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>
37 #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);
76 rds_send_drop_to(rs, NULL);
77 rds_rdma_drop_keys(rs);
78 rds_notify_queue_get(rs, NULL);
80 spin_lock_irqsave(&rds_sock_lock, flags);
81 list_del_init(&rs->rs_item);
83 spin_unlock_irqrestore(&rds_sock_lock, flags);
92 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
93 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
94 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
95 * this seems more conservative.
96 * NB - normally, one would use sk_callback_lock for this, but we can
97 * get here from interrupts, whereas the network code grabs sk_callback_lock
98 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
100 void rds_wake_sk_sleep(struct rds_sock *rs)
104 read_lock_irqsave(&rs->rs_recv_lock, flags);
105 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
106 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
109 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
110 int *uaddr_len, int peer)
112 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
113 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
115 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
117 /* racey, don't care */
119 if (!rs->rs_conn_addr)
122 sin->sin_port = rs->rs_conn_port;
123 sin->sin_addr.s_addr = rs->rs_conn_addr;
125 sin->sin_port = rs->rs_bound_port;
126 sin->sin_addr.s_addr = rs->rs_bound_addr;
129 sin->sin_family = AF_INET;
131 *uaddr_len = sizeof(*sin);
136 * RDS' poll is without a doubt the least intuitive part of the interface,
137 * as POLLIN and POLLOUT do not behave entirely as you would expect from
138 * a network protocol.
140 * POLLIN is asserted if
141 * - there is data on the receive queue.
142 * - to signal that a previously congested destination may have become
144 * - A notification has been queued to the socket (this can be a congestion
145 * update, or a RDMA completion).
147 * POLLOUT is asserted if there is room on the send queue. This does not mean
148 * however, that the next sendmsg() call will succeed. If the application tries
149 * to send to a congested destination, the system call may still fail (and
152 static unsigned int rds_poll(struct file *file, struct socket *sock,
155 struct sock *sk = sock->sk;
156 struct rds_sock *rs = rds_sk_to_rs(sk);
157 unsigned int mask = 0;
160 poll_wait(file, sk->sk_sleep, wait);
162 poll_wait(file, &rds_poll_waitq, wait);
164 read_lock_irqsave(&rs->rs_recv_lock, flags);
165 if (!rs->rs_cong_monitor) {
166 /* When a congestion map was updated, we signal POLLIN for
167 * "historical" reasons. Applications can also poll for
169 if (rds_cong_updated_since(&rs->rs_cong_track))
170 mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
172 spin_lock(&rs->rs_lock);
173 if (rs->rs_cong_notify)
174 mask |= (POLLIN | POLLRDNORM);
175 spin_unlock(&rs->rs_lock);
177 if (!list_empty(&rs->rs_recv_queue) ||
178 !list_empty(&rs->rs_notify_queue))
179 mask |= (POLLIN | POLLRDNORM);
180 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
181 mask |= (POLLOUT | POLLWRNORM);
182 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
187 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
192 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
195 struct sockaddr_in sin;
198 /* racing with another thread binding seems ok here */
199 if (rs->rs_bound_addr == 0) {
200 ret = -ENOTCONN; /* XXX not a great errno */
204 if (len < sizeof(struct sockaddr_in)) {
209 if (copy_from_user(&sin, optval, sizeof(sin))) {
214 rds_send_drop_to(rs, &sin);
219 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
224 if (optlen < sizeof(int))
226 if (get_user(value, (int __user *) optval))
232 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
237 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
239 if (rs->rs_cong_monitor) {
240 rds_cong_add_socket(rs);
242 rds_cong_remove_socket(rs);
243 rs->rs_cong_mask = 0;
244 rs->rs_cong_notify = 0;
250 static int rds_setsockopt(struct socket *sock, int level, int optname,
251 char __user *optval, unsigned int optlen)
253 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
256 if (level != SOL_RDS) {
262 case RDS_CANCEL_SENT_TO:
263 ret = rds_cancel_sent_to(rs, optval, optlen);
266 ret = rds_get_mr(rs, optval, optlen);
268 case RDS_GET_MR_FOR_DEST:
269 ret = rds_get_mr_for_dest(rs, optval, optlen);
272 ret = rds_free_mr(rs, optval, optlen);
275 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
277 case RDS_CONG_MONITOR:
278 ret = rds_cong_monitor(rs, optval, optlen);
287 static int rds_getsockopt(struct socket *sock, int level, int optname,
288 char __user *optval, int __user *optlen)
290 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
291 int ret = -ENOPROTOOPT, len;
293 if (level != SOL_RDS)
296 if (get_user(len, optlen)) {
302 case RDS_INFO_FIRST ... RDS_INFO_LAST:
303 ret = rds_info_getsockopt(sock, optname, optval,
308 if (len < sizeof(int))
311 if (put_user(rs->rs_recverr, (int __user *) optval) ||
312 put_user(sizeof(int), optlen))
326 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
327 int addr_len, int flags)
329 struct sock *sk = sock->sk;
330 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
331 struct rds_sock *rs = rds_sk_to_rs(sk);
336 if (addr_len != sizeof(struct sockaddr_in)) {
341 if (sin->sin_family != AF_INET) {
346 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
351 rs->rs_conn_addr = sin->sin_addr.s_addr;
352 rs->rs_conn_port = sin->sin_port;
359 static struct proto rds_proto = {
361 .owner = THIS_MODULE,
362 .obj_size = sizeof(struct rds_sock),
365 static const struct proto_ops rds_proto_ops = {
367 .owner = THIS_MODULE,
368 .release = rds_release,
370 .connect = rds_connect,
371 .socketpair = sock_no_socketpair,
372 .accept = sock_no_accept,
373 .getname = rds_getname,
376 .listen = sock_no_listen,
377 .shutdown = sock_no_shutdown,
378 .setsockopt = rds_setsockopt,
379 .getsockopt = rds_getsockopt,
380 .sendmsg = rds_sendmsg,
381 .recvmsg = rds_recvmsg,
382 .mmap = sock_no_mmap,
383 .sendpage = sock_no_sendpage,
386 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
391 sock_init_data(sock, sk);
392 sock->ops = &rds_proto_ops;
393 sk->sk_protocol = protocol;
395 rs = rds_sk_to_rs(sk);
396 spin_lock_init(&rs->rs_lock);
397 rwlock_init(&rs->rs_recv_lock);
398 INIT_LIST_HEAD(&rs->rs_send_queue);
399 INIT_LIST_HEAD(&rs->rs_recv_queue);
400 INIT_LIST_HEAD(&rs->rs_notify_queue);
401 INIT_LIST_HEAD(&rs->rs_cong_list);
402 spin_lock_init(&rs->rs_rdma_lock);
403 rs->rs_rdma_keys = RB_ROOT;
405 spin_lock_irqsave(&rds_sock_lock, flags);
406 list_add_tail(&rs->rs_item, &rds_sock_list);
408 spin_unlock_irqrestore(&rds_sock_lock, flags);
413 static int rds_create(struct net *net, struct socket *sock, int protocol,
418 if (sock->type != SOCK_SEQPACKET || protocol)
419 return -ESOCKTNOSUPPORT;
421 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
425 return __rds_create(sock, sk, protocol);
428 void rds_sock_addref(struct rds_sock *rs)
430 sock_hold(rds_rs_to_sk(rs));
433 void rds_sock_put(struct rds_sock *rs)
435 sock_put(rds_rs_to_sk(rs));
438 static const struct net_proto_family rds_family_ops = {
440 .create = rds_create,
441 .owner = THIS_MODULE,
444 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
445 struct rds_info_iterator *iter,
446 struct rds_info_lengths *lens)
450 struct rds_incoming *inc;
452 unsigned int total = 0;
454 len /= sizeof(struct rds_info_message);
456 spin_lock_irqsave(&rds_sock_lock, flags);
458 list_for_each_entry(rs, &rds_sock_list, rs_item) {
459 sk = rds_rs_to_sk(rs);
460 read_lock(&rs->rs_recv_lock);
462 /* XXX too lazy to maintain counts.. */
463 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
466 rds_inc_info_copy(inc, iter, inc->i_saddr,
467 rs->rs_bound_addr, 1);
470 read_unlock(&rs->rs_recv_lock);
473 spin_unlock_irqrestore(&rds_sock_lock, flags);
476 lens->each = sizeof(struct rds_info_message);
479 static void rds_sock_info(struct socket *sock, unsigned int len,
480 struct rds_info_iterator *iter,
481 struct rds_info_lengths *lens)
483 struct rds_info_socket sinfo;
487 len /= sizeof(struct rds_info_socket);
489 spin_lock_irqsave(&rds_sock_lock, flags);
491 if (len < rds_sock_count)
494 list_for_each_entry(rs, &rds_sock_list, rs_item) {
495 sinfo.sndbuf = rds_sk_sndbuf(rs);
496 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
497 sinfo.bound_addr = rs->rs_bound_addr;
498 sinfo.connected_addr = rs->rs_conn_addr;
499 sinfo.bound_port = rs->rs_bound_port;
500 sinfo.connected_port = rs->rs_conn_port;
501 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
503 rds_info_copy(iter, &sinfo, sizeof(sinfo));
507 lens->nr = rds_sock_count;
508 lens->each = sizeof(struct rds_info_socket);
510 spin_unlock_irqrestore(&rds_sock_lock, flags);
513 static void __exit rds_exit(void)
515 sock_unregister(rds_family_ops.family);
516 proto_unregister(&rds_proto);
523 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
524 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
526 module_exit(rds_exit);
528 static int __init rds_init(void)
532 ret = rds_conn_init();
535 ret = rds_threads_init();
538 ret = rds_sysctl_init();
541 ret = rds_stats_init();
544 ret = proto_register(&rds_proto, 1);
547 ret = sock_register(&rds_family_ops);
551 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
552 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
557 proto_unregister(&rds_proto);
571 module_init(rds_init);
573 #define DRV_VERSION "4.0"
574 #define DRV_RELDATE "Feb 12, 2009"
577 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
578 " v" DRV_VERSION " (" DRV_RELDATE ")");
579 MODULE_VERSION(DRV_VERSION);
580 MODULE_LICENSE("Dual BSD/GPL");
581 MODULE_ALIAS_NETPROTO(PF_RDS);