?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / net / rxrpc / input.c
blobea506a77f3c811950b8c6ebbf8203c8d797766ca
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
30 static 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 */
42 static 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
159 resume_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;
168 out:
169 cumulative_acks = 0;
170 out_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
180 packet_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
186 send_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 */
203 static 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 */
219 static bool 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 bool rot_last = false;
224 int ix;
225 u8 annotation;
226
227 if (call->acks_lowest_nak == call->tx_hard_ack) {
228 call->acks_lowest_nak = to;
229 } else if (before_eq(call->acks_lowest_nak, to)) {
230 summary->new_low_nack = true;
231 call->acks_lowest_nak = to;
232 }
233
234 spin_lock(&call->lock);
235
236 while (before(call->tx_hard_ack, to)) {
237 call->tx_hard_ack++;
238 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
239 skb = call->rxtx_buffer[ix];
240 annotation = call->rxtx_annotations[ix];
241 rxrpc_see_skb(skb, rxrpc_skb_tx_rotated);
242 call->rxtx_buffer[ix] = NULL;
243 call->rxtx_annotations[ix] = 0;
244 skb->next = list;
245 list = skb;
246
247 if (annotation & RXRPC_TX_ANNO_LAST) {
248 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
249 rot_last = true;
250 }
251 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
252 summary->nr_rot_new_acks++;
253 }
254
255 spin_unlock(&call->lock);
256
257 trace_rxrpc_transmit(call, (rot_last ?
258 rxrpc_transmit_rotate_last :
259 rxrpc_transmit_rotate));
260 wake_up(&call->waitq);
261
262 while (list) {
263 skb = list;
264 list = skb->next;
265 skb->next = NULL;
266 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
267 }
268
269 return rot_last;
270 }
271
272 /*
273 * End the transmission phase of a call.
274 *
275 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
276 * or a final ACK packet.
277 */
278 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
279 const char *abort_why)
280 {
281
282 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
283
284 write_lock(&call->state_lock);
285
286 switch (call->state) {
287 case RXRPC_CALL_CLIENT_SEND_REQUEST:
288 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
289 if (reply_begun)
290 call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
291 else
292 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
293 break;
294
295 case RXRPC_CALL_SERVER_AWAIT_ACK:
296 __rxrpc_call_completed(call);
297 rxrpc_notify_socket(call);
298 break;
299
300 default:
301 goto bad_state;
302 }
303
304 write_unlock(&call->state_lock);
305 if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) {
306 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, 0, false, true,
307 rxrpc_propose_ack_client_tx_end);
308 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
309 } else {
310 trace_rxrpc_transmit(call, rxrpc_transmit_end);
311 }
312 _leave(" = ok");
313 return true;
314
315 bad_state:
316 write_unlock(&call->state_lock);
317 kdebug("end_tx %s", rxrpc_call_states[call->state]);
318 rxrpc_proto_abort(abort_why, call, call->tx_top);
319 return false;
320 }
321
322 /*
323 * Begin the reply reception phase of a call.
324 */
325 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
326 {
327 struct rxrpc_ack_summary summary = { 0 };
328 rxrpc_seq_t top = READ_ONCE(call->tx_top);
329
330 if (call->ackr_reason) {
331 spin_lock_bh(&call->lock);
332 call->ackr_reason = 0;
333 call->resend_at = call->expire_at;
334 call->ack_at = call->expire_at;
335 spin_unlock_bh(&call->lock);
336 rxrpc_set_timer(call, rxrpc_timer_init_for_reply,
337 ktime_get_real());
338 }
339
340 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
341 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
342 rxrpc_proto_abort("TXL", call, top);
343 return false;
344 }
345 }
346 if (!rxrpc_end_tx_phase(call, true, "ETD"))
347 return false;
348 call->tx_phase = false;
349 return true;
350 }
351
352 /*
353 * Scan a jumbo packet to validate its structure and to work out how many
354 * subpackets it contains.
355 *
356 * A jumbo packet is a collection of consecutive packets glued together with
357 * little headers between that indicate how to change the initial header for
358 * each subpacket.
359 *
360 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
361 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
362 * size.
363 */
364 static bool rxrpc_validate_jumbo(struct sk_buff *skb)
365 {
366 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
367 unsigned int offset = sizeof(struct rxrpc_wire_header);
368 unsigned int len = skb->len;
369 int nr_jumbo = 1;
370 u8 flags = sp->hdr.flags;
371
372 do {
373 nr_jumbo++;
374 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
375 goto protocol_error;
376 if (flags & RXRPC_LAST_PACKET)
377 goto protocol_error;
378 offset += RXRPC_JUMBO_DATALEN;
379 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
380 goto protocol_error;
381 offset += sizeof(struct rxrpc_jumbo_header);
382 } while (flags & RXRPC_JUMBO_PACKET);
383
384 sp->nr_jumbo = nr_jumbo;
385 return true;
386
387 protocol_error:
388 return false;
389 }
390
391 /*
392 * Handle reception of a duplicate packet.
393 *
394 * We have to take care to avoid an attack here whereby we're given a series of
395 * jumbograms, each with a sequence number one before the preceding one and
396 * filled up to maximum UDP size. If they never send us the first packet in
397 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
398 * space until the call times out.
399 *
400 * We limit the space usage by only accepting three duplicate jumbo packets per
401 * call. After that, we tell the other side we're no longer accepting jumbos
402 * (that information is encoded in the ACK packet).
403 */
404 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
405 u8 annotation, bool *_jumbo_bad)
406 {
407 /* Discard normal packets that are duplicates. */
408 if (annotation == 0)
409 return;
410
411 /* Skip jumbo subpackets that are duplicates. When we've had three or
412 * more partially duplicate jumbo packets, we refuse to take any more
413 * jumbos for this call.
414 */
415 if (!*_jumbo_bad) {
416 call->nr_jumbo_bad++;
417 *_jumbo_bad = true;
418 }
419 }
420
421 /*
422 * Process a DATA packet, adding the packet to the Rx ring.
423 */
424 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
425 u16 skew)
426 {
427 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
428 enum rxrpc_call_state state;
429 unsigned int offset = sizeof(struct rxrpc_wire_header);
430 unsigned int ix;
431 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
432 rxrpc_seq_t seq = sp->hdr.seq, hard_ack;
433 bool immediate_ack = false, jumbo_bad = false, queued;
434 u16 len;
435 u8 ack = 0, flags, annotation = 0;
436
437 _enter("{%u,%u},{%u,%u}",
438 call->rx_hard_ack, call->rx_top, skb->len, seq);
439
440 _proto("Rx DATA %%%u { #%u f=%02x }",
441 sp->hdr.serial, seq, sp->hdr.flags);
442
443 state = READ_ONCE(call->state);
444 if (state >= RXRPC_CALL_COMPLETE)
445 return;
446
447 /* Received data implicitly ACKs all of the request packets we sent
448 * when we're acting as a client.
449 */
450 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
451 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
452 !rxrpc_receiving_reply(call))
453 return;
454
455 call->ackr_prev_seq = seq;
456
457 hard_ack = READ_ONCE(call->rx_hard_ack);
458 if (after(seq, hard_ack + call->rx_winsize)) {
459 ack = RXRPC_ACK_EXCEEDS_WINDOW;
460 ack_serial = serial;
461 goto ack;
462 }
463
464 flags = sp->hdr.flags;
465 if (flags & RXRPC_JUMBO_PACKET) {
466 if (call->nr_jumbo_bad > 3) {
467 ack = RXRPC_ACK_NOSPACE;
468 ack_serial = serial;
469 goto ack;
470 }
471 annotation = 1;
472 }
473
474 next_subpacket:
475 queued = false;
476 ix = seq & RXRPC_RXTX_BUFF_MASK;
477 len = skb->len;
478 if (flags & RXRPC_JUMBO_PACKET)
479 len = RXRPC_JUMBO_DATALEN;
480
481 if (flags & RXRPC_LAST_PACKET) {
482 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
483 seq != call->rx_top)
484 return rxrpc_proto_abort("LSN", call, seq);
485 } else {
486 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
487 after_eq(seq, call->rx_top))
488 return rxrpc_proto_abort("LSA", call, seq);
489 }
490
491 trace_rxrpc_rx_data(call, seq, serial, flags, annotation);
492 if (before_eq(seq, hard_ack)) {
493 ack = RXRPC_ACK_DUPLICATE;
494 ack_serial = serial;
495 goto skip;
496 }
497
498 if (flags & RXRPC_REQUEST_ACK && !ack) {
499 ack = RXRPC_ACK_REQUESTED;
500 ack_serial = serial;
501 }
502
503 if (call->rxtx_buffer[ix]) {
504 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
505 if (ack != RXRPC_ACK_DUPLICATE) {
506 ack = RXRPC_ACK_DUPLICATE;
507 ack_serial = serial;
508 }
509 immediate_ack = true;
510 goto skip;
511 }
512
513 /* Queue the packet. We use a couple of memory barriers here as need
514 * to make sure that rx_top is perceived to be set after the buffer
515 * pointer and that the buffer pointer is set after the annotation and
516 * the skb data.
517 *
518 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
519 * and also rxrpc_fill_out_ack().
520 */
521 rxrpc_get_skb(skb, rxrpc_skb_rx_got);
522 call->rxtx_annotations[ix] = annotation;
523 smp_wmb();
524 call->rxtx_buffer[ix] = skb;
525 if (after(seq, call->rx_top)) {
526 smp_store_release(&call->rx_top, seq);
527 } else if (before(seq, call->rx_top)) {
528 /* Send an immediate ACK if we fill in a hole */
529 if (!ack) {
530 ack = RXRPC_ACK_DELAY;
531 ack_serial = serial;
532 }
533 immediate_ack = true;
534 }
535 if (flags & RXRPC_LAST_PACKET) {
536 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
537 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
538 } else {
539 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
540 }
541 queued = true;
542
543 if (after_eq(seq, call->rx_expect_next)) {
544 if (after(seq, call->rx_expect_next)) {
545 _net("OOS %u > %u", seq, call->rx_expect_next);
546 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
547 ack_serial = serial;
548 }
549 call->rx_expect_next = seq + 1;
550 }
551
552 skip:
553 offset += len;
554 if (flags & RXRPC_JUMBO_PACKET) {
555 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
556 return rxrpc_proto_abort("XJF", call, seq);
557 offset += sizeof(struct rxrpc_jumbo_header);
558 seq++;
559 serial++;
560 annotation++;
561 if (flags & RXRPC_JUMBO_PACKET)
562 annotation |= RXRPC_RX_ANNO_JLAST;
563 if (after(seq, hard_ack + call->rx_winsize)) {
564 ack = RXRPC_ACK_EXCEEDS_WINDOW;
565 ack_serial = serial;
566 if (!jumbo_bad) {
567 call->nr_jumbo_bad++;
568 jumbo_bad = true;
569 }
570 goto ack;
571 }
572
573 _proto("Rx DATA Jumbo %%%u", serial);
574 goto next_subpacket;
575 }
576
577 if (queued && flags & RXRPC_LAST_PACKET && !ack) {
578 ack = RXRPC_ACK_DELAY;
579 ack_serial = serial;
580 }
581
582 ack:
583 if (ack)
584 rxrpc_propose_ACK(call, ack, skew, ack_serial,
585 immediate_ack, true,
586 rxrpc_propose_ack_input_data);
587
588 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1)
589 rxrpc_notify_socket(call);
590 _leave(" [queued]");
591 }
592
593 /*
594 * Process a requested ACK.
595 */
596 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
597 ktime_t resp_time,
598 rxrpc_serial_t orig_serial,
599 rxrpc_serial_t ack_serial)
600 {
601 struct rxrpc_skb_priv *sp;
602 struct sk_buff *skb;
603 ktime_t sent_at;
604 int ix;
605
606 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
607 skb = call->rxtx_buffer[ix];
608 if (!skb)
609 continue;
610
611 sp = rxrpc_skb(skb);
612 if (sp->hdr.serial != orig_serial)
613 continue;
614 smp_rmb();
615 sent_at = skb->tstamp;
616 goto found;
617 }
618 return;
619
620 found:
621 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
622 orig_serial, ack_serial, sent_at, resp_time);
623 }
624
625 /*
626 * Process a ping response.
627 */
628 static void rxrpc_input_ping_response(struct rxrpc_call *call,
629 ktime_t resp_time,
630 rxrpc_serial_t orig_serial,
631 rxrpc_serial_t ack_serial)
632 {
633 rxrpc_serial_t ping_serial;
634 ktime_t ping_time;
635
636 ping_time = call->ping_time;
637 smp_rmb();
638 ping_serial = call->ping_serial;
639
640 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) ||
641 before(orig_serial, ping_serial))
642 return;
643 clear_bit(RXRPC_CALL_PINGING, &call->flags);
644 if (after(orig_serial, ping_serial))
645 return;
646
647 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
648 orig_serial, ack_serial, ping_time, resp_time);
649 }
650
651 /*
652 * Process the extra information that may be appended to an ACK packet
653 */
654 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
655 struct rxrpc_ackinfo *ackinfo)
656 {
657 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
658 struct rxrpc_peer *peer;
659 unsigned int mtu;
660 bool wake = false;
661 u32 rwind = ntohl(ackinfo->rwind);
662
663 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
664 sp->hdr.serial,
665 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
666 rwind, ntohl(ackinfo->jumbo_max));
667
668 if (call->tx_winsize != rwind) {
669 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
670 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
671 if (rwind > call->tx_winsize)
672 wake = true;
673 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
674 ntohl(ackinfo->rwind), wake);
675 call->tx_winsize = rwind;
676 }
677
678 if (call->cong_ssthresh > rwind)
679 call->cong_ssthresh = rwind;
680
681 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
682
683 peer = call->peer;
684 if (mtu < peer->maxdata) {
685 spin_lock_bh(&peer->lock);
686 peer->maxdata = mtu;
687 peer->mtu = mtu + peer->hdrsize;
688 spin_unlock_bh(&peer->lock);
689 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
690 }
691
692 if (wake)
693 wake_up(&call->waitq);
694 }
695
696 /*
697 * Process individual soft ACKs.
698 *
699 * Each ACK in the array corresponds to one packet and can be either an ACK or
700 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
701 * packets that lie beyond the end of the ACK list are scheduled for resend by
702 * the timer on the basis that the peer might just not have processed them at
703 * the time the ACK was sent.
704 */
705 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
706 rxrpc_seq_t seq, int nr_acks,
707 struct rxrpc_ack_summary *summary)
708 {
709 int ix;
710 u8 annotation, anno_type;
711
712 for (; nr_acks > 0; nr_acks--, seq++) {
713 ix = seq & RXRPC_RXTX_BUFF_MASK;
714 annotation = call->rxtx_annotations[ix];
715 anno_type = annotation & RXRPC_TX_ANNO_MASK;
716 annotation &= ~RXRPC_TX_ANNO_MASK;
717 switch (*acks++) {
718 case RXRPC_ACK_TYPE_ACK:
719 summary->nr_acks++;
720 if (anno_type == RXRPC_TX_ANNO_ACK)
721 continue;
722 summary->nr_new_acks++;
723 call->rxtx_annotations[ix] =
724 RXRPC_TX_ANNO_ACK | annotation;
725 break;
726 case RXRPC_ACK_TYPE_NACK:
727 if (!summary->nr_nacks &&
728 call->acks_lowest_nak != seq) {
729 call->acks_lowest_nak = seq;
730 summary->new_low_nack = true;
731 }
732 summary->nr_nacks++;
733 if (anno_type == RXRPC_TX_ANNO_NAK)
734 continue;
735 summary->nr_new_nacks++;
736 if (anno_type == RXRPC_TX_ANNO_RETRANS)
737 continue;
738 call->rxtx_annotations[ix] =
739 RXRPC_TX_ANNO_NAK | annotation;
740 break;
741 default:
742 return rxrpc_proto_abort("SFT", call, 0);
743 }
744 }
745 }
746
747 /*
748 * Process an ACK packet.
749 *
750 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
751 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
752 *
753 * A hard-ACK means that a packet has been processed and may be discarded; a
754 * soft-ACK means that the packet may be discarded and retransmission
755 * requested. A phase is complete when all packets are hard-ACK'd.
756 */
757 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
758 u16 skew)
759 {
760 struct rxrpc_ack_summary summary = { 0 };
761 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
762 union {
763 struct rxrpc_ackpacket ack;
764 struct rxrpc_ackinfo info;
765 u8 acks[RXRPC_MAXACKS];
766 } buf;
767 rxrpc_serial_t acked_serial;
768 rxrpc_seq_t first_soft_ack, hard_ack;
769 int nr_acks, offset, ioffset;
770
771 _enter("");
772
773 offset = sizeof(struct rxrpc_wire_header);
774 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
775 _debug("extraction failure");
776 return rxrpc_proto_abort("XAK", call, 0);
777 }
778 offset += sizeof(buf.ack);
779
780 acked_serial = ntohl(buf.ack.serial);
781 first_soft_ack = ntohl(buf.ack.firstPacket);
782 hard_ack = first_soft_ack - 1;
783 nr_acks = buf.ack.nAcks;
784 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
785 buf.ack.reason : RXRPC_ACK__INVALID);
786
787 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
788 first_soft_ack, ntohl(buf.ack.previousPacket),
789 summary.ack_reason, nr_acks);
790
791 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
792 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
793 sp->hdr.serial);
794 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
795 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
796 sp->hdr.serial);
797
798 if (buf.ack.reason == RXRPC_ACK_PING) {
799 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
800 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
801 skew, sp->hdr.serial, true, true,
802 rxrpc_propose_ack_respond_to_ping);
803 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
804 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
805 skew, sp->hdr.serial, true, true,
806 rxrpc_propose_ack_respond_to_ack);
807 }
808
809 /* Discard any out-of-order or duplicate ACKs. */
810 if (before_eq(sp->hdr.serial, call->acks_latest)) {
811 _debug("discard ACK %d <= %d",
812 sp->hdr.serial, call->acks_latest);
813 return;
814 }
815 call->acks_latest_ts = skb->tstamp;
816 call->acks_latest = sp->hdr.serial;
817
818 /* Parse rwind and mtu sizes if provided. */
819 ioffset = offset + nr_acks + 3;
820 if (skb->len >= ioffset + sizeof(buf.info)) {
821 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
822 return rxrpc_proto_abort("XAI", call, 0);
823 rxrpc_input_ackinfo(call, skb, &buf.info);
824 }
825
826 if (first_soft_ack == 0)
827 return rxrpc_proto_abort("AK0", call, 0);
828
829 /* Ignore ACKs unless we are or have just been transmitting. */
830 switch (READ_ONCE(call->state)) {
831 case RXRPC_CALL_CLIENT_SEND_REQUEST:
832 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
833 case RXRPC_CALL_SERVER_SEND_REPLY:
834 case RXRPC_CALL_SERVER_AWAIT_ACK:
835 break;
836 default:
837 return;
838 }
839
840 if (before(hard_ack, call->tx_hard_ack) ||
841 after(hard_ack, call->tx_top))
842 return rxrpc_proto_abort("AKW", call, 0);
843 if (nr_acks > call->tx_top - hard_ack)
844 return rxrpc_proto_abort("AKN", call, 0);
845
846 if (after(hard_ack, call->tx_hard_ack)) {
847 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
848 rxrpc_end_tx_phase(call, false, "ETA");
849 return;
850 }
851 }
852
853 if (nr_acks > 0) {
854 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0)
855 return rxrpc_proto_abort("XSA", call, 0);
856 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
857 &summary);
858 }
859
860 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
861 RXRPC_TX_ANNO_LAST &&
862 summary.nr_acks == call->tx_top - hard_ack &&
863 rxrpc_is_client_call(call))
864 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
865 false, true,
866 rxrpc_propose_ack_ping_for_lost_reply);
867
868 return rxrpc_congestion_management(call, skb, &summary, acked_serial);
869 }
870
871 /*
872 * Process an ACKALL packet.
873 */
874 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
875 {
876 struct rxrpc_ack_summary summary = { 0 };
877 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
878
879 _proto("Rx ACKALL %%%u", sp->hdr.serial);
880
881 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
882 rxrpc_end_tx_phase(call, false, "ETL");
883 }
884
885 /*
886 * Process an ABORT packet directed at a call.
887 */
888 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
889 {
890 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
891 __be32 wtmp;
892 u32 abort_code = RX_CALL_DEAD;
893
894 _enter("");
895
896 if (skb->len >= 4 &&
897 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
898 &wtmp, sizeof(wtmp)) >= 0)
899 abort_code = ntohl(wtmp);
900
901 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
902
903 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
904
905 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
906 abort_code, -ECONNABORTED))
907 rxrpc_notify_socket(call);
908 }
909
910 /*
911 * Process an incoming call packet.
912 */
913 static void rxrpc_input_call_packet(struct rxrpc_call *call,
914 struct sk_buff *skb, u16 skew)
915 {
916 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
917
918 _enter("%p,%p", call, skb);
919
920 switch (sp->hdr.type) {
921 case RXRPC_PACKET_TYPE_DATA:
922 rxrpc_input_data(call, skb, skew);
923 break;
924
925 case RXRPC_PACKET_TYPE_ACK:
926 rxrpc_input_ack(call, skb, skew);
927 break;
928
929 case RXRPC_PACKET_TYPE_BUSY:
930 _proto("Rx BUSY %%%u", sp->hdr.serial);
931
932 /* Just ignore BUSY packets from the server; the retry and
933 * lifespan timers will take care of business. BUSY packets
934 * from the client don't make sense.
935 */
936 break;
937
938 case RXRPC_PACKET_TYPE_ABORT:
939 rxrpc_input_abort(call, skb);
940 break;
941
942 case RXRPC_PACKET_TYPE_ACKALL:
943 rxrpc_input_ackall(call, skb);
944 break;
945
946 default:
947 break;
948 }
949
950 _leave("");
951 }
952
953 /*
954 * Handle a new call on a channel implicitly completing the preceding call on
955 * that channel.
956 *
957 * TODO: If callNumber > call_id + 1, renegotiate security.
958 */
959 static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn,
960 struct rxrpc_call *call)
961 {
962 switch (READ_ONCE(call->state)) {
963 case RXRPC_CALL_SERVER_AWAIT_ACK:
964 rxrpc_call_completed(call);
965 break;
966 case RXRPC_CALL_COMPLETE:
967 break;
968 default:
969 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
970 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
971 rxrpc_queue_call(call);
972 }
973 break;
974 }
975
976 trace_rxrpc_improper_term(call);
977 __rxrpc_disconnect_call(conn, call);
978 rxrpc_notify_socket(call);
979 }
980
981 /*
982 * post connection-level events to the connection
983 * - this includes challenges, responses, some aborts and call terminal packet
984 * retransmission.
985 */
986 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
987 struct sk_buff *skb)
988 {
989 _enter("%p,%p", conn, skb);
990
991 skb_queue_tail(&conn->rx_queue, skb);
992 rxrpc_queue_conn(conn);
993 }
994
995 /*
996 * post endpoint-level events to the local endpoint
997 * - this includes debug and version messages
998 */
999 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1000 struct sk_buff *skb)
1001 {
1002 _enter("%p,%p", local, skb);
1003
1004 skb_queue_tail(&local->event_queue, skb);
1005 rxrpc_queue_local(local);
1006 }
1007
1008 /*
1009 * put a packet up for transport-level abort
1010 */
1011 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1012 {
1013 CHECK_SLAB_OKAY(&local->usage);
1014
1015 skb_queue_tail(&local->reject_queue, skb);
1016 rxrpc_queue_local(local);
1017 }
1018
1019 /*
1020 * Extract the wire header from a packet and translate the byte order.
1021 */
1022 static noinline
1023 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1024 {
1025 struct rxrpc_wire_header whdr;
1026
1027 /* dig out the RxRPC connection details */
1028 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1029 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1030 tracepoint_string("bad_hdr"));
1031 return -EBADMSG;
1032 }
1033
1034 memset(sp, 0, sizeof(*sp));
1035 sp->hdr.epoch = ntohl(whdr.epoch);
1036 sp->hdr.cid = ntohl(whdr.cid);
1037 sp->hdr.callNumber = ntohl(whdr.callNumber);
1038 sp->hdr.seq = ntohl(whdr.seq);
1039 sp->hdr.serial = ntohl(whdr.serial);
1040 sp->hdr.flags = whdr.flags;
1041 sp->hdr.type = whdr.type;
1042 sp->hdr.userStatus = whdr.userStatus;
1043 sp->hdr.securityIndex = whdr.securityIndex;
1044 sp->hdr._rsvd = ntohs(whdr._rsvd);
1045 sp->hdr.serviceId = ntohs(whdr.serviceId);
1046 return 0;
1047 }
1048
1049 /*
1050 * handle data received on the local endpoint
1051 * - may be called in interrupt context
1052 *
1053 * The socket is locked by the caller and this prevents the socket from being
1054 * shut down and the local endpoint from going away, thus sk_user_data will not
1055 * be cleared until this function returns.
1056 */
1057 void rxrpc_data_ready(struct sock *udp_sk)
1058 {
1059 struct rxrpc_connection *conn;
1060 struct rxrpc_channel *chan;
1061 struct rxrpc_call *call;
1062 struct rxrpc_skb_priv *sp;
1063 struct rxrpc_local *local = udp_sk->sk_user_data;
1064 struct sk_buff *skb;
1065 unsigned int channel;
1066 int ret, skew;
1067
1068 _enter("%p", udp_sk);
1069
1070 ASSERT(!irqs_disabled());
1071
1072 skb = skb_recv_udp(udp_sk, 0, 1, &ret);
1073 if (!skb) {
1074 if (ret == -EAGAIN)
1075 return;
1076 _debug("UDP socket error %d", ret);
1077 return;
1078 }
1079
1080 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
1081
1082 _net("recv skb %p", skb);
1083
1084 /* we'll probably need to checksum it (didn't call sock_recvmsg) */
1085 if (skb_checksum_complete(skb)) {
1086 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1087 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
1088 _leave(" [CSUM failed]");
1089 return;
1090 }
1091
1092 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0);
1093
1094 /* The UDP protocol already released all skb resources;
1095 * we are free to add our own data there.
1096 */
1097 sp = rxrpc_skb(skb);
1098
1099 /* dig out the RxRPC connection details */
1100 if (rxrpc_extract_header(sp, skb) < 0)
1101 goto bad_message;
1102
1103 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1104 static int lose;
1105 if ((lose++ & 7) == 7) {
1106 trace_rxrpc_rx_lose(sp);
1107 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost);
1108 return;
1109 }
1110 }
1111
1112 trace_rxrpc_rx_packet(sp);
1113
1114 _net("Rx RxRPC %s ep=%x call=%x:%x",
1115 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient",
1116 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber);
1117
1118 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES ||
1119 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) {
1120 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1121 goto bad_message;
1122 }
1123
1124 switch (sp->hdr.type) {
1125 case RXRPC_PACKET_TYPE_VERSION:
1126 rxrpc_post_packet_to_local(local, skb);
1127 goto out;
1128
1129 case RXRPC_PACKET_TYPE_BUSY:
1130 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED)
1131 goto discard;
1132
1133 case RXRPC_PACKET_TYPE_DATA:
1134 if (sp->hdr.callNumber == 0)
1135 goto bad_message;
1136 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
1137 !rxrpc_validate_jumbo(skb))
1138 goto bad_message;
1139 break;
1140 }
1141
1142 rcu_read_lock();
1143
1144 conn = rxrpc_find_connection_rcu(local, skb);
1145 if (conn) {
1146 if (sp->hdr.securityIndex != conn->security_ix)
1147 goto wrong_security;
1148
1149 if (sp->hdr.serviceId != conn->service_id) {
1150 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) ||
1151 conn->service_id != conn->params.service_id)
1152 goto reupgrade;
1153 conn->service_id = sp->hdr.serviceId;
1154 }
1155
1156 if (sp->hdr.callNumber == 0) {
1157 /* Connection-level packet */
1158 _debug("CONN %p {%d}", conn, conn->debug_id);
1159 rxrpc_post_packet_to_conn(conn, skb);
1160 goto out_unlock;
1161 }
1162
1163 /* Note the serial number skew here */
1164 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
1165 if (skew >= 0) {
1166 if (skew > 0)
1167 conn->hi_serial = sp->hdr.serial;
1168 } else {
1169 skew = -skew;
1170 skew = min(skew, 65535);
1171 }
1172
1173 /* Call-bound packets are routed by connection channel. */
1174 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1175 chan = &conn->channels[channel];
1176
1177 /* Ignore really old calls */
1178 if (sp->hdr.callNumber < chan->last_call)
1179 goto discard_unlock;
1180
1181 if (sp->hdr.callNumber == chan->last_call) {
1182 if (chan->call ||
1183 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1184 goto discard_unlock;
1185
1186 /* For the previous service call, if completed
1187 * successfully, we discard all further packets.
1188 */
1189 if (rxrpc_conn_is_service(conn) &&
1190 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1191 goto discard_unlock;
1192
1193 /* But otherwise we need to retransmit the final packet
1194 * from data cached in the connection record.
1195 */
1196 rxrpc_post_packet_to_conn(conn, skb);
1197 goto out_unlock;
1198 }
1199
1200 call = rcu_dereference(chan->call);
1201
1202 if (sp->hdr.callNumber > chan->call_id) {
1203 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) {
1204 rcu_read_unlock();
1205 goto reject_packet;
1206 }
1207 if (call)
1208 rxrpc_input_implicit_end_call(conn, call);
1209 call = NULL;
1210 }
1211
1212 if (call && sp->hdr.serviceId != call->service_id)
1213 call->service_id = sp->hdr.serviceId;
1214 } else {
1215 skew = 0;
1216 call = NULL;
1217 }
1218
1219 if (!call || atomic_read(&call->usage) == 0) {
1220 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) ||
1221 sp->hdr.callNumber == 0 ||
1222 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1223 goto bad_message_unlock;
1224 if (sp->hdr.seq != 1)
1225 goto discard_unlock;
1226 call = rxrpc_new_incoming_call(local, conn, skb);
1227 if (!call) {
1228 rcu_read_unlock();
1229 goto reject_packet;
1230 }
1231 rxrpc_send_ping(call, skb, skew);
1232 mutex_unlock(&call->user_mutex);
1233 }
1234
1235 rxrpc_input_call_packet(call, skb, skew);
1236 goto discard_unlock;
1237
1238 discard_unlock:
1239 rcu_read_unlock();
1240 discard:
1241 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1242 out:
1243 trace_rxrpc_rx_done(0, 0);
1244 return;
1245
1246 out_unlock:
1247 rcu_read_unlock();
1248 goto out;
1249
1250 wrong_security:
1251 rcu_read_unlock();
1252 trace_rxrpc_abort("SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1253 RXKADINCONSISTENCY, EBADMSG);
1254 skb->priority = RXKADINCONSISTENCY;
1255 goto post_abort;
1256
1257 reupgrade:
1258 rcu_read_unlock();
1259 trace_rxrpc_abort("UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1260 RX_PROTOCOL_ERROR, EBADMSG);
1261 goto protocol_error;
1262
1263 bad_message_unlock:
1264 rcu_read_unlock();
1265 bad_message:
1266 trace_rxrpc_abort("BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1267 RX_PROTOCOL_ERROR, EBADMSG);
1268 protocol_error:
1269 skb->priority = RX_PROTOCOL_ERROR;
1270 post_abort:
1271 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
1272 reject_packet:
1273 trace_rxrpc_rx_done(skb->mark, skb->priority);
1274 rxrpc_reject_packet(local, skb);
1275 _leave(" [badmsg]");
1276 }
Linux with added FPC-III support