?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Input: tsc2005 - set up bus type in input device
[linux-2.6/kvm.git] / net / core / sock.c
blobe5af8d5d5b505d79e3102a0aa879ba3d3e17f82b
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 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
96 #include <linux/in.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
117
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
127 #include <net/cls_cgroup.h>
128
129 #include <linux/filter.h>
130
131 #ifdef CONFIG_INET
132 #include <net/tcp.h>
133 #endif
134
135 /*
136 * Each address family might have different locking rules, so we have
137 * one slock key per address family:
138 */
139 static struct lock_class_key af_family_keys[AF_MAX];
140 static struct lock_class_key af_family_slock_keys[AF_MAX];
141
142 /*
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
145 * locks is fast):
146 */
147 static const char *const af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
160 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" ,
161 "sk_lock-AF_MAX"
162 };
163 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
164 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
165 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
166 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
167 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
168 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
169 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
170 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
171 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
172 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
173 "slock-27" , "slock-28" , "slock-AF_CAN" ,
174 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
175 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
176 "slock-AF_IEEE802154", "slock-AF_CAIF" ,
177 "slock-AF_MAX"
178 };
179 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
180 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
181 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
182 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
183 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
184 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
185 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
186 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
187 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
188 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
189 "clock-27" , "clock-28" , "clock-AF_CAN" ,
190 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
191 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
192 "clock-AF_IEEE802154", "clock-AF_CAIF" ,
193 "clock-AF_MAX"
194 };
195
196 /*
197 * sk_callback_lock locking rules are per-address-family,
198 * so split the lock classes by using a per-AF key:
199 */
200 static struct lock_class_key af_callback_keys[AF_MAX];
201
202 /* Take into consideration the size of the struct sk_buff overhead in the
203 * determination of these values, since that is non-constant across
204 * platforms. This makes socket queueing behavior and performance
205 * not depend upon such differences.
206 */
207 #define _SK_MEM_PACKETS 256
208 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
209 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
210 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
211
212 /* Run time adjustable parameters. */
213 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
215 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
216 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
217
218 /* Maximal space eaten by iovec or ancilliary data plus some space */
219 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
220 EXPORT_SYMBOL(sysctl_optmem_max);
221
222 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
223 int net_cls_subsys_id = -1;
224 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
225 #endif
226
227 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
228 {
229 struct timeval tv;
230
231 if (optlen < sizeof(tv))
232 return -EINVAL;
233 if (copy_from_user(&tv, optval, sizeof(tv)))
234 return -EFAULT;
235 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
236 return -EDOM;
237
238 if (tv.tv_sec < 0) {
239 static int warned __read_mostly;
240
241 *timeo_p = 0;
242 if (warned < 10 && net_ratelimit()) {
243 warned++;
244 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
245 "tries to set negative timeout\n",
246 current->comm, task_pid_nr(current));
247 }
248 return 0;
249 }
250 *timeo_p = MAX_SCHEDULE_TIMEOUT;
251 if (tv.tv_sec == 0 && tv.tv_usec == 0)
252 return 0;
253 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
254 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
255 return 0;
256 }
257
258 static void sock_warn_obsolete_bsdism(const char *name)
259 {
260 static int warned;
261 static char warncomm[TASK_COMM_LEN];
262 if (strcmp(warncomm, current->comm) && warned < 5) {
263 strcpy(warncomm, current->comm);
264 printk(KERN_WARNING "process `%s' is using obsolete "
265 "%s SO_BSDCOMPAT\n", warncomm, name);
266 warned++;
267 }
268 }
269
270 static void sock_disable_timestamp(struct sock *sk, int flag)
271 {
272 if (sock_flag(sk, flag)) {
273 sock_reset_flag(sk, flag);
274 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
275 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
276 net_disable_timestamp();
277 }
278 }
279 }
280
281
282 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
283 {
284 int err;
285 int skb_len;
286 unsigned long flags;
287 struct sk_buff_head *list = &sk->sk_receive_queue;
288
289 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
290 number of warnings when compiling with -W --ANK
291 */
292 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
293 (unsigned)sk->sk_rcvbuf) {
294 atomic_inc(&sk->sk_drops);
295 return -ENOMEM;
296 }
297
298 err = sk_filter(sk, skb);
299 if (err)
300 return err;
301
302 if (!sk_rmem_schedule(sk, skb->truesize)) {
303 atomic_inc(&sk->sk_drops);
304 return -ENOBUFS;
305 }
306
307 skb->dev = NULL;
308 skb_set_owner_r(skb, sk);
309
310 /* Cache the SKB length before we tack it onto the receive
311 * queue. Once it is added it no longer belongs to us and
312 * may be freed by other threads of control pulling packets
313 * from the queue.
314 */
315 skb_len = skb->len;
316
317 /* we escape from rcu protected region, make sure we dont leak
318 * a norefcounted dst
319 */
320 skb_dst_force(skb);
321
322 spin_lock_irqsave(&list->lock, flags);
323 skb->dropcount = atomic_read(&sk->sk_drops);
324 __skb_queue_tail(list, skb);
325 spin_unlock_irqrestore(&list->lock, flags);
326
327 if (!sock_flag(sk, SOCK_DEAD))
328 sk->sk_data_ready(sk, skb_len);
329 return 0;
330 }
331 EXPORT_SYMBOL(sock_queue_rcv_skb);
332
333 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
334 {
335 int rc = NET_RX_SUCCESS;
336
337 if (sk_filter(sk, skb))
338 goto discard_and_relse;
339
340 skb->dev = NULL;
341
342 if (sk_rcvqueues_full(sk, skb)) {
343 atomic_inc(&sk->sk_drops);
344 goto discard_and_relse;
345 }
346 if (nested)
347 bh_lock_sock_nested(sk);
348 else
349 bh_lock_sock(sk);
350 if (!sock_owned_by_user(sk)) {
351 /*
352 * trylock + unlock semantics:
353 */
354 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
355
356 rc = sk_backlog_rcv(sk, skb);
357
358 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
359 } else if (sk_add_backlog(sk, skb)) {
360 bh_unlock_sock(sk);
361 atomic_inc(&sk->sk_drops);
362 goto discard_and_relse;
363 }
364
365 bh_unlock_sock(sk);
366 out:
367 sock_put(sk);
368 return rc;
369 discard_and_relse:
370 kfree_skb(skb);
371 goto out;
372 }
373 EXPORT_SYMBOL(sk_receive_skb);
374
375 void sk_reset_txq(struct sock *sk)
376 {
377 sk_tx_queue_clear(sk);
378 }
379 EXPORT_SYMBOL(sk_reset_txq);
380
381 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
382 {
383 struct dst_entry *dst = __sk_dst_get(sk);
384
385 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
386 sk_tx_queue_clear(sk);
387 rcu_assign_pointer(sk->sk_dst_cache, NULL);
388 dst_release(dst);
389 return NULL;
390 }
391
392 return dst;
393 }
394 EXPORT_SYMBOL(__sk_dst_check);
395
396 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
397 {
398 struct dst_entry *dst = sk_dst_get(sk);
399
400 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
401 sk_dst_reset(sk);
402 dst_release(dst);
403 return NULL;
404 }
405
406 return dst;
407 }
408 EXPORT_SYMBOL(sk_dst_check);
409
410 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
411 {
412 int ret = -ENOPROTOOPT;
413 #ifdef CONFIG_NETDEVICES
414 struct net *net = sock_net(sk);
415 char devname[IFNAMSIZ];
416 int index;
417
418 /* Sorry... */
419 ret = -EPERM;
420 if (!capable(CAP_NET_RAW))
421 goto out;
422
423 ret = -EINVAL;
424 if (optlen < 0)
425 goto out;
426
427 /* Bind this socket to a particular device like "eth0",
428 * as specified in the passed interface name. If the
429 * name is "" or the option length is zero the socket
430 * is not bound.
431 */
432 if (optlen > IFNAMSIZ - 1)
433 optlen = IFNAMSIZ - 1;
434 memset(devname, 0, sizeof(devname));
435
436 ret = -EFAULT;
437 if (copy_from_user(devname, optval, optlen))
438 goto out;
439
440 index = 0;
441 if (devname[0] != '\0') {
442 struct net_device *dev;
443
444 rcu_read_lock();
445 dev = dev_get_by_name_rcu(net, devname);
446 if (dev)
447 index = dev->ifindex;
448 rcu_read_unlock();
449 ret = -ENODEV;
450 if (!dev)
451 goto out;
452 }
453
454 lock_sock(sk);
455 sk->sk_bound_dev_if = index;
456 sk_dst_reset(sk);
457 release_sock(sk);
458
459 ret = 0;
460
461 out:
462 #endif
463
464 return ret;
465 }
466
467 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
468 {
469 if (valbool)
470 sock_set_flag(sk, bit);
471 else
472 sock_reset_flag(sk, bit);
473 }
474
475 /*
476 * This is meant for all protocols to use and covers goings on
477 * at the socket level. Everything here is generic.
478 */
479
480 int sock_setsockopt(struct socket *sock, int level, int optname,
481 char __user *optval, unsigned int optlen)
482 {
483 struct sock *sk = sock->sk;
484 int val;
485 int valbool;
486 struct linger ling;
487 int ret = 0;
488
489 /*
490 * Options without arguments
491 */
492
493 if (optname == SO_BINDTODEVICE)
494 return sock_bindtodevice(sk, optval, optlen);
495
496 if (optlen < sizeof(int))
497 return -EINVAL;
498
499 if (get_user(val, (int __user *)optval))
500 return -EFAULT;
501
502 valbool = val ? 1 : 0;
503
504 lock_sock(sk);
505
506 switch (optname) {
507 case SO_DEBUG:
508 if (val && !capable(CAP_NET_ADMIN))
509 ret = -EACCES;
510 else
511 sock_valbool_flag(sk, SOCK_DBG, valbool);
512 break;
513 case SO_REUSEADDR:
514 sk->sk_reuse = valbool;
515 break;
516 case SO_TYPE:
517 case SO_PROTOCOL:
518 case SO_DOMAIN:
519 case SO_ERROR:
520 ret = -ENOPROTOOPT;
521 break;
522 case SO_DONTROUTE:
523 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
524 break;
525 case SO_BROADCAST:
526 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
527 break;
528 case SO_SNDBUF:
529 /* Don't error on this BSD doesn't and if you think
530 about it this is right. Otherwise apps have to
531 play 'guess the biggest size' games. RCVBUF/SNDBUF
532 are treated in BSD as hints */
533
534 if (val > sysctl_wmem_max)
535 val = sysctl_wmem_max;
536 set_sndbuf:
537 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
538 if ((val * 2) < SOCK_MIN_SNDBUF)
539 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
540 else
541 sk->sk_sndbuf = val * 2;
542
543 /*
544 * Wake up sending tasks if we
545 * upped the value.
546 */
547 sk->sk_write_space(sk);
548 break;
549
550 case SO_SNDBUFFORCE:
551 if (!capable(CAP_NET_ADMIN)) {
552 ret = -EPERM;
553 break;
554 }
555 goto set_sndbuf;
556
557 case SO_RCVBUF:
558 /* Don't error on this BSD doesn't and if you think
559 about it this is right. Otherwise apps have to
560 play 'guess the biggest size' games. RCVBUF/SNDBUF
561 are treated in BSD as hints */
562
563 if (val > sysctl_rmem_max)
564 val = sysctl_rmem_max;
565 set_rcvbuf:
566 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
567 /*
568 * We double it on the way in to account for
569 * "struct sk_buff" etc. overhead. Applications
570 * assume that the SO_RCVBUF setting they make will
571 * allow that much actual data to be received on that
572 * socket.
573 *
574 * Applications are unaware that "struct sk_buff" and
575 * other overheads allocate from the receive buffer
576 * during socket buffer allocation.
577 *
578 * And after considering the possible alternatives,
579 * returning the value we actually used in getsockopt
580 * is the most desirable behavior.
581 */
582 if ((val * 2) < SOCK_MIN_RCVBUF)
583 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
584 else
585 sk->sk_rcvbuf = val * 2;
586 break;
587
588 case SO_RCVBUFFORCE:
589 if (!capable(CAP_NET_ADMIN)) {
590 ret = -EPERM;
591 break;
592 }
593 goto set_rcvbuf;
594
595 case SO_KEEPALIVE:
596 #ifdef CONFIG_INET
597 if (sk->sk_protocol == IPPROTO_TCP)
598 tcp_set_keepalive(sk, valbool);
599 #endif
600 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
601 break;
602
603 case SO_OOBINLINE:
604 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
605 break;
606
607 case SO_NO_CHECK:
608 sk->sk_no_check = valbool;
609 break;
610
611 case SO_PRIORITY:
612 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
613 sk->sk_priority = val;
614 else
615 ret = -EPERM;
616 break;
617
618 case SO_LINGER:
619 if (optlen < sizeof(ling)) {
620 ret = -EINVAL; /* 1003.1g */
621 break;
622 }
623 if (copy_from_user(&ling, optval, sizeof(ling))) {
624 ret = -EFAULT;
625 break;
626 }
627 if (!ling.l_onoff)
628 sock_reset_flag(sk, SOCK_LINGER);
629 else {
630 #if (BITS_PER_LONG == 32)
631 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
632 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
633 else
634 #endif
635 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
636 sock_set_flag(sk, SOCK_LINGER);
637 }
638 break;
639
640 case SO_BSDCOMPAT:
641 sock_warn_obsolete_bsdism("setsockopt");
642 break;
643
644 case SO_PASSCRED:
645 if (valbool)
646 set_bit(SOCK_PASSCRED, &sock->flags);
647 else
648 clear_bit(SOCK_PASSCRED, &sock->flags);
649 break;
650
651 case SO_TIMESTAMP:
652 case SO_TIMESTAMPNS:
653 if (valbool) {
654 if (optname == SO_TIMESTAMP)
655 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
656 else
657 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
658 sock_set_flag(sk, SOCK_RCVTSTAMP);
659 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
660 } else {
661 sock_reset_flag(sk, SOCK_RCVTSTAMP);
662 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
663 }
664 break;
665
666 case SO_TIMESTAMPING:
667 if (val & ~SOF_TIMESTAMPING_MASK) {
668 ret = -EINVAL;
669 break;
670 }
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
672 val & SOF_TIMESTAMPING_TX_HARDWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
674 val & SOF_TIMESTAMPING_TX_SOFTWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
676 val & SOF_TIMESTAMPING_RX_HARDWARE);
677 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
678 sock_enable_timestamp(sk,
679 SOCK_TIMESTAMPING_RX_SOFTWARE);
680 else
681 sock_disable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
684 val & SOF_TIMESTAMPING_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
686 val & SOF_TIMESTAMPING_SYS_HARDWARE);
687 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
688 val & SOF_TIMESTAMPING_RAW_HARDWARE);
689 break;
690
691 case SO_RCVLOWAT:
692 if (val < 0)
693 val = INT_MAX;
694 sk->sk_rcvlowat = val ? : 1;
695 break;
696
697 case SO_RCVTIMEO:
698 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
699 break;
700
701 case SO_SNDTIMEO:
702 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
703 break;
704
705 case SO_ATTACH_FILTER:
706 ret = -EINVAL;
707 if (optlen == sizeof(struct sock_fprog)) {
708 struct sock_fprog fprog;
709
710 ret = -EFAULT;
711 if (copy_from_user(&fprog, optval, sizeof(fprog)))
712 break;
713
714 ret = sk_attach_filter(&fprog, sk);
715 }
716 break;
717
718 case SO_DETACH_FILTER:
719 ret = sk_detach_filter(sk);
720 break;
721
722 case SO_PASSSEC:
723 if (valbool)
724 set_bit(SOCK_PASSSEC, &sock->flags);
725 else
726 clear_bit(SOCK_PASSSEC, &sock->flags);
727 break;
728 case SO_MARK:
729 if (!capable(CAP_NET_ADMIN))
730 ret = -EPERM;
731 else
732 sk->sk_mark = val;
733 break;
734
735 /* We implement the SO_SNDLOWAT etc to
736 not be settable (1003.1g 5.3) */
737 case SO_RXQ_OVFL:
738 if (valbool)
739 sock_set_flag(sk, SOCK_RXQ_OVFL);
740 else
741 sock_reset_flag(sk, SOCK_RXQ_OVFL);
742 break;
743 default:
744 ret = -ENOPROTOOPT;
745 break;
746 }
747 release_sock(sk);
748 return ret;
749 }
750 EXPORT_SYMBOL(sock_setsockopt);
751
752
753 void cred_to_ucred(struct pid *pid, const struct cred *cred,
754 struct ucred *ucred)
755 {
756 ucred->pid = pid_vnr(pid);
757 ucred->uid = ucred->gid = -1;
758 if (cred) {
759 struct user_namespace *current_ns = current_user_ns();
760
761 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
762 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
763 }
764 }
765 EXPORT_SYMBOL_GPL(cred_to_ucred);
766
767 int sock_getsockopt(struct socket *sock, int level, int optname,
768 char __user *optval, int __user *optlen)
769 {
770 struct sock *sk = sock->sk;
771
772 union {
773 int val;
774 struct linger ling;
775 struct timeval tm;
776 } v;
777
778 int lv = sizeof(int);
779 int len;
780
781 if (get_user(len, optlen))
782 return -EFAULT;
783 if (len < 0)
784 return -EINVAL;
785
786 memset(&v, 0, sizeof(v));
787
788 switch (optname) {
789 case SO_DEBUG:
790 v.val = sock_flag(sk, SOCK_DBG);
791 break;
792
793 case SO_DONTROUTE:
794 v.val = sock_flag(sk, SOCK_LOCALROUTE);
795 break;
796
797 case SO_BROADCAST:
798 v.val = !!sock_flag(sk, SOCK_BROADCAST);
799 break;
800
801 case SO_SNDBUF:
802 v.val = sk->sk_sndbuf;
803 break;
804
805 case SO_RCVBUF:
806 v.val = sk->sk_rcvbuf;
807 break;
808
809 case SO_REUSEADDR:
810 v.val = sk->sk_reuse;
811 break;
812
813 case SO_KEEPALIVE:
814 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
815 break;
816
817 case SO_TYPE:
818 v.val = sk->sk_type;
819 break;
820
821 case SO_PROTOCOL:
822 v.val = sk->sk_protocol;
823 break;
824
825 case SO_DOMAIN:
826 v.val = sk->sk_family;
827 break;
828
829 case SO_ERROR:
830 v.val = -sock_error(sk);
831 if (v.val == 0)
832 v.val = xchg(&sk->sk_err_soft, 0);
833 break;
834
835 case SO_OOBINLINE:
836 v.val = !!sock_flag(sk, SOCK_URGINLINE);
837 break;
838
839 case SO_NO_CHECK:
840 v.val = sk->sk_no_check;
841 break;
842
843 case SO_PRIORITY:
844 v.val = sk->sk_priority;
845 break;
846
847 case SO_LINGER:
848 lv = sizeof(v.ling);
849 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
850 v.ling.l_linger = sk->sk_lingertime / HZ;
851 break;
852
853 case SO_BSDCOMPAT:
854 sock_warn_obsolete_bsdism("getsockopt");
855 break;
856
857 case SO_TIMESTAMP:
858 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
859 !sock_flag(sk, SOCK_RCVTSTAMPNS);
860 break;
861
862 case SO_TIMESTAMPNS:
863 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
864 break;
865
866 case SO_TIMESTAMPING:
867 v.val = 0;
868 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
869 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
870 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
871 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
872 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
873 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
874 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
875 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
876 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
877 v.val |= SOF_TIMESTAMPING_SOFTWARE;
878 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
879 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
882 break;
883
884 case SO_RCVTIMEO:
885 lv = sizeof(struct timeval);
886 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
887 v.tm.tv_sec = 0;
888 v.tm.tv_usec = 0;
889 } else {
890 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
891 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892 }
893 break;
894
895 case SO_SNDTIMEO:
896 lv = sizeof(struct timeval);
897 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
898 v.tm.tv_sec = 0;
899 v.tm.tv_usec = 0;
900 } else {
901 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
902 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903 }
904 break;
905
906 case SO_RCVLOWAT:
907 v.val = sk->sk_rcvlowat;
908 break;
909
910 case SO_SNDLOWAT:
911 v.val = 1;
912 break;
913
914 case SO_PASSCRED:
915 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916 break;
917
918 case SO_PEERCRED:
919 {
920 struct ucred peercred;
921 if (len > sizeof(peercred))
922 len = sizeof(peercred);
923 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
924 if (copy_to_user(optval, &peercred, len))
925 return -EFAULT;
926 goto lenout;
927 }
928
929 case SO_PEERNAME:
930 {
931 char address[128];
932
933 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
934 return -ENOTCONN;
935 if (lv < len)
936 return -EINVAL;
937 if (copy_to_user(optval, address, len))
938 return -EFAULT;
939 goto lenout;
940 }
941
942 /* Dubious BSD thing... Probably nobody even uses it, but
943 * the UNIX standard wants it for whatever reason... -DaveM
944 */
945 case SO_ACCEPTCONN:
946 v.val = sk->sk_state == TCP_LISTEN;
947 break;
948
949 case SO_PASSSEC:
950 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
951 break;
952
953 case SO_PEERSEC:
954 return security_socket_getpeersec_stream(sock, optval, optlen, len);
955
956 case SO_MARK:
957 v.val = sk->sk_mark;
958 break;
959
960 case SO_RXQ_OVFL:
961 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
962 break;
963
964 default:
965 return -ENOPROTOOPT;
966 }
967
968 if (len > lv)
969 len = lv;
970 if (copy_to_user(optval, &v, len))
971 return -EFAULT;
972 lenout:
973 if (put_user(len, optlen))
974 return -EFAULT;
975 return 0;
976 }
977
978 /*
979 * Initialize an sk_lock.
980 *
981 * (We also register the sk_lock with the lock validator.)
982 */
983 static inline void sock_lock_init(struct sock *sk)
984 {
985 sock_lock_init_class_and_name(sk,
986 af_family_slock_key_strings[sk->sk_family],
987 af_family_slock_keys + sk->sk_family,
988 af_family_key_strings[sk->sk_family],
989 af_family_keys + sk->sk_family);
990 }
991
992 /*
993 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
994 * even temporarly, because of RCU lookups. sk_node should also be left as is.
995 */
996 static void sock_copy(struct sock *nsk, const struct sock *osk)
997 {
998 #ifdef CONFIG_SECURITY_NETWORK
999 void *sptr = nsk->sk_security;
1000 #endif
1001 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
1002 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
1003 sizeof(osk->sk_tx_queue_mapping));
1004 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1005 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
1006 #ifdef CONFIG_SECURITY_NETWORK
1007 nsk->sk_security = sptr;
1008 security_sk_clone(osk, nsk);
1009 #endif
1010 }
1011
1012 /*
1013 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1014 * un-modified. Special care is taken when initializing object to zero.
1015 */
1016 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1017 {
1018 if (offsetof(struct sock, sk_node.next) != 0)
1019 memset(sk, 0, offsetof(struct sock, sk_node.next));
1020 memset(&sk->sk_node.pprev, 0,
1021 size - offsetof(struct sock, sk_node.pprev));
1022 }
1023
1024 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1025 {
1026 unsigned long nulls1, nulls2;
1027
1028 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1029 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1030 if (nulls1 > nulls2)
1031 swap(nulls1, nulls2);
1032
1033 if (nulls1 != 0)
1034 memset((char *)sk, 0, nulls1);
1035 memset((char *)sk + nulls1 + sizeof(void *), 0,
1036 nulls2 - nulls1 - sizeof(void *));
1037 memset((char *)sk + nulls2 + sizeof(void *), 0,
1038 size - nulls2 - sizeof(void *));
1039 }
1040 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1041
1042 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1043 int family)
1044 {
1045 struct sock *sk;
1046 struct kmem_cache *slab;
1047
1048 slab = prot->slab;
1049 if (slab != NULL) {
1050 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1051 if (!sk)
1052 return sk;
1053 if (priority & __GFP_ZERO) {
1054 if (prot->clear_sk)
1055 prot->clear_sk(sk, prot->obj_size);
1056 else
1057 sk_prot_clear_nulls(sk, prot->obj_size);
1058 }
1059 } else
1060 sk = kmalloc(prot->obj_size, priority);
1061
1062 if (sk != NULL) {
1063 kmemcheck_annotate_bitfield(sk, flags);
1064
1065 if (security_sk_alloc(sk, family, priority))
1066 goto out_free;
1067
1068 if (!try_module_get(prot->owner))
1069 goto out_free_sec;
1070 sk_tx_queue_clear(sk);
1071 }
1072
1073 return sk;
1074
1075 out_free_sec:
1076 security_sk_free(sk);
1077 out_free:
1078 if (slab != NULL)
1079 kmem_cache_free(slab, sk);
1080 else
1081 kfree(sk);
1082 return NULL;
1083 }
1084
1085 static void sk_prot_free(struct proto *prot, struct sock *sk)
1086 {
1087 struct kmem_cache *slab;
1088 struct module *owner;
1089
1090 owner = prot->owner;
1091 slab = prot->slab;
1092
1093 security_sk_free(sk);
1094 if (slab != NULL)
1095 kmem_cache_free(slab, sk);
1096 else
1097 kfree(sk);
1098 module_put(owner);
1099 }
1100
1101 #ifdef CONFIG_CGROUPS
1102 void sock_update_classid(struct sock *sk)
1103 {
1104 u32 classid;
1105
1106 rcu_read_lock(); /* doing current task, which cannot vanish. */
1107 classid = task_cls_classid(current);
1108 rcu_read_unlock();
1109 if (classid && classid != sk->sk_classid)
1110 sk->sk_classid = classid;
1111 }
1112 EXPORT_SYMBOL(sock_update_classid);
1113 #endif
1114
1115 /**
1116 * sk_alloc - All socket objects are allocated here
1117 * @net: the applicable net namespace
1118 * @family: protocol family
1119 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1120 * @prot: struct proto associated with this new sock instance
1121 */
1122 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1123 struct proto *prot)
1124 {
1125 struct sock *sk;
1126
1127 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1128 if (sk) {
1129 sk->sk_family = family;
1130 /*
1131 * See comment in struct sock definition to understand
1132 * why we need sk_prot_creator -acme
1133 */
1134 sk->sk_prot = sk->sk_prot_creator = prot;
1135 sock_lock_init(sk);
1136 sock_net_set(sk, get_net(net));
1137 atomic_set(&sk->sk_wmem_alloc, 1);
1138
1139 sock_update_classid(sk);
1140 }
1141
1142 return sk;
1143 }
1144 EXPORT_SYMBOL(sk_alloc);
1145
1146 static void __sk_free(struct sock *sk)
1147 {
1148 struct sk_filter *filter;
1149
1150 if (sk->sk_destruct)
1151 sk->sk_destruct(sk);
1152
1153 filter = rcu_dereference_check(sk->sk_filter,
1154 atomic_read(&sk->sk_wmem_alloc) == 0);
1155 if (filter) {
1156 sk_filter_uncharge(sk, filter);
1157 rcu_assign_pointer(sk->sk_filter, NULL);
1158 }
1159
1160 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1161 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1162
1163 if (atomic_read(&sk->sk_omem_alloc))
1164 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1165 __func__, atomic_read(&sk->sk_omem_alloc));
1166
1167 if (sk->sk_peer_cred)
1168 put_cred(sk->sk_peer_cred);
1169 put_pid(sk->sk_peer_pid);
1170 put_net(sock_net(sk));
1171 sk_prot_free(sk->sk_prot_creator, sk);
1172 }
1173
1174 void sk_free(struct sock *sk)
1175 {
1176 /*
1177 * We substract one from sk_wmem_alloc and can know if
1178 * some packets are still in some tx queue.
1179 * If not null, sock_wfree() will call __sk_free(sk) later
1180 */
1181 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1182 __sk_free(sk);
1183 }
1184 EXPORT_SYMBOL(sk_free);
1185
1186 /*
1187 * Last sock_put should drop referrence to sk->sk_net. It has already
1188 * been dropped in sk_change_net. Taking referrence to stopping namespace
1189 * is not an option.
1190 * Take referrence to a socket to remove it from hash _alive_ and after that
1191 * destroy it in the context of init_net.
1192 */
1193 void sk_release_kernel(struct sock *sk)
1194 {
1195 if (sk == NULL || sk->sk_socket == NULL)
1196 return;
1197
1198 sock_hold(sk);
1199 sock_release(sk->sk_socket);
1200 release_net(sock_net(sk));
1201 sock_net_set(sk, get_net(&init_net));
1202 sock_put(sk);
1203 }
1204 EXPORT_SYMBOL(sk_release_kernel);
1205
1206 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1207 {
1208 struct sock *newsk;
1209
1210 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1211 if (newsk != NULL) {
1212 struct sk_filter *filter;
1213
1214 sock_copy(newsk, sk);
1215
1216 /* SANITY */
1217 get_net(sock_net(newsk));
1218 sk_node_init(&newsk->sk_node);
1219 sock_lock_init(newsk);
1220 bh_lock_sock(newsk);
1221 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1222 newsk->sk_backlog.len = 0;
1223
1224 atomic_set(&newsk->sk_rmem_alloc, 0);
1225 /*
1226 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1227 */
1228 atomic_set(&newsk->sk_wmem_alloc, 1);
1229 atomic_set(&newsk->sk_omem_alloc, 0);
1230 skb_queue_head_init(&newsk->sk_receive_queue);
1231 skb_queue_head_init(&newsk->sk_write_queue);
1232 #ifdef CONFIG_NET_DMA
1233 skb_queue_head_init(&newsk->sk_async_wait_queue);
1234 #endif
1235
1236 spin_lock_init(&newsk->sk_dst_lock);
1237 rwlock_init(&newsk->sk_callback_lock);
1238 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1239 af_callback_keys + newsk->sk_family,
1240 af_family_clock_key_strings[newsk->sk_family]);
1241
1242 newsk->sk_dst_cache = NULL;
1243 newsk->sk_wmem_queued = 0;
1244 newsk->sk_forward_alloc = 0;
1245 newsk->sk_send_head = NULL;
1246 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1247
1248 sock_reset_flag(newsk, SOCK_DONE);
1249 skb_queue_head_init(&newsk->sk_error_queue);
1250
1251 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1252 if (filter != NULL)
1253 sk_filter_charge(newsk, filter);
1254
1255 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1256 /* It is still raw copy of parent, so invalidate
1257 * destructor and make plain sk_free() */
1258 newsk->sk_destruct = NULL;
1259 sk_free(newsk);
1260 newsk = NULL;
1261 goto out;
1262 }
1263
1264 newsk->sk_err = 0;
1265 newsk->sk_priority = 0;
1266 /*
1267 * Before updating sk_refcnt, we must commit prior changes to memory
1268 * (Documentation/RCU/rculist_nulls.txt for details)
1269 */
1270 smp_wmb();
1271 atomic_set(&newsk->sk_refcnt, 2);
1272
1273 /*
1274 * Increment the counter in the same struct proto as the master
1275 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1276 * is the same as sk->sk_prot->socks, as this field was copied
1277 * with memcpy).
1278 *
1279 * This _changes_ the previous behaviour, where
1280 * tcp_create_openreq_child always was incrementing the
1281 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1282 * to be taken into account in all callers. -acme
1283 */
1284 sk_refcnt_debug_inc(newsk);
1285 sk_set_socket(newsk, NULL);
1286 newsk->sk_wq = NULL;
1287
1288 if (newsk->sk_prot->sockets_allocated)
1289 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1290
1291 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1292 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1293 net_enable_timestamp();
1294 }
1295 out:
1296 return newsk;
1297 }
1298 EXPORT_SYMBOL_GPL(sk_clone);
1299
1300 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1301 {
1302 __sk_dst_set(sk, dst);
1303 sk->sk_route_caps = dst->dev->features;
1304 if (sk->sk_route_caps & NETIF_F_GSO)
1305 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1306 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1307 if (sk_can_gso(sk)) {
1308 if (dst->header_len) {
1309 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1310 } else {
1311 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1312 sk->sk_gso_max_size = dst->dev->gso_max_size;
1313 }
1314 }
1315 }
1316 EXPORT_SYMBOL_GPL(sk_setup_caps);
1317
1318 void __init sk_init(void)
1319 {
1320 if (totalram_pages <= 4096) {
1321 sysctl_wmem_max = 32767;
1322 sysctl_rmem_max = 32767;
1323 sysctl_wmem_default = 32767;
1324 sysctl_rmem_default = 32767;
1325 } else if (totalram_pages >= 131072) {
1326 sysctl_wmem_max = 131071;
1327 sysctl_rmem_max = 131071;
1328 }
1329 }
1330
1331 /*
1332 * Simple resource managers for sockets.
1333 */
1334
1335
1336 /*
1337 * Write buffer destructor automatically called from kfree_skb.
1338 */
1339 void sock_wfree(struct sk_buff *skb)
1340 {
1341 struct sock *sk = skb->sk;
1342 unsigned int len = skb->truesize;
1343
1344 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1345 /*
1346 * Keep a reference on sk_wmem_alloc, this will be released
1347 * after sk_write_space() call
1348 */
1349 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1350 sk->sk_write_space(sk);
1351 len = 1;
1352 }
1353 /*
1354 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1355 * could not do because of in-flight packets
1356 */
1357 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1358 __sk_free(sk);
1359 }
1360 EXPORT_SYMBOL(sock_wfree);
1361
1362 /*
1363 * Read buffer destructor automatically called from kfree_skb.
1364 */
1365 void sock_rfree(struct sk_buff *skb)
1366 {
1367 struct sock *sk = skb->sk;
1368 unsigned int len = skb->truesize;
1369
1370 atomic_sub(len, &sk->sk_rmem_alloc);
1371 sk_mem_uncharge(sk, len);
1372 }
1373 EXPORT_SYMBOL(sock_rfree);
1374
1375
1376 int sock_i_uid(struct sock *sk)
1377 {
1378 int uid;
1379
1380 read_lock_bh(&sk->sk_callback_lock);
1381 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1382 read_unlock_bh(&sk->sk_callback_lock);
1383 return uid;
1384 }
1385 EXPORT_SYMBOL(sock_i_uid);
1386
1387 unsigned long sock_i_ino(struct sock *sk)
1388 {
1389 unsigned long ino;
1390
1391 read_lock_bh(&sk->sk_callback_lock);
1392 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1393 read_unlock_bh(&sk->sk_callback_lock);
1394 return ino;
1395 }
1396 EXPORT_SYMBOL(sock_i_ino);
1397
1398 /*
1399 * Allocate a skb from the socket's send buffer.
1400 */
1401 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1402 gfp_t priority)
1403 {
1404 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1405 struct sk_buff *skb = alloc_skb(size, priority);
1406 if (skb) {
1407 skb_set_owner_w(skb, sk);
1408 return skb;
1409 }
1410 }
1411 return NULL;
1412 }
1413 EXPORT_SYMBOL(sock_wmalloc);
1414
1415 /*
1416 * Allocate a skb from the socket's receive buffer.
1417 */
1418 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1419 gfp_t priority)
1420 {
1421 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1422 struct sk_buff *skb = alloc_skb(size, priority);
1423 if (skb) {
1424 skb_set_owner_r(skb, sk);
1425 return skb;
1426 }
1427 }
1428 return NULL;
1429 }
1430
1431 /*
1432 * Allocate a memory block from the socket's option memory buffer.
1433 */
1434 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1435 {
1436 if ((unsigned)size <= sysctl_optmem_max &&
1437 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1438 void *mem;
1439 /* First do the add, to avoid the race if kmalloc
1440 * might sleep.
1441 */
1442 atomic_add(size, &sk->sk_omem_alloc);
1443 mem = kmalloc(size, priority);
1444 if (mem)
1445 return mem;
1446 atomic_sub(size, &sk->sk_omem_alloc);
1447 }
1448 return NULL;
1449 }
1450 EXPORT_SYMBOL(sock_kmalloc);
1451
1452 /*
1453 * Free an option memory block.
1454 */
1455 void sock_kfree_s(struct sock *sk, void *mem, int size)
1456 {
1457 kfree(mem);
1458 atomic_sub(size, &sk->sk_omem_alloc);
1459 }
1460 EXPORT_SYMBOL(sock_kfree_s);
1461
1462 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1463 I think, these locks should be removed for datagram sockets.
1464 */
1465 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1466 {
1467 DEFINE_WAIT(wait);
1468
1469 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1470 for (;;) {
1471 if (!timeo)
1472 break;
1473 if (signal_pending(current))
1474 break;
1475 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1476 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1477 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1478 break;
1479 if (sk->sk_shutdown & SEND_SHUTDOWN)
1480 break;
1481 if (sk->sk_err)
1482 break;
1483 timeo = schedule_timeout(timeo);
1484 }
1485 finish_wait(sk_sleep(sk), &wait);
1486 return timeo;
1487 }
1488
1489
1490 /*
1491 * Generic send/receive buffer handlers
1492 */
1493
1494 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1495 unsigned long data_len, int noblock,
1496 int *errcode)
1497 {
1498 struct sk_buff *skb;
1499 gfp_t gfp_mask;
1500 long timeo;
1501 int err;
1502
1503 gfp_mask = sk->sk_allocation;
1504 if (gfp_mask & __GFP_WAIT)
1505 gfp_mask |= __GFP_REPEAT;
1506
1507 timeo = sock_sndtimeo(sk, noblock);
1508 while (1) {
1509 err = sock_error(sk);
1510 if (err != 0)
1511 goto failure;
1512
1513 err = -EPIPE;
1514 if (sk->sk_shutdown & SEND_SHUTDOWN)
1515 goto failure;
1516
1517 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1518 skb = alloc_skb(header_len, gfp_mask);
1519 if (skb) {
1520 int npages;
1521 int i;
1522
1523 /* No pages, we're done... */
1524 if (!data_len)
1525 break;
1526
1527 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1528 skb->truesize += data_len;
1529 skb_shinfo(skb)->nr_frags = npages;
1530 for (i = 0; i < npages; i++) {
1531 struct page *page;
1532 skb_frag_t *frag;
1533
1534 page = alloc_pages(sk->sk_allocation, 0);
1535 if (!page) {
1536 err = -ENOBUFS;
1537 skb_shinfo(skb)->nr_frags = i;
1538 kfree_skb(skb);
1539 goto failure;
1540 }
1541
1542 frag = &skb_shinfo(skb)->frags[i];
1543 frag->page = page;
1544 frag->page_offset = 0;
1545 frag->size = (data_len >= PAGE_SIZE ?
1546 PAGE_SIZE :
1547 data_len);
1548 data_len -= PAGE_SIZE;
1549 }
1550
1551 /* Full success... */
1552 break;
1553 }
1554 err = -ENOBUFS;
1555 goto failure;
1556 }
1557 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1558 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1559 err = -EAGAIN;
1560 if (!timeo)
1561 goto failure;
1562 if (signal_pending(current))
1563 goto interrupted;
1564 timeo = sock_wait_for_wmem(sk, timeo);
1565 }
1566
1567 skb_set_owner_w(skb, sk);
1568 return skb;
1569
1570 interrupted:
1571 err = sock_intr_errno(timeo);
1572 failure:
1573 *errcode = err;
1574 return NULL;
1575 }
1576 EXPORT_SYMBOL(sock_alloc_send_pskb);
1577
1578 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1579 int noblock, int *errcode)
1580 {
1581 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1582 }
1583 EXPORT_SYMBOL(sock_alloc_send_skb);
1584
1585 static void __lock_sock(struct sock *sk)
1586 __releases(&sk->sk_lock.slock)
1587 __acquires(&sk->sk_lock.slock)
1588 {
1589 DEFINE_WAIT(wait);
1590
1591 for (;;) {
1592 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1593 TASK_UNINTERRUPTIBLE);
1594 spin_unlock_bh(&sk->sk_lock.slock);
1595 schedule();
1596 spin_lock_bh(&sk->sk_lock.slock);
1597 if (!sock_owned_by_user(sk))
1598 break;
1599 }
1600 finish_wait(&sk->sk_lock.wq, &wait);
1601 }
1602
1603 static void __release_sock(struct sock *sk)
1604 __releases(&sk->sk_lock.slock)
1605 __acquires(&sk->sk_lock.slock)
1606 {
1607 struct sk_buff *skb = sk->sk_backlog.head;
1608
1609 do {
1610 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1611 bh_unlock_sock(sk);
1612
1613 do {
1614 struct sk_buff *next = skb->next;
1615
1616 WARN_ON_ONCE(skb_dst_is_noref(skb));
1617 skb->next = NULL;
1618 sk_backlog_rcv(sk, skb);
1619
1620 /*
1621 * We are in process context here with softirqs
1622 * disabled, use cond_resched_softirq() to preempt.
1623 * This is safe to do because we've taken the backlog
1624 * queue private:
1625 */
1626 cond_resched_softirq();
1627
1628 skb = next;
1629 } while (skb != NULL);
1630
1631 bh_lock_sock(sk);
1632 } while ((skb = sk->sk_backlog.head) != NULL);
1633
1634 /*
1635 * Doing the zeroing here guarantee we can not loop forever
1636 * while a wild producer attempts to flood us.
1637 */
1638 sk->sk_backlog.len = 0;
1639 }
1640
1641 /**
1642 * sk_wait_data - wait for data to arrive at sk_receive_queue
1643 * @sk: sock to wait on
1644 * @timeo: for how long
1645 *
1646 * Now socket state including sk->sk_err is changed only under lock,
1647 * hence we may omit checks after joining wait queue.
1648 * We check receive queue before schedule() only as optimization;
1649 * it is very likely that release_sock() added new data.
1650 */
1651 int sk_wait_data(struct sock *sk, long *timeo)
1652 {
1653 int rc;
1654 DEFINE_WAIT(wait);
1655
1656 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1657 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1658 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1659 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1660 finish_wait(sk_sleep(sk), &wait);
1661 return rc;
1662 }
1663 EXPORT_SYMBOL(sk_wait_data);
1664
1665 /**
1666 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1667 * @sk: socket
1668 * @size: memory size to allocate
1669 * @kind: allocation type
1670 *
1671 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1672 * rmem allocation. This function assumes that protocols which have
1673 * memory_pressure use sk_wmem_queued as write buffer accounting.
1674 */
1675 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1676 {
1677 struct proto *prot = sk->sk_prot;
1678 int amt = sk_mem_pages(size);
1679 long allocated;
1680
1681 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1682 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1683
1684 /* Under limit. */
1685 if (allocated <= prot->sysctl_mem[0]) {
1686 if (prot->memory_pressure && *prot->memory_pressure)
1687 *prot->memory_pressure = 0;
1688 return 1;
1689 }
1690
1691 /* Under pressure. */
1692 if (allocated > prot->sysctl_mem[1])
1693 if (prot->enter_memory_pressure)
1694 prot->enter_memory_pressure(sk);
1695
1696 /* Over hard limit. */
1697 if (allocated > prot->sysctl_mem[2])
1698 goto suppress_allocation;
1699
1700 /* guarantee minimum buffer size under pressure */
1701 if (kind == SK_MEM_RECV) {
1702 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1703 return 1;
1704 } else { /* SK_MEM_SEND */
1705 if (sk->sk_type == SOCK_STREAM) {
1706 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1707 return 1;
1708 } else if (atomic_read(&sk->sk_wmem_alloc) <
1709 prot->sysctl_wmem[0])
1710 return 1;
1711 }
1712
1713 if (prot->memory_pressure) {
1714 int alloc;
1715
1716 if (!*prot->memory_pressure)
1717 return 1;
1718 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1719 if (prot->sysctl_mem[2] > alloc *
1720 sk_mem_pages(sk->sk_wmem_queued +
1721 atomic_read(&sk->sk_rmem_alloc) +
1722 sk->sk_forward_alloc))
1723 return 1;
1724 }
1725
1726 suppress_allocation:
1727
1728 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1729 sk_stream_moderate_sndbuf(sk);
1730
1731 /* Fail only if socket is _under_ its sndbuf.
1732 * In this case we cannot block, so that we have to fail.
1733 */
1734 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1735 return 1;
1736 }
1737
1738 /* Alas. Undo changes. */
1739 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1740 atomic_long_sub(amt, prot->memory_allocated);
1741 return 0;
1742 }
1743 EXPORT_SYMBOL(__sk_mem_schedule);
1744
1745 /**
1746 * __sk_reclaim - reclaim memory_allocated
1747 * @sk: socket
1748 */
1749 void __sk_mem_reclaim(struct sock *sk)
1750 {
1751 struct proto *prot = sk->sk_prot;
1752
1753 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1754 prot->memory_allocated);
1755 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1756
1757 if (prot->memory_pressure && *prot->memory_pressure &&
1758 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1759 *prot->memory_pressure = 0;
1760 }
1761 EXPORT_SYMBOL(__sk_mem_reclaim);
1762
1763
1764 /*
1765 * Set of default routines for initialising struct proto_ops when
1766 * the protocol does not support a particular function. In certain
1767 * cases where it makes no sense for a protocol to have a "do nothing"
1768 * function, some default processing is provided.
1769 */
1770
1771 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1772 {
1773 return -EOPNOTSUPP;
1774 }
1775 EXPORT_SYMBOL(sock_no_bind);
1776
1777 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1778 int len, int flags)
1779 {
1780 return -EOPNOTSUPP;
1781 }
1782 EXPORT_SYMBOL(sock_no_connect);
1783
1784 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1785 {
1786 return -EOPNOTSUPP;
1787 }
1788 EXPORT_SYMBOL(sock_no_socketpair);
1789
1790 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1791 {
1792 return -EOPNOTSUPP;
1793 }
1794 EXPORT_SYMBOL(sock_no_accept);
1795
1796 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1797 int *len, int peer)
1798 {
1799 return -EOPNOTSUPP;
1800 }
1801 EXPORT_SYMBOL(sock_no_getname);
1802
1803 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1804 {
1805 return 0;
1806 }
1807 EXPORT_SYMBOL(sock_no_poll);
1808
1809 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1810 {
1811 return -EOPNOTSUPP;
1812 }
1813 EXPORT_SYMBOL(sock_no_ioctl);
1814
1815 int sock_no_listen(struct socket *sock, int backlog)
1816 {
1817 return -EOPNOTSUPP;
1818 }
1819 EXPORT_SYMBOL(sock_no_listen);
1820
1821 int sock_no_shutdown(struct socket *sock, int how)
1822 {
1823 return -EOPNOTSUPP;
1824 }
1825 EXPORT_SYMBOL(sock_no_shutdown);
1826
1827 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1828 char __user *optval, unsigned int optlen)
1829 {
1830 return -EOPNOTSUPP;
1831 }
1832 EXPORT_SYMBOL(sock_no_setsockopt);
1833
1834 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1835 char __user *optval, int __user *optlen)
1836 {
1837 return -EOPNOTSUPP;
1838 }
1839 EXPORT_SYMBOL(sock_no_getsockopt);
1840
1841 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1842 size_t len)
1843 {
1844 return -EOPNOTSUPP;
1845 }
1846 EXPORT_SYMBOL(sock_no_sendmsg);
1847
1848 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1849 size_t len, int flags)
1850 {
1851 return -EOPNOTSUPP;
1852 }
1853 EXPORT_SYMBOL(sock_no_recvmsg);
1854
1855 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1856 {
1857 /* Mirror missing mmap method error code */
1858 return -ENODEV;
1859 }
1860 EXPORT_SYMBOL(sock_no_mmap);
1861
1862 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1863 {
1864 ssize_t res;
1865 struct msghdr msg = {.msg_flags = flags};
1866 struct kvec iov;
1867 char *kaddr = kmap(page);
1868 iov.iov_base = kaddr + offset;
1869 iov.iov_len = size;
1870 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1871 kunmap(page);
1872 return res;
1873 }
1874 EXPORT_SYMBOL(sock_no_sendpage);
1875
1876 /*
1877 * Default Socket Callbacks
1878 */
1879
1880 static void sock_def_wakeup(struct sock *sk)
1881 {
1882 struct socket_wq *wq;
1883
1884 rcu_read_lock();
1885 wq = rcu_dereference(sk->sk_wq);
1886 if (wq_has_sleeper(wq))
1887 wake_up_interruptible_all(&wq->wait);
1888 rcu_read_unlock();
1889 }
1890
1891 static void sock_def_error_report(struct sock *sk)
1892 {
1893 struct socket_wq *wq;
1894
1895 rcu_read_lock();
1896 wq = rcu_dereference(sk->sk_wq);
1897 if (wq_has_sleeper(wq))
1898 wake_up_interruptible_poll(&wq->wait, POLLERR);
1899 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1900 rcu_read_unlock();
1901 }
1902
1903 static void sock_def_readable(struct sock *sk, int len)
1904 {
1905 struct socket_wq *wq;
1906
1907 rcu_read_lock();
1908 wq = rcu_dereference(sk->sk_wq);
1909 if (wq_has_sleeper(wq))
1910 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1911 POLLRDNORM | POLLRDBAND);
1912 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1913 rcu_read_unlock();
1914 }
1915
1916 static void sock_def_write_space(struct sock *sk)
1917 {
1918 struct socket_wq *wq;
1919
1920 rcu_read_lock();
1921
1922 /* Do not wake up a writer until he can make "significant"
1923 * progress. --DaveM
1924 */
1925 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1926 wq = rcu_dereference(sk->sk_wq);
1927 if (wq_has_sleeper(wq))
1928 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1929 POLLWRNORM | POLLWRBAND);
1930
1931 /* Should agree with poll, otherwise some programs break */
1932 if (sock_writeable(sk))
1933 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1934 }
1935
1936 rcu_read_unlock();
1937 }
1938
1939 static void sock_def_destruct(struct sock *sk)
1940 {
1941 kfree(sk->sk_protinfo);
1942 }
1943
1944 void sk_send_sigurg(struct sock *sk)
1945 {
1946 if (sk->sk_socket && sk->sk_socket->file)
1947 if (send_sigurg(&sk->sk_socket->file->f_owner))
1948 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1949 }
1950 EXPORT_SYMBOL(sk_send_sigurg);
1951
1952 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1953 unsigned long expires)
1954 {
1955 if (!mod_timer(timer, expires))
1956 sock_hold(sk);
1957 }
1958 EXPORT_SYMBOL(sk_reset_timer);
1959
1960 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1961 {
1962 if (timer_pending(timer) && del_timer(timer))
1963 __sock_put(sk);
1964 }
1965 EXPORT_SYMBOL(sk_stop_timer);
1966
1967 void sock_init_data(struct socket *sock, struct sock *sk)
1968 {
1969 skb_queue_head_init(&sk->sk_receive_queue);
1970 skb_queue_head_init(&sk->sk_write_queue);
1971 skb_queue_head_init(&sk->sk_error_queue);
1972 #ifdef CONFIG_NET_DMA
1973 skb_queue_head_init(&sk->sk_async_wait_queue);
1974 #endif
1975
1976 sk->sk_send_head = NULL;
1977
1978 init_timer(&sk->sk_timer);
1979
1980 sk->sk_allocation = GFP_KERNEL;
1981 sk->sk_rcvbuf = sysctl_rmem_default;
1982 sk->sk_sndbuf = sysctl_wmem_default;
1983 sk->sk_state = TCP_CLOSE;
1984 sk_set_socket(sk, sock);
1985
1986 sock_set_flag(sk, SOCK_ZAPPED);
1987
1988 if (sock) {
1989 sk->sk_type = sock->type;
1990 sk->sk_wq = sock->wq;
1991 sock->sk = sk;
1992 } else
1993 sk->sk_wq = NULL;
1994
1995 spin_lock_init(&sk->sk_dst_lock);
1996 rwlock_init(&sk->sk_callback_lock);
1997 lockdep_set_class_and_name(&sk->sk_callback_lock,
1998 af_callback_keys + sk->sk_family,
1999 af_family_clock_key_strings[sk->sk_family]);
2000
2001 sk->sk_state_change = sock_def_wakeup;
2002 sk->sk_data_ready = sock_def_readable;
2003 sk->sk_write_space = sock_def_write_space;
2004 sk->sk_error_report = sock_def_error_report;
2005 sk->sk_destruct = sock_def_destruct;
2006
2007 sk->sk_sndmsg_page = NULL;
2008 sk->sk_sndmsg_off = 0;
2009
2010 sk->sk_peer_pid = NULL;
2011 sk->sk_peer_cred = NULL;
2012 sk->sk_write_pending = 0;
2013 sk->sk_rcvlowat = 1;
2014 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2015 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2016
2017 sk->sk_stamp = ktime_set(-1L, 0);
2018
2019 /*
2020 * Before updating sk_refcnt, we must commit prior changes to memory
2021 * (Documentation/RCU/rculist_nulls.txt for details)
2022 */
2023 smp_wmb();
2024 atomic_set(&sk->sk_refcnt, 1);
2025 atomic_set(&sk->sk_drops, 0);
2026 }
2027 EXPORT_SYMBOL(sock_init_data);
2028
2029 void lock_sock_nested(struct sock *sk, int subclass)
2030 {
2031 might_sleep();
2032 spin_lock_bh(&sk->sk_lock.slock);
2033 if (sk->sk_lock.owned)
2034 __lock_sock(sk);
2035 sk->sk_lock.owned = 1;
2036 spin_unlock(&sk->sk_lock.slock);
2037 /*
2038 * The sk_lock has mutex_lock() semantics here:
2039 */
2040 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2041 local_bh_enable();
2042 }
2043 EXPORT_SYMBOL(lock_sock_nested);
2044
2045 void release_sock(struct sock *sk)
2046 {
2047 /*
2048 * The sk_lock has mutex_unlock() semantics:
2049 */
2050 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2051
2052 spin_lock_bh(&sk->sk_lock.slock);
2053 if (sk->sk_backlog.tail)
2054 __release_sock(sk);
2055 sk->sk_lock.owned = 0;
2056 if (waitqueue_active(&sk->sk_lock.wq))
2057 wake_up(&sk->sk_lock.wq);
2058 spin_unlock_bh(&sk->sk_lock.slock);
2059 }
2060 EXPORT_SYMBOL(release_sock);
2061
2062 /**
2063 * lock_sock_fast - fast version of lock_sock
2064 * @sk: socket
2065 *
2066 * This version should be used for very small section, where process wont block
2067 * return false if fast path is taken
2068 * sk_lock.slock locked, owned = 0, BH disabled
2069 * return true if slow path is taken
2070 * sk_lock.slock unlocked, owned = 1, BH enabled
2071 */
2072 bool lock_sock_fast(struct sock *sk)
2073 {
2074 might_sleep();
2075 spin_lock_bh(&sk->sk_lock.slock);
2076
2077 if (!sk->sk_lock.owned)
2078 /*
2079 * Note : We must disable BH
2080 */
2081 return false;
2082
2083 __lock_sock(sk);
2084 sk->sk_lock.owned = 1;
2085 spin_unlock(&sk->sk_lock.slock);
2086 /*
2087 * The sk_lock has mutex_lock() semantics here:
2088 */
2089 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2090 local_bh_enable();
2091 return true;
2092 }
2093 EXPORT_SYMBOL(lock_sock_fast);
2094
2095 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2096 {
2097 struct timeval tv;
2098 if (!sock_flag(sk, SOCK_TIMESTAMP))
2099 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2100 tv = ktime_to_timeval(sk->sk_stamp);
2101 if (tv.tv_sec == -1)
2102 return -ENOENT;
2103 if (tv.tv_sec == 0) {
2104 sk->sk_stamp = ktime_get_real();
2105 tv = ktime_to_timeval(sk->sk_stamp);
2106 }
2107 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2108 }
2109 EXPORT_SYMBOL(sock_get_timestamp);
2110
2111 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2112 {
2113 struct timespec ts;
2114 if (!sock_flag(sk, SOCK_TIMESTAMP))
2115 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2116 ts = ktime_to_timespec(sk->sk_stamp);
2117 if (ts.tv_sec == -1)
2118 return -ENOENT;
2119 if (ts.tv_sec == 0) {
2120 sk->sk_stamp = ktime_get_real();
2121 ts = ktime_to_timespec(sk->sk_stamp);
2122 }
2123 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2124 }
2125 EXPORT_SYMBOL(sock_get_timestampns);
2126
2127 void sock_enable_timestamp(struct sock *sk, int flag)
2128 {
2129 if (!sock_flag(sk, flag)) {
2130 sock_set_flag(sk, flag);
2131 /*
2132 * we just set one of the two flags which require net
2133 * time stamping, but time stamping might have been on
2134 * already because of the other one
2135 */
2136 if (!sock_flag(sk,
2137 flag == SOCK_TIMESTAMP ?
2138 SOCK_TIMESTAMPING_RX_SOFTWARE :
2139 SOCK_TIMESTAMP))
2140 net_enable_timestamp();
2141 }
2142 }
2143
2144 /*
2145 * Get a socket option on an socket.
2146 *
2147 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2148 * asynchronous errors should be reported by getsockopt. We assume
2149 * this means if you specify SO_ERROR (otherwise whats the point of it).
2150 */
2151 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2152 char __user *optval, int __user *optlen)
2153 {
2154 struct sock *sk = sock->sk;
2155
2156 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2157 }
2158 EXPORT_SYMBOL(sock_common_getsockopt);
2159
2160 #ifdef CONFIG_COMPAT
2161 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2162 char __user *optval, int __user *optlen)
2163 {
2164 struct sock *sk = sock->sk;
2165
2166 if (sk->sk_prot->compat_getsockopt != NULL)
2167 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2168 optval, optlen);
2169 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2170 }
2171 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2172 #endif
2173
2174 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2175 struct msghdr *msg, size_t size, int flags)
2176 {
2177 struct sock *sk = sock->sk;
2178 int addr_len = 0;
2179 int err;
2180
2181 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2182 flags & ~MSG_DONTWAIT, &addr_len);
2183 if (err >= 0)
2184 msg->msg_namelen = addr_len;
2185 return err;
2186 }
2187 EXPORT_SYMBOL(sock_common_recvmsg);
2188
2189 /*
2190 * Set socket options on an inet socket.
2191 */
2192 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2193 char __user *optval, unsigned int optlen)
2194 {
2195 struct sock *sk = sock->sk;
2196
2197 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2198 }
2199 EXPORT_SYMBOL(sock_common_setsockopt);
2200
2201 #ifdef CONFIG_COMPAT
2202 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2203 char __user *optval, unsigned int optlen)
2204 {
2205 struct sock *sk = sock->sk;
2206
2207 if (sk->sk_prot->compat_setsockopt != NULL)
2208 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2209 optval, optlen);
2210 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2211 }
2212 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2213 #endif
2214
2215 void sk_common_release(struct sock *sk)
2216 {
2217 if (sk->sk_prot->destroy)
2218 sk->sk_prot->destroy(sk);
2219
2220 /*
2221 * Observation: when sock_common_release is called, processes have
2222 * no access to socket. But net still has.
2223 * Step one, detach it from networking:
2224 *
2225 * A. Remove from hash tables.
2226 */
2227
2228 sk->sk_prot->unhash(sk);
2229
2230 /*
2231 * In this point socket cannot receive new packets, but it is possible
2232 * that some packets are in flight because some CPU runs receiver and
2233 * did hash table lookup before we unhashed socket. They will achieve
2234 * receive queue and will be purged by socket destructor.
2235 *
2236 * Also we still have packets pending on receive queue and probably,
2237 * our own packets waiting in device queues. sock_destroy will drain
2238 * receive queue, but transmitted packets will delay socket destruction
2239 * until the last reference will be released.
2240 */
2241
2242 sock_orphan(sk);
2243
2244 xfrm_sk_free_policy(sk);
2245
2246 sk_refcnt_debug_release(sk);
2247 sock_put(sk);
2248 }
2249 EXPORT_SYMBOL(sk_common_release);
2250
2251 static DEFINE_RWLOCK(proto_list_lock);
2252 static LIST_HEAD(proto_list);
2253
2254 #ifdef CONFIG_PROC_FS
2255 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2256 struct prot_inuse {
2257 int val[PROTO_INUSE_NR];
2258 };
2259
2260 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2261
2262 #ifdef CONFIG_NET_NS
2263 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2264 {
2265 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2266 }
2267 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2268
2269 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2270 {
2271 int cpu, idx = prot->inuse_idx;
2272 int res = 0;
2273
2274 for_each_possible_cpu(cpu)
2275 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2276
2277 return res >= 0 ? res : 0;
2278 }
2279 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2280
2281 static int __net_init sock_inuse_init_net(struct net *net)
2282 {
2283 net->core.inuse = alloc_percpu(struct prot_inuse);
2284 return net->core.inuse ? 0 : -ENOMEM;
2285 }
2286
2287 static void __net_exit sock_inuse_exit_net(struct net *net)
2288 {
2289 free_percpu(net->core.inuse);
2290 }
2291
2292 static struct pernet_operations net_inuse_ops = {
2293 .init = sock_inuse_init_net,
2294 .exit = sock_inuse_exit_net,
2295 };
2296
2297 static __init int net_inuse_init(void)
2298 {
2299 if (register_pernet_subsys(&net_inuse_ops))
2300 panic("Cannot initialize net inuse counters");
2301
2302 return 0;
2303 }
2304
2305 core_initcall(net_inuse_init);
2306 #else
2307 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2308
2309 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2310 {
2311 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2312 }
2313 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2314
2315 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2316 {
2317 int cpu, idx = prot->inuse_idx;
2318 int res = 0;
2319
2320 for_each_possible_cpu(cpu)
2321 res += per_cpu(prot_inuse, cpu).val[idx];
2322
2323 return res >= 0 ? res : 0;
2324 }
2325 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2326 #endif
2327
2328 static void assign_proto_idx(struct proto *prot)
2329 {
2330 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2331
2332 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2333 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2334 return;
2335 }
2336
2337 set_bit(prot->inuse_idx, proto_inuse_idx);
2338 }
2339
2340 static void release_proto_idx(struct proto *prot)
2341 {
2342 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2343 clear_bit(prot->inuse_idx, proto_inuse_idx);
2344 }
2345 #else
2346 static inline void assign_proto_idx(struct proto *prot)
2347 {
2348 }
2349
2350 static inline void release_proto_idx(struct proto *prot)
2351 {
2352 }
2353 #endif
2354
2355 int proto_register(struct proto *prot, int alloc_slab)
2356 {
2357 if (alloc_slab) {
2358 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2359 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2360 NULL);
2361
2362 if (prot->slab == NULL) {
2363 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2364 prot->name);
2365 goto out;
2366 }
2367
2368 if (prot->rsk_prot != NULL) {
2369 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2370 if (prot->rsk_prot->slab_name == NULL)
2371 goto out_free_sock_slab;
2372
2373 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2374 prot->rsk_prot->obj_size, 0,
2375 SLAB_HWCACHE_ALIGN, NULL);
2376
2377 if (prot->rsk_prot->slab == NULL) {
2378 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2379 prot->name);
2380 goto out_free_request_sock_slab_name;
2381 }
2382 }
2383
2384 if (prot->twsk_prot != NULL) {
2385 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2386
2387 if (prot->twsk_prot->twsk_slab_name == NULL)
2388 goto out_free_request_sock_slab;
2389
2390 prot->twsk_prot->twsk_slab =
2391 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2392 prot->twsk_prot->twsk_obj_size,
2393 0,
2394 SLAB_HWCACHE_ALIGN |
2395 prot->slab_flags,
2396 NULL);
2397 if (prot->twsk_prot->twsk_slab == NULL)
2398 goto out_free_timewait_sock_slab_name;
2399 }
2400 }
2401
2402 write_lock(&proto_list_lock);
2403 list_add(&prot->node, &proto_list);
2404 assign_proto_idx(prot);
2405 write_unlock(&proto_list_lock);
2406 return 0;
2407
2408 out_free_timewait_sock_slab_name:
2409 kfree(prot->twsk_prot->twsk_slab_name);
2410 out_free_request_sock_slab:
2411 if (prot->rsk_prot && prot->rsk_prot->slab) {
2412 kmem_cache_destroy(prot->rsk_prot->slab);
2413 prot->rsk_prot->slab = NULL;
2414 }
2415 out_free_request_sock_slab_name:
2416 if (prot->rsk_prot)
2417 kfree(prot->rsk_prot->slab_name);
2418 out_free_sock_slab:
2419 kmem_cache_destroy(prot->slab);
2420 prot->slab = NULL;
2421 out:
2422 return -ENOBUFS;
2423 }
2424 EXPORT_SYMBOL(proto_register);
2425
2426 void proto_unregister(struct proto *prot)
2427 {
2428 write_lock(&proto_list_lock);
2429 release_proto_idx(prot);
2430 list_del(&prot->node);
2431 write_unlock(&proto_list_lock);
2432
2433 if (prot->slab != NULL) {
2434 kmem_cache_destroy(prot->slab);
2435 prot->slab = NULL;
2436 }
2437
2438 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2439 kmem_cache_destroy(prot->rsk_prot->slab);
2440 kfree(prot->rsk_prot->slab_name);
2441 prot->rsk_prot->slab = NULL;
2442 }
2443
2444 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2445 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2446 kfree(prot->twsk_prot->twsk_slab_name);
2447 prot->twsk_prot->twsk_slab = NULL;
2448 }
2449 }
2450 EXPORT_SYMBOL(proto_unregister);
2451
2452 #ifdef CONFIG_PROC_FS
2453 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2454 __acquires(proto_list_lock)
2455 {
2456 read_lock(&proto_list_lock);
2457 return seq_list_start_head(&proto_list, *pos);
2458 }
2459
2460 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2461 {
2462 return seq_list_next(v, &proto_list, pos);
2463 }
2464
2465 static void proto_seq_stop(struct seq_file *seq, void *v)
2466 __releases(proto_list_lock)
2467 {
2468 read_unlock(&proto_list_lock);
2469 }
2470
2471 static char proto_method_implemented(const void *method)
2472 {
2473 return method == NULL ? 'n' : 'y';
2474 }
2475
2476 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2477 {
2478 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2479 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2480 proto->name,
2481 proto->obj_size,
2482 sock_prot_inuse_get(seq_file_net(seq), proto),
2483 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2484 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2485 proto->max_header,
2486 proto->slab == NULL ? "no" : "yes",
2487 module_name(proto->owner),
2488 proto_method_implemented(proto->close),
2489 proto_method_implemented(proto->connect),
2490 proto_method_implemented(proto->disconnect),
2491 proto_method_implemented(proto->accept),
2492 proto_method_implemented(proto->ioctl),
2493 proto_method_implemented(proto->init),
2494 proto_method_implemented(proto->destroy),
2495 proto_method_implemented(proto->shutdown),
2496 proto_method_implemented(proto->setsockopt),
2497 proto_method_implemented(proto->getsockopt),
2498 proto_method_implemented(proto->sendmsg),
2499 proto_method_implemented(proto->recvmsg),
2500 proto_method_implemented(proto->sendpage),
2501 proto_method_implemented(proto->bind),
2502 proto_method_implemented(proto->backlog_rcv),
2503 proto_method_implemented(proto->hash),
2504 proto_method_implemented(proto->unhash),
2505 proto_method_implemented(proto->get_port),
2506 proto_method_implemented(proto->enter_memory_pressure));
2507 }
2508
2509 static int proto_seq_show(struct seq_file *seq, void *v)
2510 {
2511 if (v == &proto_list)
2512 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2513 "protocol",
2514 "size",
2515 "sockets",
2516 "memory",
2517 "press",
2518 "maxhdr",
2519 "slab",
2520 "module",
2521 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2522 else
2523 proto_seq_printf(seq, list_entry(v, struct proto, node));
2524 return 0;
2525 }
2526
2527 static const struct seq_operations proto_seq_ops = {
2528 .start = proto_seq_start,
2529 .next = proto_seq_next,
2530 .stop = proto_seq_stop,
2531 .show = proto_seq_show,
2532 };
2533
2534 static int proto_seq_open(struct inode *inode, struct file *file)
2535 {
2536 return seq_open_net(inode, file, &proto_seq_ops,
2537 sizeof(struct seq_net_private));
2538 }
2539
2540 static const struct file_operations proto_seq_fops = {
2541 .owner = THIS_MODULE,
2542 .open = proto_seq_open,
2543 .read = seq_read,
2544 .llseek = seq_lseek,
2545 .release = seq_release_net,
2546 };
2547
2548 static __net_init int proto_init_net(struct net *net)
2549 {
2550 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2551 return -ENOMEM;
2552
2553 return 0;
2554 }
2555
2556 static __net_exit void proto_exit_net(struct net *net)
2557 {
2558 proc_net_remove(net, "protocols");
2559 }
2560
2561
2562 static __net_initdata struct pernet_operations proto_net_ops = {
2563 .init = proto_init_net,
2564 .exit = proto_exit_net,
2565 };
2566
2567 static int __init proto_init(void)
2568 {
2569 return register_pernet_subsys(&proto_net_ops);
2570 }
2571
2572 subsys_initcall(proto_init);
2573
2574 #endif /* PROC_FS */
kernel-based virtual machine