2 * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/ipv6.h>
39 #include <linux/poll.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;
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);
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);
83 spin_lock_bh(&rds_sock_lock);
84 list_del_init(&rs->rs_item);
86 spin_unlock_bh(&rds_sock_lock);
88 rds_trans_put(rs->rs_transport);
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.
105 void rds_wake_sk_sleep(struct rds_sock *rs)
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);
114 static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
117 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
118 struct sockaddr_in6 *sin6;
119 struct sockaddr_in *sin;
122 /* racey, don't care */
124 if (ipv6_addr_any(&rs->rs_conn_addr))
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);
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);
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.
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;
159 #if IS_ENABLED(CONFIG_IPV6)
160 if (!(ipv6_addr_type(&rs->rs_conn_addr) &
162 sin6 = (struct sockaddr_in6 *)uaddr;
163 memset(sin6, 0, sizeof(*sin6));
164 sin6->sin6_family = AF_INET6;
165 return sizeof(*sin6);
169 sin = (struct sockaddr_in *)uaddr;
170 memset(sin, 0, sizeof(*sin));
171 sin->sin_family = AF_INET;
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);
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);
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.
200 * EPOLLIN is asserted if
201 * - there is data on the receive queue.
202 * - to signal that a previously congested destination may have become
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).
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
212 static __poll_t rds_poll(struct file *file, struct socket *sock,
215 struct sock *sk = sock->sk;
216 struct rds_sock *rs = rds_sk_to_rs(sk);
220 poll_wait(file, sk_sleep(sk), wait);
222 if (rs->rs_seen_congestion)
223 poll_wait(file, &rds_poll_waitq, wait);
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
230 if (rds_cong_updated_since(&rs->rs_cong_track))
231 mask |= (EPOLLIN | EPOLLRDNORM | EPOLLWRBAND);
233 spin_lock(&rs->rs_lock);
234 if (rs->rs_cong_notify)
235 mask |= (EPOLLIN | EPOLLRDNORM);
236 spin_unlock(&rs->rs_lock);
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))
246 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
248 /* clear state any time we wake a seen-congested socket */
250 rs->rs_seen_congestion = 0;
255 static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
260 static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
263 struct sockaddr_in6 sin6;
264 struct sockaddr_in sin;
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 */
273 if (len < sizeof(struct sockaddr_in)) {
276 } else if (len < sizeof(struct sockaddr_in6)) {
278 if (copy_from_user(&sin, optval, sizeof(struct sockaddr_in))) {
282 ipv6_addr_set_v4mapped(sin.sin_addr.s_addr, &sin6.sin6_addr);
283 sin6.sin6_port = sin.sin_port;
285 if (copy_from_user(&sin6, optval,
286 sizeof(struct sockaddr_in6))) {
292 rds_send_drop_to(rs, &sin6);
297 static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
302 if (optlen < sizeof(int))
304 if (get_user(value, (int __user *) optval))
310 static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
315 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
317 if (rs->rs_cong_monitor) {
318 rds_cong_add_socket(rs);
320 rds_cong_remove_socket(rs);
321 rs->rs_cong_mask = 0;
322 rs->rs_cong_notify = 0;
328 static int rds_set_transport(struct rds_sock *rs, char __user *optval,
333 if (rs->rs_transport)
334 return -EOPNOTSUPP; /* previously attached to transport */
336 if (optlen != sizeof(int))
339 if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
342 if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
345 rs->rs_transport = rds_trans_get(t_type);
347 return rs->rs_transport ? 0 : -ENOPROTOOPT;
350 static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
355 if (optlen != sizeof(int))
358 if (get_user(val, (int __user *)optval))
361 valbool = val ? 1 : 0;
364 sock_set_flag(sk, SOCK_RCVTSTAMP);
366 sock_reset_flag(sk, SOCK_RCVTSTAMP);
371 static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval,
374 struct rds_rx_trace_so trace;
377 if (optlen != sizeof(struct rds_rx_trace_so))
380 if (copy_from_user(&trace, optval, sizeof(trace)))
383 if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
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;
392 rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
398 static int rds_setsockopt(struct socket *sock, int level, int optname,
399 char __user *optval, unsigned int optlen)
401 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
404 if (level != SOL_RDS) {
410 case RDS_CANCEL_SENT_TO:
411 ret = rds_cancel_sent_to(rs, optval, optlen);
414 ret = rds_get_mr(rs, optval, optlen);
416 case RDS_GET_MR_FOR_DEST:
417 ret = rds_get_mr_for_dest(rs, optval, optlen);
420 ret = rds_free_mr(rs, optval, optlen);
423 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
425 case RDS_CONG_MONITOR:
426 ret = rds_cong_monitor(rs, optval, optlen);
428 case SO_RDS_TRANSPORT:
430 ret = rds_set_transport(rs, optval, optlen);
431 release_sock(sock->sk);
435 ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
436 release_sock(sock->sk);
438 case SO_RDS_MSG_RXPATH_LATENCY:
439 ret = rds_recv_track_latency(rs, optval, optlen);
448 static int rds_getsockopt(struct socket *sock, int level, int optname,
449 char __user *optval, int __user *optlen)
451 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
452 int ret = -ENOPROTOOPT, len;
455 if (level != SOL_RDS)
458 if (get_user(len, optlen)) {
464 case RDS_INFO_FIRST ... RDS_INFO_LAST:
465 ret = rds_info_getsockopt(sock, optname, optval,
470 if (len < sizeof(int))
473 if (put_user(rs->rs_recverr, (int __user *) optval) ||
474 put_user(sizeof(int), optlen))
479 case SO_RDS_TRANSPORT:
480 if (len < sizeof(int)) {
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))
501 static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
502 int addr_len, int flags)
504 struct sock *sk = sock->sk;
505 struct sockaddr_in *sin;
506 struct rds_sock *rs = rds_sk_to_rs(sk);
509 if (addr_len < offsetofend(struct sockaddr, sa_family))
514 switch (uaddr->sa_family) {
516 sin = (struct sockaddr_in *)uaddr;
517 if (addr_len < sizeof(struct sockaddr_in)) {
521 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
525 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) ||
526 sin->sin_addr.s_addr == htonl(INADDR_BROADCAST)) {
530 ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &rs->rs_conn_addr);
531 rs->rs_conn_port = sin->sin_port;
534 #if IS_ENABLED(CONFIG_IPV6)
536 struct sockaddr_in6 *sin6;
539 sin6 = (struct sockaddr_in6 *)uaddr;
540 if (addr_len < sizeof(struct sockaddr_in6)) {
544 addr_type = ipv6_addr_type(&sin6->sin6_addr);
545 if (!(addr_type & IPV6_ADDR_UNICAST)) {
548 if (!(addr_type & IPV6_ADDR_MAPPED)) {
553 /* It is a mapped address. Need to do some sanity
556 addr4 = sin6->sin6_addr.s6_addr32[3];
557 if (addr4 == htonl(INADDR_ANY) ||
558 addr4 == htonl(INADDR_BROADCAST) ||
559 IN_MULTICAST(ntohl(addr4))) {
565 if (addr_type & IPV6_ADDR_LINKLOCAL) {
566 /* If socket is arleady bound to a link local address,
567 * the peer address must be on the same link.
569 if (sin6->sin6_scope_id == 0 ||
570 (!ipv6_addr_any(&rs->rs_bound_addr) &&
571 rs->rs_bound_scope_id &&
572 sin6->sin6_scope_id != rs->rs_bound_scope_id)) {
576 /* Remember the connected address scope ID. It will
577 * be checked against the binding local address when
578 * the socket is bound.
580 rs->rs_bound_scope_id = sin6->sin6_scope_id;
582 rs->rs_conn_addr = sin6->sin6_addr;
583 rs->rs_conn_port = sin6->sin6_port;
597 static struct proto rds_proto = {
599 .owner = THIS_MODULE,
600 .obj_size = sizeof(struct rds_sock),
603 static const struct proto_ops rds_proto_ops = {
605 .owner = THIS_MODULE,
606 .release = rds_release,
608 .connect = rds_connect,
609 .socketpair = sock_no_socketpair,
610 .accept = sock_no_accept,
611 .getname = rds_getname,
614 .listen = sock_no_listen,
615 .shutdown = sock_no_shutdown,
616 .setsockopt = rds_setsockopt,
617 .getsockopt = rds_getsockopt,
618 .sendmsg = rds_sendmsg,
619 .recvmsg = rds_recvmsg,
620 .mmap = sock_no_mmap,
621 .sendpage = sock_no_sendpage,
624 static void rds_sock_destruct(struct sock *sk)
626 struct rds_sock *rs = rds_sk_to_rs(sk);
628 WARN_ON((&rs->rs_item != rs->rs_item.next ||
629 &rs->rs_item != rs->rs_item.prev));
632 static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
636 sock_init_data(sock, sk);
637 sock->ops = &rds_proto_ops;
638 sk->sk_protocol = protocol;
639 sk->sk_destruct = rds_sock_destruct;
641 rs = rds_sk_to_rs(sk);
642 spin_lock_init(&rs->rs_lock);
643 rwlock_init(&rs->rs_recv_lock);
644 INIT_LIST_HEAD(&rs->rs_send_queue);
645 INIT_LIST_HEAD(&rs->rs_recv_queue);
646 INIT_LIST_HEAD(&rs->rs_notify_queue);
647 INIT_LIST_HEAD(&rs->rs_cong_list);
648 rds_message_zcopy_queue_init(&rs->rs_zcookie_queue);
649 spin_lock_init(&rs->rs_rdma_lock);
650 rs->rs_rdma_keys = RB_ROOT;
651 rs->rs_rx_traces = 0;
653 spin_lock_bh(&rds_sock_lock);
654 list_add_tail(&rs->rs_item, &rds_sock_list);
656 spin_unlock_bh(&rds_sock_lock);
661 static int rds_create(struct net *net, struct socket *sock, int protocol,
666 if (sock->type != SOCK_SEQPACKET || protocol)
667 return -ESOCKTNOSUPPORT;
669 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
673 return __rds_create(sock, sk, protocol);
676 void rds_sock_addref(struct rds_sock *rs)
678 sock_hold(rds_rs_to_sk(rs));
681 void rds_sock_put(struct rds_sock *rs)
683 sock_put(rds_rs_to_sk(rs));
686 static const struct net_proto_family rds_family_ops = {
688 .create = rds_create,
689 .owner = THIS_MODULE,
692 static void rds_sock_inc_info(struct socket *sock, unsigned int len,
693 struct rds_info_iterator *iter,
694 struct rds_info_lengths *lens)
697 struct rds_incoming *inc;
698 unsigned int total = 0;
700 len /= sizeof(struct rds_info_message);
702 spin_lock_bh(&rds_sock_lock);
704 list_for_each_entry(rs, &rds_sock_list, rs_item) {
705 read_lock(&rs->rs_recv_lock);
707 /* XXX too lazy to maintain counts.. */
708 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
711 rds_inc_info_copy(inc, iter,
712 inc->i_saddr.s6_addr32[3],
713 rs->rs_bound_addr_v4,
717 read_unlock(&rs->rs_recv_lock);
720 spin_unlock_bh(&rds_sock_lock);
723 lens->each = sizeof(struct rds_info_message);
726 static void rds_sock_info(struct socket *sock, unsigned int len,
727 struct rds_info_iterator *iter,
728 struct rds_info_lengths *lens)
730 struct rds_info_socket sinfo;
733 len /= sizeof(struct rds_info_socket);
735 spin_lock_bh(&rds_sock_lock);
737 if (len < rds_sock_count)
740 list_for_each_entry(rs, &rds_sock_list, rs_item) {
741 sinfo.sndbuf = rds_sk_sndbuf(rs);
742 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
743 sinfo.bound_addr = rs->rs_bound_addr_v4;
744 sinfo.connected_addr = rs->rs_conn_addr_v4;
745 sinfo.bound_port = rs->rs_bound_port;
746 sinfo.connected_port = rs->rs_conn_port;
747 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
749 rds_info_copy(iter, &sinfo, sizeof(sinfo));
753 lens->nr = rds_sock_count;
754 lens->each = sizeof(struct rds_info_socket);
756 spin_unlock_bh(&rds_sock_lock);
759 static void rds_exit(void)
761 sock_unregister(rds_family_ops.family);
762 proto_unregister(&rds_proto);
769 rds_bind_lock_destroy();
770 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
771 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
773 module_exit(rds_exit);
777 static int rds_init(void)
781 net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
783 ret = rds_bind_lock_init();
787 ret = rds_conn_init();
791 ret = rds_threads_init();
794 ret = rds_sysctl_init();
797 ret = rds_stats_init();
800 ret = proto_register(&rds_proto, 1);
803 ret = sock_register(&rds_family_ops);
807 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
808 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
813 proto_unregister(&rds_proto);
825 rds_bind_lock_destroy();
829 module_init(rds_init);
831 #define DRV_VERSION "4.0"
832 #define DRV_RELDATE "Feb 12, 2009"
834 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
835 MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
836 " v" DRV_VERSION " (" DRV_RELDATE ")");
837 MODULE_VERSION(DRV_VERSION);
838 MODULE_LICENSE("Dual BSD/GPL");
839 MODULE_ALIAS_NETPROTO(PF_RDS);