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 <asm/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->readlock))
665 return -EINTR;
666
667 sk->sk_peek_off = val;
668 mutex_unlock(&u->readlock);
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_write_space = unix_write_space;
773 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
774 sk->sk_destruct = unix_sock_destructor;
775 u = unix_sk(sk);
776 u->path.dentry = NULL;
777 u->path.mnt = NULL;
778 spin_lock_init(&u->lock);
779 atomic_long_set(&u->inflight, 0);
780 INIT_LIST_HEAD(&u->link);
781 mutex_init(&u->readlock); /* single task reading lock */
782 init_waitqueue_head(&u->peer_wait);
783 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
784 unix_insert_socket(unix_sockets_unbound(sk), sk);
785out:
786 if (sk == NULL)
787 atomic_long_dec(&unix_nr_socks);
788 else {
789 local_bh_disable();
790 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
791 local_bh_enable();
792 }
793 return sk;
794}
795
796static int unix_create(struct net *net, struct socket *sock, int protocol,
797 int kern)
798{
799 if (protocol && protocol != PF_UNIX)
800 return -EPROTONOSUPPORT;
801
802 sock->state = SS_UNCONNECTED;
803
804 switch (sock->type) {
805 case SOCK_STREAM:
806 sock->ops = &unix_stream_ops;
807 break;
808 /*
809 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
810 * nothing uses it.
811 */
812 case SOCK_RAW:
813 sock->type = SOCK_DGRAM;
814 case SOCK_DGRAM:
815 sock->ops = &unix_dgram_ops;
816 break;
817 case SOCK_SEQPACKET:
818 sock->ops = &unix_seqpacket_ops;
819 break;
820 default:
821 return -ESOCKTNOSUPPORT;
822 }
823
824 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
825}
826
827static int unix_release(struct socket *sock)
828{
829 struct sock *sk = sock->sk;
830
831 if (!sk)
832 return 0;
833
834 unix_release_sock(sk, 0);
835 sock->sk = NULL;
836
837 return 0;
838}
839
840static int unix_autobind(struct socket *sock)
841{
842 struct sock *sk = sock->sk;
843 struct net *net = sock_net(sk);
844 struct unix_sock *u = unix_sk(sk);
845 static u32 ordernum = 1;
846 struct unix_address *addr;
847 int err;
848 unsigned int retries = 0;
849
850 err = mutex_lock_interruptible(&u->readlock);
851 if (err)
852 return err;
853
854 err = 0;
855 if (u->addr)
856 goto out;
857
858 err = -ENOMEM;
859 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
860 if (!addr)
861 goto out;
862
863 addr->name->sun_family = AF_UNIX;
864 atomic_set(&addr->refcnt, 1);
865
866retry:
867 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
868 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
869
870 spin_lock(&unix_table_lock);
871 ordernum = (ordernum+1)&0xFFFFF;
872
873 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
874 addr->hash)) {
875 spin_unlock(&unix_table_lock);
876 /*
877 * __unix_find_socket_byname() may take long time if many names
878 * are already in use.
879 */
880 cond_resched();
881 /* Give up if all names seems to be in use. */
882 if (retries++ == 0xFFFFF) {
883 err = -ENOSPC;
884 kfree(addr);
885 goto out;
886 }
887 goto retry;
888 }
889 addr->hash ^= sk->sk_type;
890
891 __unix_remove_socket(sk);
892 u->addr = addr;
893 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
894 spin_unlock(&unix_table_lock);
895 err = 0;
896
897out: mutex_unlock(&u->readlock);
898 return err;
899}
900
901static struct sock *unix_find_other(struct net *net,
902 struct sockaddr_un *sunname, int len,
903 int type, unsigned int hash, int *error)
904{
905 struct sock *u;
906 struct path path;
907 int err = 0;
908
909 if (sunname->sun_path[0]) {
910 struct inode *inode;
911 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
912 if (err)
913 goto fail;
914 inode = d_backing_inode(path.dentry);
915 err = inode_permission(inode, MAY_WRITE);
916 if (err)
917 goto put_fail;
918
919 err = -ECONNREFUSED;
920 if (!S_ISSOCK(inode->i_mode))
921 goto put_fail;
922 u = unix_find_socket_byinode(inode);
923 if (!u)
924 goto put_fail;
925
926 if (u->sk_type == type)
927 touch_atime(&path);
928
929 path_put(&path);
930
931 err = -EPROTOTYPE;
932 if (u->sk_type != type) {
933 sock_put(u);
934 goto fail;
935 }
936 } else {
937 err = -ECONNREFUSED;
938 u = unix_find_socket_byname(net, sunname, len, type, hash);
939 if (u) {
940 struct dentry *dentry;
941 dentry = unix_sk(u)->path.dentry;
942 if (dentry)
943 touch_atime(&unix_sk(u)->path);
944 } else
945 goto fail;
946 }
947 return u;
948
949put_fail:
950 path_put(&path);
951fail:
952 *error = err;
953 return NULL;
954}
955
956static int unix_mknod(struct dentry *dentry, struct path *path, umode_t mode,
957 struct path *res)
958{
959 int err;
960
961 err = security_path_mknod(path, dentry, mode, 0);
962 if (!err) {
963 err = vfs_mknod(d_inode(path->dentry), dentry, mode, 0);
964 if (!err) {
965 res->mnt = mntget(path->mnt);
966 res->dentry = dget(dentry);
967 }
968 }
969
970 return err;
971}
972
973static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
974{
975 struct sock *sk = sock->sk;
976 struct net *net = sock_net(sk);
977 struct unix_sock *u = unix_sk(sk);
978 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
979 char *sun_path = sunaddr->sun_path;
980 int err, name_err;
981 unsigned int hash;
982 struct unix_address *addr;
983 struct hlist_head *list;
984 struct path path;
985 struct dentry *dentry;
986
987 err = -EINVAL;
988 if (sunaddr->sun_family != AF_UNIX)
989 goto out;
990
991 if (addr_len == sizeof(short)) {
992 err = unix_autobind(sock);
993 goto out;
994 }
995
996 err = unix_mkname(sunaddr, addr_len, &hash);
997 if (err < 0)
998 goto out;
999 addr_len = err;
1000
1001 name_err = 0;
1002 dentry = NULL;
1003 if (sun_path[0]) {
1004 /* Get the parent directory, calculate the hash for last
1005 * component.
1006 */
1007 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
1008
1009 if (IS_ERR(dentry)) {
1010 /* delay report until after 'already bound' check */
1011 name_err = PTR_ERR(dentry);
1012 dentry = NULL;
1013 }
1014 }
1015
1016 err = mutex_lock_interruptible(&u->readlock);
1017 if (err)
1018 goto out_path;
1019
1020 err = -EINVAL;
1021 if (u->addr)
1022 goto out_up;
1023
1024 if (name_err) {
1025 err = name_err == -EEXIST ? -EADDRINUSE : name_err;
1026 goto out_up;
1027 }
1028
1029 err = -ENOMEM;
1030 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1031 if (!addr)
1032 goto out_up;
1033
1034 memcpy(addr->name, sunaddr, addr_len);
1035 addr->len = addr_len;
1036 addr->hash = hash ^ sk->sk_type;
1037 atomic_set(&addr->refcnt, 1);
1038
1039 if (dentry) {
1040 struct path u_path;
1041 umode_t mode = S_IFSOCK |
1042 (SOCK_INODE(sock)->i_mode & ~current_umask());
1043 err = unix_mknod(dentry, &path, mode, &u_path);
1044 if (err) {
1045 if (err == -EEXIST)
1046 err = -EADDRINUSE;
1047 unix_release_addr(addr);
1048 goto out_up;
1049 }
1050 addr->hash = UNIX_HASH_SIZE;
1051 hash = d_backing_inode(dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1052 spin_lock(&unix_table_lock);
1053 u->path = u_path;
1054 list = &unix_socket_table[hash];
1055 } else {
1056 spin_lock(&unix_table_lock);
1057 err = -EADDRINUSE;
1058 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1059 sk->sk_type, hash)) {
1060 unix_release_addr(addr);
1061 goto out_unlock;
1062 }
1063
1064 list = &unix_socket_table[addr->hash];
1065 }
1066
1067 err = 0;
1068 __unix_remove_socket(sk);
1069 u->addr = addr;
1070 __unix_insert_socket(list, sk);
1071
1072out_unlock:
1073 spin_unlock(&unix_table_lock);
1074out_up:
1075 mutex_unlock(&u->readlock);
1076out_path:
1077 if (dentry)
1078 done_path_create(&path, dentry);
1079
1080out:
1081 return err;
1082}
1083
1084static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1085{
1086 if (unlikely(sk1 == sk2) || !sk2) {
1087 unix_state_lock(sk1);
1088 return;
1089 }
1090 if (sk1 < sk2) {
1091 unix_state_lock(sk1);
1092 unix_state_lock_nested(sk2);
1093 } else {
1094 unix_state_lock(sk2);
1095 unix_state_lock_nested(sk1);
1096 }
1097}
1098
1099static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1100{
1101 if (unlikely(sk1 == sk2) || !sk2) {
1102 unix_state_unlock(sk1);
1103 return;
1104 }
1105 unix_state_unlock(sk1);
1106 unix_state_unlock(sk2);
1107}
1108
1109static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1110 int alen, int flags)
1111{
1112 struct sock *sk = sock->sk;
1113 struct net *net = sock_net(sk);
1114 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1115 struct sock *other;
1116 unsigned int hash;
1117 int err;
1118
1119 if (addr->sa_family != AF_UNSPEC) {
1120 err = unix_mkname(sunaddr, alen, &hash);
1121 if (err < 0)
1122 goto out;
1123 alen = err;
1124
1125 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1126 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1127 goto out;
1128
1129restart:
1130 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1131 if (!other)
1132 goto out;
1133
1134 unix_state_double_lock(sk, other);
1135
1136 /* Apparently VFS overslept socket death. Retry. */
1137 if (sock_flag(other, SOCK_DEAD)) {
1138 unix_state_double_unlock(sk, other);
1139 sock_put(other);
1140 goto restart;
1141 }
1142
1143 err = -EPERM;
1144 if (!unix_may_send(sk, other))
1145 goto out_unlock;
1146
1147 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1148 if (err)
1149 goto out_unlock;
1150
1151 } else {
1152 /*
1153 * 1003.1g breaking connected state with AF_UNSPEC
1154 */
1155 other = NULL;
1156 unix_state_double_lock(sk, other);
1157 }
1158
1159 /*
1160 * If it was connected, reconnect.
1161 */
1162 if (unix_peer(sk)) {
1163 struct sock *old_peer = unix_peer(sk);
1164 unix_peer(sk) = other;
1165 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1166
1167 unix_state_double_unlock(sk, other);
1168
1169 if (other != old_peer)
1170 unix_dgram_disconnected(sk, old_peer);
1171 sock_put(old_peer);
1172 } else {
1173 unix_peer(sk) = other;
1174 unix_state_double_unlock(sk, other);
1175 }
1176 return 0;
1177
1178out_unlock:
1179 unix_state_double_unlock(sk, other);
1180 sock_put(other);
1181out:
1182 return err;
1183}
1184
1185static long unix_wait_for_peer(struct sock *other, long timeo)
1186{
1187 struct unix_sock *u = unix_sk(other);
1188 int sched;
1189 DEFINE_WAIT(wait);
1190
1191 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1192
1193 sched = !sock_flag(other, SOCK_DEAD) &&
1194 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1195 unix_recvq_full(other);
1196
1197 unix_state_unlock(other);
1198
1199 if (sched)
1200 timeo = schedule_timeout(timeo);
1201
1202 finish_wait(&u->peer_wait, &wait);
1203 return timeo;
1204}
1205
1206static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1207 int addr_len, int flags)
1208{
1209 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1210 struct sock *sk = sock->sk;
1211 struct net *net = sock_net(sk);
1212 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1213 struct sock *newsk = NULL;
1214 struct sock *other = NULL;
1215 struct sk_buff *skb = NULL;
1216 unsigned int hash;
1217 int st;
1218 int err;
1219 long timeo;
1220
1221 err = unix_mkname(sunaddr, addr_len, &hash);
1222 if (err < 0)
1223 goto out;
1224 addr_len = err;
1225
1226 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1227 (err = unix_autobind(sock)) != 0)
1228 goto out;
1229
1230 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1231
1232 /* First of all allocate resources.
1233 If we will make it after state is locked,
1234 we will have to recheck all again in any case.
1235 */
1236
1237 err = -ENOMEM;
1238
1239 /* create new sock for complete connection */
1240 newsk = unix_create1(sock_net(sk), NULL, 0);
1241 if (newsk == NULL)
1242 goto out;
1243
1244 /* Allocate skb for sending to listening sock */
1245 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1246 if (skb == NULL)
1247 goto out;
1248
1249restart:
1250 /* Find listening sock. */
1251 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1252 if (!other)
1253 goto out;
1254
1255 /* Latch state of peer */
1256 unix_state_lock(other);
1257
1258 /* Apparently VFS overslept socket death. Retry. */
1259 if (sock_flag(other, SOCK_DEAD)) {
1260 unix_state_unlock(other);
1261 sock_put(other);
1262 goto restart;
1263 }
1264
1265 err = -ECONNREFUSED;
1266 if (other->sk_state != TCP_LISTEN)
1267 goto out_unlock;
1268 if (other->sk_shutdown & RCV_SHUTDOWN)
1269 goto out_unlock;
1270
1271 if (unix_recvq_full(other)) {
1272 err = -EAGAIN;
1273 if (!timeo)
1274 goto out_unlock;
1275
1276 timeo = unix_wait_for_peer(other, timeo);
1277
1278 err = sock_intr_errno(timeo);
1279 if (signal_pending(current))
1280 goto out;
1281 sock_put(other);
1282 goto restart;
1283 }
1284
1285 /* Latch our state.
1286
1287 It is tricky place. We need to grab our state lock and cannot
1288 drop lock on peer. It is dangerous because deadlock is
1289 possible. Connect to self case and simultaneous
1290 attempt to connect are eliminated by checking socket
1291 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1292 check this before attempt to grab lock.
1293
1294 Well, and we have to recheck the state after socket locked.
1295 */
1296 st = sk->sk_state;
1297
1298 switch (st) {
1299 case TCP_CLOSE:
1300 /* This is ok... continue with connect */
1301 break;
1302 case TCP_ESTABLISHED:
1303 /* Socket is already connected */
1304 err = -EISCONN;
1305 goto out_unlock;
1306 default:
1307 err = -EINVAL;
1308 goto out_unlock;
1309 }
1310
1311 unix_state_lock_nested(sk);
1312
1313 if (sk->sk_state != st) {
1314 unix_state_unlock(sk);
1315 unix_state_unlock(other);
1316 sock_put(other);
1317 goto restart;
1318 }
1319
1320 err = security_unix_stream_connect(sk, other, newsk);
1321 if (err) {
1322 unix_state_unlock(sk);
1323 goto out_unlock;
1324 }
1325
1326 /* The way is open! Fastly set all the necessary fields... */
1327
1328 sock_hold(sk);
1329 unix_peer(newsk) = sk;
1330 newsk->sk_state = TCP_ESTABLISHED;
1331 newsk->sk_type = sk->sk_type;
1332 init_peercred(newsk);
1333 newu = unix_sk(newsk);
1334 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1335 otheru = unix_sk(other);
1336
1337 /* copy address information from listening to new sock*/
1338 if (otheru->addr) {
1339 atomic_inc(&otheru->addr->refcnt);
1340 newu->addr = otheru->addr;
1341 }
1342 if (otheru->path.dentry) {
1343 path_get(&otheru->path);
1344 newu->path = otheru->path;
1345 }
1346
1347 /* Set credentials */
1348 copy_peercred(sk, other);
1349
1350 sock->state = SS_CONNECTED;
1351 sk->sk_state = TCP_ESTABLISHED;
1352 sock_hold(newsk);
1353
1354 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1355 unix_peer(sk) = newsk;
1356
1357 unix_state_unlock(sk);
1358
1359 /* take ten and and send info to listening sock */
1360 spin_lock(&other->sk_receive_queue.lock);
1361 __skb_queue_tail(&other->sk_receive_queue, skb);
1362 spin_unlock(&other->sk_receive_queue.lock);
1363 unix_state_unlock(other);
1364 other->sk_data_ready(other);
1365 sock_put(other);
1366 return 0;
1367
1368out_unlock:
1369 if (other)
1370 unix_state_unlock(other);
1371
1372out:
1373 kfree_skb(skb);
1374 if (newsk)
1375 unix_release_sock(newsk, 0);
1376 if (other)
1377 sock_put(other);
1378 return err;
1379}
1380
1381static int unix_socketpair(struct socket *socka, struct socket *sockb)
1382{
1383 struct sock *ska = socka->sk, *skb = sockb->sk;
1384
1385 /* Join our sockets back to back */
1386 sock_hold(ska);
1387 sock_hold(skb);
1388 unix_peer(ska) = skb;
1389 unix_peer(skb) = ska;
1390 init_peercred(ska);
1391 init_peercred(skb);
1392
1393 if (ska->sk_type != SOCK_DGRAM) {
1394 ska->sk_state = TCP_ESTABLISHED;
1395 skb->sk_state = TCP_ESTABLISHED;
1396 socka->state = SS_CONNECTED;
1397 sockb->state = SS_CONNECTED;
1398 }
1399 return 0;
1400}
1401
1402static void unix_sock_inherit_flags(const struct socket *old,
1403 struct socket *new)
1404{
1405 if (test_bit(SOCK_PASSCRED, &old->flags))
1406 set_bit(SOCK_PASSCRED, &new->flags);
1407 if (test_bit(SOCK_PASSSEC, &old->flags))
1408 set_bit(SOCK_PASSSEC, &new->flags);
1409}
1410
1411static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1412{
1413 struct sock *sk = sock->sk;
1414 struct sock *tsk;
1415 struct sk_buff *skb;
1416 int err;
1417
1418 err = -EOPNOTSUPP;
1419 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1420 goto out;
1421
1422 err = -EINVAL;
1423 if (sk->sk_state != TCP_LISTEN)
1424 goto out;
1425
1426 /* If socket state is TCP_LISTEN it cannot change (for now...),
1427 * so that no locks are necessary.
1428 */
1429
1430 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1431 if (!skb) {
1432 /* This means receive shutdown. */
1433 if (err == 0)
1434 err = -EINVAL;
1435 goto out;
1436 }
1437
1438 tsk = skb->sk;
1439 skb_free_datagram(sk, skb);
1440 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1441
1442 /* attach accepted sock to socket */
1443 unix_state_lock(tsk);
1444 newsock->state = SS_CONNECTED;
1445 unix_sock_inherit_flags(sock, newsock);
1446 sock_graft(tsk, newsock);
1447 unix_state_unlock(tsk);
1448 return 0;
1449
1450out:
1451 return err;
1452}
1453
1454
1455static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1456{
1457 struct sock *sk = sock->sk;
1458 struct unix_sock *u;
1459 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1460 int err = 0;
1461
1462 if (peer) {
1463 sk = unix_peer_get(sk);
1464
1465 err = -ENOTCONN;
1466 if (!sk)
1467 goto out;
1468 err = 0;
1469 } else {
1470 sock_hold(sk);
1471 }
1472
1473 u = unix_sk(sk);
1474 unix_state_lock(sk);
1475 if (!u->addr) {
1476 sunaddr->sun_family = AF_UNIX;
1477 sunaddr->sun_path[0] = 0;
1478 *uaddr_len = sizeof(short);
1479 } else {
1480 struct unix_address *addr = u->addr;
1481
1482 *uaddr_len = addr->len;
1483 memcpy(sunaddr, addr->name, *uaddr_len);
1484 }
1485 unix_state_unlock(sk);
1486 sock_put(sk);
1487out:
1488 return err;
1489}
1490
1491static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1492{
1493 int i;
1494
1495 scm->fp = UNIXCB(skb).fp;
1496 UNIXCB(skb).fp = NULL;
1497
1498 for (i = scm->fp->count-1; i >= 0; i--)
1499 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1500}
1501
1502static void unix_destruct_scm(struct sk_buff *skb)
1503{
1504 struct scm_cookie scm;
1505 memset(&scm, 0, sizeof(scm));
1506 scm.pid = UNIXCB(skb).pid;
1507 if (UNIXCB(skb).fp)
1508 unix_detach_fds(&scm, skb);
1509
1510 /* Alas, it calls VFS */
1511 /* So fscking what? fput() had been SMP-safe since the last Summer */
1512 scm_destroy(&scm);
1513 sock_wfree(skb);
1514}
1515
1516/*
1517 * The "user->unix_inflight" variable is protected by the garbage
1518 * collection lock, and we just read it locklessly here. If you go
1519 * over the limit, there might be a tiny race in actually noticing
1520 * it across threads. Tough.
1521 */
1522static inline bool too_many_unix_fds(struct task_struct *p)
1523{
1524 struct user_struct *user = current_user();
1525
1526 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1527 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1528 return false;
1529}
1530
1531#define MAX_RECURSION_LEVEL 4
1532
1533static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1534{
1535 int i;
1536 unsigned char max_level = 0;
1537
1538 if (too_many_unix_fds(current))
1539 return -ETOOMANYREFS;
1540
1541 for (i = scm->fp->count - 1; i >= 0; i--) {
1542 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1543
1544 if (sk)
1545 max_level = max(max_level,
1546 unix_sk(sk)->recursion_level);
1547 }
1548 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1549 return -ETOOMANYREFS;
1550
1551 /*
1552 * Need to duplicate file references for the sake of garbage
1553 * collection. Otherwise a socket in the fps might become a
1554 * candidate for GC while the skb is not yet queued.
1555 */
1556 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1557 if (!UNIXCB(skb).fp)
1558 return -ENOMEM;
1559
1560 for (i = scm->fp->count - 1; i >= 0; i--)
1561 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1562 return max_level;
1563}
1564
1565static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1566{
1567 int err = 0;
1568
1569 UNIXCB(skb).pid = get_pid(scm->pid);
1570 UNIXCB(skb).uid = scm->creds.uid;
1571 UNIXCB(skb).gid = scm->creds.gid;
1572 UNIXCB(skb).fp = NULL;
1573 unix_get_secdata(scm, skb);
1574 if (scm->fp && send_fds)
1575 err = unix_attach_fds(scm, skb);
1576
1577 skb->destructor = unix_destruct_scm;
1578 return err;
1579}
1580
1581static bool unix_passcred_enabled(const struct socket *sock,
1582 const struct sock *other)
1583{
1584 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1585 !other->sk_socket ||
1586 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1587}
1588
1589/*
1590 * Some apps rely on write() giving SCM_CREDENTIALS
1591 * We include credentials if source or destination socket
1592 * asserted SOCK_PASSCRED.
1593 */
1594static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1595 const struct sock *other)
1596{
1597 if (UNIXCB(skb).pid)
1598 return;
1599 if (unix_passcred_enabled(sock, other)) {
1600 UNIXCB(skb).pid = get_pid(task_tgid(current));
1601 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1602 }
1603}
1604
1605static int maybe_init_creds(struct scm_cookie *scm,
1606 struct socket *socket,
1607 const struct sock *other)
1608{
1609 int err;
1610 struct msghdr msg = { .msg_controllen = 0 };
1611
1612 err = scm_send(socket, &msg, scm, false);
1613 if (err)
1614 return err;
1615
1616 if (unix_passcred_enabled(socket, other)) {
1617 scm->pid = get_pid(task_tgid(current));
1618 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1619 }
1620 return err;
1621}
1622
1623static bool unix_skb_scm_eq(struct sk_buff *skb,
1624 struct scm_cookie *scm)
1625{
1626 const struct unix_skb_parms *u = &UNIXCB(skb);
1627
1628 return u->pid == scm->pid &&
1629 uid_eq(u->uid, scm->creds.uid) &&
1630 gid_eq(u->gid, scm->creds.gid) &&
1631 unix_secdata_eq(scm, skb);
1632}
1633
1634/*
1635 * Send AF_UNIX data.
1636 */
1637
1638static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1639 size_t len)
1640{
1641 struct sock *sk = sock->sk;
1642 struct net *net = sock_net(sk);
1643 struct unix_sock *u = unix_sk(sk);
1644 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1645 struct sock *other = NULL;
1646 int namelen = 0; /* fake GCC */
1647 int err;
1648 unsigned int hash;
1649 struct sk_buff *skb;
1650 long timeo;
1651 struct scm_cookie scm;
1652 int max_level;
1653 int data_len = 0;
1654 int sk_locked;
1655
1656 wait_for_unix_gc();
1657 err = scm_send(sock, msg, &scm, false);
1658 if (err < 0)
1659 return err;
1660
1661 err = -EOPNOTSUPP;
1662 if (msg->msg_flags&MSG_OOB)
1663 goto out;
1664
1665 if (msg->msg_namelen) {
1666 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1667 if (err < 0)
1668 goto out;
1669 namelen = err;
1670 } else {
1671 sunaddr = NULL;
1672 err = -ENOTCONN;
1673 other = unix_peer_get(sk);
1674 if (!other)
1675 goto out;
1676 }
1677
1678 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1679 && (err = unix_autobind(sock)) != 0)
1680 goto out;
1681
1682 err = -EMSGSIZE;
1683 if (len > sk->sk_sndbuf - 32)
1684 goto out;
1685
1686 if (len > SKB_MAX_ALLOC) {
1687 data_len = min_t(size_t,
1688 len - SKB_MAX_ALLOC,
1689 MAX_SKB_FRAGS * PAGE_SIZE);
1690 data_len = PAGE_ALIGN(data_len);
1691
1692 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1693 }
1694
1695 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1696 msg->msg_flags & MSG_DONTWAIT, &err,
1697 PAGE_ALLOC_COSTLY_ORDER);
1698 if (skb == NULL)
1699 goto out;
1700
1701 err = unix_scm_to_skb(&scm, skb, true);
1702 if (err < 0)
1703 goto out_free;
1704 max_level = err + 1;
1705
1706 skb_put(skb, len - data_len);
1707 skb->data_len = data_len;
1708 skb->len = len;
1709 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1710 if (err)
1711 goto out_free;
1712
1713 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1714
1715restart:
1716 if (!other) {
1717 err = -ECONNRESET;
1718 if (sunaddr == NULL)
1719 goto out_free;
1720
1721 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1722 hash, &err);
1723 if (other == NULL)
1724 goto out_free;
1725 }
1726
1727 if (sk_filter(other, skb) < 0) {
1728 /* Toss the packet but do not return any error to the sender */
1729 err = len;
1730 goto out_free;
1731 }
1732
1733 sk_locked = 0;
1734 unix_state_lock(other);
1735restart_locked:
1736 err = -EPERM;
1737 if (!unix_may_send(sk, other))
1738 goto out_unlock;
1739
1740 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1741 /*
1742 * Check with 1003.1g - what should
1743 * datagram error
1744 */
1745 unix_state_unlock(other);
1746 sock_put(other);
1747
1748 if (!sk_locked)
1749 unix_state_lock(sk);
1750
1751 err = 0;
1752 if (unix_peer(sk) == other) {
1753 unix_peer(sk) = NULL;
1754 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1755
1756 unix_state_unlock(sk);
1757
1758 unix_dgram_disconnected(sk, other);
1759 sock_put(other);
1760 err = -ECONNREFUSED;
1761 } else {
1762 unix_state_unlock(sk);
1763 }
1764
1765 other = NULL;
1766 if (err)
1767 goto out_free;
1768 goto restart;
1769 }
1770
1771 err = -EPIPE;
1772 if (other->sk_shutdown & RCV_SHUTDOWN)
1773 goto out_unlock;
1774
1775 if (sk->sk_type != SOCK_SEQPACKET) {
1776 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1777 if (err)
1778 goto out_unlock;
1779 }
1780
1781 /* other == sk && unix_peer(other) != sk if
1782 * - unix_peer(sk) == NULL, destination address bound to sk
1783 * - unix_peer(sk) == sk by time of get but disconnected before lock
1784 */
1785 if (other != sk &&
1786 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1787 if (timeo) {
1788 timeo = unix_wait_for_peer(other, timeo);
1789
1790 err = sock_intr_errno(timeo);
1791 if (signal_pending(current))
1792 goto out_free;
1793
1794 goto restart;
1795 }
1796
1797 if (!sk_locked) {
1798 unix_state_unlock(other);
1799 unix_state_double_lock(sk, other);
1800 }
1801
1802 if (unix_peer(sk) != other ||
1803 unix_dgram_peer_wake_me(sk, other)) {
1804 err = -EAGAIN;
1805 sk_locked = 1;
1806 goto out_unlock;
1807 }
1808
1809 if (!sk_locked) {
1810 sk_locked = 1;
1811 goto restart_locked;
1812 }
1813 }
1814
1815 if (unlikely(sk_locked))
1816 unix_state_unlock(sk);
1817
1818 if (sock_flag(other, SOCK_RCVTSTAMP))
1819 __net_timestamp(skb);
1820 maybe_add_creds(skb, sock, other);
1821 skb_queue_tail(&other->sk_receive_queue, skb);
1822 if (max_level > unix_sk(other)->recursion_level)
1823 unix_sk(other)->recursion_level = max_level;
1824 unix_state_unlock(other);
1825 other->sk_data_ready(other);
1826 sock_put(other);
1827 scm_destroy(&scm);
1828 return len;
1829
1830out_unlock:
1831 if (sk_locked)
1832 unix_state_unlock(sk);
1833 unix_state_unlock(other);
1834out_free:
1835 kfree_skb(skb);
1836out:
1837 if (other)
1838 sock_put(other);
1839 scm_destroy(&scm);
1840 return err;
1841}
1842
1843/* We use paged skbs for stream sockets, and limit occupancy to 32768
1844 * bytes, and a minimun of a full page.
1845 */
1846#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1847
1848static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1849 size_t len)
1850{
1851 struct sock *sk = sock->sk;
1852 struct sock *other = NULL;
1853 int err, size;
1854 struct sk_buff *skb;
1855 int sent = 0;
1856 struct scm_cookie scm;
1857 bool fds_sent = false;
1858 int max_level;
1859 int data_len;
1860
1861 wait_for_unix_gc();
1862 err = scm_send(sock, msg, &scm, false);
1863 if (err < 0)
1864 return err;
1865
1866 err = -EOPNOTSUPP;
1867 if (msg->msg_flags&MSG_OOB)
1868 goto out_err;
1869
1870 if (msg->msg_namelen) {
1871 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1872 goto out_err;
1873 } else {
1874 err = -ENOTCONN;
1875 other = unix_peer(sk);
1876 if (!other)
1877 goto out_err;
1878 }
1879
1880 if (sk->sk_shutdown & SEND_SHUTDOWN)
1881 goto pipe_err;
1882
1883 while (sent < len) {
1884 size = len - sent;
1885
1886 /* Keep two messages in the pipe so it schedules better */
1887 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1888
1889 /* allow fallback to order-0 allocations */
1890 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1891
1892 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1893
1894 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1895
1896 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1897 msg->msg_flags & MSG_DONTWAIT, &err,
1898 get_order(UNIX_SKB_FRAGS_SZ));
1899 if (!skb)
1900 goto out_err;
1901
1902 /* Only send the fds in the first buffer */
1903 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1904 if (err < 0) {
1905 kfree_skb(skb);
1906 goto out_err;
1907 }
1908 max_level = err + 1;
1909 fds_sent = true;
1910
1911 skb_put(skb, size - data_len);
1912 skb->data_len = data_len;
1913 skb->len = size;
1914 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1915 if (err) {
1916 kfree_skb(skb);
1917 goto out_err;
1918 }
1919
1920 unix_state_lock(other);
1921
1922 if (sock_flag(other, SOCK_DEAD) ||
1923 (other->sk_shutdown & RCV_SHUTDOWN))
1924 goto pipe_err_free;
1925
1926 maybe_add_creds(skb, sock, other);
1927 skb_queue_tail(&other->sk_receive_queue, skb);
1928 if (max_level > unix_sk(other)->recursion_level)
1929 unix_sk(other)->recursion_level = max_level;
1930 unix_state_unlock(other);
1931 other->sk_data_ready(other);
1932 sent += size;
1933 }
1934
1935 scm_destroy(&scm);
1936
1937 return sent;
1938
1939pipe_err_free:
1940 unix_state_unlock(other);
1941 kfree_skb(skb);
1942pipe_err:
1943 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1944 send_sig(SIGPIPE, current, 0);
1945 err = -EPIPE;
1946out_err:
1947 scm_destroy(&scm);
1948 return sent ? : err;
1949}
1950
1951static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1952 int offset, size_t size, int flags)
1953{
1954 int err;
1955 bool send_sigpipe = false;
1956 bool init_scm = true;
1957 struct scm_cookie scm;
1958 struct sock *other, *sk = socket->sk;
1959 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1960
1961 if (flags & MSG_OOB)
1962 return -EOPNOTSUPP;
1963
1964 other = unix_peer(sk);
1965 if (!other || sk->sk_state != TCP_ESTABLISHED)
1966 return -ENOTCONN;
1967
1968 if (false) {
1969alloc_skb:
1970 unix_state_unlock(other);
1971 mutex_unlock(&unix_sk(other)->readlock);
1972 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1973 &err, 0);
1974 if (!newskb)
1975 goto err;
1976 }
1977
1978 /* we must acquire readlock as we modify already present
1979 * skbs in the sk_receive_queue and mess with skb->len
1980 */
1981 err = mutex_lock_interruptible(&unix_sk(other)->readlock);
1982 if (err) {
1983 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1984 goto err;
1985 }
1986
1987 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1988 err = -EPIPE;
1989 send_sigpipe = true;
1990 goto err_unlock;
1991 }
1992
1993 unix_state_lock(other);
1994
1995 if (sock_flag(other, SOCK_DEAD) ||
1996 other->sk_shutdown & RCV_SHUTDOWN) {
1997 err = -EPIPE;
1998 send_sigpipe = true;
1999 goto err_state_unlock;
2000 }
2001
2002 if (init_scm) {
2003 err = maybe_init_creds(&scm, socket, other);
2004 if (err)
2005 goto err_state_unlock;
2006 init_scm = false;
2007 }
2008
2009 skb = skb_peek_tail(&other->sk_receive_queue);
2010 if (tail && tail == skb) {
2011 skb = newskb;
2012 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2013 if (newskb) {
2014 skb = newskb;
2015 } else {
2016 tail = skb;
2017 goto alloc_skb;
2018 }
2019 } else if (newskb) {
2020 /* this is fast path, we don't necessarily need to
2021 * call to kfree_skb even though with newskb == NULL
2022 * this - does no harm
2023 */
2024 consume_skb(newskb);
2025 newskb = NULL;
2026 }
2027
2028 if (skb_append_pagefrags(skb, page, offset, size)) {
2029 tail = skb;
2030 goto alloc_skb;
2031 }
2032
2033 skb->len += size;
2034 skb->data_len += size;
2035 skb->truesize += size;
2036 atomic_add(size, &sk->sk_wmem_alloc);
2037
2038 if (newskb) {
2039 err = unix_scm_to_skb(&scm, skb, false);
2040 if (err)
2041 goto err_state_unlock;
2042 spin_lock(&other->sk_receive_queue.lock);
2043 __skb_queue_tail(&other->sk_receive_queue, newskb);
2044 spin_unlock(&other->sk_receive_queue.lock);
2045 }
2046
2047 unix_state_unlock(other);
2048 mutex_unlock(&unix_sk(other)->readlock);
2049
2050 other->sk_data_ready(other);
2051 scm_destroy(&scm);
2052 return size;
2053
2054err_state_unlock:
2055 unix_state_unlock(other);
2056err_unlock:
2057 mutex_unlock(&unix_sk(other)->readlock);
2058err:
2059 kfree_skb(newskb);
2060 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2061 send_sig(SIGPIPE, current, 0);
2062 if (!init_scm)
2063 scm_destroy(&scm);
2064 return err;
2065}
2066
2067static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2068 size_t len)
2069{
2070 int err;
2071 struct sock *sk = sock->sk;
2072
2073 err = sock_error(sk);
2074 if (err)
2075 return err;
2076
2077 if (sk->sk_state != TCP_ESTABLISHED)
2078 return -ENOTCONN;
2079
2080 if (msg->msg_namelen)
2081 msg->msg_namelen = 0;
2082
2083 return unix_dgram_sendmsg(sock, msg, len);
2084}
2085
2086static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2087 size_t size, int flags)
2088{
2089 struct sock *sk = sock->sk;
2090
2091 if (sk->sk_state != TCP_ESTABLISHED)
2092 return -ENOTCONN;
2093
2094 return unix_dgram_recvmsg(sock, msg, size, flags);
2095}
2096
2097static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2098{
2099 struct unix_sock *u = unix_sk(sk);
2100
2101 if (u->addr) {
2102 msg->msg_namelen = u->addr->len;
2103 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2104 }
2105}
2106
2107static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2108 size_t size, int flags)
2109{
2110 struct scm_cookie scm;
2111 struct sock *sk = sock->sk;
2112 struct unix_sock *u = unix_sk(sk);
2113 struct sk_buff *skb, *last;
2114 long timeo;
2115 int err;
2116 int peeked, skip;
2117
2118 err = -EOPNOTSUPP;
2119 if (flags&MSG_OOB)
2120 goto out;
2121
2122 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2123
2124 do {
2125 mutex_lock(&u->readlock);
2126
2127 skip = sk_peek_offset(sk, flags);
2128 skb = __skb_try_recv_datagram(sk, flags, &peeked, &skip, &err,
2129 &last);
2130 if (skb)
2131 break;
2132
2133 mutex_unlock(&u->readlock);
2134
2135 if (err != -EAGAIN)
2136 break;
2137 } while (timeo &&
2138 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2139
2140 if (!skb) { /* implies readlock unlocked */
2141 unix_state_lock(sk);
2142 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2143 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2144 (sk->sk_shutdown & RCV_SHUTDOWN))
2145 err = 0;
2146 unix_state_unlock(sk);
2147 goto out;
2148 }
2149
2150 if (wq_has_sleeper(&u->peer_wait))
2151 wake_up_interruptible_sync_poll(&u->peer_wait,
2152 POLLOUT | POLLWRNORM |
2153 POLLWRBAND);
2154
2155 if (msg->msg_name)
2156 unix_copy_addr(msg, skb->sk);
2157
2158 if (size > skb->len - skip)
2159 size = skb->len - skip;
2160 else if (size < skb->len - skip)
2161 msg->msg_flags |= MSG_TRUNC;
2162
2163 err = skb_copy_datagram_msg(skb, skip, msg, size);
2164 if (err)
2165 goto out_free;
2166
2167 if (sock_flag(sk, SOCK_RCVTSTAMP))
2168 __sock_recv_timestamp(msg, sk, skb);
2169
2170 memset(&scm, 0, sizeof(scm));
2171
2172 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2173 unix_set_secdata(&scm, skb);
2174
2175 if (!(flags & MSG_PEEK)) {
2176 if (UNIXCB(skb).fp)
2177 unix_detach_fds(&scm, skb);
2178
2179 sk_peek_offset_bwd(sk, skb->len);
2180 } else {
2181 /* It is questionable: on PEEK we could:
2182 - do not return fds - good, but too simple 8)
2183 - return fds, and do not return them on read (old strategy,
2184 apparently wrong)
2185 - clone fds (I chose it for now, it is the most universal
2186 solution)
2187
2188 POSIX 1003.1g does not actually define this clearly
2189 at all. POSIX 1003.1g doesn't define a lot of things
2190 clearly however!
2191
2192 */
2193
2194 sk_peek_offset_fwd(sk, size);
2195
2196 if (UNIXCB(skb).fp)
2197 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2198 }
2199 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2200
2201 scm_recv(sock, msg, &scm, flags);
2202
2203out_free:
2204 skb_free_datagram(sk, skb);
2205 mutex_unlock(&u->readlock);
2206out:
2207 return err;
2208}
2209
2210/*
2211 * Sleep until more data has arrived. But check for races..
2212 */
2213static long unix_stream_data_wait(struct sock *sk, long timeo,
2214 struct sk_buff *last, unsigned int last_len)
2215{
2216 struct sk_buff *tail;
2217 DEFINE_WAIT(wait);
2218
2219 unix_state_lock(sk);
2220
2221 for (;;) {
2222 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2223
2224 tail = skb_peek_tail(&sk->sk_receive_queue);
2225 if (tail != last ||
2226 (tail && tail->len != last_len) ||
2227 sk->sk_err ||
2228 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2229 signal_pending(current) ||
2230 !timeo)
2231 break;
2232
2233 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2234 unix_state_unlock(sk);
2235 timeo = freezable_schedule_timeout(timeo);
2236 unix_state_lock(sk);
2237
2238 if (sock_flag(sk, SOCK_DEAD))
2239 break;
2240
2241 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2242 }
2243
2244 finish_wait(sk_sleep(sk), &wait);
2245 unix_state_unlock(sk);
2246 return timeo;
2247}
2248
2249static unsigned int unix_skb_len(const struct sk_buff *skb)
2250{
2251 return skb->len - UNIXCB(skb).consumed;
2252}
2253
2254struct unix_stream_read_state {
2255 int (*recv_actor)(struct sk_buff *, int, int,
2256 struct unix_stream_read_state *);
2257 struct socket *socket;
2258 struct msghdr *msg;
2259 struct pipe_inode_info *pipe;
2260 size_t size;
2261 int flags;
2262 unsigned int splice_flags;
2263};
2264
2265static int unix_stream_read_generic(struct unix_stream_read_state *state)
2266{
2267 struct scm_cookie scm;
2268 struct socket *sock = state->socket;
2269 struct sock *sk = sock->sk;
2270 struct unix_sock *u = unix_sk(sk);
2271 int copied = 0;
2272 int flags = state->flags;
2273 int noblock = flags & MSG_DONTWAIT;
2274 bool check_creds = false;
2275 int target;
2276 int err = 0;
2277 long timeo;
2278 int skip;
2279 size_t size = state->size;
2280 unsigned int last_len;
2281
2282 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2283 err = -EINVAL;
2284 goto out;
2285 }
2286
2287 if (unlikely(flags & MSG_OOB)) {
2288 err = -EOPNOTSUPP;
2289 goto out;
2290 }
2291
2292 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2293 timeo = sock_rcvtimeo(sk, noblock);
2294
2295 memset(&scm, 0, sizeof(scm));
2296
2297 /* Lock the socket to prevent queue disordering
2298 * while sleeps in memcpy_tomsg
2299 */
2300 mutex_lock(&u->readlock);
2301
2302 if (flags & MSG_PEEK)
2303 skip = sk_peek_offset(sk, flags);
2304 else
2305 skip = 0;
2306
2307 do {
2308 int chunk;
2309 bool drop_skb;
2310 struct sk_buff *skb, *last;
2311
2312redo:
2313 unix_state_lock(sk);
2314 if (sock_flag(sk, SOCK_DEAD)) {
2315 err = -ECONNRESET;
2316 goto unlock;
2317 }
2318 last = skb = skb_peek(&sk->sk_receive_queue);
2319 last_len = last ? last->len : 0;
2320again:
2321 if (skb == NULL) {
2322 unix_sk(sk)->recursion_level = 0;
2323 if (copied >= target)
2324 goto unlock;
2325
2326 /*
2327 * POSIX 1003.1g mandates this order.
2328 */
2329
2330 err = sock_error(sk);
2331 if (err)
2332 goto unlock;
2333 if (sk->sk_shutdown & RCV_SHUTDOWN)
2334 goto unlock;
2335
2336 unix_state_unlock(sk);
2337 if (!timeo) {
2338 err = -EAGAIN;
2339 break;
2340 }
2341
2342 mutex_unlock(&u->readlock);
2343
2344 timeo = unix_stream_data_wait(sk, timeo, last,
2345 last_len);
2346
2347 if (signal_pending(current)) {
2348 err = sock_intr_errno(timeo);
2349 scm_destroy(&scm);
2350 goto out;
2351 }
2352
2353 mutex_lock(&u->readlock);
2354 goto redo;
2355unlock:
2356 unix_state_unlock(sk);
2357 break;
2358 }
2359
2360 while (skip >= unix_skb_len(skb)) {
2361 skip -= unix_skb_len(skb);
2362 last = skb;
2363 last_len = skb->len;
2364 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2365 if (!skb)
2366 goto again;
2367 }
2368
2369 unix_state_unlock(sk);
2370
2371 if (check_creds) {
2372 /* Never glue messages from different writers */
2373 if (!unix_skb_scm_eq(skb, &scm))
2374 break;
2375 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2376 /* Copy credentials */
2377 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2378 unix_set_secdata(&scm, skb);
2379 check_creds = true;
2380 }
2381
2382 /* Copy address just once */
2383 if (state->msg && state->msg->msg_name) {
2384 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2385 state->msg->msg_name);
2386 unix_copy_addr(state->msg, skb->sk);
2387 sunaddr = NULL;
2388 }
2389
2390 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2391 skb_get(skb);
2392 chunk = state->recv_actor(skb, skip, chunk, state);
2393 drop_skb = !unix_skb_len(skb);
2394 /* skb is only safe to use if !drop_skb */
2395 consume_skb(skb);
2396 if (chunk < 0) {
2397 if (copied == 0)
2398 copied = -EFAULT;
2399 break;
2400 }
2401 copied += chunk;
2402 size -= chunk;
2403
2404 if (drop_skb) {
2405 /* the skb was touched by a concurrent reader;
2406 * we should not expect anything from this skb
2407 * anymore and assume it invalid - we can be
2408 * sure it was dropped from the socket queue
2409 *
2410 * let's report a short read
2411 */
2412 err = 0;
2413 break;
2414 }
2415
2416 /* Mark read part of skb as used */
2417 if (!(flags & MSG_PEEK)) {
2418 UNIXCB(skb).consumed += chunk;
2419
2420 sk_peek_offset_bwd(sk, chunk);
2421
2422 if (UNIXCB(skb).fp)
2423 unix_detach_fds(&scm, skb);
2424
2425 if (unix_skb_len(skb))
2426 break;
2427
2428 skb_unlink(skb, &sk->sk_receive_queue);
2429 consume_skb(skb);
2430
2431 if (scm.fp)
2432 break;
2433 } else {
2434 /* It is questionable, see note in unix_dgram_recvmsg.
2435 */
2436 if (UNIXCB(skb).fp)
2437 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2438
2439 sk_peek_offset_fwd(sk, chunk);
2440
2441 if (UNIXCB(skb).fp)
2442 break;
2443
2444 skip = 0;
2445 last = skb;
2446 last_len = skb->len;
2447 unix_state_lock(sk);
2448 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2449 if (skb)
2450 goto again;
2451 unix_state_unlock(sk);
2452 break;
2453 }
2454 } while (size);
2455
2456 mutex_unlock(&u->readlock);
2457 if (state->msg)
2458 scm_recv(sock, state->msg, &scm, flags);
2459 else
2460 scm_destroy(&scm);
2461out:
2462 return copied ? : err;
2463}
2464
2465static int unix_stream_read_actor(struct sk_buff *skb,
2466 int skip, int chunk,
2467 struct unix_stream_read_state *state)
2468{
2469 int ret;
2470
2471 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2472 state->msg, chunk);
2473 return ret ?: chunk;
2474}
2475
2476static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2477 size_t size, int flags)
2478{
2479 struct unix_stream_read_state state = {
2480 .recv_actor = unix_stream_read_actor,
2481 .socket = sock,
2482 .msg = msg,
2483 .size = size,
2484 .flags = flags
2485 };
2486
2487 return unix_stream_read_generic(&state);
2488}
2489
2490static ssize_t skb_unix_socket_splice(struct sock *sk,
2491 struct pipe_inode_info *pipe,
2492 struct splice_pipe_desc *spd)
2493{
2494 int ret;
2495 struct unix_sock *u = unix_sk(sk);
2496
2497 mutex_unlock(&u->readlock);
2498 ret = splice_to_pipe(pipe, spd);
2499 mutex_lock(&u->readlock);
2500
2501 return ret;
2502}
2503
2504static int unix_stream_splice_actor(struct sk_buff *skb,
2505 int skip, int chunk,
2506 struct unix_stream_read_state *state)
2507{
2508 return skb_splice_bits(skb, state->socket->sk,
2509 UNIXCB(skb).consumed + skip,
2510 state->pipe, chunk, state->splice_flags,
2511 skb_unix_socket_splice);
2512}
2513
2514static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2515 struct pipe_inode_info *pipe,
2516 size_t size, unsigned int flags)
2517{
2518 struct unix_stream_read_state state = {
2519 .recv_actor = unix_stream_splice_actor,
2520 .socket = sock,
2521 .pipe = pipe,
2522 .size = size,
2523 .splice_flags = flags,
2524 };
2525
2526 if (unlikely(*ppos))
2527 return -ESPIPE;
2528
2529 if (sock->file->f_flags & O_NONBLOCK ||
2530 flags & SPLICE_F_NONBLOCK)
2531 state.flags = MSG_DONTWAIT;
2532
2533 return unix_stream_read_generic(&state);
2534}
2535
2536static int unix_shutdown(struct socket *sock, int mode)
2537{
2538 struct sock *sk = sock->sk;
2539 struct sock *other;
2540
2541 if (mode < SHUT_RD || mode > SHUT_RDWR)
2542 return -EINVAL;
2543 /* This maps:
2544 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2545 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2546 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2547 */
2548 ++mode;
2549
2550 unix_state_lock(sk);
2551 sk->sk_shutdown |= mode;
2552 other = unix_peer(sk);
2553 if (other)
2554 sock_hold(other);
2555 unix_state_unlock(sk);
2556 sk->sk_state_change(sk);
2557
2558 if (other &&
2559 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2560
2561 int peer_mode = 0;
2562
2563 if (mode&RCV_SHUTDOWN)
2564 peer_mode |= SEND_SHUTDOWN;
2565 if (mode&SEND_SHUTDOWN)
2566 peer_mode |= RCV_SHUTDOWN;
2567 unix_state_lock(other);
2568 other->sk_shutdown |= peer_mode;
2569 unix_state_unlock(other);
2570 other->sk_state_change(other);
2571 if (peer_mode == SHUTDOWN_MASK)
2572 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2573 else if (peer_mode & RCV_SHUTDOWN)
2574 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2575 }
2576 if (other)
2577 sock_put(other);
2578
2579 return 0;
2580}
2581
2582long unix_inq_len(struct sock *sk)
2583{
2584 struct sk_buff *skb;
2585 long amount = 0;
2586
2587 if (sk->sk_state == TCP_LISTEN)
2588 return -EINVAL;
2589
2590 spin_lock(&sk->sk_receive_queue.lock);
2591 if (sk->sk_type == SOCK_STREAM ||
2592 sk->sk_type == SOCK_SEQPACKET) {
2593 skb_queue_walk(&sk->sk_receive_queue, skb)
2594 amount += unix_skb_len(skb);
2595 } else {
2596 skb = skb_peek(&sk->sk_receive_queue);
2597 if (skb)
2598 amount = skb->len;
2599 }
2600 spin_unlock(&sk->sk_receive_queue.lock);
2601
2602 return amount;
2603}
2604EXPORT_SYMBOL_GPL(unix_inq_len);
2605
2606long unix_outq_len(struct sock *sk)
2607{
2608 return sk_wmem_alloc_get(sk);
2609}
2610EXPORT_SYMBOL_GPL(unix_outq_len);
2611
2612static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2613{
2614 struct sock *sk = sock->sk;
2615 long amount = 0;
2616 int err;
2617
2618 switch (cmd) {
2619 case SIOCOUTQ:
2620 amount = unix_outq_len(sk);
2621 err = put_user(amount, (int __user *)arg);
2622 break;
2623 case SIOCINQ:
2624 amount = unix_inq_len(sk);
2625 if (amount < 0)
2626 err = amount;
2627 else
2628 err = put_user(amount, (int __user *)arg);
2629 break;
2630 default:
2631 err = -ENOIOCTLCMD;
2632 break;
2633 }
2634 return err;
2635}
2636
2637static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2638{
2639 struct sock *sk = sock->sk;
2640 unsigned int mask;
2641
2642 sock_poll_wait(file, sk_sleep(sk), wait);
2643 mask = 0;
2644
2645 /* exceptional events? */
2646 if (sk->sk_err)
2647 mask |= POLLERR;
2648 if (sk->sk_shutdown == SHUTDOWN_MASK)
2649 mask |= POLLHUP;
2650 if (sk->sk_shutdown & RCV_SHUTDOWN)
2651 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2652
2653 /* readable? */
2654 if (!skb_queue_empty(&sk->sk_receive_queue))
2655 mask |= POLLIN | POLLRDNORM;
2656
2657 /* Connection-based need to check for termination and startup */
2658 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2659 sk->sk_state == TCP_CLOSE)
2660 mask |= POLLHUP;
2661
2662 /*
2663 * we set writable also when the other side has shut down the
2664 * connection. This prevents stuck sockets.
2665 */
2666 if (unix_writable(sk))
2667 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2668
2669 return mask;
2670}
2671
2672static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2673 poll_table *wait)
2674{
2675 struct sock *sk = sock->sk, *other;
2676 unsigned int mask, writable;
2677
2678 sock_poll_wait(file, sk_sleep(sk), wait);
2679 mask = 0;
2680
2681 /* exceptional events? */
2682 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2683 mask |= POLLERR |
2684 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2685
2686 if (sk->sk_shutdown & RCV_SHUTDOWN)
2687 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2688 if (sk->sk_shutdown == SHUTDOWN_MASK)
2689 mask |= POLLHUP;
2690
2691 /* readable? */
2692 if (!skb_queue_empty(&sk->sk_receive_queue))
2693 mask |= POLLIN | POLLRDNORM;
2694
2695 /* Connection-based need to check for termination and startup */
2696 if (sk->sk_type == SOCK_SEQPACKET) {
2697 if (sk->sk_state == TCP_CLOSE)
2698 mask |= POLLHUP;
2699 /* connection hasn't started yet? */
2700 if (sk->sk_state == TCP_SYN_SENT)
2701 return mask;
2702 }
2703
2704 /* No write status requested, avoid expensive OUT tests. */
2705 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2706 return mask;
2707
2708 writable = unix_writable(sk);
2709 if (writable) {
2710 unix_state_lock(sk);
2711
2712 other = unix_peer(sk);
2713 if (other && unix_peer(other) != sk &&
2714 unix_recvq_full(other) &&
2715 unix_dgram_peer_wake_me(sk, other))
2716 writable = 0;
2717
2718 unix_state_unlock(sk);
2719 }
2720
2721 if (writable)
2722 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2723 else
2724 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2725
2726 return mask;
2727}
2728
2729#ifdef CONFIG_PROC_FS
2730
2731#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2732
2733#define get_bucket(x) ((x) >> BUCKET_SPACE)
2734#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2735#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2736
2737static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2738{
2739 unsigned long offset = get_offset(*pos);
2740 unsigned long bucket = get_bucket(*pos);
2741 struct sock *sk;
2742 unsigned long count = 0;
2743
2744 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2745 if (sock_net(sk) != seq_file_net(seq))
2746 continue;
2747 if (++count == offset)
2748 break;
2749 }
2750
2751 return sk;
2752}
2753
2754static struct sock *unix_next_socket(struct seq_file *seq,
2755 struct sock *sk,
2756 loff_t *pos)
2757{
2758 unsigned long bucket;
2759
2760 while (sk > (struct sock *)SEQ_START_TOKEN) {
2761 sk = sk_next(sk);
2762 if (!sk)
2763 goto next_bucket;
2764 if (sock_net(sk) == seq_file_net(seq))
2765 return sk;
2766 }
2767
2768 do {
2769 sk = unix_from_bucket(seq, pos);
2770 if (sk)
2771 return sk;
2772
2773next_bucket:
2774 bucket = get_bucket(*pos) + 1;
2775 *pos = set_bucket_offset(bucket, 1);
2776 } while (bucket < ARRAY_SIZE(unix_socket_table));
2777
2778 return NULL;
2779}
2780
2781static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2782 __acquires(unix_table_lock)
2783{
2784 spin_lock(&unix_table_lock);
2785
2786 if (!*pos)
2787 return SEQ_START_TOKEN;
2788
2789 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2790 return NULL;
2791
2792 return unix_next_socket(seq, NULL, pos);
2793}
2794
2795static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2796{
2797 ++*pos;
2798 return unix_next_socket(seq, v, pos);
2799}
2800
2801static void unix_seq_stop(struct seq_file *seq, void *v)
2802 __releases(unix_table_lock)
2803{
2804 spin_unlock(&unix_table_lock);
2805}
2806
2807static int unix_seq_show(struct seq_file *seq, void *v)
2808{
2809
2810 if (v == SEQ_START_TOKEN)
2811 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2812 "Inode Path\n");
2813 else {
2814 struct sock *s = v;
2815 struct unix_sock *u = unix_sk(s);
2816 unix_state_lock(s);
2817
2818 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2819 s,
2820 atomic_read(&s->sk_refcnt),
2821 0,
2822 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2823 s->sk_type,
2824 s->sk_socket ?
2825 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2826 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2827 sock_i_ino(s));
2828
2829 if (u->addr) {
2830 int i, len;
2831 seq_putc(seq, ' ');
2832
2833 i = 0;
2834 len = u->addr->len - sizeof(short);
2835 if (!UNIX_ABSTRACT(s))
2836 len--;
2837 else {
2838 seq_putc(seq, '@');
2839 i++;
2840 }
2841 for ( ; i < len; i++)
2842 seq_putc(seq, u->addr->name->sun_path[i]);
2843 }
2844 unix_state_unlock(s);
2845 seq_putc(seq, '\n');
2846 }
2847
2848 return 0;
2849}
2850
2851static const struct seq_operations unix_seq_ops = {
2852 .start = unix_seq_start,
2853 .next = unix_seq_next,
2854 .stop = unix_seq_stop,
2855 .show = unix_seq_show,
2856};
2857
2858static int unix_seq_open(struct inode *inode, struct file *file)
2859{
2860 return seq_open_net(inode, file, &unix_seq_ops,
2861 sizeof(struct seq_net_private));
2862}
2863
2864static const struct file_operations unix_seq_fops = {
2865 .owner = THIS_MODULE,
2866 .open = unix_seq_open,
2867 .read = seq_read,
2868 .llseek = seq_lseek,
2869 .release = seq_release_net,
2870};
2871
2872#endif
2873
2874static const struct net_proto_family unix_family_ops = {
2875 .family = PF_UNIX,
2876 .create = unix_create,
2877 .owner = THIS_MODULE,
2878};
2879
2880
2881static int __net_init unix_net_init(struct net *net)
2882{
2883 int error = -ENOMEM;
2884
2885 net->unx.sysctl_max_dgram_qlen = 10;
2886 if (unix_sysctl_register(net))
2887 goto out;
2888
2889#ifdef CONFIG_PROC_FS
2890 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2891 unix_sysctl_unregister(net);
2892 goto out;
2893 }
2894#endif
2895 error = 0;
2896out:
2897 return error;
2898}
2899
2900static void __net_exit unix_net_exit(struct net *net)
2901{
2902 unix_sysctl_unregister(net);
2903 remove_proc_entry("unix", net->proc_net);
2904}
2905
2906static struct pernet_operations unix_net_ops = {
2907 .init = unix_net_init,
2908 .exit = unix_net_exit,
2909};
2910
2911static int __init af_unix_init(void)
2912{
2913 int rc = -1;
2914
2915 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2916
2917 rc = proto_register(&unix_proto, 1);
2918 if (rc != 0) {
2919 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2920 goto out;
2921 }
2922
2923 sock_register(&unix_family_ops);
2924 register_pernet_subsys(&unix_net_ops);
2925out:
2926 return rc;
2927}
2928
2929static void __exit af_unix_exit(void)
2930{
2931 sock_unregister(PF_UNIX);
2932 proto_unregister(&unix_proto);
2933 unregister_pernet_subsys(&unix_net_ops);
2934}
2935
2936/* Earlier than device_initcall() so that other drivers invoking
2937 request_module() don't end up in a loop when modprobe tries
2938 to use a UNIX socket. But later than subsys_initcall() because
2939 we depend on stuff initialised there */
2940fs_initcall(af_unix_init);
2941module_exit(af_unix_exit);
2942
2943MODULE_LICENSE("GPL");
2944MODULE_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/freezer.h>
116#include <linux/file.h>
117#include <linux/btf_ids.h>
118
119#include "scm.h"
120
121static atomic_long_t unix_nr_socks;
122static struct hlist_head bsd_socket_buckets[UNIX_HASH_SIZE / 2];
123static spinlock_t bsd_socket_locks[UNIX_HASH_SIZE / 2];
124
125/* SMP locking strategy:
126 * hash table is protected with spinlock.
127 * each socket state is protected by separate spinlock.
128 */
129
130static unsigned int unix_unbound_hash(struct sock *sk)
131{
132 unsigned long hash = (unsigned long)sk;
133
134 hash ^= hash >> 16;
135 hash ^= hash >> 8;
136 hash ^= sk->sk_type;
137
138 return hash & UNIX_HASH_MOD;
139}
140
141static unsigned int unix_bsd_hash(struct inode *i)
142{
143 return i->i_ino & UNIX_HASH_MOD;
144}
145
146static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
147 int addr_len, int type)
148{
149 __wsum csum = csum_partial(sunaddr, addr_len, 0);
150 unsigned int hash;
151
152 hash = (__force unsigned int)csum_fold(csum);
153 hash ^= hash >> 8;
154 hash ^= type;
155
156 return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
157}
158
159static void unix_table_double_lock(struct net *net,
160 unsigned int hash1, unsigned int hash2)
161{
162 if (hash1 == hash2) {
163 spin_lock(&net->unx.table.locks[hash1]);
164 return;
165 }
166
167 if (hash1 > hash2)
168 swap(hash1, hash2);
169
170 spin_lock(&net->unx.table.locks[hash1]);
171 spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
172}
173
174static void unix_table_double_unlock(struct net *net,
175 unsigned int hash1, unsigned int hash2)
176{
177 if (hash1 == hash2) {
178 spin_unlock(&net->unx.table.locks[hash1]);
179 return;
180 }
181
182 spin_unlock(&net->unx.table.locks[hash1]);
183 spin_unlock(&net->unx.table.locks[hash2]);
184}
185
186#ifdef CONFIG_SECURITY_NETWORK
187static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
188{
189 UNIXCB(skb).secid = scm->secid;
190}
191
192static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
193{
194 scm->secid = UNIXCB(skb).secid;
195}
196
197static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
198{
199 return (scm->secid == UNIXCB(skb).secid);
200}
201#else
202static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
203{ }
204
205static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
206{ }
207
208static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
209{
210 return true;
211}
212#endif /* CONFIG_SECURITY_NETWORK */
213
214#define unix_peer(sk) (unix_sk(sk)->peer)
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 void unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
292{
293 /* This may look like an off by one error but it is a bit more
294 * subtle. 108 is the longest valid AF_UNIX path for a binding.
295 * sun_path[108] doesn't as such exist. However in kernel space
296 * we are guaranteed that it is a valid memory location in our
297 * kernel address buffer because syscall functions always pass
298 * a pointer of struct sockaddr_storage which has a bigger buffer
299 * than 108.
300 */
301 ((char *)sunaddr)[addr_len] = 0;
302}
303
304static void __unix_remove_socket(struct sock *sk)
305{
306 sk_del_node_init(sk);
307}
308
309static void __unix_insert_socket(struct net *net, struct sock *sk)
310{
311 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
312 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
313}
314
315static void __unix_set_addr_hash(struct net *net, struct sock *sk,
316 struct unix_address *addr, unsigned int hash)
317{
318 __unix_remove_socket(sk);
319 smp_store_release(&unix_sk(sk)->addr, addr);
320
321 sk->sk_hash = hash;
322 __unix_insert_socket(net, sk);
323}
324
325static void unix_remove_socket(struct net *net, struct sock *sk)
326{
327 spin_lock(&net->unx.table.locks[sk->sk_hash]);
328 __unix_remove_socket(sk);
329 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
330}
331
332static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
333{
334 spin_lock(&net->unx.table.locks[sk->sk_hash]);
335 __unix_insert_socket(net, sk);
336 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
337}
338
339static void unix_insert_bsd_socket(struct sock *sk)
340{
341 spin_lock(&bsd_socket_locks[sk->sk_hash]);
342 sk_add_bind_node(sk, &bsd_socket_buckets[sk->sk_hash]);
343 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
344}
345
346static void unix_remove_bsd_socket(struct sock *sk)
347{
348 if (!hlist_unhashed(&sk->sk_bind_node)) {
349 spin_lock(&bsd_socket_locks[sk->sk_hash]);
350 __sk_del_bind_node(sk);
351 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
352
353 sk_node_init(&sk->sk_bind_node);
354 }
355}
356
357static struct sock *__unix_find_socket_byname(struct net *net,
358 struct sockaddr_un *sunname,
359 int len, unsigned int hash)
360{
361 struct sock *s;
362
363 sk_for_each(s, &net->unx.table.buckets[hash]) {
364 struct unix_sock *u = unix_sk(s);
365
366 if (u->addr->len == len &&
367 !memcmp(u->addr->name, sunname, len))
368 return s;
369 }
370 return NULL;
371}
372
373static inline struct sock *unix_find_socket_byname(struct net *net,
374 struct sockaddr_un *sunname,
375 int len, unsigned int hash)
376{
377 struct sock *s;
378
379 spin_lock(&net->unx.table.locks[hash]);
380 s = __unix_find_socket_byname(net, sunname, len, hash);
381 if (s)
382 sock_hold(s);
383 spin_unlock(&net->unx.table.locks[hash]);
384 return s;
385}
386
387static struct sock *unix_find_socket_byinode(struct inode *i)
388{
389 unsigned int hash = unix_bsd_hash(i);
390 struct sock *s;
391
392 spin_lock(&bsd_socket_locks[hash]);
393 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
394 struct dentry *dentry = unix_sk(s)->path.dentry;
395
396 if (dentry && d_backing_inode(dentry) == i) {
397 sock_hold(s);
398 spin_unlock(&bsd_socket_locks[hash]);
399 return s;
400 }
401 }
402 spin_unlock(&bsd_socket_locks[hash]);
403 return NULL;
404}
405
406/* Support code for asymmetrically connected dgram sockets
407 *
408 * If a datagram socket is connected to a socket not itself connected
409 * to the first socket (eg, /dev/log), clients may only enqueue more
410 * messages if the present receive queue of the server socket is not
411 * "too large". This means there's a second writeability condition
412 * poll and sendmsg need to test. The dgram recv code will do a wake
413 * up on the peer_wait wait queue of a socket upon reception of a
414 * datagram which needs to be propagated to sleeping would-be writers
415 * since these might not have sent anything so far. This can't be
416 * accomplished via poll_wait because the lifetime of the server
417 * socket might be less than that of its clients if these break their
418 * association with it or if the server socket is closed while clients
419 * are still connected to it and there's no way to inform "a polling
420 * implementation" that it should let go of a certain wait queue
421 *
422 * In order to propagate a wake up, a wait_queue_entry_t of the client
423 * socket is enqueued on the peer_wait queue of the server socket
424 * whose wake function does a wake_up on the ordinary client socket
425 * wait queue. This connection is established whenever a write (or
426 * poll for write) hit the flow control condition and broken when the
427 * association to the server socket is dissolved or after a wake up
428 * was relayed.
429 */
430
431static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
432 void *key)
433{
434 struct unix_sock *u;
435 wait_queue_head_t *u_sleep;
436
437 u = container_of(q, struct unix_sock, peer_wake);
438
439 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
440 q);
441 u->peer_wake.private = NULL;
442
443 /* relaying can only happen while the wq still exists */
444 u_sleep = sk_sleep(&u->sk);
445 if (u_sleep)
446 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
447
448 return 0;
449}
450
451static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
452{
453 struct unix_sock *u, *u_other;
454 int rc;
455
456 u = unix_sk(sk);
457 u_other = unix_sk(other);
458 rc = 0;
459 spin_lock(&u_other->peer_wait.lock);
460
461 if (!u->peer_wake.private) {
462 u->peer_wake.private = other;
463 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
464
465 rc = 1;
466 }
467
468 spin_unlock(&u_other->peer_wait.lock);
469 return rc;
470}
471
472static void unix_dgram_peer_wake_disconnect(struct sock *sk,
473 struct sock *other)
474{
475 struct unix_sock *u, *u_other;
476
477 u = unix_sk(sk);
478 u_other = unix_sk(other);
479 spin_lock(&u_other->peer_wait.lock);
480
481 if (u->peer_wake.private == other) {
482 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
483 u->peer_wake.private = NULL;
484 }
485
486 spin_unlock(&u_other->peer_wait.lock);
487}
488
489static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
490 struct sock *other)
491{
492 unix_dgram_peer_wake_disconnect(sk, other);
493 wake_up_interruptible_poll(sk_sleep(sk),
494 EPOLLOUT |
495 EPOLLWRNORM |
496 EPOLLWRBAND);
497}
498
499/* preconditions:
500 * - unix_peer(sk) == other
501 * - association is stable
502 */
503static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
504{
505 int connected;
506
507 connected = unix_dgram_peer_wake_connect(sk, other);
508
509 /* If other is SOCK_DEAD, we want to make sure we signal
510 * POLLOUT, such that a subsequent write() can get a
511 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
512 * to other and its full, we will hang waiting for POLLOUT.
513 */
514 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
515 return 1;
516
517 if (connected)
518 unix_dgram_peer_wake_disconnect(sk, other);
519
520 return 0;
521}
522
523static int unix_writable(const struct sock *sk)
524{
525 return sk->sk_state != TCP_LISTEN &&
526 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
527}
528
529static void unix_write_space(struct sock *sk)
530{
531 struct socket_wq *wq;
532
533 rcu_read_lock();
534 if (unix_writable(sk)) {
535 wq = rcu_dereference(sk->sk_wq);
536 if (skwq_has_sleeper(wq))
537 wake_up_interruptible_sync_poll(&wq->wait,
538 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
539 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
540 }
541 rcu_read_unlock();
542}
543
544/* When dgram socket disconnects (or changes its peer), we clear its receive
545 * queue of packets arrived from previous peer. First, it allows to do
546 * flow control based only on wmem_alloc; second, sk connected to peer
547 * may receive messages only from that peer. */
548static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
549{
550 if (!skb_queue_empty(&sk->sk_receive_queue)) {
551 skb_queue_purge(&sk->sk_receive_queue);
552 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
553
554 /* If one link of bidirectional dgram pipe is disconnected,
555 * we signal error. Messages are lost. Do not make this,
556 * when peer was not connected to us.
557 */
558 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
559 other->sk_err = ECONNRESET;
560 sk_error_report(other);
561 }
562 }
563 other->sk_state = TCP_CLOSE;
564}
565
566static void unix_sock_destructor(struct sock *sk)
567{
568 struct unix_sock *u = unix_sk(sk);
569
570 skb_queue_purge(&sk->sk_receive_queue);
571
572 DEBUG_NET_WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
573 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
574 DEBUG_NET_WARN_ON_ONCE(sk->sk_socket);
575 if (!sock_flag(sk, SOCK_DEAD)) {
576 pr_info("Attempt to release alive unix socket: %p\n", sk);
577 return;
578 }
579
580 if (u->addr)
581 unix_release_addr(u->addr);
582
583 atomic_long_dec(&unix_nr_socks);
584 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
585#ifdef UNIX_REFCNT_DEBUG
586 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
587 atomic_long_read(&unix_nr_socks));
588#endif
589}
590
591static void unix_release_sock(struct sock *sk, int embrion)
592{
593 struct unix_sock *u = unix_sk(sk);
594 struct sock *skpair;
595 struct sk_buff *skb;
596 struct path path;
597 int state;
598
599 unix_remove_socket(sock_net(sk), sk);
600 unix_remove_bsd_socket(sk);
601
602 /* Clear state */
603 unix_state_lock(sk);
604 sock_orphan(sk);
605 sk->sk_shutdown = SHUTDOWN_MASK;
606 path = u->path;
607 u->path.dentry = NULL;
608 u->path.mnt = NULL;
609 state = sk->sk_state;
610 sk->sk_state = TCP_CLOSE;
611
612 skpair = unix_peer(sk);
613 unix_peer(sk) = NULL;
614
615 unix_state_unlock(sk);
616
617#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
618 if (u->oob_skb) {
619 kfree_skb(u->oob_skb);
620 u->oob_skb = NULL;
621 }
622#endif
623
624 wake_up_interruptible_all(&u->peer_wait);
625
626 if (skpair != NULL) {
627 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
628 unix_state_lock(skpair);
629 /* No more writes */
630 skpair->sk_shutdown = SHUTDOWN_MASK;
631 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
632 skpair->sk_err = ECONNRESET;
633 unix_state_unlock(skpair);
634 skpair->sk_state_change(skpair);
635 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
636 }
637
638 unix_dgram_peer_wake_disconnect(sk, skpair);
639 sock_put(skpair); /* It may now die */
640 }
641
642 /* Try to flush out this socket. Throw out buffers at least */
643
644 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
645 if (state == TCP_LISTEN)
646 unix_release_sock(skb->sk, 1);
647 /* passed fds are erased in the kfree_skb hook */
648 UNIXCB(skb).consumed = skb->len;
649 kfree_skb(skb);
650 }
651
652 if (path.dentry)
653 path_put(&path);
654
655 sock_put(sk);
656
657 /* ---- Socket is dead now and most probably destroyed ---- */
658
659 /*
660 * Fixme: BSD difference: In BSD all sockets connected to us get
661 * ECONNRESET and we die on the spot. In Linux we behave
662 * like files and pipes do and wait for the last
663 * dereference.
664 *
665 * Can't we simply set sock->err?
666 *
667 * What the above comment does talk about? --ANK(980817)
668 */
669
670 if (unix_tot_inflight)
671 unix_gc(); /* Garbage collect fds */
672}
673
674static void init_peercred(struct sock *sk)
675{
676 const struct cred *old_cred;
677 struct pid *old_pid;
678
679 spin_lock(&sk->sk_peer_lock);
680 old_pid = sk->sk_peer_pid;
681 old_cred = sk->sk_peer_cred;
682 sk->sk_peer_pid = get_pid(task_tgid(current));
683 sk->sk_peer_cred = get_current_cred();
684 spin_unlock(&sk->sk_peer_lock);
685
686 put_pid(old_pid);
687 put_cred(old_cred);
688}
689
690static void copy_peercred(struct sock *sk, struct sock *peersk)
691{
692 const struct cred *old_cred;
693 struct pid *old_pid;
694
695 if (sk < peersk) {
696 spin_lock(&sk->sk_peer_lock);
697 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
698 } else {
699 spin_lock(&peersk->sk_peer_lock);
700 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
701 }
702 old_pid = sk->sk_peer_pid;
703 old_cred = sk->sk_peer_cred;
704 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
705 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
706
707 spin_unlock(&sk->sk_peer_lock);
708 spin_unlock(&peersk->sk_peer_lock);
709
710 put_pid(old_pid);
711 put_cred(old_cred);
712}
713
714static int unix_listen(struct socket *sock, int backlog)
715{
716 int err;
717 struct sock *sk = sock->sk;
718 struct unix_sock *u = unix_sk(sk);
719
720 err = -EOPNOTSUPP;
721 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
722 goto out; /* Only stream/seqpacket sockets accept */
723 err = -EINVAL;
724 if (!u->addr)
725 goto out; /* No listens on an unbound socket */
726 unix_state_lock(sk);
727 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
728 goto out_unlock;
729 if (backlog > sk->sk_max_ack_backlog)
730 wake_up_interruptible_all(&u->peer_wait);
731 sk->sk_max_ack_backlog = backlog;
732 sk->sk_state = TCP_LISTEN;
733 /* set credentials so connect can copy them */
734 init_peercred(sk);
735 err = 0;
736
737out_unlock:
738 unix_state_unlock(sk);
739out:
740 return err;
741}
742
743static int unix_release(struct socket *);
744static int unix_bind(struct socket *, struct sockaddr *, int);
745static int unix_stream_connect(struct socket *, struct sockaddr *,
746 int addr_len, int flags);
747static int unix_socketpair(struct socket *, struct socket *);
748static int unix_accept(struct socket *, struct socket *, int, bool);
749static int unix_getname(struct socket *, struct sockaddr *, int);
750static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
751static __poll_t unix_dgram_poll(struct file *, struct socket *,
752 poll_table *);
753static int unix_ioctl(struct socket *, unsigned int, unsigned long);
754#ifdef CONFIG_COMPAT
755static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
756#endif
757static int unix_shutdown(struct socket *, int);
758static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
759static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
760static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
761 size_t size, int flags);
762static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
763 struct pipe_inode_info *, size_t size,
764 unsigned int flags);
765static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
766static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
767static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
768static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
769static int unix_dgram_connect(struct socket *, struct sockaddr *,
770 int, int);
771static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
772static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
773 int);
774
775static int unix_set_peek_off(struct sock *sk, int val)
776{
777 struct unix_sock *u = unix_sk(sk);
778
779 if (mutex_lock_interruptible(&u->iolock))
780 return -EINTR;
781
782 sk->sk_peek_off = val;
783 mutex_unlock(&u->iolock);
784
785 return 0;
786}
787
788#ifdef CONFIG_PROC_FS
789static int unix_count_nr_fds(struct sock *sk)
790{
791 struct sk_buff *skb;
792 struct unix_sock *u;
793 int nr_fds = 0;
794
795 spin_lock(&sk->sk_receive_queue.lock);
796 skb = skb_peek(&sk->sk_receive_queue);
797 while (skb) {
798 u = unix_sk(skb->sk);
799 nr_fds += atomic_read(&u->scm_stat.nr_fds);
800 skb = skb_peek_next(skb, &sk->sk_receive_queue);
801 }
802 spin_unlock(&sk->sk_receive_queue.lock);
803
804 return nr_fds;
805}
806
807static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
808{
809 struct sock *sk = sock->sk;
810 struct unix_sock *u;
811 int nr_fds;
812
813 if (sk) {
814 u = unix_sk(sk);
815 if (sock->type == SOCK_DGRAM) {
816 nr_fds = atomic_read(&u->scm_stat.nr_fds);
817 goto out_print;
818 }
819
820 unix_state_lock(sk);
821 if (sk->sk_state != TCP_LISTEN)
822 nr_fds = atomic_read(&u->scm_stat.nr_fds);
823 else
824 nr_fds = unix_count_nr_fds(sk);
825 unix_state_unlock(sk);
826out_print:
827 seq_printf(m, "scm_fds: %u\n", nr_fds);
828 }
829}
830#else
831#define unix_show_fdinfo NULL
832#endif
833
834static const struct proto_ops unix_stream_ops = {
835 .family = PF_UNIX,
836 .owner = THIS_MODULE,
837 .release = unix_release,
838 .bind = unix_bind,
839 .connect = unix_stream_connect,
840 .socketpair = unix_socketpair,
841 .accept = unix_accept,
842 .getname = unix_getname,
843 .poll = unix_poll,
844 .ioctl = unix_ioctl,
845#ifdef CONFIG_COMPAT
846 .compat_ioctl = unix_compat_ioctl,
847#endif
848 .listen = unix_listen,
849 .shutdown = unix_shutdown,
850 .sendmsg = unix_stream_sendmsg,
851 .recvmsg = unix_stream_recvmsg,
852 .read_skb = unix_stream_read_skb,
853 .mmap = sock_no_mmap,
854 .sendpage = unix_stream_sendpage,
855 .splice_read = unix_stream_splice_read,
856 .set_peek_off = unix_set_peek_off,
857 .show_fdinfo = unix_show_fdinfo,
858};
859
860static const struct proto_ops unix_dgram_ops = {
861 .family = PF_UNIX,
862 .owner = THIS_MODULE,
863 .release = unix_release,
864 .bind = unix_bind,
865 .connect = unix_dgram_connect,
866 .socketpair = unix_socketpair,
867 .accept = sock_no_accept,
868 .getname = unix_getname,
869 .poll = unix_dgram_poll,
870 .ioctl = unix_ioctl,
871#ifdef CONFIG_COMPAT
872 .compat_ioctl = unix_compat_ioctl,
873#endif
874 .listen = sock_no_listen,
875 .shutdown = unix_shutdown,
876 .sendmsg = unix_dgram_sendmsg,
877 .read_skb = unix_read_skb,
878 .recvmsg = unix_dgram_recvmsg,
879 .mmap = sock_no_mmap,
880 .sendpage = sock_no_sendpage,
881 .set_peek_off = unix_set_peek_off,
882 .show_fdinfo = unix_show_fdinfo,
883};
884
885static const struct proto_ops unix_seqpacket_ops = {
886 .family = PF_UNIX,
887 .owner = THIS_MODULE,
888 .release = unix_release,
889 .bind = unix_bind,
890 .connect = unix_stream_connect,
891 .socketpair = unix_socketpair,
892 .accept = unix_accept,
893 .getname = unix_getname,
894 .poll = unix_dgram_poll,
895 .ioctl = unix_ioctl,
896#ifdef CONFIG_COMPAT
897 .compat_ioctl = unix_compat_ioctl,
898#endif
899 .listen = unix_listen,
900 .shutdown = unix_shutdown,
901 .sendmsg = unix_seqpacket_sendmsg,
902 .recvmsg = unix_seqpacket_recvmsg,
903 .mmap = sock_no_mmap,
904 .sendpage = sock_no_sendpage,
905 .set_peek_off = unix_set_peek_off,
906 .show_fdinfo = unix_show_fdinfo,
907};
908
909static void unix_close(struct sock *sk, long timeout)
910{
911 /* Nothing to do here, unix socket does not need a ->close().
912 * This is merely for sockmap.
913 */
914}
915
916static void unix_unhash(struct sock *sk)
917{
918 /* Nothing to do here, unix socket does not need a ->unhash().
919 * This is merely for sockmap.
920 */
921}
922
923struct proto unix_dgram_proto = {
924 .name = "UNIX",
925 .owner = THIS_MODULE,
926 .obj_size = sizeof(struct unix_sock),
927 .close = unix_close,
928#ifdef CONFIG_BPF_SYSCALL
929 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
930#endif
931};
932
933struct proto unix_stream_proto = {
934 .name = "UNIX-STREAM",
935 .owner = THIS_MODULE,
936 .obj_size = sizeof(struct unix_sock),
937 .close = unix_close,
938 .unhash = unix_unhash,
939#ifdef CONFIG_BPF_SYSCALL
940 .psock_update_sk_prot = unix_stream_bpf_update_proto,
941#endif
942};
943
944static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
945{
946 struct unix_sock *u;
947 struct sock *sk;
948 int err;
949
950 atomic_long_inc(&unix_nr_socks);
951 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
952 err = -ENFILE;
953 goto err;
954 }
955
956 if (type == SOCK_STREAM)
957 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
958 else /*dgram and seqpacket */
959 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
960
961 if (!sk) {
962 err = -ENOMEM;
963 goto err;
964 }
965
966 sock_init_data(sock, sk);
967
968 sk->sk_hash = unix_unbound_hash(sk);
969 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
970 sk->sk_write_space = unix_write_space;
971 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
972 sk->sk_destruct = unix_sock_destructor;
973 u = unix_sk(sk);
974 u->path.dentry = NULL;
975 u->path.mnt = NULL;
976 spin_lock_init(&u->lock);
977 atomic_long_set(&u->inflight, 0);
978 INIT_LIST_HEAD(&u->link);
979 mutex_init(&u->iolock); /* single task reading lock */
980 mutex_init(&u->bindlock); /* single task binding lock */
981 init_waitqueue_head(&u->peer_wait);
982 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
983 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
984 unix_insert_unbound_socket(net, sk);
985
986 sock_prot_inuse_add(net, sk->sk_prot, 1);
987
988 return sk;
989
990err:
991 atomic_long_dec(&unix_nr_socks);
992 return ERR_PTR(err);
993}
994
995static int unix_create(struct net *net, struct socket *sock, int protocol,
996 int kern)
997{
998 struct sock *sk;
999
1000 if (protocol && protocol != PF_UNIX)
1001 return -EPROTONOSUPPORT;
1002
1003 sock->state = SS_UNCONNECTED;
1004
1005 switch (sock->type) {
1006 case SOCK_STREAM:
1007 sock->ops = &unix_stream_ops;
1008 break;
1009 /*
1010 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
1011 * nothing uses it.
1012 */
1013 case SOCK_RAW:
1014 sock->type = SOCK_DGRAM;
1015 fallthrough;
1016 case SOCK_DGRAM:
1017 sock->ops = &unix_dgram_ops;
1018 break;
1019 case SOCK_SEQPACKET:
1020 sock->ops = &unix_seqpacket_ops;
1021 break;
1022 default:
1023 return -ESOCKTNOSUPPORT;
1024 }
1025
1026 sk = unix_create1(net, sock, kern, sock->type);
1027 if (IS_ERR(sk))
1028 return PTR_ERR(sk);
1029
1030 return 0;
1031}
1032
1033static int unix_release(struct socket *sock)
1034{
1035 struct sock *sk = sock->sk;
1036
1037 if (!sk)
1038 return 0;
1039
1040 sk->sk_prot->close(sk, 0);
1041 unix_release_sock(sk, 0);
1042 sock->sk = NULL;
1043
1044 return 0;
1045}
1046
1047static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1048 int type)
1049{
1050 struct inode *inode;
1051 struct path path;
1052 struct sock *sk;
1053 int err;
1054
1055 unix_mkname_bsd(sunaddr, addr_len);
1056 err = kern_path(sunaddr->sun_path, LOOKUP_FOLLOW, &path);
1057 if (err)
1058 goto fail;
1059
1060 err = path_permission(&path, MAY_WRITE);
1061 if (err)
1062 goto path_put;
1063
1064 err = -ECONNREFUSED;
1065 inode = d_backing_inode(path.dentry);
1066 if (!S_ISSOCK(inode->i_mode))
1067 goto path_put;
1068
1069 sk = unix_find_socket_byinode(inode);
1070 if (!sk)
1071 goto path_put;
1072
1073 err = -EPROTOTYPE;
1074 if (sk->sk_type == type)
1075 touch_atime(&path);
1076 else
1077 goto sock_put;
1078
1079 path_put(&path);
1080
1081 return sk;
1082
1083sock_put:
1084 sock_put(sk);
1085path_put:
1086 path_put(&path);
1087fail:
1088 return ERR_PTR(err);
1089}
1090
1091static struct sock *unix_find_abstract(struct net *net,
1092 struct sockaddr_un *sunaddr,
1093 int addr_len, int type)
1094{
1095 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
1096 struct dentry *dentry;
1097 struct sock *sk;
1098
1099 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
1100 if (!sk)
1101 return ERR_PTR(-ECONNREFUSED);
1102
1103 dentry = unix_sk(sk)->path.dentry;
1104 if (dentry)
1105 touch_atime(&unix_sk(sk)->path);
1106
1107 return sk;
1108}
1109
1110static struct sock *unix_find_other(struct net *net,
1111 struct sockaddr_un *sunaddr,
1112 int addr_len, int type)
1113{
1114 struct sock *sk;
1115
1116 if (sunaddr->sun_path[0])
1117 sk = unix_find_bsd(sunaddr, addr_len, type);
1118 else
1119 sk = unix_find_abstract(net, sunaddr, addr_len, type);
1120
1121 return sk;
1122}
1123
1124static int unix_autobind(struct sock *sk)
1125{
1126 unsigned int new_hash, old_hash = sk->sk_hash;
1127 struct unix_sock *u = unix_sk(sk);
1128 struct net *net = sock_net(sk);
1129 struct unix_address *addr;
1130 u32 lastnum, ordernum;
1131 int err;
1132
1133 err = mutex_lock_interruptible(&u->bindlock);
1134 if (err)
1135 return err;
1136
1137 if (u->addr)
1138 goto out;
1139
1140 err = -ENOMEM;
1141 addr = kzalloc(sizeof(*addr) +
1142 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1143 if (!addr)
1144 goto out;
1145
1146 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1147 addr->name->sun_family = AF_UNIX;
1148 refcount_set(&addr->refcnt, 1);
1149
1150 ordernum = get_random_u32();
1151 lastnum = ordernum & 0xFFFFF;
1152retry:
1153 ordernum = (ordernum + 1) & 0xFFFFF;
1154 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1155
1156 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1157 unix_table_double_lock(net, old_hash, new_hash);
1158
1159 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1160 unix_table_double_unlock(net, old_hash, new_hash);
1161
1162 /* __unix_find_socket_byname() may take long time if many names
1163 * are already in use.
1164 */
1165 cond_resched();
1166
1167 if (ordernum == lastnum) {
1168 /* Give up if all names seems to be in use. */
1169 err = -ENOSPC;
1170 unix_release_addr(addr);
1171 goto out;
1172 }
1173
1174 goto retry;
1175 }
1176
1177 __unix_set_addr_hash(net, sk, addr, new_hash);
1178 unix_table_double_unlock(net, old_hash, new_hash);
1179 err = 0;
1180
1181out: mutex_unlock(&u->bindlock);
1182 return err;
1183}
1184
1185static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1186 int addr_len)
1187{
1188 umode_t mode = S_IFSOCK |
1189 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1190 unsigned int new_hash, old_hash = sk->sk_hash;
1191 struct unix_sock *u = unix_sk(sk);
1192 struct net *net = sock_net(sk);
1193 struct user_namespace *ns; // barf...
1194 struct unix_address *addr;
1195 struct dentry *dentry;
1196 struct path parent;
1197 int err;
1198
1199 unix_mkname_bsd(sunaddr, addr_len);
1200 addr_len = strlen(sunaddr->sun_path) +
1201 offsetof(struct sockaddr_un, sun_path) + 1;
1202
1203 addr = unix_create_addr(sunaddr, addr_len);
1204 if (!addr)
1205 return -ENOMEM;
1206
1207 /*
1208 * Get the parent directory, calculate the hash for last
1209 * component.
1210 */
1211 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1212 if (IS_ERR(dentry)) {
1213 err = PTR_ERR(dentry);
1214 goto out;
1215 }
1216
1217 /*
1218 * All right, let's create it.
1219 */
1220 ns = mnt_user_ns(parent.mnt);
1221 err = security_path_mknod(&parent, dentry, mode, 0);
1222 if (!err)
1223 err = vfs_mknod(ns, d_inode(parent.dentry), dentry, mode, 0);
1224 if (err)
1225 goto out_path;
1226 err = mutex_lock_interruptible(&u->bindlock);
1227 if (err)
1228 goto out_unlink;
1229 if (u->addr)
1230 goto out_unlock;
1231
1232 new_hash = unix_bsd_hash(d_backing_inode(dentry));
1233 unix_table_double_lock(net, old_hash, new_hash);
1234 u->path.mnt = mntget(parent.mnt);
1235 u->path.dentry = dget(dentry);
1236 __unix_set_addr_hash(net, sk, addr, new_hash);
1237 unix_table_double_unlock(net, old_hash, new_hash);
1238 unix_insert_bsd_socket(sk);
1239 mutex_unlock(&u->bindlock);
1240 done_path_create(&parent, dentry);
1241 return 0;
1242
1243out_unlock:
1244 mutex_unlock(&u->bindlock);
1245 err = -EINVAL;
1246out_unlink:
1247 /* failed after successful mknod? unlink what we'd created... */
1248 vfs_unlink(ns, d_inode(parent.dentry), dentry, NULL);
1249out_path:
1250 done_path_create(&parent, dentry);
1251out:
1252 unix_release_addr(addr);
1253 return err == -EEXIST ? -EADDRINUSE : err;
1254}
1255
1256static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1257 int addr_len)
1258{
1259 unsigned int new_hash, old_hash = sk->sk_hash;
1260 struct unix_sock *u = unix_sk(sk);
1261 struct net *net = sock_net(sk);
1262 struct unix_address *addr;
1263 int err;
1264
1265 addr = unix_create_addr(sunaddr, addr_len);
1266 if (!addr)
1267 return -ENOMEM;
1268
1269 err = mutex_lock_interruptible(&u->bindlock);
1270 if (err)
1271 goto out;
1272
1273 if (u->addr) {
1274 err = -EINVAL;
1275 goto out_mutex;
1276 }
1277
1278 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1279 unix_table_double_lock(net, old_hash, new_hash);
1280
1281 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
1282 goto out_spin;
1283
1284 __unix_set_addr_hash(net, sk, addr, new_hash);
1285 unix_table_double_unlock(net, old_hash, new_hash);
1286 mutex_unlock(&u->bindlock);
1287 return 0;
1288
1289out_spin:
1290 unix_table_double_unlock(net, old_hash, new_hash);
1291 err = -EADDRINUSE;
1292out_mutex:
1293 mutex_unlock(&u->bindlock);
1294out:
1295 unix_release_addr(addr);
1296 return err;
1297}
1298
1299static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1300{
1301 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1302 struct sock *sk = sock->sk;
1303 int err;
1304
1305 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1306 sunaddr->sun_family == AF_UNIX)
1307 return unix_autobind(sk);
1308
1309 err = unix_validate_addr(sunaddr, addr_len);
1310 if (err)
1311 return err;
1312
1313 if (sunaddr->sun_path[0])
1314 err = unix_bind_bsd(sk, sunaddr, addr_len);
1315 else
1316 err = unix_bind_abstract(sk, sunaddr, addr_len);
1317
1318 return err;
1319}
1320
1321static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1322{
1323 if (unlikely(sk1 == sk2) || !sk2) {
1324 unix_state_lock(sk1);
1325 return;
1326 }
1327 if (sk1 < sk2) {
1328 unix_state_lock(sk1);
1329 unix_state_lock_nested(sk2);
1330 } else {
1331 unix_state_lock(sk2);
1332 unix_state_lock_nested(sk1);
1333 }
1334}
1335
1336static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1337{
1338 if (unlikely(sk1 == sk2) || !sk2) {
1339 unix_state_unlock(sk1);
1340 return;
1341 }
1342 unix_state_unlock(sk1);
1343 unix_state_unlock(sk2);
1344}
1345
1346static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1347 int alen, int flags)
1348{
1349 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1350 struct sock *sk = sock->sk;
1351 struct sock *other;
1352 int err;
1353
1354 err = -EINVAL;
1355 if (alen < offsetofend(struct sockaddr, sa_family))
1356 goto out;
1357
1358 if (addr->sa_family != AF_UNSPEC) {
1359 err = unix_validate_addr(sunaddr, alen);
1360 if (err)
1361 goto out;
1362
1363 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1364 !unix_sk(sk)->addr) {
1365 err = unix_autobind(sk);
1366 if (err)
1367 goto out;
1368 }
1369
1370restart:
1371 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
1372 if (IS_ERR(other)) {
1373 err = PTR_ERR(other);
1374 goto out;
1375 }
1376
1377 unix_state_double_lock(sk, other);
1378
1379 /* Apparently VFS overslept socket death. Retry. */
1380 if (sock_flag(other, SOCK_DEAD)) {
1381 unix_state_double_unlock(sk, other);
1382 sock_put(other);
1383 goto restart;
1384 }
1385
1386 err = -EPERM;
1387 if (!unix_may_send(sk, other))
1388 goto out_unlock;
1389
1390 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1391 if (err)
1392 goto out_unlock;
1393
1394 sk->sk_state = other->sk_state = TCP_ESTABLISHED;
1395 } else {
1396 /*
1397 * 1003.1g breaking connected state with AF_UNSPEC
1398 */
1399 other = NULL;
1400 unix_state_double_lock(sk, other);
1401 }
1402
1403 /*
1404 * If it was connected, reconnect.
1405 */
1406 if (unix_peer(sk)) {
1407 struct sock *old_peer = unix_peer(sk);
1408
1409 unix_peer(sk) = other;
1410 if (!other)
1411 sk->sk_state = TCP_CLOSE;
1412 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1413
1414 unix_state_double_unlock(sk, other);
1415
1416 if (other != old_peer)
1417 unix_dgram_disconnected(sk, old_peer);
1418 sock_put(old_peer);
1419 } else {
1420 unix_peer(sk) = other;
1421 unix_state_double_unlock(sk, other);
1422 }
1423
1424 return 0;
1425
1426out_unlock:
1427 unix_state_double_unlock(sk, other);
1428 sock_put(other);
1429out:
1430 return err;
1431}
1432
1433static long unix_wait_for_peer(struct sock *other, long timeo)
1434 __releases(&unix_sk(other)->lock)
1435{
1436 struct unix_sock *u = unix_sk(other);
1437 int sched;
1438 DEFINE_WAIT(wait);
1439
1440 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1441
1442 sched = !sock_flag(other, SOCK_DEAD) &&
1443 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1444 unix_recvq_full(other);
1445
1446 unix_state_unlock(other);
1447
1448 if (sched)
1449 timeo = schedule_timeout(timeo);
1450
1451 finish_wait(&u->peer_wait, &wait);
1452 return timeo;
1453}
1454
1455static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1456 int addr_len, int flags)
1457{
1458 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1459 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1460 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1461 struct net *net = sock_net(sk);
1462 struct sk_buff *skb = NULL;
1463 long timeo;
1464 int err;
1465 int st;
1466
1467 err = unix_validate_addr(sunaddr, addr_len);
1468 if (err)
1469 goto out;
1470
1471 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr) {
1472 err = unix_autobind(sk);
1473 if (err)
1474 goto out;
1475 }
1476
1477 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1478
1479 /* First of all allocate resources.
1480 If we will make it after state is locked,
1481 we will have to recheck all again in any case.
1482 */
1483
1484 /* create new sock for complete connection */
1485 newsk = unix_create1(net, NULL, 0, sock->type);
1486 if (IS_ERR(newsk)) {
1487 err = PTR_ERR(newsk);
1488 newsk = NULL;
1489 goto out;
1490 }
1491
1492 err = -ENOMEM;
1493
1494 /* Allocate skb for sending to listening sock */
1495 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1496 if (skb == NULL)
1497 goto out;
1498
1499restart:
1500 /* Find listening sock. */
1501 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
1502 if (IS_ERR(other)) {
1503 err = PTR_ERR(other);
1504 other = NULL;
1505 goto out;
1506 }
1507
1508 /* Latch state of peer */
1509 unix_state_lock(other);
1510
1511 /* Apparently VFS overslept socket death. Retry. */
1512 if (sock_flag(other, SOCK_DEAD)) {
1513 unix_state_unlock(other);
1514 sock_put(other);
1515 goto restart;
1516 }
1517
1518 err = -ECONNREFUSED;
1519 if (other->sk_state != TCP_LISTEN)
1520 goto out_unlock;
1521 if (other->sk_shutdown & RCV_SHUTDOWN)
1522 goto out_unlock;
1523
1524 if (unix_recvq_full(other)) {
1525 err = -EAGAIN;
1526 if (!timeo)
1527 goto out_unlock;
1528
1529 timeo = unix_wait_for_peer(other, timeo);
1530
1531 err = sock_intr_errno(timeo);
1532 if (signal_pending(current))
1533 goto out;
1534 sock_put(other);
1535 goto restart;
1536 }
1537
1538 /* Latch our state.
1539
1540 It is tricky place. We need to grab our state lock and cannot
1541 drop lock on peer. It is dangerous because deadlock is
1542 possible. Connect to self case and simultaneous
1543 attempt to connect are eliminated by checking socket
1544 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1545 check this before attempt to grab lock.
1546
1547 Well, and we have to recheck the state after socket locked.
1548 */
1549 st = sk->sk_state;
1550
1551 switch (st) {
1552 case TCP_CLOSE:
1553 /* This is ok... continue with connect */
1554 break;
1555 case TCP_ESTABLISHED:
1556 /* Socket is already connected */
1557 err = -EISCONN;
1558 goto out_unlock;
1559 default:
1560 err = -EINVAL;
1561 goto out_unlock;
1562 }
1563
1564 unix_state_lock_nested(sk);
1565
1566 if (sk->sk_state != st) {
1567 unix_state_unlock(sk);
1568 unix_state_unlock(other);
1569 sock_put(other);
1570 goto restart;
1571 }
1572
1573 err = security_unix_stream_connect(sk, other, newsk);
1574 if (err) {
1575 unix_state_unlock(sk);
1576 goto out_unlock;
1577 }
1578
1579 /* The way is open! Fastly set all the necessary fields... */
1580
1581 sock_hold(sk);
1582 unix_peer(newsk) = sk;
1583 newsk->sk_state = TCP_ESTABLISHED;
1584 newsk->sk_type = sk->sk_type;
1585 init_peercred(newsk);
1586 newu = unix_sk(newsk);
1587 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1588 otheru = unix_sk(other);
1589
1590 /* copy address information from listening to new sock
1591 *
1592 * The contents of *(otheru->addr) and otheru->path
1593 * are seen fully set up here, since we have found
1594 * otheru in hash under its lock. Insertion into the
1595 * hash chain we'd found it in had been done in an
1596 * earlier critical area protected by the chain's lock,
1597 * the same one where we'd set *(otheru->addr) contents,
1598 * as well as otheru->path and otheru->addr itself.
1599 *
1600 * Using smp_store_release() here to set newu->addr
1601 * is enough to make those stores, as well as stores
1602 * to newu->path visible to anyone who gets newu->addr
1603 * by smp_load_acquire(). IOW, the same warranties
1604 * as for unix_sock instances bound in unix_bind() or
1605 * in unix_autobind().
1606 */
1607 if (otheru->path.dentry) {
1608 path_get(&otheru->path);
1609 newu->path = otheru->path;
1610 }
1611 refcount_inc(&otheru->addr->refcnt);
1612 smp_store_release(&newu->addr, otheru->addr);
1613
1614 /* Set credentials */
1615 copy_peercred(sk, other);
1616
1617 sock->state = SS_CONNECTED;
1618 sk->sk_state = TCP_ESTABLISHED;
1619 sock_hold(newsk);
1620
1621 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1622 unix_peer(sk) = newsk;
1623
1624 unix_state_unlock(sk);
1625
1626 /* take ten and send info to listening sock */
1627 spin_lock(&other->sk_receive_queue.lock);
1628 __skb_queue_tail(&other->sk_receive_queue, skb);
1629 spin_unlock(&other->sk_receive_queue.lock);
1630 unix_state_unlock(other);
1631 other->sk_data_ready(other);
1632 sock_put(other);
1633 return 0;
1634
1635out_unlock:
1636 if (other)
1637 unix_state_unlock(other);
1638
1639out:
1640 kfree_skb(skb);
1641 if (newsk)
1642 unix_release_sock(newsk, 0);
1643 if (other)
1644 sock_put(other);
1645 return err;
1646}
1647
1648static int unix_socketpair(struct socket *socka, struct socket *sockb)
1649{
1650 struct sock *ska = socka->sk, *skb = sockb->sk;
1651
1652 /* Join our sockets back to back */
1653 sock_hold(ska);
1654 sock_hold(skb);
1655 unix_peer(ska) = skb;
1656 unix_peer(skb) = ska;
1657 init_peercred(ska);
1658 init_peercred(skb);
1659
1660 ska->sk_state = TCP_ESTABLISHED;
1661 skb->sk_state = TCP_ESTABLISHED;
1662 socka->state = SS_CONNECTED;
1663 sockb->state = SS_CONNECTED;
1664 return 0;
1665}
1666
1667static void unix_sock_inherit_flags(const struct socket *old,
1668 struct socket *new)
1669{
1670 if (test_bit(SOCK_PASSCRED, &old->flags))
1671 set_bit(SOCK_PASSCRED, &new->flags);
1672 if (test_bit(SOCK_PASSSEC, &old->flags))
1673 set_bit(SOCK_PASSSEC, &new->flags);
1674}
1675
1676static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1677 bool kern)
1678{
1679 struct sock *sk = sock->sk;
1680 struct sock *tsk;
1681 struct sk_buff *skb;
1682 int err;
1683
1684 err = -EOPNOTSUPP;
1685 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1686 goto out;
1687
1688 err = -EINVAL;
1689 if (sk->sk_state != TCP_LISTEN)
1690 goto out;
1691
1692 /* If socket state is TCP_LISTEN it cannot change (for now...),
1693 * so that no locks are necessary.
1694 */
1695
1696 skb = skb_recv_datagram(sk, (flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1697 &err);
1698 if (!skb) {
1699 /* This means receive shutdown. */
1700 if (err == 0)
1701 err = -EINVAL;
1702 goto out;
1703 }
1704
1705 tsk = skb->sk;
1706 skb_free_datagram(sk, skb);
1707 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1708
1709 /* attach accepted sock to socket */
1710 unix_state_lock(tsk);
1711 newsock->state = SS_CONNECTED;
1712 unix_sock_inherit_flags(sock, newsock);
1713 sock_graft(tsk, newsock);
1714 unix_state_unlock(tsk);
1715 return 0;
1716
1717out:
1718 return err;
1719}
1720
1721
1722static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1723{
1724 struct sock *sk = sock->sk;
1725 struct unix_address *addr;
1726 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1727 int err = 0;
1728
1729 if (peer) {
1730 sk = unix_peer_get(sk);
1731
1732 err = -ENOTCONN;
1733 if (!sk)
1734 goto out;
1735 err = 0;
1736 } else {
1737 sock_hold(sk);
1738 }
1739
1740 addr = smp_load_acquire(&unix_sk(sk)->addr);
1741 if (!addr) {
1742 sunaddr->sun_family = AF_UNIX;
1743 sunaddr->sun_path[0] = 0;
1744 err = offsetof(struct sockaddr_un, sun_path);
1745 } else {
1746 err = addr->len;
1747 memcpy(sunaddr, addr->name, addr->len);
1748 }
1749 sock_put(sk);
1750out:
1751 return err;
1752}
1753
1754static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1755{
1756 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1757
1758 /*
1759 * Garbage collection of unix sockets starts by selecting a set of
1760 * candidate sockets which have reference only from being in flight
1761 * (total_refs == inflight_refs). This condition is checked once during
1762 * the candidate collection phase, and candidates are marked as such, so
1763 * that non-candidates can later be ignored. While inflight_refs is
1764 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1765 * is an instantaneous decision.
1766 *
1767 * Once a candidate, however, the socket must not be reinstalled into a
1768 * file descriptor while the garbage collection is in progress.
1769 *
1770 * If the above conditions are met, then the directed graph of
1771 * candidates (*) does not change while unix_gc_lock is held.
1772 *
1773 * Any operations that changes the file count through file descriptors
1774 * (dup, close, sendmsg) does not change the graph since candidates are
1775 * not installed in fds.
1776 *
1777 * Dequeing a candidate via recvmsg would install it into an fd, but
1778 * that takes unix_gc_lock to decrement the inflight count, so it's
1779 * serialized with garbage collection.
1780 *
1781 * MSG_PEEK is special in that it does not change the inflight count,
1782 * yet does install the socket into an fd. The following lock/unlock
1783 * pair is to ensure serialization with garbage collection. It must be
1784 * done between incrementing the file count and installing the file into
1785 * an fd.
1786 *
1787 * If garbage collection starts after the barrier provided by the
1788 * lock/unlock, then it will see the elevated refcount and not mark this
1789 * as a candidate. If a garbage collection is already in progress
1790 * before the file count was incremented, then the lock/unlock pair will
1791 * ensure that garbage collection is finished before progressing to
1792 * installing the fd.
1793 *
1794 * (*) A -> B where B is on the queue of A or B is on the queue of C
1795 * which is on the queue of listening socket A.
1796 */
1797 spin_lock(&unix_gc_lock);
1798 spin_unlock(&unix_gc_lock);
1799}
1800
1801static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1802{
1803 int err = 0;
1804
1805 UNIXCB(skb).pid = get_pid(scm->pid);
1806 UNIXCB(skb).uid = scm->creds.uid;
1807 UNIXCB(skb).gid = scm->creds.gid;
1808 UNIXCB(skb).fp = NULL;
1809 unix_get_secdata(scm, skb);
1810 if (scm->fp && send_fds)
1811 err = unix_attach_fds(scm, skb);
1812
1813 skb->destructor = unix_destruct_scm;
1814 return err;
1815}
1816
1817static bool unix_passcred_enabled(const struct socket *sock,
1818 const struct sock *other)
1819{
1820 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1821 !other->sk_socket ||
1822 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1823}
1824
1825/*
1826 * Some apps rely on write() giving SCM_CREDENTIALS
1827 * We include credentials if source or destination socket
1828 * asserted SOCK_PASSCRED.
1829 */
1830static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1831 const struct sock *other)
1832{
1833 if (UNIXCB(skb).pid)
1834 return;
1835 if (unix_passcred_enabled(sock, other)) {
1836 UNIXCB(skb).pid = get_pid(task_tgid(current));
1837 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1838 }
1839}
1840
1841static int maybe_init_creds(struct scm_cookie *scm,
1842 struct socket *socket,
1843 const struct sock *other)
1844{
1845 int err;
1846 struct msghdr msg = { .msg_controllen = 0 };
1847
1848 err = scm_send(socket, &msg, scm, false);
1849 if (err)
1850 return err;
1851
1852 if (unix_passcred_enabled(socket, other)) {
1853 scm->pid = get_pid(task_tgid(current));
1854 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1855 }
1856 return err;
1857}
1858
1859static bool unix_skb_scm_eq(struct sk_buff *skb,
1860 struct scm_cookie *scm)
1861{
1862 return UNIXCB(skb).pid == scm->pid &&
1863 uid_eq(UNIXCB(skb).uid, scm->creds.uid) &&
1864 gid_eq(UNIXCB(skb).gid, scm->creds.gid) &&
1865 unix_secdata_eq(scm, skb);
1866}
1867
1868static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1869{
1870 struct scm_fp_list *fp = UNIXCB(skb).fp;
1871 struct unix_sock *u = unix_sk(sk);
1872
1873 if (unlikely(fp && fp->count))
1874 atomic_add(fp->count, &u->scm_stat.nr_fds);
1875}
1876
1877static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1878{
1879 struct scm_fp_list *fp = UNIXCB(skb).fp;
1880 struct unix_sock *u = unix_sk(sk);
1881
1882 if (unlikely(fp && fp->count))
1883 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1884}
1885
1886/*
1887 * Send AF_UNIX data.
1888 */
1889
1890static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1891 size_t len)
1892{
1893 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1894 struct sock *sk = sock->sk, *other = NULL;
1895 struct unix_sock *u = unix_sk(sk);
1896 struct scm_cookie scm;
1897 struct sk_buff *skb;
1898 int data_len = 0;
1899 int sk_locked;
1900 long timeo;
1901 int err;
1902
1903 wait_for_unix_gc();
1904 err = scm_send(sock, msg, &scm, false);
1905 if (err < 0)
1906 return err;
1907
1908 err = -EOPNOTSUPP;
1909 if (msg->msg_flags&MSG_OOB)
1910 goto out;
1911
1912 if (msg->msg_namelen) {
1913 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1914 if (err)
1915 goto out;
1916 } else {
1917 sunaddr = NULL;
1918 err = -ENOTCONN;
1919 other = unix_peer_get(sk);
1920 if (!other)
1921 goto out;
1922 }
1923
1924 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr) {
1925 err = unix_autobind(sk);
1926 if (err)
1927 goto out;
1928 }
1929
1930 err = -EMSGSIZE;
1931 if (len > sk->sk_sndbuf - 32)
1932 goto out;
1933
1934 if (len > SKB_MAX_ALLOC) {
1935 data_len = min_t(size_t,
1936 len - SKB_MAX_ALLOC,
1937 MAX_SKB_FRAGS * PAGE_SIZE);
1938 data_len = PAGE_ALIGN(data_len);
1939
1940 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1941 }
1942
1943 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1944 msg->msg_flags & MSG_DONTWAIT, &err,
1945 PAGE_ALLOC_COSTLY_ORDER);
1946 if (skb == NULL)
1947 goto out;
1948
1949 err = unix_scm_to_skb(&scm, skb, true);
1950 if (err < 0)
1951 goto out_free;
1952
1953 skb_put(skb, len - data_len);
1954 skb->data_len = data_len;
1955 skb->len = len;
1956 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1957 if (err)
1958 goto out_free;
1959
1960 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1961
1962restart:
1963 if (!other) {
1964 err = -ECONNRESET;
1965 if (sunaddr == NULL)
1966 goto out_free;
1967
1968 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
1969 sk->sk_type);
1970 if (IS_ERR(other)) {
1971 err = PTR_ERR(other);
1972 other = NULL;
1973 goto out_free;
1974 }
1975 }
1976
1977 if (sk_filter(other, skb) < 0) {
1978 /* Toss the packet but do not return any error to the sender */
1979 err = len;
1980 goto out_free;
1981 }
1982
1983 sk_locked = 0;
1984 unix_state_lock(other);
1985restart_locked:
1986 err = -EPERM;
1987 if (!unix_may_send(sk, other))
1988 goto out_unlock;
1989
1990 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1991 /*
1992 * Check with 1003.1g - what should
1993 * datagram error
1994 */
1995 unix_state_unlock(other);
1996 sock_put(other);
1997
1998 if (!sk_locked)
1999 unix_state_lock(sk);
2000
2001 err = 0;
2002 if (sk->sk_type == SOCK_SEQPACKET) {
2003 /* We are here only when racing with unix_release_sock()
2004 * is clearing @other. Never change state to TCP_CLOSE
2005 * unlike SOCK_DGRAM wants.
2006 */
2007 unix_state_unlock(sk);
2008 err = -EPIPE;
2009 } else if (unix_peer(sk) == other) {
2010 unix_peer(sk) = NULL;
2011 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2012
2013 sk->sk_state = TCP_CLOSE;
2014 unix_state_unlock(sk);
2015
2016 unix_dgram_disconnected(sk, other);
2017 sock_put(other);
2018 err = -ECONNREFUSED;
2019 } else {
2020 unix_state_unlock(sk);
2021 }
2022
2023 other = NULL;
2024 if (err)
2025 goto out_free;
2026 goto restart;
2027 }
2028
2029 err = -EPIPE;
2030 if (other->sk_shutdown & RCV_SHUTDOWN)
2031 goto out_unlock;
2032
2033 if (sk->sk_type != SOCK_SEQPACKET) {
2034 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
2035 if (err)
2036 goto out_unlock;
2037 }
2038
2039 /* other == sk && unix_peer(other) != sk if
2040 * - unix_peer(sk) == NULL, destination address bound to sk
2041 * - unix_peer(sk) == sk by time of get but disconnected before lock
2042 */
2043 if (other != sk &&
2044 unlikely(unix_peer(other) != sk &&
2045 unix_recvq_full_lockless(other))) {
2046 if (timeo) {
2047 timeo = unix_wait_for_peer(other, timeo);
2048
2049 err = sock_intr_errno(timeo);
2050 if (signal_pending(current))
2051 goto out_free;
2052
2053 goto restart;
2054 }
2055
2056 if (!sk_locked) {
2057 unix_state_unlock(other);
2058 unix_state_double_lock(sk, other);
2059 }
2060
2061 if (unix_peer(sk) != other ||
2062 unix_dgram_peer_wake_me(sk, other)) {
2063 err = -EAGAIN;
2064 sk_locked = 1;
2065 goto out_unlock;
2066 }
2067
2068 if (!sk_locked) {
2069 sk_locked = 1;
2070 goto restart_locked;
2071 }
2072 }
2073
2074 if (unlikely(sk_locked))
2075 unix_state_unlock(sk);
2076
2077 if (sock_flag(other, SOCK_RCVTSTAMP))
2078 __net_timestamp(skb);
2079 maybe_add_creds(skb, sock, other);
2080 scm_stat_add(other, skb);
2081 skb_queue_tail(&other->sk_receive_queue, skb);
2082 unix_state_unlock(other);
2083 other->sk_data_ready(other);
2084 sock_put(other);
2085 scm_destroy(&scm);
2086 return len;
2087
2088out_unlock:
2089 if (sk_locked)
2090 unix_state_unlock(sk);
2091 unix_state_unlock(other);
2092out_free:
2093 kfree_skb(skb);
2094out:
2095 if (other)
2096 sock_put(other);
2097 scm_destroy(&scm);
2098 return err;
2099}
2100
2101/* We use paged skbs for stream sockets, and limit occupancy to 32768
2102 * bytes, and a minimum of a full page.
2103 */
2104#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
2105
2106#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2107static int queue_oob(struct socket *sock, struct msghdr *msg, struct sock *other)
2108{
2109 struct unix_sock *ousk = unix_sk(other);
2110 struct sk_buff *skb;
2111 int err = 0;
2112
2113 skb = sock_alloc_send_skb(sock->sk, 1, msg->msg_flags & MSG_DONTWAIT, &err);
2114
2115 if (!skb)
2116 return err;
2117
2118 skb_put(skb, 1);
2119 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
2120
2121 if (err) {
2122 kfree_skb(skb);
2123 return err;
2124 }
2125
2126 unix_state_lock(other);
2127
2128 if (sock_flag(other, SOCK_DEAD) ||
2129 (other->sk_shutdown & RCV_SHUTDOWN)) {
2130 unix_state_unlock(other);
2131 kfree_skb(skb);
2132 return -EPIPE;
2133 }
2134
2135 maybe_add_creds(skb, sock, other);
2136 skb_get(skb);
2137
2138 if (ousk->oob_skb)
2139 consume_skb(ousk->oob_skb);
2140
2141 WRITE_ONCE(ousk->oob_skb, skb);
2142
2143 scm_stat_add(other, skb);
2144 skb_queue_tail(&other->sk_receive_queue, skb);
2145 sk_send_sigurg(other);
2146 unix_state_unlock(other);
2147 other->sk_data_ready(other);
2148
2149 return err;
2150}
2151#endif
2152
2153static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2154 size_t len)
2155{
2156 struct sock *sk = sock->sk;
2157 struct sock *other = NULL;
2158 int err, size;
2159 struct sk_buff *skb;
2160 int sent = 0;
2161 struct scm_cookie scm;
2162 bool fds_sent = false;
2163 int data_len;
2164
2165 wait_for_unix_gc();
2166 err = scm_send(sock, msg, &scm, false);
2167 if (err < 0)
2168 return err;
2169
2170 err = -EOPNOTSUPP;
2171 if (msg->msg_flags & MSG_OOB) {
2172#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2173 if (len)
2174 len--;
2175 else
2176#endif
2177 goto out_err;
2178 }
2179
2180 if (msg->msg_namelen) {
2181 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2182 goto out_err;
2183 } else {
2184 err = -ENOTCONN;
2185 other = unix_peer(sk);
2186 if (!other)
2187 goto out_err;
2188 }
2189
2190 if (sk->sk_shutdown & SEND_SHUTDOWN)
2191 goto pipe_err;
2192
2193 while (sent < len) {
2194 size = len - sent;
2195
2196 /* Keep two messages in the pipe so it schedules better */
2197 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
2198
2199 /* allow fallback to order-0 allocations */
2200 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
2201
2202 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
2203
2204 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2205
2206 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
2207 msg->msg_flags & MSG_DONTWAIT, &err,
2208 get_order(UNIX_SKB_FRAGS_SZ));
2209 if (!skb)
2210 goto out_err;
2211
2212 /* Only send the fds in the first buffer */
2213 err = unix_scm_to_skb(&scm, skb, !fds_sent);
2214 if (err < 0) {
2215 kfree_skb(skb);
2216 goto out_err;
2217 }
2218 fds_sent = true;
2219
2220 skb_put(skb, size - data_len);
2221 skb->data_len = data_len;
2222 skb->len = size;
2223 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
2224 if (err) {
2225 kfree_skb(skb);
2226 goto out_err;
2227 }
2228
2229 unix_state_lock(other);
2230
2231 if (sock_flag(other, SOCK_DEAD) ||
2232 (other->sk_shutdown & RCV_SHUTDOWN))
2233 goto pipe_err_free;
2234
2235 maybe_add_creds(skb, sock, other);
2236 scm_stat_add(other, skb);
2237 skb_queue_tail(&other->sk_receive_queue, skb);
2238 unix_state_unlock(other);
2239 other->sk_data_ready(other);
2240 sent += size;
2241 }
2242
2243#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2244 if (msg->msg_flags & MSG_OOB) {
2245 err = queue_oob(sock, msg, other);
2246 if (err)
2247 goto out_err;
2248 sent++;
2249 }
2250#endif
2251
2252 scm_destroy(&scm);
2253
2254 return sent;
2255
2256pipe_err_free:
2257 unix_state_unlock(other);
2258 kfree_skb(skb);
2259pipe_err:
2260 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
2261 send_sig(SIGPIPE, current, 0);
2262 err = -EPIPE;
2263out_err:
2264 scm_destroy(&scm);
2265 return sent ? : err;
2266}
2267
2268static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
2269 int offset, size_t size, int flags)
2270{
2271 int err;
2272 bool send_sigpipe = false;
2273 bool init_scm = true;
2274 struct scm_cookie scm;
2275 struct sock *other, *sk = socket->sk;
2276 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
2277
2278 if (flags & MSG_OOB)
2279 return -EOPNOTSUPP;
2280
2281 other = unix_peer(sk);
2282 if (!other || sk->sk_state != TCP_ESTABLISHED)
2283 return -ENOTCONN;
2284
2285 if (false) {
2286alloc_skb:
2287 unix_state_unlock(other);
2288 mutex_unlock(&unix_sk(other)->iolock);
2289 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
2290 &err, 0);
2291 if (!newskb)
2292 goto err;
2293 }
2294
2295 /* we must acquire iolock as we modify already present
2296 * skbs in the sk_receive_queue and mess with skb->len
2297 */
2298 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
2299 if (err) {
2300 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
2301 goto err;
2302 }
2303
2304 if (sk->sk_shutdown & SEND_SHUTDOWN) {
2305 err = -EPIPE;
2306 send_sigpipe = true;
2307 goto err_unlock;
2308 }
2309
2310 unix_state_lock(other);
2311
2312 if (sock_flag(other, SOCK_DEAD) ||
2313 other->sk_shutdown & RCV_SHUTDOWN) {
2314 err = -EPIPE;
2315 send_sigpipe = true;
2316 goto err_state_unlock;
2317 }
2318
2319 if (init_scm) {
2320 err = maybe_init_creds(&scm, socket, other);
2321 if (err)
2322 goto err_state_unlock;
2323 init_scm = false;
2324 }
2325
2326 skb = skb_peek_tail(&other->sk_receive_queue);
2327 if (tail && tail == skb) {
2328 skb = newskb;
2329 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2330 if (newskb) {
2331 skb = newskb;
2332 } else {
2333 tail = skb;
2334 goto alloc_skb;
2335 }
2336 } else if (newskb) {
2337 /* this is fast path, we don't necessarily need to
2338 * call to kfree_skb even though with newskb == NULL
2339 * this - does no harm
2340 */
2341 consume_skb(newskb);
2342 newskb = NULL;
2343 }
2344
2345 if (skb_append_pagefrags(skb, page, offset, size)) {
2346 tail = skb;
2347 goto alloc_skb;
2348 }
2349
2350 skb->len += size;
2351 skb->data_len += size;
2352 skb->truesize += size;
2353 refcount_add(size, &sk->sk_wmem_alloc);
2354
2355 if (newskb) {
2356 err = unix_scm_to_skb(&scm, skb, false);
2357 if (err)
2358 goto err_state_unlock;
2359 spin_lock(&other->sk_receive_queue.lock);
2360 __skb_queue_tail(&other->sk_receive_queue, newskb);
2361 spin_unlock(&other->sk_receive_queue.lock);
2362 }
2363
2364 unix_state_unlock(other);
2365 mutex_unlock(&unix_sk(other)->iolock);
2366
2367 other->sk_data_ready(other);
2368 scm_destroy(&scm);
2369 return size;
2370
2371err_state_unlock:
2372 unix_state_unlock(other);
2373err_unlock:
2374 mutex_unlock(&unix_sk(other)->iolock);
2375err:
2376 kfree_skb(newskb);
2377 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2378 send_sig(SIGPIPE, current, 0);
2379 if (!init_scm)
2380 scm_destroy(&scm);
2381 return err;
2382}
2383
2384static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2385 size_t len)
2386{
2387 int err;
2388 struct sock *sk = sock->sk;
2389
2390 err = sock_error(sk);
2391 if (err)
2392 return err;
2393
2394 if (sk->sk_state != TCP_ESTABLISHED)
2395 return -ENOTCONN;
2396
2397 if (msg->msg_namelen)
2398 msg->msg_namelen = 0;
2399
2400 return unix_dgram_sendmsg(sock, msg, len);
2401}
2402
2403static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2404 size_t size, int flags)
2405{
2406 struct sock *sk = sock->sk;
2407
2408 if (sk->sk_state != TCP_ESTABLISHED)
2409 return -ENOTCONN;
2410
2411 return unix_dgram_recvmsg(sock, msg, size, flags);
2412}
2413
2414static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2415{
2416 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2417
2418 if (addr) {
2419 msg->msg_namelen = addr->len;
2420 memcpy(msg->msg_name, addr->name, addr->len);
2421 }
2422}
2423
2424int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
2425 int flags)
2426{
2427 struct scm_cookie scm;
2428 struct socket *sock = sk->sk_socket;
2429 struct unix_sock *u = unix_sk(sk);
2430 struct sk_buff *skb, *last;
2431 long timeo;
2432 int skip;
2433 int err;
2434
2435 err = -EOPNOTSUPP;
2436 if (flags&MSG_OOB)
2437 goto out;
2438
2439 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2440
2441 do {
2442 mutex_lock(&u->iolock);
2443
2444 skip = sk_peek_offset(sk, flags);
2445 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2446 &skip, &err, &last);
2447 if (skb) {
2448 if (!(flags & MSG_PEEK))
2449 scm_stat_del(sk, skb);
2450 break;
2451 }
2452
2453 mutex_unlock(&u->iolock);
2454
2455 if (err != -EAGAIN)
2456 break;
2457 } while (timeo &&
2458 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2459 &err, &timeo, last));
2460
2461 if (!skb) { /* implies iolock unlocked */
2462 unix_state_lock(sk);
2463 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2464 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2465 (sk->sk_shutdown & RCV_SHUTDOWN))
2466 err = 0;
2467 unix_state_unlock(sk);
2468 goto out;
2469 }
2470
2471 if (wq_has_sleeper(&u->peer_wait))
2472 wake_up_interruptible_sync_poll(&u->peer_wait,
2473 EPOLLOUT | EPOLLWRNORM |
2474 EPOLLWRBAND);
2475
2476 if (msg->msg_name)
2477 unix_copy_addr(msg, skb->sk);
2478
2479 if (size > skb->len - skip)
2480 size = skb->len - skip;
2481 else if (size < skb->len - skip)
2482 msg->msg_flags |= MSG_TRUNC;
2483
2484 err = skb_copy_datagram_msg(skb, skip, msg, size);
2485 if (err)
2486 goto out_free;
2487
2488 if (sock_flag(sk, SOCK_RCVTSTAMP))
2489 __sock_recv_timestamp(msg, sk, skb);
2490
2491 memset(&scm, 0, sizeof(scm));
2492
2493 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2494 unix_set_secdata(&scm, skb);
2495
2496 if (!(flags & MSG_PEEK)) {
2497 if (UNIXCB(skb).fp)
2498 unix_detach_fds(&scm, skb);
2499
2500 sk_peek_offset_bwd(sk, skb->len);
2501 } else {
2502 /* It is questionable: on PEEK we could:
2503 - do not return fds - good, but too simple 8)
2504 - return fds, and do not return them on read (old strategy,
2505 apparently wrong)
2506 - clone fds (I chose it for now, it is the most universal
2507 solution)
2508
2509 POSIX 1003.1g does not actually define this clearly
2510 at all. POSIX 1003.1g doesn't define a lot of things
2511 clearly however!
2512
2513 */
2514
2515 sk_peek_offset_fwd(sk, size);
2516
2517 if (UNIXCB(skb).fp)
2518 unix_peek_fds(&scm, skb);
2519 }
2520 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2521
2522 scm_recv(sock, msg, &scm, flags);
2523
2524out_free:
2525 skb_free_datagram(sk, skb);
2526 mutex_unlock(&u->iolock);
2527out:
2528 return err;
2529}
2530
2531static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2532 int flags)
2533{
2534 struct sock *sk = sock->sk;
2535
2536#ifdef CONFIG_BPF_SYSCALL
2537 const struct proto *prot = READ_ONCE(sk->sk_prot);
2538
2539 if (prot != &unix_dgram_proto)
2540 return prot->recvmsg(sk, msg, size, flags, NULL);
2541#endif
2542 return __unix_dgram_recvmsg(sk, msg, size, flags);
2543}
2544
2545static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2546{
2547 struct unix_sock *u = unix_sk(sk);
2548 struct sk_buff *skb;
2549 int err, copied;
2550
2551 mutex_lock(&u->iolock);
2552 skb = skb_recv_datagram(sk, MSG_DONTWAIT, &err);
2553 mutex_unlock(&u->iolock);
2554 if (!skb)
2555 return err;
2556
2557 copied = recv_actor(sk, skb);
2558 kfree_skb(skb);
2559
2560 return copied;
2561}
2562
2563/*
2564 * Sleep until more data has arrived. But check for races..
2565 */
2566static long unix_stream_data_wait(struct sock *sk, long timeo,
2567 struct sk_buff *last, unsigned int last_len,
2568 bool freezable)
2569{
2570 unsigned int state = TASK_INTERRUPTIBLE | freezable * TASK_FREEZABLE;
2571 struct sk_buff *tail;
2572 DEFINE_WAIT(wait);
2573
2574 unix_state_lock(sk);
2575
2576 for (;;) {
2577 prepare_to_wait(sk_sleep(sk), &wait, state);
2578
2579 tail = skb_peek_tail(&sk->sk_receive_queue);
2580 if (tail != last ||
2581 (tail && tail->len != last_len) ||
2582 sk->sk_err ||
2583 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2584 signal_pending(current) ||
2585 !timeo)
2586 break;
2587
2588 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2589 unix_state_unlock(sk);
2590 timeo = schedule_timeout(timeo);
2591 unix_state_lock(sk);
2592
2593 if (sock_flag(sk, SOCK_DEAD))
2594 break;
2595
2596 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2597 }
2598
2599 finish_wait(sk_sleep(sk), &wait);
2600 unix_state_unlock(sk);
2601 return timeo;
2602}
2603
2604static unsigned int unix_skb_len(const struct sk_buff *skb)
2605{
2606 return skb->len - UNIXCB(skb).consumed;
2607}
2608
2609struct unix_stream_read_state {
2610 int (*recv_actor)(struct sk_buff *, int, int,
2611 struct unix_stream_read_state *);
2612 struct socket *socket;
2613 struct msghdr *msg;
2614 struct pipe_inode_info *pipe;
2615 size_t size;
2616 int flags;
2617 unsigned int splice_flags;
2618};
2619
2620#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2621static int unix_stream_recv_urg(struct unix_stream_read_state *state)
2622{
2623 struct socket *sock = state->socket;
2624 struct sock *sk = sock->sk;
2625 struct unix_sock *u = unix_sk(sk);
2626 int chunk = 1;
2627 struct sk_buff *oob_skb;
2628
2629 mutex_lock(&u->iolock);
2630 unix_state_lock(sk);
2631
2632 if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb) {
2633 unix_state_unlock(sk);
2634 mutex_unlock(&u->iolock);
2635 return -EINVAL;
2636 }
2637
2638 oob_skb = u->oob_skb;
2639
2640 if (!(state->flags & MSG_PEEK))
2641 WRITE_ONCE(u->oob_skb, NULL);
2642
2643 unix_state_unlock(sk);
2644
2645 chunk = state->recv_actor(oob_skb, 0, chunk, state);
2646
2647 if (!(state->flags & MSG_PEEK)) {
2648 UNIXCB(oob_skb).consumed += 1;
2649 kfree_skb(oob_skb);
2650 }
2651
2652 mutex_unlock(&u->iolock);
2653
2654 if (chunk < 0)
2655 return -EFAULT;
2656
2657 state->msg->msg_flags |= MSG_OOB;
2658 return 1;
2659}
2660
2661static struct sk_buff *manage_oob(struct sk_buff *skb, struct sock *sk,
2662 int flags, int copied)
2663{
2664 struct unix_sock *u = unix_sk(sk);
2665
2666 if (!unix_skb_len(skb) && !(flags & MSG_PEEK)) {
2667 skb_unlink(skb, &sk->sk_receive_queue);
2668 consume_skb(skb);
2669 skb = NULL;
2670 } else {
2671 if (skb == u->oob_skb) {
2672 if (copied) {
2673 skb = NULL;
2674 } else if (sock_flag(sk, SOCK_URGINLINE)) {
2675 if (!(flags & MSG_PEEK)) {
2676 WRITE_ONCE(u->oob_skb, NULL);
2677 consume_skb(skb);
2678 }
2679 } else if (!(flags & MSG_PEEK)) {
2680 skb_unlink(skb, &sk->sk_receive_queue);
2681 consume_skb(skb);
2682 skb = skb_peek(&sk->sk_receive_queue);
2683 }
2684 }
2685 }
2686 return skb;
2687}
2688#endif
2689
2690static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2691{
2692 if (unlikely(sk->sk_state != TCP_ESTABLISHED))
2693 return -ENOTCONN;
2694
2695 return unix_read_skb(sk, recv_actor);
2696}
2697
2698static int unix_stream_read_generic(struct unix_stream_read_state *state,
2699 bool freezable)
2700{
2701 struct scm_cookie scm;
2702 struct socket *sock = state->socket;
2703 struct sock *sk = sock->sk;
2704 struct unix_sock *u = unix_sk(sk);
2705 int copied = 0;
2706 int flags = state->flags;
2707 int noblock = flags & MSG_DONTWAIT;
2708 bool check_creds = false;
2709 int target;
2710 int err = 0;
2711 long timeo;
2712 int skip;
2713 size_t size = state->size;
2714 unsigned int last_len;
2715
2716 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2717 err = -EINVAL;
2718 goto out;
2719 }
2720
2721 if (unlikely(flags & MSG_OOB)) {
2722 err = -EOPNOTSUPP;
2723#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2724 err = unix_stream_recv_urg(state);
2725#endif
2726 goto out;
2727 }
2728
2729 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2730 timeo = sock_rcvtimeo(sk, noblock);
2731
2732 memset(&scm, 0, sizeof(scm));
2733
2734 /* Lock the socket to prevent queue disordering
2735 * while sleeps in memcpy_tomsg
2736 */
2737 mutex_lock(&u->iolock);
2738
2739 skip = max(sk_peek_offset(sk, flags), 0);
2740
2741 do {
2742 int chunk;
2743 bool drop_skb;
2744 struct sk_buff *skb, *last;
2745
2746redo:
2747 unix_state_lock(sk);
2748 if (sock_flag(sk, SOCK_DEAD)) {
2749 err = -ECONNRESET;
2750 goto unlock;
2751 }
2752 last = skb = skb_peek(&sk->sk_receive_queue);
2753 last_len = last ? last->len : 0;
2754
2755#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2756 if (skb) {
2757 skb = manage_oob(skb, sk, flags, copied);
2758 if (!skb) {
2759 unix_state_unlock(sk);
2760 if (copied)
2761 break;
2762 goto redo;
2763 }
2764 }
2765#endif
2766again:
2767 if (skb == NULL) {
2768 if (copied >= target)
2769 goto unlock;
2770
2771 /*
2772 * POSIX 1003.1g mandates this order.
2773 */
2774
2775 err = sock_error(sk);
2776 if (err)
2777 goto unlock;
2778 if (sk->sk_shutdown & RCV_SHUTDOWN)
2779 goto unlock;
2780
2781 unix_state_unlock(sk);
2782 if (!timeo) {
2783 err = -EAGAIN;
2784 break;
2785 }
2786
2787 mutex_unlock(&u->iolock);
2788
2789 timeo = unix_stream_data_wait(sk, timeo, last,
2790 last_len, freezable);
2791
2792 if (signal_pending(current)) {
2793 err = sock_intr_errno(timeo);
2794 scm_destroy(&scm);
2795 goto out;
2796 }
2797
2798 mutex_lock(&u->iolock);
2799 goto redo;
2800unlock:
2801 unix_state_unlock(sk);
2802 break;
2803 }
2804
2805 while (skip >= unix_skb_len(skb)) {
2806 skip -= unix_skb_len(skb);
2807 last = skb;
2808 last_len = skb->len;
2809 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2810 if (!skb)
2811 goto again;
2812 }
2813
2814 unix_state_unlock(sk);
2815
2816 if (check_creds) {
2817 /* Never glue messages from different writers */
2818 if (!unix_skb_scm_eq(skb, &scm))
2819 break;
2820 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2821 /* Copy credentials */
2822 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2823 unix_set_secdata(&scm, skb);
2824 check_creds = true;
2825 }
2826
2827 /* Copy address just once */
2828 if (state->msg && state->msg->msg_name) {
2829 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2830 state->msg->msg_name);
2831 unix_copy_addr(state->msg, skb->sk);
2832 sunaddr = NULL;
2833 }
2834
2835 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2836 skb_get(skb);
2837 chunk = state->recv_actor(skb, skip, chunk, state);
2838 drop_skb = !unix_skb_len(skb);
2839 /* skb is only safe to use if !drop_skb */
2840 consume_skb(skb);
2841 if (chunk < 0) {
2842 if (copied == 0)
2843 copied = -EFAULT;
2844 break;
2845 }
2846 copied += chunk;
2847 size -= chunk;
2848
2849 if (drop_skb) {
2850 /* the skb was touched by a concurrent reader;
2851 * we should not expect anything from this skb
2852 * anymore and assume it invalid - we can be
2853 * sure it was dropped from the socket queue
2854 *
2855 * let's report a short read
2856 */
2857 err = 0;
2858 break;
2859 }
2860
2861 /* Mark read part of skb as used */
2862 if (!(flags & MSG_PEEK)) {
2863 UNIXCB(skb).consumed += chunk;
2864
2865 sk_peek_offset_bwd(sk, chunk);
2866
2867 if (UNIXCB(skb).fp) {
2868 scm_stat_del(sk, skb);
2869 unix_detach_fds(&scm, skb);
2870 }
2871
2872 if (unix_skb_len(skb))
2873 break;
2874
2875 skb_unlink(skb, &sk->sk_receive_queue);
2876 consume_skb(skb);
2877
2878 if (scm.fp)
2879 break;
2880 } else {
2881 /* It is questionable, see note in unix_dgram_recvmsg.
2882 */
2883 if (UNIXCB(skb).fp)
2884 unix_peek_fds(&scm, skb);
2885
2886 sk_peek_offset_fwd(sk, chunk);
2887
2888 if (UNIXCB(skb).fp)
2889 break;
2890
2891 skip = 0;
2892 last = skb;
2893 last_len = skb->len;
2894 unix_state_lock(sk);
2895 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2896 if (skb)
2897 goto again;
2898 unix_state_unlock(sk);
2899 break;
2900 }
2901 } while (size);
2902
2903 mutex_unlock(&u->iolock);
2904 if (state->msg)
2905 scm_recv(sock, state->msg, &scm, flags);
2906 else
2907 scm_destroy(&scm);
2908out:
2909 return copied ? : err;
2910}
2911
2912static int unix_stream_read_actor(struct sk_buff *skb,
2913 int skip, int chunk,
2914 struct unix_stream_read_state *state)
2915{
2916 int ret;
2917
2918 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2919 state->msg, chunk);
2920 return ret ?: chunk;
2921}
2922
2923int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2924 size_t size, int flags)
2925{
2926 struct unix_stream_read_state state = {
2927 .recv_actor = unix_stream_read_actor,
2928 .socket = sk->sk_socket,
2929 .msg = msg,
2930 .size = size,
2931 .flags = flags
2932 };
2933
2934 return unix_stream_read_generic(&state, true);
2935}
2936
2937static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2938 size_t size, int flags)
2939{
2940 struct unix_stream_read_state state = {
2941 .recv_actor = unix_stream_read_actor,
2942 .socket = sock,
2943 .msg = msg,
2944 .size = size,
2945 .flags = flags
2946 };
2947
2948#ifdef CONFIG_BPF_SYSCALL
2949 struct sock *sk = sock->sk;
2950 const struct proto *prot = READ_ONCE(sk->sk_prot);
2951
2952 if (prot != &unix_stream_proto)
2953 return prot->recvmsg(sk, msg, size, flags, NULL);
2954#endif
2955 return unix_stream_read_generic(&state, true);
2956}
2957
2958static int unix_stream_splice_actor(struct sk_buff *skb,
2959 int skip, int chunk,
2960 struct unix_stream_read_state *state)
2961{
2962 return skb_splice_bits(skb, state->socket->sk,
2963 UNIXCB(skb).consumed + skip,
2964 state->pipe, chunk, state->splice_flags);
2965}
2966
2967static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2968 struct pipe_inode_info *pipe,
2969 size_t size, unsigned int flags)
2970{
2971 struct unix_stream_read_state state = {
2972 .recv_actor = unix_stream_splice_actor,
2973 .socket = sock,
2974 .pipe = pipe,
2975 .size = size,
2976 .splice_flags = flags,
2977 };
2978
2979 if (unlikely(*ppos))
2980 return -ESPIPE;
2981
2982 if (sock->file->f_flags & O_NONBLOCK ||
2983 flags & SPLICE_F_NONBLOCK)
2984 state.flags = MSG_DONTWAIT;
2985
2986 return unix_stream_read_generic(&state, false);
2987}
2988
2989static int unix_shutdown(struct socket *sock, int mode)
2990{
2991 struct sock *sk = sock->sk;
2992 struct sock *other;
2993
2994 if (mode < SHUT_RD || mode > SHUT_RDWR)
2995 return -EINVAL;
2996 /* This maps:
2997 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2998 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2999 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
3000 */
3001 ++mode;
3002
3003 unix_state_lock(sk);
3004 sk->sk_shutdown |= mode;
3005 other = unix_peer(sk);
3006 if (other)
3007 sock_hold(other);
3008 unix_state_unlock(sk);
3009 sk->sk_state_change(sk);
3010
3011 if (other &&
3012 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
3013
3014 int peer_mode = 0;
3015 const struct proto *prot = READ_ONCE(other->sk_prot);
3016
3017 if (prot->unhash)
3018 prot->unhash(other);
3019 if (mode&RCV_SHUTDOWN)
3020 peer_mode |= SEND_SHUTDOWN;
3021 if (mode&SEND_SHUTDOWN)
3022 peer_mode |= RCV_SHUTDOWN;
3023 unix_state_lock(other);
3024 other->sk_shutdown |= peer_mode;
3025 unix_state_unlock(other);
3026 other->sk_state_change(other);
3027 if (peer_mode == SHUTDOWN_MASK)
3028 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
3029 else if (peer_mode & RCV_SHUTDOWN)
3030 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
3031 }
3032 if (other)
3033 sock_put(other);
3034
3035 return 0;
3036}
3037
3038long unix_inq_len(struct sock *sk)
3039{
3040 struct sk_buff *skb;
3041 long amount = 0;
3042
3043 if (sk->sk_state == TCP_LISTEN)
3044 return -EINVAL;
3045
3046 spin_lock(&sk->sk_receive_queue.lock);
3047 if (sk->sk_type == SOCK_STREAM ||
3048 sk->sk_type == SOCK_SEQPACKET) {
3049 skb_queue_walk(&sk->sk_receive_queue, skb)
3050 amount += unix_skb_len(skb);
3051 } else {
3052 skb = skb_peek(&sk->sk_receive_queue);
3053 if (skb)
3054 amount = skb->len;
3055 }
3056 spin_unlock(&sk->sk_receive_queue.lock);
3057
3058 return amount;
3059}
3060EXPORT_SYMBOL_GPL(unix_inq_len);
3061
3062long unix_outq_len(struct sock *sk)
3063{
3064 return sk_wmem_alloc_get(sk);
3065}
3066EXPORT_SYMBOL_GPL(unix_outq_len);
3067
3068static int unix_open_file(struct sock *sk)
3069{
3070 struct path path;
3071 struct file *f;
3072 int fd;
3073
3074 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3075 return -EPERM;
3076
3077 if (!smp_load_acquire(&unix_sk(sk)->addr))
3078 return -ENOENT;
3079
3080 path = unix_sk(sk)->path;
3081 if (!path.dentry)
3082 return -ENOENT;
3083
3084 path_get(&path);
3085
3086 fd = get_unused_fd_flags(O_CLOEXEC);
3087 if (fd < 0)
3088 goto out;
3089
3090 f = dentry_open(&path, O_PATH, current_cred());
3091 if (IS_ERR(f)) {
3092 put_unused_fd(fd);
3093 fd = PTR_ERR(f);
3094 goto out;
3095 }
3096
3097 fd_install(fd, f);
3098out:
3099 path_put(&path);
3100
3101 return fd;
3102}
3103
3104static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3105{
3106 struct sock *sk = sock->sk;
3107 long amount = 0;
3108 int err;
3109
3110 switch (cmd) {
3111 case SIOCOUTQ:
3112 amount = unix_outq_len(sk);
3113 err = put_user(amount, (int __user *)arg);
3114 break;
3115 case SIOCINQ:
3116 amount = unix_inq_len(sk);
3117 if (amount < 0)
3118 err = amount;
3119 else
3120 err = put_user(amount, (int __user *)arg);
3121 break;
3122 case SIOCUNIXFILE:
3123 err = unix_open_file(sk);
3124 break;
3125#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3126 case SIOCATMARK:
3127 {
3128 struct sk_buff *skb;
3129 int answ = 0;
3130
3131 skb = skb_peek(&sk->sk_receive_queue);
3132 if (skb && skb == READ_ONCE(unix_sk(sk)->oob_skb))
3133 answ = 1;
3134 err = put_user(answ, (int __user *)arg);
3135 }
3136 break;
3137#endif
3138 default:
3139 err = -ENOIOCTLCMD;
3140 break;
3141 }
3142 return err;
3143}
3144
3145#ifdef CONFIG_COMPAT
3146static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3147{
3148 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
3149}
3150#endif
3151
3152static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
3153{
3154 struct sock *sk = sock->sk;
3155 __poll_t mask;
3156
3157 sock_poll_wait(file, sock, wait);
3158 mask = 0;
3159
3160 /* exceptional events? */
3161 if (sk->sk_err)
3162 mask |= EPOLLERR;
3163 if (sk->sk_shutdown == SHUTDOWN_MASK)
3164 mask |= EPOLLHUP;
3165 if (sk->sk_shutdown & RCV_SHUTDOWN)
3166 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3167
3168 /* readable? */
3169 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3170 mask |= EPOLLIN | EPOLLRDNORM;
3171 if (sk_is_readable(sk))
3172 mask |= EPOLLIN | EPOLLRDNORM;
3173#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3174 if (READ_ONCE(unix_sk(sk)->oob_skb))
3175 mask |= EPOLLPRI;
3176#endif
3177
3178 /* Connection-based need to check for termination and startup */
3179 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3180 sk->sk_state == TCP_CLOSE)
3181 mask |= EPOLLHUP;
3182
3183 /*
3184 * we set writable also when the other side has shut down the
3185 * connection. This prevents stuck sockets.
3186 */
3187 if (unix_writable(sk))
3188 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3189
3190 return mask;
3191}
3192
3193static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3194 poll_table *wait)
3195{
3196 struct sock *sk = sock->sk, *other;
3197 unsigned int writable;
3198 __poll_t mask;
3199
3200 sock_poll_wait(file, sock, wait);
3201 mask = 0;
3202
3203 /* exceptional events? */
3204 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
3205 mask |= EPOLLERR |
3206 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
3207
3208 if (sk->sk_shutdown & RCV_SHUTDOWN)
3209 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3210 if (sk->sk_shutdown == SHUTDOWN_MASK)
3211 mask |= EPOLLHUP;
3212
3213 /* readable? */
3214 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3215 mask |= EPOLLIN | EPOLLRDNORM;
3216 if (sk_is_readable(sk))
3217 mask |= EPOLLIN | EPOLLRDNORM;
3218
3219 /* Connection-based need to check for termination and startup */
3220 if (sk->sk_type == SOCK_SEQPACKET) {
3221 if (sk->sk_state == TCP_CLOSE)
3222 mask |= EPOLLHUP;
3223 /* connection hasn't started yet? */
3224 if (sk->sk_state == TCP_SYN_SENT)
3225 return mask;
3226 }
3227
3228 /* No write status requested, avoid expensive OUT tests. */
3229 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
3230 return mask;
3231
3232 writable = unix_writable(sk);
3233 if (writable) {
3234 unix_state_lock(sk);
3235
3236 other = unix_peer(sk);
3237 if (other && unix_peer(other) != sk &&
3238 unix_recvq_full_lockless(other) &&
3239 unix_dgram_peer_wake_me(sk, other))
3240 writable = 0;
3241
3242 unix_state_unlock(sk);
3243 }
3244
3245 if (writable)
3246 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3247 else
3248 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3249
3250 return mask;
3251}
3252
3253#ifdef CONFIG_PROC_FS
3254
3255#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3256
3257#define get_bucket(x) ((x) >> BUCKET_SPACE)
3258#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
3259#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
3260
3261static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
3262{
3263 unsigned long offset = get_offset(*pos);
3264 unsigned long bucket = get_bucket(*pos);
3265 unsigned long count = 0;
3266 struct sock *sk;
3267
3268 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3269 sk; sk = sk_next(sk)) {
3270 if (++count == offset)
3271 break;
3272 }
3273
3274 return sk;
3275}
3276
3277static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
3278{
3279 unsigned long bucket = get_bucket(*pos);
3280 struct net *net = seq_file_net(seq);
3281 struct sock *sk;
3282
3283 while (bucket < UNIX_HASH_SIZE) {
3284 spin_lock(&net->unx.table.locks[bucket]);
3285
3286 sk = unix_from_bucket(seq, pos);
3287 if (sk)
3288 return sk;
3289
3290 spin_unlock(&net->unx.table.locks[bucket]);
3291
3292 *pos = set_bucket_offset(++bucket, 1);
3293 }
3294
3295 return NULL;
3296}
3297
3298static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
3299 loff_t *pos)
3300{
3301 unsigned long bucket = get_bucket(*pos);
3302
3303 sk = sk_next(sk);
3304 if (sk)
3305 return sk;
3306
3307
3308 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
3309
3310 *pos = set_bucket_offset(++bucket, 1);
3311
3312 return unix_get_first(seq, pos);
3313}
3314
3315static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3316{
3317 if (!*pos)
3318 return SEQ_START_TOKEN;
3319
3320 return unix_get_first(seq, pos);
3321}
3322
3323static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3324{
3325 ++*pos;
3326
3327 if (v == SEQ_START_TOKEN)
3328 return unix_get_first(seq, pos);
3329
3330 return unix_get_next(seq, v, pos);
3331}
3332
3333static void unix_seq_stop(struct seq_file *seq, void *v)
3334{
3335 struct sock *sk = v;
3336
3337 if (sk)
3338 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
3339}
3340
3341static int unix_seq_show(struct seq_file *seq, void *v)
3342{
3343
3344 if (v == SEQ_START_TOKEN)
3345 seq_puts(seq, "Num RefCount Protocol Flags Type St "
3346 "Inode Path\n");
3347 else {
3348 struct sock *s = v;
3349 struct unix_sock *u = unix_sk(s);
3350 unix_state_lock(s);
3351
3352 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
3353 s,
3354 refcount_read(&s->sk_refcnt),
3355 0,
3356 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
3357 s->sk_type,
3358 s->sk_socket ?
3359 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
3360 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
3361 sock_i_ino(s));
3362
3363 if (u->addr) { // under a hash table lock here
3364 int i, len;
3365 seq_putc(seq, ' ');
3366
3367 i = 0;
3368 len = u->addr->len -
3369 offsetof(struct sockaddr_un, sun_path);
3370 if (u->addr->name->sun_path[0]) {
3371 len--;
3372 } else {
3373 seq_putc(seq, '@');
3374 i++;
3375 }
3376 for ( ; i < len; i++)
3377 seq_putc(seq, u->addr->name->sun_path[i] ?:
3378 '@');
3379 }
3380 unix_state_unlock(s);
3381 seq_putc(seq, '\n');
3382 }
3383
3384 return 0;
3385}
3386
3387static const struct seq_operations unix_seq_ops = {
3388 .start = unix_seq_start,
3389 .next = unix_seq_next,
3390 .stop = unix_seq_stop,
3391 .show = unix_seq_show,
3392};
3393
3394#if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL)
3395struct bpf_unix_iter_state {
3396 struct seq_net_private p;
3397 unsigned int cur_sk;
3398 unsigned int end_sk;
3399 unsigned int max_sk;
3400 struct sock **batch;
3401 bool st_bucket_done;
3402};
3403
3404struct bpf_iter__unix {
3405 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3406 __bpf_md_ptr(struct unix_sock *, unix_sk);
3407 uid_t uid __aligned(8);
3408};
3409
3410static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3411 struct unix_sock *unix_sk, uid_t uid)
3412{
3413 struct bpf_iter__unix ctx;
3414
3415 meta->seq_num--; /* skip SEQ_START_TOKEN */
3416 ctx.meta = meta;
3417 ctx.unix_sk = unix_sk;
3418 ctx.uid = uid;
3419 return bpf_iter_run_prog(prog, &ctx);
3420}
3421
3422static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
3423
3424{
3425 struct bpf_unix_iter_state *iter = seq->private;
3426 unsigned int expected = 1;
3427 struct sock *sk;
3428
3429 sock_hold(start_sk);
3430 iter->batch[iter->end_sk++] = start_sk;
3431
3432 for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
3433 if (iter->end_sk < iter->max_sk) {
3434 sock_hold(sk);
3435 iter->batch[iter->end_sk++] = sk;
3436 }
3437
3438 expected++;
3439 }
3440
3441 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
3442
3443 return expected;
3444}
3445
3446static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
3447{
3448 while (iter->cur_sk < iter->end_sk)
3449 sock_put(iter->batch[iter->cur_sk++]);
3450}
3451
3452static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
3453 unsigned int new_batch_sz)
3454{
3455 struct sock **new_batch;
3456
3457 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3458 GFP_USER | __GFP_NOWARN);
3459 if (!new_batch)
3460 return -ENOMEM;
3461
3462 bpf_iter_unix_put_batch(iter);
3463 kvfree(iter->batch);
3464 iter->batch = new_batch;
3465 iter->max_sk = new_batch_sz;
3466
3467 return 0;
3468}
3469
3470static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
3471 loff_t *pos)
3472{
3473 struct bpf_unix_iter_state *iter = seq->private;
3474 unsigned int expected;
3475 bool resized = false;
3476 struct sock *sk;
3477
3478 if (iter->st_bucket_done)
3479 *pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
3480
3481again:
3482 /* Get a new batch */
3483 iter->cur_sk = 0;
3484 iter->end_sk = 0;
3485
3486 sk = unix_get_first(seq, pos);
3487 if (!sk)
3488 return NULL; /* Done */
3489
3490 expected = bpf_iter_unix_hold_batch(seq, sk);
3491
3492 if (iter->end_sk == expected) {
3493 iter->st_bucket_done = true;
3494 return sk;
3495 }
3496
3497 if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
3498 resized = true;
3499 goto again;
3500 }
3501
3502 return sk;
3503}
3504
3505static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
3506{
3507 if (!*pos)
3508 return SEQ_START_TOKEN;
3509
3510 /* bpf iter does not support lseek, so it always
3511 * continue from where it was stop()-ped.
3512 */
3513 return bpf_iter_unix_batch(seq, pos);
3514}
3515
3516static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3517{
3518 struct bpf_unix_iter_state *iter = seq->private;
3519 struct sock *sk;
3520
3521 /* Whenever seq_next() is called, the iter->cur_sk is
3522 * done with seq_show(), so advance to the next sk in
3523 * the batch.
3524 */
3525 if (iter->cur_sk < iter->end_sk)
3526 sock_put(iter->batch[iter->cur_sk++]);
3527
3528 ++*pos;
3529
3530 if (iter->cur_sk < iter->end_sk)
3531 sk = iter->batch[iter->cur_sk];
3532 else
3533 sk = bpf_iter_unix_batch(seq, pos);
3534
3535 return sk;
3536}
3537
3538static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
3539{
3540 struct bpf_iter_meta meta;
3541 struct bpf_prog *prog;
3542 struct sock *sk = v;
3543 uid_t uid;
3544 bool slow;
3545 int ret;
3546
3547 if (v == SEQ_START_TOKEN)
3548 return 0;
3549
3550 slow = lock_sock_fast(sk);
3551
3552 if (unlikely(sk_unhashed(sk))) {
3553 ret = SEQ_SKIP;
3554 goto unlock;
3555 }
3556
3557 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3558 meta.seq = seq;
3559 prog = bpf_iter_get_info(&meta, false);
3560 ret = unix_prog_seq_show(prog, &meta, v, uid);
3561unlock:
3562 unlock_sock_fast(sk, slow);
3563 return ret;
3564}
3565
3566static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3567{
3568 struct bpf_unix_iter_state *iter = seq->private;
3569 struct bpf_iter_meta meta;
3570 struct bpf_prog *prog;
3571
3572 if (!v) {
3573 meta.seq = seq;
3574 prog = bpf_iter_get_info(&meta, true);
3575 if (prog)
3576 (void)unix_prog_seq_show(prog, &meta, v, 0);
3577 }
3578
3579 if (iter->cur_sk < iter->end_sk)
3580 bpf_iter_unix_put_batch(iter);
3581}
3582
3583static const struct seq_operations bpf_iter_unix_seq_ops = {
3584 .start = bpf_iter_unix_seq_start,
3585 .next = bpf_iter_unix_seq_next,
3586 .stop = bpf_iter_unix_seq_stop,
3587 .show = bpf_iter_unix_seq_show,
3588};
3589#endif
3590#endif
3591
3592static const struct net_proto_family unix_family_ops = {
3593 .family = PF_UNIX,
3594 .create = unix_create,
3595 .owner = THIS_MODULE,
3596};
3597
3598
3599static int __net_init unix_net_init(struct net *net)
3600{
3601 int i;
3602
3603 net->unx.sysctl_max_dgram_qlen = 10;
3604 if (unix_sysctl_register(net))
3605 goto out;
3606
3607#ifdef CONFIG_PROC_FS
3608 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
3609 sizeof(struct seq_net_private)))
3610 goto err_sysctl;
3611#endif
3612
3613 net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
3614 sizeof(spinlock_t), GFP_KERNEL);
3615 if (!net->unx.table.locks)
3616 goto err_proc;
3617
3618 net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
3619 sizeof(struct hlist_head),
3620 GFP_KERNEL);
3621 if (!net->unx.table.buckets)
3622 goto free_locks;
3623
3624 for (i = 0; i < UNIX_HASH_SIZE; i++) {
3625 spin_lock_init(&net->unx.table.locks[i]);
3626 INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
3627 }
3628
3629 return 0;
3630
3631free_locks:
3632 kvfree(net->unx.table.locks);
3633err_proc:
3634#ifdef CONFIG_PROC_FS
3635 remove_proc_entry("unix", net->proc_net);
3636err_sysctl:
3637#endif
3638 unix_sysctl_unregister(net);
3639out:
3640 return -ENOMEM;
3641}
3642
3643static void __net_exit unix_net_exit(struct net *net)
3644{
3645 kvfree(net->unx.table.buckets);
3646 kvfree(net->unx.table.locks);
3647 unix_sysctl_unregister(net);
3648 remove_proc_entry("unix", net->proc_net);
3649}
3650
3651static struct pernet_operations unix_net_ops = {
3652 .init = unix_net_init,
3653 .exit = unix_net_exit,
3654};
3655
3656#if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3657DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
3658 struct unix_sock *unix_sk, uid_t uid)
3659
3660#define INIT_BATCH_SZ 16
3661
3662static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
3663{
3664 struct bpf_unix_iter_state *iter = priv_data;
3665 int err;
3666
3667 err = bpf_iter_init_seq_net(priv_data, aux);
3668 if (err)
3669 return err;
3670
3671 err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
3672 if (err) {
3673 bpf_iter_fini_seq_net(priv_data);
3674 return err;
3675 }
3676
3677 return 0;
3678}
3679
3680static void bpf_iter_fini_unix(void *priv_data)
3681{
3682 struct bpf_unix_iter_state *iter = priv_data;
3683
3684 bpf_iter_fini_seq_net(priv_data);
3685 kvfree(iter->batch);
3686}
3687
3688static const struct bpf_iter_seq_info unix_seq_info = {
3689 .seq_ops = &bpf_iter_unix_seq_ops,
3690 .init_seq_private = bpf_iter_init_unix,
3691 .fini_seq_private = bpf_iter_fini_unix,
3692 .seq_priv_size = sizeof(struct bpf_unix_iter_state),
3693};
3694
3695static const struct bpf_func_proto *
3696bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
3697 const struct bpf_prog *prog)
3698{
3699 switch (func_id) {
3700 case BPF_FUNC_setsockopt:
3701 return &bpf_sk_setsockopt_proto;
3702 case BPF_FUNC_getsockopt:
3703 return &bpf_sk_getsockopt_proto;
3704 default:
3705 return NULL;
3706 }
3707}
3708
3709static struct bpf_iter_reg unix_reg_info = {
3710 .target = "unix",
3711 .ctx_arg_info_size = 1,
3712 .ctx_arg_info = {
3713 { offsetof(struct bpf_iter__unix, unix_sk),
3714 PTR_TO_BTF_ID_OR_NULL },
3715 },
3716 .get_func_proto = bpf_iter_unix_get_func_proto,
3717 .seq_info = &unix_seq_info,
3718};
3719
3720static void __init bpf_iter_register(void)
3721{
3722 unix_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UNIX];
3723 if (bpf_iter_reg_target(&unix_reg_info))
3724 pr_warn("Warning: could not register bpf iterator unix\n");
3725}
3726#endif
3727
3728static int __init af_unix_init(void)
3729{
3730 int i, rc = -1;
3731
3732 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
3733
3734 for (i = 0; i < UNIX_HASH_SIZE / 2; i++) {
3735 spin_lock_init(&bsd_socket_locks[i]);
3736 INIT_HLIST_HEAD(&bsd_socket_buckets[i]);
3737 }
3738
3739 rc = proto_register(&unix_dgram_proto, 1);
3740 if (rc != 0) {
3741 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3742 goto out;
3743 }
3744
3745 rc = proto_register(&unix_stream_proto, 1);
3746 if (rc != 0) {
3747 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3748 proto_unregister(&unix_dgram_proto);
3749 goto out;
3750 }
3751
3752 sock_register(&unix_family_ops);
3753 register_pernet_subsys(&unix_net_ops);
3754 unix_bpf_build_proto();
3755
3756#if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3757 bpf_iter_register();
3758#endif
3759
3760out:
3761 return rc;
3762}
3763
3764static void __exit af_unix_exit(void)
3765{
3766 sock_unregister(PF_UNIX);
3767 proto_unregister(&unix_dgram_proto);
3768 proto_unregister(&unix_stream_proto);
3769 unregister_pernet_subsys(&unix_net_ops);
3770}
3771
3772/* Earlier than device_initcall() so that other drivers invoking
3773 request_module() don't end up in a loop when modprobe tries
3774 to use a UNIX socket. But later than subsys_initcall() because
3775 we depend on stuff initialised there */
3776fs_initcall(af_unix_init);
3777module_exit(af_unix_exit);
3778
3779MODULE_LICENSE("GPL");
3780MODULE_ALIAS_NETPROTO(PF_UNIX);