1 /* Management of Tx window, Tx resend, ACKs and out-of-sequence reception
2 *
3 * Copyright (C) 2007 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/circ_buf.h>
16 #include <linux/net.h>
17 #include <linux/skbuff.h>
18 #include <linux/slab.h>
19 #include <linux/udp.h>
20 #include <net/sock.h>
21 #include <net/af_rxrpc.h>
22 #include "ar-internal.h"
23
24 /*
25 * Propose a PING ACK be sent.
26 */
rxrpc_propose_ping(struct rxrpc_call * call,bool immediate,bool background)27 static void rxrpc_propose_ping(struct rxrpc_call *call,
28 bool immediate, bool background)
29 {
30 if (immediate) {
31 if (background &&
32 !test_and_set_bit(RXRPC_CALL_EV_PING, &call->events))
33 rxrpc_queue_call(call);
34 } else {
35 unsigned long now = jiffies;
36 unsigned long ping_at = now + rxrpc_idle_ack_delay;
37
38 if (time_before(ping_at, call->ping_at)) {
39 WRITE_ONCE(call->ping_at, ping_at);
40 rxrpc_reduce_call_timer(call, ping_at, now,
41 rxrpc_timer_set_for_ping);
42 }
43 }
44 }
45
46 /*
47 * propose an ACK be sent
48 */
__rxrpc_propose_ACK(struct rxrpc_call * call,u8 ack_reason,u16 skew,u32 serial,bool immediate,bool background,enum rxrpc_propose_ack_trace why)49 static void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
50 u16 skew, u32 serial, bool immediate,
51 bool background,
52 enum rxrpc_propose_ack_trace why)
53 {
54 enum rxrpc_propose_ack_outcome outcome = rxrpc_propose_ack_use;
55 unsigned long expiry = rxrpc_soft_ack_delay;
56 s8 prior = rxrpc_ack_priority[ack_reason];
57
58 /* Pings are handled specially because we don't want to accidentally
59 * lose a ping response by subsuming it into a ping.
60 */
61 if (ack_reason == RXRPC_ACK_PING) {
62 rxrpc_propose_ping(call, immediate, background);
63 goto trace;
64 }
65
66 /* Update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
67 * numbers, but we don't alter the timeout.
68 */
69 _debug("prior %u %u vs %u %u",
70 ack_reason, prior,
71 call->ackr_reason, rxrpc_ack_priority[call->ackr_reason]);
72 if (ack_reason == call->ackr_reason) {
73 if (RXRPC_ACK_UPDATEABLE & (1 << ack_reason)) {
74 outcome = rxrpc_propose_ack_update;
75 call->ackr_serial = serial;
76 call->ackr_skew = skew;
77 }
78 if (!immediate)
79 goto trace;
80 } else if (prior > rxrpc_ack_priority[call->ackr_reason]) {
81 call->ackr_reason = ack_reason;
82 call->ackr_serial = serial;
83 call->ackr_skew = skew;
84 } else {
85 outcome = rxrpc_propose_ack_subsume;
86 }
87
88 switch (ack_reason) {
89 case RXRPC_ACK_REQUESTED:
90 if (rxrpc_requested_ack_delay < expiry)
91 expiry = rxrpc_requested_ack_delay;
92 if (serial == 1)
93 immediate = false;
94 break;
95
96 case RXRPC_ACK_DELAY:
97 if (rxrpc_soft_ack_delay < expiry)
98 expiry = rxrpc_soft_ack_delay;
99 break;
100
101 case RXRPC_ACK_IDLE:
102 if (rxrpc_idle_ack_delay < expiry)
103 expiry = rxrpc_idle_ack_delay;
104 break;
105
106 default:
107 immediate = true;
108 break;
109 }
110
111 if (test_bit(RXRPC_CALL_EV_ACK, &call->events)) {
112 _debug("already scheduled");
113 } else if (immediate || expiry == 0) {
114 _debug("immediate ACK %lx", call->events);
115 if (!test_and_set_bit(RXRPC_CALL_EV_ACK, &call->events) &&
116 background)
117 rxrpc_queue_call(call);
118 } else {
119 unsigned long now = jiffies, ack_at;
120
121 if (call->peer->rtt_usage > 0)
122 ack_at = nsecs_to_jiffies(call->peer->rtt);
123 else
124 ack_at = expiry;
125
126 ack_at += now;
127 if (time_before(ack_at, call->ack_at)) {
128 WRITE_ONCE(call->ack_at, ack_at);
129 rxrpc_reduce_call_timer(call, ack_at, now,
130 rxrpc_timer_set_for_ack);
131 }
132 }
133
134 trace:
135 trace_rxrpc_propose_ack(call, why, ack_reason, serial, immediate,
136 background, outcome);
137 }
138
139 /*
140 * propose an ACK be sent, locking the call structure
141 */
rxrpc_propose_ACK(struct rxrpc_call * call,u8 ack_reason,u16 skew,u32 serial,bool immediate,bool background,enum rxrpc_propose_ack_trace why)142 void rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
143 u16 skew, u32 serial, bool immediate, bool background,
144 enum rxrpc_propose_ack_trace why)
145 {
146 spin_lock_bh(&call->lock);
147 __rxrpc_propose_ACK(call, ack_reason, skew, serial,
148 immediate, background, why);
149 spin_unlock_bh(&call->lock);
150 }
151
152 /*
153 * Handle congestion being detected by the retransmit timeout.
154 */
rxrpc_congestion_timeout(struct rxrpc_call * call)155 static void rxrpc_congestion_timeout(struct rxrpc_call *call)
156 {
157 set_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags);
158 }
159
160 /*
161 * Perform retransmission of NAK'd and unack'd packets.
162 */
rxrpc_resend(struct rxrpc_call * call,unsigned long now_j)163 static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
164 {
165 struct sk_buff *skb;
166 unsigned long resend_at;
167 rxrpc_seq_t cursor, seq, top;
168 ktime_t now, max_age, oldest, ack_ts, timeout, min_timeo;
169 int ix;
170 u8 annotation, anno_type, retrans = 0, unacked = 0;
171
172 _enter("{%d,%d}", call->tx_hard_ack, call->tx_top);
173
174 if (call->peer->rtt_usage > 1)
175 timeout = ns_to_ktime(call->peer->rtt * 3 / 2);
176 else
177 timeout = ms_to_ktime(rxrpc_resend_timeout);
178 min_timeo = ns_to_ktime((1000000000 / HZ) * 4);
179 if (ktime_before(timeout, min_timeo))
180 timeout = min_timeo;
181
182 now = ktime_get_real();
183 max_age = ktime_sub(now, timeout);
184
185 spin_lock_bh(&call->lock);
186
187 cursor = call->tx_hard_ack;
188 top = call->tx_top;
189 ASSERT(before_eq(cursor, top));
190 if (cursor == top)
191 goto out_unlock;
192
193 /* Scan the packet list without dropping the lock and decide which of
194 * the packets in the Tx buffer we're going to resend and what the new
195 * resend timeout will be.
196 */
197 trace_rxrpc_resend(call, (cursor + 1) & RXRPC_RXTX_BUFF_MASK);
198 oldest = now;
199 for (seq = cursor + 1; before_eq(seq, top); seq++) {
200 ix = seq & RXRPC_RXTX_BUFF_MASK;
201 annotation = call->rxtx_annotations[ix];
202 anno_type = annotation & RXRPC_TX_ANNO_MASK;
203 annotation &= ~RXRPC_TX_ANNO_MASK;
204 if (anno_type == RXRPC_TX_ANNO_ACK)
205 continue;
206
207 skb = call->rxtx_buffer[ix];
208 rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
209
210 if (anno_type == RXRPC_TX_ANNO_UNACK) {
211 if (ktime_after(skb->tstamp, max_age)) {
212 if (ktime_before(skb->tstamp, oldest))
213 oldest = skb->tstamp;
214 continue;
215 }
216 if (!(annotation & RXRPC_TX_ANNO_RESENT))
217 unacked++;
218 }
219
220 /* Okay, we need to retransmit a packet. */
221 call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS | annotation;
222 retrans++;
223 trace_rxrpc_retransmit(call, seq, annotation | anno_type,
224 ktime_to_ns(ktime_sub(skb->tstamp, max_age)));
225 }
226
227 resend_at = nsecs_to_jiffies(ktime_to_ns(ktime_sub(now, oldest)));
228 resend_at += jiffies + rxrpc_resend_timeout;
229 WRITE_ONCE(call->resend_at, resend_at);
230
231 if (unacked)
232 rxrpc_congestion_timeout(call);
233
234 /* If there was nothing that needed retransmission then it's likely
235 * that an ACK got lost somewhere. Send a ping to find out instead of
236 * retransmitting data.
237 */
238 if (!retrans) {
239 rxrpc_reduce_call_timer(call, resend_at, now_j,
240 rxrpc_timer_set_for_resend);
241 spin_unlock_bh(&call->lock);
242 ack_ts = ktime_sub(now, call->acks_latest_ts);
243 if (ktime_to_ns(ack_ts) < call->peer->rtt)
244 goto out;
245 rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, 0, true, false,
246 rxrpc_propose_ack_ping_for_lost_ack);
247 rxrpc_send_ack_packet(call, true, NULL);
248 goto out;
249 }
250
251 /* Now go through the Tx window and perform the retransmissions. We
252 * have to drop the lock for each send. If an ACK comes in whilst the
253 * lock is dropped, it may clear some of the retransmission markers for
254 * packets that it soft-ACKs.
255 */
256 for (seq = cursor + 1; before_eq(seq, top); seq++) {
257 ix = seq & RXRPC_RXTX_BUFF_MASK;
258 annotation = call->rxtx_annotations[ix];
259 anno_type = annotation & RXRPC_TX_ANNO_MASK;
260 if (anno_type != RXRPC_TX_ANNO_RETRANS)
261 continue;
262
263 skb = call->rxtx_buffer[ix];
264 rxrpc_get_skb(skb, rxrpc_skb_tx_got);
265 spin_unlock_bh(&call->lock);
266
267 if (rxrpc_send_data_packet(call, skb, true) < 0) {
268 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
269 return;
270 }
271
272 if (rxrpc_is_client_call(call))
273 rxrpc_expose_client_call(call);
274
275 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
276 spin_lock_bh(&call->lock);
277
278 /* We need to clear the retransmit state, but there are two
279 * things we need to be aware of: A new ACK/NAK might have been
280 * received and the packet might have been hard-ACK'd (in which
281 * case it will no longer be in the buffer).
282 */
283 if (after(seq, call->tx_hard_ack)) {
284 annotation = call->rxtx_annotations[ix];
285 anno_type = annotation & RXRPC_TX_ANNO_MASK;
286 if (anno_type == RXRPC_TX_ANNO_RETRANS ||
287 anno_type == RXRPC_TX_ANNO_NAK) {
288 annotation &= ~RXRPC_TX_ANNO_MASK;
289 annotation |= RXRPC_TX_ANNO_UNACK;
290 }
291 annotation |= RXRPC_TX_ANNO_RESENT;
292 call->rxtx_annotations[ix] = annotation;
293 }
294
295 if (after(call->tx_hard_ack, seq))
296 seq = call->tx_hard_ack;
297 }
298
299 out_unlock:
300 spin_unlock_bh(&call->lock);
301 out:
302 _leave("");
303 }
304
305 /*
306 * Handle retransmission and deferred ACK/abort generation.
307 */
rxrpc_process_call(struct work_struct * work)308 void rxrpc_process_call(struct work_struct *work)
309 {
310 struct rxrpc_call *call =
311 container_of(work, struct rxrpc_call, processor);
312 rxrpc_serial_t *send_ack;
313 unsigned long now, next, t;
314
315 rxrpc_see_call(call);
316
317 //printk("\n--------------------\n");
318 _enter("{%d,%s,%lx}",
319 call->debug_id, rxrpc_call_states[call->state], call->events);
320
321 recheck_state:
322 if (test_and_clear_bit(RXRPC_CALL_EV_ABORT, &call->events)) {
323 rxrpc_send_abort_packet(call);
324 goto recheck_state;
325 }
326
327 if (call->state == RXRPC_CALL_COMPLETE) {
328 del_timer_sync(&call->timer);
329 rxrpc_notify_socket(call);
330 goto out_put;
331 }
332
333 /* Work out if any timeouts tripped */
334 now = jiffies;
335 t = READ_ONCE(call->expect_rx_by);
336 if (time_after_eq(now, t)) {
337 trace_rxrpc_timer(call, rxrpc_timer_exp_normal, now);
338 set_bit(RXRPC_CALL_EV_EXPIRED, &call->events);
339 }
340
341 t = READ_ONCE(call->expect_req_by);
342 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST &&
343 time_after_eq(now, t)) {
344 trace_rxrpc_timer(call, rxrpc_timer_exp_idle, now);
345 set_bit(RXRPC_CALL_EV_EXPIRED, &call->events);
346 }
347
348 t = READ_ONCE(call->expect_term_by);
349 if (time_after_eq(now, t)) {
350 trace_rxrpc_timer(call, rxrpc_timer_exp_hard, now);
351 set_bit(RXRPC_CALL_EV_EXPIRED, &call->events);
352 }
353
354 t = READ_ONCE(call->ack_at);
355 if (time_after_eq(now, t)) {
356 trace_rxrpc_timer(call, rxrpc_timer_exp_ack, now);
357 cmpxchg(&call->ack_at, t, now + MAX_JIFFY_OFFSET);
358 set_bit(RXRPC_CALL_EV_ACK, &call->events);
359 }
360
361 t = READ_ONCE(call->ack_lost_at);
362 if (time_after_eq(now, t)) {
363 trace_rxrpc_timer(call, rxrpc_timer_exp_lost_ack, now);
364 cmpxchg(&call->ack_lost_at, t, now + MAX_JIFFY_OFFSET);
365 set_bit(RXRPC_CALL_EV_ACK_LOST, &call->events);
366 }
367
368 t = READ_ONCE(call->keepalive_at);
369 if (time_after_eq(now, t)) {
370 trace_rxrpc_timer(call, rxrpc_timer_exp_keepalive, now);
371 cmpxchg(&call->keepalive_at, t, now + MAX_JIFFY_OFFSET);
372 rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, 0, true, true,
373 rxrpc_propose_ack_ping_for_keepalive);
374 set_bit(RXRPC_CALL_EV_PING, &call->events);
375 }
376
377 t = READ_ONCE(call->ping_at);
378 if (time_after_eq(now, t)) {
379 trace_rxrpc_timer(call, rxrpc_timer_exp_ping, now);
380 cmpxchg(&call->ping_at, t, now + MAX_JIFFY_OFFSET);
381 set_bit(RXRPC_CALL_EV_PING, &call->events);
382 }
383
384 t = READ_ONCE(call->resend_at);
385 if (time_after_eq(now, t)) {
386 trace_rxrpc_timer(call, rxrpc_timer_exp_resend, now);
387 cmpxchg(&call->resend_at, t, now + MAX_JIFFY_OFFSET);
388 set_bit(RXRPC_CALL_EV_RESEND, &call->events);
389 }
390
391 /* Process events */
392 if (test_and_clear_bit(RXRPC_CALL_EV_EXPIRED, &call->events)) {
393 if (test_bit(RXRPC_CALL_RX_HEARD, &call->flags) &&
394 (int)call->conn->hi_serial - (int)call->rx_serial > 0) {
395 trace_rxrpc_call_reset(call);
396 rxrpc_abort_call("EXP", call, 0, RX_USER_ABORT, -ECONNRESET);
397 } else {
398 rxrpc_abort_call("EXP", call, 0, RX_USER_ABORT, -ETIME);
399 }
400 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
401 goto recheck_state;
402 }
403
404 send_ack = NULL;
405 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events)) {
406 call->acks_lost_top = call->tx_top;
407 rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, 0, true, false,
408 rxrpc_propose_ack_ping_for_lost_ack);
409 send_ack = &call->acks_lost_ping;
410 }
411
412 if (test_and_clear_bit(RXRPC_CALL_EV_ACK, &call->events) ||
413 send_ack) {
414 if (call->ackr_reason) {
415 rxrpc_send_ack_packet(call, false, send_ack);
416 goto recheck_state;
417 }
418 }
419
420 if (test_and_clear_bit(RXRPC_CALL_EV_PING, &call->events)) {
421 rxrpc_send_ack_packet(call, true, NULL);
422 goto recheck_state;
423 }
424
425 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND, &call->events)) {
426 rxrpc_resend(call, now);
427 goto recheck_state;
428 }
429
430 /* Make sure the timer is restarted */
431 next = call->expect_rx_by;
432
433 #define set(T) { t = READ_ONCE(T); if (time_before(t, next)) next = t; }
434
435 set(call->expect_req_by);
436 set(call->expect_term_by);
437 set(call->ack_at);
438 set(call->ack_lost_at);
439 set(call->resend_at);
440 set(call->keepalive_at);
441 set(call->ping_at);
442
443 now = jiffies;
444 if (time_after_eq(now, next))
445 goto recheck_state;
446
447 rxrpc_reduce_call_timer(call, next, now, rxrpc_timer_restart);
448
449 /* other events may have been raised since we started checking */
450 if (call->events && call->state < RXRPC_CALL_COMPLETE) {
451 __rxrpc_queue_call(call);
452 goto out;
453 }
454
455 out_put:
456 rxrpc_put_call(call, rxrpc_call_put);
457 out:
458 _leave("");
459 }
460
