Loading...
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/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/poll.h>
39#include <net/sock.h>
40
41#include "rds.h"
42
43/* this is just used for stats gathering :/ */
44static DEFINE_SPINLOCK(rds_sock_lock);
45static unsigned long rds_sock_count;
46static LIST_HEAD(rds_sock_list);
47DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
48
49/*
50 * This is called as the final descriptor referencing this socket is closed.
51 * We have to unbind the socket so that another socket can be bound to the
52 * address it was using.
53 *
54 * We have to be careful about racing with the incoming path. sock_orphan()
55 * sets SOCK_DEAD and we use that as an indicator to the rx path that new
56 * messages shouldn't be queued.
57 */
58static int rds_release(struct socket *sock)
59{
60 struct sock *sk = sock->sk;
61 struct rds_sock *rs;
62
63 if (!sk)
64 goto out;
65
66 rs = rds_sk_to_rs(sk);
67
68 sock_orphan(sk);
69 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
70 * that ensures the recv path has completed messing
71 * with the socket. */
72 rds_clear_recv_queue(rs);
73 rds_cong_remove_socket(rs);
74
75 rds_remove_bound(rs);
76
77 rds_send_drop_to(rs, NULL);
78 rds_rdma_drop_keys(rs);
79 rds_notify_queue_get(rs, NULL);
80
81 spin_lock_bh(&rds_sock_lock);
82 list_del_init(&rs->rs_item);
83 rds_sock_count--;
84 spin_unlock_bh(&rds_sock_lock);
85
86 rds_trans_put(rs->rs_transport);
87
88 sock->sk = NULL;
89 sock_put(sk);
90out:
91 return 0;
92}
93
94/*
95 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
96 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
97 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
98 * this seems more conservative.
99 * NB - normally, one would use sk_callback_lock for this, but we can
100 * get here from interrupts, whereas the network code grabs sk_callback_lock
101 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
102 */
103void rds_wake_sk_sleep(struct rds_sock *rs)
104{
105 unsigned long flags;
106
107 read_lock_irqsave(&rs->rs_recv_lock, flags);
108 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
109 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
110}
111
112static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
113 int *uaddr_len, int peer)
114{
115 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
116 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
117
118 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
119
120 /* racey, don't care */
121 if (peer) {
122 if (!rs->rs_conn_addr)
123 return -ENOTCONN;
124
125 sin->sin_port = rs->rs_conn_port;
126 sin->sin_addr.s_addr = rs->rs_conn_addr;
127 } else {
128 sin->sin_port = rs->rs_bound_port;
129 sin->sin_addr.s_addr = rs->rs_bound_addr;
130 }
131
132 sin->sin_family = AF_INET;
133
134 *uaddr_len = sizeof(*sin);
135 return 0;
136}
137
138/*
139 * RDS' poll is without a doubt the least intuitive part of the interface,
140 * as POLLIN and POLLOUT do not behave entirely as you would expect from
141 * a network protocol.
142 *
143 * POLLIN is asserted if
144 * - there is data on the receive queue.
145 * - to signal that a previously congested destination may have become
146 * uncongested
147 * - A notification has been queued to the socket (this can be a congestion
148 * update, or a RDMA completion).
149 *
150 * POLLOUT is asserted if there is room on the send queue. This does not mean
151 * however, that the next sendmsg() call will succeed. If the application tries
152 * to send to a congested destination, the system call may still fail (and
153 * return ENOBUFS).
154 */
155static unsigned int rds_poll(struct file *file, struct socket *sock,
156 poll_table *wait)
157{
158 struct sock *sk = sock->sk;
159 struct rds_sock *rs = rds_sk_to_rs(sk);
160 unsigned int mask = 0;
161 unsigned long flags;
162
163 poll_wait(file, sk_sleep(sk), wait);
164
165 if (rs->rs_seen_congestion)
166 poll_wait(file, &rds_poll_waitq, wait);
167
168 read_lock_irqsave(&rs->rs_recv_lock, flags);
169 if (!rs->rs_cong_monitor) {
170 /* When a congestion map was updated, we signal POLLIN for
171 * "historical" reasons. Applications can also poll for
172 * WRBAND instead. */
173 if (rds_cong_updated_since(&rs->rs_cong_track))
174 mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
175 } else {
176 spin_lock(&rs->rs_lock);
177 if (rs->rs_cong_notify)
178 mask |= (POLLIN | POLLRDNORM);
179 spin_unlock(&rs->rs_lock);
180 }
181 if (!list_empty(&rs->rs_recv_queue) ||
182 !list_empty(&rs->rs_notify_queue))
183 mask |= (POLLIN | POLLRDNORM);
184 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
185 mask |= (POLLOUT | POLLWRNORM);
186 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
187
188 /* clear state any time we wake a seen-congested socket */
189 if (mask)
190 rs->rs_seen_congestion = 0;
191
192 return mask;
193}
194
195static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
196{
197 return -ENOIOCTLCMD;
198}
199
200static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
201 int len)
202{
203 struct sockaddr_in sin;
204 int ret = 0;
205
206 /* racing with another thread binding seems ok here */
207 if (rs->rs_bound_addr == 0) {
208 ret = -ENOTCONN; /* XXX not a great errno */
209 goto out;
210 }
211
212 if (len < sizeof(struct sockaddr_in)) {
213 ret = -EINVAL;
214 goto out;
215 }
216
217 if (copy_from_user(&sin, optval, sizeof(sin))) {
218 ret = -EFAULT;
219 goto out;
220 }
221
222 rds_send_drop_to(rs, &sin);
223out:
224 return ret;
225}
226
227static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
228 int optlen)
229{
230 int value;
231
232 if (optlen < sizeof(int))
233 return -EINVAL;
234 if (get_user(value, (int __user *) optval))
235 return -EFAULT;
236 *optvar = !!value;
237 return 0;
238}
239
240static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
241 int optlen)
242{
243 int ret;
244
245 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
246 if (ret == 0) {
247 if (rs->rs_cong_monitor) {
248 rds_cong_add_socket(rs);
249 } else {
250 rds_cong_remove_socket(rs);
251 rs->rs_cong_mask = 0;
252 rs->rs_cong_notify = 0;
253 }
254 }
255 return ret;
256}
257
258static int rds_set_transport(struct rds_sock *rs, char __user *optval,
259 int optlen)
260{
261 int t_type;
262
263 if (rs->rs_transport)
264 return -EOPNOTSUPP; /* previously attached to transport */
265
266 if (optlen != sizeof(int))
267 return -EINVAL;
268
269 if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
270 return -EFAULT;
271
272 if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
273 return -EINVAL;
274
275 rs->rs_transport = rds_trans_get(t_type);
276
277 return rs->rs_transport ? 0 : -ENOPROTOOPT;
278}
279
280static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
281 int optlen)
282{
283 int val, valbool;
284
285 if (optlen != sizeof(int))
286 return -EFAULT;
287
288 if (get_user(val, (int __user *)optval))
289 return -EFAULT;
290
291 valbool = val ? 1 : 0;
292
293 if (valbool)
294 sock_set_flag(sk, SOCK_RCVTSTAMP);
295 else
296 sock_reset_flag(sk, SOCK_RCVTSTAMP);
297
298 return 0;
299}
300
301static int rds_setsockopt(struct socket *sock, int level, int optname,
302 char __user *optval, unsigned int optlen)
303{
304 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
305 int ret;
306
307 if (level != SOL_RDS) {
308 ret = -ENOPROTOOPT;
309 goto out;
310 }
311
312 switch (optname) {
313 case RDS_CANCEL_SENT_TO:
314 ret = rds_cancel_sent_to(rs, optval, optlen);
315 break;
316 case RDS_GET_MR:
317 ret = rds_get_mr(rs, optval, optlen);
318 break;
319 case RDS_GET_MR_FOR_DEST:
320 ret = rds_get_mr_for_dest(rs, optval, optlen);
321 break;
322 case RDS_FREE_MR:
323 ret = rds_free_mr(rs, optval, optlen);
324 break;
325 case RDS_RECVERR:
326 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
327 break;
328 case RDS_CONG_MONITOR:
329 ret = rds_cong_monitor(rs, optval, optlen);
330 break;
331 case SO_RDS_TRANSPORT:
332 lock_sock(sock->sk);
333 ret = rds_set_transport(rs, optval, optlen);
334 release_sock(sock->sk);
335 break;
336 case SO_TIMESTAMP:
337 lock_sock(sock->sk);
338 ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
339 release_sock(sock->sk);
340 break;
341 default:
342 ret = -ENOPROTOOPT;
343 }
344out:
345 return ret;
346}
347
348static int rds_getsockopt(struct socket *sock, int level, int optname,
349 char __user *optval, int __user *optlen)
350{
351 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
352 int ret = -ENOPROTOOPT, len;
353 int trans;
354
355 if (level != SOL_RDS)
356 goto out;
357
358 if (get_user(len, optlen)) {
359 ret = -EFAULT;
360 goto out;
361 }
362
363 switch (optname) {
364 case RDS_INFO_FIRST ... RDS_INFO_LAST:
365 ret = rds_info_getsockopt(sock, optname, optval,
366 optlen);
367 break;
368
369 case RDS_RECVERR:
370 if (len < sizeof(int))
371 ret = -EINVAL;
372 else
373 if (put_user(rs->rs_recverr, (int __user *) optval) ||
374 put_user(sizeof(int), optlen))
375 ret = -EFAULT;
376 else
377 ret = 0;
378 break;
379 case SO_RDS_TRANSPORT:
380 if (len < sizeof(int)) {
381 ret = -EINVAL;
382 break;
383 }
384 trans = (rs->rs_transport ? rs->rs_transport->t_type :
385 RDS_TRANS_NONE); /* unbound */
386 if (put_user(trans, (int __user *)optval) ||
387 put_user(sizeof(int), optlen))
388 ret = -EFAULT;
389 else
390 ret = 0;
391 break;
392 default:
393 break;
394 }
395
396out:
397 return ret;
398
399}
400
401static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
402 int addr_len, int flags)
403{
404 struct sock *sk = sock->sk;
405 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
406 struct rds_sock *rs = rds_sk_to_rs(sk);
407 int ret = 0;
408
409 lock_sock(sk);
410
411 if (addr_len != sizeof(struct sockaddr_in)) {
412 ret = -EINVAL;
413 goto out;
414 }
415
416 if (sin->sin_family != AF_INET) {
417 ret = -EAFNOSUPPORT;
418 goto out;
419 }
420
421 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
422 ret = -EDESTADDRREQ;
423 goto out;
424 }
425
426 rs->rs_conn_addr = sin->sin_addr.s_addr;
427 rs->rs_conn_port = sin->sin_port;
428
429out:
430 release_sock(sk);
431 return ret;
432}
433
434static struct proto rds_proto = {
435 .name = "RDS",
436 .owner = THIS_MODULE,
437 .obj_size = sizeof(struct rds_sock),
438};
439
440static const struct proto_ops rds_proto_ops = {
441 .family = AF_RDS,
442 .owner = THIS_MODULE,
443 .release = rds_release,
444 .bind = rds_bind,
445 .connect = rds_connect,
446 .socketpair = sock_no_socketpair,
447 .accept = sock_no_accept,
448 .getname = rds_getname,
449 .poll = rds_poll,
450 .ioctl = rds_ioctl,
451 .listen = sock_no_listen,
452 .shutdown = sock_no_shutdown,
453 .setsockopt = rds_setsockopt,
454 .getsockopt = rds_getsockopt,
455 .sendmsg = rds_sendmsg,
456 .recvmsg = rds_recvmsg,
457 .mmap = sock_no_mmap,
458 .sendpage = sock_no_sendpage,
459};
460
461static void rds_sock_destruct(struct sock *sk)
462{
463 struct rds_sock *rs = rds_sk_to_rs(sk);
464
465 WARN_ON((&rs->rs_item != rs->rs_item.next ||
466 &rs->rs_item != rs->rs_item.prev));
467}
468
469static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
470{
471 struct rds_sock *rs;
472
473 sock_init_data(sock, sk);
474 sock->ops = &rds_proto_ops;
475 sk->sk_protocol = protocol;
476 sk->sk_destruct = rds_sock_destruct;
477
478 rs = rds_sk_to_rs(sk);
479 spin_lock_init(&rs->rs_lock);
480 rwlock_init(&rs->rs_recv_lock);
481 INIT_LIST_HEAD(&rs->rs_send_queue);
482 INIT_LIST_HEAD(&rs->rs_recv_queue);
483 INIT_LIST_HEAD(&rs->rs_notify_queue);
484 INIT_LIST_HEAD(&rs->rs_cong_list);
485 spin_lock_init(&rs->rs_rdma_lock);
486 rs->rs_rdma_keys = RB_ROOT;
487
488 spin_lock_bh(&rds_sock_lock);
489 list_add_tail(&rs->rs_item, &rds_sock_list);
490 rds_sock_count++;
491 spin_unlock_bh(&rds_sock_lock);
492
493 return 0;
494}
495
496static int rds_create(struct net *net, struct socket *sock, int protocol,
497 int kern)
498{
499 struct sock *sk;
500
501 if (sock->type != SOCK_SEQPACKET || protocol)
502 return -ESOCKTNOSUPPORT;
503
504 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
505 if (!sk)
506 return -ENOMEM;
507
508 return __rds_create(sock, sk, protocol);
509}
510
511void rds_sock_addref(struct rds_sock *rs)
512{
513 sock_hold(rds_rs_to_sk(rs));
514}
515
516void rds_sock_put(struct rds_sock *rs)
517{
518 sock_put(rds_rs_to_sk(rs));
519}
520
521static const struct net_proto_family rds_family_ops = {
522 .family = AF_RDS,
523 .create = rds_create,
524 .owner = THIS_MODULE,
525};
526
527static void rds_sock_inc_info(struct socket *sock, unsigned int len,
528 struct rds_info_iterator *iter,
529 struct rds_info_lengths *lens)
530{
531 struct rds_sock *rs;
532 struct rds_incoming *inc;
533 unsigned int total = 0;
534
535 len /= sizeof(struct rds_info_message);
536
537 spin_lock_bh(&rds_sock_lock);
538
539 list_for_each_entry(rs, &rds_sock_list, rs_item) {
540 read_lock(&rs->rs_recv_lock);
541
542 /* XXX too lazy to maintain counts.. */
543 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
544 total++;
545 if (total <= len)
546 rds_inc_info_copy(inc, iter, inc->i_saddr,
547 rs->rs_bound_addr, 1);
548 }
549
550 read_unlock(&rs->rs_recv_lock);
551 }
552
553 spin_unlock_bh(&rds_sock_lock);
554
555 lens->nr = total;
556 lens->each = sizeof(struct rds_info_message);
557}
558
559static void rds_sock_info(struct socket *sock, unsigned int len,
560 struct rds_info_iterator *iter,
561 struct rds_info_lengths *lens)
562{
563 struct rds_info_socket sinfo;
564 struct rds_sock *rs;
565
566 len /= sizeof(struct rds_info_socket);
567
568 spin_lock_bh(&rds_sock_lock);
569
570 if (len < rds_sock_count)
571 goto out;
572
573 list_for_each_entry(rs, &rds_sock_list, rs_item) {
574 sinfo.sndbuf = rds_sk_sndbuf(rs);
575 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
576 sinfo.bound_addr = rs->rs_bound_addr;
577 sinfo.connected_addr = rs->rs_conn_addr;
578 sinfo.bound_port = rs->rs_bound_port;
579 sinfo.connected_port = rs->rs_conn_port;
580 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
581
582 rds_info_copy(iter, &sinfo, sizeof(sinfo));
583 }
584
585out:
586 lens->nr = rds_sock_count;
587 lens->each = sizeof(struct rds_info_socket);
588
589 spin_unlock_bh(&rds_sock_lock);
590}
591
592static void rds_exit(void)
593{
594 sock_unregister(rds_family_ops.family);
595 proto_unregister(&rds_proto);
596 rds_conn_exit();
597 rds_cong_exit();
598 rds_sysctl_exit();
599 rds_threads_exit();
600 rds_stats_exit();
601 rds_page_exit();
602 rds_bind_lock_destroy();
603 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
604 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
605}
606module_exit(rds_exit);
607
608static int rds_init(void)
609{
610 int ret;
611
612 ret = rds_bind_lock_init();
613 if (ret)
614 goto out;
615
616 ret = rds_conn_init();
617 if (ret)
618 goto out_bind;
619
620 ret = rds_threads_init();
621 if (ret)
622 goto out_conn;
623 ret = rds_sysctl_init();
624 if (ret)
625 goto out_threads;
626 ret = rds_stats_init();
627 if (ret)
628 goto out_sysctl;
629 ret = proto_register(&rds_proto, 1);
630 if (ret)
631 goto out_stats;
632 ret = sock_register(&rds_family_ops);
633 if (ret)
634 goto out_proto;
635
636 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
637 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
638
639 goto out;
640
641out_proto:
642 proto_unregister(&rds_proto);
643out_stats:
644 rds_stats_exit();
645out_sysctl:
646 rds_sysctl_exit();
647out_threads:
648 rds_threads_exit();
649out_conn:
650 rds_conn_exit();
651 rds_cong_exit();
652 rds_page_exit();
653out_bind:
654 rds_bind_lock_destroy();
655out:
656 return ret;
657}
658module_init(rds_init);
659
660#define DRV_VERSION "4.0"
661#define DRV_RELDATE "Feb 12, 2009"
662
663MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
664MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
665 " v" DRV_VERSION " (" DRV_RELDATE ")");
666MODULE_VERSION(DRV_VERSION);
667MODULE_LICENSE("Dual BSD/GPL");
668MODULE_ALIAS_NETPROTO(PF_RDS);
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/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/poll.h>
39#include <net/sock.h>
40
41#include "rds.h"
42
43char *rds_str_array(char **array, size_t elements, size_t index)
44{
45 if ((index < elements) && array[index])
46 return array[index];
47 else
48 return "unknown";
49}
50EXPORT_SYMBOL(rds_str_array);
51
52/* this is just used for stats gathering :/ */
53static DEFINE_SPINLOCK(rds_sock_lock);
54static unsigned long rds_sock_count;
55static LIST_HEAD(rds_sock_list);
56DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
57
58/*
59 * This is called as the final descriptor referencing this socket is closed.
60 * We have to unbind the socket so that another socket can be bound to the
61 * address it was using.
62 *
63 * We have to be careful about racing with the incoming path. sock_orphan()
64 * sets SOCK_DEAD and we use that as an indicator to the rx path that new
65 * messages shouldn't be queued.
66 */
67static int rds_release(struct socket *sock)
68{
69 struct sock *sk = sock->sk;
70 struct rds_sock *rs;
71
72 if (!sk)
73 goto out;
74
75 rs = rds_sk_to_rs(sk);
76
77 sock_orphan(sk);
78 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
79 * that ensures the recv path has completed messing
80 * with the socket. */
81 rds_clear_recv_queue(rs);
82 rds_cong_remove_socket(rs);
83
84 /*
85 * the binding lookup hash uses rcu, we need to
86 * make sure we sychronize_rcu before we free our
87 * entry
88 */
89 rds_remove_bound(rs);
90 synchronize_rcu();
91
92 rds_send_drop_to(rs, NULL);
93 rds_rdma_drop_keys(rs);
94 rds_notify_queue_get(rs, NULL);
95
96 spin_lock_bh(&rds_sock_lock);
97 list_del_init(&rs->rs_item);
98 rds_sock_count--;
99 spin_unlock_bh(&rds_sock_lock);
100
101 rds_trans_put(rs->rs_transport);
102
103 sock->sk = NULL;
104 sock_put(sk);
105out:
106 return 0;
107}
108
109/*
110 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
111 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
112 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
113 * this seems more conservative.
114 * NB - normally, one would use sk_callback_lock for this, but we can
115 * get here from interrupts, whereas the network code grabs sk_callback_lock
116 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
117 */
118void rds_wake_sk_sleep(struct rds_sock *rs)
119{
120 unsigned long flags;
121
122 read_lock_irqsave(&rs->rs_recv_lock, flags);
123 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
124 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
125}
126
127static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
128 int *uaddr_len, int peer)
129{
130 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
131 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
132
133 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
134
135 /* racey, don't care */
136 if (peer) {
137 if (!rs->rs_conn_addr)
138 return -ENOTCONN;
139
140 sin->sin_port = rs->rs_conn_port;
141 sin->sin_addr.s_addr = rs->rs_conn_addr;
142 } else {
143 sin->sin_port = rs->rs_bound_port;
144 sin->sin_addr.s_addr = rs->rs_bound_addr;
145 }
146
147 sin->sin_family = AF_INET;
148
149 *uaddr_len = sizeof(*sin);
150 return 0;
151}
152
153/*
154 * RDS' poll is without a doubt the least intuitive part of the interface,
155 * as POLLIN and POLLOUT do not behave entirely as you would expect from
156 * a network protocol.
157 *
158 * POLLIN is asserted if
159 * - there is data on the receive queue.
160 * - to signal that a previously congested destination may have become
161 * uncongested
162 * - A notification has been queued to the socket (this can be a congestion
163 * update, or a RDMA completion).
164 *
165 * POLLOUT is asserted if there is room on the send queue. This does not mean
166 * however, that the next sendmsg() call will succeed. If the application tries
167 * to send to a congested destination, the system call may still fail (and
168 * return ENOBUFS).
169 */
170static unsigned int rds_poll(struct file *file, struct socket *sock,
171 poll_table *wait)
172{
173 struct sock *sk = sock->sk;
174 struct rds_sock *rs = rds_sk_to_rs(sk);
175 unsigned int mask = 0;
176 unsigned long flags;
177
178 poll_wait(file, sk_sleep(sk), wait);
179
180 if (rs->rs_seen_congestion)
181 poll_wait(file, &rds_poll_waitq, wait);
182
183 read_lock_irqsave(&rs->rs_recv_lock, flags);
184 if (!rs->rs_cong_monitor) {
185 /* When a congestion map was updated, we signal POLLIN for
186 * "historical" reasons. Applications can also poll for
187 * WRBAND instead. */
188 if (rds_cong_updated_since(&rs->rs_cong_track))
189 mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
190 } else {
191 spin_lock(&rs->rs_lock);
192 if (rs->rs_cong_notify)
193 mask |= (POLLIN | POLLRDNORM);
194 spin_unlock(&rs->rs_lock);
195 }
196 if (!list_empty(&rs->rs_recv_queue) ||
197 !list_empty(&rs->rs_notify_queue))
198 mask |= (POLLIN | POLLRDNORM);
199 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
200 mask |= (POLLOUT | POLLWRNORM);
201 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
202
203 /* clear state any time we wake a seen-congested socket */
204 if (mask)
205 rs->rs_seen_congestion = 0;
206
207 return mask;
208}
209
210static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
211{
212 return -ENOIOCTLCMD;
213}
214
215static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
216 int len)
217{
218 struct sockaddr_in sin;
219 int ret = 0;
220
221 /* racing with another thread binding seems ok here */
222 if (rs->rs_bound_addr == 0) {
223 ret = -ENOTCONN; /* XXX not a great errno */
224 goto out;
225 }
226
227 if (len < sizeof(struct sockaddr_in)) {
228 ret = -EINVAL;
229 goto out;
230 }
231
232 if (copy_from_user(&sin, optval, sizeof(sin))) {
233 ret = -EFAULT;
234 goto out;
235 }
236
237 rds_send_drop_to(rs, &sin);
238out:
239 return ret;
240}
241
242static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
243 int optlen)
244{
245 int value;
246
247 if (optlen < sizeof(int))
248 return -EINVAL;
249 if (get_user(value, (int __user *) optval))
250 return -EFAULT;
251 *optvar = !!value;
252 return 0;
253}
254
255static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
256 int optlen)
257{
258 int ret;
259
260 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
261 if (ret == 0) {
262 if (rs->rs_cong_monitor) {
263 rds_cong_add_socket(rs);
264 } else {
265 rds_cong_remove_socket(rs);
266 rs->rs_cong_mask = 0;
267 rs->rs_cong_notify = 0;
268 }
269 }
270 return ret;
271}
272
273static int rds_setsockopt(struct socket *sock, int level, int optname,
274 char __user *optval, unsigned int optlen)
275{
276 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
277 int ret;
278
279 if (level != SOL_RDS) {
280 ret = -ENOPROTOOPT;
281 goto out;
282 }
283
284 switch (optname) {
285 case RDS_CANCEL_SENT_TO:
286 ret = rds_cancel_sent_to(rs, optval, optlen);
287 break;
288 case RDS_GET_MR:
289 ret = rds_get_mr(rs, optval, optlen);
290 break;
291 case RDS_GET_MR_FOR_DEST:
292 ret = rds_get_mr_for_dest(rs, optval, optlen);
293 break;
294 case RDS_FREE_MR:
295 ret = rds_free_mr(rs, optval, optlen);
296 break;
297 case RDS_RECVERR:
298 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
299 break;
300 case RDS_CONG_MONITOR:
301 ret = rds_cong_monitor(rs, optval, optlen);
302 break;
303 default:
304 ret = -ENOPROTOOPT;
305 }
306out:
307 return ret;
308}
309
310static int rds_getsockopt(struct socket *sock, int level, int optname,
311 char __user *optval, int __user *optlen)
312{
313 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
314 int ret = -ENOPROTOOPT, len;
315
316 if (level != SOL_RDS)
317 goto out;
318
319 if (get_user(len, optlen)) {
320 ret = -EFAULT;
321 goto out;
322 }
323
324 switch (optname) {
325 case RDS_INFO_FIRST ... RDS_INFO_LAST:
326 ret = rds_info_getsockopt(sock, optname, optval,
327 optlen);
328 break;
329
330 case RDS_RECVERR:
331 if (len < sizeof(int))
332 ret = -EINVAL;
333 else
334 if (put_user(rs->rs_recverr, (int __user *) optval) ||
335 put_user(sizeof(int), optlen))
336 ret = -EFAULT;
337 else
338 ret = 0;
339 break;
340 default:
341 break;
342 }
343
344out:
345 return ret;
346
347}
348
349static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
350 int addr_len, int flags)
351{
352 struct sock *sk = sock->sk;
353 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
354 struct rds_sock *rs = rds_sk_to_rs(sk);
355 int ret = 0;
356
357 lock_sock(sk);
358
359 if (addr_len != sizeof(struct sockaddr_in)) {
360 ret = -EINVAL;
361 goto out;
362 }
363
364 if (sin->sin_family != AF_INET) {
365 ret = -EAFNOSUPPORT;
366 goto out;
367 }
368
369 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
370 ret = -EDESTADDRREQ;
371 goto out;
372 }
373
374 rs->rs_conn_addr = sin->sin_addr.s_addr;
375 rs->rs_conn_port = sin->sin_port;
376
377out:
378 release_sock(sk);
379 return ret;
380}
381
382static struct proto rds_proto = {
383 .name = "RDS",
384 .owner = THIS_MODULE,
385 .obj_size = sizeof(struct rds_sock),
386};
387
388static const struct proto_ops rds_proto_ops = {
389 .family = AF_RDS,
390 .owner = THIS_MODULE,
391 .release = rds_release,
392 .bind = rds_bind,
393 .connect = rds_connect,
394 .socketpair = sock_no_socketpair,
395 .accept = sock_no_accept,
396 .getname = rds_getname,
397 .poll = rds_poll,
398 .ioctl = rds_ioctl,
399 .listen = sock_no_listen,
400 .shutdown = sock_no_shutdown,
401 .setsockopt = rds_setsockopt,
402 .getsockopt = rds_getsockopt,
403 .sendmsg = rds_sendmsg,
404 .recvmsg = rds_recvmsg,
405 .mmap = sock_no_mmap,
406 .sendpage = sock_no_sendpage,
407};
408
409static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
410{
411 struct rds_sock *rs;
412
413 sock_init_data(sock, sk);
414 sock->ops = &rds_proto_ops;
415 sk->sk_protocol = protocol;
416
417 rs = rds_sk_to_rs(sk);
418 spin_lock_init(&rs->rs_lock);
419 rwlock_init(&rs->rs_recv_lock);
420 INIT_LIST_HEAD(&rs->rs_send_queue);
421 INIT_LIST_HEAD(&rs->rs_recv_queue);
422 INIT_LIST_HEAD(&rs->rs_notify_queue);
423 INIT_LIST_HEAD(&rs->rs_cong_list);
424 spin_lock_init(&rs->rs_rdma_lock);
425 rs->rs_rdma_keys = RB_ROOT;
426
427 spin_lock_bh(&rds_sock_lock);
428 list_add_tail(&rs->rs_item, &rds_sock_list);
429 rds_sock_count++;
430 spin_unlock_bh(&rds_sock_lock);
431
432 return 0;
433}
434
435static int rds_create(struct net *net, struct socket *sock, int protocol,
436 int kern)
437{
438 struct sock *sk;
439
440 if (sock->type != SOCK_SEQPACKET || protocol)
441 return -ESOCKTNOSUPPORT;
442
443 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
444 if (!sk)
445 return -ENOMEM;
446
447 return __rds_create(sock, sk, protocol);
448}
449
450void rds_sock_addref(struct rds_sock *rs)
451{
452 sock_hold(rds_rs_to_sk(rs));
453}
454
455void rds_sock_put(struct rds_sock *rs)
456{
457 sock_put(rds_rs_to_sk(rs));
458}
459
460static const struct net_proto_family rds_family_ops = {
461 .family = AF_RDS,
462 .create = rds_create,
463 .owner = THIS_MODULE,
464};
465
466static void rds_sock_inc_info(struct socket *sock, unsigned int len,
467 struct rds_info_iterator *iter,
468 struct rds_info_lengths *lens)
469{
470 struct rds_sock *rs;
471 struct rds_incoming *inc;
472 unsigned int total = 0;
473
474 len /= sizeof(struct rds_info_message);
475
476 spin_lock_bh(&rds_sock_lock);
477
478 list_for_each_entry(rs, &rds_sock_list, rs_item) {
479 read_lock(&rs->rs_recv_lock);
480
481 /* XXX too lazy to maintain counts.. */
482 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
483 total++;
484 if (total <= len)
485 rds_inc_info_copy(inc, iter, inc->i_saddr,
486 rs->rs_bound_addr, 1);
487 }
488
489 read_unlock(&rs->rs_recv_lock);
490 }
491
492 spin_unlock_bh(&rds_sock_lock);
493
494 lens->nr = total;
495 lens->each = sizeof(struct rds_info_message);
496}
497
498static void rds_sock_info(struct socket *sock, unsigned int len,
499 struct rds_info_iterator *iter,
500 struct rds_info_lengths *lens)
501{
502 struct rds_info_socket sinfo;
503 struct rds_sock *rs;
504
505 len /= sizeof(struct rds_info_socket);
506
507 spin_lock_bh(&rds_sock_lock);
508
509 if (len < rds_sock_count)
510 goto out;
511
512 list_for_each_entry(rs, &rds_sock_list, rs_item) {
513 sinfo.sndbuf = rds_sk_sndbuf(rs);
514 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
515 sinfo.bound_addr = rs->rs_bound_addr;
516 sinfo.connected_addr = rs->rs_conn_addr;
517 sinfo.bound_port = rs->rs_bound_port;
518 sinfo.connected_port = rs->rs_conn_port;
519 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
520
521 rds_info_copy(iter, &sinfo, sizeof(sinfo));
522 }
523
524out:
525 lens->nr = rds_sock_count;
526 lens->each = sizeof(struct rds_info_socket);
527
528 spin_unlock_bh(&rds_sock_lock);
529}
530
531static void rds_exit(void)
532{
533 sock_unregister(rds_family_ops.family);
534 proto_unregister(&rds_proto);
535 rds_conn_exit();
536 rds_cong_exit();
537 rds_sysctl_exit();
538 rds_threads_exit();
539 rds_stats_exit();
540 rds_page_exit();
541 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
542 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
543}
544module_exit(rds_exit);
545
546static int rds_init(void)
547{
548 int ret;
549
550 ret = rds_conn_init();
551 if (ret)
552 goto out;
553 ret = rds_threads_init();
554 if (ret)
555 goto out_conn;
556 ret = rds_sysctl_init();
557 if (ret)
558 goto out_threads;
559 ret = rds_stats_init();
560 if (ret)
561 goto out_sysctl;
562 ret = proto_register(&rds_proto, 1);
563 if (ret)
564 goto out_stats;
565 ret = sock_register(&rds_family_ops);
566 if (ret)
567 goto out_proto;
568
569 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
570 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
571
572 goto out;
573
574out_proto:
575 proto_unregister(&rds_proto);
576out_stats:
577 rds_stats_exit();
578out_sysctl:
579 rds_sysctl_exit();
580out_threads:
581 rds_threads_exit();
582out_conn:
583 rds_conn_exit();
584 rds_cong_exit();
585 rds_page_exit();
586out:
587 return ret;
588}
589module_init(rds_init);
590
591#define DRV_VERSION "4.0"
592#define DRV_RELDATE "Feb 12, 2009"
593
594MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
595MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
596 " v" DRV_VERSION " (" DRV_RELDATE ")");
597MODULE_VERSION(DRV_VERSION);
598MODULE_LICENSE("Dual BSD/GPL");
599MODULE_ALIAS_NETPROTO(PF_RDS);