]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/rxrpc/input.c
Merge tag 'iommu-fix-v4.11-rc0' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-artful-kernel.git] / net / rxrpc / input.c
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 void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
220 struct rxrpc_ack_summary *summary)
221 {
222 struct sk_buff *skb, *list = NULL;
223 int ix;
224 u8 annotation;
225
226 if (call->acks_lowest_nak == call->tx_hard_ack) {
227 call->acks_lowest_nak = to;
228 } else if (before_eq(call->acks_lowest_nak, to)) {
229 summary->new_low_nack = true;
230 call->acks_lowest_nak = to;
231 }
232
233 spin_lock(&call->lock);
234
235 while (before(call->tx_hard_ack, to)) {
236 call->tx_hard_ack++;
237 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
238 skb = call->rxtx_buffer[ix];
239 annotation = call->rxtx_annotations[ix];
240 rxrpc_see_skb(skb, rxrpc_skb_tx_rotated);
241 call->rxtx_buffer[ix] = NULL;
242 call->rxtx_annotations[ix] = 0;
243 skb->next = list;
244 list = skb;
245
246 if (annotation & RXRPC_TX_ANNO_LAST)
247 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
248 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
249 summary->nr_rot_new_acks++;
250 }
251
252 spin_unlock(&call->lock);
253
254 trace_rxrpc_transmit(call, (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ?
255 rxrpc_transmit_rotate_last :
256 rxrpc_transmit_rotate));
257 wake_up(&call->waitq);
258
259 while (list) {
260 skb = list;
261 list = skb->next;
262 skb->next = NULL;
263 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
264 }
265 }
266
267 /*
268 * End the transmission phase of a call.
269 *
270 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
271 * or a final ACK packet.
272 */
273 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
274 const char *abort_why)
275 {
276
277 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
278
279 write_lock(&call->state_lock);
280
281 switch (call->state) {
282 case RXRPC_CALL_CLIENT_SEND_REQUEST:
283 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
284 if (reply_begun)
285 call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
286 else
287 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
288 break;
289
290 case RXRPC_CALL_SERVER_AWAIT_ACK:
291 __rxrpc_call_completed(call);
292 rxrpc_notify_socket(call);
293 break;
294
295 default:
296 goto bad_state;
297 }
298
299 write_unlock(&call->state_lock);
300 if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) {
301 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, 0, false, true,
302 rxrpc_propose_ack_client_tx_end);
303 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
304 } else {
305 trace_rxrpc_transmit(call, rxrpc_transmit_end);
306 }
307 _leave(" = ok");
308 return true;
309
310 bad_state:
311 write_unlock(&call->state_lock);
312 kdebug("end_tx %s", rxrpc_call_states[call->state]);
313 rxrpc_proto_abort(abort_why, call, call->tx_top);
314 return false;
315 }
316
317 /*
318 * Begin the reply reception phase of a call.
319 */
320 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
321 {
322 struct rxrpc_ack_summary summary = { 0 };
323 rxrpc_seq_t top = READ_ONCE(call->tx_top);
324
325 if (call->ackr_reason) {
326 spin_lock_bh(&call->lock);
327 call->ackr_reason = 0;
328 call->resend_at = call->expire_at;
329 call->ack_at = call->expire_at;
330 spin_unlock_bh(&call->lock);
331 rxrpc_set_timer(call, rxrpc_timer_init_for_reply,
332 ktime_get_real());
333 }
334
335 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags))
336 rxrpc_rotate_tx_window(call, top, &summary);
337 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
338 rxrpc_proto_abort("TXL", call, top);
339 return false;
340 }
341 if (!rxrpc_end_tx_phase(call, true, "ETD"))
342 return false;
343 call->tx_phase = false;
344 return true;
345 }
346
347 /*
348 * Scan a jumbo packet to validate its structure and to work out how many
349 * subpackets it contains.
350 *
351 * A jumbo packet is a collection of consecutive packets glued together with
352 * little headers between that indicate how to change the initial header for
353 * each subpacket.
354 *
355 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
356 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
357 * size.
358 */
359 static bool rxrpc_validate_jumbo(struct sk_buff *skb)
360 {
361 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
362 unsigned int offset = sizeof(struct rxrpc_wire_header);
363 unsigned int len = skb->len;
364 int nr_jumbo = 1;
365 u8 flags = sp->hdr.flags;
366
367 do {
368 nr_jumbo++;
369 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
370 goto protocol_error;
371 if (flags & RXRPC_LAST_PACKET)
372 goto protocol_error;
373 offset += RXRPC_JUMBO_DATALEN;
374 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
375 goto protocol_error;
376 offset += sizeof(struct rxrpc_jumbo_header);
377 } while (flags & RXRPC_JUMBO_PACKET);
378
379 sp->nr_jumbo = nr_jumbo;
380 return true;
381
382 protocol_error:
383 return false;
384 }
385
386 /*
387 * Handle reception of a duplicate packet.
388 *
389 * We have to take care to avoid an attack here whereby we're given a series of
390 * jumbograms, each with a sequence number one before the preceding one and
391 * filled up to maximum UDP size. If they never send us the first packet in
392 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
393 * space until the call times out.
394 *
395 * We limit the space usage by only accepting three duplicate jumbo packets per
396 * call. After that, we tell the other side we're no longer accepting jumbos
397 * (that information is encoded in the ACK packet).
398 */
399 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
400 u8 annotation, bool *_jumbo_bad)
401 {
402 /* Discard normal packets that are duplicates. */
403 if (annotation == 0)
404 return;
405
406 /* Skip jumbo subpackets that are duplicates. When we've had three or
407 * more partially duplicate jumbo packets, we refuse to take any more
408 * jumbos for this call.
409 */
410 if (!*_jumbo_bad) {
411 call->nr_jumbo_bad++;
412 *_jumbo_bad = true;
413 }
414 }
415
416 /*
417 * Process a DATA packet, adding the packet to the Rx ring.
418 */
419 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
420 u16 skew)
421 {
422 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
423 unsigned int offset = sizeof(struct rxrpc_wire_header);
424 unsigned int ix;
425 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
426 rxrpc_seq_t seq = sp->hdr.seq, hard_ack;
427 bool immediate_ack = false, jumbo_bad = false, queued;
428 u16 len;
429 u8 ack = 0, flags, annotation = 0;
430
431 _enter("{%u,%u},{%u,%u}",
432 call->rx_hard_ack, call->rx_top, skb->len, seq);
433
434 _proto("Rx DATA %%%u { #%u f=%02x }",
435 sp->hdr.serial, seq, sp->hdr.flags);
436
437 if (call->state >= RXRPC_CALL_COMPLETE)
438 return;
439
440 /* Received data implicitly ACKs all of the request packets we sent
441 * when we're acting as a client.
442 */
443 if ((call->state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
444 call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
445 !rxrpc_receiving_reply(call))
446 return;
447
448 call->ackr_prev_seq = seq;
449
450 hard_ack = READ_ONCE(call->rx_hard_ack);
451 if (after(seq, hard_ack + call->rx_winsize)) {
452 ack = RXRPC_ACK_EXCEEDS_WINDOW;
453 ack_serial = serial;
454 goto ack;
455 }
456
457 flags = sp->hdr.flags;
458 if (flags & RXRPC_JUMBO_PACKET) {
459 if (call->nr_jumbo_bad > 3) {
460 ack = RXRPC_ACK_NOSPACE;
461 ack_serial = serial;
462 goto ack;
463 }
464 annotation = 1;
465 }
466
467 next_subpacket:
468 queued = false;
469 ix = seq & RXRPC_RXTX_BUFF_MASK;
470 len = skb->len;
471 if (flags & RXRPC_JUMBO_PACKET)
472 len = RXRPC_JUMBO_DATALEN;
473
474 if (flags & RXRPC_LAST_PACKET) {
475 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
476 seq != call->rx_top)
477 return rxrpc_proto_abort("LSN", call, seq);
478 } else {
479 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
480 after_eq(seq, call->rx_top))
481 return rxrpc_proto_abort("LSA", call, seq);
482 }
483
484 trace_rxrpc_rx_data(call, seq, serial, flags, annotation);
485 if (before_eq(seq, hard_ack)) {
486 ack = RXRPC_ACK_DUPLICATE;
487 ack_serial = serial;
488 goto skip;
489 }
490
491 if (flags & RXRPC_REQUEST_ACK && !ack) {
492 ack = RXRPC_ACK_REQUESTED;
493 ack_serial = serial;
494 }
495
496 if (call->rxtx_buffer[ix]) {
497 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
498 if (ack != RXRPC_ACK_DUPLICATE) {
499 ack = RXRPC_ACK_DUPLICATE;
500 ack_serial = serial;
501 }
502 immediate_ack = true;
503 goto skip;
504 }
505
506 /* Queue the packet. We use a couple of memory barriers here as need
507 * to make sure that rx_top is perceived to be set after the buffer
508 * pointer and that the buffer pointer is set after the annotation and
509 * the skb data.
510 *
511 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
512 * and also rxrpc_fill_out_ack().
513 */
514 rxrpc_get_skb(skb, rxrpc_skb_rx_got);
515 call->rxtx_annotations[ix] = annotation;
516 smp_wmb();
517 call->rxtx_buffer[ix] = skb;
518 if (after(seq, call->rx_top)) {
519 smp_store_release(&call->rx_top, seq);
520 } else if (before(seq, call->rx_top)) {
521 /* Send an immediate ACK if we fill in a hole */
522 if (!ack) {
523 ack = RXRPC_ACK_DELAY;
524 ack_serial = serial;
525 }
526 immediate_ack = true;
527 }
528 if (flags & RXRPC_LAST_PACKET) {
529 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
530 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
531 } else {
532 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
533 }
534 queued = true;
535
536 if (after_eq(seq, call->rx_expect_next)) {
537 if (after(seq, call->rx_expect_next)) {
538 _net("OOS %u > %u", seq, call->rx_expect_next);
539 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
540 ack_serial = serial;
541 }
542 call->rx_expect_next = seq + 1;
543 }
544
545 skip:
546 offset += len;
547 if (flags & RXRPC_JUMBO_PACKET) {
548 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
549 return rxrpc_proto_abort("XJF", call, seq);
550 offset += sizeof(struct rxrpc_jumbo_header);
551 seq++;
552 serial++;
553 annotation++;
554 if (flags & RXRPC_JUMBO_PACKET)
555 annotation |= RXRPC_RX_ANNO_JLAST;
556 if (after(seq, hard_ack + call->rx_winsize)) {
557 ack = RXRPC_ACK_EXCEEDS_WINDOW;
558 ack_serial = serial;
559 if (!jumbo_bad) {
560 call->nr_jumbo_bad++;
561 jumbo_bad = true;
562 }
563 goto ack;
564 }
565
566 _proto("Rx DATA Jumbo %%%u", serial);
567 goto next_subpacket;
568 }
569
570 if (queued && flags & RXRPC_LAST_PACKET && !ack) {
571 ack = RXRPC_ACK_DELAY;
572 ack_serial = serial;
573 }
574
575 ack:
576 if (ack)
577 rxrpc_propose_ACK(call, ack, skew, ack_serial,
578 immediate_ack, true,
579 rxrpc_propose_ack_input_data);
580
581 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1)
582 rxrpc_notify_socket(call);
583 _leave(" [queued]");
584 }
585
586 /*
587 * Process a requested ACK.
588 */
589 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
590 ktime_t resp_time,
591 rxrpc_serial_t orig_serial,
592 rxrpc_serial_t ack_serial)
593 {
594 struct rxrpc_skb_priv *sp;
595 struct sk_buff *skb;
596 ktime_t sent_at;
597 int ix;
598
599 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
600 skb = call->rxtx_buffer[ix];
601 if (!skb)
602 continue;
603
604 sp = rxrpc_skb(skb);
605 if (sp->hdr.serial != orig_serial)
606 continue;
607 smp_rmb();
608 sent_at = skb->tstamp;
609 goto found;
610 }
611 return;
612
613 found:
614 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
615 orig_serial, ack_serial, sent_at, resp_time);
616 }
617
618 /*
619 * Process a ping response.
620 */
621 static void rxrpc_input_ping_response(struct rxrpc_call *call,
622 ktime_t resp_time,
623 rxrpc_serial_t orig_serial,
624 rxrpc_serial_t ack_serial)
625 {
626 rxrpc_serial_t ping_serial;
627 ktime_t ping_time;
628
629 ping_time = call->ping_time;
630 smp_rmb();
631 ping_serial = call->ping_serial;
632
633 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) ||
634 before(orig_serial, ping_serial))
635 return;
636 clear_bit(RXRPC_CALL_PINGING, &call->flags);
637 if (after(orig_serial, ping_serial))
638 return;
639
640 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
641 orig_serial, ack_serial, ping_time, resp_time);
642 }
643
644 /*
645 * Process the extra information that may be appended to an ACK packet
646 */
647 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
648 struct rxrpc_ackinfo *ackinfo)
649 {
650 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
651 struct rxrpc_peer *peer;
652 unsigned int mtu;
653 u32 rwind = ntohl(ackinfo->rwind);
654
655 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
656 sp->hdr.serial,
657 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
658 rwind, ntohl(ackinfo->jumbo_max));
659
660 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
661 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
662 call->tx_winsize = rwind;
663 if (call->cong_ssthresh > rwind)
664 call->cong_ssthresh = rwind;
665
666 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
667
668 peer = call->peer;
669 if (mtu < peer->maxdata) {
670 spin_lock_bh(&peer->lock);
671 peer->maxdata = mtu;
672 peer->mtu = mtu + peer->hdrsize;
673 spin_unlock_bh(&peer->lock);
674 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
675 }
676 }
677
678 /*
679 * Process individual soft ACKs.
680 *
681 * Each ACK in the array corresponds to one packet and can be either an ACK or
682 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
683 * packets that lie beyond the end of the ACK list are scheduled for resend by
684 * the timer on the basis that the peer might just not have processed them at
685 * the time the ACK was sent.
686 */
687 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
688 rxrpc_seq_t seq, int nr_acks,
689 struct rxrpc_ack_summary *summary)
690 {
691 int ix;
692 u8 annotation, anno_type;
693
694 for (; nr_acks > 0; nr_acks--, seq++) {
695 ix = seq & RXRPC_RXTX_BUFF_MASK;
696 annotation = call->rxtx_annotations[ix];
697 anno_type = annotation & RXRPC_TX_ANNO_MASK;
698 annotation &= ~RXRPC_TX_ANNO_MASK;
699 switch (*acks++) {
700 case RXRPC_ACK_TYPE_ACK:
701 summary->nr_acks++;
702 if (anno_type == RXRPC_TX_ANNO_ACK)
703 continue;
704 summary->nr_new_acks++;
705 call->rxtx_annotations[ix] =
706 RXRPC_TX_ANNO_ACK | annotation;
707 break;
708 case RXRPC_ACK_TYPE_NACK:
709 if (!summary->nr_nacks &&
710 call->acks_lowest_nak != seq) {
711 call->acks_lowest_nak = seq;
712 summary->new_low_nack = true;
713 }
714 summary->nr_nacks++;
715 if (anno_type == RXRPC_TX_ANNO_NAK)
716 continue;
717 summary->nr_new_nacks++;
718 if (anno_type == RXRPC_TX_ANNO_RETRANS)
719 continue;
720 call->rxtx_annotations[ix] =
721 RXRPC_TX_ANNO_NAK | annotation;
722 break;
723 default:
724 return rxrpc_proto_abort("SFT", call, 0);
725 }
726 }
727 }
728
729 /*
730 * Process an ACK packet.
731 *
732 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
733 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
734 *
735 * A hard-ACK means that a packet has been processed and may be discarded; a
736 * soft-ACK means that the packet may be discarded and retransmission
737 * requested. A phase is complete when all packets are hard-ACK'd.
738 */
739 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
740 u16 skew)
741 {
742 struct rxrpc_ack_summary summary = { 0 };
743 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
744 union {
745 struct rxrpc_ackpacket ack;
746 struct rxrpc_ackinfo info;
747 u8 acks[RXRPC_MAXACKS];
748 } buf;
749 rxrpc_serial_t acked_serial;
750 rxrpc_seq_t first_soft_ack, hard_ack;
751 int nr_acks, offset, ioffset;
752
753 _enter("");
754
755 offset = sizeof(struct rxrpc_wire_header);
756 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
757 _debug("extraction failure");
758 return rxrpc_proto_abort("XAK", call, 0);
759 }
760 offset += sizeof(buf.ack);
761
762 acked_serial = ntohl(buf.ack.serial);
763 first_soft_ack = ntohl(buf.ack.firstPacket);
764 hard_ack = first_soft_ack - 1;
765 nr_acks = buf.ack.nAcks;
766 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
767 buf.ack.reason : RXRPC_ACK__INVALID);
768
769 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
770 first_soft_ack, ntohl(buf.ack.previousPacket),
771 summary.ack_reason, nr_acks);
772
773 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
774 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
775 sp->hdr.serial);
776 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
777 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
778 sp->hdr.serial);
779
780 if (buf.ack.reason == RXRPC_ACK_PING) {
781 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
782 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
783 skew, sp->hdr.serial, true, true,
784 rxrpc_propose_ack_respond_to_ping);
785 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
786 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
787 skew, sp->hdr.serial, true, true,
788 rxrpc_propose_ack_respond_to_ack);
789 }
790
791 ioffset = offset + nr_acks + 3;
792 if (skb->len >= ioffset + sizeof(buf.info)) {
793 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
794 return rxrpc_proto_abort("XAI", call, 0);
795 rxrpc_input_ackinfo(call, skb, &buf.info);
796 }
797
798 if (first_soft_ack == 0)
799 return rxrpc_proto_abort("AK0", call, 0);
800
801 /* Ignore ACKs unless we are or have just been transmitting. */
802 switch (call->state) {
803 case RXRPC_CALL_CLIENT_SEND_REQUEST:
804 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
805 case RXRPC_CALL_SERVER_SEND_REPLY:
806 case RXRPC_CALL_SERVER_AWAIT_ACK:
807 break;
808 default:
809 return;
810 }
811
812 /* Discard any out-of-order or duplicate ACKs. */
813 if (before_eq(sp->hdr.serial, call->acks_latest)) {
814 _debug("discard ACK %d <= %d",
815 sp->hdr.serial, call->acks_latest);
816 return;
817 }
818 call->acks_latest_ts = skb->tstamp;
819 call->acks_latest = sp->hdr.serial;
820
821 if (before(hard_ack, call->tx_hard_ack) ||
822 after(hard_ack, call->tx_top))
823 return rxrpc_proto_abort("AKW", call, 0);
824 if (nr_acks > call->tx_top - hard_ack)
825 return rxrpc_proto_abort("AKN", call, 0);
826
827 if (after(hard_ack, call->tx_hard_ack))
828 rxrpc_rotate_tx_window(call, hard_ack, &summary);
829
830 if (nr_acks > 0) {
831 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0)
832 return rxrpc_proto_abort("XSA", call, 0);
833 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
834 &summary);
835 }
836
837 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
838 rxrpc_end_tx_phase(call, false, "ETA");
839 return;
840 }
841
842 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
843 RXRPC_TX_ANNO_LAST &&
844 summary.nr_acks == call->tx_top - hard_ack &&
845 rxrpc_is_client_call(call))
846 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
847 false, true,
848 rxrpc_propose_ack_ping_for_lost_reply);
849
850 return rxrpc_congestion_management(call, skb, &summary, acked_serial);
851 }
852
853 /*
854 * Process an ACKALL packet.
855 */
856 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
857 {
858 struct rxrpc_ack_summary summary = { 0 };
859 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
860
861 _proto("Rx ACKALL %%%u", sp->hdr.serial);
862
863 rxrpc_rotate_tx_window(call, call->tx_top, &summary);
864 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags))
865 rxrpc_end_tx_phase(call, false, "ETL");
866 }
867
868 /*
869 * Process an ABORT packet.
870 */
871 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
872 {
873 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
874 __be32 wtmp;
875 u32 abort_code = RX_CALL_DEAD;
876
877 _enter("");
878
879 if (skb->len >= 4 &&
880 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
881 &wtmp, sizeof(wtmp)) >= 0)
882 abort_code = ntohl(wtmp);
883
884 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
885
886 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
887 abort_code, ECONNABORTED))
888 rxrpc_notify_socket(call);
889 }
890
891 /*
892 * Process an incoming call packet.
893 */
894 static void rxrpc_input_call_packet(struct rxrpc_call *call,
895 struct sk_buff *skb, u16 skew)
896 {
897 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
898
899 _enter("%p,%p", call, skb);
900
901 switch (sp->hdr.type) {
902 case RXRPC_PACKET_TYPE_DATA:
903 rxrpc_input_data(call, skb, skew);
904 break;
905
906 case RXRPC_PACKET_TYPE_ACK:
907 rxrpc_input_ack(call, skb, skew);
908 break;
909
910 case RXRPC_PACKET_TYPE_BUSY:
911 _proto("Rx BUSY %%%u", sp->hdr.serial);
912
913 /* Just ignore BUSY packets from the server; the retry and
914 * lifespan timers will take care of business. BUSY packets
915 * from the client don't make sense.
916 */
917 break;
918
919 case RXRPC_PACKET_TYPE_ABORT:
920 rxrpc_input_abort(call, skb);
921 break;
922
923 case RXRPC_PACKET_TYPE_ACKALL:
924 rxrpc_input_ackall(call, skb);
925 break;
926
927 default:
928 break;
929 }
930
931 _leave("");
932 }
933
934 /*
935 * Handle a new call on a channel implicitly completing the preceding call on
936 * that channel.
937 *
938 * TODO: If callNumber > call_id + 1, renegotiate security.
939 */
940 static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn,
941 struct rxrpc_call *call)
942 {
943 switch (call->state) {
944 case RXRPC_CALL_SERVER_AWAIT_ACK:
945 rxrpc_call_completed(call);
946 break;
947 case RXRPC_CALL_COMPLETE:
948 break;
949 default:
950 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, ESHUTDOWN)) {
951 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
952 rxrpc_queue_call(call);
953 }
954 break;
955 }
956
957 trace_rxrpc_improper_term(call);
958 __rxrpc_disconnect_call(conn, call);
959 rxrpc_notify_socket(call);
960 }
961
962 /*
963 * post connection-level events to the connection
964 * - this includes challenges, responses, some aborts and call terminal packet
965 * retransmission.
966 */
967 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
968 struct sk_buff *skb)
969 {
970 _enter("%p,%p", conn, skb);
971
972 skb_queue_tail(&conn->rx_queue, skb);
973 rxrpc_queue_conn(conn);
974 }
975
976 /*
977 * post endpoint-level events to the local endpoint
978 * - this includes debug and version messages
979 */
980 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
981 struct sk_buff *skb)
982 {
983 _enter("%p,%p", local, skb);
984
985 skb_queue_tail(&local->event_queue, skb);
986 rxrpc_queue_local(local);
987 }
988
989 /*
990 * put a packet up for transport-level abort
991 */
992 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
993 {
994 CHECK_SLAB_OKAY(&local->usage);
995
996 skb_queue_tail(&local->reject_queue, skb);
997 rxrpc_queue_local(local);
998 }
999
1000 /*
1001 * Extract the wire header from a packet and translate the byte order.
1002 */
1003 static noinline
1004 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1005 {
1006 struct rxrpc_wire_header whdr;
1007
1008 /* dig out the RxRPC connection details */
1009 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0)
1010 return -EBADMSG;
1011
1012 memset(sp, 0, sizeof(*sp));
1013 sp->hdr.epoch = ntohl(whdr.epoch);
1014 sp->hdr.cid = ntohl(whdr.cid);
1015 sp->hdr.callNumber = ntohl(whdr.callNumber);
1016 sp->hdr.seq = ntohl(whdr.seq);
1017 sp->hdr.serial = ntohl(whdr.serial);
1018 sp->hdr.flags = whdr.flags;
1019 sp->hdr.type = whdr.type;
1020 sp->hdr.userStatus = whdr.userStatus;
1021 sp->hdr.securityIndex = whdr.securityIndex;
1022 sp->hdr._rsvd = ntohs(whdr._rsvd);
1023 sp->hdr.serviceId = ntohs(whdr.serviceId);
1024 return 0;
1025 }
1026
1027 /*
1028 * handle data received on the local endpoint
1029 * - may be called in interrupt context
1030 *
1031 * The socket is locked by the caller and this prevents the socket from being
1032 * shut down and the local endpoint from going away, thus sk_user_data will not
1033 * be cleared until this function returns.
1034 */
1035 void rxrpc_data_ready(struct sock *udp_sk)
1036 {
1037 struct rxrpc_connection *conn;
1038 struct rxrpc_channel *chan;
1039 struct rxrpc_call *call;
1040 struct rxrpc_skb_priv *sp;
1041 struct rxrpc_local *local = udp_sk->sk_user_data;
1042 struct sk_buff *skb;
1043 unsigned int channel;
1044 int ret, skew;
1045
1046 _enter("%p", udp_sk);
1047
1048 ASSERT(!irqs_disabled());
1049
1050 skb = skb_recv_udp(udp_sk, 0, 1, &ret);
1051 if (!skb) {
1052 if (ret == -EAGAIN)
1053 return;
1054 _debug("UDP socket error %d", ret);
1055 return;
1056 }
1057
1058 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
1059
1060 _net("recv skb %p", skb);
1061
1062 /* we'll probably need to checksum it (didn't call sock_recvmsg) */
1063 if (skb_checksum_complete(skb)) {
1064 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1065 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
1066 _leave(" [CSUM failed]");
1067 return;
1068 }
1069
1070 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0);
1071
1072 /* The UDP protocol already released all skb resources;
1073 * we are free to add our own data there.
1074 */
1075 sp = rxrpc_skb(skb);
1076
1077 /* dig out the RxRPC connection details */
1078 if (rxrpc_extract_header(sp, skb) < 0)
1079 goto bad_message;
1080
1081 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1082 static int lose;
1083 if ((lose++ & 7) == 7) {
1084 trace_rxrpc_rx_lose(sp);
1085 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost);
1086 return;
1087 }
1088 }
1089
1090 trace_rxrpc_rx_packet(sp);
1091
1092 _net("Rx RxRPC %s ep=%x call=%x:%x",
1093 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient",
1094 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber);
1095
1096 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES ||
1097 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) {
1098 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1099 goto bad_message;
1100 }
1101
1102 switch (sp->hdr.type) {
1103 case RXRPC_PACKET_TYPE_VERSION:
1104 rxrpc_post_packet_to_local(local, skb);
1105 goto out;
1106
1107 case RXRPC_PACKET_TYPE_BUSY:
1108 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED)
1109 goto discard;
1110
1111 case RXRPC_PACKET_TYPE_DATA:
1112 if (sp->hdr.callNumber == 0)
1113 goto bad_message;
1114 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
1115 !rxrpc_validate_jumbo(skb))
1116 goto bad_message;
1117 break;
1118 }
1119
1120 rcu_read_lock();
1121
1122 conn = rxrpc_find_connection_rcu(local, skb);
1123 if (conn) {
1124 if (sp->hdr.securityIndex != conn->security_ix)
1125 goto wrong_security;
1126
1127 if (sp->hdr.callNumber == 0) {
1128 /* Connection-level packet */
1129 _debug("CONN %p {%d}", conn, conn->debug_id);
1130 rxrpc_post_packet_to_conn(conn, skb);
1131 goto out_unlock;
1132 }
1133
1134 /* Note the serial number skew here */
1135 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
1136 if (skew >= 0) {
1137 if (skew > 0)
1138 conn->hi_serial = sp->hdr.serial;
1139 } else {
1140 skew = -skew;
1141 skew = min(skew, 65535);
1142 }
1143
1144 /* Call-bound packets are routed by connection channel. */
1145 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1146 chan = &conn->channels[channel];
1147
1148 /* Ignore really old calls */
1149 if (sp->hdr.callNumber < chan->last_call)
1150 goto discard_unlock;
1151
1152 if (sp->hdr.callNumber == chan->last_call) {
1153 /* For the previous service call, if completed successfully, we
1154 * discard all further packets.
1155 */
1156 if (rxrpc_conn_is_service(conn) &&
1157 (chan->last_type == RXRPC_PACKET_TYPE_ACK ||
1158 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT))
1159 goto discard_unlock;
1160
1161 /* But otherwise we need to retransmit the final packet from
1162 * data cached in the connection record.
1163 */
1164 rxrpc_post_packet_to_conn(conn, skb);
1165 goto out_unlock;
1166 }
1167
1168 call = rcu_dereference(chan->call);
1169
1170 if (sp->hdr.callNumber > chan->call_id) {
1171 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) {
1172 rcu_read_unlock();
1173 goto reject_packet;
1174 }
1175 if (call)
1176 rxrpc_input_implicit_end_call(conn, call);
1177 call = NULL;
1178 }
1179 } else {
1180 skew = 0;
1181 call = NULL;
1182 }
1183
1184 if (!call || atomic_read(&call->usage) == 0) {
1185 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) ||
1186 sp->hdr.callNumber == 0 ||
1187 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1188 goto bad_message_unlock;
1189 if (sp->hdr.seq != 1)
1190 goto discard_unlock;
1191 call = rxrpc_new_incoming_call(local, conn, skb);
1192 if (!call) {
1193 rcu_read_unlock();
1194 goto reject_packet;
1195 }
1196 rxrpc_send_ping(call, skb, skew);
1197 }
1198
1199 rxrpc_input_call_packet(call, skb, skew);
1200 goto discard_unlock;
1201
1202 discard_unlock:
1203 rcu_read_unlock();
1204 discard:
1205 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1206 out:
1207 trace_rxrpc_rx_done(0, 0);
1208 return;
1209
1210 out_unlock:
1211 rcu_read_unlock();
1212 goto out;
1213
1214 wrong_security:
1215 rcu_read_unlock();
1216 trace_rxrpc_abort("SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1217 RXKADINCONSISTENCY, EBADMSG);
1218 skb->priority = RXKADINCONSISTENCY;
1219 goto post_abort;
1220
1221 bad_message_unlock:
1222 rcu_read_unlock();
1223 bad_message:
1224 trace_rxrpc_abort("BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1225 RX_PROTOCOL_ERROR, EBADMSG);
1226 skb->priority = RX_PROTOCOL_ERROR;
1227 post_abort:
1228 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
1229 reject_packet:
1230 trace_rxrpc_rx_done(skb->mark, skb->priority);
1231 rxrpc_reject_packet(local, skb);
1232 _leave(" [badmsg]");
1233 }