Loading...
1/*
2 * NET4: Implementation of BSD Unix domain sockets.
3 *
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * Fixes:
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
21 * Mike Shaver's work.
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
28 * reference counting
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
31 * Lots of bug fixes.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
43 * dgram receiver.
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
51 *
52 *
53 * Known differences from reference BSD that was tested:
54 *
55 * [TO FIX]
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
60 * [NOT TO FIX]
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
68 *
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
73 *
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
80 * with BSD names.
81 */
82
83#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84
85#include <linux/module.h>
86#include <linux/kernel.h>
87#include <linux/signal.h>
88#include <linux/sched.h>
89#include <linux/errno.h>
90#include <linux/string.h>
91#include <linux/stat.h>
92#include <linux/dcache.h>
93#include <linux/namei.h>
94#include <linux/socket.h>
95#include <linux/un.h>
96#include <linux/fcntl.h>
97#include <linux/termios.h>
98#include <linux/sockios.h>
99#include <linux/net.h>
100#include <linux/in.h>
101#include <linux/fs.h>
102#include <linux/slab.h>
103#include <linux/uaccess.h>
104#include <linux/skbuff.h>
105#include <linux/netdevice.h>
106#include <net/net_namespace.h>
107#include <net/sock.h>
108#include <net/tcp_states.h>
109#include <net/af_unix.h>
110#include <linux/proc_fs.h>
111#include <linux/seq_file.h>
112#include <net/scm.h>
113#include <linux/init.h>
114#include <linux/poll.h>
115#include <linux/rtnetlink.h>
116#include <linux/mount.h>
117#include <net/checksum.h>
118#include <linux/security.h>
119#include <linux/freezer.h>
120
121struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
122EXPORT_SYMBOL_GPL(unix_socket_table);
123DEFINE_SPINLOCK(unix_table_lock);
124EXPORT_SYMBOL_GPL(unix_table_lock);
125static atomic_long_t unix_nr_socks;
126
127
128static struct hlist_head *unix_sockets_unbound(void *addr)
129{
130 unsigned long hash = (unsigned long)addr;
131
132 hash ^= hash >> 16;
133 hash ^= hash >> 8;
134 hash %= UNIX_HASH_SIZE;
135 return &unix_socket_table[UNIX_HASH_SIZE + hash];
136}
137
138#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
139
140#ifdef CONFIG_SECURITY_NETWORK
141static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142{
143 UNIXCB(skb).secid = scm->secid;
144}
145
146static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
147{
148 scm->secid = UNIXCB(skb).secid;
149}
150
151static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
152{
153 return (scm->secid == UNIXCB(skb).secid);
154}
155#else
156static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
157{ }
158
159static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
160{ }
161
162static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
163{
164 return true;
165}
166#endif /* CONFIG_SECURITY_NETWORK */
167
168/*
169 * SMP locking strategy:
170 * hash table is protected with spinlock unix_table_lock
171 * each socket state is protected by separate spin lock.
172 */
173
174static inline unsigned int unix_hash_fold(__wsum n)
175{
176 unsigned int hash = (__force unsigned int)csum_fold(n);
177
178 hash ^= hash>>8;
179 return hash&(UNIX_HASH_SIZE-1);
180}
181
182#define unix_peer(sk) (unix_sk(sk)->peer)
183
184static inline int unix_our_peer(struct sock *sk, struct sock *osk)
185{
186 return unix_peer(osk) == sk;
187}
188
189static inline int unix_may_send(struct sock *sk, struct sock *osk)
190{
191 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
192}
193
194static inline int unix_recvq_full(struct sock const *sk)
195{
196 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
197}
198
199struct sock *unix_peer_get(struct sock *s)
200{
201 struct sock *peer;
202
203 unix_state_lock(s);
204 peer = unix_peer(s);
205 if (peer)
206 sock_hold(peer);
207 unix_state_unlock(s);
208 return peer;
209}
210EXPORT_SYMBOL_GPL(unix_peer_get);
211
212static inline void unix_release_addr(struct unix_address *addr)
213{
214 if (atomic_dec_and_test(&addr->refcnt))
215 kfree(addr);
216}
217
218/*
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
223 */
224
225static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
226{
227 if (len <= sizeof(short) || len > sizeof(*sunaddr))
228 return -EINVAL;
229 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
230 return -EINVAL;
231 if (sunaddr->sun_path[0]) {
232 /*
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesn't as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
238 */
239 ((char *)sunaddr)[len] = 0;
240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
241 return len;
242 }
243
244 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
245 return len;
246}
247
248static void __unix_remove_socket(struct sock *sk)
249{
250 sk_del_node_init(sk);
251}
252
253static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
254{
255 WARN_ON(!sk_unhashed(sk));
256 sk_add_node(sk, list);
257}
258
259static inline void unix_remove_socket(struct sock *sk)
260{
261 spin_lock(&unix_table_lock);
262 __unix_remove_socket(sk);
263 spin_unlock(&unix_table_lock);
264}
265
266static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
267{
268 spin_lock(&unix_table_lock);
269 __unix_insert_socket(list, sk);
270 spin_unlock(&unix_table_lock);
271}
272
273static struct sock *__unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
275 int len, int type, unsigned int hash)
276{
277 struct sock *s;
278
279 sk_for_each(s, &unix_socket_table[hash ^ type]) {
280 struct unix_sock *u = unix_sk(s);
281
282 if (!net_eq(sock_net(s), net))
283 continue;
284
285 if (u->addr->len == len &&
286 !memcmp(u->addr->name, sunname, len))
287 goto found;
288 }
289 s = NULL;
290found:
291 return s;
292}
293
294static inline struct sock *unix_find_socket_byname(struct net *net,
295 struct sockaddr_un *sunname,
296 int len, int type,
297 unsigned int hash)
298{
299 struct sock *s;
300
301 spin_lock(&unix_table_lock);
302 s = __unix_find_socket_byname(net, sunname, len, type, hash);
303 if (s)
304 sock_hold(s);
305 spin_unlock(&unix_table_lock);
306 return s;
307}
308
309static struct sock *unix_find_socket_byinode(struct inode *i)
310{
311 struct sock *s;
312
313 spin_lock(&unix_table_lock);
314 sk_for_each(s,
315 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
316 struct dentry *dentry = unix_sk(s)->path.dentry;
317
318 if (dentry && d_backing_inode(dentry) == i) {
319 sock_hold(s);
320 goto found;
321 }
322 }
323 s = NULL;
324found:
325 spin_unlock(&unix_table_lock);
326 return s;
327}
328
329/* Support code for asymmetrically connected dgram sockets
330 *
331 * If a datagram socket is connected to a socket not itself connected
332 * to the first socket (eg, /dev/log), clients may only enqueue more
333 * messages if the present receive queue of the server socket is not
334 * "too large". This means there's a second writeability condition
335 * poll and sendmsg need to test. The dgram recv code will do a wake
336 * up on the peer_wait wait queue of a socket upon reception of a
337 * datagram which needs to be propagated to sleeping would-be writers
338 * since these might not have sent anything so far. This can't be
339 * accomplished via poll_wait because the lifetime of the server
340 * socket might be less than that of its clients if these break their
341 * association with it or if the server socket is closed while clients
342 * are still connected to it and there's no way to inform "a polling
343 * implementation" that it should let go of a certain wait queue
344 *
345 * In order to propagate a wake up, a wait_queue_t of the client
346 * socket is enqueued on the peer_wait queue of the server socket
347 * whose wake function does a wake_up on the ordinary client socket
348 * wait queue. This connection is established whenever a write (or
349 * poll for write) hit the flow control condition and broken when the
350 * association to the server socket is dissolved or after a wake up
351 * was relayed.
352 */
353
354static int unix_dgram_peer_wake_relay(wait_queue_t *q, unsigned mode, int flags,
355 void *key)
356{
357 struct unix_sock *u;
358 wait_queue_head_t *u_sleep;
359
360 u = container_of(q, struct unix_sock, peer_wake);
361
362 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
363 q);
364 u->peer_wake.private = NULL;
365
366 /* relaying can only happen while the wq still exists */
367 u_sleep = sk_sleep(&u->sk);
368 if (u_sleep)
369 wake_up_interruptible_poll(u_sleep, key);
370
371 return 0;
372}
373
374static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
375{
376 struct unix_sock *u, *u_other;
377 int rc;
378
379 u = unix_sk(sk);
380 u_other = unix_sk(other);
381 rc = 0;
382 spin_lock(&u_other->peer_wait.lock);
383
384 if (!u->peer_wake.private) {
385 u->peer_wake.private = other;
386 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
387
388 rc = 1;
389 }
390
391 spin_unlock(&u_other->peer_wait.lock);
392 return rc;
393}
394
395static void unix_dgram_peer_wake_disconnect(struct sock *sk,
396 struct sock *other)
397{
398 struct unix_sock *u, *u_other;
399
400 u = unix_sk(sk);
401 u_other = unix_sk(other);
402 spin_lock(&u_other->peer_wait.lock);
403
404 if (u->peer_wake.private == other) {
405 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
406 u->peer_wake.private = NULL;
407 }
408
409 spin_unlock(&u_other->peer_wait.lock);
410}
411
412static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
413 struct sock *other)
414{
415 unix_dgram_peer_wake_disconnect(sk, other);
416 wake_up_interruptible_poll(sk_sleep(sk),
417 POLLOUT |
418 POLLWRNORM |
419 POLLWRBAND);
420}
421
422/* preconditions:
423 * - unix_peer(sk) == other
424 * - association is stable
425 */
426static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
427{
428 int connected;
429
430 connected = unix_dgram_peer_wake_connect(sk, other);
431
432 if (unix_recvq_full(other))
433 return 1;
434
435 if (connected)
436 unix_dgram_peer_wake_disconnect(sk, other);
437
438 return 0;
439}
440
441static int unix_writable(const struct sock *sk)
442{
443 return sk->sk_state != TCP_LISTEN &&
444 (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
445}
446
447static void unix_write_space(struct sock *sk)
448{
449 struct socket_wq *wq;
450
451 rcu_read_lock();
452 if (unix_writable(sk)) {
453 wq = rcu_dereference(sk->sk_wq);
454 if (skwq_has_sleeper(wq))
455 wake_up_interruptible_sync_poll(&wq->wait,
456 POLLOUT | POLLWRNORM | POLLWRBAND);
457 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
458 }
459 rcu_read_unlock();
460}
461
462/* When dgram socket disconnects (or changes its peer), we clear its receive
463 * queue of packets arrived from previous peer. First, it allows to do
464 * flow control based only on wmem_alloc; second, sk connected to peer
465 * may receive messages only from that peer. */
466static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
467{
468 if (!skb_queue_empty(&sk->sk_receive_queue)) {
469 skb_queue_purge(&sk->sk_receive_queue);
470 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
471
472 /* If one link of bidirectional dgram pipe is disconnected,
473 * we signal error. Messages are lost. Do not make this,
474 * when peer was not connected to us.
475 */
476 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
477 other->sk_err = ECONNRESET;
478 other->sk_error_report(other);
479 }
480 }
481}
482
483static void unix_sock_destructor(struct sock *sk)
484{
485 struct unix_sock *u = unix_sk(sk);
486
487 skb_queue_purge(&sk->sk_receive_queue);
488
489 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
490 WARN_ON(!sk_unhashed(sk));
491 WARN_ON(sk->sk_socket);
492 if (!sock_flag(sk, SOCK_DEAD)) {
493 pr_info("Attempt to release alive unix socket: %p\n", sk);
494 return;
495 }
496
497 if (u->addr)
498 unix_release_addr(u->addr);
499
500 atomic_long_dec(&unix_nr_socks);
501 local_bh_disable();
502 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
503 local_bh_enable();
504#ifdef UNIX_REFCNT_DEBUG
505 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
506 atomic_long_read(&unix_nr_socks));
507#endif
508}
509
510static void unix_release_sock(struct sock *sk, int embrion)
511{
512 struct unix_sock *u = unix_sk(sk);
513 struct path path;
514 struct sock *skpair;
515 struct sk_buff *skb;
516 int state;
517
518 unix_remove_socket(sk);
519
520 /* Clear state */
521 unix_state_lock(sk);
522 sock_orphan(sk);
523 sk->sk_shutdown = SHUTDOWN_MASK;
524 path = u->path;
525 u->path.dentry = NULL;
526 u->path.mnt = NULL;
527 state = sk->sk_state;
528 sk->sk_state = TCP_CLOSE;
529 unix_state_unlock(sk);
530
531 wake_up_interruptible_all(&u->peer_wait);
532
533 skpair = unix_peer(sk);
534
535 if (skpair != NULL) {
536 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
537 unix_state_lock(skpair);
538 /* No more writes */
539 skpair->sk_shutdown = SHUTDOWN_MASK;
540 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
541 skpair->sk_err = ECONNRESET;
542 unix_state_unlock(skpair);
543 skpair->sk_state_change(skpair);
544 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
545 }
546
547 unix_dgram_peer_wake_disconnect(sk, skpair);
548 sock_put(skpair); /* It may now die */
549 unix_peer(sk) = NULL;
550 }
551
552 /* Try to flush out this socket. Throw out buffers at least */
553
554 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
555 if (state == TCP_LISTEN)
556 unix_release_sock(skb->sk, 1);
557 /* passed fds are erased in the kfree_skb hook */
558 UNIXCB(skb).consumed = skb->len;
559 kfree_skb(skb);
560 }
561
562 if (path.dentry)
563 path_put(&path);
564
565 sock_put(sk);
566
567 /* ---- Socket is dead now and most probably destroyed ---- */
568
569 /*
570 * Fixme: BSD difference: In BSD all sockets connected to us get
571 * ECONNRESET and we die on the spot. In Linux we behave
572 * like files and pipes do and wait for the last
573 * dereference.
574 *
575 * Can't we simply set sock->err?
576 *
577 * What the above comment does talk about? --ANK(980817)
578 */
579
580 if (unix_tot_inflight)
581 unix_gc(); /* Garbage collect fds */
582}
583
584static void init_peercred(struct sock *sk)
585{
586 put_pid(sk->sk_peer_pid);
587 if (sk->sk_peer_cred)
588 put_cred(sk->sk_peer_cred);
589 sk->sk_peer_pid = get_pid(task_tgid(current));
590 sk->sk_peer_cred = get_current_cred();
591}
592
593static void copy_peercred(struct sock *sk, struct sock *peersk)
594{
595 put_pid(sk->sk_peer_pid);
596 if (sk->sk_peer_cred)
597 put_cred(sk->sk_peer_cred);
598 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
599 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
600}
601
602static int unix_listen(struct socket *sock, int backlog)
603{
604 int err;
605 struct sock *sk = sock->sk;
606 struct unix_sock *u = unix_sk(sk);
607 struct pid *old_pid = NULL;
608
609 err = -EOPNOTSUPP;
610 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
611 goto out; /* Only stream/seqpacket sockets accept */
612 err = -EINVAL;
613 if (!u->addr)
614 goto out; /* No listens on an unbound socket */
615 unix_state_lock(sk);
616 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
617 goto out_unlock;
618 if (backlog > sk->sk_max_ack_backlog)
619 wake_up_interruptible_all(&u->peer_wait);
620 sk->sk_max_ack_backlog = backlog;
621 sk->sk_state = TCP_LISTEN;
622 /* set credentials so connect can copy them */
623 init_peercred(sk);
624 err = 0;
625
626out_unlock:
627 unix_state_unlock(sk);
628 put_pid(old_pid);
629out:
630 return err;
631}
632
633static int unix_release(struct socket *);
634static int unix_bind(struct socket *, struct sockaddr *, int);
635static int unix_stream_connect(struct socket *, struct sockaddr *,
636 int addr_len, int flags);
637static int unix_socketpair(struct socket *, struct socket *);
638static int unix_accept(struct socket *, struct socket *, int);
639static int unix_getname(struct socket *, struct sockaddr *, int *, int);
640static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
641static unsigned int unix_dgram_poll(struct file *, struct socket *,
642 poll_table *);
643static int unix_ioctl(struct socket *, unsigned int, unsigned long);
644static int unix_shutdown(struct socket *, int);
645static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
646static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
647static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
648 size_t size, int flags);
649static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
650 struct pipe_inode_info *, size_t size,
651 unsigned int flags);
652static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
653static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
654static int unix_dgram_connect(struct socket *, struct sockaddr *,
655 int, int);
656static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
657static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
658 int);
659
660static int unix_set_peek_off(struct sock *sk, int val)
661{
662 struct unix_sock *u = unix_sk(sk);
663
664 if (mutex_lock_interruptible(&u->iolock))
665 return -EINTR;
666
667 sk->sk_peek_off = val;
668 mutex_unlock(&u->iolock);
669
670 return 0;
671}
672
673
674static const struct proto_ops unix_stream_ops = {
675 .family = PF_UNIX,
676 .owner = THIS_MODULE,
677 .release = unix_release,
678 .bind = unix_bind,
679 .connect = unix_stream_connect,
680 .socketpair = unix_socketpair,
681 .accept = unix_accept,
682 .getname = unix_getname,
683 .poll = unix_poll,
684 .ioctl = unix_ioctl,
685 .listen = unix_listen,
686 .shutdown = unix_shutdown,
687 .setsockopt = sock_no_setsockopt,
688 .getsockopt = sock_no_getsockopt,
689 .sendmsg = unix_stream_sendmsg,
690 .recvmsg = unix_stream_recvmsg,
691 .mmap = sock_no_mmap,
692 .sendpage = unix_stream_sendpage,
693 .splice_read = unix_stream_splice_read,
694 .set_peek_off = unix_set_peek_off,
695};
696
697static const struct proto_ops unix_dgram_ops = {
698 .family = PF_UNIX,
699 .owner = THIS_MODULE,
700 .release = unix_release,
701 .bind = unix_bind,
702 .connect = unix_dgram_connect,
703 .socketpair = unix_socketpair,
704 .accept = sock_no_accept,
705 .getname = unix_getname,
706 .poll = unix_dgram_poll,
707 .ioctl = unix_ioctl,
708 .listen = sock_no_listen,
709 .shutdown = unix_shutdown,
710 .setsockopt = sock_no_setsockopt,
711 .getsockopt = sock_no_getsockopt,
712 .sendmsg = unix_dgram_sendmsg,
713 .recvmsg = unix_dgram_recvmsg,
714 .mmap = sock_no_mmap,
715 .sendpage = sock_no_sendpage,
716 .set_peek_off = unix_set_peek_off,
717};
718
719static const struct proto_ops unix_seqpacket_ops = {
720 .family = PF_UNIX,
721 .owner = THIS_MODULE,
722 .release = unix_release,
723 .bind = unix_bind,
724 .connect = unix_stream_connect,
725 .socketpair = unix_socketpair,
726 .accept = unix_accept,
727 .getname = unix_getname,
728 .poll = unix_dgram_poll,
729 .ioctl = unix_ioctl,
730 .listen = unix_listen,
731 .shutdown = unix_shutdown,
732 .setsockopt = sock_no_setsockopt,
733 .getsockopt = sock_no_getsockopt,
734 .sendmsg = unix_seqpacket_sendmsg,
735 .recvmsg = unix_seqpacket_recvmsg,
736 .mmap = sock_no_mmap,
737 .sendpage = sock_no_sendpage,
738 .set_peek_off = unix_set_peek_off,
739};
740
741static struct proto unix_proto = {
742 .name = "UNIX",
743 .owner = THIS_MODULE,
744 .obj_size = sizeof(struct unix_sock),
745};
746
747/*
748 * AF_UNIX sockets do not interact with hardware, hence they
749 * dont trigger interrupts - so it's safe for them to have
750 * bh-unsafe locking for their sk_receive_queue.lock. Split off
751 * this special lock-class by reinitializing the spinlock key:
752 */
753static struct lock_class_key af_unix_sk_receive_queue_lock_key;
754
755static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
756{
757 struct sock *sk = NULL;
758 struct unix_sock *u;
759
760 atomic_long_inc(&unix_nr_socks);
761 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
762 goto out;
763
764 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
765 if (!sk)
766 goto out;
767
768 sock_init_data(sock, sk);
769 lockdep_set_class(&sk->sk_receive_queue.lock,
770 &af_unix_sk_receive_queue_lock_key);
771
772 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
773 sk->sk_write_space = unix_write_space;
774 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
775 sk->sk_destruct = unix_sock_destructor;
776 u = unix_sk(sk);
777 u->path.dentry = NULL;
778 u->path.mnt = NULL;
779 spin_lock_init(&u->lock);
780 atomic_long_set(&u->inflight, 0);
781 INIT_LIST_HEAD(&u->link);
782 mutex_init(&u->iolock); /* single task reading lock */
783 mutex_init(&u->bindlock); /* single task binding lock */
784 init_waitqueue_head(&u->peer_wait);
785 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
786 unix_insert_socket(unix_sockets_unbound(sk), sk);
787out:
788 if (sk == NULL)
789 atomic_long_dec(&unix_nr_socks);
790 else {
791 local_bh_disable();
792 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
793 local_bh_enable();
794 }
795 return sk;
796}
797
798static int unix_create(struct net *net, struct socket *sock, int protocol,
799 int kern)
800{
801 if (protocol && protocol != PF_UNIX)
802 return -EPROTONOSUPPORT;
803
804 sock->state = SS_UNCONNECTED;
805
806 switch (sock->type) {
807 case SOCK_STREAM:
808 sock->ops = &unix_stream_ops;
809 break;
810 /*
811 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
812 * nothing uses it.
813 */
814 case SOCK_RAW:
815 sock->type = SOCK_DGRAM;
816 case SOCK_DGRAM:
817 sock->ops = &unix_dgram_ops;
818 break;
819 case SOCK_SEQPACKET:
820 sock->ops = &unix_seqpacket_ops;
821 break;
822 default:
823 return -ESOCKTNOSUPPORT;
824 }
825
826 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
827}
828
829static int unix_release(struct socket *sock)
830{
831 struct sock *sk = sock->sk;
832
833 if (!sk)
834 return 0;
835
836 unix_release_sock(sk, 0);
837 sock->sk = NULL;
838
839 return 0;
840}
841
842static int unix_autobind(struct socket *sock)
843{
844 struct sock *sk = sock->sk;
845 struct net *net = sock_net(sk);
846 struct unix_sock *u = unix_sk(sk);
847 static u32 ordernum = 1;
848 struct unix_address *addr;
849 int err;
850 unsigned int retries = 0;
851
852 err = mutex_lock_interruptible(&u->bindlock);
853 if (err)
854 return err;
855
856 err = 0;
857 if (u->addr)
858 goto out;
859
860 err = -ENOMEM;
861 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
862 if (!addr)
863 goto out;
864
865 addr->name->sun_family = AF_UNIX;
866 atomic_set(&addr->refcnt, 1);
867
868retry:
869 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
870 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
871
872 spin_lock(&unix_table_lock);
873 ordernum = (ordernum+1)&0xFFFFF;
874
875 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
876 addr->hash)) {
877 spin_unlock(&unix_table_lock);
878 /*
879 * __unix_find_socket_byname() may take long time if many names
880 * are already in use.
881 */
882 cond_resched();
883 /* Give up if all names seems to be in use. */
884 if (retries++ == 0xFFFFF) {
885 err = -ENOSPC;
886 kfree(addr);
887 goto out;
888 }
889 goto retry;
890 }
891 addr->hash ^= sk->sk_type;
892
893 __unix_remove_socket(sk);
894 u->addr = addr;
895 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
896 spin_unlock(&unix_table_lock);
897 err = 0;
898
899out: mutex_unlock(&u->bindlock);
900 return err;
901}
902
903static struct sock *unix_find_other(struct net *net,
904 struct sockaddr_un *sunname, int len,
905 int type, unsigned int hash, int *error)
906{
907 struct sock *u;
908 struct path path;
909 int err = 0;
910
911 if (sunname->sun_path[0]) {
912 struct inode *inode;
913 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
914 if (err)
915 goto fail;
916 inode = d_backing_inode(path.dentry);
917 err = inode_permission(inode, MAY_WRITE);
918 if (err)
919 goto put_fail;
920
921 err = -ECONNREFUSED;
922 if (!S_ISSOCK(inode->i_mode))
923 goto put_fail;
924 u = unix_find_socket_byinode(inode);
925 if (!u)
926 goto put_fail;
927
928 if (u->sk_type == type)
929 touch_atime(&path);
930
931 path_put(&path);
932
933 err = -EPROTOTYPE;
934 if (u->sk_type != type) {
935 sock_put(u);
936 goto fail;
937 }
938 } else {
939 err = -ECONNREFUSED;
940 u = unix_find_socket_byname(net, sunname, len, type, hash);
941 if (u) {
942 struct dentry *dentry;
943 dentry = unix_sk(u)->path.dentry;
944 if (dentry)
945 touch_atime(&unix_sk(u)->path);
946 } else
947 goto fail;
948 }
949 return u;
950
951put_fail:
952 path_put(&path);
953fail:
954 *error = err;
955 return NULL;
956}
957
958static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
959{
960 struct dentry *dentry;
961 struct path path;
962 int err = 0;
963 /*
964 * Get the parent directory, calculate the hash for last
965 * component.
966 */
967 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
968 err = PTR_ERR(dentry);
969 if (IS_ERR(dentry))
970 return err;
971
972 /*
973 * All right, let's create it.
974 */
975 err = security_path_mknod(&path, dentry, mode, 0);
976 if (!err) {
977 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
978 if (!err) {
979 res->mnt = mntget(path.mnt);
980 res->dentry = dget(dentry);
981 }
982 }
983 done_path_create(&path, dentry);
984 return err;
985}
986
987static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
988{
989 struct sock *sk = sock->sk;
990 struct net *net = sock_net(sk);
991 struct unix_sock *u = unix_sk(sk);
992 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
993 char *sun_path = sunaddr->sun_path;
994 int err;
995 unsigned int hash;
996 struct unix_address *addr;
997 struct hlist_head *list;
998 struct path path = { NULL, NULL };
999
1000 err = -EINVAL;
1001 if (sunaddr->sun_family != AF_UNIX)
1002 goto out;
1003
1004 if (addr_len == sizeof(short)) {
1005 err = unix_autobind(sock);
1006 goto out;
1007 }
1008
1009 err = unix_mkname(sunaddr, addr_len, &hash);
1010 if (err < 0)
1011 goto out;
1012 addr_len = err;
1013
1014 if (sun_path[0]) {
1015 umode_t mode = S_IFSOCK |
1016 (SOCK_INODE(sock)->i_mode & ~current_umask());
1017 err = unix_mknod(sun_path, mode, &path);
1018 if (err) {
1019 if (err == -EEXIST)
1020 err = -EADDRINUSE;
1021 goto out;
1022 }
1023 }
1024
1025 err = mutex_lock_interruptible(&u->bindlock);
1026 if (err)
1027 goto out_put;
1028
1029 err = -EINVAL;
1030 if (u->addr)
1031 goto out_up;
1032
1033 err = -ENOMEM;
1034 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1035 if (!addr)
1036 goto out_up;
1037
1038 memcpy(addr->name, sunaddr, addr_len);
1039 addr->len = addr_len;
1040 addr->hash = hash ^ sk->sk_type;
1041 atomic_set(&addr->refcnt, 1);
1042
1043 if (sun_path[0]) {
1044 addr->hash = UNIX_HASH_SIZE;
1045 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1046 spin_lock(&unix_table_lock);
1047 u->path = path;
1048 list = &unix_socket_table[hash];
1049 } else {
1050 spin_lock(&unix_table_lock);
1051 err = -EADDRINUSE;
1052 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1053 sk->sk_type, hash)) {
1054 unix_release_addr(addr);
1055 goto out_unlock;
1056 }
1057
1058 list = &unix_socket_table[addr->hash];
1059 }
1060
1061 err = 0;
1062 __unix_remove_socket(sk);
1063 u->addr = addr;
1064 __unix_insert_socket(list, sk);
1065
1066out_unlock:
1067 spin_unlock(&unix_table_lock);
1068out_up:
1069 mutex_unlock(&u->bindlock);
1070out_put:
1071 if (err)
1072 path_put(&path);
1073out:
1074 return err;
1075}
1076
1077static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1078{
1079 if (unlikely(sk1 == sk2) || !sk2) {
1080 unix_state_lock(sk1);
1081 return;
1082 }
1083 if (sk1 < sk2) {
1084 unix_state_lock(sk1);
1085 unix_state_lock_nested(sk2);
1086 } else {
1087 unix_state_lock(sk2);
1088 unix_state_lock_nested(sk1);
1089 }
1090}
1091
1092static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1093{
1094 if (unlikely(sk1 == sk2) || !sk2) {
1095 unix_state_unlock(sk1);
1096 return;
1097 }
1098 unix_state_unlock(sk1);
1099 unix_state_unlock(sk2);
1100}
1101
1102static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1103 int alen, int flags)
1104{
1105 struct sock *sk = sock->sk;
1106 struct net *net = sock_net(sk);
1107 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1108 struct sock *other;
1109 unsigned int hash;
1110 int err;
1111
1112 if (addr->sa_family != AF_UNSPEC) {
1113 err = unix_mkname(sunaddr, alen, &hash);
1114 if (err < 0)
1115 goto out;
1116 alen = err;
1117
1118 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1119 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1120 goto out;
1121
1122restart:
1123 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1124 if (!other)
1125 goto out;
1126
1127 unix_state_double_lock(sk, other);
1128
1129 /* Apparently VFS overslept socket death. Retry. */
1130 if (sock_flag(other, SOCK_DEAD)) {
1131 unix_state_double_unlock(sk, other);
1132 sock_put(other);
1133 goto restart;
1134 }
1135
1136 err = -EPERM;
1137 if (!unix_may_send(sk, other))
1138 goto out_unlock;
1139
1140 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1141 if (err)
1142 goto out_unlock;
1143
1144 } else {
1145 /*
1146 * 1003.1g breaking connected state with AF_UNSPEC
1147 */
1148 other = NULL;
1149 unix_state_double_lock(sk, other);
1150 }
1151
1152 /*
1153 * If it was connected, reconnect.
1154 */
1155 if (unix_peer(sk)) {
1156 struct sock *old_peer = unix_peer(sk);
1157 unix_peer(sk) = other;
1158 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1159
1160 unix_state_double_unlock(sk, other);
1161
1162 if (other != old_peer)
1163 unix_dgram_disconnected(sk, old_peer);
1164 sock_put(old_peer);
1165 } else {
1166 unix_peer(sk) = other;
1167 unix_state_double_unlock(sk, other);
1168 }
1169 return 0;
1170
1171out_unlock:
1172 unix_state_double_unlock(sk, other);
1173 sock_put(other);
1174out:
1175 return err;
1176}
1177
1178static long unix_wait_for_peer(struct sock *other, long timeo)
1179{
1180 struct unix_sock *u = unix_sk(other);
1181 int sched;
1182 DEFINE_WAIT(wait);
1183
1184 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1185
1186 sched = !sock_flag(other, SOCK_DEAD) &&
1187 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1188 unix_recvq_full(other);
1189
1190 unix_state_unlock(other);
1191
1192 if (sched)
1193 timeo = schedule_timeout(timeo);
1194
1195 finish_wait(&u->peer_wait, &wait);
1196 return timeo;
1197}
1198
1199static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1200 int addr_len, int flags)
1201{
1202 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1203 struct sock *sk = sock->sk;
1204 struct net *net = sock_net(sk);
1205 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1206 struct sock *newsk = NULL;
1207 struct sock *other = NULL;
1208 struct sk_buff *skb = NULL;
1209 unsigned int hash;
1210 int st;
1211 int err;
1212 long timeo;
1213
1214 err = unix_mkname(sunaddr, addr_len, &hash);
1215 if (err < 0)
1216 goto out;
1217 addr_len = err;
1218
1219 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1220 (err = unix_autobind(sock)) != 0)
1221 goto out;
1222
1223 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1224
1225 /* First of all allocate resources.
1226 If we will make it after state is locked,
1227 we will have to recheck all again in any case.
1228 */
1229
1230 err = -ENOMEM;
1231
1232 /* create new sock for complete connection */
1233 newsk = unix_create1(sock_net(sk), NULL, 0);
1234 if (newsk == NULL)
1235 goto out;
1236
1237 /* Allocate skb for sending to listening sock */
1238 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1239 if (skb == NULL)
1240 goto out;
1241
1242restart:
1243 /* Find listening sock. */
1244 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1245 if (!other)
1246 goto out;
1247
1248 /* Latch state of peer */
1249 unix_state_lock(other);
1250
1251 /* Apparently VFS overslept socket death. Retry. */
1252 if (sock_flag(other, SOCK_DEAD)) {
1253 unix_state_unlock(other);
1254 sock_put(other);
1255 goto restart;
1256 }
1257
1258 err = -ECONNREFUSED;
1259 if (other->sk_state != TCP_LISTEN)
1260 goto out_unlock;
1261 if (other->sk_shutdown & RCV_SHUTDOWN)
1262 goto out_unlock;
1263
1264 if (unix_recvq_full(other)) {
1265 err = -EAGAIN;
1266 if (!timeo)
1267 goto out_unlock;
1268
1269 timeo = unix_wait_for_peer(other, timeo);
1270
1271 err = sock_intr_errno(timeo);
1272 if (signal_pending(current))
1273 goto out;
1274 sock_put(other);
1275 goto restart;
1276 }
1277
1278 /* Latch our state.
1279
1280 It is tricky place. We need to grab our state lock and cannot
1281 drop lock on peer. It is dangerous because deadlock is
1282 possible. Connect to self case and simultaneous
1283 attempt to connect are eliminated by checking socket
1284 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1285 check this before attempt to grab lock.
1286
1287 Well, and we have to recheck the state after socket locked.
1288 */
1289 st = sk->sk_state;
1290
1291 switch (st) {
1292 case TCP_CLOSE:
1293 /* This is ok... continue with connect */
1294 break;
1295 case TCP_ESTABLISHED:
1296 /* Socket is already connected */
1297 err = -EISCONN;
1298 goto out_unlock;
1299 default:
1300 err = -EINVAL;
1301 goto out_unlock;
1302 }
1303
1304 unix_state_lock_nested(sk);
1305
1306 if (sk->sk_state != st) {
1307 unix_state_unlock(sk);
1308 unix_state_unlock(other);
1309 sock_put(other);
1310 goto restart;
1311 }
1312
1313 err = security_unix_stream_connect(sk, other, newsk);
1314 if (err) {
1315 unix_state_unlock(sk);
1316 goto out_unlock;
1317 }
1318
1319 /* The way is open! Fastly set all the necessary fields... */
1320
1321 sock_hold(sk);
1322 unix_peer(newsk) = sk;
1323 newsk->sk_state = TCP_ESTABLISHED;
1324 newsk->sk_type = sk->sk_type;
1325 init_peercred(newsk);
1326 newu = unix_sk(newsk);
1327 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1328 otheru = unix_sk(other);
1329
1330 /* copy address information from listening to new sock*/
1331 if (otheru->addr) {
1332 atomic_inc(&otheru->addr->refcnt);
1333 newu->addr = otheru->addr;
1334 }
1335 if (otheru->path.dentry) {
1336 path_get(&otheru->path);
1337 newu->path = otheru->path;
1338 }
1339
1340 /* Set credentials */
1341 copy_peercred(sk, other);
1342
1343 sock->state = SS_CONNECTED;
1344 sk->sk_state = TCP_ESTABLISHED;
1345 sock_hold(newsk);
1346
1347 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1348 unix_peer(sk) = newsk;
1349
1350 unix_state_unlock(sk);
1351
1352 /* take ten and and send info to listening sock */
1353 spin_lock(&other->sk_receive_queue.lock);
1354 __skb_queue_tail(&other->sk_receive_queue, skb);
1355 spin_unlock(&other->sk_receive_queue.lock);
1356 unix_state_unlock(other);
1357 other->sk_data_ready(other);
1358 sock_put(other);
1359 return 0;
1360
1361out_unlock:
1362 if (other)
1363 unix_state_unlock(other);
1364
1365out:
1366 kfree_skb(skb);
1367 if (newsk)
1368 unix_release_sock(newsk, 0);
1369 if (other)
1370 sock_put(other);
1371 return err;
1372}
1373
1374static int unix_socketpair(struct socket *socka, struct socket *sockb)
1375{
1376 struct sock *ska = socka->sk, *skb = sockb->sk;
1377
1378 /* Join our sockets back to back */
1379 sock_hold(ska);
1380 sock_hold(skb);
1381 unix_peer(ska) = skb;
1382 unix_peer(skb) = ska;
1383 init_peercred(ska);
1384 init_peercred(skb);
1385
1386 if (ska->sk_type != SOCK_DGRAM) {
1387 ska->sk_state = TCP_ESTABLISHED;
1388 skb->sk_state = TCP_ESTABLISHED;
1389 socka->state = SS_CONNECTED;
1390 sockb->state = SS_CONNECTED;
1391 }
1392 return 0;
1393}
1394
1395static void unix_sock_inherit_flags(const struct socket *old,
1396 struct socket *new)
1397{
1398 if (test_bit(SOCK_PASSCRED, &old->flags))
1399 set_bit(SOCK_PASSCRED, &new->flags);
1400 if (test_bit(SOCK_PASSSEC, &old->flags))
1401 set_bit(SOCK_PASSSEC, &new->flags);
1402}
1403
1404static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1405{
1406 struct sock *sk = sock->sk;
1407 struct sock *tsk;
1408 struct sk_buff *skb;
1409 int err;
1410
1411 err = -EOPNOTSUPP;
1412 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1413 goto out;
1414
1415 err = -EINVAL;
1416 if (sk->sk_state != TCP_LISTEN)
1417 goto out;
1418
1419 /* If socket state is TCP_LISTEN it cannot change (for now...),
1420 * so that no locks are necessary.
1421 */
1422
1423 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1424 if (!skb) {
1425 /* This means receive shutdown. */
1426 if (err == 0)
1427 err = -EINVAL;
1428 goto out;
1429 }
1430
1431 tsk = skb->sk;
1432 skb_free_datagram(sk, skb);
1433 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1434
1435 /* attach accepted sock to socket */
1436 unix_state_lock(tsk);
1437 newsock->state = SS_CONNECTED;
1438 unix_sock_inherit_flags(sock, newsock);
1439 sock_graft(tsk, newsock);
1440 unix_state_unlock(tsk);
1441 return 0;
1442
1443out:
1444 return err;
1445}
1446
1447
1448static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1449{
1450 struct sock *sk = sock->sk;
1451 struct unix_sock *u;
1452 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1453 int err = 0;
1454
1455 if (peer) {
1456 sk = unix_peer_get(sk);
1457
1458 err = -ENOTCONN;
1459 if (!sk)
1460 goto out;
1461 err = 0;
1462 } else {
1463 sock_hold(sk);
1464 }
1465
1466 u = unix_sk(sk);
1467 unix_state_lock(sk);
1468 if (!u->addr) {
1469 sunaddr->sun_family = AF_UNIX;
1470 sunaddr->sun_path[0] = 0;
1471 *uaddr_len = sizeof(short);
1472 } else {
1473 struct unix_address *addr = u->addr;
1474
1475 *uaddr_len = addr->len;
1476 memcpy(sunaddr, addr->name, *uaddr_len);
1477 }
1478 unix_state_unlock(sk);
1479 sock_put(sk);
1480out:
1481 return err;
1482}
1483
1484static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1485{
1486 int i;
1487
1488 scm->fp = UNIXCB(skb).fp;
1489 UNIXCB(skb).fp = NULL;
1490
1491 for (i = scm->fp->count-1; i >= 0; i--)
1492 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1493}
1494
1495static void unix_destruct_scm(struct sk_buff *skb)
1496{
1497 struct scm_cookie scm;
1498 memset(&scm, 0, sizeof(scm));
1499 scm.pid = UNIXCB(skb).pid;
1500 if (UNIXCB(skb).fp)
1501 unix_detach_fds(&scm, skb);
1502
1503 /* Alas, it calls VFS */
1504 /* So fscking what? fput() had been SMP-safe since the last Summer */
1505 scm_destroy(&scm);
1506 sock_wfree(skb);
1507}
1508
1509/*
1510 * The "user->unix_inflight" variable is protected by the garbage
1511 * collection lock, and we just read it locklessly here. If you go
1512 * over the limit, there might be a tiny race in actually noticing
1513 * it across threads. Tough.
1514 */
1515static inline bool too_many_unix_fds(struct task_struct *p)
1516{
1517 struct user_struct *user = current_user();
1518
1519 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1520 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1521 return false;
1522}
1523
1524#define MAX_RECURSION_LEVEL 4
1525
1526static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1527{
1528 int i;
1529 unsigned char max_level = 0;
1530
1531 if (too_many_unix_fds(current))
1532 return -ETOOMANYREFS;
1533
1534 for (i = scm->fp->count - 1; i >= 0; i--) {
1535 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1536
1537 if (sk)
1538 max_level = max(max_level,
1539 unix_sk(sk)->recursion_level);
1540 }
1541 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1542 return -ETOOMANYREFS;
1543
1544 /*
1545 * Need to duplicate file references for the sake of garbage
1546 * collection. Otherwise a socket in the fps might become a
1547 * candidate for GC while the skb is not yet queued.
1548 */
1549 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1550 if (!UNIXCB(skb).fp)
1551 return -ENOMEM;
1552
1553 for (i = scm->fp->count - 1; i >= 0; i--)
1554 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1555 return max_level;
1556}
1557
1558static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1559{
1560 int err = 0;
1561
1562 UNIXCB(skb).pid = get_pid(scm->pid);
1563 UNIXCB(skb).uid = scm->creds.uid;
1564 UNIXCB(skb).gid = scm->creds.gid;
1565 UNIXCB(skb).fp = NULL;
1566 unix_get_secdata(scm, skb);
1567 if (scm->fp && send_fds)
1568 err = unix_attach_fds(scm, skb);
1569
1570 skb->destructor = unix_destruct_scm;
1571 return err;
1572}
1573
1574static bool unix_passcred_enabled(const struct socket *sock,
1575 const struct sock *other)
1576{
1577 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1578 !other->sk_socket ||
1579 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1580}
1581
1582/*
1583 * Some apps rely on write() giving SCM_CREDENTIALS
1584 * We include credentials if source or destination socket
1585 * asserted SOCK_PASSCRED.
1586 */
1587static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1588 const struct sock *other)
1589{
1590 if (UNIXCB(skb).pid)
1591 return;
1592 if (unix_passcred_enabled(sock, other)) {
1593 UNIXCB(skb).pid = get_pid(task_tgid(current));
1594 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1595 }
1596}
1597
1598static int maybe_init_creds(struct scm_cookie *scm,
1599 struct socket *socket,
1600 const struct sock *other)
1601{
1602 int err;
1603 struct msghdr msg = { .msg_controllen = 0 };
1604
1605 err = scm_send(socket, &msg, scm, false);
1606 if (err)
1607 return err;
1608
1609 if (unix_passcred_enabled(socket, other)) {
1610 scm->pid = get_pid(task_tgid(current));
1611 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1612 }
1613 return err;
1614}
1615
1616static bool unix_skb_scm_eq(struct sk_buff *skb,
1617 struct scm_cookie *scm)
1618{
1619 const struct unix_skb_parms *u = &UNIXCB(skb);
1620
1621 return u->pid == scm->pid &&
1622 uid_eq(u->uid, scm->creds.uid) &&
1623 gid_eq(u->gid, scm->creds.gid) &&
1624 unix_secdata_eq(scm, skb);
1625}
1626
1627/*
1628 * Send AF_UNIX data.
1629 */
1630
1631static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1632 size_t len)
1633{
1634 struct sock *sk = sock->sk;
1635 struct net *net = sock_net(sk);
1636 struct unix_sock *u = unix_sk(sk);
1637 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1638 struct sock *other = NULL;
1639 int namelen = 0; /* fake GCC */
1640 int err;
1641 unsigned int hash;
1642 struct sk_buff *skb;
1643 long timeo;
1644 struct scm_cookie scm;
1645 int max_level;
1646 int data_len = 0;
1647 int sk_locked;
1648
1649 wait_for_unix_gc();
1650 err = scm_send(sock, msg, &scm, false);
1651 if (err < 0)
1652 return err;
1653
1654 err = -EOPNOTSUPP;
1655 if (msg->msg_flags&MSG_OOB)
1656 goto out;
1657
1658 if (msg->msg_namelen) {
1659 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1660 if (err < 0)
1661 goto out;
1662 namelen = err;
1663 } else {
1664 sunaddr = NULL;
1665 err = -ENOTCONN;
1666 other = unix_peer_get(sk);
1667 if (!other)
1668 goto out;
1669 }
1670
1671 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1672 && (err = unix_autobind(sock)) != 0)
1673 goto out;
1674
1675 err = -EMSGSIZE;
1676 if (len > sk->sk_sndbuf - 32)
1677 goto out;
1678
1679 if (len > SKB_MAX_ALLOC) {
1680 data_len = min_t(size_t,
1681 len - SKB_MAX_ALLOC,
1682 MAX_SKB_FRAGS * PAGE_SIZE);
1683 data_len = PAGE_ALIGN(data_len);
1684
1685 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1686 }
1687
1688 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1689 msg->msg_flags & MSG_DONTWAIT, &err,
1690 PAGE_ALLOC_COSTLY_ORDER);
1691 if (skb == NULL)
1692 goto out;
1693
1694 err = unix_scm_to_skb(&scm, skb, true);
1695 if (err < 0)
1696 goto out_free;
1697 max_level = err + 1;
1698
1699 skb_put(skb, len - data_len);
1700 skb->data_len = data_len;
1701 skb->len = len;
1702 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1703 if (err)
1704 goto out_free;
1705
1706 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1707
1708restart:
1709 if (!other) {
1710 err = -ECONNRESET;
1711 if (sunaddr == NULL)
1712 goto out_free;
1713
1714 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1715 hash, &err);
1716 if (other == NULL)
1717 goto out_free;
1718 }
1719
1720 if (sk_filter(other, skb) < 0) {
1721 /* Toss the packet but do not return any error to the sender */
1722 err = len;
1723 goto out_free;
1724 }
1725
1726 sk_locked = 0;
1727 unix_state_lock(other);
1728restart_locked:
1729 err = -EPERM;
1730 if (!unix_may_send(sk, other))
1731 goto out_unlock;
1732
1733 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1734 /*
1735 * Check with 1003.1g - what should
1736 * datagram error
1737 */
1738 unix_state_unlock(other);
1739 sock_put(other);
1740
1741 if (!sk_locked)
1742 unix_state_lock(sk);
1743
1744 err = 0;
1745 if (unix_peer(sk) == other) {
1746 unix_peer(sk) = NULL;
1747 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1748
1749 unix_state_unlock(sk);
1750
1751 unix_dgram_disconnected(sk, other);
1752 sock_put(other);
1753 err = -ECONNREFUSED;
1754 } else {
1755 unix_state_unlock(sk);
1756 }
1757
1758 other = NULL;
1759 if (err)
1760 goto out_free;
1761 goto restart;
1762 }
1763
1764 err = -EPIPE;
1765 if (other->sk_shutdown & RCV_SHUTDOWN)
1766 goto out_unlock;
1767
1768 if (sk->sk_type != SOCK_SEQPACKET) {
1769 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1770 if (err)
1771 goto out_unlock;
1772 }
1773
1774 /* other == sk && unix_peer(other) != sk if
1775 * - unix_peer(sk) == NULL, destination address bound to sk
1776 * - unix_peer(sk) == sk by time of get but disconnected before lock
1777 */
1778 if (other != sk &&
1779 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1780 if (timeo) {
1781 timeo = unix_wait_for_peer(other, timeo);
1782
1783 err = sock_intr_errno(timeo);
1784 if (signal_pending(current))
1785 goto out_free;
1786
1787 goto restart;
1788 }
1789
1790 if (!sk_locked) {
1791 unix_state_unlock(other);
1792 unix_state_double_lock(sk, other);
1793 }
1794
1795 if (unix_peer(sk) != other ||
1796 unix_dgram_peer_wake_me(sk, other)) {
1797 err = -EAGAIN;
1798 sk_locked = 1;
1799 goto out_unlock;
1800 }
1801
1802 if (!sk_locked) {
1803 sk_locked = 1;
1804 goto restart_locked;
1805 }
1806 }
1807
1808 if (unlikely(sk_locked))
1809 unix_state_unlock(sk);
1810
1811 if (sock_flag(other, SOCK_RCVTSTAMP))
1812 __net_timestamp(skb);
1813 maybe_add_creds(skb, sock, other);
1814 skb_queue_tail(&other->sk_receive_queue, skb);
1815 if (max_level > unix_sk(other)->recursion_level)
1816 unix_sk(other)->recursion_level = max_level;
1817 unix_state_unlock(other);
1818 other->sk_data_ready(other);
1819 sock_put(other);
1820 scm_destroy(&scm);
1821 return len;
1822
1823out_unlock:
1824 if (sk_locked)
1825 unix_state_unlock(sk);
1826 unix_state_unlock(other);
1827out_free:
1828 kfree_skb(skb);
1829out:
1830 if (other)
1831 sock_put(other);
1832 scm_destroy(&scm);
1833 return err;
1834}
1835
1836/* We use paged skbs for stream sockets, and limit occupancy to 32768
1837 * bytes, and a minimun of a full page.
1838 */
1839#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1840
1841static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1842 size_t len)
1843{
1844 struct sock *sk = sock->sk;
1845 struct sock *other = NULL;
1846 int err, size;
1847 struct sk_buff *skb;
1848 int sent = 0;
1849 struct scm_cookie scm;
1850 bool fds_sent = false;
1851 int max_level;
1852 int data_len;
1853
1854 wait_for_unix_gc();
1855 err = scm_send(sock, msg, &scm, false);
1856 if (err < 0)
1857 return err;
1858
1859 err = -EOPNOTSUPP;
1860 if (msg->msg_flags&MSG_OOB)
1861 goto out_err;
1862
1863 if (msg->msg_namelen) {
1864 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1865 goto out_err;
1866 } else {
1867 err = -ENOTCONN;
1868 other = unix_peer(sk);
1869 if (!other)
1870 goto out_err;
1871 }
1872
1873 if (sk->sk_shutdown & SEND_SHUTDOWN)
1874 goto pipe_err;
1875
1876 while (sent < len) {
1877 size = len - sent;
1878
1879 /* Keep two messages in the pipe so it schedules better */
1880 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1881
1882 /* allow fallback to order-0 allocations */
1883 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1884
1885 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1886
1887 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1888
1889 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1890 msg->msg_flags & MSG_DONTWAIT, &err,
1891 get_order(UNIX_SKB_FRAGS_SZ));
1892 if (!skb)
1893 goto out_err;
1894
1895 /* Only send the fds in the first buffer */
1896 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1897 if (err < 0) {
1898 kfree_skb(skb);
1899 goto out_err;
1900 }
1901 max_level = err + 1;
1902 fds_sent = true;
1903
1904 skb_put(skb, size - data_len);
1905 skb->data_len = data_len;
1906 skb->len = size;
1907 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1908 if (err) {
1909 kfree_skb(skb);
1910 goto out_err;
1911 }
1912
1913 unix_state_lock(other);
1914
1915 if (sock_flag(other, SOCK_DEAD) ||
1916 (other->sk_shutdown & RCV_SHUTDOWN))
1917 goto pipe_err_free;
1918
1919 maybe_add_creds(skb, sock, other);
1920 skb_queue_tail(&other->sk_receive_queue, skb);
1921 if (max_level > unix_sk(other)->recursion_level)
1922 unix_sk(other)->recursion_level = max_level;
1923 unix_state_unlock(other);
1924 other->sk_data_ready(other);
1925 sent += size;
1926 }
1927
1928 scm_destroy(&scm);
1929
1930 return sent;
1931
1932pipe_err_free:
1933 unix_state_unlock(other);
1934 kfree_skb(skb);
1935pipe_err:
1936 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1937 send_sig(SIGPIPE, current, 0);
1938 err = -EPIPE;
1939out_err:
1940 scm_destroy(&scm);
1941 return sent ? : err;
1942}
1943
1944static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1945 int offset, size_t size, int flags)
1946{
1947 int err;
1948 bool send_sigpipe = false;
1949 bool init_scm = true;
1950 struct scm_cookie scm;
1951 struct sock *other, *sk = socket->sk;
1952 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1953
1954 if (flags & MSG_OOB)
1955 return -EOPNOTSUPP;
1956
1957 other = unix_peer(sk);
1958 if (!other || sk->sk_state != TCP_ESTABLISHED)
1959 return -ENOTCONN;
1960
1961 if (false) {
1962alloc_skb:
1963 unix_state_unlock(other);
1964 mutex_unlock(&unix_sk(other)->iolock);
1965 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1966 &err, 0);
1967 if (!newskb)
1968 goto err;
1969 }
1970
1971 /* we must acquire iolock as we modify already present
1972 * skbs in the sk_receive_queue and mess with skb->len
1973 */
1974 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1975 if (err) {
1976 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1977 goto err;
1978 }
1979
1980 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1981 err = -EPIPE;
1982 send_sigpipe = true;
1983 goto err_unlock;
1984 }
1985
1986 unix_state_lock(other);
1987
1988 if (sock_flag(other, SOCK_DEAD) ||
1989 other->sk_shutdown & RCV_SHUTDOWN) {
1990 err = -EPIPE;
1991 send_sigpipe = true;
1992 goto err_state_unlock;
1993 }
1994
1995 if (init_scm) {
1996 err = maybe_init_creds(&scm, socket, other);
1997 if (err)
1998 goto err_state_unlock;
1999 init_scm = false;
2000 }
2001
2002 skb = skb_peek_tail(&other->sk_receive_queue);
2003 if (tail && tail == skb) {
2004 skb = newskb;
2005 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2006 if (newskb) {
2007 skb = newskb;
2008 } else {
2009 tail = skb;
2010 goto alloc_skb;
2011 }
2012 } else if (newskb) {
2013 /* this is fast path, we don't necessarily need to
2014 * call to kfree_skb even though with newskb == NULL
2015 * this - does no harm
2016 */
2017 consume_skb(newskb);
2018 newskb = NULL;
2019 }
2020
2021 if (skb_append_pagefrags(skb, page, offset, size)) {
2022 tail = skb;
2023 goto alloc_skb;
2024 }
2025
2026 skb->len += size;
2027 skb->data_len += size;
2028 skb->truesize += size;
2029 atomic_add(size, &sk->sk_wmem_alloc);
2030
2031 if (newskb) {
2032 err = unix_scm_to_skb(&scm, skb, false);
2033 if (err)
2034 goto err_state_unlock;
2035 spin_lock(&other->sk_receive_queue.lock);
2036 __skb_queue_tail(&other->sk_receive_queue, newskb);
2037 spin_unlock(&other->sk_receive_queue.lock);
2038 }
2039
2040 unix_state_unlock(other);
2041 mutex_unlock(&unix_sk(other)->iolock);
2042
2043 other->sk_data_ready(other);
2044 scm_destroy(&scm);
2045 return size;
2046
2047err_state_unlock:
2048 unix_state_unlock(other);
2049err_unlock:
2050 mutex_unlock(&unix_sk(other)->iolock);
2051err:
2052 kfree_skb(newskb);
2053 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2054 send_sig(SIGPIPE, current, 0);
2055 if (!init_scm)
2056 scm_destroy(&scm);
2057 return err;
2058}
2059
2060static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2061 size_t len)
2062{
2063 int err;
2064 struct sock *sk = sock->sk;
2065
2066 err = sock_error(sk);
2067 if (err)
2068 return err;
2069
2070 if (sk->sk_state != TCP_ESTABLISHED)
2071 return -ENOTCONN;
2072
2073 if (msg->msg_namelen)
2074 msg->msg_namelen = 0;
2075
2076 return unix_dgram_sendmsg(sock, msg, len);
2077}
2078
2079static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2080 size_t size, int flags)
2081{
2082 struct sock *sk = sock->sk;
2083
2084 if (sk->sk_state != TCP_ESTABLISHED)
2085 return -ENOTCONN;
2086
2087 return unix_dgram_recvmsg(sock, msg, size, flags);
2088}
2089
2090static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2091{
2092 struct unix_sock *u = unix_sk(sk);
2093
2094 if (u->addr) {
2095 msg->msg_namelen = u->addr->len;
2096 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2097 }
2098}
2099
2100static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2101 size_t size, int flags)
2102{
2103 struct scm_cookie scm;
2104 struct sock *sk = sock->sk;
2105 struct unix_sock *u = unix_sk(sk);
2106 struct sk_buff *skb, *last;
2107 long timeo;
2108 int err;
2109 int peeked, skip;
2110
2111 err = -EOPNOTSUPP;
2112 if (flags&MSG_OOB)
2113 goto out;
2114
2115 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2116
2117 do {
2118 mutex_lock(&u->iolock);
2119
2120 skip = sk_peek_offset(sk, flags);
2121 skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2122 &err, &last);
2123 if (skb)
2124 break;
2125
2126 mutex_unlock(&u->iolock);
2127
2128 if (err != -EAGAIN)
2129 break;
2130 } while (timeo &&
2131 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2132
2133 if (!skb) { /* implies iolock unlocked */
2134 unix_state_lock(sk);
2135 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2136 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2137 (sk->sk_shutdown & RCV_SHUTDOWN))
2138 err = 0;
2139 unix_state_unlock(sk);
2140 goto out;
2141 }
2142
2143 if (wq_has_sleeper(&u->peer_wait))
2144 wake_up_interruptible_sync_poll(&u->peer_wait,
2145 POLLOUT | POLLWRNORM |
2146 POLLWRBAND);
2147
2148 if (msg->msg_name)
2149 unix_copy_addr(msg, skb->sk);
2150
2151 if (size > skb->len - skip)
2152 size = skb->len - skip;
2153 else if (size < skb->len - skip)
2154 msg->msg_flags |= MSG_TRUNC;
2155
2156 err = skb_copy_datagram_msg(skb, skip, msg, size);
2157 if (err)
2158 goto out_free;
2159
2160 if (sock_flag(sk, SOCK_RCVTSTAMP))
2161 __sock_recv_timestamp(msg, sk, skb);
2162
2163 memset(&scm, 0, sizeof(scm));
2164
2165 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2166 unix_set_secdata(&scm, skb);
2167
2168 if (!(flags & MSG_PEEK)) {
2169 if (UNIXCB(skb).fp)
2170 unix_detach_fds(&scm, skb);
2171
2172 sk_peek_offset_bwd(sk, skb->len);
2173 } else {
2174 /* It is questionable: on PEEK we could:
2175 - do not return fds - good, but too simple 8)
2176 - return fds, and do not return them on read (old strategy,
2177 apparently wrong)
2178 - clone fds (I chose it for now, it is the most universal
2179 solution)
2180
2181 POSIX 1003.1g does not actually define this clearly
2182 at all. POSIX 1003.1g doesn't define a lot of things
2183 clearly however!
2184
2185 */
2186
2187 sk_peek_offset_fwd(sk, size);
2188
2189 if (UNIXCB(skb).fp)
2190 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2191 }
2192 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2193
2194 scm_recv(sock, msg, &scm, flags);
2195
2196out_free:
2197 skb_free_datagram(sk, skb);
2198 mutex_unlock(&u->iolock);
2199out:
2200 return err;
2201}
2202
2203/*
2204 * Sleep until more data has arrived. But check for races..
2205 */
2206static long unix_stream_data_wait(struct sock *sk, long timeo,
2207 struct sk_buff *last, unsigned int last_len,
2208 bool freezable)
2209{
2210 struct sk_buff *tail;
2211 DEFINE_WAIT(wait);
2212
2213 unix_state_lock(sk);
2214
2215 for (;;) {
2216 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2217
2218 tail = skb_peek_tail(&sk->sk_receive_queue);
2219 if (tail != last ||
2220 (tail && tail->len != last_len) ||
2221 sk->sk_err ||
2222 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2223 signal_pending(current) ||
2224 !timeo)
2225 break;
2226
2227 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2228 unix_state_unlock(sk);
2229 if (freezable)
2230 timeo = freezable_schedule_timeout(timeo);
2231 else
2232 timeo = schedule_timeout(timeo);
2233 unix_state_lock(sk);
2234
2235 if (sock_flag(sk, SOCK_DEAD))
2236 break;
2237
2238 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2239 }
2240
2241 finish_wait(sk_sleep(sk), &wait);
2242 unix_state_unlock(sk);
2243 return timeo;
2244}
2245
2246static unsigned int unix_skb_len(const struct sk_buff *skb)
2247{
2248 return skb->len - UNIXCB(skb).consumed;
2249}
2250
2251struct unix_stream_read_state {
2252 int (*recv_actor)(struct sk_buff *, int, int,
2253 struct unix_stream_read_state *);
2254 struct socket *socket;
2255 struct msghdr *msg;
2256 struct pipe_inode_info *pipe;
2257 size_t size;
2258 int flags;
2259 unsigned int splice_flags;
2260};
2261
2262static int unix_stream_read_generic(struct unix_stream_read_state *state,
2263 bool freezable)
2264{
2265 struct scm_cookie scm;
2266 struct socket *sock = state->socket;
2267 struct sock *sk = sock->sk;
2268 struct unix_sock *u = unix_sk(sk);
2269 int copied = 0;
2270 int flags = state->flags;
2271 int noblock = flags & MSG_DONTWAIT;
2272 bool check_creds = false;
2273 int target;
2274 int err = 0;
2275 long timeo;
2276 int skip;
2277 size_t size = state->size;
2278 unsigned int last_len;
2279
2280 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2281 err = -EINVAL;
2282 goto out;
2283 }
2284
2285 if (unlikely(flags & MSG_OOB)) {
2286 err = -EOPNOTSUPP;
2287 goto out;
2288 }
2289
2290 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2291 timeo = sock_rcvtimeo(sk, noblock);
2292
2293 memset(&scm, 0, sizeof(scm));
2294
2295 /* Lock the socket to prevent queue disordering
2296 * while sleeps in memcpy_tomsg
2297 */
2298 mutex_lock(&u->iolock);
2299
2300 if (flags & MSG_PEEK)
2301 skip = sk_peek_offset(sk, flags);
2302 else
2303 skip = 0;
2304
2305 do {
2306 int chunk;
2307 bool drop_skb;
2308 struct sk_buff *skb, *last;
2309
2310redo:
2311 unix_state_lock(sk);
2312 if (sock_flag(sk, SOCK_DEAD)) {
2313 err = -ECONNRESET;
2314 goto unlock;
2315 }
2316 last = skb = skb_peek(&sk->sk_receive_queue);
2317 last_len = last ? last->len : 0;
2318again:
2319 if (skb == NULL) {
2320 unix_sk(sk)->recursion_level = 0;
2321 if (copied >= target)
2322 goto unlock;
2323
2324 /*
2325 * POSIX 1003.1g mandates this order.
2326 */
2327
2328 err = sock_error(sk);
2329 if (err)
2330 goto unlock;
2331 if (sk->sk_shutdown & RCV_SHUTDOWN)
2332 goto unlock;
2333
2334 unix_state_unlock(sk);
2335 if (!timeo) {
2336 err = -EAGAIN;
2337 break;
2338 }
2339
2340 mutex_unlock(&u->iolock);
2341
2342 timeo = unix_stream_data_wait(sk, timeo, last,
2343 last_len, freezable);
2344
2345 if (signal_pending(current)) {
2346 err = sock_intr_errno(timeo);
2347 scm_destroy(&scm);
2348 goto out;
2349 }
2350
2351 mutex_lock(&u->iolock);
2352 goto redo;
2353unlock:
2354 unix_state_unlock(sk);
2355 break;
2356 }
2357
2358 while (skip >= unix_skb_len(skb)) {
2359 skip -= unix_skb_len(skb);
2360 last = skb;
2361 last_len = skb->len;
2362 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2363 if (!skb)
2364 goto again;
2365 }
2366
2367 unix_state_unlock(sk);
2368
2369 if (check_creds) {
2370 /* Never glue messages from different writers */
2371 if (!unix_skb_scm_eq(skb, &scm))
2372 break;
2373 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2374 /* Copy credentials */
2375 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2376 unix_set_secdata(&scm, skb);
2377 check_creds = true;
2378 }
2379
2380 /* Copy address just once */
2381 if (state->msg && state->msg->msg_name) {
2382 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2383 state->msg->msg_name);
2384 unix_copy_addr(state->msg, skb->sk);
2385 sunaddr = NULL;
2386 }
2387
2388 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2389 skb_get(skb);
2390 chunk = state->recv_actor(skb, skip, chunk, state);
2391 drop_skb = !unix_skb_len(skb);
2392 /* skb is only safe to use if !drop_skb */
2393 consume_skb(skb);
2394 if (chunk < 0) {
2395 if (copied == 0)
2396 copied = -EFAULT;
2397 break;
2398 }
2399 copied += chunk;
2400 size -= chunk;
2401
2402 if (drop_skb) {
2403 /* the skb was touched by a concurrent reader;
2404 * we should not expect anything from this skb
2405 * anymore and assume it invalid - we can be
2406 * sure it was dropped from the socket queue
2407 *
2408 * let's report a short read
2409 */
2410 err = 0;
2411 break;
2412 }
2413
2414 /* Mark read part of skb as used */
2415 if (!(flags & MSG_PEEK)) {
2416 UNIXCB(skb).consumed += chunk;
2417
2418 sk_peek_offset_bwd(sk, chunk);
2419
2420 if (UNIXCB(skb).fp)
2421 unix_detach_fds(&scm, skb);
2422
2423 if (unix_skb_len(skb))
2424 break;
2425
2426 skb_unlink(skb, &sk->sk_receive_queue);
2427 consume_skb(skb);
2428
2429 if (scm.fp)
2430 break;
2431 } else {
2432 /* It is questionable, see note in unix_dgram_recvmsg.
2433 */
2434 if (UNIXCB(skb).fp)
2435 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2436
2437 sk_peek_offset_fwd(sk, chunk);
2438
2439 if (UNIXCB(skb).fp)
2440 break;
2441
2442 skip = 0;
2443 last = skb;
2444 last_len = skb->len;
2445 unix_state_lock(sk);
2446 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2447 if (skb)
2448 goto again;
2449 unix_state_unlock(sk);
2450 break;
2451 }
2452 } while (size);
2453
2454 mutex_unlock(&u->iolock);
2455 if (state->msg)
2456 scm_recv(sock, state->msg, &scm, flags);
2457 else
2458 scm_destroy(&scm);
2459out:
2460 return copied ? : err;
2461}
2462
2463static int unix_stream_read_actor(struct sk_buff *skb,
2464 int skip, int chunk,
2465 struct unix_stream_read_state *state)
2466{
2467 int ret;
2468
2469 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2470 state->msg, chunk);
2471 return ret ?: chunk;
2472}
2473
2474static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2475 size_t size, int flags)
2476{
2477 struct unix_stream_read_state state = {
2478 .recv_actor = unix_stream_read_actor,
2479 .socket = sock,
2480 .msg = msg,
2481 .size = size,
2482 .flags = flags
2483 };
2484
2485 return unix_stream_read_generic(&state, true);
2486}
2487
2488static int unix_stream_splice_actor(struct sk_buff *skb,
2489 int skip, int chunk,
2490 struct unix_stream_read_state *state)
2491{
2492 return skb_splice_bits(skb, state->socket->sk,
2493 UNIXCB(skb).consumed + skip,
2494 state->pipe, chunk, state->splice_flags);
2495}
2496
2497static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2498 struct pipe_inode_info *pipe,
2499 size_t size, unsigned int flags)
2500{
2501 struct unix_stream_read_state state = {
2502 .recv_actor = unix_stream_splice_actor,
2503 .socket = sock,
2504 .pipe = pipe,
2505 .size = size,
2506 .splice_flags = flags,
2507 };
2508
2509 if (unlikely(*ppos))
2510 return -ESPIPE;
2511
2512 if (sock->file->f_flags & O_NONBLOCK ||
2513 flags & SPLICE_F_NONBLOCK)
2514 state.flags = MSG_DONTWAIT;
2515
2516 return unix_stream_read_generic(&state, false);
2517}
2518
2519static int unix_shutdown(struct socket *sock, int mode)
2520{
2521 struct sock *sk = sock->sk;
2522 struct sock *other;
2523
2524 if (mode < SHUT_RD || mode > SHUT_RDWR)
2525 return -EINVAL;
2526 /* This maps:
2527 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2528 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2529 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2530 */
2531 ++mode;
2532
2533 unix_state_lock(sk);
2534 sk->sk_shutdown |= mode;
2535 other = unix_peer(sk);
2536 if (other)
2537 sock_hold(other);
2538 unix_state_unlock(sk);
2539 sk->sk_state_change(sk);
2540
2541 if (other &&
2542 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2543
2544 int peer_mode = 0;
2545
2546 if (mode&RCV_SHUTDOWN)
2547 peer_mode |= SEND_SHUTDOWN;
2548 if (mode&SEND_SHUTDOWN)
2549 peer_mode |= RCV_SHUTDOWN;
2550 unix_state_lock(other);
2551 other->sk_shutdown |= peer_mode;
2552 unix_state_unlock(other);
2553 other->sk_state_change(other);
2554 if (peer_mode == SHUTDOWN_MASK)
2555 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2556 else if (peer_mode & RCV_SHUTDOWN)
2557 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2558 }
2559 if (other)
2560 sock_put(other);
2561
2562 return 0;
2563}
2564
2565long unix_inq_len(struct sock *sk)
2566{
2567 struct sk_buff *skb;
2568 long amount = 0;
2569
2570 if (sk->sk_state == TCP_LISTEN)
2571 return -EINVAL;
2572
2573 spin_lock(&sk->sk_receive_queue.lock);
2574 if (sk->sk_type == SOCK_STREAM ||
2575 sk->sk_type == SOCK_SEQPACKET) {
2576 skb_queue_walk(&sk->sk_receive_queue, skb)
2577 amount += unix_skb_len(skb);
2578 } else {
2579 skb = skb_peek(&sk->sk_receive_queue);
2580 if (skb)
2581 amount = skb->len;
2582 }
2583 spin_unlock(&sk->sk_receive_queue.lock);
2584
2585 return amount;
2586}
2587EXPORT_SYMBOL_GPL(unix_inq_len);
2588
2589long unix_outq_len(struct sock *sk)
2590{
2591 return sk_wmem_alloc_get(sk);
2592}
2593EXPORT_SYMBOL_GPL(unix_outq_len);
2594
2595static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2596{
2597 struct sock *sk = sock->sk;
2598 long amount = 0;
2599 int err;
2600
2601 switch (cmd) {
2602 case SIOCOUTQ:
2603 amount = unix_outq_len(sk);
2604 err = put_user(amount, (int __user *)arg);
2605 break;
2606 case SIOCINQ:
2607 amount = unix_inq_len(sk);
2608 if (amount < 0)
2609 err = amount;
2610 else
2611 err = put_user(amount, (int __user *)arg);
2612 break;
2613 default:
2614 err = -ENOIOCTLCMD;
2615 break;
2616 }
2617 return err;
2618}
2619
2620static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2621{
2622 struct sock *sk = sock->sk;
2623 unsigned int mask;
2624
2625 sock_poll_wait(file, sk_sleep(sk), wait);
2626 mask = 0;
2627
2628 /* exceptional events? */
2629 if (sk->sk_err)
2630 mask |= POLLERR;
2631 if (sk->sk_shutdown == SHUTDOWN_MASK)
2632 mask |= POLLHUP;
2633 if (sk->sk_shutdown & RCV_SHUTDOWN)
2634 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2635
2636 /* readable? */
2637 if (!skb_queue_empty(&sk->sk_receive_queue))
2638 mask |= POLLIN | POLLRDNORM;
2639
2640 /* Connection-based need to check for termination and startup */
2641 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2642 sk->sk_state == TCP_CLOSE)
2643 mask |= POLLHUP;
2644
2645 /*
2646 * we set writable also when the other side has shut down the
2647 * connection. This prevents stuck sockets.
2648 */
2649 if (unix_writable(sk))
2650 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2651
2652 return mask;
2653}
2654
2655static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2656 poll_table *wait)
2657{
2658 struct sock *sk = sock->sk, *other;
2659 unsigned int mask, writable;
2660
2661 sock_poll_wait(file, sk_sleep(sk), wait);
2662 mask = 0;
2663
2664 /* exceptional events? */
2665 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2666 mask |= POLLERR |
2667 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2668
2669 if (sk->sk_shutdown & RCV_SHUTDOWN)
2670 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2671 if (sk->sk_shutdown == SHUTDOWN_MASK)
2672 mask |= POLLHUP;
2673
2674 /* readable? */
2675 if (!skb_queue_empty(&sk->sk_receive_queue))
2676 mask |= POLLIN | POLLRDNORM;
2677
2678 /* Connection-based need to check for termination and startup */
2679 if (sk->sk_type == SOCK_SEQPACKET) {
2680 if (sk->sk_state == TCP_CLOSE)
2681 mask |= POLLHUP;
2682 /* connection hasn't started yet? */
2683 if (sk->sk_state == TCP_SYN_SENT)
2684 return mask;
2685 }
2686
2687 /* No write status requested, avoid expensive OUT tests. */
2688 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2689 return mask;
2690
2691 writable = unix_writable(sk);
2692 if (writable) {
2693 unix_state_lock(sk);
2694
2695 other = unix_peer(sk);
2696 if (other && unix_peer(other) != sk &&
2697 unix_recvq_full(other) &&
2698 unix_dgram_peer_wake_me(sk, other))
2699 writable = 0;
2700
2701 unix_state_unlock(sk);
2702 }
2703
2704 if (writable)
2705 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2706 else
2707 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2708
2709 return mask;
2710}
2711
2712#ifdef CONFIG_PROC_FS
2713
2714#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2715
2716#define get_bucket(x) ((x) >> BUCKET_SPACE)
2717#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2718#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2719
2720static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2721{
2722 unsigned long offset = get_offset(*pos);
2723 unsigned long bucket = get_bucket(*pos);
2724 struct sock *sk;
2725 unsigned long count = 0;
2726
2727 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2728 if (sock_net(sk) != seq_file_net(seq))
2729 continue;
2730 if (++count == offset)
2731 break;
2732 }
2733
2734 return sk;
2735}
2736
2737static struct sock *unix_next_socket(struct seq_file *seq,
2738 struct sock *sk,
2739 loff_t *pos)
2740{
2741 unsigned long bucket;
2742
2743 while (sk > (struct sock *)SEQ_START_TOKEN) {
2744 sk = sk_next(sk);
2745 if (!sk)
2746 goto next_bucket;
2747 if (sock_net(sk) == seq_file_net(seq))
2748 return sk;
2749 }
2750
2751 do {
2752 sk = unix_from_bucket(seq, pos);
2753 if (sk)
2754 return sk;
2755
2756next_bucket:
2757 bucket = get_bucket(*pos) + 1;
2758 *pos = set_bucket_offset(bucket, 1);
2759 } while (bucket < ARRAY_SIZE(unix_socket_table));
2760
2761 return NULL;
2762}
2763
2764static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2765 __acquires(unix_table_lock)
2766{
2767 spin_lock(&unix_table_lock);
2768
2769 if (!*pos)
2770 return SEQ_START_TOKEN;
2771
2772 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2773 return NULL;
2774
2775 return unix_next_socket(seq, NULL, pos);
2776}
2777
2778static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2779{
2780 ++*pos;
2781 return unix_next_socket(seq, v, pos);
2782}
2783
2784static void unix_seq_stop(struct seq_file *seq, void *v)
2785 __releases(unix_table_lock)
2786{
2787 spin_unlock(&unix_table_lock);
2788}
2789
2790static int unix_seq_show(struct seq_file *seq, void *v)
2791{
2792
2793 if (v == SEQ_START_TOKEN)
2794 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2795 "Inode Path\n");
2796 else {
2797 struct sock *s = v;
2798 struct unix_sock *u = unix_sk(s);
2799 unix_state_lock(s);
2800
2801 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2802 s,
2803 atomic_read(&s->sk_refcnt),
2804 0,
2805 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2806 s->sk_type,
2807 s->sk_socket ?
2808 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2809 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2810 sock_i_ino(s));
2811
2812 if (u->addr) {
2813 int i, len;
2814 seq_putc(seq, ' ');
2815
2816 i = 0;
2817 len = u->addr->len - sizeof(short);
2818 if (!UNIX_ABSTRACT(s))
2819 len--;
2820 else {
2821 seq_putc(seq, '@');
2822 i++;
2823 }
2824 for ( ; i < len; i++)
2825 seq_putc(seq, u->addr->name->sun_path[i] ?:
2826 '@');
2827 }
2828 unix_state_unlock(s);
2829 seq_putc(seq, '\n');
2830 }
2831
2832 return 0;
2833}
2834
2835static const struct seq_operations unix_seq_ops = {
2836 .start = unix_seq_start,
2837 .next = unix_seq_next,
2838 .stop = unix_seq_stop,
2839 .show = unix_seq_show,
2840};
2841
2842static int unix_seq_open(struct inode *inode, struct file *file)
2843{
2844 return seq_open_net(inode, file, &unix_seq_ops,
2845 sizeof(struct seq_net_private));
2846}
2847
2848static const struct file_operations unix_seq_fops = {
2849 .owner = THIS_MODULE,
2850 .open = unix_seq_open,
2851 .read = seq_read,
2852 .llseek = seq_lseek,
2853 .release = seq_release_net,
2854};
2855
2856#endif
2857
2858static const struct net_proto_family unix_family_ops = {
2859 .family = PF_UNIX,
2860 .create = unix_create,
2861 .owner = THIS_MODULE,
2862};
2863
2864
2865static int __net_init unix_net_init(struct net *net)
2866{
2867 int error = -ENOMEM;
2868
2869 net->unx.sysctl_max_dgram_qlen = 10;
2870 if (unix_sysctl_register(net))
2871 goto out;
2872
2873#ifdef CONFIG_PROC_FS
2874 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2875 unix_sysctl_unregister(net);
2876 goto out;
2877 }
2878#endif
2879 error = 0;
2880out:
2881 return error;
2882}
2883
2884static void __net_exit unix_net_exit(struct net *net)
2885{
2886 unix_sysctl_unregister(net);
2887 remove_proc_entry("unix", net->proc_net);
2888}
2889
2890static struct pernet_operations unix_net_ops = {
2891 .init = unix_net_init,
2892 .exit = unix_net_exit,
2893};
2894
2895static int __init af_unix_init(void)
2896{
2897 int rc = -1;
2898
2899 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2900
2901 rc = proto_register(&unix_proto, 1);
2902 if (rc != 0) {
2903 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2904 goto out;
2905 }
2906
2907 sock_register(&unix_family_ops);
2908 register_pernet_subsys(&unix_net_ops);
2909out:
2910 return rc;
2911}
2912
2913static void __exit af_unix_exit(void)
2914{
2915 sock_unregister(PF_UNIX);
2916 proto_unregister(&unix_proto);
2917 unregister_pernet_subsys(&unix_net_ops);
2918}
2919
2920/* Earlier than device_initcall() so that other drivers invoking
2921 request_module() don't end up in a loop when modprobe tries
2922 to use a UNIX socket. But later than subsys_initcall() because
2923 we depend on stuff initialised there */
2924fs_initcall(af_unix_init);
2925module_exit(af_unix_exit);
2926
2927MODULE_LICENSE("GPL");
2928MODULE_ALIAS_NETPROTO(PF_UNIX);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET4: Implementation of BSD Unix domain sockets.
4 *
5 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Fixes:
8 * Linus Torvalds : Assorted bug cures.
9 * Niibe Yutaka : async I/O support.
10 * Carsten Paeth : PF_UNIX check, address fixes.
11 * Alan Cox : Limit size of allocated blocks.
12 * Alan Cox : Fixed the stupid socketpair bug.
13 * Alan Cox : BSD compatibility fine tuning.
14 * Alan Cox : Fixed a bug in connect when interrupted.
15 * Alan Cox : Sorted out a proper draft version of
16 * file descriptor passing hacked up from
17 * Mike Shaver's work.
18 * Marty Leisner : Fixes to fd passing
19 * Nick Nevin : recvmsg bugfix.
20 * Alan Cox : Started proper garbage collector
21 * Heiko EiBfeldt : Missing verify_area check
22 * Alan Cox : Started POSIXisms
23 * Andreas Schwab : Replace inode by dentry for proper
24 * reference counting
25 * Kirk Petersen : Made this a module
26 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
27 * Lots of bug fixes.
28 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
29 * by above two patches.
30 * Andrea Arcangeli : If possible we block in connect(2)
31 * if the max backlog of the listen socket
32 * is been reached. This won't break
33 * old apps and it will avoid huge amount
34 * of socks hashed (this for unix_gc()
35 * performances reasons).
36 * Security fix that limits the max
37 * number of socks to 2*max_files and
38 * the number of skb queueable in the
39 * dgram receiver.
40 * Artur Skawina : Hash function optimizations
41 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
42 * Malcolm Beattie : Set peercred for socketpair
43 * Michal Ostrowski : Module initialization cleanup.
44 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
45 * the core infrastructure is doing that
46 * for all net proto families now (2.5.69+)
47 *
48 * Known differences from reference BSD that was tested:
49 *
50 * [TO FIX]
51 * ECONNREFUSED is not returned from one end of a connected() socket to the
52 * other the moment one end closes.
53 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
54 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
55 * [NOT TO FIX]
56 * accept() returns a path name even if the connecting socket has closed
57 * in the meantime (BSD loses the path and gives up).
58 * accept() returns 0 length path for an unbound connector. BSD returns 16
59 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
60 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
61 * BSD af_unix apparently has connect forgetting to block properly.
62 * (need to check this with the POSIX spec in detail)
63 *
64 * Differences from 2.0.0-11-... (ANK)
65 * Bug fixes and improvements.
66 * - client shutdown killed server socket.
67 * - removed all useless cli/sti pairs.
68 *
69 * Semantic changes/extensions.
70 * - generic control message passing.
71 * - SCM_CREDENTIALS control message.
72 * - "Abstract" (not FS based) socket bindings.
73 * Abstract names are sequences of bytes (not zero terminated)
74 * started by 0, so that this name space does not intersect
75 * with BSD names.
76 */
77
78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
80#include <linux/module.h>
81#include <linux/kernel.h>
82#include <linux/signal.h>
83#include <linux/sched/signal.h>
84#include <linux/errno.h>
85#include <linux/string.h>
86#include <linux/stat.h>
87#include <linux/dcache.h>
88#include <linux/namei.h>
89#include <linux/socket.h>
90#include <linux/un.h>
91#include <linux/fcntl.h>
92#include <linux/filter.h>
93#include <linux/termios.h>
94#include <linux/sockios.h>
95#include <linux/net.h>
96#include <linux/in.h>
97#include <linux/fs.h>
98#include <linux/slab.h>
99#include <linux/uaccess.h>
100#include <linux/skbuff.h>
101#include <linux/netdevice.h>
102#include <net/net_namespace.h>
103#include <net/sock.h>
104#include <net/tcp_states.h>
105#include <net/af_unix.h>
106#include <linux/proc_fs.h>
107#include <linux/seq_file.h>
108#include <net/scm.h>
109#include <linux/init.h>
110#include <linux/poll.h>
111#include <linux/rtnetlink.h>
112#include <linux/mount.h>
113#include <net/checksum.h>
114#include <linux/security.h>
115#include <linux/splice.h>
116#include <linux/freezer.h>
117#include <linux/file.h>
118#include <linux/btf_ids.h>
119#include <linux/bpf-cgroup.h>
120
121#include "scm.h"
122
123static atomic_long_t unix_nr_socks;
124static struct hlist_head bsd_socket_buckets[UNIX_HASH_SIZE / 2];
125static spinlock_t bsd_socket_locks[UNIX_HASH_SIZE / 2];
126
127/* SMP locking strategy:
128 * hash table is protected with spinlock.
129 * each socket state is protected by separate spinlock.
130 */
131
132static unsigned int unix_unbound_hash(struct sock *sk)
133{
134 unsigned long hash = (unsigned long)sk;
135
136 hash ^= hash >> 16;
137 hash ^= hash >> 8;
138 hash ^= sk->sk_type;
139
140 return hash & UNIX_HASH_MOD;
141}
142
143static unsigned int unix_bsd_hash(struct inode *i)
144{
145 return i->i_ino & UNIX_HASH_MOD;
146}
147
148static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
149 int addr_len, int type)
150{
151 __wsum csum = csum_partial(sunaddr, addr_len, 0);
152 unsigned int hash;
153
154 hash = (__force unsigned int)csum_fold(csum);
155 hash ^= hash >> 8;
156 hash ^= type;
157
158 return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
159}
160
161static void unix_table_double_lock(struct net *net,
162 unsigned int hash1, unsigned int hash2)
163{
164 if (hash1 == hash2) {
165 spin_lock(&net->unx.table.locks[hash1]);
166 return;
167 }
168
169 if (hash1 > hash2)
170 swap(hash1, hash2);
171
172 spin_lock(&net->unx.table.locks[hash1]);
173 spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
174}
175
176static void unix_table_double_unlock(struct net *net,
177 unsigned int hash1, unsigned int hash2)
178{
179 if (hash1 == hash2) {
180 spin_unlock(&net->unx.table.locks[hash1]);
181 return;
182 }
183
184 spin_unlock(&net->unx.table.locks[hash1]);
185 spin_unlock(&net->unx.table.locks[hash2]);
186}
187
188#ifdef CONFIG_SECURITY_NETWORK
189static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
190{
191 UNIXCB(skb).secid = scm->secid;
192}
193
194static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
195{
196 scm->secid = UNIXCB(skb).secid;
197}
198
199static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
200{
201 return (scm->secid == UNIXCB(skb).secid);
202}
203#else
204static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
205{ }
206
207static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
208{ }
209
210static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
211{
212 return true;
213}
214#endif /* CONFIG_SECURITY_NETWORK */
215
216static inline int unix_our_peer(struct sock *sk, struct sock *osk)
217{
218 return unix_peer(osk) == sk;
219}
220
221static inline int unix_may_send(struct sock *sk, struct sock *osk)
222{
223 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
224}
225
226static inline int unix_recvq_full(const struct sock *sk)
227{
228 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
229}
230
231static inline int unix_recvq_full_lockless(const struct sock *sk)
232{
233 return skb_queue_len_lockless(&sk->sk_receive_queue) >
234 READ_ONCE(sk->sk_max_ack_backlog);
235}
236
237struct sock *unix_peer_get(struct sock *s)
238{
239 struct sock *peer;
240
241 unix_state_lock(s);
242 peer = unix_peer(s);
243 if (peer)
244 sock_hold(peer);
245 unix_state_unlock(s);
246 return peer;
247}
248EXPORT_SYMBOL_GPL(unix_peer_get);
249
250static struct unix_address *unix_create_addr(struct sockaddr_un *sunaddr,
251 int addr_len)
252{
253 struct unix_address *addr;
254
255 addr = kmalloc(sizeof(*addr) + addr_len, GFP_KERNEL);
256 if (!addr)
257 return NULL;
258
259 refcount_set(&addr->refcnt, 1);
260 addr->len = addr_len;
261 memcpy(addr->name, sunaddr, addr_len);
262
263 return addr;
264}
265
266static inline void unix_release_addr(struct unix_address *addr)
267{
268 if (refcount_dec_and_test(&addr->refcnt))
269 kfree(addr);
270}
271
272/*
273 * Check unix socket name:
274 * - should be not zero length.
275 * - if started by not zero, should be NULL terminated (FS object)
276 * - if started by zero, it is abstract name.
277 */
278
279static int unix_validate_addr(struct sockaddr_un *sunaddr, int addr_len)
280{
281 if (addr_len <= offsetof(struct sockaddr_un, sun_path) ||
282 addr_len > sizeof(*sunaddr))
283 return -EINVAL;
284
285 if (sunaddr->sun_family != AF_UNIX)
286 return -EINVAL;
287
288 return 0;
289}
290
291static int unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
292{
293 struct sockaddr_storage *addr = (struct sockaddr_storage *)sunaddr;
294 short offset = offsetof(struct sockaddr_storage, __data);
295
296 BUILD_BUG_ON(offset != offsetof(struct sockaddr_un, sun_path));
297
298 /* This may look like an off by one error but it is a bit more
299 * subtle. 108 is the longest valid AF_UNIX path for a binding.
300 * sun_path[108] doesn't as such exist. However in kernel space
301 * we are guaranteed that it is a valid memory location in our
302 * kernel address buffer because syscall functions always pass
303 * a pointer of struct sockaddr_storage which has a bigger buffer
304 * than 108. Also, we must terminate sun_path for strlen() in
305 * getname_kernel().
306 */
307 addr->__data[addr_len - offset] = 0;
308
309 /* Don't pass sunaddr->sun_path to strlen(). Otherwise, 108 will
310 * cause panic if CONFIG_FORTIFY_SOURCE=y. Let __fortify_strlen()
311 * know the actual buffer.
312 */
313 return strlen(addr->__data) + offset + 1;
314}
315
316static void __unix_remove_socket(struct sock *sk)
317{
318 sk_del_node_init(sk);
319}
320
321static void __unix_insert_socket(struct net *net, struct sock *sk)
322{
323 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
324 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
325}
326
327static void __unix_set_addr_hash(struct net *net, struct sock *sk,
328 struct unix_address *addr, unsigned int hash)
329{
330 __unix_remove_socket(sk);
331 smp_store_release(&unix_sk(sk)->addr, addr);
332
333 sk->sk_hash = hash;
334 __unix_insert_socket(net, sk);
335}
336
337static void unix_remove_socket(struct net *net, struct sock *sk)
338{
339 spin_lock(&net->unx.table.locks[sk->sk_hash]);
340 __unix_remove_socket(sk);
341 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
342}
343
344static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
345{
346 spin_lock(&net->unx.table.locks[sk->sk_hash]);
347 __unix_insert_socket(net, sk);
348 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
349}
350
351static void unix_insert_bsd_socket(struct sock *sk)
352{
353 spin_lock(&bsd_socket_locks[sk->sk_hash]);
354 sk_add_bind_node(sk, &bsd_socket_buckets[sk->sk_hash]);
355 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
356}
357
358static void unix_remove_bsd_socket(struct sock *sk)
359{
360 if (!hlist_unhashed(&sk->sk_bind_node)) {
361 spin_lock(&bsd_socket_locks[sk->sk_hash]);
362 __sk_del_bind_node(sk);
363 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
364
365 sk_node_init(&sk->sk_bind_node);
366 }
367}
368
369static struct sock *__unix_find_socket_byname(struct net *net,
370 struct sockaddr_un *sunname,
371 int len, unsigned int hash)
372{
373 struct sock *s;
374
375 sk_for_each(s, &net->unx.table.buckets[hash]) {
376 struct unix_sock *u = unix_sk(s);
377
378 if (u->addr->len == len &&
379 !memcmp(u->addr->name, sunname, len))
380 return s;
381 }
382 return NULL;
383}
384
385static inline struct sock *unix_find_socket_byname(struct net *net,
386 struct sockaddr_un *sunname,
387 int len, unsigned int hash)
388{
389 struct sock *s;
390
391 spin_lock(&net->unx.table.locks[hash]);
392 s = __unix_find_socket_byname(net, sunname, len, hash);
393 if (s)
394 sock_hold(s);
395 spin_unlock(&net->unx.table.locks[hash]);
396 return s;
397}
398
399static struct sock *unix_find_socket_byinode(struct inode *i)
400{
401 unsigned int hash = unix_bsd_hash(i);
402 struct sock *s;
403
404 spin_lock(&bsd_socket_locks[hash]);
405 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
406 struct dentry *dentry = unix_sk(s)->path.dentry;
407
408 if (dentry && d_backing_inode(dentry) == i) {
409 sock_hold(s);
410 spin_unlock(&bsd_socket_locks[hash]);
411 return s;
412 }
413 }
414 spin_unlock(&bsd_socket_locks[hash]);
415 return NULL;
416}
417
418/* Support code for asymmetrically connected dgram sockets
419 *
420 * If a datagram socket is connected to a socket not itself connected
421 * to the first socket (eg, /dev/log), clients may only enqueue more
422 * messages if the present receive queue of the server socket is not
423 * "too large". This means there's a second writeability condition
424 * poll and sendmsg need to test. The dgram recv code will do a wake
425 * up on the peer_wait wait queue of a socket upon reception of a
426 * datagram which needs to be propagated to sleeping would-be writers
427 * since these might not have sent anything so far. This can't be
428 * accomplished via poll_wait because the lifetime of the server
429 * socket might be less than that of its clients if these break their
430 * association with it or if the server socket is closed while clients
431 * are still connected to it and there's no way to inform "a polling
432 * implementation" that it should let go of a certain wait queue
433 *
434 * In order to propagate a wake up, a wait_queue_entry_t of the client
435 * socket is enqueued on the peer_wait queue of the server socket
436 * whose wake function does a wake_up on the ordinary client socket
437 * wait queue. This connection is established whenever a write (or
438 * poll for write) hit the flow control condition and broken when the
439 * association to the server socket is dissolved or after a wake up
440 * was relayed.
441 */
442
443static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
444 void *key)
445{
446 struct unix_sock *u;
447 wait_queue_head_t *u_sleep;
448
449 u = container_of(q, struct unix_sock, peer_wake);
450
451 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
452 q);
453 u->peer_wake.private = NULL;
454
455 /* relaying can only happen while the wq still exists */
456 u_sleep = sk_sleep(&u->sk);
457 if (u_sleep)
458 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
459
460 return 0;
461}
462
463static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
464{
465 struct unix_sock *u, *u_other;
466 int rc;
467
468 u = unix_sk(sk);
469 u_other = unix_sk(other);
470 rc = 0;
471 spin_lock(&u_other->peer_wait.lock);
472
473 if (!u->peer_wake.private) {
474 u->peer_wake.private = other;
475 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
476
477 rc = 1;
478 }
479
480 spin_unlock(&u_other->peer_wait.lock);
481 return rc;
482}
483
484static void unix_dgram_peer_wake_disconnect(struct sock *sk,
485 struct sock *other)
486{
487 struct unix_sock *u, *u_other;
488
489 u = unix_sk(sk);
490 u_other = unix_sk(other);
491 spin_lock(&u_other->peer_wait.lock);
492
493 if (u->peer_wake.private == other) {
494 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
495 u->peer_wake.private = NULL;
496 }
497
498 spin_unlock(&u_other->peer_wait.lock);
499}
500
501static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
502 struct sock *other)
503{
504 unix_dgram_peer_wake_disconnect(sk, other);
505 wake_up_interruptible_poll(sk_sleep(sk),
506 EPOLLOUT |
507 EPOLLWRNORM |
508 EPOLLWRBAND);
509}
510
511/* preconditions:
512 * - unix_peer(sk) == other
513 * - association is stable
514 */
515static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
516{
517 int connected;
518
519 connected = unix_dgram_peer_wake_connect(sk, other);
520
521 /* If other is SOCK_DEAD, we want to make sure we signal
522 * POLLOUT, such that a subsequent write() can get a
523 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
524 * to other and its full, we will hang waiting for POLLOUT.
525 */
526 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
527 return 1;
528
529 if (connected)
530 unix_dgram_peer_wake_disconnect(sk, other);
531
532 return 0;
533}
534
535static int unix_writable(const struct sock *sk)
536{
537 return sk->sk_state != TCP_LISTEN &&
538 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
539}
540
541static void unix_write_space(struct sock *sk)
542{
543 struct socket_wq *wq;
544
545 rcu_read_lock();
546 if (unix_writable(sk)) {
547 wq = rcu_dereference(sk->sk_wq);
548 if (skwq_has_sleeper(wq))
549 wake_up_interruptible_sync_poll(&wq->wait,
550 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
551 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
552 }
553 rcu_read_unlock();
554}
555
556/* When dgram socket disconnects (or changes its peer), we clear its receive
557 * queue of packets arrived from previous peer. First, it allows to do
558 * flow control based only on wmem_alloc; second, sk connected to peer
559 * may receive messages only from that peer. */
560static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
561{
562 if (!skb_queue_empty(&sk->sk_receive_queue)) {
563 skb_queue_purge(&sk->sk_receive_queue);
564 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
565
566 /* If one link of bidirectional dgram pipe is disconnected,
567 * we signal error. Messages are lost. Do not make this,
568 * when peer was not connected to us.
569 */
570 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
571 WRITE_ONCE(other->sk_err, ECONNRESET);
572 sk_error_report(other);
573 }
574 }
575 other->sk_state = TCP_CLOSE;
576}
577
578static void unix_sock_destructor(struct sock *sk)
579{
580 struct unix_sock *u = unix_sk(sk);
581
582 skb_queue_purge(&sk->sk_receive_queue);
583
584 DEBUG_NET_WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
585 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
586 DEBUG_NET_WARN_ON_ONCE(sk->sk_socket);
587 if (!sock_flag(sk, SOCK_DEAD)) {
588 pr_info("Attempt to release alive unix socket: %p\n", sk);
589 return;
590 }
591
592 if (u->addr)
593 unix_release_addr(u->addr);
594
595 atomic_long_dec(&unix_nr_socks);
596 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
597#ifdef UNIX_REFCNT_DEBUG
598 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
599 atomic_long_read(&unix_nr_socks));
600#endif
601}
602
603static void unix_release_sock(struct sock *sk, int embrion)
604{
605 struct unix_sock *u = unix_sk(sk);
606 struct sock *skpair;
607 struct sk_buff *skb;
608 struct path path;
609 int state;
610
611 unix_remove_socket(sock_net(sk), sk);
612 unix_remove_bsd_socket(sk);
613
614 /* Clear state */
615 unix_state_lock(sk);
616 sock_orphan(sk);
617 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
618 path = u->path;
619 u->path.dentry = NULL;
620 u->path.mnt = NULL;
621 state = sk->sk_state;
622 sk->sk_state = TCP_CLOSE;
623
624 skpair = unix_peer(sk);
625 unix_peer(sk) = NULL;
626
627 unix_state_unlock(sk);
628
629#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
630 if (u->oob_skb) {
631 kfree_skb(u->oob_skb);
632 u->oob_skb = NULL;
633 }
634#endif
635
636 wake_up_interruptible_all(&u->peer_wait);
637
638 if (skpair != NULL) {
639 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
640 unix_state_lock(skpair);
641 /* No more writes */
642 WRITE_ONCE(skpair->sk_shutdown, SHUTDOWN_MASK);
643 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
644 WRITE_ONCE(skpair->sk_err, ECONNRESET);
645 unix_state_unlock(skpair);
646 skpair->sk_state_change(skpair);
647 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
648 }
649
650 unix_dgram_peer_wake_disconnect(sk, skpair);
651 sock_put(skpair); /* It may now die */
652 }
653
654 /* Try to flush out this socket. Throw out buffers at least */
655
656 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
657 if (state == TCP_LISTEN)
658 unix_release_sock(skb->sk, 1);
659 /* passed fds are erased in the kfree_skb hook */
660 UNIXCB(skb).consumed = skb->len;
661 kfree_skb(skb);
662 }
663
664 if (path.dentry)
665 path_put(&path);
666
667 sock_put(sk);
668
669 /* ---- Socket is dead now and most probably destroyed ---- */
670
671 /*
672 * Fixme: BSD difference: In BSD all sockets connected to us get
673 * ECONNRESET and we die on the spot. In Linux we behave
674 * like files and pipes do and wait for the last
675 * dereference.
676 *
677 * Can't we simply set sock->err?
678 *
679 * What the above comment does talk about? --ANK(980817)
680 */
681
682 if (READ_ONCE(unix_tot_inflight))
683 unix_gc(); /* Garbage collect fds */
684}
685
686static void init_peercred(struct sock *sk)
687{
688 const struct cred *old_cred;
689 struct pid *old_pid;
690
691 spin_lock(&sk->sk_peer_lock);
692 old_pid = sk->sk_peer_pid;
693 old_cred = sk->sk_peer_cred;
694 sk->sk_peer_pid = get_pid(task_tgid(current));
695 sk->sk_peer_cred = get_current_cred();
696 spin_unlock(&sk->sk_peer_lock);
697
698 put_pid(old_pid);
699 put_cred(old_cred);
700}
701
702static void copy_peercred(struct sock *sk, struct sock *peersk)
703{
704 const struct cred *old_cred;
705 struct pid *old_pid;
706
707 if (sk < peersk) {
708 spin_lock(&sk->sk_peer_lock);
709 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
710 } else {
711 spin_lock(&peersk->sk_peer_lock);
712 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
713 }
714 old_pid = sk->sk_peer_pid;
715 old_cred = sk->sk_peer_cred;
716 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
717 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
718
719 spin_unlock(&sk->sk_peer_lock);
720 spin_unlock(&peersk->sk_peer_lock);
721
722 put_pid(old_pid);
723 put_cred(old_cred);
724}
725
726static int unix_listen(struct socket *sock, int backlog)
727{
728 int err;
729 struct sock *sk = sock->sk;
730 struct unix_sock *u = unix_sk(sk);
731
732 err = -EOPNOTSUPP;
733 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
734 goto out; /* Only stream/seqpacket sockets accept */
735 err = -EINVAL;
736 if (!u->addr)
737 goto out; /* No listens on an unbound socket */
738 unix_state_lock(sk);
739 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
740 goto out_unlock;
741 if (backlog > sk->sk_max_ack_backlog)
742 wake_up_interruptible_all(&u->peer_wait);
743 sk->sk_max_ack_backlog = backlog;
744 sk->sk_state = TCP_LISTEN;
745 /* set credentials so connect can copy them */
746 init_peercred(sk);
747 err = 0;
748
749out_unlock:
750 unix_state_unlock(sk);
751out:
752 return err;
753}
754
755static int unix_release(struct socket *);
756static int unix_bind(struct socket *, struct sockaddr *, int);
757static int unix_stream_connect(struct socket *, struct sockaddr *,
758 int addr_len, int flags);
759static int unix_socketpair(struct socket *, struct socket *);
760static int unix_accept(struct socket *, struct socket *, int, bool);
761static int unix_getname(struct socket *, struct sockaddr *, int);
762static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
763static __poll_t unix_dgram_poll(struct file *, struct socket *,
764 poll_table *);
765static int unix_ioctl(struct socket *, unsigned int, unsigned long);
766#ifdef CONFIG_COMPAT
767static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
768#endif
769static int unix_shutdown(struct socket *, int);
770static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
771static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
772static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
773 struct pipe_inode_info *, size_t size,
774 unsigned int flags);
775static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
776static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
777static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
778static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
779static int unix_dgram_connect(struct socket *, struct sockaddr *,
780 int, int);
781static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
782static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
783 int);
784
785#ifdef CONFIG_PROC_FS
786static int unix_count_nr_fds(struct sock *sk)
787{
788 struct sk_buff *skb;
789 struct unix_sock *u;
790 int nr_fds = 0;
791
792 spin_lock(&sk->sk_receive_queue.lock);
793 skb = skb_peek(&sk->sk_receive_queue);
794 while (skb) {
795 u = unix_sk(skb->sk);
796 nr_fds += atomic_read(&u->scm_stat.nr_fds);
797 skb = skb_peek_next(skb, &sk->sk_receive_queue);
798 }
799 spin_unlock(&sk->sk_receive_queue.lock);
800
801 return nr_fds;
802}
803
804static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
805{
806 struct sock *sk = sock->sk;
807 unsigned char s_state;
808 struct unix_sock *u;
809 int nr_fds = 0;
810
811 if (sk) {
812 s_state = READ_ONCE(sk->sk_state);
813 u = unix_sk(sk);
814
815 /* SOCK_STREAM and SOCK_SEQPACKET sockets never change their
816 * sk_state after switching to TCP_ESTABLISHED or TCP_LISTEN.
817 * SOCK_DGRAM is ordinary. So, no lock is needed.
818 */
819 if (sock->type == SOCK_DGRAM || s_state == TCP_ESTABLISHED)
820 nr_fds = atomic_read(&u->scm_stat.nr_fds);
821 else if (s_state == TCP_LISTEN)
822 nr_fds = unix_count_nr_fds(sk);
823
824 seq_printf(m, "scm_fds: %u\n", nr_fds);
825 }
826}
827#else
828#define unix_show_fdinfo NULL
829#endif
830
831static const struct proto_ops unix_stream_ops = {
832 .family = PF_UNIX,
833 .owner = THIS_MODULE,
834 .release = unix_release,
835 .bind = unix_bind,
836 .connect = unix_stream_connect,
837 .socketpair = unix_socketpair,
838 .accept = unix_accept,
839 .getname = unix_getname,
840 .poll = unix_poll,
841 .ioctl = unix_ioctl,
842#ifdef CONFIG_COMPAT
843 .compat_ioctl = unix_compat_ioctl,
844#endif
845 .listen = unix_listen,
846 .shutdown = unix_shutdown,
847 .sendmsg = unix_stream_sendmsg,
848 .recvmsg = unix_stream_recvmsg,
849 .read_skb = unix_stream_read_skb,
850 .mmap = sock_no_mmap,
851 .splice_read = unix_stream_splice_read,
852 .set_peek_off = sk_set_peek_off,
853 .show_fdinfo = unix_show_fdinfo,
854};
855
856static const struct proto_ops unix_dgram_ops = {
857 .family = PF_UNIX,
858 .owner = THIS_MODULE,
859 .release = unix_release,
860 .bind = unix_bind,
861 .connect = unix_dgram_connect,
862 .socketpair = unix_socketpair,
863 .accept = sock_no_accept,
864 .getname = unix_getname,
865 .poll = unix_dgram_poll,
866 .ioctl = unix_ioctl,
867#ifdef CONFIG_COMPAT
868 .compat_ioctl = unix_compat_ioctl,
869#endif
870 .listen = sock_no_listen,
871 .shutdown = unix_shutdown,
872 .sendmsg = unix_dgram_sendmsg,
873 .read_skb = unix_read_skb,
874 .recvmsg = unix_dgram_recvmsg,
875 .mmap = sock_no_mmap,
876 .set_peek_off = sk_set_peek_off,
877 .show_fdinfo = unix_show_fdinfo,
878};
879
880static const struct proto_ops unix_seqpacket_ops = {
881 .family = PF_UNIX,
882 .owner = THIS_MODULE,
883 .release = unix_release,
884 .bind = unix_bind,
885 .connect = unix_stream_connect,
886 .socketpair = unix_socketpair,
887 .accept = unix_accept,
888 .getname = unix_getname,
889 .poll = unix_dgram_poll,
890 .ioctl = unix_ioctl,
891#ifdef CONFIG_COMPAT
892 .compat_ioctl = unix_compat_ioctl,
893#endif
894 .listen = unix_listen,
895 .shutdown = unix_shutdown,
896 .sendmsg = unix_seqpacket_sendmsg,
897 .recvmsg = unix_seqpacket_recvmsg,
898 .mmap = sock_no_mmap,
899 .set_peek_off = sk_set_peek_off,
900 .show_fdinfo = unix_show_fdinfo,
901};
902
903static void unix_close(struct sock *sk, long timeout)
904{
905 /* Nothing to do here, unix socket does not need a ->close().
906 * This is merely for sockmap.
907 */
908}
909
910static void unix_unhash(struct sock *sk)
911{
912 /* Nothing to do here, unix socket does not need a ->unhash().
913 * This is merely for sockmap.
914 */
915}
916
917static bool unix_bpf_bypass_getsockopt(int level, int optname)
918{
919 if (level == SOL_SOCKET) {
920 switch (optname) {
921 case SO_PEERPIDFD:
922 return true;
923 default:
924 return false;
925 }
926 }
927
928 return false;
929}
930
931struct proto unix_dgram_proto = {
932 .name = "UNIX",
933 .owner = THIS_MODULE,
934 .obj_size = sizeof(struct unix_sock),
935 .close = unix_close,
936 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
937#ifdef CONFIG_BPF_SYSCALL
938 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
939#endif
940};
941
942struct proto unix_stream_proto = {
943 .name = "UNIX-STREAM",
944 .owner = THIS_MODULE,
945 .obj_size = sizeof(struct unix_sock),
946 .close = unix_close,
947 .unhash = unix_unhash,
948 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
949#ifdef CONFIG_BPF_SYSCALL
950 .psock_update_sk_prot = unix_stream_bpf_update_proto,
951#endif
952};
953
954static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
955{
956 struct unix_sock *u;
957 struct sock *sk;
958 int err;
959
960 atomic_long_inc(&unix_nr_socks);
961 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
962 err = -ENFILE;
963 goto err;
964 }
965
966 if (type == SOCK_STREAM)
967 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
968 else /*dgram and seqpacket */
969 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
970
971 if (!sk) {
972 err = -ENOMEM;
973 goto err;
974 }
975
976 sock_init_data(sock, sk);
977
978 sk->sk_hash = unix_unbound_hash(sk);
979 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
980 sk->sk_write_space = unix_write_space;
981 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
982 sk->sk_destruct = unix_sock_destructor;
983 u = unix_sk(sk);
984 u->path.dentry = NULL;
985 u->path.mnt = NULL;
986 spin_lock_init(&u->lock);
987 atomic_long_set(&u->inflight, 0);
988 INIT_LIST_HEAD(&u->link);
989 mutex_init(&u->iolock); /* single task reading lock */
990 mutex_init(&u->bindlock); /* single task binding lock */
991 init_waitqueue_head(&u->peer_wait);
992 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
993 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
994 unix_insert_unbound_socket(net, sk);
995
996 sock_prot_inuse_add(net, sk->sk_prot, 1);
997
998 return sk;
999
1000err:
1001 atomic_long_dec(&unix_nr_socks);
1002 return ERR_PTR(err);
1003}
1004
1005static int unix_create(struct net *net, struct socket *sock, int protocol,
1006 int kern)
1007{
1008 struct sock *sk;
1009
1010 if (protocol && protocol != PF_UNIX)
1011 return -EPROTONOSUPPORT;
1012
1013 sock->state = SS_UNCONNECTED;
1014
1015 switch (sock->type) {
1016 case SOCK_STREAM:
1017 sock->ops = &unix_stream_ops;
1018 break;
1019 /*
1020 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
1021 * nothing uses it.
1022 */
1023 case SOCK_RAW:
1024 sock->type = SOCK_DGRAM;
1025 fallthrough;
1026 case SOCK_DGRAM:
1027 sock->ops = &unix_dgram_ops;
1028 break;
1029 case SOCK_SEQPACKET:
1030 sock->ops = &unix_seqpacket_ops;
1031 break;
1032 default:
1033 return -ESOCKTNOSUPPORT;
1034 }
1035
1036 sk = unix_create1(net, sock, kern, sock->type);
1037 if (IS_ERR(sk))
1038 return PTR_ERR(sk);
1039
1040 return 0;
1041}
1042
1043static int unix_release(struct socket *sock)
1044{
1045 struct sock *sk = sock->sk;
1046
1047 if (!sk)
1048 return 0;
1049
1050 sk->sk_prot->close(sk, 0);
1051 unix_release_sock(sk, 0);
1052 sock->sk = NULL;
1053
1054 return 0;
1055}
1056
1057static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1058 int type)
1059{
1060 struct inode *inode;
1061 struct path path;
1062 struct sock *sk;
1063 int err;
1064
1065 unix_mkname_bsd(sunaddr, addr_len);
1066 err = kern_path(sunaddr->sun_path, LOOKUP_FOLLOW, &path);
1067 if (err)
1068 goto fail;
1069
1070 err = path_permission(&path, MAY_WRITE);
1071 if (err)
1072 goto path_put;
1073
1074 err = -ECONNREFUSED;
1075 inode = d_backing_inode(path.dentry);
1076 if (!S_ISSOCK(inode->i_mode))
1077 goto path_put;
1078
1079 sk = unix_find_socket_byinode(inode);
1080 if (!sk)
1081 goto path_put;
1082
1083 err = -EPROTOTYPE;
1084 if (sk->sk_type == type)
1085 touch_atime(&path);
1086 else
1087 goto sock_put;
1088
1089 path_put(&path);
1090
1091 return sk;
1092
1093sock_put:
1094 sock_put(sk);
1095path_put:
1096 path_put(&path);
1097fail:
1098 return ERR_PTR(err);
1099}
1100
1101static struct sock *unix_find_abstract(struct net *net,
1102 struct sockaddr_un *sunaddr,
1103 int addr_len, int type)
1104{
1105 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
1106 struct dentry *dentry;
1107 struct sock *sk;
1108
1109 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
1110 if (!sk)
1111 return ERR_PTR(-ECONNREFUSED);
1112
1113 dentry = unix_sk(sk)->path.dentry;
1114 if (dentry)
1115 touch_atime(&unix_sk(sk)->path);
1116
1117 return sk;
1118}
1119
1120static struct sock *unix_find_other(struct net *net,
1121 struct sockaddr_un *sunaddr,
1122 int addr_len, int type)
1123{
1124 struct sock *sk;
1125
1126 if (sunaddr->sun_path[0])
1127 sk = unix_find_bsd(sunaddr, addr_len, type);
1128 else
1129 sk = unix_find_abstract(net, sunaddr, addr_len, type);
1130
1131 return sk;
1132}
1133
1134static int unix_autobind(struct sock *sk)
1135{
1136 unsigned int new_hash, old_hash = sk->sk_hash;
1137 struct unix_sock *u = unix_sk(sk);
1138 struct net *net = sock_net(sk);
1139 struct unix_address *addr;
1140 u32 lastnum, ordernum;
1141 int err;
1142
1143 err = mutex_lock_interruptible(&u->bindlock);
1144 if (err)
1145 return err;
1146
1147 if (u->addr)
1148 goto out;
1149
1150 err = -ENOMEM;
1151 addr = kzalloc(sizeof(*addr) +
1152 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1153 if (!addr)
1154 goto out;
1155
1156 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1157 addr->name->sun_family = AF_UNIX;
1158 refcount_set(&addr->refcnt, 1);
1159
1160 ordernum = get_random_u32();
1161 lastnum = ordernum & 0xFFFFF;
1162retry:
1163 ordernum = (ordernum + 1) & 0xFFFFF;
1164 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1165
1166 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1167 unix_table_double_lock(net, old_hash, new_hash);
1168
1169 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1170 unix_table_double_unlock(net, old_hash, new_hash);
1171
1172 /* __unix_find_socket_byname() may take long time if many names
1173 * are already in use.
1174 */
1175 cond_resched();
1176
1177 if (ordernum == lastnum) {
1178 /* Give up if all names seems to be in use. */
1179 err = -ENOSPC;
1180 unix_release_addr(addr);
1181 goto out;
1182 }
1183
1184 goto retry;
1185 }
1186
1187 __unix_set_addr_hash(net, sk, addr, new_hash);
1188 unix_table_double_unlock(net, old_hash, new_hash);
1189 err = 0;
1190
1191out: mutex_unlock(&u->bindlock);
1192 return err;
1193}
1194
1195static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1196 int addr_len)
1197{
1198 umode_t mode = S_IFSOCK |
1199 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1200 unsigned int new_hash, old_hash = sk->sk_hash;
1201 struct unix_sock *u = unix_sk(sk);
1202 struct net *net = sock_net(sk);
1203 struct mnt_idmap *idmap;
1204 struct unix_address *addr;
1205 struct dentry *dentry;
1206 struct path parent;
1207 int err;
1208
1209 addr_len = unix_mkname_bsd(sunaddr, addr_len);
1210 addr = unix_create_addr(sunaddr, addr_len);
1211 if (!addr)
1212 return -ENOMEM;
1213
1214 /*
1215 * Get the parent directory, calculate the hash for last
1216 * component.
1217 */
1218 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1219 if (IS_ERR(dentry)) {
1220 err = PTR_ERR(dentry);
1221 goto out;
1222 }
1223
1224 /*
1225 * All right, let's create it.
1226 */
1227 idmap = mnt_idmap(parent.mnt);
1228 err = security_path_mknod(&parent, dentry, mode, 0);
1229 if (!err)
1230 err = vfs_mknod(idmap, d_inode(parent.dentry), dentry, mode, 0);
1231 if (err)
1232 goto out_path;
1233 err = mutex_lock_interruptible(&u->bindlock);
1234 if (err)
1235 goto out_unlink;
1236 if (u->addr)
1237 goto out_unlock;
1238
1239 new_hash = unix_bsd_hash(d_backing_inode(dentry));
1240 unix_table_double_lock(net, old_hash, new_hash);
1241 u->path.mnt = mntget(parent.mnt);
1242 u->path.dentry = dget(dentry);
1243 __unix_set_addr_hash(net, sk, addr, new_hash);
1244 unix_table_double_unlock(net, old_hash, new_hash);
1245 unix_insert_bsd_socket(sk);
1246 mutex_unlock(&u->bindlock);
1247 done_path_create(&parent, dentry);
1248 return 0;
1249
1250out_unlock:
1251 mutex_unlock(&u->bindlock);
1252 err = -EINVAL;
1253out_unlink:
1254 /* failed after successful mknod? unlink what we'd created... */
1255 vfs_unlink(idmap, d_inode(parent.dentry), dentry, NULL);
1256out_path:
1257 done_path_create(&parent, dentry);
1258out:
1259 unix_release_addr(addr);
1260 return err == -EEXIST ? -EADDRINUSE : err;
1261}
1262
1263static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1264 int addr_len)
1265{
1266 unsigned int new_hash, old_hash = sk->sk_hash;
1267 struct unix_sock *u = unix_sk(sk);
1268 struct net *net = sock_net(sk);
1269 struct unix_address *addr;
1270 int err;
1271
1272 addr = unix_create_addr(sunaddr, addr_len);
1273 if (!addr)
1274 return -ENOMEM;
1275
1276 err = mutex_lock_interruptible(&u->bindlock);
1277 if (err)
1278 goto out;
1279
1280 if (u->addr) {
1281 err = -EINVAL;
1282 goto out_mutex;
1283 }
1284
1285 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1286 unix_table_double_lock(net, old_hash, new_hash);
1287
1288 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
1289 goto out_spin;
1290
1291 __unix_set_addr_hash(net, sk, addr, new_hash);
1292 unix_table_double_unlock(net, old_hash, new_hash);
1293 mutex_unlock(&u->bindlock);
1294 return 0;
1295
1296out_spin:
1297 unix_table_double_unlock(net, old_hash, new_hash);
1298 err = -EADDRINUSE;
1299out_mutex:
1300 mutex_unlock(&u->bindlock);
1301out:
1302 unix_release_addr(addr);
1303 return err;
1304}
1305
1306static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1307{
1308 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1309 struct sock *sk = sock->sk;
1310 int err;
1311
1312 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1313 sunaddr->sun_family == AF_UNIX)
1314 return unix_autobind(sk);
1315
1316 err = unix_validate_addr(sunaddr, addr_len);
1317 if (err)
1318 return err;
1319
1320 if (sunaddr->sun_path[0])
1321 err = unix_bind_bsd(sk, sunaddr, addr_len);
1322 else
1323 err = unix_bind_abstract(sk, sunaddr, addr_len);
1324
1325 return err;
1326}
1327
1328static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1329{
1330 if (unlikely(sk1 == sk2) || !sk2) {
1331 unix_state_lock(sk1);
1332 return;
1333 }
1334 if (sk1 > sk2)
1335 swap(sk1, sk2);
1336
1337 unix_state_lock(sk1);
1338 unix_state_lock_nested(sk2, U_LOCK_SECOND);
1339}
1340
1341static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1342{
1343 if (unlikely(sk1 == sk2) || !sk2) {
1344 unix_state_unlock(sk1);
1345 return;
1346 }
1347 unix_state_unlock(sk1);
1348 unix_state_unlock(sk2);
1349}
1350
1351static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1352 int alen, int flags)
1353{
1354 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1355 struct sock *sk = sock->sk;
1356 struct sock *other;
1357 int err;
1358
1359 err = -EINVAL;
1360 if (alen < offsetofend(struct sockaddr, sa_family))
1361 goto out;
1362
1363 if (addr->sa_family != AF_UNSPEC) {
1364 err = unix_validate_addr(sunaddr, alen);
1365 if (err)
1366 goto out;
1367
1368 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, addr, &alen);
1369 if (err)
1370 goto out;
1371
1372 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1373 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1374 !unix_sk(sk)->addr) {
1375 err = unix_autobind(sk);
1376 if (err)
1377 goto out;
1378 }
1379
1380restart:
1381 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
1382 if (IS_ERR(other)) {
1383 err = PTR_ERR(other);
1384 goto out;
1385 }
1386
1387 unix_state_double_lock(sk, other);
1388
1389 /* Apparently VFS overslept socket death. Retry. */
1390 if (sock_flag(other, SOCK_DEAD)) {
1391 unix_state_double_unlock(sk, other);
1392 sock_put(other);
1393 goto restart;
1394 }
1395
1396 err = -EPERM;
1397 if (!unix_may_send(sk, other))
1398 goto out_unlock;
1399
1400 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1401 if (err)
1402 goto out_unlock;
1403
1404 sk->sk_state = other->sk_state = TCP_ESTABLISHED;
1405 } else {
1406 /*
1407 * 1003.1g breaking connected state with AF_UNSPEC
1408 */
1409 other = NULL;
1410 unix_state_double_lock(sk, other);
1411 }
1412
1413 /*
1414 * If it was connected, reconnect.
1415 */
1416 if (unix_peer(sk)) {
1417 struct sock *old_peer = unix_peer(sk);
1418
1419 unix_peer(sk) = other;
1420 if (!other)
1421 sk->sk_state = TCP_CLOSE;
1422 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1423
1424 unix_state_double_unlock(sk, other);
1425
1426 if (other != old_peer)
1427 unix_dgram_disconnected(sk, old_peer);
1428 sock_put(old_peer);
1429 } else {
1430 unix_peer(sk) = other;
1431 unix_state_double_unlock(sk, other);
1432 }
1433
1434 return 0;
1435
1436out_unlock:
1437 unix_state_double_unlock(sk, other);
1438 sock_put(other);
1439out:
1440 return err;
1441}
1442
1443static long unix_wait_for_peer(struct sock *other, long timeo)
1444 __releases(&unix_sk(other)->lock)
1445{
1446 struct unix_sock *u = unix_sk(other);
1447 int sched;
1448 DEFINE_WAIT(wait);
1449
1450 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1451
1452 sched = !sock_flag(other, SOCK_DEAD) &&
1453 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1454 unix_recvq_full_lockless(other);
1455
1456 unix_state_unlock(other);
1457
1458 if (sched)
1459 timeo = schedule_timeout(timeo);
1460
1461 finish_wait(&u->peer_wait, &wait);
1462 return timeo;
1463}
1464
1465static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1466 int addr_len, int flags)
1467{
1468 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1469 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1470 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1471 struct net *net = sock_net(sk);
1472 struct sk_buff *skb = NULL;
1473 long timeo;
1474 int err;
1475 int st;
1476
1477 err = unix_validate_addr(sunaddr, addr_len);
1478 if (err)
1479 goto out;
1480
1481 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, &addr_len);
1482 if (err)
1483 goto out;
1484
1485 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1486 test_bit(SOCK_PASSPIDFD, &sock->flags)) && !u->addr) {
1487 err = unix_autobind(sk);
1488 if (err)
1489 goto out;
1490 }
1491
1492 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1493
1494 /* First of all allocate resources.
1495 If we will make it after state is locked,
1496 we will have to recheck all again in any case.
1497 */
1498
1499 /* create new sock for complete connection */
1500 newsk = unix_create1(net, NULL, 0, sock->type);
1501 if (IS_ERR(newsk)) {
1502 err = PTR_ERR(newsk);
1503 newsk = NULL;
1504 goto out;
1505 }
1506
1507 err = -ENOMEM;
1508
1509 /* Allocate skb for sending to listening sock */
1510 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1511 if (skb == NULL)
1512 goto out;
1513
1514restart:
1515 /* Find listening sock. */
1516 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
1517 if (IS_ERR(other)) {
1518 err = PTR_ERR(other);
1519 other = NULL;
1520 goto out;
1521 }
1522
1523 /* Latch state of peer */
1524 unix_state_lock(other);
1525
1526 /* Apparently VFS overslept socket death. Retry. */
1527 if (sock_flag(other, SOCK_DEAD)) {
1528 unix_state_unlock(other);
1529 sock_put(other);
1530 goto restart;
1531 }
1532
1533 err = -ECONNREFUSED;
1534 if (other->sk_state != TCP_LISTEN)
1535 goto out_unlock;
1536 if (other->sk_shutdown & RCV_SHUTDOWN)
1537 goto out_unlock;
1538
1539 if (unix_recvq_full(other)) {
1540 err = -EAGAIN;
1541 if (!timeo)
1542 goto out_unlock;
1543
1544 timeo = unix_wait_for_peer(other, timeo);
1545
1546 err = sock_intr_errno(timeo);
1547 if (signal_pending(current))
1548 goto out;
1549 sock_put(other);
1550 goto restart;
1551 }
1552
1553 /* Latch our state.
1554
1555 It is tricky place. We need to grab our state lock and cannot
1556 drop lock on peer. It is dangerous because deadlock is
1557 possible. Connect to self case and simultaneous
1558 attempt to connect are eliminated by checking socket
1559 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1560 check this before attempt to grab lock.
1561
1562 Well, and we have to recheck the state after socket locked.
1563 */
1564 st = sk->sk_state;
1565
1566 switch (st) {
1567 case TCP_CLOSE:
1568 /* This is ok... continue with connect */
1569 break;
1570 case TCP_ESTABLISHED:
1571 /* Socket is already connected */
1572 err = -EISCONN;
1573 goto out_unlock;
1574 default:
1575 err = -EINVAL;
1576 goto out_unlock;
1577 }
1578
1579 unix_state_lock_nested(sk, U_LOCK_SECOND);
1580
1581 if (sk->sk_state != st) {
1582 unix_state_unlock(sk);
1583 unix_state_unlock(other);
1584 sock_put(other);
1585 goto restart;
1586 }
1587
1588 err = security_unix_stream_connect(sk, other, newsk);
1589 if (err) {
1590 unix_state_unlock(sk);
1591 goto out_unlock;
1592 }
1593
1594 /* The way is open! Fastly set all the necessary fields... */
1595
1596 sock_hold(sk);
1597 unix_peer(newsk) = sk;
1598 newsk->sk_state = TCP_ESTABLISHED;
1599 newsk->sk_type = sk->sk_type;
1600 init_peercred(newsk);
1601 newu = unix_sk(newsk);
1602 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1603 otheru = unix_sk(other);
1604
1605 /* copy address information from listening to new sock
1606 *
1607 * The contents of *(otheru->addr) and otheru->path
1608 * are seen fully set up here, since we have found
1609 * otheru in hash under its lock. Insertion into the
1610 * hash chain we'd found it in had been done in an
1611 * earlier critical area protected by the chain's lock,
1612 * the same one where we'd set *(otheru->addr) contents,
1613 * as well as otheru->path and otheru->addr itself.
1614 *
1615 * Using smp_store_release() here to set newu->addr
1616 * is enough to make those stores, as well as stores
1617 * to newu->path visible to anyone who gets newu->addr
1618 * by smp_load_acquire(). IOW, the same warranties
1619 * as for unix_sock instances bound in unix_bind() or
1620 * in unix_autobind().
1621 */
1622 if (otheru->path.dentry) {
1623 path_get(&otheru->path);
1624 newu->path = otheru->path;
1625 }
1626 refcount_inc(&otheru->addr->refcnt);
1627 smp_store_release(&newu->addr, otheru->addr);
1628
1629 /* Set credentials */
1630 copy_peercred(sk, other);
1631
1632 sock->state = SS_CONNECTED;
1633 sk->sk_state = TCP_ESTABLISHED;
1634 sock_hold(newsk);
1635
1636 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1637 unix_peer(sk) = newsk;
1638
1639 unix_state_unlock(sk);
1640
1641 /* take ten and send info to listening sock */
1642 spin_lock(&other->sk_receive_queue.lock);
1643 __skb_queue_tail(&other->sk_receive_queue, skb);
1644 spin_unlock(&other->sk_receive_queue.lock);
1645 unix_state_unlock(other);
1646 other->sk_data_ready(other);
1647 sock_put(other);
1648 return 0;
1649
1650out_unlock:
1651 if (other)
1652 unix_state_unlock(other);
1653
1654out:
1655 kfree_skb(skb);
1656 if (newsk)
1657 unix_release_sock(newsk, 0);
1658 if (other)
1659 sock_put(other);
1660 return err;
1661}
1662
1663static int unix_socketpair(struct socket *socka, struct socket *sockb)
1664{
1665 struct sock *ska = socka->sk, *skb = sockb->sk;
1666
1667 /* Join our sockets back to back */
1668 sock_hold(ska);
1669 sock_hold(skb);
1670 unix_peer(ska) = skb;
1671 unix_peer(skb) = ska;
1672 init_peercred(ska);
1673 init_peercred(skb);
1674
1675 ska->sk_state = TCP_ESTABLISHED;
1676 skb->sk_state = TCP_ESTABLISHED;
1677 socka->state = SS_CONNECTED;
1678 sockb->state = SS_CONNECTED;
1679 return 0;
1680}
1681
1682static void unix_sock_inherit_flags(const struct socket *old,
1683 struct socket *new)
1684{
1685 if (test_bit(SOCK_PASSCRED, &old->flags))
1686 set_bit(SOCK_PASSCRED, &new->flags);
1687 if (test_bit(SOCK_PASSPIDFD, &old->flags))
1688 set_bit(SOCK_PASSPIDFD, &new->flags);
1689 if (test_bit(SOCK_PASSSEC, &old->flags))
1690 set_bit(SOCK_PASSSEC, &new->flags);
1691}
1692
1693static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1694 bool kern)
1695{
1696 struct sock *sk = sock->sk;
1697 struct sock *tsk;
1698 struct sk_buff *skb;
1699 int err;
1700
1701 err = -EOPNOTSUPP;
1702 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1703 goto out;
1704
1705 err = -EINVAL;
1706 if (sk->sk_state != TCP_LISTEN)
1707 goto out;
1708
1709 /* If socket state is TCP_LISTEN it cannot change (for now...),
1710 * so that no locks are necessary.
1711 */
1712
1713 skb = skb_recv_datagram(sk, (flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1714 &err);
1715 if (!skb) {
1716 /* This means receive shutdown. */
1717 if (err == 0)
1718 err = -EINVAL;
1719 goto out;
1720 }
1721
1722 tsk = skb->sk;
1723 skb_free_datagram(sk, skb);
1724 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1725
1726 /* attach accepted sock to socket */
1727 unix_state_lock(tsk);
1728 newsock->state = SS_CONNECTED;
1729 unix_sock_inherit_flags(sock, newsock);
1730 sock_graft(tsk, newsock);
1731 unix_state_unlock(tsk);
1732 return 0;
1733
1734out:
1735 return err;
1736}
1737
1738
1739static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1740{
1741 struct sock *sk = sock->sk;
1742 struct unix_address *addr;
1743 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1744 int err = 0;
1745
1746 if (peer) {
1747 sk = unix_peer_get(sk);
1748
1749 err = -ENOTCONN;
1750 if (!sk)
1751 goto out;
1752 err = 0;
1753 } else {
1754 sock_hold(sk);
1755 }
1756
1757 addr = smp_load_acquire(&unix_sk(sk)->addr);
1758 if (!addr) {
1759 sunaddr->sun_family = AF_UNIX;
1760 sunaddr->sun_path[0] = 0;
1761 err = offsetof(struct sockaddr_un, sun_path);
1762 } else {
1763 err = addr->len;
1764 memcpy(sunaddr, addr->name, addr->len);
1765
1766 if (peer)
1767 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1768 CGROUP_UNIX_GETPEERNAME);
1769 else
1770 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1771 CGROUP_UNIX_GETSOCKNAME);
1772 }
1773 sock_put(sk);
1774out:
1775 return err;
1776}
1777
1778static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1779{
1780 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1781
1782 /*
1783 * Garbage collection of unix sockets starts by selecting a set of
1784 * candidate sockets which have reference only from being in flight
1785 * (total_refs == inflight_refs). This condition is checked once during
1786 * the candidate collection phase, and candidates are marked as such, so
1787 * that non-candidates can later be ignored. While inflight_refs is
1788 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1789 * is an instantaneous decision.
1790 *
1791 * Once a candidate, however, the socket must not be reinstalled into a
1792 * file descriptor while the garbage collection is in progress.
1793 *
1794 * If the above conditions are met, then the directed graph of
1795 * candidates (*) does not change while unix_gc_lock is held.
1796 *
1797 * Any operations that changes the file count through file descriptors
1798 * (dup, close, sendmsg) does not change the graph since candidates are
1799 * not installed in fds.
1800 *
1801 * Dequeing a candidate via recvmsg would install it into an fd, but
1802 * that takes unix_gc_lock to decrement the inflight count, so it's
1803 * serialized with garbage collection.
1804 *
1805 * MSG_PEEK is special in that it does not change the inflight count,
1806 * yet does install the socket into an fd. The following lock/unlock
1807 * pair is to ensure serialization with garbage collection. It must be
1808 * done between incrementing the file count and installing the file into
1809 * an fd.
1810 *
1811 * If garbage collection starts after the barrier provided by the
1812 * lock/unlock, then it will see the elevated refcount and not mark this
1813 * as a candidate. If a garbage collection is already in progress
1814 * before the file count was incremented, then the lock/unlock pair will
1815 * ensure that garbage collection is finished before progressing to
1816 * installing the fd.
1817 *
1818 * (*) A -> B where B is on the queue of A or B is on the queue of C
1819 * which is on the queue of listening socket A.
1820 */
1821 spin_lock(&unix_gc_lock);
1822 spin_unlock(&unix_gc_lock);
1823}
1824
1825static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1826{
1827 int err = 0;
1828
1829 UNIXCB(skb).pid = get_pid(scm->pid);
1830 UNIXCB(skb).uid = scm->creds.uid;
1831 UNIXCB(skb).gid = scm->creds.gid;
1832 UNIXCB(skb).fp = NULL;
1833 unix_get_secdata(scm, skb);
1834 if (scm->fp && send_fds)
1835 err = unix_attach_fds(scm, skb);
1836
1837 skb->destructor = unix_destruct_scm;
1838 return err;
1839}
1840
1841static bool unix_passcred_enabled(const struct socket *sock,
1842 const struct sock *other)
1843{
1844 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1845 test_bit(SOCK_PASSPIDFD, &sock->flags) ||
1846 !other->sk_socket ||
1847 test_bit(SOCK_PASSCRED, &other->sk_socket->flags) ||
1848 test_bit(SOCK_PASSPIDFD, &other->sk_socket->flags);
1849}
1850
1851/*
1852 * Some apps rely on write() giving SCM_CREDENTIALS
1853 * We include credentials if source or destination socket
1854 * asserted SOCK_PASSCRED.
1855 */
1856static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1857 const struct sock *other)
1858{
1859 if (UNIXCB(skb).pid)
1860 return;
1861 if (unix_passcred_enabled(sock, other)) {
1862 UNIXCB(skb).pid = get_pid(task_tgid(current));
1863 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1864 }
1865}
1866
1867static bool unix_skb_scm_eq(struct sk_buff *skb,
1868 struct scm_cookie *scm)
1869{
1870 return UNIXCB(skb).pid == scm->pid &&
1871 uid_eq(UNIXCB(skb).uid, scm->creds.uid) &&
1872 gid_eq(UNIXCB(skb).gid, scm->creds.gid) &&
1873 unix_secdata_eq(scm, skb);
1874}
1875
1876static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1877{
1878 struct scm_fp_list *fp = UNIXCB(skb).fp;
1879 struct unix_sock *u = unix_sk(sk);
1880
1881 if (unlikely(fp && fp->count))
1882 atomic_add(fp->count, &u->scm_stat.nr_fds);
1883}
1884
1885static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1886{
1887 struct scm_fp_list *fp = UNIXCB(skb).fp;
1888 struct unix_sock *u = unix_sk(sk);
1889
1890 if (unlikely(fp && fp->count))
1891 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1892}
1893
1894/*
1895 * Send AF_UNIX data.
1896 */
1897
1898static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1899 size_t len)
1900{
1901 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1902 struct sock *sk = sock->sk, *other = NULL;
1903 struct unix_sock *u = unix_sk(sk);
1904 struct scm_cookie scm;
1905 struct sk_buff *skb;
1906 int data_len = 0;
1907 int sk_locked;
1908 long timeo;
1909 int err;
1910
1911 wait_for_unix_gc();
1912 err = scm_send(sock, msg, &scm, false);
1913 if (err < 0)
1914 return err;
1915
1916 err = -EOPNOTSUPP;
1917 if (msg->msg_flags&MSG_OOB)
1918 goto out;
1919
1920 if (msg->msg_namelen) {
1921 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1922 if (err)
1923 goto out;
1924
1925 err = BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk,
1926 msg->msg_name,
1927 &msg->msg_namelen,
1928 NULL);
1929 if (err)
1930 goto out;
1931 } else {
1932 sunaddr = NULL;
1933 err = -ENOTCONN;
1934 other = unix_peer_get(sk);
1935 if (!other)
1936 goto out;
1937 }
1938
1939 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1940 test_bit(SOCK_PASSPIDFD, &sock->flags)) && !u->addr) {
1941 err = unix_autobind(sk);
1942 if (err)
1943 goto out;
1944 }
1945
1946 err = -EMSGSIZE;
1947 if (len > sk->sk_sndbuf - 32)
1948 goto out;
1949
1950 if (len > SKB_MAX_ALLOC) {
1951 data_len = min_t(size_t,
1952 len - SKB_MAX_ALLOC,
1953 MAX_SKB_FRAGS * PAGE_SIZE);
1954 data_len = PAGE_ALIGN(data_len);
1955
1956 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1957 }
1958
1959 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1960 msg->msg_flags & MSG_DONTWAIT, &err,
1961 PAGE_ALLOC_COSTLY_ORDER);
1962 if (skb == NULL)
1963 goto out;
1964
1965 err = unix_scm_to_skb(&scm, skb, true);
1966 if (err < 0)
1967 goto out_free;
1968
1969 skb_put(skb, len - data_len);
1970 skb->data_len = data_len;
1971 skb->len = len;
1972 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1973 if (err)
1974 goto out_free;
1975
1976 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1977
1978restart:
1979 if (!other) {
1980 err = -ECONNRESET;
1981 if (sunaddr == NULL)
1982 goto out_free;
1983
1984 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
1985 sk->sk_type);
1986 if (IS_ERR(other)) {
1987 err = PTR_ERR(other);
1988 other = NULL;
1989 goto out_free;
1990 }
1991 }
1992
1993 if (sk_filter(other, skb) < 0) {
1994 /* Toss the packet but do not return any error to the sender */
1995 err = len;
1996 goto out_free;
1997 }
1998
1999 sk_locked = 0;
2000 unix_state_lock(other);
2001restart_locked:
2002 err = -EPERM;
2003 if (!unix_may_send(sk, other))
2004 goto out_unlock;
2005
2006 if (unlikely(sock_flag(other, SOCK_DEAD))) {
2007 /*
2008 * Check with 1003.1g - what should
2009 * datagram error
2010 */
2011 unix_state_unlock(other);
2012 sock_put(other);
2013
2014 if (!sk_locked)
2015 unix_state_lock(sk);
2016
2017 err = 0;
2018 if (sk->sk_type == SOCK_SEQPACKET) {
2019 /* We are here only when racing with unix_release_sock()
2020 * is clearing @other. Never change state to TCP_CLOSE
2021 * unlike SOCK_DGRAM wants.
2022 */
2023 unix_state_unlock(sk);
2024 err = -EPIPE;
2025 } else if (unix_peer(sk) == other) {
2026 unix_peer(sk) = NULL;
2027 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2028
2029 sk->sk_state = TCP_CLOSE;
2030 unix_state_unlock(sk);
2031
2032 unix_dgram_disconnected(sk, other);
2033 sock_put(other);
2034 err = -ECONNREFUSED;
2035 } else {
2036 unix_state_unlock(sk);
2037 }
2038
2039 other = NULL;
2040 if (err)
2041 goto out_free;
2042 goto restart;
2043 }
2044
2045 err = -EPIPE;
2046 if (other->sk_shutdown & RCV_SHUTDOWN)
2047 goto out_unlock;
2048
2049 if (sk->sk_type != SOCK_SEQPACKET) {
2050 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
2051 if (err)
2052 goto out_unlock;
2053 }
2054
2055 /* other == sk && unix_peer(other) != sk if
2056 * - unix_peer(sk) == NULL, destination address bound to sk
2057 * - unix_peer(sk) == sk by time of get but disconnected before lock
2058 */
2059 if (other != sk &&
2060 unlikely(unix_peer(other) != sk &&
2061 unix_recvq_full_lockless(other))) {
2062 if (timeo) {
2063 timeo = unix_wait_for_peer(other, timeo);
2064
2065 err = sock_intr_errno(timeo);
2066 if (signal_pending(current))
2067 goto out_free;
2068
2069 goto restart;
2070 }
2071
2072 if (!sk_locked) {
2073 unix_state_unlock(other);
2074 unix_state_double_lock(sk, other);
2075 }
2076
2077 if (unix_peer(sk) != other ||
2078 unix_dgram_peer_wake_me(sk, other)) {
2079 err = -EAGAIN;
2080 sk_locked = 1;
2081 goto out_unlock;
2082 }
2083
2084 if (!sk_locked) {
2085 sk_locked = 1;
2086 goto restart_locked;
2087 }
2088 }
2089
2090 if (unlikely(sk_locked))
2091 unix_state_unlock(sk);
2092
2093 if (sock_flag(other, SOCK_RCVTSTAMP))
2094 __net_timestamp(skb);
2095 maybe_add_creds(skb, sock, other);
2096 scm_stat_add(other, skb);
2097 skb_queue_tail(&other->sk_receive_queue, skb);
2098 unix_state_unlock(other);
2099 other->sk_data_ready(other);
2100 sock_put(other);
2101 scm_destroy(&scm);
2102 return len;
2103
2104out_unlock:
2105 if (sk_locked)
2106 unix_state_unlock(sk);
2107 unix_state_unlock(other);
2108out_free:
2109 kfree_skb(skb);
2110out:
2111 if (other)
2112 sock_put(other);
2113 scm_destroy(&scm);
2114 return err;
2115}
2116
2117/* We use paged skbs for stream sockets, and limit occupancy to 32768
2118 * bytes, and a minimum of a full page.
2119 */
2120#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
2121
2122#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2123static int queue_oob(struct socket *sock, struct msghdr *msg, struct sock *other,
2124 struct scm_cookie *scm, bool fds_sent)
2125{
2126 struct unix_sock *ousk = unix_sk(other);
2127 struct sk_buff *skb;
2128 int err = 0;
2129
2130 skb = sock_alloc_send_skb(sock->sk, 1, msg->msg_flags & MSG_DONTWAIT, &err);
2131
2132 if (!skb)
2133 return err;
2134
2135 err = unix_scm_to_skb(scm, skb, !fds_sent);
2136 if (err < 0) {
2137 kfree_skb(skb);
2138 return err;
2139 }
2140 skb_put(skb, 1);
2141 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
2142
2143 if (err) {
2144 kfree_skb(skb);
2145 return err;
2146 }
2147
2148 unix_state_lock(other);
2149
2150 if (sock_flag(other, SOCK_DEAD) ||
2151 (other->sk_shutdown & RCV_SHUTDOWN)) {
2152 unix_state_unlock(other);
2153 kfree_skb(skb);
2154 return -EPIPE;
2155 }
2156
2157 maybe_add_creds(skb, sock, other);
2158 skb_get(skb);
2159
2160 if (ousk->oob_skb)
2161 consume_skb(ousk->oob_skb);
2162
2163 WRITE_ONCE(ousk->oob_skb, skb);
2164
2165 scm_stat_add(other, skb);
2166 skb_queue_tail(&other->sk_receive_queue, skb);
2167 sk_send_sigurg(other);
2168 unix_state_unlock(other);
2169 other->sk_data_ready(other);
2170
2171 return err;
2172}
2173#endif
2174
2175static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2176 size_t len)
2177{
2178 struct sock *sk = sock->sk;
2179 struct sock *other = NULL;
2180 int err, size;
2181 struct sk_buff *skb;
2182 int sent = 0;
2183 struct scm_cookie scm;
2184 bool fds_sent = false;
2185 int data_len;
2186
2187 wait_for_unix_gc();
2188 err = scm_send(sock, msg, &scm, false);
2189 if (err < 0)
2190 return err;
2191
2192 err = -EOPNOTSUPP;
2193 if (msg->msg_flags & MSG_OOB) {
2194#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2195 if (len)
2196 len--;
2197 else
2198#endif
2199 goto out_err;
2200 }
2201
2202 if (msg->msg_namelen) {
2203 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2204 goto out_err;
2205 } else {
2206 err = -ENOTCONN;
2207 other = unix_peer(sk);
2208 if (!other)
2209 goto out_err;
2210 }
2211
2212 if (sk->sk_shutdown & SEND_SHUTDOWN)
2213 goto pipe_err;
2214
2215 while (sent < len) {
2216 size = len - sent;
2217
2218 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2219 skb = sock_alloc_send_pskb(sk, 0, 0,
2220 msg->msg_flags & MSG_DONTWAIT,
2221 &err, 0);
2222 } else {
2223 /* Keep two messages in the pipe so it schedules better */
2224 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
2225
2226 /* allow fallback to order-0 allocations */
2227 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
2228
2229 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
2230
2231 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2232
2233 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
2234 msg->msg_flags & MSG_DONTWAIT, &err,
2235 get_order(UNIX_SKB_FRAGS_SZ));
2236 }
2237 if (!skb)
2238 goto out_err;
2239
2240 /* Only send the fds in the first buffer */
2241 err = unix_scm_to_skb(&scm, skb, !fds_sent);
2242 if (err < 0) {
2243 kfree_skb(skb);
2244 goto out_err;
2245 }
2246 fds_sent = true;
2247
2248 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2249 err = skb_splice_from_iter(skb, &msg->msg_iter, size,
2250 sk->sk_allocation);
2251 if (err < 0) {
2252 kfree_skb(skb);
2253 goto out_err;
2254 }
2255 size = err;
2256 refcount_add(size, &sk->sk_wmem_alloc);
2257 } else {
2258 skb_put(skb, size - data_len);
2259 skb->data_len = data_len;
2260 skb->len = size;
2261 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
2262 if (err) {
2263 kfree_skb(skb);
2264 goto out_err;
2265 }
2266 }
2267
2268 unix_state_lock(other);
2269
2270 if (sock_flag(other, SOCK_DEAD) ||
2271 (other->sk_shutdown & RCV_SHUTDOWN))
2272 goto pipe_err_free;
2273
2274 maybe_add_creds(skb, sock, other);
2275 scm_stat_add(other, skb);
2276 skb_queue_tail(&other->sk_receive_queue, skb);
2277 unix_state_unlock(other);
2278 other->sk_data_ready(other);
2279 sent += size;
2280 }
2281
2282#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2283 if (msg->msg_flags & MSG_OOB) {
2284 err = queue_oob(sock, msg, other, &scm, fds_sent);
2285 if (err)
2286 goto out_err;
2287 sent++;
2288 }
2289#endif
2290
2291 scm_destroy(&scm);
2292
2293 return sent;
2294
2295pipe_err_free:
2296 unix_state_unlock(other);
2297 kfree_skb(skb);
2298pipe_err:
2299 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
2300 send_sig(SIGPIPE, current, 0);
2301 err = -EPIPE;
2302out_err:
2303 scm_destroy(&scm);
2304 return sent ? : err;
2305}
2306
2307static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2308 size_t len)
2309{
2310 int err;
2311 struct sock *sk = sock->sk;
2312
2313 err = sock_error(sk);
2314 if (err)
2315 return err;
2316
2317 if (sk->sk_state != TCP_ESTABLISHED)
2318 return -ENOTCONN;
2319
2320 if (msg->msg_namelen)
2321 msg->msg_namelen = 0;
2322
2323 return unix_dgram_sendmsg(sock, msg, len);
2324}
2325
2326static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2327 size_t size, int flags)
2328{
2329 struct sock *sk = sock->sk;
2330
2331 if (sk->sk_state != TCP_ESTABLISHED)
2332 return -ENOTCONN;
2333
2334 return unix_dgram_recvmsg(sock, msg, size, flags);
2335}
2336
2337static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2338{
2339 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2340
2341 if (addr) {
2342 msg->msg_namelen = addr->len;
2343 memcpy(msg->msg_name, addr->name, addr->len);
2344 }
2345}
2346
2347int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
2348 int flags)
2349{
2350 struct scm_cookie scm;
2351 struct socket *sock = sk->sk_socket;
2352 struct unix_sock *u = unix_sk(sk);
2353 struct sk_buff *skb, *last;
2354 long timeo;
2355 int skip;
2356 int err;
2357
2358 err = -EOPNOTSUPP;
2359 if (flags&MSG_OOB)
2360 goto out;
2361
2362 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2363
2364 do {
2365 mutex_lock(&u->iolock);
2366
2367 skip = sk_peek_offset(sk, flags);
2368 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2369 &skip, &err, &last);
2370 if (skb) {
2371 if (!(flags & MSG_PEEK))
2372 scm_stat_del(sk, skb);
2373 break;
2374 }
2375
2376 mutex_unlock(&u->iolock);
2377
2378 if (err != -EAGAIN)
2379 break;
2380 } while (timeo &&
2381 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2382 &err, &timeo, last));
2383
2384 if (!skb) { /* implies iolock unlocked */
2385 unix_state_lock(sk);
2386 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2387 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2388 (sk->sk_shutdown & RCV_SHUTDOWN))
2389 err = 0;
2390 unix_state_unlock(sk);
2391 goto out;
2392 }
2393
2394 if (wq_has_sleeper(&u->peer_wait))
2395 wake_up_interruptible_sync_poll(&u->peer_wait,
2396 EPOLLOUT | EPOLLWRNORM |
2397 EPOLLWRBAND);
2398
2399 if (msg->msg_name) {
2400 unix_copy_addr(msg, skb->sk);
2401
2402 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2403 msg->msg_name,
2404 &msg->msg_namelen);
2405 }
2406
2407 if (size > skb->len - skip)
2408 size = skb->len - skip;
2409 else if (size < skb->len - skip)
2410 msg->msg_flags |= MSG_TRUNC;
2411
2412 err = skb_copy_datagram_msg(skb, skip, msg, size);
2413 if (err)
2414 goto out_free;
2415
2416 if (sock_flag(sk, SOCK_RCVTSTAMP))
2417 __sock_recv_timestamp(msg, sk, skb);
2418
2419 memset(&scm, 0, sizeof(scm));
2420
2421 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2422 unix_set_secdata(&scm, skb);
2423
2424 if (!(flags & MSG_PEEK)) {
2425 if (UNIXCB(skb).fp)
2426 unix_detach_fds(&scm, skb);
2427
2428 sk_peek_offset_bwd(sk, skb->len);
2429 } else {
2430 /* It is questionable: on PEEK we could:
2431 - do not return fds - good, but too simple 8)
2432 - return fds, and do not return them on read (old strategy,
2433 apparently wrong)
2434 - clone fds (I chose it for now, it is the most universal
2435 solution)
2436
2437 POSIX 1003.1g does not actually define this clearly
2438 at all. POSIX 1003.1g doesn't define a lot of things
2439 clearly however!
2440
2441 */
2442
2443 sk_peek_offset_fwd(sk, size);
2444
2445 if (UNIXCB(skb).fp)
2446 unix_peek_fds(&scm, skb);
2447 }
2448 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2449
2450 scm_recv_unix(sock, msg, &scm, flags);
2451
2452out_free:
2453 skb_free_datagram(sk, skb);
2454 mutex_unlock(&u->iolock);
2455out:
2456 return err;
2457}
2458
2459static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2460 int flags)
2461{
2462 struct sock *sk = sock->sk;
2463
2464#ifdef CONFIG_BPF_SYSCALL
2465 const struct proto *prot = READ_ONCE(sk->sk_prot);
2466
2467 if (prot != &unix_dgram_proto)
2468 return prot->recvmsg(sk, msg, size, flags, NULL);
2469#endif
2470 return __unix_dgram_recvmsg(sk, msg, size, flags);
2471}
2472
2473static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2474{
2475 struct unix_sock *u = unix_sk(sk);
2476 struct sk_buff *skb;
2477 int err;
2478
2479 mutex_lock(&u->iolock);
2480 skb = skb_recv_datagram(sk, MSG_DONTWAIT, &err);
2481 mutex_unlock(&u->iolock);
2482 if (!skb)
2483 return err;
2484
2485 return recv_actor(sk, skb);
2486}
2487
2488/*
2489 * Sleep until more data has arrived. But check for races..
2490 */
2491static long unix_stream_data_wait(struct sock *sk, long timeo,
2492 struct sk_buff *last, unsigned int last_len,
2493 bool freezable)
2494{
2495 unsigned int state = TASK_INTERRUPTIBLE | freezable * TASK_FREEZABLE;
2496 struct sk_buff *tail;
2497 DEFINE_WAIT(wait);
2498
2499 unix_state_lock(sk);
2500
2501 for (;;) {
2502 prepare_to_wait(sk_sleep(sk), &wait, state);
2503
2504 tail = skb_peek_tail(&sk->sk_receive_queue);
2505 if (tail != last ||
2506 (tail && tail->len != last_len) ||
2507 sk->sk_err ||
2508 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2509 signal_pending(current) ||
2510 !timeo)
2511 break;
2512
2513 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2514 unix_state_unlock(sk);
2515 timeo = schedule_timeout(timeo);
2516 unix_state_lock(sk);
2517
2518 if (sock_flag(sk, SOCK_DEAD))
2519 break;
2520
2521 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2522 }
2523
2524 finish_wait(sk_sleep(sk), &wait);
2525 unix_state_unlock(sk);
2526 return timeo;
2527}
2528
2529static unsigned int unix_skb_len(const struct sk_buff *skb)
2530{
2531 return skb->len - UNIXCB(skb).consumed;
2532}
2533
2534struct unix_stream_read_state {
2535 int (*recv_actor)(struct sk_buff *, int, int,
2536 struct unix_stream_read_state *);
2537 struct socket *socket;
2538 struct msghdr *msg;
2539 struct pipe_inode_info *pipe;
2540 size_t size;
2541 int flags;
2542 unsigned int splice_flags;
2543};
2544
2545#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2546static int unix_stream_recv_urg(struct unix_stream_read_state *state)
2547{
2548 struct socket *sock = state->socket;
2549 struct sock *sk = sock->sk;
2550 struct unix_sock *u = unix_sk(sk);
2551 int chunk = 1;
2552 struct sk_buff *oob_skb;
2553
2554 mutex_lock(&u->iolock);
2555 unix_state_lock(sk);
2556
2557 if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb) {
2558 unix_state_unlock(sk);
2559 mutex_unlock(&u->iolock);
2560 return -EINVAL;
2561 }
2562
2563 oob_skb = u->oob_skb;
2564
2565 if (!(state->flags & MSG_PEEK))
2566 WRITE_ONCE(u->oob_skb, NULL);
2567 else
2568 skb_get(oob_skb);
2569 unix_state_unlock(sk);
2570
2571 chunk = state->recv_actor(oob_skb, 0, chunk, state);
2572
2573 if (!(state->flags & MSG_PEEK))
2574 UNIXCB(oob_skb).consumed += 1;
2575
2576 consume_skb(oob_skb);
2577
2578 mutex_unlock(&u->iolock);
2579
2580 if (chunk < 0)
2581 return -EFAULT;
2582
2583 state->msg->msg_flags |= MSG_OOB;
2584 return 1;
2585}
2586
2587static struct sk_buff *manage_oob(struct sk_buff *skb, struct sock *sk,
2588 int flags, int copied)
2589{
2590 struct unix_sock *u = unix_sk(sk);
2591
2592 if (!unix_skb_len(skb) && !(flags & MSG_PEEK)) {
2593 skb_unlink(skb, &sk->sk_receive_queue);
2594 consume_skb(skb);
2595 skb = NULL;
2596 } else {
2597 if (skb == u->oob_skb) {
2598 if (copied) {
2599 skb = NULL;
2600 } else if (sock_flag(sk, SOCK_URGINLINE)) {
2601 if (!(flags & MSG_PEEK)) {
2602 WRITE_ONCE(u->oob_skb, NULL);
2603 consume_skb(skb);
2604 }
2605 } else if (!(flags & MSG_PEEK)) {
2606 skb_unlink(skb, &sk->sk_receive_queue);
2607 consume_skb(skb);
2608 skb = skb_peek(&sk->sk_receive_queue);
2609 }
2610 }
2611 }
2612 return skb;
2613}
2614#endif
2615
2616static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2617{
2618 if (unlikely(sk->sk_state != TCP_ESTABLISHED))
2619 return -ENOTCONN;
2620
2621 return unix_read_skb(sk, recv_actor);
2622}
2623
2624static int unix_stream_read_generic(struct unix_stream_read_state *state,
2625 bool freezable)
2626{
2627 struct scm_cookie scm;
2628 struct socket *sock = state->socket;
2629 struct sock *sk = sock->sk;
2630 struct unix_sock *u = unix_sk(sk);
2631 int copied = 0;
2632 int flags = state->flags;
2633 int noblock = flags & MSG_DONTWAIT;
2634 bool check_creds = false;
2635 int target;
2636 int err = 0;
2637 long timeo;
2638 int skip;
2639 size_t size = state->size;
2640 unsigned int last_len;
2641
2642 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2643 err = -EINVAL;
2644 goto out;
2645 }
2646
2647 if (unlikely(flags & MSG_OOB)) {
2648 err = -EOPNOTSUPP;
2649#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2650 err = unix_stream_recv_urg(state);
2651#endif
2652 goto out;
2653 }
2654
2655 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2656 timeo = sock_rcvtimeo(sk, noblock);
2657
2658 memset(&scm, 0, sizeof(scm));
2659
2660 /* Lock the socket to prevent queue disordering
2661 * while sleeps in memcpy_tomsg
2662 */
2663 mutex_lock(&u->iolock);
2664
2665 skip = max(sk_peek_offset(sk, flags), 0);
2666
2667 do {
2668 int chunk;
2669 bool drop_skb;
2670 struct sk_buff *skb, *last;
2671
2672redo:
2673 unix_state_lock(sk);
2674 if (sock_flag(sk, SOCK_DEAD)) {
2675 err = -ECONNRESET;
2676 goto unlock;
2677 }
2678 last = skb = skb_peek(&sk->sk_receive_queue);
2679 last_len = last ? last->len : 0;
2680
2681#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2682 if (skb) {
2683 skb = manage_oob(skb, sk, flags, copied);
2684 if (!skb) {
2685 unix_state_unlock(sk);
2686 if (copied)
2687 break;
2688 goto redo;
2689 }
2690 }
2691#endif
2692again:
2693 if (skb == NULL) {
2694 if (copied >= target)
2695 goto unlock;
2696
2697 /*
2698 * POSIX 1003.1g mandates this order.
2699 */
2700
2701 err = sock_error(sk);
2702 if (err)
2703 goto unlock;
2704 if (sk->sk_shutdown & RCV_SHUTDOWN)
2705 goto unlock;
2706
2707 unix_state_unlock(sk);
2708 if (!timeo) {
2709 err = -EAGAIN;
2710 break;
2711 }
2712
2713 mutex_unlock(&u->iolock);
2714
2715 timeo = unix_stream_data_wait(sk, timeo, last,
2716 last_len, freezable);
2717
2718 if (signal_pending(current)) {
2719 err = sock_intr_errno(timeo);
2720 scm_destroy(&scm);
2721 goto out;
2722 }
2723
2724 mutex_lock(&u->iolock);
2725 goto redo;
2726unlock:
2727 unix_state_unlock(sk);
2728 break;
2729 }
2730
2731 while (skip >= unix_skb_len(skb)) {
2732 skip -= unix_skb_len(skb);
2733 last = skb;
2734 last_len = skb->len;
2735 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2736 if (!skb)
2737 goto again;
2738 }
2739
2740 unix_state_unlock(sk);
2741
2742 if (check_creds) {
2743 /* Never glue messages from different writers */
2744 if (!unix_skb_scm_eq(skb, &scm))
2745 break;
2746 } else if (test_bit(SOCK_PASSCRED, &sock->flags) ||
2747 test_bit(SOCK_PASSPIDFD, &sock->flags)) {
2748 /* Copy credentials */
2749 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2750 unix_set_secdata(&scm, skb);
2751 check_creds = true;
2752 }
2753
2754 /* Copy address just once */
2755 if (state->msg && state->msg->msg_name) {
2756 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2757 state->msg->msg_name);
2758 unix_copy_addr(state->msg, skb->sk);
2759
2760 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2761 state->msg->msg_name,
2762 &state->msg->msg_namelen);
2763
2764 sunaddr = NULL;
2765 }
2766
2767 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2768 skb_get(skb);
2769 chunk = state->recv_actor(skb, skip, chunk, state);
2770 drop_skb = !unix_skb_len(skb);
2771 /* skb is only safe to use if !drop_skb */
2772 consume_skb(skb);
2773 if (chunk < 0) {
2774 if (copied == 0)
2775 copied = -EFAULT;
2776 break;
2777 }
2778 copied += chunk;
2779 size -= chunk;
2780
2781 if (drop_skb) {
2782 /* the skb was touched by a concurrent reader;
2783 * we should not expect anything from this skb
2784 * anymore and assume it invalid - we can be
2785 * sure it was dropped from the socket queue
2786 *
2787 * let's report a short read
2788 */
2789 err = 0;
2790 break;
2791 }
2792
2793 /* Mark read part of skb as used */
2794 if (!(flags & MSG_PEEK)) {
2795 UNIXCB(skb).consumed += chunk;
2796
2797 sk_peek_offset_bwd(sk, chunk);
2798
2799 if (UNIXCB(skb).fp) {
2800 scm_stat_del(sk, skb);
2801 unix_detach_fds(&scm, skb);
2802 }
2803
2804 if (unix_skb_len(skb))
2805 break;
2806
2807 skb_unlink(skb, &sk->sk_receive_queue);
2808 consume_skb(skb);
2809
2810 if (scm.fp)
2811 break;
2812 } else {
2813 /* It is questionable, see note in unix_dgram_recvmsg.
2814 */
2815 if (UNIXCB(skb).fp)
2816 unix_peek_fds(&scm, skb);
2817
2818 sk_peek_offset_fwd(sk, chunk);
2819
2820 if (UNIXCB(skb).fp)
2821 break;
2822
2823 skip = 0;
2824 last = skb;
2825 last_len = skb->len;
2826 unix_state_lock(sk);
2827 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2828 if (skb)
2829 goto again;
2830 unix_state_unlock(sk);
2831 break;
2832 }
2833 } while (size);
2834
2835 mutex_unlock(&u->iolock);
2836 if (state->msg)
2837 scm_recv_unix(sock, state->msg, &scm, flags);
2838 else
2839 scm_destroy(&scm);
2840out:
2841 return copied ? : err;
2842}
2843
2844static int unix_stream_read_actor(struct sk_buff *skb,
2845 int skip, int chunk,
2846 struct unix_stream_read_state *state)
2847{
2848 int ret;
2849
2850 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2851 state->msg, chunk);
2852 return ret ?: chunk;
2853}
2854
2855int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2856 size_t size, int flags)
2857{
2858 struct unix_stream_read_state state = {
2859 .recv_actor = unix_stream_read_actor,
2860 .socket = sk->sk_socket,
2861 .msg = msg,
2862 .size = size,
2863 .flags = flags
2864 };
2865
2866 return unix_stream_read_generic(&state, true);
2867}
2868
2869static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2870 size_t size, int flags)
2871{
2872 struct unix_stream_read_state state = {
2873 .recv_actor = unix_stream_read_actor,
2874 .socket = sock,
2875 .msg = msg,
2876 .size = size,
2877 .flags = flags
2878 };
2879
2880#ifdef CONFIG_BPF_SYSCALL
2881 struct sock *sk = sock->sk;
2882 const struct proto *prot = READ_ONCE(sk->sk_prot);
2883
2884 if (prot != &unix_stream_proto)
2885 return prot->recvmsg(sk, msg, size, flags, NULL);
2886#endif
2887 return unix_stream_read_generic(&state, true);
2888}
2889
2890static int unix_stream_splice_actor(struct sk_buff *skb,
2891 int skip, int chunk,
2892 struct unix_stream_read_state *state)
2893{
2894 return skb_splice_bits(skb, state->socket->sk,
2895 UNIXCB(skb).consumed + skip,
2896 state->pipe, chunk, state->splice_flags);
2897}
2898
2899static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2900 struct pipe_inode_info *pipe,
2901 size_t size, unsigned int flags)
2902{
2903 struct unix_stream_read_state state = {
2904 .recv_actor = unix_stream_splice_actor,
2905 .socket = sock,
2906 .pipe = pipe,
2907 .size = size,
2908 .splice_flags = flags,
2909 };
2910
2911 if (unlikely(*ppos))
2912 return -ESPIPE;
2913
2914 if (sock->file->f_flags & O_NONBLOCK ||
2915 flags & SPLICE_F_NONBLOCK)
2916 state.flags = MSG_DONTWAIT;
2917
2918 return unix_stream_read_generic(&state, false);
2919}
2920
2921static int unix_shutdown(struct socket *sock, int mode)
2922{
2923 struct sock *sk = sock->sk;
2924 struct sock *other;
2925
2926 if (mode < SHUT_RD || mode > SHUT_RDWR)
2927 return -EINVAL;
2928 /* This maps:
2929 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2930 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2931 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2932 */
2933 ++mode;
2934
2935 unix_state_lock(sk);
2936 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | mode);
2937 other = unix_peer(sk);
2938 if (other)
2939 sock_hold(other);
2940 unix_state_unlock(sk);
2941 sk->sk_state_change(sk);
2942
2943 if (other &&
2944 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2945
2946 int peer_mode = 0;
2947 const struct proto *prot = READ_ONCE(other->sk_prot);
2948
2949 if (prot->unhash)
2950 prot->unhash(other);
2951 if (mode&RCV_SHUTDOWN)
2952 peer_mode |= SEND_SHUTDOWN;
2953 if (mode&SEND_SHUTDOWN)
2954 peer_mode |= RCV_SHUTDOWN;
2955 unix_state_lock(other);
2956 WRITE_ONCE(other->sk_shutdown, other->sk_shutdown | peer_mode);
2957 unix_state_unlock(other);
2958 other->sk_state_change(other);
2959 if (peer_mode == SHUTDOWN_MASK)
2960 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2961 else if (peer_mode & RCV_SHUTDOWN)
2962 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2963 }
2964 if (other)
2965 sock_put(other);
2966
2967 return 0;
2968}
2969
2970long unix_inq_len(struct sock *sk)
2971{
2972 struct sk_buff *skb;
2973 long amount = 0;
2974
2975 if (sk->sk_state == TCP_LISTEN)
2976 return -EINVAL;
2977
2978 spin_lock(&sk->sk_receive_queue.lock);
2979 if (sk->sk_type == SOCK_STREAM ||
2980 sk->sk_type == SOCK_SEQPACKET) {
2981 skb_queue_walk(&sk->sk_receive_queue, skb)
2982 amount += unix_skb_len(skb);
2983 } else {
2984 skb = skb_peek(&sk->sk_receive_queue);
2985 if (skb)
2986 amount = skb->len;
2987 }
2988 spin_unlock(&sk->sk_receive_queue.lock);
2989
2990 return amount;
2991}
2992EXPORT_SYMBOL_GPL(unix_inq_len);
2993
2994long unix_outq_len(struct sock *sk)
2995{
2996 return sk_wmem_alloc_get(sk);
2997}
2998EXPORT_SYMBOL_GPL(unix_outq_len);
2999
3000static int unix_open_file(struct sock *sk)
3001{
3002 struct path path;
3003 struct file *f;
3004 int fd;
3005
3006 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3007 return -EPERM;
3008
3009 if (!smp_load_acquire(&unix_sk(sk)->addr))
3010 return -ENOENT;
3011
3012 path = unix_sk(sk)->path;
3013 if (!path.dentry)
3014 return -ENOENT;
3015
3016 path_get(&path);
3017
3018 fd = get_unused_fd_flags(O_CLOEXEC);
3019 if (fd < 0)
3020 goto out;
3021
3022 f = dentry_open(&path, O_PATH, current_cred());
3023 if (IS_ERR(f)) {
3024 put_unused_fd(fd);
3025 fd = PTR_ERR(f);
3026 goto out;
3027 }
3028
3029 fd_install(fd, f);
3030out:
3031 path_put(&path);
3032
3033 return fd;
3034}
3035
3036static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3037{
3038 struct sock *sk = sock->sk;
3039 long amount = 0;
3040 int err;
3041
3042 switch (cmd) {
3043 case SIOCOUTQ:
3044 amount = unix_outq_len(sk);
3045 err = put_user(amount, (int __user *)arg);
3046 break;
3047 case SIOCINQ:
3048 amount = unix_inq_len(sk);
3049 if (amount < 0)
3050 err = amount;
3051 else
3052 err = put_user(amount, (int __user *)arg);
3053 break;
3054 case SIOCUNIXFILE:
3055 err = unix_open_file(sk);
3056 break;
3057#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3058 case SIOCATMARK:
3059 {
3060 struct sk_buff *skb;
3061 int answ = 0;
3062
3063 skb = skb_peek(&sk->sk_receive_queue);
3064 if (skb && skb == READ_ONCE(unix_sk(sk)->oob_skb))
3065 answ = 1;
3066 err = put_user(answ, (int __user *)arg);
3067 }
3068 break;
3069#endif
3070 default:
3071 err = -ENOIOCTLCMD;
3072 break;
3073 }
3074 return err;
3075}
3076
3077#ifdef CONFIG_COMPAT
3078static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3079{
3080 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
3081}
3082#endif
3083
3084static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
3085{
3086 struct sock *sk = sock->sk;
3087 __poll_t mask;
3088 u8 shutdown;
3089
3090 sock_poll_wait(file, sock, wait);
3091 mask = 0;
3092 shutdown = READ_ONCE(sk->sk_shutdown);
3093
3094 /* exceptional events? */
3095 if (READ_ONCE(sk->sk_err))
3096 mask |= EPOLLERR;
3097 if (shutdown == SHUTDOWN_MASK)
3098 mask |= EPOLLHUP;
3099 if (shutdown & RCV_SHUTDOWN)
3100 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3101
3102 /* readable? */
3103 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3104 mask |= EPOLLIN | EPOLLRDNORM;
3105 if (sk_is_readable(sk))
3106 mask |= EPOLLIN | EPOLLRDNORM;
3107#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3108 if (READ_ONCE(unix_sk(sk)->oob_skb))
3109 mask |= EPOLLPRI;
3110#endif
3111
3112 /* Connection-based need to check for termination and startup */
3113 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3114 sk->sk_state == TCP_CLOSE)
3115 mask |= EPOLLHUP;
3116
3117 /*
3118 * we set writable also when the other side has shut down the
3119 * connection. This prevents stuck sockets.
3120 */
3121 if (unix_writable(sk))
3122 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3123
3124 return mask;
3125}
3126
3127static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3128 poll_table *wait)
3129{
3130 struct sock *sk = sock->sk, *other;
3131 unsigned int writable;
3132 __poll_t mask;
3133 u8 shutdown;
3134
3135 sock_poll_wait(file, sock, wait);
3136 mask = 0;
3137 shutdown = READ_ONCE(sk->sk_shutdown);
3138
3139 /* exceptional events? */
3140 if (READ_ONCE(sk->sk_err) ||
3141 !skb_queue_empty_lockless(&sk->sk_error_queue))
3142 mask |= EPOLLERR |
3143 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
3144
3145 if (shutdown & RCV_SHUTDOWN)
3146 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3147 if (shutdown == SHUTDOWN_MASK)
3148 mask |= EPOLLHUP;
3149
3150 /* readable? */
3151 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3152 mask |= EPOLLIN | EPOLLRDNORM;
3153 if (sk_is_readable(sk))
3154 mask |= EPOLLIN | EPOLLRDNORM;
3155
3156 /* Connection-based need to check for termination and startup */
3157 if (sk->sk_type == SOCK_SEQPACKET) {
3158 if (sk->sk_state == TCP_CLOSE)
3159 mask |= EPOLLHUP;
3160 /* connection hasn't started yet? */
3161 if (sk->sk_state == TCP_SYN_SENT)
3162 return mask;
3163 }
3164
3165 /* No write status requested, avoid expensive OUT tests. */
3166 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
3167 return mask;
3168
3169 writable = unix_writable(sk);
3170 if (writable) {
3171 unix_state_lock(sk);
3172
3173 other = unix_peer(sk);
3174 if (other && unix_peer(other) != sk &&
3175 unix_recvq_full_lockless(other) &&
3176 unix_dgram_peer_wake_me(sk, other))
3177 writable = 0;
3178
3179 unix_state_unlock(sk);
3180 }
3181
3182 if (writable)
3183 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3184 else
3185 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3186
3187 return mask;
3188}
3189
3190#ifdef CONFIG_PROC_FS
3191
3192#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3193
3194#define get_bucket(x) ((x) >> BUCKET_SPACE)
3195#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
3196#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
3197
3198static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
3199{
3200 unsigned long offset = get_offset(*pos);
3201 unsigned long bucket = get_bucket(*pos);
3202 unsigned long count = 0;
3203 struct sock *sk;
3204
3205 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3206 sk; sk = sk_next(sk)) {
3207 if (++count == offset)
3208 break;
3209 }
3210
3211 return sk;
3212}
3213
3214static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
3215{
3216 unsigned long bucket = get_bucket(*pos);
3217 struct net *net = seq_file_net(seq);
3218 struct sock *sk;
3219
3220 while (bucket < UNIX_HASH_SIZE) {
3221 spin_lock(&net->unx.table.locks[bucket]);
3222
3223 sk = unix_from_bucket(seq, pos);
3224 if (sk)
3225 return sk;
3226
3227 spin_unlock(&net->unx.table.locks[bucket]);
3228
3229 *pos = set_bucket_offset(++bucket, 1);
3230 }
3231
3232 return NULL;
3233}
3234
3235static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
3236 loff_t *pos)
3237{
3238 unsigned long bucket = get_bucket(*pos);
3239
3240 sk = sk_next(sk);
3241 if (sk)
3242 return sk;
3243
3244
3245 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
3246
3247 *pos = set_bucket_offset(++bucket, 1);
3248
3249 return unix_get_first(seq, pos);
3250}
3251
3252static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3253{
3254 if (!*pos)
3255 return SEQ_START_TOKEN;
3256
3257 return unix_get_first(seq, pos);
3258}
3259
3260static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3261{
3262 ++*pos;
3263
3264 if (v == SEQ_START_TOKEN)
3265 return unix_get_first(seq, pos);
3266
3267 return unix_get_next(seq, v, pos);
3268}
3269
3270static void unix_seq_stop(struct seq_file *seq, void *v)
3271{
3272 struct sock *sk = v;
3273
3274 if (sk)
3275 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
3276}
3277
3278static int unix_seq_show(struct seq_file *seq, void *v)
3279{
3280
3281 if (v == SEQ_START_TOKEN)
3282 seq_puts(seq, "Num RefCount Protocol Flags Type St "
3283 "Inode Path\n");
3284 else {
3285 struct sock *s = v;
3286 struct unix_sock *u = unix_sk(s);
3287 unix_state_lock(s);
3288
3289 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
3290 s,
3291 refcount_read(&s->sk_refcnt),
3292 0,
3293 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
3294 s->sk_type,
3295 s->sk_socket ?
3296 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
3297 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
3298 sock_i_ino(s));
3299
3300 if (u->addr) { // under a hash table lock here
3301 int i, len;
3302 seq_putc(seq, ' ');
3303
3304 i = 0;
3305 len = u->addr->len -
3306 offsetof(struct sockaddr_un, sun_path);
3307 if (u->addr->name->sun_path[0]) {
3308 len--;
3309 } else {
3310 seq_putc(seq, '@');
3311 i++;
3312 }
3313 for ( ; i < len; i++)
3314 seq_putc(seq, u->addr->name->sun_path[i] ?:
3315 '@');
3316 }
3317 unix_state_unlock(s);
3318 seq_putc(seq, '\n');
3319 }
3320
3321 return 0;
3322}
3323
3324static const struct seq_operations unix_seq_ops = {
3325 .start = unix_seq_start,
3326 .next = unix_seq_next,
3327 .stop = unix_seq_stop,
3328 .show = unix_seq_show,
3329};
3330
3331#ifdef CONFIG_BPF_SYSCALL
3332struct bpf_unix_iter_state {
3333 struct seq_net_private p;
3334 unsigned int cur_sk;
3335 unsigned int end_sk;
3336 unsigned int max_sk;
3337 struct sock **batch;
3338 bool st_bucket_done;
3339};
3340
3341struct bpf_iter__unix {
3342 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3343 __bpf_md_ptr(struct unix_sock *, unix_sk);
3344 uid_t uid __aligned(8);
3345};
3346
3347static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3348 struct unix_sock *unix_sk, uid_t uid)
3349{
3350 struct bpf_iter__unix ctx;
3351
3352 meta->seq_num--; /* skip SEQ_START_TOKEN */
3353 ctx.meta = meta;
3354 ctx.unix_sk = unix_sk;
3355 ctx.uid = uid;
3356 return bpf_iter_run_prog(prog, &ctx);
3357}
3358
3359static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
3360
3361{
3362 struct bpf_unix_iter_state *iter = seq->private;
3363 unsigned int expected = 1;
3364 struct sock *sk;
3365
3366 sock_hold(start_sk);
3367 iter->batch[iter->end_sk++] = start_sk;
3368
3369 for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
3370 if (iter->end_sk < iter->max_sk) {
3371 sock_hold(sk);
3372 iter->batch[iter->end_sk++] = sk;
3373 }
3374
3375 expected++;
3376 }
3377
3378 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
3379
3380 return expected;
3381}
3382
3383static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
3384{
3385 while (iter->cur_sk < iter->end_sk)
3386 sock_put(iter->batch[iter->cur_sk++]);
3387}
3388
3389static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
3390 unsigned int new_batch_sz)
3391{
3392 struct sock **new_batch;
3393
3394 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3395 GFP_USER | __GFP_NOWARN);
3396 if (!new_batch)
3397 return -ENOMEM;
3398
3399 bpf_iter_unix_put_batch(iter);
3400 kvfree(iter->batch);
3401 iter->batch = new_batch;
3402 iter->max_sk = new_batch_sz;
3403
3404 return 0;
3405}
3406
3407static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
3408 loff_t *pos)
3409{
3410 struct bpf_unix_iter_state *iter = seq->private;
3411 unsigned int expected;
3412 bool resized = false;
3413 struct sock *sk;
3414
3415 if (iter->st_bucket_done)
3416 *pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
3417
3418again:
3419 /* Get a new batch */
3420 iter->cur_sk = 0;
3421 iter->end_sk = 0;
3422
3423 sk = unix_get_first(seq, pos);
3424 if (!sk)
3425 return NULL; /* Done */
3426
3427 expected = bpf_iter_unix_hold_batch(seq, sk);
3428
3429 if (iter->end_sk == expected) {
3430 iter->st_bucket_done = true;
3431 return sk;
3432 }
3433
3434 if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
3435 resized = true;
3436 goto again;
3437 }
3438
3439 return sk;
3440}
3441
3442static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
3443{
3444 if (!*pos)
3445 return SEQ_START_TOKEN;
3446
3447 /* bpf iter does not support lseek, so it always
3448 * continue from where it was stop()-ped.
3449 */
3450 return bpf_iter_unix_batch(seq, pos);
3451}
3452
3453static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3454{
3455 struct bpf_unix_iter_state *iter = seq->private;
3456 struct sock *sk;
3457
3458 /* Whenever seq_next() is called, the iter->cur_sk is
3459 * done with seq_show(), so advance to the next sk in
3460 * the batch.
3461 */
3462 if (iter->cur_sk < iter->end_sk)
3463 sock_put(iter->batch[iter->cur_sk++]);
3464
3465 ++*pos;
3466
3467 if (iter->cur_sk < iter->end_sk)
3468 sk = iter->batch[iter->cur_sk];
3469 else
3470 sk = bpf_iter_unix_batch(seq, pos);
3471
3472 return sk;
3473}
3474
3475static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
3476{
3477 struct bpf_iter_meta meta;
3478 struct bpf_prog *prog;
3479 struct sock *sk = v;
3480 uid_t uid;
3481 bool slow;
3482 int ret;
3483
3484 if (v == SEQ_START_TOKEN)
3485 return 0;
3486
3487 slow = lock_sock_fast(sk);
3488
3489 if (unlikely(sk_unhashed(sk))) {
3490 ret = SEQ_SKIP;
3491 goto unlock;
3492 }
3493
3494 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3495 meta.seq = seq;
3496 prog = bpf_iter_get_info(&meta, false);
3497 ret = unix_prog_seq_show(prog, &meta, v, uid);
3498unlock:
3499 unlock_sock_fast(sk, slow);
3500 return ret;
3501}
3502
3503static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3504{
3505 struct bpf_unix_iter_state *iter = seq->private;
3506 struct bpf_iter_meta meta;
3507 struct bpf_prog *prog;
3508
3509 if (!v) {
3510 meta.seq = seq;
3511 prog = bpf_iter_get_info(&meta, true);
3512 if (prog)
3513 (void)unix_prog_seq_show(prog, &meta, v, 0);
3514 }
3515
3516 if (iter->cur_sk < iter->end_sk)
3517 bpf_iter_unix_put_batch(iter);
3518}
3519
3520static const struct seq_operations bpf_iter_unix_seq_ops = {
3521 .start = bpf_iter_unix_seq_start,
3522 .next = bpf_iter_unix_seq_next,
3523 .stop = bpf_iter_unix_seq_stop,
3524 .show = bpf_iter_unix_seq_show,
3525};
3526#endif
3527#endif
3528
3529static const struct net_proto_family unix_family_ops = {
3530 .family = PF_UNIX,
3531 .create = unix_create,
3532 .owner = THIS_MODULE,
3533};
3534
3535
3536static int __net_init unix_net_init(struct net *net)
3537{
3538 int i;
3539
3540 net->unx.sysctl_max_dgram_qlen = 10;
3541 if (unix_sysctl_register(net))
3542 goto out;
3543
3544#ifdef CONFIG_PROC_FS
3545 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
3546 sizeof(struct seq_net_private)))
3547 goto err_sysctl;
3548#endif
3549
3550 net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
3551 sizeof(spinlock_t), GFP_KERNEL);
3552 if (!net->unx.table.locks)
3553 goto err_proc;
3554
3555 net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
3556 sizeof(struct hlist_head),
3557 GFP_KERNEL);
3558 if (!net->unx.table.buckets)
3559 goto free_locks;
3560
3561 for (i = 0; i < UNIX_HASH_SIZE; i++) {
3562 spin_lock_init(&net->unx.table.locks[i]);
3563 INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
3564 }
3565
3566 return 0;
3567
3568free_locks:
3569 kvfree(net->unx.table.locks);
3570err_proc:
3571#ifdef CONFIG_PROC_FS
3572 remove_proc_entry("unix", net->proc_net);
3573err_sysctl:
3574#endif
3575 unix_sysctl_unregister(net);
3576out:
3577 return -ENOMEM;
3578}
3579
3580static void __net_exit unix_net_exit(struct net *net)
3581{
3582 kvfree(net->unx.table.buckets);
3583 kvfree(net->unx.table.locks);
3584 unix_sysctl_unregister(net);
3585 remove_proc_entry("unix", net->proc_net);
3586}
3587
3588static struct pernet_operations unix_net_ops = {
3589 .init = unix_net_init,
3590 .exit = unix_net_exit,
3591};
3592
3593#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3594DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
3595 struct unix_sock *unix_sk, uid_t uid)
3596
3597#define INIT_BATCH_SZ 16
3598
3599static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
3600{
3601 struct bpf_unix_iter_state *iter = priv_data;
3602 int err;
3603
3604 err = bpf_iter_init_seq_net(priv_data, aux);
3605 if (err)
3606 return err;
3607
3608 err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
3609 if (err) {
3610 bpf_iter_fini_seq_net(priv_data);
3611 return err;
3612 }
3613
3614 return 0;
3615}
3616
3617static void bpf_iter_fini_unix(void *priv_data)
3618{
3619 struct bpf_unix_iter_state *iter = priv_data;
3620
3621 bpf_iter_fini_seq_net(priv_data);
3622 kvfree(iter->batch);
3623}
3624
3625static const struct bpf_iter_seq_info unix_seq_info = {
3626 .seq_ops = &bpf_iter_unix_seq_ops,
3627 .init_seq_private = bpf_iter_init_unix,
3628 .fini_seq_private = bpf_iter_fini_unix,
3629 .seq_priv_size = sizeof(struct bpf_unix_iter_state),
3630};
3631
3632static const struct bpf_func_proto *
3633bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
3634 const struct bpf_prog *prog)
3635{
3636 switch (func_id) {
3637 case BPF_FUNC_setsockopt:
3638 return &bpf_sk_setsockopt_proto;
3639 case BPF_FUNC_getsockopt:
3640 return &bpf_sk_getsockopt_proto;
3641 default:
3642 return NULL;
3643 }
3644}
3645
3646static struct bpf_iter_reg unix_reg_info = {
3647 .target = "unix",
3648 .ctx_arg_info_size = 1,
3649 .ctx_arg_info = {
3650 { offsetof(struct bpf_iter__unix, unix_sk),
3651 PTR_TO_BTF_ID_OR_NULL },
3652 },
3653 .get_func_proto = bpf_iter_unix_get_func_proto,
3654 .seq_info = &unix_seq_info,
3655};
3656
3657static void __init bpf_iter_register(void)
3658{
3659 unix_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UNIX];
3660 if (bpf_iter_reg_target(&unix_reg_info))
3661 pr_warn("Warning: could not register bpf iterator unix\n");
3662}
3663#endif
3664
3665static int __init af_unix_init(void)
3666{
3667 int i, rc = -1;
3668
3669 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
3670
3671 for (i = 0; i < UNIX_HASH_SIZE / 2; i++) {
3672 spin_lock_init(&bsd_socket_locks[i]);
3673 INIT_HLIST_HEAD(&bsd_socket_buckets[i]);
3674 }
3675
3676 rc = proto_register(&unix_dgram_proto, 1);
3677 if (rc != 0) {
3678 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3679 goto out;
3680 }
3681
3682 rc = proto_register(&unix_stream_proto, 1);
3683 if (rc != 0) {
3684 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3685 proto_unregister(&unix_dgram_proto);
3686 goto out;
3687 }
3688
3689 sock_register(&unix_family_ops);
3690 register_pernet_subsys(&unix_net_ops);
3691 unix_bpf_build_proto();
3692
3693#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3694 bpf_iter_register();
3695#endif
3696
3697out:
3698 return rc;
3699}
3700
3701/* Later than subsys_initcall() because we depend on stuff initialised there */
3702fs_initcall(af_unix_init);