xref: /linux/net/rxrpc/io_thread.c (revision 06a130e42a5bfc84795464bff023bff4c16f58c5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC packet reception
3  *
4  * Copyright (C) 2007, 2016, 2022 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 int rxrpc_input_packet_on_conn(struct rxrpc_connection *conn,
13 				      struct sockaddr_rxrpc *peer_srx,
14 				      struct sk_buff *skb);
15 
16 /*
17  * handle data received on the local endpoint
18  * - may be called in interrupt context
19  *
20  * [!] Note that as this is called from the encap_rcv hook, the socket is not
21  * held locked by the caller and nothing prevents sk_user_data on the UDP from
22  * being cleared in the middle of processing this function.
23  *
24  * Called with the RCU read lock held from the IP layer via UDP.
25  */
26 int rxrpc_encap_rcv(struct sock *udp_sk, struct sk_buff *skb)
27 {
28 	struct sk_buff_head *rx_queue;
29 	struct rxrpc_local *local = rcu_dereference_sk_user_data(udp_sk);
30 	struct task_struct *io_thread;
31 
32 	if (unlikely(!local)) {
33 		kfree_skb(skb);
34 		return 0;
35 	}
36 	io_thread = READ_ONCE(local->io_thread);
37 	if (!io_thread) {
38 		kfree_skb(skb);
39 		return 0;
40 	}
41 	if (skb->tstamp == 0)
42 		skb->tstamp = ktime_get_real();
43 
44 	skb->mark = RXRPC_SKB_MARK_PACKET;
45 	rxrpc_new_skb(skb, rxrpc_skb_new_encap_rcv);
46 	rx_queue = &local->rx_queue;
47 #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY
48 	if (rxrpc_inject_rx_delay ||
49 	    !skb_queue_empty(&local->rx_delay_queue)) {
50 		skb->tstamp = ktime_add_ms(skb->tstamp, rxrpc_inject_rx_delay);
51 		rx_queue = &local->rx_delay_queue;
52 	}
53 #endif
54 
55 	skb_queue_tail(rx_queue, skb);
56 	wake_up_process(io_thread);
57 	return 0;
58 }
59 
60 /*
61  * Handle an error received on the local endpoint.
62  */
63 void rxrpc_error_report(struct sock *sk)
64 {
65 	struct rxrpc_local *local;
66 	struct sk_buff *skb;
67 
68 	rcu_read_lock();
69 	local = rcu_dereference_sk_user_data(sk);
70 	if (unlikely(!local)) {
71 		rcu_read_unlock();
72 		return;
73 	}
74 
75 	while ((skb = skb_dequeue(&sk->sk_error_queue))) {
76 		skb->mark = RXRPC_SKB_MARK_ERROR;
77 		rxrpc_new_skb(skb, rxrpc_skb_new_error_report);
78 		skb_queue_tail(&local->rx_queue, skb);
79 	}
80 
81 	rxrpc_wake_up_io_thread(local);
82 	rcu_read_unlock();
83 }
84 
85 /*
86  * Directly produce an abort from a packet.
87  */
88 bool rxrpc_direct_abort(struct sk_buff *skb, enum rxrpc_abort_reason why,
89 			s32 abort_code, int err)
90 {
91 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
92 
93 	trace_rxrpc_abort(0, why, sp->hdr.cid, sp->hdr.callNumber, sp->hdr.seq,
94 			  abort_code, err);
95 	skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
96 	skb->priority = abort_code;
97 	return false;
98 }
99 
100 static bool rxrpc_bad_message(struct sk_buff *skb, enum rxrpc_abort_reason why)
101 {
102 	return rxrpc_direct_abort(skb, why, RX_PROTOCOL_ERROR, -EBADMSG);
103 }
104 
105 #define just_discard true
106 
107 /*
108  * Process event packets targeted at a local endpoint.
109  */
110 static bool rxrpc_input_version(struct rxrpc_local *local, struct sk_buff *skb)
111 {
112 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
113 	char v;
114 
115 	_enter("");
116 
117 	rxrpc_see_skb(skb, rxrpc_skb_see_version);
118 	if (skb_copy_bits(skb, sizeof(struct rxrpc_wire_header), &v, 1) >= 0) {
119 		if (v == 0)
120 			rxrpc_send_version_request(local, &sp->hdr, skb);
121 	}
122 
123 	return true;
124 }
125 
126 /*
127  * Extract the wire header from a packet and translate the byte order.
128  */
129 static bool rxrpc_extract_header(struct rxrpc_skb_priv *sp,
130 				 struct sk_buff *skb)
131 {
132 	struct rxrpc_wire_header whdr;
133 	struct rxrpc_ackpacket ack;
134 
135 	/* dig out the RxRPC connection details */
136 	if (skb_copy_bits(skb, 0, &whdr, sizeof(whdr)) < 0)
137 		return rxrpc_bad_message(skb, rxrpc_badmsg_short_hdr);
138 
139 	memset(sp, 0, sizeof(*sp));
140 	sp->hdr.epoch		= ntohl(whdr.epoch);
141 	sp->hdr.cid		= ntohl(whdr.cid);
142 	sp->hdr.callNumber	= ntohl(whdr.callNumber);
143 	sp->hdr.seq		= ntohl(whdr.seq);
144 	sp->hdr.serial		= ntohl(whdr.serial);
145 	sp->hdr.flags		= whdr.flags;
146 	sp->hdr.type		= whdr.type;
147 	sp->hdr.userStatus	= whdr.userStatus;
148 	sp->hdr.securityIndex	= whdr.securityIndex;
149 	sp->hdr._rsvd		= ntohs(whdr._rsvd);
150 	sp->hdr.serviceId	= ntohs(whdr.serviceId);
151 
152 	if (sp->hdr.type == RXRPC_PACKET_TYPE_ACK) {
153 		if (skb_copy_bits(skb, sizeof(whdr), &ack, sizeof(ack)) < 0)
154 			return rxrpc_bad_message(skb, rxrpc_badmsg_short_ack);
155 		sp->ack.first_ack	= ntohl(ack.firstPacket);
156 		sp->ack.prev_ack	= ntohl(ack.previousPacket);
157 		sp->ack.acked_serial	= ntohl(ack.serial);
158 		sp->ack.reason		= ack.reason;
159 		sp->ack.nr_acks		= ack.nAcks;
160 	}
161 	return true;
162 }
163 
164 /*
165  * Extract the abort code from an ABORT packet and stash it in skb->priority.
166  */
167 static bool rxrpc_extract_abort(struct sk_buff *skb)
168 {
169 	__be32 wtmp;
170 
171 	if (skb_copy_bits(skb, sizeof(struct rxrpc_wire_header),
172 			  &wtmp, sizeof(wtmp)) < 0)
173 		return false;
174 	skb->priority = ntohl(wtmp);
175 	return true;
176 }
177 
178 /*
179  * Process packets received on the local endpoint
180  */
181 static bool rxrpc_input_packet(struct rxrpc_local *local, struct sk_buff **_skb)
182 {
183 	struct rxrpc_connection *conn;
184 	struct sockaddr_rxrpc peer_srx;
185 	struct rxrpc_skb_priv *sp;
186 	struct rxrpc_peer *peer = NULL;
187 	struct sk_buff *skb = *_skb;
188 	bool ret = false;
189 
190 	skb_pull(skb, sizeof(struct udphdr));
191 
192 	sp = rxrpc_skb(skb);
193 
194 	/* dig out the RxRPC connection details */
195 	if (!rxrpc_extract_header(sp, skb))
196 		return just_discard;
197 
198 	if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) {
199 		static int lose;
200 		if ((lose++ & 7) == 7) {
201 			trace_rxrpc_rx_lose(sp);
202 			return just_discard;
203 		}
204 	}
205 
206 	trace_rxrpc_rx_packet(sp);
207 
208 	switch (sp->hdr.type) {
209 	case RXRPC_PACKET_TYPE_VERSION:
210 		if (rxrpc_to_client(sp))
211 			return just_discard;
212 		return rxrpc_input_version(local, skb);
213 
214 	case RXRPC_PACKET_TYPE_BUSY:
215 		if (rxrpc_to_server(sp))
216 			return just_discard;
217 		fallthrough;
218 	case RXRPC_PACKET_TYPE_ACK:
219 	case RXRPC_PACKET_TYPE_ACKALL:
220 		if (sp->hdr.callNumber == 0)
221 			return rxrpc_bad_message(skb, rxrpc_badmsg_zero_call);
222 		break;
223 	case RXRPC_PACKET_TYPE_ABORT:
224 		if (!rxrpc_extract_abort(skb))
225 			return just_discard; /* Just discard if malformed */
226 		break;
227 
228 	case RXRPC_PACKET_TYPE_DATA:
229 		if (sp->hdr.callNumber == 0)
230 			return rxrpc_bad_message(skb, rxrpc_badmsg_zero_call);
231 		if (sp->hdr.seq == 0)
232 			return rxrpc_bad_message(skb, rxrpc_badmsg_zero_seq);
233 
234 		/* Unshare the packet so that it can be modified for in-place
235 		 * decryption.
236 		 */
237 		if (sp->hdr.securityIndex != 0) {
238 			skb = skb_unshare(skb, GFP_ATOMIC);
239 			if (!skb) {
240 				rxrpc_eaten_skb(*_skb, rxrpc_skb_eaten_by_unshare_nomem);
241 				*_skb = NULL;
242 				return just_discard;
243 			}
244 
245 			if (skb != *_skb) {
246 				rxrpc_eaten_skb(*_skb, rxrpc_skb_eaten_by_unshare);
247 				*_skb = skb;
248 				rxrpc_new_skb(skb, rxrpc_skb_new_unshared);
249 				sp = rxrpc_skb(skb);
250 			}
251 		}
252 		break;
253 
254 	case RXRPC_PACKET_TYPE_CHALLENGE:
255 		if (rxrpc_to_server(sp))
256 			return just_discard;
257 		break;
258 	case RXRPC_PACKET_TYPE_RESPONSE:
259 		if (rxrpc_to_client(sp))
260 			return just_discard;
261 		break;
262 
263 		/* Packet types 9-11 should just be ignored. */
264 	case RXRPC_PACKET_TYPE_PARAMS:
265 	case RXRPC_PACKET_TYPE_10:
266 	case RXRPC_PACKET_TYPE_11:
267 		return just_discard;
268 
269 	default:
270 		return rxrpc_bad_message(skb, rxrpc_badmsg_unsupported_packet);
271 	}
272 
273 	if (sp->hdr.serviceId == 0)
274 		return rxrpc_bad_message(skb, rxrpc_badmsg_zero_service);
275 
276 	if (WARN_ON_ONCE(rxrpc_extract_addr_from_skb(&peer_srx, skb) < 0))
277 		return just_discard; /* Unsupported address type. */
278 
279 	if (peer_srx.transport.family != local->srx.transport.family &&
280 	    (peer_srx.transport.family == AF_INET &&
281 	     local->srx.transport.family != AF_INET6)) {
282 		pr_warn_ratelimited("AF_RXRPC: Protocol mismatch %u not %u\n",
283 				    peer_srx.transport.family,
284 				    local->srx.transport.family);
285 		return just_discard; /* Wrong address type. */
286 	}
287 
288 	if (rxrpc_to_client(sp)) {
289 		rcu_read_lock();
290 		conn = rxrpc_find_client_connection_rcu(local, &peer_srx, skb);
291 		conn = rxrpc_get_connection_maybe(conn, rxrpc_conn_get_call_input);
292 		rcu_read_unlock();
293 		if (!conn)
294 			return rxrpc_protocol_error(skb, rxrpc_eproto_no_client_conn);
295 
296 		ret = rxrpc_input_packet_on_conn(conn, &peer_srx, skb);
297 		rxrpc_put_connection(conn, rxrpc_conn_put_call_input);
298 		return ret;
299 	}
300 
301 	/* We need to look up service connections by the full protocol
302 	 * parameter set.  We look up the peer first as an intermediate step
303 	 * and then the connection from the peer's tree.
304 	 */
305 	rcu_read_lock();
306 
307 	peer = rxrpc_lookup_peer_rcu(local, &peer_srx);
308 	if (!peer) {
309 		rcu_read_unlock();
310 		return rxrpc_new_incoming_call(local, NULL, NULL, &peer_srx, skb);
311 	}
312 
313 	conn = rxrpc_find_service_conn_rcu(peer, skb);
314 	conn = rxrpc_get_connection_maybe(conn, rxrpc_conn_get_call_input);
315 	if (conn) {
316 		rcu_read_unlock();
317 		ret = rxrpc_input_packet_on_conn(conn, &peer_srx, skb);
318 		rxrpc_put_connection(conn, rxrpc_conn_put_call_input);
319 		return ret;
320 	}
321 
322 	peer = rxrpc_get_peer_maybe(peer, rxrpc_peer_get_input);
323 	rcu_read_unlock();
324 
325 	ret = rxrpc_new_incoming_call(local, peer, NULL, &peer_srx, skb);
326 	rxrpc_put_peer(peer, rxrpc_peer_put_input);
327 	return ret;
328 }
329 
330 /*
331  * Deal with a packet that's associated with an extant connection.
332  */
333 static int rxrpc_input_packet_on_conn(struct rxrpc_connection *conn,
334 				      struct sockaddr_rxrpc *peer_srx,
335 				      struct sk_buff *skb)
336 {
337 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
338 	struct rxrpc_channel *chan;
339 	struct rxrpc_call *call = NULL;
340 	unsigned int channel;
341 	bool ret;
342 
343 	if (sp->hdr.securityIndex != conn->security_ix)
344 		return rxrpc_direct_abort(skb, rxrpc_eproto_wrong_security,
345 					  RXKADINCONSISTENCY, -EBADMSG);
346 
347 	if (sp->hdr.serviceId != conn->service_id) {
348 		int old_id;
349 
350 		if (!test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags))
351 			return rxrpc_protocol_error(skb, rxrpc_eproto_reupgrade);
352 
353 		old_id = cmpxchg(&conn->service_id, conn->orig_service_id,
354 				 sp->hdr.serviceId);
355 		if (old_id != conn->orig_service_id &&
356 		    old_id != sp->hdr.serviceId)
357 			return rxrpc_protocol_error(skb, rxrpc_eproto_bad_upgrade);
358 	}
359 
360 	if (after(sp->hdr.serial, conn->hi_serial))
361 		conn->hi_serial = sp->hdr.serial;
362 
363 	/* It's a connection-level packet if the call number is 0. */
364 	if (sp->hdr.callNumber == 0)
365 		return rxrpc_input_conn_packet(conn, skb);
366 
367 	/* Call-bound packets are routed by connection channel. */
368 	channel = sp->hdr.cid & RXRPC_CHANNELMASK;
369 	chan = &conn->channels[channel];
370 
371 	/* Ignore really old calls */
372 	if (sp->hdr.callNumber < chan->last_call)
373 		return just_discard;
374 
375 	if (sp->hdr.callNumber == chan->last_call) {
376 		if (chan->call ||
377 		    sp->hdr.type == RXRPC_PACKET_TYPE_ABORT)
378 			return just_discard;
379 
380 		/* For the previous service call, if completed successfully, we
381 		 * discard all further packets.
382 		 */
383 		if (rxrpc_conn_is_service(conn) &&
384 		    chan->last_type == RXRPC_PACKET_TYPE_ACK)
385 			return just_discard;
386 
387 		/* But otherwise we need to retransmit the final packet from
388 		 * data cached in the connection record.
389 		 */
390 		if (sp->hdr.type == RXRPC_PACKET_TYPE_DATA)
391 			trace_rxrpc_rx_data(chan->call_debug_id,
392 					    sp->hdr.seq,
393 					    sp->hdr.serial,
394 					    sp->hdr.flags);
395 		rxrpc_conn_retransmit_call(conn, skb, channel);
396 		return just_discard;
397 	}
398 
399 	call = rxrpc_try_get_call(chan->call, rxrpc_call_get_input);
400 
401 	if (sp->hdr.callNumber > chan->call_id) {
402 		if (rxrpc_to_client(sp)) {
403 			rxrpc_put_call(call, rxrpc_call_put_input);
404 			return rxrpc_protocol_error(skb,
405 						    rxrpc_eproto_unexpected_implicit_end);
406 		}
407 
408 		if (call) {
409 			rxrpc_implicit_end_call(call, skb);
410 			rxrpc_put_call(call, rxrpc_call_put_input);
411 			call = NULL;
412 		}
413 	}
414 
415 	if (!call) {
416 		if (rxrpc_to_client(sp))
417 			return rxrpc_protocol_error(skb, rxrpc_eproto_no_client_call);
418 		return rxrpc_new_incoming_call(conn->local, conn->peer, conn,
419 					       peer_srx, skb);
420 	}
421 
422 	ret = rxrpc_input_call_event(call, skb);
423 	rxrpc_put_call(call, rxrpc_call_put_input);
424 	return ret;
425 }
426 
427 /*
428  * I/O and event handling thread.
429  */
430 int rxrpc_io_thread(void *data)
431 {
432 	struct rxrpc_connection *conn;
433 	struct sk_buff_head rx_queue;
434 	struct rxrpc_local *local = data;
435 	struct rxrpc_call *call;
436 	struct sk_buff *skb;
437 #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY
438 	ktime_t now;
439 #endif
440 	bool should_stop;
441 
442 	complete(&local->io_thread_ready);
443 
444 	skb_queue_head_init(&rx_queue);
445 
446 	set_user_nice(current, MIN_NICE);
447 
448 	for (;;) {
449 		rxrpc_inc_stat(local->rxnet, stat_io_loop);
450 
451 		/* Deal with connections that want immediate attention. */
452 		conn = list_first_entry_or_null(&local->conn_attend_q,
453 						struct rxrpc_connection,
454 						attend_link);
455 		if (conn) {
456 			spin_lock_bh(&local->lock);
457 			list_del_init(&conn->attend_link);
458 			spin_unlock_bh(&local->lock);
459 
460 			rxrpc_input_conn_event(conn, NULL);
461 			rxrpc_put_connection(conn, rxrpc_conn_put_poke);
462 			continue;
463 		}
464 
465 		if (test_and_clear_bit(RXRPC_CLIENT_CONN_REAP_TIMER,
466 				       &local->client_conn_flags))
467 			rxrpc_discard_expired_client_conns(local);
468 
469 		/* Deal with calls that want immediate attention. */
470 		if ((call = list_first_entry_or_null(&local->call_attend_q,
471 						     struct rxrpc_call,
472 						     attend_link))) {
473 			spin_lock_bh(&local->lock);
474 			list_del_init(&call->attend_link);
475 			spin_unlock_bh(&local->lock);
476 
477 			trace_rxrpc_call_poked(call);
478 			rxrpc_input_call_event(call, NULL);
479 			rxrpc_put_call(call, rxrpc_call_put_poke);
480 			continue;
481 		}
482 
483 		if (!list_empty(&local->new_client_calls))
484 			rxrpc_connect_client_calls(local);
485 
486 		/* Process received packets and errors. */
487 		if ((skb = __skb_dequeue(&rx_queue))) {
488 			struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
489 			switch (skb->mark) {
490 			case RXRPC_SKB_MARK_PACKET:
491 				skb->priority = 0;
492 				if (!rxrpc_input_packet(local, &skb))
493 					rxrpc_reject_packet(local, skb);
494 				trace_rxrpc_rx_done(skb->mark, skb->priority);
495 				rxrpc_free_skb(skb, rxrpc_skb_put_input);
496 				break;
497 			case RXRPC_SKB_MARK_ERROR:
498 				rxrpc_input_error(local, skb);
499 				rxrpc_free_skb(skb, rxrpc_skb_put_error_report);
500 				break;
501 			case RXRPC_SKB_MARK_SERVICE_CONN_SECURED:
502 				rxrpc_input_conn_event(sp->conn, skb);
503 				rxrpc_put_connection(sp->conn, rxrpc_conn_put_poke);
504 				rxrpc_free_skb(skb, rxrpc_skb_put_conn_secured);
505 				break;
506 			default:
507 				WARN_ON_ONCE(1);
508 				rxrpc_free_skb(skb, rxrpc_skb_put_unknown);
509 				break;
510 			}
511 			continue;
512 		}
513 
514 		/* Inject a delay into packets if requested. */
515 #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY
516 		now = ktime_get_real();
517 		while ((skb = skb_peek(&local->rx_delay_queue))) {
518 			if (ktime_before(now, skb->tstamp))
519 				break;
520 			skb = skb_dequeue(&local->rx_delay_queue);
521 			skb_queue_tail(&local->rx_queue, skb);
522 		}
523 #endif
524 
525 		if (!skb_queue_empty(&local->rx_queue)) {
526 			spin_lock_irq(&local->rx_queue.lock);
527 			skb_queue_splice_tail_init(&local->rx_queue, &rx_queue);
528 			spin_unlock_irq(&local->rx_queue.lock);
529 			continue;
530 		}
531 
532 		set_current_state(TASK_INTERRUPTIBLE);
533 		should_stop = kthread_should_stop();
534 		if (!skb_queue_empty(&local->rx_queue) ||
535 		    !list_empty(&local->call_attend_q) ||
536 		    !list_empty(&local->conn_attend_q) ||
537 		    !list_empty(&local->new_client_calls) ||
538 		    test_bit(RXRPC_CLIENT_CONN_REAP_TIMER,
539 			     &local->client_conn_flags)) {
540 			__set_current_state(TASK_RUNNING);
541 			continue;
542 		}
543 
544 		if (should_stop)
545 			break;
546 
547 #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY
548 		skb = skb_peek(&local->rx_delay_queue);
549 		if (skb) {
550 			unsigned long timeout;
551 			ktime_t tstamp = skb->tstamp;
552 			ktime_t now = ktime_get_real();
553 			s64 delay_ns = ktime_to_ns(ktime_sub(tstamp, now));
554 
555 			if (delay_ns <= 0) {
556 				__set_current_state(TASK_RUNNING);
557 				continue;
558 			}
559 
560 			timeout = nsecs_to_jiffies(delay_ns);
561 			timeout = max(timeout, 1UL);
562 			schedule_timeout(timeout);
563 			__set_current_state(TASK_RUNNING);
564 			continue;
565 		}
566 #endif
567 
568 		schedule();
569 	}
570 
571 	__set_current_state(TASK_RUNNING);
572 	rxrpc_see_local(local, rxrpc_local_stop);
573 	rxrpc_destroy_local(local);
574 	WRITE_ONCE(local->io_thread, NULL);
575 	rxrpc_see_local(local, rxrpc_local_stopped);
576 	return 0;
577 }
578