]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/sock.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sock: add net to prot->enter_memory_pressure callback
[net-next-2.6.git] / net / core / sock.c
CommitLineData
1da177e4
LT		 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 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 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
4ec93edb 35 * code. The ACK stuff can wait and needs major
1da177e4
LT		 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
4fc268d2 92#include <linux/capability.h>
1da177e4
LT		 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>
1da177e4
LT		 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>
a1f8e7f7 112#include <linux/highmem.h>
1da177e4
LT		 113
114#include <asm/uaccess.h>
115#include <asm/system.h>
116
117#include <linux/netdevice.h>
118#include <net/protocol.h>
119#include <linux/skbuff.h>
457c4cbc 120#include <net/net_namespace.h>
2e6599cb 121#include <net/request_sock.h>
1da177e4
LT		 122#include <net/sock.h>
123#include <net/xfrm.h>
124#include <linux/ipsec.h>
125
126#include <linux/filter.h>
127
128#ifdef CONFIG_INET
129#include <net/tcp.h>
130#endif
131
da21f24d
IM		 132/*
133 * Each address family might have different locking rules, so we have
134 * one slock key per address family:
135 */
a5b5bb9a
IM		 136static struct lock_class_key af_family_keys[AF_MAX];
137static struct lock_class_key af_family_slock_keys[AF_MAX];
138
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140/*
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
143 * locks is fast):
144 */
145static const char *af_family_key_strings[AF_MAX+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 155 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79
DH		 156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
157 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
a5b5bb9a
IM		 158};
159static const char *af_family_slock_key_strings[AF_MAX+1] = {
160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 169 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79
DH		 170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
171 "slock-AF_RXRPC" , "slock-AF_MAX"
a5b5bb9a 172};
443aef0e
PZ		 173static const char *af_family_clock_key_strings[AF_MAX+1] = {
174 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
175 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
176 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
177 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
178 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
179 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
180 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
181 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
182 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
183 "clock-27" , "clock-28" , "clock-29" ,
e51f802b
DH		 184 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
185 "clock-AF_RXRPC" , "clock-AF_MAX"
443aef0e 186};
a5b5bb9a 187#endif
da21f24d
IM		 188
189/*
190 * sk_callback_lock locking rules are per-address-family,
191 * so split the lock classes by using a per-AF key:
192 */
193static struct lock_class_key af_callback_keys[AF_MAX];
194
1da177e4
LT		 195/* Take into consideration the size of the struct sk_buff overhead in the
196 * determination of these values, since that is non-constant across
197 * platforms. This makes socket queueing behavior and performance
198 * not depend upon such differences.
199 */
200#define _SK_MEM_PACKETS 256
201#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
202#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
203#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
204
205/* Run time adjustable parameters. */
ab32ea5d
BH		 206__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
207__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
208__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
209__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4
LT		 210
211/* Maximal space eaten by iovec or ancilliary data plus some space */
ab32ea5d 212int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
1da177e4
LT		 213
214static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
215{
216 struct timeval tv;
217
218 if (optlen < sizeof(tv))
219 return -EINVAL;
220 if (copy_from_user(&tv, optval, sizeof(tv)))
221 return -EFAULT;
ba78073e
VA		 222 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
223 return -EDOM;
1da177e4 224
ba78073e 225 if (tv.tv_sec < 0) {
6f11df83
AM		 226 static int warned __read_mostly;
227
ba78073e 228 *timeo_p = 0;
50aab54f 229 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 230 warned++;
231 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
232 "tries to set negative timeout\n",
ba25f9dc 233 current->comm, task_pid_nr(current));
50aab54f 234 }
ba78073e
VA		 235 return 0;
236 }
1da177e4
LT		 237 *timeo_p = MAX_SCHEDULE_TIMEOUT;
238 if (tv.tv_sec == 0 && tv.tv_usec == 0)
239 return 0;
240 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
241 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
242 return 0;
243}
244
245static void sock_warn_obsolete_bsdism(const char *name)
246{
247 static int warned;
248 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 249 if (strcmp(warncomm, current->comm) && warned < 5) {
250 strcpy(warncomm, current->comm);
1da177e4
LT		 251 printk(KERN_WARNING "process `%s' is using obsolete "
252 "%s SO_BSDCOMPAT\n", warncomm, name);
253 warned++;
254 }
255}
256
257static void sock_disable_timestamp(struct sock *sk)
4ec93edb
YH		 258{
259 if (sock_flag(sk, SOCK_TIMESTAMP)) {
1da177e4
LT		 260 sock_reset_flag(sk, SOCK_TIMESTAMP);
261 net_disable_timestamp();
262 }
263}
264
265
f0088a50
DV		 266int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
267{
268 int err = 0;
269 int skb_len;
270
9ee6b7f1 271 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
f0088a50
DV		 272 number of warnings when compiling with -W --ANK
273 */
274 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
275 (unsigned)sk->sk_rcvbuf) {
276 err = -ENOMEM;
277 goto out;
278 }
279
fda9ef5d 280 err = sk_filter(sk, skb);
f0088a50
DV		 281 if (err)
282 goto out;
283
3ab224be
HA		 284 if (!sk_rmem_schedule(sk, skb->truesize)) {
285 err = -ENOBUFS;
286 goto out;
287 }
288
f0088a50
DV		 289 skb->dev = NULL;
290 skb_set_owner_r(skb, sk);
291
292 /* Cache the SKB length before we tack it onto the receive
293 * queue. Once it is added it no longer belongs to us and
294 * may be freed by other threads of control pulling packets
295 * from the queue.
296 */
297 skb_len = skb->len;
298
299 skb_queue_tail(&sk->sk_receive_queue, skb);
300
301 if (!sock_flag(sk, SOCK_DEAD))
302 sk->sk_data_ready(sk, skb_len);
303out:
304 return err;
305}
306EXPORT_SYMBOL(sock_queue_rcv_skb);
307
58a5a7b9 308int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 309{
310 int rc = NET_RX_SUCCESS;
311
fda9ef5d 312 if (sk_filter(sk, skb))
f0088a50
DV		 313 goto discard_and_relse;
314
315 skb->dev = NULL;
316
58a5a7b9
ACM		 317 if (nested)
318 bh_lock_sock_nested(sk);
319 else
320 bh_lock_sock(sk);
a5b5bb9a
IM		 321 if (!sock_owned_by_user(sk)) {
322 /*
323 * trylock + unlock semantics:
324 */
325 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
326
f0088a50 327 rc = sk->sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 328
329 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
330 } else
f0088a50
DV		 331 sk_add_backlog(sk, skb);
332 bh_unlock_sock(sk);
333out:
334 sock_put(sk);
335 return rc;
336discard_and_relse:
337 kfree_skb(skb);
338 goto out;
339}
340EXPORT_SYMBOL(sk_receive_skb);
341
342struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
343{
344 struct dst_entry *dst = sk->sk_dst_cache;
345
346 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
347 sk->sk_dst_cache = NULL;
348 dst_release(dst);
349 return NULL;
350 }
351
352 return dst;
353}
354EXPORT_SYMBOL(__sk_dst_check);
355
356struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
357{
358 struct dst_entry *dst = sk_dst_get(sk);
359
360 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
361 sk_dst_reset(sk);
362 dst_release(dst);
363 return NULL;
364 }
365
366 return dst;
367}
368EXPORT_SYMBOL(sk_dst_check);
369
4878809f
DM		 370static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
371{
372 int ret = -ENOPROTOOPT;
373#ifdef CONFIG_NETDEVICES
3b1e0a65 374 struct net *net = sock_net(sk);
4878809f
DM		 375 char devname[IFNAMSIZ];
376 int index;
377
378 /* Sorry... */
379 ret = -EPERM;
380 if (!capable(CAP_NET_RAW))
381 goto out;
382
383 ret = -EINVAL;
384 if (optlen < 0)
385 goto out;
386
387 /* Bind this socket to a particular device like "eth0",
388 * as specified in the passed interface name. If the
389 * name is "" or the option length is zero the socket
390 * is not bound.
391 */
392 if (optlen > IFNAMSIZ - 1)
393 optlen = IFNAMSIZ - 1;
394 memset(devname, 0, sizeof(devname));
395
396 ret = -EFAULT;
397 if (copy_from_user(devname, optval, optlen))
398 goto out;
399
400 if (devname[0] == '\0') {
401 index = 0;
402 } else {
881d966b 403 struct net_device *dev = dev_get_by_name(net, devname);
4878809f
DM		 404
405 ret = -ENODEV;
406 if (!dev)
407 goto out;
408
409 index = dev->ifindex;
410 dev_put(dev);
411 }
412
413 lock_sock(sk);
414 sk->sk_bound_dev_if = index;
415 sk_dst_reset(sk);
416 release_sock(sk);
417
418 ret = 0;
419
420out:
421#endif
422
423 return ret;
424}
425
c0ef877b
PE		 426static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
427{
428 if (valbool)
429 sock_set_flag(sk, bit);
430 else
431 sock_reset_flag(sk, bit);
432}
433
1da177e4
LT		 434/*
435 * This is meant for all protocols to use and covers goings on
436 * at the socket level. Everything here is generic.
437 */
438
439int sock_setsockopt(struct socket *sock, int level, int optname,
440 char __user *optval, int optlen)
441{
442 struct sock *sk=sock->sk;
1da177e4
LT		 443 int val;
444 int valbool;
445 struct linger ling;
446 int ret = 0;
4ec93edb 447
1da177e4
LT		 448 /*
449 * Options without arguments
450 */
451
4878809f
DM		 452 if (optname == SO_BINDTODEVICE)
453 return sock_bindtodevice(sk, optval, optlen);
454
e71a4783
SH		 455 if (optlen < sizeof(int))
456 return -EINVAL;
4ec93edb 457
1da177e4
LT		 458 if (get_user(val, (int __user *)optval))
459 return -EFAULT;
4ec93edb
YH		 460
461 valbool = val?1:0;
1da177e4
LT		 462
463 lock_sock(sk);
464
e71a4783
SH		 465 switch(optname) {
466 case SO_DEBUG:
467 if (val && !capable(CAP_NET_ADMIN)) {
468 ret = -EACCES;
c0ef877b
PE		 469 } else
470 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 471 break;
472 case SO_REUSEADDR:
473 sk->sk_reuse = valbool;
474 break;
475 case SO_TYPE:
476 case SO_ERROR:
477 ret = -ENOPROTOOPT;
478 break;
479 case SO_DONTROUTE:
c0ef877b 480 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 481 break;
482 case SO_BROADCAST:
483 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
484 break;
485 case SO_SNDBUF:
486 /* Don't error on this BSD doesn't and if you think
487 about it this is right. Otherwise apps have to
488 play 'guess the biggest size' games. RCVBUF/SNDBUF
489 are treated in BSD as hints */
490
491 if (val > sysctl_wmem_max)
492 val = sysctl_wmem_max;
b0573dea 493set_sndbuf:
e71a4783
SH		 494 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
495 if ((val * 2) < SOCK_MIN_SNDBUF)
496 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
497 else
498 sk->sk_sndbuf = val * 2;
1da177e4 499
e71a4783
SH		 500 /*
501 * Wake up sending tasks if we
502 * upped the value.
503 */
504 sk->sk_write_space(sk);
505 break;
1da177e4 506
e71a4783
SH		 507 case SO_SNDBUFFORCE:
508 if (!capable(CAP_NET_ADMIN)) {
509 ret = -EPERM;
510 break;
511 }
512 goto set_sndbuf;
b0573dea 513
e71a4783
SH		 514 case SO_RCVBUF:
515 /* Don't error on this BSD doesn't and if you think
516 about it this is right. Otherwise apps have to
517 play 'guess the biggest size' games. RCVBUF/SNDBUF
518 are treated in BSD as hints */
4ec93edb 519
e71a4783
SH		 520 if (val > sysctl_rmem_max)
521 val = sysctl_rmem_max;
b0573dea 522set_rcvbuf:
e71a4783
SH		 523 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
524 /*
525 * We double it on the way in to account for
526 * "struct sk_buff" etc. overhead. Applications
527 * assume that the SO_RCVBUF setting they make will
528 * allow that much actual data to be received on that
529 * socket.
530 *
531 * Applications are unaware that "struct sk_buff" and
532 * other overheads allocate from the receive buffer
533 * during socket buffer allocation.
534 *
535 * And after considering the possible alternatives,
536 * returning the value we actually used in getsockopt
537 * is the most desirable behavior.
538 */
539 if ((val * 2) < SOCK_MIN_RCVBUF)
540 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
541 else
542 sk->sk_rcvbuf = val * 2;
543 break;
544
545 case SO_RCVBUFFORCE:
546 if (!capable(CAP_NET_ADMIN)) {
547 ret = -EPERM;
1da177e4 548 break;
e71a4783
SH		 549 }
550 goto set_rcvbuf;
1da177e4 551
e71a4783 552 case SO_KEEPALIVE:
1da177e4 553#ifdef CONFIG_INET
e71a4783
SH		 554 if (sk->sk_protocol == IPPROTO_TCP)
555 tcp_set_keepalive(sk, valbool);
1da177e4 556#endif
e71a4783
SH		 557 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
558 break;
559
560 case SO_OOBINLINE:
561 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
562 break;
563
564 case SO_NO_CHECK:
565 sk->sk_no_check = valbool;
566 break;
567
568 case SO_PRIORITY:
569 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
570 sk->sk_priority = val;
571 else
572 ret = -EPERM;
573 break;
574
575 case SO_LINGER:
576 if (optlen < sizeof(ling)) {
577 ret = -EINVAL; /* 1003.1g */
1da177e4 578 break;
e71a4783
SH		 579 }
580 if (copy_from_user(&ling,optval,sizeof(ling))) {
581 ret = -EFAULT;
1da177e4 582 break;
e71a4783
SH		 583 }
584 if (!ling.l_onoff)
585 sock_reset_flag(sk, SOCK_LINGER);
586 else {
1da177e4 587#if (BITS_PER_LONG == 32)
e71a4783
SH		 588 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
589 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 590 else
e71a4783
SH		 591#endif
592 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
593 sock_set_flag(sk, SOCK_LINGER);
594 }
595 break;
596
597 case SO_BSDCOMPAT:
598 sock_warn_obsolete_bsdism("setsockopt");
599 break;
600
601 case SO_PASSCRED:
602 if (valbool)
603 set_bit(SOCK_PASSCRED, &sock->flags);
604 else
605 clear_bit(SOCK_PASSCRED, &sock->flags);
606 break;
607
608 case SO_TIMESTAMP:
92f37fd2 609 case SO_TIMESTAMPNS:
e71a4783 610 if (valbool) {
92f37fd2
ED		 611 if (optname == SO_TIMESTAMP)
612 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
613 else
614 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 615 sock_set_flag(sk, SOCK_RCVTSTAMP);
616 sock_enable_timestamp(sk);
92f37fd2 617 } else {
e71a4783 618 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 619 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
620 }
e71a4783
SH		 621 break;
622
623 case SO_RCVLOWAT:
624 if (val < 0)
625 val = INT_MAX;
626 sk->sk_rcvlowat = val ? : 1;
627 break;
628
629 case SO_RCVTIMEO:
630 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
631 break;
632
633 case SO_SNDTIMEO:
634 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
635 break;
1da177e4 636
e71a4783
SH		 637 case SO_ATTACH_FILTER:
638 ret = -EINVAL;
639 if (optlen == sizeof(struct sock_fprog)) {
640 struct sock_fprog fprog;
1da177e4 641
e71a4783
SH		 642 ret = -EFAULT;
643 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 644 break;
e71a4783
SH		 645
646 ret = sk_attach_filter(&fprog, sk);
647 }
648 break;
649
650 case SO_DETACH_FILTER:
55b33325 651 ret = sk_detach_filter(sk);
e71a4783 652 break;
1da177e4 653
e71a4783
SH		 654 case SO_PASSSEC:
655 if (valbool)
656 set_bit(SOCK_PASSSEC, &sock->flags);
657 else
658 clear_bit(SOCK_PASSSEC, &sock->flags);
659 break;
4a19ec58
LAT		 660 case SO_MARK:
661 if (!capable(CAP_NET_ADMIN))
662 ret = -EPERM;
663 else {
664 sk->sk_mark = val;
665 }
666 break;
877ce7c1 667
1da177e4
LT		 668 /* We implement the SO_SNDLOWAT etc to
669 not be settable (1003.1g 5.3) */
e71a4783
SH		 670 default:
671 ret = -ENOPROTOOPT;
672 break;
4ec93edb 673 }
1da177e4
LT		 674 release_sock(sk);
675 return ret;
676}
677
678
679int sock_getsockopt(struct socket *sock, int level, int optname,
680 char __user *optval, int __user *optlen)
681{
682 struct sock *sk = sock->sk;
4ec93edb 683
e71a4783 684 union {
4ec93edb
YH		 685 int val;
686 struct linger ling;
1da177e4
LT		 687 struct timeval tm;
688 } v;
4ec93edb 689
1da177e4
LT		 690 unsigned int lv = sizeof(int);
691 int len;
4ec93edb 692
e71a4783 693 if (get_user(len, optlen))
4ec93edb 694 return -EFAULT;
e71a4783 695 if (len < 0)
1da177e4 696 return -EINVAL;
4ec93edb 697
e71a4783
SH		 698 switch(optname) {
699 case SO_DEBUG:
700 v.val = sock_flag(sk, SOCK_DBG);
701 break;
702
703 case SO_DONTROUTE:
704 v.val = sock_flag(sk, SOCK_LOCALROUTE);
705 break;
706
707 case SO_BROADCAST:
708 v.val = !!sock_flag(sk, SOCK_BROADCAST);
709 break;
710
711 case SO_SNDBUF:
712 v.val = sk->sk_sndbuf;
713 break;
714
715 case SO_RCVBUF:
716 v.val = sk->sk_rcvbuf;
717 break;
718
719 case SO_REUSEADDR:
720 v.val = sk->sk_reuse;
721 break;
722
723 case SO_KEEPALIVE:
724 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
725 break;
726
727 case SO_TYPE:
728 v.val = sk->sk_type;
729 break;
730
731 case SO_ERROR:
732 v.val = -sock_error(sk);
733 if (v.val==0)
734 v.val = xchg(&sk->sk_err_soft, 0);
735 break;
736
737 case SO_OOBINLINE:
738 v.val = !!sock_flag(sk, SOCK_URGINLINE);
739 break;
740
741 case SO_NO_CHECK:
742 v.val = sk->sk_no_check;
743 break;
744
745 case SO_PRIORITY:
746 v.val = sk->sk_priority;
747 break;
748
749 case SO_LINGER:
750 lv = sizeof(v.ling);
751 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
752 v.ling.l_linger = sk->sk_lingertime / HZ;
753 break;
754
755 case SO_BSDCOMPAT:
756 sock_warn_obsolete_bsdism("getsockopt");
757 break;
758
759 case SO_TIMESTAMP:
92f37fd2
ED		 760 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
761 !sock_flag(sk, SOCK_RCVTSTAMPNS);
762 break;
763
764 case SO_TIMESTAMPNS:
765 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 766 break;
767
768 case SO_RCVTIMEO:
769 lv=sizeof(struct timeval);
770 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
771 v.tm.tv_sec = 0;
772 v.tm.tv_usec = 0;
773 } else {
774 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
775 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
776 }
777 break;
778
779 case SO_SNDTIMEO:
780 lv=sizeof(struct timeval);
781 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
782 v.tm.tv_sec = 0;
783 v.tm.tv_usec = 0;
784 } else {
785 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
786 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
787 }
788 break;
1da177e4 789
e71a4783
SH		 790 case SO_RCVLOWAT:
791 v.val = sk->sk_rcvlowat;
792 break;
1da177e4 793
e71a4783
SH		 794 case SO_SNDLOWAT:
795 v.val=1;
796 break;
1da177e4 797
e71a4783
SH		 798 case SO_PASSCRED:
799 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
800 break;
1da177e4 801
e71a4783
SH		 802 case SO_PEERCRED:
803 if (len > sizeof(sk->sk_peercred))
804 len = sizeof(sk->sk_peercred);
805 if (copy_to_user(optval, &sk->sk_peercred, len))
806 return -EFAULT;
807 goto lenout;
1da177e4 808
e71a4783
SH		 809 case SO_PEERNAME:
810 {
811 char address[128];
812
813 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
814 return -ENOTCONN;
815 if (lv < len)
816 return -EINVAL;
817 if (copy_to_user(optval, address, len))
818 return -EFAULT;
819 goto lenout;
820 }
1da177e4 821
e71a4783
SH		 822 /* Dubious BSD thing... Probably nobody even uses it, but
823 * the UNIX standard wants it for whatever reason... -DaveM
824 */
825 case SO_ACCEPTCONN:
826 v.val = sk->sk_state == TCP_LISTEN;
827 break;
1da177e4 828
e71a4783
SH		 829 case SO_PASSSEC:
830 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
831 break;
877ce7c1 832
e71a4783
SH		 833 case SO_PEERSEC:
834 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 835
4a19ec58
LAT		 836 case SO_MARK:
837 v.val = sk->sk_mark;
838 break;
839
e71a4783
SH		 840 default:
841 return -ENOPROTOOPT;
1da177e4 842 }
e71a4783 843
1da177e4
LT		 844 if (len > lv)
845 len = lv;
846 if (copy_to_user(optval, &v, len))
847 return -EFAULT;
848lenout:
4ec93edb
YH		 849 if (put_user(len, optlen))
850 return -EFAULT;
851 return 0;
1da177e4
LT		 852}
853
a5b5bb9a
IM		 854/*
855 * Initialize an sk_lock.
856 *
857 * (We also register the sk_lock with the lock validator.)
858 */
b6f99a21 859static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 860{
ed07536e
PZ		 861 sock_lock_init_class_and_name(sk,
862 af_family_slock_key_strings[sk->sk_family],
863 af_family_slock_keys + sk->sk_family,
864 af_family_key_strings[sk->sk_family],
865 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 866}
867
f1a6c4da
PE		 868static void sock_copy(struct sock *nsk, const struct sock *osk)
869{
870#ifdef CONFIG_SECURITY_NETWORK
871 void *sptr = nsk->sk_security;
872#endif
873
874 memcpy(nsk, osk, osk->sk_prot->obj_size);
f1a6c4da
PE		 875#ifdef CONFIG_SECURITY_NETWORK
876 nsk->sk_security = sptr;
877 security_sk_clone(osk, nsk);
878#endif
879}
880
2e4afe7b
PE		 881static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
882 int family)
c308c1b2
PE		 883{
884 struct sock *sk;
885 struct kmem_cache *slab;
886
887 slab = prot->slab;
888 if (slab != NULL)
889 sk = kmem_cache_alloc(slab, priority);
890 else
891 sk = kmalloc(prot->obj_size, priority);
892
2e4afe7b
PE		 893 if (sk != NULL) {
894 if (security_sk_alloc(sk, family, priority))
895 goto out_free;
896
897 if (!try_module_get(prot->owner))
898 goto out_free_sec;
899 }
900
c308c1b2 901 return sk;
2e4afe7b
PE		 902
903out_free_sec:
904 security_sk_free(sk);
905out_free:
906 if (slab != NULL)
907 kmem_cache_free(slab, sk);
908 else
909 kfree(sk);
910 return NULL;
c308c1b2
PE		 911}
912
913static void sk_prot_free(struct proto *prot, struct sock *sk)
914{
915 struct kmem_cache *slab;
2e4afe7b 916 struct module *owner;
c308c1b2 917
2e4afe7b 918 owner = prot->owner;
c308c1b2 919 slab = prot->slab;
2e4afe7b
PE		 920
921 security_sk_free(sk);
c308c1b2
PE		 922 if (slab != NULL)
923 kmem_cache_free(slab, sk);
924 else
925 kfree(sk);
2e4afe7b 926 module_put(owner);
c308c1b2
PE		 927}
928
1da177e4
LT		 929/**
930 * sk_alloc - All socket objects are allocated here
c4ea43c5 931 * @net: the applicable net namespace
4dc3b16b
PP		 932 * @family: protocol family
933 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
934 * @prot: struct proto associated with this new sock instance
1da177e4 935 */
1b8d7ae4 936struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 937 struct proto *prot)
1da177e4 938{
c308c1b2 939 struct sock *sk;
1da177e4 940
154adbc8 941 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 942 if (sk) {
154adbc8
PE		 943 sk->sk_family = family;
944 /*
945 * See comment in struct sock definition to understand
946 * why we need sk_prot_creator -acme
947 */
948 sk->sk_prot = sk->sk_prot_creator = prot;
949 sock_lock_init(sk);
3b1e0a65 950 sock_net_set(sk, get_net(net));
1da177e4 951 }
a79af59e 952
2e4afe7b 953 return sk;
1da177e4
LT		 954}
955
956void sk_free(struct sock *sk)
957{
958 struct sk_filter *filter;
1da177e4
LT		 959
960 if (sk->sk_destruct)
961 sk->sk_destruct(sk);
962
fda9ef5d 963 filter = rcu_dereference(sk->sk_filter);
1da177e4 964 if (filter) {
309dd5fc 965 sk_filter_uncharge(sk, filter);
fda9ef5d 966 rcu_assign_pointer(sk->sk_filter, NULL);
1da177e4
LT		 967 }
968
969 sock_disable_timestamp(sk);
970
971 if (atomic_read(&sk->sk_omem_alloc))
972 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 973 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 974
3b1e0a65 975 put_net(sock_net(sk));
c308c1b2 976 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4
LT		 977}
978
edf02087
DL		 979/*
980 * Last sock_put should drop referrence to sk->sk_net. It has already
981 * been dropped in sk_change_net. Taking referrence to stopping namespace
982 * is not an option.
983 * Take referrence to a socket to remove it from hash _alive_ and after that
984 * destroy it in the context of init_net.
985 */
986void sk_release_kernel(struct sock *sk)
987{
988 if (sk == NULL || sk->sk_socket == NULL)
989 return;
990
991 sock_hold(sk);
992 sock_release(sk->sk_socket);
65a18ec5 993 release_net(sock_net(sk));
3b1e0a65 994 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 995 sock_put(sk);
996}
45af1754 997EXPORT_SYMBOL(sk_release_kernel);
edf02087 998
dd0fc66f 999struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
87d11ceb 1000{
8fd1d178 1001 struct sock *newsk;
87d11ceb 1002
8fd1d178 1003 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1004 if (newsk != NULL) {
1005 struct sk_filter *filter;
1006
892c141e 1007 sock_copy(newsk, sk);
87d11ceb
ACM		 1008
1009 /* SANITY */
3b1e0a65 1010 get_net(sock_net(newsk));
87d11ceb
ACM		 1011 sk_node_init(&newsk->sk_node);
1012 sock_lock_init(newsk);
1013 bh_lock_sock(newsk);
fa438ccf 1014 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
87d11ceb
ACM		 1015
1016 atomic_set(&newsk->sk_rmem_alloc, 0);
1017 atomic_set(&newsk->sk_wmem_alloc, 0);
1018 atomic_set(&newsk->sk_omem_alloc, 0);
1019 skb_queue_head_init(&newsk->sk_receive_queue);
1020 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1021#ifdef CONFIG_NET_DMA
1022 skb_queue_head_init(&newsk->sk_async_wait_queue);
1023#endif
87d11ceb
ACM		 1024
1025 rwlock_init(&newsk->sk_dst_lock);
1026 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1027 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1028 af_callback_keys + newsk->sk_family,
1029 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1030
1031 newsk->sk_dst_cache = NULL;
1032 newsk->sk_wmem_queued = 0;
1033 newsk->sk_forward_alloc = 0;
1034 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1035 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1036
1037 sock_reset_flag(newsk, SOCK_DONE);
1038 skb_queue_head_init(&newsk->sk_error_queue);
1039
1040 filter = newsk->sk_filter;
1041 if (filter != NULL)
1042 sk_filter_charge(newsk, filter);
1043
1044 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1045 /* It is still raw copy of parent, so invalidate
1046 * destructor and make plain sk_free() */
1047 newsk->sk_destruct = NULL;
1048 sk_free(newsk);
1049 newsk = NULL;
1050 goto out;
1051 }
1052
1053 newsk->sk_err = 0;
1054 newsk->sk_priority = 0;
1055 atomic_set(&newsk->sk_refcnt, 2);
1056
1057 /*
1058 * Increment the counter in the same struct proto as the master
1059 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1060 * is the same as sk->sk_prot->socks, as this field was copied
1061 * with memcpy).
1062 *
1063 * This _changes_ the previous behaviour, where
1064 * tcp_create_openreq_child always was incrementing the
1065 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1066 * to be taken into account in all callers. -acme
1067 */
1068 sk_refcnt_debug_inc(newsk);
972692e0 1069 sk_set_socket(newsk, NULL);
87d11ceb
ACM		 1070 newsk->sk_sleep = NULL;
1071
1072 if (newsk->sk_prot->sockets_allocated)
1073 atomic_inc(newsk->sk_prot->sockets_allocated);
1074 }
1075out:
1076 return newsk;
1077}
1078
1079EXPORT_SYMBOL_GPL(sk_clone);
1080
9958089a
AK		 1081void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1082{
1083 __sk_dst_set(sk, dst);
1084 sk->sk_route_caps = dst->dev->features;
1085 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1086 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
9958089a 1087 if (sk_can_gso(sk)) {
82cc1a7a 1088 if (dst->header_len) {
9958089a 1089 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1090 } else {
9958089a 1091 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1092 sk->sk_gso_max_size = dst->dev->gso_max_size;
1093 }
9958089a
AK		 1094 }
1095}
1096EXPORT_SYMBOL_GPL(sk_setup_caps);
1097
1da177e4
LT		 1098void __init sk_init(void)
1099{
1100 if (num_physpages <= 4096) {
1101 sysctl_wmem_max = 32767;
1102 sysctl_rmem_max = 32767;
1103 sysctl_wmem_default = 32767;
1104 sysctl_rmem_default = 32767;
1105 } else if (num_physpages >= 131072) {
1106 sysctl_wmem_max = 131071;
1107 sysctl_rmem_max = 131071;
1108 }
1109}
1110
1111/*
1112 * Simple resource managers for sockets.
1113 */
1114
1115
4ec93edb
YH		 1116/*
1117 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1118 */
1119void sock_wfree(struct sk_buff *skb)
1120{
1121 struct sock *sk = skb->sk;
1122
1123 /* In case it might be waiting for more memory. */
1124 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1125 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1126 sk->sk_write_space(sk);
1127 sock_put(sk);
1128}
1129
4ec93edb
YH		 1130/*
1131 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1132 */
1133void sock_rfree(struct sk_buff *skb)
1134{
1135 struct sock *sk = skb->sk;
1136
3ab224be 1137 skb_truesize_check(skb);
1da177e4 1138 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3ab224be 1139 sk_mem_uncharge(skb->sk, skb->truesize);
1da177e4
LT		 1140}
1141
1142
1143int sock_i_uid(struct sock *sk)
1144{
1145 int uid;
1146
1147 read_lock(&sk->sk_callback_lock);
1148 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1149 read_unlock(&sk->sk_callback_lock);
1150 return uid;
1151}
1152
1153unsigned long sock_i_ino(struct sock *sk)
1154{
1155 unsigned long ino;
1156
1157 read_lock(&sk->sk_callback_lock);
1158 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1159 read_unlock(&sk->sk_callback_lock);
1160 return ino;
1161}
1162
1163/*
1164 * Allocate a skb from the socket's send buffer.
1165 */
86a76caf 1166struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1167 gfp_t priority)
1da177e4
LT		 1168{
1169 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1170 struct sk_buff * skb = alloc_skb(size, priority);
1171 if (skb) {
1172 skb_set_owner_w(skb, sk);
1173 return skb;
1174 }
1175 }
1176 return NULL;
1177}
1178
1179/*
1180 * Allocate a skb from the socket's receive buffer.
4ec93edb 1181 */
86a76caf 1182struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1183 gfp_t priority)
1da177e4
LT		 1184{
1185 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1186 struct sk_buff *skb = alloc_skb(size, priority);
1187 if (skb) {
1188 skb_set_owner_r(skb, sk);
1189 return skb;
1190 }
1191 }
1192 return NULL;
1193}
1194
4ec93edb 1195/*
1da177e4 1196 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1197 */
dd0fc66f 1198void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1199{
1200 if ((unsigned)size <= sysctl_optmem_max &&
1201 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1202 void *mem;
1203 /* First do the add, to avoid the race if kmalloc
4ec93edb 1204 * might sleep.
1da177e4
LT		 1205 */
1206 atomic_add(size, &sk->sk_omem_alloc);
1207 mem = kmalloc(size, priority);
1208 if (mem)
1209 return mem;
1210 atomic_sub(size, &sk->sk_omem_alloc);
1211 }
1212 return NULL;
1213}
1214
1215/*
1216 * Free an option memory block.
1217 */
1218void sock_kfree_s(struct sock *sk, void *mem, int size)
1219{
1220 kfree(mem);
1221 atomic_sub(size, &sk->sk_omem_alloc);
1222}
1223
1224/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1225 I think, these locks should be removed for datagram sockets.
1226 */
1227static long sock_wait_for_wmem(struct sock * sk, long timeo)
1228{
1229 DEFINE_WAIT(wait);
1230
1231 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1232 for (;;) {
1233 if (!timeo)
1234 break;
1235 if (signal_pending(current))
1236 break;
1237 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1238 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1239 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1240 break;
1241 if (sk->sk_shutdown & SEND_SHUTDOWN)
1242 break;
1243 if (sk->sk_err)
1244 break;
1245 timeo = schedule_timeout(timeo);
1246 }
1247 finish_wait(sk->sk_sleep, &wait);
1248 return timeo;
1249}
1250
1251
1252/*
1253 * Generic send/receive buffer handlers
1254 */
1255
1256static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1257 unsigned long header_len,
1258 unsigned long data_len,
1259 int noblock, int *errcode)
1260{
1261 struct sk_buff *skb;
7d877f3b 1262 gfp_t gfp_mask;
1da177e4
LT		 1263 long timeo;
1264 int err;
1265
1266 gfp_mask = sk->sk_allocation;
1267 if (gfp_mask & __GFP_WAIT)
1268 gfp_mask |= __GFP_REPEAT;
1269
1270 timeo = sock_sndtimeo(sk, noblock);
1271 while (1) {
1272 err = sock_error(sk);
1273 if (err != 0)
1274 goto failure;
1275
1276 err = -EPIPE;
1277 if (sk->sk_shutdown & SEND_SHUTDOWN)
1278 goto failure;
1279
1280 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1281 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1282 if (skb) {
1283 int npages;
1284 int i;
1285
1286 /* No pages, we're done... */
1287 if (!data_len)
1288 break;
1289
1290 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1291 skb->truesize += data_len;
1292 skb_shinfo(skb)->nr_frags = npages;
1293 for (i = 0; i < npages; i++) {
1294 struct page *page;
1295 skb_frag_t *frag;
1296
1297 page = alloc_pages(sk->sk_allocation, 0);
1298 if (!page) {
1299 err = -ENOBUFS;
1300 skb_shinfo(skb)->nr_frags = i;
1301 kfree_skb(skb);
1302 goto failure;
1303 }
1304
1305 frag = &skb_shinfo(skb)->frags[i];
1306 frag->page = page;
1307 frag->page_offset = 0;
1308 frag->size = (data_len >= PAGE_SIZE ?
1309 PAGE_SIZE :
1310 data_len);
1311 data_len -= PAGE_SIZE;
1312 }
1313
1314 /* Full success... */
1315 break;
1316 }
1317 err = -ENOBUFS;
1318 goto failure;
1319 }
1320 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1321 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1322 err = -EAGAIN;
1323 if (!timeo)
1324 goto failure;
1325 if (signal_pending(current))
1326 goto interrupted;
1327 timeo = sock_wait_for_wmem(sk, timeo);
1328 }
1329
1330 skb_set_owner_w(skb, sk);
1331 return skb;
1332
1333interrupted:
1334 err = sock_intr_errno(timeo);
1335failure:
1336 *errcode = err;
1337 return NULL;
1338}
1339
4ec93edb 1340struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1341 int noblock, int *errcode)
1342{
1343 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1344}
1345
1346static void __lock_sock(struct sock *sk)
1347{
1348 DEFINE_WAIT(wait);
1349
e71a4783 1350 for (;;) {
1da177e4
LT		 1351 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1352 TASK_UNINTERRUPTIBLE);
1353 spin_unlock_bh(&sk->sk_lock.slock);
1354 schedule();
1355 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1356 if (!sock_owned_by_user(sk))
1da177e4
LT		 1357 break;
1358 }
1359 finish_wait(&sk->sk_lock.wq, &wait);
1360}
1361
1362static void __release_sock(struct sock *sk)
1363{
1364 struct sk_buff *skb = sk->sk_backlog.head;
1365
1366 do {
1367 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1368 bh_unlock_sock(sk);
1369
1370 do {
1371 struct sk_buff *next = skb->next;
1372
1373 skb->next = NULL;
1374 sk->sk_backlog_rcv(sk, skb);
1375
1376 /*
1377 * We are in process context here with softirqs
1378 * disabled, use cond_resched_softirq() to preempt.
1379 * This is safe to do because we've taken the backlog
1380 * queue private:
1381 */
1382 cond_resched_softirq();
1383
1384 skb = next;
1385 } while (skb != NULL);
1386
1387 bh_lock_sock(sk);
e71a4783 1388 } while ((skb = sk->sk_backlog.head) != NULL);
1da177e4
LT		 1389}
1390
1391/**
1392 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1393 * @sk: sock to wait on
1394 * @timeo: for how long
1da177e4
LT		 1395 *
1396 * Now socket state including sk->sk_err is changed only under lock,
1397 * hence we may omit checks after joining wait queue.
1398 * We check receive queue before schedule() only as optimization;
1399 * it is very likely that release_sock() added new data.
1400 */
1401int sk_wait_data(struct sock *sk, long *timeo)
1402{
1403 int rc;
1404 DEFINE_WAIT(wait);
1405
1406 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1407 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1408 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1409 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1410 finish_wait(sk->sk_sleep, &wait);
1411 return rc;
1412}
1413
1414EXPORT_SYMBOL(sk_wait_data);
1415
3ab224be
HA		 1416/**
1417 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1418 * @sk: socket
1419 * @size: memory size to allocate
1420 * @kind: allocation type
1421 *
1422 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1423 * rmem allocation. This function assumes that protocols which have
1424 * memory_pressure use sk_wmem_queued as write buffer accounting.
1425 */
1426int __sk_mem_schedule(struct sock *sk, int size, int kind)
1427{
1428 struct proto *prot = sk->sk_prot;
1429 int amt = sk_mem_pages(size);
1430 int allocated;
1431
1432 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1433 allocated = atomic_add_return(amt, prot->memory_allocated);
1434
1435 /* Under limit. */
1436 if (allocated <= prot->sysctl_mem[0]) {
1437 if (prot->memory_pressure && *prot->memory_pressure)
1438 *prot->memory_pressure = 0;
1439 return 1;
1440 }
1441
1442 /* Under pressure. */
1443 if (allocated > prot->sysctl_mem[1])
1444 if (prot->enter_memory_pressure)
5c52ba17 1445 prot->enter_memory_pressure(sk);
3ab224be
HA		 1446
1447 /* Over hard limit. */
1448 if (allocated > prot->sysctl_mem[2])
1449 goto suppress_allocation;
1450
1451 /* guarantee minimum buffer size under pressure */
1452 if (kind == SK_MEM_RECV) {
1453 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1454 return 1;
1455 } else { /* SK_MEM_SEND */
1456 if (sk->sk_type == SOCK_STREAM) {
1457 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1458 return 1;
1459 } else if (atomic_read(&sk->sk_wmem_alloc) <
1460 prot->sysctl_wmem[0])
1461 return 1;
1462 }
1463
1464 if (prot->memory_pressure) {
1465 if (!*prot->memory_pressure ||
1466 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1467 sk_mem_pages(sk->sk_wmem_queued +
1468 atomic_read(&sk->sk_rmem_alloc) +
1469 sk->sk_forward_alloc))
1470 return 1;
1471 }
1472
1473suppress_allocation:
1474
1475 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1476 sk_stream_moderate_sndbuf(sk);
1477
1478 /* Fail only if socket is _under_ its sndbuf.
1479 * In this case we cannot block, so that we have to fail.
1480 */
1481 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1482 return 1;
1483 }
1484
1485 /* Alas. Undo changes. */
1486 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1487 atomic_sub(amt, prot->memory_allocated);
1488 return 0;
1489}
1490
1491EXPORT_SYMBOL(__sk_mem_schedule);
1492
1493/**
1494 * __sk_reclaim - reclaim memory_allocated
1495 * @sk: socket
1496 */
1497void __sk_mem_reclaim(struct sock *sk)
1498{
1499 struct proto *prot = sk->sk_prot;
1500
680a5a50 1501 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1502 prot->memory_allocated);
1503 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1504
1505 if (prot->memory_pressure && *prot->memory_pressure &&
1506 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1507 *prot->memory_pressure = 0;
1508}
1509
1510EXPORT_SYMBOL(__sk_mem_reclaim);
1511
1512
1da177e4
LT		 1513/*
1514 * Set of default routines for initialising struct proto_ops when
1515 * the protocol does not support a particular function. In certain
1516 * cases where it makes no sense for a protocol to have a "do nothing"
1517 * function, some default processing is provided.
1518 */
1519
1520int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1521{
1522 return -EOPNOTSUPP;
1523}
1524
4ec93edb 1525int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1526 int len, int flags)
1527{
1528 return -EOPNOTSUPP;
1529}
1530
1531int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1532{
1533 return -EOPNOTSUPP;
1534}
1535
1536int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1537{
1538 return -EOPNOTSUPP;
1539}
1540
4ec93edb 1541int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1542 int *len, int peer)
1543{
1544 return -EOPNOTSUPP;
1545}
1546
1547unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1548{
1549 return 0;
1550}
1551
1552int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1553{
1554 return -EOPNOTSUPP;
1555}
1556
1557int sock_no_listen(struct socket *sock, int backlog)
1558{
1559 return -EOPNOTSUPP;
1560}
1561
1562int sock_no_shutdown(struct socket *sock, int how)
1563{
1564 return -EOPNOTSUPP;
1565}
1566
1567int sock_no_setsockopt(struct socket *sock, int level, int optname,
1568 char __user *optval, int optlen)
1569{
1570 return -EOPNOTSUPP;
1571}
1572
1573int sock_no_getsockopt(struct socket *sock, int level, int optname,
1574 char __user *optval, int __user *optlen)
1575{
1576 return -EOPNOTSUPP;
1577}
1578
1579int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1580 size_t len)
1581{
1582 return -EOPNOTSUPP;
1583}
1584
1585int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1586 size_t len, int flags)
1587{
1588 return -EOPNOTSUPP;
1589}
1590
1591int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1592{
1593 /* Mirror missing mmap method error code */
1594 return -ENODEV;
1595}
1596
1597ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1598{
1599 ssize_t res;
1600 struct msghdr msg = {.msg_flags = flags};
1601 struct kvec iov;
1602 char *kaddr = kmap(page);
1603 iov.iov_base = kaddr + offset;
1604 iov.iov_len = size;
1605 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1606 kunmap(page);
1607 return res;
1608}
1609
1610/*
1611 * Default Socket Callbacks
1612 */
1613
1614static void sock_def_wakeup(struct sock *sk)
1615{
1616 read_lock(&sk->sk_callback_lock);
1617 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1618 wake_up_interruptible_all(sk->sk_sleep);
1619 read_unlock(&sk->sk_callback_lock);
1620}
1621
1622static void sock_def_error_report(struct sock *sk)
1623{
1624 read_lock(&sk->sk_callback_lock);
1625 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1626 wake_up_interruptible(sk->sk_sleep);
8d8ad9d7 1627 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1da177e4
LT		 1628 read_unlock(&sk->sk_callback_lock);
1629}
1630
1631static void sock_def_readable(struct sock *sk, int len)
1632{
1633 read_lock(&sk->sk_callback_lock);
1634 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
6f3d0929 1635 wake_up_interruptible_sync(sk->sk_sleep);
8d8ad9d7 1636 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1da177e4
LT		 1637 read_unlock(&sk->sk_callback_lock);
1638}
1639
1640static void sock_def_write_space(struct sock *sk)
1641{
1642 read_lock(&sk->sk_callback_lock);
1643
1644 /* Do not wake up a writer until he can make "significant"
1645 * progress. --DaveM
1646 */
e71a4783 1647 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1da177e4 1648 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
6f3d0929 1649 wake_up_interruptible_sync(sk->sk_sleep);
1da177e4
LT		 1650
1651 /* Should agree with poll, otherwise some programs break */
1652 if (sock_writeable(sk))
8d8ad9d7 1653 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1654 }
1655
1656 read_unlock(&sk->sk_callback_lock);
1657}
1658
1659static void sock_def_destruct(struct sock *sk)
1660{
a51482bd 1661 kfree(sk->sk_protinfo);
1da177e4
LT		 1662}
1663
1664void sk_send_sigurg(struct sock *sk)
1665{
1666 if (sk->sk_socket && sk->sk_socket->file)
1667 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1668 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4
LT		 1669}
1670
1671void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1672 unsigned long expires)
1673{
1674 if (!mod_timer(timer, expires))
1675 sock_hold(sk);
1676}
1677
1678EXPORT_SYMBOL(sk_reset_timer);
1679
1680void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1681{
1682 if (timer_pending(timer) && del_timer(timer))
1683 __sock_put(sk);
1684}
1685
1686EXPORT_SYMBOL(sk_stop_timer);
1687
1688void sock_init_data(struct socket *sock, struct sock *sk)
1689{
1690 skb_queue_head_init(&sk->sk_receive_queue);
1691 skb_queue_head_init(&sk->sk_write_queue);
1692 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 1693#ifdef CONFIG_NET_DMA
1694 skb_queue_head_init(&sk->sk_async_wait_queue);
1695#endif
1da177e4
LT		 1696
1697 sk->sk_send_head = NULL;
1698
1699 init_timer(&sk->sk_timer);
4ec93edb 1700
1da177e4
LT		 1701 sk->sk_allocation = GFP_KERNEL;
1702 sk->sk_rcvbuf = sysctl_rmem_default;
1703 sk->sk_sndbuf = sysctl_wmem_default;
1704 sk->sk_state = TCP_CLOSE;
972692e0 1705 sk_set_socket(sk, sock);
1da177e4
LT		 1706
1707 sock_set_flag(sk, SOCK_ZAPPED);
1708
e71a4783 1709 if (sock) {
1da177e4
LT		 1710 sk->sk_type = sock->type;
1711 sk->sk_sleep = &sock->wait;
1712 sock->sk = sk;
1713 } else
1714 sk->sk_sleep = NULL;
1715
1716 rwlock_init(&sk->sk_dst_lock);
1717 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 1718 lockdep_set_class_and_name(&sk->sk_callback_lock,
1719 af_callback_keys + sk->sk_family,
1720 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 1721
1722 sk->sk_state_change = sock_def_wakeup;
1723 sk->sk_data_ready = sock_def_readable;
1724 sk->sk_write_space = sock_def_write_space;
1725 sk->sk_error_report = sock_def_error_report;
1726 sk->sk_destruct = sock_def_destruct;
1727
1728 sk->sk_sndmsg_page = NULL;
1729 sk->sk_sndmsg_off = 0;
1730
1731 sk->sk_peercred.pid = 0;
1732 sk->sk_peercred.uid = -1;
1733 sk->sk_peercred.gid = -1;
1734 sk->sk_write_pending = 0;
1735 sk->sk_rcvlowat = 1;
1736 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1737 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1738
f37f0afb 1739 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4
LT		 1740
1741 atomic_set(&sk->sk_refcnt, 1);
33c732c3 1742 atomic_set(&sk->sk_drops, 0);
1da177e4
LT		 1743}
1744
b5606c2d 1745void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 1746{
1747 might_sleep();
a5b5bb9a 1748 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 1749 if (sk->sk_lock.owned)
1da177e4 1750 __lock_sock(sk);
d2e9117c 1751 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 1752 spin_unlock(&sk->sk_lock.slock);
1753 /*
1754 * The sk_lock has mutex_lock() semantics here:
1755 */
fcc70d5f 1756 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 1757 local_bh_enable();
1da177e4
LT		 1758}
1759
fcc70d5f 1760EXPORT_SYMBOL(lock_sock_nested);
1da177e4 1761
b5606c2d 1762void release_sock(struct sock *sk)
1da177e4 1763{
a5b5bb9a
IM		 1764 /*
1765 * The sk_lock has mutex_unlock() semantics:
1766 */
1767 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1768
1769 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 1770 if (sk->sk_backlog.tail)
1771 __release_sock(sk);
d2e9117c 1772 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 1773 if (waitqueue_active(&sk->sk_lock.wq))
1774 wake_up(&sk->sk_lock.wq);
1775 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 1776}
1777EXPORT_SYMBOL(release_sock);
1778
1779int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 1780{
b7aa0bf7 1781 struct timeval tv;
1da177e4
LT		 1782 if (!sock_flag(sk, SOCK_TIMESTAMP))
1783 sock_enable_timestamp(sk);
b7aa0bf7
ED		 1784 tv = ktime_to_timeval(sk->sk_stamp);
1785 if (tv.tv_sec == -1)
1da177e4 1786 return -ENOENT;
b7aa0bf7
ED		 1787 if (tv.tv_sec == 0) {
1788 sk->sk_stamp = ktime_get_real();
1789 tv = ktime_to_timeval(sk->sk_stamp);
1790 }
1791 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 1792}
1da177e4
LT		 1793EXPORT_SYMBOL(sock_get_timestamp);
1794
ae40eb1e
ED		 1795int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1796{
1797 struct timespec ts;
1798 if (!sock_flag(sk, SOCK_TIMESTAMP))
1799 sock_enable_timestamp(sk);
1800 ts = ktime_to_timespec(sk->sk_stamp);
1801 if (ts.tv_sec == -1)
1802 return -ENOENT;
1803 if (ts.tv_sec == 0) {
1804 sk->sk_stamp = ktime_get_real();
1805 ts = ktime_to_timespec(sk->sk_stamp);
1806 }
1807 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1808}
1809EXPORT_SYMBOL(sock_get_timestampns);
1810
1da177e4 1811void sock_enable_timestamp(struct sock *sk)
4ec93edb
YH		 1812{
1813 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1da177e4
LT		 1814 sock_set_flag(sk, SOCK_TIMESTAMP);
1815 net_enable_timestamp();
1816 }
1817}
1da177e4
LT		 1818
1819/*
1820 * Get a socket option on an socket.
1821 *
1822 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1823 * asynchronous errors should be reported by getsockopt. We assume
1824 * this means if you specify SO_ERROR (otherwise whats the point of it).
1825 */
1826int sock_common_getsockopt(struct socket *sock, int level, int optname,
1827 char __user *optval, int __user *optlen)
1828{
1829 struct sock *sk = sock->sk;
1830
1831 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1832}
1833
1834EXPORT_SYMBOL(sock_common_getsockopt);
1835
3fdadf7d 1836#ifdef CONFIG_COMPAT
543d9cfe
ACM		 1837int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1838 char __user *optval, int __user *optlen)
3fdadf7d
DM		 1839{
1840 struct sock *sk = sock->sk;
1841
1e51f951 1842 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 1843 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1844 optval, optlen);
3fdadf7d
DM		 1845 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1846}
1847EXPORT_SYMBOL(compat_sock_common_getsockopt);
1848#endif
1849
1da177e4
LT		 1850int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1851 struct msghdr *msg, size_t size, int flags)
1852{
1853 struct sock *sk = sock->sk;
1854 int addr_len = 0;
1855 int err;
1856
1857 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1858 flags & ~MSG_DONTWAIT, &addr_len);
1859 if (err >= 0)
1860 msg->msg_namelen = addr_len;
1861 return err;
1862}
1863
1864EXPORT_SYMBOL(sock_common_recvmsg);
1865
1866/*
1867 * Set socket options on an inet socket.
1868 */
1869int sock_common_setsockopt(struct socket *sock, int level, int optname,
1870 char __user *optval, int optlen)
1871{
1872 struct sock *sk = sock->sk;
1873
1874 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1875}
1876
1877EXPORT_SYMBOL(sock_common_setsockopt);
1878
3fdadf7d 1879#ifdef CONFIG_COMPAT
543d9cfe
ACM		 1880int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1881 char __user *optval, int optlen)
3fdadf7d
DM		 1882{
1883 struct sock *sk = sock->sk;
1884
543d9cfe
ACM		 1885 if (sk->sk_prot->compat_setsockopt != NULL)
1886 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1887 optval, optlen);
3fdadf7d
DM		 1888 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1889}
1890EXPORT_SYMBOL(compat_sock_common_setsockopt);
1891#endif
1892
1da177e4
LT		 1893void sk_common_release(struct sock *sk)
1894{
1895 if (sk->sk_prot->destroy)
1896 sk->sk_prot->destroy(sk);
1897
1898 /*
1899 * Observation: when sock_common_release is called, processes have
1900 * no access to socket. But net still has.
1901 * Step one, detach it from networking:
1902 *
1903 * A. Remove from hash tables.
1904 */
1905
1906 sk->sk_prot->unhash(sk);
1907
1908 /*
1909 * In this point socket cannot receive new packets, but it is possible
1910 * that some packets are in flight because some CPU runs receiver and
1911 * did hash table lookup before we unhashed socket. They will achieve
1912 * receive queue and will be purged by socket destructor.
1913 *
1914 * Also we still have packets pending on receive queue and probably,
1915 * our own packets waiting in device queues. sock_destroy will drain
1916 * receive queue, but transmitted packets will delay socket destruction
1917 * until the last reference will be released.
1918 */
1919
1920 sock_orphan(sk);
1921
1922 xfrm_sk_free_policy(sk);
1923
e6848976 1924 sk_refcnt_debug_release(sk);
1da177e4
LT		 1925 sock_put(sk);
1926}
1927
1928EXPORT_SYMBOL(sk_common_release);
1929
1930static DEFINE_RWLOCK(proto_list_lock);
1931static LIST_HEAD(proto_list);
1932
13ff3d6f
PE		 1933#ifdef CONFIG_PROC_FS
1934#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 1935struct prot_inuse {
1936 int val[PROTO_INUSE_NR];
1937};
13ff3d6f
PE		 1938
1939static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 1940
1941#ifdef CONFIG_NET_NS
1942void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1943{
1944 int cpu = smp_processor_id();
1945 per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1946}
1947EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1948
1949int sock_prot_inuse_get(struct net *net, struct proto *prot)
1950{
1951 int cpu, idx = prot->inuse_idx;
1952 int res = 0;
1953
1954 for_each_possible_cpu(cpu)
1955 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1956
1957 return res >= 0 ? res : 0;
1958}
1959EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1960
1961static int sock_inuse_init_net(struct net *net)
1962{
1963 net->core.inuse = alloc_percpu(struct prot_inuse);
1964 return net->core.inuse ? 0 : -ENOMEM;
1965}
1966
1967static void sock_inuse_exit_net(struct net *net)
1968{
1969 free_percpu(net->core.inuse);
1970}
1971
1972static struct pernet_operations net_inuse_ops = {
1973 .init = sock_inuse_init_net,
1974 .exit = sock_inuse_exit_net,
1975};
1976
1977static __init int net_inuse_init(void)
1978{
1979 if (register_pernet_subsys(&net_inuse_ops))
1980 panic("Cannot initialize net inuse counters");
1981
1982 return 0;
1983}
1984
1985core_initcall(net_inuse_init);
1986#else
1338d466
PE		 1987static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1988
c29a0bc4 1989void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466
PE		 1990{
1991 __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1992}
1993EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1994
c29a0bc4 1995int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 1996{
1997 int cpu, idx = prot->inuse_idx;
1998 int res = 0;
1999
2000 for_each_possible_cpu(cpu)
2001 res += per_cpu(prot_inuse, cpu).val[idx];
2002
2003 return res >= 0 ? res : 0;
2004}
2005EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2006#endif
13ff3d6f
PE		 2007
2008static void assign_proto_idx(struct proto *prot)
2009{
2010 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2011
2012 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2013 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2014 return;
2015 }
2016
2017 set_bit(prot->inuse_idx, proto_inuse_idx);
2018}
2019
2020static void release_proto_idx(struct proto *prot)
2021{
2022 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2023 clear_bit(prot->inuse_idx, proto_inuse_idx);
2024}
2025#else
2026static inline void assign_proto_idx(struct proto *prot)
2027{
2028}
2029
2030static inline void release_proto_idx(struct proto *prot)
2031{
2032}
2033#endif
2034
b733c007
PE		 2035int proto_register(struct proto *prot, int alloc_slab)
2036{
2037 char *request_sock_slab_name = NULL;
2038 char *timewait_sock_slab_name;
2039
1da177e4
LT		 2040 if (alloc_slab) {
2041 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
20c2df83 2042 SLAB_HWCACHE_ALIGN, NULL);
1da177e4
LT		 2043
2044 if (prot->slab == NULL) {
2045 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2046 prot->name);
60e7663d 2047 goto out;
1da177e4 2048 }
2e6599cb
ACM		 2049
2050 if (prot->rsk_prot != NULL) {
2051 static const char mask[] = "request_sock_%s";
2052
2053 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2054 if (request_sock_slab_name == NULL)
2055 goto out_free_sock_slab;
2056
2057 sprintf(request_sock_slab_name, mask, prot->name);
2058 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
2059 prot->rsk_prot->obj_size, 0,
20c2df83 2060 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2061
2062 if (prot->rsk_prot->slab == NULL) {
2063 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2064 prot->name);
2065 goto out_free_request_sock_slab_name;
2066 }
2067 }
8feaf0c0 2068
6d6ee43e 2069 if (prot->twsk_prot != NULL) {
8feaf0c0
ACM		 2070 static const char mask[] = "tw_sock_%s";
2071
2072 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2073
2074 if (timewait_sock_slab_name == NULL)
2075 goto out_free_request_sock_slab;
2076
2077 sprintf(timewait_sock_slab_name, mask, prot->name);
6d6ee43e
ACM		 2078 prot->twsk_prot->twsk_slab =
2079 kmem_cache_create(timewait_sock_slab_name,
2080 prot->twsk_prot->twsk_obj_size,
2081 0, SLAB_HWCACHE_ALIGN,
20c2df83 2082 NULL);
6d6ee43e 2083 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2084 goto out_free_timewait_sock_slab_name;
2085 }
1da177e4
LT		 2086 }
2087
2a278051 2088 write_lock(&proto_list_lock);
1da177e4 2089 list_add(&prot->node, &proto_list);
13ff3d6f 2090 assign_proto_idx(prot);
1da177e4 2091 write_unlock(&proto_list_lock);
b733c007
PE		 2092 return 0;
2093
8feaf0c0
ACM		 2094out_free_timewait_sock_slab_name:
2095 kfree(timewait_sock_slab_name);
2096out_free_request_sock_slab:
2097 if (prot->rsk_prot && prot->rsk_prot->slab) {
2098 kmem_cache_destroy(prot->rsk_prot->slab);
2099 prot->rsk_prot->slab = NULL;
2100 }
2e6599cb
ACM		 2101out_free_request_sock_slab_name:
2102 kfree(request_sock_slab_name);
2103out_free_sock_slab:
2104 kmem_cache_destroy(prot->slab);
2105 prot->slab = NULL;
b733c007
PE		 2106out:
2107 return -ENOBUFS;
1da177e4
LT		 2108}
2109
2110EXPORT_SYMBOL(proto_register);
2111
2112void proto_unregister(struct proto *prot)
2113{
2114 write_lock(&proto_list_lock);
13ff3d6f 2115 release_proto_idx(prot);
0a3f4358
PM		 2116 list_del(&prot->node);
2117 write_unlock(&proto_list_lock);
1da177e4
LT		 2118
2119 if (prot->slab != NULL) {
2120 kmem_cache_destroy(prot->slab);
2121 prot->slab = NULL;
2122 }
2123
2e6599cb
ACM		 2124 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2125 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2126
2127 kmem_cache_destroy(prot->rsk_prot->slab);
2128 kfree(name);
2129 prot->rsk_prot->slab = NULL;
2130 }
2131
6d6ee43e
ACM		 2132 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2133 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
8feaf0c0 2134
6d6ee43e 2135 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
8feaf0c0 2136 kfree(name);
6d6ee43e 2137 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2138 }
1da177e4
LT		 2139}
2140
2141EXPORT_SYMBOL(proto_unregister);
2142
2143#ifdef CONFIG_PROC_FS
1da177e4 2144static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2145 __acquires(proto_list_lock)
1da177e4
LT		 2146{
2147 read_lock(&proto_list_lock);
60f0438a 2148 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2149}
2150
2151static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2152{
60f0438a 2153 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2154}
2155
2156static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2157 __releases(proto_list_lock)
1da177e4
LT		 2158{
2159 read_unlock(&proto_list_lock);
2160}
2161
2162static char proto_method_implemented(const void *method)
2163{
2164 return method == NULL ? 'n' : 'y';
2165}
2166
2167static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2168{
2169 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2170 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2171 proto->name,
2172 proto->obj_size,
2173 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2174 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2175 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2176 proto->max_header,
2177 proto->slab == NULL ? "no" : "yes",
2178 module_name(proto->owner),
2179 proto_method_implemented(proto->close),
2180 proto_method_implemented(proto->connect),
2181 proto_method_implemented(proto->disconnect),
2182 proto_method_implemented(proto->accept),
2183 proto_method_implemented(proto->ioctl),
2184 proto_method_implemented(proto->init),
2185 proto_method_implemented(proto->destroy),
2186 proto_method_implemented(proto->shutdown),
2187 proto_method_implemented(proto->setsockopt),
2188 proto_method_implemented(proto->getsockopt),
2189 proto_method_implemented(proto->sendmsg),
2190 proto_method_implemented(proto->recvmsg),
2191 proto_method_implemented(proto->sendpage),
2192 proto_method_implemented(proto->bind),
2193 proto_method_implemented(proto->backlog_rcv),
2194 proto_method_implemented(proto->hash),
2195 proto_method_implemented(proto->unhash),
2196 proto_method_implemented(proto->get_port),
2197 proto_method_implemented(proto->enter_memory_pressure));
2198}
2199
2200static int proto_seq_show(struct seq_file *seq, void *v)
2201{
60f0438a 2202 if (v == &proto_list)
1da177e4
LT		 2203 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2204 "protocol",
2205 "size",
2206 "sockets",
2207 "memory",
2208 "press",
2209 "maxhdr",
2210 "slab",
2211 "module",
2212 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2213 else
60f0438a 2214 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2215 return 0;
2216}
2217
f690808e 2218static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2219 .start = proto_seq_start,
2220 .next = proto_seq_next,
2221 .stop = proto_seq_stop,
2222 .show = proto_seq_show,
2223};
2224
2225static int proto_seq_open(struct inode *inode, struct file *file)
2226{
2227 return seq_open(file, &proto_seq_ops);
2228}
2229
9a32144e 2230static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2231 .owner = THIS_MODULE,
2232 .open = proto_seq_open,
2233 .read = seq_read,
2234 .llseek = seq_lseek,
2235 .release = seq_release,
2236};
2237
2238static int __init proto_init(void)
2239{
2240 /* register /proc/net/protocols */
457c4cbc 2241 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1da177e4
LT		 2242}
2243
2244subsys_initcall(proto_init);
2245
2246#endif /* PROC_FS */
2247
2248EXPORT_SYMBOL(sk_alloc);
2249EXPORT_SYMBOL(sk_free);
2250EXPORT_SYMBOL(sk_send_sigurg);
2251EXPORT_SYMBOL(sock_alloc_send_skb);
2252EXPORT_SYMBOL(sock_init_data);
2253EXPORT_SYMBOL(sock_kfree_s);
2254EXPORT_SYMBOL(sock_kmalloc);
2255EXPORT_SYMBOL(sock_no_accept);
2256EXPORT_SYMBOL(sock_no_bind);
2257EXPORT_SYMBOL(sock_no_connect);
2258EXPORT_SYMBOL(sock_no_getname);
2259EXPORT_SYMBOL(sock_no_getsockopt);
2260EXPORT_SYMBOL(sock_no_ioctl);
2261EXPORT_SYMBOL(sock_no_listen);
2262EXPORT_SYMBOL(sock_no_mmap);
2263EXPORT_SYMBOL(sock_no_poll);
2264EXPORT_SYMBOL(sock_no_recvmsg);
2265EXPORT_SYMBOL(sock_no_sendmsg);
2266EXPORT_SYMBOL(sock_no_sendpage);
2267EXPORT_SYMBOL(sock_no_setsockopt);
2268EXPORT_SYMBOL(sock_no_shutdown);
2269EXPORT_SYMBOL(sock_no_socketpair);
2270EXPORT_SYMBOL(sock_rfree);
2271EXPORT_SYMBOL(sock_setsockopt);
2272EXPORT_SYMBOL(sock_wfree);
2273EXPORT_SYMBOL(sock_wmalloc);
2274EXPORT_SYMBOL(sock_i_uid);
2275EXPORT_SYMBOL(sock_i_ino);
1da177e4 2276EXPORT_SYMBOL(sysctl_optmem_max);
Clone of davem's net-next-2.6 tree