?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
cgroup: net_cls: Remove rcu_read_lock/unlock
[linux-2.6.git] / net / core / sock.c
blob9fedbbfb0708022cd7547866e06ffbdac9701456
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 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
144
145 #ifdef CONFIG_MEMCG_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
147 {
148 struct proto *proto;
149 int ret = 0;
150
151 mutex_lock(&proto_list_mutex);
152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
154 ret = proto->init_cgroup(memcg, ss);
155 if (ret)
156 goto out;
157 }
158 }
159
160 mutex_unlock(&proto_list_mutex);
161 return ret;
162 out:
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
165 proto->destroy_cgroup(memcg);
166 mutex_unlock(&proto_list_mutex);
167 return ret;
168 }
169
170 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
171 {
172 struct proto *proto;
173
174 mutex_lock(&proto_list_mutex);
175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
177 proto->destroy_cgroup(memcg);
178 mutex_unlock(&proto_list_mutex);
179 }
180 #endif
181
182 /*
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
185 */
186 static struct lock_class_key af_family_keys[AF_MAX];
187 static struct lock_class_key af_family_slock_keys[AF_MAX];
188
189 struct static_key memcg_socket_limit_enabled;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled);
191
192 /*
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
195 * locks is fast):
196 */
197 static const char *const af_family_key_strings[AF_MAX+1] = {
198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
212 };
213 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
227 "slock-AF_NFC" , "slock-AF_MAX"
228 };
229 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
243 "clock-AF_NFC" , "clock-AF_MAX"
244 };
245
246 /*
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
249 */
250 static struct lock_class_key af_callback_keys[AF_MAX];
251
252 /* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
256 */
257 #define _SK_MEM_PACKETS 256
258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264 EXPORT_SYMBOL(sysctl_wmem_max);
265 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266 EXPORT_SYMBOL(sysctl_rmem_max);
267 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
269
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272 EXPORT_SYMBOL(sysctl_optmem_max);
273
274 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
275 EXPORT_SYMBOL_GPL(memalloc_socks);
276
277 /**
278 * sk_set_memalloc - sets %SOCK_MEMALLOC
279 * @sk: socket to set it on
280 *
281 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
282 * It's the responsibility of the admin to adjust min_free_kbytes
283 * to meet the requirements
284 */
285 void sk_set_memalloc(struct sock *sk)
286 {
287 sock_set_flag(sk, SOCK_MEMALLOC);
288 sk->sk_allocation |= __GFP_MEMALLOC;
289 static_key_slow_inc(&memalloc_socks);
290 }
291 EXPORT_SYMBOL_GPL(sk_set_memalloc);
292
293 void sk_clear_memalloc(struct sock *sk)
294 {
295 sock_reset_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation &= ~__GFP_MEMALLOC;
297 static_key_slow_dec(&memalloc_socks);
298
299 /*
300 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
301 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
302 * it has rmem allocations there is a risk that the user of the
303 * socket cannot make forward progress due to exceeding the rmem
304 * limits. By rights, sk_clear_memalloc() should only be called
305 * on sockets being torn down but warn and reset the accounting if
306 * that assumption breaks.
307 */
308 if (WARN_ON(sk->sk_forward_alloc))
309 sk_mem_reclaim(sk);
310 }
311 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
312
313 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
314 {
315 int ret;
316 unsigned long pflags = current->flags;
317
318 /* these should have been dropped before queueing */
319 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
320
321 current->flags |= PF_MEMALLOC;
322 ret = sk->sk_backlog_rcv(sk, skb);
323 tsk_restore_flags(current, pflags, PF_MEMALLOC);
324
325 return ret;
326 }
327 EXPORT_SYMBOL(__sk_backlog_rcv);
328
329 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
330 {
331 struct timeval tv;
332
333 if (optlen < sizeof(tv))
334 return -EINVAL;
335 if (copy_from_user(&tv, optval, sizeof(tv)))
336 return -EFAULT;
337 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
338 return -EDOM;
339
340 if (tv.tv_sec < 0) {
341 static int warned __read_mostly;
342
343 *timeo_p = 0;
344 if (warned < 10 && net_ratelimit()) {
345 warned++;
346 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
347 __func__, current->comm, task_pid_nr(current));
348 }
349 return 0;
350 }
351 *timeo_p = MAX_SCHEDULE_TIMEOUT;
352 if (tv.tv_sec == 0 && tv.tv_usec == 0)
353 return 0;
354 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
355 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
356 return 0;
357 }
358
359 static void sock_warn_obsolete_bsdism(const char *name)
360 {
361 static int warned;
362 static char warncomm[TASK_COMM_LEN];
363 if (strcmp(warncomm, current->comm) && warned < 5) {
364 strcpy(warncomm, current->comm);
365 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
366 warncomm, name);
367 warned++;
368 }
369 }
370
371 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
372
373 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
374 {
375 if (sk->sk_flags & flags) {
376 sk->sk_flags &= ~flags;
377 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
378 net_disable_timestamp();
379 }
380 }
381
382
383 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
384 {
385 int err;
386 int skb_len;
387 unsigned long flags;
388 struct sk_buff_head *list = &sk->sk_receive_queue;
389
390 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
391 atomic_inc(&sk->sk_drops);
392 trace_sock_rcvqueue_full(sk, skb);
393 return -ENOMEM;
394 }
395
396 err = sk_filter(sk, skb);
397 if (err)
398 return err;
399
400 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
401 atomic_inc(&sk->sk_drops);
402 return -ENOBUFS;
403 }
404
405 skb->dev = NULL;
406 skb_set_owner_r(skb, sk);
407
408 /* Cache the SKB length before we tack it onto the receive
409 * queue. Once it is added it no longer belongs to us and
410 * may be freed by other threads of control pulling packets
411 * from the queue.
412 */
413 skb_len = skb->len;
414
415 /* we escape from rcu protected region, make sure we dont leak
416 * a norefcounted dst
417 */
418 skb_dst_force(skb);
419
420 spin_lock_irqsave(&list->lock, flags);
421 skb->dropcount = atomic_read(&sk->sk_drops);
422 __skb_queue_tail(list, skb);
423 spin_unlock_irqrestore(&list->lock, flags);
424
425 if (!sock_flag(sk, SOCK_DEAD))
426 sk->sk_data_ready(sk, skb_len);
427 return 0;
428 }
429 EXPORT_SYMBOL(sock_queue_rcv_skb);
430
431 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
432 {
433 int rc = NET_RX_SUCCESS;
434
435 if (sk_filter(sk, skb))
436 goto discard_and_relse;
437
438 skb->dev = NULL;
439
440 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
441 atomic_inc(&sk->sk_drops);
442 goto discard_and_relse;
443 }
444 if (nested)
445 bh_lock_sock_nested(sk);
446 else
447 bh_lock_sock(sk);
448 if (!sock_owned_by_user(sk)) {
449 /*
450 * trylock + unlock semantics:
451 */
452 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
453
454 rc = sk_backlog_rcv(sk, skb);
455
456 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
457 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
458 bh_unlock_sock(sk);
459 atomic_inc(&sk->sk_drops);
460 goto discard_and_relse;
461 }
462
463 bh_unlock_sock(sk);
464 out:
465 sock_put(sk);
466 return rc;
467 discard_and_relse:
468 kfree_skb(skb);
469 goto out;
470 }
471 EXPORT_SYMBOL(sk_receive_skb);
472
473 void sk_reset_txq(struct sock *sk)
474 {
475 sk_tx_queue_clear(sk);
476 }
477 EXPORT_SYMBOL(sk_reset_txq);
478
479 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
480 {
481 struct dst_entry *dst = __sk_dst_get(sk);
482
483 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
484 sk_tx_queue_clear(sk);
485 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
486 dst_release(dst);
487 return NULL;
488 }
489
490 return dst;
491 }
492 EXPORT_SYMBOL(__sk_dst_check);
493
494 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
495 {
496 struct dst_entry *dst = sk_dst_get(sk);
497
498 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
499 sk_dst_reset(sk);
500 dst_release(dst);
501 return NULL;
502 }
503
504 return dst;
505 }
506 EXPORT_SYMBOL(sk_dst_check);
507
508 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
509 {
510 int ret = -ENOPROTOOPT;
511 #ifdef CONFIG_NETDEVICES
512 struct net *net = sock_net(sk);
513 char devname[IFNAMSIZ];
514 int index;
515
516 /* Sorry... */
517 ret = -EPERM;
518 if (!capable(CAP_NET_RAW))
519 goto out;
520
521 ret = -EINVAL;
522 if (optlen < 0)
523 goto out;
524
525 /* Bind this socket to a particular device like "eth0",
526 * as specified in the passed interface name. If the
527 * name is "" or the option length is zero the socket
528 * is not bound.
529 */
530 if (optlen > IFNAMSIZ - 1)
531 optlen = IFNAMSIZ - 1;
532 memset(devname, 0, sizeof(devname));
533
534 ret = -EFAULT;
535 if (copy_from_user(devname, optval, optlen))
536 goto out;
537
538 index = 0;
539 if (devname[0] != '\0') {
540 struct net_device *dev;
541
542 rcu_read_lock();
543 dev = dev_get_by_name_rcu(net, devname);
544 if (dev)
545 index = dev->ifindex;
546 rcu_read_unlock();
547 ret = -ENODEV;
548 if (!dev)
549 goto out;
550 }
551
552 lock_sock(sk);
553 sk->sk_bound_dev_if = index;
554 sk_dst_reset(sk);
555 release_sock(sk);
556
557 ret = 0;
558
559 out:
560 #endif
561
562 return ret;
563 }
564
565 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
566 {
567 if (valbool)
568 sock_set_flag(sk, bit);
569 else
570 sock_reset_flag(sk, bit);
571 }
572
573 /*
574 * This is meant for all protocols to use and covers goings on
575 * at the socket level. Everything here is generic.
576 */
577
578 int sock_setsockopt(struct socket *sock, int level, int optname,
579 char __user *optval, unsigned int optlen)
580 {
581 struct sock *sk = sock->sk;
582 int val;
583 int valbool;
584 struct linger ling;
585 int ret = 0;
586
587 /*
588 * Options without arguments
589 */
590
591 if (optname == SO_BINDTODEVICE)
592 return sock_bindtodevice(sk, optval, optlen);
593
594 if (optlen < sizeof(int))
595 return -EINVAL;
596
597 if (get_user(val, (int __user *)optval))
598 return -EFAULT;
599
600 valbool = val ? 1 : 0;
601
602 lock_sock(sk);
603
604 switch (optname) {
605 case SO_DEBUG:
606 if (val && !capable(CAP_NET_ADMIN))
607 ret = -EACCES;
608 else
609 sock_valbool_flag(sk, SOCK_DBG, valbool);
610 break;
611 case SO_REUSEADDR:
612 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
613 break;
614 case SO_TYPE:
615 case SO_PROTOCOL:
616 case SO_DOMAIN:
617 case SO_ERROR:
618 ret = -ENOPROTOOPT;
619 break;
620 case SO_DONTROUTE:
621 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
622 break;
623 case SO_BROADCAST:
624 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
625 break;
626 case SO_SNDBUF:
627 /* Don't error on this BSD doesn't and if you think
628 * about it this is right. Otherwise apps have to
629 * play 'guess the biggest size' games. RCVBUF/SNDBUF
630 * are treated in BSD as hints
631 */
632 val = min_t(u32, val, sysctl_wmem_max);
633 set_sndbuf:
634 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
635 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
636 /* Wake up sending tasks if we upped the value. */
637 sk->sk_write_space(sk);
638 break;
639
640 case SO_SNDBUFFORCE:
641 if (!capable(CAP_NET_ADMIN)) {
642 ret = -EPERM;
643 break;
644 }
645 goto set_sndbuf;
646
647 case SO_RCVBUF:
648 /* Don't error on this BSD doesn't and if you think
649 * about it this is right. Otherwise apps have to
650 * play 'guess the biggest size' games. RCVBUF/SNDBUF
651 * are treated in BSD as hints
652 */
653 val = min_t(u32, val, sysctl_rmem_max);
654 set_rcvbuf:
655 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
656 /*
657 * We double it on the way in to account for
658 * "struct sk_buff" etc. overhead. Applications
659 * assume that the SO_RCVBUF setting they make will
660 * allow that much actual data to be received on that
661 * socket.
662 *
663 * Applications are unaware that "struct sk_buff" and
664 * other overheads allocate from the receive buffer
665 * during socket buffer allocation.
666 *
667 * And after considering the possible alternatives,
668 * returning the value we actually used in getsockopt
669 * is the most desirable behavior.
670 */
671 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
672 break;
673
674 case SO_RCVBUFFORCE:
675 if (!capable(CAP_NET_ADMIN)) {
676 ret = -EPERM;
677 break;
678 }
679 goto set_rcvbuf;
680
681 case SO_KEEPALIVE:
682 #ifdef CONFIG_INET
683 if (sk->sk_protocol == IPPROTO_TCP &&
684 sk->sk_type == SOCK_STREAM)
685 tcp_set_keepalive(sk, valbool);
686 #endif
687 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
688 break;
689
690 case SO_OOBINLINE:
691 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
692 break;
693
694 case SO_NO_CHECK:
695 sk->sk_no_check = valbool;
696 break;
697
698 case SO_PRIORITY:
699 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
700 sk->sk_priority = val;
701 else
702 ret = -EPERM;
703 break;
704
705 case SO_LINGER:
706 if (optlen < sizeof(ling)) {
707 ret = -EINVAL; /* 1003.1g */
708 break;
709 }
710 if (copy_from_user(&ling, optval, sizeof(ling))) {
711 ret = -EFAULT;
712 break;
713 }
714 if (!ling.l_onoff)
715 sock_reset_flag(sk, SOCK_LINGER);
716 else {
717 #if (BITS_PER_LONG == 32)
718 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
719 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
720 else
721 #endif
722 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
723 sock_set_flag(sk, SOCK_LINGER);
724 }
725 break;
726
727 case SO_BSDCOMPAT:
728 sock_warn_obsolete_bsdism("setsockopt");
729 break;
730
731 case SO_PASSCRED:
732 if (valbool)
733 set_bit(SOCK_PASSCRED, &sock->flags);
734 else
735 clear_bit(SOCK_PASSCRED, &sock->flags);
736 break;
737
738 case SO_TIMESTAMP:
739 case SO_TIMESTAMPNS:
740 if (valbool) {
741 if (optname == SO_TIMESTAMP)
742 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
743 else
744 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
745 sock_set_flag(sk, SOCK_RCVTSTAMP);
746 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
747 } else {
748 sock_reset_flag(sk, SOCK_RCVTSTAMP);
749 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
750 }
751 break;
752
753 case SO_TIMESTAMPING:
754 if (val & ~SOF_TIMESTAMPING_MASK) {
755 ret = -EINVAL;
756 break;
757 }
758 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
759 val & SOF_TIMESTAMPING_TX_HARDWARE);
760 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
761 val & SOF_TIMESTAMPING_TX_SOFTWARE);
762 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
763 val & SOF_TIMESTAMPING_RX_HARDWARE);
764 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
765 sock_enable_timestamp(sk,
766 SOCK_TIMESTAMPING_RX_SOFTWARE);
767 else
768 sock_disable_timestamp(sk,
769 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
770 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
771 val & SOF_TIMESTAMPING_SOFTWARE);
772 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
773 val & SOF_TIMESTAMPING_SYS_HARDWARE);
774 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
775 val & SOF_TIMESTAMPING_RAW_HARDWARE);
776 break;
777
778 case SO_RCVLOWAT:
779 if (val < 0)
780 val = INT_MAX;
781 sk->sk_rcvlowat = val ? : 1;
782 break;
783
784 case SO_RCVTIMEO:
785 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
786 break;
787
788 case SO_SNDTIMEO:
789 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
790 break;
791
792 case SO_ATTACH_FILTER:
793 ret = -EINVAL;
794 if (optlen == sizeof(struct sock_fprog)) {
795 struct sock_fprog fprog;
796
797 ret = -EFAULT;
798 if (copy_from_user(&fprog, optval, sizeof(fprog)))
799 break;
800
801 ret = sk_attach_filter(&fprog, sk);
802 }
803 break;
804
805 case SO_DETACH_FILTER:
806 ret = sk_detach_filter(sk);
807 break;
808
809 case SO_PASSSEC:
810 if (valbool)
811 set_bit(SOCK_PASSSEC, &sock->flags);
812 else
813 clear_bit(SOCK_PASSSEC, &sock->flags);
814 break;
815 case SO_MARK:
816 if (!capable(CAP_NET_ADMIN))
817 ret = -EPERM;
818 else
819 sk->sk_mark = val;
820 break;
821
822 /* We implement the SO_SNDLOWAT etc to
823 not be settable (1003.1g 5.3) */
824 case SO_RXQ_OVFL:
825 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
826 break;
827
828 case SO_WIFI_STATUS:
829 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
830 break;
831
832 case SO_PEEK_OFF:
833 if (sock->ops->set_peek_off)
834 sock->ops->set_peek_off(sk, val);
835 else
836 ret = -EOPNOTSUPP;
837 break;
838
839 case SO_NOFCS:
840 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
841 break;
842
843 default:
844 ret = -ENOPROTOOPT;
845 break;
846 }
847 release_sock(sk);
848 return ret;
849 }
850 EXPORT_SYMBOL(sock_setsockopt);
851
852
853 void cred_to_ucred(struct pid *pid, const struct cred *cred,
854 struct ucred *ucred)
855 {
856 ucred->pid = pid_vnr(pid);
857 ucred->uid = ucred->gid = -1;
858 if (cred) {
859 struct user_namespace *current_ns = current_user_ns();
860
861 ucred->uid = from_kuid_munged(current_ns, cred->euid);
862 ucred->gid = from_kgid_munged(current_ns, cred->egid);
863 }
864 }
865 EXPORT_SYMBOL_GPL(cred_to_ucred);
866
867 int sock_getsockopt(struct socket *sock, int level, int optname,
868 char __user *optval, int __user *optlen)
869 {
870 struct sock *sk = sock->sk;
871
872 union {
873 int val;
874 struct linger ling;
875 struct timeval tm;
876 } v;
877
878 int lv = sizeof(int);
879 int len;
880
881 if (get_user(len, optlen))
882 return -EFAULT;
883 if (len < 0)
884 return -EINVAL;
885
886 memset(&v, 0, sizeof(v));
887
888 switch (optname) {
889 case SO_DEBUG:
890 v.val = sock_flag(sk, SOCK_DBG);
891 break;
892
893 case SO_DONTROUTE:
894 v.val = sock_flag(sk, SOCK_LOCALROUTE);
895 break;
896
897 case SO_BROADCAST:
898 v.val = sock_flag(sk, SOCK_BROADCAST);
899 break;
900
901 case SO_SNDBUF:
902 v.val = sk->sk_sndbuf;
903 break;
904
905 case SO_RCVBUF:
906 v.val = sk->sk_rcvbuf;
907 break;
908
909 case SO_REUSEADDR:
910 v.val = sk->sk_reuse;
911 break;
912
913 case SO_KEEPALIVE:
914 v.val = sock_flag(sk, SOCK_KEEPOPEN);
915 break;
916
917 case SO_TYPE:
918 v.val = sk->sk_type;
919 break;
920
921 case SO_PROTOCOL:
922 v.val = sk->sk_protocol;
923 break;
924
925 case SO_DOMAIN:
926 v.val = sk->sk_family;
927 break;
928
929 case SO_ERROR:
930 v.val = -sock_error(sk);
931 if (v.val == 0)
932 v.val = xchg(&sk->sk_err_soft, 0);
933 break;
934
935 case SO_OOBINLINE:
936 v.val = sock_flag(sk, SOCK_URGINLINE);
937 break;
938
939 case SO_NO_CHECK:
940 v.val = sk->sk_no_check;
941 break;
942
943 case SO_PRIORITY:
944 v.val = sk->sk_priority;
945 break;
946
947 case SO_LINGER:
948 lv = sizeof(v.ling);
949 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
950 v.ling.l_linger = sk->sk_lingertime / HZ;
951 break;
952
953 case SO_BSDCOMPAT:
954 sock_warn_obsolete_bsdism("getsockopt");
955 break;
956
957 case SO_TIMESTAMP:
958 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
959 !sock_flag(sk, SOCK_RCVTSTAMPNS);
960 break;
961
962 case SO_TIMESTAMPNS:
963 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
964 break;
965
966 case SO_TIMESTAMPING:
967 v.val = 0;
968 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
969 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
970 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
971 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
972 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
973 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
974 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
975 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
976 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
977 v.val |= SOF_TIMESTAMPING_SOFTWARE;
978 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
979 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
980 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
981 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
982 break;
983
984 case SO_RCVTIMEO:
985 lv = sizeof(struct timeval);
986 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
987 v.tm.tv_sec = 0;
988 v.tm.tv_usec = 0;
989 } else {
990 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
991 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
992 }
993 break;
994
995 case SO_SNDTIMEO:
996 lv = sizeof(struct timeval);
997 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
998 v.tm.tv_sec = 0;
999 v.tm.tv_usec = 0;
1000 } else {
1001 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1002 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1003 }
1004 break;
1005
1006 case SO_RCVLOWAT:
1007 v.val = sk->sk_rcvlowat;
1008 break;
1009
1010 case SO_SNDLOWAT:
1011 v.val = 1;
1012 break;
1013
1014 case SO_PASSCRED:
1015 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1016 break;
1017
1018 case SO_PEERCRED:
1019 {
1020 struct ucred peercred;
1021 if (len > sizeof(peercred))
1022 len = sizeof(peercred);
1023 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1024 if (copy_to_user(optval, &peercred, len))
1025 return -EFAULT;
1026 goto lenout;
1027 }
1028
1029 case SO_PEERNAME:
1030 {
1031 char address[128];
1032
1033 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1034 return -ENOTCONN;
1035 if (lv < len)
1036 return -EINVAL;
1037 if (copy_to_user(optval, address, len))
1038 return -EFAULT;
1039 goto lenout;
1040 }
1041
1042 /* Dubious BSD thing... Probably nobody even uses it, but
1043 * the UNIX standard wants it for whatever reason... -DaveM
1044 */
1045 case SO_ACCEPTCONN:
1046 v.val = sk->sk_state == TCP_LISTEN;
1047 break;
1048
1049 case SO_PASSSEC:
1050 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1051 break;
1052
1053 case SO_PEERSEC:
1054 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1055
1056 case SO_MARK:
1057 v.val = sk->sk_mark;
1058 break;
1059
1060 case SO_RXQ_OVFL:
1061 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1062 break;
1063
1064 case SO_WIFI_STATUS:
1065 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1066 break;
1067
1068 case SO_PEEK_OFF:
1069 if (!sock->ops->set_peek_off)
1070 return -EOPNOTSUPP;
1071
1072 v.val = sk->sk_peek_off;
1073 break;
1074 case SO_NOFCS:
1075 v.val = sock_flag(sk, SOCK_NOFCS);
1076 break;
1077 case SO_BINDTODEVICE:
1078 v.val = sk->sk_bound_dev_if;
1079 break;
1080 default:
1081 return -ENOPROTOOPT;
1082 }
1083
1084 if (len > lv)
1085 len = lv;
1086 if (copy_to_user(optval, &v, len))
1087 return -EFAULT;
1088 lenout:
1089 if (put_user(len, optlen))
1090 return -EFAULT;
1091 return 0;
1092 }
1093
1094 /*
1095 * Initialize an sk_lock.
1096 *
1097 * (We also register the sk_lock with the lock validator.)
1098 */
1099 static inline void sock_lock_init(struct sock *sk)
1100 {
1101 sock_lock_init_class_and_name(sk,
1102 af_family_slock_key_strings[sk->sk_family],
1103 af_family_slock_keys + sk->sk_family,
1104 af_family_key_strings[sk->sk_family],
1105 af_family_keys + sk->sk_family);
1106 }
1107
1108 /*
1109 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1110 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1111 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1112 */
1113 static void sock_copy(struct sock *nsk, const struct sock *osk)
1114 {
1115 #ifdef CONFIG_SECURITY_NETWORK
1116 void *sptr = nsk->sk_security;
1117 #endif
1118 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1119
1120 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1121 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1122
1123 #ifdef CONFIG_SECURITY_NETWORK
1124 nsk->sk_security = sptr;
1125 security_sk_clone(osk, nsk);
1126 #endif
1127 }
1128
1129 /*
1130 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1131 * un-modified. Special care is taken when initializing object to zero.
1132 */
1133 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1134 {
1135 if (offsetof(struct sock, sk_node.next) != 0)
1136 memset(sk, 0, offsetof(struct sock, sk_node.next));
1137 memset(&sk->sk_node.pprev, 0,
1138 size - offsetof(struct sock, sk_node.pprev));
1139 }
1140
1141 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1142 {
1143 unsigned long nulls1, nulls2;
1144
1145 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1146 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1147 if (nulls1 > nulls2)
1148 swap(nulls1, nulls2);
1149
1150 if (nulls1 != 0)
1151 memset((char *)sk, 0, nulls1);
1152 memset((char *)sk + nulls1 + sizeof(void *), 0,
1153 nulls2 - nulls1 - sizeof(void *));
1154 memset((char *)sk + nulls2 + sizeof(void *), 0,
1155 size - nulls2 - sizeof(void *));
1156 }
1157 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1158
1159 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1160 int family)
1161 {
1162 struct sock *sk;
1163 struct kmem_cache *slab;
1164
1165 slab = prot->slab;
1166 if (slab != NULL) {
1167 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1168 if (!sk)
1169 return sk;
1170 if (priority & __GFP_ZERO) {
1171 if (prot->clear_sk)
1172 prot->clear_sk(sk, prot->obj_size);
1173 else
1174 sk_prot_clear_nulls(sk, prot->obj_size);
1175 }
1176 } else
1177 sk = kmalloc(prot->obj_size, priority);
1178
1179 if (sk != NULL) {
1180 kmemcheck_annotate_bitfield(sk, flags);
1181
1182 if (security_sk_alloc(sk, family, priority))
1183 goto out_free;
1184
1185 if (!try_module_get(prot->owner))
1186 goto out_free_sec;
1187 sk_tx_queue_clear(sk);
1188 }
1189
1190 return sk;
1191
1192 out_free_sec:
1193 security_sk_free(sk);
1194 out_free:
1195 if (slab != NULL)
1196 kmem_cache_free(slab, sk);
1197 else
1198 kfree(sk);
1199 return NULL;
1200 }
1201
1202 static void sk_prot_free(struct proto *prot, struct sock *sk)
1203 {
1204 struct kmem_cache *slab;
1205 struct module *owner;
1206
1207 owner = prot->owner;
1208 slab = prot->slab;
1209
1210 security_sk_free(sk);
1211 if (slab != NULL)
1212 kmem_cache_free(slab, sk);
1213 else
1214 kfree(sk);
1215 module_put(owner);
1216 }
1217
1218 #ifdef CONFIG_CGROUPS
1219 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1220 void sock_update_classid(struct sock *sk)
1221 {
1222 u32 classid;
1223
1224 classid = task_cls_classid(current);
1225 if (classid != sk->sk_classid)
1226 sk->sk_classid = classid;
1227 }
1228 EXPORT_SYMBOL(sock_update_classid);
1229 #endif
1230
1231 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1232 void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
1233 {
1234 if (in_interrupt())
1235 return;
1236
1237 sk->sk_cgrp_prioidx = task_netprioidx(task);
1238 }
1239 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1240 #endif
1241 #endif
1242
1243 /**
1244 * sk_alloc - All socket objects are allocated here
1245 * @net: the applicable net namespace
1246 * @family: protocol family
1247 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1248 * @prot: struct proto associated with this new sock instance
1249 */
1250 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1251 struct proto *prot)
1252 {
1253 struct sock *sk;
1254
1255 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1256 if (sk) {
1257 sk->sk_family = family;
1258 /*
1259 * See comment in struct sock definition to understand
1260 * why we need sk_prot_creator -acme
1261 */
1262 sk->sk_prot = sk->sk_prot_creator = prot;
1263 sock_lock_init(sk);
1264 sock_net_set(sk, get_net(net));
1265 atomic_set(&sk->sk_wmem_alloc, 1);
1266
1267 sock_update_classid(sk);
1268 sock_update_netprioidx(sk, current);
1269 }
1270
1271 return sk;
1272 }
1273 EXPORT_SYMBOL(sk_alloc);
1274
1275 static void __sk_free(struct sock *sk)
1276 {
1277 struct sk_filter *filter;
1278
1279 if (sk->sk_destruct)
1280 sk->sk_destruct(sk);
1281
1282 filter = rcu_dereference_check(sk->sk_filter,
1283 atomic_read(&sk->sk_wmem_alloc) == 0);
1284 if (filter) {
1285 sk_filter_uncharge(sk, filter);
1286 RCU_INIT_POINTER(sk->sk_filter, NULL);
1287 }
1288
1289 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1290
1291 if (atomic_read(&sk->sk_omem_alloc))
1292 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1293 __func__, atomic_read(&sk->sk_omem_alloc));
1294
1295 if (sk->sk_peer_cred)
1296 put_cred(sk->sk_peer_cred);
1297 put_pid(sk->sk_peer_pid);
1298 put_net(sock_net(sk));
1299 sk_prot_free(sk->sk_prot_creator, sk);
1300 }
1301
1302 void sk_free(struct sock *sk)
1303 {
1304 /*
1305 * We subtract one from sk_wmem_alloc and can know if
1306 * some packets are still in some tx queue.
1307 * If not null, sock_wfree() will call __sk_free(sk) later
1308 */
1309 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1310 __sk_free(sk);
1311 }
1312 EXPORT_SYMBOL(sk_free);
1313
1314 /*
1315 * Last sock_put should drop reference to sk->sk_net. It has already
1316 * been dropped in sk_change_net. Taking reference to stopping namespace
1317 * is not an option.
1318 * Take reference to a socket to remove it from hash _alive_ and after that
1319 * destroy it in the context of init_net.
1320 */
1321 void sk_release_kernel(struct sock *sk)
1322 {
1323 if (sk == NULL || sk->sk_socket == NULL)
1324 return;
1325
1326 sock_hold(sk);
1327 sock_release(sk->sk_socket);
1328 release_net(sock_net(sk));
1329 sock_net_set(sk, get_net(&init_net));
1330 sock_put(sk);
1331 }
1332 EXPORT_SYMBOL(sk_release_kernel);
1333
1334 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1335 {
1336 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1337 sock_update_memcg(newsk);
1338 }
1339
1340 /**
1341 * sk_clone_lock - clone a socket, and lock its clone
1342 * @sk: the socket to clone
1343 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1344 *
1345 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1346 */
1347 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1348 {
1349 struct sock *newsk;
1350
1351 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1352 if (newsk != NULL) {
1353 struct sk_filter *filter;
1354
1355 sock_copy(newsk, sk);
1356
1357 /* SANITY */
1358 get_net(sock_net(newsk));
1359 sk_node_init(&newsk->sk_node);
1360 sock_lock_init(newsk);
1361 bh_lock_sock(newsk);
1362 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1363 newsk->sk_backlog.len = 0;
1364
1365 atomic_set(&newsk->sk_rmem_alloc, 0);
1366 /*
1367 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1368 */
1369 atomic_set(&newsk->sk_wmem_alloc, 1);
1370 atomic_set(&newsk->sk_omem_alloc, 0);
1371 skb_queue_head_init(&newsk->sk_receive_queue);
1372 skb_queue_head_init(&newsk->sk_write_queue);
1373 #ifdef CONFIG_NET_DMA
1374 skb_queue_head_init(&newsk->sk_async_wait_queue);
1375 #endif
1376
1377 spin_lock_init(&newsk->sk_dst_lock);
1378 rwlock_init(&newsk->sk_callback_lock);
1379 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1380 af_callback_keys + newsk->sk_family,
1381 af_family_clock_key_strings[newsk->sk_family]);
1382
1383 newsk->sk_dst_cache = NULL;
1384 newsk->sk_wmem_queued = 0;
1385 newsk->sk_forward_alloc = 0;
1386 newsk->sk_send_head = NULL;
1387 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1388
1389 sock_reset_flag(newsk, SOCK_DONE);
1390 skb_queue_head_init(&newsk->sk_error_queue);
1391
1392 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1393 if (filter != NULL)
1394 sk_filter_charge(newsk, filter);
1395
1396 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1397 /* It is still raw copy of parent, so invalidate
1398 * destructor and make plain sk_free() */
1399 newsk->sk_destruct = NULL;
1400 bh_unlock_sock(newsk);
1401 sk_free(newsk);
1402 newsk = NULL;
1403 goto out;
1404 }
1405
1406 newsk->sk_err = 0;
1407 newsk->sk_priority = 0;
1408 /*
1409 * Before updating sk_refcnt, we must commit prior changes to memory
1410 * (Documentation/RCU/rculist_nulls.txt for details)
1411 */
1412 smp_wmb();
1413 atomic_set(&newsk->sk_refcnt, 2);
1414
1415 /*
1416 * Increment the counter in the same struct proto as the master
1417 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1418 * is the same as sk->sk_prot->socks, as this field was copied
1419 * with memcpy).
1420 *
1421 * This _changes_ the previous behaviour, where
1422 * tcp_create_openreq_child always was incrementing the
1423 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1424 * to be taken into account in all callers. -acme
1425 */
1426 sk_refcnt_debug_inc(newsk);
1427 sk_set_socket(newsk, NULL);
1428 newsk->sk_wq = NULL;
1429
1430 sk_update_clone(sk, newsk);
1431
1432 if (newsk->sk_prot->sockets_allocated)
1433 sk_sockets_allocated_inc(newsk);
1434
1435 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1436 net_enable_timestamp();
1437 }
1438 out:
1439 return newsk;
1440 }
1441 EXPORT_SYMBOL_GPL(sk_clone_lock);
1442
1443 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1444 {
1445 __sk_dst_set(sk, dst);
1446 sk->sk_route_caps = dst->dev->features;
1447 if (sk->sk_route_caps & NETIF_F_GSO)
1448 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1449 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1450 if (sk_can_gso(sk)) {
1451 if (dst->header_len) {
1452 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1453 } else {
1454 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1455 sk->sk_gso_max_size = dst->dev->gso_max_size;
1456 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1457 }
1458 }
1459 }
1460 EXPORT_SYMBOL_GPL(sk_setup_caps);
1461
1462 /*
1463 * Simple resource managers for sockets.
1464 */
1465
1466
1467 /*
1468 * Write buffer destructor automatically called from kfree_skb.
1469 */
1470 void sock_wfree(struct sk_buff *skb)
1471 {
1472 struct sock *sk = skb->sk;
1473 unsigned int len = skb->truesize;
1474
1475 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1476 /*
1477 * Keep a reference on sk_wmem_alloc, this will be released
1478 * after sk_write_space() call
1479 */
1480 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1481 sk->sk_write_space(sk);
1482 len = 1;
1483 }
1484 /*
1485 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1486 * could not do because of in-flight packets
1487 */
1488 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1489 __sk_free(sk);
1490 }
1491 EXPORT_SYMBOL(sock_wfree);
1492
1493 /*
1494 * Read buffer destructor automatically called from kfree_skb.
1495 */
1496 void sock_rfree(struct sk_buff *skb)
1497 {
1498 struct sock *sk = skb->sk;
1499 unsigned int len = skb->truesize;
1500
1501 atomic_sub(len, &sk->sk_rmem_alloc);
1502 sk_mem_uncharge(sk, len);
1503 }
1504 EXPORT_SYMBOL(sock_rfree);
1505
1506 void sock_edemux(struct sk_buff *skb)
1507 {
1508 struct sock *sk = skb->sk;
1509
1510 #ifdef CONFIG_INET
1511 if (sk->sk_state == TCP_TIME_WAIT)
1512 inet_twsk_put(inet_twsk(sk));
1513 else
1514 #endif
1515 sock_put(sk);
1516 }
1517 EXPORT_SYMBOL(sock_edemux);
1518
1519 kuid_t sock_i_uid(struct sock *sk)
1520 {
1521 kuid_t uid;
1522
1523 read_lock_bh(&sk->sk_callback_lock);
1524 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1525 read_unlock_bh(&sk->sk_callback_lock);
1526 return uid;
1527 }
1528 EXPORT_SYMBOL(sock_i_uid);
1529
1530 unsigned long sock_i_ino(struct sock *sk)
1531 {
1532 unsigned long ino;
1533
1534 read_lock_bh(&sk->sk_callback_lock);
1535 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1536 read_unlock_bh(&sk->sk_callback_lock);
1537 return ino;
1538 }
1539 EXPORT_SYMBOL(sock_i_ino);
1540
1541 /*
1542 * Allocate a skb from the socket's send buffer.
1543 */
1544 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1545 gfp_t priority)
1546 {
1547 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1548 struct sk_buff *skb = alloc_skb(size, priority);
1549 if (skb) {
1550 skb_set_owner_w(skb, sk);
1551 return skb;
1552 }
1553 }
1554 return NULL;
1555 }
1556 EXPORT_SYMBOL(sock_wmalloc);
1557
1558 /*
1559 * Allocate a skb from the socket's receive buffer.
1560 */
1561 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1562 gfp_t priority)
1563 {
1564 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1565 struct sk_buff *skb = alloc_skb(size, priority);
1566 if (skb) {
1567 skb_set_owner_r(skb, sk);
1568 return skb;
1569 }
1570 }
1571 return NULL;
1572 }
1573
1574 /*
1575 * Allocate a memory block from the socket's option memory buffer.
1576 */
1577 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1578 {
1579 if ((unsigned int)size <= sysctl_optmem_max &&
1580 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1581 void *mem;
1582 /* First do the add, to avoid the race if kmalloc
1583 * might sleep.
1584 */
1585 atomic_add(size, &sk->sk_omem_alloc);
1586 mem = kmalloc(size, priority);
1587 if (mem)
1588 return mem;
1589 atomic_sub(size, &sk->sk_omem_alloc);
1590 }
1591 return NULL;
1592 }
1593 EXPORT_SYMBOL(sock_kmalloc);
1594
1595 /*
1596 * Free an option memory block.
1597 */
1598 void sock_kfree_s(struct sock *sk, void *mem, int size)
1599 {
1600 kfree(mem);
1601 atomic_sub(size, &sk->sk_omem_alloc);
1602 }
1603 EXPORT_SYMBOL(sock_kfree_s);
1604
1605 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1606 I think, these locks should be removed for datagram sockets.
1607 */
1608 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1609 {
1610 DEFINE_WAIT(wait);
1611
1612 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1613 for (;;) {
1614 if (!timeo)
1615 break;
1616 if (signal_pending(current))
1617 break;
1618 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1619 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1620 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1621 break;
1622 if (sk->sk_shutdown & SEND_SHUTDOWN)
1623 break;
1624 if (sk->sk_err)
1625 break;
1626 timeo = schedule_timeout(timeo);
1627 }
1628 finish_wait(sk_sleep(sk), &wait);
1629 return timeo;
1630 }
1631
1632
1633 /*
1634 * Generic send/receive buffer handlers
1635 */
1636
1637 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1638 unsigned long data_len, int noblock,
1639 int *errcode)
1640 {
1641 struct sk_buff *skb;
1642 gfp_t gfp_mask;
1643 long timeo;
1644 int err;
1645 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1646
1647 err = -EMSGSIZE;
1648 if (npages > MAX_SKB_FRAGS)
1649 goto failure;
1650
1651 gfp_mask = sk->sk_allocation;
1652 if (gfp_mask & __GFP_WAIT)
1653 gfp_mask |= __GFP_REPEAT;
1654
1655 timeo = sock_sndtimeo(sk, noblock);
1656 while (1) {
1657 err = sock_error(sk);
1658 if (err != 0)
1659 goto failure;
1660
1661 err = -EPIPE;
1662 if (sk->sk_shutdown & SEND_SHUTDOWN)
1663 goto failure;
1664
1665 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1666 skb = alloc_skb(header_len, gfp_mask);
1667 if (skb) {
1668 int i;
1669
1670 /* No pages, we're done... */
1671 if (!data_len)
1672 break;
1673
1674 skb->truesize += data_len;
1675 skb_shinfo(skb)->nr_frags = npages;
1676 for (i = 0; i < npages; i++) {
1677 struct page *page;
1678
1679 page = alloc_pages(sk->sk_allocation, 0);
1680 if (!page) {
1681 err = -ENOBUFS;
1682 skb_shinfo(skb)->nr_frags = i;
1683 kfree_skb(skb);
1684 goto failure;
1685 }
1686
1687 __skb_fill_page_desc(skb, i,
1688 page, 0,
1689 (data_len >= PAGE_SIZE ?
1690 PAGE_SIZE :
1691 data_len));
1692 data_len -= PAGE_SIZE;
1693 }
1694
1695 /* Full success... */
1696 break;
1697 }
1698 err = -ENOBUFS;
1699 goto failure;
1700 }
1701 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1702 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1703 err = -EAGAIN;
1704 if (!timeo)
1705 goto failure;
1706 if (signal_pending(current))
1707 goto interrupted;
1708 timeo = sock_wait_for_wmem(sk, timeo);
1709 }
1710
1711 skb_set_owner_w(skb, sk);
1712 return skb;
1713
1714 interrupted:
1715 err = sock_intr_errno(timeo);
1716 failure:
1717 *errcode = err;
1718 return NULL;
1719 }
1720 EXPORT_SYMBOL(sock_alloc_send_pskb);
1721
1722 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1723 int noblock, int *errcode)
1724 {
1725 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1726 }
1727 EXPORT_SYMBOL(sock_alloc_send_skb);
1728
1729 /* On 32bit arches, an skb frag is limited to 2^15 */
1730 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1731
1732 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1733 {
1734 int order;
1735
1736 if (pfrag->page) {
1737 if (atomic_read(&pfrag->page->_count) == 1) {
1738 pfrag->offset = 0;
1739 return true;
1740 }
1741 if (pfrag->offset < pfrag->size)
1742 return true;
1743 put_page(pfrag->page);
1744 }
1745
1746 /* We restrict high order allocations to users that can afford to wait */
1747 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1748
1749 do {
1750 gfp_t gfp = sk->sk_allocation;
1751
1752 if (order)
1753 gfp |= __GFP_COMP | __GFP_NOWARN;
1754 pfrag->page = alloc_pages(gfp, order);
1755 if (likely(pfrag->page)) {
1756 pfrag->offset = 0;
1757 pfrag->size = PAGE_SIZE << order;
1758 return true;
1759 }
1760 } while (--order >= 0);
1761
1762 sk_enter_memory_pressure(sk);
1763 sk_stream_moderate_sndbuf(sk);
1764 return false;
1765 }
1766 EXPORT_SYMBOL(sk_page_frag_refill);
1767
1768 static void __lock_sock(struct sock *sk)
1769 __releases(&sk->sk_lock.slock)
1770 __acquires(&sk->sk_lock.slock)
1771 {
1772 DEFINE_WAIT(wait);
1773
1774 for (;;) {
1775 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1776 TASK_UNINTERRUPTIBLE);
1777 spin_unlock_bh(&sk->sk_lock.slock);
1778 schedule();
1779 spin_lock_bh(&sk->sk_lock.slock);
1780 if (!sock_owned_by_user(sk))
1781 break;
1782 }
1783 finish_wait(&sk->sk_lock.wq, &wait);
1784 }
1785
1786 static void __release_sock(struct sock *sk)
1787 __releases(&sk->sk_lock.slock)
1788 __acquires(&sk->sk_lock.slock)
1789 {
1790 struct sk_buff *skb = sk->sk_backlog.head;
1791
1792 do {
1793 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1794 bh_unlock_sock(sk);
1795
1796 do {
1797 struct sk_buff *next = skb->next;
1798
1799 prefetch(next);
1800 WARN_ON_ONCE(skb_dst_is_noref(skb));
1801 skb->next = NULL;
1802 sk_backlog_rcv(sk, skb);
1803
1804 /*
1805 * We are in process context here with softirqs
1806 * disabled, use cond_resched_softirq() to preempt.
1807 * This is safe to do because we've taken the backlog
1808 * queue private:
1809 */
1810 cond_resched_softirq();
1811
1812 skb = next;
1813 } while (skb != NULL);
1814
1815 bh_lock_sock(sk);
1816 } while ((skb = sk->sk_backlog.head) != NULL);
1817
1818 /*
1819 * Doing the zeroing here guarantee we can not loop forever
1820 * while a wild producer attempts to flood us.
1821 */
1822 sk->sk_backlog.len = 0;
1823 }
1824
1825 /**
1826 * sk_wait_data - wait for data to arrive at sk_receive_queue
1827 * @sk: sock to wait on
1828 * @timeo: for how long
1829 *
1830 * Now socket state including sk->sk_err is changed only under lock,
1831 * hence we may omit checks after joining wait queue.
1832 * We check receive queue before schedule() only as optimization;
1833 * it is very likely that release_sock() added new data.
1834 */
1835 int sk_wait_data(struct sock *sk, long *timeo)
1836 {
1837 int rc;
1838 DEFINE_WAIT(wait);
1839
1840 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1841 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1842 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1843 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1844 finish_wait(sk_sleep(sk), &wait);
1845 return rc;
1846 }
1847 EXPORT_SYMBOL(sk_wait_data);
1848
1849 /**
1850 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1851 * @sk: socket
1852 * @size: memory size to allocate
1853 * @kind: allocation type
1854 *
1855 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1856 * rmem allocation. This function assumes that protocols which have
1857 * memory_pressure use sk_wmem_queued as write buffer accounting.
1858 */
1859 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1860 {
1861 struct proto *prot = sk->sk_prot;
1862 int amt = sk_mem_pages(size);
1863 long allocated;
1864 int parent_status = UNDER_LIMIT;
1865
1866 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1867
1868 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1869
1870 /* Under limit. */
1871 if (parent_status == UNDER_LIMIT &&
1872 allocated <= sk_prot_mem_limits(sk, 0)) {
1873 sk_leave_memory_pressure(sk);
1874 return 1;
1875 }
1876
1877 /* Under pressure. (we or our parents) */
1878 if ((parent_status > SOFT_LIMIT) ||
1879 allocated > sk_prot_mem_limits(sk, 1))
1880 sk_enter_memory_pressure(sk);
1881
1882 /* Over hard limit (we or our parents) */
1883 if ((parent_status == OVER_LIMIT) ||
1884 (allocated > sk_prot_mem_limits(sk, 2)))
1885 goto suppress_allocation;
1886
1887 /* guarantee minimum buffer size under pressure */
1888 if (kind == SK_MEM_RECV) {
1889 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1890 return 1;
1891
1892 } else { /* SK_MEM_SEND */
1893 if (sk->sk_type == SOCK_STREAM) {
1894 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1895 return 1;
1896 } else if (atomic_read(&sk->sk_wmem_alloc) <
1897 prot->sysctl_wmem[0])
1898 return 1;
1899 }
1900
1901 if (sk_has_memory_pressure(sk)) {
1902 int alloc;
1903
1904 if (!sk_under_memory_pressure(sk))
1905 return 1;
1906 alloc = sk_sockets_allocated_read_positive(sk);
1907 if (sk_prot_mem_limits(sk, 2) > alloc *
1908 sk_mem_pages(sk->sk_wmem_queued +
1909 atomic_read(&sk->sk_rmem_alloc) +
1910 sk->sk_forward_alloc))
1911 return 1;
1912 }
1913
1914 suppress_allocation:
1915
1916 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1917 sk_stream_moderate_sndbuf(sk);
1918
1919 /* Fail only if socket is _under_ its sndbuf.
1920 * In this case we cannot block, so that we have to fail.
1921 */
1922 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1923 return 1;
1924 }
1925
1926 trace_sock_exceed_buf_limit(sk, prot, allocated);
1927
1928 /* Alas. Undo changes. */
1929 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1930
1931 sk_memory_allocated_sub(sk, amt);
1932
1933 return 0;
1934 }
1935 EXPORT_SYMBOL(__sk_mem_schedule);
1936
1937 /**
1938 * __sk_reclaim - reclaim memory_allocated
1939 * @sk: socket
1940 */
1941 void __sk_mem_reclaim(struct sock *sk)
1942 {
1943 sk_memory_allocated_sub(sk,
1944 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1945 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1946
1947 if (sk_under_memory_pressure(sk) &&
1948 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1949 sk_leave_memory_pressure(sk);
1950 }
1951 EXPORT_SYMBOL(__sk_mem_reclaim);
1952
1953
1954 /*
1955 * Set of default routines for initialising struct proto_ops when
1956 * the protocol does not support a particular function. In certain
1957 * cases where it makes no sense for a protocol to have a "do nothing"
1958 * function, some default processing is provided.
1959 */
1960
1961 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1962 {
1963 return -EOPNOTSUPP;
1964 }
1965 EXPORT_SYMBOL(sock_no_bind);
1966
1967 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1968 int len, int flags)
1969 {
1970 return -EOPNOTSUPP;
1971 }
1972 EXPORT_SYMBOL(sock_no_connect);
1973
1974 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1975 {
1976 return -EOPNOTSUPP;
1977 }
1978 EXPORT_SYMBOL(sock_no_socketpair);
1979
1980 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1981 {
1982 return -EOPNOTSUPP;
1983 }
1984 EXPORT_SYMBOL(sock_no_accept);
1985
1986 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1987 int *len, int peer)
1988 {
1989 return -EOPNOTSUPP;
1990 }
1991 EXPORT_SYMBOL(sock_no_getname);
1992
1993 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1994 {
1995 return 0;
1996 }
1997 EXPORT_SYMBOL(sock_no_poll);
1998
1999 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2000 {
2001 return -EOPNOTSUPP;
2002 }
2003 EXPORT_SYMBOL(sock_no_ioctl);
2004
2005 int sock_no_listen(struct socket *sock, int backlog)
2006 {
2007 return -EOPNOTSUPP;
2008 }
2009 EXPORT_SYMBOL(sock_no_listen);
2010
2011 int sock_no_shutdown(struct socket *sock, int how)
2012 {
2013 return -EOPNOTSUPP;
2014 }
2015 EXPORT_SYMBOL(sock_no_shutdown);
2016
2017 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2018 char __user *optval, unsigned int optlen)
2019 {
2020 return -EOPNOTSUPP;
2021 }
2022 EXPORT_SYMBOL(sock_no_setsockopt);
2023
2024 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2025 char __user *optval, int __user *optlen)
2026 {
2027 return -EOPNOTSUPP;
2028 }
2029 EXPORT_SYMBOL(sock_no_getsockopt);
2030
2031 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2032 size_t len)
2033 {
2034 return -EOPNOTSUPP;
2035 }
2036 EXPORT_SYMBOL(sock_no_sendmsg);
2037
2038 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2039 size_t len, int flags)
2040 {
2041 return -EOPNOTSUPP;
2042 }
2043 EXPORT_SYMBOL(sock_no_recvmsg);
2044
2045 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2046 {
2047 /* Mirror missing mmap method error code */
2048 return -ENODEV;
2049 }
2050 EXPORT_SYMBOL(sock_no_mmap);
2051
2052 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2053 {
2054 ssize_t res;
2055 struct msghdr msg = {.msg_flags = flags};
2056 struct kvec iov;
2057 char *kaddr = kmap(page);
2058 iov.iov_base = kaddr + offset;
2059 iov.iov_len = size;
2060 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2061 kunmap(page);
2062 return res;
2063 }
2064 EXPORT_SYMBOL(sock_no_sendpage);
2065
2066 /*
2067 * Default Socket Callbacks
2068 */
2069
2070 static void sock_def_wakeup(struct sock *sk)
2071 {
2072 struct socket_wq *wq;
2073
2074 rcu_read_lock();
2075 wq = rcu_dereference(sk->sk_wq);
2076 if (wq_has_sleeper(wq))
2077 wake_up_interruptible_all(&wq->wait);
2078 rcu_read_unlock();
2079 }
2080
2081 static void sock_def_error_report(struct sock *sk)
2082 {
2083 struct socket_wq *wq;
2084
2085 rcu_read_lock();
2086 wq = rcu_dereference(sk->sk_wq);
2087 if (wq_has_sleeper(wq))
2088 wake_up_interruptible_poll(&wq->wait, POLLERR);
2089 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2090 rcu_read_unlock();
2091 }
2092
2093 static void sock_def_readable(struct sock *sk, int len)
2094 {
2095 struct socket_wq *wq;
2096
2097 rcu_read_lock();
2098 wq = rcu_dereference(sk->sk_wq);
2099 if (wq_has_sleeper(wq))
2100 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2101 POLLRDNORM | POLLRDBAND);
2102 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2103 rcu_read_unlock();
2104 }
2105
2106 static void sock_def_write_space(struct sock *sk)
2107 {
2108 struct socket_wq *wq;
2109
2110 rcu_read_lock();
2111
2112 /* Do not wake up a writer until he can make "significant"
2113 * progress. --DaveM
2114 */
2115 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2116 wq = rcu_dereference(sk->sk_wq);
2117 if (wq_has_sleeper(wq))
2118 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2119 POLLWRNORM | POLLWRBAND);
2120
2121 /* Should agree with poll, otherwise some programs break */
2122 if (sock_writeable(sk))
2123 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2124 }
2125
2126 rcu_read_unlock();
2127 }
2128
2129 static void sock_def_destruct(struct sock *sk)
2130 {
2131 kfree(sk->sk_protinfo);
2132 }
2133
2134 void sk_send_sigurg(struct sock *sk)
2135 {
2136 if (sk->sk_socket && sk->sk_socket->file)
2137 if (send_sigurg(&sk->sk_socket->file->f_owner))
2138 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2139 }
2140 EXPORT_SYMBOL(sk_send_sigurg);
2141
2142 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2143 unsigned long expires)
2144 {
2145 if (!mod_timer(timer, expires))
2146 sock_hold(sk);
2147 }
2148 EXPORT_SYMBOL(sk_reset_timer);
2149
2150 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2151 {
2152 if (timer_pending(timer) && del_timer(timer))
2153 __sock_put(sk);
2154 }
2155 EXPORT_SYMBOL(sk_stop_timer);
2156
2157 void sock_init_data(struct socket *sock, struct sock *sk)
2158 {
2159 skb_queue_head_init(&sk->sk_receive_queue);
2160 skb_queue_head_init(&sk->sk_write_queue);
2161 skb_queue_head_init(&sk->sk_error_queue);
2162 #ifdef CONFIG_NET_DMA
2163 skb_queue_head_init(&sk->sk_async_wait_queue);
2164 #endif
2165
2166 sk->sk_send_head = NULL;
2167
2168 init_timer(&sk->sk_timer);
2169
2170 sk->sk_allocation = GFP_KERNEL;
2171 sk->sk_rcvbuf = sysctl_rmem_default;
2172 sk->sk_sndbuf = sysctl_wmem_default;
2173 sk->sk_state = TCP_CLOSE;
2174 sk_set_socket(sk, sock);
2175
2176 sock_set_flag(sk, SOCK_ZAPPED);
2177
2178 if (sock) {
2179 sk->sk_type = sock->type;
2180 sk->sk_wq = sock->wq;
2181 sock->sk = sk;
2182 } else
2183 sk->sk_wq = NULL;
2184
2185 spin_lock_init(&sk->sk_dst_lock);
2186 rwlock_init(&sk->sk_callback_lock);
2187 lockdep_set_class_and_name(&sk->sk_callback_lock,
2188 af_callback_keys + sk->sk_family,
2189 af_family_clock_key_strings[sk->sk_family]);
2190
2191 sk->sk_state_change = sock_def_wakeup;
2192 sk->sk_data_ready = sock_def_readable;
2193 sk->sk_write_space = sock_def_write_space;
2194 sk->sk_error_report = sock_def_error_report;
2195 sk->sk_destruct = sock_def_destruct;
2196
2197 sk->sk_frag.page = NULL;
2198 sk->sk_frag.offset = 0;
2199 sk->sk_peek_off = -1;
2200
2201 sk->sk_peer_pid = NULL;
2202 sk->sk_peer_cred = NULL;
2203 sk->sk_write_pending = 0;
2204 sk->sk_rcvlowat = 1;
2205 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2206 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2207
2208 sk->sk_stamp = ktime_set(-1L, 0);
2209
2210 /*
2211 * Before updating sk_refcnt, we must commit prior changes to memory
2212 * (Documentation/RCU/rculist_nulls.txt for details)
2213 */
2214 smp_wmb();
2215 atomic_set(&sk->sk_refcnt, 1);
2216 atomic_set(&sk->sk_drops, 0);
2217 }
2218 EXPORT_SYMBOL(sock_init_data);
2219
2220 void lock_sock_nested(struct sock *sk, int subclass)
2221 {
2222 might_sleep();
2223 spin_lock_bh(&sk->sk_lock.slock);
2224 if (sk->sk_lock.owned)
2225 __lock_sock(sk);
2226 sk->sk_lock.owned = 1;
2227 spin_unlock(&sk->sk_lock.slock);
2228 /*
2229 * The sk_lock has mutex_lock() semantics here:
2230 */
2231 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2232 local_bh_enable();
2233 }
2234 EXPORT_SYMBOL(lock_sock_nested);
2235
2236 void release_sock(struct sock *sk)
2237 {
2238 /*
2239 * The sk_lock has mutex_unlock() semantics:
2240 */
2241 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2242
2243 spin_lock_bh(&sk->sk_lock.slock);
2244 if (sk->sk_backlog.tail)
2245 __release_sock(sk);
2246
2247 if (sk->sk_prot->release_cb)
2248 sk->sk_prot->release_cb(sk);
2249
2250 sk->sk_lock.owned = 0;
2251 if (waitqueue_active(&sk->sk_lock.wq))
2252 wake_up(&sk->sk_lock.wq);
2253 spin_unlock_bh(&sk->sk_lock.slock);
2254 }
2255 EXPORT_SYMBOL(release_sock);
2256
2257 /**
2258 * lock_sock_fast - fast version of lock_sock
2259 * @sk: socket
2260 *
2261 * This version should be used for very small section, where process wont block
2262 * return false if fast path is taken
2263 * sk_lock.slock locked, owned = 0, BH disabled
2264 * return true if slow path is taken
2265 * sk_lock.slock unlocked, owned = 1, BH enabled
2266 */
2267 bool lock_sock_fast(struct sock *sk)
2268 {
2269 might_sleep();
2270 spin_lock_bh(&sk->sk_lock.slock);
2271
2272 if (!sk->sk_lock.owned)
2273 /*
2274 * Note : We must disable BH
2275 */
2276 return false;
2277
2278 __lock_sock(sk);
2279 sk->sk_lock.owned = 1;
2280 spin_unlock(&sk->sk_lock.slock);
2281 /*
2282 * The sk_lock has mutex_lock() semantics here:
2283 */
2284 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2285 local_bh_enable();
2286 return true;
2287 }
2288 EXPORT_SYMBOL(lock_sock_fast);
2289
2290 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2291 {
2292 struct timeval tv;
2293 if (!sock_flag(sk, SOCK_TIMESTAMP))
2294 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2295 tv = ktime_to_timeval(sk->sk_stamp);
2296 if (tv.tv_sec == -1)
2297 return -ENOENT;
2298 if (tv.tv_sec == 0) {
2299 sk->sk_stamp = ktime_get_real();
2300 tv = ktime_to_timeval(sk->sk_stamp);
2301 }
2302 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2303 }
2304 EXPORT_SYMBOL(sock_get_timestamp);
2305
2306 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2307 {
2308 struct timespec ts;
2309 if (!sock_flag(sk, SOCK_TIMESTAMP))
2310 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2311 ts = ktime_to_timespec(sk->sk_stamp);
2312 if (ts.tv_sec == -1)
2313 return -ENOENT;
2314 if (ts.tv_sec == 0) {
2315 sk->sk_stamp = ktime_get_real();
2316 ts = ktime_to_timespec(sk->sk_stamp);
2317 }
2318 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2319 }
2320 EXPORT_SYMBOL(sock_get_timestampns);
2321
2322 void sock_enable_timestamp(struct sock *sk, int flag)
2323 {
2324 if (!sock_flag(sk, flag)) {
2325 unsigned long previous_flags = sk->sk_flags;
2326
2327 sock_set_flag(sk, flag);
2328 /*
2329 * we just set one of the two flags which require net
2330 * time stamping, but time stamping might have been on
2331 * already because of the other one
2332 */
2333 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2334 net_enable_timestamp();
2335 }
2336 }
2337
2338 /*
2339 * Get a socket option on an socket.
2340 *
2341 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2342 * asynchronous errors should be reported by getsockopt. We assume
2343 * this means if you specify SO_ERROR (otherwise whats the point of it).
2344 */
2345 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2346 char __user *optval, int __user *optlen)
2347 {
2348 struct sock *sk = sock->sk;
2349
2350 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2351 }
2352 EXPORT_SYMBOL(sock_common_getsockopt);
2353
2354 #ifdef CONFIG_COMPAT
2355 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2356 char __user *optval, int __user *optlen)
2357 {
2358 struct sock *sk = sock->sk;
2359
2360 if (sk->sk_prot->compat_getsockopt != NULL)
2361 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2362 optval, optlen);
2363 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2364 }
2365 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2366 #endif
2367
2368 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2369 struct msghdr *msg, size_t size, int flags)
2370 {
2371 struct sock *sk = sock->sk;
2372 int addr_len = 0;
2373 int err;
2374
2375 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2376 flags & ~MSG_DONTWAIT, &addr_len);
2377 if (err >= 0)
2378 msg->msg_namelen = addr_len;
2379 return err;
2380 }
2381 EXPORT_SYMBOL(sock_common_recvmsg);
2382
2383 /*
2384 * Set socket options on an inet socket.
2385 */
2386 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2387 char __user *optval, unsigned int optlen)
2388 {
2389 struct sock *sk = sock->sk;
2390
2391 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2392 }
2393 EXPORT_SYMBOL(sock_common_setsockopt);
2394
2395 #ifdef CONFIG_COMPAT
2396 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2397 char __user *optval, unsigned int optlen)
2398 {
2399 struct sock *sk = sock->sk;
2400
2401 if (sk->sk_prot->compat_setsockopt != NULL)
2402 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2403 optval, optlen);
2404 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2405 }
2406 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2407 #endif
2408
2409 void sk_common_release(struct sock *sk)
2410 {
2411 if (sk->sk_prot->destroy)
2412 sk->sk_prot->destroy(sk);
2413
2414 /*
2415 * Observation: when sock_common_release is called, processes have
2416 * no access to socket. But net still has.
2417 * Step one, detach it from networking:
2418 *
2419 * A. Remove from hash tables.
2420 */
2421
2422 sk->sk_prot->unhash(sk);
2423
2424 /*
2425 * In this point socket cannot receive new packets, but it is possible
2426 * that some packets are in flight because some CPU runs receiver and
2427 * did hash table lookup before we unhashed socket. They will achieve
2428 * receive queue and will be purged by socket destructor.
2429 *
2430 * Also we still have packets pending on receive queue and probably,
2431 * our own packets waiting in device queues. sock_destroy will drain
2432 * receive queue, but transmitted packets will delay socket destruction
2433 * until the last reference will be released.
2434 */
2435
2436 sock_orphan(sk);
2437
2438 xfrm_sk_free_policy(sk);
2439
2440 sk_refcnt_debug_release(sk);
2441
2442 if (sk->sk_frag.page) {
2443 put_page(sk->sk_frag.page);
2444 sk->sk_frag.page = NULL;
2445 }
2446
2447 sock_put(sk);
2448 }
2449 EXPORT_SYMBOL(sk_common_release);
2450
2451 #ifdef CONFIG_PROC_FS
2452 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2453 struct prot_inuse {
2454 int val[PROTO_INUSE_NR];
2455 };
2456
2457 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2458
2459 #ifdef CONFIG_NET_NS
2460 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2461 {
2462 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2463 }
2464 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2465
2466 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2467 {
2468 int cpu, idx = prot->inuse_idx;
2469 int res = 0;
2470
2471 for_each_possible_cpu(cpu)
2472 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2473
2474 return res >= 0 ? res : 0;
2475 }
2476 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2477
2478 static int __net_init sock_inuse_init_net(struct net *net)
2479 {
2480 net->core.inuse = alloc_percpu(struct prot_inuse);
2481 return net->core.inuse ? 0 : -ENOMEM;
2482 }
2483
2484 static void __net_exit sock_inuse_exit_net(struct net *net)
2485 {
2486 free_percpu(net->core.inuse);
2487 }
2488
2489 static struct pernet_operations net_inuse_ops = {
2490 .init = sock_inuse_init_net,
2491 .exit = sock_inuse_exit_net,
2492 };
2493
2494 static __init int net_inuse_init(void)
2495 {
2496 if (register_pernet_subsys(&net_inuse_ops))
2497 panic("Cannot initialize net inuse counters");
2498
2499 return 0;
2500 }
2501
2502 core_initcall(net_inuse_init);
2503 #else
2504 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2505
2506 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2507 {
2508 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2509 }
2510 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2511
2512 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2513 {
2514 int cpu, idx = prot->inuse_idx;
2515 int res = 0;
2516
2517 for_each_possible_cpu(cpu)
2518 res += per_cpu(prot_inuse, cpu).val[idx];
2519
2520 return res >= 0 ? res : 0;
2521 }
2522 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2523 #endif
2524
2525 static void assign_proto_idx(struct proto *prot)
2526 {
2527 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2528
2529 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2530 pr_err("PROTO_INUSE_NR exhausted\n");
2531 return;
2532 }
2533
2534 set_bit(prot->inuse_idx, proto_inuse_idx);
2535 }
2536
2537 static void release_proto_idx(struct proto *prot)
2538 {
2539 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2540 clear_bit(prot->inuse_idx, proto_inuse_idx);
2541 }
2542 #else
2543 static inline void assign_proto_idx(struct proto *prot)
2544 {
2545 }
2546
2547 static inline void release_proto_idx(struct proto *prot)
2548 {
2549 }
2550 #endif
2551
2552 int proto_register(struct proto *prot, int alloc_slab)
2553 {
2554 if (alloc_slab) {
2555 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2556 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2557 NULL);
2558
2559 if (prot->slab == NULL) {
2560 pr_crit("%s: Can't create sock SLAB cache!\n",
2561 prot->name);
2562 goto out;
2563 }
2564
2565 if (prot->rsk_prot != NULL) {
2566 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2567 if (prot->rsk_prot->slab_name == NULL)
2568 goto out_free_sock_slab;
2569
2570 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2571 prot->rsk_prot->obj_size, 0,
2572 SLAB_HWCACHE_ALIGN, NULL);
2573
2574 if (prot->rsk_prot->slab == NULL) {
2575 pr_crit("%s: Can't create request sock SLAB cache!\n",
2576 prot->name);
2577 goto out_free_request_sock_slab_name;
2578 }
2579 }
2580
2581 if (prot->twsk_prot != NULL) {
2582 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2583
2584 if (prot->twsk_prot->twsk_slab_name == NULL)
2585 goto out_free_request_sock_slab;
2586
2587 prot->twsk_prot->twsk_slab =
2588 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2589 prot->twsk_prot->twsk_obj_size,
2590 0,
2591 SLAB_HWCACHE_ALIGN |
2592 prot->slab_flags,
2593 NULL);
2594 if (prot->twsk_prot->twsk_slab == NULL)
2595 goto out_free_timewait_sock_slab_name;
2596 }
2597 }
2598
2599 mutex_lock(&proto_list_mutex);
2600 list_add(&prot->node, &proto_list);
2601 assign_proto_idx(prot);
2602 mutex_unlock(&proto_list_mutex);
2603 return 0;
2604
2605 out_free_timewait_sock_slab_name:
2606 kfree(prot->twsk_prot->twsk_slab_name);
2607 out_free_request_sock_slab:
2608 if (prot->rsk_prot && prot->rsk_prot->slab) {
2609 kmem_cache_destroy(prot->rsk_prot->slab);
2610 prot->rsk_prot->slab = NULL;
2611 }
2612 out_free_request_sock_slab_name:
2613 if (prot->rsk_prot)
2614 kfree(prot->rsk_prot->slab_name);
2615 out_free_sock_slab:
2616 kmem_cache_destroy(prot->slab);
2617 prot->slab = NULL;
2618 out:
2619 return -ENOBUFS;
2620 }
2621 EXPORT_SYMBOL(proto_register);
2622
2623 void proto_unregister(struct proto *prot)
2624 {
2625 mutex_lock(&proto_list_mutex);
2626 release_proto_idx(prot);
2627 list_del(&prot->node);
2628 mutex_unlock(&proto_list_mutex);
2629
2630 if (prot->slab != NULL) {
2631 kmem_cache_destroy(prot->slab);
2632 prot->slab = NULL;
2633 }
2634
2635 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2636 kmem_cache_destroy(prot->rsk_prot->slab);
2637 kfree(prot->rsk_prot->slab_name);
2638 prot->rsk_prot->slab = NULL;
2639 }
2640
2641 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2642 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2643 kfree(prot->twsk_prot->twsk_slab_name);
2644 prot->twsk_prot->twsk_slab = NULL;
2645 }
2646 }
2647 EXPORT_SYMBOL(proto_unregister);
2648
2649 #ifdef CONFIG_PROC_FS
2650 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2651 __acquires(proto_list_mutex)
2652 {
2653 mutex_lock(&proto_list_mutex);
2654 return seq_list_start_head(&proto_list, *pos);
2655 }
2656
2657 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2658 {
2659 return seq_list_next(v, &proto_list, pos);
2660 }
2661
2662 static void proto_seq_stop(struct seq_file *seq, void *v)
2663 __releases(proto_list_mutex)
2664 {
2665 mutex_unlock(&proto_list_mutex);
2666 }
2667
2668 static char proto_method_implemented(const void *method)
2669 {
2670 return method == NULL ? 'n' : 'y';
2671 }
2672 static long sock_prot_memory_allocated(struct proto *proto)
2673 {
2674 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2675 }
2676
2677 static char *sock_prot_memory_pressure(struct proto *proto)
2678 {
2679 return proto->memory_pressure != NULL ?
2680 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2681 }
2682
2683 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2684 {
2685
2686 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2687 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2688 proto->name,
2689 proto->obj_size,
2690 sock_prot_inuse_get(seq_file_net(seq), proto),
2691 sock_prot_memory_allocated(proto),
2692 sock_prot_memory_pressure(proto),
2693 proto->max_header,
2694 proto->slab == NULL ? "no" : "yes",
2695 module_name(proto->owner),
2696 proto_method_implemented(proto->close),
2697 proto_method_implemented(proto->connect),
2698 proto_method_implemented(proto->disconnect),
2699 proto_method_implemented(proto->accept),
2700 proto_method_implemented(proto->ioctl),
2701 proto_method_implemented(proto->init),
2702 proto_method_implemented(proto->destroy),
2703 proto_method_implemented(proto->shutdown),
2704 proto_method_implemented(proto->setsockopt),
2705 proto_method_implemented(proto->getsockopt),
2706 proto_method_implemented(proto->sendmsg),
2707 proto_method_implemented(proto->recvmsg),
2708 proto_method_implemented(proto->sendpage),
2709 proto_method_implemented(proto->bind),
2710 proto_method_implemented(proto->backlog_rcv),
2711 proto_method_implemented(proto->hash),
2712 proto_method_implemented(proto->unhash),
2713 proto_method_implemented(proto->get_port),
2714 proto_method_implemented(proto->enter_memory_pressure));
2715 }
2716
2717 static int proto_seq_show(struct seq_file *seq, void *v)
2718 {
2719 if (v == &proto_list)
2720 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2721 "protocol",
2722 "size",
2723 "sockets",
2724 "memory",
2725 "press",
2726 "maxhdr",
2727 "slab",
2728 "module",
2729 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2730 else
2731 proto_seq_printf(seq, list_entry(v, struct proto, node));
2732 return 0;
2733 }
2734
2735 static const struct seq_operations proto_seq_ops = {
2736 .start = proto_seq_start,
2737 .next = proto_seq_next,
2738 .stop = proto_seq_stop,
2739 .show = proto_seq_show,
2740 };
2741
2742 static int proto_seq_open(struct inode *inode, struct file *file)
2743 {
2744 return seq_open_net(inode, file, &proto_seq_ops,
2745 sizeof(struct seq_net_private));
2746 }
2747
2748 static const struct file_operations proto_seq_fops = {
2749 .owner = THIS_MODULE,
2750 .open = proto_seq_open,
2751 .read = seq_read,
2752 .llseek = seq_lseek,
2753 .release = seq_release_net,
2754 };
2755
2756 static __net_init int proto_init_net(struct net *net)
2757 {
2758 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2759 return -ENOMEM;
2760
2761 return 0;
2762 }
2763
2764 static __net_exit void proto_exit_net(struct net *net)
2765 {
2766 proc_net_remove(net, "protocols");
2767 }
2768
2769
2770 static __net_initdata struct pernet_operations proto_net_ops = {
2771 .init = proto_init_net,
2772 .exit = proto_exit_net,
2773 };
2774
2775 static int __init proto_init(void)
2776 {
2777 return register_pernet_subsys(&proto_net_ops);
2778 }
2779
2780 subsys_initcall(proto_init);
2781
2782 #endif /* PROC_FS */
Linus' kernel tree