]> Git Repo - linux.git/blame - net/core/sock.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'x86/cleanups' into perf/uprobes
[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, <[email protected]>
12 * Florian La Roche, <[email protected]>
13 * Alan Cox, <[email protected]>
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>
e1aab161 114#include <linux/jump_label.h>
3969eb38 115#include <linux/memcontrol.h>
1da177e4
LT		 116
117#include <asm/uaccess.h>
118#include <asm/system.h>
119
120#include <linux/netdevice.h>
121#include <net/protocol.h>
122#include <linux/skbuff.h>
457c4cbc 123#include <net/net_namespace.h>
2e6599cb 124#include <net/request_sock.h>
1da177e4 125#include <net/sock.h>
20d49473 126#include <linux/net_tstamp.h>
1da177e4
LT		 127#include <net/xfrm.h>
128#include <linux/ipsec.h>
f8451725 129#include <net/cls_cgroup.h>
5bc1421e 130#include <net/netprio_cgroup.h>
1da177e4
LT		 131
132#include <linux/filter.h>
133
3847ce32
SM		 134#include <trace/events/sock.h>
135
1da177e4
LT		 136#ifdef CONFIG_INET
137#include <net/tcp.h>
138#endif
139
36b77a52 140static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 141static LIST_HEAD(proto_list);
142
143#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
144int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
145{
146 struct proto *proto;
147 int ret = 0;
148
36b77a52 149 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 150 list_for_each_entry(proto, &proto_list, node) {
151 if (proto->init_cgroup) {
152 ret = proto->init_cgroup(cgrp, ss);
153 if (ret)
154 goto out;
155 }
156 }
157
36b77a52 158 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 159 return ret;
160out:
161 list_for_each_entry_continue_reverse(proto, &proto_list, node)
162 if (proto->destroy_cgroup)
163 proto->destroy_cgroup(cgrp, ss);
36b77a52 164 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 165 return ret;
166}
167
168void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss)
169{
170 struct proto *proto;
171
36b77a52 172 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 173 list_for_each_entry_reverse(proto, &proto_list, node)
174 if (proto->destroy_cgroup)
175 proto->destroy_cgroup(cgrp, ss);
36b77a52 176 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 177}
178#endif
179
da21f24d
IM		 180/*
181 * Each address family might have different locking rules, so we have
182 * one slock key per address family:
183 */
a5b5bb9a
IM		 184static struct lock_class_key af_family_keys[AF_MAX];
185static struct lock_class_key af_family_slock_keys[AF_MAX];
186
e1aab161
GC		 187struct jump_label_key memcg_socket_limit_enabled;
188EXPORT_SYMBOL(memcg_socket_limit_enabled);
189
a5b5bb9a
IM		 190/*
191 * Make lock validator output more readable. (we pre-construct these
192 * strings build-time, so that runtime initialization of socket
193 * locks is fast):
194 */
36cbd3dc 195static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 196 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
197 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
198 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
199 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
200 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
201 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
202 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 203 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 204 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 205 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 206 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 207 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 208 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 209 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 210};
36cbd3dc 211static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 212 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
213 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
214 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
215 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
216 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
217 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
218 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 219 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 220 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 221 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 222 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 223 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 224 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 225 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 226};
36cbd3dc 227static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 228 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
229 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
230 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
231 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
232 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
233 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
234 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 235 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 236 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 237 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 238 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 239 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 240 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 241 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 242};
da21f24d
IM		 243
244/*
245 * sk_callback_lock locking rules are per-address-family,
246 * so split the lock classes by using a per-AF key:
247 */
248static struct lock_class_key af_callback_keys[AF_MAX];
249
1da177e4
LT		 250/* Take into consideration the size of the struct sk_buff overhead in the
251 * determination of these values, since that is non-constant across
252 * platforms. This makes socket queueing behavior and performance
253 * not depend upon such differences.
254 */
255#define _SK_MEM_PACKETS 256
87fb4b7b 256#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 257#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
258#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
259
260/* Run time adjustable parameters. */
ab32ea5d
BH		 261__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
262__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
263__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
264__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 265
25985edc 266/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 267int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 268EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 269
5bc1421e
NH		 270#if defined(CONFIG_CGROUPS)
271#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 272int net_cls_subsys_id = -1;
273EXPORT_SYMBOL_GPL(net_cls_subsys_id);
274#endif
5bc1421e
NH		 275#if !defined(CONFIG_NETPRIO_CGROUP)
276int net_prio_subsys_id = -1;
277EXPORT_SYMBOL_GPL(net_prio_subsys_id);
278#endif
279#endif
f8451725 280
1da177e4
LT		 281static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
282{
283 struct timeval tv;
284
285 if (optlen < sizeof(tv))
286 return -EINVAL;
287 if (copy_from_user(&tv, optval, sizeof(tv)))
288 return -EFAULT;
ba78073e
VA		 289 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
290 return -EDOM;
1da177e4 291
ba78073e 292 if (tv.tv_sec < 0) {
6f11df83
AM		 293 static int warned __read_mostly;
294
ba78073e 295 *timeo_p = 0;
50aab54f 296 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 297 warned++;
298 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
299 "tries to set negative timeout\n",
ba25f9dc 300 current->comm, task_pid_nr(current));
50aab54f 301 }
ba78073e
VA		 302 return 0;
303 }
1da177e4
LT		 304 *timeo_p = MAX_SCHEDULE_TIMEOUT;
305 if (tv.tv_sec == 0 && tv.tv_usec == 0)
306 return 0;
307 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
308 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
309 return 0;
310}
311
312static void sock_warn_obsolete_bsdism(const char *name)
313{
314 static int warned;
315 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 316 if (strcmp(warncomm, current->comm) && warned < 5) {
317 strcpy(warncomm, current->comm);
1da177e4
LT		 318 printk(KERN_WARNING "process `%s' is using obsolete "
319 "%s SO_BSDCOMPAT\n", warncomm, name);
320 warned++;
321 }
322}
323
08e29af3
ED		 324#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
325
326static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 327{
08e29af3
ED		 328 if (sk->sk_flags & flags) {
329 sk->sk_flags &= ~flags;
330 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 331 net_disable_timestamp();
1da177e4
LT		 332 }
333}
334
335
f0088a50
DV		 336int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
337{
766e9037 338 int err;
f0088a50 339 int skb_len;
3b885787
NH		 340 unsigned long flags;
341 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 342
0fd7bac6 343 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 344 atomic_inc(&sk->sk_drops);
3847ce32 345 trace_sock_rcvqueue_full(sk, skb);
766e9037 346 return -ENOMEM;
f0088a50
DV		 347 }
348
fda9ef5d 349 err = sk_filter(sk, skb);
f0088a50 350 if (err)
766e9037 351 return err;
f0088a50 352
3ab224be 353 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 354 atomic_inc(&sk->sk_drops);
355 return -ENOBUFS;
3ab224be
HA		 356 }
357
f0088a50
DV		 358 skb->dev = NULL;
359 skb_set_owner_r(skb, sk);
49ad9599 360
f0088a50
DV		 361 /* Cache the SKB length before we tack it onto the receive
362 * queue. Once it is added it no longer belongs to us and
363 * may be freed by other threads of control pulling packets
364 * from the queue.
365 */
366 skb_len = skb->len;
367
7fee226a
ED		 368 /* we escape from rcu protected region, make sure we dont leak
369 * a norefcounted dst
370 */
371 skb_dst_force(skb);
372
3b885787
NH		 373 spin_lock_irqsave(&list->lock, flags);
374 skb->dropcount = atomic_read(&sk->sk_drops);
375 __skb_queue_tail(list, skb);
376 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 377
378 if (!sock_flag(sk, SOCK_DEAD))
379 sk->sk_data_ready(sk, skb_len);
766e9037 380 return 0;
f0088a50
DV		 381}
382EXPORT_SYMBOL(sock_queue_rcv_skb);
383
58a5a7b9 384int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 385{
386 int rc = NET_RX_SUCCESS;
387
fda9ef5d 388 if (sk_filter(sk, skb))
f0088a50
DV		 389 goto discard_and_relse;
390
391 skb->dev = NULL;
392
c377411f
ED		 393 if (sk_rcvqueues_full(sk, skb)) {
394 atomic_inc(&sk->sk_drops);
395 goto discard_and_relse;
396 }
58a5a7b9
ACM		 397 if (nested)
398 bh_lock_sock_nested(sk);
399 else
400 bh_lock_sock(sk);
a5b5bb9a
IM		 401 if (!sock_owned_by_user(sk)) {
402 /*
403 * trylock + unlock semantics:
404 */
405 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
406
c57943a1 407 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 408
409 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 410 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 411 bh_unlock_sock(sk);
412 atomic_inc(&sk->sk_drops);
413 goto discard_and_relse;
414 }
415
f0088a50
DV		 416 bh_unlock_sock(sk);
417out:
418 sock_put(sk);
419 return rc;
420discard_and_relse:
421 kfree_skb(skb);
422 goto out;
423}
424EXPORT_SYMBOL(sk_receive_skb);
425
ea94ff3b
KK		 426void sk_reset_txq(struct sock *sk)
427{
428 sk_tx_queue_clear(sk);
429}
430EXPORT_SYMBOL(sk_reset_txq);
431
f0088a50
DV		 432struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
433{
b6c6712a 434 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 435
436 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 437 sk_tx_queue_clear(sk);
a9b3cd7f 438 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 439 dst_release(dst);
440 return NULL;
441 }
442
443 return dst;
444}
445EXPORT_SYMBOL(__sk_dst_check);
446
447struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
448{
449 struct dst_entry *dst = sk_dst_get(sk);
450
451 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
452 sk_dst_reset(sk);
453 dst_release(dst);
454 return NULL;
455 }
456
457 return dst;
458}
459EXPORT_SYMBOL(sk_dst_check);
460
4878809f
DM		 461static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
462{
463 int ret = -ENOPROTOOPT;
464#ifdef CONFIG_NETDEVICES
3b1e0a65 465 struct net *net = sock_net(sk);
4878809f
DM		 466 char devname[IFNAMSIZ];
467 int index;
468
469 /* Sorry... */
470 ret = -EPERM;
471 if (!capable(CAP_NET_RAW))
472 goto out;
473
474 ret = -EINVAL;
475 if (optlen < 0)
476 goto out;
477
478 /* Bind this socket to a particular device like "eth0",
479 * as specified in the passed interface name. If the
480 * name is "" or the option length is zero the socket
481 * is not bound.
482 */
483 if (optlen > IFNAMSIZ - 1)
484 optlen = IFNAMSIZ - 1;
485 memset(devname, 0, sizeof(devname));
486
487 ret = -EFAULT;
488 if (copy_from_user(devname, optval, optlen))
489 goto out;
490
000ba2e4
DM		 491 index = 0;
492 if (devname[0] != '\0') {
bf8e56bf 493 struct net_device *dev;
4878809f 494
bf8e56bf
ED		 495 rcu_read_lock();
496 dev = dev_get_by_name_rcu(net, devname);
497 if (dev)
498 index = dev->ifindex;
499 rcu_read_unlock();
4878809f
DM		 500 ret = -ENODEV;
501 if (!dev)
502 goto out;
4878809f
DM		 503 }
504
505 lock_sock(sk);
506 sk->sk_bound_dev_if = index;
507 sk_dst_reset(sk);
508 release_sock(sk);
509
510 ret = 0;
511
512out:
513#endif
514
515 return ret;
516}
517
c0ef877b
PE		 518static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
519{
520 if (valbool)
521 sock_set_flag(sk, bit);
522 else
523 sock_reset_flag(sk, bit);
524}
525
1da177e4
LT		 526/*
527 * This is meant for all protocols to use and covers goings on
528 * at the socket level. Everything here is generic.
529 */
530
531int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 532 char __user *optval, unsigned int optlen)
1da177e4 533{
2a91525c 534 struct sock *sk = sock->sk;
1da177e4
LT		 535 int val;
536 int valbool;
537 struct linger ling;
538 int ret = 0;
4ec93edb 539
1da177e4
LT		 540 /*
541 * Options without arguments
542 */
543
4878809f
DM		 544 if (optname == SO_BINDTODEVICE)
545 return sock_bindtodevice(sk, optval, optlen);
546
e71a4783
SH		 547 if (optlen < sizeof(int))
548 return -EINVAL;
4ec93edb 549
1da177e4
LT		 550 if (get_user(val, (int __user *)optval))
551 return -EFAULT;
4ec93edb 552
2a91525c 553 valbool = val ? 1 : 0;
1da177e4
LT		 554
555 lock_sock(sk);
556
2a91525c 557 switch (optname) {
e71a4783 558 case SO_DEBUG:
2a91525c 559 if (val && !capable(CAP_NET_ADMIN))
e71a4783 560 ret = -EACCES;
2a91525c 561 else
c0ef877b 562 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 563 break;
564 case SO_REUSEADDR:
565 sk->sk_reuse = valbool;
566 break;
567 case SO_TYPE:
49c794e9 568 case SO_PROTOCOL:
0d6038ee 569 case SO_DOMAIN:
e71a4783
SH		 570 case SO_ERROR:
571 ret = -ENOPROTOOPT;
572 break;
573 case SO_DONTROUTE:
c0ef877b 574 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 575 break;
576 case SO_BROADCAST:
577 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
578 break;
579 case SO_SNDBUF:
580 /* Don't error on this BSD doesn't and if you think
581 about it this is right. Otherwise apps have to
582 play 'guess the biggest size' games. RCVBUF/SNDBUF
583 are treated in BSD as hints */
584
585 if (val > sysctl_wmem_max)
586 val = sysctl_wmem_max;
b0573dea 587set_sndbuf:
e71a4783
SH		 588 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
589 if ((val * 2) < SOCK_MIN_SNDBUF)
590 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
591 else
592 sk->sk_sndbuf = val * 2;
1da177e4 593
e71a4783
SH		 594 /*
595 * Wake up sending tasks if we
596 * upped the value.
597 */
598 sk->sk_write_space(sk);
599 break;
1da177e4 600
e71a4783
SH		 601 case SO_SNDBUFFORCE:
602 if (!capable(CAP_NET_ADMIN)) {
603 ret = -EPERM;
604 break;
605 }
606 goto set_sndbuf;
b0573dea 607
e71a4783
SH		 608 case SO_RCVBUF:
609 /* Don't error on this BSD doesn't and if you think
610 about it this is right. Otherwise apps have to
611 play 'guess the biggest size' games. RCVBUF/SNDBUF
612 are treated in BSD as hints */
4ec93edb 613
e71a4783
SH		 614 if (val > sysctl_rmem_max)
615 val = sysctl_rmem_max;
b0573dea 616set_rcvbuf:
e71a4783
SH		 617 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
618 /*
619 * We double it on the way in to account for
620 * "struct sk_buff" etc. overhead. Applications
621 * assume that the SO_RCVBUF setting they make will
622 * allow that much actual data to be received on that
623 * socket.
624 *
625 * Applications are unaware that "struct sk_buff" and
626 * other overheads allocate from the receive buffer
627 * during socket buffer allocation.
628 *
629 * And after considering the possible alternatives,
630 * returning the value we actually used in getsockopt
631 * is the most desirable behavior.
632 */
633 if ((val * 2) < SOCK_MIN_RCVBUF)
634 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
635 else
636 sk->sk_rcvbuf = val * 2;
637 break;
638
639 case SO_RCVBUFFORCE:
640 if (!capable(CAP_NET_ADMIN)) {
641 ret = -EPERM;
1da177e4 642 break;
e71a4783
SH		 643 }
644 goto set_rcvbuf;
1da177e4 645
e71a4783 646 case SO_KEEPALIVE:
1da177e4 647#ifdef CONFIG_INET
e71a4783
SH		 648 if (sk->sk_protocol == IPPROTO_TCP)
649 tcp_set_keepalive(sk, valbool);
1da177e4 650#endif
e71a4783
SH		 651 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
652 break;
653
654 case SO_OOBINLINE:
655 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
656 break;
657
658 case SO_NO_CHECK:
659 sk->sk_no_check = valbool;
660 break;
661
662 case SO_PRIORITY:
663 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
664 sk->sk_priority = val;
665 else
666 ret = -EPERM;
667 break;
668
669 case SO_LINGER:
670 if (optlen < sizeof(ling)) {
671 ret = -EINVAL; /* 1003.1g */
1da177e4 672 break;
e71a4783 673 }
2a91525c 674 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 675 ret = -EFAULT;
1da177e4 676 break;
e71a4783
SH		 677 }
678 if (!ling.l_onoff)
679 sock_reset_flag(sk, SOCK_LINGER);
680 else {
1da177e4 681#if (BITS_PER_LONG == 32)
e71a4783
SH		 682 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
683 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 684 else
e71a4783
SH		 685#endif
686 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
687 sock_set_flag(sk, SOCK_LINGER);
688 }
689 break;
690
691 case SO_BSDCOMPAT:
692 sock_warn_obsolete_bsdism("setsockopt");
693 break;
694
695 case SO_PASSCRED:
696 if (valbool)
697 set_bit(SOCK_PASSCRED, &sock->flags);
698 else
699 clear_bit(SOCK_PASSCRED, &sock->flags);
700 break;
701
702 case SO_TIMESTAMP:
92f37fd2 703 case SO_TIMESTAMPNS:
e71a4783 704 if (valbool) {
92f37fd2
ED		 705 if (optname == SO_TIMESTAMP)
706 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
707 else
708 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 709 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 710 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 711 } else {
e71a4783 712 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 713 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
714 }
e71a4783
SH		 715 break;
716
20d49473
PO		 717 case SO_TIMESTAMPING:
718 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 719 ret = -EINVAL;
20d49473
PO		 720 break;
721 }
722 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
723 val & SOF_TIMESTAMPING_TX_HARDWARE);
724 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
725 val & SOF_TIMESTAMPING_TX_SOFTWARE);
726 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
727 val & SOF_TIMESTAMPING_RX_HARDWARE);
728 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
729 sock_enable_timestamp(sk,
730 SOCK_TIMESTAMPING_RX_SOFTWARE);
731 else
732 sock_disable_timestamp(sk,
08e29af3 733 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 734 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
735 val & SOF_TIMESTAMPING_SOFTWARE);
736 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
737 val & SOF_TIMESTAMPING_SYS_HARDWARE);
738 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
739 val & SOF_TIMESTAMPING_RAW_HARDWARE);
740 break;
741
e71a4783
SH		 742 case SO_RCVLOWAT:
743 if (val < 0)
744 val = INT_MAX;
745 sk->sk_rcvlowat = val ? : 1;
746 break;
747
748 case SO_RCVTIMEO:
749 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
750 break;
751
752 case SO_SNDTIMEO:
753 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
754 break;
1da177e4 755
e71a4783
SH		 756 case SO_ATTACH_FILTER:
757 ret = -EINVAL;
758 if (optlen == sizeof(struct sock_fprog)) {
759 struct sock_fprog fprog;
1da177e4 760
e71a4783
SH		 761 ret = -EFAULT;
762 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 763 break;
e71a4783
SH		 764
765 ret = sk_attach_filter(&fprog, sk);
766 }
767 break;
768
769 case SO_DETACH_FILTER:
55b33325 770 ret = sk_detach_filter(sk);
e71a4783 771 break;
1da177e4 772
e71a4783
SH		 773 case SO_PASSSEC:
774 if (valbool)
775 set_bit(SOCK_PASSSEC, &sock->flags);
776 else
777 clear_bit(SOCK_PASSSEC, &sock->flags);
778 break;
4a19ec58
LAT		 779 case SO_MARK:
780 if (!capable(CAP_NET_ADMIN))
781 ret = -EPERM;
2a91525c 782 else
4a19ec58 783 sk->sk_mark = val;
4a19ec58 784 break;
877ce7c1 785
1da177e4
LT		 786 /* We implement the SO_SNDLOWAT etc to
787 not be settable (1003.1g 5.3) */
3b885787 788 case SO_RXQ_OVFL:
8083f0fc 789 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 790 break;
6e3e939f
JB		 791
792 case SO_WIFI_STATUS:
793 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
794 break;
795
e71a4783
SH		 796 default:
797 ret = -ENOPROTOOPT;
798 break;
4ec93edb 799 }
1da177e4
LT		 800 release_sock(sk);
801 return ret;
802}
2a91525c 803EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 804
805
3f551f94
EB		 806void cred_to_ucred(struct pid *pid, const struct cred *cred,
807 struct ucred *ucred)
808{
809 ucred->pid = pid_vnr(pid);
810 ucred->uid = ucred->gid = -1;
811 if (cred) {
812 struct user_namespace *current_ns = current_user_ns();
813
814 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
815 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
816 }
817}
3924773a 818EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 819
1da177e4
LT		 820int sock_getsockopt(struct socket *sock, int level, int optname,
821 char __user *optval, int __user *optlen)
822{
823 struct sock *sk = sock->sk;
4ec93edb 824
e71a4783 825 union {
4ec93edb
YH		 826 int val;
827 struct linger ling;
1da177e4
LT		 828 struct timeval tm;
829 } v;
4ec93edb 830
4d0392be 831 int lv = sizeof(int);
1da177e4 832 int len;
4ec93edb 833
e71a4783 834 if (get_user(len, optlen))
4ec93edb 835 return -EFAULT;
e71a4783 836 if (len < 0)
1da177e4 837 return -EINVAL;
4ec93edb 838
50fee1de 839 memset(&v, 0, sizeof(v));
df0bca04 840
2a91525c 841 switch (optname) {
e71a4783
SH		 842 case SO_DEBUG:
843 v.val = sock_flag(sk, SOCK_DBG);
844 break;
845
846 case SO_DONTROUTE:
847 v.val = sock_flag(sk, SOCK_LOCALROUTE);
848 break;
849
850 case SO_BROADCAST:
851 v.val = !!sock_flag(sk, SOCK_BROADCAST);
852 break;
853
854 case SO_SNDBUF:
855 v.val = sk->sk_sndbuf;
856 break;
857
858 case SO_RCVBUF:
859 v.val = sk->sk_rcvbuf;
860 break;
861
862 case SO_REUSEADDR:
863 v.val = sk->sk_reuse;
864 break;
865
866 case SO_KEEPALIVE:
867 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
868 break;
869
870 case SO_TYPE:
871 v.val = sk->sk_type;
872 break;
873
49c794e9
JE		 874 case SO_PROTOCOL:
875 v.val = sk->sk_protocol;
876 break;
877
0d6038ee
JE		 878 case SO_DOMAIN:
879 v.val = sk->sk_family;
880 break;
881
e71a4783
SH		 882 case SO_ERROR:
883 v.val = -sock_error(sk);
2a91525c 884 if (v.val == 0)
e71a4783
SH		 885 v.val = xchg(&sk->sk_err_soft, 0);
886 break;
887
888 case SO_OOBINLINE:
889 v.val = !!sock_flag(sk, SOCK_URGINLINE);
890 break;
891
892 case SO_NO_CHECK:
893 v.val = sk->sk_no_check;
894 break;
895
896 case SO_PRIORITY:
897 v.val = sk->sk_priority;
898 break;
899
900 case SO_LINGER:
901 lv = sizeof(v.ling);
902 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
903 v.ling.l_linger = sk->sk_lingertime / HZ;
904 break;
905
906 case SO_BSDCOMPAT:
907 sock_warn_obsolete_bsdism("getsockopt");
908 break;
909
910 case SO_TIMESTAMP:
92f37fd2
ED		 911 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
912 !sock_flag(sk, SOCK_RCVTSTAMPNS);
913 break;
914
915 case SO_TIMESTAMPNS:
916 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 917 break;
918
20d49473
PO		 919 case SO_TIMESTAMPING:
920 v.val = 0;
921 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
922 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
924 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
925 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
926 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
928 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
932 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
934 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
935 break;
936
e71a4783 937 case SO_RCVTIMEO:
2a91525c 938 lv = sizeof(struct timeval);
e71a4783
SH		 939 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
940 v.tm.tv_sec = 0;
941 v.tm.tv_usec = 0;
942 } else {
943 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
944 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
945 }
946 break;
947
948 case SO_SNDTIMEO:
2a91525c 949 lv = sizeof(struct timeval);
e71a4783
SH		 950 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
951 v.tm.tv_sec = 0;
952 v.tm.tv_usec = 0;
953 } else {
954 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
955 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
956 }
957 break;
1da177e4 958
e71a4783
SH		 959 case SO_RCVLOWAT:
960 v.val = sk->sk_rcvlowat;
961 break;
1da177e4 962
e71a4783 963 case SO_SNDLOWAT:
2a91525c 964 v.val = 1;
e71a4783 965 break;
1da177e4 966
e71a4783
SH		 967 case SO_PASSCRED:
968 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
969 break;
1da177e4 970
e71a4783 971 case SO_PEERCRED:
109f6e39
EB		 972 {
973 struct ucred peercred;
974 if (len > sizeof(peercred))
975 len = sizeof(peercred);
976 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
977 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 978 return -EFAULT;
979 goto lenout;
109f6e39 980 }
1da177e4 981
e71a4783
SH		 982 case SO_PEERNAME:
983 {
984 char address[128];
985
986 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
987 return -ENOTCONN;
988 if (lv < len)
989 return -EINVAL;
990 if (copy_to_user(optval, address, len))
991 return -EFAULT;
992 goto lenout;
993 }
1da177e4 994
e71a4783
SH		 995 /* Dubious BSD thing... Probably nobody even uses it, but
996 * the UNIX standard wants it for whatever reason... -DaveM
997 */
998 case SO_ACCEPTCONN:
999 v.val = sk->sk_state == TCP_LISTEN;
1000 break;
1da177e4 1001
e71a4783
SH		 1002 case SO_PASSSEC:
1003 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1004 break;
877ce7c1 1005
e71a4783
SH		 1006 case SO_PEERSEC:
1007 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1008
4a19ec58
LAT		 1009 case SO_MARK:
1010 v.val = sk->sk_mark;
1011 break;
1012
3b885787
NH		 1013 case SO_RXQ_OVFL:
1014 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1015 break;
1016
6e3e939f
JB		 1017 case SO_WIFI_STATUS:
1018 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
1019 break;
1020
e71a4783
SH		 1021 default:
1022 return -ENOPROTOOPT;
1da177e4 1023 }
e71a4783 1024
1da177e4
LT		 1025 if (len > lv)
1026 len = lv;
1027 if (copy_to_user(optval, &v, len))
1028 return -EFAULT;
1029lenout:
4ec93edb
YH		 1030 if (put_user(len, optlen))
1031 return -EFAULT;
1032 return 0;
1da177e4
LT		 1033}
1034
a5b5bb9a
IM		 1035/*
1036 * Initialize an sk_lock.
1037 *
1038 * (We also register the sk_lock with the lock validator.)
1039 */
b6f99a21 1040static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1041{
ed07536e
PZ		 1042 sock_lock_init_class_and_name(sk,
1043 af_family_slock_key_strings[sk->sk_family],
1044 af_family_slock_keys + sk->sk_family,
1045 af_family_key_strings[sk->sk_family],
1046 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1047}
1048
4dc6dc71
ED		 1049/*
1050 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1051 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1052 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1053 */
f1a6c4da
PE		 1054static void sock_copy(struct sock *nsk, const struct sock *osk)
1055{
1056#ifdef CONFIG_SECURITY_NETWORK
1057 void *sptr = nsk->sk_security;
1058#endif
68835aba
ED		 1059 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1060
1061 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1062 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1063
f1a6c4da
PE		 1064#ifdef CONFIG_SECURITY_NETWORK
1065 nsk->sk_security = sptr;
1066 security_sk_clone(osk, nsk);
1067#endif
1068}
1069
fcbdf09d
OP		 1070/*
1071 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1072 * un-modified. Special care is taken when initializing object to zero.
1073 */
1074static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1075{
1076 if (offsetof(struct sock, sk_node.next) != 0)
1077 memset(sk, 0, offsetof(struct sock, sk_node.next));
1078 memset(&sk->sk_node.pprev, 0,
1079 size - offsetof(struct sock, sk_node.pprev));
1080}
1081
1082void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1083{
1084 unsigned long nulls1, nulls2;
1085
1086 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1087 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1088 if (nulls1 > nulls2)
1089 swap(nulls1, nulls2);
1090
1091 if (nulls1 != 0)
1092 memset((char *)sk, 0, nulls1);
1093 memset((char *)sk + nulls1 + sizeof(void *), 0,
1094 nulls2 - nulls1 - sizeof(void *));
1095 memset((char *)sk + nulls2 + sizeof(void *), 0,
1096 size - nulls2 - sizeof(void *));
1097}
1098EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1099
2e4afe7b
PE		 1100static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1101 int family)
c308c1b2
PE		 1102{
1103 struct sock *sk;
1104 struct kmem_cache *slab;
1105
1106 slab = prot->slab;
e912b114
ED		 1107 if (slab != NULL) {
1108 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1109 if (!sk)
1110 return sk;
1111 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1112 if (prot->clear_sk)
1113 prot->clear_sk(sk, prot->obj_size);
1114 else
1115 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1116 }
fcbdf09d 1117 } else
c308c1b2
PE		 1118 sk = kmalloc(prot->obj_size, priority);
1119
2e4afe7b 1120 if (sk != NULL) {
a98b65a3
VN		 1121 kmemcheck_annotate_bitfield(sk, flags);
1122
2e4afe7b
PE		 1123 if (security_sk_alloc(sk, family, priority))
1124 goto out_free;
1125
1126 if (!try_module_get(prot->owner))
1127 goto out_free_sec;
e022f0b4 1128 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1129 }
1130
c308c1b2 1131 return sk;
2e4afe7b
PE		 1132
1133out_free_sec:
1134 security_sk_free(sk);
1135out_free:
1136 if (slab != NULL)
1137 kmem_cache_free(slab, sk);
1138 else
1139 kfree(sk);
1140 return NULL;
c308c1b2
PE		 1141}
1142
1143static void sk_prot_free(struct proto *prot, struct sock *sk)
1144{
1145 struct kmem_cache *slab;
2e4afe7b 1146 struct module *owner;
c308c1b2 1147
2e4afe7b 1148 owner = prot->owner;
c308c1b2 1149 slab = prot->slab;
2e4afe7b
PE		 1150
1151 security_sk_free(sk);
c308c1b2
PE		 1152 if (slab != NULL)
1153 kmem_cache_free(slab, sk);
1154 else
1155 kfree(sk);
2e4afe7b 1156 module_put(owner);
c308c1b2
PE		 1157}
1158
f8451725
HX		 1159#ifdef CONFIG_CGROUPS
1160void sock_update_classid(struct sock *sk)
1161{
1144182a 1162 u32 classid;
f8451725 1163
1144182a
PM		 1164 rcu_read_lock(); /* doing current task, which cannot vanish. */
1165 classid = task_cls_classid(current);
1166 rcu_read_unlock();
f8451725
HX		 1167 if (classid && classid != sk->sk_classid)
1168 sk->sk_classid = classid;
1169}
82862742 1170EXPORT_SYMBOL(sock_update_classid);
5bc1421e
NH		 1171
1172void sock_update_netprioidx(struct sock *sk)
1173{
5bc1421e
NH		 1174 if (in_interrupt())
1175 return;
2b73bc65
NH		 1176
1177 sk->sk_cgrp_prioidx = task_netprioidx(current);
5bc1421e
NH		 1178}
1179EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1180#endif
1181
1da177e4
LT		 1182/**
1183 * sk_alloc - All socket objects are allocated here
c4ea43c5 1184 * @net: the applicable net namespace
4dc3b16b
PP		 1185 * @family: protocol family
1186 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1187 * @prot: struct proto associated with this new sock instance
1da177e4 1188 */
1b8d7ae4 1189struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1190 struct proto *prot)
1da177e4 1191{
c308c1b2 1192 struct sock *sk;
1da177e4 1193
154adbc8 1194 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1195 if (sk) {
154adbc8
PE		 1196 sk->sk_family = family;
1197 /*
1198 * See comment in struct sock definition to understand
1199 * why we need sk_prot_creator -acme
1200 */
1201 sk->sk_prot = sk->sk_prot_creator = prot;
1202 sock_lock_init(sk);
3b1e0a65 1203 sock_net_set(sk, get_net(net));
d66ee058 1204 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1205
1206 sock_update_classid(sk);
5bc1421e 1207 sock_update_netprioidx(sk);
1da177e4 1208 }
a79af59e 1209
2e4afe7b 1210 return sk;
1da177e4 1211}
2a91525c 1212EXPORT_SYMBOL(sk_alloc);
1da177e4 1213
2b85a34e 1214static void __sk_free(struct sock *sk)
1da177e4
LT		 1215{
1216 struct sk_filter *filter;
1da177e4
LT		 1217
1218 if (sk->sk_destruct)
1219 sk->sk_destruct(sk);
1220
a898def2
PM		 1221 filter = rcu_dereference_check(sk->sk_filter,
1222 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1223 if (filter) {
309dd5fc 1224 sk_filter_uncharge(sk, filter);
a9b3cd7f 1225 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1226 }
1227
08e29af3 1228 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1229
1230 if (atomic_read(&sk->sk_omem_alloc))
1231 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1232 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1233
109f6e39
EB		 1234 if (sk->sk_peer_cred)
1235 put_cred(sk->sk_peer_cred);
1236 put_pid(sk->sk_peer_pid);
3b1e0a65 1237 put_net(sock_net(sk));
c308c1b2 1238 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1239}
2b85a34e
ED		 1240
1241void sk_free(struct sock *sk)
1242{
1243 /*
25985edc 1244 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1245 * some packets are still in some tx queue.
1246 * If not null, sock_wfree() will call __sk_free(sk) later
1247 */
1248 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1249 __sk_free(sk);
1250}
2a91525c 1251EXPORT_SYMBOL(sk_free);
1da177e4 1252
edf02087 1253/*
25985edc
LDM		 1254 * Last sock_put should drop reference to sk->sk_net. It has already
1255 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1256 * is not an option.
25985edc 1257 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1258 * destroy it in the context of init_net.
1259 */
1260void sk_release_kernel(struct sock *sk)
1261{
1262 if (sk == NULL || sk->sk_socket == NULL)
1263 return;
1264
1265 sock_hold(sk);
1266 sock_release(sk->sk_socket);
65a18ec5 1267 release_net(sock_net(sk));
3b1e0a65 1268 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1269 sock_put(sk);
1270}
45af1754 1271EXPORT_SYMBOL(sk_release_kernel);
edf02087 1272
475f1b52
SR		 1273static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1274{
1275 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1276 sock_update_memcg(newsk);
1277}
1278
e56c57d0
ED		 1279/**
1280 * sk_clone_lock - clone a socket, and lock its clone
1281 * @sk: the socket to clone
1282 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1283 *
1284 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1285 */
1286struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1287{
8fd1d178 1288 struct sock *newsk;
87d11ceb 1289
8fd1d178 1290 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1291 if (newsk != NULL) {
1292 struct sk_filter *filter;
1293
892c141e 1294 sock_copy(newsk, sk);
87d11ceb
ACM		 1295
1296 /* SANITY */
3b1e0a65 1297 get_net(sock_net(newsk));
87d11ceb
ACM		 1298 sk_node_init(&newsk->sk_node);
1299 sock_lock_init(newsk);
1300 bh_lock_sock(newsk);
fa438ccf 1301 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1302 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1303
1304 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1305 /*
1306 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1307 */
1308 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1309 atomic_set(&newsk->sk_omem_alloc, 0);
1310 skb_queue_head_init(&newsk->sk_receive_queue);
1311 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1312#ifdef CONFIG_NET_DMA
1313 skb_queue_head_init(&newsk->sk_async_wait_queue);
1314#endif
87d11ceb 1315
b6c6712a 1316 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1317 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1318 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1319 af_callback_keys + newsk->sk_family,
1320 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1321
1322 newsk->sk_dst_cache = NULL;
1323 newsk->sk_wmem_queued = 0;
1324 newsk->sk_forward_alloc = 0;
1325 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1326 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1327
1328 sock_reset_flag(newsk, SOCK_DONE);
1329 skb_queue_head_init(&newsk->sk_error_queue);
1330
0d7da9dd 1331 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1332 if (filter != NULL)
1333 sk_filter_charge(newsk, filter);
1334
1335 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1336 /* It is still raw copy of parent, so invalidate
1337 * destructor and make plain sk_free() */
1338 newsk->sk_destruct = NULL;
b0691c8e 1339 bh_unlock_sock(newsk);
87d11ceb
ACM		 1340 sk_free(newsk);
1341 newsk = NULL;
1342 goto out;
1343 }
1344
1345 newsk->sk_err = 0;
1346 newsk->sk_priority = 0;
4dc6dc71
ED		 1347 /*
1348 * Before updating sk_refcnt, we must commit prior changes to memory
1349 * (Documentation/RCU/rculist_nulls.txt for details)
1350 */
1351 smp_wmb();
87d11ceb
ACM		 1352 atomic_set(&newsk->sk_refcnt, 2);
1353
1354 /*
1355 * Increment the counter in the same struct proto as the master
1356 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1357 * is the same as sk->sk_prot->socks, as this field was copied
1358 * with memcpy).
1359 *
1360 * This _changes_ the previous behaviour, where
1361 * tcp_create_openreq_child always was incrementing the
1362 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1363 * to be taken into account in all callers. -acme
1364 */
1365 sk_refcnt_debug_inc(newsk);
972692e0 1366 sk_set_socket(newsk, NULL);
43815482 1367 newsk->sk_wq = NULL;
87d11ceb 1368
f3f511e1
GC		 1369 sk_update_clone(sk, newsk);
1370
87d11ceb 1371 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1372 sk_sockets_allocated_inc(newsk);
704da560 1373
08e29af3 1374 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1375 net_enable_timestamp();
87d11ceb
ACM		 1376 }
1377out:
1378 return newsk;
1379}
e56c57d0 1380EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1381
9958089a
AK		 1382void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1383{
1384 __sk_dst_set(sk, dst);
1385 sk->sk_route_caps = dst->dev->features;
1386 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1387 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1388 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1389 if (sk_can_gso(sk)) {
82cc1a7a 1390 if (dst->header_len) {
9958089a 1391 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1392 } else {
9958089a 1393 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1394 sk->sk_gso_max_size = dst->dev->gso_max_size;
1395 }
9958089a
AK		 1396 }
1397}
1398EXPORT_SYMBOL_GPL(sk_setup_caps);
1399
1da177e4
LT		 1400void __init sk_init(void)
1401{
4481374c 1402 if (totalram_pages <= 4096) {
1da177e4
LT		 1403 sysctl_wmem_max = 32767;
1404 sysctl_rmem_max = 32767;
1405 sysctl_wmem_default = 32767;
1406 sysctl_rmem_default = 32767;
4481374c 1407 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1408 sysctl_wmem_max = 131071;
1409 sysctl_rmem_max = 131071;
1410 }
1411}
1412
1413/*
1414 * Simple resource managers for sockets.
1415 */
1416
1417
4ec93edb
YH		 1418/*
1419 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1420 */
1421void sock_wfree(struct sk_buff *skb)
1422{
1423 struct sock *sk = skb->sk;
d99927f4 1424 unsigned int len = skb->truesize;
1da177e4 1425
d99927f4
ED		 1426 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1427 /*
1428 * Keep a reference on sk_wmem_alloc, this will be released
1429 * after sk_write_space() call
1430 */
1431 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1432 sk->sk_write_space(sk);
d99927f4
ED		 1433 len = 1;
1434 }
2b85a34e 1435 /*
d99927f4
ED		 1436 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1437 * could not do because of in-flight packets
2b85a34e 1438 */
d99927f4 1439 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1440 __sk_free(sk);
1da177e4 1441}
2a91525c 1442EXPORT_SYMBOL(sock_wfree);
1da177e4 1443
4ec93edb
YH		 1444/*
1445 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1446 */
1447void sock_rfree(struct sk_buff *skb)
1448{
1449 struct sock *sk = skb->sk;
d361fd59 1450 unsigned int len = skb->truesize;
1da177e4 1451
d361fd59
ED		 1452 atomic_sub(len, &sk->sk_rmem_alloc);
1453 sk_mem_uncharge(sk, len);
1da177e4 1454}
2a91525c 1455EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1456
1457
1458int sock_i_uid(struct sock *sk)
1459{
1460 int uid;
1461
f064af1e 1462 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1463 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1464 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1465 return uid;
1466}
2a91525c 1467EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1468
1469unsigned long sock_i_ino(struct sock *sk)
1470{
1471 unsigned long ino;
1472
f064af1e 1473 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1474 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1475 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1476 return ino;
1477}
2a91525c 1478EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1479
1480/*
1481 * Allocate a skb from the socket's send buffer.
1482 */
86a76caf 1483struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1484 gfp_t priority)
1da177e4
LT		 1485{
1486 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1487 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1488 if (skb) {
1489 skb_set_owner_w(skb, sk);
1490 return skb;
1491 }
1492 }
1493 return NULL;
1494}
2a91525c 1495EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1496
1497/*
1498 * Allocate a skb from the socket's receive buffer.
4ec93edb 1499 */
86a76caf 1500struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1501 gfp_t priority)
1da177e4
LT		 1502{
1503 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1504 struct sk_buff *skb = alloc_skb(size, priority);
1505 if (skb) {
1506 skb_set_owner_r(skb, sk);
1507 return skb;
1508 }
1509 }
1510 return NULL;
1511}
1512
4ec93edb 1513/*
1da177e4 1514 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1515 */
dd0fc66f 1516void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1517{
1518 if ((unsigned)size <= sysctl_optmem_max &&
1519 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1520 void *mem;
1521 /* First do the add, to avoid the race if kmalloc
4ec93edb 1522 * might sleep.
1da177e4
LT		 1523 */
1524 atomic_add(size, &sk->sk_omem_alloc);
1525 mem = kmalloc(size, priority);
1526 if (mem)
1527 return mem;
1528 atomic_sub(size, &sk->sk_omem_alloc);
1529 }
1530 return NULL;
1531}
2a91525c 1532EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1533
1534/*
1535 * Free an option memory block.
1536 */
1537void sock_kfree_s(struct sock *sk, void *mem, int size)
1538{
1539 kfree(mem);
1540 atomic_sub(size, &sk->sk_omem_alloc);
1541}
2a91525c 1542EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1543
1544/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1545 I think, these locks should be removed for datagram sockets.
1546 */
2a91525c 1547static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1548{
1549 DEFINE_WAIT(wait);
1550
1551 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1552 for (;;) {
1553 if (!timeo)
1554 break;
1555 if (signal_pending(current))
1556 break;
1557 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1558 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1559 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1560 break;
1561 if (sk->sk_shutdown & SEND_SHUTDOWN)
1562 break;
1563 if (sk->sk_err)
1564 break;
1565 timeo = schedule_timeout(timeo);
1566 }
aa395145 1567 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1568 return timeo;
1569}
1570
1571
1572/*
1573 * Generic send/receive buffer handlers
1574 */
1575
4cc7f68d
HX		 1576struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1577 unsigned long data_len, int noblock,
1578 int *errcode)
1da177e4
LT		 1579{
1580 struct sk_buff *skb;
7d877f3b 1581 gfp_t gfp_mask;
1da177e4
LT		 1582 long timeo;
1583 int err;
1584
1585 gfp_mask = sk->sk_allocation;
1586 if (gfp_mask & __GFP_WAIT)
1587 gfp_mask |= __GFP_REPEAT;
1588
1589 timeo = sock_sndtimeo(sk, noblock);
1590 while (1) {
1591 err = sock_error(sk);
1592 if (err != 0)
1593 goto failure;
1594
1595 err = -EPIPE;
1596 if (sk->sk_shutdown & SEND_SHUTDOWN)
1597 goto failure;
1598
1599 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1600 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1601 if (skb) {
1602 int npages;
1603 int i;
1604
1605 /* No pages, we're done... */
1606 if (!data_len)
1607 break;
1608
1609 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1610 skb->truesize += data_len;
1611 skb_shinfo(skb)->nr_frags = npages;
1612 for (i = 0; i < npages; i++) {
1613 struct page *page;
1da177e4
LT		 1614
1615 page = alloc_pages(sk->sk_allocation, 0);
1616 if (!page) {
1617 err = -ENOBUFS;
1618 skb_shinfo(skb)->nr_frags = i;
1619 kfree_skb(skb);
1620 goto failure;
1621 }
1622
ea2ab693
IC		 1623 __skb_fill_page_desc(skb, i,
1624 page, 0,
1625 (data_len >= PAGE_SIZE ?
1626 PAGE_SIZE :
1627 data_len));
1da177e4
LT		 1628 data_len -= PAGE_SIZE;
1629 }
1630
1631 /* Full success... */
1632 break;
1633 }
1634 err = -ENOBUFS;
1635 goto failure;
1636 }
1637 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1638 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1639 err = -EAGAIN;
1640 if (!timeo)
1641 goto failure;
1642 if (signal_pending(current))
1643 goto interrupted;
1644 timeo = sock_wait_for_wmem(sk, timeo);
1645 }
1646
1647 skb_set_owner_w(skb, sk);
1648 return skb;
1649
1650interrupted:
1651 err = sock_intr_errno(timeo);
1652failure:
1653 *errcode = err;
1654 return NULL;
1655}
4cc7f68d 1656EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1657
4ec93edb 1658struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1659 int noblock, int *errcode)
1660{
1661 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1662}
2a91525c 1663EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1664
1665static void __lock_sock(struct sock *sk)
f39234d6
NK		 1666 __releases(&sk->sk_lock.slock)
1667 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1668{
1669 DEFINE_WAIT(wait);
1670
e71a4783 1671 for (;;) {
1da177e4
LT		 1672 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1673 TASK_UNINTERRUPTIBLE);
1674 spin_unlock_bh(&sk->sk_lock.slock);
1675 schedule();
1676 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1677 if (!sock_owned_by_user(sk))
1da177e4
LT		 1678 break;
1679 }
1680 finish_wait(&sk->sk_lock.wq, &wait);
1681}
1682
1683static void __release_sock(struct sock *sk)
f39234d6
NK		 1684 __releases(&sk->sk_lock.slock)
1685 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1686{
1687 struct sk_buff *skb = sk->sk_backlog.head;
1688
1689 do {
1690 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1691 bh_unlock_sock(sk);
1692
1693 do {
1694 struct sk_buff *next = skb->next;
1695
7fee226a 1696 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1697 skb->next = NULL;
c57943a1 1698 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1699
1700 /*
1701 * We are in process context here with softirqs
1702 * disabled, use cond_resched_softirq() to preempt.
1703 * This is safe to do because we've taken the backlog
1704 * queue private:
1705 */
1706 cond_resched_softirq();
1707
1708 skb = next;
1709 } while (skb != NULL);
1710
1711 bh_lock_sock(sk);
e71a4783 1712 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1713
1714 /*
1715 * Doing the zeroing here guarantee we can not loop forever
1716 * while a wild producer attempts to flood us.
1717 */
1718 sk->sk_backlog.len = 0;
1da177e4
LT		 1719}
1720
1721/**
1722 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1723 * @sk: sock to wait on
1724 * @timeo: for how long
1da177e4
LT		 1725 *
1726 * Now socket state including sk->sk_err is changed only under lock,
1727 * hence we may omit checks after joining wait queue.
1728 * We check receive queue before schedule() only as optimization;
1729 * it is very likely that release_sock() added new data.
1730 */
1731int sk_wait_data(struct sock *sk, long *timeo)
1732{
1733 int rc;
1734 DEFINE_WAIT(wait);
1735
aa395145 1736 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1737 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1738 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1739 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1740 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1741 return rc;
1742}
1da177e4
LT		 1743EXPORT_SYMBOL(sk_wait_data);
1744
3ab224be
HA		 1745/**
1746 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1747 * @sk: socket
1748 * @size: memory size to allocate
1749 * @kind: allocation type
1750 *
1751 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1752 * rmem allocation. This function assumes that protocols which have
1753 * memory_pressure use sk_wmem_queued as write buffer accounting.
1754 */
1755int __sk_mem_schedule(struct sock *sk, int size, int kind)
1756{
1757 struct proto *prot = sk->sk_prot;
1758 int amt = sk_mem_pages(size);
8d987e5c 1759 long allocated;
e1aab161 1760 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1761
1762 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1763
e1aab161 1764 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1765
1766 /* Under limit. */
e1aab161
GC		 1767 if (parent_status == UNDER_LIMIT &&
1768 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1769 sk_leave_memory_pressure(sk);
3ab224be
HA		 1770 return 1;
1771 }
1772
e1aab161
GC		 1773 /* Under pressure. (we or our parents) */
1774 if ((parent_status > SOFT_LIMIT) ||
1775 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1776 sk_enter_memory_pressure(sk);
3ab224be 1777
e1aab161
GC		 1778 /* Over hard limit (we or our parents) */
1779 if ((parent_status == OVER_LIMIT) ||
1780 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1781 goto suppress_allocation;
1782
1783 /* guarantee minimum buffer size under pressure */
1784 if (kind == SK_MEM_RECV) {
1785 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1786 return 1;
180d8cd9 1787
3ab224be
HA		 1788 } else { /* SK_MEM_SEND */
1789 if (sk->sk_type == SOCK_STREAM) {
1790 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1791 return 1;
1792 } else if (atomic_read(&sk->sk_wmem_alloc) <
1793 prot->sysctl_wmem[0])
1794 return 1;
1795 }
1796
180d8cd9 1797 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1798 int alloc;
1799
180d8cd9 1800 if (!sk_under_memory_pressure(sk))
1748376b 1801 return 1;
180d8cd9
GC		 1802 alloc = sk_sockets_allocated_read_positive(sk);
1803 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1804 sk_mem_pages(sk->sk_wmem_queued +
1805 atomic_read(&sk->sk_rmem_alloc) +
1806 sk->sk_forward_alloc))
1807 return 1;
1808 }
1809
1810suppress_allocation:
1811
1812 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1813 sk_stream_moderate_sndbuf(sk);
1814
1815 /* Fail only if socket is _under_ its sndbuf.
1816 * In this case we cannot block, so that we have to fail.
1817 */
1818 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1819 return 1;
1820 }
1821
3847ce32
SM		 1822 trace_sock_exceed_buf_limit(sk, prot, allocated);
1823
3ab224be
HA		 1824 /* Alas. Undo changes. */
1825 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1826
0e90b31f 1827 sk_memory_allocated_sub(sk, amt);
180d8cd9 1828
3ab224be
HA		 1829 return 0;
1830}
3ab224be
HA		 1831EXPORT_SYMBOL(__sk_mem_schedule);
1832
1833/**
1834 * __sk_reclaim - reclaim memory_allocated
1835 * @sk: socket
1836 */
1837void __sk_mem_reclaim(struct sock *sk)
1838{
180d8cd9 1839 sk_memory_allocated_sub(sk,
0e90b31f 1840 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
3ab224be
HA		 1841 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1842
180d8cd9
GC		 1843 if (sk_under_memory_pressure(sk) &&
1844 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1845 sk_leave_memory_pressure(sk);
3ab224be 1846}
3ab224be
HA		 1847EXPORT_SYMBOL(__sk_mem_reclaim);
1848
1849
1da177e4
LT		 1850/*
1851 * Set of default routines for initialising struct proto_ops when
1852 * the protocol does not support a particular function. In certain
1853 * cases where it makes no sense for a protocol to have a "do nothing"
1854 * function, some default processing is provided.
1855 */
1856
1857int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1858{
1859 return -EOPNOTSUPP;
1860}
2a91525c 1861EXPORT_SYMBOL(sock_no_bind);
1da177e4 1862
4ec93edb 1863int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1864 int len, int flags)
1865{
1866 return -EOPNOTSUPP;
1867}
2a91525c 1868EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1869
1870int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1871{
1872 return -EOPNOTSUPP;
1873}
2a91525c 1874EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1875
1876int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1877{
1878 return -EOPNOTSUPP;
1879}
2a91525c 1880EXPORT_SYMBOL(sock_no_accept);
1da177e4 1881
4ec93edb 1882int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1883 int *len, int peer)
1884{
1885 return -EOPNOTSUPP;
1886}
2a91525c 1887EXPORT_SYMBOL(sock_no_getname);
1da177e4 1888
2a91525c 1889unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1890{
1891 return 0;
1892}
2a91525c 1893EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1894
1895int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1896{
1897 return -EOPNOTSUPP;
1898}
2a91525c 1899EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1900
1901int sock_no_listen(struct socket *sock, int backlog)
1902{
1903 return -EOPNOTSUPP;
1904}
2a91525c 1905EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1906
1907int sock_no_shutdown(struct socket *sock, int how)
1908{
1909 return -EOPNOTSUPP;
1910}
2a91525c 1911EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1912
1913int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1914 char __user *optval, unsigned int optlen)
1da177e4
LT		 1915{
1916 return -EOPNOTSUPP;
1917}
2a91525c 1918EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1919
1920int sock_no_getsockopt(struct socket *sock, int level, int optname,
1921 char __user *optval, int __user *optlen)
1922{
1923 return -EOPNOTSUPP;
1924}
2a91525c 1925EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1926
1927int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1928 size_t len)
1929{
1930 return -EOPNOTSUPP;
1931}
2a91525c 1932EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1933
1934int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1935 size_t len, int flags)
1936{
1937 return -EOPNOTSUPP;
1938}
2a91525c 1939EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1940
1941int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1942{
1943 /* Mirror missing mmap method error code */
1944 return -ENODEV;
1945}
2a91525c 1946EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1947
1948ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1949{
1950 ssize_t res;
1951 struct msghdr msg = {.msg_flags = flags};
1952 struct kvec iov;
1953 char *kaddr = kmap(page);
1954 iov.iov_base = kaddr + offset;
1955 iov.iov_len = size;
1956 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1957 kunmap(page);
1958 return res;
1959}
2a91525c 1960EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1961
1962/*
1963 * Default Socket Callbacks
1964 */
1965
1966static void sock_def_wakeup(struct sock *sk)
1967{
43815482
ED		 1968 struct socket_wq *wq;
1969
1970 rcu_read_lock();
1971 wq = rcu_dereference(sk->sk_wq);
1972 if (wq_has_sleeper(wq))
1973 wake_up_interruptible_all(&wq->wait);
1974 rcu_read_unlock();
1da177e4
LT		 1975}
1976
1977static void sock_def_error_report(struct sock *sk)
1978{
43815482
ED		 1979 struct socket_wq *wq;
1980
1981 rcu_read_lock();
1982 wq = rcu_dereference(sk->sk_wq);
1983 if (wq_has_sleeper(wq))
1984 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1985 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1986 rcu_read_unlock();
1da177e4
LT		 1987}
1988
1989static void sock_def_readable(struct sock *sk, int len)
1990{
43815482
ED		 1991 struct socket_wq *wq;
1992
1993 rcu_read_lock();
1994 wq = rcu_dereference(sk->sk_wq);
1995 if (wq_has_sleeper(wq))
2c6607c6 1996 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1997 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1998 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1999 rcu_read_unlock();
1da177e4
LT		 2000}
2001
2002static void sock_def_write_space(struct sock *sk)
2003{
43815482
ED		 2004 struct socket_wq *wq;
2005
2006 rcu_read_lock();
1da177e4
LT		 2007
2008 /* Do not wake up a writer until he can make "significant"
2009 * progress. --DaveM
2010 */
e71a4783 2011 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2012 wq = rcu_dereference(sk->sk_wq);
2013 if (wq_has_sleeper(wq))
2014 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2015 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2016
2017 /* Should agree with poll, otherwise some programs break */
2018 if (sock_writeable(sk))
8d8ad9d7 2019 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2020 }
2021
43815482 2022 rcu_read_unlock();
1da177e4
LT		 2023}
2024
2025static void sock_def_destruct(struct sock *sk)
2026{
a51482bd 2027 kfree(sk->sk_protinfo);
1da177e4
LT		 2028}
2029
2030void sk_send_sigurg(struct sock *sk)
2031{
2032 if (sk->sk_socket && sk->sk_socket->file)
2033 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2034 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2035}
2a91525c 2036EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2037
2038void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2039 unsigned long expires)
2040{
2041 if (!mod_timer(timer, expires))
2042 sock_hold(sk);
2043}
1da177e4
LT		 2044EXPORT_SYMBOL(sk_reset_timer);
2045
2046void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2047{
2048 if (timer_pending(timer) && del_timer(timer))
2049 __sock_put(sk);
2050}
1da177e4
LT		 2051EXPORT_SYMBOL(sk_stop_timer);
2052
2053void sock_init_data(struct socket *sock, struct sock *sk)
2054{
2055 skb_queue_head_init(&sk->sk_receive_queue);
2056 skb_queue_head_init(&sk->sk_write_queue);
2057 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2058#ifdef CONFIG_NET_DMA
2059 skb_queue_head_init(&sk->sk_async_wait_queue);
2060#endif
1da177e4
LT		 2061
2062 sk->sk_send_head = NULL;
2063
2064 init_timer(&sk->sk_timer);
4ec93edb 2065
1da177e4
LT		 2066 sk->sk_allocation = GFP_KERNEL;
2067 sk->sk_rcvbuf = sysctl_rmem_default;
2068 sk->sk_sndbuf = sysctl_wmem_default;
2069 sk->sk_state = TCP_CLOSE;
972692e0 2070 sk_set_socket(sk, sock);
1da177e4
LT		 2071
2072 sock_set_flag(sk, SOCK_ZAPPED);
2073
e71a4783 2074 if (sock) {
1da177e4 2075 sk->sk_type = sock->type;
43815482 2076 sk->sk_wq = sock->wq;
1da177e4
LT		 2077 sock->sk = sk;
2078 } else
43815482 2079 sk->sk_wq = NULL;
1da177e4 2080
b6c6712a 2081 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2082 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2083 lockdep_set_class_and_name(&sk->sk_callback_lock,
2084 af_callback_keys + sk->sk_family,
2085 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2086
2087 sk->sk_state_change = sock_def_wakeup;
2088 sk->sk_data_ready = sock_def_readable;
2089 sk->sk_write_space = sock_def_write_space;
2090 sk->sk_error_report = sock_def_error_report;
2091 sk->sk_destruct = sock_def_destruct;
2092
2093 sk->sk_sndmsg_page = NULL;
2094 sk->sk_sndmsg_off = 0;
2095
109f6e39
EB		 2096 sk->sk_peer_pid = NULL;
2097 sk->sk_peer_cred = NULL;
1da177e4
LT		 2098 sk->sk_write_pending = 0;
2099 sk->sk_rcvlowat = 1;
2100 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2101 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2102
f37f0afb 2103 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2104
4dc6dc71
ED		 2105 /*
2106 * Before updating sk_refcnt, we must commit prior changes to memory
2107 * (Documentation/RCU/rculist_nulls.txt for details)
2108 */
2109 smp_wmb();
1da177e4 2110 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2111 atomic_set(&sk->sk_drops, 0);
1da177e4 2112}
2a91525c 2113EXPORT_SYMBOL(sock_init_data);
1da177e4 2114
b5606c2d 2115void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2116{
2117 might_sleep();
a5b5bb9a 2118 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2119 if (sk->sk_lock.owned)
1da177e4 2120 __lock_sock(sk);
d2e9117c 2121 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2122 spin_unlock(&sk->sk_lock.slock);
2123 /*
2124 * The sk_lock has mutex_lock() semantics here:
2125 */
fcc70d5f 2126 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2127 local_bh_enable();
1da177e4 2128}
fcc70d5f 2129EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2130
b5606c2d 2131void release_sock(struct sock *sk)
1da177e4 2132{
a5b5bb9a
IM		 2133 /*
2134 * The sk_lock has mutex_unlock() semantics:
2135 */
2136 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2137
2138 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2139 if (sk->sk_backlog.tail)
2140 __release_sock(sk);
d2e9117c 2141 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2142 if (waitqueue_active(&sk->sk_lock.wq))
2143 wake_up(&sk->sk_lock.wq);
2144 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2145}
2146EXPORT_SYMBOL(release_sock);
2147
8a74ad60
ED		 2148/**
2149 * lock_sock_fast - fast version of lock_sock
2150 * @sk: socket
2151 *
2152 * This version should be used for very small section, where process wont block
2153 * return false if fast path is taken
2154 * sk_lock.slock locked, owned = 0, BH disabled
2155 * return true if slow path is taken
2156 * sk_lock.slock unlocked, owned = 1, BH enabled
2157 */
2158bool lock_sock_fast(struct sock *sk)
2159{
2160 might_sleep();
2161 spin_lock_bh(&sk->sk_lock.slock);
2162
2163 if (!sk->sk_lock.owned)
2164 /*
2165 * Note : We must disable BH
2166 */
2167 return false;
2168
2169 __lock_sock(sk);
2170 sk->sk_lock.owned = 1;
2171 spin_unlock(&sk->sk_lock.slock);
2172 /*
2173 * The sk_lock has mutex_lock() semantics here:
2174 */
2175 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2176 local_bh_enable();
2177 return true;
2178}
2179EXPORT_SYMBOL(lock_sock_fast);
2180
1da177e4 2181int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2182{
b7aa0bf7 2183 struct timeval tv;
1da177e4 2184 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2185 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2186 tv = ktime_to_timeval(sk->sk_stamp);
2187 if (tv.tv_sec == -1)
1da177e4 2188 return -ENOENT;
b7aa0bf7
ED		 2189 if (tv.tv_sec == 0) {
2190 sk->sk_stamp = ktime_get_real();
2191 tv = ktime_to_timeval(sk->sk_stamp);
2192 }
2193 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2194}
1da177e4
LT		 2195EXPORT_SYMBOL(sock_get_timestamp);
2196
ae40eb1e
ED		 2197int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2198{
2199 struct timespec ts;
2200 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2201 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2202 ts = ktime_to_timespec(sk->sk_stamp);
2203 if (ts.tv_sec == -1)
2204 return -ENOENT;
2205 if (ts.tv_sec == 0) {
2206 sk->sk_stamp = ktime_get_real();
2207 ts = ktime_to_timespec(sk->sk_stamp);
2208 }
2209 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2210}
2211EXPORT_SYMBOL(sock_get_timestampns);
2212
20d49473 2213void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2214{
20d49473 2215 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2216 unsigned long previous_flags = sk->sk_flags;
2217
20d49473
PO		 2218 sock_set_flag(sk, flag);
2219 /*
2220 * we just set one of the two flags which require net
2221 * time stamping, but time stamping might have been on
2222 * already because of the other one
2223 */
08e29af3 2224 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2225 net_enable_timestamp();
1da177e4
LT		 2226 }
2227}
1da177e4
LT		 2228
2229/*
2230 * Get a socket option on an socket.
2231 *
2232 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2233 * asynchronous errors should be reported by getsockopt. We assume
2234 * this means if you specify SO_ERROR (otherwise whats the point of it).
2235 */
2236int sock_common_getsockopt(struct socket *sock, int level, int optname,
2237 char __user *optval, int __user *optlen)
2238{
2239 struct sock *sk = sock->sk;
2240
2241 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2242}
1da177e4
LT		 2243EXPORT_SYMBOL(sock_common_getsockopt);
2244
3fdadf7d 2245#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2246int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2247 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2248{
2249 struct sock *sk = sock->sk;
2250
1e51f951 2251 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2252 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2253 optval, optlen);
3fdadf7d
DM		 2254 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2255}
2256EXPORT_SYMBOL(compat_sock_common_getsockopt);
2257#endif
2258
1da177e4
LT		 2259int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2260 struct msghdr *msg, size_t size, int flags)
2261{
2262 struct sock *sk = sock->sk;
2263 int addr_len = 0;
2264 int err;
2265
2266 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2267 flags & ~MSG_DONTWAIT, &addr_len);
2268 if (err >= 0)
2269 msg->msg_namelen = addr_len;
2270 return err;
2271}
1da177e4
LT		 2272EXPORT_SYMBOL(sock_common_recvmsg);
2273
2274/*
2275 * Set socket options on an inet socket.
2276 */
2277int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2278 char __user *optval, unsigned int optlen)
1da177e4
LT		 2279{
2280 struct sock *sk = sock->sk;
2281
2282 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2283}
1da177e4
LT		 2284EXPORT_SYMBOL(sock_common_setsockopt);
2285
3fdadf7d 2286#ifdef CONFIG_COMPAT
543d9cfe 2287int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2288 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2289{
2290 struct sock *sk = sock->sk;
2291
543d9cfe
ACM		 2292 if (sk->sk_prot->compat_setsockopt != NULL)
2293 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2294 optval, optlen);
3fdadf7d
DM		 2295 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2296}
2297EXPORT_SYMBOL(compat_sock_common_setsockopt);
2298#endif
2299
1da177e4
LT		 2300void sk_common_release(struct sock *sk)
2301{
2302 if (sk->sk_prot->destroy)
2303 sk->sk_prot->destroy(sk);
2304
2305 /*
2306 * Observation: when sock_common_release is called, processes have
2307 * no access to socket. But net still has.
2308 * Step one, detach it from networking:
2309 *
2310 * A. Remove from hash tables.
2311 */
2312
2313 sk->sk_prot->unhash(sk);
2314
2315 /*
2316 * In this point socket cannot receive new packets, but it is possible
2317 * that some packets are in flight because some CPU runs receiver and
2318 * did hash table lookup before we unhashed socket. They will achieve
2319 * receive queue and will be purged by socket destructor.
2320 *
2321 * Also we still have packets pending on receive queue and probably,
2322 * our own packets waiting in device queues. sock_destroy will drain
2323 * receive queue, but transmitted packets will delay socket destruction
2324 * until the last reference will be released.
2325 */
2326
2327 sock_orphan(sk);
2328
2329 xfrm_sk_free_policy(sk);
2330
e6848976 2331 sk_refcnt_debug_release(sk);
1da177e4
LT		 2332 sock_put(sk);
2333}
1da177e4
LT		 2334EXPORT_SYMBOL(sk_common_release);
2335
13ff3d6f
PE		 2336#ifdef CONFIG_PROC_FS
2337#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2338struct prot_inuse {
2339 int val[PROTO_INUSE_NR];
2340};
13ff3d6f
PE		 2341
2342static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2343
2344#ifdef CONFIG_NET_NS
2345void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2346{
d6d9ca0f 2347 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2348}
2349EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2350
2351int sock_prot_inuse_get(struct net *net, struct proto *prot)
2352{
2353 int cpu, idx = prot->inuse_idx;
2354 int res = 0;
2355
2356 for_each_possible_cpu(cpu)
2357 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2358
2359 return res >= 0 ? res : 0;
2360}
2361EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2362
2c8c1e72 2363static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2364{
2365 net->core.inuse = alloc_percpu(struct prot_inuse);
2366 return net->core.inuse ? 0 : -ENOMEM;
2367}
2368
2c8c1e72 2369static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2370{
2371 free_percpu(net->core.inuse);
2372}
2373
2374static struct pernet_operations net_inuse_ops = {
2375 .init = sock_inuse_init_net,
2376 .exit = sock_inuse_exit_net,
2377};
2378
2379static __init int net_inuse_init(void)
2380{
2381 if (register_pernet_subsys(&net_inuse_ops))
2382 panic("Cannot initialize net inuse counters");
2383
2384 return 0;
2385}
2386
2387core_initcall(net_inuse_init);
2388#else
1338d466
PE		 2389static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2390
c29a0bc4 2391void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2392{
d6d9ca0f 2393 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2394}
2395EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2396
c29a0bc4 2397int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2398{
2399 int cpu, idx = prot->inuse_idx;
2400 int res = 0;
2401
2402 for_each_possible_cpu(cpu)
2403 res += per_cpu(prot_inuse, cpu).val[idx];
2404
2405 return res >= 0 ? res : 0;
2406}
2407EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2408#endif
13ff3d6f
PE		 2409
2410static void assign_proto_idx(struct proto *prot)
2411{
2412 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2413
2414 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2415 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2416 return;
2417 }
2418
2419 set_bit(prot->inuse_idx, proto_inuse_idx);
2420}
2421
2422static void release_proto_idx(struct proto *prot)
2423{
2424 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2425 clear_bit(prot->inuse_idx, proto_inuse_idx);
2426}
2427#else
2428static inline void assign_proto_idx(struct proto *prot)
2429{
2430}
2431
2432static inline void release_proto_idx(struct proto *prot)
2433{
2434}
2435#endif
2436
b733c007
PE		 2437int proto_register(struct proto *prot, int alloc_slab)
2438{
1da177e4
LT		 2439 if (alloc_slab) {
2440 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2441 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2442 NULL);
1da177e4
LT		 2443
2444 if (prot->slab == NULL) {
2445 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2446 prot->name);
60e7663d 2447 goto out;
1da177e4 2448 }
2e6599cb
ACM		 2449
2450 if (prot->rsk_prot != NULL) {
faf23422 2451 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2452 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2453 goto out_free_sock_slab;
2454
7e56b5d6 2455 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2456 prot->rsk_prot->obj_size, 0,
20c2df83 2457 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2458
2459 if (prot->rsk_prot->slab == NULL) {
2460 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2461 prot->name);
2462 goto out_free_request_sock_slab_name;
2463 }
2464 }
8feaf0c0 2465
6d6ee43e 2466 if (prot->twsk_prot != NULL) {
faf23422 2467 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2468
7e56b5d6 2469 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2470 goto out_free_request_sock_slab;
2471
6d6ee43e 2472 prot->twsk_prot->twsk_slab =
7e56b5d6 2473 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2474 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2475 0,
2476 SLAB_HWCACHE_ALIGN |
2477 prot->slab_flags,
20c2df83 2478 NULL);
6d6ee43e 2479 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2480 goto out_free_timewait_sock_slab_name;
2481 }
1da177e4
LT		 2482 }
2483
36b77a52 2484 mutex_lock(&proto_list_mutex);
1da177e4 2485 list_add(&prot->node, &proto_list);
13ff3d6f 2486 assign_proto_idx(prot);
36b77a52 2487 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2488 return 0;
2489
8feaf0c0 2490out_free_timewait_sock_slab_name:
7e56b5d6 2491 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2492out_free_request_sock_slab:
2493 if (prot->rsk_prot && prot->rsk_prot->slab) {
2494 kmem_cache_destroy(prot->rsk_prot->slab);
2495 prot->rsk_prot->slab = NULL;
2496 }
2e6599cb 2497out_free_request_sock_slab_name:
72150e9b
DC		 2498 if (prot->rsk_prot)
2499 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2500out_free_sock_slab:
2501 kmem_cache_destroy(prot->slab);
2502 prot->slab = NULL;
b733c007
PE		 2503out:
2504 return -ENOBUFS;
1da177e4 2505}
1da177e4
LT		 2506EXPORT_SYMBOL(proto_register);
2507
2508void proto_unregister(struct proto *prot)
2509{
36b77a52 2510 mutex_lock(&proto_list_mutex);
13ff3d6f 2511 release_proto_idx(prot);
0a3f4358 2512 list_del(&prot->node);
36b77a52 2513 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2514
2515 if (prot->slab != NULL) {
2516 kmem_cache_destroy(prot->slab);
2517 prot->slab = NULL;
2518 }
2519
2e6599cb 2520 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2521 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2522 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2523 prot->rsk_prot->slab = NULL;
2524 }
2525
6d6ee43e 2526 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2527 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2528 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2529 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2530 }
1da177e4 2531}
1da177e4
LT		 2532EXPORT_SYMBOL(proto_unregister);
2533
2534#ifdef CONFIG_PROC_FS
1da177e4 2535static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2536 __acquires(proto_list_mutex)
1da177e4 2537{
36b77a52 2538 mutex_lock(&proto_list_mutex);
60f0438a 2539 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2540}
2541
2542static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2543{
60f0438a 2544 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2545}
2546
2547static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2548 __releases(proto_list_mutex)
1da177e4 2549{
36b77a52 2550 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2551}
2552
2553static char proto_method_implemented(const void *method)
2554{
2555 return method == NULL ? 'n' : 'y';
2556}
180d8cd9
GC		 2557static long sock_prot_memory_allocated(struct proto *proto)
2558{
2559 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2560}
2561
2562static char *sock_prot_memory_pressure(struct proto *proto)
2563{
2564 return proto->memory_pressure != NULL ?
2565 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2566}
1da177e4
LT		 2567
2568static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2569{
180d8cd9 2570
8d987e5c 2571 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2572 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2573 proto->name,
2574 proto->obj_size,
14e943db 2575 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2576 sock_prot_memory_allocated(proto),
2577 sock_prot_memory_pressure(proto),
1da177e4
LT		 2578 proto->max_header,
2579 proto->slab == NULL ? "no" : "yes",
2580 module_name(proto->owner),
2581 proto_method_implemented(proto->close),
2582 proto_method_implemented(proto->connect),
2583 proto_method_implemented(proto->disconnect),
2584 proto_method_implemented(proto->accept),
2585 proto_method_implemented(proto->ioctl),
2586 proto_method_implemented(proto->init),
2587 proto_method_implemented(proto->destroy),
2588 proto_method_implemented(proto->shutdown),
2589 proto_method_implemented(proto->setsockopt),
2590 proto_method_implemented(proto->getsockopt),
2591 proto_method_implemented(proto->sendmsg),
2592 proto_method_implemented(proto->recvmsg),
2593 proto_method_implemented(proto->sendpage),
2594 proto_method_implemented(proto->bind),
2595 proto_method_implemented(proto->backlog_rcv),
2596 proto_method_implemented(proto->hash),
2597 proto_method_implemented(proto->unhash),
2598 proto_method_implemented(proto->get_port),
2599 proto_method_implemented(proto->enter_memory_pressure));
2600}
2601
2602static int proto_seq_show(struct seq_file *seq, void *v)
2603{
60f0438a 2604 if (v == &proto_list)
1da177e4
LT		 2605 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2606 "protocol",
2607 "size",
2608 "sockets",
2609 "memory",
2610 "press",
2611 "maxhdr",
2612 "slab",
2613 "module",
2614 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2615 else
60f0438a 2616 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2617 return 0;
2618}
2619
f690808e 2620static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2621 .start = proto_seq_start,
2622 .next = proto_seq_next,
2623 .stop = proto_seq_stop,
2624 .show = proto_seq_show,
2625};
2626
2627static int proto_seq_open(struct inode *inode, struct file *file)
2628{
14e943db
ED		 2629 return seq_open_net(inode, file, &proto_seq_ops,
2630 sizeof(struct seq_net_private));
1da177e4
LT		 2631}
2632
9a32144e 2633static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2634 .owner = THIS_MODULE,
2635 .open = proto_seq_open,
2636 .read = seq_read,
2637 .llseek = seq_lseek,
14e943db
ED		 2638 .release = seq_release_net,
2639};
2640
2641static __net_init int proto_init_net(struct net *net)
2642{
2643 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2644 return -ENOMEM;
2645
2646 return 0;
2647}
2648
2649static __net_exit void proto_exit_net(struct net *net)
2650{
2651 proc_net_remove(net, "protocols");
2652}
2653
2654
2655static __net_initdata struct pernet_operations proto_net_ops = {
2656 .init = proto_init_net,
2657 .exit = proto_exit_net,
1da177e4
LT		 2658};
2659
2660static int __init proto_init(void)
2661{
14e943db 2662 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2663}
2664
2665subsys_initcall(proto_init);
2666
2667#endif /* PROC_FS */
Empty description
This page took 1.103393 seconds and 4 git commands to generate.