]> git.ipfire.org Git - people/ms/linux.git/blob - net/core/sock.c
projects / people / ms / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Don't reset ->total_link_count on nested calls of vfs_path_lookup()
[people/ms/linux.git] / net / core / sock.c
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
138 #include <trace/events/sock.h>
139
140 #ifdef CONFIG_INET
141 #include <net/tcp.h>
142 #endif
143
144 #include <net/busy_poll.h>
145
146 static DEFINE_MUTEX(proto_list_mutex);
147 static LIST_HEAD(proto_list);
148
149 /**
150 * sk_ns_capable - General socket capability test
151 * @sk: Socket to use a capability on or through
152 * @user_ns: The user namespace of the capability to use
153 * @cap: The capability to use
154 *
155 * Test to see if the opener of the socket had when the socket was
156 * created and the current process has the capability @cap in the user
157 * namespace @user_ns.
158 */
159 bool sk_ns_capable(const struct sock *sk,
160 struct user_namespace *user_ns, int cap)
161 {
162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
163 ns_capable(user_ns, cap);
164 }
165 EXPORT_SYMBOL(sk_ns_capable);
166
167 /**
168 * sk_capable - Socket global capability test
169 * @sk: Socket to use a capability on or through
170 * @cap: The global capability to use
171 *
172 * Test to see if the opener of the socket had when the socket was
173 * created and the current process has the capability @cap in all user
174 * namespaces.
175 */
176 bool sk_capable(const struct sock *sk, int cap)
177 {
178 return sk_ns_capable(sk, &init_user_ns, cap);
179 }
180 EXPORT_SYMBOL(sk_capable);
181
182 /**
183 * sk_net_capable - Network namespace socket capability test
184 * @sk: Socket to use a capability on or through
185 * @cap: The capability to use
186 *
187 * Test to see if the opener of the socket had when the socket was created
188 * and the current process has the capability @cap over the network namespace
189 * the socket is a member of.
190 */
191 bool sk_net_capable(const struct sock *sk, int cap)
192 {
193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
194 }
195 EXPORT_SYMBOL(sk_net_capable);
196
197
198 #ifdef CONFIG_MEMCG_KMEM
199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
200 {
201 struct proto *proto;
202 int ret = 0;
203
204 mutex_lock(&proto_list_mutex);
205 list_for_each_entry(proto, &proto_list, node) {
206 if (proto->init_cgroup) {
207 ret = proto->init_cgroup(memcg, ss);
208 if (ret)
209 goto out;
210 }
211 }
212
213 mutex_unlock(&proto_list_mutex);
214 return ret;
215 out:
216 list_for_each_entry_continue_reverse(proto, &proto_list, node)
217 if (proto->destroy_cgroup)
218 proto->destroy_cgroup(memcg);
219 mutex_unlock(&proto_list_mutex);
220 return ret;
221 }
222
223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
224 {
225 struct proto *proto;
226
227 mutex_lock(&proto_list_mutex);
228 list_for_each_entry_reverse(proto, &proto_list, node)
229 if (proto->destroy_cgroup)
230 proto->destroy_cgroup(memcg);
231 mutex_unlock(&proto_list_mutex);
232 }
233 #endif
234
235 /*
236 * Each address family might have different locking rules, so we have
237 * one slock key per address family:
238 */
239 static struct lock_class_key af_family_keys[AF_MAX];
240 static struct lock_class_key af_family_slock_keys[AF_MAX];
241
242 #if defined(CONFIG_MEMCG_KMEM)
243 struct static_key memcg_socket_limit_enabled;
244 EXPORT_SYMBOL(memcg_socket_limit_enabled);
245 #endif
246
247 /*
248 * Make lock validator output more readable. (we pre-construct these
249 * strings build-time, so that runtime initialization of socket
250 * locks is fast):
251 */
252 static const char *const af_family_key_strings[AF_MAX+1] = {
253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
267 };
268 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
278 "slock-27" , "slock-28" , "slock-AF_CAN" ,
279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
283 };
284 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
294 "clock-27" , "clock-28" , "clock-AF_CAN" ,
295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
299 };
300
301 /*
302 * sk_callback_lock locking rules are per-address-family,
303 * so split the lock classes by using a per-AF key:
304 */
305 static struct lock_class_key af_callback_keys[AF_MAX];
306
307 /* Take into consideration the size of the struct sk_buff overhead in the
308 * determination of these values, since that is non-constant across
309 * platforms. This makes socket queueing behavior and performance
310 * not depend upon such differences.
311 */
312 #define _SK_MEM_PACKETS 256
313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
316
317 /* Run time adjustable parameters. */
318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
319 EXPORT_SYMBOL(sysctl_wmem_max);
320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
321 EXPORT_SYMBOL(sysctl_rmem_max);
322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
324
325 /* Maximal space eaten by iovec or ancillary data plus some space */
326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
327 EXPORT_SYMBOL(sysctl_optmem_max);
328
329 int sysctl_tstamp_allow_data __read_mostly = 1;
330
331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
332 EXPORT_SYMBOL_GPL(memalloc_socks);
333
334 /**
335 * sk_set_memalloc - sets %SOCK_MEMALLOC
336 * @sk: socket to set it on
337 *
338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
339 * It's the responsibility of the admin to adjust min_free_kbytes
340 * to meet the requirements
341 */
342 void sk_set_memalloc(struct sock *sk)
343 {
344 sock_set_flag(sk, SOCK_MEMALLOC);
345 sk->sk_allocation |= __GFP_MEMALLOC;
346 static_key_slow_inc(&memalloc_socks);
347 }
348 EXPORT_SYMBOL_GPL(sk_set_memalloc);
349
350 void sk_clear_memalloc(struct sock *sk)
351 {
352 sock_reset_flag(sk, SOCK_MEMALLOC);
353 sk->sk_allocation &= ~__GFP_MEMALLOC;
354 static_key_slow_dec(&memalloc_socks);
355
356 /*
357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
358 * progress of swapping. SOCK_MEMALLOC may be cleared while
359 * it has rmem allocations due to the last swapfile being deactivated
360 * but there is a risk that the socket is unusable due to exceeding
361 * the rmem limits. Reclaim the reserves and obey rmem limits again.
362 */
363 sk_mem_reclaim(sk);
364 }
365 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
366
367 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
368 {
369 int ret;
370 unsigned long pflags = current->flags;
371
372 /* these should have been dropped before queueing */
373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
374
375 current->flags |= PF_MEMALLOC;
376 ret = sk->sk_backlog_rcv(sk, skb);
377 tsk_restore_flags(current, pflags, PF_MEMALLOC);
378
379 return ret;
380 }
381 EXPORT_SYMBOL(__sk_backlog_rcv);
382
383 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
384 {
385 struct timeval tv;
386
387 if (optlen < sizeof(tv))
388 return -EINVAL;
389 if (copy_from_user(&tv, optval, sizeof(tv)))
390 return -EFAULT;
391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
392 return -EDOM;
393
394 if (tv.tv_sec < 0) {
395 static int warned __read_mostly;
396
397 *timeo_p = 0;
398 if (warned < 10 && net_ratelimit()) {
399 warned++;
400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
401 __func__, current->comm, task_pid_nr(current));
402 }
403 return 0;
404 }
405 *timeo_p = MAX_SCHEDULE_TIMEOUT;
406 if (tv.tv_sec == 0 && tv.tv_usec == 0)
407 return 0;
408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
410 return 0;
411 }
412
413 static void sock_warn_obsolete_bsdism(const char *name)
414 {
415 static int warned;
416 static char warncomm[TASK_COMM_LEN];
417 if (strcmp(warncomm, current->comm) && warned < 5) {
418 strcpy(warncomm, current->comm);
419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
420 warncomm, name);
421 warned++;
422 }
423 }
424
425 static bool sock_needs_netstamp(const struct sock *sk)
426 {
427 switch (sk->sk_family) {
428 case AF_UNSPEC:
429 case AF_UNIX:
430 return false;
431 default:
432 return true;
433 }
434 }
435
436 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
437
438 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
439 {
440 if (sk->sk_flags & flags) {
441 sk->sk_flags &= ~flags;
442 if (sock_needs_netstamp(sk) &&
443 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
444 net_disable_timestamp();
445 }
446 }
447
448
449 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
450 {
451 int err;
452 unsigned long flags;
453 struct sk_buff_head *list = &sk->sk_receive_queue;
454
455 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
456 atomic_inc(&sk->sk_drops);
457 trace_sock_rcvqueue_full(sk, skb);
458 return -ENOMEM;
459 }
460
461 err = sk_filter(sk, skb);
462 if (err)
463 return err;
464
465 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
466 atomic_inc(&sk->sk_drops);
467 return -ENOBUFS;
468 }
469
470 skb->dev = NULL;
471 skb_set_owner_r(skb, sk);
472
473 /* we escape from rcu protected region, make sure we dont leak
474 * a norefcounted dst
475 */
476 skb_dst_force(skb);
477
478 spin_lock_irqsave(&list->lock, flags);
479 sock_skb_set_dropcount(sk, skb);
480 __skb_queue_tail(list, skb);
481 spin_unlock_irqrestore(&list->lock, flags);
482
483 if (!sock_flag(sk, SOCK_DEAD))
484 sk->sk_data_ready(sk);
485 return 0;
486 }
487 EXPORT_SYMBOL(sock_queue_rcv_skb);
488
489 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
490 {
491 int rc = NET_RX_SUCCESS;
492
493 if (sk_filter(sk, skb))
494 goto discard_and_relse;
495
496 skb->dev = NULL;
497
498 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
499 atomic_inc(&sk->sk_drops);
500 goto discard_and_relse;
501 }
502 if (nested)
503 bh_lock_sock_nested(sk);
504 else
505 bh_lock_sock(sk);
506 if (!sock_owned_by_user(sk)) {
507 /*
508 * trylock + unlock semantics:
509 */
510 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
511
512 rc = sk_backlog_rcv(sk, skb);
513
514 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
515 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
516 bh_unlock_sock(sk);
517 atomic_inc(&sk->sk_drops);
518 goto discard_and_relse;
519 }
520
521 bh_unlock_sock(sk);
522 out:
523 sock_put(sk);
524 return rc;
525 discard_and_relse:
526 kfree_skb(skb);
527 goto out;
528 }
529 EXPORT_SYMBOL(sk_receive_skb);
530
531 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
532 {
533 struct dst_entry *dst = __sk_dst_get(sk);
534
535 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
536 sk_tx_queue_clear(sk);
537 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
538 dst_release(dst);
539 return NULL;
540 }
541
542 return dst;
543 }
544 EXPORT_SYMBOL(__sk_dst_check);
545
546 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
547 {
548 struct dst_entry *dst = sk_dst_get(sk);
549
550 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
551 sk_dst_reset(sk);
552 dst_release(dst);
553 return NULL;
554 }
555
556 return dst;
557 }
558 EXPORT_SYMBOL(sk_dst_check);
559
560 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
561 int optlen)
562 {
563 int ret = -ENOPROTOOPT;
564 #ifdef CONFIG_NETDEVICES
565 struct net *net = sock_net(sk);
566 char devname[IFNAMSIZ];
567 int index;
568
569 /* Sorry... */
570 ret = -EPERM;
571 if (!ns_capable(net->user_ns, CAP_NET_RAW))
572 goto out;
573
574 ret = -EINVAL;
575 if (optlen < 0)
576 goto out;
577
578 /* Bind this socket to a particular device like "eth0",
579 * as specified in the passed interface name. If the
580 * name is "" or the option length is zero the socket
581 * is not bound.
582 */
583 if (optlen > IFNAMSIZ - 1)
584 optlen = IFNAMSIZ - 1;
585 memset(devname, 0, sizeof(devname));
586
587 ret = -EFAULT;
588 if (copy_from_user(devname, optval, optlen))
589 goto out;
590
591 index = 0;
592 if (devname[0] != '\0') {
593 struct net_device *dev;
594
595 rcu_read_lock();
596 dev = dev_get_by_name_rcu(net, devname);
597 if (dev)
598 index = dev->ifindex;
599 rcu_read_unlock();
600 ret = -ENODEV;
601 if (!dev)
602 goto out;
603 }
604
605 lock_sock(sk);
606 sk->sk_bound_dev_if = index;
607 sk_dst_reset(sk);
608 release_sock(sk);
609
610 ret = 0;
611
612 out:
613 #endif
614
615 return ret;
616 }
617
618 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
619 int __user *optlen, int len)
620 {
621 int ret = -ENOPROTOOPT;
622 #ifdef CONFIG_NETDEVICES
623 struct net *net = sock_net(sk);
624 char devname[IFNAMSIZ];
625
626 if (sk->sk_bound_dev_if == 0) {
627 len = 0;
628 goto zero;
629 }
630
631 ret = -EINVAL;
632 if (len < IFNAMSIZ)
633 goto out;
634
635 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
636 if (ret)
637 goto out;
638
639 len = strlen(devname) + 1;
640
641 ret = -EFAULT;
642 if (copy_to_user(optval, devname, len))
643 goto out;
644
645 zero:
646 ret = -EFAULT;
647 if (put_user(len, optlen))
648 goto out;
649
650 ret = 0;
651
652 out:
653 #endif
654
655 return ret;
656 }
657
658 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
659 {
660 if (valbool)
661 sock_set_flag(sk, bit);
662 else
663 sock_reset_flag(sk, bit);
664 }
665
666 bool sk_mc_loop(struct sock *sk)
667 {
668 if (dev_recursion_level())
669 return false;
670 if (!sk)
671 return true;
672 switch (sk->sk_family) {
673 case AF_INET:
674 return inet_sk(sk)->mc_loop;
675 #if IS_ENABLED(CONFIG_IPV6)
676 case AF_INET6:
677 return inet6_sk(sk)->mc_loop;
678 #endif
679 }
680 WARN_ON(1);
681 return true;
682 }
683 EXPORT_SYMBOL(sk_mc_loop);
684
685 /*
686 * This is meant for all protocols to use and covers goings on
687 * at the socket level. Everything here is generic.
688 */
689
690 int sock_setsockopt(struct socket *sock, int level, int optname,
691 char __user *optval, unsigned int optlen)
692 {
693 struct sock *sk = sock->sk;
694 int val;
695 int valbool;
696 struct linger ling;
697 int ret = 0;
698
699 /*
700 * Options without arguments
701 */
702
703 if (optname == SO_BINDTODEVICE)
704 return sock_setbindtodevice(sk, optval, optlen);
705
706 if (optlen < sizeof(int))
707 return -EINVAL;
708
709 if (get_user(val, (int __user *)optval))
710 return -EFAULT;
711
712 valbool = val ? 1 : 0;
713
714 lock_sock(sk);
715
716 switch (optname) {
717 case SO_DEBUG:
718 if (val && !capable(CAP_NET_ADMIN))
719 ret = -EACCES;
720 else
721 sock_valbool_flag(sk, SOCK_DBG, valbool);
722 break;
723 case SO_REUSEADDR:
724 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
725 break;
726 case SO_REUSEPORT:
727 sk->sk_reuseport = valbool;
728 break;
729 case SO_TYPE:
730 case SO_PROTOCOL:
731 case SO_DOMAIN:
732 case SO_ERROR:
733 ret = -ENOPROTOOPT;
734 break;
735 case SO_DONTROUTE:
736 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
737 break;
738 case SO_BROADCAST:
739 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
740 break;
741 case SO_SNDBUF:
742 /* Don't error on this BSD doesn't and if you think
743 * about it this is right. Otherwise apps have to
744 * play 'guess the biggest size' games. RCVBUF/SNDBUF
745 * are treated in BSD as hints
746 */
747 val = min_t(u32, val, sysctl_wmem_max);
748 set_sndbuf:
749 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
750 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
751 /* Wake up sending tasks if we upped the value. */
752 sk->sk_write_space(sk);
753 break;
754
755 case SO_SNDBUFFORCE:
756 if (!capable(CAP_NET_ADMIN)) {
757 ret = -EPERM;
758 break;
759 }
760 goto set_sndbuf;
761
762 case SO_RCVBUF:
763 /* Don't error on this BSD doesn't and if you think
764 * about it this is right. Otherwise apps have to
765 * play 'guess the biggest size' games. RCVBUF/SNDBUF
766 * are treated in BSD as hints
767 */
768 val = min_t(u32, val, sysctl_rmem_max);
769 set_rcvbuf:
770 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
771 /*
772 * We double it on the way in to account for
773 * "struct sk_buff" etc. overhead. Applications
774 * assume that the SO_RCVBUF setting they make will
775 * allow that much actual data to be received on that
776 * socket.
777 *
778 * Applications are unaware that "struct sk_buff" and
779 * other overheads allocate from the receive buffer
780 * during socket buffer allocation.
781 *
782 * And after considering the possible alternatives,
783 * returning the value we actually used in getsockopt
784 * is the most desirable behavior.
785 */
786 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
787 break;
788
789 case SO_RCVBUFFORCE:
790 if (!capable(CAP_NET_ADMIN)) {
791 ret = -EPERM;
792 break;
793 }
794 goto set_rcvbuf;
795
796 case SO_KEEPALIVE:
797 #ifdef CONFIG_INET
798 if (sk->sk_protocol == IPPROTO_TCP &&
799 sk->sk_type == SOCK_STREAM)
800 tcp_set_keepalive(sk, valbool);
801 #endif
802 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
803 break;
804
805 case SO_OOBINLINE:
806 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
807 break;
808
809 case SO_NO_CHECK:
810 sk->sk_no_check_tx = valbool;
811 break;
812
813 case SO_PRIORITY:
814 if ((val >= 0 && val <= 6) ||
815 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
816 sk->sk_priority = val;
817 else
818 ret = -EPERM;
819 break;
820
821 case SO_LINGER:
822 if (optlen < sizeof(ling)) {
823 ret = -EINVAL; /* 1003.1g */
824 break;
825 }
826 if (copy_from_user(&ling, optval, sizeof(ling))) {
827 ret = -EFAULT;
828 break;
829 }
830 if (!ling.l_onoff)
831 sock_reset_flag(sk, SOCK_LINGER);
832 else {
833 #if (BITS_PER_LONG == 32)
834 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
835 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
836 else
837 #endif
838 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
839 sock_set_flag(sk, SOCK_LINGER);
840 }
841 break;
842
843 case SO_BSDCOMPAT:
844 sock_warn_obsolete_bsdism("setsockopt");
845 break;
846
847 case SO_PASSCRED:
848 if (valbool)
849 set_bit(SOCK_PASSCRED, &sock->flags);
850 else
851 clear_bit(SOCK_PASSCRED, &sock->flags);
852 break;
853
854 case SO_TIMESTAMP:
855 case SO_TIMESTAMPNS:
856 if (valbool) {
857 if (optname == SO_TIMESTAMP)
858 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
859 else
860 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
861 sock_set_flag(sk, SOCK_RCVTSTAMP);
862 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
863 } else {
864 sock_reset_flag(sk, SOCK_RCVTSTAMP);
865 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
866 }
867 break;
868
869 case SO_TIMESTAMPING:
870 if (val & ~SOF_TIMESTAMPING_MASK) {
871 ret = -EINVAL;
872 break;
873 }
874
875 if (val & SOF_TIMESTAMPING_OPT_ID &&
876 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
877 if (sk->sk_protocol == IPPROTO_TCP) {
878 if (sk->sk_state != TCP_ESTABLISHED) {
879 ret = -EINVAL;
880 break;
881 }
882 sk->sk_tskey = tcp_sk(sk)->snd_una;
883 } else {
884 sk->sk_tskey = 0;
885 }
886 }
887 sk->sk_tsflags = val;
888 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
889 sock_enable_timestamp(sk,
890 SOCK_TIMESTAMPING_RX_SOFTWARE);
891 else
892 sock_disable_timestamp(sk,
893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
894 break;
895
896 case SO_RCVLOWAT:
897 if (val < 0)
898 val = INT_MAX;
899 sk->sk_rcvlowat = val ? : 1;
900 break;
901
902 case SO_RCVTIMEO:
903 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
904 break;
905
906 case SO_SNDTIMEO:
907 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
908 break;
909
910 case SO_ATTACH_FILTER:
911 ret = -EINVAL;
912 if (optlen == sizeof(struct sock_fprog)) {
913 struct sock_fprog fprog;
914
915 ret = -EFAULT;
916 if (copy_from_user(&fprog, optval, sizeof(fprog)))
917 break;
918
919 ret = sk_attach_filter(&fprog, sk);
920 }
921 break;
922
923 case SO_ATTACH_BPF:
924 ret = -EINVAL;
925 if (optlen == sizeof(u32)) {
926 u32 ufd;
927
928 ret = -EFAULT;
929 if (copy_from_user(&ufd, optval, sizeof(ufd)))
930 break;
931
932 ret = sk_attach_bpf(ufd, sk);
933 }
934 break;
935
936 case SO_DETACH_FILTER:
937 ret = sk_detach_filter(sk);
938 break;
939
940 case SO_LOCK_FILTER:
941 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
942 ret = -EPERM;
943 else
944 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
945 break;
946
947 case SO_PASSSEC:
948 if (valbool)
949 set_bit(SOCK_PASSSEC, &sock->flags);
950 else
951 clear_bit(SOCK_PASSSEC, &sock->flags);
952 break;
953 case SO_MARK:
954 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
955 ret = -EPERM;
956 else
957 sk->sk_mark = val;
958 break;
959
960 case SO_RXQ_OVFL:
961 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
962 break;
963
964 case SO_WIFI_STATUS:
965 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
966 break;
967
968 case SO_PEEK_OFF:
969 if (sock->ops->set_peek_off)
970 ret = sock->ops->set_peek_off(sk, val);
971 else
972 ret = -EOPNOTSUPP;
973 break;
974
975 case SO_NOFCS:
976 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
977 break;
978
979 case SO_SELECT_ERR_QUEUE:
980 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
981 break;
982
983 #ifdef CONFIG_NET_RX_BUSY_POLL
984 case SO_BUSY_POLL:
985 /* allow unprivileged users to decrease the value */
986 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
987 ret = -EPERM;
988 else {
989 if (val < 0)
990 ret = -EINVAL;
991 else
992 sk->sk_ll_usec = val;
993 }
994 break;
995 #endif
996
997 case SO_MAX_PACING_RATE:
998 sk->sk_max_pacing_rate = val;
999 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1000 sk->sk_max_pacing_rate);
1001 break;
1002
1003 case SO_INCOMING_CPU:
1004 sk->sk_incoming_cpu = val;
1005 break;
1006
1007 default:
1008 ret = -ENOPROTOOPT;
1009 break;
1010 }
1011 release_sock(sk);
1012 return ret;
1013 }
1014 EXPORT_SYMBOL(sock_setsockopt);
1015
1016
1017 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1018 struct ucred *ucred)
1019 {
1020 ucred->pid = pid_vnr(pid);
1021 ucred->uid = ucred->gid = -1;
1022 if (cred) {
1023 struct user_namespace *current_ns = current_user_ns();
1024
1025 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1026 ucred->gid = from_kgid_munged(current_ns, cred->egid);
1027 }
1028 }
1029
1030 int sock_getsockopt(struct socket *sock, int level, int optname,
1031 char __user *optval, int __user *optlen)
1032 {
1033 struct sock *sk = sock->sk;
1034
1035 union {
1036 int val;
1037 struct linger ling;
1038 struct timeval tm;
1039 } v;
1040
1041 int lv = sizeof(int);
1042 int len;
1043
1044 if (get_user(len, optlen))
1045 return -EFAULT;
1046 if (len < 0)
1047 return -EINVAL;
1048
1049 memset(&v, 0, sizeof(v));
1050
1051 switch (optname) {
1052 case SO_DEBUG:
1053 v.val = sock_flag(sk, SOCK_DBG);
1054 break;
1055
1056 case SO_DONTROUTE:
1057 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1058 break;
1059
1060 case SO_BROADCAST:
1061 v.val = sock_flag(sk, SOCK_BROADCAST);
1062 break;
1063
1064 case SO_SNDBUF:
1065 v.val = sk->sk_sndbuf;
1066 break;
1067
1068 case SO_RCVBUF:
1069 v.val = sk->sk_rcvbuf;
1070 break;
1071
1072 case SO_REUSEADDR:
1073 v.val = sk->sk_reuse;
1074 break;
1075
1076 case SO_REUSEPORT:
1077 v.val = sk->sk_reuseport;
1078 break;
1079
1080 case SO_KEEPALIVE:
1081 v.val = sock_flag(sk, SOCK_KEEPOPEN);
1082 break;
1083
1084 case SO_TYPE:
1085 v.val = sk->sk_type;
1086 break;
1087
1088 case SO_PROTOCOL:
1089 v.val = sk->sk_protocol;
1090 break;
1091
1092 case SO_DOMAIN:
1093 v.val = sk->sk_family;
1094 break;
1095
1096 case SO_ERROR:
1097 v.val = -sock_error(sk);
1098 if (v.val == 0)
1099 v.val = xchg(&sk->sk_err_soft, 0);
1100 break;
1101
1102 case SO_OOBINLINE:
1103 v.val = sock_flag(sk, SOCK_URGINLINE);
1104 break;
1105
1106 case SO_NO_CHECK:
1107 v.val = sk->sk_no_check_tx;
1108 break;
1109
1110 case SO_PRIORITY:
1111 v.val = sk->sk_priority;
1112 break;
1113
1114 case SO_LINGER:
1115 lv = sizeof(v.ling);
1116 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1117 v.ling.l_linger = sk->sk_lingertime / HZ;
1118 break;
1119
1120 case SO_BSDCOMPAT:
1121 sock_warn_obsolete_bsdism("getsockopt");
1122 break;
1123
1124 case SO_TIMESTAMP:
1125 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1126 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1127 break;
1128
1129 case SO_TIMESTAMPNS:
1130 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1131 break;
1132
1133 case SO_TIMESTAMPING:
1134 v.val = sk->sk_tsflags;
1135 break;
1136
1137 case SO_RCVTIMEO:
1138 lv = sizeof(struct timeval);
1139 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1140 v.tm.tv_sec = 0;
1141 v.tm.tv_usec = 0;
1142 } else {
1143 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1144 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1145 }
1146 break;
1147
1148 case SO_SNDTIMEO:
1149 lv = sizeof(struct timeval);
1150 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1151 v.tm.tv_sec = 0;
1152 v.tm.tv_usec = 0;
1153 } else {
1154 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1155 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1156 }
1157 break;
1158
1159 case SO_RCVLOWAT:
1160 v.val = sk->sk_rcvlowat;
1161 break;
1162
1163 case SO_SNDLOWAT:
1164 v.val = 1;
1165 break;
1166
1167 case SO_PASSCRED:
1168 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1169 break;
1170
1171 case SO_PEERCRED:
1172 {
1173 struct ucred peercred;
1174 if (len > sizeof(peercred))
1175 len = sizeof(peercred);
1176 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1177 if (copy_to_user(optval, &peercred, len))
1178 return -EFAULT;
1179 goto lenout;
1180 }
1181
1182 case SO_PEERNAME:
1183 {
1184 char address[128];
1185
1186 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1187 return -ENOTCONN;
1188 if (lv < len)
1189 return -EINVAL;
1190 if (copy_to_user(optval, address, len))
1191 return -EFAULT;
1192 goto lenout;
1193 }
1194
1195 /* Dubious BSD thing... Probably nobody even uses it, but
1196 * the UNIX standard wants it for whatever reason... -DaveM
1197 */
1198 case SO_ACCEPTCONN:
1199 v.val = sk->sk_state == TCP_LISTEN;
1200 break;
1201
1202 case SO_PASSSEC:
1203 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1204 break;
1205
1206 case SO_PEERSEC:
1207 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1208
1209 case SO_MARK:
1210 v.val = sk->sk_mark;
1211 break;
1212
1213 case SO_RXQ_OVFL:
1214 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1215 break;
1216
1217 case SO_WIFI_STATUS:
1218 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1219 break;
1220
1221 case SO_PEEK_OFF:
1222 if (!sock->ops->set_peek_off)
1223 return -EOPNOTSUPP;
1224
1225 v.val = sk->sk_peek_off;
1226 break;
1227 case SO_NOFCS:
1228 v.val = sock_flag(sk, SOCK_NOFCS);
1229 break;
1230
1231 case SO_BINDTODEVICE:
1232 return sock_getbindtodevice(sk, optval, optlen, len);
1233
1234 case SO_GET_FILTER:
1235 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1236 if (len < 0)
1237 return len;
1238
1239 goto lenout;
1240
1241 case SO_LOCK_FILTER:
1242 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1243 break;
1244
1245 case SO_BPF_EXTENSIONS:
1246 v.val = bpf_tell_extensions();
1247 break;
1248
1249 case SO_SELECT_ERR_QUEUE:
1250 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1251 break;
1252
1253 #ifdef CONFIG_NET_RX_BUSY_POLL
1254 case SO_BUSY_POLL:
1255 v.val = sk->sk_ll_usec;
1256 break;
1257 #endif
1258
1259 case SO_MAX_PACING_RATE:
1260 v.val = sk->sk_max_pacing_rate;
1261 break;
1262
1263 case SO_INCOMING_CPU:
1264 v.val = sk->sk_incoming_cpu;
1265 break;
1266
1267 default:
1268 /* We implement the SO_SNDLOWAT etc to not be settable
1269 * (1003.1g 7).
1270 */
1271 return -ENOPROTOOPT;
1272 }
1273
1274 if (len > lv)
1275 len = lv;
1276 if (copy_to_user(optval, &v, len))
1277 return -EFAULT;
1278 lenout:
1279 if (put_user(len, optlen))
1280 return -EFAULT;
1281 return 0;
1282 }
1283
1284 /*
1285 * Initialize an sk_lock.
1286 *
1287 * (We also register the sk_lock with the lock validator.)
1288 */
1289 static inline void sock_lock_init(struct sock *sk)
1290 {
1291 sock_lock_init_class_and_name(sk,
1292 af_family_slock_key_strings[sk->sk_family],
1293 af_family_slock_keys + sk->sk_family,
1294 af_family_key_strings[sk->sk_family],
1295 af_family_keys + sk->sk_family);
1296 }
1297
1298 /*
1299 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1300 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1301 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1302 */
1303 static void sock_copy(struct sock *nsk, const struct sock *osk)
1304 {
1305 #ifdef CONFIG_SECURITY_NETWORK
1306 void *sptr = nsk->sk_security;
1307 #endif
1308 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1309
1310 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1311 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1312
1313 #ifdef CONFIG_SECURITY_NETWORK
1314 nsk->sk_security = sptr;
1315 security_sk_clone(osk, nsk);
1316 #endif
1317 }
1318
1319 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1320 {
1321 unsigned long nulls1, nulls2;
1322
1323 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1324 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1325 if (nulls1 > nulls2)
1326 swap(nulls1, nulls2);
1327
1328 if (nulls1 != 0)
1329 memset((char *)sk, 0, nulls1);
1330 memset((char *)sk + nulls1 + sizeof(void *), 0,
1331 nulls2 - nulls1 - sizeof(void *));
1332 memset((char *)sk + nulls2 + sizeof(void *), 0,
1333 size - nulls2 - sizeof(void *));
1334 }
1335 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1336
1337 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1338 int family)
1339 {
1340 struct sock *sk;
1341 struct kmem_cache *slab;
1342
1343 slab = prot->slab;
1344 if (slab != NULL) {
1345 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1346 if (!sk)
1347 return sk;
1348 if (priority & __GFP_ZERO) {
1349 if (prot->clear_sk)
1350 prot->clear_sk(sk, prot->obj_size);
1351 else
1352 sk_prot_clear_nulls(sk, prot->obj_size);
1353 }
1354 } else
1355 sk = kmalloc(prot->obj_size, priority);
1356
1357 if (sk != NULL) {
1358 kmemcheck_annotate_bitfield(sk, flags);
1359
1360 if (security_sk_alloc(sk, family, priority))
1361 goto out_free;
1362
1363 if (!try_module_get(prot->owner))
1364 goto out_free_sec;
1365 sk_tx_queue_clear(sk);
1366 }
1367
1368 return sk;
1369
1370 out_free_sec:
1371 security_sk_free(sk);
1372 out_free:
1373 if (slab != NULL)
1374 kmem_cache_free(slab, sk);
1375 else
1376 kfree(sk);
1377 return NULL;
1378 }
1379
1380 static void sk_prot_free(struct proto *prot, struct sock *sk)
1381 {
1382 struct kmem_cache *slab;
1383 struct module *owner;
1384
1385 owner = prot->owner;
1386 slab = prot->slab;
1387
1388 security_sk_free(sk);
1389 if (slab != NULL)
1390 kmem_cache_free(slab, sk);
1391 else
1392 kfree(sk);
1393 module_put(owner);
1394 }
1395
1396 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1397 void sock_update_netprioidx(struct sock *sk)
1398 {
1399 if (in_interrupt())
1400 return;
1401
1402 sk->sk_cgrp_prioidx = task_netprioidx(current);
1403 }
1404 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1405 #endif
1406
1407 /**
1408 * sk_alloc - All socket objects are allocated here
1409 * @net: the applicable net namespace
1410 * @family: protocol family
1411 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1412 * @prot: struct proto associated with this new sock instance
1413 * @kern: is this to be a kernel socket?
1414 */
1415 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1416 struct proto *prot, int kern)
1417 {
1418 struct sock *sk;
1419
1420 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1421 if (sk) {
1422 sk->sk_family = family;
1423 /*
1424 * See comment in struct sock definition to understand
1425 * why we need sk_prot_creator -acme
1426 */
1427 sk->sk_prot = sk->sk_prot_creator = prot;
1428 sock_lock_init(sk);
1429 sk->sk_net_refcnt = kern ? 0 : 1;
1430 if (likely(sk->sk_net_refcnt))
1431 get_net(net);
1432 sock_net_set(sk, net);
1433 atomic_set(&sk->sk_wmem_alloc, 1);
1434
1435 sock_update_classid(sk);
1436 sock_update_netprioidx(sk);
1437 }
1438
1439 return sk;
1440 }
1441 EXPORT_SYMBOL(sk_alloc);
1442
1443 void sk_destruct(struct sock *sk)
1444 {
1445 struct sk_filter *filter;
1446
1447 if (sk->sk_destruct)
1448 sk->sk_destruct(sk);
1449
1450 filter = rcu_dereference_check(sk->sk_filter,
1451 atomic_read(&sk->sk_wmem_alloc) == 0);
1452 if (filter) {
1453 sk_filter_uncharge(sk, filter);
1454 RCU_INIT_POINTER(sk->sk_filter, NULL);
1455 }
1456
1457 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1458
1459 if (atomic_read(&sk->sk_omem_alloc))
1460 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1461 __func__, atomic_read(&sk->sk_omem_alloc));
1462
1463 if (sk->sk_peer_cred)
1464 put_cred(sk->sk_peer_cred);
1465 put_pid(sk->sk_peer_pid);
1466 if (likely(sk->sk_net_refcnt))
1467 put_net(sock_net(sk));
1468 sk_prot_free(sk->sk_prot_creator, sk);
1469 }
1470
1471 static void __sk_free(struct sock *sk)
1472 {
1473 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1474 sock_diag_broadcast_destroy(sk);
1475 else
1476 sk_destruct(sk);
1477 }
1478
1479 void sk_free(struct sock *sk)
1480 {
1481 /*
1482 * We subtract one from sk_wmem_alloc and can know if
1483 * some packets are still in some tx queue.
1484 * If not null, sock_wfree() will call __sk_free(sk) later
1485 */
1486 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1487 __sk_free(sk);
1488 }
1489 EXPORT_SYMBOL(sk_free);
1490
1491 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1492 {
1493 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1494 sock_update_memcg(newsk);
1495 }
1496
1497 /**
1498 * sk_clone_lock - clone a socket, and lock its clone
1499 * @sk: the socket to clone
1500 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1501 *
1502 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1503 */
1504 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1505 {
1506 struct sock *newsk;
1507 bool is_charged = true;
1508
1509 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1510 if (newsk != NULL) {
1511 struct sk_filter *filter;
1512
1513 sock_copy(newsk, sk);
1514
1515 /* SANITY */
1516 if (likely(newsk->sk_net_refcnt))
1517 get_net(sock_net(newsk));
1518 sk_node_init(&newsk->sk_node);
1519 sock_lock_init(newsk);
1520 bh_lock_sock(newsk);
1521 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1522 newsk->sk_backlog.len = 0;
1523
1524 atomic_set(&newsk->sk_rmem_alloc, 0);
1525 /*
1526 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1527 */
1528 atomic_set(&newsk->sk_wmem_alloc, 1);
1529 atomic_set(&newsk->sk_omem_alloc, 0);
1530 skb_queue_head_init(&newsk->sk_receive_queue);
1531 skb_queue_head_init(&newsk->sk_write_queue);
1532
1533 spin_lock_init(&newsk->sk_dst_lock);
1534 rwlock_init(&newsk->sk_callback_lock);
1535 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1536 af_callback_keys + newsk->sk_family,
1537 af_family_clock_key_strings[newsk->sk_family]);
1538
1539 newsk->sk_dst_cache = NULL;
1540 newsk->sk_wmem_queued = 0;
1541 newsk->sk_forward_alloc = 0;
1542 newsk->sk_send_head = NULL;
1543 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1544
1545 sock_reset_flag(newsk, SOCK_DONE);
1546 skb_queue_head_init(&newsk->sk_error_queue);
1547
1548 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1549 if (filter != NULL)
1550 /* though it's an empty new sock, the charging may fail
1551 * if sysctl_optmem_max was changed between creation of
1552 * original socket and cloning
1553 */
1554 is_charged = sk_filter_charge(newsk, filter);
1555
1556 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) {
1557 /* It is still raw copy of parent, so invalidate
1558 * destructor and make plain sk_free() */
1559 newsk->sk_destruct = NULL;
1560 bh_unlock_sock(newsk);
1561 sk_free(newsk);
1562 newsk = NULL;
1563 goto out;
1564 }
1565
1566 newsk->sk_err = 0;
1567 newsk->sk_priority = 0;
1568 newsk->sk_incoming_cpu = raw_smp_processor_id();
1569 atomic64_set(&newsk->sk_cookie, 0);
1570 /*
1571 * Before updating sk_refcnt, we must commit prior changes to memory
1572 * (Documentation/RCU/rculist_nulls.txt for details)
1573 */
1574 smp_wmb();
1575 atomic_set(&newsk->sk_refcnt, 2);
1576
1577 /*
1578 * Increment the counter in the same struct proto as the master
1579 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1580 * is the same as sk->sk_prot->socks, as this field was copied
1581 * with memcpy).
1582 *
1583 * This _changes_ the previous behaviour, where
1584 * tcp_create_openreq_child always was incrementing the
1585 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1586 * to be taken into account in all callers. -acme
1587 */
1588 sk_refcnt_debug_inc(newsk);
1589 sk_set_socket(newsk, NULL);
1590 newsk->sk_wq = NULL;
1591
1592 sk_update_clone(sk, newsk);
1593
1594 if (newsk->sk_prot->sockets_allocated)
1595 sk_sockets_allocated_inc(newsk);
1596
1597 if (sock_needs_netstamp(sk) &&
1598 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1599 net_enable_timestamp();
1600 }
1601 out:
1602 return newsk;
1603 }
1604 EXPORT_SYMBOL_GPL(sk_clone_lock);
1605
1606 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1607 {
1608 u32 max_segs = 1;
1609
1610 __sk_dst_set(sk, dst);
1611 sk->sk_route_caps = dst->dev->features;
1612 if (sk->sk_route_caps & NETIF_F_GSO)
1613 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1614 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1615 if (sk_can_gso(sk)) {
1616 if (dst->header_len) {
1617 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1618 } else {
1619 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1620 sk->sk_gso_max_size = dst->dev->gso_max_size;
1621 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1622 }
1623 }
1624 sk->sk_gso_max_segs = max_segs;
1625 }
1626 EXPORT_SYMBOL_GPL(sk_setup_caps);
1627
1628 /*
1629 * Simple resource managers for sockets.
1630 */
1631
1632
1633 /*
1634 * Write buffer destructor automatically called from kfree_skb.
1635 */
1636 void sock_wfree(struct sk_buff *skb)
1637 {
1638 struct sock *sk = skb->sk;
1639 unsigned int len = skb->truesize;
1640
1641 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1642 /*
1643 * Keep a reference on sk_wmem_alloc, this will be released
1644 * after sk_write_space() call
1645 */
1646 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1647 sk->sk_write_space(sk);
1648 len = 1;
1649 }
1650 /*
1651 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1652 * could not do because of in-flight packets
1653 */
1654 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1655 __sk_free(sk);
1656 }
1657 EXPORT_SYMBOL(sock_wfree);
1658
1659 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1660 {
1661 skb_orphan(skb);
1662 skb->sk = sk;
1663 #ifdef CONFIG_INET
1664 if (unlikely(!sk_fullsock(sk))) {
1665 skb->destructor = sock_edemux;
1666 sock_hold(sk);
1667 return;
1668 }
1669 #endif
1670 skb->destructor = sock_wfree;
1671 skb_set_hash_from_sk(skb, sk);
1672 /*
1673 * We used to take a refcount on sk, but following operation
1674 * is enough to guarantee sk_free() wont free this sock until
1675 * all in-flight packets are completed
1676 */
1677 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1678 }
1679 EXPORT_SYMBOL(skb_set_owner_w);
1680
1681 void skb_orphan_partial(struct sk_buff *skb)
1682 {
1683 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1684 * so we do not completely orphan skb, but transfert all
1685 * accounted bytes but one, to avoid unexpected reorders.
1686 */
1687 if (skb->destructor == sock_wfree
1688 #ifdef CONFIG_INET
1689 || skb->destructor == tcp_wfree
1690 #endif
1691 ) {
1692 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1693 skb->truesize = 1;
1694 } else {
1695 skb_orphan(skb);
1696 }
1697 }
1698 EXPORT_SYMBOL(skb_orphan_partial);
1699
1700 /*
1701 * Read buffer destructor automatically called from kfree_skb.
1702 */
1703 void sock_rfree(struct sk_buff *skb)
1704 {
1705 struct sock *sk = skb->sk;
1706 unsigned int len = skb->truesize;
1707
1708 atomic_sub(len, &sk->sk_rmem_alloc);
1709 sk_mem_uncharge(sk, len);
1710 }
1711 EXPORT_SYMBOL(sock_rfree);
1712
1713 /*
1714 * Buffer destructor for skbs that are not used directly in read or write
1715 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1716 */
1717 void sock_efree(struct sk_buff *skb)
1718 {
1719 sock_put(skb->sk);
1720 }
1721 EXPORT_SYMBOL(sock_efree);
1722
1723 kuid_t sock_i_uid(struct sock *sk)
1724 {
1725 kuid_t uid;
1726
1727 read_lock_bh(&sk->sk_callback_lock);
1728 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1729 read_unlock_bh(&sk->sk_callback_lock);
1730 return uid;
1731 }
1732 EXPORT_SYMBOL(sock_i_uid);
1733
1734 unsigned long sock_i_ino(struct sock *sk)
1735 {
1736 unsigned long ino;
1737
1738 read_lock_bh(&sk->sk_callback_lock);
1739 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1740 read_unlock_bh(&sk->sk_callback_lock);
1741 return ino;
1742 }
1743 EXPORT_SYMBOL(sock_i_ino);
1744
1745 /*
1746 * Allocate a skb from the socket's send buffer.
1747 */
1748 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1749 gfp_t priority)
1750 {
1751 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1752 struct sk_buff *skb = alloc_skb(size, priority);
1753 if (skb) {
1754 skb_set_owner_w(skb, sk);
1755 return skb;
1756 }
1757 }
1758 return NULL;
1759 }
1760 EXPORT_SYMBOL(sock_wmalloc);
1761
1762 /*
1763 * Allocate a memory block from the socket's option memory buffer.
1764 */
1765 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1766 {
1767 if ((unsigned int)size <= sysctl_optmem_max &&
1768 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1769 void *mem;
1770 /* First do the add, to avoid the race if kmalloc
1771 * might sleep.
1772 */
1773 atomic_add(size, &sk->sk_omem_alloc);
1774 mem = kmalloc(size, priority);
1775 if (mem)
1776 return mem;
1777 atomic_sub(size, &sk->sk_omem_alloc);
1778 }
1779 return NULL;
1780 }
1781 EXPORT_SYMBOL(sock_kmalloc);
1782
1783 /* Free an option memory block. Note, we actually want the inline
1784 * here as this allows gcc to detect the nullify and fold away the
1785 * condition entirely.
1786 */
1787 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1788 const bool nullify)
1789 {
1790 if (WARN_ON_ONCE(!mem))
1791 return;
1792 if (nullify)
1793 kzfree(mem);
1794 else
1795 kfree(mem);
1796 atomic_sub(size, &sk->sk_omem_alloc);
1797 }
1798
1799 void sock_kfree_s(struct sock *sk, void *mem, int size)
1800 {
1801 __sock_kfree_s(sk, mem, size, false);
1802 }
1803 EXPORT_SYMBOL(sock_kfree_s);
1804
1805 void sock_kzfree_s(struct sock *sk, void *mem, int size)
1806 {
1807 __sock_kfree_s(sk, mem, size, true);
1808 }
1809 EXPORT_SYMBOL(sock_kzfree_s);
1810
1811 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1812 I think, these locks should be removed for datagram sockets.
1813 */
1814 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1815 {
1816 DEFINE_WAIT(wait);
1817
1818 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1819 for (;;) {
1820 if (!timeo)
1821 break;
1822 if (signal_pending(current))
1823 break;
1824 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1825 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1826 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1827 break;
1828 if (sk->sk_shutdown & SEND_SHUTDOWN)
1829 break;
1830 if (sk->sk_err)
1831 break;
1832 timeo = schedule_timeout(timeo);
1833 }
1834 finish_wait(sk_sleep(sk), &wait);
1835 return timeo;
1836 }
1837
1838
1839 /*
1840 * Generic send/receive buffer handlers
1841 */
1842
1843 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1844 unsigned long data_len, int noblock,
1845 int *errcode, int max_page_order)
1846 {
1847 struct sk_buff *skb;
1848 long timeo;
1849 int err;
1850
1851 timeo = sock_sndtimeo(sk, noblock);
1852 for (;;) {
1853 err = sock_error(sk);
1854 if (err != 0)
1855 goto failure;
1856
1857 err = -EPIPE;
1858 if (sk->sk_shutdown & SEND_SHUTDOWN)
1859 goto failure;
1860
1861 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1862 break;
1863
1864 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1865 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1866 err = -EAGAIN;
1867 if (!timeo)
1868 goto failure;
1869 if (signal_pending(current))
1870 goto interrupted;
1871 timeo = sock_wait_for_wmem(sk, timeo);
1872 }
1873 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1874 errcode, sk->sk_allocation);
1875 if (skb)
1876 skb_set_owner_w(skb, sk);
1877 return skb;
1878
1879 interrupted:
1880 err = sock_intr_errno(timeo);
1881 failure:
1882 *errcode = err;
1883 return NULL;
1884 }
1885 EXPORT_SYMBOL(sock_alloc_send_pskb);
1886
1887 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1888 int noblock, int *errcode)
1889 {
1890 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1891 }
1892 EXPORT_SYMBOL(sock_alloc_send_skb);
1893
1894 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1895 struct sockcm_cookie *sockc)
1896 {
1897 struct cmsghdr *cmsg;
1898
1899 for_each_cmsghdr(cmsg, msg) {
1900 if (!CMSG_OK(msg, cmsg))
1901 return -EINVAL;
1902 if (cmsg->cmsg_level != SOL_SOCKET)
1903 continue;
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 default:
1913 return -EINVAL;
1914 }
1915 }
1916 return 0;
1917 }
1918 EXPORT_SYMBOL(sock_cmsg_send);
1919
1920 /* On 32bit arches, an skb frag is limited to 2^15 */
1921 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1922
1923 /**
1924 * skb_page_frag_refill - check that a page_frag contains enough room
1925 * @sz: minimum size of the fragment we want to get
1926 * @pfrag: pointer to page_frag
1927 * @gfp: priority for memory allocation
1928 *
1929 * Note: While this allocator tries to use high order pages, there is
1930 * no guarantee that allocations succeed. Therefore, @sz MUST be
1931 * less or equal than PAGE_SIZE.
1932 */
1933 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
1934 {
1935 if (pfrag->page) {
1936 if (atomic_read(&pfrag->page->_count) == 1) {
1937 pfrag->offset = 0;
1938 return true;
1939 }
1940 if (pfrag->offset + sz <= pfrag->size)
1941 return true;
1942 put_page(pfrag->page);
1943 }
1944
1945 pfrag->offset = 0;
1946 if (SKB_FRAG_PAGE_ORDER) {
1947 /* Avoid direct reclaim but allow kswapd to wake */
1948 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1949 __GFP_COMP | __GFP_NOWARN |
1950 __GFP_NORETRY,
1951 SKB_FRAG_PAGE_ORDER);
1952 if (likely(pfrag->page)) {
1953 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
1954 return true;
1955 }
1956 }
1957 pfrag->page = alloc_page(gfp);
1958 if (likely(pfrag->page)) {
1959 pfrag->size = PAGE_SIZE;
1960 return true;
1961 }
1962 return false;
1963 }
1964 EXPORT_SYMBOL(skb_page_frag_refill);
1965
1966 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1967 {
1968 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1969 return true;
1970
1971 sk_enter_memory_pressure(sk);
1972 sk_stream_moderate_sndbuf(sk);
1973 return false;
1974 }
1975 EXPORT_SYMBOL(sk_page_frag_refill);
1976
1977 static void __lock_sock(struct sock *sk)
1978 __releases(&sk->sk_lock.slock)
1979 __acquires(&sk->sk_lock.slock)
1980 {
1981 DEFINE_WAIT(wait);
1982
1983 for (;;) {
1984 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1985 TASK_UNINTERRUPTIBLE);
1986 spin_unlock_bh(&sk->sk_lock.slock);
1987 schedule();
1988 spin_lock_bh(&sk->sk_lock.slock);
1989 if (!sock_owned_by_user(sk))
1990 break;
1991 }
1992 finish_wait(&sk->sk_lock.wq, &wait);
1993 }
1994
1995 static void __release_sock(struct sock *sk)
1996 __releases(&sk->sk_lock.slock)
1997 __acquires(&sk->sk_lock.slock)
1998 {
1999 struct sk_buff *skb = sk->sk_backlog.head;
2000
2001 do {
2002 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2003 bh_unlock_sock(sk);
2004
2005 do {
2006 struct sk_buff *next = skb->next;
2007
2008 prefetch(next);
2009 WARN_ON_ONCE(skb_dst_is_noref(skb));
2010 skb->next = NULL;
2011 sk_backlog_rcv(sk, skb);
2012
2013 /*
2014 * We are in process context here with softirqs
2015 * disabled, use cond_resched_softirq() to preempt.
2016 * This is safe to do because we've taken the backlog
2017 * queue private:
2018 */
2019 cond_resched_softirq();
2020
2021 skb = next;
2022 } while (skb != NULL);
2023
2024 bh_lock_sock(sk);
2025 } while ((skb = sk->sk_backlog.head) != NULL);
2026
2027 /*
2028 * Doing the zeroing here guarantee we can not loop forever
2029 * while a wild producer attempts to flood us.
2030 */
2031 sk->sk_backlog.len = 0;
2032 }
2033
2034 /**
2035 * sk_wait_data - wait for data to arrive at sk_receive_queue
2036 * @sk: sock to wait on
2037 * @timeo: for how long
2038 * @skb: last skb seen on sk_receive_queue
2039 *
2040 * Now socket state including sk->sk_err is changed only under lock,
2041 * hence we may omit checks after joining wait queue.
2042 * We check receive queue before schedule() only as optimization;
2043 * it is very likely that release_sock() added new data.
2044 */
2045 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2046 {
2047 int rc;
2048 DEFINE_WAIT(wait);
2049
2050 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2051 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2052 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
2053 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2054 finish_wait(sk_sleep(sk), &wait);
2055 return rc;
2056 }
2057 EXPORT_SYMBOL(sk_wait_data);
2058
2059 /**
2060 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2061 * @sk: socket
2062 * @size: memory size to allocate
2063 * @kind: allocation type
2064 *
2065 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2066 * rmem allocation. This function assumes that protocols which have
2067 * memory_pressure use sk_wmem_queued as write buffer accounting.
2068 */
2069 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2070 {
2071 struct proto *prot = sk->sk_prot;
2072 int amt = sk_mem_pages(size);
2073 long allocated;
2074 int parent_status = UNDER_LIMIT;
2075
2076 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
2077
2078 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
2079
2080 /* Under limit. */
2081 if (parent_status == UNDER_LIMIT &&
2082 allocated <= sk_prot_mem_limits(sk, 0)) {
2083 sk_leave_memory_pressure(sk);
2084 return 1;
2085 }
2086
2087 /* Under pressure. (we or our parents) */
2088 if ((parent_status > SOFT_LIMIT) ||
2089 allocated > sk_prot_mem_limits(sk, 1))
2090 sk_enter_memory_pressure(sk);
2091
2092 /* Over hard limit (we or our parents) */
2093 if ((parent_status == OVER_LIMIT) ||
2094 (allocated > sk_prot_mem_limits(sk, 2)))
2095 goto suppress_allocation;
2096
2097 /* guarantee minimum buffer size under pressure */
2098 if (kind == SK_MEM_RECV) {
2099 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2100 return 1;
2101
2102 } else { /* SK_MEM_SEND */
2103 if (sk->sk_type == SOCK_STREAM) {
2104 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2105 return 1;
2106 } else if (atomic_read(&sk->sk_wmem_alloc) <
2107 prot->sysctl_wmem[0])
2108 return 1;
2109 }
2110
2111 if (sk_has_memory_pressure(sk)) {
2112 int alloc;
2113
2114 if (!sk_under_memory_pressure(sk))
2115 return 1;
2116 alloc = sk_sockets_allocated_read_positive(sk);
2117 if (sk_prot_mem_limits(sk, 2) > alloc *
2118 sk_mem_pages(sk->sk_wmem_queued +
2119 atomic_read(&sk->sk_rmem_alloc) +
2120 sk->sk_forward_alloc))
2121 return 1;
2122 }
2123
2124 suppress_allocation:
2125
2126 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2127 sk_stream_moderate_sndbuf(sk);
2128
2129 /* Fail only if socket is _under_ its sndbuf.
2130 * In this case we cannot block, so that we have to fail.
2131 */
2132 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2133 return 1;
2134 }
2135
2136 trace_sock_exceed_buf_limit(sk, prot, allocated);
2137
2138 /* Alas. Undo changes. */
2139 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2140
2141 sk_memory_allocated_sub(sk, amt);
2142
2143 return 0;
2144 }
2145 EXPORT_SYMBOL(__sk_mem_schedule);
2146
2147 /**
2148 * __sk_mem_reclaim - reclaim memory_allocated
2149 * @sk: socket
2150 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2151 */
2152 void __sk_mem_reclaim(struct sock *sk, int amount)
2153 {
2154 amount >>= SK_MEM_QUANTUM_SHIFT;
2155 sk_memory_allocated_sub(sk, amount);
2156 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2157
2158 if (sk_under_memory_pressure(sk) &&
2159 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2160 sk_leave_memory_pressure(sk);
2161 }
2162 EXPORT_SYMBOL(__sk_mem_reclaim);
2163
2164
2165 /*
2166 * Set of default routines for initialising struct proto_ops when
2167 * the protocol does not support a particular function. In certain
2168 * cases where it makes no sense for a protocol to have a "do nothing"
2169 * function, some default processing is provided.
2170 */
2171
2172 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2173 {
2174 return -EOPNOTSUPP;
2175 }
2176 EXPORT_SYMBOL(sock_no_bind);
2177
2178 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2179 int len, int flags)
2180 {
2181 return -EOPNOTSUPP;
2182 }
2183 EXPORT_SYMBOL(sock_no_connect);
2184
2185 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2186 {
2187 return -EOPNOTSUPP;
2188 }
2189 EXPORT_SYMBOL(sock_no_socketpair);
2190
2191 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2192 {
2193 return -EOPNOTSUPP;
2194 }
2195 EXPORT_SYMBOL(sock_no_accept);
2196
2197 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2198 int *len, int peer)
2199 {
2200 return -EOPNOTSUPP;
2201 }
2202 EXPORT_SYMBOL(sock_no_getname);
2203
2204 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2205 {
2206 return 0;
2207 }
2208 EXPORT_SYMBOL(sock_no_poll);
2209
2210 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2211 {
2212 return -EOPNOTSUPP;
2213 }
2214 EXPORT_SYMBOL(sock_no_ioctl);
2215
2216 int sock_no_listen(struct socket *sock, int backlog)
2217 {
2218 return -EOPNOTSUPP;
2219 }
2220 EXPORT_SYMBOL(sock_no_listen);
2221
2222 int sock_no_shutdown(struct socket *sock, int how)
2223 {
2224 return -EOPNOTSUPP;
2225 }
2226 EXPORT_SYMBOL(sock_no_shutdown);
2227
2228 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2229 char __user *optval, unsigned int optlen)
2230 {
2231 return -EOPNOTSUPP;
2232 }
2233 EXPORT_SYMBOL(sock_no_setsockopt);
2234
2235 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2236 char __user *optval, int __user *optlen)
2237 {
2238 return -EOPNOTSUPP;
2239 }
2240 EXPORT_SYMBOL(sock_no_getsockopt);
2241
2242 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2243 {
2244 return -EOPNOTSUPP;
2245 }
2246 EXPORT_SYMBOL(sock_no_sendmsg);
2247
2248 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2249 int flags)
2250 {
2251 return -EOPNOTSUPP;
2252 }
2253 EXPORT_SYMBOL(sock_no_recvmsg);
2254
2255 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2256 {
2257 /* Mirror missing mmap method error code */
2258 return -ENODEV;
2259 }
2260 EXPORT_SYMBOL(sock_no_mmap);
2261
2262 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2263 {
2264 ssize_t res;
2265 struct msghdr msg = {.msg_flags = flags};
2266 struct kvec iov;
2267 char *kaddr = kmap(page);
2268 iov.iov_base = kaddr + offset;
2269 iov.iov_len = size;
2270 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2271 kunmap(page);
2272 return res;
2273 }
2274 EXPORT_SYMBOL(sock_no_sendpage);
2275
2276 /*
2277 * Default Socket Callbacks
2278 */
2279
2280 static void sock_def_wakeup(struct sock *sk)
2281 {
2282 struct socket_wq *wq;
2283
2284 rcu_read_lock();
2285 wq = rcu_dereference(sk->sk_wq);
2286 if (wq_has_sleeper(wq))
2287 wake_up_interruptible_all(&wq->wait);
2288 rcu_read_unlock();
2289 }
2290
2291 static void sock_def_error_report(struct sock *sk)
2292 {
2293 struct socket_wq *wq;
2294
2295 rcu_read_lock();
2296 wq = rcu_dereference(sk->sk_wq);
2297 if (wq_has_sleeper(wq))
2298 wake_up_interruptible_poll(&wq->wait, POLLERR);
2299 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2300 rcu_read_unlock();
2301 }
2302
2303 static void sock_def_readable(struct sock *sk)
2304 {
2305 struct socket_wq *wq;
2306
2307 rcu_read_lock();
2308 wq = rcu_dereference(sk->sk_wq);
2309 if (wq_has_sleeper(wq))
2310 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2311 POLLRDNORM | POLLRDBAND);
2312 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2313 rcu_read_unlock();
2314 }
2315
2316 static void sock_def_write_space(struct sock *sk)
2317 {
2318 struct socket_wq *wq;
2319
2320 rcu_read_lock();
2321
2322 /* Do not wake up a writer until he can make "significant"
2323 * progress. --DaveM
2324 */
2325 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2326 wq = rcu_dereference(sk->sk_wq);
2327 if (wq_has_sleeper(wq))
2328 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2329 POLLWRNORM | POLLWRBAND);
2330
2331 /* Should agree with poll, otherwise some programs break */
2332 if (sock_writeable(sk))
2333 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2334 }
2335
2336 rcu_read_unlock();
2337 }
2338
2339 static void sock_def_destruct(struct sock *sk)
2340 {
2341 }
2342
2343 void sk_send_sigurg(struct sock *sk)
2344 {
2345 if (sk->sk_socket && sk->sk_socket->file)
2346 if (send_sigurg(&sk->sk_socket->file->f_owner))
2347 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2348 }
2349 EXPORT_SYMBOL(sk_send_sigurg);
2350
2351 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2352 unsigned long expires)
2353 {
2354 if (!mod_timer(timer, expires))
2355 sock_hold(sk);
2356 }
2357 EXPORT_SYMBOL(sk_reset_timer);
2358
2359 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2360 {
2361 if (del_timer(timer))
2362 __sock_put(sk);
2363 }
2364 EXPORT_SYMBOL(sk_stop_timer);
2365
2366 void sock_init_data(struct socket *sock, struct sock *sk)
2367 {
2368 skb_queue_head_init(&sk->sk_receive_queue);
2369 skb_queue_head_init(&sk->sk_write_queue);
2370 skb_queue_head_init(&sk->sk_error_queue);
2371
2372 sk->sk_send_head = NULL;
2373
2374 init_timer(&sk->sk_timer);
2375
2376 sk->sk_allocation = GFP_KERNEL;
2377 sk->sk_rcvbuf = sysctl_rmem_default;
2378 sk->sk_sndbuf = sysctl_wmem_default;
2379 sk->sk_state = TCP_CLOSE;
2380 sk_set_socket(sk, sock);
2381
2382 sock_set_flag(sk, SOCK_ZAPPED);
2383
2384 if (sock) {
2385 sk->sk_type = sock->type;
2386 sk->sk_wq = sock->wq;
2387 sock->sk = sk;
2388 } else
2389 sk->sk_wq = NULL;
2390
2391 spin_lock_init(&sk->sk_dst_lock);
2392 rwlock_init(&sk->sk_callback_lock);
2393 lockdep_set_class_and_name(&sk->sk_callback_lock,
2394 af_callback_keys + sk->sk_family,
2395 af_family_clock_key_strings[sk->sk_family]);
2396
2397 sk->sk_state_change = sock_def_wakeup;
2398 sk->sk_data_ready = sock_def_readable;
2399 sk->sk_write_space = sock_def_write_space;
2400 sk->sk_error_report = sock_def_error_report;
2401 sk->sk_destruct = sock_def_destruct;
2402
2403 sk->sk_frag.page = NULL;
2404 sk->sk_frag.offset = 0;
2405 sk->sk_peek_off = -1;
2406
2407 sk->sk_peer_pid = NULL;
2408 sk->sk_peer_cred = NULL;
2409 sk->sk_write_pending = 0;
2410 sk->sk_rcvlowat = 1;
2411 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2412 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2413
2414 sk->sk_stamp = ktime_set(-1L, 0);
2415
2416 #ifdef CONFIG_NET_RX_BUSY_POLL
2417 sk->sk_napi_id = 0;
2418 sk->sk_ll_usec = sysctl_net_busy_read;
2419 #endif
2420
2421 sk->sk_max_pacing_rate = ~0U;
2422 sk->sk_pacing_rate = ~0U;
2423 sk->sk_incoming_cpu = -1;
2424 /*
2425 * Before updating sk_refcnt, we must commit prior changes to memory
2426 * (Documentation/RCU/rculist_nulls.txt for details)
2427 */
2428 smp_wmb();
2429 atomic_set(&sk->sk_refcnt, 1);
2430 atomic_set(&sk->sk_drops, 0);
2431 }
2432 EXPORT_SYMBOL(sock_init_data);
2433
2434 void lock_sock_nested(struct sock *sk, int subclass)
2435 {
2436 might_sleep();
2437 spin_lock_bh(&sk->sk_lock.slock);
2438 if (sk->sk_lock.owned)
2439 __lock_sock(sk);
2440 sk->sk_lock.owned = 1;
2441 spin_unlock(&sk->sk_lock.slock);
2442 /*
2443 * The sk_lock has mutex_lock() semantics here:
2444 */
2445 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2446 local_bh_enable();
2447 }
2448 EXPORT_SYMBOL(lock_sock_nested);
2449
2450 void release_sock(struct sock *sk)
2451 {
2452 /*
2453 * The sk_lock has mutex_unlock() semantics:
2454 */
2455 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2456
2457 spin_lock_bh(&sk->sk_lock.slock);
2458 if (sk->sk_backlog.tail)
2459 __release_sock(sk);
2460
2461 /* Warning : release_cb() might need to release sk ownership,
2462 * ie call sock_release_ownership(sk) before us.
2463 */
2464 if (sk->sk_prot->release_cb)
2465 sk->sk_prot->release_cb(sk);
2466
2467 sock_release_ownership(sk);
2468 if (waitqueue_active(&sk->sk_lock.wq))
2469 wake_up(&sk->sk_lock.wq);
2470 spin_unlock_bh(&sk->sk_lock.slock);
2471 }
2472 EXPORT_SYMBOL(release_sock);
2473
2474 /**
2475 * lock_sock_fast - fast version of lock_sock
2476 * @sk: socket
2477 *
2478 * This version should be used for very small section, where process wont block
2479 * return false if fast path is taken
2480 * sk_lock.slock locked, owned = 0, BH disabled
2481 * return true if slow path is taken
2482 * sk_lock.slock unlocked, owned = 1, BH enabled
2483 */
2484 bool lock_sock_fast(struct sock *sk)
2485 {
2486 might_sleep();
2487 spin_lock_bh(&sk->sk_lock.slock);
2488
2489 if (!sk->sk_lock.owned)
2490 /*
2491 * Note : We must disable BH
2492 */
2493 return false;
2494
2495 __lock_sock(sk);
2496 sk->sk_lock.owned = 1;
2497 spin_unlock(&sk->sk_lock.slock);
2498 /*
2499 * The sk_lock has mutex_lock() semantics here:
2500 */
2501 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2502 local_bh_enable();
2503 return true;
2504 }
2505 EXPORT_SYMBOL(lock_sock_fast);
2506
2507 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2508 {
2509 struct timeval tv;
2510 if (!sock_flag(sk, SOCK_TIMESTAMP))
2511 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2512 tv = ktime_to_timeval(sk->sk_stamp);
2513 if (tv.tv_sec == -1)
2514 return -ENOENT;
2515 if (tv.tv_sec == 0) {
2516 sk->sk_stamp = ktime_get_real();
2517 tv = ktime_to_timeval(sk->sk_stamp);
2518 }
2519 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2520 }
2521 EXPORT_SYMBOL(sock_get_timestamp);
2522
2523 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2524 {
2525 struct timespec ts;
2526 if (!sock_flag(sk, SOCK_TIMESTAMP))
2527 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2528 ts = ktime_to_timespec(sk->sk_stamp);
2529 if (ts.tv_sec == -1)
2530 return -ENOENT;
2531 if (ts.tv_sec == 0) {
2532 sk->sk_stamp = ktime_get_real();
2533 ts = ktime_to_timespec(sk->sk_stamp);
2534 }
2535 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2536 }
2537 EXPORT_SYMBOL(sock_get_timestampns);
2538
2539 void sock_enable_timestamp(struct sock *sk, int flag)
2540 {
2541 if (!sock_flag(sk, flag)) {
2542 unsigned long previous_flags = sk->sk_flags;
2543
2544 sock_set_flag(sk, flag);
2545 /*
2546 * we just set one of the two flags which require net
2547 * time stamping, but time stamping might have been on
2548 * already because of the other one
2549 */
2550 if (sock_needs_netstamp(sk) &&
2551 !(previous_flags & SK_FLAGS_TIMESTAMP))
2552 net_enable_timestamp();
2553 }
2554 }
2555
2556 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2557 int level, int type)
2558 {
2559 struct sock_exterr_skb *serr;
2560 struct sk_buff *skb;
2561 int copied, err;
2562
2563 err = -EAGAIN;
2564 skb = sock_dequeue_err_skb(sk);
2565 if (skb == NULL)
2566 goto out;
2567
2568 copied = skb->len;
2569 if (copied > len) {
2570 msg->msg_flags |= MSG_TRUNC;
2571 copied = len;
2572 }
2573 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2574 if (err)
2575 goto out_free_skb;
2576
2577 sock_recv_timestamp(msg, sk, skb);
2578
2579 serr = SKB_EXT_ERR(skb);
2580 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2581
2582 msg->msg_flags |= MSG_ERRQUEUE;
2583 err = copied;
2584
2585 out_free_skb:
2586 kfree_skb(skb);
2587 out:
2588 return err;
2589 }
2590 EXPORT_SYMBOL(sock_recv_errqueue);
2591
2592 /*
2593 * Get a socket option on an socket.
2594 *
2595 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2596 * asynchronous errors should be reported by getsockopt. We assume
2597 * this means if you specify SO_ERROR (otherwise whats the point of it).
2598 */
2599 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2600 char __user *optval, int __user *optlen)
2601 {
2602 struct sock *sk = sock->sk;
2603
2604 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2605 }
2606 EXPORT_SYMBOL(sock_common_getsockopt);
2607
2608 #ifdef CONFIG_COMPAT
2609 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2610 char __user *optval, int __user *optlen)
2611 {
2612 struct sock *sk = sock->sk;
2613
2614 if (sk->sk_prot->compat_getsockopt != NULL)
2615 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2616 optval, optlen);
2617 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2618 }
2619 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2620 #endif
2621
2622 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2623 int flags)
2624 {
2625 struct sock *sk = sock->sk;
2626 int addr_len = 0;
2627 int err;
2628
2629 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2630 flags & ~MSG_DONTWAIT, &addr_len);
2631 if (err >= 0)
2632 msg->msg_namelen = addr_len;
2633 return err;
2634 }
2635 EXPORT_SYMBOL(sock_common_recvmsg);
2636
2637 /*
2638 * Set socket options on an inet socket.
2639 */
2640 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2641 char __user *optval, unsigned int optlen)
2642 {
2643 struct sock *sk = sock->sk;
2644
2645 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2646 }
2647 EXPORT_SYMBOL(sock_common_setsockopt);
2648
2649 #ifdef CONFIG_COMPAT
2650 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2651 char __user *optval, unsigned int optlen)
2652 {
2653 struct sock *sk = sock->sk;
2654
2655 if (sk->sk_prot->compat_setsockopt != NULL)
2656 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2657 optval, optlen);
2658 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2659 }
2660 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2661 #endif
2662
2663 void sk_common_release(struct sock *sk)
2664 {
2665 if (sk->sk_prot->destroy)
2666 sk->sk_prot->destroy(sk);
2667
2668 /*
2669 * Observation: when sock_common_release is called, processes have
2670 * no access to socket. But net still has.
2671 * Step one, detach it from networking:
2672 *
2673 * A. Remove from hash tables.
2674 */
2675
2676 sk->sk_prot->unhash(sk);
2677
2678 /*
2679 * In this point socket cannot receive new packets, but it is possible
2680 * that some packets are in flight because some CPU runs receiver and
2681 * did hash table lookup before we unhashed socket. They will achieve
2682 * receive queue and will be purged by socket destructor.
2683 *
2684 * Also we still have packets pending on receive queue and probably,
2685 * our own packets waiting in device queues. sock_destroy will drain
2686 * receive queue, but transmitted packets will delay socket destruction
2687 * until the last reference will be released.
2688 */
2689
2690 sock_orphan(sk);
2691
2692 xfrm_sk_free_policy(sk);
2693
2694 sk_refcnt_debug_release(sk);
2695
2696 if (sk->sk_frag.page) {
2697 put_page(sk->sk_frag.page);
2698 sk->sk_frag.page = NULL;
2699 }
2700
2701 sock_put(sk);
2702 }
2703 EXPORT_SYMBOL(sk_common_release);
2704
2705 #ifdef CONFIG_PROC_FS
2706 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2707 struct prot_inuse {
2708 int val[PROTO_INUSE_NR];
2709 };
2710
2711 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2712
2713 #ifdef CONFIG_NET_NS
2714 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2715 {
2716 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2717 }
2718 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2719
2720 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2721 {
2722 int cpu, idx = prot->inuse_idx;
2723 int res = 0;
2724
2725 for_each_possible_cpu(cpu)
2726 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2727
2728 return res >= 0 ? res : 0;
2729 }
2730 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2731
2732 static int __net_init sock_inuse_init_net(struct net *net)
2733 {
2734 net->core.inuse = alloc_percpu(struct prot_inuse);
2735 return net->core.inuse ? 0 : -ENOMEM;
2736 }
2737
2738 static void __net_exit sock_inuse_exit_net(struct net *net)
2739 {
2740 free_percpu(net->core.inuse);
2741 }
2742
2743 static struct pernet_operations net_inuse_ops = {
2744 .init = sock_inuse_init_net,
2745 .exit = sock_inuse_exit_net,
2746 };
2747
2748 static __init int net_inuse_init(void)
2749 {
2750 if (register_pernet_subsys(&net_inuse_ops))
2751 panic("Cannot initialize net inuse counters");
2752
2753 return 0;
2754 }
2755
2756 core_initcall(net_inuse_init);
2757 #else
2758 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2759
2760 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2761 {
2762 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2763 }
2764 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2765
2766 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2767 {
2768 int cpu, idx = prot->inuse_idx;
2769 int res = 0;
2770
2771 for_each_possible_cpu(cpu)
2772 res += per_cpu(prot_inuse, cpu).val[idx];
2773
2774 return res >= 0 ? res : 0;
2775 }
2776 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2777 #endif
2778
2779 static void assign_proto_idx(struct proto *prot)
2780 {
2781 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2782
2783 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2784 pr_err("PROTO_INUSE_NR exhausted\n");
2785 return;
2786 }
2787
2788 set_bit(prot->inuse_idx, proto_inuse_idx);
2789 }
2790
2791 static void release_proto_idx(struct proto *prot)
2792 {
2793 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2794 clear_bit(prot->inuse_idx, proto_inuse_idx);
2795 }
2796 #else
2797 static inline void assign_proto_idx(struct proto *prot)
2798 {
2799 }
2800
2801 static inline void release_proto_idx(struct proto *prot)
2802 {
2803 }
2804 #endif
2805
2806 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2807 {
2808 if (!rsk_prot)
2809 return;
2810 kfree(rsk_prot->slab_name);
2811 rsk_prot->slab_name = NULL;
2812 kmem_cache_destroy(rsk_prot->slab);
2813 rsk_prot->slab = NULL;
2814 }
2815
2816 static int req_prot_init(const struct proto *prot)
2817 {
2818 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2819
2820 if (!rsk_prot)
2821 return 0;
2822
2823 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2824 prot->name);
2825 if (!rsk_prot->slab_name)
2826 return -ENOMEM;
2827
2828 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2829 rsk_prot->obj_size, 0,
2830 prot->slab_flags, NULL);
2831
2832 if (!rsk_prot->slab) {
2833 pr_crit("%s: Can't create request sock SLAB cache!\n",
2834 prot->name);
2835 return -ENOMEM;
2836 }
2837 return 0;
2838 }
2839
2840 int proto_register(struct proto *prot, int alloc_slab)
2841 {
2842 if (alloc_slab) {
2843 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2844 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2845 NULL);
2846
2847 if (prot->slab == NULL) {
2848 pr_crit("%s: Can't create sock SLAB cache!\n",
2849 prot->name);
2850 goto out;
2851 }
2852
2853 if (req_prot_init(prot))
2854 goto out_free_request_sock_slab;
2855
2856 if (prot->twsk_prot != NULL) {
2857 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2858
2859 if (prot->twsk_prot->twsk_slab_name == NULL)
2860 goto out_free_request_sock_slab;
2861
2862 prot->twsk_prot->twsk_slab =
2863 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2864 prot->twsk_prot->twsk_obj_size,
2865 0,
2866 prot->slab_flags,
2867 NULL);
2868 if (prot->twsk_prot->twsk_slab == NULL)
2869 goto out_free_timewait_sock_slab_name;
2870 }
2871 }
2872
2873 mutex_lock(&proto_list_mutex);
2874 list_add(&prot->node, &proto_list);
2875 assign_proto_idx(prot);
2876 mutex_unlock(&proto_list_mutex);
2877 return 0;
2878
2879 out_free_timewait_sock_slab_name:
2880 kfree(prot->twsk_prot->twsk_slab_name);
2881 out_free_request_sock_slab:
2882 req_prot_cleanup(prot->rsk_prot);
2883
2884 kmem_cache_destroy(prot->slab);
2885 prot->slab = NULL;
2886 out:
2887 return -ENOBUFS;
2888 }
2889 EXPORT_SYMBOL(proto_register);
2890
2891 void proto_unregister(struct proto *prot)
2892 {
2893 mutex_lock(&proto_list_mutex);
2894 release_proto_idx(prot);
2895 list_del(&prot->node);
2896 mutex_unlock(&proto_list_mutex);
2897
2898 kmem_cache_destroy(prot->slab);
2899 prot->slab = NULL;
2900
2901 req_prot_cleanup(prot->rsk_prot);
2902
2903 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2904 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2905 kfree(prot->twsk_prot->twsk_slab_name);
2906 prot->twsk_prot->twsk_slab = NULL;
2907 }
2908 }
2909 EXPORT_SYMBOL(proto_unregister);
2910
2911 #ifdef CONFIG_PROC_FS
2912 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2913 __acquires(proto_list_mutex)
2914 {
2915 mutex_lock(&proto_list_mutex);
2916 return seq_list_start_head(&proto_list, *pos);
2917 }
2918
2919 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2920 {
2921 return seq_list_next(v, &proto_list, pos);
2922 }
2923
2924 static void proto_seq_stop(struct seq_file *seq, void *v)
2925 __releases(proto_list_mutex)
2926 {
2927 mutex_unlock(&proto_list_mutex);
2928 }
2929
2930 static char proto_method_implemented(const void *method)
2931 {
2932 return method == NULL ? 'n' : 'y';
2933 }
2934 static long sock_prot_memory_allocated(struct proto *proto)
2935 {
2936 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2937 }
2938
2939 static char *sock_prot_memory_pressure(struct proto *proto)
2940 {
2941 return proto->memory_pressure != NULL ?
2942 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2943 }
2944
2945 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2946 {
2947
2948 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2949 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2950 proto->name,
2951 proto->obj_size,
2952 sock_prot_inuse_get(seq_file_net(seq), proto),
2953 sock_prot_memory_allocated(proto),
2954 sock_prot_memory_pressure(proto),
2955 proto->max_header,
2956 proto->slab == NULL ? "no" : "yes",
2957 module_name(proto->owner),
2958 proto_method_implemented(proto->close),
2959 proto_method_implemented(proto->connect),
2960 proto_method_implemented(proto->disconnect),
2961 proto_method_implemented(proto->accept),
2962 proto_method_implemented(proto->ioctl),
2963 proto_method_implemented(proto->init),
2964 proto_method_implemented(proto->destroy),
2965 proto_method_implemented(proto->shutdown),
2966 proto_method_implemented(proto->setsockopt),
2967 proto_method_implemented(proto->getsockopt),
2968 proto_method_implemented(proto->sendmsg),
2969 proto_method_implemented(proto->recvmsg),
2970 proto_method_implemented(proto->sendpage),
2971 proto_method_implemented(proto->bind),
2972 proto_method_implemented(proto->backlog_rcv),
2973 proto_method_implemented(proto->hash),
2974 proto_method_implemented(proto->unhash),
2975 proto_method_implemented(proto->get_port),
2976 proto_method_implemented(proto->enter_memory_pressure));
2977 }
2978
2979 static int proto_seq_show(struct seq_file *seq, void *v)
2980 {
2981 if (v == &proto_list)
2982 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2983 "protocol",
2984 "size",
2985 "sockets",
2986 "memory",
2987 "press",
2988 "maxhdr",
2989 "slab",
2990 "module",
2991 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2992 else
2993 proto_seq_printf(seq, list_entry(v, struct proto, node));
2994 return 0;
2995 }
2996
2997 static const struct seq_operations proto_seq_ops = {
2998 .start = proto_seq_start,
2999 .next = proto_seq_next,
3000 .stop = proto_seq_stop,
3001 .show = proto_seq_show,
3002 };
3003
3004 static int proto_seq_open(struct inode *inode, struct file *file)
3005 {
3006 return seq_open_net(inode, file, &proto_seq_ops,
3007 sizeof(struct seq_net_private));
3008 }
3009
3010 static const struct file_operations proto_seq_fops = {
3011 .owner = THIS_MODULE,
3012 .open = proto_seq_open,
3013 .read = seq_read,
3014 .llseek = seq_lseek,
3015 .release = seq_release_net,
3016 };
3017
3018 static __net_init int proto_init_net(struct net *net)
3019 {
3020 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3021 return -ENOMEM;
3022
3023 return 0;
3024 }
3025
3026 static __net_exit void proto_exit_net(struct net *net)
3027 {
3028 remove_proc_entry("protocols", net->proc_net);
3029 }
3030
3031
3032 static __net_initdata struct pernet_operations proto_net_ops = {
3033 .init = proto_init_net,
3034 .exit = proto_exit_net,
3035 };
3036
3037 static int __init proto_init(void)
3038 {
3039 return register_pernet_subsys(&proto_net_ops);
3040 }
3041
3042 subsys_initcall(proto_init);
3043
3044 #endif /* PROC_FS */
Michael's Linux tree.