?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: add low latency socket poll
[linux-2.6.git] / net / core / sock.c
blob788c0da5eed17fedee804c1654bff2b33d7d0261
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/ll_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 struct net_device *dev;
577 char devname[IFNAMSIZ];
578 unsigned seq;
579
580 if (sk->sk_bound_dev_if == 0) {
581 len = 0;
582 goto zero;
583 }
584
585 ret = -EINVAL;
586 if (len < IFNAMSIZ)
587 goto out;
588
589 retry:
590 seq = read_seqcount_begin(&devnet_rename_seq);
591 rcu_read_lock();
592 dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
593 ret = -ENODEV;
594 if (!dev) {
595 rcu_read_unlock();
596 goto out;
597 }
598
599 strcpy(devname, dev->name);
600 rcu_read_unlock();
601 if (read_seqcount_retry(&devnet_rename_seq, seq))
602 goto retry;
603
604 len = strlen(devname) + 1;
605
606 ret = -EFAULT;
607 if (copy_to_user(optval, devname, len))
608 goto out;
609
610 zero:
611 ret = -EFAULT;
612 if (put_user(len, optlen))
613 goto out;
614
615 ret = 0;
616
617 out:
618 #endif
619
620 return ret;
621 }
622
623 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
624 {
625 if (valbool)
626 sock_set_flag(sk, bit);
627 else
628 sock_reset_flag(sk, bit);
629 }
630
631 /*
632 * This is meant for all protocols to use and covers goings on
633 * at the socket level. Everything here is generic.
634 */
635
636 int sock_setsockopt(struct socket *sock, int level, int optname,
637 char __user *optval, unsigned int optlen)
638 {
639 struct sock *sk = sock->sk;
640 int val;
641 int valbool;
642 struct linger ling;
643 int ret = 0;
644
645 /*
646 * Options without arguments
647 */
648
649 if (optname == SO_BINDTODEVICE)
650 return sock_setbindtodevice(sk, optval, optlen);
651
652 if (optlen < sizeof(int))
653 return -EINVAL;
654
655 if (get_user(val, (int __user *)optval))
656 return -EFAULT;
657
658 valbool = val ? 1 : 0;
659
660 lock_sock(sk);
661
662 switch (optname) {
663 case SO_DEBUG:
664 if (val && !capable(CAP_NET_ADMIN))
665 ret = -EACCES;
666 else
667 sock_valbool_flag(sk, SOCK_DBG, valbool);
668 break;
669 case SO_REUSEADDR:
670 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
671 break;
672 case SO_REUSEPORT:
673 sk->sk_reuseport = valbool;
674 break;
675 case SO_TYPE:
676 case SO_PROTOCOL:
677 case SO_DOMAIN:
678 case SO_ERROR:
679 ret = -ENOPROTOOPT;
680 break;
681 case SO_DONTROUTE:
682 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
683 break;
684 case SO_BROADCAST:
685 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
686 break;
687 case SO_SNDBUF:
688 /* Don't error on this BSD doesn't and if you think
689 * about it this is right. Otherwise apps have to
690 * play 'guess the biggest size' games. RCVBUF/SNDBUF
691 * are treated in BSD as hints
692 */
693 val = min_t(u32, val, sysctl_wmem_max);
694 set_sndbuf:
695 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
696 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
697 /* Wake up sending tasks if we upped the value. */
698 sk->sk_write_space(sk);
699 break;
700
701 case SO_SNDBUFFORCE:
702 if (!capable(CAP_NET_ADMIN)) {
703 ret = -EPERM;
704 break;
705 }
706 goto set_sndbuf;
707
708 case SO_RCVBUF:
709 /* Don't error on this BSD doesn't and if you think
710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
713 */
714 val = min_t(u32, val, sysctl_rmem_max);
715 set_rcvbuf:
716 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
717 /*
718 * We double it on the way in to account for
719 * "struct sk_buff" etc. overhead. Applications
720 * assume that the SO_RCVBUF setting they make will
721 * allow that much actual data to be received on that
722 * socket.
723 *
724 * Applications are unaware that "struct sk_buff" and
725 * other overheads allocate from the receive buffer
726 * during socket buffer allocation.
727 *
728 * And after considering the possible alternatives,
729 * returning the value we actually used in getsockopt
730 * is the most desirable behavior.
731 */
732 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
733 break;
734
735 case SO_RCVBUFFORCE:
736 if (!capable(CAP_NET_ADMIN)) {
737 ret = -EPERM;
738 break;
739 }
740 goto set_rcvbuf;
741
742 case SO_KEEPALIVE:
743 #ifdef CONFIG_INET
744 if (sk->sk_protocol == IPPROTO_TCP &&
745 sk->sk_type == SOCK_STREAM)
746 tcp_set_keepalive(sk, valbool);
747 #endif
748 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
749 break;
750
751 case SO_OOBINLINE:
752 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
753 break;
754
755 case SO_NO_CHECK:
756 sk->sk_no_check = valbool;
757 break;
758
759 case SO_PRIORITY:
760 if ((val >= 0 && val <= 6) ||
761 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
762 sk->sk_priority = val;
763 else
764 ret = -EPERM;
765 break;
766
767 case SO_LINGER:
768 if (optlen < sizeof(ling)) {
769 ret = -EINVAL; /* 1003.1g */
770 break;
771 }
772 if (copy_from_user(&ling, optval, sizeof(ling))) {
773 ret = -EFAULT;
774 break;
775 }
776 if (!ling.l_onoff)
777 sock_reset_flag(sk, SOCK_LINGER);
778 else {
779 #if (BITS_PER_LONG == 32)
780 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
781 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
782 else
783 #endif
784 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
785 sock_set_flag(sk, SOCK_LINGER);
786 }
787 break;
788
789 case SO_BSDCOMPAT:
790 sock_warn_obsolete_bsdism("setsockopt");
791 break;
792
793 case SO_PASSCRED:
794 if (valbool)
795 set_bit(SOCK_PASSCRED, &sock->flags);
796 else
797 clear_bit(SOCK_PASSCRED, &sock->flags);
798 break;
799
800 case SO_TIMESTAMP:
801 case SO_TIMESTAMPNS:
802 if (valbool) {
803 if (optname == SO_TIMESTAMP)
804 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
805 else
806 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
807 sock_set_flag(sk, SOCK_RCVTSTAMP);
808 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
809 } else {
810 sock_reset_flag(sk, SOCK_RCVTSTAMP);
811 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
812 }
813 break;
814
815 case SO_TIMESTAMPING:
816 if (val & ~SOF_TIMESTAMPING_MASK) {
817 ret = -EINVAL;
818 break;
819 }
820 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
821 val & SOF_TIMESTAMPING_TX_HARDWARE);
822 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
823 val & SOF_TIMESTAMPING_TX_SOFTWARE);
824 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
825 val & SOF_TIMESTAMPING_RX_HARDWARE);
826 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
827 sock_enable_timestamp(sk,
828 SOCK_TIMESTAMPING_RX_SOFTWARE);
829 else
830 sock_disable_timestamp(sk,
831 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
832 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
833 val & SOF_TIMESTAMPING_SOFTWARE);
834 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
835 val & SOF_TIMESTAMPING_SYS_HARDWARE);
836 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
837 val & SOF_TIMESTAMPING_RAW_HARDWARE);
838 break;
839
840 case SO_RCVLOWAT:
841 if (val < 0)
842 val = INT_MAX;
843 sk->sk_rcvlowat = val ? : 1;
844 break;
845
846 case SO_RCVTIMEO:
847 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
848 break;
849
850 case SO_SNDTIMEO:
851 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
852 break;
853
854 case SO_ATTACH_FILTER:
855 ret = -EINVAL;
856 if (optlen == sizeof(struct sock_fprog)) {
857 struct sock_fprog fprog;
858
859 ret = -EFAULT;
860 if (copy_from_user(&fprog, optval, sizeof(fprog)))
861 break;
862
863 ret = sk_attach_filter(&fprog, sk);
864 }
865 break;
866
867 case SO_DETACH_FILTER:
868 ret = sk_detach_filter(sk);
869 break;
870
871 case SO_LOCK_FILTER:
872 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
873 ret = -EPERM;
874 else
875 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
876 break;
877
878 case SO_PASSSEC:
879 if (valbool)
880 set_bit(SOCK_PASSSEC, &sock->flags);
881 else
882 clear_bit(SOCK_PASSSEC, &sock->flags);
883 break;
884 case SO_MARK:
885 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
886 ret = -EPERM;
887 else
888 sk->sk_mark = val;
889 break;
890
891 /* We implement the SO_SNDLOWAT etc to
892 not be settable (1003.1g 5.3) */
893 case SO_RXQ_OVFL:
894 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
895 break;
896
897 case SO_WIFI_STATUS:
898 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
899 break;
900
901 case SO_PEEK_OFF:
902 if (sock->ops->set_peek_off)
903 sock->ops->set_peek_off(sk, val);
904 else
905 ret = -EOPNOTSUPP;
906 break;
907
908 case SO_NOFCS:
909 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
910 break;
911
912 case SO_SELECT_ERR_QUEUE:
913 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
914 break;
915
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 default:
1174 return -ENOPROTOOPT;
1175 }
1176
1177 if (len > lv)
1178 len = lv;
1179 if (copy_to_user(optval, &v, len))
1180 return -EFAULT;
1181 lenout:
1182 if (put_user(len, optlen))
1183 return -EFAULT;
1184 return 0;
1185 }
1186
1187 /*
1188 * Initialize an sk_lock.
1189 *
1190 * (We also register the sk_lock with the lock validator.)
1191 */
1192 static inline void sock_lock_init(struct sock *sk)
1193 {
1194 sock_lock_init_class_and_name(sk,
1195 af_family_slock_key_strings[sk->sk_family],
1196 af_family_slock_keys + sk->sk_family,
1197 af_family_key_strings[sk->sk_family],
1198 af_family_keys + sk->sk_family);
1199 }
1200
1201 /*
1202 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1203 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1204 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1205 */
1206 static void sock_copy(struct sock *nsk, const struct sock *osk)
1207 {
1208 #ifdef CONFIG_SECURITY_NETWORK
1209 void *sptr = nsk->sk_security;
1210 #endif
1211 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1212
1213 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1214 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1215
1216 #ifdef CONFIG_SECURITY_NETWORK
1217 nsk->sk_security = sptr;
1218 security_sk_clone(osk, nsk);
1219 #endif
1220 }
1221
1222 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1223 {
1224 unsigned long nulls1, nulls2;
1225
1226 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1227 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1228 if (nulls1 > nulls2)
1229 swap(nulls1, nulls2);
1230
1231 if (nulls1 != 0)
1232 memset((char *)sk, 0, nulls1);
1233 memset((char *)sk + nulls1 + sizeof(void *), 0,
1234 nulls2 - nulls1 - sizeof(void *));
1235 memset((char *)sk + nulls2 + sizeof(void *), 0,
1236 size - nulls2 - sizeof(void *));
1237 }
1238 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1239
1240 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1241 int family)
1242 {
1243 struct sock *sk;
1244 struct kmem_cache *slab;
1245
1246 slab = prot->slab;
1247 if (slab != NULL) {
1248 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1249 if (!sk)
1250 return sk;
1251 if (priority & __GFP_ZERO) {
1252 if (prot->clear_sk)
1253 prot->clear_sk(sk, prot->obj_size);
1254 else
1255 sk_prot_clear_nulls(sk, prot->obj_size);
1256 }
1257 } else
1258 sk = kmalloc(prot->obj_size, priority);
1259
1260 if (sk != NULL) {
1261 kmemcheck_annotate_bitfield(sk, flags);
1262
1263 if (security_sk_alloc(sk, family, priority))
1264 goto out_free;
1265
1266 if (!try_module_get(prot->owner))
1267 goto out_free_sec;
1268 sk_tx_queue_clear(sk);
1269 }
1270
1271 return sk;
1272
1273 out_free_sec:
1274 security_sk_free(sk);
1275 out_free:
1276 if (slab != NULL)
1277 kmem_cache_free(slab, sk);
1278 else
1279 kfree(sk);
1280 return NULL;
1281 }
1282
1283 static void sk_prot_free(struct proto *prot, struct sock *sk)
1284 {
1285 struct kmem_cache *slab;
1286 struct module *owner;
1287
1288 owner = prot->owner;
1289 slab = prot->slab;
1290
1291 security_sk_free(sk);
1292 if (slab != NULL)
1293 kmem_cache_free(slab, sk);
1294 else
1295 kfree(sk);
1296 module_put(owner);
1297 }
1298
1299 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1300 void sock_update_classid(struct sock *sk)
1301 {
1302 u32 classid;
1303
1304 classid = task_cls_classid(current);
1305 if (classid != sk->sk_classid)
1306 sk->sk_classid = classid;
1307 }
1308 EXPORT_SYMBOL(sock_update_classid);
1309 #endif
1310
1311 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1312 void sock_update_netprioidx(struct sock *sk)
1313 {
1314 if (in_interrupt())
1315 return;
1316
1317 sk->sk_cgrp_prioidx = task_netprioidx(current);
1318 }
1319 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1320 #endif
1321
1322 /**
1323 * sk_alloc - All socket objects are allocated here
1324 * @net: the applicable net namespace
1325 * @family: protocol family
1326 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1327 * @prot: struct proto associated with this new sock instance
1328 */
1329 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1330 struct proto *prot)
1331 {
1332 struct sock *sk;
1333
1334 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1335 if (sk) {
1336 sk->sk_family = family;
1337 /*
1338 * See comment in struct sock definition to understand
1339 * why we need sk_prot_creator -acme
1340 */
1341 sk->sk_prot = sk->sk_prot_creator = prot;
1342 sock_lock_init(sk);
1343 sock_net_set(sk, get_net(net));
1344 atomic_set(&sk->sk_wmem_alloc, 1);
1345
1346 sock_update_classid(sk);
1347 sock_update_netprioidx(sk);
1348 }
1349
1350 return sk;
1351 }
1352 EXPORT_SYMBOL(sk_alloc);
1353
1354 static void __sk_free(struct sock *sk)
1355 {
1356 struct sk_filter *filter;
1357
1358 if (sk->sk_destruct)
1359 sk->sk_destruct(sk);
1360
1361 filter = rcu_dereference_check(sk->sk_filter,
1362 atomic_read(&sk->sk_wmem_alloc) == 0);
1363 if (filter) {
1364 sk_filter_uncharge(sk, filter);
1365 RCU_INIT_POINTER(sk->sk_filter, NULL);
1366 }
1367
1368 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1369
1370 if (atomic_read(&sk->sk_omem_alloc))
1371 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1372 __func__, atomic_read(&sk->sk_omem_alloc));
1373
1374 if (sk->sk_peer_cred)
1375 put_cred(sk->sk_peer_cred);
1376 put_pid(sk->sk_peer_pid);
1377 put_net(sock_net(sk));
1378 sk_prot_free(sk->sk_prot_creator, sk);
1379 }
1380
1381 void sk_free(struct sock *sk)
1382 {
1383 /*
1384 * We subtract one from sk_wmem_alloc and can know if
1385 * some packets are still in some tx queue.
1386 * If not null, sock_wfree() will call __sk_free(sk) later
1387 */
1388 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1389 __sk_free(sk);
1390 }
1391 EXPORT_SYMBOL(sk_free);
1392
1393 /*
1394 * Last sock_put should drop reference to sk->sk_net. It has already
1395 * been dropped in sk_change_net. Taking reference to stopping namespace
1396 * is not an option.
1397 * Take reference to a socket to remove it from hash _alive_ and after that
1398 * destroy it in the context of init_net.
1399 */
1400 void sk_release_kernel(struct sock *sk)
1401 {
1402 if (sk == NULL || sk->sk_socket == NULL)
1403 return;
1404
1405 sock_hold(sk);
1406 sock_release(sk->sk_socket);
1407 release_net(sock_net(sk));
1408 sock_net_set(sk, get_net(&init_net));
1409 sock_put(sk);
1410 }
1411 EXPORT_SYMBOL(sk_release_kernel);
1412
1413 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1414 {
1415 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1416 sock_update_memcg(newsk);
1417 }
1418
1419 /**
1420 * sk_clone_lock - clone a socket, and lock its clone
1421 * @sk: the socket to clone
1422 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1423 *
1424 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1425 */
1426 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1427 {
1428 struct sock *newsk;
1429
1430 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1431 if (newsk != NULL) {
1432 struct sk_filter *filter;
1433
1434 sock_copy(newsk, sk);
1435
1436 /* SANITY */
1437 get_net(sock_net(newsk));
1438 sk_node_init(&newsk->sk_node);
1439 sock_lock_init(newsk);
1440 bh_lock_sock(newsk);
1441 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1442 newsk->sk_backlog.len = 0;
1443
1444 atomic_set(&newsk->sk_rmem_alloc, 0);
1445 /*
1446 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1447 */
1448 atomic_set(&newsk->sk_wmem_alloc, 1);
1449 atomic_set(&newsk->sk_omem_alloc, 0);
1450 skb_queue_head_init(&newsk->sk_receive_queue);
1451 skb_queue_head_init(&newsk->sk_write_queue);
1452 #ifdef CONFIG_NET_DMA
1453 skb_queue_head_init(&newsk->sk_async_wait_queue);
1454 #endif
1455
1456 spin_lock_init(&newsk->sk_dst_lock);
1457 rwlock_init(&newsk->sk_callback_lock);
1458 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1459 af_callback_keys + newsk->sk_family,
1460 af_family_clock_key_strings[newsk->sk_family]);
1461
1462 newsk->sk_dst_cache = NULL;
1463 newsk->sk_wmem_queued = 0;
1464 newsk->sk_forward_alloc = 0;
1465 newsk->sk_send_head = NULL;
1466 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1467
1468 sock_reset_flag(newsk, SOCK_DONE);
1469 skb_queue_head_init(&newsk->sk_error_queue);
1470
1471 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1472 if (filter != NULL)
1473 sk_filter_charge(newsk, filter);
1474
1475 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1476 /* It is still raw copy of parent, so invalidate
1477 * destructor and make plain sk_free() */
1478 newsk->sk_destruct = NULL;
1479 bh_unlock_sock(newsk);
1480 sk_free(newsk);
1481 newsk = NULL;
1482 goto out;
1483 }
1484
1485 newsk->sk_err = 0;
1486 newsk->sk_priority = 0;
1487 /*
1488 * Before updating sk_refcnt, we must commit prior changes to memory
1489 * (Documentation/RCU/rculist_nulls.txt for details)
1490 */
1491 smp_wmb();
1492 atomic_set(&newsk->sk_refcnt, 2);
1493
1494 /*
1495 * Increment the counter in the same struct proto as the master
1496 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1497 * is the same as sk->sk_prot->socks, as this field was copied
1498 * with memcpy).
1499 *
1500 * This _changes_ the previous behaviour, where
1501 * tcp_create_openreq_child always was incrementing the
1502 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1503 * to be taken into account in all callers. -acme
1504 */
1505 sk_refcnt_debug_inc(newsk);
1506 sk_set_socket(newsk, NULL);
1507 newsk->sk_wq = NULL;
1508
1509 sk_update_clone(sk, newsk);
1510
1511 if (newsk->sk_prot->sockets_allocated)
1512 sk_sockets_allocated_inc(newsk);
1513
1514 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1515 net_enable_timestamp();
1516 }
1517 out:
1518 return newsk;
1519 }
1520 EXPORT_SYMBOL_GPL(sk_clone_lock);
1521
1522 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1523 {
1524 __sk_dst_set(sk, dst);
1525 sk->sk_route_caps = dst->dev->features;
1526 if (sk->sk_route_caps & NETIF_F_GSO)
1527 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1528 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1529 if (sk_can_gso(sk)) {
1530 if (dst->header_len) {
1531 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1532 } else {
1533 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1534 sk->sk_gso_max_size = dst->dev->gso_max_size;
1535 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1536 }
1537 }
1538 }
1539 EXPORT_SYMBOL_GPL(sk_setup_caps);
1540
1541 /*
1542 * Simple resource managers for sockets.
1543 */
1544
1545
1546 /*
1547 * Write buffer destructor automatically called from kfree_skb.
1548 */
1549 void sock_wfree(struct sk_buff *skb)
1550 {
1551 struct sock *sk = skb->sk;
1552 unsigned int len = skb->truesize;
1553
1554 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1555 /*
1556 * Keep a reference on sk_wmem_alloc, this will be released
1557 * after sk_write_space() call
1558 */
1559 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1560 sk->sk_write_space(sk);
1561 len = 1;
1562 }
1563 /*
1564 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1565 * could not do because of in-flight packets
1566 */
1567 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1568 __sk_free(sk);
1569 }
1570 EXPORT_SYMBOL(sock_wfree);
1571
1572 /*
1573 * Read buffer destructor automatically called from kfree_skb.
1574 */
1575 void sock_rfree(struct sk_buff *skb)
1576 {
1577 struct sock *sk = skb->sk;
1578 unsigned int len = skb->truesize;
1579
1580 atomic_sub(len, &sk->sk_rmem_alloc);
1581 sk_mem_uncharge(sk, len);
1582 }
1583 EXPORT_SYMBOL(sock_rfree);
1584
1585 void sock_edemux(struct sk_buff *skb)
1586 {
1587 struct sock *sk = skb->sk;
1588
1589 #ifdef CONFIG_INET
1590 if (sk->sk_state == TCP_TIME_WAIT)
1591 inet_twsk_put(inet_twsk(sk));
1592 else
1593 #endif
1594 sock_put(sk);
1595 }
1596 EXPORT_SYMBOL(sock_edemux);
1597
1598 kuid_t sock_i_uid(struct sock *sk)
1599 {
1600 kuid_t uid;
1601
1602 read_lock_bh(&sk->sk_callback_lock);
1603 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1604 read_unlock_bh(&sk->sk_callback_lock);
1605 return uid;
1606 }
1607 EXPORT_SYMBOL(sock_i_uid);
1608
1609 unsigned long sock_i_ino(struct sock *sk)
1610 {
1611 unsigned long ino;
1612
1613 read_lock_bh(&sk->sk_callback_lock);
1614 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1615 read_unlock_bh(&sk->sk_callback_lock);
1616 return ino;
1617 }
1618 EXPORT_SYMBOL(sock_i_ino);
1619
1620 /*
1621 * Allocate a skb from the socket's send buffer.
1622 */
1623 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1624 gfp_t priority)
1625 {
1626 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1627 struct sk_buff *skb = alloc_skb(size, priority);
1628 if (skb) {
1629 skb_set_owner_w(skb, sk);
1630 return skb;
1631 }
1632 }
1633 return NULL;
1634 }
1635 EXPORT_SYMBOL(sock_wmalloc);
1636
1637 /*
1638 * Allocate a skb from the socket's receive buffer.
1639 */
1640 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1641 gfp_t priority)
1642 {
1643 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1644 struct sk_buff *skb = alloc_skb(size, priority);
1645 if (skb) {
1646 skb_set_owner_r(skb, sk);
1647 return skb;
1648 }
1649 }
1650 return NULL;
1651 }
1652
1653 /*
1654 * Allocate a memory block from the socket's option memory buffer.
1655 */
1656 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1657 {
1658 if ((unsigned int)size <= sysctl_optmem_max &&
1659 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1660 void *mem;
1661 /* First do the add, to avoid the race if kmalloc
1662 * might sleep.
1663 */
1664 atomic_add(size, &sk->sk_omem_alloc);
1665 mem = kmalloc(size, priority);
1666 if (mem)
1667 return mem;
1668 atomic_sub(size, &sk->sk_omem_alloc);
1669 }
1670 return NULL;
1671 }
1672 EXPORT_SYMBOL(sock_kmalloc);
1673
1674 /*
1675 * Free an option memory block.
1676 */
1677 void sock_kfree_s(struct sock *sk, void *mem, int size)
1678 {
1679 kfree(mem);
1680 atomic_sub(size, &sk->sk_omem_alloc);
1681 }
1682 EXPORT_SYMBOL(sock_kfree_s);
1683
1684 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1685 I think, these locks should be removed for datagram sockets.
1686 */
1687 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1688 {
1689 DEFINE_WAIT(wait);
1690
1691 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1692 for (;;) {
1693 if (!timeo)
1694 break;
1695 if (signal_pending(current))
1696 break;
1697 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1698 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1699 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1700 break;
1701 if (sk->sk_shutdown & SEND_SHUTDOWN)
1702 break;
1703 if (sk->sk_err)
1704 break;
1705 timeo = schedule_timeout(timeo);
1706 }
1707 finish_wait(sk_sleep(sk), &wait);
1708 return timeo;
1709 }
1710
1711
1712 /*
1713 * Generic send/receive buffer handlers
1714 */
1715
1716 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1717 unsigned long data_len, int noblock,
1718 int *errcode)
1719 {
1720 struct sk_buff *skb;
1721 gfp_t gfp_mask;
1722 long timeo;
1723 int err;
1724 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1725
1726 err = -EMSGSIZE;
1727 if (npages > MAX_SKB_FRAGS)
1728 goto failure;
1729
1730 gfp_mask = sk->sk_allocation;
1731 if (gfp_mask & __GFP_WAIT)
1732 gfp_mask |= __GFP_REPEAT;
1733
1734 timeo = sock_sndtimeo(sk, noblock);
1735 while (1) {
1736 err = sock_error(sk);
1737 if (err != 0)
1738 goto failure;
1739
1740 err = -EPIPE;
1741 if (sk->sk_shutdown & SEND_SHUTDOWN)
1742 goto failure;
1743
1744 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1745 skb = alloc_skb(header_len, gfp_mask);
1746 if (skb) {
1747 int i;
1748
1749 /* No pages, we're done... */
1750 if (!data_len)
1751 break;
1752
1753 skb->truesize += data_len;
1754 skb_shinfo(skb)->nr_frags = npages;
1755 for (i = 0; i < npages; i++) {
1756 struct page *page;
1757
1758 page = alloc_pages(sk->sk_allocation, 0);
1759 if (!page) {
1760 err = -ENOBUFS;
1761 skb_shinfo(skb)->nr_frags = i;
1762 kfree_skb(skb);
1763 goto failure;
1764 }
1765
1766 __skb_fill_page_desc(skb, i,
1767 page, 0,
1768 (data_len >= PAGE_SIZE ?
1769 PAGE_SIZE :
1770 data_len));
1771 data_len -= PAGE_SIZE;
1772 }
1773
1774 /* Full success... */
1775 break;
1776 }
1777 err = -ENOBUFS;
1778 goto failure;
1779 }
1780 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1781 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1782 err = -EAGAIN;
1783 if (!timeo)
1784 goto failure;
1785 if (signal_pending(current))
1786 goto interrupted;
1787 timeo = sock_wait_for_wmem(sk, timeo);
1788 }
1789
1790 skb_set_owner_w(skb, sk);
1791 return skb;
1792
1793 interrupted:
1794 err = sock_intr_errno(timeo);
1795 failure:
1796 *errcode = err;
1797 return NULL;
1798 }
1799 EXPORT_SYMBOL(sock_alloc_send_pskb);
1800
1801 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1802 int noblock, int *errcode)
1803 {
1804 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1805 }
1806 EXPORT_SYMBOL(sock_alloc_send_skb);
1807
1808 /* On 32bit arches, an skb frag is limited to 2^15 */
1809 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1810
1811 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1812 {
1813 int order;
1814
1815 if (pfrag->page) {
1816 if (atomic_read(&pfrag->page->_count) == 1) {
1817 pfrag->offset = 0;
1818 return true;
1819 }
1820 if (pfrag->offset < pfrag->size)
1821 return true;
1822 put_page(pfrag->page);
1823 }
1824
1825 /* We restrict high order allocations to users that can afford to wait */
1826 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1827
1828 do {
1829 gfp_t gfp = sk->sk_allocation;
1830
1831 if (order)
1832 gfp |= __GFP_COMP | __GFP_NOWARN;
1833 pfrag->page = alloc_pages(gfp, order);
1834 if (likely(pfrag->page)) {
1835 pfrag->offset = 0;
1836 pfrag->size = PAGE_SIZE << order;
1837 return true;
1838 }
1839 } while (--order >= 0);
1840
1841 sk_enter_memory_pressure(sk);
1842 sk_stream_moderate_sndbuf(sk);
1843 return false;
1844 }
1845 EXPORT_SYMBOL(sk_page_frag_refill);
1846
1847 static void __lock_sock(struct sock *sk)
1848 __releases(&sk->sk_lock.slock)
1849 __acquires(&sk->sk_lock.slock)
1850 {
1851 DEFINE_WAIT(wait);
1852
1853 for (;;) {
1854 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1855 TASK_UNINTERRUPTIBLE);
1856 spin_unlock_bh(&sk->sk_lock.slock);
1857 schedule();
1858 spin_lock_bh(&sk->sk_lock.slock);
1859 if (!sock_owned_by_user(sk))
1860 break;
1861 }
1862 finish_wait(&sk->sk_lock.wq, &wait);
1863 }
1864
1865 static void __release_sock(struct sock *sk)
1866 __releases(&sk->sk_lock.slock)
1867 __acquires(&sk->sk_lock.slock)
1868 {
1869 struct sk_buff *skb = sk->sk_backlog.head;
1870
1871 do {
1872 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1873 bh_unlock_sock(sk);
1874
1875 do {
1876 struct sk_buff *next = skb->next;
1877
1878 prefetch(next);
1879 WARN_ON_ONCE(skb_dst_is_noref(skb));
1880 skb->next = NULL;
1881 sk_backlog_rcv(sk, skb);
1882
1883 /*
1884 * We are in process context here with softirqs
1885 * disabled, use cond_resched_softirq() to preempt.
1886 * This is safe to do because we've taken the backlog
1887 * queue private:
1888 */
1889 cond_resched_softirq();
1890
1891 skb = next;
1892 } while (skb != NULL);
1893
1894 bh_lock_sock(sk);
1895 } while ((skb = sk->sk_backlog.head) != NULL);
1896
1897 /*
1898 * Doing the zeroing here guarantee we can not loop forever
1899 * while a wild producer attempts to flood us.
1900 */
1901 sk->sk_backlog.len = 0;
1902 }
1903
1904 /**
1905 * sk_wait_data - wait for data to arrive at sk_receive_queue
1906 * @sk: sock to wait on
1907 * @timeo: for how long
1908 *
1909 * Now socket state including sk->sk_err is changed only under lock,
1910 * hence we may omit checks after joining wait queue.
1911 * We check receive queue before schedule() only as optimization;
1912 * it is very likely that release_sock() added new data.
1913 */
1914 int sk_wait_data(struct sock *sk, long *timeo)
1915 {
1916 int rc;
1917 DEFINE_WAIT(wait);
1918
1919 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1920 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1921 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1922 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1923 finish_wait(sk_sleep(sk), &wait);
1924 return rc;
1925 }
1926 EXPORT_SYMBOL(sk_wait_data);
1927
1928 /**
1929 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1930 * @sk: socket
1931 * @size: memory size to allocate
1932 * @kind: allocation type
1933 *
1934 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1935 * rmem allocation. This function assumes that protocols which have
1936 * memory_pressure use sk_wmem_queued as write buffer accounting.
1937 */
1938 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1939 {
1940 struct proto *prot = sk->sk_prot;
1941 int amt = sk_mem_pages(size);
1942 long allocated;
1943 int parent_status = UNDER_LIMIT;
1944
1945 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1946
1947 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1948
1949 /* Under limit. */
1950 if (parent_status == UNDER_LIMIT &&
1951 allocated <= sk_prot_mem_limits(sk, 0)) {
1952 sk_leave_memory_pressure(sk);
1953 return 1;
1954 }
1955
1956 /* Under pressure. (we or our parents) */
1957 if ((parent_status > SOFT_LIMIT) ||
1958 allocated > sk_prot_mem_limits(sk, 1))
1959 sk_enter_memory_pressure(sk);
1960
1961 /* Over hard limit (we or our parents) */
1962 if ((parent_status == OVER_LIMIT) ||
1963 (allocated > sk_prot_mem_limits(sk, 2)))
1964 goto suppress_allocation;
1965
1966 /* guarantee minimum buffer size under pressure */
1967 if (kind == SK_MEM_RECV) {
1968 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1969 return 1;
1970
1971 } else { /* SK_MEM_SEND */
1972 if (sk->sk_type == SOCK_STREAM) {
1973 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1974 return 1;
1975 } else if (atomic_read(&sk->sk_wmem_alloc) <
1976 prot->sysctl_wmem[0])
1977 return 1;
1978 }
1979
1980 if (sk_has_memory_pressure(sk)) {
1981 int alloc;
1982
1983 if (!sk_under_memory_pressure(sk))
1984 return 1;
1985 alloc = sk_sockets_allocated_read_positive(sk);
1986 if (sk_prot_mem_limits(sk, 2) > alloc *
1987 sk_mem_pages(sk->sk_wmem_queued +
1988 atomic_read(&sk->sk_rmem_alloc) +
1989 sk->sk_forward_alloc))
1990 return 1;
1991 }
1992
1993 suppress_allocation:
1994
1995 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1996 sk_stream_moderate_sndbuf(sk);
1997
1998 /* Fail only if socket is _under_ its sndbuf.
1999 * In this case we cannot block, so that we have to fail.
2000 */
2001 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2002 return 1;
2003 }
2004
2005 trace_sock_exceed_buf_limit(sk, prot, allocated);
2006
2007 /* Alas. Undo changes. */
2008 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2009
2010 sk_memory_allocated_sub(sk, amt);
2011
2012 return 0;
2013 }
2014 EXPORT_SYMBOL(__sk_mem_schedule);
2015
2016 /**
2017 * __sk_reclaim - reclaim memory_allocated
2018 * @sk: socket
2019 */
2020 void __sk_mem_reclaim(struct sock *sk)
2021 {
2022 sk_memory_allocated_sub(sk,
2023 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2024 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2025
2026 if (sk_under_memory_pressure(sk) &&
2027 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2028 sk_leave_memory_pressure(sk);
2029 }
2030 EXPORT_SYMBOL(__sk_mem_reclaim);
2031
2032
2033 /*
2034 * Set of default routines for initialising struct proto_ops when
2035 * the protocol does not support a particular function. In certain
2036 * cases where it makes no sense for a protocol to have a "do nothing"
2037 * function, some default processing is provided.
2038 */
2039
2040 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2041 {
2042 return -EOPNOTSUPP;
2043 }
2044 EXPORT_SYMBOL(sock_no_bind);
2045
2046 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2047 int len, int flags)
2048 {
2049 return -EOPNOTSUPP;
2050 }
2051 EXPORT_SYMBOL(sock_no_connect);
2052
2053 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2054 {
2055 return -EOPNOTSUPP;
2056 }
2057 EXPORT_SYMBOL(sock_no_socketpair);
2058
2059 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2060 {
2061 return -EOPNOTSUPP;
2062 }
2063 EXPORT_SYMBOL(sock_no_accept);
2064
2065 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2066 int *len, int peer)
2067 {
2068 return -EOPNOTSUPP;
2069 }
2070 EXPORT_SYMBOL(sock_no_getname);
2071
2072 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2073 {
2074 return 0;
2075 }
2076 EXPORT_SYMBOL(sock_no_poll);
2077
2078 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2079 {
2080 return -EOPNOTSUPP;
2081 }
2082 EXPORT_SYMBOL(sock_no_ioctl);
2083
2084 int sock_no_listen(struct socket *sock, int backlog)
2085 {
2086 return -EOPNOTSUPP;
2087 }
2088 EXPORT_SYMBOL(sock_no_listen);
2089
2090 int sock_no_shutdown(struct socket *sock, int how)
2091 {
2092 return -EOPNOTSUPP;
2093 }
2094 EXPORT_SYMBOL(sock_no_shutdown);
2095
2096 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2097 char __user *optval, unsigned int optlen)
2098 {
2099 return -EOPNOTSUPP;
2100 }
2101 EXPORT_SYMBOL(sock_no_setsockopt);
2102
2103 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2104 char __user *optval, int __user *optlen)
2105 {
2106 return -EOPNOTSUPP;
2107 }
2108 EXPORT_SYMBOL(sock_no_getsockopt);
2109
2110 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2111 size_t len)
2112 {
2113 return -EOPNOTSUPP;
2114 }
2115 EXPORT_SYMBOL(sock_no_sendmsg);
2116
2117 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2118 size_t len, int flags)
2119 {
2120 return -EOPNOTSUPP;
2121 }
2122 EXPORT_SYMBOL(sock_no_recvmsg);
2123
2124 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2125 {
2126 /* Mirror missing mmap method error code */
2127 return -ENODEV;
2128 }
2129 EXPORT_SYMBOL(sock_no_mmap);
2130
2131 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2132 {
2133 ssize_t res;
2134 struct msghdr msg = {.msg_flags = flags};
2135 struct kvec iov;
2136 char *kaddr = kmap(page);
2137 iov.iov_base = kaddr + offset;
2138 iov.iov_len = size;
2139 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2140 kunmap(page);
2141 return res;
2142 }
2143 EXPORT_SYMBOL(sock_no_sendpage);
2144
2145 /*
2146 * Default Socket Callbacks
2147 */
2148
2149 static void sock_def_wakeup(struct sock *sk)
2150 {
2151 struct socket_wq *wq;
2152
2153 rcu_read_lock();
2154 wq = rcu_dereference(sk->sk_wq);
2155 if (wq_has_sleeper(wq))
2156 wake_up_interruptible_all(&wq->wait);
2157 rcu_read_unlock();
2158 }
2159
2160 static void sock_def_error_report(struct sock *sk)
2161 {
2162 struct socket_wq *wq;
2163
2164 rcu_read_lock();
2165 wq = rcu_dereference(sk->sk_wq);
2166 if (wq_has_sleeper(wq))
2167 wake_up_interruptible_poll(&wq->wait, POLLERR);
2168 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2169 rcu_read_unlock();
2170 }
2171
2172 static void sock_def_readable(struct sock *sk, int len)
2173 {
2174 struct socket_wq *wq;
2175
2176 rcu_read_lock();
2177 wq = rcu_dereference(sk->sk_wq);
2178 if (wq_has_sleeper(wq))
2179 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2180 POLLRDNORM | POLLRDBAND);
2181 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2182 rcu_read_unlock();
2183 }
2184
2185 static void sock_def_write_space(struct sock *sk)
2186 {
2187 struct socket_wq *wq;
2188
2189 rcu_read_lock();
2190
2191 /* Do not wake up a writer until he can make "significant"
2192 * progress. --DaveM
2193 */
2194 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2195 wq = rcu_dereference(sk->sk_wq);
2196 if (wq_has_sleeper(wq))
2197 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2198 POLLWRNORM | POLLWRBAND);
2199
2200 /* Should agree with poll, otherwise some programs break */
2201 if (sock_writeable(sk))
2202 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2203 }
2204
2205 rcu_read_unlock();
2206 }
2207
2208 static void sock_def_destruct(struct sock *sk)
2209 {
2210 kfree(sk->sk_protinfo);
2211 }
2212
2213 void sk_send_sigurg(struct sock *sk)
2214 {
2215 if (sk->sk_socket && sk->sk_socket->file)
2216 if (send_sigurg(&sk->sk_socket->file->f_owner))
2217 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2218 }
2219 EXPORT_SYMBOL(sk_send_sigurg);
2220
2221 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2222 unsigned long expires)
2223 {
2224 if (!mod_timer(timer, expires))
2225 sock_hold(sk);
2226 }
2227 EXPORT_SYMBOL(sk_reset_timer);
2228
2229 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2230 {
2231 if (del_timer(timer))
2232 __sock_put(sk);
2233 }
2234 EXPORT_SYMBOL(sk_stop_timer);
2235
2236 void sock_init_data(struct socket *sock, struct sock *sk)
2237 {
2238 skb_queue_head_init(&sk->sk_receive_queue);
2239 skb_queue_head_init(&sk->sk_write_queue);
2240 skb_queue_head_init(&sk->sk_error_queue);
2241 #ifdef CONFIG_NET_DMA
2242 skb_queue_head_init(&sk->sk_async_wait_queue);
2243 #endif
2244
2245 sk->sk_send_head = NULL;
2246
2247 init_timer(&sk->sk_timer);
2248
2249 sk->sk_allocation = GFP_KERNEL;
2250 sk->sk_rcvbuf = sysctl_rmem_default;
2251 sk->sk_sndbuf = sysctl_wmem_default;
2252 sk->sk_state = TCP_CLOSE;
2253 sk_set_socket(sk, sock);
2254
2255 sock_set_flag(sk, SOCK_ZAPPED);
2256
2257 if (sock) {
2258 sk->sk_type = sock->type;
2259 sk->sk_wq = sock->wq;
2260 sock->sk = sk;
2261 } else
2262 sk->sk_wq = NULL;
2263
2264 spin_lock_init(&sk->sk_dst_lock);
2265 rwlock_init(&sk->sk_callback_lock);
2266 lockdep_set_class_and_name(&sk->sk_callback_lock,
2267 af_callback_keys + sk->sk_family,
2268 af_family_clock_key_strings[sk->sk_family]);
2269
2270 sk->sk_state_change = sock_def_wakeup;
2271 sk->sk_data_ready = sock_def_readable;
2272 sk->sk_write_space = sock_def_write_space;
2273 sk->sk_error_report = sock_def_error_report;
2274 sk->sk_destruct = sock_def_destruct;
2275
2276 sk->sk_frag.page = NULL;
2277 sk->sk_frag.offset = 0;
2278 sk->sk_peek_off = -1;
2279
2280 sk->sk_peer_pid = NULL;
2281 sk->sk_peer_cred = NULL;
2282 sk->sk_write_pending = 0;
2283 sk->sk_rcvlowat = 1;
2284 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2285 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2286
2287 sk->sk_stamp = ktime_set(-1L, 0);
2288
2289 #ifdef CONFIG_NET_LL_RX_POLL
2290 sk->sk_napi_id = 0;
2291 #endif
2292
2293 /*
2294 * Before updating sk_refcnt, we must commit prior changes to memory
2295 * (Documentation/RCU/rculist_nulls.txt for details)
2296 */
2297 smp_wmb();
2298 atomic_set(&sk->sk_refcnt, 1);
2299 atomic_set(&sk->sk_drops, 0);
2300 }
2301 EXPORT_SYMBOL(sock_init_data);
2302
2303 void lock_sock_nested(struct sock *sk, int subclass)
2304 {
2305 might_sleep();
2306 spin_lock_bh(&sk->sk_lock.slock);
2307 if (sk->sk_lock.owned)
2308 __lock_sock(sk);
2309 sk->sk_lock.owned = 1;
2310 spin_unlock(&sk->sk_lock.slock);
2311 /*
2312 * The sk_lock has mutex_lock() semantics here:
2313 */
2314 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2315 local_bh_enable();
2316 }
2317 EXPORT_SYMBOL(lock_sock_nested);
2318
2319 void release_sock(struct sock *sk)
2320 {
2321 /*
2322 * The sk_lock has mutex_unlock() semantics:
2323 */
2324 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2325
2326 spin_lock_bh(&sk->sk_lock.slock);
2327 if (sk->sk_backlog.tail)
2328 __release_sock(sk);
2329
2330 if (sk->sk_prot->release_cb)
2331 sk->sk_prot->release_cb(sk);
2332
2333 sk->sk_lock.owned = 0;
2334 if (waitqueue_active(&sk->sk_lock.wq))
2335 wake_up(&sk->sk_lock.wq);
2336 spin_unlock_bh(&sk->sk_lock.slock);
2337 }
2338 EXPORT_SYMBOL(release_sock);
2339
2340 /**
2341 * lock_sock_fast - fast version of lock_sock
2342 * @sk: socket
2343 *
2344 * This version should be used for very small section, where process wont block
2345 * return false if fast path is taken
2346 * sk_lock.slock locked, owned = 0, BH disabled
2347 * return true if slow path is taken
2348 * sk_lock.slock unlocked, owned = 1, BH enabled
2349 */
2350 bool lock_sock_fast(struct sock *sk)
2351 {
2352 might_sleep();
2353 spin_lock_bh(&sk->sk_lock.slock);
2354
2355 if (!sk->sk_lock.owned)
2356 /*
2357 * Note : We must disable BH
2358 */
2359 return false;
2360
2361 __lock_sock(sk);
2362 sk->sk_lock.owned = 1;
2363 spin_unlock(&sk->sk_lock.slock);
2364 /*
2365 * The sk_lock has mutex_lock() semantics here:
2366 */
2367 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2368 local_bh_enable();
2369 return true;
2370 }
2371 EXPORT_SYMBOL(lock_sock_fast);
2372
2373 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2374 {
2375 struct timeval tv;
2376 if (!sock_flag(sk, SOCK_TIMESTAMP))
2377 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2378 tv = ktime_to_timeval(sk->sk_stamp);
2379 if (tv.tv_sec == -1)
2380 return -ENOENT;
2381 if (tv.tv_sec == 0) {
2382 sk->sk_stamp = ktime_get_real();
2383 tv = ktime_to_timeval(sk->sk_stamp);
2384 }
2385 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2386 }
2387 EXPORT_SYMBOL(sock_get_timestamp);
2388
2389 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2390 {
2391 struct timespec ts;
2392 if (!sock_flag(sk, SOCK_TIMESTAMP))
2393 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2394 ts = ktime_to_timespec(sk->sk_stamp);
2395 if (ts.tv_sec == -1)
2396 return -ENOENT;
2397 if (ts.tv_sec == 0) {
2398 sk->sk_stamp = ktime_get_real();
2399 ts = ktime_to_timespec(sk->sk_stamp);
2400 }
2401 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2402 }
2403 EXPORT_SYMBOL(sock_get_timestampns);
2404
2405 void sock_enable_timestamp(struct sock *sk, int flag)
2406 {
2407 if (!sock_flag(sk, flag)) {
2408 unsigned long previous_flags = sk->sk_flags;
2409
2410 sock_set_flag(sk, flag);
2411 /*
2412 * we just set one of the two flags which require net
2413 * time stamping, but time stamping might have been on
2414 * already because of the other one
2415 */
2416 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2417 net_enable_timestamp();
2418 }
2419 }
2420
2421 /*
2422 * Get a socket option on an socket.
2423 *
2424 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2425 * asynchronous errors should be reported by getsockopt. We assume
2426 * this means if you specify SO_ERROR (otherwise whats the point of it).
2427 */
2428 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2429 char __user *optval, int __user *optlen)
2430 {
2431 struct sock *sk = sock->sk;
2432
2433 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2434 }
2435 EXPORT_SYMBOL(sock_common_getsockopt);
2436
2437 #ifdef CONFIG_COMPAT
2438 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2439 char __user *optval, int __user *optlen)
2440 {
2441 struct sock *sk = sock->sk;
2442
2443 if (sk->sk_prot->compat_getsockopt != NULL)
2444 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2445 optval, optlen);
2446 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2447 }
2448 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2449 #endif
2450
2451 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2452 struct msghdr *msg, size_t size, int flags)
2453 {
2454 struct sock *sk = sock->sk;
2455 int addr_len = 0;
2456 int err;
2457
2458 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2459 flags & ~MSG_DONTWAIT, &addr_len);
2460 if (err >= 0)
2461 msg->msg_namelen = addr_len;
2462 return err;
2463 }
2464 EXPORT_SYMBOL(sock_common_recvmsg);
2465
2466 /*
2467 * Set socket options on an inet socket.
2468 */
2469 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2470 char __user *optval, unsigned int optlen)
2471 {
2472 struct sock *sk = sock->sk;
2473
2474 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2475 }
2476 EXPORT_SYMBOL(sock_common_setsockopt);
2477
2478 #ifdef CONFIG_COMPAT
2479 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2480 char __user *optval, unsigned int optlen)
2481 {
2482 struct sock *sk = sock->sk;
2483
2484 if (sk->sk_prot->compat_setsockopt != NULL)
2485 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2486 optval, optlen);
2487 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2488 }
2489 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2490 #endif
2491
2492 void sk_common_release(struct sock *sk)
2493 {
2494 if (sk->sk_prot->destroy)
2495 sk->sk_prot->destroy(sk);
2496
2497 /*
2498 * Observation: when sock_common_release is called, processes have
2499 * no access to socket. But net still has.
2500 * Step one, detach it from networking:
2501 *
2502 * A. Remove from hash tables.
2503 */
2504
2505 sk->sk_prot->unhash(sk);
2506
2507 /*
2508 * In this point socket cannot receive new packets, but it is possible
2509 * that some packets are in flight because some CPU runs receiver and
2510 * did hash table lookup before we unhashed socket. They will achieve
2511 * receive queue and will be purged by socket destructor.
2512 *
2513 * Also we still have packets pending on receive queue and probably,
2514 * our own packets waiting in device queues. sock_destroy will drain
2515 * receive queue, but transmitted packets will delay socket destruction
2516 * until the last reference will be released.
2517 */
2518
2519 sock_orphan(sk);
2520
2521 xfrm_sk_free_policy(sk);
2522
2523 sk_refcnt_debug_release(sk);
2524
2525 if (sk->sk_frag.page) {
2526 put_page(sk->sk_frag.page);
2527 sk->sk_frag.page = NULL;
2528 }
2529
2530 sock_put(sk);
2531 }
2532 EXPORT_SYMBOL(sk_common_release);
2533
2534 #ifdef CONFIG_PROC_FS
2535 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2536 struct prot_inuse {
2537 int val[PROTO_INUSE_NR];
2538 };
2539
2540 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2541
2542 #ifdef CONFIG_NET_NS
2543 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2544 {
2545 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2546 }
2547 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2548
2549 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2550 {
2551 int cpu, idx = prot->inuse_idx;
2552 int res = 0;
2553
2554 for_each_possible_cpu(cpu)
2555 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2556
2557 return res >= 0 ? res : 0;
2558 }
2559 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2560
2561 static int __net_init sock_inuse_init_net(struct net *net)
2562 {
2563 net->core.inuse = alloc_percpu(struct prot_inuse);
2564 return net->core.inuse ? 0 : -ENOMEM;
2565 }
2566
2567 static void __net_exit sock_inuse_exit_net(struct net *net)
2568 {
2569 free_percpu(net->core.inuse);
2570 }
2571
2572 static struct pernet_operations net_inuse_ops = {
2573 .init = sock_inuse_init_net,
2574 .exit = sock_inuse_exit_net,
2575 };
2576
2577 static __init int net_inuse_init(void)
2578 {
2579 if (register_pernet_subsys(&net_inuse_ops))
2580 panic("Cannot initialize net inuse counters");
2581
2582 return 0;
2583 }
2584
2585 core_initcall(net_inuse_init);
2586 #else
2587 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2588
2589 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2590 {
2591 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2592 }
2593 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2594
2595 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2596 {
2597 int cpu, idx = prot->inuse_idx;
2598 int res = 0;
2599
2600 for_each_possible_cpu(cpu)
2601 res += per_cpu(prot_inuse, cpu).val[idx];
2602
2603 return res >= 0 ? res : 0;
2604 }
2605 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2606 #endif
2607
2608 static void assign_proto_idx(struct proto *prot)
2609 {
2610 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2611
2612 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2613 pr_err("PROTO_INUSE_NR exhausted\n");
2614 return;
2615 }
2616
2617 set_bit(prot->inuse_idx, proto_inuse_idx);
2618 }
2619
2620 static void release_proto_idx(struct proto *prot)
2621 {
2622 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2623 clear_bit(prot->inuse_idx, proto_inuse_idx);
2624 }
2625 #else
2626 static inline void assign_proto_idx(struct proto *prot)
2627 {
2628 }
2629
2630 static inline void release_proto_idx(struct proto *prot)
2631 {
2632 }
2633 #endif
2634
2635 int proto_register(struct proto *prot, int alloc_slab)
2636 {
2637 if (alloc_slab) {
2638 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2639 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2640 NULL);
2641
2642 if (prot->slab == NULL) {
2643 pr_crit("%s: Can't create sock SLAB cache!\n",
2644 prot->name);
2645 goto out;
2646 }
2647
2648 if (prot->rsk_prot != NULL) {
2649 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2650 if (prot->rsk_prot->slab_name == NULL)
2651 goto out_free_sock_slab;
2652
2653 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2654 prot->rsk_prot->obj_size, 0,
2655 SLAB_HWCACHE_ALIGN, NULL);
2656
2657 if (prot->rsk_prot->slab == NULL) {
2658 pr_crit("%s: Can't create request sock SLAB cache!\n",
2659 prot->name);
2660 goto out_free_request_sock_slab_name;
2661 }
2662 }
2663
2664 if (prot->twsk_prot != NULL) {
2665 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2666
2667 if (prot->twsk_prot->twsk_slab_name == NULL)
2668 goto out_free_request_sock_slab;
2669
2670 prot->twsk_prot->twsk_slab =
2671 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2672 prot->twsk_prot->twsk_obj_size,
2673 0,
2674 SLAB_HWCACHE_ALIGN |
2675 prot->slab_flags,
2676 NULL);
2677 if (prot->twsk_prot->twsk_slab == NULL)
2678 goto out_free_timewait_sock_slab_name;
2679 }
2680 }
2681
2682 mutex_lock(&proto_list_mutex);
2683 list_add(&prot->node, &proto_list);
2684 assign_proto_idx(prot);
2685 mutex_unlock(&proto_list_mutex);
2686 return 0;
2687
2688 out_free_timewait_sock_slab_name:
2689 kfree(prot->twsk_prot->twsk_slab_name);
2690 out_free_request_sock_slab:
2691 if (prot->rsk_prot && prot->rsk_prot->slab) {
2692 kmem_cache_destroy(prot->rsk_prot->slab);
2693 prot->rsk_prot->slab = NULL;
2694 }
2695 out_free_request_sock_slab_name:
2696 if (prot->rsk_prot)
2697 kfree(prot->rsk_prot->slab_name);
2698 out_free_sock_slab:
2699 kmem_cache_destroy(prot->slab);
2700 prot->slab = NULL;
2701 out:
2702 return -ENOBUFS;
2703 }
2704 EXPORT_SYMBOL(proto_register);
2705
2706 void proto_unregister(struct proto *prot)
2707 {
2708 mutex_lock(&proto_list_mutex);
2709 release_proto_idx(prot);
2710 list_del(&prot->node);
2711 mutex_unlock(&proto_list_mutex);
2712
2713 if (prot->slab != NULL) {
2714 kmem_cache_destroy(prot->slab);
2715 prot->slab = NULL;
2716 }
2717
2718 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2719 kmem_cache_destroy(prot->rsk_prot->slab);
2720 kfree(prot->rsk_prot->slab_name);
2721 prot->rsk_prot->slab = NULL;
2722 }
2723
2724 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2725 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2726 kfree(prot->twsk_prot->twsk_slab_name);
2727 prot->twsk_prot->twsk_slab = NULL;
2728 }
2729 }
2730 EXPORT_SYMBOL(proto_unregister);
2731
2732 #ifdef CONFIG_PROC_FS
2733 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2734 __acquires(proto_list_mutex)
2735 {
2736 mutex_lock(&proto_list_mutex);
2737 return seq_list_start_head(&proto_list, *pos);
2738 }
2739
2740 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2741 {
2742 return seq_list_next(v, &proto_list, pos);
2743 }
2744
2745 static void proto_seq_stop(struct seq_file *seq, void *v)
2746 __releases(proto_list_mutex)
2747 {
2748 mutex_unlock(&proto_list_mutex);
2749 }
2750
2751 static char proto_method_implemented(const void *method)
2752 {
2753 return method == NULL ? 'n' : 'y';
2754 }
2755 static long sock_prot_memory_allocated(struct proto *proto)
2756 {
2757 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2758 }
2759
2760 static char *sock_prot_memory_pressure(struct proto *proto)
2761 {
2762 return proto->memory_pressure != NULL ?
2763 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2764 }
2765
2766 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2767 {
2768
2769 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2770 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2771 proto->name,
2772 proto->obj_size,
2773 sock_prot_inuse_get(seq_file_net(seq), proto),
2774 sock_prot_memory_allocated(proto),
2775 sock_prot_memory_pressure(proto),
2776 proto->max_header,
2777 proto->slab == NULL ? "no" : "yes",
2778 module_name(proto->owner),
2779 proto_method_implemented(proto->close),
2780 proto_method_implemented(proto->connect),
2781 proto_method_implemented(proto->disconnect),
2782 proto_method_implemented(proto->accept),
2783 proto_method_implemented(proto->ioctl),
2784 proto_method_implemented(proto->init),
2785 proto_method_implemented(proto->destroy),
2786 proto_method_implemented(proto->shutdown),
2787 proto_method_implemented(proto->setsockopt),
2788 proto_method_implemented(proto->getsockopt),
2789 proto_method_implemented(proto->sendmsg),
2790 proto_method_implemented(proto->recvmsg),
2791 proto_method_implemented(proto->sendpage),
2792 proto_method_implemented(proto->bind),
2793 proto_method_implemented(proto->backlog_rcv),
2794 proto_method_implemented(proto->hash),
2795 proto_method_implemented(proto->unhash),
2796 proto_method_implemented(proto->get_port),
2797 proto_method_implemented(proto->enter_memory_pressure));
2798 }
2799
2800 static int proto_seq_show(struct seq_file *seq, void *v)
2801 {
2802 if (v == &proto_list)
2803 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2804 "protocol",
2805 "size",
2806 "sockets",
2807 "memory",
2808 "press",
2809 "maxhdr",
2810 "slab",
2811 "module",
2812 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2813 else
2814 proto_seq_printf(seq, list_entry(v, struct proto, node));
2815 return 0;
2816 }
2817
2818 static const struct seq_operations proto_seq_ops = {
2819 .start = proto_seq_start,
2820 .next = proto_seq_next,
2821 .stop = proto_seq_stop,
2822 .show = proto_seq_show,
2823 };
2824
2825 static int proto_seq_open(struct inode *inode, struct file *file)
2826 {
2827 return seq_open_net(inode, file, &proto_seq_ops,
2828 sizeof(struct seq_net_private));
2829 }
2830
2831 static const struct file_operations proto_seq_fops = {
2832 .owner = THIS_MODULE,
2833 .open = proto_seq_open,
2834 .read = seq_read,
2835 .llseek = seq_lseek,
2836 .release = seq_release_net,
2837 };
2838
2839 static __net_init int proto_init_net(struct net *net)
2840 {
2841 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2842 return -ENOMEM;
2843
2844 return 0;
2845 }
2846
2847 static __net_exit void proto_exit_net(struct net *net)
2848 {
2849 remove_proc_entry("protocols", net->proc_net);
2850 }
2851
2852
2853 static __net_initdata struct pernet_operations proto_net_ops = {
2854 .init = proto_init_net,
2855 .exit = proto_exit_net,
2856 };
2857
2858 static int __init proto_init(void)
2859 {
2860 return register_pernet_subsys(&proto_net_ops);
2861 }
2862
2863 subsys_initcall(proto_init);
2864
2865 #endif /* PROC_FS */
Linus' kernel tree