summaryrefslogtreecommitdiff
path: root/net/rds/recv.c
blob: b58f50571782b1921036696d262bba145456dac1 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
/*
 * Copyright (c) 2006 Oracle.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/slab.h>
#include <net/sock.h>
#include <linux/in.h>
#include <linux/export.h>
#include <linux/time.h>
#include <linux/rds.h>

#include "rds.h"

void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
		  __be32 saddr)
{
	atomic_set(&inc->i_refcount, 1);
	INIT_LIST_HEAD(&inc->i_item);
	inc->i_conn = conn;
	inc->i_saddr = saddr;
	inc->i_rdma_cookie = 0;
	inc->i_rx_tstamp.tv_sec = 0;
	inc->i_rx_tstamp.tv_usec = 0;
}
EXPORT_SYMBOL_GPL(rds_inc_init);

void rds_inc_path_init(struct rds_incoming *inc, struct rds_conn_path *cp,
		       __be32 saddr)
{
	atomic_set(&inc->i_refcount, 1);
	INIT_LIST_HEAD(&inc->i_item);
	inc->i_conn = cp->cp_conn;
	inc->i_conn_path = cp;
	inc->i_saddr = saddr;
	inc->i_rdma_cookie = 0;
	inc->i_rx_tstamp.tv_sec = 0;
	inc->i_rx_tstamp.tv_usec = 0;
}
EXPORT_SYMBOL_GPL(rds_inc_path_init);

static void rds_inc_addref(struct rds_incoming *inc)
{
	rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
	atomic_inc(&inc->i_refcount);
}

void rds_inc_put(struct rds_incoming *inc)
{
	rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount));
	if (atomic_dec_and_test(&inc->i_refcount)) {
		BUG_ON(!list_empty(&inc->i_item));

		inc->i_conn->c_trans->inc_free(inc);
	}
}
EXPORT_SYMBOL_GPL(rds_inc_put);

static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk,
				  struct rds_cong_map *map,
				  int delta, __be16 port)
{
	int now_congested;

	if (delta == 0)
		return;

	rs->rs_rcv_bytes += delta;
	now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs);

	rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d "
	  "now_cong %d delta %d\n",
	  rs, &rs->rs_bound_addr,
	  ntohs(rs->rs_bound_port), rs->rs_rcv_bytes,
	  rds_sk_rcvbuf(rs), now_congested, delta);

	/* wasn't -> am congested */
	if (!rs->rs_congested && now_congested) {
		rs->rs_congested = 1;
		rds_cong_set_bit(map, port);
		rds_cong_queue_updates(map);
	}
	/* was -> aren't congested */
	/* Require more free space before reporting uncongested to prevent
	   bouncing cong/uncong state too often */
	else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) {
		rs->rs_congested = 0;
		rds_cong_clear_bit(map, port);
		rds_cong_queue_updates(map);
	}

	/* do nothing if no change in cong state */
}

/*
 * Process all extension headers that come with this message.
 */
static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs)
{
	struct rds_header *hdr = &inc->i_hdr;
	unsigned int pos = 0, type, len;
	union {
		struct rds_ext_header_version version;
		struct rds_ext_header_rdma rdma;
		struct rds_ext_header_rdma_dest rdma_dest;
	} buffer;

	while (1) {
		len = sizeof(buffer);
		type = rds_message_next_extension(hdr, &pos, &buffer, &len);
		if (type == RDS_EXTHDR_NONE)
			break;
		/* Process extension header here */
		switch (type) {
		case RDS_EXTHDR_RDMA:
			rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0);
			break;

		case RDS_EXTHDR_RDMA_DEST:
			/* We ignore the size for now. We could stash it
			 * somewhere and use it for error checking. */
			inc->i_rdma_cookie = rds_rdma_make_cookie(
					be32_to_cpu(buffer.rdma_dest.h_rdma_rkey),
					be32_to_cpu(buffer.rdma_dest.h_rdma_offset));

			break;
		}
	}
}

/*
 * The transport must make sure that this is serialized against other
 * rx and conn reset on this specific conn.
 *
 * We currently assert that only one fragmented message will be sent
 * down a connection at a time.  This lets us reassemble in the conn
 * instead of per-flow which means that we don't have to go digging through
 * flows to tear down partial reassembly progress on conn failure and
 * we save flow lookup and locking for each frag arrival.  It does mean
 * that small messages will wait behind large ones.  Fragmenting at all
 * is only to reduce the memory consumption of pre-posted buffers.
 *
 * The caller passes in saddr and daddr instead of us getting it from the
 * conn.  This lets loopback, who only has one conn for both directions,
 * tell us which roles the addrs in the conn are playing for this message.
 */
void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
		       struct rds_incoming *inc, gfp_t gfp)
{
	struct rds_sock *rs = NULL;
	struct sock *sk;
	unsigned long flags;
	struct rds_conn_path *cp;

	inc->i_conn = conn;
	inc->i_rx_jiffies = jiffies;
	if (conn->c_trans->t_mp_capable)
		cp = inc->i_conn_path;
	else
		cp = &conn->c_path[0];

	rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u "
		 "flags 0x%x rx_jiffies %lu\n", conn,
		 (unsigned long long)cp->cp_next_rx_seq,
		 inc,
		 (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence),
		 be32_to_cpu(inc->i_hdr.h_len),
		 be16_to_cpu(inc->i_hdr.h_sport),
		 be16_to_cpu(inc->i_hdr.h_dport),
		 inc->i_hdr.h_flags,
		 inc->i_rx_jiffies);

	/*
	 * Sequence numbers should only increase.  Messages get their
	 * sequence number as they're queued in a sending conn.  They
	 * can be dropped, though, if the sending socket is closed before
	 * they hit the wire.  So sequence numbers can skip forward
	 * under normal operation.  They can also drop back in the conn
	 * failover case as previously sent messages are resent down the
	 * new instance of a conn.  We drop those, otherwise we have
	 * to assume that the next valid seq does not come after a
	 * hole in the fragment stream.
	 *
	 * The headers don't give us a way to realize if fragments of
	 * a message have been dropped.  We assume that frags that arrive
	 * to a flow are part of the current message on the flow that is
	 * being reassembled.  This means that senders can't drop messages
	 * from the sending conn until all their frags are sent.
	 *
	 * XXX we could spend more on the wire to get more robust failure
	 * detection, arguably worth it to avoid data corruption.
	 */
	if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq &&
	    (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) {
		rds_stats_inc(s_recv_drop_old_seq);
		goto out;
	}
	cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1;

	if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) {
		rds_stats_inc(s_recv_ping);
		rds_send_pong(cp, inc->i_hdr.h_sport);
		goto out;
	}

	rs = rds_find_bound(daddr, inc->i_hdr.h_dport);
	if (!rs) {
		rds_stats_inc(s_recv_drop_no_sock);
		goto out;
	}

	/* Process extension headers */
	rds_recv_incoming_exthdrs(inc, rs);

	/* We can be racing with rds_release() which marks the socket dead. */
	sk = rds_rs_to_sk(rs);

	/* serialize with rds_release -> sock_orphan */
	write_lock_irqsave(&rs->rs_recv_lock, flags);
	if (!sock_flag(sk, SOCK_DEAD)) {
		rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs);
		rds_stats_inc(s_recv_queued);
		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
				      be32_to_cpu(inc->i_hdr.h_len),
				      inc->i_hdr.h_dport);
		if (sock_flag(sk, SOCK_RCVTSTAMP))
			do_gettimeofday(&inc->i_rx_tstamp);
		rds_inc_addref(inc);
		list_add_tail(&inc->i_item, &rs->rs_recv_queue);
		__rds_wake_sk_sleep(sk);
	} else {
		rds_stats_inc(s_recv_drop_dead_sock);
	}
	write_unlock_irqrestore(&rs->rs_recv_lock, flags);

out:
	if (rs)
		rds_sock_put(rs);
}
EXPORT_SYMBOL_GPL(rds_recv_incoming);

/*
 * be very careful here.  This is being called as the condition in
 * wait_event_*() needs to cope with being called many times.
 */
static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc)
{
	unsigned long flags;

	if (!*inc) {
		read_lock_irqsave(&rs->rs_recv_lock, flags);
		if (!list_empty(&rs->rs_recv_queue)) {
			*inc = list_entry(rs->rs_recv_queue.next,
					  struct rds_incoming,
					  i_item);
			rds_inc_addref(*inc);
		}
		read_unlock_irqrestore(&rs->rs_recv_lock, flags);
	}

	return *inc != NULL;
}

static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc,
			    int drop)
{
	struct sock *sk = rds_rs_to_sk(rs);
	int ret = 0;
	unsigned long flags;

	write_lock_irqsave(&rs->rs_recv_lock, flags);
	if (!list_empty(&inc->i_item)) {
		ret = 1;
		if (drop) {
			/* XXX make sure this i_conn is reliable */
			rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
					      -be32_to_cpu(inc->i_hdr.h_len),
					      inc->i_hdr.h_dport);
			list_del_init(&inc->i_item);
			rds_inc_put(inc);
		}
	}
	write_unlock_irqrestore(&rs->rs_recv_lock, flags);

	rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop);
	return ret;
}

/*
 * Pull errors off the error queue.
 * If msghdr is NULL, we will just purge the error queue.
 */
int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr)
{
	struct rds_notifier *notifier;
	struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */
	unsigned int count = 0, max_messages = ~0U;
	unsigned long flags;
	LIST_HEAD(copy);
	int err = 0;


	/* put_cmsg copies to user space and thus may sleep. We can't do this
	 * with rs_lock held, so first grab as many notifications as we can stuff
	 * in the user provided cmsg buffer. We don't try to copy more, to avoid
	 * losing notifications - except when the buffer is so small that it wouldn't
	 * even hold a single notification. Then we give him as much of this single
	 * msg as we can squeeze in, and set MSG_CTRUNC.
	 */
	if (msghdr) {
		max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg));
		if (!max_messages)
			max_messages = 1;
	}

	spin_lock_irqsave(&rs->rs_lock, flags);
	while (!list_empty(&rs->rs_notify_queue) && count < max_messages) {
		notifier = list_entry(rs->rs_notify_queue.next,
				struct rds_notifier, n_list);
		list_move(&notifier->n_list, &copy);
		count++;
	}
	spin_unlock_irqrestore(&rs->rs_lock, flags);

	if (!count)
		return 0;

	while (!list_empty(&copy)) {
		notifier = list_entry(copy.next, struct rds_notifier, n_list);

		if (msghdr) {
			cmsg.user_token = notifier->n_user_token;
			cmsg.status = notifier->n_status;

			err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS,
				       sizeof(cmsg), &cmsg);
			if (err)
				break;
		}

		list_del_init(&notifier->n_list);
		kfree(notifier);
	}

	/* If we bailed out because of an error in put_cmsg,
	 * we may be left with one or more notifications that we
	 * didn't process. Return them to the head of the list. */
	if (!list_empty(&copy)) {
		spin_lock_irqsave(&rs->rs_lock, flags);
		list_splice(&copy, &rs->rs_notify_queue);
		spin_unlock_irqrestore(&rs->rs_lock, flags);
	}

	return err;
}

/*
 * Queue a congestion notification
 */
static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr)
{
	uint64_t notify = rs->rs_cong_notify;
	unsigned long flags;
	int err;

	err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE,
			sizeof(notify), &notify);
	if (err)
		return err;

	spin_lock_irqsave(&rs->rs_lock, flags);
	rs->rs_cong_notify &= ~notify;
	spin_unlock_irqrestore(&rs->rs_lock, flags);

	return 0;
}

/*
 * Receive any control messages.
 */
static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg,
			 struct rds_sock *rs)
{
	int ret = 0;

	if (inc->i_rdma_cookie) {
		ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST,
				sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie);
		if (ret)
			return ret;
	}

	if ((inc->i_rx_tstamp.tv_sec != 0) &&
	    sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) {
		ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
			       sizeof(struct timeval),
			       &inc->i_rx_tstamp);
		if (ret)
			return ret;
	}

	return 0;
}

int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
		int msg_flags)
{
	struct sock *sk = sock->sk;
	struct rds_sock *rs = rds_sk_to_rs(sk);
	long timeo;
	int ret = 0, nonblock = msg_flags & MSG_DONTWAIT;
	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
	struct rds_incoming *inc = NULL;

	/* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */
	timeo = sock_rcvtimeo(sk, nonblock);

	rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo);

	if (msg_flags & MSG_OOB)
		goto out;

	while (1) {
		struct iov_iter save;
		/* If there are pending notifications, do those - and nothing else */
		if (!list_empty(&rs->rs_notify_queue)) {
			ret = rds_notify_queue_get(rs, msg);
			break;
		}

		if (rs->rs_cong_notify) {
			ret = rds_notify_cong(rs, msg);
			break;
		}

		if (!rds_next_incoming(rs, &inc)) {
			if (nonblock) {
				ret = -EAGAIN;
				break;
			}

			timeo = wait_event_interruptible_timeout(*sk_sleep(sk),
					(!list_empty(&rs->rs_notify_queue) ||
					 rs->rs_cong_notify ||
					 rds_next_incoming(rs, &inc)), timeo);
			rdsdebug("recvmsg woke inc %p timeo %ld\n", inc,
				 timeo);
			if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT)
				continue;

			ret = timeo;
			if (ret == 0)
				ret = -ETIMEDOUT;
			break;
		}

		rdsdebug("copying inc %p from %pI4:%u to user\n", inc,
			 &inc->i_conn->c_faddr,
			 ntohs(inc->i_hdr.h_sport));
		save = msg->msg_iter;
		ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter);
		if (ret < 0)
			break;

		/*
		 * if the message we just copied isn't at the head of the
		 * recv queue then someone else raced us to return it, try
		 * to get the next message.
		 */
		if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) {
			rds_inc_put(inc);
			inc = NULL;
			rds_stats_inc(s_recv_deliver_raced);
			msg->msg_iter = save;
			continue;
		}

		if (ret < be32_to_cpu(inc->i_hdr.h_len)) {
			if (msg_flags & MSG_TRUNC)
				ret = be32_to_cpu(inc->i_hdr.h_len);
			msg->msg_flags |= MSG_TRUNC;
		}

		if (rds_cmsg_recv(inc, msg, rs)) {
			ret = -EFAULT;
			goto out;
		}

		rds_stats_inc(s_recv_delivered);

		if (sin) {
			sin->sin_family = AF_INET;
			sin->sin_port = inc->i_hdr.h_sport;
			sin->sin_addr.s_addr = inc->i_saddr;
			memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
			msg->msg_namelen = sizeof(*sin);
		}
		break;
	}

	if (inc)
		rds_inc_put(inc);

out:
	return ret;
}

/*
 * The socket is being shut down and we're asked to drop messages that were
 * queued for recvmsg.  The caller has unbound the socket so the receive path
 * won't queue any more incoming fragments or messages on the socket.
 */
void rds_clear_recv_queue(struct rds_sock *rs)
{
	struct sock *sk = rds_rs_to_sk(rs);
	struct rds_incoming *inc, *tmp;
	unsigned long flags;

	write_lock_irqsave(&rs->rs_recv_lock, flags);
	list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) {
		rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong,
				      -be32_to_cpu(inc->i_hdr.h_len),
				      inc->i_hdr.h_dport);
		list_del_init(&inc->i_item);
		rds_inc_put(inc);
	}
	write_unlock_irqrestore(&rs->rs_recv_lock, flags);
}

/*
 * inc->i_saddr isn't used here because it is only set in the receive
 * path.
 */
void rds_inc_info_copy(struct rds_incoming *inc,
		       struct rds_info_iterator *iter,
		       __be32 saddr, __be32 daddr, int flip)
{
	struct rds_info_message minfo;

	minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence);
	minfo.len = be32_to_cpu(inc->i_hdr.h_len);

	if (flip) {
		minfo.laddr = daddr;
		minfo.faddr = saddr;
		minfo.lport = inc->i_hdr.h_dport;
		minfo.fport = inc->i_hdr.h_sport;
	} else {
		minfo.laddr = saddr;
		minfo.faddr = daddr;
		minfo.lport = inc->i_hdr.h_sport;
		minfo.fport = inc->i_hdr.h_dport;
	}

	minfo.flags = 0;

	rds_info_copy(iter, &minfo, sizeof(minfo));
}