Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Processing of received RxRPC packets
3 *
4 * Copyright (C) 2020 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 "ar-internal.h"
11
12static void rxrpc_proto_abort(struct rxrpc_call *call, rxrpc_seq_t seq,
13 enum rxrpc_abort_reason why)
14{
15 rxrpc_abort_call(call, seq, RX_PROTOCOL_ERROR, -EBADMSG, why);
16}
17
18/*
19 * Do TCP-style congestion management [RFC 5681].
20 */
21static void rxrpc_congestion_management(struct rxrpc_call *call,
22 struct sk_buff *skb,
23 struct rxrpc_ack_summary *summary,
24 rxrpc_serial_t acked_serial)
25{
26 enum rxrpc_congest_change change = rxrpc_cong_no_change;
27 unsigned int cumulative_acks = call->cong_cumul_acks;
28 unsigned int cwnd = call->cong_cwnd;
29 bool resend = false;
30
31 summary->flight_size =
32 (call->tx_top - call->acks_hard_ack) - summary->nr_acks;
33
34 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
35 summary->retrans_timeo = true;
36 call->cong_ssthresh = max_t(unsigned int,
37 summary->flight_size / 2, 2);
38 cwnd = 1;
39 if (cwnd >= call->cong_ssthresh &&
40 call->cong_mode == RXRPC_CALL_SLOW_START) {
41 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
42 call->cong_tstamp = skb->tstamp;
43 cumulative_acks = 0;
44 }
45 }
46
47 cumulative_acks += summary->nr_new_acks;
48 cumulative_acks += summary->nr_rot_new_acks;
49 if (cumulative_acks > 255)
50 cumulative_acks = 255;
51
52 summary->mode = call->cong_mode;
53 summary->cwnd = call->cong_cwnd;
54 summary->ssthresh = call->cong_ssthresh;
55 summary->cumulative_acks = cumulative_acks;
56 summary->dup_acks = call->cong_dup_acks;
57
58 switch (call->cong_mode) {
59 case RXRPC_CALL_SLOW_START:
60 if (summary->saw_nacks)
61 goto packet_loss_detected;
62 if (summary->cumulative_acks > 0)
63 cwnd += 1;
64 if (cwnd >= call->cong_ssthresh) {
65 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
66 call->cong_tstamp = skb->tstamp;
67 }
68 goto out;
69
70 case RXRPC_CALL_CONGEST_AVOIDANCE:
71 if (summary->saw_nacks)
72 goto packet_loss_detected;
73
74 /* We analyse the number of packets that get ACK'd per RTT
75 * period and increase the window if we managed to fill it.
76 */
77 if (call->peer->rtt_count == 0)
78 goto out;
79 if (ktime_before(skb->tstamp,
80 ktime_add_us(call->cong_tstamp,
81 call->peer->srtt_us >> 3)))
82 goto out_no_clear_ca;
83 change = rxrpc_cong_rtt_window_end;
84 call->cong_tstamp = skb->tstamp;
85 if (cumulative_acks >= cwnd)
86 cwnd++;
87 goto out;
88
89 case RXRPC_CALL_PACKET_LOSS:
90 if (!summary->saw_nacks)
91 goto resume_normality;
92
93 if (summary->new_low_nack) {
94 change = rxrpc_cong_new_low_nack;
95 call->cong_dup_acks = 1;
96 if (call->cong_extra > 1)
97 call->cong_extra = 1;
98 goto send_extra_data;
99 }
100
101 call->cong_dup_acks++;
102 if (call->cong_dup_acks < 3)
103 goto send_extra_data;
104
105 change = rxrpc_cong_begin_retransmission;
106 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
107 call->cong_ssthresh = max_t(unsigned int,
108 summary->flight_size / 2, 2);
109 cwnd = call->cong_ssthresh + 3;
110 call->cong_extra = 0;
111 call->cong_dup_acks = 0;
112 resend = true;
113 goto out;
114
115 case RXRPC_CALL_FAST_RETRANSMIT:
116 if (!summary->new_low_nack) {
117 if (summary->nr_new_acks == 0)
118 cwnd += 1;
119 call->cong_dup_acks++;
120 if (call->cong_dup_acks == 2) {
121 change = rxrpc_cong_retransmit_again;
122 call->cong_dup_acks = 0;
123 resend = true;
124 }
125 } else {
126 change = rxrpc_cong_progress;
127 cwnd = call->cong_ssthresh;
128 if (!summary->saw_nacks)
129 goto resume_normality;
130 }
131 goto out;
132
133 default:
134 BUG();
135 goto out;
136 }
137
138resume_normality:
139 change = rxrpc_cong_cleared_nacks;
140 call->cong_dup_acks = 0;
141 call->cong_extra = 0;
142 call->cong_tstamp = skb->tstamp;
143 if (cwnd < call->cong_ssthresh)
144 call->cong_mode = RXRPC_CALL_SLOW_START;
145 else
146 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
147out:
148 cumulative_acks = 0;
149out_no_clear_ca:
150 if (cwnd >= RXRPC_TX_MAX_WINDOW)
151 cwnd = RXRPC_TX_MAX_WINDOW;
152 call->cong_cwnd = cwnd;
153 call->cong_cumul_acks = cumulative_acks;
154 trace_rxrpc_congest(call, summary, acked_serial, change);
155 if (resend)
156 rxrpc_resend(call, skb);
157 return;
158
159packet_loss_detected:
160 change = rxrpc_cong_saw_nack;
161 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
162 call->cong_dup_acks = 0;
163 goto send_extra_data;
164
165send_extra_data:
166 /* Send some previously unsent DATA if we have some to advance the ACK
167 * state.
168 */
169 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
170 summary->nr_acks != call->tx_top - call->acks_hard_ack) {
171 call->cong_extra++;
172 wake_up(&call->waitq);
173 }
174 goto out_no_clear_ca;
175}
176
177/*
178 * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
179 */
180void rxrpc_congestion_degrade(struct rxrpc_call *call)
181{
182 ktime_t rtt, now;
183
184 if (call->cong_mode != RXRPC_CALL_SLOW_START &&
185 call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
186 return;
187 if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY)
188 return;
189
190 rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
191 now = ktime_get_real();
192 if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
193 return;
194
195 trace_rxrpc_reset_cwnd(call, now);
196 rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
197 call->tx_last_sent = now;
198 call->cong_mode = RXRPC_CALL_SLOW_START;
199 call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
200 call->cong_cwnd * 3 / 4);
201 call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
202}
203
204/*
205 * Apply a hard ACK by advancing the Tx window.
206 */
207static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
208 struct rxrpc_ack_summary *summary)
209{
210 struct rxrpc_txbuf *txb;
211 bool rot_last = false;
212
213 list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
214 if (before_eq(txb->seq, call->acks_hard_ack))
215 continue;
216 summary->nr_rot_new_acks++;
217 if (test_bit(RXRPC_TXBUF_LAST, &txb->flags)) {
218 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
219 rot_last = true;
220 }
221 if (txb->seq == to)
222 break;
223 }
224
225 if (rot_last)
226 set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
227
228 _enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
229
230 if (call->acks_lowest_nak == call->acks_hard_ack) {
231 call->acks_lowest_nak = to;
232 } else if (after(to, call->acks_lowest_nak)) {
233 summary->new_low_nack = true;
234 call->acks_lowest_nak = to;
235 }
236
237 smp_store_release(&call->acks_hard_ack, to);
238
239 trace_rxrpc_txqueue(call, (rot_last ?
240 rxrpc_txqueue_rotate_last :
241 rxrpc_txqueue_rotate));
242 wake_up(&call->waitq);
243 return rot_last;
244}
245
246/*
247 * End the transmission phase of a call.
248 *
249 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
250 * or a final ACK packet.
251 */
252static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
253 enum rxrpc_abort_reason abort_why)
254{
255 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
256
257 switch (__rxrpc_call_state(call)) {
258 case RXRPC_CALL_CLIENT_SEND_REQUEST:
259 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
260 if (reply_begun) {
261 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
262 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
263 break;
264 }
265
266 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
267 trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
268 break;
269
270 case RXRPC_CALL_SERVER_AWAIT_ACK:
271 rxrpc_call_completed(call);
272 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
273 break;
274
275 default:
276 kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
277 rxrpc_proto_abort(call, call->tx_top, abort_why);
278 break;
279 }
280}
281
282/*
283 * Begin the reply reception phase of a call.
284 */
285static bool rxrpc_receiving_reply(struct rxrpc_call *call)
286{
287 struct rxrpc_ack_summary summary = { 0 };
288 unsigned long now, timo;
289 rxrpc_seq_t top = READ_ONCE(call->tx_top);
290
291 if (call->ackr_reason) {
292 now = jiffies;
293 timo = now + MAX_JIFFY_OFFSET;
294
295 WRITE_ONCE(call->delay_ack_at, timo);
296 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
297 }
298
299 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
300 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
301 rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
302 return false;
303 }
304 }
305
306 rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
307 return true;
308}
309
310/*
311 * End the packet reception phase.
312 */
313static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
314{
315 rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
316
317 _enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
318
319 trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
320
321 switch (__rxrpc_call_state(call)) {
322 case RXRPC_CALL_CLIENT_RECV_REPLY:
323 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
324 rxrpc_call_completed(call);
325 break;
326
327 case RXRPC_CALL_SERVER_RECV_REQUEST:
328 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
329 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
330 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
331 break;
332
333 default:
334 break;
335 }
336}
337
338static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
339 rxrpc_seq_t window, rxrpc_seq_t wtop)
340{
341 atomic64_set_release(&call->ackr_window, ((u64)wtop) << 32 | window);
342}
343
344/*
345 * Push a DATA packet onto the Rx queue.
346 */
347static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
348 rxrpc_seq_t window, rxrpc_seq_t wtop,
349 enum rxrpc_receive_trace why)
350{
351 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
352 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
353
354 __skb_queue_tail(&call->recvmsg_queue, skb);
355 rxrpc_input_update_ack_window(call, window, wtop);
356 trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
357 if (last)
358 rxrpc_end_rx_phase(call, sp->hdr.serial);
359}
360
361/*
362 * Process a DATA packet.
363 */
364static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
365 bool *_notify)
366{
367 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
368 struct sk_buff *oos;
369 rxrpc_serial_t serial = sp->hdr.serial;
370 u64 win = atomic64_read(&call->ackr_window);
371 rxrpc_seq_t window = lower_32_bits(win);
372 rxrpc_seq_t wtop = upper_32_bits(win);
373 rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
374 rxrpc_seq_t seq = sp->hdr.seq;
375 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
376 int ack_reason = -1;
377
378 rxrpc_inc_stat(call->rxnet, stat_rx_data);
379 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
380 rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
381 if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
382 rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
383
384 if (last) {
385 if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
386 seq + 1 != wtop)
387 return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
388 } else {
389 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
390 after_eq(seq, wtop)) {
391 pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
392 call->debug_id, seq, window, wtop, wlimit);
393 return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
394 }
395 }
396
397 if (after(seq, call->rx_highest_seq))
398 call->rx_highest_seq = seq;
399
400 trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
401
402 if (before(seq, window)) {
403 ack_reason = RXRPC_ACK_DUPLICATE;
404 goto send_ack;
405 }
406 if (after(seq, wlimit)) {
407 ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
408 goto send_ack;
409 }
410
411 /* Queue the packet. */
412 if (seq == window) {
413 rxrpc_seq_t reset_from;
414 bool reset_sack = false;
415
416 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
417 ack_reason = RXRPC_ACK_REQUESTED;
418 /* Send an immediate ACK if we fill in a hole */
419 else if (!skb_queue_empty(&call->rx_oos_queue))
420 ack_reason = RXRPC_ACK_DELAY;
421 else
422 atomic_inc_return(&call->ackr_nr_unacked);
423
424 window++;
425 if (after(window, wtop))
426 wtop = window;
427
428 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
429
430 spin_lock(&call->recvmsg_queue.lock);
431 rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
432 *_notify = true;
433
434 while ((oos = skb_peek(&call->rx_oos_queue))) {
435 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
436
437 if (after(osp->hdr.seq, window))
438 break;
439
440 __skb_unlink(oos, &call->rx_oos_queue);
441 last = osp->hdr.flags & RXRPC_LAST_PACKET;
442 seq = osp->hdr.seq;
443 if (!reset_sack) {
444 reset_from = seq;
445 reset_sack = true;
446 }
447
448 window++;
449 rxrpc_input_queue_data(call, oos, window, wtop,
450 rxrpc_receive_queue_oos);
451 }
452
453 spin_unlock(&call->recvmsg_queue.lock);
454
455 if (reset_sack) {
456 do {
457 call->ackr_sack_table[reset_from % RXRPC_SACK_SIZE] = 0;
458 } while (reset_from++, before(reset_from, window));
459 }
460 } else {
461 bool keep = false;
462
463 ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
464
465 if (!call->ackr_sack_table[seq % RXRPC_SACK_SIZE]) {
466 call->ackr_sack_table[seq % RXRPC_SACK_SIZE] = 1;
467 keep = 1;
468 }
469
470 if (after(seq + 1, wtop)) {
471 wtop = seq + 1;
472 rxrpc_input_update_ack_window(call, window, wtop);
473 }
474
475 if (!keep) {
476 ack_reason = RXRPC_ACK_DUPLICATE;
477 goto send_ack;
478 }
479
480 skb_queue_walk(&call->rx_oos_queue, oos) {
481 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
482
483 if (after(osp->hdr.seq, seq)) {
484 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
485 __skb_queue_before(&call->rx_oos_queue, oos, skb);
486 goto oos_queued;
487 }
488 }
489
490 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
491 __skb_queue_tail(&call->rx_oos_queue, skb);
492 oos_queued:
493 trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
494 sp->hdr.serial, sp->hdr.seq);
495 }
496
497send_ack:
498 if (ack_reason >= 0)
499 rxrpc_send_ACK(call, ack_reason, serial,
500 rxrpc_propose_ack_input_data);
501 else
502 rxrpc_propose_delay_ACK(call, serial,
503 rxrpc_propose_ack_input_data);
504}
505
506/*
507 * Split a jumbo packet and file the bits separately.
508 */
509static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
510{
511 struct rxrpc_jumbo_header jhdr;
512 struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
513 struct sk_buff *jskb;
514 unsigned int offset = sizeof(struct rxrpc_wire_header);
515 unsigned int len = skb->len - offset;
516 bool notify = false;
517
518 while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
519 if (len < RXRPC_JUMBO_SUBPKTLEN)
520 goto protocol_error;
521 if (sp->hdr.flags & RXRPC_LAST_PACKET)
522 goto protocol_error;
523 if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
524 &jhdr, sizeof(jhdr)) < 0)
525 goto protocol_error;
526
527 jskb = skb_clone(skb, GFP_NOFS);
528 if (!jskb) {
529 kdebug("couldn't clone");
530 return false;
531 }
532 rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
533 jsp = rxrpc_skb(jskb);
534 jsp->offset = offset;
535 jsp->len = RXRPC_JUMBO_DATALEN;
536 rxrpc_input_data_one(call, jskb, ¬ify);
537 rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
538
539 sp->hdr.flags = jhdr.flags;
540 sp->hdr._rsvd = ntohs(jhdr._rsvd);
541 sp->hdr.seq++;
542 sp->hdr.serial++;
543 offset += RXRPC_JUMBO_SUBPKTLEN;
544 len -= RXRPC_JUMBO_SUBPKTLEN;
545 }
546
547 sp->offset = offset;
548 sp->len = len;
549 rxrpc_input_data_one(call, skb, ¬ify);
550 if (notify) {
551 trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
552 rxrpc_notify_socket(call);
553 }
554 return true;
555
556protocol_error:
557 return false;
558}
559
560/*
561 * Process a DATA packet, adding the packet to the Rx ring. The caller's
562 * packet ref must be passed on or discarded.
563 */
564static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
565{
566 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
567 rxrpc_serial_t serial = sp->hdr.serial;
568 rxrpc_seq_t seq0 = sp->hdr.seq;
569
570 _enter("{%llx,%x},{%u,%x}",
571 atomic64_read(&call->ackr_window), call->rx_highest_seq,
572 skb->len, seq0);
573
574 if (__rxrpc_call_is_complete(call))
575 return;
576
577 switch (__rxrpc_call_state(call)) {
578 case RXRPC_CALL_CLIENT_SEND_REQUEST:
579 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
580 /* Received data implicitly ACKs all of the request
581 * packets we sent when we're acting as a client.
582 */
583 if (!rxrpc_receiving_reply(call))
584 goto out_notify;
585 break;
586
587 case RXRPC_CALL_SERVER_RECV_REQUEST: {
588 unsigned long timo = READ_ONCE(call->next_req_timo);
589 unsigned long now, expect_req_by;
590
591 if (timo) {
592 now = jiffies;
593 expect_req_by = now + timo;
594 WRITE_ONCE(call->expect_req_by, expect_req_by);
595 rxrpc_reduce_call_timer(call, expect_req_by, now,
596 rxrpc_timer_set_for_idle);
597 }
598 break;
599 }
600
601 default:
602 break;
603 }
604
605 if (!rxrpc_input_split_jumbo(call, skb)) {
606 rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
607 goto out_notify;
608 }
609 skb = NULL;
610
611out_notify:
612 trace_rxrpc_notify_socket(call->debug_id, serial);
613 rxrpc_notify_socket(call);
614 _leave(" [queued]");
615}
616
617/*
618 * See if there's a cached RTT probe to complete.
619 */
620static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
621 ktime_t resp_time,
622 rxrpc_serial_t acked_serial,
623 rxrpc_serial_t ack_serial,
624 enum rxrpc_rtt_rx_trace type)
625{
626 rxrpc_serial_t orig_serial;
627 unsigned long avail;
628 ktime_t sent_at;
629 bool matched = false;
630 int i;
631
632 avail = READ_ONCE(call->rtt_avail);
633 smp_rmb(); /* Read avail bits before accessing data. */
634
635 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
636 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
637 continue;
638
639 sent_at = call->rtt_sent_at[i];
640 orig_serial = call->rtt_serial[i];
641
642 if (orig_serial == acked_serial) {
643 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
644 smp_mb(); /* Read data before setting avail bit */
645 set_bit(i, &call->rtt_avail);
646 if (type != rxrpc_rtt_rx_cancel)
647 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
648 sent_at, resp_time);
649 else
650 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
651 orig_serial, acked_serial, 0, 0);
652 matched = true;
653 }
654
655 /* If a later serial is being acked, then mark this slot as
656 * being available.
657 */
658 if (after(acked_serial, orig_serial)) {
659 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
660 orig_serial, acked_serial, 0, 0);
661 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
662 smp_wmb();
663 set_bit(i, &call->rtt_avail);
664 }
665 }
666
667 if (!matched)
668 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
669}
670
671/*
672 * Process the extra information that may be appended to an ACK packet
673 */
674static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
675 struct rxrpc_ackinfo *ackinfo)
676{
677 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
678 struct rxrpc_peer *peer;
679 unsigned int mtu;
680 bool wake = false;
681 u32 rwind = ntohl(ackinfo->rwind);
682
683 if (rwind > RXRPC_TX_MAX_WINDOW)
684 rwind = RXRPC_TX_MAX_WINDOW;
685 if (call->tx_winsize != rwind) {
686 if (rwind > call->tx_winsize)
687 wake = true;
688 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
689 call->tx_winsize = rwind;
690 }
691
692 if (call->cong_ssthresh > rwind)
693 call->cong_ssthresh = rwind;
694
695 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
696
697 peer = call->peer;
698 if (mtu < peer->maxdata) {
699 spin_lock(&peer->lock);
700 peer->maxdata = mtu;
701 peer->mtu = mtu + peer->hdrsize;
702 spin_unlock(&peer->lock);
703 }
704
705 if (wake)
706 wake_up(&call->waitq);
707}
708
709/*
710 * Process individual soft ACKs.
711 *
712 * Each ACK in the array corresponds to one packet and can be either an ACK or
713 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
714 * packets that lie beyond the end of the ACK list are scheduled for resend by
715 * the timer on the basis that the peer might just not have processed them at
716 * the time the ACK was sent.
717 */
718static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
719 rxrpc_seq_t seq, int nr_acks,
720 struct rxrpc_ack_summary *summary)
721{
722 unsigned int i;
723
724 for (i = 0; i < nr_acks; i++) {
725 if (acks[i] == RXRPC_ACK_TYPE_ACK) {
726 summary->nr_acks++;
727 summary->nr_new_acks++;
728 } else {
729 if (!summary->saw_nacks &&
730 call->acks_lowest_nak != seq + i) {
731 call->acks_lowest_nak = seq + i;
732 summary->new_low_nack = true;
733 }
734 summary->saw_nacks = true;
735 }
736 }
737}
738
739/*
740 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
741 * with respect to the ack state conveyed by preceding ACKs.
742 */
743static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
744 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
745{
746 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
747
748 if (after(first_pkt, base))
749 return true; /* The window advanced */
750
751 if (before(first_pkt, base))
752 return false; /* firstPacket regressed */
753
754 if (after_eq(prev_pkt, call->acks_prev_seq))
755 return true; /* previousPacket hasn't regressed. */
756
757 /* Some rx implementations put a serial number in previousPacket. */
758 if (after_eq(prev_pkt, base + call->tx_winsize))
759 return false;
760 return true;
761}
762
763/*
764 * Process an ACK packet.
765 *
766 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
767 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
768 *
769 * A hard-ACK means that a packet has been processed and may be discarded; a
770 * soft-ACK means that the packet may be discarded and retransmission
771 * requested. A phase is complete when all packets are hard-ACK'd.
772 */
773static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
774{
775 struct rxrpc_ack_summary summary = { 0 };
776 struct rxrpc_ackpacket ack;
777 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
778 struct rxrpc_ackinfo info;
779 rxrpc_serial_t ack_serial, acked_serial;
780 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
781 int nr_acks, offset, ioffset;
782
783 _enter("");
784
785 offset = sizeof(struct rxrpc_wire_header);
786 if (skb_copy_bits(skb, offset, &ack, sizeof(ack)) < 0)
787 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack);
788 offset += sizeof(ack);
789
790 ack_serial = sp->hdr.serial;
791 acked_serial = ntohl(ack.serial);
792 first_soft_ack = ntohl(ack.firstPacket);
793 prev_pkt = ntohl(ack.previousPacket);
794 hard_ack = first_soft_ack - 1;
795 nr_acks = ack.nAcks;
796 summary.ack_reason = (ack.reason < RXRPC_ACK__INVALID ?
797 ack.reason : RXRPC_ACK__INVALID);
798
799 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
800 first_soft_ack, prev_pkt,
801 summary.ack_reason, nr_acks);
802 rxrpc_inc_stat(call->rxnet, stat_rx_acks[ack.reason]);
803
804 switch (ack.reason) {
805 case RXRPC_ACK_PING_RESPONSE:
806 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
807 rxrpc_rtt_rx_ping_response);
808 break;
809 case RXRPC_ACK_REQUESTED:
810 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
811 rxrpc_rtt_rx_requested_ack);
812 break;
813 default:
814 if (acked_serial != 0)
815 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
816 rxrpc_rtt_rx_cancel);
817 break;
818 }
819
820 if (ack.reason == RXRPC_ACK_PING) {
821 rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
822 rxrpc_propose_ack_respond_to_ping);
823 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
824 rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
825 rxrpc_propose_ack_respond_to_ack);
826 }
827
828 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
829 * indicates that the client address changed due to NAT. The server
830 * lost the call because it switched to a different peer.
831 */
832 if (unlikely(ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
833 first_soft_ack == 1 &&
834 prev_pkt == 0 &&
835 rxrpc_is_client_call(call)) {
836 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
837 0, -ENETRESET);
838 return;
839 }
840
841 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
842 * indicate a change of address. However, we can retransmit the call
843 * if we still have it buffered to the beginning.
844 */
845 if (unlikely(ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
846 first_soft_ack == 1 &&
847 prev_pkt == 0 &&
848 call->acks_hard_ack == 0 &&
849 rxrpc_is_client_call(call)) {
850 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
851 0, -ENETRESET);
852 return;
853 }
854
855 /* Discard any out-of-order or duplicate ACKs (outside lock). */
856 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
857 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
858 first_soft_ack, call->acks_first_seq,
859 prev_pkt, call->acks_prev_seq);
860 return;
861 }
862
863 info.rxMTU = 0;
864 ioffset = offset + nr_acks + 3;
865 if (skb->len >= ioffset + sizeof(info) &&
866 skb_copy_bits(skb, ioffset, &info, sizeof(info)) < 0)
867 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_info);
868
869 if (nr_acks > 0)
870 skb_condense(skb);
871
872 call->acks_latest_ts = skb->tstamp;
873 call->acks_first_seq = first_soft_ack;
874 call->acks_prev_seq = prev_pkt;
875
876 switch (ack.reason) {
877 case RXRPC_ACK_PING:
878 break;
879 default:
880 if (after(acked_serial, call->acks_highest_serial))
881 call->acks_highest_serial = acked_serial;
882 break;
883 }
884
885 /* Parse rwind and mtu sizes if provided. */
886 if (info.rxMTU)
887 rxrpc_input_ackinfo(call, skb, &info);
888
889 if (first_soft_ack == 0)
890 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
891
892 /* Ignore ACKs unless we are or have just been transmitting. */
893 switch (__rxrpc_call_state(call)) {
894 case RXRPC_CALL_CLIENT_SEND_REQUEST:
895 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
896 case RXRPC_CALL_SERVER_SEND_REPLY:
897 case RXRPC_CALL_SERVER_AWAIT_ACK:
898 break;
899 default:
900 return;
901 }
902
903 if (before(hard_ack, call->acks_hard_ack) ||
904 after(hard_ack, call->tx_top))
905 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
906 if (nr_acks > call->tx_top - hard_ack)
907 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
908
909 if (after(hard_ack, call->acks_hard_ack)) {
910 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
911 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
912 return;
913 }
914 }
915
916 if (nr_acks > 0) {
917 if (offset > (int)skb->len - nr_acks)
918 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
919 rxrpc_input_soft_acks(call, skb->data + offset, first_soft_ack,
920 nr_acks, &summary);
921 }
922
923 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
924 summary.nr_acks == call->tx_top - hard_ack &&
925 rxrpc_is_client_call(call))
926 rxrpc_propose_ping(call, ack_serial,
927 rxrpc_propose_ack_ping_for_lost_reply);
928
929 rxrpc_congestion_management(call, skb, &summary, acked_serial);
930}
931
932/*
933 * Process an ACKALL packet.
934 */
935static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
936{
937 struct rxrpc_ack_summary summary = { 0 };
938
939 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
940 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
941}
942
943/*
944 * Process an ABORT packet directed at a call.
945 */
946static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
947{
948 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
949
950 trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
951
952 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
953 skb->priority, -ECONNABORTED);
954}
955
956/*
957 * Process an incoming call packet.
958 */
959void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
960{
961 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
962 unsigned long timo;
963
964 _enter("%p,%p", call, skb);
965
966 if (sp->hdr.serviceId != call->dest_srx.srx_service)
967 call->dest_srx.srx_service = sp->hdr.serviceId;
968 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
969 call->rx_serial = sp->hdr.serial;
970 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
971 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
972
973 timo = READ_ONCE(call->next_rx_timo);
974 if (timo) {
975 unsigned long now = jiffies, expect_rx_by;
976
977 expect_rx_by = now + timo;
978 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
979 rxrpc_reduce_call_timer(call, expect_rx_by, now,
980 rxrpc_timer_set_for_normal);
981 }
982
983 switch (sp->hdr.type) {
984 case RXRPC_PACKET_TYPE_DATA:
985 return rxrpc_input_data(call, skb);
986
987 case RXRPC_PACKET_TYPE_ACK:
988 return rxrpc_input_ack(call, skb);
989
990 case RXRPC_PACKET_TYPE_BUSY:
991 /* Just ignore BUSY packets from the server; the retry and
992 * lifespan timers will take care of business. BUSY packets
993 * from the client don't make sense.
994 */
995 return;
996
997 case RXRPC_PACKET_TYPE_ABORT:
998 return rxrpc_input_abort(call, skb);
999
1000 case RXRPC_PACKET_TYPE_ACKALL:
1001 return rxrpc_input_ackall(call, skb);
1002
1003 default:
1004 break;
1005 }
1006}
1007
1008/*
1009 * Handle a new service call on a channel implicitly completing the preceding
1010 * call on that channel. This does not apply to client conns.
1011 *
1012 * TODO: If callNumber > call_id + 1, renegotiate security.
1013 */
1014void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1015{
1016 switch (__rxrpc_call_state(call)) {
1017 case RXRPC_CALL_SERVER_AWAIT_ACK:
1018 rxrpc_call_completed(call);
1019 fallthrough;
1020 case RXRPC_CALL_COMPLETE:
1021 break;
1022 default:
1023 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1024 rxrpc_eproto_improper_term);
1025 trace_rxrpc_improper_term(call);
1026 break;
1027 }
1028
1029 rxrpc_input_call_event(call, skb);
1030}
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_usage == 0)
95 goto out;
96 if (ktime_before(skb->tstamp,
97 ktime_add_ns(call->cong_tstamp,
98 call->peer->rtt)))
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 * Ping the other end to fill our RTT cache and to retrieve the rwind
197 * and MTU parameters.
198 */
199static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb)
200{
201 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
202 ktime_t now = skb->tstamp;
203
204 if (call->peer->rtt_usage < 3 ||
205 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now))
206 rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
207 true, true,
208 rxrpc_propose_ack_ping_for_params);
209}
210
211/*
212 * Apply a hard ACK by advancing the Tx window.
213 */
214static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
215 struct rxrpc_ack_summary *summary)
216{
217 struct sk_buff *skb, *list = NULL;
218 bool rot_last = false;
219 int ix;
220 u8 annotation;
221
222 if (call->acks_lowest_nak == call->tx_hard_ack) {
223 call->acks_lowest_nak = to;
224 } else if (before_eq(call->acks_lowest_nak, to)) {
225 summary->new_low_nack = true;
226 call->acks_lowest_nak = to;
227 }
228
229 spin_lock(&call->lock);
230
231 while (before(call->tx_hard_ack, to)) {
232 call->tx_hard_ack++;
233 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
234 skb = call->rxtx_buffer[ix];
235 annotation = call->rxtx_annotations[ix];
236 rxrpc_see_skb(skb, rxrpc_skb_rotated);
237 call->rxtx_buffer[ix] = NULL;
238 call->rxtx_annotations[ix] = 0;
239 skb->next = list;
240 list = skb;
241
242 if (annotation & RXRPC_TX_ANNO_LAST) {
243 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
244 rot_last = true;
245 }
246 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
247 summary->nr_rot_new_acks++;
248 }
249
250 spin_unlock(&call->lock);
251
252 trace_rxrpc_transmit(call, (rot_last ?
253 rxrpc_transmit_rotate_last :
254 rxrpc_transmit_rotate));
255 wake_up(&call->waitq);
256
257 while (list) {
258 skb = list;
259 list = skb->next;
260 skb_mark_not_on_list(skb);
261 rxrpc_free_skb(skb, rxrpc_skb_freed);
262 }
263
264 return rot_last;
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 unsigned int state;
277
278 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
279
280 write_lock(&call->state_lock);
281
282 state = call->state;
283 switch (state) {
284 case RXRPC_CALL_CLIENT_SEND_REQUEST:
285 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
286 if (reply_begun)
287 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
288 else
289 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
290 break;
291
292 case RXRPC_CALL_SERVER_AWAIT_ACK:
293 __rxrpc_call_completed(call);
294 rxrpc_notify_socket(call);
295 state = call->state;
296 break;
297
298 default:
299 goto bad_state;
300 }
301
302 write_unlock(&call->state_lock);
303 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
304 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
305 else
306 trace_rxrpc_transmit(call, rxrpc_transmit_end);
307 _leave(" = ok");
308 return true;
309
310bad_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 */
320static bool rxrpc_receiving_reply(struct rxrpc_call *call)
321{
322 struct rxrpc_ack_summary summary = { 0 };
323 unsigned long now, timo;
324 rxrpc_seq_t top = READ_ONCE(call->tx_top);
325
326 if (call->ackr_reason) {
327 spin_lock_bh(&call->lock);
328 call->ackr_reason = 0;
329 spin_unlock_bh(&call->lock);
330 now = jiffies;
331 timo = now + MAX_JIFFY_OFFSET;
332 WRITE_ONCE(call->resend_at, timo);
333 WRITE_ONCE(call->ack_at, timo);
334 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
335 }
336
337 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
338 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
339 rxrpc_proto_abort("TXL", call, top);
340 return false;
341 }
342 }
343 if (!rxrpc_end_tx_phase(call, true, "ETD"))
344 return false;
345 call->tx_phase = false;
346 return true;
347}
348
349/*
350 * Scan a data packet to validate its structure and to work out how many
351 * subpackets it contains.
352 *
353 * A jumbo packet is a collection of consecutive packets glued together with
354 * little headers between that indicate how to change the initial header for
355 * each subpacket.
356 *
357 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
358 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
359 * size.
360 */
361static bool rxrpc_validate_data(struct sk_buff *skb)
362{
363 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
364 unsigned int offset = sizeof(struct rxrpc_wire_header);
365 unsigned int len = skb->len;
366 u8 flags = sp->hdr.flags;
367
368 for (;;) {
369 if (flags & RXRPC_REQUEST_ACK)
370 __set_bit(sp->nr_subpackets, sp->rx_req_ack);
371 sp->nr_subpackets++;
372
373 if (!(flags & RXRPC_JUMBO_PACKET))
374 break;
375
376 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
377 goto protocol_error;
378 if (flags & RXRPC_LAST_PACKET)
379 goto protocol_error;
380 offset += RXRPC_JUMBO_DATALEN;
381 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
382 goto protocol_error;
383 offset += sizeof(struct rxrpc_jumbo_header);
384 }
385
386 if (flags & RXRPC_LAST_PACKET)
387 sp->rx_flags |= RXRPC_SKB_INCL_LAST;
388 return true;
389
390protocol_error:
391 return false;
392}
393
394/*
395 * Handle reception of a duplicate packet.
396 *
397 * We have to take care to avoid an attack here whereby we're given a series of
398 * jumbograms, each with a sequence number one before the preceding one and
399 * filled up to maximum UDP size. If they never send us the first packet in
400 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
401 * space until the call times out.
402 *
403 * We limit the space usage by only accepting three duplicate jumbo packets per
404 * call. After that, we tell the other side we're no longer accepting jumbos
405 * (that information is encoded in the ACK packet).
406 */
407static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
408 bool is_jumbo, bool *_jumbo_bad)
409{
410 /* Discard normal packets that are duplicates. */
411 if (is_jumbo)
412 return;
413
414 /* Skip jumbo subpackets that are duplicates. When we've had three or
415 * more partially duplicate jumbo packets, we refuse to take any more
416 * jumbos for this call.
417 */
418 if (!*_jumbo_bad) {
419 call->nr_jumbo_bad++;
420 *_jumbo_bad = true;
421 }
422}
423
424/*
425 * Process a DATA packet, adding the packet to the Rx ring. The caller's
426 * packet ref must be passed on or discarded.
427 */
428static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
429{
430 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
431 enum rxrpc_call_state state;
432 unsigned int j;
433 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
434 rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
435 bool immediate_ack = false, jumbo_bad = false;
436 u8 ack = 0;
437
438 _enter("{%u,%u},{%u,%u}",
439 call->rx_hard_ack, call->rx_top, skb->len, seq0);
440
441 _proto("Rx DATA %%%u { #%u f=%02x n=%u }",
442 sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
443
444 state = READ_ONCE(call->state);
445 if (state >= RXRPC_CALL_COMPLETE) {
446 rxrpc_free_skb(skb, rxrpc_skb_freed);
447 return;
448 }
449
450 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
451 unsigned long timo = READ_ONCE(call->next_req_timo);
452 unsigned long now, expect_req_by;
453
454 if (timo) {
455 now = jiffies;
456 expect_req_by = now + timo;
457 WRITE_ONCE(call->expect_req_by, expect_req_by);
458 rxrpc_reduce_call_timer(call, expect_req_by, now,
459 rxrpc_timer_set_for_idle);
460 }
461 }
462
463 spin_lock(&call->input_lock);
464
465 /* Received data implicitly ACKs all of the request packets we sent
466 * when we're acting as a client.
467 */
468 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
469 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
470 !rxrpc_receiving_reply(call))
471 goto unlock;
472
473 call->ackr_prev_seq = seq0;
474 hard_ack = READ_ONCE(call->rx_hard_ack);
475
476 if (sp->nr_subpackets > 1) {
477 if (call->nr_jumbo_bad > 3) {
478 ack = RXRPC_ACK_NOSPACE;
479 ack_serial = serial;
480 goto ack;
481 }
482 }
483
484 for (j = 0; j < sp->nr_subpackets; j++) {
485 rxrpc_serial_t serial = sp->hdr.serial + j;
486 rxrpc_seq_t seq = seq0 + j;
487 unsigned int ix = seq & RXRPC_RXTX_BUFF_MASK;
488 bool terminal = (j == sp->nr_subpackets - 1);
489 bool last = terminal && (sp->rx_flags & RXRPC_SKB_INCL_LAST);
490 u8 flags, annotation = j;
491
492 _proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
493 j, serial, seq, terminal, last);
494
495 if (last) {
496 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
497 seq != call->rx_top) {
498 rxrpc_proto_abort("LSN", call, seq);
499 goto unlock;
500 }
501 } else {
502 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
503 after_eq(seq, call->rx_top)) {
504 rxrpc_proto_abort("LSA", call, seq);
505 goto unlock;
506 }
507 }
508
509 flags = 0;
510 if (last)
511 flags |= RXRPC_LAST_PACKET;
512 if (!terminal)
513 flags |= RXRPC_JUMBO_PACKET;
514 if (test_bit(j, sp->rx_req_ack))
515 flags |= RXRPC_REQUEST_ACK;
516 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
517
518 if (before_eq(seq, hard_ack)) {
519 ack = RXRPC_ACK_DUPLICATE;
520 ack_serial = serial;
521 continue;
522 }
523
524 if (call->rxtx_buffer[ix]) {
525 rxrpc_input_dup_data(call, seq, sp->nr_subpackets > 1,
526 &jumbo_bad);
527 if (ack != RXRPC_ACK_DUPLICATE) {
528 ack = RXRPC_ACK_DUPLICATE;
529 ack_serial = serial;
530 }
531 immediate_ack = true;
532 continue;
533 }
534
535 if (after(seq, hard_ack + call->rx_winsize)) {
536 ack = RXRPC_ACK_EXCEEDS_WINDOW;
537 ack_serial = serial;
538 if (flags & RXRPC_JUMBO_PACKET) {
539 if (!jumbo_bad) {
540 call->nr_jumbo_bad++;
541 jumbo_bad = true;
542 }
543 }
544
545 goto ack;
546 }
547
548 if (flags & RXRPC_REQUEST_ACK && !ack) {
549 ack = RXRPC_ACK_REQUESTED;
550 ack_serial = serial;
551 }
552
553 /* Queue the packet. We use a couple of memory barriers here as need
554 * to make sure that rx_top is perceived to be set after the buffer
555 * pointer and that the buffer pointer is set after the annotation and
556 * the skb data.
557 *
558 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
559 * and also rxrpc_fill_out_ack().
560 */
561 if (!terminal)
562 rxrpc_get_skb(skb, rxrpc_skb_got);
563 call->rxtx_annotations[ix] = annotation;
564 smp_wmb();
565 call->rxtx_buffer[ix] = skb;
566 if (after(seq, call->rx_top)) {
567 smp_store_release(&call->rx_top, seq);
568 } else if (before(seq, call->rx_top)) {
569 /* Send an immediate ACK if we fill in a hole */
570 if (!ack) {
571 ack = RXRPC_ACK_DELAY;
572 ack_serial = serial;
573 }
574 immediate_ack = true;
575 }
576
577 if (terminal) {
578 /* From this point on, we're not allowed to touch the
579 * packet any longer as its ref now belongs to the Rx
580 * ring.
581 */
582 skb = NULL;
583 }
584
585 if (last) {
586 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
587 if (!ack) {
588 ack = RXRPC_ACK_DELAY;
589 ack_serial = serial;
590 }
591 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
592 } else {
593 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
594 }
595
596 if (after_eq(seq, call->rx_expect_next)) {
597 if (after(seq, call->rx_expect_next)) {
598 _net("OOS %u > %u", seq, call->rx_expect_next);
599 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
600 ack_serial = serial;
601 }
602 call->rx_expect_next = seq + 1;
603 }
604 }
605
606ack:
607 if (ack)
608 rxrpc_propose_ACK(call, ack, ack_serial,
609 immediate_ack, true,
610 rxrpc_propose_ack_input_data);
611 else
612 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
613 false, true,
614 rxrpc_propose_ack_input_data);
615
616 if (seq0 == READ_ONCE(call->rx_hard_ack) + 1) {
617 trace_rxrpc_notify_socket(call->debug_id, serial);
618 rxrpc_notify_socket(call);
619 }
620
621unlock:
622 spin_unlock(&call->input_lock);
623 rxrpc_free_skb(skb, rxrpc_skb_freed);
624 _leave(" [queued]");
625}
626
627/*
628 * Process a requested ACK.
629 */
630static void rxrpc_input_requested_ack(struct rxrpc_call *call,
631 ktime_t resp_time,
632 rxrpc_serial_t orig_serial,
633 rxrpc_serial_t ack_serial)
634{
635 struct rxrpc_skb_priv *sp;
636 struct sk_buff *skb;
637 ktime_t sent_at;
638 int ix;
639
640 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
641 skb = call->rxtx_buffer[ix];
642 if (!skb)
643 continue;
644
645 sent_at = skb->tstamp;
646 smp_rmb(); /* Read timestamp before serial. */
647 sp = rxrpc_skb(skb);
648 if (sp->hdr.serial != orig_serial)
649 continue;
650 goto found;
651 }
652
653 return;
654
655found:
656 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
657 orig_serial, ack_serial, sent_at, resp_time);
658}
659
660/*
661 * Process the response to a ping that we sent to find out if we lost an ACK.
662 *
663 * If we got back a ping response that indicates a lower tx_top than what we
664 * had at the time of the ping transmission, we adjudge all the DATA packets
665 * sent between the response tx_top and the ping-time tx_top to have been lost.
666 */
667static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
668{
669 rxrpc_seq_t top, bottom, seq;
670 bool resend = false;
671
672 spin_lock_bh(&call->lock);
673
674 bottom = call->tx_hard_ack + 1;
675 top = call->acks_lost_top;
676 if (before(bottom, top)) {
677 for (seq = bottom; before_eq(seq, top); seq++) {
678 int ix = seq & RXRPC_RXTX_BUFF_MASK;
679 u8 annotation = call->rxtx_annotations[ix];
680 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
681
682 if (anno_type != RXRPC_TX_ANNO_UNACK)
683 continue;
684 annotation &= ~RXRPC_TX_ANNO_MASK;
685 annotation |= RXRPC_TX_ANNO_RETRANS;
686 call->rxtx_annotations[ix] = annotation;
687 resend = true;
688 }
689 }
690
691 spin_unlock_bh(&call->lock);
692
693 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
694 rxrpc_queue_call(call);
695}
696
697/*
698 * Process a ping response.
699 */
700static void rxrpc_input_ping_response(struct rxrpc_call *call,
701 ktime_t resp_time,
702 rxrpc_serial_t orig_serial,
703 rxrpc_serial_t ack_serial)
704{
705 rxrpc_serial_t ping_serial;
706 ktime_t ping_time;
707
708 ping_time = call->ping_time;
709 smp_rmb();
710 ping_serial = READ_ONCE(call->ping_serial);
711
712 if (orig_serial == call->acks_lost_ping)
713 rxrpc_input_check_for_lost_ack(call);
714
715 if (before(orig_serial, ping_serial) ||
716 !test_and_clear_bit(RXRPC_CALL_PINGING, &call->flags))
717 return;
718 if (after(orig_serial, ping_serial))
719 return;
720
721 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
722 orig_serial, ack_serial, ping_time, resp_time);
723}
724
725/*
726 * Process the extra information that may be appended to an ACK packet
727 */
728static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
729 struct rxrpc_ackinfo *ackinfo)
730{
731 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
732 struct rxrpc_peer *peer;
733 unsigned int mtu;
734 bool wake = false;
735 u32 rwind = ntohl(ackinfo->rwind);
736
737 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
738 sp->hdr.serial,
739 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
740 rwind, ntohl(ackinfo->jumbo_max));
741
742 if (call->tx_winsize != rwind) {
743 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
744 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
745 if (rwind > call->tx_winsize)
746 wake = true;
747 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
748 ntohl(ackinfo->rwind), wake);
749 call->tx_winsize = rwind;
750 }
751
752 if (call->cong_ssthresh > rwind)
753 call->cong_ssthresh = rwind;
754
755 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
756
757 peer = call->peer;
758 if (mtu < peer->maxdata) {
759 spin_lock_bh(&peer->lock);
760 peer->maxdata = mtu;
761 peer->mtu = mtu + peer->hdrsize;
762 spin_unlock_bh(&peer->lock);
763 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
764 }
765
766 if (wake)
767 wake_up(&call->waitq);
768}
769
770/*
771 * Process individual soft ACKs.
772 *
773 * Each ACK in the array corresponds to one packet and can be either an ACK or
774 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
775 * packets that lie beyond the end of the ACK list are scheduled for resend by
776 * the timer on the basis that the peer might just not have processed them at
777 * the time the ACK was sent.
778 */
779static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
780 rxrpc_seq_t seq, int nr_acks,
781 struct rxrpc_ack_summary *summary)
782{
783 int ix;
784 u8 annotation, anno_type;
785
786 for (; nr_acks > 0; nr_acks--, seq++) {
787 ix = seq & RXRPC_RXTX_BUFF_MASK;
788 annotation = call->rxtx_annotations[ix];
789 anno_type = annotation & RXRPC_TX_ANNO_MASK;
790 annotation &= ~RXRPC_TX_ANNO_MASK;
791 switch (*acks++) {
792 case RXRPC_ACK_TYPE_ACK:
793 summary->nr_acks++;
794 if (anno_type == RXRPC_TX_ANNO_ACK)
795 continue;
796 summary->nr_new_acks++;
797 call->rxtx_annotations[ix] =
798 RXRPC_TX_ANNO_ACK | annotation;
799 break;
800 case RXRPC_ACK_TYPE_NACK:
801 if (!summary->nr_nacks &&
802 call->acks_lowest_nak != seq) {
803 call->acks_lowest_nak = seq;
804 summary->new_low_nack = true;
805 }
806 summary->nr_nacks++;
807 if (anno_type == RXRPC_TX_ANNO_NAK)
808 continue;
809 summary->nr_new_nacks++;
810 if (anno_type == RXRPC_TX_ANNO_RETRANS)
811 continue;
812 call->rxtx_annotations[ix] =
813 RXRPC_TX_ANNO_NAK | annotation;
814 break;
815 default:
816 return rxrpc_proto_abort("SFT", call, 0);
817 }
818 }
819}
820
821/*
822 * Process an ACK packet.
823 *
824 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
825 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
826 *
827 * A hard-ACK means that a packet has been processed and may be discarded; a
828 * soft-ACK means that the packet may be discarded and retransmission
829 * requested. A phase is complete when all packets are hard-ACK'd.
830 */
831static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
832{
833 struct rxrpc_ack_summary summary = { 0 };
834 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
835 union {
836 struct rxrpc_ackpacket ack;
837 struct rxrpc_ackinfo info;
838 u8 acks[RXRPC_MAXACKS];
839 } buf;
840 rxrpc_serial_t acked_serial;
841 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
842 int nr_acks, offset, ioffset;
843
844 _enter("");
845
846 offset = sizeof(struct rxrpc_wire_header);
847 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
848 _debug("extraction failure");
849 return rxrpc_proto_abort("XAK", call, 0);
850 }
851 offset += sizeof(buf.ack);
852
853 acked_serial = ntohl(buf.ack.serial);
854 first_soft_ack = ntohl(buf.ack.firstPacket);
855 prev_pkt = ntohl(buf.ack.previousPacket);
856 hard_ack = first_soft_ack - 1;
857 nr_acks = buf.ack.nAcks;
858 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
859 buf.ack.reason : RXRPC_ACK__INVALID);
860
861 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
862 first_soft_ack, prev_pkt,
863 summary.ack_reason, nr_acks);
864
865 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
866 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
867 sp->hdr.serial);
868 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
869 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
870 sp->hdr.serial);
871
872 if (buf.ack.reason == RXRPC_ACK_PING) {
873 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
874 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
875 sp->hdr.serial, true, true,
876 rxrpc_propose_ack_respond_to_ping);
877 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
878 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
879 sp->hdr.serial, true, true,
880 rxrpc_propose_ack_respond_to_ack);
881 }
882
883 /* Discard any out-of-order or duplicate ACKs (outside lock). */
884 if (before(first_soft_ack, call->ackr_first_seq) ||
885 before(prev_pkt, call->ackr_prev_seq))
886 return;
887
888 buf.info.rxMTU = 0;
889 ioffset = offset + nr_acks + 3;
890 if (skb->len >= ioffset + sizeof(buf.info) &&
891 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
892 return rxrpc_proto_abort("XAI", call, 0);
893
894 spin_lock(&call->input_lock);
895
896 /* Discard any out-of-order or duplicate ACKs (inside lock). */
897 if (before(first_soft_ack, call->ackr_first_seq) ||
898 before(prev_pkt, call->ackr_prev_seq))
899 goto out;
900 call->acks_latest_ts = skb->tstamp;
901 call->acks_latest = sp->hdr.serial;
902
903 call->ackr_first_seq = first_soft_ack;
904 call->ackr_prev_seq = prev_pkt;
905
906 /* Parse rwind and mtu sizes if provided. */
907 if (buf.info.rxMTU)
908 rxrpc_input_ackinfo(call, skb, &buf.info);
909
910 if (first_soft_ack == 0) {
911 rxrpc_proto_abort("AK0", call, 0);
912 goto out;
913 }
914
915 /* Ignore ACKs unless we are or have just been transmitting. */
916 switch (READ_ONCE(call->state)) {
917 case RXRPC_CALL_CLIENT_SEND_REQUEST:
918 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
919 case RXRPC_CALL_SERVER_SEND_REPLY:
920 case RXRPC_CALL_SERVER_AWAIT_ACK:
921 break;
922 default:
923 goto out;
924 }
925
926 if (before(hard_ack, call->tx_hard_ack) ||
927 after(hard_ack, call->tx_top)) {
928 rxrpc_proto_abort("AKW", call, 0);
929 goto out;
930 }
931 if (nr_acks > call->tx_top - hard_ack) {
932 rxrpc_proto_abort("AKN", call, 0);
933 goto out;
934 }
935
936 if (after(hard_ack, call->tx_hard_ack)) {
937 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
938 rxrpc_end_tx_phase(call, false, "ETA");
939 goto out;
940 }
941 }
942
943 if (nr_acks > 0) {
944 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
945 rxrpc_proto_abort("XSA", call, 0);
946 goto out;
947 }
948 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
949 &summary);
950 }
951
952 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
953 RXRPC_TX_ANNO_LAST &&
954 summary.nr_acks == call->tx_top - hard_ack &&
955 rxrpc_is_client_call(call))
956 rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
957 false, true,
958 rxrpc_propose_ack_ping_for_lost_reply);
959
960 rxrpc_congestion_management(call, skb, &summary, acked_serial);
961out:
962 spin_unlock(&call->input_lock);
963}
964
965/*
966 * Process an ACKALL packet.
967 */
968static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
969{
970 struct rxrpc_ack_summary summary = { 0 };
971 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
972
973 _proto("Rx ACKALL %%%u", sp->hdr.serial);
974
975 spin_lock(&call->input_lock);
976
977 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
978 rxrpc_end_tx_phase(call, false, "ETL");
979
980 spin_unlock(&call->input_lock);
981}
982
983/*
984 * Process an ABORT packet directed at a call.
985 */
986static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
987{
988 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
989 __be32 wtmp;
990 u32 abort_code = RX_CALL_DEAD;
991
992 _enter("");
993
994 if (skb->len >= 4 &&
995 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
996 &wtmp, sizeof(wtmp)) >= 0)
997 abort_code = ntohl(wtmp);
998
999 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1000
1001 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1002
1003 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1004 abort_code, -ECONNABORTED))
1005 rxrpc_notify_socket(call);
1006}
1007
1008/*
1009 * Process an incoming call packet.
1010 */
1011static void rxrpc_input_call_packet(struct rxrpc_call *call,
1012 struct sk_buff *skb)
1013{
1014 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1015 unsigned long timo;
1016
1017 _enter("%p,%p", call, skb);
1018
1019 timo = READ_ONCE(call->next_rx_timo);
1020 if (timo) {
1021 unsigned long now = jiffies, expect_rx_by;
1022
1023 expect_rx_by = now + timo;
1024 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1025 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1026 rxrpc_timer_set_for_normal);
1027 }
1028
1029 switch (sp->hdr.type) {
1030 case RXRPC_PACKET_TYPE_DATA:
1031 rxrpc_input_data(call, skb);
1032 goto no_free;
1033
1034 case RXRPC_PACKET_TYPE_ACK:
1035 rxrpc_input_ack(call, skb);
1036 break;
1037
1038 case RXRPC_PACKET_TYPE_BUSY:
1039 _proto("Rx BUSY %%%u", sp->hdr.serial);
1040
1041 /* Just ignore BUSY packets from the server; the retry and
1042 * lifespan timers will take care of business. BUSY packets
1043 * from the client don't make sense.
1044 */
1045 break;
1046
1047 case RXRPC_PACKET_TYPE_ABORT:
1048 rxrpc_input_abort(call, skb);
1049 break;
1050
1051 case RXRPC_PACKET_TYPE_ACKALL:
1052 rxrpc_input_ackall(call, skb);
1053 break;
1054
1055 default:
1056 break;
1057 }
1058
1059 rxrpc_free_skb(skb, rxrpc_skb_freed);
1060no_free:
1061 _leave("");
1062}
1063
1064/*
1065 * Handle a new service call on a channel implicitly completing the preceding
1066 * call on that channel. This does not apply to client conns.
1067 *
1068 * TODO: If callNumber > call_id + 1, renegotiate security.
1069 */
1070static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1071 struct rxrpc_connection *conn,
1072 struct rxrpc_call *call)
1073{
1074 switch (READ_ONCE(call->state)) {
1075 case RXRPC_CALL_SERVER_AWAIT_ACK:
1076 rxrpc_call_completed(call);
1077 /* Fall through */
1078 case RXRPC_CALL_COMPLETE:
1079 break;
1080 default:
1081 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1082 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1083 rxrpc_queue_call(call);
1084 }
1085 trace_rxrpc_improper_term(call);
1086 break;
1087 }
1088
1089 spin_lock(&rx->incoming_lock);
1090 __rxrpc_disconnect_call(conn, call);
1091 spin_unlock(&rx->incoming_lock);
1092 rxrpc_notify_socket(call);
1093}
1094
1095/*
1096 * post connection-level events to the connection
1097 * - this includes challenges, responses, some aborts and call terminal packet
1098 * retransmission.
1099 */
1100static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1101 struct sk_buff *skb)
1102{
1103 _enter("%p,%p", conn, skb);
1104
1105 skb_queue_tail(&conn->rx_queue, skb);
1106 rxrpc_queue_conn(conn);
1107}
1108
1109/*
1110 * post endpoint-level events to the local endpoint
1111 * - this includes debug and version messages
1112 */
1113static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1114 struct sk_buff *skb)
1115{
1116 _enter("%p,%p", local, skb);
1117
1118 if (rxrpc_get_local_maybe(local)) {
1119 skb_queue_tail(&local->event_queue, skb);
1120 rxrpc_queue_local(local);
1121 } else {
1122 rxrpc_free_skb(skb, rxrpc_skb_freed);
1123 }
1124}
1125
1126/*
1127 * put a packet up for transport-level abort
1128 */
1129static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1130{
1131 CHECK_SLAB_OKAY(&local->usage);
1132
1133 if (rxrpc_get_local_maybe(local)) {
1134 skb_queue_tail(&local->reject_queue, skb);
1135 rxrpc_queue_local(local);
1136 } else {
1137 rxrpc_free_skb(skb, rxrpc_skb_freed);
1138 }
1139}
1140
1141/*
1142 * Extract the wire header from a packet and translate the byte order.
1143 */
1144static noinline
1145int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1146{
1147 struct rxrpc_wire_header whdr;
1148
1149 /* dig out the RxRPC connection details */
1150 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1151 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1152 tracepoint_string("bad_hdr"));
1153 return -EBADMSG;
1154 }
1155
1156 memset(sp, 0, sizeof(*sp));
1157 sp->hdr.epoch = ntohl(whdr.epoch);
1158 sp->hdr.cid = ntohl(whdr.cid);
1159 sp->hdr.callNumber = ntohl(whdr.callNumber);
1160 sp->hdr.seq = ntohl(whdr.seq);
1161 sp->hdr.serial = ntohl(whdr.serial);
1162 sp->hdr.flags = whdr.flags;
1163 sp->hdr.type = whdr.type;
1164 sp->hdr.userStatus = whdr.userStatus;
1165 sp->hdr.securityIndex = whdr.securityIndex;
1166 sp->hdr._rsvd = ntohs(whdr._rsvd);
1167 sp->hdr.serviceId = ntohs(whdr.serviceId);
1168 return 0;
1169}
1170
1171/*
1172 * handle data received on the local endpoint
1173 * - may be called in interrupt context
1174 *
1175 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1176 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1177 * being cleared in the middle of processing this function.
1178 *
1179 * Called with the RCU read lock held from the IP layer via UDP.
1180 */
1181int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1182{
1183 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1184 struct rxrpc_connection *conn;
1185 struct rxrpc_channel *chan;
1186 struct rxrpc_call *call = NULL;
1187 struct rxrpc_skb_priv *sp;
1188 struct rxrpc_peer *peer = NULL;
1189 struct rxrpc_sock *rx = NULL;
1190 unsigned int channel;
1191
1192 _enter("%p", udp_sk);
1193
1194 if (unlikely(!local)) {
1195 kfree_skb(skb);
1196 return 0;
1197 }
1198 if (skb->tstamp == 0)
1199 skb->tstamp = ktime_get_real();
1200
1201 rxrpc_new_skb(skb, rxrpc_skb_received);
1202
1203 skb_pull(skb, sizeof(struct udphdr));
1204
1205 /* The UDP protocol already released all skb resources;
1206 * we are free to add our own data there.
1207 */
1208 sp = rxrpc_skb(skb);
1209
1210 /* dig out the RxRPC connection details */
1211 if (rxrpc_extract_header(sp, skb) < 0)
1212 goto bad_message;
1213
1214 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1215 static int lose;
1216 if ((lose++ & 7) == 7) {
1217 trace_rxrpc_rx_lose(sp);
1218 rxrpc_free_skb(skb, rxrpc_skb_lost);
1219 return 0;
1220 }
1221 }
1222
1223 if (skb->tstamp == 0)
1224 skb->tstamp = ktime_get_real();
1225 trace_rxrpc_rx_packet(sp);
1226
1227 switch (sp->hdr.type) {
1228 case RXRPC_PACKET_TYPE_VERSION:
1229 if (rxrpc_to_client(sp))
1230 goto discard;
1231 rxrpc_post_packet_to_local(local, skb);
1232 goto out;
1233
1234 case RXRPC_PACKET_TYPE_BUSY:
1235 if (rxrpc_to_server(sp))
1236 goto discard;
1237 /* Fall through */
1238 case RXRPC_PACKET_TYPE_ACK:
1239 case RXRPC_PACKET_TYPE_ACKALL:
1240 if (sp->hdr.callNumber == 0)
1241 goto bad_message;
1242 /* Fall through */
1243 case RXRPC_PACKET_TYPE_ABORT:
1244 break;
1245
1246 case RXRPC_PACKET_TYPE_DATA:
1247 if (sp->hdr.callNumber == 0 ||
1248 sp->hdr.seq == 0)
1249 goto bad_message;
1250 if (!rxrpc_validate_data(skb))
1251 goto bad_message;
1252
1253 /* Unshare the packet so that it can be modified for in-place
1254 * decryption.
1255 */
1256 if (sp->hdr.securityIndex != 0) {
1257 struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1258 if (!nskb) {
1259 rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1260 goto out;
1261 }
1262
1263 if (nskb != skb) {
1264 rxrpc_eaten_skb(skb, rxrpc_skb_received);
1265 skb = nskb;
1266 rxrpc_new_skb(skb, rxrpc_skb_unshared);
1267 sp = rxrpc_skb(skb);
1268 }
1269 }
1270 break;
1271
1272 case RXRPC_PACKET_TYPE_CHALLENGE:
1273 if (rxrpc_to_server(sp))
1274 goto discard;
1275 break;
1276 case RXRPC_PACKET_TYPE_RESPONSE:
1277 if (rxrpc_to_client(sp))
1278 goto discard;
1279 break;
1280
1281 /* Packet types 9-11 should just be ignored. */
1282 case RXRPC_PACKET_TYPE_PARAMS:
1283 case RXRPC_PACKET_TYPE_10:
1284 case RXRPC_PACKET_TYPE_11:
1285 goto discard;
1286
1287 default:
1288 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1289 goto bad_message;
1290 }
1291
1292 if (sp->hdr.serviceId == 0)
1293 goto bad_message;
1294
1295 if (rxrpc_to_server(sp)) {
1296 /* Weed out packets to services we're not offering. Packets
1297 * that would begin a call are explicitly rejected and the rest
1298 * are just discarded.
1299 */
1300 rx = rcu_dereference(local->service);
1301 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1302 sp->hdr.serviceId != rx->second_service)) {
1303 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1304 sp->hdr.seq == 1)
1305 goto unsupported_service;
1306 goto discard;
1307 }
1308 }
1309
1310 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1311 if (conn) {
1312 if (sp->hdr.securityIndex != conn->security_ix)
1313 goto wrong_security;
1314
1315 if (sp->hdr.serviceId != conn->service_id) {
1316 int old_id;
1317
1318 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1319 goto reupgrade;
1320 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1321 sp->hdr.serviceId);
1322
1323 if (old_id != conn->params.service_id &&
1324 old_id != sp->hdr.serviceId)
1325 goto reupgrade;
1326 }
1327
1328 if (sp->hdr.callNumber == 0) {
1329 /* Connection-level packet */
1330 _debug("CONN %p {%d}", conn, conn->debug_id);
1331 rxrpc_post_packet_to_conn(conn, skb);
1332 goto out;
1333 }
1334
1335 if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1336 conn->hi_serial = sp->hdr.serial;
1337
1338 /* Call-bound packets are routed by connection channel. */
1339 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1340 chan = &conn->channels[channel];
1341
1342 /* Ignore really old calls */
1343 if (sp->hdr.callNumber < chan->last_call)
1344 goto discard;
1345
1346 if (sp->hdr.callNumber == chan->last_call) {
1347 if (chan->call ||
1348 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1349 goto discard;
1350
1351 /* For the previous service call, if completed
1352 * successfully, we discard all further packets.
1353 */
1354 if (rxrpc_conn_is_service(conn) &&
1355 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1356 goto discard;
1357
1358 /* But otherwise we need to retransmit the final packet
1359 * from data cached in the connection record.
1360 */
1361 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1362 trace_rxrpc_rx_data(chan->call_debug_id,
1363 sp->hdr.seq,
1364 sp->hdr.serial,
1365 sp->hdr.flags, 0);
1366 rxrpc_post_packet_to_conn(conn, skb);
1367 goto out;
1368 }
1369
1370 call = rcu_dereference(chan->call);
1371
1372 if (sp->hdr.callNumber > chan->call_id) {
1373 if (rxrpc_to_client(sp))
1374 goto reject_packet;
1375 if (call)
1376 rxrpc_input_implicit_end_call(rx, conn, call);
1377 call = NULL;
1378 }
1379
1380 if (call) {
1381 if (sp->hdr.serviceId != call->service_id)
1382 call->service_id = sp->hdr.serviceId;
1383 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1384 call->rx_serial = sp->hdr.serial;
1385 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1386 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1387 }
1388 }
1389
1390 if (!call || atomic_read(&call->usage) == 0) {
1391 if (rxrpc_to_client(sp) ||
1392 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1393 goto bad_message;
1394 if (sp->hdr.seq != 1)
1395 goto discard;
1396 call = rxrpc_new_incoming_call(local, rx, skb);
1397 if (!call)
1398 goto reject_packet;
1399 rxrpc_send_ping(call, skb);
1400 mutex_unlock(&call->user_mutex);
1401 }
1402
1403 /* Process a call packet; this either discards or passes on the ref
1404 * elsewhere.
1405 */
1406 rxrpc_input_call_packet(call, skb);
1407 goto out;
1408
1409discard:
1410 rxrpc_free_skb(skb, rxrpc_skb_freed);
1411out:
1412 trace_rxrpc_rx_done(0, 0);
1413 return 0;
1414
1415wrong_security:
1416 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1417 RXKADINCONSISTENCY, EBADMSG);
1418 skb->priority = RXKADINCONSISTENCY;
1419 goto post_abort;
1420
1421unsupported_service:
1422 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1423 RX_INVALID_OPERATION, EOPNOTSUPP);
1424 skb->priority = RX_INVALID_OPERATION;
1425 goto post_abort;
1426
1427reupgrade:
1428 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1429 RX_PROTOCOL_ERROR, EBADMSG);
1430 goto protocol_error;
1431
1432bad_message:
1433 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1434 RX_PROTOCOL_ERROR, EBADMSG);
1435protocol_error:
1436 skb->priority = RX_PROTOCOL_ERROR;
1437post_abort:
1438 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1439reject_packet:
1440 trace_rxrpc_rx_done(skb->mark, skb->priority);
1441 rxrpc_reject_packet(local, skb);
1442 _leave(" [badmsg]");
1443 return 0;
1444}