1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Management of Tx window, Tx resend, ACKs and out-of-sequence reception
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/module.h>
11 #include <linux/circ_buf.h>
12 #include <linux/net.h>
13 #include <linux/skbuff.h>
14 #include <linux/slab.h>
15 #include <linux/udp.h>
16 #include <net/sock.h>
17 #include <net/af_rxrpc.h>
18 #include "ar-internal.h"
19
20 /*
21 * Propose a PING ACK be sent.
22 */
23 void rxrpc_propose_ping(struct rxrpc_call *call, u32 serial,
24 enum rxrpc_propose_ack_trace why)
25 {
26 ktime_t delay = ms_to_ktime(READ_ONCE(rxrpc_idle_ack_delay));
27 ktime_t now = ktime_get_real();
28 ktime_t ping_at = ktime_add(now, delay);
29
30 trace_rxrpc_propose_ack(call, why, RXRPC_ACK_PING, serial);
31 if (ktime_before(ping_at, call->ping_at)) {
32 call->ping_at = ping_at;
33 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_ping);
34 }
35 }
36
37 /*
38 * Propose a DELAY ACK be sent in the future.
39 */
40 void rxrpc_propose_delay_ACK(struct rxrpc_call *call, rxrpc_serial_t serial,
41 enum rxrpc_propose_ack_trace why)
42 {
43 ktime_t now = ktime_get_real(), delay;
44
45 trace_rxrpc_propose_ack(call, why, RXRPC_ACK_DELAY, serial);
46
47 if (call->peer->srtt_us)
48 delay = (call->peer->srtt_us >> 3) * NSEC_PER_USEC;
49 else
50 delay = ms_to_ktime(READ_ONCE(rxrpc_soft_ack_delay));
51 ktime_add_ms(delay, call->tx_backoff);
52
53 call->delay_ack_at = ktime_add(now, delay);
54 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_delayed_ack);
55 }
56
57 /*
58 * Handle congestion being detected by the retransmit timeout.
59 */
60 static void rxrpc_congestion_timeout(struct rxrpc_call *call)
61 {
62 set_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags);
63 }
64
65 /*
66 * Perform retransmission of NAK'd and unack'd packets.
67 */
68 void rxrpc_resend(struct rxrpc_call *call, struct sk_buff *ack_skb)
69 {
70 struct rxrpc_ackpacket *ack = NULL;
71 struct rxrpc_skb_priv *sp;
72 struct rxrpc_txbuf *txb;
73 rxrpc_seq_t transmitted = call->tx_transmitted;
74 ktime_t next_resend = KTIME_MAX, rto = ns_to_ktime(call->peer->rto_us * NSEC_PER_USEC);
75 ktime_t resend_at = KTIME_MAX, now, delay;
76 bool unacked = false, did_send = false;
77 unsigned int i;
78
79 _enter("{%d,%d}", call->acks_hard_ack, call->tx_top);
80
81 now = ktime_get_real();
82
83 if (list_empty(&call->tx_buffer))
84 goto no_resend;
85
86 trace_rxrpc_resend(call, ack_skb);
87 txb = list_first_entry(&call->tx_buffer, struct rxrpc_txbuf, call_link);
88
89 /* Scan the soft ACK table without dropping the lock and resend any
90 * explicitly NAK'd packets.
91 */
92 if (ack_skb) {
93 sp = rxrpc_skb(ack_skb);
94 ack = (void *)ack_skb->data + sizeof(struct rxrpc_wire_header);
95
96 for (i = 0; i < sp->ack.nr_acks; i++) {
97 rxrpc_seq_t seq;
98
99 if (ack->acks[i] & 1)
100 continue;
101 seq = sp->ack.first_ack + i;
102 if (after(txb->seq, transmitted))
103 break;
104 if (after(txb->seq, seq))
105 continue; /* A new hard ACK probably came in */
106 list_for_each_entry_from(txb, &call->tx_buffer, call_link) {
107 if (txb->seq == seq)
108 goto found_txb;
109 }
110 goto no_further_resend;
111
112 found_txb:
113 resend_at = ktime_add(txb->last_sent, rto);
114 if (after(txb->serial, call->acks_highest_serial)) {
115 if (ktime_after(resend_at, now) &&
116 ktime_before(resend_at, next_resend))
117 next_resend = resend_at;
118 continue; /* Ack point not yet reached */
119 }
120
121 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_unacked);
122
123 trace_rxrpc_retransmit(call, txb->seq, txb->serial,
124 ktime_sub(resend_at, now));
125
126 txb->flags |= RXRPC_TXBUF_RESENT;
127 rxrpc_transmit_one(call, txb);
128 did_send = true;
129 now = ktime_get_real();
130
131 if (list_is_last(&txb->call_link, &call->tx_buffer))
132 goto no_further_resend;
133 txb = list_next_entry(txb, call_link);
134 }
135 }
136
137 /* Fast-forward through the Tx queue to the point the peer says it has
138 * seen. Anything between the soft-ACK table and that point will get
139 * ACK'd or NACK'd in due course, so don't worry about it here; here we
140 * need to consider retransmitting anything beyond that point.
141 */
142 if (after_eq(call->acks_prev_seq, call->tx_transmitted))
143 goto no_further_resend;
144
145 list_for_each_entry_from(txb, &call->tx_buffer, call_link) {
146 resend_at = ktime_add(txb->last_sent, rto);
147
148 if (before_eq(txb->seq, call->acks_prev_seq))
149 continue;
150 if (after(txb->seq, call->tx_transmitted))
151 break; /* Not transmitted yet */
152
153 if (ack && ack->reason == RXRPC_ACK_PING_RESPONSE &&
154 before(txb->serial, ntohl(ack->serial)))
155 goto do_resend; /* Wasn't accounted for by a more recent ping. */
156
157 if (ktime_after(resend_at, now)) {
158 if (ktime_before(resend_at, next_resend))
159 next_resend = resend_at;
160 continue;
161 }
162
163 do_resend:
164 unacked = true;
165
166 txb->flags |= RXRPC_TXBUF_RESENT;
167 rxrpc_transmit_one(call, txb);
168 did_send = true;
169 rxrpc_inc_stat(call->rxnet, stat_tx_data_retrans);
170 now = ktime_get_real();
171 }
172
173 no_further_resend:
174 no_resend:
175 if (resend_at < KTIME_MAX) {
176 delay = rxrpc_get_rto_backoff(call->peer, did_send);
177 resend_at = ktime_add(resend_at, delay);
178 trace_rxrpc_timer_set(call, resend_at - now, rxrpc_timer_trace_resend_reset);
179 }
180 call->resend_at = resend_at;
181
182 if (unacked)
183 rxrpc_congestion_timeout(call);
184
185 /* If there was nothing that needed retransmission then it's likely
186 * that an ACK got lost somewhere. Send a ping to find out instead of
187 * retransmitting data.
188 */
189 if (!did_send) {
190 ktime_t next_ping = ktime_add_us(call->acks_latest_ts,
191 call->peer->srtt_us >> 3);
192
193 if (ktime_sub(next_ping, now) <= 0)
194 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
195 rxrpc_propose_ack_ping_for_0_retrans);
196 }
197
198 _leave("");
199 }
200
201 /*
202 * Start transmitting the reply to a service. This cancels the need to ACK the
203 * request if we haven't yet done so.
204 */
205 static void rxrpc_begin_service_reply(struct rxrpc_call *call)
206 {
207 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_SEND_REPLY);
208 if (call->ackr_reason == RXRPC_ACK_DELAY)
209 call->ackr_reason = 0;
210 call->delay_ack_at = KTIME_MAX;
211 trace_rxrpc_timer_can(call, rxrpc_timer_trace_delayed_ack);
212 }
213
214 /*
215 * Close the transmission phase. After this point there is no more data to be
216 * transmitted in the call.
217 */
218 static void rxrpc_close_tx_phase(struct rxrpc_call *call)
219 {
220 _debug("________awaiting reply/ACK__________");
221
222 switch (__rxrpc_call_state(call)) {
223 case RXRPC_CALL_CLIENT_SEND_REQUEST:
224 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
225 break;
226 case RXRPC_CALL_SERVER_SEND_REPLY:
227 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_AWAIT_ACK);
228 break;
229 default:
230 break;
231 }
232 }
233
234 static bool rxrpc_tx_window_has_space(struct rxrpc_call *call)
235 {
236 unsigned int winsize = min_t(unsigned int, call->tx_winsize,
237 call->cong_cwnd + call->cong_extra);
238 rxrpc_seq_t window = call->acks_hard_ack, wtop = window + winsize;
239 rxrpc_seq_t tx_top = call->tx_top;
240 int space;
241
242 space = wtop - tx_top;
243 return space > 0;
244 }
245
246 /*
247 * Decant some if the sendmsg prepared queue into the transmission buffer.
248 */
249 static void rxrpc_decant_prepared_tx(struct rxrpc_call *call)
250 {
251 struct rxrpc_txbuf *txb;
252
253 if (!test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
254 if (list_empty(&call->tx_sendmsg))
255 return;
256 rxrpc_expose_client_call(call);
257 }
258
259 while ((txb = list_first_entry_or_null(&call->tx_sendmsg,
260 struct rxrpc_txbuf, call_link))) {
261 spin_lock(&call->tx_lock);
262 list_del(&txb->call_link);
263 spin_unlock(&call->tx_lock);
264
265 call->tx_top = txb->seq;
266 list_add_tail(&txb->call_link, &call->tx_buffer);
267
268 if (txb->flags & RXRPC_LAST_PACKET)
269 rxrpc_close_tx_phase(call);
270
271 rxrpc_transmit_one(call, txb);
272
273 if (!rxrpc_tx_window_has_space(call))
274 break;
275 }
276 }
277
278 static void rxrpc_transmit_some_data(struct rxrpc_call *call)
279 {
280 switch (__rxrpc_call_state(call)) {
281 case RXRPC_CALL_SERVER_ACK_REQUEST:
282 if (list_empty(&call->tx_sendmsg))
283 return;
284 rxrpc_begin_service_reply(call);
285 fallthrough;
286
287 case RXRPC_CALL_SERVER_SEND_REPLY:
288 case RXRPC_CALL_CLIENT_SEND_REQUEST:
289 if (!rxrpc_tx_window_has_space(call))
290 return;
291 if (list_empty(&call->tx_sendmsg)) {
292 rxrpc_inc_stat(call->rxnet, stat_tx_data_underflow);
293 return;
294 }
295 rxrpc_decant_prepared_tx(call);
296 break;
297 default:
298 return;
299 }
300 }
301
302 /*
303 * Ping the other end to fill our RTT cache and to retrieve the rwind
304 * and MTU parameters.
305 */
306 static void rxrpc_send_initial_ping(struct rxrpc_call *call)
307 {
308 if (call->peer->rtt_count < 3 ||
309 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
310 ktime_get_real()))
311 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
312 rxrpc_propose_ack_ping_for_params);
313 }
314
315 /*
316 * Handle retransmission and deferred ACK/abort generation.
317 */
318 bool rxrpc_input_call_event(struct rxrpc_call *call, struct sk_buff *skb)
319 {
320 ktime_t now, t;
321 bool resend = false;
322 s32 abort_code;
323
324 rxrpc_see_call(call, rxrpc_call_see_input);
325
326 //printk("\n--------------------\n");
327 _enter("{%d,%s,%lx}",
328 call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)],
329 call->events);
330
331 if (__rxrpc_call_is_complete(call))
332 goto out;
333
334 /* Handle abort request locklessly, vs rxrpc_propose_abort(). */
335 abort_code = smp_load_acquire(&call->send_abort);
336 if (abort_code) {
337 rxrpc_abort_call(call, 0, call->send_abort, call->send_abort_err,
338 call->send_abort_why);
339 goto out;
340 }
341
342 if (skb && skb->mark == RXRPC_SKB_MARK_ERROR)
343 goto out;
344
345 if (skb)
346 rxrpc_input_call_packet(call, skb);
347
348 /* If we see our async-event poke, check for timeout trippage. */
349 now = ktime_get_real();
350 t = ktime_sub(call->expect_rx_by, now);
351 if (t <= 0) {
352 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_expect_rx);
353 goto expired;
354 }
355
356 t = ktime_sub(call->expect_req_by, now);
357 if (t <= 0) {
358 call->expect_req_by = KTIME_MAX;
359 if (__rxrpc_call_state(call) == RXRPC_CALL_SERVER_RECV_REQUEST) {
360 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_idle);
361 goto expired;
362 }
363 }
364
365 t = ktime_sub(READ_ONCE(call->expect_term_by), now);
366 if (t <= 0) {
367 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_hard);
368 goto expired;
369 }
370
371 t = ktime_sub(call->delay_ack_at, now);
372 if (t <= 0) {
373 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_delayed_ack);
374 call->delay_ack_at = KTIME_MAX;
375 rxrpc_send_ACK(call, RXRPC_ACK_DELAY, 0,
376 rxrpc_propose_ack_delayed_ack);
377 }
378
379 t = ktime_sub(call->ack_lost_at, now);
380 if (t <= 0) {
381 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_lost_ack);
382 call->ack_lost_at = KTIME_MAX;
383 set_bit(RXRPC_CALL_EV_ACK_LOST, &call->events);
384 }
385
386 t = ktime_sub(call->ping_at, now);
387 if (t <= 0) {
388 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_ping);
389 call->ping_at = KTIME_MAX;
390 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
391 rxrpc_propose_ack_ping_for_keepalive);
392 }
393
394 t = ktime_sub(call->resend_at, now);
395 if (t <= 0) {
396 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_resend);
397 call->resend_at = KTIME_MAX;
398 resend = true;
399 }
400
401 rxrpc_transmit_some_data(call);
402
403 now = ktime_get_real();
404 t = ktime_sub(call->keepalive_at, now);
405 if (t <= 0) {
406 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_keepalive);
407 call->keepalive_at = KTIME_MAX;
408 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
409 rxrpc_propose_ack_ping_for_keepalive);
410 }
411
412 if (skb) {
413 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
414
415 if (sp->hdr.type == RXRPC_PACKET_TYPE_ACK)
416 rxrpc_congestion_degrade(call);
417 }
418
419 if (test_and_clear_bit(RXRPC_CALL_EV_INITIAL_PING, &call->events))
420 rxrpc_send_initial_ping(call);
421
422 /* Process events */
423 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events))
424 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
425 rxrpc_propose_ack_ping_for_lost_ack);
426
427 if (resend &&
428 __rxrpc_call_state(call) != RXRPC_CALL_CLIENT_RECV_REPLY &&
429 !test_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags))
430 rxrpc_resend(call, NULL);
431
432 if (test_and_clear_bit(RXRPC_CALL_RX_IS_IDLE, &call->flags))
433 rxrpc_send_ACK(call, RXRPC_ACK_IDLE, 0,
434 rxrpc_propose_ack_rx_idle);
435
436 if (call->ackr_nr_unacked > 2) {
437 if (call->peer->rtt_count < 3)
438 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
439 rxrpc_propose_ack_ping_for_rtt);
440 else if (ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
441 ktime_get_real()))
442 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0,
443 rxrpc_propose_ack_ping_for_old_rtt);
444 else
445 rxrpc_send_ACK(call, RXRPC_ACK_IDLE, 0,
446 rxrpc_propose_ack_input_data);
447 }
448
449 /* Make sure the timer is restarted */
450 if (!__rxrpc_call_is_complete(call)) {
451 ktime_t next = READ_ONCE(call->expect_term_by), delay;
452
453 #define set(T) { ktime_t _t = (T); if (ktime_before(_t, next)) next = _t; }
454
455 set(call->expect_req_by);
456 set(call->expect_rx_by);
457 set(call->delay_ack_at);
458 set(call->ack_lost_at);
459 set(call->resend_at);
460 set(call->keepalive_at);
461 set(call->ping_at);
462
463 now = ktime_get_real();
464 delay = ktime_sub(next, now);
465 if (delay <= 0) {
466 rxrpc_poke_call(call, rxrpc_call_poke_timer_now);
467 } else {
468 unsigned long nowj = jiffies, delayj, nextj;
469
470 delayj = max(nsecs_to_jiffies(delay), 1);
471 nextj = nowj + delayj;
472 if (time_before(nextj, call->timer.expires) ||
473 !timer_pending(&call->timer)) {
474 trace_rxrpc_timer_restart(call, delay, delayj);
475 timer_reduce(&call->timer, nextj);
476 }
477 }
478 }
479
480 out:
481 if (__rxrpc_call_is_complete(call)) {
482 del_timer_sync(&call->timer);
483 if (!test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
484 rxrpc_disconnect_call(call);
485 if (call->security)
486 call->security->free_call_crypto(call);
487 }
488 if (call->acks_hard_ack != call->tx_bottom)
489 rxrpc_shrink_call_tx_buffer(call);
490 _leave("");
491 return true;
492
493 expired:
494 if (test_bit(RXRPC_CALL_RX_HEARD, &call->flags) &&
495 (int)call->conn->hi_serial - (int)call->rx_serial > 0) {
496 trace_rxrpc_call_reset(call);
497 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ECONNRESET,
498 rxrpc_abort_call_reset);
499 } else {
500 rxrpc_abort_call(call, 0, RX_CALL_TIMEOUT, -ETIME,
501 rxrpc_abort_call_timeout);
502 }
503 goto out;
504 }
505
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.