rxrpc: Send an immediate ACK if we fill in a hole
[linux-block.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  * Ping the other end to fill our RTT cache and to retrieve the rwind
41  * and MTU parameters.
42  */
43 static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb,
44                             int skew)
45 {
46         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
47         ktime_t now = skb->tstamp;
48
49         if (call->peer->rtt_usage < 3 ||
50             ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now))
51                 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
52                                   true, true,
53                                   rxrpc_propose_ack_ping_for_params);
54 }
55
56 /*
57  * Apply a hard ACK by advancing the Tx window.
58  */
59 static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to)
60 {
61         struct sk_buff *skb, *list = NULL;
62         int ix;
63         u8 annotation;
64
65         spin_lock(&call->lock);
66
67         while (before(call->tx_hard_ack, to)) {
68                 call->tx_hard_ack++;
69                 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
70                 skb = call->rxtx_buffer[ix];
71                 annotation = call->rxtx_annotations[ix];
72                 rxrpc_see_skb(skb, rxrpc_skb_tx_rotated);
73                 call->rxtx_buffer[ix] = NULL;
74                 call->rxtx_annotations[ix] = 0;
75                 skb->next = list;
76                 list = skb;
77
78                 if (annotation & RXRPC_TX_ANNO_LAST)
79                         set_bit(RXRPC_CALL_TX_LAST, &call->flags);
80         }
81
82         spin_unlock(&call->lock);
83
84         trace_rxrpc_transmit(call, (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ?
85                                     rxrpc_transmit_rotate_last :
86                                     rxrpc_transmit_rotate));
87         wake_up(&call->waitq);
88
89         while (list) {
90                 skb = list;
91                 list = skb->next;
92                 skb->next = NULL;
93                 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
94         }
95 }
96
97 /*
98  * End the transmission phase of a call.
99  *
100  * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
101  * or a final ACK packet.
102  */
103 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
104                                const char *abort_why)
105 {
106
107         ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
108
109         write_lock(&call->state_lock);
110
111         switch (call->state) {
112         case RXRPC_CALL_CLIENT_SEND_REQUEST:
113         case RXRPC_CALL_CLIENT_AWAIT_REPLY:
114                 if (reply_begun)
115                         call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
116                 else
117                         call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
118                 break;
119
120         case RXRPC_CALL_SERVER_AWAIT_ACK:
121                 __rxrpc_call_completed(call);
122                 rxrpc_notify_socket(call);
123                 break;
124
125         default:
126                 goto bad_state;
127         }
128
129         write_unlock(&call->state_lock);
130         if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) {
131                 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
132         } else {
133                 trace_rxrpc_transmit(call, rxrpc_transmit_end);
134         }
135         _leave(" = ok");
136         return true;
137
138 bad_state:
139         write_unlock(&call->state_lock);
140         kdebug("end_tx %s", rxrpc_call_states[call->state]);
141         rxrpc_proto_abort(abort_why, call, call->tx_top);
142         return false;
143 }
144
145 /*
146  * Begin the reply reception phase of a call.
147  */
148 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
149 {
150         rxrpc_seq_t top = READ_ONCE(call->tx_top);
151
152         if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags))
153                 rxrpc_rotate_tx_window(call, top);
154         if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
155                 rxrpc_proto_abort("TXL", call, top);
156                 return false;
157         }
158         if (!rxrpc_end_tx_phase(call, true, "ETD"))
159                 return false;
160         call->tx_phase = false;
161         return true;
162 }
163
164 /*
165  * Scan a jumbo packet to validate its structure and to work out how many
166  * subpackets it contains.
167  *
168  * A jumbo packet is a collection of consecutive packets glued together with
169  * little headers between that indicate how to change the initial header for
170  * each subpacket.
171  *
172  * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
173  * the last are RXRPC_JUMBO_DATALEN in size.  The last subpacket may be of any
174  * size.
175  */
176 static bool rxrpc_validate_jumbo(struct sk_buff *skb)
177 {
178         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
179         unsigned int offset = sp->offset;
180         unsigned int len = skb->len;
181         int nr_jumbo = 1;
182         u8 flags = sp->hdr.flags;
183
184         do {
185                 nr_jumbo++;
186                 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
187                         goto protocol_error;
188                 if (flags & RXRPC_LAST_PACKET)
189                         goto protocol_error;
190                 offset += RXRPC_JUMBO_DATALEN;
191                 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
192                         goto protocol_error;
193                 offset += sizeof(struct rxrpc_jumbo_header);
194         } while (flags & RXRPC_JUMBO_PACKET);
195
196         sp->nr_jumbo = nr_jumbo;
197         return true;
198
199 protocol_error:
200         return false;
201 }
202
203 /*
204  * Handle reception of a duplicate packet.
205  *
206  * We have to take care to avoid an attack here whereby we're given a series of
207  * jumbograms, each with a sequence number one before the preceding one and
208  * filled up to maximum UDP size.  If they never send us the first packet in
209  * the sequence, they can cause us to have to hold on to around 2MiB of kernel
210  * space until the call times out.
211  *
212  * We limit the space usage by only accepting three duplicate jumbo packets per
213  * call.  After that, we tell the other side we're no longer accepting jumbos
214  * (that information is encoded in the ACK packet).
215  */
216 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
217                                  u8 annotation, bool *_jumbo_bad)
218 {
219         /* Discard normal packets that are duplicates. */
220         if (annotation == 0)
221                 return;
222
223         /* Skip jumbo subpackets that are duplicates.  When we've had three or
224          * more partially duplicate jumbo packets, we refuse to take any more
225          * jumbos for this call.
226          */
227         if (!*_jumbo_bad) {
228                 call->nr_jumbo_bad++;
229                 *_jumbo_bad = true;
230         }
231 }
232
233 /*
234  * Process a DATA packet, adding the packet to the Rx ring.
235  */
236 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
237                              u16 skew)
238 {
239         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
240         unsigned int offset = sp->offset;
241         unsigned int ix;
242         rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
243         rxrpc_seq_t seq = sp->hdr.seq, hard_ack;
244         bool immediate_ack = false, jumbo_bad = false, queued;
245         u16 len;
246         u8 ack = 0, flags, annotation = 0;
247
248         _enter("{%u,%u},{%u,%u}",
249                call->rx_hard_ack, call->rx_top, skb->len, seq);
250
251         _proto("Rx DATA %%%u { #%u f=%02x }",
252                sp->hdr.serial, seq, sp->hdr.flags);
253
254         if (call->state >= RXRPC_CALL_COMPLETE)
255                 return;
256
257         /* Received data implicitly ACKs all of the request packets we sent
258          * when we're acting as a client.
259          */
260         if ((call->state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
261              call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
262             !rxrpc_receiving_reply(call))
263                 return;
264
265         call->ackr_prev_seq = seq;
266
267         hard_ack = READ_ONCE(call->rx_hard_ack);
268         if (after(seq, hard_ack + call->rx_winsize)) {
269                 ack = RXRPC_ACK_EXCEEDS_WINDOW;
270                 ack_serial = serial;
271                 goto ack;
272         }
273
274         flags = sp->hdr.flags;
275         if (flags & RXRPC_JUMBO_PACKET) {
276                 if (call->nr_jumbo_bad > 3) {
277                         ack = RXRPC_ACK_NOSPACE;
278                         ack_serial = serial;
279                         goto ack;
280                 }
281                 annotation = 1;
282         }
283
284 next_subpacket:
285         queued = false;
286         ix = seq & RXRPC_RXTX_BUFF_MASK;
287         len = skb->len;
288         if (flags & RXRPC_JUMBO_PACKET)
289                 len = RXRPC_JUMBO_DATALEN;
290
291         if (flags & RXRPC_LAST_PACKET) {
292                 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
293                     seq != call->rx_top)
294                         return rxrpc_proto_abort("LSN", call, seq);
295         } else {
296                 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
297                     after_eq(seq, call->rx_top))
298                         return rxrpc_proto_abort("LSA", call, seq);
299         }
300
301         if (before_eq(seq, hard_ack)) {
302                 ack = RXRPC_ACK_DUPLICATE;
303                 ack_serial = serial;
304                 goto skip;
305         }
306
307         if (flags & RXRPC_REQUEST_ACK && !ack) {
308                 ack = RXRPC_ACK_REQUESTED;
309                 ack_serial = serial;
310         }
311
312         if (call->rxtx_buffer[ix]) {
313                 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
314                 if (ack != RXRPC_ACK_DUPLICATE) {
315                         ack = RXRPC_ACK_DUPLICATE;
316                         ack_serial = serial;
317                 }
318                 immediate_ack = true;
319                 goto skip;
320         }
321
322         /* Queue the packet.  We use a couple of memory barriers here as need
323          * to make sure that rx_top is perceived to be set after the buffer
324          * pointer and that the buffer pointer is set after the annotation and
325          * the skb data.
326          *
327          * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
328          * and also rxrpc_fill_out_ack().
329          */
330         rxrpc_get_skb(skb, rxrpc_skb_rx_got);
331         call->rxtx_annotations[ix] = annotation;
332         smp_wmb();
333         call->rxtx_buffer[ix] = skb;
334         if (after(seq, call->rx_top)) {
335                 smp_store_release(&call->rx_top, seq);
336         } else if (before(seq, call->rx_top)) {
337                 /* Send an immediate ACK if we fill in a hole */
338                 if (!ack) {
339                         ack = RXRPC_ACK_DELAY;
340                         ack_serial = serial;
341                 }
342                 immediate_ack = true;
343         }
344         if (flags & RXRPC_LAST_PACKET) {
345                 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
346                 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
347         } else {
348                 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
349         }
350         queued = true;
351
352         if (after_eq(seq, call->rx_expect_next)) {
353                 if (after(seq, call->rx_expect_next)) {
354                         _net("OOS %u > %u", seq, call->rx_expect_next);
355                         ack = RXRPC_ACK_OUT_OF_SEQUENCE;
356                         ack_serial = serial;
357                 }
358                 call->rx_expect_next = seq + 1;
359         }
360
361 skip:
362         offset += len;
363         if (flags & RXRPC_JUMBO_PACKET) {
364                 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
365                         return rxrpc_proto_abort("XJF", call, seq);
366                 offset += sizeof(struct rxrpc_jumbo_header);
367                 seq++;
368                 serial++;
369                 annotation++;
370                 if (flags & RXRPC_JUMBO_PACKET)
371                         annotation |= RXRPC_RX_ANNO_JLAST;
372                 if (after(seq, hard_ack + call->rx_winsize)) {
373                         ack = RXRPC_ACK_EXCEEDS_WINDOW;
374                         ack_serial = serial;
375                         if (!jumbo_bad) {
376                                 call->nr_jumbo_bad++;
377                                 jumbo_bad = true;
378                         }
379                         goto ack;
380                 }
381
382                 _proto("Rx DATA Jumbo %%%u", serial);
383                 goto next_subpacket;
384         }
385
386         if (queued && flags & RXRPC_LAST_PACKET && !ack) {
387                 ack = RXRPC_ACK_DELAY;
388                 ack_serial = serial;
389         }
390
391 ack:
392         if (ack)
393                 rxrpc_propose_ACK(call, ack, skew, ack_serial,
394                                   immediate_ack, true,
395                                   rxrpc_propose_ack_input_data);
396
397         if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1)
398                 rxrpc_notify_socket(call);
399         _leave(" [queued]");
400 }
401
402 /*
403  * Process a requested ACK.
404  */
405 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
406                                       ktime_t resp_time,
407                                       rxrpc_serial_t orig_serial,
408                                       rxrpc_serial_t ack_serial)
409 {
410         struct rxrpc_skb_priv *sp;
411         struct sk_buff *skb;
412         ktime_t sent_at;
413         int ix;
414
415         for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
416                 skb = call->rxtx_buffer[ix];
417                 if (!skb)
418                         continue;
419
420                 sp = rxrpc_skb(skb);
421                 if (sp->hdr.serial != orig_serial)
422                         continue;
423                 smp_rmb();
424                 sent_at = skb->tstamp;
425                 goto found;
426         }
427         return;
428
429 found:
430         rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
431                            orig_serial, ack_serial, sent_at, resp_time);
432 }
433
434 /*
435  * Process a ping response.
436  */
437 static void rxrpc_input_ping_response(struct rxrpc_call *call,
438                                       ktime_t resp_time,
439                                       rxrpc_serial_t orig_serial,
440                                       rxrpc_serial_t ack_serial)
441 {
442         rxrpc_serial_t ping_serial;
443         ktime_t ping_time;
444
445         ping_time = call->ackr_ping_time;
446         smp_rmb();
447         ping_serial = call->ackr_ping;
448
449         if (!test_bit(RXRPC_CALL_PINGING, &call->flags) ||
450             before(orig_serial, ping_serial))
451                 return;
452         clear_bit(RXRPC_CALL_PINGING, &call->flags);
453         if (after(orig_serial, ping_serial))
454                 return;
455
456         rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
457                            orig_serial, ack_serial, ping_time, resp_time);
458 }
459
460 /*
461  * Process the extra information that may be appended to an ACK packet
462  */
463 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
464                                 struct rxrpc_ackinfo *ackinfo)
465 {
466         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
467         struct rxrpc_peer *peer;
468         unsigned int mtu;
469         u32 rwind = ntohl(ackinfo->rwind);
470
471         _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
472                sp->hdr.serial,
473                ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
474                rwind, ntohl(ackinfo->jumbo_max));
475
476         if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
477                 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
478         call->tx_winsize = rwind;
479
480         mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
481
482         peer = call->peer;
483         if (mtu < peer->maxdata) {
484                 spin_lock_bh(&peer->lock);
485                 peer->maxdata = mtu;
486                 peer->mtu = mtu + peer->hdrsize;
487                 spin_unlock_bh(&peer->lock);
488                 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
489         }
490 }
491
492 /*
493  * Process individual soft ACKs.
494  *
495  * Each ACK in the array corresponds to one packet and can be either an ACK or
496  * a NAK.  If we get find an explicitly NAK'd packet we resend immediately;
497  * packets that lie beyond the end of the ACK list are scheduled for resend by
498  * the timer on the basis that the peer might just not have processed them at
499  * the time the ACK was sent.
500  */
501 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
502                                   rxrpc_seq_t seq, int nr_acks)
503 {
504         bool resend = false;
505         int ix;
506         u8 annotation, anno_type;
507
508         for (; nr_acks > 0; nr_acks--, seq++) {
509                 ix = seq & RXRPC_RXTX_BUFF_MASK;
510                 annotation = call->rxtx_annotations[ix];
511                 anno_type = annotation & RXRPC_TX_ANNO_MASK;
512                 annotation &= ~RXRPC_TX_ANNO_MASK;
513                 switch (*acks++) {
514                 case RXRPC_ACK_TYPE_ACK:
515                         if (anno_type == RXRPC_TX_ANNO_ACK)
516                                 continue;
517                         call->rxtx_annotations[ix] =
518                                 RXRPC_TX_ANNO_ACK | annotation;
519                         break;
520                 case RXRPC_ACK_TYPE_NACK:
521                         if (anno_type == RXRPC_TX_ANNO_NAK)
522                                 continue;
523                         if (anno_type == RXRPC_TX_ANNO_RETRANS)
524                                 continue;
525                         call->rxtx_annotations[ix] =
526                                 RXRPC_TX_ANNO_NAK | annotation;
527                         resend = true;
528                         break;
529                 default:
530                         return rxrpc_proto_abort("SFT", call, 0);
531                 }
532         }
533
534         if (resend &&
535             !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
536                 rxrpc_queue_call(call);
537 }
538
539 /*
540  * Process an ACK packet.
541  *
542  * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
543  * in the ACK array.  Anything before that is hard-ACK'd and may be discarded.
544  *
545  * A hard-ACK means that a packet has been processed and may be discarded; a
546  * soft-ACK means that the packet may be discarded and retransmission
547  * requested.  A phase is complete when all packets are hard-ACK'd.
548  */
549 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
550                             u16 skew)
551 {
552         u8 ack_reason;
553         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
554         union {
555                 struct rxrpc_ackpacket ack;
556                 struct rxrpc_ackinfo info;
557                 u8 acks[RXRPC_MAXACKS];
558         } buf;
559         rxrpc_serial_t acked_serial;
560         rxrpc_seq_t first_soft_ack, hard_ack;
561         int nr_acks, offset;
562
563         _enter("");
564
565         if (skb_copy_bits(skb, sp->offset, &buf.ack, sizeof(buf.ack)) < 0) {
566                 _debug("extraction failure");
567                 return rxrpc_proto_abort("XAK", call, 0);
568         }
569         sp->offset += sizeof(buf.ack);
570
571         acked_serial = ntohl(buf.ack.serial);
572         first_soft_ack = ntohl(buf.ack.firstPacket);
573         hard_ack = first_soft_ack - 1;
574         nr_acks = buf.ack.nAcks;
575         ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
576                       buf.ack.reason : RXRPC_ACK__INVALID);
577
578         trace_rxrpc_rx_ack(call, first_soft_ack, ack_reason, nr_acks);
579
580         _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
581                sp->hdr.serial,
582                ntohs(buf.ack.maxSkew),
583                first_soft_ack,
584                ntohl(buf.ack.previousPacket),
585                acked_serial,
586                rxrpc_ack_names[ack_reason],
587                buf.ack.nAcks);
588
589         if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
590                 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
591                                           sp->hdr.serial);
592         if (buf.ack.reason == RXRPC_ACK_REQUESTED)
593                 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
594                                           sp->hdr.serial);
595
596         if (buf.ack.reason == RXRPC_ACK_PING) {
597                 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
598                 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
599                                   skew, sp->hdr.serial, true, true,
600                                   rxrpc_propose_ack_respond_to_ping);
601         } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
602                 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
603                                   skew, sp->hdr.serial, true, true,
604                                   rxrpc_propose_ack_respond_to_ack);
605         }
606
607         offset = sp->offset + nr_acks + 3;
608         if (skb->len >= offset + sizeof(buf.info)) {
609                 if (skb_copy_bits(skb, offset, &buf.info, sizeof(buf.info)) < 0)
610                         return rxrpc_proto_abort("XAI", call, 0);
611                 rxrpc_input_ackinfo(call, skb, &buf.info);
612         }
613
614         if (first_soft_ack == 0)
615                 return rxrpc_proto_abort("AK0", call, 0);
616
617         /* Ignore ACKs unless we are or have just been transmitting. */
618         switch (call->state) {
619         case RXRPC_CALL_CLIENT_SEND_REQUEST:
620         case RXRPC_CALL_CLIENT_AWAIT_REPLY:
621         case RXRPC_CALL_SERVER_SEND_REPLY:
622         case RXRPC_CALL_SERVER_AWAIT_ACK:
623                 break;
624         default:
625                 return;
626         }
627
628         /* Discard any out-of-order or duplicate ACKs. */
629         if (before_eq(sp->hdr.serial, call->acks_latest)) {
630                 _debug("discard ACK %d <= %d",
631                        sp->hdr.serial, call->acks_latest);
632                 return;
633         }
634         call->acks_latest = sp->hdr.serial;
635
636         if (before(hard_ack, call->tx_hard_ack) ||
637             after(hard_ack, call->tx_top))
638                 return rxrpc_proto_abort("AKW", call, 0);
639         if (nr_acks > call->tx_top - hard_ack)
640                 return rxrpc_proto_abort("AKN", call, 0);
641
642         if (after(hard_ack, call->tx_hard_ack))
643                 rxrpc_rotate_tx_window(call, hard_ack);
644
645         if (nr_acks > 0) {
646                 if (skb_copy_bits(skb, sp->offset, buf.acks, nr_acks) < 0)
647                         return rxrpc_proto_abort("XSA", call, 0);
648                 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks);
649         }
650
651         if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
652                 rxrpc_end_tx_phase(call, false, "ETA");
653                 return;
654         }
655
656 }
657
658 /*
659  * Process an ACKALL packet.
660  */
661 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
662 {
663         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
664
665         _proto("Rx ACKALL %%%u", sp->hdr.serial);
666
667         rxrpc_rotate_tx_window(call, call->tx_top);
668         if (test_bit(RXRPC_CALL_TX_LAST, &call->flags))
669                 rxrpc_end_tx_phase(call, false, "ETL");
670 }
671
672 /*
673  * Process an ABORT packet.
674  */
675 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
676 {
677         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
678         __be32 wtmp;
679         u32 abort_code = RX_CALL_DEAD;
680
681         _enter("");
682
683         if (skb->len >= 4 &&
684             skb_copy_bits(skb, sp->offset, &wtmp, sizeof(wtmp)) >= 0)
685                 abort_code = ntohl(wtmp);
686
687         _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
688
689         if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
690                                       abort_code, ECONNABORTED))
691                 rxrpc_notify_socket(call);
692 }
693
694 /*
695  * Process an incoming call packet.
696  */
697 static void rxrpc_input_call_packet(struct rxrpc_call *call,
698                                     struct sk_buff *skb, u16 skew)
699 {
700         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
701
702         _enter("%p,%p", call, skb);
703
704         switch (sp->hdr.type) {
705         case RXRPC_PACKET_TYPE_DATA:
706                 rxrpc_input_data(call, skb, skew);
707                 break;
708
709         case RXRPC_PACKET_TYPE_ACK:
710                 rxrpc_input_ack(call, skb, skew);
711                 break;
712
713         case RXRPC_PACKET_TYPE_BUSY:
714                 _proto("Rx BUSY %%%u", sp->hdr.serial);
715
716                 /* Just ignore BUSY packets from the server; the retry and
717                  * lifespan timers will take care of business.  BUSY packets
718                  * from the client don't make sense.
719                  */
720                 break;
721
722         case RXRPC_PACKET_TYPE_ABORT:
723                 rxrpc_input_abort(call, skb);
724                 break;
725
726         case RXRPC_PACKET_TYPE_ACKALL:
727                 rxrpc_input_ackall(call, skb);
728                 break;
729
730         default:
731                 _proto("Rx %s %%%u", rxrpc_pkts[sp->hdr.type], sp->hdr.serial);
732                 break;
733         }
734
735         _leave("");
736 }
737
738 /*
739  * post connection-level events to the connection
740  * - this includes challenges, responses, some aborts and call terminal packet
741  *   retransmission.
742  */
743 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
744                                       struct sk_buff *skb)
745 {
746         _enter("%p,%p", conn, skb);
747
748         skb_queue_tail(&conn->rx_queue, skb);
749         rxrpc_queue_conn(conn);
750 }
751
752 /*
753  * post endpoint-level events to the local endpoint
754  * - this includes debug and version messages
755  */
756 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
757                                        struct sk_buff *skb)
758 {
759         _enter("%p,%p", local, skb);
760
761         skb_queue_tail(&local->event_queue, skb);
762         rxrpc_queue_local(local);
763 }
764
765 /*
766  * put a packet up for transport-level abort
767  */
768 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
769 {
770         CHECK_SLAB_OKAY(&local->usage);
771
772         skb_queue_tail(&local->reject_queue, skb);
773         rxrpc_queue_local(local);
774 }
775
776 /*
777  * Extract the wire header from a packet and translate the byte order.
778  */
779 static noinline
780 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
781 {
782         struct rxrpc_wire_header whdr;
783
784         /* dig out the RxRPC connection details */
785         if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0)
786                 return -EBADMSG;
787
788         memset(sp, 0, sizeof(*sp));
789         sp->hdr.epoch           = ntohl(whdr.epoch);
790         sp->hdr.cid             = ntohl(whdr.cid);
791         sp->hdr.callNumber      = ntohl(whdr.callNumber);
792         sp->hdr.seq             = ntohl(whdr.seq);
793         sp->hdr.serial          = ntohl(whdr.serial);
794         sp->hdr.flags           = whdr.flags;
795         sp->hdr.type            = whdr.type;
796         sp->hdr.userStatus      = whdr.userStatus;
797         sp->hdr.securityIndex   = whdr.securityIndex;
798         sp->hdr._rsvd           = ntohs(whdr._rsvd);
799         sp->hdr.serviceId       = ntohs(whdr.serviceId);
800         sp->offset = sizeof(whdr);
801         return 0;
802 }
803
804 /*
805  * handle data received on the local endpoint
806  * - may be called in interrupt context
807  *
808  * The socket is locked by the caller and this prevents the socket from being
809  * shut down and the local endpoint from going away, thus sk_user_data will not
810  * be cleared until this function returns.
811  */
812 void rxrpc_data_ready(struct sock *udp_sk)
813 {
814         struct rxrpc_connection *conn;
815         struct rxrpc_channel *chan;
816         struct rxrpc_call *call;
817         struct rxrpc_skb_priv *sp;
818         struct rxrpc_local *local = udp_sk->sk_user_data;
819         struct sk_buff *skb;
820         unsigned int channel;
821         int ret, skew;
822
823         _enter("%p", udp_sk);
824
825         ASSERT(!irqs_disabled());
826
827         skb = skb_recv_datagram(udp_sk, 0, 1, &ret);
828         if (!skb) {
829                 if (ret == -EAGAIN)
830                         return;
831                 _debug("UDP socket error %d", ret);
832                 return;
833         }
834
835         rxrpc_new_skb(skb, rxrpc_skb_rx_received);
836
837         _net("recv skb %p", skb);
838
839         /* we'll probably need to checksum it (didn't call sock_recvmsg) */
840         if (skb_checksum_complete(skb)) {
841                 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
842                 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
843                 _leave(" [CSUM failed]");
844                 return;
845         }
846
847         __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0);
848
849         /* The socket buffer we have is owned by UDP, with UDP's data all over
850          * it, but we really want our own data there.
851          */
852         skb_orphan(skb);
853         sp = rxrpc_skb(skb);
854
855         /* dig out the RxRPC connection details */
856         if (rxrpc_extract_header(sp, skb) < 0)
857                 goto bad_message;
858
859         if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
860                 static int lose;
861                 if ((lose++ & 7) == 7) {
862                         trace_rxrpc_rx_lose(sp);
863                         rxrpc_lose_skb(skb, rxrpc_skb_rx_lost);
864                         return;
865                 }
866         }
867
868         trace_rxrpc_rx_packet(sp);
869
870         _net("Rx RxRPC %s ep=%x call=%x:%x",
871              sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient",
872              sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber);
873
874         if (sp->hdr.type >= RXRPC_N_PACKET_TYPES ||
875             !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) {
876                 _proto("Rx Bad Packet Type %u", sp->hdr.type);
877                 goto bad_message;
878         }
879
880         switch (sp->hdr.type) {
881         case RXRPC_PACKET_TYPE_VERSION:
882                 rxrpc_post_packet_to_local(local, skb);
883                 goto out;
884
885         case RXRPC_PACKET_TYPE_BUSY:
886                 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED)
887                         goto discard;
888
889         case RXRPC_PACKET_TYPE_DATA:
890                 if (sp->hdr.callNumber == 0)
891                         goto bad_message;
892                 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
893                     !rxrpc_validate_jumbo(skb))
894                         goto bad_message;
895                 break;
896         }
897
898         rcu_read_lock();
899
900         conn = rxrpc_find_connection_rcu(local, skb);
901         if (conn) {
902                 if (sp->hdr.securityIndex != conn->security_ix)
903                         goto wrong_security;
904
905                 if (sp->hdr.callNumber == 0) {
906                         /* Connection-level packet */
907                         _debug("CONN %p {%d}", conn, conn->debug_id);
908                         rxrpc_post_packet_to_conn(conn, skb);
909                         goto out_unlock;
910                 }
911
912                 /* Note the serial number skew here */
913                 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
914                 if (skew >= 0) {
915                         if (skew > 0)
916                                 conn->hi_serial = sp->hdr.serial;
917                 } else {
918                         skew = -skew;
919                         skew = min(skew, 65535);
920                 }
921
922                 /* Call-bound packets are routed by connection channel. */
923                 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
924                 chan = &conn->channels[channel];
925
926                 /* Ignore really old calls */
927                 if (sp->hdr.callNumber < chan->last_call)
928                         goto discard_unlock;
929
930                 if (sp->hdr.callNumber == chan->last_call) {
931                         /* For the previous service call, if completed successfully, we
932                          * discard all further packets.
933                          */
934                         if (rxrpc_conn_is_service(conn) &&
935                             (chan->last_type == RXRPC_PACKET_TYPE_ACK ||
936                              sp->hdr.type == RXRPC_PACKET_TYPE_ABORT))
937                                 goto discard_unlock;
938
939                         /* But otherwise we need to retransmit the final packet from
940                          * data cached in the connection record.
941                          */
942                         rxrpc_post_packet_to_conn(conn, skb);
943                         goto out_unlock;
944                 }
945
946                 call = rcu_dereference(chan->call);
947         } else {
948                 skew = 0;
949                 call = NULL;
950         }
951
952         if (!call || atomic_read(&call->usage) == 0) {
953                 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) ||
954                     sp->hdr.callNumber == 0 ||
955                     sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
956                         goto bad_message_unlock;
957                 if (sp->hdr.seq != 1)
958                         goto discard_unlock;
959                 call = rxrpc_new_incoming_call(local, conn, skb);
960                 if (!call) {
961                         rcu_read_unlock();
962                         goto reject_packet;
963                 }
964                 rxrpc_send_ping(call, skb, skew);
965         }
966
967         rxrpc_input_call_packet(call, skb, skew);
968         goto discard_unlock;
969
970 discard_unlock:
971         rcu_read_unlock();
972 discard:
973         rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
974 out:
975         trace_rxrpc_rx_done(0, 0);
976         return;
977
978 out_unlock:
979         rcu_read_unlock();
980         goto out;
981
982 wrong_security:
983         rcu_read_unlock();
984         trace_rxrpc_abort("SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
985                           RXKADINCONSISTENCY, EBADMSG);
986         skb->priority = RXKADINCONSISTENCY;
987         goto post_abort;
988
989 bad_message_unlock:
990         rcu_read_unlock();
991 bad_message:
992         trace_rxrpc_abort("BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
993                           RX_PROTOCOL_ERROR, EBADMSG);
994         skb->priority = RX_PROTOCOL_ERROR;
995 post_abort:
996         skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
997 reject_packet:
998         trace_rxrpc_rx_done(skb->mark, skb->priority);
999         rxrpc_reject_packet(local, skb);
1000         _leave(" [badmsg]");
1001 }