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