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/filter.h>
93#include <linux/termios.h>
94#include <linux/sockios.h>
95#include <linux/net.h>
96#include <linux/in.h>
97#include <linux/fs.h>
98#include <linux/slab.h>
99#include <linux/uaccess.h>
100#include <linux/skbuff.h>
101#include <linux/netdevice.h>
102#include <net/net_namespace.h>
103#include <net/sock.h>
104#include <net/tcp_states.h>
105#include <net/af_unix.h>
106#include <linux/proc_fs.h>
107#include <linux/seq_file.h>
108#include <net/scm.h>
109#include <linux/init.h>
110#include <linux/poll.h>
111#include <linux/rtnetlink.h>
112#include <linux/mount.h>
113#include <net/checksum.h>
114#include <linux/security.h>
115#include <linux/splice.h>
116#include <linux/freezer.h>
117#include <linux/file.h>
118#include <linux/btf_ids.h>
119#include <linux/bpf-cgroup.h>
120
121static atomic_long_t unix_nr_socks;
122static struct hlist_head bsd_socket_buckets[UNIX_HASH_SIZE / 2];
123static spinlock_t bsd_socket_locks[UNIX_HASH_SIZE / 2];
124
125/* SMP locking strategy:
126 * hash table is protected with spinlock.
127 * each socket state is protected by separate spinlock.
128 */
129
130static unsigned int unix_unbound_hash(struct sock *sk)
131{
132 unsigned long hash = (unsigned long)sk;
133
134 hash ^= hash >> 16;
135 hash ^= hash >> 8;
136 hash ^= sk->sk_type;
137
138 return hash & UNIX_HASH_MOD;
139}
140
141static unsigned int unix_bsd_hash(struct inode *i)
142{
143 return i->i_ino & UNIX_HASH_MOD;
144}
145
146static unsigned int unix_abstract_hash(struct sockaddr_un *sunaddr,
147 int addr_len, int type)
148{
149 __wsum csum = csum_partial(sunaddr, addr_len, 0);
150 unsigned int hash;
151
152 hash = (__force unsigned int)csum_fold(csum);
153 hash ^= hash >> 8;
154 hash ^= type;
155
156 return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
157}
158
159static void unix_table_double_lock(struct net *net,
160 unsigned int hash1, unsigned int hash2)
161{
162 if (hash1 == hash2) {
163 spin_lock(&net->unx.table.locks[hash1]);
164 return;
165 }
166
167 if (hash1 > hash2)
168 swap(hash1, hash2);
169
170 spin_lock(&net->unx.table.locks[hash1]);
171 spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
172}
173
174static void unix_table_double_unlock(struct net *net,
175 unsigned int hash1, unsigned int hash2)
176{
177 if (hash1 == hash2) {
178 spin_unlock(&net->unx.table.locks[hash1]);
179 return;
180 }
181
182 spin_unlock(&net->unx.table.locks[hash1]);
183 spin_unlock(&net->unx.table.locks[hash2]);
184}
185
186#ifdef CONFIG_SECURITY_NETWORK
187static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
188{
189 UNIXCB(skb).secid = scm->secid;
190}
191
192static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
193{
194 scm->secid = UNIXCB(skb).secid;
195}
196
197static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
198{
199 return (scm->secid == UNIXCB(skb).secid);
200}
201#else
202static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
203{ }
204
205static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
206{ }
207
208static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
209{
210 return true;
211}
212#endif /* CONFIG_SECURITY_NETWORK */
213
214static inline int unix_our_peer(struct sock *sk, struct sock *osk)
215{
216 return unix_peer(osk) == sk;
217}
218
219static inline int unix_may_send(struct sock *sk, struct sock *osk)
220{
221 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
222}
223
224static inline int unix_recvq_full(const struct sock *sk)
225{
226 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
227}
228
229static inline int unix_recvq_full_lockless(const struct sock *sk)
230{
231 return skb_queue_len_lockless(&sk->sk_receive_queue) >
232 READ_ONCE(sk->sk_max_ack_backlog);
233}
234
235struct sock *unix_peer_get(struct sock *s)
236{
237 struct sock *peer;
238
239 unix_state_lock(s);
240 peer = unix_peer(s);
241 if (peer)
242 sock_hold(peer);
243 unix_state_unlock(s);
244 return peer;
245}
246EXPORT_SYMBOL_GPL(unix_peer_get);
247
248static struct unix_address *unix_create_addr(struct sockaddr_un *sunaddr,
249 int addr_len)
250{
251 struct unix_address *addr;
252
253 addr = kmalloc(sizeof(*addr) + addr_len, GFP_KERNEL);
254 if (!addr)
255 return NULL;
256
257 refcount_set(&addr->refcnt, 1);
258 addr->len = addr_len;
259 memcpy(addr->name, sunaddr, addr_len);
260
261 return addr;
262}
263
264static inline void unix_release_addr(struct unix_address *addr)
265{
266 if (refcount_dec_and_test(&addr->refcnt))
267 kfree(addr);
268}
269
270/*
271 * Check unix socket name:
272 * - should be not zero length.
273 * - if started by not zero, should be NULL terminated (FS object)
274 * - if started by zero, it is abstract name.
275 */
276
277static int unix_validate_addr(struct sockaddr_un *sunaddr, int addr_len)
278{
279 if (addr_len <= offsetof(struct sockaddr_un, sun_path) ||
280 addr_len > sizeof(*sunaddr))
281 return -EINVAL;
282
283 if (sunaddr->sun_family != AF_UNIX)
284 return -EINVAL;
285
286 return 0;
287}
288
289static int unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
290{
291 struct sockaddr_storage *addr = (struct sockaddr_storage *)sunaddr;
292 short offset = offsetof(struct sockaddr_storage, __data);
293
294 BUILD_BUG_ON(offset != offsetof(struct sockaddr_un, sun_path));
295
296 /* This may look like an off by one error but it is a bit more
297 * subtle. 108 is the longest valid AF_UNIX path for a binding.
298 * sun_path[108] doesn't as such exist. However in kernel space
299 * we are guaranteed that it is a valid memory location in our
300 * kernel address buffer because syscall functions always pass
301 * a pointer of struct sockaddr_storage which has a bigger buffer
302 * than 108. Also, we must terminate sun_path for strlen() in
303 * getname_kernel().
304 */
305 addr->__data[addr_len - offset] = 0;
306
307 /* Don't pass sunaddr->sun_path to strlen(). Otherwise, 108 will
308 * cause panic if CONFIG_FORTIFY_SOURCE=y. Let __fortify_strlen()
309 * know the actual buffer.
310 */
311 return strlen(addr->__data) + offset + 1;
312}
313
314static void __unix_remove_socket(struct sock *sk)
315{
316 sk_del_node_init(sk);
317}
318
319static void __unix_insert_socket(struct net *net, struct sock *sk)
320{
321 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
322 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
323}
324
325static void __unix_set_addr_hash(struct net *net, struct sock *sk,
326 struct unix_address *addr, unsigned int hash)
327{
328 __unix_remove_socket(sk);
329 smp_store_release(&unix_sk(sk)->addr, addr);
330
331 sk->sk_hash = hash;
332 __unix_insert_socket(net, sk);
333}
334
335static void unix_remove_socket(struct net *net, struct sock *sk)
336{
337 spin_lock(&net->unx.table.locks[sk->sk_hash]);
338 __unix_remove_socket(sk);
339 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
340}
341
342static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
343{
344 spin_lock(&net->unx.table.locks[sk->sk_hash]);
345 __unix_insert_socket(net, sk);
346 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
347}
348
349static void unix_insert_bsd_socket(struct sock *sk)
350{
351 spin_lock(&bsd_socket_locks[sk->sk_hash]);
352 sk_add_bind_node(sk, &bsd_socket_buckets[sk->sk_hash]);
353 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
354}
355
356static void unix_remove_bsd_socket(struct sock *sk)
357{
358 if (!hlist_unhashed(&sk->sk_bind_node)) {
359 spin_lock(&bsd_socket_locks[sk->sk_hash]);
360 __sk_del_bind_node(sk);
361 spin_unlock(&bsd_socket_locks[sk->sk_hash]);
362
363 sk_node_init(&sk->sk_bind_node);
364 }
365}
366
367static struct sock *__unix_find_socket_byname(struct net *net,
368 struct sockaddr_un *sunname,
369 int len, unsigned int hash)
370{
371 struct sock *s;
372
373 sk_for_each(s, &net->unx.table.buckets[hash]) {
374 struct unix_sock *u = unix_sk(s);
375
376 if (u->addr->len == len &&
377 !memcmp(u->addr->name, sunname, len))
378 return s;
379 }
380 return NULL;
381}
382
383static inline struct sock *unix_find_socket_byname(struct net *net,
384 struct sockaddr_un *sunname,
385 int len, unsigned int hash)
386{
387 struct sock *s;
388
389 spin_lock(&net->unx.table.locks[hash]);
390 s = __unix_find_socket_byname(net, sunname, len, hash);
391 if (s)
392 sock_hold(s);
393 spin_unlock(&net->unx.table.locks[hash]);
394 return s;
395}
396
397static struct sock *unix_find_socket_byinode(struct inode *i)
398{
399 unsigned int hash = unix_bsd_hash(i);
400 struct sock *s;
401
402 spin_lock(&bsd_socket_locks[hash]);
403 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
404 struct dentry *dentry = unix_sk(s)->path.dentry;
405
406 if (dentry && d_backing_inode(dentry) == i) {
407 sock_hold(s);
408 spin_unlock(&bsd_socket_locks[hash]);
409 return s;
410 }
411 }
412 spin_unlock(&bsd_socket_locks[hash]);
413 return NULL;
414}
415
416/* Support code for asymmetrically connected dgram sockets
417 *
418 * If a datagram socket is connected to a socket not itself connected
419 * to the first socket (eg, /dev/log), clients may only enqueue more
420 * messages if the present receive queue of the server socket is not
421 * "too large". This means there's a second writeability condition
422 * poll and sendmsg need to test. The dgram recv code will do a wake
423 * up on the peer_wait wait queue of a socket upon reception of a
424 * datagram which needs to be propagated to sleeping would-be writers
425 * since these might not have sent anything so far. This can't be
426 * accomplished via poll_wait because the lifetime of the server
427 * socket might be less than that of its clients if these break their
428 * association with it or if the server socket is closed while clients
429 * are still connected to it and there's no way to inform "a polling
430 * implementation" that it should let go of a certain wait queue
431 *
432 * In order to propagate a wake up, a wait_queue_entry_t of the client
433 * socket is enqueued on the peer_wait queue of the server socket
434 * whose wake function does a wake_up on the ordinary client socket
435 * wait queue. This connection is established whenever a write (or
436 * poll for write) hit the flow control condition and broken when the
437 * association to the server socket is dissolved or after a wake up
438 * was relayed.
439 */
440
441static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
442 void *key)
443{
444 struct unix_sock *u;
445 wait_queue_head_t *u_sleep;
446
447 u = container_of(q, struct unix_sock, peer_wake);
448
449 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
450 q);
451 u->peer_wake.private = NULL;
452
453 /* relaying can only happen while the wq still exists */
454 u_sleep = sk_sleep(&u->sk);
455 if (u_sleep)
456 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
457
458 return 0;
459}
460
461static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
462{
463 struct unix_sock *u, *u_other;
464 int rc;
465
466 u = unix_sk(sk);
467 u_other = unix_sk(other);
468 rc = 0;
469 spin_lock(&u_other->peer_wait.lock);
470
471 if (!u->peer_wake.private) {
472 u->peer_wake.private = other;
473 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
474
475 rc = 1;
476 }
477
478 spin_unlock(&u_other->peer_wait.lock);
479 return rc;
480}
481
482static void unix_dgram_peer_wake_disconnect(struct sock *sk,
483 struct sock *other)
484{
485 struct unix_sock *u, *u_other;
486
487 u = unix_sk(sk);
488 u_other = unix_sk(other);
489 spin_lock(&u_other->peer_wait.lock);
490
491 if (u->peer_wake.private == other) {
492 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
493 u->peer_wake.private = NULL;
494 }
495
496 spin_unlock(&u_other->peer_wait.lock);
497}
498
499static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
500 struct sock *other)
501{
502 unix_dgram_peer_wake_disconnect(sk, other);
503 wake_up_interruptible_poll(sk_sleep(sk),
504 EPOLLOUT |
505 EPOLLWRNORM |
506 EPOLLWRBAND);
507}
508
509/* preconditions:
510 * - unix_peer(sk) == other
511 * - association is stable
512 */
513static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
514{
515 int connected;
516
517 connected = unix_dgram_peer_wake_connect(sk, other);
518
519 /* If other is SOCK_DEAD, we want to make sure we signal
520 * POLLOUT, such that a subsequent write() can get a
521 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
522 * to other and its full, we will hang waiting for POLLOUT.
523 */
524 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
525 return 1;
526
527 if (connected)
528 unix_dgram_peer_wake_disconnect(sk, other);
529
530 return 0;
531}
532
533static int unix_writable(const struct sock *sk)
534{
535 return sk->sk_state != TCP_LISTEN &&
536 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
537}
538
539static void unix_write_space(struct sock *sk)
540{
541 struct socket_wq *wq;
542
543 rcu_read_lock();
544 if (unix_writable(sk)) {
545 wq = rcu_dereference(sk->sk_wq);
546 if (skwq_has_sleeper(wq))
547 wake_up_interruptible_sync_poll(&wq->wait,
548 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
549 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
550 }
551 rcu_read_unlock();
552}
553
554/* When dgram socket disconnects (or changes its peer), we clear its receive
555 * queue of packets arrived from previous peer. First, it allows to do
556 * flow control based only on wmem_alloc; second, sk connected to peer
557 * may receive messages only from that peer. */
558static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
559{
560 if (!skb_queue_empty(&sk->sk_receive_queue)) {
561 skb_queue_purge(&sk->sk_receive_queue);
562 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
563
564 /* If one link of bidirectional dgram pipe is disconnected,
565 * we signal error. Messages are lost. Do not make this,
566 * when peer was not connected to us.
567 */
568 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
569 WRITE_ONCE(other->sk_err, ECONNRESET);
570 sk_error_report(other);
571 }
572 }
573 other->sk_state = TCP_CLOSE;
574}
575
576static void unix_sock_destructor(struct sock *sk)
577{
578 struct unix_sock *u = unix_sk(sk);
579
580 skb_queue_purge(&sk->sk_receive_queue);
581
582 DEBUG_NET_WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
583 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
584 DEBUG_NET_WARN_ON_ONCE(sk->sk_socket);
585 if (!sock_flag(sk, SOCK_DEAD)) {
586 pr_info("Attempt to release alive unix socket: %p\n", sk);
587 return;
588 }
589
590 if (u->addr)
591 unix_release_addr(u->addr);
592
593 atomic_long_dec(&unix_nr_socks);
594 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
595#ifdef UNIX_REFCNT_DEBUG
596 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
597 atomic_long_read(&unix_nr_socks));
598#endif
599}
600
601static void unix_release_sock(struct sock *sk, int embrion)
602{
603 struct unix_sock *u = unix_sk(sk);
604 struct sock *skpair;
605 struct sk_buff *skb;
606 struct path path;
607 int state;
608
609 unix_remove_socket(sock_net(sk), sk);
610 unix_remove_bsd_socket(sk);
611
612 /* Clear state */
613 unix_state_lock(sk);
614 sock_orphan(sk);
615 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
616 path = u->path;
617 u->path.dentry = NULL;
618 u->path.mnt = NULL;
619 state = sk->sk_state;
620 sk->sk_state = TCP_CLOSE;
621
622 skpair = unix_peer(sk);
623 unix_peer(sk) = NULL;
624
625 unix_state_unlock(sk);
626
627#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
628 if (u->oob_skb) {
629 kfree_skb(u->oob_skb);
630 u->oob_skb = NULL;
631 }
632#endif
633
634 wake_up_interruptible_all(&u->peer_wait);
635
636 if (skpair != NULL) {
637 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
638 unix_state_lock(skpair);
639 /* No more writes */
640 WRITE_ONCE(skpair->sk_shutdown, SHUTDOWN_MASK);
641 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
642 WRITE_ONCE(skpair->sk_err, ECONNRESET);
643 unix_state_unlock(skpair);
644 skpair->sk_state_change(skpair);
645 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
646 }
647
648 unix_dgram_peer_wake_disconnect(sk, skpair);
649 sock_put(skpair); /* It may now die */
650 }
651
652 /* Try to flush out this socket. Throw out buffers at least */
653
654 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
655 if (state == TCP_LISTEN)
656 unix_release_sock(skb->sk, 1);
657 /* passed fds are erased in the kfree_skb hook */
658 UNIXCB(skb).consumed = skb->len;
659 kfree_skb(skb);
660 }
661
662 if (path.dentry)
663 path_put(&path);
664
665 sock_put(sk);
666
667 /* ---- Socket is dead now and most probably destroyed ---- */
668
669 /*
670 * Fixme: BSD difference: In BSD all sockets connected to us get
671 * ECONNRESET and we die on the spot. In Linux we behave
672 * like files and pipes do and wait for the last
673 * dereference.
674 *
675 * Can't we simply set sock->err?
676 *
677 * What the above comment does talk about? --ANK(980817)
678 */
679
680 if (READ_ONCE(unix_tot_inflight))
681 unix_gc(); /* Garbage collect fds */
682}
683
684static void init_peercred(struct sock *sk)
685{
686 const struct cred *old_cred;
687 struct pid *old_pid;
688
689 spin_lock(&sk->sk_peer_lock);
690 old_pid = sk->sk_peer_pid;
691 old_cred = sk->sk_peer_cred;
692 sk->sk_peer_pid = get_pid(task_tgid(current));
693 sk->sk_peer_cred = get_current_cred();
694 spin_unlock(&sk->sk_peer_lock);
695
696 put_pid(old_pid);
697 put_cred(old_cred);
698}
699
700static void copy_peercred(struct sock *sk, struct sock *peersk)
701{
702 const struct cred *old_cred;
703 struct pid *old_pid;
704
705 if (sk < peersk) {
706 spin_lock(&sk->sk_peer_lock);
707 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
708 } else {
709 spin_lock(&peersk->sk_peer_lock);
710 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
711 }
712 old_pid = sk->sk_peer_pid;
713 old_cred = sk->sk_peer_cred;
714 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
715 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
716
717 spin_unlock(&sk->sk_peer_lock);
718 spin_unlock(&peersk->sk_peer_lock);
719
720 put_pid(old_pid);
721 put_cred(old_cred);
722}
723
724static int unix_listen(struct socket *sock, int backlog)
725{
726 int err;
727 struct sock *sk = sock->sk;
728 struct unix_sock *u = unix_sk(sk);
729
730 err = -EOPNOTSUPP;
731 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
732 goto out; /* Only stream/seqpacket sockets accept */
733 err = -EINVAL;
734 if (!READ_ONCE(u->addr))
735 goto out; /* No listens on an unbound socket */
736 unix_state_lock(sk);
737 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
738 goto out_unlock;
739 if (backlog > sk->sk_max_ack_backlog)
740 wake_up_interruptible_all(&u->peer_wait);
741 sk->sk_max_ack_backlog = backlog;
742 sk->sk_state = TCP_LISTEN;
743 /* set credentials so connect can copy them */
744 init_peercred(sk);
745 err = 0;
746
747out_unlock:
748 unix_state_unlock(sk);
749out:
750 return err;
751}
752
753static int unix_release(struct socket *);
754static int unix_bind(struct socket *, struct sockaddr *, int);
755static int unix_stream_connect(struct socket *, struct sockaddr *,
756 int addr_len, int flags);
757static int unix_socketpair(struct socket *, struct socket *);
758static int unix_accept(struct socket *, struct socket *, int, bool);
759static int unix_getname(struct socket *, struct sockaddr *, int);
760static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
761static __poll_t unix_dgram_poll(struct file *, struct socket *,
762 poll_table *);
763static int unix_ioctl(struct socket *, unsigned int, unsigned long);
764#ifdef CONFIG_COMPAT
765static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
766#endif
767static int unix_shutdown(struct socket *, int);
768static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
769static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
770static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
771 struct pipe_inode_info *, size_t size,
772 unsigned int flags);
773static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
774static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
775static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
776static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
777static int unix_dgram_connect(struct socket *, struct sockaddr *,
778 int, int);
779static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
780static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
781 int);
782
783#ifdef CONFIG_PROC_FS
784static int unix_count_nr_fds(struct sock *sk)
785{
786 struct sk_buff *skb;
787 struct unix_sock *u;
788 int nr_fds = 0;
789
790 spin_lock(&sk->sk_receive_queue.lock);
791 skb = skb_peek(&sk->sk_receive_queue);
792 while (skb) {
793 u = unix_sk(skb->sk);
794 nr_fds += atomic_read(&u->scm_stat.nr_fds);
795 skb = skb_peek_next(skb, &sk->sk_receive_queue);
796 }
797 spin_unlock(&sk->sk_receive_queue.lock);
798
799 return nr_fds;
800}
801
802static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
803{
804 struct sock *sk = sock->sk;
805 unsigned char s_state;
806 struct unix_sock *u;
807 int nr_fds = 0;
808
809 if (sk) {
810 s_state = READ_ONCE(sk->sk_state);
811 u = unix_sk(sk);
812
813 /* SOCK_STREAM and SOCK_SEQPACKET sockets never change their
814 * sk_state after switching to TCP_ESTABLISHED or TCP_LISTEN.
815 * SOCK_DGRAM is ordinary. So, no lock is needed.
816 */
817 if (sock->type == SOCK_DGRAM || s_state == TCP_ESTABLISHED)
818 nr_fds = atomic_read(&u->scm_stat.nr_fds);
819 else if (s_state == TCP_LISTEN)
820 nr_fds = unix_count_nr_fds(sk);
821
822 seq_printf(m, "scm_fds: %u\n", nr_fds);
823 }
824}
825#else
826#define unix_show_fdinfo NULL
827#endif
828
829static const struct proto_ops unix_stream_ops = {
830 .family = PF_UNIX,
831 .owner = THIS_MODULE,
832 .release = unix_release,
833 .bind = unix_bind,
834 .connect = unix_stream_connect,
835 .socketpair = unix_socketpair,
836 .accept = unix_accept,
837 .getname = unix_getname,
838 .poll = unix_poll,
839 .ioctl = unix_ioctl,
840#ifdef CONFIG_COMPAT
841 .compat_ioctl = unix_compat_ioctl,
842#endif
843 .listen = unix_listen,
844 .shutdown = unix_shutdown,
845 .sendmsg = unix_stream_sendmsg,
846 .recvmsg = unix_stream_recvmsg,
847 .read_skb = unix_stream_read_skb,
848 .mmap = sock_no_mmap,
849 .splice_read = unix_stream_splice_read,
850 .set_peek_off = sk_set_peek_off,
851 .show_fdinfo = unix_show_fdinfo,
852};
853
854static const struct proto_ops unix_dgram_ops = {
855 .family = PF_UNIX,
856 .owner = THIS_MODULE,
857 .release = unix_release,
858 .bind = unix_bind,
859 .connect = unix_dgram_connect,
860 .socketpair = unix_socketpair,
861 .accept = sock_no_accept,
862 .getname = unix_getname,
863 .poll = unix_dgram_poll,
864 .ioctl = unix_ioctl,
865#ifdef CONFIG_COMPAT
866 .compat_ioctl = unix_compat_ioctl,
867#endif
868 .listen = sock_no_listen,
869 .shutdown = unix_shutdown,
870 .sendmsg = unix_dgram_sendmsg,
871 .read_skb = unix_read_skb,
872 .recvmsg = unix_dgram_recvmsg,
873 .mmap = sock_no_mmap,
874 .set_peek_off = sk_set_peek_off,
875 .show_fdinfo = unix_show_fdinfo,
876};
877
878static const struct proto_ops unix_seqpacket_ops = {
879 .family = PF_UNIX,
880 .owner = THIS_MODULE,
881 .release = unix_release,
882 .bind = unix_bind,
883 .connect = unix_stream_connect,
884 .socketpair = unix_socketpair,
885 .accept = unix_accept,
886 .getname = unix_getname,
887 .poll = unix_dgram_poll,
888 .ioctl = unix_ioctl,
889#ifdef CONFIG_COMPAT
890 .compat_ioctl = unix_compat_ioctl,
891#endif
892 .listen = unix_listen,
893 .shutdown = unix_shutdown,
894 .sendmsg = unix_seqpacket_sendmsg,
895 .recvmsg = unix_seqpacket_recvmsg,
896 .mmap = sock_no_mmap,
897 .set_peek_off = sk_set_peek_off,
898 .show_fdinfo = unix_show_fdinfo,
899};
900
901static void unix_close(struct sock *sk, long timeout)
902{
903 /* Nothing to do here, unix socket does not need a ->close().
904 * This is merely for sockmap.
905 */
906}
907
908static void unix_unhash(struct sock *sk)
909{
910 /* Nothing to do here, unix socket does not need a ->unhash().
911 * This is merely for sockmap.
912 */
913}
914
915static bool unix_bpf_bypass_getsockopt(int level, int optname)
916{
917 if (level == SOL_SOCKET) {
918 switch (optname) {
919 case SO_PEERPIDFD:
920 return true;
921 default:
922 return false;
923 }
924 }
925
926 return false;
927}
928
929struct proto unix_dgram_proto = {
930 .name = "UNIX",
931 .owner = THIS_MODULE,
932 .obj_size = sizeof(struct unix_sock),
933 .close = unix_close,
934 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
935#ifdef CONFIG_BPF_SYSCALL
936 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
937#endif
938};
939
940struct proto unix_stream_proto = {
941 .name = "UNIX-STREAM",
942 .owner = THIS_MODULE,
943 .obj_size = sizeof(struct unix_sock),
944 .close = unix_close,
945 .unhash = unix_unhash,
946 .bpf_bypass_getsockopt = unix_bpf_bypass_getsockopt,
947#ifdef CONFIG_BPF_SYSCALL
948 .psock_update_sk_prot = unix_stream_bpf_update_proto,
949#endif
950};
951
952static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
953{
954 struct unix_sock *u;
955 struct sock *sk;
956 int err;
957
958 atomic_long_inc(&unix_nr_socks);
959 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
960 err = -ENFILE;
961 goto err;
962 }
963
964 if (type == SOCK_STREAM)
965 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_stream_proto, kern);
966 else /*dgram and seqpacket */
967 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_dgram_proto, kern);
968
969 if (!sk) {
970 err = -ENOMEM;
971 goto err;
972 }
973
974 sock_init_data(sock, sk);
975
976 sk->sk_hash = unix_unbound_hash(sk);
977 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
978 sk->sk_write_space = unix_write_space;
979 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
980 sk->sk_destruct = unix_sock_destructor;
981 u = unix_sk(sk);
982 u->inflight = 0;
983 u->path.dentry = NULL;
984 u->path.mnt = NULL;
985 spin_lock_init(&u->lock);
986 INIT_LIST_HEAD(&u->link);
987 mutex_init(&u->iolock); /* single task reading lock */
988 mutex_init(&u->bindlock); /* single task binding lock */
989 init_waitqueue_head(&u->peer_wait);
990 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
991 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
992 unix_insert_unbound_socket(net, sk);
993
994 sock_prot_inuse_add(net, sk->sk_prot, 1);
995
996 return sk;
997
998err:
999 atomic_long_dec(&unix_nr_socks);
1000 return ERR_PTR(err);
1001}
1002
1003static int unix_create(struct net *net, struct socket *sock, int protocol,
1004 int kern)
1005{
1006 struct sock *sk;
1007
1008 if (protocol && protocol != PF_UNIX)
1009 return -EPROTONOSUPPORT;
1010
1011 sock->state = SS_UNCONNECTED;
1012
1013 switch (sock->type) {
1014 case SOCK_STREAM:
1015 sock->ops = &unix_stream_ops;
1016 break;
1017 /*
1018 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
1019 * nothing uses it.
1020 */
1021 case SOCK_RAW:
1022 sock->type = SOCK_DGRAM;
1023 fallthrough;
1024 case SOCK_DGRAM:
1025 sock->ops = &unix_dgram_ops;
1026 break;
1027 case SOCK_SEQPACKET:
1028 sock->ops = &unix_seqpacket_ops;
1029 break;
1030 default:
1031 return -ESOCKTNOSUPPORT;
1032 }
1033
1034 sk = unix_create1(net, sock, kern, sock->type);
1035 if (IS_ERR(sk))
1036 return PTR_ERR(sk);
1037
1038 return 0;
1039}
1040
1041static int unix_release(struct socket *sock)
1042{
1043 struct sock *sk = sock->sk;
1044
1045 if (!sk)
1046 return 0;
1047
1048 sk->sk_prot->close(sk, 0);
1049 unix_release_sock(sk, 0);
1050 sock->sk = NULL;
1051
1052 return 0;
1053}
1054
1055static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1056 int type)
1057{
1058 struct inode *inode;
1059 struct path path;
1060 struct sock *sk;
1061 int err;
1062
1063 unix_mkname_bsd(sunaddr, addr_len);
1064 err = kern_path(sunaddr->sun_path, LOOKUP_FOLLOW, &path);
1065 if (err)
1066 goto fail;
1067
1068 err = path_permission(&path, MAY_WRITE);
1069 if (err)
1070 goto path_put;
1071
1072 err = -ECONNREFUSED;
1073 inode = d_backing_inode(path.dentry);
1074 if (!S_ISSOCK(inode->i_mode))
1075 goto path_put;
1076
1077 sk = unix_find_socket_byinode(inode);
1078 if (!sk)
1079 goto path_put;
1080
1081 err = -EPROTOTYPE;
1082 if (sk->sk_type == type)
1083 touch_atime(&path);
1084 else
1085 goto sock_put;
1086
1087 path_put(&path);
1088
1089 return sk;
1090
1091sock_put:
1092 sock_put(sk);
1093path_put:
1094 path_put(&path);
1095fail:
1096 return ERR_PTR(err);
1097}
1098
1099static struct sock *unix_find_abstract(struct net *net,
1100 struct sockaddr_un *sunaddr,
1101 int addr_len, int type)
1102{
1103 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
1104 struct dentry *dentry;
1105 struct sock *sk;
1106
1107 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
1108 if (!sk)
1109 return ERR_PTR(-ECONNREFUSED);
1110
1111 dentry = unix_sk(sk)->path.dentry;
1112 if (dentry)
1113 touch_atime(&unix_sk(sk)->path);
1114
1115 return sk;
1116}
1117
1118static struct sock *unix_find_other(struct net *net,
1119 struct sockaddr_un *sunaddr,
1120 int addr_len, int type)
1121{
1122 struct sock *sk;
1123
1124 if (sunaddr->sun_path[0])
1125 sk = unix_find_bsd(sunaddr, addr_len, type);
1126 else
1127 sk = unix_find_abstract(net, sunaddr, addr_len, type);
1128
1129 return sk;
1130}
1131
1132static int unix_autobind(struct sock *sk)
1133{
1134 struct unix_sock *u = unix_sk(sk);
1135 unsigned int new_hash, old_hash;
1136 struct net *net = sock_net(sk);
1137 struct unix_address *addr;
1138 u32 lastnum, ordernum;
1139 int err;
1140
1141 err = mutex_lock_interruptible(&u->bindlock);
1142 if (err)
1143 return err;
1144
1145 if (u->addr)
1146 goto out;
1147
1148 err = -ENOMEM;
1149 addr = kzalloc(sizeof(*addr) +
1150 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1151 if (!addr)
1152 goto out;
1153
1154 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1155 addr->name->sun_family = AF_UNIX;
1156 refcount_set(&addr->refcnt, 1);
1157
1158 old_hash = sk->sk_hash;
1159 ordernum = get_random_u32();
1160 lastnum = ordernum & 0xFFFFF;
1161retry:
1162 ordernum = (ordernum + 1) & 0xFFFFF;
1163 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1164
1165 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1166 unix_table_double_lock(net, old_hash, new_hash);
1167
1168 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1169 unix_table_double_unlock(net, old_hash, new_hash);
1170
1171 /* __unix_find_socket_byname() may take long time if many names
1172 * are already in use.
1173 */
1174 cond_resched();
1175
1176 if (ordernum == lastnum) {
1177 /* Give up if all names seems to be in use. */
1178 err = -ENOSPC;
1179 unix_release_addr(addr);
1180 goto out;
1181 }
1182
1183 goto retry;
1184 }
1185
1186 __unix_set_addr_hash(net, sk, addr, new_hash);
1187 unix_table_double_unlock(net, old_hash, new_hash);
1188 err = 0;
1189
1190out: mutex_unlock(&u->bindlock);
1191 return err;
1192}
1193
1194static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1195 int addr_len)
1196{
1197 umode_t mode = S_IFSOCK |
1198 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1199 struct unix_sock *u = unix_sk(sk);
1200 unsigned int new_hash, old_hash;
1201 struct net *net = sock_net(sk);
1202 struct mnt_idmap *idmap;
1203 struct unix_address *addr;
1204 struct dentry *dentry;
1205 struct path parent;
1206 int err;
1207
1208 addr_len = unix_mkname_bsd(sunaddr, addr_len);
1209 addr = unix_create_addr(sunaddr, addr_len);
1210 if (!addr)
1211 return -ENOMEM;
1212
1213 /*
1214 * Get the parent directory, calculate the hash for last
1215 * component.
1216 */
1217 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1218 if (IS_ERR(dentry)) {
1219 err = PTR_ERR(dentry);
1220 goto out;
1221 }
1222
1223 /*
1224 * All right, let's create it.
1225 */
1226 idmap = mnt_idmap(parent.mnt);
1227 err = security_path_mknod(&parent, dentry, mode, 0);
1228 if (!err)
1229 err = vfs_mknod(idmap, d_inode(parent.dentry), dentry, mode, 0);
1230 if (err)
1231 goto out_path;
1232 err = mutex_lock_interruptible(&u->bindlock);
1233 if (err)
1234 goto out_unlink;
1235 if (u->addr)
1236 goto out_unlock;
1237
1238 old_hash = sk->sk_hash;
1239 new_hash = unix_bsd_hash(d_backing_inode(dentry));
1240 unix_table_double_lock(net, old_hash, new_hash);
1241 u->path.mnt = mntget(parent.mnt);
1242 u->path.dentry = dget(dentry);
1243 __unix_set_addr_hash(net, sk, addr, new_hash);
1244 unix_table_double_unlock(net, old_hash, new_hash);
1245 unix_insert_bsd_socket(sk);
1246 mutex_unlock(&u->bindlock);
1247 done_path_create(&parent, dentry);
1248 return 0;
1249
1250out_unlock:
1251 mutex_unlock(&u->bindlock);
1252 err = -EINVAL;
1253out_unlink:
1254 /* failed after successful mknod? unlink what we'd created... */
1255 vfs_unlink(idmap, d_inode(parent.dentry), dentry, NULL);
1256out_path:
1257 done_path_create(&parent, dentry);
1258out:
1259 unix_release_addr(addr);
1260 return err == -EEXIST ? -EADDRINUSE : err;
1261}
1262
1263static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1264 int addr_len)
1265{
1266 struct unix_sock *u = unix_sk(sk);
1267 unsigned int new_hash, old_hash;
1268 struct net *net = sock_net(sk);
1269 struct unix_address *addr;
1270 int err;
1271
1272 addr = unix_create_addr(sunaddr, addr_len);
1273 if (!addr)
1274 return -ENOMEM;
1275
1276 err = mutex_lock_interruptible(&u->bindlock);
1277 if (err)
1278 goto out;
1279
1280 if (u->addr) {
1281 err = -EINVAL;
1282 goto out_mutex;
1283 }
1284
1285 old_hash = sk->sk_hash;
1286 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
1287 unix_table_double_lock(net, old_hash, new_hash);
1288
1289 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
1290 goto out_spin;
1291
1292 __unix_set_addr_hash(net, sk, addr, new_hash);
1293 unix_table_double_unlock(net, old_hash, new_hash);
1294 mutex_unlock(&u->bindlock);
1295 return 0;
1296
1297out_spin:
1298 unix_table_double_unlock(net, old_hash, new_hash);
1299 err = -EADDRINUSE;
1300out_mutex:
1301 mutex_unlock(&u->bindlock);
1302out:
1303 unix_release_addr(addr);
1304 return err;
1305}
1306
1307static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1308{
1309 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1310 struct sock *sk = sock->sk;
1311 int err;
1312
1313 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1314 sunaddr->sun_family == AF_UNIX)
1315 return unix_autobind(sk);
1316
1317 err = unix_validate_addr(sunaddr, addr_len);
1318 if (err)
1319 return err;
1320
1321 if (sunaddr->sun_path[0])
1322 err = unix_bind_bsd(sk, sunaddr, addr_len);
1323 else
1324 err = unix_bind_abstract(sk, sunaddr, addr_len);
1325
1326 return err;
1327}
1328
1329static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1330{
1331 if (unlikely(sk1 == sk2) || !sk2) {
1332 unix_state_lock(sk1);
1333 return;
1334 }
1335 if (sk1 > sk2)
1336 swap(sk1, sk2);
1337
1338 unix_state_lock(sk1);
1339 unix_state_lock_nested(sk2, U_LOCK_SECOND);
1340}
1341
1342static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1343{
1344 if (unlikely(sk1 == sk2) || !sk2) {
1345 unix_state_unlock(sk1);
1346 return;
1347 }
1348 unix_state_unlock(sk1);
1349 unix_state_unlock(sk2);
1350}
1351
1352static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1353 int alen, int flags)
1354{
1355 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1356 struct sock *sk = sock->sk;
1357 struct sock *other;
1358 int err;
1359
1360 err = -EINVAL;
1361 if (alen < offsetofend(struct sockaddr, sa_family))
1362 goto out;
1363
1364 if (addr->sa_family != AF_UNSPEC) {
1365 err = unix_validate_addr(sunaddr, alen);
1366 if (err)
1367 goto out;
1368
1369 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, addr, &alen);
1370 if (err)
1371 goto out;
1372
1373 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1374 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1375 !READ_ONCE(unix_sk(sk)->addr)) {
1376 err = unix_autobind(sk);
1377 if (err)
1378 goto out;
1379 }
1380
1381restart:
1382 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
1383 if (IS_ERR(other)) {
1384 err = PTR_ERR(other);
1385 goto out;
1386 }
1387
1388 unix_state_double_lock(sk, other);
1389
1390 /* Apparently VFS overslept socket death. Retry. */
1391 if (sock_flag(other, SOCK_DEAD)) {
1392 unix_state_double_unlock(sk, other);
1393 sock_put(other);
1394 goto restart;
1395 }
1396
1397 err = -EPERM;
1398 if (!unix_may_send(sk, other))
1399 goto out_unlock;
1400
1401 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1402 if (err)
1403 goto out_unlock;
1404
1405 sk->sk_state = other->sk_state = TCP_ESTABLISHED;
1406 } else {
1407 /*
1408 * 1003.1g breaking connected state with AF_UNSPEC
1409 */
1410 other = NULL;
1411 unix_state_double_lock(sk, other);
1412 }
1413
1414 /*
1415 * If it was connected, reconnect.
1416 */
1417 if (unix_peer(sk)) {
1418 struct sock *old_peer = unix_peer(sk);
1419
1420 unix_peer(sk) = other;
1421 if (!other)
1422 sk->sk_state = TCP_CLOSE;
1423 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1424
1425 unix_state_double_unlock(sk, other);
1426
1427 if (other != old_peer)
1428 unix_dgram_disconnected(sk, old_peer);
1429 sock_put(old_peer);
1430 } else {
1431 unix_peer(sk) = other;
1432 unix_state_double_unlock(sk, other);
1433 }
1434
1435 return 0;
1436
1437out_unlock:
1438 unix_state_double_unlock(sk, other);
1439 sock_put(other);
1440out:
1441 return err;
1442}
1443
1444static long unix_wait_for_peer(struct sock *other, long timeo)
1445 __releases(&unix_sk(other)->lock)
1446{
1447 struct unix_sock *u = unix_sk(other);
1448 int sched;
1449 DEFINE_WAIT(wait);
1450
1451 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1452
1453 sched = !sock_flag(other, SOCK_DEAD) &&
1454 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1455 unix_recvq_full_lockless(other);
1456
1457 unix_state_unlock(other);
1458
1459 if (sched)
1460 timeo = schedule_timeout(timeo);
1461
1462 finish_wait(&u->peer_wait, &wait);
1463 return timeo;
1464}
1465
1466static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1467 int addr_len, int flags)
1468{
1469 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1470 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1471 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1472 struct net *net = sock_net(sk);
1473 struct sk_buff *skb = NULL;
1474 long timeo;
1475 int err;
1476 int st;
1477
1478 err = unix_validate_addr(sunaddr, addr_len);
1479 if (err)
1480 goto out;
1481
1482 err = BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, &addr_len);
1483 if (err)
1484 goto out;
1485
1486 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
1487 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
1488 !READ_ONCE(u->addr)) {
1489 err = unix_autobind(sk);
1490 if (err)
1491 goto out;
1492 }
1493
1494 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1495
1496 /* First of all allocate resources.
1497 If we will make it after state is locked,
1498 we will have to recheck all again in any case.
1499 */
1500
1501 /* create new sock for complete connection */
1502 newsk = unix_create1(net, NULL, 0, sock->type);
1503 if (IS_ERR(newsk)) {
1504 err = PTR_ERR(newsk);
1505 newsk = NULL;
1506 goto out;
1507 }
1508
1509 err = -ENOMEM;
1510
1511 /* Allocate skb for sending to listening sock */
1512 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1513 if (skb == NULL)
1514 goto out;
1515
1516restart:
1517 /* Find listening sock. */
1518 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
1519 if (IS_ERR(other)) {
1520 err = PTR_ERR(other);
1521 other = NULL;
1522 goto out;
1523 }
1524
1525 /* Latch state of peer */
1526 unix_state_lock(other);
1527
1528 /* Apparently VFS overslept socket death. Retry. */
1529 if (sock_flag(other, SOCK_DEAD)) {
1530 unix_state_unlock(other);
1531 sock_put(other);
1532 goto restart;
1533 }
1534
1535 err = -ECONNREFUSED;
1536 if (other->sk_state != TCP_LISTEN)
1537 goto out_unlock;
1538 if (other->sk_shutdown & RCV_SHUTDOWN)
1539 goto out_unlock;
1540
1541 if (unix_recvq_full(other)) {
1542 err = -EAGAIN;
1543 if (!timeo)
1544 goto out_unlock;
1545
1546 timeo = unix_wait_for_peer(other, timeo);
1547
1548 err = sock_intr_errno(timeo);
1549 if (signal_pending(current))
1550 goto out;
1551 sock_put(other);
1552 goto restart;
1553 }
1554
1555 /* Latch our state.
1556
1557 It is tricky place. We need to grab our state lock and cannot
1558 drop lock on peer. It is dangerous because deadlock is
1559 possible. Connect to self case and simultaneous
1560 attempt to connect are eliminated by checking socket
1561 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1562 check this before attempt to grab lock.
1563
1564 Well, and we have to recheck the state after socket locked.
1565 */
1566 st = sk->sk_state;
1567
1568 switch (st) {
1569 case TCP_CLOSE:
1570 /* This is ok... continue with connect */
1571 break;
1572 case TCP_ESTABLISHED:
1573 /* Socket is already connected */
1574 err = -EISCONN;
1575 goto out_unlock;
1576 default:
1577 err = -EINVAL;
1578 goto out_unlock;
1579 }
1580
1581 unix_state_lock_nested(sk, U_LOCK_SECOND);
1582
1583 if (sk->sk_state != st) {
1584 unix_state_unlock(sk);
1585 unix_state_unlock(other);
1586 sock_put(other);
1587 goto restart;
1588 }
1589
1590 err = security_unix_stream_connect(sk, other, newsk);
1591 if (err) {
1592 unix_state_unlock(sk);
1593 goto out_unlock;
1594 }
1595
1596 /* The way is open! Fastly set all the necessary fields... */
1597
1598 sock_hold(sk);
1599 unix_peer(newsk) = sk;
1600 newsk->sk_state = TCP_ESTABLISHED;
1601 newsk->sk_type = sk->sk_type;
1602 init_peercred(newsk);
1603 newu = unix_sk(newsk);
1604 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1605 otheru = unix_sk(other);
1606
1607 /* copy address information from listening to new sock
1608 *
1609 * The contents of *(otheru->addr) and otheru->path
1610 * are seen fully set up here, since we have found
1611 * otheru in hash under its lock. Insertion into the
1612 * hash chain we'd found it in had been done in an
1613 * earlier critical area protected by the chain's lock,
1614 * the same one where we'd set *(otheru->addr) contents,
1615 * as well as otheru->path and otheru->addr itself.
1616 *
1617 * Using smp_store_release() here to set newu->addr
1618 * is enough to make those stores, as well as stores
1619 * to newu->path visible to anyone who gets newu->addr
1620 * by smp_load_acquire(). IOW, the same warranties
1621 * as for unix_sock instances bound in unix_bind() or
1622 * in unix_autobind().
1623 */
1624 if (otheru->path.dentry) {
1625 path_get(&otheru->path);
1626 newu->path = otheru->path;
1627 }
1628 refcount_inc(&otheru->addr->refcnt);
1629 smp_store_release(&newu->addr, otheru->addr);
1630
1631 /* Set credentials */
1632 copy_peercred(sk, other);
1633
1634 sock->state = SS_CONNECTED;
1635 sk->sk_state = TCP_ESTABLISHED;
1636 sock_hold(newsk);
1637
1638 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1639 unix_peer(sk) = newsk;
1640
1641 unix_state_unlock(sk);
1642
1643 /* take ten and send info to listening sock */
1644 spin_lock(&other->sk_receive_queue.lock);
1645 __skb_queue_tail(&other->sk_receive_queue, skb);
1646 spin_unlock(&other->sk_receive_queue.lock);
1647 unix_state_unlock(other);
1648 other->sk_data_ready(other);
1649 sock_put(other);
1650 return 0;
1651
1652out_unlock:
1653 if (other)
1654 unix_state_unlock(other);
1655
1656out:
1657 kfree_skb(skb);
1658 if (newsk)
1659 unix_release_sock(newsk, 0);
1660 if (other)
1661 sock_put(other);
1662 return err;
1663}
1664
1665static int unix_socketpair(struct socket *socka, struct socket *sockb)
1666{
1667 struct sock *ska = socka->sk, *skb = sockb->sk;
1668
1669 /* Join our sockets back to back */
1670 sock_hold(ska);
1671 sock_hold(skb);
1672 unix_peer(ska) = skb;
1673 unix_peer(skb) = ska;
1674 init_peercred(ska);
1675 init_peercred(skb);
1676
1677 ska->sk_state = TCP_ESTABLISHED;
1678 skb->sk_state = TCP_ESTABLISHED;
1679 socka->state = SS_CONNECTED;
1680 sockb->state = SS_CONNECTED;
1681 return 0;
1682}
1683
1684static void unix_sock_inherit_flags(const struct socket *old,
1685 struct socket *new)
1686{
1687 if (test_bit(SOCK_PASSCRED, &old->flags))
1688 set_bit(SOCK_PASSCRED, &new->flags);
1689 if (test_bit(SOCK_PASSPIDFD, &old->flags))
1690 set_bit(SOCK_PASSPIDFD, &new->flags);
1691 if (test_bit(SOCK_PASSSEC, &old->flags))
1692 set_bit(SOCK_PASSSEC, &new->flags);
1693}
1694
1695static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1696 bool kern)
1697{
1698 struct sock *sk = sock->sk;
1699 struct sock *tsk;
1700 struct sk_buff *skb;
1701 int err;
1702
1703 err = -EOPNOTSUPP;
1704 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1705 goto out;
1706
1707 err = -EINVAL;
1708 if (sk->sk_state != TCP_LISTEN)
1709 goto out;
1710
1711 /* If socket state is TCP_LISTEN it cannot change (for now...),
1712 * so that no locks are necessary.
1713 */
1714
1715 skb = skb_recv_datagram(sk, (flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1716 &err);
1717 if (!skb) {
1718 /* This means receive shutdown. */
1719 if (err == 0)
1720 err = -EINVAL;
1721 goto out;
1722 }
1723
1724 tsk = skb->sk;
1725 skb_free_datagram(sk, skb);
1726 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1727
1728 /* attach accepted sock to socket */
1729 unix_state_lock(tsk);
1730 newsock->state = SS_CONNECTED;
1731 unix_sock_inherit_flags(sock, newsock);
1732 sock_graft(tsk, newsock);
1733 unix_state_unlock(tsk);
1734 return 0;
1735
1736out:
1737 return err;
1738}
1739
1740
1741static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1742{
1743 struct sock *sk = sock->sk;
1744 struct unix_address *addr;
1745 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1746 int err = 0;
1747
1748 if (peer) {
1749 sk = unix_peer_get(sk);
1750
1751 err = -ENOTCONN;
1752 if (!sk)
1753 goto out;
1754 err = 0;
1755 } else {
1756 sock_hold(sk);
1757 }
1758
1759 addr = smp_load_acquire(&unix_sk(sk)->addr);
1760 if (!addr) {
1761 sunaddr->sun_family = AF_UNIX;
1762 sunaddr->sun_path[0] = 0;
1763 err = offsetof(struct sockaddr_un, sun_path);
1764 } else {
1765 err = addr->len;
1766 memcpy(sunaddr, addr->name, addr->len);
1767
1768 if (peer)
1769 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1770 CGROUP_UNIX_GETPEERNAME);
1771 else
1772 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, &err,
1773 CGROUP_UNIX_GETSOCKNAME);
1774 }
1775 sock_put(sk);
1776out:
1777 return err;
1778}
1779
1780/* The "user->unix_inflight" variable is protected by the garbage
1781 * collection lock, and we just read it locklessly here. If you go
1782 * over the limit, there might be a tiny race in actually noticing
1783 * it across threads. Tough.
1784 */
1785static inline bool too_many_unix_fds(struct task_struct *p)
1786{
1787 struct user_struct *user = current_user();
1788
1789 if (unlikely(READ_ONCE(user->unix_inflight) > task_rlimit(p, RLIMIT_NOFILE)))
1790 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1791 return false;
1792}
1793
1794static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1795{
1796 int i;
1797
1798 if (too_many_unix_fds(current))
1799 return -ETOOMANYREFS;
1800
1801 /* Need to duplicate file references for the sake of garbage
1802 * collection. Otherwise a socket in the fps might become a
1803 * candidate for GC while the skb is not yet queued.
1804 */
1805 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1806 if (!UNIXCB(skb).fp)
1807 return -ENOMEM;
1808
1809 for (i = scm->fp->count - 1; i >= 0; i--)
1810 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1811
1812 return 0;
1813}
1814
1815static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1816{
1817 int i;
1818
1819 scm->fp = UNIXCB(skb).fp;
1820 UNIXCB(skb).fp = NULL;
1821
1822 for (i = scm->fp->count - 1; i >= 0; i--)
1823 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1824}
1825
1826static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1827{
1828 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1829
1830 /*
1831 * Garbage collection of unix sockets starts by selecting a set of
1832 * candidate sockets which have reference only from being in flight
1833 * (total_refs == inflight_refs). This condition is checked once during
1834 * the candidate collection phase, and candidates are marked as such, so
1835 * that non-candidates can later be ignored. While inflight_refs is
1836 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1837 * is an instantaneous decision.
1838 *
1839 * Once a candidate, however, the socket must not be reinstalled into a
1840 * file descriptor while the garbage collection is in progress.
1841 *
1842 * If the above conditions are met, then the directed graph of
1843 * candidates (*) does not change while unix_gc_lock is held.
1844 *
1845 * Any operations that changes the file count through file descriptors
1846 * (dup, close, sendmsg) does not change the graph since candidates are
1847 * not installed in fds.
1848 *
1849 * Dequeing a candidate via recvmsg would install it into an fd, but
1850 * that takes unix_gc_lock to decrement the inflight count, so it's
1851 * serialized with garbage collection.
1852 *
1853 * MSG_PEEK is special in that it does not change the inflight count,
1854 * yet does install the socket into an fd. The following lock/unlock
1855 * pair is to ensure serialization with garbage collection. It must be
1856 * done between incrementing the file count and installing the file into
1857 * an fd.
1858 *
1859 * If garbage collection starts after the barrier provided by the
1860 * lock/unlock, then it will see the elevated refcount and not mark this
1861 * as a candidate. If a garbage collection is already in progress
1862 * before the file count was incremented, then the lock/unlock pair will
1863 * ensure that garbage collection is finished before progressing to
1864 * installing the fd.
1865 *
1866 * (*) A -> B where B is on the queue of A or B is on the queue of C
1867 * which is on the queue of listening socket A.
1868 */
1869 spin_lock(&unix_gc_lock);
1870 spin_unlock(&unix_gc_lock);
1871}
1872
1873static void unix_destruct_scm(struct sk_buff *skb)
1874{
1875 struct scm_cookie scm;
1876
1877 memset(&scm, 0, sizeof(scm));
1878 scm.pid = UNIXCB(skb).pid;
1879 if (UNIXCB(skb).fp)
1880 unix_detach_fds(&scm, skb);
1881
1882 /* Alas, it calls VFS */
1883 /* So fscking what? fput() had been SMP-safe since the last Summer */
1884 scm_destroy(&scm);
1885 sock_wfree(skb);
1886}
1887
1888static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1889{
1890 int err = 0;
1891
1892 UNIXCB(skb).pid = get_pid(scm->pid);
1893 UNIXCB(skb).uid = scm->creds.uid;
1894 UNIXCB(skb).gid = scm->creds.gid;
1895 UNIXCB(skb).fp = NULL;
1896 unix_get_secdata(scm, skb);
1897 if (scm->fp && send_fds)
1898 err = unix_attach_fds(scm, skb);
1899
1900 skb->destructor = unix_destruct_scm;
1901 return err;
1902}
1903
1904static bool unix_passcred_enabled(const struct socket *sock,
1905 const struct sock *other)
1906{
1907 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1908 test_bit(SOCK_PASSPIDFD, &sock->flags) ||
1909 !other->sk_socket ||
1910 test_bit(SOCK_PASSCRED, &other->sk_socket->flags) ||
1911 test_bit(SOCK_PASSPIDFD, &other->sk_socket->flags);
1912}
1913
1914/*
1915 * Some apps rely on write() giving SCM_CREDENTIALS
1916 * We include credentials if source or destination socket
1917 * asserted SOCK_PASSCRED.
1918 */
1919static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1920 const struct sock *other)
1921{
1922 if (UNIXCB(skb).pid)
1923 return;
1924 if (unix_passcred_enabled(sock, other)) {
1925 UNIXCB(skb).pid = get_pid(task_tgid(current));
1926 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1927 }
1928}
1929
1930static bool unix_skb_scm_eq(struct sk_buff *skb,
1931 struct scm_cookie *scm)
1932{
1933 return UNIXCB(skb).pid == scm->pid &&
1934 uid_eq(UNIXCB(skb).uid, scm->creds.uid) &&
1935 gid_eq(UNIXCB(skb).gid, scm->creds.gid) &&
1936 unix_secdata_eq(scm, skb);
1937}
1938
1939static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1940{
1941 struct scm_fp_list *fp = UNIXCB(skb).fp;
1942 struct unix_sock *u = unix_sk(sk);
1943
1944 if (unlikely(fp && fp->count))
1945 atomic_add(fp->count, &u->scm_stat.nr_fds);
1946}
1947
1948static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1949{
1950 struct scm_fp_list *fp = UNIXCB(skb).fp;
1951 struct unix_sock *u = unix_sk(sk);
1952
1953 if (unlikely(fp && fp->count))
1954 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1955}
1956
1957/*
1958 * Send AF_UNIX data.
1959 */
1960
1961static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1962 size_t len)
1963{
1964 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1965 struct sock *sk = sock->sk, *other = NULL;
1966 struct unix_sock *u = unix_sk(sk);
1967 struct scm_cookie scm;
1968 struct sk_buff *skb;
1969 int data_len = 0;
1970 int sk_locked;
1971 long timeo;
1972 int err;
1973
1974 err = scm_send(sock, msg, &scm, false);
1975 if (err < 0)
1976 return err;
1977
1978 wait_for_unix_gc(scm.fp);
1979
1980 err = -EOPNOTSUPP;
1981 if (msg->msg_flags&MSG_OOB)
1982 goto out;
1983
1984 if (msg->msg_namelen) {
1985 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1986 if (err)
1987 goto out;
1988
1989 err = BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk,
1990 msg->msg_name,
1991 &msg->msg_namelen,
1992 NULL);
1993 if (err)
1994 goto out;
1995 } else {
1996 sunaddr = NULL;
1997 err = -ENOTCONN;
1998 other = unix_peer_get(sk);
1999 if (!other)
2000 goto out;
2001 }
2002
2003 if ((test_bit(SOCK_PASSCRED, &sock->flags) ||
2004 test_bit(SOCK_PASSPIDFD, &sock->flags)) &&
2005 !READ_ONCE(u->addr)) {
2006 err = unix_autobind(sk);
2007 if (err)
2008 goto out;
2009 }
2010
2011 err = -EMSGSIZE;
2012 if (len > sk->sk_sndbuf - 32)
2013 goto out;
2014
2015 if (len > SKB_MAX_ALLOC) {
2016 data_len = min_t(size_t,
2017 len - SKB_MAX_ALLOC,
2018 MAX_SKB_FRAGS * PAGE_SIZE);
2019 data_len = PAGE_ALIGN(data_len);
2020
2021 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
2022 }
2023
2024 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
2025 msg->msg_flags & MSG_DONTWAIT, &err,
2026 PAGE_ALLOC_COSTLY_ORDER);
2027 if (skb == NULL)
2028 goto out;
2029
2030 err = unix_scm_to_skb(&scm, skb, true);
2031 if (err < 0)
2032 goto out_free;
2033
2034 skb_put(skb, len - data_len);
2035 skb->data_len = data_len;
2036 skb->len = len;
2037 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
2038 if (err)
2039 goto out_free;
2040
2041 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
2042
2043restart:
2044 if (!other) {
2045 err = -ECONNRESET;
2046 if (sunaddr == NULL)
2047 goto out_free;
2048
2049 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
2050 sk->sk_type);
2051 if (IS_ERR(other)) {
2052 err = PTR_ERR(other);
2053 other = NULL;
2054 goto out_free;
2055 }
2056 }
2057
2058 if (sk_filter(other, skb) < 0) {
2059 /* Toss the packet but do not return any error to the sender */
2060 err = len;
2061 goto out_free;
2062 }
2063
2064 sk_locked = 0;
2065 unix_state_lock(other);
2066restart_locked:
2067 err = -EPERM;
2068 if (!unix_may_send(sk, other))
2069 goto out_unlock;
2070
2071 if (unlikely(sock_flag(other, SOCK_DEAD))) {
2072 /*
2073 * Check with 1003.1g - what should
2074 * datagram error
2075 */
2076 unix_state_unlock(other);
2077 sock_put(other);
2078
2079 if (!sk_locked)
2080 unix_state_lock(sk);
2081
2082 err = 0;
2083 if (sk->sk_type == SOCK_SEQPACKET) {
2084 /* We are here only when racing with unix_release_sock()
2085 * is clearing @other. Never change state to TCP_CLOSE
2086 * unlike SOCK_DGRAM wants.
2087 */
2088 unix_state_unlock(sk);
2089 err = -EPIPE;
2090 } else if (unix_peer(sk) == other) {
2091 unix_peer(sk) = NULL;
2092 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2093
2094 sk->sk_state = TCP_CLOSE;
2095 unix_state_unlock(sk);
2096
2097 unix_dgram_disconnected(sk, other);
2098 sock_put(other);
2099 err = -ECONNREFUSED;
2100 } else {
2101 unix_state_unlock(sk);
2102 }
2103
2104 other = NULL;
2105 if (err)
2106 goto out_free;
2107 goto restart;
2108 }
2109
2110 err = -EPIPE;
2111 if (other->sk_shutdown & RCV_SHUTDOWN)
2112 goto out_unlock;
2113
2114 if (sk->sk_type != SOCK_SEQPACKET) {
2115 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
2116 if (err)
2117 goto out_unlock;
2118 }
2119
2120 /* other == sk && unix_peer(other) != sk if
2121 * - unix_peer(sk) == NULL, destination address bound to sk
2122 * - unix_peer(sk) == sk by time of get but disconnected before lock
2123 */
2124 if (other != sk &&
2125 unlikely(unix_peer(other) != sk &&
2126 unix_recvq_full_lockless(other))) {
2127 if (timeo) {
2128 timeo = unix_wait_for_peer(other, timeo);
2129
2130 err = sock_intr_errno(timeo);
2131 if (signal_pending(current))
2132 goto out_free;
2133
2134 goto restart;
2135 }
2136
2137 if (!sk_locked) {
2138 unix_state_unlock(other);
2139 unix_state_double_lock(sk, other);
2140 }
2141
2142 if (unix_peer(sk) != other ||
2143 unix_dgram_peer_wake_me(sk, other)) {
2144 err = -EAGAIN;
2145 sk_locked = 1;
2146 goto out_unlock;
2147 }
2148
2149 if (!sk_locked) {
2150 sk_locked = 1;
2151 goto restart_locked;
2152 }
2153 }
2154
2155 if (unlikely(sk_locked))
2156 unix_state_unlock(sk);
2157
2158 if (sock_flag(other, SOCK_RCVTSTAMP))
2159 __net_timestamp(skb);
2160 maybe_add_creds(skb, sock, other);
2161 scm_stat_add(other, skb);
2162 skb_queue_tail(&other->sk_receive_queue, skb);
2163 unix_state_unlock(other);
2164 other->sk_data_ready(other);
2165 sock_put(other);
2166 scm_destroy(&scm);
2167 return len;
2168
2169out_unlock:
2170 if (sk_locked)
2171 unix_state_unlock(sk);
2172 unix_state_unlock(other);
2173out_free:
2174 kfree_skb(skb);
2175out:
2176 if (other)
2177 sock_put(other);
2178 scm_destroy(&scm);
2179 return err;
2180}
2181
2182/* We use paged skbs for stream sockets, and limit occupancy to 32768
2183 * bytes, and a minimum of a full page.
2184 */
2185#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
2186
2187#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2188static int queue_oob(struct socket *sock, struct msghdr *msg, struct sock *other,
2189 struct scm_cookie *scm, bool fds_sent)
2190{
2191 struct unix_sock *ousk = unix_sk(other);
2192 struct sk_buff *skb;
2193 int err = 0;
2194
2195 skb = sock_alloc_send_skb(sock->sk, 1, msg->msg_flags & MSG_DONTWAIT, &err);
2196
2197 if (!skb)
2198 return err;
2199
2200 err = unix_scm_to_skb(scm, skb, !fds_sent);
2201 if (err < 0) {
2202 kfree_skb(skb);
2203 return err;
2204 }
2205 skb_put(skb, 1);
2206 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
2207
2208 if (err) {
2209 kfree_skb(skb);
2210 return err;
2211 }
2212
2213 unix_state_lock(other);
2214
2215 if (sock_flag(other, SOCK_DEAD) ||
2216 (other->sk_shutdown & RCV_SHUTDOWN)) {
2217 unix_state_unlock(other);
2218 kfree_skb(skb);
2219 return -EPIPE;
2220 }
2221
2222 maybe_add_creds(skb, sock, other);
2223 skb_get(skb);
2224
2225 scm_stat_add(other, skb);
2226
2227 spin_lock(&other->sk_receive_queue.lock);
2228 if (ousk->oob_skb)
2229 consume_skb(ousk->oob_skb);
2230 WRITE_ONCE(ousk->oob_skb, skb);
2231 __skb_queue_tail(&other->sk_receive_queue, skb);
2232 spin_unlock(&other->sk_receive_queue.lock);
2233
2234 sk_send_sigurg(other);
2235 unix_state_unlock(other);
2236 other->sk_data_ready(other);
2237
2238 return err;
2239}
2240#endif
2241
2242static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2243 size_t len)
2244{
2245 struct sock *sk = sock->sk;
2246 struct sock *other = NULL;
2247 int err, size;
2248 struct sk_buff *skb;
2249 int sent = 0;
2250 struct scm_cookie scm;
2251 bool fds_sent = false;
2252 int data_len;
2253
2254 err = scm_send(sock, msg, &scm, false);
2255 if (err < 0)
2256 return err;
2257
2258 wait_for_unix_gc(scm.fp);
2259
2260 err = -EOPNOTSUPP;
2261 if (msg->msg_flags & MSG_OOB) {
2262#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2263 if (len)
2264 len--;
2265 else
2266#endif
2267 goto out_err;
2268 }
2269
2270 if (msg->msg_namelen) {
2271 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2272 goto out_err;
2273 } else {
2274 err = -ENOTCONN;
2275 other = unix_peer(sk);
2276 if (!other)
2277 goto out_err;
2278 }
2279
2280 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2281 goto pipe_err;
2282
2283 while (sent < len) {
2284 size = len - sent;
2285
2286 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2287 skb = sock_alloc_send_pskb(sk, 0, 0,
2288 msg->msg_flags & MSG_DONTWAIT,
2289 &err, 0);
2290 } else {
2291 /* Keep two messages in the pipe so it schedules better */
2292 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
2293
2294 /* allow fallback to order-0 allocations */
2295 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
2296
2297 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
2298
2299 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2300
2301 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
2302 msg->msg_flags & MSG_DONTWAIT, &err,
2303 get_order(UNIX_SKB_FRAGS_SZ));
2304 }
2305 if (!skb)
2306 goto out_err;
2307
2308 /* Only send the fds in the first buffer */
2309 err = unix_scm_to_skb(&scm, skb, !fds_sent);
2310 if (err < 0) {
2311 kfree_skb(skb);
2312 goto out_err;
2313 }
2314 fds_sent = true;
2315
2316 if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
2317 err = skb_splice_from_iter(skb, &msg->msg_iter, size,
2318 sk->sk_allocation);
2319 if (err < 0) {
2320 kfree_skb(skb);
2321 goto out_err;
2322 }
2323 size = err;
2324 refcount_add(size, &sk->sk_wmem_alloc);
2325 } else {
2326 skb_put(skb, size - data_len);
2327 skb->data_len = data_len;
2328 skb->len = size;
2329 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
2330 if (err) {
2331 kfree_skb(skb);
2332 goto out_err;
2333 }
2334 }
2335
2336 unix_state_lock(other);
2337
2338 if (sock_flag(other, SOCK_DEAD) ||
2339 (other->sk_shutdown & RCV_SHUTDOWN))
2340 goto pipe_err_free;
2341
2342 maybe_add_creds(skb, sock, other);
2343 scm_stat_add(other, skb);
2344 skb_queue_tail(&other->sk_receive_queue, skb);
2345 unix_state_unlock(other);
2346 other->sk_data_ready(other);
2347 sent += size;
2348 }
2349
2350#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2351 if (msg->msg_flags & MSG_OOB) {
2352 err = queue_oob(sock, msg, other, &scm, fds_sent);
2353 if (err)
2354 goto out_err;
2355 sent++;
2356 }
2357#endif
2358
2359 scm_destroy(&scm);
2360
2361 return sent;
2362
2363pipe_err_free:
2364 unix_state_unlock(other);
2365 kfree_skb(skb);
2366pipe_err:
2367 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
2368 send_sig(SIGPIPE, current, 0);
2369 err = -EPIPE;
2370out_err:
2371 scm_destroy(&scm);
2372 return sent ? : err;
2373}
2374
2375static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2376 size_t len)
2377{
2378 int err;
2379 struct sock *sk = sock->sk;
2380
2381 err = sock_error(sk);
2382 if (err)
2383 return err;
2384
2385 if (sk->sk_state != TCP_ESTABLISHED)
2386 return -ENOTCONN;
2387
2388 if (msg->msg_namelen)
2389 msg->msg_namelen = 0;
2390
2391 return unix_dgram_sendmsg(sock, msg, len);
2392}
2393
2394static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2395 size_t size, int flags)
2396{
2397 struct sock *sk = sock->sk;
2398
2399 if (sk->sk_state != TCP_ESTABLISHED)
2400 return -ENOTCONN;
2401
2402 return unix_dgram_recvmsg(sock, msg, size, flags);
2403}
2404
2405static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2406{
2407 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2408
2409 if (addr) {
2410 msg->msg_namelen = addr->len;
2411 memcpy(msg->msg_name, addr->name, addr->len);
2412 }
2413}
2414
2415int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
2416 int flags)
2417{
2418 struct scm_cookie scm;
2419 struct socket *sock = sk->sk_socket;
2420 struct unix_sock *u = unix_sk(sk);
2421 struct sk_buff *skb, *last;
2422 long timeo;
2423 int skip;
2424 int err;
2425
2426 err = -EOPNOTSUPP;
2427 if (flags&MSG_OOB)
2428 goto out;
2429
2430 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2431
2432 do {
2433 mutex_lock(&u->iolock);
2434
2435 skip = sk_peek_offset(sk, flags);
2436 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2437 &skip, &err, &last);
2438 if (skb) {
2439 if (!(flags & MSG_PEEK))
2440 scm_stat_del(sk, skb);
2441 break;
2442 }
2443
2444 mutex_unlock(&u->iolock);
2445
2446 if (err != -EAGAIN)
2447 break;
2448 } while (timeo &&
2449 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2450 &err, &timeo, last));
2451
2452 if (!skb) { /* implies iolock unlocked */
2453 unix_state_lock(sk);
2454 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2455 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2456 (sk->sk_shutdown & RCV_SHUTDOWN))
2457 err = 0;
2458 unix_state_unlock(sk);
2459 goto out;
2460 }
2461
2462 if (wq_has_sleeper(&u->peer_wait))
2463 wake_up_interruptible_sync_poll(&u->peer_wait,
2464 EPOLLOUT | EPOLLWRNORM |
2465 EPOLLWRBAND);
2466
2467 if (msg->msg_name) {
2468 unix_copy_addr(msg, skb->sk);
2469
2470 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2471 msg->msg_name,
2472 &msg->msg_namelen);
2473 }
2474
2475 if (size > skb->len - skip)
2476 size = skb->len - skip;
2477 else if (size < skb->len - skip)
2478 msg->msg_flags |= MSG_TRUNC;
2479
2480 err = skb_copy_datagram_msg(skb, skip, msg, size);
2481 if (err)
2482 goto out_free;
2483
2484 if (sock_flag(sk, SOCK_RCVTSTAMP))
2485 __sock_recv_timestamp(msg, sk, skb);
2486
2487 memset(&scm, 0, sizeof(scm));
2488
2489 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2490 unix_set_secdata(&scm, skb);
2491
2492 if (!(flags & MSG_PEEK)) {
2493 if (UNIXCB(skb).fp)
2494 unix_detach_fds(&scm, skb);
2495
2496 sk_peek_offset_bwd(sk, skb->len);
2497 } else {
2498 /* It is questionable: on PEEK we could:
2499 - do not return fds - good, but too simple 8)
2500 - return fds, and do not return them on read (old strategy,
2501 apparently wrong)
2502 - clone fds (I chose it for now, it is the most universal
2503 solution)
2504
2505 POSIX 1003.1g does not actually define this clearly
2506 at all. POSIX 1003.1g doesn't define a lot of things
2507 clearly however!
2508
2509 */
2510
2511 sk_peek_offset_fwd(sk, size);
2512
2513 if (UNIXCB(skb).fp)
2514 unix_peek_fds(&scm, skb);
2515 }
2516 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2517
2518 scm_recv_unix(sock, msg, &scm, flags);
2519
2520out_free:
2521 skb_free_datagram(sk, skb);
2522 mutex_unlock(&u->iolock);
2523out:
2524 return err;
2525}
2526
2527static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2528 int flags)
2529{
2530 struct sock *sk = sock->sk;
2531
2532#ifdef CONFIG_BPF_SYSCALL
2533 const struct proto *prot = READ_ONCE(sk->sk_prot);
2534
2535 if (prot != &unix_dgram_proto)
2536 return prot->recvmsg(sk, msg, size, flags, NULL);
2537#endif
2538 return __unix_dgram_recvmsg(sk, msg, size, flags);
2539}
2540
2541static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2542{
2543 struct unix_sock *u = unix_sk(sk);
2544 struct sk_buff *skb;
2545 int err;
2546
2547 mutex_lock(&u->iolock);
2548 skb = skb_recv_datagram(sk, MSG_DONTWAIT, &err);
2549 mutex_unlock(&u->iolock);
2550 if (!skb)
2551 return err;
2552
2553 return recv_actor(sk, skb);
2554}
2555
2556/*
2557 * Sleep until more data has arrived. But check for races..
2558 */
2559static long unix_stream_data_wait(struct sock *sk, long timeo,
2560 struct sk_buff *last, unsigned int last_len,
2561 bool freezable)
2562{
2563 unsigned int state = TASK_INTERRUPTIBLE | freezable * TASK_FREEZABLE;
2564 struct sk_buff *tail;
2565 DEFINE_WAIT(wait);
2566
2567 unix_state_lock(sk);
2568
2569 for (;;) {
2570 prepare_to_wait(sk_sleep(sk), &wait, state);
2571
2572 tail = skb_peek_tail(&sk->sk_receive_queue);
2573 if (tail != last ||
2574 (tail && tail->len != last_len) ||
2575 sk->sk_err ||
2576 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2577 signal_pending(current) ||
2578 !timeo)
2579 break;
2580
2581 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2582 unix_state_unlock(sk);
2583 timeo = schedule_timeout(timeo);
2584 unix_state_lock(sk);
2585
2586 if (sock_flag(sk, SOCK_DEAD))
2587 break;
2588
2589 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2590 }
2591
2592 finish_wait(sk_sleep(sk), &wait);
2593 unix_state_unlock(sk);
2594 return timeo;
2595}
2596
2597static unsigned int unix_skb_len(const struct sk_buff *skb)
2598{
2599 return skb->len - UNIXCB(skb).consumed;
2600}
2601
2602struct unix_stream_read_state {
2603 int (*recv_actor)(struct sk_buff *, int, int,
2604 struct unix_stream_read_state *);
2605 struct socket *socket;
2606 struct msghdr *msg;
2607 struct pipe_inode_info *pipe;
2608 size_t size;
2609 int flags;
2610 unsigned int splice_flags;
2611};
2612
2613#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2614static int unix_stream_recv_urg(struct unix_stream_read_state *state)
2615{
2616 struct socket *sock = state->socket;
2617 struct sock *sk = sock->sk;
2618 struct unix_sock *u = unix_sk(sk);
2619 int chunk = 1;
2620 struct sk_buff *oob_skb;
2621
2622 mutex_lock(&u->iolock);
2623 unix_state_lock(sk);
2624 spin_lock(&sk->sk_receive_queue.lock);
2625
2626 if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb) {
2627 spin_unlock(&sk->sk_receive_queue.lock);
2628 unix_state_unlock(sk);
2629 mutex_unlock(&u->iolock);
2630 return -EINVAL;
2631 }
2632
2633 oob_skb = u->oob_skb;
2634
2635 if (!(state->flags & MSG_PEEK))
2636 WRITE_ONCE(u->oob_skb, NULL);
2637 else
2638 skb_get(oob_skb);
2639
2640 spin_unlock(&sk->sk_receive_queue.lock);
2641 unix_state_unlock(sk);
2642
2643 chunk = state->recv_actor(oob_skb, 0, chunk, state);
2644
2645 if (!(state->flags & MSG_PEEK))
2646 UNIXCB(oob_skb).consumed += 1;
2647
2648 consume_skb(oob_skb);
2649
2650 mutex_unlock(&u->iolock);
2651
2652 if (chunk < 0)
2653 return -EFAULT;
2654
2655 state->msg->msg_flags |= MSG_OOB;
2656 return 1;
2657}
2658
2659static struct sk_buff *manage_oob(struct sk_buff *skb, struct sock *sk,
2660 int flags, int copied)
2661{
2662 struct unix_sock *u = unix_sk(sk);
2663
2664 if (!unix_skb_len(skb) && !(flags & MSG_PEEK)) {
2665 skb_unlink(skb, &sk->sk_receive_queue);
2666 consume_skb(skb);
2667 skb = NULL;
2668 } else {
2669 struct sk_buff *unlinked_skb = NULL;
2670
2671 spin_lock(&sk->sk_receive_queue.lock);
2672
2673 if (skb == u->oob_skb) {
2674 if (copied) {
2675 skb = NULL;
2676 } else if (sock_flag(sk, SOCK_URGINLINE)) {
2677 if (!(flags & MSG_PEEK)) {
2678 WRITE_ONCE(u->oob_skb, NULL);
2679 consume_skb(skb);
2680 }
2681 } else if (flags & MSG_PEEK) {
2682 skb = NULL;
2683 } else {
2684 __skb_unlink(skb, &sk->sk_receive_queue);
2685 WRITE_ONCE(u->oob_skb, NULL);
2686 unlinked_skb = skb;
2687 skb = skb_peek(&sk->sk_receive_queue);
2688 }
2689 }
2690
2691 spin_unlock(&sk->sk_receive_queue.lock);
2692
2693 if (unlinked_skb) {
2694 WARN_ON_ONCE(skb_unref(unlinked_skb));
2695 kfree_skb(unlinked_skb);
2696 }
2697 }
2698 return skb;
2699}
2700#endif
2701
2702static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
2703{
2704 if (unlikely(sk->sk_state != TCP_ESTABLISHED))
2705 return -ENOTCONN;
2706
2707 return unix_read_skb(sk, recv_actor);
2708}
2709
2710static int unix_stream_read_generic(struct unix_stream_read_state *state,
2711 bool freezable)
2712{
2713 struct scm_cookie scm;
2714 struct socket *sock = state->socket;
2715 struct sock *sk = sock->sk;
2716 struct unix_sock *u = unix_sk(sk);
2717 int copied = 0;
2718 int flags = state->flags;
2719 int noblock = flags & MSG_DONTWAIT;
2720 bool check_creds = false;
2721 int target;
2722 int err = 0;
2723 long timeo;
2724 int skip;
2725 size_t size = state->size;
2726 unsigned int last_len;
2727
2728 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2729 err = -EINVAL;
2730 goto out;
2731 }
2732
2733 if (unlikely(flags & MSG_OOB)) {
2734 err = -EOPNOTSUPP;
2735#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2736 err = unix_stream_recv_urg(state);
2737#endif
2738 goto out;
2739 }
2740
2741 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2742 timeo = sock_rcvtimeo(sk, noblock);
2743
2744 memset(&scm, 0, sizeof(scm));
2745
2746 /* Lock the socket to prevent queue disordering
2747 * while sleeps in memcpy_tomsg
2748 */
2749 mutex_lock(&u->iolock);
2750
2751 skip = max(sk_peek_offset(sk, flags), 0);
2752
2753 do {
2754 int chunk;
2755 bool drop_skb;
2756 struct sk_buff *skb, *last;
2757
2758redo:
2759 unix_state_lock(sk);
2760 if (sock_flag(sk, SOCK_DEAD)) {
2761 err = -ECONNRESET;
2762 goto unlock;
2763 }
2764 last = skb = skb_peek(&sk->sk_receive_queue);
2765 last_len = last ? last->len : 0;
2766
2767again:
2768#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
2769 if (skb) {
2770 skb = manage_oob(skb, sk, flags, copied);
2771 if (!skb && copied) {
2772 unix_state_unlock(sk);
2773 break;
2774 }
2775 }
2776#endif
2777 if (skb == NULL) {
2778 if (copied >= target)
2779 goto unlock;
2780
2781 /*
2782 * POSIX 1003.1g mandates this order.
2783 */
2784
2785 err = sock_error(sk);
2786 if (err)
2787 goto unlock;
2788 if (sk->sk_shutdown & RCV_SHUTDOWN)
2789 goto unlock;
2790
2791 unix_state_unlock(sk);
2792 if (!timeo) {
2793 err = -EAGAIN;
2794 break;
2795 }
2796
2797 mutex_unlock(&u->iolock);
2798
2799 timeo = unix_stream_data_wait(sk, timeo, last,
2800 last_len, freezable);
2801
2802 if (signal_pending(current)) {
2803 err = sock_intr_errno(timeo);
2804 scm_destroy(&scm);
2805 goto out;
2806 }
2807
2808 mutex_lock(&u->iolock);
2809 goto redo;
2810unlock:
2811 unix_state_unlock(sk);
2812 break;
2813 }
2814
2815 while (skip >= unix_skb_len(skb)) {
2816 skip -= unix_skb_len(skb);
2817 last = skb;
2818 last_len = skb->len;
2819 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2820 if (!skb)
2821 goto again;
2822 }
2823
2824 unix_state_unlock(sk);
2825
2826 if (check_creds) {
2827 /* Never glue messages from different writers */
2828 if (!unix_skb_scm_eq(skb, &scm))
2829 break;
2830 } else if (test_bit(SOCK_PASSCRED, &sock->flags) ||
2831 test_bit(SOCK_PASSPIDFD, &sock->flags)) {
2832 /* Copy credentials */
2833 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2834 unix_set_secdata(&scm, skb);
2835 check_creds = true;
2836 }
2837
2838 /* Copy address just once */
2839 if (state->msg && state->msg->msg_name) {
2840 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2841 state->msg->msg_name);
2842 unix_copy_addr(state->msg, skb->sk);
2843
2844 BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk,
2845 state->msg->msg_name,
2846 &state->msg->msg_namelen);
2847
2848 sunaddr = NULL;
2849 }
2850
2851 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2852 skb_get(skb);
2853 chunk = state->recv_actor(skb, skip, chunk, state);
2854 drop_skb = !unix_skb_len(skb);
2855 /* skb is only safe to use if !drop_skb */
2856 consume_skb(skb);
2857 if (chunk < 0) {
2858 if (copied == 0)
2859 copied = -EFAULT;
2860 break;
2861 }
2862 copied += chunk;
2863 size -= chunk;
2864
2865 if (drop_skb) {
2866 /* the skb was touched by a concurrent reader;
2867 * we should not expect anything from this skb
2868 * anymore and assume it invalid - we can be
2869 * sure it was dropped from the socket queue
2870 *
2871 * let's report a short read
2872 */
2873 err = 0;
2874 break;
2875 }
2876
2877 /* Mark read part of skb as used */
2878 if (!(flags & MSG_PEEK)) {
2879 UNIXCB(skb).consumed += chunk;
2880
2881 sk_peek_offset_bwd(sk, chunk);
2882
2883 if (UNIXCB(skb).fp) {
2884 scm_stat_del(sk, skb);
2885 unix_detach_fds(&scm, skb);
2886 }
2887
2888 if (unix_skb_len(skb))
2889 break;
2890
2891 skb_unlink(skb, &sk->sk_receive_queue);
2892 consume_skb(skb);
2893
2894 if (scm.fp)
2895 break;
2896 } else {
2897 /* It is questionable, see note in unix_dgram_recvmsg.
2898 */
2899 if (UNIXCB(skb).fp)
2900 unix_peek_fds(&scm, skb);
2901
2902 sk_peek_offset_fwd(sk, chunk);
2903
2904 if (UNIXCB(skb).fp)
2905 break;
2906
2907 skip = 0;
2908 last = skb;
2909 last_len = skb->len;
2910 unix_state_lock(sk);
2911 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2912 if (skb)
2913 goto again;
2914 unix_state_unlock(sk);
2915 break;
2916 }
2917 } while (size);
2918
2919 mutex_unlock(&u->iolock);
2920 if (state->msg)
2921 scm_recv_unix(sock, state->msg, &scm, flags);
2922 else
2923 scm_destroy(&scm);
2924out:
2925 return copied ? : err;
2926}
2927
2928static int unix_stream_read_actor(struct sk_buff *skb,
2929 int skip, int chunk,
2930 struct unix_stream_read_state *state)
2931{
2932 int ret;
2933
2934 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2935 state->msg, chunk);
2936 return ret ?: chunk;
2937}
2938
2939int __unix_stream_recvmsg(struct sock *sk, struct msghdr *msg,
2940 size_t size, int flags)
2941{
2942 struct unix_stream_read_state state = {
2943 .recv_actor = unix_stream_read_actor,
2944 .socket = sk->sk_socket,
2945 .msg = msg,
2946 .size = size,
2947 .flags = flags
2948 };
2949
2950 return unix_stream_read_generic(&state, true);
2951}
2952
2953static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2954 size_t size, int flags)
2955{
2956 struct unix_stream_read_state state = {
2957 .recv_actor = unix_stream_read_actor,
2958 .socket = sock,
2959 .msg = msg,
2960 .size = size,
2961 .flags = flags
2962 };
2963
2964#ifdef CONFIG_BPF_SYSCALL
2965 struct sock *sk = sock->sk;
2966 const struct proto *prot = READ_ONCE(sk->sk_prot);
2967
2968 if (prot != &unix_stream_proto)
2969 return prot->recvmsg(sk, msg, size, flags, NULL);
2970#endif
2971 return unix_stream_read_generic(&state, true);
2972}
2973
2974static int unix_stream_splice_actor(struct sk_buff *skb,
2975 int skip, int chunk,
2976 struct unix_stream_read_state *state)
2977{
2978 return skb_splice_bits(skb, state->socket->sk,
2979 UNIXCB(skb).consumed + skip,
2980 state->pipe, chunk, state->splice_flags);
2981}
2982
2983static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2984 struct pipe_inode_info *pipe,
2985 size_t size, unsigned int flags)
2986{
2987 struct unix_stream_read_state state = {
2988 .recv_actor = unix_stream_splice_actor,
2989 .socket = sock,
2990 .pipe = pipe,
2991 .size = size,
2992 .splice_flags = flags,
2993 };
2994
2995 if (unlikely(*ppos))
2996 return -ESPIPE;
2997
2998 if (sock->file->f_flags & O_NONBLOCK ||
2999 flags & SPLICE_F_NONBLOCK)
3000 state.flags = MSG_DONTWAIT;
3001
3002 return unix_stream_read_generic(&state, false);
3003}
3004
3005static int unix_shutdown(struct socket *sock, int mode)
3006{
3007 struct sock *sk = sock->sk;
3008 struct sock *other;
3009
3010 if (mode < SHUT_RD || mode > SHUT_RDWR)
3011 return -EINVAL;
3012 /* This maps:
3013 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
3014 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
3015 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
3016 */
3017 ++mode;
3018
3019 unix_state_lock(sk);
3020 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | mode);
3021 other = unix_peer(sk);
3022 if (other)
3023 sock_hold(other);
3024 unix_state_unlock(sk);
3025 sk->sk_state_change(sk);
3026
3027 if (other &&
3028 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
3029
3030 int peer_mode = 0;
3031 const struct proto *prot = READ_ONCE(other->sk_prot);
3032
3033 if (prot->unhash)
3034 prot->unhash(other);
3035 if (mode&RCV_SHUTDOWN)
3036 peer_mode |= SEND_SHUTDOWN;
3037 if (mode&SEND_SHUTDOWN)
3038 peer_mode |= RCV_SHUTDOWN;
3039 unix_state_lock(other);
3040 WRITE_ONCE(other->sk_shutdown, other->sk_shutdown | peer_mode);
3041 unix_state_unlock(other);
3042 other->sk_state_change(other);
3043 if (peer_mode == SHUTDOWN_MASK)
3044 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
3045 else if (peer_mode & RCV_SHUTDOWN)
3046 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
3047 }
3048 if (other)
3049 sock_put(other);
3050
3051 return 0;
3052}
3053
3054long unix_inq_len(struct sock *sk)
3055{
3056 struct sk_buff *skb;
3057 long amount = 0;
3058
3059 if (sk->sk_state == TCP_LISTEN)
3060 return -EINVAL;
3061
3062 spin_lock(&sk->sk_receive_queue.lock);
3063 if (sk->sk_type == SOCK_STREAM ||
3064 sk->sk_type == SOCK_SEQPACKET) {
3065 skb_queue_walk(&sk->sk_receive_queue, skb)
3066 amount += unix_skb_len(skb);
3067 } else {
3068 skb = skb_peek(&sk->sk_receive_queue);
3069 if (skb)
3070 amount = skb->len;
3071 }
3072 spin_unlock(&sk->sk_receive_queue.lock);
3073
3074 return amount;
3075}
3076EXPORT_SYMBOL_GPL(unix_inq_len);
3077
3078long unix_outq_len(struct sock *sk)
3079{
3080 return sk_wmem_alloc_get(sk);
3081}
3082EXPORT_SYMBOL_GPL(unix_outq_len);
3083
3084static int unix_open_file(struct sock *sk)
3085{
3086 struct path path;
3087 struct file *f;
3088 int fd;
3089
3090 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
3091 return -EPERM;
3092
3093 if (!smp_load_acquire(&unix_sk(sk)->addr))
3094 return -ENOENT;
3095
3096 path = unix_sk(sk)->path;
3097 if (!path.dentry)
3098 return -ENOENT;
3099
3100 path_get(&path);
3101
3102 fd = get_unused_fd_flags(O_CLOEXEC);
3103 if (fd < 0)
3104 goto out;
3105
3106 f = dentry_open(&path, O_PATH, current_cred());
3107 if (IS_ERR(f)) {
3108 put_unused_fd(fd);
3109 fd = PTR_ERR(f);
3110 goto out;
3111 }
3112
3113 fd_install(fd, f);
3114out:
3115 path_put(&path);
3116
3117 return fd;
3118}
3119
3120static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3121{
3122 struct sock *sk = sock->sk;
3123 long amount = 0;
3124 int err;
3125
3126 switch (cmd) {
3127 case SIOCOUTQ:
3128 amount = unix_outq_len(sk);
3129 err = put_user(amount, (int __user *)arg);
3130 break;
3131 case SIOCINQ:
3132 amount = unix_inq_len(sk);
3133 if (amount < 0)
3134 err = amount;
3135 else
3136 err = put_user(amount, (int __user *)arg);
3137 break;
3138 case SIOCUNIXFILE:
3139 err = unix_open_file(sk);
3140 break;
3141#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3142 case SIOCATMARK:
3143 {
3144 struct sk_buff *skb;
3145 int answ = 0;
3146
3147 skb = skb_peek(&sk->sk_receive_queue);
3148 if (skb && skb == READ_ONCE(unix_sk(sk)->oob_skb))
3149 answ = 1;
3150 err = put_user(answ, (int __user *)arg);
3151 }
3152 break;
3153#endif
3154 default:
3155 err = -ENOIOCTLCMD;
3156 break;
3157 }
3158 return err;
3159}
3160
3161#ifdef CONFIG_COMPAT
3162static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3163{
3164 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
3165}
3166#endif
3167
3168static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
3169{
3170 struct sock *sk = sock->sk;
3171 __poll_t mask;
3172 u8 shutdown;
3173
3174 sock_poll_wait(file, sock, wait);
3175 mask = 0;
3176 shutdown = READ_ONCE(sk->sk_shutdown);
3177
3178 /* exceptional events? */
3179 if (READ_ONCE(sk->sk_err))
3180 mask |= EPOLLERR;
3181 if (shutdown == SHUTDOWN_MASK)
3182 mask |= EPOLLHUP;
3183 if (shutdown & RCV_SHUTDOWN)
3184 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3185
3186 /* readable? */
3187 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3188 mask |= EPOLLIN | EPOLLRDNORM;
3189 if (sk_is_readable(sk))
3190 mask |= EPOLLIN | EPOLLRDNORM;
3191#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3192 if (READ_ONCE(unix_sk(sk)->oob_skb))
3193 mask |= EPOLLPRI;
3194#endif
3195
3196 /* Connection-based need to check for termination and startup */
3197 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3198 sk->sk_state == TCP_CLOSE)
3199 mask |= EPOLLHUP;
3200
3201 /*
3202 * we set writable also when the other side has shut down the
3203 * connection. This prevents stuck sockets.
3204 */
3205 if (unix_writable(sk))
3206 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3207
3208 return mask;
3209}
3210
3211static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3212 poll_table *wait)
3213{
3214 struct sock *sk = sock->sk, *other;
3215 unsigned int writable;
3216 __poll_t mask;
3217 u8 shutdown;
3218
3219 sock_poll_wait(file, sock, wait);
3220 mask = 0;
3221 shutdown = READ_ONCE(sk->sk_shutdown);
3222
3223 /* exceptional events? */
3224 if (READ_ONCE(sk->sk_err) ||
3225 !skb_queue_empty_lockless(&sk->sk_error_queue))
3226 mask |= EPOLLERR |
3227 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
3228
3229 if (shutdown & RCV_SHUTDOWN)
3230 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3231 if (shutdown == SHUTDOWN_MASK)
3232 mask |= EPOLLHUP;
3233
3234 /* readable? */
3235 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
3236 mask |= EPOLLIN | EPOLLRDNORM;
3237 if (sk_is_readable(sk))
3238 mask |= EPOLLIN | EPOLLRDNORM;
3239
3240 /* Connection-based need to check for termination and startup */
3241 if (sk->sk_type == SOCK_SEQPACKET) {
3242 if (sk->sk_state == TCP_CLOSE)
3243 mask |= EPOLLHUP;
3244 /* connection hasn't started yet? */
3245 if (sk->sk_state == TCP_SYN_SENT)
3246 return mask;
3247 }
3248
3249 /* No write status requested, avoid expensive OUT tests. */
3250 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
3251 return mask;
3252
3253 writable = unix_writable(sk);
3254 if (writable) {
3255 unix_state_lock(sk);
3256
3257 other = unix_peer(sk);
3258 if (other && unix_peer(other) != sk &&
3259 unix_recvq_full_lockless(other) &&
3260 unix_dgram_peer_wake_me(sk, other))
3261 writable = 0;
3262
3263 unix_state_unlock(sk);
3264 }
3265
3266 if (writable)
3267 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3268 else
3269 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3270
3271 return mask;
3272}
3273
3274#ifdef CONFIG_PROC_FS
3275
3276#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3277
3278#define get_bucket(x) ((x) >> BUCKET_SPACE)
3279#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
3280#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
3281
3282static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
3283{
3284 unsigned long offset = get_offset(*pos);
3285 unsigned long bucket = get_bucket(*pos);
3286 unsigned long count = 0;
3287 struct sock *sk;
3288
3289 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3290 sk; sk = sk_next(sk)) {
3291 if (++count == offset)
3292 break;
3293 }
3294
3295 return sk;
3296}
3297
3298static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
3299{
3300 unsigned long bucket = get_bucket(*pos);
3301 struct net *net = seq_file_net(seq);
3302 struct sock *sk;
3303
3304 while (bucket < UNIX_HASH_SIZE) {
3305 spin_lock(&net->unx.table.locks[bucket]);
3306
3307 sk = unix_from_bucket(seq, pos);
3308 if (sk)
3309 return sk;
3310
3311 spin_unlock(&net->unx.table.locks[bucket]);
3312
3313 *pos = set_bucket_offset(++bucket, 1);
3314 }
3315
3316 return NULL;
3317}
3318
3319static struct sock *unix_get_next(struct seq_file *seq, struct sock *sk,
3320 loff_t *pos)
3321{
3322 unsigned long bucket = get_bucket(*pos);
3323
3324 sk = sk_next(sk);
3325 if (sk)
3326 return sk;
3327
3328
3329 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
3330
3331 *pos = set_bucket_offset(++bucket, 1);
3332
3333 return unix_get_first(seq, pos);
3334}
3335
3336static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3337{
3338 if (!*pos)
3339 return SEQ_START_TOKEN;
3340
3341 return unix_get_first(seq, pos);
3342}
3343
3344static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3345{
3346 ++*pos;
3347
3348 if (v == SEQ_START_TOKEN)
3349 return unix_get_first(seq, pos);
3350
3351 return unix_get_next(seq, v, pos);
3352}
3353
3354static void unix_seq_stop(struct seq_file *seq, void *v)
3355{
3356 struct sock *sk = v;
3357
3358 if (sk)
3359 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
3360}
3361
3362static int unix_seq_show(struct seq_file *seq, void *v)
3363{
3364
3365 if (v == SEQ_START_TOKEN)
3366 seq_puts(seq, "Num RefCount Protocol Flags Type St "
3367 "Inode Path\n");
3368 else {
3369 struct sock *s = v;
3370 struct unix_sock *u = unix_sk(s);
3371 unix_state_lock(s);
3372
3373 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
3374 s,
3375 refcount_read(&s->sk_refcnt),
3376 0,
3377 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
3378 s->sk_type,
3379 s->sk_socket ?
3380 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
3381 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
3382 sock_i_ino(s));
3383
3384 if (u->addr) { // under a hash table lock here
3385 int i, len;
3386 seq_putc(seq, ' ');
3387
3388 i = 0;
3389 len = u->addr->len -
3390 offsetof(struct sockaddr_un, sun_path);
3391 if (u->addr->name->sun_path[0]) {
3392 len--;
3393 } else {
3394 seq_putc(seq, '@');
3395 i++;
3396 }
3397 for ( ; i < len; i++)
3398 seq_putc(seq, u->addr->name->sun_path[i] ?:
3399 '@');
3400 }
3401 unix_state_unlock(s);
3402 seq_putc(seq, '\n');
3403 }
3404
3405 return 0;
3406}
3407
3408static const struct seq_operations unix_seq_ops = {
3409 .start = unix_seq_start,
3410 .next = unix_seq_next,
3411 .stop = unix_seq_stop,
3412 .show = unix_seq_show,
3413};
3414
3415#ifdef CONFIG_BPF_SYSCALL
3416struct bpf_unix_iter_state {
3417 struct seq_net_private p;
3418 unsigned int cur_sk;
3419 unsigned int end_sk;
3420 unsigned int max_sk;
3421 struct sock **batch;
3422 bool st_bucket_done;
3423};
3424
3425struct bpf_iter__unix {
3426 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3427 __bpf_md_ptr(struct unix_sock *, unix_sk);
3428 uid_t uid __aligned(8);
3429};
3430
3431static int unix_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3432 struct unix_sock *unix_sk, uid_t uid)
3433{
3434 struct bpf_iter__unix ctx;
3435
3436 meta->seq_num--; /* skip SEQ_START_TOKEN */
3437 ctx.meta = meta;
3438 ctx.unix_sk = unix_sk;
3439 ctx.uid = uid;
3440 return bpf_iter_run_prog(prog, &ctx);
3441}
3442
3443static int bpf_iter_unix_hold_batch(struct seq_file *seq, struct sock *start_sk)
3444
3445{
3446 struct bpf_unix_iter_state *iter = seq->private;
3447 unsigned int expected = 1;
3448 struct sock *sk;
3449
3450 sock_hold(start_sk);
3451 iter->batch[iter->end_sk++] = start_sk;
3452
3453 for (sk = sk_next(start_sk); sk; sk = sk_next(sk)) {
3454 if (iter->end_sk < iter->max_sk) {
3455 sock_hold(sk);
3456 iter->batch[iter->end_sk++] = sk;
3457 }
3458
3459 expected++;
3460 }
3461
3462 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
3463
3464 return expected;
3465}
3466
3467static void bpf_iter_unix_put_batch(struct bpf_unix_iter_state *iter)
3468{
3469 while (iter->cur_sk < iter->end_sk)
3470 sock_put(iter->batch[iter->cur_sk++]);
3471}
3472
3473static int bpf_iter_unix_realloc_batch(struct bpf_unix_iter_state *iter,
3474 unsigned int new_batch_sz)
3475{
3476 struct sock **new_batch;
3477
3478 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3479 GFP_USER | __GFP_NOWARN);
3480 if (!new_batch)
3481 return -ENOMEM;
3482
3483 bpf_iter_unix_put_batch(iter);
3484 kvfree(iter->batch);
3485 iter->batch = new_batch;
3486 iter->max_sk = new_batch_sz;
3487
3488 return 0;
3489}
3490
3491static struct sock *bpf_iter_unix_batch(struct seq_file *seq,
3492 loff_t *pos)
3493{
3494 struct bpf_unix_iter_state *iter = seq->private;
3495 unsigned int expected;
3496 bool resized = false;
3497 struct sock *sk;
3498
3499 if (iter->st_bucket_done)
3500 *pos = set_bucket_offset(get_bucket(*pos) + 1, 1);
3501
3502again:
3503 /* Get a new batch */
3504 iter->cur_sk = 0;
3505 iter->end_sk = 0;
3506
3507 sk = unix_get_first(seq, pos);
3508 if (!sk)
3509 return NULL; /* Done */
3510
3511 expected = bpf_iter_unix_hold_batch(seq, sk);
3512
3513 if (iter->end_sk == expected) {
3514 iter->st_bucket_done = true;
3515 return sk;
3516 }
3517
3518 if (!resized && !bpf_iter_unix_realloc_batch(iter, expected * 3 / 2)) {
3519 resized = true;
3520 goto again;
3521 }
3522
3523 return sk;
3524}
3525
3526static void *bpf_iter_unix_seq_start(struct seq_file *seq, loff_t *pos)
3527{
3528 if (!*pos)
3529 return SEQ_START_TOKEN;
3530
3531 /* bpf iter does not support lseek, so it always
3532 * continue from where it was stop()-ped.
3533 */
3534 return bpf_iter_unix_batch(seq, pos);
3535}
3536
3537static void *bpf_iter_unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3538{
3539 struct bpf_unix_iter_state *iter = seq->private;
3540 struct sock *sk;
3541
3542 /* Whenever seq_next() is called, the iter->cur_sk is
3543 * done with seq_show(), so advance to the next sk in
3544 * the batch.
3545 */
3546 if (iter->cur_sk < iter->end_sk)
3547 sock_put(iter->batch[iter->cur_sk++]);
3548
3549 ++*pos;
3550
3551 if (iter->cur_sk < iter->end_sk)
3552 sk = iter->batch[iter->cur_sk];
3553 else
3554 sk = bpf_iter_unix_batch(seq, pos);
3555
3556 return sk;
3557}
3558
3559static int bpf_iter_unix_seq_show(struct seq_file *seq, void *v)
3560{
3561 struct bpf_iter_meta meta;
3562 struct bpf_prog *prog;
3563 struct sock *sk = v;
3564 uid_t uid;
3565 bool slow;
3566 int ret;
3567
3568 if (v == SEQ_START_TOKEN)
3569 return 0;
3570
3571 slow = lock_sock_fast(sk);
3572
3573 if (unlikely(sk_unhashed(sk))) {
3574 ret = SEQ_SKIP;
3575 goto unlock;
3576 }
3577
3578 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3579 meta.seq = seq;
3580 prog = bpf_iter_get_info(&meta, false);
3581 ret = unix_prog_seq_show(prog, &meta, v, uid);
3582unlock:
3583 unlock_sock_fast(sk, slow);
3584 return ret;
3585}
3586
3587static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3588{
3589 struct bpf_unix_iter_state *iter = seq->private;
3590 struct bpf_iter_meta meta;
3591 struct bpf_prog *prog;
3592
3593 if (!v) {
3594 meta.seq = seq;
3595 prog = bpf_iter_get_info(&meta, true);
3596 if (prog)
3597 (void)unix_prog_seq_show(prog, &meta, v, 0);
3598 }
3599
3600 if (iter->cur_sk < iter->end_sk)
3601 bpf_iter_unix_put_batch(iter);
3602}
3603
3604static const struct seq_operations bpf_iter_unix_seq_ops = {
3605 .start = bpf_iter_unix_seq_start,
3606 .next = bpf_iter_unix_seq_next,
3607 .stop = bpf_iter_unix_seq_stop,
3608 .show = bpf_iter_unix_seq_show,
3609};
3610#endif
3611#endif
3612
3613static const struct net_proto_family unix_family_ops = {
3614 .family = PF_UNIX,
3615 .create = unix_create,
3616 .owner = THIS_MODULE,
3617};
3618
3619
3620static int __net_init unix_net_init(struct net *net)
3621{
3622 int i;
3623
3624 net->unx.sysctl_max_dgram_qlen = 10;
3625 if (unix_sysctl_register(net))
3626 goto out;
3627
3628#ifdef CONFIG_PROC_FS
3629 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
3630 sizeof(struct seq_net_private)))
3631 goto err_sysctl;
3632#endif
3633
3634 net->unx.table.locks = kvmalloc_array(UNIX_HASH_SIZE,
3635 sizeof(spinlock_t), GFP_KERNEL);
3636 if (!net->unx.table.locks)
3637 goto err_proc;
3638
3639 net->unx.table.buckets = kvmalloc_array(UNIX_HASH_SIZE,
3640 sizeof(struct hlist_head),
3641 GFP_KERNEL);
3642 if (!net->unx.table.buckets)
3643 goto free_locks;
3644
3645 for (i = 0; i < UNIX_HASH_SIZE; i++) {
3646 spin_lock_init(&net->unx.table.locks[i]);
3647 INIT_HLIST_HEAD(&net->unx.table.buckets[i]);
3648 }
3649
3650 return 0;
3651
3652free_locks:
3653 kvfree(net->unx.table.locks);
3654err_proc:
3655#ifdef CONFIG_PROC_FS
3656 remove_proc_entry("unix", net->proc_net);
3657err_sysctl:
3658#endif
3659 unix_sysctl_unregister(net);
3660out:
3661 return -ENOMEM;
3662}
3663
3664static void __net_exit unix_net_exit(struct net *net)
3665{
3666 kvfree(net->unx.table.buckets);
3667 kvfree(net->unx.table.locks);
3668 unix_sysctl_unregister(net);
3669 remove_proc_entry("unix", net->proc_net);
3670}
3671
3672static struct pernet_operations unix_net_ops = {
3673 .init = unix_net_init,
3674 .exit = unix_net_exit,
3675};
3676
3677#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3678DEFINE_BPF_ITER_FUNC(unix, struct bpf_iter_meta *meta,
3679 struct unix_sock *unix_sk, uid_t uid)
3680
3681#define INIT_BATCH_SZ 16
3682
3683static int bpf_iter_init_unix(void *priv_data, struct bpf_iter_aux_info *aux)
3684{
3685 struct bpf_unix_iter_state *iter = priv_data;
3686 int err;
3687
3688 err = bpf_iter_init_seq_net(priv_data, aux);
3689 if (err)
3690 return err;
3691
3692 err = bpf_iter_unix_realloc_batch(iter, INIT_BATCH_SZ);
3693 if (err) {
3694 bpf_iter_fini_seq_net(priv_data);
3695 return err;
3696 }
3697
3698 return 0;
3699}
3700
3701static void bpf_iter_fini_unix(void *priv_data)
3702{
3703 struct bpf_unix_iter_state *iter = priv_data;
3704
3705 bpf_iter_fini_seq_net(priv_data);
3706 kvfree(iter->batch);
3707}
3708
3709static const struct bpf_iter_seq_info unix_seq_info = {
3710 .seq_ops = &bpf_iter_unix_seq_ops,
3711 .init_seq_private = bpf_iter_init_unix,
3712 .fini_seq_private = bpf_iter_fini_unix,
3713 .seq_priv_size = sizeof(struct bpf_unix_iter_state),
3714};
3715
3716static const struct bpf_func_proto *
3717bpf_iter_unix_get_func_proto(enum bpf_func_id func_id,
3718 const struct bpf_prog *prog)
3719{
3720 switch (func_id) {
3721 case BPF_FUNC_setsockopt:
3722 return &bpf_sk_setsockopt_proto;
3723 case BPF_FUNC_getsockopt:
3724 return &bpf_sk_getsockopt_proto;
3725 default:
3726 return NULL;
3727 }
3728}
3729
3730static struct bpf_iter_reg unix_reg_info = {
3731 .target = "unix",
3732 .ctx_arg_info_size = 1,
3733 .ctx_arg_info = {
3734 { offsetof(struct bpf_iter__unix, unix_sk),
3735 PTR_TO_BTF_ID_OR_NULL },
3736 },
3737 .get_func_proto = bpf_iter_unix_get_func_proto,
3738 .seq_info = &unix_seq_info,
3739};
3740
3741static void __init bpf_iter_register(void)
3742{
3743 unix_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UNIX];
3744 if (bpf_iter_reg_target(&unix_reg_info))
3745 pr_warn("Warning: could not register bpf iterator unix\n");
3746}
3747#endif
3748
3749static int __init af_unix_init(void)
3750{
3751 int i, rc = -1;
3752
3753 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
3754
3755 for (i = 0; i < UNIX_HASH_SIZE / 2; i++) {
3756 spin_lock_init(&bsd_socket_locks[i]);
3757 INIT_HLIST_HEAD(&bsd_socket_buckets[i]);
3758 }
3759
3760 rc = proto_register(&unix_dgram_proto, 1);
3761 if (rc != 0) {
3762 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3763 goto out;
3764 }
3765
3766 rc = proto_register(&unix_stream_proto, 1);
3767 if (rc != 0) {
3768 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
3769 proto_unregister(&unix_dgram_proto);
3770 goto out;
3771 }
3772
3773 sock_register(&unix_family_ops);
3774 register_pernet_subsys(&unix_net_ops);
3775 unix_bpf_build_proto();
3776
3777#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3778 bpf_iter_register();
3779#endif
3780
3781out:
3782 return rc;
3783}
3784
3785/* Later than subsys_initcall() because we depend on stuff initialised there */
3786fs_initcall(af_unix_init);