc_proto_abort("AKW", call, 0);
890 if (nr_acks > call->tx_top - hard_ack)
891 return rxrpc_proto_abort("AKN", call, 0);
892
893 if (after(hard_ack, call->tx_hard_ack))
894 rxrpc_rotate_tx_window(call, hard_ack, &summary);
895
896 if (nr_acks > 0) {
897 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0)
898 return rxrpc_proto_abort("XSA", call, 0);
899 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
900 &summary);
901 }
902
903 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
904 rxrpc_end_tx_phase(call, false, "ETA");
905 return;
906 }
907
908 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
909 RXRPC_TX_ANNO_LAST &&
910 summary.nr_acks == call->tx_top - hard_ack &&
911 rxrpc_is_client_call(call))
912 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
913 false, true,
914 rxrpc_propose_ack_ping_for_lost_reply);
915
916 return rxrpc_congestion_management(call, skb, &summary, acked_serial);
917}
918
919/*
920 * Process an ACKALL packet.
921 */
922static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
923{
924 struct rxrpc_ack_summary summary = { 0 };
925 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
926
927 _proto("Rx ACKALL %%%u", sp->hdr.serial);
928
929 rxrpc_rotate_tx_window(call, call->tx_top, &summary);
930 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags))
931 rxrpc_end_tx_phase(call, false, "ETL");
932}
933
934/*
935 * Process an ABORT packet directed at a call.
936 */
937static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
938{
939 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
940 __be32 wtmp;
941 u32 abort_code = RX_CALL_DEAD;
942
943 _enter("");
944
945 if (skb->len >= 4 &&
946 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
947 &wtmp, sizeof(wtmp)) >= 0)
948 abort_code = ntohl(wtmp);
949
950 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
951
952 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
953
954 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
955 abort_code, -ECONNABORTED))
956 rxrpc_notify_socket(call);
957}
958
959/*
960 * Process an incoming call packet.
961 */
962static void rxrpc_input_call_packet(struct rxrpc_call *call,
963 struct sk_buff *skb, u16 skew)
964{
965 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
966 unsigned long timo;
967
968 _enter("%p,%p", call, skb);
969
970 timo = READ_ONCE(call->next_rx_timo);
971 if (timo) {
972 unsigned long now = jiffies, expect_rx_by;
973
974 expect_rx_by = now + timo;
975 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
976 rxrpc_reduce_call_timer(call, expect_rx_by, now,
977 rxrpc_timer_set_for_normal);
978 }
979
980 switch (sp->hdr.type) {
981 case RXRPC_PACKET_TYPE_DATA:
982 rxrpc_input_data(call, skb, skew);
983 break;
984
985 case RXRPC_PACKET_TYPE_ACK:
986 rxrpc_input_ack(call, skb, skew);
987 break;
988
989 case RXRPC_PACKET_TYPE_BUSY:
990 _proto("Rx BUSY %%%u", sp->hdr.serial);
991
992 /* Just ignore BUSY packets from the server; the retry and
993 * lifespan timers will take care of business. BUSY packets
994 * from the client don't make sense.
995 */
996 break;
997
998 case RXRPC_PACKET_TYPE_ABORT:
999 rxrpc_input_abort(call, skb);
1000 break;
1001
1002 case RXRPC_PACKET_TYPE_ACKALL:
1003 rxrpc_input_ackall(call, skb);
1004 break;
1005
1006 default:
1007 break;
1008 }
1009
1010 _leave("");
1011}
1012
1013/*
1014 * Handle a new call on a channel implicitly completing the preceding call on
1015 * that channel.
1016 *
1017 * TODO: If callNumber > call_id + 1, renegotiate security.
1018 */
1019static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn,
1020 struct rxrpc_call *call)
1021{
1022 switch (READ_ONCE(call->state)) {
1023 case RXRPC_CALL_SERVER_AWAIT_ACK:
1024 rxrpc_call_completed(call);
1025 break;
1026 case RXRPC_CALL_COMPLETE:
1027 break;
1028 default:
1029 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1030 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1031 rxrpc_queue_call(call);
1032 }
1033 break;
1034 }
1035
1036 trace_rxrpc_improper_term(call);
1037 __rxrpc_disconnect_call(conn, call);
1038 rxrpc_notify_socket(call);
1039}
1040
1041/*
1042 * post connection-level events to the connection
1043 * - this includes challenges, responses, some aborts and call terminal packet
1044 * retransmission.
1045 */
1046static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1047 struct sk_buff *skb)
1048{
1049 _enter("%p,%p", conn, skb);
1050
1051 skb_queue_tail(&conn->rx_queue, skb);
1052 rxrpc_queue_conn(conn);
1053}
1054
1055/*
1056 * post endpoint-level events to the local endpoint
1057 * - this includes debug and version messages
1058 */
1059static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1060 struct sk_buff *skb)
1061{
1062 _enter("%p,%p", local, skb);
1063
1064 skb_queue_tail(&local->event_queue, skb);
1065 rxrpc_queue_local(local);
1066}
1067
1068/*
1069 * put a packet up for transport-level abort
1070 */
1071static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1072{
1073 CHECK_SLAB_OKAY(&local->usage);
1074
1075 skb_queue_tail(&local->reject_queue, skb);
1076 rxrpc_queue_local(local);
1077}
1078
1079/*
1080 * Extract the wire header from a packet and translate the byte order.
1081 */
1082static noinline
1083int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1084{
1085 struct rxrpc_wire_header whdr;
1086
1087 /* dig out the RxRPC connection details */
1088 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1089 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1090 tracepoint_string("bad_hdr"));
1091 return -EBADMSG;
1092 }
1093
1094 memset(sp, 0, sizeof(*sp));
1095 sp->hdr.epoch = ntohl(whdr.epoch);
1096 sp->hdr.cid = ntohl(whdr.cid);
1097 sp->hdr.callNumber = ntohl(whdr.callNumber);
1098 sp->hdr.seq = ntohl(whdr.seq);
1099 sp->hdr.serial = ntohl(whdr.serial);
1100 sp->hdr.flags = whdr.flags;
1101 sp->hdr.type = whdr.type;
1102 sp->hdr.userStatus = whdr.userStatus;
1103 sp->hdr.securityIndex = whdr.securityIndex;
1104 sp->hdr._rsvd = ntohs(whdr._rsvd);
1105 sp->hdr.serviceId = ntohs(whdr.serviceId);
1106 return 0;
1107}
1108
1109/*
1110 * handle data received on the local endpoint
1111 * - may be called in interrupt context
1112 *
1113 * The socket is locked by the caller and this prevents the socket from being
1114 * shut down and the local endpoint from going away, thus sk_user_data will not
1115 * be cleared until this function returns.
1116 */
1117void rxrpc_data_ready(struct sock *udp_sk)
1118{
1119 struct rxrpc_connection *conn;
1120 struct rxrpc_channel *chan;
1121 struct rxrpc_call *call;
1122 struct rxrpc_skb_priv *sp;
1123 struct rxrpc_local *local = udp_sk->sk_user_data;
1124 struct sk_buff *skb;
1125 unsigned int channel;
1126 int ret, skew;
1127
1128 _enter("%p", udp_sk);
1129
1130 ASSERT(!irqs_disabled());
1131
1132 skb = skb_recv_udp(udp_sk, 0, 1, &ret);
1133 if (!skb) {
1134 if (ret == -EAGAIN)
1135 return;
1136 _debug("UDP socket error %d", ret);
1137 return;
1138 }
1139
1140 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
1141
1142 _net("recv skb %p", skb);
1143
1144 /* we'll probably need to checksum it (didn't call sock_recvmsg) */
1145 if (skb_checksum_complete(skb)) {
1146 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1147 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
1148 _leave(" [CSUM failed]");
1149 return;
1150 }
1151
1152 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0);
1153
1154 /* The UDP protocol already released all skb resources;
1155 * we are free to add our own data there.
1156 */
1157 sp = rxrpc_skb(skb);
1158
1159 /* dig out the RxRPC connection details */
1160 if (rxrpc_extract_header(sp, skb) < 0)
1161 goto bad_message;
1162
1163 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1164 static int lose;
1165 if ((lose++ & 7) == 7) {
1166 trace_rxrpc_rx_lose(sp);
1167 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost);
1168 return;
1169 }
1170 }
1171
1172 trace_rxrpc_rx_packet(sp);
1173
1174 _net("Rx RxRPC %s ep=%x call=%x:%x",
1175 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient",
1176 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber);
1177
1178 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES ||
1179 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) {
1180 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1181 goto bad_message;
1182 }
1183
1184 switch (sp->hdr.type) {
1185 case RXRPC_PACKET_TYPE_VERSION:
1186 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED))
1187 goto discard;
1188 rxrpc_post_packet_to_local(local, skb);
1189 goto out;
1190
1191 case RXRPC_PACKET_TYPE_BUSY:
1192 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED)
1193 goto discard;
1194 /* Fall through */
1195
1196 case RXRPC_PACKET_TYPE_DATA:
1197 if (sp->hdr.callNumber == 0)
1198 goto bad_message;
1199 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
1200 !rxrpc_validate_jumbo(skb))
1201 goto bad_message;
1202 break;
1203
1204 /* Packet types 9-11 should just be ignored. */
1205 case RXRPC_PACKET_TYPE_PARAMS:
1206 case RXRPC_PACKET_TYPE_10:
1207 case RXRPC_PACKET_TYPE_11:
1208 goto discard;
1209 }
1210
1211 rcu_read_lock();
1212
1213 conn = rxrpc_find_connection_rcu(local, skb);
1214 if (conn) {
1215 if (sp->hdr.securityIndex != conn->security_ix)
1216 goto wrong_security;
1217
1218 if (sp->hdr.serviceId != conn->service_id) {
1219 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) ||
1220 conn->service_id != conn->params.service_id)
1221 goto reupgrade;
1222 conn->service_id = sp->hdr.serviceId;
1223 }
1224
1225 if (sp->hdr.callNumber == 0) {
1226 /* Connection-level packet */
1227 _debug("CONN %p {%d}", conn, conn->debug_id);
1228 rxrpc_post_packet_to_conn(conn, skb);
1229 goto out_unlock;
1230 }
1231
1232 /* Note the serial number skew here */
1233 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
1234 if (skew >= 0) {
1235 if (skew > 0)
1236 conn->hi_serial = sp->hdr.serial;
1237 } else {
1238 skew = -skew;
1239 skew = min(skew, 65535);
1240 }
1241
1242 /* Call-bound packets are routed by connection channel. */
1243 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1244 chan = &conn->channels[channel];
1245
1246 /* Ignore really old calls */
1247 if (sp->hdr.callNumber < chan->last_call)
1248 goto discard_unlock;
1249
1250 if (sp->hdr.callNumber == chan->last_call) {
1251 if (chan->call ||
1252 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1253 goto discard_unlock;
1254
1255 /* For the previous service call, if completed
1256 * successfully, we discard all further packets.
1257 */
1258 if (rxrpc_conn_is_service(conn) &&
1259 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1260 goto discard_unlock;
1261
1262 /* But otherwise we need to retransmit the final packet
1263 * from data cached in the connection record.
1264 */
1265 rxrpc_post_packet_to_conn(conn, skb);
1266 goto out_unlock;
1267 }
1268
1269 call = rcu_dereference(chan->call);
1270
1271 if (sp->hdr.callNumber > chan->call_id) {
1272 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) {
1273 rcu_read_unlock();
1274 goto reject_packet;
1275 }
1276 if (call)
1277 rxrpc_input_implicit_end_call(conn, call);
1278 call = NULL;
1279 }
1280
1281 if (call && sp->hdr.serviceId != call->service_id)
1282 call->service_id = sp->hdr.serviceId;
1283 } else {
1284 skew = 0;
1285 call = NULL;
1286 }
1287
1288 if (!call || atomic_read(&call->usage) == 0) {
1289 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) ||
1290 sp->hdr.callNumber == 0 ||
1291 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1292 goto bad_message_unlock;
1293 if (sp->hdr.seq != 1)
1294 goto discard_unlock;
1295 call = rxrpc_new_incoming_call(local, conn, skb);
1296 if (!call) {
1297 rcu_read_unlock();
1298 goto reject_packet;
1299 }
1300 rxrpc_send_ping(call, skb, skew);
1301 mutex_unlock(&call->user_mutex);
1302 }
1303
1304 rxrpc_input_call_packet(call, skb, skew);
1305 goto discard_unlock;
1306
1307discard_unlock:
1308 rcu_read_unlock();
1309discard:
1310 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1311out:
1312 trace_rxrpc_rx_done(0, 0);
1313 return;
1314
1315out_unlock:
1316 rcu_read_unlock();
1317 goto out;
1318
1319wrong_security:
1320 rcu_read_unlock();
1321 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1322 RXKADINCONSISTENCY, EBADMSG);
1323 skb->priority = RXKADINCONSISTENCY;
1324 goto post_abort;
1325
1326reupgrade:
1327 rcu_read_unlock();
1328 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1329 RX_PROTOCOL_ERROR, EBADMSG);
1330 goto protocol_error;
1331
1332bad_message_unlock:
1333 rcu_read_unlock();
1334bad_message:
1335 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1336 RX_PROTOCOL_ERROR, EBADMSG);
1337protocol_error:
1338 skb->priority = RX_PROTOCOL_ERROR;
1339post_abort:
1340 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
1341reject_packet:
1342 trace_rxrpc_rx_done(skb->mark, skb->priority);
1343 rxrpc_reject_packet(local, skb);
1344 _leave(" [badmsg]");
1345}