Loading...
1/*
2 * NET4: Implementation of BSD Unix domain sockets.
3 *
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * Fixes:
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
21 * Mike Shaver's work.
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
28 * reference counting
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
31 * Lots of bug fixes.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
43 * dgram receiver.
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
51 *
52 *
53 * Known differences from reference BSD that was tested:
54 *
55 * [TO FIX]
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
60 * [NOT TO FIX]
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
68 *
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
73 *
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
80 * with BSD names.
81 */
82
83#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84
85#include <linux/module.h>
86#include <linux/kernel.h>
87#include <linux/signal.h>
88#include <linux/sched.h>
89#include <linux/errno.h>
90#include <linux/string.h>
91#include <linux/stat.h>
92#include <linux/dcache.h>
93#include <linux/namei.h>
94#include <linux/socket.h>
95#include <linux/un.h>
96#include <linux/fcntl.h>
97#include <linux/termios.h>
98#include <linux/sockios.h>
99#include <linux/net.h>
100#include <linux/in.h>
101#include <linux/fs.h>
102#include <linux/slab.h>
103#include <linux/uaccess.h>
104#include <linux/skbuff.h>
105#include <linux/netdevice.h>
106#include <net/net_namespace.h>
107#include <net/sock.h>
108#include <net/tcp_states.h>
109#include <net/af_unix.h>
110#include <linux/proc_fs.h>
111#include <linux/seq_file.h>
112#include <net/scm.h>
113#include <linux/init.h>
114#include <linux/poll.h>
115#include <linux/rtnetlink.h>
116#include <linux/mount.h>
117#include <net/checksum.h>
118#include <linux/security.h>
119#include <linux/freezer.h>
120
121struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
122EXPORT_SYMBOL_GPL(unix_socket_table);
123DEFINE_SPINLOCK(unix_table_lock);
124EXPORT_SYMBOL_GPL(unix_table_lock);
125static atomic_long_t unix_nr_socks;
126
127
128static struct hlist_head *unix_sockets_unbound(void *addr)
129{
130 unsigned long hash = (unsigned long)addr;
131
132 hash ^= hash >> 16;
133 hash ^= hash >> 8;
134 hash %= UNIX_HASH_SIZE;
135 return &unix_socket_table[UNIX_HASH_SIZE + hash];
136}
137
138#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
139
140#ifdef CONFIG_SECURITY_NETWORK
141static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142{
143 UNIXCB(skb).secid = scm->secid;
144}
145
146static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
147{
148 scm->secid = UNIXCB(skb).secid;
149}
150
151static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
152{
153 return (scm->secid == UNIXCB(skb).secid);
154}
155#else
156static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
157{ }
158
159static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
160{ }
161
162static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
163{
164 return true;
165}
166#endif /* CONFIG_SECURITY_NETWORK */
167
168/*
169 * SMP locking strategy:
170 * hash table is protected with spinlock unix_table_lock
171 * each socket state is protected by separate spin lock.
172 */
173
174static inline unsigned int unix_hash_fold(__wsum n)
175{
176 unsigned int hash = (__force unsigned int)csum_fold(n);
177
178 hash ^= hash>>8;
179 return hash&(UNIX_HASH_SIZE-1);
180}
181
182#define unix_peer(sk) (unix_sk(sk)->peer)
183
184static inline int unix_our_peer(struct sock *sk, struct sock *osk)
185{
186 return unix_peer(osk) == sk;
187}
188
189static inline int unix_may_send(struct sock *sk, struct sock *osk)
190{
191 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
192}
193
194static inline int unix_recvq_full(struct sock const *sk)
195{
196 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
197}
198
199struct sock *unix_peer_get(struct sock *s)
200{
201 struct sock *peer;
202
203 unix_state_lock(s);
204 peer = unix_peer(s);
205 if (peer)
206 sock_hold(peer);
207 unix_state_unlock(s);
208 return peer;
209}
210EXPORT_SYMBOL_GPL(unix_peer_get);
211
212static inline void unix_release_addr(struct unix_address *addr)
213{
214 if (atomic_dec_and_test(&addr->refcnt))
215 kfree(addr);
216}
217
218/*
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
223 */
224
225static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
226{
227 if (len <= sizeof(short) || len > sizeof(*sunaddr))
228 return -EINVAL;
229 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
230 return -EINVAL;
231 if (sunaddr->sun_path[0]) {
232 /*
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesn't as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
238 */
239 ((char *)sunaddr)[len] = 0;
240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
241 return len;
242 }
243
244 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
245 return len;
246}
247
248static void __unix_remove_socket(struct sock *sk)
249{
250 sk_del_node_init(sk);
251}
252
253static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
254{
255 WARN_ON(!sk_unhashed(sk));
256 sk_add_node(sk, list);
257}
258
259static inline void unix_remove_socket(struct sock *sk)
260{
261 spin_lock(&unix_table_lock);
262 __unix_remove_socket(sk);
263 spin_unlock(&unix_table_lock);
264}
265
266static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
267{
268 spin_lock(&unix_table_lock);
269 __unix_insert_socket(list, sk);
270 spin_unlock(&unix_table_lock);
271}
272
273static struct sock *__unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
275 int len, int type, unsigned int hash)
276{
277 struct sock *s;
278
279 sk_for_each(s, &unix_socket_table[hash ^ type]) {
280 struct unix_sock *u = unix_sk(s);
281
282 if (!net_eq(sock_net(s), net))
283 continue;
284
285 if (u->addr->len == len &&
286 !memcmp(u->addr->name, sunname, len))
287 goto found;
288 }
289 s = NULL;
290found:
291 return s;
292}
293
294static inline struct sock *unix_find_socket_byname(struct net *net,
295 struct sockaddr_un *sunname,
296 int len, int type,
297 unsigned int hash)
298{
299 struct sock *s;
300
301 spin_lock(&unix_table_lock);
302 s = __unix_find_socket_byname(net, sunname, len, type, hash);
303 if (s)
304 sock_hold(s);
305 spin_unlock(&unix_table_lock);
306 return s;
307}
308
309static struct sock *unix_find_socket_byinode(struct inode *i)
310{
311 struct sock *s;
312
313 spin_lock(&unix_table_lock);
314 sk_for_each(s,
315 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
316 struct dentry *dentry = unix_sk(s)->path.dentry;
317
318 if (dentry && d_backing_inode(dentry) == i) {
319 sock_hold(s);
320 goto found;
321 }
322 }
323 s = NULL;
324found:
325 spin_unlock(&unix_table_lock);
326 return s;
327}
328
329/* Support code for asymmetrically connected dgram sockets
330 *
331 * If a datagram socket is connected to a socket not itself connected
332 * to the first socket (eg, /dev/log), clients may only enqueue more
333 * messages if the present receive queue of the server socket is not
334 * "too large". This means there's a second writeability condition
335 * poll and sendmsg need to test. The dgram recv code will do a wake
336 * up on the peer_wait wait queue of a socket upon reception of a
337 * datagram which needs to be propagated to sleeping would-be writers
338 * since these might not have sent anything so far. This can't be
339 * accomplished via poll_wait because the lifetime of the server
340 * socket might be less than that of its clients if these break their
341 * association with it or if the server socket is closed while clients
342 * are still connected to it and there's no way to inform "a polling
343 * implementation" that it should let go of a certain wait queue
344 *
345 * In order to propagate a wake up, a wait_queue_t of the client
346 * socket is enqueued on the peer_wait queue of the server socket
347 * whose wake function does a wake_up on the ordinary client socket
348 * wait queue. This connection is established whenever a write (or
349 * poll for write) hit the flow control condition and broken when the
350 * association to the server socket is dissolved or after a wake up
351 * was relayed.
352 */
353
354static int unix_dgram_peer_wake_relay(wait_queue_t *q, unsigned mode, int flags,
355 void *key)
356{
357 struct unix_sock *u;
358 wait_queue_head_t *u_sleep;
359
360 u = container_of(q, struct unix_sock, peer_wake);
361
362 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
363 q);
364 u->peer_wake.private = NULL;
365
366 /* relaying can only happen while the wq still exists */
367 u_sleep = sk_sleep(&u->sk);
368 if (u_sleep)
369 wake_up_interruptible_poll(u_sleep, key);
370
371 return 0;
372}
373
374static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
375{
376 struct unix_sock *u, *u_other;
377 int rc;
378
379 u = unix_sk(sk);
380 u_other = unix_sk(other);
381 rc = 0;
382 spin_lock(&u_other->peer_wait.lock);
383
384 if (!u->peer_wake.private) {
385 u->peer_wake.private = other;
386 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
387
388 rc = 1;
389 }
390
391 spin_unlock(&u_other->peer_wait.lock);
392 return rc;
393}
394
395static void unix_dgram_peer_wake_disconnect(struct sock *sk,
396 struct sock *other)
397{
398 struct unix_sock *u, *u_other;
399
400 u = unix_sk(sk);
401 u_other = unix_sk(other);
402 spin_lock(&u_other->peer_wait.lock);
403
404 if (u->peer_wake.private == other) {
405 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
406 u->peer_wake.private = NULL;
407 }
408
409 spin_unlock(&u_other->peer_wait.lock);
410}
411
412static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
413 struct sock *other)
414{
415 unix_dgram_peer_wake_disconnect(sk, other);
416 wake_up_interruptible_poll(sk_sleep(sk),
417 POLLOUT |
418 POLLWRNORM |
419 POLLWRBAND);
420}
421
422/* preconditions:
423 * - unix_peer(sk) == other
424 * - association is stable
425 */
426static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
427{
428 int connected;
429
430 connected = unix_dgram_peer_wake_connect(sk, other);
431
432 if (unix_recvq_full(other))
433 return 1;
434
435 if (connected)
436 unix_dgram_peer_wake_disconnect(sk, other);
437
438 return 0;
439}
440
441static int unix_writable(const struct sock *sk)
442{
443 return sk->sk_state != TCP_LISTEN &&
444 (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
445}
446
447static void unix_write_space(struct sock *sk)
448{
449 struct socket_wq *wq;
450
451 rcu_read_lock();
452 if (unix_writable(sk)) {
453 wq = rcu_dereference(sk->sk_wq);
454 if (skwq_has_sleeper(wq))
455 wake_up_interruptible_sync_poll(&wq->wait,
456 POLLOUT | POLLWRNORM | POLLWRBAND);
457 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
458 }
459 rcu_read_unlock();
460}
461
462/* When dgram socket disconnects (or changes its peer), we clear its receive
463 * queue of packets arrived from previous peer. First, it allows to do
464 * flow control based only on wmem_alloc; second, sk connected to peer
465 * may receive messages only from that peer. */
466static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
467{
468 if (!skb_queue_empty(&sk->sk_receive_queue)) {
469 skb_queue_purge(&sk->sk_receive_queue);
470 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
471
472 /* If one link of bidirectional dgram pipe is disconnected,
473 * we signal error. Messages are lost. Do not make this,
474 * when peer was not connected to us.
475 */
476 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
477 other->sk_err = ECONNRESET;
478 other->sk_error_report(other);
479 }
480 }
481}
482
483static void unix_sock_destructor(struct sock *sk)
484{
485 struct unix_sock *u = unix_sk(sk);
486
487 skb_queue_purge(&sk->sk_receive_queue);
488
489 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
490 WARN_ON(!sk_unhashed(sk));
491 WARN_ON(sk->sk_socket);
492 if (!sock_flag(sk, SOCK_DEAD)) {
493 pr_info("Attempt to release alive unix socket: %p\n", sk);
494 return;
495 }
496
497 if (u->addr)
498 unix_release_addr(u->addr);
499
500 atomic_long_dec(&unix_nr_socks);
501 local_bh_disable();
502 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
503 local_bh_enable();
504#ifdef UNIX_REFCNT_DEBUG
505 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
506 atomic_long_read(&unix_nr_socks));
507#endif
508}
509
510static void unix_release_sock(struct sock *sk, int embrion)
511{
512 struct unix_sock *u = unix_sk(sk);
513 struct path path;
514 struct sock *skpair;
515 struct sk_buff *skb;
516 int state;
517
518 unix_remove_socket(sk);
519
520 /* Clear state */
521 unix_state_lock(sk);
522 sock_orphan(sk);
523 sk->sk_shutdown = SHUTDOWN_MASK;
524 path = u->path;
525 u->path.dentry = NULL;
526 u->path.mnt = NULL;
527 state = sk->sk_state;
528 sk->sk_state = TCP_CLOSE;
529 unix_state_unlock(sk);
530
531 wake_up_interruptible_all(&u->peer_wait);
532
533 skpair = unix_peer(sk);
534
535 if (skpair != NULL) {
536 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
537 unix_state_lock(skpair);
538 /* No more writes */
539 skpair->sk_shutdown = SHUTDOWN_MASK;
540 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
541 skpair->sk_err = ECONNRESET;
542 unix_state_unlock(skpair);
543 skpair->sk_state_change(skpair);
544 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
545 }
546
547 unix_dgram_peer_wake_disconnect(sk, skpair);
548 sock_put(skpair); /* It may now die */
549 unix_peer(sk) = NULL;
550 }
551
552 /* Try to flush out this socket. Throw out buffers at least */
553
554 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
555 if (state == TCP_LISTEN)
556 unix_release_sock(skb->sk, 1);
557 /* passed fds are erased in the kfree_skb hook */
558 UNIXCB(skb).consumed = skb->len;
559 kfree_skb(skb);
560 }
561
562 if (path.dentry)
563 path_put(&path);
564
565 sock_put(sk);
566
567 /* ---- Socket is dead now and most probably destroyed ---- */
568
569 /*
570 * Fixme: BSD difference: In BSD all sockets connected to us get
571 * ECONNRESET and we die on the spot. In Linux we behave
572 * like files and pipes do and wait for the last
573 * dereference.
574 *
575 * Can't we simply set sock->err?
576 *
577 * What the above comment does talk about? --ANK(980817)
578 */
579
580 if (unix_tot_inflight)
581 unix_gc(); /* Garbage collect fds */
582}
583
584static void init_peercred(struct sock *sk)
585{
586 put_pid(sk->sk_peer_pid);
587 if (sk->sk_peer_cred)
588 put_cred(sk->sk_peer_cred);
589 sk->sk_peer_pid = get_pid(task_tgid(current));
590 sk->sk_peer_cred = get_current_cred();
591}
592
593static void copy_peercred(struct sock *sk, struct sock *peersk)
594{
595 put_pid(sk->sk_peer_pid);
596 if (sk->sk_peer_cred)
597 put_cred(sk->sk_peer_cred);
598 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
599 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
600}
601
602static int unix_listen(struct socket *sock, int backlog)
603{
604 int err;
605 struct sock *sk = sock->sk;
606 struct unix_sock *u = unix_sk(sk);
607 struct pid *old_pid = NULL;
608
609 err = -EOPNOTSUPP;
610 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
611 goto out; /* Only stream/seqpacket sockets accept */
612 err = -EINVAL;
613 if (!u->addr)
614 goto out; /* No listens on an unbound socket */
615 unix_state_lock(sk);
616 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
617 goto out_unlock;
618 if (backlog > sk->sk_max_ack_backlog)
619 wake_up_interruptible_all(&u->peer_wait);
620 sk->sk_max_ack_backlog = backlog;
621 sk->sk_state = TCP_LISTEN;
622 /* set credentials so connect can copy them */
623 init_peercred(sk);
624 err = 0;
625
626out_unlock:
627 unix_state_unlock(sk);
628 put_pid(old_pid);
629out:
630 return err;
631}
632
633static int unix_release(struct socket *);
634static int unix_bind(struct socket *, struct sockaddr *, int);
635static int unix_stream_connect(struct socket *, struct sockaddr *,
636 int addr_len, int flags);
637static int unix_socketpair(struct socket *, struct socket *);
638static int unix_accept(struct socket *, struct socket *, int);
639static int unix_getname(struct socket *, struct sockaddr *, int *, int);
640static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
641static unsigned int unix_dgram_poll(struct file *, struct socket *,
642 poll_table *);
643static int unix_ioctl(struct socket *, unsigned int, unsigned long);
644static int unix_shutdown(struct socket *, int);
645static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
646static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
647static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
648 size_t size, int flags);
649static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
650 struct pipe_inode_info *, size_t size,
651 unsigned int flags);
652static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
653static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
654static int unix_dgram_connect(struct socket *, struct sockaddr *,
655 int, int);
656static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
657static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
658 int);
659
660static int unix_set_peek_off(struct sock *sk, int val)
661{
662 struct unix_sock *u = unix_sk(sk);
663
664 if (mutex_lock_interruptible(&u->iolock))
665 return -EINTR;
666
667 sk->sk_peek_off = val;
668 mutex_unlock(&u->iolock);
669
670 return 0;
671}
672
673
674static const struct proto_ops unix_stream_ops = {
675 .family = PF_UNIX,
676 .owner = THIS_MODULE,
677 .release = unix_release,
678 .bind = unix_bind,
679 .connect = unix_stream_connect,
680 .socketpair = unix_socketpair,
681 .accept = unix_accept,
682 .getname = unix_getname,
683 .poll = unix_poll,
684 .ioctl = unix_ioctl,
685 .listen = unix_listen,
686 .shutdown = unix_shutdown,
687 .setsockopt = sock_no_setsockopt,
688 .getsockopt = sock_no_getsockopt,
689 .sendmsg = unix_stream_sendmsg,
690 .recvmsg = unix_stream_recvmsg,
691 .mmap = sock_no_mmap,
692 .sendpage = unix_stream_sendpage,
693 .splice_read = unix_stream_splice_read,
694 .set_peek_off = unix_set_peek_off,
695};
696
697static const struct proto_ops unix_dgram_ops = {
698 .family = PF_UNIX,
699 .owner = THIS_MODULE,
700 .release = unix_release,
701 .bind = unix_bind,
702 .connect = unix_dgram_connect,
703 .socketpair = unix_socketpair,
704 .accept = sock_no_accept,
705 .getname = unix_getname,
706 .poll = unix_dgram_poll,
707 .ioctl = unix_ioctl,
708 .listen = sock_no_listen,
709 .shutdown = unix_shutdown,
710 .setsockopt = sock_no_setsockopt,
711 .getsockopt = sock_no_getsockopt,
712 .sendmsg = unix_dgram_sendmsg,
713 .recvmsg = unix_dgram_recvmsg,
714 .mmap = sock_no_mmap,
715 .sendpage = sock_no_sendpage,
716 .set_peek_off = unix_set_peek_off,
717};
718
719static const struct proto_ops unix_seqpacket_ops = {
720 .family = PF_UNIX,
721 .owner = THIS_MODULE,
722 .release = unix_release,
723 .bind = unix_bind,
724 .connect = unix_stream_connect,
725 .socketpair = unix_socketpair,
726 .accept = unix_accept,
727 .getname = unix_getname,
728 .poll = unix_dgram_poll,
729 .ioctl = unix_ioctl,
730 .listen = unix_listen,
731 .shutdown = unix_shutdown,
732 .setsockopt = sock_no_setsockopt,
733 .getsockopt = sock_no_getsockopt,
734 .sendmsg = unix_seqpacket_sendmsg,
735 .recvmsg = unix_seqpacket_recvmsg,
736 .mmap = sock_no_mmap,
737 .sendpage = sock_no_sendpage,
738 .set_peek_off = unix_set_peek_off,
739};
740
741static struct proto unix_proto = {
742 .name = "UNIX",
743 .owner = THIS_MODULE,
744 .obj_size = sizeof(struct unix_sock),
745};
746
747/*
748 * AF_UNIX sockets do not interact with hardware, hence they
749 * dont trigger interrupts - so it's safe for them to have
750 * bh-unsafe locking for their sk_receive_queue.lock. Split off
751 * this special lock-class by reinitializing the spinlock key:
752 */
753static struct lock_class_key af_unix_sk_receive_queue_lock_key;
754
755static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
756{
757 struct sock *sk = NULL;
758 struct unix_sock *u;
759
760 atomic_long_inc(&unix_nr_socks);
761 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
762 goto out;
763
764 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
765 if (!sk)
766 goto out;
767
768 sock_init_data(sock, sk);
769 lockdep_set_class(&sk->sk_receive_queue.lock,
770 &af_unix_sk_receive_queue_lock_key);
771
772 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
773 sk->sk_write_space = unix_write_space;
774 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
775 sk->sk_destruct = unix_sock_destructor;
776 u = unix_sk(sk);
777 u->path.dentry = NULL;
778 u->path.mnt = NULL;
779 spin_lock_init(&u->lock);
780 atomic_long_set(&u->inflight, 0);
781 INIT_LIST_HEAD(&u->link);
782 mutex_init(&u->iolock); /* single task reading lock */
783 mutex_init(&u->bindlock); /* single task binding lock */
784 init_waitqueue_head(&u->peer_wait);
785 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
786 unix_insert_socket(unix_sockets_unbound(sk), sk);
787out:
788 if (sk == NULL)
789 atomic_long_dec(&unix_nr_socks);
790 else {
791 local_bh_disable();
792 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
793 local_bh_enable();
794 }
795 return sk;
796}
797
798static int unix_create(struct net *net, struct socket *sock, int protocol,
799 int kern)
800{
801 if (protocol && protocol != PF_UNIX)
802 return -EPROTONOSUPPORT;
803
804 sock->state = SS_UNCONNECTED;
805
806 switch (sock->type) {
807 case SOCK_STREAM:
808 sock->ops = &unix_stream_ops;
809 break;
810 /*
811 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
812 * nothing uses it.
813 */
814 case SOCK_RAW:
815 sock->type = SOCK_DGRAM;
816 case SOCK_DGRAM:
817 sock->ops = &unix_dgram_ops;
818 break;
819 case SOCK_SEQPACKET:
820 sock->ops = &unix_seqpacket_ops;
821 break;
822 default:
823 return -ESOCKTNOSUPPORT;
824 }
825
826 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
827}
828
829static int unix_release(struct socket *sock)
830{
831 struct sock *sk = sock->sk;
832
833 if (!sk)
834 return 0;
835
836 unix_release_sock(sk, 0);
837 sock->sk = NULL;
838
839 return 0;
840}
841
842static int unix_autobind(struct socket *sock)
843{
844 struct sock *sk = sock->sk;
845 struct net *net = sock_net(sk);
846 struct unix_sock *u = unix_sk(sk);
847 static u32 ordernum = 1;
848 struct unix_address *addr;
849 int err;
850 unsigned int retries = 0;
851
852 err = mutex_lock_interruptible(&u->bindlock);
853 if (err)
854 return err;
855
856 err = 0;
857 if (u->addr)
858 goto out;
859
860 err = -ENOMEM;
861 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
862 if (!addr)
863 goto out;
864
865 addr->name->sun_family = AF_UNIX;
866 atomic_set(&addr->refcnt, 1);
867
868retry:
869 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
870 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
871
872 spin_lock(&unix_table_lock);
873 ordernum = (ordernum+1)&0xFFFFF;
874
875 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
876 addr->hash)) {
877 spin_unlock(&unix_table_lock);
878 /*
879 * __unix_find_socket_byname() may take long time if many names
880 * are already in use.
881 */
882 cond_resched();
883 /* Give up if all names seems to be in use. */
884 if (retries++ == 0xFFFFF) {
885 err = -ENOSPC;
886 kfree(addr);
887 goto out;
888 }
889 goto retry;
890 }
891 addr->hash ^= sk->sk_type;
892
893 __unix_remove_socket(sk);
894 u->addr = addr;
895 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
896 spin_unlock(&unix_table_lock);
897 err = 0;
898
899out: mutex_unlock(&u->bindlock);
900 return err;
901}
902
903static struct sock *unix_find_other(struct net *net,
904 struct sockaddr_un *sunname, int len,
905 int type, unsigned int hash, int *error)
906{
907 struct sock *u;
908 struct path path;
909 int err = 0;
910
911 if (sunname->sun_path[0]) {
912 struct inode *inode;
913 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
914 if (err)
915 goto fail;
916 inode = d_backing_inode(path.dentry);
917 err = inode_permission(inode, MAY_WRITE);
918 if (err)
919 goto put_fail;
920
921 err = -ECONNREFUSED;
922 if (!S_ISSOCK(inode->i_mode))
923 goto put_fail;
924 u = unix_find_socket_byinode(inode);
925 if (!u)
926 goto put_fail;
927
928 if (u->sk_type == type)
929 touch_atime(&path);
930
931 path_put(&path);
932
933 err = -EPROTOTYPE;
934 if (u->sk_type != type) {
935 sock_put(u);
936 goto fail;
937 }
938 } else {
939 err = -ECONNREFUSED;
940 u = unix_find_socket_byname(net, sunname, len, type, hash);
941 if (u) {
942 struct dentry *dentry;
943 dentry = unix_sk(u)->path.dentry;
944 if (dentry)
945 touch_atime(&unix_sk(u)->path);
946 } else
947 goto fail;
948 }
949 return u;
950
951put_fail:
952 path_put(&path);
953fail:
954 *error = err;
955 return NULL;
956}
957
958static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
959{
960 struct dentry *dentry;
961 struct path path;
962 int err = 0;
963 /*
964 * Get the parent directory, calculate the hash for last
965 * component.
966 */
967 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
968 err = PTR_ERR(dentry);
969 if (IS_ERR(dentry))
970 return err;
971
972 /*
973 * All right, let's create it.
974 */
975 err = security_path_mknod(&path, dentry, mode, 0);
976 if (!err) {
977 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
978 if (!err) {
979 res->mnt = mntget(path.mnt);
980 res->dentry = dget(dentry);
981 }
982 }
983 done_path_create(&path, dentry);
984 return err;
985}
986
987static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
988{
989 struct sock *sk = sock->sk;
990 struct net *net = sock_net(sk);
991 struct unix_sock *u = unix_sk(sk);
992 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
993 char *sun_path = sunaddr->sun_path;
994 int err;
995 unsigned int hash;
996 struct unix_address *addr;
997 struct hlist_head *list;
998 struct path path = { NULL, NULL };
999
1000 err = -EINVAL;
1001 if (sunaddr->sun_family != AF_UNIX)
1002 goto out;
1003
1004 if (addr_len == sizeof(short)) {
1005 err = unix_autobind(sock);
1006 goto out;
1007 }
1008
1009 err = unix_mkname(sunaddr, addr_len, &hash);
1010 if (err < 0)
1011 goto out;
1012 addr_len = err;
1013
1014 if (sun_path[0]) {
1015 umode_t mode = S_IFSOCK |
1016 (SOCK_INODE(sock)->i_mode & ~current_umask());
1017 err = unix_mknod(sun_path, mode, &path);
1018 if (err) {
1019 if (err == -EEXIST)
1020 err = -EADDRINUSE;
1021 goto out;
1022 }
1023 }
1024
1025 err = mutex_lock_interruptible(&u->bindlock);
1026 if (err)
1027 goto out_put;
1028
1029 err = -EINVAL;
1030 if (u->addr)
1031 goto out_up;
1032
1033 err = -ENOMEM;
1034 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1035 if (!addr)
1036 goto out_up;
1037
1038 memcpy(addr->name, sunaddr, addr_len);
1039 addr->len = addr_len;
1040 addr->hash = hash ^ sk->sk_type;
1041 atomic_set(&addr->refcnt, 1);
1042
1043 if (sun_path[0]) {
1044 addr->hash = UNIX_HASH_SIZE;
1045 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1046 spin_lock(&unix_table_lock);
1047 u->path = path;
1048 list = &unix_socket_table[hash];
1049 } else {
1050 spin_lock(&unix_table_lock);
1051 err = -EADDRINUSE;
1052 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1053 sk->sk_type, hash)) {
1054 unix_release_addr(addr);
1055 goto out_unlock;
1056 }
1057
1058 list = &unix_socket_table[addr->hash];
1059 }
1060
1061 err = 0;
1062 __unix_remove_socket(sk);
1063 u->addr = addr;
1064 __unix_insert_socket(list, sk);
1065
1066out_unlock:
1067 spin_unlock(&unix_table_lock);
1068out_up:
1069 mutex_unlock(&u->bindlock);
1070out_put:
1071 if (err)
1072 path_put(&path);
1073out:
1074 return err;
1075}
1076
1077static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1078{
1079 if (unlikely(sk1 == sk2) || !sk2) {
1080 unix_state_lock(sk1);
1081 return;
1082 }
1083 if (sk1 < sk2) {
1084 unix_state_lock(sk1);
1085 unix_state_lock_nested(sk2);
1086 } else {
1087 unix_state_lock(sk2);
1088 unix_state_lock_nested(sk1);
1089 }
1090}
1091
1092static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1093{
1094 if (unlikely(sk1 == sk2) || !sk2) {
1095 unix_state_unlock(sk1);
1096 return;
1097 }
1098 unix_state_unlock(sk1);
1099 unix_state_unlock(sk2);
1100}
1101
1102static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1103 int alen, int flags)
1104{
1105 struct sock *sk = sock->sk;
1106 struct net *net = sock_net(sk);
1107 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1108 struct sock *other;
1109 unsigned int hash;
1110 int err;
1111
1112 if (addr->sa_family != AF_UNSPEC) {
1113 err = unix_mkname(sunaddr, alen, &hash);
1114 if (err < 0)
1115 goto out;
1116 alen = err;
1117
1118 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1119 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1120 goto out;
1121
1122restart:
1123 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1124 if (!other)
1125 goto out;
1126
1127 unix_state_double_lock(sk, other);
1128
1129 /* Apparently VFS overslept socket death. Retry. */
1130 if (sock_flag(other, SOCK_DEAD)) {
1131 unix_state_double_unlock(sk, other);
1132 sock_put(other);
1133 goto restart;
1134 }
1135
1136 err = -EPERM;
1137 if (!unix_may_send(sk, other))
1138 goto out_unlock;
1139
1140 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1141 if (err)
1142 goto out_unlock;
1143
1144 } else {
1145 /*
1146 * 1003.1g breaking connected state with AF_UNSPEC
1147 */
1148 other = NULL;
1149 unix_state_double_lock(sk, other);
1150 }
1151
1152 /*
1153 * If it was connected, reconnect.
1154 */
1155 if (unix_peer(sk)) {
1156 struct sock *old_peer = unix_peer(sk);
1157 unix_peer(sk) = other;
1158 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1159
1160 unix_state_double_unlock(sk, other);
1161
1162 if (other != old_peer)
1163 unix_dgram_disconnected(sk, old_peer);
1164 sock_put(old_peer);
1165 } else {
1166 unix_peer(sk) = other;
1167 unix_state_double_unlock(sk, other);
1168 }
1169 return 0;
1170
1171out_unlock:
1172 unix_state_double_unlock(sk, other);
1173 sock_put(other);
1174out:
1175 return err;
1176}
1177
1178static long unix_wait_for_peer(struct sock *other, long timeo)
1179{
1180 struct unix_sock *u = unix_sk(other);
1181 int sched;
1182 DEFINE_WAIT(wait);
1183
1184 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1185
1186 sched = !sock_flag(other, SOCK_DEAD) &&
1187 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1188 unix_recvq_full(other);
1189
1190 unix_state_unlock(other);
1191
1192 if (sched)
1193 timeo = schedule_timeout(timeo);
1194
1195 finish_wait(&u->peer_wait, &wait);
1196 return timeo;
1197}
1198
1199static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1200 int addr_len, int flags)
1201{
1202 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1203 struct sock *sk = sock->sk;
1204 struct net *net = sock_net(sk);
1205 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1206 struct sock *newsk = NULL;
1207 struct sock *other = NULL;
1208 struct sk_buff *skb = NULL;
1209 unsigned int hash;
1210 int st;
1211 int err;
1212 long timeo;
1213
1214 err = unix_mkname(sunaddr, addr_len, &hash);
1215 if (err < 0)
1216 goto out;
1217 addr_len = err;
1218
1219 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1220 (err = unix_autobind(sock)) != 0)
1221 goto out;
1222
1223 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1224
1225 /* First of all allocate resources.
1226 If we will make it after state is locked,
1227 we will have to recheck all again in any case.
1228 */
1229
1230 err = -ENOMEM;
1231
1232 /* create new sock for complete connection */
1233 newsk = unix_create1(sock_net(sk), NULL, 0);
1234 if (newsk == NULL)
1235 goto out;
1236
1237 /* Allocate skb for sending to listening sock */
1238 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1239 if (skb == NULL)
1240 goto out;
1241
1242restart:
1243 /* Find listening sock. */
1244 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1245 if (!other)
1246 goto out;
1247
1248 /* Latch state of peer */
1249 unix_state_lock(other);
1250
1251 /* Apparently VFS overslept socket death. Retry. */
1252 if (sock_flag(other, SOCK_DEAD)) {
1253 unix_state_unlock(other);
1254 sock_put(other);
1255 goto restart;
1256 }
1257
1258 err = -ECONNREFUSED;
1259 if (other->sk_state != TCP_LISTEN)
1260 goto out_unlock;
1261 if (other->sk_shutdown & RCV_SHUTDOWN)
1262 goto out_unlock;
1263
1264 if (unix_recvq_full(other)) {
1265 err = -EAGAIN;
1266 if (!timeo)
1267 goto out_unlock;
1268
1269 timeo = unix_wait_for_peer(other, timeo);
1270
1271 err = sock_intr_errno(timeo);
1272 if (signal_pending(current))
1273 goto out;
1274 sock_put(other);
1275 goto restart;
1276 }
1277
1278 /* Latch our state.
1279
1280 It is tricky place. We need to grab our state lock and cannot
1281 drop lock on peer. It is dangerous because deadlock is
1282 possible. Connect to self case and simultaneous
1283 attempt to connect are eliminated by checking socket
1284 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1285 check this before attempt to grab lock.
1286
1287 Well, and we have to recheck the state after socket locked.
1288 */
1289 st = sk->sk_state;
1290
1291 switch (st) {
1292 case TCP_CLOSE:
1293 /* This is ok... continue with connect */
1294 break;
1295 case TCP_ESTABLISHED:
1296 /* Socket is already connected */
1297 err = -EISCONN;
1298 goto out_unlock;
1299 default:
1300 err = -EINVAL;
1301 goto out_unlock;
1302 }
1303
1304 unix_state_lock_nested(sk);
1305
1306 if (sk->sk_state != st) {
1307 unix_state_unlock(sk);
1308 unix_state_unlock(other);
1309 sock_put(other);
1310 goto restart;
1311 }
1312
1313 err = security_unix_stream_connect(sk, other, newsk);
1314 if (err) {
1315 unix_state_unlock(sk);
1316 goto out_unlock;
1317 }
1318
1319 /* The way is open! Fastly set all the necessary fields... */
1320
1321 sock_hold(sk);
1322 unix_peer(newsk) = sk;
1323 newsk->sk_state = TCP_ESTABLISHED;
1324 newsk->sk_type = sk->sk_type;
1325 init_peercred(newsk);
1326 newu = unix_sk(newsk);
1327 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1328 otheru = unix_sk(other);
1329
1330 /* copy address information from listening to new sock*/
1331 if (otheru->addr) {
1332 atomic_inc(&otheru->addr->refcnt);
1333 newu->addr = otheru->addr;
1334 }
1335 if (otheru->path.dentry) {
1336 path_get(&otheru->path);
1337 newu->path = otheru->path;
1338 }
1339
1340 /* Set credentials */
1341 copy_peercred(sk, other);
1342
1343 sock->state = SS_CONNECTED;
1344 sk->sk_state = TCP_ESTABLISHED;
1345 sock_hold(newsk);
1346
1347 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1348 unix_peer(sk) = newsk;
1349
1350 unix_state_unlock(sk);
1351
1352 /* take ten and and send info to listening sock */
1353 spin_lock(&other->sk_receive_queue.lock);
1354 __skb_queue_tail(&other->sk_receive_queue, skb);
1355 spin_unlock(&other->sk_receive_queue.lock);
1356 unix_state_unlock(other);
1357 other->sk_data_ready(other);
1358 sock_put(other);
1359 return 0;
1360
1361out_unlock:
1362 if (other)
1363 unix_state_unlock(other);
1364
1365out:
1366 kfree_skb(skb);
1367 if (newsk)
1368 unix_release_sock(newsk, 0);
1369 if (other)
1370 sock_put(other);
1371 return err;
1372}
1373
1374static int unix_socketpair(struct socket *socka, struct socket *sockb)
1375{
1376 struct sock *ska = socka->sk, *skb = sockb->sk;
1377
1378 /* Join our sockets back to back */
1379 sock_hold(ska);
1380 sock_hold(skb);
1381 unix_peer(ska) = skb;
1382 unix_peer(skb) = ska;
1383 init_peercred(ska);
1384 init_peercred(skb);
1385
1386 if (ska->sk_type != SOCK_DGRAM) {
1387 ska->sk_state = TCP_ESTABLISHED;
1388 skb->sk_state = TCP_ESTABLISHED;
1389 socka->state = SS_CONNECTED;
1390 sockb->state = SS_CONNECTED;
1391 }
1392 return 0;
1393}
1394
1395static void unix_sock_inherit_flags(const struct socket *old,
1396 struct socket *new)
1397{
1398 if (test_bit(SOCK_PASSCRED, &old->flags))
1399 set_bit(SOCK_PASSCRED, &new->flags);
1400 if (test_bit(SOCK_PASSSEC, &old->flags))
1401 set_bit(SOCK_PASSSEC, &new->flags);
1402}
1403
1404static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1405{
1406 struct sock *sk = sock->sk;
1407 struct sock *tsk;
1408 struct sk_buff *skb;
1409 int err;
1410
1411 err = -EOPNOTSUPP;
1412 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1413 goto out;
1414
1415 err = -EINVAL;
1416 if (sk->sk_state != TCP_LISTEN)
1417 goto out;
1418
1419 /* If socket state is TCP_LISTEN it cannot change (for now...),
1420 * so that no locks are necessary.
1421 */
1422
1423 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1424 if (!skb) {
1425 /* This means receive shutdown. */
1426 if (err == 0)
1427 err = -EINVAL;
1428 goto out;
1429 }
1430
1431 tsk = skb->sk;
1432 skb_free_datagram(sk, skb);
1433 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1434
1435 /* attach accepted sock to socket */
1436 unix_state_lock(tsk);
1437 newsock->state = SS_CONNECTED;
1438 unix_sock_inherit_flags(sock, newsock);
1439 sock_graft(tsk, newsock);
1440 unix_state_unlock(tsk);
1441 return 0;
1442
1443out:
1444 return err;
1445}
1446
1447
1448static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1449{
1450 struct sock *sk = sock->sk;
1451 struct unix_sock *u;
1452 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1453 int err = 0;
1454
1455 if (peer) {
1456 sk = unix_peer_get(sk);
1457
1458 err = -ENOTCONN;
1459 if (!sk)
1460 goto out;
1461 err = 0;
1462 } else {
1463 sock_hold(sk);
1464 }
1465
1466 u = unix_sk(sk);
1467 unix_state_lock(sk);
1468 if (!u->addr) {
1469 sunaddr->sun_family = AF_UNIX;
1470 sunaddr->sun_path[0] = 0;
1471 *uaddr_len = sizeof(short);
1472 } else {
1473 struct unix_address *addr = u->addr;
1474
1475 *uaddr_len = addr->len;
1476 memcpy(sunaddr, addr->name, *uaddr_len);
1477 }
1478 unix_state_unlock(sk);
1479 sock_put(sk);
1480out:
1481 return err;
1482}
1483
1484static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1485{
1486 int i;
1487
1488 scm->fp = UNIXCB(skb).fp;
1489 UNIXCB(skb).fp = NULL;
1490
1491 for (i = scm->fp->count-1; i >= 0; i--)
1492 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1493}
1494
1495static void unix_destruct_scm(struct sk_buff *skb)
1496{
1497 struct scm_cookie scm;
1498 memset(&scm, 0, sizeof(scm));
1499 scm.pid = UNIXCB(skb).pid;
1500 if (UNIXCB(skb).fp)
1501 unix_detach_fds(&scm, skb);
1502
1503 /* Alas, it calls VFS */
1504 /* So fscking what? fput() had been SMP-safe since the last Summer */
1505 scm_destroy(&scm);
1506 sock_wfree(skb);
1507}
1508
1509/*
1510 * The "user->unix_inflight" variable is protected by the garbage
1511 * collection lock, and we just read it locklessly here. If you go
1512 * over the limit, there might be a tiny race in actually noticing
1513 * it across threads. Tough.
1514 */
1515static inline bool too_many_unix_fds(struct task_struct *p)
1516{
1517 struct user_struct *user = current_user();
1518
1519 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1520 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1521 return false;
1522}
1523
1524#define MAX_RECURSION_LEVEL 4
1525
1526static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1527{
1528 int i;
1529 unsigned char max_level = 0;
1530
1531 if (too_many_unix_fds(current))
1532 return -ETOOMANYREFS;
1533
1534 for (i = scm->fp->count - 1; i >= 0; i--) {
1535 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1536
1537 if (sk)
1538 max_level = max(max_level,
1539 unix_sk(sk)->recursion_level);
1540 }
1541 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1542 return -ETOOMANYREFS;
1543
1544 /*
1545 * Need to duplicate file references for the sake of garbage
1546 * collection. Otherwise a socket in the fps might become a
1547 * candidate for GC while the skb is not yet queued.
1548 */
1549 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1550 if (!UNIXCB(skb).fp)
1551 return -ENOMEM;
1552
1553 for (i = scm->fp->count - 1; i >= 0; i--)
1554 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1555 return max_level;
1556}
1557
1558static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1559{
1560 int err = 0;
1561
1562 UNIXCB(skb).pid = get_pid(scm->pid);
1563 UNIXCB(skb).uid = scm->creds.uid;
1564 UNIXCB(skb).gid = scm->creds.gid;
1565 UNIXCB(skb).fp = NULL;
1566 unix_get_secdata(scm, skb);
1567 if (scm->fp && send_fds)
1568 err = unix_attach_fds(scm, skb);
1569
1570 skb->destructor = unix_destruct_scm;
1571 return err;
1572}
1573
1574static bool unix_passcred_enabled(const struct socket *sock,
1575 const struct sock *other)
1576{
1577 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1578 !other->sk_socket ||
1579 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1580}
1581
1582/*
1583 * Some apps rely on write() giving SCM_CREDENTIALS
1584 * We include credentials if source or destination socket
1585 * asserted SOCK_PASSCRED.
1586 */
1587static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1588 const struct sock *other)
1589{
1590 if (UNIXCB(skb).pid)
1591 return;
1592 if (unix_passcred_enabled(sock, other)) {
1593 UNIXCB(skb).pid = get_pid(task_tgid(current));
1594 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1595 }
1596}
1597
1598static int maybe_init_creds(struct scm_cookie *scm,
1599 struct socket *socket,
1600 const struct sock *other)
1601{
1602 int err;
1603 struct msghdr msg = { .msg_controllen = 0 };
1604
1605 err = scm_send(socket, &msg, scm, false);
1606 if (err)
1607 return err;
1608
1609 if (unix_passcred_enabled(socket, other)) {
1610 scm->pid = get_pid(task_tgid(current));
1611 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1612 }
1613 return err;
1614}
1615
1616static bool unix_skb_scm_eq(struct sk_buff *skb,
1617 struct scm_cookie *scm)
1618{
1619 const struct unix_skb_parms *u = &UNIXCB(skb);
1620
1621 return u->pid == scm->pid &&
1622 uid_eq(u->uid, scm->creds.uid) &&
1623 gid_eq(u->gid, scm->creds.gid) &&
1624 unix_secdata_eq(scm, skb);
1625}
1626
1627/*
1628 * Send AF_UNIX data.
1629 */
1630
1631static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1632 size_t len)
1633{
1634 struct sock *sk = sock->sk;
1635 struct net *net = sock_net(sk);
1636 struct unix_sock *u = unix_sk(sk);
1637 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1638 struct sock *other = NULL;
1639 int namelen = 0; /* fake GCC */
1640 int err;
1641 unsigned int hash;
1642 struct sk_buff *skb;
1643 long timeo;
1644 struct scm_cookie scm;
1645 int max_level;
1646 int data_len = 0;
1647 int sk_locked;
1648
1649 wait_for_unix_gc();
1650 err = scm_send(sock, msg, &scm, false);
1651 if (err < 0)
1652 return err;
1653
1654 err = -EOPNOTSUPP;
1655 if (msg->msg_flags&MSG_OOB)
1656 goto out;
1657
1658 if (msg->msg_namelen) {
1659 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1660 if (err < 0)
1661 goto out;
1662 namelen = err;
1663 } else {
1664 sunaddr = NULL;
1665 err = -ENOTCONN;
1666 other = unix_peer_get(sk);
1667 if (!other)
1668 goto out;
1669 }
1670
1671 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1672 && (err = unix_autobind(sock)) != 0)
1673 goto out;
1674
1675 err = -EMSGSIZE;
1676 if (len > sk->sk_sndbuf - 32)
1677 goto out;
1678
1679 if (len > SKB_MAX_ALLOC) {
1680 data_len = min_t(size_t,
1681 len - SKB_MAX_ALLOC,
1682 MAX_SKB_FRAGS * PAGE_SIZE);
1683 data_len = PAGE_ALIGN(data_len);
1684
1685 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1686 }
1687
1688 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1689 msg->msg_flags & MSG_DONTWAIT, &err,
1690 PAGE_ALLOC_COSTLY_ORDER);
1691 if (skb == NULL)
1692 goto out;
1693
1694 err = unix_scm_to_skb(&scm, skb, true);
1695 if (err < 0)
1696 goto out_free;
1697 max_level = err + 1;
1698
1699 skb_put(skb, len - data_len);
1700 skb->data_len = data_len;
1701 skb->len = len;
1702 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1703 if (err)
1704 goto out_free;
1705
1706 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1707
1708restart:
1709 if (!other) {
1710 err = -ECONNRESET;
1711 if (sunaddr == NULL)
1712 goto out_free;
1713
1714 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1715 hash, &err);
1716 if (other == NULL)
1717 goto out_free;
1718 }
1719
1720 if (sk_filter(other, skb) < 0) {
1721 /* Toss the packet but do not return any error to the sender */
1722 err = len;
1723 goto out_free;
1724 }
1725
1726 sk_locked = 0;
1727 unix_state_lock(other);
1728restart_locked:
1729 err = -EPERM;
1730 if (!unix_may_send(sk, other))
1731 goto out_unlock;
1732
1733 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1734 /*
1735 * Check with 1003.1g - what should
1736 * datagram error
1737 */
1738 unix_state_unlock(other);
1739 sock_put(other);
1740
1741 if (!sk_locked)
1742 unix_state_lock(sk);
1743
1744 err = 0;
1745 if (unix_peer(sk) == other) {
1746 unix_peer(sk) = NULL;
1747 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1748
1749 unix_state_unlock(sk);
1750
1751 unix_dgram_disconnected(sk, other);
1752 sock_put(other);
1753 err = -ECONNREFUSED;
1754 } else {
1755 unix_state_unlock(sk);
1756 }
1757
1758 other = NULL;
1759 if (err)
1760 goto out_free;
1761 goto restart;
1762 }
1763
1764 err = -EPIPE;
1765 if (other->sk_shutdown & RCV_SHUTDOWN)
1766 goto out_unlock;
1767
1768 if (sk->sk_type != SOCK_SEQPACKET) {
1769 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1770 if (err)
1771 goto out_unlock;
1772 }
1773
1774 /* other == sk && unix_peer(other) != sk if
1775 * - unix_peer(sk) == NULL, destination address bound to sk
1776 * - unix_peer(sk) == sk by time of get but disconnected before lock
1777 */
1778 if (other != sk &&
1779 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1780 if (timeo) {
1781 timeo = unix_wait_for_peer(other, timeo);
1782
1783 err = sock_intr_errno(timeo);
1784 if (signal_pending(current))
1785 goto out_free;
1786
1787 goto restart;
1788 }
1789
1790 if (!sk_locked) {
1791 unix_state_unlock(other);
1792 unix_state_double_lock(sk, other);
1793 }
1794
1795 if (unix_peer(sk) != other ||
1796 unix_dgram_peer_wake_me(sk, other)) {
1797 err = -EAGAIN;
1798 sk_locked = 1;
1799 goto out_unlock;
1800 }
1801
1802 if (!sk_locked) {
1803 sk_locked = 1;
1804 goto restart_locked;
1805 }
1806 }
1807
1808 if (unlikely(sk_locked))
1809 unix_state_unlock(sk);
1810
1811 if (sock_flag(other, SOCK_RCVTSTAMP))
1812 __net_timestamp(skb);
1813 maybe_add_creds(skb, sock, other);
1814 skb_queue_tail(&other->sk_receive_queue, skb);
1815 if (max_level > unix_sk(other)->recursion_level)
1816 unix_sk(other)->recursion_level = max_level;
1817 unix_state_unlock(other);
1818 other->sk_data_ready(other);
1819 sock_put(other);
1820 scm_destroy(&scm);
1821 return len;
1822
1823out_unlock:
1824 if (sk_locked)
1825 unix_state_unlock(sk);
1826 unix_state_unlock(other);
1827out_free:
1828 kfree_skb(skb);
1829out:
1830 if (other)
1831 sock_put(other);
1832 scm_destroy(&scm);
1833 return err;
1834}
1835
1836/* We use paged skbs for stream sockets, and limit occupancy to 32768
1837 * bytes, and a minimun of a full page.
1838 */
1839#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1840
1841static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1842 size_t len)
1843{
1844 struct sock *sk = sock->sk;
1845 struct sock *other = NULL;
1846 int err, size;
1847 struct sk_buff *skb;
1848 int sent = 0;
1849 struct scm_cookie scm;
1850 bool fds_sent = false;
1851 int max_level;
1852 int data_len;
1853
1854 wait_for_unix_gc();
1855 err = scm_send(sock, msg, &scm, false);
1856 if (err < 0)
1857 return err;
1858
1859 err = -EOPNOTSUPP;
1860 if (msg->msg_flags&MSG_OOB)
1861 goto out_err;
1862
1863 if (msg->msg_namelen) {
1864 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1865 goto out_err;
1866 } else {
1867 err = -ENOTCONN;
1868 other = unix_peer(sk);
1869 if (!other)
1870 goto out_err;
1871 }
1872
1873 if (sk->sk_shutdown & SEND_SHUTDOWN)
1874 goto pipe_err;
1875
1876 while (sent < len) {
1877 size = len - sent;
1878
1879 /* Keep two messages in the pipe so it schedules better */
1880 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1881
1882 /* allow fallback to order-0 allocations */
1883 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1884
1885 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1886
1887 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1888
1889 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1890 msg->msg_flags & MSG_DONTWAIT, &err,
1891 get_order(UNIX_SKB_FRAGS_SZ));
1892 if (!skb)
1893 goto out_err;
1894
1895 /* Only send the fds in the first buffer */
1896 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1897 if (err < 0) {
1898 kfree_skb(skb);
1899 goto out_err;
1900 }
1901 max_level = err + 1;
1902 fds_sent = true;
1903
1904 skb_put(skb, size - data_len);
1905 skb->data_len = data_len;
1906 skb->len = size;
1907 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1908 if (err) {
1909 kfree_skb(skb);
1910 goto out_err;
1911 }
1912
1913 unix_state_lock(other);
1914
1915 if (sock_flag(other, SOCK_DEAD) ||
1916 (other->sk_shutdown & RCV_SHUTDOWN))
1917 goto pipe_err_free;
1918
1919 maybe_add_creds(skb, sock, other);
1920 skb_queue_tail(&other->sk_receive_queue, skb);
1921 if (max_level > unix_sk(other)->recursion_level)
1922 unix_sk(other)->recursion_level = max_level;
1923 unix_state_unlock(other);
1924 other->sk_data_ready(other);
1925 sent += size;
1926 }
1927
1928 scm_destroy(&scm);
1929
1930 return sent;
1931
1932pipe_err_free:
1933 unix_state_unlock(other);
1934 kfree_skb(skb);
1935pipe_err:
1936 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1937 send_sig(SIGPIPE, current, 0);
1938 err = -EPIPE;
1939out_err:
1940 scm_destroy(&scm);
1941 return sent ? : err;
1942}
1943
1944static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1945 int offset, size_t size, int flags)
1946{
1947 int err;
1948 bool send_sigpipe = false;
1949 bool init_scm = true;
1950 struct scm_cookie scm;
1951 struct sock *other, *sk = socket->sk;
1952 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1953
1954 if (flags & MSG_OOB)
1955 return -EOPNOTSUPP;
1956
1957 other = unix_peer(sk);
1958 if (!other || sk->sk_state != TCP_ESTABLISHED)
1959 return -ENOTCONN;
1960
1961 if (false) {
1962alloc_skb:
1963 unix_state_unlock(other);
1964 mutex_unlock(&unix_sk(other)->iolock);
1965 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1966 &err, 0);
1967 if (!newskb)
1968 goto err;
1969 }
1970
1971 /* we must acquire iolock as we modify already present
1972 * skbs in the sk_receive_queue and mess with skb->len
1973 */
1974 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1975 if (err) {
1976 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1977 goto err;
1978 }
1979
1980 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1981 err = -EPIPE;
1982 send_sigpipe = true;
1983 goto err_unlock;
1984 }
1985
1986 unix_state_lock(other);
1987
1988 if (sock_flag(other, SOCK_DEAD) ||
1989 other->sk_shutdown & RCV_SHUTDOWN) {
1990 err = -EPIPE;
1991 send_sigpipe = true;
1992 goto err_state_unlock;
1993 }
1994
1995 if (init_scm) {
1996 err = maybe_init_creds(&scm, socket, other);
1997 if (err)
1998 goto err_state_unlock;
1999 init_scm = false;
2000 }
2001
2002 skb = skb_peek_tail(&other->sk_receive_queue);
2003 if (tail && tail == skb) {
2004 skb = newskb;
2005 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2006 if (newskb) {
2007 skb = newskb;
2008 } else {
2009 tail = skb;
2010 goto alloc_skb;
2011 }
2012 } else if (newskb) {
2013 /* this is fast path, we don't necessarily need to
2014 * call to kfree_skb even though with newskb == NULL
2015 * this - does no harm
2016 */
2017 consume_skb(newskb);
2018 newskb = NULL;
2019 }
2020
2021 if (skb_append_pagefrags(skb, page, offset, size)) {
2022 tail = skb;
2023 goto alloc_skb;
2024 }
2025
2026 skb->len += size;
2027 skb->data_len += size;
2028 skb->truesize += size;
2029 atomic_add(size, &sk->sk_wmem_alloc);
2030
2031 if (newskb) {
2032 err = unix_scm_to_skb(&scm, skb, false);
2033 if (err)
2034 goto err_state_unlock;
2035 spin_lock(&other->sk_receive_queue.lock);
2036 __skb_queue_tail(&other->sk_receive_queue, newskb);
2037 spin_unlock(&other->sk_receive_queue.lock);
2038 }
2039
2040 unix_state_unlock(other);
2041 mutex_unlock(&unix_sk(other)->iolock);
2042
2043 other->sk_data_ready(other);
2044 scm_destroy(&scm);
2045 return size;
2046
2047err_state_unlock:
2048 unix_state_unlock(other);
2049err_unlock:
2050 mutex_unlock(&unix_sk(other)->iolock);
2051err:
2052 kfree_skb(newskb);
2053 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2054 send_sig(SIGPIPE, current, 0);
2055 if (!init_scm)
2056 scm_destroy(&scm);
2057 return err;
2058}
2059
2060static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2061 size_t len)
2062{
2063 int err;
2064 struct sock *sk = sock->sk;
2065
2066 err = sock_error(sk);
2067 if (err)
2068 return err;
2069
2070 if (sk->sk_state != TCP_ESTABLISHED)
2071 return -ENOTCONN;
2072
2073 if (msg->msg_namelen)
2074 msg->msg_namelen = 0;
2075
2076 return unix_dgram_sendmsg(sock, msg, len);
2077}
2078
2079static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2080 size_t size, int flags)
2081{
2082 struct sock *sk = sock->sk;
2083
2084 if (sk->sk_state != TCP_ESTABLISHED)
2085 return -ENOTCONN;
2086
2087 return unix_dgram_recvmsg(sock, msg, size, flags);
2088}
2089
2090static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2091{
2092 struct unix_sock *u = unix_sk(sk);
2093
2094 if (u->addr) {
2095 msg->msg_namelen = u->addr->len;
2096 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2097 }
2098}
2099
2100static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2101 size_t size, int flags)
2102{
2103 struct scm_cookie scm;
2104 struct sock *sk = sock->sk;
2105 struct unix_sock *u = unix_sk(sk);
2106 struct sk_buff *skb, *last;
2107 long timeo;
2108 int err;
2109 int peeked, skip;
2110
2111 err = -EOPNOTSUPP;
2112 if (flags&MSG_OOB)
2113 goto out;
2114
2115 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2116
2117 do {
2118 mutex_lock(&u->iolock);
2119
2120 skip = sk_peek_offset(sk, flags);
2121 skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2122 &err, &last);
2123 if (skb)
2124 break;
2125
2126 mutex_unlock(&u->iolock);
2127
2128 if (err != -EAGAIN)
2129 break;
2130 } while (timeo &&
2131 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2132
2133 if (!skb) { /* implies iolock unlocked */
2134 unix_state_lock(sk);
2135 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2136 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2137 (sk->sk_shutdown & RCV_SHUTDOWN))
2138 err = 0;
2139 unix_state_unlock(sk);
2140 goto out;
2141 }
2142
2143 if (wq_has_sleeper(&u->peer_wait))
2144 wake_up_interruptible_sync_poll(&u->peer_wait,
2145 POLLOUT | POLLWRNORM |
2146 POLLWRBAND);
2147
2148 if (msg->msg_name)
2149 unix_copy_addr(msg, skb->sk);
2150
2151 if (size > skb->len - skip)
2152 size = skb->len - skip;
2153 else if (size < skb->len - skip)
2154 msg->msg_flags |= MSG_TRUNC;
2155
2156 err = skb_copy_datagram_msg(skb, skip, msg, size);
2157 if (err)
2158 goto out_free;
2159
2160 if (sock_flag(sk, SOCK_RCVTSTAMP))
2161 __sock_recv_timestamp(msg, sk, skb);
2162
2163 memset(&scm, 0, sizeof(scm));
2164
2165 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2166 unix_set_secdata(&scm, skb);
2167
2168 if (!(flags & MSG_PEEK)) {
2169 if (UNIXCB(skb).fp)
2170 unix_detach_fds(&scm, skb);
2171
2172 sk_peek_offset_bwd(sk, skb->len);
2173 } else {
2174 /* It is questionable: on PEEK we could:
2175 - do not return fds - good, but too simple 8)
2176 - return fds, and do not return them on read (old strategy,
2177 apparently wrong)
2178 - clone fds (I chose it for now, it is the most universal
2179 solution)
2180
2181 POSIX 1003.1g does not actually define this clearly
2182 at all. POSIX 1003.1g doesn't define a lot of things
2183 clearly however!
2184
2185 */
2186
2187 sk_peek_offset_fwd(sk, size);
2188
2189 if (UNIXCB(skb).fp)
2190 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2191 }
2192 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2193
2194 scm_recv(sock, msg, &scm, flags);
2195
2196out_free:
2197 skb_free_datagram(sk, skb);
2198 mutex_unlock(&u->iolock);
2199out:
2200 return err;
2201}
2202
2203/*
2204 * Sleep until more data has arrived. But check for races..
2205 */
2206static long unix_stream_data_wait(struct sock *sk, long timeo,
2207 struct sk_buff *last, unsigned int last_len,
2208 bool freezable)
2209{
2210 struct sk_buff *tail;
2211 DEFINE_WAIT(wait);
2212
2213 unix_state_lock(sk);
2214
2215 for (;;) {
2216 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2217
2218 tail = skb_peek_tail(&sk->sk_receive_queue);
2219 if (tail != last ||
2220 (tail && tail->len != last_len) ||
2221 sk->sk_err ||
2222 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2223 signal_pending(current) ||
2224 !timeo)
2225 break;
2226
2227 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2228 unix_state_unlock(sk);
2229 if (freezable)
2230 timeo = freezable_schedule_timeout(timeo);
2231 else
2232 timeo = schedule_timeout(timeo);
2233 unix_state_lock(sk);
2234
2235 if (sock_flag(sk, SOCK_DEAD))
2236 break;
2237
2238 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2239 }
2240
2241 finish_wait(sk_sleep(sk), &wait);
2242 unix_state_unlock(sk);
2243 return timeo;
2244}
2245
2246static unsigned int unix_skb_len(const struct sk_buff *skb)
2247{
2248 return skb->len - UNIXCB(skb).consumed;
2249}
2250
2251struct unix_stream_read_state {
2252 int (*recv_actor)(struct sk_buff *, int, int,
2253 struct unix_stream_read_state *);
2254 struct socket *socket;
2255 struct msghdr *msg;
2256 struct pipe_inode_info *pipe;
2257 size_t size;
2258 int flags;
2259 unsigned int splice_flags;
2260};
2261
2262static int unix_stream_read_generic(struct unix_stream_read_state *state,
2263 bool freezable)
2264{
2265 struct scm_cookie scm;
2266 struct socket *sock = state->socket;
2267 struct sock *sk = sock->sk;
2268 struct unix_sock *u = unix_sk(sk);
2269 int copied = 0;
2270 int flags = state->flags;
2271 int noblock = flags & MSG_DONTWAIT;
2272 bool check_creds = false;
2273 int target;
2274 int err = 0;
2275 long timeo;
2276 int skip;
2277 size_t size = state->size;
2278 unsigned int last_len;
2279
2280 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2281 err = -EINVAL;
2282 goto out;
2283 }
2284
2285 if (unlikely(flags & MSG_OOB)) {
2286 err = -EOPNOTSUPP;
2287 goto out;
2288 }
2289
2290 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2291 timeo = sock_rcvtimeo(sk, noblock);
2292
2293 memset(&scm, 0, sizeof(scm));
2294
2295 /* Lock the socket to prevent queue disordering
2296 * while sleeps in memcpy_tomsg
2297 */
2298 mutex_lock(&u->iolock);
2299
2300 if (flags & MSG_PEEK)
2301 skip = sk_peek_offset(sk, flags);
2302 else
2303 skip = 0;
2304
2305 do {
2306 int chunk;
2307 bool drop_skb;
2308 struct sk_buff *skb, *last;
2309
2310redo:
2311 unix_state_lock(sk);
2312 if (sock_flag(sk, SOCK_DEAD)) {
2313 err = -ECONNRESET;
2314 goto unlock;
2315 }
2316 last = skb = skb_peek(&sk->sk_receive_queue);
2317 last_len = last ? last->len : 0;
2318again:
2319 if (skb == NULL) {
2320 unix_sk(sk)->recursion_level = 0;
2321 if (copied >= target)
2322 goto unlock;
2323
2324 /*
2325 * POSIX 1003.1g mandates this order.
2326 */
2327
2328 err = sock_error(sk);
2329 if (err)
2330 goto unlock;
2331 if (sk->sk_shutdown & RCV_SHUTDOWN)
2332 goto unlock;
2333
2334 unix_state_unlock(sk);
2335 if (!timeo) {
2336 err = -EAGAIN;
2337 break;
2338 }
2339
2340 mutex_unlock(&u->iolock);
2341
2342 timeo = unix_stream_data_wait(sk, timeo, last,
2343 last_len, freezable);
2344
2345 if (signal_pending(current)) {
2346 err = sock_intr_errno(timeo);
2347 scm_destroy(&scm);
2348 goto out;
2349 }
2350
2351 mutex_lock(&u->iolock);
2352 goto redo;
2353unlock:
2354 unix_state_unlock(sk);
2355 break;
2356 }
2357
2358 while (skip >= unix_skb_len(skb)) {
2359 skip -= unix_skb_len(skb);
2360 last = skb;
2361 last_len = skb->len;
2362 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2363 if (!skb)
2364 goto again;
2365 }
2366
2367 unix_state_unlock(sk);
2368
2369 if (check_creds) {
2370 /* Never glue messages from different writers */
2371 if (!unix_skb_scm_eq(skb, &scm))
2372 break;
2373 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2374 /* Copy credentials */
2375 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2376 unix_set_secdata(&scm, skb);
2377 check_creds = true;
2378 }
2379
2380 /* Copy address just once */
2381 if (state->msg && state->msg->msg_name) {
2382 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2383 state->msg->msg_name);
2384 unix_copy_addr(state->msg, skb->sk);
2385 sunaddr = NULL;
2386 }
2387
2388 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2389 skb_get(skb);
2390 chunk = state->recv_actor(skb, skip, chunk, state);
2391 drop_skb = !unix_skb_len(skb);
2392 /* skb is only safe to use if !drop_skb */
2393 consume_skb(skb);
2394 if (chunk < 0) {
2395 if (copied == 0)
2396 copied = -EFAULT;
2397 break;
2398 }
2399 copied += chunk;
2400 size -= chunk;
2401
2402 if (drop_skb) {
2403 /* the skb was touched by a concurrent reader;
2404 * we should not expect anything from this skb
2405 * anymore and assume it invalid - we can be
2406 * sure it was dropped from the socket queue
2407 *
2408 * let's report a short read
2409 */
2410 err = 0;
2411 break;
2412 }
2413
2414 /* Mark read part of skb as used */
2415 if (!(flags & MSG_PEEK)) {
2416 UNIXCB(skb).consumed += chunk;
2417
2418 sk_peek_offset_bwd(sk, chunk);
2419
2420 if (UNIXCB(skb).fp)
2421 unix_detach_fds(&scm, skb);
2422
2423 if (unix_skb_len(skb))
2424 break;
2425
2426 skb_unlink(skb, &sk->sk_receive_queue);
2427 consume_skb(skb);
2428
2429 if (scm.fp)
2430 break;
2431 } else {
2432 /* It is questionable, see note in unix_dgram_recvmsg.
2433 */
2434 if (UNIXCB(skb).fp)
2435 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2436
2437 sk_peek_offset_fwd(sk, chunk);
2438
2439 if (UNIXCB(skb).fp)
2440 break;
2441
2442 skip = 0;
2443 last = skb;
2444 last_len = skb->len;
2445 unix_state_lock(sk);
2446 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2447 if (skb)
2448 goto again;
2449 unix_state_unlock(sk);
2450 break;
2451 }
2452 } while (size);
2453
2454 mutex_unlock(&u->iolock);
2455 if (state->msg)
2456 scm_recv(sock, state->msg, &scm, flags);
2457 else
2458 scm_destroy(&scm);
2459out:
2460 return copied ? : err;
2461}
2462
2463static int unix_stream_read_actor(struct sk_buff *skb,
2464 int skip, int chunk,
2465 struct unix_stream_read_state *state)
2466{
2467 int ret;
2468
2469 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2470 state->msg, chunk);
2471 return ret ?: chunk;
2472}
2473
2474static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2475 size_t size, int flags)
2476{
2477 struct unix_stream_read_state state = {
2478 .recv_actor = unix_stream_read_actor,
2479 .socket = sock,
2480 .msg = msg,
2481 .size = size,
2482 .flags = flags
2483 };
2484
2485 return unix_stream_read_generic(&state, true);
2486}
2487
2488static int unix_stream_splice_actor(struct sk_buff *skb,
2489 int skip, int chunk,
2490 struct unix_stream_read_state *state)
2491{
2492 return skb_splice_bits(skb, state->socket->sk,
2493 UNIXCB(skb).consumed + skip,
2494 state->pipe, chunk, state->splice_flags);
2495}
2496
2497static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2498 struct pipe_inode_info *pipe,
2499 size_t size, unsigned int flags)
2500{
2501 struct unix_stream_read_state state = {
2502 .recv_actor = unix_stream_splice_actor,
2503 .socket = sock,
2504 .pipe = pipe,
2505 .size = size,
2506 .splice_flags = flags,
2507 };
2508
2509 if (unlikely(*ppos))
2510 return -ESPIPE;
2511
2512 if (sock->file->f_flags & O_NONBLOCK ||
2513 flags & SPLICE_F_NONBLOCK)
2514 state.flags = MSG_DONTWAIT;
2515
2516 return unix_stream_read_generic(&state, false);
2517}
2518
2519static int unix_shutdown(struct socket *sock, int mode)
2520{
2521 struct sock *sk = sock->sk;
2522 struct sock *other;
2523
2524 if (mode < SHUT_RD || mode > SHUT_RDWR)
2525 return -EINVAL;
2526 /* This maps:
2527 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2528 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2529 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2530 */
2531 ++mode;
2532
2533 unix_state_lock(sk);
2534 sk->sk_shutdown |= mode;
2535 other = unix_peer(sk);
2536 if (other)
2537 sock_hold(other);
2538 unix_state_unlock(sk);
2539 sk->sk_state_change(sk);
2540
2541 if (other &&
2542 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2543
2544 int peer_mode = 0;
2545
2546 if (mode&RCV_SHUTDOWN)
2547 peer_mode |= SEND_SHUTDOWN;
2548 if (mode&SEND_SHUTDOWN)
2549 peer_mode |= RCV_SHUTDOWN;
2550 unix_state_lock(other);
2551 other->sk_shutdown |= peer_mode;
2552 unix_state_unlock(other);
2553 other->sk_state_change(other);
2554 if (peer_mode == SHUTDOWN_MASK)
2555 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2556 else if (peer_mode & RCV_SHUTDOWN)
2557 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2558 }
2559 if (other)
2560 sock_put(other);
2561
2562 return 0;
2563}
2564
2565long unix_inq_len(struct sock *sk)
2566{
2567 struct sk_buff *skb;
2568 long amount = 0;
2569
2570 if (sk->sk_state == TCP_LISTEN)
2571 return -EINVAL;
2572
2573 spin_lock(&sk->sk_receive_queue.lock);
2574 if (sk->sk_type == SOCK_STREAM ||
2575 sk->sk_type == SOCK_SEQPACKET) {
2576 skb_queue_walk(&sk->sk_receive_queue, skb)
2577 amount += unix_skb_len(skb);
2578 } else {
2579 skb = skb_peek(&sk->sk_receive_queue);
2580 if (skb)
2581 amount = skb->len;
2582 }
2583 spin_unlock(&sk->sk_receive_queue.lock);
2584
2585 return amount;
2586}
2587EXPORT_SYMBOL_GPL(unix_inq_len);
2588
2589long unix_outq_len(struct sock *sk)
2590{
2591 return sk_wmem_alloc_get(sk);
2592}
2593EXPORT_SYMBOL_GPL(unix_outq_len);
2594
2595static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2596{
2597 struct sock *sk = sock->sk;
2598 long amount = 0;
2599 int err;
2600
2601 switch (cmd) {
2602 case SIOCOUTQ:
2603 amount = unix_outq_len(sk);
2604 err = put_user(amount, (int __user *)arg);
2605 break;
2606 case SIOCINQ:
2607 amount = unix_inq_len(sk);
2608 if (amount < 0)
2609 err = amount;
2610 else
2611 err = put_user(amount, (int __user *)arg);
2612 break;
2613 default:
2614 err = -ENOIOCTLCMD;
2615 break;
2616 }
2617 return err;
2618}
2619
2620static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2621{
2622 struct sock *sk = sock->sk;
2623 unsigned int mask;
2624
2625 sock_poll_wait(file, sk_sleep(sk), wait);
2626 mask = 0;
2627
2628 /* exceptional events? */
2629 if (sk->sk_err)
2630 mask |= POLLERR;
2631 if (sk->sk_shutdown == SHUTDOWN_MASK)
2632 mask |= POLLHUP;
2633 if (sk->sk_shutdown & RCV_SHUTDOWN)
2634 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2635
2636 /* readable? */
2637 if (!skb_queue_empty(&sk->sk_receive_queue))
2638 mask |= POLLIN | POLLRDNORM;
2639
2640 /* Connection-based need to check for termination and startup */
2641 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2642 sk->sk_state == TCP_CLOSE)
2643 mask |= POLLHUP;
2644
2645 /*
2646 * we set writable also when the other side has shut down the
2647 * connection. This prevents stuck sockets.
2648 */
2649 if (unix_writable(sk))
2650 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2651
2652 return mask;
2653}
2654
2655static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2656 poll_table *wait)
2657{
2658 struct sock *sk = sock->sk, *other;
2659 unsigned int mask, writable;
2660
2661 sock_poll_wait(file, sk_sleep(sk), wait);
2662 mask = 0;
2663
2664 /* exceptional events? */
2665 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2666 mask |= POLLERR |
2667 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2668
2669 if (sk->sk_shutdown & RCV_SHUTDOWN)
2670 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2671 if (sk->sk_shutdown == SHUTDOWN_MASK)
2672 mask |= POLLHUP;
2673
2674 /* readable? */
2675 if (!skb_queue_empty(&sk->sk_receive_queue))
2676 mask |= POLLIN | POLLRDNORM;
2677
2678 /* Connection-based need to check for termination and startup */
2679 if (sk->sk_type == SOCK_SEQPACKET) {
2680 if (sk->sk_state == TCP_CLOSE)
2681 mask |= POLLHUP;
2682 /* connection hasn't started yet? */
2683 if (sk->sk_state == TCP_SYN_SENT)
2684 return mask;
2685 }
2686
2687 /* No write status requested, avoid expensive OUT tests. */
2688 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2689 return mask;
2690
2691 writable = unix_writable(sk);
2692 if (writable) {
2693 unix_state_lock(sk);
2694
2695 other = unix_peer(sk);
2696 if (other && unix_peer(other) != sk &&
2697 unix_recvq_full(other) &&
2698 unix_dgram_peer_wake_me(sk, other))
2699 writable = 0;
2700
2701 unix_state_unlock(sk);
2702 }
2703
2704 if (writable)
2705 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2706 else
2707 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2708
2709 return mask;
2710}
2711
2712#ifdef CONFIG_PROC_FS
2713
2714#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2715
2716#define get_bucket(x) ((x) >> BUCKET_SPACE)
2717#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2718#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2719
2720static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2721{
2722 unsigned long offset = get_offset(*pos);
2723 unsigned long bucket = get_bucket(*pos);
2724 struct sock *sk;
2725 unsigned long count = 0;
2726
2727 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2728 if (sock_net(sk) != seq_file_net(seq))
2729 continue;
2730 if (++count == offset)
2731 break;
2732 }
2733
2734 return sk;
2735}
2736
2737static struct sock *unix_next_socket(struct seq_file *seq,
2738 struct sock *sk,
2739 loff_t *pos)
2740{
2741 unsigned long bucket;
2742
2743 while (sk > (struct sock *)SEQ_START_TOKEN) {
2744 sk = sk_next(sk);
2745 if (!sk)
2746 goto next_bucket;
2747 if (sock_net(sk) == seq_file_net(seq))
2748 return sk;
2749 }
2750
2751 do {
2752 sk = unix_from_bucket(seq, pos);
2753 if (sk)
2754 return sk;
2755
2756next_bucket:
2757 bucket = get_bucket(*pos) + 1;
2758 *pos = set_bucket_offset(bucket, 1);
2759 } while (bucket < ARRAY_SIZE(unix_socket_table));
2760
2761 return NULL;
2762}
2763
2764static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2765 __acquires(unix_table_lock)
2766{
2767 spin_lock(&unix_table_lock);
2768
2769 if (!*pos)
2770 return SEQ_START_TOKEN;
2771
2772 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2773 return NULL;
2774
2775 return unix_next_socket(seq, NULL, pos);
2776}
2777
2778static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2779{
2780 ++*pos;
2781 return unix_next_socket(seq, v, pos);
2782}
2783
2784static void unix_seq_stop(struct seq_file *seq, void *v)
2785 __releases(unix_table_lock)
2786{
2787 spin_unlock(&unix_table_lock);
2788}
2789
2790static int unix_seq_show(struct seq_file *seq, void *v)
2791{
2792
2793 if (v == SEQ_START_TOKEN)
2794 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2795 "Inode Path\n");
2796 else {
2797 struct sock *s = v;
2798 struct unix_sock *u = unix_sk(s);
2799 unix_state_lock(s);
2800
2801 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2802 s,
2803 atomic_read(&s->sk_refcnt),
2804 0,
2805 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2806 s->sk_type,
2807 s->sk_socket ?
2808 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2809 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2810 sock_i_ino(s));
2811
2812 if (u->addr) {
2813 int i, len;
2814 seq_putc(seq, ' ');
2815
2816 i = 0;
2817 len = u->addr->len - sizeof(short);
2818 if (!UNIX_ABSTRACT(s))
2819 len--;
2820 else {
2821 seq_putc(seq, '@');
2822 i++;
2823 }
2824 for ( ; i < len; i++)
2825 seq_putc(seq, u->addr->name->sun_path[i] ?:
2826 '@');
2827 }
2828 unix_state_unlock(s);
2829 seq_putc(seq, '\n');
2830 }
2831
2832 return 0;
2833}
2834
2835static const struct seq_operations unix_seq_ops = {
2836 .start = unix_seq_start,
2837 .next = unix_seq_next,
2838 .stop = unix_seq_stop,
2839 .show = unix_seq_show,
2840};
2841
2842static int unix_seq_open(struct inode *inode, struct file *file)
2843{
2844 return seq_open_net(inode, file, &unix_seq_ops,
2845 sizeof(struct seq_net_private));
2846}
2847
2848static const struct file_operations unix_seq_fops = {
2849 .owner = THIS_MODULE,
2850 .open = unix_seq_open,
2851 .read = seq_read,
2852 .llseek = seq_lseek,
2853 .release = seq_release_net,
2854};
2855
2856#endif
2857
2858static const struct net_proto_family unix_family_ops = {
2859 .family = PF_UNIX,
2860 .create = unix_create,
2861 .owner = THIS_MODULE,
2862};
2863
2864
2865static int __net_init unix_net_init(struct net *net)
2866{
2867 int error = -ENOMEM;
2868
2869 net->unx.sysctl_max_dgram_qlen = 10;
2870 if (unix_sysctl_register(net))
2871 goto out;
2872
2873#ifdef CONFIG_PROC_FS
2874 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2875 unix_sysctl_unregister(net);
2876 goto out;
2877 }
2878#endif
2879 error = 0;
2880out:
2881 return error;
2882}
2883
2884static void __net_exit unix_net_exit(struct net *net)
2885{
2886 unix_sysctl_unregister(net);
2887 remove_proc_entry("unix", net->proc_net);
2888}
2889
2890static struct pernet_operations unix_net_ops = {
2891 .init = unix_net_init,
2892 .exit = unix_net_exit,
2893};
2894
2895static int __init af_unix_init(void)
2896{
2897 int rc = -1;
2898
2899 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2900
2901 rc = proto_register(&unix_proto, 1);
2902 if (rc != 0) {
2903 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2904 goto out;
2905 }
2906
2907 sock_register(&unix_family_ops);
2908 register_pernet_subsys(&unix_net_ops);
2909out:
2910 return rc;
2911}
2912
2913static void __exit af_unix_exit(void)
2914{
2915 sock_unregister(PF_UNIX);
2916 proto_unregister(&unix_proto);
2917 unregister_pernet_subsys(&unix_net_ops);
2918}
2919
2920/* Earlier than device_initcall() so that other drivers invoking
2921 request_module() don't end up in a loop when modprobe tries
2922 to use a UNIX socket. But later than subsys_initcall() because
2923 we depend on stuff initialised there */
2924fs_initcall(af_unix_init);
2925module_exit(af_unix_exit);
2926
2927MODULE_LICENSE("GPL");
2928MODULE_ALIAS_NETPROTO(PF_UNIX);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET4: Implementation of BSD Unix domain sockets.
4 *
5 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Fixes:
8 * Linus Torvalds : Assorted bug cures.
9 * Niibe Yutaka : async I/O support.
10 * Carsten Paeth : PF_UNIX check, address fixes.
11 * Alan Cox : Limit size of allocated blocks.
12 * Alan Cox : Fixed the stupid socketpair bug.
13 * Alan Cox : BSD compatibility fine tuning.
14 * Alan Cox : Fixed a bug in connect when interrupted.
15 * Alan Cox : Sorted out a proper draft version of
16 * file descriptor passing hacked up from
17 * Mike Shaver's work.
18 * Marty Leisner : Fixes to fd passing
19 * Nick Nevin : recvmsg bugfix.
20 * Alan Cox : Started proper garbage collector
21 * Heiko EiBfeldt : Missing verify_area check
22 * Alan Cox : Started POSIXisms
23 * Andreas Schwab : Replace inode by dentry for proper
24 * reference counting
25 * Kirk Petersen : Made this a module
26 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
27 * Lots of bug fixes.
28 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
29 * by above two patches.
30 * Andrea Arcangeli : If possible we block in connect(2)
31 * if the max backlog of the listen socket
32 * is been reached. This won't break
33 * old apps and it will avoid huge amount
34 * of socks hashed (this for unix_gc()
35 * performances reasons).
36 * Security fix that limits the max
37 * number of socks to 2*max_files and
38 * the number of skb queueable in the
39 * dgram receiver.
40 * Artur Skawina : Hash function optimizations
41 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
42 * Malcolm Beattie : Set peercred for socketpair
43 * Michal Ostrowski : Module initialization cleanup.
44 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
45 * the core infrastructure is doing that
46 * for all net proto families now (2.5.69+)
47 *
48 * Known differences from reference BSD that was tested:
49 *
50 * [TO FIX]
51 * ECONNREFUSED is not returned from one end of a connected() socket to the
52 * other the moment one end closes.
53 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
54 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
55 * [NOT TO FIX]
56 * accept() returns a path name even if the connecting socket has closed
57 * in the meantime (BSD loses the path and gives up).
58 * accept() returns 0 length path for an unbound connector. BSD returns 16
59 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
60 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
61 * BSD af_unix apparently has connect forgetting to block properly.
62 * (need to check this with the POSIX spec in detail)
63 *
64 * Differences from 2.0.0-11-... (ANK)
65 * Bug fixes and improvements.
66 * - client shutdown killed server socket.
67 * - removed all useless cli/sti pairs.
68 *
69 * Semantic changes/extensions.
70 * - generic control message passing.
71 * - SCM_CREDENTIALS control message.
72 * - "Abstract" (not FS based) socket bindings.
73 * Abstract names are sequences of bytes (not zero terminated)
74 * started by 0, so that this name space does not intersect
75 * with BSD names.
76 */
77
78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
80#include <linux/module.h>
81#include <linux/kernel.h>
82#include <linux/signal.h>
83#include <linux/sched/signal.h>
84#include <linux/errno.h>
85#include <linux/string.h>
86#include <linux/stat.h>
87#include <linux/dcache.h>
88#include <linux/namei.h>
89#include <linux/socket.h>
90#include <linux/un.h>
91#include <linux/fcntl.h>
92#include <linux/filter.h>
93#include <linux/termios.h>
94#include <linux/sockios.h>
95#include <linux/net.h>
96#include <linux/in.h>
97#include <linux/fs.h>
98#include <linux/slab.h>
99#include <linux/uaccess.h>
100#include <linux/skbuff.h>
101#include <linux/netdevice.h>
102#include <net/net_namespace.h>
103#include <net/sock.h>
104#include <net/tcp_states.h>
105#include <net/af_unix.h>
106#include <linux/proc_fs.h>
107#include <linux/seq_file.h>
108#include <net/scm.h>
109#include <linux/init.h>
110#include <linux/poll.h>
111#include <linux/rtnetlink.h>
112#include <linux/mount.h>
113#include <net/checksum.h>
114#include <linux/security.h>
115#include <linux/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);