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
12 /* Override priority when generating ACKs for received DATA */
13 static const u8 rxrpc_ack_priority[RXRPC_ACK__INVALID] = {
14 [RXRPC_ACK_IDLE] = 1,
15 [RXRPC_ACK_DELAY] = 2,
16 [RXRPC_ACK_REQUESTED] = 3,
17 [RXRPC_ACK_DUPLICATE] = 4,
18 [RXRPC_ACK_EXCEEDS_WINDOW] = 5,
19 [RXRPC_ACK_NOSPACE] = 6,
20 [RXRPC_ACK_OUT_OF_SEQUENCE] = 7,
21 };
22
rxrpc_proto_abort(struct rxrpc_call * call,rxrpc_seq_t seq,enum rxrpc_abort_reason why)23 static void rxrpc_proto_abort(struct rxrpc_call *call, rxrpc_seq_t seq,
24 enum rxrpc_abort_reason why)
25 {
26 rxrpc_abort_call(call, seq, RX_PROTOCOL_ERROR, -EBADMSG, why);
27 }
28
29 /*
30 * Do TCP-style congestion management [RFC 5681].
31 */
rxrpc_congestion_management(struct rxrpc_call * call,struct sk_buff * skb,struct rxrpc_ack_summary * summary,rxrpc_serial_t acked_serial)32 static void rxrpc_congestion_management(struct rxrpc_call *call,
33 struct sk_buff *skb,
34 struct rxrpc_ack_summary *summary,
35 rxrpc_serial_t acked_serial)
36 {
37 enum rxrpc_congest_change change = rxrpc_cong_no_change;
38 unsigned int cumulative_acks = call->cong_cumul_acks;
39 unsigned int cwnd = call->cong_cwnd;
40 bool resend = false;
41
42 summary->flight_size =
43 (call->tx_top - call->acks_hard_ack) - summary->nr_acks;
44
45 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
46 summary->retrans_timeo = true;
47 call->cong_ssthresh = max_t(unsigned int,
48 summary->flight_size / 2, 2);
49 cwnd = 1;
50 if (cwnd >= call->cong_ssthresh &&
51 call->cong_mode == RXRPC_CALL_SLOW_START) {
52 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
53 call->cong_tstamp = skb->tstamp;
54 cumulative_acks = 0;
55 }
56 }
57
58 cumulative_acks += summary->nr_new_acks;
59 if (cumulative_acks > 255)
60 cumulative_acks = 255;
61
62 summary->cwnd = call->cong_cwnd;
63 summary->ssthresh = call->cong_ssthresh;
64 summary->cumulative_acks = cumulative_acks;
65 summary->dup_acks = call->cong_dup_acks;
66
67 switch (call->cong_mode) {
68 case RXRPC_CALL_SLOW_START:
69 if (summary->saw_nacks)
70 goto packet_loss_detected;
71 if (summary->cumulative_acks > 0)
72 cwnd += 1;
73 if (cwnd >= call->cong_ssthresh) {
74 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
75 call->cong_tstamp = skb->tstamp;
76 }
77 goto out;
78
79 case RXRPC_CALL_CONGEST_AVOIDANCE:
80 if (summary->saw_nacks)
81 goto packet_loss_detected;
82
83 /* We analyse the number of packets that get ACK'd per RTT
84 * period and increase the window if we managed to fill it.
85 */
86 if (call->peer->rtt_count == 0)
87 goto out;
88 if (ktime_before(skb->tstamp,
89 ktime_add_us(call->cong_tstamp,
90 call->peer->srtt_us >> 3)))
91 goto out_no_clear_ca;
92 change = rxrpc_cong_rtt_window_end;
93 call->cong_tstamp = skb->tstamp;
94 if (cumulative_acks >= cwnd)
95 cwnd++;
96 goto out;
97
98 case RXRPC_CALL_PACKET_LOSS:
99 if (!summary->saw_nacks)
100 goto resume_normality;
101
102 if (summary->new_low_nack) {
103 change = rxrpc_cong_new_low_nack;
104 call->cong_dup_acks = 1;
105 if (call->cong_extra > 1)
106 call->cong_extra = 1;
107 goto send_extra_data;
108 }
109
110 call->cong_dup_acks++;
111 if (call->cong_dup_acks < 3)
112 goto send_extra_data;
113
114 change = rxrpc_cong_begin_retransmission;
115 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
116 call->cong_ssthresh = max_t(unsigned int,
117 summary->flight_size / 2, 2);
118 cwnd = call->cong_ssthresh + 3;
119 call->cong_extra = 0;
120 call->cong_dup_acks = 0;
121 resend = true;
122 goto out;
123
124 case RXRPC_CALL_FAST_RETRANSMIT:
125 if (!summary->new_low_nack) {
126 if (summary->nr_new_acks == 0)
127 cwnd += 1;
128 call->cong_dup_acks++;
129 if (call->cong_dup_acks == 2) {
130 change = rxrpc_cong_retransmit_again;
131 call->cong_dup_acks = 0;
132 resend = true;
133 }
134 } else {
135 change = rxrpc_cong_progress;
136 cwnd = call->cong_ssthresh;
137 if (!summary->saw_nacks)
138 goto resume_normality;
139 }
140 goto out;
141
142 default:
143 BUG();
144 goto out;
145 }
146
147 resume_normality:
148 change = rxrpc_cong_cleared_nacks;
149 call->cong_dup_acks = 0;
150 call->cong_extra = 0;
151 call->cong_tstamp = skb->tstamp;
152 if (cwnd < call->cong_ssthresh)
153 call->cong_mode = RXRPC_CALL_SLOW_START;
154 else
155 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
156 out:
157 cumulative_acks = 0;
158 out_no_clear_ca:
159 if (cwnd >= RXRPC_TX_MAX_WINDOW)
160 cwnd = RXRPC_TX_MAX_WINDOW;
161 call->cong_cwnd = cwnd;
162 call->cong_cumul_acks = cumulative_acks;
163 summary->mode = call->cong_mode;
164 trace_rxrpc_congest(call, summary, acked_serial, change);
165 if (resend)
166 rxrpc_resend(call, skb);
167 return;
168
169 packet_loss_detected:
170 change = rxrpc_cong_saw_nack;
171 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
172 call->cong_dup_acks = 0;
173 goto send_extra_data;
174
175 send_extra_data:
176 /* Send some previously unsent DATA if we have some to advance the ACK
177 * state.
178 */
179 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
180 summary->nr_acks != call->tx_top - call->acks_hard_ack) {
181 call->cong_extra++;
182 wake_up(&call->waitq);
183 }
184 goto out_no_clear_ca;
185 }
186
187 /*
188 * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
189 */
rxrpc_congestion_degrade(struct rxrpc_call * call)190 void rxrpc_congestion_degrade(struct rxrpc_call *call)
191 {
192 ktime_t rtt, now;
193
194 if (call->cong_mode != RXRPC_CALL_SLOW_START &&
195 call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
196 return;
197 if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY)
198 return;
199
200 rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
201 now = ktime_get_real();
202 if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
203 return;
204
205 trace_rxrpc_reset_cwnd(call, now);
206 rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
207 call->tx_last_sent = now;
208 call->cong_mode = RXRPC_CALL_SLOW_START;
209 call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
210 call->cong_cwnd * 3 / 4);
211 call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
212 }
213
214 /*
215 * Apply a hard ACK by advancing the Tx window.
216 */
rxrpc_rotate_tx_window(struct rxrpc_call * call,rxrpc_seq_t to,struct rxrpc_ack_summary * summary)217 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
218 struct rxrpc_ack_summary *summary)
219 {
220 struct rxrpc_txbuf *txb;
221 bool rot_last = false;
222
223 list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
224 if (before_eq(txb->seq, call->acks_hard_ack))
225 continue;
226 if (txb->flags & RXRPC_LAST_PACKET) {
227 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
228 rot_last = true;
229 }
230 if (txb->seq == to)
231 break;
232 }
233
234 if (rot_last)
235 set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
236
237 _enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
238
239 if (call->acks_lowest_nak == call->acks_hard_ack) {
240 call->acks_lowest_nak = to;
241 } else if (after(to, call->acks_lowest_nak)) {
242 summary->new_low_nack = true;
243 call->acks_lowest_nak = to;
244 }
245
246 smp_store_release(&call->acks_hard_ack, to);
247
248 trace_rxrpc_txqueue(call, (rot_last ?
249 rxrpc_txqueue_rotate_last :
250 rxrpc_txqueue_rotate));
251 wake_up(&call->waitq);
252 return rot_last;
253 }
254
255 /*
256 * End the transmission phase of a call.
257 *
258 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
259 * or a final ACK packet.
260 */
rxrpc_end_tx_phase(struct rxrpc_call * call,bool reply_begun,enum rxrpc_abort_reason abort_why)261 static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
262 enum rxrpc_abort_reason abort_why)
263 {
264 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
265
266 call->resend_at = KTIME_MAX;
267 trace_rxrpc_timer_can(call, rxrpc_timer_trace_resend);
268
269 if (unlikely(call->cong_last_nack)) {
270 rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
271 call->cong_last_nack = NULL;
272 }
273
274 switch (__rxrpc_call_state(call)) {
275 case RXRPC_CALL_CLIENT_SEND_REQUEST:
276 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
277 if (reply_begun) {
278 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
279 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
280 break;
281 }
282
283 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
284 trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
285 break;
286
287 case RXRPC_CALL_SERVER_AWAIT_ACK:
288 rxrpc_call_completed(call);
289 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
290 break;
291
292 default:
293 kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
294 rxrpc_proto_abort(call, call->tx_top, abort_why);
295 break;
296 }
297 }
298
299 /*
300 * Begin the reply reception phase of a call.
301 */
rxrpc_receiving_reply(struct rxrpc_call * call)302 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
303 {
304 struct rxrpc_ack_summary summary = { 0 };
305 rxrpc_seq_t top = READ_ONCE(call->tx_top);
306
307 if (call->ackr_reason) {
308 call->delay_ack_at = KTIME_MAX;
309 trace_rxrpc_timer_can(call, rxrpc_timer_trace_delayed_ack);
310 }
311
312 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
313 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
314 rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
315 return false;
316 }
317 }
318
319 rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
320 return true;
321 }
322
323 /*
324 * End the packet reception phase.
325 */
rxrpc_end_rx_phase(struct rxrpc_call * call,rxrpc_serial_t serial)326 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
327 {
328 rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
329
330 _enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
331
332 trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
333
334 switch (__rxrpc_call_state(call)) {
335 case RXRPC_CALL_CLIENT_RECV_REPLY:
336 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
337 rxrpc_call_completed(call);
338 break;
339
340 case RXRPC_CALL_SERVER_RECV_REQUEST:
341 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
342 call->expect_req_by = KTIME_MAX;
343 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
344 break;
345
346 default:
347 break;
348 }
349 }
350
rxrpc_input_update_ack_window(struct rxrpc_call * call,rxrpc_seq_t window,rxrpc_seq_t wtop)351 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
352 rxrpc_seq_t window, rxrpc_seq_t wtop)
353 {
354 call->ackr_window = window;
355 call->ackr_wtop = wtop;
356 }
357
358 /*
359 * Push a DATA packet onto the Rx queue.
360 */
rxrpc_input_queue_data(struct rxrpc_call * call,struct sk_buff * skb,rxrpc_seq_t window,rxrpc_seq_t wtop,enum rxrpc_receive_trace why)361 static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
362 rxrpc_seq_t window, rxrpc_seq_t wtop,
363 enum rxrpc_receive_trace why)
364 {
365 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
366 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
367
368 __skb_queue_tail(&call->recvmsg_queue, skb);
369 rxrpc_input_update_ack_window(call, window, wtop);
370 trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
371 if (last)
372 rxrpc_end_rx_phase(call, sp->hdr.serial);
373 }
374
375 /*
376 * Process a DATA packet.
377 */
rxrpc_input_data_one(struct rxrpc_call * call,struct sk_buff * skb,bool * _notify,rxrpc_serial_t * _ack_serial,int * _ack_reason)378 static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
379 bool *_notify, rxrpc_serial_t *_ack_serial, int *_ack_reason)
380 {
381 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
382 struct sk_buff *oos;
383 rxrpc_serial_t serial = sp->hdr.serial;
384 unsigned int sack = call->ackr_sack_base;
385 rxrpc_seq_t window = call->ackr_window;
386 rxrpc_seq_t wtop = call->ackr_wtop;
387 rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
388 rxrpc_seq_t seq = sp->hdr.seq;
389 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
390 int ack_reason = -1;
391
392 rxrpc_inc_stat(call->rxnet, stat_rx_data);
393 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
394 rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
395 if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
396 rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
397
398 if (last) {
399 if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
400 seq + 1 != wtop)
401 return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
402 } else {
403 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
404 after_eq(seq, wtop)) {
405 pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
406 call->debug_id, seq, window, wtop, wlimit);
407 return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
408 }
409 }
410
411 if (after(seq, call->rx_highest_seq))
412 call->rx_highest_seq = seq;
413
414 trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
415
416 if (before(seq, window)) {
417 ack_reason = RXRPC_ACK_DUPLICATE;
418 goto send_ack;
419 }
420 if (after(seq, wlimit)) {
421 ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
422 goto send_ack;
423 }
424
425 /* Queue the packet. */
426 if (seq == window) {
427 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
428 ack_reason = RXRPC_ACK_REQUESTED;
429 /* Send an immediate ACK if we fill in a hole */
430 else if (!skb_queue_empty(&call->rx_oos_queue))
431 ack_reason = RXRPC_ACK_DELAY;
432
433 window++;
434 if (after(window, wtop)) {
435 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_none);
436 wtop = window;
437 } else {
438 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_advance);
439 sack = (sack + 1) % RXRPC_SACK_SIZE;
440 }
441
442
443 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
444
445 spin_lock(&call->recvmsg_queue.lock);
446 rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
447 *_notify = true;
448
449 while ((oos = skb_peek(&call->rx_oos_queue))) {
450 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
451
452 if (after(osp->hdr.seq, window))
453 break;
454
455 __skb_unlink(oos, &call->rx_oos_queue);
456 last = osp->hdr.flags & RXRPC_LAST_PACKET;
457 seq = osp->hdr.seq;
458 call->ackr_sack_table[sack] = 0;
459 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_fill);
460 sack = (sack + 1) % RXRPC_SACK_SIZE;
461
462 window++;
463 rxrpc_input_queue_data(call, oos, window, wtop,
464 rxrpc_receive_queue_oos);
465 }
466
467 spin_unlock(&call->recvmsg_queue.lock);
468
469 call->ackr_sack_base = sack;
470 } else {
471 unsigned int slot;
472
473 ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
474
475 slot = seq - window;
476 sack = (sack + slot) % RXRPC_SACK_SIZE;
477
478 if (call->ackr_sack_table[sack % RXRPC_SACK_SIZE]) {
479 ack_reason = RXRPC_ACK_DUPLICATE;
480 goto send_ack;
481 }
482
483 call->ackr_sack_table[sack % RXRPC_SACK_SIZE] |= 1;
484 trace_rxrpc_sack(call, seq, sack, rxrpc_sack_oos);
485
486 if (after(seq + 1, wtop)) {
487 wtop = seq + 1;
488 rxrpc_input_update_ack_window(call, window, wtop);
489 }
490
491 skb_queue_walk(&call->rx_oos_queue, oos) {
492 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
493
494 if (after(osp->hdr.seq, seq)) {
495 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
496 __skb_queue_before(&call->rx_oos_queue, oos, skb);
497 goto oos_queued;
498 }
499 }
500
501 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
502 __skb_queue_tail(&call->rx_oos_queue, skb);
503 oos_queued:
504 trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
505 sp->hdr.serial, sp->hdr.seq);
506 }
507
508 send_ack:
509 if (ack_reason >= 0) {
510 if (rxrpc_ack_priority[ack_reason] > rxrpc_ack_priority[*_ack_reason]) {
511 *_ack_serial = serial;
512 *_ack_reason = ack_reason;
513 } else if (rxrpc_ack_priority[ack_reason] == rxrpc_ack_priority[*_ack_reason] &&
514 ack_reason == RXRPC_ACK_REQUESTED) {
515 *_ack_serial = serial;
516 *_ack_reason = ack_reason;
517 }
518 }
519 }
520
521 /*
522 * Split a jumbo packet and file the bits separately.
523 */
rxrpc_input_split_jumbo(struct rxrpc_call * call,struct sk_buff * skb)524 static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
525 {
526 struct rxrpc_jumbo_header jhdr;
527 struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
528 struct sk_buff *jskb;
529 rxrpc_serial_t ack_serial = 0;
530 unsigned int offset = sizeof(struct rxrpc_wire_header);
531 unsigned int len = skb->len - offset;
532 bool notify = false;
533 int ack_reason = 0;
534
535 while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
536 if (len < RXRPC_JUMBO_SUBPKTLEN)
537 goto protocol_error;
538 if (sp->hdr.flags & RXRPC_LAST_PACKET)
539 goto protocol_error;
540 if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
541 &jhdr, sizeof(jhdr)) < 0)
542 goto protocol_error;
543
544 jskb = skb_clone(skb, GFP_NOFS);
545 if (!jskb) {
546 kdebug("couldn't clone");
547 return false;
548 }
549 rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
550 jsp = rxrpc_skb(jskb);
551 jsp->offset = offset;
552 jsp->len = RXRPC_JUMBO_DATALEN;
553 rxrpc_input_data_one(call, jskb, ¬ify, &ack_serial, &ack_reason);
554 rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
555
556 sp->hdr.flags = jhdr.flags;
557 sp->hdr._rsvd = ntohs(jhdr._rsvd);
558 sp->hdr.seq++;
559 sp->hdr.serial++;
560 offset += RXRPC_JUMBO_SUBPKTLEN;
561 len -= RXRPC_JUMBO_SUBPKTLEN;
562 }
563
564 sp->offset = offset;
565 sp->len = len;
566 rxrpc_input_data_one(call, skb, ¬ify, &ack_serial, &ack_reason);
567
568 if (ack_reason > 0) {
569 rxrpc_send_ACK(call, ack_reason, ack_serial,
570 rxrpc_propose_ack_input_data);
571 } else {
572 call->ackr_nr_unacked++;
573 rxrpc_propose_delay_ACK(call, sp->hdr.serial,
574 rxrpc_propose_ack_input_data);
575 }
576 if (notify) {
577 trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
578 rxrpc_notify_socket(call);
579 }
580 return true;
581
582 protocol_error:
583 return false;
584 }
585
586 /*
587 * Process a DATA packet, adding the packet to the Rx ring. The caller's
588 * packet ref must be passed on or discarded.
589 */
rxrpc_input_data(struct rxrpc_call * call,struct sk_buff * skb)590 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
591 {
592 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
593 rxrpc_serial_t serial = sp->hdr.serial;
594 rxrpc_seq_t seq0 = sp->hdr.seq;
595
596 _enter("{%x,%x,%x},{%u,%x}",
597 call->ackr_window, call->ackr_wtop, call->rx_highest_seq,
598 skb->len, seq0);
599
600 if (__rxrpc_call_is_complete(call))
601 return;
602
603 switch (__rxrpc_call_state(call)) {
604 case RXRPC_CALL_CLIENT_SEND_REQUEST:
605 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
606 /* Received data implicitly ACKs all of the request
607 * packets we sent when we're acting as a client.
608 */
609 if (!rxrpc_receiving_reply(call))
610 goto out_notify;
611 break;
612
613 case RXRPC_CALL_SERVER_RECV_REQUEST: {
614 unsigned long timo = READ_ONCE(call->next_req_timo);
615
616 if (timo) {
617 ktime_t delay = ms_to_ktime(timo);
618
619 call->expect_req_by = ktime_add(ktime_get_real(), delay);
620 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_idle);
621 }
622 break;
623 }
624
625 default:
626 break;
627 }
628
629 if (!rxrpc_input_split_jumbo(call, skb)) {
630 rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
631 goto out_notify;
632 }
633 return;
634
635 out_notify:
636 trace_rxrpc_notify_socket(call->debug_id, serial);
637 rxrpc_notify_socket(call);
638 _leave(" [queued]");
639 }
640
641 /*
642 * See if there's a cached RTT probe to complete.
643 */
rxrpc_complete_rtt_probe(struct rxrpc_call * call,ktime_t resp_time,rxrpc_serial_t acked_serial,rxrpc_serial_t ack_serial,enum rxrpc_rtt_rx_trace type)644 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
645 ktime_t resp_time,
646 rxrpc_serial_t acked_serial,
647 rxrpc_serial_t ack_serial,
648 enum rxrpc_rtt_rx_trace type)
649 {
650 rxrpc_serial_t orig_serial;
651 unsigned long avail;
652 ktime_t sent_at;
653 bool matched = false;
654 int i;
655
656 avail = READ_ONCE(call->rtt_avail);
657 smp_rmb(); /* Read avail bits before accessing data. */
658
659 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
660 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
661 continue;
662
663 sent_at = call->rtt_sent_at[i];
664 orig_serial = call->rtt_serial[i];
665
666 if (orig_serial == acked_serial) {
667 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
668 smp_mb(); /* Read data before setting avail bit */
669 set_bit(i, &call->rtt_avail);
670 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
671 sent_at, resp_time);
672 matched = true;
673 }
674
675 /* If a later serial is being acked, then mark this slot as
676 * being available.
677 */
678 if (after(acked_serial, orig_serial)) {
679 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
680 orig_serial, acked_serial, 0, 0);
681 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
682 smp_wmb();
683 set_bit(i, &call->rtt_avail);
684 }
685 }
686
687 if (!matched)
688 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
689 }
690
691 /*
692 * Process the extra information that may be appended to an ACK packet
693 */
rxrpc_input_ack_trailer(struct rxrpc_call * call,struct sk_buff * skb,struct rxrpc_acktrailer * trailer)694 static void rxrpc_input_ack_trailer(struct rxrpc_call *call, struct sk_buff *skb,
695 struct rxrpc_acktrailer *trailer)
696 {
697 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
698 struct rxrpc_peer *peer;
699 unsigned int mtu;
700 bool wake = false;
701 u32 rwind = ntohl(trailer->rwind);
702
703 if (rwind > RXRPC_TX_MAX_WINDOW)
704 rwind = RXRPC_TX_MAX_WINDOW;
705 if (call->tx_winsize != rwind) {
706 if (rwind > call->tx_winsize)
707 wake = true;
708 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
709 call->tx_winsize = rwind;
710 }
711
712 mtu = min(ntohl(trailer->maxMTU), ntohl(trailer->ifMTU));
713
714 peer = call->peer;
715 if (mtu < peer->maxdata) {
716 spin_lock(&peer->lock);
717 peer->maxdata = mtu;
718 peer->mtu = mtu + peer->hdrsize;
719 spin_unlock(&peer->lock);
720 }
721
722 if (wake)
723 wake_up(&call->waitq);
724 }
725
726 /*
727 * Determine how many nacks from the previous ACK have now been satisfied.
728 */
rxrpc_input_check_prev_ack(struct rxrpc_call * call,struct rxrpc_ack_summary * summary,rxrpc_seq_t seq)729 static rxrpc_seq_t rxrpc_input_check_prev_ack(struct rxrpc_call *call,
730 struct rxrpc_ack_summary *summary,
731 rxrpc_seq_t seq)
732 {
733 struct sk_buff *skb = call->cong_last_nack;
734 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
735 unsigned int i, new_acks = 0, retained_nacks = 0;
736 rxrpc_seq_t old_seq = sp->ack.first_ack;
737 u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
738
739 if (after_eq(seq, old_seq + sp->ack.nr_acks)) {
740 summary->nr_new_acks += sp->ack.nr_nacks;
741 summary->nr_new_acks += seq - (old_seq + sp->ack.nr_acks);
742 summary->nr_retained_nacks = 0;
743 } else if (seq == old_seq) {
744 summary->nr_retained_nacks = sp->ack.nr_nacks;
745 } else {
746 for (i = 0; i < sp->ack.nr_acks; i++) {
747 if (acks[i] == RXRPC_ACK_TYPE_NACK) {
748 if (before(old_seq + i, seq))
749 new_acks++;
750 else
751 retained_nacks++;
752 }
753 }
754
755 summary->nr_new_acks += new_acks;
756 summary->nr_retained_nacks = retained_nacks;
757 }
758
759 return old_seq + sp->ack.nr_acks;
760 }
761
762 /*
763 * Process individual soft ACKs.
764 *
765 * Each ACK in the array corresponds to one packet and can be either an ACK or
766 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
767 * packets that lie beyond the end of the ACK list are scheduled for resend by
768 * the timer on the basis that the peer might just not have processed them at
769 * the time the ACK was sent.
770 */
rxrpc_input_soft_acks(struct rxrpc_call * call,struct rxrpc_ack_summary * summary,struct sk_buff * skb,rxrpc_seq_t seq,rxrpc_seq_t since)771 static void rxrpc_input_soft_acks(struct rxrpc_call *call,
772 struct rxrpc_ack_summary *summary,
773 struct sk_buff *skb,
774 rxrpc_seq_t seq,
775 rxrpc_seq_t since)
776 {
777 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
778 unsigned int i, old_nacks = 0;
779 rxrpc_seq_t lowest_nak = seq + sp->ack.nr_acks;
780 u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
781
782 for (i = 0; i < sp->ack.nr_acks; i++) {
783 if (acks[i] == RXRPC_ACK_TYPE_ACK) {
784 summary->nr_acks++;
785 if (after_eq(seq, since))
786 summary->nr_new_acks++;
787 } else {
788 summary->saw_nacks = true;
789 if (before(seq, since)) {
790 /* Overlap with previous ACK */
791 old_nacks++;
792 } else {
793 summary->nr_new_nacks++;
794 sp->ack.nr_nacks++;
795 }
796
797 if (before(seq, lowest_nak))
798 lowest_nak = seq;
799 }
800 seq++;
801 }
802
803 if (lowest_nak != call->acks_lowest_nak) {
804 call->acks_lowest_nak = lowest_nak;
805 summary->new_low_nack = true;
806 }
807
808 /* We *can* have more nacks than we did - the peer is permitted to drop
809 * packets it has soft-acked and re-request them. Further, it is
810 * possible for the nack distribution to change whilst the number of
811 * nacks stays the same or goes down.
812 */
813 if (old_nacks < summary->nr_retained_nacks)
814 summary->nr_new_acks += summary->nr_retained_nacks - old_nacks;
815 summary->nr_retained_nacks = old_nacks;
816 }
817
818 /*
819 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
820 * with respect to the ack state conveyed by preceding ACKs.
821 */
rxrpc_is_ack_valid(struct rxrpc_call * call,rxrpc_seq_t first_pkt,rxrpc_seq_t prev_pkt)822 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
823 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
824 {
825 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
826
827 if (after(first_pkt, base))
828 return true; /* The window advanced */
829
830 if (before(first_pkt, base))
831 return false; /* firstPacket regressed */
832
833 if (after_eq(prev_pkt, call->acks_prev_seq))
834 return true; /* previousPacket hasn't regressed. */
835
836 /* Some rx implementations put a serial number in previousPacket. */
837 if (after_eq(prev_pkt, base + call->tx_winsize))
838 return false;
839 return true;
840 }
841
842 /*
843 * Process an ACK packet.
844 *
845 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
846 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
847 *
848 * A hard-ACK means that a packet has been processed and may be discarded; a
849 * soft-ACK means that the packet may be discarded and retransmission
850 * requested. A phase is complete when all packets are hard-ACK'd.
851 */
rxrpc_input_ack(struct rxrpc_call * call,struct sk_buff * skb)852 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
853 {
854 struct rxrpc_ack_summary summary = { 0 };
855 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
856 struct rxrpc_acktrailer trailer;
857 rxrpc_serial_t ack_serial, acked_serial;
858 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt, since;
859 int nr_acks, offset, ioffset;
860
861 _enter("");
862
863 offset = sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
864
865 ack_serial = sp->hdr.serial;
866 acked_serial = sp->ack.acked_serial;
867 first_soft_ack = sp->ack.first_ack;
868 prev_pkt = sp->ack.prev_ack;
869 nr_acks = sp->ack.nr_acks;
870 hard_ack = first_soft_ack - 1;
871 summary.ack_reason = (sp->ack.reason < RXRPC_ACK__INVALID ?
872 sp->ack.reason : RXRPC_ACK__INVALID);
873
874 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
875 first_soft_ack, prev_pkt,
876 summary.ack_reason, nr_acks);
877 rxrpc_inc_stat(call->rxnet, stat_rx_acks[summary.ack_reason]);
878
879 if (acked_serial != 0) {
880 switch (summary.ack_reason) {
881 case RXRPC_ACK_PING_RESPONSE:
882 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
883 rxrpc_rtt_rx_ping_response);
884 break;
885 case RXRPC_ACK_REQUESTED:
886 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
887 rxrpc_rtt_rx_requested_ack);
888 break;
889 default:
890 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
891 rxrpc_rtt_rx_other_ack);
892 break;
893 }
894 }
895
896 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
897 * indicates that the client address changed due to NAT. The server
898 * lost the call because it switched to a different peer.
899 */
900 if (unlikely(summary.ack_reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
901 first_soft_ack == 1 &&
902 prev_pkt == 0 &&
903 rxrpc_is_client_call(call)) {
904 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
905 0, -ENETRESET);
906 goto send_response;
907 }
908
909 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
910 * indicate a change of address. However, we can retransmit the call
911 * if we still have it buffered to the beginning.
912 */
913 if (unlikely(summary.ack_reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
914 first_soft_ack == 1 &&
915 prev_pkt == 0 &&
916 call->acks_hard_ack == 0 &&
917 rxrpc_is_client_call(call)) {
918 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
919 0, -ENETRESET);
920 goto send_response;
921 }
922
923 /* Discard any out-of-order or duplicate ACKs (outside lock). */
924 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
925 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
926 first_soft_ack, call->acks_first_seq,
927 prev_pkt, call->acks_prev_seq);
928 goto send_response;
929 }
930
931 trailer.maxMTU = 0;
932 ioffset = offset + nr_acks + 3;
933 if (skb->len >= ioffset + sizeof(trailer) &&
934 skb_copy_bits(skb, ioffset, &trailer, sizeof(trailer)) < 0)
935 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_trailer);
936
937 if (nr_acks > 0)
938 skb_condense(skb);
939
940 if (call->cong_last_nack) {
941 since = rxrpc_input_check_prev_ack(call, &summary, first_soft_ack);
942 rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
943 call->cong_last_nack = NULL;
944 } else {
945 summary.nr_new_acks = first_soft_ack - call->acks_first_seq;
946 call->acks_lowest_nak = first_soft_ack + nr_acks;
947 since = first_soft_ack;
948 }
949
950 call->acks_latest_ts = skb->tstamp;
951 call->acks_first_seq = first_soft_ack;
952 call->acks_prev_seq = prev_pkt;
953
954 switch (summary.ack_reason) {
955 case RXRPC_ACK_PING:
956 break;
957 default:
958 if (acked_serial && after(acked_serial, call->acks_highest_serial))
959 call->acks_highest_serial = acked_serial;
960 break;
961 }
962
963 /* Parse rwind and mtu sizes if provided. */
964 if (trailer.maxMTU)
965 rxrpc_input_ack_trailer(call, skb, &trailer);
966
967 if (first_soft_ack == 0)
968 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
969
970 /* Ignore ACKs unless we are or have just been transmitting. */
971 switch (__rxrpc_call_state(call)) {
972 case RXRPC_CALL_CLIENT_SEND_REQUEST:
973 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
974 case RXRPC_CALL_SERVER_SEND_REPLY:
975 case RXRPC_CALL_SERVER_AWAIT_ACK:
976 break;
977 default:
978 goto send_response;
979 }
980
981 if (before(hard_ack, call->acks_hard_ack) ||
982 after(hard_ack, call->tx_top))
983 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
984 if (nr_acks > call->tx_top - hard_ack)
985 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
986
987 if (after(hard_ack, call->acks_hard_ack)) {
988 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
989 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
990 goto send_response;
991 }
992 }
993
994 if (nr_acks > 0) {
995 if (offset > (int)skb->len - nr_acks)
996 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
997 rxrpc_input_soft_acks(call, &summary, skb, first_soft_ack, since);
998 rxrpc_get_skb(skb, rxrpc_skb_get_last_nack);
999 call->cong_last_nack = skb;
1000 }
1001
1002 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
1003 summary.nr_acks == call->tx_top - hard_ack &&
1004 rxrpc_is_client_call(call))
1005 rxrpc_propose_ping(call, ack_serial,
1006 rxrpc_propose_ack_ping_for_lost_reply);
1007
1008 rxrpc_congestion_management(call, skb, &summary, acked_serial);
1009
1010 send_response:
1011 if (summary.ack_reason == RXRPC_ACK_PING)
1012 rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
1013 rxrpc_propose_ack_respond_to_ping);
1014 else if (sp->hdr.flags & RXRPC_REQUEST_ACK)
1015 rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
1016 rxrpc_propose_ack_respond_to_ack);
1017 }
1018
1019 /*
1020 * Process an ACKALL packet.
1021 */
rxrpc_input_ackall(struct rxrpc_call * call,struct sk_buff * skb)1022 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1023 {
1024 struct rxrpc_ack_summary summary = { 0 };
1025
1026 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1027 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
1028 }
1029
1030 /*
1031 * Process an ABORT packet directed at a call.
1032 */
rxrpc_input_abort(struct rxrpc_call * call,struct sk_buff * skb)1033 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1034 {
1035 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1036
1037 trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
1038
1039 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1040 skb->priority, -ECONNABORTED);
1041 }
1042
1043 /*
1044 * Process an incoming call packet.
1045 */
rxrpc_input_call_packet(struct rxrpc_call * call,struct sk_buff * skb)1046 void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
1047 {
1048 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1049 unsigned long timo;
1050
1051 _enter("%p,%p", call, skb);
1052
1053 if (sp->hdr.serviceId != call->dest_srx.srx_service)
1054 call->dest_srx.srx_service = sp->hdr.serviceId;
1055 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1056 call->rx_serial = sp->hdr.serial;
1057 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1058 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1059
1060 timo = READ_ONCE(call->next_rx_timo);
1061 if (timo) {
1062 ktime_t delay = ms_to_ktime(timo);
1063
1064 call->expect_rx_by = ktime_add(ktime_get_real(), delay);
1065 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_expect_rx);
1066 }
1067
1068 switch (sp->hdr.type) {
1069 case RXRPC_PACKET_TYPE_DATA:
1070 return rxrpc_input_data(call, skb);
1071
1072 case RXRPC_PACKET_TYPE_ACK:
1073 return rxrpc_input_ack(call, skb);
1074
1075 case RXRPC_PACKET_TYPE_BUSY:
1076 /* Just ignore BUSY packets from the server; the retry and
1077 * lifespan timers will take care of business. BUSY packets
1078 * from the client don't make sense.
1079 */
1080 return;
1081
1082 case RXRPC_PACKET_TYPE_ABORT:
1083 return rxrpc_input_abort(call, skb);
1084
1085 case RXRPC_PACKET_TYPE_ACKALL:
1086 return rxrpc_input_ackall(call, skb);
1087
1088 default:
1089 break;
1090 }
1091 }
1092
1093 /*
1094 * Handle a new service call on a channel implicitly completing the preceding
1095 * call on that channel. This does not apply to client conns.
1096 *
1097 * TODO: If callNumber > call_id + 1, renegotiate security.
1098 */
rxrpc_implicit_end_call(struct rxrpc_call * call,struct sk_buff * skb)1099 void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1100 {
1101 switch (__rxrpc_call_state(call)) {
1102 case RXRPC_CALL_SERVER_AWAIT_ACK:
1103 rxrpc_call_completed(call);
1104 fallthrough;
1105 case RXRPC_CALL_COMPLETE:
1106 break;
1107 default:
1108 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1109 rxrpc_eproto_improper_term);
1110 trace_rxrpc_improper_term(call);
1111 break;
1112 }
1113
1114 rxrpc_input_call_event(call, skb);
1115 }
1116