xref: /openbmc/linux/net/rxrpc/input.c (revision 62a9bbf2)
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 	cumulative_acks += summary->nr_rot_new_acks;
49 	if (cumulative_acks > 255)
50 		cumulative_acks = 255;
51 
52 	summary->mode = call->cong_mode;
53 	summary->cwnd = call->cong_cwnd;
54 	summary->ssthresh = call->cong_ssthresh;
55 	summary->cumulative_acks = cumulative_acks;
56 	summary->dup_acks = call->cong_dup_acks;
57 
58 	switch (call->cong_mode) {
59 	case RXRPC_CALL_SLOW_START:
60 		if (summary->saw_nacks)
61 			goto packet_loss_detected;
62 		if (summary->cumulative_acks > 0)
63 			cwnd += 1;
64 		if (cwnd >= call->cong_ssthresh) {
65 			call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
66 			call->cong_tstamp = skb->tstamp;
67 		}
68 		goto out;
69 
70 	case RXRPC_CALL_CONGEST_AVOIDANCE:
71 		if (summary->saw_nacks)
72 			goto packet_loss_detected;
73 
74 		/* We analyse the number of packets that get ACK'd per RTT
75 		 * period and increase the window if we managed to fill it.
76 		 */
77 		if (call->peer->rtt_count == 0)
78 			goto out;
79 		if (ktime_before(skb->tstamp,
80 				 ktime_add_us(call->cong_tstamp,
81 					      call->peer->srtt_us >> 3)))
82 			goto out_no_clear_ca;
83 		change = rxrpc_cong_rtt_window_end;
84 		call->cong_tstamp = skb->tstamp;
85 		if (cumulative_acks >= cwnd)
86 			cwnd++;
87 		goto out;
88 
89 	case RXRPC_CALL_PACKET_LOSS:
90 		if (!summary->saw_nacks)
91 			goto resume_normality;
92 
93 		if (summary->new_low_nack) {
94 			change = rxrpc_cong_new_low_nack;
95 			call->cong_dup_acks = 1;
96 			if (call->cong_extra > 1)
97 				call->cong_extra = 1;
98 			goto send_extra_data;
99 		}
100 
101 		call->cong_dup_acks++;
102 		if (call->cong_dup_acks < 3)
103 			goto send_extra_data;
104 
105 		change = rxrpc_cong_begin_retransmission;
106 		call->cong_mode = RXRPC_CALL_FAST_RETRANSMIT;
107 		call->cong_ssthresh = max_t(unsigned int,
108 					    summary->flight_size / 2, 2);
109 		cwnd = call->cong_ssthresh + 3;
110 		call->cong_extra = 0;
111 		call->cong_dup_acks = 0;
112 		resend = true;
113 		goto out;
114 
115 	case RXRPC_CALL_FAST_RETRANSMIT:
116 		if (!summary->new_low_nack) {
117 			if (summary->nr_new_acks == 0)
118 				cwnd += 1;
119 			call->cong_dup_acks++;
120 			if (call->cong_dup_acks == 2) {
121 				change = rxrpc_cong_retransmit_again;
122 				call->cong_dup_acks = 0;
123 				resend = true;
124 			}
125 		} else {
126 			change = rxrpc_cong_progress;
127 			cwnd = call->cong_ssthresh;
128 			if (!summary->saw_nacks)
129 				goto resume_normality;
130 		}
131 		goto out;
132 
133 	default:
134 		BUG();
135 		goto out;
136 	}
137 
138 resume_normality:
139 	change = rxrpc_cong_cleared_nacks;
140 	call->cong_dup_acks = 0;
141 	call->cong_extra = 0;
142 	call->cong_tstamp = skb->tstamp;
143 	if (cwnd < call->cong_ssthresh)
144 		call->cong_mode = RXRPC_CALL_SLOW_START;
145 	else
146 		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
147 out:
148 	cumulative_acks = 0;
149 out_no_clear_ca:
150 	if (cwnd >= RXRPC_TX_MAX_WINDOW)
151 		cwnd = RXRPC_TX_MAX_WINDOW;
152 	call->cong_cwnd = cwnd;
153 	call->cong_cumul_acks = cumulative_acks;
154 	trace_rxrpc_congest(call, summary, acked_serial, change);
155 	if (resend)
156 		rxrpc_resend(call, skb);
157 	return;
158 
159 packet_loss_detected:
160 	change = rxrpc_cong_saw_nack;
161 	call->cong_mode = RXRPC_CALL_PACKET_LOSS;
162 	call->cong_dup_acks = 0;
163 	goto send_extra_data;
164 
165 send_extra_data:
166 	/* Send some previously unsent DATA if we have some to advance the ACK
167 	 * state.
168 	 */
169 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) ||
170 	    summary->nr_acks != call->tx_top - call->acks_hard_ack) {
171 		call->cong_extra++;
172 		wake_up(&call->waitq);
173 	}
174 	goto out_no_clear_ca;
175 }
176 
177 /*
178  * Degrade the congestion window if we haven't transmitted a packet for >1RTT.
179  */
180 void rxrpc_congestion_degrade(struct rxrpc_call *call)
181 {
182 	ktime_t rtt, now;
183 
184 	if (call->cong_mode != RXRPC_CALL_SLOW_START &&
185 	    call->cong_mode != RXRPC_CALL_CONGEST_AVOIDANCE)
186 		return;
187 	if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY)
188 		return;
189 
190 	rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8));
191 	now = ktime_get_real();
192 	if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now))
193 		return;
194 
195 	trace_rxrpc_reset_cwnd(call, now);
196 	rxrpc_inc_stat(call->rxnet, stat_tx_data_cwnd_reset);
197 	call->tx_last_sent = now;
198 	call->cong_mode = RXRPC_CALL_SLOW_START;
199 	call->cong_ssthresh = max_t(unsigned int, call->cong_ssthresh,
200 				    call->cong_cwnd * 3 / 4);
201 	call->cong_cwnd = max_t(unsigned int, call->cong_cwnd / 2, RXRPC_MIN_CWND);
202 }
203 
204 /*
205  * Apply a hard ACK by advancing the Tx window.
206  */
207 static bool rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to,
208 				   struct rxrpc_ack_summary *summary)
209 {
210 	struct rxrpc_txbuf *txb;
211 	bool rot_last = false;
212 
213 	list_for_each_entry_rcu(txb, &call->tx_buffer, call_link, false) {
214 		if (before_eq(txb->seq, call->acks_hard_ack))
215 			continue;
216 		summary->nr_rot_new_acks++;
217 		if (test_bit(RXRPC_TXBUF_LAST, &txb->flags)) {
218 			set_bit(RXRPC_CALL_TX_LAST, &call->flags);
219 			rot_last = true;
220 		}
221 		if (txb->seq == to)
222 			break;
223 	}
224 
225 	if (rot_last)
226 		set_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags);
227 
228 	_enter("%x,%x,%x,%d", to, call->acks_hard_ack, call->tx_top, rot_last);
229 
230 	if (call->acks_lowest_nak == call->acks_hard_ack) {
231 		call->acks_lowest_nak = to;
232 	} else if (after(to, call->acks_lowest_nak)) {
233 		summary->new_low_nack = true;
234 		call->acks_lowest_nak = to;
235 	}
236 
237 	smp_store_release(&call->acks_hard_ack, to);
238 
239 	trace_rxrpc_txqueue(call, (rot_last ?
240 				   rxrpc_txqueue_rotate_last :
241 				   rxrpc_txqueue_rotate));
242 	wake_up(&call->waitq);
243 	return rot_last;
244 }
245 
246 /*
247  * End the transmission phase of a call.
248  *
249  * This occurs when we get an ACKALL packet, the first DATA packet of a reply,
250  * or a final ACK packet.
251  */
252 static void rxrpc_end_tx_phase(struct rxrpc_call *call, bool reply_begun,
253 			       enum rxrpc_abort_reason abort_why)
254 {
255 	ASSERT(test_bit(RXRPC_CALL_TX_LAST, &call->flags));
256 
257 	switch (__rxrpc_call_state(call)) {
258 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
259 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
260 		if (reply_begun) {
261 			rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_RECV_REPLY);
262 			trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
263 			break;
264 		}
265 
266 		rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY);
267 		trace_rxrpc_txqueue(call, rxrpc_txqueue_await_reply);
268 		break;
269 
270 	case RXRPC_CALL_SERVER_AWAIT_ACK:
271 		rxrpc_call_completed(call);
272 		trace_rxrpc_txqueue(call, rxrpc_txqueue_end);
273 		break;
274 
275 	default:
276 		kdebug("end_tx %s", rxrpc_call_states[__rxrpc_call_state(call)]);
277 		rxrpc_proto_abort(call, call->tx_top, abort_why);
278 		break;
279 	}
280 }
281 
282 /*
283  * Begin the reply reception phase of a call.
284  */
285 static bool rxrpc_receiving_reply(struct rxrpc_call *call)
286 {
287 	struct rxrpc_ack_summary summary = { 0 };
288 	unsigned long now, timo;
289 	rxrpc_seq_t top = READ_ONCE(call->tx_top);
290 
291 	if (call->ackr_reason) {
292 		now = jiffies;
293 		timo = now + MAX_JIFFY_OFFSET;
294 
295 		WRITE_ONCE(call->delay_ack_at, timo);
296 		trace_rxrpc_timer(call, rxrpc_timer_init_for_reply, now);
297 	}
298 
299 	if (!test_bit(RXRPC_CALL_TX_LAST, &call->flags)) {
300 		if (!rxrpc_rotate_tx_window(call, top, &summary)) {
301 			rxrpc_proto_abort(call, top, rxrpc_eproto_early_reply);
302 			return false;
303 		}
304 	}
305 
306 	rxrpc_end_tx_phase(call, true, rxrpc_eproto_unexpected_reply);
307 	return true;
308 }
309 
310 /*
311  * End the packet reception phase.
312  */
313 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
314 {
315 	rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
316 
317 	_enter("%d,%s", call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)]);
318 
319 	trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
320 
321 	switch (__rxrpc_call_state(call)) {
322 	case RXRPC_CALL_CLIENT_RECV_REPLY:
323 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
324 		rxrpc_call_completed(call);
325 		break;
326 
327 	case RXRPC_CALL_SERVER_RECV_REQUEST:
328 		rxrpc_set_call_state(call, RXRPC_CALL_SERVER_ACK_REQUEST);
329 		call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
330 		rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_processing_op);
331 		break;
332 
333 	default:
334 		break;
335 	}
336 }
337 
338 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
339 					  rxrpc_seq_t window, rxrpc_seq_t wtop)
340 {
341 	atomic64_set_release(&call->ackr_window, ((u64)wtop) << 32 | window);
342 }
343 
344 /*
345  * Push a DATA packet onto the Rx queue.
346  */
347 static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
348 				   rxrpc_seq_t window, rxrpc_seq_t wtop,
349 				   enum rxrpc_receive_trace why)
350 {
351 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
352 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
353 
354 	__skb_queue_tail(&call->recvmsg_queue, skb);
355 	rxrpc_input_update_ack_window(call, window, wtop);
356 	trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
357 	if (last)
358 		rxrpc_end_rx_phase(call, sp->hdr.serial);
359 }
360 
361 /*
362  * Process a DATA packet.
363  */
364 static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb,
365 				 bool *_notify)
366 {
367 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
368 	struct sk_buff *oos;
369 	rxrpc_serial_t serial = sp->hdr.serial;
370 	u64 win = atomic64_read(&call->ackr_window);
371 	rxrpc_seq_t window = lower_32_bits(win);
372 	rxrpc_seq_t wtop = upper_32_bits(win);
373 	rxrpc_seq_t wlimit = window + call->rx_winsize - 1;
374 	rxrpc_seq_t seq = sp->hdr.seq;
375 	bool last = sp->hdr.flags & RXRPC_LAST_PACKET;
376 	int ack_reason = -1;
377 
378 	rxrpc_inc_stat(call->rxnet, stat_rx_data);
379 	if (sp->hdr.flags & RXRPC_REQUEST_ACK)
380 		rxrpc_inc_stat(call->rxnet, stat_rx_data_reqack);
381 	if (sp->hdr.flags & RXRPC_JUMBO_PACKET)
382 		rxrpc_inc_stat(call->rxnet, stat_rx_data_jumbo);
383 
384 	if (last) {
385 		if (test_and_set_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
386 		    seq + 1 != wtop)
387 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_different_last);
388 	} else {
389 		if (test_bit(RXRPC_CALL_RX_LAST, &call->flags) &&
390 		    after_eq(seq, wtop)) {
391 			pr_warn("Packet beyond last: c=%x q=%x window=%x-%x wlimit=%x\n",
392 				call->debug_id, seq, window, wtop, wlimit);
393 			return rxrpc_proto_abort(call, seq, rxrpc_eproto_data_after_last);
394 		}
395 	}
396 
397 	if (after(seq, call->rx_highest_seq))
398 		call->rx_highest_seq = seq;
399 
400 	trace_rxrpc_rx_data(call->debug_id, seq, serial, sp->hdr.flags);
401 
402 	if (before(seq, window)) {
403 		ack_reason = RXRPC_ACK_DUPLICATE;
404 		goto send_ack;
405 	}
406 	if (after(seq, wlimit)) {
407 		ack_reason = RXRPC_ACK_EXCEEDS_WINDOW;
408 		goto send_ack;
409 	}
410 
411 	/* Queue the packet. */
412 	if (seq == window) {
413 		rxrpc_seq_t reset_from;
414 		bool reset_sack = false;
415 
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 			atomic_inc_return(&call->ackr_nr_unacked);
423 
424 		window++;
425 		if (after(window, wtop))
426 			wtop = window;
427 
428 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg);
429 
430 		spin_lock(&call->recvmsg_queue.lock);
431 		rxrpc_input_queue_data(call, skb, window, wtop, rxrpc_receive_queue);
432 		*_notify = true;
433 
434 		while ((oos = skb_peek(&call->rx_oos_queue))) {
435 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
436 
437 			if (after(osp->hdr.seq, window))
438 				break;
439 
440 			__skb_unlink(oos, &call->rx_oos_queue);
441 			last = osp->hdr.flags & RXRPC_LAST_PACKET;
442 			seq = osp->hdr.seq;
443 			if (!reset_sack) {
444 				reset_from = seq;
445 				reset_sack = true;
446 			}
447 
448 			window++;
449 			rxrpc_input_queue_data(call, oos, window, wtop,
450 						 rxrpc_receive_queue_oos);
451 		}
452 
453 		spin_unlock(&call->recvmsg_queue.lock);
454 
455 		if (reset_sack) {
456 			do {
457 				call->ackr_sack_table[reset_from % RXRPC_SACK_SIZE] = 0;
458 			} while (reset_from++, before(reset_from, window));
459 		}
460 	} else {
461 		bool keep = false;
462 
463 		ack_reason = RXRPC_ACK_OUT_OF_SEQUENCE;
464 
465 		if (!call->ackr_sack_table[seq % RXRPC_SACK_SIZE]) {
466 			call->ackr_sack_table[seq % RXRPC_SACK_SIZE] = 1;
467 			keep = 1;
468 		}
469 
470 		if (after(seq + 1, wtop)) {
471 			wtop = seq + 1;
472 			rxrpc_input_update_ack_window(call, window, wtop);
473 		}
474 
475 		if (!keep) {
476 			ack_reason = RXRPC_ACK_DUPLICATE;
477 			goto send_ack;
478 		}
479 
480 		skb_queue_walk(&call->rx_oos_queue, oos) {
481 			struct rxrpc_skb_priv *osp = rxrpc_skb(oos);
482 
483 			if (after(osp->hdr.seq, seq)) {
484 				rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
485 				__skb_queue_before(&call->rx_oos_queue, oos, skb);
486 				goto oos_queued;
487 			}
488 		}
489 
490 		rxrpc_get_skb(skb, rxrpc_skb_get_to_recvmsg_oos);
491 		__skb_queue_tail(&call->rx_oos_queue, skb);
492 	oos_queued:
493 		trace_rxrpc_receive(call, last ? rxrpc_receive_oos_last : rxrpc_receive_oos,
494 				    sp->hdr.serial, sp->hdr.seq);
495 	}
496 
497 send_ack:
498 	if (ack_reason >= 0)
499 		rxrpc_send_ACK(call, ack_reason, serial,
500 			       rxrpc_propose_ack_input_data);
501 	else
502 		rxrpc_propose_delay_ACK(call, serial,
503 					rxrpc_propose_ack_input_data);
504 }
505 
506 /*
507  * Split a jumbo packet and file the bits separately.
508  */
509 static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb)
510 {
511 	struct rxrpc_jumbo_header jhdr;
512 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp;
513 	struct sk_buff *jskb;
514 	unsigned int offset = sizeof(struct rxrpc_wire_header);
515 	unsigned int len = skb->len - offset;
516 	bool notify = false;
517 
518 	while (sp->hdr.flags & RXRPC_JUMBO_PACKET) {
519 		if (len < RXRPC_JUMBO_SUBPKTLEN)
520 			goto protocol_error;
521 		if (sp->hdr.flags & RXRPC_LAST_PACKET)
522 			goto protocol_error;
523 		if (skb_copy_bits(skb, offset + RXRPC_JUMBO_DATALEN,
524 				  &jhdr, sizeof(jhdr)) < 0)
525 			goto protocol_error;
526 
527 		jskb = skb_clone(skb, GFP_NOFS);
528 		if (!jskb) {
529 			kdebug("couldn't clone");
530 			return false;
531 		}
532 		rxrpc_new_skb(jskb, rxrpc_skb_new_jumbo_subpacket);
533 		jsp = rxrpc_skb(jskb);
534 		jsp->offset = offset;
535 		jsp->len = RXRPC_JUMBO_DATALEN;
536 		rxrpc_input_data_one(call, jskb, &notify);
537 		rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket);
538 
539 		sp->hdr.flags = jhdr.flags;
540 		sp->hdr._rsvd = ntohs(jhdr._rsvd);
541 		sp->hdr.seq++;
542 		sp->hdr.serial++;
543 		offset += RXRPC_JUMBO_SUBPKTLEN;
544 		len -= RXRPC_JUMBO_SUBPKTLEN;
545 	}
546 
547 	sp->offset = offset;
548 	sp->len    = len;
549 	rxrpc_input_data_one(call, skb, &notify);
550 	if (notify) {
551 		trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial);
552 		rxrpc_notify_socket(call);
553 	}
554 	return true;
555 
556 protocol_error:
557 	return false;
558 }
559 
560 /*
561  * Process a DATA packet, adding the packet to the Rx ring.  The caller's
562  * packet ref must be passed on or discarded.
563  */
564 static void rxrpc_input_data(struct rxrpc_call *call, struct sk_buff *skb)
565 {
566 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
567 	rxrpc_serial_t serial = sp->hdr.serial;
568 	rxrpc_seq_t seq0 = sp->hdr.seq;
569 
570 	_enter("{%llx,%x},{%u,%x}",
571 	       atomic64_read(&call->ackr_window), call->rx_highest_seq,
572 	       skb->len, seq0);
573 
574 	if (__rxrpc_call_is_complete(call))
575 		return;
576 
577 	switch (__rxrpc_call_state(call)) {
578 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
579 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
580 		/* Received data implicitly ACKs all of the request
581 		 * packets we sent when we're acting as a client.
582 		 */
583 		if (!rxrpc_receiving_reply(call))
584 			goto out_notify;
585 		break;
586 
587 	case RXRPC_CALL_SERVER_RECV_REQUEST: {
588 		unsigned long timo = READ_ONCE(call->next_req_timo);
589 		unsigned long now, expect_req_by;
590 
591 		if (timo) {
592 			now = jiffies;
593 			expect_req_by = now + timo;
594 			WRITE_ONCE(call->expect_req_by, expect_req_by);
595 			rxrpc_reduce_call_timer(call, expect_req_by, now,
596 						rxrpc_timer_set_for_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 	skb = NULL;
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 			if (type != rxrpc_rtt_rx_cancel)
647 				rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial,
648 						   sent_at, resp_time);
649 			else
650 				trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_cancel, i,
651 						   orig_serial, acked_serial, 0, 0);
652 			matched = true;
653 		}
654 
655 		/* If a later serial is being acked, then mark this slot as
656 		 * being available.
657 		 */
658 		if (after(acked_serial, orig_serial)) {
659 			trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_obsolete, i,
660 					   orig_serial, acked_serial, 0, 0);
661 			clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail);
662 			smp_wmb();
663 			set_bit(i, &call->rtt_avail);
664 		}
665 	}
666 
667 	if (!matched)
668 		trace_rxrpc_rtt_rx(call, rxrpc_rtt_rx_lost, 9, 0, acked_serial, 0, 0);
669 }
670 
671 /*
672  * Process the extra information that may be appended to an ACK packet
673  */
674 static void rxrpc_input_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
675 				struct rxrpc_ackinfo *ackinfo)
676 {
677 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
678 	struct rxrpc_peer *peer;
679 	unsigned int mtu;
680 	bool wake = false;
681 	u32 rwind = ntohl(ackinfo->rwind);
682 
683 	if (rwind > RXRPC_TX_MAX_WINDOW)
684 		rwind = RXRPC_TX_MAX_WINDOW;
685 	if (call->tx_winsize != rwind) {
686 		if (rwind > call->tx_winsize)
687 			wake = true;
688 		trace_rxrpc_rx_rwind_change(call, sp->hdr.serial, rwind, wake);
689 		call->tx_winsize = rwind;
690 	}
691 
692 	if (call->cong_ssthresh > rwind)
693 		call->cong_ssthresh = rwind;
694 
695 	mtu = min(ntohl(ackinfo->rxMTU), ntohl(ackinfo->maxMTU));
696 
697 	peer = call->peer;
698 	if (mtu < peer->maxdata) {
699 		spin_lock(&peer->lock);
700 		peer->maxdata = mtu;
701 		peer->mtu = mtu + peer->hdrsize;
702 		spin_unlock(&peer->lock);
703 	}
704 
705 	if (wake)
706 		wake_up(&call->waitq);
707 }
708 
709 /*
710  * Process individual soft ACKs.
711  *
712  * Each ACK in the array corresponds to one packet and can be either an ACK or
713  * a NAK.  If we get find an explicitly NAK'd packet we resend immediately;
714  * packets that lie beyond the end of the ACK list are scheduled for resend by
715  * the timer on the basis that the peer might just not have processed them at
716  * the time the ACK was sent.
717  */
718 static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
719 				  rxrpc_seq_t seq, int nr_acks,
720 				  struct rxrpc_ack_summary *summary)
721 {
722 	unsigned int i;
723 
724 	for (i = 0; i < nr_acks; i++) {
725 		if (acks[i] == RXRPC_ACK_TYPE_ACK) {
726 			summary->nr_acks++;
727 			summary->nr_new_acks++;
728 		} else {
729 			if (!summary->saw_nacks &&
730 			    call->acks_lowest_nak != seq + i) {
731 				call->acks_lowest_nak = seq + i;
732 				summary->new_low_nack = true;
733 			}
734 			summary->saw_nacks = true;
735 		}
736 	}
737 }
738 
739 /*
740  * Return true if the ACK is valid - ie. it doesn't appear to have regressed
741  * with respect to the ack state conveyed by preceding ACKs.
742  */
743 static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
744 			       rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
745 {
746 	rxrpc_seq_t base = READ_ONCE(call->acks_first_seq);
747 
748 	if (after(first_pkt, base))
749 		return true; /* The window advanced */
750 
751 	if (before(first_pkt, base))
752 		return false; /* firstPacket regressed */
753 
754 	if (after_eq(prev_pkt, call->acks_prev_seq))
755 		return true; /* previousPacket hasn't regressed. */
756 
757 	/* Some rx implementations put a serial number in previousPacket. */
758 	if (after_eq(prev_pkt, base + call->tx_winsize))
759 		return false;
760 	return true;
761 }
762 
763 /*
764  * Process an ACK packet.
765  *
766  * ack.firstPacket is the sequence number of the first soft-ACK'd/NAK'd packet
767  * in the ACK array.  Anything before that is hard-ACK'd and may be discarded.
768  *
769  * A hard-ACK means that a packet has been processed and may be discarded; a
770  * soft-ACK means that the packet may be discarded and retransmission
771  * requested.  A phase is complete when all packets are hard-ACK'd.
772  */
773 static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
774 {
775 	struct rxrpc_ack_summary summary = { 0 };
776 	struct rxrpc_ackpacket ack;
777 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
778 	struct rxrpc_ackinfo info;
779 	rxrpc_serial_t ack_serial, acked_serial;
780 	rxrpc_seq_t first_soft_ack, hard_ack, prev_pkt;
781 	int nr_acks, offset, ioffset;
782 
783 	_enter("");
784 
785 	offset = sizeof(struct rxrpc_wire_header);
786 	if (skb_copy_bits(skb, offset, &ack, sizeof(ack)) < 0)
787 		return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack);
788 	offset += sizeof(ack);
789 
790 	ack_serial = sp->hdr.serial;
791 	acked_serial = ntohl(ack.serial);
792 	first_soft_ack = ntohl(ack.firstPacket);
793 	prev_pkt = ntohl(ack.previousPacket);
794 	hard_ack = first_soft_ack - 1;
795 	nr_acks = ack.nAcks;
796 	summary.ack_reason = (ack.reason < RXRPC_ACK__INVALID ?
797 			      ack.reason : RXRPC_ACK__INVALID);
798 
799 	trace_rxrpc_rx_ack(call, ack_serial, acked_serial,
800 			   first_soft_ack, prev_pkt,
801 			   summary.ack_reason, nr_acks);
802 	rxrpc_inc_stat(call->rxnet, stat_rx_acks[ack.reason]);
803 
804 	switch (ack.reason) {
805 	case RXRPC_ACK_PING_RESPONSE:
806 		rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
807 					 rxrpc_rtt_rx_ping_response);
808 		break;
809 	case RXRPC_ACK_REQUESTED:
810 		rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
811 					 rxrpc_rtt_rx_requested_ack);
812 		break;
813 	default:
814 		if (acked_serial != 0)
815 			rxrpc_complete_rtt_probe(call, skb->tstamp, acked_serial, ack_serial,
816 						 rxrpc_rtt_rx_cancel);
817 		break;
818 	}
819 
820 	if (ack.reason == RXRPC_ACK_PING) {
821 		rxrpc_send_ACK(call, RXRPC_ACK_PING_RESPONSE, ack_serial,
822 			       rxrpc_propose_ack_respond_to_ping);
823 	} else if (sp->hdr.flags & RXRPC_REQUEST_ACK) {
824 		rxrpc_send_ACK(call, RXRPC_ACK_REQUESTED, ack_serial,
825 			       rxrpc_propose_ack_respond_to_ack);
826 	}
827 
828 	/* If we get an EXCEEDS_WINDOW ACK from the server, it probably
829 	 * indicates that the client address changed due to NAT.  The server
830 	 * lost the call because it switched to a different peer.
831 	 */
832 	if (unlikely(ack.reason == RXRPC_ACK_EXCEEDS_WINDOW) &&
833 	    first_soft_ack == 1 &&
834 	    prev_pkt == 0 &&
835 	    rxrpc_is_client_call(call)) {
836 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
837 					  0, -ENETRESET);
838 		return;
839 	}
840 
841 	/* If we get an OUT_OF_SEQUENCE ACK from the server, that can also
842 	 * indicate a change of address.  However, we can retransmit the call
843 	 * if we still have it buffered to the beginning.
844 	 */
845 	if (unlikely(ack.reason == RXRPC_ACK_OUT_OF_SEQUENCE) &&
846 	    first_soft_ack == 1 &&
847 	    prev_pkt == 0 &&
848 	    call->acks_hard_ack == 0 &&
849 	    rxrpc_is_client_call(call)) {
850 		rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
851 					  0, -ENETRESET);
852 		return;
853 	}
854 
855 	/* Discard any out-of-order or duplicate ACKs (outside lock). */
856 	if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
857 		trace_rxrpc_rx_discard_ack(call->debug_id, ack_serial,
858 					   first_soft_ack, call->acks_first_seq,
859 					   prev_pkt, call->acks_prev_seq);
860 		return;
861 	}
862 
863 	info.rxMTU = 0;
864 	ioffset = offset + nr_acks + 3;
865 	if (skb->len >= ioffset + sizeof(info) &&
866 	    skb_copy_bits(skb, ioffset, &info, sizeof(info)) < 0)
867 		return rxrpc_proto_abort(call, 0, rxrpc_badmsg_short_ack_info);
868 
869 	if (nr_acks > 0)
870 		skb_condense(skb);
871 
872 	call->acks_latest_ts = skb->tstamp;
873 	call->acks_first_seq = first_soft_ack;
874 	call->acks_prev_seq = prev_pkt;
875 
876 	switch (ack.reason) {
877 	case RXRPC_ACK_PING:
878 		break;
879 	default:
880 		if (after(acked_serial, call->acks_highest_serial))
881 			call->acks_highest_serial = acked_serial;
882 		break;
883 	}
884 
885 	/* Parse rwind and mtu sizes if provided. */
886 	if (info.rxMTU)
887 		rxrpc_input_ackinfo(call, skb, &info);
888 
889 	if (first_soft_ack == 0)
890 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_zero);
891 
892 	/* Ignore ACKs unless we are or have just been transmitting. */
893 	switch (__rxrpc_call_state(call)) {
894 	case RXRPC_CALL_CLIENT_SEND_REQUEST:
895 	case RXRPC_CALL_CLIENT_AWAIT_REPLY:
896 	case RXRPC_CALL_SERVER_SEND_REPLY:
897 	case RXRPC_CALL_SERVER_AWAIT_ACK:
898 		break;
899 	default:
900 		return;
901 	}
902 
903 	if (before(hard_ack, call->acks_hard_ack) ||
904 	    after(hard_ack, call->tx_top))
905 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_outside_window);
906 	if (nr_acks > call->tx_top - hard_ack)
907 		return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_sack_overflow);
908 
909 	if (after(hard_ack, call->acks_hard_ack)) {
910 		if (rxrpc_rotate_tx_window(call, hard_ack, &summary)) {
911 			rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ack);
912 			return;
913 		}
914 	}
915 
916 	if (nr_acks > 0) {
917 		if (offset > (int)skb->len - nr_acks)
918 			return rxrpc_proto_abort(call, 0, rxrpc_eproto_ackr_short_sack);
919 		rxrpc_input_soft_acks(call, skb->data + offset, first_soft_ack,
920 				      nr_acks, &summary);
921 	}
922 
923 	if (test_bit(RXRPC_CALL_TX_LAST, &call->flags) &&
924 	    summary.nr_acks == call->tx_top - hard_ack &&
925 	    rxrpc_is_client_call(call))
926 		rxrpc_propose_ping(call, ack_serial,
927 				   rxrpc_propose_ack_ping_for_lost_reply);
928 
929 	rxrpc_congestion_management(call, skb, &summary, acked_serial);
930 }
931 
932 /*
933  * Process an ACKALL packet.
934  */
935 static void rxrpc_input_ackall(struct rxrpc_call *call, struct sk_buff *skb)
936 {
937 	struct rxrpc_ack_summary summary = { 0 };
938 
939 	if (rxrpc_rotate_tx_window(call, call->tx_top, &summary))
940 		rxrpc_end_tx_phase(call, false, rxrpc_eproto_unexpected_ackall);
941 }
942 
943 /*
944  * Process an ABORT packet directed at a call.
945  */
946 static void rxrpc_input_abort(struct rxrpc_call *call, struct sk_buff *skb)
947 {
948 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
949 
950 	trace_rxrpc_rx_abort(call, sp->hdr.serial, skb->priority);
951 
952 	rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
953 				  skb->priority, -ECONNABORTED);
954 }
955 
956 /*
957  * Process an incoming call packet.
958  */
959 void rxrpc_input_call_packet(struct rxrpc_call *call, struct sk_buff *skb)
960 {
961 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
962 	unsigned long timo;
963 
964 	_enter("%p,%p", call, skb);
965 
966 	if (sp->hdr.serviceId != call->dest_srx.srx_service)
967 		call->dest_srx.srx_service = sp->hdr.serviceId;
968 	if ((int)sp->hdr.serial - (int)call->rx_serial > 0)
969 		call->rx_serial = sp->hdr.serial;
970 	if (!test_bit(RXRPC_CALL_RX_HEARD, &call->flags))
971 		set_bit(RXRPC_CALL_RX_HEARD, &call->flags);
972 
973 	timo = READ_ONCE(call->next_rx_timo);
974 	if (timo) {
975 		unsigned long now = jiffies, expect_rx_by;
976 
977 		expect_rx_by = now + timo;
978 		WRITE_ONCE(call->expect_rx_by, expect_rx_by);
979 		rxrpc_reduce_call_timer(call, expect_rx_by, now,
980 					rxrpc_timer_set_for_normal);
981 	}
982 
983 	switch (sp->hdr.type) {
984 	case RXRPC_PACKET_TYPE_DATA:
985 		return rxrpc_input_data(call, skb);
986 
987 	case RXRPC_PACKET_TYPE_ACK:
988 		return rxrpc_input_ack(call, skb);
989 
990 	case RXRPC_PACKET_TYPE_BUSY:
991 		/* Just ignore BUSY packets from the server; the retry and
992 		 * lifespan timers will take care of business.  BUSY packets
993 		 * from the client don't make sense.
994 		 */
995 		return;
996 
997 	case RXRPC_PACKET_TYPE_ABORT:
998 		return rxrpc_input_abort(call, skb);
999 
1000 	case RXRPC_PACKET_TYPE_ACKALL:
1001 		return rxrpc_input_ackall(call, skb);
1002 
1003 	default:
1004 		break;
1005 	}
1006 }
1007 
1008 /*
1009  * Handle a new service call on a channel implicitly completing the preceding
1010  * call on that channel.  This does not apply to client conns.
1011  *
1012  * TODO: If callNumber > call_id + 1, renegotiate security.
1013  */
1014 void rxrpc_implicit_end_call(struct rxrpc_call *call, struct sk_buff *skb)
1015 {
1016 	switch (__rxrpc_call_state(call)) {
1017 	case RXRPC_CALL_SERVER_AWAIT_ACK:
1018 		rxrpc_call_completed(call);
1019 		fallthrough;
1020 	case RXRPC_CALL_COMPLETE:
1021 		break;
1022 	default:
1023 		rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ESHUTDOWN,
1024 				 rxrpc_eproto_improper_term);
1025 		trace_rxrpc_improper_term(call);
1026 		break;
1027 	}
1028 
1029 	rxrpc_input_call_event(call, skb);
1030 }
1031