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 "ar-internal.h"
12 static void rxrpc_proto_abort(const char *why,
13 struct rxrpc_call *call, rxrpc_seq_t seq)
15 if (rxrpc_abort_call(why, call, seq, RX_PROTOCOL_ERROR, -EBADMSG)) {
16 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
17 rxrpc_queue_call(call);
22 * Do TCP-style congestion management [RFC 5681].
24 static void rxrpc_congestion_management(struct rxrpc_call *call,
26 struct rxrpc_ack_summary *summary,
27 rxrpc_serial_t acked_serial)
29 enum rxrpc_congest_change change = rxrpc_cong_no_change;
30 unsigned int cumulative_acks = call->cong_cumul_acks;
31 unsigned int cwnd = call->cong_cwnd;
34 summary->flight_size =
35 (call->tx_top - call->tx_hard_ack) - summary->nr_acks;
37 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
38 summary->retrans_timeo = true;
39 call->cong_ssthresh = max_t(unsigned int,
40 summary->flight_size / 2, 2);
42 if (cwnd >= call->cong_ssthresh &&
43 call->cong_mode == RXRPC_CALL_SLOW_START) {
44 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
45 call->cong_tstamp = skb->tstamp;
50 cumulative_acks += summary->nr_new_acks;
51 cumulative_acks += summary->nr_rot_new_acks;
52 if (cumulative_acks > 255)
53 cumulative_acks = 255;
55 summary->mode = call->cong_mode;
56 summary->cwnd = call->cong_cwnd;
57 summary->ssthresh = call->cong_ssthresh;
58 summary->cumulative_acks = cumulative_acks;
59 summary->dup_acks = call->cong_dup_acks;
61 switch (call->cong_mode) {
62 case RXRPC_CALL_SLOW_START:
63 if (summary->nr_nacks > 0)
64 goto packet_loss_detected;
65 if (summary->cumulative_acks > 0)
67 if (cwnd >= call->cong_ssthresh) {
68 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
69 call->cong_tstamp = skb->tstamp;
73 case RXRPC_CALL_CONGEST_AVOIDANCE:
74 if (summary->nr_nacks > 0)
75 goto packet_loss_detected;
77 /* We analyse the number of packets that get ACK'd per RTT
78 * period and increase the window if we managed to fill it.
80 if (call->peer->rtt_count == 0)
82 if (ktime_before(skb->tstamp,
83 ktime_add_us(call->cong_tstamp,
84 call->peer->srtt_us >> 3)))
86 change = rxrpc_cong_rtt_window_end;
87 call->cong_tstamp = skb->tstamp;
88 if (cumulative_acks >= cwnd)
92 case RXRPC_CALL_PACKET_LOSS:
93 if (summary->nr_nacks == 0)
94 goto resume_normality;
96 if (summary->new_low_nack) {
97 change = rxrpc_cong_new_low_nack;
98 call->cong_dup_acks = 1;
99 if (call->cong_extra > 1)
100 call->cong_extra = 1;
101 goto send_extra_data;
104 call->cong_dup_acks++;
105 if (call->cong_dup_acks < 3)
106 goto send_extra_data;
108 change = rxrpc_cong_begin_retransmission;
109 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
110 call->cong_ssthresh = max_t(unsigned int,
111 summary->flight_size / 2, 2);
112 cwnd = call->cong_ssthresh + 3;
113 call->cong_extra = 0;
114 call->cong_dup_acks = 0;
118 case RXRPC_CALL_FAST_RETRANSMIT:
119 if (!summary->new_low_nack) {
120 if (summary->nr_new_acks == 0)
122 call->cong_dup_acks++;
123 if (call->cong_dup_acks == 2) {
124 change = rxrpc_cong_retransmit_again;
125 call->cong_dup_acks = 0;
129 change = rxrpc_cong_progress;
130 cwnd = call->cong_ssthresh;
131 if (summary->nr_nacks == 0)
132 goto resume_normality;
142 change = rxrpc_cong_cleared_nacks;
143 call->cong_dup_acks = 0;
144 call->cong_extra = 0;
145 call->cong_tstamp = skb->tstamp;
146 if (cwnd < call->cong_ssthresh)
147 call->cong_mode = RXRPC_CALL_SLOW_START;
149 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
153 if (cwnd >= RXRPC_RXTX_BUFF_SIZE - 1)
154 cwnd = RXRPC_RXTX_BUFF_SIZE - 1;
155 call->cong_cwnd = cwnd;
156 call->cong_cumul_acks = cumulative_acks;
157 trace_rxrpc_congest(call, summary, acked_serial, change);
158 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
159 rxrpc_queue_call(call);
162 packet_loss_detected:
163 change = rxrpc_cong_saw_nack;
164 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
165 call->cong_dup_acks = 0;
166 goto send_extra_data;
169 /* Send some previously unsent DATA if we have some to advance the ACK
172 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
173 RXRPC_TX_ANNO_LAST ||
174 summary->nr_acks != call->tx_top - call->tx_hard_ack) {
176 wake_up(&call->waitq);
178 goto out_no_clear_ca;
182 * Apply a hard ACK by advancing the Tx window.
184 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
185 struct rxrpc_ack_summary *summary)
187 struct sk_buff *skb, *list = NULL;
188 bool rot_last = false;
192 if (call->acks_lowest_nak == call->tx_hard_ack) {
193 call->acks_lowest_nak = to;
194 } else if (before_eq(call->acks_lowest_nak, to)) {
195 summary->new_low_nack = true;
196 call->acks_lowest_nak = to;
199 spin_lock(&call->lock);
201 while (before(call->tx_hard_ack, to)) {
203 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
204 skb = call->rxtx_buffer[ix];
205 annotation = call->rxtx_annotations[ix];
206 rxrpc_see_skb(skb, rxrpc_skb_rotated);
207 call->rxtx_buffer[ix] = NULL;
208 call->rxtx_annotations[ix] = 0;
212 if (annotation & RXRPC_TX_ANNO_LAST) {
213 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
216 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
217 summary->nr_rot_new_acks++;
220 spin_unlock(&call->lock);
222 trace_rxrpc_transmit(call, (rot_last ?
223 rxrpc_transmit_rotate_last :
224 rxrpc_transmit_rotate));
225 wake_up(&call->waitq);
230 skb_mark_not_on_list(skb);
231 rxrpc_free_skb(skb, rxrpc_skb_freed);
238 * End the transmission phase of a call.
240 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
241 * or a final ACK packet.
243 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
244 const char *abort_why)
248 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
250 write_lock(&call->state_lock);
254 case RXRPC_CALL_CLIENT_SEND_REQUEST:
255 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
257 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
259 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
262 case RXRPC_CALL_SERVER_AWAIT_ACK:
263 __rxrpc_call_completed(call);
271 write_unlock(&call->state_lock);
272 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
273 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
275 trace_rxrpc_transmit(call, rxrpc_transmit_end);
280 write_unlock(&call->state_lock);
281 kdebug("end_tx %s", rxrpc_call_states[call->state]);
282 rxrpc_proto_abort(abort_why, call, call->tx_top);
287 * Begin the reply reception phase of a call.
289 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
291 struct rxrpc_ack_summary summary = { 0 };
292 unsigned long now, timo;
293 rxrpc_seq_t top = READ_ONCE(call->tx_top);
295 if (call->ackr_reason) {
296 spin_lock_bh(&call->lock);
297 call->ackr_reason = 0;
298 spin_unlock_bh(&call->lock);
300 timo = now + MAX_JIFFY_OFFSET;
301 WRITE_ONCE(call->resend_at, timo);
302 WRITE_ONCE(call->ack_at, timo);
303 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
306 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
307 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
308 rxrpc_proto_abort("TXL", call, top);
312 return rxrpc_end_tx_phase(call, true, "ETD");
316 * Scan a data packet to validate its structure and to work out how many
317 * subpackets it contains.
319 * A jumbo packet is a collection of consecutive packets glued together with
320 * little headers between that indicate how to change the initial header for
323 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
324 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
327 static bool rxrpc_validate_data(struct sk_buff *skb)
329 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
330 unsigned int offset = sizeof(struct rxrpc_wire_header);
331 unsigned int len = skb->len;
332 u8 flags = sp->hdr.flags;
335 if (flags & RXRPC_REQUEST_ACK)
336 __set_bit(sp->nr_subpackets, sp->rx_req_ack);
339 if (!(flags & RXRPC_JUMBO_PACKET))
342 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
344 if (flags & RXRPC_LAST_PACKET)
346 offset += RXRPC_JUMBO_DATALEN;
347 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
349 offset += sizeof(struct rxrpc_jumbo_header);
352 if (flags & RXRPC_LAST_PACKET)
353 sp->rx_flags |= RXRPC_SKB_INCL_LAST;
361 * Handle reception of a duplicate packet.
363 * We have to take care to avoid an attack here whereby we're given a series of
364 * jumbograms, each with a sequence number one before the preceding one and
365 * filled up to maximum UDP size. If they never send us the first packet in
366 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
367 * space until the call times out.
369 * We limit the space usage by only accepting three duplicate jumbo packets per
370 * call. After that, we tell the other side we're no longer accepting jumbos
371 * (that information is encoded in the ACK packet).
373 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
374 bool is_jumbo, bool *_jumbo_bad)
376 /* Discard normal packets that are duplicates. */
380 /* Skip jumbo subpackets that are duplicates. When we've had three or
381 * more partially duplicate jumbo packets, we refuse to take any more
382 * jumbos for this call.
385 call->nr_jumbo_bad++;
391 * Process a DATA packet, adding the packet to the Rx ring. The caller's
392 * packet ref must be passed on or discarded.
394 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
396 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
397 enum rxrpc_call_state state;
398 unsigned int j, nr_subpackets, nr_unacked = 0;
399 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = serial;
400 rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
401 bool immediate_ack = false, jumbo_bad = false;
404 _enter("{%u,%u},{%u,%u}",
405 call->rx_hard_ack, call->rx_top, skb->len, seq0);
407 _proto("Rx DATA %%%u { #%u f=%02x n=%u }",
408 sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
410 state = READ_ONCE(call->state);
411 if (state >= RXRPC_CALL_COMPLETE) {
412 rxrpc_free_skb(skb, rxrpc_skb_freed);
416 if (state == RXRPC_CALL_SERVER_RECV_REQUEST) {
417 unsigned long timo = READ_ONCE(call->next_req_timo);
418 unsigned long now, expect_req_by;
422 expect_req_by = now + timo;
423 WRITE_ONCE(call->expect_req_by, expect_req_by);
424 rxrpc_reduce_call_timer(call, expect_req_by, now,
425 rxrpc_timer_set_for_idle);
429 rxrpc_inc_stat(call->rxnet, stat_rx_data);
430 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
431 rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
432 if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
433 rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
435 spin_lock(&call->input_lock);
437 /* Received data implicitly ACKs all of the request packets we sent
438 * when we're acting as a client.
440 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
441 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
442 !rxrpc_receiving_reply(call))
445 hard_ack = READ_ONCE(call->rx_hard_ack);
447 nr_subpackets = sp->nr_subpackets;
448 if (nr_subpackets > 1) {
449 if (call->nr_jumbo_bad > 3) {
450 ack = RXRPC_ACK_NOSPACE;
456 for (j = 0; j < nr_subpackets; j++) {
457 rxrpc_serial_t serial = sp->hdr.serial + j;
458 rxrpc_seq_t seq = seq0 + j;
459 unsigned int ix = seq & RXRPC_RXTX_BUFF_MASK;
460 bool terminal = (j == nr_subpackets - 1);
461 bool last = terminal && (sp->rx_flags & RXRPC_SKB_INCL_LAST);
462 u8 flags, annotation = j;
464 _proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
465 j, serial, seq, terminal, last);
468 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
469 seq != call->rx_top) {
470 rxrpc_proto_abort("LSN", call, seq);
474 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
475 after_eq(seq, call->rx_top)) {
476 rxrpc_proto_abort("LSA", call, seq);
483 flags |= RXRPC_LAST_PACKET;
485 flags |= RXRPC_JUMBO_PACKET;
486 if (test_bit(j, sp->rx_req_ack))
487 flags |= RXRPC_REQUEST_ACK;
488 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
490 if (before_eq(seq, hard_ack)) {
491 ack = RXRPC_ACK_DUPLICATE;
496 if (call->rxtx_buffer[ix]) {
497 rxrpc_input_dup_data(call, seq, nr_subpackets > 1,
499 if (ack != RXRPC_ACK_DUPLICATE) {
500 ack = RXRPC_ACK_DUPLICATE;
503 immediate_ack = true;
507 if (after(seq, hard_ack + call->rx_winsize)) {
508 ack = RXRPC_ACK_EXCEEDS_WINDOW;
510 if (flags & RXRPC_JUMBO_PACKET) {
512 call->nr_jumbo_bad++;
520 if (flags & RXRPC_REQUEST_ACK && !ack) {
521 ack = RXRPC_ACK_REQUESTED;
525 if (after(seq0, call->ackr_highest_seq))
526 call->ackr_highest_seq = seq0;
528 /* Queue the packet. We use a couple of memory barriers here as need
529 * to make sure that rx_top is perceived to be set after the buffer
530 * pointer and that the buffer pointer is set after the annotation and
533 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
534 * and also rxrpc_fill_out_ack().
537 rxrpc_get_skb(skb, rxrpc_skb_got);
538 call->rxtx_annotations[ix] = annotation;
540 call->rxtx_buffer[ix] = skb;
541 if (after(seq, call->rx_top)) {
542 smp_store_release(&call->rx_top, seq);
543 } else if (before(seq, call->rx_top)) {
544 /* Send an immediate ACK if we fill in a hole */
546 ack = RXRPC_ACK_DELAY;
549 immediate_ack = true;
553 /* From this point on, we're not allowed to touch the
554 * packet any longer as its ref now belongs to the Rx
564 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
566 ack = RXRPC_ACK_DELAY;
569 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
571 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
574 if (after_eq(seq, call->rx_expect_next)) {
575 if (after(seq, call->rx_expect_next)) {
576 _net("OOS %u > %u", seq, call->rx_expect_next);
577 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
580 call->rx_expect_next = seq + 1;
587 if (atomic_add_return(nr_unacked, &call->ackr_nr_unacked) > 2 && !ack)
588 ack = RXRPC_ACK_IDLE;
591 rxrpc_propose_ACK(call, ack, ack_serial,
593 rxrpc_propose_ack_input_data);
595 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
597 rxrpc_propose_ack_input_data);
599 trace_rxrpc_notify_socket(call->debug_id, serial);
600 rxrpc_notify_socket(call);
603 spin_unlock(&call->input_lock);
604 rxrpc_free_skb(skb, rxrpc_skb_freed);
609 * See if there's a cached RTT probe to complete.
611 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
613 rxrpc_serial_t acked_serial,
614 rxrpc_serial_t ack_serial,
615 enum rxrpc_rtt_rx_trace type)
617 rxrpc_serial_t orig_serial;
620 bool matched = false;
623 avail = READ_ONCE(call->rtt_avail);
624 smp_rmb(); /* Read avail bits before accessing data. */
626 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
627 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
630 sent_at = call->rtt_sent_at[i];
631 orig_serial = call->rtt_serial[i];
633 if (orig_serial == acked_serial) {
634 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
635 smp_mb(); /* Read data before setting avail bit */
636 set_bit(i, &call->rtt_avail);
637 if (type != rxrpc_rtt_rx_cancel)
638 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
641 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
642 orig_serial, acked_serial, 0, 0);
646 /* If a later serial is being acked, then mark this slot as
649 if (after(acked_serial, orig_serial)) {
650 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
651 orig_serial, acked_serial, 0, 0);
652 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
654 set_bit(i, &call->rtt_avail);
659 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
663 * Process the response to a ping that we sent to find out if we lost an ACK.
665 * If we got back a ping response that indicates a lower tx_top than what we
666 * had at the time of the ping transmission, we adjudge all the DATA packets
667 * sent between the response tx_top and the ping-time tx_top to have been lost.
669 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
671 rxrpc_seq_t top, bottom, seq;
674 spin_lock_bh(&call->lock);
676 bottom = call->tx_hard_ack + 1;
677 top = call->acks_lost_top;
678 if (before(bottom, top)) {
679 for (seq = bottom; before_eq(seq, top); seq++) {
680 int ix = seq & RXRPC_RXTX_BUFF_MASK;
681 u8 annotation = call->rxtx_annotations[ix];
682 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
684 if (anno_type != RXRPC_TX_ANNO_UNACK)
686 annotation &= ~RXRPC_TX_ANNO_MASK;
687 annotation |= RXRPC_TX_ANNO_RETRANS;
688 call->rxtx_annotations[ix] = annotation;
693 spin_unlock_bh(&call->lock);
695 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
696 rxrpc_queue_call(call);
700 * Process a ping response.
702 static void rxrpc_input_ping_response(struct rxrpc_call *call,
704 rxrpc_serial_t acked_serial,
705 rxrpc_serial_t ack_serial)
707 if (acked_serial == call->acks_lost_ping)
708 rxrpc_input_check_for_lost_ack(call);
712 * Process the extra information that may be appended to an ACK packet
714 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
715 struct rxrpc_ackinfo *ackinfo)
717 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
718 struct rxrpc_peer *peer;
721 u32 rwind = ntohl(ackinfo->rwind);
723 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
725 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
726 rwind, ntohl(ackinfo->jumbo_max));
728 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
729 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
730 if (call->tx_winsize != rwind) {
731 if (rwind > call->tx_winsize)
733 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
734 call->tx_winsize = rwind;
737 if (call->cong_ssthresh > rwind)
738 call->cong_ssthresh = rwind;
740 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
743 if (mtu < peer->maxdata) {
744 spin_lock_bh(&peer->lock);
746 peer->mtu = mtu + peer->hdrsize;
747 spin_unlock_bh(&peer->lock);
748 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
752 wake_up(&call->waitq);
756 * Process individual soft ACKs.
758 * Each ACK in the array corresponds to one packet and can be either an ACK or
759 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
760 * packets that lie beyond the end of the ACK list are scheduled for resend by
761 * the timer on the basis that the peer might just not have processed them at
762 * the time the ACK was sent.
764 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
765 rxrpc_seq_t seq, int nr_acks,
766 struct rxrpc_ack_summary *summary)
769 u8 annotation, anno_type;
771 for (; nr_acks > 0; nr_acks--, seq++) {
772 ix = seq & RXRPC_RXTX_BUFF_MASK;
773 annotation = call->rxtx_annotations[ix];
774 anno_type = annotation & RXRPC_TX_ANNO_MASK;
775 annotation &= ~RXRPC_TX_ANNO_MASK;
777 case RXRPC_ACK_TYPE_ACK:
779 if (anno_type == RXRPC_TX_ANNO_ACK)
781 summary->nr_new_acks++;
782 call->rxtx_annotations[ix] =
783 RXRPC_TX_ANNO_ACK | annotation;
785 case RXRPC_ACK_TYPE_NACK:
786 if (!summary->nr_nacks &&
787 call->acks_lowest_nak != seq) {
788 call->acks_lowest_nak = seq;
789 summary->new_low_nack = true;
792 if (anno_type == RXRPC_TX_ANNO_NAK)
794 summary->nr_new_nacks++;
795 if (anno_type == RXRPC_TX_ANNO_RETRANS)
797 call->rxtx_annotations[ix] =
798 RXRPC_TX_ANNO_NAK | annotation;
801 return rxrpc_proto_abort("SFT", call, 0);
807 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
808 * with respect to the ack state conveyed by preceding ACKs.
810 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
811 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
813 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
815 if (after(first_pkt, base))
816 return true; /* The window advanced */
818 if (before(first_pkt, base))
819 return false; /* firstPacket regressed */
821 if (after_eq(prev_pkt, call->acks_prev_seq))
822 return true; /* previousPacket hasn't regressed. */
824 /* Some rx implementations put a serial number in previousPacket. */
825 if (after_eq(prev_pkt, base + call->tx_winsize))
831 * Process an ACK packet.
833 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
834 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
836 * A hard-ACK means that a packet has been processed and may be discarded; a
837 * soft-ACK means that the packet may be discarded and retransmission
838 * requested. A phase is complete when all packets are hard-ACK'd.
840 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
842 struct rxrpc_ack_summary summary = { 0 };
843 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
845 struct rxrpc_ackpacket ack;
846 struct rxrpc_ackinfo info;
847 u8 acks[RXRPC_MAXACKS];
849 rxrpc_serial_t ack_serial, acked_serial;
850 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
851 int nr_acks, offset, ioffset;
855 offset = sizeof(struct rxrpc_wire_header);
856 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
857 _debug("extraction failure");
858 return rxrpc_proto_abort("XAK", call, 0);
860 offset += sizeof(buf.ack);
862 ack_serial = sp->hdr.serial;
863 acked_serial = ntohl(buf.ack.serial);
864 first_soft_ack = ntohl(buf.ack.firstPacket);
865 prev_pkt = ntohl(buf.ack.previousPacket);
866 hard_ack = first_soft_ack - 1;
867 nr_acks = buf.ack.nAcks;
868 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
869 buf.ack.reason : RXRPC_ACK__INVALID);
871 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
872 first_soft_ack, prev_pkt,
873 summary.ack_reason, nr_acks);
874 rxrpc_inc_stat(call->rxnet, stat_rx_acks[buf.ack.reason]);
876 switch (buf.ack.reason) {
877 case RXRPC_ACK_PING_RESPONSE:
878 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
880 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
881 rxrpc_rtt_rx_ping_response);
883 case RXRPC_ACK_REQUESTED:
884 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
885 rxrpc_rtt_rx_requested_ack);
888 if (acked_serial != 0)
889 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
890 rxrpc_rtt_rx_cancel);
894 if (buf.ack.reason == RXRPC_ACK_PING) {
895 _proto("Rx ACK %%%u PING Request", ack_serial);
896 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
897 ack_serial, true, true,
898 rxrpc_propose_ack_respond_to_ping);
899 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
900 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
901 ack_serial, true, true,
902 rxrpc_propose_ack_respond_to_ack);
905 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
906 * indicates that the client address changed due to NAT. The server
907 * lost the call because it switched to a different peer.
909 if (unlikely(buf.ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
910 first_soft_ack == 1 &&
912 rxrpc_is_client_call(call)) {
913 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
918 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
919 * indicate a change of address. However, we can retransmit the call
920 * if we still have it buffered to the beginning.
922 if (unlikely(buf.ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
923 first_soft_ack == 1 &&
925 call->tx_hard_ack == 0 &&
926 rxrpc_is_client_call(call)) {
927 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
932 /* Discard any out-of-order or duplicate ACKs (outside lock). */
933 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
934 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
935 first_soft_ack, call->acks_first_seq,
936 prev_pkt, call->acks_prev_seq);
941 ioffset = offset + nr_acks + 3;
942 if (skb->len >= ioffset + sizeof(buf.info) &&
943 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
944 return rxrpc_proto_abort("XAI", call, 0);
946 spin_lock(&call->input_lock);
948 /* Discard any out-of-order or duplicate ACKs (inside lock). */
949 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
950 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
951 first_soft_ack, call->acks_first_seq,
952 prev_pkt, call->acks_prev_seq);
955 call->acks_latest_ts = skb->tstamp;
957 call->acks_first_seq = first_soft_ack;
958 call->acks_prev_seq = prev_pkt;
960 if (buf.ack.reason != RXRPC_ACK_PING &&
961 after(acked_serial, call->acks_highest_serial))
962 call->acks_highest_serial = acked_serial;
964 /* Parse rwind and mtu sizes if provided. */
966 rxrpc_input_ackinfo(call, skb, &buf.info);
968 if (first_soft_ack == 0) {
969 rxrpc_proto_abort("AK0", call, 0);
973 /* Ignore ACKs unless we are or have just been transmitting. */
974 switch (READ_ONCE(call->state)) {
975 case RXRPC_CALL_CLIENT_SEND_REQUEST:
976 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
977 case RXRPC_CALL_SERVER_SEND_REPLY:
978 case RXRPC_CALL_SERVER_AWAIT_ACK:
984 if (before(hard_ack, call->tx_hard_ack) ||
985 after(hard_ack, call->tx_top)) {
986 rxrpc_proto_abort("AKW", call, 0);
989 if (nr_acks > call->tx_top - hard_ack) {
990 rxrpc_proto_abort("AKN", call, 0);
994 if (after(hard_ack, call->tx_hard_ack)) {
995 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
996 rxrpc_end_tx_phase(call, false, "ETA");
1002 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
1003 rxrpc_proto_abort("XSA", call, 0);
1006 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
1010 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
1011 RXRPC_TX_ANNO_LAST &&
1012 summary.nr_acks == call->tx_top - hard_ack &&
1013 rxrpc_is_client_call(call))
1014 rxrpc_propose_ACK(call, RXRPC_ACK_PING, ack_serial,
1016 rxrpc_propose_ack_ping_for_lost_reply);
1018 rxrpc_congestion_management(call, skb, &summary, acked_serial);
1020 spin_unlock(&call->input_lock);
1024 * Process an ACKALL packet.
1026 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1028 struct rxrpc_ack_summary summary = { 0 };
1029 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1031 _proto("Rx ACKALL %%%u", sp->hdr.serial);
1033 spin_lock(&call->input_lock);
1035 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1036 rxrpc_end_tx_phase(call, false, "ETL");
1038 spin_unlock(&call->input_lock);
1042 * Process an ABORT packet directed at a call.
1044 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1046 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1048 u32 abort_code = RX_CALL_DEAD;
1052 if (skb->len >= 4 &&
1053 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
1054 &wtmp, sizeof(wtmp)) >= 0)
1055 abort_code = ntohl(wtmp);
1057 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1059 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1061 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1062 abort_code, -ECONNABORTED);
1066 * Process an incoming call packet.
1068 static void rxrpc_input_call_packet(struct rxrpc_call *call,
1069 struct sk_buff *skb)
1071 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1074 _enter("%p,%p", call, skb);
1076 timo = READ_ONCE(call->next_rx_timo);
1078 unsigned long now = jiffies, expect_rx_by;
1080 expect_rx_by = now + timo;
1081 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1082 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1083 rxrpc_timer_set_for_normal);
1086 switch (sp->hdr.type) {
1087 case RXRPC_PACKET_TYPE_DATA:
1088 rxrpc_input_data(call, skb);
1091 case RXRPC_PACKET_TYPE_ACK:
1092 rxrpc_input_ack(call, skb);
1095 case RXRPC_PACKET_TYPE_BUSY:
1096 _proto("Rx BUSY %%%u", sp->hdr.serial);
1098 /* Just ignore BUSY packets from the server; the retry and
1099 * lifespan timers will take care of business. BUSY packets
1100 * from the client don't make sense.
1104 case RXRPC_PACKET_TYPE_ABORT:
1105 rxrpc_input_abort(call, skb);
1108 case RXRPC_PACKET_TYPE_ACKALL:
1109 rxrpc_input_ackall(call, skb);
1116 rxrpc_free_skb(skb, rxrpc_skb_freed);
1122 * Handle a new service call on a channel implicitly completing the preceding
1123 * call on that channel. This does not apply to client conns.
1125 * TODO: If callNumber > call_id + 1, renegotiate security.
1127 static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1128 struct rxrpc_connection *conn,
1129 struct rxrpc_call *call)
1131 switch (READ_ONCE(call->state)) {
1132 case RXRPC_CALL_SERVER_AWAIT_ACK:
1133 rxrpc_call_completed(call);
1135 case RXRPC_CALL_COMPLETE:
1138 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1139 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1140 rxrpc_queue_call(call);
1142 trace_rxrpc_improper_term(call);
1146 spin_lock(&rx->incoming_lock);
1147 __rxrpc_disconnect_call(conn, call);
1148 spin_unlock(&rx->incoming_lock);
1152 * post connection-level events to the connection
1153 * - this includes challenges, responses, some aborts and call terminal packet
1156 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1157 struct sk_buff *skb)
1159 _enter("%p,%p", conn, skb);
1161 skb_queue_tail(&conn->rx_queue, skb);
1162 rxrpc_queue_conn(conn);
1166 * post endpoint-level events to the local endpoint
1167 * - this includes debug and version messages
1169 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1170 struct sk_buff *skb)
1172 _enter("%p,%p", local, skb);
1174 if (rxrpc_get_local_maybe(local)) {
1175 skb_queue_tail(&local->event_queue, skb);
1176 rxrpc_queue_local(local);
1178 rxrpc_free_skb(skb, rxrpc_skb_freed);
1183 * put a packet up for transport-level abort
1185 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1187 if (rxrpc_get_local_maybe(local)) {
1188 skb_queue_tail(&local->reject_queue, skb);
1189 rxrpc_queue_local(local);
1191 rxrpc_free_skb(skb, rxrpc_skb_freed);
1196 * Extract the wire header from a packet and translate the byte order.
1199 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1201 struct rxrpc_wire_header whdr;
1203 /* dig out the RxRPC connection details */
1204 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1205 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1206 tracepoint_string("bad_hdr"));
1210 memset(sp, 0, sizeof(*sp));
1211 sp->hdr.epoch = ntohl(whdr.epoch);
1212 sp->hdr.cid = ntohl(whdr.cid);
1213 sp->hdr.callNumber = ntohl(whdr.callNumber);
1214 sp->hdr.seq = ntohl(whdr.seq);
1215 sp->hdr.serial = ntohl(whdr.serial);
1216 sp->hdr.flags = whdr.flags;
1217 sp->hdr.type = whdr.type;
1218 sp->hdr.userStatus = whdr.userStatus;
1219 sp->hdr.securityIndex = whdr.securityIndex;
1220 sp->hdr._rsvd = ntohs(whdr._rsvd);
1221 sp->hdr.serviceId = ntohs(whdr.serviceId);
1226 * handle data received on the local endpoint
1227 * - may be called in interrupt context
1229 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1230 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1231 * being cleared in the middle of processing this function.
1233 * Called with the RCU read lock held from the IP layer via UDP.
1235 int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1237 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1238 struct rxrpc_connection *conn;
1239 struct rxrpc_channel *chan;
1240 struct rxrpc_call *call = NULL;
1241 struct rxrpc_skb_priv *sp;
1242 struct rxrpc_peer *peer = NULL;
1243 struct rxrpc_sock *rx = NULL;
1244 unsigned int channel;
1246 _enter("%p", udp_sk);
1248 if (unlikely(!local)) {
1252 if (skb->tstamp == 0)
1253 skb->tstamp = ktime_get_real();
1255 rxrpc_new_skb(skb, rxrpc_skb_received);
1257 skb_pull(skb, sizeof(struct udphdr));
1259 /* The UDP protocol already released all skb resources;
1260 * we are free to add our own data there.
1262 sp = rxrpc_skb(skb);
1264 /* dig out the RxRPC connection details */
1265 if (rxrpc_extract_header(sp, skb) < 0)
1268 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1270 if ((lose++ & 7) == 7) {
1271 trace_rxrpc_rx_lose(sp);
1272 rxrpc_free_skb(skb, rxrpc_skb_lost);
1277 if (skb->tstamp == 0)
1278 skb->tstamp = ktime_get_real();
1279 trace_rxrpc_rx_packet(sp);
1281 switch (sp->hdr.type) {
1282 case RXRPC_PACKET_TYPE_VERSION:
1283 if (rxrpc_to_client(sp))
1285 rxrpc_post_packet_to_local(local, skb);
1288 case RXRPC_PACKET_TYPE_BUSY:
1289 if (rxrpc_to_server(sp))
1292 case RXRPC_PACKET_TYPE_ACK:
1293 case RXRPC_PACKET_TYPE_ACKALL:
1294 if (sp->hdr.callNumber == 0)
1297 case RXRPC_PACKET_TYPE_ABORT:
1300 case RXRPC_PACKET_TYPE_DATA:
1301 if (sp->hdr.callNumber == 0 ||
1304 if (!rxrpc_validate_data(skb))
1307 /* Unshare the packet so that it can be modified for in-place
1310 if (sp->hdr.securityIndex != 0) {
1311 struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1313 rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1318 rxrpc_eaten_skb(skb, rxrpc_skb_received);
1320 rxrpc_new_skb(skb, rxrpc_skb_unshared);
1321 sp = rxrpc_skb(skb);
1326 case RXRPC_PACKET_TYPE_CHALLENGE:
1327 if (rxrpc_to_server(sp))
1330 case RXRPC_PACKET_TYPE_RESPONSE:
1331 if (rxrpc_to_client(sp))
1335 /* Packet types 9-11 should just be ignored. */
1336 case RXRPC_PACKET_TYPE_PARAMS:
1337 case RXRPC_PACKET_TYPE_10:
1338 case RXRPC_PACKET_TYPE_11:
1342 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1346 if (sp->hdr.serviceId == 0)
1349 if (rxrpc_to_server(sp)) {
1350 /* Weed out packets to services we're not offering. Packets
1351 * that would begin a call are explicitly rejected and the rest
1352 * are just discarded.
1354 rx = rcu_dereference(local->service);
1355 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1356 sp->hdr.serviceId != rx->second_service)) {
1357 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1359 goto unsupported_service;
1364 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1366 if (sp->hdr.securityIndex != conn->security_ix)
1367 goto wrong_security;
1369 if (sp->hdr.serviceId != conn->service_id) {
1372 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1374 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1377 if (old_id != conn->params.service_id &&
1378 old_id != sp->hdr.serviceId)
1382 if (sp->hdr.callNumber == 0) {
1383 /* Connection-level packet */
1384 _debug("CONN %p {%d}", conn, conn->debug_id);
1385 rxrpc_post_packet_to_conn(conn, skb);
1389 if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1390 conn->hi_serial = sp->hdr.serial;
1392 /* Call-bound packets are routed by connection channel. */
1393 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1394 chan = &conn->channels[channel];
1396 /* Ignore really old calls */
1397 if (sp->hdr.callNumber < chan->last_call)
1400 if (sp->hdr.callNumber == chan->last_call) {
1402 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1405 /* For the previous service call, if completed
1406 * successfully, we discard all further packets.
1408 if (rxrpc_conn_is_service(conn) &&
1409 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1412 /* But otherwise we need to retransmit the final packet
1413 * from data cached in the connection record.
1415 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1416 trace_rxrpc_rx_data(chan->call_debug_id,
1420 rxrpc_post_packet_to_conn(conn, skb);
1424 call = rcu_dereference(chan->call);
1426 if (sp->hdr.callNumber > chan->call_id) {
1427 if (rxrpc_to_client(sp))
1430 rxrpc_input_implicit_end_call(rx, conn, call);
1435 if (sp->hdr.serviceId != call->service_id)
1436 call->service_id = sp->hdr.serviceId;
1437 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1438 call->rx_serial = sp->hdr.serial;
1439 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1440 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1444 if (!call || refcount_read(&call->ref) == 0) {
1445 if (rxrpc_to_client(sp) ||
1446 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1448 if (sp->hdr.seq != 1)
1450 call = rxrpc_new_incoming_call(local, rx, skb);
1455 /* Process a call packet; this either discards or passes on the ref
1458 rxrpc_input_call_packet(call, skb);
1462 rxrpc_free_skb(skb, rxrpc_skb_freed);
1464 trace_rxrpc_rx_done(0, 0);
1468 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1469 RXKADINCONSISTENCY, EBADMSG);
1470 skb->priority = RXKADINCONSISTENCY;
1473 unsupported_service:
1474 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1475 RX_INVALID_OPERATION, EOPNOTSUPP);
1476 skb->priority = RX_INVALID_OPERATION;
1480 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1481 RX_PROTOCOL_ERROR, EBADMSG);
1482 goto protocol_error;
1485 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1486 RX_PROTOCOL_ERROR, EBADMSG);
1488 skb->priority = RX_PROTOCOL_ERROR;
1490 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1492 trace_rxrpc_rx_done(skb->mark, skb->priority);
1493 rxrpc_reject_packet(local, skb);
1494 _leave(" [badmsg]");