projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jesse/openvswitch
[deliverable/linux.git] / net / core / sock.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
101 #include <linux/module.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <linux/sched.h>
105 #include <linux/timer.h>
106 #include <linux/string.h>
107 #include <linux/sockios.h>
108 #include <linux/net.h>
109 #include <linux/mm.h>
110 #include <linux/slab.h>
111 #include <linux/interrupt.h>
112 #include <linux/poll.h>
113 #include <linux/tcp.h>
114 #include <linux/init.h>
115 #include <linux/highmem.h>
116 #include <linux/user_namespace.h>
117 #include <linux/static_key.h>
118 #include <linux/memcontrol.h>
119 #include <linux/prefetch.h>
120
121 #include <asm/uaccess.h>
122
123 #include <linux/netdevice.h>
124 #include <net/protocol.h>
125 #include <linux/skbuff.h>
126 #include <net/net_namespace.h>
127 #include <net/request_sock.h>
128 #include <net/sock.h>
129 #include <linux/net_tstamp.h>
130 #include <net/xfrm.h>
131 #include <linux/ipsec.h>
132 #include <net/cls_cgroup.h>
133 #include <net/netprio_cgroup.h>
134
135 #include <linux/filter.h>
136
137 #include <trace/events/sock.h>
138
139 #ifdef CONFIG_INET
140 #include <net/tcp.h>
141 #endif
142
143 #include <net/busy_poll.h>
144
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
147
148 #ifdef CONFIG_MEMCG_KMEM
149 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
150 {
151 struct proto *proto;
152 int ret = 0;
153
154 mutex_lock(&proto_list_mutex);
155 list_for_each_entry(proto, &proto_list, node) {
156 if (proto->init_cgroup) {
157 ret = proto->init_cgroup(memcg, ss);
158 if (ret)
159 goto out;
160 }
161 }
162
163 mutex_unlock(&proto_list_mutex);
164 return ret;
165 out:
166 list_for_each_entry_continue_reverse(proto, &proto_list, node)
167 if (proto->destroy_cgroup)
168 proto->destroy_cgroup(memcg);
169 mutex_unlock(&proto_list_mutex);
170 return ret;
171 }
172
173 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
174 {
175 struct proto *proto;
176
177 mutex_lock(&proto_list_mutex);
178 list_for_each_entry_reverse(proto, &proto_list, node)
179 if (proto->destroy_cgroup)
180 proto->destroy_cgroup(memcg);
181 mutex_unlock(&proto_list_mutex);
182 }
183 #endif
184
185 /*
186 * Each address family might have different locking rules, so we have
187 * one slock key per address family:
188 */
189 static struct lock_class_key af_family_keys[AF_MAX];
190 static struct lock_class_key af_family_slock_keys[AF_MAX];
191
192 #if defined(CONFIG_MEMCG_KMEM)
193 struct static_key memcg_socket_limit_enabled;
194 EXPORT_SYMBOL(memcg_socket_limit_enabled);
195 #endif
196
197 /*
198 * Make lock validator output more readable. (we pre-construct these
199 * strings build-time, so that runtime initialization of socket
200 * locks is fast):
201 */
202 static const char *const af_family_key_strings[AF_MAX+1] = {
203 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
204 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
205 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
206 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
207 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
208 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
209 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
210 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
211 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
212 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
213 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
214 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
215 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
216 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX"
217 };
218 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
219 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
220 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
221 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
222 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
223 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
224 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
225 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
226 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
227 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
228 "slock-27" , "slock-28" , "slock-AF_CAN" ,
229 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
230 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
231 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
232 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX"
233 };
234 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
235 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
236 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
237 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
238 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
239 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
240 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
241 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
242 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
243 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
244 "clock-27" , "clock-28" , "clock-AF_CAN" ,
245 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
246 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
247 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
248 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX"
249 };
250
251 /*
252 * sk_callback_lock locking rules are per-address-family,
253 * so split the lock classes by using a per-AF key:
254 */
255 static struct lock_class_key af_callback_keys[AF_MAX];
256
257 /* Take into consideration the size of the struct sk_buff overhead in the
258 * determination of these values, since that is non-constant across
259 * platforms. This makes socket queueing behavior and performance
260 * not depend upon such differences.
261 */
262 #define _SK_MEM_PACKETS 256
263 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
264 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
265 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
266
267 /* Run time adjustable parameters. */
268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
269 EXPORT_SYMBOL(sysctl_wmem_max);
270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
271 EXPORT_SYMBOL(sysctl_rmem_max);
272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
274
275 /* Maximal space eaten by iovec or ancillary data plus some space */
276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
277 EXPORT_SYMBOL(sysctl_optmem_max);
278
279 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
280 EXPORT_SYMBOL_GPL(memalloc_socks);
281
282 /**
283 * sk_set_memalloc - sets %SOCK_MEMALLOC
284 * @sk: socket to set it on
285 *
286 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
287 * It's the responsibility of the admin to adjust min_free_kbytes
288 * to meet the requirements
289 */
290 void sk_set_memalloc(struct sock *sk)
291 {
292 sock_set_flag(sk, SOCK_MEMALLOC);
293 sk->sk_allocation |= __GFP_MEMALLOC;
294 static_key_slow_inc(&memalloc_socks);
295 }
296 EXPORT_SYMBOL_GPL(sk_set_memalloc);
297
298 void sk_clear_memalloc(struct sock *sk)
299 {
300 sock_reset_flag(sk, SOCK_MEMALLOC);
301 sk->sk_allocation &= ~__GFP_MEMALLOC;
302 static_key_slow_dec(&memalloc_socks);
303
304 /*
305 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
306 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
307 * it has rmem allocations there is a risk that the user of the
308 * socket cannot make forward progress due to exceeding the rmem
309 * limits. By rights, sk_clear_memalloc() should only be called
310 * on sockets being torn down but warn and reset the accounting if
311 * that assumption breaks.
312 */
313 if (WARN_ON(sk->sk_forward_alloc))
314 sk_mem_reclaim(sk);
315 }
316 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
317
318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
319 {
320 int ret;
321 unsigned long pflags = current->flags;
322
323 /* these should have been dropped before queueing */
324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
325
326 current->flags |= PF_MEMALLOC;
327 ret = sk->sk_backlog_rcv(sk, skb);
328 tsk_restore_flags(current, pflags, PF_MEMALLOC);
329
330 return ret;
331 }
332 EXPORT_SYMBOL(__sk_backlog_rcv);
333
334 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
335 {
336 struct timeval tv;
337
338 if (optlen < sizeof(tv))
339 return -EINVAL;
340 if (copy_from_user(&tv, optval, sizeof(tv)))
341 return -EFAULT;
342 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
343 return -EDOM;
344
345 if (tv.tv_sec < 0) {
346 static int warned __read_mostly;
347
348 *timeo_p = 0;
349 if (warned < 10 && net_ratelimit()) {
350 warned++;
351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
352 __func__, current->comm, task_pid_nr(current));
353 }
354 return 0;
355 }
356 *timeo_p = MAX_SCHEDULE_TIMEOUT;
357 if (tv.tv_sec == 0 && tv.tv_usec == 0)
358 return 0;
359 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
360 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
361 return 0;
362 }
363
364 static void sock_warn_obsolete_bsdism(const char *name)
365 {
366 static int warned;
367 static char warncomm[TASK_COMM_LEN];
368 if (strcmp(warncomm, current->comm) && warned < 5) {
369 strcpy(warncomm, current->comm);
370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
371 warncomm, name);
372 warned++;
373 }
374 }
375
376 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
377
378 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
379 {
380 if (sk->sk_flags & flags) {
381 sk->sk_flags &= ~flags;
382 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
383 net_disable_timestamp();
384 }
385 }
386
387
388 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
389 {
390 int err;
391 int skb_len;
392 unsigned long flags;
393 struct sk_buff_head *list = &sk->sk_receive_queue;
394
395 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
396 atomic_inc(&sk->sk_drops);
397 trace_sock_rcvqueue_full(sk, skb);
398 return -ENOMEM;
399 }
400
401 err = sk_filter(sk, skb);
402 if (err)
403 return err;
404
405 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
406 atomic_inc(&sk->sk_drops);
407 return -ENOBUFS;
408 }
409
410 skb->dev = NULL;
411 skb_set_owner_r(skb, sk);
412
413 /* Cache the SKB length before we tack it onto the receive
414 * queue. Once it is added it no longer belongs to us and
415 * may be freed by other threads of control pulling packets
416 * from the queue.
417 */
418 skb_len = skb->len;
419
420 /* we escape from rcu protected region, make sure we dont leak
421 * a norefcounted dst
422 */
423 skb_dst_force(skb);
424
425 spin_lock_irqsave(&list->lock, flags);
426 skb->dropcount = atomic_read(&sk->sk_drops);
427 __skb_queue_tail(list, skb);
428 spin_unlock_irqrestore(&list->lock, flags);
429
430 if (!sock_flag(sk, SOCK_DEAD))
431 sk->sk_data_ready(sk, skb_len);
432 return 0;
433 }
434 EXPORT_SYMBOL(sock_queue_rcv_skb);
435
436 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
437 {
438 int rc = NET_RX_SUCCESS;
439
440 if (sk_filter(sk, skb))
441 goto discard_and_relse;
442
443 skb->dev = NULL;
444
445 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
446 atomic_inc(&sk->sk_drops);
447 goto discard_and_relse;
448 }
449 if (nested)
450 bh_lock_sock_nested(sk);
451 else
452 bh_lock_sock(sk);
453 if (!sock_owned_by_user(sk)) {
454 /*
455 * trylock + unlock semantics:
456 */
457 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
458
459 rc = sk_backlog_rcv(sk, skb);
460
461 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
462 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
463 bh_unlock_sock(sk);
464 atomic_inc(&sk->sk_drops);
465 goto discard_and_relse;
466 }
467
468 bh_unlock_sock(sk);
469 out:
470 sock_put(sk);
471 return rc;
472 discard_and_relse:
473 kfree_skb(skb);
474 goto out;
475 }
476 EXPORT_SYMBOL(sk_receive_skb);
477
478 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
479 {
480 struct dst_entry *dst = __sk_dst_get(sk);
481
482 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
483 sk_tx_queue_clear(sk);
484 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
485 dst_release(dst);
486 return NULL;
487 }
488
489 return dst;
490 }
491 EXPORT_SYMBOL(__sk_dst_check);
492
493 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
494 {
495 struct dst_entry *dst = sk_dst_get(sk);
496
497 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
498 sk_dst_reset(sk);
499 dst_release(dst);
500 return NULL;
501 }
502
503 return dst;
504 }
505 EXPORT_SYMBOL(sk_dst_check);
506
507 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
508 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 (!ns_capable(net->user_ns, 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 int sock_getbindtodevice(struct sock *sk, char __user *optval,
566 int __user *optlen, int len)
567 {
568 int ret = -ENOPROTOOPT;
569 #ifdef CONFIG_NETDEVICES
570 struct net *net = sock_net(sk);
571 char devname[IFNAMSIZ];
572
573 if (sk->sk_bound_dev_if == 0) {
574 len = 0;
575 goto zero;
576 }
577
578 ret = -EINVAL;
579 if (len < IFNAMSIZ)
580 goto out;
581
582 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
583 if (ret)
584 goto out;
585
586 len = strlen(devname) + 1;
587
588 ret = -EFAULT;
589 if (copy_to_user(optval, devname, len))
590 goto out;
591
592 zero:
593 ret = -EFAULT;
594 if (put_user(len, optlen))
595 goto out;
596
597 ret = 0;
598
599 out:
600 #endif
601
602 return ret;
603 }
604
605 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
606 {
607 if (valbool)
608 sock_set_flag(sk, bit);
609 else
610 sock_reset_flag(sk, bit);
611 }
612
613 /*
614 * This is meant for all protocols to use and covers goings on
615 * at the socket level. Everything here is generic.
616 */
617
618 int sock_setsockopt(struct socket *sock, int level, int optname,
619 char __user *optval, unsigned int optlen)
620 {
621 struct sock *sk = sock->sk;
622 int val;
623 int valbool;
624 struct linger ling;
625 int ret = 0;
626
627 /*
628 * Options without arguments
629 */
630
631 if (optname == SO_BINDTODEVICE)
632 return sock_setbindtodevice(sk, optval, optlen);
633
634 if (optlen < sizeof(int))
635 return -EINVAL;
636
637 if (get_user(val, (int __user *)optval))
638 return -EFAULT;
639
640 valbool = val ? 1 : 0;
641
642 lock_sock(sk);
643
644 switch (optname) {
645 case SO_DEBUG:
646 if (val && !capable(CAP_NET_ADMIN))
647 ret = -EACCES;
648 else
649 sock_valbool_flag(sk, SOCK_DBG, valbool);
650 break;
651 case SO_REUSEADDR:
652 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
653 break;
654 case SO_REUSEPORT:
655 sk->sk_reuseport = valbool;
656 break;
657 case SO_TYPE:
658 case SO_PROTOCOL:
659 case SO_DOMAIN:
660 case SO_ERROR:
661 ret = -ENOPROTOOPT;
662 break;
663 case SO_DONTROUTE:
664 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
665 break;
666 case SO_BROADCAST:
667 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
668 break;
669 case SO_SNDBUF:
670 /* Don't error on this BSD doesn't and if you think
671 * about it this is right. Otherwise apps have to
672 * play 'guess the biggest size' games. RCVBUF/SNDBUF
673 * are treated in BSD as hints
674 */
675 val = min_t(u32, val, sysctl_wmem_max);
676 set_sndbuf:
677 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
678 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
679 /* Wake up sending tasks if we upped the value. */
680 sk->sk_write_space(sk);
681 break;
682
683 case SO_SNDBUFFORCE:
684 if (!capable(CAP_NET_ADMIN)) {
685 ret = -EPERM;
686 break;
687 }
688 goto set_sndbuf;
689
690 case SO_RCVBUF:
691 /* Don't error on this BSD doesn't and if you think
692 * about it this is right. Otherwise apps have to
693 * play 'guess the biggest size' games. RCVBUF/SNDBUF
694 * are treated in BSD as hints
695 */
696 val = min_t(u32, val, sysctl_rmem_max);
697 set_rcvbuf:
698 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
699 /*
700 * We double it on the way in to account for
701 * "struct sk_buff" etc. overhead. Applications
702 * assume that the SO_RCVBUF setting they make will
703 * allow that much actual data to be received on that
704 * socket.
705 *
706 * Applications are unaware that "struct sk_buff" and
707 * other overheads allocate from the receive buffer
708 * during socket buffer allocation.
709 *
710 * And after considering the possible alternatives,
711 * returning the value we actually used in getsockopt
712 * is the most desirable behavior.
713 */
714 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
715 break;
716
717 case SO_RCVBUFFORCE:
718 if (!capable(CAP_NET_ADMIN)) {
719 ret = -EPERM;
720 break;
721 }
722 goto set_rcvbuf;
723
724 case SO_KEEPALIVE:
725 #ifdef CONFIG_INET
726 if (sk->sk_protocol == IPPROTO_TCP &&
727 sk->sk_type == SOCK_STREAM)
728 tcp_set_keepalive(sk, valbool);
729 #endif
730 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
731 break;
732
733 case SO_OOBINLINE:
734 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
735 break;
736
737 case SO_NO_CHECK:
738 sk->sk_no_check = valbool;
739 break;
740
741 case SO_PRIORITY:
742 if ((val >= 0 && val <= 6) ||
743 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
744 sk->sk_priority = val;
745 else
746 ret = -EPERM;
747 break;
748
749 case SO_LINGER:
750 if (optlen < sizeof(ling)) {
751 ret = -EINVAL; /* 1003.1g */
752 break;
753 }
754 if (copy_from_user(&ling, optval, sizeof(ling))) {
755 ret = -EFAULT;
756 break;
757 }
758 if (!ling.l_onoff)
759 sock_reset_flag(sk, SOCK_LINGER);
760 else {
761 #if (BITS_PER_LONG == 32)
762 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
763 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
764 else
765 #endif
766 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
767 sock_set_flag(sk, SOCK_LINGER);
768 }
769 break;
770
771 case SO_BSDCOMPAT:
772 sock_warn_obsolete_bsdism("setsockopt");
773 break;
774
775 case SO_PASSCRED:
776 if (valbool)
777 set_bit(SOCK_PASSCRED, &sock->flags);
778 else
779 clear_bit(SOCK_PASSCRED, &sock->flags);
780 break;
781
782 case SO_TIMESTAMP:
783 case SO_TIMESTAMPNS:
784 if (valbool) {
785 if (optname == SO_TIMESTAMP)
786 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
787 else
788 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
789 sock_set_flag(sk, SOCK_RCVTSTAMP);
790 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
791 } else {
792 sock_reset_flag(sk, SOCK_RCVTSTAMP);
793 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
794 }
795 break;
796
797 case SO_TIMESTAMPING:
798 if (val & ~SOF_TIMESTAMPING_MASK) {
799 ret = -EINVAL;
800 break;
801 }
802 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
803 val & SOF_TIMESTAMPING_TX_HARDWARE);
804 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
805 val & SOF_TIMESTAMPING_TX_SOFTWARE);
806 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
807 val & SOF_TIMESTAMPING_RX_HARDWARE);
808 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
809 sock_enable_timestamp(sk,
810 SOCK_TIMESTAMPING_RX_SOFTWARE);
811 else
812 sock_disable_timestamp(sk,
813 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
814 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
815 val & SOF_TIMESTAMPING_SOFTWARE);
816 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
817 val & SOF_TIMESTAMPING_SYS_HARDWARE);
818 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
819 val & SOF_TIMESTAMPING_RAW_HARDWARE);
820 break;
821
822 case SO_RCVLOWAT:
823 if (val < 0)
824 val = INT_MAX;
825 sk->sk_rcvlowat = val ? : 1;
826 break;
827
828 case SO_RCVTIMEO:
829 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
830 break;
831
832 case SO_SNDTIMEO:
833 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
834 break;
835
836 case SO_ATTACH_FILTER:
837 ret = -EINVAL;
838 if (optlen == sizeof(struct sock_fprog)) {
839 struct sock_fprog fprog;
840
841 ret = -EFAULT;
842 if (copy_from_user(&fprog, optval, sizeof(fprog)))
843 break;
844
845 ret = sk_attach_filter(&fprog, sk);
846 }
847 break;
848
849 case SO_DETACH_FILTER:
850 ret = sk_detach_filter(sk);
851 break;
852
853 case SO_LOCK_FILTER:
854 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
855 ret = -EPERM;
856 else
857 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
858 break;
859
860 case SO_PASSSEC:
861 if (valbool)
862 set_bit(SOCK_PASSSEC, &sock->flags);
863 else
864 clear_bit(SOCK_PASSSEC, &sock->flags);
865 break;
866 case SO_MARK:
867 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
868 ret = -EPERM;
869 else
870 sk->sk_mark = val;
871 break;
872
873 /* We implement the SO_SNDLOWAT etc to
874 not be settable (1003.1g 5.3) */
875 case SO_RXQ_OVFL:
876 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
877 break;
878
879 case SO_WIFI_STATUS:
880 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
881 break;
882
883 case SO_PEEK_OFF:
884 if (sock->ops->set_peek_off)
885 ret = sock->ops->set_peek_off(sk, val);
886 else
887 ret = -EOPNOTSUPP;
888 break;
889
890 case SO_NOFCS:
891 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
892 break;
893
894 case SO_SELECT_ERR_QUEUE:
895 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
896 break;
897
898 #ifdef CONFIG_NET_RX_BUSY_POLL
899 case SO_BUSY_POLL:
900 /* allow unprivileged users to decrease the value */
901 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
902 ret = -EPERM;
903 else {
904 if (val < 0)
905 ret = -EINVAL;
906 else
907 sk->sk_ll_usec = val;
908 }
909 break;
910 #endif
911
912 case SO_MAX_PACING_RATE:
913 sk->sk_max_pacing_rate = val;
914 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
915 sk->sk_max_pacing_rate);
916 break;
917
918 default:
919 ret = -ENOPROTOOPT;
920 break;
921 }
922 release_sock(sk);
923 return ret;
924 }
925 EXPORT_SYMBOL(sock_setsockopt);
926
927
928 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
929 struct ucred *ucred)
930 {
931 ucred->pid = pid_vnr(pid);
932 ucred->uid = ucred->gid = -1;
933 if (cred) {
934 struct user_namespace *current_ns = current_user_ns();
935
936 ucred->uid = from_kuid_munged(current_ns, cred->euid);
937 ucred->gid = from_kgid_munged(current_ns, cred->egid);
938 }
939 }
940
941 int sock_getsockopt(struct socket *sock, int level, int optname,
942 char __user *optval, int __user *optlen)
943 {
944 struct sock *sk = sock->sk;
945
946 union {
947 int val;
948 struct linger ling;
949 struct timeval tm;
950 } v;
951
952 int lv = sizeof(int);
953 int len;
954
955 if (get_user(len, optlen))
956 return -EFAULT;
957 if (len < 0)
958 return -EINVAL;
959
960 memset(&v, 0, sizeof(v));
961
962 switch (optname) {
963 case SO_DEBUG:
964 v.val = sock_flag(sk, SOCK_DBG);
965 break;
966
967 case SO_DONTROUTE:
968 v.val = sock_flag(sk, SOCK_LOCALROUTE);
969 break;
970
971 case SO_BROADCAST:
972 v.val = sock_flag(sk, SOCK_BROADCAST);
973 break;
974
975 case SO_SNDBUF:
976 v.val = sk->sk_sndbuf;
977 break;
978
979 case SO_RCVBUF:
980 v.val = sk->sk_rcvbuf;
981 break;
982
983 case SO_REUSEADDR:
984 v.val = sk->sk_reuse;
985 break;
986
987 case SO_REUSEPORT:
988 v.val = sk->sk_reuseport;
989 break;
990
991 case SO_KEEPALIVE:
992 v.val = sock_flag(sk, SOCK_KEEPOPEN);
993 break;
994
995 case SO_TYPE:
996 v.val = sk->sk_type;
997 break;
998
999 case SO_PROTOCOL:
1000 v.val = sk->sk_protocol;
1001 break;
1002
1003 case SO_DOMAIN:
1004 v.val = sk->sk_family;
1005 break;
1006
1007 case SO_ERROR:
1008 v.val = -sock_error(sk);
1009 if (v.val == 0)
1010 v.val = xchg(&sk->sk_err_soft, 0);
1011 break;
1012
1013 case SO_OOBINLINE:
1014 v.val = sock_flag(sk, SOCK_URGINLINE);
1015 break;
1016
1017 case SO_NO_CHECK:
1018 v.val = sk->sk_no_check;
1019 break;
1020
1021 case SO_PRIORITY:
1022 v.val = sk->sk_priority;
1023 break;
1024
1025 case SO_LINGER:
1026 lv = sizeof(v.ling);
1027 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
1028 v.ling.l_linger = sk->sk_lingertime / HZ;
1029 break;
1030
1031 case SO_BSDCOMPAT:
1032 sock_warn_obsolete_bsdism("getsockopt");
1033 break;
1034
1035 case SO_TIMESTAMP:
1036 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1037 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1038 break;
1039
1040 case SO_TIMESTAMPNS:
1041 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1042 break;
1043
1044 case SO_TIMESTAMPING:
1045 v.val = 0;
1046 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1047 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1048 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1049 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1050 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1051 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1052 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1053 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1054 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1055 v.val |= SOF_TIMESTAMPING_SOFTWARE;
1056 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1057 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1058 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1059 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1060 break;
1061
1062 case SO_RCVTIMEO:
1063 lv = sizeof(struct timeval);
1064 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1065 v.tm.tv_sec = 0;
1066 v.tm.tv_usec = 0;
1067 } else {
1068 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1069 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1070 }
1071 break;
1072
1073 case SO_SNDTIMEO:
1074 lv = sizeof(struct timeval);
1075 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1076 v.tm.tv_sec = 0;
1077 v.tm.tv_usec = 0;
1078 } else {
1079 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1080 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1081 }
1082 break;
1083
1084 case SO_RCVLOWAT:
1085 v.val = sk->sk_rcvlowat;
1086 break;
1087
1088 case SO_SNDLOWAT:
1089 v.val = 1;
1090 break;
1091
1092 case SO_PASSCRED:
1093 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1094 break;
1095
1096 case SO_PEERCRED:
1097 {
1098 struct ucred peercred;
1099 if (len > sizeof(peercred))
1100 len = sizeof(peercred);
1101 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1102 if (copy_to_user(optval, &peercred, len))
1103 return -EFAULT;
1104 goto lenout;
1105 }
1106
1107 case SO_PEERNAME:
1108 {
1109 char address[128];
1110
1111 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1112 return -ENOTCONN;
1113 if (lv < len)
1114 return -EINVAL;
1115 if (copy_to_user(optval, address, len))
1116 return -EFAULT;
1117 goto lenout;
1118 }
1119
1120 /* Dubious BSD thing... Probably nobody even uses it, but
1121 * the UNIX standard wants it for whatever reason... -DaveM
1122 */
1123 case SO_ACCEPTCONN:
1124 v.val = sk->sk_state == TCP_LISTEN;
1125 break;
1126
1127 case SO_PASSSEC:
1128 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1129 break;
1130
1131 case SO_PEERSEC:
1132 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1133
1134 case SO_MARK:
1135 v.val = sk->sk_mark;
1136 break;
1137
1138 case SO_RXQ_OVFL:
1139 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1140 break;
1141
1142 case SO_WIFI_STATUS:
1143 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1144 break;
1145
1146 case SO_PEEK_OFF:
1147 if (!sock->ops->set_peek_off)
1148 return -EOPNOTSUPP;
1149
1150 v.val = sk->sk_peek_off;
1151 break;
1152 case SO_NOFCS:
1153 v.val = sock_flag(sk, SOCK_NOFCS);
1154 break;
1155
1156 case SO_BINDTODEVICE:
1157 return sock_getbindtodevice(sk, optval, optlen, len);
1158
1159 case SO_GET_FILTER:
1160 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1161 if (len < 0)
1162 return len;
1163
1164 goto lenout;
1165
1166 case SO_LOCK_FILTER:
1167 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1168 break;
1169
1170 case SO_SELECT_ERR_QUEUE:
1171 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1172 break;
1173
1174 #ifdef CONFIG_NET_RX_BUSY_POLL
1175 case SO_BUSY_POLL:
1176 v.val = sk->sk_ll_usec;
1177 break;
1178 #endif
1179
1180 case SO_MAX_PACING_RATE:
1181 v.val = sk->sk_max_pacing_rate;
1182 break;
1183
1184 default:
1185 return -ENOPROTOOPT;
1186 }
1187
1188 if (len > lv)
1189 len = lv;
1190 if (copy_to_user(optval, &v, len))
1191 return -EFAULT;
1192 lenout:
1193 if (put_user(len, optlen))
1194 return -EFAULT;
1195 return 0;
1196 }
1197
1198 /*
1199 * Initialize an sk_lock.
1200 *
1201 * (We also register the sk_lock with the lock validator.)
1202 */
1203 static inline void sock_lock_init(struct sock *sk)
1204 {
1205 sock_lock_init_class_and_name(sk,
1206 af_family_slock_key_strings[sk->sk_family],
1207 af_family_slock_keys + sk->sk_family,
1208 af_family_key_strings[sk->sk_family],
1209 af_family_keys + sk->sk_family);
1210 }
1211
1212 /*
1213 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1214 * even temporarly, because of RCU lookups. sk_node should also be left as is.
1215 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1216 */
1217 static void sock_copy(struct sock *nsk, const struct sock *osk)
1218 {
1219 #ifdef CONFIG_SECURITY_NETWORK
1220 void *sptr = nsk->sk_security;
1221 #endif
1222 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1223
1224 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1225 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1226
1227 #ifdef CONFIG_SECURITY_NETWORK
1228 nsk->sk_security = sptr;
1229 security_sk_clone(osk, nsk);
1230 #endif
1231 }
1232
1233 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1234 {
1235 unsigned long nulls1, nulls2;
1236
1237 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1238 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1239 if (nulls1 > nulls2)
1240 swap(nulls1, nulls2);
1241
1242 if (nulls1 != 0)
1243 memset((char *)sk, 0, nulls1);
1244 memset((char *)sk + nulls1 + sizeof(void *), 0,
1245 nulls2 - nulls1 - sizeof(void *));
1246 memset((char *)sk + nulls2 + sizeof(void *), 0,
1247 size - nulls2 - sizeof(void *));
1248 }
1249 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1250
1251 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1252 int family)
1253 {
1254 struct sock *sk;
1255 struct kmem_cache *slab;
1256
1257 slab = prot->slab;
1258 if (slab != NULL) {
1259 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1260 if (!sk)
1261 return sk;
1262 if (priority & __GFP_ZERO) {
1263 if (prot->clear_sk)
1264 prot->clear_sk(sk, prot->obj_size);
1265 else
1266 sk_prot_clear_nulls(sk, prot->obj_size);
1267 }
1268 } else
1269 sk = kmalloc(prot->obj_size, priority);
1270
1271 if (sk != NULL) {
1272 kmemcheck_annotate_bitfield(sk, flags);
1273
1274 if (security_sk_alloc(sk, family, priority))
1275 goto out_free;
1276
1277 if (!try_module_get(prot->owner))
1278 goto out_free_sec;
1279 sk_tx_queue_clear(sk);
1280 }
1281
1282 return sk;
1283
1284 out_free_sec:
1285 security_sk_free(sk);
1286 out_free:
1287 if (slab != NULL)
1288 kmem_cache_free(slab, sk);
1289 else
1290 kfree(sk);
1291 return NULL;
1292 }
1293
1294 static void sk_prot_free(struct proto *prot, struct sock *sk)
1295 {
1296 struct kmem_cache *slab;
1297 struct module *owner;
1298
1299 owner = prot->owner;
1300 slab = prot->slab;
1301
1302 security_sk_free(sk);
1303 if (slab != NULL)
1304 kmem_cache_free(slab, sk);
1305 else
1306 kfree(sk);
1307 module_put(owner);
1308 }
1309
1310 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
1311 void sock_update_netprioidx(struct sock *sk)
1312 {
1313 if (in_interrupt())
1314 return;
1315
1316 sk->sk_cgrp_prioidx = task_netprioidx(current);
1317 }
1318 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1319 #endif
1320
1321 /**
1322 * sk_alloc - All socket objects are allocated here
1323 * @net: the applicable net namespace
1324 * @family: protocol family
1325 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1326 * @prot: struct proto associated with this new sock instance
1327 */
1328 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1329 struct proto *prot)
1330 {
1331 struct sock *sk;
1332
1333 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1334 if (sk) {
1335 sk->sk_family = family;
1336 /*
1337 * See comment in struct sock definition to understand
1338 * why we need sk_prot_creator -acme
1339 */
1340 sk->sk_prot = sk->sk_prot_creator = prot;
1341 sock_lock_init(sk);
1342 sock_net_set(sk, get_net(net));
1343 atomic_set(&sk->sk_wmem_alloc, 1);
1344
1345 sock_update_classid(sk);
1346 sock_update_netprioidx(sk);
1347 }
1348
1349 return sk;
1350 }
1351 EXPORT_SYMBOL(sk_alloc);
1352
1353 static void __sk_free(struct sock *sk)
1354 {
1355 struct sk_filter *filter;
1356
1357 if (sk->sk_destruct)
1358 sk->sk_destruct(sk);
1359
1360 filter = rcu_dereference_check(sk->sk_filter,
1361 atomic_read(&sk->sk_wmem_alloc) == 0);
1362 if (filter) {
1363 sk_filter_uncharge(sk, filter);
1364 RCU_INIT_POINTER(sk->sk_filter, NULL);
1365 }
1366
1367 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1368
1369 if (atomic_read(&sk->sk_omem_alloc))
1370 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1371 __func__, atomic_read(&sk->sk_omem_alloc));
1372
1373 if (sk->sk_peer_cred)
1374 put_cred(sk->sk_peer_cred);
1375 put_pid(sk->sk_peer_pid);
1376 put_net(sock_net(sk));
1377 sk_prot_free(sk->sk_prot_creator, sk);
1378 }
1379
1380 void sk_free(struct sock *sk)
1381 {
1382 /*
1383 * We subtract one from sk_wmem_alloc and can know if
1384 * some packets are still in some tx queue.
1385 * If not null, sock_wfree() will call __sk_free(sk) later
1386 */
1387 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1388 __sk_free(sk);
1389 }
1390 EXPORT_SYMBOL(sk_free);
1391
1392 /*
1393 * Last sock_put should drop reference to sk->sk_net. It has already
1394 * been dropped in sk_change_net. Taking reference to stopping namespace
1395 * is not an option.
1396 * Take reference to a socket to remove it from hash _alive_ and after that
1397 * destroy it in the context of init_net.
1398 */
1399 void sk_release_kernel(struct sock *sk)
1400 {
1401 if (sk == NULL || sk->sk_socket == NULL)
1402 return;
1403
1404 sock_hold(sk);
1405 sock_release(sk->sk_socket);
1406 release_net(sock_net(sk));
1407 sock_net_set(sk, get_net(&init_net));
1408 sock_put(sk);
1409 }
1410 EXPORT_SYMBOL(sk_release_kernel);
1411
1412 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1413 {
1414 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1415 sock_update_memcg(newsk);
1416 }
1417
1418 /**
1419 * sk_clone_lock - clone a socket, and lock its clone
1420 * @sk: the socket to clone
1421 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1422 *
1423 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1424 */
1425 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1426 {
1427 struct sock *newsk;
1428
1429 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1430 if (newsk != NULL) {
1431 struct sk_filter *filter;
1432
1433 sock_copy(newsk, sk);
1434
1435 /* SANITY */
1436 get_net(sock_net(newsk));
1437 sk_node_init(&newsk->sk_node);
1438 sock_lock_init(newsk);
1439 bh_lock_sock(newsk);
1440 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1441 newsk->sk_backlog.len = 0;
1442
1443 atomic_set(&newsk->sk_rmem_alloc, 0);
1444 /*
1445 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1446 */
1447 atomic_set(&newsk->sk_wmem_alloc, 1);
1448 atomic_set(&newsk->sk_omem_alloc, 0);
1449 skb_queue_head_init(&newsk->sk_receive_queue);
1450 skb_queue_head_init(&newsk->sk_write_queue);
1451 #ifdef CONFIG_NET_DMA
1452 skb_queue_head_init(&newsk->sk_async_wait_queue);
1453 #endif
1454
1455 spin_lock_init(&newsk->sk_dst_lock);
1456 rwlock_init(&newsk->sk_callback_lock);
1457 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1458 af_callback_keys + newsk->sk_family,
1459 af_family_clock_key_strings[newsk->sk_family]);
1460
1461 newsk->sk_dst_cache = NULL;
1462 newsk->sk_wmem_queued = 0;
1463 newsk->sk_forward_alloc = 0;
1464 newsk->sk_send_head = NULL;
1465 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1466
1467 sock_reset_flag(newsk, SOCK_DONE);
1468 skb_queue_head_init(&newsk->sk_error_queue);
1469
1470 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1471 if (filter != NULL)
1472 sk_filter_charge(newsk, filter);
1473
1474 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1475 /* It is still raw copy of parent, so invalidate
1476 * destructor and make plain sk_free() */
1477 newsk->sk_destruct = NULL;
1478 bh_unlock_sock(newsk);
1479 sk_free(newsk);
1480 newsk = NULL;
1481 goto out;
1482 }
1483
1484 newsk->sk_err = 0;
1485 newsk->sk_priority = 0;
1486 /*
1487 * Before updating sk_refcnt, we must commit prior changes to memory
1488 * (Documentation/RCU/rculist_nulls.txt for details)
1489 */
1490 smp_wmb();
1491 atomic_set(&newsk->sk_refcnt, 2);
1492
1493 /*
1494 * Increment the counter in the same struct proto as the master
1495 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1496 * is the same as sk->sk_prot->socks, as this field was copied
1497 * with memcpy).
1498 *
1499 * This _changes_ the previous behaviour, where
1500 * tcp_create_openreq_child always was incrementing the
1501 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1502 * to be taken into account in all callers. -acme
1503 */
1504 sk_refcnt_debug_inc(newsk);
1505 sk_set_socket(newsk, NULL);
1506 newsk->sk_wq = NULL;
1507
1508 sk_update_clone(sk, newsk);
1509
1510 if (newsk->sk_prot->sockets_allocated)
1511 sk_sockets_allocated_inc(newsk);
1512
1513 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1514 net_enable_timestamp();
1515 }
1516 out:
1517 return newsk;
1518 }
1519 EXPORT_SYMBOL_GPL(sk_clone_lock);
1520
1521 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1522 {
1523 __sk_dst_set(sk, dst);
1524 sk->sk_route_caps = dst->dev->features;
1525 if (sk->sk_route_caps & NETIF_F_GSO)
1526 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1527 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1528 if (sk_can_gso(sk)) {
1529 if (dst->header_len) {
1530 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1531 } else {
1532 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1533 sk->sk_gso_max_size = dst->dev->gso_max_size;
1534 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1535 }
1536 }
1537 }
1538 EXPORT_SYMBOL_GPL(sk_setup_caps);
1539
1540 /*
1541 * Simple resource managers for sockets.
1542 */
1543
1544
1545 /*
1546 * Write buffer destructor automatically called from kfree_skb.
1547 */
1548 void sock_wfree(struct sk_buff *skb)
1549 {
1550 struct sock *sk = skb->sk;
1551 unsigned int len = skb->truesize;
1552
1553 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1554 /*
1555 * Keep a reference on sk_wmem_alloc, this will be released
1556 * after sk_write_space() call
1557 */
1558 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1559 sk->sk_write_space(sk);
1560 len = 1;
1561 }
1562 /*
1563 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1564 * could not do because of in-flight packets
1565 */
1566 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1567 __sk_free(sk);
1568 }
1569 EXPORT_SYMBOL(sock_wfree);
1570
1571 void skb_orphan_partial(struct sk_buff *skb)
1572 {
1573 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1574 * so we do not completely orphan skb, but transfert all
1575 * accounted bytes but one, to avoid unexpected reorders.
1576 */
1577 if (skb->destructor == sock_wfree
1578 #ifdef CONFIG_INET
1579 || skb->destructor == tcp_wfree
1580 #endif
1581 ) {
1582 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1583 skb->truesize = 1;
1584 } else {
1585 skb_orphan(skb);
1586 }
1587 }
1588 EXPORT_SYMBOL(skb_orphan_partial);
1589
1590 /*
1591 * Read buffer destructor automatically called from kfree_skb.
1592 */
1593 void sock_rfree(struct sk_buff *skb)
1594 {
1595 struct sock *sk = skb->sk;
1596 unsigned int len = skb->truesize;
1597
1598 atomic_sub(len, &sk->sk_rmem_alloc);
1599 sk_mem_uncharge(sk, len);
1600 }
1601 EXPORT_SYMBOL(sock_rfree);
1602
1603 void sock_edemux(struct sk_buff *skb)
1604 {
1605 struct sock *sk = skb->sk;
1606
1607 #ifdef CONFIG_INET
1608 if (sk->sk_state == TCP_TIME_WAIT)
1609 inet_twsk_put(inet_twsk(sk));
1610 else
1611 #endif
1612 sock_put(sk);
1613 }
1614 EXPORT_SYMBOL(sock_edemux);
1615
1616 kuid_t sock_i_uid(struct sock *sk)
1617 {
1618 kuid_t uid;
1619
1620 read_lock_bh(&sk->sk_callback_lock);
1621 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1622 read_unlock_bh(&sk->sk_callback_lock);
1623 return uid;
1624 }
1625 EXPORT_SYMBOL(sock_i_uid);
1626
1627 unsigned long sock_i_ino(struct sock *sk)
1628 {
1629 unsigned long ino;
1630
1631 read_lock_bh(&sk->sk_callback_lock);
1632 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1633 read_unlock_bh(&sk->sk_callback_lock);
1634 return ino;
1635 }
1636 EXPORT_SYMBOL(sock_i_ino);
1637
1638 /*
1639 * Allocate a skb from the socket's send buffer.
1640 */
1641 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1642 gfp_t priority)
1643 {
1644 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1645 struct sk_buff *skb = alloc_skb(size, priority);
1646 if (skb) {
1647 skb_set_owner_w(skb, sk);
1648 return skb;
1649 }
1650 }
1651 return NULL;
1652 }
1653 EXPORT_SYMBOL(sock_wmalloc);
1654
1655 /*
1656 * Allocate a memory block from the socket's option memory buffer.
1657 */
1658 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1659 {
1660 if ((unsigned int)size <= sysctl_optmem_max &&
1661 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1662 void *mem;
1663 /* First do the add, to avoid the race if kmalloc
1664 * might sleep.
1665 */
1666 atomic_add(size, &sk->sk_omem_alloc);
1667 mem = kmalloc(size, priority);
1668 if (mem)
1669 return mem;
1670 atomic_sub(size, &sk->sk_omem_alloc);
1671 }
1672 return NULL;
1673 }
1674 EXPORT_SYMBOL(sock_kmalloc);
1675
1676 /*
1677 * Free an option memory block.
1678 */
1679 void sock_kfree_s(struct sock *sk, void *mem, int size)
1680 {
1681 kfree(mem);
1682 atomic_sub(size, &sk->sk_omem_alloc);
1683 }
1684 EXPORT_SYMBOL(sock_kfree_s);
1685
1686 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1687 I think, these locks should be removed for datagram sockets.
1688 */
1689 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1690 {
1691 DEFINE_WAIT(wait);
1692
1693 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1694 for (;;) {
1695 if (!timeo)
1696 break;
1697 if (signal_pending(current))
1698 break;
1699 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1700 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1701 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1702 break;
1703 if (sk->sk_shutdown & SEND_SHUTDOWN)
1704 break;
1705 if (sk->sk_err)
1706 break;
1707 timeo = schedule_timeout(timeo);
1708 }
1709 finish_wait(sk_sleep(sk), &wait);
1710 return timeo;
1711 }
1712
1713
1714 /*
1715 * Generic send/receive buffer handlers
1716 */
1717
1718 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1719 unsigned long data_len, int noblock,
1720 int *errcode, int max_page_order)
1721 {
1722 struct sk_buff *skb = NULL;
1723 unsigned long chunk;
1724 gfp_t gfp_mask;
1725 long timeo;
1726 int err;
1727 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1728 struct page *page;
1729 int i;
1730
1731 err = -EMSGSIZE;
1732 if (npages > MAX_SKB_FRAGS)
1733 goto failure;
1734
1735 timeo = sock_sndtimeo(sk, noblock);
1736 while (!skb) {
1737 err = sock_error(sk);
1738 if (err != 0)
1739 goto failure;
1740
1741 err = -EPIPE;
1742 if (sk->sk_shutdown & SEND_SHUTDOWN)
1743 goto failure;
1744
1745 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) {
1746 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1747 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1748 err = -EAGAIN;
1749 if (!timeo)
1750 goto failure;
1751 if (signal_pending(current))
1752 goto interrupted;
1753 timeo = sock_wait_for_wmem(sk, timeo);
1754 continue;
1755 }
1756
1757 err = -ENOBUFS;
1758 gfp_mask = sk->sk_allocation;
1759 if (gfp_mask & __GFP_WAIT)
1760 gfp_mask |= __GFP_REPEAT;
1761
1762 skb = alloc_skb(header_len, gfp_mask);
1763 if (!skb)
1764 goto failure;
1765
1766 skb->truesize += data_len;
1767
1768 for (i = 0; npages > 0; i++) {
1769 int order = max_page_order;
1770
1771 while (order) {
1772 if (npages >= 1 << order) {
1773 page = alloc_pages(sk->sk_allocation |
1774 __GFP_COMP | __GFP_NOWARN,
1775 order);
1776 if (page)
1777 goto fill_page;
1778 }
1779 order--;
1780 }
1781 page = alloc_page(sk->sk_allocation);
1782 if (!page)
1783 goto failure;
1784 fill_page:
1785 chunk = min_t(unsigned long, data_len,
1786 PAGE_SIZE << order);
1787 skb_fill_page_desc(skb, i, page, 0, chunk);
1788 data_len -= chunk;
1789 npages -= 1 << order;
1790 }
1791 }
1792
1793 skb_set_owner_w(skb, sk);
1794 return skb;
1795
1796 interrupted:
1797 err = sock_intr_errno(timeo);
1798 failure:
1799 kfree_skb(skb);
1800 *errcode = err;
1801 return NULL;
1802 }
1803 EXPORT_SYMBOL(sock_alloc_send_pskb);
1804
1805 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1806 int noblock, int *errcode)
1807 {
1808 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1809 }
1810 EXPORT_SYMBOL(sock_alloc_send_skb);
1811
1812 /* On 32bit arches, an skb frag is limited to 2^15 */
1813 #define SKB_FRAG_PAGE_ORDER get_order(32768)
1814
1815 /**
1816 * skb_page_frag_refill - check that a page_frag contains enough room
1817 * @sz: minimum size of the fragment we want to get
1818 * @pfrag: pointer to page_frag
1819 * @prio: priority for memory allocation
1820 *
1821 * Note: While this allocator tries to use high order pages, there is
1822 * no guarantee that allocations succeed. Therefore, @sz MUST be
1823 * less or equal than PAGE_SIZE.
1824 */
1825 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio)
1826 {
1827 int order;
1828
1829 if (pfrag->page) {
1830 if (atomic_read(&pfrag->page->_count) == 1) {
1831 pfrag->offset = 0;
1832 return true;
1833 }
1834 if (pfrag->offset + sz <= pfrag->size)
1835 return true;
1836 put_page(pfrag->page);
1837 }
1838
1839 /* We restrict high order allocations to users that can afford to wait */
1840 order = (prio & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1841
1842 do {
1843 gfp_t gfp = prio;
1844
1845 if (order)
1846 gfp |= __GFP_COMP | __GFP_NOWARN;
1847 pfrag->page = alloc_pages(gfp, order);
1848 if (likely(pfrag->page)) {
1849 pfrag->offset = 0;
1850 pfrag->size = PAGE_SIZE << order;
1851 return true;
1852 }
1853 } while (--order >= 0);
1854
1855 return false;
1856 }
1857 EXPORT_SYMBOL(skb_page_frag_refill);
1858
1859 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1860 {
1861 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
1862 return true;
1863
1864 sk_enter_memory_pressure(sk);
1865 sk_stream_moderate_sndbuf(sk);
1866 return false;
1867 }
1868 EXPORT_SYMBOL(sk_page_frag_refill);
1869
1870 static void __lock_sock(struct sock *sk)
1871 __releases(&sk->sk_lock.slock)
1872 __acquires(&sk->sk_lock.slock)
1873 {
1874 DEFINE_WAIT(wait);
1875
1876 for (;;) {
1877 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1878 TASK_UNINTERRUPTIBLE);
1879 spin_unlock_bh(&sk->sk_lock.slock);
1880 schedule();
1881 spin_lock_bh(&sk->sk_lock.slock);
1882 if (!sock_owned_by_user(sk))
1883 break;
1884 }
1885 finish_wait(&sk->sk_lock.wq, &wait);
1886 }
1887
1888 static void __release_sock(struct sock *sk)
1889 __releases(&sk->sk_lock.slock)
1890 __acquires(&sk->sk_lock.slock)
1891 {
1892 struct sk_buff *skb = sk->sk_backlog.head;
1893
1894 do {
1895 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1896 bh_unlock_sock(sk);
1897
1898 do {
1899 struct sk_buff *next = skb->next;
1900
1901 prefetch(next);
1902 WARN_ON_ONCE(skb_dst_is_noref(skb));
1903 skb->next = NULL;
1904 sk_backlog_rcv(sk, skb);
1905
1906 /*
1907 * We are in process context here with softirqs
1908 * disabled, use cond_resched_softirq() to preempt.
1909 * This is safe to do because we've taken the backlog
1910 * queue private:
1911 */
1912 cond_resched_softirq();
1913
1914 skb = next;
1915 } while (skb != NULL);
1916
1917 bh_lock_sock(sk);
1918 } while ((skb = sk->sk_backlog.head) != NULL);
1919
1920 /*
1921 * Doing the zeroing here guarantee we can not loop forever
1922 * while a wild producer attempts to flood us.
1923 */
1924 sk->sk_backlog.len = 0;
1925 }
1926
1927 /**
1928 * sk_wait_data - wait for data to arrive at sk_receive_queue
1929 * @sk: sock to wait on
1930 * @timeo: for how long
1931 *
1932 * Now socket state including sk->sk_err is changed only under lock,
1933 * hence we may omit checks after joining wait queue.
1934 * We check receive queue before schedule() only as optimization;
1935 * it is very likely that release_sock() added new data.
1936 */
1937 int sk_wait_data(struct sock *sk, long *timeo)
1938 {
1939 int rc;
1940 DEFINE_WAIT(wait);
1941
1942 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1943 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1944 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1945 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1946 finish_wait(sk_sleep(sk), &wait);
1947 return rc;
1948 }
1949 EXPORT_SYMBOL(sk_wait_data);
1950
1951 /**
1952 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1953 * @sk: socket
1954 * @size: memory size to allocate
1955 * @kind: allocation type
1956 *
1957 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1958 * rmem allocation. This function assumes that protocols which have
1959 * memory_pressure use sk_wmem_queued as write buffer accounting.
1960 */
1961 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1962 {
1963 struct proto *prot = sk->sk_prot;
1964 int amt = sk_mem_pages(size);
1965 long allocated;
1966 int parent_status = UNDER_LIMIT;
1967
1968 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1969
1970 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1971
1972 /* Under limit. */
1973 if (parent_status == UNDER_LIMIT &&
1974 allocated <= sk_prot_mem_limits(sk, 0)) {
1975 sk_leave_memory_pressure(sk);
1976 return 1;
1977 }
1978
1979 /* Under pressure. (we or our parents) */
1980 if ((parent_status > SOFT_LIMIT) ||
1981 allocated > sk_prot_mem_limits(sk, 1))
1982 sk_enter_memory_pressure(sk);
1983
1984 /* Over hard limit (we or our parents) */
1985 if ((parent_status == OVER_LIMIT) ||
1986 (allocated > sk_prot_mem_limits(sk, 2)))
1987 goto suppress_allocation;
1988
1989 /* guarantee minimum buffer size under pressure */
1990 if (kind == SK_MEM_RECV) {
1991 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1992 return 1;
1993
1994 } else { /* SK_MEM_SEND */
1995 if (sk->sk_type == SOCK_STREAM) {
1996 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1997 return 1;
1998 } else if (atomic_read(&sk->sk_wmem_alloc) <
1999 prot->sysctl_wmem[0])
2000 return 1;
2001 }
2002
2003 if (sk_has_memory_pressure(sk)) {
2004 int alloc;
2005
2006 if (!sk_under_memory_pressure(sk))
2007 return 1;
2008 alloc = sk_sockets_allocated_read_positive(sk);
2009 if (sk_prot_mem_limits(sk, 2) > alloc *
2010 sk_mem_pages(sk->sk_wmem_queued +
2011 atomic_read(&sk->sk_rmem_alloc) +
2012 sk->sk_forward_alloc))
2013 return 1;
2014 }
2015
2016 suppress_allocation:
2017
2018 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2019 sk_stream_moderate_sndbuf(sk);
2020
2021 /* Fail only if socket is _under_ its sndbuf.
2022 * In this case we cannot block, so that we have to fail.
2023 */
2024 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2025 return 1;
2026 }
2027
2028 trace_sock_exceed_buf_limit(sk, prot, allocated);
2029
2030 /* Alas. Undo changes. */
2031 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2032
2033 sk_memory_allocated_sub(sk, amt);
2034
2035 return 0;
2036 }
2037 EXPORT_SYMBOL(__sk_mem_schedule);
2038
2039 /**
2040 * __sk_reclaim - reclaim memory_allocated
2041 * @sk: socket
2042 */
2043 void __sk_mem_reclaim(struct sock *sk)
2044 {
2045 sk_memory_allocated_sub(sk,
2046 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2047 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2048
2049 if (sk_under_memory_pressure(sk) &&
2050 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2051 sk_leave_memory_pressure(sk);
2052 }
2053 EXPORT_SYMBOL(__sk_mem_reclaim);
2054
2055
2056 /*
2057 * Set of default routines for initialising struct proto_ops when
2058 * the protocol does not support a particular function. In certain
2059 * cases where it makes no sense for a protocol to have a "do nothing"
2060 * function, some default processing is provided.
2061 */
2062
2063 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2064 {
2065 return -EOPNOTSUPP;
2066 }
2067 EXPORT_SYMBOL(sock_no_bind);
2068
2069 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2070 int len, int flags)
2071 {
2072 return -EOPNOTSUPP;
2073 }
2074 EXPORT_SYMBOL(sock_no_connect);
2075
2076 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2077 {
2078 return -EOPNOTSUPP;
2079 }
2080 EXPORT_SYMBOL(sock_no_socketpair);
2081
2082 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2083 {
2084 return -EOPNOTSUPP;
2085 }
2086 EXPORT_SYMBOL(sock_no_accept);
2087
2088 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2089 int *len, int peer)
2090 {
2091 return -EOPNOTSUPP;
2092 }
2093 EXPORT_SYMBOL(sock_no_getname);
2094
2095 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2096 {
2097 return 0;
2098 }
2099 EXPORT_SYMBOL(sock_no_poll);
2100
2101 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2102 {
2103 return -EOPNOTSUPP;
2104 }
2105 EXPORT_SYMBOL(sock_no_ioctl);
2106
2107 int sock_no_listen(struct socket *sock, int backlog)
2108 {
2109 return -EOPNOTSUPP;
2110 }
2111 EXPORT_SYMBOL(sock_no_listen);
2112
2113 int sock_no_shutdown(struct socket *sock, int how)
2114 {
2115 return -EOPNOTSUPP;
2116 }
2117 EXPORT_SYMBOL(sock_no_shutdown);
2118
2119 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2120 char __user *optval, unsigned int optlen)
2121 {
2122 return -EOPNOTSUPP;
2123 }
2124 EXPORT_SYMBOL(sock_no_setsockopt);
2125
2126 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2127 char __user *optval, int __user *optlen)
2128 {
2129 return -EOPNOTSUPP;
2130 }
2131 EXPORT_SYMBOL(sock_no_getsockopt);
2132
2133 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2134 size_t len)
2135 {
2136 return -EOPNOTSUPP;
2137 }
2138 EXPORT_SYMBOL(sock_no_sendmsg);
2139
2140 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2141 size_t len, int flags)
2142 {
2143 return -EOPNOTSUPP;
2144 }
2145 EXPORT_SYMBOL(sock_no_recvmsg);
2146
2147 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2148 {
2149 /* Mirror missing mmap method error code */
2150 return -ENODEV;
2151 }
2152 EXPORT_SYMBOL(sock_no_mmap);
2153
2154 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2155 {
2156 ssize_t res;
2157 struct msghdr msg = {.msg_flags = flags};
2158 struct kvec iov;
2159 char *kaddr = kmap(page);
2160 iov.iov_base = kaddr + offset;
2161 iov.iov_len = size;
2162 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2163 kunmap(page);
2164 return res;
2165 }
2166 EXPORT_SYMBOL(sock_no_sendpage);
2167
2168 /*
2169 * Default Socket Callbacks
2170 */
2171
2172 static void sock_def_wakeup(struct sock *sk)
2173 {
2174 struct socket_wq *wq;
2175
2176 rcu_read_lock();
2177 wq = rcu_dereference(sk->sk_wq);
2178 if (wq_has_sleeper(wq))
2179 wake_up_interruptible_all(&wq->wait);
2180 rcu_read_unlock();
2181 }
2182
2183 static void sock_def_error_report(struct sock *sk)
2184 {
2185 struct socket_wq *wq;
2186
2187 rcu_read_lock();
2188 wq = rcu_dereference(sk->sk_wq);
2189 if (wq_has_sleeper(wq))
2190 wake_up_interruptible_poll(&wq->wait, POLLERR);
2191 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2192 rcu_read_unlock();
2193 }
2194
2195 static void sock_def_readable(struct sock *sk, int len)
2196 {
2197 struct socket_wq *wq;
2198
2199 rcu_read_lock();
2200 wq = rcu_dereference(sk->sk_wq);
2201 if (wq_has_sleeper(wq))
2202 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2203 POLLRDNORM | POLLRDBAND);
2204 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2205 rcu_read_unlock();
2206 }
2207
2208 static void sock_def_write_space(struct sock *sk)
2209 {
2210 struct socket_wq *wq;
2211
2212 rcu_read_lock();
2213
2214 /* Do not wake up a writer until he can make "significant"
2215 * progress. --DaveM
2216 */
2217 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2218 wq = rcu_dereference(sk->sk_wq);
2219 if (wq_has_sleeper(wq))
2220 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2221 POLLWRNORM | POLLWRBAND);
2222
2223 /* Should agree with poll, otherwise some programs break */
2224 if (sock_writeable(sk))
2225 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2226 }
2227
2228 rcu_read_unlock();
2229 }
2230
2231 static void sock_def_destruct(struct sock *sk)
2232 {
2233 kfree(sk->sk_protinfo);
2234 }
2235
2236 void sk_send_sigurg(struct sock *sk)
2237 {
2238 if (sk->sk_socket && sk->sk_socket->file)
2239 if (send_sigurg(&sk->sk_socket->file->f_owner))
2240 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2241 }
2242 EXPORT_SYMBOL(sk_send_sigurg);
2243
2244 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2245 unsigned long expires)
2246 {
2247 if (!mod_timer(timer, expires))
2248 sock_hold(sk);
2249 }
2250 EXPORT_SYMBOL(sk_reset_timer);
2251
2252 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2253 {
2254 if (del_timer(timer))
2255 __sock_put(sk);
2256 }
2257 EXPORT_SYMBOL(sk_stop_timer);
2258
2259 void sock_init_data(struct socket *sock, struct sock *sk)
2260 {
2261 skb_queue_head_init(&sk->sk_receive_queue);
2262 skb_queue_head_init(&sk->sk_write_queue);
2263 skb_queue_head_init(&sk->sk_error_queue);
2264 #ifdef CONFIG_NET_DMA
2265 skb_queue_head_init(&sk->sk_async_wait_queue);
2266 #endif
2267
2268 sk->sk_send_head = NULL;
2269
2270 init_timer(&sk->sk_timer);
2271
2272 sk->sk_allocation = GFP_KERNEL;
2273 sk->sk_rcvbuf = sysctl_rmem_default;
2274 sk->sk_sndbuf = sysctl_wmem_default;
2275 sk->sk_state = TCP_CLOSE;
2276 sk_set_socket(sk, sock);
2277
2278 sock_set_flag(sk, SOCK_ZAPPED);
2279
2280 if (sock) {
2281 sk->sk_type = sock->type;
2282 sk->sk_wq = sock->wq;
2283 sock->sk = sk;
2284 } else
2285 sk->sk_wq = NULL;
2286
2287 spin_lock_init(&sk->sk_dst_lock);
2288 rwlock_init(&sk->sk_callback_lock);
2289 lockdep_set_class_and_name(&sk->sk_callback_lock,
2290 af_callback_keys + sk->sk_family,
2291 af_family_clock_key_strings[sk->sk_family]);
2292
2293 sk->sk_state_change = sock_def_wakeup;
2294 sk->sk_data_ready = sock_def_readable;
2295 sk->sk_write_space = sock_def_write_space;
2296 sk->sk_error_report = sock_def_error_report;
2297 sk->sk_destruct = sock_def_destruct;
2298
2299 sk->sk_frag.page = NULL;
2300 sk->sk_frag.offset = 0;
2301 sk->sk_peek_off = -1;
2302
2303 sk->sk_peer_pid = NULL;
2304 sk->sk_peer_cred = NULL;
2305 sk->sk_write_pending = 0;
2306 sk->sk_rcvlowat = 1;
2307 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2308 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2309
2310 sk->sk_stamp = ktime_set(-1L, 0);
2311
2312 #ifdef CONFIG_NET_RX_BUSY_POLL
2313 sk->sk_napi_id = 0;
2314 sk->sk_ll_usec = sysctl_net_busy_read;
2315 #endif
2316
2317 sk->sk_max_pacing_rate = ~0U;
2318 sk->sk_pacing_rate = ~0U;
2319 /*
2320 * Before updating sk_refcnt, we must commit prior changes to memory
2321 * (Documentation/RCU/rculist_nulls.txt for details)
2322 */
2323 smp_wmb();
2324 atomic_set(&sk->sk_refcnt, 1);
2325 atomic_set(&sk->sk_drops, 0);
2326 }
2327 EXPORT_SYMBOL(sock_init_data);
2328
2329 void lock_sock_nested(struct sock *sk, int subclass)
2330 {
2331 might_sleep();
2332 spin_lock_bh(&sk->sk_lock.slock);
2333 if (sk->sk_lock.owned)
2334 __lock_sock(sk);
2335 sk->sk_lock.owned = 1;
2336 spin_unlock(&sk->sk_lock.slock);
2337 /*
2338 * The sk_lock has mutex_lock() semantics here:
2339 */
2340 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2341 local_bh_enable();
2342 }
2343 EXPORT_SYMBOL(lock_sock_nested);
2344
2345 void release_sock(struct sock *sk)
2346 {
2347 /*
2348 * The sk_lock has mutex_unlock() semantics:
2349 */
2350 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2351
2352 spin_lock_bh(&sk->sk_lock.slock);
2353 if (sk->sk_backlog.tail)
2354 __release_sock(sk);
2355
2356 if (sk->sk_prot->release_cb)
2357 sk->sk_prot->release_cb(sk);
2358
2359 sk->sk_lock.owned = 0;
2360 if (waitqueue_active(&sk->sk_lock.wq))
2361 wake_up(&sk->sk_lock.wq);
2362 spin_unlock_bh(&sk->sk_lock.slock);
2363 }
2364 EXPORT_SYMBOL(release_sock);
2365
2366 /**
2367 * lock_sock_fast - fast version of lock_sock
2368 * @sk: socket
2369 *
2370 * This version should be used for very small section, where process wont block
2371 * return false if fast path is taken
2372 * sk_lock.slock locked, owned = 0, BH disabled
2373 * return true if slow path is taken
2374 * sk_lock.slock unlocked, owned = 1, BH enabled
2375 */
2376 bool lock_sock_fast(struct sock *sk)
2377 {
2378 might_sleep();
2379 spin_lock_bh(&sk->sk_lock.slock);
2380
2381 if (!sk->sk_lock.owned)
2382 /*
2383 * Note : We must disable BH
2384 */
2385 return false;
2386
2387 __lock_sock(sk);
2388 sk->sk_lock.owned = 1;
2389 spin_unlock(&sk->sk_lock.slock);
2390 /*
2391 * The sk_lock has mutex_lock() semantics here:
2392 */
2393 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2394 local_bh_enable();
2395 return true;
2396 }
2397 EXPORT_SYMBOL(lock_sock_fast);
2398
2399 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2400 {
2401 struct timeval tv;
2402 if (!sock_flag(sk, SOCK_TIMESTAMP))
2403 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2404 tv = ktime_to_timeval(sk->sk_stamp);
2405 if (tv.tv_sec == -1)
2406 return -ENOENT;
2407 if (tv.tv_sec == 0) {
2408 sk->sk_stamp = ktime_get_real();
2409 tv = ktime_to_timeval(sk->sk_stamp);
2410 }
2411 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2412 }
2413 EXPORT_SYMBOL(sock_get_timestamp);
2414
2415 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2416 {
2417 struct timespec ts;
2418 if (!sock_flag(sk, SOCK_TIMESTAMP))
2419 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2420 ts = ktime_to_timespec(sk->sk_stamp);
2421 if (ts.tv_sec == -1)
2422 return -ENOENT;
2423 if (ts.tv_sec == 0) {
2424 sk->sk_stamp = ktime_get_real();
2425 ts = ktime_to_timespec(sk->sk_stamp);
2426 }
2427 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2428 }
2429 EXPORT_SYMBOL(sock_get_timestampns);
2430
2431 void sock_enable_timestamp(struct sock *sk, int flag)
2432 {
2433 if (!sock_flag(sk, flag)) {
2434 unsigned long previous_flags = sk->sk_flags;
2435
2436 sock_set_flag(sk, flag);
2437 /*
2438 * we just set one of the two flags which require net
2439 * time stamping, but time stamping might have been on
2440 * already because of the other one
2441 */
2442 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2443 net_enable_timestamp();
2444 }
2445 }
2446
2447 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2448 int level, int type)
2449 {
2450 struct sock_exterr_skb *serr;
2451 struct sk_buff *skb, *skb2;
2452 int copied, err;
2453
2454 err = -EAGAIN;
2455 skb = skb_dequeue(&sk->sk_error_queue);
2456 if (skb == NULL)
2457 goto out;
2458
2459 copied = skb->len;
2460 if (copied > len) {
2461 msg->msg_flags |= MSG_TRUNC;
2462 copied = len;
2463 }
2464 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2465 if (err)
2466 goto out_free_skb;
2467
2468 sock_recv_timestamp(msg, sk, skb);
2469
2470 serr = SKB_EXT_ERR(skb);
2471 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2472
2473 msg->msg_flags |= MSG_ERRQUEUE;
2474 err = copied;
2475
2476 /* Reset and regenerate socket error */
2477 spin_lock_bh(&sk->sk_error_queue.lock);
2478 sk->sk_err = 0;
2479 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2480 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2481 spin_unlock_bh(&sk->sk_error_queue.lock);
2482 sk->sk_error_report(sk);
2483 } else
2484 spin_unlock_bh(&sk->sk_error_queue.lock);
2485
2486 out_free_skb:
2487 kfree_skb(skb);
2488 out:
2489 return err;
2490 }
2491 EXPORT_SYMBOL(sock_recv_errqueue);
2492
2493 /*
2494 * Get a socket option on an socket.
2495 *
2496 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2497 * asynchronous errors should be reported by getsockopt. We assume
2498 * this means if you specify SO_ERROR (otherwise whats the point of it).
2499 */
2500 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2501 char __user *optval, int __user *optlen)
2502 {
2503 struct sock *sk = sock->sk;
2504
2505 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2506 }
2507 EXPORT_SYMBOL(sock_common_getsockopt);
2508
2509 #ifdef CONFIG_COMPAT
2510 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2511 char __user *optval, int __user *optlen)
2512 {
2513 struct sock *sk = sock->sk;
2514
2515 if (sk->sk_prot->compat_getsockopt != NULL)
2516 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2517 optval, optlen);
2518 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2519 }
2520 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2521 #endif
2522
2523 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2524 struct msghdr *msg, size_t size, int flags)
2525 {
2526 struct sock *sk = sock->sk;
2527 int addr_len = 0;
2528 int err;
2529
2530 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2531 flags & ~MSG_DONTWAIT, &addr_len);
2532 if (err >= 0)
2533 msg->msg_namelen = addr_len;
2534 return err;
2535 }
2536 EXPORT_SYMBOL(sock_common_recvmsg);
2537
2538 /*
2539 * Set socket options on an inet socket.
2540 */
2541 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2542 char __user *optval, unsigned int optlen)
2543 {
2544 struct sock *sk = sock->sk;
2545
2546 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2547 }
2548 EXPORT_SYMBOL(sock_common_setsockopt);
2549
2550 #ifdef CONFIG_COMPAT
2551 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2552 char __user *optval, unsigned int optlen)
2553 {
2554 struct sock *sk = sock->sk;
2555
2556 if (sk->sk_prot->compat_setsockopt != NULL)
2557 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2558 optval, optlen);
2559 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2560 }
2561 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2562 #endif
2563
2564 void sk_common_release(struct sock *sk)
2565 {
2566 if (sk->sk_prot->destroy)
2567 sk->sk_prot->destroy(sk);
2568
2569 /*
2570 * Observation: when sock_common_release is called, processes have
2571 * no access to socket. But net still has.
2572 * Step one, detach it from networking:
2573 *
2574 * A. Remove from hash tables.
2575 */
2576
2577 sk->sk_prot->unhash(sk);
2578
2579 /*
2580 * In this point socket cannot receive new packets, but it is possible
2581 * that some packets are in flight because some CPU runs receiver and
2582 * did hash table lookup before we unhashed socket. They will achieve
2583 * receive queue and will be purged by socket destructor.
2584 *
2585 * Also we still have packets pending on receive queue and probably,
2586 * our own packets waiting in device queues. sock_destroy will drain
2587 * receive queue, but transmitted packets will delay socket destruction
2588 * until the last reference will be released.
2589 */
2590
2591 sock_orphan(sk);
2592
2593 xfrm_sk_free_policy(sk);
2594
2595 sk_refcnt_debug_release(sk);
2596
2597 if (sk->sk_frag.page) {
2598 put_page(sk->sk_frag.page);
2599 sk->sk_frag.page = NULL;
2600 }
2601
2602 sock_put(sk);
2603 }
2604 EXPORT_SYMBOL(sk_common_release);
2605
2606 #ifdef CONFIG_PROC_FS
2607 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2608 struct prot_inuse {
2609 int val[PROTO_INUSE_NR];
2610 };
2611
2612 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2613
2614 #ifdef CONFIG_NET_NS
2615 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2616 {
2617 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2618 }
2619 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2620
2621 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2622 {
2623 int cpu, idx = prot->inuse_idx;
2624 int res = 0;
2625
2626 for_each_possible_cpu(cpu)
2627 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2628
2629 return res >= 0 ? res : 0;
2630 }
2631 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2632
2633 static int __net_init sock_inuse_init_net(struct net *net)
2634 {
2635 net->core.inuse = alloc_percpu(struct prot_inuse);
2636 return net->core.inuse ? 0 : -ENOMEM;
2637 }
2638
2639 static void __net_exit sock_inuse_exit_net(struct net *net)
2640 {
2641 free_percpu(net->core.inuse);
2642 }
2643
2644 static struct pernet_operations net_inuse_ops = {
2645 .init = sock_inuse_init_net,
2646 .exit = sock_inuse_exit_net,
2647 };
2648
2649 static __init int net_inuse_init(void)
2650 {
2651 if (register_pernet_subsys(&net_inuse_ops))
2652 panic("Cannot initialize net inuse counters");
2653
2654 return 0;
2655 }
2656
2657 core_initcall(net_inuse_init);
2658 #else
2659 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2660
2661 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2662 {
2663 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2664 }
2665 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2666
2667 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2668 {
2669 int cpu, idx = prot->inuse_idx;
2670 int res = 0;
2671
2672 for_each_possible_cpu(cpu)
2673 res += per_cpu(prot_inuse, cpu).val[idx];
2674
2675 return res >= 0 ? res : 0;
2676 }
2677 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2678 #endif
2679
2680 static void assign_proto_idx(struct proto *prot)
2681 {
2682 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2683
2684 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2685 pr_err("PROTO_INUSE_NR exhausted\n");
2686 return;
2687 }
2688
2689 set_bit(prot->inuse_idx, proto_inuse_idx);
2690 }
2691
2692 static void release_proto_idx(struct proto *prot)
2693 {
2694 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2695 clear_bit(prot->inuse_idx, proto_inuse_idx);
2696 }
2697 #else
2698 static inline void assign_proto_idx(struct proto *prot)
2699 {
2700 }
2701
2702 static inline void release_proto_idx(struct proto *prot)
2703 {
2704 }
2705 #endif
2706
2707 int proto_register(struct proto *prot, int alloc_slab)
2708 {
2709 if (alloc_slab) {
2710 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2711 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2712 NULL);
2713
2714 if (prot->slab == NULL) {
2715 pr_crit("%s: Can't create sock SLAB cache!\n",
2716 prot->name);
2717 goto out;
2718 }
2719
2720 if (prot->rsk_prot != NULL) {
2721 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2722 if (prot->rsk_prot->slab_name == NULL)
2723 goto out_free_sock_slab;
2724
2725 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2726 prot->rsk_prot->obj_size, 0,
2727 SLAB_HWCACHE_ALIGN, NULL);
2728
2729 if (prot->rsk_prot->slab == NULL) {
2730 pr_crit("%s: Can't create request sock SLAB cache!\n",
2731 prot->name);
2732 goto out_free_request_sock_slab_name;
2733 }
2734 }
2735
2736 if (prot->twsk_prot != NULL) {
2737 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2738
2739 if (prot->twsk_prot->twsk_slab_name == NULL)
2740 goto out_free_request_sock_slab;
2741
2742 prot->twsk_prot->twsk_slab =
2743 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2744 prot->twsk_prot->twsk_obj_size,
2745 0,
2746 SLAB_HWCACHE_ALIGN |
2747 prot->slab_flags,
2748 NULL);
2749 if (prot->twsk_prot->twsk_slab == NULL)
2750 goto out_free_timewait_sock_slab_name;
2751 }
2752 }
2753
2754 mutex_lock(&proto_list_mutex);
2755 list_add(&prot->node, &proto_list);
2756 assign_proto_idx(prot);
2757 mutex_unlock(&proto_list_mutex);
2758 return 0;
2759
2760 out_free_timewait_sock_slab_name:
2761 kfree(prot->twsk_prot->twsk_slab_name);
2762 out_free_request_sock_slab:
2763 if (prot->rsk_prot && prot->rsk_prot->slab) {
2764 kmem_cache_destroy(prot->rsk_prot->slab);
2765 prot->rsk_prot->slab = NULL;
2766 }
2767 out_free_request_sock_slab_name:
2768 if (prot->rsk_prot)
2769 kfree(prot->rsk_prot->slab_name);
2770 out_free_sock_slab:
2771 kmem_cache_destroy(prot->slab);
2772 prot->slab = NULL;
2773 out:
2774 return -ENOBUFS;
2775 }
2776 EXPORT_SYMBOL(proto_register);
2777
2778 void proto_unregister(struct proto *prot)
2779 {
2780 mutex_lock(&proto_list_mutex);
2781 release_proto_idx(prot);
2782 list_del(&prot->node);
2783 mutex_unlock(&proto_list_mutex);
2784
2785 if (prot->slab != NULL) {
2786 kmem_cache_destroy(prot->slab);
2787 prot->slab = NULL;
2788 }
2789
2790 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2791 kmem_cache_destroy(prot->rsk_prot->slab);
2792 kfree(prot->rsk_prot->slab_name);
2793 prot->rsk_prot->slab = NULL;
2794 }
2795
2796 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2797 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2798 kfree(prot->twsk_prot->twsk_slab_name);
2799 prot->twsk_prot->twsk_slab = NULL;
2800 }
2801 }
2802 EXPORT_SYMBOL(proto_unregister);
2803
2804 #ifdef CONFIG_PROC_FS
2805 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2806 __acquires(proto_list_mutex)
2807 {
2808 mutex_lock(&proto_list_mutex);
2809 return seq_list_start_head(&proto_list, *pos);
2810 }
2811
2812 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2813 {
2814 return seq_list_next(v, &proto_list, pos);
2815 }
2816
2817 static void proto_seq_stop(struct seq_file *seq, void *v)
2818 __releases(proto_list_mutex)
2819 {
2820 mutex_unlock(&proto_list_mutex);
2821 }
2822
2823 static char proto_method_implemented(const void *method)
2824 {
2825 return method == NULL ? 'n' : 'y';
2826 }
2827 static long sock_prot_memory_allocated(struct proto *proto)
2828 {
2829 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2830 }
2831
2832 static char *sock_prot_memory_pressure(struct proto *proto)
2833 {
2834 return proto->memory_pressure != NULL ?
2835 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2836 }
2837
2838 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2839 {
2840
2841 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2842 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2843 proto->name,
2844 proto->obj_size,
2845 sock_prot_inuse_get(seq_file_net(seq), proto),
2846 sock_prot_memory_allocated(proto),
2847 sock_prot_memory_pressure(proto),
2848 proto->max_header,
2849 proto->slab == NULL ? "no" : "yes",
2850 module_name(proto->owner),
2851 proto_method_implemented(proto->close),
2852 proto_method_implemented(proto->connect),
2853 proto_method_implemented(proto->disconnect),
2854 proto_method_implemented(proto->accept),
2855 proto_method_implemented(proto->ioctl),
2856 proto_method_implemented(proto->init),
2857 proto_method_implemented(proto->destroy),
2858 proto_method_implemented(proto->shutdown),
2859 proto_method_implemented(proto->setsockopt),
2860 proto_method_implemented(proto->getsockopt),
2861 proto_method_implemented(proto->sendmsg),
2862 proto_method_implemented(proto->recvmsg),
2863 proto_method_implemented(proto->sendpage),
2864 proto_method_implemented(proto->bind),
2865 proto_method_implemented(proto->backlog_rcv),
2866 proto_method_implemented(proto->hash),
2867 proto_method_implemented(proto->unhash),
2868 proto_method_implemented(proto->get_port),
2869 proto_method_implemented(proto->enter_memory_pressure));
2870 }
2871
2872 static int proto_seq_show(struct seq_file *seq, void *v)
2873 {
2874 if (v == &proto_list)
2875 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2876 "protocol",
2877 "size",
2878 "sockets",
2879 "memory",
2880 "press",
2881 "maxhdr",
2882 "slab",
2883 "module",
2884 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2885 else
2886 proto_seq_printf(seq, list_entry(v, struct proto, node));
2887 return 0;
2888 }
2889
2890 static const struct seq_operations proto_seq_ops = {
2891 .start = proto_seq_start,
2892 .next = proto_seq_next,
2893 .stop = proto_seq_stop,
2894 .show = proto_seq_show,
2895 };
2896
2897 static int proto_seq_open(struct inode *inode, struct file *file)
2898 {
2899 return seq_open_net(inode, file, &proto_seq_ops,
2900 sizeof(struct seq_net_private));
2901 }
2902
2903 static const struct file_operations proto_seq_fops = {
2904 .owner = THIS_MODULE,
2905 .open = proto_seq_open,
2906 .read = seq_read,
2907 .llseek = seq_lseek,
2908 .release = seq_release_net,
2909 };
2910
2911 static __net_init int proto_init_net(struct net *net)
2912 {
2913 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2914 return -ENOMEM;
2915
2916 return 0;
2917 }
2918
2919 static __net_exit void proto_exit_net(struct net *net)
2920 {
2921 remove_proc_entry("protocols", net->proc_net);
2922 }
2923
2924
2925 static __net_initdata struct pernet_operations proto_net_ops = {
2926 .init = proto_init_net,
2927 .exit = proto_exit_net,
2928 };
2929
2930 static int __init proto_init(void)
2931 {
2932 return register_pernet_subsys(&proto_net_ops);
2933 }
2934
2935 subsys_initcall(proto_init);
2936
2937 #endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.08638 seconds and 6 git commands to generate.