projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge git://github.com/Jkirsher/net-next
[deliverable/linux.git] / net / core / sock.c
CommitLineData
1da177e4
LT		 1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
4ec93edb 35 * code. The ACK stuff can wait and needs major
1da177e4
LT		 36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
4fc268d2 92#include <linux/capability.h>
1da177e4
LT		 93#include <linux/errno.h>
94#include <linux/types.h>
95#include <linux/socket.h>
96#include <linux/in.h>
97#include <linux/kernel.h>
1da177e4
LT		 98#include <linux/module.h>
99#include <linux/proc_fs.h>
100#include <linux/seq_file.h>
101#include <linux/sched.h>
102#include <linux/timer.h>
103#include <linux/string.h>
104#include <linux/sockios.h>
105#include <linux/net.h>
106#include <linux/mm.h>
107#include <linux/slab.h>
108#include <linux/interrupt.h>
109#include <linux/poll.h>
110#include <linux/tcp.h>
111#include <linux/init.h>
a1f8e7f7 112#include <linux/highmem.h>
3f551f94 113#include <linux/user_namespace.h>
1da177e4
LT		 114
115#include <asm/uaccess.h>
116#include <asm/system.h>
117
118#include <linux/netdevice.h>
119#include <net/protocol.h>
120#include <linux/skbuff.h>
457c4cbc 121#include <net/net_namespace.h>
2e6599cb 122#include <net/request_sock.h>
1da177e4 123#include <net/sock.h>
20d49473 124#include <linux/net_tstamp.h>
1da177e4
LT		 125#include <net/xfrm.h>
126#include <linux/ipsec.h>
f8451725 127#include <net/cls_cgroup.h>
1da177e4
LT		 128
129#include <linux/filter.h>
130
3847ce32
SM		 131#include <trace/events/sock.h>
132
1da177e4
LT		 133#ifdef CONFIG_INET
134#include <net/tcp.h>
135#endif
136
da21f24d
IM		 137/*
138 * Each address family might have different locking rules, so we have
139 * one slock key per address family:
140 */
a5b5bb9a
IM		 141static struct lock_class_key af_family_keys[AF_MAX];
142static struct lock_class_key af_family_slock_keys[AF_MAX];
143
a5b5bb9a
IM		 144/*
145 * Make lock validator output more readable. (we pre-construct these
146 * strings build-time, so that runtime initialization of socket
147 * locks is fast):
148 */
36cbd3dc 149static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 164};
36cbd3dc 165static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 175 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 179 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 180};
36cbd3dc 181static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 191 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 195 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 196};
da21f24d
IM		 197
198/*
199 * sk_callback_lock locking rules are per-address-family,
200 * so split the lock classes by using a per-AF key:
201 */
202static struct lock_class_key af_callback_keys[AF_MAX];
203
1da177e4
LT		 204/* Take into consideration the size of the struct sk_buff overhead in the
205 * determination of these values, since that is non-constant across
206 * platforms. This makes socket queueing behavior and performance
207 * not depend upon such differences.
208 */
209#define _SK_MEM_PACKETS 256
210#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
211#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213
214/* Run time adjustable parameters. */
ab32ea5d
BH		 215__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 219
25985edc 220/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 221int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 222EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 223
f8451725
HX		 224#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225int net_cls_subsys_id = -1;
226EXPORT_SYMBOL_GPL(net_cls_subsys_id);
227#endif
228
1da177e4
LT		 229static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
230{
231 struct timeval tv;
232
233 if (optlen < sizeof(tv))
234 return -EINVAL;
235 if (copy_from_user(&tv, optval, sizeof(tv)))
236 return -EFAULT;
ba78073e
VA		 237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
238 return -EDOM;
1da177e4 239
ba78073e 240 if (tv.tv_sec < 0) {
6f11df83
AM		 241 static int warned __read_mostly;
242
ba78073e 243 *timeo_p = 0;
50aab54f 244 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 245 warned++;
246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247 "tries to set negative timeout\n",
ba25f9dc 248 current->comm, task_pid_nr(current));
50aab54f 249 }
ba78073e
VA		 250 return 0;
251 }
1da177e4
LT		 252 *timeo_p = MAX_SCHEDULE_TIMEOUT;
253 if (tv.tv_sec == 0 && tv.tv_usec == 0)
254 return 0;
255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
257 return 0;
258}
259
260static void sock_warn_obsolete_bsdism(const char *name)
261{
262 static int warned;
263 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 264 if (strcmp(warncomm, current->comm) && warned < 5) {
265 strcpy(warncomm, current->comm);
1da177e4
LT		 266 printk(KERN_WARNING "process `%s' is using obsolete "
267 "%s SO_BSDCOMPAT\n", warncomm, name);
268 warned++;
269 }
270}
271
20d49473 272static void sock_disable_timestamp(struct sock *sk, int flag)
4ec93edb 273{
20d49473
PO		 274 if (sock_flag(sk, flag)) {
275 sock_reset_flag(sk, flag);
276 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278 net_disable_timestamp();
279 }
1da177e4
LT		 280 }
281}
282
283
f0088a50
DV		 284int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
285{
766e9037 286 int err;
f0088a50 287 int skb_len;
3b885787
NH		 288 unsigned long flags;
289 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 290
9ee6b7f1 291 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
f0088a50
DV		 292 number of warnings when compiling with -W --ANK
293 */
294 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
295 (unsigned)sk->sk_rcvbuf) {
766e9037 296 atomic_inc(&sk->sk_drops);
3847ce32 297 trace_sock_rcvqueue_full(sk, skb);
766e9037 298 return -ENOMEM;
f0088a50
DV		 299 }
300
fda9ef5d 301 err = sk_filter(sk, skb);
f0088a50 302 if (err)
766e9037 303 return err;
f0088a50 304
3ab224be 305 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 306 atomic_inc(&sk->sk_drops);
307 return -ENOBUFS;
3ab224be
HA		 308 }
309
f0088a50
DV		 310 skb->dev = NULL;
311 skb_set_owner_r(skb, sk);
49ad9599 312
f0088a50
DV		 313 /* Cache the SKB length before we tack it onto the receive
314 * queue. Once it is added it no longer belongs to us and
315 * may be freed by other threads of control pulling packets
316 * from the queue.
317 */
318 skb_len = skb->len;
319
7fee226a
ED		 320 /* we escape from rcu protected region, make sure we dont leak
321 * a norefcounted dst
322 */
323 skb_dst_force(skb);
324
3b885787
NH		 325 spin_lock_irqsave(&list->lock, flags);
326 skb->dropcount = atomic_read(&sk->sk_drops);
327 __skb_queue_tail(list, skb);
328 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 329
330 if (!sock_flag(sk, SOCK_DEAD))
331 sk->sk_data_ready(sk, skb_len);
766e9037 332 return 0;
f0088a50
DV		 333}
334EXPORT_SYMBOL(sock_queue_rcv_skb);
335
58a5a7b9 336int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 337{
338 int rc = NET_RX_SUCCESS;
339
fda9ef5d 340 if (sk_filter(sk, skb))
f0088a50
DV		 341 goto discard_and_relse;
342
343 skb->dev = NULL;
344
c377411f
ED		 345 if (sk_rcvqueues_full(sk, skb)) {
346 atomic_inc(&sk->sk_drops);
347 goto discard_and_relse;
348 }
58a5a7b9
ACM		 349 if (nested)
350 bh_lock_sock_nested(sk);
351 else
352 bh_lock_sock(sk);
a5b5bb9a
IM		 353 if (!sock_owned_by_user(sk)) {
354 /*
355 * trylock + unlock semantics:
356 */
357 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
358
c57943a1 359 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 360
361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 362 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 363 bh_unlock_sock(sk);
364 atomic_inc(&sk->sk_drops);
365 goto discard_and_relse;
366 }
367
f0088a50
DV		 368 bh_unlock_sock(sk);
369out:
370 sock_put(sk);
371 return rc;
372discard_and_relse:
373 kfree_skb(skb);
374 goto out;
375}
376EXPORT_SYMBOL(sk_receive_skb);
377
ea94ff3b
KK		 378void sk_reset_txq(struct sock *sk)
379{
380 sk_tx_queue_clear(sk);
381}
382EXPORT_SYMBOL(sk_reset_txq);
383
f0088a50
DV		 384struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
385{
b6c6712a 386 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 387
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 389 sk_tx_queue_clear(sk);
a9b3cd7f 390 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 391 dst_release(dst);
392 return NULL;
393 }
394
395 return dst;
396}
397EXPORT_SYMBOL(__sk_dst_check);
398
399struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
400{
401 struct dst_entry *dst = sk_dst_get(sk);
402
403 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
404 sk_dst_reset(sk);
405 dst_release(dst);
406 return NULL;
407 }
408
409 return dst;
410}
411EXPORT_SYMBOL(sk_dst_check);
412
4878809f
DM		 413static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
414{
415 int ret = -ENOPROTOOPT;
416#ifdef CONFIG_NETDEVICES
3b1e0a65 417 struct net *net = sock_net(sk);
4878809f
DM		 418 char devname[IFNAMSIZ];
419 int index;
420
421 /* Sorry... */
422 ret = -EPERM;
423 if (!capable(CAP_NET_RAW))
424 goto out;
425
426 ret = -EINVAL;
427 if (optlen < 0)
428 goto out;
429
430 /* Bind this socket to a particular device like "eth0",
431 * as specified in the passed interface name. If the
432 * name is "" or the option length is zero the socket
433 * is not bound.
434 */
435 if (optlen > IFNAMSIZ - 1)
436 optlen = IFNAMSIZ - 1;
437 memset(devname, 0, sizeof(devname));
438
439 ret = -EFAULT;
440 if (copy_from_user(devname, optval, optlen))
441 goto out;
442
000ba2e4
DM		 443 index = 0;
444 if (devname[0] != '\0') {
bf8e56bf 445 struct net_device *dev;
4878809f 446
bf8e56bf
ED		 447 rcu_read_lock();
448 dev = dev_get_by_name_rcu(net, devname);
449 if (dev)
450 index = dev->ifindex;
451 rcu_read_unlock();
4878809f
DM		 452 ret = -ENODEV;
453 if (!dev)
454 goto out;
4878809f
DM		 455 }
456
457 lock_sock(sk);
458 sk->sk_bound_dev_if = index;
459 sk_dst_reset(sk);
460 release_sock(sk);
461
462 ret = 0;
463
464out:
465#endif
466
467 return ret;
468}
469
c0ef877b
PE		 470static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
471{
472 if (valbool)
473 sock_set_flag(sk, bit);
474 else
475 sock_reset_flag(sk, bit);
476}
477
1da177e4
LT		 478/*
479 * This is meant for all protocols to use and covers goings on
480 * at the socket level. Everything here is generic.
481 */
482
483int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 484 char __user *optval, unsigned int optlen)
1da177e4 485{
2a91525c 486 struct sock *sk = sock->sk;
1da177e4
LT		 487 int val;
488 int valbool;
489 struct linger ling;
490 int ret = 0;
4ec93edb 491
1da177e4
LT		 492 /*
493 * Options without arguments
494 */
495
4878809f
DM		 496 if (optname == SO_BINDTODEVICE)
497 return sock_bindtodevice(sk, optval, optlen);
498
e71a4783
SH		 499 if (optlen < sizeof(int))
500 return -EINVAL;
4ec93edb 501
1da177e4
LT		 502 if (get_user(val, (int __user *)optval))
503 return -EFAULT;
4ec93edb 504
2a91525c 505 valbool = val ? 1 : 0;
1da177e4
LT		 506
507 lock_sock(sk);
508
2a91525c 509 switch (optname) {
e71a4783 510 case SO_DEBUG:
2a91525c 511 if (val && !capable(CAP_NET_ADMIN))
e71a4783 512 ret = -EACCES;
2a91525c 513 else
c0ef877b 514 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 515 break;
516 case SO_REUSEADDR:
517 sk->sk_reuse = valbool;
518 break;
519 case SO_TYPE:
49c794e9 520 case SO_PROTOCOL:
0d6038ee 521 case SO_DOMAIN:
e71a4783
SH		 522 case SO_ERROR:
523 ret = -ENOPROTOOPT;
524 break;
525 case SO_DONTROUTE:
c0ef877b 526 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 527 break;
528 case SO_BROADCAST:
529 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
530 break;
531 case SO_SNDBUF:
532 /* Don't error on this BSD doesn't and if you think
533 about it this is right. Otherwise apps have to
534 play 'guess the biggest size' games. RCVBUF/SNDBUF
535 are treated in BSD as hints */
536
537 if (val > sysctl_wmem_max)
538 val = sysctl_wmem_max;
b0573dea 539set_sndbuf:
e71a4783
SH		 540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
541 if ((val * 2) < SOCK_MIN_SNDBUF)
542 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
543 else
544 sk->sk_sndbuf = val * 2;
1da177e4 545
e71a4783
SH		 546 /*
547 * Wake up sending tasks if we
548 * upped the value.
549 */
550 sk->sk_write_space(sk);
551 break;
1da177e4 552
e71a4783
SH		 553 case SO_SNDBUFFORCE:
554 if (!capable(CAP_NET_ADMIN)) {
555 ret = -EPERM;
556 break;
557 }
558 goto set_sndbuf;
b0573dea 559
e71a4783
SH		 560 case SO_RCVBUF:
561 /* Don't error on this BSD doesn't and if you think
562 about it this is right. Otherwise apps have to
563 play 'guess the biggest size' games. RCVBUF/SNDBUF
564 are treated in BSD as hints */
4ec93edb 565
e71a4783
SH		 566 if (val > sysctl_rmem_max)
567 val = sysctl_rmem_max;
b0573dea 568set_rcvbuf:
e71a4783
SH		 569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
570 /*
571 * We double it on the way in to account for
572 * "struct sk_buff" etc. overhead. Applications
573 * assume that the SO_RCVBUF setting they make will
574 * allow that much actual data to be received on that
575 * socket.
576 *
577 * Applications are unaware that "struct sk_buff" and
578 * other overheads allocate from the receive buffer
579 * during socket buffer allocation.
580 *
581 * And after considering the possible alternatives,
582 * returning the value we actually used in getsockopt
583 * is the most desirable behavior.
584 */
585 if ((val * 2) < SOCK_MIN_RCVBUF)
586 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
587 else
588 sk->sk_rcvbuf = val * 2;
589 break;
590
591 case SO_RCVBUFFORCE:
592 if (!capable(CAP_NET_ADMIN)) {
593 ret = -EPERM;
1da177e4 594 break;
e71a4783
SH		 595 }
596 goto set_rcvbuf;
1da177e4 597
e71a4783 598 case SO_KEEPALIVE:
1da177e4 599#ifdef CONFIG_INET
e71a4783
SH		 600 if (sk->sk_protocol == IPPROTO_TCP)
601 tcp_set_keepalive(sk, valbool);
1da177e4 602#endif
e71a4783
SH		 603 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
604 break;
605
606 case SO_OOBINLINE:
607 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
608 break;
609
610 case SO_NO_CHECK:
611 sk->sk_no_check = valbool;
612 break;
613
614 case SO_PRIORITY:
615 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
616 sk->sk_priority = val;
617 else
618 ret = -EPERM;
619 break;
620
621 case SO_LINGER:
622 if (optlen < sizeof(ling)) {
623 ret = -EINVAL; /* 1003.1g */
1da177e4 624 break;
e71a4783 625 }
2a91525c 626 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 627 ret = -EFAULT;
1da177e4 628 break;
e71a4783
SH		 629 }
630 if (!ling.l_onoff)
631 sock_reset_flag(sk, SOCK_LINGER);
632 else {
1da177e4 633#if (BITS_PER_LONG == 32)
e71a4783
SH		 634 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
635 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 636 else
e71a4783
SH		 637#endif
638 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
639 sock_set_flag(sk, SOCK_LINGER);
640 }
641 break;
642
643 case SO_BSDCOMPAT:
644 sock_warn_obsolete_bsdism("setsockopt");
645 break;
646
647 case SO_PASSCRED:
648 if (valbool)
649 set_bit(SOCK_PASSCRED, &sock->flags);
650 else
651 clear_bit(SOCK_PASSCRED, &sock->flags);
652 break;
653
654 case SO_TIMESTAMP:
92f37fd2 655 case SO_TIMESTAMPNS:
e71a4783 656 if (valbool) {
92f37fd2
ED		 657 if (optname == SO_TIMESTAMP)
658 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
659 else
660 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 661 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 662 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 663 } else {
e71a4783 664 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 }
e71a4783
SH		 667 break;
668
20d49473
PO		 669 case SO_TIMESTAMPING:
670 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 671 ret = -EINVAL;
20d49473
PO		 672 break;
673 }
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
675 val & SOF_TIMESTAMPING_TX_HARDWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
677 val & SOF_TIMESTAMPING_TX_SOFTWARE);
678 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
679 val & SOF_TIMESTAMPING_RX_HARDWARE);
680 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
681 sock_enable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 else
684 sock_disable_timestamp(sk,
685 SOCK_TIMESTAMPING_RX_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
687 val & SOF_TIMESTAMPING_SOFTWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
689 val & SOF_TIMESTAMPING_SYS_HARDWARE);
690 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
691 val & SOF_TIMESTAMPING_RAW_HARDWARE);
692 break;
693
e71a4783
SH		 694 case SO_RCVLOWAT:
695 if (val < 0)
696 val = INT_MAX;
697 sk->sk_rcvlowat = val ? : 1;
698 break;
699
700 case SO_RCVTIMEO:
701 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
702 break;
703
704 case SO_SNDTIMEO:
705 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
706 break;
1da177e4 707
e71a4783
SH		 708 case SO_ATTACH_FILTER:
709 ret = -EINVAL;
710 if (optlen == sizeof(struct sock_fprog)) {
711 struct sock_fprog fprog;
1da177e4 712
e71a4783
SH		 713 ret = -EFAULT;
714 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 715 break;
e71a4783
SH		 716
717 ret = sk_attach_filter(&fprog, sk);
718 }
719 break;
720
721 case SO_DETACH_FILTER:
55b33325 722 ret = sk_detach_filter(sk);
e71a4783 723 break;
1da177e4 724
e71a4783
SH		 725 case SO_PASSSEC:
726 if (valbool)
727 set_bit(SOCK_PASSSEC, &sock->flags);
728 else
729 clear_bit(SOCK_PASSSEC, &sock->flags);
730 break;
4a19ec58
LAT		 731 case SO_MARK:
732 if (!capable(CAP_NET_ADMIN))
733 ret = -EPERM;
2a91525c 734 else
4a19ec58 735 sk->sk_mark = val;
4a19ec58 736 break;
877ce7c1 737
1da177e4
LT		 738 /* We implement the SO_SNDLOWAT etc to
739 not be settable (1003.1g 5.3) */
3b885787
NH		 740 case SO_RXQ_OVFL:
741 if (valbool)
742 sock_set_flag(sk, SOCK_RXQ_OVFL);
743 else
744 sock_reset_flag(sk, SOCK_RXQ_OVFL);
745 break;
e71a4783
SH		 746 default:
747 ret = -ENOPROTOOPT;
748 break;
4ec93edb 749 }
1da177e4
LT		 750 release_sock(sk);
751 return ret;
752}
2a91525c 753EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 754
755
3f551f94
EB		 756void cred_to_ucred(struct pid *pid, const struct cred *cred,
757 struct ucred *ucred)
758{
759 ucred->pid = pid_vnr(pid);
760 ucred->uid = ucred->gid = -1;
761 if (cred) {
762 struct user_namespace *current_ns = current_user_ns();
763
764 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
765 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
766 }
767}
3924773a 768EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 769
1da177e4
LT		 770int sock_getsockopt(struct socket *sock, int level, int optname,
771 char __user *optval, int __user *optlen)
772{
773 struct sock *sk = sock->sk;
4ec93edb 774
e71a4783 775 union {
4ec93edb
YH		 776 int val;
777 struct linger ling;
1da177e4
LT		 778 struct timeval tm;
779 } v;
4ec93edb 780
4d0392be 781 int lv = sizeof(int);
1da177e4 782 int len;
4ec93edb 783
e71a4783 784 if (get_user(len, optlen))
4ec93edb 785 return -EFAULT;
e71a4783 786 if (len < 0)
1da177e4 787 return -EINVAL;
4ec93edb 788
50fee1de 789 memset(&v, 0, sizeof(v));
df0bca04 790
2a91525c 791 switch (optname) {
e71a4783
SH		 792 case SO_DEBUG:
793 v.val = sock_flag(sk, SOCK_DBG);
794 break;
795
796 case SO_DONTROUTE:
797 v.val = sock_flag(sk, SOCK_LOCALROUTE);
798 break;
799
800 case SO_BROADCAST:
801 v.val = !!sock_flag(sk, SOCK_BROADCAST);
802 break;
803
804 case SO_SNDBUF:
805 v.val = sk->sk_sndbuf;
806 break;
807
808 case SO_RCVBUF:
809 v.val = sk->sk_rcvbuf;
810 break;
811
812 case SO_REUSEADDR:
813 v.val = sk->sk_reuse;
814 break;
815
816 case SO_KEEPALIVE:
817 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
818 break;
819
820 case SO_TYPE:
821 v.val = sk->sk_type;
822 break;
823
49c794e9
JE		 824 case SO_PROTOCOL:
825 v.val = sk->sk_protocol;
826 break;
827
0d6038ee
JE		 828 case SO_DOMAIN:
829 v.val = sk->sk_family;
830 break;
831
e71a4783
SH		 832 case SO_ERROR:
833 v.val = -sock_error(sk);
2a91525c 834 if (v.val == 0)
e71a4783
SH		 835 v.val = xchg(&sk->sk_err_soft, 0);
836 break;
837
838 case SO_OOBINLINE:
839 v.val = !!sock_flag(sk, SOCK_URGINLINE);
840 break;
841
842 case SO_NO_CHECK:
843 v.val = sk->sk_no_check;
844 break;
845
846 case SO_PRIORITY:
847 v.val = sk->sk_priority;
848 break;
849
850 case SO_LINGER:
851 lv = sizeof(v.ling);
852 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
853 v.ling.l_linger = sk->sk_lingertime / HZ;
854 break;
855
856 case SO_BSDCOMPAT:
857 sock_warn_obsolete_bsdism("getsockopt");
858 break;
859
860 case SO_TIMESTAMP:
92f37fd2
ED		 861 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
862 !sock_flag(sk, SOCK_RCVTSTAMPNS);
863 break;
864
865 case SO_TIMESTAMPNS:
866 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 867 break;
868
20d49473
PO		 869 case SO_TIMESTAMPING:
870 v.val = 0;
871 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
872 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
873 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
874 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
876 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
878 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
880 v.val |= SOF_TIMESTAMPING_SOFTWARE;
881 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
882 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
883 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
884 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
885 break;
886
e71a4783 887 case SO_RCVTIMEO:
2a91525c 888 lv = sizeof(struct timeval);
e71a4783
SH		 889 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
890 v.tm.tv_sec = 0;
891 v.tm.tv_usec = 0;
892 } else {
893 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
894 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
895 }
896 break;
897
898 case SO_SNDTIMEO:
2a91525c 899 lv = sizeof(struct timeval);
e71a4783
SH		 900 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
901 v.tm.tv_sec = 0;
902 v.tm.tv_usec = 0;
903 } else {
904 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
905 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
906 }
907 break;
1da177e4 908
e71a4783
SH		 909 case SO_RCVLOWAT:
910 v.val = sk->sk_rcvlowat;
911 break;
1da177e4 912
e71a4783 913 case SO_SNDLOWAT:
2a91525c 914 v.val = 1;
e71a4783 915 break;
1da177e4 916
e71a4783
SH		 917 case SO_PASSCRED:
918 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
919 break;
1da177e4 920
e71a4783 921 case SO_PEERCRED:
109f6e39
EB		 922 {
923 struct ucred peercred;
924 if (len > sizeof(peercred))
925 len = sizeof(peercred);
926 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
927 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 928 return -EFAULT;
929 goto lenout;
109f6e39 930 }
1da177e4 931
e71a4783
SH		 932 case SO_PEERNAME:
933 {
934 char address[128];
935
936 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
937 return -ENOTCONN;
938 if (lv < len)
939 return -EINVAL;
940 if (copy_to_user(optval, address, len))
941 return -EFAULT;
942 goto lenout;
943 }
1da177e4 944
e71a4783
SH		 945 /* Dubious BSD thing... Probably nobody even uses it, but
946 * the UNIX standard wants it for whatever reason... -DaveM
947 */
948 case SO_ACCEPTCONN:
949 v.val = sk->sk_state == TCP_LISTEN;
950 break;
1da177e4 951
e71a4783
SH		 952 case SO_PASSSEC:
953 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
954 break;
877ce7c1 955
e71a4783
SH		 956 case SO_PEERSEC:
957 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 958
4a19ec58
LAT		 959 case SO_MARK:
960 v.val = sk->sk_mark;
961 break;
962
3b885787
NH		 963 case SO_RXQ_OVFL:
964 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
965 break;
966
e71a4783
SH		 967 default:
968 return -ENOPROTOOPT;
1da177e4 969 }
e71a4783 970
1da177e4
LT		 971 if (len > lv)
972 len = lv;
973 if (copy_to_user(optval, &v, len))
974 return -EFAULT;
975lenout:
4ec93edb
YH		 976 if (put_user(len, optlen))
977 return -EFAULT;
978 return 0;
1da177e4
LT		 979}
980
a5b5bb9a
IM		 981/*
982 * Initialize an sk_lock.
983 *
984 * (We also register the sk_lock with the lock validator.)
985 */
b6f99a21 986static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 987{
ed07536e
PZ		 988 sock_lock_init_class_and_name(sk,
989 af_family_slock_key_strings[sk->sk_family],
990 af_family_slock_keys + sk->sk_family,
991 af_family_key_strings[sk->sk_family],
992 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 993}
994
4dc6dc71
ED		 995/*
996 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
997 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 998 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 999 */
f1a6c4da
PE		 1000static void sock_copy(struct sock *nsk, const struct sock *osk)
1001{
1002#ifdef CONFIG_SECURITY_NETWORK
1003 void *sptr = nsk->sk_security;
1004#endif
68835aba
ED		 1005 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1006
1007 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1008 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1009
f1a6c4da
PE		 1010#ifdef CONFIG_SECURITY_NETWORK
1011 nsk->sk_security = sptr;
1012 security_sk_clone(osk, nsk);
1013#endif
1014}
1015
fcbdf09d
OP		 1016/*
1017 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1018 * un-modified. Special care is taken when initializing object to zero.
1019 */
1020static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1021{
1022 if (offsetof(struct sock, sk_node.next) != 0)
1023 memset(sk, 0, offsetof(struct sock, sk_node.next));
1024 memset(&sk->sk_node.pprev, 0,
1025 size - offsetof(struct sock, sk_node.pprev));
1026}
1027
1028void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1029{
1030 unsigned long nulls1, nulls2;
1031
1032 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1033 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1034 if (nulls1 > nulls2)
1035 swap(nulls1, nulls2);
1036
1037 if (nulls1 != 0)
1038 memset((char *)sk, 0, nulls1);
1039 memset((char *)sk + nulls1 + sizeof(void *), 0,
1040 nulls2 - nulls1 - sizeof(void *));
1041 memset((char *)sk + nulls2 + sizeof(void *), 0,
1042 size - nulls2 - sizeof(void *));
1043}
1044EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1045
2e4afe7b
PE		 1046static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1047 int family)
c308c1b2
PE		 1048{
1049 struct sock *sk;
1050 struct kmem_cache *slab;
1051
1052 slab = prot->slab;
e912b114
ED		 1053 if (slab != NULL) {
1054 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1055 if (!sk)
1056 return sk;
1057 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1058 if (prot->clear_sk)
1059 prot->clear_sk(sk, prot->obj_size);
1060 else
1061 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1062 }
fcbdf09d 1063 } else
c308c1b2
PE		 1064 sk = kmalloc(prot->obj_size, priority);
1065
2e4afe7b 1066 if (sk != NULL) {
a98b65a3
VN		 1067 kmemcheck_annotate_bitfield(sk, flags);
1068
2e4afe7b
PE		 1069 if (security_sk_alloc(sk, family, priority))
1070 goto out_free;
1071
1072 if (!try_module_get(prot->owner))
1073 goto out_free_sec;
e022f0b4 1074 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1075 }
1076
c308c1b2 1077 return sk;
2e4afe7b
PE		 1078
1079out_free_sec:
1080 security_sk_free(sk);
1081out_free:
1082 if (slab != NULL)
1083 kmem_cache_free(slab, sk);
1084 else
1085 kfree(sk);
1086 return NULL;
c308c1b2
PE		 1087}
1088
1089static void sk_prot_free(struct proto *prot, struct sock *sk)
1090{
1091 struct kmem_cache *slab;
2e4afe7b 1092 struct module *owner;
c308c1b2 1093
2e4afe7b 1094 owner = prot->owner;
c308c1b2 1095 slab = prot->slab;
2e4afe7b
PE		 1096
1097 security_sk_free(sk);
c308c1b2
PE		 1098 if (slab != NULL)
1099 kmem_cache_free(slab, sk);
1100 else
1101 kfree(sk);
2e4afe7b 1102 module_put(owner);
c308c1b2
PE		 1103}
1104
f8451725
HX		 1105#ifdef CONFIG_CGROUPS
1106void sock_update_classid(struct sock *sk)
1107{
1144182a 1108 u32 classid;
f8451725 1109
1144182a
PM		 1110 rcu_read_lock(); /* doing current task, which cannot vanish. */
1111 classid = task_cls_classid(current);
1112 rcu_read_unlock();
f8451725
HX		 1113 if (classid && classid != sk->sk_classid)
1114 sk->sk_classid = classid;
1115}
82862742 1116EXPORT_SYMBOL(sock_update_classid);
f8451725
HX		 1117#endif
1118
1da177e4
LT		 1119/**
1120 * sk_alloc - All socket objects are allocated here
c4ea43c5 1121 * @net: the applicable net namespace
4dc3b16b
PP		 1122 * @family: protocol family
1123 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1124 * @prot: struct proto associated with this new sock instance
1da177e4 1125 */
1b8d7ae4 1126struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1127 struct proto *prot)
1da177e4 1128{
c308c1b2 1129 struct sock *sk;
1da177e4 1130
154adbc8 1131 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1132 if (sk) {
154adbc8
PE		 1133 sk->sk_family = family;
1134 /*
1135 * See comment in struct sock definition to understand
1136 * why we need sk_prot_creator -acme
1137 */
1138 sk->sk_prot = sk->sk_prot_creator = prot;
1139 sock_lock_init(sk);
3b1e0a65 1140 sock_net_set(sk, get_net(net));
d66ee058 1141 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1142
1143 sock_update_classid(sk);
1da177e4 1144 }
a79af59e 1145
2e4afe7b 1146 return sk;
1da177e4 1147}
2a91525c 1148EXPORT_SYMBOL(sk_alloc);
1da177e4 1149
2b85a34e 1150static void __sk_free(struct sock *sk)
1da177e4
LT		 1151{
1152 struct sk_filter *filter;
1da177e4
LT		 1153
1154 if (sk->sk_destruct)
1155 sk->sk_destruct(sk);
1156
a898def2
PM		 1157 filter = rcu_dereference_check(sk->sk_filter,
1158 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1159 if (filter) {
309dd5fc 1160 sk_filter_uncharge(sk, filter);
a9b3cd7f 1161 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1162 }
1163
20d49473
PO		 1164 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1165 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1da177e4
LT		 1166
1167 if (atomic_read(&sk->sk_omem_alloc))
1168 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1169 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1170
109f6e39
EB		 1171 if (sk->sk_peer_cred)
1172 put_cred(sk->sk_peer_cred);
1173 put_pid(sk->sk_peer_pid);
3b1e0a65 1174 put_net(sock_net(sk));
c308c1b2 1175 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1176}
2b85a34e
ED		 1177
1178void sk_free(struct sock *sk)
1179{
1180 /*
25985edc 1181 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1182 * some packets are still in some tx queue.
1183 * If not null, sock_wfree() will call __sk_free(sk) later
1184 */
1185 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1186 __sk_free(sk);
1187}
2a91525c 1188EXPORT_SYMBOL(sk_free);
1da177e4 1189
edf02087 1190/*
25985edc
LDM		 1191 * Last sock_put should drop reference to sk->sk_net. It has already
1192 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1193 * is not an option.
25985edc 1194 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1195 * destroy it in the context of init_net.
1196 */
1197void sk_release_kernel(struct sock *sk)
1198{
1199 if (sk == NULL || sk->sk_socket == NULL)
1200 return;
1201
1202 sock_hold(sk);
1203 sock_release(sk->sk_socket);
65a18ec5 1204 release_net(sock_net(sk));
3b1e0a65 1205 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1206 sock_put(sk);
1207}
45af1754 1208EXPORT_SYMBOL(sk_release_kernel);
edf02087 1209
dd0fc66f 1210struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
87d11ceb 1211{
8fd1d178 1212 struct sock *newsk;
87d11ceb 1213
8fd1d178 1214 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1215 if (newsk != NULL) {
1216 struct sk_filter *filter;
1217
892c141e 1218 sock_copy(newsk, sk);
87d11ceb
ACM		 1219
1220 /* SANITY */
3b1e0a65 1221 get_net(sock_net(newsk));
87d11ceb
ACM		 1222 sk_node_init(&newsk->sk_node);
1223 sock_lock_init(newsk);
1224 bh_lock_sock(newsk);
fa438ccf 1225 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1226 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1227
1228 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1229 /*
1230 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1231 */
1232 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1233 atomic_set(&newsk->sk_omem_alloc, 0);
1234 skb_queue_head_init(&newsk->sk_receive_queue);
1235 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1236#ifdef CONFIG_NET_DMA
1237 skb_queue_head_init(&newsk->sk_async_wait_queue);
1238#endif
87d11ceb 1239
b6c6712a 1240 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1241 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1242 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1243 af_callback_keys + newsk->sk_family,
1244 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1245
1246 newsk->sk_dst_cache = NULL;
1247 newsk->sk_wmem_queued = 0;
1248 newsk->sk_forward_alloc = 0;
1249 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1250 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1251
1252 sock_reset_flag(newsk, SOCK_DONE);
1253 skb_queue_head_init(&newsk->sk_error_queue);
1254
0d7da9dd 1255 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1256 if (filter != NULL)
1257 sk_filter_charge(newsk, filter);
1258
1259 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1260 /* It is still raw copy of parent, so invalidate
1261 * destructor and make plain sk_free() */
1262 newsk->sk_destruct = NULL;
1263 sk_free(newsk);
1264 newsk = NULL;
1265 goto out;
1266 }
1267
1268 newsk->sk_err = 0;
1269 newsk->sk_priority = 0;
4dc6dc71
ED		 1270 /*
1271 * Before updating sk_refcnt, we must commit prior changes to memory
1272 * (Documentation/RCU/rculist_nulls.txt for details)
1273 */
1274 smp_wmb();
87d11ceb
ACM		 1275 atomic_set(&newsk->sk_refcnt, 2);
1276
1277 /*
1278 * Increment the counter in the same struct proto as the master
1279 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1280 * is the same as sk->sk_prot->socks, as this field was copied
1281 * with memcpy).
1282 *
1283 * This _changes_ the previous behaviour, where
1284 * tcp_create_openreq_child always was incrementing the
1285 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1286 * to be taken into account in all callers. -acme
1287 */
1288 sk_refcnt_debug_inc(newsk);
972692e0 1289 sk_set_socket(newsk, NULL);
43815482 1290 newsk->sk_wq = NULL;
87d11ceb
ACM		 1291
1292 if (newsk->sk_prot->sockets_allocated)
1748376b 1293 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560
OP		 1294
1295 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1296 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1297 net_enable_timestamp();
87d11ceb
ACM		 1298 }
1299out:
1300 return newsk;
1301}
87d11ceb
ACM		 1302EXPORT_SYMBOL_GPL(sk_clone);
1303
9958089a
AK		 1304void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1305{
1306 __sk_dst_set(sk, dst);
1307 sk->sk_route_caps = dst->dev->features;
1308 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1309 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1310 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1311 if (sk_can_gso(sk)) {
82cc1a7a 1312 if (dst->header_len) {
9958089a 1313 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1314 } else {
9958089a 1315 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1316 sk->sk_gso_max_size = dst->dev->gso_max_size;
1317 }
9958089a
AK		 1318 }
1319}
1320EXPORT_SYMBOL_GPL(sk_setup_caps);
1321
1da177e4
LT		 1322void __init sk_init(void)
1323{
4481374c 1324 if (totalram_pages <= 4096) {
1da177e4
LT		 1325 sysctl_wmem_max = 32767;
1326 sysctl_rmem_max = 32767;
1327 sysctl_wmem_default = 32767;
1328 sysctl_rmem_default = 32767;
4481374c 1329 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1330 sysctl_wmem_max = 131071;
1331 sysctl_rmem_max = 131071;
1332 }
1333}
1334
1335/*
1336 * Simple resource managers for sockets.
1337 */
1338
1339
4ec93edb
YH		 1340/*
1341 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1342 */
1343void sock_wfree(struct sk_buff *skb)
1344{
1345 struct sock *sk = skb->sk;
d99927f4 1346 unsigned int len = skb->truesize;
1da177e4 1347
d99927f4
ED		 1348 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1349 /*
1350 * Keep a reference on sk_wmem_alloc, this will be released
1351 * after sk_write_space() call
1352 */
1353 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1354 sk->sk_write_space(sk);
d99927f4
ED		 1355 len = 1;
1356 }
2b85a34e 1357 /*
d99927f4
ED		 1358 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1359 * could not do because of in-flight packets
2b85a34e 1360 */
d99927f4 1361 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1362 __sk_free(sk);
1da177e4 1363}
2a91525c 1364EXPORT_SYMBOL(sock_wfree);
1da177e4 1365
4ec93edb
YH		 1366/*
1367 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1368 */
1369void sock_rfree(struct sk_buff *skb)
1370{
1371 struct sock *sk = skb->sk;
d361fd59 1372 unsigned int len = skb->truesize;
1da177e4 1373
d361fd59
ED		 1374 atomic_sub(len, &sk->sk_rmem_alloc);
1375 sk_mem_uncharge(sk, len);
1da177e4 1376}
2a91525c 1377EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1378
1379
1380int sock_i_uid(struct sock *sk)
1381{
1382 int uid;
1383
f064af1e 1384 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1385 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1386 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1387 return uid;
1388}
2a91525c 1389EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1390
1391unsigned long sock_i_ino(struct sock *sk)
1392{
1393 unsigned long ino;
1394
f064af1e 1395 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1396 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1397 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1398 return ino;
1399}
2a91525c 1400EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1401
1402/*
1403 * Allocate a skb from the socket's send buffer.
1404 */
86a76caf 1405struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1406 gfp_t priority)
1da177e4
LT		 1407{
1408 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1409 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1410 if (skb) {
1411 skb_set_owner_w(skb, sk);
1412 return skb;
1413 }
1414 }
1415 return NULL;
1416}
2a91525c 1417EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1418
1419/*
1420 * Allocate a skb from the socket's receive buffer.
4ec93edb 1421 */
86a76caf 1422struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1423 gfp_t priority)
1da177e4
LT		 1424{
1425 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1426 struct sk_buff *skb = alloc_skb(size, priority);
1427 if (skb) {
1428 skb_set_owner_r(skb, sk);
1429 return skb;
1430 }
1431 }
1432 return NULL;
1433}
1434
4ec93edb 1435/*
1da177e4 1436 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1437 */
dd0fc66f 1438void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1439{
1440 if ((unsigned)size <= sysctl_optmem_max &&
1441 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1442 void *mem;
1443 /* First do the add, to avoid the race if kmalloc
4ec93edb 1444 * might sleep.
1da177e4
LT		 1445 */
1446 atomic_add(size, &sk->sk_omem_alloc);
1447 mem = kmalloc(size, priority);
1448 if (mem)
1449 return mem;
1450 atomic_sub(size, &sk->sk_omem_alloc);
1451 }
1452 return NULL;
1453}
2a91525c 1454EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1455
1456/*
1457 * Free an option memory block.
1458 */
1459void sock_kfree_s(struct sock *sk, void *mem, int size)
1460{
1461 kfree(mem);
1462 atomic_sub(size, &sk->sk_omem_alloc);
1463}
2a91525c 1464EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1465
1466/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1467 I think, these locks should be removed for datagram sockets.
1468 */
2a91525c 1469static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1470{
1471 DEFINE_WAIT(wait);
1472
1473 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1474 for (;;) {
1475 if (!timeo)
1476 break;
1477 if (signal_pending(current))
1478 break;
1479 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1480 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1481 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1482 break;
1483 if (sk->sk_shutdown & SEND_SHUTDOWN)
1484 break;
1485 if (sk->sk_err)
1486 break;
1487 timeo = schedule_timeout(timeo);
1488 }
aa395145 1489 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1490 return timeo;
1491}
1492
1493
1494/*
1495 * Generic send/receive buffer handlers
1496 */
1497
4cc7f68d
HX		 1498struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1499 unsigned long data_len, int noblock,
1500 int *errcode)
1da177e4
LT		 1501{
1502 struct sk_buff *skb;
7d877f3b 1503 gfp_t gfp_mask;
1da177e4
LT		 1504 long timeo;
1505 int err;
1506
1507 gfp_mask = sk->sk_allocation;
1508 if (gfp_mask & __GFP_WAIT)
1509 gfp_mask |= __GFP_REPEAT;
1510
1511 timeo = sock_sndtimeo(sk, noblock);
1512 while (1) {
1513 err = sock_error(sk);
1514 if (err != 0)
1515 goto failure;
1516
1517 err = -EPIPE;
1518 if (sk->sk_shutdown & SEND_SHUTDOWN)
1519 goto failure;
1520
1521 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1522 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1523 if (skb) {
1524 int npages;
1525 int i;
1526
1527 /* No pages, we're done... */
1528 if (!data_len)
1529 break;
1530
1531 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1532 skb->truesize += data_len;
1533 skb_shinfo(skb)->nr_frags = npages;
1534 for (i = 0; i < npages; i++) {
1535 struct page *page;
1da177e4
LT		 1536
1537 page = alloc_pages(sk->sk_allocation, 0);
1538 if (!page) {
1539 err = -ENOBUFS;
1540 skb_shinfo(skb)->nr_frags = i;
1541 kfree_skb(skb);
1542 goto failure;
1543 }
1544
ea2ab693
IC		 1545 __skb_fill_page_desc(skb, i,
1546 page, 0,
1547 (data_len >= PAGE_SIZE ?
1548 PAGE_SIZE :
1549 data_len));
1da177e4
LT		 1550 data_len -= PAGE_SIZE;
1551 }
1552
1553 /* Full success... */
1554 break;
1555 }
1556 err = -ENOBUFS;
1557 goto failure;
1558 }
1559 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1560 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1561 err = -EAGAIN;
1562 if (!timeo)
1563 goto failure;
1564 if (signal_pending(current))
1565 goto interrupted;
1566 timeo = sock_wait_for_wmem(sk, timeo);
1567 }
1568
1569 skb_set_owner_w(skb, sk);
1570 return skb;
1571
1572interrupted:
1573 err = sock_intr_errno(timeo);
1574failure:
1575 *errcode = err;
1576 return NULL;
1577}
4cc7f68d 1578EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1579
4ec93edb 1580struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1581 int noblock, int *errcode)
1582{
1583 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1584}
2a91525c 1585EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1586
1587static void __lock_sock(struct sock *sk)
f39234d6
NK		 1588 __releases(&sk->sk_lock.slock)
1589 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1590{
1591 DEFINE_WAIT(wait);
1592
e71a4783 1593 for (;;) {
1da177e4
LT		 1594 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1595 TASK_UNINTERRUPTIBLE);
1596 spin_unlock_bh(&sk->sk_lock.slock);
1597 schedule();
1598 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1599 if (!sock_owned_by_user(sk))
1da177e4
LT		 1600 break;
1601 }
1602 finish_wait(&sk->sk_lock.wq, &wait);
1603}
1604
1605static void __release_sock(struct sock *sk)
f39234d6
NK		 1606 __releases(&sk->sk_lock.slock)
1607 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1608{
1609 struct sk_buff *skb = sk->sk_backlog.head;
1610
1611 do {
1612 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1613 bh_unlock_sock(sk);
1614
1615 do {
1616 struct sk_buff *next = skb->next;
1617
7fee226a 1618 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1619 skb->next = NULL;
c57943a1 1620 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1621
1622 /*
1623 * We are in process context here with softirqs
1624 * disabled, use cond_resched_softirq() to preempt.
1625 * This is safe to do because we've taken the backlog
1626 * queue private:
1627 */
1628 cond_resched_softirq();
1629
1630 skb = next;
1631 } while (skb != NULL);
1632
1633 bh_lock_sock(sk);
e71a4783 1634 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1635
1636 /*
1637 * Doing the zeroing here guarantee we can not loop forever
1638 * while a wild producer attempts to flood us.
1639 */
1640 sk->sk_backlog.len = 0;
1da177e4
LT		 1641}
1642
1643/**
1644 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1645 * @sk: sock to wait on
1646 * @timeo: for how long
1da177e4
LT		 1647 *
1648 * Now socket state including sk->sk_err is changed only under lock,
1649 * hence we may omit checks after joining wait queue.
1650 * We check receive queue before schedule() only as optimization;
1651 * it is very likely that release_sock() added new data.
1652 */
1653int sk_wait_data(struct sock *sk, long *timeo)
1654{
1655 int rc;
1656 DEFINE_WAIT(wait);
1657
aa395145 1658 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1659 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1660 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1661 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1662 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1663 return rc;
1664}
1da177e4
LT		 1665EXPORT_SYMBOL(sk_wait_data);
1666
3ab224be
HA		 1667/**
1668 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1669 * @sk: socket
1670 * @size: memory size to allocate
1671 * @kind: allocation type
1672 *
1673 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1674 * rmem allocation. This function assumes that protocols which have
1675 * memory_pressure use sk_wmem_queued as write buffer accounting.
1676 */
1677int __sk_mem_schedule(struct sock *sk, int size, int kind)
1678{
1679 struct proto *prot = sk->sk_prot;
1680 int amt = sk_mem_pages(size);
8d987e5c 1681 long allocated;
3ab224be
HA		 1682
1683 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
8d987e5c 1684 allocated = atomic_long_add_return(amt, prot->memory_allocated);
3ab224be
HA		 1685
1686 /* Under limit. */
1687 if (allocated <= prot->sysctl_mem[0]) {
1688 if (prot->memory_pressure && *prot->memory_pressure)
1689 *prot->memory_pressure = 0;
1690 return 1;
1691 }
1692
1693 /* Under pressure. */
1694 if (allocated > prot->sysctl_mem[1])
1695 if (prot->enter_memory_pressure)
5c52ba17 1696 prot->enter_memory_pressure(sk);
3ab224be
HA		 1697
1698 /* Over hard limit. */
1699 if (allocated > prot->sysctl_mem[2])
1700 goto suppress_allocation;
1701
1702 /* guarantee minimum buffer size under pressure */
1703 if (kind == SK_MEM_RECV) {
1704 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1705 return 1;
1706 } else { /* SK_MEM_SEND */
1707 if (sk->sk_type == SOCK_STREAM) {
1708 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1709 return 1;
1710 } else if (atomic_read(&sk->sk_wmem_alloc) <
1711 prot->sysctl_wmem[0])
1712 return 1;
1713 }
1714
1715 if (prot->memory_pressure) {
1748376b
ED		 1716 int alloc;
1717
1718 if (!*prot->memory_pressure)
1719 return 1;
1720 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1721 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1722 sk_mem_pages(sk->sk_wmem_queued +
1723 atomic_read(&sk->sk_rmem_alloc) +
1724 sk->sk_forward_alloc))
1725 return 1;
1726 }
1727
1728suppress_allocation:
1729
1730 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1731 sk_stream_moderate_sndbuf(sk);
1732
1733 /* Fail only if socket is _under_ its sndbuf.
1734 * In this case we cannot block, so that we have to fail.
1735 */
1736 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1737 return 1;
1738 }
1739
3847ce32
SM		 1740 trace_sock_exceed_buf_limit(sk, prot, allocated);
1741
3ab224be
HA		 1742 /* Alas. Undo changes. */
1743 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
8d987e5c 1744 atomic_long_sub(amt, prot->memory_allocated);
3ab224be
HA		 1745 return 0;
1746}
3ab224be
HA		 1747EXPORT_SYMBOL(__sk_mem_schedule);
1748
1749/**
1750 * __sk_reclaim - reclaim memory_allocated
1751 * @sk: socket
1752 */
1753void __sk_mem_reclaim(struct sock *sk)
1754{
1755 struct proto *prot = sk->sk_prot;
1756
8d987e5c 1757 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1758 prot->memory_allocated);
1759 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1760
1761 if (prot->memory_pressure && *prot->memory_pressure &&
8d987e5c 1762 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
3ab224be
HA		 1763 *prot->memory_pressure = 0;
1764}
3ab224be
HA		 1765EXPORT_SYMBOL(__sk_mem_reclaim);
1766
1767
1da177e4
LT		 1768/*
1769 * Set of default routines for initialising struct proto_ops when
1770 * the protocol does not support a particular function. In certain
1771 * cases where it makes no sense for a protocol to have a "do nothing"
1772 * function, some default processing is provided.
1773 */
1774
1775int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1776{
1777 return -EOPNOTSUPP;
1778}
2a91525c 1779EXPORT_SYMBOL(sock_no_bind);
1da177e4 1780
4ec93edb 1781int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1782 int len, int flags)
1783{
1784 return -EOPNOTSUPP;
1785}
2a91525c 1786EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1787
1788int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1789{
1790 return -EOPNOTSUPP;
1791}
2a91525c 1792EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1793
1794int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1795{
1796 return -EOPNOTSUPP;
1797}
2a91525c 1798EXPORT_SYMBOL(sock_no_accept);
1da177e4 1799
4ec93edb 1800int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1801 int *len, int peer)
1802{
1803 return -EOPNOTSUPP;
1804}
2a91525c 1805EXPORT_SYMBOL(sock_no_getname);
1da177e4 1806
2a91525c 1807unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1808{
1809 return 0;
1810}
2a91525c 1811EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1812
1813int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1814{
1815 return -EOPNOTSUPP;
1816}
2a91525c 1817EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1818
1819int sock_no_listen(struct socket *sock, int backlog)
1820{
1821 return -EOPNOTSUPP;
1822}
2a91525c 1823EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1824
1825int sock_no_shutdown(struct socket *sock, int how)
1826{
1827 return -EOPNOTSUPP;
1828}
2a91525c 1829EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1830
1831int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1832 char __user *optval, unsigned int optlen)
1da177e4
LT		 1833{
1834 return -EOPNOTSUPP;
1835}
2a91525c 1836EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1837
1838int sock_no_getsockopt(struct socket *sock, int level, int optname,
1839 char __user *optval, int __user *optlen)
1840{
1841 return -EOPNOTSUPP;
1842}
2a91525c 1843EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1844
1845int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1846 size_t len)
1847{
1848 return -EOPNOTSUPP;
1849}
2a91525c 1850EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1851
1852int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1853 size_t len, int flags)
1854{
1855 return -EOPNOTSUPP;
1856}
2a91525c 1857EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1858
1859int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1860{
1861 /* Mirror missing mmap method error code */
1862 return -ENODEV;
1863}
2a91525c 1864EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1865
1866ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1867{
1868 ssize_t res;
1869 struct msghdr msg = {.msg_flags = flags};
1870 struct kvec iov;
1871 char *kaddr = kmap(page);
1872 iov.iov_base = kaddr + offset;
1873 iov.iov_len = size;
1874 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1875 kunmap(page);
1876 return res;
1877}
2a91525c 1878EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1879
1880/*
1881 * Default Socket Callbacks
1882 */
1883
1884static void sock_def_wakeup(struct sock *sk)
1885{
43815482
ED		 1886 struct socket_wq *wq;
1887
1888 rcu_read_lock();
1889 wq = rcu_dereference(sk->sk_wq);
1890 if (wq_has_sleeper(wq))
1891 wake_up_interruptible_all(&wq->wait);
1892 rcu_read_unlock();
1da177e4
LT		 1893}
1894
1895static void sock_def_error_report(struct sock *sk)
1896{
43815482
ED		 1897 struct socket_wq *wq;
1898
1899 rcu_read_lock();
1900 wq = rcu_dereference(sk->sk_wq);
1901 if (wq_has_sleeper(wq))
1902 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1903 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1904 rcu_read_unlock();
1da177e4
LT		 1905}
1906
1907static void sock_def_readable(struct sock *sk, int len)
1908{
43815482
ED		 1909 struct socket_wq *wq;
1910
1911 rcu_read_lock();
1912 wq = rcu_dereference(sk->sk_wq);
1913 if (wq_has_sleeper(wq))
2c6607c6 1914 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1915 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1916 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1917 rcu_read_unlock();
1da177e4
LT		 1918}
1919
1920static void sock_def_write_space(struct sock *sk)
1921{
43815482
ED		 1922 struct socket_wq *wq;
1923
1924 rcu_read_lock();
1da177e4
LT		 1925
1926 /* Do not wake up a writer until he can make "significant"
1927 * progress. --DaveM
1928 */
e71a4783 1929 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1930 wq = rcu_dereference(sk->sk_wq);
1931 if (wq_has_sleeper(wq))
1932 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1933 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1934
1935 /* Should agree with poll, otherwise some programs break */
1936 if (sock_writeable(sk))
8d8ad9d7 1937 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1938 }
1939
43815482 1940 rcu_read_unlock();
1da177e4
LT		 1941}
1942
1943static void sock_def_destruct(struct sock *sk)
1944{
a51482bd 1945 kfree(sk->sk_protinfo);
1da177e4
LT		 1946}
1947
1948void sk_send_sigurg(struct sock *sk)
1949{
1950 if (sk->sk_socket && sk->sk_socket->file)
1951 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1952 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1953}
2a91525c 1954EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1955
1956void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1957 unsigned long expires)
1958{
1959 if (!mod_timer(timer, expires))
1960 sock_hold(sk);
1961}
1da177e4
LT		 1962EXPORT_SYMBOL(sk_reset_timer);
1963
1964void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1965{
1966 if (timer_pending(timer) && del_timer(timer))
1967 __sock_put(sk);
1968}
1da177e4
LT		 1969EXPORT_SYMBOL(sk_stop_timer);
1970
1971void sock_init_data(struct socket *sock, struct sock *sk)
1972{
1973 skb_queue_head_init(&sk->sk_receive_queue);
1974 skb_queue_head_init(&sk->sk_write_queue);
1975 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 1976#ifdef CONFIG_NET_DMA
1977 skb_queue_head_init(&sk->sk_async_wait_queue);
1978#endif
1da177e4
LT		 1979
1980 sk->sk_send_head = NULL;
1981
1982 init_timer(&sk->sk_timer);
4ec93edb 1983
1da177e4
LT		 1984 sk->sk_allocation = GFP_KERNEL;
1985 sk->sk_rcvbuf = sysctl_rmem_default;
1986 sk->sk_sndbuf = sysctl_wmem_default;
1987 sk->sk_state = TCP_CLOSE;
972692e0 1988 sk_set_socket(sk, sock);
1da177e4
LT		 1989
1990 sock_set_flag(sk, SOCK_ZAPPED);
1991
e71a4783 1992 if (sock) {
1da177e4 1993 sk->sk_type = sock->type;
43815482 1994 sk->sk_wq = sock->wq;
1da177e4
LT		 1995 sock->sk = sk;
1996 } else
43815482 1997 sk->sk_wq = NULL;
1da177e4 1998
b6c6712a 1999 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2000 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2001 lockdep_set_class_and_name(&sk->sk_callback_lock,
2002 af_callback_keys + sk->sk_family,
2003 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2004
2005 sk->sk_state_change = sock_def_wakeup;
2006 sk->sk_data_ready = sock_def_readable;
2007 sk->sk_write_space = sock_def_write_space;
2008 sk->sk_error_report = sock_def_error_report;
2009 sk->sk_destruct = sock_def_destruct;
2010
2011 sk->sk_sndmsg_page = NULL;
2012 sk->sk_sndmsg_off = 0;
2013
109f6e39
EB		 2014 sk->sk_peer_pid = NULL;
2015 sk->sk_peer_cred = NULL;
1da177e4
LT		 2016 sk->sk_write_pending = 0;
2017 sk->sk_rcvlowat = 1;
2018 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2019 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2020
f37f0afb 2021 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2022
4dc6dc71
ED		 2023 /*
2024 * Before updating sk_refcnt, we must commit prior changes to memory
2025 * (Documentation/RCU/rculist_nulls.txt for details)
2026 */
2027 smp_wmb();
1da177e4 2028 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2029 atomic_set(&sk->sk_drops, 0);
1da177e4 2030}
2a91525c 2031EXPORT_SYMBOL(sock_init_data);
1da177e4 2032
b5606c2d 2033void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2034{
2035 might_sleep();
a5b5bb9a 2036 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2037 if (sk->sk_lock.owned)
1da177e4 2038 __lock_sock(sk);
d2e9117c 2039 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2040 spin_unlock(&sk->sk_lock.slock);
2041 /*
2042 * The sk_lock has mutex_lock() semantics here:
2043 */
fcc70d5f 2044 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2045 local_bh_enable();
1da177e4 2046}
fcc70d5f 2047EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2048
b5606c2d 2049void release_sock(struct sock *sk)
1da177e4 2050{
a5b5bb9a
IM		 2051 /*
2052 * The sk_lock has mutex_unlock() semantics:
2053 */
2054 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2055
2056 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2057 if (sk->sk_backlog.tail)
2058 __release_sock(sk);
d2e9117c 2059 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2060 if (waitqueue_active(&sk->sk_lock.wq))
2061 wake_up(&sk->sk_lock.wq);
2062 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2063}
2064EXPORT_SYMBOL(release_sock);
2065
8a74ad60
ED		 2066/**
2067 * lock_sock_fast - fast version of lock_sock
2068 * @sk: socket
2069 *
2070 * This version should be used for very small section, where process wont block
2071 * return false if fast path is taken
2072 * sk_lock.slock locked, owned = 0, BH disabled
2073 * return true if slow path is taken
2074 * sk_lock.slock unlocked, owned = 1, BH enabled
2075 */
2076bool lock_sock_fast(struct sock *sk)
2077{
2078 might_sleep();
2079 spin_lock_bh(&sk->sk_lock.slock);
2080
2081 if (!sk->sk_lock.owned)
2082 /*
2083 * Note : We must disable BH
2084 */
2085 return false;
2086
2087 __lock_sock(sk);
2088 sk->sk_lock.owned = 1;
2089 spin_unlock(&sk->sk_lock.slock);
2090 /*
2091 * The sk_lock has mutex_lock() semantics here:
2092 */
2093 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2094 local_bh_enable();
2095 return true;
2096}
2097EXPORT_SYMBOL(lock_sock_fast);
2098
1da177e4 2099int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2100{
b7aa0bf7 2101 struct timeval tv;
1da177e4 2102 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2103 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2104 tv = ktime_to_timeval(sk->sk_stamp);
2105 if (tv.tv_sec == -1)
1da177e4 2106 return -ENOENT;
b7aa0bf7
ED		 2107 if (tv.tv_sec == 0) {
2108 sk->sk_stamp = ktime_get_real();
2109 tv = ktime_to_timeval(sk->sk_stamp);
2110 }
2111 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2112}
1da177e4
LT		 2113EXPORT_SYMBOL(sock_get_timestamp);
2114
ae40eb1e
ED		 2115int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2116{
2117 struct timespec ts;
2118 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2119 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2120 ts = ktime_to_timespec(sk->sk_stamp);
2121 if (ts.tv_sec == -1)
2122 return -ENOENT;
2123 if (ts.tv_sec == 0) {
2124 sk->sk_stamp = ktime_get_real();
2125 ts = ktime_to_timespec(sk->sk_stamp);
2126 }
2127 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2128}
2129EXPORT_SYMBOL(sock_get_timestampns);
2130
20d49473 2131void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2132{
20d49473
PO		 2133 if (!sock_flag(sk, flag)) {
2134 sock_set_flag(sk, flag);
2135 /*
2136 * we just set one of the two flags which require net
2137 * time stamping, but time stamping might have been on
2138 * already because of the other one
2139 */
2140 if (!sock_flag(sk,
2141 flag == SOCK_TIMESTAMP ?
2142 SOCK_TIMESTAMPING_RX_SOFTWARE :
2143 SOCK_TIMESTAMP))
2144 net_enable_timestamp();
1da177e4
LT		 2145 }
2146}
1da177e4
LT		 2147
2148/*
2149 * Get a socket option on an socket.
2150 *
2151 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2152 * asynchronous errors should be reported by getsockopt. We assume
2153 * this means if you specify SO_ERROR (otherwise whats the point of it).
2154 */
2155int sock_common_getsockopt(struct socket *sock, int level, int optname,
2156 char __user *optval, int __user *optlen)
2157{
2158 struct sock *sk = sock->sk;
2159
2160 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2161}
1da177e4
LT		 2162EXPORT_SYMBOL(sock_common_getsockopt);
2163
3fdadf7d 2164#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2165int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2166 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2167{
2168 struct sock *sk = sock->sk;
2169
1e51f951 2170 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2171 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2172 optval, optlen);
3fdadf7d
DM		 2173 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2174}
2175EXPORT_SYMBOL(compat_sock_common_getsockopt);
2176#endif
2177
1da177e4
LT		 2178int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2179 struct msghdr *msg, size_t size, int flags)
2180{
2181 struct sock *sk = sock->sk;
2182 int addr_len = 0;
2183 int err;
2184
2185 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2186 flags & ~MSG_DONTWAIT, &addr_len);
2187 if (err >= 0)
2188 msg->msg_namelen = addr_len;
2189 return err;
2190}
1da177e4
LT		 2191EXPORT_SYMBOL(sock_common_recvmsg);
2192
2193/*
2194 * Set socket options on an inet socket.
2195 */
2196int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2197 char __user *optval, unsigned int optlen)
1da177e4
LT		 2198{
2199 struct sock *sk = sock->sk;
2200
2201 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2202}
1da177e4
LT		 2203EXPORT_SYMBOL(sock_common_setsockopt);
2204
3fdadf7d 2205#ifdef CONFIG_COMPAT
543d9cfe 2206int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2207 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2208{
2209 struct sock *sk = sock->sk;
2210
543d9cfe
ACM		 2211 if (sk->sk_prot->compat_setsockopt != NULL)
2212 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2213 optval, optlen);
3fdadf7d
DM		 2214 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2215}
2216EXPORT_SYMBOL(compat_sock_common_setsockopt);
2217#endif
2218
1da177e4
LT		 2219void sk_common_release(struct sock *sk)
2220{
2221 if (sk->sk_prot->destroy)
2222 sk->sk_prot->destroy(sk);
2223
2224 /*
2225 * Observation: when sock_common_release is called, processes have
2226 * no access to socket. But net still has.
2227 * Step one, detach it from networking:
2228 *
2229 * A. Remove from hash tables.
2230 */
2231
2232 sk->sk_prot->unhash(sk);
2233
2234 /*
2235 * In this point socket cannot receive new packets, but it is possible
2236 * that some packets are in flight because some CPU runs receiver and
2237 * did hash table lookup before we unhashed socket. They will achieve
2238 * receive queue and will be purged by socket destructor.
2239 *
2240 * Also we still have packets pending on receive queue and probably,
2241 * our own packets waiting in device queues. sock_destroy will drain
2242 * receive queue, but transmitted packets will delay socket destruction
2243 * until the last reference will be released.
2244 */
2245
2246 sock_orphan(sk);
2247
2248 xfrm_sk_free_policy(sk);
2249
e6848976 2250 sk_refcnt_debug_release(sk);
1da177e4
LT		 2251 sock_put(sk);
2252}
1da177e4
LT		 2253EXPORT_SYMBOL(sk_common_release);
2254
2255static DEFINE_RWLOCK(proto_list_lock);
2256static LIST_HEAD(proto_list);
2257
13ff3d6f
PE		 2258#ifdef CONFIG_PROC_FS
2259#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2260struct prot_inuse {
2261 int val[PROTO_INUSE_NR];
2262};
13ff3d6f
PE		 2263
2264static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2265
2266#ifdef CONFIG_NET_NS
2267void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2268{
d6d9ca0f 2269 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2270}
2271EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2272
2273int sock_prot_inuse_get(struct net *net, struct proto *prot)
2274{
2275 int cpu, idx = prot->inuse_idx;
2276 int res = 0;
2277
2278 for_each_possible_cpu(cpu)
2279 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2280
2281 return res >= 0 ? res : 0;
2282}
2283EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2284
2c8c1e72 2285static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2286{
2287 net->core.inuse = alloc_percpu(struct prot_inuse);
2288 return net->core.inuse ? 0 : -ENOMEM;
2289}
2290
2c8c1e72 2291static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2292{
2293 free_percpu(net->core.inuse);
2294}
2295
2296static struct pernet_operations net_inuse_ops = {
2297 .init = sock_inuse_init_net,
2298 .exit = sock_inuse_exit_net,
2299};
2300
2301static __init int net_inuse_init(void)
2302{
2303 if (register_pernet_subsys(&net_inuse_ops))
2304 panic("Cannot initialize net inuse counters");
2305
2306 return 0;
2307}
2308
2309core_initcall(net_inuse_init);
2310#else
1338d466
PE		 2311static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2312
c29a0bc4 2313void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2314{
d6d9ca0f 2315 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2316}
2317EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2318
c29a0bc4 2319int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2320{
2321 int cpu, idx = prot->inuse_idx;
2322 int res = 0;
2323
2324 for_each_possible_cpu(cpu)
2325 res += per_cpu(prot_inuse, cpu).val[idx];
2326
2327 return res >= 0 ? res : 0;
2328}
2329EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2330#endif
13ff3d6f
PE		 2331
2332static void assign_proto_idx(struct proto *prot)
2333{
2334 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2335
2336 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2337 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2338 return;
2339 }
2340
2341 set_bit(prot->inuse_idx, proto_inuse_idx);
2342}
2343
2344static void release_proto_idx(struct proto *prot)
2345{
2346 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2347 clear_bit(prot->inuse_idx, proto_inuse_idx);
2348}
2349#else
2350static inline void assign_proto_idx(struct proto *prot)
2351{
2352}
2353
2354static inline void release_proto_idx(struct proto *prot)
2355{
2356}
2357#endif
2358
b733c007
PE		 2359int proto_register(struct proto *prot, int alloc_slab)
2360{
1da177e4
LT		 2361 if (alloc_slab) {
2362 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2363 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2364 NULL);
1da177e4
LT		 2365
2366 if (prot->slab == NULL) {
2367 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2368 prot->name);
60e7663d 2369 goto out;
1da177e4 2370 }
2e6599cb
ACM		 2371
2372 if (prot->rsk_prot != NULL) {
faf23422 2373 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2374 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2375 goto out_free_sock_slab;
2376
7e56b5d6 2377 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2378 prot->rsk_prot->obj_size, 0,
20c2df83 2379 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2380
2381 if (prot->rsk_prot->slab == NULL) {
2382 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2383 prot->name);
2384 goto out_free_request_sock_slab_name;
2385 }
2386 }
8feaf0c0 2387
6d6ee43e 2388 if (prot->twsk_prot != NULL) {
faf23422 2389 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2390
7e56b5d6 2391 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2392 goto out_free_request_sock_slab;
2393
6d6ee43e 2394 prot->twsk_prot->twsk_slab =
7e56b5d6 2395 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2396 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2397 0,
2398 SLAB_HWCACHE_ALIGN |
2399 prot->slab_flags,
20c2df83 2400 NULL);
6d6ee43e 2401 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2402 goto out_free_timewait_sock_slab_name;
2403 }
1da177e4
LT		 2404 }
2405
2a278051 2406 write_lock(&proto_list_lock);
1da177e4 2407 list_add(&prot->node, &proto_list);
13ff3d6f 2408 assign_proto_idx(prot);
1da177e4 2409 write_unlock(&proto_list_lock);
b733c007
PE		 2410 return 0;
2411
8feaf0c0 2412out_free_timewait_sock_slab_name:
7e56b5d6 2413 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2414out_free_request_sock_slab:
2415 if (prot->rsk_prot && prot->rsk_prot->slab) {
2416 kmem_cache_destroy(prot->rsk_prot->slab);
2417 prot->rsk_prot->slab = NULL;
2418 }
2e6599cb 2419out_free_request_sock_slab_name:
72150e9b
DC		 2420 if (prot->rsk_prot)
2421 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2422out_free_sock_slab:
2423 kmem_cache_destroy(prot->slab);
2424 prot->slab = NULL;
b733c007
PE		 2425out:
2426 return -ENOBUFS;
1da177e4 2427}
1da177e4
LT		 2428EXPORT_SYMBOL(proto_register);
2429
2430void proto_unregister(struct proto *prot)
2431{
2432 write_lock(&proto_list_lock);
13ff3d6f 2433 release_proto_idx(prot);
0a3f4358
PM		 2434 list_del(&prot->node);
2435 write_unlock(&proto_list_lock);
1da177e4
LT		 2436
2437 if (prot->slab != NULL) {
2438 kmem_cache_destroy(prot->slab);
2439 prot->slab = NULL;
2440 }
2441
2e6599cb 2442 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2443 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2444 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2445 prot->rsk_prot->slab = NULL;
2446 }
2447
6d6ee43e 2448 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2449 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2450 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2451 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2452 }
1da177e4 2453}
1da177e4
LT		 2454EXPORT_SYMBOL(proto_unregister);
2455
2456#ifdef CONFIG_PROC_FS
1da177e4 2457static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2458 __acquires(proto_list_lock)
1da177e4
LT		 2459{
2460 read_lock(&proto_list_lock);
60f0438a 2461 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2462}
2463
2464static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2465{
60f0438a 2466 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2467}
2468
2469static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2470 __releases(proto_list_lock)
1da177e4
LT		 2471{
2472 read_unlock(&proto_list_lock);
2473}
2474
2475static char proto_method_implemented(const void *method)
2476{
2477 return method == NULL ? 'n' : 'y';
2478}
2479
2480static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2481{
8d987e5c 2482 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2483 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2484 proto->name,
2485 proto->obj_size,
14e943db 2486 sock_prot_inuse_get(seq_file_net(seq), proto),
8d987e5c 2487 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
1da177e4
LT		 2488 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2489 proto->max_header,
2490 proto->slab == NULL ? "no" : "yes",
2491 module_name(proto->owner),
2492 proto_method_implemented(proto->close),
2493 proto_method_implemented(proto->connect),
2494 proto_method_implemented(proto->disconnect),
2495 proto_method_implemented(proto->accept),
2496 proto_method_implemented(proto->ioctl),
2497 proto_method_implemented(proto->init),
2498 proto_method_implemented(proto->destroy),
2499 proto_method_implemented(proto->shutdown),
2500 proto_method_implemented(proto->setsockopt),
2501 proto_method_implemented(proto->getsockopt),
2502 proto_method_implemented(proto->sendmsg),
2503 proto_method_implemented(proto->recvmsg),
2504 proto_method_implemented(proto->sendpage),
2505 proto_method_implemented(proto->bind),
2506 proto_method_implemented(proto->backlog_rcv),
2507 proto_method_implemented(proto->hash),
2508 proto_method_implemented(proto->unhash),
2509 proto_method_implemented(proto->get_port),
2510 proto_method_implemented(proto->enter_memory_pressure));
2511}
2512
2513static int proto_seq_show(struct seq_file *seq, void *v)
2514{
60f0438a 2515 if (v == &proto_list)
1da177e4
LT		 2516 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2517 "protocol",
2518 "size",
2519 "sockets",
2520 "memory",
2521 "press",
2522 "maxhdr",
2523 "slab",
2524 "module",
2525 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2526 else
60f0438a 2527 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2528 return 0;
2529}
2530
f690808e 2531static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2532 .start = proto_seq_start,
2533 .next = proto_seq_next,
2534 .stop = proto_seq_stop,
2535 .show = proto_seq_show,
2536};
2537
2538static int proto_seq_open(struct inode *inode, struct file *file)
2539{
14e943db
ED		 2540 return seq_open_net(inode, file, &proto_seq_ops,
2541 sizeof(struct seq_net_private));
1da177e4
LT		 2542}
2543
9a32144e 2544static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2545 .owner = THIS_MODULE,
2546 .open = proto_seq_open,
2547 .read = seq_read,
2548 .llseek = seq_lseek,
14e943db
ED		 2549 .release = seq_release_net,
2550};
2551
2552static __net_init int proto_init_net(struct net *net)
2553{
2554 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2555 return -ENOMEM;
2556
2557 return 0;
2558}
2559
2560static __net_exit void proto_exit_net(struct net *net)
2561{
2562 proc_net_remove(net, "protocols");
2563}
2564
2565
2566static __net_initdata struct pernet_operations proto_net_ops = {
2567 .init = proto_init_net,
2568 .exit = proto_exit_net,
1da177e4
LT		 2569};
2570
2571static int __init proto_init(void)
2572{
14e943db 2573 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2574}
2575
2576subsys_initcall(proto_init);
2577
2578#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.814982 seconds and 5 git commands to generate.