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