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_count == 0)
96 if (ktime_before(skb->tstamp,
97 ktime_add_us(call->cong_tstamp,
98 call->peer->srtt_us >> 3)))
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 * Apply a hard ACK by advancing the Tx window.
198 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
199 struct rxrpc_ack_summary *summary)
201 struct sk_buff *skb, *list = NULL;
202 bool rot_last = false;
206 if (call->acks_lowest_nak == call->tx_hard_ack) {
207 call->acks_lowest_nak = to;
208 } else if (before_eq(call->acks_lowest_nak, to)) {
209 summary->new_low_nack = true;
210 call->acks_lowest_nak = to;
213 spin_lock(&call->lock);
215 while (before(call->tx_hard_ack, to)) {
217 ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
218 skb = call->rxtx_buffer[ix];
219 annotation = call->rxtx_annotations[ix];
220 rxrpc_see_skb(skb, rxrpc_skb_rotated);
221 call->rxtx_buffer[ix] = NULL;
222 call->rxtx_annotations[ix] = 0;
226 if (annotation & RXRPC_TX_ANNO_LAST) {
227 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
230 if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
231 summary->nr_rot_new_acks++;
234 spin_unlock(&call->lock);
236 trace_rxrpc_transmit(call, (rot_last ?
237 rxrpc_transmit_rotate_last :
238 rxrpc_transmit_rotate));
239 wake_up(&call->waitq);
244 skb_mark_not_on_list(skb);
245 rxrpc_free_skb(skb, rxrpc_skb_freed);
252 * End the transmission phase of a call.
254 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
255 * or a final ACK packet.
257 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
258 const char *abort_why)
262 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
264 write_lock(&call->state_lock);
268 case RXRPC_CALL_CLIENT_SEND_REQUEST:
269 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
271 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
273 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
276 case RXRPC_CALL_SERVER_AWAIT_ACK:
277 __rxrpc_call_completed(call);
285 write_unlock(&call->state_lock);
286 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
287 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
289 trace_rxrpc_transmit(call, rxrpc_transmit_end);
294 write_unlock(&call->state_lock);
295 kdebug("end_tx %s", rxrpc_call_states[call->state]);
296 rxrpc_proto_abort(abort_why, call, call->tx_top);
301 * Begin the reply reception phase of a call.
303 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
305 struct rxrpc_ack_summary summary = { 0 };
306 unsigned long now, timo;
307 rxrpc_seq_t top = READ_ONCE(call->tx_top);
309 if (call->ackr_reason) {
310 spin_lock_bh(&call->lock);
311 call->ackr_reason = 0;
312 spin_unlock_bh(&call->lock);
314 timo = now + MAX_JIFFY_OFFSET;
315 WRITE_ONCE(call->resend_at, timo);
316 WRITE_ONCE(call->ack_at, timo);
317 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
320 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
321 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
322 rxrpc_proto_abort("TXL", call, top);
326 if (!rxrpc_end_tx_phase(call, true, "ETD"))
328 call->tx_phase = false;
333 * Scan a data packet to validate its structure and to work out how many
334 * subpackets it contains.
336 * A jumbo packet is a collection of consecutive packets glued together with
337 * little headers between that indicate how to change the initial header for
340 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
341 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
344 static bool rxrpc_validate_data(struct sk_buff *skb)
346 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
347 unsigned int offset = sizeof(struct rxrpc_wire_header);
348 unsigned int len = skb->len;
349 u8 flags = sp->hdr.flags;
352 if (flags & RXRPC_REQUEST_ACK)
353 __set_bit(sp->nr_subpackets, sp->rx_req_ack);
356 if (!(flags & RXRPC_JUMBO_PACKET))
359 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
361 if (flags & RXRPC_LAST_PACKET)
363 offset += RXRPC_JUMBO_DATALEN;
364 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
366 offset += sizeof(struct rxrpc_jumbo_header);
369 if (flags & RXRPC_LAST_PACKET)
370 sp->rx_flags |= RXRPC_SKB_INCL_LAST;
378 * Handle reception of a duplicate packet.
380 * We have to take care to avoid an attack here whereby we're given a series of
381 * jumbograms, each with a sequence number one before the preceding one and
382 * filled up to maximum UDP size. If they never send us the first packet in
383 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
384 * space until the call times out.
386 * We limit the space usage by only accepting three duplicate jumbo packets per
387 * call. After that, we tell the other side we're no longer accepting jumbos
388 * (that information is encoded in the ACK packet).
390 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
391 bool is_jumbo, bool *_jumbo_bad)
393 /* Discard normal packets that are duplicates. */
397 /* Skip jumbo subpackets that are duplicates. When we've had three or
398 * more partially duplicate jumbo packets, we refuse to take any more
399 * jumbos for this call.
402 call->nr_jumbo_bad++;
408 * Process a DATA packet, adding the packet to the Rx ring. The caller's
409 * packet ref must be passed on or discarded.
411 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
413 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
414 enum rxrpc_call_state state;
415 unsigned int j, nr_subpackets, nr_unacked = 0;
416 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = serial;
417 rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
418 bool immediate_ack = false, jumbo_bad = false;
421 _enter("{%u,%u},{%u,%u}",
422 call->rx_hard_ack, call->rx_top, skb->len, seq0);
424 _proto("Rx DATA %%%u { #%u f=%02x n=%u }",
425 sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
427 state = READ_ONCE(call->state);
428 if (state >= RXRPC_CALL_COMPLETE) {
429 rxrpc_free_skb(skb, rxrpc_skb_freed);
433 if (state == RXRPC_CALL_SERVER_RECV_REQUEST) {
434 unsigned long timo = READ_ONCE(call->next_req_timo);
435 unsigned long now, expect_req_by;
439 expect_req_by = now + timo;
440 WRITE_ONCE(call->expect_req_by, expect_req_by);
441 rxrpc_reduce_call_timer(call, expect_req_by, now,
442 rxrpc_timer_set_for_idle);
446 spin_lock(&call->input_lock);
448 /* Received data implicitly ACKs all of the request packets we sent
449 * when we're acting as a client.
451 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
452 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
453 !rxrpc_receiving_reply(call))
456 hard_ack = READ_ONCE(call->rx_hard_ack);
458 nr_subpackets = sp->nr_subpackets;
459 if (nr_subpackets > 1) {
460 if (call->nr_jumbo_bad > 3) {
461 ack = RXRPC_ACK_NOSPACE;
467 for (j = 0; j < nr_subpackets; j++) {
468 rxrpc_serial_t serial = sp->hdr.serial + j;
469 rxrpc_seq_t seq = seq0 + j;
470 unsigned int ix = seq & RXRPC_RXTX_BUFF_MASK;
471 bool terminal = (j == nr_subpackets - 1);
472 bool last = terminal && (sp->rx_flags & RXRPC_SKB_INCL_LAST);
473 u8 flags, annotation = j;
475 _proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
476 j, serial, seq, terminal, last);
479 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
480 seq != call->rx_top) {
481 rxrpc_proto_abort("LSN", call, seq);
485 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
486 after_eq(seq, call->rx_top)) {
487 rxrpc_proto_abort("LSA", call, seq);
494 flags |= RXRPC_LAST_PACKET;
496 flags |= RXRPC_JUMBO_PACKET;
497 if (test_bit(j, sp->rx_req_ack))
498 flags |= RXRPC_REQUEST_ACK;
499 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
501 if (before_eq(seq, hard_ack)) {
502 ack = RXRPC_ACK_DUPLICATE;
507 if (call->rxtx_buffer[ix]) {
508 rxrpc_input_dup_data(call, seq, nr_subpackets > 1,
510 if (ack != RXRPC_ACK_DUPLICATE) {
511 ack = RXRPC_ACK_DUPLICATE;
514 immediate_ack = true;
518 if (after(seq, hard_ack + call->rx_winsize)) {
519 ack = RXRPC_ACK_EXCEEDS_WINDOW;
521 if (flags & RXRPC_JUMBO_PACKET) {
523 call->nr_jumbo_bad++;
531 if (flags & RXRPC_REQUEST_ACK && !ack) {
532 ack = RXRPC_ACK_REQUESTED;
536 if (after(seq0, call->ackr_highest_seq))
537 call->ackr_highest_seq = seq0;
539 /* Queue the packet. We use a couple of memory barriers here as need
540 * to make sure that rx_top is perceived to be set after the buffer
541 * pointer and that the buffer pointer is set after the annotation and
544 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
545 * and also rxrpc_fill_out_ack().
548 rxrpc_get_skb(skb, rxrpc_skb_got);
549 call->rxtx_annotations[ix] = annotation;
551 call->rxtx_buffer[ix] = skb;
552 if (after(seq, call->rx_top)) {
553 smp_store_release(&call->rx_top, seq);
554 } else if (before(seq, call->rx_top)) {
555 /* Send an immediate ACK if we fill in a hole */
557 ack = RXRPC_ACK_DELAY;
560 immediate_ack = true;
564 /* From this point on, we're not allowed to touch the
565 * packet any longer as its ref now belongs to the Rx
575 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
577 ack = RXRPC_ACK_DELAY;
580 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
582 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
585 if (after_eq(seq, call->rx_expect_next)) {
586 if (after(seq, call->rx_expect_next)) {
587 _net("OOS %u > %u", seq, call->rx_expect_next);
588 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
591 call->rx_expect_next = seq + 1;
598 if (atomic_add_return(nr_unacked, &call->ackr_nr_unacked) > 2 && !ack)
599 ack = RXRPC_ACK_IDLE;
602 rxrpc_propose_ACK(call, ack, ack_serial,
604 rxrpc_propose_ack_input_data);
606 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
608 rxrpc_propose_ack_input_data);
610 trace_rxrpc_notify_socket(call->debug_id, serial);
611 rxrpc_notify_socket(call);
614 spin_unlock(&call->input_lock);
615 rxrpc_free_skb(skb, rxrpc_skb_freed);
620 * See if there's a cached RTT probe to complete.
622 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
624 rxrpc_serial_t acked_serial,
625 rxrpc_serial_t ack_serial,
626 enum rxrpc_rtt_rx_trace type)
628 rxrpc_serial_t orig_serial;
631 bool matched = false;
634 avail = READ_ONCE(call->rtt_avail);
635 smp_rmb(); /* Read avail bits before accessing data. */
637 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
638 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
641 sent_at = call->rtt_sent_at[i];
642 orig_serial = call->rtt_serial[i];
644 if (orig_serial == acked_serial) {
645 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
646 smp_mb(); /* Read data before setting avail bit */
647 set_bit(i, &call->rtt_avail);
648 if (type != rxrpc_rtt_rx_cancel)
649 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
652 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
653 orig_serial, acked_serial, 0, 0);
657 /* If a later serial is being acked, then mark this slot as
660 if (after(acked_serial, orig_serial)) {
661 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
662 orig_serial, acked_serial, 0, 0);
663 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
665 set_bit(i, &call->rtt_avail);
670 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
674 * Process the response to a ping that we sent to find out if we lost an ACK.
676 * If we got back a ping response that indicates a lower tx_top than what we
677 * had at the time of the ping transmission, we adjudge all the DATA packets
678 * sent between the response tx_top and the ping-time tx_top to have been lost.
680 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
682 rxrpc_seq_t top, bottom, seq;
685 spin_lock_bh(&call->lock);
687 bottom = call->tx_hard_ack + 1;
688 top = call->acks_lost_top;
689 if (before(bottom, top)) {
690 for (seq = bottom; before_eq(seq, top); seq++) {
691 int ix = seq & RXRPC_RXTX_BUFF_MASK;
692 u8 annotation = call->rxtx_annotations[ix];
693 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
695 if (anno_type != RXRPC_TX_ANNO_UNACK)
697 annotation &= ~RXRPC_TX_ANNO_MASK;
698 annotation |= RXRPC_TX_ANNO_RETRANS;
699 call->rxtx_annotations[ix] = annotation;
704 spin_unlock_bh(&call->lock);
706 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
707 rxrpc_queue_call(call);
711 * Process a ping response.
713 static void rxrpc_input_ping_response(struct rxrpc_call *call,
715 rxrpc_serial_t acked_serial,
716 rxrpc_serial_t ack_serial)
718 if (acked_serial == call->acks_lost_ping)
719 rxrpc_input_check_for_lost_ack(call);
723 * Process the extra information that may be appended to an ACK packet
725 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
726 struct rxrpc_ackinfo *ackinfo)
728 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
729 struct rxrpc_peer *peer;
732 u32 rwind = ntohl(ackinfo->rwind);
734 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
736 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
737 rwind, ntohl(ackinfo->jumbo_max));
739 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
740 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
741 if (call->tx_winsize != rwind) {
742 if (rwind > call->tx_winsize)
744 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
745 call->tx_winsize = rwind;
748 if (call->cong_ssthresh > rwind)
749 call->cong_ssthresh = rwind;
751 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
754 if (mtu < peer->maxdata) {
755 spin_lock_bh(&peer->lock);
757 peer->mtu = mtu + peer->hdrsize;
758 spin_unlock_bh(&peer->lock);
759 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
763 wake_up(&call->waitq);
767 * Process individual soft ACKs.
769 * Each ACK in the array corresponds to one packet and can be either an ACK or
770 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
771 * packets that lie beyond the end of the ACK list are scheduled for resend by
772 * the timer on the basis that the peer might just not have processed them at
773 * the time the ACK was sent.
775 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
776 rxrpc_seq_t seq, int nr_acks,
777 struct rxrpc_ack_summary *summary)
780 u8 annotation, anno_type;
782 for (; nr_acks > 0; nr_acks--, seq++) {
783 ix = seq & RXRPC_RXTX_BUFF_MASK;
784 annotation = call->rxtx_annotations[ix];
785 anno_type = annotation & RXRPC_TX_ANNO_MASK;
786 annotation &= ~RXRPC_TX_ANNO_MASK;
788 case RXRPC_ACK_TYPE_ACK:
790 if (anno_type == RXRPC_TX_ANNO_ACK)
792 summary->nr_new_acks++;
793 call->rxtx_annotations[ix] =
794 RXRPC_TX_ANNO_ACK | annotation;
796 case RXRPC_ACK_TYPE_NACK:
797 if (!summary->nr_nacks &&
798 call->acks_lowest_nak != seq) {
799 call->acks_lowest_nak = seq;
800 summary->new_low_nack = true;
803 if (anno_type == RXRPC_TX_ANNO_NAK)
805 summary->nr_new_nacks++;
806 if (anno_type == RXRPC_TX_ANNO_RETRANS)
808 call->rxtx_annotations[ix] =
809 RXRPC_TX_ANNO_NAK | annotation;
812 return rxrpc_proto_abort("SFT", call, 0);
818 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
819 * with respect to the ack state conveyed by preceding ACKs.
821 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
822 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
824 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
826 if (after(first_pkt, base))
827 return true; /* The window advanced */
829 if (before(first_pkt, base))
830 return false; /* firstPacket regressed */
832 if (after_eq(prev_pkt, call->acks_prev_seq))
833 return true; /* previousPacket hasn't regressed. */
835 /* Some rx implementations put a serial number in previousPacket. */
836 if (after_eq(prev_pkt, base + call->tx_winsize))
842 * Process an ACK packet.
844 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
845 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
847 * A hard-ACK means that a packet has been processed and may be discarded; a
848 * soft-ACK means that the packet may be discarded and retransmission
849 * requested. A phase is complete when all packets are hard-ACK'd.
851 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
853 struct rxrpc_ack_summary summary = { 0 };
854 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
856 struct rxrpc_ackpacket ack;
857 struct rxrpc_ackinfo info;
858 u8 acks[RXRPC_MAXACKS];
860 rxrpc_serial_t ack_serial, acked_serial;
861 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
862 int nr_acks, offset, ioffset;
866 offset = sizeof(struct rxrpc_wire_header);
867 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
868 _debug("extraction failure");
869 return rxrpc_proto_abort("XAK", call, 0);
871 offset += sizeof(buf.ack);
873 ack_serial = sp->hdr.serial;
874 acked_serial = ntohl(buf.ack.serial);
875 first_soft_ack = ntohl(buf.ack.firstPacket);
876 prev_pkt = ntohl(buf.ack.previousPacket);
877 hard_ack = first_soft_ack - 1;
878 nr_acks = buf.ack.nAcks;
879 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
880 buf.ack.reason : RXRPC_ACK__INVALID);
882 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
883 first_soft_ack, prev_pkt,
884 summary.ack_reason, nr_acks);
886 switch (buf.ack.reason) {
887 case RXRPC_ACK_PING_RESPONSE:
888 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
890 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
891 rxrpc_rtt_rx_ping_response);
893 case RXRPC_ACK_REQUESTED:
894 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
895 rxrpc_rtt_rx_requested_ack);
898 if (acked_serial != 0)
899 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
900 rxrpc_rtt_rx_cancel);
904 if (buf.ack.reason == RXRPC_ACK_PING) {
905 _proto("Rx ACK %%%u PING Request", ack_serial);
906 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
907 ack_serial, true, true,
908 rxrpc_propose_ack_respond_to_ping);
909 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
910 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
911 ack_serial, true, true,
912 rxrpc_propose_ack_respond_to_ack);
915 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
916 * indicates that the client address changed due to NAT. The server
917 * lost the call because it switched to a different peer.
919 if (unlikely(buf.ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
920 first_soft_ack == 1 &&
922 rxrpc_is_client_call(call)) {
923 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
928 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
929 * indicate a change of address. However, we can retransmit the call
930 * if we still have it buffered to the beginning.
932 if (unlikely(buf.ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
933 first_soft_ack == 1 &&
935 call->tx_hard_ack == 0 &&
936 rxrpc_is_client_call(call)) {
937 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
942 /* Discard any out-of-order or duplicate ACKs (outside lock). */
943 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
944 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
945 first_soft_ack, call->acks_first_seq,
946 prev_pkt, call->acks_prev_seq);
951 ioffset = offset + nr_acks + 3;
952 if (skb->len >= ioffset + sizeof(buf.info) &&
953 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
954 return rxrpc_proto_abort("XAI", call, 0);
956 spin_lock(&call->input_lock);
958 /* Discard any out-of-order or duplicate ACKs (inside lock). */
959 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
960 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
961 first_soft_ack, call->acks_first_seq,
962 prev_pkt, call->acks_prev_seq);
965 call->acks_latest_ts = skb->tstamp;
967 call->acks_first_seq = first_soft_ack;
968 call->acks_prev_seq = prev_pkt;
970 /* Parse rwind and mtu sizes if provided. */
972 rxrpc_input_ackinfo(call, skb, &buf.info);
974 if (first_soft_ack == 0) {
975 rxrpc_proto_abort("AK0", call, 0);
979 /* Ignore ACKs unless we are or have just been transmitting. */
980 switch (READ_ONCE(call->state)) {
981 case RXRPC_CALL_CLIENT_SEND_REQUEST:
982 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
983 case RXRPC_CALL_SERVER_SEND_REPLY:
984 case RXRPC_CALL_SERVER_AWAIT_ACK:
990 if (before(hard_ack, call->tx_hard_ack) ||
991 after(hard_ack, call->tx_top)) {
992 rxrpc_proto_abort("AKW", call, 0);
995 if (nr_acks > call->tx_top - hard_ack) {
996 rxrpc_proto_abort("AKN", call, 0);
1000 if (after(hard_ack, call->tx_hard_ack)) {
1001 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
1002 rxrpc_end_tx_phase(call, false, "ETA");
1008 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
1009 rxrpc_proto_abort("XSA", call, 0);
1012 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
1016 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
1017 RXRPC_TX_ANNO_LAST &&
1018 summary.nr_acks == call->tx_top - hard_ack &&
1019 rxrpc_is_client_call(call))
1020 rxrpc_propose_ACK(call, RXRPC_ACK_PING, ack_serial,
1022 rxrpc_propose_ack_ping_for_lost_reply);
1024 rxrpc_congestion_management(call, skb, &summary, acked_serial);
1026 spin_unlock(&call->input_lock);
1030 * Process an ACKALL packet.
1032 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1034 struct rxrpc_ack_summary summary = { 0 };
1035 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1037 _proto("Rx ACKALL %%%u", sp->hdr.serial);
1039 spin_lock(&call->input_lock);
1041 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1042 rxrpc_end_tx_phase(call, false, "ETL");
1044 spin_unlock(&call->input_lock);
1048 * Process an ABORT packet directed at a call.
1050 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1052 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1054 u32 abort_code = RX_CALL_DEAD;
1058 if (skb->len >= 4 &&
1059 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
1060 &wtmp, sizeof(wtmp)) >= 0)
1061 abort_code = ntohl(wtmp);
1063 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1065 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1067 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1068 abort_code, -ECONNABORTED);
1072 * Process an incoming call packet.
1074 static void rxrpc_input_call_packet(struct rxrpc_call *call,
1075 struct sk_buff *skb)
1077 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1080 _enter("%p,%p", call, skb);
1082 timo = READ_ONCE(call->next_rx_timo);
1084 unsigned long now = jiffies, expect_rx_by;
1086 expect_rx_by = now + timo;
1087 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1088 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1089 rxrpc_timer_set_for_normal);
1092 switch (sp->hdr.type) {
1093 case RXRPC_PACKET_TYPE_DATA:
1094 rxrpc_input_data(call, skb);
1097 case RXRPC_PACKET_TYPE_ACK:
1098 rxrpc_input_ack(call, skb);
1101 case RXRPC_PACKET_TYPE_BUSY:
1102 _proto("Rx BUSY %%%u", sp->hdr.serial);
1104 /* Just ignore BUSY packets from the server; the retry and
1105 * lifespan timers will take care of business. BUSY packets
1106 * from the client don't make sense.
1110 case RXRPC_PACKET_TYPE_ABORT:
1111 rxrpc_input_abort(call, skb);
1114 case RXRPC_PACKET_TYPE_ACKALL:
1115 rxrpc_input_ackall(call, skb);
1122 rxrpc_free_skb(skb, rxrpc_skb_freed);
1128 * Handle a new service call on a channel implicitly completing the preceding
1129 * call on that channel. This does not apply to client conns.
1131 * TODO: If callNumber > call_id + 1, renegotiate security.
1133 static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1134 struct rxrpc_connection *conn,
1135 struct rxrpc_call *call)
1137 switch (READ_ONCE(call->state)) {
1138 case RXRPC_CALL_SERVER_AWAIT_ACK:
1139 rxrpc_call_completed(call);
1141 case RXRPC_CALL_COMPLETE:
1144 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1145 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1146 rxrpc_queue_call(call);
1148 trace_rxrpc_improper_term(call);
1152 spin_lock(&rx->incoming_lock);
1153 __rxrpc_disconnect_call(conn, call);
1154 spin_unlock(&rx->incoming_lock);
1158 * post connection-level events to the connection
1159 * - this includes challenges, responses, some aborts and call terminal packet
1162 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1163 struct sk_buff *skb)
1165 _enter("%p,%p", conn, skb);
1167 skb_queue_tail(&conn->rx_queue, skb);
1168 rxrpc_queue_conn(conn);
1172 * post endpoint-level events to the local endpoint
1173 * - this includes debug and version messages
1175 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1176 struct sk_buff *skb)
1178 _enter("%p,%p", local, skb);
1180 if (rxrpc_get_local_maybe(local)) {
1181 skb_queue_tail(&local->event_queue, skb);
1182 rxrpc_queue_local(local);
1184 rxrpc_free_skb(skb, rxrpc_skb_freed);
1189 * put a packet up for transport-level abort
1191 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1193 if (rxrpc_get_local_maybe(local)) {
1194 skb_queue_tail(&local->reject_queue, skb);
1195 rxrpc_queue_local(local);
1197 rxrpc_free_skb(skb, rxrpc_skb_freed);
1202 * Extract the wire header from a packet and translate the byte order.
1205 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1207 struct rxrpc_wire_header whdr;
1209 /* dig out the RxRPC connection details */
1210 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1211 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1212 tracepoint_string("bad_hdr"));
1216 memset(sp, 0, sizeof(*sp));
1217 sp->hdr.epoch = ntohl(whdr.epoch);
1218 sp->hdr.cid = ntohl(whdr.cid);
1219 sp->hdr.callNumber = ntohl(whdr.callNumber);
1220 sp->hdr.seq = ntohl(whdr.seq);
1221 sp->hdr.serial = ntohl(whdr.serial);
1222 sp->hdr.flags = whdr.flags;
1223 sp->hdr.type = whdr.type;
1224 sp->hdr.userStatus = whdr.userStatus;
1225 sp->hdr.securityIndex = whdr.securityIndex;
1226 sp->hdr._rsvd = ntohs(whdr._rsvd);
1227 sp->hdr.serviceId = ntohs(whdr.serviceId);
1232 * handle data received on the local endpoint
1233 * - may be called in interrupt context
1235 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1236 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1237 * being cleared in the middle of processing this function.
1239 * Called with the RCU read lock held from the IP layer via UDP.
1241 int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1243 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1244 struct rxrpc_connection *conn;
1245 struct rxrpc_channel *chan;
1246 struct rxrpc_call *call = NULL;
1247 struct rxrpc_skb_priv *sp;
1248 struct rxrpc_peer *peer = NULL;
1249 struct rxrpc_sock *rx = NULL;
1250 unsigned int channel;
1252 _enter("%p", udp_sk);
1254 if (unlikely(!local)) {
1258 if (skb->tstamp == 0)
1259 skb->tstamp = ktime_get_real();
1261 rxrpc_new_skb(skb, rxrpc_skb_received);
1263 skb_pull(skb, sizeof(struct udphdr));
1265 /* The UDP protocol already released all skb resources;
1266 * we are free to add our own data there.
1268 sp = rxrpc_skb(skb);
1270 /* dig out the RxRPC connection details */
1271 if (rxrpc_extract_header(sp, skb) < 0)
1274 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1276 if ((lose++ & 7) == 7) {
1277 trace_rxrpc_rx_lose(sp);
1278 rxrpc_free_skb(skb, rxrpc_skb_lost);
1283 if (skb->tstamp == 0)
1284 skb->tstamp = ktime_get_real();
1285 trace_rxrpc_rx_packet(sp);
1287 switch (sp->hdr.type) {
1288 case RXRPC_PACKET_TYPE_VERSION:
1289 if (rxrpc_to_client(sp))
1291 rxrpc_post_packet_to_local(local, skb);
1294 case RXRPC_PACKET_TYPE_BUSY:
1295 if (rxrpc_to_server(sp))
1298 case RXRPC_PACKET_TYPE_ACK:
1299 case RXRPC_PACKET_TYPE_ACKALL:
1300 if (sp->hdr.callNumber == 0)
1303 case RXRPC_PACKET_TYPE_ABORT:
1306 case RXRPC_PACKET_TYPE_DATA:
1307 if (sp->hdr.callNumber == 0 ||
1310 if (!rxrpc_validate_data(skb))
1313 /* Unshare the packet so that it can be modified for in-place
1316 if (sp->hdr.securityIndex != 0) {
1317 struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1319 rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1324 rxrpc_eaten_skb(skb, rxrpc_skb_received);
1326 rxrpc_new_skb(skb, rxrpc_skb_unshared);
1327 sp = rxrpc_skb(skb);
1332 case RXRPC_PACKET_TYPE_CHALLENGE:
1333 if (rxrpc_to_server(sp))
1336 case RXRPC_PACKET_TYPE_RESPONSE:
1337 if (rxrpc_to_client(sp))
1341 /* Packet types 9-11 should just be ignored. */
1342 case RXRPC_PACKET_TYPE_PARAMS:
1343 case RXRPC_PACKET_TYPE_10:
1344 case RXRPC_PACKET_TYPE_11:
1348 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1352 if (sp->hdr.serviceId == 0)
1355 if (rxrpc_to_server(sp)) {
1356 /* Weed out packets to services we're not offering. Packets
1357 * that would begin a call are explicitly rejected and the rest
1358 * are just discarded.
1360 rx = rcu_dereference(local->service);
1361 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1362 sp->hdr.serviceId != rx->second_service)) {
1363 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1365 goto unsupported_service;
1370 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1372 if (sp->hdr.securityIndex != conn->security_ix)
1373 goto wrong_security;
1375 if (sp->hdr.serviceId != conn->service_id) {
1378 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1380 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1383 if (old_id != conn->params.service_id &&
1384 old_id != sp->hdr.serviceId)
1388 if (sp->hdr.callNumber == 0) {
1389 /* Connection-level packet */
1390 _debug("CONN %p {%d}", conn, conn->debug_id);
1391 rxrpc_post_packet_to_conn(conn, skb);
1395 if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1396 conn->hi_serial = sp->hdr.serial;
1398 /* Call-bound packets are routed by connection channel. */
1399 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1400 chan = &conn->channels[channel];
1402 /* Ignore really old calls */
1403 if (sp->hdr.callNumber < chan->last_call)
1406 if (sp->hdr.callNumber == chan->last_call) {
1408 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1411 /* For the previous service call, if completed
1412 * successfully, we discard all further packets.
1414 if (rxrpc_conn_is_service(conn) &&
1415 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1418 /* But otherwise we need to retransmit the final packet
1419 * from data cached in the connection record.
1421 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1422 trace_rxrpc_rx_data(chan->call_debug_id,
1426 rxrpc_post_packet_to_conn(conn, skb);
1430 call = rcu_dereference(chan->call);
1432 if (sp->hdr.callNumber > chan->call_id) {
1433 if (rxrpc_to_client(sp))
1436 rxrpc_input_implicit_end_call(rx, conn, call);
1441 if (sp->hdr.serviceId != call->service_id)
1442 call->service_id = sp->hdr.serviceId;
1443 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1444 call->rx_serial = sp->hdr.serial;
1445 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1446 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1450 if (!call || refcount_read(&call->ref) == 0) {
1451 if (rxrpc_to_client(sp) ||
1452 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1454 if (sp->hdr.seq != 1)
1456 call = rxrpc_new_incoming_call(local, rx, skb);
1461 /* Process a call packet; this either discards or passes on the ref
1464 rxrpc_input_call_packet(call, skb);
1468 rxrpc_free_skb(skb, rxrpc_skb_freed);
1470 trace_rxrpc_rx_done(0, 0);
1474 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1475 RXKADINCONSISTENCY, EBADMSG);
1476 skb->priority = RXKADINCONSISTENCY;
1479 unsupported_service:
1480 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1481 RX_INVALID_OPERATION, EOPNOTSUPP);
1482 skb->priority = RX_INVALID_OPERATION;
1486 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1487 RX_PROTOCOL_ERROR, EBADMSG);
1488 goto protocol_error;
1491 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1492 RX_PROTOCOL_ERROR, EBADMSG);
1494 skb->priority = RX_PROTOCOL_ERROR;
1496 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1498 trace_rxrpc_rx_done(skb->mark, skb->priority);
1499 rxrpc_reject_packet(local, skb);
1500 _leave(" [badmsg]");