1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Processing of received RxRPC packets
4 * Copyright (C) 2020 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(struct rxrpc_call *call, rxrpc_seq_t seq,
13 enum rxrpc_abort_reason why)
15 rxrpc_abort_call(call, seq, RX_PROTOCOL_ERROR, -EBADMSG, why);
19 * Do TCP-style congestion management [RFC 5681].
21 static void rxrpc_congestion_management(struct rxrpc_call *call,
23 struct rxrpc_ack_summary *summary,
24 rxrpc_serial_t acked_serial)
26 enum rxrpc_congest_change change = rxrpc_cong_no_change;
27 unsigned int cumulative_acks = call->cong_cumul_acks;
28 unsigned int cwnd = call->cong_cwnd;
31 summary->flight_size =
32 (call->tx_top - call->acks_hard_ack) - summary->nr_acks;
34 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT, &call->flags)) {
35 summary->retrans_timeo = true;
36 call->cong_ssthresh = max_t(unsigned int,
37 summary->flight_size / 2, 2);
39 if (cwnd >= call->cong_ssthresh &&
40 call->cong_mode == RXRPC_CALL_SLOW_START) {
41 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
42 call->cong_tstamp = skb->tstamp;
47 cumulative_acks += summary->nr_new_acks;
48 cumulative_acks += summary->nr_rot_new_acks;
49 if (cumulative_acks > 255)
50 cumulative_acks = 255;
52 summary->mode = call->cong_mode;
53 summary->cwnd = call->cong_cwnd;
54 summary->ssthresh = call->cong_ssthresh;
55 summary->cumulative_acks = cumulative_acks;
56 summary->dup_acks = call->cong_dup_acks;
58 switch (call->cong_mode) {
59 case RXRPC_CALL_SLOW_START:
60 if (summary->saw_nacks)
61 goto packet_loss_detected;
62 if (summary->cumulative_acks > 0)
64 if (cwnd >= call->cong_ssthresh) {
65 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
66 call->cong_tstamp = skb->tstamp;
70 case RXRPC_CALL_CONGEST_AVOIDANCE:
71 if (summary->saw_nacks)
72 goto packet_loss_detected;
74 /* We analyse the number of packets that get ACK'd per RTT
75 * period and increase the window if we managed to fill it.
77 if (call->peer->rtt_count == 0)
79 if (ktime_before(skb->tstamp,
80 ktime_add_us(call->cong_tstamp,
81 call->peer->srtt_us >> 3)))
83 change = rxrpc_cong_rtt_window_end;
84 call->cong_tstamp = skb->tstamp;
85 if (cumulative_acks >= cwnd)
89 case RXRPC_CALL_PACKET_LOSS:
90 if (!summary->saw_nacks)
91 goto resume_normality;
93 if (summary->new_low_nack) {
94 change = rxrpc_cong_new_low_nack;
95 call->cong_dup_acks = 1;
96 if (call->cong_extra > 1)
101 call->cong_dup_acks++;
102 if (call->cong_dup_acks < 3)
103 goto send_extra_data;
105 change = rxrpc_cong_begin_retransmission;
106 call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
107 call->cong_ssthresh = max_t(unsigned int,
108 summary->flight_size / 2, 2);
109 cwnd = call->cong_ssthresh + 3;
110 call->cong_extra = 0;
111 call->cong_dup_acks = 0;
115 case RXRPC_CALL_FAST_RETRANSMIT:
116 if (!summary->new_low_nack) {
117 if (summary->nr_new_acks == 0)
119 call->cong_dup_acks++;
120 if (call->cong_dup_acks == 2) {
121 change = rxrpc_cong_retransmit_again;
122 call->cong_dup_acks = 0;
126 change = rxrpc_cong_progress;
127 cwnd = call->cong_ssthresh;
128 if (!summary->saw_nacks)
129 goto resume_normality;
139 change = rxrpc_cong_cleared_nacks;
140 call->cong_dup_acks = 0;
141 call->cong_extra = 0;
142 call->cong_tstamp = skb->tstamp;
143 if (cwnd < call->cong_ssthresh)
144 call->cong_mode = RXRPC_CALL_SLOW_START;
146 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
150 if (cwnd >= RXRPC_TX_MAX_WINDOW)
151 cwnd = RXRPC_TX_MAX_WINDOW;
152 call->cong_cwnd = cwnd;
153 call->cong_cumul_acks = cumulative_acks;
154 trace_rxrpc_congest(call, summary, acked_serial, change);
156 rxrpc_resend(call, skb);
159 packet_loss_detected:
160 change = rxrpc_cong_saw_nack;
161 call->cong_mode = RXRPC_CALL_PACKET_LOSS;
162 call->cong_dup_acks = 0;
163 goto send_extra_data;
166 /* Send some previously unsent DATA if we have some to advance the ACK
169 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
170 summary->nr_acks != call->tx_top - call->acks_hard_ack) {
172 wake_up(&call->waitq);
174 goto out_no_clear_ca;
178 * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
180 void rxrpc_congestion_degrade(struct rxrpc_call *call)
184 if (call->cong_mode != RXRPC_CALL_SLOW_START &&
185 call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
187 if (call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
190 rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
191 now = ktime_get_real();
192 if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
195 trace_rxrpc_reset_cwnd(call, now);
196 rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
197 call->tx_last_sent = now;
198 call->cong_mode = RXRPC_CALL_SLOW_START;
199 call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
200 call->cong_cwnd * 3 / 4);
201 call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
205 * Apply a hard ACK by advancing the Tx window.
207 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
208 struct rxrpc_ack_summary *summary)
210 struct rxrpc_txbuf *txb;
211 bool rot_last = false;
213 list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
214 if (before_eq(txb->seq, call->acks_hard_ack))
216 summary->nr_rot_new_acks++;
217 if (test_bit(RXRPC_TXBUF_LAST, &txb->flags)) {
218 set_bit(RXRPC_CALL_TX_LAST, &call->flags);
226 set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
228 _enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
230 if (call->acks_lowest_nak == call->acks_hard_ack) {
231 call->acks_lowest_nak = to;
232 } else if (after(to, call->acks_lowest_nak)) {
233 summary->new_low_nack = true;
234 call->acks_lowest_nak = to;
237 smp_store_release(&call->acks_hard_ack, to);
239 trace_rxrpc_txqueue(call, (rot_last ?
240 rxrpc_txqueue_rotate_last :
241 rxrpc_txqueue_rotate));
242 wake_up(&call->waitq);
247 * End the transmission phase of a call.
249 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
250 * or a final ACK packet.
252 static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
253 enum rxrpc_abort_reason abort_why)
257 ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
259 write_lock(&call->state_lock);
263 case RXRPC_CALL_CLIENT_SEND_REQUEST:
264 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
266 call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
268 call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
271 case RXRPC_CALL_SERVER_AWAIT_ACK:
272 __rxrpc_call_completed(call);
280 write_unlock(&call->state_lock);
281 if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
282 trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
284 trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
289 write_unlock(&call->state_lock);
290 kdebug("end_tx %s", rxrpc_call_states[call->state]);
291 rxrpc_proto_abort(call, call->tx_top, abort_why);
295 * Begin the reply reception phase of a call.
297 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
299 struct rxrpc_ack_summary summary = { 0 };
300 unsigned long now, timo;
301 rxrpc_seq_t top = READ_ONCE(call->tx_top);
303 if (call->ackr_reason) {
305 timo = now + MAX_JIFFY_OFFSET;
306 WRITE_ONCE(call->resend_at, timo);
307 WRITE_ONCE(call->delay_ack_at, timo);
308 trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
311 if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
312 if (!rxrpc_rotate_tx_window(call, top, &summary)) {
313 rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
318 rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
323 * End the packet reception phase.
325 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
327 rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
329 _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
331 trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
333 if (rxrpc_call_state(call) == RXRPC_CALL_CLIENT_RECV_REPLY)
334 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
336 write_lock(&call->state_lock);
338 switch (call->state) {
339 case RXRPC_CALL_CLIENT_RECV_REPLY:
340 __rxrpc_call_completed(call);
341 write_unlock(&call->state_lock);
344 case RXRPC_CALL_SERVER_RECV_REQUEST:
345 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
346 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
347 write_unlock(&call->state_lock);
348 rxrpc_propose_delay_ACK(call, serial,
349 rxrpc_propose_ack_processing_op);
352 write_unlock(&call->state_lock);
357 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
358 rxrpc_seq_t window, rxrpc_seq_t wtop)
360 atomic64_set_release(&call->ackr_window, ((u64)wtop) << 32 | window);
364 * Push a DATA packet onto the Rx queue.
366 static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
367 rxrpc_seq_t window, rxrpc_seq_t wtop,
368 enum rxrpc_receive_trace why)
370 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
371 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
373 __skb_queue_tail(&call->recvmsg_queue, skb);
374 rxrpc_input_update_ack_window(call, window, wtop);
375 trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
377 rxrpc_end_rx_phase(call, sp->hdr.serial);
381 * Process a DATA packet.
383 static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
386 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
388 rxrpc_serial_t serial = sp->hdr.serial;
389 u64 win = atomic64_read(&call->ackr_window);
390 rxrpc_seq_t window = lower_32_bits(win);
391 rxrpc_seq_t wtop = upper_32_bits(win);
392 rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
393 rxrpc_seq_t seq = sp->hdr.seq;
394 bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
397 rxrpc_inc_stat(call->rxnet, stat_rx_data);
398 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
399 rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
400 if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
401 rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
404 if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
406 return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
408 if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
409 after_eq(seq, wtop)) {
410 pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
411 call->debug_id, seq, window, wtop, wlimit);
412 return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
416 if (after(seq, call->rx_highest_seq))
417 call->rx_highest_seq = seq;
419 trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
421 if (before(seq, window)) {
422 ack_reason = RXRPC_ACK_DUPLICATE;
425 if (after(seq, wlimit)) {
426 ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
430 /* Queue the packet. */
432 rxrpc_seq_t reset_from;
433 bool reset_sack = false;
435 if (sp->hdr.flags & RXRPC_REQUEST_ACK)
436 ack_reason = RXRPC_ACK_REQUESTED;
437 /* Send an immediate ACK if we fill in a hole */
438 else if (!skb_queue_empty(&call->rx_oos_queue))
439 ack_reason = RXRPC_ACK_DELAY;
441 atomic_inc_return(&call->ackr_nr_unacked);
444 if (after(window, wtop))
447 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
449 spin_lock(&call->recvmsg_queue.lock);
450 rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
453 while ((oos = skb_peek(&call->rx_oos_queue))) {
454 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
456 if (after(osp->hdr.seq, window))
459 __skb_unlink(oos, &call->rx_oos_queue);
460 last = osp->hdr.flags & RXRPC_LAST_PACKET;
468 rxrpc_input_queue_data(call, oos, window, wtop,
469 rxrpc_receive_queue_oos);
472 spin_unlock(&call->recvmsg_queue.lock);
476 call->ackr_sack_table[reset_from % RXRPC_SACK_SIZE] = 0;
477 } while (reset_from++, before(reset_from, window));
482 ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
484 if (!call->ackr_sack_table[seq % RXRPC_SACK_SIZE]) {
485 call->ackr_sack_table[seq % RXRPC_SACK_SIZE] = 1;
489 if (after(seq + 1, wtop)) {
491 rxrpc_input_update_ack_window(call, window, wtop);
495 ack_reason = RXRPC_ACK_DUPLICATE;
499 skb_queue_walk(&call->rx_oos_queue, oos) {
500 struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
502 if (after(osp->hdr.seq, seq)) {
503 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
504 __skb_queue_before(&call->rx_oos_queue, oos, skb);
509 rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
510 __skb_queue_tail(&call->rx_oos_queue, skb);
512 trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
513 sp->hdr.serial, sp->hdr.seq);
518 rxrpc_send_ACK(call, ack_reason, serial,
519 rxrpc_propose_ack_input_data);
521 rxrpc_propose_delay_ACK(call, serial,
522 rxrpc_propose_ack_input_data);
526 * Split a jumbo packet and file the bits separately.
528 static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
530 struct rxrpc_jumbo_header jhdr;
531 struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
532 struct sk_buff *jskb;
533 unsigned int offset = sizeof(struct rxrpc_wire_header);
534 unsigned int len = skb->len - offset;
537 while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
538 if (len < RXRPC_JUMBO_SUBPKTLEN)
540 if (sp->hdr.flags & RXRPC_LAST_PACKET)
542 if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
543 &jhdr, sizeof(jhdr)) < 0)
546 jskb = skb_clone(skb, GFP_NOFS);
548 kdebug("couldn't clone");
551 rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
552 jsp = rxrpc_skb(jskb);
553 jsp->offset = offset;
554 jsp->len = RXRPC_JUMBO_DATALEN;
555 rxrpc_input_data_one(call, jskb, ¬ify);
556 rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
558 sp->hdr.flags = jhdr.flags;
559 sp->hdr._rsvd = ntohs(jhdr._rsvd);
562 offset += RXRPC_JUMBO_SUBPKTLEN;
563 len -= RXRPC_JUMBO_SUBPKTLEN;
568 rxrpc_input_data_one(call, skb, ¬ify);
570 trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
571 rxrpc_notify_socket(call);
580 * Process a DATA packet, adding the packet to the Rx ring. The caller's
581 * packet ref must be passed on or discarded.
583 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
585 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
586 enum rxrpc_call_state state;
587 rxrpc_serial_t serial = sp->hdr.serial;
588 rxrpc_seq_t seq0 = sp->hdr.seq;
590 _enter("{%llx,%x},{%u,%x}",
591 atomic64_read(&call->ackr_window), call->rx_highest_seq,
594 state = READ_ONCE(call->state);
595 if (state >= RXRPC_CALL_COMPLETE)
598 if (state == RXRPC_CALL_SERVER_RECV_REQUEST) {
599 unsigned long timo = READ_ONCE(call->next_req_timo);
600 unsigned long now, expect_req_by;
604 expect_req_by = now + timo;
605 WRITE_ONCE(call->expect_req_by, expect_req_by);
606 rxrpc_reduce_call_timer(call, expect_req_by, now,
607 rxrpc_timer_set_for_idle);
611 /* Received data implicitly ACKs all of the request packets we sent
612 * when we're acting as a client.
614 if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
615 state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
616 !rxrpc_receiving_reply(call))
619 if (!rxrpc_input_split_jumbo(call, skb)) {
620 rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
626 trace_rxrpc_notify_socket(call->debug_id, serial);
627 rxrpc_notify_socket(call);
632 * See if there's a cached RTT probe to complete.
634 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
636 rxrpc_serial_t acked_serial,
637 rxrpc_serial_t ack_serial,
638 enum rxrpc_rtt_rx_trace type)
640 rxrpc_serial_t orig_serial;
643 bool matched = false;
646 avail = READ_ONCE(call->rtt_avail);
647 smp_rmb(); /* Read avail bits before accessing data. */
649 for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
650 if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
653 sent_at = call->rtt_sent_at[i];
654 orig_serial = call->rtt_serial[i];
656 if (orig_serial == acked_serial) {
657 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
658 smp_mb(); /* Read data before setting avail bit */
659 set_bit(i, &call->rtt_avail);
660 if (type != rxrpc_rtt_rx_cancel)
661 rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
664 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
665 orig_serial, acked_serial, 0, 0);
669 /* If a later serial is being acked, then mark this slot as
672 if (after(acked_serial, orig_serial)) {
673 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
674 orig_serial, acked_serial, 0, 0);
675 clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
677 set_bit(i, &call->rtt_avail);
682 trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
686 * Process the extra information that may be appended to an ACK packet
688 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
689 struct rxrpc_ackinfo *ackinfo)
691 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
692 struct rxrpc_peer *peer;
695 u32 rwind = ntohl(ackinfo->rwind);
697 if (rwind > RXRPC_TX_MAX_WINDOW)
698 rwind = RXRPC_TX_MAX_WINDOW;
699 if (call->tx_winsize != rwind) {
700 if (rwind > call->tx_winsize)
702 trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
703 call->tx_winsize = rwind;
706 if (call->cong_ssthresh > rwind)
707 call->cong_ssthresh = rwind;
709 mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
712 if (mtu < peer->maxdata) {
713 spin_lock(&peer->lock);
715 peer->mtu = mtu + peer->hdrsize;
716 spin_unlock(&peer->lock);
720 wake_up(&call->waitq);
724 * Process individual soft ACKs.
726 * Each ACK in the array corresponds to one packet and can be either an ACK or
727 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
728 * packets that lie beyond the end of the ACK list are scheduled for resend by
729 * the timer on the basis that the peer might just not have processed them at
730 * the time the ACK was sent.
732 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
733 rxrpc_seq_t seq, int nr_acks,
734 struct rxrpc_ack_summary *summary)
738 for (i = 0; i < nr_acks; i++) {
739 if (acks[i] == RXRPC_ACK_TYPE_ACK) {
741 summary->nr_new_acks++;
743 if (!summary->saw_nacks &&
744 call->acks_lowest_nak != seq + i) {
745 call->acks_lowest_nak = seq + i;
746 summary->new_low_nack = true;
748 summary->saw_nacks = true;
754 * Return true if the ACK is valid - ie. it doesn't appear to have regressed
755 * with respect to the ack state conveyed by preceding ACKs.
757 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
758 rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
760 rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
762 if (after(first_pkt, base))
763 return true; /* The window advanced */
765 if (before(first_pkt, base))
766 return false; /* firstPacket regressed */
768 if (after_eq(prev_pkt, call->acks_prev_seq))
769 return true; /* previousPacket hasn't regressed. */
771 /* Some rx implementations put a serial number in previousPacket. */
772 if (after_eq(prev_pkt, base + call->tx_winsize))
778 * Process an ACK packet.
780 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
781 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
783 * A hard-ACK means that a packet has been processed and may be discarded; a
784 * soft-ACK means that the packet may be discarded and retransmission
785 * requested. A phase is complete when all packets are hard-ACK'd.
787 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
789 struct rxrpc_ack_summary summary = { 0 };
790 struct rxrpc_ackpacket ack;
791 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
792 struct rxrpc_ackinfo info;
793 rxrpc_serial_t ack_serial, acked_serial;
794 rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
795 int nr_acks, offset, ioffset;
799 offset = sizeof(struct rxrpc_wire_header);
800 if (skb_copy_bits(skb, offset, &ack, sizeof(ack)) < 0)
801 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack);
802 offset += sizeof(ack);
804 ack_serial = sp->hdr.serial;
805 acked_serial = ntohl(ack.serial);
806 first_soft_ack = ntohl(ack.firstPacket);
807 prev_pkt = ntohl(ack.previousPacket);
808 hard_ack = first_soft_ack - 1;
810 summary.ack_reason = (ack.reason < RXRPC_ACK__INVALID ?
811 ack.reason : RXRPC_ACK__INVALID);
813 trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
814 first_soft_ack, prev_pkt,
815 summary.ack_reason, nr_acks);
816 rxrpc_inc_stat(call->rxnet, stat_rx_acks[ack.reason]);
818 switch (ack.reason) {
819 case RXRPC_ACK_PING_RESPONSE:
820 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
821 rxrpc_rtt_rx_ping_response);
823 case RXRPC_ACK_REQUESTED:
824 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
825 rxrpc_rtt_rx_requested_ack);
828 if (acked_serial != 0)
829 rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
830 rxrpc_rtt_rx_cancel);
834 if (ack.reason == RXRPC_ACK_PING) {
835 rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
836 rxrpc_propose_ack_respond_to_ping);
837 } else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
838 rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
839 rxrpc_propose_ack_respond_to_ack);
842 /* If we get an EXCEEDS_WINDOW ACK from the server, it probably
843 * indicates that the client address changed due to NAT. The server
844 * lost the call because it switched to a different peer.
846 if (unlikely(ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
847 first_soft_ack == 1 &&
849 rxrpc_is_client_call(call)) {
850 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
855 /* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
856 * indicate a change of address. However, we can retransmit the call
857 * if we still have it buffered to the beginning.
859 if (unlikely(ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
860 first_soft_ack == 1 &&
862 call->acks_hard_ack == 0 &&
863 rxrpc_is_client_call(call)) {
864 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
869 /* Discard any out-of-order or duplicate ACKs (outside lock). */
870 if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
871 trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
872 first_soft_ack, call->acks_first_seq,
873 prev_pkt, call->acks_prev_seq);
878 ioffset = offset + nr_acks + 3;
879 if (skb->len >= ioffset + sizeof(info) &&
880 skb_copy_bits(skb, ioffset, &info, sizeof(info)) < 0)
881 return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_info);
886 call->acks_latest_ts = skb->tstamp;
887 call->acks_first_seq = first_soft_ack;
888 call->acks_prev_seq = prev_pkt;
890 switch (ack.reason) {
894 if (after(acked_serial, call->acks_highest_serial))
895 call->acks_highest_serial = acked_serial;
899 /* Parse rwind and mtu sizes if provided. */
901 rxrpc_input_ackinfo(call, skb, &info);
903 if (first_soft_ack == 0)
904 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
906 /* Ignore ACKs unless we are or have just been transmitting. */
907 switch (READ_ONCE(call->state)) {
908 case RXRPC_CALL_CLIENT_SEND_REQUEST:
909 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
910 case RXRPC_CALL_SERVER_SEND_REPLY:
911 case RXRPC_CALL_SERVER_AWAIT_ACK:
917 if (before(hard_ack, call->acks_hard_ack) ||
918 after(hard_ack, call->tx_top))
919 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
920 if (nr_acks > call->tx_top - hard_ack)
921 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
923 if (after(hard_ack, call->acks_hard_ack)) {
924 if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
925 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
931 if (offset > (int)skb->len - nr_acks)
932 return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
933 rxrpc_input_soft_acks(call, skb->data + offset, first_soft_ack,
937 if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
938 summary.nr_acks == call->tx_top - hard_ack &&
939 rxrpc_is_client_call(call))
940 rxrpc_propose_ping(call, ack_serial,
941 rxrpc_propose_ack_ping_for_lost_reply);
943 rxrpc_congestion_management(call, skb, &summary, acked_serial);
947 * Process an ACKALL packet.
949 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
951 struct rxrpc_ack_summary summary = { 0 };
953 if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
954 rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
958 * Process an ABORT packet directed at a call.
960 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
962 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
964 trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
966 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
967 skb->priority, -ECONNABORTED);
971 * Process an incoming call packet.
973 void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
975 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
978 _enter("%p,%p", call, skb);
980 if (sp->hdr.serviceId != call->dest_srx.srx_service)
981 call->dest_srx.srx_service = sp->hdr.serviceId;
982 if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
983 call->rx_serial = sp->hdr.serial;
984 if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
985 set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
987 timo = READ_ONCE(call->next_rx_timo);
989 unsigned long now = jiffies, expect_rx_by;
991 expect_rx_by = now + timo;
992 WRITE_ONCE(call->expect_rx_by, expect_rx_by);
993 rxrpc_reduce_call_timer(call, expect_rx_by, now,
994 rxrpc_timer_set_for_normal);
997 switch (sp->hdr.type) {
998 case RXRPC_PACKET_TYPE_DATA:
999 return rxrpc_input_data(call, skb);
1001 case RXRPC_PACKET_TYPE_ACK:
1002 return rxrpc_input_ack(call, skb);
1004 case RXRPC_PACKET_TYPE_BUSY:
1005 /* Just ignore BUSY packets from the server; the retry and
1006 * lifespan timers will take care of business. BUSY packets
1007 * from the client don't make sense.
1011 case RXRPC_PACKET_TYPE_ABORT:
1012 return rxrpc_input_abort(call, skb);
1014 case RXRPC_PACKET_TYPE_ACKALL:
1015 return rxrpc_input_ackall(call, skb);
1023 * Handle a new service call on a channel implicitly completing the preceding
1024 * call on that channel. This does not apply to client conns.
1026 * TODO: If callNumber > call_id + 1, renegotiate security.
1028 void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1030 switch (READ_ONCE(call->state)) {
1031 case RXRPC_CALL_SERVER_AWAIT_ACK:
1032 rxrpc_call_completed(call);
1034 case RXRPC_CALL_COMPLETE:
1037 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1038 rxrpc_eproto_improper_term);
1039 trace_rxrpc_improper_term(call);
1043 rxrpc_input_call_event(call, skb);