projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs: port vfs_*() helpers to struct mnt_idmap
[linux-block.git] / net / unix / af_unix.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 * NET4: Implementation of BSD Unix domain sockets.
4 *
113aa838 5 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
1da177e4 6 *
1da177e4
LT		 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 *
1da177e4
LT		 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
5cc208be 78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
1da177e4 80#include <linux/module.h>
1da177e4 81#include <linux/kernel.h>
1da177e4 82#include <linux/signal.h>
3f07c014 83#include <linux/sched/signal.h>
1da177e4
LT		 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>
b6459415 92#include <linux/filter.h>
1da177e4
LT		 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>
7c0f6ba6 99#include <linux/uaccess.h>
1da177e4
LT		 100#include <linux/skbuff.h>
101#include <linux/netdevice.h>
457c4cbc 102#include <net/net_namespace.h>
1da177e4 103#include <net/sock.h>
c752f073 104#include <net/tcp_states.h>
1da177e4
LT		 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>
1da177e4
LT		 111#include <linux/rtnetlink.h>
112#include <linux/mount.h>
113#include <net/checksum.h>
114#include <linux/security.h>
2b15af6f 115#include <linux/freezer.h>
ba94f308 116#include <linux/file.h>
2c860a43 117#include <linux/btf_ids.h>
1da177e4 118
f4e65870
JA		 119#include "scm.h"
120
518de9b3 121static atomic_long_t unix_nr_socks;
51bae889
KI		 122static struct hlist_head bsd_socket_buckets[UNIX_HASH_SIZE / 2];
123static spinlock_t bsd_socket_locks[UNIX_HASH_SIZE / 2];
1da177e4 124
f452be49 125/* SMP locking strategy:
2f7ca90a
KI		 126 * hash table is protected with spinlock.
127 * each socket state is protected by separate spinlock.
f452be49 128 */
1da177e4 129
f452be49 130static unsigned int unix_unbound_hash(struct sock *sk)
7123aaa3 131{
f452be49 132 unsigned long hash = (unsigned long)sk;
7123aaa3
ED		 133
134 hash ^= hash >> 16;
135 hash ^= hash >> 8;
f452be49
KI		 136 hash ^= sk->sk_type;
137
cf21b355 138 return hash & UNIX_HASH_MOD;
f452be49
KI		 139}
140
141static unsigned int unix_bsd_hash(struct inode *i)
142{
f302d180 143 return i->i_ino & UNIX_HASH_MOD;
f452be49
KI		 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
cf21b355 156 return UNIX_HASH_MOD + 1 + (hash & UNIX_HASH_MOD);
7123aaa3
ED		 157}
158
79b05bea
KI		 159static void unix_table_double_lock(struct net *net,
160 unsigned int hash1, unsigned int hash2)
afd20b92 161{
cf21b355
KI		 162 if (hash1 == hash2) {
163 spin_lock(&net->unx.table.locks[hash1]);
164 return;
165 }
166
afd20b92
KI		 167 if (hash1 > hash2)
168 swap(hash1, hash2);
169
79b05bea
KI		 170 spin_lock(&net->unx.table.locks[hash1]);
171 spin_lock_nested(&net->unx.table.locks[hash2], SINGLE_DEPTH_NESTING);
afd20b92
KI		 172}
173
79b05bea
KI		 174static void unix_table_double_unlock(struct net *net,
175 unsigned int hash1, unsigned int hash2)
afd20b92 176{
cf21b355
KI		 177 if (hash1 == hash2) {
178 spin_unlock(&net->unx.table.locks[hash1]);
179 return;
180 }
181
79b05bea
KI		 182 spin_unlock(&net->unx.table.locks[hash1]);
183 spin_unlock(&net->unx.table.locks[hash2]);
afd20b92
KI		 184}
185
877ce7c1 186#ifdef CONFIG_SECURITY_NETWORK
dc49c1f9 187static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
877ce7c1 188{
37a9a8df 189 UNIXCB(skb).secid = scm->secid;
877ce7c1
CZ		 190}
191
192static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
193{
37a9a8df
SS		 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);
877ce7c1
CZ		 200}
201#else
dc49c1f9 202static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
877ce7c1
CZ		 203{ }
204
205static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
206{ }
37a9a8df
SS		 207
208static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
209{
210 return true;
211}
877ce7c1
CZ		 212#endif /* CONFIG_SECURITY_NETWORK */
213
1da177e4
LT		 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{
6eba6a37 223 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
1da177e4
LT		 224}
225
86b18aaa 226static inline int unix_recvq_full(const struct sock *sk)
3c73419c
RW		 227{
228 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
229}
230
86b18aaa
QC		 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
fa7ff56f 237struct sock *unix_peer_get(struct sock *s)
1da177e4
LT		 238{
239 struct sock *peer;
240
1c92b4e5 241 unix_state_lock(s);
1da177e4
LT		 242 peer = unix_peer(s);
243 if (peer)
244 sock_hold(peer);
1c92b4e5 245 unix_state_unlock(s);
1da177e4
LT		 246 return peer;
247}
fa7ff56f 248EXPORT_SYMBOL_GPL(unix_peer_get);
1da177e4 249
12f21c49
KI		 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
1da177e4
LT		 266static inline void unix_release_addr(struct unix_address *addr)
267{
8c9814b9 268 if (refcount_dec_and_test(&addr->refcnt))
1da177e4
LT		 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 */
ac7bfa62 278
b8a58aa6
KI		 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
d2d8c9fd
KI		 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
1da177e4
LT		 304static void __unix_remove_socket(struct sock *sk)
305{
306 sk_del_node_init(sk);
307}
308
cf2f225e 309static void __unix_insert_socket(struct net *net, struct sock *sk)
1da177e4 310{
dd29c67d 311 DEBUG_NET_WARN_ON_ONCE(!sk_unhashed(sk));
cf2f225e 312 sk_add_node(sk, &net->unx.table.buckets[sk->sk_hash]);
1da177e4
LT		 313}
314
cf2f225e
KI		 315static void __unix_set_addr_hash(struct net *net, struct sock *sk,
316 struct unix_address *addr, unsigned int hash)
185ab886
AV		 317{
318 __unix_remove_socket(sk);
319 smp_store_release(&unix_sk(sk)->addr, addr);
e6b4b873
KI		 320
321 sk->sk_hash = hash;
cf2f225e 322 __unix_insert_socket(net, sk);
185ab886
AV		 323}
324
79b05bea 325static void unix_remove_socket(struct net *net, struct sock *sk)
1da177e4 326{
79b05bea 327 spin_lock(&net->unx.table.locks[sk->sk_hash]);
1da177e4 328 __unix_remove_socket(sk);
79b05bea 329 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
1da177e4
LT		 330}
331
79b05bea 332static void unix_insert_unbound_socket(struct net *net, struct sock *sk)
1da177e4 333{
79b05bea 334 spin_lock(&net->unx.table.locks[sk->sk_hash]);
cf2f225e 335 __unix_insert_socket(net, sk);
79b05bea 336 spin_unlock(&net->unx.table.locks[sk->sk_hash]);
1da177e4
LT		 337}
338
51bae889
KI		 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
097e66c5
DL		 357static struct sock *__unix_find_socket_byname(struct net *net,
358 struct sockaddr_un *sunname,
be752283 359 int len, unsigned int hash)
1da177e4
LT		 360{
361 struct sock *s;
1da177e4 362
cf2f225e 363 sk_for_each(s, &net->unx.table.buckets[hash]) {
1da177e4
LT		 364 struct unix_sock *u = unix_sk(s);
365
366 if (u->addr->len == len &&
367 !memcmp(u->addr->name, sunname, len))
262ce0af 368 return s;
1da177e4 369 }
262ce0af 370 return NULL;
1da177e4
LT		 371}
372
097e66c5
DL		 373static inline struct sock *unix_find_socket_byname(struct net *net,
374 struct sockaddr_un *sunname,
be752283 375 int len, unsigned int hash)
1da177e4
LT		 376{
377 struct sock *s;
378
79b05bea 379 spin_lock(&net->unx.table.locks[hash]);
be752283 380 s = __unix_find_socket_byname(net, sunname, len, hash);
1da177e4
LT		 381 if (s)
382 sock_hold(s);
79b05bea 383 spin_unlock(&net->unx.table.locks[hash]);
1da177e4
LT		 384 return s;
385}
386
51bae889 387static struct sock *unix_find_socket_byinode(struct inode *i)
1da177e4 388{
f452be49 389 unsigned int hash = unix_bsd_hash(i);
1da177e4 390 struct sock *s;
1da177e4 391
51bae889
KI		 392 spin_lock(&bsd_socket_locks[hash]);
393 sk_for_each_bound(s, &bsd_socket_buckets[hash]) {
40ffe67d 394 struct dentry *dentry = unix_sk(s)->path.dentry;
1da177e4 395
beef5121 396 if (dentry && d_backing_inode(dentry) == i) {
1da177e4 397 sock_hold(s);
51bae889 398 spin_unlock(&bsd_socket_locks[hash]);
afd20b92 399 return s;
1da177e4
LT		 400 }
401 }
51bae889 402 spin_unlock(&bsd_socket_locks[hash]);
afd20b92 403 return NULL;
1da177e4
LT		 404}
405
7d267278
RW		 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 *
ac6424b9 422 * In order to propagate a wake up, a wait_queue_entry_t of the client
7d267278
RW		 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
ac6424b9 431static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
7d267278
RW		 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)
3ad6f93e 446 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
7d267278
RW		 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),
a9a08845
LT		 494 EPOLLOUT |
495 EPOLLWRNORM |
496 EPOLLWRBAND);
7d267278
RW		 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
51f7e951
JB		 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 */
662a8094 514 if (unix_recvq_full_lockless(other) && !sock_flag(other, SOCK_DEAD))
7d267278
RW		 515 return 1;
516
517 if (connected)
518 unix_dgram_peer_wake_disconnect(sk, other);
519
520 return 0;
521}
522
1586a587 523static int unix_writable(const struct sock *sk)
1da177e4 524{
1586a587 525 return sk->sk_state != TCP_LISTEN &&
14afee4b 526 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
1da177e4
LT		 527}
528
529static void unix_write_space(struct sock *sk)
530{
43815482
ED		 531 struct socket_wq *wq;
532
533 rcu_read_lock();
1da177e4 534 if (unix_writable(sk)) {
43815482 535 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 536 if (skwq_has_sleeper(wq))
67426b75 537 wake_up_interruptible_sync_poll(&wq->wait,
a9a08845 538 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
8d8ad9d7 539 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4 540 }
43815482 541 rcu_read_unlock();
1da177e4
LT		 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{
b03efcfb 550 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1da177e4
LT		 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;
e3ae2365 560 sk_error_report(other);
1da177e4
LT		 561 }
562 }
dc56ad70 563 other->sk_state = TCP_CLOSE;
1da177e4
LT		 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
dd29c67d
ED		 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);
1da177e4 575 if (!sock_flag(sk, SOCK_DEAD)) {
5cc208be 576 pr_info("Attempt to release alive unix socket: %p\n", sk);
1da177e4
LT		 577 return;
578 }
579
580 if (u->addr)
581 unix_release_addr(u->addr);
582
518de9b3 583 atomic_long_dec(&unix_nr_socks);
a8076d8d 584 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1da177e4 585#ifdef UNIX_REFCNT_DEBUG
5cc208be 586 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
518de9b3 587 atomic_long_read(&unix_nr_socks));
1da177e4
LT		 588#endif
589}
590
ded34e0f 591static void unix_release_sock(struct sock *sk, int embrion)
1da177e4
LT		 592{
593 struct unix_sock *u = unix_sk(sk);
1da177e4
LT		 594 struct sock *skpair;
595 struct sk_buff *skb;
79b05bea 596 struct path path;
1da177e4
LT		 597 int state;
598
79b05bea 599 unix_remove_socket(sock_net(sk), sk);
51bae889 600 unix_remove_bsd_socket(sk);
1da177e4
LT		 601
602 /* Clear state */
1c92b4e5 603 unix_state_lock(sk);
1da177e4
LT		 604 sock_orphan(sk);
605 sk->sk_shutdown = SHUTDOWN_MASK;
40ffe67d
AV		 606 path = u->path;
607 u->path.dentry = NULL;
608 u->path.mnt = NULL;
1da177e4
LT		 609 state = sk->sk_state;
610 sk->sk_state = TCP_CLOSE;
a494bd64
ED		 611
612 skpair = unix_peer(sk);
613 unix_peer(sk) = NULL;
614
1c92b4e5 615 unix_state_unlock(sk);
1da177e4 616
7a62ed61
KI		 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
1da177e4
LT		 624 wake_up_interruptible_all(&u->peer_wait);
625
e27dfcea 626 if (skpair != NULL) {
1da177e4 627 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
1c92b4e5 628 unix_state_lock(skpair);
1da177e4
LT		 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;
1c92b4e5 633 unix_state_unlock(skpair);
1da177e4 634 skpair->sk_state_change(skpair);
8d8ad9d7 635 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
1da177e4 636 }
7d267278
RW		 637
638 unix_dgram_peer_wake_disconnect(sk, skpair);
1da177e4 639 sock_put(skpair); /* It may now die */
1da177e4
LT		 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) {
e27dfcea 645 if (state == TCP_LISTEN)
1da177e4
LT		 646 unix_release_sock(skb->sk, 1);
647 /* passed fds are erased in the kfree_skb hook */
73ed5d25 648 UNIXCB(skb).consumed = skb->len;
1da177e4
LT		 649 kfree_skb(skb);
650 }
651
40ffe67d
AV		 652 if (path.dentry)
653 path_put(&path);
1da177e4
LT		 654
655 sock_put(sk);
656
657 /* ---- Socket is dead now and most probably destroyed ---- */
658
659 /*
e04dae84 660 * Fixme: BSD difference: In BSD all sockets connected to us get
1da177e4
LT		 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
9305cfa4 670 if (unix_tot_inflight)
ac7bfa62 671 unix_gc(); /* Garbage collect fds */
1da177e4
LT		 672}
673
109f6e39
EB		 674static void init_peercred(struct sock *sk)
675{
35306eb2
ED		 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;
109f6e39
EB		 682 sk->sk_peer_pid = get_pid(task_tgid(current));
683 sk->sk_peer_cred = get_current_cred();
35306eb2
ED		 684 spin_unlock(&sk->sk_peer_lock);
685
686 put_pid(old_pid);
687 put_cred(old_cred);
109f6e39
EB		 688}
689
690static void copy_peercred(struct sock *sk, struct sock *peersk)
691{
35306eb2
ED		 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;
109f6e39
EB		 704 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
705 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
35306eb2
ED		 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);
109f6e39
EB		 712}
713
1da177e4
LT		 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;
6eba6a37
ED		 721 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
722 goto out; /* Only stream/seqpacket sockets accept */
1da177e4
LT		 723 err = -EINVAL;
724 if (!u->addr)
6eba6a37 725 goto out; /* No listens on an unbound socket */
1c92b4e5 726 unix_state_lock(sk);
1da177e4
LT		 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 */
109f6e39 734 init_peercred(sk);
1da177e4
LT		 735 err = 0;
736
737out_unlock:
1c92b4e5 738 unix_state_unlock(sk);
1da177e4
LT		 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 *);
cdfbabfb 748static int unix_accept(struct socket *, struct socket *, int, bool);
9b2c45d4 749static int unix_getname(struct socket *, struct sockaddr *, int);
a11e1d43
LT		 750static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
751static __poll_t unix_dgram_poll(struct file *, struct socket *,
752 poll_table *);
1da177e4 753static int unix_ioctl(struct socket *, unsigned int, unsigned long);
5f6beb9e
AB		 754#ifdef CONFIG_COMPAT
755static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
756#endif
1da177e4 757static int unix_shutdown(struct socket *, int);
1b784140
YX		 758static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
759static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
869e7c62
HFS		 760static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
761 size_t size, int flags);
2b514574
HFS		 762static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
763 struct pipe_inode_info *, size_t size,
764 unsigned int flags);
1b784140
YX		 765static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
766static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
965b57b4
CW		 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);
1da177e4
LT		 769static int unix_dgram_connect(struct socket *, struct sockaddr *,
770 int, int);
1b784140
YX		 771static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
772static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
773 int);
1da177e4 774
12663bfc 775static int unix_set_peek_off(struct sock *sk, int val)
f55bb7f9
PE		 776{
777 struct unix_sock *u = unix_sk(sk);
778
6e1ce3c3 779 if (mutex_lock_interruptible(&u->iolock))
12663bfc
SL		 780 return -EINTR;
781
f55bb7f9 782 sk->sk_peek_off = val;
6e1ce3c3 783 mutex_unlock(&u->iolock);
12663bfc
SL		 784
785 return 0;
f55bb7f9
PE		 786}
787
5c05a164 788#ifdef CONFIG_PROC_FS
de437089
KT		 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
3c32da19
KT		 807static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
808{
809 struct sock *sk = sock->sk;
810 struct unix_sock *u;
de437089 811 int nr_fds;
3c32da19
KT		 812
813 if (sk) {
de437089
KT		 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);
3c32da19
KT		 828 }
829}
3a12500e
TK		 830#else
831#define unix_show_fdinfo NULL
832#endif
f55bb7f9 833
90ddc4f0 834static const struct proto_ops unix_stream_ops = {
1da177e4
LT		 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,
a11e1d43 843 .poll = unix_poll,
1da177e4 844 .ioctl = unix_ioctl,
5f6beb9e
AB		 845#ifdef CONFIG_COMPAT
846 .compat_ioctl = unix_compat_ioctl,
847#endif
1da177e4
LT		 848 .listen = unix_listen,
849 .shutdown = unix_shutdown,
1da177e4
LT		 850 .sendmsg = unix_stream_sendmsg,
851 .recvmsg = unix_stream_recvmsg,
965b57b4 852 .read_skb = unix_stream_read_skb,
1da177e4 853 .mmap = sock_no_mmap,
869e7c62 854 .sendpage = unix_stream_sendpage,
2b514574 855 .splice_read = unix_stream_splice_read,
fc0d7536 856 .set_peek_off = unix_set_peek_off,
3c32da19 857 .show_fdinfo = unix_show_fdinfo,
1da177e4
LT		 858};
859
90ddc4f0 860static const struct proto_ops unix_dgram_ops = {
1da177e4
LT		 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,
a11e1d43 869 .poll = unix_dgram_poll,
1da177e4 870 .ioctl = unix_ioctl,
5f6beb9e
AB		 871#ifdef CONFIG_COMPAT
872 .compat_ioctl = unix_compat_ioctl,
873#endif
1da177e4
LT		 874 .listen = sock_no_listen,
875 .shutdown = unix_shutdown,
1da177e4 876 .sendmsg = unix_dgram_sendmsg,
965b57b4 877 .read_skb = unix_read_skb,
1da177e4
LT		 878 .recvmsg = unix_dgram_recvmsg,
879 .mmap = sock_no_mmap,
880 .sendpage = sock_no_sendpage,
f55bb7f9 881 .set_peek_off = unix_set_peek_off,
3c32da19 882 .show_fdinfo = unix_show_fdinfo,
1da177e4
LT		 883};
884
90ddc4f0 885static const struct proto_ops unix_seqpacket_ops = {
1da177e4
LT		 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,
a11e1d43 894 .poll = unix_dgram_poll,
1da177e4 895 .ioctl = unix_ioctl,
5f6beb9e
AB		 896#ifdef CONFIG_COMPAT
897 .compat_ioctl = unix_compat_ioctl,
898#endif
1da177e4
LT		 899 .listen = unix_listen,
900 .shutdown = unix_shutdown,
1da177e4 901 .sendmsg = unix_seqpacket_sendmsg,
a05d2ad1 902 .recvmsg = unix_seqpacket_recvmsg,
1da177e4
LT		 903 .mmap = sock_no_mmap,
904 .sendpage = sock_no_sendpage,
f55bb7f9 905 .set_peek_off = unix_set_peek_off,
3c32da19 906 .show_fdinfo = unix_show_fdinfo,
1da177e4
LT		 907};
908
c7272e15
CW		 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
94531cfc
JW		 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 = {
0edf0824 924 .name = "UNIX",
248969ae 925 .owner = THIS_MODULE,
248969ae 926 .obj_size = sizeof(struct unix_sock),
c7272e15 927 .close = unix_close,
c6382918 928#ifdef CONFIG_BPF_SYSCALL
94531cfc 929 .psock_update_sk_prot = unix_dgram_bpf_update_proto,
c6382918 930#endif
1da177e4
LT		 931};
932
94531cfc
JW		 933struct proto unix_stream_proto = {
934 .name = "UNIX-STREAM",
248969ae 935 .owner = THIS_MODULE,
248969ae 936 .obj_size = sizeof(struct unix_sock),
c7272e15 937 .close = unix_close,
94531cfc 938 .unhash = unix_unhash,
c6382918 939#ifdef CONFIG_BPF_SYSCALL
94531cfc 940 .psock_update_sk_prot = unix_stream_bpf_update_proto,
c6382918 941#endif
1da177e4
LT		 942};
943
94531cfc 944static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, int type)
1da177e4 945{
1da177e4 946 struct unix_sock *u;
f4bd73b5
KI		 947 struct sock *sk;
948 int err;
1da177e4 949
518de9b3 950 atomic_long_inc(&unix_nr_socks);
f4bd73b5
KI		 951 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) {
952 err = -ENFILE;
953 goto err;
954 }
1da177e4 955
94531cfc
JW		 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
f4bd73b5
KI		 961 if (!sk) {
962 err = -ENOMEM;
963 goto err;
964 }
1da177e4 965
6eba6a37 966 sock_init_data(sock, sk);
1da177e4 967
e6b4b873 968 sk->sk_hash = unix_unbound_hash(sk);
3aa9799e 969 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
1da177e4 970 sk->sk_write_space = unix_write_space;
a0a53c8b 971 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
1da177e4
LT		 972 sk->sk_destruct = unix_sock_destructor;
973 u = unix_sk(sk);
40ffe67d
AV		 974 u->path.dentry = NULL;
975 u->path.mnt = NULL;
fd19f329 976 spin_lock_init(&u->lock);
516e0cc5 977 atomic_long_set(&u->inflight, 0);
1fd05ba5 978 INIT_LIST_HEAD(&u->link);
6e1ce3c3
LT		 979 mutex_init(&u->iolock); /* single task reading lock */
980 mutex_init(&u->bindlock); /* single task binding lock */
1da177e4 981 init_waitqueue_head(&u->peer_wait);
7d267278 982 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
3c32da19 983 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
79b05bea 984 unix_insert_unbound_socket(net, sk);
f4bd73b5 985
340c3d33 986 sock_prot_inuse_add(net, sk->sk_prot, 1);
f4bd73b5 987
1da177e4 988 return sk;
f4bd73b5
KI		 989
990err:
991 atomic_long_dec(&unix_nr_socks);
992 return ERR_PTR(err);
1da177e4
LT		 993}
994
3f378b68
EP		 995static int unix_create(struct net *net, struct socket *sock, int protocol,
996 int kern)
1da177e4 997{
f4bd73b5
KI		 998 struct sock *sk;
999
1da177e4
LT		 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:
e27dfcea 1014 sock->type = SOCK_DGRAM;
df561f66 1015 fallthrough;
1da177e4
LT		 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
f4bd73b5
KI		 1026 sk = unix_create1(net, sock, kern, sock->type);
1027 if (IS_ERR(sk))
1028 return PTR_ERR(sk);
1029
1030 return 0;
1da177e4
LT		 1031}
1032
1033static int unix_release(struct socket *sock)
1034{
1035 struct sock *sk = sock->sk;
1036
1037 if (!sk)
1038 return 0;
1039
c7272e15 1040 sk->sk_prot->close(sk, 0);
ded34e0f 1041 unix_release_sock(sk, 0);
1da177e4
LT		 1042 sock->sk = NULL;
1043
ded34e0f 1044 return 0;
1da177e4
LT		 1045}
1046
51bae889
KI		 1047static struct sock *unix_find_bsd(struct sockaddr_un *sunaddr, int addr_len,
1048 int type)
fa39ef0e
KI		 1049{
1050 struct inode *inode;
1051 struct path path;
1052 struct sock *sk;
1053 int err;
1054
d2d8c9fd 1055 unix_mkname_bsd(sunaddr, addr_len);
fa39ef0e
KI		 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
51bae889 1069 sk = unix_find_socket_byinode(inode);
fa39ef0e
KI		 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:
aed26f55 1088 return ERR_PTR(err);
fa39ef0e
KI		 1089}
1090
1091static struct sock *unix_find_abstract(struct net *net,
1092 struct sockaddr_un *sunaddr,
d2d8c9fd 1093 int addr_len, int type)
fa39ef0e 1094{
f452be49 1095 unsigned int hash = unix_abstract_hash(sunaddr, addr_len, type);
fa39ef0e
KI		 1096 struct dentry *dentry;
1097 struct sock *sk;
1098
f452be49 1099 sk = unix_find_socket_byname(net, sunaddr, addr_len, hash);
aed26f55
KI		 1100 if (!sk)
1101 return ERR_PTR(-ECONNREFUSED);
fa39ef0e
KI		 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,
d2d8c9fd 1112 int addr_len, int type)
fa39ef0e
KI		 1113{
1114 struct sock *sk;
1115
1116 if (sunaddr->sun_path[0])
51bae889 1117 sk = unix_find_bsd(sunaddr, addr_len, type);
fa39ef0e 1118 else
d2d8c9fd 1119 sk = unix_find_abstract(net, sunaddr, addr_len, type);
fa39ef0e
KI		 1120
1121 return sk;
1122}
1123
f7ed31f4 1124static int unix_autobind(struct sock *sk)
1da177e4 1125{
afd20b92 1126 unsigned int new_hash, old_hash = sk->sk_hash;
1da177e4 1127 struct unix_sock *u = unix_sk(sk);
79b05bea 1128 struct net *net = sock_net(sk);
6eba6a37 1129 struct unix_address *addr;
9acbc584 1130 u32 lastnum, ordernum;
f7ed31f4 1131 int err;
1da177e4 1132
6e1ce3c3 1133 err = mutex_lock_interruptible(&u->bindlock);
37ab4fa7
SL		 1134 if (err)
1135 return err;
1da177e4 1136
1da177e4
LT		 1137 if (u->addr)
1138 goto out;
1139
1140 err = -ENOMEM;
755662ce
KI		 1141 addr = kzalloc(sizeof(*addr) +
1142 offsetof(struct sockaddr_un, sun_path) + 16, GFP_KERNEL);
1da177e4
LT		 1143 if (!addr)
1144 goto out;
1145
9acbc584 1146 addr->len = offsetof(struct sockaddr_un, sun_path) + 6;
1da177e4 1147 addr->name->sun_family = AF_UNIX;
8c9814b9 1148 refcount_set(&addr->refcnt, 1);
1da177e4 1149
a251c17a 1150 ordernum = get_random_u32();
9acbc584 1151 lastnum = ordernum & 0xFFFFF;
1da177e4 1152retry:
9acbc584
KI		 1153 ordernum = (ordernum + 1) & 0xFFFFF;
1154 sprintf(addr->name->sun_path + 1, "%05x", ordernum);
1da177e4 1155
e6b4b873 1156 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
79b05bea 1157 unix_table_double_lock(net, old_hash, new_hash);
1da177e4 1158
79b05bea
KI		 1159 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash)) {
1160 unix_table_double_unlock(net, old_hash, new_hash);
afd20b92 1161
9acbc584 1162 /* __unix_find_socket_byname() may take long time if many names
8df73ff9
TH		 1163 * are already in use.
1164 */
1165 cond_resched();
9acbc584
KI		 1166
1167 if (ordernum == lastnum) {
1168 /* Give up if all names seems to be in use. */
8df73ff9 1169 err = -ENOSPC;
9acbc584 1170 unix_release_addr(addr);
8df73ff9
TH		 1171 goto out;
1172 }
9acbc584 1173
1da177e4
LT		 1174 goto retry;
1175 }
1da177e4 1176
cf2f225e 1177 __unix_set_addr_hash(net, sk, addr, new_hash);
79b05bea 1178 unix_table_double_unlock(net, old_hash, new_hash);
1da177e4
LT		 1179 err = 0;
1180
6e1ce3c3 1181out: mutex_unlock(&u->bindlock);
1da177e4
LT		 1182 return err;
1183}
1184
12f21c49
KI		 1185static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
1186 int addr_len)
faf02010 1187{
71e6be6f
AV		 1188 umode_t mode = S_IFSOCK |
1189 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
afd20b92 1190 unsigned int new_hash, old_hash = sk->sk_hash;
12f21c49 1191 struct unix_sock *u = unix_sk(sk);
79b05bea 1192 struct net *net = sock_net(sk);
abf08576 1193 struct mnt_idmap *idmap;
12f21c49 1194 struct unix_address *addr;
38f7bd94 1195 struct dentry *dentry;
12f21c49 1196 struct path parent;
71e6be6f
AV		 1197 int err;
1198
12f21c49
KI		 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
38f7bd94
LT		 1207 /*
1208 * Get the parent directory, calculate the hash for last
1209 * component.
1210 */
71e6be6f 1211 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
12f21c49
KI		 1212 if (IS_ERR(dentry)) {
1213 err = PTR_ERR(dentry);
1214 goto out;
1215 }
faf02010 1216
38f7bd94
LT		 1217 /*
1218 * All right, let's create it.
1219 */
abf08576 1220 idmap = mnt_idmap(parent.mnt);
71e6be6f 1221 err = security_path_mknod(&parent, dentry, mode, 0);
56c1731b 1222 if (!err)
abf08576 1223 err = vfs_mknod(idmap, d_inode(parent.dentry), dentry, mode, 0);
c0c3b8d3 1224 if (err)
12f21c49 1225 goto out_path;
fa42d910 1226 err = mutex_lock_interruptible(&u->bindlock);
c0c3b8d3
AV		 1227 if (err)
1228 goto out_unlink;
1229 if (u->addr)
1230 goto out_unlock;
fa42d910 1231
e6b4b873 1232 new_hash = unix_bsd_hash(d_backing_inode(dentry));
79b05bea 1233 unix_table_double_lock(net, old_hash, new_hash);
56c1731b
AV		 1234 u->path.mnt = mntget(parent.mnt);
1235 u->path.dentry = dget(dentry);
cf2f225e 1236 __unix_set_addr_hash(net, sk, addr, new_hash);
79b05bea 1237 unix_table_double_unlock(net, old_hash, new_hash);
51bae889 1238 unix_insert_bsd_socket(sk);
fa42d910 1239 mutex_unlock(&u->bindlock);
56c1731b 1240 done_path_create(&parent, dentry);
fa42d910 1241 return 0;
c0c3b8d3
AV		 1242
1243out_unlock:
1244 mutex_unlock(&u->bindlock);
1245 err = -EINVAL;
1246out_unlink:
1247 /* failed after successful mknod? unlink what we'd created... */
abf08576 1248 vfs_unlink(idmap, d_inode(parent.dentry), dentry, NULL);
12f21c49 1249out_path:
c0c3b8d3 1250 done_path_create(&parent, dentry);
12f21c49
KI		 1251out:
1252 unix_release_addr(addr);
1253 return err == -EEXIST ? -EADDRINUSE : err;
fa42d910
AV		 1254}
1255
12f21c49
KI		 1256static int unix_bind_abstract(struct sock *sk, struct sockaddr_un *sunaddr,
1257 int addr_len)
fa42d910 1258{
afd20b92 1259 unsigned int new_hash, old_hash = sk->sk_hash;
fa42d910 1260 struct unix_sock *u = unix_sk(sk);
79b05bea 1261 struct net *net = sock_net(sk);
12f21c49 1262 struct unix_address *addr;
fa42d910
AV		 1263 int err;
1264
12f21c49
KI		 1265 addr = unix_create_addr(sunaddr, addr_len);
1266 if (!addr)
1267 return -ENOMEM;
1268
fa42d910
AV		 1269 err = mutex_lock_interruptible(&u->bindlock);
1270 if (err)
12f21c49 1271 goto out;
fa42d910
AV		 1272
1273 if (u->addr) {
12f21c49
KI		 1274 err = -EINVAL;
1275 goto out_mutex;
fa42d910
AV		 1276 }
1277
e6b4b873 1278 new_hash = unix_abstract_hash(addr->name, addr->len, sk->sk_type);
79b05bea 1279 unix_table_double_lock(net, old_hash, new_hash);
12f21c49 1280
79b05bea 1281 if (__unix_find_socket_byname(net, addr->name, addr->len, new_hash))
12f21c49
KI		 1282 goto out_spin;
1283
cf2f225e 1284 __unix_set_addr_hash(net, sk, addr, new_hash);
79b05bea 1285 unix_table_double_unlock(net, old_hash, new_hash);
fa42d910
AV		 1286 mutex_unlock(&u->bindlock);
1287 return 0;
12f21c49
KI		 1288
1289out_spin:
79b05bea 1290 unix_table_double_unlock(net, old_hash, new_hash);
12f21c49
KI		 1291 err = -EADDRINUSE;
1292out_mutex:
1293 mutex_unlock(&u->bindlock);
1294out:
1295 unix_release_addr(addr);
1296 return err;
fa42d910
AV		 1297}
1298
1da177e4
LT		 1299static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1300{
e27dfcea 1301 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
5c32a3ed 1302 struct sock *sk = sock->sk;
5c32a3ed 1303 int err;
1da177e4 1304
b8a58aa6
KI		 1305 if (addr_len == offsetof(struct sockaddr_un, sun_path) &&
1306 sunaddr->sun_family == AF_UNIX)
f7ed31f4 1307 return unix_autobind(sk);
1da177e4 1308
b8a58aa6
KI		 1309 err = unix_validate_addr(sunaddr, addr_len);
1310 if (err)
1311 return err;
1312
12f21c49
KI		 1313 if (sunaddr->sun_path[0])
1314 err = unix_bind_bsd(sk, sunaddr, addr_len);
fa42d910 1315 else
12f21c49
KI		 1316 err = unix_bind_abstract(sk, sunaddr, addr_len);
1317
1318 return err;
1da177e4
LT		 1319}
1320
278a3de5
DM		 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
1da177e4
LT		 1346static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1347 int alen, int flags)
1348{
e27dfcea 1349 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
340c3d33 1350 struct sock *sk = sock->sk;
1da177e4 1351 struct sock *other;
1da177e4
LT		 1352 int err;
1353
defbcf2d
MJ		 1354 err = -EINVAL;
1355 if (alen < offsetofend(struct sockaddr, sa_family))
1356 goto out;
1357
1da177e4 1358 if (addr->sa_family != AF_UNSPEC) {
b8a58aa6
KI		 1359 err = unix_validate_addr(sunaddr, alen);
1360 if (err)
1361 goto out;
1362
1da177e4 1363 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
f7ed31f4
KI		 1364 !unix_sk(sk)->addr) {
1365 err = unix_autobind(sk);
1366 if (err)
1367 goto out;
1368 }
1da177e4 1369
278a3de5 1370restart:
340c3d33 1371 other = unix_find_other(sock_net(sk), sunaddr, alen, sock->type);
aed26f55
KI		 1372 if (IS_ERR(other)) {
1373 err = PTR_ERR(other);
1da177e4 1374 goto out;
aed26f55 1375 }
1da177e4 1376
278a3de5
DM		 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 }
1da177e4
LT		 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
dc56ad70 1394 sk->sk_state = other->sk_state = TCP_ESTABLISHED;
1da177e4
LT		 1395 } else {
1396 /*
1397 * 1003.1g breaking connected state with AF_UNSPEC
1398 */
1399 other = NULL;
278a3de5 1400 unix_state_double_lock(sk, other);
1da177e4
LT		 1401 }
1402
1403 /*
1404 * If it was connected, reconnect.
1405 */
1406 if (unix_peer(sk)) {
1407 struct sock *old_peer = unix_peer(sk);
dc56ad70 1408
e27dfcea 1409 unix_peer(sk) = other;
dc56ad70
ED		 1410 if (!other)
1411 sk->sk_state = TCP_CLOSE;
7d267278
RW		 1412 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1413
278a3de5 1414 unix_state_double_unlock(sk, other);
1da177e4
LT		 1415
1416 if (other != old_peer)
1417 unix_dgram_disconnected(sk, old_peer);
1418 sock_put(old_peer);
1419 } else {
e27dfcea 1420 unix_peer(sk) = other;
278a3de5 1421 unix_state_double_unlock(sk, other);
1da177e4 1422 }
83301b53 1423
ac7bfa62 1424 return 0;
1da177e4
LT		 1425
1426out_unlock:
278a3de5 1427 unix_state_double_unlock(sk, other);
1da177e4
LT		 1428 sock_put(other);
1429out:
1430 return err;
1431}
1432
1433static long unix_wait_for_peer(struct sock *other, long timeo)
48851e9e 1434 __releases(&unix_sk(other)->lock)
1da177e4
LT		 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) &&
3c73419c 1444 unix_recvq_full(other);
1da177e4 1445
1c92b4e5 1446 unix_state_unlock(other);
1da177e4
LT		 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{
e27dfcea 1458 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
340c3d33 1459 struct sock *sk = sock->sk, *newsk = NULL, *other = NULL;
1da177e4 1460 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
340c3d33 1461 struct net *net = sock_net(sk);
1da177e4 1462 struct sk_buff *skb = NULL;
1da177e4 1463 long timeo;
340c3d33
KI		 1464 int err;
1465 int st;
1da177e4 1466
b8a58aa6
KI		 1467 err = unix_validate_addr(sunaddr, addr_len);
1468 if (err)
1469 goto out;
1470
f7ed31f4
KI		 1471 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr) {
1472 err = unix_autobind(sk);
1473 if (err)
1474 goto out;
1475 }
1da177e4
LT		 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
1da177e4 1484 /* create new sock for complete connection */
340c3d33 1485 newsk = unix_create1(net, NULL, 0, sock->type);
f4bd73b5
KI		 1486 if (IS_ERR(newsk)) {
1487 err = PTR_ERR(newsk);
1488 newsk = NULL;
1da177e4 1489 goto out;
f4bd73b5
KI		 1490 }
1491
1492 err = -ENOMEM;
1da177e4
LT		 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. */
d2d8c9fd 1501 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type);
aed26f55
KI		 1502 if (IS_ERR(other)) {
1503 err = PTR_ERR(other);
1504 other = NULL;
1da177e4 1505 goto out;
aed26f55 1506 }
1da177e4
LT		 1507
1508 /* Latch state of peer */
1c92b4e5 1509 unix_state_lock(other);
1da177e4
LT		 1510
1511 /* Apparently VFS overslept socket death. Retry. */
1512 if (sock_flag(other, SOCK_DEAD)) {
1c92b4e5 1513 unix_state_unlock(other);
1da177e4
LT		 1514 sock_put(other);
1515 goto restart;
1516 }
1517
1518 err = -ECONNREFUSED;
1519 if (other->sk_state != TCP_LISTEN)
1520 goto out_unlock;
77238f2b
TS		 1521 if (other->sk_shutdown & RCV_SHUTDOWN)
1522 goto out_unlock;
1da177e4 1523
3c73419c 1524 if (unix_recvq_full(other)) {
1da177e4
LT		 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;
ac7bfa62 1536 }
1da177e4
LT		 1537
1538 /* Latch our state.
1539
e5537bfc 1540 It is tricky place. We need to grab our state lock and cannot
1da177e4
LT		 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
1c92b4e5 1564 unix_state_lock_nested(sk);
1da177e4
LT		 1565
1566 if (sk->sk_state != st) {
1c92b4e5
DM		 1567 unix_state_unlock(sk);
1568 unix_state_unlock(other);
1da177e4
LT		 1569 sock_put(other);
1570 goto restart;
1571 }
1572
3610cda5 1573 err = security_unix_stream_connect(sk, other, newsk);
1da177e4 1574 if (err) {
1c92b4e5 1575 unix_state_unlock(sk);
1da177e4
LT		 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;
109f6e39 1585 init_peercred(newsk);
1da177e4 1586 newu = unix_sk(newsk);
eaefd110 1587 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1da177e4
LT		 1588 otheru = unix_sk(other);
1589
ae3b5641
AV		 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
2f7ca90a
KI		 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,
ae3b5641
AV		 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 */
40ffe67d
AV		 1607 if (otheru->path.dentry) {
1608 path_get(&otheru->path);
1609 newu->path = otheru->path;
1da177e4 1610 }
ae3b5641
AV		 1611 refcount_inc(&otheru->addr->refcnt);
1612 smp_store_release(&newu->addr, otheru->addr);
1da177e4
LT		 1613
1614 /* Set credentials */
109f6e39 1615 copy_peercred(sk, other);
1da177e4 1616
1da177e4
LT		 1617 sock->state = SS_CONNECTED;
1618 sk->sk_state = TCP_ESTABLISHED;
830a1e5c
BL		 1619 sock_hold(newsk);
1620
4e857c58 1621 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
830a1e5c 1622 unix_peer(sk) = newsk;
1da177e4 1623
1c92b4e5 1624 unix_state_unlock(sk);
1da177e4 1625
4e03d073 1626 /* take ten and send info to listening sock */
1da177e4
LT		 1627 spin_lock(&other->sk_receive_queue.lock);
1628 __skb_queue_tail(&other->sk_receive_queue, skb);
1da177e4 1629 spin_unlock(&other->sk_receive_queue.lock);
1c92b4e5 1630 unix_state_unlock(other);
676d2369 1631 other->sk_data_ready(other);
1da177e4
LT		 1632 sock_put(other);
1633 return 0;
1634
1635out_unlock:
1636 if (other)
1c92b4e5 1637 unix_state_unlock(other);
1da177e4
LT		 1638
1639out:
40d44446 1640 kfree_skb(skb);
1da177e4
LT		 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{
e27dfcea 1650 struct sock *ska = socka->sk, *skb = sockb->sk;
1da177e4
LT		 1651
1652 /* Join our sockets back to back */
1653 sock_hold(ska);
1654 sock_hold(skb);
e27dfcea
JK		 1655 unix_peer(ska) = skb;
1656 unix_peer(skb) = ska;
109f6e39
EB		 1657 init_peercred(ska);
1658 init_peercred(skb);
1da177e4 1659
83301b53
CW		 1660 ska->sk_state = TCP_ESTABLISHED;
1661 skb->sk_state = TCP_ESTABLISHED;
1662 socka->state = SS_CONNECTED;
1663 sockb->state = SS_CONNECTED;
1da177e4
LT		 1664 return 0;
1665}
1666
90c6bd34
DB		 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
cdfbabfb
DH		 1676static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1677 bool kern)
1da177e4
LT		 1678{
1679 struct sock *sk = sock->sk;
1680 struct sock *tsk;
1681 struct sk_buff *skb;
1682 int err;
1683
1684 err = -EOPNOTSUPP;
6eba6a37 1685 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1da177e4
LT		 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
f4b41f06
OH		 1696 skb = skb_recv_datagram(sk, (flags & O_NONBLOCK) ? MSG_DONTWAIT : 0,
1697 &err);
1da177e4
LT		 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 */
1c92b4e5 1710 unix_state_lock(tsk);
1da177e4 1711 newsock->state = SS_CONNECTED;
90c6bd34 1712 unix_sock_inherit_flags(sock, newsock);
1da177e4 1713 sock_graft(tsk, newsock);
1c92b4e5 1714 unix_state_unlock(tsk);
1da177e4
LT		 1715 return 0;
1716
1717out:
1718 return err;
1719}
1720
1721
9b2c45d4 1722static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1da177e4
LT		 1723{
1724 struct sock *sk = sock->sk;
ae3b5641 1725 struct unix_address *addr;
13cfa97b 1726 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1da177e4
LT		 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
ae3b5641
AV		 1740 addr = smp_load_acquire(&unix_sk(sk)->addr);
1741 if (!addr) {
1da177e4
LT		 1742 sunaddr->sun_family = AF_UNIX;
1743 sunaddr->sun_path[0] = 0;
755662ce 1744 err = offsetof(struct sockaddr_un, sun_path);
1da177e4 1745 } else {
9b2c45d4
DV		 1746 err = addr->len;
1747 memcpy(sunaddr, addr->name, addr->len);
1da177e4 1748 }
1da177e4
LT		 1749 sock_put(sk);
1750out:
1751 return err;
1752}
1753
cbcf0112
MS		 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
f78a5fda 1801static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
7361c36c
EB		 1802{
1803 int err = 0;
16e57262 1804
f78a5fda 1805 UNIXCB(skb).pid = get_pid(scm->pid);
6b0ee8c0
EB		 1806 UNIXCB(skb).uid = scm->creds.uid;
1807 UNIXCB(skb).gid = scm->creds.gid;
7361c36c 1808 UNIXCB(skb).fp = NULL;
37a9a8df 1809 unix_get_secdata(scm, skb);
7361c36c
EB		 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
9490f886
HFS		 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
16e57262
ED		 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{
6b0ee8c0 1833 if (UNIXCB(skb).pid)
16e57262 1834 return;
9490f886 1835 if (unix_passcred_enabled(sock, other)) {
16e57262 1836 UNIXCB(skb).pid = get_pid(task_tgid(current));
6e0895c2 1837 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
16e57262
ED		 1838 }
1839}
1840
9490f886
HFS		 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{
b146cbf2
KC		 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) &&
9490f886
HFS		 1865 unix_secdata_eq(scm, skb);
1866}
1867
3c32da19
KT		 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
3c32da19 1873 if (unlikely(fp && fp->count))
7782040b 1874 atomic_add(fp->count, &u->scm_stat.nr_fds);
3c32da19
KT		 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
3c32da19 1882 if (unlikely(fp && fp->count))
7782040b 1883 atomic_sub(fp->count, &u->scm_stat.nr_fds);
3c32da19
KT		 1884}
1885
1da177e4
LT		 1886/*
1887 * Send AF_UNIX data.
1888 */
1889
1b784140
YX		 1890static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1891 size_t len)
1da177e4 1892{
342dfc30 1893 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
340c3d33
KI		 1894 struct sock *sk = sock->sk, *other = NULL;
1895 struct unix_sock *u = unix_sk(sk);
7cc05662 1896 struct scm_cookie scm;
340c3d33 1897 struct sk_buff *skb;
eb6a2481 1898 int data_len = 0;
7d267278 1899 int sk_locked;
340c3d33
KI		 1900 long timeo;
1901 int err;
1da177e4 1902
5f23b734 1903 wait_for_unix_gc();
7cc05662 1904 err = scm_send(sock, msg, &scm, false);
1da177e4
LT		 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) {
b8a58aa6
KI		 1913 err = unix_validate_addr(sunaddr, msg->msg_namelen);
1914 if (err)
1915 goto out;
1da177e4
LT		 1916 } else {
1917 sunaddr = NULL;
1918 err = -ENOTCONN;
1919 other = unix_peer_get(sk);
1920 if (!other)
1921 goto out;
1922 }
1923
f7ed31f4
KI		 1924 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr) {
1925 err = unix_autobind(sk);
1926 if (err)
1927 goto out;
1928 }
1da177e4
LT		 1929
1930 err = -EMSGSIZE;
1931 if (len > sk->sk_sndbuf - 32)
1932 goto out;
1933
31ff6aa5 1934 if (len > SKB_MAX_ALLOC) {
eb6a2481
ED		 1935 data_len = min_t(size_t,
1936 len - SKB_MAX_ALLOC,
1937 MAX_SKB_FRAGS * PAGE_SIZE);
31ff6aa5
KT		 1938 data_len = PAGE_ALIGN(data_len);
1939
1940 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1941 }
eb6a2481
ED		 1942
1943 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
28d64271
ED		 1944 msg->msg_flags & MSG_DONTWAIT, &err,
1945 PAGE_ALLOC_COSTLY_ORDER);
e27dfcea 1946 if (skb == NULL)
1da177e4
LT		 1947 goto out;
1948
7cc05662 1949 err = unix_scm_to_skb(&scm, skb, true);
25888e30 1950 if (err < 0)
7361c36c 1951 goto out_free;
877ce7c1 1952
eb6a2481
ED		 1953 skb_put(skb, len - data_len);
1954 skb->data_len = data_len;
1955 skb->len = len;
c0371da6 1956 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1da177e4
LT		 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
340c3d33 1968 other = unix_find_other(sock_net(sk), sunaddr, msg->msg_namelen,
d2d8c9fd 1969 sk->sk_type);
aed26f55
KI		 1970 if (IS_ERR(other)) {
1971 err = PTR_ERR(other);
1972 other = NULL;
1da177e4 1973 goto out_free;
aed26f55 1974 }
1da177e4
LT		 1975 }
1976
d6ae3bae
AC		 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
7d267278 1983 sk_locked = 0;
1c92b4e5 1984 unix_state_lock(other);
7d267278 1985restart_locked:
1da177e4
LT		 1986 err = -EPERM;
1987 if (!unix_may_send(sk, other))
1988 goto out_unlock;
1989
7d267278 1990 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1da177e4
LT		 1991 /*
1992 * Check with 1003.1g - what should
1993 * datagram error
1994 */
1c92b4e5 1995 unix_state_unlock(other);
1da177e4
LT		 1996 sock_put(other);
1997
7d267278
RW		 1998 if (!sk_locked)
1999 unix_state_lock(sk);
2000
1da177e4 2001 err = 0;
3ff8bff7
KT		 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) {
e27dfcea 2010 unix_peer(sk) = NULL;
7d267278
RW		 2011 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
2012
3ff8bff7 2013 sk->sk_state = TCP_CLOSE;
1c92b4e5 2014 unix_state_unlock(sk);
1da177e4
LT		 2015
2016 unix_dgram_disconnected(sk, other);
2017 sock_put(other);
2018 err = -ECONNREFUSED;
2019 } else {
1c92b4e5 2020 unix_state_unlock(sk);
1da177e4
LT		 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
a5527dda
RW		 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 &&
86b18aaa
QC		 2044 unlikely(unix_peer(other) != sk &&
2045 unix_recvq_full_lockless(other))) {
7d267278
RW		 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;
1da177e4
LT		 2054 }
2055
7d267278
RW		 2056 if (!sk_locked) {
2057 unix_state_unlock(other);
2058 unix_state_double_lock(sk, other);
2059 }
1da177e4 2060
7d267278
RW		 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 }
1da177e4 2067
7d267278
RW		 2068 if (!sk_locked) {
2069 sk_locked = 1;
2070 goto restart_locked;
2071 }
1da177e4
LT		 2072 }
2073
7d267278
RW		 2074 if (unlikely(sk_locked))
2075 unix_state_unlock(sk);
2076
3f66116e
AC		 2077 if (sock_flag(other, SOCK_RCVTSTAMP))
2078 __net_timestamp(skb);
16e57262 2079 maybe_add_creds(skb, sock, other);
3c32da19 2080 scm_stat_add(other, skb);
7782040b 2081 skb_queue_tail(&other->sk_receive_queue, skb);
1c92b4e5 2082 unix_state_unlock(other);
676d2369 2083 other->sk_data_ready(other);
1da177e4 2084 sock_put(other);
7cc05662 2085 scm_destroy(&scm);
1da177e4
LT		 2086 return len;
2087
2088out_unlock:
7d267278
RW		 2089 if (sk_locked)
2090 unix_state_unlock(sk);
1c92b4e5 2091 unix_state_unlock(other);
1da177e4
LT		 2092out_free:
2093 kfree_skb(skb);
2094out:
2095 if (other)
2096 sock_put(other);
7cc05662 2097 scm_destroy(&scm);
1da177e4
LT		 2098 return err;
2099}
2100
e370a723 2101/* We use paged skbs for stream sockets, and limit occupancy to 32768
d4e9a408 2102 * bytes, and a minimum of a full page.
e370a723
ED		 2103 */
2104#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
ac7bfa62 2105
4edf21aa 2106#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
314001f0
RS		 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);
314001f0
RS		 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);
19eed721
RS		 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
314001f0
RS		 2135 maybe_add_creds(skb, sock, other);
2136 skb_get(skb);
2137
2138 if (ousk->oob_skb)
19eed721 2139 consume_skb(ousk->oob_skb);
314001f0 2140
e82025c6 2141 WRITE_ONCE(ousk->oob_skb, skb);
314001f0
RS		 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
1b784140
YX		 2153static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
2154 size_t len)
1da177e4 2155{
1da177e4
LT		 2156 struct sock *sk = sock->sk;
2157 struct sock *other = NULL;
6eba6a37 2158 int err, size;
f78a5fda 2159 struct sk_buff *skb;
e27dfcea 2160 int sent = 0;
7cc05662 2161 struct scm_cookie scm;
8ba69ba6 2162 bool fds_sent = false;
e370a723 2163 int data_len;
1da177e4 2164
5f23b734 2165 wait_for_unix_gc();
7cc05662 2166 err = scm_send(sock, msg, &scm, false);
1da177e4
LT		 2167 if (err < 0)
2168 return err;
2169
2170 err = -EOPNOTSUPP;
314001f0 2171 if (msg->msg_flags & MSG_OOB) {
4edf21aa 2172#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
314001f0
RS		 2173 if (len)
2174 len--;
2175 else
2176#endif
2177 goto out_err;
2178 }
1da177e4
LT		 2179
2180 if (msg->msg_namelen) {
2181 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
2182 goto out_err;
2183 } else {
1da177e4 2184 err = -ENOTCONN;
830a1e5c 2185 other = unix_peer(sk);
1da177e4
LT		 2186 if (!other)
2187 goto out_err;
2188 }
2189
2190 if (sk->sk_shutdown & SEND_SHUTDOWN)
2191 goto pipe_err;
2192
6eba6a37 2193 while (sent < len) {
e370a723 2194 size = len - sent;
1da177e4
LT		 2195
2196 /* Keep two messages in the pipe so it schedules better */
e370a723 2197 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1da177e4 2198
e370a723
ED		 2199 /* allow fallback to order-0 allocations */
2200 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
ac7bfa62 2201
e370a723 2202 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1da177e4 2203
31ff6aa5
KT		 2204 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
2205
e370a723 2206 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
28d64271
ED		 2207 msg->msg_flags & MSG_DONTWAIT, &err,
2208 get_order(UNIX_SKB_FRAGS_SZ));
e370a723 2209 if (!skb)
1da177e4
LT		 2210 goto out_err;
2211
f78a5fda 2212 /* Only send the fds in the first buffer */
7cc05662 2213 err = unix_scm_to_skb(&scm, skb, !fds_sent);
25888e30 2214 if (err < 0) {
7361c36c 2215 kfree_skb(skb);
f78a5fda 2216 goto out_err;
6209344f 2217 }
7361c36c 2218 fds_sent = true;
1da177e4 2219
e370a723
ED		 2220 skb_put(skb, size - data_len);
2221 skb->data_len = data_len;
2222 skb->len = size;
c0371da6 2223 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
6eba6a37 2224 if (err) {
1da177e4 2225 kfree_skb(skb);
f78a5fda 2226 goto out_err;
1da177e4
LT		 2227 }
2228
1c92b4e5 2229 unix_state_lock(other);
1da177e4
LT		 2230
2231 if (sock_flag(other, SOCK_DEAD) ||
2232 (other->sk_shutdown & RCV_SHUTDOWN))
2233 goto pipe_err_free;
2234
16e57262 2235 maybe_add_creds(skb, sock, other);
3c32da19 2236 scm_stat_add(other, skb);
7782040b 2237 skb_queue_tail(&other->sk_receive_queue, skb);
1c92b4e5 2238 unix_state_unlock(other);
676d2369 2239 other->sk_data_ready(other);
e27dfcea 2240 sent += size;
1da177e4 2241 }
1da177e4 2242
4edf21aa 2243#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
314001f0
RS		 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
7cc05662 2252 scm_destroy(&scm);
1da177e4
LT		 2253
2254 return sent;
2255
2256pipe_err_free:
1c92b4e5 2257 unix_state_unlock(other);
1da177e4
LT		 2258 kfree_skb(skb);
2259pipe_err:
6eba6a37
ED		 2260 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
2261 send_sig(SIGPIPE, current, 0);
1da177e4
LT		 2262 err = -EPIPE;
2263out_err:
7cc05662 2264 scm_destroy(&scm);
1da177e4
LT		 2265 return sent ? : err;
2266}
2267
869e7c62
HFS		 2268static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
2269 int offset, size_t size, int flags)
2270{
9490f886
HFS		 2271 int err;
2272 bool send_sigpipe = false;
2273 bool init_scm = true;
2274 struct scm_cookie scm;
869e7c62
HFS		 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);
6e1ce3c3 2288 mutex_unlock(&unix_sk(other)->iolock);
869e7c62
HFS		 2289 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
2290 &err, 0);
2291 if (!newskb)
9490f886 2292 goto err;
869e7c62
HFS		 2293 }
2294
6e1ce3c3 2295 /* we must acquire iolock as we modify already present
869e7c62
HFS		 2296 * skbs in the sk_receive_queue and mess with skb->len
2297 */
6e1ce3c3 2298 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
869e7c62
HFS		 2299 if (err) {
2300 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
869e7c62
HFS		 2301 goto err;
2302 }
2303
2304 if (sk->sk_shutdown & SEND_SHUTDOWN) {
2305 err = -EPIPE;
9490f886 2306 send_sigpipe = true;
869e7c62
HFS		 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;
9490f886 2315 send_sigpipe = true;
869e7c62
HFS		 2316 goto err_state_unlock;
2317 }
2318
9490f886
HFS		 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
869e7c62
HFS		 2326 skb = skb_peek_tail(&other->sk_receive_queue);
2327 if (tail && tail == skb) {
2328 skb = newskb;
9490f886
HFS		 2329 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2330 if (newskb) {
869e7c62 2331 skb = newskb;
9490f886
HFS		 2332 } else {
2333 tail = skb;
869e7c62 2334 goto alloc_skb;
9490f886 2335 }
869e7c62
HFS		 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);
8844f972 2342 newskb = NULL;
869e7c62
HFS		 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;
14afee4b 2353 refcount_add(size, &sk->sk_wmem_alloc);
869e7c62 2354
a3a116e0 2355 if (newskb) {
9490f886
HFS		 2356 err = unix_scm_to_skb(&scm, skb, false);
2357 if (err)
2358 goto err_state_unlock;
a3a116e0 2359 spin_lock(&other->sk_receive_queue.lock);
869e7c62 2360 __skb_queue_tail(&other->sk_receive_queue, newskb);
a3a116e0
HFS		 2361 spin_unlock(&other->sk_receive_queue.lock);
2362 }
869e7c62
HFS		 2363
2364 unix_state_unlock(other);
6e1ce3c3 2365 mutex_unlock(&unix_sk(other)->iolock);
869e7c62
HFS		 2366
2367 other->sk_data_ready(other);
9490f886 2368 scm_destroy(&scm);
869e7c62
HFS		 2369 return size;
2370
2371err_state_unlock:
2372 unix_state_unlock(other);
2373err_unlock:
6e1ce3c3 2374 mutex_unlock(&unix_sk(other)->iolock);
869e7c62
HFS		 2375err:
2376 kfree_skb(newskb);
2377 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2378 send_sig(SIGPIPE, current, 0);
9490f886
HFS		 2379 if (!init_scm)
2380 scm_destroy(&scm);
869e7c62
HFS		 2381 return err;
2382}
2383
1b784140
YX		 2384static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2385 size_t len)
1da177e4
LT		 2386{
2387 int err;
2388 struct sock *sk = sock->sk;
ac7bfa62 2389
1da177e4
LT		 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
1b784140 2400 return unix_dgram_sendmsg(sock, msg, len);
1da177e4 2401}
ac7bfa62 2402
1b784140
YX		 2403static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2404 size_t size, int flags)
a05d2ad1
EB		 2405{
2406 struct sock *sk = sock->sk;
2407
2408 if (sk->sk_state != TCP_ESTABLISHED)
2409 return -ENOTCONN;
2410
1b784140 2411 return unix_dgram_recvmsg(sock, msg, size, flags);
a05d2ad1
EB		 2412}
2413
1da177e4
LT		 2414static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2415{
ae3b5641 2416 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
1da177e4 2417
ae3b5641
AV		 2418 if (addr) {
2419 msg->msg_namelen = addr->len;
2420 memcpy(msg->msg_name, addr->name, addr->len);
1da177e4
LT		 2421 }
2422}
2423
9825d866
CW		 2424int __unix_dgram_recvmsg(struct sock *sk, struct msghdr *msg, size_t size,
2425 int flags)
1da177e4 2426{
7cc05662 2427 struct scm_cookie scm;
9825d866 2428 struct socket *sock = sk->sk_socket;
1da177e4 2429 struct unix_sock *u = unix_sk(sk);
64874280
RW		 2430 struct sk_buff *skb, *last;
2431 long timeo;
fd69c399 2432 int skip;
1da177e4
LT		 2433 int err;
2434
2435 err = -EOPNOTSUPP;
2436 if (flags&MSG_OOB)
2437 goto out;
2438
64874280 2439 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1da177e4 2440
64874280 2441 do {
6e1ce3c3 2442 mutex_lock(&u->iolock);
f55bb7f9 2443
64874280 2444 skip = sk_peek_offset(sk, flags);
b50b0580 2445 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
e427cad6
PA		 2446 &skip, &err, &last);
2447 if (skb) {
2448 if (!(flags & MSG_PEEK))
2449 scm_stat_del(sk, skb);
64874280 2450 break;
e427cad6 2451 }
64874280 2452
6e1ce3c3 2453 mutex_unlock(&u->iolock);
64874280
RW		 2454
2455 if (err != -EAGAIN)
2456 break;
2457 } while (timeo &&
b50b0580
SD		 2458 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2459 &err, &timeo, last));
64874280 2460
6e1ce3c3 2461 if (!skb) { /* implies iolock unlocked */
0a112258
FZ		 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);
64874280 2468 goto out;
0a112258 2469 }
1da177e4 2470
77b75f4d
RW		 2471 if (wq_has_sleeper(&u->peer_wait))
2472 wake_up_interruptible_sync_poll(&u->peer_wait,
a9a08845
LT		 2473 EPOLLOUT | EPOLLWRNORM |
2474 EPOLLWRBAND);
1da177e4
LT		 2475
2476 if (msg->msg_name)
2477 unix_copy_addr(msg, skb->sk);
2478
f55bb7f9
PE		 2479 if (size > skb->len - skip)
2480 size = skb->len - skip;
2481 else if (size < skb->len - skip)
1da177e4
LT		 2482 msg->msg_flags |= MSG_TRUNC;
2483
51f3d02b 2484 err = skb_copy_datagram_msg(skb, skip, msg, size);
1da177e4
LT		 2485 if (err)
2486 goto out_free;
2487
3f66116e
AC		 2488 if (sock_flag(sk, SOCK_RCVTSTAMP))
2489 __sock_recv_timestamp(msg, sk, skb);
2490
7cc05662
CH		 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);
1da177e4 2495
6eba6a37 2496 if (!(flags & MSG_PEEK)) {
1da177e4 2497 if (UNIXCB(skb).fp)
7cc05662 2498 unix_detach_fds(&scm, skb);
f55bb7f9
PE		 2499
2500 sk_peek_offset_bwd(sk, skb->len);
6eba6a37 2501 } else {
1da177e4
LT		 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)
ac7bfa62
YH		 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
1da177e4 2513 */
f55bb7f9
PE		 2514
2515 sk_peek_offset_fwd(sk, size);
2516
1da177e4 2517 if (UNIXCB(skb).fp)
cbcf0112 2518 unix_peek_fds(&scm, skb);
1da177e4 2519 }
9f6f9af7 2520 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1da177e4 2521
7cc05662 2522 scm_recv(sock, msg, &scm, flags);
1da177e4
LT		 2523
2524out_free:
6eba6a37 2525 skb_free_datagram(sk, skb);
6e1ce3c3 2526 mutex_unlock(&u->iolock);
1da177e4
LT		 2527out:
2528 return err;
2529}
29df44fa 2530
9825d866
CW		 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
94531cfc
JW		 2537 const struct proto *prot = READ_ONCE(sk->sk_prot);
2538
2539 if (prot != &unix_dgram_proto)
ec095263 2540 return prot->recvmsg(sk, msg, size, flags, NULL);
9825d866
CW		 2541#endif
2542 return __unix_dgram_recvmsg(sk, msg, size, flags);
2543}
2544
965b57b4 2545static int unix_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
29df44fa 2546{
d6e3b27c
PY		 2547 struct unix_sock *u = unix_sk(sk);
2548 struct sk_buff *skb;
2549 int err, copied;
29df44fa 2550
d6e3b27c
PY		 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;
29df44fa 2556
d6e3b27c
PY		 2557 copied = recv_actor(sk, skb);
2558 kfree_skb(skb);
29df44fa
CW		 2559
2560 return copied;
2561}
1da177e4
LT		 2562
2563/*
79f632c7 2564 * Sleep until more data has arrived. But check for races..
1da177e4 2565 */
79f632c7 2566static long unix_stream_data_wait(struct sock *sk, long timeo,
06a77b07
WC		 2567 struct sk_buff *last, unsigned int last_len,
2568 bool freezable)
1da177e4 2569{
f5d39b02 2570 unsigned int state = TASK_INTERRUPTIBLE | freezable * TASK_FREEZABLE;
2b514574 2571 struct sk_buff *tail;
1da177e4
LT		 2572 DEFINE_WAIT(wait);
2573
1c92b4e5 2574 unix_state_lock(sk);
1da177e4
LT		 2575
2576 for (;;) {
f5d39b02 2577 prepare_to_wait(sk_sleep(sk), &wait, state);
1da177e4 2578
2b514574
HFS		 2579 tail = skb_peek_tail(&sk->sk_receive_queue);
2580 if (tail != last ||
2581 (tail && tail->len != last_len) ||
1da177e4
LT		 2582 sk->sk_err ||
2583 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2584 signal_pending(current) ||
2585 !timeo)
2586 break;
2587
9cd3e072 2588 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1c92b4e5 2589 unix_state_unlock(sk);
f5d39b02 2590 timeo = schedule_timeout(timeo);
1c92b4e5 2591 unix_state_lock(sk);
b48732e4
MS		 2592
2593 if (sock_flag(sk, SOCK_DEAD))
2594 break;
2595
9cd3e072 2596 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1da177e4
LT		 2597 }
2598
aa395145 2599 finish_wait(sk_sleep(sk), &wait);
1c92b4e5 2600 unix_state_unlock(sk);
1da177e4
LT		 2601 return timeo;
2602}
2603
e370a723
ED		 2604static unsigned int unix_skb_len(const struct sk_buff *skb)
2605{
2606 return skb->len - UNIXCB(skb).consumed;
2607}
2608
2b514574
HFS		 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
314001f0
RS		 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;
876c14ad 2627 struct sk_buff *oob_skb;
314001f0 2628
876c14ad
RS		 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);
314001f0 2635 return -EINVAL;
876c14ad 2636 }
314001f0 2637
876c14ad 2638 oob_skb = u->oob_skb;
314001f0 2639
e82025c6
KI		 2640 if (!(state->flags & MSG_PEEK))
2641 WRITE_ONCE(u->oob_skb, NULL);
876c14ad
RS		 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
314001f0
RS		 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)) {
e82025c6 2676 WRITE_ONCE(u->oob_skb, NULL);
314001f0
RS		 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
965b57b4 2690static int unix_stream_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
77462de1
JW		 2691{
2692 if (unlikely(sk->sk_state != TCP_ESTABLISHED))
2693 return -ENOTCONN;
2694
965b57b4 2695 return unix_read_skb(sk, recv_actor);
77462de1
JW		 2696}
2697
06a77b07
WC		 2698static int unix_stream_read_generic(struct unix_stream_read_state *state,
2699 bool freezable)
1da177e4 2700{
7cc05662 2701 struct scm_cookie scm;
2b514574 2702 struct socket *sock = state->socket;
1da177e4
LT		 2703 struct sock *sk = sock->sk;
2704 struct unix_sock *u = unix_sk(sk);
1da177e4 2705 int copied = 0;
2b514574 2706 int flags = state->flags;
de144391 2707 int noblock = flags & MSG_DONTWAIT;
2b514574 2708 bool check_creds = false;
1da177e4
LT		 2709 int target;
2710 int err = 0;
2711 long timeo;
fc0d7536 2712 int skip;
2b514574
HFS		 2713 size_t size = state->size;
2714 unsigned int last_len;
1da177e4 2715
1b92ee3d
RW		 2716 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2717 err = -EINVAL;
1da177e4 2718 goto out;
1b92ee3d 2719 }
1da177e4 2720
1b92ee3d
RW		 2721 if (unlikely(flags & MSG_OOB)) {
2722 err = -EOPNOTSUPP;
314001f0 2723#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
314001f0 2724 err = unix_stream_recv_urg(state);
314001f0 2725#endif
1da177e4 2726 goto out;
1b92ee3d 2727 }
1da177e4 2728
2b514574 2729 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
de144391 2730 timeo = sock_rcvtimeo(sk, noblock);
1da177e4 2731
2b514574
HFS		 2732 memset(&scm, 0, sizeof(scm));
2733
1da177e4
LT		 2734 /* Lock the socket to prevent queue disordering
2735 * while sleeps in memcpy_tomsg
2736 */
6e1ce3c3 2737 mutex_lock(&u->iolock);
1da177e4 2738
a0917e0b 2739 skip = max(sk_peek_offset(sk, flags), 0);
e9193d60 2740
6eba6a37 2741 do {
1da177e4 2742 int chunk;
73ed5d25 2743 bool drop_skb;
79f632c7 2744 struct sk_buff *skb, *last;
1da177e4 2745
18eceb81 2746redo:
3c0d2f37 2747 unix_state_lock(sk);
b48732e4
MS		 2748 if (sock_flag(sk, SOCK_DEAD)) {
2749 err = -ECONNRESET;
2750 goto unlock;
2751 }
79f632c7 2752 last = skb = skb_peek(&sk->sk_receive_queue);
2b514574 2753 last_len = last ? last->len : 0;
314001f0
RS		 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
fc0d7536 2766again:
6eba6a37 2767 if (skb == NULL) {
1da177e4 2768 if (copied >= target)
3c0d2f37 2769 goto unlock;
1da177e4
LT		 2770
2771 /*
2772 * POSIX 1003.1g mandates this order.
2773 */
ac7bfa62 2774
6eba6a37
ED		 2775 err = sock_error(sk);
2776 if (err)
3c0d2f37 2777 goto unlock;
1da177e4 2778 if (sk->sk_shutdown & RCV_SHUTDOWN)
3c0d2f37
MS		 2779 goto unlock;
2780
2781 unix_state_unlock(sk);
1b92ee3d
RW		 2782 if (!timeo) {
2783 err = -EAGAIN;
1da177e4 2784 break;
1b92ee3d
RW		 2785 }
2786
6e1ce3c3 2787 mutex_unlock(&u->iolock);
1da177e4 2788
2b514574 2789 timeo = unix_stream_data_wait(sk, timeo, last,
06a77b07 2790 last_len, freezable);
1da177e4 2791
3822b5c2 2792 if (signal_pending(current)) {
1da177e4 2793 err = sock_intr_errno(timeo);
fa0dc04d 2794 scm_destroy(&scm);
1da177e4
LT		 2795 goto out;
2796 }
b3ca9b02 2797
6e1ce3c3 2798 mutex_lock(&u->iolock);
18eceb81 2799 goto redo;
2b514574 2800unlock:
3c0d2f37
MS		 2801 unix_state_unlock(sk);
2802 break;
1da177e4 2803 }
fc0d7536 2804
e370a723
ED		 2805 while (skip >= unix_skb_len(skb)) {
2806 skip -= unix_skb_len(skb);
79f632c7 2807 last = skb;
2b514574 2808 last_len = skb->len;
fc0d7536 2809 skb = skb_peek_next(skb, &sk->sk_receive_queue);
79f632c7
BP		 2810 if (!skb)
2811 goto again;
fc0d7536
PE		 2812 }
2813
3c0d2f37 2814 unix_state_unlock(sk);
1da177e4
LT		 2815
2816 if (check_creds) {
2817 /* Never glue messages from different writers */
9490f886 2818 if (!unix_skb_scm_eq(skb, &scm))
1da177e4 2819 break;
0e82e7f6 2820 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
1da177e4 2821 /* Copy credentials */
7cc05662 2822 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
37a9a8df 2823 unix_set_secdata(&scm, skb);
2b514574 2824 check_creds = true;
1da177e4
LT		 2825 }
2826
2827 /* Copy address just once */
2b514574
HFS		 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);
1da177e4
LT		 2832 sunaddr = NULL;
2833 }
2834
e370a723 2835 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
73ed5d25 2836 skb_get(skb);
2b514574 2837 chunk = state->recv_actor(skb, skip, chunk, state);
73ed5d25
HFS		 2838 drop_skb = !unix_skb_len(skb);
2839 /* skb is only safe to use if !drop_skb */
2840 consume_skb(skb);
2b514574 2841 if (chunk < 0) {
1da177e4
LT		 2842 if (copied == 0)
2843 copied = -EFAULT;
2844 break;
2845 }
2846 copied += chunk;
2847 size -= chunk;
2848
73ed5d25
HFS		 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
1da177e4 2861 /* Mark read part of skb as used */
6eba6a37 2862 if (!(flags & MSG_PEEK)) {
e370a723 2863 UNIXCB(skb).consumed += chunk;
1da177e4 2864
fc0d7536
PE		 2865 sk_peek_offset_bwd(sk, chunk);
2866
3c32da19 2867 if (UNIXCB(skb).fp) {
3c32da19 2868 scm_stat_del(sk, skb);
7cc05662 2869 unix_detach_fds(&scm, skb);
3c32da19 2870 }
1da177e4 2871
e370a723 2872 if (unix_skb_len(skb))
1da177e4 2873 break;
1da177e4 2874
6f01fd6e 2875 skb_unlink(skb, &sk->sk_receive_queue);
70d4bf6d 2876 consume_skb(skb);
1da177e4 2877
7cc05662 2878 if (scm.fp)
1da177e4 2879 break;
6eba6a37 2880 } else {
1da177e4
LT		 2881 /* It is questionable, see note in unix_dgram_recvmsg.
2882 */
2883 if (UNIXCB(skb).fp)
cbcf0112 2884 unix_peek_fds(&scm, skb);
1da177e4 2885
e9193d60 2886 sk_peek_offset_fwd(sk, chunk);
fc0d7536 2887
9f389e35
AC		 2888 if (UNIXCB(skb).fp)
2889 break;
2890
e9193d60 2891 skip = 0;
9f389e35
AC		 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);
1da177e4
LT		 2899 break;
2900 }
2901 } while (size);
2902
6e1ce3c3 2903 mutex_unlock(&u->iolock);
2b514574
HFS		 2904 if (state->msg)
2905 scm_recv(sock, state->msg, &scm, flags);
2906 else
2907 scm_destroy(&scm);
1da177e4
LT		 2908out:
2909 return copied ? : err;
2910}
2911
2b514574
HFS		 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
94531cfc
JW		 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
2b514574
HFS		 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
94531cfc
JW		 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)
ec095263 2953 return prot->recvmsg(sk, msg, size, flags, NULL);
94531cfc 2954#endif
06a77b07 2955 return unix_stream_read_generic(&state, true);
2b514574
HFS		 2956}
2957
2b514574
HFS		 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,
25869262 2964 state->pipe, chunk, state->splice_flags);
2b514574
HFS		 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
06a77b07 2986 return unix_stream_read_generic(&state, false);
2b514574
HFS		 2987}
2988
1da177e4
LT		 2989static int unix_shutdown(struct socket *sock, int mode)
2990{
2991 struct sock *sk = sock->sk;
2992 struct sock *other;
2993
fc61b928
XW		 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;
7180a031
AC		 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;
94531cfc 3015 const struct proto *prot = READ_ONCE(other->sk_prot);
7180a031 3016
d359902d
JW		 3017 if (prot->unhash)
3018 prot->unhash(other);
7180a031
AC		 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);
d0c6416b 3027 if (peer_mode == SHUTDOWN_MASK)
7180a031 3028 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
d0c6416b 3029 else if (peer_mode & RCV_SHUTDOWN)
7180a031 3030 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1da177e4 3031 }
7180a031
AC		 3032 if (other)
3033 sock_put(other);
3034
1da177e4
LT		 3035 return 0;
3036}
3037
885ee74d
PE		 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)
e370a723 3050 amount += unix_skb_len(skb);
885ee74d
PE		 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
ba94f308
AV		 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
ae3b5641
AV		 3077 if (!smp_load_acquire(&unix_sk(sk)->addr))
3078 return -ENOENT;
3079
ba94f308 3080 path = unix_sk(sk)->path;
ae3b5641 3081 if (!path.dentry)
ba94f308 3082 return -ENOENT;
ba94f308
AV		 3083
3084 path_get(&path);
ba94f308
AV		 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
1da177e4
LT		 3104static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3105{
3106 struct sock *sk = sock->sk;
e27dfcea 3107 long amount = 0;
1da177e4
LT		 3108 int err;
3109
6eba6a37
ED		 3110 switch (cmd) {
3111 case SIOCOUTQ:
885ee74d 3112 amount = unix_outq_len(sk);
6eba6a37
ED		 3113 err = put_user(amount, (int __user *)arg);
3114 break;
3115 case SIOCINQ:
885ee74d
PE		 3116 amount = unix_inq_len(sk);
3117 if (amount < 0)
3118 err = amount;
3119 else
1da177e4 3120 err = put_user(amount, (int __user *)arg);
885ee74d 3121 break;
ba94f308
AV		 3122 case SIOCUNIXFILE:
3123 err = unix_open_file(sk);
3124 break;
314001f0
RS		 3125#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3126 case SIOCATMARK:
3127 {
3128 struct sk_buff *skb;
314001f0
RS		 3129 int answ = 0;
3130
3131 skb = skb_peek(&sk->sk_receive_queue);
e82025c6 3132 if (skb && skb == READ_ONCE(unix_sk(sk)->oob_skb))
314001f0
RS		 3133 answ = 1;
3134 err = put_user(answ, (int __user *)arg);
3135 }
3136 break;
3137#endif
6eba6a37
ED		 3138 default:
3139 err = -ENOIOCTLCMD;
3140 break;
1da177e4
LT		 3141 }
3142 return err;
3143}
3144
5f6beb9e
AB		 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
a11e1d43 3152static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1da177e4
LT		 3153{
3154 struct sock *sk = sock->sk;
a11e1d43
LT		 3155 __poll_t mask;
3156
89ab066d 3157 sock_poll_wait(file, sock, wait);
a11e1d43 3158 mask = 0;
1da177e4
LT		 3159
3160 /* exceptional events? */
3161 if (sk->sk_err)
a9a08845 3162 mask |= EPOLLERR;
1da177e4 3163 if (sk->sk_shutdown == SHUTDOWN_MASK)
a9a08845 3164 mask |= EPOLLHUP;
f348d70a 3165 if (sk->sk_shutdown & RCV_SHUTDOWN)
a9a08845 3166 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
1da177e4
LT		 3167
3168 /* readable? */
3ef7cf57 3169 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
a9a08845 3170 mask |= EPOLLIN | EPOLLRDNORM;
af493388
CW		 3171 if (sk_is_readable(sk))
3172 mask |= EPOLLIN | EPOLLRDNORM;
d9a232d4
KI		 3173#if IS_ENABLED(CONFIG_AF_UNIX_OOB)
3174 if (READ_ONCE(unix_sk(sk)->oob_skb))
3175 mask |= EPOLLPRI;
3176#endif
1da177e4
LT		 3177
3178 /* Connection-based need to check for termination and startup */
6eba6a37
ED		 3179 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
3180 sk->sk_state == TCP_CLOSE)
a9a08845 3181 mask |= EPOLLHUP;
1da177e4
LT		 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))
a9a08845 3188 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1da177e4
LT		 3189
3190 return mask;
3191}
3192
a11e1d43
LT		 3193static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
3194 poll_table *wait)
3c73419c 3195{
ec0d215f 3196 struct sock *sk = sock->sk, *other;
a11e1d43
LT		 3197 unsigned int writable;
3198 __poll_t mask;
3199
89ab066d 3200 sock_poll_wait(file, sock, wait);
a11e1d43 3201 mask = 0;
3c73419c
RW		 3202
3203 /* exceptional events? */
3ef7cf57 3204 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
a9a08845
LT		 3205 mask |= EPOLLERR |
3206 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
7d4c04fc 3207
3c73419c 3208 if (sk->sk_shutdown & RCV_SHUTDOWN)
a9a08845 3209 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
3c73419c 3210 if (sk->sk_shutdown == SHUTDOWN_MASK)
a9a08845 3211 mask |= EPOLLHUP;
3c73419c
RW		 3212
3213 /* readable? */
3ef7cf57 3214 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
a9a08845 3215 mask |= EPOLLIN | EPOLLRDNORM;
af493388
CW		 3216 if (sk_is_readable(sk))
3217 mask |= EPOLLIN | EPOLLRDNORM;
3c73419c
RW		 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)
a9a08845 3222 mask |= EPOLLHUP;
3c73419c
RW		 3223 /* connection hasn't started yet? */
3224 if (sk->sk_state == TCP_SYN_SENT)
3225 return mask;
3226 }
3227
973a34aa 3228 /* No write status requested, avoid expensive OUT tests. */
a11e1d43 3229 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
973a34aa
ED		 3230 return mask;
3231
ec0d215f 3232 writable = unix_writable(sk);
7d267278
RW		 3233 if (writable) {
3234 unix_state_lock(sk);
3235
3236 other = unix_peer(sk);
3237 if (other && unix_peer(other) != sk &&
04f08eb4 3238 unix_recvq_full_lockless(other) &&
7d267278
RW		 3239 unix_dgram_peer_wake_me(sk, other))
3240 writable = 0;
3241
3242 unix_state_unlock(sk);
ec0d215f
RW		 3243 }
3244
3245 if (writable)
a9a08845 3246 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
3c73419c 3247 else
9cd3e072 3248 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
3c73419c 3249
3c73419c
RW		 3250 return mask;
3251}
1da177e4
LT		 3252
3253#ifdef CONFIG_PROC_FS
a53eb3fe 3254
7123aaa3
ED		 3255#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
3256
3257#define get_bucket(x) ((x) >> BUCKET_SPACE)
afd20b92 3258#define get_offset(x) ((x) & ((1UL << BUCKET_SPACE) - 1))
7123aaa3 3259#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
a53eb3fe 3260
7123aaa3 3261static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
1da177e4 3262{
7123aaa3
ED		 3263 unsigned long offset = get_offset(*pos);
3264 unsigned long bucket = get_bucket(*pos);
7123aaa3 3265 unsigned long count = 0;
cf2f225e 3266 struct sock *sk;
1da177e4 3267
cf2f225e
KI		 3268 for (sk = sk_head(&seq_file_net(seq)->unx.table.buckets[bucket]);
3269 sk; sk = sk_next(sk)) {
7123aaa3
ED		 3270 if (++count == offset)
3271 break;
3272 }
3273
3274 return sk;
3275}
3276
4408d55a 3277static struct sock *unix_get_first(struct seq_file *seq, loff_t *pos)
7123aaa3 3278{
afd20b92 3279 unsigned long bucket = get_bucket(*pos);
79b05bea 3280 struct net *net = seq_file_net(seq);
4408d55a 3281 struct sock *sk;
7123aaa3 3282
f302d180 3283 while (bucket < UNIX_HASH_SIZE) {
79b05bea 3284 spin_lock(&net->unx.table.locks[bucket]);
4408d55a 3285
7123aaa3
ED		 3286 sk = unix_from_bucket(seq, pos);
3287 if (sk)
3288 return sk;
3289
79b05bea 3290 spin_unlock(&net->unx.table.locks[bucket]);
4408d55a
KI		 3291
3292 *pos = set_bucket_offset(++bucket, 1);
3293 }
7123aaa3 3294
1da177e4
LT		 3295 return NULL;
3296}
3297
4408d55a
KI		 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
cf2f225e
KI		 3303 sk = sk_next(sk);
3304 if (sk)
3305 return sk;
3306
4408d55a 3307
cf2f225e 3308 spin_unlock(&seq_file_net(seq)->unx.table.locks[bucket]);
4408d55a
KI		 3309
3310 *pos = set_bucket_offset(++bucket, 1);
3311
3312 return unix_get_first(seq, pos);
3313}
3314
1da177e4
LT		 3315static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
3316{
7123aaa3
ED		 3317 if (!*pos)
3318 return SEQ_START_TOKEN;
3319
4408d55a 3320 return unix_get_first(seq, pos);
1da177e4
LT		 3321}
3322
3323static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3324{
3325 ++*pos;
4408d55a
KI		 3326
3327 if (v == SEQ_START_TOKEN)
3328 return unix_get_first(seq, pos);
3329
3330 return unix_get_next(seq, v, pos);
1da177e4
LT		 3331}
3332
3333static void unix_seq_stop(struct seq_file *seq, void *v)
3334{
afd20b92
KI		 3335 struct sock *sk = v;
3336
2f7ca90a 3337 if (sk)
79b05bea 3338 spin_unlock(&seq_file_net(seq)->unx.table.locks[sk->sk_hash]);
1da177e4
LT		 3339}
3340
3341static int unix_seq_show(struct seq_file *seq, void *v)
3342{
ac7bfa62 3343
b9f3124f 3344 if (v == SEQ_START_TOKEN)
1da177e4
LT		 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);
1c92b4e5 3350 unix_state_lock(s);
1da177e4 3351
71338aa7 3352 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
1da177e4 3353 s,
41c6d650 3354 refcount_read(&s->sk_refcnt),
1da177e4
LT		 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
2f7ca90a 3363 if (u->addr) { // under a hash table lock here
1da177e4
LT		 3364 int i, len;
3365 seq_putc(seq, ' ');
3366
3367 i = 0;
755662ce
KI		 3368 len = u->addr->len -
3369 offsetof(struct sockaddr_un, sun_path);
5ce7ab49 3370 if (u->addr->name->sun_path[0]) {
1da177e4 3371 len--;
5ce7ab49 3372 } else {
1da177e4
LT		 3373 seq_putc(seq, '@');
3374 i++;
3375 }
3376 for ( ; i < len; i++)
e7947ea7
IB		 3377 seq_putc(seq, u->addr->name->sun_path[i] ?:
3378 '@');
1da177e4 3379 }
1c92b4e5 3380 unix_state_unlock(s);
1da177e4
LT		 3381 seq_putc(seq, '\n');
3382 }
3383
3384 return 0;
3385}
3386
56b3d975 3387static const struct seq_operations unix_seq_ops = {
1da177e4
LT		 3388 .start = unix_seq_start,
3389 .next = unix_seq_next,
3390 .stop = unix_seq_stop,
3391 .show = unix_seq_show,
3392};
2c860a43
KI		 3393
3394#if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL)
855d8e77
KI		 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
2c860a43
KI		 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
855d8e77
KI		 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)) {
855d8e77
KI		 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
cf2f225e 3441 spin_unlock(&seq_file_net(seq)->unx.table.locks[start_sk->sk_hash]);
855d8e77
KI		 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
2c860a43
KI		 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;
855d8e77
KI		 3544 bool slow;
3545 int ret;
2c860a43
KI		 3546
3547 if (v == SEQ_START_TOKEN)
3548 return 0;
3549
855d8e77
KI		 3550 slow = lock_sock_fast(sk);
3551
3552 if (unlikely(sk_unhashed(sk))) {
3553 ret = SEQ_SKIP;
3554 goto unlock;
3555 }
3556
2c860a43
KI		 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);
855d8e77
KI		 3560 ret = unix_prog_seq_show(prog, &meta, v, uid);
3561unlock:
3562 unlock_sock_fast(sk, slow);
3563 return ret;
2c860a43
KI		 3564}
3565
3566static void bpf_iter_unix_seq_stop(struct seq_file *seq, void *v)
3567{
855d8e77 3568 struct bpf_unix_iter_state *iter = seq->private;
2c860a43
KI		 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
855d8e77
KI		 3579 if (iter->cur_sk < iter->end_sk)
3580 bpf_iter_unix_put_batch(iter);
2c860a43
KI		 3581}
3582
3583static const struct seq_operations bpf_iter_unix_seq_ops = {
855d8e77
KI		 3584 .start = bpf_iter_unix_seq_start,
3585 .next = bpf_iter_unix_seq_next,
2c860a43
KI		 3586 .stop = bpf_iter_unix_seq_stop,
3587 .show = bpf_iter_unix_seq_show,
3588};
3589#endif
1da177e4
LT		 3590#endif
3591
ec1b4cf7 3592static const struct net_proto_family unix_family_ops = {
1da177e4
LT		 3593 .family = PF_UNIX,
3594 .create = unix_create,
3595 .owner = THIS_MODULE,
3596};
3597
097e66c5 3598
2c8c1e72 3599static int __net_init unix_net_init(struct net *net)
097e66c5 3600{
b6e81138 3601 int i;
097e66c5 3602
a0a53c8b 3603 net->unx.sysctl_max_dgram_qlen = 10;
1597fbc0
PE		 3604 if (unix_sysctl_register(net))
3605 goto out;
d392e497 3606
097e66c5 3607#ifdef CONFIG_PROC_FS
c3506372 3608 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
b6e81138
KI		 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]);
1597fbc0 3627 }
b6e81138
KI		 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:
097e66c5 3637#endif
b6e81138 3638 unix_sysctl_unregister(net);
097e66c5 3639out:
b6e81138 3640 return -ENOMEM;
097e66c5
DL		 3641}
3642
2c8c1e72 3643static void __net_exit unix_net_exit(struct net *net)
097e66c5 3644{
b6e81138
KI		 3645 kvfree(net->unx.table.buckets);
3646 kvfree(net->unx.table.locks);
1597fbc0 3647 unix_sysctl_unregister(net);
ece31ffd 3648 remove_proc_entry("unix", net->proc_net);
097e66c5
DL		 3649}
3650
3651static struct pernet_operations unix_net_ops = {
3652 .init = unix_net_init,
3653 .exit = unix_net_exit,
3654};
3655
2c860a43
KI		 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
855d8e77
KI		 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
2c860a43
KI		 3688static const struct bpf_iter_seq_info unix_seq_info = {
3689 .seq_ops = &bpf_iter_unix_seq_ops,
855d8e77
KI		 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),
2c860a43
KI		 3693};
3694
eb7d8f1d
KI		 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
2c860a43
KI		 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 },
eb7d8f1d 3716 .get_func_proto = bpf_iter_unix_get_func_proto,
2c860a43
KI		 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
1da177e4
LT		 3728static int __init af_unix_init(void)
3729{
51bae889 3730 int i, rc = -1;
1da177e4 3731
c593642c 3732 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
1da177e4 3733
51bae889
KI		 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
94531cfc
JW		 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);
ac7bfa62 3746 if (rc != 0) {
5cc208be 3747 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
73e341e0 3748 proto_unregister(&unix_dgram_proto);
1da177e4
LT		 3749 goto out;
3750 }
3751
3752 sock_register(&unix_family_ops);
097e66c5 3753 register_pernet_subsys(&unix_net_ops);
c6382918 3754 unix_bpf_build_proto();
2c860a43
KI		 3755
3756#if IS_BUILTIN(CONFIG_UNIX) && defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3757 bpf_iter_register();
3758#endif
3759
1da177e4
LT		 3760out:
3761 return rc;
3762}
3763
3764static void __exit af_unix_exit(void)
3765{
3766 sock_unregister(PF_UNIX);
94531cfc
JW		 3767 proto_unregister(&unix_dgram_proto);
3768 proto_unregister(&unix_stream_proto);
097e66c5 3769 unregister_pernet_subsys(&unix_net_ops);
1da177e4
LT		 3770}
3771
3d366960
DW		 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);
1da177e4
LT		 3777module_exit(af_unix_exit);
3778
3779MODULE_LICENSE("GPL");
3780MODULE_ALIAS_NETPROTO(PF_UNIX);
Linux 6.x block layer and io_uring tree(s)