Loading...
1/*
2 * Copyright (c) 2006, 2019 Oracle and/or its affiliates. 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/ipv6.h>
39#include <linux/poll.h>
40#include <net/sock.h>
41
42#include "rds.h"
43
44/* this is just used for stats gathering :/ */
45static DEFINE_SPINLOCK(rds_sock_lock);
46static unsigned long rds_sock_count;
47static LIST_HEAD(rds_sock_list);
48DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
49
50/*
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.
54 *
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.
58 */
59static int rds_release(struct socket *sock)
60{
61 struct sock *sk = sock->sk;
62 struct rds_sock *rs;
63
64 if (!sk)
65 goto out;
66
67 rs = rds_sk_to_rs(sk);
68
69 sock_orphan(sk);
70 /* Note - rds_clear_recv_queue grabs rs_recv_lock, so
71 * that ensures the recv path has completed messing
72 * with the socket. */
73 rds_clear_recv_queue(rs);
74 rds_cong_remove_socket(rs);
75
76 rds_remove_bound(rs);
77
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);
82
83 spin_lock_bh(&rds_sock_lock);
84 list_del_init(&rs->rs_item);
85 rds_sock_count--;
86 spin_unlock_bh(&rds_sock_lock);
87
88 rds_trans_put(rs->rs_transport);
89
90 sock->sk = NULL;
91 sock_put(sk);
92out:
93 return 0;
94}
95
96/*
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.
104 */
105void rds_wake_sk_sleep(struct rds_sock *rs)
106{
107 unsigned long flags;
108
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);
112}
113
114static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
115 int peer)
116{
117 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
118 struct sockaddr_in6 *sin6;
119 struct sockaddr_in *sin;
120 int uaddr_len;
121
122 /* racey, don't care */
123 if (peer) {
124 if (ipv6_addr_any(&rs->rs_conn_addr))
125 return -ENOTCONN;
126
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);
134 } else {
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);
143 }
144 } else {
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.
150 */
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;
156 return sizeof(*sin);
157 }
158
159#if IS_ENABLED(CONFIG_IPV6)
160 if (!(ipv6_addr_type(&rs->rs_conn_addr) &
161 IPV6_ADDR_MAPPED)) {
162 sin6 = (struct sockaddr_in6 *)uaddr;
163 memset(sin6, 0, sizeof(*sin6));
164 sin6->sin6_family = AF_INET6;
165 return sizeof(*sin6);
166 }
167#endif
168
169 sin = (struct sockaddr_in *)uaddr;
170 memset(sin, 0, sizeof(*sin));
171 sin->sin_family = AF_INET;
172 return sizeof(*sin);
173 }
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);
181 } else {
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);
189 }
190 }
191
192 return uaddr_len;
193}
194
195/*
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.
199 *
200 * EPOLLIN is asserted if
201 * - there is data on the receive queue.
202 * - to signal that a previously congested destination may have become
203 * uncongested
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).
206 *
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
210 * return ENOBUFS).
211 */
212static __poll_t rds_poll(struct file *file, struct socket *sock,
213 poll_table *wait)
214{
215 struct sock *sk = sock->sk;
216 struct rds_sock *rs = rds_sk_to_rs(sk);
217 __poll_t mask = 0;
218 unsigned long flags;
219
220 poll_wait(file, sk_sleep(sk), wait);
221
222 if (rs->rs_seen_congestion)
223 poll_wait(file, &rds_poll_waitq, wait);
224
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
229 * WRBAND instead. */
230 if (rds_cong_updated_since(&rs->rs_cong_track))
231 mask |= (EPOLLIN | EPOLLRDNORM | EPOLLWRBAND);
232 } else {
233 spin_lock(&rs->rs_lock);
234 if (rs->rs_cong_notify)
235 mask |= (EPOLLIN | EPOLLRDNORM);
236 spin_unlock(&rs->rs_lock);
237 }
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))
245 mask |= POLLERR;
246 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
247
248 /* clear state any time we wake a seen-congested socket */
249 if (mask)
250 rs->rs_seen_congestion = 0;
251
252 return mask;
253}
254
255static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
256{
257 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
258 rds_tos_t utos, tos = 0;
259
260 switch (cmd) {
261 case SIOCRDSSETTOS:
262 if (get_user(utos, (rds_tos_t __user *)arg))
263 return -EFAULT;
264
265 if (rs->rs_transport &&
266 rs->rs_transport->get_tos_map)
267 tos = rs->rs_transport->get_tos_map(utos);
268 else
269 return -ENOIOCTLCMD;
270
271 spin_lock_bh(&rds_sock_lock);
272 if (rs->rs_tos || rs->rs_conn) {
273 spin_unlock_bh(&rds_sock_lock);
274 return -EINVAL;
275 }
276 rs->rs_tos = tos;
277 spin_unlock_bh(&rds_sock_lock);
278 break;
279 case SIOCRDSGETTOS:
280 spin_lock_bh(&rds_sock_lock);
281 tos = rs->rs_tos;
282 spin_unlock_bh(&rds_sock_lock);
283 if (put_user(tos, (rds_tos_t __user *)arg))
284 return -EFAULT;
285 break;
286 default:
287 return -ENOIOCTLCMD;
288 }
289
290 return 0;
291}
292
293static int rds_cancel_sent_to(struct rds_sock *rs, sockptr_t optval, int len)
294{
295 struct sockaddr_in6 sin6;
296 struct sockaddr_in sin;
297 int ret = 0;
298
299 /* racing with another thread binding seems ok here */
300 if (ipv6_addr_any(&rs->rs_bound_addr)) {
301 ret = -ENOTCONN; /* XXX not a great errno */
302 goto out;
303 }
304
305 if (len < sizeof(struct sockaddr_in)) {
306 ret = -EINVAL;
307 goto out;
308 } else if (len < sizeof(struct sockaddr_in6)) {
309 /* Assume IPv4 */
310 if (copy_from_sockptr(&sin, optval,
311 sizeof(struct sockaddr_in))) {
312 ret = -EFAULT;
313 goto out;
314 }
315 ipv6_addr_set_v4mapped(sin.sin_addr.s_addr, &sin6.sin6_addr);
316 sin6.sin6_port = sin.sin_port;
317 } else {
318 if (copy_from_sockptr(&sin6, optval,
319 sizeof(struct sockaddr_in6))) {
320 ret = -EFAULT;
321 goto out;
322 }
323 }
324
325 rds_send_drop_to(rs, &sin6);
326out:
327 return ret;
328}
329
330static int rds_set_bool_option(unsigned char *optvar, sockptr_t optval,
331 int optlen)
332{
333 int value;
334
335 if (optlen < sizeof(int))
336 return -EINVAL;
337 if (copy_from_sockptr(&value, optval, sizeof(int)))
338 return -EFAULT;
339 *optvar = !!value;
340 return 0;
341}
342
343static int rds_cong_monitor(struct rds_sock *rs, sockptr_t optval, int optlen)
344{
345 int ret;
346
347 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
348 if (ret == 0) {
349 if (rs->rs_cong_monitor) {
350 rds_cong_add_socket(rs);
351 } else {
352 rds_cong_remove_socket(rs);
353 rs->rs_cong_mask = 0;
354 rs->rs_cong_notify = 0;
355 }
356 }
357 return ret;
358}
359
360static int rds_set_transport(struct rds_sock *rs, sockptr_t optval, int optlen)
361{
362 int t_type;
363
364 if (rs->rs_transport)
365 return -EOPNOTSUPP; /* previously attached to transport */
366
367 if (optlen != sizeof(int))
368 return -EINVAL;
369
370 if (copy_from_sockptr(&t_type, optval, sizeof(t_type)))
371 return -EFAULT;
372
373 if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
374 return -EINVAL;
375
376 rs->rs_transport = rds_trans_get(t_type);
377
378 return rs->rs_transport ? 0 : -ENOPROTOOPT;
379}
380
381static int rds_enable_recvtstamp(struct sock *sk, sockptr_t optval,
382 int optlen, int optname)
383{
384 int val, valbool;
385
386 if (optlen != sizeof(int))
387 return -EFAULT;
388
389 if (copy_from_sockptr(&val, optval, sizeof(int)))
390 return -EFAULT;
391
392 valbool = val ? 1 : 0;
393
394 if (optname == SO_TIMESTAMP_NEW)
395 sock_set_flag(sk, SOCK_TSTAMP_NEW);
396
397 if (valbool)
398 sock_set_flag(sk, SOCK_RCVTSTAMP);
399 else
400 sock_reset_flag(sk, SOCK_RCVTSTAMP);
401
402 return 0;
403}
404
405static int rds_recv_track_latency(struct rds_sock *rs, sockptr_t optval,
406 int optlen)
407{
408 struct rds_rx_trace_so trace;
409 int i;
410
411 if (optlen != sizeof(struct rds_rx_trace_so))
412 return -EFAULT;
413
414 if (copy_from_sockptr(&trace, optval, sizeof(trace)))
415 return -EFAULT;
416
417 if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
418 return -EFAULT;
419
420 rs->rs_rx_traces = trace.rx_traces;
421 for (i = 0; i < rs->rs_rx_traces; i++) {
422 if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
423 rs->rs_rx_traces = 0;
424 return -EFAULT;
425 }
426 rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
427 }
428
429 return 0;
430}
431
432static int rds_setsockopt(struct socket *sock, int level, int optname,
433 sockptr_t optval, unsigned int optlen)
434{
435 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
436 int ret;
437
438 if (level != SOL_RDS) {
439 ret = -ENOPROTOOPT;
440 goto out;
441 }
442
443 switch (optname) {
444 case RDS_CANCEL_SENT_TO:
445 ret = rds_cancel_sent_to(rs, optval, optlen);
446 break;
447 case RDS_GET_MR:
448 ret = rds_get_mr(rs, optval, optlen);
449 break;
450 case RDS_GET_MR_FOR_DEST:
451 ret = rds_get_mr_for_dest(rs, optval, optlen);
452 break;
453 case RDS_FREE_MR:
454 ret = rds_free_mr(rs, optval, optlen);
455 break;
456 case RDS_RECVERR:
457 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
458 break;
459 case RDS_CONG_MONITOR:
460 ret = rds_cong_monitor(rs, optval, optlen);
461 break;
462 case SO_RDS_TRANSPORT:
463 lock_sock(sock->sk);
464 ret = rds_set_transport(rs, optval, optlen);
465 release_sock(sock->sk);
466 break;
467 case SO_TIMESTAMP_OLD:
468 case SO_TIMESTAMP_NEW:
469 lock_sock(sock->sk);
470 ret = rds_enable_recvtstamp(sock->sk, optval, optlen, optname);
471 release_sock(sock->sk);
472 break;
473 case SO_RDS_MSG_RXPATH_LATENCY:
474 ret = rds_recv_track_latency(rs, optval, optlen);
475 break;
476 default:
477 ret = -ENOPROTOOPT;
478 }
479out:
480 return ret;
481}
482
483static int rds_getsockopt(struct socket *sock, int level, int optname,
484 char __user *optval, int __user *optlen)
485{
486 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
487 int ret = -ENOPROTOOPT, len;
488 int trans;
489
490 if (level != SOL_RDS)
491 goto out;
492
493 if (get_user(len, optlen)) {
494 ret = -EFAULT;
495 goto out;
496 }
497
498 switch (optname) {
499 case RDS_INFO_FIRST ... RDS_INFO_LAST:
500 ret = rds_info_getsockopt(sock, optname, optval,
501 optlen);
502 break;
503
504 case RDS_RECVERR:
505 if (len < sizeof(int))
506 ret = -EINVAL;
507 else
508 if (put_user(rs->rs_recverr, (int __user *) optval) ||
509 put_user(sizeof(int), optlen))
510 ret = -EFAULT;
511 else
512 ret = 0;
513 break;
514 case SO_RDS_TRANSPORT:
515 if (len < sizeof(int)) {
516 ret = -EINVAL;
517 break;
518 }
519 trans = (rs->rs_transport ? rs->rs_transport->t_type :
520 RDS_TRANS_NONE); /* unbound */
521 if (put_user(trans, (int __user *)optval) ||
522 put_user(sizeof(int), optlen))
523 ret = -EFAULT;
524 else
525 ret = 0;
526 break;
527 default:
528 break;
529 }
530
531out:
532 return ret;
533
534}
535
536static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
537 int addr_len, int flags)
538{
539 struct sock *sk = sock->sk;
540 struct sockaddr_in *sin;
541 struct rds_sock *rs = rds_sk_to_rs(sk);
542 int ret = 0;
543
544 if (addr_len < offsetofend(struct sockaddr, sa_family))
545 return -EINVAL;
546
547 lock_sock(sk);
548
549 switch (uaddr->sa_family) {
550 case AF_INET:
551 sin = (struct sockaddr_in *)uaddr;
552 if (addr_len < sizeof(struct sockaddr_in)) {
553 ret = -EINVAL;
554 break;
555 }
556 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
557 ret = -EDESTADDRREQ;
558 break;
559 }
560 if (ipv4_is_multicast(sin->sin_addr.s_addr) ||
561 sin->sin_addr.s_addr == htonl(INADDR_BROADCAST)) {
562 ret = -EINVAL;
563 break;
564 }
565 ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &rs->rs_conn_addr);
566 rs->rs_conn_port = sin->sin_port;
567 break;
568
569#if IS_ENABLED(CONFIG_IPV6)
570 case AF_INET6: {
571 struct sockaddr_in6 *sin6;
572 int addr_type;
573
574 sin6 = (struct sockaddr_in6 *)uaddr;
575 if (addr_len < sizeof(struct sockaddr_in6)) {
576 ret = -EINVAL;
577 break;
578 }
579 addr_type = ipv6_addr_type(&sin6->sin6_addr);
580 if (!(addr_type & IPV6_ADDR_UNICAST)) {
581 __be32 addr4;
582
583 if (!(addr_type & IPV6_ADDR_MAPPED)) {
584 ret = -EPROTOTYPE;
585 break;
586 }
587
588 /* It is a mapped address. Need to do some sanity
589 * checks.
590 */
591 addr4 = sin6->sin6_addr.s6_addr32[3];
592 if (addr4 == htonl(INADDR_ANY) ||
593 addr4 == htonl(INADDR_BROADCAST) ||
594 ipv4_is_multicast(addr4)) {
595 ret = -EPROTOTYPE;
596 break;
597 }
598 }
599
600 if (addr_type & IPV6_ADDR_LINKLOCAL) {
601 /* If socket is arleady bound to a link local address,
602 * the peer address must be on the same link.
603 */
604 if (sin6->sin6_scope_id == 0 ||
605 (!ipv6_addr_any(&rs->rs_bound_addr) &&
606 rs->rs_bound_scope_id &&
607 sin6->sin6_scope_id != rs->rs_bound_scope_id)) {
608 ret = -EINVAL;
609 break;
610 }
611 /* Remember the connected address scope ID. It will
612 * be checked against the binding local address when
613 * the socket is bound.
614 */
615 rs->rs_bound_scope_id = sin6->sin6_scope_id;
616 }
617 rs->rs_conn_addr = sin6->sin6_addr;
618 rs->rs_conn_port = sin6->sin6_port;
619 break;
620 }
621#endif
622
623 default:
624 ret = -EAFNOSUPPORT;
625 break;
626 }
627
628 release_sock(sk);
629 return ret;
630}
631
632static struct proto rds_proto = {
633 .name = "RDS",
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct rds_sock),
636};
637
638static const struct proto_ops rds_proto_ops = {
639 .family = AF_RDS,
640 .owner = THIS_MODULE,
641 .release = rds_release,
642 .bind = rds_bind,
643 .connect = rds_connect,
644 .socketpair = sock_no_socketpair,
645 .accept = sock_no_accept,
646 .getname = rds_getname,
647 .poll = rds_poll,
648 .ioctl = rds_ioctl,
649 .listen = sock_no_listen,
650 .shutdown = sock_no_shutdown,
651 .setsockopt = rds_setsockopt,
652 .getsockopt = rds_getsockopt,
653 .sendmsg = rds_sendmsg,
654 .recvmsg = rds_recvmsg,
655 .mmap = sock_no_mmap,
656 .sendpage = sock_no_sendpage,
657};
658
659static void rds_sock_destruct(struct sock *sk)
660{
661 struct rds_sock *rs = rds_sk_to_rs(sk);
662
663 WARN_ON((&rs->rs_item != rs->rs_item.next ||
664 &rs->rs_item != rs->rs_item.prev));
665}
666
667static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
668{
669 struct rds_sock *rs;
670
671 sock_init_data(sock, sk);
672 sock->ops = &rds_proto_ops;
673 sk->sk_protocol = protocol;
674 sk->sk_destruct = rds_sock_destruct;
675
676 rs = rds_sk_to_rs(sk);
677 spin_lock_init(&rs->rs_lock);
678 rwlock_init(&rs->rs_recv_lock);
679 INIT_LIST_HEAD(&rs->rs_send_queue);
680 INIT_LIST_HEAD(&rs->rs_recv_queue);
681 INIT_LIST_HEAD(&rs->rs_notify_queue);
682 INIT_LIST_HEAD(&rs->rs_cong_list);
683 rds_message_zcopy_queue_init(&rs->rs_zcookie_queue);
684 spin_lock_init(&rs->rs_rdma_lock);
685 rs->rs_rdma_keys = RB_ROOT;
686 rs->rs_rx_traces = 0;
687 rs->rs_tos = 0;
688 rs->rs_conn = NULL;
689
690 spin_lock_bh(&rds_sock_lock);
691 list_add_tail(&rs->rs_item, &rds_sock_list);
692 rds_sock_count++;
693 spin_unlock_bh(&rds_sock_lock);
694
695 return 0;
696}
697
698static int rds_create(struct net *net, struct socket *sock, int protocol,
699 int kern)
700{
701 struct sock *sk;
702
703 if (sock->type != SOCK_SEQPACKET || protocol)
704 return -ESOCKTNOSUPPORT;
705
706 sk = sk_alloc(net, AF_RDS, GFP_KERNEL, &rds_proto, kern);
707 if (!sk)
708 return -ENOMEM;
709
710 return __rds_create(sock, sk, protocol);
711}
712
713void rds_sock_addref(struct rds_sock *rs)
714{
715 sock_hold(rds_rs_to_sk(rs));
716}
717
718void rds_sock_put(struct rds_sock *rs)
719{
720 sock_put(rds_rs_to_sk(rs));
721}
722
723static const struct net_proto_family rds_family_ops = {
724 .family = AF_RDS,
725 .create = rds_create,
726 .owner = THIS_MODULE,
727};
728
729static void rds_sock_inc_info(struct socket *sock, unsigned int len,
730 struct rds_info_iterator *iter,
731 struct rds_info_lengths *lens)
732{
733 struct rds_sock *rs;
734 struct rds_incoming *inc;
735 unsigned int total = 0;
736
737 len /= sizeof(struct rds_info_message);
738
739 spin_lock_bh(&rds_sock_lock);
740
741 list_for_each_entry(rs, &rds_sock_list, rs_item) {
742 /* This option only supports IPv4 sockets. */
743 if (!ipv6_addr_v4mapped(&rs->rs_bound_addr))
744 continue;
745
746 read_lock(&rs->rs_recv_lock);
747
748 /* XXX too lazy to maintain counts.. */
749 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
750 total++;
751 if (total <= len)
752 rds_inc_info_copy(inc, iter,
753 inc->i_saddr.s6_addr32[3],
754 rs->rs_bound_addr_v4,
755 1);
756 }
757
758 read_unlock(&rs->rs_recv_lock);
759 }
760
761 spin_unlock_bh(&rds_sock_lock);
762
763 lens->nr = total;
764 lens->each = sizeof(struct rds_info_message);
765}
766
767#if IS_ENABLED(CONFIG_IPV6)
768static void rds6_sock_inc_info(struct socket *sock, unsigned int len,
769 struct rds_info_iterator *iter,
770 struct rds_info_lengths *lens)
771{
772 struct rds_incoming *inc;
773 unsigned int total = 0;
774 struct rds_sock *rs;
775
776 len /= sizeof(struct rds6_info_message);
777
778 spin_lock_bh(&rds_sock_lock);
779
780 list_for_each_entry(rs, &rds_sock_list, rs_item) {
781 read_lock(&rs->rs_recv_lock);
782
783 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
784 total++;
785 if (total <= len)
786 rds6_inc_info_copy(inc, iter, &inc->i_saddr,
787 &rs->rs_bound_addr, 1);
788 }
789
790 read_unlock(&rs->rs_recv_lock);
791 }
792
793 spin_unlock_bh(&rds_sock_lock);
794
795 lens->nr = total;
796 lens->each = sizeof(struct rds6_info_message);
797}
798#endif
799
800static void rds_sock_info(struct socket *sock, unsigned int len,
801 struct rds_info_iterator *iter,
802 struct rds_info_lengths *lens)
803{
804 struct rds_info_socket sinfo;
805 unsigned int cnt = 0;
806 struct rds_sock *rs;
807
808 len /= sizeof(struct rds_info_socket);
809
810 spin_lock_bh(&rds_sock_lock);
811
812 if (len < rds_sock_count) {
813 cnt = rds_sock_count;
814 goto out;
815 }
816
817 list_for_each_entry(rs, &rds_sock_list, rs_item) {
818 /* This option only supports IPv4 sockets. */
819 if (!ipv6_addr_v4mapped(&rs->rs_bound_addr))
820 continue;
821 sinfo.sndbuf = rds_sk_sndbuf(rs);
822 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
823 sinfo.bound_addr = rs->rs_bound_addr_v4;
824 sinfo.connected_addr = rs->rs_conn_addr_v4;
825 sinfo.bound_port = rs->rs_bound_port;
826 sinfo.connected_port = rs->rs_conn_port;
827 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
828
829 rds_info_copy(iter, &sinfo, sizeof(sinfo));
830 cnt++;
831 }
832
833out:
834 lens->nr = cnt;
835 lens->each = sizeof(struct rds_info_socket);
836
837 spin_unlock_bh(&rds_sock_lock);
838}
839
840#if IS_ENABLED(CONFIG_IPV6)
841static void rds6_sock_info(struct socket *sock, unsigned int len,
842 struct rds_info_iterator *iter,
843 struct rds_info_lengths *lens)
844{
845 struct rds6_info_socket sinfo6;
846 struct rds_sock *rs;
847
848 len /= sizeof(struct rds6_info_socket);
849
850 spin_lock_bh(&rds_sock_lock);
851
852 if (len < rds_sock_count)
853 goto out;
854
855 list_for_each_entry(rs, &rds_sock_list, rs_item) {
856 sinfo6.sndbuf = rds_sk_sndbuf(rs);
857 sinfo6.rcvbuf = rds_sk_rcvbuf(rs);
858 sinfo6.bound_addr = rs->rs_bound_addr;
859 sinfo6.connected_addr = rs->rs_conn_addr;
860 sinfo6.bound_port = rs->rs_bound_port;
861 sinfo6.connected_port = rs->rs_conn_port;
862 sinfo6.inum = sock_i_ino(rds_rs_to_sk(rs));
863
864 rds_info_copy(iter, &sinfo6, sizeof(sinfo6));
865 }
866
867 out:
868 lens->nr = rds_sock_count;
869 lens->each = sizeof(struct rds6_info_socket);
870
871 spin_unlock_bh(&rds_sock_lock);
872}
873#endif
874
875static void rds_exit(void)
876{
877 sock_unregister(rds_family_ops.family);
878 proto_unregister(&rds_proto);
879 rds_conn_exit();
880 rds_cong_exit();
881 rds_sysctl_exit();
882 rds_threads_exit();
883 rds_stats_exit();
884 rds_page_exit();
885 rds_bind_lock_destroy();
886 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
887 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
888#if IS_ENABLED(CONFIG_IPV6)
889 rds_info_deregister_func(RDS6_INFO_SOCKETS, rds6_sock_info);
890 rds_info_deregister_func(RDS6_INFO_RECV_MESSAGES, rds6_sock_inc_info);
891#endif
892}
893module_exit(rds_exit);
894
895u32 rds_gen_num;
896
897static int rds_init(void)
898{
899 int ret;
900
901 net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
902
903 ret = rds_bind_lock_init();
904 if (ret)
905 goto out;
906
907 ret = rds_conn_init();
908 if (ret)
909 goto out_bind;
910
911 ret = rds_threads_init();
912 if (ret)
913 goto out_conn;
914 ret = rds_sysctl_init();
915 if (ret)
916 goto out_threads;
917 ret = rds_stats_init();
918 if (ret)
919 goto out_sysctl;
920 ret = proto_register(&rds_proto, 1);
921 if (ret)
922 goto out_stats;
923 ret = sock_register(&rds_family_ops);
924 if (ret)
925 goto out_proto;
926
927 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
928 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
929#if IS_ENABLED(CONFIG_IPV6)
930 rds_info_register_func(RDS6_INFO_SOCKETS, rds6_sock_info);
931 rds_info_register_func(RDS6_INFO_RECV_MESSAGES, rds6_sock_inc_info);
932#endif
933
934 goto out;
935
936out_proto:
937 proto_unregister(&rds_proto);
938out_stats:
939 rds_stats_exit();
940out_sysctl:
941 rds_sysctl_exit();
942out_threads:
943 rds_threads_exit();
944out_conn:
945 rds_conn_exit();
946 rds_cong_exit();
947 rds_page_exit();
948out_bind:
949 rds_bind_lock_destroy();
950out:
951 return ret;
952}
953module_init(rds_init);
954
955#define DRV_VERSION "4.0"
956#define DRV_RELDATE "Feb 12, 2009"
957
958MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
959MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
960 " v" DRV_VERSION " (" DRV_RELDATE ")");
961MODULE_VERSION(DRV_VERSION);
962MODULE_LICENSE("Dual BSD/GPL");
963MODULE_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
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 rds_notify_msg_zcopy_purge(&rs->rs_zcookie_queue);
81
82 spin_lock_bh(&rds_sock_lock);
83 list_del_init(&rs->rs_item);
84 rds_sock_count--;
85 spin_unlock_bh(&rds_sock_lock);
86
87 rds_trans_put(rs->rs_transport);
88
89 sock->sk = NULL;
90 sock_put(sk);
91out:
92 return 0;
93}
94
95/*
96 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
97 * _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
98 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
99 * this seems more conservative.
100 * NB - normally, one would use sk_callback_lock for this, but we can
101 * get here from interrupts, whereas the network code grabs sk_callback_lock
102 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
103 */
104void rds_wake_sk_sleep(struct rds_sock *rs)
105{
106 unsigned long flags;
107
108 read_lock_irqsave(&rs->rs_recv_lock, flags);
109 __rds_wake_sk_sleep(rds_rs_to_sk(rs));
110 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
111}
112
113static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
114 int peer)
115{
116 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
117 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
118
119 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
120
121 /* racey, don't care */
122 if (peer) {
123 if (!rs->rs_conn_addr)
124 return -ENOTCONN;
125
126 sin->sin_port = rs->rs_conn_port;
127 sin->sin_addr.s_addr = rs->rs_conn_addr;
128 } else {
129 sin->sin_port = rs->rs_bound_port;
130 sin->sin_addr.s_addr = rs->rs_bound_addr;
131 }
132
133 sin->sin_family = AF_INET;
134
135 return sizeof(*sin);
136}
137
138/*
139 * RDS' poll is without a doubt the least intuitive part of the interface,
140 * as EPOLLIN and EPOLLOUT do not behave entirely as you would expect from
141 * a network protocol.
142 *
143 * EPOLLIN 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, or a MSG_ZEROCOPY completion).
149 *
150 * EPOLLOUT 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 __poll_t 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 __poll_t 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 EPOLLIN for
171 * "historical" reasons. Applications can also poll for
172 * WRBAND instead. */
173 if (rds_cong_updated_since(&rs->rs_cong_track))
174 mask |= (EPOLLIN | EPOLLRDNORM | EPOLLWRBAND);
175 } else {
176 spin_lock(&rs->rs_lock);
177 if (rs->rs_cong_notify)
178 mask |= (EPOLLIN | EPOLLRDNORM);
179 spin_unlock(&rs->rs_lock);
180 }
181 if (!list_empty(&rs->rs_recv_queue) ||
182 !list_empty(&rs->rs_notify_queue) ||
183 !list_empty(&rs->rs_zcookie_queue.zcookie_head))
184 mask |= (EPOLLIN | EPOLLRDNORM);
185 if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
186 mask |= (EPOLLOUT | EPOLLWRNORM);
187 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
188 mask |= POLLERR;
189 read_unlock_irqrestore(&rs->rs_recv_lock, flags);
190
191 /* clear state any time we wake a seen-congested socket */
192 if (mask)
193 rs->rs_seen_congestion = 0;
194
195 return mask;
196}
197
198static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
199{
200 return -ENOIOCTLCMD;
201}
202
203static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
204 int len)
205{
206 struct sockaddr_in sin;
207 int ret = 0;
208
209 /* racing with another thread binding seems ok here */
210 if (rs->rs_bound_addr == 0) {
211 ret = -ENOTCONN; /* XXX not a great errno */
212 goto out;
213 }
214
215 if (len < sizeof(struct sockaddr_in)) {
216 ret = -EINVAL;
217 goto out;
218 }
219
220 if (copy_from_user(&sin, optval, sizeof(sin))) {
221 ret = -EFAULT;
222 goto out;
223 }
224
225 rds_send_drop_to(rs, &sin);
226out:
227 return ret;
228}
229
230static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
231 int optlen)
232{
233 int value;
234
235 if (optlen < sizeof(int))
236 return -EINVAL;
237 if (get_user(value, (int __user *) optval))
238 return -EFAULT;
239 *optvar = !!value;
240 return 0;
241}
242
243static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
244 int optlen)
245{
246 int ret;
247
248 ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
249 if (ret == 0) {
250 if (rs->rs_cong_monitor) {
251 rds_cong_add_socket(rs);
252 } else {
253 rds_cong_remove_socket(rs);
254 rs->rs_cong_mask = 0;
255 rs->rs_cong_notify = 0;
256 }
257 }
258 return ret;
259}
260
261static int rds_set_transport(struct rds_sock *rs, char __user *optval,
262 int optlen)
263{
264 int t_type;
265
266 if (rs->rs_transport)
267 return -EOPNOTSUPP; /* previously attached to transport */
268
269 if (optlen != sizeof(int))
270 return -EINVAL;
271
272 if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
273 return -EFAULT;
274
275 if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
276 return -EINVAL;
277
278 rs->rs_transport = rds_trans_get(t_type);
279
280 return rs->rs_transport ? 0 : -ENOPROTOOPT;
281}
282
283static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
284 int optlen)
285{
286 int val, valbool;
287
288 if (optlen != sizeof(int))
289 return -EFAULT;
290
291 if (get_user(val, (int __user *)optval))
292 return -EFAULT;
293
294 valbool = val ? 1 : 0;
295
296 if (valbool)
297 sock_set_flag(sk, SOCK_RCVTSTAMP);
298 else
299 sock_reset_flag(sk, SOCK_RCVTSTAMP);
300
301 return 0;
302}
303
304static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval,
305 int optlen)
306{
307 struct rds_rx_trace_so trace;
308 int i;
309
310 if (optlen != sizeof(struct rds_rx_trace_so))
311 return -EFAULT;
312
313 if (copy_from_user(&trace, optval, sizeof(trace)))
314 return -EFAULT;
315
316 if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
317 return -EFAULT;
318
319 rs->rs_rx_traces = trace.rx_traces;
320 for (i = 0; i < rs->rs_rx_traces; i++) {
321 if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
322 rs->rs_rx_traces = 0;
323 return -EFAULT;
324 }
325 rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
326 }
327
328 return 0;
329}
330
331static int rds_setsockopt(struct socket *sock, int level, int optname,
332 char __user *optval, unsigned int optlen)
333{
334 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
335 int ret;
336
337 if (level != SOL_RDS) {
338 ret = -ENOPROTOOPT;
339 goto out;
340 }
341
342 switch (optname) {
343 case RDS_CANCEL_SENT_TO:
344 ret = rds_cancel_sent_to(rs, optval, optlen);
345 break;
346 case RDS_GET_MR:
347 ret = rds_get_mr(rs, optval, optlen);
348 break;
349 case RDS_GET_MR_FOR_DEST:
350 ret = rds_get_mr_for_dest(rs, optval, optlen);
351 break;
352 case RDS_FREE_MR:
353 ret = rds_free_mr(rs, optval, optlen);
354 break;
355 case RDS_RECVERR:
356 ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
357 break;
358 case RDS_CONG_MONITOR:
359 ret = rds_cong_monitor(rs, optval, optlen);
360 break;
361 case SO_RDS_TRANSPORT:
362 lock_sock(sock->sk);
363 ret = rds_set_transport(rs, optval, optlen);
364 release_sock(sock->sk);
365 break;
366 case SO_TIMESTAMP:
367 lock_sock(sock->sk);
368 ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
369 release_sock(sock->sk);
370 break;
371 case SO_RDS_MSG_RXPATH_LATENCY:
372 ret = rds_recv_track_latency(rs, optval, optlen);
373 break;
374 default:
375 ret = -ENOPROTOOPT;
376 }
377out:
378 return ret;
379}
380
381static int rds_getsockopt(struct socket *sock, int level, int optname,
382 char __user *optval, int __user *optlen)
383{
384 struct rds_sock *rs = rds_sk_to_rs(sock->sk);
385 int ret = -ENOPROTOOPT, len;
386 int trans;
387
388 if (level != SOL_RDS)
389 goto out;
390
391 if (get_user(len, optlen)) {
392 ret = -EFAULT;
393 goto out;
394 }
395
396 switch (optname) {
397 case RDS_INFO_FIRST ... RDS_INFO_LAST:
398 ret = rds_info_getsockopt(sock, optname, optval,
399 optlen);
400 break;
401
402 case RDS_RECVERR:
403 if (len < sizeof(int))
404 ret = -EINVAL;
405 else
406 if (put_user(rs->rs_recverr, (int __user *) optval) ||
407 put_user(sizeof(int), optlen))
408 ret = -EFAULT;
409 else
410 ret = 0;
411 break;
412 case SO_RDS_TRANSPORT:
413 if (len < sizeof(int)) {
414 ret = -EINVAL;
415 break;
416 }
417 trans = (rs->rs_transport ? rs->rs_transport->t_type :
418 RDS_TRANS_NONE); /* unbound */
419 if (put_user(trans, (int __user *)optval) ||
420 put_user(sizeof(int), optlen))
421 ret = -EFAULT;
422 else
423 ret = 0;
424 break;
425 default:
426 break;
427 }
428
429out:
430 return ret;
431
432}
433
434static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
435 int addr_len, int flags)
436{
437 struct sock *sk = sock->sk;
438 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
439 struct rds_sock *rs = rds_sk_to_rs(sk);
440 int ret = 0;
441
442 lock_sock(sk);
443
444 if (addr_len != sizeof(struct sockaddr_in)) {
445 ret = -EINVAL;
446 goto out;
447 }
448
449 if (sin->sin_family != AF_INET) {
450 ret = -EAFNOSUPPORT;
451 goto out;
452 }
453
454 if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
455 ret = -EDESTADDRREQ;
456 goto out;
457 }
458
459 rs->rs_conn_addr = sin->sin_addr.s_addr;
460 rs->rs_conn_port = sin->sin_port;
461
462out:
463 release_sock(sk);
464 return ret;
465}
466
467static struct proto rds_proto = {
468 .name = "RDS",
469 .owner = THIS_MODULE,
470 .obj_size = sizeof(struct rds_sock),
471};
472
473static const struct proto_ops rds_proto_ops = {
474 .family = AF_RDS,
475 .owner = THIS_MODULE,
476 .release = rds_release,
477 .bind = rds_bind,
478 .connect = rds_connect,
479 .socketpair = sock_no_socketpair,
480 .accept = sock_no_accept,
481 .getname = rds_getname,
482 .poll = rds_poll,
483 .ioctl = rds_ioctl,
484 .listen = sock_no_listen,
485 .shutdown = sock_no_shutdown,
486 .setsockopt = rds_setsockopt,
487 .getsockopt = rds_getsockopt,
488 .sendmsg = rds_sendmsg,
489 .recvmsg = rds_recvmsg,
490 .mmap = sock_no_mmap,
491 .sendpage = sock_no_sendpage,
492};
493
494static void rds_sock_destruct(struct sock *sk)
495{
496 struct rds_sock *rs = rds_sk_to_rs(sk);
497
498 WARN_ON((&rs->rs_item != rs->rs_item.next ||
499 &rs->rs_item != rs->rs_item.prev));
500}
501
502static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
503{
504 struct rds_sock *rs;
505
506 sock_init_data(sock, sk);
507 sock->ops = &rds_proto_ops;
508 sk->sk_protocol = protocol;
509 sk->sk_destruct = rds_sock_destruct;
510
511 rs = rds_sk_to_rs(sk);
512 spin_lock_init(&rs->rs_lock);
513 rwlock_init(&rs->rs_recv_lock);
514 INIT_LIST_HEAD(&rs->rs_send_queue);
515 INIT_LIST_HEAD(&rs->rs_recv_queue);
516 INIT_LIST_HEAD(&rs->rs_notify_queue);
517 INIT_LIST_HEAD(&rs->rs_cong_list);
518 rds_message_zcopy_queue_init(&rs->rs_zcookie_queue);
519 spin_lock_init(&rs->rs_rdma_lock);
520 rs->rs_rdma_keys = RB_ROOT;
521 rs->rs_rx_traces = 0;
522
523 spin_lock_bh(&rds_sock_lock);
524 list_add_tail(&rs->rs_item, &rds_sock_list);
525 rds_sock_count++;
526 spin_unlock_bh(&rds_sock_lock);
527
528 return 0;
529}
530
531static int rds_create(struct net *net, struct socket *sock, int protocol,
532 int kern)
533{
534 struct sock *sk;
535
536 if (sock->type != SOCK_SEQPACKET || protocol)
537 return -ESOCKTNOSUPPORT;
538
539 sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
540 if (!sk)
541 return -ENOMEM;
542
543 return __rds_create(sock, sk, protocol);
544}
545
546void rds_sock_addref(struct rds_sock *rs)
547{
548 sock_hold(rds_rs_to_sk(rs));
549}
550
551void rds_sock_put(struct rds_sock *rs)
552{
553 sock_put(rds_rs_to_sk(rs));
554}
555
556static const struct net_proto_family rds_family_ops = {
557 .family = AF_RDS,
558 .create = rds_create,
559 .owner = THIS_MODULE,
560};
561
562static void rds_sock_inc_info(struct socket *sock, unsigned int len,
563 struct rds_info_iterator *iter,
564 struct rds_info_lengths *lens)
565{
566 struct rds_sock *rs;
567 struct rds_incoming *inc;
568 unsigned int total = 0;
569
570 len /= sizeof(struct rds_info_message);
571
572 spin_lock_bh(&rds_sock_lock);
573
574 list_for_each_entry(rs, &rds_sock_list, rs_item) {
575 read_lock(&rs->rs_recv_lock);
576
577 /* XXX too lazy to maintain counts.. */
578 list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
579 total++;
580 if (total <= len)
581 rds_inc_info_copy(inc, iter, inc->i_saddr,
582 rs->rs_bound_addr, 1);
583 }
584
585 read_unlock(&rs->rs_recv_lock);
586 }
587
588 spin_unlock_bh(&rds_sock_lock);
589
590 lens->nr = total;
591 lens->each = sizeof(struct rds_info_message);
592}
593
594static void rds_sock_info(struct socket *sock, unsigned int len,
595 struct rds_info_iterator *iter,
596 struct rds_info_lengths *lens)
597{
598 struct rds_info_socket sinfo;
599 struct rds_sock *rs;
600
601 len /= sizeof(struct rds_info_socket);
602
603 spin_lock_bh(&rds_sock_lock);
604
605 if (len < rds_sock_count)
606 goto out;
607
608 list_for_each_entry(rs, &rds_sock_list, rs_item) {
609 sinfo.sndbuf = rds_sk_sndbuf(rs);
610 sinfo.rcvbuf = rds_sk_rcvbuf(rs);
611 sinfo.bound_addr = rs->rs_bound_addr;
612 sinfo.connected_addr = rs->rs_conn_addr;
613 sinfo.bound_port = rs->rs_bound_port;
614 sinfo.connected_port = rs->rs_conn_port;
615 sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
616
617 rds_info_copy(iter, &sinfo, sizeof(sinfo));
618 }
619
620out:
621 lens->nr = rds_sock_count;
622 lens->each = sizeof(struct rds_info_socket);
623
624 spin_unlock_bh(&rds_sock_lock);
625}
626
627static void rds_exit(void)
628{
629 sock_unregister(rds_family_ops.family);
630 proto_unregister(&rds_proto);
631 rds_conn_exit();
632 rds_cong_exit();
633 rds_sysctl_exit();
634 rds_threads_exit();
635 rds_stats_exit();
636 rds_page_exit();
637 rds_bind_lock_destroy();
638 rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
639 rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
640}
641module_exit(rds_exit);
642
643u32 rds_gen_num;
644
645static int rds_init(void)
646{
647 int ret;
648
649 net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
650
651 ret = rds_bind_lock_init();
652 if (ret)
653 goto out;
654
655 ret = rds_conn_init();
656 if (ret)
657 goto out_bind;
658
659 ret = rds_threads_init();
660 if (ret)
661 goto out_conn;
662 ret = rds_sysctl_init();
663 if (ret)
664 goto out_threads;
665 ret = rds_stats_init();
666 if (ret)
667 goto out_sysctl;
668 ret = proto_register(&rds_proto, 1);
669 if (ret)
670 goto out_stats;
671 ret = sock_register(&rds_family_ops);
672 if (ret)
673 goto out_proto;
674
675 rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
676 rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
677
678 goto out;
679
680out_proto:
681 proto_unregister(&rds_proto);
682out_stats:
683 rds_stats_exit();
684out_sysctl:
685 rds_sysctl_exit();
686out_threads:
687 rds_threads_exit();
688out_conn:
689 rds_conn_exit();
690 rds_cong_exit();
691 rds_page_exit();
692out_bind:
693 rds_bind_lock_destroy();
694out:
695 return ret;
696}
697module_init(rds_init);
698
699#define DRV_VERSION "4.0"
700#define DRV_RELDATE "Feb 12, 2009"
701
702MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
703MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
704 " v" DRV_VERSION " (" DRV_RELDATE ")");
705MODULE_VERSION(DRV_VERSION);
706MODULE_LICENSE("Dual BSD/GPL");
707MODULE_ALIAS_NETPROTO(PF_RDS);