projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
inet: Sanitize inet{,6} protocol demux.
[deliverable/linux.git] / net / core / sock.c
CommitLineData
1da177e4
LT		 1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
4ec93edb 35 * code. The ACK stuff can wait and needs major
1da177e4
LT		 36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
e005d193
JP		 92#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
4fc268d2 94#include <linux/capability.h>
1da177e4
LT		 95#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
1da177e4
LT		 100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
a1f8e7f7 114#include <linux/highmem.h>
3f551f94 115#include <linux/user_namespace.h>
c5905afb 116#include <linux/static_key.h>
3969eb38 117#include <linux/memcontrol.h>
8c1ae10d 118#include <linux/prefetch.h>
1da177e4
LT		 119
120#include <asm/uaccess.h>
1da177e4
LT		 121
122#include <linux/netdevice.h>
123#include <net/protocol.h>
124#include <linux/skbuff.h>
457c4cbc 125#include <net/net_namespace.h>
2e6599cb 126#include <net/request_sock.h>
1da177e4 127#include <net/sock.h>
20d49473 128#include <linux/net_tstamp.h>
1da177e4
LT		 129#include <net/xfrm.h>
130#include <linux/ipsec.h>
f8451725 131#include <net/cls_cgroup.h>
5bc1421e 132#include <net/netprio_cgroup.h>
1da177e4
LT		 133
134#include <linux/filter.h>
135
3847ce32
SM		 136#include <trace/events/sock.h>
137
1da177e4
LT		 138#ifdef CONFIG_INET
139#include <net/tcp.h>
140#endif
141
36b77a52 142static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 143static LIST_HEAD(proto_list);
144
145#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
1d62e436 146int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
d1a4c0b3
GC		 147{
148 struct proto *proto;
149 int ret = 0;
150
36b77a52 151 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
1d62e436 154 ret = proto->init_cgroup(memcg, ss);
d1a4c0b3
GC		 155 if (ret)
156 goto out;
157 }
158 }
159
36b77a52 160 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 161 return ret;
162out:
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
1d62e436 165 proto->destroy_cgroup(memcg);
36b77a52 166 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 167 return ret;
168}
169
1d62e436 170void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
d1a4c0b3
GC		 171{
172 struct proto *proto;
173
36b77a52 174 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
1d62e436 177 proto->destroy_cgroup(memcg);
36b77a52 178 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 179}
180#endif
181
da21f24d
IM		 182/*
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
185 */
a5b5bb9a
IM		 186static struct lock_class_key af_family_keys[AF_MAX];
187static struct lock_class_key af_family_slock_keys[AF_MAX];
188
c5905afb 189struct static_key memcg_socket_limit_enabled;
e1aab161
GC		 190EXPORT_SYMBOL(memcg_socket_limit_enabled);
191
a5b5bb9a
IM		 192/*
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
195 * locks is fast):
196 */
36cbd3dc 197static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 212};
36cbd3dc 213static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 227 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 228};
36cbd3dc 229static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 243 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 244};
da21f24d
IM		 245
246/*
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
249 */
250static struct lock_class_key af_callback_keys[AF_MAX];
251
1da177e4
LT		 252/* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
256 */
257#define _SK_MEM_PACKETS 256
87fb4b7b 258#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 259#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261
262/* Run time adjustable parameters. */
ab32ea5d 263__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
6d8ebc8a 264EXPORT_SYMBOL(sysctl_wmem_max);
ab32ea5d 265__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
6d8ebc8a 266EXPORT_SYMBOL(sysctl_rmem_max);
ab32ea5d
BH		 267__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 269
25985edc 270/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 271int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 272EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 273
5bc1421e
NH		 274#if defined(CONFIG_CGROUPS)
275#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 276int net_cls_subsys_id = -1;
277EXPORT_SYMBOL_GPL(net_cls_subsys_id);
278#endif
5bc1421e
NH		 279#if !defined(CONFIG_NETPRIO_CGROUP)
280int net_prio_subsys_id = -1;
281EXPORT_SYMBOL_GPL(net_prio_subsys_id);
282#endif
283#endif
f8451725 284
1da177e4
LT		 285static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
286{
287 struct timeval tv;
288
289 if (optlen < sizeof(tv))
290 return -EINVAL;
291 if (copy_from_user(&tv, optval, sizeof(tv)))
292 return -EFAULT;
ba78073e
VA		 293 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
294 return -EDOM;
1da177e4 295
ba78073e 296 if (tv.tv_sec < 0) {
6f11df83
AM		 297 static int warned __read_mostly;
298
ba78073e 299 *timeo_p = 0;
50aab54f 300 if (warned < 10 && net_ratelimit()) {
ba78073e 301 warned++;
e005d193
JP		 302 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
303 __func__, current->comm, task_pid_nr(current));
50aab54f 304 }
ba78073e
VA		 305 return 0;
306 }
1da177e4
LT		 307 *timeo_p = MAX_SCHEDULE_TIMEOUT;
308 if (tv.tv_sec == 0 && tv.tv_usec == 0)
309 return 0;
310 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
311 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
312 return 0;
313}
314
315static void sock_warn_obsolete_bsdism(const char *name)
316{
317 static int warned;
318 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 319 if (strcmp(warncomm, current->comm) && warned < 5) {
320 strcpy(warncomm, current->comm);
e005d193
JP		 321 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
322 warncomm, name);
1da177e4
LT		 323 warned++;
324 }
325}
326
08e29af3
ED		 327#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
328
329static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 330{
08e29af3
ED		 331 if (sk->sk_flags & flags) {
332 sk->sk_flags &= ~flags;
333 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 334 net_disable_timestamp();
1da177e4
LT		 335 }
336}
337
338
f0088a50
DV		 339int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
340{
766e9037 341 int err;
f0088a50 342 int skb_len;
3b885787
NH		 343 unsigned long flags;
344 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 345
0fd7bac6 346 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 347 atomic_inc(&sk->sk_drops);
3847ce32 348 trace_sock_rcvqueue_full(sk, skb);
766e9037 349 return -ENOMEM;
f0088a50
DV		 350 }
351
fda9ef5d 352 err = sk_filter(sk, skb);
f0088a50 353 if (err)
766e9037 354 return err;
f0088a50 355
3ab224be 356 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 357 atomic_inc(&sk->sk_drops);
358 return -ENOBUFS;
3ab224be
HA		 359 }
360
f0088a50
DV		 361 skb->dev = NULL;
362 skb_set_owner_r(skb, sk);
49ad9599 363
f0088a50
DV		 364 /* Cache the SKB length before we tack it onto the receive
365 * queue. Once it is added it no longer belongs to us and
366 * may be freed by other threads of control pulling packets
367 * from the queue.
368 */
369 skb_len = skb->len;
370
7fee226a
ED		 371 /* we escape from rcu protected region, make sure we dont leak
372 * a norefcounted dst
373 */
374 skb_dst_force(skb);
375
3b885787
NH		 376 spin_lock_irqsave(&list->lock, flags);
377 skb->dropcount = atomic_read(&sk->sk_drops);
378 __skb_queue_tail(list, skb);
379 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 380
381 if (!sock_flag(sk, SOCK_DEAD))
382 sk->sk_data_ready(sk, skb_len);
766e9037 383 return 0;
f0088a50
DV		 384}
385EXPORT_SYMBOL(sock_queue_rcv_skb);
386
58a5a7b9 387int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 388{
389 int rc = NET_RX_SUCCESS;
390
fda9ef5d 391 if (sk_filter(sk, skb))
f0088a50
DV		 392 goto discard_and_relse;
393
394 skb->dev = NULL;
395
f545a38f 396 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
c377411f
ED		 397 atomic_inc(&sk->sk_drops);
398 goto discard_and_relse;
399 }
58a5a7b9
ACM		 400 if (nested)
401 bh_lock_sock_nested(sk);
402 else
403 bh_lock_sock(sk);
a5b5bb9a
IM		 404 if (!sock_owned_by_user(sk)) {
405 /*
406 * trylock + unlock semantics:
407 */
408 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
409
c57943a1 410 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 411
412 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
f545a38f 413 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
8eae939f
ZY		 414 bh_unlock_sock(sk);
415 atomic_inc(&sk->sk_drops);
416 goto discard_and_relse;
417 }
418
f0088a50
DV		 419 bh_unlock_sock(sk);
420out:
421 sock_put(sk);
422 return rc;
423discard_and_relse:
424 kfree_skb(skb);
425 goto out;
426}
427EXPORT_SYMBOL(sk_receive_skb);
428
ea94ff3b
KK		 429void sk_reset_txq(struct sock *sk)
430{
431 sk_tx_queue_clear(sk);
432}
433EXPORT_SYMBOL(sk_reset_txq);
434
f0088a50
DV		 435struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
436{
b6c6712a 437 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 438
439 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 440 sk_tx_queue_clear(sk);
a9b3cd7f 441 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 442 dst_release(dst);
443 return NULL;
444 }
445
446 return dst;
447}
448EXPORT_SYMBOL(__sk_dst_check);
449
450struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
451{
452 struct dst_entry *dst = sk_dst_get(sk);
453
454 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
455 sk_dst_reset(sk);
456 dst_release(dst);
457 return NULL;
458 }
459
460 return dst;
461}
462EXPORT_SYMBOL(sk_dst_check);
463
4878809f
DM		 464static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
465{
466 int ret = -ENOPROTOOPT;
467#ifdef CONFIG_NETDEVICES
3b1e0a65 468 struct net *net = sock_net(sk);
4878809f
DM		 469 char devname[IFNAMSIZ];
470 int index;
471
472 /* Sorry... */
473 ret = -EPERM;
474 if (!capable(CAP_NET_RAW))
475 goto out;
476
477 ret = -EINVAL;
478 if (optlen < 0)
479 goto out;
480
481 /* Bind this socket to a particular device like "eth0",
482 * as specified in the passed interface name. If the
483 * name is "" or the option length is zero the socket
484 * is not bound.
485 */
486 if (optlen > IFNAMSIZ - 1)
487 optlen = IFNAMSIZ - 1;
488 memset(devname, 0, sizeof(devname));
489
490 ret = -EFAULT;
491 if (copy_from_user(devname, optval, optlen))
492 goto out;
493
000ba2e4
DM		 494 index = 0;
495 if (devname[0] != '\0') {
bf8e56bf 496 struct net_device *dev;
4878809f 497
bf8e56bf
ED		 498 rcu_read_lock();
499 dev = dev_get_by_name_rcu(net, devname);
500 if (dev)
501 index = dev->ifindex;
502 rcu_read_unlock();
4878809f
DM		 503 ret = -ENODEV;
504 if (!dev)
505 goto out;
4878809f
DM		 506 }
507
508 lock_sock(sk);
509 sk->sk_bound_dev_if = index;
510 sk_dst_reset(sk);
511 release_sock(sk);
512
513 ret = 0;
514
515out:
516#endif
517
518 return ret;
519}
520
c0ef877b
PE		 521static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
522{
523 if (valbool)
524 sock_set_flag(sk, bit);
525 else
526 sock_reset_flag(sk, bit);
527}
528
1da177e4
LT		 529/*
530 * This is meant for all protocols to use and covers goings on
531 * at the socket level. Everything here is generic.
532 */
533
534int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 535 char __user *optval, unsigned int optlen)
1da177e4 536{
2a91525c 537 struct sock *sk = sock->sk;
1da177e4
LT		 538 int val;
539 int valbool;
540 struct linger ling;
541 int ret = 0;
4ec93edb 542
1da177e4
LT		 543 /*
544 * Options without arguments
545 */
546
4878809f
DM		 547 if (optname == SO_BINDTODEVICE)
548 return sock_bindtodevice(sk, optval, optlen);
549
e71a4783
SH		 550 if (optlen < sizeof(int))
551 return -EINVAL;
4ec93edb 552
1da177e4
LT		 553 if (get_user(val, (int __user *)optval))
554 return -EFAULT;
4ec93edb 555
2a91525c 556 valbool = val ? 1 : 0;
1da177e4
LT		 557
558 lock_sock(sk);
559
2a91525c 560 switch (optname) {
e71a4783 561 case SO_DEBUG:
2a91525c 562 if (val && !capable(CAP_NET_ADMIN))
e71a4783 563 ret = -EACCES;
2a91525c 564 else
c0ef877b 565 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 566 break;
567 case SO_REUSEADDR:
4a17fd52 568 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
e71a4783
SH		 569 break;
570 case SO_TYPE:
49c794e9 571 case SO_PROTOCOL:
0d6038ee 572 case SO_DOMAIN:
e71a4783
SH		 573 case SO_ERROR:
574 ret = -ENOPROTOOPT;
575 break;
576 case SO_DONTROUTE:
c0ef877b 577 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 578 break;
579 case SO_BROADCAST:
580 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
581 break;
582 case SO_SNDBUF:
583 /* Don't error on this BSD doesn't and if you think
82981930
ED		 584 * about it this is right. Otherwise apps have to
585 * play 'guess the biggest size' games. RCVBUF/SNDBUF
586 * are treated in BSD as hints
587 */
588 val = min_t(u32, val, sysctl_wmem_max);
b0573dea 589set_sndbuf:
e71a4783 590 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
82981930
ED		 591 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
592 /* Wake up sending tasks if we upped the value. */
e71a4783
SH		 593 sk->sk_write_space(sk);
594 break;
1da177e4 595
e71a4783
SH		 596 case SO_SNDBUFFORCE:
597 if (!capable(CAP_NET_ADMIN)) {
598 ret = -EPERM;
599 break;
600 }
601 goto set_sndbuf;
b0573dea 602
e71a4783
SH		 603 case SO_RCVBUF:
604 /* Don't error on this BSD doesn't and if you think
82981930
ED		 605 * about it this is right. Otherwise apps have to
606 * play 'guess the biggest size' games. RCVBUF/SNDBUF
607 * are treated in BSD as hints
608 */
609 val = min_t(u32, val, sysctl_rmem_max);
b0573dea 610set_rcvbuf:
e71a4783
SH		 611 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
612 /*
613 * We double it on the way in to account for
614 * "struct sk_buff" etc. overhead. Applications
615 * assume that the SO_RCVBUF setting they make will
616 * allow that much actual data to be received on that
617 * socket.
618 *
619 * Applications are unaware that "struct sk_buff" and
620 * other overheads allocate from the receive buffer
621 * during socket buffer allocation.
622 *
623 * And after considering the possible alternatives,
624 * returning the value we actually used in getsockopt
625 * is the most desirable behavior.
626 */
82981930 627 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
e71a4783
SH		 628 break;
629
630 case SO_RCVBUFFORCE:
631 if (!capable(CAP_NET_ADMIN)) {
632 ret = -EPERM;
1da177e4 633 break;
e71a4783
SH		 634 }
635 goto set_rcvbuf;
1da177e4 636
e71a4783 637 case SO_KEEPALIVE:
1da177e4 638#ifdef CONFIG_INET
e71a4783
SH		 639 if (sk->sk_protocol == IPPROTO_TCP)
640 tcp_set_keepalive(sk, valbool);
1da177e4 641#endif
e71a4783
SH		 642 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
643 break;
644
645 case SO_OOBINLINE:
646 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
647 break;
648
649 case SO_NO_CHECK:
650 sk->sk_no_check = valbool;
651 break;
652
653 case SO_PRIORITY:
654 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
655 sk->sk_priority = val;
656 else
657 ret = -EPERM;
658 break;
659
660 case SO_LINGER:
661 if (optlen < sizeof(ling)) {
662 ret = -EINVAL; /* 1003.1g */
1da177e4 663 break;
e71a4783 664 }
2a91525c 665 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 666 ret = -EFAULT;
1da177e4 667 break;
e71a4783
SH		 668 }
669 if (!ling.l_onoff)
670 sock_reset_flag(sk, SOCK_LINGER);
671 else {
1da177e4 672#if (BITS_PER_LONG == 32)
e71a4783
SH		 673 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
674 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 675 else
e71a4783
SH		 676#endif
677 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
678 sock_set_flag(sk, SOCK_LINGER);
679 }
680 break;
681
682 case SO_BSDCOMPAT:
683 sock_warn_obsolete_bsdism("setsockopt");
684 break;
685
686 case SO_PASSCRED:
687 if (valbool)
688 set_bit(SOCK_PASSCRED, &sock->flags);
689 else
690 clear_bit(SOCK_PASSCRED, &sock->flags);
691 break;
692
693 case SO_TIMESTAMP:
92f37fd2 694 case SO_TIMESTAMPNS:
e71a4783 695 if (valbool) {
92f37fd2
ED		 696 if (optname == SO_TIMESTAMP)
697 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
698 else
699 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 700 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 701 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 702 } else {
e71a4783 703 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 704 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
705 }
e71a4783
SH		 706 break;
707
20d49473
PO		 708 case SO_TIMESTAMPING:
709 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 710 ret = -EINVAL;
20d49473
PO		 711 break;
712 }
713 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
714 val & SOF_TIMESTAMPING_TX_HARDWARE);
715 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
716 val & SOF_TIMESTAMPING_TX_SOFTWARE);
717 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
718 val & SOF_TIMESTAMPING_RX_HARDWARE);
719 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
720 sock_enable_timestamp(sk,
721 SOCK_TIMESTAMPING_RX_SOFTWARE);
722 else
723 sock_disable_timestamp(sk,
08e29af3 724 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
726 val & SOF_TIMESTAMPING_SOFTWARE);
727 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
728 val & SOF_TIMESTAMPING_SYS_HARDWARE);
729 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
730 val & SOF_TIMESTAMPING_RAW_HARDWARE);
731 break;
732
e71a4783
SH		 733 case SO_RCVLOWAT:
734 if (val < 0)
735 val = INT_MAX;
736 sk->sk_rcvlowat = val ? : 1;
737 break;
738
739 case SO_RCVTIMEO:
740 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
741 break;
742
743 case SO_SNDTIMEO:
744 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
745 break;
1da177e4 746
e71a4783
SH		 747 case SO_ATTACH_FILTER:
748 ret = -EINVAL;
749 if (optlen == sizeof(struct sock_fprog)) {
750 struct sock_fprog fprog;
1da177e4 751
e71a4783
SH		 752 ret = -EFAULT;
753 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 754 break;
e71a4783
SH		 755
756 ret = sk_attach_filter(&fprog, sk);
757 }
758 break;
759
760 case SO_DETACH_FILTER:
55b33325 761 ret = sk_detach_filter(sk);
e71a4783 762 break;
1da177e4 763
e71a4783
SH		 764 case SO_PASSSEC:
765 if (valbool)
766 set_bit(SOCK_PASSSEC, &sock->flags);
767 else
768 clear_bit(SOCK_PASSSEC, &sock->flags);
769 break;
4a19ec58
LAT		 770 case SO_MARK:
771 if (!capable(CAP_NET_ADMIN))
772 ret = -EPERM;
2a91525c 773 else
4a19ec58 774 sk->sk_mark = val;
4a19ec58 775 break;
877ce7c1 776
1da177e4
LT		 777 /* We implement the SO_SNDLOWAT etc to
778 not be settable (1003.1g 5.3) */
3b885787 779 case SO_RXQ_OVFL:
8083f0fc 780 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 781 break;
6e3e939f
JB		 782
783 case SO_WIFI_STATUS:
784 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
785 break;
786
ef64a54f
PE		 787 case SO_PEEK_OFF:
788 if (sock->ops->set_peek_off)
789 sock->ops->set_peek_off(sk, val);
790 else
791 ret = -EOPNOTSUPP;
792 break;
3bdc0eba
BG		 793
794 case SO_NOFCS:
795 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
796 break;
797
e71a4783
SH		 798 default:
799 ret = -ENOPROTOOPT;
800 break;
4ec93edb 801 }
1da177e4
LT		 802 release_sock(sk);
803 return ret;
804}
2a91525c 805EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 806
807
3f551f94
EB		 808void cred_to_ucred(struct pid *pid, const struct cred *cred,
809 struct ucred *ucred)
810{
811 ucred->pid = pid_vnr(pid);
812 ucred->uid = ucred->gid = -1;
813 if (cred) {
814 struct user_namespace *current_ns = current_user_ns();
815
76b6db01
EB		 816 ucred->uid = from_kuid(current_ns, cred->euid);
817 ucred->gid = from_kgid(current_ns, cred->egid);
3f551f94
EB		 818 }
819}
3924773a 820EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 821
1da177e4
LT		 822int sock_getsockopt(struct socket *sock, int level, int optname,
823 char __user *optval, int __user *optlen)
824{
825 struct sock *sk = sock->sk;
4ec93edb 826
e71a4783 827 union {
4ec93edb
YH		 828 int val;
829 struct linger ling;
1da177e4
LT		 830 struct timeval tm;
831 } v;
4ec93edb 832
4d0392be 833 int lv = sizeof(int);
1da177e4 834 int len;
4ec93edb 835
e71a4783 836 if (get_user(len, optlen))
4ec93edb 837 return -EFAULT;
e71a4783 838 if (len < 0)
1da177e4 839 return -EINVAL;
4ec93edb 840
50fee1de 841 memset(&v, 0, sizeof(v));
df0bca04 842
2a91525c 843 switch (optname) {
e71a4783
SH		 844 case SO_DEBUG:
845 v.val = sock_flag(sk, SOCK_DBG);
846 break;
847
848 case SO_DONTROUTE:
849 v.val = sock_flag(sk, SOCK_LOCALROUTE);
850 break;
851
852 case SO_BROADCAST:
1b23a5df 853 v.val = sock_flag(sk, SOCK_BROADCAST);
e71a4783
SH		 854 break;
855
856 case SO_SNDBUF:
857 v.val = sk->sk_sndbuf;
858 break;
859
860 case SO_RCVBUF:
861 v.val = sk->sk_rcvbuf;
862 break;
863
864 case SO_REUSEADDR:
865 v.val = sk->sk_reuse;
866 break;
867
868 case SO_KEEPALIVE:
1b23a5df 869 v.val = sock_flag(sk, SOCK_KEEPOPEN);
e71a4783
SH		 870 break;
871
872 case SO_TYPE:
873 v.val = sk->sk_type;
874 break;
875
49c794e9
JE		 876 case SO_PROTOCOL:
877 v.val = sk->sk_protocol;
878 break;
879
0d6038ee
JE		 880 case SO_DOMAIN:
881 v.val = sk->sk_family;
882 break;
883
e71a4783
SH		 884 case SO_ERROR:
885 v.val = -sock_error(sk);
2a91525c 886 if (v.val == 0)
e71a4783
SH		 887 v.val = xchg(&sk->sk_err_soft, 0);
888 break;
889
890 case SO_OOBINLINE:
1b23a5df 891 v.val = sock_flag(sk, SOCK_URGINLINE);
e71a4783
SH		 892 break;
893
894 case SO_NO_CHECK:
895 v.val = sk->sk_no_check;
896 break;
897
898 case SO_PRIORITY:
899 v.val = sk->sk_priority;
900 break;
901
902 case SO_LINGER:
903 lv = sizeof(v.ling);
1b23a5df 904 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
e71a4783
SH		 905 v.ling.l_linger = sk->sk_lingertime / HZ;
906 break;
907
908 case SO_BSDCOMPAT:
909 sock_warn_obsolete_bsdism("getsockopt");
910 break;
911
912 case SO_TIMESTAMP:
92f37fd2
ED		 913 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
914 !sock_flag(sk, SOCK_RCVTSTAMPNS);
915 break;
916
917 case SO_TIMESTAMPNS:
918 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 919 break;
920
20d49473
PO		 921 case SO_TIMESTAMPING:
922 v.val = 0;
923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
924 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
925 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
926 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
928 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
929 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
930 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
931 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
932 v.val |= SOF_TIMESTAMPING_SOFTWARE;
933 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
934 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
935 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
936 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
937 break;
938
e71a4783 939 case SO_RCVTIMEO:
2a91525c 940 lv = sizeof(struct timeval);
e71a4783
SH		 941 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
942 v.tm.tv_sec = 0;
943 v.tm.tv_usec = 0;
944 } else {
945 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
946 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
947 }
948 break;
949
950 case SO_SNDTIMEO:
2a91525c 951 lv = sizeof(struct timeval);
e71a4783
SH		 952 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
953 v.tm.tv_sec = 0;
954 v.tm.tv_usec = 0;
955 } else {
956 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
957 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
958 }
959 break;
1da177e4 960
e71a4783
SH		 961 case SO_RCVLOWAT:
962 v.val = sk->sk_rcvlowat;
963 break;
1da177e4 964
e71a4783 965 case SO_SNDLOWAT:
2a91525c 966 v.val = 1;
e71a4783 967 break;
1da177e4 968
e71a4783 969 case SO_PASSCRED:
82981930 970 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
e71a4783 971 break;
1da177e4 972
e71a4783 973 case SO_PEERCRED:
109f6e39
EB		 974 {
975 struct ucred peercred;
976 if (len > sizeof(peercred))
977 len = sizeof(peercred);
978 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
979 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 980 return -EFAULT;
981 goto lenout;
109f6e39 982 }
1da177e4 983
e71a4783
SH		 984 case SO_PEERNAME:
985 {
986 char address[128];
987
988 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
989 return -ENOTCONN;
990 if (lv < len)
991 return -EINVAL;
992 if (copy_to_user(optval, address, len))
993 return -EFAULT;
994 goto lenout;
995 }
1da177e4 996
e71a4783
SH		 997 /* Dubious BSD thing... Probably nobody even uses it, but
998 * the UNIX standard wants it for whatever reason... -DaveM
999 */
1000 case SO_ACCEPTCONN:
1001 v.val = sk->sk_state == TCP_LISTEN;
1002 break;
1da177e4 1003
e71a4783 1004 case SO_PASSSEC:
82981930 1005 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
e71a4783 1006 break;
877ce7c1 1007
e71a4783
SH		 1008 case SO_PEERSEC:
1009 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1010
4a19ec58
LAT		 1011 case SO_MARK:
1012 v.val = sk->sk_mark;
1013 break;
1014
3b885787 1015 case SO_RXQ_OVFL:
1b23a5df 1016 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
3b885787
NH		 1017 break;
1018
6e3e939f 1019 case SO_WIFI_STATUS:
1b23a5df 1020 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
6e3e939f
JB		 1021 break;
1022
ef64a54f
PE		 1023 case SO_PEEK_OFF:
1024 if (!sock->ops->set_peek_off)
1025 return -EOPNOTSUPP;
1026
1027 v.val = sk->sk_peek_off;
1028 break;
bc2f7996 1029 case SO_NOFCS:
1b23a5df 1030 v.val = sock_flag(sk, SOCK_NOFCS);
bc2f7996 1031 break;
e71a4783
SH		 1032 default:
1033 return -ENOPROTOOPT;
1da177e4 1034 }
e71a4783 1035
1da177e4
LT		 1036 if (len > lv)
1037 len = lv;
1038 if (copy_to_user(optval, &v, len))
1039 return -EFAULT;
1040lenout:
4ec93edb
YH		 1041 if (put_user(len, optlen))
1042 return -EFAULT;
1043 return 0;
1da177e4
LT		 1044}
1045
a5b5bb9a
IM		 1046/*
1047 * Initialize an sk_lock.
1048 *
1049 * (We also register the sk_lock with the lock validator.)
1050 */
b6f99a21 1051static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1052{
ed07536e
PZ		 1053 sock_lock_init_class_and_name(sk,
1054 af_family_slock_key_strings[sk->sk_family],
1055 af_family_slock_keys + sk->sk_family,
1056 af_family_key_strings[sk->sk_family],
1057 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1058}
1059
4dc6dc71
ED		 1060/*
1061 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1062 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1063 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1064 */
f1a6c4da
PE		 1065static void sock_copy(struct sock *nsk, const struct sock *osk)
1066{
1067#ifdef CONFIG_SECURITY_NETWORK
1068 void *sptr = nsk->sk_security;
1069#endif
68835aba
ED		 1070 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1071
1072 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1073 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1074
f1a6c4da
PE		 1075#ifdef CONFIG_SECURITY_NETWORK
1076 nsk->sk_security = sptr;
1077 security_sk_clone(osk, nsk);
1078#endif
1079}
1080
fcbdf09d
OP		 1081/*
1082 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1083 * un-modified. Special care is taken when initializing object to zero.
1084 */
1085static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1086{
1087 if (offsetof(struct sock, sk_node.next) != 0)
1088 memset(sk, 0, offsetof(struct sock, sk_node.next));
1089 memset(&sk->sk_node.pprev, 0,
1090 size - offsetof(struct sock, sk_node.pprev));
1091}
1092
1093void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1094{
1095 unsigned long nulls1, nulls2;
1096
1097 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1098 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1099 if (nulls1 > nulls2)
1100 swap(nulls1, nulls2);
1101
1102 if (nulls1 != 0)
1103 memset((char *)sk, 0, nulls1);
1104 memset((char *)sk + nulls1 + sizeof(void *), 0,
1105 nulls2 - nulls1 - sizeof(void *));
1106 memset((char *)sk + nulls2 + sizeof(void *), 0,
1107 size - nulls2 - sizeof(void *));
1108}
1109EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1110
2e4afe7b
PE		 1111static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1112 int family)
c308c1b2
PE		 1113{
1114 struct sock *sk;
1115 struct kmem_cache *slab;
1116
1117 slab = prot->slab;
e912b114
ED		 1118 if (slab != NULL) {
1119 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1120 if (!sk)
1121 return sk;
1122 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1123 if (prot->clear_sk)
1124 prot->clear_sk(sk, prot->obj_size);
1125 else
1126 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1127 }
fcbdf09d 1128 } else
c308c1b2
PE		 1129 sk = kmalloc(prot->obj_size, priority);
1130
2e4afe7b 1131 if (sk != NULL) {
a98b65a3
VN		 1132 kmemcheck_annotate_bitfield(sk, flags);
1133
2e4afe7b
PE		 1134 if (security_sk_alloc(sk, family, priority))
1135 goto out_free;
1136
1137 if (!try_module_get(prot->owner))
1138 goto out_free_sec;
e022f0b4 1139 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1140 }
1141
c308c1b2 1142 return sk;
2e4afe7b
PE		 1143
1144out_free_sec:
1145 security_sk_free(sk);
1146out_free:
1147 if (slab != NULL)
1148 kmem_cache_free(slab, sk);
1149 else
1150 kfree(sk);
1151 return NULL;
c308c1b2
PE		 1152}
1153
1154static void sk_prot_free(struct proto *prot, struct sock *sk)
1155{
1156 struct kmem_cache *slab;
2e4afe7b 1157 struct module *owner;
c308c1b2 1158
2e4afe7b 1159 owner = prot->owner;
c308c1b2 1160 slab = prot->slab;
2e4afe7b
PE		 1161
1162 security_sk_free(sk);
c308c1b2
PE		 1163 if (slab != NULL)
1164 kmem_cache_free(slab, sk);
1165 else
1166 kfree(sk);
2e4afe7b 1167 module_put(owner);
c308c1b2
PE		 1168}
1169
f8451725
HX		 1170#ifdef CONFIG_CGROUPS
1171void sock_update_classid(struct sock *sk)
1172{
1144182a 1173 u32 classid;
f8451725 1174
1144182a
PM		 1175 rcu_read_lock(); /* doing current task, which cannot vanish. */
1176 classid = task_cls_classid(current);
1177 rcu_read_unlock();
f8451725
HX		 1178 if (classid && classid != sk->sk_classid)
1179 sk->sk_classid = classid;
1180}
82862742 1181EXPORT_SYMBOL(sock_update_classid);
5bc1421e
NH		 1182
1183void sock_update_netprioidx(struct sock *sk)
1184{
5bc1421e
NH		 1185 if (in_interrupt())
1186 return;
2b73bc65
NH		 1187
1188 sk->sk_cgrp_prioidx = task_netprioidx(current);
5bc1421e
NH		 1189}
1190EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1191#endif
1192
1da177e4
LT		 1193/**
1194 * sk_alloc - All socket objects are allocated here
c4ea43c5 1195 * @net: the applicable net namespace
4dc3b16b
PP		 1196 * @family: protocol family
1197 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1198 * @prot: struct proto associated with this new sock instance
1da177e4 1199 */
1b8d7ae4 1200struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1201 struct proto *prot)
1da177e4 1202{
c308c1b2 1203 struct sock *sk;
1da177e4 1204
154adbc8 1205 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1206 if (sk) {
154adbc8
PE		 1207 sk->sk_family = family;
1208 /*
1209 * See comment in struct sock definition to understand
1210 * why we need sk_prot_creator -acme
1211 */
1212 sk->sk_prot = sk->sk_prot_creator = prot;
1213 sock_lock_init(sk);
3b1e0a65 1214 sock_net_set(sk, get_net(net));
d66ee058 1215 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1216
1217 sock_update_classid(sk);
5bc1421e 1218 sock_update_netprioidx(sk);
1da177e4 1219 }
a79af59e 1220
2e4afe7b 1221 return sk;
1da177e4 1222}
2a91525c 1223EXPORT_SYMBOL(sk_alloc);
1da177e4 1224
2b85a34e 1225static void __sk_free(struct sock *sk)
1da177e4
LT		 1226{
1227 struct sk_filter *filter;
1da177e4
LT		 1228
1229 if (sk->sk_destruct)
1230 sk->sk_destruct(sk);
1231
a898def2
PM		 1232 filter = rcu_dereference_check(sk->sk_filter,
1233 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1234 if (filter) {
309dd5fc 1235 sk_filter_uncharge(sk, filter);
a9b3cd7f 1236 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1237 }
1238
08e29af3 1239 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1240
1241 if (atomic_read(&sk->sk_omem_alloc))
e005d193
JP		 1242 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1243 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1244
109f6e39
EB		 1245 if (sk->sk_peer_cred)
1246 put_cred(sk->sk_peer_cred);
1247 put_pid(sk->sk_peer_pid);
3b1e0a65 1248 put_net(sock_net(sk));
c308c1b2 1249 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1250}
2b85a34e
ED		 1251
1252void sk_free(struct sock *sk)
1253{
1254 /*
25985edc 1255 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1256 * some packets are still in some tx queue.
1257 * If not null, sock_wfree() will call __sk_free(sk) later
1258 */
1259 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1260 __sk_free(sk);
1261}
2a91525c 1262EXPORT_SYMBOL(sk_free);
1da177e4 1263
edf02087 1264/*
25985edc
LDM		 1265 * Last sock_put should drop reference to sk->sk_net. It has already
1266 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1267 * is not an option.
25985edc 1268 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1269 * destroy it in the context of init_net.
1270 */
1271void sk_release_kernel(struct sock *sk)
1272{
1273 if (sk == NULL || sk->sk_socket == NULL)
1274 return;
1275
1276 sock_hold(sk);
1277 sock_release(sk->sk_socket);
65a18ec5 1278 release_net(sock_net(sk));
3b1e0a65 1279 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1280 sock_put(sk);
1281}
45af1754 1282EXPORT_SYMBOL(sk_release_kernel);
edf02087 1283
475f1b52
SR		 1284static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1285{
1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1287 sock_update_memcg(newsk);
1288}
1289
e56c57d0
ED		 1290/**
1291 * sk_clone_lock - clone a socket, and lock its clone
1292 * @sk: the socket to clone
1293 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1294 *
1295 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1296 */
1297struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1298{
8fd1d178 1299 struct sock *newsk;
87d11ceb 1300
8fd1d178 1301 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1302 if (newsk != NULL) {
1303 struct sk_filter *filter;
1304
892c141e 1305 sock_copy(newsk, sk);
87d11ceb
ACM		 1306
1307 /* SANITY */
3b1e0a65 1308 get_net(sock_net(newsk));
87d11ceb
ACM		 1309 sk_node_init(&newsk->sk_node);
1310 sock_lock_init(newsk);
1311 bh_lock_sock(newsk);
fa438ccf 1312 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1313 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1314
1315 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1316 /*
1317 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1318 */
1319 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1320 atomic_set(&newsk->sk_omem_alloc, 0);
1321 skb_queue_head_init(&newsk->sk_receive_queue);
1322 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1323#ifdef CONFIG_NET_DMA
1324 skb_queue_head_init(&newsk->sk_async_wait_queue);
1325#endif
87d11ceb 1326
b6c6712a 1327 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1328 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1329 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1330 af_callback_keys + newsk->sk_family,
1331 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1332
1333 newsk->sk_dst_cache = NULL;
1334 newsk->sk_wmem_queued = 0;
1335 newsk->sk_forward_alloc = 0;
1336 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1337 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1338
1339 sock_reset_flag(newsk, SOCK_DONE);
1340 skb_queue_head_init(&newsk->sk_error_queue);
1341
0d7da9dd 1342 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1343 if (filter != NULL)
1344 sk_filter_charge(newsk, filter);
1345
1346 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1347 /* It is still raw copy of parent, so invalidate
1348 * destructor and make plain sk_free() */
1349 newsk->sk_destruct = NULL;
b0691c8e 1350 bh_unlock_sock(newsk);
87d11ceb
ACM		 1351 sk_free(newsk);
1352 newsk = NULL;
1353 goto out;
1354 }
1355
1356 newsk->sk_err = 0;
1357 newsk->sk_priority = 0;
4dc6dc71
ED		 1358 /*
1359 * Before updating sk_refcnt, we must commit prior changes to memory
1360 * (Documentation/RCU/rculist_nulls.txt for details)
1361 */
1362 smp_wmb();
87d11ceb
ACM		 1363 atomic_set(&newsk->sk_refcnt, 2);
1364
1365 /*
1366 * Increment the counter in the same struct proto as the master
1367 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1368 * is the same as sk->sk_prot->socks, as this field was copied
1369 * with memcpy).
1370 *
1371 * This _changes_ the previous behaviour, where
1372 * tcp_create_openreq_child always was incrementing the
1373 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1374 * to be taken into account in all callers. -acme
1375 */
1376 sk_refcnt_debug_inc(newsk);
972692e0 1377 sk_set_socket(newsk, NULL);
43815482 1378 newsk->sk_wq = NULL;
87d11ceb 1379
f3f511e1
GC		 1380 sk_update_clone(sk, newsk);
1381
87d11ceb 1382 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1383 sk_sockets_allocated_inc(newsk);
704da560 1384
08e29af3 1385 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1386 net_enable_timestamp();
87d11ceb
ACM		 1387 }
1388out:
1389 return newsk;
1390}
e56c57d0 1391EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1392
9958089a
AK		 1393void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1394{
1395 __sk_dst_set(sk, dst);
1396 sk->sk_route_caps = dst->dev->features;
1397 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1398 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1399 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1400 if (sk_can_gso(sk)) {
82cc1a7a 1401 if (dst->header_len) {
9958089a 1402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1403 } else {
9958089a 1404 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1405 sk->sk_gso_max_size = dst->dev->gso_max_size;
1406 }
9958089a
AK		 1407 }
1408}
1409EXPORT_SYMBOL_GPL(sk_setup_caps);
1410
1da177e4
LT		 1411void __init sk_init(void)
1412{
4481374c 1413 if (totalram_pages <= 4096) {
1da177e4
LT		 1414 sysctl_wmem_max = 32767;
1415 sysctl_rmem_max = 32767;
1416 sysctl_wmem_default = 32767;
1417 sysctl_rmem_default = 32767;
4481374c 1418 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1419 sysctl_wmem_max = 131071;
1420 sysctl_rmem_max = 131071;
1421 }
1422}
1423
1424/*
1425 * Simple resource managers for sockets.
1426 */
1427
1428
4ec93edb
YH		 1429/*
1430 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1431 */
1432void sock_wfree(struct sk_buff *skb)
1433{
1434 struct sock *sk = skb->sk;
d99927f4 1435 unsigned int len = skb->truesize;
1da177e4 1436
d99927f4
ED		 1437 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1438 /*
1439 * Keep a reference on sk_wmem_alloc, this will be released
1440 * after sk_write_space() call
1441 */
1442 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1443 sk->sk_write_space(sk);
d99927f4
ED		 1444 len = 1;
1445 }
2b85a34e 1446 /*
d99927f4
ED		 1447 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1448 * could not do because of in-flight packets
2b85a34e 1449 */
d99927f4 1450 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1451 __sk_free(sk);
1da177e4 1452}
2a91525c 1453EXPORT_SYMBOL(sock_wfree);
1da177e4 1454
4ec93edb
YH		 1455/*
1456 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1457 */
1458void sock_rfree(struct sk_buff *skb)
1459{
1460 struct sock *sk = skb->sk;
d361fd59 1461 unsigned int len = skb->truesize;
1da177e4 1462
d361fd59
ED		 1463 atomic_sub(len, &sk->sk_rmem_alloc);
1464 sk_mem_uncharge(sk, len);
1da177e4 1465}
2a91525c 1466EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1467
1468
1469int sock_i_uid(struct sock *sk)
1470{
1471 int uid;
1472
f064af1e 1473 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1474 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1475 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1476 return uid;
1477}
2a91525c 1478EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1479
1480unsigned long sock_i_ino(struct sock *sk)
1481{
1482 unsigned long ino;
1483
f064af1e 1484 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1485 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1486 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1487 return ino;
1488}
2a91525c 1489EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1490
1491/*
1492 * Allocate a skb from the socket's send buffer.
1493 */
86a76caf 1494struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1495 gfp_t priority)
1da177e4
LT		 1496{
1497 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1498 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1499 if (skb) {
1500 skb_set_owner_w(skb, sk);
1501 return skb;
1502 }
1503 }
1504 return NULL;
1505}
2a91525c 1506EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1507
1508/*
1509 * Allocate a skb from the socket's receive buffer.
4ec93edb 1510 */
86a76caf 1511struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1512 gfp_t priority)
1da177e4
LT		 1513{
1514 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1515 struct sk_buff *skb = alloc_skb(size, priority);
1516 if (skb) {
1517 skb_set_owner_r(skb, sk);
1518 return skb;
1519 }
1520 }
1521 return NULL;
1522}
1523
4ec93edb 1524/*
1da177e4 1525 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1526 */
dd0fc66f 1527void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4 1528{
95c96174 1529 if ((unsigned int)size <= sysctl_optmem_max &&
1da177e4
LT		 1530 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1531 void *mem;
1532 /* First do the add, to avoid the race if kmalloc
4ec93edb 1533 * might sleep.
1da177e4
LT		 1534 */
1535 atomic_add(size, &sk->sk_omem_alloc);
1536 mem = kmalloc(size, priority);
1537 if (mem)
1538 return mem;
1539 atomic_sub(size, &sk->sk_omem_alloc);
1540 }
1541 return NULL;
1542}
2a91525c 1543EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1544
1545/*
1546 * Free an option memory block.
1547 */
1548void sock_kfree_s(struct sock *sk, void *mem, int size)
1549{
1550 kfree(mem);
1551 atomic_sub(size, &sk->sk_omem_alloc);
1552}
2a91525c 1553EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1554
1555/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1556 I think, these locks should be removed for datagram sockets.
1557 */
2a91525c 1558static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1559{
1560 DEFINE_WAIT(wait);
1561
1562 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1563 for (;;) {
1564 if (!timeo)
1565 break;
1566 if (signal_pending(current))
1567 break;
1568 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1569 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1570 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1571 break;
1572 if (sk->sk_shutdown & SEND_SHUTDOWN)
1573 break;
1574 if (sk->sk_err)
1575 break;
1576 timeo = schedule_timeout(timeo);
1577 }
aa395145 1578 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1579 return timeo;
1580}
1581
1582
1583/*
1584 * Generic send/receive buffer handlers
1585 */
1586
4cc7f68d
HX		 1587struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1588 unsigned long data_len, int noblock,
1589 int *errcode)
1da177e4
LT		 1590{
1591 struct sk_buff *skb;
7d877f3b 1592 gfp_t gfp_mask;
1da177e4
LT		 1593 long timeo;
1594 int err;
cc9b17ad
JW		 1595 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1596
1597 err = -EMSGSIZE;
1598 if (npages > MAX_SKB_FRAGS)
1599 goto failure;
1da177e4
LT		 1600
1601 gfp_mask = sk->sk_allocation;
1602 if (gfp_mask & __GFP_WAIT)
1603 gfp_mask |= __GFP_REPEAT;
1604
1605 timeo = sock_sndtimeo(sk, noblock);
1606 while (1) {
1607 err = sock_error(sk);
1608 if (err != 0)
1609 goto failure;
1610
1611 err = -EPIPE;
1612 if (sk->sk_shutdown & SEND_SHUTDOWN)
1613 goto failure;
1614
1615 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1616 skb = alloc_skb(header_len, gfp_mask);
1da177e4 1617 if (skb) {
1da177e4
LT		 1618 int i;
1619
1620 /* No pages, we're done... */
1621 if (!data_len)
1622 break;
1623
1da177e4
LT		 1624 skb->truesize += data_len;
1625 skb_shinfo(skb)->nr_frags = npages;
1626 for (i = 0; i < npages; i++) {
1627 struct page *page;
1da177e4
LT		 1628
1629 page = alloc_pages(sk->sk_allocation, 0);
1630 if (!page) {
1631 err = -ENOBUFS;
1632 skb_shinfo(skb)->nr_frags = i;
1633 kfree_skb(skb);
1634 goto failure;
1635 }
1636
ea2ab693
IC		 1637 __skb_fill_page_desc(skb, i,
1638 page, 0,
1639 (data_len >= PAGE_SIZE ?
1640 PAGE_SIZE :
1641 data_len));
1da177e4
LT		 1642 data_len -= PAGE_SIZE;
1643 }
1644
1645 /* Full success... */
1646 break;
1647 }
1648 err = -ENOBUFS;
1649 goto failure;
1650 }
1651 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1652 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1653 err = -EAGAIN;
1654 if (!timeo)
1655 goto failure;
1656 if (signal_pending(current))
1657 goto interrupted;
1658 timeo = sock_wait_for_wmem(sk, timeo);
1659 }
1660
1661 skb_set_owner_w(skb, sk);
1662 return skb;
1663
1664interrupted:
1665 err = sock_intr_errno(timeo);
1666failure:
1667 *errcode = err;
1668 return NULL;
1669}
4cc7f68d 1670EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1671
4ec93edb 1672struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1673 int noblock, int *errcode)
1674{
1675 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1676}
2a91525c 1677EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1678
1679static void __lock_sock(struct sock *sk)
f39234d6
NK		 1680 __releases(&sk->sk_lock.slock)
1681 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1682{
1683 DEFINE_WAIT(wait);
1684
e71a4783 1685 for (;;) {
1da177e4
LT		 1686 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1687 TASK_UNINTERRUPTIBLE);
1688 spin_unlock_bh(&sk->sk_lock.slock);
1689 schedule();
1690 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1691 if (!sock_owned_by_user(sk))
1da177e4
LT		 1692 break;
1693 }
1694 finish_wait(&sk->sk_lock.wq, &wait);
1695}
1696
1697static void __release_sock(struct sock *sk)
f39234d6
NK		 1698 __releases(&sk->sk_lock.slock)
1699 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1700{
1701 struct sk_buff *skb = sk->sk_backlog.head;
1702
1703 do {
1704 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1705 bh_unlock_sock(sk);
1706
1707 do {
1708 struct sk_buff *next = skb->next;
1709
e4cbb02a 1710 prefetch(next);
7fee226a 1711 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1712 skb->next = NULL;
c57943a1 1713 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1714
1715 /*
1716 * We are in process context here with softirqs
1717 * disabled, use cond_resched_softirq() to preempt.
1718 * This is safe to do because we've taken the backlog
1719 * queue private:
1720 */
1721 cond_resched_softirq();
1722
1723 skb = next;
1724 } while (skb != NULL);
1725
1726 bh_lock_sock(sk);
e71a4783 1727 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1728
1729 /*
1730 * Doing the zeroing here guarantee we can not loop forever
1731 * while a wild producer attempts to flood us.
1732 */
1733 sk->sk_backlog.len = 0;
1da177e4
LT		 1734}
1735
1736/**
1737 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1738 * @sk: sock to wait on
1739 * @timeo: for how long
1da177e4
LT		 1740 *
1741 * Now socket state including sk->sk_err is changed only under lock,
1742 * hence we may omit checks after joining wait queue.
1743 * We check receive queue before schedule() only as optimization;
1744 * it is very likely that release_sock() added new data.
1745 */
1746int sk_wait_data(struct sock *sk, long *timeo)
1747{
1748 int rc;
1749 DEFINE_WAIT(wait);
1750
aa395145 1751 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1752 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1753 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1754 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1755 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1756 return rc;
1757}
1da177e4
LT		 1758EXPORT_SYMBOL(sk_wait_data);
1759
3ab224be
HA		 1760/**
1761 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1762 * @sk: socket
1763 * @size: memory size to allocate
1764 * @kind: allocation type
1765 *
1766 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1767 * rmem allocation. This function assumes that protocols which have
1768 * memory_pressure use sk_wmem_queued as write buffer accounting.
1769 */
1770int __sk_mem_schedule(struct sock *sk, int size, int kind)
1771{
1772 struct proto *prot = sk->sk_prot;
1773 int amt = sk_mem_pages(size);
8d987e5c 1774 long allocated;
e1aab161 1775 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1776
1777 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1778
e1aab161 1779 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1780
1781 /* Under limit. */
e1aab161
GC		 1782 if (parent_status == UNDER_LIMIT &&
1783 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1784 sk_leave_memory_pressure(sk);
3ab224be
HA		 1785 return 1;
1786 }
1787
e1aab161
GC		 1788 /* Under pressure. (we or our parents) */
1789 if ((parent_status > SOFT_LIMIT) ||
1790 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1791 sk_enter_memory_pressure(sk);
3ab224be 1792
e1aab161
GC		 1793 /* Over hard limit (we or our parents) */
1794 if ((parent_status == OVER_LIMIT) ||
1795 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1796 goto suppress_allocation;
1797
1798 /* guarantee minimum buffer size under pressure */
1799 if (kind == SK_MEM_RECV) {
1800 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1801 return 1;
180d8cd9 1802
3ab224be
HA		 1803 } else { /* SK_MEM_SEND */
1804 if (sk->sk_type == SOCK_STREAM) {
1805 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1806 return 1;
1807 } else if (atomic_read(&sk->sk_wmem_alloc) <
1808 prot->sysctl_wmem[0])
1809 return 1;
1810 }
1811
180d8cd9 1812 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1813 int alloc;
1814
180d8cd9 1815 if (!sk_under_memory_pressure(sk))
1748376b 1816 return 1;
180d8cd9
GC		 1817 alloc = sk_sockets_allocated_read_positive(sk);
1818 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1819 sk_mem_pages(sk->sk_wmem_queued +
1820 atomic_read(&sk->sk_rmem_alloc) +
1821 sk->sk_forward_alloc))
1822 return 1;
1823 }
1824
1825suppress_allocation:
1826
1827 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1828 sk_stream_moderate_sndbuf(sk);
1829
1830 /* Fail only if socket is _under_ its sndbuf.
1831 * In this case we cannot block, so that we have to fail.
1832 */
1833 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1834 return 1;
1835 }
1836
3847ce32
SM		 1837 trace_sock_exceed_buf_limit(sk, prot, allocated);
1838
3ab224be
HA		 1839 /* Alas. Undo changes. */
1840 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1841
0e90b31f 1842 sk_memory_allocated_sub(sk, amt);
180d8cd9 1843
3ab224be
HA		 1844 return 0;
1845}
3ab224be
HA		 1846EXPORT_SYMBOL(__sk_mem_schedule);
1847
1848/**
1849 * __sk_reclaim - reclaim memory_allocated
1850 * @sk: socket
1851 */
1852void __sk_mem_reclaim(struct sock *sk)
1853{
180d8cd9 1854 sk_memory_allocated_sub(sk,
0e90b31f 1855 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
3ab224be
HA		 1856 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1857
180d8cd9
GC		 1858 if (sk_under_memory_pressure(sk) &&
1859 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1860 sk_leave_memory_pressure(sk);
3ab224be 1861}
3ab224be
HA		 1862EXPORT_SYMBOL(__sk_mem_reclaim);
1863
1864
1da177e4
LT		 1865/*
1866 * Set of default routines for initialising struct proto_ops when
1867 * the protocol does not support a particular function. In certain
1868 * cases where it makes no sense for a protocol to have a "do nothing"
1869 * function, some default processing is provided.
1870 */
1871
1872int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1873{
1874 return -EOPNOTSUPP;
1875}
2a91525c 1876EXPORT_SYMBOL(sock_no_bind);
1da177e4 1877
4ec93edb 1878int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1879 int len, int flags)
1880{
1881 return -EOPNOTSUPP;
1882}
2a91525c 1883EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1884
1885int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1886{
1887 return -EOPNOTSUPP;
1888}
2a91525c 1889EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1890
1891int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1892{
1893 return -EOPNOTSUPP;
1894}
2a91525c 1895EXPORT_SYMBOL(sock_no_accept);
1da177e4 1896
4ec93edb 1897int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1898 int *len, int peer)
1899{
1900 return -EOPNOTSUPP;
1901}
2a91525c 1902EXPORT_SYMBOL(sock_no_getname);
1da177e4 1903
2a91525c 1904unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1905{
1906 return 0;
1907}
2a91525c 1908EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1909
1910int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1911{
1912 return -EOPNOTSUPP;
1913}
2a91525c 1914EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1915
1916int sock_no_listen(struct socket *sock, int backlog)
1917{
1918 return -EOPNOTSUPP;
1919}
2a91525c 1920EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1921
1922int sock_no_shutdown(struct socket *sock, int how)
1923{
1924 return -EOPNOTSUPP;
1925}
2a91525c 1926EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1927
1928int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1929 char __user *optval, unsigned int optlen)
1da177e4
LT		 1930{
1931 return -EOPNOTSUPP;
1932}
2a91525c 1933EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1934
1935int sock_no_getsockopt(struct socket *sock, int level, int optname,
1936 char __user *optval, int __user *optlen)
1937{
1938 return -EOPNOTSUPP;
1939}
2a91525c 1940EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1941
1942int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1943 size_t len)
1944{
1945 return -EOPNOTSUPP;
1946}
2a91525c 1947EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1948
1949int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1950 size_t len, int flags)
1951{
1952 return -EOPNOTSUPP;
1953}
2a91525c 1954EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1955
1956int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1957{
1958 /* Mirror missing mmap method error code */
1959 return -ENODEV;
1960}
2a91525c 1961EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1962
1963ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1964{
1965 ssize_t res;
1966 struct msghdr msg = {.msg_flags = flags};
1967 struct kvec iov;
1968 char *kaddr = kmap(page);
1969 iov.iov_base = kaddr + offset;
1970 iov.iov_len = size;
1971 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1972 kunmap(page);
1973 return res;
1974}
2a91525c 1975EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1976
1977/*
1978 * Default Socket Callbacks
1979 */
1980
1981static void sock_def_wakeup(struct sock *sk)
1982{
43815482
ED		 1983 struct socket_wq *wq;
1984
1985 rcu_read_lock();
1986 wq = rcu_dereference(sk->sk_wq);
1987 if (wq_has_sleeper(wq))
1988 wake_up_interruptible_all(&wq->wait);
1989 rcu_read_unlock();
1da177e4
LT		 1990}
1991
1992static void sock_def_error_report(struct sock *sk)
1993{
43815482
ED		 1994 struct socket_wq *wq;
1995
1996 rcu_read_lock();
1997 wq = rcu_dereference(sk->sk_wq);
1998 if (wq_has_sleeper(wq))
1999 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 2000 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 2001 rcu_read_unlock();
1da177e4
LT		 2002}
2003
2004static void sock_def_readable(struct sock *sk, int len)
2005{
43815482
ED		 2006 struct socket_wq *wq;
2007
2008 rcu_read_lock();
2009 wq = rcu_dereference(sk->sk_wq);
2010 if (wq_has_sleeper(wq))
2c6607c6 2011 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 2012 POLLRDNORM | POLLRDBAND);
8d8ad9d7 2013 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 2014 rcu_read_unlock();
1da177e4
LT		 2015}
2016
2017static void sock_def_write_space(struct sock *sk)
2018{
43815482
ED		 2019 struct socket_wq *wq;
2020
2021 rcu_read_lock();
1da177e4
LT		 2022
2023 /* Do not wake up a writer until he can make "significant"
2024 * progress. --DaveM
2025 */
e71a4783 2026 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2027 wq = rcu_dereference(sk->sk_wq);
2028 if (wq_has_sleeper(wq))
2029 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2030 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2031
2032 /* Should agree with poll, otherwise some programs break */
2033 if (sock_writeable(sk))
8d8ad9d7 2034 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2035 }
2036
43815482 2037 rcu_read_unlock();
1da177e4
LT		 2038}
2039
2040static void sock_def_destruct(struct sock *sk)
2041{
a51482bd 2042 kfree(sk->sk_protinfo);
1da177e4
LT		 2043}
2044
2045void sk_send_sigurg(struct sock *sk)
2046{
2047 if (sk->sk_socket && sk->sk_socket->file)
2048 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2049 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2050}
2a91525c 2051EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2052
2053void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2054 unsigned long expires)
2055{
2056 if (!mod_timer(timer, expires))
2057 sock_hold(sk);
2058}
1da177e4
LT		 2059EXPORT_SYMBOL(sk_reset_timer);
2060
2061void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2062{
2063 if (timer_pending(timer) && del_timer(timer))
2064 __sock_put(sk);
2065}
1da177e4
LT		 2066EXPORT_SYMBOL(sk_stop_timer);
2067
2068void sock_init_data(struct socket *sock, struct sock *sk)
2069{
2070 skb_queue_head_init(&sk->sk_receive_queue);
2071 skb_queue_head_init(&sk->sk_write_queue);
2072 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2073#ifdef CONFIG_NET_DMA
2074 skb_queue_head_init(&sk->sk_async_wait_queue);
2075#endif
1da177e4
LT		 2076
2077 sk->sk_send_head = NULL;
2078
2079 init_timer(&sk->sk_timer);
4ec93edb 2080
1da177e4
LT		 2081 sk->sk_allocation = GFP_KERNEL;
2082 sk->sk_rcvbuf = sysctl_rmem_default;
2083 sk->sk_sndbuf = sysctl_wmem_default;
2084 sk->sk_state = TCP_CLOSE;
972692e0 2085 sk_set_socket(sk, sock);
1da177e4
LT		 2086
2087 sock_set_flag(sk, SOCK_ZAPPED);
2088
e71a4783 2089 if (sock) {
1da177e4 2090 sk->sk_type = sock->type;
43815482 2091 sk->sk_wq = sock->wq;
1da177e4
LT		 2092 sock->sk = sk;
2093 } else
43815482 2094 sk->sk_wq = NULL;
1da177e4 2095
b6c6712a 2096 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2097 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2098 lockdep_set_class_and_name(&sk->sk_callback_lock,
2099 af_callback_keys + sk->sk_family,
2100 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2101
2102 sk->sk_state_change = sock_def_wakeup;
2103 sk->sk_data_ready = sock_def_readable;
2104 sk->sk_write_space = sock_def_write_space;
2105 sk->sk_error_report = sock_def_error_report;
2106 sk->sk_destruct = sock_def_destruct;
2107
2108 sk->sk_sndmsg_page = NULL;
2109 sk->sk_sndmsg_off = 0;
ef64a54f 2110 sk->sk_peek_off = -1;
1da177e4 2111
109f6e39
EB		 2112 sk->sk_peer_pid = NULL;
2113 sk->sk_peer_cred = NULL;
1da177e4
LT		 2114 sk->sk_write_pending = 0;
2115 sk->sk_rcvlowat = 1;
2116 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2117 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2118
f37f0afb 2119 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2120
4dc6dc71
ED		 2121 /*
2122 * Before updating sk_refcnt, we must commit prior changes to memory
2123 * (Documentation/RCU/rculist_nulls.txt for details)
2124 */
2125 smp_wmb();
1da177e4 2126 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2127 atomic_set(&sk->sk_drops, 0);
1da177e4 2128}
2a91525c 2129EXPORT_SYMBOL(sock_init_data);
1da177e4 2130
b5606c2d 2131void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2132{
2133 might_sleep();
a5b5bb9a 2134 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2135 if (sk->sk_lock.owned)
1da177e4 2136 __lock_sock(sk);
d2e9117c 2137 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2138 spin_unlock(&sk->sk_lock.slock);
2139 /*
2140 * The sk_lock has mutex_lock() semantics here:
2141 */
fcc70d5f 2142 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2143 local_bh_enable();
1da177e4 2144}
fcc70d5f 2145EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2146
b5606c2d 2147void release_sock(struct sock *sk)
1da177e4 2148{
a5b5bb9a
IM		 2149 /*
2150 * The sk_lock has mutex_unlock() semantics:
2151 */
2152 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2153
2154 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2155 if (sk->sk_backlog.tail)
2156 __release_sock(sk);
d2e9117c 2157 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2158 if (waitqueue_active(&sk->sk_lock.wq))
2159 wake_up(&sk->sk_lock.wq);
2160 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2161}
2162EXPORT_SYMBOL(release_sock);
2163
8a74ad60
ED		 2164/**
2165 * lock_sock_fast - fast version of lock_sock
2166 * @sk: socket
2167 *
2168 * This version should be used for very small section, where process wont block
2169 * return false if fast path is taken
2170 * sk_lock.slock locked, owned = 0, BH disabled
2171 * return true if slow path is taken
2172 * sk_lock.slock unlocked, owned = 1, BH enabled
2173 */
2174bool lock_sock_fast(struct sock *sk)
2175{
2176 might_sleep();
2177 spin_lock_bh(&sk->sk_lock.slock);
2178
2179 if (!sk->sk_lock.owned)
2180 /*
2181 * Note : We must disable BH
2182 */
2183 return false;
2184
2185 __lock_sock(sk);
2186 sk->sk_lock.owned = 1;
2187 spin_unlock(&sk->sk_lock.slock);
2188 /*
2189 * The sk_lock has mutex_lock() semantics here:
2190 */
2191 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2192 local_bh_enable();
2193 return true;
2194}
2195EXPORT_SYMBOL(lock_sock_fast);
2196
1da177e4 2197int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2198{
b7aa0bf7 2199 struct timeval tv;
1da177e4 2200 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2201 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2202 tv = ktime_to_timeval(sk->sk_stamp);
2203 if (tv.tv_sec == -1)
1da177e4 2204 return -ENOENT;
b7aa0bf7
ED		 2205 if (tv.tv_sec == 0) {
2206 sk->sk_stamp = ktime_get_real();
2207 tv = ktime_to_timeval(sk->sk_stamp);
2208 }
2209 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2210}
1da177e4
LT		 2211EXPORT_SYMBOL(sock_get_timestamp);
2212
ae40eb1e
ED		 2213int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2214{
2215 struct timespec ts;
2216 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2217 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2218 ts = ktime_to_timespec(sk->sk_stamp);
2219 if (ts.tv_sec == -1)
2220 return -ENOENT;
2221 if (ts.tv_sec == 0) {
2222 sk->sk_stamp = ktime_get_real();
2223 ts = ktime_to_timespec(sk->sk_stamp);
2224 }
2225 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2226}
2227EXPORT_SYMBOL(sock_get_timestampns);
2228
20d49473 2229void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2230{
20d49473 2231 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2232 unsigned long previous_flags = sk->sk_flags;
2233
20d49473
PO		 2234 sock_set_flag(sk, flag);
2235 /*
2236 * we just set one of the two flags which require net
2237 * time stamping, but time stamping might have been on
2238 * already because of the other one
2239 */
08e29af3 2240 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2241 net_enable_timestamp();
1da177e4
LT		 2242 }
2243}
1da177e4
LT		 2244
2245/*
2246 * Get a socket option on an socket.
2247 *
2248 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2249 * asynchronous errors should be reported by getsockopt. We assume
2250 * this means if you specify SO_ERROR (otherwise whats the point of it).
2251 */
2252int sock_common_getsockopt(struct socket *sock, int level, int optname,
2253 char __user *optval, int __user *optlen)
2254{
2255 struct sock *sk = sock->sk;
2256
2257 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2258}
1da177e4
LT		 2259EXPORT_SYMBOL(sock_common_getsockopt);
2260
3fdadf7d 2261#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2262int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2263 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2264{
2265 struct sock *sk = sock->sk;
2266
1e51f951 2267 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2268 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2269 optval, optlen);
3fdadf7d
DM		 2270 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2271}
2272EXPORT_SYMBOL(compat_sock_common_getsockopt);
2273#endif
2274
1da177e4
LT		 2275int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2276 struct msghdr *msg, size_t size, int flags)
2277{
2278 struct sock *sk = sock->sk;
2279 int addr_len = 0;
2280 int err;
2281
2282 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2283 flags & ~MSG_DONTWAIT, &addr_len);
2284 if (err >= 0)
2285 msg->msg_namelen = addr_len;
2286 return err;
2287}
1da177e4
LT		 2288EXPORT_SYMBOL(sock_common_recvmsg);
2289
2290/*
2291 * Set socket options on an inet socket.
2292 */
2293int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2294 char __user *optval, unsigned int optlen)
1da177e4
LT		 2295{
2296 struct sock *sk = sock->sk;
2297
2298 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2299}
1da177e4
LT		 2300EXPORT_SYMBOL(sock_common_setsockopt);
2301
3fdadf7d 2302#ifdef CONFIG_COMPAT
543d9cfe 2303int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2304 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2305{
2306 struct sock *sk = sock->sk;
2307
543d9cfe
ACM		 2308 if (sk->sk_prot->compat_setsockopt != NULL)
2309 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2310 optval, optlen);
3fdadf7d
DM		 2311 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2312}
2313EXPORT_SYMBOL(compat_sock_common_setsockopt);
2314#endif
2315
1da177e4
LT		 2316void sk_common_release(struct sock *sk)
2317{
2318 if (sk->sk_prot->destroy)
2319 sk->sk_prot->destroy(sk);
2320
2321 /*
2322 * Observation: when sock_common_release is called, processes have
2323 * no access to socket. But net still has.
2324 * Step one, detach it from networking:
2325 *
2326 * A. Remove from hash tables.
2327 */
2328
2329 sk->sk_prot->unhash(sk);
2330
2331 /*
2332 * In this point socket cannot receive new packets, but it is possible
2333 * that some packets are in flight because some CPU runs receiver and
2334 * did hash table lookup before we unhashed socket. They will achieve
2335 * receive queue and will be purged by socket destructor.
2336 *
2337 * Also we still have packets pending on receive queue and probably,
2338 * our own packets waiting in device queues. sock_destroy will drain
2339 * receive queue, but transmitted packets will delay socket destruction
2340 * until the last reference will be released.
2341 */
2342
2343 sock_orphan(sk);
2344
2345 xfrm_sk_free_policy(sk);
2346
e6848976 2347 sk_refcnt_debug_release(sk);
1da177e4
LT		 2348 sock_put(sk);
2349}
1da177e4
LT		 2350EXPORT_SYMBOL(sk_common_release);
2351
13ff3d6f
PE		 2352#ifdef CONFIG_PROC_FS
2353#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2354struct prot_inuse {
2355 int val[PROTO_INUSE_NR];
2356};
13ff3d6f
PE		 2357
2358static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2359
2360#ifdef CONFIG_NET_NS
2361void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2362{
d6d9ca0f 2363 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2364}
2365EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2366
2367int sock_prot_inuse_get(struct net *net, struct proto *prot)
2368{
2369 int cpu, idx = prot->inuse_idx;
2370 int res = 0;
2371
2372 for_each_possible_cpu(cpu)
2373 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2374
2375 return res >= 0 ? res : 0;
2376}
2377EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2378
2c8c1e72 2379static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2380{
2381 net->core.inuse = alloc_percpu(struct prot_inuse);
2382 return net->core.inuse ? 0 : -ENOMEM;
2383}
2384
2c8c1e72 2385static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2386{
2387 free_percpu(net->core.inuse);
2388}
2389
2390static struct pernet_operations net_inuse_ops = {
2391 .init = sock_inuse_init_net,
2392 .exit = sock_inuse_exit_net,
2393};
2394
2395static __init int net_inuse_init(void)
2396{
2397 if (register_pernet_subsys(&net_inuse_ops))
2398 panic("Cannot initialize net inuse counters");
2399
2400 return 0;
2401}
2402
2403core_initcall(net_inuse_init);
2404#else
1338d466
PE		 2405static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2406
c29a0bc4 2407void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2408{
d6d9ca0f 2409 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2410}
2411EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2412
c29a0bc4 2413int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2414{
2415 int cpu, idx = prot->inuse_idx;
2416 int res = 0;
2417
2418 for_each_possible_cpu(cpu)
2419 res += per_cpu(prot_inuse, cpu).val[idx];
2420
2421 return res >= 0 ? res : 0;
2422}
2423EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2424#endif
13ff3d6f
PE		 2425
2426static void assign_proto_idx(struct proto *prot)
2427{
2428 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2429
2430 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
e005d193 2431 pr_err("PROTO_INUSE_NR exhausted\n");
13ff3d6f
PE		 2432 return;
2433 }
2434
2435 set_bit(prot->inuse_idx, proto_inuse_idx);
2436}
2437
2438static void release_proto_idx(struct proto *prot)
2439{
2440 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2441 clear_bit(prot->inuse_idx, proto_inuse_idx);
2442}
2443#else
2444static inline void assign_proto_idx(struct proto *prot)
2445{
2446}
2447
2448static inline void release_proto_idx(struct proto *prot)
2449{
2450}
2451#endif
2452
b733c007
PE		 2453int proto_register(struct proto *prot, int alloc_slab)
2454{
1da177e4
LT		 2455 if (alloc_slab) {
2456 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2457 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2458 NULL);
1da177e4
LT		 2459
2460 if (prot->slab == NULL) {
e005d193
JP		 2461 pr_crit("%s: Can't create sock SLAB cache!\n",
2462 prot->name);
60e7663d 2463 goto out;
1da177e4 2464 }
2e6599cb
ACM		 2465
2466 if (prot->rsk_prot != NULL) {
faf23422 2467 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2468 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2469 goto out_free_sock_slab;
2470
7e56b5d6 2471 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2472 prot->rsk_prot->obj_size, 0,
20c2df83 2473 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2474
2475 if (prot->rsk_prot->slab == NULL) {
e005d193
JP		 2476 pr_crit("%s: Can't create request sock SLAB cache!\n",
2477 prot->name);
2e6599cb
ACM		 2478 goto out_free_request_sock_slab_name;
2479 }
2480 }
8feaf0c0 2481
6d6ee43e 2482 if (prot->twsk_prot != NULL) {
faf23422 2483 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2484
7e56b5d6 2485 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2486 goto out_free_request_sock_slab;
2487
6d6ee43e 2488 prot->twsk_prot->twsk_slab =
7e56b5d6 2489 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2490 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2491 0,
2492 SLAB_HWCACHE_ALIGN |
2493 prot->slab_flags,
20c2df83 2494 NULL);
6d6ee43e 2495 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2496 goto out_free_timewait_sock_slab_name;
2497 }
1da177e4
LT		 2498 }
2499
36b77a52 2500 mutex_lock(&proto_list_mutex);
1da177e4 2501 list_add(&prot->node, &proto_list);
13ff3d6f 2502 assign_proto_idx(prot);
36b77a52 2503 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2504 return 0;
2505
8feaf0c0 2506out_free_timewait_sock_slab_name:
7e56b5d6 2507 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2508out_free_request_sock_slab:
2509 if (prot->rsk_prot && prot->rsk_prot->slab) {
2510 kmem_cache_destroy(prot->rsk_prot->slab);
2511 prot->rsk_prot->slab = NULL;
2512 }
2e6599cb 2513out_free_request_sock_slab_name:
72150e9b
DC		 2514 if (prot->rsk_prot)
2515 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2516out_free_sock_slab:
2517 kmem_cache_destroy(prot->slab);
2518 prot->slab = NULL;
b733c007
PE		 2519out:
2520 return -ENOBUFS;
1da177e4 2521}
1da177e4
LT		 2522EXPORT_SYMBOL(proto_register);
2523
2524void proto_unregister(struct proto *prot)
2525{
36b77a52 2526 mutex_lock(&proto_list_mutex);
13ff3d6f 2527 release_proto_idx(prot);
0a3f4358 2528 list_del(&prot->node);
36b77a52 2529 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2530
2531 if (prot->slab != NULL) {
2532 kmem_cache_destroy(prot->slab);
2533 prot->slab = NULL;
2534 }
2535
2e6599cb 2536 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2537 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2538 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2539 prot->rsk_prot->slab = NULL;
2540 }
2541
6d6ee43e 2542 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2543 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2544 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2545 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2546 }
1da177e4 2547}
1da177e4
LT		 2548EXPORT_SYMBOL(proto_unregister);
2549
2550#ifdef CONFIG_PROC_FS
1da177e4 2551static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2552 __acquires(proto_list_mutex)
1da177e4 2553{
36b77a52 2554 mutex_lock(&proto_list_mutex);
60f0438a 2555 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2556}
2557
2558static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2559{
60f0438a 2560 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2561}
2562
2563static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2564 __releases(proto_list_mutex)
1da177e4 2565{
36b77a52 2566 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2567}
2568
2569static char proto_method_implemented(const void *method)
2570{
2571 return method == NULL ? 'n' : 'y';
2572}
180d8cd9
GC		 2573static long sock_prot_memory_allocated(struct proto *proto)
2574{
cb75a36c 2575 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
180d8cd9
GC		 2576}
2577
2578static char *sock_prot_memory_pressure(struct proto *proto)
2579{
2580 return proto->memory_pressure != NULL ?
2581 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2582}
1da177e4
LT		 2583
2584static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2585{
180d8cd9 2586
8d987e5c 2587 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2588 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2589 proto->name,
2590 proto->obj_size,
14e943db 2591 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2592 sock_prot_memory_allocated(proto),
2593 sock_prot_memory_pressure(proto),
1da177e4
LT		 2594 proto->max_header,
2595 proto->slab == NULL ? "no" : "yes",
2596 module_name(proto->owner),
2597 proto_method_implemented(proto->close),
2598 proto_method_implemented(proto->connect),
2599 proto_method_implemented(proto->disconnect),
2600 proto_method_implemented(proto->accept),
2601 proto_method_implemented(proto->ioctl),
2602 proto_method_implemented(proto->init),
2603 proto_method_implemented(proto->destroy),
2604 proto_method_implemented(proto->shutdown),
2605 proto_method_implemented(proto->setsockopt),
2606 proto_method_implemented(proto->getsockopt),
2607 proto_method_implemented(proto->sendmsg),
2608 proto_method_implemented(proto->recvmsg),
2609 proto_method_implemented(proto->sendpage),
2610 proto_method_implemented(proto->bind),
2611 proto_method_implemented(proto->backlog_rcv),
2612 proto_method_implemented(proto->hash),
2613 proto_method_implemented(proto->unhash),
2614 proto_method_implemented(proto->get_port),
2615 proto_method_implemented(proto->enter_memory_pressure));
2616}
2617
2618static int proto_seq_show(struct seq_file *seq, void *v)
2619{
60f0438a 2620 if (v == &proto_list)
1da177e4
LT		 2621 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2622 "protocol",
2623 "size",
2624 "sockets",
2625 "memory",
2626 "press",
2627 "maxhdr",
2628 "slab",
2629 "module",
2630 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2631 else
60f0438a 2632 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2633 return 0;
2634}
2635
f690808e 2636static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2637 .start = proto_seq_start,
2638 .next = proto_seq_next,
2639 .stop = proto_seq_stop,
2640 .show = proto_seq_show,
2641};
2642
2643static int proto_seq_open(struct inode *inode, struct file *file)
2644{
14e943db
ED		 2645 return seq_open_net(inode, file, &proto_seq_ops,
2646 sizeof(struct seq_net_private));
1da177e4
LT		 2647}
2648
9a32144e 2649static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2650 .owner = THIS_MODULE,
2651 .open = proto_seq_open,
2652 .read = seq_read,
2653 .llseek = seq_lseek,
14e943db
ED		 2654 .release = seq_release_net,
2655};
2656
2657static __net_init int proto_init_net(struct net *net)
2658{
2659 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2660 return -ENOMEM;
2661
2662 return 0;
2663}
2664
2665static __net_exit void proto_exit_net(struct net *net)
2666{
2667 proc_net_remove(net, "protocols");
2668}
2669
2670
2671static __net_initdata struct pernet_operations proto_net_ops = {
2672 .init = proto_init_net,
2673 .exit = proto_exit_net,
1da177e4
LT		 2674};
2675
2676static int __init proto_init(void)
2677{
14e943db 2678 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2679}
2680
2681subsys_initcall(proto_init);
2682
2683#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.81638 seconds and 5 git commands to generate.