1 /* RxRPC packet reception
3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/module.h>
15 #include <linux/net.h>
16 #include <linux/skbuff.h>
17 #include <linux/errqueue.h>
18 #include <linux/udp.h>
20 #include <linux/in6.h>
21 #include <linux/icmp.h>
22 #include <linux/gfp.h>
24 #include <net/af_rxrpc.h>
27 #include <net/net_namespace.h>
28 #include "ar-internal.h"
30 static void rxrpc_proto_abort(const char *why
,
31 struct rxrpc_call
*call
, rxrpc_seq_t seq
)
33 if (rxrpc_abort_call(why
, call
, seq
, RX_PROTOCOL_ERROR
, -EBADMSG
)) {
34 set_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
35 rxrpc_queue_call(call
);
40 * Do TCP-style congestion management [RFC 5681].
42 static void rxrpc_congestion_management(struct rxrpc_call
*call
,
44 struct rxrpc_ack_summary
*summary
,
45 rxrpc_serial_t acked_serial
)
47 enum rxrpc_congest_change change
= rxrpc_cong_no_change
;
48 unsigned int cumulative_acks
= call
->cong_cumul_acks
;
49 unsigned int cwnd
= call
->cong_cwnd
;
52 summary
->flight_size
=
53 (call
->tx_top
- call
->tx_hard_ack
) - summary
->nr_acks
;
55 if (test_and_clear_bit(RXRPC_CALL_RETRANS_TIMEOUT
, &call
->flags
)) {
56 summary
->retrans_timeo
= true;
57 call
->cong_ssthresh
= max_t(unsigned int,
58 summary
->flight_size
/ 2, 2);
60 if (cwnd
>= call
->cong_ssthresh
&&
61 call
->cong_mode
== RXRPC_CALL_SLOW_START
) {
62 call
->cong_mode
= RXRPC_CALL_CONGEST_AVOIDANCE
;
63 call
->cong_tstamp
= skb
->tstamp
;
68 cumulative_acks
+= summary
->nr_new_acks
;
69 cumulative_acks
+= summary
->nr_rot_new_acks
;
70 if (cumulative_acks
> 255)
71 cumulative_acks
= 255;
73 summary
->mode
= call
->cong_mode
;
74 summary
->cwnd
= call
->cong_cwnd
;
75 summary
->ssthresh
= call
->cong_ssthresh
;
76 summary
->cumulative_acks
= cumulative_acks
;
77 summary
->dup_acks
= call
->cong_dup_acks
;
79 switch (call
->cong_mode
) {
80 case RXRPC_CALL_SLOW_START
:
81 if (summary
->nr_nacks
> 0)
82 goto packet_loss_detected
;
83 if (summary
->cumulative_acks
> 0)
85 if (cwnd
>= call
->cong_ssthresh
) {
86 call
->cong_mode
= RXRPC_CALL_CONGEST_AVOIDANCE
;
87 call
->cong_tstamp
= skb
->tstamp
;
91 case RXRPC_CALL_CONGEST_AVOIDANCE
:
92 if (summary
->nr_nacks
> 0)
93 goto packet_loss_detected
;
95 /* We analyse the number of packets that get ACK'd per RTT
96 * period and increase the window if we managed to fill it.
98 if (call
->peer
->rtt_usage
== 0)
100 if (ktime_before(skb
->tstamp
,
101 ktime_add_ns(call
->cong_tstamp
,
103 goto out_no_clear_ca
;
104 change
= rxrpc_cong_rtt_window_end
;
105 call
->cong_tstamp
= skb
->tstamp
;
106 if (cumulative_acks
>= cwnd
)
110 case RXRPC_CALL_PACKET_LOSS
:
111 if (summary
->nr_nacks
== 0)
112 goto resume_normality
;
114 if (summary
->new_low_nack
) {
115 change
= rxrpc_cong_new_low_nack
;
116 call
->cong_dup_acks
= 1;
117 if (call
->cong_extra
> 1)
118 call
->cong_extra
= 1;
119 goto send_extra_data
;
122 call
->cong_dup_acks
++;
123 if (call
->cong_dup_acks
< 3)
124 goto send_extra_data
;
126 change
= rxrpc_cong_begin_retransmission
;
127 call
->cong_mode
= RXRPC_CALL_FAST_RETRANSMIT
;
128 call
->cong_ssthresh
= max_t(unsigned int,
129 summary
->flight_size
/ 2, 2);
130 cwnd
= call
->cong_ssthresh
+ 3;
131 call
->cong_extra
= 0;
132 call
->cong_dup_acks
= 0;
136 case RXRPC_CALL_FAST_RETRANSMIT
:
137 if (!summary
->new_low_nack
) {
138 if (summary
->nr_new_acks
== 0)
140 call
->cong_dup_acks
++;
141 if (call
->cong_dup_acks
== 2) {
142 change
= rxrpc_cong_retransmit_again
;
143 call
->cong_dup_acks
= 0;
147 change
= rxrpc_cong_progress
;
148 cwnd
= call
->cong_ssthresh
;
149 if (summary
->nr_nacks
== 0)
150 goto resume_normality
;
160 change
= rxrpc_cong_cleared_nacks
;
161 call
->cong_dup_acks
= 0;
162 call
->cong_extra
= 0;
163 call
->cong_tstamp
= skb
->tstamp
;
164 if (cwnd
< call
->cong_ssthresh
)
165 call
->cong_mode
= RXRPC_CALL_SLOW_START
;
167 call
->cong_mode
= RXRPC_CALL_CONGEST_AVOIDANCE
;
171 if (cwnd
>= RXRPC_RXTX_BUFF_SIZE
- 1)
172 cwnd
= RXRPC_RXTX_BUFF_SIZE
- 1;
173 call
->cong_cwnd
= cwnd
;
174 call
->cong_cumul_acks
= cumulative_acks
;
175 trace_rxrpc_congest(call
, summary
, acked_serial
, change
);
176 if (resend
&& !test_and_set_bit(RXRPC_CALL_EV_RESEND
, &call
->events
))
177 rxrpc_queue_call(call
);
180 packet_loss_detected
:
181 change
= rxrpc_cong_saw_nack
;
182 call
->cong_mode
= RXRPC_CALL_PACKET_LOSS
;
183 call
->cong_dup_acks
= 0;
184 goto send_extra_data
;
187 /* Send some previously unsent DATA if we have some to advance the ACK
190 if (call
->rxtx_annotations
[call
->tx_top
& RXRPC_RXTX_BUFF_MASK
] &
191 RXRPC_TX_ANNO_LAST
||
192 summary
->nr_acks
!= call
->tx_top
- call
->tx_hard_ack
) {
194 wake_up(&call
->waitq
);
196 goto out_no_clear_ca
;
200 * Ping the other end to fill our RTT cache and to retrieve the rwind
201 * and MTU parameters.
203 static void rxrpc_send_ping(struct rxrpc_call
*call
, struct sk_buff
*skb
,
206 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
207 ktime_t now
= skb
->tstamp
;
209 if (call
->peer
->rtt_usage
< 3 ||
210 ktime_before(ktime_add_ms(call
->peer
->rtt_last_req
, 1000), now
))
211 rxrpc_propose_ACK(call
, RXRPC_ACK_PING
, skew
, sp
->hdr
.serial
,
213 rxrpc_propose_ack_ping_for_params
);
217 * Apply a hard ACK by advancing the Tx window.
219 static void rxrpc_rotate_tx_window(struct rxrpc_call
*call
, rxrpc_seq_t to
,
220 struct rxrpc_ack_summary
*summary
)
222 struct sk_buff
*skb
, *list
= NULL
;
226 if (call
->acks_lowest_nak
== call
->tx_hard_ack
) {
227 call
->acks_lowest_nak
= to
;
228 } else if (before_eq(call
->acks_lowest_nak
, to
)) {
229 summary
->new_low_nack
= true;
230 call
->acks_lowest_nak
= to
;
233 spin_lock(&call
->lock
);
235 while (before(call
->tx_hard_ack
, to
)) {
237 ix
= call
->tx_hard_ack
& RXRPC_RXTX_BUFF_MASK
;
238 skb
= call
->rxtx_buffer
[ix
];
239 annotation
= call
->rxtx_annotations
[ix
];
240 rxrpc_see_skb(skb
, rxrpc_skb_tx_rotated
);
241 call
->rxtx_buffer
[ix
] = NULL
;
242 call
->rxtx_annotations
[ix
] = 0;
246 if (annotation
& RXRPC_TX_ANNO_LAST
)
247 set_bit(RXRPC_CALL_TX_LAST
, &call
->flags
);
248 if ((annotation
& RXRPC_TX_ANNO_MASK
) != RXRPC_TX_ANNO_ACK
)
249 summary
->nr_rot_new_acks
++;
252 spin_unlock(&call
->lock
);
254 trace_rxrpc_transmit(call
, (test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
) ?
255 rxrpc_transmit_rotate_last
:
256 rxrpc_transmit_rotate
));
257 wake_up(&call
->waitq
);
263 rxrpc_free_skb(skb
, rxrpc_skb_tx_freed
);
268 * End the transmission phase of a call.
270 * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
271 * or a final ACK packet.
273 static bool rxrpc_end_tx_phase(struct rxrpc_call
*call
, bool reply_begun
,
274 const char *abort_why
)
277 ASSERT(test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
));
279 write_lock(&call
->state_lock
);
281 switch (call
->state
) {
282 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
283 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
285 call
->state
= RXRPC_CALL_CLIENT_RECV_REPLY
;
287 call
->state
= RXRPC_CALL_CLIENT_AWAIT_REPLY
;
290 case RXRPC_CALL_SERVER_AWAIT_ACK
:
291 __rxrpc_call_completed(call
);
292 rxrpc_notify_socket(call
);
299 write_unlock(&call
->state_lock
);
300 if (call
->state
== RXRPC_CALL_CLIENT_AWAIT_REPLY
) {
301 trace_rxrpc_transmit(call
, rxrpc_transmit_await_reply
);
303 trace_rxrpc_transmit(call
, rxrpc_transmit_end
);
309 write_unlock(&call
->state_lock
);
310 kdebug("end_tx %s", rxrpc_call_states
[call
->state
]);
311 rxrpc_proto_abort(abort_why
, call
, call
->tx_top
);
316 * Begin the reply reception phase of a call.
318 static bool rxrpc_receiving_reply(struct rxrpc_call
*call
)
320 struct rxrpc_ack_summary summary
= { 0 };
321 unsigned long now
, timo
;
322 rxrpc_seq_t top
= READ_ONCE(call
->tx_top
);
324 if (call
->ackr_reason
) {
325 spin_lock_bh(&call
->lock
);
326 call
->ackr_reason
= 0;
327 spin_unlock_bh(&call
->lock
);
329 timo
= now
+ MAX_JIFFY_OFFSET
;
330 WRITE_ONCE(call
->resend_at
, timo
);
331 WRITE_ONCE(call
->ack_at
, timo
);
332 trace_rxrpc_timer(call
, rxrpc_timer_init_for_reply
, now
);
335 if (!test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
))
336 rxrpc_rotate_tx_window(call
, top
, &summary
);
337 if (!test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
)) {
338 rxrpc_proto_abort("TXL", call
, top
);
341 if (!rxrpc_end_tx_phase(call
, true, "ETD"))
343 call
->tx_phase
= false;
348 * Scan a jumbo packet to validate its structure and to work out how many
349 * subpackets it contains.
351 * A jumbo packet is a collection of consecutive packets glued together with
352 * little headers between that indicate how to change the initial header for
355 * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
356 * the last are RXRPC_JUMBO_DATALEN in size. The last subpacket may be of any
359 static bool rxrpc_validate_jumbo(struct sk_buff
*skb
)
361 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
362 unsigned int offset
= sizeof(struct rxrpc_wire_header
);
363 unsigned int len
= skb
->len
;
365 u8 flags
= sp
->hdr
.flags
;
369 if (len
- offset
< RXRPC_JUMBO_SUBPKTLEN
)
371 if (flags
& RXRPC_LAST_PACKET
)
373 offset
+= RXRPC_JUMBO_DATALEN
;
374 if (skb_copy_bits(skb
, offset
, &flags
, 1) < 0)
376 offset
+= sizeof(struct rxrpc_jumbo_header
);
377 } while (flags
& RXRPC_JUMBO_PACKET
);
379 sp
->nr_jumbo
= nr_jumbo
;
387 * Handle reception of a duplicate packet.
389 * We have to take care to avoid an attack here whereby we're given a series of
390 * jumbograms, each with a sequence number one before the preceding one and
391 * filled up to maximum UDP size. If they never send us the first packet in
392 * the sequence, they can cause us to have to hold on to around 2MiB of kernel
393 * space until the call times out.
395 * We limit the space usage by only accepting three duplicate jumbo packets per
396 * call. After that, we tell the other side we're no longer accepting jumbos
397 * (that information is encoded in the ACK packet).
399 static void rxrpc_input_dup_data(struct rxrpc_call
*call
, rxrpc_seq_t seq
,
400 u8 annotation
, bool *_jumbo_bad
)
402 /* Discard normal packets that are duplicates. */
406 /* Skip jumbo subpackets that are duplicates. When we've had three or
407 * more partially duplicate jumbo packets, we refuse to take any more
408 * jumbos for this call.
411 call
->nr_jumbo_bad
++;
417 * Process a DATA packet, adding the packet to the Rx ring.
419 static void rxrpc_input_data(struct rxrpc_call
*call
, struct sk_buff
*skb
,
422 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
423 enum rxrpc_call_state state
;
424 unsigned int offset
= sizeof(struct rxrpc_wire_header
);
426 rxrpc_serial_t serial
= sp
->hdr
.serial
, ack_serial
= 0;
427 rxrpc_seq_t seq
= sp
->hdr
.seq
, hard_ack
;
428 bool immediate_ack
= false, jumbo_bad
= false, queued
;
430 u8 ack
= 0, flags
, annotation
= 0;
432 _enter("{%u,%u},{%u,%u}",
433 call
->rx_hard_ack
, call
->rx_top
, skb
->len
, seq
);
435 _proto("Rx DATA %%%u { #%u f=%02x }",
436 sp
->hdr
.serial
, seq
, sp
->hdr
.flags
);
438 state
= READ_ONCE(call
->state
);
439 if (state
>= RXRPC_CALL_COMPLETE
)
442 if (call
->state
== RXRPC_CALL_SERVER_RECV_REQUEST
) {
443 unsigned long timo
= READ_ONCE(call
->next_req_timo
);
444 unsigned long now
, expect_req_by
;
448 expect_req_by
= now
+ timo
;
449 WRITE_ONCE(call
->expect_req_by
, expect_req_by
);
450 rxrpc_reduce_call_timer(call
, expect_req_by
, now
,
451 rxrpc_timer_set_for_idle
);
455 /* Received data implicitly ACKs all of the request packets we sent
456 * when we're acting as a client.
458 if ((state
== RXRPC_CALL_CLIENT_SEND_REQUEST
||
459 state
== RXRPC_CALL_CLIENT_AWAIT_REPLY
) &&
460 !rxrpc_receiving_reply(call
))
463 call
->ackr_prev_seq
= seq
;
465 hard_ack
= READ_ONCE(call
->rx_hard_ack
);
466 if (after(seq
, hard_ack
+ call
->rx_winsize
)) {
467 ack
= RXRPC_ACK_EXCEEDS_WINDOW
;
472 flags
= sp
->hdr
.flags
;
473 if (flags
& RXRPC_JUMBO_PACKET
) {
474 if (call
->nr_jumbo_bad
> 3) {
475 ack
= RXRPC_ACK_NOSPACE
;
484 ix
= seq
& RXRPC_RXTX_BUFF_MASK
;
486 if (flags
& RXRPC_JUMBO_PACKET
)
487 len
= RXRPC_JUMBO_DATALEN
;
489 if (flags
& RXRPC_LAST_PACKET
) {
490 if (test_bit(RXRPC_CALL_RX_LAST
, &call
->flags
) &&
492 return rxrpc_proto_abort("LSN", call
, seq
);
494 if (test_bit(RXRPC_CALL_RX_LAST
, &call
->flags
) &&
495 after_eq(seq
, call
->rx_top
))
496 return rxrpc_proto_abort("LSA", call
, seq
);
499 trace_rxrpc_rx_data(call
, seq
, serial
, flags
, annotation
);
500 if (before_eq(seq
, hard_ack
)) {
501 ack
= RXRPC_ACK_DUPLICATE
;
506 if (flags
& RXRPC_REQUEST_ACK
&& !ack
) {
507 ack
= RXRPC_ACK_REQUESTED
;
511 if (call
->rxtx_buffer
[ix
]) {
512 rxrpc_input_dup_data(call
, seq
, annotation
, &jumbo_bad
);
513 if (ack
!= RXRPC_ACK_DUPLICATE
) {
514 ack
= RXRPC_ACK_DUPLICATE
;
517 immediate_ack
= true;
521 /* Queue the packet. We use a couple of memory barriers here as need
522 * to make sure that rx_top is perceived to be set after the buffer
523 * pointer and that the buffer pointer is set after the annotation and
526 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
527 * and also rxrpc_fill_out_ack().
529 rxrpc_get_skb(skb
, rxrpc_skb_rx_got
);
530 call
->rxtx_annotations
[ix
] = annotation
;
532 call
->rxtx_buffer
[ix
] = skb
;
533 if (after(seq
, call
->rx_top
)) {
534 smp_store_release(&call
->rx_top
, seq
);
535 } else if (before(seq
, call
->rx_top
)) {
536 /* Send an immediate ACK if we fill in a hole */
538 ack
= RXRPC_ACK_DELAY
;
541 immediate_ack
= true;
543 if (flags
& RXRPC_LAST_PACKET
) {
544 set_bit(RXRPC_CALL_RX_LAST
, &call
->flags
);
545 trace_rxrpc_receive(call
, rxrpc_receive_queue_last
, serial
, seq
);
547 trace_rxrpc_receive(call
, rxrpc_receive_queue
, serial
, seq
);
551 if (after_eq(seq
, call
->rx_expect_next
)) {
552 if (after(seq
, call
->rx_expect_next
)) {
553 _net("OOS %u > %u", seq
, call
->rx_expect_next
);
554 ack
= RXRPC_ACK_OUT_OF_SEQUENCE
;
557 call
->rx_expect_next
= seq
+ 1;
562 if (flags
& RXRPC_JUMBO_PACKET
) {
563 if (skb_copy_bits(skb
, offset
, &flags
, 1) < 0)
564 return rxrpc_proto_abort("XJF", call
, seq
);
565 offset
+= sizeof(struct rxrpc_jumbo_header
);
569 if (flags
& RXRPC_JUMBO_PACKET
)
570 annotation
|= RXRPC_RX_ANNO_JLAST
;
571 if (after(seq
, hard_ack
+ call
->rx_winsize
)) {
572 ack
= RXRPC_ACK_EXCEEDS_WINDOW
;
575 call
->nr_jumbo_bad
++;
581 _proto("Rx DATA Jumbo %%%u", serial
);
585 if (queued
&& flags
& RXRPC_LAST_PACKET
&& !ack
) {
586 ack
= RXRPC_ACK_DELAY
;
592 rxrpc_propose_ACK(call
, ack
, skew
, ack_serial
,
594 rxrpc_propose_ack_input_data
);
596 if (sp
->hdr
.seq
== READ_ONCE(call
->rx_hard_ack
) + 1)
597 rxrpc_notify_socket(call
);
602 * Process a requested ACK.
604 static void rxrpc_input_requested_ack(struct rxrpc_call
*call
,
606 rxrpc_serial_t orig_serial
,
607 rxrpc_serial_t ack_serial
)
609 struct rxrpc_skb_priv
*sp
;
614 for (ix
= 0; ix
< RXRPC_RXTX_BUFF_SIZE
; ix
++) {
615 skb
= call
->rxtx_buffer
[ix
];
620 if (sp
->hdr
.serial
!= orig_serial
)
623 sent_at
= skb
->tstamp
;
629 rxrpc_peer_add_rtt(call
, rxrpc_rtt_rx_requested_ack
,
630 orig_serial
, ack_serial
, sent_at
, resp_time
);
634 * Process the response to a ping that we sent to find out if we lost an ACK.
636 * If we got back a ping response that indicates a lower tx_top than what we
637 * had at the time of the ping transmission, we adjudge all the DATA packets
638 * sent between the response tx_top and the ping-time tx_top to have been lost.
640 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call
*call
)
642 rxrpc_seq_t top
, bottom
, seq
;
645 spin_lock_bh(&call
->lock
);
647 bottom
= call
->tx_hard_ack
+ 1;
648 top
= call
->acks_lost_top
;
649 if (before(bottom
, top
)) {
650 for (seq
= bottom
; before_eq(seq
, top
); seq
++) {
651 int ix
= seq
& RXRPC_RXTX_BUFF_MASK
;
652 u8 annotation
= call
->rxtx_annotations
[ix
];
653 u8 anno_type
= annotation
& RXRPC_TX_ANNO_MASK
;
655 if (anno_type
!= RXRPC_TX_ANNO_UNACK
)
657 annotation
&= ~RXRPC_TX_ANNO_MASK
;
658 annotation
|= RXRPC_TX_ANNO_RETRANS
;
659 call
->rxtx_annotations
[ix
] = annotation
;
664 spin_unlock_bh(&call
->lock
);
666 if (resend
&& !test_and_set_bit(RXRPC_CALL_EV_RESEND
, &call
->events
))
667 rxrpc_queue_call(call
);
671 * Process a ping response.
673 static void rxrpc_input_ping_response(struct rxrpc_call
*call
,
675 rxrpc_serial_t orig_serial
,
676 rxrpc_serial_t ack_serial
)
678 rxrpc_serial_t ping_serial
;
681 ping_time
= call
->ping_time
;
683 ping_serial
= call
->ping_serial
;
685 if (orig_serial
== call
->acks_lost_ping
)
686 rxrpc_input_check_for_lost_ack(call
);
688 if (!test_bit(RXRPC_CALL_PINGING
, &call
->flags
) ||
689 before(orig_serial
, ping_serial
))
691 clear_bit(RXRPC_CALL_PINGING
, &call
->flags
);
692 if (after(orig_serial
, ping_serial
))
695 rxrpc_peer_add_rtt(call
, rxrpc_rtt_rx_ping_response
,
696 orig_serial
, ack_serial
, ping_time
, resp_time
);
700 * Process the extra information that may be appended to an ACK packet
702 static void rxrpc_input_ackinfo(struct rxrpc_call
*call
, struct sk_buff
*skb
,
703 struct rxrpc_ackinfo
*ackinfo
)
705 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
706 struct rxrpc_peer
*peer
;
709 u32 rwind
= ntohl(ackinfo
->rwind
);
711 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
713 ntohl(ackinfo
->rxMTU
), ntohl(ackinfo
->maxMTU
),
714 rwind
, ntohl(ackinfo
->jumbo_max
));
716 if (call
->tx_winsize
!= rwind
) {
717 if (rwind
> RXRPC_RXTX_BUFF_SIZE
- 1)
718 rwind
= RXRPC_RXTX_BUFF_SIZE
- 1;
719 if (rwind
> call
->tx_winsize
)
721 trace_rxrpc_rx_rwind_change(call
, sp
->hdr
.serial
,
722 ntohl(ackinfo
->rwind
), wake
);
723 call
->tx_winsize
= rwind
;
726 if (call
->cong_ssthresh
> rwind
)
727 call
->cong_ssthresh
= rwind
;
729 mtu
= min(ntohl(ackinfo
->rxMTU
), ntohl(ackinfo
->maxMTU
));
732 if (mtu
< peer
->maxdata
) {
733 spin_lock_bh(&peer
->lock
);
735 peer
->mtu
= mtu
+ peer
->hdrsize
;
736 spin_unlock_bh(&peer
->lock
);
737 _net("Net MTU %u (maxdata %u)", peer
->mtu
, peer
->maxdata
);
741 wake_up(&call
->waitq
);
745 * Process individual soft ACKs.
747 * Each ACK in the array corresponds to one packet and can be either an ACK or
748 * a NAK. If we get find an explicitly NAK'd packet we resend immediately;
749 * packets that lie beyond the end of the ACK list are scheduled for resend by
750 * the timer on the basis that the peer might just not have processed them at
751 * the time the ACK was sent.
753 static void rxrpc_input_soft_acks(struct rxrpc_call
*call
, u8
*acks
,
754 rxrpc_seq_t seq
, int nr_acks
,
755 struct rxrpc_ack_summary
*summary
)
758 u8 annotation
, anno_type
;
760 for (; nr_acks
> 0; nr_acks
--, seq
++) {
761 ix
= seq
& RXRPC_RXTX_BUFF_MASK
;
762 annotation
= call
->rxtx_annotations
[ix
];
763 anno_type
= annotation
& RXRPC_TX_ANNO_MASK
;
764 annotation
&= ~RXRPC_TX_ANNO_MASK
;
766 case RXRPC_ACK_TYPE_ACK
:
768 if (anno_type
== RXRPC_TX_ANNO_ACK
)
770 summary
->nr_new_acks
++;
771 call
->rxtx_annotations
[ix
] =
772 RXRPC_TX_ANNO_ACK
| annotation
;
774 case RXRPC_ACK_TYPE_NACK
:
775 if (!summary
->nr_nacks
&&
776 call
->acks_lowest_nak
!= seq
) {
777 call
->acks_lowest_nak
= seq
;
778 summary
->new_low_nack
= true;
781 if (anno_type
== RXRPC_TX_ANNO_NAK
)
783 summary
->nr_new_nacks
++;
784 if (anno_type
== RXRPC_TX_ANNO_RETRANS
)
786 call
->rxtx_annotations
[ix
] =
787 RXRPC_TX_ANNO_NAK
| annotation
;
790 return rxrpc_proto_abort("SFT", call
, 0);
796 * Process an ACK packet.
798 * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
799 * in the ACK array. Anything before that is hard-ACK'd and may be discarded.
801 * A hard-ACK means that a packet has been processed and may be discarded; a
802 * soft-ACK means that the packet may be discarded and retransmission
803 * requested. A phase is complete when all packets are hard-ACK'd.
805 static void rxrpc_input_ack(struct rxrpc_call
*call
, struct sk_buff
*skb
,
808 struct rxrpc_ack_summary summary
= { 0 };
809 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
811 struct rxrpc_ackpacket ack
;
812 struct rxrpc_ackinfo info
;
813 u8 acks
[RXRPC_MAXACKS
];
815 rxrpc_serial_t acked_serial
;
816 rxrpc_seq_t first_soft_ack
, hard_ack
;
817 int nr_acks
, offset
, ioffset
;
821 offset
= sizeof(struct rxrpc_wire_header
);
822 if (skb_copy_bits(skb
, offset
, &buf
.ack
, sizeof(buf
.ack
)) < 0) {
823 _debug("extraction failure");
824 return rxrpc_proto_abort("XAK", call
, 0);
826 offset
+= sizeof(buf
.ack
);
828 acked_serial
= ntohl(buf
.ack
.serial
);
829 first_soft_ack
= ntohl(buf
.ack
.firstPacket
);
830 hard_ack
= first_soft_ack
- 1;
831 nr_acks
= buf
.ack
.nAcks
;
832 summary
.ack_reason
= (buf
.ack
.reason
< RXRPC_ACK__INVALID
?
833 buf
.ack
.reason
: RXRPC_ACK__INVALID
);
835 trace_rxrpc_rx_ack(call
, sp
->hdr
.serial
, acked_serial
,
836 first_soft_ack
, ntohl(buf
.ack
.previousPacket
),
837 summary
.ack_reason
, nr_acks
);
839 if (buf
.ack
.reason
== RXRPC_ACK_PING_RESPONSE
)
840 rxrpc_input_ping_response(call
, skb
->tstamp
, acked_serial
,
842 if (buf
.ack
.reason
== RXRPC_ACK_REQUESTED
)
843 rxrpc_input_requested_ack(call
, skb
->tstamp
, acked_serial
,
846 if (buf
.ack
.reason
== RXRPC_ACK_PING
) {
847 _proto("Rx ACK %%%u PING Request", sp
->hdr
.serial
);
848 rxrpc_propose_ACK(call
, RXRPC_ACK_PING_RESPONSE
,
849 skew
, sp
->hdr
.serial
, true, true,
850 rxrpc_propose_ack_respond_to_ping
);
851 } else if (sp
->hdr
.flags
& RXRPC_REQUEST_ACK
) {
852 rxrpc_propose_ACK(call
, RXRPC_ACK_REQUESTED
,
853 skew
, sp
->hdr
.serial
, true, true,
854 rxrpc_propose_ack_respond_to_ack
);
857 ioffset
= offset
+ nr_acks
+ 3;
858 if (skb
->len
>= ioffset
+ sizeof(buf
.info
)) {
859 if (skb_copy_bits(skb
, ioffset
, &buf
.info
, sizeof(buf
.info
)) < 0)
860 return rxrpc_proto_abort("XAI", call
, 0);
861 rxrpc_input_ackinfo(call
, skb
, &buf
.info
);
864 if (first_soft_ack
== 0)
865 return rxrpc_proto_abort("AK0", call
, 0);
867 /* Ignore ACKs unless we are or have just been transmitting. */
868 switch (READ_ONCE(call
->state
)) {
869 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
870 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
871 case RXRPC_CALL_SERVER_SEND_REPLY
:
872 case RXRPC_CALL_SERVER_AWAIT_ACK
:
878 /* Discard any out-of-order or duplicate ACKs. */
879 if (before_eq(sp
->hdr
.serial
, call
->acks_latest
)) {
880 _debug("discard ACK %d <= %d",
881 sp
->hdr
.serial
, call
->acks_latest
);
884 call
->acks_latest_ts
= skb
->tstamp
;
885 call
->acks_latest
= sp
->hdr
.serial
;
887 if (before(hard_ack
, call
->tx_hard_ack
) ||
888 after(hard_ack
, call
->tx_top
))
889 return rxrpc_proto_abort("AKW", call
, 0);
890 if (nr_acks
> call
->tx_top
- hard_ack
)
891 return rxrpc_proto_abort("AKN", call
, 0);
893 if (after(hard_ack
, call
->tx_hard_ack
))
894 rxrpc_rotate_tx_window(call
, hard_ack
, &summary
);
897 if (skb_copy_bits(skb
, offset
, buf
.acks
, nr_acks
) < 0)
898 return rxrpc_proto_abort("XSA", call
, 0);
899 rxrpc_input_soft_acks(call
, buf
.acks
, first_soft_ack
, nr_acks
,
903 if (test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
)) {
904 rxrpc_end_tx_phase(call
, false, "ETA");
908 if (call
->rxtx_annotations
[call
->tx_top
& RXRPC_RXTX_BUFF_MASK
] &
909 RXRPC_TX_ANNO_LAST
&&
910 summary
.nr_acks
== call
->tx_top
- hard_ack
&&
911 rxrpc_is_client_call(call
))
912 rxrpc_propose_ACK(call
, RXRPC_ACK_PING
, skew
, sp
->hdr
.serial
,
914 rxrpc_propose_ack_ping_for_lost_reply
);
916 return rxrpc_congestion_management(call
, skb
, &summary
, acked_serial
);
920 * Process an ACKALL packet.
922 static void rxrpc_input_ackall(struct rxrpc_call
*call
, struct sk_buff
*skb
)
924 struct rxrpc_ack_summary summary
= { 0 };
925 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
927 _proto("Rx ACKALL %%%u", sp
->hdr
.serial
);
929 rxrpc_rotate_tx_window(call
, call
->tx_top
, &summary
);
930 if (test_bit(RXRPC_CALL_TX_LAST
, &call
->flags
))
931 rxrpc_end_tx_phase(call
, false, "ETL");
935 * Process an ABORT packet directed at a call.
937 static void rxrpc_input_abort(struct rxrpc_call
*call
, struct sk_buff
*skb
)
939 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
941 u32 abort_code
= RX_CALL_DEAD
;
946 skb_copy_bits(skb
, sizeof(struct rxrpc_wire_header
),
947 &wtmp
, sizeof(wtmp
)) >= 0)
948 abort_code
= ntohl(wtmp
);
950 trace_rxrpc_rx_abort(call
, sp
->hdr
.serial
, abort_code
);
952 _proto("Rx ABORT %%%u { %x }", sp
->hdr
.serial
, abort_code
);
954 if (rxrpc_set_call_completion(call
, RXRPC_CALL_REMOTELY_ABORTED
,
955 abort_code
, -ECONNABORTED
))
956 rxrpc_notify_socket(call
);
960 * Process an incoming call packet.
962 static void rxrpc_input_call_packet(struct rxrpc_call
*call
,
963 struct sk_buff
*skb
, u16 skew
)
965 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
968 _enter("%p,%p", call
, skb
);
970 timo
= READ_ONCE(call
->next_rx_timo
);
972 unsigned long now
= jiffies
, expect_rx_by
;
974 expect_rx_by
= jiffies
+ timo
;
975 WRITE_ONCE(call
->expect_rx_by
, expect_rx_by
);
976 rxrpc_reduce_call_timer(call
, expect_rx_by
, now
,
977 rxrpc_timer_set_for_normal
);
980 switch (sp
->hdr
.type
) {
981 case RXRPC_PACKET_TYPE_DATA
:
982 rxrpc_input_data(call
, skb
, skew
);
985 case RXRPC_PACKET_TYPE_ACK
:
986 rxrpc_input_ack(call
, skb
, skew
);
989 case RXRPC_PACKET_TYPE_BUSY
:
990 _proto("Rx BUSY %%%u", sp
->hdr
.serial
);
992 /* Just ignore BUSY packets from the server; the retry and
993 * lifespan timers will take care of business. BUSY packets
994 * from the client don't make sense.
998 case RXRPC_PACKET_TYPE_ABORT
:
999 rxrpc_input_abort(call
, skb
);
1002 case RXRPC_PACKET_TYPE_ACKALL
:
1003 rxrpc_input_ackall(call
, skb
);
1014 * Handle a new call on a channel implicitly completing the preceding call on
1017 * TODO: If callNumber > call_id + 1, renegotiate security.
1019 static void rxrpc_input_implicit_end_call(struct rxrpc_connection
*conn
,
1020 struct rxrpc_call
*call
)
1022 switch (READ_ONCE(call
->state
)) {
1023 case RXRPC_CALL_SERVER_AWAIT_ACK
:
1024 rxrpc_call_completed(call
);
1026 case RXRPC_CALL_COMPLETE
:
1029 if (rxrpc_abort_call("IMP", call
, 0, RX_CALL_DEAD
, -ESHUTDOWN
)) {
1030 set_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
1031 rxrpc_queue_call(call
);
1036 trace_rxrpc_improper_term(call
);
1037 __rxrpc_disconnect_call(conn
, call
);
1038 rxrpc_notify_socket(call
);
1042 * post connection-level events to the connection
1043 * - this includes challenges, responses, some aborts and call terminal packet
1046 static void rxrpc_post_packet_to_conn(struct rxrpc_connection
*conn
,
1047 struct sk_buff
*skb
)
1049 _enter("%p,%p", conn
, skb
);
1051 skb_queue_tail(&conn
->rx_queue
, skb
);
1052 rxrpc_queue_conn(conn
);
1056 * post endpoint-level events to the local endpoint
1057 * - this includes debug and version messages
1059 static void rxrpc_post_packet_to_local(struct rxrpc_local
*local
,
1060 struct sk_buff
*skb
)
1062 _enter("%p,%p", local
, skb
);
1064 skb_queue_tail(&local
->event_queue
, skb
);
1065 rxrpc_queue_local(local
);
1069 * put a packet up for transport-level abort
1071 static void rxrpc_reject_packet(struct rxrpc_local
*local
, struct sk_buff
*skb
)
1073 CHECK_SLAB_OKAY(&local
->usage
);
1075 skb_queue_tail(&local
->reject_queue
, skb
);
1076 rxrpc_queue_local(local
);
1080 * Extract the wire header from a packet and translate the byte order.
1083 int rxrpc_extract_header(struct rxrpc_skb_priv
*sp
, struct sk_buff
*skb
)
1085 struct rxrpc_wire_header whdr
;
1087 /* dig out the RxRPC connection details */
1088 if (skb_copy_bits(skb
, 0, &whdr
, sizeof(whdr
)) < 0) {
1089 trace_rxrpc_rx_eproto(NULL
, sp
->hdr
.serial
,
1090 tracepoint_string("bad_hdr"));
1094 memset(sp
, 0, sizeof(*sp
));
1095 sp
->hdr
.epoch
= ntohl(whdr
.epoch
);
1096 sp
->hdr
.cid
= ntohl(whdr
.cid
);
1097 sp
->hdr
.callNumber
= ntohl(whdr
.callNumber
);
1098 sp
->hdr
.seq
= ntohl(whdr
.seq
);
1099 sp
->hdr
.serial
= ntohl(whdr
.serial
);
1100 sp
->hdr
.flags
= whdr
.flags
;
1101 sp
->hdr
.type
= whdr
.type
;
1102 sp
->hdr
.userStatus
= whdr
.userStatus
;
1103 sp
->hdr
.securityIndex
= whdr
.securityIndex
;
1104 sp
->hdr
._rsvd
= ntohs(whdr
._rsvd
);
1105 sp
->hdr
.serviceId
= ntohs(whdr
.serviceId
);
1110 * handle data received on the local endpoint
1111 * - may be called in interrupt context
1113 * The socket is locked by the caller and this prevents the socket from being
1114 * shut down and the local endpoint from going away, thus sk_user_data will not
1115 * be cleared until this function returns.
1117 void rxrpc_data_ready(struct sock
*udp_sk
)
1119 struct rxrpc_connection
*conn
;
1120 struct rxrpc_channel
*chan
;
1121 struct rxrpc_call
*call
;
1122 struct rxrpc_skb_priv
*sp
;
1123 struct rxrpc_local
*local
= udp_sk
->sk_user_data
;
1124 struct sk_buff
*skb
;
1125 unsigned int channel
;
1128 _enter("%p", udp_sk
);
1130 ASSERT(!irqs_disabled());
1132 skb
= skb_recv_udp(udp_sk
, 0, 1, &ret
);
1136 _debug("UDP socket error %d", ret
);
1140 rxrpc_new_skb(skb
, rxrpc_skb_rx_received
);
1142 _net("recv skb %p", skb
);
1144 /* we'll probably need to checksum it (didn't call sock_recvmsg) */
1145 if (skb_checksum_complete(skb
)) {
1146 rxrpc_free_skb(skb
, rxrpc_skb_rx_freed
);
1147 __UDP_INC_STATS(&init_net
, UDP_MIB_INERRORS
, 0);
1148 _leave(" [CSUM failed]");
1152 __UDP_INC_STATS(&init_net
, UDP_MIB_INDATAGRAMS
, 0);
1154 /* The UDP protocol already released all skb resources;
1155 * we are free to add our own data there.
1157 sp
= rxrpc_skb(skb
);
1159 /* dig out the RxRPC connection details */
1160 if (rxrpc_extract_header(sp
, skb
) < 0)
1163 if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS
)) {
1165 if ((lose
++ & 7) == 7) {
1166 trace_rxrpc_rx_lose(sp
);
1167 rxrpc_lose_skb(skb
, rxrpc_skb_rx_lost
);
1172 trace_rxrpc_rx_packet(sp
);
1174 _net("Rx RxRPC %s ep=%x call=%x:%x",
1175 sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
? "ToServer" : "ToClient",
1176 sp
->hdr
.epoch
, sp
->hdr
.cid
, sp
->hdr
.callNumber
);
1178 if (sp
->hdr
.type
>= RXRPC_N_PACKET_TYPES
||
1179 !((RXRPC_SUPPORTED_PACKET_TYPES
>> sp
->hdr
.type
) & 1)) {
1180 _proto("Rx Bad Packet Type %u", sp
->hdr
.type
);
1184 switch (sp
->hdr
.type
) {
1185 case RXRPC_PACKET_TYPE_VERSION
:
1186 rxrpc_post_packet_to_local(local
, skb
);
1189 case RXRPC_PACKET_TYPE_BUSY
:
1190 if (sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
)
1194 case RXRPC_PACKET_TYPE_DATA
:
1195 if (sp
->hdr
.callNumber
== 0)
1197 if (sp
->hdr
.flags
& RXRPC_JUMBO_PACKET
&&
1198 !rxrpc_validate_jumbo(skb
))
1205 conn
= rxrpc_find_connection_rcu(local
, skb
);
1207 if (sp
->hdr
.securityIndex
!= conn
->security_ix
)
1208 goto wrong_security
;
1210 if (sp
->hdr
.serviceId
!= conn
->service_id
) {
1211 if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE
, &conn
->flags
) ||
1212 conn
->service_id
!= conn
->params
.service_id
)
1214 conn
->service_id
= sp
->hdr
.serviceId
;
1217 if (sp
->hdr
.callNumber
== 0) {
1218 /* Connection-level packet */
1219 _debug("CONN %p {%d}", conn
, conn
->debug_id
);
1220 rxrpc_post_packet_to_conn(conn
, skb
);
1224 /* Note the serial number skew here */
1225 skew
= (int)sp
->hdr
.serial
- (int)conn
->hi_serial
;
1228 conn
->hi_serial
= sp
->hdr
.serial
;
1231 skew
= min(skew
, 65535);
1234 /* Call-bound packets are routed by connection channel. */
1235 channel
= sp
->hdr
.cid
& RXRPC_CHANNELMASK
;
1236 chan
= &conn
->channels
[channel
];
1238 /* Ignore really old calls */
1239 if (sp
->hdr
.callNumber
< chan
->last_call
)
1240 goto discard_unlock
;
1242 if (sp
->hdr
.callNumber
== chan
->last_call
) {
1243 /* For the previous service call, if completed successfully, we
1244 * discard all further packets.
1246 if (rxrpc_conn_is_service(conn
) &&
1247 (chan
->last_type
== RXRPC_PACKET_TYPE_ACK
||
1248 sp
->hdr
.type
== RXRPC_PACKET_TYPE_ABORT
))
1249 goto discard_unlock
;
1251 /* But otherwise we need to retransmit the final packet from
1252 * data cached in the connection record.
1254 rxrpc_post_packet_to_conn(conn
, skb
);
1258 call
= rcu_dereference(chan
->call
);
1260 if (sp
->hdr
.callNumber
> chan
->call_id
) {
1261 if (!(sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
)) {
1266 rxrpc_input_implicit_end_call(conn
, call
);
1270 if (call
&& sp
->hdr
.serviceId
!= call
->service_id
)
1271 call
->service_id
= sp
->hdr
.serviceId
;
1277 if (!call
|| atomic_read(&call
->usage
) == 0) {
1278 if (!(sp
->hdr
.type
& RXRPC_CLIENT_INITIATED
) ||
1279 sp
->hdr
.callNumber
== 0 ||
1280 sp
->hdr
.type
!= RXRPC_PACKET_TYPE_DATA
)
1281 goto bad_message_unlock
;
1282 if (sp
->hdr
.seq
!= 1)
1283 goto discard_unlock
;
1284 call
= rxrpc_new_incoming_call(local
, conn
, skb
);
1289 rxrpc_send_ping(call
, skb
, skew
);
1290 mutex_unlock(&call
->user_mutex
);
1293 rxrpc_input_call_packet(call
, skb
, skew
);
1294 goto discard_unlock
;
1299 rxrpc_free_skb(skb
, rxrpc_skb_rx_freed
);
1301 trace_rxrpc_rx_done(0, 0);
1310 trace_rxrpc_abort("SEC", sp
->hdr
.cid
, sp
->hdr
.callNumber
, sp
->hdr
.seq
,
1311 RXKADINCONSISTENCY
, EBADMSG
);
1312 skb
->priority
= RXKADINCONSISTENCY
;
1317 trace_rxrpc_abort("UPG", sp
->hdr
.cid
, sp
->hdr
.callNumber
, sp
->hdr
.seq
,
1318 RX_PROTOCOL_ERROR
, EBADMSG
);
1319 goto protocol_error
;
1324 trace_rxrpc_abort("BAD", sp
->hdr
.cid
, sp
->hdr
.callNumber
, sp
->hdr
.seq
,
1325 RX_PROTOCOL_ERROR
, EBADMSG
);
1327 skb
->priority
= RX_PROTOCOL_ERROR
;
1329 skb
->mark
= RXRPC_SKB_MARK_LOCAL_ABORT
;
1331 trace_rxrpc_rx_done(skb
->mark
, skb
->priority
);
1332 rxrpc_reject_packet(local
, skb
);
1333 _leave(" [badmsg]");