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#ifdef CONFIG_PROVE_LOCKING
130#define cmp_ptr(l, r) (((l) > (r)) - ((l) < (r)))
131
132static int unix_table_lock_cmp_fn(const struct lockdep_map *a,
133 const struct lockdep_map *b)
134{
135 return cmp_ptr(a, b);
136}
137
138static int unix_state_lock_cmp_fn(const struct lockdep_map *_a,
139 const struct lockdep_map *_b)
140{
141 const struct unix_sock *a, *b;
142
143 a = container_of(_a, struct unix_sock, lock.dep_map);
144 b = container_of(_b, struct unix_sock, lock.dep_map);
145
146 if (a->sk.sk_state == TCP_LISTEN) {
147 /* unix_stream_connect(): Before the 2nd unix_state_lock(),
148 *
149 * 1. a is TCP_LISTEN.
150 * 2. b is not a.
151 * 3. concurrent connect(b -> a) must fail.
152 *
153 * Except for 2. & 3., the b's state can be any possible
154 * value due to concurrent connect() or listen().
155 *
156 * 2. is detected in debug_spin_lock_before(), and 3. cannot
157 * be expressed as lock_cmp_fn.
158 */
159 switch (b->sk.sk_state) {
160 case TCP_CLOSE:
161 case TCP_ESTABLISHED:
162 case TCP_LISTEN:
163 return -1;
164 default:
165 /* Invalid case. */
166 return 0;
167 }
168 }
169
170 /* Should never happen. Just to be symmetric. */
171 if (b->sk.sk_state == TCP_LISTEN) {
172 switch (b->sk.sk_state) {
173 case TCP_CLOSE:
174 case TCP_ESTABLISHED:
175 return 1;
176 default:
177 return 0;
178 }
179 }
180
181 /* unix_state_double_lock(): ascending address order. */
182 return cmp_ptr(a, b);
183}
184
185static int unix_recvq_lock_cmp_fn(const struct lockdep_map *_a,
186 const struct lockdep_map *_b)
187{
188 const struct sock *a, *b;
189
190 a = container_of(_a, struct sock, sk_receive_queue.lock.dep_map);
191 b = container_of(_b, struct sock, sk_receive_queue.lock.dep_map);
192
193 /* unix_collect_skb(): listener -> embryo order. */
194 if (a->sk_state == TCP_LISTEN && unix_sk(b)->listener == a)
195 return -1;
196
197 /* Should never happen. Just to be symmetric. */
198 if (b->sk_state == TCP_LISTEN && unix_sk(a)->listener == b)
199 return 1;
200
201 return 0;
202}
203#endif
204
205static unsigned int unix_unbound_hash(struct sock *sk)
206{
207 unsigned long hash = (unsigned long)sk;
208
209 hash ^= hash >> 16;
210 hash ^= hash >> 8;
211 hash ^= sk->sk_type;
212
213 return hash & UNIX_HASH_MOD;
214}
215
216static unsigned int unix_bsd_hash(struct inode *i)
217{
218 return i->i_ino & UNIX_HASH_MOD;
219}
220
221static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
222 int addr_len, int type)
223{
224 __wsum csum = csum_partial(sunaddr, addr_len, 0);
225 unsigned int hash;
226
227 hash = (__force unsigned int)csum_fold(csum);
228 hash ^= hash >> 8;
229 hash ^= type;
230
231 return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
232}
233
234static void unix_table_double_lock(struct net *net,
235 unsigned int hash1, unsigned int hash2)
236{
237 if (hash1 == hash2) {
238 spin_lock(&net->unx.table.locks[hash1]);
239 return;
240 }
241
242 if (hash1 > hash2)
243 swap(hash1, hash2);
244
245 spin_lock(&net->unx.table.locks[hash1]);
246 spin_lock(&net->unx.table.locks[hash2]);
247}
248
249static void unix_table_double_unlock(struct net *net,
250 unsigned int hash1, unsigned int hash2)
251{
252 if (hash1 == hash2) {
253 spin_unlock(&net->unx.table.locks[hash1]);
254 return;
255 }
256
257 spin_unlock(&net->unx.table.locks[hash1]);
258 spin_unlock(&net->unx.table.locks[hash2]);
259}
260
261#ifdef CONFIG_SECURITY_NETWORK
262static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
263{
264 UNIXCB(skb).secid = scm->secid;
265}
266
267static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
268{
269 scm->secid = UNIXCB(skb).secid;
270}
271
272static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
273{
274 return (scm->secid == UNIXCB(skb).secid);
275}
276#else
277static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
278{ }
279
280static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
281{ }
282
283static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
284{
285 return true;
286}
287#endif /* CONFIG_SECURITY_NETWORK */
288
289static inline int unix_our_peer(struct sock *sk, struct sock *osk)
290{
291 return unix_peer(osk) == sk;
292}
293
294static inline int unix_may_send(struct sock *sk, struct sock *osk)
295{
296 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
297}
298
299static inline int unix_recvq_full_lockless(const struct sock *sk)
300{
301 return skb_queue_len_lockless(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
302}
303
304struct sock *unix_peer_get(struct sock *s)
305{
306 struct sock *peer;
307
308 unix_state_lock(s);
309 peer = unix_peer(s);
310 if (peer)
311 sock_hold(peer);
312 unix_state_unlock(s);
313 return peer;
314}
315EXPORT_SYMBOL_GPL(unix_peer_get);
316
317static struct unix_address *unix_create_addr(struct sockaddr_un *sunaddr,
318 int addr_len)
319{
320 struct unix_address *addr;
321
322 addr = kmalloc(sizeof(*addr) + addr_len, GFP_KERNEL);
323 if (!addr)
324 return NULL;
325
326 refcount_set(&addr->refcnt, 1);
327 addr->len = addr_len;
328 memcpy(addr->name, sunaddr, addr_len);
329
330 return addr;
331}
332
333static inline void unix_release_addr(struct unix_address *addr)
334{
335 if (refcount_dec_and_test(&addr->refcnt))
336 kfree(addr);
337}
338
339/*
340 * Check unix socket name:
341 * - should be not zero length.
342 * - if started by not zero, should be NULL terminated (FS object)
343 * - if started by zero, it is abstract name.
344 */
345
346static int unix_validate_addr(struct sockaddr_un *sunaddr, int addr_len)
347{
348 if (addr_len <= offsetof(struct sockaddr_un, sun_path) ||
349 addr_len > sizeof(*sunaddr))
350 return -EINVAL;
351
352 if (sunaddr->sun_family != AF_UNIX)
353 return -EINVAL;
354
355 return 0;
356}
357
358static int unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
359{
360 struct sockaddr_storage *addr = (struct sockaddr_storage *)sunaddr;
361 short offset = offsetof(struct sockaddr_storage, __data);
362
363 BUILD_BUG_ON(offset != offsetof(struct sockaddr_un, sun_path));
364
365 /* This may look like an off by one error but it is a bit more
366 * subtle. 108 is the longest valid AF_UNIX path for a binding.
367 * sun_path[108] doesn't as such exist. However in kernel space
368 * we are guaranteed that it is a valid memory location in our
369 * kernel address buffer because syscall functions always pass
370 * a pointer of struct sockaddr_storage which has a bigger buffer
371 * than 108. Also, we must terminate sun_path for strlen() in
372 * getname_kernel().
373 */
374 addr->__data[addr_len - offset] = 0;
375
376 /* Don't pass sunaddr->sun_path to strlen(). Otherwise, 108 will
377 * cause panic if CONFIG_FORTIFY_SOURCE=y. Let __fortify_strlen()
378 * know the actual buffer.
379 */
380 return strlen(addr->__data) + offset + 1;
381}
382
383static void __unix_remove_socket(struct sock *sk)
384{
385 sk_del_node_init(sk);
386}
387
388static void __unix_insert_socket(struct net *net, struct sock *sk)
389{
390 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
391 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
392}
393
394static void __unix_set_addr_hash(struct net *net, struct sock *sk,
395 struct unix_address *addr, unsigned int hash)
396{
397 __unix_remove_socket(sk);
398 smp_store_release(&unix_sk(sk)->addr, addr);
399
400 sk->sk_hash = hash;
401 __unix_insert_socket(net, sk);
402}
403
404static void unix_remove_socket(struct net *net, struct sock *sk)
405{
406 spin_lock(&net->unx.table.locks[sk->sk_hash]);
407 __unix_remove_socket(sk);
408 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
409}
410
411static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
412{
413 spin_lock(&net->unx.table.locks[sk->sk_hash]);
414 __unix_insert_socket(net, sk);
415 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
416}
417
418static void unix_insert_bsd_socket(struct sock *sk)
419{
420 spin_lock(&bsd_socket_locks[sk->sk_hash]);
421 sk_add_bind_node(sk, &bsd_socket_buckets[sk->sk_hash]);
422 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
423}
424
425static void unix_remove_bsd_socket(struct sock *sk)
426{
427 if (!hlist_unhashed(&sk->sk_bind_node)) {
428 spin_lock(&bsd_socket_locks[sk->sk_hash]);
429 __sk_del_bind_node(sk);
430 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
431
432 sk_node_init(&sk->sk_bind_node);
433 }
434}
435
436static struct sock *__unix_find_socket_byname(struct net *net,
437 struct sockaddr_un *sunname,
438 int len, unsigned int hash)
439{
440 struct sock *s;
441
442 sk_for_each(s, &net->unx.table.buckets[hash]) {
443 struct unix_sock *u = unix_sk(s);
444
445 if (u->addr->len == len &&
446 !memcmp(u->addr->name, sunname, len))
447 return s;
448 }
449 return NULL;
450}
451
452static inline struct sock *unix_find_socket_byname(struct net *net,
453 struct sockaddr_un *sunname,
454 int len, unsigned int hash)
455{
456 struct sock *s;
457
458 spin_lock(&net->unx.table.locks[hash]);
459 s = __unix_find_socket_byname(net, sunname, len, hash);
460 if (s)
461 sock_hold(s);
462 spin_unlock(&net->unx.table.locks[hash]);
463 return s;
464}
465
466static struct sock *unix_find_socket_byinode(struct inode *i)
467{
468 unsigned int hash = unix_bsd_hash(i);
469 struct sock *s;
470
471 spin_lock(&bsd_socket_locks[hash]);
472 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
473 struct dentry *dentry = unix_sk(s)->path.dentry;
474
475 if (dentry && d_backing_inode(dentry) == i) {
476 sock_hold(s);
477 spin_unlock(&bsd_socket_locks[hash]);
478 return s;
479 }
480 }
481 spin_unlock(&bsd_socket_locks[hash]);
482 return NULL;
483}
484
485/* Support code for asymmetrically connected dgram sockets
486 *
487 * If a datagram socket is connected to a socket not itself connected
488 * to the first socket (eg, /dev/log), clients may only enqueue more
489 * messages if the present receive queue of the server socket is not
490 * "too large". This means there's a second writeability condition
491 * poll and sendmsg need to test. The dgram recv code will do a wake
492 * up on the peer_wait wait queue of a socket upon reception of a
493 * datagram which needs to be propagated to sleeping would-be writers
494 * since these might not have sent anything so far. This can't be
495 * accomplished via poll_wait because the lifetime of the server
496 * socket might be less than that of its clients if these break their
497 * association with it or if the server socket is closed while clients
498 * are still connected to it and there's no way to inform "a polling
499 * implementation" that it should let go of a certain wait queue
500 *
501 * In order to propagate a wake up, a wait_queue_entry_t of the client
502 * socket is enqueued on the peer_wait queue of the server socket
503 * whose wake function does a wake_up on the ordinary client socket
504 * wait queue. This connection is established whenever a write (or
505 * poll for write) hit the flow control condition and broken when the
506 * association to the server socket is dissolved or after a wake up
507 * was relayed.
508 */
509
510static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
511 void *key)
512{
513 struct unix_sock *u;
514 wait_queue_head_t *u_sleep;
515
516 u = container_of(q, struct unix_sock, peer_wake);
517
518 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
519 q);
520 u->peer_wake.private = NULL;
521
522 /* relaying can only happen while the wq still exists */
523 u_sleep = sk_sleep(&u->sk);
524 if (u_sleep)
525 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
526
527 return 0;
528}
529
530static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
531{
532 struct unix_sock *u, *u_other;
533 int rc;
534
535 u = unix_sk(sk);
536 u_other = unix_sk(other);
537 rc = 0;
538 spin_lock(&u_other->peer_wait.lock);
539
540 if (!u->peer_wake.private) {
541 u->peer_wake.private = other;
542 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
543
544 rc = 1;
545 }
546
547 spin_unlock(&u_other->peer_wait.lock);
548 return rc;
549}
550
551static void unix_dgram_peer_wake_disconnect(struct sock *sk,
552 struct sock *other)
553{
554 struct unix_sock *u, *u_other;
555
556 u = unix_sk(sk);
557 u_other = unix_sk(other);
558 spin_lock(&u_other->peer_wait.lock);
559
560 if (u->peer_wake.private == other) {
561 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
562 u->peer_wake.private = NULL;
563 }
564
565 spin_unlock(&u_other->peer_wait.lock);
566}
567
568static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
569 struct sock *other)
570{
571 unix_dgram_peer_wake_disconnect(sk, other);
572 wake_up_interruptible_poll(sk_sleep(sk),
573 EPOLLOUT |
574 EPOLLWRNORM |
575 EPOLLWRBAND);
576}
577
578/* preconditions:
579 * - unix_peer(sk) == other
580 * - association is stable
581 */
582static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
583{
584 int connected;
585
586 connected = unix_dgram_peer_wake_connect(sk, other);
587
588 /* If other is SOCK_DEAD, we want to make sure we signal
589 * POLLOUT, such that a subsequent write() can get a
590 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
591 * to other and its full, we will hang waiting for POLLOUT.
592 */
593 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
594 return 1;
595
596 if (connected)
597 unix_dgram_peer_wake_disconnect(sk, other);
598
599 return 0;
600}
601
602static int unix_writable(const struct sock *sk, unsigned char state)
603{
604 return state != TCP_LISTEN &&
605 (refcount_read(&sk->sk_wmem_alloc) << 2) <= READ_ONCE(sk->sk_sndbuf);
606}
607
608static void unix_write_space(struct sock *sk)
609{
610 struct socket_wq *wq;
611
612 rcu_read_lock();
613 if (unix_writable(sk, READ_ONCE(sk->sk_state))) {
614 wq = rcu_dereference(sk->sk_wq);
615 if (skwq_has_sleeper(wq))
616 wake_up_interruptible_sync_poll(&wq->wait,
617 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
618 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
619 }
620 rcu_read_unlock();
621}
622
623/* When dgram socket disconnects (or changes its peer), we clear its receive
624 * queue of packets arrived from previous peer. First, it allows to do
625 * flow control based only on wmem_alloc; second, sk connected to peer
626 * may receive messages only from that peer. */
627static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
628{
629 if (!skb_queue_empty(&sk->sk_receive_queue)) {
630 skb_queue_purge(&sk->sk_receive_queue);
631 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
632
633 /* If one link of bidirectional dgram pipe is disconnected,
634 * we signal error. Messages are lost. Do not make this,
635 * when peer was not connected to us.
636 */
637 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
638 WRITE_ONCE(other->sk_err, ECONNRESET);
639 sk_error_report(other);
640 }
641 }
642}
643
644static void unix_sock_destructor(struct sock *sk)
645{
646 struct unix_sock *u = unix_sk(sk);
647
648 skb_queue_purge(&sk->sk_receive_queue);
649
650 DEBUG_NET_WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
651 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
652 DEBUG_NET_WARN_ON_ONCE(sk->sk_socket);
653 if (!sock_flag(sk, SOCK_DEAD)) {
654 pr_info("Attempt to release alive unix socket: %p\n", sk);
655 return;
656 }
657
658 if (u->addr)
659 unix_release_addr(u->addr);
660
661 atomic_long_dec(&unix_nr_socks);
662 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
663#ifdef UNIX_REFCNT_DEBUG
664 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
665 atomic_long_read(&unix_nr_socks));
666#endif
667}
668
669static void unix_release_sock(struct sock *sk, int embrion)
670{
671 struct unix_sock *u = unix_sk(sk);
672 struct sock *skpair;
673 struct sk_buff *skb;
674 struct path path;
675 int state;
676
677 unix_remove_socket(sock_net(sk), sk);
678 unix_remove_bsd_socket(sk);
679
680 /* Clear state */
681 unix_state_lock(sk);
682 sock_orphan(sk);
683 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
684 path = u->path;
685 u->path.dentry = NULL;
686 u->path.mnt = NULL;
687 state = sk->sk_state;
688 WRITE_ONCE(sk->sk_state, TCP_CLOSE);
689
690 skpair = unix_peer(sk);
691 unix_peer(sk) = NULL;
692
693 unix_state_unlock(sk);
694
695#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
696 u->oob_skb = NULL;
697#endif
698
699 wake_up_interruptible_all(&u->peer_wait);
700
701 if (skpair != NULL) {
702 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
703 unix_state_lock(skpair);
704 /* No more writes */
705 WRITE_ONCE(skpair->sk_shutdown, SHUTDOWN_MASK);
706 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) || embrion)
707 WRITE_ONCE(skpair->sk_err, ECONNRESET);
708 unix_state_unlock(skpair);
709 skpair->sk_state_change(skpair);
710 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
711 }
712
713 unix_dgram_peer_wake_disconnect(sk, skpair);
714 sock_put(skpair); /* It may now die */
715 }
716
717 /* Try to flush out this socket. Throw out buffers at least */
718
719 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
720 if (state == TCP_LISTEN)
721 unix_release_sock(skb->sk, 1);
722
723 /* passed fds are erased in the kfree_skb hook */
724 kfree_skb(skb);
725 }
726
727 if (path.dentry)
728 path_put(&path);
729
730 sock_put(sk);
731
732 /* ---- Socket is dead now and most probably destroyed ---- */
733
734 /*
735 * Fixme: BSD difference: In BSD all sockets connected to us get
736 * ECONNRESET and we die on the spot. In Linux we behave
737 * like files and pipes do and wait for the last
738 * dereference.
739 *
740 * Can't we simply set sock->err?
741 *
742 * What the above comment does talk about? --ANK(980817)
743 */
744
745 if (READ_ONCE(unix_tot_inflight))
746 unix_gc(); /* Garbage collect fds */
747}
748
749static void init_peercred(struct sock *sk)
750{
751 sk->sk_peer_pid = get_pid(task_tgid(current));
752 sk->sk_peer_cred = get_current_cred();
753}
754
755static void update_peercred(struct sock *sk)
756{
757 const struct cred *old_cred;
758 struct pid *old_pid;
759
760 spin_lock(&sk->sk_peer_lock);
761 old_pid = sk->sk_peer_pid;
762 old_cred = sk->sk_peer_cred;
763 init_peercred(sk);
764 spin_unlock(&sk->sk_peer_lock);
765
766 put_pid(old_pid);
767 put_cred(old_cred);
768}
769
770static void copy_peercred(struct sock *sk, struct sock *peersk)
771{
772 lockdep_assert_held(&unix_sk(peersk)->lock);
773
774 spin_lock(&sk->sk_peer_lock);
775 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
776 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
777 spin_unlock(&sk->sk_peer_lock);
778}
779
780static int unix_listen(struct socket *sock, int backlog)
781{
782 int err;
783 struct sock *sk = sock->sk;
784 struct unix_sock *u = unix_sk(sk);
785
786 err = -EOPNOTSUPP;
787 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
788 goto out; /* Only stream/seqpacket sockets accept */
789 err = -EINVAL;
790 if (!READ_ONCE(u->addr))
791 goto out; /* No listens on an unbound socket */
792 unix_state_lock(sk);
793 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
794 goto out_unlock;
795 if (backlog > sk->sk_max_ack_backlog)
796 wake_up_interruptible_all(&u->peer_wait);
797 sk->sk_max_ack_backlog = backlog;
798 WRITE_ONCE(sk->sk_state, TCP_LISTEN);
799
800 /* set credentials so connect can copy them */
801 update_peercred(sk);
802 err = 0;
803
804out_unlock:
805 unix_state_unlock(sk);
806out:
807 return err;
808}
809
810static int unix_release(struct socket *);
811static int unix_bind(struct socket *, struct sockaddr *, int);
812static int unix_stream_connect(struct socket *, struct sockaddr *,
813 int addr_len, int flags);
814static int unix_socketpair(struct socket *, struct socket *);
815static int unix_accept(struct socket *, struct socket *, struct proto_accept_arg *arg);
816static int unix_getname(struct socket *, struct sockaddr *, int);
817static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
818static __poll_t unix_dgram_poll(struct file *, struct socket *,
819 poll_table *);
820static int unix_ioctl(struct socket *, unsigned int, unsigned long);
821#ifdef CONFIG_COMPAT
822static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
823#endif
824static int unix_shutdown(struct socket *, int);
825static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
826static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
827static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
828 struct pipe_inode_info *, size_t size,
829 unsigned int flags);
830static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
831static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
832static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
833static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
834static int unix_dgram_connect(struct socket *, struct sockaddr *,
835 int, int);
836static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
837static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
838 int);
839
840#ifdef CONFIG_PROC_FS
841static int unix_count_nr_fds(struct sock *sk)
842{
843 struct sk_buff *skb;
844 struct unix_sock *u;
845 int nr_fds = 0;
846
847 spin_lock(&sk->sk_receive_queue.lock);
848 skb = skb_peek(&sk->sk_receive_queue);
849 while (skb) {
850 u = unix_sk(skb->sk);
851 nr_fds += atomic_read(&u->scm_stat.nr_fds);
852 skb = skb_peek_next(skb, &sk->sk_receive_queue);
853 }
854 spin_unlock(&sk->sk_receive_queue.lock);
855
856 return nr_fds;
857}
858
859static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
860{
861 struct sock *sk = sock->sk;
862 unsigned char s_state;
863 struct unix_sock *u;
864 int nr_fds = 0;
865
866 if (sk) {
867 s_state = READ_ONCE(sk->sk_state);
868 u = unix_sk(sk);
869
870 /* SOCK_STREAM and SOCK_SEQPACKET sockets never change their
871 * sk_state after switching to TCP_ESTABLISHED or TCP_LISTEN.
872 * SOCK_DGRAM is ordinary. So, no lock is needed.
873 */
874 if (sock->type == SOCK_DGRAM || s_state == TCP_ESTABLISHED)
875 nr_fds = atomic_read(&u->scm_stat.nr_fds);
876 else if (s_state == TCP_LISTEN)
877 nr_fds = unix_count_nr_fds(sk);
878
879 seq_printf(m, "scm_fds: %u\n", nr_fds);
880 }
881}
882#else
883#define unix_show_fdinfo NULL
884#endif
885
886static const struct proto_ops unix_stream_ops = {
887 .family = PF_UNIX,
888 .owner = THIS_MODULE,
889 .release = unix_release,
890 .bind = unix_bind,
891 .connect = unix_stream_connect,
892 .socketpair = unix_socketpair,
893 .accept = unix_accept,
894 .getname = unix_getname,
895 .poll = unix_poll,
896 .ioctl = unix_ioctl,
897#ifdef CONFIG_COMPAT
898 .compat_ioctl = unix_compat_ioctl,
899#endif
900 .listen = unix_listen,
901 .shutdown = unix_shutdown,
902 .sendmsg = unix_stream_sendmsg,
903 .recvmsg = unix_stream_recvmsg,
904 .read_skb = unix_stream_read_skb,
905 .mmap = sock_no_mmap,
906 .splice_read = unix_stream_splice_read,
907 .set_peek_off = sk_set_peek_off,
908 .show_fdinfo = unix_show_fdinfo,
909};
910
911static const struct proto_ops unix_dgram_ops = {
912 .family = PF_UNIX,
913 .owner = THIS_MODULE,
914 .release = unix_release,
915 .bind = unix_bind,
916 .connect = unix_dgram_connect,
917 .socketpair = unix_socketpair,
918 .accept = sock_no_accept,
919 .getname = unix_getname,
920 .poll = unix_dgram_poll,
921 .ioctl = unix_ioctl,
922#ifdef CONFIG_COMPAT
923 .compat_ioctl = unix_compat_ioctl,
924#endif
925 .listen = sock_no_listen,
926 .shutdown = unix_shutdown,
927 .sendmsg = unix_dgram_sendmsg,
928 .read_skb = unix_read_skb,
929 .recvmsg = unix_dgram_recvmsg,
930 .mmap = sock_no_mmap,
931 .set_peek_off = sk_set_peek_off,
932 .show_fdinfo = unix_show_fdinfo,
933};
934
935static const struct proto_ops unix_seqpacket_ops = {
936 .family = PF_UNIX,
937 .owner = THIS_MODULE,
938 .release = unix_release,
939 .bind = unix_bind,
940 .connect = unix_stream_connect,
941 .socketpair = unix_socketpair,
942 .accept = unix_accept,
943 .getname = unix_getname,
944 .poll = unix_dgram_poll,
945 .ioctl = unix_ioctl,
946#ifdef CONFIG_COMPAT
947 .compat_ioctl = unix_compat_ioctl,
948#endif
949 .listen = unix_listen,
950 .shutdown = unix_shutdown,
951 .sendmsg = unix_seqpacket_sendmsg,
952 .recvmsg = unix_seqpacket_recvmsg,
953 .mmap = sock_no_mmap,
954 .set_peek_off = sk_set_peek_off,
955 .show_fdinfo = unix_show_fdinfo,
956};
957
958static void unix_close(struct sock *sk, long timeout)
959{
960 /* Nothing to do here, unix socket does not need a ->close().
961 * This is merely for sockmap.
962 */
963}
964
965static void unix_unhash(struct sock *sk)
966{
967 /* Nothing to do here, unix socket does not need a ->unhash().
968 * This is merely for sockmap.
969 */
970}
971
972static bool unix_bpf_bypass_getsockopt(int level, int optname)
973{
974 if (level == SOL_SOCKET) {
975 switch (optname) {
976 case SO_PEERPIDFD:
977 return true;
978 default:
979 return false;
980 }
981 }
982
983 return false;
984}
985
986struct proto unix_dgram_proto = {
987 .name = "UNIX",
988 .owner = THIS_MODULE,
989 .obj_size = sizeof(struct unix_sock),
990 .close = unix_close,
991 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
992#ifdef CONFIG_BPF_SYSCALL
993 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
994#endif
995};
996
997struct proto unix_stream_proto = {
998 .name = "UNIX-STREAM",
999 .owner = THIS_MODULE,
1000 .obj_size = sizeof(struct unix_sock),
1001 .close = unix_close,
1002 .unhash = unix_unhash,
1003 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
1004#ifdef CONFIG_BPF_SYSCALL
1005 .psock_update_sk_prot = unix_stream_bpf_update_proto,
1006#endif
1007};
1008
1009static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
1010{
1011 struct unix_sock *u;
1012 struct sock *sk;
1013 int err;
1014
1015 atomic_long_inc(&unix_nr_socks);
1016 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
1017 err = -ENFILE;
1018 goto err;
1019 }
1020
1021 if (type == SOCK_STREAM)
1022 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
1023 else /*dgram and seqpacket */
1024 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
1025
1026 if (!sk) {
1027 err = -ENOMEM;
1028 goto err;
1029 }
1030
1031 sock_init_data(sock, sk);
1032
1033 sk->sk_hash = unix_unbound_hash(sk);
1034 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
1035 sk->sk_write_space = unix_write_space;
1036 sk->sk_max_ack_backlog = READ_ONCE(net->unx.sysctl_max_dgram_qlen);
1037 sk->sk_destruct = unix_sock_destructor;
1038 lock_set_cmp_fn(&sk->sk_receive_queue.lock, unix_recvq_lock_cmp_fn, NULL);
1039
1040 u = unix_sk(sk);
1041 u->listener = NULL;
1042 u->vertex = NULL;
1043 u->path.dentry = NULL;
1044 u->path.mnt = NULL;
1045 spin_lock_init(&u->lock);
1046 lock_set_cmp_fn(&u->lock, unix_state_lock_cmp_fn, NULL);
1047 mutex_init(&u->iolock); /* single task reading lock */
1048 mutex_init(&u->bindlock); /* single task binding lock */
1049 init_waitqueue_head(&u->peer_wait);
1050 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
1051 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
1052 unix_insert_unbound_socket(net, sk);
1053
1054 sock_prot_inuse_add(net, sk->sk_prot, 1);
1055
1056 return sk;
1057
1058err:
1059 atomic_long_dec(&unix_nr_socks);
1060 return ERR_PTR(err);
1061}
1062
1063static int unix_create(struct net *net, struct socket *sock, int protocol,
1064 int kern)
1065{
1066 struct sock *sk;
1067
1068 if (protocol && protocol != PF_UNIX)
1069 return -EPROTONOSUPPORT;
1070
1071 sock->state = SS_UNCONNECTED;
1072
1073 switch (sock->type) {
1074 case SOCK_STREAM:
1075 sock->ops = &unix_stream_ops;
1076 break;
1077 /*
1078 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
1079 * nothing uses it.
1080 */
1081 case SOCK_RAW:
1082 sock->type = SOCK_DGRAM;
1083 fallthrough;
1084 case SOCK_DGRAM:
1085 sock->ops = &unix_dgram_ops;
1086 break;
1087 case SOCK_SEQPACKET:
1088 sock->ops = &unix_seqpacket_ops;
1089 break;
1090 default:
1091 return -ESOCKTNOSUPPORT;
1092 }
1093
1094 sk = unix_create1(net, sock, kern, sock->type);
1095 if (IS_ERR(sk))
1096 return PTR_ERR(sk);
1097
1098 return 0;
1099}
1100
1101static int unix_release(struct socket *sock)
1102{
1103 struct sock *sk = sock->sk;
1104
1105 if (!sk)
1106 return 0;
1107
1108 sk->sk_prot->close(sk, 0);
1109 unix_release_sock(sk, 0);
1110 sock->sk = NULL;
1111
1112 return 0;
1113}
1114
1115static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1116 int type)
1117{
1118 struct inode *inode;
1119 struct path path;
1120 struct sock *sk;
1121 int err;
1122
1123 unix_mkname_bsd(sunaddr, addr_len);
1124 err = kern_path(sunaddr->sun_path, LOOKUP_FOLLOW, &path);
1125 if (err)
1126 goto fail;
1127
1128 err = path_permission(&path, MAY_WRITE);
1129 if (err)
1130 goto path_put;
1131
1132 err = -ECONNREFUSED;
1133 inode = d_backing_inode(path.dentry);
1134 if (!S_ISSOCK(inode->i_mode))
1135 goto path_put;
1136
1137 sk = unix_find_socket_byinode(inode);
1138 if (!sk)
1139 goto path_put;
1140
1141 err = -EPROTOTYPE;
1142 if (sk->sk_type == type)
1143 touch_atime(&path);
1144 else
1145 goto sock_put;
1146
1147 path_put(&path);
1148
1149 return sk;
1150
1151sock_put:
1152 sock_put(sk);
1153path_put:
1154 path_put(&path);
1155fail:
1156 return ERR_PTR(err);
1157}
1158
1159static struct sock *unix_find_abstract(struct net *net,
1160 struct sockaddr_un *sunaddr,
1161 int addr_len, int type)
1162{
1163 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
1164 struct dentry *dentry;
1165 struct sock *sk;
1166
1167 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
1168 if (!sk)
1169 return ERR_PTR(-ECONNREFUSED);
1170
1171 dentry = unix_sk(sk)->path.dentry;
1172 if (dentry)
1173 touch_atime(&unix_sk(sk)->path);
1174
1175 return sk;
1176}
1177
1178static struct sock *unix_find_other(struct net *net,
1179 struct sockaddr_un *sunaddr,
1180 int addr_len, int type)
1181{
1182 struct sock *sk;
1183
1184 if (sunaddr->sun_path[0])
1185 sk = unix_find_bsd(sunaddr, addr_len, type);
1186 else
1187 sk = unix_find_abstract(net, sunaddr, addr_len, type);
1188
1189 return sk;
1190}
1191
1192static int unix_autobind(struct sock *sk)
1193{
1194 struct unix_sock *u = unix_sk(sk);
1195 unsigned int new_hash, old_hash;
1196 struct net *net = sock_net(sk);
1197 struct unix_address *addr;
1198 u32 lastnum, ordernum;
1199 int err;
1200
1201 err = mutex_lock_interruptible(&u->bindlock);
1202 if (err)
1203 return err;
1204
1205 if (u->addr)
1206 goto out;
1207
1208 err = -ENOMEM;
1209 addr = kzalloc(sizeof(*addr) +
1210 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1211 if (!addr)
1212 goto out;
1213
1214 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1215 addr->name->sun_family = AF_UNIX;
1216 refcount_set(&addr->refcnt, 1);
1217
1218 old_hash = sk->sk_hash;
1219 ordernum = get_random_u32();
1220 lastnum = ordernum & 0xFFFFF;
1221retry:
1222 ordernum = (ordernum + 1) & 0xFFFFF;
1223 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1224
1225 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1226 unix_table_double_lock(net, old_hash, new_hash);
1227
1228 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1229 unix_table_double_unlock(net, old_hash, new_hash);
1230
1231 /* __unix_find_socket_byname() may take long time if many names
1232 * are already in use.
1233 */
1234 cond_resched();
1235
1236 if (ordernum == lastnum) {
1237 /* Give up if all names seems to be in use. */
1238 err = -ENOSPC;
1239 unix_release_addr(addr);
1240 goto out;
1241 }
1242
1243 goto retry;
1244 }
1245
1246 __unix_set_addr_hash(net, sk, addr, new_hash);
1247 unix_table_double_unlock(net, old_hash, new_hash);
1248 err = 0;
1249
1250out: mutex_unlock(&u->bindlock);
1251 return err;
1252}
1253
1254static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1255 int addr_len)
1256{
1257 umode_t mode = S_IFSOCK |
1258 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1259 struct unix_sock *u = unix_sk(sk);
1260 unsigned int new_hash, old_hash;
1261 struct net *net = sock_net(sk);
1262 struct mnt_idmap *idmap;
1263 struct unix_address *addr;
1264 struct dentry *dentry;
1265 struct path parent;
1266 int err;
1267
1268 addr_len = unix_mkname_bsd(sunaddr, addr_len);
1269 addr = unix_create_addr(sunaddr, addr_len);
1270 if (!addr)
1271 return -ENOMEM;
1272
1273 /*
1274 * Get the parent directory, calculate the hash for last
1275 * component.
1276 */
1277 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1278 if (IS_ERR(dentry)) {
1279 err = PTR_ERR(dentry);
1280 goto out;
1281 }
1282
1283 /*
1284 * All right, let's create it.
1285 */
1286 idmap = mnt_idmap(parent.mnt);
1287 err = security_path_mknod(&parent, dentry, mode, 0);
1288 if (!err)
1289 err = vfs_mknod(idmap, d_inode(parent.dentry), dentry, mode, 0);
1290 if (err)
1291 goto out_path;
1292 err = mutex_lock_interruptible(&u->bindlock);
1293 if (err)
1294 goto out_unlink;
1295 if (u->addr)
1296 goto out_unlock;
1297
1298 old_hash = sk->sk_hash;
1299 new_hash = unix_bsd_hash(d_backing_inode(dentry));
1300 unix_table_double_lock(net, old_hash, new_hash);
1301 u->path.mnt = mntget(parent.mnt);
1302 u->path.dentry = dget(dentry);
1303 __unix_set_addr_hash(net, sk, addr, new_hash);
1304 unix_table_double_unlock(net, old_hash, new_hash);
1305 unix_insert_bsd_socket(sk);
1306 mutex_unlock(&u->bindlock);
1307 done_path_create(&parent, dentry);
1308 return 0;
1309
1310out_unlock:
1311 mutex_unlock(&u->bindlock);
1312 err = -EINVAL;
1313out_unlink:
1314 /* failed after successful mknod? unlink what we'd created... */
1315 vfs_unlink(idmap, d_inode(parent.dentry), dentry, NULL);
1316out_path:
1317 done_path_create(&parent, dentry);
1318out:
1319 unix_release_addr(addr);
1320 return err == -EEXIST ? -EADDRINUSE : err;
1321}
1322
1323static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1324 int addr_len)
1325{
1326 struct unix_sock *u = unix_sk(sk);
1327 unsigned int new_hash, old_hash;
1328 struct net *net = sock_net(sk);
1329 struct unix_address *addr;
1330 int err;
1331
1332 addr = unix_create_addr(sunaddr, addr_len);
1333 if (!addr)
1334 return -ENOMEM;
1335
1336 err = mutex_lock_interruptible(&u->bindlock);
1337 if (err)
1338 goto out;
1339
1340 if (u->addr) {
1341 err = -EINVAL;
1342 goto out_mutex;
1343 }
1344
1345 old_hash = sk->sk_hash;
1346 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1347 unix_table_double_lock(net, old_hash, new_hash);
1348
1349 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
1350 goto out_spin;
1351
1352 __unix_set_addr_hash(net, sk, addr, new_hash);
1353 unix_table_double_unlock(net, old_hash, new_hash);
1354 mutex_unlock(&u->bindlock);
1355 return 0;
1356
1357out_spin:
1358 unix_table_double_unlock(net, old_hash, new_hash);
1359 err = -EADDRINUSE;
1360out_mutex:
1361 mutex_unlock(&u->bindlock);
1362out:
1363 unix_release_addr(addr);
1364 return err;
1365}
1366
1367static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1368{
1369 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1370 struct sock *sk = sock->sk;
1371 int err;
1372
1373 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1374 sunaddr->sun_family == AF_UNIX)
1375 return unix_autobind(sk);
1376
1377 err = unix_validate_addr(sunaddr, addr_len);
1378 if (err)
1379 return err;
1380
1381 if (sunaddr->sun_path[0])
1382 err = unix_bind_bsd(sk, sunaddr, addr_len);
1383 else
1384 err = unix_bind_abstract(sk, sunaddr, addr_len);
1385
1386 return err;
1387}
1388
1389static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1390{
1391 if (unlikely(sk1 == sk2) || !sk2) {
1392 unix_state_lock(sk1);
1393 return;
1394 }
1395
1396 if (sk1 > sk2)
1397 swap(sk1, sk2);
1398
1399 unix_state_lock(sk1);
1400 unix_state_lock(sk2);
1401}
1402
1403static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1404{
1405 if (unlikely(sk1 == sk2) || !sk2) {
1406 unix_state_unlock(sk1);
1407 return;
1408 }
1409 unix_state_unlock(sk1);
1410 unix_state_unlock(sk2);
1411}
1412
1413static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1414 int alen, int flags)
1415{
1416 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1417 struct sock *sk = sock->sk;
1418 struct sock *other;
1419 int err;
1420
1421 err = -EINVAL;
1422 if (alen < offsetofend(struct sockaddr, sa_family))
1423 goto out;
1424
1425 if (addr->sa_family != AF_UNSPEC) {
1426 err = unix_validate_addr(sunaddr, alen);
1427 if (err)
1428 goto out;
1429
1430 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, addr, &alen);
1431 if (err)
1432 goto out;
1433
1434 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1435 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1436 !READ_ONCE(unix_sk(sk)->addr)) {
1437 err = unix_autobind(sk);
1438 if (err)
1439 goto out;
1440 }
1441
1442restart:
1443 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
1444 if (IS_ERR(other)) {
1445 err = PTR_ERR(other);
1446 goto out;
1447 }
1448
1449 unix_state_double_lock(sk, other);
1450
1451 /* Apparently VFS overslept socket death. Retry. */
1452 if (sock_flag(other, SOCK_DEAD)) {
1453 unix_state_double_unlock(sk, other);
1454 sock_put(other);
1455 goto restart;
1456 }
1457
1458 err = -EPERM;
1459 if (!unix_may_send(sk, other))
1460 goto out_unlock;
1461
1462 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1463 if (err)
1464 goto out_unlock;
1465
1466 WRITE_ONCE(sk->sk_state, TCP_ESTABLISHED);
1467 WRITE_ONCE(other->sk_state, TCP_ESTABLISHED);
1468 } else {
1469 /*
1470 * 1003.1g breaking connected state with AF_UNSPEC
1471 */
1472 other = NULL;
1473 unix_state_double_lock(sk, other);
1474 }
1475
1476 /*
1477 * If it was connected, reconnect.
1478 */
1479 if (unix_peer(sk)) {
1480 struct sock *old_peer = unix_peer(sk);
1481
1482 unix_peer(sk) = other;
1483 if (!other)
1484 WRITE_ONCE(sk->sk_state, TCP_CLOSE);
1485 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1486
1487 unix_state_double_unlock(sk, other);
1488
1489 if (other != old_peer) {
1490 unix_dgram_disconnected(sk, old_peer);
1491
1492 unix_state_lock(old_peer);
1493 if (!unix_peer(old_peer))
1494 WRITE_ONCE(old_peer->sk_state, TCP_CLOSE);
1495 unix_state_unlock(old_peer);
1496 }
1497
1498 sock_put(old_peer);
1499 } else {
1500 unix_peer(sk) = other;
1501 unix_state_double_unlock(sk, other);
1502 }
1503
1504 return 0;
1505
1506out_unlock:
1507 unix_state_double_unlock(sk, other);
1508 sock_put(other);
1509out:
1510 return err;
1511}
1512
1513static long unix_wait_for_peer(struct sock *other, long timeo)
1514 __releases(&unix_sk(other)->lock)
1515{
1516 struct unix_sock *u = unix_sk(other);
1517 int sched;
1518 DEFINE_WAIT(wait);
1519
1520 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1521
1522 sched = !sock_flag(other, SOCK_DEAD) &&
1523 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1524 unix_recvq_full_lockless(other);
1525
1526 unix_state_unlock(other);
1527
1528 if (sched)
1529 timeo = schedule_timeout(timeo);
1530
1531 finish_wait(&u->peer_wait, &wait);
1532 return timeo;
1533}
1534
1535static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1536 int addr_len, int flags)
1537{
1538 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1539 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1540 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1541 struct net *net = sock_net(sk);
1542 struct sk_buff *skb = NULL;
1543 unsigned char state;
1544 long timeo;
1545 int err;
1546
1547 err = unix_validate_addr(sunaddr, addr_len);
1548 if (err)
1549 goto out;
1550
1551 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, &addr_len);
1552 if (err)
1553 goto out;
1554
1555 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1556 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1557 !READ_ONCE(u->addr)) {
1558 err = unix_autobind(sk);
1559 if (err)
1560 goto out;
1561 }
1562
1563 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1564
1565 /* First of all allocate resources.
1566 If we will make it after state is locked,
1567 we will have to recheck all again in any case.
1568 */
1569
1570 /* create new sock for complete connection */
1571 newsk = unix_create1(net, NULL, 0, sock->type);
1572 if (IS_ERR(newsk)) {
1573 err = PTR_ERR(newsk);
1574 newsk = NULL;
1575 goto out;
1576 }
1577
1578 err = -ENOMEM;
1579
1580 /* Allocate skb for sending to listening sock */
1581 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1582 if (skb == NULL)
1583 goto out;
1584
1585restart:
1586 /* Find listening sock. */
1587 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
1588 if (IS_ERR(other)) {
1589 err = PTR_ERR(other);
1590 other = NULL;
1591 goto out;
1592 }
1593
1594 unix_state_lock(other);
1595
1596 /* Apparently VFS overslept socket death. Retry. */
1597 if (sock_flag(other, SOCK_DEAD)) {
1598 unix_state_unlock(other);
1599 sock_put(other);
1600 goto restart;
1601 }
1602
1603 err = -ECONNREFUSED;
1604 if (other->sk_state != TCP_LISTEN)
1605 goto out_unlock;
1606 if (other->sk_shutdown & RCV_SHUTDOWN)
1607 goto out_unlock;
1608
1609 if (unix_recvq_full_lockless(other)) {
1610 err = -EAGAIN;
1611 if (!timeo)
1612 goto out_unlock;
1613
1614 timeo = unix_wait_for_peer(other, timeo);
1615
1616 err = sock_intr_errno(timeo);
1617 if (signal_pending(current))
1618 goto out;
1619 sock_put(other);
1620 goto restart;
1621 }
1622
1623 /* self connect and simultaneous connect are eliminated
1624 * by rejecting TCP_LISTEN socket to avoid deadlock.
1625 */
1626 state = READ_ONCE(sk->sk_state);
1627 if (unlikely(state != TCP_CLOSE)) {
1628 err = state == TCP_ESTABLISHED ? -EISCONN : -EINVAL;
1629 goto out_unlock;
1630 }
1631
1632 unix_state_lock(sk);
1633
1634 if (unlikely(sk->sk_state != TCP_CLOSE)) {
1635 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EINVAL;
1636 unix_state_unlock(sk);
1637 goto out_unlock;
1638 }
1639
1640 err = security_unix_stream_connect(sk, other, newsk);
1641 if (err) {
1642 unix_state_unlock(sk);
1643 goto out_unlock;
1644 }
1645
1646 /* The way is open! Fastly set all the necessary fields... */
1647
1648 sock_hold(sk);
1649 unix_peer(newsk) = sk;
1650 newsk->sk_state = TCP_ESTABLISHED;
1651 newsk->sk_type = sk->sk_type;
1652 init_peercred(newsk);
1653 newu = unix_sk(newsk);
1654 newu->listener = other;
1655 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1656 otheru = unix_sk(other);
1657
1658 /* copy address information from listening to new sock
1659 *
1660 * The contents of *(otheru->addr) and otheru->path
1661 * are seen fully set up here, since we have found
1662 * otheru in hash under its lock. Insertion into the
1663 * hash chain we'd found it in had been done in an
1664 * earlier critical area protected by the chain's lock,
1665 * the same one where we'd set *(otheru->addr) contents,
1666 * as well as otheru->path and otheru->addr itself.
1667 *
1668 * Using smp_store_release() here to set newu->addr
1669 * is enough to make those stores, as well as stores
1670 * to newu->path visible to anyone who gets newu->addr
1671 * by smp_load_acquire(). IOW, the same warranties
1672 * as for unix_sock instances bound in unix_bind() or
1673 * in unix_autobind().
1674 */
1675 if (otheru->path.dentry) {
1676 path_get(&otheru->path);
1677 newu->path = otheru->path;
1678 }
1679 refcount_inc(&otheru->addr->refcnt);
1680 smp_store_release(&newu->addr, otheru->addr);
1681
1682 /* Set credentials */
1683 copy_peercred(sk, other);
1684
1685 sock->state = SS_CONNECTED;
1686 WRITE_ONCE(sk->sk_state, TCP_ESTABLISHED);
1687 sock_hold(newsk);
1688
1689 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1690 unix_peer(sk) = newsk;
1691
1692 unix_state_unlock(sk);
1693
1694 /* take ten and send info to listening sock */
1695 spin_lock(&other->sk_receive_queue.lock);
1696 __skb_queue_tail(&other->sk_receive_queue, skb);
1697 spin_unlock(&other->sk_receive_queue.lock);
1698 unix_state_unlock(other);
1699 other->sk_data_ready(other);
1700 sock_put(other);
1701 return 0;
1702
1703out_unlock:
1704 if (other)
1705 unix_state_unlock(other);
1706
1707out:
1708 kfree_skb(skb);
1709 if (newsk)
1710 unix_release_sock(newsk, 0);
1711 if (other)
1712 sock_put(other);
1713 return err;
1714}
1715
1716static int unix_socketpair(struct socket *socka, struct socket *sockb)
1717{
1718 struct sock *ska = socka->sk, *skb = sockb->sk;
1719
1720 /* Join our sockets back to back */
1721 sock_hold(ska);
1722 sock_hold(skb);
1723 unix_peer(ska) = skb;
1724 unix_peer(skb) = ska;
1725 init_peercred(ska);
1726 init_peercred(skb);
1727
1728 ska->sk_state = TCP_ESTABLISHED;
1729 skb->sk_state = TCP_ESTABLISHED;
1730 socka->state = SS_CONNECTED;
1731 sockb->state = SS_CONNECTED;
1732 return 0;
1733}
1734
1735static void unix_sock_inherit_flags(const struct socket *old,
1736 struct socket *new)
1737{
1738 if (test_bit(SOCK_PASSCRED, &old->flags))
1739 set_bit(SOCK_PASSCRED, &new->flags);
1740 if (test_bit(SOCK_PASSPIDFD, &old->flags))
1741 set_bit(SOCK_PASSPIDFD, &new->flags);
1742 if (test_bit(SOCK_PASSSEC, &old->flags))
1743 set_bit(SOCK_PASSSEC, &new->flags);
1744}
1745
1746static int unix_accept(struct socket *sock, struct socket *newsock,
1747 struct proto_accept_arg *arg)
1748{
1749 struct sock *sk = sock->sk;
1750 struct sk_buff *skb;
1751 struct sock *tsk;
1752
1753 arg->err = -EOPNOTSUPP;
1754 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1755 goto out;
1756
1757 arg->err = -EINVAL;
1758 if (READ_ONCE(sk->sk_state) != TCP_LISTEN)
1759 goto out;
1760
1761 /* If socket state is TCP_LISTEN it cannot change (for now...),
1762 * so that no locks are necessary.
1763 */
1764
1765 skb = skb_recv_datagram(sk, (arg->flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1766 &arg->err);
1767 if (!skb) {
1768 /* This means receive shutdown. */
1769 if (arg->err == 0)
1770 arg->err = -EINVAL;
1771 goto out;
1772 }
1773
1774 tsk = skb->sk;
1775 skb_free_datagram(sk, skb);
1776 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1777
1778 /* attach accepted sock to socket */
1779 unix_state_lock(tsk);
1780 unix_update_edges(unix_sk(tsk));
1781 newsock->state = SS_CONNECTED;
1782 unix_sock_inherit_flags(sock, newsock);
1783 sock_graft(tsk, newsock);
1784 unix_state_unlock(tsk);
1785 return 0;
1786
1787out:
1788 return arg->err;
1789}
1790
1791
1792static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1793{
1794 struct sock *sk = sock->sk;
1795 struct unix_address *addr;
1796 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1797 int err = 0;
1798
1799 if (peer) {
1800 sk = unix_peer_get(sk);
1801
1802 err = -ENOTCONN;
1803 if (!sk)
1804 goto out;
1805 err = 0;
1806 } else {
1807 sock_hold(sk);
1808 }
1809
1810 addr = smp_load_acquire(&unix_sk(sk)->addr);
1811 if (!addr) {
1812 sunaddr->sun_family = AF_UNIX;
1813 sunaddr->sun_path[0] = 0;
1814 err = offsetof(struct sockaddr_un, sun_path);
1815 } else {
1816 err = addr->len;
1817 memcpy(sunaddr, addr->name, addr->len);
1818
1819 if (peer)
1820 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1821 CGROUP_UNIX_GETPEERNAME);
1822 else
1823 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1824 CGROUP_UNIX_GETSOCKNAME);
1825 }
1826 sock_put(sk);
1827out:
1828 return err;
1829}
1830
1831/* The "user->unix_inflight" variable is protected by the garbage
1832 * collection lock, and we just read it locklessly here. If you go
1833 * over the limit, there might be a tiny race in actually noticing
1834 * it across threads. Tough.
1835 */
1836static inline bool too_many_unix_fds(struct task_struct *p)
1837{
1838 struct user_struct *user = current_user();
1839
1840 if (unlikely(READ_ONCE(user->unix_inflight) > task_rlimit(p, RLIMIT_NOFILE)))
1841 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1842 return false;
1843}
1844
1845static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1846{
1847 if (too_many_unix_fds(current))
1848 return -ETOOMANYREFS;
1849
1850 UNIXCB(skb).fp = scm->fp;
1851 scm->fp = NULL;
1852
1853 if (unix_prepare_fpl(UNIXCB(skb).fp))
1854 return -ENOMEM;
1855
1856 return 0;
1857}
1858
1859static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1860{
1861 scm->fp = UNIXCB(skb).fp;
1862 UNIXCB(skb).fp = NULL;
1863
1864 unix_destroy_fpl(scm->fp);
1865}
1866
1867static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1868{
1869 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1870}
1871
1872static void unix_destruct_scm(struct sk_buff *skb)
1873{
1874 struct scm_cookie scm;
1875
1876 memset(&scm, 0, sizeof(scm));
1877 scm.pid = UNIXCB(skb).pid;
1878 if (UNIXCB(skb).fp)
1879 unix_detach_fds(&scm, skb);
1880
1881 /* Alas, it calls VFS */
1882 /* So fscking what? fput() had been SMP-safe since the last Summer */
1883 scm_destroy(&scm);
1884 sock_wfree(skb);
1885}
1886
1887static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1888{
1889 int err = 0;
1890
1891 UNIXCB(skb).pid = get_pid(scm->pid);
1892 UNIXCB(skb).uid = scm->creds.uid;
1893 UNIXCB(skb).gid = scm->creds.gid;
1894 UNIXCB(skb).fp = NULL;
1895 unix_get_secdata(scm, skb);
1896 if (scm->fp && send_fds)
1897 err = unix_attach_fds(scm, skb);
1898
1899 skb->destructor = unix_destruct_scm;
1900 return err;
1901}
1902
1903static bool unix_passcred_enabled(const struct socket *sock,
1904 const struct sock *other)
1905{
1906 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1907 test_bit(SOCK_PASSPIDFD, &sock->flags) ||
1908 !other->sk_socket ||
1909 test_bit(SOCK_PASSCRED, &other->sk_socket->flags) ||
1910 test_bit(SOCK_PASSPIDFD, &other->sk_socket->flags);
1911}
1912
1913/*
1914 * Some apps rely on write() giving SCM_CREDENTIALS
1915 * We include credentials if source or destination socket
1916 * asserted SOCK_PASSCRED.
1917 */
1918static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1919 const struct sock *other)
1920{
1921 if (UNIXCB(skb).pid)
1922 return;
1923 if (unix_passcred_enabled(sock, other)) {
1924 UNIXCB(skb).pid = get_pid(task_tgid(current));
1925 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1926 }
1927}
1928
1929static bool unix_skb_scm_eq(struct sk_buff *skb,
1930 struct scm_cookie *scm)
1931{
1932 return UNIXCB(skb).pid == scm->pid &&
1933 uid_eq(UNIXCB(skb).uid, scm->creds.uid) &&
1934 gid_eq(UNIXCB(skb).gid, scm->creds.gid) &&
1935 unix_secdata_eq(scm, skb);
1936}
1937
1938static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1939{
1940 struct scm_fp_list *fp = UNIXCB(skb).fp;
1941 struct unix_sock *u = unix_sk(sk);
1942
1943 if (unlikely(fp && fp->count)) {
1944 atomic_add(fp->count, &u->scm_stat.nr_fds);
1945 unix_add_edges(fp, u);
1946 }
1947}
1948
1949static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1950{
1951 struct scm_fp_list *fp = UNIXCB(skb).fp;
1952 struct unix_sock *u = unix_sk(sk);
1953
1954 if (unlikely(fp && fp->count)) {
1955 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1956 unix_del_edges(fp);
1957 }
1958}
1959
1960/*
1961 * Send AF_UNIX data.
1962 */
1963
1964static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1965 size_t len)
1966{
1967 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1968 struct sock *sk = sock->sk, *other = NULL;
1969 struct unix_sock *u = unix_sk(sk);
1970 struct scm_cookie scm;
1971 struct sk_buff *skb;
1972 int data_len = 0;
1973 int sk_locked;
1974 long timeo;
1975 int err;
1976
1977 err = scm_send(sock, msg, &scm, false);
1978 if (err < 0)
1979 return err;
1980
1981 wait_for_unix_gc(scm.fp);
1982
1983 err = -EOPNOTSUPP;
1984 if (msg->msg_flags&MSG_OOB)
1985 goto out;
1986
1987 if (msg->msg_namelen) {
1988 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1989 if (err)
1990 goto out;
1991
1992 err = BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk,
1993 msg->msg_name,
1994 &msg->msg_namelen,
1995 NULL);
1996 if (err)
1997 goto out;
1998 } else {
1999 sunaddr = NULL;
2000 err = -ENOTCONN;
2001 other = unix_peer_get(sk);
2002 if (!other)
2003 goto out;
2004 }
2005
2006 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
2007 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
2008 !READ_ONCE(u->addr)) {
2009 err = unix_autobind(sk);
2010 if (err)
2011 goto out;
2012 }
2013
2014 err = -EMSGSIZE;
2015 if (len > READ_ONCE(sk->sk_sndbuf) - 32)
2016 goto out;
2017
2018 if (len > SKB_MAX_ALLOC) {
2019 data_len = min_t(size_t,
2020 len - SKB_MAX_ALLOC,
2021 MAX_SKB_FRAGS * PAGE_SIZE);
2022 data_len = PAGE_ALIGN(data_len);
2023
2024 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
2025 }
2026
2027 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
2028 msg->msg_flags & MSG_DONTWAIT, &err,
2029 PAGE_ALLOC_COSTLY_ORDER);
2030 if (skb == NULL)
2031 goto out;
2032
2033 err = unix_scm_to_skb(&scm, skb, true);
2034 if (err < 0)
2035 goto out_free;
2036
2037 skb_put(skb, len - data_len);
2038 skb->data_len = data_len;
2039 skb->len = len;
2040 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
2041 if (err)
2042 goto out_free;
2043
2044 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
2045
2046restart:
2047 if (!other) {
2048 err = -ECONNRESET;
2049 if (sunaddr == NULL)
2050 goto out_free;
2051
2052 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
2053 sk->sk_type);
2054 if (IS_ERR(other)) {
2055 err = PTR_ERR(other);
2056 other = NULL;
2057 goto out_free;
2058 }
2059 }
2060
2061 if (sk_filter(other, skb) < 0) {
2062 /* Toss the packet but do not return any error to the sender */
2063 err = len;
2064 goto out_free;
2065 }
2066
2067 sk_locked = 0;
2068 unix_state_lock(other);
2069restart_locked:
2070 err = -EPERM;
2071 if (!unix_may_send(sk, other))
2072 goto out_unlock;
2073
2074 if (unlikely(sock_flag(other, SOCK_DEAD))) {
2075 /*
2076 * Check with 1003.1g - what should
2077 * datagram error
2078 */
2079 unix_state_unlock(other);
2080 sock_put(other);
2081
2082 if (!sk_locked)
2083 unix_state_lock(sk);
2084
2085 err = 0;
2086 if (sk->sk_type == SOCK_SEQPACKET) {
2087 /* We are here only when racing with unix_release_sock()
2088 * is clearing @other. Never change state to TCP_CLOSE
2089 * unlike SOCK_DGRAM wants.
2090 */
2091 unix_state_unlock(sk);
2092 err = -EPIPE;
2093 } else if (unix_peer(sk) == other) {
2094 unix_peer(sk) = NULL;
2095 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2096
2097 WRITE_ONCE(sk->sk_state, TCP_CLOSE);
2098 unix_state_unlock(sk);
2099
2100 unix_dgram_disconnected(sk, other);
2101 sock_put(other);
2102 err = -ECONNREFUSED;
2103 } else {
2104 unix_state_unlock(sk);
2105 }
2106
2107 other = NULL;
2108 if (err)
2109 goto out_free;
2110 goto restart;
2111 }
2112
2113 err = -EPIPE;
2114 if (other->sk_shutdown & RCV_SHUTDOWN)
2115 goto out_unlock;
2116
2117 if (sk->sk_type != SOCK_SEQPACKET) {
2118 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
2119 if (err)
2120 goto out_unlock;
2121 }
2122
2123 /* other == sk && unix_peer(other) != sk if
2124 * - unix_peer(sk) == NULL, destination address bound to sk
2125 * - unix_peer(sk) == sk by time of get but disconnected before lock
2126 */
2127 if (other != sk &&
2128 unlikely(unix_peer(other) != sk &&
2129 unix_recvq_full_lockless(other))) {
2130 if (timeo) {
2131 timeo = unix_wait_for_peer(other, timeo);
2132
2133 err = sock_intr_errno(timeo);
2134 if (signal_pending(current))
2135 goto out_free;
2136
2137 goto restart;
2138 }
2139
2140 if (!sk_locked) {
2141 unix_state_unlock(other);
2142 unix_state_double_lock(sk, other);
2143 }
2144
2145 if (unix_peer(sk) != other ||
2146 unix_dgram_peer_wake_me(sk, other)) {
2147 err = -EAGAIN;
2148 sk_locked = 1;
2149 goto out_unlock;
2150 }
2151
2152 if (!sk_locked) {
2153 sk_locked = 1;
2154 goto restart_locked;
2155 }
2156 }
2157
2158 if (unlikely(sk_locked))
2159 unix_state_unlock(sk);
2160
2161 if (sock_flag(other, SOCK_RCVTSTAMP))
2162 __net_timestamp(skb);
2163 maybe_add_creds(skb, sock, other);
2164 scm_stat_add(other, skb);
2165 skb_queue_tail(&other->sk_receive_queue, skb);
2166 unix_state_unlock(other);
2167 other->sk_data_ready(other);
2168 sock_put(other);
2169 scm_destroy(&scm);
2170 return len;
2171
2172out_unlock:
2173 if (sk_locked)
2174 unix_state_unlock(sk);
2175 unix_state_unlock(other);
2176out_free:
2177 kfree_skb(skb);
2178out:
2179 if (other)
2180 sock_put(other);
2181 scm_destroy(&scm);
2182 return err;
2183}
2184
2185/* We use paged skbs for stream sockets, and limit occupancy to 32768
2186 * bytes, and a minimum of a full page.
2187 */
2188#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
2189
2190#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2191static int queue_oob(struct socket *sock, struct msghdr *msg, struct sock *other,
2192 struct scm_cookie *scm, bool fds_sent)
2193{
2194 struct unix_sock *ousk = unix_sk(other);
2195 struct sk_buff *skb;
2196 int err = 0;
2197
2198 skb = sock_alloc_send_skb(sock->sk, 1, msg->msg_flags & MSG_DONTWAIT, &err);
2199
2200 if (!skb)
2201 return err;
2202
2203 err = unix_scm_to_skb(scm, skb, !fds_sent);
2204 if (err < 0) {
2205 kfree_skb(skb);
2206 return err;
2207 }
2208 skb_put(skb, 1);
2209 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
2210
2211 if (err) {
2212 kfree_skb(skb);
2213 return err;
2214 }
2215
2216 unix_state_lock(other);
2217
2218 if (sock_flag(other, SOCK_DEAD) ||
2219 (other->sk_shutdown & RCV_SHUTDOWN)) {
2220 unix_state_unlock(other);
2221 kfree_skb(skb);
2222 return -EPIPE;
2223 }
2224
2225 maybe_add_creds(skb, sock, other);
2226 scm_stat_add(other, skb);
2227
2228 spin_lock(&other->sk_receive_queue.lock);
2229 WRITE_ONCE(ousk->oob_skb, skb);
2230 __skb_queue_tail(&other->sk_receive_queue, skb);
2231 spin_unlock(&other->sk_receive_queue.lock);
2232
2233 sk_send_sigurg(other);
2234 unix_state_unlock(other);
2235 other->sk_data_ready(other);
2236
2237 return err;
2238}
2239#endif
2240
2241static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2242 size_t len)
2243{
2244 struct sock *sk = sock->sk;
2245 struct sock *other = NULL;
2246 int err, size;
2247 struct sk_buff *skb;
2248 int sent = 0;
2249 struct scm_cookie scm;
2250 bool fds_sent = false;
2251 int data_len;
2252
2253 err = scm_send(sock, msg, &scm, false);
2254 if (err < 0)
2255 return err;
2256
2257 wait_for_unix_gc(scm.fp);
2258
2259 err = -EOPNOTSUPP;
2260 if (msg->msg_flags & MSG_OOB) {
2261#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2262 if (len)
2263 len--;
2264 else
2265#endif
2266 goto out_err;
2267 }
2268
2269 if (msg->msg_namelen) {
2270 err = READ_ONCE(sk->sk_state) == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2271 goto out_err;
2272 } else {
2273 err = -ENOTCONN;
2274 other = unix_peer(sk);
2275 if (!other)
2276 goto out_err;
2277 }
2278
2279 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2280 goto pipe_err;
2281
2282 while (sent < len) {
2283 size = len - sent;
2284
2285 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2286 skb = sock_alloc_send_pskb(sk, 0, 0,
2287 msg->msg_flags & MSG_DONTWAIT,
2288 &err, 0);
2289 } else {
2290 /* Keep two messages in the pipe so it schedules better */
2291 size = min_t(int, size, (READ_ONCE(sk->sk_sndbuf) >> 1) - 64);
2292
2293 /* allow fallback to order-0 allocations */
2294 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
2295
2296 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
2297
2298 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2299
2300 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
2301 msg->msg_flags & MSG_DONTWAIT, &err,
2302 get_order(UNIX_SKB_FRAGS_SZ));
2303 }
2304 if (!skb)
2305 goto out_err;
2306
2307 /* Only send the fds in the first buffer */
2308 err = unix_scm_to_skb(&scm, skb, !fds_sent);
2309 if (err < 0) {
2310 kfree_skb(skb);
2311 goto out_err;
2312 }
2313 fds_sent = true;
2314
2315 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2316 skb->ip_summed = CHECKSUM_UNNECESSARY;
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 (READ_ONCE(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 (READ_ONCE(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
2638 spin_unlock(&sk->sk_receive_queue.lock);
2639 unix_state_unlock(sk);
2640
2641 chunk = state->recv_actor(oob_skb, 0, chunk, state);
2642
2643 if (!(state->flags & MSG_PEEK))
2644 UNIXCB(oob_skb).consumed += 1;
2645
2646 mutex_unlock(&u->iolock);
2647
2648 if (chunk < 0)
2649 return -EFAULT;
2650
2651 state->msg->msg_flags |= MSG_OOB;
2652 return 1;
2653}
2654
2655static struct sk_buff *manage_oob(struct sk_buff *skb, struct sock *sk,
2656 int flags, int copied)
2657{
2658 struct sk_buff *read_skb = NULL, *unread_skb = NULL;
2659 struct unix_sock *u = unix_sk(sk);
2660
2661 if (likely(unix_skb_len(skb) && skb != READ_ONCE(u->oob_skb)))
2662 return skb;
2663
2664 spin_lock(&sk->sk_receive_queue.lock);
2665
2666 if (!unix_skb_len(skb)) {
2667 if (copied && (!u->oob_skb || skb == u->oob_skb)) {
2668 skb = NULL;
2669 } else if (flags & MSG_PEEK) {
2670 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2671 } else {
2672 read_skb = skb;
2673 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2674 __skb_unlink(read_skb, &sk->sk_receive_queue);
2675 }
2676
2677 if (!skb)
2678 goto unlock;
2679 }
2680
2681 if (skb != u->oob_skb)
2682 goto unlock;
2683
2684 if (copied) {
2685 skb = NULL;
2686 } else if (!(flags & MSG_PEEK)) {
2687 WRITE_ONCE(u->oob_skb, NULL);
2688
2689 if (!sock_flag(sk, SOCK_URGINLINE)) {
2690 __skb_unlink(skb, &sk->sk_receive_queue);
2691 unread_skb = skb;
2692 skb = skb_peek(&sk->sk_receive_queue);
2693 }
2694 } else if (!sock_flag(sk, SOCK_URGINLINE)) {
2695 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2696 }
2697
2698unlock:
2699 spin_unlock(&sk->sk_receive_queue.lock);
2700
2701 consume_skb(read_skb);
2702 kfree_skb(unread_skb);
2703
2704 return skb;
2705}
2706#endif
2707
2708static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2709{
2710 struct unix_sock *u = unix_sk(sk);
2711 struct sk_buff *skb;
2712 int err;
2713
2714 if (unlikely(READ_ONCE(sk->sk_state) != TCP_ESTABLISHED))
2715 return -ENOTCONN;
2716
2717 mutex_lock(&u->iolock);
2718 skb = skb_recv_datagram(sk, MSG_DONTWAIT, &err);
2719 mutex_unlock(&u->iolock);
2720 if (!skb)
2721 return err;
2722
2723#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2724 if (unlikely(skb == READ_ONCE(u->oob_skb))) {
2725 bool drop = false;
2726
2727 unix_state_lock(sk);
2728
2729 if (sock_flag(sk, SOCK_DEAD)) {
2730 unix_state_unlock(sk);
2731 kfree_skb(skb);
2732 return -ECONNRESET;
2733 }
2734
2735 spin_lock(&sk->sk_receive_queue.lock);
2736 if (likely(skb == u->oob_skb)) {
2737 WRITE_ONCE(u->oob_skb, NULL);
2738 drop = true;
2739 }
2740 spin_unlock(&sk->sk_receive_queue.lock);
2741
2742 unix_state_unlock(sk);
2743
2744 if (drop) {
2745 kfree_skb(skb);
2746 return -EAGAIN;
2747 }
2748 }
2749#endif
2750
2751 return recv_actor(sk, skb);
2752}
2753
2754static int unix_stream_read_generic(struct unix_stream_read_state *state,
2755 bool freezable)
2756{
2757 struct scm_cookie scm;
2758 struct socket *sock = state->socket;
2759 struct sock *sk = sock->sk;
2760 struct unix_sock *u = unix_sk(sk);
2761 int copied = 0;
2762 int flags = state->flags;
2763 int noblock = flags & MSG_DONTWAIT;
2764 bool check_creds = false;
2765 int target;
2766 int err = 0;
2767 long timeo;
2768 int skip;
2769 size_t size = state->size;
2770 unsigned int last_len;
2771
2772 if (unlikely(READ_ONCE(sk->sk_state) != TCP_ESTABLISHED)) {
2773 err = -EINVAL;
2774 goto out;
2775 }
2776
2777 if (unlikely(flags & MSG_OOB)) {
2778 err = -EOPNOTSUPP;
2779#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2780 err = unix_stream_recv_urg(state);
2781#endif
2782 goto out;
2783 }
2784
2785 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2786 timeo = sock_rcvtimeo(sk, noblock);
2787
2788 memset(&scm, 0, sizeof(scm));
2789
2790 /* Lock the socket to prevent queue disordering
2791 * while sleeps in memcpy_tomsg
2792 */
2793 mutex_lock(&u->iolock);
2794
2795 skip = max(sk_peek_offset(sk, flags), 0);
2796
2797 do {
2798 struct sk_buff *skb, *last;
2799 int chunk;
2800
2801redo:
2802 unix_state_lock(sk);
2803 if (sock_flag(sk, SOCK_DEAD)) {
2804 err = -ECONNRESET;
2805 goto unlock;
2806 }
2807 last = skb = skb_peek(&sk->sk_receive_queue);
2808 last_len = last ? last->len : 0;
2809
2810again:
2811#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2812 if (skb) {
2813 skb = manage_oob(skb, sk, flags, copied);
2814 if (!skb && copied) {
2815 unix_state_unlock(sk);
2816 break;
2817 }
2818 }
2819#endif
2820 if (skb == NULL) {
2821 if (copied >= target)
2822 goto unlock;
2823
2824 /*
2825 * POSIX 1003.1g mandates this order.
2826 */
2827
2828 err = sock_error(sk);
2829 if (err)
2830 goto unlock;
2831 if (sk->sk_shutdown & RCV_SHUTDOWN)
2832 goto unlock;
2833
2834 unix_state_unlock(sk);
2835 if (!timeo) {
2836 err = -EAGAIN;
2837 break;
2838 }
2839
2840 mutex_unlock(&u->iolock);
2841
2842 timeo = unix_stream_data_wait(sk, timeo, last,
2843 last_len, freezable);
2844
2845 if (signal_pending(current)) {
2846 err = sock_intr_errno(timeo);
2847 scm_destroy(&scm);
2848 goto out;
2849 }
2850
2851 mutex_lock(&u->iolock);
2852 goto redo;
2853unlock:
2854 unix_state_unlock(sk);
2855 break;
2856 }
2857
2858 while (skip >= unix_skb_len(skb)) {
2859 skip -= unix_skb_len(skb);
2860 last = skb;
2861 last_len = skb->len;
2862 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2863 if (!skb)
2864 goto again;
2865 }
2866
2867 unix_state_unlock(sk);
2868
2869 if (check_creds) {
2870 /* Never glue messages from different writers */
2871 if (!unix_skb_scm_eq(skb, &scm))
2872 break;
2873 } else if (test_bit(SOCK_PASSCRED, &sock->flags) ||
2874 test_bit(SOCK_PASSPIDFD, &sock->flags)) {
2875 /* Copy credentials */
2876 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2877 unix_set_secdata(&scm, skb);
2878 check_creds = true;
2879 }
2880
2881 /* Copy address just once */
2882 if (state->msg && state->msg->msg_name) {
2883 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2884 state->msg->msg_name);
2885 unix_copy_addr(state->msg, skb->sk);
2886
2887 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2888 state->msg->msg_name,
2889 &state->msg->msg_namelen);
2890
2891 sunaddr = NULL;
2892 }
2893
2894 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2895 chunk = state->recv_actor(skb, skip, chunk, state);
2896 if (chunk < 0) {
2897 if (copied == 0)
2898 copied = -EFAULT;
2899 break;
2900 }
2901 copied += chunk;
2902 size -= chunk;
2903
2904 /* Mark read part of skb as used */
2905 if (!(flags & MSG_PEEK)) {
2906 UNIXCB(skb).consumed += chunk;
2907
2908 sk_peek_offset_bwd(sk, chunk);
2909
2910 if (UNIXCB(skb).fp) {
2911 scm_stat_del(sk, skb);
2912 unix_detach_fds(&scm, skb);
2913 }
2914
2915 if (unix_skb_len(skb))
2916 break;
2917
2918 skb_unlink(skb, &sk->sk_receive_queue);
2919 consume_skb(skb);
2920
2921 if (scm.fp)
2922 break;
2923 } else {
2924 /* It is questionable, see note in unix_dgram_recvmsg.
2925 */
2926 if (UNIXCB(skb).fp)
2927 unix_peek_fds(&scm, skb);
2928
2929 sk_peek_offset_fwd(sk, chunk);
2930
2931 if (UNIXCB(skb).fp)
2932 break;
2933
2934 skip = 0;
2935 last = skb;
2936 last_len = skb->len;
2937 unix_state_lock(sk);
2938 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2939 if (skb)
2940 goto again;
2941 unix_state_unlock(sk);
2942 break;
2943 }
2944 } while (size);
2945
2946 mutex_unlock(&u->iolock);
2947 if (state->msg)
2948 scm_recv_unix(sock, state->msg, &scm, flags);
2949 else
2950 scm_destroy(&scm);
2951out:
2952 return copied ? : err;
2953}
2954
2955static int unix_stream_read_actor(struct sk_buff *skb,
2956 int skip, int chunk,
2957 struct unix_stream_read_state *state)
2958{
2959 int ret;
2960
2961 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2962 state->msg, chunk);
2963 return ret ?: chunk;
2964}
2965
2966int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2967 size_t size, int flags)
2968{
2969 struct unix_stream_read_state state = {
2970 .recv_actor = unix_stream_read_actor,
2971 .socket = sk->sk_socket,
2972 .msg = msg,
2973 .size = size,
2974 .flags = flags
2975 };
2976
2977 return unix_stream_read_generic(&state, true);
2978}
2979
2980static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2981 size_t size, int flags)
2982{
2983 struct unix_stream_read_state state = {
2984 .recv_actor = unix_stream_read_actor,
2985 .socket = sock,
2986 .msg = msg,
2987 .size = size,
2988 .flags = flags
2989 };
2990
2991#ifdef CONFIG_BPF_SYSCALL
2992 struct sock *sk = sock->sk;
2993 const struct proto *prot = READ_ONCE(sk->sk_prot);
2994
2995 if (prot != &unix_stream_proto)
2996 return prot->recvmsg(sk, msg, size, flags, NULL);
2997#endif
2998 return unix_stream_read_generic(&state, true);
2999}
3000
3001static int unix_stream_splice_actor(struct sk_buff *skb,
3002 int skip, int chunk,
3003 struct unix_stream_read_state *state)
3004{
3005 return skb_splice_bits(skb, state->socket->sk,
3006 UNIXCB(skb).consumed + skip,
3007 state->pipe, chunk, state->splice_flags);
3008}
3009
3010static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
3011 struct pipe_inode_info *pipe,
3012 size_t size, unsigned int flags)
3013{
3014 struct unix_stream_read_state state = {
3015 .recv_actor = unix_stream_splice_actor,
3016 .socket = sock,
3017 .pipe = pipe,
3018 .size = size,
3019 .splice_flags = flags,
3020 };
3021
3022 if (unlikely(*ppos))
3023 return -ESPIPE;
3024
3025 if (sock->file->f_flags & O_NONBLOCK ||
3026 flags & SPLICE_F_NONBLOCK)
3027 state.flags = MSG_DONTWAIT;
3028
3029 return unix_stream_read_generic(&state, false);
3030}
3031
3032static int unix_shutdown(struct socket *sock, int mode)
3033{
3034 struct sock *sk = sock->sk;
3035 struct sock *other;
3036
3037 if (mode < SHUT_RD || mode > SHUT_RDWR)
3038 return -EINVAL;
3039 /* This maps:
3040 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
3041 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
3042 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
3043 */
3044 ++mode;
3045
3046 unix_state_lock(sk);
3047 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | mode);
3048 other = unix_peer(sk);
3049 if (other)
3050 sock_hold(other);
3051 unix_state_unlock(sk);
3052 sk->sk_state_change(sk);
3053
3054 if (other &&
3055 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
3056
3057 int peer_mode = 0;
3058 const struct proto *prot = READ_ONCE(other->sk_prot);
3059
3060 if (prot->unhash)
3061 prot->unhash(other);
3062 if (mode&RCV_SHUTDOWN)
3063 peer_mode |= SEND_SHUTDOWN;
3064 if (mode&SEND_SHUTDOWN)
3065 peer_mode |= RCV_SHUTDOWN;
3066 unix_state_lock(other);
3067 WRITE_ONCE(other->sk_shutdown, other->sk_shutdown | peer_mode);
3068 unix_state_unlock(other);
3069 other->sk_state_change(other);
3070 if (peer_mode == SHUTDOWN_MASK)
3071 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
3072 else if (peer_mode & RCV_SHUTDOWN)
3073 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
3074 }
3075 if (other)
3076 sock_put(other);
3077
3078 return 0;
3079}
3080
3081long unix_inq_len(struct sock *sk)
3082{
3083 struct sk_buff *skb;
3084 long amount = 0;
3085
3086 if (READ_ONCE(sk->sk_state) == TCP_LISTEN)
3087 return -EINVAL;
3088
3089 spin_lock(&sk->sk_receive_queue.lock);
3090 if (sk->sk_type == SOCK_STREAM ||
3091 sk->sk_type == SOCK_SEQPACKET) {
3092 skb_queue_walk(&sk->sk_receive_queue, skb)
3093 amount += unix_skb_len(skb);
3094 } else {
3095 skb = skb_peek(&sk->sk_receive_queue);
3096 if (skb)
3097 amount = skb->len;
3098 }
3099 spin_unlock(&sk->sk_receive_queue.lock);
3100
3101 return amount;
3102}
3103EXPORT_SYMBOL_GPL(unix_inq_len);
3104
3105long unix_outq_len(struct sock *sk)
3106{
3107 return sk_wmem_alloc_get(sk);
3108}
3109EXPORT_SYMBOL_GPL(unix_outq_len);
3110
3111static int unix_open_file(struct sock *sk)
3112{
3113 struct path path;
3114 struct file *f;
3115 int fd;
3116
3117 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3118 return -EPERM;
3119
3120 if (!smp_load_acquire(&unix_sk(sk)->addr))
3121 return -ENOENT;
3122
3123 path = unix_sk(sk)->path;
3124 if (!path.dentry)
3125 return -ENOENT;
3126
3127 path_get(&path);
3128
3129 fd = get_unused_fd_flags(O_CLOEXEC);
3130 if (fd < 0)
3131 goto out;
3132
3133 f = dentry_open(&path, O_PATH, current_cred());
3134 if (IS_ERR(f)) {
3135 put_unused_fd(fd);
3136 fd = PTR_ERR(f);
3137 goto out;
3138 }
3139
3140 fd_install(fd, f);
3141out:
3142 path_put(&path);
3143
3144 return fd;
3145}
3146
3147static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3148{
3149 struct sock *sk = sock->sk;
3150 long amount = 0;
3151 int err;
3152
3153 switch (cmd) {
3154 case SIOCOUTQ:
3155 amount = unix_outq_len(sk);
3156 err = put_user(amount, (int __user *)arg);
3157 break;
3158 case SIOCINQ:
3159 amount = unix_inq_len(sk);
3160 if (amount < 0)
3161 err = amount;
3162 else
3163 err = put_user(amount, (int __user *)arg);
3164 break;
3165 case SIOCUNIXFILE:
3166 err = unix_open_file(sk);
3167 break;
3168#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3169 case SIOCATMARK:
3170 {
3171 struct unix_sock *u = unix_sk(sk);
3172 struct sk_buff *skb;
3173 int answ = 0;
3174
3175 mutex_lock(&u->iolock);
3176
3177 skb = skb_peek(&sk->sk_receive_queue);
3178 if (skb) {
3179 struct sk_buff *oob_skb = READ_ONCE(u->oob_skb);
3180 struct sk_buff *next_skb;
3181
3182 next_skb = skb_peek_next(skb, &sk->sk_receive_queue);
3183
3184 if (skb == oob_skb ||
3185 (!unix_skb_len(skb) &&
3186 (!oob_skb || next_skb == oob_skb)))
3187 answ = 1;
3188 }
3189
3190 mutex_unlock(&u->iolock);
3191
3192 err = put_user(answ, (int __user *)arg);
3193 }
3194 break;
3195#endif
3196 default:
3197 err = -ENOIOCTLCMD;
3198 break;
3199 }
3200 return err;
3201}
3202
3203#ifdef CONFIG_COMPAT
3204static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3205{
3206 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
3207}
3208#endif
3209
3210static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
3211{
3212 struct sock *sk = sock->sk;
3213 unsigned char state;
3214 __poll_t mask;
3215 u8 shutdown;
3216
3217 sock_poll_wait(file, sock, wait);
3218 mask = 0;
3219 shutdown = READ_ONCE(sk->sk_shutdown);
3220 state = READ_ONCE(sk->sk_state);
3221
3222 /* exceptional events? */
3223 if (READ_ONCE(sk->sk_err))
3224 mask |= EPOLLERR;
3225 if (shutdown == SHUTDOWN_MASK)
3226 mask |= EPOLLHUP;
3227 if (shutdown & RCV_SHUTDOWN)
3228 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3229
3230 /* readable? */
3231 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3232 mask |= EPOLLIN | EPOLLRDNORM;
3233 if (sk_is_readable(sk))
3234 mask |= EPOLLIN | EPOLLRDNORM;
3235#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3236 if (READ_ONCE(unix_sk(sk)->oob_skb))
3237 mask |= EPOLLPRI;
3238#endif
3239
3240 /* Connection-based need to check for termination and startup */
3241 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3242 state == TCP_CLOSE)
3243 mask |= EPOLLHUP;
3244
3245 /*
3246 * we set writable also when the other side has shut down the
3247 * connection. This prevents stuck sockets.
3248 */
3249 if (unix_writable(sk, state))
3250 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3251
3252 return mask;
3253}
3254
3255static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3256 poll_table *wait)
3257{
3258 struct sock *sk = sock->sk, *other;
3259 unsigned int writable;
3260 unsigned char state;
3261 __poll_t mask;
3262 u8 shutdown;
3263
3264 sock_poll_wait(file, sock, wait);
3265 mask = 0;
3266 shutdown = READ_ONCE(sk->sk_shutdown);
3267 state = READ_ONCE(sk->sk_state);
3268
3269 /* exceptional events? */
3270 if (READ_ONCE(sk->sk_err) ||
3271 !skb_queue_empty_lockless(&sk->sk_error_queue))
3272 mask |= EPOLLERR |
3273 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
3274
3275 if (shutdown & RCV_SHUTDOWN)
3276 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3277 if (shutdown == SHUTDOWN_MASK)
3278 mask |= EPOLLHUP;
3279
3280 /* readable? */
3281 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3282 mask |= EPOLLIN | EPOLLRDNORM;
3283 if (sk_is_readable(sk))
3284 mask |= EPOLLIN | EPOLLRDNORM;
3285
3286 /* Connection-based need to check for termination and startup */
3287 if (sk->sk_type == SOCK_SEQPACKET && state == TCP_CLOSE)
3288 mask |= EPOLLHUP;
3289
3290 /* No write status requested, avoid expensive OUT tests. */
3291 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
3292 return mask;
3293
3294 writable = unix_writable(sk, state);
3295 if (writable) {
3296 unix_state_lock(sk);
3297
3298 other = unix_peer(sk);
3299 if (other && unix_peer(other) != sk &&
3300 unix_recvq_full_lockless(other) &&
3301 unix_dgram_peer_wake_me(sk, other))
3302 writable = 0;
3303
3304 unix_state_unlock(sk);
3305 }
3306
3307 if (writable)
3308 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3309 else
3310 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3311
3312 return mask;
3313}
3314
3315#ifdef CONFIG_PROC_FS
3316
3317#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3318
3319#define get_bucket(x) ((x) >> BUCKET_SPACE)
3320#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
3321#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
3322
3323static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
3324{
3325 unsigned long offset = get_offset(*pos);
3326 unsigned long bucket = get_bucket(*pos);
3327 unsigned long count = 0;
3328 struct sock *sk;
3329
3330 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3331 sk; sk = sk_next(sk)) {
3332 if (++count == offset)
3333 break;
3334 }
3335
3336 return sk;
3337}
3338
3339static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
3340{
3341 unsigned long bucket = get_bucket(*pos);
3342 struct net *net = seq_file_net(seq);
3343 struct sock *sk;
3344
3345 while (bucket < UNIX_HASH_SIZE) {
3346 spin_lock(&net->unx.table.locks[bucket]);
3347
3348 sk = unix_from_bucket(seq, pos);
3349 if (sk)
3350 return sk;
3351
3352 spin_unlock(&net->unx.table.locks[bucket]);
3353
3354 *pos = set_bucket_offset(++bucket, 1);
3355 }
3356
3357 return NULL;
3358}
3359
3360static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
3361 loff_t *pos)
3362{
3363 unsigned long bucket = get_bucket(*pos);
3364
3365 sk = sk_next(sk);
3366 if (sk)
3367 return sk;
3368
3369
3370 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
3371
3372 *pos = set_bucket_offset(++bucket, 1);
3373
3374 return unix_get_first(seq, pos);
3375}
3376
3377static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3378{
3379 if (!*pos)
3380 return SEQ_START_TOKEN;
3381
3382 return unix_get_first(seq, pos);
3383}
3384
3385static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3386{
3387 ++*pos;
3388
3389 if (v == SEQ_START_TOKEN)
3390 return unix_get_first(seq, pos);
3391
3392 return unix_get_next(seq, v, pos);
3393}
3394
3395static void unix_seq_stop(struct seq_file *seq, void *v)
3396{
3397 struct sock *sk = v;
3398
3399 if (sk)
3400 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
3401}
3402
3403static int unix_seq_show(struct seq_file *seq, void *v)
3404{
3405
3406 if (v == SEQ_START_TOKEN)
3407 seq_puts(seq, "Num RefCount Protocol Flags Type St "
3408 "Inode Path\n");
3409 else {
3410 struct sock *s = v;
3411 struct unix_sock *u = unix_sk(s);
3412 unix_state_lock(s);
3413
3414 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
3415 s,
3416 refcount_read(&s->sk_refcnt),
3417 0,
3418 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
3419 s->sk_type,
3420 s->sk_socket ?
3421 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
3422 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
3423 sock_i_ino(s));
3424
3425 if (u->addr) { // under a hash table lock here
3426 int i, len;
3427 seq_putc(seq, ' ');
3428
3429 i = 0;
3430 len = u->addr->len -
3431 offsetof(struct sockaddr_un, sun_path);
3432 if (u->addr->name->sun_path[0]) {
3433 len--;
3434 } else {
3435 seq_putc(seq, '@');
3436 i++;
3437 }
3438 for ( ; i < len; i++)
3439 seq_putc(seq, u->addr->name->sun_path[i] ?:
3440 '@');
3441 }
3442 unix_state_unlock(s);
3443 seq_putc(seq, '\n');
3444 }
3445
3446 return 0;
3447}
3448
3449static const struct seq_operations unix_seq_ops = {
3450 .start = unix_seq_start,
3451 .next = unix_seq_next,
3452 .stop = unix_seq_stop,
3453 .show = unix_seq_show,
3454};
3455
3456#ifdef CONFIG_BPF_SYSCALL
3457struct bpf_unix_iter_state {
3458 struct seq_net_private p;
3459 unsigned int cur_sk;
3460 unsigned int end_sk;
3461 unsigned int max_sk;
3462 struct sock **batch;
3463 bool st_bucket_done;
3464};
3465
3466struct bpf_iter__unix {
3467 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3468 __bpf_md_ptr(struct unix_sock *, unix_sk);
3469 uid_t uid __aligned(8);
3470};
3471
3472static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3473 struct unix_sock *unix_sk, uid_t uid)
3474{
3475 struct bpf_iter__unix ctx;
3476
3477 meta->seq_num--; /* skip SEQ_START_TOKEN */
3478 ctx.meta = meta;
3479 ctx.unix_sk = unix_sk;
3480 ctx.uid = uid;
3481 return bpf_iter_run_prog(prog, &ctx);
3482}
3483
3484static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
3485
3486{
3487 struct bpf_unix_iter_state *iter = seq->private;
3488 unsigned int expected = 1;
3489 struct sock *sk;
3490
3491 sock_hold(start_sk);
3492 iter->batch[iter->end_sk++] = start_sk;
3493
3494 for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
3495 if (iter->end_sk < iter->max_sk) {
3496 sock_hold(sk);
3497 iter->batch[iter->end_sk++] = sk;
3498 }
3499
3500 expected++;
3501 }
3502
3503 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
3504
3505 return expected;
3506}
3507
3508static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
3509{
3510 while (iter->cur_sk < iter->end_sk)
3511 sock_put(iter->batch[iter->cur_sk++]);
3512}
3513
3514static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
3515 unsigned int new_batch_sz)
3516{
3517 struct sock **new_batch;
3518
3519 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3520 GFP_USER | __GFP_NOWARN);
3521 if (!new_batch)
3522 return -ENOMEM;
3523
3524 bpf_iter_unix_put_batch(iter);
3525 kvfree(iter->batch);
3526 iter->batch = new_batch;
3527 iter->max_sk = new_batch_sz;
3528
3529 return 0;
3530}
3531
3532static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
3533 loff_t *pos)
3534{
3535 struct bpf_unix_iter_state *iter = seq->private;
3536 unsigned int expected;
3537 bool resized = false;
3538 struct sock *sk;
3539
3540 if (iter->st_bucket_done)
3541 *pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
3542
3543again:
3544 /* Get a new batch */
3545 iter->cur_sk = 0;
3546 iter->end_sk = 0;
3547
3548 sk = unix_get_first(seq, pos);
3549 if (!sk)
3550 return NULL; /* Done */
3551
3552 expected = bpf_iter_unix_hold_batch(seq, sk);
3553
3554 if (iter->end_sk == expected) {
3555 iter->st_bucket_done = true;
3556 return sk;
3557 }
3558
3559 if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
3560 resized = true;
3561 goto again;
3562 }
3563
3564 return sk;
3565}
3566
3567static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
3568{
3569 if (!*pos)
3570 return SEQ_START_TOKEN;
3571
3572 /* bpf iter does not support lseek, so it always
3573 * continue from where it was stop()-ped.
3574 */
3575 return bpf_iter_unix_batch(seq, pos);
3576}
3577
3578static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3579{
3580 struct bpf_unix_iter_state *iter = seq->private;
3581 struct sock *sk;
3582
3583 /* Whenever seq_next() is called, the iter->cur_sk is
3584 * done with seq_show(), so advance to the next sk in
3585 * the batch.
3586 */
3587 if (iter->cur_sk < iter->end_sk)
3588 sock_put(iter->batch[iter->cur_sk++]);
3589
3590 ++*pos;
3591
3592 if (iter->cur_sk < iter->end_sk)
3593 sk = iter->batch[iter->cur_sk];
3594 else
3595 sk = bpf_iter_unix_batch(seq, pos);
3596
3597 return sk;
3598}
3599
3600static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
3601{
3602 struct bpf_iter_meta meta;
3603 struct bpf_prog *prog;
3604 struct sock *sk = v;
3605 uid_t uid;
3606 bool slow;
3607 int ret;
3608
3609 if (v == SEQ_START_TOKEN)
3610 return 0;
3611
3612 slow = lock_sock_fast(sk);
3613
3614 if (unlikely(sk_unhashed(sk))) {
3615 ret = SEQ_SKIP;
3616 goto unlock;
3617 }
3618
3619 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3620 meta.seq = seq;
3621 prog = bpf_iter_get_info(&meta, false);
3622 ret = unix_prog_seq_show(prog, &meta, v, uid);
3623unlock:
3624 unlock_sock_fast(sk, slow);
3625 return ret;
3626}
3627
3628static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3629{
3630 struct bpf_unix_iter_state *iter = seq->private;
3631 struct bpf_iter_meta meta;
3632 struct bpf_prog *prog;
3633
3634 if (!v) {
3635 meta.seq = seq;
3636 prog = bpf_iter_get_info(&meta, true);
3637 if (prog)
3638 (void)unix_prog_seq_show(prog, &meta, v, 0);
3639 }
3640
3641 if (iter->cur_sk < iter->end_sk)
3642 bpf_iter_unix_put_batch(iter);
3643}
3644
3645static const struct seq_operations bpf_iter_unix_seq_ops = {
3646 .start = bpf_iter_unix_seq_start,
3647 .next = bpf_iter_unix_seq_next,
3648 .stop = bpf_iter_unix_seq_stop,
3649 .show = bpf_iter_unix_seq_show,
3650};
3651#endif
3652#endif
3653
3654static const struct net_proto_family unix_family_ops = {
3655 .family = PF_UNIX,
3656 .create = unix_create,
3657 .owner = THIS_MODULE,
3658};
3659
3660
3661static int __net_init unix_net_init(struct net *net)
3662{
3663 int i;
3664
3665 net->unx.sysctl_max_dgram_qlen = 10;
3666 if (unix_sysctl_register(net))
3667 goto out;
3668
3669#ifdef CONFIG_PROC_FS
3670 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
3671 sizeof(struct seq_net_private)))
3672 goto err_sysctl;
3673#endif
3674
3675 net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
3676 sizeof(spinlock_t), GFP_KERNEL);
3677 if (!net->unx.table.locks)
3678 goto err_proc;
3679
3680 net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
3681 sizeof(struct hlist_head),
3682 GFP_KERNEL);
3683 if (!net->unx.table.buckets)
3684 goto free_locks;
3685
3686 for (i = 0; i < UNIX_HASH_SIZE; i++) {
3687 spin_lock_init(&net->unx.table.locks[i]);
3688 lock_set_cmp_fn(&net->unx.table.locks[i], unix_table_lock_cmp_fn, NULL);
3689 INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
3690 }
3691
3692 return 0;
3693
3694free_locks:
3695 kvfree(net->unx.table.locks);
3696err_proc:
3697#ifdef CONFIG_PROC_FS
3698 remove_proc_entry("unix", net->proc_net);
3699err_sysctl:
3700#endif
3701 unix_sysctl_unregister(net);
3702out:
3703 return -ENOMEM;
3704}
3705
3706static void __net_exit unix_net_exit(struct net *net)
3707{
3708 kvfree(net->unx.table.buckets);
3709 kvfree(net->unx.table.locks);
3710 unix_sysctl_unregister(net);
3711 remove_proc_entry("unix", net->proc_net);
3712}
3713
3714static struct pernet_operations unix_net_ops = {
3715 .init = unix_net_init,
3716 .exit = unix_net_exit,
3717};
3718
3719#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3720DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
3721 struct unix_sock *unix_sk, uid_t uid)
3722
3723#define INIT_BATCH_SZ 16
3724
3725static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
3726{
3727 struct bpf_unix_iter_state *iter = priv_data;
3728 int err;
3729
3730 err = bpf_iter_init_seq_net(priv_data, aux);
3731 if (err)
3732 return err;
3733
3734 err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
3735 if (err) {
3736 bpf_iter_fini_seq_net(priv_data);
3737 return err;
3738 }
3739
3740 return 0;
3741}
3742
3743static void bpf_iter_fini_unix(void *priv_data)
3744{
3745 struct bpf_unix_iter_state *iter = priv_data;
3746
3747 bpf_iter_fini_seq_net(priv_data);
3748 kvfree(iter->batch);
3749}
3750
3751static const struct bpf_iter_seq_info unix_seq_info = {
3752 .seq_ops = &bpf_iter_unix_seq_ops,
3753 .init_seq_private = bpf_iter_init_unix,
3754 .fini_seq_private = bpf_iter_fini_unix,
3755 .seq_priv_size = sizeof(struct bpf_unix_iter_state),
3756};
3757
3758static const struct bpf_func_proto *
3759bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
3760 const struct bpf_prog *prog)
3761{
3762 switch (func_id) {
3763 case BPF_FUNC_setsockopt:
3764 return &bpf_sk_setsockopt_proto;
3765 case BPF_FUNC_getsockopt:
3766 return &bpf_sk_getsockopt_proto;
3767 default:
3768 return NULL;
3769 }
3770}
3771
3772static struct bpf_iter_reg unix_reg_info = {
3773 .target = "unix",
3774 .ctx_arg_info_size = 1,
3775 .ctx_arg_info = {
3776 { offsetof(struct bpf_iter__unix, unix_sk),
3777 PTR_TO_BTF_ID_OR_NULL },
3778 },
3779 .get_func_proto = bpf_iter_unix_get_func_proto,
3780 .seq_info = &unix_seq_info,
3781};
3782
3783static void __init bpf_iter_register(void)
3784{
3785 unix_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UNIX];
3786 if (bpf_iter_reg_target(&unix_reg_info))
3787 pr_warn("Warning: could not register bpf iterator unix\n");
3788}
3789#endif
3790
3791static int __init af_unix_init(void)
3792{
3793 int i, rc = -1;
3794
3795 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
3796
3797 for (i = 0; i < UNIX_HASH_SIZE / 2; i++) {
3798 spin_lock_init(&bsd_socket_locks[i]);
3799 INIT_HLIST_HEAD(&bsd_socket_buckets[i]);
3800 }
3801
3802 rc = proto_register(&unix_dgram_proto, 1);
3803 if (rc != 0) {
3804 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3805 goto out;
3806 }
3807
3808 rc = proto_register(&unix_stream_proto, 1);
3809 if (rc != 0) {
3810 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3811 proto_unregister(&unix_dgram_proto);
3812 goto out;
3813 }
3814
3815 sock_register(&unix_family_ops);
3816 register_pernet_subsys(&unix_net_ops);
3817 unix_bpf_build_proto();
3818
3819#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3820 bpf_iter_register();
3821#endif
3822
3823out:
3824 return rc;
3825}
3826
3827/* Later than subsys_initcall() because we depend on stuff initialised there */
3828fs_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 void __unix_set_addr(struct sock *sk, struct unix_address *addr,
266 unsigned hash)
267{
268 __unix_remove_socket(sk);
269 smp_store_release(&unix_sk(sk)->addr, addr);
270 __unix_insert_socket(&unix_socket_table[hash], sk);
271}
272
273static inline void unix_remove_socket(struct sock *sk)
274{
275 spin_lock(&unix_table_lock);
276 __unix_remove_socket(sk);
277 spin_unlock(&unix_table_lock);
278}
279
280static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
281{
282 spin_lock(&unix_table_lock);
283 __unix_insert_socket(list, sk);
284 spin_unlock(&unix_table_lock);
285}
286
287static struct sock *__unix_find_socket_byname(struct net *net,
288 struct sockaddr_un *sunname,
289 int len, unsigned int hash)
290{
291 struct sock *s;
292
293 sk_for_each(s, &unix_socket_table[hash]) {
294 struct unix_sock *u = unix_sk(s);
295
296 if (!net_eq(sock_net(s), net))
297 continue;
298
299 if (u->addr->len == len &&
300 !memcmp(u->addr->name, sunname, len))
301 return s;
302 }
303 return NULL;
304}
305
306static inline struct sock *unix_find_socket_byname(struct net *net,
307 struct sockaddr_un *sunname,
308 int len, unsigned int hash)
309{
310 struct sock *s;
311
312 spin_lock(&unix_table_lock);
313 s = __unix_find_socket_byname(net, sunname, len, hash);
314 if (s)
315 sock_hold(s);
316 spin_unlock(&unix_table_lock);
317 return s;
318}
319
320static struct sock *unix_find_socket_byinode(struct inode *i)
321{
322 struct sock *s;
323
324 spin_lock(&unix_table_lock);
325 sk_for_each(s,
326 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
327 struct dentry *dentry = unix_sk(s)->path.dentry;
328
329 if (dentry && d_backing_inode(dentry) == i) {
330 sock_hold(s);
331 goto found;
332 }
333 }
334 s = NULL;
335found:
336 spin_unlock(&unix_table_lock);
337 return s;
338}
339
340/* Support code for asymmetrically connected dgram sockets
341 *
342 * If a datagram socket is connected to a socket not itself connected
343 * to the first socket (eg, /dev/log), clients may only enqueue more
344 * messages if the present receive queue of the server socket is not
345 * "too large". This means there's a second writeability condition
346 * poll and sendmsg need to test. The dgram recv code will do a wake
347 * up on the peer_wait wait queue of a socket upon reception of a
348 * datagram which needs to be propagated to sleeping would-be writers
349 * since these might not have sent anything so far. This can't be
350 * accomplished via poll_wait because the lifetime of the server
351 * socket might be less than that of its clients if these break their
352 * association with it or if the server socket is closed while clients
353 * are still connected to it and there's no way to inform "a polling
354 * implementation" that it should let go of a certain wait queue
355 *
356 * In order to propagate a wake up, a wait_queue_entry_t of the client
357 * socket is enqueued on the peer_wait queue of the server socket
358 * whose wake function does a wake_up on the ordinary client socket
359 * wait queue. This connection is established whenever a write (or
360 * poll for write) hit the flow control condition and broken when the
361 * association to the server socket is dissolved or after a wake up
362 * was relayed.
363 */
364
365static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
366 void *key)
367{
368 struct unix_sock *u;
369 wait_queue_head_t *u_sleep;
370
371 u = container_of(q, struct unix_sock, peer_wake);
372
373 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
374 q);
375 u->peer_wake.private = NULL;
376
377 /* relaying can only happen while the wq still exists */
378 u_sleep = sk_sleep(&u->sk);
379 if (u_sleep)
380 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
381
382 return 0;
383}
384
385static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
386{
387 struct unix_sock *u, *u_other;
388 int rc;
389
390 u = unix_sk(sk);
391 u_other = unix_sk(other);
392 rc = 0;
393 spin_lock(&u_other->peer_wait.lock);
394
395 if (!u->peer_wake.private) {
396 u->peer_wake.private = other;
397 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
398
399 rc = 1;
400 }
401
402 spin_unlock(&u_other->peer_wait.lock);
403 return rc;
404}
405
406static void unix_dgram_peer_wake_disconnect(struct sock *sk,
407 struct sock *other)
408{
409 struct unix_sock *u, *u_other;
410
411 u = unix_sk(sk);
412 u_other = unix_sk(other);
413 spin_lock(&u_other->peer_wait.lock);
414
415 if (u->peer_wake.private == other) {
416 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
417 u->peer_wake.private = NULL;
418 }
419
420 spin_unlock(&u_other->peer_wait.lock);
421}
422
423static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
424 struct sock *other)
425{
426 unix_dgram_peer_wake_disconnect(sk, other);
427 wake_up_interruptible_poll(sk_sleep(sk),
428 EPOLLOUT |
429 EPOLLWRNORM |
430 EPOLLWRBAND);
431}
432
433/* preconditions:
434 * - unix_peer(sk) == other
435 * - association is stable
436 */
437static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
438{
439 int connected;
440
441 connected = unix_dgram_peer_wake_connect(sk, other);
442
443 /* If other is SOCK_DEAD, we want to make sure we signal
444 * POLLOUT, such that a subsequent write() can get a
445 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
446 * to other and its full, we will hang waiting for POLLOUT.
447 */
448 if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
449 return 1;
450
451 if (connected)
452 unix_dgram_peer_wake_disconnect(sk, other);
453
454 return 0;
455}
456
457static int unix_writable(const struct sock *sk)
458{
459 return sk->sk_state != TCP_LISTEN &&
460 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
461}
462
463static void unix_write_space(struct sock *sk)
464{
465 struct socket_wq *wq;
466
467 rcu_read_lock();
468 if (unix_writable(sk)) {
469 wq = rcu_dereference(sk->sk_wq);
470 if (skwq_has_sleeper(wq))
471 wake_up_interruptible_sync_poll(&wq->wait,
472 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
473 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
474 }
475 rcu_read_unlock();
476}
477
478/* When dgram socket disconnects (or changes its peer), we clear its receive
479 * queue of packets arrived from previous peer. First, it allows to do
480 * flow control based only on wmem_alloc; second, sk connected to peer
481 * may receive messages only from that peer. */
482static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
483{
484 if (!skb_queue_empty(&sk->sk_receive_queue)) {
485 skb_queue_purge(&sk->sk_receive_queue);
486 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
487
488 /* If one link of bidirectional dgram pipe is disconnected,
489 * we signal error. Messages are lost. Do not make this,
490 * when peer was not connected to us.
491 */
492 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
493 other->sk_err = ECONNRESET;
494 sk_error_report(other);
495 }
496 }
497}
498
499static void unix_sock_destructor(struct sock *sk)
500{
501 struct unix_sock *u = unix_sk(sk);
502
503 skb_queue_purge(&sk->sk_receive_queue);
504
505 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
506 WARN_ON(!sk_unhashed(sk));
507 WARN_ON(sk->sk_socket);
508 if (!sock_flag(sk, SOCK_DEAD)) {
509 pr_info("Attempt to release alive unix socket: %p\n", sk);
510 return;
511 }
512
513 if (u->addr)
514 unix_release_addr(u->addr);
515
516 atomic_long_dec(&unix_nr_socks);
517 local_bh_disable();
518 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
519 local_bh_enable();
520#ifdef UNIX_REFCNT_DEBUG
521 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
522 atomic_long_read(&unix_nr_socks));
523#endif
524}
525
526static void unix_release_sock(struct sock *sk, int embrion)
527{
528 struct unix_sock *u = unix_sk(sk);
529 struct path path;
530 struct sock *skpair;
531 struct sk_buff *skb;
532 int state;
533
534 unix_remove_socket(sk);
535
536 /* Clear state */
537 unix_state_lock(sk);
538 sock_orphan(sk);
539 sk->sk_shutdown = SHUTDOWN_MASK;
540 path = u->path;
541 u->path.dentry = NULL;
542 u->path.mnt = NULL;
543 state = sk->sk_state;
544 sk->sk_state = TCP_CLOSE;
545
546 skpair = unix_peer(sk);
547 unix_peer(sk) = NULL;
548
549 unix_state_unlock(sk);
550
551 wake_up_interruptible_all(&u->peer_wait);
552
553 if (skpair != NULL) {
554 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
555 unix_state_lock(skpair);
556 /* No more writes */
557 skpair->sk_shutdown = SHUTDOWN_MASK;
558 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
559 skpair->sk_err = ECONNRESET;
560 unix_state_unlock(skpair);
561 skpair->sk_state_change(skpair);
562 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
563 }
564
565 unix_dgram_peer_wake_disconnect(sk, skpair);
566 sock_put(skpair); /* It may now die */
567 }
568
569 /* Try to flush out this socket. Throw out buffers at least */
570
571 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
572 if (state == TCP_LISTEN)
573 unix_release_sock(skb->sk, 1);
574 /* passed fds are erased in the kfree_skb hook */
575 UNIXCB(skb).consumed = skb->len;
576 kfree_skb(skb);
577 }
578
579 if (path.dentry)
580 path_put(&path);
581
582 sock_put(sk);
583
584 /* ---- Socket is dead now and most probably destroyed ---- */
585
586 /*
587 * Fixme: BSD difference: In BSD all sockets connected to us get
588 * ECONNRESET and we die on the spot. In Linux we behave
589 * like files and pipes do and wait for the last
590 * dereference.
591 *
592 * Can't we simply set sock->err?
593 *
594 * What the above comment does talk about? --ANK(980817)
595 */
596
597 if (unix_tot_inflight)
598 unix_gc(); /* Garbage collect fds */
599}
600
601static void init_peercred(struct sock *sk)
602{
603 const struct cred *old_cred;
604 struct pid *old_pid;
605
606 spin_lock(&sk->sk_peer_lock);
607 old_pid = sk->sk_peer_pid;
608 old_cred = sk->sk_peer_cred;
609 sk->sk_peer_pid = get_pid(task_tgid(current));
610 sk->sk_peer_cred = get_current_cred();
611 spin_unlock(&sk->sk_peer_lock);
612
613 put_pid(old_pid);
614 put_cred(old_cred);
615}
616
617static void copy_peercred(struct sock *sk, struct sock *peersk)
618{
619 const struct cred *old_cred;
620 struct pid *old_pid;
621
622 if (sk < peersk) {
623 spin_lock(&sk->sk_peer_lock);
624 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
625 } else {
626 spin_lock(&peersk->sk_peer_lock);
627 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
628 }
629 old_pid = sk->sk_peer_pid;
630 old_cred = sk->sk_peer_cred;
631 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
632 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
633
634 spin_unlock(&sk->sk_peer_lock);
635 spin_unlock(&peersk->sk_peer_lock);
636
637 put_pid(old_pid);
638 put_cred(old_cred);
639}
640
641static int unix_listen(struct socket *sock, int backlog)
642{
643 int err;
644 struct sock *sk = sock->sk;
645 struct unix_sock *u = unix_sk(sk);
646
647 err = -EOPNOTSUPP;
648 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
649 goto out; /* Only stream/seqpacket sockets accept */
650 err = -EINVAL;
651 if (!u->addr)
652 goto out; /* No listens on an unbound socket */
653 unix_state_lock(sk);
654 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
655 goto out_unlock;
656 if (backlog > sk->sk_max_ack_backlog)
657 wake_up_interruptible_all(&u->peer_wait);
658 sk->sk_max_ack_backlog = backlog;
659 sk->sk_state = TCP_LISTEN;
660 /* set credentials so connect can copy them */
661 init_peercred(sk);
662 err = 0;
663
664out_unlock:
665 unix_state_unlock(sk);
666out:
667 return err;
668}
669
670static int unix_release(struct socket *);
671static int unix_bind(struct socket *, struct sockaddr *, int);
672static int unix_stream_connect(struct socket *, struct sockaddr *,
673 int addr_len, int flags);
674static int unix_socketpair(struct socket *, struct socket *);
675static int unix_accept(struct socket *, struct socket *, int, bool);
676static int unix_getname(struct socket *, struct sockaddr *, int);
677static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
678static __poll_t unix_dgram_poll(struct file *, struct socket *,
679 poll_table *);
680static int unix_ioctl(struct socket *, unsigned int, unsigned long);
681#ifdef CONFIG_COMPAT
682static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
683#endif
684static int unix_shutdown(struct socket *, int);
685static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
686static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
687static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
688 size_t size, int flags);
689static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
690 struct pipe_inode_info *, size_t size,
691 unsigned int flags);
692static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
693static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
694static int unix_dgram_connect(struct socket *, struct sockaddr *,
695 int, int);
696static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
697static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
698 int);
699
700static int unix_set_peek_off(struct sock *sk, int val)
701{
702 struct unix_sock *u = unix_sk(sk);
703
704 if (mutex_lock_interruptible(&u->iolock))
705 return -EINTR;
706
707 sk->sk_peek_off = val;
708 mutex_unlock(&u->iolock);
709
710 return 0;
711}
712
713#ifdef CONFIG_PROC_FS
714static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
715{
716 struct sock *sk = sock->sk;
717 struct unix_sock *u;
718
719 if (sk) {
720 u = unix_sk(sock->sk);
721 seq_printf(m, "scm_fds: %u\n",
722 atomic_read(&u->scm_stat.nr_fds));
723 }
724}
725#else
726#define unix_show_fdinfo NULL
727#endif
728
729static const struct proto_ops unix_stream_ops = {
730 .family = PF_UNIX,
731 .owner = THIS_MODULE,
732 .release = unix_release,
733 .bind = unix_bind,
734 .connect = unix_stream_connect,
735 .socketpair = unix_socketpair,
736 .accept = unix_accept,
737 .getname = unix_getname,
738 .poll = unix_poll,
739 .ioctl = unix_ioctl,
740#ifdef CONFIG_COMPAT
741 .compat_ioctl = unix_compat_ioctl,
742#endif
743 .listen = unix_listen,
744 .shutdown = unix_shutdown,
745 .sendmsg = unix_stream_sendmsg,
746 .recvmsg = unix_stream_recvmsg,
747 .mmap = sock_no_mmap,
748 .sendpage = unix_stream_sendpage,
749 .splice_read = unix_stream_splice_read,
750 .set_peek_off = unix_set_peek_off,
751 .show_fdinfo = unix_show_fdinfo,
752};
753
754static const struct proto_ops unix_dgram_ops = {
755 .family = PF_UNIX,
756 .owner = THIS_MODULE,
757 .release = unix_release,
758 .bind = unix_bind,
759 .connect = unix_dgram_connect,
760 .socketpair = unix_socketpair,
761 .accept = sock_no_accept,
762 .getname = unix_getname,
763 .poll = unix_dgram_poll,
764 .ioctl = unix_ioctl,
765#ifdef CONFIG_COMPAT
766 .compat_ioctl = unix_compat_ioctl,
767#endif
768 .listen = sock_no_listen,
769 .shutdown = unix_shutdown,
770 .sendmsg = unix_dgram_sendmsg,
771 .recvmsg = unix_dgram_recvmsg,
772 .mmap = sock_no_mmap,
773 .sendpage = sock_no_sendpage,
774 .set_peek_off = unix_set_peek_off,
775 .show_fdinfo = unix_show_fdinfo,
776};
777
778static const struct proto_ops unix_seqpacket_ops = {
779 .family = PF_UNIX,
780 .owner = THIS_MODULE,
781 .release = unix_release,
782 .bind = unix_bind,
783 .connect = unix_stream_connect,
784 .socketpair = unix_socketpair,
785 .accept = unix_accept,
786 .getname = unix_getname,
787 .poll = unix_dgram_poll,
788 .ioctl = unix_ioctl,
789#ifdef CONFIG_COMPAT
790 .compat_ioctl = unix_compat_ioctl,
791#endif
792 .listen = unix_listen,
793 .shutdown = unix_shutdown,
794 .sendmsg = unix_seqpacket_sendmsg,
795 .recvmsg = unix_seqpacket_recvmsg,
796 .mmap = sock_no_mmap,
797 .sendpage = sock_no_sendpage,
798 .set_peek_off = unix_set_peek_off,
799 .show_fdinfo = unix_show_fdinfo,
800};
801
802static struct proto unix_proto = {
803 .name = "UNIX",
804 .owner = THIS_MODULE,
805 .obj_size = sizeof(struct unix_sock),
806};
807
808static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
809{
810 struct sock *sk = NULL;
811 struct unix_sock *u;
812
813 atomic_long_inc(&unix_nr_socks);
814 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
815 goto out;
816
817 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
818 if (!sk)
819 goto out;
820
821 sock_init_data(sock, sk);
822
823 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
824 sk->sk_write_space = unix_write_space;
825 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
826 sk->sk_destruct = unix_sock_destructor;
827 u = unix_sk(sk);
828 u->path.dentry = NULL;
829 u->path.mnt = NULL;
830 spin_lock_init(&u->lock);
831 atomic_long_set(&u->inflight, 0);
832 INIT_LIST_HEAD(&u->link);
833 mutex_init(&u->iolock); /* single task reading lock */
834 mutex_init(&u->bindlock); /* single task binding lock */
835 init_waitqueue_head(&u->peer_wait);
836 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
837 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
838 unix_insert_socket(unix_sockets_unbound(sk), sk);
839out:
840 if (sk == NULL)
841 atomic_long_dec(&unix_nr_socks);
842 else {
843 local_bh_disable();
844 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
845 local_bh_enable();
846 }
847 return sk;
848}
849
850static int unix_create(struct net *net, struct socket *sock, int protocol,
851 int kern)
852{
853 if (protocol && protocol != PF_UNIX)
854 return -EPROTONOSUPPORT;
855
856 sock->state = SS_UNCONNECTED;
857
858 switch (sock->type) {
859 case SOCK_STREAM:
860 sock->ops = &unix_stream_ops;
861 break;
862 /*
863 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
864 * nothing uses it.
865 */
866 case SOCK_RAW:
867 sock->type = SOCK_DGRAM;
868 fallthrough;
869 case SOCK_DGRAM:
870 sock->ops = &unix_dgram_ops;
871 break;
872 case SOCK_SEQPACKET:
873 sock->ops = &unix_seqpacket_ops;
874 break;
875 default:
876 return -ESOCKTNOSUPPORT;
877 }
878
879 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
880}
881
882static int unix_release(struct socket *sock)
883{
884 struct sock *sk = sock->sk;
885
886 if (!sk)
887 return 0;
888
889 unix_release_sock(sk, 0);
890 sock->sk = NULL;
891
892 return 0;
893}
894
895static int unix_autobind(struct socket *sock)
896{
897 struct sock *sk = sock->sk;
898 struct net *net = sock_net(sk);
899 struct unix_sock *u = unix_sk(sk);
900 static u32 ordernum = 1;
901 struct unix_address *addr;
902 int err;
903 unsigned int retries = 0;
904
905 err = mutex_lock_interruptible(&u->bindlock);
906 if (err)
907 return err;
908
909 if (u->addr)
910 goto out;
911
912 err = -ENOMEM;
913 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
914 if (!addr)
915 goto out;
916
917 addr->name->sun_family = AF_UNIX;
918 refcount_set(&addr->refcnt, 1);
919
920retry:
921 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
922 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
923 addr->hash ^= sk->sk_type;
924
925 spin_lock(&unix_table_lock);
926 ordernum = (ordernum+1)&0xFFFFF;
927
928 if (__unix_find_socket_byname(net, addr->name, addr->len, addr->hash)) {
929 spin_unlock(&unix_table_lock);
930 /*
931 * __unix_find_socket_byname() may take long time if many names
932 * are already in use.
933 */
934 cond_resched();
935 /* Give up if all names seems to be in use. */
936 if (retries++ == 0xFFFFF) {
937 err = -ENOSPC;
938 kfree(addr);
939 goto out;
940 }
941 goto retry;
942 }
943
944 __unix_set_addr(sk, addr, addr->hash);
945 spin_unlock(&unix_table_lock);
946 err = 0;
947
948out: mutex_unlock(&u->bindlock);
949 return err;
950}
951
952static struct sock *unix_find_other(struct net *net,
953 struct sockaddr_un *sunname, int len,
954 int type, unsigned int hash, int *error)
955{
956 struct sock *u;
957 struct path path;
958 int err = 0;
959
960 if (sunname->sun_path[0]) {
961 struct inode *inode;
962 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
963 if (err)
964 goto fail;
965 inode = d_backing_inode(path.dentry);
966 err = path_permission(&path, MAY_WRITE);
967 if (err)
968 goto put_fail;
969
970 err = -ECONNREFUSED;
971 if (!S_ISSOCK(inode->i_mode))
972 goto put_fail;
973 u = unix_find_socket_byinode(inode);
974 if (!u)
975 goto put_fail;
976
977 if (u->sk_type == type)
978 touch_atime(&path);
979
980 path_put(&path);
981
982 err = -EPROTOTYPE;
983 if (u->sk_type != type) {
984 sock_put(u);
985 goto fail;
986 }
987 } else {
988 err = -ECONNREFUSED;
989 u = unix_find_socket_byname(net, sunname, len, type ^ hash);
990 if (u) {
991 struct dentry *dentry;
992 dentry = unix_sk(u)->path.dentry;
993 if (dentry)
994 touch_atime(&unix_sk(u)->path);
995 } else
996 goto fail;
997 }
998 return u;
999
1000put_fail:
1001 path_put(&path);
1002fail:
1003 *error = err;
1004 return NULL;
1005}
1006
1007static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)
1008{
1009 struct unix_sock *u = unix_sk(sk);
1010 umode_t mode = S_IFSOCK |
1011 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1012 struct user_namespace *ns; // barf...
1013 struct path parent;
1014 struct dentry *dentry;
1015 unsigned int hash;
1016 int err;
1017
1018 /*
1019 * Get the parent directory, calculate the hash for last
1020 * component.
1021 */
1022 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1023 if (IS_ERR(dentry))
1024 return PTR_ERR(dentry);
1025 ns = mnt_user_ns(parent.mnt);
1026
1027 /*
1028 * All right, let's create it.
1029 */
1030 err = security_path_mknod(&parent, dentry, mode, 0);
1031 if (!err)
1032 err = vfs_mknod(ns, d_inode(parent.dentry), dentry, mode, 0);
1033 if (err)
1034 goto out;
1035 err = mutex_lock_interruptible(&u->bindlock);
1036 if (err)
1037 goto out_unlink;
1038 if (u->addr)
1039 goto out_unlock;
1040
1041 addr->hash = UNIX_HASH_SIZE;
1042 hash = d_backing_inode(dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1043 spin_lock(&unix_table_lock);
1044 u->path.mnt = mntget(parent.mnt);
1045 u->path.dentry = dget(dentry);
1046 __unix_set_addr(sk, addr, hash);
1047 spin_unlock(&unix_table_lock);
1048 mutex_unlock(&u->bindlock);
1049 done_path_create(&parent, dentry);
1050 return 0;
1051
1052out_unlock:
1053 mutex_unlock(&u->bindlock);
1054 err = -EINVAL;
1055out_unlink:
1056 /* failed after successful mknod? unlink what we'd created... */
1057 vfs_unlink(ns, d_inode(parent.dentry), dentry, NULL);
1058out:
1059 done_path_create(&parent, dentry);
1060 return err;
1061}
1062
1063static int unix_bind_abstract(struct sock *sk, struct unix_address *addr)
1064{
1065 struct unix_sock *u = unix_sk(sk);
1066 int err;
1067
1068 err = mutex_lock_interruptible(&u->bindlock);
1069 if (err)
1070 return err;
1071
1072 if (u->addr) {
1073 mutex_unlock(&u->bindlock);
1074 return -EINVAL;
1075 }
1076
1077 spin_lock(&unix_table_lock);
1078 if (__unix_find_socket_byname(sock_net(sk), addr->name, addr->len,
1079 addr->hash)) {
1080 spin_unlock(&unix_table_lock);
1081 mutex_unlock(&u->bindlock);
1082 return -EADDRINUSE;
1083 }
1084 __unix_set_addr(sk, addr, addr->hash);
1085 spin_unlock(&unix_table_lock);
1086 mutex_unlock(&u->bindlock);
1087 return 0;
1088}
1089
1090static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1091{
1092 struct sock *sk = sock->sk;
1093 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1094 char *sun_path = sunaddr->sun_path;
1095 int err;
1096 unsigned int hash;
1097 struct unix_address *addr;
1098
1099 if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1100 sunaddr->sun_family != AF_UNIX)
1101 return -EINVAL;
1102
1103 if (addr_len == sizeof(short))
1104 return unix_autobind(sock);
1105
1106 err = unix_mkname(sunaddr, addr_len, &hash);
1107 if (err < 0)
1108 return err;
1109 addr_len = err;
1110 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1111 if (!addr)
1112 return -ENOMEM;
1113
1114 memcpy(addr->name, sunaddr, addr_len);
1115 addr->len = addr_len;
1116 addr->hash = hash ^ sk->sk_type;
1117 refcount_set(&addr->refcnt, 1);
1118
1119 if (sun_path[0])
1120 err = unix_bind_bsd(sk, addr);
1121 else
1122 err = unix_bind_abstract(sk, addr);
1123 if (err)
1124 unix_release_addr(addr);
1125 return err == -EEXIST ? -EADDRINUSE : err;
1126}
1127
1128static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1129{
1130 if (unlikely(sk1 == sk2) || !sk2) {
1131 unix_state_lock(sk1);
1132 return;
1133 }
1134 if (sk1 < sk2) {
1135 unix_state_lock(sk1);
1136 unix_state_lock_nested(sk2);
1137 } else {
1138 unix_state_lock(sk2);
1139 unix_state_lock_nested(sk1);
1140 }
1141}
1142
1143static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1144{
1145 if (unlikely(sk1 == sk2) || !sk2) {
1146 unix_state_unlock(sk1);
1147 return;
1148 }
1149 unix_state_unlock(sk1);
1150 unix_state_unlock(sk2);
1151}
1152
1153static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1154 int alen, int flags)
1155{
1156 struct sock *sk = sock->sk;
1157 struct net *net = sock_net(sk);
1158 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1159 struct sock *other;
1160 unsigned int hash;
1161 int err;
1162
1163 err = -EINVAL;
1164 if (alen < offsetofend(struct sockaddr, sa_family))
1165 goto out;
1166
1167 if (addr->sa_family != AF_UNSPEC) {
1168 err = unix_mkname(sunaddr, alen, &hash);
1169 if (err < 0)
1170 goto out;
1171 alen = err;
1172
1173 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1174 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1175 goto out;
1176
1177restart:
1178 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1179 if (!other)
1180 goto out;
1181
1182 unix_state_double_lock(sk, other);
1183
1184 /* Apparently VFS overslept socket death. Retry. */
1185 if (sock_flag(other, SOCK_DEAD)) {
1186 unix_state_double_unlock(sk, other);
1187 sock_put(other);
1188 goto restart;
1189 }
1190
1191 err = -EPERM;
1192 if (!unix_may_send(sk, other))
1193 goto out_unlock;
1194
1195 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1196 if (err)
1197 goto out_unlock;
1198
1199 } else {
1200 /*
1201 * 1003.1g breaking connected state with AF_UNSPEC
1202 */
1203 other = NULL;
1204 unix_state_double_lock(sk, other);
1205 }
1206
1207 /*
1208 * If it was connected, reconnect.
1209 */
1210 if (unix_peer(sk)) {
1211 struct sock *old_peer = unix_peer(sk);
1212 unix_peer(sk) = other;
1213 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1214
1215 unix_state_double_unlock(sk, other);
1216
1217 if (other != old_peer)
1218 unix_dgram_disconnected(sk, old_peer);
1219 sock_put(old_peer);
1220 } else {
1221 unix_peer(sk) = other;
1222 unix_state_double_unlock(sk, other);
1223 }
1224 return 0;
1225
1226out_unlock:
1227 unix_state_double_unlock(sk, other);
1228 sock_put(other);
1229out:
1230 return err;
1231}
1232
1233static long unix_wait_for_peer(struct sock *other, long timeo)
1234 __releases(&unix_sk(other)->lock)
1235{
1236 struct unix_sock *u = unix_sk(other);
1237 int sched;
1238 DEFINE_WAIT(wait);
1239
1240 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1241
1242 sched = !sock_flag(other, SOCK_DEAD) &&
1243 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1244 unix_recvq_full(other);
1245
1246 unix_state_unlock(other);
1247
1248 if (sched)
1249 timeo = schedule_timeout(timeo);
1250
1251 finish_wait(&u->peer_wait, &wait);
1252 return timeo;
1253}
1254
1255static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1256 int addr_len, int flags)
1257{
1258 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1259 struct sock *sk = sock->sk;
1260 struct net *net = sock_net(sk);
1261 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1262 struct sock *newsk = NULL;
1263 struct sock *other = NULL;
1264 struct sk_buff *skb = NULL;
1265 unsigned int hash;
1266 int st;
1267 int err;
1268 long timeo;
1269
1270 err = unix_mkname(sunaddr, addr_len, &hash);
1271 if (err < 0)
1272 goto out;
1273 addr_len = err;
1274
1275 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1276 (err = unix_autobind(sock)) != 0)
1277 goto out;
1278
1279 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1280
1281 /* First of all allocate resources.
1282 If we will make it after state is locked,
1283 we will have to recheck all again in any case.
1284 */
1285
1286 err = -ENOMEM;
1287
1288 /* create new sock for complete connection */
1289 newsk = unix_create1(sock_net(sk), NULL, 0);
1290 if (newsk == NULL)
1291 goto out;
1292
1293 /* Allocate skb for sending to listening sock */
1294 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1295 if (skb == NULL)
1296 goto out;
1297
1298restart:
1299 /* Find listening sock. */
1300 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1301 if (!other)
1302 goto out;
1303
1304 /* Latch state of peer */
1305 unix_state_lock(other);
1306
1307 /* Apparently VFS overslept socket death. Retry. */
1308 if (sock_flag(other, SOCK_DEAD)) {
1309 unix_state_unlock(other);
1310 sock_put(other);
1311 goto restart;
1312 }
1313
1314 err = -ECONNREFUSED;
1315 if (other->sk_state != TCP_LISTEN)
1316 goto out_unlock;
1317 if (other->sk_shutdown & RCV_SHUTDOWN)
1318 goto out_unlock;
1319
1320 if (unix_recvq_full(other)) {
1321 err = -EAGAIN;
1322 if (!timeo)
1323 goto out_unlock;
1324
1325 timeo = unix_wait_for_peer(other, timeo);
1326
1327 err = sock_intr_errno(timeo);
1328 if (signal_pending(current))
1329 goto out;
1330 sock_put(other);
1331 goto restart;
1332 }
1333
1334 /* Latch our state.
1335
1336 It is tricky place. We need to grab our state lock and cannot
1337 drop lock on peer. It is dangerous because deadlock is
1338 possible. Connect to self case and simultaneous
1339 attempt to connect are eliminated by checking socket
1340 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1341 check this before attempt to grab lock.
1342
1343 Well, and we have to recheck the state after socket locked.
1344 */
1345 st = sk->sk_state;
1346
1347 switch (st) {
1348 case TCP_CLOSE:
1349 /* This is ok... continue with connect */
1350 break;
1351 case TCP_ESTABLISHED:
1352 /* Socket is already connected */
1353 err = -EISCONN;
1354 goto out_unlock;
1355 default:
1356 err = -EINVAL;
1357 goto out_unlock;
1358 }
1359
1360 unix_state_lock_nested(sk);
1361
1362 if (sk->sk_state != st) {
1363 unix_state_unlock(sk);
1364 unix_state_unlock(other);
1365 sock_put(other);
1366 goto restart;
1367 }
1368
1369 err = security_unix_stream_connect(sk, other, newsk);
1370 if (err) {
1371 unix_state_unlock(sk);
1372 goto out_unlock;
1373 }
1374
1375 /* The way is open! Fastly set all the necessary fields... */
1376
1377 sock_hold(sk);
1378 unix_peer(newsk) = sk;
1379 newsk->sk_state = TCP_ESTABLISHED;
1380 newsk->sk_type = sk->sk_type;
1381 init_peercred(newsk);
1382 newu = unix_sk(newsk);
1383 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1384 otheru = unix_sk(other);
1385
1386 /* copy address information from listening to new sock
1387 *
1388 * The contents of *(otheru->addr) and otheru->path
1389 * are seen fully set up here, since we have found
1390 * otheru in hash under unix_table_lock. Insertion
1391 * into the hash chain we'd found it in had been done
1392 * in an earlier critical area protected by unix_table_lock,
1393 * the same one where we'd set *(otheru->addr) contents,
1394 * as well as otheru->path and otheru->addr itself.
1395 *
1396 * Using smp_store_release() here to set newu->addr
1397 * is enough to make those stores, as well as stores
1398 * to newu->path visible to anyone who gets newu->addr
1399 * by smp_load_acquire(). IOW, the same warranties
1400 * as for unix_sock instances bound in unix_bind() or
1401 * in unix_autobind().
1402 */
1403 if (otheru->path.dentry) {
1404 path_get(&otheru->path);
1405 newu->path = otheru->path;
1406 }
1407 refcount_inc(&otheru->addr->refcnt);
1408 smp_store_release(&newu->addr, otheru->addr);
1409
1410 /* Set credentials */
1411 copy_peercred(sk, other);
1412
1413 sock->state = SS_CONNECTED;
1414 sk->sk_state = TCP_ESTABLISHED;
1415 sock_hold(newsk);
1416
1417 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1418 unix_peer(sk) = newsk;
1419
1420 unix_state_unlock(sk);
1421
1422 /* take ten and send info to listening sock */
1423 spin_lock(&other->sk_receive_queue.lock);
1424 __skb_queue_tail(&other->sk_receive_queue, skb);
1425 spin_unlock(&other->sk_receive_queue.lock);
1426 unix_state_unlock(other);
1427 other->sk_data_ready(other);
1428 sock_put(other);
1429 return 0;
1430
1431out_unlock:
1432 if (other)
1433 unix_state_unlock(other);
1434
1435out:
1436 kfree_skb(skb);
1437 if (newsk)
1438 unix_release_sock(newsk, 0);
1439 if (other)
1440 sock_put(other);
1441 return err;
1442}
1443
1444static int unix_socketpair(struct socket *socka, struct socket *sockb)
1445{
1446 struct sock *ska = socka->sk, *skb = sockb->sk;
1447
1448 /* Join our sockets back to back */
1449 sock_hold(ska);
1450 sock_hold(skb);
1451 unix_peer(ska) = skb;
1452 unix_peer(skb) = ska;
1453 init_peercred(ska);
1454 init_peercred(skb);
1455
1456 if (ska->sk_type != SOCK_DGRAM) {
1457 ska->sk_state = TCP_ESTABLISHED;
1458 skb->sk_state = TCP_ESTABLISHED;
1459 socka->state = SS_CONNECTED;
1460 sockb->state = SS_CONNECTED;
1461 }
1462 return 0;
1463}
1464
1465static void unix_sock_inherit_flags(const struct socket *old,
1466 struct socket *new)
1467{
1468 if (test_bit(SOCK_PASSCRED, &old->flags))
1469 set_bit(SOCK_PASSCRED, &new->flags);
1470 if (test_bit(SOCK_PASSSEC, &old->flags))
1471 set_bit(SOCK_PASSSEC, &new->flags);
1472}
1473
1474static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1475 bool kern)
1476{
1477 struct sock *sk = sock->sk;
1478 struct sock *tsk;
1479 struct sk_buff *skb;
1480 int err;
1481
1482 err = -EOPNOTSUPP;
1483 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1484 goto out;
1485
1486 err = -EINVAL;
1487 if (sk->sk_state != TCP_LISTEN)
1488 goto out;
1489
1490 /* If socket state is TCP_LISTEN it cannot change (for now...),
1491 * so that no locks are necessary.
1492 */
1493
1494 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1495 if (!skb) {
1496 /* This means receive shutdown. */
1497 if (err == 0)
1498 err = -EINVAL;
1499 goto out;
1500 }
1501
1502 tsk = skb->sk;
1503 skb_free_datagram(sk, skb);
1504 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1505
1506 /* attach accepted sock to socket */
1507 unix_state_lock(tsk);
1508 newsock->state = SS_CONNECTED;
1509 unix_sock_inherit_flags(sock, newsock);
1510 sock_graft(tsk, newsock);
1511 unix_state_unlock(tsk);
1512 return 0;
1513
1514out:
1515 return err;
1516}
1517
1518
1519static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1520{
1521 struct sock *sk = sock->sk;
1522 struct unix_address *addr;
1523 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1524 int err = 0;
1525
1526 if (peer) {
1527 sk = unix_peer_get(sk);
1528
1529 err = -ENOTCONN;
1530 if (!sk)
1531 goto out;
1532 err = 0;
1533 } else {
1534 sock_hold(sk);
1535 }
1536
1537 addr = smp_load_acquire(&unix_sk(sk)->addr);
1538 if (!addr) {
1539 sunaddr->sun_family = AF_UNIX;
1540 sunaddr->sun_path[0] = 0;
1541 err = sizeof(short);
1542 } else {
1543 err = addr->len;
1544 memcpy(sunaddr, addr->name, addr->len);
1545 }
1546 sock_put(sk);
1547out:
1548 return err;
1549}
1550
1551static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1552{
1553 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1554
1555 /*
1556 * Garbage collection of unix sockets starts by selecting a set of
1557 * candidate sockets which have reference only from being in flight
1558 * (total_refs == inflight_refs). This condition is checked once during
1559 * the candidate collection phase, and candidates are marked as such, so
1560 * that non-candidates can later be ignored. While inflight_refs is
1561 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1562 * is an instantaneous decision.
1563 *
1564 * Once a candidate, however, the socket must not be reinstalled into a
1565 * file descriptor while the garbage collection is in progress.
1566 *
1567 * If the above conditions are met, then the directed graph of
1568 * candidates (*) does not change while unix_gc_lock is held.
1569 *
1570 * Any operations that changes the file count through file descriptors
1571 * (dup, close, sendmsg) does not change the graph since candidates are
1572 * not installed in fds.
1573 *
1574 * Dequeing a candidate via recvmsg would install it into an fd, but
1575 * that takes unix_gc_lock to decrement the inflight count, so it's
1576 * serialized with garbage collection.
1577 *
1578 * MSG_PEEK is special in that it does not change the inflight count,
1579 * yet does install the socket into an fd. The following lock/unlock
1580 * pair is to ensure serialization with garbage collection. It must be
1581 * done between incrementing the file count and installing the file into
1582 * an fd.
1583 *
1584 * If garbage collection starts after the barrier provided by the
1585 * lock/unlock, then it will see the elevated refcount and not mark this
1586 * as a candidate. If a garbage collection is already in progress
1587 * before the file count was incremented, then the lock/unlock pair will
1588 * ensure that garbage collection is finished before progressing to
1589 * installing the fd.
1590 *
1591 * (*) A -> B where B is on the queue of A or B is on the queue of C
1592 * which is on the queue of listening socket A.
1593 */
1594 spin_lock(&unix_gc_lock);
1595 spin_unlock(&unix_gc_lock);
1596}
1597
1598static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1599{
1600 int err = 0;
1601
1602 UNIXCB(skb).pid = get_pid(scm->pid);
1603 UNIXCB(skb).uid = scm->creds.uid;
1604 UNIXCB(skb).gid = scm->creds.gid;
1605 UNIXCB(skb).fp = NULL;
1606 unix_get_secdata(scm, skb);
1607 if (scm->fp && send_fds)
1608 err = unix_attach_fds(scm, skb);
1609
1610 skb->destructor = unix_destruct_scm;
1611 return err;
1612}
1613
1614static bool unix_passcred_enabled(const struct socket *sock,
1615 const struct sock *other)
1616{
1617 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1618 !other->sk_socket ||
1619 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1620}
1621
1622/*
1623 * Some apps rely on write() giving SCM_CREDENTIALS
1624 * We include credentials if source or destination socket
1625 * asserted SOCK_PASSCRED.
1626 */
1627static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1628 const struct sock *other)
1629{
1630 if (UNIXCB(skb).pid)
1631 return;
1632 if (unix_passcred_enabled(sock, other)) {
1633 UNIXCB(skb).pid = get_pid(task_tgid(current));
1634 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1635 }
1636}
1637
1638static int maybe_init_creds(struct scm_cookie *scm,
1639 struct socket *socket,
1640 const struct sock *other)
1641{
1642 int err;
1643 struct msghdr msg = { .msg_controllen = 0 };
1644
1645 err = scm_send(socket, &msg, scm, false);
1646 if (err)
1647 return err;
1648
1649 if (unix_passcred_enabled(socket, other)) {
1650 scm->pid = get_pid(task_tgid(current));
1651 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1652 }
1653 return err;
1654}
1655
1656static bool unix_skb_scm_eq(struct sk_buff *skb,
1657 struct scm_cookie *scm)
1658{
1659 const struct unix_skb_parms *u = &UNIXCB(skb);
1660
1661 return u->pid == scm->pid &&
1662 uid_eq(u->uid, scm->creds.uid) &&
1663 gid_eq(u->gid, scm->creds.gid) &&
1664 unix_secdata_eq(scm, skb);
1665}
1666
1667static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1668{
1669 struct scm_fp_list *fp = UNIXCB(skb).fp;
1670 struct unix_sock *u = unix_sk(sk);
1671
1672 if (unlikely(fp && fp->count))
1673 atomic_add(fp->count, &u->scm_stat.nr_fds);
1674}
1675
1676static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1677{
1678 struct scm_fp_list *fp = UNIXCB(skb).fp;
1679 struct unix_sock *u = unix_sk(sk);
1680
1681 if (unlikely(fp && fp->count))
1682 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1683}
1684
1685/*
1686 * Send AF_UNIX data.
1687 */
1688
1689static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1690 size_t len)
1691{
1692 struct sock *sk = sock->sk;
1693 struct net *net = sock_net(sk);
1694 struct unix_sock *u = unix_sk(sk);
1695 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1696 struct sock *other = NULL;
1697 int namelen = 0; /* fake GCC */
1698 int err;
1699 unsigned int hash;
1700 struct sk_buff *skb;
1701 long timeo;
1702 struct scm_cookie scm;
1703 int data_len = 0;
1704 int sk_locked;
1705
1706 wait_for_unix_gc();
1707 err = scm_send(sock, msg, &scm, false);
1708 if (err < 0)
1709 return err;
1710
1711 err = -EOPNOTSUPP;
1712 if (msg->msg_flags&MSG_OOB)
1713 goto out;
1714
1715 if (msg->msg_namelen) {
1716 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1717 if (err < 0)
1718 goto out;
1719 namelen = err;
1720 } else {
1721 sunaddr = NULL;
1722 err = -ENOTCONN;
1723 other = unix_peer_get(sk);
1724 if (!other)
1725 goto out;
1726 }
1727
1728 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1729 && (err = unix_autobind(sock)) != 0)
1730 goto out;
1731
1732 err = -EMSGSIZE;
1733 if (len > sk->sk_sndbuf - 32)
1734 goto out;
1735
1736 if (len > SKB_MAX_ALLOC) {
1737 data_len = min_t(size_t,
1738 len - SKB_MAX_ALLOC,
1739 MAX_SKB_FRAGS * PAGE_SIZE);
1740 data_len = PAGE_ALIGN(data_len);
1741
1742 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1743 }
1744
1745 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1746 msg->msg_flags & MSG_DONTWAIT, &err,
1747 PAGE_ALLOC_COSTLY_ORDER);
1748 if (skb == NULL)
1749 goto out;
1750
1751 err = unix_scm_to_skb(&scm, skb, true);
1752 if (err < 0)
1753 goto out_free;
1754
1755 skb_put(skb, len - data_len);
1756 skb->data_len = data_len;
1757 skb->len = len;
1758 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1759 if (err)
1760 goto out_free;
1761
1762 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1763
1764restart:
1765 if (!other) {
1766 err = -ECONNRESET;
1767 if (sunaddr == NULL)
1768 goto out_free;
1769
1770 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1771 hash, &err);
1772 if (other == NULL)
1773 goto out_free;
1774 }
1775
1776 if (sk_filter(other, skb) < 0) {
1777 /* Toss the packet but do not return any error to the sender */
1778 err = len;
1779 goto out_free;
1780 }
1781
1782 sk_locked = 0;
1783 unix_state_lock(other);
1784restart_locked:
1785 err = -EPERM;
1786 if (!unix_may_send(sk, other))
1787 goto out_unlock;
1788
1789 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1790 /*
1791 * Check with 1003.1g - what should
1792 * datagram error
1793 */
1794 unix_state_unlock(other);
1795 sock_put(other);
1796
1797 if (!sk_locked)
1798 unix_state_lock(sk);
1799
1800 err = 0;
1801 if (unix_peer(sk) == other) {
1802 unix_peer(sk) = NULL;
1803 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1804
1805 unix_state_unlock(sk);
1806
1807 unix_dgram_disconnected(sk, other);
1808 sock_put(other);
1809 err = -ECONNREFUSED;
1810 } else {
1811 unix_state_unlock(sk);
1812 }
1813
1814 other = NULL;
1815 if (err)
1816 goto out_free;
1817 goto restart;
1818 }
1819
1820 err = -EPIPE;
1821 if (other->sk_shutdown & RCV_SHUTDOWN)
1822 goto out_unlock;
1823
1824 if (sk->sk_type != SOCK_SEQPACKET) {
1825 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1826 if (err)
1827 goto out_unlock;
1828 }
1829
1830 /* other == sk && unix_peer(other) != sk if
1831 * - unix_peer(sk) == NULL, destination address bound to sk
1832 * - unix_peer(sk) == sk by time of get but disconnected before lock
1833 */
1834 if (other != sk &&
1835 unlikely(unix_peer(other) != sk &&
1836 unix_recvq_full_lockless(other))) {
1837 if (timeo) {
1838 timeo = unix_wait_for_peer(other, timeo);
1839
1840 err = sock_intr_errno(timeo);
1841 if (signal_pending(current))
1842 goto out_free;
1843
1844 goto restart;
1845 }
1846
1847 if (!sk_locked) {
1848 unix_state_unlock(other);
1849 unix_state_double_lock(sk, other);
1850 }
1851
1852 if (unix_peer(sk) != other ||
1853 unix_dgram_peer_wake_me(sk, other)) {
1854 err = -EAGAIN;
1855 sk_locked = 1;
1856 goto out_unlock;
1857 }
1858
1859 if (!sk_locked) {
1860 sk_locked = 1;
1861 goto restart_locked;
1862 }
1863 }
1864
1865 if (unlikely(sk_locked))
1866 unix_state_unlock(sk);
1867
1868 if (sock_flag(other, SOCK_RCVTSTAMP))
1869 __net_timestamp(skb);
1870 maybe_add_creds(skb, sock, other);
1871 scm_stat_add(other, skb);
1872 skb_queue_tail(&other->sk_receive_queue, skb);
1873 unix_state_unlock(other);
1874 other->sk_data_ready(other);
1875 sock_put(other);
1876 scm_destroy(&scm);
1877 return len;
1878
1879out_unlock:
1880 if (sk_locked)
1881 unix_state_unlock(sk);
1882 unix_state_unlock(other);
1883out_free:
1884 kfree_skb(skb);
1885out:
1886 if (other)
1887 sock_put(other);
1888 scm_destroy(&scm);
1889 return err;
1890}
1891
1892/* We use paged skbs for stream sockets, and limit occupancy to 32768
1893 * bytes, and a minimum of a full page.
1894 */
1895#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1896
1897static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1898 size_t len)
1899{
1900 struct sock *sk = sock->sk;
1901 struct sock *other = NULL;
1902 int err, size;
1903 struct sk_buff *skb;
1904 int sent = 0;
1905 struct scm_cookie scm;
1906 bool fds_sent = false;
1907 int data_len;
1908
1909 wait_for_unix_gc();
1910 err = scm_send(sock, msg, &scm, false);
1911 if (err < 0)
1912 return err;
1913
1914 err = -EOPNOTSUPP;
1915 if (msg->msg_flags&MSG_OOB)
1916 goto out_err;
1917
1918 if (msg->msg_namelen) {
1919 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1920 goto out_err;
1921 } else {
1922 err = -ENOTCONN;
1923 other = unix_peer(sk);
1924 if (!other)
1925 goto out_err;
1926 }
1927
1928 if (sk->sk_shutdown & SEND_SHUTDOWN)
1929 goto pipe_err;
1930
1931 while (sent < len) {
1932 size = len - sent;
1933
1934 /* Keep two messages in the pipe so it schedules better */
1935 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1936
1937 /* allow fallback to order-0 allocations */
1938 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1939
1940 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1941
1942 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1943
1944 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1945 msg->msg_flags & MSG_DONTWAIT, &err,
1946 get_order(UNIX_SKB_FRAGS_SZ));
1947 if (!skb)
1948 goto out_err;
1949
1950 /* Only send the fds in the first buffer */
1951 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1952 if (err < 0) {
1953 kfree_skb(skb);
1954 goto out_err;
1955 }
1956 fds_sent = true;
1957
1958 skb_put(skb, size - data_len);
1959 skb->data_len = data_len;
1960 skb->len = size;
1961 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1962 if (err) {
1963 kfree_skb(skb);
1964 goto out_err;
1965 }
1966
1967 unix_state_lock(other);
1968
1969 if (sock_flag(other, SOCK_DEAD) ||
1970 (other->sk_shutdown & RCV_SHUTDOWN))
1971 goto pipe_err_free;
1972
1973 maybe_add_creds(skb, sock, other);
1974 scm_stat_add(other, skb);
1975 skb_queue_tail(&other->sk_receive_queue, skb);
1976 unix_state_unlock(other);
1977 other->sk_data_ready(other);
1978 sent += size;
1979 }
1980
1981 scm_destroy(&scm);
1982
1983 return sent;
1984
1985pipe_err_free:
1986 unix_state_unlock(other);
1987 kfree_skb(skb);
1988pipe_err:
1989 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1990 send_sig(SIGPIPE, current, 0);
1991 err = -EPIPE;
1992out_err:
1993 scm_destroy(&scm);
1994 return sent ? : err;
1995}
1996
1997static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1998 int offset, size_t size, int flags)
1999{
2000 int err;
2001 bool send_sigpipe = false;
2002 bool init_scm = true;
2003 struct scm_cookie scm;
2004 struct sock *other, *sk = socket->sk;
2005 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
2006
2007 if (flags & MSG_OOB)
2008 return -EOPNOTSUPP;
2009
2010 other = unix_peer(sk);
2011 if (!other || sk->sk_state != TCP_ESTABLISHED)
2012 return -ENOTCONN;
2013
2014 if (false) {
2015alloc_skb:
2016 unix_state_unlock(other);
2017 mutex_unlock(&unix_sk(other)->iolock);
2018 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
2019 &err, 0);
2020 if (!newskb)
2021 goto err;
2022 }
2023
2024 /* we must acquire iolock as we modify already present
2025 * skbs in the sk_receive_queue and mess with skb->len
2026 */
2027 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
2028 if (err) {
2029 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
2030 goto err;
2031 }
2032
2033 if (sk->sk_shutdown & SEND_SHUTDOWN) {
2034 err = -EPIPE;
2035 send_sigpipe = true;
2036 goto err_unlock;
2037 }
2038
2039 unix_state_lock(other);
2040
2041 if (sock_flag(other, SOCK_DEAD) ||
2042 other->sk_shutdown & RCV_SHUTDOWN) {
2043 err = -EPIPE;
2044 send_sigpipe = true;
2045 goto err_state_unlock;
2046 }
2047
2048 if (init_scm) {
2049 err = maybe_init_creds(&scm, socket, other);
2050 if (err)
2051 goto err_state_unlock;
2052 init_scm = false;
2053 }
2054
2055 skb = skb_peek_tail(&other->sk_receive_queue);
2056 if (tail && tail == skb) {
2057 skb = newskb;
2058 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2059 if (newskb) {
2060 skb = newskb;
2061 } else {
2062 tail = skb;
2063 goto alloc_skb;
2064 }
2065 } else if (newskb) {
2066 /* this is fast path, we don't necessarily need to
2067 * call to kfree_skb even though with newskb == NULL
2068 * this - does no harm
2069 */
2070 consume_skb(newskb);
2071 newskb = NULL;
2072 }
2073
2074 if (skb_append_pagefrags(skb, page, offset, size)) {
2075 tail = skb;
2076 goto alloc_skb;
2077 }
2078
2079 skb->len += size;
2080 skb->data_len += size;
2081 skb->truesize += size;
2082 refcount_add(size, &sk->sk_wmem_alloc);
2083
2084 if (newskb) {
2085 err = unix_scm_to_skb(&scm, skb, false);
2086 if (err)
2087 goto err_state_unlock;
2088 spin_lock(&other->sk_receive_queue.lock);
2089 __skb_queue_tail(&other->sk_receive_queue, newskb);
2090 spin_unlock(&other->sk_receive_queue.lock);
2091 }
2092
2093 unix_state_unlock(other);
2094 mutex_unlock(&unix_sk(other)->iolock);
2095
2096 other->sk_data_ready(other);
2097 scm_destroy(&scm);
2098 return size;
2099
2100err_state_unlock:
2101 unix_state_unlock(other);
2102err_unlock:
2103 mutex_unlock(&unix_sk(other)->iolock);
2104err:
2105 kfree_skb(newskb);
2106 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2107 send_sig(SIGPIPE, current, 0);
2108 if (!init_scm)
2109 scm_destroy(&scm);
2110 return err;
2111}
2112
2113static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2114 size_t len)
2115{
2116 int err;
2117 struct sock *sk = sock->sk;
2118
2119 err = sock_error(sk);
2120 if (err)
2121 return err;
2122
2123 if (sk->sk_state != TCP_ESTABLISHED)
2124 return -ENOTCONN;
2125
2126 if (msg->msg_namelen)
2127 msg->msg_namelen = 0;
2128
2129 return unix_dgram_sendmsg(sock, msg, len);
2130}
2131
2132static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2133 size_t size, int flags)
2134{
2135 struct sock *sk = sock->sk;
2136
2137 if (sk->sk_state != TCP_ESTABLISHED)
2138 return -ENOTCONN;
2139
2140 return unix_dgram_recvmsg(sock, msg, size, flags);
2141}
2142
2143static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2144{
2145 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2146
2147 if (addr) {
2148 msg->msg_namelen = addr->len;
2149 memcpy(msg->msg_name, addr->name, addr->len);
2150 }
2151}
2152
2153static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2154 size_t size, int flags)
2155{
2156 struct scm_cookie scm;
2157 struct sock *sk = sock->sk;
2158 struct unix_sock *u = unix_sk(sk);
2159 struct sk_buff *skb, *last;
2160 long timeo;
2161 int skip;
2162 int err;
2163
2164 err = -EOPNOTSUPP;
2165 if (flags&MSG_OOB)
2166 goto out;
2167
2168 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2169
2170 do {
2171 mutex_lock(&u->iolock);
2172
2173 skip = sk_peek_offset(sk, flags);
2174 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2175 &skip, &err, &last);
2176 if (skb) {
2177 if (!(flags & MSG_PEEK))
2178 scm_stat_del(sk, skb);
2179 break;
2180 }
2181
2182 mutex_unlock(&u->iolock);
2183
2184 if (err != -EAGAIN)
2185 break;
2186 } while (timeo &&
2187 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2188 &err, &timeo, last));
2189
2190 if (!skb) { /* implies iolock unlocked */
2191 unix_state_lock(sk);
2192 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2193 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2194 (sk->sk_shutdown & RCV_SHUTDOWN))
2195 err = 0;
2196 unix_state_unlock(sk);
2197 goto out;
2198 }
2199
2200 if (wq_has_sleeper(&u->peer_wait))
2201 wake_up_interruptible_sync_poll(&u->peer_wait,
2202 EPOLLOUT | EPOLLWRNORM |
2203 EPOLLWRBAND);
2204
2205 if (msg->msg_name)
2206 unix_copy_addr(msg, skb->sk);
2207
2208 if (size > skb->len - skip)
2209 size = skb->len - skip;
2210 else if (size < skb->len - skip)
2211 msg->msg_flags |= MSG_TRUNC;
2212
2213 err = skb_copy_datagram_msg(skb, skip, msg, size);
2214 if (err)
2215 goto out_free;
2216
2217 if (sock_flag(sk, SOCK_RCVTSTAMP))
2218 __sock_recv_timestamp(msg, sk, skb);
2219
2220 memset(&scm, 0, sizeof(scm));
2221
2222 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2223 unix_set_secdata(&scm, skb);
2224
2225 if (!(flags & MSG_PEEK)) {
2226 if (UNIXCB(skb).fp)
2227 unix_detach_fds(&scm, skb);
2228
2229 sk_peek_offset_bwd(sk, skb->len);
2230 } else {
2231 /* It is questionable: on PEEK we could:
2232 - do not return fds - good, but too simple 8)
2233 - return fds, and do not return them on read (old strategy,
2234 apparently wrong)
2235 - clone fds (I chose it for now, it is the most universal
2236 solution)
2237
2238 POSIX 1003.1g does not actually define this clearly
2239 at all. POSIX 1003.1g doesn't define a lot of things
2240 clearly however!
2241
2242 */
2243
2244 sk_peek_offset_fwd(sk, size);
2245
2246 if (UNIXCB(skb).fp)
2247 unix_peek_fds(&scm, skb);
2248 }
2249 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2250
2251 scm_recv(sock, msg, &scm, flags);
2252
2253out_free:
2254 skb_free_datagram(sk, skb);
2255 mutex_unlock(&u->iolock);
2256out:
2257 return err;
2258}
2259
2260/*
2261 * Sleep until more data has arrived. But check for races..
2262 */
2263static long unix_stream_data_wait(struct sock *sk, long timeo,
2264 struct sk_buff *last, unsigned int last_len,
2265 bool freezable)
2266{
2267 struct sk_buff *tail;
2268 DEFINE_WAIT(wait);
2269
2270 unix_state_lock(sk);
2271
2272 for (;;) {
2273 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2274
2275 tail = skb_peek_tail(&sk->sk_receive_queue);
2276 if (tail != last ||
2277 (tail && tail->len != last_len) ||
2278 sk->sk_err ||
2279 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2280 signal_pending(current) ||
2281 !timeo)
2282 break;
2283
2284 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2285 unix_state_unlock(sk);
2286 if (freezable)
2287 timeo = freezable_schedule_timeout(timeo);
2288 else
2289 timeo = schedule_timeout(timeo);
2290 unix_state_lock(sk);
2291
2292 if (sock_flag(sk, SOCK_DEAD))
2293 break;
2294
2295 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2296 }
2297
2298 finish_wait(sk_sleep(sk), &wait);
2299 unix_state_unlock(sk);
2300 return timeo;
2301}
2302
2303static unsigned int unix_skb_len(const struct sk_buff *skb)
2304{
2305 return skb->len - UNIXCB(skb).consumed;
2306}
2307
2308struct unix_stream_read_state {
2309 int (*recv_actor)(struct sk_buff *, int, int,
2310 struct unix_stream_read_state *);
2311 struct socket *socket;
2312 struct msghdr *msg;
2313 struct pipe_inode_info *pipe;
2314 size_t size;
2315 int flags;
2316 unsigned int splice_flags;
2317};
2318
2319static int unix_stream_read_generic(struct unix_stream_read_state *state,
2320 bool freezable)
2321{
2322 struct scm_cookie scm;
2323 struct socket *sock = state->socket;
2324 struct sock *sk = sock->sk;
2325 struct unix_sock *u = unix_sk(sk);
2326 int copied = 0;
2327 int flags = state->flags;
2328 int noblock = flags & MSG_DONTWAIT;
2329 bool check_creds = false;
2330 int target;
2331 int err = 0;
2332 long timeo;
2333 int skip;
2334 size_t size = state->size;
2335 unsigned int last_len;
2336
2337 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2338 err = -EINVAL;
2339 goto out;
2340 }
2341
2342 if (unlikely(flags & MSG_OOB)) {
2343 err = -EOPNOTSUPP;
2344 goto out;
2345 }
2346
2347 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2348 timeo = sock_rcvtimeo(sk, noblock);
2349
2350 memset(&scm, 0, sizeof(scm));
2351
2352 /* Lock the socket to prevent queue disordering
2353 * while sleeps in memcpy_tomsg
2354 */
2355 mutex_lock(&u->iolock);
2356
2357 skip = max(sk_peek_offset(sk, flags), 0);
2358
2359 do {
2360 int chunk;
2361 bool drop_skb;
2362 struct sk_buff *skb, *last;
2363
2364redo:
2365 unix_state_lock(sk);
2366 if (sock_flag(sk, SOCK_DEAD)) {
2367 err = -ECONNRESET;
2368 goto unlock;
2369 }
2370 last = skb = skb_peek(&sk->sk_receive_queue);
2371 last_len = last ? last->len : 0;
2372again:
2373 if (skb == NULL) {
2374 if (copied >= target)
2375 goto unlock;
2376
2377 /*
2378 * POSIX 1003.1g mandates this order.
2379 */
2380
2381 err = sock_error(sk);
2382 if (err)
2383 goto unlock;
2384 if (sk->sk_shutdown & RCV_SHUTDOWN)
2385 goto unlock;
2386
2387 unix_state_unlock(sk);
2388 if (!timeo) {
2389 err = -EAGAIN;
2390 break;
2391 }
2392
2393 mutex_unlock(&u->iolock);
2394
2395 timeo = unix_stream_data_wait(sk, timeo, last,
2396 last_len, freezable);
2397
2398 if (signal_pending(current)) {
2399 err = sock_intr_errno(timeo);
2400 scm_destroy(&scm);
2401 goto out;
2402 }
2403
2404 mutex_lock(&u->iolock);
2405 goto redo;
2406unlock:
2407 unix_state_unlock(sk);
2408 break;
2409 }
2410
2411 while (skip >= unix_skb_len(skb)) {
2412 skip -= unix_skb_len(skb);
2413 last = skb;
2414 last_len = skb->len;
2415 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2416 if (!skb)
2417 goto again;
2418 }
2419
2420 unix_state_unlock(sk);
2421
2422 if (check_creds) {
2423 /* Never glue messages from different writers */
2424 if (!unix_skb_scm_eq(skb, &scm))
2425 break;
2426 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2427 /* Copy credentials */
2428 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2429 unix_set_secdata(&scm, skb);
2430 check_creds = true;
2431 }
2432
2433 /* Copy address just once */
2434 if (state->msg && state->msg->msg_name) {
2435 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2436 state->msg->msg_name);
2437 unix_copy_addr(state->msg, skb->sk);
2438 sunaddr = NULL;
2439 }
2440
2441 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2442 skb_get(skb);
2443 chunk = state->recv_actor(skb, skip, chunk, state);
2444 drop_skb = !unix_skb_len(skb);
2445 /* skb is only safe to use if !drop_skb */
2446 consume_skb(skb);
2447 if (chunk < 0) {
2448 if (copied == 0)
2449 copied = -EFAULT;
2450 break;
2451 }
2452 copied += chunk;
2453 size -= chunk;
2454
2455 if (drop_skb) {
2456 /* the skb was touched by a concurrent reader;
2457 * we should not expect anything from this skb
2458 * anymore and assume it invalid - we can be
2459 * sure it was dropped from the socket queue
2460 *
2461 * let's report a short read
2462 */
2463 err = 0;
2464 break;
2465 }
2466
2467 /* Mark read part of skb as used */
2468 if (!(flags & MSG_PEEK)) {
2469 UNIXCB(skb).consumed += chunk;
2470
2471 sk_peek_offset_bwd(sk, chunk);
2472
2473 if (UNIXCB(skb).fp) {
2474 scm_stat_del(sk, skb);
2475 unix_detach_fds(&scm, skb);
2476 }
2477
2478 if (unix_skb_len(skb))
2479 break;
2480
2481 skb_unlink(skb, &sk->sk_receive_queue);
2482 consume_skb(skb);
2483
2484 if (scm.fp)
2485 break;
2486 } else {
2487 /* It is questionable, see note in unix_dgram_recvmsg.
2488 */
2489 if (UNIXCB(skb).fp)
2490 unix_peek_fds(&scm, skb);
2491
2492 sk_peek_offset_fwd(sk, chunk);
2493
2494 if (UNIXCB(skb).fp)
2495 break;
2496
2497 skip = 0;
2498 last = skb;
2499 last_len = skb->len;
2500 unix_state_lock(sk);
2501 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2502 if (skb)
2503 goto again;
2504 unix_state_unlock(sk);
2505 break;
2506 }
2507 } while (size);
2508
2509 mutex_unlock(&u->iolock);
2510 if (state->msg)
2511 scm_recv(sock, state->msg, &scm, flags);
2512 else
2513 scm_destroy(&scm);
2514out:
2515 return copied ? : err;
2516}
2517
2518static int unix_stream_read_actor(struct sk_buff *skb,
2519 int skip, int chunk,
2520 struct unix_stream_read_state *state)
2521{
2522 int ret;
2523
2524 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2525 state->msg, chunk);
2526 return ret ?: chunk;
2527}
2528
2529static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2530 size_t size, int flags)
2531{
2532 struct unix_stream_read_state state = {
2533 .recv_actor = unix_stream_read_actor,
2534 .socket = sock,
2535 .msg = msg,
2536 .size = size,
2537 .flags = flags
2538 };
2539
2540 return unix_stream_read_generic(&state, true);
2541}
2542
2543static int unix_stream_splice_actor(struct sk_buff *skb,
2544 int skip, int chunk,
2545 struct unix_stream_read_state *state)
2546{
2547 return skb_splice_bits(skb, state->socket->sk,
2548 UNIXCB(skb).consumed + skip,
2549 state->pipe, chunk, state->splice_flags);
2550}
2551
2552static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2553 struct pipe_inode_info *pipe,
2554 size_t size, unsigned int flags)
2555{
2556 struct unix_stream_read_state state = {
2557 .recv_actor = unix_stream_splice_actor,
2558 .socket = sock,
2559 .pipe = pipe,
2560 .size = size,
2561 .splice_flags = flags,
2562 };
2563
2564 if (unlikely(*ppos))
2565 return -ESPIPE;
2566
2567 if (sock->file->f_flags & O_NONBLOCK ||
2568 flags & SPLICE_F_NONBLOCK)
2569 state.flags = MSG_DONTWAIT;
2570
2571 return unix_stream_read_generic(&state, false);
2572}
2573
2574static int unix_shutdown(struct socket *sock, int mode)
2575{
2576 struct sock *sk = sock->sk;
2577 struct sock *other;
2578
2579 if (mode < SHUT_RD || mode > SHUT_RDWR)
2580 return -EINVAL;
2581 /* This maps:
2582 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2583 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2584 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2585 */
2586 ++mode;
2587
2588 unix_state_lock(sk);
2589 sk->sk_shutdown |= mode;
2590 other = unix_peer(sk);
2591 if (other)
2592 sock_hold(other);
2593 unix_state_unlock(sk);
2594 sk->sk_state_change(sk);
2595
2596 if (other &&
2597 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2598
2599 int peer_mode = 0;
2600
2601 if (mode&RCV_SHUTDOWN)
2602 peer_mode |= SEND_SHUTDOWN;
2603 if (mode&SEND_SHUTDOWN)
2604 peer_mode |= RCV_SHUTDOWN;
2605 unix_state_lock(other);
2606 other->sk_shutdown |= peer_mode;
2607 unix_state_unlock(other);
2608 other->sk_state_change(other);
2609 if (peer_mode == SHUTDOWN_MASK)
2610 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2611 else if (peer_mode & RCV_SHUTDOWN)
2612 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2613 }
2614 if (other)
2615 sock_put(other);
2616
2617 return 0;
2618}
2619
2620long unix_inq_len(struct sock *sk)
2621{
2622 struct sk_buff *skb;
2623 long amount = 0;
2624
2625 if (sk->sk_state == TCP_LISTEN)
2626 return -EINVAL;
2627
2628 spin_lock(&sk->sk_receive_queue.lock);
2629 if (sk->sk_type == SOCK_STREAM ||
2630 sk->sk_type == SOCK_SEQPACKET) {
2631 skb_queue_walk(&sk->sk_receive_queue, skb)
2632 amount += unix_skb_len(skb);
2633 } else {
2634 skb = skb_peek(&sk->sk_receive_queue);
2635 if (skb)
2636 amount = skb->len;
2637 }
2638 spin_unlock(&sk->sk_receive_queue.lock);
2639
2640 return amount;
2641}
2642EXPORT_SYMBOL_GPL(unix_inq_len);
2643
2644long unix_outq_len(struct sock *sk)
2645{
2646 return sk_wmem_alloc_get(sk);
2647}
2648EXPORT_SYMBOL_GPL(unix_outq_len);
2649
2650static int unix_open_file(struct sock *sk)
2651{
2652 struct path path;
2653 struct file *f;
2654 int fd;
2655
2656 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2657 return -EPERM;
2658
2659 if (!smp_load_acquire(&unix_sk(sk)->addr))
2660 return -ENOENT;
2661
2662 path = unix_sk(sk)->path;
2663 if (!path.dentry)
2664 return -ENOENT;
2665
2666 path_get(&path);
2667
2668 fd = get_unused_fd_flags(O_CLOEXEC);
2669 if (fd < 0)
2670 goto out;
2671
2672 f = dentry_open(&path, O_PATH, current_cred());
2673 if (IS_ERR(f)) {
2674 put_unused_fd(fd);
2675 fd = PTR_ERR(f);
2676 goto out;
2677 }
2678
2679 fd_install(fd, f);
2680out:
2681 path_put(&path);
2682
2683 return fd;
2684}
2685
2686static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2687{
2688 struct sock *sk = sock->sk;
2689 long amount = 0;
2690 int err;
2691
2692 switch (cmd) {
2693 case SIOCOUTQ:
2694 amount = unix_outq_len(sk);
2695 err = put_user(amount, (int __user *)arg);
2696 break;
2697 case SIOCINQ:
2698 amount = unix_inq_len(sk);
2699 if (amount < 0)
2700 err = amount;
2701 else
2702 err = put_user(amount, (int __user *)arg);
2703 break;
2704 case SIOCUNIXFILE:
2705 err = unix_open_file(sk);
2706 break;
2707 default:
2708 err = -ENOIOCTLCMD;
2709 break;
2710 }
2711 return err;
2712}
2713
2714#ifdef CONFIG_COMPAT
2715static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2716{
2717 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2718}
2719#endif
2720
2721static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2722{
2723 struct sock *sk = sock->sk;
2724 __poll_t mask;
2725
2726 sock_poll_wait(file, sock, wait);
2727 mask = 0;
2728
2729 /* exceptional events? */
2730 if (sk->sk_err)
2731 mask |= EPOLLERR;
2732 if (sk->sk_shutdown == SHUTDOWN_MASK)
2733 mask |= EPOLLHUP;
2734 if (sk->sk_shutdown & RCV_SHUTDOWN)
2735 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2736
2737 /* readable? */
2738 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2739 mask |= EPOLLIN | EPOLLRDNORM;
2740
2741 /* Connection-based need to check for termination and startup */
2742 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2743 sk->sk_state == TCP_CLOSE)
2744 mask |= EPOLLHUP;
2745
2746 /*
2747 * we set writable also when the other side has shut down the
2748 * connection. This prevents stuck sockets.
2749 */
2750 if (unix_writable(sk))
2751 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2752
2753 return mask;
2754}
2755
2756static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2757 poll_table *wait)
2758{
2759 struct sock *sk = sock->sk, *other;
2760 unsigned int writable;
2761 __poll_t mask;
2762
2763 sock_poll_wait(file, sock, wait);
2764 mask = 0;
2765
2766 /* exceptional events? */
2767 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2768 mask |= EPOLLERR |
2769 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2770
2771 if (sk->sk_shutdown & RCV_SHUTDOWN)
2772 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2773 if (sk->sk_shutdown == SHUTDOWN_MASK)
2774 mask |= EPOLLHUP;
2775
2776 /* readable? */
2777 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2778 mask |= EPOLLIN | EPOLLRDNORM;
2779
2780 /* Connection-based need to check for termination and startup */
2781 if (sk->sk_type == SOCK_SEQPACKET) {
2782 if (sk->sk_state == TCP_CLOSE)
2783 mask |= EPOLLHUP;
2784 /* connection hasn't started yet? */
2785 if (sk->sk_state == TCP_SYN_SENT)
2786 return mask;
2787 }
2788
2789 /* No write status requested, avoid expensive OUT tests. */
2790 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2791 return mask;
2792
2793 writable = unix_writable(sk);
2794 if (writable) {
2795 unix_state_lock(sk);
2796
2797 other = unix_peer(sk);
2798 if (other && unix_peer(other) != sk &&
2799 unix_recvq_full_lockless(other) &&
2800 unix_dgram_peer_wake_me(sk, other))
2801 writable = 0;
2802
2803 unix_state_unlock(sk);
2804 }
2805
2806 if (writable)
2807 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2808 else
2809 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2810
2811 return mask;
2812}
2813
2814#ifdef CONFIG_PROC_FS
2815
2816#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2817
2818#define get_bucket(x) ((x) >> BUCKET_SPACE)
2819#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2820#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2821
2822static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2823{
2824 unsigned long offset = get_offset(*pos);
2825 unsigned long bucket = get_bucket(*pos);
2826 struct sock *sk;
2827 unsigned long count = 0;
2828
2829 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2830 if (sock_net(sk) != seq_file_net(seq))
2831 continue;
2832 if (++count == offset)
2833 break;
2834 }
2835
2836 return sk;
2837}
2838
2839static struct sock *unix_next_socket(struct seq_file *seq,
2840 struct sock *sk,
2841 loff_t *pos)
2842{
2843 unsigned long bucket;
2844
2845 while (sk > (struct sock *)SEQ_START_TOKEN) {
2846 sk = sk_next(sk);
2847 if (!sk)
2848 goto next_bucket;
2849 if (sock_net(sk) == seq_file_net(seq))
2850 return sk;
2851 }
2852
2853 do {
2854 sk = unix_from_bucket(seq, pos);
2855 if (sk)
2856 return sk;
2857
2858next_bucket:
2859 bucket = get_bucket(*pos) + 1;
2860 *pos = set_bucket_offset(bucket, 1);
2861 } while (bucket < ARRAY_SIZE(unix_socket_table));
2862
2863 return NULL;
2864}
2865
2866static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2867 __acquires(unix_table_lock)
2868{
2869 spin_lock(&unix_table_lock);
2870
2871 if (!*pos)
2872 return SEQ_START_TOKEN;
2873
2874 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2875 return NULL;
2876
2877 return unix_next_socket(seq, NULL, pos);
2878}
2879
2880static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2881{
2882 ++*pos;
2883 return unix_next_socket(seq, v, pos);
2884}
2885
2886static void unix_seq_stop(struct seq_file *seq, void *v)
2887 __releases(unix_table_lock)
2888{
2889 spin_unlock(&unix_table_lock);
2890}
2891
2892static int unix_seq_show(struct seq_file *seq, void *v)
2893{
2894
2895 if (v == SEQ_START_TOKEN)
2896 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2897 "Inode Path\n");
2898 else {
2899 struct sock *s = v;
2900 struct unix_sock *u = unix_sk(s);
2901 unix_state_lock(s);
2902
2903 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2904 s,
2905 refcount_read(&s->sk_refcnt),
2906 0,
2907 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2908 s->sk_type,
2909 s->sk_socket ?
2910 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2911 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2912 sock_i_ino(s));
2913
2914 if (u->addr) { // under unix_table_lock here
2915 int i, len;
2916 seq_putc(seq, ' ');
2917
2918 i = 0;
2919 len = u->addr->len - sizeof(short);
2920 if (!UNIX_ABSTRACT(s))
2921 len--;
2922 else {
2923 seq_putc(seq, '@');
2924 i++;
2925 }
2926 for ( ; i < len; i++)
2927 seq_putc(seq, u->addr->name->sun_path[i] ?:
2928 '@');
2929 }
2930 unix_state_unlock(s);
2931 seq_putc(seq, '\n');
2932 }
2933
2934 return 0;
2935}
2936
2937static const struct seq_operations unix_seq_ops = {
2938 .start = unix_seq_start,
2939 .next = unix_seq_next,
2940 .stop = unix_seq_stop,
2941 .show = unix_seq_show,
2942};
2943#endif
2944
2945static const struct net_proto_family unix_family_ops = {
2946 .family = PF_UNIX,
2947 .create = unix_create,
2948 .owner = THIS_MODULE,
2949};
2950
2951
2952static int __net_init unix_net_init(struct net *net)
2953{
2954 int error = -ENOMEM;
2955
2956 net->unx.sysctl_max_dgram_qlen = 10;
2957 if (unix_sysctl_register(net))
2958 goto out;
2959
2960#ifdef CONFIG_PROC_FS
2961 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2962 sizeof(struct seq_net_private))) {
2963 unix_sysctl_unregister(net);
2964 goto out;
2965 }
2966#endif
2967 error = 0;
2968out:
2969 return error;
2970}
2971
2972static void __net_exit unix_net_exit(struct net *net)
2973{
2974 unix_sysctl_unregister(net);
2975 remove_proc_entry("unix", net->proc_net);
2976}
2977
2978static struct pernet_operations unix_net_ops = {
2979 .init = unix_net_init,
2980 .exit = unix_net_exit,
2981};
2982
2983static int __init af_unix_init(void)
2984{
2985 int rc = -1;
2986
2987 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
2988
2989 rc = proto_register(&unix_proto, 1);
2990 if (rc != 0) {
2991 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2992 goto out;
2993 }
2994
2995 sock_register(&unix_family_ops);
2996 register_pernet_subsys(&unix_net_ops);
2997out:
2998 return rc;
2999}
3000
3001static void __exit af_unix_exit(void)
3002{
3003 sock_unregister(PF_UNIX);
3004 proto_unregister(&unix_proto);
3005 unregister_pernet_subsys(&unix_net_ops);
3006}
3007
3008/* Earlier than device_initcall() so that other drivers invoking
3009 request_module() don't end up in a loop when modprobe tries
3010 to use a UNIX socket. But later than subsys_initcall() because
3011 we depend on stuff initialised there */
3012fs_initcall(af_unix_init);
3013module_exit(af_unix_exit);
3014
3015MODULE_LICENSE("GPL");
3016MODULE_ALIAS_NETPROTO(PF_UNIX);