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>
42 #include "rdma_transport.h"
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);
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.
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.
59 static int rds_release(struct socket *sock)
61 struct sock *sk = sock->sk;
68 rs = rds_sk_to_rs(sk);
71 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
72 * that ensures the recv path has completed messing
74 rds_clear_recv_queue(rs);
75 rds_cong_remove_socket(rs);
77 rds_send_drop_to(rs, NULL);
78 rds_rdma_drop_keys(rs);
79 rds_notify_queue_get(rs, NULL);
81 spin_lock_irqsave(&rds_sock_lock, flags);
82 list_del_init(&rs->rs_item);
84 spin_unlock_irqrestore(&rds_sock_lock, flags);
93 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
94 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
95 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
96 * this seems more conservative.
97 * NB - normally, one would use sk_callback_lock for this, but we can
98 * get here from interrupts, whereas the network code grabs sk_callback_lock
99 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
101 void rds_wake_sk_sleep(struct rds_sock *rs)
105 read_lock_irqsave(&rs->rs_recv_lock, flags);
106 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
107 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
110 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
111 int *uaddr_len, int peer)
113 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
114 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
116 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
118 /* racey, don't care */
120 if (!rs->rs_conn_addr)
123 sin->sin_port = rs->rs_conn_port;
124 sin->sin_addr.s_addr = rs->rs_conn_addr;
126 sin->sin_port = rs->rs_bound_port;
127 sin->sin_addr.s_addr = rs->rs_bound_addr;
130 sin->sin_family = AF_INET;
132 *uaddr_len = sizeof(*sin);
137 * RDS' poll is without a doubt the least intuitive part of the interface,
138 * as POLLIN and POLLOUT do not behave entirely as you would expect from
139 * a network protocol.
141 * POLLIN is asserted if
142 * - there is data on the receive queue.
143 * - to signal that a previously congested destination may have become
145 * - A notification has been queued to the socket (this can be a congestion
146 * update, or a RDMA completion).
148 * POLLOUT is asserted if there is room on the send queue. This does not mean
149 * however, that the next sendmsg() call will succeed. If the application tries
150 * to send to a congested destination, the system call may still fail (and
153 static unsigned int rds_poll(struct file *file, struct socket *sock,
156 struct sock *sk = sock->sk;
157 struct rds_sock *rs = rds_sk_to_rs(sk);
158 unsigned int mask = 0;
161 poll_wait(file, sk->sk_sleep, wait);
163 poll_wait(file, &rds_poll_waitq, wait);
165 read_lock_irqsave(&rs->rs_recv_lock, flags);
166 if (!rs->rs_cong_monitor) {
167 /* When a congestion map was updated, we signal POLLIN for
168 * "historical" reasons. Applications can also poll for
170 if (rds_cong_updated_since(&rs->rs_cong_track))
171 mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
173 spin_lock(&rs->rs_lock);
174 if (rs->rs_cong_notify)
175 mask |= (POLLIN | POLLRDNORM);
176 spin_unlock(&rs->rs_lock);
178 if (!list_empty(&rs->rs_recv_queue)
179 || !list_empty(&rs->rs_notify_queue))
180 mask |= (POLLIN | POLLRDNORM);
181 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
182 mask |= (POLLOUT | POLLWRNORM);
183 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
188 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
193 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
196 struct sockaddr_in sin;
199 /* racing with another thread binding seems ok here */
200 if (rs->rs_bound_addr == 0) {
201 ret = -ENOTCONN; /* XXX not a great errno */
205 if (len < sizeof(struct sockaddr_in)) {
210 if (copy_from_user(&sin, optval, sizeof(sin))) {
215 rds_send_drop_to(rs, &sin);
220 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
225 if (optlen < sizeof(int))
227 if (get_user(value, (int __user *) optval))
233 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
238 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
240 if (rs->rs_cong_monitor) {
241 rds_cong_add_socket(rs);
243 rds_cong_remove_socket(rs);
244 rs->rs_cong_mask = 0;
245 rs->rs_cong_notify = 0;
251 static int rds_setsockopt(struct socket *sock, int level, int optname,
252 char __user *optval, int optlen)
254 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
257 if (level != SOL_RDS) {
263 case RDS_CANCEL_SENT_TO:
264 ret = rds_cancel_sent_to(rs, optval, optlen);
267 ret = rds_get_mr(rs, optval, optlen);
270 ret = rds_free_mr(rs, optval, optlen);
273 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
275 case RDS_CONG_MONITOR:
276 ret = rds_cong_monitor(rs, optval, optlen);
285 static int rds_getsockopt(struct socket *sock, int level, int optname,
286 char __user *optval, int __user *optlen)
288 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
289 int ret = -ENOPROTOOPT, len;
291 if (level != SOL_RDS)
294 if (get_user(len, optlen)) {
300 case RDS_INFO_FIRST ... RDS_INFO_LAST:
301 ret = rds_info_getsockopt(sock, optname, optval,
306 if (len < sizeof(int))
309 if (put_user(rs->rs_recverr, (int __user *) optval)
310 || put_user(sizeof(int), optlen))
324 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
325 int addr_len, int flags)
327 struct sock *sk = sock->sk;
328 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
329 struct rds_sock *rs = rds_sk_to_rs(sk);
334 if (addr_len != sizeof(struct sockaddr_in)) {
339 if (sin->sin_family != AF_INET) {
344 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
349 rs->rs_conn_addr = sin->sin_addr.s_addr;
350 rs->rs_conn_port = sin->sin_port;
357 static struct proto rds_proto = {
359 .owner = THIS_MODULE,
360 .obj_size = sizeof(struct rds_sock),
363 static struct proto_ops rds_proto_ops = {
365 .owner = THIS_MODULE,
366 .release = rds_release,
368 .connect = rds_connect,
369 .socketpair = sock_no_socketpair,
370 .accept = sock_no_accept,
371 .getname = rds_getname,
374 .listen = sock_no_listen,
375 .shutdown = sock_no_shutdown,
376 .setsockopt = rds_setsockopt,
377 .getsockopt = rds_getsockopt,
378 .sendmsg = rds_sendmsg,
379 .recvmsg = rds_recvmsg,
380 .mmap = sock_no_mmap,
381 .sendpage = sock_no_sendpage,
384 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
389 sock_init_data(sock, sk);
390 sock->ops = &rds_proto_ops;
391 sk->sk_protocol = protocol;
393 rs = rds_sk_to_rs(sk);
394 spin_lock_init(&rs->rs_lock);
395 rwlock_init(&rs->rs_recv_lock);
396 INIT_LIST_HEAD(&rs->rs_send_queue);
397 INIT_LIST_HEAD(&rs->rs_recv_queue);
398 INIT_LIST_HEAD(&rs->rs_notify_queue);
399 INIT_LIST_HEAD(&rs->rs_cong_list);
400 spin_lock_init(&rs->rs_rdma_lock);
401 rs->rs_rdma_keys = RB_ROOT;
403 spin_lock_irqsave(&rds_sock_lock, flags);
404 list_add_tail(&rs->rs_item, &rds_sock_list);
406 spin_unlock_irqrestore(&rds_sock_lock, flags);
411 static int rds_create(struct net *net, struct socket *sock, int protocol)
415 if (sock->type != SOCK_SEQPACKET || protocol)
416 return -ESOCKTNOSUPPORT;
418 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto);
422 return __rds_create(sock, sk, protocol);
425 void rds_sock_addref(struct rds_sock *rs)
427 sock_hold(rds_rs_to_sk(rs));
430 void rds_sock_put(struct rds_sock *rs)
432 sock_put(rds_rs_to_sk(rs));
435 static struct net_proto_family rds_family_ops = {
437 .create = rds_create,
438 .owner = THIS_MODULE,
441 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
442 struct rds_info_iterator *iter,
443 struct rds_info_lengths *lens)
447 struct rds_incoming *inc;
449 unsigned int total = 0;
451 len /= sizeof(struct rds_info_message);
453 spin_lock_irqsave(&rds_sock_lock, flags);
455 list_for_each_entry(rs, &rds_sock_list, rs_item) {
456 sk = rds_rs_to_sk(rs);
457 read_lock(&rs->rs_recv_lock);
459 /* XXX too lazy to maintain counts.. */
460 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
463 rds_inc_info_copy(inc, iter, inc->i_saddr,
464 rs->rs_bound_addr, 1);
467 read_unlock(&rs->rs_recv_lock);
470 spin_unlock_irqrestore(&rds_sock_lock, flags);
473 lens->each = sizeof(struct rds_info_message);
476 static void rds_sock_info(struct socket *sock, unsigned int len,
477 struct rds_info_iterator *iter,
478 struct rds_info_lengths *lens)
480 struct rds_info_socket sinfo;
484 len /= sizeof(struct rds_info_socket);
486 spin_lock_irqsave(&rds_sock_lock, flags);
488 if (len < rds_sock_count)
491 list_for_each_entry(rs, &rds_sock_list, rs_item) {
492 sinfo.sndbuf = rds_sk_sndbuf(rs);
493 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
494 sinfo.bound_addr = rs->rs_bound_addr;
495 sinfo.connected_addr = rs->rs_conn_addr;
496 sinfo.bound_port = rs->rs_bound_port;
497 sinfo.connected_port = rs->rs_conn_port;
498 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
500 rds_info_copy(iter, &sinfo, sizeof(sinfo));
504 lens->nr = rds_sock_count;
505 lens->each = sizeof(struct rds_info_socket);
507 spin_unlock_irqrestore(&rds_sock_lock, flags);
510 static void __exit rds_exit(void)
513 sock_unregister(rds_family_ops.family);
514 proto_unregister(&rds_proto);
521 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
522 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
524 module_exit(rds_exit);
526 static int __init rds_init(void)
530 ret = rds_conn_init();
533 ret = rds_threads_init();
536 ret = rds_sysctl_init();
539 ret = rds_stats_init();
542 ret = proto_register(&rds_proto, 1);
545 ret = sock_register(&rds_family_ops);
549 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
550 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
552 /* ib/iwarp transports currently compiled-in */
553 ret = rds_rdma_init();
559 sock_unregister(rds_family_ops.family);
561 proto_unregister(&rds_proto);
575 module_init(rds_init);
577 #define DRV_VERSION "4.0"
578 #define DRV_RELDATE "Feb 12, 2009"
580 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
581 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
582 " v" DRV_VERSION " (" DRV_RELDATE ")");
583 MODULE_VERSION(DRV_VERSION);
584 MODULE_LICENSE("Dual BSD/GPL");
585 MODULE_ALIAS_NETPROTO(PF_RDS);