2 * Copyright (c) 2006 Oracle. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/module.h>
34 #include <linux/errno.h>
35 #include <linux/kernel.h>
36 #include <linux/gfp.h>
38 #include <linux/poll.h>
43 /* this is just used for stats gathering :/ */
44 static DEFINE_SPINLOCK(rds_sock_lock);
45 static unsigned long rds_sock_count;
46 static LIST_HEAD(rds_sock_list);
47 DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
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.
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.
58 static int rds_release(struct socket *sock)
60 struct sock *sk = sock->sk;
67 rs = rds_sk_to_rs(sk);
70 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
71 * that ensures the recv path has completed messing
73 rds_clear_recv_queue(rs);
74 rds_cong_remove_socket(rs);
77 * the binding lookup hash uses rcu, we need to
78 * make sure we sychronize_rcu before we free our
84 rds_send_drop_to(rs, NULL);
85 rds_rdma_drop_keys(rs);
86 rds_notify_queue_get(rs, NULL);
88 spin_lock_irqsave(&rds_sock_lock, flags);
89 list_del_init(&rs->rs_item);
91 spin_unlock_irqrestore(&rds_sock_lock, flags);
100 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
101 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
102 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
103 * this seems more conservative.
104 * NB - normally, one would use sk_callback_lock for this, but we can
105 * get here from interrupts, whereas the network code grabs sk_callback_lock
106 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
108 void rds_wake_sk_sleep(struct rds_sock *rs)
112 read_lock_irqsave(&rs->rs_recv_lock, flags);
113 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
114 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
117 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
118 int *uaddr_len, int peer)
120 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
121 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
123 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
125 /* racey, don't care */
127 if (!rs->rs_conn_addr)
130 sin->sin_port = rs->rs_conn_port;
131 sin->sin_addr.s_addr = rs->rs_conn_addr;
133 sin->sin_port = rs->rs_bound_port;
134 sin->sin_addr.s_addr = rs->rs_bound_addr;
137 sin->sin_family = AF_INET;
139 *uaddr_len = sizeof(*sin);
144 * RDS' poll is without a doubt the least intuitive part of the interface,
145 * as POLLIN and POLLOUT do not behave entirely as you would expect from
146 * a network protocol.
148 * POLLIN is asserted if
149 * - there is data on the receive queue.
150 * - to signal that a previously congested destination may have become
152 * - A notification has been queued to the socket (this can be a congestion
153 * update, or a RDMA completion).
155 * POLLOUT is asserted if there is room on the send queue. This does not mean
156 * however, that the next sendmsg() call will succeed. If the application tries
157 * to send to a congested destination, the system call may still fail (and
160 static unsigned int rds_poll(struct file *file, struct socket *sock,
163 struct sock *sk = sock->sk;
164 struct rds_sock *rs = rds_sk_to_rs(sk);
165 unsigned int mask = 0;
168 poll_wait(file, sk_sleep(sk), wait);
170 if (rs->rs_seen_congestion)
171 poll_wait(file, &rds_poll_waitq, wait);
173 read_lock_irqsave(&rs->rs_recv_lock, flags);
174 if (!rs->rs_cong_monitor) {
175 /* When a congestion map was updated, we signal POLLIN for
176 * "historical" reasons. Applications can also poll for
178 if (rds_cong_updated_since(&rs->rs_cong_track))
179 mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
181 spin_lock(&rs->rs_lock);
182 if (rs->rs_cong_notify)
183 mask |= (POLLIN | POLLRDNORM);
184 spin_unlock(&rs->rs_lock);
186 if (!list_empty(&rs->rs_recv_queue) ||
187 !list_empty(&rs->rs_notify_queue))
188 mask |= (POLLIN | POLLRDNORM);
189 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
190 mask |= (POLLOUT | POLLWRNORM);
191 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
193 /* clear state any time we wake a seen-congested socket */
195 rs->rs_seen_congestion = 0;
200 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
205 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
208 struct sockaddr_in sin;
211 /* racing with another thread binding seems ok here */
212 if (rs->rs_bound_addr == 0) {
213 ret = -ENOTCONN; /* XXX not a great errno */
217 if (len < sizeof(struct sockaddr_in)) {
222 if (copy_from_user(&sin, optval, sizeof(sin))) {
227 rds_send_drop_to(rs, &sin);
232 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
237 if (optlen < sizeof(int))
239 if (get_user(value, (int __user *) optval))
245 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
250 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
252 if (rs->rs_cong_monitor) {
253 rds_cong_add_socket(rs);
255 rds_cong_remove_socket(rs);
256 rs->rs_cong_mask = 0;
257 rs->rs_cong_notify = 0;
263 static int rds_setsockopt(struct socket *sock, int level, int optname,
264 char __user *optval, unsigned int optlen)
266 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
269 if (level != SOL_RDS) {
275 case RDS_CANCEL_SENT_TO:
276 ret = rds_cancel_sent_to(rs, optval, optlen);
279 ret = rds_get_mr(rs, optval, optlen);
281 case RDS_GET_MR_FOR_DEST:
282 ret = rds_get_mr_for_dest(rs, optval, optlen);
285 ret = rds_free_mr(rs, optval, optlen);
288 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
290 case RDS_CONG_MONITOR:
291 ret = rds_cong_monitor(rs, optval, optlen);
300 static int rds_getsockopt(struct socket *sock, int level, int optname,
301 char __user *optval, int __user *optlen)
303 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
304 int ret = -ENOPROTOOPT, len;
306 if (level != SOL_RDS)
309 if (get_user(len, optlen)) {
315 case RDS_INFO_FIRST ... RDS_INFO_LAST:
316 ret = rds_info_getsockopt(sock, optname, optval,
321 if (len < sizeof(int))
324 if (put_user(rs->rs_recverr, (int __user *) optval) ||
325 put_user(sizeof(int), optlen))
339 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
340 int addr_len, int flags)
342 struct sock *sk = sock->sk;
343 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
344 struct rds_sock *rs = rds_sk_to_rs(sk);
349 if (addr_len != sizeof(struct sockaddr_in)) {
354 if (sin->sin_family != AF_INET) {
359 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
364 rs->rs_conn_addr = sin->sin_addr.s_addr;
365 rs->rs_conn_port = sin->sin_port;
372 static struct proto rds_proto = {
374 .owner = THIS_MODULE,
375 .obj_size = sizeof(struct rds_sock),
378 static const struct proto_ops rds_proto_ops = {
380 .owner = THIS_MODULE,
381 .release = rds_release,
383 .connect = rds_connect,
384 .socketpair = sock_no_socketpair,
385 .accept = sock_no_accept,
386 .getname = rds_getname,
389 .listen = sock_no_listen,
390 .shutdown = sock_no_shutdown,
391 .setsockopt = rds_setsockopt,
392 .getsockopt = rds_getsockopt,
393 .sendmsg = rds_sendmsg,
394 .recvmsg = rds_recvmsg,
395 .mmap = sock_no_mmap,
396 .sendpage = sock_no_sendpage,
399 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
404 sock_init_data(sock, sk);
405 sock->ops = &rds_proto_ops;
406 sk->sk_protocol = protocol;
408 rs = rds_sk_to_rs(sk);
409 spin_lock_init(&rs->rs_lock);
410 rwlock_init(&rs->rs_recv_lock);
411 INIT_LIST_HEAD(&rs->rs_send_queue);
412 INIT_LIST_HEAD(&rs->rs_recv_queue);
413 INIT_LIST_HEAD(&rs->rs_notify_queue);
414 INIT_LIST_HEAD(&rs->rs_cong_list);
415 spin_lock_init(&rs->rs_rdma_lock);
416 rs->rs_rdma_keys = RB_ROOT;
418 spin_lock_irqsave(&rds_sock_lock, flags);
419 list_add_tail(&rs->rs_item, &rds_sock_list);
421 spin_unlock_irqrestore(&rds_sock_lock, flags);
426 static int rds_create(struct net *net, struct socket *sock, int protocol,
431 if (sock->type != SOCK_SEQPACKET || protocol)
432 return -ESOCKTNOSUPPORT;
434 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
438 return __rds_create(sock, sk, protocol);
441 void rds_sock_addref(struct rds_sock *rs)
443 sock_hold(rds_rs_to_sk(rs));
446 void rds_sock_put(struct rds_sock *rs)
448 sock_put(rds_rs_to_sk(rs));
451 static const struct net_proto_family rds_family_ops = {
453 .create = rds_create,
454 .owner = THIS_MODULE,
457 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
458 struct rds_info_iterator *iter,
459 struct rds_info_lengths *lens)
462 struct rds_incoming *inc;
464 unsigned int total = 0;
466 len /= sizeof(struct rds_info_message);
468 spin_lock_irqsave(&rds_sock_lock, flags);
470 list_for_each_entry(rs, &rds_sock_list, rs_item) {
471 read_lock(&rs->rs_recv_lock);
473 /* XXX too lazy to maintain counts.. */
474 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
477 rds_inc_info_copy(inc, iter, inc->i_saddr,
478 rs->rs_bound_addr, 1);
481 read_unlock(&rs->rs_recv_lock);
484 spin_unlock_irqrestore(&rds_sock_lock, flags);
487 lens->each = sizeof(struct rds_info_message);
490 static void rds_sock_info(struct socket *sock, unsigned int len,
491 struct rds_info_iterator *iter,
492 struct rds_info_lengths *lens)
494 struct rds_info_socket sinfo;
498 len /= sizeof(struct rds_info_socket);
500 spin_lock_irqsave(&rds_sock_lock, flags);
502 if (len < rds_sock_count)
505 list_for_each_entry(rs, &rds_sock_list, rs_item) {
506 sinfo.sndbuf = rds_sk_sndbuf(rs);
507 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
508 sinfo.bound_addr = rs->rs_bound_addr;
509 sinfo.connected_addr = rs->rs_conn_addr;
510 sinfo.bound_port = rs->rs_bound_port;
511 sinfo.connected_port = rs->rs_conn_port;
512 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
514 rds_info_copy(iter, &sinfo, sizeof(sinfo));
518 lens->nr = rds_sock_count;
519 lens->each = sizeof(struct rds_info_socket);
521 spin_unlock_irqrestore(&rds_sock_lock, flags);
524 static void __exit rds_exit(void)
526 sock_unregister(rds_family_ops.family);
527 proto_unregister(&rds_proto);
534 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
535 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
537 module_exit(rds_exit);
539 static int __init rds_init(void)
543 ret = rds_conn_init();
546 ret = rds_threads_init();
549 ret = rds_sysctl_init();
552 ret = rds_stats_init();
555 ret = proto_register(&rds_proto, 1);
558 ret = sock_register(&rds_family_ops);
562 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
563 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
568 proto_unregister(&rds_proto);
582 module_init(rds_init);
584 #define DRV_VERSION "4.0"
585 #define DRV_RELDATE "Feb 12, 2009"
587 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
588 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
589 " v" DRV_VERSION " (" DRV_RELDATE ")");
590 MODULE_VERSION(DRV_VERSION);
591 MODULE_LICENSE("Dual BSD/GPL");
592 MODULE_ALIAS_NETPROTO(PF_RDS);