projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
rps: fix insufficient bounds checking in store_rps_dev_flow_table_cnt()
[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 *
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
87fb4b7b 210#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 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 740 case SO_RXQ_OVFL:
8083f0fc 741 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 742 break;
e71a4783
SH		 743 default:
744 ret = -ENOPROTOOPT;
745 break;
4ec93edb 746 }
1da177e4
LT		 747 release_sock(sk);
748 return ret;
749}
2a91525c 750EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 751
752
3f551f94
EB		 753void cred_to_ucred(struct pid *pid, const struct cred *cred,
754 struct ucred *ucred)
755{
756 ucred->pid = pid_vnr(pid);
757 ucred->uid = ucred->gid = -1;
758 if (cred) {
759 struct user_namespace *current_ns = current_user_ns();
760
761 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
762 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
763 }
764}
3924773a 765EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 766
1da177e4
LT		 767int sock_getsockopt(struct socket *sock, int level, int optname,
768 char __user *optval, int __user *optlen)
769{
770 struct sock *sk = sock->sk;
4ec93edb 771
e71a4783 772 union {
4ec93edb
YH		 773 int val;
774 struct linger ling;
1da177e4
LT		 775 struct timeval tm;
776 } v;
4ec93edb 777
4d0392be 778 int lv = sizeof(int);
1da177e4 779 int len;
4ec93edb 780
e71a4783 781 if (get_user(len, optlen))
4ec93edb 782 return -EFAULT;
e71a4783 783 if (len < 0)
1da177e4 784 return -EINVAL;
4ec93edb 785
50fee1de 786 memset(&v, 0, sizeof(v));
df0bca04 787
2a91525c 788 switch (optname) {
e71a4783
SH		 789 case SO_DEBUG:
790 v.val = sock_flag(sk, SOCK_DBG);
791 break;
792
793 case SO_DONTROUTE:
794 v.val = sock_flag(sk, SOCK_LOCALROUTE);
795 break;
796
797 case SO_BROADCAST:
798 v.val = !!sock_flag(sk, SOCK_BROADCAST);
799 break;
800
801 case SO_SNDBUF:
802 v.val = sk->sk_sndbuf;
803 break;
804
805 case SO_RCVBUF:
806 v.val = sk->sk_rcvbuf;
807 break;
808
809 case SO_REUSEADDR:
810 v.val = sk->sk_reuse;
811 break;
812
813 case SO_KEEPALIVE:
814 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
815 break;
816
817 case SO_TYPE:
818 v.val = sk->sk_type;
819 break;
820
49c794e9
JE		 821 case SO_PROTOCOL:
822 v.val = sk->sk_protocol;
823 break;
824
0d6038ee
JE		 825 case SO_DOMAIN:
826 v.val = sk->sk_family;
827 break;
828
e71a4783
SH		 829 case SO_ERROR:
830 v.val = -sock_error(sk);
2a91525c 831 if (v.val == 0)
e71a4783
SH		 832 v.val = xchg(&sk->sk_err_soft, 0);
833 break;
834
835 case SO_OOBINLINE:
836 v.val = !!sock_flag(sk, SOCK_URGINLINE);
837 break;
838
839 case SO_NO_CHECK:
840 v.val = sk->sk_no_check;
841 break;
842
843 case SO_PRIORITY:
844 v.val = sk->sk_priority;
845 break;
846
847 case SO_LINGER:
848 lv = sizeof(v.ling);
849 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
850 v.ling.l_linger = sk->sk_lingertime / HZ;
851 break;
852
853 case SO_BSDCOMPAT:
854 sock_warn_obsolete_bsdism("getsockopt");
855 break;
856
857 case SO_TIMESTAMP:
92f37fd2
ED		 858 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
859 !sock_flag(sk, SOCK_RCVTSTAMPNS);
860 break;
861
862 case SO_TIMESTAMPNS:
863 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 864 break;
865
20d49473
PO		 866 case SO_TIMESTAMPING:
867 v.val = 0;
868 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
869 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
870 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
871 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
872 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
873 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
874 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
875 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
876 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
877 v.val |= SOF_TIMESTAMPING_SOFTWARE;
878 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
879 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
882 break;
883
e71a4783 884 case SO_RCVTIMEO:
2a91525c 885 lv = sizeof(struct timeval);
e71a4783
SH		 886 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
887 v.tm.tv_sec = 0;
888 v.tm.tv_usec = 0;
889 } else {
890 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
891 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892 }
893 break;
894
895 case SO_SNDTIMEO:
2a91525c 896 lv = sizeof(struct timeval);
e71a4783
SH		 897 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
898 v.tm.tv_sec = 0;
899 v.tm.tv_usec = 0;
900 } else {
901 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
902 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903 }
904 break;
1da177e4 905
e71a4783
SH		 906 case SO_RCVLOWAT:
907 v.val = sk->sk_rcvlowat;
908 break;
1da177e4 909
e71a4783 910 case SO_SNDLOWAT:
2a91525c 911 v.val = 1;
e71a4783 912 break;
1da177e4 913
e71a4783
SH		 914 case SO_PASSCRED:
915 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916 break;
1da177e4 917
e71a4783 918 case SO_PEERCRED:
109f6e39
EB		 919 {
920 struct ucred peercred;
921 if (len > sizeof(peercred))
922 len = sizeof(peercred);
923 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
924 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 925 return -EFAULT;
926 goto lenout;
109f6e39 927 }
1da177e4 928
e71a4783
SH		 929 case SO_PEERNAME:
930 {
931 char address[128];
932
933 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
934 return -ENOTCONN;
935 if (lv < len)
936 return -EINVAL;
937 if (copy_to_user(optval, address, len))
938 return -EFAULT;
939 goto lenout;
940 }
1da177e4 941
e71a4783
SH		 942 /* Dubious BSD thing... Probably nobody even uses it, but
943 * the UNIX standard wants it for whatever reason... -DaveM
944 */
945 case SO_ACCEPTCONN:
946 v.val = sk->sk_state == TCP_LISTEN;
947 break;
1da177e4 948
e71a4783
SH		 949 case SO_PASSSEC:
950 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
951 break;
877ce7c1 952
e71a4783
SH		 953 case SO_PEERSEC:
954 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 955
4a19ec58
LAT		 956 case SO_MARK:
957 v.val = sk->sk_mark;
958 break;
959
3b885787
NH		 960 case SO_RXQ_OVFL:
961 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
962 break;
963
e71a4783
SH		 964 default:
965 return -ENOPROTOOPT;
1da177e4 966 }
e71a4783 967
1da177e4
LT		 968 if (len > lv)
969 len = lv;
970 if (copy_to_user(optval, &v, len))
971 return -EFAULT;
972lenout:
4ec93edb
YH		 973 if (put_user(len, optlen))
974 return -EFAULT;
975 return 0;
1da177e4
LT		 976}
977
a5b5bb9a
IM		 978/*
979 * Initialize an sk_lock.
980 *
981 * (We also register the sk_lock with the lock validator.)
982 */
b6f99a21 983static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 984{
ed07536e
PZ		 985 sock_lock_init_class_and_name(sk,
986 af_family_slock_key_strings[sk->sk_family],
987 af_family_slock_keys + sk->sk_family,
988 af_family_key_strings[sk->sk_family],
989 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 990}
991
4dc6dc71
ED		 992/*
993 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
994 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 995 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 996 */
f1a6c4da
PE		 997static void sock_copy(struct sock *nsk, const struct sock *osk)
998{
999#ifdef CONFIG_SECURITY_NETWORK
1000 void *sptr = nsk->sk_security;
1001#endif
68835aba
ED		 1002 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1003
1004 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1005 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1006
f1a6c4da
PE		 1007#ifdef CONFIG_SECURITY_NETWORK
1008 nsk->sk_security = sptr;
1009 security_sk_clone(osk, nsk);
1010#endif
1011}
1012
fcbdf09d
OP		 1013/*
1014 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1015 * un-modified. Special care is taken when initializing object to zero.
1016 */
1017static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1018{
1019 if (offsetof(struct sock, sk_node.next) != 0)
1020 memset(sk, 0, offsetof(struct sock, sk_node.next));
1021 memset(&sk->sk_node.pprev, 0,
1022 size - offsetof(struct sock, sk_node.pprev));
1023}
1024
1025void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1026{
1027 unsigned long nulls1, nulls2;
1028
1029 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1030 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1031 if (nulls1 > nulls2)
1032 swap(nulls1, nulls2);
1033
1034 if (nulls1 != 0)
1035 memset((char *)sk, 0, nulls1);
1036 memset((char *)sk + nulls1 + sizeof(void *), 0,
1037 nulls2 - nulls1 - sizeof(void *));
1038 memset((char *)sk + nulls2 + sizeof(void *), 0,
1039 size - nulls2 - sizeof(void *));
1040}
1041EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1042
2e4afe7b
PE		 1043static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1044 int family)
c308c1b2
PE		 1045{
1046 struct sock *sk;
1047 struct kmem_cache *slab;
1048
1049 slab = prot->slab;
e912b114
ED		 1050 if (slab != NULL) {
1051 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1052 if (!sk)
1053 return sk;
1054 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1055 if (prot->clear_sk)
1056 prot->clear_sk(sk, prot->obj_size);
1057 else
1058 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1059 }
fcbdf09d 1060 } else
c308c1b2
PE		 1061 sk = kmalloc(prot->obj_size, priority);
1062
2e4afe7b 1063 if (sk != NULL) {
a98b65a3
VN		 1064 kmemcheck_annotate_bitfield(sk, flags);
1065
2e4afe7b
PE		 1066 if (security_sk_alloc(sk, family, priority))
1067 goto out_free;
1068
1069 if (!try_module_get(prot->owner))
1070 goto out_free_sec;
e022f0b4 1071 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1072 }
1073
c308c1b2 1074 return sk;
2e4afe7b
PE		 1075
1076out_free_sec:
1077 security_sk_free(sk);
1078out_free:
1079 if (slab != NULL)
1080 kmem_cache_free(slab, sk);
1081 else
1082 kfree(sk);
1083 return NULL;
c308c1b2
PE		 1084}
1085
1086static void sk_prot_free(struct proto *prot, struct sock *sk)
1087{
1088 struct kmem_cache *slab;
2e4afe7b 1089 struct module *owner;
c308c1b2 1090
2e4afe7b 1091 owner = prot->owner;
c308c1b2 1092 slab = prot->slab;
2e4afe7b
PE		 1093
1094 security_sk_free(sk);
c308c1b2
PE		 1095 if (slab != NULL)
1096 kmem_cache_free(slab, sk);
1097 else
1098 kfree(sk);
2e4afe7b 1099 module_put(owner);
c308c1b2
PE		 1100}
1101
f8451725
HX		 1102#ifdef CONFIG_CGROUPS
1103void sock_update_classid(struct sock *sk)
1104{
1144182a 1105 u32 classid;
f8451725 1106
1144182a
PM		 1107 rcu_read_lock(); /* doing current task, which cannot vanish. */
1108 classid = task_cls_classid(current);
1109 rcu_read_unlock();
f8451725
HX		 1110 if (classid && classid != sk->sk_classid)
1111 sk->sk_classid = classid;
1112}
82862742 1113EXPORT_SYMBOL(sock_update_classid);
f8451725
HX		 1114#endif
1115
1da177e4
LT		 1116/**
1117 * sk_alloc - All socket objects are allocated here
c4ea43c5 1118 * @net: the applicable net namespace
4dc3b16b
PP		 1119 * @family: protocol family
1120 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1121 * @prot: struct proto associated with this new sock instance
1da177e4 1122 */
1b8d7ae4 1123struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1124 struct proto *prot)
1da177e4 1125{
c308c1b2 1126 struct sock *sk;
1da177e4 1127
154adbc8 1128 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1129 if (sk) {
154adbc8
PE		 1130 sk->sk_family = family;
1131 /*
1132 * See comment in struct sock definition to understand
1133 * why we need sk_prot_creator -acme
1134 */
1135 sk->sk_prot = sk->sk_prot_creator = prot;
1136 sock_lock_init(sk);
3b1e0a65 1137 sock_net_set(sk, get_net(net));
d66ee058 1138 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1139
1140 sock_update_classid(sk);
1da177e4 1141 }
a79af59e 1142
2e4afe7b 1143 return sk;
1da177e4 1144}
2a91525c 1145EXPORT_SYMBOL(sk_alloc);
1da177e4 1146
2b85a34e 1147static void __sk_free(struct sock *sk)
1da177e4
LT		 1148{
1149 struct sk_filter *filter;
1da177e4
LT		 1150
1151 if (sk->sk_destruct)
1152 sk->sk_destruct(sk);
1153
a898def2
PM		 1154 filter = rcu_dereference_check(sk->sk_filter,
1155 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1156 if (filter) {
309dd5fc 1157 sk_filter_uncharge(sk, filter);
a9b3cd7f 1158 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1159 }
1160
20d49473
PO		 1161 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1162 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1da177e4
LT		 1163
1164 if (atomic_read(&sk->sk_omem_alloc))
1165 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1166 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1167
109f6e39
EB		 1168 if (sk->sk_peer_cred)
1169 put_cred(sk->sk_peer_cred);
1170 put_pid(sk->sk_peer_pid);
3b1e0a65 1171 put_net(sock_net(sk));
c308c1b2 1172 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1173}
2b85a34e
ED		 1174
1175void sk_free(struct sock *sk)
1176{
1177 /*
25985edc 1178 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1179 * some packets are still in some tx queue.
1180 * If not null, sock_wfree() will call __sk_free(sk) later
1181 */
1182 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1183 __sk_free(sk);
1184}
2a91525c 1185EXPORT_SYMBOL(sk_free);
1da177e4 1186
edf02087 1187/*
25985edc
LDM		 1188 * Last sock_put should drop reference to sk->sk_net. It has already
1189 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1190 * is not an option.
25985edc 1191 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1192 * destroy it in the context of init_net.
1193 */
1194void sk_release_kernel(struct sock *sk)
1195{
1196 if (sk == NULL || sk->sk_socket == NULL)
1197 return;
1198
1199 sock_hold(sk);
1200 sock_release(sk->sk_socket);
65a18ec5 1201 release_net(sock_net(sk));
3b1e0a65 1202 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1203 sock_put(sk);
1204}
45af1754 1205EXPORT_SYMBOL(sk_release_kernel);
edf02087 1206
dd0fc66f 1207struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
87d11ceb 1208{
8fd1d178 1209 struct sock *newsk;
87d11ceb 1210
8fd1d178 1211 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1212 if (newsk != NULL) {
1213 struct sk_filter *filter;
1214
892c141e 1215 sock_copy(newsk, sk);
87d11ceb
ACM		 1216
1217 /* SANITY */
3b1e0a65 1218 get_net(sock_net(newsk));
87d11ceb
ACM		 1219 sk_node_init(&newsk->sk_node);
1220 sock_lock_init(newsk);
1221 bh_lock_sock(newsk);
fa438ccf 1222 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1223 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1224
1225 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1226 /*
1227 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1228 */
1229 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1230 atomic_set(&newsk->sk_omem_alloc, 0);
1231 skb_queue_head_init(&newsk->sk_receive_queue);
1232 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1233#ifdef CONFIG_NET_DMA
1234 skb_queue_head_init(&newsk->sk_async_wait_queue);
1235#endif
87d11ceb 1236
b6c6712a 1237 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1238 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1239 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1240 af_callback_keys + newsk->sk_family,
1241 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1242
1243 newsk->sk_dst_cache = NULL;
1244 newsk->sk_wmem_queued = 0;
1245 newsk->sk_forward_alloc = 0;
1246 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1247 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1248
1249 sock_reset_flag(newsk, SOCK_DONE);
1250 skb_queue_head_init(&newsk->sk_error_queue);
1251
0d7da9dd 1252 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1253 if (filter != NULL)
1254 sk_filter_charge(newsk, filter);
1255
1256 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1257 /* It is still raw copy of parent, so invalidate
1258 * destructor and make plain sk_free() */
1259 newsk->sk_destruct = NULL;
b0691c8e 1260 bh_unlock_sock(newsk);
87d11ceb
ACM		 1261 sk_free(newsk);
1262 newsk = NULL;
1263 goto out;
1264 }
1265
1266 newsk->sk_err = 0;
1267 newsk->sk_priority = 0;
4dc6dc71
ED		 1268 /*
1269 * Before updating sk_refcnt, we must commit prior changes to memory
1270 * (Documentation/RCU/rculist_nulls.txt for details)
1271 */
1272 smp_wmb();
87d11ceb
ACM		 1273 atomic_set(&newsk->sk_refcnt, 2);
1274
1275 /*
1276 * Increment the counter in the same struct proto as the master
1277 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1278 * is the same as sk->sk_prot->socks, as this field was copied
1279 * with memcpy).
1280 *
1281 * This _changes_ the previous behaviour, where
1282 * tcp_create_openreq_child always was incrementing the
1283 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1284 * to be taken into account in all callers. -acme
1285 */
1286 sk_refcnt_debug_inc(newsk);
972692e0 1287 sk_set_socket(newsk, NULL);
43815482 1288 newsk->sk_wq = NULL;
87d11ceb
ACM		 1289
1290 if (newsk->sk_prot->sockets_allocated)
1748376b 1291 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560
OP		 1292
1293 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1294 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1295 net_enable_timestamp();
87d11ceb
ACM		 1296 }
1297out:
1298 return newsk;
1299}
87d11ceb
ACM		 1300EXPORT_SYMBOL_GPL(sk_clone);
1301
9958089a
AK		 1302void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1303{
1304 __sk_dst_set(sk, dst);
1305 sk->sk_route_caps = dst->dev->features;
1306 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1307 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1308 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1309 if (sk_can_gso(sk)) {
82cc1a7a 1310 if (dst->header_len) {
9958089a 1311 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1312 } else {
9958089a 1313 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1314 sk->sk_gso_max_size = dst->dev->gso_max_size;
1315 }
9958089a
AK		 1316 }
1317}
1318EXPORT_SYMBOL_GPL(sk_setup_caps);
1319
1da177e4
LT		 1320void __init sk_init(void)
1321{
4481374c 1322 if (totalram_pages <= 4096) {
1da177e4
LT		 1323 sysctl_wmem_max = 32767;
1324 sysctl_rmem_max = 32767;
1325 sysctl_wmem_default = 32767;
1326 sysctl_rmem_default = 32767;
4481374c 1327 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1328 sysctl_wmem_max = 131071;
1329 sysctl_rmem_max = 131071;
1330 }
1331}
1332
1333/*
1334 * Simple resource managers for sockets.
1335 */
1336
1337
4ec93edb
YH		 1338/*
1339 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1340 */
1341void sock_wfree(struct sk_buff *skb)
1342{
1343 struct sock *sk = skb->sk;
d99927f4 1344 unsigned int len = skb->truesize;
1da177e4 1345
d99927f4
ED		 1346 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1347 /*
1348 * Keep a reference on sk_wmem_alloc, this will be released
1349 * after sk_write_space() call
1350 */
1351 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1352 sk->sk_write_space(sk);
d99927f4
ED		 1353 len = 1;
1354 }
2b85a34e 1355 /*
d99927f4
ED		 1356 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1357 * could not do because of in-flight packets
2b85a34e 1358 */
d99927f4 1359 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1360 __sk_free(sk);
1da177e4 1361}
2a91525c 1362EXPORT_SYMBOL(sock_wfree);
1da177e4 1363
4ec93edb
YH		 1364/*
1365 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1366 */
1367void sock_rfree(struct sk_buff *skb)
1368{
1369 struct sock *sk = skb->sk;
d361fd59 1370 unsigned int len = skb->truesize;
1da177e4 1371
d361fd59
ED		 1372 atomic_sub(len, &sk->sk_rmem_alloc);
1373 sk_mem_uncharge(sk, len);
1da177e4 1374}
2a91525c 1375EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1376
1377
1378int sock_i_uid(struct sock *sk)
1379{
1380 int uid;
1381
f064af1e 1382 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1383 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1384 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1385 return uid;
1386}
2a91525c 1387EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1388
1389unsigned long sock_i_ino(struct sock *sk)
1390{
1391 unsigned long ino;
1392
f064af1e 1393 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1394 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1395 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1396 return ino;
1397}
2a91525c 1398EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1399
1400/*
1401 * Allocate a skb from the socket's send buffer.
1402 */
86a76caf 1403struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1404 gfp_t priority)
1da177e4
LT		 1405{
1406 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1407 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1408 if (skb) {
1409 skb_set_owner_w(skb, sk);
1410 return skb;
1411 }
1412 }
1413 return NULL;
1414}
2a91525c 1415EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1416
1417/*
1418 * Allocate a skb from the socket's receive buffer.
4ec93edb 1419 */
86a76caf 1420struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1421 gfp_t priority)
1da177e4
LT		 1422{
1423 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1424 struct sk_buff *skb = alloc_skb(size, priority);
1425 if (skb) {
1426 skb_set_owner_r(skb, sk);
1427 return skb;
1428 }
1429 }
1430 return NULL;
1431}
1432
4ec93edb 1433/*
1da177e4 1434 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1435 */
dd0fc66f 1436void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1437{
1438 if ((unsigned)size <= sysctl_optmem_max &&
1439 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1440 void *mem;
1441 /* First do the add, to avoid the race if kmalloc
4ec93edb 1442 * might sleep.
1da177e4
LT		 1443 */
1444 atomic_add(size, &sk->sk_omem_alloc);
1445 mem = kmalloc(size, priority);
1446 if (mem)
1447 return mem;
1448 atomic_sub(size, &sk->sk_omem_alloc);
1449 }
1450 return NULL;
1451}
2a91525c 1452EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1453
1454/*
1455 * Free an option memory block.
1456 */
1457void sock_kfree_s(struct sock *sk, void *mem, int size)
1458{
1459 kfree(mem);
1460 atomic_sub(size, &sk->sk_omem_alloc);
1461}
2a91525c 1462EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1463
1464/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1465 I think, these locks should be removed for datagram sockets.
1466 */
2a91525c 1467static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1468{
1469 DEFINE_WAIT(wait);
1470
1471 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1472 for (;;) {
1473 if (!timeo)
1474 break;
1475 if (signal_pending(current))
1476 break;
1477 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1478 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1479 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1480 break;
1481 if (sk->sk_shutdown & SEND_SHUTDOWN)
1482 break;
1483 if (sk->sk_err)
1484 break;
1485 timeo = schedule_timeout(timeo);
1486 }
aa395145 1487 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1488 return timeo;
1489}
1490
1491
1492/*
1493 * Generic send/receive buffer handlers
1494 */
1495
4cc7f68d
HX		 1496struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1497 unsigned long data_len, int noblock,
1498 int *errcode)
1da177e4
LT		 1499{
1500 struct sk_buff *skb;
7d877f3b 1501 gfp_t gfp_mask;
1da177e4
LT		 1502 long timeo;
1503 int err;
1504
1505 gfp_mask = sk->sk_allocation;
1506 if (gfp_mask & __GFP_WAIT)
1507 gfp_mask |= __GFP_REPEAT;
1508
1509 timeo = sock_sndtimeo(sk, noblock);
1510 while (1) {
1511 err = sock_error(sk);
1512 if (err != 0)
1513 goto failure;
1514
1515 err = -EPIPE;
1516 if (sk->sk_shutdown & SEND_SHUTDOWN)
1517 goto failure;
1518
1519 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1520 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1521 if (skb) {
1522 int npages;
1523 int i;
1524
1525 /* No pages, we're done... */
1526 if (!data_len)
1527 break;
1528
1529 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1530 skb->truesize += data_len;
1531 skb_shinfo(skb)->nr_frags = npages;
1532 for (i = 0; i < npages; i++) {
1533 struct page *page;
1da177e4
LT		 1534
1535 page = alloc_pages(sk->sk_allocation, 0);
1536 if (!page) {
1537 err = -ENOBUFS;
1538 skb_shinfo(skb)->nr_frags = i;
1539 kfree_skb(skb);
1540 goto failure;
1541 }
1542
ea2ab693
IC		 1543 __skb_fill_page_desc(skb, i,
1544 page, 0,
1545 (data_len >= PAGE_SIZE ?
1546 PAGE_SIZE :
1547 data_len));
1da177e4
LT		 1548 data_len -= PAGE_SIZE;
1549 }
1550
1551 /* Full success... */
1552 break;
1553 }
1554 err = -ENOBUFS;
1555 goto failure;
1556 }
1557 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1558 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1559 err = -EAGAIN;
1560 if (!timeo)
1561 goto failure;
1562 if (signal_pending(current))
1563 goto interrupted;
1564 timeo = sock_wait_for_wmem(sk, timeo);
1565 }
1566
1567 skb_set_owner_w(skb, sk);
1568 return skb;
1569
1570interrupted:
1571 err = sock_intr_errno(timeo);
1572failure:
1573 *errcode = err;
1574 return NULL;
1575}
4cc7f68d 1576EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1577
4ec93edb 1578struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1579 int noblock, int *errcode)
1580{
1581 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1582}
2a91525c 1583EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1584
1585static void __lock_sock(struct sock *sk)
f39234d6
NK		 1586 __releases(&sk->sk_lock.slock)
1587 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1588{
1589 DEFINE_WAIT(wait);
1590
e71a4783 1591 for (;;) {
1da177e4
LT		 1592 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1593 TASK_UNINTERRUPTIBLE);
1594 spin_unlock_bh(&sk->sk_lock.slock);
1595 schedule();
1596 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1597 if (!sock_owned_by_user(sk))
1da177e4
LT		 1598 break;
1599 }
1600 finish_wait(&sk->sk_lock.wq, &wait);
1601}
1602
1603static void __release_sock(struct sock *sk)
f39234d6
NK		 1604 __releases(&sk->sk_lock.slock)
1605 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1606{
1607 struct sk_buff *skb = sk->sk_backlog.head;
1608
1609 do {
1610 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1611 bh_unlock_sock(sk);
1612
1613 do {
1614 struct sk_buff *next = skb->next;
1615
7fee226a 1616 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1617 skb->next = NULL;
c57943a1 1618 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1619
1620 /*
1621 * We are in process context here with softirqs
1622 * disabled, use cond_resched_softirq() to preempt.
1623 * This is safe to do because we've taken the backlog
1624 * queue private:
1625 */
1626 cond_resched_softirq();
1627
1628 skb = next;
1629 } while (skb != NULL);
1630
1631 bh_lock_sock(sk);
e71a4783 1632 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1633
1634 /*
1635 * Doing the zeroing here guarantee we can not loop forever
1636 * while a wild producer attempts to flood us.
1637 */
1638 sk->sk_backlog.len = 0;
1da177e4
LT		 1639}
1640
1641/**
1642 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1643 * @sk: sock to wait on
1644 * @timeo: for how long
1da177e4
LT		 1645 *
1646 * Now socket state including sk->sk_err is changed only under lock,
1647 * hence we may omit checks after joining wait queue.
1648 * We check receive queue before schedule() only as optimization;
1649 * it is very likely that release_sock() added new data.
1650 */
1651int sk_wait_data(struct sock *sk, long *timeo)
1652{
1653 int rc;
1654 DEFINE_WAIT(wait);
1655
aa395145 1656 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1657 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1658 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1659 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1660 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1661 return rc;
1662}
1da177e4
LT		 1663EXPORT_SYMBOL(sk_wait_data);
1664
3ab224be
HA		 1665/**
1666 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1667 * @sk: socket
1668 * @size: memory size to allocate
1669 * @kind: allocation type
1670 *
1671 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1672 * rmem allocation. This function assumes that protocols which have
1673 * memory_pressure use sk_wmem_queued as write buffer accounting.
1674 */
1675int __sk_mem_schedule(struct sock *sk, int size, int kind)
1676{
1677 struct proto *prot = sk->sk_prot;
1678 int amt = sk_mem_pages(size);
8d987e5c 1679 long allocated;
3ab224be
HA		 1680
1681 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
8d987e5c 1682 allocated = atomic_long_add_return(amt, prot->memory_allocated);
3ab224be
HA		 1683
1684 /* Under limit. */
1685 if (allocated <= prot->sysctl_mem[0]) {
1686 if (prot->memory_pressure && *prot->memory_pressure)
1687 *prot->memory_pressure = 0;
1688 return 1;
1689 }
1690
1691 /* Under pressure. */
1692 if (allocated > prot->sysctl_mem[1])
1693 if (prot->enter_memory_pressure)
5c52ba17 1694 prot->enter_memory_pressure(sk);
3ab224be
HA		 1695
1696 /* Over hard limit. */
1697 if (allocated > prot->sysctl_mem[2])
1698 goto suppress_allocation;
1699
1700 /* guarantee minimum buffer size under pressure */
1701 if (kind == SK_MEM_RECV) {
1702 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1703 return 1;
1704 } else { /* SK_MEM_SEND */
1705 if (sk->sk_type == SOCK_STREAM) {
1706 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1707 return 1;
1708 } else if (atomic_read(&sk->sk_wmem_alloc) <
1709 prot->sysctl_wmem[0])
1710 return 1;
1711 }
1712
1713 if (prot->memory_pressure) {
1748376b
ED		 1714 int alloc;
1715
1716 if (!*prot->memory_pressure)
1717 return 1;
1718 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1719 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1720 sk_mem_pages(sk->sk_wmem_queued +
1721 atomic_read(&sk->sk_rmem_alloc) +
1722 sk->sk_forward_alloc))
1723 return 1;
1724 }
1725
1726suppress_allocation:
1727
1728 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1729 sk_stream_moderate_sndbuf(sk);
1730
1731 /* Fail only if socket is _under_ its sndbuf.
1732 * In this case we cannot block, so that we have to fail.
1733 */
1734 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1735 return 1;
1736 }
1737
3847ce32
SM		 1738 trace_sock_exceed_buf_limit(sk, prot, allocated);
1739
3ab224be
HA		 1740 /* Alas. Undo changes. */
1741 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
8d987e5c 1742 atomic_long_sub(amt, prot->memory_allocated);
3ab224be
HA		 1743 return 0;
1744}
3ab224be
HA		 1745EXPORT_SYMBOL(__sk_mem_schedule);
1746
1747/**
1748 * __sk_reclaim - reclaim memory_allocated
1749 * @sk: socket
1750 */
1751void __sk_mem_reclaim(struct sock *sk)
1752{
1753 struct proto *prot = sk->sk_prot;
1754
8d987e5c 1755 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1756 prot->memory_allocated);
1757 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1758
1759 if (prot->memory_pressure && *prot->memory_pressure &&
8d987e5c 1760 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
3ab224be
HA		 1761 *prot->memory_pressure = 0;
1762}
3ab224be
HA		 1763EXPORT_SYMBOL(__sk_mem_reclaim);
1764
1765
1da177e4
LT		 1766/*
1767 * Set of default routines for initialising struct proto_ops when
1768 * the protocol does not support a particular function. In certain
1769 * cases where it makes no sense for a protocol to have a "do nothing"
1770 * function, some default processing is provided.
1771 */
1772
1773int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1774{
1775 return -EOPNOTSUPP;
1776}
2a91525c 1777EXPORT_SYMBOL(sock_no_bind);
1da177e4 1778
4ec93edb 1779int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1780 int len, int flags)
1781{
1782 return -EOPNOTSUPP;
1783}
2a91525c 1784EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1785
1786int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1787{
1788 return -EOPNOTSUPP;
1789}
2a91525c 1790EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1791
1792int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1793{
1794 return -EOPNOTSUPP;
1795}
2a91525c 1796EXPORT_SYMBOL(sock_no_accept);
1da177e4 1797
4ec93edb 1798int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1799 int *len, int peer)
1800{
1801 return -EOPNOTSUPP;
1802}
2a91525c 1803EXPORT_SYMBOL(sock_no_getname);
1da177e4 1804
2a91525c 1805unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1806{
1807 return 0;
1808}
2a91525c 1809EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1810
1811int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1812{
1813 return -EOPNOTSUPP;
1814}
2a91525c 1815EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1816
1817int sock_no_listen(struct socket *sock, int backlog)
1818{
1819 return -EOPNOTSUPP;
1820}
2a91525c 1821EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1822
1823int sock_no_shutdown(struct socket *sock, int how)
1824{
1825 return -EOPNOTSUPP;
1826}
2a91525c 1827EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1828
1829int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1830 char __user *optval, unsigned int optlen)
1da177e4
LT		 1831{
1832 return -EOPNOTSUPP;
1833}
2a91525c 1834EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1835
1836int sock_no_getsockopt(struct socket *sock, int level, int optname,
1837 char __user *optval, int __user *optlen)
1838{
1839 return -EOPNOTSUPP;
1840}
2a91525c 1841EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1842
1843int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1844 size_t len)
1845{
1846 return -EOPNOTSUPP;
1847}
2a91525c 1848EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1849
1850int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1851 size_t len, int flags)
1852{
1853 return -EOPNOTSUPP;
1854}
2a91525c 1855EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1856
1857int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1858{
1859 /* Mirror missing mmap method error code */
1860 return -ENODEV;
1861}
2a91525c 1862EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1863
1864ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1865{
1866 ssize_t res;
1867 struct msghdr msg = {.msg_flags = flags};
1868 struct kvec iov;
1869 char *kaddr = kmap(page);
1870 iov.iov_base = kaddr + offset;
1871 iov.iov_len = size;
1872 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1873 kunmap(page);
1874 return res;
1875}
2a91525c 1876EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1877
1878/*
1879 * Default Socket Callbacks
1880 */
1881
1882static void sock_def_wakeup(struct sock *sk)
1883{
43815482
ED		 1884 struct socket_wq *wq;
1885
1886 rcu_read_lock();
1887 wq = rcu_dereference(sk->sk_wq);
1888 if (wq_has_sleeper(wq))
1889 wake_up_interruptible_all(&wq->wait);
1890 rcu_read_unlock();
1da177e4
LT		 1891}
1892
1893static void sock_def_error_report(struct sock *sk)
1894{
43815482
ED		 1895 struct socket_wq *wq;
1896
1897 rcu_read_lock();
1898 wq = rcu_dereference(sk->sk_wq);
1899 if (wq_has_sleeper(wq))
1900 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1901 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1902 rcu_read_unlock();
1da177e4
LT		 1903}
1904
1905static void sock_def_readable(struct sock *sk, int len)
1906{
43815482
ED		 1907 struct socket_wq *wq;
1908
1909 rcu_read_lock();
1910 wq = rcu_dereference(sk->sk_wq);
1911 if (wq_has_sleeper(wq))
2c6607c6 1912 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1913 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1914 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1915 rcu_read_unlock();
1da177e4
LT		 1916}
1917
1918static void sock_def_write_space(struct sock *sk)
1919{
43815482
ED		 1920 struct socket_wq *wq;
1921
1922 rcu_read_lock();
1da177e4
LT		 1923
1924 /* Do not wake up a writer until he can make "significant"
1925 * progress. --DaveM
1926 */
e71a4783 1927 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1928 wq = rcu_dereference(sk->sk_wq);
1929 if (wq_has_sleeper(wq))
1930 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1931 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1932
1933 /* Should agree with poll, otherwise some programs break */
1934 if (sock_writeable(sk))
8d8ad9d7 1935 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1936 }
1937
43815482 1938 rcu_read_unlock();
1da177e4
LT		 1939}
1940
1941static void sock_def_destruct(struct sock *sk)
1942{
a51482bd 1943 kfree(sk->sk_protinfo);
1da177e4
LT		 1944}
1945
1946void sk_send_sigurg(struct sock *sk)
1947{
1948 if (sk->sk_socket && sk->sk_socket->file)
1949 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1950 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1951}
2a91525c 1952EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1953
1954void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1955 unsigned long expires)
1956{
1957 if (!mod_timer(timer, expires))
1958 sock_hold(sk);
1959}
1da177e4
LT		 1960EXPORT_SYMBOL(sk_reset_timer);
1961
1962void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1963{
1964 if (timer_pending(timer) && del_timer(timer))
1965 __sock_put(sk);
1966}
1da177e4
LT		 1967EXPORT_SYMBOL(sk_stop_timer);
1968
1969void sock_init_data(struct socket *sock, struct sock *sk)
1970{
1971 skb_queue_head_init(&sk->sk_receive_queue);
1972 skb_queue_head_init(&sk->sk_write_queue);
1973 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 1974#ifdef CONFIG_NET_DMA
1975 skb_queue_head_init(&sk->sk_async_wait_queue);
1976#endif
1da177e4
LT		 1977
1978 sk->sk_send_head = NULL;
1979
1980 init_timer(&sk->sk_timer);
4ec93edb 1981
1da177e4
LT		 1982 sk->sk_allocation = GFP_KERNEL;
1983 sk->sk_rcvbuf = sysctl_rmem_default;
1984 sk->sk_sndbuf = sysctl_wmem_default;
1985 sk->sk_state = TCP_CLOSE;
972692e0 1986 sk_set_socket(sk, sock);
1da177e4
LT		 1987
1988 sock_set_flag(sk, SOCK_ZAPPED);
1989
e71a4783 1990 if (sock) {
1da177e4 1991 sk->sk_type = sock->type;
43815482 1992 sk->sk_wq = sock->wq;
1da177e4
LT		 1993 sock->sk = sk;
1994 } else
43815482 1995 sk->sk_wq = NULL;
1da177e4 1996
b6c6712a 1997 spin_lock_init(&sk->sk_dst_lock);
1da177e4 1998 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 1999 lockdep_set_class_and_name(&sk->sk_callback_lock,
2000 af_callback_keys + sk->sk_family,
2001 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2002
2003 sk->sk_state_change = sock_def_wakeup;
2004 sk->sk_data_ready = sock_def_readable;
2005 sk->sk_write_space = sock_def_write_space;
2006 sk->sk_error_report = sock_def_error_report;
2007 sk->sk_destruct = sock_def_destruct;
2008
2009 sk->sk_sndmsg_page = NULL;
2010 sk->sk_sndmsg_off = 0;
2011
109f6e39
EB		 2012 sk->sk_peer_pid = NULL;
2013 sk->sk_peer_cred = NULL;
1da177e4
LT		 2014 sk->sk_write_pending = 0;
2015 sk->sk_rcvlowat = 1;
2016 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2017 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2018
f37f0afb 2019 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2020
4dc6dc71
ED		 2021 /*
2022 * Before updating sk_refcnt, we must commit prior changes to memory
2023 * (Documentation/RCU/rculist_nulls.txt for details)
2024 */
2025 smp_wmb();
1da177e4 2026 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2027 atomic_set(&sk->sk_drops, 0);
1da177e4 2028}
2a91525c 2029EXPORT_SYMBOL(sock_init_data);
1da177e4 2030
b5606c2d 2031void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2032{
2033 might_sleep();
a5b5bb9a 2034 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2035 if (sk->sk_lock.owned)
1da177e4 2036 __lock_sock(sk);
d2e9117c 2037 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2038 spin_unlock(&sk->sk_lock.slock);
2039 /*
2040 * The sk_lock has mutex_lock() semantics here:
2041 */
fcc70d5f 2042 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2043 local_bh_enable();
1da177e4 2044}
fcc70d5f 2045EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2046
b5606c2d 2047void release_sock(struct sock *sk)
1da177e4 2048{
a5b5bb9a
IM		 2049 /*
2050 * The sk_lock has mutex_unlock() semantics:
2051 */
2052 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2053
2054 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2055 if (sk->sk_backlog.tail)
2056 __release_sock(sk);
d2e9117c 2057 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2058 if (waitqueue_active(&sk->sk_lock.wq))
2059 wake_up(&sk->sk_lock.wq);
2060 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2061}
2062EXPORT_SYMBOL(release_sock);
2063
8a74ad60
ED		 2064/**
2065 * lock_sock_fast - fast version of lock_sock
2066 * @sk: socket
2067 *
2068 * This version should be used for very small section, where process wont block
2069 * return false if fast path is taken
2070 * sk_lock.slock locked, owned = 0, BH disabled
2071 * return true if slow path is taken
2072 * sk_lock.slock unlocked, owned = 1, BH enabled
2073 */
2074bool lock_sock_fast(struct sock *sk)
2075{
2076 might_sleep();
2077 spin_lock_bh(&sk->sk_lock.slock);
2078
2079 if (!sk->sk_lock.owned)
2080 /*
2081 * Note : We must disable BH
2082 */
2083 return false;
2084
2085 __lock_sock(sk);
2086 sk->sk_lock.owned = 1;
2087 spin_unlock(&sk->sk_lock.slock);
2088 /*
2089 * The sk_lock has mutex_lock() semantics here:
2090 */
2091 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2092 local_bh_enable();
2093 return true;
2094}
2095EXPORT_SYMBOL(lock_sock_fast);
2096
1da177e4 2097int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2098{
b7aa0bf7 2099 struct timeval tv;
1da177e4 2100 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2101 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2102 tv = ktime_to_timeval(sk->sk_stamp);
2103 if (tv.tv_sec == -1)
1da177e4 2104 return -ENOENT;
b7aa0bf7
ED		 2105 if (tv.tv_sec == 0) {
2106 sk->sk_stamp = ktime_get_real();
2107 tv = ktime_to_timeval(sk->sk_stamp);
2108 }
2109 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2110}
1da177e4
LT		 2111EXPORT_SYMBOL(sock_get_timestamp);
2112
ae40eb1e
ED		 2113int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2114{
2115 struct timespec ts;
2116 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2117 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2118 ts = ktime_to_timespec(sk->sk_stamp);
2119 if (ts.tv_sec == -1)
2120 return -ENOENT;
2121 if (ts.tv_sec == 0) {
2122 sk->sk_stamp = ktime_get_real();
2123 ts = ktime_to_timespec(sk->sk_stamp);
2124 }
2125 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2126}
2127EXPORT_SYMBOL(sock_get_timestampns);
2128
20d49473 2129void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2130{
20d49473
PO		 2131 if (!sock_flag(sk, flag)) {
2132 sock_set_flag(sk, flag);
2133 /*
2134 * we just set one of the two flags which require net
2135 * time stamping, but time stamping might have been on
2136 * already because of the other one
2137 */
2138 if (!sock_flag(sk,
2139 flag == SOCK_TIMESTAMP ?
2140 SOCK_TIMESTAMPING_RX_SOFTWARE :
2141 SOCK_TIMESTAMP))
2142 net_enable_timestamp();
1da177e4
LT		 2143 }
2144}
1da177e4
LT		 2145
2146/*
2147 * Get a socket option on an socket.
2148 *
2149 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2150 * asynchronous errors should be reported by getsockopt. We assume
2151 * this means if you specify SO_ERROR (otherwise whats the point of it).
2152 */
2153int sock_common_getsockopt(struct socket *sock, int level, int optname,
2154 char __user *optval, int __user *optlen)
2155{
2156 struct sock *sk = sock->sk;
2157
2158 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2159}
1da177e4
LT		 2160EXPORT_SYMBOL(sock_common_getsockopt);
2161
3fdadf7d 2162#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2163int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2164 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2165{
2166 struct sock *sk = sock->sk;
2167
1e51f951 2168 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2169 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2170 optval, optlen);
3fdadf7d
DM		 2171 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2172}
2173EXPORT_SYMBOL(compat_sock_common_getsockopt);
2174#endif
2175
1da177e4
LT		 2176int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2177 struct msghdr *msg, size_t size, int flags)
2178{
2179 struct sock *sk = sock->sk;
2180 int addr_len = 0;
2181 int err;
2182
2183 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2184 flags & ~MSG_DONTWAIT, &addr_len);
2185 if (err >= 0)
2186 msg->msg_namelen = addr_len;
2187 return err;
2188}
1da177e4
LT		 2189EXPORT_SYMBOL(sock_common_recvmsg);
2190
2191/*
2192 * Set socket options on an inet socket.
2193 */
2194int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2195 char __user *optval, unsigned int optlen)
1da177e4
LT		 2196{
2197 struct sock *sk = sock->sk;
2198
2199 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2200}
1da177e4
LT		 2201EXPORT_SYMBOL(sock_common_setsockopt);
2202
3fdadf7d 2203#ifdef CONFIG_COMPAT
543d9cfe 2204int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2205 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2206{
2207 struct sock *sk = sock->sk;
2208
543d9cfe
ACM		 2209 if (sk->sk_prot->compat_setsockopt != NULL)
2210 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2211 optval, optlen);
3fdadf7d
DM		 2212 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2213}
2214EXPORT_SYMBOL(compat_sock_common_setsockopt);
2215#endif
2216
1da177e4
LT		 2217void sk_common_release(struct sock *sk)
2218{
2219 if (sk->sk_prot->destroy)
2220 sk->sk_prot->destroy(sk);
2221
2222 /*
2223 * Observation: when sock_common_release is called, processes have
2224 * no access to socket. But net still has.
2225 * Step one, detach it from networking:
2226 *
2227 * A. Remove from hash tables.
2228 */
2229
2230 sk->sk_prot->unhash(sk);
2231
2232 /*
2233 * In this point socket cannot receive new packets, but it is possible
2234 * that some packets are in flight because some CPU runs receiver and
2235 * did hash table lookup before we unhashed socket. They will achieve
2236 * receive queue and will be purged by socket destructor.
2237 *
2238 * Also we still have packets pending on receive queue and probably,
2239 * our own packets waiting in device queues. sock_destroy will drain
2240 * receive queue, but transmitted packets will delay socket destruction
2241 * until the last reference will be released.
2242 */
2243
2244 sock_orphan(sk);
2245
2246 xfrm_sk_free_policy(sk);
2247
e6848976 2248 sk_refcnt_debug_release(sk);
1da177e4
LT		 2249 sock_put(sk);
2250}
1da177e4
LT		 2251EXPORT_SYMBOL(sk_common_release);
2252
2253static DEFINE_RWLOCK(proto_list_lock);
2254static LIST_HEAD(proto_list);
2255
13ff3d6f
PE		 2256#ifdef CONFIG_PROC_FS
2257#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2258struct prot_inuse {
2259 int val[PROTO_INUSE_NR];
2260};
13ff3d6f
PE		 2261
2262static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2263
2264#ifdef CONFIG_NET_NS
2265void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2266{
d6d9ca0f 2267 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2268}
2269EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2270
2271int sock_prot_inuse_get(struct net *net, struct proto *prot)
2272{
2273 int cpu, idx = prot->inuse_idx;
2274 int res = 0;
2275
2276 for_each_possible_cpu(cpu)
2277 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2278
2279 return res >= 0 ? res : 0;
2280}
2281EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2282
2c8c1e72 2283static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2284{
2285 net->core.inuse = alloc_percpu(struct prot_inuse);
2286 return net->core.inuse ? 0 : -ENOMEM;
2287}
2288
2c8c1e72 2289static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2290{
2291 free_percpu(net->core.inuse);
2292}
2293
2294static struct pernet_operations net_inuse_ops = {
2295 .init = sock_inuse_init_net,
2296 .exit = sock_inuse_exit_net,
2297};
2298
2299static __init int net_inuse_init(void)
2300{
2301 if (register_pernet_subsys(&net_inuse_ops))
2302 panic("Cannot initialize net inuse counters");
2303
2304 return 0;
2305}
2306
2307core_initcall(net_inuse_init);
2308#else
1338d466
PE		 2309static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2310
c29a0bc4 2311void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2312{
d6d9ca0f 2313 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2314}
2315EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2316
c29a0bc4 2317int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2318{
2319 int cpu, idx = prot->inuse_idx;
2320 int res = 0;
2321
2322 for_each_possible_cpu(cpu)
2323 res += per_cpu(prot_inuse, cpu).val[idx];
2324
2325 return res >= 0 ? res : 0;
2326}
2327EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2328#endif
13ff3d6f
PE		 2329
2330static void assign_proto_idx(struct proto *prot)
2331{
2332 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2333
2334 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2335 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2336 return;
2337 }
2338
2339 set_bit(prot->inuse_idx, proto_inuse_idx);
2340}
2341
2342static void release_proto_idx(struct proto *prot)
2343{
2344 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2345 clear_bit(prot->inuse_idx, proto_inuse_idx);
2346}
2347#else
2348static inline void assign_proto_idx(struct proto *prot)
2349{
2350}
2351
2352static inline void release_proto_idx(struct proto *prot)
2353{
2354}
2355#endif
2356
b733c007
PE		 2357int proto_register(struct proto *prot, int alloc_slab)
2358{
1da177e4
LT		 2359 if (alloc_slab) {
2360 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2361 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2362 NULL);
1da177e4
LT		 2363
2364 if (prot->slab == NULL) {
2365 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2366 prot->name);
60e7663d 2367 goto out;
1da177e4 2368 }
2e6599cb
ACM		 2369
2370 if (prot->rsk_prot != NULL) {
faf23422 2371 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2372 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2373 goto out_free_sock_slab;
2374
7e56b5d6 2375 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2376 prot->rsk_prot->obj_size, 0,
20c2df83 2377 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2378
2379 if (prot->rsk_prot->slab == NULL) {
2380 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2381 prot->name);
2382 goto out_free_request_sock_slab_name;
2383 }
2384 }
8feaf0c0 2385
6d6ee43e 2386 if (prot->twsk_prot != NULL) {
faf23422 2387 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2388
7e56b5d6 2389 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2390 goto out_free_request_sock_slab;
2391
6d6ee43e 2392 prot->twsk_prot->twsk_slab =
7e56b5d6 2393 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2394 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2395 0,
2396 SLAB_HWCACHE_ALIGN |
2397 prot->slab_flags,
20c2df83 2398 NULL);
6d6ee43e 2399 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2400 goto out_free_timewait_sock_slab_name;
2401 }
1da177e4
LT		 2402 }
2403
2a278051 2404 write_lock(&proto_list_lock);
1da177e4 2405 list_add(&prot->node, &proto_list);
13ff3d6f 2406 assign_proto_idx(prot);
1da177e4 2407 write_unlock(&proto_list_lock);
b733c007
PE		 2408 return 0;
2409
8feaf0c0 2410out_free_timewait_sock_slab_name:
7e56b5d6 2411 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2412out_free_request_sock_slab:
2413 if (prot->rsk_prot && prot->rsk_prot->slab) {
2414 kmem_cache_destroy(prot->rsk_prot->slab);
2415 prot->rsk_prot->slab = NULL;
2416 }
2e6599cb 2417out_free_request_sock_slab_name:
72150e9b
DC		 2418 if (prot->rsk_prot)
2419 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2420out_free_sock_slab:
2421 kmem_cache_destroy(prot->slab);
2422 prot->slab = NULL;
b733c007
PE		 2423out:
2424 return -ENOBUFS;
1da177e4 2425}
1da177e4
LT		 2426EXPORT_SYMBOL(proto_register);
2427
2428void proto_unregister(struct proto *prot)
2429{
2430 write_lock(&proto_list_lock);
13ff3d6f 2431 release_proto_idx(prot);
0a3f4358
PM		 2432 list_del(&prot->node);
2433 write_unlock(&proto_list_lock);
1da177e4
LT		 2434
2435 if (prot->slab != NULL) {
2436 kmem_cache_destroy(prot->slab);
2437 prot->slab = NULL;
2438 }
2439
2e6599cb 2440 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2441 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2442 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2443 prot->rsk_prot->slab = NULL;
2444 }
2445
6d6ee43e 2446 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2447 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2448 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2449 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2450 }
1da177e4 2451}
1da177e4
LT		 2452EXPORT_SYMBOL(proto_unregister);
2453
2454#ifdef CONFIG_PROC_FS
1da177e4 2455static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2456 __acquires(proto_list_lock)
1da177e4
LT		 2457{
2458 read_lock(&proto_list_lock);
60f0438a 2459 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2460}
2461
2462static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2463{
60f0438a 2464 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2465}
2466
2467static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2468 __releases(proto_list_lock)
1da177e4
LT		 2469{
2470 read_unlock(&proto_list_lock);
2471}
2472
2473static char proto_method_implemented(const void *method)
2474{
2475 return method == NULL ? 'n' : 'y';
2476}
2477
2478static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2479{
8d987e5c 2480 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2481 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2482 proto->name,
2483 proto->obj_size,
14e943db 2484 sock_prot_inuse_get(seq_file_net(seq), proto),
8d987e5c 2485 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
1da177e4
LT		 2486 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2487 proto->max_header,
2488 proto->slab == NULL ? "no" : "yes",
2489 module_name(proto->owner),
2490 proto_method_implemented(proto->close),
2491 proto_method_implemented(proto->connect),
2492 proto_method_implemented(proto->disconnect),
2493 proto_method_implemented(proto->accept),
2494 proto_method_implemented(proto->ioctl),
2495 proto_method_implemented(proto->init),
2496 proto_method_implemented(proto->destroy),
2497 proto_method_implemented(proto->shutdown),
2498 proto_method_implemented(proto->setsockopt),
2499 proto_method_implemented(proto->getsockopt),
2500 proto_method_implemented(proto->sendmsg),
2501 proto_method_implemented(proto->recvmsg),
2502 proto_method_implemented(proto->sendpage),
2503 proto_method_implemented(proto->bind),
2504 proto_method_implemented(proto->backlog_rcv),
2505 proto_method_implemented(proto->hash),
2506 proto_method_implemented(proto->unhash),
2507 proto_method_implemented(proto->get_port),
2508 proto_method_implemented(proto->enter_memory_pressure));
2509}
2510
2511static int proto_seq_show(struct seq_file *seq, void *v)
2512{
60f0438a 2513 if (v == &proto_list)
1da177e4
LT		 2514 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2515 "protocol",
2516 "size",
2517 "sockets",
2518 "memory",
2519 "press",
2520 "maxhdr",
2521 "slab",
2522 "module",
2523 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2524 else
60f0438a 2525 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2526 return 0;
2527}
2528
f690808e 2529static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2530 .start = proto_seq_start,
2531 .next = proto_seq_next,
2532 .stop = proto_seq_stop,
2533 .show = proto_seq_show,
2534};
2535
2536static int proto_seq_open(struct inode *inode, struct file *file)
2537{
14e943db
ED		 2538 return seq_open_net(inode, file, &proto_seq_ops,
2539 sizeof(struct seq_net_private));
1da177e4
LT		 2540}
2541
9a32144e 2542static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2543 .owner = THIS_MODULE,
2544 .open = proto_seq_open,
2545 .read = seq_read,
2546 .llseek = seq_lseek,
14e943db
ED		 2547 .release = seq_release_net,
2548};
2549
2550static __net_init int proto_init_net(struct net *net)
2551{
2552 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2553 return -ENOMEM;
2554
2555 return 0;
2556}
2557
2558static __net_exit void proto_exit_net(struct net *net)
2559{
2560 proc_net_remove(net, "protocols");
2561}
2562
2563
2564static __net_initdata struct pernet_operations proto_net_ops = {
2565 .init = proto_init_net,
2566 .exit = proto_exit_net,
1da177e4
LT		 2567};
2568
2569static int __init proto_init(void)
2570{
14e943db 2571 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2572}
2573
2574subsys_initcall(proto_init);
2575
2576#endif /* PROC_FS */
Linux 6.x block layer and io_uring tree(s)