projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6-block.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 *
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
11 *
02c30a84 12 * Authors: Ross Biro
1da177e4
LT		 13 * 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
4ec93edb 37 * code. The ACK stuff can wait and needs major
1da177e4
LT		 38 * 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
4fc268d2 94#include <linux/capability.h>
1da177e4
LT		 95#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
1da177e4
LT		 100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
a1f8e7f7 114#include <linux/highmem.h>
1da177e4
LT		 115
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>
457c4cbc 122#include <net/net_namespace.h>
2e6599cb 123#include <net/request_sock.h>
1da177e4
LT		 124#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
da21f24d
IM		 134/*
135 * Each address family might have different locking rules, so we have
136 * one slock key per address family:
137 */
a5b5bb9a
IM		 138static 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" ,
cd05acfe 157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79
DH		 158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
a5b5bb9a
IM		 160};
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" ,
cd05acfe 171 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79
DH		 172 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
173 "slock-AF_RXRPC" , "slock-AF_MAX"
a5b5bb9a 174};
443aef0e
PZ		 175static 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" ,
e51f802b
DH		 186 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
187 "clock-AF_RXRPC" , "clock-AF_MAX"
443aef0e 188};
a5b5bb9a 189#endif
da21f24d
IM		 190
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
1da177e4
LT		 197/* 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. */
ab32ea5d
BH		 208__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;
1da177e4
LT		 212
213/* Maximal space eaten by iovec or ancilliary data plus some space */
ab32ea5d 214int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
1da177e4
LT		 215
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;
ba78073e
VA		 224 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225 return -EDOM;
1da177e4 226
ba78073e 227 if (tv.tv_sec < 0) {
6f11df83
AM		 228 static int warned __read_mostly;
229
ba78073e
VA		 230 *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",
ba25f9dc 235 current->comm, task_pid_nr(current));
ba78073e
VA		 236 return 0;
237 }
1da177e4
LT		 238 *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];
4ec93edb
YH		 250 if (strcmp(warncomm, current->comm) && warned < 5) {
251 strcpy(warncomm, current->comm);
1da177e4
LT		 252 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)
4ec93edb
YH		 259{
260 if (sock_flag(sk, SOCK_TIMESTAMP)) {
1da177e4
LT		 261 sock_reset_flag(sk, SOCK_TIMESTAMP);
262 net_disable_timestamp();
263 }
264}
265
266
f0088a50
DV		 267int 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
fda9ef5d 281 err = sk_filter(sk, skb);
f0088a50
DV		 282 if (err)
283 goto out;
284
3ab224be
HA		 285 if (!sk_rmem_schedule(sk, skb->truesize)) {
286 err = -ENOBUFS;
287 goto out;
288 }
289
f0088a50
DV		 290 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
58a5a7b9 309int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 310{
311 int rc = NET_RX_SUCCESS;
312
fda9ef5d 313 if (sk_filter(sk, skb))
f0088a50
DV		 314 goto discard_and_relse;
315
316 skb->dev = NULL;
317
58a5a7b9
ACM		 318 if (nested)
319 bh_lock_sock_nested(sk);
320 else
321 bh_lock_sock(sk);
a5b5bb9a
IM		 322 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
f0088a50 328 rc = sk->sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 329
330 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
331 } else
f0088a50
DV		 332 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
4878809f
DM		 371static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
372{
373 int ret = -ENOPROTOOPT;
374#ifdef CONFIG_NETDEVICES
881d966b 375 struct net *net = sk->sk_net;
4878809f
DM		 376 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 {
881d966b 404 struct net_device *dev = dev_get_by_name(net, devname);
4878809f
DM		 405
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
c0ef877b
PE		 427static 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
1da177e4
LT		 435/*
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;
1da177e4
LT		 444 int val;
445 int valbool;
446 struct linger ling;
447 int ret = 0;
4ec93edb 448
1da177e4
LT		 449 /*
450 * Options without arguments
451 */
452
453#ifdef SO_DONTLINGER /* Compatibility item... */
a77be819
KM		 454 if (optname == SO_DONTLINGER) {
455 lock_sock(sk);
456 sock_reset_flag(sk, SOCK_LINGER);
457 release_sock(sk);
458 return 0;
1da177e4 459 }
a77be819 460#endif
4ec93edb 461
4878809f
DM		 462 if (optname == SO_BINDTODEVICE)
463 return sock_bindtodevice(sk, optval, optlen);
464
e71a4783
SH		 465 if (optlen < sizeof(int))
466 return -EINVAL;
4ec93edb 467
1da177e4
LT		 468 if (get_user(val, (int __user *)optval))
469 return -EFAULT;
4ec93edb
YH		 470
471 valbool = val?1:0;
1da177e4
LT		 472
473 lock_sock(sk);
474
e71a4783
SH		 475 switch(optname) {
476 case SO_DEBUG:
477 if (val && !capable(CAP_NET_ADMIN)) {
478 ret = -EACCES;
c0ef877b
PE		 479 } else
480 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 481 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:
c0ef877b 490 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 491 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 */
500
501 if (val > sysctl_wmem_max)
502 val = sysctl_wmem_max;
b0573dea 503set_sndbuf:
e71a4783
SH		 504 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;
1da177e4 509
e71a4783
SH		 510 /*
511 * Wake up sending tasks if we
512 * upped the value.
513 */
514 sk->sk_write_space(sk);
515 break;
1da177e4 516
e71a4783
SH		 517 case SO_SNDBUFFORCE:
518 if (!capable(CAP_NET_ADMIN)) {
519 ret = -EPERM;
520 break;
521 }
522 goto set_sndbuf;
b0573dea 523
e71a4783
SH		 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 */
4ec93edb 529
e71a4783
SH		 530 if (val > sysctl_rmem_max)
531 val = sysctl_rmem_max;
b0573dea 532set_rcvbuf:
e71a4783
SH		 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;
1da177e4 558 break;
e71a4783
SH		 559 }
560 goto set_rcvbuf;
1da177e4 561
e71a4783 562 case SO_KEEPALIVE:
1da177e4 563#ifdef CONFIG_INET
e71a4783
SH		 564 if (sk->sk_protocol == IPPROTO_TCP)
565 tcp_set_keepalive(sk, valbool);
1da177e4 566#endif
e71a4783
SH		 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 */
1da177e4 588 break;
e71a4783
SH		 589 }
590 if (copy_from_user(&ling,optval,sizeof(ling))) {
591 ret = -EFAULT;
1da177e4 592 break;
e71a4783
SH		 593 }
594 if (!ling.l_onoff)
595 sock_reset_flag(sk, SOCK_LINGER);
596 else {
1da177e4 597#if (BITS_PER_LONG == 32)
e71a4783
SH		 598 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
599 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 600 else
e71a4783
SH		 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:
92f37fd2 619 case SO_TIMESTAMPNS:
e71a4783 620 if (valbool) {
92f37fd2
ED		 621 if (optname == SO_TIMESTAMP)
622 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
623 else
624 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 625 sock_set_flag(sk, SOCK_RCVTSTAMP);
626 sock_enable_timestamp(sk);
92f37fd2 627 } else {
e71a4783 628 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 629 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
630 }
e71a4783
SH		 631 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;
1da177e4 646
e71a4783
SH		 647 case SO_ATTACH_FILTER:
648 ret = -EINVAL;
649 if (optlen == sizeof(struct sock_fprog)) {
650 struct sock_fprog fprog;
1da177e4 651
e71a4783
SH		 652 ret = -EFAULT;
653 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 654 break;
e71a4783
SH		 655
656 ret = sk_attach_filter(&fprog, sk);
657 }
658 break;
659
660 case SO_DETACH_FILTER:
55b33325 661 ret = sk_detach_filter(sk);
e71a4783 662 break;
1da177e4 663
e71a4783
SH		 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;
4a19ec58
LAT		 670 case SO_MARK:
671 if (!capable(CAP_NET_ADMIN))
672 ret = -EPERM;
673 else {
674 sk->sk_mark = val;
675 }
676 break;
877ce7c1 677
1da177e4
LT		 678 /* We implement the SO_SNDLOWAT etc to
679 not be settable (1003.1g 5.3) */
e71a4783
SH		 680 default:
681 ret = -ENOPROTOOPT;
682 break;
4ec93edb 683 }
1da177e4
LT		 684 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;
4ec93edb 693
e71a4783 694 union {
4ec93edb
YH		 695 int val;
696 struct linger ling;
1da177e4
LT		 697 struct timeval tm;
698 } v;
4ec93edb 699
1da177e4
LT		 700 unsigned int lv = sizeof(int);
701 int len;
4ec93edb 702
e71a4783 703 if (get_user(len, optlen))
4ec93edb 704 return -EFAULT;
e71a4783 705 if (len < 0)
1da177e4 706 return -EINVAL;
4ec93edb 707
e71a4783
SH		 708 switch(optname) {
709 case SO_DEBUG:
710 v.val = sock_flag(sk, SOCK_DBG);
711 break;
712
713 case SO_DONTROUTE:
714 v.val = sock_flag(sk, SOCK_LOCALROUTE);
715 break;
716
717 case SO_BROADCAST:
718 v.val = !!sock_flag(sk, SOCK_BROADCAST);
719 break;
720
721 case SO_SNDBUF:
722 v.val = sk->sk_sndbuf;
723 break;
724
725 case SO_RCVBUF:
726 v.val = sk->sk_rcvbuf;
727 break;
728
729 case SO_REUSEADDR:
730 v.val = sk->sk_reuse;
731 break;
732
733 case SO_KEEPALIVE:
734 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
735 break;
736
737 case SO_TYPE:
738 v.val = sk->sk_type;
739 break;
740
741 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;
746
747 case SO_OOBINLINE:
748 v.val = !!sock_flag(sk, SOCK_URGINLINE);
749 break;
750
751 case SO_NO_CHECK:
752 v.val = sk->sk_no_check;
753 break;
754
755 case SO_PRIORITY:
756 v.val = sk->sk_priority;
757 break;
758
759 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;
764
765 case SO_BSDCOMPAT:
766 sock_warn_obsolete_bsdism("getsockopt");
767 break;
768
769 case SO_TIMESTAMP:
92f37fd2
ED		 770 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);
e71a4783
SH		 776 break;
777
778 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;
786 }
787 break;
788
789 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;
1da177e4 799
e71a4783
SH		 800 case SO_RCVLOWAT:
801 v.val = sk->sk_rcvlowat;
802 break;
1da177e4 803
e71a4783
SH		 804 case SO_SNDLOWAT:
805 v.val=1;
806 break;
1da177e4 807
e71a4783
SH		 808 case SO_PASSCRED:
809 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
810 break;
1da177e4 811
e71a4783
SH		 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;
1da177e4 818
e71a4783
SH		 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;
830 }
1da177e4 831
e71a4783
SH		 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;
1da177e4 838
e71a4783
SH		 839 case SO_PASSSEC:
840 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
841 break;
877ce7c1 842
e71a4783
SH		 843 case SO_PEERSEC:
844 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 845
4a19ec58
LAT		 846 case SO_MARK:
847 v.val = sk->sk_mark;
848 break;
849
e71a4783
SH		 850 default:
851 return -ENOPROTOOPT;
1da177e4 852 }
e71a4783 853
1da177e4
LT		 854 if (len > lv)
855 len = lv;
856 if (copy_to_user(optval, &v, len))
857 return -EFAULT;
858lenout:
4ec93edb
YH		 859 if (put_user(len, optlen))
860 return -EFAULT;
861 return 0;
1da177e4
LT		 862}
863
a5b5bb9a
IM		 864/*
865 * Initialize an sk_lock.
866 *
867 * (We also register the sk_lock with the lock validator.)
868 */
b6f99a21 869static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 870{
ed07536e
PZ		 871 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);
a5b5bb9a
IM		 876}
877
f1a6c4da
PE		 878static 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);
f1a6c4da
PE		 885#ifdef CONFIG_SECURITY_NETWORK
886 nsk->sk_security = sptr;
887 security_sk_clone(osk, nsk);
888#endif
889}
890
2e4afe7b
PE		 891static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
892 int family)
c308c1b2
PE		 893{
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
2e4afe7b
PE		 903 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
c308c1b2 911 return sk;
2e4afe7b
PE		 912
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;
c308c1b2
PE		 921}
922
923static void sk_prot_free(struct proto *prot, struct sock *sk)
924{
925 struct kmem_cache *slab;
2e4afe7b 926 struct module *owner;
c308c1b2 927
2e4afe7b 928 owner = prot->owner;
c308c1b2 929 slab = prot->slab;
2e4afe7b
PE		 930
931 security_sk_free(sk);
c308c1b2
PE		 932 if (slab != NULL)
933 kmem_cache_free(slab, sk);
934 else
935 kfree(sk);
2e4afe7b 936 module_put(owner);
c308c1b2
PE		 937}
938
1da177e4
LT		 939/**
940 * sk_alloc - All socket objects are allocated here
c4ea43c5 941 * @net: the applicable net namespace
4dc3b16b
PP		 942 * @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
1da177e4 946 */
1b8d7ae4 947struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 948 struct proto *prot)
1da177e4 949{
c308c1b2 950 struct sock *sk;
1da177e4 951
154adbc8 952 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 953 if (sk) {
154adbc8
PE		 954 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);
1da177e4 962 }
a79af59e 963
2e4afe7b 964 return sk;
1da177e4
LT		 965}
966
967void sk_free(struct sock *sk)
968{
969 struct sk_filter *filter;
1da177e4
LT		 970
971 if (sk->sk_destruct)
972 sk->sk_destruct(sk);
973
fda9ef5d 974 filter = rcu_dereference(sk->sk_filter);
1da177e4 975 if (filter) {
309dd5fc 976 sk_filter_uncharge(sk, filter);
fda9ef5d 977 rcu_assign_pointer(sk->sk_filter, NULL);
1da177e4
LT		 978 }
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",
0dc47877 984 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 985
1b8d7ae4 986 put_net(sk->sk_net);
c308c1b2 987 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4
LT		 988}
989
edf02087
DL		 990/*
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}
45af1754 1007EXPORT_SYMBOL(sk_release_kernel);
edf02087 1008
dd0fc66f 1009struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
87d11ceb 1010{
8fd1d178 1011 struct sock *newsk;
87d11ceb 1012
8fd1d178 1013 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1014 if (newsk != NULL) {
1015 struct sk_filter *filter;
1016
892c141e 1017 sock_copy(newsk, sk);
87d11ceb
ACM		 1018
1019 /* SANITY */
1e2e6b89 1020 get_net(newsk->sk_net);
87d11ceb
ACM		 1021 sk_node_init(&newsk->sk_node);
1022 sock_lock_init(newsk);
1023 bh_lock_sock(newsk);
fa438ccf 1024 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
87d11ceb
ACM		 1025
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);
97fc2f08
CL		 1031#ifdef CONFIG_NET_DMA
1032 skb_queue_head_init(&newsk->sk_async_wait_queue);
1033#endif
87d11ceb
ACM		 1034
1035 rwlock_init(&newsk->sk_dst_lock);
1036 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1037 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]);
87d11ceb
ACM		 1040
1041 newsk->sk_dst_cache = NULL;
1042 newsk->sk_wmem_queued = 0;
1043 newsk->sk_forward_alloc = 0;
1044 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1045 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
9958089a
AK		 1091void 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)
4fcd6b99 1096 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
9958089a
AK		 1097 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
1da177e4
LT		 1106void __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
4ec93edb
YH		 1124/*
1125 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1126 */
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
4ec93edb
YH		 1138/*
1139 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1140 */
1141void sock_rfree(struct sk_buff *skb)
1142{
1143 struct sock *sk = skb->sk;
1144
3ab224be 1145 skb_truesize_check(skb);
1da177e4 1146 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3ab224be 1147 sk_mem_uncharge(skb->sk, skb->truesize);
1da177e4
LT		 1148}
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 */
86a76caf 1174struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1175 gfp_t priority)
1da177e4
LT		 1176{
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.
4ec93edb 1189 */
86a76caf 1190struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1191 gfp_t priority)
1da177e4
LT		 1192{
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
4ec93edb 1203/*
1da177e4 1204 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1205 */
dd0fc66f 1206void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1207{
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
4ec93edb 1212 * might sleep.
1da177e4
LT		 1213 */
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;
7d877f3b 1270 gfp_t gfp_mask;
1da177e4
LT		 1271 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) {
db38c179 1289 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1290 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
4ec93edb 1348struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1349 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
e71a4783 1358 for (;;) {
1da177e4
LT		 1359 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);
e71a4783 1364 if (!sock_owned_by_user(sk))
1da177e4
LT		 1365 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);
e71a4783 1396 } while ((skb = sk->sk_backlog.head) != NULL);
1da177e4
LT		 1397}
1398
1399/**
1400 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1401 * @sk: sock to wait on
1402 * @timeo: for how long
1da177e4
LT		 1403 *
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
3ab224be
HA		 1424/**
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
680a5a50 1509 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1510 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
1da177e4
LT		 1521/*
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
4ec93edb 1533int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1534 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
4ec93edb 1549int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1550 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);
8d8ad9d7 1635 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1da177e4
LT		 1636 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);
8d8ad9d7 1644 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1da177e4
LT		 1645 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 */
e71a4783 1655 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1da177e4
LT		 1656 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))
8d8ad9d7 1661 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1662 }
1663
1664 read_unlock(&sk->sk_callback_lock);
1665}
1666
1667static void sock_def_destruct(struct sock *sk)
1668{
a51482bd 1669 kfree(sk->sk_protinfo);
1da177e4
LT		 1670}
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))
8d8ad9d7 1676 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4
LT		 1677}
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);
97fc2f08
CL		 1701#ifdef CONFIG_NET_DMA
1702 skb_queue_head_init(&sk->sk_async_wait_queue);
1703#endif
1da177e4
LT		 1704
1705 sk->sk_send_head = NULL;
1706
1707 init_timer(&sk->sk_timer);
4ec93edb 1708
1da177e4
LT		 1709 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
e71a4783 1717 if (sock) {
1da177e4
LT		 1718 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);
443aef0e
PZ		 1726 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]);
1da177e4
LT		 1729
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
b7aa0bf7 1747 sk->sk_stamp = ktime_set(-1L, -1L);
1da177e4
LT		 1748
1749 atomic_set(&sk->sk_refcnt, 1);
33c732c3 1750 atomic_set(&sk->sk_drops, 0);
1da177e4
LT		 1751}
1752
b5606c2d 1753void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 1754{
1755 might_sleep();
a5b5bb9a 1756 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 1757 if (sk->sk_lock.owned)
1da177e4 1758 __lock_sock(sk);
d2e9117c 1759 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 1760 spin_unlock(&sk->sk_lock.slock);
1761 /*
1762 * The sk_lock has mutex_lock() semantics here:
1763 */
fcc70d5f 1764 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 1765 local_bh_enable();
1da177e4
LT		 1766}
1767
fcc70d5f 1768EXPORT_SYMBOL(lock_sock_nested);
1da177e4 1769
b5606c2d 1770void release_sock(struct sock *sk)
1da177e4 1771{
a5b5bb9a
IM		 1772 /*
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);
1da177e4
LT		 1778 if (sk->sk_backlog.tail)
1779 __release_sock(sk);
d2e9117c 1780 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 1781 if (waitqueue_active(&sk->sk_lock.wq))
1782 wake_up(&sk->sk_lock.wq);
1783 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 1784}
1785EXPORT_SYMBOL(release_sock);
1786
1787int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 1788{
b7aa0bf7 1789 struct timeval tv;
1da177e4
LT		 1790 if (!sock_flag(sk, SOCK_TIMESTAMP))
1791 sock_enable_timestamp(sk);
b7aa0bf7
ED		 1792 tv = ktime_to_timeval(sk->sk_stamp);
1793 if (tv.tv_sec == -1)
1da177e4 1794 return -ENOENT;
b7aa0bf7
ED		 1795 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;
4ec93edb 1800}
1da177e4
LT		 1801EXPORT_SYMBOL(sock_get_timestamp);
1802
ae40eb1e
ED		 1803int 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
1da177e4 1819void sock_enable_timestamp(struct sock *sk)
4ec93edb
YH		 1820{
1821 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1da177e4
LT		 1822 sock_set_flag(sk, SOCK_TIMESTAMP);
1823 net_enable_timestamp();
1824 }
1825}
1da177e4
LT		 1826
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
3fdadf7d 1844#ifdef CONFIG_COMPAT
543d9cfe
ACM		 1845int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1846 char __user *optval, int __user *optlen)
3fdadf7d
DM		 1847{
1848 struct sock *sk = sock->sk;
1849
1e51f951 1850 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 1851 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1852 optval, optlen);
3fdadf7d
DM		 1853 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1854}
1855EXPORT_SYMBOL(compat_sock_common_getsockopt);
1856#endif
1857
1da177e4
LT		 1858int 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
3fdadf7d 1887#ifdef CONFIG_COMPAT
543d9cfe
ACM		 1888int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1889 char __user *optval, int optlen)
3fdadf7d
DM		 1890{
1891 struct sock *sk = sock->sk;
1892
543d9cfe
ACM		 1893 if (sk->sk_prot->compat_setsockopt != NULL)
1894 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1895 optval, optlen);
3fdadf7d
DM		 1896 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1897}
1898EXPORT_SYMBOL(compat_sock_common_setsockopt);
1899#endif
1900
1da177e4
LT		 1901void 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
e6848976 1932 sk_refcnt_debug_release(sk);
1da177e4
LT		 1933 sock_put(sk);
1934}
1935
1936EXPORT_SYMBOL(sk_common_release);
1937
1938static DEFINE_RWLOCK(proto_list_lock);
1939static LIST_HEAD(proto_list);
1940
b733c007
PE		 1941int proto_register(struct proto *prot, int alloc_slab)
1942{
1943 char *request_sock_slab_name = NULL;
1944 char *timewait_sock_slab_name;
1945
65f76517 1946 if (sock_prot_inuse_init(prot) != 0) {
ebb53d75 1947 printk(KERN_CRIT "%s: Can't alloc inuse counters!\n", prot->name);
b733c007 1948 goto out;
ebb53d75 1949 }
b733c007 1950
1da177e4
LT		 1951 if (alloc_slab) {
1952 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
20c2df83 1953 SLAB_HWCACHE_ALIGN, NULL);
1da177e4
LT		 1954
1955 if (prot->slab == NULL) {
1956 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1957 prot->name);
286ab3d4 1958 goto out_free_inuse;
1da177e4 1959 }
2e6599cb
ACM		 1960
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,
20c2df83 1971 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 1972
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 }
8feaf0c0 1979
6d6ee43e 1980 if (prot->twsk_prot != NULL) {
8feaf0c0
ACM		 1981 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);
6d6ee43e
ACM		 1989 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,
20c2df83 1993 NULL);
6d6ee43e 1994 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 1995 goto out_free_timewait_sock_slab_name;
1996 }
1da177e4
LT		 1997 }
1998
2a278051 1999 write_lock(&proto_list_lock);
1da177e4 2000 list_add(&prot->node, &proto_list);
1da177e4 2001 write_unlock(&proto_list_lock);
b733c007
PE		 2002 return 0;
2003
8feaf0c0
ACM		 2004out_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 }
2e6599cb
ACM		 2011out_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;
286ab3d4 2016out_free_inuse:
65f76517 2017 sock_prot_inuse_free(prot);
b733c007
PE		 2018out:
2019 return -ENOBUFS;
1da177e4
LT		 2020}
2021
2022EXPORT_SYMBOL(proto_register);
2023
2024void proto_unregister(struct proto *prot)
2025{
2026 write_lock(&proto_list_lock);
0a3f4358
PM		 2027 list_del(&prot->node);
2028 write_unlock(&proto_list_lock);
1da177e4 2029
65f76517 2030 sock_prot_inuse_free(prot);
ebb53d75 2031
1da177e4
LT		 2032 if (prot->slab != NULL) {
2033 kmem_cache_destroy(prot->slab);
2034 prot->slab = NULL;
2035 }
2036
2e6599cb
ACM		 2037 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
6d6ee43e
ACM		 2045 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2046 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
8feaf0c0 2047
6d6ee43e 2048 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
8feaf0c0 2049 kfree(name);
6d6ee43e 2050 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2051 }
1da177e4
LT		 2052}
2053
2054EXPORT_SYMBOL(proto_unregister);
2055
2056#ifdef CONFIG_PROC_FS
1da177e4 2057static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2058 __acquires(proto_list_lock)
1da177e4
LT		 2059{
2060 read_lock(&proto_list_lock);
60f0438a 2061 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2062}
2063
2064static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2065{
60f0438a 2066 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2067}
2068
2069static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2070 __releases(proto_list_lock)
1da177e4
LT		 2071{
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{
60f0438a 2115 if (v == &proto_list)
1da177e4
LT		 2116 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
60f0438a 2127 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2128 return 0;
2129}
2130
f690808e 2131static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2132 .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
9a32144e 2143static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2144 .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 */
457c4cbc 2154 return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1da177e4
LT		 2155}
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);
1da177e4 2189EXPORT_SYMBOL(sysctl_optmem_max);
Linux 6.x block layer and io_uring tree(s)