xref: /openbmc/linux/net/rxrpc/input.c (revision f20c7d91)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC packet reception
3  *
4  * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
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>
15 #include <linux/in.h>
16 #include <linux/in6.h>
17 #include <linux/icmp.h>
18 #include <linux/gfp.h>
19 #include <net/sock.h>
20 #include <net/af_rxrpc.h>
21 #include <net/ip.h>
22 #include <net/udp.h>
23 #include <net/net_namespace.h>
24 #include "ar-internal.h"
25 
26 static void rxrpc_proto_abort(const char *why,
27 			      struct rxrpc_call *call, rxrpc_seq_t seq)
28 {
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);
32 	}
33 }
34 
35 /*
36  * Do TCP-style congestion management [RFC 5681].
37  */
38 static void rxrpc_congestion_management(struct rxrpc_call *call,
39 					struct sk_buff *skb,
40 					struct rxrpc_ack_summary *summary,
41 					rxrpc_serial_t acked_serial)
42 {
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;
46 	bool resend = false;
47 
48 	summary->flight_size =
49 		(call->tx_top - call->tx_hard_ack) - summary->nr_acks;
50 
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);
55 		cwnd = 1;
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;
60 			cumulative_acks = 0;
61 		}
62 	}
63 
64 	cumulative_acks += summary->nr_new_acks;
65 	cumulative_acks += summary->nr_rot_new_acks;
66 	if (cumulative_acks > 255)
67 		cumulative_acks = 255;
68 
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;
74 
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)
80 			cwnd += 1;
81 		if (cwnd >= call->cong_ssthresh) {
82 			call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
83 			call->cong_tstamp = skb->tstamp;
84 		}
85 		goto out;
86 
87 	case RXRPC_CALL_CONGEST_AVOIDANCE:
88 		if (summary->nr_nacks > 0)
89 			goto packet_loss_detected;
90 
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.
93 		 */
94 		if (call->peer->rtt_count == 0)
95 			goto out;
96 		if (ktime_before(skb->tstamp,
97 				 ktime_add_us(call->cong_tstamp,
98 					      call->peer->srtt_us >> 3)))
99 			goto out_no_clear_ca;
100 		change = rxrpc_cong_rtt_window_end;
101 		call->cong_tstamp = skb->tstamp;
102 		if (cumulative_acks >= cwnd)
103 			cwnd++;
104 		goto out;
105 
106 	case RXRPC_CALL_PACKET_LOSS:
107 		if (summary->nr_nacks == 0)
108 			goto resume_normality;
109 
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;
116 		}
117 
118 		call->cong_dup_acks++;
119 		if (call->cong_dup_acks < 3)
120 			goto send_extra_data;
121 
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;
129 		resend = true;
130 		goto out;
131 
132 	case RXRPC_CALL_FAST_RETRANSMIT:
133 		if (!summary->new_low_nack) {
134 			if (summary->nr_new_acks == 0)
135 				cwnd += 1;
136 			call->cong_dup_acks++;
137 			if (call->cong_dup_acks == 2) {
138 				change = rxrpc_cong_retransmit_again;
139 				call->cong_dup_acks = 0;
140 				resend = true;
141 			}
142 		} else {
143 			change = rxrpc_cong_progress;
144 			cwnd = call->cong_ssthresh;
145 			if (summary->nr_nacks == 0)
146 				goto resume_normality;
147 		}
148 		goto out;
149 
150 	default:
151 		BUG();
152 		goto out;
153 	}
154 
155 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;
162 	else
163 		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
164 out:
165 	cumulative_acks = 0;
166 out_no_clear_ca:
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);
174 	return;
175 
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;
181 
182 send_extra_data:
183 	/* Send some previously unsent DATA if we have some to advance the ACK
184 	 * state.
185 	 */
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) {
189 		call->cong_extra++;
190 		wake_up(&call->waitq);
191 	}
192 	goto out_no_clear_ca;
193 }
194 
195 /*
196  * Apply a hard ACK by advancing the Tx window.
197  */
198 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
199 				   struct rxrpc_ack_summary *summary)
200 {
201 	struct sk_buff *skb, *list = NULL;
202 	bool rot_last = false;
203 	int ix;
204 	u8 annotation;
205 
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;
211 	}
212 
213 	spin_lock(&call->lock);
214 
215 	while (before(call->tx_hard_ack, to)) {
216 		call->tx_hard_ack++;
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;
223 		skb->next = list;
224 		list = skb;
225 
226 		if (annotation & RXRPC_TX_ANNO_LAST) {
227 			set_bit(RXRPC_CALL_TX_LAST, &call->flags);
228 			rot_last = true;
229 		}
230 		if ((annotation & RXRPC_TX_ANNO_MASK) != RXRPC_TX_ANNO_ACK)
231 			summary->nr_rot_new_acks++;
232 	}
233 
234 	spin_unlock(&call->lock);
235 
236 	trace_rxrpc_transmit(call, (rot_last ?
237 				    rxrpc_transmit_rotate_last :
238 				    rxrpc_transmit_rotate));
239 	wake_up(&call->waitq);
240 
241 	while (list) {
242 		skb = list;
243 		list = skb->next;
244 		skb_mark_not_on_list(skb);
245 		rxrpc_free_skb(skb, rxrpc_skb_freed);
246 	}
247 
248 	return rot_last;
249 }
250 
251 /*
252  * End the transmission phase of a call.
253  *
254  * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
255  * or a final ACK packet.
256  */
257 static bool rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
258 			       const char *abort_why)
259 {
260 	unsigned int state;
261 
262 	ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
263 
264 	write_lock(&call->state_lock);
265 
266 	state = call->state;
267 	switch (state) {
268 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
269 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
270 		if (reply_begun)
271 			call->state = state = RXRPC_CALL_CLIENT_RECV_REPLY;
272 		else
273 			call->state = state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
274 		break;
275 
276 	case RXRPC_CALL_SERVER_AWAIT_ACK:
277 		__rxrpc_call_completed(call);
278 		state = call->state;
279 		break;
280 
281 	default:
282 		goto bad_state;
283 	}
284 
285 	write_unlock(&call->state_lock);
286 	if (state == RXRPC_CALL_CLIENT_AWAIT_REPLY)
287 		trace_rxrpc_transmit(call, rxrpc_transmit_await_reply);
288 	else
289 		trace_rxrpc_transmit(call, rxrpc_transmit_end);
290 	_leave(" = ok");
291 	return true;
292 
293 bad_state:
294 	write_unlock(&call->state_lock);
295 	kdebug("end_tx %s", rxrpc_call_states[call->state]);
296 	rxrpc_proto_abort(abort_why, call, call->tx_top);
297 	return false;
298 }
299 
300 /*
301  * Begin the reply reception phase of a call.
302  */
303 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
304 {
305 	struct rxrpc_ack_summary summary = { 0 };
306 	unsigned long now, timo;
307 	rxrpc_seq_t top = READ_ONCE(call->tx_top);
308 
309 	if (call->ackr_reason) {
310 		spin_lock_bh(&call->lock);
311 		call->ackr_reason = 0;
312 		spin_unlock_bh(&call->lock);
313 		now = jiffies;
314 		timo = now + MAX_JIFFY_OFFSET;
315 		WRITE_ONCE(call->resend_at, timo);
316 		WRITE_ONCE(call->ack_at, timo);
317 		trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
318 	}
319 
320 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
321 		if (!rxrpc_rotate_tx_window(call, top, &summary)) {
322 			rxrpc_proto_abort("TXL", call, top);
323 			return false;
324 		}
325 	}
326 	if (!rxrpc_end_tx_phase(call, true, "ETD"))
327 		return false;
328 	call->tx_phase = false;
329 	return true;
330 }
331 
332 /*
333  * Scan a data packet to validate its structure and to work out how many
334  * subpackets it contains.
335  *
336  * A jumbo packet is a collection of consecutive packets glued together with
337  * little headers between that indicate how to change the initial header for
338  * each subpacket.
339  *
340  * RXRPC_JUMBO_PACKET must be set on all but the last subpacket - and all but
341  * the last are RXRPC_JUMBO_DATALEN in size.  The last subpacket may be of any
342  * size.
343  */
344 static bool rxrpc_validate_data(struct sk_buff *skb)
345 {
346 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
347 	unsigned int offset = sizeof(struct rxrpc_wire_header);
348 	unsigned int len = skb->len;
349 	u8 flags = sp->hdr.flags;
350 
351 	for (;;) {
352 		if (flags & RXRPC_REQUEST_ACK)
353 			__set_bit(sp->nr_subpackets, sp->rx_req_ack);
354 		sp->nr_subpackets++;
355 
356 		if (!(flags & RXRPC_JUMBO_PACKET))
357 			break;
358 
359 		if (len - offset < RXRPC_JUMBO_SUBPKTLEN)
360 			goto protocol_error;
361 		if (flags & RXRPC_LAST_PACKET)
362 			goto protocol_error;
363 		offset += RXRPC_JUMBO_DATALEN;
364 		if (skb_copy_bits(skb, offset, &flags, 1) < 0)
365 			goto protocol_error;
366 		offset += sizeof(struct rxrpc_jumbo_header);
367 	}
368 
369 	if (flags & RXRPC_LAST_PACKET)
370 		sp->rx_flags |= RXRPC_SKB_INCL_LAST;
371 	return true;
372 
373 protocol_error:
374 	return false;
375 }
376 
377 /*
378  * Handle reception of a duplicate packet.
379  *
380  * We have to take care to avoid an attack here whereby we're given a series of
381  * jumbograms, each with a sequence number one before the preceding one and
382  * filled up to maximum UDP size.  If they never send us the first packet in
383  * the sequence, they can cause us to have to hold on to around 2MiB of kernel
384  * space until the call times out.
385  *
386  * We limit the space usage by only accepting three duplicate jumbo packets per
387  * call.  After that, we tell the other side we're no longer accepting jumbos
388  * (that information is encoded in the ACK packet).
389  */
390 static void rxrpc_input_dup_data(struct rxrpc_call *call, rxrpc_seq_t seq,
391 				 bool is_jumbo, bool *_jumbo_bad)
392 {
393 	/* Discard normal packets that are duplicates. */
394 	if (is_jumbo)
395 		return;
396 
397 	/* Skip jumbo subpackets that are duplicates.  When we've had three or
398 	 * more partially duplicate jumbo packets, we refuse to take any more
399 	 * jumbos for this call.
400 	 */
401 	if (!*_jumbo_bad) {
402 		call->nr_jumbo_bad++;
403 		*_jumbo_bad = true;
404 	}
405 }
406 
407 /*
408  * Process a DATA packet, adding the packet to the Rx ring.  The caller's
409  * packet ref must be passed on or discarded.
410  */
411 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
412 {
413 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
414 	enum rxrpc_call_state state;
415 	unsigned int j, nr_subpackets;
416 	rxrpc_serial_t serial = sp->hdr.serial, ack_serial = 0;
417 	rxrpc_seq_t seq0 = sp->hdr.seq, hard_ack;
418 	bool immediate_ack = false, jumbo_bad = false;
419 	u8 ack = 0;
420 
421 	_enter("{%u,%u},{%u,%u}",
422 	       call->rx_hard_ack, call->rx_top, skb->len, seq0);
423 
424 	_proto("Rx DATA %%%u { #%u f=%02x n=%u }",
425 	       sp->hdr.serial, seq0, sp->hdr.flags, sp->nr_subpackets);
426 
427 	state = READ_ONCE(call->state);
428 	if (state >= RXRPC_CALL_COMPLETE) {
429 		rxrpc_free_skb(skb, rxrpc_skb_freed);
430 		return;
431 	}
432 
433 	if (call->state == RXRPC_CALL_SERVER_RECV_REQUEST) {
434 		unsigned long timo = READ_ONCE(call->next_req_timo);
435 		unsigned long now, expect_req_by;
436 
437 		if (timo) {
438 			now = jiffies;
439 			expect_req_by = now + timo;
440 			WRITE_ONCE(call->expect_req_by, expect_req_by);
441 			rxrpc_reduce_call_timer(call, expect_req_by, now,
442 						rxrpc_timer_set_for_idle);
443 		}
444 	}
445 
446 	spin_lock(&call->input_lock);
447 
448 	/* Received data implicitly ACKs all of the request packets we sent
449 	 * when we're acting as a client.
450 	 */
451 	if ((state == RXRPC_CALL_CLIENT_SEND_REQUEST ||
452 	     state == RXRPC_CALL_CLIENT_AWAIT_REPLY) &&
453 	    !rxrpc_receiving_reply(call))
454 		goto unlock;
455 
456 	call->ackr_prev_seq = seq0;
457 	hard_ack = READ_ONCE(call->rx_hard_ack);
458 
459 	nr_subpackets = sp->nr_subpackets;
460 	if (nr_subpackets > 1) {
461 		if (call->nr_jumbo_bad > 3) {
462 			ack = RXRPC_ACK_NOSPACE;
463 			ack_serial = serial;
464 			goto ack;
465 		}
466 	}
467 
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;
475 
476 		_proto("Rx DATA+%u %%%u { #%x t=%u l=%u }",
477 		     j, serial, seq, terminal, last);
478 
479 		if (last) {
480 			if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
481 			    seq != call->rx_top) {
482 				rxrpc_proto_abort("LSN", call, seq);
483 				goto unlock;
484 			}
485 		} else {
486 			if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
487 			    after_eq(seq, call->rx_top)) {
488 				rxrpc_proto_abort("LSA", call, seq);
489 				goto unlock;
490 			}
491 		}
492 
493 		flags = 0;
494 		if (last)
495 			flags |= RXRPC_LAST_PACKET;
496 		if (!terminal)
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);
501 
502 		if (before_eq(seq, hard_ack)) {
503 			ack = RXRPC_ACK_DUPLICATE;
504 			ack_serial = serial;
505 			continue;
506 		}
507 
508 		if (call->rxtx_buffer[ix]) {
509 			rxrpc_input_dup_data(call, seq, nr_subpackets > 1,
510 					     &jumbo_bad);
511 			if (ack != RXRPC_ACK_DUPLICATE) {
512 				ack = RXRPC_ACK_DUPLICATE;
513 				ack_serial = serial;
514 			}
515 			immediate_ack = true;
516 			continue;
517 		}
518 
519 		if (after(seq, hard_ack + call->rx_winsize)) {
520 			ack = RXRPC_ACK_EXCEEDS_WINDOW;
521 			ack_serial = serial;
522 			if (flags & RXRPC_JUMBO_PACKET) {
523 				if (!jumbo_bad) {
524 					call->nr_jumbo_bad++;
525 					jumbo_bad = true;
526 				}
527 			}
528 
529 			goto ack;
530 		}
531 
532 		if (flags & RXRPC_REQUEST_ACK && !ack) {
533 			ack = RXRPC_ACK_REQUESTED;
534 			ack_serial = serial;
535 		}
536 
537 		/* Queue the packet.  We use a couple of memory barriers here as need
538 		 * to make sure that rx_top is perceived to be set after the buffer
539 		 * pointer and that the buffer pointer is set after the annotation and
540 		 * the skb data.
541 		 *
542 		 * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
543 		 * and also rxrpc_fill_out_ack().
544 		 */
545 		if (!terminal)
546 			rxrpc_get_skb(skb, rxrpc_skb_got);
547 		call->rxtx_annotations[ix] = annotation;
548 		smp_wmb();
549 		call->rxtx_buffer[ix] = skb;
550 		if (after(seq, call->rx_top)) {
551 			smp_store_release(&call->rx_top, seq);
552 		} else if (before(seq, call->rx_top)) {
553 			/* Send an immediate ACK if we fill in a hole */
554 			if (!ack) {
555 				ack = RXRPC_ACK_DELAY;
556 				ack_serial = serial;
557 			}
558 			immediate_ack = true;
559 		}
560 
561 		if (terminal) {
562 			/* From this point on, we're not allowed to touch the
563 			 * packet any longer as its ref now belongs to the Rx
564 			 * ring.
565 			 */
566 			skb = NULL;
567 			sp = NULL;
568 		}
569 
570 		if (last) {
571 			set_bit(RXRPC_CALL_RX_LAST, &call->flags);
572 			if (!ack) {
573 				ack = RXRPC_ACK_DELAY;
574 				ack_serial = serial;
575 			}
576 			trace_rxrpc_receive(call, rxrpc_receive_queue_last, serial, seq);
577 		} else {
578 			trace_rxrpc_receive(call, rxrpc_receive_queue, serial, seq);
579 		}
580 
581 		if (after_eq(seq, call->rx_expect_next)) {
582 			if (after(seq, call->rx_expect_next)) {
583 				_net("OOS %u > %u", seq, call->rx_expect_next);
584 				ack = RXRPC_ACK_OUT_OF_SEQUENCE;
585 				ack_serial = serial;
586 			}
587 			call->rx_expect_next = seq + 1;
588 		}
589 	}
590 
591 ack:
592 	if (ack)
593 		rxrpc_propose_ACK(call, ack, ack_serial,
594 				  immediate_ack, true,
595 				  rxrpc_propose_ack_input_data);
596 	else
597 		rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
598 				  false, true,
599 				  rxrpc_propose_ack_input_data);
600 
601 	trace_rxrpc_notify_socket(call->debug_id, serial);
602 	rxrpc_notify_socket(call);
603 
604 unlock:
605 	spin_unlock(&call->input_lock);
606 	rxrpc_free_skb(skb, rxrpc_skb_freed);
607 	_leave(" [queued]");
608 }
609 
610 /*
611  * Process a requested ACK.
612  */
613 static void rxrpc_input_requested_ack(struct rxrpc_call *call,
614 				      ktime_t resp_time,
615 				      rxrpc_serial_t orig_serial,
616 				      rxrpc_serial_t ack_serial)
617 {
618 	struct rxrpc_skb_priv *sp;
619 	struct sk_buff *skb;
620 	ktime_t sent_at;
621 	int ix;
622 
623 	for (ix = 0; ix < RXRPC_RXTX_BUFF_SIZE; ix++) {
624 		skb = call->rxtx_buffer[ix];
625 		if (!skb)
626 			continue;
627 
628 		sent_at = skb->tstamp;
629 		smp_rmb(); /* Read timestamp before serial. */
630 		sp = rxrpc_skb(skb);
631 		if (sp->hdr.serial != orig_serial)
632 			continue;
633 		goto found;
634 	}
635 
636 	return;
637 
638 found:
639 	rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_requested_ack,
640 			   orig_serial, ack_serial, sent_at, resp_time);
641 }
642 
643 /*
644  * Process the response to a ping that we sent to find out if we lost an ACK.
645  *
646  * If we got back a ping response that indicates a lower tx_top than what we
647  * had at the time of the ping transmission, we adjudge all the DATA packets
648  * sent between the response tx_top and the ping-time tx_top to have been lost.
649  */
650 static void rxrpc_input_check_for_lost_ack(struct rxrpc_call *call)
651 {
652 	rxrpc_seq_t top, bottom, seq;
653 	bool resend = false;
654 
655 	spin_lock_bh(&call->lock);
656 
657 	bottom = call->tx_hard_ack + 1;
658 	top = call->acks_lost_top;
659 	if (before(bottom, top)) {
660 		for (seq = bottom; before_eq(seq, top); seq++) {
661 			int ix = seq & RXRPC_RXTX_BUFF_MASK;
662 			u8 annotation = call->rxtx_annotations[ix];
663 			u8 anno_type = annotation & RXRPC_TX_ANNO_MASK;
664 
665 			if (anno_type != RXRPC_TX_ANNO_UNACK)
666 				continue;
667 			annotation &= ~RXRPC_TX_ANNO_MASK;
668 			annotation |= RXRPC_TX_ANNO_RETRANS;
669 			call->rxtx_annotations[ix] = annotation;
670 			resend = true;
671 		}
672 	}
673 
674 	spin_unlock_bh(&call->lock);
675 
676 	if (resend && !test_and_set_bit(RXRPC_CALL_EV_RESEND, &call->events))
677 		rxrpc_queue_call(call);
678 }
679 
680 /*
681  * Process a ping response.
682  */
683 static void rxrpc_input_ping_response(struct rxrpc_call *call,
684 				      ktime_t resp_time,
685 				      rxrpc_serial_t orig_serial,
686 				      rxrpc_serial_t ack_serial)
687 {
688 	rxrpc_serial_t ping_serial;
689 	ktime_t ping_time;
690 
691 	ping_time = call->ping_time;
692 	smp_rmb();
693 	ping_serial = READ_ONCE(call->ping_serial);
694 
695 	if (orig_serial == call->acks_lost_ping)
696 		rxrpc_input_check_for_lost_ack(call);
697 
698 	if (before(orig_serial, ping_serial) ||
699 	    !test_and_clear_bit(RXRPC_CALL_PINGING, &call->flags))
700 		return;
701 	if (after(orig_serial, ping_serial))
702 		return;
703 
704 	rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_ping_response,
705 			   orig_serial, ack_serial, ping_time, resp_time);
706 }
707 
708 /*
709  * Process the extra information that may be appended to an ACK packet
710  */
711 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
712 				struct rxrpc_ackinfo *ackinfo)
713 {
714 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
715 	struct rxrpc_peer *peer;
716 	unsigned int mtu;
717 	bool wake = false;
718 	u32 rwind = ntohl(ackinfo->rwind);
719 
720 	_proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
721 	       sp->hdr.serial,
722 	       ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU),
723 	       rwind, ntohl(ackinfo->jumbo_max));
724 
725 	if (call->tx_winsize != rwind) {
726 		if (rwind > RXRPC_RXTX_BUFF_SIZE - 1)
727 			rwind = RXRPC_RXTX_BUFF_SIZE - 1;
728 		if (rwind > call->tx_winsize)
729 			wake = true;
730 		trace_rxrpc_rx_rwind_change(call, sp->hdr.serial,
731 					    ntohl(ackinfo->rwind), wake);
732 		call->tx_winsize = rwind;
733 	}
734 
735 	if (call->cong_ssthresh > rwind)
736 		call->cong_ssthresh = rwind;
737 
738 	mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
739 
740 	peer = call->peer;
741 	if (mtu < peer->maxdata) {
742 		spin_lock_bh(&peer->lock);
743 		peer->maxdata = mtu;
744 		peer->mtu = mtu + peer->hdrsize;
745 		spin_unlock_bh(&peer->lock);
746 		_net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
747 	}
748 
749 	if (wake)
750 		wake_up(&call->waitq);
751 }
752 
753 /*
754  * Process individual soft ACKs.
755  *
756  * Each ACK in the array corresponds to one packet and can be either an ACK or
757  * a NAK.  If we get find an explicitly NAK'd packet we resend immediately;
758  * packets that lie beyond the end of the ACK list are scheduled for resend by
759  * the timer on the basis that the peer might just not have processed them at
760  * the time the ACK was sent.
761  */
762 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
763 				  rxrpc_seq_t seq, int nr_acks,
764 				  struct rxrpc_ack_summary *summary)
765 {
766 	int ix;
767 	u8 annotation, anno_type;
768 
769 	for (; nr_acks > 0; nr_acks--, seq++) {
770 		ix = seq & RXRPC_RXTX_BUFF_MASK;
771 		annotation = call->rxtx_annotations[ix];
772 		anno_type = annotation & RXRPC_TX_ANNO_MASK;
773 		annotation &= ~RXRPC_TX_ANNO_MASK;
774 		switch (*acks++) {
775 		case RXRPC_ACK_TYPE_ACK:
776 			summary->nr_acks++;
777 			if (anno_type == RXRPC_TX_ANNO_ACK)
778 				continue;
779 			summary->nr_new_acks++;
780 			call->rxtx_annotations[ix] =
781 				RXRPC_TX_ANNO_ACK | annotation;
782 			break;
783 		case RXRPC_ACK_TYPE_NACK:
784 			if (!summary->nr_nacks &&
785 			    call->acks_lowest_nak != seq) {
786 				call->acks_lowest_nak = seq;
787 				summary->new_low_nack = true;
788 			}
789 			summary->nr_nacks++;
790 			if (anno_type == RXRPC_TX_ANNO_NAK)
791 				continue;
792 			summary->nr_new_nacks++;
793 			if (anno_type == RXRPC_TX_ANNO_RETRANS)
794 				continue;
795 			call->rxtx_annotations[ix] =
796 				RXRPC_TX_ANNO_NAK | annotation;
797 			break;
798 		default:
799 			return rxrpc_proto_abort("SFT", call, 0);
800 		}
801 	}
802 }
803 
804 /*
805  * Return true if the ACK is valid - ie. it doesn't appear to have regressed
806  * with respect to the ack state conveyed by preceding ACKs.
807  */
808 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
809 			       rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
810 {
811 	rxrpc_seq_t base = READ_ONCE(call->ackr_first_seq);
812 
813 	if (after(first_pkt, base))
814 		return true; /* The window advanced */
815 
816 	if (before(first_pkt, base))
817 		return false; /* firstPacket regressed */
818 
819 	if (after_eq(prev_pkt, call->ackr_prev_seq))
820 		return true; /* previousPacket hasn't regressed. */
821 
822 	/* Some rx implementations put a serial number in previousPacket. */
823 	if (after_eq(prev_pkt, base + call->tx_winsize))
824 		return false;
825 	return true;
826 }
827 
828 /*
829  * Process an ACK packet.
830  *
831  * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
832  * in the ACK array.  Anything before that is hard-ACK'd and may be discarded.
833  *
834  * A hard-ACK means that a packet has been processed and may be discarded; a
835  * soft-ACK means that the packet may be discarded and retransmission
836  * requested.  A phase is complete when all packets are hard-ACK'd.
837  */
838 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
839 {
840 	struct rxrpc_ack_summary summary = { 0 };
841 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
842 	union {
843 		struct rxrpc_ackpacket ack;
844 		struct rxrpc_ackinfo info;
845 		u8 acks[RXRPC_MAXACKS];
846 	} buf;
847 	rxrpc_serial_t acked_serial;
848 	rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
849 	int nr_acks, offset, ioffset;
850 
851 	_enter("");
852 
853 	offset = sizeof(struct rxrpc_wire_header);
854 	if (skb_copy_bits(skb, offset, &buf.ack, sizeof(buf.ack)) < 0) {
855 		_debug("extraction failure");
856 		return rxrpc_proto_abort("XAK", call, 0);
857 	}
858 	offset += sizeof(buf.ack);
859 
860 	acked_serial = ntohl(buf.ack.serial);
861 	first_soft_ack = ntohl(buf.ack.firstPacket);
862 	prev_pkt = ntohl(buf.ack.previousPacket);
863 	hard_ack = first_soft_ack - 1;
864 	nr_acks = buf.ack.nAcks;
865 	summary.ack_reason = (buf.ack.reason < RXRPC_ACK__INVALID ?
866 			      buf.ack.reason : RXRPC_ACK__INVALID);
867 
868 	trace_rxrpc_rx_ack(call, sp->hdr.serial, acked_serial,
869 			   first_soft_ack, prev_pkt,
870 			   summary.ack_reason, nr_acks);
871 
872 	if (buf.ack.reason == RXRPC_ACK_PING_RESPONSE)
873 		rxrpc_input_ping_response(call, skb->tstamp, acked_serial,
874 					  sp->hdr.serial);
875 	if (buf.ack.reason == RXRPC_ACK_REQUESTED)
876 		rxrpc_input_requested_ack(call, skb->tstamp, acked_serial,
877 					  sp->hdr.serial);
878 
879 	if (buf.ack.reason == RXRPC_ACK_PING) {
880 		_proto("Rx ACK %%%u PING Request", sp->hdr.serial);
881 		rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
882 				  sp->hdr.serial, true, true,
883 				  rxrpc_propose_ack_respond_to_ping);
884 	} else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
885 		rxrpc_propose_ACK(call, RXRPC_ACK_REQUESTED,
886 				  sp->hdr.serial, true, true,
887 				  rxrpc_propose_ack_respond_to_ack);
888 	}
889 
890 	/* Discard any out-of-order or duplicate ACKs (outside lock). */
891 	if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
892 		trace_rxrpc_rx_discard_ack(call->debug_id, sp->hdr.serial,
893 					   first_soft_ack, call->ackr_first_seq,
894 					   prev_pkt, call->ackr_prev_seq);
895 		return;
896 	}
897 
898 	buf.info.rxMTU = 0;
899 	ioffset = offset + nr_acks + 3;
900 	if (skb->len >= ioffset + sizeof(buf.info) &&
901 	    skb_copy_bits(skb, ioffset, &buf.info, sizeof(buf.info)) < 0)
902 		return rxrpc_proto_abort("XAI", call, 0);
903 
904 	spin_lock(&call->input_lock);
905 
906 	/* Discard any out-of-order or duplicate ACKs (inside lock). */
907 	if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
908 		trace_rxrpc_rx_discard_ack(call->debug_id, sp->hdr.serial,
909 					   first_soft_ack, call->ackr_first_seq,
910 					   prev_pkt, call->ackr_prev_seq);
911 		goto out;
912 	}
913 	call->acks_latest_ts = skb->tstamp;
914 
915 	call->ackr_first_seq = first_soft_ack;
916 	call->ackr_prev_seq = prev_pkt;
917 
918 	/* Parse rwind and mtu sizes if provided. */
919 	if (buf.info.rxMTU)
920 		rxrpc_input_ackinfo(call, skb, &buf.info);
921 
922 	if (first_soft_ack == 0) {
923 		rxrpc_proto_abort("AK0", call, 0);
924 		goto out;
925 	}
926 
927 	/* Ignore ACKs unless we are or have just been transmitting. */
928 	switch (READ_ONCE(call->state)) {
929 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
930 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
931 	case RXRPC_CALL_SERVER_SEND_REPLY:
932 	case RXRPC_CALL_SERVER_AWAIT_ACK:
933 		break;
934 	default:
935 		goto out;
936 	}
937 
938 	if (before(hard_ack, call->tx_hard_ack) ||
939 	    after(hard_ack, call->tx_top)) {
940 		rxrpc_proto_abort("AKW", call, 0);
941 		goto out;
942 	}
943 	if (nr_acks > call->tx_top - hard_ack) {
944 		rxrpc_proto_abort("AKN", call, 0);
945 		goto out;
946 	}
947 
948 	if (after(hard_ack, call->tx_hard_ack)) {
949 		if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
950 			rxrpc_end_tx_phase(call, false, "ETA");
951 			goto out;
952 		}
953 	}
954 
955 	if (nr_acks > 0) {
956 		if (skb_copy_bits(skb, offset, buf.acks, nr_acks) < 0) {
957 			rxrpc_proto_abort("XSA", call, 0);
958 			goto out;
959 		}
960 		rxrpc_input_soft_acks(call, buf.acks, first_soft_ack, nr_acks,
961 				      &summary);
962 	}
963 
964 	if (call->rxtx_annotations[call->tx_top & RXRPC_RXTX_BUFF_MASK] &
965 	    RXRPC_TX_ANNO_LAST &&
966 	    summary.nr_acks == call->tx_top - hard_ack &&
967 	    rxrpc_is_client_call(call))
968 		rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
969 				  false, true,
970 				  rxrpc_propose_ack_ping_for_lost_reply);
971 
972 	rxrpc_congestion_management(call, skb, &summary, acked_serial);
973 out:
974 	spin_unlock(&call->input_lock);
975 }
976 
977 /*
978  * Process an ACKALL packet.
979  */
980 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
981 {
982 	struct rxrpc_ack_summary summary = { 0 };
983 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
984 
985 	_proto("Rx ACKALL %%%u", sp->hdr.serial);
986 
987 	spin_lock(&call->input_lock);
988 
989 	if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
990 		rxrpc_end_tx_phase(call, false, "ETL");
991 
992 	spin_unlock(&call->input_lock);
993 }
994 
995 /*
996  * Process an ABORT packet directed at a call.
997  */
998 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
999 {
1000 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1001 	__be32 wtmp;
1002 	u32 abort_code = RX_CALL_DEAD;
1003 
1004 	_enter("");
1005 
1006 	if (skb->len >= 4 &&
1007 	    skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
1008 			  &wtmp, sizeof(wtmp)) >= 0)
1009 		abort_code = ntohl(wtmp);
1010 
1011 	trace_rxrpc_rx_abort(call, sp->hdr.serial, abort_code);
1012 
1013 	_proto("Rx ABORT %%%u { %x }", sp->hdr.serial, abort_code);
1014 
1015 	rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1016 				  abort_code, -ECONNABORTED);
1017 }
1018 
1019 /*
1020  * Process an incoming call packet.
1021  */
1022 static void rxrpc_input_call_packet(struct rxrpc_call *call,
1023 				    struct sk_buff *skb)
1024 {
1025 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1026 	unsigned long timo;
1027 
1028 	_enter("%p,%p", call, skb);
1029 
1030 	timo = READ_ONCE(call->next_rx_timo);
1031 	if (timo) {
1032 		unsigned long now = jiffies, expect_rx_by;
1033 
1034 		expect_rx_by = now + timo;
1035 		WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1036 		rxrpc_reduce_call_timer(call, expect_rx_by, now,
1037 					rxrpc_timer_set_for_normal);
1038 	}
1039 
1040 	switch (sp->hdr.type) {
1041 	case RXRPC_PACKET_TYPE_DATA:
1042 		rxrpc_input_data(call, skb);
1043 		goto no_free;
1044 
1045 	case RXRPC_PACKET_TYPE_ACK:
1046 		rxrpc_input_ack(call, skb);
1047 		break;
1048 
1049 	case RXRPC_PACKET_TYPE_BUSY:
1050 		_proto("Rx BUSY %%%u", sp->hdr.serial);
1051 
1052 		/* Just ignore BUSY packets from the server; the retry and
1053 		 * lifespan timers will take care of business.  BUSY packets
1054 		 * from the client don't make sense.
1055 		 */
1056 		break;
1057 
1058 	case RXRPC_PACKET_TYPE_ABORT:
1059 		rxrpc_input_abort(call, skb);
1060 		break;
1061 
1062 	case RXRPC_PACKET_TYPE_ACKALL:
1063 		rxrpc_input_ackall(call, skb);
1064 		break;
1065 
1066 	default:
1067 		break;
1068 	}
1069 
1070 	rxrpc_free_skb(skb, rxrpc_skb_freed);
1071 no_free:
1072 	_leave("");
1073 }
1074 
1075 /*
1076  * Handle a new service call on a channel implicitly completing the preceding
1077  * call on that channel.  This does not apply to client conns.
1078  *
1079  * TODO: If callNumber > call_id + 1, renegotiate security.
1080  */
1081 static void rxrpc_input_implicit_end_call(struct rxrpc_sock *rx,
1082 					  struct rxrpc_connection *conn,
1083 					  struct rxrpc_call *call)
1084 {
1085 	switch (READ_ONCE(call->state)) {
1086 	case RXRPC_CALL_SERVER_AWAIT_ACK:
1087 		rxrpc_call_completed(call);
1088 		/* Fall through */
1089 	case RXRPC_CALL_COMPLETE:
1090 		break;
1091 	default:
1092 		if (rxrpc_abort_call("IMP", call, 0, RX_CALL_DEAD, -ESHUTDOWN)) {
1093 			set_bit(RXRPC_CALL_EV_ABORT, &call->events);
1094 			rxrpc_queue_call(call);
1095 		}
1096 		trace_rxrpc_improper_term(call);
1097 		break;
1098 	}
1099 
1100 	spin_lock(&rx->incoming_lock);
1101 	__rxrpc_disconnect_call(conn, call);
1102 	spin_unlock(&rx->incoming_lock);
1103 }
1104 
1105 /*
1106  * post connection-level events to the connection
1107  * - this includes challenges, responses, some aborts and call terminal packet
1108  *   retransmission.
1109  */
1110 static void rxrpc_post_packet_to_conn(struct rxrpc_connection *conn,
1111 				      struct sk_buff *skb)
1112 {
1113 	_enter("%p,%p", conn, skb);
1114 
1115 	skb_queue_tail(&conn->rx_queue, skb);
1116 	rxrpc_queue_conn(conn);
1117 }
1118 
1119 /*
1120  * post endpoint-level events to the local endpoint
1121  * - this includes debug and version messages
1122  */
1123 static void rxrpc_post_packet_to_local(struct rxrpc_local *local,
1124 				       struct sk_buff *skb)
1125 {
1126 	_enter("%p,%p", local, skb);
1127 
1128 	if (rxrpc_get_local_maybe(local)) {
1129 		skb_queue_tail(&local->event_queue, skb);
1130 		rxrpc_queue_local(local);
1131 	} else {
1132 		rxrpc_free_skb(skb, rxrpc_skb_freed);
1133 	}
1134 }
1135 
1136 /*
1137  * put a packet up for transport-level abort
1138  */
1139 static void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb)
1140 {
1141 	CHECK_SLAB_OKAY(&local->usage);
1142 
1143 	if (rxrpc_get_local_maybe(local)) {
1144 		skb_queue_tail(&local->reject_queue, skb);
1145 		rxrpc_queue_local(local);
1146 	} else {
1147 		rxrpc_free_skb(skb, rxrpc_skb_freed);
1148 	}
1149 }
1150 
1151 /*
1152  * Extract the wire header from a packet and translate the byte order.
1153  */
1154 static noinline
1155 int rxrpc_extract_header(struct rxrpc_skb_priv *sp, struct sk_buff *skb)
1156 {
1157 	struct rxrpc_wire_header whdr;
1158 
1159 	/* dig out the RxRPC connection details */
1160 	if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0) {
1161 		trace_rxrpc_rx_eproto(NULL, sp->hdr.serial,
1162 				      tracepoint_string("bad_hdr"));
1163 		return -EBADMSG;
1164 	}
1165 
1166 	memset(sp, 0, sizeof(*sp));
1167 	sp->hdr.epoch		= ntohl(whdr.epoch);
1168 	sp->hdr.cid		= ntohl(whdr.cid);
1169 	sp->hdr.callNumber	= ntohl(whdr.callNumber);
1170 	sp->hdr.seq		= ntohl(whdr.seq);
1171 	sp->hdr.serial		= ntohl(whdr.serial);
1172 	sp->hdr.flags		= whdr.flags;
1173 	sp->hdr.type		= whdr.type;
1174 	sp->hdr.userStatus	= whdr.userStatus;
1175 	sp->hdr.securityIndex	= whdr.securityIndex;
1176 	sp->hdr._rsvd		= ntohs(whdr._rsvd);
1177 	sp->hdr.serviceId	= ntohs(whdr.serviceId);
1178 	return 0;
1179 }
1180 
1181 /*
1182  * handle data received on the local endpoint
1183  * - may be called in interrupt context
1184  *
1185  * [!] Note that as this is called from the encap_rcv hook, the socket is not
1186  * held locked by the caller and nothing prevents sk_user_data on the UDP from
1187  * being cleared in the middle of processing this function.
1188  *
1189  * Called with the RCU read lock held from the IP layer via UDP.
1190  */
1191 int rxrpc_input_packet(struct sock *udp_sk, struct sk_buff *skb)
1192 {
1193 	struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
1194 	struct rxrpc_connection *conn;
1195 	struct rxrpc_channel *chan;
1196 	struct rxrpc_call *call = NULL;
1197 	struct rxrpc_skb_priv *sp;
1198 	struct rxrpc_peer *peer = NULL;
1199 	struct rxrpc_sock *rx = NULL;
1200 	unsigned int channel;
1201 
1202 	_enter("%p", udp_sk);
1203 
1204 	if (unlikely(!local)) {
1205 		kfree_skb(skb);
1206 		return 0;
1207 	}
1208 	if (skb->tstamp == 0)
1209 		skb->tstamp = ktime_get_real();
1210 
1211 	rxrpc_new_skb(skb, rxrpc_skb_received);
1212 
1213 	skb_pull(skb, sizeof(struct udphdr));
1214 
1215 	/* The UDP protocol already released all skb resources;
1216 	 * we are free to add our own data there.
1217 	 */
1218 	sp = rxrpc_skb(skb);
1219 
1220 	/* dig out the RxRPC connection details */
1221 	if (rxrpc_extract_header(sp, skb) < 0)
1222 		goto bad_message;
1223 
1224 	if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
1225 		static int lose;
1226 		if ((lose++ & 7) == 7) {
1227 			trace_rxrpc_rx_lose(sp);
1228 			rxrpc_free_skb(skb, rxrpc_skb_lost);
1229 			return 0;
1230 		}
1231 	}
1232 
1233 	if (skb->tstamp == 0)
1234 		skb->tstamp = ktime_get_real();
1235 	trace_rxrpc_rx_packet(sp);
1236 
1237 	switch (sp->hdr.type) {
1238 	case RXRPC_PACKET_TYPE_VERSION:
1239 		if (rxrpc_to_client(sp))
1240 			goto discard;
1241 		rxrpc_post_packet_to_local(local, skb);
1242 		goto out;
1243 
1244 	case RXRPC_PACKET_TYPE_BUSY:
1245 		if (rxrpc_to_server(sp))
1246 			goto discard;
1247 		/* Fall through */
1248 	case RXRPC_PACKET_TYPE_ACK:
1249 	case RXRPC_PACKET_TYPE_ACKALL:
1250 		if (sp->hdr.callNumber == 0)
1251 			goto bad_message;
1252 		/* Fall through */
1253 	case RXRPC_PACKET_TYPE_ABORT:
1254 		break;
1255 
1256 	case RXRPC_PACKET_TYPE_DATA:
1257 		if (sp->hdr.callNumber == 0 ||
1258 		    sp->hdr.seq == 0)
1259 			goto bad_message;
1260 		if (!rxrpc_validate_data(skb))
1261 			goto bad_message;
1262 
1263 		/* Unshare the packet so that it can be modified for in-place
1264 		 * decryption.
1265 		 */
1266 		if (sp->hdr.securityIndex != 0) {
1267 			struct sk_buff *nskb = skb_unshare(skb, GFP_ATOMIC);
1268 			if (!nskb) {
1269 				rxrpc_eaten_skb(skb, rxrpc_skb_unshared_nomem);
1270 				goto out;
1271 			}
1272 
1273 			if (nskb != skb) {
1274 				rxrpc_eaten_skb(skb, rxrpc_skb_received);
1275 				skb = nskb;
1276 				rxrpc_new_skb(skb, rxrpc_skb_unshared);
1277 				sp = rxrpc_skb(skb);
1278 			}
1279 		}
1280 		break;
1281 
1282 	case RXRPC_PACKET_TYPE_CHALLENGE:
1283 		if (rxrpc_to_server(sp))
1284 			goto discard;
1285 		break;
1286 	case RXRPC_PACKET_TYPE_RESPONSE:
1287 		if (rxrpc_to_client(sp))
1288 			goto discard;
1289 		break;
1290 
1291 		/* Packet types 9-11 should just be ignored. */
1292 	case RXRPC_PACKET_TYPE_PARAMS:
1293 	case RXRPC_PACKET_TYPE_10:
1294 	case RXRPC_PACKET_TYPE_11:
1295 		goto discard;
1296 
1297 	default:
1298 		_proto("Rx Bad Packet Type %u", sp->hdr.type);
1299 		goto bad_message;
1300 	}
1301 
1302 	if (sp->hdr.serviceId == 0)
1303 		goto bad_message;
1304 
1305 	if (rxrpc_to_server(sp)) {
1306 		/* Weed out packets to services we're not offering.  Packets
1307 		 * that would begin a call are explicitly rejected and the rest
1308 		 * are just discarded.
1309 		 */
1310 		rx = rcu_dereference(local->service);
1311 		if (!rx || (sp->hdr.serviceId != rx->srx.srx_service &&
1312 			    sp->hdr.serviceId != rx->second_service)) {
1313 			if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA &&
1314 			    sp->hdr.seq == 1)
1315 				goto unsupported_service;
1316 			goto discard;
1317 		}
1318 	}
1319 
1320 	conn = rxrpc_find_connection_rcu(local, skb, &peer);
1321 	if (conn) {
1322 		if (sp->hdr.securityIndex != conn->security_ix)
1323 			goto wrong_security;
1324 
1325 		if (sp->hdr.serviceId != conn->service_id) {
1326 			int old_id;
1327 
1328 			if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
1329 				goto reupgrade;
1330 			old_id = cmpxchg(&conn->service_id, conn->params.service_id,
1331 					 sp->hdr.serviceId);
1332 
1333 			if (old_id != conn->params.service_id &&
1334 			    old_id != sp->hdr.serviceId)
1335 				goto reupgrade;
1336 		}
1337 
1338 		if (sp->hdr.callNumber == 0) {
1339 			/* Connection-level packet */
1340 			_debug("CONN %p {%d}", conn, conn->debug_id);
1341 			rxrpc_post_packet_to_conn(conn, skb);
1342 			goto out;
1343 		}
1344 
1345 		if ((int)sp->hdr.serial - (int)conn->hi_serial > 0)
1346 			conn->hi_serial = sp->hdr.serial;
1347 
1348 		/* Call-bound packets are routed by connection channel. */
1349 		channel = sp->hdr.cid & RXRPC_CHANNELMASK;
1350 		chan = &conn->channels[channel];
1351 
1352 		/* Ignore really old calls */
1353 		if (sp->hdr.callNumber < chan->last_call)
1354 			goto discard;
1355 
1356 		if (sp->hdr.callNumber == chan->last_call) {
1357 			if (chan->call ||
1358 			    sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
1359 				goto discard;
1360 
1361 			/* For the previous service call, if completed
1362 			 * successfully, we discard all further packets.
1363 			 */
1364 			if (rxrpc_conn_is_service(conn) &&
1365 			    chan->last_type == RXRPC_PACKET_TYPE_ACK)
1366 				goto discard;
1367 
1368 			/* But otherwise we need to retransmit the final packet
1369 			 * from data cached in the connection record.
1370 			 */
1371 			if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
1372 				trace_rxrpc_rx_data(chan->call_debug_id,
1373 						    sp->hdr.seq,
1374 						    sp->hdr.serial,
1375 						    sp->hdr.flags, 0);
1376 			rxrpc_post_packet_to_conn(conn, skb);
1377 			goto out;
1378 		}
1379 
1380 		call = rcu_dereference(chan->call);
1381 
1382 		if (sp->hdr.callNumber > chan->call_id) {
1383 			if (rxrpc_to_client(sp))
1384 				goto reject_packet;
1385 			if (call)
1386 				rxrpc_input_implicit_end_call(rx, conn, call);
1387 			call = NULL;
1388 		}
1389 
1390 		if (call) {
1391 			if (sp->hdr.serviceId != call->service_id)
1392 				call->service_id = sp->hdr.serviceId;
1393 			if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1394 				call->rx_serial = sp->hdr.serial;
1395 			if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1396 				set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1397 		}
1398 	}
1399 
1400 	if (!call || atomic_read(&call->usage) == 0) {
1401 		if (rxrpc_to_client(sp) ||
1402 		    sp->hdr.type != RXRPC_PACKET_TYPE_DATA)
1403 			goto bad_message;
1404 		if (sp->hdr.seq != 1)
1405 			goto discard;
1406 		call = rxrpc_new_incoming_call(local, rx, skb);
1407 		if (!call)
1408 			goto reject_packet;
1409 	}
1410 
1411 	/* Process a call packet; this either discards or passes on the ref
1412 	 * elsewhere.
1413 	 */
1414 	rxrpc_input_call_packet(call, skb);
1415 	goto out;
1416 
1417 discard:
1418 	rxrpc_free_skb(skb, rxrpc_skb_freed);
1419 out:
1420 	trace_rxrpc_rx_done(0, 0);
1421 	return 0;
1422 
1423 wrong_security:
1424 	trace_rxrpc_abort(0, "SEC", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1425 			  RXKADINCONSISTENCY, EBADMSG);
1426 	skb->priority = RXKADINCONSISTENCY;
1427 	goto post_abort;
1428 
1429 unsupported_service:
1430 	trace_rxrpc_abort(0, "INV", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1431 			  RX_INVALID_OPERATION, EOPNOTSUPP);
1432 	skb->priority = RX_INVALID_OPERATION;
1433 	goto post_abort;
1434 
1435 reupgrade:
1436 	trace_rxrpc_abort(0, "UPG", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1437 			  RX_PROTOCOL_ERROR, EBADMSG);
1438 	goto protocol_error;
1439 
1440 bad_message:
1441 	trace_rxrpc_abort(0, "BAD", sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
1442 			  RX_PROTOCOL_ERROR, EBADMSG);
1443 protocol_error:
1444 	skb->priority = RX_PROTOCOL_ERROR;
1445 post_abort:
1446 	skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
1447 reject_packet:
1448 	trace_rxrpc_rx_done(skb->mark, skb->priority);
1449 	rxrpc_reject_packet(local, skb);
1450 	_leave(" [badmsg]");
1451 	return 0;
1452 }
1453