xref: /openbmc/linux/net/rxrpc/input.c (revision 95973a9b)
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 (test_bit(RXRPC_TXBUF_LAST, &txb->flags)) {
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 	if (unlikely(call->cong_last_nack)) {
256 		rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
257 		call->cong_last_nack = NULL;
258 	}
259 
260 	switch (__rxrpc_call_state(call)) {
261 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
262 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
263 		if (reply_begun) {
264 			rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
265 			trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
266 			break;
267 		}
268 
269 		rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
270 		trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
271 		break;
272 
273 	case RXRPC_CALL_SERVER_AWAIT_ACK:
274 		rxrpc_call_completed(call);
275 		trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
276 		break;
277 
278 	default:
279 		kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
280 		rxrpc_proto_abort(call, call->tx_top, abort_why);
281 		break;
282 	}
283 }
284 
285 /*
286  * Begin the reply reception phase of a call.
287  */
288 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
289 {
290 	struct rxrpc_ack_summary summary = { 0 };
291 	unsigned long now, timo;
292 	rxrpc_seq_t top = READ_ONCE(call->tx_top);
293 
294 	if (call->ackr_reason) {
295 		now = jiffies;
296 		timo = now + MAX_JIFFY_OFFSET;
297 
298 		WRITE_ONCE(call->delay_ack_at, timo);
299 		trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
300 	}
301 
302 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
303 		if (!rxrpc_rotate_tx_window(call, top, &summary)) {
304 			rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
305 			return false;
306 		}
307 	}
308 
309 	rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
310 	return true;
311 }
312 
313 /*
314  * End the packet reception phase.
315  */
316 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
317 {
318 	rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
319 
320 	_enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
321 
322 	trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
323 
324 	switch (__rxrpc_call_state(call)) {
325 	case RXRPC_CALL_CLIENT_RECV_REPLY:
326 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
327 		rxrpc_call_completed(call);
328 		break;
329 
330 	case RXRPC_CALL_SERVER_RECV_REQUEST:
331 		rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
332 		call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
333 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
334 		break;
335 
336 	default:
337 		break;
338 	}
339 }
340 
341 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
342 					  rxrpc_seq_t window, rxrpc_seq_t wtop)
343 {
344 	call->ackr_window = window;
345 	call->ackr_wtop = wtop;
346 }
347 
348 /*
349  * Push a DATA packet onto the Rx queue.
350  */
351 static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
352 				   rxrpc_seq_t window, rxrpc_seq_t wtop,
353 				   enum rxrpc_receive_trace why)
354 {
355 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
356 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
357 
358 	__skb_queue_tail(&call->recvmsg_queue, skb);
359 	rxrpc_input_update_ack_window(call, window, wtop);
360 	trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
361 	if (last)
362 		rxrpc_end_rx_phase(call, sp->hdr.serial);
363 }
364 
365 /*
366  * Process a DATA packet.
367  */
368 static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
369 				 bool *_notify)
370 {
371 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
372 	struct sk_buff *oos;
373 	rxrpc_serial_t serial = sp->hdr.serial;
374 	unsigned int sack = call->ackr_sack_base;
375 	rxrpc_seq_t window = call->ackr_window;
376 	rxrpc_seq_t wtop = call->ackr_wtop;
377 	rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
378 	rxrpc_seq_t seq = sp->hdr.seq;
379 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
380 	int ack_reason = -1;
381 
382 	rxrpc_inc_stat(call->rxnet, stat_rx_data);
383 	if (sp->hdr.flags & RXRPC_REQUEST_ACK)
384 		rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
385 	if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
386 		rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
387 
388 	if (last) {
389 		if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
390 		    seq + 1 != wtop)
391 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
392 	} else {
393 		if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
394 		    after_eq(seq, wtop)) {
395 			pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
396 				call->debug_id, seq, window, wtop, wlimit);
397 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
398 		}
399 	}
400 
401 	if (after(seq, call->rx_highest_seq))
402 		call->rx_highest_seq = seq;
403 
404 	trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
405 
406 	if (before(seq, window)) {
407 		ack_reason = RXRPC_ACK_DUPLICATE;
408 		goto send_ack;
409 	}
410 	if (after(seq, wlimit)) {
411 		ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
412 		goto send_ack;
413 	}
414 
415 	/* Queue the packet. */
416 	if (seq == window) {
417 		if (sp->hdr.flags & RXRPC_REQUEST_ACK)
418 			ack_reason = RXRPC_ACK_REQUESTED;
419 		/* Send an immediate ACK if we fill in a hole */
420 		else if (!skb_queue_empty(&call->rx_oos_queue))
421 			ack_reason = RXRPC_ACK_DELAY;
422 		else
423 			call->ackr_nr_unacked++;
424 
425 		window++;
426 		if (after(window, wtop)) {
427 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_none);
428 			wtop = window;
429 		} else {
430 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_advance);
431 			sack = (sack + 1) % RXRPC_SACK_SIZE;
432 		}
433 
434 
435 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
436 
437 		spin_lock(&call->recvmsg_queue.lock);
438 		rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
439 		*_notify = true;
440 
441 		while ((oos = skb_peek(&call->rx_oos_queue))) {
442 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
443 
444 			if (after(osp->hdr.seq, window))
445 				break;
446 
447 			__skb_unlink(oos, &call->rx_oos_queue);
448 			last = osp->hdr.flags & RXRPC_LAST_PACKET;
449 			seq = osp->hdr.seq;
450 			call->ackr_sack_table[sack] = 0;
451 			trace_rxrpc_sack(call, seq, sack, rxrpc_sack_fill);
452 			sack = (sack + 1) % RXRPC_SACK_SIZE;
453 
454 			window++;
455 			rxrpc_input_queue_data(call, oos, window, wtop,
456 					       rxrpc_receive_queue_oos);
457 		}
458 
459 		spin_unlock(&call->recvmsg_queue.lock);
460 
461 		call->ackr_sack_base = sack;
462 	} else {
463 		unsigned int slot;
464 
465 		ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
466 
467 		slot = seq - window;
468 		sack = (sack + slot) % RXRPC_SACK_SIZE;
469 
470 		if (call->ackr_sack_table[sack % RXRPC_SACK_SIZE]) {
471 			ack_reason = RXRPC_ACK_DUPLICATE;
472 			goto send_ack;
473 		}
474 
475 		call->ackr_sack_table[sack % RXRPC_SACK_SIZE] |= 1;
476 		trace_rxrpc_sack(call, seq, sack, rxrpc_sack_oos);
477 
478 		if (after(seq + 1, wtop)) {
479 			wtop = seq + 1;
480 			rxrpc_input_update_ack_window(call, window, wtop);
481 		}
482 
483 		skb_queue_walk(&call->rx_oos_queue, oos) {
484 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
485 
486 			if (after(osp->hdr.seq, seq)) {
487 				rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
488 				__skb_queue_before(&call->rx_oos_queue, oos, skb);
489 				goto oos_queued;
490 			}
491 		}
492 
493 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
494 		__skb_queue_tail(&call->rx_oos_queue, skb);
495 	oos_queued:
496 		trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
497 				    sp->hdr.serial, sp->hdr.seq);
498 	}
499 
500 send_ack:
501 	if (ack_reason >= 0)
502 		rxrpc_send_ACK(call, ack_reason, serial,
503 			       rxrpc_propose_ack_input_data);
504 	else
505 		rxrpc_propose_delay_ACK(call, serial,
506 					rxrpc_propose_ack_input_data);
507 }
508 
509 /*
510  * Split a jumbo packet and file the bits separately.
511  */
512 static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
513 {
514 	struct rxrpc_jumbo_header jhdr;
515 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
516 	struct sk_buff *jskb;
517 	unsigned int offset = sizeof(struct rxrpc_wire_header);
518 	unsigned int len = skb->len - offset;
519 	bool notify = false;
520 
521 	while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
522 		if (len < RXRPC_JUMBO_SUBPKTLEN)
523 			goto protocol_error;
524 		if (sp->hdr.flags & RXRPC_LAST_PACKET)
525 			goto protocol_error;
526 		if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
527 				  &jhdr, sizeof(jhdr)) < 0)
528 			goto protocol_error;
529 
530 		jskb = skb_clone(skb, GFP_NOFS);
531 		if (!jskb) {
532 			kdebug("couldn't clone");
533 			return false;
534 		}
535 		rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
536 		jsp = rxrpc_skb(jskb);
537 		jsp->offset = offset;
538 		jsp->len = RXRPC_JUMBO_DATALEN;
539 		rxrpc_input_data_one(call, jskb, &notify);
540 		rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
541 
542 		sp->hdr.flags = jhdr.flags;
543 		sp->hdr._rsvd = ntohs(jhdr._rsvd);
544 		sp->hdr.seq++;
545 		sp->hdr.serial++;
546 		offset += RXRPC_JUMBO_SUBPKTLEN;
547 		len -= RXRPC_JUMBO_SUBPKTLEN;
548 	}
549 
550 	sp->offset = offset;
551 	sp->len    = len;
552 	rxrpc_input_data_one(call, skb, &notify);
553 	if (notify) {
554 		trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
555 		rxrpc_notify_socket(call);
556 	}
557 	return true;
558 
559 protocol_error:
560 	return false;
561 }
562 
563 /*
564  * Process a DATA packet, adding the packet to the Rx ring.  The caller's
565  * packet ref must be passed on or discarded.
566  */
567 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
568 {
569 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
570 	rxrpc_serial_t serial = sp->hdr.serial;
571 	rxrpc_seq_t seq0 = sp->hdr.seq;
572 
573 	_enter("{%x,%x,%x},{%u,%x}",
574 	       call->ackr_window, call->ackr_wtop, call->rx_highest_seq,
575 	       skb->len, seq0);
576 
577 	if (__rxrpc_call_is_complete(call))
578 		return;
579 
580 	switch (__rxrpc_call_state(call)) {
581 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
582 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
583 		/* Received data implicitly ACKs all of the request
584 		 * packets we sent when we're acting as a client.
585 		 */
586 		if (!rxrpc_receiving_reply(call))
587 			goto out_notify;
588 		break;
589 
590 	case RXRPC_CALL_SERVER_RECV_REQUEST: {
591 		unsigned long timo = READ_ONCE(call->next_req_timo);
592 		unsigned long now, expect_req_by;
593 
594 		if (timo) {
595 			now = jiffies;
596 			expect_req_by = now + timo;
597 			WRITE_ONCE(call->expect_req_by, expect_req_by);
598 			rxrpc_reduce_call_timer(call, expect_req_by, now,
599 						rxrpc_timer_set_for_idle);
600 		}
601 		break;
602 	}
603 
604 	default:
605 		break;
606 	}
607 
608 	if (!rxrpc_input_split_jumbo(call, skb)) {
609 		rxrpc_proto_abort(call, sp->hdr.seq, rxrpc_badmsg_bad_jumbo);
610 		goto out_notify;
611 	}
612 	return;
613 
614 out_notify:
615 	trace_rxrpc_notify_socket(call->debug_id, serial);
616 	rxrpc_notify_socket(call);
617 	_leave(" [queued]");
618 }
619 
620 /*
621  * See if there's a cached RTT probe to complete.
622  */
623 static void rxrpc_complete_rtt_probe(struct rxrpc_call *call,
624 				     ktime_t resp_time,
625 				     rxrpc_serial_t acked_serial,
626 				     rxrpc_serial_t ack_serial,
627 				     enum rxrpc_rtt_rx_trace type)
628 {
629 	rxrpc_serial_t orig_serial;
630 	unsigned long avail;
631 	ktime_t sent_at;
632 	bool matched = false;
633 	int i;
634 
635 	avail = READ_ONCE(call->rtt_avail);
636 	smp_rmb(); /* Read avail bits before accessing data. */
637 
638 	for (i = 0; i < ARRAY_SIZE(call->rtt_serial); i++) {
639 		if (!test_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &avail))
640 			continue;
641 
642 		sent_at = call->rtt_sent_at[i];
643 		orig_serial = call->rtt_serial[i];
644 
645 		if (orig_serial == acked_serial) {
646 			clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
647 			smp_mb(); /* Read data before setting avail bit */
648 			set_bit(i, &call->rtt_avail);
649 			rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
650 					   sent_at, resp_time);
651 			matched = true;
652 		}
653 
654 		/* If a later serial is being acked, then mark this slot as
655 		 * being available.
656 		 */
657 		if (after(acked_serial, orig_serial)) {
658 			trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
659 					   orig_serial, acked_serial, 0, 0);
660 			clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
661 			smp_wmb();
662 			set_bit(i, &call->rtt_avail);
663 		}
664 	}
665 
666 	if (!matched)
667 		trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
668 }
669 
670 /*
671  * Process the extra information that may be appended to an ACK packet
672  */
673 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
674 				struct rxrpc_ackinfo *ackinfo)
675 {
676 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
677 	struct rxrpc_peer *peer;
678 	unsigned int mtu;
679 	bool wake = false;
680 	u32 rwind = ntohl(ackinfo->rwind);
681 
682 	if (rwind > RXRPC_TX_MAX_WINDOW)
683 		rwind = RXRPC_TX_MAX_WINDOW;
684 	if (call->tx_winsize != rwind) {
685 		if (rwind > call->tx_winsize)
686 			wake = true;
687 		trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
688 		call->tx_winsize = rwind;
689 	}
690 
691 	if (call->cong_ssthresh > rwind)
692 		call->cong_ssthresh = rwind;
693 
694 	mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
695 
696 	peer = call->peer;
697 	if (mtu < peer->maxdata) {
698 		spin_lock(&peer->lock);
699 		peer->maxdata = mtu;
700 		peer->mtu = mtu + peer->hdrsize;
701 		spin_unlock(&peer->lock);
702 	}
703 
704 	if (wake)
705 		wake_up(&call->waitq);
706 }
707 
708 /*
709  * Determine how many nacks from the previous ACK have now been satisfied.
710  */
711 static rxrpc_seq_t rxrpc_input_check_prev_ack(struct rxrpc_call *call,
712 					      struct rxrpc_ack_summary *summary,
713 					      rxrpc_seq_t seq)
714 {
715 	struct sk_buff *skb = call->cong_last_nack;
716 	struct rxrpc_ackpacket ack;
717 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
718 	unsigned int i, new_acks = 0, retained_nacks = 0;
719 	rxrpc_seq_t old_seq = sp->first_ack;
720 	u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(ack);
721 
722 	if (after_eq(seq, old_seq + sp->nr_acks)) {
723 		summary->nr_new_acks += sp->nr_nacks;
724 		summary->nr_new_acks += seq - (old_seq + sp->nr_acks);
725 		summary->nr_retained_nacks = 0;
726 	} else if (seq == old_seq) {
727 		summary->nr_retained_nacks = sp->nr_nacks;
728 	} else {
729 		for (i = 0; i < sp->nr_acks; i++) {
730 			if (acks[i] == RXRPC_ACK_TYPE_NACK) {
731 				if (before(old_seq + i, seq))
732 					new_acks++;
733 				else
734 					retained_nacks++;
735 			}
736 		}
737 
738 		summary->nr_new_acks += new_acks;
739 		summary->nr_retained_nacks = retained_nacks;
740 	}
741 
742 	return old_seq + sp->nr_acks;
743 }
744 
745 /*
746  * Process individual soft ACKs.
747  *
748  * Each ACK in the array corresponds to one packet and can be either an ACK or
749  * a NAK.  If we get find an explicitly NAK'd packet we resend immediately;
750  * packets that lie beyond the end of the ACK list are scheduled for resend by
751  * the timer on the basis that the peer might just not have processed them at
752  * the time the ACK was sent.
753  */
754 static void rxrpc_input_soft_acks(struct rxrpc_call *call,
755 				  struct rxrpc_ack_summary *summary,
756 				  struct sk_buff *skb,
757 				  rxrpc_seq_t seq,
758 				  rxrpc_seq_t since)
759 {
760 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
761 	unsigned int i, old_nacks = 0;
762 	rxrpc_seq_t lowest_nak = seq + sp->nr_acks;
763 	u8 *acks = skb->data + sizeof(struct rxrpc_wire_header) + sizeof(struct rxrpc_ackpacket);
764 
765 	for (i = 0; i < sp->nr_acks; i++) {
766 		if (acks[i] == RXRPC_ACK_TYPE_ACK) {
767 			summary->nr_acks++;
768 			if (after_eq(seq, since))
769 				summary->nr_new_acks++;
770 		} else {
771 			summary->saw_nacks = true;
772 			if (before(seq, since)) {
773 				/* Overlap with previous ACK */
774 				old_nacks++;
775 			} else {
776 				summary->nr_new_nacks++;
777 				sp->nr_nacks++;
778 			}
779 
780 			if (before(seq, lowest_nak))
781 				lowest_nak = seq;
782 		}
783 		seq++;
784 	}
785 
786 	if (lowest_nak != call->acks_lowest_nak) {
787 		call->acks_lowest_nak = lowest_nak;
788 		summary->new_low_nack = true;
789 	}
790 
791 	/* We *can* have more nacks than we did - the peer is permitted to drop
792 	 * packets it has soft-acked and re-request them.  Further, it is
793 	 * possible for the nack distribution to change whilst the number of
794 	 * nacks stays the same or goes down.
795 	 */
796 	if (old_nacks < summary->nr_retained_nacks)
797 		summary->nr_new_acks += summary->nr_retained_nacks - old_nacks;
798 	summary->nr_retained_nacks = old_nacks;
799 }
800 
801 /*
802  * Return true if the ACK is valid - ie. it doesn't appear to have regressed
803  * with respect to the ack state conveyed by preceding ACKs.
804  */
805 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
806 			       rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
807 {
808 	rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
809 
810 	if (after(first_pkt, base))
811 		return true; /* The window advanced */
812 
813 	if (before(first_pkt, base))
814 		return false; /* firstPacket regressed */
815 
816 	if (after_eq(prev_pkt, call->acks_prev_seq))
817 		return true; /* previousPacket hasn't regressed. */
818 
819 	/* Some rx implementations put a serial number in previousPacket. */
820 	if (after_eq(prev_pkt, base + call->tx_winsize))
821 		return false;
822 	return true;
823 }
824 
825 /*
826  * Process an ACK packet.
827  *
828  * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
829  * in the ACK array.  Anything before that is hard-ACK'd and may be discarded.
830  *
831  * A hard-ACK means that a packet has been processed and may be discarded; a
832  * soft-ACK means that the packet may be discarded and retransmission
833  * requested.  A phase is complete when all packets are hard-ACK'd.
834  */
835 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
836 {
837 	struct rxrpc_ack_summary summary = { 0 };
838 	struct rxrpc_ackpacket ack;
839 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
840 	struct rxrpc_ackinfo info;
841 	rxrpc_serial_t ack_serial, acked_serial;
842 	rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt, since;
843 	int nr_acks, offset, ioffset;
844 
845 	_enter("");
846 
847 	offset = sizeof(struct rxrpc_wire_header);
848 	if (skb_copy_bits(skb, offset, &ack, sizeof(ack)) < 0)
849 		return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack);
850 	offset += sizeof(ack);
851 
852 	ack_serial = sp->hdr.serial;
853 	acked_serial = ntohl(ack.serial);
854 	first_soft_ack = ntohl(ack.firstPacket);
855 	prev_pkt = ntohl(ack.previousPacket);
856 	hard_ack = first_soft_ack - 1;
857 	nr_acks = ack.nAcks;
858 	sp->first_ack = first_soft_ack;
859 	sp->nr_acks = nr_acks;
860 	summary.ack_reason = (ack.reason < RXRPC_ACK__INVALID ?
861 			      ack.reason : RXRPC_ACK__INVALID);
862 
863 	trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
864 			   first_soft_ack, prev_pkt,
865 			   summary.ack_reason, nr_acks);
866 	rxrpc_inc_stat(call->rxnet, stat_rx_acks[ack.reason]);
867 
868 	if (acked_serial != 0) {
869 		switch (ack.reason) {
870 		case RXRPC_ACK_PING_RESPONSE:
871 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
872 						 rxrpc_rtt_rx_ping_response);
873 			break;
874 		case RXRPC_ACK_REQUESTED:
875 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
876 						 rxrpc_rtt_rx_requested_ack);
877 			break;
878 		default:
879 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
880 						 rxrpc_rtt_rx_other_ack);
881 			break;
882 		}
883 	}
884 
885 	/* If we get an EXCEEDS_WINDOW ACK from the server, it probably
886 	 * indicates that the client address changed due to NAT.  The server
887 	 * lost the call because it switched to a different peer.
888 	 */
889 	if (unlikely(ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
890 	    first_soft_ack == 1 &&
891 	    prev_pkt == 0 &&
892 	    rxrpc_is_client_call(call)) {
893 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
894 					  0, -ENETRESET);
895 		goto send_response;
896 	}
897 
898 	/* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
899 	 * indicate a change of address.  However, we can retransmit the call
900 	 * if we still have it buffered to the beginning.
901 	 */
902 	if (unlikely(ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
903 	    first_soft_ack == 1 &&
904 	    prev_pkt == 0 &&
905 	    call->acks_hard_ack == 0 &&
906 	    rxrpc_is_client_call(call)) {
907 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
908 					  0, -ENETRESET);
909 		goto send_response;
910 	}
911 
912 	/* Discard any out-of-order or duplicate ACKs (outside lock). */
913 	if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
914 		trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
915 					   first_soft_ack, call->acks_first_seq,
916 					   prev_pkt, call->acks_prev_seq);
917 		goto send_response;
918 	}
919 
920 	info.rxMTU = 0;
921 	ioffset = offset + nr_acks + 3;
922 	if (skb->len >= ioffset + sizeof(info) &&
923 	    skb_copy_bits(skb, ioffset, &info, sizeof(info)) < 0)
924 		return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_info);
925 
926 	if (nr_acks > 0)
927 		skb_condense(skb);
928 
929 	if (call->cong_last_nack) {
930 		since = rxrpc_input_check_prev_ack(call, &summary, first_soft_ack);
931 		rxrpc_free_skb(call->cong_last_nack, rxrpc_skb_put_last_nack);
932 		call->cong_last_nack = NULL;
933 	} else {
934 		summary.nr_new_acks = first_soft_ack - call->acks_first_seq;
935 		call->acks_lowest_nak = first_soft_ack + nr_acks;
936 		since = first_soft_ack;
937 	}
938 
939 	call->acks_latest_ts = skb->tstamp;
940 	call->acks_first_seq = first_soft_ack;
941 	call->acks_prev_seq = prev_pkt;
942 
943 	switch (ack.reason) {
944 	case RXRPC_ACK_PING:
945 		break;
946 	default:
947 		if (acked_serial && after(acked_serial, call->acks_highest_serial))
948 			call->acks_highest_serial = acked_serial;
949 		break;
950 	}
951 
952 	/* Parse rwind and mtu sizes if provided. */
953 	if (info.rxMTU)
954 		rxrpc_input_ackinfo(call, skb, &info);
955 
956 	if (first_soft_ack == 0)
957 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
958 
959 	/* Ignore ACKs unless we are or have just been transmitting. */
960 	switch (__rxrpc_call_state(call)) {
961 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
962 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
963 	case RXRPC_CALL_SERVER_SEND_REPLY:
964 	case RXRPC_CALL_SERVER_AWAIT_ACK:
965 		break;
966 	default:
967 		goto send_response;
968 	}
969 
970 	if (before(hard_ack, call->acks_hard_ack) ||
971 	    after(hard_ack, call->tx_top))
972 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
973 	if (nr_acks > call->tx_top - hard_ack)
974 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
975 
976 	if (after(hard_ack, call->acks_hard_ack)) {
977 		if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
978 			rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
979 			goto send_response;
980 		}
981 	}
982 
983 	if (nr_acks > 0) {
984 		if (offset > (int)skb->len - nr_acks)
985 			return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
986 		rxrpc_input_soft_acks(call, &summary, skb, first_soft_ack, since);
987 		rxrpc_get_skb(skb, rxrpc_skb_get_last_nack);
988 		call->cong_last_nack = skb;
989 	}
990 
991 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
992 	    summary.nr_acks == call->tx_top - hard_ack &&
993 	    rxrpc_is_client_call(call))
994 		rxrpc_propose_ping(call, ack_serial,
995 				   rxrpc_propose_ack_ping_for_lost_reply);
996 
997 	rxrpc_congestion_management(call, skb, &summary, acked_serial);
998 
999 send_response:
1000 	if (ack.reason == RXRPC_ACK_PING)
1001 		rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
1002 			       rxrpc_propose_ack_respond_to_ping);
1003 	else if (sp->hdr.flags & RXRPC_REQUEST_ACK)
1004 		rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
1005 			       rxrpc_propose_ack_respond_to_ack);
1006 }
1007 
1008 /*
1009  * Process an ACKALL packet.
1010  */
1011 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
1012 {
1013 	struct rxrpc_ack_summary summary = { 0 };
1014 
1015 	if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
1016 		rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
1017 }
1018 
1019 /*
1020  * Process an ABORT packet directed at a call.
1021  */
1022 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
1023 {
1024 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1025 
1026 	trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
1027 
1028 	rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
1029 				  skb->priority, -ECONNABORTED);
1030 }
1031 
1032 /*
1033  * Process an incoming call packet.
1034  */
1035 void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
1036 {
1037 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
1038 	unsigned long timo;
1039 
1040 	_enter("%p,%p", call, skb);
1041 
1042 	if (sp->hdr.serviceId != call->dest_srx.srx_service)
1043 		call->dest_srx.srx_service = sp->hdr.serviceId;
1044 	if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
1045 		call->rx_serial = sp->hdr.serial;
1046 	if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
1047 		set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
1048 
1049 	timo = READ_ONCE(call->next_rx_timo);
1050 	if (timo) {
1051 		unsigned long now = jiffies, expect_rx_by;
1052 
1053 		expect_rx_by = now + timo;
1054 		WRITE_ONCE(call->expect_rx_by, expect_rx_by);
1055 		rxrpc_reduce_call_timer(call, expect_rx_by, now,
1056 					rxrpc_timer_set_for_normal);
1057 	}
1058 
1059 	switch (sp->hdr.type) {
1060 	case RXRPC_PACKET_TYPE_DATA:
1061 		return rxrpc_input_data(call, skb);
1062 
1063 	case RXRPC_PACKET_TYPE_ACK:
1064 		return rxrpc_input_ack(call, skb);
1065 
1066 	case RXRPC_PACKET_TYPE_BUSY:
1067 		/* Just ignore BUSY packets from the server; the retry and
1068 		 * lifespan timers will take care of business.  BUSY packets
1069 		 * from the client don't make sense.
1070 		 */
1071 		return;
1072 
1073 	case RXRPC_PACKET_TYPE_ABORT:
1074 		return rxrpc_input_abort(call, skb);
1075 
1076 	case RXRPC_PACKET_TYPE_ACKALL:
1077 		return rxrpc_input_ackall(call, skb);
1078 
1079 	default:
1080 		break;
1081 	}
1082 }
1083 
1084 /*
1085  * Handle a new service call on a channel implicitly completing the preceding
1086  * call on that channel.  This does not apply to client conns.
1087  *
1088  * TODO: If callNumber > call_id + 1, renegotiate security.
1089  */
1090 void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1091 {
1092 	switch (__rxrpc_call_state(call)) {
1093 	case RXRPC_CALL_SERVER_AWAIT_ACK:
1094 		rxrpc_call_completed(call);
1095 		fallthrough;
1096 	case RXRPC_CALL_COMPLETE:
1097 		break;
1098 	default:
1099 		rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1100 				 rxrpc_eproto_improper_term);
1101 		trace_rxrpc_improper_term(call);
1102 		break;
1103 	}
1104 
1105 	rxrpc_input_call_event(call, skb);
1106 }
1107