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