?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
cls_cgroup: remove task_struct parameter from sock_update_classid()
[linux-2.6.git] / net / core / sock.c
blob0e1d2fe827ec8c3a985d7cc322d058c53a119b5d
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 #if defined(CONFIG_MEMCG_KMEM)
190 struct static_key memcg_socket_limit_enabled;
191 EXPORT_SYMBOL(memcg_socket_limit_enabled);
192 #endif
193
194 /*
195 * Make lock validator output more readable. (we pre-construct these
196 * strings build-time, so that runtime initialization of socket
197 * locks is fast):
198 */
199 static const char *const af_family_key_strings[AF_MAX+1] = {
200 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
201 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
202 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
203 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
204 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
205 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
206 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
207 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
208 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
209 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
210 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
211 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
212 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
213 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
214 };
215 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
216 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
217 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
218 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
219 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
220 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
221 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
222 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
223 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
224 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
225 "slock-27" , "slock-28" , "slock-AF_CAN" ,
226 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
227 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
228 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
229 "slock-AF_NFC" , "slock-AF_MAX"
230 };
231 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
232 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
233 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
234 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
235 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
236 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
237 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
238 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
239 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
240 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
241 "clock-27" , "clock-28" , "clock-AF_CAN" ,
242 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
243 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
244 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
245 "clock-AF_NFC" , "clock-AF_MAX"
246 };
247
248 /*
249 * sk_callback_lock locking rules are per-address-family,
250 * so split the lock classes by using a per-AF key:
251 */
252 static struct lock_class_key af_callback_keys[AF_MAX];
253
254 /* Take into consideration the size of the struct sk_buff overhead in the
255 * determination of these values, since that is non-constant across
256 * platforms. This makes socket queueing behavior and performance
257 * not depend upon such differences.
258 */
259 #define _SK_MEM_PACKETS 256
260 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
261 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
263
264 /* Run time adjustable parameters. */
265 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
266 EXPORT_SYMBOL(sysctl_wmem_max);
267 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
268 EXPORT_SYMBOL(sysctl_rmem_max);
269 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
270 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
271
272 /* Maximal space eaten by iovec or ancillary data plus some space */
273 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
274 EXPORT_SYMBOL(sysctl_optmem_max);
275
276 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
277 EXPORT_SYMBOL_GPL(memalloc_socks);
278
279 /**
280 * sk_set_memalloc - sets %SOCK_MEMALLOC
281 * @sk: socket to set it on
282 *
283 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
284 * It's the responsibility of the admin to adjust min_free_kbytes
285 * to meet the requirements
286 */
287 void sk_set_memalloc(struct sock *sk)
288 {
289 sock_set_flag(sk, SOCK_MEMALLOC);
290 sk->sk_allocation |= __GFP_MEMALLOC;
291 static_key_slow_inc(&memalloc_socks);
292 }
293 EXPORT_SYMBOL_GPL(sk_set_memalloc);
294
295 void sk_clear_memalloc(struct sock *sk)
296 {
297 sock_reset_flag(sk, SOCK_MEMALLOC);
298 sk->sk_allocation &= ~__GFP_MEMALLOC;
299 static_key_slow_dec(&memalloc_socks);
300
301 /*
302 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
303 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
304 * it has rmem allocations there is a risk that the user of the
305 * socket cannot make forward progress due to exceeding the rmem
306 * limits. By rights, sk_clear_memalloc() should only be called
307 * on sockets being torn down but warn and reset the accounting if
308 * that assumption breaks.
309 */
310 if (WARN_ON(sk->sk_forward_alloc))
311 sk_mem_reclaim(sk);
312 }
313 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
314
315 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
316 {
317 int ret;
318 unsigned long pflags = current->flags;
319
320 /* these should have been dropped before queueing */
321 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
322
323 current->flags |= PF_MEMALLOC;
324 ret = sk->sk_backlog_rcv(sk, skb);
325 tsk_restore_flags(current, pflags, PF_MEMALLOC);
326
327 return ret;
328 }
329 EXPORT_SYMBOL(__sk_backlog_rcv);
330
331 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
332 {
333 struct timeval tv;
334
335 if (optlen < sizeof(tv))
336 return -EINVAL;
337 if (copy_from_user(&tv, optval, sizeof(tv)))
338 return -EFAULT;
339 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
340 return -EDOM;
341
342 if (tv.tv_sec < 0) {
343 static int warned __read_mostly;
344
345 *timeo_p = 0;
346 if (warned < 10 && net_ratelimit()) {
347 warned++;
348 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
349 __func__, current->comm, task_pid_nr(current));
350 }
351 return 0;
352 }
353 *timeo_p = MAX_SCHEDULE_TIMEOUT;
354 if (tv.tv_sec == 0 && tv.tv_usec == 0)
355 return 0;
356 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
357 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
358 return 0;
359 }
360
361 static void sock_warn_obsolete_bsdism(const char *name)
362 {
363 static int warned;
364 static char warncomm[TASK_COMM_LEN];
365 if (strcmp(warncomm, current->comm) && warned < 5) {
366 strcpy(warncomm, current->comm);
367 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
368 warncomm, name);
369 warned++;
370 }
371 }
372
373 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
374
375 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
376 {
377 if (sk->sk_flags & flags) {
378 sk->sk_flags &= ~flags;
379 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
380 net_disable_timestamp();
381 }
382 }
383
384
385 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
386 {
387 int err;
388 int skb_len;
389 unsigned long flags;
390 struct sk_buff_head *list = &sk->sk_receive_queue;
391
392 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
393 atomic_inc(&sk->sk_drops);
394 trace_sock_rcvqueue_full(sk, skb);
395 return -ENOMEM;
396 }
397
398 err = sk_filter(sk, skb);
399 if (err)
400 return err;
401
402 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
403 atomic_inc(&sk->sk_drops);
404 return -ENOBUFS;
405 }
406
407 skb->dev = NULL;
408 skb_set_owner_r(skb, sk);
409
410 /* Cache the SKB length before we tack it onto the receive
411 * queue. Once it is added it no longer belongs to us and
412 * may be freed by other threads of control pulling packets
413 * from the queue.
414 */
415 skb_len = skb->len;
416
417 /* we escape from rcu protected region, make sure we dont leak
418 * a norefcounted dst
419 */
420 skb_dst_force(skb);
421
422 spin_lock_irqsave(&list->lock, flags);
423 skb->dropcount = atomic_read(&sk->sk_drops);
424 __skb_queue_tail(list, skb);
425 spin_unlock_irqrestore(&list->lock, flags);
426
427 if (!sock_flag(sk, SOCK_DEAD))
428 sk->sk_data_ready(sk, skb_len);
429 return 0;
430 }
431 EXPORT_SYMBOL(sock_queue_rcv_skb);
432
433 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
434 {
435 int rc = NET_RX_SUCCESS;
436
437 if (sk_filter(sk, skb))
438 goto discard_and_relse;
439
440 skb->dev = NULL;
441
442 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
443 atomic_inc(&sk->sk_drops);
444 goto discard_and_relse;
445 }
446 if (nested)
447 bh_lock_sock_nested(sk);
448 else
449 bh_lock_sock(sk);
450 if (!sock_owned_by_user(sk)) {
451 /*
452 * trylock + unlock semantics:
453 */
454 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
455
456 rc = sk_backlog_rcv(sk, skb);
457
458 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
459 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
460 bh_unlock_sock(sk);
461 atomic_inc(&sk->sk_drops);
462 goto discard_and_relse;
463 }
464
465 bh_unlock_sock(sk);
466 out:
467 sock_put(sk);
468 return rc;
469 discard_and_relse:
470 kfree_skb(skb);
471 goto out;
472 }
473 EXPORT_SYMBOL(sk_receive_skb);
474
475 void sk_reset_txq(struct sock *sk)
476 {
477 sk_tx_queue_clear(sk);
478 }
479 EXPORT_SYMBOL(sk_reset_txq);
480
481 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
482 {
483 struct dst_entry *dst = __sk_dst_get(sk);
484
485 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
486 sk_tx_queue_clear(sk);
487 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
488 dst_release(dst);
489 return NULL;
490 }
491
492 return dst;
493 }
494 EXPORT_SYMBOL(__sk_dst_check);
495
496 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
497 {
498 struct dst_entry *dst = sk_dst_get(sk);
499
500 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
501 sk_dst_reset(sk);
502 dst_release(dst);
503 return NULL;
504 }
505
506 return dst;
507 }
508 EXPORT_SYMBOL(sk_dst_check);
509
510 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
511 int optlen)
512 {
513 int ret = -ENOPROTOOPT;
514 #ifdef CONFIG_NETDEVICES
515 struct net *net = sock_net(sk);
516 char devname[IFNAMSIZ];
517 int index;
518
519 /* Sorry... */
520 ret = -EPERM;
521 if (!ns_capable(net->user_ns, CAP_NET_RAW))
522 goto out;
523
524 ret = -EINVAL;
525 if (optlen < 0)
526 goto out;
527
528 /* Bind this socket to a particular device like "eth0",
529 * as specified in the passed interface name. If the
530 * name is "" or the option length is zero the socket
531 * is not bound.
532 */
533 if (optlen > IFNAMSIZ - 1)
534 optlen = IFNAMSIZ - 1;
535 memset(devname, 0, sizeof(devname));
536
537 ret = -EFAULT;
538 if (copy_from_user(devname, optval, optlen))
539 goto out;
540
541 index = 0;
542 if (devname[0] != '\0') {
543 struct net_device *dev;
544
545 rcu_read_lock();
546 dev = dev_get_by_name_rcu(net, devname);
547 if (dev)
548 index = dev->ifindex;
549 rcu_read_unlock();
550 ret = -ENODEV;
551 if (!dev)
552 goto out;
553 }
554
555 lock_sock(sk);
556 sk->sk_bound_dev_if = index;
557 sk_dst_reset(sk);
558 release_sock(sk);
559
560 ret = 0;
561
562 out:
563 #endif
564
565 return ret;
566 }
567
568 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
569 int __user *optlen, int len)
570 {
571 int ret = -ENOPROTOOPT;
572 #ifdef CONFIG_NETDEVICES
573 struct net *net = sock_net(sk);
574 struct net_device *dev;
575 char devname[IFNAMSIZ];
576 unsigned seq;
577
578 if (sk->sk_bound_dev_if == 0) {
579 len = 0;
580 goto zero;
581 }
582
583 ret = -EINVAL;
584 if (len < IFNAMSIZ)
585 goto out;
586
587 retry:
588 seq = read_seqcount_begin(&devnet_rename_seq);
589 rcu_read_lock();
590 dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
591 ret = -ENODEV;
592 if (!dev) {
593 rcu_read_unlock();
594 goto out;
595 }
596
597 strcpy(devname, dev->name);
598 rcu_read_unlock();
599 if (read_seqcount_retry(&devnet_rename_seq, seq))
600 goto retry;
601
602 len = strlen(devname) + 1;
603
604 ret = -EFAULT;
605 if (copy_to_user(optval, devname, len))
606 goto out;
607
608 zero:
609 ret = -EFAULT;
610 if (put_user(len, optlen))
611 goto out;
612
613 ret = 0;
614
615 out:
616 #endif
617
618 return ret;
619 }
620
621 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
622 {
623 if (valbool)
624 sock_set_flag(sk, bit);
625 else
626 sock_reset_flag(sk, bit);
627 }
628
629 /*
630 * This is meant for all protocols to use and covers goings on
631 * at the socket level. Everything here is generic.
632 */
633
634 int sock_setsockopt(struct socket *sock, int level, int optname,
635 char __user *optval, unsigned int optlen)
636 {
637 struct sock *sk = sock->sk;
638 int val;
639 int valbool;
640 struct linger ling;
641 int ret = 0;
642
643 /*
644 * Options without arguments
645 */
646
647 if (optname == SO_BINDTODEVICE)
648 return sock_setbindtodevice(sk, optval, optlen);
649
650 if (optlen < sizeof(int))
651 return -EINVAL;
652
653 if (get_user(val, (int __user *)optval))
654 return -EFAULT;
655
656 valbool = val ? 1 : 0;
657
658 lock_sock(sk);
659
660 switch (optname) {
661 case SO_DEBUG:
662 if (val && !capable(CAP_NET_ADMIN))
663 ret = -EACCES;
664 else
665 sock_valbool_flag(sk, SOCK_DBG, valbool);
666 break;
667 case SO_REUSEADDR:
668 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
669 break;
670 case SO_REUSEPORT:
671 sk->sk_reuseport = valbool;
672 break;
673 case SO_TYPE:
674 case SO_PROTOCOL:
675 case SO_DOMAIN:
676 case SO_ERROR:
677 ret = -ENOPROTOOPT;
678 break;
679 case SO_DONTROUTE:
680 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
681 break;
682 case SO_BROADCAST:
683 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
684 break;
685 case SO_SNDBUF:
686 /* Don't error on this BSD doesn't and if you think
687 * about it this is right. Otherwise apps have to
688 * play 'guess the biggest size' games. RCVBUF/SNDBUF
689 * are treated in BSD as hints
690 */
691 val = min_t(u32, val, sysctl_wmem_max);
692 set_sndbuf:
693 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
694 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
695 /* Wake up sending tasks if we upped the value. */
696 sk->sk_write_space(sk);
697 break;
698
699 case SO_SNDBUFFORCE:
700 if (!capable(CAP_NET_ADMIN)) {
701 ret = -EPERM;
702 break;
703 }
704 goto set_sndbuf;
705
706 case SO_RCVBUF:
707 /* Don't error on this BSD doesn't and if you think
708 * about it this is right. Otherwise apps have to
709 * play 'guess the biggest size' games. RCVBUF/SNDBUF
710 * are treated in BSD as hints
711 */
712 val = min_t(u32, val, sysctl_rmem_max);
713 set_rcvbuf:
714 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
715 /*
716 * We double it on the way in to account for
717 * "struct sk_buff" etc. overhead. Applications
718 * assume that the SO_RCVBUF setting they make will
719 * allow that much actual data to be received on that
720 * socket.
721 *
722 * Applications are unaware that "struct sk_buff" and
723 * other overheads allocate from the receive buffer
724 * during socket buffer allocation.
725 *
726 * And after considering the possible alternatives,
727 * returning the value we actually used in getsockopt
728 * is the most desirable behavior.
729 */
730 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
731 break;
732
733 case SO_RCVBUFFORCE:
734 if (!capable(CAP_NET_ADMIN)) {
735 ret = -EPERM;
736 break;
737 }
738 goto set_rcvbuf;
739
740 case SO_KEEPALIVE:
741 #ifdef CONFIG_INET
742 if (sk->sk_protocol == IPPROTO_TCP &&
743 sk->sk_type == SOCK_STREAM)
744 tcp_set_keepalive(sk, valbool);
745 #endif
746 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
747 break;
748
749 case SO_OOBINLINE:
750 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
751 break;
752
753 case SO_NO_CHECK:
754 sk->sk_no_check = valbool;
755 break;
756
757 case SO_PRIORITY:
758 if ((val >= 0 && val <= 6) ||
759 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
760 sk->sk_priority = val;
761 else
762 ret = -EPERM;
763 break;
764
765 case SO_LINGER:
766 if (optlen < sizeof(ling)) {
767 ret = -EINVAL; /* 1003.1g */
768 break;
769 }
770 if (copy_from_user(&ling, optval, sizeof(ling))) {
771 ret = -EFAULT;
772 break;
773 }
774 if (!ling.l_onoff)
775 sock_reset_flag(sk, SOCK_LINGER);
776 else {
777 #if (BITS_PER_LONG == 32)
778 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
779 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
780 else
781 #endif
782 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
783 sock_set_flag(sk, SOCK_LINGER);
784 }
785 break;
786
787 case SO_BSDCOMPAT:
788 sock_warn_obsolete_bsdism("setsockopt");
789 break;
790
791 case SO_PASSCRED:
792 if (valbool)
793 set_bit(SOCK_PASSCRED, &sock->flags);
794 else
795 clear_bit(SOCK_PASSCRED, &sock->flags);
796 break;
797
798 case SO_TIMESTAMP:
799 case SO_TIMESTAMPNS:
800 if (valbool) {
801 if (optname == SO_TIMESTAMP)
802 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
803 else
804 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
805 sock_set_flag(sk, SOCK_RCVTSTAMP);
806 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
807 } else {
808 sock_reset_flag(sk, SOCK_RCVTSTAMP);
809 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
810 }
811 break;
812
813 case SO_TIMESTAMPING:
814 if (val & ~SOF_TIMESTAMPING_MASK) {
815 ret = -EINVAL;
816 break;
817 }
818 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
819 val & SOF_TIMESTAMPING_TX_HARDWARE);
820 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
821 val & SOF_TIMESTAMPING_TX_SOFTWARE);
822 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
823 val & SOF_TIMESTAMPING_RX_HARDWARE);
824 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
825 sock_enable_timestamp(sk,
826 SOCK_TIMESTAMPING_RX_SOFTWARE);
827 else
828 sock_disable_timestamp(sk,
829 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
830 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
831 val & SOF_TIMESTAMPING_SOFTWARE);
832 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
833 val & SOF_TIMESTAMPING_SYS_HARDWARE);
834 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
835 val & SOF_TIMESTAMPING_RAW_HARDWARE);
836 break;
837
838 case SO_RCVLOWAT:
839 if (val < 0)
840 val = INT_MAX;
841 sk->sk_rcvlowat = val ? : 1;
842 break;
843
844 case SO_RCVTIMEO:
845 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
846 break;
847
848 case SO_SNDTIMEO:
849 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
850 break;
851
852 case SO_ATTACH_FILTER:
853 ret = -EINVAL;
854 if (optlen == sizeof(struct sock_fprog)) {
855 struct sock_fprog fprog;
856
857 ret = -EFAULT;
858 if (copy_from_user(&fprog, optval, sizeof(fprog)))
859 break;
860
861 ret = sk_attach_filter(&fprog, sk);
862 }
863 break;
864
865 case SO_DETACH_FILTER:
866 ret = sk_detach_filter(sk);
867 break;
868
869 case SO_LOCK_FILTER:
870 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
871 ret = -EPERM;
872 else
873 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
874 break;
875
876 case SO_PASSSEC:
877 if (valbool)
878 set_bit(SOCK_PASSSEC, &sock->flags);
879 else
880 clear_bit(SOCK_PASSSEC, &sock->flags);
881 break;
882 case SO_MARK:
883 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
884 ret = -EPERM;
885 else
886 sk->sk_mark = val;
887 break;
888
889 /* We implement the SO_SNDLOWAT etc to
890 not be settable (1003.1g 5.3) */
891 case SO_RXQ_OVFL:
892 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
893 break;
894
895 case SO_WIFI_STATUS:
896 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
897 break;
898
899 case SO_PEEK_OFF:
900 if (sock->ops->set_peek_off)
901 sock->ops->set_peek_off(sk, val);
902 else
903 ret = -EOPNOTSUPP;
904 break;
905
906 case SO_NOFCS:
907 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
908 break;
909
910 case SO_SELECT_ERR_QUEUE:
911 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
912 break;
913
914 default:
915 ret = -ENOPROTOOPT;
916 break;
917 }
918 release_sock(sk);
919 return ret;
920 }
921 EXPORT_SYMBOL(sock_setsockopt);
922
923
924 void cred_to_ucred(struct pid *pid, const struct cred *cred,
925 struct ucred *ucred)
926 {
927 ucred->pid = pid_vnr(pid);
928 ucred->uid = ucred->gid = -1;
929 if (cred) {
930 struct user_namespace *current_ns = current_user_ns();
931
932 ucred->uid = from_kuid_munged(current_ns, cred->euid);
933 ucred->gid = from_kgid_munged(current_ns, cred->egid);
934 }
935 }
936 EXPORT_SYMBOL_GPL(cred_to_ucred);
937
938 int sock_getsockopt(struct socket *sock, int level, int optname,
939 char __user *optval, int __user *optlen)
940 {
941 struct sock *sk = sock->sk;
942
943 union {
944 int val;
945 struct linger ling;
946 struct timeval tm;
947 } v;
948
949 int lv = sizeof(int);
950 int len;
951
952 if (get_user(len, optlen))
953 return -EFAULT;
954 if (len < 0)
955 return -EINVAL;
956
957 memset(&v, 0, sizeof(v));
958
959 switch (optname) {
960 case SO_DEBUG:
961 v.val = sock_flag(sk, SOCK_DBG);
962 break;
963
964 case SO_DONTROUTE:
965 v.val = sock_flag(sk, SOCK_LOCALROUTE);
966 break;
967
968 case SO_BROADCAST:
969 v.val = sock_flag(sk, SOCK_BROADCAST);
970 break;
971
972 case SO_SNDBUF:
973 v.val = sk->sk_sndbuf;
974 break;
975
976 case SO_RCVBUF:
977 v.val = sk->sk_rcvbuf;
978 break;
979
980 case SO_REUSEADDR:
981 v.val = sk->sk_reuse;
982 break;
983
984 case SO_REUSEPORT:
985 v.val = sk->sk_reuseport;
986 break;
987
988 case SO_KEEPALIVE:
989 v.val = sock_flag(sk, SOCK_KEEPOPEN);
990 break;
991
992 case SO_TYPE:
993 v.val = sk->sk_type;
994 break;
995
996 case SO_PROTOCOL:
997 v.val = sk->sk_protocol;
998 break;
999
1000 case SO_DOMAIN:
1001 v.val = sk->sk_family;
1002 break;
1003
1004 case SO_ERROR:
1005 v.val = -sock_error(sk);
1006 if (v.val == 0)
1007 v.val = xchg(&sk->sk_err_soft, 0);
1008 break;
1009
1010 case SO_OOBINLINE:
1011 v.val = sock_flag(sk, SOCK_URGINLINE);
1012 break;
1013
1014 case SO_NO_CHECK:
1015 v.val = sk->sk_no_check;
1016 break;
1017
1018 case SO_PRIORITY:
1019 v.val = sk->sk_priority;
1020 break;
1021
1022 case SO_LINGER:
1023 lv = sizeof(v.ling);
1024 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1025 v.ling.l_linger = sk->sk_lingertime / HZ;
1026 break;
1027
1028 case SO_BSDCOMPAT:
1029 sock_warn_obsolete_bsdism("getsockopt");
1030 break;
1031
1032 case SO_TIMESTAMP:
1033 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1034 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1035 break;
1036
1037 case SO_TIMESTAMPNS:
1038 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1039 break;
1040
1041 case SO_TIMESTAMPING:
1042 v.val = 0;
1043 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1044 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1045 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1046 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1047 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1048 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1049 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1050 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1051 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1052 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1053 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1054 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1055 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1056 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1057 break;
1058
1059 case SO_RCVTIMEO:
1060 lv = sizeof(struct timeval);
1061 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1062 v.tm.tv_sec = 0;
1063 v.tm.tv_usec = 0;
1064 } else {
1065 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1066 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1067 }
1068 break;
1069
1070 case SO_SNDTIMEO:
1071 lv = sizeof(struct timeval);
1072 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1073 v.tm.tv_sec = 0;
1074 v.tm.tv_usec = 0;
1075 } else {
1076 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1077 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1078 }
1079 break;
1080
1081 case SO_RCVLOWAT:
1082 v.val = sk->sk_rcvlowat;
1083 break;
1084
1085 case SO_SNDLOWAT:
1086 v.val = 1;
1087 break;
1088
1089 case SO_PASSCRED:
1090 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1091 break;
1092
1093 case SO_PEERCRED:
1094 {
1095 struct ucred peercred;
1096 if (len > sizeof(peercred))
1097 len = sizeof(peercred);
1098 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1099 if (copy_to_user(optval, &peercred, len))
1100 return -EFAULT;
1101 goto lenout;
1102 }
1103
1104 case SO_PEERNAME:
1105 {
1106 char address[128];
1107
1108 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1109 return -ENOTCONN;
1110 if (lv < len)
1111 return -EINVAL;
1112 if (copy_to_user(optval, address, len))
1113 return -EFAULT;
1114 goto lenout;
1115 }
1116
1117 /* Dubious BSD thing... Probably nobody even uses it, but
1118 * the UNIX standard wants it for whatever reason... -DaveM
1119 */
1120 case SO_ACCEPTCONN:
1121 v.val = sk->sk_state == TCP_LISTEN;
1122 break;
1123
1124 case SO_PASSSEC:
1125 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1126 break;
1127
1128 case SO_PEERSEC:
1129 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1130
1131 case SO_MARK:
1132 v.val = sk->sk_mark;
1133 break;
1134
1135 case SO_RXQ_OVFL:
1136 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1137 break;
1138
1139 case SO_WIFI_STATUS:
1140 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1141 break;
1142
1143 case SO_PEEK_OFF:
1144 if (!sock->ops->set_peek_off)
1145 return -EOPNOTSUPP;
1146
1147 v.val = sk->sk_peek_off;
1148 break;
1149 case SO_NOFCS:
1150 v.val = sock_flag(sk, SOCK_NOFCS);
1151 break;
1152
1153 case SO_BINDTODEVICE:
1154 return sock_getbindtodevice(sk, optval, optlen, len);
1155
1156 case SO_GET_FILTER:
1157 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1158 if (len < 0)
1159 return len;
1160
1161 goto lenout;
1162
1163 case SO_LOCK_FILTER:
1164 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1165 break;
1166
1167 case SO_SELECT_ERR_QUEUE:
1168 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1169 break;
1170
1171 default:
1172 return -ENOPROTOOPT;
1173 }
1174
1175 if (len > lv)
1176 len = lv;
1177 if (copy_to_user(optval, &v, len))
1178 return -EFAULT;
1179 lenout:
1180 if (put_user(len, optlen))
1181 return -EFAULT;
1182 return 0;
1183 }
1184
1185 /*
1186 * Initialize an sk_lock.
1187 *
1188 * (We also register the sk_lock with the lock validator.)
1189 */
1190 static inline void sock_lock_init(struct sock *sk)
1191 {
1192 sock_lock_init_class_and_name(sk,
1193 af_family_slock_key_strings[sk->sk_family],
1194 af_family_slock_keys + sk->sk_family,
1195 af_family_key_strings[sk->sk_family],
1196 af_family_keys + sk->sk_family);
1197 }
1198
1199 /*
1200 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1201 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1202 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1203 */
1204 static void sock_copy(struct sock *nsk, const struct sock *osk)
1205 {
1206 #ifdef CONFIG_SECURITY_NETWORK
1207 void *sptr = nsk->sk_security;
1208 #endif
1209 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1210
1211 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1212 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1213
1214 #ifdef CONFIG_SECURITY_NETWORK
1215 nsk->sk_security = sptr;
1216 security_sk_clone(osk, nsk);
1217 #endif
1218 }
1219
1220 /*
1221 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1222 * un-modified. Special care is taken when initializing object to zero.
1223 */
1224 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1225 {
1226 if (offsetof(struct sock, sk_node.next) != 0)
1227 memset(sk, 0, offsetof(struct sock, sk_node.next));
1228 memset(&sk->sk_node.pprev, 0,
1229 size - offsetof(struct sock, sk_node.pprev));
1230 }
1231
1232 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1233 {
1234 unsigned long nulls1, nulls2;
1235
1236 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1237 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1238 if (nulls1 > nulls2)
1239 swap(nulls1, nulls2);
1240
1241 if (nulls1 != 0)
1242 memset((char *)sk, 0, nulls1);
1243 memset((char *)sk + nulls1 + sizeof(void *), 0,
1244 nulls2 - nulls1 - sizeof(void *));
1245 memset((char *)sk + nulls2 + sizeof(void *), 0,
1246 size - nulls2 - sizeof(void *));
1247 }
1248 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1249
1250 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1251 int family)
1252 {
1253 struct sock *sk;
1254 struct kmem_cache *slab;
1255
1256 slab = prot->slab;
1257 if (slab != NULL) {
1258 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1259 if (!sk)
1260 return sk;
1261 if (priority & __GFP_ZERO) {
1262 if (prot->clear_sk)
1263 prot->clear_sk(sk, prot->obj_size);
1264 else
1265 sk_prot_clear_nulls(sk, prot->obj_size);
1266 }
1267 } else
1268 sk = kmalloc(prot->obj_size, priority);
1269
1270 if (sk != NULL) {
1271 kmemcheck_annotate_bitfield(sk, flags);
1272
1273 if (security_sk_alloc(sk, family, priority))
1274 goto out_free;
1275
1276 if (!try_module_get(prot->owner))
1277 goto out_free_sec;
1278 sk_tx_queue_clear(sk);
1279 }
1280
1281 return sk;
1282
1283 out_free_sec:
1284 security_sk_free(sk);
1285 out_free:
1286 if (slab != NULL)
1287 kmem_cache_free(slab, sk);
1288 else
1289 kfree(sk);
1290 return NULL;
1291 }
1292
1293 static void sk_prot_free(struct proto *prot, struct sock *sk)
1294 {
1295 struct kmem_cache *slab;
1296 struct module *owner;
1297
1298 owner = prot->owner;
1299 slab = prot->slab;
1300
1301 security_sk_free(sk);
1302 if (slab != NULL)
1303 kmem_cache_free(slab, sk);
1304 else
1305 kfree(sk);
1306 module_put(owner);
1307 }
1308
1309 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1310 void sock_update_classid(struct sock *sk)
1311 {
1312 u32 classid;
1313
1314 classid = task_cls_classid(current);
1315 if (classid != sk->sk_classid)
1316 sk->sk_classid = classid;
1317 }
1318 EXPORT_SYMBOL(sock_update_classid);
1319 #endif
1320
1321 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1322 void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
1323 {
1324 if (in_interrupt())
1325 return;
1326
1327 sk->sk_cgrp_prioidx = task_netprioidx(task);
1328 }
1329 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1330 #endif
1331
1332 /**
1333 * sk_alloc - All socket objects are allocated here
1334 * @net: the applicable net namespace
1335 * @family: protocol family
1336 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1337 * @prot: struct proto associated with this new sock instance
1338 */
1339 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1340 struct proto *prot)
1341 {
1342 struct sock *sk;
1343
1344 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1345 if (sk) {
1346 sk->sk_family = family;
1347 /*
1348 * See comment in struct sock definition to understand
1349 * why we need sk_prot_creator -acme
1350 */
1351 sk->sk_prot = sk->sk_prot_creator = prot;
1352 sock_lock_init(sk);
1353 sock_net_set(sk, get_net(net));
1354 atomic_set(&sk->sk_wmem_alloc, 1);
1355
1356 sock_update_classid(sk);
1357 sock_update_netprioidx(sk, current);
1358 }
1359
1360 return sk;
1361 }
1362 EXPORT_SYMBOL(sk_alloc);
1363
1364 static void __sk_free(struct sock *sk)
1365 {
1366 struct sk_filter *filter;
1367
1368 if (sk->sk_destruct)
1369 sk->sk_destruct(sk);
1370
1371 filter = rcu_dereference_check(sk->sk_filter,
1372 atomic_read(&sk->sk_wmem_alloc) == 0);
1373 if (filter) {
1374 sk_filter_uncharge(sk, filter);
1375 RCU_INIT_POINTER(sk->sk_filter, NULL);
1376 }
1377
1378 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1379
1380 if (atomic_read(&sk->sk_omem_alloc))
1381 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1382 __func__, atomic_read(&sk->sk_omem_alloc));
1383
1384 if (sk->sk_peer_cred)
1385 put_cred(sk->sk_peer_cred);
1386 put_pid(sk->sk_peer_pid);
1387 put_net(sock_net(sk));
1388 sk_prot_free(sk->sk_prot_creator, sk);
1389 }
1390
1391 void sk_free(struct sock *sk)
1392 {
1393 /*
1394 * We subtract one from sk_wmem_alloc and can know if
1395 * some packets are still in some tx queue.
1396 * If not null, sock_wfree() will call __sk_free(sk) later
1397 */
1398 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1399 __sk_free(sk);
1400 }
1401 EXPORT_SYMBOL(sk_free);
1402
1403 /*
1404 * Last sock_put should drop reference to sk->sk_net. It has already
1405 * been dropped in sk_change_net. Taking reference to stopping namespace
1406 * is not an option.
1407 * Take reference to a socket to remove it from hash _alive_ and after that
1408 * destroy it in the context of init_net.
1409 */
1410 void sk_release_kernel(struct sock *sk)
1411 {
1412 if (sk == NULL || sk->sk_socket == NULL)
1413 return;
1414
1415 sock_hold(sk);
1416 sock_release(sk->sk_socket);
1417 release_net(sock_net(sk));
1418 sock_net_set(sk, get_net(&init_net));
1419 sock_put(sk);
1420 }
1421 EXPORT_SYMBOL(sk_release_kernel);
1422
1423 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1424 {
1425 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1426 sock_update_memcg(newsk);
1427 }
1428
1429 /**
1430 * sk_clone_lock - clone a socket, and lock its clone
1431 * @sk: the socket to clone
1432 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1433 *
1434 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1435 */
1436 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1437 {
1438 struct sock *newsk;
1439
1440 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1441 if (newsk != NULL) {
1442 struct sk_filter *filter;
1443
1444 sock_copy(newsk, sk);
1445
1446 /* SANITY */
1447 get_net(sock_net(newsk));
1448 sk_node_init(&newsk->sk_node);
1449 sock_lock_init(newsk);
1450 bh_lock_sock(newsk);
1451 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1452 newsk->sk_backlog.len = 0;
1453
1454 atomic_set(&newsk->sk_rmem_alloc, 0);
1455 /*
1456 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1457 */
1458 atomic_set(&newsk->sk_wmem_alloc, 1);
1459 atomic_set(&newsk->sk_omem_alloc, 0);
1460 skb_queue_head_init(&newsk->sk_receive_queue);
1461 skb_queue_head_init(&newsk->sk_write_queue);
1462 #ifdef CONFIG_NET_DMA
1463 skb_queue_head_init(&newsk->sk_async_wait_queue);
1464 #endif
1465
1466 spin_lock_init(&newsk->sk_dst_lock);
1467 rwlock_init(&newsk->sk_callback_lock);
1468 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1469 af_callback_keys + newsk->sk_family,
1470 af_family_clock_key_strings[newsk->sk_family]);
1471
1472 newsk->sk_dst_cache = NULL;
1473 newsk->sk_wmem_queued = 0;
1474 newsk->sk_forward_alloc = 0;
1475 newsk->sk_send_head = NULL;
1476 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1477
1478 sock_reset_flag(newsk, SOCK_DONE);
1479 skb_queue_head_init(&newsk->sk_error_queue);
1480
1481 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1482 if (filter != NULL)
1483 sk_filter_charge(newsk, filter);
1484
1485 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1486 /* It is still raw copy of parent, so invalidate
1487 * destructor and make plain sk_free() */
1488 newsk->sk_destruct = NULL;
1489 bh_unlock_sock(newsk);
1490 sk_free(newsk);
1491 newsk = NULL;
1492 goto out;
1493 }
1494
1495 newsk->sk_err = 0;
1496 newsk->sk_priority = 0;
1497 /*
1498 * Before updating sk_refcnt, we must commit prior changes to memory
1499 * (Documentation/RCU/rculist_nulls.txt for details)
1500 */
1501 smp_wmb();
1502 atomic_set(&newsk->sk_refcnt, 2);
1503
1504 /*
1505 * Increment the counter in the same struct proto as the master
1506 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1507 * is the same as sk->sk_prot->socks, as this field was copied
1508 * with memcpy).
1509 *
1510 * This _changes_ the previous behaviour, where
1511 * tcp_create_openreq_child always was incrementing the
1512 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1513 * to be taken into account in all callers. -acme
1514 */
1515 sk_refcnt_debug_inc(newsk);
1516 sk_set_socket(newsk, NULL);
1517 newsk->sk_wq = NULL;
1518
1519 sk_update_clone(sk, newsk);
1520
1521 if (newsk->sk_prot->sockets_allocated)
1522 sk_sockets_allocated_inc(newsk);
1523
1524 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1525 net_enable_timestamp();
1526 }
1527 out:
1528 return newsk;
1529 }
1530 EXPORT_SYMBOL_GPL(sk_clone_lock);
1531
1532 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1533 {
1534 __sk_dst_set(sk, dst);
1535 sk->sk_route_caps = dst->dev->features;
1536 if (sk->sk_route_caps & NETIF_F_GSO)
1537 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1538 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1539 if (sk_can_gso(sk)) {
1540 if (dst->header_len) {
1541 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1542 } else {
1543 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1544 sk->sk_gso_max_size = dst->dev->gso_max_size;
1545 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1546 }
1547 }
1548 }
1549 EXPORT_SYMBOL_GPL(sk_setup_caps);
1550
1551 /*
1552 * Simple resource managers for sockets.
1553 */
1554
1555
1556 /*
1557 * Write buffer destructor automatically called from kfree_skb.
1558 */
1559 void sock_wfree(struct sk_buff *skb)
1560 {
1561 struct sock *sk = skb->sk;
1562 unsigned int len = skb->truesize;
1563
1564 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1565 /*
1566 * Keep a reference on sk_wmem_alloc, this will be released
1567 * after sk_write_space() call
1568 */
1569 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1570 sk->sk_write_space(sk);
1571 len = 1;
1572 }
1573 /*
1574 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1575 * could not do because of in-flight packets
1576 */
1577 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1578 __sk_free(sk);
1579 }
1580 EXPORT_SYMBOL(sock_wfree);
1581
1582 /*
1583 * Read buffer destructor automatically called from kfree_skb.
1584 */
1585 void sock_rfree(struct sk_buff *skb)
1586 {
1587 struct sock *sk = skb->sk;
1588 unsigned int len = skb->truesize;
1589
1590 atomic_sub(len, &sk->sk_rmem_alloc);
1591 sk_mem_uncharge(sk, len);
1592 }
1593 EXPORT_SYMBOL(sock_rfree);
1594
1595 void sock_edemux(struct sk_buff *skb)
1596 {
1597 struct sock *sk = skb->sk;
1598
1599 #ifdef CONFIG_INET
1600 if (sk->sk_state == TCP_TIME_WAIT)
1601 inet_twsk_put(inet_twsk(sk));
1602 else
1603 #endif
1604 sock_put(sk);
1605 }
1606 EXPORT_SYMBOL(sock_edemux);
1607
1608 kuid_t sock_i_uid(struct sock *sk)
1609 {
1610 kuid_t uid;
1611
1612 read_lock_bh(&sk->sk_callback_lock);
1613 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1614 read_unlock_bh(&sk->sk_callback_lock);
1615 return uid;
1616 }
1617 EXPORT_SYMBOL(sock_i_uid);
1618
1619 unsigned long sock_i_ino(struct sock *sk)
1620 {
1621 unsigned long ino;
1622
1623 read_lock_bh(&sk->sk_callback_lock);
1624 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1625 read_unlock_bh(&sk->sk_callback_lock);
1626 return ino;
1627 }
1628 EXPORT_SYMBOL(sock_i_ino);
1629
1630 /*
1631 * Allocate a skb from the socket's send buffer.
1632 */
1633 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1634 gfp_t priority)
1635 {
1636 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1637 struct sk_buff *skb = alloc_skb(size, priority);
1638 if (skb) {
1639 skb_set_owner_w(skb, sk);
1640 return skb;
1641 }
1642 }
1643 return NULL;
1644 }
1645 EXPORT_SYMBOL(sock_wmalloc);
1646
1647 /*
1648 * Allocate a skb from the socket's receive buffer.
1649 */
1650 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1651 gfp_t priority)
1652 {
1653 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1654 struct sk_buff *skb = alloc_skb(size, priority);
1655 if (skb) {
1656 skb_set_owner_r(skb, sk);
1657 return skb;
1658 }
1659 }
1660 return NULL;
1661 }
1662
1663 /*
1664 * Allocate a memory block from the socket's option memory buffer.
1665 */
1666 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1667 {
1668 if ((unsigned int)size <= sysctl_optmem_max &&
1669 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1670 void *mem;
1671 /* First do the add, to avoid the race if kmalloc
1672 * might sleep.
1673 */
1674 atomic_add(size, &sk->sk_omem_alloc);
1675 mem = kmalloc(size, priority);
1676 if (mem)
1677 return mem;
1678 atomic_sub(size, &sk->sk_omem_alloc);
1679 }
1680 return NULL;
1681 }
1682 EXPORT_SYMBOL(sock_kmalloc);
1683
1684 /*
1685 * Free an option memory block.
1686 */
1687 void sock_kfree_s(struct sock *sk, void *mem, int size)
1688 {
1689 kfree(mem);
1690 atomic_sub(size, &sk->sk_omem_alloc);
1691 }
1692 EXPORT_SYMBOL(sock_kfree_s);
1693
1694 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1695 I think, these locks should be removed for datagram sockets.
1696 */
1697 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1698 {
1699 DEFINE_WAIT(wait);
1700
1701 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1702 for (;;) {
1703 if (!timeo)
1704 break;
1705 if (signal_pending(current))
1706 break;
1707 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1708 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1709 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1710 break;
1711 if (sk->sk_shutdown & SEND_SHUTDOWN)
1712 break;
1713 if (sk->sk_err)
1714 break;
1715 timeo = schedule_timeout(timeo);
1716 }
1717 finish_wait(sk_sleep(sk), &wait);
1718 return timeo;
1719 }
1720
1721
1722 /*
1723 * Generic send/receive buffer handlers
1724 */
1725
1726 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1727 unsigned long data_len, int noblock,
1728 int *errcode)
1729 {
1730 struct sk_buff *skb;
1731 gfp_t gfp_mask;
1732 long timeo;
1733 int err;
1734 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1735
1736 err = -EMSGSIZE;
1737 if (npages > MAX_SKB_FRAGS)
1738 goto failure;
1739
1740 gfp_mask = sk->sk_allocation;
1741 if (gfp_mask & __GFP_WAIT)
1742 gfp_mask |= __GFP_REPEAT;
1743
1744 timeo = sock_sndtimeo(sk, noblock);
1745 while (1) {
1746 err = sock_error(sk);
1747 if (err != 0)
1748 goto failure;
1749
1750 err = -EPIPE;
1751 if (sk->sk_shutdown & SEND_SHUTDOWN)
1752 goto failure;
1753
1754 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1755 skb = alloc_skb(header_len, gfp_mask);
1756 if (skb) {
1757 int i;
1758
1759 /* No pages, we're done... */
1760 if (!data_len)
1761 break;
1762
1763 skb->truesize += data_len;
1764 skb_shinfo(skb)->nr_frags = npages;
1765 for (i = 0; i < npages; i++) {
1766 struct page *page;
1767
1768 page = alloc_pages(sk->sk_allocation, 0);
1769 if (!page) {
1770 err = -ENOBUFS;
1771 skb_shinfo(skb)->nr_frags = i;
1772 kfree_skb(skb);
1773 goto failure;
1774 }
1775
1776 __skb_fill_page_desc(skb, i,
1777 page, 0,
1778 (data_len >= PAGE_SIZE ?
1779 PAGE_SIZE :
1780 data_len));
1781 data_len -= PAGE_SIZE;
1782 }
1783
1784 /* Full success... */
1785 break;
1786 }
1787 err = -ENOBUFS;
1788 goto failure;
1789 }
1790 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1791 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1792 err = -EAGAIN;
1793 if (!timeo)
1794 goto failure;
1795 if (signal_pending(current))
1796 goto interrupted;
1797 timeo = sock_wait_for_wmem(sk, timeo);
1798 }
1799
1800 skb_set_owner_w(skb, sk);
1801 return skb;
1802
1803 interrupted:
1804 err = sock_intr_errno(timeo);
1805 failure:
1806 *errcode = err;
1807 return NULL;
1808 }
1809 EXPORT_SYMBOL(sock_alloc_send_pskb);
1810
1811 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1812 int noblock, int *errcode)
1813 {
1814 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1815 }
1816 EXPORT_SYMBOL(sock_alloc_send_skb);
1817
1818 /* On 32bit arches, an skb frag is limited to 2^15 */
1819 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1820
1821 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1822 {
1823 int order;
1824
1825 if (pfrag->page) {
1826 if (atomic_read(&pfrag->page->_count) == 1) {
1827 pfrag->offset = 0;
1828 return true;
1829 }
1830 if (pfrag->offset < pfrag->size)
1831 return true;
1832 put_page(pfrag->page);
1833 }
1834
1835 /* We restrict high order allocations to users that can afford to wait */
1836 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1837
1838 do {
1839 gfp_t gfp = sk->sk_allocation;
1840
1841 if (order)
1842 gfp |= __GFP_COMP | __GFP_NOWARN;
1843 pfrag->page = alloc_pages(gfp, order);
1844 if (likely(pfrag->page)) {
1845 pfrag->offset = 0;
1846 pfrag->size = PAGE_SIZE << order;
1847 return true;
1848 }
1849 } while (--order >= 0);
1850
1851 sk_enter_memory_pressure(sk);
1852 sk_stream_moderate_sndbuf(sk);
1853 return false;
1854 }
1855 EXPORT_SYMBOL(sk_page_frag_refill);
1856
1857 static void __lock_sock(struct sock *sk)
1858 __releases(&sk->sk_lock.slock)
1859 __acquires(&sk->sk_lock.slock)
1860 {
1861 DEFINE_WAIT(wait);
1862
1863 for (;;) {
1864 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1865 TASK_UNINTERRUPTIBLE);
1866 spin_unlock_bh(&sk->sk_lock.slock);
1867 schedule();
1868 spin_lock_bh(&sk->sk_lock.slock);
1869 if (!sock_owned_by_user(sk))
1870 break;
1871 }
1872 finish_wait(&sk->sk_lock.wq, &wait);
1873 }
1874
1875 static void __release_sock(struct sock *sk)
1876 __releases(&sk->sk_lock.slock)
1877 __acquires(&sk->sk_lock.slock)
1878 {
1879 struct sk_buff *skb = sk->sk_backlog.head;
1880
1881 do {
1882 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1883 bh_unlock_sock(sk);
1884
1885 do {
1886 struct sk_buff *next = skb->next;
1887
1888 prefetch(next);
1889 WARN_ON_ONCE(skb_dst_is_noref(skb));
1890 skb->next = NULL;
1891 sk_backlog_rcv(sk, skb);
1892
1893 /*
1894 * We are in process context here with softirqs
1895 * disabled, use cond_resched_softirq() to preempt.
1896 * This is safe to do because we've taken the backlog
1897 * queue private:
1898 */
1899 cond_resched_softirq();
1900
1901 skb = next;
1902 } while (skb != NULL);
1903
1904 bh_lock_sock(sk);
1905 } while ((skb = sk->sk_backlog.head) != NULL);
1906
1907 /*
1908 * Doing the zeroing here guarantee we can not loop forever
1909 * while a wild producer attempts to flood us.
1910 */
1911 sk->sk_backlog.len = 0;
1912 }
1913
1914 /**
1915 * sk_wait_data - wait for data to arrive at sk_receive_queue
1916 * @sk: sock to wait on
1917 * @timeo: for how long
1918 *
1919 * Now socket state including sk->sk_err is changed only under lock,
1920 * hence we may omit checks after joining wait queue.
1921 * We check receive queue before schedule() only as optimization;
1922 * it is very likely that release_sock() added new data.
1923 */
1924 int sk_wait_data(struct sock *sk, long *timeo)
1925 {
1926 int rc;
1927 DEFINE_WAIT(wait);
1928
1929 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1930 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1931 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1932 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1933 finish_wait(sk_sleep(sk), &wait);
1934 return rc;
1935 }
1936 EXPORT_SYMBOL(sk_wait_data);
1937
1938 /**
1939 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1940 * @sk: socket
1941 * @size: memory size to allocate
1942 * @kind: allocation type
1943 *
1944 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1945 * rmem allocation. This function assumes that protocols which have
1946 * memory_pressure use sk_wmem_queued as write buffer accounting.
1947 */
1948 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1949 {
1950 struct proto *prot = sk->sk_prot;
1951 int amt = sk_mem_pages(size);
1952 long allocated;
1953 int parent_status = UNDER_LIMIT;
1954
1955 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1956
1957 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1958
1959 /* Under limit. */
1960 if (parent_status == UNDER_LIMIT &&
1961 allocated <= sk_prot_mem_limits(sk, 0)) {
1962 sk_leave_memory_pressure(sk);
1963 return 1;
1964 }
1965
1966 /* Under pressure. (we or our parents) */
1967 if ((parent_status > SOFT_LIMIT) ||
1968 allocated > sk_prot_mem_limits(sk, 1))
1969 sk_enter_memory_pressure(sk);
1970
1971 /* Over hard limit (we or our parents) */
1972 if ((parent_status == OVER_LIMIT) ||
1973 (allocated > sk_prot_mem_limits(sk, 2)))
1974 goto suppress_allocation;
1975
1976 /* guarantee minimum buffer size under pressure */
1977 if (kind == SK_MEM_RECV) {
1978 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1979 return 1;
1980
1981 } else { /* SK_MEM_SEND */
1982 if (sk->sk_type == SOCK_STREAM) {
1983 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1984 return 1;
1985 } else if (atomic_read(&sk->sk_wmem_alloc) <
1986 prot->sysctl_wmem[0])
1987 return 1;
1988 }
1989
1990 if (sk_has_memory_pressure(sk)) {
1991 int alloc;
1992
1993 if (!sk_under_memory_pressure(sk))
1994 return 1;
1995 alloc = sk_sockets_allocated_read_positive(sk);
1996 if (sk_prot_mem_limits(sk, 2) > alloc *
1997 sk_mem_pages(sk->sk_wmem_queued +
1998 atomic_read(&sk->sk_rmem_alloc) +
1999 sk->sk_forward_alloc))
2000 return 1;
2001 }
2002
2003 suppress_allocation:
2004
2005 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2006 sk_stream_moderate_sndbuf(sk);
2007
2008 /* Fail only if socket is _under_ its sndbuf.
2009 * In this case we cannot block, so that we have to fail.
2010 */
2011 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2012 return 1;
2013 }
2014
2015 trace_sock_exceed_buf_limit(sk, prot, allocated);
2016
2017 /* Alas. Undo changes. */
2018 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2019
2020 sk_memory_allocated_sub(sk, amt);
2021
2022 return 0;
2023 }
2024 EXPORT_SYMBOL(__sk_mem_schedule);
2025
2026 /**
2027 * __sk_reclaim - reclaim memory_allocated
2028 * @sk: socket
2029 */
2030 void __sk_mem_reclaim(struct sock *sk)
2031 {
2032 sk_memory_allocated_sub(sk,
2033 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2034 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2035
2036 if (sk_under_memory_pressure(sk) &&
2037 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2038 sk_leave_memory_pressure(sk);
2039 }
2040 EXPORT_SYMBOL(__sk_mem_reclaim);
2041
2042
2043 /*
2044 * Set of default routines for initialising struct proto_ops when
2045 * the protocol does not support a particular function. In certain
2046 * cases where it makes no sense for a protocol to have a "do nothing"
2047 * function, some default processing is provided.
2048 */
2049
2050 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2051 {
2052 return -EOPNOTSUPP;
2053 }
2054 EXPORT_SYMBOL(sock_no_bind);
2055
2056 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2057 int len, int flags)
2058 {
2059 return -EOPNOTSUPP;
2060 }
2061 EXPORT_SYMBOL(sock_no_connect);
2062
2063 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2064 {
2065 return -EOPNOTSUPP;
2066 }
2067 EXPORT_SYMBOL(sock_no_socketpair);
2068
2069 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2070 {
2071 return -EOPNOTSUPP;
2072 }
2073 EXPORT_SYMBOL(sock_no_accept);
2074
2075 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2076 int *len, int peer)
2077 {
2078 return -EOPNOTSUPP;
2079 }
2080 EXPORT_SYMBOL(sock_no_getname);
2081
2082 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2083 {
2084 return 0;
2085 }
2086 EXPORT_SYMBOL(sock_no_poll);
2087
2088 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2089 {
2090 return -EOPNOTSUPP;
2091 }
2092 EXPORT_SYMBOL(sock_no_ioctl);
2093
2094 int sock_no_listen(struct socket *sock, int backlog)
2095 {
2096 return -EOPNOTSUPP;
2097 }
2098 EXPORT_SYMBOL(sock_no_listen);
2099
2100 int sock_no_shutdown(struct socket *sock, int how)
2101 {
2102 return -EOPNOTSUPP;
2103 }
2104 EXPORT_SYMBOL(sock_no_shutdown);
2105
2106 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2107 char __user *optval, unsigned int optlen)
2108 {
2109 return -EOPNOTSUPP;
2110 }
2111 EXPORT_SYMBOL(sock_no_setsockopt);
2112
2113 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2114 char __user *optval, int __user *optlen)
2115 {
2116 return -EOPNOTSUPP;
2117 }
2118 EXPORT_SYMBOL(sock_no_getsockopt);
2119
2120 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2121 size_t len)
2122 {
2123 return -EOPNOTSUPP;
2124 }
2125 EXPORT_SYMBOL(sock_no_sendmsg);
2126
2127 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2128 size_t len, int flags)
2129 {
2130 return -EOPNOTSUPP;
2131 }
2132 EXPORT_SYMBOL(sock_no_recvmsg);
2133
2134 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2135 {
2136 /* Mirror missing mmap method error code */
2137 return -ENODEV;
2138 }
2139 EXPORT_SYMBOL(sock_no_mmap);
2140
2141 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2142 {
2143 ssize_t res;
2144 struct msghdr msg = {.msg_flags = flags};
2145 struct kvec iov;
2146 char *kaddr = kmap(page);
2147 iov.iov_base = kaddr + offset;
2148 iov.iov_len = size;
2149 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2150 kunmap(page);
2151 return res;
2152 }
2153 EXPORT_SYMBOL(sock_no_sendpage);
2154
2155 /*
2156 * Default Socket Callbacks
2157 */
2158
2159 static void sock_def_wakeup(struct sock *sk)
2160 {
2161 struct socket_wq *wq;
2162
2163 rcu_read_lock();
2164 wq = rcu_dereference(sk->sk_wq);
2165 if (wq_has_sleeper(wq))
2166 wake_up_interruptible_all(&wq->wait);
2167 rcu_read_unlock();
2168 }
2169
2170 static void sock_def_error_report(struct sock *sk)
2171 {
2172 struct socket_wq *wq;
2173
2174 rcu_read_lock();
2175 wq = rcu_dereference(sk->sk_wq);
2176 if (wq_has_sleeper(wq))
2177 wake_up_interruptible_poll(&wq->wait, POLLERR);
2178 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2179 rcu_read_unlock();
2180 }
2181
2182 static void sock_def_readable(struct sock *sk, int len)
2183 {
2184 struct socket_wq *wq;
2185
2186 rcu_read_lock();
2187 wq = rcu_dereference(sk->sk_wq);
2188 if (wq_has_sleeper(wq))
2189 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2190 POLLRDNORM | POLLRDBAND);
2191 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2192 rcu_read_unlock();
2193 }
2194
2195 static void sock_def_write_space(struct sock *sk)
2196 {
2197 struct socket_wq *wq;
2198
2199 rcu_read_lock();
2200
2201 /* Do not wake up a writer until he can make "significant"
2202 * progress. --DaveM
2203 */
2204 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2205 wq = rcu_dereference(sk->sk_wq);
2206 if (wq_has_sleeper(wq))
2207 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2208 POLLWRNORM | POLLWRBAND);
2209
2210 /* Should agree with poll, otherwise some programs break */
2211 if (sock_writeable(sk))
2212 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2213 }
2214
2215 rcu_read_unlock();
2216 }
2217
2218 static void sock_def_destruct(struct sock *sk)
2219 {
2220 kfree(sk->sk_protinfo);
2221 }
2222
2223 void sk_send_sigurg(struct sock *sk)
2224 {
2225 if (sk->sk_socket && sk->sk_socket->file)
2226 if (send_sigurg(&sk->sk_socket->file->f_owner))
2227 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2228 }
2229 EXPORT_SYMBOL(sk_send_sigurg);
2230
2231 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2232 unsigned long expires)
2233 {
2234 if (!mod_timer(timer, expires))
2235 sock_hold(sk);
2236 }
2237 EXPORT_SYMBOL(sk_reset_timer);
2238
2239 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2240 {
2241 if (del_timer(timer))
2242 __sock_put(sk);
2243 }
2244 EXPORT_SYMBOL(sk_stop_timer);
2245
2246 void sock_init_data(struct socket *sock, struct sock *sk)
2247 {
2248 skb_queue_head_init(&sk->sk_receive_queue);
2249 skb_queue_head_init(&sk->sk_write_queue);
2250 skb_queue_head_init(&sk->sk_error_queue);
2251 #ifdef CONFIG_NET_DMA
2252 skb_queue_head_init(&sk->sk_async_wait_queue);
2253 #endif
2254
2255 sk->sk_send_head = NULL;
2256
2257 init_timer(&sk->sk_timer);
2258
2259 sk->sk_allocation = GFP_KERNEL;
2260 sk->sk_rcvbuf = sysctl_rmem_default;
2261 sk->sk_sndbuf = sysctl_wmem_default;
2262 sk->sk_state = TCP_CLOSE;
2263 sk_set_socket(sk, sock);
2264
2265 sock_set_flag(sk, SOCK_ZAPPED);
2266
2267 if (sock) {
2268 sk->sk_type = sock->type;
2269 sk->sk_wq = sock->wq;
2270 sock->sk = sk;
2271 } else
2272 sk->sk_wq = NULL;
2273
2274 spin_lock_init(&sk->sk_dst_lock);
2275 rwlock_init(&sk->sk_callback_lock);
2276 lockdep_set_class_and_name(&sk->sk_callback_lock,
2277 af_callback_keys + sk->sk_family,
2278 af_family_clock_key_strings[sk->sk_family]);
2279
2280 sk->sk_state_change = sock_def_wakeup;
2281 sk->sk_data_ready = sock_def_readable;
2282 sk->sk_write_space = sock_def_write_space;
2283 sk->sk_error_report = sock_def_error_report;
2284 sk->sk_destruct = sock_def_destruct;
2285
2286 sk->sk_frag.page = NULL;
2287 sk->sk_frag.offset = 0;
2288 sk->sk_peek_off = -1;
2289
2290 sk->sk_peer_pid = NULL;
2291 sk->sk_peer_cred = NULL;
2292 sk->sk_write_pending = 0;
2293 sk->sk_rcvlowat = 1;
2294 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2295 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2296
2297 sk->sk_stamp = ktime_set(-1L, 0);
2298
2299 /*
2300 * Before updating sk_refcnt, we must commit prior changes to memory
2301 * (Documentation/RCU/rculist_nulls.txt for details)
2302 */
2303 smp_wmb();
2304 atomic_set(&sk->sk_refcnt, 1);
2305 atomic_set(&sk->sk_drops, 0);
2306 }
2307 EXPORT_SYMBOL(sock_init_data);
2308
2309 void lock_sock_nested(struct sock *sk, int subclass)
2310 {
2311 might_sleep();
2312 spin_lock_bh(&sk->sk_lock.slock);
2313 if (sk->sk_lock.owned)
2314 __lock_sock(sk);
2315 sk->sk_lock.owned = 1;
2316 spin_unlock(&sk->sk_lock.slock);
2317 /*
2318 * The sk_lock has mutex_lock() semantics here:
2319 */
2320 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2321 local_bh_enable();
2322 }
2323 EXPORT_SYMBOL(lock_sock_nested);
2324
2325 void release_sock(struct sock *sk)
2326 {
2327 /*
2328 * The sk_lock has mutex_unlock() semantics:
2329 */
2330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2331
2332 spin_lock_bh(&sk->sk_lock.slock);
2333 if (sk->sk_backlog.tail)
2334 __release_sock(sk);
2335
2336 if (sk->sk_prot->release_cb)
2337 sk->sk_prot->release_cb(sk);
2338
2339 sk->sk_lock.owned = 0;
2340 if (waitqueue_active(&sk->sk_lock.wq))
2341 wake_up(&sk->sk_lock.wq);
2342 spin_unlock_bh(&sk->sk_lock.slock);
2343 }
2344 EXPORT_SYMBOL(release_sock);
2345
2346 /**
2347 * lock_sock_fast - fast version of lock_sock
2348 * @sk: socket
2349 *
2350 * This version should be used for very small section, where process wont block
2351 * return false if fast path is taken
2352 * sk_lock.slock locked, owned = 0, BH disabled
2353 * return true if slow path is taken
2354 * sk_lock.slock unlocked, owned = 1, BH enabled
2355 */
2356 bool lock_sock_fast(struct sock *sk)
2357 {
2358 might_sleep();
2359 spin_lock_bh(&sk->sk_lock.slock);
2360
2361 if (!sk->sk_lock.owned)
2362 /*
2363 * Note : We must disable BH
2364 */
2365 return false;
2366
2367 __lock_sock(sk);
2368 sk->sk_lock.owned = 1;
2369 spin_unlock(&sk->sk_lock.slock);
2370 /*
2371 * The sk_lock has mutex_lock() semantics here:
2372 */
2373 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2374 local_bh_enable();
2375 return true;
2376 }
2377 EXPORT_SYMBOL(lock_sock_fast);
2378
2379 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2380 {
2381 struct timeval tv;
2382 if (!sock_flag(sk, SOCK_TIMESTAMP))
2383 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2384 tv = ktime_to_timeval(sk->sk_stamp);
2385 if (tv.tv_sec == -1)
2386 return -ENOENT;
2387 if (tv.tv_sec == 0) {
2388 sk->sk_stamp = ktime_get_real();
2389 tv = ktime_to_timeval(sk->sk_stamp);
2390 }
2391 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2392 }
2393 EXPORT_SYMBOL(sock_get_timestamp);
2394
2395 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2396 {
2397 struct timespec ts;
2398 if (!sock_flag(sk, SOCK_TIMESTAMP))
2399 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2400 ts = ktime_to_timespec(sk->sk_stamp);
2401 if (ts.tv_sec == -1)
2402 return -ENOENT;
2403 if (ts.tv_sec == 0) {
2404 sk->sk_stamp = ktime_get_real();
2405 ts = ktime_to_timespec(sk->sk_stamp);
2406 }
2407 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2408 }
2409 EXPORT_SYMBOL(sock_get_timestampns);
2410
2411 void sock_enable_timestamp(struct sock *sk, int flag)
2412 {
2413 if (!sock_flag(sk, flag)) {
2414 unsigned long previous_flags = sk->sk_flags;
2415
2416 sock_set_flag(sk, flag);
2417 /*
2418 * we just set one of the two flags which require net
2419 * time stamping, but time stamping might have been on
2420 * already because of the other one
2421 */
2422 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2423 net_enable_timestamp();
2424 }
2425 }
2426
2427 /*
2428 * Get a socket option on an socket.
2429 *
2430 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2431 * asynchronous errors should be reported by getsockopt. We assume
2432 * this means if you specify SO_ERROR (otherwise whats the point of it).
2433 */
2434 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2435 char __user *optval, int __user *optlen)
2436 {
2437 struct sock *sk = sock->sk;
2438
2439 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2440 }
2441 EXPORT_SYMBOL(sock_common_getsockopt);
2442
2443 #ifdef CONFIG_COMPAT
2444 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2445 char __user *optval, int __user *optlen)
2446 {
2447 struct sock *sk = sock->sk;
2448
2449 if (sk->sk_prot->compat_getsockopt != NULL)
2450 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2451 optval, optlen);
2452 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2453 }
2454 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2455 #endif
2456
2457 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2458 struct msghdr *msg, size_t size, int flags)
2459 {
2460 struct sock *sk = sock->sk;
2461 int addr_len = 0;
2462 int err;
2463
2464 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2465 flags & ~MSG_DONTWAIT, &addr_len);
2466 if (err >= 0)
2467 msg->msg_namelen = addr_len;
2468 return err;
2469 }
2470 EXPORT_SYMBOL(sock_common_recvmsg);
2471
2472 /*
2473 * Set socket options on an inet socket.
2474 */
2475 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2476 char __user *optval, unsigned int optlen)
2477 {
2478 struct sock *sk = sock->sk;
2479
2480 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2481 }
2482 EXPORT_SYMBOL(sock_common_setsockopt);
2483
2484 #ifdef CONFIG_COMPAT
2485 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2486 char __user *optval, unsigned int optlen)
2487 {
2488 struct sock *sk = sock->sk;
2489
2490 if (sk->sk_prot->compat_setsockopt != NULL)
2491 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2492 optval, optlen);
2493 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2494 }
2495 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2496 #endif
2497
2498 void sk_common_release(struct sock *sk)
2499 {
2500 if (sk->sk_prot->destroy)
2501 sk->sk_prot->destroy(sk);
2502
2503 /*
2504 * Observation: when sock_common_release is called, processes have
2505 * no access to socket. But net still has.
2506 * Step one, detach it from networking:
2507 *
2508 * A. Remove from hash tables.
2509 */
2510
2511 sk->sk_prot->unhash(sk);
2512
2513 /*
2514 * In this point socket cannot receive new packets, but it is possible
2515 * that some packets are in flight because some CPU runs receiver and
2516 * did hash table lookup before we unhashed socket. They will achieve
2517 * receive queue and will be purged by socket destructor.
2518 *
2519 * Also we still have packets pending on receive queue and probably,
2520 * our own packets waiting in device queues. sock_destroy will drain
2521 * receive queue, but transmitted packets will delay socket destruction
2522 * until the last reference will be released.
2523 */
2524
2525 sock_orphan(sk);
2526
2527 xfrm_sk_free_policy(sk);
2528
2529 sk_refcnt_debug_release(sk);
2530
2531 if (sk->sk_frag.page) {
2532 put_page(sk->sk_frag.page);
2533 sk->sk_frag.page = NULL;
2534 }
2535
2536 sock_put(sk);
2537 }
2538 EXPORT_SYMBOL(sk_common_release);
2539
2540 #ifdef CONFIG_PROC_FS
2541 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2542 struct prot_inuse {
2543 int val[PROTO_INUSE_NR];
2544 };
2545
2546 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2547
2548 #ifdef CONFIG_NET_NS
2549 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2550 {
2551 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2552 }
2553 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2554
2555 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2556 {
2557 int cpu, idx = prot->inuse_idx;
2558 int res = 0;
2559
2560 for_each_possible_cpu(cpu)
2561 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2562
2563 return res >= 0 ? res : 0;
2564 }
2565 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2566
2567 static int __net_init sock_inuse_init_net(struct net *net)
2568 {
2569 net->core.inuse = alloc_percpu(struct prot_inuse);
2570 return net->core.inuse ? 0 : -ENOMEM;
2571 }
2572
2573 static void __net_exit sock_inuse_exit_net(struct net *net)
2574 {
2575 free_percpu(net->core.inuse);
2576 }
2577
2578 static struct pernet_operations net_inuse_ops = {
2579 .init = sock_inuse_init_net,
2580 .exit = sock_inuse_exit_net,
2581 };
2582
2583 static __init int net_inuse_init(void)
2584 {
2585 if (register_pernet_subsys(&net_inuse_ops))
2586 panic("Cannot initialize net inuse counters");
2587
2588 return 0;
2589 }
2590
2591 core_initcall(net_inuse_init);
2592 #else
2593 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2594
2595 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2596 {
2597 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2598 }
2599 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2600
2601 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2602 {
2603 int cpu, idx = prot->inuse_idx;
2604 int res = 0;
2605
2606 for_each_possible_cpu(cpu)
2607 res += per_cpu(prot_inuse, cpu).val[idx];
2608
2609 return res >= 0 ? res : 0;
2610 }
2611 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2612 #endif
2613
2614 static void assign_proto_idx(struct proto *prot)
2615 {
2616 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2617
2618 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2619 pr_err("PROTO_INUSE_NR exhausted\n");
2620 return;
2621 }
2622
2623 set_bit(prot->inuse_idx, proto_inuse_idx);
2624 }
2625
2626 static void release_proto_idx(struct proto *prot)
2627 {
2628 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2629 clear_bit(prot->inuse_idx, proto_inuse_idx);
2630 }
2631 #else
2632 static inline void assign_proto_idx(struct proto *prot)
2633 {
2634 }
2635
2636 static inline void release_proto_idx(struct proto *prot)
2637 {
2638 }
2639 #endif
2640
2641 int proto_register(struct proto *prot, int alloc_slab)
2642 {
2643 if (alloc_slab) {
2644 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2645 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2646 NULL);
2647
2648 if (prot->slab == NULL) {
2649 pr_crit("%s: Can't create sock SLAB cache!\n",
2650 prot->name);
2651 goto out;
2652 }
2653
2654 if (prot->rsk_prot != NULL) {
2655 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2656 if (prot->rsk_prot->slab_name == NULL)
2657 goto out_free_sock_slab;
2658
2659 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2660 prot->rsk_prot->obj_size, 0,
2661 SLAB_HWCACHE_ALIGN, NULL);
2662
2663 if (prot->rsk_prot->slab == NULL) {
2664 pr_crit("%s: Can't create request sock SLAB cache!\n",
2665 prot->name);
2666 goto out_free_request_sock_slab_name;
2667 }
2668 }
2669
2670 if (prot->twsk_prot != NULL) {
2671 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2672
2673 if (prot->twsk_prot->twsk_slab_name == NULL)
2674 goto out_free_request_sock_slab;
2675
2676 prot->twsk_prot->twsk_slab =
2677 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2678 prot->twsk_prot->twsk_obj_size,
2679 0,
2680 SLAB_HWCACHE_ALIGN |
2681 prot->slab_flags,
2682 NULL);
2683 if (prot->twsk_prot->twsk_slab == NULL)
2684 goto out_free_timewait_sock_slab_name;
2685 }
2686 }
2687
2688 mutex_lock(&proto_list_mutex);
2689 list_add(&prot->node, &proto_list);
2690 assign_proto_idx(prot);
2691 mutex_unlock(&proto_list_mutex);
2692 return 0;
2693
2694 out_free_timewait_sock_slab_name:
2695 kfree(prot->twsk_prot->twsk_slab_name);
2696 out_free_request_sock_slab:
2697 if (prot->rsk_prot && prot->rsk_prot->slab) {
2698 kmem_cache_destroy(prot->rsk_prot->slab);
2699 prot->rsk_prot->slab = NULL;
2700 }
2701 out_free_request_sock_slab_name:
2702 if (prot->rsk_prot)
2703 kfree(prot->rsk_prot->slab_name);
2704 out_free_sock_slab:
2705 kmem_cache_destroy(prot->slab);
2706 prot->slab = NULL;
2707 out:
2708 return -ENOBUFS;
2709 }
2710 EXPORT_SYMBOL(proto_register);
2711
2712 void proto_unregister(struct proto *prot)
2713 {
2714 mutex_lock(&proto_list_mutex);
2715 release_proto_idx(prot);
2716 list_del(&prot->node);
2717 mutex_unlock(&proto_list_mutex);
2718
2719 if (prot->slab != NULL) {
2720 kmem_cache_destroy(prot->slab);
2721 prot->slab = NULL;
2722 }
2723
2724 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2725 kmem_cache_destroy(prot->rsk_prot->slab);
2726 kfree(prot->rsk_prot->slab_name);
2727 prot->rsk_prot->slab = NULL;
2728 }
2729
2730 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2731 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2732 kfree(prot->twsk_prot->twsk_slab_name);
2733 prot->twsk_prot->twsk_slab = NULL;
2734 }
2735 }
2736 EXPORT_SYMBOL(proto_unregister);
2737
2738 #ifdef CONFIG_PROC_FS
2739 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2740 __acquires(proto_list_mutex)
2741 {
2742 mutex_lock(&proto_list_mutex);
2743 return seq_list_start_head(&proto_list, *pos);
2744 }
2745
2746 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2747 {
2748 return seq_list_next(v, &proto_list, pos);
2749 }
2750
2751 static void proto_seq_stop(struct seq_file *seq, void *v)
2752 __releases(proto_list_mutex)
2753 {
2754 mutex_unlock(&proto_list_mutex);
2755 }
2756
2757 static char proto_method_implemented(const void *method)
2758 {
2759 return method == NULL ? 'n' : 'y';
2760 }
2761 static long sock_prot_memory_allocated(struct proto *proto)
2762 {
2763 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2764 }
2765
2766 static char *sock_prot_memory_pressure(struct proto *proto)
2767 {
2768 return proto->memory_pressure != NULL ?
2769 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2770 }
2771
2772 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2773 {
2774
2775 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2776 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2777 proto->name,
2778 proto->obj_size,
2779 sock_prot_inuse_get(seq_file_net(seq), proto),
2780 sock_prot_memory_allocated(proto),
2781 sock_prot_memory_pressure(proto),
2782 proto->max_header,
2783 proto->slab == NULL ? "no" : "yes",
2784 module_name(proto->owner),
2785 proto_method_implemented(proto->close),
2786 proto_method_implemented(proto->connect),
2787 proto_method_implemented(proto->disconnect),
2788 proto_method_implemented(proto->accept),
2789 proto_method_implemented(proto->ioctl),
2790 proto_method_implemented(proto->init),
2791 proto_method_implemented(proto->destroy),
2792 proto_method_implemented(proto->shutdown),
2793 proto_method_implemented(proto->setsockopt),
2794 proto_method_implemented(proto->getsockopt),
2795 proto_method_implemented(proto->sendmsg),
2796 proto_method_implemented(proto->recvmsg),
2797 proto_method_implemented(proto->sendpage),
2798 proto_method_implemented(proto->bind),
2799 proto_method_implemented(proto->backlog_rcv),
2800 proto_method_implemented(proto->hash),
2801 proto_method_implemented(proto->unhash),
2802 proto_method_implemented(proto->get_port),
2803 proto_method_implemented(proto->enter_memory_pressure));
2804 }
2805
2806 static int proto_seq_show(struct seq_file *seq, void *v)
2807 {
2808 if (v == &proto_list)
2809 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2810 "protocol",
2811 "size",
2812 "sockets",
2813 "memory",
2814 "press",
2815 "maxhdr",
2816 "slab",
2817 "module",
2818 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2819 else
2820 proto_seq_printf(seq, list_entry(v, struct proto, node));
2821 return 0;
2822 }
2823
2824 static const struct seq_operations proto_seq_ops = {
2825 .start = proto_seq_start,
2826 .next = proto_seq_next,
2827 .stop = proto_seq_stop,
2828 .show = proto_seq_show,
2829 };
2830
2831 static int proto_seq_open(struct inode *inode, struct file *file)
2832 {
2833 return seq_open_net(inode, file, &proto_seq_ops,
2834 sizeof(struct seq_net_private));
2835 }
2836
2837 static const struct file_operations proto_seq_fops = {
2838 .owner = THIS_MODULE,
2839 .open = proto_seq_open,
2840 .read = seq_read,
2841 .llseek = seq_lseek,
2842 .release = seq_release_net,
2843 };
2844
2845 static __net_init int proto_init_net(struct net *net)
2846 {
2847 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2848 return -ENOMEM;
2849
2850 return 0;
2851 }
2852
2853 static __net_exit void proto_exit_net(struct net *net)
2854 {
2855 remove_proc_entry("protocols", net->proc_net);
2856 }
2857
2858
2859 static __net_initdata struct pernet_operations proto_net_ops = {
2860 .init = proto_init_net,
2861 .exit = proto_exit_net,
2862 };
2863
2864 static int __init proto_init(void)
2865 {
2866 return register_pernet_subsys(&proto_net_ops);
2867 }
2868
2869 subsys_initcall(proto_init);
2870
2871 #endif /* PROC_FS */
Linus' kernel tree