blob: 0ca069738021f47d14552c160648ba3d448d7aa2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
11 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070012 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
16 *
17 * Fixes:
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +090037 * code. The ACK stuff can wait and needs major
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 * TCP layer surgery.
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
66 * (compatibility fix)
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
84 *
85 * To Fix:
86 *
87 *
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
92 */
93
Randy Dunlap4fc268d2006-01-11 12:17:47 -080094#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -0400114#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116#include <asm/uaccess.h>
117#include <asm/system.h>
118
119#include <linux/netdevice.h>
120#include <net/protocol.h>
121#include <linux/skbuff.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200122#include <net/net_namespace.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700123#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124#include <net/sock.h>
125#include <net/xfrm.h>
126#include <linux/ipsec.h>
127
128#include <linux/filter.h>
129
130#ifdef CONFIG_INET
131#include <net/tcp.h>
132#endif
133
Ingo Molnarda21f242006-07-03 00:25:12 -0700134/*
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
137 */
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700138static struct lock_class_key af_family_keys[AF_MAX];
139static struct lock_class_key af_family_slock_keys[AF_MAX];
140
141#ifdef CONFIG_DEBUG_LOCK_ALLOC
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 */
147static const char *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-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
Oliver Hartkoppcd05acf2007-12-16 15:59:24 -0800157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
David Howells17926a72007-04-26 15:48:28 -0700158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700160};
161static const char *af_family_slock_key_strings[AF_MAX+1] = {
162 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
163 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
164 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
165 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
166 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
167 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
168 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
169 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
170 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
Oliver Hartkoppcd05acf2007-12-16 15:59:24 -0800171 "slock-27" , "slock-28" , "slock-AF_CAN" ,
David Howells17926a72007-04-26 15:48:28 -0700172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700174};
Peter Zijlstra443aef02007-07-19 01:49:00 -0700175static const char *af_family_clock_key_strings[AF_MAX+1] = {
176 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
177 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
178 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
179 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
180 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
181 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
182 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
183 "clock-21" , "clock-AF_SNA" , "clock-AF_IRDA" ,
184 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
185 "clock-27" , "clock-28" , "clock-29" ,
David Howellse51f8022007-07-21 19:30:16 -0700186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
Peter Zijlstra443aef02007-07-19 01:49:00 -0700188};
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700189#endif
Ingo Molnarda21f242006-07-03 00:25:12 -0700190
191/*
192 * sk_callback_lock locking rules are per-address-family,
193 * so split the lock classes by using a per-AF key:
194 */
195static struct lock_class_key af_callback_keys[AF_MAX];
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197/* Take into consideration the size of the struct sk_buff overhead in the
198 * determination of these values, since that is non-constant across
199 * platforms. This makes socket queueing behavior and performance
200 * not depend upon such differences.
201 */
202#define _SK_MEM_PACKETS 256
203#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
204#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206
207/* Run time adjustable parameters. */
Brian Haleyab32ea52006-09-22 14:15:41 -0700208__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213/* Maximal space eaten by iovec or ancilliary data plus some space */
Brian Haleyab32ea52006-09-22 14:15:41 -0700214int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217{
218 struct timeval tv;
219
220 if (optlen < sizeof(tv))
221 return -EINVAL;
222 if (copy_from_user(&tv, optval, sizeof(tv)))
223 return -EFAULT;
Vasily Averinba780732007-05-24 16:58:54 -0700224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225 return -EDOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Vasily Averinba780732007-05-24 16:58:54 -0700227 if (tv.tv_sec < 0) {
Andrew Morton6f11df82007-07-09 13:16:00 -0700228 static int warned __read_mostly;
229
Vasily Averinba780732007-05-24 16:58:54 -0700230 *timeo_p = 0;
231 if (warned < 10 && net_ratelimit())
232 warned++;
233 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234 "tries to set negative timeout\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700235 current->comm, task_pid_nr(current));
Vasily Averinba780732007-05-24 16:58:54 -0700236 return 0;
237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 *timeo_p = MAX_SCHEDULE_TIMEOUT;
239 if (tv.tv_sec == 0 && tv.tv_usec == 0)
240 return 0;
241 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
243 return 0;
244}
245
246static void sock_warn_obsolete_bsdism(const char *name)
247{
248 static int warned;
249 static char warncomm[TASK_COMM_LEN];
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 printk(KERN_WARNING "process `%s' is using obsolete "
253 "%s SO_BSDCOMPAT\n", warncomm, name);
254 warned++;
255 }
256}
257
258static void sock_disable_timestamp(struct sock *sk)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900259{
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
263 }
264}
265
266
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800267int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
268{
269 int err = 0;
270 int skb_len;
271
272 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
273 number of warnings when compiling with -W --ANK
274 */
275 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276 (unsigned)sk->sk_rcvbuf) {
277 err = -ENOMEM;
278 goto out;
279 }
280
Dmitry Mishinfda9ef52006-08-31 15:28:39 -0700281 err = sk_filter(sk, skb);
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800282 if (err)
283 goto out;
284
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800285 if (!sk_rmem_schedule(sk, skb->truesize)) {
286 err = -ENOBUFS;
287 goto out;
288 }
289
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800290 skb->dev = NULL;
291 skb_set_owner_r(skb, sk);
292
293 /* Cache the SKB length before we tack it onto the receive
294 * queue. Once it is added it no longer belongs to us and
295 * may be freed by other threads of control pulling packets
296 * from the queue.
297 */
298 skb_len = skb->len;
299
300 skb_queue_tail(&sk->sk_receive_queue, skb);
301
302 if (!sock_flag(sk, SOCK_DEAD))
303 sk->sk_data_ready(sk, skb_len);
304out:
305 return err;
306}
307EXPORT_SYMBOL(sock_queue_rcv_skb);
308
Arnaldo Carvalho de Melo58a5a7b2006-11-16 14:06:06 -0200309int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800310{
311 int rc = NET_RX_SUCCESS;
312
Dmitry Mishinfda9ef52006-08-31 15:28:39 -0700313 if (sk_filter(sk, skb))
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800314 goto discard_and_relse;
315
316 skb->dev = NULL;
317
Arnaldo Carvalho de Melo58a5a7b2006-11-16 14:06:06 -0200318 if (nested)
319 bh_lock_sock_nested(sk);
320 else
321 bh_lock_sock(sk);
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700322 if (!sock_owned_by_user(sk)) {
323 /*
324 * trylock + unlock semantics:
325 */
326 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
327
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800328 rc = sk->sk_backlog_rcv(sk, skb);
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700329
330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
331 } else
Denis Vlasenkof0088a52006-03-28 01:08:21 -0800332 sk_add_backlog(sk, skb);
333 bh_unlock_sock(sk);
334out:
335 sock_put(sk);
336 return rc;
337discard_and_relse:
338 kfree_skb(skb);
339 goto out;
340}
341EXPORT_SYMBOL(sk_receive_skb);
342
343struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
344{
345 struct dst_entry *dst = sk->sk_dst_cache;
346
347 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
348 sk->sk_dst_cache = NULL;
349 dst_release(dst);
350 return NULL;
351 }
352
353 return dst;
354}
355EXPORT_SYMBOL(__sk_dst_check);
356
357struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
358{
359 struct dst_entry *dst = sk_dst_get(sk);
360
361 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
362 sk_dst_reset(sk);
363 dst_release(dst);
364 return NULL;
365 }
366
367 return dst;
368}
369EXPORT_SYMBOL(sk_dst_check);
370
David S. Miller48788092007-09-14 16:41:03 -0700371static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
372{
373 int ret = -ENOPROTOOPT;
374#ifdef CONFIG_NETDEVICES
Eric W. Biederman881d9662007-09-17 11:56:21 -0700375 struct net *net = sk->sk_net;
David S. Miller48788092007-09-14 16:41:03 -0700376 char devname[IFNAMSIZ];
377 int index;
378
379 /* Sorry... */
380 ret = -EPERM;
381 if (!capable(CAP_NET_RAW))
382 goto out;
383
384 ret = -EINVAL;
385 if (optlen < 0)
386 goto out;
387
388 /* Bind this socket to a particular device like "eth0",
389 * as specified in the passed interface name. If the
390 * name is "" or the option length is zero the socket
391 * is not bound.
392 */
393 if (optlen > IFNAMSIZ - 1)
394 optlen = IFNAMSIZ - 1;
395 memset(devname, 0, sizeof(devname));
396
397 ret = -EFAULT;
398 if (copy_from_user(devname, optval, optlen))
399 goto out;
400
401 if (devname[0] == '\0') {
402 index = 0;
403 } else {
Eric W. Biederman881d9662007-09-17 11:56:21 -0700404 struct net_device *dev = dev_get_by_name(net, devname);
David S. Miller48788092007-09-14 16:41:03 -0700405
406 ret = -ENODEV;
407 if (!dev)
408 goto out;
409
410 index = dev->ifindex;
411 dev_put(dev);
412 }
413
414 lock_sock(sk);
415 sk->sk_bound_dev_if = index;
416 sk_dst_reset(sk);
417 release_sock(sk);
418
419 ret = 0;
420
421out:
422#endif
423
424 return ret;
425}
426
Pavel Emelyanovc0ef8772007-11-15 03:03:19 -0800427static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
428{
429 if (valbool)
430 sock_set_flag(sk, bit);
431 else
432 sock_reset_flag(sk, bit);
433}
434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435/*
436 * This is meant for all protocols to use and covers goings on
437 * at the socket level. Everything here is generic.
438 */
439
440int sock_setsockopt(struct socket *sock, int level, int optname,
441 char __user *optval, int optlen)
442{
443 struct sock *sk=sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 int val;
445 int valbool;
446 struct linger ling;
447 int ret = 0;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 /*
450 * Options without arguments
451 */
452
453#ifdef SO_DONTLINGER /* Compatibility item... */
Kyle Moffetta77be812005-07-27 14:22:30 -0700454 if (optname == SO_DONTLINGER) {
455 lock_sock(sk);
456 sock_reset_flag(sk, SOCK_LINGER);
457 release_sock(sk);
458 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Kyle Moffetta77be812005-07-27 14:22:30 -0700460#endif
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900461
David S. Miller48788092007-09-14 16:41:03 -0700462 if (optname == SO_BINDTODEVICE)
463 return sock_bindtodevice(sk, optval, optlen);
464
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700465 if (optlen < sizeof(int))
466 return -EINVAL;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (get_user(val, (int __user *)optval))
469 return -EFAULT;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900470
471 valbool = val?1:0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 lock_sock(sk);
474
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700475 switch(optname) {
476 case SO_DEBUG:
477 if (val && !capable(CAP_NET_ADMIN)) {
478 ret = -EACCES;
Pavel Emelyanovc0ef8772007-11-15 03:03:19 -0800479 } else
480 sock_valbool_flag(sk, SOCK_DBG, valbool);
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700481 break;
482 case SO_REUSEADDR:
483 sk->sk_reuse = valbool;
484 break;
485 case SO_TYPE:
486 case SO_ERROR:
487 ret = -ENOPROTOOPT;
488 break;
489 case SO_DONTROUTE:
Pavel Emelyanovc0ef8772007-11-15 03:03:19 -0800490 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700491 break;
492 case SO_BROADCAST:
493 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
494 break;
495 case SO_SNDBUF:
496 /* Don't error on this BSD doesn't and if you think
497 about it this is right. Otherwise apps have to
498 play 'guess the biggest size' games. RCVBUF/SNDBUF
499 are treated in BSD as hints */
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900500
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700501 if (val > sysctl_wmem_max)
502 val = sysctl_wmem_max;
Patrick McHardyb0573de2005-08-09 19:30:51 -0700503set_sndbuf:
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700504 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
505 if ((val * 2) < SOCK_MIN_SNDBUF)
506 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
507 else
508 sk->sk_sndbuf = val * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700510 /*
511 * Wake up sending tasks if we
512 * upped the value.
513 */
514 sk->sk_write_space(sk);
515 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700517 case SO_SNDBUFFORCE:
518 if (!capable(CAP_NET_ADMIN)) {
519 ret = -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 break;
521 }
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700522 goto set_sndbuf;
523
524 case SO_RCVBUF:
525 /* Don't error on this BSD doesn't and if you think
526 about it this is right. Otherwise apps have to
527 play 'guess the biggest size' games. RCVBUF/SNDBUF
528 are treated in BSD as hints */
529
530 if (val > sysctl_rmem_max)
531 val = sysctl_rmem_max;
532set_rcvbuf:
533 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
534 /*
535 * We double it on the way in to account for
536 * "struct sk_buff" etc. overhead. Applications
537 * assume that the SO_RCVBUF setting they make will
538 * allow that much actual data to be received on that
539 * socket.
540 *
541 * Applications are unaware that "struct sk_buff" and
542 * other overheads allocate from the receive buffer
543 * during socket buffer allocation.
544 *
545 * And after considering the possible alternatives,
546 * returning the value we actually used in getsockopt
547 * is the most desirable behavior.
548 */
549 if ((val * 2) < SOCK_MIN_RCVBUF)
550 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
551 else
552 sk->sk_rcvbuf = val * 2;
553 break;
554
555 case SO_RCVBUFFORCE:
556 if (!capable(CAP_NET_ADMIN)) {
557 ret = -EPERM;
558 break;
559 }
560 goto set_rcvbuf;
561
562 case SO_KEEPALIVE:
563#ifdef CONFIG_INET
564 if (sk->sk_protocol == IPPROTO_TCP)
565 tcp_set_keepalive(sk, valbool);
566#endif
567 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
568 break;
569
570 case SO_OOBINLINE:
571 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
572 break;
573
574 case SO_NO_CHECK:
575 sk->sk_no_check = valbool;
576 break;
577
578 case SO_PRIORITY:
579 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
580 sk->sk_priority = val;
581 else
582 ret = -EPERM;
583 break;
584
585 case SO_LINGER:
586 if (optlen < sizeof(ling)) {
587 ret = -EINVAL; /* 1003.1g */
588 break;
589 }
590 if (copy_from_user(&ling,optval,sizeof(ling))) {
591 ret = -EFAULT;
592 break;
593 }
594 if (!ling.l_onoff)
595 sock_reset_flag(sk, SOCK_LINGER);
596 else {
597#if (BITS_PER_LONG == 32)
598 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
599 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
600 else
601#endif
602 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
603 sock_set_flag(sk, SOCK_LINGER);
604 }
605 break;
606
607 case SO_BSDCOMPAT:
608 sock_warn_obsolete_bsdism("setsockopt");
609 break;
610
611 case SO_PASSCRED:
612 if (valbool)
613 set_bit(SOCK_PASSCRED, &sock->flags);
614 else
615 clear_bit(SOCK_PASSCRED, &sock->flags);
616 break;
617
618 case SO_TIMESTAMP:
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700619 case SO_TIMESTAMPNS:
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700620 if (valbool) {
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700621 if (optname == SO_TIMESTAMP)
622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
623 else
624 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700625 sock_set_flag(sk, SOCK_RCVTSTAMP);
626 sock_enable_timestamp(sk);
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700627 } else {
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700628 sock_reset_flag(sk, SOCK_RCVTSTAMP);
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700629 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
630 }
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700631 break;
632
633 case SO_RCVLOWAT:
634 if (val < 0)
635 val = INT_MAX;
636 sk->sk_rcvlowat = val ? : 1;
637 break;
638
639 case SO_RCVTIMEO:
640 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
641 break;
642
643 case SO_SNDTIMEO:
644 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
645 break;
646
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700647 case SO_ATTACH_FILTER:
648 ret = -EINVAL;
649 if (optlen == sizeof(struct sock_fprog)) {
650 struct sock_fprog fprog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700652 ret = -EFAULT;
653 if (copy_from_user(&fprog, optval, sizeof(fprog)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700656 ret = sk_attach_filter(&fprog, sk);
657 }
658 break;
659
660 case SO_DETACH_FILTER:
Pavel Emelyanov55b33322007-10-17 21:21:26 -0700661 ret = sk_detach_filter(sk);
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700662 break;
663
664 case SO_PASSSEC:
665 if (valbool)
666 set_bit(SOCK_PASSSEC, &sock->flags);
667 else
668 clear_bit(SOCK_PASSSEC, &sock->flags);
669 break;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800670 case SO_MARK:
671 if (!capable(CAP_NET_ADMIN))
672 ret = -EPERM;
673 else {
674 sk->sk_mark = val;
675 }
676 break;
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 /* We implement the SO_SNDLOWAT etc to
679 not be settable (1003.1g 5.3) */
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700680 default:
681 ret = -ENOPROTOOPT;
682 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 release_sock(sk);
685 return ret;
686}
687
688
689int sock_getsockopt(struct socket *sock, int level, int optname,
690 char __user *optval, int __user *optlen)
691{
692 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900693
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700694 union {
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900695 int val;
696 struct linger ling;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 struct timeval tm;
698 } v;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 unsigned int lv = sizeof(int);
701 int len;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900702
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700703 if (get_user(len, optlen))
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900704 return -EFAULT;
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700705 if (len < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return -EINVAL;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900707
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700708 switch(optname) {
709 case SO_DEBUG:
710 v.val = sock_flag(sk, SOCK_DBG);
711 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900712
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700713 case SO_DONTROUTE:
714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
715 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900716
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700717 case SO_BROADCAST:
718 v.val = !!sock_flag(sk, SOCK_BROADCAST);
719 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700721 case SO_SNDBUF:
722 v.val = sk->sk_sndbuf;
723 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900724
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700725 case SO_RCVBUF:
726 v.val = sk->sk_rcvbuf;
727 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700729 case SO_REUSEADDR:
730 v.val = sk->sk_reuse;
731 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700733 case SO_KEEPALIVE:
734 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
735 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700737 case SO_TYPE:
738 v.val = sk->sk_type;
739 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700741 case SO_ERROR:
742 v.val = -sock_error(sk);
743 if (v.val==0)
744 v.val = xchg(&sk->sk_err_soft, 0);
745 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700747 case SO_OOBINLINE:
748 v.val = !!sock_flag(sk, SOCK_URGINLINE);
749 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900750
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700751 case SO_NO_CHECK:
752 v.val = sk->sk_no_check;
753 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700755 case SO_PRIORITY:
756 v.val = sk->sk_priority;
757 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900758
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700759 case SO_LINGER:
760 lv = sizeof(v.ling);
761 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
762 v.ling.l_linger = sk->sk_lingertime / HZ;
763 break;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900764
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700765 case SO_BSDCOMPAT:
766 sock_warn_obsolete_bsdism("getsockopt");
767 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700769 case SO_TIMESTAMP:
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700770 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
771 !sock_flag(sk, SOCK_RCVTSTAMPNS);
772 break;
773
774 case SO_TIMESTAMPNS:
775 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700776 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700778 case SO_RCVTIMEO:
779 lv=sizeof(struct timeval);
780 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
781 v.tm.tv_sec = 0;
782 v.tm.tv_usec = 0;
783 } else {
784 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
785 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700787 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700789 case SO_SNDTIMEO:
790 lv=sizeof(struct timeval);
791 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
792 v.tm.tv_sec = 0;
793 v.tm.tv_usec = 0;
794 } else {
795 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
796 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
797 }
798 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700800 case SO_RCVLOWAT:
801 v.val = sk->sk_rcvlowat;
802 break;
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700803
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700804 case SO_SNDLOWAT:
805 v.val=1;
806 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700808 case SO_PASSCRED:
809 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
810 break;
811
812 case SO_PEERCRED:
813 if (len > sizeof(sk->sk_peercred))
814 len = sizeof(sk->sk_peercred);
815 if (copy_to_user(optval, &sk->sk_peercred, len))
816 return -EFAULT;
817 goto lenout;
818
819 case SO_PEERNAME:
820 {
821 char address[128];
822
823 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
824 return -ENOTCONN;
825 if (lv < len)
826 return -EINVAL;
827 if (copy_to_user(optval, address, len))
828 return -EFAULT;
829 goto lenout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700831
832 /* Dubious BSD thing... Probably nobody even uses it, but
833 * the UNIX standard wants it for whatever reason... -DaveM
834 */
835 case SO_ACCEPTCONN:
836 v.val = sk->sk_state == TCP_LISTEN;
837 break;
838
839 case SO_PASSSEC:
840 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
841 break;
842
843 case SO_PEERSEC:
844 return security_socket_getpeersec_stream(sock, optval, optlen, len);
845
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800846 case SO_MARK:
847 v.val = sk->sk_mark;
848 break;
849
Stephen Hemmingere71a4782007-04-10 20:10:33 -0700850 default:
851 return -ENOPROTOOPT;
852 }
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (len > lv)
855 len = lv;
856 if (copy_to_user(optval, &v, len))
857 return -EFAULT;
858lenout:
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900859 if (put_user(len, optlen))
860 return -EFAULT;
861 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
863
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700864/*
865 * Initialize an sk_lock.
866 *
867 * (We also register the sk_lock with the lock validator.)
868 */
Dave Jonesb6f99a22007-03-22 12:27:49 -0700869static inline void sock_lock_init(struct sock *sk)
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700870{
Peter Zijlstraed075362006-12-06 20:35:24 -0800871 sock_lock_init_class_and_name(sk,
872 af_family_slock_key_strings[sk->sk_family],
873 af_family_slock_keys + sk->sk_family,
874 af_family_key_strings[sk->sk_family],
875 af_family_keys + sk->sk_family);
Ingo Molnara5b5bb92006-07-03 00:25:35 -0700876}
877
Pavel Emelyanovf1a6c4d2007-11-01 00:29:45 -0700878static void sock_copy(struct sock *nsk, const struct sock *osk)
879{
880#ifdef CONFIG_SECURITY_NETWORK
881 void *sptr = nsk->sk_security;
882#endif
883
884 memcpy(nsk, osk, osk->sk_prot->obj_size);
Pavel Emelyanovf1a6c4d2007-11-01 00:29:45 -0700885#ifdef CONFIG_SECURITY_NETWORK
886 nsk->sk_security = sptr;
887 security_sk_clone(osk, nsk);
888#endif
889}
890
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700891static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
892 int family)
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700893{
894 struct sock *sk;
895 struct kmem_cache *slab;
896
897 slab = prot->slab;
898 if (slab != NULL)
899 sk = kmem_cache_alloc(slab, priority);
900 else
901 sk = kmalloc(prot->obj_size, priority);
902
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700903 if (sk != NULL) {
904 if (security_sk_alloc(sk, family, priority))
905 goto out_free;
906
907 if (!try_module_get(prot->owner))
908 goto out_free_sec;
909 }
910
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700911 return sk;
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700912
913out_free_sec:
914 security_sk_free(sk);
915out_free:
916 if (slab != NULL)
917 kmem_cache_free(slab, sk);
918 else
919 kfree(sk);
920 return NULL;
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700921}
922
923static void sk_prot_free(struct proto *prot, struct sock *sk)
924{
925 struct kmem_cache *slab;
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700926 struct module *owner;
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700927
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700928 owner = prot->owner;
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700929 slab = prot->slab;
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700930
931 security_sk_free(sk);
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700932 if (slab != NULL)
933 kmem_cache_free(slab, sk);
934 else
935 kfree(sk);
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700936 module_put(owner);
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700937}
938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939/**
940 * sk_alloc - All socket objects are allocated here
Randy Dunlapc4ea43c2007-10-12 21:17:49 -0700941 * @net: the applicable net namespace
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700942 * @family: protocol family
943 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
944 * @prot: struct proto associated with this new sock instance
945 * @zero_it: if we should zero the newly allocated sock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700947struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
Pavel Emelyanov6257ff22007-11-01 00:39:31 -0700948 struct proto *prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700950 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Pavel Emelyanov154adbc2007-11-01 00:38:43 -0700952 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (sk) {
Pavel Emelyanov154adbc2007-11-01 00:38:43 -0700954 sk->sk_family = family;
955 /*
956 * See comment in struct sock definition to understand
957 * why we need sk_prot_creator -acme
958 */
959 sk->sk_prot = sk->sk_prot_creator = prot;
960 sock_lock_init(sk);
961 sk->sk_net = get_net(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
Frank Filza79af592005-09-27 15:23:38 -0700963
Pavel Emelyanov2e4afe72007-11-01 00:36:26 -0700964 return sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
967void sk_free(struct sock *sk)
968{
969 struct sk_filter *filter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 if (sk->sk_destruct)
972 sk->sk_destruct(sk);
973
Dmitry Mishinfda9ef52006-08-31 15:28:39 -0700974 filter = rcu_dereference(sk->sk_filter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (filter) {
Pavel Emelyanov309dd5f2007-10-17 21:21:51 -0700976 sk_filter_uncharge(sk, filter);
Dmitry Mishinfda9ef52006-08-31 15:28:39 -0700977 rcu_assign_pointer(sk->sk_filter, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 }
979
980 sock_disable_timestamp(sk);
981
982 if (atomic_read(&sk->sk_omem_alloc))
983 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
984 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
985
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700986 put_net(sk->sk_net);
Pavel Emelyanovc308c1b2007-11-01 00:33:50 -0700987 sk_prot_free(sk->sk_prot_creator, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988}
989
Denis V. Lunevedf02082008-02-29 11:18:32 -0800990/*
991 * Last sock_put should drop referrence to sk->sk_net. It has already
992 * been dropped in sk_change_net. Taking referrence to stopping namespace
993 * is not an option.
994 * Take referrence to a socket to remove it from hash _alive_ and after that
995 * destroy it in the context of init_net.
996 */
997void sk_release_kernel(struct sock *sk)
998{
999 if (sk == NULL || sk->sk_socket == NULL)
1000 return;
1001
1002 sock_hold(sk);
1003 sock_release(sk->sk_socket);
1004 sk->sk_net = get_net(&init_net);
1005 sock_put(sk);
1006}
David S. Miller45af1752008-02-29 11:33:19 -08001007EXPORT_SYMBOL(sk_release_kernel);
Denis V. Lunevedf02082008-02-29 11:18:32 -08001008
Al Virodd0fc662005-10-07 07:46:04 +01001009struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001010{
Pavel Emelyanov8fd1d172007-11-01 00:37:32 -07001011 struct sock *newsk;
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001012
Pavel Emelyanov8fd1d172007-11-01 00:37:32 -07001013 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001014 if (newsk != NULL) {
1015 struct sk_filter *filter;
1016
Venkat Yekkirala892c1412006-08-04 23:08:56 -07001017 sock_copy(newsk, sk);
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001018
1019 /* SANITY */
Pavel Emelyanov1e2e6b82007-11-01 00:31:26 -07001020 get_net(newsk->sk_net);
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001021 sk_node_init(&newsk->sk_node);
1022 sock_lock_init(newsk);
1023 bh_lock_sock(newsk);
Eric Dumazetfa438cc2007-03-04 16:05:44 -08001024 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001025
1026 atomic_set(&newsk->sk_rmem_alloc, 0);
1027 atomic_set(&newsk->sk_wmem_alloc, 0);
1028 atomic_set(&newsk->sk_omem_alloc, 0);
1029 skb_queue_head_init(&newsk->sk_receive_queue);
1030 skb_queue_head_init(&newsk->sk_write_queue);
Chris Leech97fc2f02006-05-23 17:55:33 -07001031#ifdef CONFIG_NET_DMA
1032 skb_queue_head_init(&newsk->sk_async_wait_queue);
1033#endif
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001034
1035 rwlock_init(&newsk->sk_dst_lock);
1036 rwlock_init(&newsk->sk_callback_lock);
Peter Zijlstra443aef02007-07-19 01:49:00 -07001037 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1038 af_callback_keys + newsk->sk_family,
1039 af_family_clock_key_strings[newsk->sk_family]);
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001040
1041 newsk->sk_dst_cache = NULL;
1042 newsk->sk_wmem_queued = 0;
1043 newsk->sk_forward_alloc = 0;
1044 newsk->sk_send_head = NULL;
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -07001045 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1046
1047 sock_reset_flag(newsk, SOCK_DONE);
1048 skb_queue_head_init(&newsk->sk_error_queue);
1049
1050 filter = newsk->sk_filter;
1051 if (filter != NULL)
1052 sk_filter_charge(newsk, filter);
1053
1054 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1055 /* It is still raw copy of parent, so invalidate
1056 * destructor and make plain sk_free() */
1057 newsk->sk_destruct = NULL;
1058 sk_free(newsk);
1059 newsk = NULL;
1060 goto out;
1061 }
1062
1063 newsk->sk_err = 0;
1064 newsk->sk_priority = 0;
1065 atomic_set(&newsk->sk_refcnt, 2);
1066
1067 /*
1068 * Increment the counter in the same struct proto as the master
1069 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1070 * is the same as sk->sk_prot->socks, as this field was copied
1071 * with memcpy).
1072 *
1073 * This _changes_ the previous behaviour, where
1074 * tcp_create_openreq_child always was incrementing the
1075 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1076 * to be taken into account in all callers. -acme
1077 */
1078 sk_refcnt_debug_inc(newsk);
1079 newsk->sk_socket = NULL;
1080 newsk->sk_sleep = NULL;
1081
1082 if (newsk->sk_prot->sockets_allocated)
1083 atomic_inc(newsk->sk_prot->sockets_allocated);
1084 }
1085out:
1086 return newsk;
1087}
1088
1089EXPORT_SYMBOL_GPL(sk_clone);
1090
Andi Kleen99580892007-04-20 17:12:43 -07001091void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1092{
1093 __sk_dst_set(sk, dst);
1094 sk->sk_route_caps = dst->dev->features;
1095 if (sk->sk_route_caps & NETIF_F_GSO)
Herbert Xu4fcd6b92007-05-31 22:15:50 -07001096 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
Andi Kleen99580892007-04-20 17:12:43 -07001097 if (sk_can_gso(sk)) {
1098 if (dst->header_len)
1099 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1100 else
1101 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1102 }
1103}
1104EXPORT_SYMBOL_GPL(sk_setup_caps);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106void __init sk_init(void)
1107{
1108 if (num_physpages <= 4096) {
1109 sysctl_wmem_max = 32767;
1110 sysctl_rmem_max = 32767;
1111 sysctl_wmem_default = 32767;
1112 sysctl_rmem_default = 32767;
1113 } else if (num_physpages >= 131072) {
1114 sysctl_wmem_max = 131071;
1115 sysctl_rmem_max = 131071;
1116 }
1117}
1118
1119/*
1120 * Simple resource managers for sockets.
1121 */
1122
1123
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001124/*
1125 * Write buffer destructor automatically called from kfree_skb.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 */
1127void sock_wfree(struct sk_buff *skb)
1128{
1129 struct sock *sk = skb->sk;
1130
1131 /* In case it might be waiting for more memory. */
1132 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1133 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1134 sk->sk_write_space(sk);
1135 sock_put(sk);
1136}
1137
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001138/*
1139 * Read buffer destructor automatically called from kfree_skb.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 */
1141void sock_rfree(struct sk_buff *skb)
1142{
1143 struct sock *sk = skb->sk;
1144
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001145 skb_truesize_check(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001147 sk_mem_uncharge(skb->sk, skb->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148}
1149
1150
1151int sock_i_uid(struct sock *sk)
1152{
1153 int uid;
1154
1155 read_lock(&sk->sk_callback_lock);
1156 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1157 read_unlock(&sk->sk_callback_lock);
1158 return uid;
1159}
1160
1161unsigned long sock_i_ino(struct sock *sk)
1162{
1163 unsigned long ino;
1164
1165 read_lock(&sk->sk_callback_lock);
1166 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1167 read_unlock(&sk->sk_callback_lock);
1168 return ino;
1169}
1170
1171/*
1172 * Allocate a skb from the socket's send buffer.
1173 */
Victor Fusco86a76ca2005-07-08 14:57:47 -07001174struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
Al Virodd0fc662005-10-07 07:46:04 +01001175 gfp_t priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
1177 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1178 struct sk_buff * skb = alloc_skb(size, priority);
1179 if (skb) {
1180 skb_set_owner_w(skb, sk);
1181 return skb;
1182 }
1183 }
1184 return NULL;
1185}
1186
1187/*
1188 * Allocate a skb from the socket's receive buffer.
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001189 */
Victor Fusco86a76ca2005-07-08 14:57:47 -07001190struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
Al Virodd0fc662005-10-07 07:46:04 +01001191 gfp_t priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
1193 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1194 struct sk_buff *skb = alloc_skb(size, priority);
1195 if (skb) {
1196 skb_set_owner_r(skb, sk);
1197 return skb;
1198 }
1199 }
1200 return NULL;
1201}
1202
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001203/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 * Allocate a memory block from the socket's option memory buffer.
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001205 */
Al Virodd0fc662005-10-07 07:46:04 +01001206void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207{
1208 if ((unsigned)size <= sysctl_optmem_max &&
1209 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1210 void *mem;
1211 /* First do the add, to avoid the race if kmalloc
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001212 * might sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 */
1214 atomic_add(size, &sk->sk_omem_alloc);
1215 mem = kmalloc(size, priority);
1216 if (mem)
1217 return mem;
1218 atomic_sub(size, &sk->sk_omem_alloc);
1219 }
1220 return NULL;
1221}
1222
1223/*
1224 * Free an option memory block.
1225 */
1226void sock_kfree_s(struct sock *sk, void *mem, int size)
1227{
1228 kfree(mem);
1229 atomic_sub(size, &sk->sk_omem_alloc);
1230}
1231
1232/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1233 I think, these locks should be removed for datagram sockets.
1234 */
1235static long sock_wait_for_wmem(struct sock * sk, long timeo)
1236{
1237 DEFINE_WAIT(wait);
1238
1239 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1240 for (;;) {
1241 if (!timeo)
1242 break;
1243 if (signal_pending(current))
1244 break;
1245 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1246 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1247 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1248 break;
1249 if (sk->sk_shutdown & SEND_SHUTDOWN)
1250 break;
1251 if (sk->sk_err)
1252 break;
1253 timeo = schedule_timeout(timeo);
1254 }
1255 finish_wait(sk->sk_sleep, &wait);
1256 return timeo;
1257}
1258
1259
1260/*
1261 * Generic send/receive buffer handlers
1262 */
1263
1264static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1265 unsigned long header_len,
1266 unsigned long data_len,
1267 int noblock, int *errcode)
1268{
1269 struct sk_buff *skb;
Al Viro7d877f32005-10-21 03:20:43 -04001270 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 long timeo;
1272 int err;
1273
1274 gfp_mask = sk->sk_allocation;
1275 if (gfp_mask & __GFP_WAIT)
1276 gfp_mask |= __GFP_REPEAT;
1277
1278 timeo = sock_sndtimeo(sk, noblock);
1279 while (1) {
1280 err = sock_error(sk);
1281 if (err != 0)
1282 goto failure;
1283
1284 err = -EPIPE;
1285 if (sk->sk_shutdown & SEND_SHUTDOWN)
1286 goto failure;
1287
1288 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
Larry Woodmandb38c1792006-11-03 16:05:45 -08001289 skb = alloc_skb(header_len, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 if (skb) {
1291 int npages;
1292 int i;
1293
1294 /* No pages, we're done... */
1295 if (!data_len)
1296 break;
1297
1298 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1299 skb->truesize += data_len;
1300 skb_shinfo(skb)->nr_frags = npages;
1301 for (i = 0; i < npages; i++) {
1302 struct page *page;
1303 skb_frag_t *frag;
1304
1305 page = alloc_pages(sk->sk_allocation, 0);
1306 if (!page) {
1307 err = -ENOBUFS;
1308 skb_shinfo(skb)->nr_frags = i;
1309 kfree_skb(skb);
1310 goto failure;
1311 }
1312
1313 frag = &skb_shinfo(skb)->frags[i];
1314 frag->page = page;
1315 frag->page_offset = 0;
1316 frag->size = (data_len >= PAGE_SIZE ?
1317 PAGE_SIZE :
1318 data_len);
1319 data_len -= PAGE_SIZE;
1320 }
1321
1322 /* Full success... */
1323 break;
1324 }
1325 err = -ENOBUFS;
1326 goto failure;
1327 }
1328 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1329 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1330 err = -EAGAIN;
1331 if (!timeo)
1332 goto failure;
1333 if (signal_pending(current))
1334 goto interrupted;
1335 timeo = sock_wait_for_wmem(sk, timeo);
1336 }
1337
1338 skb_set_owner_w(skb, sk);
1339 return skb;
1340
1341interrupted:
1342 err = sock_intr_errno(timeo);
1343failure:
1344 *errcode = err;
1345 return NULL;
1346}
1347
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001348struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 int noblock, int *errcode)
1350{
1351 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1352}
1353
1354static void __lock_sock(struct sock *sk)
1355{
1356 DEFINE_WAIT(wait);
1357
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001358 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1360 TASK_UNINTERRUPTIBLE);
1361 spin_unlock_bh(&sk->sk_lock.slock);
1362 schedule();
1363 spin_lock_bh(&sk->sk_lock.slock);
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001364 if (!sock_owned_by_user(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 break;
1366 }
1367 finish_wait(&sk->sk_lock.wq, &wait);
1368}
1369
1370static void __release_sock(struct sock *sk)
1371{
1372 struct sk_buff *skb = sk->sk_backlog.head;
1373
1374 do {
1375 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1376 bh_unlock_sock(sk);
1377
1378 do {
1379 struct sk_buff *next = skb->next;
1380
1381 skb->next = NULL;
1382 sk->sk_backlog_rcv(sk, skb);
1383
1384 /*
1385 * We are in process context here with softirqs
1386 * disabled, use cond_resched_softirq() to preempt.
1387 * This is safe to do because we've taken the backlog
1388 * queue private:
1389 */
1390 cond_resched_softirq();
1391
1392 skb = next;
1393 } while (skb != NULL);
1394
1395 bh_lock_sock(sk);
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001396 } while ((skb = sk->sk_backlog.head) != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397}
1398
1399/**
1400 * sk_wait_data - wait for data to arrive at sk_receive_queue
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001401 * @sk: sock to wait on
1402 * @timeo: for how long
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 *
1404 * Now socket state including sk->sk_err is changed only under lock,
1405 * hence we may omit checks after joining wait queue.
1406 * We check receive queue before schedule() only as optimization;
1407 * it is very likely that release_sock() added new data.
1408 */
1409int sk_wait_data(struct sock *sk, long *timeo)
1410{
1411 int rc;
1412 DEFINE_WAIT(wait);
1413
1414 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1415 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1416 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1417 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1418 finish_wait(sk->sk_sleep, &wait);
1419 return rc;
1420}
1421
1422EXPORT_SYMBOL(sk_wait_data);
1423
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001424/**
1425 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1426 * @sk: socket
1427 * @size: memory size to allocate
1428 * @kind: allocation type
1429 *
1430 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1431 * rmem allocation. This function assumes that protocols which have
1432 * memory_pressure use sk_wmem_queued as write buffer accounting.
1433 */
1434int __sk_mem_schedule(struct sock *sk, int size, int kind)
1435{
1436 struct proto *prot = sk->sk_prot;
1437 int amt = sk_mem_pages(size);
1438 int allocated;
1439
1440 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1441 allocated = atomic_add_return(amt, prot->memory_allocated);
1442
1443 /* Under limit. */
1444 if (allocated <= prot->sysctl_mem[0]) {
1445 if (prot->memory_pressure && *prot->memory_pressure)
1446 *prot->memory_pressure = 0;
1447 return 1;
1448 }
1449
1450 /* Under pressure. */
1451 if (allocated > prot->sysctl_mem[1])
1452 if (prot->enter_memory_pressure)
1453 prot->enter_memory_pressure();
1454
1455 /* Over hard limit. */
1456 if (allocated > prot->sysctl_mem[2])
1457 goto suppress_allocation;
1458
1459 /* guarantee minimum buffer size under pressure */
1460 if (kind == SK_MEM_RECV) {
1461 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1462 return 1;
1463 } else { /* SK_MEM_SEND */
1464 if (sk->sk_type == SOCK_STREAM) {
1465 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1466 return 1;
1467 } else if (atomic_read(&sk->sk_wmem_alloc) <
1468 prot->sysctl_wmem[0])
1469 return 1;
1470 }
1471
1472 if (prot->memory_pressure) {
1473 if (!*prot->memory_pressure ||
1474 prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1475 sk_mem_pages(sk->sk_wmem_queued +
1476 atomic_read(&sk->sk_rmem_alloc) +
1477 sk->sk_forward_alloc))
1478 return 1;
1479 }
1480
1481suppress_allocation:
1482
1483 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1484 sk_stream_moderate_sndbuf(sk);
1485
1486 /* Fail only if socket is _under_ its sndbuf.
1487 * In this case we cannot block, so that we have to fail.
1488 */
1489 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1490 return 1;
1491 }
1492
1493 /* Alas. Undo changes. */
1494 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1495 atomic_sub(amt, prot->memory_allocated);
1496 return 0;
1497}
1498
1499EXPORT_SYMBOL(__sk_mem_schedule);
1500
1501/**
1502 * __sk_reclaim - reclaim memory_allocated
1503 * @sk: socket
1504 */
1505void __sk_mem_reclaim(struct sock *sk)
1506{
1507 struct proto *prot = sk->sk_prot;
1508
Eric Dumazet680a5a52007-12-31 15:00:50 -08001509 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001510 prot->memory_allocated);
1511 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1512
1513 if (prot->memory_pressure && *prot->memory_pressure &&
1514 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1515 *prot->memory_pressure = 0;
1516}
1517
1518EXPORT_SYMBOL(__sk_mem_reclaim);
1519
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521/*
1522 * Set of default routines for initialising struct proto_ops when
1523 * the protocol does not support a particular function. In certain
1524 * cases where it makes no sense for a protocol to have a "do nothing"
1525 * function, some default processing is provided.
1526 */
1527
1528int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1529{
1530 return -EOPNOTSUPP;
1531}
1532
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001533int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 int len, int flags)
1535{
1536 return -EOPNOTSUPP;
1537}
1538
1539int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1540{
1541 return -EOPNOTSUPP;
1542}
1543
1544int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1545{
1546 return -EOPNOTSUPP;
1547}
1548
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001549int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 int *len, int peer)
1551{
1552 return -EOPNOTSUPP;
1553}
1554
1555unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1556{
1557 return 0;
1558}
1559
1560int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1561{
1562 return -EOPNOTSUPP;
1563}
1564
1565int sock_no_listen(struct socket *sock, int backlog)
1566{
1567 return -EOPNOTSUPP;
1568}
1569
1570int sock_no_shutdown(struct socket *sock, int how)
1571{
1572 return -EOPNOTSUPP;
1573}
1574
1575int sock_no_setsockopt(struct socket *sock, int level, int optname,
1576 char __user *optval, int optlen)
1577{
1578 return -EOPNOTSUPP;
1579}
1580
1581int sock_no_getsockopt(struct socket *sock, int level, int optname,
1582 char __user *optval, int __user *optlen)
1583{
1584 return -EOPNOTSUPP;
1585}
1586
1587int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1588 size_t len)
1589{
1590 return -EOPNOTSUPP;
1591}
1592
1593int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1594 size_t len, int flags)
1595{
1596 return -EOPNOTSUPP;
1597}
1598
1599int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1600{
1601 /* Mirror missing mmap method error code */
1602 return -ENODEV;
1603}
1604
1605ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1606{
1607 ssize_t res;
1608 struct msghdr msg = {.msg_flags = flags};
1609 struct kvec iov;
1610 char *kaddr = kmap(page);
1611 iov.iov_base = kaddr + offset;
1612 iov.iov_len = size;
1613 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1614 kunmap(page);
1615 return res;
1616}
1617
1618/*
1619 * Default Socket Callbacks
1620 */
1621
1622static void sock_def_wakeup(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_all(sk->sk_sleep);
1627 read_unlock(&sk->sk_callback_lock);
1628}
1629
1630static void sock_def_error_report(struct sock *sk)
1631{
1632 read_lock(&sk->sk_callback_lock);
1633 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1634 wake_up_interruptible(sk->sk_sleep);
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001635 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 read_unlock(&sk->sk_callback_lock);
1637}
1638
1639static void sock_def_readable(struct sock *sk, int len)
1640{
1641 read_lock(&sk->sk_callback_lock);
1642 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1643 wake_up_interruptible(sk->sk_sleep);
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001644 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 read_unlock(&sk->sk_callback_lock);
1646}
1647
1648static void sock_def_write_space(struct sock *sk)
1649{
1650 read_lock(&sk->sk_callback_lock);
1651
1652 /* Do not wake up a writer until he can make "significant"
1653 * progress. --DaveM
1654 */
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001655 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1657 wake_up_interruptible(sk->sk_sleep);
1658
1659 /* Should agree with poll, otherwise some programs break */
1660 if (sock_writeable(sk))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001661 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 }
1663
1664 read_unlock(&sk->sk_callback_lock);
1665}
1666
1667static void sock_def_destruct(struct sock *sk)
1668{
Jesper Juhla51482b2005-11-08 09:41:34 -08001669 kfree(sk->sk_protinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670}
1671
1672void sk_send_sigurg(struct sock *sk)
1673{
1674 if (sk->sk_socket && sk->sk_socket->file)
1675 if (send_sigurg(&sk->sk_socket->file->f_owner))
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001676 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677}
1678
1679void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1680 unsigned long expires)
1681{
1682 if (!mod_timer(timer, expires))
1683 sock_hold(sk);
1684}
1685
1686EXPORT_SYMBOL(sk_reset_timer);
1687
1688void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1689{
1690 if (timer_pending(timer) && del_timer(timer))
1691 __sock_put(sk);
1692}
1693
1694EXPORT_SYMBOL(sk_stop_timer);
1695
1696void sock_init_data(struct socket *sock, struct sock *sk)
1697{
1698 skb_queue_head_init(&sk->sk_receive_queue);
1699 skb_queue_head_init(&sk->sk_write_queue);
1700 skb_queue_head_init(&sk->sk_error_queue);
Chris Leech97fc2f02006-05-23 17:55:33 -07001701#ifdef CONFIG_NET_DMA
1702 skb_queue_head_init(&sk->sk_async_wait_queue);
1703#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 sk->sk_send_head = NULL;
1706
1707 init_timer(&sk->sk_timer);
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 sk->sk_allocation = GFP_KERNEL;
1710 sk->sk_rcvbuf = sysctl_rmem_default;
1711 sk->sk_sndbuf = sysctl_wmem_default;
1712 sk->sk_state = TCP_CLOSE;
1713 sk->sk_socket = sock;
1714
1715 sock_set_flag(sk, SOCK_ZAPPED);
1716
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001717 if (sock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 sk->sk_type = sock->type;
1719 sk->sk_sleep = &sock->wait;
1720 sock->sk = sk;
1721 } else
1722 sk->sk_sleep = NULL;
1723
1724 rwlock_init(&sk->sk_dst_lock);
1725 rwlock_init(&sk->sk_callback_lock);
Peter Zijlstra443aef02007-07-19 01:49:00 -07001726 lockdep_set_class_and_name(&sk->sk_callback_lock,
1727 af_callback_keys + sk->sk_family,
1728 af_family_clock_key_strings[sk->sk_family]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730 sk->sk_state_change = sock_def_wakeup;
1731 sk->sk_data_ready = sock_def_readable;
1732 sk->sk_write_space = sock_def_write_space;
1733 sk->sk_error_report = sock_def_error_report;
1734 sk->sk_destruct = sock_def_destruct;
1735
1736 sk->sk_sndmsg_page = NULL;
1737 sk->sk_sndmsg_off = 0;
1738
1739 sk->sk_peercred.pid = 0;
1740 sk->sk_peercred.uid = -1;
1741 sk->sk_peercred.gid = -1;
1742 sk->sk_write_pending = 0;
1743 sk->sk_rcvlowat = 1;
1744 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1745 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1746
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001747 sk->sk_stamp = ktime_set(-1L, -1L);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
1749 atomic_set(&sk->sk_refcnt, 1);
Wang Chen33c732c2007-11-13 20:30:01 -08001750 atomic_set(&sk->sk_drops, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751}
1752
Harvey Harrisonb5606c22008-02-13 15:03:16 -08001753void lock_sock_nested(struct sock *sk, int subclass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754{
1755 might_sleep();
Ingo Molnara5b5bb92006-07-03 00:25:35 -07001756 spin_lock_bh(&sk->sk_lock.slock);
John Heffnerd2e91172007-09-12 10:44:19 +02001757 if (sk->sk_lock.owned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 __lock_sock(sk);
John Heffnerd2e91172007-09-12 10:44:19 +02001759 sk->sk_lock.owned = 1;
Ingo Molnara5b5bb92006-07-03 00:25:35 -07001760 spin_unlock(&sk->sk_lock.slock);
1761 /*
1762 * The sk_lock has mutex_lock() semantics here:
1763 */
Peter Zijlstrafcc70d52006-11-08 22:44:35 -08001764 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
Ingo Molnara5b5bb92006-07-03 00:25:35 -07001765 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
Peter Zijlstrafcc70d52006-11-08 22:44:35 -08001768EXPORT_SYMBOL(lock_sock_nested);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Harvey Harrisonb5606c22008-02-13 15:03:16 -08001770void release_sock(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771{
Ingo Molnara5b5bb92006-07-03 00:25:35 -07001772 /*
1773 * The sk_lock has mutex_unlock() semantics:
1774 */
1775 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1776
1777 spin_lock_bh(&sk->sk_lock.slock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 if (sk->sk_backlog.tail)
1779 __release_sock(sk);
John Heffnerd2e91172007-09-12 10:44:19 +02001780 sk->sk_lock.owned = 0;
Ingo Molnara5b5bb92006-07-03 00:25:35 -07001781 if (waitqueue_active(&sk->sk_lock.wq))
1782 wake_up(&sk->sk_lock.wq);
1783 spin_unlock_bh(&sk->sk_lock.slock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784}
1785EXPORT_SYMBOL(release_sock);
1786
1787int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001788{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001789 struct timeval tv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 if (!sock_flag(sk, SOCK_TIMESTAMP))
1791 sock_enable_timestamp(sk);
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001792 tv = ktime_to_timeval(sk->sk_stamp);
1793 if (tv.tv_sec == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 return -ENOENT;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07001795 if (tv.tv_sec == 0) {
1796 sk->sk_stamp = ktime_get_real();
1797 tv = ktime_to_timeval(sk->sk_stamp);
1798 }
1799 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001800}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801EXPORT_SYMBOL(sock_get_timestamp);
1802
Eric Dumazetae40eb12007-03-18 17:33:16 -07001803int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1804{
1805 struct timespec ts;
1806 if (!sock_flag(sk, SOCK_TIMESTAMP))
1807 sock_enable_timestamp(sk);
1808 ts = ktime_to_timespec(sk->sk_stamp);
1809 if (ts.tv_sec == -1)
1810 return -ENOENT;
1811 if (ts.tv_sec == 0) {
1812 sk->sk_stamp = ktime_get_real();
1813 ts = ktime_to_timespec(sk->sk_stamp);
1814 }
1815 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1816}
1817EXPORT_SYMBOL(sock_get_timestampns);
1818
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819void sock_enable_timestamp(struct sock *sk)
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +09001820{
1821 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 sock_set_flag(sk, SOCK_TIMESTAMP);
1823 net_enable_timestamp();
1824 }
1825}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827/*
1828 * Get a socket option on an socket.
1829 *
1830 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1831 * asynchronous errors should be reported by getsockopt. We assume
1832 * this means if you specify SO_ERROR (otherwise whats the point of it).
1833 */
1834int sock_common_getsockopt(struct socket *sock, int level, int optname,
1835 char __user *optval, int __user *optlen)
1836{
1837 struct sock *sk = sock->sk;
1838
1839 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1840}
1841
1842EXPORT_SYMBOL(sock_common_getsockopt);
1843
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001844#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001845int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1846 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001847{
1848 struct sock *sk = sock->sk;
1849
Johannes Berg1e51f952007-03-06 13:44:06 -08001850 if (sk->sk_prot->compat_getsockopt != NULL)
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001851 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1852 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001853 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1854}
1855EXPORT_SYMBOL(compat_sock_common_getsockopt);
1856#endif
1857
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1859 struct msghdr *msg, size_t size, int flags)
1860{
1861 struct sock *sk = sock->sk;
1862 int addr_len = 0;
1863 int err;
1864
1865 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1866 flags & ~MSG_DONTWAIT, &addr_len);
1867 if (err >= 0)
1868 msg->msg_namelen = addr_len;
1869 return err;
1870}
1871
1872EXPORT_SYMBOL(sock_common_recvmsg);
1873
1874/*
1875 * Set socket options on an inet socket.
1876 */
1877int sock_common_setsockopt(struct socket *sock, int level, int optname,
1878 char __user *optval, int optlen)
1879{
1880 struct sock *sk = sock->sk;
1881
1882 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1883}
1884
1885EXPORT_SYMBOL(sock_common_setsockopt);
1886
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001887#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001888int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1889 char __user *optval, int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001890{
1891 struct sock *sk = sock->sk;
1892
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001893 if (sk->sk_prot->compat_setsockopt != NULL)
1894 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1895 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001896 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1897}
1898EXPORT_SYMBOL(compat_sock_common_setsockopt);
1899#endif
1900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901void sk_common_release(struct sock *sk)
1902{
1903 if (sk->sk_prot->destroy)
1904 sk->sk_prot->destroy(sk);
1905
1906 /*
1907 * Observation: when sock_common_release is called, processes have
1908 * no access to socket. But net still has.
1909 * Step one, detach it from networking:
1910 *
1911 * A. Remove from hash tables.
1912 */
1913
1914 sk->sk_prot->unhash(sk);
1915
1916 /*
1917 * In this point socket cannot receive new packets, but it is possible
1918 * that some packets are in flight because some CPU runs receiver and
1919 * did hash table lookup before we unhashed socket. They will achieve
1920 * receive queue and will be purged by socket destructor.
1921 *
1922 * Also we still have packets pending on receive queue and probably,
1923 * our own packets waiting in device queues. sock_destroy will drain
1924 * receive queue, but transmitted packets will delay socket destruction
1925 * until the last reference will be released.
1926 */
1927
1928 sock_orphan(sk);
1929
1930 xfrm_sk_free_policy(sk);
1931
Arnaldo Carvalho de Meloe6848972005-08-09 19:45:38 -07001932 sk_refcnt_debug_release(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 sock_put(sk);
1934}
1935
1936EXPORT_SYMBOL(sk_common_release);
1937
1938static DEFINE_RWLOCK(proto_list_lock);
1939static LIST_HEAD(proto_list);
1940
1941int proto_register(struct proto *prot, int alloc_slab)
1942{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001943 char *request_sock_slab_name = NULL;
1944 char *timewait_sock_slab_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Eric Dumazet65f76512008-01-03 20:46:48 -08001946 if (sock_prot_inuse_init(prot) != 0) {
Arnaldo Carvalho de Meloebb53d72007-11-21 22:08:50 +08001947 printk(KERN_CRIT "%s: Can't alloc inuse counters!\n", prot->name);
Pavel Emelyanovb733c002007-11-07 02:23:38 -08001948 goto out;
Arnaldo Carvalho de Meloebb53d72007-11-21 22:08:50 +08001949 }
Pavel Emelyanovb733c002007-11-07 02:23:38 -08001950
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 if (alloc_slab) {
1952 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001953 SLAB_HWCACHE_ALIGN, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 if (prot->slab == NULL) {
1956 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1957 prot->name);
Eric Dumazet286ab3d2007-11-05 23:38:39 -08001958 goto out_free_inuse;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 }
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001960
1961 if (prot->rsk_prot != NULL) {
1962 static const char mask[] = "request_sock_%s";
1963
1964 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1965 if (request_sock_slab_name == NULL)
1966 goto out_free_sock_slab;
1967
1968 sprintf(request_sock_slab_name, mask, prot->name);
1969 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1970 prot->rsk_prot->obj_size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001971 SLAB_HWCACHE_ALIGN, NULL);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001972
1973 if (prot->rsk_prot->slab == NULL) {
1974 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1975 prot->name);
1976 goto out_free_request_sock_slab_name;
1977 }
1978 }
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001979
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001980 if (prot->twsk_prot != NULL) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001981 static const char mask[] = "tw_sock_%s";
1982
1983 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1984
1985 if (timewait_sock_slab_name == NULL)
1986 goto out_free_request_sock_slab;
1987
1988 sprintf(timewait_sock_slab_name, mask, prot->name);
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001989 prot->twsk_prot->twsk_slab =
1990 kmem_cache_create(timewait_sock_slab_name,
1991 prot->twsk_prot->twsk_obj_size,
1992 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001993 NULL);
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001994 if (prot->twsk_prot->twsk_slab == NULL)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001995 goto out_free_timewait_sock_slab_name;
1996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 }
1998
Arnaldo Carvalho de Melo2a278052005-04-16 15:24:09 -07001999 write_lock(&proto_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 list_add(&prot->node, &proto_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 write_unlock(&proto_list_lock);
Pavel Emelyanovb733c002007-11-07 02:23:38 -08002002 return 0;
2003
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002004out_free_timewait_sock_slab_name:
2005 kfree(timewait_sock_slab_name);
2006out_free_request_sock_slab:
2007 if (prot->rsk_prot && prot->rsk_prot->slab) {
2008 kmem_cache_destroy(prot->rsk_prot->slab);
2009 prot->rsk_prot->slab = NULL;
2010 }
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002011out_free_request_sock_slab_name:
2012 kfree(request_sock_slab_name);
2013out_free_sock_slab:
2014 kmem_cache_destroy(prot->slab);
2015 prot->slab = NULL;
Eric Dumazet286ab3d2007-11-05 23:38:39 -08002016out_free_inuse:
Eric Dumazet65f76512008-01-03 20:46:48 -08002017 sock_prot_inuse_free(prot);
Pavel Emelyanovb733c002007-11-07 02:23:38 -08002018out:
2019 return -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020}
2021
2022EXPORT_SYMBOL(proto_register);
2023
2024void proto_unregister(struct proto *prot)
2025{
2026 write_lock(&proto_list_lock);
Patrick McHardy0a3f4352005-09-06 19:47:50 -07002027 list_del(&prot->node);
2028 write_unlock(&proto_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Eric Dumazet65f76512008-01-03 20:46:48 -08002030 sock_prot_inuse_free(prot);
Arnaldo Carvalho de Meloebb53d72007-11-21 22:08:50 +08002031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 if (prot->slab != NULL) {
2033 kmem_cache_destroy(prot->slab);
2034 prot->slab = NULL;
2035 }
2036
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002037 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2038 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2039
2040 kmem_cache_destroy(prot->rsk_prot->slab);
2041 kfree(name);
2042 prot->rsk_prot->slab = NULL;
2043 }
2044
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002045 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2046 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002047
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002048 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002049 kfree(name);
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002050 prot->twsk_prot->twsk_slab = NULL;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002051 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
2053
2054EXPORT_SYMBOL(proto_unregister);
2055
2056#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002058 __acquires(proto_list_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
2060 read_lock(&proto_list_lock);
Pavel Emelianov60f04382007-07-09 13:15:14 -07002061 return seq_list_start_head(&proto_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
2063
2064static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2065{
Pavel Emelianov60f04382007-07-09 13:15:14 -07002066 return seq_list_next(v, &proto_list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067}
2068
2069static void proto_seq_stop(struct seq_file *seq, void *v)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002070 __releases(proto_list_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071{
2072 read_unlock(&proto_list_lock);
2073}
2074
2075static char proto_method_implemented(const void *method)
2076{
2077 return method == NULL ? 'n' : 'y';
2078}
2079
2080static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2081{
2082 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2083 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2084 proto->name,
2085 proto->obj_size,
2086 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2087 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2088 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2089 proto->max_header,
2090 proto->slab == NULL ? "no" : "yes",
2091 module_name(proto->owner),
2092 proto_method_implemented(proto->close),
2093 proto_method_implemented(proto->connect),
2094 proto_method_implemented(proto->disconnect),
2095 proto_method_implemented(proto->accept),
2096 proto_method_implemented(proto->ioctl),
2097 proto_method_implemented(proto->init),
2098 proto_method_implemented(proto->destroy),
2099 proto_method_implemented(proto->shutdown),
2100 proto_method_implemented(proto->setsockopt),
2101 proto_method_implemented(proto->getsockopt),
2102 proto_method_implemented(proto->sendmsg),
2103 proto_method_implemented(proto->recvmsg),
2104 proto_method_implemented(proto->sendpage),
2105 proto_method_implemented(proto->bind),
2106 proto_method_implemented(proto->backlog_rcv),
2107 proto_method_implemented(proto->hash),
2108 proto_method_implemented(proto->unhash),
2109 proto_method_implemented(proto->get_port),
2110 proto_method_implemented(proto->enter_memory_pressure));
2111}
2112
2113static int proto_seq_show(struct seq_file *seq, void *v)
2114{
Pavel Emelianov60f04382007-07-09 13:15:14 -07002115 if (v == &proto_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2117 "protocol",
2118 "size",
2119 "sockets",
2120 "memory",
2121 "press",
2122 "maxhdr",
2123 "slab",
2124 "module",
2125 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2126 else
Pavel Emelianov60f04382007-07-09 13:15:14 -07002127 proto_seq_printf(seq, list_entry(v, struct proto, node));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return 0;
2129}
2130
Stephen Hemmingerf6908082007-03-12 14:34:29 -07002131static const struct seq_operations proto_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 .start = proto_seq_start,
2133 .next = proto_seq_next,
2134 .stop = proto_seq_stop,
2135 .show = proto_seq_show,
2136};
2137
2138static int proto_seq_open(struct inode *inode, struct file *file)
2139{
2140 return seq_open(file, &proto_seq_ops);
2141}
2142
Arjan van de Ven9a321442007-02-12 00:55:35 -08002143static const struct file_operations proto_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 .owner = THIS_MODULE,
2145 .open = proto_seq_open,
2146 .read = seq_read,
2147 .llseek = seq_lseek,
2148 .release = seq_release,
2149};
2150
2151static int __init proto_init(void)
2152{
2153 /* register /proc/net/protocols */
Eric W. Biederman457c4cb2007-09-12 12:01:34 +02002154 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155}
2156
2157subsys_initcall(proto_init);
2158
2159#endif /* PROC_FS */
2160
2161EXPORT_SYMBOL(sk_alloc);
2162EXPORT_SYMBOL(sk_free);
2163EXPORT_SYMBOL(sk_send_sigurg);
2164EXPORT_SYMBOL(sock_alloc_send_skb);
2165EXPORT_SYMBOL(sock_init_data);
2166EXPORT_SYMBOL(sock_kfree_s);
2167EXPORT_SYMBOL(sock_kmalloc);
2168EXPORT_SYMBOL(sock_no_accept);
2169EXPORT_SYMBOL(sock_no_bind);
2170EXPORT_SYMBOL(sock_no_connect);
2171EXPORT_SYMBOL(sock_no_getname);
2172EXPORT_SYMBOL(sock_no_getsockopt);
2173EXPORT_SYMBOL(sock_no_ioctl);
2174EXPORT_SYMBOL(sock_no_listen);
2175EXPORT_SYMBOL(sock_no_mmap);
2176EXPORT_SYMBOL(sock_no_poll);
2177EXPORT_SYMBOL(sock_no_recvmsg);
2178EXPORT_SYMBOL(sock_no_sendmsg);
2179EXPORT_SYMBOL(sock_no_sendpage);
2180EXPORT_SYMBOL(sock_no_setsockopt);
2181EXPORT_SYMBOL(sock_no_shutdown);
2182EXPORT_SYMBOL(sock_no_socketpair);
2183EXPORT_SYMBOL(sock_rfree);
2184EXPORT_SYMBOL(sock_setsockopt);
2185EXPORT_SYMBOL(sock_wfree);
2186EXPORT_SYMBOL(sock_wmalloc);
2187EXPORT_SYMBOL(sock_i_uid);
2188EXPORT_SYMBOL(sock_i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189EXPORT_SYMBOL(sysctl_optmem_max);