projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux...
[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>
e1aab161 114#include <linux/jump_label.h>
1da177e4
LT		 115
116#include <asm/uaccess.h>
117#include <asm/system.h>
118
119#include <linux/netdevice.h>
120#include <net/protocol.h>
121#include <linux/skbuff.h>
457c4cbc 122#include <net/net_namespace.h>
2e6599cb 123#include <net/request_sock.h>
1da177e4 124#include <net/sock.h>
20d49473 125#include <linux/net_tstamp.h>
1da177e4
LT		 126#include <net/xfrm.h>
127#include <linux/ipsec.h>
f8451725 128#include <net/cls_cgroup.h>
5bc1421e 129#include <net/netprio_cgroup.h>
1da177e4
LT		 130
131#include <linux/filter.h>
132
3847ce32
SM		 133#include <trace/events/sock.h>
134
1da177e4
LT		 135#ifdef CONFIG_INET
136#include <net/tcp.h>
137#endif
138
36b77a52 139static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 140static LIST_HEAD(proto_list);
141
142#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
143int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
144{
145 struct proto *proto;
146 int ret = 0;
147
36b77a52 148 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 149 list_for_each_entry(proto, &proto_list, node) {
150 if (proto->init_cgroup) {
151 ret = proto->init_cgroup(cgrp, ss);
152 if (ret)
153 goto out;
154 }
155 }
156
36b77a52 157 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 158 return ret;
159out:
160 list_for_each_entry_continue_reverse(proto, &proto_list, node)
161 if (proto->destroy_cgroup)
162 proto->destroy_cgroup(cgrp, ss);
36b77a52 163 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 164 return ret;
165}
166
167void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss)
168{
169 struct proto *proto;
170
36b77a52 171 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 172 list_for_each_entry_reverse(proto, &proto_list, node)
173 if (proto->destroy_cgroup)
174 proto->destroy_cgroup(cgrp, ss);
36b77a52 175 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 176}
177#endif
178
da21f24d
IM		 179/*
180 * Each address family might have different locking rules, so we have
181 * one slock key per address family:
182 */
a5b5bb9a
IM		 183static struct lock_class_key af_family_keys[AF_MAX];
184static struct lock_class_key af_family_slock_keys[AF_MAX];
185
e1aab161
GC		 186struct jump_label_key memcg_socket_limit_enabled;
187EXPORT_SYMBOL(memcg_socket_limit_enabled);
188
a5b5bb9a
IM		 189/*
190 * Make lock validator output more readable. (we pre-construct these
191 * strings build-time, so that runtime initialization of socket
192 * locks is fast):
193 */
36cbd3dc 194static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 195 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
196 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
197 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
198 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
199 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
200 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
201 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 202 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 203 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 204 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 205 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 206 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 207 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 208 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 209};
36cbd3dc 210static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 211 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
212 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
213 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
214 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
215 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
216 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
217 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 218 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 219 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 220 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 221 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 222 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 223 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 224 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 225};
36cbd3dc 226static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 227 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
228 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
229 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
230 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
231 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
232 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
233 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 234 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 235 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 236 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 237 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 238 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 239 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 240 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 241};
da21f24d
IM		 242
243/*
244 * sk_callback_lock locking rules are per-address-family,
245 * so split the lock classes by using a per-AF key:
246 */
247static struct lock_class_key af_callback_keys[AF_MAX];
248
1da177e4
LT		 249/* Take into consideration the size of the struct sk_buff overhead in the
250 * determination of these values, since that is non-constant across
251 * platforms. This makes socket queueing behavior and performance
252 * not depend upon such differences.
253 */
254#define _SK_MEM_PACKETS 256
87fb4b7b 255#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 256#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
257#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
258
259/* Run time adjustable parameters. */
ab32ea5d
BH		 260__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
261__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
262__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
263__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 264
25985edc 265/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 266int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 267EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 268
5bc1421e
NH		 269#if defined(CONFIG_CGROUPS)
270#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 271int net_cls_subsys_id = -1;
272EXPORT_SYMBOL_GPL(net_cls_subsys_id);
273#endif
5bc1421e
NH		 274#if !defined(CONFIG_NETPRIO_CGROUP)
275int net_prio_subsys_id = -1;
276EXPORT_SYMBOL_GPL(net_prio_subsys_id);
277#endif
278#endif
f8451725 279
1da177e4
LT		 280static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
281{
282 struct timeval tv;
283
284 if (optlen < sizeof(tv))
285 return -EINVAL;
286 if (copy_from_user(&tv, optval, sizeof(tv)))
287 return -EFAULT;
ba78073e
VA		 288 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
289 return -EDOM;
1da177e4 290
ba78073e 291 if (tv.tv_sec < 0) {
6f11df83
AM		 292 static int warned __read_mostly;
293
ba78073e 294 *timeo_p = 0;
50aab54f 295 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 296 warned++;
297 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
298 "tries to set negative timeout\n",
ba25f9dc 299 current->comm, task_pid_nr(current));
50aab54f 300 }
ba78073e
VA		 301 return 0;
302 }
1da177e4
LT		 303 *timeo_p = MAX_SCHEDULE_TIMEOUT;
304 if (tv.tv_sec == 0 && tv.tv_usec == 0)
305 return 0;
306 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
307 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
308 return 0;
309}
310
311static void sock_warn_obsolete_bsdism(const char *name)
312{
313 static int warned;
314 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 315 if (strcmp(warncomm, current->comm) && warned < 5) {
316 strcpy(warncomm, current->comm);
1da177e4
LT		 317 printk(KERN_WARNING "process `%s' is using obsolete "
318 "%s SO_BSDCOMPAT\n", warncomm, name);
319 warned++;
320 }
321}
322
08e29af3
ED		 323#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
324
325static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 326{
08e29af3
ED		 327 if (sk->sk_flags & flags) {
328 sk->sk_flags &= ~flags;
329 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 330 net_disable_timestamp();
1da177e4
LT		 331 }
332}
333
334
f0088a50
DV		 335int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
336{
766e9037 337 int err;
f0088a50 338 int skb_len;
3b885787
NH		 339 unsigned long flags;
340 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 341
0fd7bac6 342 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 343 atomic_inc(&sk->sk_drops);
3847ce32 344 trace_sock_rcvqueue_full(sk, skb);
766e9037 345 return -ENOMEM;
f0088a50
DV		 346 }
347
fda9ef5d 348 err = sk_filter(sk, skb);
f0088a50 349 if (err)
766e9037 350 return err;
f0088a50 351
3ab224be 352 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 353 atomic_inc(&sk->sk_drops);
354 return -ENOBUFS;
3ab224be
HA		 355 }
356
f0088a50
DV		 357 skb->dev = NULL;
358 skb_set_owner_r(skb, sk);
49ad9599 359
f0088a50
DV		 360 /* Cache the SKB length before we tack it onto the receive
361 * queue. Once it is added it no longer belongs to us and
362 * may be freed by other threads of control pulling packets
363 * from the queue.
364 */
365 skb_len = skb->len;
366
7fee226a
ED		 367 /* we escape from rcu protected region, make sure we dont leak
368 * a norefcounted dst
369 */
370 skb_dst_force(skb);
371
3b885787
NH		 372 spin_lock_irqsave(&list->lock, flags);
373 skb->dropcount = atomic_read(&sk->sk_drops);
374 __skb_queue_tail(list, skb);
375 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 376
377 if (!sock_flag(sk, SOCK_DEAD))
378 sk->sk_data_ready(sk, skb_len);
766e9037 379 return 0;
f0088a50
DV		 380}
381EXPORT_SYMBOL(sock_queue_rcv_skb);
382
58a5a7b9 383int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 384{
385 int rc = NET_RX_SUCCESS;
386
fda9ef5d 387 if (sk_filter(sk, skb))
f0088a50
DV		 388 goto discard_and_relse;
389
390 skb->dev = NULL;
391
c377411f
ED		 392 if (sk_rcvqueues_full(sk, skb)) {
393 atomic_inc(&sk->sk_drops);
394 goto discard_and_relse;
395 }
58a5a7b9
ACM		 396 if (nested)
397 bh_lock_sock_nested(sk);
398 else
399 bh_lock_sock(sk);
a5b5bb9a
IM		 400 if (!sock_owned_by_user(sk)) {
401 /*
402 * trylock + unlock semantics:
403 */
404 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
405
c57943a1 406 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 407
408 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 409 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 410 bh_unlock_sock(sk);
411 atomic_inc(&sk->sk_drops);
412 goto discard_and_relse;
413 }
414
f0088a50
DV		 415 bh_unlock_sock(sk);
416out:
417 sock_put(sk);
418 return rc;
419discard_and_relse:
420 kfree_skb(skb);
421 goto out;
422}
423EXPORT_SYMBOL(sk_receive_skb);
424
ea94ff3b
KK		 425void sk_reset_txq(struct sock *sk)
426{
427 sk_tx_queue_clear(sk);
428}
429EXPORT_SYMBOL(sk_reset_txq);
430
f0088a50
DV		 431struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
432{
b6c6712a 433 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 434
435 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 436 sk_tx_queue_clear(sk);
a9b3cd7f 437 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 438 dst_release(dst);
439 return NULL;
440 }
441
442 return dst;
443}
444EXPORT_SYMBOL(__sk_dst_check);
445
446struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
447{
448 struct dst_entry *dst = sk_dst_get(sk);
449
450 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
451 sk_dst_reset(sk);
452 dst_release(dst);
453 return NULL;
454 }
455
456 return dst;
457}
458EXPORT_SYMBOL(sk_dst_check);
459
4878809f
DM		 460static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
461{
462 int ret = -ENOPROTOOPT;
463#ifdef CONFIG_NETDEVICES
3b1e0a65 464 struct net *net = sock_net(sk);
4878809f
DM		 465 char devname[IFNAMSIZ];
466 int index;
467
468 /* Sorry... */
469 ret = -EPERM;
470 if (!capable(CAP_NET_RAW))
471 goto out;
472
473 ret = -EINVAL;
474 if (optlen < 0)
475 goto out;
476
477 /* Bind this socket to a particular device like "eth0",
478 * as specified in the passed interface name. If the
479 * name is "" or the option length is zero the socket
480 * is not bound.
481 */
482 if (optlen > IFNAMSIZ - 1)
483 optlen = IFNAMSIZ - 1;
484 memset(devname, 0, sizeof(devname));
485
486 ret = -EFAULT;
487 if (copy_from_user(devname, optval, optlen))
488 goto out;
489
000ba2e4
DM		 490 index = 0;
491 if (devname[0] != '\0') {
bf8e56bf 492 struct net_device *dev;
4878809f 493
bf8e56bf
ED		 494 rcu_read_lock();
495 dev = dev_get_by_name_rcu(net, devname);
496 if (dev)
497 index = dev->ifindex;
498 rcu_read_unlock();
4878809f
DM		 499 ret = -ENODEV;
500 if (!dev)
501 goto out;
4878809f
DM		 502 }
503
504 lock_sock(sk);
505 sk->sk_bound_dev_if = index;
506 sk_dst_reset(sk);
507 release_sock(sk);
508
509 ret = 0;
510
511out:
512#endif
513
514 return ret;
515}
516
c0ef877b
PE		 517static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
518{
519 if (valbool)
520 sock_set_flag(sk, bit);
521 else
522 sock_reset_flag(sk, bit);
523}
524
1da177e4
LT		 525/*
526 * This is meant for all protocols to use and covers goings on
527 * at the socket level. Everything here is generic.
528 */
529
530int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 531 char __user *optval, unsigned int optlen)
1da177e4 532{
2a91525c 533 struct sock *sk = sock->sk;
1da177e4
LT		 534 int val;
535 int valbool;
536 struct linger ling;
537 int ret = 0;
4ec93edb 538
1da177e4
LT		 539 /*
540 * Options without arguments
541 */
542
4878809f
DM		 543 if (optname == SO_BINDTODEVICE)
544 return sock_bindtodevice(sk, optval, optlen);
545
e71a4783
SH		 546 if (optlen < sizeof(int))
547 return -EINVAL;
4ec93edb 548
1da177e4
LT		 549 if (get_user(val, (int __user *)optval))
550 return -EFAULT;
4ec93edb 551
2a91525c 552 valbool = val ? 1 : 0;
1da177e4
LT		 553
554 lock_sock(sk);
555
2a91525c 556 switch (optname) {
e71a4783 557 case SO_DEBUG:
2a91525c 558 if (val && !capable(CAP_NET_ADMIN))
e71a4783 559 ret = -EACCES;
2a91525c 560 else
c0ef877b 561 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 562 break;
563 case SO_REUSEADDR:
564 sk->sk_reuse = valbool;
565 break;
566 case SO_TYPE:
49c794e9 567 case SO_PROTOCOL:
0d6038ee 568 case SO_DOMAIN:
e71a4783
SH		 569 case SO_ERROR:
570 ret = -ENOPROTOOPT;
571 break;
572 case SO_DONTROUTE:
c0ef877b 573 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 574 break;
575 case SO_BROADCAST:
576 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
577 break;
578 case SO_SNDBUF:
579 /* Don't error on this BSD doesn't and if you think
580 about it this is right. Otherwise apps have to
581 play 'guess the biggest size' games. RCVBUF/SNDBUF
582 are treated in BSD as hints */
583
584 if (val > sysctl_wmem_max)
585 val = sysctl_wmem_max;
b0573dea 586set_sndbuf:
e71a4783
SH		 587 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
588 if ((val * 2) < SOCK_MIN_SNDBUF)
589 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
590 else
591 sk->sk_sndbuf = val * 2;
1da177e4 592
e71a4783
SH		 593 /*
594 * Wake up sending tasks if we
595 * upped the value.
596 */
597 sk->sk_write_space(sk);
598 break;
1da177e4 599
e71a4783
SH		 600 case SO_SNDBUFFORCE:
601 if (!capable(CAP_NET_ADMIN)) {
602 ret = -EPERM;
603 break;
604 }
605 goto set_sndbuf;
b0573dea 606
e71a4783
SH		 607 case SO_RCVBUF:
608 /* Don't error on this BSD doesn't and if you think
609 about it this is right. Otherwise apps have to
610 play 'guess the biggest size' games. RCVBUF/SNDBUF
611 are treated in BSD as hints */
4ec93edb 612
e71a4783
SH		 613 if (val > sysctl_rmem_max)
614 val = sysctl_rmem_max;
b0573dea 615set_rcvbuf:
e71a4783
SH		 616 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
617 /*
618 * We double it on the way in to account for
619 * "struct sk_buff" etc. overhead. Applications
620 * assume that the SO_RCVBUF setting they make will
621 * allow that much actual data to be received on that
622 * socket.
623 *
624 * Applications are unaware that "struct sk_buff" and
625 * other overheads allocate from the receive buffer
626 * during socket buffer allocation.
627 *
628 * And after considering the possible alternatives,
629 * returning the value we actually used in getsockopt
630 * is the most desirable behavior.
631 */
632 if ((val * 2) < SOCK_MIN_RCVBUF)
633 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
634 else
635 sk->sk_rcvbuf = val * 2;
636 break;
637
638 case SO_RCVBUFFORCE:
639 if (!capable(CAP_NET_ADMIN)) {
640 ret = -EPERM;
1da177e4 641 break;
e71a4783
SH		 642 }
643 goto set_rcvbuf;
1da177e4 644
e71a4783 645 case SO_KEEPALIVE:
1da177e4 646#ifdef CONFIG_INET
e71a4783
SH		 647 if (sk->sk_protocol == IPPROTO_TCP)
648 tcp_set_keepalive(sk, valbool);
1da177e4 649#endif
e71a4783
SH		 650 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
651 break;
652
653 case SO_OOBINLINE:
654 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
655 break;
656
657 case SO_NO_CHECK:
658 sk->sk_no_check = valbool;
659 break;
660
661 case SO_PRIORITY:
662 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
663 sk->sk_priority = val;
664 else
665 ret = -EPERM;
666 break;
667
668 case SO_LINGER:
669 if (optlen < sizeof(ling)) {
670 ret = -EINVAL; /* 1003.1g */
1da177e4 671 break;
e71a4783 672 }
2a91525c 673 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 674 ret = -EFAULT;
1da177e4 675 break;
e71a4783
SH		 676 }
677 if (!ling.l_onoff)
678 sock_reset_flag(sk, SOCK_LINGER);
679 else {
1da177e4 680#if (BITS_PER_LONG == 32)
e71a4783
SH		 681 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
682 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 683 else
e71a4783
SH		 684#endif
685 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
686 sock_set_flag(sk, SOCK_LINGER);
687 }
688 break;
689
690 case SO_BSDCOMPAT:
691 sock_warn_obsolete_bsdism("setsockopt");
692 break;
693
694 case SO_PASSCRED:
695 if (valbool)
696 set_bit(SOCK_PASSCRED, &sock->flags);
697 else
698 clear_bit(SOCK_PASSCRED, &sock->flags);
699 break;
700
701 case SO_TIMESTAMP:
92f37fd2 702 case SO_TIMESTAMPNS:
e71a4783 703 if (valbool) {
92f37fd2
ED		 704 if (optname == SO_TIMESTAMP)
705 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
706 else
707 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 708 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 709 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 710 } else {
e71a4783 711 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 712 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
713 }
e71a4783
SH		 714 break;
715
20d49473
PO		 716 case SO_TIMESTAMPING:
717 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 718 ret = -EINVAL;
20d49473
PO		 719 break;
720 }
721 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
722 val & SOF_TIMESTAMPING_TX_HARDWARE);
723 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
724 val & SOF_TIMESTAMPING_TX_SOFTWARE);
725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
726 val & SOF_TIMESTAMPING_RX_HARDWARE);
727 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
728 sock_enable_timestamp(sk,
729 SOCK_TIMESTAMPING_RX_SOFTWARE);
730 else
731 sock_disable_timestamp(sk,
08e29af3 732 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 733 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
734 val & SOF_TIMESTAMPING_SOFTWARE);
735 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
736 val & SOF_TIMESTAMPING_SYS_HARDWARE);
737 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
738 val & SOF_TIMESTAMPING_RAW_HARDWARE);
739 break;
740
e71a4783
SH		 741 case SO_RCVLOWAT:
742 if (val < 0)
743 val = INT_MAX;
744 sk->sk_rcvlowat = val ? : 1;
745 break;
746
747 case SO_RCVTIMEO:
748 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
749 break;
750
751 case SO_SNDTIMEO:
752 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
753 break;
1da177e4 754
e71a4783
SH		 755 case SO_ATTACH_FILTER:
756 ret = -EINVAL;
757 if (optlen == sizeof(struct sock_fprog)) {
758 struct sock_fprog fprog;
1da177e4 759
e71a4783
SH		 760 ret = -EFAULT;
761 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 762 break;
e71a4783
SH		 763
764 ret = sk_attach_filter(&fprog, sk);
765 }
766 break;
767
768 case SO_DETACH_FILTER:
55b33325 769 ret = sk_detach_filter(sk);
e71a4783 770 break;
1da177e4 771
e71a4783
SH		 772 case SO_PASSSEC:
773 if (valbool)
774 set_bit(SOCK_PASSSEC, &sock->flags);
775 else
776 clear_bit(SOCK_PASSSEC, &sock->flags);
777 break;
4a19ec58
LAT		 778 case SO_MARK:
779 if (!capable(CAP_NET_ADMIN))
780 ret = -EPERM;
2a91525c 781 else
4a19ec58 782 sk->sk_mark = val;
4a19ec58 783 break;
877ce7c1 784
1da177e4
LT		 785 /* We implement the SO_SNDLOWAT etc to
786 not be settable (1003.1g 5.3) */
3b885787 787 case SO_RXQ_OVFL:
8083f0fc 788 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 789 break;
6e3e939f
JB		 790
791 case SO_WIFI_STATUS:
792 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
793 break;
794
e71a4783
SH		 795 default:
796 ret = -ENOPROTOOPT;
797 break;
4ec93edb 798 }
1da177e4
LT		 799 release_sock(sk);
800 return ret;
801}
2a91525c 802EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 803
804
3f551f94
EB		 805void cred_to_ucred(struct pid *pid, const struct cred *cred,
806 struct ucred *ucred)
807{
808 ucred->pid = pid_vnr(pid);
809 ucred->uid = ucred->gid = -1;
810 if (cred) {
811 struct user_namespace *current_ns = current_user_ns();
812
813 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
814 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
815 }
816}
3924773a 817EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 818
1da177e4
LT		 819int sock_getsockopt(struct socket *sock, int level, int optname,
820 char __user *optval, int __user *optlen)
821{
822 struct sock *sk = sock->sk;
4ec93edb 823
e71a4783 824 union {
4ec93edb
YH		 825 int val;
826 struct linger ling;
1da177e4
LT		 827 struct timeval tm;
828 } v;
4ec93edb 829
4d0392be 830 int lv = sizeof(int);
1da177e4 831 int len;
4ec93edb 832
e71a4783 833 if (get_user(len, optlen))
4ec93edb 834 return -EFAULT;
e71a4783 835 if (len < 0)
1da177e4 836 return -EINVAL;
4ec93edb 837
50fee1de 838 memset(&v, 0, sizeof(v));
df0bca04 839
2a91525c 840 switch (optname) {
e71a4783
SH		 841 case SO_DEBUG:
842 v.val = sock_flag(sk, SOCK_DBG);
843 break;
844
845 case SO_DONTROUTE:
846 v.val = sock_flag(sk, SOCK_LOCALROUTE);
847 break;
848
849 case SO_BROADCAST:
850 v.val = !!sock_flag(sk, SOCK_BROADCAST);
851 break;
852
853 case SO_SNDBUF:
854 v.val = sk->sk_sndbuf;
855 break;
856
857 case SO_RCVBUF:
858 v.val = sk->sk_rcvbuf;
859 break;
860
861 case SO_REUSEADDR:
862 v.val = sk->sk_reuse;
863 break;
864
865 case SO_KEEPALIVE:
866 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
867 break;
868
869 case SO_TYPE:
870 v.val = sk->sk_type;
871 break;
872
49c794e9
JE		 873 case SO_PROTOCOL:
874 v.val = sk->sk_protocol;
875 break;
876
0d6038ee
JE		 877 case SO_DOMAIN:
878 v.val = sk->sk_family;
879 break;
880
e71a4783
SH		 881 case SO_ERROR:
882 v.val = -sock_error(sk);
2a91525c 883 if (v.val == 0)
e71a4783
SH		 884 v.val = xchg(&sk->sk_err_soft, 0);
885 break;
886
887 case SO_OOBINLINE:
888 v.val = !!sock_flag(sk, SOCK_URGINLINE);
889 break;
890
891 case SO_NO_CHECK:
892 v.val = sk->sk_no_check;
893 break;
894
895 case SO_PRIORITY:
896 v.val = sk->sk_priority;
897 break;
898
899 case SO_LINGER:
900 lv = sizeof(v.ling);
901 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
902 v.ling.l_linger = sk->sk_lingertime / HZ;
903 break;
904
905 case SO_BSDCOMPAT:
906 sock_warn_obsolete_bsdism("getsockopt");
907 break;
908
909 case SO_TIMESTAMP:
92f37fd2
ED		 910 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
911 !sock_flag(sk, SOCK_RCVTSTAMPNS);
912 break;
913
914 case SO_TIMESTAMPNS:
915 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 916 break;
917
20d49473
PO		 918 case SO_TIMESTAMPING:
919 v.val = 0;
920 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
921 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
922 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
923 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
924 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
925 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
926 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
927 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
928 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
929 v.val |= SOF_TIMESTAMPING_SOFTWARE;
930 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
931 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
932 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
933 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
934 break;
935
e71a4783 936 case SO_RCVTIMEO:
2a91525c 937 lv = sizeof(struct timeval);
e71a4783
SH		 938 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
939 v.tm.tv_sec = 0;
940 v.tm.tv_usec = 0;
941 } else {
942 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
943 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
944 }
945 break;
946
947 case SO_SNDTIMEO:
2a91525c 948 lv = sizeof(struct timeval);
e71a4783
SH		 949 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
950 v.tm.tv_sec = 0;
951 v.tm.tv_usec = 0;
952 } else {
953 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
954 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
955 }
956 break;
1da177e4 957
e71a4783
SH		 958 case SO_RCVLOWAT:
959 v.val = sk->sk_rcvlowat;
960 break;
1da177e4 961
e71a4783 962 case SO_SNDLOWAT:
2a91525c 963 v.val = 1;
e71a4783 964 break;
1da177e4 965
e71a4783
SH		 966 case SO_PASSCRED:
967 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
968 break;
1da177e4 969
e71a4783 970 case SO_PEERCRED:
109f6e39
EB		 971 {
972 struct ucred peercred;
973 if (len > sizeof(peercred))
974 len = sizeof(peercred);
975 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
976 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 977 return -EFAULT;
978 goto lenout;
109f6e39 979 }
1da177e4 980
e71a4783
SH		 981 case SO_PEERNAME:
982 {
983 char address[128];
984
985 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
986 return -ENOTCONN;
987 if (lv < len)
988 return -EINVAL;
989 if (copy_to_user(optval, address, len))
990 return -EFAULT;
991 goto lenout;
992 }
1da177e4 993
e71a4783
SH		 994 /* Dubious BSD thing... Probably nobody even uses it, but
995 * the UNIX standard wants it for whatever reason... -DaveM
996 */
997 case SO_ACCEPTCONN:
998 v.val = sk->sk_state == TCP_LISTEN;
999 break;
1da177e4 1000
e71a4783
SH		 1001 case SO_PASSSEC:
1002 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
1003 break;
877ce7c1 1004
e71a4783
SH		 1005 case SO_PEERSEC:
1006 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1007
4a19ec58
LAT		 1008 case SO_MARK:
1009 v.val = sk->sk_mark;
1010 break;
1011
3b885787
NH		 1012 case SO_RXQ_OVFL:
1013 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
1014 break;
1015
6e3e939f
JB		 1016 case SO_WIFI_STATUS:
1017 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
1018 break;
1019
e71a4783
SH		 1020 default:
1021 return -ENOPROTOOPT;
1da177e4 1022 }
e71a4783 1023
1da177e4
LT		 1024 if (len > lv)
1025 len = lv;
1026 if (copy_to_user(optval, &v, len))
1027 return -EFAULT;
1028lenout:
4ec93edb
YH		 1029 if (put_user(len, optlen))
1030 return -EFAULT;
1031 return 0;
1da177e4
LT		 1032}
1033
a5b5bb9a
IM		 1034/*
1035 * Initialize an sk_lock.
1036 *
1037 * (We also register the sk_lock with the lock validator.)
1038 */
b6f99a21 1039static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1040{
ed07536e
PZ		 1041 sock_lock_init_class_and_name(sk,
1042 af_family_slock_key_strings[sk->sk_family],
1043 af_family_slock_keys + sk->sk_family,
1044 af_family_key_strings[sk->sk_family],
1045 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1046}
1047
4dc6dc71
ED		 1048/*
1049 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1050 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1051 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1052 */
f1a6c4da
PE		 1053static void sock_copy(struct sock *nsk, const struct sock *osk)
1054{
1055#ifdef CONFIG_SECURITY_NETWORK
1056 void *sptr = nsk->sk_security;
1057#endif
68835aba
ED		 1058 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1059
1060 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1061 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1062
f1a6c4da
PE		 1063#ifdef CONFIG_SECURITY_NETWORK
1064 nsk->sk_security = sptr;
1065 security_sk_clone(osk, nsk);
1066#endif
1067}
1068
fcbdf09d
OP		 1069/*
1070 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1071 * un-modified. Special care is taken when initializing object to zero.
1072 */
1073static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1074{
1075 if (offsetof(struct sock, sk_node.next) != 0)
1076 memset(sk, 0, offsetof(struct sock, sk_node.next));
1077 memset(&sk->sk_node.pprev, 0,
1078 size - offsetof(struct sock, sk_node.pprev));
1079}
1080
1081void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1082{
1083 unsigned long nulls1, nulls2;
1084
1085 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1086 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1087 if (nulls1 > nulls2)
1088 swap(nulls1, nulls2);
1089
1090 if (nulls1 != 0)
1091 memset((char *)sk, 0, nulls1);
1092 memset((char *)sk + nulls1 + sizeof(void *), 0,
1093 nulls2 - nulls1 - sizeof(void *));
1094 memset((char *)sk + nulls2 + sizeof(void *), 0,
1095 size - nulls2 - sizeof(void *));
1096}
1097EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1098
2e4afe7b
PE		 1099static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1100 int family)
c308c1b2
PE		 1101{
1102 struct sock *sk;
1103 struct kmem_cache *slab;
1104
1105 slab = prot->slab;
e912b114
ED		 1106 if (slab != NULL) {
1107 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1108 if (!sk)
1109 return sk;
1110 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1111 if (prot->clear_sk)
1112 prot->clear_sk(sk, prot->obj_size);
1113 else
1114 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1115 }
fcbdf09d 1116 } else
c308c1b2
PE		 1117 sk = kmalloc(prot->obj_size, priority);
1118
2e4afe7b 1119 if (sk != NULL) {
a98b65a3
VN		 1120 kmemcheck_annotate_bitfield(sk, flags);
1121
2e4afe7b
PE		 1122 if (security_sk_alloc(sk, family, priority))
1123 goto out_free;
1124
1125 if (!try_module_get(prot->owner))
1126 goto out_free_sec;
e022f0b4 1127 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1128 }
1129
c308c1b2 1130 return sk;
2e4afe7b
PE		 1131
1132out_free_sec:
1133 security_sk_free(sk);
1134out_free:
1135 if (slab != NULL)
1136 kmem_cache_free(slab, sk);
1137 else
1138 kfree(sk);
1139 return NULL;
c308c1b2
PE		 1140}
1141
1142static void sk_prot_free(struct proto *prot, struct sock *sk)
1143{
1144 struct kmem_cache *slab;
2e4afe7b 1145 struct module *owner;
c308c1b2 1146
2e4afe7b 1147 owner = prot->owner;
c308c1b2 1148 slab = prot->slab;
2e4afe7b
PE		 1149
1150 security_sk_free(sk);
c308c1b2
PE		 1151 if (slab != NULL)
1152 kmem_cache_free(slab, sk);
1153 else
1154 kfree(sk);
2e4afe7b 1155 module_put(owner);
c308c1b2
PE		 1156}
1157
f8451725
HX		 1158#ifdef CONFIG_CGROUPS
1159void sock_update_classid(struct sock *sk)
1160{
1144182a 1161 u32 classid;
f8451725 1162
1144182a
PM		 1163 rcu_read_lock(); /* doing current task, which cannot vanish. */
1164 classid = task_cls_classid(current);
1165 rcu_read_unlock();
f8451725
HX		 1166 if (classid && classid != sk->sk_classid)
1167 sk->sk_classid = classid;
1168}
82862742 1169EXPORT_SYMBOL(sock_update_classid);
5bc1421e
NH		 1170
1171void sock_update_netprioidx(struct sock *sk)
1172{
1173 struct cgroup_netprio_state *state;
1174 if (in_interrupt())
1175 return;
1176 rcu_read_lock();
1177 state = task_netprio_state(current);
1178 sk->sk_cgrp_prioidx = state ? state->prioidx : 0;
1179 rcu_read_unlock();
1180}
1181EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1182#endif
1183
1da177e4
LT		 1184/**
1185 * sk_alloc - All socket objects are allocated here
c4ea43c5 1186 * @net: the applicable net namespace
4dc3b16b
PP		 1187 * @family: protocol family
1188 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1189 * @prot: struct proto associated with this new sock instance
1da177e4 1190 */
1b8d7ae4 1191struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1192 struct proto *prot)
1da177e4 1193{
c308c1b2 1194 struct sock *sk;
1da177e4 1195
154adbc8 1196 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1197 if (sk) {
154adbc8
PE		 1198 sk->sk_family = family;
1199 /*
1200 * See comment in struct sock definition to understand
1201 * why we need sk_prot_creator -acme
1202 */
1203 sk->sk_prot = sk->sk_prot_creator = prot;
1204 sock_lock_init(sk);
3b1e0a65 1205 sock_net_set(sk, get_net(net));
d66ee058 1206 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1207
1208 sock_update_classid(sk);
5bc1421e 1209 sock_update_netprioidx(sk);
1da177e4 1210 }
a79af59e 1211
2e4afe7b 1212 return sk;
1da177e4 1213}
2a91525c 1214EXPORT_SYMBOL(sk_alloc);
1da177e4 1215
2b85a34e 1216static void __sk_free(struct sock *sk)
1da177e4
LT		 1217{
1218 struct sk_filter *filter;
1da177e4
LT		 1219
1220 if (sk->sk_destruct)
1221 sk->sk_destruct(sk);
1222
a898def2
PM		 1223 filter = rcu_dereference_check(sk->sk_filter,
1224 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1225 if (filter) {
309dd5fc 1226 sk_filter_uncharge(sk, filter);
a9b3cd7f 1227 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1228 }
1229
08e29af3 1230 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1231
1232 if (atomic_read(&sk->sk_omem_alloc))
1233 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1234 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1235
109f6e39
EB		 1236 if (sk->sk_peer_cred)
1237 put_cred(sk->sk_peer_cred);
1238 put_pid(sk->sk_peer_pid);
3b1e0a65 1239 put_net(sock_net(sk));
c308c1b2 1240 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1241}
2b85a34e
ED		 1242
1243void sk_free(struct sock *sk)
1244{
1245 /*
25985edc 1246 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1247 * some packets are still in some tx queue.
1248 * If not null, sock_wfree() will call __sk_free(sk) later
1249 */
1250 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1251 __sk_free(sk);
1252}
2a91525c 1253EXPORT_SYMBOL(sk_free);
1da177e4 1254
edf02087 1255/*
25985edc
LDM		 1256 * Last sock_put should drop reference to sk->sk_net. It has already
1257 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1258 * is not an option.
25985edc 1259 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1260 * destroy it in the context of init_net.
1261 */
1262void sk_release_kernel(struct sock *sk)
1263{
1264 if (sk == NULL || sk->sk_socket == NULL)
1265 return;
1266
1267 sock_hold(sk);
1268 sock_release(sk->sk_socket);
65a18ec5 1269 release_net(sock_net(sk));
3b1e0a65 1270 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1271 sock_put(sk);
1272}
45af1754 1273EXPORT_SYMBOL(sk_release_kernel);
edf02087 1274
e56c57d0
ED		 1275/**
1276 * sk_clone_lock - clone a socket, and lock its clone
1277 * @sk: the socket to clone
1278 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1279 *
1280 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1281 */
1282struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1283{
8fd1d178 1284 struct sock *newsk;
87d11ceb 1285
8fd1d178 1286 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1287 if (newsk != NULL) {
1288 struct sk_filter *filter;
1289
892c141e 1290 sock_copy(newsk, sk);
87d11ceb
ACM		 1291
1292 /* SANITY */
3b1e0a65 1293 get_net(sock_net(newsk));
87d11ceb
ACM		 1294 sk_node_init(&newsk->sk_node);
1295 sock_lock_init(newsk);
1296 bh_lock_sock(newsk);
fa438ccf 1297 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1298 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1299
1300 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1301 /*
1302 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1303 */
1304 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1305 atomic_set(&newsk->sk_omem_alloc, 0);
1306 skb_queue_head_init(&newsk->sk_receive_queue);
1307 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1308#ifdef CONFIG_NET_DMA
1309 skb_queue_head_init(&newsk->sk_async_wait_queue);
1310#endif
87d11ceb 1311
b6c6712a 1312 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1313 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1314 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1315 af_callback_keys + newsk->sk_family,
1316 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1317
1318 newsk->sk_dst_cache = NULL;
1319 newsk->sk_wmem_queued = 0;
1320 newsk->sk_forward_alloc = 0;
1321 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1322 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1323
1324 sock_reset_flag(newsk, SOCK_DONE);
1325 skb_queue_head_init(&newsk->sk_error_queue);
1326
0d7da9dd 1327 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1328 if (filter != NULL)
1329 sk_filter_charge(newsk, filter);
1330
1331 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1332 /* It is still raw copy of parent, so invalidate
1333 * destructor and make plain sk_free() */
1334 newsk->sk_destruct = NULL;
b0691c8e 1335 bh_unlock_sock(newsk);
87d11ceb
ACM		 1336 sk_free(newsk);
1337 newsk = NULL;
1338 goto out;
1339 }
1340
1341 newsk->sk_err = 0;
1342 newsk->sk_priority = 0;
4dc6dc71
ED		 1343 /*
1344 * Before updating sk_refcnt, we must commit prior changes to memory
1345 * (Documentation/RCU/rculist_nulls.txt for details)
1346 */
1347 smp_wmb();
87d11ceb
ACM		 1348 atomic_set(&newsk->sk_refcnt, 2);
1349
1350 /*
1351 * Increment the counter in the same struct proto as the master
1352 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1353 * is the same as sk->sk_prot->socks, as this field was copied
1354 * with memcpy).
1355 *
1356 * This _changes_ the previous behaviour, where
1357 * tcp_create_openreq_child always was incrementing the
1358 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1359 * to be taken into account in all callers. -acme
1360 */
1361 sk_refcnt_debug_inc(newsk);
972692e0 1362 sk_set_socket(newsk, NULL);
43815482 1363 newsk->sk_wq = NULL;
87d11ceb
ACM		 1364
1365 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1366 sk_sockets_allocated_inc(newsk);
704da560 1367
08e29af3 1368 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1369 net_enable_timestamp();
87d11ceb
ACM		 1370 }
1371out:
1372 return newsk;
1373}
e56c57d0 1374EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1375
9958089a
AK		 1376void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1377{
1378 __sk_dst_set(sk, dst);
1379 sk->sk_route_caps = dst->dev->features;
1380 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1381 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1382 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1383 if (sk_can_gso(sk)) {
82cc1a7a 1384 if (dst->header_len) {
9958089a 1385 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1386 } else {
9958089a 1387 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1388 sk->sk_gso_max_size = dst->dev->gso_max_size;
1389 }
9958089a
AK		 1390 }
1391}
1392EXPORT_SYMBOL_GPL(sk_setup_caps);
1393
1da177e4
LT		 1394void __init sk_init(void)
1395{
4481374c 1396 if (totalram_pages <= 4096) {
1da177e4
LT		 1397 sysctl_wmem_max = 32767;
1398 sysctl_rmem_max = 32767;
1399 sysctl_wmem_default = 32767;
1400 sysctl_rmem_default = 32767;
4481374c 1401 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1402 sysctl_wmem_max = 131071;
1403 sysctl_rmem_max = 131071;
1404 }
1405}
1406
1407/*
1408 * Simple resource managers for sockets.
1409 */
1410
1411
4ec93edb
YH		 1412/*
1413 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1414 */
1415void sock_wfree(struct sk_buff *skb)
1416{
1417 struct sock *sk = skb->sk;
d99927f4 1418 unsigned int len = skb->truesize;
1da177e4 1419
d99927f4
ED		 1420 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1421 /*
1422 * Keep a reference on sk_wmem_alloc, this will be released
1423 * after sk_write_space() call
1424 */
1425 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1426 sk->sk_write_space(sk);
d99927f4
ED		 1427 len = 1;
1428 }
2b85a34e 1429 /*
d99927f4
ED		 1430 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1431 * could not do because of in-flight packets
2b85a34e 1432 */
d99927f4 1433 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1434 __sk_free(sk);
1da177e4 1435}
2a91525c 1436EXPORT_SYMBOL(sock_wfree);
1da177e4 1437
4ec93edb
YH		 1438/*
1439 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1440 */
1441void sock_rfree(struct sk_buff *skb)
1442{
1443 struct sock *sk = skb->sk;
d361fd59 1444 unsigned int len = skb->truesize;
1da177e4 1445
d361fd59
ED		 1446 atomic_sub(len, &sk->sk_rmem_alloc);
1447 sk_mem_uncharge(sk, len);
1da177e4 1448}
2a91525c 1449EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1450
1451
1452int sock_i_uid(struct sock *sk)
1453{
1454 int uid;
1455
f064af1e 1456 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1457 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1458 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1459 return uid;
1460}
2a91525c 1461EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1462
1463unsigned long sock_i_ino(struct sock *sk)
1464{
1465 unsigned long ino;
1466
f064af1e 1467 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1468 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1469 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1470 return ino;
1471}
2a91525c 1472EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1473
1474/*
1475 * Allocate a skb from the socket's send buffer.
1476 */
86a76caf 1477struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1478 gfp_t priority)
1da177e4
LT		 1479{
1480 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1481 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1482 if (skb) {
1483 skb_set_owner_w(skb, sk);
1484 return skb;
1485 }
1486 }
1487 return NULL;
1488}
2a91525c 1489EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1490
1491/*
1492 * Allocate a skb from the socket's receive buffer.
4ec93edb 1493 */
86a76caf 1494struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1495 gfp_t priority)
1da177e4
LT		 1496{
1497 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1498 struct sk_buff *skb = alloc_skb(size, priority);
1499 if (skb) {
1500 skb_set_owner_r(skb, sk);
1501 return skb;
1502 }
1503 }
1504 return NULL;
1505}
1506
4ec93edb 1507/*
1da177e4 1508 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1509 */
dd0fc66f 1510void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1511{
1512 if ((unsigned)size <= sysctl_optmem_max &&
1513 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1514 void *mem;
1515 /* First do the add, to avoid the race if kmalloc
4ec93edb 1516 * might sleep.
1da177e4
LT		 1517 */
1518 atomic_add(size, &sk->sk_omem_alloc);
1519 mem = kmalloc(size, priority);
1520 if (mem)
1521 return mem;
1522 atomic_sub(size, &sk->sk_omem_alloc);
1523 }
1524 return NULL;
1525}
2a91525c 1526EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1527
1528/*
1529 * Free an option memory block.
1530 */
1531void sock_kfree_s(struct sock *sk, void *mem, int size)
1532{
1533 kfree(mem);
1534 atomic_sub(size, &sk->sk_omem_alloc);
1535}
2a91525c 1536EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1537
1538/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1539 I think, these locks should be removed for datagram sockets.
1540 */
2a91525c 1541static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1542{
1543 DEFINE_WAIT(wait);
1544
1545 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1546 for (;;) {
1547 if (!timeo)
1548 break;
1549 if (signal_pending(current))
1550 break;
1551 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1552 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1553 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1554 break;
1555 if (sk->sk_shutdown & SEND_SHUTDOWN)
1556 break;
1557 if (sk->sk_err)
1558 break;
1559 timeo = schedule_timeout(timeo);
1560 }
aa395145 1561 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1562 return timeo;
1563}
1564
1565
1566/*
1567 * Generic send/receive buffer handlers
1568 */
1569
4cc7f68d
HX		 1570struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1571 unsigned long data_len, int noblock,
1572 int *errcode)
1da177e4
LT		 1573{
1574 struct sk_buff *skb;
7d877f3b 1575 gfp_t gfp_mask;
1da177e4
LT		 1576 long timeo;
1577 int err;
1578
1579 gfp_mask = sk->sk_allocation;
1580 if (gfp_mask & __GFP_WAIT)
1581 gfp_mask |= __GFP_REPEAT;
1582
1583 timeo = sock_sndtimeo(sk, noblock);
1584 while (1) {
1585 err = sock_error(sk);
1586 if (err != 0)
1587 goto failure;
1588
1589 err = -EPIPE;
1590 if (sk->sk_shutdown & SEND_SHUTDOWN)
1591 goto failure;
1592
1593 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1594 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1595 if (skb) {
1596 int npages;
1597 int i;
1598
1599 /* No pages, we're done... */
1600 if (!data_len)
1601 break;
1602
1603 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1604 skb->truesize += data_len;
1605 skb_shinfo(skb)->nr_frags = npages;
1606 for (i = 0; i < npages; i++) {
1607 struct page *page;
1da177e4
LT		 1608
1609 page = alloc_pages(sk->sk_allocation, 0);
1610 if (!page) {
1611 err = -ENOBUFS;
1612 skb_shinfo(skb)->nr_frags = i;
1613 kfree_skb(skb);
1614 goto failure;
1615 }
1616
ea2ab693
IC		 1617 __skb_fill_page_desc(skb, i,
1618 page, 0,
1619 (data_len >= PAGE_SIZE ?
1620 PAGE_SIZE :
1621 data_len));
1da177e4
LT		 1622 data_len -= PAGE_SIZE;
1623 }
1624
1625 /* Full success... */
1626 break;
1627 }
1628 err = -ENOBUFS;
1629 goto failure;
1630 }
1631 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1632 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1633 err = -EAGAIN;
1634 if (!timeo)
1635 goto failure;
1636 if (signal_pending(current))
1637 goto interrupted;
1638 timeo = sock_wait_for_wmem(sk, timeo);
1639 }
1640
1641 skb_set_owner_w(skb, sk);
1642 return skb;
1643
1644interrupted:
1645 err = sock_intr_errno(timeo);
1646failure:
1647 *errcode = err;
1648 return NULL;
1649}
4cc7f68d 1650EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1651
4ec93edb 1652struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1653 int noblock, int *errcode)
1654{
1655 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1656}
2a91525c 1657EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1658
1659static void __lock_sock(struct sock *sk)
f39234d6
NK		 1660 __releases(&sk->sk_lock.slock)
1661 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1662{
1663 DEFINE_WAIT(wait);
1664
e71a4783 1665 for (;;) {
1da177e4
LT		 1666 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1667 TASK_UNINTERRUPTIBLE);
1668 spin_unlock_bh(&sk->sk_lock.slock);
1669 schedule();
1670 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1671 if (!sock_owned_by_user(sk))
1da177e4
LT		 1672 break;
1673 }
1674 finish_wait(&sk->sk_lock.wq, &wait);
1675}
1676
1677static void __release_sock(struct sock *sk)
f39234d6
NK		 1678 __releases(&sk->sk_lock.slock)
1679 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1680{
1681 struct sk_buff *skb = sk->sk_backlog.head;
1682
1683 do {
1684 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1685 bh_unlock_sock(sk);
1686
1687 do {
1688 struct sk_buff *next = skb->next;
1689
7fee226a 1690 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1691 skb->next = NULL;
c57943a1 1692 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1693
1694 /*
1695 * We are in process context here with softirqs
1696 * disabled, use cond_resched_softirq() to preempt.
1697 * This is safe to do because we've taken the backlog
1698 * queue private:
1699 */
1700 cond_resched_softirq();
1701
1702 skb = next;
1703 } while (skb != NULL);
1704
1705 bh_lock_sock(sk);
e71a4783 1706 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1707
1708 /*
1709 * Doing the zeroing here guarantee we can not loop forever
1710 * while a wild producer attempts to flood us.
1711 */
1712 sk->sk_backlog.len = 0;
1da177e4
LT		 1713}
1714
1715/**
1716 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1717 * @sk: sock to wait on
1718 * @timeo: for how long
1da177e4
LT		 1719 *
1720 * Now socket state including sk->sk_err is changed only under lock,
1721 * hence we may omit checks after joining wait queue.
1722 * We check receive queue before schedule() only as optimization;
1723 * it is very likely that release_sock() added new data.
1724 */
1725int sk_wait_data(struct sock *sk, long *timeo)
1726{
1727 int rc;
1728 DEFINE_WAIT(wait);
1729
aa395145 1730 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1731 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1732 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1733 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1734 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1735 return rc;
1736}
1da177e4
LT		 1737EXPORT_SYMBOL(sk_wait_data);
1738
3ab224be
HA		 1739/**
1740 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1741 * @sk: socket
1742 * @size: memory size to allocate
1743 * @kind: allocation type
1744 *
1745 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1746 * rmem allocation. This function assumes that protocols which have
1747 * memory_pressure use sk_wmem_queued as write buffer accounting.
1748 */
1749int __sk_mem_schedule(struct sock *sk, int size, int kind)
1750{
1751 struct proto *prot = sk->sk_prot;
1752 int amt = sk_mem_pages(size);
8d987e5c 1753 long allocated;
e1aab161 1754 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1755
1756 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1757
e1aab161 1758 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1759
1760 /* Under limit. */
e1aab161
GC		 1761 if (parent_status == UNDER_LIMIT &&
1762 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1763 sk_leave_memory_pressure(sk);
3ab224be
HA		 1764 return 1;
1765 }
1766
e1aab161
GC		 1767 /* Under pressure. (we or our parents) */
1768 if ((parent_status > SOFT_LIMIT) ||
1769 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1770 sk_enter_memory_pressure(sk);
3ab224be 1771
e1aab161
GC		 1772 /* Over hard limit (we or our parents) */
1773 if ((parent_status == OVER_LIMIT) ||
1774 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1775 goto suppress_allocation;
1776
1777 /* guarantee minimum buffer size under pressure */
1778 if (kind == SK_MEM_RECV) {
1779 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1780 return 1;
180d8cd9 1781
3ab224be
HA		 1782 } else { /* SK_MEM_SEND */
1783 if (sk->sk_type == SOCK_STREAM) {
1784 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1785 return 1;
1786 } else if (atomic_read(&sk->sk_wmem_alloc) <
1787 prot->sysctl_wmem[0])
1788 return 1;
1789 }
1790
180d8cd9 1791 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1792 int alloc;
1793
180d8cd9 1794 if (!sk_under_memory_pressure(sk))
1748376b 1795 return 1;
180d8cd9
GC		 1796 alloc = sk_sockets_allocated_read_positive(sk);
1797 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1798 sk_mem_pages(sk->sk_wmem_queued +
1799 atomic_read(&sk->sk_rmem_alloc) +
1800 sk->sk_forward_alloc))
1801 return 1;
1802 }
1803
1804suppress_allocation:
1805
1806 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1807 sk_stream_moderate_sndbuf(sk);
1808
1809 /* Fail only if socket is _under_ its sndbuf.
1810 * In this case we cannot block, so that we have to fail.
1811 */
1812 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1813 return 1;
1814 }
1815
3847ce32
SM		 1816 trace_sock_exceed_buf_limit(sk, prot, allocated);
1817
3ab224be
HA		 1818 /* Alas. Undo changes. */
1819 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1820
e1aab161 1821 sk_memory_allocated_sub(sk, amt, parent_status);
180d8cd9 1822
3ab224be
HA		 1823 return 0;
1824}
3ab224be
HA		 1825EXPORT_SYMBOL(__sk_mem_schedule);
1826
1827/**
1828 * __sk_reclaim - reclaim memory_allocated
1829 * @sk: socket
1830 */
1831void __sk_mem_reclaim(struct sock *sk)
1832{
180d8cd9 1833 sk_memory_allocated_sub(sk,
e1aab161 1834 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT, 0);
3ab224be
HA		 1835 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1836
180d8cd9
GC		 1837 if (sk_under_memory_pressure(sk) &&
1838 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1839 sk_leave_memory_pressure(sk);
3ab224be 1840}
3ab224be
HA		 1841EXPORT_SYMBOL(__sk_mem_reclaim);
1842
1843
1da177e4
LT		 1844/*
1845 * Set of default routines for initialising struct proto_ops when
1846 * the protocol does not support a particular function. In certain
1847 * cases where it makes no sense for a protocol to have a "do nothing"
1848 * function, some default processing is provided.
1849 */
1850
1851int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1852{
1853 return -EOPNOTSUPP;
1854}
2a91525c 1855EXPORT_SYMBOL(sock_no_bind);
1da177e4 1856
4ec93edb 1857int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1858 int len, int flags)
1859{
1860 return -EOPNOTSUPP;
1861}
2a91525c 1862EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1863
1864int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1865{
1866 return -EOPNOTSUPP;
1867}
2a91525c 1868EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1869
1870int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1871{
1872 return -EOPNOTSUPP;
1873}
2a91525c 1874EXPORT_SYMBOL(sock_no_accept);
1da177e4 1875
4ec93edb 1876int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1877 int *len, int peer)
1878{
1879 return -EOPNOTSUPP;
1880}
2a91525c 1881EXPORT_SYMBOL(sock_no_getname);
1da177e4 1882
2a91525c 1883unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1884{
1885 return 0;
1886}
2a91525c 1887EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1888
1889int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1890{
1891 return -EOPNOTSUPP;
1892}
2a91525c 1893EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1894
1895int sock_no_listen(struct socket *sock, int backlog)
1896{
1897 return -EOPNOTSUPP;
1898}
2a91525c 1899EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1900
1901int sock_no_shutdown(struct socket *sock, int how)
1902{
1903 return -EOPNOTSUPP;
1904}
2a91525c 1905EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1906
1907int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1908 char __user *optval, unsigned int optlen)
1da177e4
LT		 1909{
1910 return -EOPNOTSUPP;
1911}
2a91525c 1912EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1913
1914int sock_no_getsockopt(struct socket *sock, int level, int optname,
1915 char __user *optval, int __user *optlen)
1916{
1917 return -EOPNOTSUPP;
1918}
2a91525c 1919EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1920
1921int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1922 size_t len)
1923{
1924 return -EOPNOTSUPP;
1925}
2a91525c 1926EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1927
1928int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1929 size_t len, int flags)
1930{
1931 return -EOPNOTSUPP;
1932}
2a91525c 1933EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1934
1935int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1936{
1937 /* Mirror missing mmap method error code */
1938 return -ENODEV;
1939}
2a91525c 1940EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1941
1942ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1943{
1944 ssize_t res;
1945 struct msghdr msg = {.msg_flags = flags};
1946 struct kvec iov;
1947 char *kaddr = kmap(page);
1948 iov.iov_base = kaddr + offset;
1949 iov.iov_len = size;
1950 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1951 kunmap(page);
1952 return res;
1953}
2a91525c 1954EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1955
1956/*
1957 * Default Socket Callbacks
1958 */
1959
1960static void sock_def_wakeup(struct sock *sk)
1961{
43815482
ED		 1962 struct socket_wq *wq;
1963
1964 rcu_read_lock();
1965 wq = rcu_dereference(sk->sk_wq);
1966 if (wq_has_sleeper(wq))
1967 wake_up_interruptible_all(&wq->wait);
1968 rcu_read_unlock();
1da177e4
LT		 1969}
1970
1971static void sock_def_error_report(struct sock *sk)
1972{
43815482
ED		 1973 struct socket_wq *wq;
1974
1975 rcu_read_lock();
1976 wq = rcu_dereference(sk->sk_wq);
1977 if (wq_has_sleeper(wq))
1978 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1979 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1980 rcu_read_unlock();
1da177e4
LT		 1981}
1982
1983static void sock_def_readable(struct sock *sk, int len)
1984{
43815482
ED		 1985 struct socket_wq *wq;
1986
1987 rcu_read_lock();
1988 wq = rcu_dereference(sk->sk_wq);
1989 if (wq_has_sleeper(wq))
2c6607c6 1990 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1991 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1992 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1993 rcu_read_unlock();
1da177e4
LT		 1994}
1995
1996static void sock_def_write_space(struct sock *sk)
1997{
43815482
ED		 1998 struct socket_wq *wq;
1999
2000 rcu_read_lock();
1da177e4
LT		 2001
2002 /* Do not wake up a writer until he can make "significant"
2003 * progress. --DaveM
2004 */
e71a4783 2005 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2006 wq = rcu_dereference(sk->sk_wq);
2007 if (wq_has_sleeper(wq))
2008 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2009 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2010
2011 /* Should agree with poll, otherwise some programs break */
2012 if (sock_writeable(sk))
8d8ad9d7 2013 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2014 }
2015
43815482 2016 rcu_read_unlock();
1da177e4
LT		 2017}
2018
2019static void sock_def_destruct(struct sock *sk)
2020{
a51482bd 2021 kfree(sk->sk_protinfo);
1da177e4
LT		 2022}
2023
2024void sk_send_sigurg(struct sock *sk)
2025{
2026 if (sk->sk_socket && sk->sk_socket->file)
2027 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2028 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2029}
2a91525c 2030EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2031
2032void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2033 unsigned long expires)
2034{
2035 if (!mod_timer(timer, expires))
2036 sock_hold(sk);
2037}
1da177e4
LT		 2038EXPORT_SYMBOL(sk_reset_timer);
2039
2040void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2041{
2042 if (timer_pending(timer) && del_timer(timer))
2043 __sock_put(sk);
2044}
1da177e4
LT		 2045EXPORT_SYMBOL(sk_stop_timer);
2046
2047void sock_init_data(struct socket *sock, struct sock *sk)
2048{
2049 skb_queue_head_init(&sk->sk_receive_queue);
2050 skb_queue_head_init(&sk->sk_write_queue);
2051 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2052#ifdef CONFIG_NET_DMA
2053 skb_queue_head_init(&sk->sk_async_wait_queue);
2054#endif
1da177e4
LT		 2055
2056 sk->sk_send_head = NULL;
2057
2058 init_timer(&sk->sk_timer);
4ec93edb 2059
1da177e4
LT		 2060 sk->sk_allocation = GFP_KERNEL;
2061 sk->sk_rcvbuf = sysctl_rmem_default;
2062 sk->sk_sndbuf = sysctl_wmem_default;
2063 sk->sk_state = TCP_CLOSE;
972692e0 2064 sk_set_socket(sk, sock);
1da177e4
LT		 2065
2066 sock_set_flag(sk, SOCK_ZAPPED);
2067
e71a4783 2068 if (sock) {
1da177e4 2069 sk->sk_type = sock->type;
43815482 2070 sk->sk_wq = sock->wq;
1da177e4
LT		 2071 sock->sk = sk;
2072 } else
43815482 2073 sk->sk_wq = NULL;
1da177e4 2074
b6c6712a 2075 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2076 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2077 lockdep_set_class_and_name(&sk->sk_callback_lock,
2078 af_callback_keys + sk->sk_family,
2079 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2080
2081 sk->sk_state_change = sock_def_wakeup;
2082 sk->sk_data_ready = sock_def_readable;
2083 sk->sk_write_space = sock_def_write_space;
2084 sk->sk_error_report = sock_def_error_report;
2085 sk->sk_destruct = sock_def_destruct;
2086
2087 sk->sk_sndmsg_page = NULL;
2088 sk->sk_sndmsg_off = 0;
2089
109f6e39
EB		 2090 sk->sk_peer_pid = NULL;
2091 sk->sk_peer_cred = NULL;
1da177e4
LT		 2092 sk->sk_write_pending = 0;
2093 sk->sk_rcvlowat = 1;
2094 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2095 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2096
f37f0afb 2097 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2098
4dc6dc71
ED		 2099 /*
2100 * Before updating sk_refcnt, we must commit prior changes to memory
2101 * (Documentation/RCU/rculist_nulls.txt for details)
2102 */
2103 smp_wmb();
1da177e4 2104 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2105 atomic_set(&sk->sk_drops, 0);
1da177e4 2106}
2a91525c 2107EXPORT_SYMBOL(sock_init_data);
1da177e4 2108
b5606c2d 2109void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2110{
2111 might_sleep();
a5b5bb9a 2112 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2113 if (sk->sk_lock.owned)
1da177e4 2114 __lock_sock(sk);
d2e9117c 2115 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2116 spin_unlock(&sk->sk_lock.slock);
2117 /*
2118 * The sk_lock has mutex_lock() semantics here:
2119 */
fcc70d5f 2120 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2121 local_bh_enable();
1da177e4 2122}
fcc70d5f 2123EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2124
b5606c2d 2125void release_sock(struct sock *sk)
1da177e4 2126{
a5b5bb9a
IM		 2127 /*
2128 * The sk_lock has mutex_unlock() semantics:
2129 */
2130 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2131
2132 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2133 if (sk->sk_backlog.tail)
2134 __release_sock(sk);
d2e9117c 2135 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2136 if (waitqueue_active(&sk->sk_lock.wq))
2137 wake_up(&sk->sk_lock.wq);
2138 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2139}
2140EXPORT_SYMBOL(release_sock);
2141
8a74ad60
ED		 2142/**
2143 * lock_sock_fast - fast version of lock_sock
2144 * @sk: socket
2145 *
2146 * This version should be used for very small section, where process wont block
2147 * return false if fast path is taken
2148 * sk_lock.slock locked, owned = 0, BH disabled
2149 * return true if slow path is taken
2150 * sk_lock.slock unlocked, owned = 1, BH enabled
2151 */
2152bool lock_sock_fast(struct sock *sk)
2153{
2154 might_sleep();
2155 spin_lock_bh(&sk->sk_lock.slock);
2156
2157 if (!sk->sk_lock.owned)
2158 /*
2159 * Note : We must disable BH
2160 */
2161 return false;
2162
2163 __lock_sock(sk);
2164 sk->sk_lock.owned = 1;
2165 spin_unlock(&sk->sk_lock.slock);
2166 /*
2167 * The sk_lock has mutex_lock() semantics here:
2168 */
2169 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2170 local_bh_enable();
2171 return true;
2172}
2173EXPORT_SYMBOL(lock_sock_fast);
2174
1da177e4 2175int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2176{
b7aa0bf7 2177 struct timeval tv;
1da177e4 2178 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2179 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2180 tv = ktime_to_timeval(sk->sk_stamp);
2181 if (tv.tv_sec == -1)
1da177e4 2182 return -ENOENT;
b7aa0bf7
ED		 2183 if (tv.tv_sec == 0) {
2184 sk->sk_stamp = ktime_get_real();
2185 tv = ktime_to_timeval(sk->sk_stamp);
2186 }
2187 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2188}
1da177e4
LT		 2189EXPORT_SYMBOL(sock_get_timestamp);
2190
ae40eb1e
ED		 2191int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2192{
2193 struct timespec ts;
2194 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2195 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2196 ts = ktime_to_timespec(sk->sk_stamp);
2197 if (ts.tv_sec == -1)
2198 return -ENOENT;
2199 if (ts.tv_sec == 0) {
2200 sk->sk_stamp = ktime_get_real();
2201 ts = ktime_to_timespec(sk->sk_stamp);
2202 }
2203 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2204}
2205EXPORT_SYMBOL(sock_get_timestampns);
2206
20d49473 2207void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2208{
20d49473 2209 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2210 unsigned long previous_flags = sk->sk_flags;
2211
20d49473
PO		 2212 sock_set_flag(sk, flag);
2213 /*
2214 * we just set one of the two flags which require net
2215 * time stamping, but time stamping might have been on
2216 * already because of the other one
2217 */
08e29af3 2218 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2219 net_enable_timestamp();
1da177e4
LT		 2220 }
2221}
1da177e4
LT		 2222
2223/*
2224 * Get a socket option on an socket.
2225 *
2226 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2227 * asynchronous errors should be reported by getsockopt. We assume
2228 * this means if you specify SO_ERROR (otherwise whats the point of it).
2229 */
2230int sock_common_getsockopt(struct socket *sock, int level, int optname,
2231 char __user *optval, int __user *optlen)
2232{
2233 struct sock *sk = sock->sk;
2234
2235 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2236}
1da177e4
LT		 2237EXPORT_SYMBOL(sock_common_getsockopt);
2238
3fdadf7d 2239#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2240int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2241 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2242{
2243 struct sock *sk = sock->sk;
2244
1e51f951 2245 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2246 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2247 optval, optlen);
3fdadf7d
DM		 2248 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2249}
2250EXPORT_SYMBOL(compat_sock_common_getsockopt);
2251#endif
2252
1da177e4
LT		 2253int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2254 struct msghdr *msg, size_t size, int flags)
2255{
2256 struct sock *sk = sock->sk;
2257 int addr_len = 0;
2258 int err;
2259
2260 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2261 flags & ~MSG_DONTWAIT, &addr_len);
2262 if (err >= 0)
2263 msg->msg_namelen = addr_len;
2264 return err;
2265}
1da177e4
LT		 2266EXPORT_SYMBOL(sock_common_recvmsg);
2267
2268/*
2269 * Set socket options on an inet socket.
2270 */
2271int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2272 char __user *optval, unsigned int optlen)
1da177e4
LT		 2273{
2274 struct sock *sk = sock->sk;
2275
2276 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2277}
1da177e4
LT		 2278EXPORT_SYMBOL(sock_common_setsockopt);
2279
3fdadf7d 2280#ifdef CONFIG_COMPAT
543d9cfe 2281int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2282 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2283{
2284 struct sock *sk = sock->sk;
2285
543d9cfe
ACM		 2286 if (sk->sk_prot->compat_setsockopt != NULL)
2287 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2288 optval, optlen);
3fdadf7d
DM		 2289 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2290}
2291EXPORT_SYMBOL(compat_sock_common_setsockopt);
2292#endif
2293
1da177e4
LT		 2294void sk_common_release(struct sock *sk)
2295{
2296 if (sk->sk_prot->destroy)
2297 sk->sk_prot->destroy(sk);
2298
2299 /*
2300 * Observation: when sock_common_release is called, processes have
2301 * no access to socket. But net still has.
2302 * Step one, detach it from networking:
2303 *
2304 * A. Remove from hash tables.
2305 */
2306
2307 sk->sk_prot->unhash(sk);
2308
2309 /*
2310 * In this point socket cannot receive new packets, but it is possible
2311 * that some packets are in flight because some CPU runs receiver and
2312 * did hash table lookup before we unhashed socket. They will achieve
2313 * receive queue and will be purged by socket destructor.
2314 *
2315 * Also we still have packets pending on receive queue and probably,
2316 * our own packets waiting in device queues. sock_destroy will drain
2317 * receive queue, but transmitted packets will delay socket destruction
2318 * until the last reference will be released.
2319 */
2320
2321 sock_orphan(sk);
2322
2323 xfrm_sk_free_policy(sk);
2324
e6848976 2325 sk_refcnt_debug_release(sk);
1da177e4
LT		 2326 sock_put(sk);
2327}
1da177e4
LT		 2328EXPORT_SYMBOL(sk_common_release);
2329
13ff3d6f
PE		 2330#ifdef CONFIG_PROC_FS
2331#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2332struct prot_inuse {
2333 int val[PROTO_INUSE_NR];
2334};
13ff3d6f
PE		 2335
2336static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2337
2338#ifdef CONFIG_NET_NS
2339void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2340{
d6d9ca0f 2341 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2342}
2343EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2344
2345int sock_prot_inuse_get(struct net *net, struct proto *prot)
2346{
2347 int cpu, idx = prot->inuse_idx;
2348 int res = 0;
2349
2350 for_each_possible_cpu(cpu)
2351 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2352
2353 return res >= 0 ? res : 0;
2354}
2355EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2356
2c8c1e72 2357static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2358{
2359 net->core.inuse = alloc_percpu(struct prot_inuse);
2360 return net->core.inuse ? 0 : -ENOMEM;
2361}
2362
2c8c1e72 2363static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2364{
2365 free_percpu(net->core.inuse);
2366}
2367
2368static struct pernet_operations net_inuse_ops = {
2369 .init = sock_inuse_init_net,
2370 .exit = sock_inuse_exit_net,
2371};
2372
2373static __init int net_inuse_init(void)
2374{
2375 if (register_pernet_subsys(&net_inuse_ops))
2376 panic("Cannot initialize net inuse counters");
2377
2378 return 0;
2379}
2380
2381core_initcall(net_inuse_init);
2382#else
1338d466
PE		 2383static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2384
c29a0bc4 2385void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2386{
d6d9ca0f 2387 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2388}
2389EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2390
c29a0bc4 2391int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2392{
2393 int cpu, idx = prot->inuse_idx;
2394 int res = 0;
2395
2396 for_each_possible_cpu(cpu)
2397 res += per_cpu(prot_inuse, cpu).val[idx];
2398
2399 return res >= 0 ? res : 0;
2400}
2401EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2402#endif
13ff3d6f
PE		 2403
2404static void assign_proto_idx(struct proto *prot)
2405{
2406 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2407
2408 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2409 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2410 return;
2411 }
2412
2413 set_bit(prot->inuse_idx, proto_inuse_idx);
2414}
2415
2416static void release_proto_idx(struct proto *prot)
2417{
2418 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2419 clear_bit(prot->inuse_idx, proto_inuse_idx);
2420}
2421#else
2422static inline void assign_proto_idx(struct proto *prot)
2423{
2424}
2425
2426static inline void release_proto_idx(struct proto *prot)
2427{
2428}
2429#endif
2430
b733c007
PE		 2431int proto_register(struct proto *prot, int alloc_slab)
2432{
1da177e4
LT		 2433 if (alloc_slab) {
2434 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2435 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2436 NULL);
1da177e4
LT		 2437
2438 if (prot->slab == NULL) {
2439 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2440 prot->name);
60e7663d 2441 goto out;
1da177e4 2442 }
2e6599cb
ACM		 2443
2444 if (prot->rsk_prot != NULL) {
faf23422 2445 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2446 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2447 goto out_free_sock_slab;
2448
7e56b5d6 2449 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2450 prot->rsk_prot->obj_size, 0,
20c2df83 2451 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2452
2453 if (prot->rsk_prot->slab == NULL) {
2454 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2455 prot->name);
2456 goto out_free_request_sock_slab_name;
2457 }
2458 }
8feaf0c0 2459
6d6ee43e 2460 if (prot->twsk_prot != NULL) {
faf23422 2461 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2462
7e56b5d6 2463 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2464 goto out_free_request_sock_slab;
2465
6d6ee43e 2466 prot->twsk_prot->twsk_slab =
7e56b5d6 2467 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2468 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2469 0,
2470 SLAB_HWCACHE_ALIGN |
2471 prot->slab_flags,
20c2df83 2472 NULL);
6d6ee43e 2473 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2474 goto out_free_timewait_sock_slab_name;
2475 }
1da177e4
LT		 2476 }
2477
36b77a52 2478 mutex_lock(&proto_list_mutex);
1da177e4 2479 list_add(&prot->node, &proto_list);
13ff3d6f 2480 assign_proto_idx(prot);
36b77a52 2481 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2482 return 0;
2483
8feaf0c0 2484out_free_timewait_sock_slab_name:
7e56b5d6 2485 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2486out_free_request_sock_slab:
2487 if (prot->rsk_prot && prot->rsk_prot->slab) {
2488 kmem_cache_destroy(prot->rsk_prot->slab);
2489 prot->rsk_prot->slab = NULL;
2490 }
2e6599cb 2491out_free_request_sock_slab_name:
72150e9b
DC		 2492 if (prot->rsk_prot)
2493 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2494out_free_sock_slab:
2495 kmem_cache_destroy(prot->slab);
2496 prot->slab = NULL;
b733c007
PE		 2497out:
2498 return -ENOBUFS;
1da177e4 2499}
1da177e4
LT		 2500EXPORT_SYMBOL(proto_register);
2501
2502void proto_unregister(struct proto *prot)
2503{
36b77a52 2504 mutex_lock(&proto_list_mutex);
13ff3d6f 2505 release_proto_idx(prot);
0a3f4358 2506 list_del(&prot->node);
36b77a52 2507 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2508
2509 if (prot->slab != NULL) {
2510 kmem_cache_destroy(prot->slab);
2511 prot->slab = NULL;
2512 }
2513
2e6599cb 2514 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2515 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2516 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2517 prot->rsk_prot->slab = NULL;
2518 }
2519
6d6ee43e 2520 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2521 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2522 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2523 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2524 }
1da177e4 2525}
1da177e4
LT		 2526EXPORT_SYMBOL(proto_unregister);
2527
2528#ifdef CONFIG_PROC_FS
1da177e4 2529static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2530 __acquires(proto_list_mutex)
1da177e4 2531{
36b77a52 2532 mutex_lock(&proto_list_mutex);
60f0438a 2533 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2534}
2535
2536static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2537{
60f0438a 2538 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2539}
2540
2541static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2542 __releases(proto_list_mutex)
1da177e4 2543{
36b77a52 2544 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2545}
2546
2547static char proto_method_implemented(const void *method)
2548{
2549 return method == NULL ? 'n' : 'y';
2550}
180d8cd9
GC		 2551static long sock_prot_memory_allocated(struct proto *proto)
2552{
2553 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2554}
2555
2556static char *sock_prot_memory_pressure(struct proto *proto)
2557{
2558 return proto->memory_pressure != NULL ?
2559 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2560}
1da177e4
LT		 2561
2562static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2563{
180d8cd9 2564
8d987e5c 2565 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2566 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2567 proto->name,
2568 proto->obj_size,
14e943db 2569 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2570 sock_prot_memory_allocated(proto),
2571 sock_prot_memory_pressure(proto),
1da177e4
LT		 2572 proto->max_header,
2573 proto->slab == NULL ? "no" : "yes",
2574 module_name(proto->owner),
2575 proto_method_implemented(proto->close),
2576 proto_method_implemented(proto->connect),
2577 proto_method_implemented(proto->disconnect),
2578 proto_method_implemented(proto->accept),
2579 proto_method_implemented(proto->ioctl),
2580 proto_method_implemented(proto->init),
2581 proto_method_implemented(proto->destroy),
2582 proto_method_implemented(proto->shutdown),
2583 proto_method_implemented(proto->setsockopt),
2584 proto_method_implemented(proto->getsockopt),
2585 proto_method_implemented(proto->sendmsg),
2586 proto_method_implemented(proto->recvmsg),
2587 proto_method_implemented(proto->sendpage),
2588 proto_method_implemented(proto->bind),
2589 proto_method_implemented(proto->backlog_rcv),
2590 proto_method_implemented(proto->hash),
2591 proto_method_implemented(proto->unhash),
2592 proto_method_implemented(proto->get_port),
2593 proto_method_implemented(proto->enter_memory_pressure));
2594}
2595
2596static int proto_seq_show(struct seq_file *seq, void *v)
2597{
60f0438a 2598 if (v == &proto_list)
1da177e4
LT		 2599 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2600 "protocol",
2601 "size",
2602 "sockets",
2603 "memory",
2604 "press",
2605 "maxhdr",
2606 "slab",
2607 "module",
2608 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2609 else
60f0438a 2610 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2611 return 0;
2612}
2613
f690808e 2614static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2615 .start = proto_seq_start,
2616 .next = proto_seq_next,
2617 .stop = proto_seq_stop,
2618 .show = proto_seq_show,
2619};
2620
2621static int proto_seq_open(struct inode *inode, struct file *file)
2622{
14e943db
ED		 2623 return seq_open_net(inode, file, &proto_seq_ops,
2624 sizeof(struct seq_net_private));
1da177e4
LT		 2625}
2626
9a32144e 2627static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2628 .owner = THIS_MODULE,
2629 .open = proto_seq_open,
2630 .read = seq_read,
2631 .llseek = seq_lseek,
14e943db
ED		 2632 .release = seq_release_net,
2633};
2634
2635static __net_init int proto_init_net(struct net *net)
2636{
2637 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2638 return -ENOMEM;
2639
2640 return 0;
2641}
2642
2643static __net_exit void proto_exit_net(struct net *net)
2644{
2645 proc_net_remove(net, "protocols");
2646}
2647
2648
2649static __net_initdata struct pernet_operations proto_net_ops = {
2650 .init = proto_init_net,
2651 .exit = proto_exit_net,
1da177e4
LT		 2652};
2653
2654static int __init proto_init(void)
2655{
14e943db 2656 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2657}
2658
2659subsys_initcall(proto_init);
2660
2661#endif /* PROC_FS */
Linux 6.x block layer and io_uring tree(s)