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