xref: /openbmc/linux/net/rds/recv.c (revision 78700c0a)
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 #include <linux/time.h>
39 #include <linux/rds.h>
40 
41 #include "rds.h"
42 
43 void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
44 		  __be32 saddr)
45 {
46 	atomic_set(&inc->i_refcount, 1);
47 	INIT_LIST_HEAD(&inc->i_item);
48 	inc->i_conn = conn;
49 	inc->i_saddr = saddr;
50 	inc->i_rdma_cookie = 0;
51 	inc->i_rx_tstamp.tv_sec = 0;
52 	inc->i_rx_tstamp.tv_usec = 0;
53 }
54 EXPORT_SYMBOL_GPL(rds_inc_init);
55 
56 static void rds_inc_addref(struct rds_incoming *inc)
57 {
58 	rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
59 	atomic_inc(&inc->i_refcount);
60 }
61 
62 void rds_inc_put(struct rds_incoming *inc)
63 {
64 	rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
65 	if (atomic_dec_and_test(&inc->i_refcount)) {
66 		BUG_ON(!list_empty(&inc->i_item));
67 
68 		inc->i_conn->c_trans->inc_free(inc);
69 	}
70 }
71 EXPORT_SYMBOL_GPL(rds_inc_put);
72 
73 static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
74 				  struct rds_cong_map *map,
75 				  int delta, __be16 port)
76 {
77 	int now_congested;
78 
79 	if (delta == 0)
80 		return;
81 
82 	rs->rs_rcv_bytes += delta;
83 	now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);
84 
85 	rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d "
86 	  "now_cong %d delta %d\n",
87 	  rs, &rs->rs_bound_addr,
88 	  ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
89 	  rds_sk_rcvbuf(rs), now_congested, delta);
90 
91 	/* wasn't -> am congested */
92 	if (!rs->rs_congested && now_congested) {
93 		rs->rs_congested = 1;
94 		rds_cong_set_bit(map, port);
95 		rds_cong_queue_updates(map);
96 	}
97 	/* was -> aren't congested */
98 	/* Require more free space before reporting uncongested to prevent
99 	   bouncing cong/uncong state too often */
100 	else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
101 		rs->rs_congested = 0;
102 		rds_cong_clear_bit(map, port);
103 		rds_cong_queue_updates(map);
104 	}
105 
106 	/* do nothing if no change in cong state */
107 }
108 
109 /*
110  * Process all extension headers that come with this message.
111  */
112 static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
113 {
114 	struct rds_header *hdr = &inc->i_hdr;
115 	unsigned int pos = 0, type, len;
116 	union {
117 		struct rds_ext_header_version version;
118 		struct rds_ext_header_rdma rdma;
119 		struct rds_ext_header_rdma_dest rdma_dest;
120 	} buffer;
121 
122 	while (1) {
123 		len = sizeof(buffer);
124 		type = rds_message_next_extension(hdr, &pos, &buffer, &len);
125 		if (type == RDS_EXTHDR_NONE)
126 			break;
127 		/* Process extension header here */
128 		switch (type) {
129 		case RDS_EXTHDR_RDMA:
130 			rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
131 			break;
132 
133 		case RDS_EXTHDR_RDMA_DEST:
134 			/* We ignore the size for now. We could stash it
135 			 * somewhere and use it for error checking. */
136 			inc->i_rdma_cookie = rds_rdma_make_cookie(
137 					be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
138 					be32_to_cpu(buffer.rdma_dest.h_rdma_offset));
139 
140 			break;
141 		}
142 	}
143 }
144 
145 /*
146  * The transport must make sure that this is serialized against other
147  * rx and conn reset on this specific conn.
148  *
149  * We currently assert that only one fragmented message will be sent
150  * down a connection at a time.  This lets us reassemble in the conn
151  * instead of per-flow which means that we don't have to go digging through
152  * flows to tear down partial reassembly progress on conn failure and
153  * we save flow lookup and locking for each frag arrival.  It does mean
154  * that small messages will wait behind large ones.  Fragmenting at all
155  * is only to reduce the memory consumption of pre-posted buffers.
156  *
157  * The caller passes in saddr and daddr instead of us getting it from the
158  * conn.  This lets loopback, who only has one conn for both directions,
159  * tell us which roles the addrs in the conn are playing for this message.
160  */
161 void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
162 		       struct rds_incoming *inc, gfp_t gfp)
163 {
164 	struct rds_sock *rs = NULL;
165 	struct sock *sk;
166 	unsigned long flags;
167 
168 	inc->i_conn = conn;
169 	inc->i_rx_jiffies = jiffies;
170 
171 	rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
172 		 "flags 0x%x rx_jiffies %lu\n", conn,
173 		 (unsigned long long)conn->c_next_rx_seq,
174 		 inc,
175 		 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
176 		 be32_to_cpu(inc->i_hdr.h_len),
177 		 be16_to_cpu(inc->i_hdr.h_sport),
178 		 be16_to_cpu(inc->i_hdr.h_dport),
179 		 inc->i_hdr.h_flags,
180 		 inc->i_rx_jiffies);
181 
182 	/*
183 	 * Sequence numbers should only increase.  Messages get their
184 	 * sequence number as they're queued in a sending conn.  They
185 	 * can be dropped, though, if the sending socket is closed before
186 	 * they hit the wire.  So sequence numbers can skip forward
187 	 * under normal operation.  They can also drop back in the conn
188 	 * failover case as previously sent messages are resent down the
189 	 * new instance of a conn.  We drop those, otherwise we have
190 	 * to assume that the next valid seq does not come after a
191 	 * hole in the fragment stream.
192 	 *
193 	 * The headers don't give us a way to realize if fragments of
194 	 * a message have been dropped.  We assume that frags that arrive
195 	 * to a flow are part of the current message on the flow that is
196 	 * being reassembled.  This means that senders can't drop messages
197 	 * from the sending conn until all their frags are sent.
198 	 *
199 	 * XXX we could spend more on the wire to get more robust failure
200 	 * detection, arguably worth it to avoid data corruption.
201 	 */
202 	if (be64_to_cpu(inc->i_hdr.h_sequence) < conn->c_next_rx_seq &&
203 	    (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
204 		rds_stats_inc(s_recv_drop_old_seq);
205 		goto out;
206 	}
207 	conn->c_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;
208 
209 	if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
210 		rds_stats_inc(s_recv_ping);
211 		rds_send_pong(conn, inc->i_hdr.h_sport);
212 		goto out;
213 	}
214 
215 	rs = rds_find_bound(daddr, inc->i_hdr.h_dport);
216 	if (!rs) {
217 		rds_stats_inc(s_recv_drop_no_sock);
218 		goto out;
219 	}
220 
221 	/* Process extension headers */
222 	rds_recv_incoming_exthdrs(inc, rs);
223 
224 	/* We can be racing with rds_release() which marks the socket dead. */
225 	sk = rds_rs_to_sk(rs);
226 
227 	/* serialize with rds_release -> sock_orphan */
228 	write_lock_irqsave(&rs->rs_recv_lock, flags);
229 	if (!sock_flag(sk, SOCK_DEAD)) {
230 		rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
231 		rds_stats_inc(s_recv_queued);
232 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
233 				      be32_to_cpu(inc->i_hdr.h_len),
234 				      inc->i_hdr.h_dport);
235 		if (sock_flag(sk, SOCK_RCVTSTAMP))
236 			do_gettimeofday(&inc->i_rx_tstamp);
237 		rds_inc_addref(inc);
238 		list_add_tail(&inc->i_item, &rs->rs_recv_queue);
239 		__rds_wake_sk_sleep(sk);
240 	} else {
241 		rds_stats_inc(s_recv_drop_dead_sock);
242 	}
243 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
244 
245 out:
246 	if (rs)
247 		rds_sock_put(rs);
248 }
249 EXPORT_SYMBOL_GPL(rds_recv_incoming);
250 
251 /*
252  * be very careful here.  This is being called as the condition in
253  * wait_event_*() needs to cope with being called many times.
254  */
255 static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
256 {
257 	unsigned long flags;
258 
259 	if (!*inc) {
260 		read_lock_irqsave(&rs->rs_recv_lock, flags);
261 		if (!list_empty(&rs->rs_recv_queue)) {
262 			*inc = list_entry(rs->rs_recv_queue.next,
263 					  struct rds_incoming,
264 					  i_item);
265 			rds_inc_addref(*inc);
266 		}
267 		read_unlock_irqrestore(&rs->rs_recv_lock, flags);
268 	}
269 
270 	return *inc != NULL;
271 }
272 
273 static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
274 			    int drop)
275 {
276 	struct sock *sk = rds_rs_to_sk(rs);
277 	int ret = 0;
278 	unsigned long flags;
279 
280 	write_lock_irqsave(&rs->rs_recv_lock, flags);
281 	if (!list_empty(&inc->i_item)) {
282 		ret = 1;
283 		if (drop) {
284 			/* XXX make sure this i_conn is reliable */
285 			rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
286 					      -be32_to_cpu(inc->i_hdr.h_len),
287 					      inc->i_hdr.h_dport);
288 			list_del_init(&inc->i_item);
289 			rds_inc_put(inc);
290 		}
291 	}
292 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
293 
294 	rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
295 	return ret;
296 }
297 
298 /*
299  * Pull errors off the error queue.
300  * If msghdr is NULL, we will just purge the error queue.
301  */
302 int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
303 {
304 	struct rds_notifier *notifier;
305 	struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
306 	unsigned int count = 0, max_messages = ~0U;
307 	unsigned long flags;
308 	LIST_HEAD(copy);
309 	int err = 0;
310 
311 
312 	/* put_cmsg copies to user space and thus may sleep. We can't do this
313 	 * with rs_lock held, so first grab as many notifications as we can stuff
314 	 * in the user provided cmsg buffer. We don't try to copy more, to avoid
315 	 * losing notifications - except when the buffer is so small that it wouldn't
316 	 * even hold a single notification. Then we give him as much of this single
317 	 * msg as we can squeeze in, and set MSG_CTRUNC.
318 	 */
319 	if (msghdr) {
320 		max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
321 		if (!max_messages)
322 			max_messages = 1;
323 	}
324 
325 	spin_lock_irqsave(&rs->rs_lock, flags);
326 	while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
327 		notifier = list_entry(rs->rs_notify_queue.next,
328 				struct rds_notifier, n_list);
329 		list_move(&notifier->n_list, &copy);
330 		count++;
331 	}
332 	spin_unlock_irqrestore(&rs->rs_lock, flags);
333 
334 	if (!count)
335 		return 0;
336 
337 	while (!list_empty(&copy)) {
338 		notifier = list_entry(copy.next, struct rds_notifier, n_list);
339 
340 		if (msghdr) {
341 			cmsg.user_token = notifier->n_user_token;
342 			cmsg.status = notifier->n_status;
343 
344 			err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
345 				       sizeof(cmsg), &cmsg);
346 			if (err)
347 				break;
348 		}
349 
350 		list_del_init(&notifier->n_list);
351 		kfree(notifier);
352 	}
353 
354 	/* If we bailed out because of an error in put_cmsg,
355 	 * we may be left with one or more notifications that we
356 	 * didn't process. Return them to the head of the list. */
357 	if (!list_empty(&copy)) {
358 		spin_lock_irqsave(&rs->rs_lock, flags);
359 		list_splice(&copy, &rs->rs_notify_queue);
360 		spin_unlock_irqrestore(&rs->rs_lock, flags);
361 	}
362 
363 	return err;
364 }
365 
366 /*
367  * Queue a congestion notification
368  */
369 static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
370 {
371 	uint64_t notify = rs->rs_cong_notify;
372 	unsigned long flags;
373 	int err;
374 
375 	err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
376 			sizeof(notify), &notify);
377 	if (err)
378 		return err;
379 
380 	spin_lock_irqsave(&rs->rs_lock, flags);
381 	rs->rs_cong_notify &= ~notify;
382 	spin_unlock_irqrestore(&rs->rs_lock, flags);
383 
384 	return 0;
385 }
386 
387 /*
388  * Receive any control messages.
389  */
390 static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
391 			 struct rds_sock *rs)
392 {
393 	int ret = 0;
394 
395 	if (inc->i_rdma_cookie) {
396 		ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
397 				sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
398 		if (ret)
399 			return ret;
400 	}
401 
402 	if ((inc->i_rx_tstamp.tv_sec != 0) &&
403 	    sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
404 		ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
405 			       sizeof(struct timeval),
406 			       &inc->i_rx_tstamp);
407 		if (ret)
408 			return ret;
409 	}
410 
411 	return 0;
412 }
413 
414 int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
415 		int msg_flags)
416 {
417 	struct sock *sk = sock->sk;
418 	struct rds_sock *rs = rds_sk_to_rs(sk);
419 	long timeo;
420 	int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
421 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
422 	struct rds_incoming *inc = NULL;
423 
424 	/* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
425 	timeo = sock_rcvtimeo(sk, nonblock);
426 
427 	rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);
428 
429 	if (msg_flags & MSG_OOB)
430 		goto out;
431 
432 	while (1) {
433 		struct iov_iter save;
434 		/* If there are pending notifications, do those - and nothing else */
435 		if (!list_empty(&rs->rs_notify_queue)) {
436 			ret = rds_notify_queue_get(rs, msg);
437 			break;
438 		}
439 
440 		if (rs->rs_cong_notify) {
441 			ret = rds_notify_cong(rs, msg);
442 			break;
443 		}
444 
445 		if (!rds_next_incoming(rs, &inc)) {
446 			if (nonblock) {
447 				ret = -EAGAIN;
448 				break;
449 			}
450 
451 			timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
452 					(!list_empty(&rs->rs_notify_queue) ||
453 					 rs->rs_cong_notify ||
454 					 rds_next_incoming(rs, &inc)), timeo);
455 			rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
456 				 timeo);
457 			if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
458 				continue;
459 
460 			ret = timeo;
461 			if (ret == 0)
462 				ret = -ETIMEDOUT;
463 			break;
464 		}
465 
466 		rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
467 			 &inc->i_conn->c_faddr,
468 			 ntohs(inc->i_hdr.h_sport));
469 		save = msg->msg_iter;
470 		ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
471 		if (ret < 0)
472 			break;
473 
474 		/*
475 		 * if the message we just copied isn't at the head of the
476 		 * recv queue then someone else raced us to return it, try
477 		 * to get the next message.
478 		 */
479 		if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
480 			rds_inc_put(inc);
481 			inc = NULL;
482 			rds_stats_inc(s_recv_deliver_raced);
483 			msg->msg_iter = save;
484 			continue;
485 		}
486 
487 		if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
488 			if (msg_flags & MSG_TRUNC)
489 				ret = be32_to_cpu(inc->i_hdr.h_len);
490 			msg->msg_flags |= MSG_TRUNC;
491 		}
492 
493 		if (rds_cmsg_recv(inc, msg, rs)) {
494 			ret = -EFAULT;
495 			goto out;
496 		}
497 
498 		rds_stats_inc(s_recv_delivered);
499 
500 		if (sin) {
501 			sin->sin_family = AF_INET;
502 			sin->sin_port = inc->i_hdr.h_sport;
503 			sin->sin_addr.s_addr = inc->i_saddr;
504 			memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
505 			msg->msg_namelen = sizeof(*sin);
506 		}
507 		break;
508 	}
509 
510 	if (inc)
511 		rds_inc_put(inc);
512 
513 out:
514 	return ret;
515 }
516 
517 /*
518  * The socket is being shut down and we're asked to drop messages that were
519  * queued for recvmsg.  The caller has unbound the socket so the receive path
520  * won't queue any more incoming fragments or messages on the socket.
521  */
522 void rds_clear_recv_queue(struct rds_sock *rs)
523 {
524 	struct sock *sk = rds_rs_to_sk(rs);
525 	struct rds_incoming *inc, *tmp;
526 	unsigned long flags;
527 
528 	write_lock_irqsave(&rs->rs_recv_lock, flags);
529 	list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
530 		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
531 				      -be32_to_cpu(inc->i_hdr.h_len),
532 				      inc->i_hdr.h_dport);
533 		list_del_init(&inc->i_item);
534 		rds_inc_put(inc);
535 	}
536 	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
537 }
538 
539 /*
540  * inc->i_saddr isn't used here because it is only set in the receive
541  * path.
542  */
543 void rds_inc_info_copy(struct rds_incoming *inc,
544 		       struct rds_info_iterator *iter,
545 		       __be32 saddr, __be32 daddr, int flip)
546 {
547 	struct rds_info_message minfo;
548 
549 	minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
550 	minfo.len = be32_to_cpu(inc->i_hdr.h_len);
551 
552 	if (flip) {
553 		minfo.laddr = daddr;
554 		minfo.faddr = saddr;
555 		minfo.lport = inc->i_hdr.h_dport;
556 		minfo.fport = inc->i_hdr.h_sport;
557 	} else {
558 		minfo.laddr = saddr;
559 		minfo.faddr = daddr;
560 		minfo.lport = inc->i_hdr.h_sport;
561 		minfo.fport = inc->i_hdr.h_dport;
562 	}
563 
564 	rds_info_copy(iter, &minfo, sizeof(minfo));
565 }
566