?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
flow_dissector: fix vlan tag handling
[linux-2.6/btrfs-unstable.git] / net / core / sock.c
blobc73e28fc9c2a45225af39f7905072456d4380c7f
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
120
121 #include <asm/uaccess.h>
122
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134 #include <linux/sock_diag.h>
135
136 #include <linux/filter.h>
137 #include <net/sock_reuseport.h>
138
139 #include <trace/events/sock.h>
140
141 #ifdef CONFIG_INET
142 #include <net/tcp.h>
143 #endif
144
145 #include <net/busy_poll.h>
146
147 static DEFINE_MUTEX(proto_list_mutex);
148 static LIST_HEAD(proto_list);
149
150 /**
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
155 *
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
159 */
160 bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
162 {
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
165 }
166 EXPORT_SYMBOL(sk_ns_capable);
167
168 /**
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
171 * @cap: The global capability to use
172 *
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
175 * namespaces.
176 */
177 bool sk_capable(const struct sock *sk, int cap)
178 {
179 return sk_ns_capable(sk, &init_user_ns, cap);
180 }
181 EXPORT_SYMBOL(sk_capable);
182
183 /**
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
187 *
188 * Test to see if the opener of the socket had when the socket was created
189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
191 */
192 bool sk_net_capable(const struct sock *sk, int cap)
193 {
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 }
196 EXPORT_SYMBOL(sk_net_capable);
197
198 /*
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
201 */
202 static struct lock_class_key af_family_keys[AF_MAX];
203 static struct lock_class_key af_family_slock_keys[AF_MAX];
204
205 /*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
210 static const char *const af_family_key_strings[AF_MAX+1] = {
211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_MAX"
226 };
227 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_MAX"
243 };
244 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_MAX"
260 };
261
262 /*
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
265 */
266 static struct lock_class_key af_callback_keys[AF_MAX];
267
268 /* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
272 */
273 #define _SK_MEM_PACKETS 256
274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
277
278 /* Run time adjustable parameters. */
279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
280 EXPORT_SYMBOL(sysctl_wmem_max);
281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
282 EXPORT_SYMBOL(sysctl_rmem_max);
283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
285
286 /* Maximal space eaten by iovec or ancillary data plus some space */
287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
288 EXPORT_SYMBOL(sysctl_optmem_max);
289
290 int sysctl_tstamp_allow_data __read_mostly = 1;
291
292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293 EXPORT_SYMBOL_GPL(memalloc_socks);
294
295 /**
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
298 *
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
302 */
303 void sk_set_memalloc(struct sock *sk)
304 {
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
307 static_key_slow_inc(&memalloc_socks);
308 }
309 EXPORT_SYMBOL_GPL(sk_set_memalloc);
310
311 void sk_clear_memalloc(struct sock *sk)
312 {
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
315 static_key_slow_dec(&memalloc_socks);
316
317 /*
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
323 */
324 sk_mem_reclaim(sk);
325 }
326 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327
328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329 {
330 int ret;
331 unsigned long pflags = current->flags;
332
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339
340 return ret;
341 }
342 EXPORT_SYMBOL(__sk_backlog_rcv);
343
344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345 {
346 struct timeval tv;
347
348 if (optlen < sizeof(tv))
349 return -EINVAL;
350 if (copy_from_user(&tv, optval, sizeof(tv)))
351 return -EFAULT;
352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 return -EDOM;
354
355 if (tv.tv_sec < 0) {
356 static int warned __read_mostly;
357
358 *timeo_p = 0;
359 if (warned < 10 && net_ratelimit()) {
360 warned++;
361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
363 }
364 return 0;
365 }
366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
368 return 0;
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 return 0;
372 }
373
374 static void sock_warn_obsolete_bsdism(const char *name)
375 {
376 static int warned;
377 static char warncomm[TASK_COMM_LEN];
378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
381 warncomm, name);
382 warned++;
383 }
384 }
385
386 static bool sock_needs_netstamp(const struct sock *sk)
387 {
388 switch (sk->sk_family) {
389 case AF_UNSPEC:
390 case AF_UNIX:
391 return false;
392 default:
393 return true;
394 }
395 }
396
397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
398 {
399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
403 net_disable_timestamp();
404 }
405 }
406
407
408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
409 {
410 unsigned long flags;
411 struct sk_buff_head *list = &sk->sk_receive_queue;
412
413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
414 atomic_inc(&sk->sk_drops);
415 trace_sock_rcvqueue_full(sk, skb);
416 return -ENOMEM;
417 }
418
419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
420 atomic_inc(&sk->sk_drops);
421 return -ENOBUFS;
422 }
423
424 skb->dev = NULL;
425 skb_set_owner_r(skb, sk);
426
427 /* we escape from rcu protected region, make sure we dont leak
428 * a norefcounted dst
429 */
430 skb_dst_force(skb);
431
432 spin_lock_irqsave(&list->lock, flags);
433 sock_skb_set_dropcount(sk, skb);
434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
436
437 if (!sock_flag(sk, SOCK_DEAD))
438 sk->sk_data_ready(sk);
439 return 0;
440 }
441 EXPORT_SYMBOL(__sock_queue_rcv_skb);
442
443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
444 {
445 int err;
446
447 err = sk_filter(sk, skb);
448 if (err)
449 return err;
450
451 return __sock_queue_rcv_skb(sk, skb);
452 }
453 EXPORT_SYMBOL(sock_queue_rcv_skb);
454
455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap)
457 {
458 int rc = NET_RX_SUCCESS;
459
460 if (sk_filter_trim_cap(sk, skb, trim_cap))
461 goto discard_and_relse;
462
463 skb->dev = NULL;
464
465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
468 }
469 if (nested)
470 bh_lock_sock_nested(sk);
471 else
472 bh_lock_sock(sk);
473 if (!sock_owned_by_user(sk)) {
474 /*
475 * trylock + unlock semantics:
476 */
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
478
479 rc = sk_backlog_rcv(sk, skb);
480
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
483 bh_unlock_sock(sk);
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
486 }
487
488 bh_unlock_sock(sk);
489 out:
490 sock_put(sk);
491 return rc;
492 discard_and_relse:
493 kfree_skb(skb);
494 goto out;
495 }
496 EXPORT_SYMBOL(__sk_receive_skb);
497
498 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
499 {
500 struct dst_entry *dst = __sk_dst_get(sk);
501
502 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
503 sk_tx_queue_clear(sk);
504 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
505 dst_release(dst);
506 return NULL;
507 }
508
509 return dst;
510 }
511 EXPORT_SYMBOL(__sk_dst_check);
512
513 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
514 {
515 struct dst_entry *dst = sk_dst_get(sk);
516
517 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
518 sk_dst_reset(sk);
519 dst_release(dst);
520 return NULL;
521 }
522
523 return dst;
524 }
525 EXPORT_SYMBOL(sk_dst_check);
526
527 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
528 int optlen)
529 {
530 int ret = -ENOPROTOOPT;
531 #ifdef CONFIG_NETDEVICES
532 struct net *net = sock_net(sk);
533 char devname[IFNAMSIZ];
534 int index;
535
536 /* Sorry... */
537 ret = -EPERM;
538 if (!ns_capable(net->user_ns, CAP_NET_RAW))
539 goto out;
540
541 ret = -EINVAL;
542 if (optlen < 0)
543 goto out;
544
545 /* Bind this socket to a particular device like "eth0",
546 * as specified in the passed interface name. If the
547 * name is "" or the option length is zero the socket
548 * is not bound.
549 */
550 if (optlen > IFNAMSIZ - 1)
551 optlen = IFNAMSIZ - 1;
552 memset(devname, 0, sizeof(devname));
553
554 ret = -EFAULT;
555 if (copy_from_user(devname, optval, optlen))
556 goto out;
557
558 index = 0;
559 if (devname[0] != '\0') {
560 struct net_device *dev;
561
562 rcu_read_lock();
563 dev = dev_get_by_name_rcu(net, devname);
564 if (dev)
565 index = dev->ifindex;
566 rcu_read_unlock();
567 ret = -ENODEV;
568 if (!dev)
569 goto out;
570 }
571
572 lock_sock(sk);
573 sk->sk_bound_dev_if = index;
574 sk_dst_reset(sk);
575 release_sock(sk);
576
577 ret = 0;
578
579 out:
580 #endif
581
582 return ret;
583 }
584
585 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
586 int __user *optlen, int len)
587 {
588 int ret = -ENOPROTOOPT;
589 #ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
591 char devname[IFNAMSIZ];
592
593 if (sk->sk_bound_dev_if == 0) {
594 len = 0;
595 goto zero;
596 }
597
598 ret = -EINVAL;
599 if (len < IFNAMSIZ)
600 goto out;
601
602 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
603 if (ret)
604 goto out;
605
606 len = strlen(devname) + 1;
607
608 ret = -EFAULT;
609 if (copy_to_user(optval, devname, len))
610 goto out;
611
612 zero:
613 ret = -EFAULT;
614 if (put_user(len, optlen))
615 goto out;
616
617 ret = 0;
618
619 out:
620 #endif
621
622 return ret;
623 }
624
625 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
626 {
627 if (valbool)
628 sock_set_flag(sk, bit);
629 else
630 sock_reset_flag(sk, bit);
631 }
632
633 bool sk_mc_loop(struct sock *sk)
634 {
635 if (dev_recursion_level())
636 return false;
637 if (!sk)
638 return true;
639 switch (sk->sk_family) {
640 case AF_INET:
641 return inet_sk(sk)->mc_loop;
642 #if IS_ENABLED(CONFIG_IPV6)
643 case AF_INET6:
644 return inet6_sk(sk)->mc_loop;
645 #endif
646 }
647 WARN_ON(1);
648 return true;
649 }
650 EXPORT_SYMBOL(sk_mc_loop);
651
652 /*
653 * This is meant for all protocols to use and covers goings on
654 * at the socket level. Everything here is generic.
655 */
656
657 int sock_setsockopt(struct socket *sock, int level, int optname,
658 char __user *optval, unsigned int optlen)
659 {
660 struct sock *sk = sock->sk;
661 int val;
662 int valbool;
663 struct linger ling;
664 int ret = 0;
665
666 /*
667 * Options without arguments
668 */
669
670 if (optname == SO_BINDTODEVICE)
671 return sock_setbindtodevice(sk, optval, optlen);
672
673 if (optlen < sizeof(int))
674 return -EINVAL;
675
676 if (get_user(val, (int __user *)optval))
677 return -EFAULT;
678
679 valbool = val ? 1 : 0;
680
681 lock_sock(sk);
682
683 switch (optname) {
684 case SO_DEBUG:
685 if (val && !capable(CAP_NET_ADMIN))
686 ret = -EACCES;
687 else
688 sock_valbool_flag(sk, SOCK_DBG, valbool);
689 break;
690 case SO_REUSEADDR:
691 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
692 break;
693 case SO_REUSEPORT:
694 sk->sk_reuseport = valbool;
695 break;
696 case SO_TYPE:
697 case SO_PROTOCOL:
698 case SO_DOMAIN:
699 case SO_ERROR:
700 ret = -ENOPROTOOPT;
701 break;
702 case SO_DONTROUTE:
703 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
704 break;
705 case SO_BROADCAST:
706 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
707 break;
708 case SO_SNDBUF:
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
713 */
714 val = min_t(u32, val, sysctl_wmem_max);
715 set_sndbuf:
716 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
717 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
718 /* Wake up sending tasks if we upped the value. */
719 sk->sk_write_space(sk);
720 break;
721
722 case SO_SNDBUFFORCE:
723 if (!capable(CAP_NET_ADMIN)) {
724 ret = -EPERM;
725 break;
726 }
727 goto set_sndbuf;
728
729 case SO_RCVBUF:
730 /* Don't error on this BSD doesn't and if you think
731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
734 */
735 val = min_t(u32, val, sysctl_rmem_max);
736 set_rcvbuf:
737 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
738 /*
739 * We double it on the way in to account for
740 * "struct sk_buff" etc. overhead. Applications
741 * assume that the SO_RCVBUF setting they make will
742 * allow that much actual data to be received on that
743 * socket.
744 *
745 * Applications are unaware that "struct sk_buff" and
746 * other overheads allocate from the receive buffer
747 * during socket buffer allocation.
748 *
749 * And after considering the possible alternatives,
750 * returning the value we actually used in getsockopt
751 * is the most desirable behavior.
752 */
753 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
754 break;
755
756 case SO_RCVBUFFORCE:
757 if (!capable(CAP_NET_ADMIN)) {
758 ret = -EPERM;
759 break;
760 }
761 goto set_rcvbuf;
762
763 case SO_KEEPALIVE:
764 #ifdef CONFIG_INET
765 if (sk->sk_protocol == IPPROTO_TCP &&
766 sk->sk_type == SOCK_STREAM)
767 tcp_set_keepalive(sk, valbool);
768 #endif
769 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
770 break;
771
772 case SO_OOBINLINE:
773 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
774 break;
775
776 case SO_NO_CHECK:
777 sk->sk_no_check_tx = valbool;
778 break;
779
780 case SO_PRIORITY:
781 if ((val >= 0 && val <= 6) ||
782 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
783 sk->sk_priority = val;
784 else
785 ret = -EPERM;
786 break;
787
788 case SO_LINGER:
789 if (optlen < sizeof(ling)) {
790 ret = -EINVAL; /* 1003.1g */
791 break;
792 }
793 if (copy_from_user(&ling, optval, sizeof(ling))) {
794 ret = -EFAULT;
795 break;
796 }
797 if (!ling.l_onoff)
798 sock_reset_flag(sk, SOCK_LINGER);
799 else {
800 #if (BITS_PER_LONG == 32)
801 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
802 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
803 else
804 #endif
805 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
806 sock_set_flag(sk, SOCK_LINGER);
807 }
808 break;
809
810 case SO_BSDCOMPAT:
811 sock_warn_obsolete_bsdism("setsockopt");
812 break;
813
814 case SO_PASSCRED:
815 if (valbool)
816 set_bit(SOCK_PASSCRED, &sock->flags);
817 else
818 clear_bit(SOCK_PASSCRED, &sock->flags);
819 break;
820
821 case SO_TIMESTAMP:
822 case SO_TIMESTAMPNS:
823 if (valbool) {
824 if (optname == SO_TIMESTAMP)
825 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
826 else
827 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
828 sock_set_flag(sk, SOCK_RCVTSTAMP);
829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
830 } else {
831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
833 }
834 break;
835
836 case SO_TIMESTAMPING:
837 if (val & ~SOF_TIMESTAMPING_MASK) {
838 ret = -EINVAL;
839 break;
840 }
841
842 if (val & SOF_TIMESTAMPING_OPT_ID &&
843 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
844 if (sk->sk_protocol == IPPROTO_TCP &&
845 sk->sk_type == SOCK_STREAM) {
846 if ((1 << sk->sk_state) &
847 (TCPF_CLOSE | TCPF_LISTEN)) {
848 ret = -EINVAL;
849 break;
850 }
851 sk->sk_tskey = tcp_sk(sk)->snd_una;
852 } else {
853 sk->sk_tskey = 0;
854 }
855 }
856 sk->sk_tsflags = val;
857 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
858 sock_enable_timestamp(sk,
859 SOCK_TIMESTAMPING_RX_SOFTWARE);
860 else
861 sock_disable_timestamp(sk,
862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
863 break;
864
865 case SO_RCVLOWAT:
866 if (val < 0)
867 val = INT_MAX;
868 sk->sk_rcvlowat = val ? : 1;
869 break;
870
871 case SO_RCVTIMEO:
872 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
873 break;
874
875 case SO_SNDTIMEO:
876 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
877 break;
878
879 case SO_ATTACH_FILTER:
880 ret = -EINVAL;
881 if (optlen == sizeof(struct sock_fprog)) {
882 struct sock_fprog fprog;
883
884 ret = -EFAULT;
885 if (copy_from_user(&fprog, optval, sizeof(fprog)))
886 break;
887
888 ret = sk_attach_filter(&fprog, sk);
889 }
890 break;
891
892 case SO_ATTACH_BPF:
893 ret = -EINVAL;
894 if (optlen == sizeof(u32)) {
895 u32 ufd;
896
897 ret = -EFAULT;
898 if (copy_from_user(&ufd, optval, sizeof(ufd)))
899 break;
900
901 ret = sk_attach_bpf(ufd, sk);
902 }
903 break;
904
905 case SO_ATTACH_REUSEPORT_CBPF:
906 ret = -EINVAL;
907 if (optlen == sizeof(struct sock_fprog)) {
908 struct sock_fprog fprog;
909
910 ret = -EFAULT;
911 if (copy_from_user(&fprog, optval, sizeof(fprog)))
912 break;
913
914 ret = sk_reuseport_attach_filter(&fprog, sk);
915 }
916 break;
917
918 case SO_ATTACH_REUSEPORT_EBPF:
919 ret = -EINVAL;
920 if (optlen == sizeof(u32)) {
921 u32 ufd;
922
923 ret = -EFAULT;
924 if (copy_from_user(&ufd, optval, sizeof(ufd)))
925 break;
926
927 ret = sk_reuseport_attach_bpf(ufd, sk);
928 }
929 break;
930
931 case SO_DETACH_FILTER:
932 ret = sk_detach_filter(sk);
933 break;
934
935 case SO_LOCK_FILTER:
936 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
937 ret = -EPERM;
938 else
939 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
940 break;
941
942 case SO_PASSSEC:
943 if (valbool)
944 set_bit(SOCK_PASSSEC, &sock->flags);
945 else
946 clear_bit(SOCK_PASSSEC, &sock->flags);
947 break;
948 case SO_MARK:
949 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
950 ret = -EPERM;
951 else
952 sk->sk_mark = val;
953 break;
954
955 case SO_RXQ_OVFL:
956 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
957 break;
958
959 case SO_WIFI_STATUS:
960 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
961 break;
962
963 case SO_PEEK_OFF:
964 if (sock->ops->set_peek_off)
965 ret = sock->ops->set_peek_off(sk, val);
966 else
967 ret = -EOPNOTSUPP;
968 break;
969
970 case SO_NOFCS:
971 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
972 break;
973
974 case SO_SELECT_ERR_QUEUE:
975 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
976 break;
977
978 #ifdef CONFIG_NET_RX_BUSY_POLL
979 case SO_BUSY_POLL:
980 /* allow unprivileged users to decrease the value */
981 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
982 ret = -EPERM;
983 else {
984 if (val < 0)
985 ret = -EINVAL;
986 else
987 sk->sk_ll_usec = val;
988 }
989 break;
990 #endif
991
992 case SO_MAX_PACING_RATE:
993 sk->sk_max_pacing_rate = val;
994 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
995 sk->sk_max_pacing_rate);
996 break;
997
998 case SO_INCOMING_CPU:
999 sk->sk_incoming_cpu = val;
1000 break;
1001
1002 case SO_CNX_ADVICE:
1003 if (val == 1)
1004 dst_negative_advice(sk);
1005 break;
1006 default:
1007 ret = -ENOPROTOOPT;
1008 break;
1009 }
1010 release_sock(sk);
1011 return ret;
1012 }
1013 EXPORT_SYMBOL(sock_setsockopt);
1014
1015
1016 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1017 struct ucred *ucred)
1018 {
1019 ucred->pid = pid_vnr(pid);
1020 ucred->uid = ucred->gid = -1;
1021 if (cred) {
1022 struct user_namespace *current_ns = current_user_ns();
1023
1024 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1025 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1026 }
1027 }
1028
1029 int sock_getsockopt(struct socket *sock, int level, int optname,
1030 char __user *optval, int __user *optlen)
1031 {
1032 struct sock *sk = sock->sk;
1033
1034 union {
1035 int val;
1036 struct linger ling;
1037 struct timeval tm;
1038 } v;
1039
1040 int lv = sizeof(int);
1041 int len;
1042
1043 if (get_user(len, optlen))
1044 return -EFAULT;
1045 if (len < 0)
1046 return -EINVAL;
1047
1048 memset(&v, 0, sizeof(v));
1049
1050 switch (optname) {
1051 case SO_DEBUG:
1052 v.val = sock_flag(sk, SOCK_DBG);
1053 break;
1054
1055 case SO_DONTROUTE:
1056 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1057 break;
1058
1059 case SO_BROADCAST:
1060 v.val = sock_flag(sk, SOCK_BROADCAST);
1061 break;
1062
1063 case SO_SNDBUF:
1064 v.val = sk->sk_sndbuf;
1065 break;
1066
1067 case SO_RCVBUF:
1068 v.val = sk->sk_rcvbuf;
1069 break;
1070
1071 case SO_REUSEADDR:
1072 v.val = sk->sk_reuse;
1073 break;
1074
1075 case SO_REUSEPORT:
1076 v.val = sk->sk_reuseport;
1077 break;
1078
1079 case SO_KEEPALIVE:
1080 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1081 break;
1082
1083 case SO_TYPE:
1084 v.val = sk->sk_type;
1085 break;
1086
1087 case SO_PROTOCOL:
1088 v.val = sk->sk_protocol;
1089 break;
1090
1091 case SO_DOMAIN:
1092 v.val = sk->sk_family;
1093 break;
1094
1095 case SO_ERROR:
1096 v.val = -sock_error(sk);
1097 if (v.val == 0)
1098 v.val = xchg(&sk->sk_err_soft, 0);
1099 break;
1100
1101 case SO_OOBINLINE:
1102 v.val = sock_flag(sk, SOCK_URGINLINE);
1103 break;
1104
1105 case SO_NO_CHECK:
1106 v.val = sk->sk_no_check_tx;
1107 break;
1108
1109 case SO_PRIORITY:
1110 v.val = sk->sk_priority;
1111 break;
1112
1113 case SO_LINGER:
1114 lv = sizeof(v.ling);
1115 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1116 v.ling.l_linger = sk->sk_lingertime / HZ;
1117 break;
1118
1119 case SO_BSDCOMPAT:
1120 sock_warn_obsolete_bsdism("getsockopt");
1121 break;
1122
1123 case SO_TIMESTAMP:
1124 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1125 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1126 break;
1127
1128 case SO_TIMESTAMPNS:
1129 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1130 break;
1131
1132 case SO_TIMESTAMPING:
1133 v.val = sk->sk_tsflags;
1134 break;
1135
1136 case SO_RCVTIMEO:
1137 lv = sizeof(struct timeval);
1138 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1139 v.tm.tv_sec = 0;
1140 v.tm.tv_usec = 0;
1141 } else {
1142 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1144 }
1145 break;
1146
1147 case SO_SNDTIMEO:
1148 lv = sizeof(struct timeval);
1149 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1150 v.tm.tv_sec = 0;
1151 v.tm.tv_usec = 0;
1152 } else {
1153 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1154 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1155 }
1156 break;
1157
1158 case SO_RCVLOWAT:
1159 v.val = sk->sk_rcvlowat;
1160 break;
1161
1162 case SO_SNDLOWAT:
1163 v.val = 1;
1164 break;
1165
1166 case SO_PASSCRED:
1167 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1168 break;
1169
1170 case SO_PEERCRED:
1171 {
1172 struct ucred peercred;
1173 if (len > sizeof(peercred))
1174 len = sizeof(peercred);
1175 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1176 if (copy_to_user(optval, &peercred, len))
1177 return -EFAULT;
1178 goto lenout;
1179 }
1180
1181 case SO_PEERNAME:
1182 {
1183 char address[128];
1184
1185 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1186 return -ENOTCONN;
1187 if (lv < len)
1188 return -EINVAL;
1189 if (copy_to_user(optval, address, len))
1190 return -EFAULT;
1191 goto lenout;
1192 }
1193
1194 /* Dubious BSD thing... Probably nobody even uses it, but
1195 * the UNIX standard wants it for whatever reason... -DaveM
1196 */
1197 case SO_ACCEPTCONN:
1198 v.val = sk->sk_state == TCP_LISTEN;
1199 break;
1200
1201 case SO_PASSSEC:
1202 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1203 break;
1204
1205 case SO_PEERSEC:
1206 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1207
1208 case SO_MARK:
1209 v.val = sk->sk_mark;
1210 break;
1211
1212 case SO_RXQ_OVFL:
1213 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1214 break;
1215
1216 case SO_WIFI_STATUS:
1217 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1218 break;
1219
1220 case SO_PEEK_OFF:
1221 if (!sock->ops->set_peek_off)
1222 return -EOPNOTSUPP;
1223
1224 v.val = sk->sk_peek_off;
1225 break;
1226 case SO_NOFCS:
1227 v.val = sock_flag(sk, SOCK_NOFCS);
1228 break;
1229
1230 case SO_BINDTODEVICE:
1231 return sock_getbindtodevice(sk, optval, optlen, len);
1232
1233 case SO_GET_FILTER:
1234 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1235 if (len < 0)
1236 return len;
1237
1238 goto lenout;
1239
1240 case SO_LOCK_FILTER:
1241 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1242 break;
1243
1244 case SO_BPF_EXTENSIONS:
1245 v.val = bpf_tell_extensions();
1246 break;
1247
1248 case SO_SELECT_ERR_QUEUE:
1249 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1250 break;
1251
1252 #ifdef CONFIG_NET_RX_BUSY_POLL
1253 case SO_BUSY_POLL:
1254 v.val = sk->sk_ll_usec;
1255 break;
1256 #endif
1257
1258 case SO_MAX_PACING_RATE:
1259 v.val = sk->sk_max_pacing_rate;
1260 break;
1261
1262 case SO_INCOMING_CPU:
1263 v.val = sk->sk_incoming_cpu;
1264 break;
1265
1266 default:
1267 /* We implement the SO_SNDLOWAT etc to not be settable
1268 * (1003.1g 7).
1269 */
1270 return -ENOPROTOOPT;
1271 }
1272
1273 if (len > lv)
1274 len = lv;
1275 if (copy_to_user(optval, &v, len))
1276 return -EFAULT;
1277 lenout:
1278 if (put_user(len, optlen))
1279 return -EFAULT;
1280 return 0;
1281 }
1282
1283 /*
1284 * Initialize an sk_lock.
1285 *
1286 * (We also register the sk_lock with the lock validator.)
1287 */
1288 static inline void sock_lock_init(struct sock *sk)
1289 {
1290 sock_lock_init_class_and_name(sk,
1291 af_family_slock_key_strings[sk->sk_family],
1292 af_family_slock_keys + sk->sk_family,
1293 af_family_key_strings[sk->sk_family],
1294 af_family_keys + sk->sk_family);
1295 }
1296
1297 /*
1298 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1300 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1301 */
1302 static void sock_copy(struct sock *nsk, const struct sock *osk)
1303 {
1304 #ifdef CONFIG_SECURITY_NETWORK
1305 void *sptr = nsk->sk_security;
1306 #endif
1307 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308
1309 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1310 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311
1312 #ifdef CONFIG_SECURITY_NETWORK
1313 nsk->sk_security = sptr;
1314 security_sk_clone(osk, nsk);
1315 #endif
1316 }
1317
1318 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1319 int family)
1320 {
1321 struct sock *sk;
1322 struct kmem_cache *slab;
1323
1324 slab = prot->slab;
1325 if (slab != NULL) {
1326 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1327 if (!sk)
1328 return sk;
1329 if (priority & __GFP_ZERO)
1330 sk_prot_clear_nulls(sk, prot->obj_size);
1331 } else
1332 sk = kmalloc(prot->obj_size, priority);
1333
1334 if (sk != NULL) {
1335 kmemcheck_annotate_bitfield(sk, flags);
1336
1337 if (security_sk_alloc(sk, family, priority))
1338 goto out_free;
1339
1340 if (!try_module_get(prot->owner))
1341 goto out_free_sec;
1342 sk_tx_queue_clear(sk);
1343 }
1344
1345 return sk;
1346
1347 out_free_sec:
1348 security_sk_free(sk);
1349 out_free:
1350 if (slab != NULL)
1351 kmem_cache_free(slab, sk);
1352 else
1353 kfree(sk);
1354 return NULL;
1355 }
1356
1357 static void sk_prot_free(struct proto *prot, struct sock *sk)
1358 {
1359 struct kmem_cache *slab;
1360 struct module *owner;
1361
1362 owner = prot->owner;
1363 slab = prot->slab;
1364
1365 cgroup_sk_free(&sk->sk_cgrp_data);
1366 mem_cgroup_sk_free(sk);
1367 security_sk_free(sk);
1368 if (slab != NULL)
1369 kmem_cache_free(slab, sk);
1370 else
1371 kfree(sk);
1372 module_put(owner);
1373 }
1374
1375 /**
1376 * sk_alloc - All socket objects are allocated here
1377 * @net: the applicable net namespace
1378 * @family: protocol family
1379 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1380 * @prot: struct proto associated with this new sock instance
1381 * @kern: is this to be a kernel socket?
1382 */
1383 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1384 struct proto *prot, int kern)
1385 {
1386 struct sock *sk;
1387
1388 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1389 if (sk) {
1390 sk->sk_family = family;
1391 /*
1392 * See comment in struct sock definition to understand
1393 * why we need sk_prot_creator -acme
1394 */
1395 sk->sk_prot = sk->sk_prot_creator = prot;
1396 sock_lock_init(sk);
1397 sk->sk_net_refcnt = kern ? 0 : 1;
1398 if (likely(sk->sk_net_refcnt))
1399 get_net(net);
1400 sock_net_set(sk, net);
1401 atomic_set(&sk->sk_wmem_alloc, 1);
1402
1403 mem_cgroup_sk_alloc(sk);
1404 cgroup_sk_alloc(&sk->sk_cgrp_data);
1405 sock_update_classid(&sk->sk_cgrp_data);
1406 sock_update_netprioidx(&sk->sk_cgrp_data);
1407 }
1408
1409 return sk;
1410 }
1411 EXPORT_SYMBOL(sk_alloc);
1412
1413 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1414 * grace period. This is the case for UDP sockets and TCP listeners.
1415 */
1416 static void __sk_destruct(struct rcu_head *head)
1417 {
1418 struct sock *sk = container_of(head, struct sock, sk_rcu);
1419 struct sk_filter *filter;
1420
1421 if (sk->sk_destruct)
1422 sk->sk_destruct(sk);
1423
1424 filter = rcu_dereference_check(sk->sk_filter,
1425 atomic_read(&sk->sk_wmem_alloc) == 0);
1426 if (filter) {
1427 sk_filter_uncharge(sk, filter);
1428 RCU_INIT_POINTER(sk->sk_filter, NULL);
1429 }
1430 if (rcu_access_pointer(sk->sk_reuseport_cb))
1431 reuseport_detach_sock(sk);
1432
1433 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1434
1435 if (atomic_read(&sk->sk_omem_alloc))
1436 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1437 __func__, atomic_read(&sk->sk_omem_alloc));
1438
1439 if (sk->sk_peer_cred)
1440 put_cred(sk->sk_peer_cred);
1441 put_pid(sk->sk_peer_pid);
1442 if (likely(sk->sk_net_refcnt))
1443 put_net(sock_net(sk));
1444 sk_prot_free(sk->sk_prot_creator, sk);
1445 }
1446
1447 void sk_destruct(struct sock *sk)
1448 {
1449 if (sock_flag(sk, SOCK_RCU_FREE))
1450 call_rcu(&sk->sk_rcu, __sk_destruct);
1451 else
1452 __sk_destruct(&sk->sk_rcu);
1453 }
1454
1455 static void __sk_free(struct sock *sk)
1456 {
1457 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1458 sock_diag_broadcast_destroy(sk);
1459 else
1460 sk_destruct(sk);
1461 }
1462
1463 void sk_free(struct sock *sk)
1464 {
1465 /*
1466 * We subtract one from sk_wmem_alloc and can know if
1467 * some packets are still in some tx queue.
1468 * If not null, sock_wfree() will call __sk_free(sk) later
1469 */
1470 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1471 __sk_free(sk);
1472 }
1473 EXPORT_SYMBOL(sk_free);
1474
1475 /**
1476 * sk_clone_lock - clone a socket, and lock its clone
1477 * @sk: the socket to clone
1478 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1479 *
1480 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1481 */
1482 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1483 {
1484 struct sock *newsk;
1485 bool is_charged = true;
1486
1487 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1488 if (newsk != NULL) {
1489 struct sk_filter *filter;
1490
1491 sock_copy(newsk, sk);
1492
1493 /* SANITY */
1494 if (likely(newsk->sk_net_refcnt))
1495 get_net(sock_net(newsk));
1496 sk_node_init(&newsk->sk_node);
1497 sock_lock_init(newsk);
1498 bh_lock_sock(newsk);
1499 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1500 newsk->sk_backlog.len = 0;
1501
1502 atomic_set(&newsk->sk_rmem_alloc, 0);
1503 /*
1504 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1505 */
1506 atomic_set(&newsk->sk_wmem_alloc, 1);
1507 atomic_set(&newsk->sk_omem_alloc, 0);
1508 skb_queue_head_init(&newsk->sk_receive_queue);
1509 skb_queue_head_init(&newsk->sk_write_queue);
1510
1511 rwlock_init(&newsk->sk_callback_lock);
1512 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1513 af_callback_keys + newsk->sk_family,
1514 af_family_clock_key_strings[newsk->sk_family]);
1515
1516 newsk->sk_dst_cache = NULL;
1517 newsk->sk_wmem_queued = 0;
1518 newsk->sk_forward_alloc = 0;
1519 atomic_set(&newsk->sk_drops, 0);
1520 newsk->sk_send_head = NULL;
1521 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1522
1523 sock_reset_flag(newsk, SOCK_DONE);
1524 skb_queue_head_init(&newsk->sk_error_queue);
1525
1526 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1527 if (filter != NULL)
1528 /* though it's an empty new sock, the charging may fail
1529 * if sysctl_optmem_max was changed between creation of
1530 * original socket and cloning
1531 */
1532 is_charged = sk_filter_charge(newsk, filter);
1533
1534 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1535 /* It is still raw copy of parent, so invalidate
1536 * destructor and make plain sk_free() */
1537 newsk->sk_destruct = NULL;
1538 bh_unlock_sock(newsk);
1539 sk_free(newsk);
1540 newsk = NULL;
1541 goto out;
1542 }
1543 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1544
1545 newsk->sk_err = 0;
1546 newsk->sk_priority = 0;
1547 newsk->sk_incoming_cpu = raw_smp_processor_id();
1548 atomic64_set(&newsk->sk_cookie, 0);
1549
1550 mem_cgroup_sk_alloc(newsk);
1551 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1552
1553 /*
1554 * Before updating sk_refcnt, we must commit prior changes to memory
1555 * (Documentation/RCU/rculist_nulls.txt for details)
1556 */
1557 smp_wmb();
1558 atomic_set(&newsk->sk_refcnt, 2);
1559
1560 /*
1561 * Increment the counter in the same struct proto as the master
1562 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1563 * is the same as sk->sk_prot->socks, as this field was copied
1564 * with memcpy).
1565 *
1566 * This _changes_ the previous behaviour, where
1567 * tcp_create_openreq_child always was incrementing the
1568 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1569 * to be taken into account in all callers. -acme
1570 */
1571 sk_refcnt_debug_inc(newsk);
1572 sk_set_socket(newsk, NULL);
1573 newsk->sk_wq = NULL;
1574
1575 if (newsk->sk_prot->sockets_allocated)
1576 sk_sockets_allocated_inc(newsk);
1577
1578 if (sock_needs_netstamp(sk) &&
1579 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1580 net_enable_timestamp();
1581 }
1582 out:
1583 return newsk;
1584 }
1585 EXPORT_SYMBOL_GPL(sk_clone_lock);
1586
1587 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1588 {
1589 u32 max_segs = 1;
1590
1591 sk_dst_set(sk, dst);
1592 sk->sk_route_caps = dst->dev->features;
1593 if (sk->sk_route_caps & NETIF_F_GSO)
1594 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1595 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1596 if (sk_can_gso(sk)) {
1597 if (dst->header_len) {
1598 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1599 } else {
1600 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1601 sk->sk_gso_max_size = dst->dev->gso_max_size;
1602 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1603 }
1604 }
1605 sk->sk_gso_max_segs = max_segs;
1606 }
1607 EXPORT_SYMBOL_GPL(sk_setup_caps);
1608
1609 /*
1610 * Simple resource managers for sockets.
1611 */
1612
1613
1614 /*
1615 * Write buffer destructor automatically called from kfree_skb.
1616 */
1617 void sock_wfree(struct sk_buff *skb)
1618 {
1619 struct sock *sk = skb->sk;
1620 unsigned int len = skb->truesize;
1621
1622 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1623 /*
1624 * Keep a reference on sk_wmem_alloc, this will be released
1625 * after sk_write_space() call
1626 */
1627 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1628 sk->sk_write_space(sk);
1629 len = 1;
1630 }
1631 /*
1632 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1633 * could not do because of in-flight packets
1634 */
1635 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1636 __sk_free(sk);
1637 }
1638 EXPORT_SYMBOL(sock_wfree);
1639
1640 /* This variant of sock_wfree() is used by TCP,
1641 * since it sets SOCK_USE_WRITE_QUEUE.
1642 */
1643 void __sock_wfree(struct sk_buff *skb)
1644 {
1645 struct sock *sk = skb->sk;
1646
1647 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1648 __sk_free(sk);
1649 }
1650
1651 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1652 {
1653 skb_orphan(skb);
1654 skb->sk = sk;
1655 #ifdef CONFIG_INET
1656 if (unlikely(!sk_fullsock(sk))) {
1657 skb->destructor = sock_edemux;
1658 sock_hold(sk);
1659 return;
1660 }
1661 #endif
1662 skb->destructor = sock_wfree;
1663 skb_set_hash_from_sk(skb, sk);
1664 /*
1665 * We used to take a refcount on sk, but following operation
1666 * is enough to guarantee sk_free() wont free this sock until
1667 * all in-flight packets are completed
1668 */
1669 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1670 }
1671 EXPORT_SYMBOL(skb_set_owner_w);
1672
1673 /* This helper is used by netem, as it can hold packets in its
1674 * delay queue. We want to allow the owner socket to send more
1675 * packets, as if they were already TX completed by a typical driver.
1676 * But we also want to keep skb->sk set because some packet schedulers
1677 * rely on it (sch_fq for example). So we set skb->truesize to a small
1678 * amount (1) and decrease sk_wmem_alloc accordingly.
1679 */
1680 void skb_orphan_partial(struct sk_buff *skb)
1681 {
1682 /* If this skb is a TCP pure ACK or already went here,
1683 * we have nothing to do. 2 is already a very small truesize.
1684 */
1685 if (skb->truesize <= 2)
1686 return;
1687
1688 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1689 * so we do not completely orphan skb, but transfert all
1690 * accounted bytes but one, to avoid unexpected reorders.
1691 */
1692 if (skb->destructor == sock_wfree
1693 #ifdef CONFIG_INET
1694 || skb->destructor == tcp_wfree
1695 #endif
1696 ) {
1697 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1698 skb->truesize = 1;
1699 } else {
1700 skb_orphan(skb);
1701 }
1702 }
1703 EXPORT_SYMBOL(skb_orphan_partial);
1704
1705 /*
1706 * Read buffer destructor automatically called from kfree_skb.
1707 */
1708 void sock_rfree(struct sk_buff *skb)
1709 {
1710 struct sock *sk = skb->sk;
1711 unsigned int len = skb->truesize;
1712
1713 atomic_sub(len, &sk->sk_rmem_alloc);
1714 sk_mem_uncharge(sk, len);
1715 }
1716 EXPORT_SYMBOL(sock_rfree);
1717
1718 /*
1719 * Buffer destructor for skbs that are not used directly in read or write
1720 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1721 */
1722 void sock_efree(struct sk_buff *skb)
1723 {
1724 sock_put(skb->sk);
1725 }
1726 EXPORT_SYMBOL(sock_efree);
1727
1728 kuid_t sock_i_uid(struct sock *sk)
1729 {
1730 kuid_t uid;
1731
1732 read_lock_bh(&sk->sk_callback_lock);
1733 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1734 read_unlock_bh(&sk->sk_callback_lock);
1735 return uid;
1736 }
1737 EXPORT_SYMBOL(sock_i_uid);
1738
1739 unsigned long sock_i_ino(struct sock *sk)
1740 {
1741 unsigned long ino;
1742
1743 read_lock_bh(&sk->sk_callback_lock);
1744 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1745 read_unlock_bh(&sk->sk_callback_lock);
1746 return ino;
1747 }
1748 EXPORT_SYMBOL(sock_i_ino);
1749
1750 /*
1751 * Allocate a skb from the socket's send buffer.
1752 */
1753 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1754 gfp_t priority)
1755 {
1756 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1757 struct sk_buff *skb = alloc_skb(size, priority);
1758 if (skb) {
1759 skb_set_owner_w(skb, sk);
1760 return skb;
1761 }
1762 }
1763 return NULL;
1764 }
1765 EXPORT_SYMBOL(sock_wmalloc);
1766
1767 /*
1768 * Allocate a memory block from the socket's option memory buffer.
1769 */
1770 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1771 {
1772 if ((unsigned int)size <= sysctl_optmem_max &&
1773 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1774 void *mem;
1775 /* First do the add, to avoid the race if kmalloc
1776 * might sleep.
1777 */
1778 atomic_add(size, &sk->sk_omem_alloc);
1779 mem = kmalloc(size, priority);
1780 if (mem)
1781 return mem;
1782 atomic_sub(size, &sk->sk_omem_alloc);
1783 }
1784 return NULL;
1785 }
1786 EXPORT_SYMBOL(sock_kmalloc);
1787
1788 /* Free an option memory block. Note, we actually want the inline
1789 * here as this allows gcc to detect the nullify and fold away the
1790 * condition entirely.
1791 */
1792 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1793 const bool nullify)
1794 {
1795 if (WARN_ON_ONCE(!mem))
1796 return;
1797 if (nullify)
1798 kzfree(mem);
1799 else
1800 kfree(mem);
1801 atomic_sub(size, &sk->sk_omem_alloc);
1802 }
1803
1804 void sock_kfree_s(struct sock *sk, void *mem, int size)
1805 {
1806 __sock_kfree_s(sk, mem, size, false);
1807 }
1808 EXPORT_SYMBOL(sock_kfree_s);
1809
1810 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1811 {
1812 __sock_kfree_s(sk, mem, size, true);
1813 }
1814 EXPORT_SYMBOL(sock_kzfree_s);
1815
1816 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1817 I think, these locks should be removed for datagram sockets.
1818 */
1819 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1820 {
1821 DEFINE_WAIT(wait);
1822
1823 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1824 for (;;) {
1825 if (!timeo)
1826 break;
1827 if (signal_pending(current))
1828 break;
1829 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1830 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1831 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1832 break;
1833 if (sk->sk_shutdown & SEND_SHUTDOWN)
1834 break;
1835 if (sk->sk_err)
1836 break;
1837 timeo = schedule_timeout(timeo);
1838 }
1839 finish_wait(sk_sleep(sk), &wait);
1840 return timeo;
1841 }
1842
1843
1844 /*
1845 * Generic send/receive buffer handlers
1846 */
1847
1848 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1849 unsigned long data_len, int noblock,
1850 int *errcode, int max_page_order)
1851 {
1852 struct sk_buff *skb;
1853 long timeo;
1854 int err;
1855
1856 timeo = sock_sndtimeo(sk, noblock);
1857 for (;;) {
1858 err = sock_error(sk);
1859 if (err != 0)
1860 goto failure;
1861
1862 err = -EPIPE;
1863 if (sk->sk_shutdown & SEND_SHUTDOWN)
1864 goto failure;
1865
1866 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1867 break;
1868
1869 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1870 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1871 err = -EAGAIN;
1872 if (!timeo)
1873 goto failure;
1874 if (signal_pending(current))
1875 goto interrupted;
1876 timeo = sock_wait_for_wmem(sk, timeo);
1877 }
1878 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1879 errcode, sk->sk_allocation);
1880 if (skb)
1881 skb_set_owner_w(skb, sk);
1882 return skb;
1883
1884 interrupted:
1885 err = sock_intr_errno(timeo);
1886 failure:
1887 *errcode = err;
1888 return NULL;
1889 }
1890 EXPORT_SYMBOL(sock_alloc_send_pskb);
1891
1892 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1893 int noblock, int *errcode)
1894 {
1895 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1896 }
1897 EXPORT_SYMBOL(sock_alloc_send_skb);
1898
1899 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1900 struct sockcm_cookie *sockc)
1901 {
1902 u32 tsflags;
1903
1904 switch (cmsg->cmsg_type) {
1905 case SO_MARK:
1906 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1907 return -EPERM;
1908 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1909 return -EINVAL;
1910 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1911 break;
1912 case SO_TIMESTAMPING:
1913 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1914 return -EINVAL;
1915
1916 tsflags = *(u32 *)CMSG_DATA(cmsg);
1917 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1918 return -EINVAL;
1919
1920 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1921 sockc->tsflags |= tsflags;
1922 break;
1923 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1924 case SCM_RIGHTS:
1925 case SCM_CREDENTIALS:
1926 break;
1927 default:
1928 return -EINVAL;
1929 }
1930 return 0;
1931 }
1932 EXPORT_SYMBOL(__sock_cmsg_send);
1933
1934 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1935 struct sockcm_cookie *sockc)
1936 {
1937 struct cmsghdr *cmsg;
1938 int ret;
1939
1940 for_each_cmsghdr(cmsg, msg) {
1941 if (!CMSG_OK(msg, cmsg))
1942 return -EINVAL;
1943 if (cmsg->cmsg_level != SOL_SOCKET)
1944 continue;
1945 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1946 if (ret)
1947 return ret;
1948 }
1949 return 0;
1950 }
1951 EXPORT_SYMBOL(sock_cmsg_send);
1952
1953 /* On 32bit arches, an skb frag is limited to 2^15 */
1954 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1955
1956 /**
1957 * skb_page_frag_refill - check that a page_frag contains enough room
1958 * @sz: minimum size of the fragment we want to get
1959 * @pfrag: pointer to page_frag
1960 * @gfp: priority for memory allocation
1961 *
1962 * Note: While this allocator tries to use high order pages, there is
1963 * no guarantee that allocations succeed. Therefore, @sz MUST be
1964 * less or equal than PAGE_SIZE.
1965 */
1966 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1967 {
1968 if (pfrag->page) {
1969 if (page_ref_count(pfrag->page) == 1) {
1970 pfrag->offset = 0;
1971 return true;
1972 }
1973 if (pfrag->offset + sz <= pfrag->size)
1974 return true;
1975 put_page(pfrag->page);
1976 }
1977
1978 pfrag->offset = 0;
1979 if (SKB_FRAG_PAGE_ORDER) {
1980 /* Avoid direct reclaim but allow kswapd to wake */
1981 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1982 __GFP_COMP | __GFP_NOWARN |
1983 __GFP_NORETRY,
1984 SKB_FRAG_PAGE_ORDER);
1985 if (likely(pfrag->page)) {
1986 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1987 return true;
1988 }
1989 }
1990 pfrag->page = alloc_page(gfp);
1991 if (likely(pfrag->page)) {
1992 pfrag->size = PAGE_SIZE;
1993 return true;
1994 }
1995 return false;
1996 }
1997 EXPORT_SYMBOL(skb_page_frag_refill);
1998
1999 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2000 {
2001 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2002 return true;
2003
2004 sk_enter_memory_pressure(sk);
2005 sk_stream_moderate_sndbuf(sk);
2006 return false;
2007 }
2008 EXPORT_SYMBOL(sk_page_frag_refill);
2009
2010 static void __lock_sock(struct sock *sk)
2011 __releases(&sk->sk_lock.slock)
2012 __acquires(&sk->sk_lock.slock)
2013 {
2014 DEFINE_WAIT(wait);
2015
2016 for (;;) {
2017 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2018 TASK_UNINTERRUPTIBLE);
2019 spin_unlock_bh(&sk->sk_lock.slock);
2020 schedule();
2021 spin_lock_bh(&sk->sk_lock.slock);
2022 if (!sock_owned_by_user(sk))
2023 break;
2024 }
2025 finish_wait(&sk->sk_lock.wq, &wait);
2026 }
2027
2028 static void __release_sock(struct sock *sk)
2029 __releases(&sk->sk_lock.slock)
2030 __acquires(&sk->sk_lock.slock)
2031 {
2032 struct sk_buff *skb, *next;
2033
2034 while ((skb = sk->sk_backlog.head) != NULL) {
2035 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2036
2037 spin_unlock_bh(&sk->sk_lock.slock);
2038
2039 do {
2040 next = skb->next;
2041 prefetch(next);
2042 WARN_ON_ONCE(skb_dst_is_noref(skb));
2043 skb->next = NULL;
2044 sk_backlog_rcv(sk, skb);
2045
2046 cond_resched();
2047
2048 skb = next;
2049 } while (skb != NULL);
2050
2051 spin_lock_bh(&sk->sk_lock.slock);
2052 }
2053
2054 /*
2055 * Doing the zeroing here guarantee we can not loop forever
2056 * while a wild producer attempts to flood us.
2057 */
2058 sk->sk_backlog.len = 0;
2059 }
2060
2061 void __sk_flush_backlog(struct sock *sk)
2062 {
2063 spin_lock_bh(&sk->sk_lock.slock);
2064 __release_sock(sk);
2065 spin_unlock_bh(&sk->sk_lock.slock);
2066 }
2067
2068 /**
2069 * sk_wait_data - wait for data to arrive at sk_receive_queue
2070 * @sk: sock to wait on
2071 * @timeo: for how long
2072 * @skb: last skb seen on sk_receive_queue
2073 *
2074 * Now socket state including sk->sk_err is changed only under lock,
2075 * hence we may omit checks after joining wait queue.
2076 * We check receive queue before schedule() only as optimization;
2077 * it is very likely that release_sock() added new data.
2078 */
2079 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2080 {
2081 int rc;
2082 DEFINE_WAIT(wait);
2083
2084 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2085 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2086 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2087 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2088 finish_wait(sk_sleep(sk), &wait);
2089 return rc;
2090 }
2091 EXPORT_SYMBOL(sk_wait_data);
2092
2093 /**
2094 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2095 * @sk: socket
2096 * @size: memory size to allocate
2097 * @kind: allocation type
2098 *
2099 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2100 * rmem allocation. This function assumes that protocols which have
2101 * memory_pressure use sk_wmem_queued as write buffer accounting.
2102 */
2103 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2104 {
2105 struct proto *prot = sk->sk_prot;
2106 int amt = sk_mem_pages(size);
2107 long allocated;
2108
2109 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2110
2111 allocated = sk_memory_allocated_add(sk, amt);
2112
2113 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2114 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2115 goto suppress_allocation;
2116
2117 /* Under limit. */
2118 if (allocated <= sk_prot_mem_limits(sk, 0)) {
2119 sk_leave_memory_pressure(sk);
2120 return 1;
2121 }
2122
2123 /* Under pressure. */
2124 if (allocated > sk_prot_mem_limits(sk, 1))
2125 sk_enter_memory_pressure(sk);
2126
2127 /* Over hard limit. */
2128 if (allocated > sk_prot_mem_limits(sk, 2))
2129 goto suppress_allocation;
2130
2131 /* guarantee minimum buffer size under pressure */
2132 if (kind == SK_MEM_RECV) {
2133 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2134 return 1;
2135
2136 } else { /* SK_MEM_SEND */
2137 if (sk->sk_type == SOCK_STREAM) {
2138 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2139 return 1;
2140 } else if (atomic_read(&sk->sk_wmem_alloc) <
2141 prot->sysctl_wmem[0])
2142 return 1;
2143 }
2144
2145 if (sk_has_memory_pressure(sk)) {
2146 int alloc;
2147
2148 if (!sk_under_memory_pressure(sk))
2149 return 1;
2150 alloc = sk_sockets_allocated_read_positive(sk);
2151 if (sk_prot_mem_limits(sk, 2) > alloc *
2152 sk_mem_pages(sk->sk_wmem_queued +
2153 atomic_read(&sk->sk_rmem_alloc) +
2154 sk->sk_forward_alloc))
2155 return 1;
2156 }
2157
2158 suppress_allocation:
2159
2160 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2161 sk_stream_moderate_sndbuf(sk);
2162
2163 /* Fail only if socket is _under_ its sndbuf.
2164 * In this case we cannot block, so that we have to fail.
2165 */
2166 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2167 return 1;
2168 }
2169
2170 trace_sock_exceed_buf_limit(sk, prot, allocated);
2171
2172 /* Alas. Undo changes. */
2173 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2174
2175 sk_memory_allocated_sub(sk, amt);
2176
2177 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2178 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2179
2180 return 0;
2181 }
2182 EXPORT_SYMBOL(__sk_mem_schedule);
2183
2184 /**
2185 * __sk_mem_reclaim - reclaim memory_allocated
2186 * @sk: socket
2187 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2188 */
2189 void __sk_mem_reclaim(struct sock *sk, int amount)
2190 {
2191 amount >>= SK_MEM_QUANTUM_SHIFT;
2192 sk_memory_allocated_sub(sk, amount);
2193 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2194
2195 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2196 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2197
2198 if (sk_under_memory_pressure(sk) &&
2199 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2200 sk_leave_memory_pressure(sk);
2201 }
2202 EXPORT_SYMBOL(__sk_mem_reclaim);
2203
2204 int sk_set_peek_off(struct sock *sk, int val)
2205 {
2206 if (val < 0)
2207 return -EINVAL;
2208
2209 sk->sk_peek_off = val;
2210 return 0;
2211 }
2212 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2213
2214 /*
2215 * Set of default routines for initialising struct proto_ops when
2216 * the protocol does not support a particular function. In certain
2217 * cases where it makes no sense for a protocol to have a "do nothing"
2218 * function, some default processing is provided.
2219 */
2220
2221 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2222 {
2223 return -EOPNOTSUPP;
2224 }
2225 EXPORT_SYMBOL(sock_no_bind);
2226
2227 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2228 int len, int flags)
2229 {
2230 return -EOPNOTSUPP;
2231 }
2232 EXPORT_SYMBOL(sock_no_connect);
2233
2234 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2235 {
2236 return -EOPNOTSUPP;
2237 }
2238 EXPORT_SYMBOL(sock_no_socketpair);
2239
2240 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2241 {
2242 return -EOPNOTSUPP;
2243 }
2244 EXPORT_SYMBOL(sock_no_accept);
2245
2246 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2247 int *len, int peer)
2248 {
2249 return -EOPNOTSUPP;
2250 }
2251 EXPORT_SYMBOL(sock_no_getname);
2252
2253 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2254 {
2255 return 0;
2256 }
2257 EXPORT_SYMBOL(sock_no_poll);
2258
2259 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2260 {
2261 return -EOPNOTSUPP;
2262 }
2263 EXPORT_SYMBOL(sock_no_ioctl);
2264
2265 int sock_no_listen(struct socket *sock, int backlog)
2266 {
2267 return -EOPNOTSUPP;
2268 }
2269 EXPORT_SYMBOL(sock_no_listen);
2270
2271 int sock_no_shutdown(struct socket *sock, int how)
2272 {
2273 return -EOPNOTSUPP;
2274 }
2275 EXPORT_SYMBOL(sock_no_shutdown);
2276
2277 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2278 char __user *optval, unsigned int optlen)
2279 {
2280 return -EOPNOTSUPP;
2281 }
2282 EXPORT_SYMBOL(sock_no_setsockopt);
2283
2284 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2285 char __user *optval, int __user *optlen)
2286 {
2287 return -EOPNOTSUPP;
2288 }
2289 EXPORT_SYMBOL(sock_no_getsockopt);
2290
2291 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2292 {
2293 return -EOPNOTSUPP;
2294 }
2295 EXPORT_SYMBOL(sock_no_sendmsg);
2296
2297 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2298 int flags)
2299 {
2300 return -EOPNOTSUPP;
2301 }
2302 EXPORT_SYMBOL(sock_no_recvmsg);
2303
2304 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2305 {
2306 /* Mirror missing mmap method error code */
2307 return -ENODEV;
2308 }
2309 EXPORT_SYMBOL(sock_no_mmap);
2310
2311 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2312 {
2313 ssize_t res;
2314 struct msghdr msg = {.msg_flags = flags};
2315 struct kvec iov;
2316 char *kaddr = kmap(page);
2317 iov.iov_base = kaddr + offset;
2318 iov.iov_len = size;
2319 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2320 kunmap(page);
2321 return res;
2322 }
2323 EXPORT_SYMBOL(sock_no_sendpage);
2324
2325 /*
2326 * Default Socket Callbacks
2327 */
2328
2329 static void sock_def_wakeup(struct sock *sk)
2330 {
2331 struct socket_wq *wq;
2332
2333 rcu_read_lock();
2334 wq = rcu_dereference(sk->sk_wq);
2335 if (skwq_has_sleeper(wq))
2336 wake_up_interruptible_all(&wq->wait);
2337 rcu_read_unlock();
2338 }
2339
2340 static void sock_def_error_report(struct sock *sk)
2341 {
2342 struct socket_wq *wq;
2343
2344 rcu_read_lock();
2345 wq = rcu_dereference(sk->sk_wq);
2346 if (skwq_has_sleeper(wq))
2347 wake_up_interruptible_poll(&wq->wait, POLLERR);
2348 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2349 rcu_read_unlock();
2350 }
2351
2352 static void sock_def_readable(struct sock *sk)
2353 {
2354 struct socket_wq *wq;
2355
2356 rcu_read_lock();
2357 wq = rcu_dereference(sk->sk_wq);
2358 if (skwq_has_sleeper(wq))
2359 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2360 POLLRDNORM | POLLRDBAND);
2361 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2362 rcu_read_unlock();
2363 }
2364
2365 static void sock_def_write_space(struct sock *sk)
2366 {
2367 struct socket_wq *wq;
2368
2369 rcu_read_lock();
2370
2371 /* Do not wake up a writer until he can make "significant"
2372 * progress. --DaveM
2373 */
2374 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2375 wq = rcu_dereference(sk->sk_wq);
2376 if (skwq_has_sleeper(wq))
2377 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2378 POLLWRNORM | POLLWRBAND);
2379
2380 /* Should agree with poll, otherwise some programs break */
2381 if (sock_writeable(sk))
2382 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2383 }
2384
2385 rcu_read_unlock();
2386 }
2387
2388 static void sock_def_destruct(struct sock *sk)
2389 {
2390 }
2391
2392 void sk_send_sigurg(struct sock *sk)
2393 {
2394 if (sk->sk_socket && sk->sk_socket->file)
2395 if (send_sigurg(&sk->sk_socket->file->f_owner))
2396 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2397 }
2398 EXPORT_SYMBOL(sk_send_sigurg);
2399
2400 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2401 unsigned long expires)
2402 {
2403 if (!mod_timer(timer, expires))
2404 sock_hold(sk);
2405 }
2406 EXPORT_SYMBOL(sk_reset_timer);
2407
2408 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2409 {
2410 if (del_timer(timer))
2411 __sock_put(sk);
2412 }
2413 EXPORT_SYMBOL(sk_stop_timer);
2414
2415 void sock_init_data(struct socket *sock, struct sock *sk)
2416 {
2417 skb_queue_head_init(&sk->sk_receive_queue);
2418 skb_queue_head_init(&sk->sk_write_queue);
2419 skb_queue_head_init(&sk->sk_error_queue);
2420
2421 sk->sk_send_head = NULL;
2422
2423 init_timer(&sk->sk_timer);
2424
2425 sk->sk_allocation = GFP_KERNEL;
2426 sk->sk_rcvbuf = sysctl_rmem_default;
2427 sk->sk_sndbuf = sysctl_wmem_default;
2428 sk->sk_state = TCP_CLOSE;
2429 sk_set_socket(sk, sock);
2430
2431 sock_set_flag(sk, SOCK_ZAPPED);
2432
2433 if (sock) {
2434 sk->sk_type = sock->type;
2435 sk->sk_wq = sock->wq;
2436 sock->sk = sk;
2437 } else
2438 sk->sk_wq = NULL;
2439
2440 rwlock_init(&sk->sk_callback_lock);
2441 lockdep_set_class_and_name(&sk->sk_callback_lock,
2442 af_callback_keys + sk->sk_family,
2443 af_family_clock_key_strings[sk->sk_family]);
2444
2445 sk->sk_state_change = sock_def_wakeup;
2446 sk->sk_data_ready = sock_def_readable;
2447 sk->sk_write_space = sock_def_write_space;
2448 sk->sk_error_report = sock_def_error_report;
2449 sk->sk_destruct = sock_def_destruct;
2450
2451 sk->sk_frag.page = NULL;
2452 sk->sk_frag.offset = 0;
2453 sk->sk_peek_off = -1;
2454
2455 sk->sk_peer_pid = NULL;
2456 sk->sk_peer_cred = NULL;
2457 sk->sk_write_pending = 0;
2458 sk->sk_rcvlowat = 1;
2459 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2460 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2461
2462 sk->sk_stamp = ktime_set(-1L, 0);
2463
2464 #ifdef CONFIG_NET_RX_BUSY_POLL
2465 sk->sk_napi_id = 0;
2466 sk->sk_ll_usec = sysctl_net_busy_read;
2467 #endif
2468
2469 sk->sk_max_pacing_rate = ~0U;
2470 sk->sk_pacing_rate = ~0U;
2471 sk->sk_incoming_cpu = -1;
2472 /*
2473 * Before updating sk_refcnt, we must commit prior changes to memory
2474 * (Documentation/RCU/rculist_nulls.txt for details)
2475 */
2476 smp_wmb();
2477 atomic_set(&sk->sk_refcnt, 1);
2478 atomic_set(&sk->sk_drops, 0);
2479 }
2480 EXPORT_SYMBOL(sock_init_data);
2481
2482 void lock_sock_nested(struct sock *sk, int subclass)
2483 {
2484 might_sleep();
2485 spin_lock_bh(&sk->sk_lock.slock);
2486 if (sk->sk_lock.owned)
2487 __lock_sock(sk);
2488 sk->sk_lock.owned = 1;
2489 spin_unlock(&sk->sk_lock.slock);
2490 /*
2491 * The sk_lock has mutex_lock() semantics here:
2492 */
2493 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2494 local_bh_enable();
2495 }
2496 EXPORT_SYMBOL(lock_sock_nested);
2497
2498 void release_sock(struct sock *sk)
2499 {
2500 spin_lock_bh(&sk->sk_lock.slock);
2501 if (sk->sk_backlog.tail)
2502 __release_sock(sk);
2503
2504 /* Warning : release_cb() might need to release sk ownership,
2505 * ie call sock_release_ownership(sk) before us.
2506 */
2507 if (sk->sk_prot->release_cb)
2508 sk->sk_prot->release_cb(sk);
2509
2510 sock_release_ownership(sk);
2511 if (waitqueue_active(&sk->sk_lock.wq))
2512 wake_up(&sk->sk_lock.wq);
2513 spin_unlock_bh(&sk->sk_lock.slock);
2514 }
2515 EXPORT_SYMBOL(release_sock);
2516
2517 /**
2518 * lock_sock_fast - fast version of lock_sock
2519 * @sk: socket
2520 *
2521 * This version should be used for very small section, where process wont block
2522 * return false if fast path is taken
2523 * sk_lock.slock locked, owned = 0, BH disabled
2524 * return true if slow path is taken
2525 * sk_lock.slock unlocked, owned = 1, BH enabled
2526 */
2527 bool lock_sock_fast(struct sock *sk)
2528 {
2529 might_sleep();
2530 spin_lock_bh(&sk->sk_lock.slock);
2531
2532 if (!sk->sk_lock.owned)
2533 /*
2534 * Note : We must disable BH
2535 */
2536 return false;
2537
2538 __lock_sock(sk);
2539 sk->sk_lock.owned = 1;
2540 spin_unlock(&sk->sk_lock.slock);
2541 /*
2542 * The sk_lock has mutex_lock() semantics here:
2543 */
2544 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2545 local_bh_enable();
2546 return true;
2547 }
2548 EXPORT_SYMBOL(lock_sock_fast);
2549
2550 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2551 {
2552 struct timeval tv;
2553 if (!sock_flag(sk, SOCK_TIMESTAMP))
2554 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2555 tv = ktime_to_timeval(sk->sk_stamp);
2556 if (tv.tv_sec == -1)
2557 return -ENOENT;
2558 if (tv.tv_sec == 0) {
2559 sk->sk_stamp = ktime_get_real();
2560 tv = ktime_to_timeval(sk->sk_stamp);
2561 }
2562 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2563 }
2564 EXPORT_SYMBOL(sock_get_timestamp);
2565
2566 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2567 {
2568 struct timespec ts;
2569 if (!sock_flag(sk, SOCK_TIMESTAMP))
2570 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2571 ts = ktime_to_timespec(sk->sk_stamp);
2572 if (ts.tv_sec == -1)
2573 return -ENOENT;
2574 if (ts.tv_sec == 0) {
2575 sk->sk_stamp = ktime_get_real();
2576 ts = ktime_to_timespec(sk->sk_stamp);
2577 }
2578 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2579 }
2580 EXPORT_SYMBOL(sock_get_timestampns);
2581
2582 void sock_enable_timestamp(struct sock *sk, int flag)
2583 {
2584 if (!sock_flag(sk, flag)) {
2585 unsigned long previous_flags = sk->sk_flags;
2586
2587 sock_set_flag(sk, flag);
2588 /*
2589 * we just set one of the two flags which require net
2590 * time stamping, but time stamping might have been on
2591 * already because of the other one
2592 */
2593 if (sock_needs_netstamp(sk) &&
2594 !(previous_flags & SK_FLAGS_TIMESTAMP))
2595 net_enable_timestamp();
2596 }
2597 }
2598
2599 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2600 int level, int type)
2601 {
2602 struct sock_exterr_skb *serr;
2603 struct sk_buff *skb;
2604 int copied, err;
2605
2606 err = -EAGAIN;
2607 skb = sock_dequeue_err_skb(sk);
2608 if (skb == NULL)
2609 goto out;
2610
2611 copied = skb->len;
2612 if (copied > len) {
2613 msg->msg_flags |= MSG_TRUNC;
2614 copied = len;
2615 }
2616 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2617 if (err)
2618 goto out_free_skb;
2619
2620 sock_recv_timestamp(msg, sk, skb);
2621
2622 serr = SKB_EXT_ERR(skb);
2623 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2624
2625 msg->msg_flags |= MSG_ERRQUEUE;
2626 err = copied;
2627
2628 out_free_skb:
2629 kfree_skb(skb);
2630 out:
2631 return err;
2632 }
2633 EXPORT_SYMBOL(sock_recv_errqueue);
2634
2635 /*
2636 * Get a socket option on an socket.
2637 *
2638 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2639 * asynchronous errors should be reported by getsockopt. We assume
2640 * this means if you specify SO_ERROR (otherwise whats the point of it).
2641 */
2642 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2643 char __user *optval, int __user *optlen)
2644 {
2645 struct sock *sk = sock->sk;
2646
2647 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2648 }
2649 EXPORT_SYMBOL(sock_common_getsockopt);
2650
2651 #ifdef CONFIG_COMPAT
2652 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2653 char __user *optval, int __user *optlen)
2654 {
2655 struct sock *sk = sock->sk;
2656
2657 if (sk->sk_prot->compat_getsockopt != NULL)
2658 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2659 optval, optlen);
2660 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2661 }
2662 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2663 #endif
2664
2665 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2666 int flags)
2667 {
2668 struct sock *sk = sock->sk;
2669 int addr_len = 0;
2670 int err;
2671
2672 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2673 flags & ~MSG_DONTWAIT, &addr_len);
2674 if (err >= 0)
2675 msg->msg_namelen = addr_len;
2676 return err;
2677 }
2678 EXPORT_SYMBOL(sock_common_recvmsg);
2679
2680 /*
2681 * Set socket options on an inet socket.
2682 */
2683 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2684 char __user *optval, unsigned int optlen)
2685 {
2686 struct sock *sk = sock->sk;
2687
2688 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2689 }
2690 EXPORT_SYMBOL(sock_common_setsockopt);
2691
2692 #ifdef CONFIG_COMPAT
2693 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2694 char __user *optval, unsigned int optlen)
2695 {
2696 struct sock *sk = sock->sk;
2697
2698 if (sk->sk_prot->compat_setsockopt != NULL)
2699 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2700 optval, optlen);
2701 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2702 }
2703 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2704 #endif
2705
2706 void sk_common_release(struct sock *sk)
2707 {
2708 if (sk->sk_prot->destroy)
2709 sk->sk_prot->destroy(sk);
2710
2711 /*
2712 * Observation: when sock_common_release is called, processes have
2713 * no access to socket. But net still has.
2714 * Step one, detach it from networking:
2715 *
2716 * A. Remove from hash tables.
2717 */
2718
2719 sk->sk_prot->unhash(sk);
2720
2721 /*
2722 * In this point socket cannot receive new packets, but it is possible
2723 * that some packets are in flight because some CPU runs receiver and
2724 * did hash table lookup before we unhashed socket. They will achieve
2725 * receive queue and will be purged by socket destructor.
2726 *
2727 * Also we still have packets pending on receive queue and probably,
2728 * our own packets waiting in device queues. sock_destroy will drain
2729 * receive queue, but transmitted packets will delay socket destruction
2730 * until the last reference will be released.
2731 */
2732
2733 sock_orphan(sk);
2734
2735 xfrm_sk_free_policy(sk);
2736
2737 sk_refcnt_debug_release(sk);
2738
2739 if (sk->sk_frag.page) {
2740 put_page(sk->sk_frag.page);
2741 sk->sk_frag.page = NULL;
2742 }
2743
2744 sock_put(sk);
2745 }
2746 EXPORT_SYMBOL(sk_common_release);
2747
2748 #ifdef CONFIG_PROC_FS
2749 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2750 struct prot_inuse {
2751 int val[PROTO_INUSE_NR];
2752 };
2753
2754 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2755
2756 #ifdef CONFIG_NET_NS
2757 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2758 {
2759 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2760 }
2761 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2762
2763 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2764 {
2765 int cpu, idx = prot->inuse_idx;
2766 int res = 0;
2767
2768 for_each_possible_cpu(cpu)
2769 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2770
2771 return res >= 0 ? res : 0;
2772 }
2773 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2774
2775 static int __net_init sock_inuse_init_net(struct net *net)
2776 {
2777 net->core.inuse = alloc_percpu(struct prot_inuse);
2778 return net->core.inuse ? 0 : -ENOMEM;
2779 }
2780
2781 static void __net_exit sock_inuse_exit_net(struct net *net)
2782 {
2783 free_percpu(net->core.inuse);
2784 }
2785
2786 static struct pernet_operations net_inuse_ops = {
2787 .init = sock_inuse_init_net,
2788 .exit = sock_inuse_exit_net,
2789 };
2790
2791 static __init int net_inuse_init(void)
2792 {
2793 if (register_pernet_subsys(&net_inuse_ops))
2794 panic("Cannot initialize net inuse counters");
2795
2796 return 0;
2797 }
2798
2799 core_initcall(net_inuse_init);
2800 #else
2801 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2802
2803 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2804 {
2805 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2806 }
2807 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2808
2809 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2810 {
2811 int cpu, idx = prot->inuse_idx;
2812 int res = 0;
2813
2814 for_each_possible_cpu(cpu)
2815 res += per_cpu(prot_inuse, cpu).val[idx];
2816
2817 return res >= 0 ? res : 0;
2818 }
2819 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2820 #endif
2821
2822 static void assign_proto_idx(struct proto *prot)
2823 {
2824 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2825
2826 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2827 pr_err("PROTO_INUSE_NR exhausted\n");
2828 return;
2829 }
2830
2831 set_bit(prot->inuse_idx, proto_inuse_idx);
2832 }
2833
2834 static void release_proto_idx(struct proto *prot)
2835 {
2836 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2837 clear_bit(prot->inuse_idx, proto_inuse_idx);
2838 }
2839 #else
2840 static inline void assign_proto_idx(struct proto *prot)
2841 {
2842 }
2843
2844 static inline void release_proto_idx(struct proto *prot)
2845 {
2846 }
2847 #endif
2848
2849 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2850 {
2851 if (!rsk_prot)
2852 return;
2853 kfree(rsk_prot->slab_name);
2854 rsk_prot->slab_name = NULL;
2855 kmem_cache_destroy(rsk_prot->slab);
2856 rsk_prot->slab = NULL;
2857 }
2858
2859 static int req_prot_init(const struct proto *prot)
2860 {
2861 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2862
2863 if (!rsk_prot)
2864 return 0;
2865
2866 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2867 prot->name);
2868 if (!rsk_prot->slab_name)
2869 return -ENOMEM;
2870
2871 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2872 rsk_prot->obj_size, 0,
2873 prot->slab_flags, NULL);
2874
2875 if (!rsk_prot->slab) {
2876 pr_crit("%s: Can't create request sock SLAB cache!\n",
2877 prot->name);
2878 return -ENOMEM;
2879 }
2880 return 0;
2881 }
2882
2883 int proto_register(struct proto *prot, int alloc_slab)
2884 {
2885 if (alloc_slab) {
2886 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2887 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2888 NULL);
2889
2890 if (prot->slab == NULL) {
2891 pr_crit("%s: Can't create sock SLAB cache!\n",
2892 prot->name);
2893 goto out;
2894 }
2895
2896 if (req_prot_init(prot))
2897 goto out_free_request_sock_slab;
2898
2899 if (prot->twsk_prot != NULL) {
2900 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2901
2902 if (prot->twsk_prot->twsk_slab_name == NULL)
2903 goto out_free_request_sock_slab;
2904
2905 prot->twsk_prot->twsk_slab =
2906 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2907 prot->twsk_prot->twsk_obj_size,
2908 0,
2909 prot->slab_flags,
2910 NULL);
2911 if (prot->twsk_prot->twsk_slab == NULL)
2912 goto out_free_timewait_sock_slab_name;
2913 }
2914 }
2915
2916 mutex_lock(&proto_list_mutex);
2917 list_add(&prot->node, &proto_list);
2918 assign_proto_idx(prot);
2919 mutex_unlock(&proto_list_mutex);
2920 return 0;
2921
2922 out_free_timewait_sock_slab_name:
2923 kfree(prot->twsk_prot->twsk_slab_name);
2924 out_free_request_sock_slab:
2925 req_prot_cleanup(prot->rsk_prot);
2926
2927 kmem_cache_destroy(prot->slab);
2928 prot->slab = NULL;
2929 out:
2930 return -ENOBUFS;
2931 }
2932 EXPORT_SYMBOL(proto_register);
2933
2934 void proto_unregister(struct proto *prot)
2935 {
2936 mutex_lock(&proto_list_mutex);
2937 release_proto_idx(prot);
2938 list_del(&prot->node);
2939 mutex_unlock(&proto_list_mutex);
2940
2941 kmem_cache_destroy(prot->slab);
2942 prot->slab = NULL;
2943
2944 req_prot_cleanup(prot->rsk_prot);
2945
2946 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2947 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2948 kfree(prot->twsk_prot->twsk_slab_name);
2949 prot->twsk_prot->twsk_slab = NULL;
2950 }
2951 }
2952 EXPORT_SYMBOL(proto_unregister);
2953
2954 #ifdef CONFIG_PROC_FS
2955 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2956 __acquires(proto_list_mutex)
2957 {
2958 mutex_lock(&proto_list_mutex);
2959 return seq_list_start_head(&proto_list, *pos);
2960 }
2961
2962 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2963 {
2964 return seq_list_next(v, &proto_list, pos);
2965 }
2966
2967 static void proto_seq_stop(struct seq_file *seq, void *v)
2968 __releases(proto_list_mutex)
2969 {
2970 mutex_unlock(&proto_list_mutex);
2971 }
2972
2973 static char proto_method_implemented(const void *method)
2974 {
2975 return method == NULL ? 'n' : 'y';
2976 }
2977 static long sock_prot_memory_allocated(struct proto *proto)
2978 {
2979 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2980 }
2981
2982 static char *sock_prot_memory_pressure(struct proto *proto)
2983 {
2984 return proto->memory_pressure != NULL ?
2985 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2986 }
2987
2988 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2989 {
2990
2991 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2992 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2993 proto->name,
2994 proto->obj_size,
2995 sock_prot_inuse_get(seq_file_net(seq), proto),
2996 sock_prot_memory_allocated(proto),
2997 sock_prot_memory_pressure(proto),
2998 proto->max_header,
2999 proto->slab == NULL ? "no" : "yes",
3000 module_name(proto->owner),
3001 proto_method_implemented(proto->close),
3002 proto_method_implemented(proto->connect),
3003 proto_method_implemented(proto->disconnect),
3004 proto_method_implemented(proto->accept),
3005 proto_method_implemented(proto->ioctl),
3006 proto_method_implemented(proto->init),
3007 proto_method_implemented(proto->destroy),
3008 proto_method_implemented(proto->shutdown),
3009 proto_method_implemented(proto->setsockopt),
3010 proto_method_implemented(proto->getsockopt),
3011 proto_method_implemented(proto->sendmsg),
3012 proto_method_implemented(proto->recvmsg),
3013 proto_method_implemented(proto->sendpage),
3014 proto_method_implemented(proto->bind),
3015 proto_method_implemented(proto->backlog_rcv),
3016 proto_method_implemented(proto->hash),
3017 proto_method_implemented(proto->unhash),
3018 proto_method_implemented(proto->get_port),
3019 proto_method_implemented(proto->enter_memory_pressure));
3020 }
3021
3022 static int proto_seq_show(struct seq_file *seq, void *v)
3023 {
3024 if (v == &proto_list)
3025 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3026 "protocol",
3027 "size",
3028 "sockets",
3029 "memory",
3030 "press",
3031 "maxhdr",
3032 "slab",
3033 "module",
3034 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3035 else
3036 proto_seq_printf(seq, list_entry(v, struct proto, node));
3037 return 0;
3038 }
3039
3040 static const struct seq_operations proto_seq_ops = {
3041 .start = proto_seq_start,
3042 .next = proto_seq_next,
3043 .stop = proto_seq_stop,
3044 .show = proto_seq_show,
3045 };
3046
3047 static int proto_seq_open(struct inode *inode, struct file *file)
3048 {
3049 return seq_open_net(inode, file, &proto_seq_ops,
3050 sizeof(struct seq_net_private));
3051 }
3052
3053 static const struct file_operations proto_seq_fops = {
3054 .owner = THIS_MODULE,
3055 .open = proto_seq_open,
3056 .read = seq_read,
3057 .llseek = seq_lseek,
3058 .release = seq_release_net,
3059 };
3060
3061 static __net_init int proto_init_net(struct net *net)
3062 {
3063 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3064 return -ENOMEM;
3065
3066 return 0;
3067 }
3068
3069 static __net_exit void proto_exit_net(struct net *net)
3070 {
3071 remove_proc_entry("protocols", net->proc_net);
3072 }
3073
3074
3075 static __net_initdata struct pernet_operations proto_net_ops = {
3076 .init = proto_init_net,
3077 .exit = proto_exit_net,
3078 };
3079
3080 static int __init proto_init(void)
3081 {
3082 return register_pernet_subsys(&proto_net_ops);
3083 }
3084
3085 subsys_initcall(proto_init);
3086
3087 #endif /* PROC_FS */
(deprecated) Btrfs devel tree