xref: /linux/net/rxrpc/input.c (revision 72bea132f3680ee51e7ed2cee62892b6f5121909)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Processing of received RxRPC packets
3  *
4  * Copyright (C) 2020 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 "ar-internal.h"
11 
12 static void rxrpc_proto_abort(struct rxrpc_call *call, rxrpc_seq_t seq,
13 			      enum rxrpc_abort_reason why)
14 {
15 	rxrpc_abort_call(call, seq, RX_PROTOCOL_ERROR, -EBADMSG, why);
16 }
17 
18 /*
19  * Do TCP-style congestion management [RFC 5681].
20  */
21 static void rxrpc_congestion_management(struct rxrpc_call *call,
22 					struct sk_buff *skb,
23 					struct rxrpc_ack_summary *summary,
24 					rxrpc_serial_t acked_serial)
25 {
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;
29 	bool resend = false;
30 
31 	summary->flight_size =
32 		(call->tx_top - call->acks_hard_ack) - summary->nr_acks;
33 
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);
38 		cwnd = 1;
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;
43 			cumulative_acks = 0;
44 		}
45 	}
46 
47 	cumulative_acks += summary->nr_new_acks;
48 	if (cumulative_acks > 255)
49 		cumulative_acks = 255;
50 
51 	summary->cwnd = call->cong_cwnd;
52 	summary->ssthresh = call->cong_ssthresh;
53 	summary->cumulative_acks = cumulative_acks;
54 	summary->dup_acks = call->cong_dup_acks;
55 
56 	switch (call->cong_mode) {
57 	case RXRPC_CALL_SLOW_START:
58 		if (summary->saw_nacks)
59 			goto packet_loss_detected;
60 		if (summary->cumulative_acks > 0)
61 			cwnd += 1;
62 		if (cwnd >= call->cong_ssthresh) {
63 			call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
64 			call->cong_tstamp = skb->tstamp;
65 		}
66 		goto out;
67 
68 	case RXRPC_CALL_CONGEST_AVOIDANCE:
69 		if (summary->saw_nacks)
70 			goto packet_loss_detected;
71 
72 		/* We analyse the number of packets that get ACK'd per RTT
73 		 * period and increase the window if we managed to fill it.
74 		 */
75 		if (call->peer->rtt_count == 0)
76 			goto out;
77 		if (ktime_before(skb->tstamp,
78 				 ktime_add_us(call->cong_tstamp,
79 					      call->peer->srtt_us >> 3)))
80 			goto out_no_clear_ca;
81 		change = rxrpc_cong_rtt_window_end;
82 		call->cong_tstamp = skb->tstamp;
83 		if (cumulative_acks >= cwnd)
84 			cwnd++;
85 		goto out;
86 
87 	case RXRPC_CALL_PACKET_LOSS:
88 		if (!summary->saw_nacks)
89 			goto resume_normality;
90 
91 		if (summary->new_low_nack) {
92 			change = rxrpc_cong_new_low_nack;
93 			call->cong_dup_acks = 1;
94 			if (call->cong_extra > 1)
95 				call->cong_extra = 1;
96 			goto send_extra_data;
97 		}
98 
99 		call->cong_dup_acks++;
100 		if (call->cong_dup_acks < 3)
101 			goto send_extra_data;
102 
103 		change = rxrpc_cong_begin_retransmission;
104 		call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
105 		call->cong_ssthresh = max_t(unsigned int,
106 					    summary->flight_size / 2, 2);
107 		cwnd = call->cong_ssthresh + 3;
108 		call->cong_extra = 0;
109 		call->cong_dup_acks = 0;
110 		resend = true;
111 		goto out;
112 
113 	case RXRPC_CALL_FAST_RETRANSMIT:
114 		if (!summary->new_low_nack) {
115 			if (summary->nr_new_acks == 0)
116 				cwnd += 1;
117 			call->cong_dup_acks++;
118 			if (call->cong_dup_acks == 2) {
119 				change = rxrpc_cong_retransmit_again;
120 				call->cong_dup_acks = 0;
121 				resend = true;
122 			}
123 		} else {
124 			change = rxrpc_cong_progress;
125 			cwnd = call->cong_ssthresh;
126 			if (!summary->saw_nacks)
127 				goto resume_normality;
128 		}
129 		goto out;
130 
131 	default:
132 		BUG();
133 		goto out;
134 	}
135 
136 resume_normality:
137 	change = rxrpc_cong_cleared_nacks;
138 	call->cong_dup_acks = 0;
139 	call->cong_extra = 0;
140 	call->cong_tstamp = skb->tstamp;
141 	if (cwnd < call->cong_ssthresh)
142 		call->cong_mode = RXRPC_CALL_SLOW_START;
143 	else
144 		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
145 out:
146 	cumulative_acks = 0;
147 out_no_clear_ca:
148 	if (cwnd >= RXRPC_TX_MAX_WINDOW)
149 		cwnd = RXRPC_TX_MAX_WINDOW;
150 	call->cong_cwnd = cwnd;
151 	call->cong_cumul_acks = cumulative_acks;
152 	summary->mode = call->cong_mode;
153 	trace_rxrpc_congest(call, summary, acked_serial, change);
154 	if (resend)
155 		rxrpc_resend(call, skb);
156 	return;
157 
158 packet_loss_detected:
159 	change = rxrpc_cong_saw_nack;
160 	call->cong_mode = RXRPC_CALL_PACKET_LOSS;
161 	call->cong_dup_acks = 0;
162 	goto send_extra_data;
163 
164 send_extra_data:
165 	/* Send some previously unsent DATA if we have some to advance the ACK
166 	 * state.
167 	 */
168 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
169 	    summary->nr_acks != call->tx_top - call->acks_hard_ack) {
170 		call->cong_extra++;
171 		wake_up(&call->waitq);
172 	}
173 	goto out_no_clear_ca;
174 }
175 
176 /*
177  * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
178  */
179 void rxrpc_congestion_degrade(struct rxrpc_call *call)
180 {
181 	ktime_t rtt, now;
182 
183 	if (call->cong_mode != RXRPC_CALL_SLOW_START &&
184 	    call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
185 		return;
186 	if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY)
187 		return;
188 
189 	rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
190 	now = ktime_get_real();
191 	if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
192 		return;
193 
194 	trace_rxrpc_reset_cwnd(call, now);
195 	rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
196 	call->tx_last_sent = now;
197 	call->cong_mode = RXRPC_CALL_SLOW_START;
198 	call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
199 				    call->cong_cwnd * 3 / 4);
200 	call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
201 }
202 
203 /*
204  * Apply a hard ACK by advancing the Tx window.
205  */
206 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
207 				   struct rxrpc_ack_summary *summary)
208 {
209 	struct rxrpc_txbuf *txb;
210 	bool rot_last = false;
211 
212 	list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
213 		if (before_eq(txb->seq, call->acks_hard_ack))
214 			continue;
215 		if (txb->flags & RXRPC_LAST_PACKET) {
216 			set_bit(RXRPC_CALL_TX_LAST, &call->flags);
217 			rot_last = true;
218 		}
219 		if (txb->seq == to)
220 			break;
221 	}
222 
223 	if (rot_last)
224 		set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
225 
226 	_enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
227 
228 	if (call->acks_lowest_nak == call->acks_hard_ack) {
229 		call->acks_lowest_nak = to;
230 	} else if (after(to, call->acks_lowest_nak)) {
231 		summary->new_low_nack = true;
232 		call->acks_lowest_nak = to;
233 	}
234 
235 	smp_store_release(&call->acks_hard_ack, to);
236 
237 	trace_rxrpc_txqueue(call, (rot_last ?
238 				   rxrpc_txqueue_rotate_last :
239 				   rxrpc_txqueue_rotate));
240 	wake_up(&call->waitq);
241 	return rot_last;
242 }
243 
244 /*
245  * End the transmission phase of a call.
246  *
247  * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
248  * or a final ACK packet.
249  */
250 static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
251 			       enum rxrpc_abort_reason abort_why)
252 {
253 	ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
254 
255 	call->resend_at = KTIME_MAX;
256 	trace_rxrpc_timer_can(call, rxrpc_timer_trace_resend);
257 
258 	if (unlikely(call->cong_last_nack)) {
259 		rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
260 		call->cong_last_nack = NULL;
261 	}
262 
263 	switch (__rxrpc_call_state(call)) {
264 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
265 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
266 		if (reply_begun) {
267 			rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
268 			trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
269 			break;
270 		}
271 
272 		rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
273 		trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
274 		break;
275 
276 	case RXRPC_CALL_SERVER_AWAIT_ACK:
277 		rxrpc_call_completed(call);
278 		trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
279 		break;
280 
281 	default:
282 		kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
283 		rxrpc_proto_abort(call, call->tx_top, abort_why);
284 		break;
285 	}
286 }
287 
288 /*
289  * Begin the reply reception phase of a call.
290  */
291 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
292 {
293 	struct rxrpc_ack_summary summary = { 0 };
294 	rxrpc_seq_t top = READ_ONCE(call->tx_top);
295 
296 	if (call->ackr_reason) {
297 		call->delay_ack_at = KTIME_MAX;
298 		trace_rxrpc_timer_can(call, rxrpc_timer_trace_delayed_ack);
299 	}
300 
301 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
302 		if (!rxrpc_rotate_tx_window(call, top, &summary)) {
303 			rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
304 			return false;
305 		}
306 	}
307 
308 	rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
309 	return true;
310 }
311 
312 /*
313  * End the packet reception phase.
314  */
315 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
316 {
317 	rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
318 
319 	_enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
320 
321 	trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
322 
323 	switch (__rxrpc_call_state(call)) {
324 	case RXRPC_CALL_CLIENT_RECV_REPLY:
325 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
326 		rxrpc_call_completed(call);
327 		break;
328 
329 	case RXRPC_CALL_SERVER_RECV_REQUEST:
330 		rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
331 		call->expect_req_by = KTIME_MAX;
332 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
333 		break;
334 
335 	default:
336 		break;
337 	}
338 }
339 
340 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
341 					  rxrpc_seq_t window, rxrpc_seq_t wtop)
342 {
343 	call->ackr_window = window;
344 	call->ackr_wtop = wtop;
345 }
346 
347 /*
348  * Push a DATA packet onto the Rx queue.
349  */
350 static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
351 				   rxrpc_seq_t window, rxrpc_seq_t wtop,
352 				   enum rxrpc_receive_trace why)
353 {
354 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
355 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
356 
357 	__skb_queue_tail(&call->recvmsg_queue, skb);
358 	rxrpc_input_update_ack_window(call, window, wtop);
359 	trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
360 	if (last)
361 		rxrpc_end_rx_phase(call, sp->hdr.serial);
362 }
363 
364 /*
365  * Process a DATA packet.
366  */
367 static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
368 				 bool *_notify)
369 {
370 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
371 	struct sk_buff *oos;
372 	rxrpc_serial_t serial = sp->hdr.serial;
373 	unsigned int sack = call->ackr_sack_base;
374 	rxrpc_seq_t window = call->ackr_window;
375 	rxrpc_seq_t wtop = call->ackr_wtop;
376 	rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
377 	rxrpc_seq_t seq = sp->hdr.seq;
378 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
379 	int ack_reason = -1;
380 
381 	rxrpc_inc_stat(call->rxnet, stat_rx_data);
382 	if (sp->hdr.flags & RXRPC_REQUEST_ACK)
383 		rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
384 	if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
385 		rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
386 
387 	if (last) {
388 		if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
389 		    seq + 1 != wtop)
390 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
391 	} else {
392 		if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
393 		    after_eq(seq, wtop)) {
394 			pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
395 				call->debug_id, seq, window, wtop, wlimit);
396 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
397 		}
398 	}
399 
400 	if (after(seq, call->rx_highest_seq))
401 		call->rx_highest_seq = seq;
402 
403 	trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
404 
405 	if (before(seq, window)) {
406 		ack_reason = RXRPC_ACK_DUPLICATE;
407 		goto send_ack;
408 	}
409 	if (after(seq, wlimit)) {
410 		ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
411 		goto send_ack;
412 	}
413 
414 	/* Queue the packet. */
415 	if (seq == window) {
416 		if (sp->hdr.flags & RXRPC_REQUEST_ACK)
417 			ack_reason = RXRPC_ACK_REQUESTED;
418 		/* Send an immediate ACK if we fill in a hole */
419 		else if (!skb_queue_empty(&call->rx_oos_queue))
420 			ack_reason = RXRPC_ACK_DELAY;
421 		else
422 			call->ackr_nr_unacked++;
423 
424 		window++;
425 		if (after(window, wtop)) {
426 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_none);
427 			wtop = window;
428 		} else {
429 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_advance);
430 			sack = (sack + 1) % RXRPC_SACK_SIZE;
431 		}
432 
433 
434 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
435 
436 		spin_lock(&call->recvmsg_queue.lock);
437 		rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
438 		*_notify = true;
439 
440 		while ((oos = skb_peek(&call->rx_oos_queue))) {
441 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
442 
443 			if (after(osp->hdr.seq, window))
444 				break;
445 
446 			__skb_unlink(oos, &call->rx_oos_queue);
447 			last = osp->hdr.flags & RXRPC_LAST_PACKET;
448 			seq = osp->hdr.seq;
449 			call->ackr_sack_table[sack] = 0;
450 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_fill);
451 			sack = (sack + 1) % RXRPC_SACK_SIZE;
452 
453 			window++;
454 			rxrpc_input_queue_data(call, oos, window, wtop,
455 					       rxrpc_receive_queue_oos);
456 		}
457 
458 		spin_unlock(&call->recvmsg_queue.lock);
459 
460 		call->ackr_sack_base = sack;
461 	} else {
462 		unsigned int slot;
463 
464 		ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
465 
466 		slot = seq - window;
467 		sack = (sack + slot) % RXRPC_SACK_SIZE;
468 
469 		if (call->ackr_sack_table[sack % RXRPC_SACK_SIZE]) {
470 			ack_reason = RXRPC_ACK_DUPLICATE;
471 			goto send_ack;
472 		}
473 
474 		call->ackr_sack_table[sack % RXRPC_SACK_SIZE] |= 1;
475 		trace_rxrpc_sack(call, seq, sack, rxrpc_sack_oos);
476 
477 		if (after(seq + 1, wtop)) {
478 			wtop = seq + 1;
479 			rxrpc_input_update_ack_window(call, window, wtop);
480 		}
481 
482 		skb_queue_walk(&call->rx_oos_queue, oos) {
483 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
484 
485 			if (after(osp->hdr.seq, seq)) {
486 				rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
487 				__skb_queue_before(&call->rx_oos_queue, oos, skb);
488 				goto oos_queued;
489 			}
490 		}
491 
492 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
493 		__skb_queue_tail(&call->rx_oos_queue, skb);
494 	oos_queued:
495 		trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
496 				    sp->hdr.serial, sp->hdr.seq);
497 	}
498 
499 send_ack:
500 	if (ack_reason >= 0)
501 		rxrpc_send_ACK(call, ack_reason, serial,
502 			       rxrpc_propose_ack_input_data);
503 	else
504 		rxrpc_propose_delay_ACK(call, serial,
505 					rxrpc_propose_ack_input_data);
506 }
507 
508 /*
509  * Split a jumbo packet and file the bits separately.
510  */
511 static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
512 {
513 	struct rxrpc_jumbo_header jhdr;
514 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
515 	struct sk_buff *jskb;
516 	unsigned int offset = sizeof(struct rxrpc_wire_header);
517 	unsigned int len = skb->len - offset;
518 	bool notify = false;
519 
520 	while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
521 		if (len < RXRPC_JUMBO_SUBPKTLEN)
522 			goto protocol_error;
523 		if (sp->hdr.flags & RXRPC_LAST_PACKET)
524 			goto protocol_error;
525 		if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
526 				  &jhdr, sizeof(jhdr)) < 0)
527 			goto protocol_error;
528 
529 		jskb = skb_clone(skb, GFP_NOFS);
530 		if (!jskb) {
531 			kdebug("couldn't clone");
532 			return false;
533 		}
534 		rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
535 		jsp = rxrpc_skb(jskb);
536 		jsp->offset = offset;
537 		jsp->len = RXRPC_JUMBO_DATALEN;
538 		rxrpc_input_data_one(call, jskb, &notify);
539 		rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
540 
541 		sp->hdr.flags = jhdr.flags;
542 		sp->hdr._rsvd = ntohs(jhdr._rsvd);
543 		sp->hdr.seq++;
544 		sp->hdr.serial++;
545 		offset += RXRPC_JUMBO_SUBPKTLEN;
546 		len -= RXRPC_JUMBO_SUBPKTLEN;
547 	}
548 
549 	sp->offset = offset;
550 	sp->len    = len;
551 	rxrpc_input_data_one(call, skb, &notify);
552 	if (notify) {
553 		trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
554 		rxrpc_notify_socket(call);
555 	}
556 	return true;
557 
558 protocol_error:
559 	return false;
560 }
561 
562 /*
563  * Process a DATA packet, adding the packet to the Rx ring.  The caller's
564  * packet ref must be passed on or discarded.
565  */
566 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
567 {
568 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
569 	rxrpc_serial_t serial = sp->hdr.serial;
570 	rxrpc_seq_t seq0 = sp->hdr.seq;
571 
572 	_enter("{%x,%x,%x},{%u,%x}",
573 	       call->ackr_window, call->ackr_wtop, call->rx_highest_seq,
574 	       skb->len, seq0);
575 
576 	if (__rxrpc_call_is_complete(call))
577 		return;
578 
579 	switch (__rxrpc_call_state(call)) {
580 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
581 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
582 		/* Received data implicitly ACKs all of the request
583 		 * packets we sent when we're acting as a client.
584 		 */
585 		if (!rxrpc_receiving_reply(call))
586 			goto out_notify;
587 		break;
588 
589 	case RXRPC_CALL_SERVER_RECV_REQUEST: {
590 		unsigned long timo = READ_ONCE(call->next_req_timo);
591 
592 		if (timo) {
593 			ktime_t delay = ms_to_ktime(timo);
594 
595 			call->expect_req_by = ktime_add(ktime_get_real(), delay);
596 			trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_idle);
597 		}
598 		break;
599 	}
600 
601 	default:
602 		break;
603 	}
604 
605 	if (!rxrpc_input_split_jumbo(call, skb)) {
606 		rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
607 		goto out_notify;
608 	}
609 	return;
610 
611 out_notify:
612 	trace_rxrpc_notify_socket(call->debug_id, serial);
613 	rxrpc_notify_socket(call);
614 	_leave(" [queued]");
615 }
616 
617 /*
618  * See if there's a cached RTT probe to complete.
619  */
620 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
621 				     ktime_t resp_time,
622 				     rxrpc_serial_t acked_serial,
623 				     rxrpc_serial_t ack_serial,
624 				     enum rxrpc_rtt_rx_trace type)
625 {
626 	rxrpc_serial_t orig_serial;
627 	unsigned long avail;
628 	ktime_t sent_at;
629 	bool matched = false;
630 	int i;
631 
632 	avail = READ_ONCE(call->rtt_avail);
633 	smp_rmb(); /* Read avail bits before accessing data. */
634 
635 	for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
636 		if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
637 			continue;
638 
639 		sent_at = call->rtt_sent_at[i];
640 		orig_serial = call->rtt_serial[i];
641 
642 		if (orig_serial == acked_serial) {
643 			clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
644 			smp_mb(); /* Read data before setting avail bit */
645 			set_bit(i, &call->rtt_avail);
646 			rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
647 					   sent_at, resp_time);
648 			matched = true;
649 		}
650 
651 		/* If a later serial is being acked, then mark this slot as
652 		 * being available.
653 		 */
654 		if (after(acked_serial, orig_serial)) {
655 			trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
656 					   orig_serial, acked_serial, 0, 0);
657 			clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
658 			smp_wmb();
659 			set_bit(i, &call->rtt_avail);
660 		}
661 	}
662 
663 	if (!matched)
664 		trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
665 }
666 
667 /*
668  * Process the extra information that may be appended to an ACK packet
669  */
670 static void rxrpc_input_ack_trailer(struct rxrpc_call *call, struct sk_buff *skb,
671 				    struct rxrpc_acktrailer *trailer)
672 {
673 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
674 	struct rxrpc_peer *peer;
675 	unsigned int mtu;
676 	bool wake = false;
677 	u32 rwind = ntohl(trailer->rwind);
678 
679 	if (rwind > RXRPC_TX_MAX_WINDOW)
680 		rwind = RXRPC_TX_MAX_WINDOW;
681 	if (call->tx_winsize != rwind) {
682 		if (rwind > call->tx_winsize)
683 			wake = true;
684 		trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
685 		call->tx_winsize = rwind;
686 	}
687 
688 	if (call->cong_ssthresh > rwind)
689 		call->cong_ssthresh = rwind;
690 
691 	mtu = min(ntohl(trailer->maxMTU), ntohl(trailer->ifMTU));
692 
693 	peer = call->peer;
694 	if (mtu < peer->maxdata) {
695 		spin_lock(&peer->lock);
696 		peer->maxdata = mtu;
697 		peer->mtu = mtu + peer->hdrsize;
698 		spin_unlock(&peer->lock);
699 	}
700 
701 	if (wake)
702 		wake_up(&call->waitq);
703 }
704 
705 /*
706  * Determine how many nacks from the previous ACK have now been satisfied.
707  */
708 static rxrpc_seq_t rxrpc_input_check_prev_ack(struct rxrpc_call *call,
709 					      struct rxrpc_ack_summary *summary,
710 					      rxrpc_seq_t seq)
711 {
712 	struct sk_buff *skb = call->cong_last_nack;
713 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
714 	unsigned int i, new_acks = 0, retained_nacks = 0;
715 	rxrpc_seq_t old_seq = sp->ack.first_ack;
716 	u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
717 
718 	if (after_eq(seq, old_seq + sp->ack.nr_acks)) {
719 		summary->nr_new_acks += sp->ack.nr_nacks;
720 		summary->nr_new_acks += seq - (old_seq + sp->ack.nr_acks);
721 		summary->nr_retained_nacks = 0;
722 	} else if (seq == old_seq) {
723 		summary->nr_retained_nacks = sp->ack.nr_nacks;
724 	} else {
725 		for (i = 0; i < sp->ack.nr_acks; i++) {
726 			if (acks[i] == RXRPC_ACK_TYPE_NACK) {
727 				if (before(old_seq + i, seq))
728 					new_acks++;
729 				else
730 					retained_nacks++;
731 			}
732 		}
733 
734 		summary->nr_new_acks += new_acks;
735 		summary->nr_retained_nacks = retained_nacks;
736 	}
737 
738 	return old_seq + sp->ack.nr_acks;
739 }
740 
741 /*
742  * Process individual soft ACKs.
743  *
744  * Each ACK in the array corresponds to one packet and can be either an ACK or
745  * a NAK.  If we get find an explicitly NAK'd packet we resend immediately;
746  * packets that lie beyond the end of the ACK list are scheduled for resend by
747  * the timer on the basis that the peer might just not have processed them at
748  * the time the ACK was sent.
749  */
750 static void rxrpc_input_soft_acks(struct rxrpc_call *call,
751 				  struct rxrpc_ack_summary *summary,
752 				  struct sk_buff *skb,
753 				  rxrpc_seq_t seq,
754 				  rxrpc_seq_t since)
755 {
756 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
757 	unsigned int i, old_nacks = 0;
758 	rxrpc_seq_t lowest_nak = seq + sp->ack.nr_acks;
759 	u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
760 
761 	for (i = 0; i < sp->ack.nr_acks; i++) {
762 		if (acks[i] == RXRPC_ACK_TYPE_ACK) {
763 			summary->nr_acks++;
764 			if (after_eq(seq, since))
765 				summary->nr_new_acks++;
766 		} else {
767 			summary->saw_nacks = true;
768 			if (before(seq, since)) {
769 				/* Overlap with previous ACK */
770 				old_nacks++;
771 			} else {
772 				summary->nr_new_nacks++;
773 				sp->ack.nr_nacks++;
774 			}
775 
776 			if (before(seq, lowest_nak))
777 				lowest_nak = seq;
778 		}
779 		seq++;
780 	}
781 
782 	if (lowest_nak != call->acks_lowest_nak) {
783 		call->acks_lowest_nak = lowest_nak;
784 		summary->new_low_nack = true;
785 	}
786 
787 	/* We *can* have more nacks than we did - the peer is permitted to drop
788 	 * packets it has soft-acked and re-request them.  Further, it is
789 	 * possible for the nack distribution to change whilst the number of
790 	 * nacks stays the same or goes down.
791 	 */
792 	if (old_nacks < summary->nr_retained_nacks)
793 		summary->nr_new_acks += summary->nr_retained_nacks - old_nacks;
794 	summary->nr_retained_nacks = old_nacks;
795 }
796 
797 /*
798  * Return true if the ACK is valid - ie. it doesn't appear to have regressed
799  * with respect to the ack state conveyed by preceding ACKs.
800  */
801 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
802 			       rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
803 {
804 	rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
805 
806 	if (after(first_pkt, base))
807 		return true; /* The window advanced */
808 
809 	if (before(first_pkt, base))
810 		return false; /* firstPacket regressed */
811 
812 	if (after_eq(prev_pkt, call->acks_prev_seq))
813 		return true; /* previousPacket hasn't regressed. */
814 
815 	/* Some rx implementations put a serial number in previousPacket. */
816 	if (after_eq(prev_pkt, base + call->tx_winsize))
817 		return false;
818 	return true;
819 }
820 
821 /*
822  * Process an ACK packet.
823  *
824  * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
825  * in the ACK array.  Anything before that is hard-ACK'd and may be discarded.
826  *
827  * A hard-ACK means that a packet has been processed and may be discarded; a
828  * soft-ACK means that the packet may be discarded and retransmission
829  * requested.  A phase is complete when all packets are hard-ACK'd.
830  */
831 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
832 {
833 	struct rxrpc_ack_summary summary = { 0 };
834 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
835 	struct rxrpc_acktrailer trailer;
836 	rxrpc_serial_t ack_serial, acked_serial;
837 	rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt, since;
838 	int nr_acks, offset, ioffset;
839 
840 	_enter("");
841 
842 	offset = sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
843 
844 	ack_serial	= sp->hdr.serial;
845 	acked_serial	= sp->ack.acked_serial;
846 	first_soft_ack	= sp->ack.first_ack;
847 	prev_pkt	= sp->ack.prev_ack;
848 	nr_acks		= sp->ack.nr_acks;
849 	hard_ack	= first_soft_ack - 1;
850 	summary.ack_reason = (sp->ack.reason < RXRPC_ACK__INVALID ?
851 			      sp->ack.reason : RXRPC_ACK__INVALID);
852 
853 	trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
854 			   first_soft_ack, prev_pkt,
855 			   summary.ack_reason, nr_acks);
856 	rxrpc_inc_stat(call->rxnet, stat_rx_acks[summary.ack_reason]);
857 
858 	if (acked_serial != 0) {
859 		switch (summary.ack_reason) {
860 		case RXRPC_ACK_PING_RESPONSE:
861 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
862 						 rxrpc_rtt_rx_ping_response);
863 			break;
864 		case RXRPC_ACK_REQUESTED:
865 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
866 						 rxrpc_rtt_rx_requested_ack);
867 			break;
868 		default:
869 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
870 						 rxrpc_rtt_rx_other_ack);
871 			break;
872 		}
873 	}
874 
875 	/* If we get an EXCEEDS_WINDOW ACK from the server, it probably
876 	 * indicates that the client address changed due to NAT.  The server
877 	 * lost the call because it switched to a different peer.
878 	 */
879 	if (unlikely(summary.ack_reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
880 	    first_soft_ack == 1 &&
881 	    prev_pkt == 0 &&
882 	    rxrpc_is_client_call(call)) {
883 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
884 					  0, -ENETRESET);
885 		goto send_response;
886 	}
887 
888 	/* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
889 	 * indicate a change of address.  However, we can retransmit the call
890 	 * if we still have it buffered to the beginning.
891 	 */
892 	if (unlikely(summary.ack_reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
893 	    first_soft_ack == 1 &&
894 	    prev_pkt == 0 &&
895 	    call->acks_hard_ack == 0 &&
896 	    rxrpc_is_client_call(call)) {
897 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
898 					  0, -ENETRESET);
899 		goto send_response;
900 	}
901 
902 	/* Discard any out-of-order or duplicate ACKs (outside lock). */
903 	if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
904 		trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
905 					   first_soft_ack, call->acks_first_seq,
906 					   prev_pkt, call->acks_prev_seq);
907 		goto send_response;
908 	}
909 
910 	trailer.maxMTU = 0;
911 	ioffset = offset + nr_acks + 3;
912 	if (skb->len >= ioffset + sizeof(trailer) &&
913 	    skb_copy_bits(skb, ioffset, &trailer, sizeof(trailer)) < 0)
914 		return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_trailer);
915 
916 	if (nr_acks > 0)
917 		skb_condense(skb);
918 
919 	if (call->cong_last_nack) {
920 		since = rxrpc_input_check_prev_ack(call, &summary, first_soft_ack);
921 		rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
922 		call->cong_last_nack = NULL;
923 	} else {
924 		summary.nr_new_acks = first_soft_ack - call->acks_first_seq;
925 		call->acks_lowest_nak = first_soft_ack + nr_acks;
926 		since = first_soft_ack;
927 	}
928 
929 	call->acks_latest_ts = skb->tstamp;
930 	call->acks_first_seq = first_soft_ack;
931 	call->acks_prev_seq = prev_pkt;
932 
933 	switch (summary.ack_reason) {
934 	case RXRPC_ACK_PING:
935 		break;
936 	default:
937 		if (acked_serial && after(acked_serial, call->acks_highest_serial))
938 			call->acks_highest_serial = acked_serial;
939 		break;
940 	}
941 
942 	/* Parse rwind and mtu sizes if provided. */
943 	if (trailer.maxMTU)
944 		rxrpc_input_ack_trailer(call, skb, &trailer);
945 
946 	if (first_soft_ack == 0)
947 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
948 
949 	/* Ignore ACKs unless we are or have just been transmitting. */
950 	switch (__rxrpc_call_state(call)) {
951 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
952 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
953 	case RXRPC_CALL_SERVER_SEND_REPLY:
954 	case RXRPC_CALL_SERVER_AWAIT_ACK:
955 		break;
956 	default:
957 		goto send_response;
958 	}
959 
960 	if (before(hard_ack, call->acks_hard_ack) ||
961 	    after(hard_ack, call->tx_top))
962 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
963 	if (nr_acks > call->tx_top - hard_ack)
964 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
965 
966 	if (after(hard_ack, call->acks_hard_ack)) {
967 		if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
968 			rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
969 			goto send_response;
970 		}
971 	}
972 
973 	if (nr_acks > 0) {
974 		if (offset > (int)skb->len - nr_acks)
975 			return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
976 		rxrpc_input_soft_acks(call, &summary, skb, first_soft_ack, since);
977 		rxrpc_get_skb(skb, rxrpc_skb_get_last_nack);
978 		call->cong_last_nack = skb;
979 	}
980 
981 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
982 	    summary.nr_acks == call->tx_top - hard_ack &&
983 	    rxrpc_is_client_call(call))
984 		rxrpc_propose_ping(call, ack_serial,
985 				   rxrpc_propose_ack_ping_for_lost_reply);
986 
987 	rxrpc_congestion_management(call, skb, &summary, acked_serial);
988 
989 send_response:
990 	if (summary.ack_reason == RXRPC_ACK_PING)
991 		rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
992 			       rxrpc_propose_ack_respond_to_ping);
993 	else if (sp->hdr.flags & RXRPC_REQUEST_ACK)
994 		rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
995 			       rxrpc_propose_ack_respond_to_ack);
996 }
997 
998 /*
999  * Process an ACKALL packet.
1000  */
1001 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1002 {
1003 	struct rxrpc_ack_summary summary = { 0 };
1004 
1005 	if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1006 		rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
1007 }
1008 
1009 /*
1010  * Process an ABORT packet directed at a call.
1011  */
1012 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1013 {
1014 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1015 
1016 	trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
1017 
1018 	rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1019 				  skb->priority, -ECONNABORTED);
1020 }
1021 
1022 /*
1023  * Process an incoming call packet.
1024  */
1025 void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
1026 {
1027 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1028 	unsigned long timo;
1029 
1030 	_enter("%p,%p", call, skb);
1031 
1032 	if (sp->hdr.serviceId != call->dest_srx.srx_service)
1033 		call->dest_srx.srx_service = sp->hdr.serviceId;
1034 	if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1035 		call->rx_serial = sp->hdr.serial;
1036 	if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1037 		set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1038 
1039 	timo = READ_ONCE(call->next_rx_timo);
1040 	if (timo) {
1041 		ktime_t delay = ms_to_ktime(timo);
1042 
1043 		call->expect_rx_by = ktime_add(ktime_get_real(), delay);
1044 		trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_expect_rx);
1045 	}
1046 
1047 	switch (sp->hdr.type) {
1048 	case RXRPC_PACKET_TYPE_DATA:
1049 		return rxrpc_input_data(call, skb);
1050 
1051 	case RXRPC_PACKET_TYPE_ACK:
1052 		return rxrpc_input_ack(call, skb);
1053 
1054 	case RXRPC_PACKET_TYPE_BUSY:
1055 		/* Just ignore BUSY packets from the server; the retry and
1056 		 * lifespan timers will take care of business.  BUSY packets
1057 		 * from the client don't make sense.
1058 		 */
1059 		return;
1060 
1061 	case RXRPC_PACKET_TYPE_ABORT:
1062 		return rxrpc_input_abort(call, skb);
1063 
1064 	case RXRPC_PACKET_TYPE_ACKALL:
1065 		return rxrpc_input_ackall(call, skb);
1066 
1067 	default:
1068 		break;
1069 	}
1070 }
1071 
1072 /*
1073  * Handle a new service call on a channel implicitly completing the preceding
1074  * call on that channel.  This does not apply to client conns.
1075  *
1076  * TODO: If callNumber > call_id + 1, renegotiate security.
1077  */
1078 void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1079 {
1080 	switch (__rxrpc_call_state(call)) {
1081 	case RXRPC_CALL_SERVER_AWAIT_ACK:
1082 		rxrpc_call_completed(call);
1083 		fallthrough;
1084 	case RXRPC_CALL_COMPLETE:
1085 		break;
1086 	default:
1087 		rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1088 				 rxrpc_eproto_improper_term);
1089 		trace_rxrpc_improper_term(call);
1090 		break;
1091 	}
1092 
1093 	rxrpc_input_call_event(call, skb);
1094 }
1095