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/* 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 if (cumulative_acks > 255)
49 cumulative_acks = 255;
50
51 summary->cwnd = call->cong_cwnd;
52 summary->ssthresh = call->cong_ssthresh;
53 summary->cumulative_acks = cumulative_acks;
54 summary->dup_acks = call->cong_dup_acks;
55
56 switch (call->cong_mode) {
57 case RXRPC_CALL_SLOW_START:
58 if (summary->saw_nacks)
59 goto packet_loss_detected;
60 if (summary->cumulative_acks > 0)
61 cwnd += 1;
62 if (cwnd >= call->cong_ssthresh) {
63 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
64 call->cong_tstamp = skb->tstamp;
65 }
66 goto out;
67
68 case RXRPC_CALL_CONGEST_AVOIDANCE:
69 if (summary->saw_nacks)
70 goto packet_loss_detected;
71
72 /* We analyse the number of packets that get ACK'd per RTT
73 * period and increase the window if we managed to fill it.
74 */
75 if (call->peer->rtt_count == 0)
76 goto out;
77 if (ktime_before(skb->tstamp,
78 ktime_add_us(call->cong_tstamp,
79 call->peer->srtt_us >> 3)))
80 goto out_no_clear_ca;
81 change = rxrpc_cong_rtt_window_end;
82 call->cong_tstamp = skb->tstamp;
83 if (cumulative_acks >= cwnd)
84 cwnd++;
85 goto out;
86
87 case RXRPC_CALL_PACKET_LOSS:
88 if (!summary->saw_nacks)
89 goto resume_normality;
90
91 if (summary->new_low_nack) {
92 change = rxrpc_cong_new_low_nack;
93 call->cong_dup_acks = 1;
94 if (call->cong_extra > 1)
95 call->cong_extra = 1;
96 goto send_extra_data;
97 }
98
99 call->cong_dup_acks++;
100 if (call->cong_dup_acks < 3)
101 goto send_extra_data;
102
103 change = rxrpc_cong_begin_retransmission;
104 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
105 call->cong_ssthresh = max_t(unsigned int,
106 summary->flight_size / 2, 2);
107 cwnd = call->cong_ssthresh + 3;
108 call->cong_extra = 0;
109 call->cong_dup_acks = 0;
110 resend = true;
111 goto out;
112
113 case RXRPC_CALL_FAST_RETRANSMIT:
114 if (!summary->new_low_nack) {
115 if (summary->nr_new_acks == 0)
116 cwnd += 1;
117 call->cong_dup_acks++;
118 if (call->cong_dup_acks == 2) {
119 change = rxrpc_cong_retransmit_again;
120 call->cong_dup_acks = 0;
121 resend = true;
122 }
123 } else {
124 change = rxrpc_cong_progress;
125 cwnd = call->cong_ssthresh;
126 if (!summary->saw_nacks)
127 goto resume_normality;
128 }
129 goto out;
130
131 default:
132 BUG();
133 goto out;
134 }
135
136resume_normality:
137 change = rxrpc_cong_cleared_nacks;
138 call->cong_dup_acks = 0;
139 call->cong_extra = 0;
140 call->cong_tstamp = skb->tstamp;
141 if (cwnd < call->cong_ssthresh)
142 call->cong_mode = RXRPC_CALL_SLOW_START;
143 else
144 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
145out:
146 cumulative_acks = 0;
147out_no_clear_ca:
148 if (cwnd >= RXRPC_TX_MAX_WINDOW)
149 cwnd = RXRPC_TX_MAX_WINDOW;
150 call->cong_cwnd = cwnd;
151 call->cong_cumul_acks = cumulative_acks;
152 summary->mode = call->cong_mode;
153 trace_rxrpc_congest(call, summary, acked_serial, change);
154 if (resend)
155 rxrpc_resend(call, skb);
156 return;
157
158packet_loss_detected:
159 change = rxrpc_cong_saw_nack;
160 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
161 call->cong_dup_acks = 0;
162 goto send_extra_data;
163
164send_extra_data:
165 /* Send some previously unsent DATA if we have some to advance the ACK
166 * state.
167 */
168 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
169 summary->nr_acks != call->tx_top - call->acks_hard_ack) {
170 call->cong_extra++;
171 wake_up(&call->waitq);
172 }
173 goto out_no_clear_ca;
174}
175
176/*
177 * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
178 */
179void rxrpc_congestion_degrade(struct rxrpc_call *call)
180{
181 ktime_t rtt, now;
182
183 if (call->cong_mode != RXRPC_CALL_SLOW_START &&
184 call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
185 return;
186 if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY)
187 return;
188
189 rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
190 now = ktime_get_real();
191 if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
192 return;
193
194 trace_rxrpc_reset_cwnd(call, now);
195 rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
196 call->tx_last_sent = now;
197 call->cong_mode = RXRPC_CALL_SLOW_START;
198 call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
199 call->cong_cwnd * 3 / 4);
200 call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
201}
202
203/*
204 * Apply a hard ACK by advancing the Tx window.
205 */
206static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
207 struct rxrpc_ack_summary *summary)
208{
209 struct rxrpc_txbuf *txb;
210 bool rot_last = false;
211
212 list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
213 if (before_eq(txb->seq, call->acks_hard_ack))
214 continue;
215 if (test_bit(RXRPC_TXBUF_LAST, &txb->flags)) {
216 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
217 rot_last = true;
218 }
219 if (txb->seq == to)
220 break;
221 }
222
223 if (rot_last)
224 set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
225
226 _enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
227
228 if (call->acks_lowest_nak == call->acks_hard_ack) {
229 call->acks_lowest_nak = to;
230 } else if (after(to, call->acks_lowest_nak)) {
231 summary->new_low_nack = true;
232 call->acks_lowest_nak = to;
233 }
234
235 smp_store_release(&call->acks_hard_ack, to);
236
237 trace_rxrpc_txqueue(call, (rot_last ?
238 rxrpc_txqueue_rotate_last :
239 rxrpc_txqueue_rotate));
240 wake_up(&call->waitq);
241 return rot_last;
242}
243
244/*
245 * End the transmission phase of a call.
246 *
247 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
248 * or a final ACK packet.
249 */
250static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
251 enum rxrpc_abort_reason abort_why)
252{
253 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
254
255 if (unlikely(call->cong_last_nack)) {
256 rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
257 call->cong_last_nack = NULL;
258 }
259
260 switch (__rxrpc_call_state(call)) {
261 case RXRPC_CALL_CLIENT_SEND_REQUEST:
262 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
263 if (reply_begun) {
264 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
265 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
266 break;
267 }
268
269 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
270 trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
271 break;
272
273 case RXRPC_CALL_SERVER_AWAIT_ACK:
274 rxrpc_call_completed(call);
275 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
276 break;
277
278 default:
279 kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
280 rxrpc_proto_abort(call, call->tx_top, abort_why);
281 break;
282 }
283}
284
285/*
286 * Begin the reply reception phase of a call.
287 */
288static bool rxrpc_receiving_reply(struct rxrpc_call *call)
289{
290 struct rxrpc_ack_summary summary = { 0 };
291 unsigned long now, timo;
292 rxrpc_seq_t top = READ_ONCE(call->tx_top);
293
294 if (call->ackr_reason) {
295 now = jiffies;
296 timo = now + MAX_JIFFY_OFFSET;
297
298 WRITE_ONCE(call->delay_ack_at, timo);
299 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
300 }
301
302 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
303 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
304 rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
305 return false;
306 }
307 }
308
309 rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
310 return true;
311}
312
313/*
314 * End the packet reception phase.
315 */
316static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
317{
318 rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
319
320 _enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
321
322 trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
323
324 switch (__rxrpc_call_state(call)) {
325 case RXRPC_CALL_CLIENT_RECV_REPLY:
326 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
327 rxrpc_call_completed(call);
328 break;
329
330 case RXRPC_CALL_SERVER_RECV_REQUEST:
331 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
332 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
333 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
334 break;
335
336 default:
337 break;
338 }
339}
340
341static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
342 rxrpc_seq_t window, rxrpc_seq_t wtop)
343{
344 call->ackr_window = window;
345 call->ackr_wtop = wtop;
346}
347
348/*
349 * Push a DATA packet onto the Rx queue.
350 */
351static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
352 rxrpc_seq_t window, rxrpc_seq_t wtop,
353 enum rxrpc_receive_trace why)
354{
355 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
356 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
357
358 __skb_queue_tail(&call->recvmsg_queue, skb);
359 rxrpc_input_update_ack_window(call, window, wtop);
360 trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
361 if (last)
362 rxrpc_end_rx_phase(call, sp->hdr.serial);
363}
364
365/*
366 * Process a DATA packet.
367 */
368static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
369 bool *_notify)
370{
371 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
372 struct sk_buff *oos;
373 rxrpc_serial_t serial = sp->hdr.serial;
374 unsigned int sack = call->ackr_sack_base;
375 rxrpc_seq_t window = call->ackr_window;
376 rxrpc_seq_t wtop = call->ackr_wtop;
377 rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
378 rxrpc_seq_t seq = sp->hdr.seq;
379 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
380 int ack_reason = -1;
381
382 rxrpc_inc_stat(call->rxnet, stat_rx_data);
383 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
384 rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
385 if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
386 rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
387
388 if (last) {
389 if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
390 seq + 1 != wtop)
391 return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
392 } else {
393 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
394 after_eq(seq, wtop)) {
395 pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
396 call->debug_id, seq, window, wtop, wlimit);
397 return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
398 }
399 }
400
401 if (after(seq, call->rx_highest_seq))
402 call->rx_highest_seq = seq;
403
404 trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
405
406 if (before(seq, window)) {
407 ack_reason = RXRPC_ACK_DUPLICATE;
408 goto send_ack;
409 }
410 if (after(seq, wlimit)) {
411 ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
412 goto send_ack;
413 }
414
415 /* Queue the packet. */
416 if (seq == window) {
417 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
418 ack_reason = RXRPC_ACK_REQUESTED;
419 /* Send an immediate ACK if we fill in a hole */
420 else if (!skb_queue_empty(&call->rx_oos_queue))
421 ack_reason = RXRPC_ACK_DELAY;
422 else
423 call->ackr_nr_unacked++;
424
425 window++;
426 if (after(window, wtop)) {
427 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_none);
428 wtop = window;
429 } else {
430 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_advance);
431 sack = (sack + 1) % RXRPC_SACK_SIZE;
432 }
433
434
435 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
436
437 spin_lock(&call->recvmsg_queue.lock);
438 rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
439 *_notify = true;
440
441 while ((oos = skb_peek(&call->rx_oos_queue))) {
442 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
443
444 if (after(osp->hdr.seq, window))
445 break;
446
447 __skb_unlink(oos, &call->rx_oos_queue);
448 last = osp->hdr.flags & RXRPC_LAST_PACKET;
449 seq = osp->hdr.seq;
450 call->ackr_sack_table[sack] = 0;
451 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_fill);
452 sack = (sack + 1) % RXRPC_SACK_SIZE;
453
454 window++;
455 rxrpc_input_queue_data(call, oos, window, wtop,
456 rxrpc_receive_queue_oos);
457 }
458
459 spin_unlock(&call->recvmsg_queue.lock);
460
461 call->ackr_sack_base = sack;
462 } else {
463 unsigned int slot;
464
465 ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
466
467 slot = seq - window;
468 sack = (sack + slot) % RXRPC_SACK_SIZE;
469
470 if (call->ackr_sack_table[sack % RXRPC_SACK_SIZE]) {
471 ack_reason = RXRPC_ACK_DUPLICATE;
472 goto send_ack;
473 }
474
475 call->ackr_sack_table[sack % RXRPC_SACK_SIZE] |= 1;
476 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_oos);
477
478 if (after(seq + 1, wtop)) {
479 wtop = seq + 1;
480 rxrpc_input_update_ack_window(call, window, wtop);
481 }
482
483 skb_queue_walk(&call->rx_oos_queue, oos) {
484 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
485
486 if (after(osp->hdr.seq, seq)) {
487 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
488 __skb_queue_before(&call->rx_oos_queue, oos, skb);
489 goto oos_queued;
490 }
491 }
492
493 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
494 __skb_queue_tail(&call->rx_oos_queue, skb);
495 oos_queued:
496 trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
497 sp->hdr.serial, sp->hdr.seq);
498 }
499
500send_ack:
501 if (ack_reason >= 0)
502 rxrpc_send_ACK(call, ack_reason, serial,
503 rxrpc_propose_ack_input_data);
504 else
505 rxrpc_propose_delay_ACK(call, serial,
506 rxrpc_propose_ack_input_data);
507}
508
509/*
510 * Split a jumbo packet and file the bits separately.
511 */
512static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
513{
514 struct rxrpc_jumbo_header jhdr;
515 struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
516 struct sk_buff *jskb;
517 unsigned int offset = sizeof(struct rxrpc_wire_header);
518 unsigned int len = skb->len - offset;
519 bool notify = false;
520
521 while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
522 if (len < RXRPC_JUMBO_SUBPKTLEN)
523 goto protocol_error;
524 if (sp->hdr.flags & RXRPC_LAST_PACKET)
525 goto protocol_error;
526 if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
527 &jhdr, sizeof(jhdr)) < 0)
528 goto protocol_error;
529
530 jskb = skb_clone(skb, GFP_NOFS);
531 if (!jskb) {
532 kdebug("couldn't clone");
533 return false;
534 }
535 rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
536 jsp = rxrpc_skb(jskb);
537 jsp->offset = offset;
538 jsp->len = RXRPC_JUMBO_DATALEN;
539 rxrpc_input_data_one(call, jskb, ¬ify);
540 rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
541
542 sp->hdr.flags = jhdr.flags;
543 sp->hdr._rsvd = ntohs(jhdr._rsvd);
544 sp->hdr.seq++;
545 sp->hdr.serial++;
546 offset += RXRPC_JUMBO_SUBPKTLEN;
547 len -= RXRPC_JUMBO_SUBPKTLEN;
548 }
549
550 sp->offset = offset;
551 sp->len = len;
552 rxrpc_input_data_one(call, skb, ¬ify);
553 if (notify) {
554 trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
555 rxrpc_notify_socket(call);
556 }
557 return true;
558
559protocol_error:
560 return false;
561}
562
563/*
564 * Process a DATA packet, adding the packet to the Rx ring. The caller's
565 * packet ref must be passed on or discarded.
566 */
567static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
568{
569 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
570 rxrpc_serial_t serial = sp->hdr.serial;
571 rxrpc_seq_t seq0 = sp->hdr.seq;
572
573 _enter("{%x,%x,%x},{%u,%x}",
574 call->ackr_window, call->ackr_wtop, call->rx_highest_seq,
575 skb->len, seq0);
576
577 if (__rxrpc_call_is_complete(call))
578 return;
579
580 switch (__rxrpc_call_state(call)) {
581 case RXRPC_CALL_CLIENT_SEND_REQUEST:
582 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
583 /* Received data implicitly ACKs all of the request
584 * packets we sent when we're acting as a client.
585 */
586 if (!rxrpc_receiving_reply(call))
587 goto out_notify;
588 break;
589
590 case RXRPC_CALL_SERVER_RECV_REQUEST: {
591 unsigned long timo = READ_ONCE(call->next_req_timo);
592 unsigned long now, expect_req_by;
593
594 if (timo) {
595 now = jiffies;
596 expect_req_by = now + timo;
597 WRITE_ONCE(call->expect_req_by, expect_req_by);
598 rxrpc_reduce_call_timer(call, expect_req_by, now,
599 rxrpc_timer_set_for_idle);
600 }
601 break;
602 }
603
604 default:
605 break;
606 }
607
608 if (!rxrpc_input_split_jumbo(call, skb)) {
609 rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
610 goto out_notify;
611 }
612 return;
613
614out_notify:
615 trace_rxrpc_notify_socket(call->debug_id, serial);
616 rxrpc_notify_socket(call);
617 _leave(" [queued]");
618}
619
620/*
621 * See if there's a cached RTT probe to complete.
622 */
623static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
624 ktime_t resp_time,
625 rxrpc_serial_t acked_serial,
626 rxrpc_serial_t ack_serial,
627 enum rxrpc_rtt_rx_trace type)
628{
629 rxrpc_serial_t orig_serial;
630 unsigned long avail;
631 ktime_t sent_at;
632 bool matched = false;
633 int i;
634
635 avail = READ_ONCE(call->rtt_avail);
636 smp_rmb(); /* Read avail bits before accessing data. */
637
638 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
639 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
640 continue;
641
642 sent_at = call->rtt_sent_at[i];
643 orig_serial = call->rtt_serial[i];
644
645 if (orig_serial == acked_serial) {
646 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
647 smp_mb(); /* Read data before setting avail bit */
648 set_bit(i, &call->rtt_avail);
649 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
650 sent_at, resp_time);
651 matched = true;
652 }
653
654 /* If a later serial is being acked, then mark this slot as
655 * being available.
656 */
657 if (after(acked_serial, orig_serial)) {
658 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
659 orig_serial, acked_serial, 0, 0);
660 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
661 smp_wmb();
662 set_bit(i, &call->rtt_avail);
663 }
664 }
665
666 if (!matched)
667 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
668}
669
670/*
671 * Process the extra information that may be appended to an ACK packet
672 */
673static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
674 struct rxrpc_ackinfo *ackinfo)
675{
676 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
677 struct rxrpc_peer *peer;
678 unsigned int mtu;
679 bool wake = false;
680 u32 rwind = ntohl(ackinfo->rwind);
681
682 if (rwind > RXRPC_TX_MAX_WINDOW)
683 rwind = RXRPC_TX_MAX_WINDOW;
684 if (call->tx_winsize != rwind) {
685 if (rwind > call->tx_winsize)
686 wake = true;
687 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
688 call->tx_winsize = rwind;
689 }
690
691 if (call->cong_ssthresh > rwind)
692 call->cong_ssthresh = rwind;
693
694 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
695
696 peer = call->peer;
697 if (mtu < peer->maxdata) {
698 spin_lock(&peer->lock);
699 peer->maxdata = mtu;
700 peer->mtu = mtu + peer->hdrsize;
701 spin_unlock(&peer->lock);
702 }
703
704 if (wake)
705 wake_up(&call->waitq);
706}
707
708/*
709 * Determine how many nacks from the previous ACK have now been satisfied.
710 */
711static rxrpc_seq_t rxrpc_input_check_prev_ack(struct rxrpc_call *call,
712 struct rxrpc_ack_summary *summary,
713 rxrpc_seq_t seq)
714{
715 struct sk_buff *skb = call->cong_last_nack;
716 struct rxrpc_ackpacket ack;
717 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
718 unsigned int i, new_acks = 0, retained_nacks = 0;
719 rxrpc_seq_t old_seq = sp->first_ack;
720 u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(ack);
721
722 if (after_eq(seq, old_seq + sp->nr_acks)) {
723 summary->nr_new_acks += sp->nr_nacks;
724 summary->nr_new_acks += seq - (old_seq + sp->nr_acks);
725 summary->nr_retained_nacks = 0;
726 } else if (seq == old_seq) {
727 summary->nr_retained_nacks = sp->nr_nacks;
728 } else {
729 for (i = 0; i < sp->nr_acks; i++) {
730 if (acks[i] == RXRPC_ACK_TYPE_NACK) {
731 if (before(old_seq + i, seq))
732 new_acks++;
733 else
734 retained_nacks++;
735 }
736 }
737
738 summary->nr_new_acks += new_acks;
739 summary->nr_retained_nacks = retained_nacks;
740 }
741
742 return old_seq + sp->nr_acks;
743}
744
745/*
746 * Process individual soft ACKs.
747 *
748 * Each ACK in the array corresponds to one packet and can be either an ACK or
749 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
750 * packets that lie beyond the end of the ACK list are scheduled for resend by
751 * the timer on the basis that the peer might just not have processed them at
752 * the time the ACK was sent.
753 */
754static void rxrpc_input_soft_acks(struct rxrpc_call *call,
755 struct rxrpc_ack_summary *summary,
756 struct sk_buff *skb,
757 rxrpc_seq_t seq,
758 rxrpc_seq_t since)
759{
760 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
761 unsigned int i, old_nacks = 0;
762 rxrpc_seq_t lowest_nak = seq + sp->nr_acks;
763 u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
764
765 for (i = 0; i < sp->nr_acks; i++) {
766 if (acks[i] == RXRPC_ACK_TYPE_ACK) {
767 summary->nr_acks++;
768 if (after_eq(seq, since))
769 summary->nr_new_acks++;
770 } else {
771 summary->saw_nacks = true;
772 if (before(seq, since)) {
773 /* Overlap with previous ACK */
774 old_nacks++;
775 } else {
776 summary->nr_new_nacks++;
777 sp->nr_nacks++;
778 }
779
780 if (before(seq, lowest_nak))
781 lowest_nak = seq;
782 }
783 seq++;
784 }
785
786 if (lowest_nak != call->acks_lowest_nak) {
787 call->acks_lowest_nak = lowest_nak;
788 summary->new_low_nack = true;
789 }
790
791 /* We *can* have more nacks than we did - the peer is permitted to drop
792 * packets it has soft-acked and re-request them. Further, it is
793 * possible for the nack distribution to change whilst the number of
794 * nacks stays the same or goes down.
795 */
796 if (old_nacks < summary->nr_retained_nacks)
797 summary->nr_new_acks += summary->nr_retained_nacks - old_nacks;
798 summary->nr_retained_nacks = old_nacks;
799}
800
801/*
802 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
803 * with respect to the ack state conveyed by preceding ACKs.
804 */
805static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
806 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
807{
808 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
809
810 if (after(first_pkt, base))
811 return true; /* The window advanced */
812
813 if (before(first_pkt, base))
814 return false; /* firstPacket regressed */
815
816 if (after_eq(prev_pkt, call->acks_prev_seq))
817 return true; /* previousPacket hasn't regressed. */
818
819 /* Some rx implementations put a serial number in previousPacket. */
820 if (after_eq(prev_pkt, base + call->tx_winsize))
821 return false;
822 return true;
823}
824
825/*
826 * Process an ACK packet.
827 *
828 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
829 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
830 *
831 * A hard-ACK means that a packet has been processed and may be discarded; a
832 * soft-ACK means that the packet may be discarded and retransmission
833 * requested. A phase is complete when all packets are hard-ACK'd.
834 */
835static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
836{
837 struct rxrpc_ack_summary summary = { 0 };
838 struct rxrpc_ackpacket ack;
839 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
840 struct rxrpc_ackinfo info;
841 rxrpc_serial_t ack_serial, acked_serial;
842 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt, since;
843 int nr_acks, offset, ioffset;
844
845 _enter("");
846
847 offset = sizeof(struct rxrpc_wire_header);
848 if (skb_copy_bits(skb, offset, &ack, sizeof(ack)) < 0)
849 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack);
850 offset += sizeof(ack);
851
852 ack_serial = sp->hdr.serial;
853 acked_serial = ntohl(ack.serial);
854 first_soft_ack = ntohl(ack.firstPacket);
855 prev_pkt = ntohl(ack.previousPacket);
856 hard_ack = first_soft_ack - 1;
857 nr_acks = ack.nAcks;
858 sp->first_ack = first_soft_ack;
859 sp->nr_acks = nr_acks;
860 summary.ack_reason = (ack.reason < RXRPC_ACK__INVALID ?
861 ack.reason : RXRPC_ACK__INVALID);
862
863 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
864 first_soft_ack, prev_pkt,
865 summary.ack_reason, nr_acks);
866 rxrpc_inc_stat(call->rxnet, stat_rx_acks[ack.reason]);
867
868 if (acked_serial != 0) {
869 switch (ack.reason) {
870 case RXRPC_ACK_PING_RESPONSE:
871 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
872 rxrpc_rtt_rx_ping_response);
873 break;
874 case RXRPC_ACK_REQUESTED:
875 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
876 rxrpc_rtt_rx_requested_ack);
877 break;
878 default:
879 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
880 rxrpc_rtt_rx_other_ack);
881 break;
882 }
883 }
884
885 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
886 * indicates that the client address changed due to NAT. The server
887 * lost the call because it switched to a different peer.
888 */
889 if (unlikely(ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
890 first_soft_ack == 1 &&
891 prev_pkt == 0 &&
892 rxrpc_is_client_call(call)) {
893 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
894 0, -ENETRESET);
895 goto send_response;
896 }
897
898 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
899 * indicate a change of address. However, we can retransmit the call
900 * if we still have it buffered to the beginning.
901 */
902 if (unlikely(ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
903 first_soft_ack == 1 &&
904 prev_pkt == 0 &&
905 call->acks_hard_ack == 0 &&
906 rxrpc_is_client_call(call)) {
907 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
908 0, -ENETRESET);
909 goto send_response;
910 }
911
912 /* Discard any out-of-order or duplicate ACKs (outside lock). */
913 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
914 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
915 first_soft_ack, call->acks_first_seq,
916 prev_pkt, call->acks_prev_seq);
917 goto send_response;
918 }
919
920 info.rxMTU = 0;
921 ioffset = offset + nr_acks + 3;
922 if (skb->len >= ioffset + sizeof(info) &&
923 skb_copy_bits(skb, ioffset, &info, sizeof(info)) < 0)
924 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_info);
925
926 if (nr_acks > 0)
927 skb_condense(skb);
928
929 if (call->cong_last_nack) {
930 since = rxrpc_input_check_prev_ack(call, &summary, first_soft_ack);
931 rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
932 call->cong_last_nack = NULL;
933 } else {
934 summary.nr_new_acks = first_soft_ack - call->acks_first_seq;
935 call->acks_lowest_nak = first_soft_ack + nr_acks;
936 since = first_soft_ack;
937 }
938
939 call->acks_latest_ts = skb->tstamp;
940 call->acks_first_seq = first_soft_ack;
941 call->acks_prev_seq = prev_pkt;
942
943 switch (ack.reason) {
944 case RXRPC_ACK_PING:
945 break;
946 default:
947 if (acked_serial && after(acked_serial, call->acks_highest_serial))
948 call->acks_highest_serial = acked_serial;
949 break;
950 }
951
952 /* Parse rwind and mtu sizes if provided. */
953 if (info.rxMTU)
954 rxrpc_input_ackinfo(call, skb, &info);
955
956 if (first_soft_ack == 0)
957 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
958
959 /* Ignore ACKs unless we are or have just been transmitting. */
960 switch (__rxrpc_call_state(call)) {
961 case RXRPC_CALL_CLIENT_SEND_REQUEST:
962 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
963 case RXRPC_CALL_SERVER_SEND_REPLY:
964 case RXRPC_CALL_SERVER_AWAIT_ACK:
965 break;
966 default:
967 goto send_response;
968 }
969
970 if (before(hard_ack, call->acks_hard_ack) ||
971 after(hard_ack, call->tx_top))
972 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
973 if (nr_acks > call->tx_top - hard_ack)
974 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
975
976 if (after(hard_ack, call->acks_hard_ack)) {
977 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
978 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
979 goto send_response;
980 }
981 }
982
983 if (nr_acks > 0) {
984 if (offset > (int)skb->len - nr_acks)
985 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
986 rxrpc_input_soft_acks(call, &summary, skb, first_soft_ack, since);
987 rxrpc_get_skb(skb, rxrpc_skb_get_last_nack);
988 call->cong_last_nack = skb;
989 }
990
991 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
992 summary.nr_acks == call->tx_top - hard_ack &&
993 rxrpc_is_client_call(call))
994 rxrpc_propose_ping(call, ack_serial,
995 rxrpc_propose_ack_ping_for_lost_reply);
996
997 rxrpc_congestion_management(call, skb, &summary, acked_serial);
998
999send_response:
1000 if (ack.reason == RXRPC_ACK_PING)
1001 rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
1002 rxrpc_propose_ack_respond_to_ping);
1003 else if (sp->hdr.flags & RXRPC_REQUEST_ACK)
1004 rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
1005 rxrpc_propose_ack_respond_to_ack);
1006}
1007
1008/*
1009 * Process an ACKALL packet.
1010 */
1011static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1012{
1013 struct rxrpc_ack_summary summary = { 0 };
1014
1015 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1016 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
1017}
1018
1019/*
1020 * Process an ABORT packet directed at a call.
1021 */
1022static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1023{
1024 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1025
1026 trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
1027
1028 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1029 skb->priority, -ECONNABORTED);
1030}
1031
1032/*
1033 * Process an incoming call packet.
1034 */
1035void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
1036{
1037 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1038 unsigned long timo;
1039
1040 _enter("%p,%p", call, skb);
1041
1042 if (sp->hdr.serviceId != call->dest_srx.srx_service)
1043 call->dest_srx.srx_service = sp->hdr.serviceId;
1044 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1045 call->rx_serial = sp->hdr.serial;
1046 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1047 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1048
1049 timo = READ_ONCE(call->next_rx_timo);
1050 if (timo) {
1051 unsigned long now = jiffies, expect_rx_by;
1052
1053 expect_rx_by = now + timo;
1054 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1055 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1056 rxrpc_timer_set_for_normal);
1057 }
1058
1059 switch (sp->hdr.type) {
1060 case RXRPC_PACKET_TYPE_DATA:
1061 return rxrpc_input_data(call, skb);
1062
1063 case RXRPC_PACKET_TYPE_ACK:
1064 return rxrpc_input_ack(call, skb);
1065
1066 case RXRPC_PACKET_TYPE_BUSY:
1067 /* Just ignore BUSY packets from the server; the retry and
1068 * lifespan timers will take care of business. BUSY packets
1069 * from the client don't make sense.
1070 */
1071 return;
1072
1073 case RXRPC_PACKET_TYPE_ABORT:
1074 return rxrpc_input_abort(call, skb);
1075
1076 case RXRPC_PACKET_TYPE_ACKALL:
1077 return rxrpc_input_ackall(call, skb);
1078
1079 default:
1080 break;
1081 }
1082}
1083
1084/*
1085 * Handle a new service call on a channel implicitly completing the preceding
1086 * call on that channel. This does not apply to client conns.
1087 *
1088 * TODO: If callNumber > call_id + 1, renegotiate security.
1089 */
1090void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1091{
1092 switch (__rxrpc_call_state(call)) {
1093 case RXRPC_CALL_SERVER_AWAIT_ACK:
1094 rxrpc_call_completed(call);
1095 fallthrough;
1096 case RXRPC_CALL_COMPLETE:
1097 break;
1098 default:
1099 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1100 rxrpc_eproto_improper_term);
1101 trace_rxrpc_improper_term(call);
1102 break;
1103 }
1104
1105 rxrpc_input_call_event(call, skb);
1106}