xref: /linux/net/rds/recv.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
1 /*
2  * Copyright (c) 2006 Oracle.  All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
35 #include <net/sock.h>
36 #include <linux/in.h>
37 #include <linux/export.h>
38 
39 #include "rds.h"
40 
41 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
42 		  __be32 saddr)
43 {
44 	atomic_set(&inc->i_refcount, 1);
45 	INIT_LIST_HEAD(&inc->i_item);
46 	inc->i_conn = conn;
47 	inc->i_saddr = saddr;
48 	inc->i_rdma_cookie = 0;
49 }
50 EXPORT_SYMBOL_GPL(rds_inc_init);
51 
52 static void rds_inc_addref(struct rds_incoming *inc)
53 {
54 	rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
55 	atomic_inc(&inc->i_refcount);
56 }
57 
58 void rds_inc_put(struct rds_incoming *inc)
59 {
60 	rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
61 	if (atomic_dec_and_test(&inc->i_refcount)) {
62 		BUG_ON(!list_empty(&inc->i_item));
63 
64 		inc->i_conn->c_trans->inc_free(inc);
65 	}
66 }
67 EXPORT_SYMBOL_GPL(rds_inc_put);
68 
69 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
70 				  struct rds_cong_map *map,
71 				  int delta, __be16 port)
72 {
73 	int now_congested;
74 
75 	if (delta == 0)
76 		return;
77 
78 	rs->rs_rcv_bytes += delta;
79 	now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
80 
81 	rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d "
82 	  "now_cong %d delta %d\n",
83 	  rs, &rs->rs_bound_addr,
84 	  ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
85 	  rds_sk_rcvbuf(rs), now_congested, delta);
86 
87 	/* wasn't -> am congested */
88 	if (!rs->rs_congested && now_congested) {
89 		rs->rs_congested = 1;
90 		rds_cong_set_bit(map, port);
91 		rds_cong_queue_updates(map);
92 	}
93 	/* was -> aren't congested */
94 	/* Require more free space before reporting uncongested to prevent
95 	   bouncing cong/uncong state too often */
96 	else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
97 		rs->rs_congested = 0;
98 		rds_cong_clear_bit(map, port);
99 		rds_cong_queue_updates(map);
100 	}
101 
102 	/* do nothing if no change in cong state */
103 }
104 
105 /*
106  * Process all extension headers that come with this message.
107  */
108 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
109 {
110 	struct rds_header *hdr = &inc->i_hdr;
111 	unsigned int pos = 0, type, len;
112 	union {
113 		struct rds_ext_header_version version;
114 		struct rds_ext_header_rdma rdma;
115 		struct rds_ext_header_rdma_dest rdma_dest;
116 	} buffer;
117 
118 	while (1) {
119 		len = sizeof(buffer);
120 		type = rds_message_next_extension(hdr, &pos, &buffer, &len);
121 		if (type == RDS_EXTHDR_NONE)
122 			break;
123 		/* Process extension header here */
124 		switch (type) {
125 		case RDS_EXTHDR_RDMA:
126 			rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
127 			break;
128 
129 		case RDS_EXTHDR_RDMA_DEST:
130 			/* We ignore the size for now. We could stash it
131 			 * somewhere and use it for error checking. */
132 			inc->i_rdma_cookie = rds_rdma_make_cookie(
133 					be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
134 					be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
135 
136 			break;
137 		}
138 	}
139 }
140 
141 /*
142  * The transport must make sure that this is serialized against other
143  * rx and conn reset on this specific conn.
144  *
145  * We currently assert that only one fragmented message will be sent
146  * down a connection at a time.  This lets us reassemble in the conn
147  * instead of per-flow which means that we don't have to go digging through
148  * flows to tear down partial reassembly progress on conn failure and
149  * we save flow lookup and locking for each frag arrival.  It does mean
150  * that small messages will wait behind large ones.  Fragmenting at all
151  * is only to reduce the memory consumption of pre-posted buffers.
152  *
153  * The caller passes in saddr and daddr instead of us getting it from the
154  * conn.  This lets loopback, who only has one conn for both directions,
155  * tell us which roles the addrs in the conn are playing for this message.
156  */
157 void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
158 		       struct rds_incoming *inc, gfp_t gfp)
159 {
160 	struct rds_sock *rs = NULL;
161 	struct sock *sk;
162 	unsigned long flags;
163 
164 	inc->i_conn = conn;
165 	inc->i_rx_jiffies = jiffies;
166 
167 	rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
168 		 "flags 0x%x rx_jiffies %lu\n", conn,
169 		 (unsigned long long)conn->c_next_rx_seq,
170 		 inc,
171 		 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
172 		 be32_to_cpu(inc->i_hdr.h_len),
173 		 be16_to_cpu(inc->i_hdr.h_sport),
174 		 be16_to_cpu(inc->i_hdr.h_dport),
175 		 inc->i_hdr.h_flags,
176 		 inc->i_rx_jiffies);
177 
178 	/*
179 	 * Sequence numbers should only increase.  Messages get their
180 	 * sequence number as they're queued in a sending conn.  They
181 	 * can be dropped, though, if the sending socket is closed before
182 	 * they hit the wire.  So sequence numbers can skip forward
183 	 * under normal operation.  They can also drop back in the conn
184 	 * failover case as previously sent messages are resent down the
185 	 * new instance of a conn.  We drop those, otherwise we have
186 	 * to assume that the next valid seq does not come after a
187 	 * hole in the fragment stream.
188 	 *
189 	 * The headers don't give us a way to realize if fragments of
190 	 * a message have been dropped.  We assume that frags that arrive
191 	 * to a flow are part of the current message on the flow that is
192 	 * being reassembled.  This means that senders can't drop messages
193 	 * from the sending conn until all their frags are sent.
194 	 *
195 	 * XXX we could spend more on the wire to get more robust failure
196 	 * detection, arguably worth it to avoid data corruption.
197 	 */
198 	if (be64_to_cpu(inc->i_hdr.h_sequence) < conn->c_next_rx_seq &&
199 	    (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
200 		rds_stats_inc(s_recv_drop_old_seq);
201 		goto out;
202 	}
203 	conn->c_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
204 
205 	if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
206 		rds_stats_inc(s_recv_ping);
207 		rds_send_pong(conn, inc->i_hdr.h_sport);
208 		goto out;
209 	}
210 
211 	rs = rds_find_bound(daddr, inc->i_hdr.h_dport);
212 	if (!rs) {
213 		rds_stats_inc(s_recv_drop_no_sock);
214 		goto out;
215 	}
216 
217 	/* Process extension headers */
218 	rds_recv_incoming_exthdrs(inc, rs);
219 
220 	/* We can be racing with rds_release() which marks the socket dead. */
221 	sk = rds_rs_to_sk(rs);
222 
223 	/* serialize with rds_release -> sock_orphan */
224 	write_lock_irqsave(&rs->rs_recv_lock, flags);
225 	if (!sock_flag(sk, SOCK_DEAD)) {
226 		rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
227 		rds_stats_inc(s_recv_queued);
228 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
229 				      be32_to_cpu(inc->i_hdr.h_len),
230 				      inc->i_hdr.h_dport);
231 		rds_inc_addref(inc);
232 		list_add_tail(&inc->i_item, &rs->rs_recv_queue);
233 		__rds_wake_sk_sleep(sk);
234 	} else {
235 		rds_stats_inc(s_recv_drop_dead_sock);
236 	}
237 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
238 
239 out:
240 	if (rs)
241 		rds_sock_put(rs);
242 }
243 EXPORT_SYMBOL_GPL(rds_recv_incoming);
244 
245 /*
246  * be very careful here.  This is being called as the condition in
247  * wait_event_*() needs to cope with being called many times.
248  */
249 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
250 {
251 	unsigned long flags;
252 
253 	if (!*inc) {
254 		read_lock_irqsave(&rs->rs_recv_lock, flags);
255 		if (!list_empty(&rs->rs_recv_queue)) {
256 			*inc = list_entry(rs->rs_recv_queue.next,
257 					  struct rds_incoming,
258 					  i_item);
259 			rds_inc_addref(*inc);
260 		}
261 		read_unlock_irqrestore(&rs->rs_recv_lock, flags);
262 	}
263 
264 	return *inc != NULL;
265 }
266 
267 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
268 			    int drop)
269 {
270 	struct sock *sk = rds_rs_to_sk(rs);
271 	int ret = 0;
272 	unsigned long flags;
273 
274 	write_lock_irqsave(&rs->rs_recv_lock, flags);
275 	if (!list_empty(&inc->i_item)) {
276 		ret = 1;
277 		if (drop) {
278 			/* XXX make sure this i_conn is reliable */
279 			rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
280 					      -be32_to_cpu(inc->i_hdr.h_len),
281 					      inc->i_hdr.h_dport);
282 			list_del_init(&inc->i_item);
283 			rds_inc_put(inc);
284 		}
285 	}
286 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
287 
288 	rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
289 	return ret;
290 }
291 
292 /*
293  * Pull errors off the error queue.
294  * If msghdr is NULL, we will just purge the error queue.
295  */
296 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
297 {
298 	struct rds_notifier *notifier;
299 	struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
300 	unsigned int count = 0, max_messages = ~0U;
301 	unsigned long flags;
302 	LIST_HEAD(copy);
303 	int err = 0;
304 
305 
306 	/* put_cmsg copies to user space and thus may sleep. We can't do this
307 	 * with rs_lock held, so first grab as many notifications as we can stuff
308 	 * in the user provided cmsg buffer. We don't try to copy more, to avoid
309 	 * losing notifications - except when the buffer is so small that it wouldn't
310 	 * even hold a single notification. Then we give him as much of this single
311 	 * msg as we can squeeze in, and set MSG_CTRUNC.
312 	 */
313 	if (msghdr) {
314 		max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
315 		if (!max_messages)
316 			max_messages = 1;
317 	}
318 
319 	spin_lock_irqsave(&rs->rs_lock, flags);
320 	while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
321 		notifier = list_entry(rs->rs_notify_queue.next,
322 				struct rds_notifier, n_list);
323 		list_move(&notifier->n_list, &copy);
324 		count++;
325 	}
326 	spin_unlock_irqrestore(&rs->rs_lock, flags);
327 
328 	if (!count)
329 		return 0;
330 
331 	while (!list_empty(&copy)) {
332 		notifier = list_entry(copy.next, struct rds_notifier, n_list);
333 
334 		if (msghdr) {
335 			cmsg.user_token = notifier->n_user_token;
336 			cmsg.status = notifier->n_status;
337 
338 			err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
339 				       sizeof(cmsg), &cmsg);
340 			if (err)
341 				break;
342 		}
343 
344 		list_del_init(&notifier->n_list);
345 		kfree(notifier);
346 	}
347 
348 	/* If we bailed out because of an error in put_cmsg,
349 	 * we may be left with one or more notifications that we
350 	 * didn't process. Return them to the head of the list. */
351 	if (!list_empty(&copy)) {
352 		spin_lock_irqsave(&rs->rs_lock, flags);
353 		list_splice(&copy, &rs->rs_notify_queue);
354 		spin_unlock_irqrestore(&rs->rs_lock, flags);
355 	}
356 
357 	return err;
358 }
359 
360 /*
361  * Queue a congestion notification
362  */
363 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
364 {
365 	uint64_t notify = rs->rs_cong_notify;
366 	unsigned long flags;
367 	int err;
368 
369 	err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
370 			sizeof(notify), &notify);
371 	if (err)
372 		return err;
373 
374 	spin_lock_irqsave(&rs->rs_lock, flags);
375 	rs->rs_cong_notify &= ~notify;
376 	spin_unlock_irqrestore(&rs->rs_lock, flags);
377 
378 	return 0;
379 }
380 
381 /*
382  * Receive any control messages.
383  */
384 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg)
385 {
386 	int ret = 0;
387 
388 	if (inc->i_rdma_cookie) {
389 		ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
390 				sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
391 		if (ret)
392 			return ret;
393 	}
394 
395 	return 0;
396 }
397 
398 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
399 		int msg_flags)
400 {
401 	struct sock *sk = sock->sk;
402 	struct rds_sock *rs = rds_sk_to_rs(sk);
403 	long timeo;
404 	int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
405 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
406 	struct rds_incoming *inc = NULL;
407 
408 	/* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
409 	timeo = sock_rcvtimeo(sk, nonblock);
410 
411 	rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
412 
413 	if (msg_flags & MSG_OOB)
414 		goto out;
415 
416 	while (1) {
417 		struct iov_iter save;
418 		/* If there are pending notifications, do those - and nothing else */
419 		if (!list_empty(&rs->rs_notify_queue)) {
420 			ret = rds_notify_queue_get(rs, msg);
421 			break;
422 		}
423 
424 		if (rs->rs_cong_notify) {
425 			ret = rds_notify_cong(rs, msg);
426 			break;
427 		}
428 
429 		if (!rds_next_incoming(rs, &inc)) {
430 			if (nonblock) {
431 				ret = -EAGAIN;
432 				break;
433 			}
434 
435 			timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
436 					(!list_empty(&rs->rs_notify_queue) ||
437 					 rs->rs_cong_notify ||
438 					 rds_next_incoming(rs, &inc)), timeo);
439 			rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
440 				 timeo);
441 			if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
442 				continue;
443 
444 			ret = timeo;
445 			if (ret == 0)
446 				ret = -ETIMEDOUT;
447 			break;
448 		}
449 
450 		rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
451 			 &inc->i_conn->c_faddr,
452 			 ntohs(inc->i_hdr.h_sport));
453 		save = msg->msg_iter;
454 		ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
455 		if (ret < 0)
456 			break;
457 
458 		/*
459 		 * if the message we just copied isn't at the head of the
460 		 * recv queue then someone else raced us to return it, try
461 		 * to get the next message.
462 		 */
463 		if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
464 			rds_inc_put(inc);
465 			inc = NULL;
466 			rds_stats_inc(s_recv_deliver_raced);
467 			msg->msg_iter = save;
468 			continue;
469 		}
470 
471 		if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
472 			if (msg_flags & MSG_TRUNC)
473 				ret = be32_to_cpu(inc->i_hdr.h_len);
474 			msg->msg_flags |= MSG_TRUNC;
475 		}
476 
477 		if (rds_cmsg_recv(inc, msg)) {
478 			ret = -EFAULT;
479 			goto out;
480 		}
481 
482 		rds_stats_inc(s_recv_delivered);
483 
484 		if (sin) {
485 			sin->sin_family = AF_INET;
486 			sin->sin_port = inc->i_hdr.h_sport;
487 			sin->sin_addr.s_addr = inc->i_saddr;
488 			memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
489 			msg->msg_namelen = sizeof(*sin);
490 		}
491 		break;
492 	}
493 
494 	if (inc)
495 		rds_inc_put(inc);
496 
497 out:
498 	return ret;
499 }
500 
501 /*
502  * The socket is being shut down and we're asked to drop messages that were
503  * queued for recvmsg.  The caller has unbound the socket so the receive path
504  * won't queue any more incoming fragments or messages on the socket.
505  */
506 void rds_clear_recv_queue(struct rds_sock *rs)
507 {
508 	struct sock *sk = rds_rs_to_sk(rs);
509 	struct rds_incoming *inc, *tmp;
510 	unsigned long flags;
511 
512 	write_lock_irqsave(&rs->rs_recv_lock, flags);
513 	list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
514 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
515 				      -be32_to_cpu(inc->i_hdr.h_len),
516 				      inc->i_hdr.h_dport);
517 		list_del_init(&inc->i_item);
518 		rds_inc_put(inc);
519 	}
520 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
521 }
522 
523 /*
524  * inc->i_saddr isn't used here because it is only set in the receive
525  * path.
526  */
527 void rds_inc_info_copy(struct rds_incoming *inc,
528 		       struct rds_info_iterator *iter,
529 		       __be32 saddr, __be32 daddr, int flip)
530 {
531 	struct rds_info_message minfo;
532 
533 	minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
534 	minfo.len = be32_to_cpu(inc->i_hdr.h_len);
535 
536 	if (flip) {
537 		minfo.laddr = daddr;
538 		minfo.faddr = saddr;
539 		minfo.lport = inc->i_hdr.h_dport;
540 		minfo.fport = inc->i_hdr.h_sport;
541 	} else {
542 		minfo.laddr = saddr;
543 		minfo.faddr = daddr;
544 		minfo.lport = inc->i_hdr.h_sport;
545 		minfo.fport = inc->i_hdr.h_dport;
546 	}
547 
548 	rds_info_copy(iter, &minfo, sizeof(minfo));
549 }
550