xref: /openbmc/linux/net/rds/af_rds.c (revision 5b4cb650)
1 /*
2  * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/module.h>
34 #include <linux/errno.h>
35 #include <linux/kernel.h>
36 #include <linux/gfp.h>
37 #include <linux/in.h>
38 #include <linux/ipv6.h>
39 #include <linux/poll.h>
40 #include <net/sock.h>
41 
42 #include "rds.h"
43 
44 /* this is just used for stats gathering :/ */
45 static DEFINE_SPINLOCK(rds_sock_lock);
46 static unsigned long rds_sock_count;
47 static LIST_HEAD(rds_sock_list);
48 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
49 
50 /*
51  * This is called as the final descriptor referencing this socket is closed.
52  * We have to unbind the socket so that another socket can be bound to the
53  * address it was using.
54  *
55  * We have to be careful about racing with the incoming path.  sock_orphan()
56  * sets SOCK_DEAD and we use that as an indicator to the rx path that new
57  * messages shouldn't be queued.
58  */
59 static int rds_release(struct socket *sock)
60 {
61 	struct sock *sk = sock->sk;
62 	struct rds_sock *rs;
63 
64 	if (!sk)
65 		goto out;
66 
67 	rs = rds_sk_to_rs(sk);
68 
69 	sock_orphan(sk);
70 	/* Note - rds_clear_recv_queue grabs rs_recv_lock, so
71 	 * that ensures the recv path has completed messing
72 	 * with the socket. */
73 	rds_clear_recv_queue(rs);
74 	rds_cong_remove_socket(rs);
75 
76 	rds_remove_bound(rs);
77 
78 	rds_send_drop_to(rs, NULL);
79 	rds_rdma_drop_keys(rs);
80 	rds_notify_queue_get(rs, NULL);
81 	rds_notify_msg_zcopy_purge(&rs->rs_zcookie_queue);
82 
83 	spin_lock_bh(&rds_sock_lock);
84 	list_del_init(&rs->rs_item);
85 	rds_sock_count--;
86 	spin_unlock_bh(&rds_sock_lock);
87 
88 	rds_trans_put(rs->rs_transport);
89 
90 	sock->sk = NULL;
91 	sock_put(sk);
92 out:
93 	return 0;
94 }
95 
96 /*
97  * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
98  * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
99  * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
100  * this seems more conservative.
101  * NB - normally, one would use sk_callback_lock for this, but we can
102  * get here from interrupts, whereas the network code grabs sk_callback_lock
103  * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
104  */
105 void rds_wake_sk_sleep(struct rds_sock *rs)
106 {
107 	unsigned long flags;
108 
109 	read_lock_irqsave(&rs->rs_recv_lock, flags);
110 	__rds_wake_sk_sleep(rds_rs_to_sk(rs));
111 	read_unlock_irqrestore(&rs->rs_recv_lock, flags);
112 }
113 
114 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
115 		       int peer)
116 {
117 	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
118 	struct sockaddr_in6 *sin6;
119 	struct sockaddr_in *sin;
120 	int uaddr_len;
121 
122 	/* racey, don't care */
123 	if (peer) {
124 		if (ipv6_addr_any(&rs->rs_conn_addr))
125 			return -ENOTCONN;
126 
127 		if (ipv6_addr_v4mapped(&rs->rs_conn_addr)) {
128 			sin = (struct sockaddr_in *)uaddr;
129 			memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
130 			sin->sin_family = AF_INET;
131 			sin->sin_port = rs->rs_conn_port;
132 			sin->sin_addr.s_addr = rs->rs_conn_addr_v4;
133 			uaddr_len = sizeof(*sin);
134 		} else {
135 			sin6 = (struct sockaddr_in6 *)uaddr;
136 			sin6->sin6_family = AF_INET6;
137 			sin6->sin6_port = rs->rs_conn_port;
138 			sin6->sin6_addr = rs->rs_conn_addr;
139 			sin6->sin6_flowinfo = 0;
140 			/* scope_id is the same as in the bound address. */
141 			sin6->sin6_scope_id = rs->rs_bound_scope_id;
142 			uaddr_len = sizeof(*sin6);
143 		}
144 	} else {
145 		/* If socket is not yet bound and the socket is connected,
146 		 * set the return address family to be the same as the
147 		 * connected address, but with 0 address value.  If it is not
148 		 * connected, set the family to be AF_UNSPEC (value 0) and
149 		 * the address size to be that of an IPv4 address.
150 		 */
151 		if (ipv6_addr_any(&rs->rs_bound_addr)) {
152 			if (ipv6_addr_any(&rs->rs_conn_addr)) {
153 				sin = (struct sockaddr_in *)uaddr;
154 				memset(sin, 0, sizeof(*sin));
155 				sin->sin_family = AF_UNSPEC;
156 				return sizeof(*sin);
157 			}
158 
159 #if IS_ENABLED(CONFIG_IPV6)
160 			if (!(ipv6_addr_type(&rs->rs_conn_addr) &
161 			      IPV6_ADDR_MAPPED)) {
162 				sin6 = (struct sockaddr_in6 *)uaddr;
163 				memset(sin6, 0, sizeof(*sin6));
164 				sin6->sin6_family = AF_INET6;
165 				return sizeof(*sin6);
166 			}
167 #endif
168 
169 			sin = (struct sockaddr_in *)uaddr;
170 			memset(sin, 0, sizeof(*sin));
171 			sin->sin_family = AF_INET;
172 			return sizeof(*sin);
173 		}
174 		if (ipv6_addr_v4mapped(&rs->rs_bound_addr)) {
175 			sin = (struct sockaddr_in *)uaddr;
176 			memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
177 			sin->sin_family = AF_INET;
178 			sin->sin_port = rs->rs_bound_port;
179 			sin->sin_addr.s_addr = rs->rs_bound_addr_v4;
180 			uaddr_len = sizeof(*sin);
181 		} else {
182 			sin6 = (struct sockaddr_in6 *)uaddr;
183 			sin6->sin6_family = AF_INET6;
184 			sin6->sin6_port = rs->rs_bound_port;
185 			sin6->sin6_addr = rs->rs_bound_addr;
186 			sin6->sin6_flowinfo = 0;
187 			sin6->sin6_scope_id = rs->rs_bound_scope_id;
188 			uaddr_len = sizeof(*sin6);
189 		}
190 	}
191 
192 	return uaddr_len;
193 }
194 
195 /*
196  * RDS' poll is without a doubt the least intuitive part of the interface,
197  * as EPOLLIN and EPOLLOUT do not behave entirely as you would expect from
198  * a network protocol.
199  *
200  * EPOLLIN is asserted if
201  *  -	there is data on the receive queue.
202  *  -	to signal that a previously congested destination may have become
203  *	uncongested
204  *  -	A notification has been queued to the socket (this can be a congestion
205  *	update, or a RDMA completion, or a MSG_ZEROCOPY completion).
206  *
207  * EPOLLOUT is asserted if there is room on the send queue. This does not mean
208  * however, that the next sendmsg() call will succeed. If the application tries
209  * to send to a congested destination, the system call may still fail (and
210  * return ENOBUFS).
211  */
212 static __poll_t rds_poll(struct file *file, struct socket *sock,
213 			     poll_table *wait)
214 {
215 	struct sock *sk = sock->sk;
216 	struct rds_sock *rs = rds_sk_to_rs(sk);
217 	__poll_t mask = 0;
218 	unsigned long flags;
219 
220 	poll_wait(file, sk_sleep(sk), wait);
221 
222 	if (rs->rs_seen_congestion)
223 		poll_wait(file, &rds_poll_waitq, wait);
224 
225 	read_lock_irqsave(&rs->rs_recv_lock, flags);
226 	if (!rs->rs_cong_monitor) {
227 		/* When a congestion map was updated, we signal EPOLLIN for
228 		 * "historical" reasons. Applications can also poll for
229 		 * WRBAND instead. */
230 		if (rds_cong_updated_since(&rs->rs_cong_track))
231 			mask |= (EPOLLIN | EPOLLRDNORM | EPOLLWRBAND);
232 	} else {
233 		spin_lock(&rs->rs_lock);
234 		if (rs->rs_cong_notify)
235 			mask |= (EPOLLIN | EPOLLRDNORM);
236 		spin_unlock(&rs->rs_lock);
237 	}
238 	if (!list_empty(&rs->rs_recv_queue) ||
239 	    !list_empty(&rs->rs_notify_queue) ||
240 	    !list_empty(&rs->rs_zcookie_queue.zcookie_head))
241 		mask |= (EPOLLIN | EPOLLRDNORM);
242 	if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
243 		mask |= (EPOLLOUT | EPOLLWRNORM);
244 	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
245 		mask |= POLLERR;
246 	read_unlock_irqrestore(&rs->rs_recv_lock, flags);
247 
248 	/* clear state any time we wake a seen-congested socket */
249 	if (mask)
250 		rs->rs_seen_congestion = 0;
251 
252 	return mask;
253 }
254 
255 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
256 {
257 	return -ENOIOCTLCMD;
258 }
259 
260 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
261 			      int len)
262 {
263 	struct sockaddr_in6 sin6;
264 	struct sockaddr_in sin;
265 	int ret = 0;
266 
267 	/* racing with another thread binding seems ok here */
268 	if (ipv6_addr_any(&rs->rs_bound_addr)) {
269 		ret = -ENOTCONN; /* XXX not a great errno */
270 		goto out;
271 	}
272 
273 	if (len < sizeof(struct sockaddr_in)) {
274 		ret = -EINVAL;
275 		goto out;
276 	} else if (len < sizeof(struct sockaddr_in6)) {
277 		/* Assume IPv4 */
278 		if (copy_from_user(&sin, optval, sizeof(struct sockaddr_in))) {
279 			ret = -EFAULT;
280 			goto out;
281 		}
282 		ipv6_addr_set_v4mapped(sin.sin_addr.s_addr, &sin6.sin6_addr);
283 		sin6.sin6_port = sin.sin_port;
284 	} else {
285 		if (copy_from_user(&sin6, optval,
286 				   sizeof(struct sockaddr_in6))) {
287 			ret = -EFAULT;
288 			goto out;
289 		}
290 	}
291 
292 	rds_send_drop_to(rs, &sin6);
293 out:
294 	return ret;
295 }
296 
297 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
298 			       int optlen)
299 {
300 	int value;
301 
302 	if (optlen < sizeof(int))
303 		return -EINVAL;
304 	if (get_user(value, (int __user *) optval))
305 		return -EFAULT;
306 	*optvar = !!value;
307 	return 0;
308 }
309 
310 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
311 			    int optlen)
312 {
313 	int ret;
314 
315 	ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
316 	if (ret == 0) {
317 		if (rs->rs_cong_monitor) {
318 			rds_cong_add_socket(rs);
319 		} else {
320 			rds_cong_remove_socket(rs);
321 			rs->rs_cong_mask = 0;
322 			rs->rs_cong_notify = 0;
323 		}
324 	}
325 	return ret;
326 }
327 
328 static int rds_set_transport(struct rds_sock *rs, char __user *optval,
329 			     int optlen)
330 {
331 	int t_type;
332 
333 	if (rs->rs_transport)
334 		return -EOPNOTSUPP; /* previously attached to transport */
335 
336 	if (optlen != sizeof(int))
337 		return -EINVAL;
338 
339 	if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
340 		return -EFAULT;
341 
342 	if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
343 		return -EINVAL;
344 
345 	rs->rs_transport = rds_trans_get(t_type);
346 
347 	return rs->rs_transport ? 0 : -ENOPROTOOPT;
348 }
349 
350 static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
351 				 int optlen)
352 {
353 	int val, valbool;
354 
355 	if (optlen != sizeof(int))
356 		return -EFAULT;
357 
358 	if (get_user(val, (int __user *)optval))
359 		return -EFAULT;
360 
361 	valbool = val ? 1 : 0;
362 
363 	if (valbool)
364 		sock_set_flag(sk, SOCK_RCVTSTAMP);
365 	else
366 		sock_reset_flag(sk, SOCK_RCVTSTAMP);
367 
368 	return 0;
369 }
370 
371 static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval,
372 				  int optlen)
373 {
374 	struct rds_rx_trace_so trace;
375 	int i;
376 
377 	if (optlen != sizeof(struct rds_rx_trace_so))
378 		return -EFAULT;
379 
380 	if (copy_from_user(&trace, optval, sizeof(trace)))
381 		return -EFAULT;
382 
383 	if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
384 		return -EFAULT;
385 
386 	rs->rs_rx_traces = trace.rx_traces;
387 	for (i = 0; i < rs->rs_rx_traces; i++) {
388 		if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
389 			rs->rs_rx_traces = 0;
390 			return -EFAULT;
391 		}
392 		rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
393 	}
394 
395 	return 0;
396 }
397 
398 static int rds_setsockopt(struct socket *sock, int level, int optname,
399 			  char __user *optval, unsigned int optlen)
400 {
401 	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
402 	int ret;
403 
404 	if (level != SOL_RDS) {
405 		ret = -ENOPROTOOPT;
406 		goto out;
407 	}
408 
409 	switch (optname) {
410 	case RDS_CANCEL_SENT_TO:
411 		ret = rds_cancel_sent_to(rs, optval, optlen);
412 		break;
413 	case RDS_GET_MR:
414 		ret = rds_get_mr(rs, optval, optlen);
415 		break;
416 	case RDS_GET_MR_FOR_DEST:
417 		ret = rds_get_mr_for_dest(rs, optval, optlen);
418 		break;
419 	case RDS_FREE_MR:
420 		ret = rds_free_mr(rs, optval, optlen);
421 		break;
422 	case RDS_RECVERR:
423 		ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
424 		break;
425 	case RDS_CONG_MONITOR:
426 		ret = rds_cong_monitor(rs, optval, optlen);
427 		break;
428 	case SO_RDS_TRANSPORT:
429 		lock_sock(sock->sk);
430 		ret = rds_set_transport(rs, optval, optlen);
431 		release_sock(sock->sk);
432 		break;
433 	case SO_TIMESTAMP:
434 		lock_sock(sock->sk);
435 		ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
436 		release_sock(sock->sk);
437 		break;
438 	case SO_RDS_MSG_RXPATH_LATENCY:
439 		ret = rds_recv_track_latency(rs, optval, optlen);
440 		break;
441 	default:
442 		ret = -ENOPROTOOPT;
443 	}
444 out:
445 	return ret;
446 }
447 
448 static int rds_getsockopt(struct socket *sock, int level, int optname,
449 			  char __user *optval, int __user *optlen)
450 {
451 	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
452 	int ret = -ENOPROTOOPT, len;
453 	int trans;
454 
455 	if (level != SOL_RDS)
456 		goto out;
457 
458 	if (get_user(len, optlen)) {
459 		ret = -EFAULT;
460 		goto out;
461 	}
462 
463 	switch (optname) {
464 	case RDS_INFO_FIRST ... RDS_INFO_LAST:
465 		ret = rds_info_getsockopt(sock, optname, optval,
466 					  optlen);
467 		break;
468 
469 	case RDS_RECVERR:
470 		if (len < sizeof(int))
471 			ret = -EINVAL;
472 		else
473 		if (put_user(rs->rs_recverr, (int __user *) optval) ||
474 		    put_user(sizeof(int), optlen))
475 			ret = -EFAULT;
476 		else
477 			ret = 0;
478 		break;
479 	case SO_RDS_TRANSPORT:
480 		if (len < sizeof(int)) {
481 			ret = -EINVAL;
482 			break;
483 		}
484 		trans = (rs->rs_transport ? rs->rs_transport->t_type :
485 			 RDS_TRANS_NONE); /* unbound */
486 		if (put_user(trans, (int __user *)optval) ||
487 		    put_user(sizeof(int), optlen))
488 			ret = -EFAULT;
489 		else
490 			ret = 0;
491 		break;
492 	default:
493 		break;
494 	}
495 
496 out:
497 	return ret;
498 
499 }
500 
501 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
502 		       int addr_len, int flags)
503 {
504 	struct sock *sk = sock->sk;
505 	struct sockaddr_in *sin;
506 	struct rds_sock *rs = rds_sk_to_rs(sk);
507 	int ret = 0;
508 
509 	lock_sock(sk);
510 
511 	switch (uaddr->sa_family) {
512 	case AF_INET:
513 		sin = (struct sockaddr_in *)uaddr;
514 		if (addr_len < sizeof(struct sockaddr_in)) {
515 			ret = -EINVAL;
516 			break;
517 		}
518 		if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
519 			ret = -EDESTADDRREQ;
520 			break;
521 		}
522 		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) ||
523 		    sin->sin_addr.s_addr == htonl(INADDR_BROADCAST)) {
524 			ret = -EINVAL;
525 			break;
526 		}
527 		ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &rs->rs_conn_addr);
528 		rs->rs_conn_port = sin->sin_port;
529 		break;
530 
531 #if IS_ENABLED(CONFIG_IPV6)
532 	case AF_INET6: {
533 		struct sockaddr_in6 *sin6;
534 		int addr_type;
535 
536 		sin6 = (struct sockaddr_in6 *)uaddr;
537 		if (addr_len < sizeof(struct sockaddr_in6)) {
538 			ret = -EINVAL;
539 			break;
540 		}
541 		addr_type = ipv6_addr_type(&sin6->sin6_addr);
542 		if (!(addr_type & IPV6_ADDR_UNICAST)) {
543 			__be32 addr4;
544 
545 			if (!(addr_type & IPV6_ADDR_MAPPED)) {
546 				ret = -EPROTOTYPE;
547 				break;
548 			}
549 
550 			/* It is a mapped address.  Need to do some sanity
551 			 * checks.
552 			 */
553 			addr4 = sin6->sin6_addr.s6_addr32[3];
554 			if (addr4 == htonl(INADDR_ANY) ||
555 			    addr4 == htonl(INADDR_BROADCAST) ||
556 			    IN_MULTICAST(ntohl(addr4))) {
557 				ret = -EPROTOTYPE;
558 				break;
559 			}
560 		}
561 
562 		if (addr_type & IPV6_ADDR_LINKLOCAL) {
563 			/* If socket is arleady bound to a link local address,
564 			 * the peer address must be on the same link.
565 			 */
566 			if (sin6->sin6_scope_id == 0 ||
567 			    (!ipv6_addr_any(&rs->rs_bound_addr) &&
568 			     rs->rs_bound_scope_id &&
569 			     sin6->sin6_scope_id != rs->rs_bound_scope_id)) {
570 				ret = -EINVAL;
571 				break;
572 			}
573 			/* Remember the connected address scope ID.  It will
574 			 * be checked against the binding local address when
575 			 * the socket is bound.
576 			 */
577 			rs->rs_bound_scope_id = sin6->sin6_scope_id;
578 		}
579 		rs->rs_conn_addr = sin6->sin6_addr;
580 		rs->rs_conn_port = sin6->sin6_port;
581 		break;
582 	}
583 #endif
584 
585 	default:
586 		ret = -EAFNOSUPPORT;
587 		break;
588 	}
589 
590 	release_sock(sk);
591 	return ret;
592 }
593 
594 static struct proto rds_proto = {
595 	.name	  = "RDS",
596 	.owner	  = THIS_MODULE,
597 	.obj_size = sizeof(struct rds_sock),
598 };
599 
600 static const struct proto_ops rds_proto_ops = {
601 	.family =	AF_RDS,
602 	.owner =	THIS_MODULE,
603 	.release =	rds_release,
604 	.bind =		rds_bind,
605 	.connect =	rds_connect,
606 	.socketpair =	sock_no_socketpair,
607 	.accept =	sock_no_accept,
608 	.getname =	rds_getname,
609 	.poll =		rds_poll,
610 	.ioctl =	rds_ioctl,
611 	.listen =	sock_no_listen,
612 	.shutdown =	sock_no_shutdown,
613 	.setsockopt =	rds_setsockopt,
614 	.getsockopt =	rds_getsockopt,
615 	.sendmsg =	rds_sendmsg,
616 	.recvmsg =	rds_recvmsg,
617 	.mmap =		sock_no_mmap,
618 	.sendpage =	sock_no_sendpage,
619 };
620 
621 static void rds_sock_destruct(struct sock *sk)
622 {
623 	struct rds_sock *rs = rds_sk_to_rs(sk);
624 
625 	WARN_ON((&rs->rs_item != rs->rs_item.next ||
626 		 &rs->rs_item != rs->rs_item.prev));
627 }
628 
629 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
630 {
631 	struct rds_sock *rs;
632 
633 	sock_init_data(sock, sk);
634 	sock->ops		= &rds_proto_ops;
635 	sk->sk_protocol		= protocol;
636 	sk->sk_destruct		= rds_sock_destruct;
637 
638 	rs = rds_sk_to_rs(sk);
639 	spin_lock_init(&rs->rs_lock);
640 	rwlock_init(&rs->rs_recv_lock);
641 	INIT_LIST_HEAD(&rs->rs_send_queue);
642 	INIT_LIST_HEAD(&rs->rs_recv_queue);
643 	INIT_LIST_HEAD(&rs->rs_notify_queue);
644 	INIT_LIST_HEAD(&rs->rs_cong_list);
645 	rds_message_zcopy_queue_init(&rs->rs_zcookie_queue);
646 	spin_lock_init(&rs->rs_rdma_lock);
647 	rs->rs_rdma_keys = RB_ROOT;
648 	rs->rs_rx_traces = 0;
649 
650 	spin_lock_bh(&rds_sock_lock);
651 	list_add_tail(&rs->rs_item, &rds_sock_list);
652 	rds_sock_count++;
653 	spin_unlock_bh(&rds_sock_lock);
654 
655 	return 0;
656 }
657 
658 static int rds_create(struct net *net, struct socket *sock, int protocol,
659 		      int kern)
660 {
661 	struct sock *sk;
662 
663 	if (sock->type != SOCK_SEQPACKET || protocol)
664 		return -ESOCKTNOSUPPORT;
665 
666 	sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
667 	if (!sk)
668 		return -ENOMEM;
669 
670 	return __rds_create(sock, sk, protocol);
671 }
672 
673 void rds_sock_addref(struct rds_sock *rs)
674 {
675 	sock_hold(rds_rs_to_sk(rs));
676 }
677 
678 void rds_sock_put(struct rds_sock *rs)
679 {
680 	sock_put(rds_rs_to_sk(rs));
681 }
682 
683 static const struct net_proto_family rds_family_ops = {
684 	.family =	AF_RDS,
685 	.create =	rds_create,
686 	.owner	=	THIS_MODULE,
687 };
688 
689 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
690 			      struct rds_info_iterator *iter,
691 			      struct rds_info_lengths *lens)
692 {
693 	struct rds_sock *rs;
694 	struct rds_incoming *inc;
695 	unsigned int total = 0;
696 
697 	len /= sizeof(struct rds_info_message);
698 
699 	spin_lock_bh(&rds_sock_lock);
700 
701 	list_for_each_entry(rs, &rds_sock_list, rs_item) {
702 		read_lock(&rs->rs_recv_lock);
703 
704 		/* XXX too lazy to maintain counts.. */
705 		list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
706 			total++;
707 			if (total <= len)
708 				rds_inc_info_copy(inc, iter,
709 						  inc->i_saddr.s6_addr32[3],
710 						  rs->rs_bound_addr_v4,
711 						  1);
712 		}
713 
714 		read_unlock(&rs->rs_recv_lock);
715 	}
716 
717 	spin_unlock_bh(&rds_sock_lock);
718 
719 	lens->nr = total;
720 	lens->each = sizeof(struct rds_info_message);
721 }
722 
723 static void rds_sock_info(struct socket *sock, unsigned int len,
724 			  struct rds_info_iterator *iter,
725 			  struct rds_info_lengths *lens)
726 {
727 	struct rds_info_socket sinfo;
728 	struct rds_sock *rs;
729 
730 	len /= sizeof(struct rds_info_socket);
731 
732 	spin_lock_bh(&rds_sock_lock);
733 
734 	if (len < rds_sock_count)
735 		goto out;
736 
737 	list_for_each_entry(rs, &rds_sock_list, rs_item) {
738 		sinfo.sndbuf = rds_sk_sndbuf(rs);
739 		sinfo.rcvbuf = rds_sk_rcvbuf(rs);
740 		sinfo.bound_addr = rs->rs_bound_addr_v4;
741 		sinfo.connected_addr = rs->rs_conn_addr_v4;
742 		sinfo.bound_port = rs->rs_bound_port;
743 		sinfo.connected_port = rs->rs_conn_port;
744 		sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
745 
746 		rds_info_copy(iter, &sinfo, sizeof(sinfo));
747 	}
748 
749 out:
750 	lens->nr = rds_sock_count;
751 	lens->each = sizeof(struct rds_info_socket);
752 
753 	spin_unlock_bh(&rds_sock_lock);
754 }
755 
756 static void rds_exit(void)
757 {
758 	sock_unregister(rds_family_ops.family);
759 	proto_unregister(&rds_proto);
760 	rds_conn_exit();
761 	rds_cong_exit();
762 	rds_sysctl_exit();
763 	rds_threads_exit();
764 	rds_stats_exit();
765 	rds_page_exit();
766 	rds_bind_lock_destroy();
767 	rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
768 	rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
769 }
770 module_exit(rds_exit);
771 
772 u32 rds_gen_num;
773 
774 static int rds_init(void)
775 {
776 	int ret;
777 
778 	net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
779 
780 	ret = rds_bind_lock_init();
781 	if (ret)
782 		goto out;
783 
784 	ret = rds_conn_init();
785 	if (ret)
786 		goto out_bind;
787 
788 	ret = rds_threads_init();
789 	if (ret)
790 		goto out_conn;
791 	ret = rds_sysctl_init();
792 	if (ret)
793 		goto out_threads;
794 	ret = rds_stats_init();
795 	if (ret)
796 		goto out_sysctl;
797 	ret = proto_register(&rds_proto, 1);
798 	if (ret)
799 		goto out_stats;
800 	ret = sock_register(&rds_family_ops);
801 	if (ret)
802 		goto out_proto;
803 
804 	rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
805 	rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
806 
807 	goto out;
808 
809 out_proto:
810 	proto_unregister(&rds_proto);
811 out_stats:
812 	rds_stats_exit();
813 out_sysctl:
814 	rds_sysctl_exit();
815 out_threads:
816 	rds_threads_exit();
817 out_conn:
818 	rds_conn_exit();
819 	rds_cong_exit();
820 	rds_page_exit();
821 out_bind:
822 	rds_bind_lock_destroy();
823 out:
824 	return ret;
825 }
826 module_init(rds_init);
827 
828 #define DRV_VERSION     "4.0"
829 #define DRV_RELDATE     "Feb 12, 2009"
830 
831 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
832 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
833 		   " v" DRV_VERSION " (" DRV_RELDATE ")");
834 MODULE_VERSION(DRV_VERSION);
835 MODULE_LICENSE("Dual BSD/GPL");
836 MODULE_ALIAS_NETPROTO(PF_RDS);
837