1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC packet reception
4 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/module.h>
11 #include <linux/net.h>
12 #include <linux/skbuff.h>
13 #include <linux/errqueue.h>
14 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/icmp.h>
18 #include <linux/gfp.h>
20 #include <net/af_rxrpc.h>
23 #include <net/net_namespace.h>
24 #include "ar-internal.h"
26 static void rxrpc_proto_abort(const char *why,
27 struct rxrpc_call *call, rxrpc_seq_t seq)
29 if (rxrpc_abort_call(why, call, seq, RX_PROTOCOL_ERROR, -EBADMSG)) {
30 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
31 rxrpc_queue_call(call);
36 * Do TCP-style congestion management [RFC 5681].
38 static void rxrpc_congestion_management(struct rxrpc_call *call,
40 struct rxrpc_ack_summary *summary,
41 rxrpc_serial_t acked_serial)
43 enum rxrpc_congest_change change = rxrpc_cong_no_change;
44 unsigned int cumulative_acks = call->cong_cumul_acks;
45 unsigned int cwnd = call->cong_cwnd;
48 summary->flight_size =
49 (call->tx_top - call->tx_hard_ack) - summary->nr_acks;
51 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
52 summary->retrans_timeo = true;
53 call->cong_ssthresh = max_t(unsigned int,
54 summary->flight_size / 2, 2);
56 if (cwnd >= call->cong_ssthresh &&
57 call->cong_mode == RXRPC_CALL_SLOW_START) {
58 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
59 call->cong_tstamp = skb->tstamp;
64 cumulative_acks += summary->nr_new_acks;
65 cumulative_acks += summary->nr_rot_new_acks;
66 if (cumulative_acks > 255)
67 cumulative_acks = 255;
69 summary->mode = call->cong_mode;
70 summary->cwnd = call->cong_cwnd;
71 summary->ssthresh = call->cong_ssthresh;
72 summary->cumulative_acks = cumulative_acks;
73 summary->dup_acks = call->cong_dup_acks;
75 switch (call->cong_mode) {
76 case RXRPC_CALL_SLOW_START:
77 if (summary->nr_nacks > 0)
78 goto packet_loss_detected;
79 if (summary->cumulative_acks > 0)
81 if (cwnd >= call->cong_ssthresh) {
82 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
83 call->cong_tstamp = skb->tstamp;
87 case RXRPC_CALL_CONGEST_AVOIDANCE:
88 if (summary->nr_nacks > 0)
89 goto packet_loss_detected;
91 /* We analyse the number of packets that get ACK'd per RTT
92 * period and increase the window if we managed to fill it.
94 if (call->peer->rtt_usage == 0)
96 if (ktime_before(skb->tstamp,
97 ktime_add_ns(call->cong_tstamp,
100 change = rxrpc_cong_rtt_window_end;
101 call->cong_tstamp = skb->tstamp;
102 if (cumulative_acks >= cwnd)
106 case RXRPC_CALL_PACKET_LOSS:
107 if (summary->nr_nacks == 0)
108 goto resume_normality;
110 if (summary->new_low_nack) {
111 change = rxrpc_cong_new_low_nack;
112 call->cong_dup_acks = 1;
113 if (call->cong_extra > 1)
114 call->cong_extra = 1;
115 goto send_extra_data;
118 call->cong_dup_acks++;
119 if (call->cong_dup_acks < 3)
120 goto send_extra_data;
122 change = rxrpc_cong_begin_retransmission;
123 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
124 call->cong_ssthresh = max_t(unsigned int,
125 summary->flight_size / 2, 2);
126 cwnd = call->cong_ssthresh + 3;
127 call->cong_extra = 0;
128 call->cong_dup_acks = 0;
132 case RXRPC_CALL_FAST_RETRANSMIT:
133 if (!summary->new_low_nack) {
134 if (summary->nr_new_acks == 0)
136 call->cong_dup_acks++;
137 if (call->cong_dup_acks == 2) {
138 change = rxrpc_cong_retransmit_again;
139 call->cong_dup_acks = 0;
143 change = rxrpc_cong_progress;
144 cwnd = call->cong_ssthresh;
145 if (summary->nr_nacks == 0)
146 goto resume_normality;
156 change = rxrpc_cong_cleared_nacks;
157 call->cong_dup_acks = 0;
158 call->cong_extra = 0;
159 call->cong_tstamp = skb->tstamp;
160 if (cwnd < call->cong_ssthresh)
161 call->cong_mode = RXRPC_CALL_SLOW_START;
163 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
167 if (cwnd >= RXRPC_RXTX_BUFF_SIZE - 1)
168 cwnd = RXRPC_RXTX_BUFF_SIZE - 1;
169 call->cong_cwnd = cwnd;
170 call->cong_cumul_acks = cumulative_acks;
171 trace_rxrpc_congest(call, summary, acked_serial, change);
172 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
173 rxrpc_queue_call(call);
176 packet_loss_detected:
177 change = rxrpc_cong_saw_nack;
178 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
179 call->cong_dup_acks = 0;
180 goto send_extra_data;
183 /* Send some previously unsent DATA if we have some to advance the ACK
186 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
187 RXRPC_TX_ANNO_LAST ||
188 summary->nr_acks != call->tx_top - call->tx_hard_ack) {
190 wake_up(&call->waitq);
192 goto out_no_clear_ca;
196 * Ping the other end to fill our RTT cache and to retrieve the rwind
197 * and MTU parameters.
199 static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb,
202 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
203 ktime_t now = skb->tstamp;
205 if (call->peer->rtt_usage < 3 ||
206 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now))
207 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
209 rxrpc_propose_ack_ping_for_params);
213 * Apply a hard ACK by advancing the Tx window.
215 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
216 struct rxrpc_ack_summary *summary)
218 struct sk_buff *skb, *list = NULL;
219 bool rot_last = false;
223 if (call->acks_lowest_nak == call->tx_hard_ack) {
224 call->acks_lowest_nak = to;
225 } else if (before_eq(call->acks_lowest_nak, to)) {
226 summary->new_low_nack = true;
227 call->acks_lowest_nak = to;
230 spin_lock(&call->lock);
232 while (before(call->tx_hard_ack, to)) {
234 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
235 skb = call->rxtx_buffer[ix];
236 annotation = call->rxtx_annotations[ix];
237 rxrpc_see_skb(skb, rxrpc_skb_tx_rotated);
238 call->rxtx_buffer[ix] = NULL;
239 call->rxtx_annotations[ix] = 0;
243 if (annotation & RXRPC_TX_ANNO_LAST) {
244 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
247 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
248 summary->nr_rot_new_acks++;
251 spin_unlock(&call->lock);
253 trace_rxrpc_transmit(call, (rot_last ?
254 rxrpc_transmit_rotate_last :
255 rxrpc_transmit_rotate));
256 wake_up(&call->waitq);
261 skb_mark_not_on_list(skb);
262 rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
269 * End the transmission phase of a call.
271 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
272 * or a final ACK packet.
274 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
275 const char *abort_why)
279 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
281 write_lock(&call->state_lock);
285 case RXRPC_CALL_CLIENT_SEND_REQUEST:
286 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
288 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
290 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
293 case RXRPC_CALL_SERVER_AWAIT_ACK:
294 __rxrpc_call_completed(call);
295 rxrpc_notify_socket(call);
303 write_unlock(&call->state_lock);
304 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
305 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
307 trace_rxrpc_transmit(call, rxrpc_transmit_end);
312 write_unlock(&call->state_lock);
313 kdebug("end_tx %s", rxrpc_call_states[call->state]);
314 rxrpc_proto_abort(abort_why, call, call->tx_top);
319 * Begin the reply reception phase of a call.
321 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
323 struct rxrpc_ack_summary summary = { 0 };
324 unsigned long now, timo;
325 rxrpc_seq_t top = READ_ONCE(call->tx_top);
327 if (call->ackr_reason) {
328 spin_lock_bh(&call->lock);
329 call->ackr_reason = 0;
330 spin_unlock_bh(&call->lock);
332 timo = now + MAX_JIFFY_OFFSET;
333 WRITE_ONCE(call->resend_at, timo);
334 WRITE_ONCE(call->ack_at, timo);
335 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
338 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
339 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
340 rxrpc_proto_abort("TXL", call, top);
344 if (!rxrpc_end_tx_phase(call, true, "ETD"))
346 call->tx_phase = false;
351 * Scan a jumbo packet to validate its structure and to work out how many
352 * subpackets it contains.
354 * A jumbo packet is a collection of consecutive packets glued together with
355 * little headers between that indicate how to change the initial header for
358 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
359 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
362 static bool rxrpc_validate_jumbo(struct sk_buff *skb)
364 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
365 unsigned int offset = sizeof(struct rxrpc_wire_header);
366 unsigned int len = skb->len;
368 u8 flags = sp->hdr.flags;
372 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
374 if (flags & RXRPC_LAST_PACKET)
376 offset += RXRPC_JUMBO_DATALEN;
377 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
379 offset += sizeof(struct rxrpc_jumbo_header);
380 } while (flags & RXRPC_JUMBO_PACKET);
382 sp->nr_jumbo = nr_jumbo;
390 * Handle reception of a duplicate packet.
392 * We have to take care to avoid an attack here whereby we're given a series of
393 * jumbograms, each with a sequence number one before the preceding one and
394 * filled up to maximum UDP size. If they never send us the first packet in
395 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
396 * space until the call times out.
398 * We limit the space usage by only accepting three duplicate jumbo packets per
399 * call. After that, we tell the other side we're no longer accepting jumbos
400 * (that information is encoded in the ACK packet).
402 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
403 u8 annotation, bool *_jumbo_bad)
405 /* Discard normal packets that are duplicates. */
409 /* Skip jumbo subpackets that are duplicates. When we've had three or
410 * more partially duplicate jumbo packets, we refuse to take any more
411 * jumbos for this call.
414 call->nr_jumbo_bad++;
420 * Process a DATA packet, adding the packet to the Rx ring.
422 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb,
425 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
426 enum rxrpc_call_state state;
427 unsigned int offset = sizeof(struct rxrpc_wire_header);
429 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
430 rxrpc_seq_t seq = sp->hdr.seq, hard_ack;
431 bool immediate_ack = false, jumbo_bad = false, queued;
433 u8 ack = 0, flags, annotation = 0;
435 _enter("{%u,%u},{%u,%u}",
436 call->rx_hard_ack, call->rx_top, skb->len, seq);
438 _proto("Rx DATA %%%u { #%u f=%02x }",
439 sp->hdr.serial, seq, sp->hdr.flags);
441 state = READ_ONCE(call->state);
442 if (state >= RXRPC_CALL_COMPLETE)
445 if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
446 unsigned long timo = READ_ONCE(call->next_req_timo);
447 unsigned long now, expect_req_by;
451 expect_req_by = now + timo;
452 WRITE_ONCE(call->expect_req_by, expect_req_by);
453 rxrpc_reduce_call_timer(call, expect_req_by, now,
454 rxrpc_timer_set_for_idle);
458 spin_lock(&call->input_lock);
460 /* Received data implicitly ACKs all of the request packets we sent
461 * when we're acting as a client.
463 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
464 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
465 !rxrpc_receiving_reply(call))
468 call->ackr_prev_seq = seq;
470 hard_ack = READ_ONCE(call->rx_hard_ack);
471 if (after(seq, hard_ack + call->rx_winsize)) {
472 ack = RXRPC_ACK_EXCEEDS_WINDOW;
477 flags = sp->hdr.flags;
478 if (flags & RXRPC_JUMBO_PACKET) {
479 if (call->nr_jumbo_bad > 3) {
480 ack = RXRPC_ACK_NOSPACE;
489 ix = seq & RXRPC_RXTX_BUFF_MASK;
491 if (flags & RXRPC_JUMBO_PACKET)
492 len = RXRPC_JUMBO_DATALEN;
494 if (flags & RXRPC_LAST_PACKET) {
495 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
496 seq != call->rx_top) {
497 rxrpc_proto_abort("LSN", call, seq);
501 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
502 after_eq(seq, call->rx_top)) {
503 rxrpc_proto_abort("LSA", call, seq);
508 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
509 if (before_eq(seq, hard_ack)) {
510 ack = RXRPC_ACK_DUPLICATE;
515 if (flags & RXRPC_REQUEST_ACK && !ack) {
516 ack = RXRPC_ACK_REQUESTED;
520 if (call->rxtx_buffer[ix]) {
521 rxrpc_input_dup_data(call, seq, annotation, &jumbo_bad);
522 if (ack != RXRPC_ACK_DUPLICATE) {
523 ack = RXRPC_ACK_DUPLICATE;
526 immediate_ack = true;
530 /* Queue the packet. We use a couple of memory barriers here as need
531 * to make sure that rx_top is perceived to be set after the buffer
532 * pointer and that the buffer pointer is set after the annotation and
535 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
536 * and also rxrpc_fill_out_ack().
538 rxrpc_get_skb(skb, rxrpc_skb_rx_got);
539 call->rxtx_annotations[ix] = annotation;
541 call->rxtx_buffer[ix] = skb;
542 if (after(seq, call->rx_top)) {
543 smp_store_release(&call->rx_top, seq);
544 } else if (before(seq, call->rx_top)) {
545 /* Send an immediate ACK if we fill in a hole */
547 ack = RXRPC_ACK_DELAY;
550 immediate_ack = true;
552 if (flags & RXRPC_LAST_PACKET) {
553 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
554 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
556 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
560 if (after_eq(seq, call->rx_expect_next)) {
561 if (after(seq, call->rx_expect_next)) {
562 _net("OOS %u > %u", seq, call->rx_expect_next);
563 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
566 call->rx_expect_next = seq + 1;
571 if (flags & RXRPC_JUMBO_PACKET) {
572 if (skb_copy_bits(skb, offset, &flags, 1) < 0) {
573 rxrpc_proto_abort("XJF", call, seq);
576 offset += sizeof(struct rxrpc_jumbo_header);
580 if (flags & RXRPC_JUMBO_PACKET)
581 annotation |= RXRPC_RX_ANNO_JLAST;
582 if (after(seq, hard_ack + call->rx_winsize)) {
583 ack = RXRPC_ACK_EXCEEDS_WINDOW;
586 call->nr_jumbo_bad++;
592 _proto("Rx DATA Jumbo %%%u", serial);
596 if (queued && flags & RXRPC_LAST_PACKET && !ack) {
597 ack = RXRPC_ACK_DELAY;
603 rxrpc_propose_ACK(call, ack, skew, ack_serial,
605 rxrpc_propose_ack_input_data);
607 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, skew, serial,
609 rxrpc_propose_ack_input_data);
611 if (sp->hdr.seq == READ_ONCE(call->rx_hard_ack) + 1) {
612 trace_rxrpc_notify_socket(call->debug_id, serial);
613 rxrpc_notify_socket(call);
617 spin_unlock(&call->input_lock);
622 * Process a requested ACK.
624 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
626 rxrpc_serial_t orig_serial,
627 rxrpc_serial_t ack_serial)
629 struct rxrpc_skb_priv *sp;
634 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
635 skb = call->rxtx_buffer[ix];
639 sent_at = skb->tstamp;
640 smp_rmb(); /* Read timestamp before serial. */
642 if (sp->hdr.serial != orig_serial)
650 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
651 orig_serial, ack_serial, sent_at, resp_time);
655 * Process the response to a ping that we sent to find out if we lost an ACK.
657 * If we got back a ping response that indicates a lower tx_top than what we
658 * had at the time of the ping transmission, we adjudge all the DATA packets
659 * sent between the response tx_top and the ping-time tx_top to have been lost.
661 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
663 rxrpc_seq_t top, bottom, seq;
666 spin_lock_bh(&call->lock);
668 bottom = call->tx_hard_ack + 1;
669 top = call->acks_lost_top;
670 if (before(bottom, top)) {
671 for (seq = bottom; before_eq(seq, top); seq++) {
672 int ix = seq & RXRPC_RXTX_BUFF_MASK;
673 u8 annotation = call->rxtx_annotations[ix];
674 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
676 if (anno_type != RXRPC_TX_ANNO_UNACK)
678 annotation &= ~RXRPC_TX_ANNO_MASK;
679 annotation |= RXRPC_TX_ANNO_RETRANS;
680 call->rxtx_annotations[ix] = annotation;
685 spin_unlock_bh(&call->lock);
687 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
688 rxrpc_queue_call(call);
692 * Process a ping response.
694 static void rxrpc_input_ping_response(struct rxrpc_call *call,
696 rxrpc_serial_t orig_serial,
697 rxrpc_serial_t ack_serial)
699 rxrpc_serial_t ping_serial;
702 ping_time = call->ping_time;
704 ping_serial = READ_ONCE(call->ping_serial);
706 if (orig_serial == call->acks_lost_ping)
707 rxrpc_input_check_for_lost_ack(call);
709 if (before(orig_serial, ping_serial) ||
710 !test_and_clear_bit(RXRPC_CALL_PINGING, &call->flags))
712 if (after(orig_serial, ping_serial))
715 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
716 orig_serial, ack_serial, ping_time, resp_time);
720 * Process the extra information that may be appended to an ACK packet
722 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
723 struct rxrpc_ackinfo *ackinfo)
725 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
726 struct rxrpc_peer *peer;
729 u32 rwind = ntohl(ackinfo->rwind);
731 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
733 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
734 rwind, ntohl(ackinfo->jumbo_max));
736 if (call->tx_winsize != rwind) {
737 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
738 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
739 if (rwind > call->tx_winsize)
741 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
742 ntohl(ackinfo->rwind), wake);
743 call->tx_winsize = rwind;
746 if (call->cong_ssthresh > rwind)
747 call->cong_ssthresh = rwind;
749 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
752 if (mtu < peer->maxdata) {
753 spin_lock_bh(&peer->lock);
755 peer->mtu = mtu + peer->hdrsize;
756 spin_unlock_bh(&peer->lock);
757 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
761 wake_up(&call->waitq);
765 * Process individual soft ACKs.
767 * Each ACK in the array corresponds to one packet and can be either an ACK or
768 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
769 * packets that lie beyond the end of the ACK list are scheduled for resend by
770 * the timer on the basis that the peer might just not have processed them at
771 * the time the ACK was sent.
773 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
774 rxrpc_seq_t seq, int nr_acks,
775 struct rxrpc_ack_summary *summary)
778 u8 annotation, anno_type;
780 for (; nr_acks > 0; nr_acks--, seq++) {
781 ix = seq & RXRPC_RXTX_BUFF_MASK;
782 annotation = call->rxtx_annotations[ix];
783 anno_type = annotation & RXRPC_TX_ANNO_MASK;
784 annotation &= ~RXRPC_TX_ANNO_MASK;
786 case RXRPC_ACK_TYPE_ACK:
788 if (anno_type == RXRPC_TX_ANNO_ACK)
790 summary->nr_new_acks++;
791 call->rxtx_annotations[ix] =
792 RXRPC_TX_ANNO_ACK | annotation;
794 case RXRPC_ACK_TYPE_NACK:
795 if (!summary->nr_nacks &&
796 call->acks_lowest_nak != seq) {
797 call->acks_lowest_nak = seq;
798 summary->new_low_nack = true;
801 if (anno_type == RXRPC_TX_ANNO_NAK)
803 summary->nr_new_nacks++;
804 if (anno_type == RXRPC_TX_ANNO_RETRANS)
806 call->rxtx_annotations[ix] =
807 RXRPC_TX_ANNO_NAK | annotation;
810 return rxrpc_proto_abort("SFT", call, 0);
816 * Process an ACK packet.
818 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
819 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
821 * A hard-ACK means that a packet has been processed and may be discarded; a
822 * soft-ACK means that the packet may be discarded and retransmission
823 * requested. A phase is complete when all packets are hard-ACK'd.
825 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb,
828 struct rxrpc_ack_summary summary = { 0 };
829 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
831 struct rxrpc_ackpacket ack;
832 struct rxrpc_ackinfo info;
833 u8 acks[RXRPC_MAXACKS];
835 rxrpc_serial_t acked_serial;
836 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
837 int nr_acks, offset, ioffset;
841 offset = sizeof(struct rxrpc_wire_header);
842 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
843 _debug("extraction failure");
844 return rxrpc_proto_abort("XAK", call, 0);
846 offset += sizeof(buf.ack);
848 acked_serial = ntohl(buf.ack.serial);
849 first_soft_ack = ntohl(buf.ack.firstPacket);
850 prev_pkt = ntohl(buf.ack.previousPacket);
851 hard_ack = first_soft_ack - 1;
852 nr_acks = buf.ack.nAcks;
853 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
854 buf.ack.reason : RXRPC_ACK__INVALID);
856 trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
857 first_soft_ack, prev_pkt,
858 summary.ack_reason, nr_acks);
860 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
861 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
863 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
864 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
867 if (buf.ack.reason == RXRPC_ACK_PING) {
868 _proto("Rx ACK %%%u PING Request", sp->hdr.serial);
869 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
870 skew, sp->hdr.serial, true, true,
871 rxrpc_propose_ack_respond_to_ping);
872 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
873 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
874 skew, sp->hdr.serial, true, true,
875 rxrpc_propose_ack_respond_to_ack);
878 /* Discard any out-of-order or duplicate ACKs (outside lock). */
879 if (before(first_soft_ack, call->ackr_first_seq) ||
880 before(prev_pkt, call->ackr_prev_seq))
884 ioffset = offset + nr_acks + 3;
885 if (skb->len >= ioffset + sizeof(buf.info) &&
886 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
887 return rxrpc_proto_abort("XAI", call, 0);
889 spin_lock(&call->input_lock);
891 /* Discard any out-of-order or duplicate ACKs (inside lock). */
892 if (before(first_soft_ack, call->ackr_first_seq) ||
893 before(prev_pkt, call->ackr_prev_seq))
895 call->acks_latest_ts = skb->tstamp;
896 call->acks_latest = sp->hdr.serial;
898 call->ackr_first_seq = first_soft_ack;
899 call->ackr_prev_seq = prev_pkt;
901 /* Parse rwind and mtu sizes if provided. */
903 rxrpc_input_ackinfo(call, skb, &buf.info);
905 if (first_soft_ack == 0) {
906 rxrpc_proto_abort("AK0", call, 0);
910 /* Ignore ACKs unless we are or have just been transmitting. */
911 switch (READ_ONCE(call->state)) {
912 case RXRPC_CALL_CLIENT_SEND_REQUEST:
913 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
914 case RXRPC_CALL_SERVER_SEND_REPLY:
915 case RXRPC_CALL_SERVER_AWAIT_ACK:
921 if (before(hard_ack, call->tx_hard_ack) ||
922 after(hard_ack, call->tx_top)) {
923 rxrpc_proto_abort("AKW", call, 0);
926 if (nr_acks > call->tx_top - hard_ack) {
927 rxrpc_proto_abort("AKN", call, 0);
931 if (after(hard_ack, call->tx_hard_ack)) {
932 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
933 rxrpc_end_tx_phase(call, false, "ETA");
939 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
940 rxrpc_proto_abort("XSA", call, 0);
943 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
947 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
948 RXRPC_TX_ANNO_LAST &&
949 summary.nr_acks == call->tx_top - hard_ack &&
950 rxrpc_is_client_call(call))
951 rxrpc_propose_ACK(call, RXRPC_ACK_PING, skew, sp->hdr.serial,
953 rxrpc_propose_ack_ping_for_lost_reply);
955 rxrpc_congestion_management(call, skb, &summary, acked_serial);
957 spin_unlock(&call->input_lock);
961 * Process an ACKALL packet.
963 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
965 struct rxrpc_ack_summary summary = { 0 };
966 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
968 _proto("Rx ACKALL %%%u", sp->hdr.serial);
970 spin_lock(&call->input_lock);
972 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
973 rxrpc_end_tx_phase(call, false, "ETL");
975 spin_unlock(&call->input_lock);
979 * Process an ABORT packet directed at a call.
981 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
983 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
985 u32 abort_code = RX_CALL_DEAD;
990 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
991 &wtmp, sizeof(wtmp)) >= 0)
992 abort_code = ntohl(wtmp);
994 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
996 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
998 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
999 abort_code, -ECONNABORTED))
1000 rxrpc_notify_socket(call);
1004 * Process an incoming call packet.
1006 static void rxrpc_input_call_packet(struct rxrpc_call *call,
1007 struct sk_buff *skb, u16 skew)
1009 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1012 _enter("%p,%p", call, skb);
1014 timo = READ_ONCE(call->next_rx_timo);
1016 unsigned long now = jiffies, expect_rx_by;
1018 expect_rx_by = now + timo;
1019 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1020 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1021 rxrpc_timer_set_for_normal);
1024 switch (sp->hdr.type) {
1025 case RXRPC_PACKET_TYPE_DATA:
1026 rxrpc_input_data(call, skb, skew);
1029 case RXRPC_PACKET_TYPE_ACK:
1030 rxrpc_input_ack(call, skb, skew);
1033 case RXRPC_PACKET_TYPE_BUSY:
1034 _proto("Rx BUSY %%%u", sp->hdr.serial);
1036 /* Just ignore BUSY packets from the server; the retry and
1037 * lifespan timers will take care of business. BUSY packets
1038 * from the client don't make sense.
1042 case RXRPC_PACKET_TYPE_ABORT:
1043 rxrpc_input_abort(call, skb);
1046 case RXRPC_PACKET_TYPE_ACKALL:
1047 rxrpc_input_ackall(call, skb);
1058 * Handle a new service call on a channel implicitly completing the preceding
1059 * call on that channel. This does not apply to client conns.
1061 * TODO: If callNumber > call_id + 1, renegotiate security.
1063 static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1064 struct rxrpc_connection *conn,
1065 struct rxrpc_call *call)
1067 switch (READ_ONCE(call->state)) {
1068 case RXRPC_CALL_SERVER_AWAIT_ACK:
1069 rxrpc_call_completed(call);
1071 case RXRPC_CALL_COMPLETE:
1074 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1075 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1076 rxrpc_queue_call(call);
1078 trace_rxrpc_improper_term(call);
1082 spin_lock(&rx->incoming_lock);
1083 __rxrpc_disconnect_call(conn, call);
1084 spin_unlock(&rx->incoming_lock);
1085 rxrpc_notify_socket(call);
1089 * post connection-level events to the connection
1090 * - this includes challenges, responses, some aborts and call terminal packet
1093 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1094 struct sk_buff *skb)
1096 _enter("%p,%p", conn, skb);
1098 skb_queue_tail(&conn->rx_queue, skb);
1099 rxrpc_queue_conn(conn);
1103 * post endpoint-level events to the local endpoint
1104 * - this includes debug and version messages
1106 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1107 struct sk_buff *skb)
1109 _enter("%p,%p", local, skb);
1111 skb_queue_tail(&local->event_queue, skb);
1112 rxrpc_queue_local(local);
1116 * put a packet up for transport-level abort
1118 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1120 CHECK_SLAB_OKAY(&local->usage);
1122 skb_queue_tail(&local->reject_queue, skb);
1123 rxrpc_queue_local(local);
1127 * Extract the wire header from a packet and translate the byte order.
1130 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1132 struct rxrpc_wire_header whdr;
1134 /* dig out the RxRPC connection details */
1135 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1136 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1137 tracepoint_string("bad_hdr"));
1141 memset(sp, 0, sizeof(*sp));
1142 sp->hdr.epoch = ntohl(whdr.epoch);
1143 sp->hdr.cid = ntohl(whdr.cid);
1144 sp->hdr.callNumber = ntohl(whdr.callNumber);
1145 sp->hdr.seq = ntohl(whdr.seq);
1146 sp->hdr.serial = ntohl(whdr.serial);
1147 sp->hdr.flags = whdr.flags;
1148 sp->hdr.type = whdr.type;
1149 sp->hdr.userStatus = whdr.userStatus;
1150 sp->hdr.securityIndex = whdr.securityIndex;
1151 sp->hdr._rsvd = ntohs(whdr._rsvd);
1152 sp->hdr.serviceId = ntohs(whdr.serviceId);
1157 * handle data received on the local endpoint
1158 * - may be called in interrupt context
1160 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1161 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1162 * being cleared in the middle of processing this function.
1164 * Called with the RCU read lock held from the IP layer via UDP.
1166 int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1168 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1169 struct rxrpc_connection *conn;
1170 struct rxrpc_channel *chan;
1171 struct rxrpc_call *call = NULL;
1172 struct rxrpc_skb_priv *sp;
1173 struct rxrpc_peer *peer = NULL;
1174 struct rxrpc_sock *rx = NULL;
1175 unsigned int channel;
1178 _enter("%p", udp_sk);
1180 if (unlikely(!local)) {
1184 if (skb->tstamp == 0)
1185 skb->tstamp = ktime_get_real();
1187 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
1189 skb_pull(skb, sizeof(struct udphdr));
1191 /* The UDP protocol already released all skb resources;
1192 * we are free to add our own data there.
1194 sp = rxrpc_skb(skb);
1196 /* dig out the RxRPC connection details */
1197 if (rxrpc_extract_header(sp, skb) < 0)
1200 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1202 if ((lose++ & 7) == 7) {
1203 trace_rxrpc_rx_lose(sp);
1204 rxrpc_free_skb(skb, rxrpc_skb_rx_lost);
1209 if (skb->tstamp == 0)
1210 skb->tstamp = ktime_get_real();
1211 trace_rxrpc_rx_packet(sp);
1213 switch (sp->hdr.type) {
1214 case RXRPC_PACKET_TYPE_VERSION:
1215 if (rxrpc_to_client(sp))
1217 rxrpc_post_packet_to_local(local, skb);
1220 case RXRPC_PACKET_TYPE_BUSY:
1221 if (rxrpc_to_server(sp))
1224 case RXRPC_PACKET_TYPE_ACK:
1225 case RXRPC_PACKET_TYPE_ACKALL:
1226 if (sp->hdr.callNumber == 0)
1229 case RXRPC_PACKET_TYPE_ABORT:
1232 case RXRPC_PACKET_TYPE_DATA:
1233 if (sp->hdr.callNumber == 0 ||
1236 if (sp->hdr.flags & RXRPC_JUMBO_PACKET &&
1237 !rxrpc_validate_jumbo(skb))
1241 case RXRPC_PACKET_TYPE_CHALLENGE:
1242 if (rxrpc_to_server(sp))
1245 case RXRPC_PACKET_TYPE_RESPONSE:
1246 if (rxrpc_to_client(sp))
1250 /* Packet types 9-11 should just be ignored. */
1251 case RXRPC_PACKET_TYPE_PARAMS:
1252 case RXRPC_PACKET_TYPE_10:
1253 case RXRPC_PACKET_TYPE_11:
1257 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1261 if (sp->hdr.serviceId == 0)
1264 if (rxrpc_to_server(sp)) {
1265 /* Weed out packets to services we're not offering. Packets
1266 * that would begin a call are explicitly rejected and the rest
1267 * are just discarded.
1269 rx = rcu_dereference(local->service);
1270 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1271 sp->hdr.serviceId != rx->second_service)) {
1272 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1274 goto unsupported_service;
1279 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1281 if (sp->hdr.securityIndex != conn->security_ix)
1282 goto wrong_security;
1284 if (sp->hdr.serviceId != conn->service_id) {
1287 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1289 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1292 if (old_id != conn->params.service_id &&
1293 old_id != sp->hdr.serviceId)
1297 if (sp->hdr.callNumber == 0) {
1298 /* Connection-level packet */
1299 _debug("CONN %p {%d}", conn, conn->debug_id);
1300 rxrpc_post_packet_to_conn(conn, skb);
1304 /* Note the serial number skew here */
1305 skew = (int)sp->hdr.serial - (int)conn->hi_serial;
1308 conn->hi_serial = sp->hdr.serial;
1311 skew = min(skew, 65535);
1314 /* Call-bound packets are routed by connection channel. */
1315 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1316 chan = &conn->channels[channel];
1318 /* Ignore really old calls */
1319 if (sp->hdr.callNumber < chan->last_call)
1322 if (sp->hdr.callNumber == chan->last_call) {
1324 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1327 /* For the previous service call, if completed
1328 * successfully, we discard all further packets.
1330 if (rxrpc_conn_is_service(conn) &&
1331 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1334 /* But otherwise we need to retransmit the final packet
1335 * from data cached in the connection record.
1337 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1338 trace_rxrpc_rx_data(chan->call_debug_id,
1342 rxrpc_post_packet_to_conn(conn, skb);
1346 call = rcu_dereference(chan->call);
1348 if (sp->hdr.callNumber > chan->call_id) {
1349 if (rxrpc_to_client(sp))
1352 rxrpc_input_implicit_end_call(rx, conn, call);
1357 if (sp->hdr.serviceId != call->service_id)
1358 call->service_id = sp->hdr.serviceId;
1359 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1360 call->rx_serial = sp->hdr.serial;
1361 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1362 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1366 if (!call || atomic_read(&call->usage) == 0) {
1367 if (rxrpc_to_client(sp) ||
1368 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1370 if (sp->hdr.seq != 1)
1372 call = rxrpc_new_incoming_call(local, rx, skb);
1375 rxrpc_send_ping(call, skb, skew);
1376 mutex_unlock(&call->user_mutex);
1379 rxrpc_input_call_packet(call, skb, skew);
1383 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
1385 trace_rxrpc_rx_done(0, 0);
1389 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1390 RXKADINCONSISTENCY, EBADMSG);
1391 skb->priority = RXKADINCONSISTENCY;
1394 unsupported_service:
1395 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1396 RX_INVALID_OPERATION, EOPNOTSUPP);
1397 skb->priority = RX_INVALID_OPERATION;
1401 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1402 RX_PROTOCOL_ERROR, EBADMSG);
1403 goto protocol_error;
1406 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1407 RX_PROTOCOL_ERROR, EBADMSG);
1409 skb->priority = RX_PROTOCOL_ERROR;
1411 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1413 trace_rxrpc_rx_done(skb->mark, skb->priority);
1414 rxrpc_reject_packet(local, skb);
1415 _leave(" [badmsg]");