?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-3.11' of git://linux-nfs.org/~bfields/linux
[linux-2.6.git] / net / core / sock.c
blob548d716c5f62ac8c9d90d0959d19022714f9585d
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/types.h>
97 #include <linux/socket.h>
98 #include <linux/in.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
119
120 #include <asm/uaccess.h>
121
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
133
134 #include <linux/filter.h>
135
136 #include <trace/events/sock.h>
137
138 #ifdef CONFIG_INET
139 #include <net/tcp.h>
140 #endif
141
142 #include <net/busy_poll.h>
143
144 static DEFINE_MUTEX(proto_list_mutex);
145 static LIST_HEAD(proto_list);
146
147 #ifdef CONFIG_MEMCG_KMEM
148 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
149 {
150 struct proto *proto;
151 int ret = 0;
152
153 mutex_lock(&proto_list_mutex);
154 list_for_each_entry(proto, &proto_list, node) {
155 if (proto->init_cgroup) {
156 ret = proto->init_cgroup(memcg, ss);
157 if (ret)
158 goto out;
159 }
160 }
161
162 mutex_unlock(&proto_list_mutex);
163 return ret;
164 out:
165 list_for_each_entry_continue_reverse(proto, &proto_list, node)
166 if (proto->destroy_cgroup)
167 proto->destroy_cgroup(memcg);
168 mutex_unlock(&proto_list_mutex);
169 return ret;
170 }
171
172 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
173 {
174 struct proto *proto;
175
176 mutex_lock(&proto_list_mutex);
177 list_for_each_entry_reverse(proto, &proto_list, node)
178 if (proto->destroy_cgroup)
179 proto->destroy_cgroup(memcg);
180 mutex_unlock(&proto_list_mutex);
181 }
182 #endif
183
184 /*
185 * Each address family might have different locking rules, so we have
186 * one slock key per address family:
187 */
188 static struct lock_class_key af_family_keys[AF_MAX];
189 static struct lock_class_key af_family_slock_keys[AF_MAX];
190
191 #if defined(CONFIG_MEMCG_KMEM)
192 struct static_key memcg_socket_limit_enabled;
193 EXPORT_SYMBOL(memcg_socket_limit_enabled);
194 #endif
195
196 /*
197 * Make lock validator output more readable. (we pre-construct these
198 * strings build-time, so that runtime initialization of socket
199 * locks is fast):
200 */
201 static const char *const af_family_key_strings[AF_MAX+1] = {
202 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
203 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
204 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
205 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
206 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
207 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
208 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
209 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
210 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
211 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
212 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
213 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
214 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
215 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
216 };
217 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
218 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
219 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
220 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
221 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
222 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
223 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
224 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
225 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
226 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
227 "slock-27" , "slock-28" , "slock-AF_CAN" ,
228 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
229 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
230 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
231 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
232 };
233 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
234 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
235 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
236 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
237 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
238 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
239 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
240 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
241 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
242 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
243 "clock-27" , "clock-28" , "clock-AF_CAN" ,
244 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
245 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
246 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
247 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
248 };
249
250 /*
251 * sk_callback_lock locking rules are per-address-family,
252 * so split the lock classes by using a per-AF key:
253 */
254 static struct lock_class_key af_callback_keys[AF_MAX];
255
256 /* Take into consideration the size of the struct sk_buff overhead in the
257 * determination of these values, since that is non-constant across
258 * platforms. This makes socket queueing behavior and performance
259 * not depend upon such differences.
260 */
261 #define _SK_MEM_PACKETS 256
262 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
263 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
264 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
265
266 /* Run time adjustable parameters. */
267 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
268 EXPORT_SYMBOL(sysctl_wmem_max);
269 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
270 EXPORT_SYMBOL(sysctl_rmem_max);
271 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
272 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
273
274 /* Maximal space eaten by iovec or ancillary data plus some space */
275 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
276 EXPORT_SYMBOL(sysctl_optmem_max);
277
278 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
279 EXPORT_SYMBOL_GPL(memalloc_socks);
280
281 /**
282 * sk_set_memalloc - sets %SOCK_MEMALLOC
283 * @sk: socket to set it on
284 *
285 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
286 * It's the responsibility of the admin to adjust min_free_kbytes
287 * to meet the requirements
288 */
289 void sk_set_memalloc(struct sock *sk)
290 {
291 sock_set_flag(sk, SOCK_MEMALLOC);
292 sk->sk_allocation |= __GFP_MEMALLOC;
293 static_key_slow_inc(&memalloc_socks);
294 }
295 EXPORT_SYMBOL_GPL(sk_set_memalloc);
296
297 void sk_clear_memalloc(struct sock *sk)
298 {
299 sock_reset_flag(sk, SOCK_MEMALLOC);
300 sk->sk_allocation &= ~__GFP_MEMALLOC;
301 static_key_slow_dec(&memalloc_socks);
302
303 /*
304 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
305 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
306 * it has rmem allocations there is a risk that the user of the
307 * socket cannot make forward progress due to exceeding the rmem
308 * limits. By rights, sk_clear_memalloc() should only be called
309 * on sockets being torn down but warn and reset the accounting if
310 * that assumption breaks.
311 */
312 if (WARN_ON(sk->sk_forward_alloc))
313 sk_mem_reclaim(sk);
314 }
315 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
316
317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
318 {
319 int ret;
320 unsigned long pflags = current->flags;
321
322 /* these should have been dropped before queueing */
323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
324
325 current->flags |= PF_MEMALLOC;
326 ret = sk->sk_backlog_rcv(sk, skb);
327 tsk_restore_flags(current, pflags, PF_MEMALLOC);
328
329 return ret;
330 }
331 EXPORT_SYMBOL(__sk_backlog_rcv);
332
333 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
334 {
335 struct timeval tv;
336
337 if (optlen < sizeof(tv))
338 return -EINVAL;
339 if (copy_from_user(&tv, optval, sizeof(tv)))
340 return -EFAULT;
341 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
342 return -EDOM;
343
344 if (tv.tv_sec < 0) {
345 static int warned __read_mostly;
346
347 *timeo_p = 0;
348 if (warned < 10 && net_ratelimit()) {
349 warned++;
350 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
351 __func__, current->comm, task_pid_nr(current));
352 }
353 return 0;
354 }
355 *timeo_p = MAX_SCHEDULE_TIMEOUT;
356 if (tv.tv_sec == 0 && tv.tv_usec == 0)
357 return 0;
358 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
359 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
360 return 0;
361 }
362
363 static void sock_warn_obsolete_bsdism(const char *name)
364 {
365 static int warned;
366 static char warncomm[TASK_COMM_LEN];
367 if (strcmp(warncomm, current->comm) && warned < 5) {
368 strcpy(warncomm, current->comm);
369 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
370 warncomm, name);
371 warned++;
372 }
373 }
374
375 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
376
377 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
378 {
379 if (sk->sk_flags & flags) {
380 sk->sk_flags &= ~flags;
381 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
382 net_disable_timestamp();
383 }
384 }
385
386
387 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
388 {
389 int err;
390 int skb_len;
391 unsigned long flags;
392 struct sk_buff_head *list = &sk->sk_receive_queue;
393
394 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
395 atomic_inc(&sk->sk_drops);
396 trace_sock_rcvqueue_full(sk, skb);
397 return -ENOMEM;
398 }
399
400 err = sk_filter(sk, skb);
401 if (err)
402 return err;
403
404 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
405 atomic_inc(&sk->sk_drops);
406 return -ENOBUFS;
407 }
408
409 skb->dev = NULL;
410 skb_set_owner_r(skb, sk);
411
412 /* Cache the SKB length before we tack it onto the receive
413 * queue. Once it is added it no longer belongs to us and
414 * may be freed by other threads of control pulling packets
415 * from the queue.
416 */
417 skb_len = skb->len;
418
419 /* we escape from rcu protected region, make sure we dont leak
420 * a norefcounted dst
421 */
422 skb_dst_force(skb);
423
424 spin_lock_irqsave(&list->lock, flags);
425 skb->dropcount = atomic_read(&sk->sk_drops);
426 __skb_queue_tail(list, skb);
427 spin_unlock_irqrestore(&list->lock, flags);
428
429 if (!sock_flag(sk, SOCK_DEAD))
430 sk->sk_data_ready(sk, skb_len);
431 return 0;
432 }
433 EXPORT_SYMBOL(sock_queue_rcv_skb);
434
435 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
436 {
437 int rc = NET_RX_SUCCESS;
438
439 if (sk_filter(sk, skb))
440 goto discard_and_relse;
441
442 skb->dev = NULL;
443
444 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
445 atomic_inc(&sk->sk_drops);
446 goto discard_and_relse;
447 }
448 if (nested)
449 bh_lock_sock_nested(sk);
450 else
451 bh_lock_sock(sk);
452 if (!sock_owned_by_user(sk)) {
453 /*
454 * trylock + unlock semantics:
455 */
456 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
457
458 rc = sk_backlog_rcv(sk, skb);
459
460 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
461 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
462 bh_unlock_sock(sk);
463 atomic_inc(&sk->sk_drops);
464 goto discard_and_relse;
465 }
466
467 bh_unlock_sock(sk);
468 out:
469 sock_put(sk);
470 return rc;
471 discard_and_relse:
472 kfree_skb(skb);
473 goto out;
474 }
475 EXPORT_SYMBOL(sk_receive_skb);
476
477 void sk_reset_txq(struct sock *sk)
478 {
479 sk_tx_queue_clear(sk);
480 }
481 EXPORT_SYMBOL(sk_reset_txq);
482
483 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
484 {
485 struct dst_entry *dst = __sk_dst_get(sk);
486
487 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
488 sk_tx_queue_clear(sk);
489 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
490 dst_release(dst);
491 return NULL;
492 }
493
494 return dst;
495 }
496 EXPORT_SYMBOL(__sk_dst_check);
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_dst_reset(sk);
504 dst_release(dst);
505 return NULL;
506 }
507
508 return dst;
509 }
510 EXPORT_SYMBOL(sk_dst_check);
511
512 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
513 int optlen)
514 {
515 int ret = -ENOPROTOOPT;
516 #ifdef CONFIG_NETDEVICES
517 struct net *net = sock_net(sk);
518 char devname[IFNAMSIZ];
519 int index;
520
521 /* Sorry... */
522 ret = -EPERM;
523 if (!ns_capable(net->user_ns, CAP_NET_RAW))
524 goto out;
525
526 ret = -EINVAL;
527 if (optlen < 0)
528 goto out;
529
530 /* Bind this socket to a particular device like "eth0",
531 * as specified in the passed interface name. If the
532 * name is "" or the option length is zero the socket
533 * is not bound.
534 */
535 if (optlen > IFNAMSIZ - 1)
536 optlen = IFNAMSIZ - 1;
537 memset(devname, 0, sizeof(devname));
538
539 ret = -EFAULT;
540 if (copy_from_user(devname, optval, optlen))
541 goto out;
542
543 index = 0;
544 if (devname[0] != '\0') {
545 struct net_device *dev;
546
547 rcu_read_lock();
548 dev = dev_get_by_name_rcu(net, devname);
549 if (dev)
550 index = dev->ifindex;
551 rcu_read_unlock();
552 ret = -ENODEV;
553 if (!dev)
554 goto out;
555 }
556
557 lock_sock(sk);
558 sk->sk_bound_dev_if = index;
559 sk_dst_reset(sk);
560 release_sock(sk);
561
562 ret = 0;
563
564 out:
565 #endif
566
567 return ret;
568 }
569
570 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
571 int __user *optlen, int len)
572 {
573 int ret = -ENOPROTOOPT;
574 #ifdef CONFIG_NETDEVICES
575 struct net *net = sock_net(sk);
576 char devname[IFNAMSIZ];
577
578 if (sk->sk_bound_dev_if == 0) {
579 len = 0;
580 goto zero;
581 }
582
583 ret = -EINVAL;
584 if (len < IFNAMSIZ)
585 goto out;
586
587 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
588 if (ret)
589 goto out;
590
591 len = strlen(devname) + 1;
592
593 ret = -EFAULT;
594 if (copy_to_user(optval, devname, len))
595 goto out;
596
597 zero:
598 ret = -EFAULT;
599 if (put_user(len, optlen))
600 goto out;
601
602 ret = 0;
603
604 out:
605 #endif
606
607 return ret;
608 }
609
610 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
611 {
612 if (valbool)
613 sock_set_flag(sk, bit);
614 else
615 sock_reset_flag(sk, bit);
616 }
617
618 /*
619 * This is meant for all protocols to use and covers goings on
620 * at the socket level. Everything here is generic.
621 */
622
623 int sock_setsockopt(struct socket *sock, int level, int optname,
624 char __user *optval, unsigned int optlen)
625 {
626 struct sock *sk = sock->sk;
627 int val;
628 int valbool;
629 struct linger ling;
630 int ret = 0;
631
632 /*
633 * Options without arguments
634 */
635
636 if (optname == SO_BINDTODEVICE)
637 return sock_setbindtodevice(sk, optval, optlen);
638
639 if (optlen < sizeof(int))
640 return -EINVAL;
641
642 if (get_user(val, (int __user *)optval))
643 return -EFAULT;
644
645 valbool = val ? 1 : 0;
646
647 lock_sock(sk);
648
649 switch (optname) {
650 case SO_DEBUG:
651 if (val && !capable(CAP_NET_ADMIN))
652 ret = -EACCES;
653 else
654 sock_valbool_flag(sk, SOCK_DBG, valbool);
655 break;
656 case SO_REUSEADDR:
657 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
658 break;
659 case SO_REUSEPORT:
660 sk->sk_reuseport = valbool;
661 break;
662 case SO_TYPE:
663 case SO_PROTOCOL:
664 case SO_DOMAIN:
665 case SO_ERROR:
666 ret = -ENOPROTOOPT;
667 break;
668 case SO_DONTROUTE:
669 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
670 break;
671 case SO_BROADCAST:
672 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
673 break;
674 case SO_SNDBUF:
675 /* Don't error on this BSD doesn't and if you think
676 * about it this is right. Otherwise apps have to
677 * play 'guess the biggest size' games. RCVBUF/SNDBUF
678 * are treated in BSD as hints
679 */
680 val = min_t(u32, val, sysctl_wmem_max);
681 set_sndbuf:
682 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
683 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
684 /* Wake up sending tasks if we upped the value. */
685 sk->sk_write_space(sk);
686 break;
687
688 case SO_SNDBUFFORCE:
689 if (!capable(CAP_NET_ADMIN)) {
690 ret = -EPERM;
691 break;
692 }
693 goto set_sndbuf;
694
695 case SO_RCVBUF:
696 /* Don't error on this BSD doesn't and if you think
697 * about it this is right. Otherwise apps have to
698 * play 'guess the biggest size' games. RCVBUF/SNDBUF
699 * are treated in BSD as hints
700 */
701 val = min_t(u32, val, sysctl_rmem_max);
702 set_rcvbuf:
703 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
704 /*
705 * We double it on the way in to account for
706 * "struct sk_buff" etc. overhead. Applications
707 * assume that the SO_RCVBUF setting they make will
708 * allow that much actual data to be received on that
709 * socket.
710 *
711 * Applications are unaware that "struct sk_buff" and
712 * other overheads allocate from the receive buffer
713 * during socket buffer allocation.
714 *
715 * And after considering the possible alternatives,
716 * returning the value we actually used in getsockopt
717 * is the most desirable behavior.
718 */
719 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
720 break;
721
722 case SO_RCVBUFFORCE:
723 if (!capable(CAP_NET_ADMIN)) {
724 ret = -EPERM;
725 break;
726 }
727 goto set_rcvbuf;
728
729 case SO_KEEPALIVE:
730 #ifdef CONFIG_INET
731 if (sk->sk_protocol == IPPROTO_TCP &&
732 sk->sk_type == SOCK_STREAM)
733 tcp_set_keepalive(sk, valbool);
734 #endif
735 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
736 break;
737
738 case SO_OOBINLINE:
739 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
740 break;
741
742 case SO_NO_CHECK:
743 sk->sk_no_check = valbool;
744 break;
745
746 case SO_PRIORITY:
747 if ((val >= 0 && val <= 6) ||
748 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
749 sk->sk_priority = val;
750 else
751 ret = -EPERM;
752 break;
753
754 case SO_LINGER:
755 if (optlen < sizeof(ling)) {
756 ret = -EINVAL; /* 1003.1g */
757 break;
758 }
759 if (copy_from_user(&ling, optval, sizeof(ling))) {
760 ret = -EFAULT;
761 break;
762 }
763 if (!ling.l_onoff)
764 sock_reset_flag(sk, SOCK_LINGER);
765 else {
766 #if (BITS_PER_LONG == 32)
767 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
768 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
769 else
770 #endif
771 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
772 sock_set_flag(sk, SOCK_LINGER);
773 }
774 break;
775
776 case SO_BSDCOMPAT:
777 sock_warn_obsolete_bsdism("setsockopt");
778 break;
779
780 case SO_PASSCRED:
781 if (valbool)
782 set_bit(SOCK_PASSCRED, &sock->flags);
783 else
784 clear_bit(SOCK_PASSCRED, &sock->flags);
785 break;
786
787 case SO_TIMESTAMP:
788 case SO_TIMESTAMPNS:
789 if (valbool) {
790 if (optname == SO_TIMESTAMP)
791 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
792 else
793 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
794 sock_set_flag(sk, SOCK_RCVTSTAMP);
795 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
796 } else {
797 sock_reset_flag(sk, SOCK_RCVTSTAMP);
798 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
799 }
800 break;
801
802 case SO_TIMESTAMPING:
803 if (val & ~SOF_TIMESTAMPING_MASK) {
804 ret = -EINVAL;
805 break;
806 }
807 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
808 val & SOF_TIMESTAMPING_TX_HARDWARE);
809 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
810 val & SOF_TIMESTAMPING_TX_SOFTWARE);
811 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
812 val & SOF_TIMESTAMPING_RX_HARDWARE);
813 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
814 sock_enable_timestamp(sk,
815 SOCK_TIMESTAMPING_RX_SOFTWARE);
816 else
817 sock_disable_timestamp(sk,
818 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
819 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
820 val & SOF_TIMESTAMPING_SOFTWARE);
821 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
822 val & SOF_TIMESTAMPING_SYS_HARDWARE);
823 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
824 val & SOF_TIMESTAMPING_RAW_HARDWARE);
825 break;
826
827 case SO_RCVLOWAT:
828 if (val < 0)
829 val = INT_MAX;
830 sk->sk_rcvlowat = val ? : 1;
831 break;
832
833 case SO_RCVTIMEO:
834 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
835 break;
836
837 case SO_SNDTIMEO:
838 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
839 break;
840
841 case SO_ATTACH_FILTER:
842 ret = -EINVAL;
843 if (optlen == sizeof(struct sock_fprog)) {
844 struct sock_fprog fprog;
845
846 ret = -EFAULT;
847 if (copy_from_user(&fprog, optval, sizeof(fprog)))
848 break;
849
850 ret = sk_attach_filter(&fprog, sk);
851 }
852 break;
853
854 case SO_DETACH_FILTER:
855 ret = sk_detach_filter(sk);
856 break;
857
858 case SO_LOCK_FILTER:
859 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
860 ret = -EPERM;
861 else
862 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
863 break;
864
865 case SO_PASSSEC:
866 if (valbool)
867 set_bit(SOCK_PASSSEC, &sock->flags);
868 else
869 clear_bit(SOCK_PASSSEC, &sock->flags);
870 break;
871 case SO_MARK:
872 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
873 ret = -EPERM;
874 else
875 sk->sk_mark = val;
876 break;
877
878 /* We implement the SO_SNDLOWAT etc to
879 not be settable (1003.1g 5.3) */
880 case SO_RXQ_OVFL:
881 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
882 break;
883
884 case SO_WIFI_STATUS:
885 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
886 break;
887
888 case SO_PEEK_OFF:
889 if (sock->ops->set_peek_off)
890 sock->ops->set_peek_off(sk, val);
891 else
892 ret = -EOPNOTSUPP;
893 break;
894
895 case SO_NOFCS:
896 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
897 break;
898
899 case SO_SELECT_ERR_QUEUE:
900 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
901 break;
902
903 #ifdef CONFIG_NET_LL_RX_POLL
904 case SO_BUSY_POLL:
905 /* allow unprivileged users to decrease the value */
906 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
907 ret = -EPERM;
908 else {
909 if (val < 0)
910 ret = -EINVAL;
911 else
912 sk->sk_ll_usec = val;
913 }
914 break;
915 #endif
916 default:
917 ret = -ENOPROTOOPT;
918 break;
919 }
920 release_sock(sk);
921 return ret;
922 }
923 EXPORT_SYMBOL(sock_setsockopt);
924
925
926 void cred_to_ucred(struct pid *pid, const struct cred *cred,
927 struct ucred *ucred)
928 {
929 ucred->pid = pid_vnr(pid);
930 ucred->uid = ucred->gid = -1;
931 if (cred) {
932 struct user_namespace *current_ns = current_user_ns();
933
934 ucred->uid = from_kuid_munged(current_ns, cred->euid);
935 ucred->gid = from_kgid_munged(current_ns, cred->egid);
936 }
937 }
938 EXPORT_SYMBOL_GPL(cred_to_ucred);
939
940 int sock_getsockopt(struct socket *sock, int level, int optname,
941 char __user *optval, int __user *optlen)
942 {
943 struct sock *sk = sock->sk;
944
945 union {
946 int val;
947 struct linger ling;
948 struct timeval tm;
949 } v;
950
951 int lv = sizeof(int);
952 int len;
953
954 if (get_user(len, optlen))
955 return -EFAULT;
956 if (len < 0)
957 return -EINVAL;
958
959 memset(&v, 0, sizeof(v));
960
961 switch (optname) {
962 case SO_DEBUG:
963 v.val = sock_flag(sk, SOCK_DBG);
964 break;
965
966 case SO_DONTROUTE:
967 v.val = sock_flag(sk, SOCK_LOCALROUTE);
968 break;
969
970 case SO_BROADCAST:
971 v.val = sock_flag(sk, SOCK_BROADCAST);
972 break;
973
974 case SO_SNDBUF:
975 v.val = sk->sk_sndbuf;
976 break;
977
978 case SO_RCVBUF:
979 v.val = sk->sk_rcvbuf;
980 break;
981
982 case SO_REUSEADDR:
983 v.val = sk->sk_reuse;
984 break;
985
986 case SO_REUSEPORT:
987 v.val = sk->sk_reuseport;
988 break;
989
990 case SO_KEEPALIVE:
991 v.val = sock_flag(sk, SOCK_KEEPOPEN);
992 break;
993
994 case SO_TYPE:
995 v.val = sk->sk_type;
996 break;
997
998 case SO_PROTOCOL:
999 v.val = sk->sk_protocol;
1000 break;
1001
1002 case SO_DOMAIN:
1003 v.val = sk->sk_family;
1004 break;
1005
1006 case SO_ERROR:
1007 v.val = -sock_error(sk);
1008 if (v.val == 0)
1009 v.val = xchg(&sk->sk_err_soft, 0);
1010 break;
1011
1012 case SO_OOBINLINE:
1013 v.val = sock_flag(sk, SOCK_URGINLINE);
1014 break;
1015
1016 case SO_NO_CHECK:
1017 v.val = sk->sk_no_check;
1018 break;
1019
1020 case SO_PRIORITY:
1021 v.val = sk->sk_priority;
1022 break;
1023
1024 case SO_LINGER:
1025 lv = sizeof(v.ling);
1026 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1027 v.ling.l_linger = sk->sk_lingertime / HZ;
1028 break;
1029
1030 case SO_BSDCOMPAT:
1031 sock_warn_obsolete_bsdism("getsockopt");
1032 break;
1033
1034 case SO_TIMESTAMP:
1035 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1036 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1037 break;
1038
1039 case SO_TIMESTAMPNS:
1040 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1041 break;
1042
1043 case SO_TIMESTAMPING:
1044 v.val = 0;
1045 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1046 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1047 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1048 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1049 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1050 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1051 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1052 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1053 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1054 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1055 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1056 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1057 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1058 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1059 break;
1060
1061 case SO_RCVTIMEO:
1062 lv = sizeof(struct timeval);
1063 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1064 v.tm.tv_sec = 0;
1065 v.tm.tv_usec = 0;
1066 } else {
1067 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1068 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1069 }
1070 break;
1071
1072 case SO_SNDTIMEO:
1073 lv = sizeof(struct timeval);
1074 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1075 v.tm.tv_sec = 0;
1076 v.tm.tv_usec = 0;
1077 } else {
1078 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1079 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1080 }
1081 break;
1082
1083 case SO_RCVLOWAT:
1084 v.val = sk->sk_rcvlowat;
1085 break;
1086
1087 case SO_SNDLOWAT:
1088 v.val = 1;
1089 break;
1090
1091 case SO_PASSCRED:
1092 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1093 break;
1094
1095 case SO_PEERCRED:
1096 {
1097 struct ucred peercred;
1098 if (len > sizeof(peercred))
1099 len = sizeof(peercred);
1100 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1101 if (copy_to_user(optval, &peercred, len))
1102 return -EFAULT;
1103 goto lenout;
1104 }
1105
1106 case SO_PEERNAME:
1107 {
1108 char address[128];
1109
1110 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1111 return -ENOTCONN;
1112 if (lv < len)
1113 return -EINVAL;
1114 if (copy_to_user(optval, address, len))
1115 return -EFAULT;
1116 goto lenout;
1117 }
1118
1119 /* Dubious BSD thing... Probably nobody even uses it, but
1120 * the UNIX standard wants it for whatever reason... -DaveM
1121 */
1122 case SO_ACCEPTCONN:
1123 v.val = sk->sk_state == TCP_LISTEN;
1124 break;
1125
1126 case SO_PASSSEC:
1127 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1128 break;
1129
1130 case SO_PEERSEC:
1131 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1132
1133 case SO_MARK:
1134 v.val = sk->sk_mark;
1135 break;
1136
1137 case SO_RXQ_OVFL:
1138 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1139 break;
1140
1141 case SO_WIFI_STATUS:
1142 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1143 break;
1144
1145 case SO_PEEK_OFF:
1146 if (!sock->ops->set_peek_off)
1147 return -EOPNOTSUPP;
1148
1149 v.val = sk->sk_peek_off;
1150 break;
1151 case SO_NOFCS:
1152 v.val = sock_flag(sk, SOCK_NOFCS);
1153 break;
1154
1155 case SO_BINDTODEVICE:
1156 return sock_getbindtodevice(sk, optval, optlen, len);
1157
1158 case SO_GET_FILTER:
1159 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1160 if (len < 0)
1161 return len;
1162
1163 goto lenout;
1164
1165 case SO_LOCK_FILTER:
1166 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1167 break;
1168
1169 case SO_SELECT_ERR_QUEUE:
1170 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1171 break;
1172
1173 #ifdef CONFIG_NET_LL_RX_POLL
1174 case SO_BUSY_POLL:
1175 v.val = sk->sk_ll_usec;
1176 break;
1177 #endif
1178
1179 default:
1180 return -ENOPROTOOPT;
1181 }
1182
1183 if (len > lv)
1184 len = lv;
1185 if (copy_to_user(optval, &v, len))
1186 return -EFAULT;
1187 lenout:
1188 if (put_user(len, optlen))
1189 return -EFAULT;
1190 return 0;
1191 }
1192
1193 /*
1194 * Initialize an sk_lock.
1195 *
1196 * (We also register the sk_lock with the lock validator.)
1197 */
1198 static inline void sock_lock_init(struct sock *sk)
1199 {
1200 sock_lock_init_class_and_name(sk,
1201 af_family_slock_key_strings[sk->sk_family],
1202 af_family_slock_keys + sk->sk_family,
1203 af_family_key_strings[sk->sk_family],
1204 af_family_keys + sk->sk_family);
1205 }
1206
1207 /*
1208 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1209 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1210 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1211 */
1212 static void sock_copy(struct sock *nsk, const struct sock *osk)
1213 {
1214 #ifdef CONFIG_SECURITY_NETWORK
1215 void *sptr = nsk->sk_security;
1216 #endif
1217 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1218
1219 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1220 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1221
1222 #ifdef CONFIG_SECURITY_NETWORK
1223 nsk->sk_security = sptr;
1224 security_sk_clone(osk, nsk);
1225 #endif
1226 }
1227
1228 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1229 {
1230 unsigned long nulls1, nulls2;
1231
1232 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1233 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1234 if (nulls1 > nulls2)
1235 swap(nulls1, nulls2);
1236
1237 if (nulls1 != 0)
1238 memset((char *)sk, 0, nulls1);
1239 memset((char *)sk + nulls1 + sizeof(void *), 0,
1240 nulls2 - nulls1 - sizeof(void *));
1241 memset((char *)sk + nulls2 + sizeof(void *), 0,
1242 size - nulls2 - sizeof(void *));
1243 }
1244 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1245
1246 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1247 int family)
1248 {
1249 struct sock *sk;
1250 struct kmem_cache *slab;
1251
1252 slab = prot->slab;
1253 if (slab != NULL) {
1254 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1255 if (!sk)
1256 return sk;
1257 if (priority & __GFP_ZERO) {
1258 if (prot->clear_sk)
1259 prot->clear_sk(sk, prot->obj_size);
1260 else
1261 sk_prot_clear_nulls(sk, prot->obj_size);
1262 }
1263 } else
1264 sk = kmalloc(prot->obj_size, priority);
1265
1266 if (sk != NULL) {
1267 kmemcheck_annotate_bitfield(sk, flags);
1268
1269 if (security_sk_alloc(sk, family, priority))
1270 goto out_free;
1271
1272 if (!try_module_get(prot->owner))
1273 goto out_free_sec;
1274 sk_tx_queue_clear(sk);
1275 }
1276
1277 return sk;
1278
1279 out_free_sec:
1280 security_sk_free(sk);
1281 out_free:
1282 if (slab != NULL)
1283 kmem_cache_free(slab, sk);
1284 else
1285 kfree(sk);
1286 return NULL;
1287 }
1288
1289 static void sk_prot_free(struct proto *prot, struct sock *sk)
1290 {
1291 struct kmem_cache *slab;
1292 struct module *owner;
1293
1294 owner = prot->owner;
1295 slab = prot->slab;
1296
1297 security_sk_free(sk);
1298 if (slab != NULL)
1299 kmem_cache_free(slab, sk);
1300 else
1301 kfree(sk);
1302 module_put(owner);
1303 }
1304
1305 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1306 void sock_update_classid(struct sock *sk)
1307 {
1308 u32 classid;
1309
1310 classid = task_cls_classid(current);
1311 if (classid != sk->sk_classid)
1312 sk->sk_classid = classid;
1313 }
1314 EXPORT_SYMBOL(sock_update_classid);
1315 #endif
1316
1317 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1318 void sock_update_netprioidx(struct sock *sk)
1319 {
1320 if (in_interrupt())
1321 return;
1322
1323 sk->sk_cgrp_prioidx = task_netprioidx(current);
1324 }
1325 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1326 #endif
1327
1328 /**
1329 * sk_alloc - All socket objects are allocated here
1330 * @net: the applicable net namespace
1331 * @family: protocol family
1332 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1333 * @prot: struct proto associated with this new sock instance
1334 */
1335 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1336 struct proto *prot)
1337 {
1338 struct sock *sk;
1339
1340 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1341 if (sk) {
1342 sk->sk_family = family;
1343 /*
1344 * See comment in struct sock definition to understand
1345 * why we need sk_prot_creator -acme
1346 */
1347 sk->sk_prot = sk->sk_prot_creator = prot;
1348 sock_lock_init(sk);
1349 sock_net_set(sk, get_net(net));
1350 atomic_set(&sk->sk_wmem_alloc, 1);
1351
1352 sock_update_classid(sk);
1353 sock_update_netprioidx(sk);
1354 }
1355
1356 return sk;
1357 }
1358 EXPORT_SYMBOL(sk_alloc);
1359
1360 static void __sk_free(struct sock *sk)
1361 {
1362 struct sk_filter *filter;
1363
1364 if (sk->sk_destruct)
1365 sk->sk_destruct(sk);
1366
1367 filter = rcu_dereference_check(sk->sk_filter,
1368 atomic_read(&sk->sk_wmem_alloc) == 0);
1369 if (filter) {
1370 sk_filter_uncharge(sk, filter);
1371 RCU_INIT_POINTER(sk->sk_filter, NULL);
1372 }
1373
1374 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1375
1376 if (atomic_read(&sk->sk_omem_alloc))
1377 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1378 __func__, atomic_read(&sk->sk_omem_alloc));
1379
1380 if (sk->sk_peer_cred)
1381 put_cred(sk->sk_peer_cred);
1382 put_pid(sk->sk_peer_pid);
1383 put_net(sock_net(sk));
1384 sk_prot_free(sk->sk_prot_creator, sk);
1385 }
1386
1387 void sk_free(struct sock *sk)
1388 {
1389 /*
1390 * We subtract one from sk_wmem_alloc and can know if
1391 * some packets are still in some tx queue.
1392 * If not null, sock_wfree() will call __sk_free(sk) later
1393 */
1394 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1395 __sk_free(sk);
1396 }
1397 EXPORT_SYMBOL(sk_free);
1398
1399 /*
1400 * Last sock_put should drop reference to sk->sk_net. It has already
1401 * been dropped in sk_change_net. Taking reference to stopping namespace
1402 * is not an option.
1403 * Take reference to a socket to remove it from hash _alive_ and after that
1404 * destroy it in the context of init_net.
1405 */
1406 void sk_release_kernel(struct sock *sk)
1407 {
1408 if (sk == NULL || sk->sk_socket == NULL)
1409 return;
1410
1411 sock_hold(sk);
1412 sock_release(sk->sk_socket);
1413 release_net(sock_net(sk));
1414 sock_net_set(sk, get_net(&init_net));
1415 sock_put(sk);
1416 }
1417 EXPORT_SYMBOL(sk_release_kernel);
1418
1419 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1420 {
1421 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1422 sock_update_memcg(newsk);
1423 }
1424
1425 /**
1426 * sk_clone_lock - clone a socket, and lock its clone
1427 * @sk: the socket to clone
1428 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1429 *
1430 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1431 */
1432 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1433 {
1434 struct sock *newsk;
1435
1436 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1437 if (newsk != NULL) {
1438 struct sk_filter *filter;
1439
1440 sock_copy(newsk, sk);
1441
1442 /* SANITY */
1443 get_net(sock_net(newsk));
1444 sk_node_init(&newsk->sk_node);
1445 sock_lock_init(newsk);
1446 bh_lock_sock(newsk);
1447 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1448 newsk->sk_backlog.len = 0;
1449
1450 atomic_set(&newsk->sk_rmem_alloc, 0);
1451 /*
1452 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1453 */
1454 atomic_set(&newsk->sk_wmem_alloc, 1);
1455 atomic_set(&newsk->sk_omem_alloc, 0);
1456 skb_queue_head_init(&newsk->sk_receive_queue);
1457 skb_queue_head_init(&newsk->sk_write_queue);
1458 #ifdef CONFIG_NET_DMA
1459 skb_queue_head_init(&newsk->sk_async_wait_queue);
1460 #endif
1461
1462 spin_lock_init(&newsk->sk_dst_lock);
1463 rwlock_init(&newsk->sk_callback_lock);
1464 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1465 af_callback_keys + newsk->sk_family,
1466 af_family_clock_key_strings[newsk->sk_family]);
1467
1468 newsk->sk_dst_cache = NULL;
1469 newsk->sk_wmem_queued = 0;
1470 newsk->sk_forward_alloc = 0;
1471 newsk->sk_send_head = NULL;
1472 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1473
1474 sock_reset_flag(newsk, SOCK_DONE);
1475 skb_queue_head_init(&newsk->sk_error_queue);
1476
1477 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1478 if (filter != NULL)
1479 sk_filter_charge(newsk, filter);
1480
1481 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1482 /* It is still raw copy of parent, so invalidate
1483 * destructor and make plain sk_free() */
1484 newsk->sk_destruct = NULL;
1485 bh_unlock_sock(newsk);
1486 sk_free(newsk);
1487 newsk = NULL;
1488 goto out;
1489 }
1490
1491 newsk->sk_err = 0;
1492 newsk->sk_priority = 0;
1493 /*
1494 * Before updating sk_refcnt, we must commit prior changes to memory
1495 * (Documentation/RCU/rculist_nulls.txt for details)
1496 */
1497 smp_wmb();
1498 atomic_set(&newsk->sk_refcnt, 2);
1499
1500 /*
1501 * Increment the counter in the same struct proto as the master
1502 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1503 * is the same as sk->sk_prot->socks, as this field was copied
1504 * with memcpy).
1505 *
1506 * This _changes_ the previous behaviour, where
1507 * tcp_create_openreq_child always was incrementing the
1508 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1509 * to be taken into account in all callers. -acme
1510 */
1511 sk_refcnt_debug_inc(newsk);
1512 sk_set_socket(newsk, NULL);
1513 newsk->sk_wq = NULL;
1514
1515 sk_update_clone(sk, newsk);
1516
1517 if (newsk->sk_prot->sockets_allocated)
1518 sk_sockets_allocated_inc(newsk);
1519
1520 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1521 net_enable_timestamp();
1522 }
1523 out:
1524 return newsk;
1525 }
1526 EXPORT_SYMBOL_GPL(sk_clone_lock);
1527
1528 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1529 {
1530 __sk_dst_set(sk, dst);
1531 sk->sk_route_caps = dst->dev->features;
1532 if (sk->sk_route_caps & NETIF_F_GSO)
1533 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1534 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1535 if (sk_can_gso(sk)) {
1536 if (dst->header_len) {
1537 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1538 } else {
1539 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1540 sk->sk_gso_max_size = dst->dev->gso_max_size;
1541 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1542 }
1543 }
1544 }
1545 EXPORT_SYMBOL_GPL(sk_setup_caps);
1546
1547 /*
1548 * Simple resource managers for sockets.
1549 */
1550
1551
1552 /*
1553 * Write buffer destructor automatically called from kfree_skb.
1554 */
1555 void sock_wfree(struct sk_buff *skb)
1556 {
1557 struct sock *sk = skb->sk;
1558 unsigned int len = skb->truesize;
1559
1560 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1561 /*
1562 * Keep a reference on sk_wmem_alloc, this will be released
1563 * after sk_write_space() call
1564 */
1565 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1566 sk->sk_write_space(sk);
1567 len = 1;
1568 }
1569 /*
1570 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1571 * could not do because of in-flight packets
1572 */
1573 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1574 __sk_free(sk);
1575 }
1576 EXPORT_SYMBOL(sock_wfree);
1577
1578 /*
1579 * Read buffer destructor automatically called from kfree_skb.
1580 */
1581 void sock_rfree(struct sk_buff *skb)
1582 {
1583 struct sock *sk = skb->sk;
1584 unsigned int len = skb->truesize;
1585
1586 atomic_sub(len, &sk->sk_rmem_alloc);
1587 sk_mem_uncharge(sk, len);
1588 }
1589 EXPORT_SYMBOL(sock_rfree);
1590
1591 void sock_edemux(struct sk_buff *skb)
1592 {
1593 struct sock *sk = skb->sk;
1594
1595 #ifdef CONFIG_INET
1596 if (sk->sk_state == TCP_TIME_WAIT)
1597 inet_twsk_put(inet_twsk(sk));
1598 else
1599 #endif
1600 sock_put(sk);
1601 }
1602 EXPORT_SYMBOL(sock_edemux);
1603
1604 kuid_t sock_i_uid(struct sock *sk)
1605 {
1606 kuid_t uid;
1607
1608 read_lock_bh(&sk->sk_callback_lock);
1609 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1610 read_unlock_bh(&sk->sk_callback_lock);
1611 return uid;
1612 }
1613 EXPORT_SYMBOL(sock_i_uid);
1614
1615 unsigned long sock_i_ino(struct sock *sk)
1616 {
1617 unsigned long ino;
1618
1619 read_lock_bh(&sk->sk_callback_lock);
1620 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1621 read_unlock_bh(&sk->sk_callback_lock);
1622 return ino;
1623 }
1624 EXPORT_SYMBOL(sock_i_ino);
1625
1626 /*
1627 * Allocate a skb from the socket's send buffer.
1628 */
1629 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1630 gfp_t priority)
1631 {
1632 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1633 struct sk_buff *skb = alloc_skb(size, priority);
1634 if (skb) {
1635 skb_set_owner_w(skb, sk);
1636 return skb;
1637 }
1638 }
1639 return NULL;
1640 }
1641 EXPORT_SYMBOL(sock_wmalloc);
1642
1643 /*
1644 * Allocate a skb from the socket's receive buffer.
1645 */
1646 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1647 gfp_t priority)
1648 {
1649 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1650 struct sk_buff *skb = alloc_skb(size, priority);
1651 if (skb) {
1652 skb_set_owner_r(skb, sk);
1653 return skb;
1654 }
1655 }
1656 return NULL;
1657 }
1658
1659 /*
1660 * Allocate a memory block from the socket's option memory buffer.
1661 */
1662 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1663 {
1664 if ((unsigned int)size <= sysctl_optmem_max &&
1665 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1666 void *mem;
1667 /* First do the add, to avoid the race if kmalloc
1668 * might sleep.
1669 */
1670 atomic_add(size, &sk->sk_omem_alloc);
1671 mem = kmalloc(size, priority);
1672 if (mem)
1673 return mem;
1674 atomic_sub(size, &sk->sk_omem_alloc);
1675 }
1676 return NULL;
1677 }
1678 EXPORT_SYMBOL(sock_kmalloc);
1679
1680 /*
1681 * Free an option memory block.
1682 */
1683 void sock_kfree_s(struct sock *sk, void *mem, int size)
1684 {
1685 kfree(mem);
1686 atomic_sub(size, &sk->sk_omem_alloc);
1687 }
1688 EXPORT_SYMBOL(sock_kfree_s);
1689
1690 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1691 I think, these locks should be removed for datagram sockets.
1692 */
1693 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1694 {
1695 DEFINE_WAIT(wait);
1696
1697 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1698 for (;;) {
1699 if (!timeo)
1700 break;
1701 if (signal_pending(current))
1702 break;
1703 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1704 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1705 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1706 break;
1707 if (sk->sk_shutdown & SEND_SHUTDOWN)
1708 break;
1709 if (sk->sk_err)
1710 break;
1711 timeo = schedule_timeout(timeo);
1712 }
1713 finish_wait(sk_sleep(sk), &wait);
1714 return timeo;
1715 }
1716
1717
1718 /*
1719 * Generic send/receive buffer handlers
1720 */
1721
1722 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1723 unsigned long data_len, int noblock,
1724 int *errcode)
1725 {
1726 struct sk_buff *skb;
1727 gfp_t gfp_mask;
1728 long timeo;
1729 int err;
1730 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1731
1732 err = -EMSGSIZE;
1733 if (npages > MAX_SKB_FRAGS)
1734 goto failure;
1735
1736 gfp_mask = sk->sk_allocation;
1737 if (gfp_mask & __GFP_WAIT)
1738 gfp_mask |= __GFP_REPEAT;
1739
1740 timeo = sock_sndtimeo(sk, noblock);
1741 while (1) {
1742 err = sock_error(sk);
1743 if (err != 0)
1744 goto failure;
1745
1746 err = -EPIPE;
1747 if (sk->sk_shutdown & SEND_SHUTDOWN)
1748 goto failure;
1749
1750 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1751 skb = alloc_skb(header_len, gfp_mask);
1752 if (skb) {
1753 int i;
1754
1755 /* No pages, we're done... */
1756 if (!data_len)
1757 break;
1758
1759 skb->truesize += data_len;
1760 skb_shinfo(skb)->nr_frags = npages;
1761 for (i = 0; i < npages; i++) {
1762 struct page *page;
1763
1764 page = alloc_pages(sk->sk_allocation, 0);
1765 if (!page) {
1766 err = -ENOBUFS;
1767 skb_shinfo(skb)->nr_frags = i;
1768 kfree_skb(skb);
1769 goto failure;
1770 }
1771
1772 __skb_fill_page_desc(skb, i,
1773 page, 0,
1774 (data_len >= PAGE_SIZE ?
1775 PAGE_SIZE :
1776 data_len));
1777 data_len -= PAGE_SIZE;
1778 }
1779
1780 /* Full success... */
1781 break;
1782 }
1783 err = -ENOBUFS;
1784 goto failure;
1785 }
1786 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1787 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1788 err = -EAGAIN;
1789 if (!timeo)
1790 goto failure;
1791 if (signal_pending(current))
1792 goto interrupted;
1793 timeo = sock_wait_for_wmem(sk, timeo);
1794 }
1795
1796 skb_set_owner_w(skb, sk);
1797 return skb;
1798
1799 interrupted:
1800 err = sock_intr_errno(timeo);
1801 failure:
1802 *errcode = err;
1803 return NULL;
1804 }
1805 EXPORT_SYMBOL(sock_alloc_send_pskb);
1806
1807 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1808 int noblock, int *errcode)
1809 {
1810 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1811 }
1812 EXPORT_SYMBOL(sock_alloc_send_skb);
1813
1814 /* On 32bit arches, an skb frag is limited to 2^15 */
1815 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1816
1817 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1818 {
1819 int order;
1820
1821 if (pfrag->page) {
1822 if (atomic_read(&pfrag->page->_count) == 1) {
1823 pfrag->offset = 0;
1824 return true;
1825 }
1826 if (pfrag->offset < pfrag->size)
1827 return true;
1828 put_page(pfrag->page);
1829 }
1830
1831 /* We restrict high order allocations to users that can afford to wait */
1832 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1833
1834 do {
1835 gfp_t gfp = sk->sk_allocation;
1836
1837 if (order)
1838 gfp |= __GFP_COMP | __GFP_NOWARN;
1839 pfrag->page = alloc_pages(gfp, order);
1840 if (likely(pfrag->page)) {
1841 pfrag->offset = 0;
1842 pfrag->size = PAGE_SIZE << order;
1843 return true;
1844 }
1845 } while (--order >= 0);
1846
1847 sk_enter_memory_pressure(sk);
1848 sk_stream_moderate_sndbuf(sk);
1849 return false;
1850 }
1851 EXPORT_SYMBOL(sk_page_frag_refill);
1852
1853 static void __lock_sock(struct sock *sk)
1854 __releases(&sk->sk_lock.slock)
1855 __acquires(&sk->sk_lock.slock)
1856 {
1857 DEFINE_WAIT(wait);
1858
1859 for (;;) {
1860 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1861 TASK_UNINTERRUPTIBLE);
1862 spin_unlock_bh(&sk->sk_lock.slock);
1863 schedule();
1864 spin_lock_bh(&sk->sk_lock.slock);
1865 if (!sock_owned_by_user(sk))
1866 break;
1867 }
1868 finish_wait(&sk->sk_lock.wq, &wait);
1869 }
1870
1871 static void __release_sock(struct sock *sk)
1872 __releases(&sk->sk_lock.slock)
1873 __acquires(&sk->sk_lock.slock)
1874 {
1875 struct sk_buff *skb = sk->sk_backlog.head;
1876
1877 do {
1878 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1879 bh_unlock_sock(sk);
1880
1881 do {
1882 struct sk_buff *next = skb->next;
1883
1884 prefetch(next);
1885 WARN_ON_ONCE(skb_dst_is_noref(skb));
1886 skb->next = NULL;
1887 sk_backlog_rcv(sk, skb);
1888
1889 /*
1890 * We are in process context here with softirqs
1891 * disabled, use cond_resched_softirq() to preempt.
1892 * This is safe to do because we've taken the backlog
1893 * queue private:
1894 */
1895 cond_resched_softirq();
1896
1897 skb = next;
1898 } while (skb != NULL);
1899
1900 bh_lock_sock(sk);
1901 } while ((skb = sk->sk_backlog.head) != NULL);
1902
1903 /*
1904 * Doing the zeroing here guarantee we can not loop forever
1905 * while a wild producer attempts to flood us.
1906 */
1907 sk->sk_backlog.len = 0;
1908 }
1909
1910 /**
1911 * sk_wait_data - wait for data to arrive at sk_receive_queue
1912 * @sk: sock to wait on
1913 * @timeo: for how long
1914 *
1915 * Now socket state including sk->sk_err is changed only under lock,
1916 * hence we may omit checks after joining wait queue.
1917 * We check receive queue before schedule() only as optimization;
1918 * it is very likely that release_sock() added new data.
1919 */
1920 int sk_wait_data(struct sock *sk, long *timeo)
1921 {
1922 int rc;
1923 DEFINE_WAIT(wait);
1924
1925 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1926 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1927 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1928 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1929 finish_wait(sk_sleep(sk), &wait);
1930 return rc;
1931 }
1932 EXPORT_SYMBOL(sk_wait_data);
1933
1934 /**
1935 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1936 * @sk: socket
1937 * @size: memory size to allocate
1938 * @kind: allocation type
1939 *
1940 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1941 * rmem allocation. This function assumes that protocols which have
1942 * memory_pressure use sk_wmem_queued as write buffer accounting.
1943 */
1944 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1945 {
1946 struct proto *prot = sk->sk_prot;
1947 int amt = sk_mem_pages(size);
1948 long allocated;
1949 int parent_status = UNDER_LIMIT;
1950
1951 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1952
1953 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1954
1955 /* Under limit. */
1956 if (parent_status == UNDER_LIMIT &&
1957 allocated <= sk_prot_mem_limits(sk, 0)) {
1958 sk_leave_memory_pressure(sk);
1959 return 1;
1960 }
1961
1962 /* Under pressure. (we or our parents) */
1963 if ((parent_status > SOFT_LIMIT) ||
1964 allocated > sk_prot_mem_limits(sk, 1))
1965 sk_enter_memory_pressure(sk);
1966
1967 /* Over hard limit (we or our parents) */
1968 if ((parent_status == OVER_LIMIT) ||
1969 (allocated > sk_prot_mem_limits(sk, 2)))
1970 goto suppress_allocation;
1971
1972 /* guarantee minimum buffer size under pressure */
1973 if (kind == SK_MEM_RECV) {
1974 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1975 return 1;
1976
1977 } else { /* SK_MEM_SEND */
1978 if (sk->sk_type == SOCK_STREAM) {
1979 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1980 return 1;
1981 } else if (atomic_read(&sk->sk_wmem_alloc) <
1982 prot->sysctl_wmem[0])
1983 return 1;
1984 }
1985
1986 if (sk_has_memory_pressure(sk)) {
1987 int alloc;
1988
1989 if (!sk_under_memory_pressure(sk))
1990 return 1;
1991 alloc = sk_sockets_allocated_read_positive(sk);
1992 if (sk_prot_mem_limits(sk, 2) > alloc *
1993 sk_mem_pages(sk->sk_wmem_queued +
1994 atomic_read(&sk->sk_rmem_alloc) +
1995 sk->sk_forward_alloc))
1996 return 1;
1997 }
1998
1999 suppress_allocation:
2000
2001 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2002 sk_stream_moderate_sndbuf(sk);
2003
2004 /* Fail only if socket is _under_ its sndbuf.
2005 * In this case we cannot block, so that we have to fail.
2006 */
2007 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2008 return 1;
2009 }
2010
2011 trace_sock_exceed_buf_limit(sk, prot, allocated);
2012
2013 /* Alas. Undo changes. */
2014 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2015
2016 sk_memory_allocated_sub(sk, amt);
2017
2018 return 0;
2019 }
2020 EXPORT_SYMBOL(__sk_mem_schedule);
2021
2022 /**
2023 * __sk_reclaim - reclaim memory_allocated
2024 * @sk: socket
2025 */
2026 void __sk_mem_reclaim(struct sock *sk)
2027 {
2028 sk_memory_allocated_sub(sk,
2029 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2030 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2031
2032 if (sk_under_memory_pressure(sk) &&
2033 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2034 sk_leave_memory_pressure(sk);
2035 }
2036 EXPORT_SYMBOL(__sk_mem_reclaim);
2037
2038
2039 /*
2040 * Set of default routines for initialising struct proto_ops when
2041 * the protocol does not support a particular function. In certain
2042 * cases where it makes no sense for a protocol to have a "do nothing"
2043 * function, some default processing is provided.
2044 */
2045
2046 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2047 {
2048 return -EOPNOTSUPP;
2049 }
2050 EXPORT_SYMBOL(sock_no_bind);
2051
2052 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2053 int len, int flags)
2054 {
2055 return -EOPNOTSUPP;
2056 }
2057 EXPORT_SYMBOL(sock_no_connect);
2058
2059 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2060 {
2061 return -EOPNOTSUPP;
2062 }
2063 EXPORT_SYMBOL(sock_no_socketpair);
2064
2065 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2066 {
2067 return -EOPNOTSUPP;
2068 }
2069 EXPORT_SYMBOL(sock_no_accept);
2070
2071 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2072 int *len, int peer)
2073 {
2074 return -EOPNOTSUPP;
2075 }
2076 EXPORT_SYMBOL(sock_no_getname);
2077
2078 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2079 {
2080 return 0;
2081 }
2082 EXPORT_SYMBOL(sock_no_poll);
2083
2084 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2085 {
2086 return -EOPNOTSUPP;
2087 }
2088 EXPORT_SYMBOL(sock_no_ioctl);
2089
2090 int sock_no_listen(struct socket *sock, int backlog)
2091 {
2092 return -EOPNOTSUPP;
2093 }
2094 EXPORT_SYMBOL(sock_no_listen);
2095
2096 int sock_no_shutdown(struct socket *sock, int how)
2097 {
2098 return -EOPNOTSUPP;
2099 }
2100 EXPORT_SYMBOL(sock_no_shutdown);
2101
2102 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2103 char __user *optval, unsigned int optlen)
2104 {
2105 return -EOPNOTSUPP;
2106 }
2107 EXPORT_SYMBOL(sock_no_setsockopt);
2108
2109 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2110 char __user *optval, int __user *optlen)
2111 {
2112 return -EOPNOTSUPP;
2113 }
2114 EXPORT_SYMBOL(sock_no_getsockopt);
2115
2116 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2117 size_t len)
2118 {
2119 return -EOPNOTSUPP;
2120 }
2121 EXPORT_SYMBOL(sock_no_sendmsg);
2122
2123 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2124 size_t len, int flags)
2125 {
2126 return -EOPNOTSUPP;
2127 }
2128 EXPORT_SYMBOL(sock_no_recvmsg);
2129
2130 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2131 {
2132 /* Mirror missing mmap method error code */
2133 return -ENODEV;
2134 }
2135 EXPORT_SYMBOL(sock_no_mmap);
2136
2137 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2138 {
2139 ssize_t res;
2140 struct msghdr msg = {.msg_flags = flags};
2141 struct kvec iov;
2142 char *kaddr = kmap(page);
2143 iov.iov_base = kaddr + offset;
2144 iov.iov_len = size;
2145 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2146 kunmap(page);
2147 return res;
2148 }
2149 EXPORT_SYMBOL(sock_no_sendpage);
2150
2151 /*
2152 * Default Socket Callbacks
2153 */
2154
2155 static void sock_def_wakeup(struct sock *sk)
2156 {
2157 struct socket_wq *wq;
2158
2159 rcu_read_lock();
2160 wq = rcu_dereference(sk->sk_wq);
2161 if (wq_has_sleeper(wq))
2162 wake_up_interruptible_all(&wq->wait);
2163 rcu_read_unlock();
2164 }
2165
2166 static void sock_def_error_report(struct sock *sk)
2167 {
2168 struct socket_wq *wq;
2169
2170 rcu_read_lock();
2171 wq = rcu_dereference(sk->sk_wq);
2172 if (wq_has_sleeper(wq))
2173 wake_up_interruptible_poll(&wq->wait, POLLERR);
2174 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2175 rcu_read_unlock();
2176 }
2177
2178 static void sock_def_readable(struct sock *sk, int len)
2179 {
2180 struct socket_wq *wq;
2181
2182 rcu_read_lock();
2183 wq = rcu_dereference(sk->sk_wq);
2184 if (wq_has_sleeper(wq))
2185 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2186 POLLRDNORM | POLLRDBAND);
2187 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2188 rcu_read_unlock();
2189 }
2190
2191 static void sock_def_write_space(struct sock *sk)
2192 {
2193 struct socket_wq *wq;
2194
2195 rcu_read_lock();
2196
2197 /* Do not wake up a writer until he can make "significant"
2198 * progress. --DaveM
2199 */
2200 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2201 wq = rcu_dereference(sk->sk_wq);
2202 if (wq_has_sleeper(wq))
2203 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2204 POLLWRNORM | POLLWRBAND);
2205
2206 /* Should agree with poll, otherwise some programs break */
2207 if (sock_writeable(sk))
2208 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2209 }
2210
2211 rcu_read_unlock();
2212 }
2213
2214 static void sock_def_destruct(struct sock *sk)
2215 {
2216 kfree(sk->sk_protinfo);
2217 }
2218
2219 void sk_send_sigurg(struct sock *sk)
2220 {
2221 if (sk->sk_socket && sk->sk_socket->file)
2222 if (send_sigurg(&sk->sk_socket->file->f_owner))
2223 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2224 }
2225 EXPORT_SYMBOL(sk_send_sigurg);
2226
2227 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2228 unsigned long expires)
2229 {
2230 if (!mod_timer(timer, expires))
2231 sock_hold(sk);
2232 }
2233 EXPORT_SYMBOL(sk_reset_timer);
2234
2235 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2236 {
2237 if (del_timer(timer))
2238 __sock_put(sk);
2239 }
2240 EXPORT_SYMBOL(sk_stop_timer);
2241
2242 void sock_init_data(struct socket *sock, struct sock *sk)
2243 {
2244 skb_queue_head_init(&sk->sk_receive_queue);
2245 skb_queue_head_init(&sk->sk_write_queue);
2246 skb_queue_head_init(&sk->sk_error_queue);
2247 #ifdef CONFIG_NET_DMA
2248 skb_queue_head_init(&sk->sk_async_wait_queue);
2249 #endif
2250
2251 sk->sk_send_head = NULL;
2252
2253 init_timer(&sk->sk_timer);
2254
2255 sk->sk_allocation = GFP_KERNEL;
2256 sk->sk_rcvbuf = sysctl_rmem_default;
2257 sk->sk_sndbuf = sysctl_wmem_default;
2258 sk->sk_state = TCP_CLOSE;
2259 sk_set_socket(sk, sock);
2260
2261 sock_set_flag(sk, SOCK_ZAPPED);
2262
2263 if (sock) {
2264 sk->sk_type = sock->type;
2265 sk->sk_wq = sock->wq;
2266 sock->sk = sk;
2267 } else
2268 sk->sk_wq = NULL;
2269
2270 spin_lock_init(&sk->sk_dst_lock);
2271 rwlock_init(&sk->sk_callback_lock);
2272 lockdep_set_class_and_name(&sk->sk_callback_lock,
2273 af_callback_keys + sk->sk_family,
2274 af_family_clock_key_strings[sk->sk_family]);
2275
2276 sk->sk_state_change = sock_def_wakeup;
2277 sk->sk_data_ready = sock_def_readable;
2278 sk->sk_write_space = sock_def_write_space;
2279 sk->sk_error_report = sock_def_error_report;
2280 sk->sk_destruct = sock_def_destruct;
2281
2282 sk->sk_frag.page = NULL;
2283 sk->sk_frag.offset = 0;
2284 sk->sk_peek_off = -1;
2285
2286 sk->sk_peer_pid = NULL;
2287 sk->sk_peer_cred = NULL;
2288 sk->sk_write_pending = 0;
2289 sk->sk_rcvlowat = 1;
2290 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2291 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2292
2293 sk->sk_stamp = ktime_set(-1L, 0);
2294
2295 #ifdef CONFIG_NET_LL_RX_POLL
2296 sk->sk_napi_id = 0;
2297 sk->sk_ll_usec = sysctl_net_busy_read;
2298 #endif
2299
2300 /*
2301 * Before updating sk_refcnt, we must commit prior changes to memory
2302 * (Documentation/RCU/rculist_nulls.txt for details)
2303 */
2304 smp_wmb();
2305 atomic_set(&sk->sk_refcnt, 1);
2306 atomic_set(&sk->sk_drops, 0);
2307 }
2308 EXPORT_SYMBOL(sock_init_data);
2309
2310 void lock_sock_nested(struct sock *sk, int subclass)
2311 {
2312 might_sleep();
2313 spin_lock_bh(&sk->sk_lock.slock);
2314 if (sk->sk_lock.owned)
2315 __lock_sock(sk);
2316 sk->sk_lock.owned = 1;
2317 spin_unlock(&sk->sk_lock.slock);
2318 /*
2319 * The sk_lock has mutex_lock() semantics here:
2320 */
2321 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2322 local_bh_enable();
2323 }
2324 EXPORT_SYMBOL(lock_sock_nested);
2325
2326 void release_sock(struct sock *sk)
2327 {
2328 /*
2329 * The sk_lock has mutex_unlock() semantics:
2330 */
2331 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2332
2333 spin_lock_bh(&sk->sk_lock.slock);
2334 if (sk->sk_backlog.tail)
2335 __release_sock(sk);
2336
2337 if (sk->sk_prot->release_cb)
2338 sk->sk_prot->release_cb(sk);
2339
2340 sk->sk_lock.owned = 0;
2341 if (waitqueue_active(&sk->sk_lock.wq))
2342 wake_up(&sk->sk_lock.wq);
2343 spin_unlock_bh(&sk->sk_lock.slock);
2344 }
2345 EXPORT_SYMBOL(release_sock);
2346
2347 /**
2348 * lock_sock_fast - fast version of lock_sock
2349 * @sk: socket
2350 *
2351 * This version should be used for very small section, where process wont block
2352 * return false if fast path is taken
2353 * sk_lock.slock locked, owned = 0, BH disabled
2354 * return true if slow path is taken
2355 * sk_lock.slock unlocked, owned = 1, BH enabled
2356 */
2357 bool lock_sock_fast(struct sock *sk)
2358 {
2359 might_sleep();
2360 spin_lock_bh(&sk->sk_lock.slock);
2361
2362 if (!sk->sk_lock.owned)
2363 /*
2364 * Note : We must disable BH
2365 */
2366 return false;
2367
2368 __lock_sock(sk);
2369 sk->sk_lock.owned = 1;
2370 spin_unlock(&sk->sk_lock.slock);
2371 /*
2372 * The sk_lock has mutex_lock() semantics here:
2373 */
2374 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2375 local_bh_enable();
2376 return true;
2377 }
2378 EXPORT_SYMBOL(lock_sock_fast);
2379
2380 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2381 {
2382 struct timeval tv;
2383 if (!sock_flag(sk, SOCK_TIMESTAMP))
2384 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2385 tv = ktime_to_timeval(sk->sk_stamp);
2386 if (tv.tv_sec == -1)
2387 return -ENOENT;
2388 if (tv.tv_sec == 0) {
2389 sk->sk_stamp = ktime_get_real();
2390 tv = ktime_to_timeval(sk->sk_stamp);
2391 }
2392 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2393 }
2394 EXPORT_SYMBOL(sock_get_timestamp);
2395
2396 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2397 {
2398 struct timespec ts;
2399 if (!sock_flag(sk, SOCK_TIMESTAMP))
2400 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2401 ts = ktime_to_timespec(sk->sk_stamp);
2402 if (ts.tv_sec == -1)
2403 return -ENOENT;
2404 if (ts.tv_sec == 0) {
2405 sk->sk_stamp = ktime_get_real();
2406 ts = ktime_to_timespec(sk->sk_stamp);
2407 }
2408 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2409 }
2410 EXPORT_SYMBOL(sock_get_timestampns);
2411
2412 void sock_enable_timestamp(struct sock *sk, int flag)
2413 {
2414 if (!sock_flag(sk, flag)) {
2415 unsigned long previous_flags = sk->sk_flags;
2416
2417 sock_set_flag(sk, flag);
2418 /*
2419 * we just set one of the two flags which require net
2420 * time stamping, but time stamping might have been on
2421 * already because of the other one
2422 */
2423 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2424 net_enable_timestamp();
2425 }
2426 }
2427
2428 /*
2429 * Get a socket option on an socket.
2430 *
2431 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2432 * asynchronous errors should be reported by getsockopt. We assume
2433 * this means if you specify SO_ERROR (otherwise whats the point of it).
2434 */
2435 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2436 char __user *optval, int __user *optlen)
2437 {
2438 struct sock *sk = sock->sk;
2439
2440 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2441 }
2442 EXPORT_SYMBOL(sock_common_getsockopt);
2443
2444 #ifdef CONFIG_COMPAT
2445 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2446 char __user *optval, int __user *optlen)
2447 {
2448 struct sock *sk = sock->sk;
2449
2450 if (sk->sk_prot->compat_getsockopt != NULL)
2451 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2452 optval, optlen);
2453 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2454 }
2455 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2456 #endif
2457
2458 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2459 struct msghdr *msg, size_t size, int flags)
2460 {
2461 struct sock *sk = sock->sk;
2462 int addr_len = 0;
2463 int err;
2464
2465 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2466 flags & ~MSG_DONTWAIT, &addr_len);
2467 if (err >= 0)
2468 msg->msg_namelen = addr_len;
2469 return err;
2470 }
2471 EXPORT_SYMBOL(sock_common_recvmsg);
2472
2473 /*
2474 * Set socket options on an inet socket.
2475 */
2476 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2477 char __user *optval, unsigned int optlen)
2478 {
2479 struct sock *sk = sock->sk;
2480
2481 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2482 }
2483 EXPORT_SYMBOL(sock_common_setsockopt);
2484
2485 #ifdef CONFIG_COMPAT
2486 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2487 char __user *optval, unsigned int optlen)
2488 {
2489 struct sock *sk = sock->sk;
2490
2491 if (sk->sk_prot->compat_setsockopt != NULL)
2492 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2493 optval, optlen);
2494 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2495 }
2496 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2497 #endif
2498
2499 void sk_common_release(struct sock *sk)
2500 {
2501 if (sk->sk_prot->destroy)
2502 sk->sk_prot->destroy(sk);
2503
2504 /*
2505 * Observation: when sock_common_release is called, processes have
2506 * no access to socket. But net still has.
2507 * Step one, detach it from networking:
2508 *
2509 * A. Remove from hash tables.
2510 */
2511
2512 sk->sk_prot->unhash(sk);
2513
2514 /*
2515 * In this point socket cannot receive new packets, but it is possible
2516 * that some packets are in flight because some CPU runs receiver and
2517 * did hash table lookup before we unhashed socket. They will achieve
2518 * receive queue and will be purged by socket destructor.
2519 *
2520 * Also we still have packets pending on receive queue and probably,
2521 * our own packets waiting in device queues. sock_destroy will drain
2522 * receive queue, but transmitted packets will delay socket destruction
2523 * until the last reference will be released.
2524 */
2525
2526 sock_orphan(sk);
2527
2528 xfrm_sk_free_policy(sk);
2529
2530 sk_refcnt_debug_release(sk);
2531
2532 if (sk->sk_frag.page) {
2533 put_page(sk->sk_frag.page);
2534 sk->sk_frag.page = NULL;
2535 }
2536
2537 sock_put(sk);
2538 }
2539 EXPORT_SYMBOL(sk_common_release);
2540
2541 #ifdef CONFIG_PROC_FS
2542 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2543 struct prot_inuse {
2544 int val[PROTO_INUSE_NR];
2545 };
2546
2547 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2548
2549 #ifdef CONFIG_NET_NS
2550 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2551 {
2552 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2553 }
2554 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2555
2556 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2557 {
2558 int cpu, idx = prot->inuse_idx;
2559 int res = 0;
2560
2561 for_each_possible_cpu(cpu)
2562 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2563
2564 return res >= 0 ? res : 0;
2565 }
2566 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2567
2568 static int __net_init sock_inuse_init_net(struct net *net)
2569 {
2570 net->core.inuse = alloc_percpu(struct prot_inuse);
2571 return net->core.inuse ? 0 : -ENOMEM;
2572 }
2573
2574 static void __net_exit sock_inuse_exit_net(struct net *net)
2575 {
2576 free_percpu(net->core.inuse);
2577 }
2578
2579 static struct pernet_operations net_inuse_ops = {
2580 .init = sock_inuse_init_net,
2581 .exit = sock_inuse_exit_net,
2582 };
2583
2584 static __init int net_inuse_init(void)
2585 {
2586 if (register_pernet_subsys(&net_inuse_ops))
2587 panic("Cannot initialize net inuse counters");
2588
2589 return 0;
2590 }
2591
2592 core_initcall(net_inuse_init);
2593 #else
2594 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2595
2596 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2597 {
2598 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2599 }
2600 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2601
2602 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2603 {
2604 int cpu, idx = prot->inuse_idx;
2605 int res = 0;
2606
2607 for_each_possible_cpu(cpu)
2608 res += per_cpu(prot_inuse, cpu).val[idx];
2609
2610 return res >= 0 ? res : 0;
2611 }
2612 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2613 #endif
2614
2615 static void assign_proto_idx(struct proto *prot)
2616 {
2617 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2618
2619 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2620 pr_err("PROTO_INUSE_NR exhausted\n");
2621 return;
2622 }
2623
2624 set_bit(prot->inuse_idx, proto_inuse_idx);
2625 }
2626
2627 static void release_proto_idx(struct proto *prot)
2628 {
2629 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2630 clear_bit(prot->inuse_idx, proto_inuse_idx);
2631 }
2632 #else
2633 static inline void assign_proto_idx(struct proto *prot)
2634 {
2635 }
2636
2637 static inline void release_proto_idx(struct proto *prot)
2638 {
2639 }
2640 #endif
2641
2642 int proto_register(struct proto *prot, int alloc_slab)
2643 {
2644 if (alloc_slab) {
2645 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2646 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2647 NULL);
2648
2649 if (prot->slab == NULL) {
2650 pr_crit("%s: Can't create sock SLAB cache!\n",
2651 prot->name);
2652 goto out;
2653 }
2654
2655 if (prot->rsk_prot != NULL) {
2656 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2657 if (prot->rsk_prot->slab_name == NULL)
2658 goto out_free_sock_slab;
2659
2660 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2661 prot->rsk_prot->obj_size, 0,
2662 SLAB_HWCACHE_ALIGN, NULL);
2663
2664 if (prot->rsk_prot->slab == NULL) {
2665 pr_crit("%s: Can't create request sock SLAB cache!\n",
2666 prot->name);
2667 goto out_free_request_sock_slab_name;
2668 }
2669 }
2670
2671 if (prot->twsk_prot != NULL) {
2672 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2673
2674 if (prot->twsk_prot->twsk_slab_name == NULL)
2675 goto out_free_request_sock_slab;
2676
2677 prot->twsk_prot->twsk_slab =
2678 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2679 prot->twsk_prot->twsk_obj_size,
2680 0,
2681 SLAB_HWCACHE_ALIGN |
2682 prot->slab_flags,
2683 NULL);
2684 if (prot->twsk_prot->twsk_slab == NULL)
2685 goto out_free_timewait_sock_slab_name;
2686 }
2687 }
2688
2689 mutex_lock(&proto_list_mutex);
2690 list_add(&prot->node, &proto_list);
2691 assign_proto_idx(prot);
2692 mutex_unlock(&proto_list_mutex);
2693 return 0;
2694
2695 out_free_timewait_sock_slab_name:
2696 kfree(prot->twsk_prot->twsk_slab_name);
2697 out_free_request_sock_slab:
2698 if (prot->rsk_prot && prot->rsk_prot->slab) {
2699 kmem_cache_destroy(prot->rsk_prot->slab);
2700 prot->rsk_prot->slab = NULL;
2701 }
2702 out_free_request_sock_slab_name:
2703 if (prot->rsk_prot)
2704 kfree(prot->rsk_prot->slab_name);
2705 out_free_sock_slab:
2706 kmem_cache_destroy(prot->slab);
2707 prot->slab = NULL;
2708 out:
2709 return -ENOBUFS;
2710 }
2711 EXPORT_SYMBOL(proto_register);
2712
2713 void proto_unregister(struct proto *prot)
2714 {
2715 mutex_lock(&proto_list_mutex);
2716 release_proto_idx(prot);
2717 list_del(&prot->node);
2718 mutex_unlock(&proto_list_mutex);
2719
2720 if (prot->slab != NULL) {
2721 kmem_cache_destroy(prot->slab);
2722 prot->slab = NULL;
2723 }
2724
2725 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2726 kmem_cache_destroy(prot->rsk_prot->slab);
2727 kfree(prot->rsk_prot->slab_name);
2728 prot->rsk_prot->slab = NULL;
2729 }
2730
2731 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2732 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2733 kfree(prot->twsk_prot->twsk_slab_name);
2734 prot->twsk_prot->twsk_slab = NULL;
2735 }
2736 }
2737 EXPORT_SYMBOL(proto_unregister);
2738
2739 #ifdef CONFIG_PROC_FS
2740 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2741 __acquires(proto_list_mutex)
2742 {
2743 mutex_lock(&proto_list_mutex);
2744 return seq_list_start_head(&proto_list, *pos);
2745 }
2746
2747 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2748 {
2749 return seq_list_next(v, &proto_list, pos);
2750 }
2751
2752 static void proto_seq_stop(struct seq_file *seq, void *v)
2753 __releases(proto_list_mutex)
2754 {
2755 mutex_unlock(&proto_list_mutex);
2756 }
2757
2758 static char proto_method_implemented(const void *method)
2759 {
2760 return method == NULL ? 'n' : 'y';
2761 }
2762 static long sock_prot_memory_allocated(struct proto *proto)
2763 {
2764 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2765 }
2766
2767 static char *sock_prot_memory_pressure(struct proto *proto)
2768 {
2769 return proto->memory_pressure != NULL ?
2770 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2771 }
2772
2773 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2774 {
2775
2776 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2777 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2778 proto->name,
2779 proto->obj_size,
2780 sock_prot_inuse_get(seq_file_net(seq), proto),
2781 sock_prot_memory_allocated(proto),
2782 sock_prot_memory_pressure(proto),
2783 proto->max_header,
2784 proto->slab == NULL ? "no" : "yes",
2785 module_name(proto->owner),
2786 proto_method_implemented(proto->close),
2787 proto_method_implemented(proto->connect),
2788 proto_method_implemented(proto->disconnect),
2789 proto_method_implemented(proto->accept),
2790 proto_method_implemented(proto->ioctl),
2791 proto_method_implemented(proto->init),
2792 proto_method_implemented(proto->destroy),
2793 proto_method_implemented(proto->shutdown),
2794 proto_method_implemented(proto->setsockopt),
2795 proto_method_implemented(proto->getsockopt),
2796 proto_method_implemented(proto->sendmsg),
2797 proto_method_implemented(proto->recvmsg),
2798 proto_method_implemented(proto->sendpage),
2799 proto_method_implemented(proto->bind),
2800 proto_method_implemented(proto->backlog_rcv),
2801 proto_method_implemented(proto->hash),
2802 proto_method_implemented(proto->unhash),
2803 proto_method_implemented(proto->get_port),
2804 proto_method_implemented(proto->enter_memory_pressure));
2805 }
2806
2807 static int proto_seq_show(struct seq_file *seq, void *v)
2808 {
2809 if (v == &proto_list)
2810 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2811 "protocol",
2812 "size",
2813 "sockets",
2814 "memory",
2815 "press",
2816 "maxhdr",
2817 "slab",
2818 "module",
2819 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2820 else
2821 proto_seq_printf(seq, list_entry(v, struct proto, node));
2822 return 0;
2823 }
2824
2825 static const struct seq_operations proto_seq_ops = {
2826 .start = proto_seq_start,
2827 .next = proto_seq_next,
2828 .stop = proto_seq_stop,
2829 .show = proto_seq_show,
2830 };
2831
2832 static int proto_seq_open(struct inode *inode, struct file *file)
2833 {
2834 return seq_open_net(inode, file, &proto_seq_ops,
2835 sizeof(struct seq_net_private));
2836 }
2837
2838 static const struct file_operations proto_seq_fops = {
2839 .owner = THIS_MODULE,
2840 .open = proto_seq_open,
2841 .read = seq_read,
2842 .llseek = seq_lseek,
2843 .release = seq_release_net,
2844 };
2845
2846 static __net_init int proto_init_net(struct net *net)
2847 {
2848 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2849 return -ENOMEM;
2850
2851 return 0;
2852 }
2853
2854 static __net_exit void proto_exit_net(struct net *net)
2855 {
2856 remove_proc_entry("protocols", net->proc_net);
2857 }
2858
2859
2860 static __net_initdata struct pernet_operations proto_net_ops = {
2861 .init = proto_init_net,
2862 .exit = proto_exit_net,
2863 };
2864
2865 static int __init proto_init(void)
2866 {
2867 return register_pernet_subsys(&proto_net_ops);
2868 }
2869
2870 subsys_initcall(proto_init);
2871
2872 #endif /* PROC_FS */
Linus' kernel tree