]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/core/sock.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
8139cp: fix missing napi_gro_flush.
[mirror_ubuntu-bionic-kernel.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 1364
f3f511e1
GC		 1365 sk_update_clone(sk, newsk);
1366
87d11ceb 1367 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1368 sk_sockets_allocated_inc(newsk);
704da560 1369
08e29af3 1370 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1371 net_enable_timestamp();
87d11ceb
ACM		 1372 }
1373out:
1374 return newsk;
1375}
e56c57d0 1376EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1377
9958089a
AK		 1378void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1379{
1380 __sk_dst_set(sk, dst);
1381 sk->sk_route_caps = dst->dev->features;
1382 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1383 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1384 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1385 if (sk_can_gso(sk)) {
82cc1a7a 1386 if (dst->header_len) {
9958089a 1387 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1388 } else {
9958089a 1389 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1390 sk->sk_gso_max_size = dst->dev->gso_max_size;
1391 }
9958089a
AK		 1392 }
1393}
1394EXPORT_SYMBOL_GPL(sk_setup_caps);
1395
1da177e4
LT		 1396void __init sk_init(void)
1397{
4481374c 1398 if (totalram_pages <= 4096) {
1da177e4
LT		 1399 sysctl_wmem_max = 32767;
1400 sysctl_rmem_max = 32767;
1401 sysctl_wmem_default = 32767;
1402 sysctl_rmem_default = 32767;
4481374c 1403 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1404 sysctl_wmem_max = 131071;
1405 sysctl_rmem_max = 131071;
1406 }
1407}
1408
1409/*
1410 * Simple resource managers for sockets.
1411 */
1412
1413
4ec93edb
YH		 1414/*
1415 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1416 */
1417void sock_wfree(struct sk_buff *skb)
1418{
1419 struct sock *sk = skb->sk;
d99927f4 1420 unsigned int len = skb->truesize;
1da177e4 1421
d99927f4
ED		 1422 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1423 /*
1424 * Keep a reference on sk_wmem_alloc, this will be released
1425 * after sk_write_space() call
1426 */
1427 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1428 sk->sk_write_space(sk);
d99927f4
ED		 1429 len = 1;
1430 }
2b85a34e 1431 /*
d99927f4
ED		 1432 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1433 * could not do because of in-flight packets
2b85a34e 1434 */
d99927f4 1435 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1436 __sk_free(sk);
1da177e4 1437}
2a91525c 1438EXPORT_SYMBOL(sock_wfree);
1da177e4 1439
4ec93edb
YH		 1440/*
1441 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1442 */
1443void sock_rfree(struct sk_buff *skb)
1444{
1445 struct sock *sk = skb->sk;
d361fd59 1446 unsigned int len = skb->truesize;
1da177e4 1447
d361fd59
ED		 1448 atomic_sub(len, &sk->sk_rmem_alloc);
1449 sk_mem_uncharge(sk, len);
1da177e4 1450}
2a91525c 1451EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1452
1453
1454int sock_i_uid(struct sock *sk)
1455{
1456 int uid;
1457
f064af1e 1458 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1459 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1460 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1461 return uid;
1462}
2a91525c 1463EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1464
1465unsigned long sock_i_ino(struct sock *sk)
1466{
1467 unsigned long ino;
1468
f064af1e 1469 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1470 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1471 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1472 return ino;
1473}
2a91525c 1474EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1475
1476/*
1477 * Allocate a skb from the socket's send buffer.
1478 */
86a76caf 1479struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1480 gfp_t priority)
1da177e4
LT		 1481{
1482 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1483 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1484 if (skb) {
1485 skb_set_owner_w(skb, sk);
1486 return skb;
1487 }
1488 }
1489 return NULL;
1490}
2a91525c 1491EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1492
1493/*
1494 * Allocate a skb from the socket's receive buffer.
4ec93edb 1495 */
86a76caf 1496struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1497 gfp_t priority)
1da177e4
LT		 1498{
1499 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1500 struct sk_buff *skb = alloc_skb(size, priority);
1501 if (skb) {
1502 skb_set_owner_r(skb, sk);
1503 return skb;
1504 }
1505 }
1506 return NULL;
1507}
1508
4ec93edb 1509/*
1da177e4 1510 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1511 */
dd0fc66f 1512void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1513{
1514 if ((unsigned)size <= sysctl_optmem_max &&
1515 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1516 void *mem;
1517 /* First do the add, to avoid the race if kmalloc
4ec93edb 1518 * might sleep.
1da177e4
LT		 1519 */
1520 atomic_add(size, &sk->sk_omem_alloc);
1521 mem = kmalloc(size, priority);
1522 if (mem)
1523 return mem;
1524 atomic_sub(size, &sk->sk_omem_alloc);
1525 }
1526 return NULL;
1527}
2a91525c 1528EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1529
1530/*
1531 * Free an option memory block.
1532 */
1533void sock_kfree_s(struct sock *sk, void *mem, int size)
1534{
1535 kfree(mem);
1536 atomic_sub(size, &sk->sk_omem_alloc);
1537}
2a91525c 1538EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1539
1540/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1541 I think, these locks should be removed for datagram sockets.
1542 */
2a91525c 1543static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1544{
1545 DEFINE_WAIT(wait);
1546
1547 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1548 for (;;) {
1549 if (!timeo)
1550 break;
1551 if (signal_pending(current))
1552 break;
1553 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1554 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1555 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1556 break;
1557 if (sk->sk_shutdown & SEND_SHUTDOWN)
1558 break;
1559 if (sk->sk_err)
1560 break;
1561 timeo = schedule_timeout(timeo);
1562 }
aa395145 1563 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1564 return timeo;
1565}
1566
1567
1568/*
1569 * Generic send/receive buffer handlers
1570 */
1571
4cc7f68d
HX		 1572struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1573 unsigned long data_len, int noblock,
1574 int *errcode)
1da177e4
LT		 1575{
1576 struct sk_buff *skb;
7d877f3b 1577 gfp_t gfp_mask;
1da177e4
LT		 1578 long timeo;
1579 int err;
1580
1581 gfp_mask = sk->sk_allocation;
1582 if (gfp_mask & __GFP_WAIT)
1583 gfp_mask |= __GFP_REPEAT;
1584
1585 timeo = sock_sndtimeo(sk, noblock);
1586 while (1) {
1587 err = sock_error(sk);
1588 if (err != 0)
1589 goto failure;
1590
1591 err = -EPIPE;
1592 if (sk->sk_shutdown & SEND_SHUTDOWN)
1593 goto failure;
1594
1595 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1596 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1597 if (skb) {
1598 int npages;
1599 int i;
1600
1601 /* No pages, we're done... */
1602 if (!data_len)
1603 break;
1604
1605 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1606 skb->truesize += data_len;
1607 skb_shinfo(skb)->nr_frags = npages;
1608 for (i = 0; i < npages; i++) {
1609 struct page *page;
1da177e4
LT		 1610
1611 page = alloc_pages(sk->sk_allocation, 0);
1612 if (!page) {
1613 err = -ENOBUFS;
1614 skb_shinfo(skb)->nr_frags = i;
1615 kfree_skb(skb);
1616 goto failure;
1617 }
1618
ea2ab693
IC		 1619 __skb_fill_page_desc(skb, i,
1620 page, 0,
1621 (data_len >= PAGE_SIZE ?
1622 PAGE_SIZE :
1623 data_len));
1da177e4
LT		 1624 data_len -= PAGE_SIZE;
1625 }
1626
1627 /* Full success... */
1628 break;
1629 }
1630 err = -ENOBUFS;
1631 goto failure;
1632 }
1633 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1634 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1635 err = -EAGAIN;
1636 if (!timeo)
1637 goto failure;
1638 if (signal_pending(current))
1639 goto interrupted;
1640 timeo = sock_wait_for_wmem(sk, timeo);
1641 }
1642
1643 skb_set_owner_w(skb, sk);
1644 return skb;
1645
1646interrupted:
1647 err = sock_intr_errno(timeo);
1648failure:
1649 *errcode = err;
1650 return NULL;
1651}
4cc7f68d 1652EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1653
4ec93edb 1654struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1655 int noblock, int *errcode)
1656{
1657 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1658}
2a91525c 1659EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1660
1661static void __lock_sock(struct sock *sk)
f39234d6
NK		 1662 __releases(&sk->sk_lock.slock)
1663 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1664{
1665 DEFINE_WAIT(wait);
1666
e71a4783 1667 for (;;) {
1da177e4
LT		 1668 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1669 TASK_UNINTERRUPTIBLE);
1670 spin_unlock_bh(&sk->sk_lock.slock);
1671 schedule();
1672 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1673 if (!sock_owned_by_user(sk))
1da177e4
LT		 1674 break;
1675 }
1676 finish_wait(&sk->sk_lock.wq, &wait);
1677}
1678
1679static void __release_sock(struct sock *sk)
f39234d6
NK		 1680 __releases(&sk->sk_lock.slock)
1681 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1682{
1683 struct sk_buff *skb = sk->sk_backlog.head;
1684
1685 do {
1686 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1687 bh_unlock_sock(sk);
1688
1689 do {
1690 struct sk_buff *next = skb->next;
1691
7fee226a 1692 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1693 skb->next = NULL;
c57943a1 1694 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1695
1696 /*
1697 * We are in process context here with softirqs
1698 * disabled, use cond_resched_softirq() to preempt.
1699 * This is safe to do because we've taken the backlog
1700 * queue private:
1701 */
1702 cond_resched_softirq();
1703
1704 skb = next;
1705 } while (skb != NULL);
1706
1707 bh_lock_sock(sk);
e71a4783 1708 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1709
1710 /*
1711 * Doing the zeroing here guarantee we can not loop forever
1712 * while a wild producer attempts to flood us.
1713 */
1714 sk->sk_backlog.len = 0;
1da177e4
LT		 1715}
1716
1717/**
1718 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1719 * @sk: sock to wait on
1720 * @timeo: for how long
1da177e4
LT		 1721 *
1722 * Now socket state including sk->sk_err is changed only under lock,
1723 * hence we may omit checks after joining wait queue.
1724 * We check receive queue before schedule() only as optimization;
1725 * it is very likely that release_sock() added new data.
1726 */
1727int sk_wait_data(struct sock *sk, long *timeo)
1728{
1729 int rc;
1730 DEFINE_WAIT(wait);
1731
aa395145 1732 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1733 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1734 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1735 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1736 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1737 return rc;
1738}
1da177e4
LT		 1739EXPORT_SYMBOL(sk_wait_data);
1740
3ab224be
HA		 1741/**
1742 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1743 * @sk: socket
1744 * @size: memory size to allocate
1745 * @kind: allocation type
1746 *
1747 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1748 * rmem allocation. This function assumes that protocols which have
1749 * memory_pressure use sk_wmem_queued as write buffer accounting.
1750 */
1751int __sk_mem_schedule(struct sock *sk, int size, int kind)
1752{
1753 struct proto *prot = sk->sk_prot;
1754 int amt = sk_mem_pages(size);
8d987e5c 1755 long allocated;
e1aab161 1756 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1757
1758 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1759
e1aab161 1760 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1761
1762 /* Under limit. */
e1aab161
GC		 1763 if (parent_status == UNDER_LIMIT &&
1764 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1765 sk_leave_memory_pressure(sk);
3ab224be
HA		 1766 return 1;
1767 }
1768
e1aab161
GC		 1769 /* Under pressure. (we or our parents) */
1770 if ((parent_status > SOFT_LIMIT) ||
1771 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1772 sk_enter_memory_pressure(sk);
3ab224be 1773
e1aab161
GC		 1774 /* Over hard limit (we or our parents) */
1775 if ((parent_status == OVER_LIMIT) ||
1776 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1777 goto suppress_allocation;
1778
1779 /* guarantee minimum buffer size under pressure */
1780 if (kind == SK_MEM_RECV) {
1781 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1782 return 1;
180d8cd9 1783
3ab224be
HA		 1784 } else { /* SK_MEM_SEND */
1785 if (sk->sk_type == SOCK_STREAM) {
1786 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1787 return 1;
1788 } else if (atomic_read(&sk->sk_wmem_alloc) <
1789 prot->sysctl_wmem[0])
1790 return 1;
1791 }
1792
180d8cd9 1793 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1794 int alloc;
1795
180d8cd9 1796 if (!sk_under_memory_pressure(sk))
1748376b 1797 return 1;
180d8cd9
GC		 1798 alloc = sk_sockets_allocated_read_positive(sk);
1799 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1800 sk_mem_pages(sk->sk_wmem_queued +
1801 atomic_read(&sk->sk_rmem_alloc) +
1802 sk->sk_forward_alloc))
1803 return 1;
1804 }
1805
1806suppress_allocation:
1807
1808 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1809 sk_stream_moderate_sndbuf(sk);
1810
1811 /* Fail only if socket is _under_ its sndbuf.
1812 * In this case we cannot block, so that we have to fail.
1813 */
1814 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1815 return 1;
1816 }
1817
3847ce32
SM		 1818 trace_sock_exceed_buf_limit(sk, prot, allocated);
1819
3ab224be
HA		 1820 /* Alas. Undo changes. */
1821 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1822
e1aab161 1823 sk_memory_allocated_sub(sk, amt, parent_status);
180d8cd9 1824
3ab224be
HA		 1825 return 0;
1826}
3ab224be
HA		 1827EXPORT_SYMBOL(__sk_mem_schedule);
1828
1829/**
1830 * __sk_reclaim - reclaim memory_allocated
1831 * @sk: socket
1832 */
1833void __sk_mem_reclaim(struct sock *sk)
1834{
180d8cd9 1835 sk_memory_allocated_sub(sk,
e1aab161 1836 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT, 0);
3ab224be
HA		 1837 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1838
180d8cd9
GC		 1839 if (sk_under_memory_pressure(sk) &&
1840 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1841 sk_leave_memory_pressure(sk);
3ab224be 1842}
3ab224be
HA		 1843EXPORT_SYMBOL(__sk_mem_reclaim);
1844
1845
1da177e4
LT		 1846/*
1847 * Set of default routines for initialising struct proto_ops when
1848 * the protocol does not support a particular function. In certain
1849 * cases where it makes no sense for a protocol to have a "do nothing"
1850 * function, some default processing is provided.
1851 */
1852
1853int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1854{
1855 return -EOPNOTSUPP;
1856}
2a91525c 1857EXPORT_SYMBOL(sock_no_bind);
1da177e4 1858
4ec93edb 1859int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1860 int len, int flags)
1861{
1862 return -EOPNOTSUPP;
1863}
2a91525c 1864EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1865
1866int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1867{
1868 return -EOPNOTSUPP;
1869}
2a91525c 1870EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1871
1872int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1873{
1874 return -EOPNOTSUPP;
1875}
2a91525c 1876EXPORT_SYMBOL(sock_no_accept);
1da177e4 1877
4ec93edb 1878int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1879 int *len, int peer)
1880{
1881 return -EOPNOTSUPP;
1882}
2a91525c 1883EXPORT_SYMBOL(sock_no_getname);
1da177e4 1884
2a91525c 1885unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1886{
1887 return 0;
1888}
2a91525c 1889EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1890
1891int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1892{
1893 return -EOPNOTSUPP;
1894}
2a91525c 1895EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1896
1897int sock_no_listen(struct socket *sock, int backlog)
1898{
1899 return -EOPNOTSUPP;
1900}
2a91525c 1901EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1902
1903int sock_no_shutdown(struct socket *sock, int how)
1904{
1905 return -EOPNOTSUPP;
1906}
2a91525c 1907EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1908
1909int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1910 char __user *optval, unsigned int optlen)
1da177e4
LT		 1911{
1912 return -EOPNOTSUPP;
1913}
2a91525c 1914EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1915
1916int sock_no_getsockopt(struct socket *sock, int level, int optname,
1917 char __user *optval, int __user *optlen)
1918{
1919 return -EOPNOTSUPP;
1920}
2a91525c 1921EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1922
1923int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1924 size_t len)
1925{
1926 return -EOPNOTSUPP;
1927}
2a91525c 1928EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1929
1930int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1931 size_t len, int flags)
1932{
1933 return -EOPNOTSUPP;
1934}
2a91525c 1935EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1936
1937int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1938{
1939 /* Mirror missing mmap method error code */
1940 return -ENODEV;
1941}
2a91525c 1942EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1943
1944ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1945{
1946 ssize_t res;
1947 struct msghdr msg = {.msg_flags = flags};
1948 struct kvec iov;
1949 char *kaddr = kmap(page);
1950 iov.iov_base = kaddr + offset;
1951 iov.iov_len = size;
1952 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1953 kunmap(page);
1954 return res;
1955}
2a91525c 1956EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1957
1958/*
1959 * Default Socket Callbacks
1960 */
1961
1962static void sock_def_wakeup(struct sock *sk)
1963{
43815482
ED		 1964 struct socket_wq *wq;
1965
1966 rcu_read_lock();
1967 wq = rcu_dereference(sk->sk_wq);
1968 if (wq_has_sleeper(wq))
1969 wake_up_interruptible_all(&wq->wait);
1970 rcu_read_unlock();
1da177e4
LT		 1971}
1972
1973static void sock_def_error_report(struct sock *sk)
1974{
43815482
ED		 1975 struct socket_wq *wq;
1976
1977 rcu_read_lock();
1978 wq = rcu_dereference(sk->sk_wq);
1979 if (wq_has_sleeper(wq))
1980 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1981 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1982 rcu_read_unlock();
1da177e4
LT		 1983}
1984
1985static void sock_def_readable(struct sock *sk, int len)
1986{
43815482
ED		 1987 struct socket_wq *wq;
1988
1989 rcu_read_lock();
1990 wq = rcu_dereference(sk->sk_wq);
1991 if (wq_has_sleeper(wq))
2c6607c6 1992 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1993 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1994 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1995 rcu_read_unlock();
1da177e4
LT		 1996}
1997
1998static void sock_def_write_space(struct sock *sk)
1999{
43815482
ED		 2000 struct socket_wq *wq;
2001
2002 rcu_read_lock();
1da177e4
LT		 2003
2004 /* Do not wake up a writer until he can make "significant"
2005 * progress. --DaveM
2006 */
e71a4783 2007 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2008 wq = rcu_dereference(sk->sk_wq);
2009 if (wq_has_sleeper(wq))
2010 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2011 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2012
2013 /* Should agree with poll, otherwise some programs break */
2014 if (sock_writeable(sk))
8d8ad9d7 2015 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2016 }
2017
43815482 2018 rcu_read_unlock();
1da177e4
LT		 2019}
2020
2021static void sock_def_destruct(struct sock *sk)
2022{
a51482bd 2023 kfree(sk->sk_protinfo);
1da177e4
LT		 2024}
2025
2026void sk_send_sigurg(struct sock *sk)
2027{
2028 if (sk->sk_socket && sk->sk_socket->file)
2029 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2030 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2031}
2a91525c 2032EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2033
2034void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2035 unsigned long expires)
2036{
2037 if (!mod_timer(timer, expires))
2038 sock_hold(sk);
2039}
1da177e4
LT		 2040EXPORT_SYMBOL(sk_reset_timer);
2041
2042void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2043{
2044 if (timer_pending(timer) && del_timer(timer))
2045 __sock_put(sk);
2046}
1da177e4
LT		 2047EXPORT_SYMBOL(sk_stop_timer);
2048
2049void sock_init_data(struct socket *sock, struct sock *sk)
2050{
2051 skb_queue_head_init(&sk->sk_receive_queue);
2052 skb_queue_head_init(&sk->sk_write_queue);
2053 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2054#ifdef CONFIG_NET_DMA
2055 skb_queue_head_init(&sk->sk_async_wait_queue);
2056#endif
1da177e4
LT		 2057
2058 sk->sk_send_head = NULL;
2059
2060 init_timer(&sk->sk_timer);
4ec93edb 2061
1da177e4
LT		 2062 sk->sk_allocation = GFP_KERNEL;
2063 sk->sk_rcvbuf = sysctl_rmem_default;
2064 sk->sk_sndbuf = sysctl_wmem_default;
2065 sk->sk_state = TCP_CLOSE;
972692e0 2066 sk_set_socket(sk, sock);
1da177e4
LT		 2067
2068 sock_set_flag(sk, SOCK_ZAPPED);
2069
e71a4783 2070 if (sock) {
1da177e4 2071 sk->sk_type = sock->type;
43815482 2072 sk->sk_wq = sock->wq;
1da177e4
LT		 2073 sock->sk = sk;
2074 } else
43815482 2075 sk->sk_wq = NULL;
1da177e4 2076
b6c6712a 2077 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2078 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2079 lockdep_set_class_and_name(&sk->sk_callback_lock,
2080 af_callback_keys + sk->sk_family,
2081 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2082
2083 sk->sk_state_change = sock_def_wakeup;
2084 sk->sk_data_ready = sock_def_readable;
2085 sk->sk_write_space = sock_def_write_space;
2086 sk->sk_error_report = sock_def_error_report;
2087 sk->sk_destruct = sock_def_destruct;
2088
2089 sk->sk_sndmsg_page = NULL;
2090 sk->sk_sndmsg_off = 0;
2091
109f6e39
EB		 2092 sk->sk_peer_pid = NULL;
2093 sk->sk_peer_cred = NULL;
1da177e4
LT		 2094 sk->sk_write_pending = 0;
2095 sk->sk_rcvlowat = 1;
2096 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2097 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2098
f37f0afb 2099 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2100
4dc6dc71
ED		 2101 /*
2102 * Before updating sk_refcnt, we must commit prior changes to memory
2103 * (Documentation/RCU/rculist_nulls.txt for details)
2104 */
2105 smp_wmb();
1da177e4 2106 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2107 atomic_set(&sk->sk_drops, 0);
1da177e4 2108}
2a91525c 2109EXPORT_SYMBOL(sock_init_data);
1da177e4 2110
b5606c2d 2111void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2112{
2113 might_sleep();
a5b5bb9a 2114 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2115 if (sk->sk_lock.owned)
1da177e4 2116 __lock_sock(sk);
d2e9117c 2117 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2118 spin_unlock(&sk->sk_lock.slock);
2119 /*
2120 * The sk_lock has mutex_lock() semantics here:
2121 */
fcc70d5f 2122 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2123 local_bh_enable();
1da177e4 2124}
fcc70d5f 2125EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2126
b5606c2d 2127void release_sock(struct sock *sk)
1da177e4 2128{
a5b5bb9a
IM		 2129 /*
2130 * The sk_lock has mutex_unlock() semantics:
2131 */
2132 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2133
2134 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2135 if (sk->sk_backlog.tail)
2136 __release_sock(sk);
d2e9117c 2137 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2138 if (waitqueue_active(&sk->sk_lock.wq))
2139 wake_up(&sk->sk_lock.wq);
2140 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2141}
2142EXPORT_SYMBOL(release_sock);
2143
8a74ad60
ED		 2144/**
2145 * lock_sock_fast - fast version of lock_sock
2146 * @sk: socket
2147 *
2148 * This version should be used for very small section, where process wont block
2149 * return false if fast path is taken
2150 * sk_lock.slock locked, owned = 0, BH disabled
2151 * return true if slow path is taken
2152 * sk_lock.slock unlocked, owned = 1, BH enabled
2153 */
2154bool lock_sock_fast(struct sock *sk)
2155{
2156 might_sleep();
2157 spin_lock_bh(&sk->sk_lock.slock);
2158
2159 if (!sk->sk_lock.owned)
2160 /*
2161 * Note : We must disable BH
2162 */
2163 return false;
2164
2165 __lock_sock(sk);
2166 sk->sk_lock.owned = 1;
2167 spin_unlock(&sk->sk_lock.slock);
2168 /*
2169 * The sk_lock has mutex_lock() semantics here:
2170 */
2171 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2172 local_bh_enable();
2173 return true;
2174}
2175EXPORT_SYMBOL(lock_sock_fast);
2176
1da177e4 2177int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2178{
b7aa0bf7 2179 struct timeval tv;
1da177e4 2180 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2181 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2182 tv = ktime_to_timeval(sk->sk_stamp);
2183 if (tv.tv_sec == -1)
1da177e4 2184 return -ENOENT;
b7aa0bf7
ED		 2185 if (tv.tv_sec == 0) {
2186 sk->sk_stamp = ktime_get_real();
2187 tv = ktime_to_timeval(sk->sk_stamp);
2188 }
2189 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2190}
1da177e4
LT		 2191EXPORT_SYMBOL(sock_get_timestamp);
2192
ae40eb1e
ED		 2193int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2194{
2195 struct timespec ts;
2196 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2197 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2198 ts = ktime_to_timespec(sk->sk_stamp);
2199 if (ts.tv_sec == -1)
2200 return -ENOENT;
2201 if (ts.tv_sec == 0) {
2202 sk->sk_stamp = ktime_get_real();
2203 ts = ktime_to_timespec(sk->sk_stamp);
2204 }
2205 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2206}
2207EXPORT_SYMBOL(sock_get_timestampns);
2208
20d49473 2209void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2210{
20d49473 2211 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2212 unsigned long previous_flags = sk->sk_flags;
2213
20d49473
PO		 2214 sock_set_flag(sk, flag);
2215 /*
2216 * we just set one of the two flags which require net
2217 * time stamping, but time stamping might have been on
2218 * already because of the other one
2219 */
08e29af3 2220 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2221 net_enable_timestamp();
1da177e4
LT		 2222 }
2223}
1da177e4
LT		 2224
2225/*
2226 * Get a socket option on an socket.
2227 *
2228 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2229 * asynchronous errors should be reported by getsockopt. We assume
2230 * this means if you specify SO_ERROR (otherwise whats the point of it).
2231 */
2232int sock_common_getsockopt(struct socket *sock, int level, int optname,
2233 char __user *optval, int __user *optlen)
2234{
2235 struct sock *sk = sock->sk;
2236
2237 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2238}
1da177e4
LT		 2239EXPORT_SYMBOL(sock_common_getsockopt);
2240
3fdadf7d 2241#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2242int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2243 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2244{
2245 struct sock *sk = sock->sk;
2246
1e51f951 2247 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2248 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2249 optval, optlen);
3fdadf7d
DM		 2250 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2251}
2252EXPORT_SYMBOL(compat_sock_common_getsockopt);
2253#endif
2254
1da177e4
LT		 2255int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2256 struct msghdr *msg, size_t size, int flags)
2257{
2258 struct sock *sk = sock->sk;
2259 int addr_len = 0;
2260 int err;
2261
2262 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2263 flags & ~MSG_DONTWAIT, &addr_len);
2264 if (err >= 0)
2265 msg->msg_namelen = addr_len;
2266 return err;
2267}
1da177e4
LT		 2268EXPORT_SYMBOL(sock_common_recvmsg);
2269
2270/*
2271 * Set socket options on an inet socket.
2272 */
2273int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2274 char __user *optval, unsigned int optlen)
1da177e4
LT		 2275{
2276 struct sock *sk = sock->sk;
2277
2278 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2279}
1da177e4
LT		 2280EXPORT_SYMBOL(sock_common_setsockopt);
2281
3fdadf7d 2282#ifdef CONFIG_COMPAT
543d9cfe 2283int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2284 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2285{
2286 struct sock *sk = sock->sk;
2287
543d9cfe
ACM		 2288 if (sk->sk_prot->compat_setsockopt != NULL)
2289 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2290 optval, optlen);
3fdadf7d
DM		 2291 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2292}
2293EXPORT_SYMBOL(compat_sock_common_setsockopt);
2294#endif
2295
1da177e4
LT		 2296void sk_common_release(struct sock *sk)
2297{
2298 if (sk->sk_prot->destroy)
2299 sk->sk_prot->destroy(sk);
2300
2301 /*
2302 * Observation: when sock_common_release is called, processes have
2303 * no access to socket. But net still has.
2304 * Step one, detach it from networking:
2305 *
2306 * A. Remove from hash tables.
2307 */
2308
2309 sk->sk_prot->unhash(sk);
2310
2311 /*
2312 * In this point socket cannot receive new packets, but it is possible
2313 * that some packets are in flight because some CPU runs receiver and
2314 * did hash table lookup before we unhashed socket. They will achieve
2315 * receive queue and will be purged by socket destructor.
2316 *
2317 * Also we still have packets pending on receive queue and probably,
2318 * our own packets waiting in device queues. sock_destroy will drain
2319 * receive queue, but transmitted packets will delay socket destruction
2320 * until the last reference will be released.
2321 */
2322
2323 sock_orphan(sk);
2324
2325 xfrm_sk_free_policy(sk);
2326
e6848976 2327 sk_refcnt_debug_release(sk);
1da177e4
LT		 2328 sock_put(sk);
2329}
1da177e4
LT		 2330EXPORT_SYMBOL(sk_common_release);
2331
13ff3d6f
PE		 2332#ifdef CONFIG_PROC_FS
2333#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2334struct prot_inuse {
2335 int val[PROTO_INUSE_NR];
2336};
13ff3d6f
PE		 2337
2338static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2339
2340#ifdef CONFIG_NET_NS
2341void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2342{
d6d9ca0f 2343 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2344}
2345EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2346
2347int sock_prot_inuse_get(struct net *net, struct proto *prot)
2348{
2349 int cpu, idx = prot->inuse_idx;
2350 int res = 0;
2351
2352 for_each_possible_cpu(cpu)
2353 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2354
2355 return res >= 0 ? res : 0;
2356}
2357EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2358
2c8c1e72 2359static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2360{
2361 net->core.inuse = alloc_percpu(struct prot_inuse);
2362 return net->core.inuse ? 0 : -ENOMEM;
2363}
2364
2c8c1e72 2365static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2366{
2367 free_percpu(net->core.inuse);
2368}
2369
2370static struct pernet_operations net_inuse_ops = {
2371 .init = sock_inuse_init_net,
2372 .exit = sock_inuse_exit_net,
2373};
2374
2375static __init int net_inuse_init(void)
2376{
2377 if (register_pernet_subsys(&net_inuse_ops))
2378 panic("Cannot initialize net inuse counters");
2379
2380 return 0;
2381}
2382
2383core_initcall(net_inuse_init);
2384#else
1338d466
PE		 2385static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2386
c29a0bc4 2387void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2388{
d6d9ca0f 2389 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2390}
2391EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2392
c29a0bc4 2393int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2394{
2395 int cpu, idx = prot->inuse_idx;
2396 int res = 0;
2397
2398 for_each_possible_cpu(cpu)
2399 res += per_cpu(prot_inuse, cpu).val[idx];
2400
2401 return res >= 0 ? res : 0;
2402}
2403EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2404#endif
13ff3d6f
PE		 2405
2406static void assign_proto_idx(struct proto *prot)
2407{
2408 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2409
2410 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2411 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2412 return;
2413 }
2414
2415 set_bit(prot->inuse_idx, proto_inuse_idx);
2416}
2417
2418static void release_proto_idx(struct proto *prot)
2419{
2420 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2421 clear_bit(prot->inuse_idx, proto_inuse_idx);
2422}
2423#else
2424static inline void assign_proto_idx(struct proto *prot)
2425{
2426}
2427
2428static inline void release_proto_idx(struct proto *prot)
2429{
2430}
2431#endif
2432
b733c007
PE		 2433int proto_register(struct proto *prot, int alloc_slab)
2434{
1da177e4
LT		 2435 if (alloc_slab) {
2436 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2437 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2438 NULL);
1da177e4
LT		 2439
2440 if (prot->slab == NULL) {
2441 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2442 prot->name);
60e7663d 2443 goto out;
1da177e4 2444 }
2e6599cb
ACM		 2445
2446 if (prot->rsk_prot != NULL) {
faf23422 2447 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2448 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2449 goto out_free_sock_slab;
2450
7e56b5d6 2451 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2452 prot->rsk_prot->obj_size, 0,
20c2df83 2453 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2454
2455 if (prot->rsk_prot->slab == NULL) {
2456 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2457 prot->name);
2458 goto out_free_request_sock_slab_name;
2459 }
2460 }
8feaf0c0 2461
6d6ee43e 2462 if (prot->twsk_prot != NULL) {
faf23422 2463 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2464
7e56b5d6 2465 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2466 goto out_free_request_sock_slab;
2467
6d6ee43e 2468 prot->twsk_prot->twsk_slab =
7e56b5d6 2469 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2470 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2471 0,
2472 SLAB_HWCACHE_ALIGN |
2473 prot->slab_flags,
20c2df83 2474 NULL);
6d6ee43e 2475 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2476 goto out_free_timewait_sock_slab_name;
2477 }
1da177e4
LT		 2478 }
2479
36b77a52 2480 mutex_lock(&proto_list_mutex);
1da177e4 2481 list_add(&prot->node, &proto_list);
13ff3d6f 2482 assign_proto_idx(prot);
36b77a52 2483 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2484 return 0;
2485
8feaf0c0 2486out_free_timewait_sock_slab_name:
7e56b5d6 2487 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2488out_free_request_sock_slab:
2489 if (prot->rsk_prot && prot->rsk_prot->slab) {
2490 kmem_cache_destroy(prot->rsk_prot->slab);
2491 prot->rsk_prot->slab = NULL;
2492 }
2e6599cb 2493out_free_request_sock_slab_name:
72150e9b
DC		 2494 if (prot->rsk_prot)
2495 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2496out_free_sock_slab:
2497 kmem_cache_destroy(prot->slab);
2498 prot->slab = NULL;
b733c007
PE		 2499out:
2500 return -ENOBUFS;
1da177e4 2501}
1da177e4
LT		 2502EXPORT_SYMBOL(proto_register);
2503
2504void proto_unregister(struct proto *prot)
2505{
36b77a52 2506 mutex_lock(&proto_list_mutex);
13ff3d6f 2507 release_proto_idx(prot);
0a3f4358 2508 list_del(&prot->node);
36b77a52 2509 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2510
2511 if (prot->slab != NULL) {
2512 kmem_cache_destroy(prot->slab);
2513 prot->slab = NULL;
2514 }
2515
2e6599cb 2516 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2517 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2518 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2519 prot->rsk_prot->slab = NULL;
2520 }
2521
6d6ee43e 2522 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2523 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2524 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2525 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2526 }
1da177e4 2527}
1da177e4
LT		 2528EXPORT_SYMBOL(proto_unregister);
2529
2530#ifdef CONFIG_PROC_FS
1da177e4 2531static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2532 __acquires(proto_list_mutex)
1da177e4 2533{
36b77a52 2534 mutex_lock(&proto_list_mutex);
60f0438a 2535 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2536}
2537
2538static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2539{
60f0438a 2540 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2541}
2542
2543static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2544 __releases(proto_list_mutex)
1da177e4 2545{
36b77a52 2546 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2547}
2548
2549static char proto_method_implemented(const void *method)
2550{
2551 return method == NULL ? 'n' : 'y';
2552}
180d8cd9
GC		 2553static long sock_prot_memory_allocated(struct proto *proto)
2554{
2555 return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
2556}
2557
2558static char *sock_prot_memory_pressure(struct proto *proto)
2559{
2560 return proto->memory_pressure != NULL ?
2561 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2562}
1da177e4
LT		 2563
2564static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2565{
180d8cd9 2566
8d987e5c 2567 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2568 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2569 proto->name,
2570 proto->obj_size,
14e943db 2571 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2572 sock_prot_memory_allocated(proto),
2573 sock_prot_memory_pressure(proto),
1da177e4
LT		 2574 proto->max_header,
2575 proto->slab == NULL ? "no" : "yes",
2576 module_name(proto->owner),
2577 proto_method_implemented(proto->close),
2578 proto_method_implemented(proto->connect),
2579 proto_method_implemented(proto->disconnect),
2580 proto_method_implemented(proto->accept),
2581 proto_method_implemented(proto->ioctl),
2582 proto_method_implemented(proto->init),
2583 proto_method_implemented(proto->destroy),
2584 proto_method_implemented(proto->shutdown),
2585 proto_method_implemented(proto->setsockopt),
2586 proto_method_implemented(proto->getsockopt),
2587 proto_method_implemented(proto->sendmsg),
2588 proto_method_implemented(proto->recvmsg),
2589 proto_method_implemented(proto->sendpage),
2590 proto_method_implemented(proto->bind),
2591 proto_method_implemented(proto->backlog_rcv),
2592 proto_method_implemented(proto->hash),
2593 proto_method_implemented(proto->unhash),
2594 proto_method_implemented(proto->get_port),
2595 proto_method_implemented(proto->enter_memory_pressure));
2596}
2597
2598static int proto_seq_show(struct seq_file *seq, void *v)
2599{
60f0438a 2600 if (v == &proto_list)
1da177e4
LT		 2601 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2602 "protocol",
2603 "size",
2604 "sockets",
2605 "memory",
2606 "press",
2607 "maxhdr",
2608 "slab",
2609 "module",
2610 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2611 else
60f0438a 2612 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2613 return 0;
2614}
2615
f690808e 2616static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2617 .start = proto_seq_start,
2618 .next = proto_seq_next,
2619 .stop = proto_seq_stop,
2620 .show = proto_seq_show,
2621};
2622
2623static int proto_seq_open(struct inode *inode, struct file *file)
2624{
14e943db
ED		 2625 return seq_open_net(inode, file, &proto_seq_ops,
2626 sizeof(struct seq_net_private));
1da177e4
LT		 2627}
2628
9a32144e 2629static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2630 .owner = THIS_MODULE,
2631 .open = proto_seq_open,
2632 .read = seq_read,
2633 .llseek = seq_lseek,
14e943db
ED		 2634 .release = seq_release_net,
2635};
2636
2637static __net_init int proto_init_net(struct net *net)
2638{
2639 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2640 return -ENOMEM;
2641
2642 return 0;
2643}
2644
2645static __net_exit void proto_exit_net(struct net *net)
2646{
2647 proc_net_remove(net, "protocols");
2648}
2649
2650
2651static __net_initdata struct pernet_operations proto_net_ops = {
2652 .init = proto_init_net,
2653 .exit = proto_exit_net,
1da177e4
LT		 2654};
2655
2656static int __init proto_init(void)
2657{
14e943db 2658 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2659}
2660
2661subsys_initcall(proto_init);
2662
2663#endif /* PROC_FS */
ubuntu-bionic-kernel mirror