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);
278 rxrpc_notify_socket(call);
286 write_unlock(&call->state_lock);
287 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
288 trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
290 trace_rxrpc_transmit(call, rxrpc_transmit_end);
295 write_unlock(&call->state_lock);
296 kdebug("end_tx %s", rxrpc_call_states[call->state]);
297 rxrpc_proto_abort(abort_why, call, call->tx_top);
302 * Begin the reply reception phase of a call.
304 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
306 struct rxrpc_ack_summary summary = { 0 };
307 unsigned long now, timo;
308 rxrpc_seq_t top = READ_ONCE(call->tx_top);
310 if (call->ackr_reason) {
311 spin_lock_bh(&call->lock);
312 call->ackr_reason = 0;
313 spin_unlock_bh(&call->lock);
315 timo = now + MAX_JIFFY_OFFSET;
316 WRITE_ONCE(call->resend_at, timo);
317 WRITE_ONCE(call->ack_at, timo);
318 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
321 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
322 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
323 rxrpc_proto_abort("TXL", call, top);
327 if (!rxrpc_end_tx_phase(call, true, "ETD"))
329 call->tx_phase = false;
334 * Scan a data packet to validate its structure and to work out how many
335 * subpackets it contains.
337 * A jumbo packet is a collection of consecutive packets glued together with
338 * little headers between that indicate how to change the initial header for
341 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
342 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
345 static bool rxrpc_validate_data(struct sk_buff *skb)
347 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
348 unsigned int offset = sizeof(struct rxrpc_wire_header);
349 unsigned int len = skb->len;
350 u8 flags = sp->hdr.flags;
353 if (flags & RXRPC_REQUEST_ACK)
354 __set_bit(sp->nr_subpackets, sp->rx_req_ack);
357 if (!(flags & RXRPC_JUMBO_PACKET))
360 if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
362 if (flags & RXRPC_LAST_PACKET)
364 offset += RXRPC_JUMBO_DATALEN;
365 if (skb_copy_bits(skb, offset, &flags, 1) < 0)
367 offset += sizeof(struct rxrpc_jumbo_header);
370 if (flags & RXRPC_LAST_PACKET)
371 sp->rx_flags |= RXRPC_SKB_INCL_LAST;
379 * Handle reception of a duplicate packet.
381 * We have to take care to avoid an attack here whereby we're given a series of
382 * jumbograms, each with a sequence number one before the preceding one and
383 * filled up to maximum UDP size. If they never send us the first packet in
384 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
385 * space until the call times out.
387 * We limit the space usage by only accepting three duplicate jumbo packets per
388 * call. After that, we tell the other side we're no longer accepting jumbos
389 * (that information is encoded in the ACK packet).
391 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
392 bool is_jumbo, bool *_jumbo_bad)
394 /* Discard normal packets that are duplicates. */
398 /* Skip jumbo subpackets that are duplicates. When we've had three or
399 * more partially duplicate jumbo packets, we refuse to take any more
400 * jumbos for this call.
403 call->nr_jumbo_bad++;
409 * Process a DATA packet, adding the packet to the Rx ring. The caller's
410 * packet ref must be passed on or discarded.
412 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
414 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
415 enum rxrpc_call_state state;
416 unsigned int j, nr_subpackets, nr_unacked = 0;
417 rxrpc_serial_t serial = sp->hdr.serial, ack_serial = serial;
418 rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
419 bool immediate_ack = false, jumbo_bad = false;
422 _enter("{%u,%u},{%u,%u}",
423 call->rx_hard_ack, call->rx_top, skb->len, seq0);
425 _proto("Rx DATA %%%u { #%u f=%02x n=%u }",
426 sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
428 state = READ_ONCE(call->state);
429 if (state >= RXRPC_CALL_COMPLETE) {
430 rxrpc_free_skb(skb, rxrpc_skb_freed);
434 if (state == RXRPC_CALL_SERVER_RECV_REQUEST) {
435 unsigned long timo = READ_ONCE(call->next_req_timo);
436 unsigned long now, expect_req_by;
440 expect_req_by = now + timo;
441 WRITE_ONCE(call->expect_req_by, expect_req_by);
442 rxrpc_reduce_call_timer(call, expect_req_by, now,
443 rxrpc_timer_set_for_idle);
447 spin_lock(&call->input_lock);
449 /* Received data implicitly ACKs all of the request packets we sent
450 * when we're acting as a client.
452 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
453 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
454 !rxrpc_receiving_reply(call))
457 hard_ack = READ_ONCE(call->rx_hard_ack);
459 nr_subpackets = sp->nr_subpackets;
460 if (nr_subpackets > 1) {
461 if (call->nr_jumbo_bad > 3) {
462 ack = RXRPC_ACK_NOSPACE;
468 for (j = 0; j < nr_subpackets; j++) {
469 rxrpc_serial_t serial = sp->hdr.serial + j;
470 rxrpc_seq_t seq = seq0 + j;
471 unsigned int ix = seq & RXRPC_RXTX_BUFF_MASK;
472 bool terminal = (j == nr_subpackets - 1);
473 bool last = terminal && (sp->rx_flags & RXRPC_SKB_INCL_LAST);
474 u8 flags, annotation = j;
476 _proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
477 j, serial, seq, terminal, last);
480 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
481 seq != call->rx_top) {
482 rxrpc_proto_abort("LSN", call, seq);
486 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
487 after_eq(seq, call->rx_top)) {
488 rxrpc_proto_abort("LSA", call, seq);
495 flags |= RXRPC_LAST_PACKET;
497 flags |= RXRPC_JUMBO_PACKET;
498 if (test_bit(j, sp->rx_req_ack))
499 flags |= RXRPC_REQUEST_ACK;
500 trace_rxrpc_rx_data(call->debug_id, seq, serial, flags, annotation);
502 if (before_eq(seq, hard_ack)) {
503 ack = RXRPC_ACK_DUPLICATE;
508 if (call->rxtx_buffer[ix]) {
509 rxrpc_input_dup_data(call, seq, nr_subpackets > 1,
511 if (ack != RXRPC_ACK_DUPLICATE) {
512 ack = RXRPC_ACK_DUPLICATE;
515 immediate_ack = true;
519 if (after(seq, hard_ack + call->rx_winsize)) {
520 ack = RXRPC_ACK_EXCEEDS_WINDOW;
522 if (flags & RXRPC_JUMBO_PACKET) {
524 call->nr_jumbo_bad++;
532 if (flags & RXRPC_REQUEST_ACK && !ack) {
533 ack = RXRPC_ACK_REQUESTED;
537 if (after(seq0, call->ackr_highest_seq))
538 call->ackr_highest_seq = seq0;
540 /* Queue the packet. We use a couple of memory barriers here as need
541 * to make sure that rx_top is perceived to be set after the buffer
542 * pointer and that the buffer pointer is set after the annotation and
545 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
546 * and also rxrpc_fill_out_ack().
549 rxrpc_get_skb(skb, rxrpc_skb_got);
550 call->rxtx_annotations[ix] = annotation;
552 call->rxtx_buffer[ix] = skb;
553 if (after(seq, call->rx_top)) {
554 smp_store_release(&call->rx_top, seq);
555 } else if (before(seq, call->rx_top)) {
556 /* Send an immediate ACK if we fill in a hole */
558 ack = RXRPC_ACK_DELAY;
561 immediate_ack = true;
565 /* From this point on, we're not allowed to touch the
566 * packet any longer as its ref now belongs to the Rx
576 set_bit(RXRPC_CALL_RX_LAST, &call->flags);
578 ack = RXRPC_ACK_DELAY;
581 trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
583 trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
586 if (after_eq(seq, call->rx_expect_next)) {
587 if (after(seq, call->rx_expect_next)) {
588 _net("OOS %u > %u", seq, call->rx_expect_next);
589 ack = RXRPC_ACK_OUT_OF_SEQUENCE;
592 call->rx_expect_next = seq + 1;
599 if (atomic_add_return(nr_unacked, &call->ackr_nr_unacked) > 2 && !ack)
600 ack = RXRPC_ACK_IDLE;
603 rxrpc_propose_ACK(call, ack, ack_serial,
605 rxrpc_propose_ack_input_data);
607 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
609 rxrpc_propose_ack_input_data);
611 trace_rxrpc_notify_socket(call->debug_id, serial);
612 rxrpc_notify_socket(call);
615 spin_unlock(&call->input_lock);
616 rxrpc_free_skb(skb, rxrpc_skb_freed);
621 * Process a requested ACK.
623 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
625 rxrpc_serial_t orig_serial,
626 rxrpc_serial_t ack_serial)
628 struct rxrpc_skb_priv *sp;
633 for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
634 skb = call->rxtx_buffer[ix];
638 sent_at = skb->tstamp;
639 smp_rmb(); /* Read timestamp before serial. */
641 if (sp->hdr.serial != orig_serial)
649 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
650 orig_serial, ack_serial, sent_at, resp_time);
654 * Process the response to a ping that we sent to find out if we lost an ACK.
656 * If we got back a ping response that indicates a lower tx_top than what we
657 * had at the time of the ping transmission, we adjudge all the DATA packets
658 * sent between the response tx_top and the ping-time tx_top to have been lost.
660 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
662 rxrpc_seq_t top, bottom, seq;
665 spin_lock_bh(&call->lock);
667 bottom = call->tx_hard_ack + 1;
668 top = call->acks_lost_top;
669 if (before(bottom, top)) {
670 for (seq = bottom; before_eq(seq, top); seq++) {
671 int ix = seq & RXRPC_RXTX_BUFF_MASK;
672 u8 annotation = call->rxtx_annotations[ix];
673 u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
675 if (anno_type != RXRPC_TX_ANNO_UNACK)
677 annotation &= ~RXRPC_TX_ANNO_MASK;
678 annotation |= RXRPC_TX_ANNO_RETRANS;
679 call->rxtx_annotations[ix] = annotation;
684 spin_unlock_bh(&call->lock);
686 if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
687 rxrpc_queue_call(call);
691 * Process a ping response.
693 static void rxrpc_input_ping_response(struct rxrpc_call *call,
695 rxrpc_serial_t orig_serial,
696 rxrpc_serial_t ack_serial)
698 rxrpc_serial_t ping_serial;
701 ping_time = call->ping_time;
703 ping_serial = READ_ONCE(call->ping_serial);
705 if (orig_serial == call->acks_lost_ping)
706 rxrpc_input_check_for_lost_ack(call);
708 if (before(orig_serial, ping_serial) ||
709 !test_and_clear_bit(RXRPC_CALL_PINGING, &call->flags))
711 if (after(orig_serial, ping_serial))
714 rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
715 orig_serial, ack_serial, ping_time, resp_time);
719 * Process the extra information that may be appended to an ACK packet
721 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
722 struct rxrpc_ackinfo *ackinfo)
724 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
725 struct rxrpc_peer *peer;
728 u32 rwind = ntohl(ackinfo->rwind);
730 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
732 ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
733 rwind, ntohl(ackinfo->jumbo_max));
735 if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
736 rwind = RXRPC_RXTX_BUFF_SIZE - 1;
737 if (call->tx_winsize != rwind) {
738 if (rwind > call->tx_winsize)
740 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
741 call->tx_winsize = rwind;
744 if (call->cong_ssthresh > rwind)
745 call->cong_ssthresh = rwind;
747 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
750 if (mtu < peer->maxdata) {
751 spin_lock_bh(&peer->lock);
753 peer->mtu = mtu + peer->hdrsize;
754 spin_unlock_bh(&peer->lock);
755 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
759 wake_up(&call->waitq);
763 * Process individual soft ACKs.
765 * Each ACK in the array corresponds to one packet and can be either an ACK or
766 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
767 * packets that lie beyond the end of the ACK list are scheduled for resend by
768 * the timer on the basis that the peer might just not have processed them at
769 * the time the ACK was sent.
771 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
772 rxrpc_seq_t seq, int nr_acks,
773 struct rxrpc_ack_summary *summary)
776 u8 annotation, anno_type;
778 for (; nr_acks > 0; nr_acks--, seq++) {
779 ix = seq & RXRPC_RXTX_BUFF_MASK;
780 annotation = call->rxtx_annotations[ix];
781 anno_type = annotation & RXRPC_TX_ANNO_MASK;
782 annotation &= ~RXRPC_TX_ANNO_MASK;
784 case RXRPC_ACK_TYPE_ACK:
786 if (anno_type == RXRPC_TX_ANNO_ACK)
788 summary->nr_new_acks++;
789 call->rxtx_annotations[ix] =
790 RXRPC_TX_ANNO_ACK | annotation;
792 case RXRPC_ACK_TYPE_NACK:
793 if (!summary->nr_nacks &&
794 call->acks_lowest_nak != seq) {
795 call->acks_lowest_nak = seq;
796 summary->new_low_nack = true;
799 if (anno_type == RXRPC_TX_ANNO_NAK)
801 summary->nr_new_nacks++;
802 if (anno_type == RXRPC_TX_ANNO_RETRANS)
804 call->rxtx_annotations[ix] =
805 RXRPC_TX_ANNO_NAK | annotation;
808 return rxrpc_proto_abort("SFT", call, 0);
814 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
815 * with respect to the ack state conveyed by preceding ACKs.
817 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
818 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
820 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
822 if (after(first_pkt, base))
823 return true; /* The window advanced */
825 if (before(first_pkt, base))
826 return false; /* firstPacket regressed */
828 if (after_eq(prev_pkt, call->acks_prev_seq))
829 return true; /* previousPacket hasn't regressed. */
831 /* Some rx implementations put a serial number in previousPacket. */
832 if (after_eq(prev_pkt, base + call->tx_winsize))
838 * Process an ACK packet.
840 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
841 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
843 * A hard-ACK means that a packet has been processed and may be discarded; a
844 * soft-ACK means that the packet may be discarded and retransmission
845 * requested. A phase is complete when all packets are hard-ACK'd.
847 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
849 struct rxrpc_ack_summary summary = { 0 };
850 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
852 struct rxrpc_ackpacket ack;
853 struct rxrpc_ackinfo info;
854 u8 acks[RXRPC_MAXACKS];
856 rxrpc_serial_t ack_serial, acked_serial;
857 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
858 int nr_acks, offset, ioffset;
862 offset = sizeof(struct rxrpc_wire_header);
863 if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
864 _debug("extraction failure");
865 return rxrpc_proto_abort("XAK", call, 0);
867 offset += sizeof(buf.ack);
869 ack_serial = sp->hdr.serial;
870 acked_serial = ntohl(buf.ack.serial);
871 first_soft_ack = ntohl(buf.ack.firstPacket);
872 prev_pkt = ntohl(buf.ack.previousPacket);
873 hard_ack = first_soft_ack - 1;
874 nr_acks = buf.ack.nAcks;
875 summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
876 buf.ack.reason : RXRPC_ACK__INVALID);
878 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
879 first_soft_ack, prev_pkt,
880 summary.ack_reason, nr_acks);
882 if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
883 rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
885 if (buf.ack.reason == RXRPC_ACK_REQUESTED)
886 rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
889 if (buf.ack.reason == RXRPC_ACK_PING) {
890 _proto("Rx ACK %%%u PING Request", ack_serial);
891 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
892 ack_serial, true, true,
893 rxrpc_propose_ack_respond_to_ping);
894 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
895 rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
896 ack_serial, true, true,
897 rxrpc_propose_ack_respond_to_ack);
900 /* Discard any out-of-order or duplicate ACKs (outside lock). */
901 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
902 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
903 first_soft_ack, call->acks_first_seq,
904 prev_pkt, call->acks_prev_seq);
909 ioffset = offset + nr_acks + 3;
910 if (skb->len >= ioffset + sizeof(buf.info) &&
911 skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
912 return rxrpc_proto_abort("XAI", call, 0);
914 spin_lock(&call->input_lock);
916 /* Discard any out-of-order or duplicate ACKs (inside lock). */
917 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
918 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
919 first_soft_ack, call->acks_first_seq,
920 prev_pkt, call->acks_prev_seq);
923 call->acks_latest_ts = skb->tstamp;
925 call->acks_first_seq = first_soft_ack;
926 call->acks_prev_seq = prev_pkt;
928 /* Parse rwind and mtu sizes if provided. */
930 rxrpc_input_ackinfo(call, skb, &buf.info);
932 if (first_soft_ack == 0) {
933 rxrpc_proto_abort("AK0", call, 0);
937 /* Ignore ACKs unless we are or have just been transmitting. */
938 switch (READ_ONCE(call->state)) {
939 case RXRPC_CALL_CLIENT_SEND_REQUEST:
940 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
941 case RXRPC_CALL_SERVER_SEND_REPLY:
942 case RXRPC_CALL_SERVER_AWAIT_ACK:
948 if (before(hard_ack, call->tx_hard_ack) ||
949 after(hard_ack, call->tx_top)) {
950 rxrpc_proto_abort("AKW", call, 0);
953 if (nr_acks > call->tx_top - hard_ack) {
954 rxrpc_proto_abort("AKN", call, 0);
958 if (after(hard_ack, call->tx_hard_ack)) {
959 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
960 rxrpc_end_tx_phase(call, false, "ETA");
966 if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
967 rxrpc_proto_abort("XSA", call, 0);
970 rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
974 if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
975 RXRPC_TX_ANNO_LAST &&
976 summary.nr_acks == call->tx_top - hard_ack &&
977 rxrpc_is_client_call(call))
978 rxrpc_propose_ACK(call, RXRPC_ACK_PING, ack_serial,
980 rxrpc_propose_ack_ping_for_lost_reply);
982 rxrpc_congestion_management(call, skb, &summary, acked_serial);
984 spin_unlock(&call->input_lock);
988 * Process an ACKALL packet.
990 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
992 struct rxrpc_ack_summary summary = { 0 };
993 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
995 _proto("Rx ACKALL %%%u", sp->hdr.serial);
997 spin_lock(&call->input_lock);
999 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1000 rxrpc_end_tx_phase(call, false, "ETL");
1002 spin_unlock(&call->input_lock);
1006 * Process an ABORT packet directed at a call.
1008 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1010 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1012 u32 abort_code = RX_CALL_DEAD;
1016 if (skb->len >= 4 &&
1017 skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
1018 &wtmp, sizeof(wtmp)) >= 0)
1019 abort_code = ntohl(wtmp);
1021 trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1023 _proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1025 if (rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1026 abort_code, -ECONNABORTED))
1027 rxrpc_notify_socket(call);
1031 * Process an incoming call packet.
1033 static void rxrpc_input_call_packet(struct rxrpc_call *call,
1034 struct sk_buff *skb)
1036 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1039 _enter("%p,%p", call, skb);
1041 timo = READ_ONCE(call->next_rx_timo);
1043 unsigned long now = jiffies, expect_rx_by;
1045 expect_rx_by = now + timo;
1046 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1047 rxrpc_reduce_call_timer(call, expect_rx_by, now,
1048 rxrpc_timer_set_for_normal);
1051 switch (sp->hdr.type) {
1052 case RXRPC_PACKET_TYPE_DATA:
1053 rxrpc_input_data(call, skb);
1056 case RXRPC_PACKET_TYPE_ACK:
1057 rxrpc_input_ack(call, skb);
1060 case RXRPC_PACKET_TYPE_BUSY:
1061 _proto("Rx BUSY %%%u", sp->hdr.serial);
1063 /* Just ignore BUSY packets from the server; the retry and
1064 * lifespan timers will take care of business. BUSY packets
1065 * from the client don't make sense.
1069 case RXRPC_PACKET_TYPE_ABORT:
1070 rxrpc_input_abort(call, skb);
1073 case RXRPC_PACKET_TYPE_ACKALL:
1074 rxrpc_input_ackall(call, skb);
1081 rxrpc_free_skb(skb, rxrpc_skb_freed);
1087 * Handle a new service call on a channel implicitly completing the preceding
1088 * call on that channel. This does not apply to client conns.
1090 * TODO: If callNumber > call_id + 1, renegotiate security.
1092 static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1093 struct rxrpc_connection *conn,
1094 struct rxrpc_call *call)
1096 switch (READ_ONCE(call->state)) {
1097 case RXRPC_CALL_SERVER_AWAIT_ACK:
1098 rxrpc_call_completed(call);
1100 case RXRPC_CALL_COMPLETE:
1103 if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1104 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1105 rxrpc_queue_call(call);
1107 trace_rxrpc_improper_term(call);
1111 spin_lock(&rx->incoming_lock);
1112 __rxrpc_disconnect_call(conn, call);
1113 spin_unlock(&rx->incoming_lock);
1114 rxrpc_notify_socket(call);
1118 * post connection-level events to the connection
1119 * - this includes challenges, responses, some aborts and call terminal packet
1122 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1123 struct sk_buff *skb)
1125 _enter("%p,%p", conn, skb);
1127 skb_queue_tail(&conn->rx_queue, skb);
1128 rxrpc_queue_conn(conn);
1132 * post endpoint-level events to the local endpoint
1133 * - this includes debug and version messages
1135 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1136 struct sk_buff *skb)
1138 _enter("%p,%p", local, skb);
1140 if (rxrpc_get_local_maybe(local)) {
1141 skb_queue_tail(&local->event_queue, skb);
1142 rxrpc_queue_local(local);
1144 rxrpc_free_skb(skb, rxrpc_skb_freed);
1149 * put a packet up for transport-level abort
1151 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1153 CHECK_SLAB_OKAY(&local->usage);
1155 if (rxrpc_get_local_maybe(local)) {
1156 skb_queue_tail(&local->reject_queue, skb);
1157 rxrpc_queue_local(local);
1159 rxrpc_free_skb(skb, rxrpc_skb_freed);
1164 * Extract the wire header from a packet and translate the byte order.
1167 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1169 struct rxrpc_wire_header whdr;
1171 /* dig out the RxRPC connection details */
1172 if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1173 trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1174 tracepoint_string("bad_hdr"));
1178 memset(sp, 0, sizeof(*sp));
1179 sp->hdr.epoch = ntohl(whdr.epoch);
1180 sp->hdr.cid = ntohl(whdr.cid);
1181 sp->hdr.callNumber = ntohl(whdr.callNumber);
1182 sp->hdr.seq = ntohl(whdr.seq);
1183 sp->hdr.serial = ntohl(whdr.serial);
1184 sp->hdr.flags = whdr.flags;
1185 sp->hdr.type = whdr.type;
1186 sp->hdr.userStatus = whdr.userStatus;
1187 sp->hdr.securityIndex = whdr.securityIndex;
1188 sp->hdr._rsvd = ntohs(whdr._rsvd);
1189 sp->hdr.serviceId = ntohs(whdr.serviceId);
1194 * handle data received on the local endpoint
1195 * - may be called in interrupt context
1197 * [!] Note that as this is called from the encap_rcv hook, the socket is not
1198 * held locked by the caller and nothing prevents sk_user_data on the UDP from
1199 * being cleared in the middle of processing this function.
1201 * Called with the RCU read lock held from the IP layer via UDP.
1203 int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1205 struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1206 struct rxrpc_connection *conn;
1207 struct rxrpc_channel *chan;
1208 struct rxrpc_call *call = NULL;
1209 struct rxrpc_skb_priv *sp;
1210 struct rxrpc_peer *peer = NULL;
1211 struct rxrpc_sock *rx = NULL;
1212 unsigned int channel;
1214 _enter("%p", udp_sk);
1216 if (unlikely(!local)) {
1220 if (skb->tstamp == 0)
1221 skb->tstamp = ktime_get_real();
1223 rxrpc_new_skb(skb, rxrpc_skb_received);
1225 skb_pull(skb, sizeof(struct udphdr));
1227 /* The UDP protocol already released all skb resources;
1228 * we are free to add our own data there.
1230 sp = rxrpc_skb(skb);
1232 /* dig out the RxRPC connection details */
1233 if (rxrpc_extract_header(sp, skb) < 0)
1236 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1238 if ((lose++ & 7) == 7) {
1239 trace_rxrpc_rx_lose(sp);
1240 rxrpc_free_skb(skb, rxrpc_skb_lost);
1245 if (skb->tstamp == 0)
1246 skb->tstamp = ktime_get_real();
1247 trace_rxrpc_rx_packet(sp);
1249 switch (sp->hdr.type) {
1250 case RXRPC_PACKET_TYPE_VERSION:
1251 if (rxrpc_to_client(sp))
1253 rxrpc_post_packet_to_local(local, skb);
1256 case RXRPC_PACKET_TYPE_BUSY:
1257 if (rxrpc_to_server(sp))
1260 case RXRPC_PACKET_TYPE_ACK:
1261 case RXRPC_PACKET_TYPE_ACKALL:
1262 if (sp->hdr.callNumber == 0)
1265 case RXRPC_PACKET_TYPE_ABORT:
1268 case RXRPC_PACKET_TYPE_DATA:
1269 if (sp->hdr.callNumber == 0 ||
1272 if (!rxrpc_validate_data(skb))
1275 /* Unshare the packet so that it can be modified for in-place
1278 if (sp->hdr.securityIndex != 0) {
1279 struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1281 rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1286 rxrpc_eaten_skb(skb, rxrpc_skb_received);
1288 rxrpc_new_skb(skb, rxrpc_skb_unshared);
1289 sp = rxrpc_skb(skb);
1294 case RXRPC_PACKET_TYPE_CHALLENGE:
1295 if (rxrpc_to_server(sp))
1298 case RXRPC_PACKET_TYPE_RESPONSE:
1299 if (rxrpc_to_client(sp))
1303 /* Packet types 9-11 should just be ignored. */
1304 case RXRPC_PACKET_TYPE_PARAMS:
1305 case RXRPC_PACKET_TYPE_10:
1306 case RXRPC_PACKET_TYPE_11:
1310 _proto("Rx Bad Packet Type %u", sp->hdr.type);
1314 if (sp->hdr.serviceId == 0)
1317 if (rxrpc_to_server(sp)) {
1318 /* Weed out packets to services we're not offering. Packets
1319 * that would begin a call are explicitly rejected and the rest
1320 * are just discarded.
1322 rx = rcu_dereference(local->service);
1323 if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1324 sp->hdr.serviceId != rx->second_service)) {
1325 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1327 goto unsupported_service;
1332 conn = rxrpc_find_connection_rcu(local, skb, &peer);
1334 if (sp->hdr.securityIndex != conn->security_ix)
1335 goto wrong_security;
1337 if (sp->hdr.serviceId != conn->service_id) {
1340 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1342 old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1345 if (old_id != conn->params.service_id &&
1346 old_id != sp->hdr.serviceId)
1350 if (sp->hdr.callNumber == 0) {
1351 /* Connection-level packet */
1352 _debug("CONN %p {%d}", conn, conn->debug_id);
1353 rxrpc_post_packet_to_conn(conn, skb);
1357 if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1358 conn->hi_serial = sp->hdr.serial;
1360 /* Call-bound packets are routed by connection channel. */
1361 channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1362 chan = &conn->channels[channel];
1364 /* Ignore really old calls */
1365 if (sp->hdr.callNumber < chan->last_call)
1368 if (sp->hdr.callNumber == chan->last_call) {
1370 sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1373 /* For the previous service call, if completed
1374 * successfully, we discard all further packets.
1376 if (rxrpc_conn_is_service(conn) &&
1377 chan->last_type == RXRPC_PACKET_TYPE_ACK)
1380 /* But otherwise we need to retransmit the final packet
1381 * from data cached in the connection record.
1383 if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1384 trace_rxrpc_rx_data(chan->call_debug_id,
1388 rxrpc_post_packet_to_conn(conn, skb);
1392 call = rcu_dereference(chan->call);
1394 if (sp->hdr.callNumber > chan->call_id) {
1395 if (rxrpc_to_client(sp))
1398 rxrpc_input_implicit_end_call(rx, conn, call);
1403 if (sp->hdr.serviceId != call->service_id)
1404 call->service_id = sp->hdr.serviceId;
1405 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1406 call->rx_serial = sp->hdr.serial;
1407 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1408 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1412 if (!call || atomic_read(&call->usage) == 0) {
1413 if (rxrpc_to_client(sp) ||
1414 sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1416 if (sp->hdr.seq != 1)
1418 call = rxrpc_new_incoming_call(local, rx, skb);
1423 /* Process a call packet; this either discards or passes on the ref
1426 rxrpc_input_call_packet(call, skb);
1430 rxrpc_free_skb(skb, rxrpc_skb_freed);
1432 trace_rxrpc_rx_done(0, 0);
1436 trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1437 RXKADINCONSISTENCY, EBADMSG);
1438 skb->priority = RXKADINCONSISTENCY;
1441 unsupported_service:
1442 trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1443 RX_INVALID_OPERATION, EOPNOTSUPP);
1444 skb->priority = RX_INVALID_OPERATION;
1448 trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1449 RX_PROTOCOL_ERROR, EBADMSG);
1450 goto protocol_error;
1453 trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1454 RX_PROTOCOL_ERROR, EBADMSG);
1456 skb->priority = RX_PROTOCOL_ERROR;
1458 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1460 trace_rxrpc_rx_done(skb->mark, skb->priority);
1461 rxrpc_reject_packet(local, skb);
1462 _leave(" [badmsg]");