1 /* RxRPC packet reception
3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
20 #include <linux/in6.h>
21 #include <linux/icmp.h>
22 #include <linux/gfp.h>
24 #include <net/af_rxrpc.h>
27 #include <net/net_namespace.h>
28 #include "ar-internal.h"
30 static void rxrpc_proto_abort(const char *why,
31 struct rxrpc_call *call, rxrpc_seq_t seq)
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);
40 * Do TCP-style congestion management [RFC 5681].
42 static void rxrpc_congestion_management(struct rxrpc_call *call,
44 struct rxrpc_ack_summary *summary,
45 rxrpc_serial_t acked_serial)
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;
52 summary->flight_size =
53 (call->tx_top - call->tx_hard_ack) - summary->nr_acks;
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);
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;
68 cumulative_acks += summary->nr_new_acks;
69 cumulative_acks += summary->nr_rot_new_acks;
70 if (cumulative_acks > 255)
71 cumulative_acks = 255;
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;
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)
85 if (cwnd >= call->cong_ssthresh) {
86 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
87 call->cong_tstamp = skb->tstamp;
91 case RXRPC_CALL_CONGEST_AVOIDANCE:
92 if (summary->nr_nacks > 0)
93 goto packet_loss_detected;
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.
98 if (call->peer->rtt_usage == 0)
100 if (ktime_before(skb->tstamp,
101 ktime_add_ns(call->cong_tstamp,
103 goto out_no_clear_ca;
104 change = rxrpc_cong_rtt_window_end;
105 call->cong_tstamp = skb->tstamp;
106 if (cumulative_acks >= cwnd)
110 case RXRPC_CALL_PACKET_LOSS:
111 if (summary->nr_nacks == 0)
112 goto resume_normality;
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;
122 call->cong_dup_acks++;
123 if (call->cong_dup_acks < 3)
124 goto send_extra_data;
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;
136 case RXRPC_CALL_FAST_RETRANSMIT:
137 if (!summary->new_low_nack) {
138 if (summary->nr_new_acks == 0)
140 call->cong_dup_acks++;
141 if (call->cong_dup_acks == 2) {
142 change = rxrpc_cong_retransmit_again;
143 call->cong_dup_acks = 0;
147 change = rxrpc_cong_progress;
148 cwnd = call->cong_ssthresh;
149 if (summary->nr_nacks == 0)
150 goto 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;
167 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
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);
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;
187 /* Send some previously unsent DATA if we have some to advance the ACK
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) {
194 wake_up(&call->waitq);
196 goto out_no_clear_ca;
200 * Ping the other end to fill our RTT cache and to retrieve the rwind
201 * and MTU parameters.
203 static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb,
206 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
207 ktime_t now = skb->tstamp;
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,
213 rxrpc_propose_ack_ping_for_params);
217 * Apply a hard ACK by advancing the Tx window.
219 static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
220 struct rxrpc_ack_summary *summary)
222 struct sk_buff *skb, *list = NULL;
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;
233 spin_lock(&call->lock);
235 while (before(call->tx_hard_ack, to)) {
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;
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++;
252 spin_unlock(&call->lock);
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);
263 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
268 * End the transmission phase of a call.
270 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
271 * or a final ACK packet.
273 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
274 const char *abort_why)
277 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
279 write_lock(&call->state_lock);
281 switch (call->state) {
282 case RXRPC_CALL_CLIENT_SEND_REQUEST:
283 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
285 call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
287 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
290 case RXRPC_CALL_SERVER_AWAIT_ACK:
291 __rxrpc_call_completed(call);
292 rxrpc_notify_socket(call);
299 write_unlock(&call->state_lock);
300 if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY) {
301 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
303 trace_rxrpc_transmit(call, rxrpc_transmit_end);
309 write_unlock(&call->state_lock);
310 kdebug("end_tx %s", rxrpc_call_states[call->state]);
311 rxrpc_proto_abort(abort_why, call, call->tx_top);
316 * Begin the reply reception phase of a call.
318 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
320 struct rxrpc_ack_summary summary = { 0 };
321 unsigned long now, timo;
322 rxrpc_seq_t top = READ_ONCE(call->tx_top);
324 if (call->ackr_reason) {
325 spin_lock_bh(&call->lock);
326 call->ackr_reason = 0;
327 spin_unlock_bh(&call->lock);
329 timo = now + MAX_JIFFY_OFFSET;
330 WRITE_ONCE(call->resend_at, timo);
331 WRITE_ONCE(call->ack_at, timo);
332 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
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);
341 if (!rxrpc_end_tx_phase(call, true, "ETD"))
343 call->tx_phase = false;
348 * Scan a jumbo packet to validate its structure and to work out how many
349 * subpackets it contains.
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
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
359 static bool rxrpc_validate_jumbo(struct sk_buff *skb)
361 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
362 unsigned int offset = sizeof(struct rxrpc_wire_header);
363 unsigned int len = skb->len;
365 u8 flags = sp->hdr.flags;
369 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
371 if (flags & RXRPC_LAST_PACKET)
373 offset += RXRPC_JUMBO_DATALEN;
374 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
376 offset += sizeof(struct rxrpc_jumbo_header);
377 } while (flags & RXRPC_JUMBO_PACKET);
379 sp->nr_jumbo = nr_jumbo;
387 * Handle reception of a duplicate packet.
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.
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).
399 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
400 u8 annotation, bool *_jumbo_bad)
402 /* Discard normal packets that are duplicates. */
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.
411 call->nr_jumbo_bad++;
417 * Process a DATA packet, adding the packet to the Rx ring.
419 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
422 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
423 enum rxrpc_call_state state;
424 unsigned int offset = sizeof(struct rxrpc_wire_header);
426 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
427 rxrpc_seq_t seq = sp->hdr.seq, hard_ack;
428 bool immediate_ack = false, jumbo_bad = false, queued;
430 u8 ack = 0, flags, annotation = 0;
432 _enter("{%u,%u},{%u,%u}",
433 call->rx_hard_ack, call->rx_top, skb->len, seq);
435 _proto("Rx DATA %%%u { #%u f=%02x }",
436 sp->hdr.serial, seq, sp->hdr.flags);
438 state = READ_ONCE(call->state);
439 if (state >= RXRPC_CALL_COMPLETE)
442 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
443 unsigned long timo = READ_ONCE(call->next_req_timo);
444 unsigned long now, expect_req_by;
448 expect_req_by = now + timo;
449 WRITE_ONCE(call->expect_req_by, expect_req_by);
450 rxrpc_reduce_call_timer(call, expect_req_by, now,
451 rxrpc_timer_set_for_idle);
455 /* Received data implicitly ACKs all of the request packets we sent
456 * when we're acting as a client.
458 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
459 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
460 !rxrpc_receiving_reply(call))
463 call->ackr_prev_seq = seq;
465 hard_ack = READ_ONCE(call->rx_hard_ack);
466 if (after(seq, hard_ack + call->rx_winsize)) {
467 ack = RXRPC_ACK_EXCEEDS_WINDOW;
472 flags = sp->hdr.flags;
473 if (flags & RXRPC_JUMBO_PACKET) {
474 if (call->nr_jumbo_bad > 3) {
475 ack = RXRPC_ACK_NOSPACE;
484 ix = seq & RXRPC_RXTX_BUFF_MASK;
486 if (flags & RXRPC_JUMBO_PACKET)
487 len = RXRPC_JUMBO_DATALEN;
489 if (flags & RXRPC_LAST_PACKET) {
490 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
492 return rxrpc_proto_abort("LSN", call, seq);
494 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
495 after_eq(seq, call->rx_top))
496 return rxrpc_proto_abort("LSA", call, seq);
499 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
500 if (before_eq(seq, hard_ack)) {
501 ack = RXRPC_ACK_DUPLICATE;
506 if (flags & RXRPC_REQUEST_ACK && !ack) {
507 ack = RXRPC_ACK_REQUESTED;
511 if (call->rxtx_buffer[ix]) {
512 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
513 if (ack != RXRPC_ACK_DUPLICATE) {
514 ack = RXRPC_ACK_DUPLICATE;
517 immediate_ack = true;
521 /* Queue the packet. We use a couple of memory barriers here as need
522 * to make sure that rx_top is perceived to be set after the buffer
523 * pointer and that the buffer pointer is set after the annotation and
526 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
527 * and also rxrpc_fill_out_ack().
529 rxrpc_get_skb(skb, rxrpc_skb_rx_got);
530 call->rxtx_annotations[ix] = annotation;
532 call->rxtx_buffer[ix] = skb;
533 if (after(seq, call->rx_top)) {
534 smp_store_release(&call->rx_top, seq);
535 } else if (before(seq, call->rx_top)) {
536 /* Send an immediate ACK if we fill in a hole */
538 ack = RXRPC_ACK_DELAY;
541 immediate_ack = true;
543 if (flags & RXRPC_LAST_PACKET) {
544 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
545 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
547 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
551 if (after_eq(seq, call->rx_expect_next)) {
552 if (after(seq, call->rx_expect_next)) {
553 _net("OOS %u > %u", seq, call->rx_expect_next);
554 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
557 call->rx_expect_next = seq + 1;
562 if (flags & RXRPC_JUMBO_PACKET) {
563 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
564 return rxrpc_proto_abort("XJF", call, seq);
565 offset += sizeof(struct rxrpc_jumbo_header);
569 if (flags & RXRPC_JUMBO_PACKET)
570 annotation |= RXRPC_RX_ANNO_JLAST;
571 if (after(seq, hard_ack + call->rx_winsize)) {
572 ack = RXRPC_ACK_EXCEEDS_WINDOW;
575 call->nr_jumbo_bad++;
581 _proto("Rx DATA Jumbo %%%u", serial);
585 if (queued && flags & RXRPC_LAST_PACKET && !ack) {
586 ack = RXRPC_ACK_DELAY;
592 rxrpc_propose_ACK(call, ack, skew, ack_serial,
594 rxrpc_propose_ack_input_data);
596 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, skew, serial,
598 rxrpc_propose_ack_input_data);
600 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1) {
601 trace_rxrpc_notify_socket(call->debug_id, serial);
602 rxrpc_notify_socket(call);
608 * Process a requested ACK.
610 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
612 rxrpc_serial_t orig_serial,
613 rxrpc_serial_t ack_serial)
615 struct rxrpc_skb_priv *sp;
620 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
621 skb = call->rxtx_buffer[ix];
626 if (sp->hdr.serial != orig_serial)
629 sent_at = skb->tstamp;
635 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
636 orig_serial, ack_serial, sent_at, resp_time);
640 * Process the response to a ping that we sent to find out if we lost an ACK.
642 * If we got back a ping response that indicates a lower tx_top than what we
643 * had at the time of the ping transmission, we adjudge all the DATA packets
644 * sent between the response tx_top and the ping-time tx_top to have been lost.
646 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
648 rxrpc_seq_t top, bottom, seq;
651 spin_lock_bh(&call->lock);
653 bottom = call->tx_hard_ack + 1;
654 top = call->acks_lost_top;
655 if (before(bottom, top)) {
656 for (seq = bottom; before_eq(seq, top); seq++) {
657 int ix = seq & RXRPC_RXTX_BUFF_MASK;
658 u8 annotation = call->rxtx_annotations[ix];
659 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
661 if (anno_type != RXRPC_TX_ANNO_UNACK)
663 annotation &= ~RXRPC_TX_ANNO_MASK;
664 annotation |= RXRPC_TX_ANNO_RETRANS;
665 call->rxtx_annotations[ix] = annotation;
670 spin_unlock_bh(&call->lock);
672 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
673 rxrpc_queue_call(call);
677 * Process a ping response.
679 static void rxrpc_input_ping_response(struct rxrpc_call *call,
681 rxrpc_serial_t orig_serial,
682 rxrpc_serial_t ack_serial)
684 rxrpc_serial_t ping_serial;
687 ping_time = call->ping_time;
689 ping_serial = call->ping_serial;
691 if (orig_serial == call->acks_lost_ping)
692 rxrpc_input_check_for_lost_ack(call);
694 if (!test_bit(RXRPC_CALL_PINGING, &call->flags) ||
695 before(orig_serial, ping_serial))
697 clear_bit(RXRPC_CALL_PINGING, &call->flags);
698 if (after(orig_serial, ping_serial))
701 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
702 orig_serial, ack_serial, ping_time, resp_time);
706 * Process the extra information that may be appended to an ACK packet
708 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
709 struct rxrpc_ackinfo *ackinfo)
711 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
712 struct rxrpc_peer *peer;
715 u32 rwind = ntohl(ackinfo->rwind);
717 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
719 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
720 rwind, ntohl(ackinfo->jumbo_max));
722 if (call->tx_winsize != rwind) {
723 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
724 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
725 if (rwind > call->tx_winsize)
727 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
728 ntohl(ackinfo->rwind), wake);
729 call->tx_winsize = rwind;
732 if (call->cong_ssthresh > rwind)
733 call->cong_ssthresh = rwind;
735 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
738 if (mtu < peer->maxdata) {
739 spin_lock_bh(&peer->lock);
741 peer->mtu = mtu + peer->hdrsize;
742 spin_unlock_bh(&peer->lock);
743 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
747 wake_up(&call->waitq);
751 * Process individual soft ACKs.
753 * Each ACK in the array corresponds to one packet and can be either an ACK or
754 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
755 * packets that lie beyond the end of the ACK list are scheduled for resend by
756 * the timer on the basis that the peer might just not have processed them at
757 * the time the ACK was sent.
759 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
760 rxrpc_seq_t seq, int nr_acks,
761 struct rxrpc_ack_summary *summary)
764 u8 annotation, anno_type;
766 for (; nr_acks > 0; nr_acks--, seq++) {
767 ix = seq & RXRPC_RXTX_BUFF_MASK;
768 annotation = call->rxtx_annotations[ix];
769 anno_type = annotation & RXRPC_TX_ANNO_MASK;
770 annotation &= ~RXRPC_TX_ANNO_MASK;
772 case RXRPC_ACK_TYPE_ACK:
774 if (anno_type == RXRPC_TX_ANNO_ACK)
776 summary->nr_new_acks++;
777 call->rxtx_annotations[ix] =
778 RXRPC_TX_ANNO_ACK | annotation;
780 case RXRPC_ACK_TYPE_NACK:
781 if (!summary->nr_nacks &&
782 call->acks_lowest_nak != seq) {
783 call->acks_lowest_nak = seq;
784 summary->new_low_nack = true;
787 if (anno_type == RXRPC_TX_ANNO_NAK)
789 summary->nr_new_nacks++;
790 if (anno_type == RXRPC_TX_ANNO_RETRANS)
792 call->rxtx_annotations[ix] =
793 RXRPC_TX_ANNO_NAK | annotation;
796 return rxrpc_proto_abort("SFT", call, 0);
802 * Process an ACK packet.
804 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
805 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
807 * A hard-ACK means that a packet has been processed and may be discarded; a
808 * soft-ACK means that the packet may be discarded and retransmission
809 * requested. A phase is complete when all packets are hard-ACK'd.
811 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
814 struct rxrpc_ack_summary summary = { 0 };
815 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
817 struct rxrpc_ackpacket ack;
818 struct rxrpc_ackinfo info;
819 u8 acks[RXRPC_MAXACKS];
821 rxrpc_serial_t acked_serial;
822 rxrpc_seq_t first_soft_ack, hard_ack;
823 int nr_acks, offset, ioffset;
827 offset = sizeof(struct rxrpc_wire_header);
828 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
829 _debug("extraction failure");
830 return rxrpc_proto_abort("XAK", call, 0);
832 offset += sizeof(buf.ack);
834 acked_serial = ntohl(buf.ack.serial);
835 first_soft_ack = ntohl(buf.ack.firstPacket);
836 hard_ack = first_soft_ack - 1;
837 nr_acks = buf.ack.nAcks;
838 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
839 buf.ack.reason : RXRPC_ACK__INVALID);
841 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
842 first_soft_ack, ntohl(buf.ack.previousPacket),
843 summary.ack_reason, nr_acks);
845 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
846 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
848 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
849 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
852 if (buf.ack.reason == RXRPC_ACK_PING) {
853 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
854 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
855 skew, sp->hdr.serial, true, true,
856 rxrpc_propose_ack_respond_to_ping);
857 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
858 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
859 skew, sp->hdr.serial, true, true,
860 rxrpc_propose_ack_respond_to_ack);
863 ioffset = offset + nr_acks + 3;
864 if (skb->len >= ioffset + sizeof(buf.info)) {
865 if (skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
866 return rxrpc_proto_abort("XAI", call, 0);
867 rxrpc_input_ackinfo(call, skb, &buf.info);
870 if (first_soft_ack == 0)
871 return rxrpc_proto_abort("AK0", call, 0);
873 /* Ignore ACKs unless we are or have just been transmitting. */
874 switch (READ_ONCE(call->state)) {
875 case RXRPC_CALL_CLIENT_SEND_REQUEST:
876 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
877 case RXRPC_CALL_SERVER_SEND_REPLY:
878 case RXRPC_CALL_SERVER_AWAIT_ACK:
884 /* Discard any out-of-order or duplicate ACKs. */
885 if (before_eq(sp->hdr.serial, call->acks_latest)) {
886 _debug("discard ACK %d <= %d",
887 sp->hdr.serial, call->acks_latest);
890 call->acks_latest_ts = skb->tstamp;
891 call->acks_latest = sp->hdr.serial;
893 if (before(hard_ack, call->tx_hard_ack) ||
894 after(hard_ack, call->tx_top))
895 return rxrpc_proto_abort("AKW", call, 0);
896 if (nr_acks > call->tx_top - hard_ack)
897 return rxrpc_proto_abort("AKN", call, 0);
899 if (after(hard_ack, call->tx_hard_ack))
900 rxrpc_rotate_tx_window(call, hard_ack, &summary);
903 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0)
904 return rxrpc_proto_abort("XSA", call, 0);
905 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
909 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
910 rxrpc_end_tx_phase(call, false, "ETA");
914 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
915 RXRPC_TX_ANNO_LAST &&
916 summary.nr_acks == call->tx_top - hard_ack &&
917 rxrpc_is_client_call(call))
918 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
920 rxrpc_propose_ack_ping_for_lost_reply);
922 return rxrpc_congestion_management(call, skb, &summary, acked_serial);
926 * Process an ACKALL packet.
928 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
930 struct rxrpc_ack_summary summary = { 0 };
931 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
933 _proto("Rx ACKALL %%%u", sp->hdr.serial);
935 rxrpc_rotate_tx_window(call, call->tx_top, &summary);
936 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags))
937 rxrpc_end_tx_phase(call, false, "ETL");
941 * Process an ABORT packet directed at a call.
943 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
945 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
947 u32 abort_code = RX_CALL_DEAD;
952 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
953 &wtmp, sizeof(wtmp)) >= 0)
954 abort_code = ntohl(wtmp);
956 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
958 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
960 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
961 abort_code, -ECONNABORTED))
962 rxrpc_notify_socket(call);
966 * Process an incoming call packet.
968 static void rxrpc_input_call_packet(struct rxrpc_call *call,
969 struct sk_buff *skb, u16 skew)
971 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
974 _enter("%p,%p", call, skb);
976 timo = READ_ONCE(call->next_rx_timo);
978 unsigned long now = jiffies, expect_rx_by;
980 expect_rx_by = now + timo;
981 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
982 rxrpc_reduce_call_timer(call, expect_rx_by, now,
983 rxrpc_timer_set_for_normal);
986 switch (sp->hdr.type) {
987 case RXRPC_PACKET_TYPE_DATA:
988 rxrpc_input_data(call, skb, skew);
991 case RXRPC_PACKET_TYPE_ACK:
992 rxrpc_input_ack(call, skb, skew);
995 case RXRPC_PACKET_TYPE_BUSY:
996 _proto("Rx BUSY %%%u", sp->hdr.serial);
998 /* Just ignore BUSY packets from the server; the retry and
999 * lifespan timers will take care of business. BUSY packets
1000 * from the client don't make sense.
1004 case RXRPC_PACKET_TYPE_ABORT:
1005 rxrpc_input_abort(call, skb);
1008 case RXRPC_PACKET_TYPE_ACKALL:
1009 rxrpc_input_ackall(call, skb);
1020 * Handle a new call on a channel implicitly completing the preceding call on
1023 * TODO: If callNumber > call_id + 1, renegotiate security.
1025 static void rxrpc_input_implicit_end_call(struct rxrpc_connection *conn,
1026 struct rxrpc_call *call)
1028 switch (READ_ONCE(call->state)) {
1029 case RXRPC_CALL_SERVER_AWAIT_ACK:
1030 rxrpc_call_completed(call);
1032 case RXRPC_CALL_COMPLETE:
1035 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1036 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1037 rxrpc_queue_call(call);
1042 trace_rxrpc_improper_term(call);
1043 __rxrpc_disconnect_call(conn, call);
1044 rxrpc_notify_socket(call);
1048 * post connection-level events to the connection
1049 * - this includes challenges, responses, some aborts and call terminal packet
1052 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1053 struct sk_buff *skb)
1055 _enter("%p,%p", conn, skb);
1057 skb_queue_tail(&conn->rx_queue, skb);
1058 rxrpc_queue_conn(conn);
1062 * post endpoint-level events to the local endpoint
1063 * - this includes debug and version messages
1065 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1066 struct sk_buff *skb)
1068 _enter("%p,%p", local, skb);
1070 skb_queue_tail(&local->event_queue, skb);
1071 rxrpc_queue_local(local);
1075 * put a packet up for transport-level abort
1077 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1079 CHECK_SLAB_OKAY(&local->usage);
1081 skb_queue_tail(&local->reject_queue, skb);
1082 rxrpc_queue_local(local);
1086 * Extract the wire header from a packet and translate the byte order.
1089 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1091 struct rxrpc_wire_header whdr;
1093 /* dig out the RxRPC connection details */
1094 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1095 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1096 tracepoint_string("bad_hdr"));
1100 memset(sp, 0, sizeof(*sp));
1101 sp->hdr.epoch = ntohl(whdr.epoch);
1102 sp->hdr.cid = ntohl(whdr.cid);
1103 sp->hdr.callNumber = ntohl(whdr.callNumber);
1104 sp->hdr.seq = ntohl(whdr.seq);
1105 sp->hdr.serial = ntohl(whdr.serial);
1106 sp->hdr.flags = whdr.flags;
1107 sp->hdr.type = whdr.type;
1108 sp->hdr.userStatus = whdr.userStatus;
1109 sp->hdr.securityIndex = whdr.securityIndex;
1110 sp->hdr._rsvd = ntohs(whdr._rsvd);
1111 sp->hdr.serviceId = ntohs(whdr.serviceId);
1116 * handle data received on the local endpoint
1117 * - may be called in interrupt context
1119 * The socket is locked by the caller and this prevents the socket from being
1120 * shut down and the local endpoint from going away, thus sk_user_data will not
1121 * be cleared until this function returns.
1123 void rxrpc_data_ready(struct sock *udp_sk)
1125 struct rxrpc_connection *conn;
1126 struct rxrpc_channel *chan;
1127 struct rxrpc_call *call;
1128 struct rxrpc_skb_priv *sp;
1129 struct rxrpc_local *local = udp_sk->sk_user_data;
1130 struct sk_buff *skb;
1131 unsigned int channel;
1134 _enter("%p", udp_sk);
1136 ASSERT(!irqs_disabled());
1138 skb = skb_recv_udp(udp_sk, 0, 1, &ret);
1142 _debug("UDP socket error %d", ret);
1146 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
1148 _net("recv skb %p", skb);
1150 /* we'll probably need to checksum it (didn't call sock_recvmsg) */
1151 if (skb_checksum_complete(skb)) {
1152 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1153 __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
1154 _leave(" [CSUM failed]");
1158 __UDP_INC_STATS(&init_net, UDP_MIB_INDATAGRAMS, 0);
1160 /* The UDP protocol already released all skb resources;
1161 * we are free to add our own data there.
1163 sp = rxrpc_skb(skb);
1165 /* dig out the RxRPC connection details */
1166 if (rxrpc_extract_header(sp, skb) < 0)
1169 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1171 if ((lose++ & 7) == 7) {
1172 trace_rxrpc_rx_lose(sp);
1173 rxrpc_lose_skb(skb, rxrpc_skb_rx_lost);
1178 trace_rxrpc_rx_packet(sp);
1180 _net("Rx RxRPC %s ep=%x call=%x:%x",
1181 sp->hdr.flags & RXRPC_CLIENT_INITIATED ? "ToServer" : "ToClient",
1182 sp->hdr.epoch, sp->hdr.cid, sp->hdr.callNumber);
1184 if (sp->hdr.type >= RXRPC_N_PACKET_TYPES ||
1185 !((RXRPC_SUPPORTED_PACKET_TYPES >> sp->hdr.type) & 1)) {
1186 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1190 switch (sp->hdr.type) {
1191 case RXRPC_PACKET_TYPE_VERSION:
1192 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED))
1194 rxrpc_post_packet_to_local(local, skb);
1197 case RXRPC_PACKET_TYPE_BUSY:
1198 if (sp->hdr.flags & RXRPC_CLIENT_INITIATED)
1202 case RXRPC_PACKET_TYPE_DATA:
1203 if (sp->hdr.callNumber == 0)
1205 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
1206 !rxrpc_validate_jumbo(skb))
1210 /* Packet types 9-11 should just be ignored. */
1211 case RXRPC_PACKET_TYPE_PARAMS:
1212 case RXRPC_PACKET_TYPE_10:
1213 case RXRPC_PACKET_TYPE_11:
1219 conn = rxrpc_find_connection_rcu(local, skb);
1221 if (sp->hdr.securityIndex != conn->security_ix)
1222 goto wrong_security;
1224 if (sp->hdr.serviceId != conn->service_id) {
1225 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) ||
1226 conn->service_id != conn->params.service_id)
1228 conn->service_id = sp->hdr.serviceId;
1231 if (sp->hdr.callNumber == 0) {
1232 /* Connection-level packet */
1233 _debug("CONN %p {%d}", conn, conn->debug_id);
1234 rxrpc_post_packet_to_conn(conn, skb);
1238 /* Note the serial number skew here */
1239 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
1242 conn->hi_serial = sp->hdr.serial;
1245 skew = min(skew, 65535);
1248 /* Call-bound packets are routed by connection channel. */
1249 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1250 chan = &conn->channels[channel];
1252 /* Ignore really old calls */
1253 if (sp->hdr.callNumber < chan->last_call)
1254 goto discard_unlock;
1256 if (sp->hdr.callNumber == chan->last_call) {
1258 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1259 goto discard_unlock;
1261 /* For the previous service call, if completed
1262 * successfully, we discard all further packets.
1264 if (rxrpc_conn_is_service(conn) &&
1265 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1266 goto discard_unlock;
1268 /* But otherwise we need to retransmit the final packet
1269 * from data cached in the connection record.
1271 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1272 trace_rxrpc_rx_data(chan->call_debug_id,
1276 rxrpc_post_packet_to_conn(conn, skb);
1280 call = rcu_dereference(chan->call);
1282 if (sp->hdr.callNumber > chan->call_id) {
1283 if (!(sp->hdr.flags & RXRPC_CLIENT_INITIATED)) {
1288 rxrpc_input_implicit_end_call(conn, call);
1293 if (sp->hdr.serviceId != call->service_id)
1294 call->service_id = sp->hdr.serviceId;
1295 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1296 call->rx_serial = sp->hdr.serial;
1297 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1298 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1305 if (!call || atomic_read(&call->usage) == 0) {
1306 if (!(sp->hdr.type & RXRPC_CLIENT_INITIATED) ||
1307 sp->hdr.callNumber == 0 ||
1308 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1309 goto bad_message_unlock;
1310 if (sp->hdr.seq != 1)
1311 goto discard_unlock;
1312 call = rxrpc_new_incoming_call(local, conn, skb);
1317 rxrpc_send_ping(call, skb, skew);
1318 mutex_unlock(&call->user_mutex);
1321 rxrpc_input_call_packet(call, skb, skew);
1322 goto discard_unlock;
1327 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1329 trace_rxrpc_rx_done(0, 0);
1338 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1339 RXKADINCONSISTENCY, EBADMSG);
1340 skb->priority = RXKADINCONSISTENCY;
1345 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1346 RX_PROTOCOL_ERROR, EBADMSG);
1347 goto protocol_error;
1352 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1353 RX_PROTOCOL_ERROR, EBADMSG);
1355 skb->priority = RX_PROTOCOL_ERROR;
1357 skb->mark = RXRPC_SKB_MARK_LOCAL_ABORT;
1359 trace_rxrpc_rx_done(skb->mark, skb->priority);
1360 rxrpc_reject_packet(local, skb);
1361 _leave(" [badmsg]");