Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET4: Implementation of BSD Unix domain sockets.
4 *
5 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Fixes:
8 * Linus Torvalds : Assorted bug cures.
9 * Niibe Yutaka : async I/O support.
10 * Carsten Paeth : PF_UNIX check, address fixes.
11 * Alan Cox : Limit size of allocated blocks.
12 * Alan Cox : Fixed the stupid socketpair bug.
13 * Alan Cox : BSD compatibility fine tuning.
14 * Alan Cox : Fixed a bug in connect when interrupted.
15 * Alan Cox : Sorted out a proper draft version of
16 * file descriptor passing hacked up from
17 * Mike Shaver's work.
18 * Marty Leisner : Fixes to fd passing
19 * Nick Nevin : recvmsg bugfix.
20 * Alan Cox : Started proper garbage collector
21 * Heiko EiBfeldt : Missing verify_area check
22 * Alan Cox : Started POSIXisms
23 * Andreas Schwab : Replace inode by dentry for proper
24 * reference counting
25 * Kirk Petersen : Made this a module
26 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
27 * Lots of bug fixes.
28 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
29 * by above two patches.
30 * Andrea Arcangeli : If possible we block in connect(2)
31 * if the max backlog of the listen socket
32 * is been reached. This won't break
33 * old apps and it will avoid huge amount
34 * of socks hashed (this for unix_gc()
35 * performances reasons).
36 * Security fix that limits the max
37 * number of socks to 2*max_files and
38 * the number of skb queueable in the
39 * dgram receiver.
40 * Artur Skawina : Hash function optimizations
41 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
42 * Malcolm Beattie : Set peercred for socketpair
43 * Michal Ostrowski : Module initialization cleanup.
44 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
45 * the core infrastructure is doing that
46 * for all net proto families now (2.5.69+)
47 *
48 * Known differences from reference BSD that was tested:
49 *
50 * [TO FIX]
51 * ECONNREFUSED is not returned from one end of a connected() socket to the
52 * other the moment one end closes.
53 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
54 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
55 * [NOT TO FIX]
56 * accept() returns a path name even if the connecting socket has closed
57 * in the meantime (BSD loses the path and gives up).
58 * accept() returns 0 length path for an unbound connector. BSD returns 16
59 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
60 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
61 * BSD af_unix apparently has connect forgetting to block properly.
62 * (need to check this with the POSIX spec in detail)
63 *
64 * Differences from 2.0.0-11-... (ANK)
65 * Bug fixes and improvements.
66 * - client shutdown killed server socket.
67 * - removed all useless cli/sti pairs.
68 *
69 * Semantic changes/extensions.
70 * - generic control message passing.
71 * - SCM_CREDENTIALS control message.
72 * - "Abstract" (not FS based) socket bindings.
73 * Abstract names are sequences of bytes (not zero terminated)
74 * started by 0, so that this name space does not intersect
75 * with BSD names.
76 */
77
78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
80#include <linux/module.h>
81#include <linux/kernel.h>
82#include <linux/signal.h>
83#include <linux/sched/signal.h>
84#include <linux/errno.h>
85#include <linux/string.h>
86#include <linux/stat.h>
87#include <linux/dcache.h>
88#include <linux/namei.h>
89#include <linux/socket.h>
90#include <linux/un.h>
91#include <linux/fcntl.h>
92#include <linux/filter.h>
93#include <linux/termios.h>
94#include <linux/sockios.h>
95#include <linux/net.h>
96#include <linux/in.h>
97#include <linux/fs.h>
98#include <linux/slab.h>
99#include <linux/uaccess.h>
100#include <linux/skbuff.h>
101#include <linux/netdevice.h>
102#include <net/net_namespace.h>
103#include <net/sock.h>
104#include <net/tcp_states.h>
105#include <net/af_unix.h>
106#include <linux/proc_fs.h>
107#include <linux/seq_file.h>
108#include <net/scm.h>
109#include <linux/init.h>
110#include <linux/poll.h>
111#include <linux/rtnetlink.h>
112#include <linux/mount.h>
113#include <net/checksum.h>
114#include <linux/security.h>
115#include <linux/splice.h>
116#include <linux/freezer.h>
117#include <linux/file.h>
118#include <linux/btf_ids.h>
119#include <linux/bpf-cgroup.h>
120
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
214static inline int unix_our_peer(struct sock *sk, struct sock *osk)
215{
216 return unix_peer(osk) == sk;
217}
218
219static inline int unix_may_send(struct sock *sk, struct sock *osk)
220{
221 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
222}
223
224static inline int unix_recvq_full(const struct sock *sk)
225{
226 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
227}
228
229static inline int unix_recvq_full_lockless(const struct sock *sk)
230{
231 return skb_queue_len_lockless(&sk->sk_receive_queue) >
232 READ_ONCE(sk->sk_max_ack_backlog);
233}
234
235struct sock *unix_peer_get(struct sock *s)
236{
237 struct sock *peer;
238
239 unix_state_lock(s);
240 peer = unix_peer(s);
241 if (peer)
242 sock_hold(peer);
243 unix_state_unlock(s);
244 return peer;
245}
246EXPORT_SYMBOL_GPL(unix_peer_get);
247
248static struct unix_address *unix_create_addr(struct sockaddr_un *sunaddr,
249 int addr_len)
250{
251 struct unix_address *addr;
252
253 addr = kmalloc(sizeof(*addr) + addr_len, GFP_KERNEL);
254 if (!addr)
255 return NULL;
256
257 refcount_set(&addr->refcnt, 1);
258 addr->len = addr_len;
259 memcpy(addr->name, sunaddr, addr_len);
260
261 return addr;
262}
263
264static inline void unix_release_addr(struct unix_address *addr)
265{
266 if (refcount_dec_and_test(&addr->refcnt))
267 kfree(addr);
268}
269
270/*
271 * Check unix socket name:
272 * - should be not zero length.
273 * - if started by not zero, should be NULL terminated (FS object)
274 * - if started by zero, it is abstract name.
275 */
276
277static int unix_validate_addr(struct sockaddr_un *sunaddr, int addr_len)
278{
279 if (addr_len <= offsetof(struct sockaddr_un, sun_path) ||
280 addr_len > sizeof(*sunaddr))
281 return -EINVAL;
282
283 if (sunaddr->sun_family != AF_UNIX)
284 return -EINVAL;
285
286 return 0;
287}
288
289static int unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
290{
291 struct sockaddr_storage *addr = (struct sockaddr_storage *)sunaddr;
292 short offset = offsetof(struct sockaddr_storage, __data);
293
294 BUILD_BUG_ON(offset != offsetof(struct sockaddr_un, sun_path));
295
296 /* This may look like an off by one error but it is a bit more
297 * subtle. 108 is the longest valid AF_UNIX path for a binding.
298 * sun_path[108] doesn't as such exist. However in kernel space
299 * we are guaranteed that it is a valid memory location in our
300 * kernel address buffer because syscall functions always pass
301 * a pointer of struct sockaddr_storage which has a bigger buffer
302 * than 108. Also, we must terminate sun_path for strlen() in
303 * getname_kernel().
304 */
305 addr->__data[addr_len - offset] = 0;
306
307 /* Don't pass sunaddr->sun_path to strlen(). Otherwise, 108 will
308 * cause panic if CONFIG_FORTIFY_SOURCE=y. Let __fortify_strlen()
309 * know the actual buffer.
310 */
311 return strlen(addr->__data) + offset + 1;
312}
313
314static void __unix_remove_socket(struct sock *sk)
315{
316 sk_del_node_init(sk);
317}
318
319static void __unix_insert_socket(struct net *net, struct sock *sk)
320{
321 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
322 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
323}
324
325static void __unix_set_addr_hash(struct net *net, struct sock *sk,
326 struct unix_address *addr, unsigned int hash)
327{
328 __unix_remove_socket(sk);
329 smp_store_release(&unix_sk(sk)->addr, addr);
330
331 sk->sk_hash = hash;
332 __unix_insert_socket(net, sk);
333}
334
335static void unix_remove_socket(struct net *net, struct sock *sk)
336{
337 spin_lock(&net->unx.table.locks[sk->sk_hash]);
338 __unix_remove_socket(sk);
339 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
340}
341
342static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
343{
344 spin_lock(&net->unx.table.locks[sk->sk_hash]);
345 __unix_insert_socket(net, sk);
346 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
347}
348
349static void unix_insert_bsd_socket(struct sock *sk)
350{
351 spin_lock(&bsd_socket_locks[sk->sk_hash]);
352 sk_add_bind_node(sk, &bsd_socket_buckets[sk->sk_hash]);
353 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
354}
355
356static void unix_remove_bsd_socket(struct sock *sk)
357{
358 if (!hlist_unhashed(&sk->sk_bind_node)) {
359 spin_lock(&bsd_socket_locks[sk->sk_hash]);
360 __sk_del_bind_node(sk);
361 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
362
363 sk_node_init(&sk->sk_bind_node);
364 }
365}
366
367static struct sock *__unix_find_socket_byname(struct net *net,
368 struct sockaddr_un *sunname,
369 int len, unsigned int hash)
370{
371 struct sock *s;
372
373 sk_for_each(s, &net->unx.table.buckets[hash]) {
374 struct unix_sock *u = unix_sk(s);
375
376 if (u->addr->len == len &&
377 !memcmp(u->addr->name, sunname, len))
378 return s;
379 }
380 return NULL;
381}
382
383static inline struct sock *unix_find_socket_byname(struct net *net,
384 struct sockaddr_un *sunname,
385 int len, unsigned int hash)
386{
387 struct sock *s;
388
389 spin_lock(&net->unx.table.locks[hash]);
390 s = __unix_find_socket_byname(net, sunname, len, hash);
391 if (s)
392 sock_hold(s);
393 spin_unlock(&net->unx.table.locks[hash]);
394 return s;
395}
396
397static struct sock *unix_find_socket_byinode(struct inode *i)
398{
399 unsigned int hash = unix_bsd_hash(i);
400 struct sock *s;
401
402 spin_lock(&bsd_socket_locks[hash]);
403 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
404 struct dentry *dentry = unix_sk(s)->path.dentry;
405
406 if (dentry && d_backing_inode(dentry) == i) {
407 sock_hold(s);
408 spin_unlock(&bsd_socket_locks[hash]);
409 return s;
410 }
411 }
412 spin_unlock(&bsd_socket_locks[hash]);
413 return NULL;
414}
415
416/* Support code for asymmetrically connected dgram sockets
417 *
418 * If a datagram socket is connected to a socket not itself connected
419 * to the first socket (eg, /dev/log), clients may only enqueue more
420 * messages if the present receive queue of the server socket is not
421 * "too large". This means there's a second writeability condition
422 * poll and sendmsg need to test. The dgram recv code will do a wake
423 * up on the peer_wait wait queue of a socket upon reception of a
424 * datagram which needs to be propagated to sleeping would-be writers
425 * since these might not have sent anything so far. This can't be
426 * accomplished via poll_wait because the lifetime of the server
427 * socket might be less than that of its clients if these break their
428 * association with it or if the server socket is closed while clients
429 * are still connected to it and there's no way to inform "a polling
430 * implementation" that it should let go of a certain wait queue
431 *
432 * In order to propagate a wake up, a wait_queue_entry_t of the client
433 * socket is enqueued on the peer_wait queue of the server socket
434 * whose wake function does a wake_up on the ordinary client socket
435 * wait queue. This connection is established whenever a write (or
436 * poll for write) hit the flow control condition and broken when the
437 * association to the server socket is dissolved or after a wake up
438 * was relayed.
439 */
440
441static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
442 void *key)
443{
444 struct unix_sock *u;
445 wait_queue_head_t *u_sleep;
446
447 u = container_of(q, struct unix_sock, peer_wake);
448
449 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
450 q);
451 u->peer_wake.private = NULL;
452
453 /* relaying can only happen while the wq still exists */
454 u_sleep = sk_sleep(&u->sk);
455 if (u_sleep)
456 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
457
458 return 0;
459}
460
461static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
462{
463 struct unix_sock *u, *u_other;
464 int rc;
465
466 u = unix_sk(sk);
467 u_other = unix_sk(other);
468 rc = 0;
469 spin_lock(&u_other->peer_wait.lock);
470
471 if (!u->peer_wake.private) {
472 u->peer_wake.private = other;
473 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
474
475 rc = 1;
476 }
477
478 spin_unlock(&u_other->peer_wait.lock);
479 return rc;
480}
481
482static void unix_dgram_peer_wake_disconnect(struct sock *sk,
483 struct sock *other)
484{
485 struct unix_sock *u, *u_other;
486
487 u = unix_sk(sk);
488 u_other = unix_sk(other);
489 spin_lock(&u_other->peer_wait.lock);
490
491 if (u->peer_wake.private == other) {
492 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
493 u->peer_wake.private = NULL;
494 }
495
496 spin_unlock(&u_other->peer_wait.lock);
497}
498
499static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
500 struct sock *other)
501{
502 unix_dgram_peer_wake_disconnect(sk, other);
503 wake_up_interruptible_poll(sk_sleep(sk),
504 EPOLLOUT |
505 EPOLLWRNORM |
506 EPOLLWRBAND);
507}
508
509/* preconditions:
510 * - unix_peer(sk) == other
511 * - association is stable
512 */
513static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
514{
515 int connected;
516
517 connected = unix_dgram_peer_wake_connect(sk, other);
518
519 /* If other is SOCK_DEAD, we want to make sure we signal
520 * POLLOUT, such that a subsequent write() can get a
521 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
522 * to other and its full, we will hang waiting for POLLOUT.
523 */
524 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
525 return 1;
526
527 if (connected)
528 unix_dgram_peer_wake_disconnect(sk, other);
529
530 return 0;
531}
532
533static int unix_writable(const struct sock *sk)
534{
535 return sk->sk_state != TCP_LISTEN &&
536 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
537}
538
539static void unix_write_space(struct sock *sk)
540{
541 struct socket_wq *wq;
542
543 rcu_read_lock();
544 if (unix_writable(sk)) {
545 wq = rcu_dereference(sk->sk_wq);
546 if (skwq_has_sleeper(wq))
547 wake_up_interruptible_sync_poll(&wq->wait,
548 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
549 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
550 }
551 rcu_read_unlock();
552}
553
554/* When dgram socket disconnects (or changes its peer), we clear its receive
555 * queue of packets arrived from previous peer. First, it allows to do
556 * flow control based only on wmem_alloc; second, sk connected to peer
557 * may receive messages only from that peer. */
558static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
559{
560 if (!skb_queue_empty(&sk->sk_receive_queue)) {
561 skb_queue_purge(&sk->sk_receive_queue);
562 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
563
564 /* If one link of bidirectional dgram pipe is disconnected,
565 * we signal error. Messages are lost. Do not make this,
566 * when peer was not connected to us.
567 */
568 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
569 WRITE_ONCE(other->sk_err, ECONNRESET);
570 sk_error_report(other);
571 }
572 }
573 other->sk_state = TCP_CLOSE;
574}
575
576static void unix_sock_destructor(struct sock *sk)
577{
578 struct unix_sock *u = unix_sk(sk);
579
580 skb_queue_purge(&sk->sk_receive_queue);
581
582 DEBUG_NET_WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
583 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
584 DEBUG_NET_WARN_ON_ONCE(sk->sk_socket);
585 if (!sock_flag(sk, SOCK_DEAD)) {
586 pr_info("Attempt to release alive unix socket: %p\n", sk);
587 return;
588 }
589
590 if (u->addr)
591 unix_release_addr(u->addr);
592
593 atomic_long_dec(&unix_nr_socks);
594 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
595#ifdef UNIX_REFCNT_DEBUG
596 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
597 atomic_long_read(&unix_nr_socks));
598#endif
599}
600
601static void unix_release_sock(struct sock *sk, int embrion)
602{
603 struct unix_sock *u = unix_sk(sk);
604 struct sock *skpair;
605 struct sk_buff *skb;
606 struct path path;
607 int state;
608
609 unix_remove_socket(sock_net(sk), sk);
610 unix_remove_bsd_socket(sk);
611
612 /* Clear state */
613 unix_state_lock(sk);
614 sock_orphan(sk);
615 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
616 path = u->path;
617 u->path.dentry = NULL;
618 u->path.mnt = NULL;
619 state = sk->sk_state;
620 sk->sk_state = TCP_CLOSE;
621
622 skpair = unix_peer(sk);
623 unix_peer(sk) = NULL;
624
625 unix_state_unlock(sk);
626
627#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
628 if (u->oob_skb) {
629 kfree_skb(u->oob_skb);
630 u->oob_skb = NULL;
631 }
632#endif
633
634 wake_up_interruptible_all(&u->peer_wait);
635
636 if (skpair != NULL) {
637 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
638 unix_state_lock(skpair);
639 /* No more writes */
640 WRITE_ONCE(skpair->sk_shutdown, SHUTDOWN_MASK);
641 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
642 WRITE_ONCE(skpair->sk_err, ECONNRESET);
643 unix_state_unlock(skpair);
644 skpair->sk_state_change(skpair);
645 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
646 }
647
648 unix_dgram_peer_wake_disconnect(sk, skpair);
649 sock_put(skpair); /* It may now die */
650 }
651
652 /* Try to flush out this socket. Throw out buffers at least */
653
654 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
655 if (state == TCP_LISTEN)
656 unix_release_sock(skb->sk, 1);
657 /* passed fds are erased in the kfree_skb hook */
658 UNIXCB(skb).consumed = skb->len;
659 kfree_skb(skb);
660 }
661
662 if (path.dentry)
663 path_put(&path);
664
665 sock_put(sk);
666
667 /* ---- Socket is dead now and most probably destroyed ---- */
668
669 /*
670 * Fixme: BSD difference: In BSD all sockets connected to us get
671 * ECONNRESET and we die on the spot. In Linux we behave
672 * like files and pipes do and wait for the last
673 * dereference.
674 *
675 * Can't we simply set sock->err?
676 *
677 * What the above comment does talk about? --ANK(980817)
678 */
679
680 if (READ_ONCE(unix_tot_inflight))
681 unix_gc(); /* Garbage collect fds */
682}
683
684static void init_peercred(struct sock *sk)
685{
686 const struct cred *old_cred;
687 struct pid *old_pid;
688
689 spin_lock(&sk->sk_peer_lock);
690 old_pid = sk->sk_peer_pid;
691 old_cred = sk->sk_peer_cred;
692 sk->sk_peer_pid = get_pid(task_tgid(current));
693 sk->sk_peer_cred = get_current_cred();
694 spin_unlock(&sk->sk_peer_lock);
695
696 put_pid(old_pid);
697 put_cred(old_cred);
698}
699
700static void copy_peercred(struct sock *sk, struct sock *peersk)
701{
702 const struct cred *old_cred;
703 struct pid *old_pid;
704
705 if (sk < peersk) {
706 spin_lock(&sk->sk_peer_lock);
707 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
708 } else {
709 spin_lock(&peersk->sk_peer_lock);
710 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
711 }
712 old_pid = sk->sk_peer_pid;
713 old_cred = sk->sk_peer_cred;
714 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
715 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
716
717 spin_unlock(&sk->sk_peer_lock);
718 spin_unlock(&peersk->sk_peer_lock);
719
720 put_pid(old_pid);
721 put_cred(old_cred);
722}
723
724static int unix_listen(struct socket *sock, int backlog)
725{
726 int err;
727 struct sock *sk = sock->sk;
728 struct unix_sock *u = unix_sk(sk);
729
730 err = -EOPNOTSUPP;
731 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
732 goto out; /* Only stream/seqpacket sockets accept */
733 err = -EINVAL;
734 if (!READ_ONCE(u->addr))
735 goto out; /* No listens on an unbound socket */
736 unix_state_lock(sk);
737 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
738 goto out_unlock;
739 if (backlog > sk->sk_max_ack_backlog)
740 wake_up_interruptible_all(&u->peer_wait);
741 sk->sk_max_ack_backlog = backlog;
742 sk->sk_state = TCP_LISTEN;
743 /* set credentials so connect can copy them */
744 init_peercred(sk);
745 err = 0;
746
747out_unlock:
748 unix_state_unlock(sk);
749out:
750 return err;
751}
752
753static int unix_release(struct socket *);
754static int unix_bind(struct socket *, struct sockaddr *, int);
755static int unix_stream_connect(struct socket *, struct sockaddr *,
756 int addr_len, int flags);
757static int unix_socketpair(struct socket *, struct socket *);
758static int unix_accept(struct socket *, struct socket *, int, bool);
759static int unix_getname(struct socket *, struct sockaddr *, int);
760static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
761static __poll_t unix_dgram_poll(struct file *, struct socket *,
762 poll_table *);
763static int unix_ioctl(struct socket *, unsigned int, unsigned long);
764#ifdef CONFIG_COMPAT
765static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
766#endif
767static int unix_shutdown(struct socket *, int);
768static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
769static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
770static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
771 struct pipe_inode_info *, size_t size,
772 unsigned int flags);
773static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
774static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
775static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
776static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
777static int unix_dgram_connect(struct socket *, struct sockaddr *,
778 int, int);
779static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
780static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
781 int);
782
783#ifdef CONFIG_PROC_FS
784static int unix_count_nr_fds(struct sock *sk)
785{
786 struct sk_buff *skb;
787 struct unix_sock *u;
788 int nr_fds = 0;
789
790 spin_lock(&sk->sk_receive_queue.lock);
791 skb = skb_peek(&sk->sk_receive_queue);
792 while (skb) {
793 u = unix_sk(skb->sk);
794 nr_fds += atomic_read(&u->scm_stat.nr_fds);
795 skb = skb_peek_next(skb, &sk->sk_receive_queue);
796 }
797 spin_unlock(&sk->sk_receive_queue.lock);
798
799 return nr_fds;
800}
801
802static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
803{
804 struct sock *sk = sock->sk;
805 unsigned char s_state;
806 struct unix_sock *u;
807 int nr_fds = 0;
808
809 if (sk) {
810 s_state = READ_ONCE(sk->sk_state);
811 u = unix_sk(sk);
812
813 /* SOCK_STREAM and SOCK_SEQPACKET sockets never change their
814 * sk_state after switching to TCP_ESTABLISHED or TCP_LISTEN.
815 * SOCK_DGRAM is ordinary. So, no lock is needed.
816 */
817 if (sock->type == SOCK_DGRAM || s_state == TCP_ESTABLISHED)
818 nr_fds = atomic_read(&u->scm_stat.nr_fds);
819 else if (s_state == TCP_LISTEN)
820 nr_fds = unix_count_nr_fds(sk);
821
822 seq_printf(m, "scm_fds: %u\n", nr_fds);
823 }
824}
825#else
826#define unix_show_fdinfo NULL
827#endif
828
829static const struct proto_ops unix_stream_ops = {
830 .family = PF_UNIX,
831 .owner = THIS_MODULE,
832 .release = unix_release,
833 .bind = unix_bind,
834 .connect = unix_stream_connect,
835 .socketpair = unix_socketpair,
836 .accept = unix_accept,
837 .getname = unix_getname,
838 .poll = unix_poll,
839 .ioctl = unix_ioctl,
840#ifdef CONFIG_COMPAT
841 .compat_ioctl = unix_compat_ioctl,
842#endif
843 .listen = unix_listen,
844 .shutdown = unix_shutdown,
845 .sendmsg = unix_stream_sendmsg,
846 .recvmsg = unix_stream_recvmsg,
847 .read_skb = unix_stream_read_skb,
848 .mmap = sock_no_mmap,
849 .splice_read = unix_stream_splice_read,
850 .set_peek_off = sk_set_peek_off,
851 .show_fdinfo = unix_show_fdinfo,
852};
853
854static const struct proto_ops unix_dgram_ops = {
855 .family = PF_UNIX,
856 .owner = THIS_MODULE,
857 .release = unix_release,
858 .bind = unix_bind,
859 .connect = unix_dgram_connect,
860 .socketpair = unix_socketpair,
861 .accept = sock_no_accept,
862 .getname = unix_getname,
863 .poll = unix_dgram_poll,
864 .ioctl = unix_ioctl,
865#ifdef CONFIG_COMPAT
866 .compat_ioctl = unix_compat_ioctl,
867#endif
868 .listen = sock_no_listen,
869 .shutdown = unix_shutdown,
870 .sendmsg = unix_dgram_sendmsg,
871 .read_skb = unix_read_skb,
872 .recvmsg = unix_dgram_recvmsg,
873 .mmap = sock_no_mmap,
874 .set_peek_off = sk_set_peek_off,
875 .show_fdinfo = unix_show_fdinfo,
876};
877
878static const struct proto_ops unix_seqpacket_ops = {
879 .family = PF_UNIX,
880 .owner = THIS_MODULE,
881 .release = unix_release,
882 .bind = unix_bind,
883 .connect = unix_stream_connect,
884 .socketpair = unix_socketpair,
885 .accept = unix_accept,
886 .getname = unix_getname,
887 .poll = unix_dgram_poll,
888 .ioctl = unix_ioctl,
889#ifdef CONFIG_COMPAT
890 .compat_ioctl = unix_compat_ioctl,
891#endif
892 .listen = unix_listen,
893 .shutdown = unix_shutdown,
894 .sendmsg = unix_seqpacket_sendmsg,
895 .recvmsg = unix_seqpacket_recvmsg,
896 .mmap = sock_no_mmap,
897 .set_peek_off = sk_set_peek_off,
898 .show_fdinfo = unix_show_fdinfo,
899};
900
901static void unix_close(struct sock *sk, long timeout)
902{
903 /* Nothing to do here, unix socket does not need a ->close().
904 * This is merely for sockmap.
905 */
906}
907
908static void unix_unhash(struct sock *sk)
909{
910 /* Nothing to do here, unix socket does not need a ->unhash().
911 * This is merely for sockmap.
912 */
913}
914
915static bool unix_bpf_bypass_getsockopt(int level, int optname)
916{
917 if (level == SOL_SOCKET) {
918 switch (optname) {
919 case SO_PEERPIDFD:
920 return true;
921 default:
922 return false;
923 }
924 }
925
926 return false;
927}
928
929struct proto unix_dgram_proto = {
930 .name = "UNIX",
931 .owner = THIS_MODULE,
932 .obj_size = sizeof(struct unix_sock),
933 .close = unix_close,
934 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
935#ifdef CONFIG_BPF_SYSCALL
936 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
937#endif
938};
939
940struct proto unix_stream_proto = {
941 .name = "UNIX-STREAM",
942 .owner = THIS_MODULE,
943 .obj_size = sizeof(struct unix_sock),
944 .close = unix_close,
945 .unhash = unix_unhash,
946 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
947#ifdef CONFIG_BPF_SYSCALL
948 .psock_update_sk_prot = unix_stream_bpf_update_proto,
949#endif
950};
951
952static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
953{
954 struct unix_sock *u;
955 struct sock *sk;
956 int err;
957
958 atomic_long_inc(&unix_nr_socks);
959 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
960 err = -ENFILE;
961 goto err;
962 }
963
964 if (type == SOCK_STREAM)
965 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
966 else /*dgram and seqpacket */
967 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
968
969 if (!sk) {
970 err = -ENOMEM;
971 goto err;
972 }
973
974 sock_init_data(sock, sk);
975
976 sk->sk_hash = unix_unbound_hash(sk);
977 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
978 sk->sk_write_space = unix_write_space;
979 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
980 sk->sk_destruct = unix_sock_destructor;
981 u = unix_sk(sk);
982 u->inflight = 0;
983 u->path.dentry = NULL;
984 u->path.mnt = NULL;
985 spin_lock_init(&u->lock);
986 INIT_LIST_HEAD(&u->link);
987 mutex_init(&u->iolock); /* single task reading lock */
988 mutex_init(&u->bindlock); /* single task binding lock */
989 init_waitqueue_head(&u->peer_wait);
990 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
991 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
992 unix_insert_unbound_socket(net, sk);
993
994 sock_prot_inuse_add(net, sk->sk_prot, 1);
995
996 return sk;
997
998err:
999 atomic_long_dec(&unix_nr_socks);
1000 return ERR_PTR(err);
1001}
1002
1003static int unix_create(struct net *net, struct socket *sock, int protocol,
1004 int kern)
1005{
1006 struct sock *sk;
1007
1008 if (protocol && protocol != PF_UNIX)
1009 return -EPROTONOSUPPORT;
1010
1011 sock->state = SS_UNCONNECTED;
1012
1013 switch (sock->type) {
1014 case SOCK_STREAM:
1015 sock->ops = &unix_stream_ops;
1016 break;
1017 /*
1018 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
1019 * nothing uses it.
1020 */
1021 case SOCK_RAW:
1022 sock->type = SOCK_DGRAM;
1023 fallthrough;
1024 case SOCK_DGRAM:
1025 sock->ops = &unix_dgram_ops;
1026 break;
1027 case SOCK_SEQPACKET:
1028 sock->ops = &unix_seqpacket_ops;
1029 break;
1030 default:
1031 return -ESOCKTNOSUPPORT;
1032 }
1033
1034 sk = unix_create1(net, sock, kern, sock->type);
1035 if (IS_ERR(sk))
1036 return PTR_ERR(sk);
1037
1038 return 0;
1039}
1040
1041static int unix_release(struct socket *sock)
1042{
1043 struct sock *sk = sock->sk;
1044
1045 if (!sk)
1046 return 0;
1047
1048 sk->sk_prot->close(sk, 0);
1049 unix_release_sock(sk, 0);
1050 sock->sk = NULL;
1051
1052 return 0;
1053}
1054
1055static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1056 int type)
1057{
1058 struct inode *inode;
1059 struct path path;
1060 struct sock *sk;
1061 int err;
1062
1063 unix_mkname_bsd(sunaddr, addr_len);
1064 err = kern_path(sunaddr->sun_path, LOOKUP_FOLLOW, &path);
1065 if (err)
1066 goto fail;
1067
1068 err = path_permission(&path, MAY_WRITE);
1069 if (err)
1070 goto path_put;
1071
1072 err = -ECONNREFUSED;
1073 inode = d_backing_inode(path.dentry);
1074 if (!S_ISSOCK(inode->i_mode))
1075 goto path_put;
1076
1077 sk = unix_find_socket_byinode(inode);
1078 if (!sk)
1079 goto path_put;
1080
1081 err = -EPROTOTYPE;
1082 if (sk->sk_type == type)
1083 touch_atime(&path);
1084 else
1085 goto sock_put;
1086
1087 path_put(&path);
1088
1089 return sk;
1090
1091sock_put:
1092 sock_put(sk);
1093path_put:
1094 path_put(&path);
1095fail:
1096 return ERR_PTR(err);
1097}
1098
1099static struct sock *unix_find_abstract(struct net *net,
1100 struct sockaddr_un *sunaddr,
1101 int addr_len, int type)
1102{
1103 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
1104 struct dentry *dentry;
1105 struct sock *sk;
1106
1107 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
1108 if (!sk)
1109 return ERR_PTR(-ECONNREFUSED);
1110
1111 dentry = unix_sk(sk)->path.dentry;
1112 if (dentry)
1113 touch_atime(&unix_sk(sk)->path);
1114
1115 return sk;
1116}
1117
1118static struct sock *unix_find_other(struct net *net,
1119 struct sockaddr_un *sunaddr,
1120 int addr_len, int type)
1121{
1122 struct sock *sk;
1123
1124 if (sunaddr->sun_path[0])
1125 sk = unix_find_bsd(sunaddr, addr_len, type);
1126 else
1127 sk = unix_find_abstract(net, sunaddr, addr_len, type);
1128
1129 return sk;
1130}
1131
1132static int unix_autobind(struct sock *sk)
1133{
1134 struct unix_sock *u = unix_sk(sk);
1135 unsigned int new_hash, old_hash;
1136 struct net *net = sock_net(sk);
1137 struct unix_address *addr;
1138 u32 lastnum, ordernum;
1139 int err;
1140
1141 err = mutex_lock_interruptible(&u->bindlock);
1142 if (err)
1143 return err;
1144
1145 if (u->addr)
1146 goto out;
1147
1148 err = -ENOMEM;
1149 addr = kzalloc(sizeof(*addr) +
1150 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1151 if (!addr)
1152 goto out;
1153
1154 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1155 addr->name->sun_family = AF_UNIX;
1156 refcount_set(&addr->refcnt, 1);
1157
1158 old_hash = sk->sk_hash;
1159 ordernum = get_random_u32();
1160 lastnum = ordernum & 0xFFFFF;
1161retry:
1162 ordernum = (ordernum + 1) & 0xFFFFF;
1163 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1164
1165 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1166 unix_table_double_lock(net, old_hash, new_hash);
1167
1168 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1169 unix_table_double_unlock(net, old_hash, new_hash);
1170
1171 /* __unix_find_socket_byname() may take long time if many names
1172 * are already in use.
1173 */
1174 cond_resched();
1175
1176 if (ordernum == lastnum) {
1177 /* Give up if all names seems to be in use. */
1178 err = -ENOSPC;
1179 unix_release_addr(addr);
1180 goto out;
1181 }
1182
1183 goto retry;
1184 }
1185
1186 __unix_set_addr_hash(net, sk, addr, new_hash);
1187 unix_table_double_unlock(net, old_hash, new_hash);
1188 err = 0;
1189
1190out: mutex_unlock(&u->bindlock);
1191 return err;
1192}
1193
1194static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1195 int addr_len)
1196{
1197 umode_t mode = S_IFSOCK |
1198 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1199 struct unix_sock *u = unix_sk(sk);
1200 unsigned int new_hash, old_hash;
1201 struct net *net = sock_net(sk);
1202 struct mnt_idmap *idmap;
1203 struct unix_address *addr;
1204 struct dentry *dentry;
1205 struct path parent;
1206 int err;
1207
1208 addr_len = unix_mkname_bsd(sunaddr, addr_len);
1209 addr = unix_create_addr(sunaddr, addr_len);
1210 if (!addr)
1211 return -ENOMEM;
1212
1213 /*
1214 * Get the parent directory, calculate the hash for last
1215 * component.
1216 */
1217 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1218 if (IS_ERR(dentry)) {
1219 err = PTR_ERR(dentry);
1220 goto out;
1221 }
1222
1223 /*
1224 * All right, let's create it.
1225 */
1226 idmap = mnt_idmap(parent.mnt);
1227 err = security_path_mknod(&parent, dentry, mode, 0);
1228 if (!err)
1229 err = vfs_mknod(idmap, d_inode(parent.dentry), dentry, mode, 0);
1230 if (err)
1231 goto out_path;
1232 err = mutex_lock_interruptible(&u->bindlock);
1233 if (err)
1234 goto out_unlink;
1235 if (u->addr)
1236 goto out_unlock;
1237
1238 old_hash = sk->sk_hash;
1239 new_hash = unix_bsd_hash(d_backing_inode(dentry));
1240 unix_table_double_lock(net, old_hash, new_hash);
1241 u->path.mnt = mntget(parent.mnt);
1242 u->path.dentry = dget(dentry);
1243 __unix_set_addr_hash(net, sk, addr, new_hash);
1244 unix_table_double_unlock(net, old_hash, new_hash);
1245 unix_insert_bsd_socket(sk);
1246 mutex_unlock(&u->bindlock);
1247 done_path_create(&parent, dentry);
1248 return 0;
1249
1250out_unlock:
1251 mutex_unlock(&u->bindlock);
1252 err = -EINVAL;
1253out_unlink:
1254 /* failed after successful mknod? unlink what we'd created... */
1255 vfs_unlink(idmap, d_inode(parent.dentry), dentry, NULL);
1256out_path:
1257 done_path_create(&parent, dentry);
1258out:
1259 unix_release_addr(addr);
1260 return err == -EEXIST ? -EADDRINUSE : err;
1261}
1262
1263static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1264 int addr_len)
1265{
1266 struct unix_sock *u = unix_sk(sk);
1267 unsigned int new_hash, old_hash;
1268 struct net *net = sock_net(sk);
1269 struct unix_address *addr;
1270 int err;
1271
1272 addr = unix_create_addr(sunaddr, addr_len);
1273 if (!addr)
1274 return -ENOMEM;
1275
1276 err = mutex_lock_interruptible(&u->bindlock);
1277 if (err)
1278 goto out;
1279
1280 if (u->addr) {
1281 err = -EINVAL;
1282 goto out_mutex;
1283 }
1284
1285 old_hash = sk->sk_hash;
1286 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1287 unix_table_double_lock(net, old_hash, new_hash);
1288
1289 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
1290 goto out_spin;
1291
1292 __unix_set_addr_hash(net, sk, addr, new_hash);
1293 unix_table_double_unlock(net, old_hash, new_hash);
1294 mutex_unlock(&u->bindlock);
1295 return 0;
1296
1297out_spin:
1298 unix_table_double_unlock(net, old_hash, new_hash);
1299 err = -EADDRINUSE;
1300out_mutex:
1301 mutex_unlock(&u->bindlock);
1302out:
1303 unix_release_addr(addr);
1304 return err;
1305}
1306
1307static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1308{
1309 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1310 struct sock *sk = sock->sk;
1311 int err;
1312
1313 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1314 sunaddr->sun_family == AF_UNIX)
1315 return unix_autobind(sk);
1316
1317 err = unix_validate_addr(sunaddr, addr_len);
1318 if (err)
1319 return err;
1320
1321 if (sunaddr->sun_path[0])
1322 err = unix_bind_bsd(sk, sunaddr, addr_len);
1323 else
1324 err = unix_bind_abstract(sk, sunaddr, addr_len);
1325
1326 return err;
1327}
1328
1329static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1330{
1331 if (unlikely(sk1 == sk2) || !sk2) {
1332 unix_state_lock(sk1);
1333 return;
1334 }
1335 if (sk1 > sk2)
1336 swap(sk1, sk2);
1337
1338 unix_state_lock(sk1);
1339 unix_state_lock_nested(sk2, U_LOCK_SECOND);
1340}
1341
1342static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1343{
1344 if (unlikely(sk1 == sk2) || !sk2) {
1345 unix_state_unlock(sk1);
1346 return;
1347 }
1348 unix_state_unlock(sk1);
1349 unix_state_unlock(sk2);
1350}
1351
1352static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1353 int alen, int flags)
1354{
1355 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1356 struct sock *sk = sock->sk;
1357 struct sock *other;
1358 int err;
1359
1360 err = -EINVAL;
1361 if (alen < offsetofend(struct sockaddr, sa_family))
1362 goto out;
1363
1364 if (addr->sa_family != AF_UNSPEC) {
1365 err = unix_validate_addr(sunaddr, alen);
1366 if (err)
1367 goto out;
1368
1369 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, addr, &alen);
1370 if (err)
1371 goto out;
1372
1373 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1374 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1375 !READ_ONCE(unix_sk(sk)->addr)) {
1376 err = unix_autobind(sk);
1377 if (err)
1378 goto out;
1379 }
1380
1381restart:
1382 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
1383 if (IS_ERR(other)) {
1384 err = PTR_ERR(other);
1385 goto out;
1386 }
1387
1388 unix_state_double_lock(sk, other);
1389
1390 /* Apparently VFS overslept socket death. Retry. */
1391 if (sock_flag(other, SOCK_DEAD)) {
1392 unix_state_double_unlock(sk, other);
1393 sock_put(other);
1394 goto restart;
1395 }
1396
1397 err = -EPERM;
1398 if (!unix_may_send(sk, other))
1399 goto out_unlock;
1400
1401 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1402 if (err)
1403 goto out_unlock;
1404
1405 sk->sk_state = other->sk_state = TCP_ESTABLISHED;
1406 } else {
1407 /*
1408 * 1003.1g breaking connected state with AF_UNSPEC
1409 */
1410 other = NULL;
1411 unix_state_double_lock(sk, other);
1412 }
1413
1414 /*
1415 * If it was connected, reconnect.
1416 */
1417 if (unix_peer(sk)) {
1418 struct sock *old_peer = unix_peer(sk);
1419
1420 unix_peer(sk) = other;
1421 if (!other)
1422 sk->sk_state = TCP_CLOSE;
1423 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1424
1425 unix_state_double_unlock(sk, other);
1426
1427 if (other != old_peer)
1428 unix_dgram_disconnected(sk, old_peer);
1429 sock_put(old_peer);
1430 } else {
1431 unix_peer(sk) = other;
1432 unix_state_double_unlock(sk, other);
1433 }
1434
1435 return 0;
1436
1437out_unlock:
1438 unix_state_double_unlock(sk, other);
1439 sock_put(other);
1440out:
1441 return err;
1442}
1443
1444static long unix_wait_for_peer(struct sock *other, long timeo)
1445 __releases(&unix_sk(other)->lock)
1446{
1447 struct unix_sock *u = unix_sk(other);
1448 int sched;
1449 DEFINE_WAIT(wait);
1450
1451 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1452
1453 sched = !sock_flag(other, SOCK_DEAD) &&
1454 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1455 unix_recvq_full_lockless(other);
1456
1457 unix_state_unlock(other);
1458
1459 if (sched)
1460 timeo = schedule_timeout(timeo);
1461
1462 finish_wait(&u->peer_wait, &wait);
1463 return timeo;
1464}
1465
1466static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1467 int addr_len, int flags)
1468{
1469 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1470 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1471 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1472 struct net *net = sock_net(sk);
1473 struct sk_buff *skb = NULL;
1474 long timeo;
1475 int err;
1476 int st;
1477
1478 err = unix_validate_addr(sunaddr, addr_len);
1479 if (err)
1480 goto out;
1481
1482 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, &addr_len);
1483 if (err)
1484 goto out;
1485
1486 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1487 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1488 !READ_ONCE(u->addr)) {
1489 err = unix_autobind(sk);
1490 if (err)
1491 goto out;
1492 }
1493
1494 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1495
1496 /* First of all allocate resources.
1497 If we will make it after state is locked,
1498 we will have to recheck all again in any case.
1499 */
1500
1501 /* create new sock for complete connection */
1502 newsk = unix_create1(net, NULL, 0, sock->type);
1503 if (IS_ERR(newsk)) {
1504 err = PTR_ERR(newsk);
1505 newsk = NULL;
1506 goto out;
1507 }
1508
1509 err = -ENOMEM;
1510
1511 /* Allocate skb for sending to listening sock */
1512 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1513 if (skb == NULL)
1514 goto out;
1515
1516restart:
1517 /* Find listening sock. */
1518 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
1519 if (IS_ERR(other)) {
1520 err = PTR_ERR(other);
1521 other = NULL;
1522 goto out;
1523 }
1524
1525 /* Latch state of peer */
1526 unix_state_lock(other);
1527
1528 /* Apparently VFS overslept socket death. Retry. */
1529 if (sock_flag(other, SOCK_DEAD)) {
1530 unix_state_unlock(other);
1531 sock_put(other);
1532 goto restart;
1533 }
1534
1535 err = -ECONNREFUSED;
1536 if (other->sk_state != TCP_LISTEN)
1537 goto out_unlock;
1538 if (other->sk_shutdown & RCV_SHUTDOWN)
1539 goto out_unlock;
1540
1541 if (unix_recvq_full(other)) {
1542 err = -EAGAIN;
1543 if (!timeo)
1544 goto out_unlock;
1545
1546 timeo = unix_wait_for_peer(other, timeo);
1547
1548 err = sock_intr_errno(timeo);
1549 if (signal_pending(current))
1550 goto out;
1551 sock_put(other);
1552 goto restart;
1553 }
1554
1555 /* Latch our state.
1556
1557 It is tricky place. We need to grab our state lock and cannot
1558 drop lock on peer. It is dangerous because deadlock is
1559 possible. Connect to self case and simultaneous
1560 attempt to connect are eliminated by checking socket
1561 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1562 check this before attempt to grab lock.
1563
1564 Well, and we have to recheck the state after socket locked.
1565 */
1566 st = sk->sk_state;
1567
1568 switch (st) {
1569 case TCP_CLOSE:
1570 /* This is ok... continue with connect */
1571 break;
1572 case TCP_ESTABLISHED:
1573 /* Socket is already connected */
1574 err = -EISCONN;
1575 goto out_unlock;
1576 default:
1577 err = -EINVAL;
1578 goto out_unlock;
1579 }
1580
1581 unix_state_lock_nested(sk, U_LOCK_SECOND);
1582
1583 if (sk->sk_state != st) {
1584 unix_state_unlock(sk);
1585 unix_state_unlock(other);
1586 sock_put(other);
1587 goto restart;
1588 }
1589
1590 err = security_unix_stream_connect(sk, other, newsk);
1591 if (err) {
1592 unix_state_unlock(sk);
1593 goto out_unlock;
1594 }
1595
1596 /* The way is open! Fastly set all the necessary fields... */
1597
1598 sock_hold(sk);
1599 unix_peer(newsk) = sk;
1600 newsk->sk_state = TCP_ESTABLISHED;
1601 newsk->sk_type = sk->sk_type;
1602 init_peercred(newsk);
1603 newu = unix_sk(newsk);
1604 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1605 otheru = unix_sk(other);
1606
1607 /* copy address information from listening to new sock
1608 *
1609 * The contents of *(otheru->addr) and otheru->path
1610 * are seen fully set up here, since we have found
1611 * otheru in hash under its lock. Insertion into the
1612 * hash chain we'd found it in had been done in an
1613 * earlier critical area protected by the chain's lock,
1614 * the same one where we'd set *(otheru->addr) contents,
1615 * as well as otheru->path and otheru->addr itself.
1616 *
1617 * Using smp_store_release() here to set newu->addr
1618 * is enough to make those stores, as well as stores
1619 * to newu->path visible to anyone who gets newu->addr
1620 * by smp_load_acquire(). IOW, the same warranties
1621 * as for unix_sock instances bound in unix_bind() or
1622 * in unix_autobind().
1623 */
1624 if (otheru->path.dentry) {
1625 path_get(&otheru->path);
1626 newu->path = otheru->path;
1627 }
1628 refcount_inc(&otheru->addr->refcnt);
1629 smp_store_release(&newu->addr, otheru->addr);
1630
1631 /* Set credentials */
1632 copy_peercred(sk, other);
1633
1634 sock->state = SS_CONNECTED;
1635 sk->sk_state = TCP_ESTABLISHED;
1636 sock_hold(newsk);
1637
1638 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1639 unix_peer(sk) = newsk;
1640
1641 unix_state_unlock(sk);
1642
1643 /* take ten and send info to listening sock */
1644 spin_lock(&other->sk_receive_queue.lock);
1645 __skb_queue_tail(&other->sk_receive_queue, skb);
1646 spin_unlock(&other->sk_receive_queue.lock);
1647 unix_state_unlock(other);
1648 other->sk_data_ready(other);
1649 sock_put(other);
1650 return 0;
1651
1652out_unlock:
1653 if (other)
1654 unix_state_unlock(other);
1655
1656out:
1657 kfree_skb(skb);
1658 if (newsk)
1659 unix_release_sock(newsk, 0);
1660 if (other)
1661 sock_put(other);
1662 return err;
1663}
1664
1665static int unix_socketpair(struct socket *socka, struct socket *sockb)
1666{
1667 struct sock *ska = socka->sk, *skb = sockb->sk;
1668
1669 /* Join our sockets back to back */
1670 sock_hold(ska);
1671 sock_hold(skb);
1672 unix_peer(ska) = skb;
1673 unix_peer(skb) = ska;
1674 init_peercred(ska);
1675 init_peercred(skb);
1676
1677 ska->sk_state = TCP_ESTABLISHED;
1678 skb->sk_state = TCP_ESTABLISHED;
1679 socka->state = SS_CONNECTED;
1680 sockb->state = SS_CONNECTED;
1681 return 0;
1682}
1683
1684static void unix_sock_inherit_flags(const struct socket *old,
1685 struct socket *new)
1686{
1687 if (test_bit(SOCK_PASSCRED, &old->flags))
1688 set_bit(SOCK_PASSCRED, &new->flags);
1689 if (test_bit(SOCK_PASSPIDFD, &old->flags))
1690 set_bit(SOCK_PASSPIDFD, &new->flags);
1691 if (test_bit(SOCK_PASSSEC, &old->flags))
1692 set_bit(SOCK_PASSSEC, &new->flags);
1693}
1694
1695static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1696 bool kern)
1697{
1698 struct sock *sk = sock->sk;
1699 struct sock *tsk;
1700 struct sk_buff *skb;
1701 int err;
1702
1703 err = -EOPNOTSUPP;
1704 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1705 goto out;
1706
1707 err = -EINVAL;
1708 if (sk->sk_state != TCP_LISTEN)
1709 goto out;
1710
1711 /* If socket state is TCP_LISTEN it cannot change (for now...),
1712 * so that no locks are necessary.
1713 */
1714
1715 skb = skb_recv_datagram(sk, (flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1716 &err);
1717 if (!skb) {
1718 /* This means receive shutdown. */
1719 if (err == 0)
1720 err = -EINVAL;
1721 goto out;
1722 }
1723
1724 tsk = skb->sk;
1725 skb_free_datagram(sk, skb);
1726 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1727
1728 /* attach accepted sock to socket */
1729 unix_state_lock(tsk);
1730 newsock->state = SS_CONNECTED;
1731 unix_sock_inherit_flags(sock, newsock);
1732 sock_graft(tsk, newsock);
1733 unix_state_unlock(tsk);
1734 return 0;
1735
1736out:
1737 return err;
1738}
1739
1740
1741static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1742{
1743 struct sock *sk = sock->sk;
1744 struct unix_address *addr;
1745 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1746 int err = 0;
1747
1748 if (peer) {
1749 sk = unix_peer_get(sk);
1750
1751 err = -ENOTCONN;
1752 if (!sk)
1753 goto out;
1754 err = 0;
1755 } else {
1756 sock_hold(sk);
1757 }
1758
1759 addr = smp_load_acquire(&unix_sk(sk)->addr);
1760 if (!addr) {
1761 sunaddr->sun_family = AF_UNIX;
1762 sunaddr->sun_path[0] = 0;
1763 err = offsetof(struct sockaddr_un, sun_path);
1764 } else {
1765 err = addr->len;
1766 memcpy(sunaddr, addr->name, addr->len);
1767
1768 if (peer)
1769 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1770 CGROUP_UNIX_GETPEERNAME);
1771 else
1772 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1773 CGROUP_UNIX_GETSOCKNAME);
1774 }
1775 sock_put(sk);
1776out:
1777 return err;
1778}
1779
1780/* The "user->unix_inflight" variable is protected by the garbage
1781 * collection lock, and we just read it locklessly here. If you go
1782 * over the limit, there might be a tiny race in actually noticing
1783 * it across threads. Tough.
1784 */
1785static inline bool too_many_unix_fds(struct task_struct *p)
1786{
1787 struct user_struct *user = current_user();
1788
1789 if (unlikely(READ_ONCE(user->unix_inflight) > task_rlimit(p, RLIMIT_NOFILE)))
1790 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1791 return false;
1792}
1793
1794static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1795{
1796 int i;
1797
1798 if (too_many_unix_fds(current))
1799 return -ETOOMANYREFS;
1800
1801 /* Need to duplicate file references for the sake of garbage
1802 * collection. Otherwise a socket in the fps might become a
1803 * candidate for GC while the skb is not yet queued.
1804 */
1805 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1806 if (!UNIXCB(skb).fp)
1807 return -ENOMEM;
1808
1809 for (i = scm->fp->count - 1; i >= 0; i--)
1810 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1811
1812 return 0;
1813}
1814
1815static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1816{
1817 int i;
1818
1819 scm->fp = UNIXCB(skb).fp;
1820 UNIXCB(skb).fp = NULL;
1821
1822 for (i = scm->fp->count - 1; i >= 0; i--)
1823 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1824}
1825
1826static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1827{
1828 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1829
1830 /*
1831 * Garbage collection of unix sockets starts by selecting a set of
1832 * candidate sockets which have reference only from being in flight
1833 * (total_refs == inflight_refs). This condition is checked once during
1834 * the candidate collection phase, and candidates are marked as such, so
1835 * that non-candidates can later be ignored. While inflight_refs is
1836 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1837 * is an instantaneous decision.
1838 *
1839 * Once a candidate, however, the socket must not be reinstalled into a
1840 * file descriptor while the garbage collection is in progress.
1841 *
1842 * If the above conditions are met, then the directed graph of
1843 * candidates (*) does not change while unix_gc_lock is held.
1844 *
1845 * Any operations that changes the file count through file descriptors
1846 * (dup, close, sendmsg) does not change the graph since candidates are
1847 * not installed in fds.
1848 *
1849 * Dequeing a candidate via recvmsg would install it into an fd, but
1850 * that takes unix_gc_lock to decrement the inflight count, so it's
1851 * serialized with garbage collection.
1852 *
1853 * MSG_PEEK is special in that it does not change the inflight count,
1854 * yet does install the socket into an fd. The following lock/unlock
1855 * pair is to ensure serialization with garbage collection. It must be
1856 * done between incrementing the file count and installing the file into
1857 * an fd.
1858 *
1859 * If garbage collection starts after the barrier provided by the
1860 * lock/unlock, then it will see the elevated refcount and not mark this
1861 * as a candidate. If a garbage collection is already in progress
1862 * before the file count was incremented, then the lock/unlock pair will
1863 * ensure that garbage collection is finished before progressing to
1864 * installing the fd.
1865 *
1866 * (*) A -> B where B is on the queue of A or B is on the queue of C
1867 * which is on the queue of listening socket A.
1868 */
1869 spin_lock(&unix_gc_lock);
1870 spin_unlock(&unix_gc_lock);
1871}
1872
1873static void unix_destruct_scm(struct sk_buff *skb)
1874{
1875 struct scm_cookie scm;
1876
1877 memset(&scm, 0, sizeof(scm));
1878 scm.pid = UNIXCB(skb).pid;
1879 if (UNIXCB(skb).fp)
1880 unix_detach_fds(&scm, skb);
1881
1882 /* Alas, it calls VFS */
1883 /* So fscking what? fput() had been SMP-safe since the last Summer */
1884 scm_destroy(&scm);
1885 sock_wfree(skb);
1886}
1887
1888static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1889{
1890 int err = 0;
1891
1892 UNIXCB(skb).pid = get_pid(scm->pid);
1893 UNIXCB(skb).uid = scm->creds.uid;
1894 UNIXCB(skb).gid = scm->creds.gid;
1895 UNIXCB(skb).fp = NULL;
1896 unix_get_secdata(scm, skb);
1897 if (scm->fp && send_fds)
1898 err = unix_attach_fds(scm, skb);
1899
1900 skb->destructor = unix_destruct_scm;
1901 return err;
1902}
1903
1904static bool unix_passcred_enabled(const struct socket *sock,
1905 const struct sock *other)
1906{
1907 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1908 test_bit(SOCK_PASSPIDFD, &sock->flags) ||
1909 !other->sk_socket ||
1910 test_bit(SOCK_PASSCRED, &other->sk_socket->flags) ||
1911 test_bit(SOCK_PASSPIDFD, &other->sk_socket->flags);
1912}
1913
1914/*
1915 * Some apps rely on write() giving SCM_CREDENTIALS
1916 * We include credentials if source or destination socket
1917 * asserted SOCK_PASSCRED.
1918 */
1919static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1920 const struct sock *other)
1921{
1922 if (UNIXCB(skb).pid)
1923 return;
1924 if (unix_passcred_enabled(sock, other)) {
1925 UNIXCB(skb).pid = get_pid(task_tgid(current));
1926 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1927 }
1928}
1929
1930static bool unix_skb_scm_eq(struct sk_buff *skb,
1931 struct scm_cookie *scm)
1932{
1933 return UNIXCB(skb).pid == scm->pid &&
1934 uid_eq(UNIXCB(skb).uid, scm->creds.uid) &&
1935 gid_eq(UNIXCB(skb).gid, scm->creds.gid) &&
1936 unix_secdata_eq(scm, skb);
1937}
1938
1939static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1940{
1941 struct scm_fp_list *fp = UNIXCB(skb).fp;
1942 struct unix_sock *u = unix_sk(sk);
1943
1944 if (unlikely(fp && fp->count))
1945 atomic_add(fp->count, &u->scm_stat.nr_fds);
1946}
1947
1948static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1949{
1950 struct scm_fp_list *fp = UNIXCB(skb).fp;
1951 struct unix_sock *u = unix_sk(sk);
1952
1953 if (unlikely(fp && fp->count))
1954 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1955}
1956
1957/*
1958 * Send AF_UNIX data.
1959 */
1960
1961static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1962 size_t len)
1963{
1964 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1965 struct sock *sk = sock->sk, *other = NULL;
1966 struct unix_sock *u = unix_sk(sk);
1967 struct scm_cookie scm;
1968 struct sk_buff *skb;
1969 int data_len = 0;
1970 int sk_locked;
1971 long timeo;
1972 int err;
1973
1974 err = scm_send(sock, msg, &scm, false);
1975 if (err < 0)
1976 return err;
1977
1978 wait_for_unix_gc(scm.fp);
1979
1980 err = -EOPNOTSUPP;
1981 if (msg->msg_flags&MSG_OOB)
1982 goto out;
1983
1984 if (msg->msg_namelen) {
1985 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1986 if (err)
1987 goto out;
1988
1989 err = BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk,
1990 msg->msg_name,
1991 &msg->msg_namelen,
1992 NULL);
1993 if (err)
1994 goto out;
1995 } else {
1996 sunaddr = NULL;
1997 err = -ENOTCONN;
1998 other = unix_peer_get(sk);
1999 if (!other)
2000 goto out;
2001 }
2002
2003 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
2004 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
2005 !READ_ONCE(u->addr)) {
2006 err = unix_autobind(sk);
2007 if (err)
2008 goto out;
2009 }
2010
2011 err = -EMSGSIZE;
2012 if (len > sk->sk_sndbuf - 32)
2013 goto out;
2014
2015 if (len > SKB_MAX_ALLOC) {
2016 data_len = min_t(size_t,
2017 len - SKB_MAX_ALLOC,
2018 MAX_SKB_FRAGS * PAGE_SIZE);
2019 data_len = PAGE_ALIGN(data_len);
2020
2021 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
2022 }
2023
2024 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
2025 msg->msg_flags & MSG_DONTWAIT, &err,
2026 PAGE_ALLOC_COSTLY_ORDER);
2027 if (skb == NULL)
2028 goto out;
2029
2030 err = unix_scm_to_skb(&scm, skb, true);
2031 if (err < 0)
2032 goto out_free;
2033
2034 skb_put(skb, len - data_len);
2035 skb->data_len = data_len;
2036 skb->len = len;
2037 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
2038 if (err)
2039 goto out_free;
2040
2041 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
2042
2043restart:
2044 if (!other) {
2045 err = -ECONNRESET;
2046 if (sunaddr == NULL)
2047 goto out_free;
2048
2049 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
2050 sk->sk_type);
2051 if (IS_ERR(other)) {
2052 err = PTR_ERR(other);
2053 other = NULL;
2054 goto out_free;
2055 }
2056 }
2057
2058 if (sk_filter(other, skb) < 0) {
2059 /* Toss the packet but do not return any error to the sender */
2060 err = len;
2061 goto out_free;
2062 }
2063
2064 sk_locked = 0;
2065 unix_state_lock(other);
2066restart_locked:
2067 err = -EPERM;
2068 if (!unix_may_send(sk, other))
2069 goto out_unlock;
2070
2071 if (unlikely(sock_flag(other, SOCK_DEAD))) {
2072 /*
2073 * Check with 1003.1g - what should
2074 * datagram error
2075 */
2076 unix_state_unlock(other);
2077 sock_put(other);
2078
2079 if (!sk_locked)
2080 unix_state_lock(sk);
2081
2082 err = 0;
2083 if (sk->sk_type == SOCK_SEQPACKET) {
2084 /* We are here only when racing with unix_release_sock()
2085 * is clearing @other. Never change state to TCP_CLOSE
2086 * unlike SOCK_DGRAM wants.
2087 */
2088 unix_state_unlock(sk);
2089 err = -EPIPE;
2090 } else if (unix_peer(sk) == other) {
2091 unix_peer(sk) = NULL;
2092 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2093
2094 sk->sk_state = TCP_CLOSE;
2095 unix_state_unlock(sk);
2096
2097 unix_dgram_disconnected(sk, other);
2098 sock_put(other);
2099 err = -ECONNREFUSED;
2100 } else {
2101 unix_state_unlock(sk);
2102 }
2103
2104 other = NULL;
2105 if (err)
2106 goto out_free;
2107 goto restart;
2108 }
2109
2110 err = -EPIPE;
2111 if (other->sk_shutdown & RCV_SHUTDOWN)
2112 goto out_unlock;
2113
2114 if (sk->sk_type != SOCK_SEQPACKET) {
2115 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
2116 if (err)
2117 goto out_unlock;
2118 }
2119
2120 /* other == sk && unix_peer(other) != sk if
2121 * - unix_peer(sk) == NULL, destination address bound to sk
2122 * - unix_peer(sk) == sk by time of get but disconnected before lock
2123 */
2124 if (other != sk &&
2125 unlikely(unix_peer(other) != sk &&
2126 unix_recvq_full_lockless(other))) {
2127 if (timeo) {
2128 timeo = unix_wait_for_peer(other, timeo);
2129
2130 err = sock_intr_errno(timeo);
2131 if (signal_pending(current))
2132 goto out_free;
2133
2134 goto restart;
2135 }
2136
2137 if (!sk_locked) {
2138 unix_state_unlock(other);
2139 unix_state_double_lock(sk, other);
2140 }
2141
2142 if (unix_peer(sk) != other ||
2143 unix_dgram_peer_wake_me(sk, other)) {
2144 err = -EAGAIN;
2145 sk_locked = 1;
2146 goto out_unlock;
2147 }
2148
2149 if (!sk_locked) {
2150 sk_locked = 1;
2151 goto restart_locked;
2152 }
2153 }
2154
2155 if (unlikely(sk_locked))
2156 unix_state_unlock(sk);
2157
2158 if (sock_flag(other, SOCK_RCVTSTAMP))
2159 __net_timestamp(skb);
2160 maybe_add_creds(skb, sock, other);
2161 scm_stat_add(other, skb);
2162 skb_queue_tail(&other->sk_receive_queue, skb);
2163 unix_state_unlock(other);
2164 other->sk_data_ready(other);
2165 sock_put(other);
2166 scm_destroy(&scm);
2167 return len;
2168
2169out_unlock:
2170 if (sk_locked)
2171 unix_state_unlock(sk);
2172 unix_state_unlock(other);
2173out_free:
2174 kfree_skb(skb);
2175out:
2176 if (other)
2177 sock_put(other);
2178 scm_destroy(&scm);
2179 return err;
2180}
2181
2182/* We use paged skbs for stream sockets, and limit occupancy to 32768
2183 * bytes, and a minimum of a full page.
2184 */
2185#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
2186
2187#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2188static int queue_oob(struct socket *sock, struct msghdr *msg, struct sock *other,
2189 struct scm_cookie *scm, bool fds_sent)
2190{
2191 struct unix_sock *ousk = unix_sk(other);
2192 struct sk_buff *skb;
2193 int err = 0;
2194
2195 skb = sock_alloc_send_skb(sock->sk, 1, msg->msg_flags & MSG_DONTWAIT, &err);
2196
2197 if (!skb)
2198 return err;
2199
2200 err = unix_scm_to_skb(scm, skb, !fds_sent);
2201 if (err < 0) {
2202 kfree_skb(skb);
2203 return err;
2204 }
2205 skb_put(skb, 1);
2206 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
2207
2208 if (err) {
2209 kfree_skb(skb);
2210 return err;
2211 }
2212
2213 unix_state_lock(other);
2214
2215 if (sock_flag(other, SOCK_DEAD) ||
2216 (other->sk_shutdown & RCV_SHUTDOWN)) {
2217 unix_state_unlock(other);
2218 kfree_skb(skb);
2219 return -EPIPE;
2220 }
2221
2222 maybe_add_creds(skb, sock, other);
2223 skb_get(skb);
2224
2225 scm_stat_add(other, skb);
2226
2227 spin_lock(&other->sk_receive_queue.lock);
2228 if (ousk->oob_skb)
2229 consume_skb(ousk->oob_skb);
2230 WRITE_ONCE(ousk->oob_skb, skb);
2231 __skb_queue_tail(&other->sk_receive_queue, skb);
2232 spin_unlock(&other->sk_receive_queue.lock);
2233
2234 sk_send_sigurg(other);
2235 unix_state_unlock(other);
2236 other->sk_data_ready(other);
2237
2238 return err;
2239}
2240#endif
2241
2242static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2243 size_t len)
2244{
2245 struct sock *sk = sock->sk;
2246 struct sock *other = NULL;
2247 int err, size;
2248 struct sk_buff *skb;
2249 int sent = 0;
2250 struct scm_cookie scm;
2251 bool fds_sent = false;
2252 int data_len;
2253
2254 err = scm_send(sock, msg, &scm, false);
2255 if (err < 0)
2256 return err;
2257
2258 wait_for_unix_gc(scm.fp);
2259
2260 err = -EOPNOTSUPP;
2261 if (msg->msg_flags & MSG_OOB) {
2262#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2263 if (len)
2264 len--;
2265 else
2266#endif
2267 goto out_err;
2268 }
2269
2270 if (msg->msg_namelen) {
2271 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2272 goto out_err;
2273 } else {
2274 err = -ENOTCONN;
2275 other = unix_peer(sk);
2276 if (!other)
2277 goto out_err;
2278 }
2279
2280 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2281 goto pipe_err;
2282
2283 while (sent < len) {
2284 size = len - sent;
2285
2286 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2287 skb = sock_alloc_send_pskb(sk, 0, 0,
2288 msg->msg_flags & MSG_DONTWAIT,
2289 &err, 0);
2290 } else {
2291 /* Keep two messages in the pipe so it schedules better */
2292 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
2293
2294 /* allow fallback to order-0 allocations */
2295 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
2296
2297 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
2298
2299 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2300
2301 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
2302 msg->msg_flags & MSG_DONTWAIT, &err,
2303 get_order(UNIX_SKB_FRAGS_SZ));
2304 }
2305 if (!skb)
2306 goto out_err;
2307
2308 /* Only send the fds in the first buffer */
2309 err = unix_scm_to_skb(&scm, skb, !fds_sent);
2310 if (err < 0) {
2311 kfree_skb(skb);
2312 goto out_err;
2313 }
2314 fds_sent = true;
2315
2316 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2317 err = skb_splice_from_iter(skb, &msg->msg_iter, size,
2318 sk->sk_allocation);
2319 if (err < 0) {
2320 kfree_skb(skb);
2321 goto out_err;
2322 }
2323 size = err;
2324 refcount_add(size, &sk->sk_wmem_alloc);
2325 } else {
2326 skb_put(skb, size - data_len);
2327 skb->data_len = data_len;
2328 skb->len = size;
2329 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
2330 if (err) {
2331 kfree_skb(skb);
2332 goto out_err;
2333 }
2334 }
2335
2336 unix_state_lock(other);
2337
2338 if (sock_flag(other, SOCK_DEAD) ||
2339 (other->sk_shutdown & RCV_SHUTDOWN))
2340 goto pipe_err_free;
2341
2342 maybe_add_creds(skb, sock, other);
2343 scm_stat_add(other, skb);
2344 skb_queue_tail(&other->sk_receive_queue, skb);
2345 unix_state_unlock(other);
2346 other->sk_data_ready(other);
2347 sent += size;
2348 }
2349
2350#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2351 if (msg->msg_flags & MSG_OOB) {
2352 err = queue_oob(sock, msg, other, &scm, fds_sent);
2353 if (err)
2354 goto out_err;
2355 sent++;
2356 }
2357#endif
2358
2359 scm_destroy(&scm);
2360
2361 return sent;
2362
2363pipe_err_free:
2364 unix_state_unlock(other);
2365 kfree_skb(skb);
2366pipe_err:
2367 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
2368 send_sig(SIGPIPE, current, 0);
2369 err = -EPIPE;
2370out_err:
2371 scm_destroy(&scm);
2372 return sent ? : err;
2373}
2374
2375static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2376 size_t len)
2377{
2378 int err;
2379 struct sock *sk = sock->sk;
2380
2381 err = sock_error(sk);
2382 if (err)
2383 return err;
2384
2385 if (sk->sk_state != TCP_ESTABLISHED)
2386 return -ENOTCONN;
2387
2388 if (msg->msg_namelen)
2389 msg->msg_namelen = 0;
2390
2391 return unix_dgram_sendmsg(sock, msg, len);
2392}
2393
2394static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2395 size_t size, int flags)
2396{
2397 struct sock *sk = sock->sk;
2398
2399 if (sk->sk_state != TCP_ESTABLISHED)
2400 return -ENOTCONN;
2401
2402 return unix_dgram_recvmsg(sock, msg, size, flags);
2403}
2404
2405static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2406{
2407 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2408
2409 if (addr) {
2410 msg->msg_namelen = addr->len;
2411 memcpy(msg->msg_name, addr->name, addr->len);
2412 }
2413}
2414
2415int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
2416 int flags)
2417{
2418 struct scm_cookie scm;
2419 struct socket *sock = sk->sk_socket;
2420 struct unix_sock *u = unix_sk(sk);
2421 struct sk_buff *skb, *last;
2422 long timeo;
2423 int skip;
2424 int err;
2425
2426 err = -EOPNOTSUPP;
2427 if (flags&MSG_OOB)
2428 goto out;
2429
2430 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2431
2432 do {
2433 mutex_lock(&u->iolock);
2434
2435 skip = sk_peek_offset(sk, flags);
2436 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2437 &skip, &err, &last);
2438 if (skb) {
2439 if (!(flags & MSG_PEEK))
2440 scm_stat_del(sk, skb);
2441 break;
2442 }
2443
2444 mutex_unlock(&u->iolock);
2445
2446 if (err != -EAGAIN)
2447 break;
2448 } while (timeo &&
2449 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2450 &err, &timeo, last));
2451
2452 if (!skb) { /* implies iolock unlocked */
2453 unix_state_lock(sk);
2454 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2455 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2456 (sk->sk_shutdown & RCV_SHUTDOWN))
2457 err = 0;
2458 unix_state_unlock(sk);
2459 goto out;
2460 }
2461
2462 if (wq_has_sleeper(&u->peer_wait))
2463 wake_up_interruptible_sync_poll(&u->peer_wait,
2464 EPOLLOUT | EPOLLWRNORM |
2465 EPOLLWRBAND);
2466
2467 if (msg->msg_name) {
2468 unix_copy_addr(msg, skb->sk);
2469
2470 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2471 msg->msg_name,
2472 &msg->msg_namelen);
2473 }
2474
2475 if (size > skb->len - skip)
2476 size = skb->len - skip;
2477 else if (size < skb->len - skip)
2478 msg->msg_flags |= MSG_TRUNC;
2479
2480 err = skb_copy_datagram_msg(skb, skip, msg, size);
2481 if (err)
2482 goto out_free;
2483
2484 if (sock_flag(sk, SOCK_RCVTSTAMP))
2485 __sock_recv_timestamp(msg, sk, skb);
2486
2487 memset(&scm, 0, sizeof(scm));
2488
2489 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2490 unix_set_secdata(&scm, skb);
2491
2492 if (!(flags & MSG_PEEK)) {
2493 if (UNIXCB(skb).fp)
2494 unix_detach_fds(&scm, skb);
2495
2496 sk_peek_offset_bwd(sk, skb->len);
2497 } else {
2498 /* It is questionable: on PEEK we could:
2499 - do not return fds - good, but too simple 8)
2500 - return fds, and do not return them on read (old strategy,
2501 apparently wrong)
2502 - clone fds (I chose it for now, it is the most universal
2503 solution)
2504
2505 POSIX 1003.1g does not actually define this clearly
2506 at all. POSIX 1003.1g doesn't define a lot of things
2507 clearly however!
2508
2509 */
2510
2511 sk_peek_offset_fwd(sk, size);
2512
2513 if (UNIXCB(skb).fp)
2514 unix_peek_fds(&scm, skb);
2515 }
2516 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2517
2518 scm_recv_unix(sock, msg, &scm, flags);
2519
2520out_free:
2521 skb_free_datagram(sk, skb);
2522 mutex_unlock(&u->iolock);
2523out:
2524 return err;
2525}
2526
2527static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2528 int flags)
2529{
2530 struct sock *sk = sock->sk;
2531
2532#ifdef CONFIG_BPF_SYSCALL
2533 const struct proto *prot = READ_ONCE(sk->sk_prot);
2534
2535 if (prot != &unix_dgram_proto)
2536 return prot->recvmsg(sk, msg, size, flags, NULL);
2537#endif
2538 return __unix_dgram_recvmsg(sk, msg, size, flags);
2539}
2540
2541static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2542{
2543 struct unix_sock *u = unix_sk(sk);
2544 struct sk_buff *skb;
2545 int err;
2546
2547 mutex_lock(&u->iolock);
2548 skb = skb_recv_datagram(sk, MSG_DONTWAIT, &err);
2549 mutex_unlock(&u->iolock);
2550 if (!skb)
2551 return err;
2552
2553 return recv_actor(sk, skb);
2554}
2555
2556/*
2557 * Sleep until more data has arrived. But check for races..
2558 */
2559static long unix_stream_data_wait(struct sock *sk, long timeo,
2560 struct sk_buff *last, unsigned int last_len,
2561 bool freezable)
2562{
2563 unsigned int state = TASK_INTERRUPTIBLE | freezable * TASK_FREEZABLE;
2564 struct sk_buff *tail;
2565 DEFINE_WAIT(wait);
2566
2567 unix_state_lock(sk);
2568
2569 for (;;) {
2570 prepare_to_wait(sk_sleep(sk), &wait, state);
2571
2572 tail = skb_peek_tail(&sk->sk_receive_queue);
2573 if (tail != last ||
2574 (tail && tail->len != last_len) ||
2575 sk->sk_err ||
2576 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2577 signal_pending(current) ||
2578 !timeo)
2579 break;
2580
2581 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2582 unix_state_unlock(sk);
2583 timeo = schedule_timeout(timeo);
2584 unix_state_lock(sk);
2585
2586 if (sock_flag(sk, SOCK_DEAD))
2587 break;
2588
2589 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2590 }
2591
2592 finish_wait(sk_sleep(sk), &wait);
2593 unix_state_unlock(sk);
2594 return timeo;
2595}
2596
2597static unsigned int unix_skb_len(const struct sk_buff *skb)
2598{
2599 return skb->len - UNIXCB(skb).consumed;
2600}
2601
2602struct unix_stream_read_state {
2603 int (*recv_actor)(struct sk_buff *, int, int,
2604 struct unix_stream_read_state *);
2605 struct socket *socket;
2606 struct msghdr *msg;
2607 struct pipe_inode_info *pipe;
2608 size_t size;
2609 int flags;
2610 unsigned int splice_flags;
2611};
2612
2613#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2614static int unix_stream_recv_urg(struct unix_stream_read_state *state)
2615{
2616 struct socket *sock = state->socket;
2617 struct sock *sk = sock->sk;
2618 struct unix_sock *u = unix_sk(sk);
2619 int chunk = 1;
2620 struct sk_buff *oob_skb;
2621
2622 mutex_lock(&u->iolock);
2623 unix_state_lock(sk);
2624 spin_lock(&sk->sk_receive_queue.lock);
2625
2626 if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb) {
2627 spin_unlock(&sk->sk_receive_queue.lock);
2628 unix_state_unlock(sk);
2629 mutex_unlock(&u->iolock);
2630 return -EINVAL;
2631 }
2632
2633 oob_skb = u->oob_skb;
2634
2635 if (!(state->flags & MSG_PEEK))
2636 WRITE_ONCE(u->oob_skb, NULL);
2637 else
2638 skb_get(oob_skb);
2639
2640 spin_unlock(&sk->sk_receive_queue.lock);
2641 unix_state_unlock(sk);
2642
2643 chunk = state->recv_actor(oob_skb, 0, chunk, state);
2644
2645 if (!(state->flags & MSG_PEEK))
2646 UNIXCB(oob_skb).consumed += 1;
2647
2648 consume_skb(oob_skb);
2649
2650 mutex_unlock(&u->iolock);
2651
2652 if (chunk < 0)
2653 return -EFAULT;
2654
2655 state->msg->msg_flags |= MSG_OOB;
2656 return 1;
2657}
2658
2659static struct sk_buff *manage_oob(struct sk_buff *skb, struct sock *sk,
2660 int flags, int copied)
2661{
2662 struct unix_sock *u = unix_sk(sk);
2663
2664 if (!unix_skb_len(skb) && !(flags & MSG_PEEK)) {
2665 skb_unlink(skb, &sk->sk_receive_queue);
2666 consume_skb(skb);
2667 skb = NULL;
2668 } else {
2669 struct sk_buff *unlinked_skb = NULL;
2670
2671 spin_lock(&sk->sk_receive_queue.lock);
2672
2673 if (skb == u->oob_skb) {
2674 if (copied) {
2675 skb = NULL;
2676 } else if (sock_flag(sk, SOCK_URGINLINE)) {
2677 if (!(flags & MSG_PEEK)) {
2678 WRITE_ONCE(u->oob_skb, NULL);
2679 consume_skb(skb);
2680 }
2681 } else if (flags & MSG_PEEK) {
2682 skb = NULL;
2683 } else {
2684 __skb_unlink(skb, &sk->sk_receive_queue);
2685 WRITE_ONCE(u->oob_skb, NULL);
2686 unlinked_skb = skb;
2687 skb = skb_peek(&sk->sk_receive_queue);
2688 }
2689 }
2690
2691 spin_unlock(&sk->sk_receive_queue.lock);
2692
2693 if (unlinked_skb) {
2694 WARN_ON_ONCE(skb_unref(unlinked_skb));
2695 kfree_skb(unlinked_skb);
2696 }
2697 }
2698 return skb;
2699}
2700#endif
2701
2702static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2703{
2704 if (unlikely(sk->sk_state != TCP_ESTABLISHED))
2705 return -ENOTCONN;
2706
2707 return unix_read_skb(sk, recv_actor);
2708}
2709
2710static int unix_stream_read_generic(struct unix_stream_read_state *state,
2711 bool freezable)
2712{
2713 struct scm_cookie scm;
2714 struct socket *sock = state->socket;
2715 struct sock *sk = sock->sk;
2716 struct unix_sock *u = unix_sk(sk);
2717 int copied = 0;
2718 int flags = state->flags;
2719 int noblock = flags & MSG_DONTWAIT;
2720 bool check_creds = false;
2721 int target;
2722 int err = 0;
2723 long timeo;
2724 int skip;
2725 size_t size = state->size;
2726 unsigned int last_len;
2727
2728 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2729 err = -EINVAL;
2730 goto out;
2731 }
2732
2733 if (unlikely(flags & MSG_OOB)) {
2734 err = -EOPNOTSUPP;
2735#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2736 err = unix_stream_recv_urg(state);
2737#endif
2738 goto out;
2739 }
2740
2741 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2742 timeo = sock_rcvtimeo(sk, noblock);
2743
2744 memset(&scm, 0, sizeof(scm));
2745
2746 /* Lock the socket to prevent queue disordering
2747 * while sleeps in memcpy_tomsg
2748 */
2749 mutex_lock(&u->iolock);
2750
2751 skip = max(sk_peek_offset(sk, flags), 0);
2752
2753 do {
2754 int chunk;
2755 bool drop_skb;
2756 struct sk_buff *skb, *last;
2757
2758redo:
2759 unix_state_lock(sk);
2760 if (sock_flag(sk, SOCK_DEAD)) {
2761 err = -ECONNRESET;
2762 goto unlock;
2763 }
2764 last = skb = skb_peek(&sk->sk_receive_queue);
2765 last_len = last ? last->len : 0;
2766
2767again:
2768#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2769 if (skb) {
2770 skb = manage_oob(skb, sk, flags, copied);
2771 if (!skb && copied) {
2772 unix_state_unlock(sk);
2773 break;
2774 }
2775 }
2776#endif
2777 if (skb == NULL) {
2778 if (copied >= target)
2779 goto unlock;
2780
2781 /*
2782 * POSIX 1003.1g mandates this order.
2783 */
2784
2785 err = sock_error(sk);
2786 if (err)
2787 goto unlock;
2788 if (sk->sk_shutdown & RCV_SHUTDOWN)
2789 goto unlock;
2790
2791 unix_state_unlock(sk);
2792 if (!timeo) {
2793 err = -EAGAIN;
2794 break;
2795 }
2796
2797 mutex_unlock(&u->iolock);
2798
2799 timeo = unix_stream_data_wait(sk, timeo, last,
2800 last_len, freezable);
2801
2802 if (signal_pending(current)) {
2803 err = sock_intr_errno(timeo);
2804 scm_destroy(&scm);
2805 goto out;
2806 }
2807
2808 mutex_lock(&u->iolock);
2809 goto redo;
2810unlock:
2811 unix_state_unlock(sk);
2812 break;
2813 }
2814
2815 while (skip >= unix_skb_len(skb)) {
2816 skip -= unix_skb_len(skb);
2817 last = skb;
2818 last_len = skb->len;
2819 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2820 if (!skb)
2821 goto again;
2822 }
2823
2824 unix_state_unlock(sk);
2825
2826 if (check_creds) {
2827 /* Never glue messages from different writers */
2828 if (!unix_skb_scm_eq(skb, &scm))
2829 break;
2830 } else if (test_bit(SOCK_PASSCRED, &sock->flags) ||
2831 test_bit(SOCK_PASSPIDFD, &sock->flags)) {
2832 /* Copy credentials */
2833 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2834 unix_set_secdata(&scm, skb);
2835 check_creds = true;
2836 }
2837
2838 /* Copy address just once */
2839 if (state->msg && state->msg->msg_name) {
2840 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2841 state->msg->msg_name);
2842 unix_copy_addr(state->msg, skb->sk);
2843
2844 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2845 state->msg->msg_name,
2846 &state->msg->msg_namelen);
2847
2848 sunaddr = NULL;
2849 }
2850
2851 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2852 skb_get(skb);
2853 chunk = state->recv_actor(skb, skip, chunk, state);
2854 drop_skb = !unix_skb_len(skb);
2855 /* skb is only safe to use if !drop_skb */
2856 consume_skb(skb);
2857 if (chunk < 0) {
2858 if (copied == 0)
2859 copied = -EFAULT;
2860 break;
2861 }
2862 copied += chunk;
2863 size -= chunk;
2864
2865 if (drop_skb) {
2866 /* the skb was touched by a concurrent reader;
2867 * we should not expect anything from this skb
2868 * anymore and assume it invalid - we can be
2869 * sure it was dropped from the socket queue
2870 *
2871 * let's report a short read
2872 */
2873 err = 0;
2874 break;
2875 }
2876
2877 /* Mark read part of skb as used */
2878 if (!(flags & MSG_PEEK)) {
2879 UNIXCB(skb).consumed += chunk;
2880
2881 sk_peek_offset_bwd(sk, chunk);
2882
2883 if (UNIXCB(skb).fp) {
2884 scm_stat_del(sk, skb);
2885 unix_detach_fds(&scm, skb);
2886 }
2887
2888 if (unix_skb_len(skb))
2889 break;
2890
2891 skb_unlink(skb, &sk->sk_receive_queue);
2892 consume_skb(skb);
2893
2894 if (scm.fp)
2895 break;
2896 } else {
2897 /* It is questionable, see note in unix_dgram_recvmsg.
2898 */
2899 if (UNIXCB(skb).fp)
2900 unix_peek_fds(&scm, skb);
2901
2902 sk_peek_offset_fwd(sk, chunk);
2903
2904 if (UNIXCB(skb).fp)
2905 break;
2906
2907 skip = 0;
2908 last = skb;
2909 last_len = skb->len;
2910 unix_state_lock(sk);
2911 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2912 if (skb)
2913 goto again;
2914 unix_state_unlock(sk);
2915 break;
2916 }
2917 } while (size);
2918
2919 mutex_unlock(&u->iolock);
2920 if (state->msg)
2921 scm_recv_unix(sock, state->msg, &scm, flags);
2922 else
2923 scm_destroy(&scm);
2924out:
2925 return copied ? : err;
2926}
2927
2928static int unix_stream_read_actor(struct sk_buff *skb,
2929 int skip, int chunk,
2930 struct unix_stream_read_state *state)
2931{
2932 int ret;
2933
2934 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2935 state->msg, chunk);
2936 return ret ?: chunk;
2937}
2938
2939int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2940 size_t size, int flags)
2941{
2942 struct unix_stream_read_state state = {
2943 .recv_actor = unix_stream_read_actor,
2944 .socket = sk->sk_socket,
2945 .msg = msg,
2946 .size = size,
2947 .flags = flags
2948 };
2949
2950 return unix_stream_read_generic(&state, true);
2951}
2952
2953static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2954 size_t size, int flags)
2955{
2956 struct unix_stream_read_state state = {
2957 .recv_actor = unix_stream_read_actor,
2958 .socket = sock,
2959 .msg = msg,
2960 .size = size,
2961 .flags = flags
2962 };
2963
2964#ifdef CONFIG_BPF_SYSCALL
2965 struct sock *sk = sock->sk;
2966 const struct proto *prot = READ_ONCE(sk->sk_prot);
2967
2968 if (prot != &unix_stream_proto)
2969 return prot->recvmsg(sk, msg, size, flags, NULL);
2970#endif
2971 return unix_stream_read_generic(&state, true);
2972}
2973
2974static int unix_stream_splice_actor(struct sk_buff *skb,
2975 int skip, int chunk,
2976 struct unix_stream_read_state *state)
2977{
2978 return skb_splice_bits(skb, state->socket->sk,
2979 UNIXCB(skb).consumed + skip,
2980 state->pipe, chunk, state->splice_flags);
2981}
2982
2983static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2984 struct pipe_inode_info *pipe,
2985 size_t size, unsigned int flags)
2986{
2987 struct unix_stream_read_state state = {
2988 .recv_actor = unix_stream_splice_actor,
2989 .socket = sock,
2990 .pipe = pipe,
2991 .size = size,
2992 .splice_flags = flags,
2993 };
2994
2995 if (unlikely(*ppos))
2996 return -ESPIPE;
2997
2998 if (sock->file->f_flags & O_NONBLOCK ||
2999 flags & SPLICE_F_NONBLOCK)
3000 state.flags = MSG_DONTWAIT;
3001
3002 return unix_stream_read_generic(&state, false);
3003}
3004
3005static int unix_shutdown(struct socket *sock, int mode)
3006{
3007 struct sock *sk = sock->sk;
3008 struct sock *other;
3009
3010 if (mode < SHUT_RD || mode > SHUT_RDWR)
3011 return -EINVAL;
3012 /* This maps:
3013 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
3014 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
3015 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
3016 */
3017 ++mode;
3018
3019 unix_state_lock(sk);
3020 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | mode);
3021 other = unix_peer(sk);
3022 if (other)
3023 sock_hold(other);
3024 unix_state_unlock(sk);
3025 sk->sk_state_change(sk);
3026
3027 if (other &&
3028 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
3029
3030 int peer_mode = 0;
3031 const struct proto *prot = READ_ONCE(other->sk_prot);
3032
3033 if (prot->unhash)
3034 prot->unhash(other);
3035 if (mode&RCV_SHUTDOWN)
3036 peer_mode |= SEND_SHUTDOWN;
3037 if (mode&SEND_SHUTDOWN)
3038 peer_mode |= RCV_SHUTDOWN;
3039 unix_state_lock(other);
3040 WRITE_ONCE(other->sk_shutdown, other->sk_shutdown | peer_mode);
3041 unix_state_unlock(other);
3042 other->sk_state_change(other);
3043 if (peer_mode == SHUTDOWN_MASK)
3044 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
3045 else if (peer_mode & RCV_SHUTDOWN)
3046 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
3047 }
3048 if (other)
3049 sock_put(other);
3050
3051 return 0;
3052}
3053
3054long unix_inq_len(struct sock *sk)
3055{
3056 struct sk_buff *skb;
3057 long amount = 0;
3058
3059 if (sk->sk_state == TCP_LISTEN)
3060 return -EINVAL;
3061
3062 spin_lock(&sk->sk_receive_queue.lock);
3063 if (sk->sk_type == SOCK_STREAM ||
3064 sk->sk_type == SOCK_SEQPACKET) {
3065 skb_queue_walk(&sk->sk_receive_queue, skb)
3066 amount += unix_skb_len(skb);
3067 } else {
3068 skb = skb_peek(&sk->sk_receive_queue);
3069 if (skb)
3070 amount = skb->len;
3071 }
3072 spin_unlock(&sk->sk_receive_queue.lock);
3073
3074 return amount;
3075}
3076EXPORT_SYMBOL_GPL(unix_inq_len);
3077
3078long unix_outq_len(struct sock *sk)
3079{
3080 return sk_wmem_alloc_get(sk);
3081}
3082EXPORT_SYMBOL_GPL(unix_outq_len);
3083
3084static int unix_open_file(struct sock *sk)
3085{
3086 struct path path;
3087 struct file *f;
3088 int fd;
3089
3090 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3091 return -EPERM;
3092
3093 if (!smp_load_acquire(&unix_sk(sk)->addr))
3094 return -ENOENT;
3095
3096 path = unix_sk(sk)->path;
3097 if (!path.dentry)
3098 return -ENOENT;
3099
3100 path_get(&path);
3101
3102 fd = get_unused_fd_flags(O_CLOEXEC);
3103 if (fd < 0)
3104 goto out;
3105
3106 f = dentry_open(&path, O_PATH, current_cred());
3107 if (IS_ERR(f)) {
3108 put_unused_fd(fd);
3109 fd = PTR_ERR(f);
3110 goto out;
3111 }
3112
3113 fd_install(fd, f);
3114out:
3115 path_put(&path);
3116
3117 return fd;
3118}
3119
3120static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3121{
3122 struct sock *sk = sock->sk;
3123 long amount = 0;
3124 int err;
3125
3126 switch (cmd) {
3127 case SIOCOUTQ:
3128 amount = unix_outq_len(sk);
3129 err = put_user(amount, (int __user *)arg);
3130 break;
3131 case SIOCINQ:
3132 amount = unix_inq_len(sk);
3133 if (amount < 0)
3134 err = amount;
3135 else
3136 err = put_user(amount, (int __user *)arg);
3137 break;
3138 case SIOCUNIXFILE:
3139 err = unix_open_file(sk);
3140 break;
3141#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3142 case SIOCATMARK:
3143 {
3144 struct sk_buff *skb;
3145 int answ = 0;
3146
3147 skb = skb_peek(&sk->sk_receive_queue);
3148 if (skb && skb == READ_ONCE(unix_sk(sk)->oob_skb))
3149 answ = 1;
3150 err = put_user(answ, (int __user *)arg);
3151 }
3152 break;
3153#endif
3154 default:
3155 err = -ENOIOCTLCMD;
3156 break;
3157 }
3158 return err;
3159}
3160
3161#ifdef CONFIG_COMPAT
3162static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3163{
3164 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
3165}
3166#endif
3167
3168static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
3169{
3170 struct sock *sk = sock->sk;
3171 __poll_t mask;
3172 u8 shutdown;
3173
3174 sock_poll_wait(file, sock, wait);
3175 mask = 0;
3176 shutdown = READ_ONCE(sk->sk_shutdown);
3177
3178 /* exceptional events? */
3179 if (READ_ONCE(sk->sk_err))
3180 mask |= EPOLLERR;
3181 if (shutdown == SHUTDOWN_MASK)
3182 mask |= EPOLLHUP;
3183 if (shutdown & RCV_SHUTDOWN)
3184 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3185
3186 /* readable? */
3187 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3188 mask |= EPOLLIN | EPOLLRDNORM;
3189 if (sk_is_readable(sk))
3190 mask |= EPOLLIN | EPOLLRDNORM;
3191#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3192 if (READ_ONCE(unix_sk(sk)->oob_skb))
3193 mask |= EPOLLPRI;
3194#endif
3195
3196 /* Connection-based need to check for termination and startup */
3197 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3198 sk->sk_state == TCP_CLOSE)
3199 mask |= EPOLLHUP;
3200
3201 /*
3202 * we set writable also when the other side has shut down the
3203 * connection. This prevents stuck sockets.
3204 */
3205 if (unix_writable(sk))
3206 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3207
3208 return mask;
3209}
3210
3211static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3212 poll_table *wait)
3213{
3214 struct sock *sk = sock->sk, *other;
3215 unsigned int writable;
3216 __poll_t mask;
3217 u8 shutdown;
3218
3219 sock_poll_wait(file, sock, wait);
3220 mask = 0;
3221 shutdown = READ_ONCE(sk->sk_shutdown);
3222
3223 /* exceptional events? */
3224 if (READ_ONCE(sk->sk_err) ||
3225 !skb_queue_empty_lockless(&sk->sk_error_queue))
3226 mask |= EPOLLERR |
3227 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
3228
3229 if (shutdown & RCV_SHUTDOWN)
3230 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3231 if (shutdown == SHUTDOWN_MASK)
3232 mask |= EPOLLHUP;
3233
3234 /* readable? */
3235 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3236 mask |= EPOLLIN | EPOLLRDNORM;
3237 if (sk_is_readable(sk))
3238 mask |= EPOLLIN | EPOLLRDNORM;
3239
3240 /* Connection-based need to check for termination and startup */
3241 if (sk->sk_type == SOCK_SEQPACKET) {
3242 if (sk->sk_state == TCP_CLOSE)
3243 mask |= EPOLLHUP;
3244 /* connection hasn't started yet? */
3245 if (sk->sk_state == TCP_SYN_SENT)
3246 return mask;
3247 }
3248
3249 /* No write status requested, avoid expensive OUT tests. */
3250 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
3251 return mask;
3252
3253 writable = unix_writable(sk);
3254 if (writable) {
3255 unix_state_lock(sk);
3256
3257 other = unix_peer(sk);
3258 if (other && unix_peer(other) != sk &&
3259 unix_recvq_full_lockless(other) &&
3260 unix_dgram_peer_wake_me(sk, other))
3261 writable = 0;
3262
3263 unix_state_unlock(sk);
3264 }
3265
3266 if (writable)
3267 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3268 else
3269 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3270
3271 return mask;
3272}
3273
3274#ifdef CONFIG_PROC_FS
3275
3276#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3277
3278#define get_bucket(x) ((x) >> BUCKET_SPACE)
3279#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
3280#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
3281
3282static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
3283{
3284 unsigned long offset = get_offset(*pos);
3285 unsigned long bucket = get_bucket(*pos);
3286 unsigned long count = 0;
3287 struct sock *sk;
3288
3289 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3290 sk; sk = sk_next(sk)) {
3291 if (++count == offset)
3292 break;
3293 }
3294
3295 return sk;
3296}
3297
3298static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
3299{
3300 unsigned long bucket = get_bucket(*pos);
3301 struct net *net = seq_file_net(seq);
3302 struct sock *sk;
3303
3304 while (bucket < UNIX_HASH_SIZE) {
3305 spin_lock(&net->unx.table.locks[bucket]);
3306
3307 sk = unix_from_bucket(seq, pos);
3308 if (sk)
3309 return sk;
3310
3311 spin_unlock(&net->unx.table.locks[bucket]);
3312
3313 *pos = set_bucket_offset(++bucket, 1);
3314 }
3315
3316 return NULL;
3317}
3318
3319static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
3320 loff_t *pos)
3321{
3322 unsigned long bucket = get_bucket(*pos);
3323
3324 sk = sk_next(sk);
3325 if (sk)
3326 return sk;
3327
3328
3329 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
3330
3331 *pos = set_bucket_offset(++bucket, 1);
3332
3333 return unix_get_first(seq, pos);
3334}
3335
3336static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3337{
3338 if (!*pos)
3339 return SEQ_START_TOKEN;
3340
3341 return unix_get_first(seq, pos);
3342}
3343
3344static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3345{
3346 ++*pos;
3347
3348 if (v == SEQ_START_TOKEN)
3349 return unix_get_first(seq, pos);
3350
3351 return unix_get_next(seq, v, pos);
3352}
3353
3354static void unix_seq_stop(struct seq_file *seq, void *v)
3355{
3356 struct sock *sk = v;
3357
3358 if (sk)
3359 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
3360}
3361
3362static int unix_seq_show(struct seq_file *seq, void *v)
3363{
3364
3365 if (v == SEQ_START_TOKEN)
3366 seq_puts(seq, "Num RefCount Protocol Flags Type St "
3367 "Inode Path\n");
3368 else {
3369 struct sock *s = v;
3370 struct unix_sock *u = unix_sk(s);
3371 unix_state_lock(s);
3372
3373 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
3374 s,
3375 refcount_read(&s->sk_refcnt),
3376 0,
3377 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
3378 s->sk_type,
3379 s->sk_socket ?
3380 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
3381 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
3382 sock_i_ino(s));
3383
3384 if (u->addr) { // under a hash table lock here
3385 int i, len;
3386 seq_putc(seq, ' ');
3387
3388 i = 0;
3389 len = u->addr->len -
3390 offsetof(struct sockaddr_un, sun_path);
3391 if (u->addr->name->sun_path[0]) {
3392 len--;
3393 } else {
3394 seq_putc(seq, '@');
3395 i++;
3396 }
3397 for ( ; i < len; i++)
3398 seq_putc(seq, u->addr->name->sun_path[i] ?:
3399 '@');
3400 }
3401 unix_state_unlock(s);
3402 seq_putc(seq, '\n');
3403 }
3404
3405 return 0;
3406}
3407
3408static const struct seq_operations unix_seq_ops = {
3409 .start = unix_seq_start,
3410 .next = unix_seq_next,
3411 .stop = unix_seq_stop,
3412 .show = unix_seq_show,
3413};
3414
3415#ifdef CONFIG_BPF_SYSCALL
3416struct bpf_unix_iter_state {
3417 struct seq_net_private p;
3418 unsigned int cur_sk;
3419 unsigned int end_sk;
3420 unsigned int max_sk;
3421 struct sock **batch;
3422 bool st_bucket_done;
3423};
3424
3425struct bpf_iter__unix {
3426 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3427 __bpf_md_ptr(struct unix_sock *, unix_sk);
3428 uid_t uid __aligned(8);
3429};
3430
3431static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3432 struct unix_sock *unix_sk, uid_t uid)
3433{
3434 struct bpf_iter__unix ctx;
3435
3436 meta->seq_num--; /* skip SEQ_START_TOKEN */
3437 ctx.meta = meta;
3438 ctx.unix_sk = unix_sk;
3439 ctx.uid = uid;
3440 return bpf_iter_run_prog(prog, &ctx);
3441}
3442
3443static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
3444
3445{
3446 struct bpf_unix_iter_state *iter = seq->private;
3447 unsigned int expected = 1;
3448 struct sock *sk;
3449
3450 sock_hold(start_sk);
3451 iter->batch[iter->end_sk++] = start_sk;
3452
3453 for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
3454 if (iter->end_sk < iter->max_sk) {
3455 sock_hold(sk);
3456 iter->batch[iter->end_sk++] = sk;
3457 }
3458
3459 expected++;
3460 }
3461
3462 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
3463
3464 return expected;
3465}
3466
3467static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
3468{
3469 while (iter->cur_sk < iter->end_sk)
3470 sock_put(iter->batch[iter->cur_sk++]);
3471}
3472
3473static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
3474 unsigned int new_batch_sz)
3475{
3476 struct sock **new_batch;
3477
3478 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3479 GFP_USER | __GFP_NOWARN);
3480 if (!new_batch)
3481 return -ENOMEM;
3482
3483 bpf_iter_unix_put_batch(iter);
3484 kvfree(iter->batch);
3485 iter->batch = new_batch;
3486 iter->max_sk = new_batch_sz;
3487
3488 return 0;
3489}
3490
3491static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
3492 loff_t *pos)
3493{
3494 struct bpf_unix_iter_state *iter = seq->private;
3495 unsigned int expected;
3496 bool resized = false;
3497 struct sock *sk;
3498
3499 if (iter->st_bucket_done)
3500 *pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
3501
3502again:
3503 /* Get a new batch */
3504 iter->cur_sk = 0;
3505 iter->end_sk = 0;
3506
3507 sk = unix_get_first(seq, pos);
3508 if (!sk)
3509 return NULL; /* Done */
3510
3511 expected = bpf_iter_unix_hold_batch(seq, sk);
3512
3513 if (iter->end_sk == expected) {
3514 iter->st_bucket_done = true;
3515 return sk;
3516 }
3517
3518 if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
3519 resized = true;
3520 goto again;
3521 }
3522
3523 return sk;
3524}
3525
3526static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
3527{
3528 if (!*pos)
3529 return SEQ_START_TOKEN;
3530
3531 /* bpf iter does not support lseek, so it always
3532 * continue from where it was stop()-ped.
3533 */
3534 return bpf_iter_unix_batch(seq, pos);
3535}
3536
3537static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3538{
3539 struct bpf_unix_iter_state *iter = seq->private;
3540 struct sock *sk;
3541
3542 /* Whenever seq_next() is called, the iter->cur_sk is
3543 * done with seq_show(), so advance to the next sk in
3544 * the batch.
3545 */
3546 if (iter->cur_sk < iter->end_sk)
3547 sock_put(iter->batch[iter->cur_sk++]);
3548
3549 ++*pos;
3550
3551 if (iter->cur_sk < iter->end_sk)
3552 sk = iter->batch[iter->cur_sk];
3553 else
3554 sk = bpf_iter_unix_batch(seq, pos);
3555
3556 return sk;
3557}
3558
3559static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
3560{
3561 struct bpf_iter_meta meta;
3562 struct bpf_prog *prog;
3563 struct sock *sk = v;
3564 uid_t uid;
3565 bool slow;
3566 int ret;
3567
3568 if (v == SEQ_START_TOKEN)
3569 return 0;
3570
3571 slow = lock_sock_fast(sk);
3572
3573 if (unlikely(sk_unhashed(sk))) {
3574 ret = SEQ_SKIP;
3575 goto unlock;
3576 }
3577
3578 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3579 meta.seq = seq;
3580 prog = bpf_iter_get_info(&meta, false);
3581 ret = unix_prog_seq_show(prog, &meta, v, uid);
3582unlock:
3583 unlock_sock_fast(sk, slow);
3584 return ret;
3585}
3586
3587static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3588{
3589 struct bpf_unix_iter_state *iter = seq->private;
3590 struct bpf_iter_meta meta;
3591 struct bpf_prog *prog;
3592
3593 if (!v) {
3594 meta.seq = seq;
3595 prog = bpf_iter_get_info(&meta, true);
3596 if (prog)
3597 (void)unix_prog_seq_show(prog, &meta, v, 0);
3598 }
3599
3600 if (iter->cur_sk < iter->end_sk)
3601 bpf_iter_unix_put_batch(iter);
3602}
3603
3604static const struct seq_operations bpf_iter_unix_seq_ops = {
3605 .start = bpf_iter_unix_seq_start,
3606 .next = bpf_iter_unix_seq_next,
3607 .stop = bpf_iter_unix_seq_stop,
3608 .show = bpf_iter_unix_seq_show,
3609};
3610#endif
3611#endif
3612
3613static const struct net_proto_family unix_family_ops = {
3614 .family = PF_UNIX,
3615 .create = unix_create,
3616 .owner = THIS_MODULE,
3617};
3618
3619
3620static int __net_init unix_net_init(struct net *net)
3621{
3622 int i;
3623
3624 net->unx.sysctl_max_dgram_qlen = 10;
3625 if (unix_sysctl_register(net))
3626 goto out;
3627
3628#ifdef CONFIG_PROC_FS
3629 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
3630 sizeof(struct seq_net_private)))
3631 goto err_sysctl;
3632#endif
3633
3634 net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
3635 sizeof(spinlock_t), GFP_KERNEL);
3636 if (!net->unx.table.locks)
3637 goto err_proc;
3638
3639 net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
3640 sizeof(struct hlist_head),
3641 GFP_KERNEL);
3642 if (!net->unx.table.buckets)
3643 goto free_locks;
3644
3645 for (i = 0; i < UNIX_HASH_SIZE; i++) {
3646 spin_lock_init(&net->unx.table.locks[i]);
3647 INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
3648 }
3649
3650 return 0;
3651
3652free_locks:
3653 kvfree(net->unx.table.locks);
3654err_proc:
3655#ifdef CONFIG_PROC_FS
3656 remove_proc_entry("unix", net->proc_net);
3657err_sysctl:
3658#endif
3659 unix_sysctl_unregister(net);
3660out:
3661 return -ENOMEM;
3662}
3663
3664static void __net_exit unix_net_exit(struct net *net)
3665{
3666 kvfree(net->unx.table.buckets);
3667 kvfree(net->unx.table.locks);
3668 unix_sysctl_unregister(net);
3669 remove_proc_entry("unix", net->proc_net);
3670}
3671
3672static struct pernet_operations unix_net_ops = {
3673 .init = unix_net_init,
3674 .exit = unix_net_exit,
3675};
3676
3677#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3678DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
3679 struct unix_sock *unix_sk, uid_t uid)
3680
3681#define INIT_BATCH_SZ 16
3682
3683static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
3684{
3685 struct bpf_unix_iter_state *iter = priv_data;
3686 int err;
3687
3688 err = bpf_iter_init_seq_net(priv_data, aux);
3689 if (err)
3690 return err;
3691
3692 err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
3693 if (err) {
3694 bpf_iter_fini_seq_net(priv_data);
3695 return err;
3696 }
3697
3698 return 0;
3699}
3700
3701static void bpf_iter_fini_unix(void *priv_data)
3702{
3703 struct bpf_unix_iter_state *iter = priv_data;
3704
3705 bpf_iter_fini_seq_net(priv_data);
3706 kvfree(iter->batch);
3707}
3708
3709static const struct bpf_iter_seq_info unix_seq_info = {
3710 .seq_ops = &bpf_iter_unix_seq_ops,
3711 .init_seq_private = bpf_iter_init_unix,
3712 .fini_seq_private = bpf_iter_fini_unix,
3713 .seq_priv_size = sizeof(struct bpf_unix_iter_state),
3714};
3715
3716static const struct bpf_func_proto *
3717bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
3718 const struct bpf_prog *prog)
3719{
3720 switch (func_id) {
3721 case BPF_FUNC_setsockopt:
3722 return &bpf_sk_setsockopt_proto;
3723 case BPF_FUNC_getsockopt:
3724 return &bpf_sk_getsockopt_proto;
3725 default:
3726 return NULL;
3727 }
3728}
3729
3730static struct bpf_iter_reg unix_reg_info = {
3731 .target = "unix",
3732 .ctx_arg_info_size = 1,
3733 .ctx_arg_info = {
3734 { offsetof(struct bpf_iter__unix, unix_sk),
3735 PTR_TO_BTF_ID_OR_NULL },
3736 },
3737 .get_func_proto = bpf_iter_unix_get_func_proto,
3738 .seq_info = &unix_seq_info,
3739};
3740
3741static void __init bpf_iter_register(void)
3742{
3743 unix_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UNIX];
3744 if (bpf_iter_reg_target(&unix_reg_info))
3745 pr_warn("Warning: could not register bpf iterator unix\n");
3746}
3747#endif
3748
3749static int __init af_unix_init(void)
3750{
3751 int i, rc = -1;
3752
3753 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
3754
3755 for (i = 0; i < UNIX_HASH_SIZE / 2; i++) {
3756 spin_lock_init(&bsd_socket_locks[i]);
3757 INIT_HLIST_HEAD(&bsd_socket_buckets[i]);
3758 }
3759
3760 rc = proto_register(&unix_dgram_proto, 1);
3761 if (rc != 0) {
3762 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3763 goto out;
3764 }
3765
3766 rc = proto_register(&unix_stream_proto, 1);
3767 if (rc != 0) {
3768 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3769 proto_unregister(&unix_dgram_proto);
3770 goto out;
3771 }
3772
3773 sock_register(&unix_family_ops);
3774 register_pernet_subsys(&unix_net_ops);
3775 unix_bpf_build_proto();
3776
3777#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3778 bpf_iter_register();
3779#endif
3780
3781out:
3782 return rc;
3783}
3784
3785/* Later than subsys_initcall() because we depend on stuff initialised there */
3786fs_initcall(af_unix_init);
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/termios.h>
93#include <linux/sockios.h>
94#include <linux/net.h>
95#include <linux/in.h>
96#include <linux/fs.h>
97#include <linux/slab.h>
98#include <linux/uaccess.h>
99#include <linux/skbuff.h>
100#include <linux/netdevice.h>
101#include <net/net_namespace.h>
102#include <net/sock.h>
103#include <net/tcp_states.h>
104#include <net/af_unix.h>
105#include <linux/proc_fs.h>
106#include <linux/seq_file.h>
107#include <net/scm.h>
108#include <linux/init.h>
109#include <linux/poll.h>
110#include <linux/rtnetlink.h>
111#include <linux/mount.h>
112#include <net/checksum.h>
113#include <linux/security.h>
114#include <linux/freezer.h>
115#include <linux/file.h>
116
117#include "scm.h"
118
119struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
120EXPORT_SYMBOL_GPL(unix_socket_table);
121DEFINE_SPINLOCK(unix_table_lock);
122EXPORT_SYMBOL_GPL(unix_table_lock);
123static atomic_long_t unix_nr_socks;
124
125
126static struct hlist_head *unix_sockets_unbound(void *addr)
127{
128 unsigned long hash = (unsigned long)addr;
129
130 hash ^= hash >> 16;
131 hash ^= hash >> 8;
132 hash %= UNIX_HASH_SIZE;
133 return &unix_socket_table[UNIX_HASH_SIZE + hash];
134}
135
136#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
137
138#ifdef CONFIG_SECURITY_NETWORK
139static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140{
141 UNIXCB(skb).secid = scm->secid;
142}
143
144static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
145{
146 scm->secid = UNIXCB(skb).secid;
147}
148
149static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
150{
151 return (scm->secid == UNIXCB(skb).secid);
152}
153#else
154static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
155{ }
156
157static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
158{ }
159
160static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
161{
162 return true;
163}
164#endif /* CONFIG_SECURITY_NETWORK */
165
166/*
167 * SMP locking strategy:
168 * hash table is protected with spinlock unix_table_lock
169 * each socket state is protected by separate spin lock.
170 */
171
172static inline unsigned int unix_hash_fold(__wsum n)
173{
174 unsigned int hash = (__force unsigned int)csum_fold(n);
175
176 hash ^= hash>>8;
177 return hash&(UNIX_HASH_SIZE-1);
178}
179
180#define unix_peer(sk) (unix_sk(sk)->peer)
181
182static inline int unix_our_peer(struct sock *sk, struct sock *osk)
183{
184 return unix_peer(osk) == sk;
185}
186
187static inline int unix_may_send(struct sock *sk, struct sock *osk)
188{
189 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
190}
191
192static inline int unix_recvq_full(const struct sock *sk)
193{
194 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
195}
196
197static inline int unix_recvq_full_lockless(const struct sock *sk)
198{
199 return skb_queue_len_lockless(&sk->sk_receive_queue) >
200 READ_ONCE(sk->sk_max_ack_backlog);
201}
202
203struct sock *unix_peer_get(struct sock *s)
204{
205 struct sock *peer;
206
207 unix_state_lock(s);
208 peer = unix_peer(s);
209 if (peer)
210 sock_hold(peer);
211 unix_state_unlock(s);
212 return peer;
213}
214EXPORT_SYMBOL_GPL(unix_peer_get);
215
216static inline void unix_release_addr(struct unix_address *addr)
217{
218 if (refcount_dec_and_test(&addr->refcnt))
219 kfree(addr);
220}
221
222/*
223 * Check unix socket name:
224 * - should be not zero length.
225 * - if started by not zero, should be NULL terminated (FS object)
226 * - if started by zero, it is abstract name.
227 */
228
229static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
230{
231 *hashp = 0;
232
233 if (len <= sizeof(short) || len > sizeof(*sunaddr))
234 return -EINVAL;
235 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
236 return -EINVAL;
237 if (sunaddr->sun_path[0]) {
238 /*
239 * This may look like an off by one error but it is a bit more
240 * subtle. 108 is the longest valid AF_UNIX path for a binding.
241 * sun_path[108] doesn't as such exist. However in kernel space
242 * we are guaranteed that it is a valid memory location in our
243 * kernel address buffer.
244 */
245 ((char *)sunaddr)[len] = 0;
246 len = strlen(sunaddr->sun_path)+1+sizeof(short);
247 return len;
248 }
249
250 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
251 return len;
252}
253
254static void __unix_remove_socket(struct sock *sk)
255{
256 sk_del_node_init(sk);
257}
258
259static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
260{
261 WARN_ON(!sk_unhashed(sk));
262 sk_add_node(sk, list);
263}
264
265static inline void unix_remove_socket(struct sock *sk)
266{
267 spin_lock(&unix_table_lock);
268 __unix_remove_socket(sk);
269 spin_unlock(&unix_table_lock);
270}
271
272static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
273{
274 spin_lock(&unix_table_lock);
275 __unix_insert_socket(list, sk);
276 spin_unlock(&unix_table_lock);
277}
278
279static struct sock *__unix_find_socket_byname(struct net *net,
280 struct sockaddr_un *sunname,
281 int len, int type, unsigned int hash)
282{
283 struct sock *s;
284
285 sk_for_each(s, &unix_socket_table[hash ^ type]) {
286 struct unix_sock *u = unix_sk(s);
287
288 if (!net_eq(sock_net(s), net))
289 continue;
290
291 if (u->addr->len == len &&
292 !memcmp(u->addr->name, sunname, len))
293 return s;
294 }
295 return NULL;
296}
297
298static inline struct sock *unix_find_socket_byname(struct net *net,
299 struct sockaddr_un *sunname,
300 int len, int type,
301 unsigned int hash)
302{
303 struct sock *s;
304
305 spin_lock(&unix_table_lock);
306 s = __unix_find_socket_byname(net, sunname, len, type, hash);
307 if (s)
308 sock_hold(s);
309 spin_unlock(&unix_table_lock);
310 return s;
311}
312
313static struct sock *unix_find_socket_byinode(struct inode *i)
314{
315 struct sock *s;
316
317 spin_lock(&unix_table_lock);
318 sk_for_each(s,
319 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
320 struct dentry *dentry = unix_sk(s)->path.dentry;
321
322 if (dentry && d_backing_inode(dentry) == i) {
323 sock_hold(s);
324 goto found;
325 }
326 }
327 s = NULL;
328found:
329 spin_unlock(&unix_table_lock);
330 return s;
331}
332
333/* Support code for asymmetrically connected dgram sockets
334 *
335 * If a datagram socket is connected to a socket not itself connected
336 * to the first socket (eg, /dev/log), clients may only enqueue more
337 * messages if the present receive queue of the server socket is not
338 * "too large". This means there's a second writeability condition
339 * poll and sendmsg need to test. The dgram recv code will do a wake
340 * up on the peer_wait wait queue of a socket upon reception of a
341 * datagram which needs to be propagated to sleeping would-be writers
342 * since these might not have sent anything so far. This can't be
343 * accomplished via poll_wait because the lifetime of the server
344 * socket might be less than that of its clients if these break their
345 * association with it or if the server socket is closed while clients
346 * are still connected to it and there's no way to inform "a polling
347 * implementation" that it should let go of a certain wait queue
348 *
349 * In order to propagate a wake up, a wait_queue_entry_t of the client
350 * socket is enqueued on the peer_wait queue of the server socket
351 * whose wake function does a wake_up on the ordinary client socket
352 * wait queue. This connection is established whenever a write (or
353 * poll for write) hit the flow control condition and broken when the
354 * association to the server socket is dissolved or after a wake up
355 * was relayed.
356 */
357
358static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
359 void *key)
360{
361 struct unix_sock *u;
362 wait_queue_head_t *u_sleep;
363
364 u = container_of(q, struct unix_sock, peer_wake);
365
366 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
367 q);
368 u->peer_wake.private = NULL;
369
370 /* relaying can only happen while the wq still exists */
371 u_sleep = sk_sleep(&u->sk);
372 if (u_sleep)
373 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
374
375 return 0;
376}
377
378static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
379{
380 struct unix_sock *u, *u_other;
381 int rc;
382
383 u = unix_sk(sk);
384 u_other = unix_sk(other);
385 rc = 0;
386 spin_lock(&u_other->peer_wait.lock);
387
388 if (!u->peer_wake.private) {
389 u->peer_wake.private = other;
390 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
391
392 rc = 1;
393 }
394
395 spin_unlock(&u_other->peer_wait.lock);
396 return rc;
397}
398
399static void unix_dgram_peer_wake_disconnect(struct sock *sk,
400 struct sock *other)
401{
402 struct unix_sock *u, *u_other;
403
404 u = unix_sk(sk);
405 u_other = unix_sk(other);
406 spin_lock(&u_other->peer_wait.lock);
407
408 if (u->peer_wake.private == other) {
409 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
410 u->peer_wake.private = NULL;
411 }
412
413 spin_unlock(&u_other->peer_wait.lock);
414}
415
416static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
417 struct sock *other)
418{
419 unix_dgram_peer_wake_disconnect(sk, other);
420 wake_up_interruptible_poll(sk_sleep(sk),
421 EPOLLOUT |
422 EPOLLWRNORM |
423 EPOLLWRBAND);
424}
425
426/* preconditions:
427 * - unix_peer(sk) == other
428 * - association is stable
429 */
430static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
431{
432 int connected;
433
434 connected = unix_dgram_peer_wake_connect(sk, other);
435
436 /* If other is SOCK_DEAD, we want to make sure we signal
437 * POLLOUT, such that a subsequent write() can get a
438 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
439 * to other and its full, we will hang waiting for POLLOUT.
440 */
441 if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
442 return 1;
443
444 if (connected)
445 unix_dgram_peer_wake_disconnect(sk, other);
446
447 return 0;
448}
449
450static int unix_writable(const struct sock *sk)
451{
452 return sk->sk_state != TCP_LISTEN &&
453 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
454}
455
456static void unix_write_space(struct sock *sk)
457{
458 struct socket_wq *wq;
459
460 rcu_read_lock();
461 if (unix_writable(sk)) {
462 wq = rcu_dereference(sk->sk_wq);
463 if (skwq_has_sleeper(wq))
464 wake_up_interruptible_sync_poll(&wq->wait,
465 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
466 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
467 }
468 rcu_read_unlock();
469}
470
471/* When dgram socket disconnects (or changes its peer), we clear its receive
472 * queue of packets arrived from previous peer. First, it allows to do
473 * flow control based only on wmem_alloc; second, sk connected to peer
474 * may receive messages only from that peer. */
475static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
476{
477 if (!skb_queue_empty(&sk->sk_receive_queue)) {
478 skb_queue_purge(&sk->sk_receive_queue);
479 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
480
481 /* If one link of bidirectional dgram pipe is disconnected,
482 * we signal error. Messages are lost. Do not make this,
483 * when peer was not connected to us.
484 */
485 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
486 other->sk_err = ECONNRESET;
487 other->sk_error_report(other);
488 }
489 }
490}
491
492static void unix_sock_destructor(struct sock *sk)
493{
494 struct unix_sock *u = unix_sk(sk);
495
496 skb_queue_purge(&sk->sk_receive_queue);
497
498 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
499 WARN_ON(!sk_unhashed(sk));
500 WARN_ON(sk->sk_socket);
501 if (!sock_flag(sk, SOCK_DEAD)) {
502 pr_info("Attempt to release alive unix socket: %p\n", sk);
503 return;
504 }
505
506 if (u->addr)
507 unix_release_addr(u->addr);
508
509 atomic_long_dec(&unix_nr_socks);
510 local_bh_disable();
511 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
512 local_bh_enable();
513#ifdef UNIX_REFCNT_DEBUG
514 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
515 atomic_long_read(&unix_nr_socks));
516#endif
517}
518
519static void unix_release_sock(struct sock *sk, int embrion)
520{
521 struct unix_sock *u = unix_sk(sk);
522 struct path path;
523 struct sock *skpair;
524 struct sk_buff *skb;
525 int state;
526
527 unix_remove_socket(sk);
528
529 /* Clear state */
530 unix_state_lock(sk);
531 sock_orphan(sk);
532 sk->sk_shutdown = SHUTDOWN_MASK;
533 path = u->path;
534 u->path.dentry = NULL;
535 u->path.mnt = NULL;
536 state = sk->sk_state;
537 sk->sk_state = TCP_CLOSE;
538 unix_state_unlock(sk);
539
540 wake_up_interruptible_all(&u->peer_wait);
541
542 skpair = unix_peer(sk);
543
544 if (skpair != NULL) {
545 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
546 unix_state_lock(skpair);
547 /* No more writes */
548 skpair->sk_shutdown = SHUTDOWN_MASK;
549 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
550 skpair->sk_err = ECONNRESET;
551 unix_state_unlock(skpair);
552 skpair->sk_state_change(skpair);
553 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
554 }
555
556 unix_dgram_peer_wake_disconnect(sk, skpair);
557 sock_put(skpair); /* It may now die */
558 unix_peer(sk) = NULL;
559 }
560
561 /* Try to flush out this socket. Throw out buffers at least */
562
563 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
564 if (state == TCP_LISTEN)
565 unix_release_sock(skb->sk, 1);
566 /* passed fds are erased in the kfree_skb hook */
567 UNIXCB(skb).consumed = skb->len;
568 kfree_skb(skb);
569 }
570
571 if (path.dentry)
572 path_put(&path);
573
574 sock_put(sk);
575
576 /* ---- Socket is dead now and most probably destroyed ---- */
577
578 /*
579 * Fixme: BSD difference: In BSD all sockets connected to us get
580 * ECONNRESET and we die on the spot. In Linux we behave
581 * like files and pipes do and wait for the last
582 * dereference.
583 *
584 * Can't we simply set sock->err?
585 *
586 * What the above comment does talk about? --ANK(980817)
587 */
588
589 if (unix_tot_inflight)
590 unix_gc(); /* Garbage collect fds */
591}
592
593static void init_peercred(struct sock *sk)
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(task_tgid(current));
599 sk->sk_peer_cred = get_current_cred();
600}
601
602static void copy_peercred(struct sock *sk, struct sock *peersk)
603{
604 put_pid(sk->sk_peer_pid);
605 if (sk->sk_peer_cred)
606 put_cred(sk->sk_peer_cred);
607 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
608 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
609}
610
611static int unix_listen(struct socket *sock, int backlog)
612{
613 int err;
614 struct sock *sk = sock->sk;
615 struct unix_sock *u = unix_sk(sk);
616 struct pid *old_pid = NULL;
617
618 err = -EOPNOTSUPP;
619 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
620 goto out; /* Only stream/seqpacket sockets accept */
621 err = -EINVAL;
622 if (!u->addr)
623 goto out; /* No listens on an unbound socket */
624 unix_state_lock(sk);
625 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
626 goto out_unlock;
627 if (backlog > sk->sk_max_ack_backlog)
628 wake_up_interruptible_all(&u->peer_wait);
629 sk->sk_max_ack_backlog = backlog;
630 sk->sk_state = TCP_LISTEN;
631 /* set credentials so connect can copy them */
632 init_peercred(sk);
633 err = 0;
634
635out_unlock:
636 unix_state_unlock(sk);
637 put_pid(old_pid);
638out:
639 return err;
640}
641
642static int unix_release(struct socket *);
643static int unix_bind(struct socket *, struct sockaddr *, int);
644static int unix_stream_connect(struct socket *, struct sockaddr *,
645 int addr_len, int flags);
646static int unix_socketpair(struct socket *, struct socket *);
647static int unix_accept(struct socket *, struct socket *, int, bool);
648static int unix_getname(struct socket *, struct sockaddr *, int);
649static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
650static __poll_t unix_dgram_poll(struct file *, struct socket *,
651 poll_table *);
652static int unix_ioctl(struct socket *, unsigned int, unsigned long);
653#ifdef CONFIG_COMPAT
654static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
655#endif
656static int unix_shutdown(struct socket *, int);
657static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
658static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
659static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
660 size_t size, int flags);
661static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
662 struct pipe_inode_info *, size_t size,
663 unsigned int flags);
664static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
665static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
666static int unix_dgram_connect(struct socket *, struct sockaddr *,
667 int, int);
668static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
669static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
670 int);
671
672static int unix_set_peek_off(struct sock *sk, int val)
673{
674 struct unix_sock *u = unix_sk(sk);
675
676 if (mutex_lock_interruptible(&u->iolock))
677 return -EINTR;
678
679 sk->sk_peek_off = val;
680 mutex_unlock(&u->iolock);
681
682 return 0;
683}
684
685#ifdef CONFIG_PROC_FS
686static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
687{
688 struct sock *sk = sock->sk;
689 struct unix_sock *u;
690
691 if (sk) {
692 u = unix_sk(sock->sk);
693 seq_printf(m, "scm_fds: %u\n",
694 atomic_read(&u->scm_stat.nr_fds));
695 }
696}
697#else
698#define unix_show_fdinfo NULL
699#endif
700
701static const struct proto_ops unix_stream_ops = {
702 .family = PF_UNIX,
703 .owner = THIS_MODULE,
704 .release = unix_release,
705 .bind = unix_bind,
706 .connect = unix_stream_connect,
707 .socketpair = unix_socketpair,
708 .accept = unix_accept,
709 .getname = unix_getname,
710 .poll = unix_poll,
711 .ioctl = unix_ioctl,
712#ifdef CONFIG_COMPAT
713 .compat_ioctl = unix_compat_ioctl,
714#endif
715 .listen = unix_listen,
716 .shutdown = unix_shutdown,
717 .sendmsg = unix_stream_sendmsg,
718 .recvmsg = unix_stream_recvmsg,
719 .mmap = sock_no_mmap,
720 .sendpage = unix_stream_sendpage,
721 .splice_read = unix_stream_splice_read,
722 .set_peek_off = unix_set_peek_off,
723 .show_fdinfo = unix_show_fdinfo,
724};
725
726static const struct proto_ops unix_dgram_ops = {
727 .family = PF_UNIX,
728 .owner = THIS_MODULE,
729 .release = unix_release,
730 .bind = unix_bind,
731 .connect = unix_dgram_connect,
732 .socketpair = unix_socketpair,
733 .accept = sock_no_accept,
734 .getname = unix_getname,
735 .poll = unix_dgram_poll,
736 .ioctl = unix_ioctl,
737#ifdef CONFIG_COMPAT
738 .compat_ioctl = unix_compat_ioctl,
739#endif
740 .listen = sock_no_listen,
741 .shutdown = unix_shutdown,
742 .sendmsg = unix_dgram_sendmsg,
743 .recvmsg = unix_dgram_recvmsg,
744 .mmap = sock_no_mmap,
745 .sendpage = sock_no_sendpage,
746 .set_peek_off = unix_set_peek_off,
747 .show_fdinfo = unix_show_fdinfo,
748};
749
750static const struct proto_ops unix_seqpacket_ops = {
751 .family = PF_UNIX,
752 .owner = THIS_MODULE,
753 .release = unix_release,
754 .bind = unix_bind,
755 .connect = unix_stream_connect,
756 .socketpair = unix_socketpair,
757 .accept = unix_accept,
758 .getname = unix_getname,
759 .poll = unix_dgram_poll,
760 .ioctl = unix_ioctl,
761#ifdef CONFIG_COMPAT
762 .compat_ioctl = unix_compat_ioctl,
763#endif
764 .listen = unix_listen,
765 .shutdown = unix_shutdown,
766 .sendmsg = unix_seqpacket_sendmsg,
767 .recvmsg = unix_seqpacket_recvmsg,
768 .mmap = sock_no_mmap,
769 .sendpage = sock_no_sendpage,
770 .set_peek_off = unix_set_peek_off,
771 .show_fdinfo = unix_show_fdinfo,
772};
773
774static struct proto unix_proto = {
775 .name = "UNIX",
776 .owner = THIS_MODULE,
777 .obj_size = sizeof(struct unix_sock),
778};
779
780static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
781{
782 struct sock *sk = NULL;
783 struct unix_sock *u;
784
785 atomic_long_inc(&unix_nr_socks);
786 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
787 goto out;
788
789 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
790 if (!sk)
791 goto out;
792
793 sock_init_data(sock, sk);
794
795 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
796 sk->sk_write_space = unix_write_space;
797 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
798 sk->sk_destruct = unix_sock_destructor;
799 u = unix_sk(sk);
800 u->path.dentry = NULL;
801 u->path.mnt = NULL;
802 spin_lock_init(&u->lock);
803 atomic_long_set(&u->inflight, 0);
804 INIT_LIST_HEAD(&u->link);
805 mutex_init(&u->iolock); /* single task reading lock */
806 mutex_init(&u->bindlock); /* single task binding lock */
807 init_waitqueue_head(&u->peer_wait);
808 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
809 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
810 unix_insert_socket(unix_sockets_unbound(sk), sk);
811out:
812 if (sk == NULL)
813 atomic_long_dec(&unix_nr_socks);
814 else {
815 local_bh_disable();
816 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
817 local_bh_enable();
818 }
819 return sk;
820}
821
822static int unix_create(struct net *net, struct socket *sock, int protocol,
823 int kern)
824{
825 if (protocol && protocol != PF_UNIX)
826 return -EPROTONOSUPPORT;
827
828 sock->state = SS_UNCONNECTED;
829
830 switch (sock->type) {
831 case SOCK_STREAM:
832 sock->ops = &unix_stream_ops;
833 break;
834 /*
835 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
836 * nothing uses it.
837 */
838 case SOCK_RAW:
839 sock->type = SOCK_DGRAM;
840 fallthrough;
841 case SOCK_DGRAM:
842 sock->ops = &unix_dgram_ops;
843 break;
844 case SOCK_SEQPACKET:
845 sock->ops = &unix_seqpacket_ops;
846 break;
847 default:
848 return -ESOCKTNOSUPPORT;
849 }
850
851 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
852}
853
854static int unix_release(struct socket *sock)
855{
856 struct sock *sk = sock->sk;
857
858 if (!sk)
859 return 0;
860
861 unix_release_sock(sk, 0);
862 sock->sk = NULL;
863
864 return 0;
865}
866
867static int unix_autobind(struct socket *sock)
868{
869 struct sock *sk = sock->sk;
870 struct net *net = sock_net(sk);
871 struct unix_sock *u = unix_sk(sk);
872 static u32 ordernum = 1;
873 struct unix_address *addr;
874 int err;
875 unsigned int retries = 0;
876
877 err = mutex_lock_interruptible(&u->bindlock);
878 if (err)
879 return err;
880
881 err = 0;
882 if (u->addr)
883 goto out;
884
885 err = -ENOMEM;
886 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
887 if (!addr)
888 goto out;
889
890 addr->name->sun_family = AF_UNIX;
891 refcount_set(&addr->refcnt, 1);
892
893retry:
894 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
895 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
896
897 spin_lock(&unix_table_lock);
898 ordernum = (ordernum+1)&0xFFFFF;
899
900 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
901 addr->hash)) {
902 spin_unlock(&unix_table_lock);
903 /*
904 * __unix_find_socket_byname() may take long time if many names
905 * are already in use.
906 */
907 cond_resched();
908 /* Give up if all names seems to be in use. */
909 if (retries++ == 0xFFFFF) {
910 err = -ENOSPC;
911 kfree(addr);
912 goto out;
913 }
914 goto retry;
915 }
916 addr->hash ^= sk->sk_type;
917
918 __unix_remove_socket(sk);
919 smp_store_release(&u->addr, addr);
920 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
921 spin_unlock(&unix_table_lock);
922 err = 0;
923
924out: mutex_unlock(&u->bindlock);
925 return err;
926}
927
928static struct sock *unix_find_other(struct net *net,
929 struct sockaddr_un *sunname, int len,
930 int type, unsigned int hash, int *error)
931{
932 struct sock *u;
933 struct path path;
934 int err = 0;
935
936 if (sunname->sun_path[0]) {
937 struct inode *inode;
938 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
939 if (err)
940 goto fail;
941 inode = d_backing_inode(path.dentry);
942 err = inode_permission(inode, MAY_WRITE);
943 if (err)
944 goto put_fail;
945
946 err = -ECONNREFUSED;
947 if (!S_ISSOCK(inode->i_mode))
948 goto put_fail;
949 u = unix_find_socket_byinode(inode);
950 if (!u)
951 goto put_fail;
952
953 if (u->sk_type == type)
954 touch_atime(&path);
955
956 path_put(&path);
957
958 err = -EPROTOTYPE;
959 if (u->sk_type != type) {
960 sock_put(u);
961 goto fail;
962 }
963 } else {
964 err = -ECONNREFUSED;
965 u = unix_find_socket_byname(net, sunname, len, type, hash);
966 if (u) {
967 struct dentry *dentry;
968 dentry = unix_sk(u)->path.dentry;
969 if (dentry)
970 touch_atime(&unix_sk(u)->path);
971 } else
972 goto fail;
973 }
974 return u;
975
976put_fail:
977 path_put(&path);
978fail:
979 *error = err;
980 return NULL;
981}
982
983static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
984{
985 struct dentry *dentry;
986 struct path path;
987 int err = 0;
988 /*
989 * Get the parent directory, calculate the hash for last
990 * component.
991 */
992 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
993 err = PTR_ERR(dentry);
994 if (IS_ERR(dentry))
995 return err;
996
997 /*
998 * All right, let's create it.
999 */
1000 err = security_path_mknod(&path, dentry, mode, 0);
1001 if (!err) {
1002 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
1003 if (!err) {
1004 res->mnt = mntget(path.mnt);
1005 res->dentry = dget(dentry);
1006 }
1007 }
1008 done_path_create(&path, dentry);
1009 return err;
1010}
1011
1012static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1013{
1014 struct sock *sk = sock->sk;
1015 struct net *net = sock_net(sk);
1016 struct unix_sock *u = unix_sk(sk);
1017 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1018 char *sun_path = sunaddr->sun_path;
1019 int err;
1020 unsigned int hash;
1021 struct unix_address *addr;
1022 struct hlist_head *list;
1023 struct path path = { };
1024
1025 err = -EINVAL;
1026 if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1027 sunaddr->sun_family != AF_UNIX)
1028 goto out;
1029
1030 if (addr_len == sizeof(short)) {
1031 err = unix_autobind(sock);
1032 goto out;
1033 }
1034
1035 err = unix_mkname(sunaddr, addr_len, &hash);
1036 if (err < 0)
1037 goto out;
1038 addr_len = err;
1039
1040 if (sun_path[0]) {
1041 umode_t mode = S_IFSOCK |
1042 (SOCK_INODE(sock)->i_mode & ~current_umask());
1043 err = unix_mknod(sun_path, mode, &path);
1044 if (err) {
1045 if (err == -EEXIST)
1046 err = -EADDRINUSE;
1047 goto out;
1048 }
1049 }
1050
1051 err = mutex_lock_interruptible(&u->bindlock);
1052 if (err)
1053 goto out_put;
1054
1055 err = -EINVAL;
1056 if (u->addr)
1057 goto out_up;
1058
1059 err = -ENOMEM;
1060 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1061 if (!addr)
1062 goto out_up;
1063
1064 memcpy(addr->name, sunaddr, addr_len);
1065 addr->len = addr_len;
1066 addr->hash = hash ^ sk->sk_type;
1067 refcount_set(&addr->refcnt, 1);
1068
1069 if (sun_path[0]) {
1070 addr->hash = UNIX_HASH_SIZE;
1071 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1072 spin_lock(&unix_table_lock);
1073 u->path = path;
1074 list = &unix_socket_table[hash];
1075 } else {
1076 spin_lock(&unix_table_lock);
1077 err = -EADDRINUSE;
1078 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1079 sk->sk_type, hash)) {
1080 unix_release_addr(addr);
1081 goto out_unlock;
1082 }
1083
1084 list = &unix_socket_table[addr->hash];
1085 }
1086
1087 err = 0;
1088 __unix_remove_socket(sk);
1089 smp_store_release(&u->addr, addr);
1090 __unix_insert_socket(list, sk);
1091
1092out_unlock:
1093 spin_unlock(&unix_table_lock);
1094out_up:
1095 mutex_unlock(&u->bindlock);
1096out_put:
1097 if (err)
1098 path_put(&path);
1099out:
1100 return err;
1101}
1102
1103static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1104{
1105 if (unlikely(sk1 == sk2) || !sk2) {
1106 unix_state_lock(sk1);
1107 return;
1108 }
1109 if (sk1 < sk2) {
1110 unix_state_lock(sk1);
1111 unix_state_lock_nested(sk2);
1112 } else {
1113 unix_state_lock(sk2);
1114 unix_state_lock_nested(sk1);
1115 }
1116}
1117
1118static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1119{
1120 if (unlikely(sk1 == sk2) || !sk2) {
1121 unix_state_unlock(sk1);
1122 return;
1123 }
1124 unix_state_unlock(sk1);
1125 unix_state_unlock(sk2);
1126}
1127
1128static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1129 int alen, int flags)
1130{
1131 struct sock *sk = sock->sk;
1132 struct net *net = sock_net(sk);
1133 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1134 struct sock *other;
1135 unsigned int hash;
1136 int err;
1137
1138 err = -EINVAL;
1139 if (alen < offsetofend(struct sockaddr, sa_family))
1140 goto out;
1141
1142 if (addr->sa_family != AF_UNSPEC) {
1143 err = unix_mkname(sunaddr, alen, &hash);
1144 if (err < 0)
1145 goto out;
1146 alen = err;
1147
1148 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1149 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1150 goto out;
1151
1152restart:
1153 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1154 if (!other)
1155 goto out;
1156
1157 unix_state_double_lock(sk, other);
1158
1159 /* Apparently VFS overslept socket death. Retry. */
1160 if (sock_flag(other, SOCK_DEAD)) {
1161 unix_state_double_unlock(sk, other);
1162 sock_put(other);
1163 goto restart;
1164 }
1165
1166 err = -EPERM;
1167 if (!unix_may_send(sk, other))
1168 goto out_unlock;
1169
1170 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1171 if (err)
1172 goto out_unlock;
1173
1174 } else {
1175 /*
1176 * 1003.1g breaking connected state with AF_UNSPEC
1177 */
1178 other = NULL;
1179 unix_state_double_lock(sk, other);
1180 }
1181
1182 /*
1183 * If it was connected, reconnect.
1184 */
1185 if (unix_peer(sk)) {
1186 struct sock *old_peer = unix_peer(sk);
1187 unix_peer(sk) = other;
1188 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1189
1190 unix_state_double_unlock(sk, other);
1191
1192 if (other != old_peer)
1193 unix_dgram_disconnected(sk, old_peer);
1194 sock_put(old_peer);
1195 } else {
1196 unix_peer(sk) = other;
1197 unix_state_double_unlock(sk, other);
1198 }
1199 return 0;
1200
1201out_unlock:
1202 unix_state_double_unlock(sk, other);
1203 sock_put(other);
1204out:
1205 return err;
1206}
1207
1208static long unix_wait_for_peer(struct sock *other, long timeo)
1209 __releases(&unix_sk(other)->lock)
1210{
1211 struct unix_sock *u = unix_sk(other);
1212 int sched;
1213 DEFINE_WAIT(wait);
1214
1215 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1216
1217 sched = !sock_flag(other, SOCK_DEAD) &&
1218 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1219 unix_recvq_full(other);
1220
1221 unix_state_unlock(other);
1222
1223 if (sched)
1224 timeo = schedule_timeout(timeo);
1225
1226 finish_wait(&u->peer_wait, &wait);
1227 return timeo;
1228}
1229
1230static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1231 int addr_len, int flags)
1232{
1233 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1234 struct sock *sk = sock->sk;
1235 struct net *net = sock_net(sk);
1236 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1237 struct sock *newsk = NULL;
1238 struct sock *other = NULL;
1239 struct sk_buff *skb = NULL;
1240 unsigned int hash;
1241 int st;
1242 int err;
1243 long timeo;
1244
1245 err = unix_mkname(sunaddr, addr_len, &hash);
1246 if (err < 0)
1247 goto out;
1248 addr_len = err;
1249
1250 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1251 (err = unix_autobind(sock)) != 0)
1252 goto out;
1253
1254 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1255
1256 /* First of all allocate resources.
1257 If we will make it after state is locked,
1258 we will have to recheck all again in any case.
1259 */
1260
1261 err = -ENOMEM;
1262
1263 /* create new sock for complete connection */
1264 newsk = unix_create1(sock_net(sk), NULL, 0);
1265 if (newsk == NULL)
1266 goto out;
1267
1268 /* Allocate skb for sending to listening sock */
1269 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1270 if (skb == NULL)
1271 goto out;
1272
1273restart:
1274 /* Find listening sock. */
1275 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1276 if (!other)
1277 goto out;
1278
1279 /* Latch state of peer */
1280 unix_state_lock(other);
1281
1282 /* Apparently VFS overslept socket death. Retry. */
1283 if (sock_flag(other, SOCK_DEAD)) {
1284 unix_state_unlock(other);
1285 sock_put(other);
1286 goto restart;
1287 }
1288
1289 err = -ECONNREFUSED;
1290 if (other->sk_state != TCP_LISTEN)
1291 goto out_unlock;
1292 if (other->sk_shutdown & RCV_SHUTDOWN)
1293 goto out_unlock;
1294
1295 if (unix_recvq_full(other)) {
1296 err = -EAGAIN;
1297 if (!timeo)
1298 goto out_unlock;
1299
1300 timeo = unix_wait_for_peer(other, timeo);
1301
1302 err = sock_intr_errno(timeo);
1303 if (signal_pending(current))
1304 goto out;
1305 sock_put(other);
1306 goto restart;
1307 }
1308
1309 /* Latch our state.
1310
1311 It is tricky place. We need to grab our state lock and cannot
1312 drop lock on peer. It is dangerous because deadlock is
1313 possible. Connect to self case and simultaneous
1314 attempt to connect are eliminated by checking socket
1315 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1316 check this before attempt to grab lock.
1317
1318 Well, and we have to recheck the state after socket locked.
1319 */
1320 st = sk->sk_state;
1321
1322 switch (st) {
1323 case TCP_CLOSE:
1324 /* This is ok... continue with connect */
1325 break;
1326 case TCP_ESTABLISHED:
1327 /* Socket is already connected */
1328 err = -EISCONN;
1329 goto out_unlock;
1330 default:
1331 err = -EINVAL;
1332 goto out_unlock;
1333 }
1334
1335 unix_state_lock_nested(sk);
1336
1337 if (sk->sk_state != st) {
1338 unix_state_unlock(sk);
1339 unix_state_unlock(other);
1340 sock_put(other);
1341 goto restart;
1342 }
1343
1344 err = security_unix_stream_connect(sk, other, newsk);
1345 if (err) {
1346 unix_state_unlock(sk);
1347 goto out_unlock;
1348 }
1349
1350 /* The way is open! Fastly set all the necessary fields... */
1351
1352 sock_hold(sk);
1353 unix_peer(newsk) = sk;
1354 newsk->sk_state = TCP_ESTABLISHED;
1355 newsk->sk_type = sk->sk_type;
1356 init_peercred(newsk);
1357 newu = unix_sk(newsk);
1358 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1359 otheru = unix_sk(other);
1360
1361 /* copy address information from listening to new sock
1362 *
1363 * The contents of *(otheru->addr) and otheru->path
1364 * are seen fully set up here, since we have found
1365 * otheru in hash under unix_table_lock. Insertion
1366 * into the hash chain we'd found it in had been done
1367 * in an earlier critical area protected by unix_table_lock,
1368 * the same one where we'd set *(otheru->addr) contents,
1369 * as well as otheru->path and otheru->addr itself.
1370 *
1371 * Using smp_store_release() here to set newu->addr
1372 * is enough to make those stores, as well as stores
1373 * to newu->path visible to anyone who gets newu->addr
1374 * by smp_load_acquire(). IOW, the same warranties
1375 * as for unix_sock instances bound in unix_bind() or
1376 * in unix_autobind().
1377 */
1378 if (otheru->path.dentry) {
1379 path_get(&otheru->path);
1380 newu->path = otheru->path;
1381 }
1382 refcount_inc(&otheru->addr->refcnt);
1383 smp_store_release(&newu->addr, otheru->addr);
1384
1385 /* Set credentials */
1386 copy_peercred(sk, other);
1387
1388 sock->state = SS_CONNECTED;
1389 sk->sk_state = TCP_ESTABLISHED;
1390 sock_hold(newsk);
1391
1392 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1393 unix_peer(sk) = newsk;
1394
1395 unix_state_unlock(sk);
1396
1397 /* take ten and and send info to listening sock */
1398 spin_lock(&other->sk_receive_queue.lock);
1399 __skb_queue_tail(&other->sk_receive_queue, skb);
1400 spin_unlock(&other->sk_receive_queue.lock);
1401 unix_state_unlock(other);
1402 other->sk_data_ready(other);
1403 sock_put(other);
1404 return 0;
1405
1406out_unlock:
1407 if (other)
1408 unix_state_unlock(other);
1409
1410out:
1411 kfree_skb(skb);
1412 if (newsk)
1413 unix_release_sock(newsk, 0);
1414 if (other)
1415 sock_put(other);
1416 return err;
1417}
1418
1419static int unix_socketpair(struct socket *socka, struct socket *sockb)
1420{
1421 struct sock *ska = socka->sk, *skb = sockb->sk;
1422
1423 /* Join our sockets back to back */
1424 sock_hold(ska);
1425 sock_hold(skb);
1426 unix_peer(ska) = skb;
1427 unix_peer(skb) = ska;
1428 init_peercred(ska);
1429 init_peercred(skb);
1430
1431 if (ska->sk_type != SOCK_DGRAM) {
1432 ska->sk_state = TCP_ESTABLISHED;
1433 skb->sk_state = TCP_ESTABLISHED;
1434 socka->state = SS_CONNECTED;
1435 sockb->state = SS_CONNECTED;
1436 }
1437 return 0;
1438}
1439
1440static void unix_sock_inherit_flags(const struct socket *old,
1441 struct socket *new)
1442{
1443 if (test_bit(SOCK_PASSCRED, &old->flags))
1444 set_bit(SOCK_PASSCRED, &new->flags);
1445 if (test_bit(SOCK_PASSSEC, &old->flags))
1446 set_bit(SOCK_PASSSEC, &new->flags);
1447}
1448
1449static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1450 bool kern)
1451{
1452 struct sock *sk = sock->sk;
1453 struct sock *tsk;
1454 struct sk_buff *skb;
1455 int err;
1456
1457 err = -EOPNOTSUPP;
1458 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1459 goto out;
1460
1461 err = -EINVAL;
1462 if (sk->sk_state != TCP_LISTEN)
1463 goto out;
1464
1465 /* If socket state is TCP_LISTEN it cannot change (for now...),
1466 * so that no locks are necessary.
1467 */
1468
1469 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1470 if (!skb) {
1471 /* This means receive shutdown. */
1472 if (err == 0)
1473 err = -EINVAL;
1474 goto out;
1475 }
1476
1477 tsk = skb->sk;
1478 skb_free_datagram(sk, skb);
1479 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1480
1481 /* attach accepted sock to socket */
1482 unix_state_lock(tsk);
1483 newsock->state = SS_CONNECTED;
1484 unix_sock_inherit_flags(sock, newsock);
1485 sock_graft(tsk, newsock);
1486 unix_state_unlock(tsk);
1487 return 0;
1488
1489out:
1490 return err;
1491}
1492
1493
1494static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1495{
1496 struct sock *sk = sock->sk;
1497 struct unix_address *addr;
1498 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1499 int err = 0;
1500
1501 if (peer) {
1502 sk = unix_peer_get(sk);
1503
1504 err = -ENOTCONN;
1505 if (!sk)
1506 goto out;
1507 err = 0;
1508 } else {
1509 sock_hold(sk);
1510 }
1511
1512 addr = smp_load_acquire(&unix_sk(sk)->addr);
1513 if (!addr) {
1514 sunaddr->sun_family = AF_UNIX;
1515 sunaddr->sun_path[0] = 0;
1516 err = sizeof(short);
1517 } else {
1518 err = addr->len;
1519 memcpy(sunaddr, addr->name, addr->len);
1520 }
1521 sock_put(sk);
1522out:
1523 return err;
1524}
1525
1526static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1527{
1528 int err = 0;
1529
1530 UNIXCB(skb).pid = get_pid(scm->pid);
1531 UNIXCB(skb).uid = scm->creds.uid;
1532 UNIXCB(skb).gid = scm->creds.gid;
1533 UNIXCB(skb).fp = NULL;
1534 unix_get_secdata(scm, skb);
1535 if (scm->fp && send_fds)
1536 err = unix_attach_fds(scm, skb);
1537
1538 skb->destructor = unix_destruct_scm;
1539 return err;
1540}
1541
1542static bool unix_passcred_enabled(const struct socket *sock,
1543 const struct sock *other)
1544{
1545 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1546 !other->sk_socket ||
1547 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1548}
1549
1550/*
1551 * Some apps rely on write() giving SCM_CREDENTIALS
1552 * We include credentials if source or destination socket
1553 * asserted SOCK_PASSCRED.
1554 */
1555static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1556 const struct sock *other)
1557{
1558 if (UNIXCB(skb).pid)
1559 return;
1560 if (unix_passcred_enabled(sock, other)) {
1561 UNIXCB(skb).pid = get_pid(task_tgid(current));
1562 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1563 }
1564}
1565
1566static int maybe_init_creds(struct scm_cookie *scm,
1567 struct socket *socket,
1568 const struct sock *other)
1569{
1570 int err;
1571 struct msghdr msg = { .msg_controllen = 0 };
1572
1573 err = scm_send(socket, &msg, scm, false);
1574 if (err)
1575 return err;
1576
1577 if (unix_passcred_enabled(socket, other)) {
1578 scm->pid = get_pid(task_tgid(current));
1579 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1580 }
1581 return err;
1582}
1583
1584static bool unix_skb_scm_eq(struct sk_buff *skb,
1585 struct scm_cookie *scm)
1586{
1587 const struct unix_skb_parms *u = &UNIXCB(skb);
1588
1589 return u->pid == scm->pid &&
1590 uid_eq(u->uid, scm->creds.uid) &&
1591 gid_eq(u->gid, scm->creds.gid) &&
1592 unix_secdata_eq(scm, skb);
1593}
1594
1595static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1596{
1597 struct scm_fp_list *fp = UNIXCB(skb).fp;
1598 struct unix_sock *u = unix_sk(sk);
1599
1600 if (unlikely(fp && fp->count))
1601 atomic_add(fp->count, &u->scm_stat.nr_fds);
1602}
1603
1604static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1605{
1606 struct scm_fp_list *fp = UNIXCB(skb).fp;
1607 struct unix_sock *u = unix_sk(sk);
1608
1609 if (unlikely(fp && fp->count))
1610 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1611}
1612
1613/*
1614 * Send AF_UNIX data.
1615 */
1616
1617static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1618 size_t len)
1619{
1620 struct sock *sk = sock->sk;
1621 struct net *net = sock_net(sk);
1622 struct unix_sock *u = unix_sk(sk);
1623 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1624 struct sock *other = NULL;
1625 int namelen = 0; /* fake GCC */
1626 int err;
1627 unsigned int hash;
1628 struct sk_buff *skb;
1629 long timeo;
1630 struct scm_cookie scm;
1631 int data_len = 0;
1632 int sk_locked;
1633
1634 wait_for_unix_gc();
1635 err = scm_send(sock, msg, &scm, false);
1636 if (err < 0)
1637 return err;
1638
1639 err = -EOPNOTSUPP;
1640 if (msg->msg_flags&MSG_OOB)
1641 goto out;
1642
1643 if (msg->msg_namelen) {
1644 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1645 if (err < 0)
1646 goto out;
1647 namelen = err;
1648 } else {
1649 sunaddr = NULL;
1650 err = -ENOTCONN;
1651 other = unix_peer_get(sk);
1652 if (!other)
1653 goto out;
1654 }
1655
1656 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1657 && (err = unix_autobind(sock)) != 0)
1658 goto out;
1659
1660 err = -EMSGSIZE;
1661 if (len > sk->sk_sndbuf - 32)
1662 goto out;
1663
1664 if (len > SKB_MAX_ALLOC) {
1665 data_len = min_t(size_t,
1666 len - SKB_MAX_ALLOC,
1667 MAX_SKB_FRAGS * PAGE_SIZE);
1668 data_len = PAGE_ALIGN(data_len);
1669
1670 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1671 }
1672
1673 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1674 msg->msg_flags & MSG_DONTWAIT, &err,
1675 PAGE_ALLOC_COSTLY_ORDER);
1676 if (skb == NULL)
1677 goto out;
1678
1679 err = unix_scm_to_skb(&scm, skb, true);
1680 if (err < 0)
1681 goto out_free;
1682
1683 skb_put(skb, len - data_len);
1684 skb->data_len = data_len;
1685 skb->len = len;
1686 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1687 if (err)
1688 goto out_free;
1689
1690 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1691
1692restart:
1693 if (!other) {
1694 err = -ECONNRESET;
1695 if (sunaddr == NULL)
1696 goto out_free;
1697
1698 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1699 hash, &err);
1700 if (other == NULL)
1701 goto out_free;
1702 }
1703
1704 if (sk_filter(other, skb) < 0) {
1705 /* Toss the packet but do not return any error to the sender */
1706 err = len;
1707 goto out_free;
1708 }
1709
1710 sk_locked = 0;
1711 unix_state_lock(other);
1712restart_locked:
1713 err = -EPERM;
1714 if (!unix_may_send(sk, other))
1715 goto out_unlock;
1716
1717 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1718 /*
1719 * Check with 1003.1g - what should
1720 * datagram error
1721 */
1722 unix_state_unlock(other);
1723 sock_put(other);
1724
1725 if (!sk_locked)
1726 unix_state_lock(sk);
1727
1728 err = 0;
1729 if (unix_peer(sk) == other) {
1730 unix_peer(sk) = NULL;
1731 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1732
1733 unix_state_unlock(sk);
1734
1735 unix_dgram_disconnected(sk, other);
1736 sock_put(other);
1737 err = -ECONNREFUSED;
1738 } else {
1739 unix_state_unlock(sk);
1740 }
1741
1742 other = NULL;
1743 if (err)
1744 goto out_free;
1745 goto restart;
1746 }
1747
1748 err = -EPIPE;
1749 if (other->sk_shutdown & RCV_SHUTDOWN)
1750 goto out_unlock;
1751
1752 if (sk->sk_type != SOCK_SEQPACKET) {
1753 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1754 if (err)
1755 goto out_unlock;
1756 }
1757
1758 /* other == sk && unix_peer(other) != sk if
1759 * - unix_peer(sk) == NULL, destination address bound to sk
1760 * - unix_peer(sk) == sk by time of get but disconnected before lock
1761 */
1762 if (other != sk &&
1763 unlikely(unix_peer(other) != sk &&
1764 unix_recvq_full_lockless(other))) {
1765 if (timeo) {
1766 timeo = unix_wait_for_peer(other, timeo);
1767
1768 err = sock_intr_errno(timeo);
1769 if (signal_pending(current))
1770 goto out_free;
1771
1772 goto restart;
1773 }
1774
1775 if (!sk_locked) {
1776 unix_state_unlock(other);
1777 unix_state_double_lock(sk, other);
1778 }
1779
1780 if (unix_peer(sk) != other ||
1781 unix_dgram_peer_wake_me(sk, other)) {
1782 err = -EAGAIN;
1783 sk_locked = 1;
1784 goto out_unlock;
1785 }
1786
1787 if (!sk_locked) {
1788 sk_locked = 1;
1789 goto restart_locked;
1790 }
1791 }
1792
1793 if (unlikely(sk_locked))
1794 unix_state_unlock(sk);
1795
1796 if (sock_flag(other, SOCK_RCVTSTAMP))
1797 __net_timestamp(skb);
1798 maybe_add_creds(skb, sock, other);
1799 scm_stat_add(other, skb);
1800 skb_queue_tail(&other->sk_receive_queue, skb);
1801 unix_state_unlock(other);
1802 other->sk_data_ready(other);
1803 sock_put(other);
1804 scm_destroy(&scm);
1805 return len;
1806
1807out_unlock:
1808 if (sk_locked)
1809 unix_state_unlock(sk);
1810 unix_state_unlock(other);
1811out_free:
1812 kfree_skb(skb);
1813out:
1814 if (other)
1815 sock_put(other);
1816 scm_destroy(&scm);
1817 return err;
1818}
1819
1820/* We use paged skbs for stream sockets, and limit occupancy to 32768
1821 * bytes, and a minimum of a full page.
1822 */
1823#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1824
1825static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1826 size_t len)
1827{
1828 struct sock *sk = sock->sk;
1829 struct sock *other = NULL;
1830 int err, size;
1831 struct sk_buff *skb;
1832 int sent = 0;
1833 struct scm_cookie scm;
1834 bool fds_sent = false;
1835 int data_len;
1836
1837 wait_for_unix_gc();
1838 err = scm_send(sock, msg, &scm, false);
1839 if (err < 0)
1840 return err;
1841
1842 err = -EOPNOTSUPP;
1843 if (msg->msg_flags&MSG_OOB)
1844 goto out_err;
1845
1846 if (msg->msg_namelen) {
1847 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1848 goto out_err;
1849 } else {
1850 err = -ENOTCONN;
1851 other = unix_peer(sk);
1852 if (!other)
1853 goto out_err;
1854 }
1855
1856 if (sk->sk_shutdown & SEND_SHUTDOWN)
1857 goto pipe_err;
1858
1859 while (sent < len) {
1860 size = len - sent;
1861
1862 /* Keep two messages in the pipe so it schedules better */
1863 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1864
1865 /* allow fallback to order-0 allocations */
1866 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1867
1868 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1869
1870 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1871
1872 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1873 msg->msg_flags & MSG_DONTWAIT, &err,
1874 get_order(UNIX_SKB_FRAGS_SZ));
1875 if (!skb)
1876 goto out_err;
1877
1878 /* Only send the fds in the first buffer */
1879 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1880 if (err < 0) {
1881 kfree_skb(skb);
1882 goto out_err;
1883 }
1884 fds_sent = true;
1885
1886 skb_put(skb, size - data_len);
1887 skb->data_len = data_len;
1888 skb->len = size;
1889 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1890 if (err) {
1891 kfree_skb(skb);
1892 goto out_err;
1893 }
1894
1895 unix_state_lock(other);
1896
1897 if (sock_flag(other, SOCK_DEAD) ||
1898 (other->sk_shutdown & RCV_SHUTDOWN))
1899 goto pipe_err_free;
1900
1901 maybe_add_creds(skb, sock, other);
1902 scm_stat_add(other, skb);
1903 skb_queue_tail(&other->sk_receive_queue, skb);
1904 unix_state_unlock(other);
1905 other->sk_data_ready(other);
1906 sent += size;
1907 }
1908
1909 scm_destroy(&scm);
1910
1911 return sent;
1912
1913pipe_err_free:
1914 unix_state_unlock(other);
1915 kfree_skb(skb);
1916pipe_err:
1917 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1918 send_sig(SIGPIPE, current, 0);
1919 err = -EPIPE;
1920out_err:
1921 scm_destroy(&scm);
1922 return sent ? : err;
1923}
1924
1925static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1926 int offset, size_t size, int flags)
1927{
1928 int err;
1929 bool send_sigpipe = false;
1930 bool init_scm = true;
1931 struct scm_cookie scm;
1932 struct sock *other, *sk = socket->sk;
1933 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1934
1935 if (flags & MSG_OOB)
1936 return -EOPNOTSUPP;
1937
1938 other = unix_peer(sk);
1939 if (!other || sk->sk_state != TCP_ESTABLISHED)
1940 return -ENOTCONN;
1941
1942 if (false) {
1943alloc_skb:
1944 unix_state_unlock(other);
1945 mutex_unlock(&unix_sk(other)->iolock);
1946 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1947 &err, 0);
1948 if (!newskb)
1949 goto err;
1950 }
1951
1952 /* we must acquire iolock as we modify already present
1953 * skbs in the sk_receive_queue and mess with skb->len
1954 */
1955 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1956 if (err) {
1957 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1958 goto err;
1959 }
1960
1961 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1962 err = -EPIPE;
1963 send_sigpipe = true;
1964 goto err_unlock;
1965 }
1966
1967 unix_state_lock(other);
1968
1969 if (sock_flag(other, SOCK_DEAD) ||
1970 other->sk_shutdown & RCV_SHUTDOWN) {
1971 err = -EPIPE;
1972 send_sigpipe = true;
1973 goto err_state_unlock;
1974 }
1975
1976 if (init_scm) {
1977 err = maybe_init_creds(&scm, socket, other);
1978 if (err)
1979 goto err_state_unlock;
1980 init_scm = false;
1981 }
1982
1983 skb = skb_peek_tail(&other->sk_receive_queue);
1984 if (tail && tail == skb) {
1985 skb = newskb;
1986 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
1987 if (newskb) {
1988 skb = newskb;
1989 } else {
1990 tail = skb;
1991 goto alloc_skb;
1992 }
1993 } else if (newskb) {
1994 /* this is fast path, we don't necessarily need to
1995 * call to kfree_skb even though with newskb == NULL
1996 * this - does no harm
1997 */
1998 consume_skb(newskb);
1999 newskb = NULL;
2000 }
2001
2002 if (skb_append_pagefrags(skb, page, offset, size)) {
2003 tail = skb;
2004 goto alloc_skb;
2005 }
2006
2007 skb->len += size;
2008 skb->data_len += size;
2009 skb->truesize += size;
2010 refcount_add(size, &sk->sk_wmem_alloc);
2011
2012 if (newskb) {
2013 err = unix_scm_to_skb(&scm, skb, false);
2014 if (err)
2015 goto err_state_unlock;
2016 spin_lock(&other->sk_receive_queue.lock);
2017 __skb_queue_tail(&other->sk_receive_queue, newskb);
2018 spin_unlock(&other->sk_receive_queue.lock);
2019 }
2020
2021 unix_state_unlock(other);
2022 mutex_unlock(&unix_sk(other)->iolock);
2023
2024 other->sk_data_ready(other);
2025 scm_destroy(&scm);
2026 return size;
2027
2028err_state_unlock:
2029 unix_state_unlock(other);
2030err_unlock:
2031 mutex_unlock(&unix_sk(other)->iolock);
2032err:
2033 kfree_skb(newskb);
2034 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2035 send_sig(SIGPIPE, current, 0);
2036 if (!init_scm)
2037 scm_destroy(&scm);
2038 return err;
2039}
2040
2041static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2042 size_t len)
2043{
2044 int err;
2045 struct sock *sk = sock->sk;
2046
2047 err = sock_error(sk);
2048 if (err)
2049 return err;
2050
2051 if (sk->sk_state != TCP_ESTABLISHED)
2052 return -ENOTCONN;
2053
2054 if (msg->msg_namelen)
2055 msg->msg_namelen = 0;
2056
2057 return unix_dgram_sendmsg(sock, msg, len);
2058}
2059
2060static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2061 size_t size, int flags)
2062{
2063 struct sock *sk = sock->sk;
2064
2065 if (sk->sk_state != TCP_ESTABLISHED)
2066 return -ENOTCONN;
2067
2068 return unix_dgram_recvmsg(sock, msg, size, flags);
2069}
2070
2071static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2072{
2073 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2074
2075 if (addr) {
2076 msg->msg_namelen = addr->len;
2077 memcpy(msg->msg_name, addr->name, addr->len);
2078 }
2079}
2080
2081static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2082 size_t size, int flags)
2083{
2084 struct scm_cookie scm;
2085 struct sock *sk = sock->sk;
2086 struct unix_sock *u = unix_sk(sk);
2087 struct sk_buff *skb, *last;
2088 long timeo;
2089 int skip;
2090 int err;
2091
2092 err = -EOPNOTSUPP;
2093 if (flags&MSG_OOB)
2094 goto out;
2095
2096 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2097
2098 do {
2099 mutex_lock(&u->iolock);
2100
2101 skip = sk_peek_offset(sk, flags);
2102 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2103 &skip, &err, &last);
2104 if (skb) {
2105 if (!(flags & MSG_PEEK))
2106 scm_stat_del(sk, skb);
2107 break;
2108 }
2109
2110 mutex_unlock(&u->iolock);
2111
2112 if (err != -EAGAIN)
2113 break;
2114 } while (timeo &&
2115 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2116 &err, &timeo, last));
2117
2118 if (!skb) { /* implies iolock unlocked */
2119 unix_state_lock(sk);
2120 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2121 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2122 (sk->sk_shutdown & RCV_SHUTDOWN))
2123 err = 0;
2124 unix_state_unlock(sk);
2125 goto out;
2126 }
2127
2128 if (wq_has_sleeper(&u->peer_wait))
2129 wake_up_interruptible_sync_poll(&u->peer_wait,
2130 EPOLLOUT | EPOLLWRNORM |
2131 EPOLLWRBAND);
2132
2133 if (msg->msg_name)
2134 unix_copy_addr(msg, skb->sk);
2135
2136 if (size > skb->len - skip)
2137 size = skb->len - skip;
2138 else if (size < skb->len - skip)
2139 msg->msg_flags |= MSG_TRUNC;
2140
2141 err = skb_copy_datagram_msg(skb, skip, msg, size);
2142 if (err)
2143 goto out_free;
2144
2145 if (sock_flag(sk, SOCK_RCVTSTAMP))
2146 __sock_recv_timestamp(msg, sk, skb);
2147
2148 memset(&scm, 0, sizeof(scm));
2149
2150 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2151 unix_set_secdata(&scm, skb);
2152
2153 if (!(flags & MSG_PEEK)) {
2154 if (UNIXCB(skb).fp)
2155 unix_detach_fds(&scm, skb);
2156
2157 sk_peek_offset_bwd(sk, skb->len);
2158 } else {
2159 /* It is questionable: on PEEK we could:
2160 - do not return fds - good, but too simple 8)
2161 - return fds, and do not return them on read (old strategy,
2162 apparently wrong)
2163 - clone fds (I chose it for now, it is the most universal
2164 solution)
2165
2166 POSIX 1003.1g does not actually define this clearly
2167 at all. POSIX 1003.1g doesn't define a lot of things
2168 clearly however!
2169
2170 */
2171
2172 sk_peek_offset_fwd(sk, size);
2173
2174 if (UNIXCB(skb).fp)
2175 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2176 }
2177 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2178
2179 scm_recv(sock, msg, &scm, flags);
2180
2181out_free:
2182 skb_free_datagram(sk, skb);
2183 mutex_unlock(&u->iolock);
2184out:
2185 return err;
2186}
2187
2188/*
2189 * Sleep until more data has arrived. But check for races..
2190 */
2191static long unix_stream_data_wait(struct sock *sk, long timeo,
2192 struct sk_buff *last, unsigned int last_len,
2193 bool freezable)
2194{
2195 struct sk_buff *tail;
2196 DEFINE_WAIT(wait);
2197
2198 unix_state_lock(sk);
2199
2200 for (;;) {
2201 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2202
2203 tail = skb_peek_tail(&sk->sk_receive_queue);
2204 if (tail != last ||
2205 (tail && tail->len != last_len) ||
2206 sk->sk_err ||
2207 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2208 signal_pending(current) ||
2209 !timeo)
2210 break;
2211
2212 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2213 unix_state_unlock(sk);
2214 if (freezable)
2215 timeo = freezable_schedule_timeout(timeo);
2216 else
2217 timeo = schedule_timeout(timeo);
2218 unix_state_lock(sk);
2219
2220 if (sock_flag(sk, SOCK_DEAD))
2221 break;
2222
2223 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2224 }
2225
2226 finish_wait(sk_sleep(sk), &wait);
2227 unix_state_unlock(sk);
2228 return timeo;
2229}
2230
2231static unsigned int unix_skb_len(const struct sk_buff *skb)
2232{
2233 return skb->len - UNIXCB(skb).consumed;
2234}
2235
2236struct unix_stream_read_state {
2237 int (*recv_actor)(struct sk_buff *, int, int,
2238 struct unix_stream_read_state *);
2239 struct socket *socket;
2240 struct msghdr *msg;
2241 struct pipe_inode_info *pipe;
2242 size_t size;
2243 int flags;
2244 unsigned int splice_flags;
2245};
2246
2247static int unix_stream_read_generic(struct unix_stream_read_state *state,
2248 bool freezable)
2249{
2250 struct scm_cookie scm;
2251 struct socket *sock = state->socket;
2252 struct sock *sk = sock->sk;
2253 struct unix_sock *u = unix_sk(sk);
2254 int copied = 0;
2255 int flags = state->flags;
2256 int noblock = flags & MSG_DONTWAIT;
2257 bool check_creds = false;
2258 int target;
2259 int err = 0;
2260 long timeo;
2261 int skip;
2262 size_t size = state->size;
2263 unsigned int last_len;
2264
2265 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2266 err = -EINVAL;
2267 goto out;
2268 }
2269
2270 if (unlikely(flags & MSG_OOB)) {
2271 err = -EOPNOTSUPP;
2272 goto out;
2273 }
2274
2275 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2276 timeo = sock_rcvtimeo(sk, noblock);
2277
2278 memset(&scm, 0, sizeof(scm));
2279
2280 /* Lock the socket to prevent queue disordering
2281 * while sleeps in memcpy_tomsg
2282 */
2283 mutex_lock(&u->iolock);
2284
2285 skip = max(sk_peek_offset(sk, flags), 0);
2286
2287 do {
2288 int chunk;
2289 bool drop_skb;
2290 struct sk_buff *skb, *last;
2291
2292redo:
2293 unix_state_lock(sk);
2294 if (sock_flag(sk, SOCK_DEAD)) {
2295 err = -ECONNRESET;
2296 goto unlock;
2297 }
2298 last = skb = skb_peek(&sk->sk_receive_queue);
2299 last_len = last ? last->len : 0;
2300again:
2301 if (skb == NULL) {
2302 if (copied >= target)
2303 goto unlock;
2304
2305 /*
2306 * POSIX 1003.1g mandates this order.
2307 */
2308
2309 err = sock_error(sk);
2310 if (err)
2311 goto unlock;
2312 if (sk->sk_shutdown & RCV_SHUTDOWN)
2313 goto unlock;
2314
2315 unix_state_unlock(sk);
2316 if (!timeo) {
2317 err = -EAGAIN;
2318 break;
2319 }
2320
2321 mutex_unlock(&u->iolock);
2322
2323 timeo = unix_stream_data_wait(sk, timeo, last,
2324 last_len, freezable);
2325
2326 if (signal_pending(current)) {
2327 err = sock_intr_errno(timeo);
2328 scm_destroy(&scm);
2329 goto out;
2330 }
2331
2332 mutex_lock(&u->iolock);
2333 goto redo;
2334unlock:
2335 unix_state_unlock(sk);
2336 break;
2337 }
2338
2339 while (skip >= unix_skb_len(skb)) {
2340 skip -= unix_skb_len(skb);
2341 last = skb;
2342 last_len = skb->len;
2343 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2344 if (!skb)
2345 goto again;
2346 }
2347
2348 unix_state_unlock(sk);
2349
2350 if (check_creds) {
2351 /* Never glue messages from different writers */
2352 if (!unix_skb_scm_eq(skb, &scm))
2353 break;
2354 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2355 /* Copy credentials */
2356 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2357 unix_set_secdata(&scm, skb);
2358 check_creds = true;
2359 }
2360
2361 /* Copy address just once */
2362 if (state->msg && state->msg->msg_name) {
2363 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2364 state->msg->msg_name);
2365 unix_copy_addr(state->msg, skb->sk);
2366 sunaddr = NULL;
2367 }
2368
2369 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2370 skb_get(skb);
2371 chunk = state->recv_actor(skb, skip, chunk, state);
2372 drop_skb = !unix_skb_len(skb);
2373 /* skb is only safe to use if !drop_skb */
2374 consume_skb(skb);
2375 if (chunk < 0) {
2376 if (copied == 0)
2377 copied = -EFAULT;
2378 break;
2379 }
2380 copied += chunk;
2381 size -= chunk;
2382
2383 if (drop_skb) {
2384 /* the skb was touched by a concurrent reader;
2385 * we should not expect anything from this skb
2386 * anymore and assume it invalid - we can be
2387 * sure it was dropped from the socket queue
2388 *
2389 * let's report a short read
2390 */
2391 err = 0;
2392 break;
2393 }
2394
2395 /* Mark read part of skb as used */
2396 if (!(flags & MSG_PEEK)) {
2397 UNIXCB(skb).consumed += chunk;
2398
2399 sk_peek_offset_bwd(sk, chunk);
2400
2401 if (UNIXCB(skb).fp) {
2402 scm_stat_del(sk, skb);
2403 unix_detach_fds(&scm, skb);
2404 }
2405
2406 if (unix_skb_len(skb))
2407 break;
2408
2409 skb_unlink(skb, &sk->sk_receive_queue);
2410 consume_skb(skb);
2411
2412 if (scm.fp)
2413 break;
2414 } else {
2415 /* It is questionable, see note in unix_dgram_recvmsg.
2416 */
2417 if (UNIXCB(skb).fp)
2418 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2419
2420 sk_peek_offset_fwd(sk, chunk);
2421
2422 if (UNIXCB(skb).fp)
2423 break;
2424
2425 skip = 0;
2426 last = skb;
2427 last_len = skb->len;
2428 unix_state_lock(sk);
2429 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2430 if (skb)
2431 goto again;
2432 unix_state_unlock(sk);
2433 break;
2434 }
2435 } while (size);
2436
2437 mutex_unlock(&u->iolock);
2438 if (state->msg)
2439 scm_recv(sock, state->msg, &scm, flags);
2440 else
2441 scm_destroy(&scm);
2442out:
2443 return copied ? : err;
2444}
2445
2446static int unix_stream_read_actor(struct sk_buff *skb,
2447 int skip, int chunk,
2448 struct unix_stream_read_state *state)
2449{
2450 int ret;
2451
2452 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2453 state->msg, chunk);
2454 return ret ?: chunk;
2455}
2456
2457static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2458 size_t size, int flags)
2459{
2460 struct unix_stream_read_state state = {
2461 .recv_actor = unix_stream_read_actor,
2462 .socket = sock,
2463 .msg = msg,
2464 .size = size,
2465 .flags = flags
2466 };
2467
2468 return unix_stream_read_generic(&state, true);
2469}
2470
2471static int unix_stream_splice_actor(struct sk_buff *skb,
2472 int skip, int chunk,
2473 struct unix_stream_read_state *state)
2474{
2475 return skb_splice_bits(skb, state->socket->sk,
2476 UNIXCB(skb).consumed + skip,
2477 state->pipe, chunk, state->splice_flags);
2478}
2479
2480static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2481 struct pipe_inode_info *pipe,
2482 size_t size, unsigned int flags)
2483{
2484 struct unix_stream_read_state state = {
2485 .recv_actor = unix_stream_splice_actor,
2486 .socket = sock,
2487 .pipe = pipe,
2488 .size = size,
2489 .splice_flags = flags,
2490 };
2491
2492 if (unlikely(*ppos))
2493 return -ESPIPE;
2494
2495 if (sock->file->f_flags & O_NONBLOCK ||
2496 flags & SPLICE_F_NONBLOCK)
2497 state.flags = MSG_DONTWAIT;
2498
2499 return unix_stream_read_generic(&state, false);
2500}
2501
2502static int unix_shutdown(struct socket *sock, int mode)
2503{
2504 struct sock *sk = sock->sk;
2505 struct sock *other;
2506
2507 if (mode < SHUT_RD || mode > SHUT_RDWR)
2508 return -EINVAL;
2509 /* This maps:
2510 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2511 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2512 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2513 */
2514 ++mode;
2515
2516 unix_state_lock(sk);
2517 sk->sk_shutdown |= mode;
2518 other = unix_peer(sk);
2519 if (other)
2520 sock_hold(other);
2521 unix_state_unlock(sk);
2522 sk->sk_state_change(sk);
2523
2524 if (other &&
2525 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2526
2527 int peer_mode = 0;
2528
2529 if (mode&RCV_SHUTDOWN)
2530 peer_mode |= SEND_SHUTDOWN;
2531 if (mode&SEND_SHUTDOWN)
2532 peer_mode |= RCV_SHUTDOWN;
2533 unix_state_lock(other);
2534 other->sk_shutdown |= peer_mode;
2535 unix_state_unlock(other);
2536 other->sk_state_change(other);
2537 if (peer_mode == SHUTDOWN_MASK)
2538 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2539 else if (peer_mode & RCV_SHUTDOWN)
2540 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2541 }
2542 if (other)
2543 sock_put(other);
2544
2545 return 0;
2546}
2547
2548long unix_inq_len(struct sock *sk)
2549{
2550 struct sk_buff *skb;
2551 long amount = 0;
2552
2553 if (sk->sk_state == TCP_LISTEN)
2554 return -EINVAL;
2555
2556 spin_lock(&sk->sk_receive_queue.lock);
2557 if (sk->sk_type == SOCK_STREAM ||
2558 sk->sk_type == SOCK_SEQPACKET) {
2559 skb_queue_walk(&sk->sk_receive_queue, skb)
2560 amount += unix_skb_len(skb);
2561 } else {
2562 skb = skb_peek(&sk->sk_receive_queue);
2563 if (skb)
2564 amount = skb->len;
2565 }
2566 spin_unlock(&sk->sk_receive_queue.lock);
2567
2568 return amount;
2569}
2570EXPORT_SYMBOL_GPL(unix_inq_len);
2571
2572long unix_outq_len(struct sock *sk)
2573{
2574 return sk_wmem_alloc_get(sk);
2575}
2576EXPORT_SYMBOL_GPL(unix_outq_len);
2577
2578static int unix_open_file(struct sock *sk)
2579{
2580 struct path path;
2581 struct file *f;
2582 int fd;
2583
2584 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2585 return -EPERM;
2586
2587 if (!smp_load_acquire(&unix_sk(sk)->addr))
2588 return -ENOENT;
2589
2590 path = unix_sk(sk)->path;
2591 if (!path.dentry)
2592 return -ENOENT;
2593
2594 path_get(&path);
2595
2596 fd = get_unused_fd_flags(O_CLOEXEC);
2597 if (fd < 0)
2598 goto out;
2599
2600 f = dentry_open(&path, O_PATH, current_cred());
2601 if (IS_ERR(f)) {
2602 put_unused_fd(fd);
2603 fd = PTR_ERR(f);
2604 goto out;
2605 }
2606
2607 fd_install(fd, f);
2608out:
2609 path_put(&path);
2610
2611 return fd;
2612}
2613
2614static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2615{
2616 struct sock *sk = sock->sk;
2617 long amount = 0;
2618 int err;
2619
2620 switch (cmd) {
2621 case SIOCOUTQ:
2622 amount = unix_outq_len(sk);
2623 err = put_user(amount, (int __user *)arg);
2624 break;
2625 case SIOCINQ:
2626 amount = unix_inq_len(sk);
2627 if (amount < 0)
2628 err = amount;
2629 else
2630 err = put_user(amount, (int __user *)arg);
2631 break;
2632 case SIOCUNIXFILE:
2633 err = unix_open_file(sk);
2634 break;
2635 default:
2636 err = -ENOIOCTLCMD;
2637 break;
2638 }
2639 return err;
2640}
2641
2642#ifdef CONFIG_COMPAT
2643static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2644{
2645 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2646}
2647#endif
2648
2649static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2650{
2651 struct sock *sk = sock->sk;
2652 __poll_t mask;
2653
2654 sock_poll_wait(file, sock, wait);
2655 mask = 0;
2656
2657 /* exceptional events? */
2658 if (sk->sk_err)
2659 mask |= EPOLLERR;
2660 if (sk->sk_shutdown == SHUTDOWN_MASK)
2661 mask |= EPOLLHUP;
2662 if (sk->sk_shutdown & RCV_SHUTDOWN)
2663 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2664
2665 /* readable? */
2666 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2667 mask |= EPOLLIN | EPOLLRDNORM;
2668
2669 /* Connection-based need to check for termination and startup */
2670 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2671 sk->sk_state == TCP_CLOSE)
2672 mask |= EPOLLHUP;
2673
2674 /*
2675 * we set writable also when the other side has shut down the
2676 * connection. This prevents stuck sockets.
2677 */
2678 if (unix_writable(sk))
2679 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2680
2681 return mask;
2682}
2683
2684static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2685 poll_table *wait)
2686{
2687 struct sock *sk = sock->sk, *other;
2688 unsigned int writable;
2689 __poll_t mask;
2690
2691 sock_poll_wait(file, sock, wait);
2692 mask = 0;
2693
2694 /* exceptional events? */
2695 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2696 mask |= EPOLLERR |
2697 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2698
2699 if (sk->sk_shutdown & RCV_SHUTDOWN)
2700 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2701 if (sk->sk_shutdown == SHUTDOWN_MASK)
2702 mask |= EPOLLHUP;
2703
2704 /* readable? */
2705 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2706 mask |= EPOLLIN | EPOLLRDNORM;
2707
2708 /* Connection-based need to check for termination and startup */
2709 if (sk->sk_type == SOCK_SEQPACKET) {
2710 if (sk->sk_state == TCP_CLOSE)
2711 mask |= EPOLLHUP;
2712 /* connection hasn't started yet? */
2713 if (sk->sk_state == TCP_SYN_SENT)
2714 return mask;
2715 }
2716
2717 /* No write status requested, avoid expensive OUT tests. */
2718 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2719 return mask;
2720
2721 writable = unix_writable(sk);
2722 if (writable) {
2723 unix_state_lock(sk);
2724
2725 other = unix_peer(sk);
2726 if (other && unix_peer(other) != sk &&
2727 unix_recvq_full(other) &&
2728 unix_dgram_peer_wake_me(sk, other))
2729 writable = 0;
2730
2731 unix_state_unlock(sk);
2732 }
2733
2734 if (writable)
2735 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2736 else
2737 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2738
2739 return mask;
2740}
2741
2742#ifdef CONFIG_PROC_FS
2743
2744#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2745
2746#define get_bucket(x) ((x) >> BUCKET_SPACE)
2747#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2748#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2749
2750static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2751{
2752 unsigned long offset = get_offset(*pos);
2753 unsigned long bucket = get_bucket(*pos);
2754 struct sock *sk;
2755 unsigned long count = 0;
2756
2757 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2758 if (sock_net(sk) != seq_file_net(seq))
2759 continue;
2760 if (++count == offset)
2761 break;
2762 }
2763
2764 return sk;
2765}
2766
2767static struct sock *unix_next_socket(struct seq_file *seq,
2768 struct sock *sk,
2769 loff_t *pos)
2770{
2771 unsigned long bucket;
2772
2773 while (sk > (struct sock *)SEQ_START_TOKEN) {
2774 sk = sk_next(sk);
2775 if (!sk)
2776 goto next_bucket;
2777 if (sock_net(sk) == seq_file_net(seq))
2778 return sk;
2779 }
2780
2781 do {
2782 sk = unix_from_bucket(seq, pos);
2783 if (sk)
2784 return sk;
2785
2786next_bucket:
2787 bucket = get_bucket(*pos) + 1;
2788 *pos = set_bucket_offset(bucket, 1);
2789 } while (bucket < ARRAY_SIZE(unix_socket_table));
2790
2791 return NULL;
2792}
2793
2794static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2795 __acquires(unix_table_lock)
2796{
2797 spin_lock(&unix_table_lock);
2798
2799 if (!*pos)
2800 return SEQ_START_TOKEN;
2801
2802 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2803 return NULL;
2804
2805 return unix_next_socket(seq, NULL, pos);
2806}
2807
2808static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2809{
2810 ++*pos;
2811 return unix_next_socket(seq, v, pos);
2812}
2813
2814static void unix_seq_stop(struct seq_file *seq, void *v)
2815 __releases(unix_table_lock)
2816{
2817 spin_unlock(&unix_table_lock);
2818}
2819
2820static int unix_seq_show(struct seq_file *seq, void *v)
2821{
2822
2823 if (v == SEQ_START_TOKEN)
2824 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2825 "Inode Path\n");
2826 else {
2827 struct sock *s = v;
2828 struct unix_sock *u = unix_sk(s);
2829 unix_state_lock(s);
2830
2831 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2832 s,
2833 refcount_read(&s->sk_refcnt),
2834 0,
2835 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2836 s->sk_type,
2837 s->sk_socket ?
2838 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2839 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2840 sock_i_ino(s));
2841
2842 if (u->addr) { // under unix_table_lock here
2843 int i, len;
2844 seq_putc(seq, ' ');
2845
2846 i = 0;
2847 len = u->addr->len - sizeof(short);
2848 if (!UNIX_ABSTRACT(s))
2849 len--;
2850 else {
2851 seq_putc(seq, '@');
2852 i++;
2853 }
2854 for ( ; i < len; i++)
2855 seq_putc(seq, u->addr->name->sun_path[i] ?:
2856 '@');
2857 }
2858 unix_state_unlock(s);
2859 seq_putc(seq, '\n');
2860 }
2861
2862 return 0;
2863}
2864
2865static const struct seq_operations unix_seq_ops = {
2866 .start = unix_seq_start,
2867 .next = unix_seq_next,
2868 .stop = unix_seq_stop,
2869 .show = unix_seq_show,
2870};
2871#endif
2872
2873static const struct net_proto_family unix_family_ops = {
2874 .family = PF_UNIX,
2875 .create = unix_create,
2876 .owner = THIS_MODULE,
2877};
2878
2879
2880static int __net_init unix_net_init(struct net *net)
2881{
2882 int error = -ENOMEM;
2883
2884 net->unx.sysctl_max_dgram_qlen = 10;
2885 if (unix_sysctl_register(net))
2886 goto out;
2887
2888#ifdef CONFIG_PROC_FS
2889 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2890 sizeof(struct seq_net_private))) {
2891 unix_sysctl_unregister(net);
2892 goto out;
2893 }
2894#endif
2895 error = 0;
2896out:
2897 return error;
2898}
2899
2900static void __net_exit unix_net_exit(struct net *net)
2901{
2902 unix_sysctl_unregister(net);
2903 remove_proc_entry("unix", net->proc_net);
2904}
2905
2906static struct pernet_operations unix_net_ops = {
2907 .init = unix_net_init,
2908 .exit = unix_net_exit,
2909};
2910
2911static int __init af_unix_init(void)
2912{
2913 int rc = -1;
2914
2915 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
2916
2917 rc = proto_register(&unix_proto, 1);
2918 if (rc != 0) {
2919 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2920 goto out;
2921 }
2922
2923 sock_register(&unix_family_ops);
2924 register_pernet_subsys(&unix_net_ops);
2925out:
2926 return rc;
2927}
2928
2929static void __exit af_unix_exit(void)
2930{
2931 sock_unregister(PF_UNIX);
2932 proto_unregister(&unix_proto);
2933 unregister_pernet_subsys(&unix_net_ops);
2934}
2935
2936/* Earlier than device_initcall() so that other drivers invoking
2937 request_module() don't end up in a loop when modprobe tries
2938 to use a UNIX socket. But later than subsys_initcall() because
2939 we depend on stuff initialised there */
2940fs_initcall(af_unix_init);
2941module_exit(af_unix_exit);
2942
2943MODULE_LICENSE("GPL");
2944MODULE_ALIAS_NETPROTO(PF_UNIX);