blob: cc9b666e58f6ae6c12cc947f05d9ac5af0f6a6f9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NET An implementation of the SOCKET network access protocol.
3 *
4 * Version: @(#)socket.c 1.1.93 18/02/95
5 *
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
Jesper Juhl02c30a82005-05-05 16:16:16 -07007 * Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
9 *
10 * Fixes:
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
12 * shutdown()
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
17 * top level.
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
22 * tty drivers).
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
25 * configurable.
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
34 * stuff.
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
40 * moment.
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
Stephen Hemminger89bddce2006-09-01 00:19:31 -070045 * Tigran Aivazian : Made listen(2) backlog sanity checks
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 * protocol-independent
47 *
48 *
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
53 *
54 *
55 * This module is effectively the top level interface to the BSD socket
Stephen Hemminger89bddce2006-09-01 00:19:31 -070056 * paradigm.
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 *
58 * Based upon Swansea University Computer Society NET3.039
59 */
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#include <linux/socket.h>
63#include <linux/file.h>
64#include <linux/net.h>
65#include <linux/interrupt.h>
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -070066#include <linux/thread_info.h>
Stephen Hemminger55737fd2006-09-01 00:23:39 -070067#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <linux/netdevice.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080071#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/wanrouter.h>
73#include <linux/if_bridge.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030074#include <linux/if_frad.h>
75#include <linux/if_vlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/init.h>
77#include <linux/poll.h>
78#include <linux/cache.h>
79#include <linux/module.h>
80#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <linux/mount.h>
82#include <linux/security.h>
83#include <linux/syscalls.h>
84#include <linux/compat.h>
85#include <linux/kmod.h>
David Woodhouse3ec3b2f2005-05-17 12:08:48 +010086#include <linux/audit.h>
Adrian Bunkd86b5e02006-01-21 00:46:55 +010087#include <linux/wireless.h>
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -070088#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90#include <asm/uaccess.h>
91#include <asm/unistd.h>
92
93#include <net/compat.h>
David S. Miller87de87d2008-06-03 09:14:03 -070094#include <net/wext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96#include <net/sock.h>
97#include <linux/netfilter.h>
98
99static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700100static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
101 unsigned long nr_segs, loff_t pos);
102static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
103 unsigned long nr_segs, loff_t pos);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700104static int sock_mmap(struct file *file, struct vm_area_struct *vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
106static int sock_close(struct inode *inode, struct file *file);
107static unsigned int sock_poll(struct file *file,
108 struct poll_table_struct *wait);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700109static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800110#ifdef CONFIG_COMPAT
111static long compat_sock_ioctl(struct file *file,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700112 unsigned int cmd, unsigned long arg);
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800113#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114static int sock_fasync(int fd, struct file *filp, int on);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115static ssize_t sock_sendpage(struct file *file, struct page *page,
116 int offset, size_t size, loff_t *ppos, int more);
Jens Axboe9c55e012007-11-06 23:30:13 -0800117static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
118 struct pipe_inode_info *pipe, size_t len,
119 unsigned int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121/*
122 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
123 * in the operation structures but are done directly via the socketcall() multiplexor.
124 */
125
Arjan van de Venda7071d2007-02-12 00:55:36 -0800126static const struct file_operations socket_file_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 .owner = THIS_MODULE,
128 .llseek = no_llseek,
129 .aio_read = sock_aio_read,
130 .aio_write = sock_aio_write,
131 .poll = sock_poll,
132 .unlocked_ioctl = sock_ioctl,
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800133#ifdef CONFIG_COMPAT
134 .compat_ioctl = compat_sock_ioctl,
135#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 .mmap = sock_mmap,
137 .open = sock_no_open, /* special open code to disallow open via /proc */
138 .release = sock_close,
139 .fasync = sock_fasync,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .sendpage = sock_sendpage,
141 .splice_write = generic_splice_sendpage,
Jens Axboe9c55e012007-11-06 23:30:13 -0800142 .splice_read = sock_splice_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143};
144
145/*
146 * The protocol list. Each protocol is registered in here.
147 */
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149static DEFINE_SPINLOCK(net_family_lock);
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -0700150static const struct net_proto_family *net_families[NPROTO] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152/*
153 * Statistics counters of the socket lists
154 */
155
156static DEFINE_PER_CPU(int, sockets_in_use) = 0;
157
158/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700159 * Support routines.
160 * Move socket addresses back and forth across the kernel/user
161 * divide and look after the messy bits.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 */
163
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700164#define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 16 for IP, 16 for IPX,
166 24 for IPv6,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700167 about 80 for AX.25
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 must be at least one bigger than
169 the AF_UNIX size (see net/unix/af_unix.c
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700170 :unix_mkname()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173/**
174 * move_addr_to_kernel - copy a socket address into kernel space
175 * @uaddr: Address in user space
176 * @kaddr: Address in kernel space
177 * @ulen: Length in user space
178 *
179 * The address is copied into kernel space. If the provided address is
180 * too long an error code of -EINVAL is returned. If the copy gives
181 * invalid addresses -EFAULT is returned. On a success 0 is returned.
182 */
183
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700184int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700186 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700188 if (ulen == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700190 if (copy_from_user(kaddr, uaddr, ulen))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +0100192 return audit_sockaddr(ulen, kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
195/**
196 * move_addr_to_user - copy an address to user space
197 * @kaddr: kernel space address
198 * @klen: length of address in kernel
199 * @uaddr: user space address
200 * @ulen: pointer to user length field
201 *
202 * The value pointed to by ulen on entry is the buffer length available.
203 * This is overwritten with the buffer space used. -EINVAL is returned
204 * if an overlong buffer is specified or a negative buffer size. -EFAULT
205 * is returned if either the buffer or the length field are not
206 * accessible.
207 * After copying the data up to the limit the user specifies, the true
208 * length of the data is written over the length limit the user
209 * specified. Zero is returned for a success.
210 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700211
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700212int move_addr_to_user(struct sockaddr *kaddr, int klen, void __user *uaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700213 int __user *ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 int err;
216 int len;
217
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700218 err = get_user(len, ulen);
219 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 return err;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700221 if (len > klen)
222 len = klen;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700223 if (len < 0 || len > sizeof(struct sockaddr_storage))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700225 if (len) {
Steve Grubbd6fe3942006-03-30 12:20:22 -0500226 if (audit_sockaddr(klen, kaddr))
227 return -ENOMEM;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700228 if (copy_to_user(uaddr, kaddr, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return -EFAULT;
230 }
231 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700232 * "fromlen shall refer to the value before truncation.."
233 * 1003.1g
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
235 return __put_user(klen, ulen);
236}
237
238#define SOCKFS_MAGIC 0x534F434B
239
Christoph Lametere18b8902006-12-06 20:33:20 -0800240static struct kmem_cache *sock_inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242static struct inode *sock_alloc_inode(struct super_block *sb)
243{
244 struct socket_alloc *ei;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700245
Christoph Lametere94b1762006-12-06 20:33:17 -0800246 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (!ei)
248 return NULL;
249 init_waitqueue_head(&ei->socket.wait);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 ei->socket.fasync_list = NULL;
252 ei->socket.state = SS_UNCONNECTED;
253 ei->socket.flags = 0;
254 ei->socket.ops = NULL;
255 ei->socket.sk = NULL;
256 ei->socket.file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 return &ei->vfs_inode;
259}
260
261static void sock_destroy_inode(struct inode *inode)
262{
263 kmem_cache_free(sock_inode_cachep,
264 container_of(inode, struct socket_alloc, vfs_inode));
265}
266
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700267static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700269 struct socket_alloc *ei = (struct socket_alloc *)foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Christoph Lametera35afb82007-05-16 22:10:57 -0700271 inode_init_once(&ei->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274static int init_inodecache(void)
275{
276 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700277 sizeof(struct socket_alloc),
278 0,
279 (SLAB_HWCACHE_ALIGN |
280 SLAB_RECLAIM_ACCOUNT |
281 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +0900282 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 if (sock_inode_cachep == NULL)
284 return -ENOMEM;
285 return 0;
286}
287
288static struct super_operations sockfs_ops = {
289 .alloc_inode = sock_alloc_inode,
290 .destroy_inode =sock_destroy_inode,
291 .statfs = simple_statfs,
292};
293
David Howells454e2392006-06-23 02:02:57 -0700294static int sockfs_get_sb(struct file_system_type *fs_type,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700295 int flags, const char *dev_name, void *data,
296 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
David Howells454e2392006-06-23 02:02:57 -0700298 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
299 mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
Eric Dumazetba899662005-08-26 12:05:31 -0700302static struct vfsmount *sock_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304static struct file_system_type sock_fs_type = {
305 .name = "sockfs",
306 .get_sb = sockfs_get_sb,
307 .kill_sb = kill_anon_super,
308};
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310static int sockfs_delete_dentry(struct dentry *dentry)
311{
Eric Dumazet304e61e2006-12-06 20:38:49 -0800312 /*
313 * At creation time, we pretended this dentry was hashed
314 * (by clearing DCACHE_UNHASHED bit in d_flags)
315 * At delete time, we restore the truth : not hashed.
316 * (so that dput() can proceed correctly)
317 */
318 dentry->d_flags |= DCACHE_UNHASHED;
319 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700321
322/*
323 * sockfs_dname() is called from d_path().
324 */
325static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
326{
327 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
328 dentry->d_inode->i_ino);
329}
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331static struct dentry_operations sockfs_dentry_operations = {
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700332 .d_delete = sockfs_delete_dentry,
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700333 .d_dname = sockfs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334};
335
336/*
337 * Obtains the first available file descriptor and sets it up for use.
338 *
David S. Miller39d8c1b2006-03-20 17:13:49 -0800339 * These functions create file structures and maps them to fd space
340 * of the current process. On success it returns file descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 * and file struct implicitly stored in sock->file.
342 * Note that another thread may close file descriptor before we return
343 * from this function. We use the fact that now we do not refer
344 * to socket after mapping. If one day we will need it, this
345 * function will increment ref. count on file by 1.
346 *
347 * In any case returned fd MAY BE not valid!
348 * This race condition is unavoidable
349 * with shared fd spaces, we cannot solve it inside kernel,
350 * but we take care of internal coherence yet.
351 */
352
Ulrich Dreppera677a032008-07-23 21:29:17 -0700353static int sock_alloc_fd(struct file **filep, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
355 int fd;
David S. Miller39d8c1b2006-03-20 17:13:49 -0800356
Ulrich Dreppera677a032008-07-23 21:29:17 -0700357 fd = get_unused_fd_flags(flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800358 if (likely(fd >= 0)) {
359 struct file *file = get_empty_filp();
360
361 *filep = file;
362 if (unlikely(!file)) {
363 put_unused_fd(fd);
364 return -ENFILE;
365 }
366 } else
367 *filep = NULL;
368 return fd;
369}
370
Ulrich Drepper77d27202008-07-23 21:29:35 -0700371static int sock_attach_fd(struct socket *sock, struct file *file, int flags)
David S. Miller39d8c1b2006-03-20 17:13:49 -0800372{
Dave Hansence8d2cd2007-10-16 23:31:13 -0700373 struct dentry *dentry;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700374 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Dave Hansence8d2cd2007-10-16 23:31:13 -0700376 dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
377 if (unlikely(!dentry))
David S. Miller39d8c1b2006-03-20 17:13:49 -0800378 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Dave Hansence8d2cd2007-10-16 23:31:13 -0700380 dentry->d_op = &sockfs_dentry_operations;
Eric Dumazet304e61e2006-12-06 20:38:49 -0800381 /*
382 * We dont want to push this dentry into global dentry hash table.
383 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
384 * This permits a working /proc/$pid/fd/XXX on sockets
385 */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700386 dentry->d_flags &= ~DCACHE_UNHASHED;
387 d_instantiate(dentry, SOCK_INODE(sock));
David S. Miller39d8c1b2006-03-20 17:13:49 -0800388
389 sock->file = file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700390 init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE,
391 &socket_file_ops);
392 SOCK_INODE(sock)->i_fop = &socket_file_ops;
Ulrich Drepper77d27202008-07-23 21:29:35 -0700393 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800394 file->f_pos = 0;
395 file->private_data = sock;
396
397 return 0;
398}
399
Ulrich Dreppera677a032008-07-23 21:29:17 -0700400int sock_map_fd(struct socket *sock, int flags)
David S. Miller39d8c1b2006-03-20 17:13:49 -0800401{
402 struct file *newfile;
Ulrich Dreppera677a032008-07-23 21:29:17 -0700403 int fd = sock_alloc_fd(&newfile, flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800404
405 if (likely(fd >= 0)) {
Ulrich Drepper77d27202008-07-23 21:29:35 -0700406 int err = sock_attach_fd(sock, newfile, flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800407
408 if (unlikely(err < 0)) {
409 put_filp(newfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 put_unused_fd(fd);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800411 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
David S. Miller39d8c1b2006-03-20 17:13:49 -0800413 fd_install(fd, newfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 return fd;
416}
417
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800418static struct socket *sock_from_file(struct file *file, int *err)
419{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800420 if (file->f_op == &socket_file_ops)
421 return file->private_data; /* set in sock_map_fd */
422
Eric Dumazet23bb80d2007-02-08 14:59:57 -0800423 *err = -ENOTSOCK;
424 return NULL;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800425}
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427/**
428 * sockfd_lookup - Go from a file number to its socket slot
429 * @fd: file handle
430 * @err: pointer to an error code return
431 *
432 * The file handle passed in is locked and the socket it is bound
433 * too is returned. If an error occurs the err pointer is overwritten
434 * with a negative errno code and NULL is returned. The function checks
435 * for both invalid handles and passing a handle which is not a socket.
436 *
437 * On a success the socket object pointer is returned.
438 */
439
440struct socket *sockfd_lookup(int fd, int *err)
441{
442 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 struct socket *sock;
444
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700445 file = fget(fd);
446 if (!file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 *err = -EBADF;
448 return NULL;
449 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700450
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800451 sock = sock_from_file(file, err);
452 if (!sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 fput(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return sock;
455}
456
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800457static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
458{
459 struct file *file;
460 struct socket *sock;
461
Hua Zhong36725582006-04-19 15:25:02 -0700462 *err = -EBADF;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800463 file = fget_light(fd, fput_needed);
464 if (file) {
465 sock = sock_from_file(file, err);
466 if (sock)
467 return sock;
468 fput_light(file, *fput_needed);
469 }
470 return NULL;
471}
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473/**
474 * sock_alloc - allocate a socket
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700475 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 * Allocate a new inode and socket object. The two are bound together
477 * and initialised. The socket is then returned. If we are out of inodes
478 * NULL is returned.
479 */
480
481static struct socket *sock_alloc(void)
482{
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700483 struct inode *inode;
484 struct socket *sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 inode = new_inode(sock_mnt->mnt_sb);
487 if (!inode)
488 return NULL;
489
490 sock = SOCKET_I(inode);
491
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700492 inode->i_mode = S_IFSOCK | S_IRWXUGO;
David Howells8192b0c2008-11-14 10:39:10 +1100493 inode->i_uid = current_fsuid();
494 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 get_cpu_var(sockets_in_use)++;
497 put_cpu_var(sockets_in_use);
498 return sock;
499}
500
501/*
502 * In theory you can't get an open on this inode, but /proc provides
503 * a back door. Remember to keep it shut otherwise you'll let the
504 * creepy crawlies in.
505 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
508{
509 return -ENXIO;
510}
511
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800512const struct file_operations bad_sock_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 .owner = THIS_MODULE,
514 .open = sock_no_open,
515};
516
517/**
518 * sock_release - close a socket
519 * @sock: socket to close
520 *
521 * The socket is released from the protocol stack if it has a release
522 * callback, and the inode is then released if the socket is bound to
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700523 * an inode not a file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526void sock_release(struct socket *sock)
527{
528 if (sock->ops) {
529 struct module *owner = sock->ops->owner;
530
531 sock->ops->release(sock);
532 sock->ops = NULL;
533 module_put(owner);
534 }
535
536 if (sock->fasync_list)
537 printk(KERN_ERR "sock_release: fasync list not empty!\n");
538
539 get_cpu_var(sockets_in_use)--;
540 put_cpu_var(sockets_in_use);
541 if (!sock->file) {
542 iput(SOCK_INODE(sock));
543 return;
544 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700545 sock->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700548static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 struct msghdr *msg, size_t size)
550{
551 struct sock_iocb *si = kiocb_to_siocb(iocb);
552 int err;
553
554 si->sock = sock;
555 si->scm = NULL;
556 si->msg = msg;
557 si->size = size;
558
559 err = security_socket_sendmsg(sock, msg, size);
560 if (err)
561 return err;
562
563 return sock->ops->sendmsg(iocb, sock, msg, size);
564}
565
566int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
567{
568 struct kiocb iocb;
569 struct sock_iocb siocb;
570 int ret;
571
572 init_sync_kiocb(&iocb, NULL);
573 iocb.private = &siocb;
574 ret = __sock_sendmsg(&iocb, sock, msg, size);
575 if (-EIOCBQUEUED == ret)
576 ret = wait_on_sync_kiocb(&iocb);
577 return ret;
578}
579
580int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
581 struct kvec *vec, size_t num, size_t size)
582{
583 mm_segment_t oldfs = get_fs();
584 int result;
585
586 set_fs(KERNEL_DS);
587 /*
588 * the following is safe, since for compiler definitions of kvec and
589 * iovec are identical, yielding the same in-core layout and alignment
590 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700591 msg->msg_iov = (struct iovec *)vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 msg->msg_iovlen = num;
593 result = sock_sendmsg(sock, msg, size);
594 set_fs(oldfs);
595 return result;
596}
597
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700598/*
599 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
600 */
601void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
602 struct sk_buff *skb)
603{
604 ktime_t kt = skb->tstamp;
605
606 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
607 struct timeval tv;
608 /* Race occurred between timestamp enabling and packet
609 receiving. Fill in the current time for now. */
610 if (kt.tv64 == 0)
611 kt = ktime_get_real();
612 skb->tstamp = kt;
613 tv = ktime_to_timeval(kt);
614 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, sizeof(tv), &tv);
615 } else {
616 struct timespec ts;
617 /* Race occurred between timestamp enabling and packet
618 receiving. Fill in the current time for now. */
619 if (kt.tv64 == 0)
620 kt = ktime_get_real();
621 skb->tstamp = kt;
622 ts = ktime_to_timespec(kt);
623 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, sizeof(ts), &ts);
624 }
625}
626
Arnaldo Carvalho de Melo7c81fd82007-03-10 00:39:35 -0300627EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
628
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700629static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 struct msghdr *msg, size_t size, int flags)
631{
632 int err;
633 struct sock_iocb *si = kiocb_to_siocb(iocb);
634
635 si->sock = sock;
636 si->scm = NULL;
637 si->msg = msg;
638 si->size = size;
639 si->flags = flags;
640
641 err = security_socket_recvmsg(sock, msg, size, flags);
642 if (err)
643 return err;
644
645 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
646}
647
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700648int sock_recvmsg(struct socket *sock, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 size_t size, int flags)
650{
651 struct kiocb iocb;
652 struct sock_iocb siocb;
653 int ret;
654
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700655 init_sync_kiocb(&iocb, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 iocb.private = &siocb;
657 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
658 if (-EIOCBQUEUED == ret)
659 ret = wait_on_sync_kiocb(&iocb);
660 return ret;
661}
662
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700663int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
664 struct kvec *vec, size_t num, size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 mm_segment_t oldfs = get_fs();
667 int result;
668
669 set_fs(KERNEL_DS);
670 /*
671 * the following is safe, since for compiler definitions of kvec and
672 * iovec are identical, yielding the same in-core layout and alignment
673 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700674 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 result = sock_recvmsg(sock, msg, size, flags);
676 set_fs(oldfs);
677 return result;
678}
679
680static void sock_aio_dtor(struct kiocb *iocb)
681{
682 kfree(iocb->private);
683}
684
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300685static ssize_t sock_sendpage(struct file *file, struct page *page,
686 int offset, size_t size, loff_t *ppos, int more)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
688 struct socket *sock;
689 int flags;
690
Eric Dumazetb69aee02005-09-06 14:42:45 -0700691 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
694 if (more)
695 flags |= MSG_MORE;
696
697 return sock->ops->sendpage(sock, page, offset, size, flags);
698}
699
Jens Axboe9c55e012007-11-06 23:30:13 -0800700static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
701 struct pipe_inode_info *pipe, size_t len,
702 unsigned int flags)
703{
704 struct socket *sock = file->private_data;
705
Rémi Denis-Courmont997b37d2008-02-15 02:35:45 -0800706 if (unlikely(!sock->ops->splice_read))
707 return -EINVAL;
708
Jens Axboe9c55e012007-11-06 23:30:13 -0800709 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
710}
711
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800712static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700713 struct sock_iocb *siocb)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800714{
715 if (!is_sync_kiocb(iocb)) {
716 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
717 if (!siocb)
718 return NULL;
719 iocb->ki_dtor = sock_aio_dtor;
720 }
721
722 siocb->kiocb = iocb;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800723 iocb->private = siocb;
724 return siocb;
725}
726
727static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -0700728 struct file *file, const struct iovec *iov,
729 unsigned long nr_segs)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800730{
731 struct socket *sock = file->private_data;
732 size_t size = 0;
733 int i;
734
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700735 for (i = 0; i < nr_segs; i++)
736 size += iov[i].iov_len;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800737
738 msg->msg_name = NULL;
739 msg->msg_namelen = 0;
740 msg->msg_control = NULL;
741 msg->msg_controllen = 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700742 msg->msg_iov = (struct iovec *)iov;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800743 msg->msg_iovlen = nr_segs;
744 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
745
746 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
747}
748
Badari Pulavarty027445c2006-09-30 23:28:46 -0700749static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
750 unsigned long nr_segs, loff_t pos)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800751{
752 struct sock_iocb siocb, *x;
753
754 if (pos != 0)
755 return -ESPIPE;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700756
757 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800758 return 0;
759
Badari Pulavarty027445c2006-09-30 23:28:46 -0700760
761 x = alloc_sock_iocb(iocb, &siocb);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800762 if (!x)
763 return -ENOMEM;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700764 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800765}
766
767static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -0700768 struct file *file, const struct iovec *iov,
769 unsigned long nr_segs)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800770{
771 struct socket *sock = file->private_data;
772 size_t size = 0;
773 int i;
774
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700775 for (i = 0; i < nr_segs; i++)
776 size += iov[i].iov_len;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800777
778 msg->msg_name = NULL;
779 msg->msg_namelen = 0;
780 msg->msg_control = NULL;
781 msg->msg_controllen = 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700782 msg->msg_iov = (struct iovec *)iov;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800783 msg->msg_iovlen = nr_segs;
784 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
785 if (sock->type == SOCK_SEQPACKET)
786 msg->msg_flags |= MSG_EOR;
787
788 return __sock_sendmsg(iocb, sock, msg, size);
789}
790
Badari Pulavarty027445c2006-09-30 23:28:46 -0700791static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
792 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800794 struct sock_iocb siocb, *x;
795
796 if (pos != 0)
797 return -ESPIPE;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700798
Badari Pulavarty027445c2006-09-30 23:28:46 -0700799 x = alloc_sock_iocb(iocb, &siocb);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800800 if (!x)
801 return -ENOMEM;
802
Badari Pulavarty027445c2006-09-30 23:28:46 -0700803 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806/*
807 * Atomic setting of ioctl hooks to avoid race
808 * with module unload.
809 */
810
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800811static DEFINE_MUTEX(br_ioctl_mutex);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700812static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Eric W. Biederman881d9662007-09-17 11:56:21 -0700814void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800816 mutex_lock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 br_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800818 mutex_unlock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821EXPORT_SYMBOL(brioctl_set);
822
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800823static DEFINE_MUTEX(vlan_ioctl_mutex);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700824static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Eric W. Biederman881d9662007-09-17 11:56:21 -0700826void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800828 mutex_lock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 vlan_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800830 mutex_unlock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833EXPORT_SYMBOL(vlan_ioctl_set);
834
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800835static DEFINE_MUTEX(dlci_ioctl_mutex);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700836static int (*dlci_ioctl_hook) (unsigned int, void __user *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700838void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800840 mutex_lock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 dlci_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800842 mutex_unlock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845EXPORT_SYMBOL(dlci_ioctl_set);
846
847/*
848 * With an ioctl, arg may well be a user mode pointer, but we don't know
849 * what to do with it - that's up to the protocol still.
850 */
851
852static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
853{
854 struct socket *sock;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700855 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 void __user *argp = (void __user *)arg;
857 int pid, err;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700858 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Eric Dumazetb69aee02005-09-06 14:42:45 -0700860 sock = file->private_data;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700861 sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900862 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
Eric W. Biederman881d9662007-09-17 11:56:21 -0700864 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 } else
Adrian Bunkd86b5e02006-01-21 00:46:55 +0100866#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
Eric W. Biederman881d9662007-09-17 11:56:21 -0700868 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 } else
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700870#endif /* CONFIG_WIRELESS_EXT */
871 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 case FIOSETOWN:
873 case SIOCSPGRP:
874 err = -EFAULT;
875 if (get_user(pid, (int __user *)argp))
876 break;
877 err = f_setown(sock->file, pid, 1);
878 break;
879 case FIOGETOWN:
880 case SIOCGPGRP:
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700881 err = put_user(f_getown(sock->file),
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700882 (int __user *)argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 break;
884 case SIOCGIFBR:
885 case SIOCSIFBR:
886 case SIOCBRADDBR:
887 case SIOCBRDELBR:
888 err = -ENOPKG;
889 if (!br_ioctl_hook)
890 request_module("bridge");
891
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800892 mutex_lock(&br_ioctl_mutex);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700893 if (br_ioctl_hook)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700894 err = br_ioctl_hook(net, cmd, argp);
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800895 mutex_unlock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 break;
897 case SIOCGIFVLAN:
898 case SIOCSIFVLAN:
899 err = -ENOPKG;
900 if (!vlan_ioctl_hook)
901 request_module("8021q");
902
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800903 mutex_lock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 if (vlan_ioctl_hook)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700905 err = vlan_ioctl_hook(net, argp);
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800906 mutex_unlock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 case SIOCADDDLCI:
909 case SIOCDELDLCI:
910 err = -ENOPKG;
911 if (!dlci_ioctl_hook)
912 request_module("dlci");
913
Pavel Emelyanov7512cbf2008-03-21 15:58:52 -0700914 mutex_lock(&dlci_ioctl_mutex);
915 if (dlci_ioctl_hook)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 err = dlci_ioctl_hook(cmd, argp);
Pavel Emelyanov7512cbf2008-03-21 15:58:52 -0700917 mutex_unlock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 break;
919 default:
920 err = sock->ops->ioctl(sock, cmd, arg);
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800921
922 /*
923 * If this ioctl is unknown try to hand it down
924 * to the NIC driver.
925 */
926 if (err == -ENOIOCTLCMD)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700927 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 break;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700929 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 return err;
931}
932
933int sock_create_lite(int family, int type, int protocol, struct socket **res)
934{
935 int err;
936 struct socket *sock = NULL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700937
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 err = security_socket_create(family, type, protocol, 1);
939 if (err)
940 goto out;
941
942 sock = sock_alloc();
943 if (!sock) {
944 err = -ENOMEM;
945 goto out;
946 }
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 sock->type = type;
Venkat Yekkirala7420ed22006-08-04 23:17:57 -0700949 err = security_socket_post_create(sock, family, type, protocol, 1);
950 if (err)
951 goto out_release;
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953out:
954 *res = sock;
955 return err;
Venkat Yekkirala7420ed22006-08-04 23:17:57 -0700956out_release:
957 sock_release(sock);
958 sock = NULL;
959 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
961
962/* No kernel lock held - perfect */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700963static unsigned int sock_poll(struct file *file, poll_table *wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
965 struct socket *sock;
966
967 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700968 * We can't return errors to poll, so it's either yes or no.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
Eric Dumazetb69aee02005-09-06 14:42:45 -0700970 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 return sock->ops->poll(file, sock, wait);
972}
973
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700974static int sock_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975{
Eric Dumazetb69aee02005-09-06 14:42:45 -0700976 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978 return sock->ops->mmap(file, sock, vma);
979}
980
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300981static int sock_close(struct inode *inode, struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
983 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700984 * It was possible the inode is NULL we were
985 * closing an unfinished socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 */
987
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700988 if (!inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 printk(KERN_DEBUG "sock_close: NULL inode\n");
990 return 0;
991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 sock_release(SOCKET_I(inode));
993 return 0;
994}
995
996/*
997 * Update the socket async list
998 *
999 * Fasync_list locking strategy.
1000 *
1001 * 1. fasync_list is modified only under process context socket lock
1002 * i.e. under semaphore.
1003 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1004 * or under socket lock.
1005 * 3. fasync_list can be used from softirq context, so that
1006 * modification under socket lock have to be enhanced with
1007 * write_lock_bh(&sk->sk_callback_lock).
1008 * --ANK (990710)
1009 */
1010
1011static int sock_fasync(int fd, struct file *filp, int on)
1012{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001013 struct fasync_struct *fa, *fna = NULL, **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 struct socket *sock;
1015 struct sock *sk;
1016
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001017 if (on) {
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08001018 fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001019 if (fna == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 return -ENOMEM;
1021 }
1022
Eric Dumazetb69aee02005-09-06 14:42:45 -07001023 sock = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001025 sk = sock->sk;
1026 if (sk == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 kfree(fna);
1028 return -EINVAL;
1029 }
1030
1031 lock_sock(sk);
1032
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001033 prev = &(sock->fasync_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001035 for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1036 if (fa->fa_file == filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 break;
1038
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001039 if (on) {
1040 if (fa != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001042 fa->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 write_unlock_bh(&sk->sk_callback_lock);
1044
1045 kfree(fna);
1046 goto out;
1047 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001048 fna->fa_file = filp;
1049 fna->fa_fd = fd;
1050 fna->magic = FASYNC_MAGIC;
1051 fna->fa_next = sock->fasync_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001053 sock->fasync_list = fna;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 write_unlock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001055 } else {
1056 if (fa != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001058 *prev = fa->fa_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 write_unlock_bh(&sk->sk_callback_lock);
1060 kfree(fa);
1061 }
1062 }
1063
1064out:
1065 release_sock(sock->sk);
1066 return 0;
1067}
1068
1069/* This function may be called only under socket lock or callback_lock */
1070
1071int sock_wake_async(struct socket *sock, int how, int band)
1072{
1073 if (!sock || !sock->fasync_list)
1074 return -1;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001075 switch (how) {
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001076 case SOCK_WAKE_WAITD:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1078 break;
1079 goto call_kill;
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001080 case SOCK_WAKE_SPACE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1082 break;
1083 /* fall through */
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001084 case SOCK_WAKE_IO:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001085call_kill:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 __kill_fasync(sock->fasync_list, SIGIO, band);
1087 break;
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001088 case SOCK_WAKE_URG:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 __kill_fasync(sock->fasync_list, SIGURG, band);
1090 }
1091 return 0;
1092}
1093
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001094static int __sock_create(struct net *net, int family, int type, int protocol,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001095 struct socket **res, int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 int err;
1098 struct socket *sock;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001099 const struct net_proto_family *pf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001102 * Check protocol is in range
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
1104 if (family < 0 || family >= NPROTO)
1105 return -EAFNOSUPPORT;
1106 if (type < 0 || type >= SOCK_MAX)
1107 return -EINVAL;
1108
1109 /* Compatibility.
1110
1111 This uglymoron is moved from INET layer to here to avoid
1112 deadlock in module load.
1113 */
1114 if (family == PF_INET && type == SOCK_PACKET) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001115 static int warned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (!warned) {
1117 warned = 1;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001118 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1119 current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 }
1121 family = PF_PACKET;
1122 }
1123
1124 err = security_socket_create(family, type, protocol, kern);
1125 if (err)
1126 return err;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001127
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001128 /*
1129 * Allocate the socket and allow the family to set things up. if
1130 * the protocol is 0, the family is instructed to select an appropriate
1131 * default.
1132 */
1133 sock = sock_alloc();
1134 if (!sock) {
1135 if (net_ratelimit())
1136 printk(KERN_WARNING "socket: no more sockets\n");
1137 return -ENFILE; /* Not exactly a match, but its the
1138 closest posix thing */
1139 }
1140
1141 sock->type = type;
1142
Johannes Berg95a5afc2008-10-16 15:24:51 -07001143#ifdef CONFIG_MODULES
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001144 /* Attempt to load a protocol module if the find failed.
1145 *
1146 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 * requested real, full-featured networking support upon configuration.
1148 * Otherwise module support will break!
1149 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001150 if (net_families[family] == NULL)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001151 request_module("net-pf-%d", family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152#endif
1153
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001154 rcu_read_lock();
1155 pf = rcu_dereference(net_families[family]);
1156 err = -EAFNOSUPPORT;
1157 if (!pf)
1158 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 /*
1161 * We will call the ->create function, that possibly is in a loadable
1162 * module, so we have to bump that loadable module refcnt first.
1163 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001164 if (!try_module_get(pf->owner))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 goto out_release;
1166
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001167 /* Now protected by module ref count */
1168 rcu_read_unlock();
1169
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001170 err = pf->create(net, sock, protocol);
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001171 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 goto out_module_put;
Frank Filza79af592005-09-27 15:23:38 -07001173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 /*
1175 * Now to bump the refcnt of the [loadable] module that owns this
1176 * socket at sock_release time we decrement its refcnt.
1177 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001178 if (!try_module_get(sock->ops->owner))
1179 goto out_module_busy;
1180
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 /*
1182 * Now that we're done with the ->create function, the [loadable]
1183 * module can have its refcnt decremented
1184 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001185 module_put(pf->owner);
Venkat Yekkirala7420ed22006-08-04 23:17:57 -07001186 err = security_socket_post_create(sock, family, type, protocol, kern);
1187 if (err)
Herbert Xu3b185522007-08-15 14:46:02 -07001188 goto out_sock_release;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001189 *res = sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001191 return 0;
1192
1193out_module_busy:
1194 err = -EAFNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195out_module_put:
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001196 sock->ops = NULL;
1197 module_put(pf->owner);
1198out_sock_release:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 sock_release(sock);
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001200 return err;
1201
1202out_release:
1203 rcu_read_unlock();
1204 goto out_sock_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207int sock_create(int family, int type, int protocol, struct socket **res)
1208{
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001209 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
1212int sock_create_kern(int family, int type, int protocol, struct socket **res)
1213{
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001214 return __sock_create(&init_net, family, type, protocol, res, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215}
1216
1217asmlinkage long sys_socket(int family, int type, int protocol)
1218{
1219 int retval;
1220 struct socket *sock;
Ulrich Dreppera677a032008-07-23 21:29:17 -07001221 int flags;
1222
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001223 /* Check the SOCK_* constants for consistency. */
1224 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1225 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1226 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1227 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1228
Ulrich Dreppera677a032008-07-23 21:29:17 -07001229 flags = type & ~SOCK_TYPE_MASK;
Ulrich Drepper77d27202008-07-23 21:29:35 -07001230 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Dreppera677a032008-07-23 21:29:17 -07001231 return -EINVAL;
1232 type &= SOCK_TYPE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001234 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1235 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 retval = sock_create(family, type, protocol, &sock);
1238 if (retval < 0)
1239 goto out;
1240
Ulrich Drepper77d27202008-07-23 21:29:35 -07001241 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (retval < 0)
1243 goto out_release;
1244
1245out:
1246 /* It may be already another descriptor 8) Not kernel problem. */
1247 return retval;
1248
1249out_release:
1250 sock_release(sock);
1251 return retval;
1252}
1253
1254/*
1255 * Create a pair of connected sockets.
1256 */
1257
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001258asmlinkage long sys_socketpair(int family, int type, int protocol,
1259 int __user *usockvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260{
1261 struct socket *sock1, *sock2;
1262 int fd1, fd2, err;
Al Virodb349502007-02-07 01:48:00 -05001263 struct file *newfile1, *newfile2;
Ulrich Dreppera677a032008-07-23 21:29:17 -07001264 int flags;
1265
1266 flags = type & ~SOCK_TYPE_MASK;
Ulrich Drepper77d27202008-07-23 21:29:35 -07001267 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Dreppera677a032008-07-23 21:29:17 -07001268 return -EINVAL;
1269 type &= SOCK_TYPE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001271 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1272 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 /*
1275 * Obtain the first socket and check if the underlying protocol
1276 * supports the socketpair call.
1277 */
1278
1279 err = sock_create(family, type, protocol, &sock1);
1280 if (err < 0)
1281 goto out;
1282
1283 err = sock_create(family, type, protocol, &sock2);
1284 if (err < 0)
1285 goto out_release_1;
1286
1287 err = sock1->ops->socketpair(sock1, sock2);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001288 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 goto out_release_both;
1290
Ulrich Dreppera677a032008-07-23 21:29:17 -07001291 fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
David S. Millerbf3c23d2007-10-29 21:54:02 -07001292 if (unlikely(fd1 < 0)) {
1293 err = fd1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 goto out_release_both;
David S. Millerbf3c23d2007-10-29 21:54:02 -07001295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Ulrich Dreppera677a032008-07-23 21:29:17 -07001297 fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
Al Virodb349502007-02-07 01:48:00 -05001298 if (unlikely(fd2 < 0)) {
David S. Millerbf3c23d2007-10-29 21:54:02 -07001299 err = fd2;
Al Virodb349502007-02-07 01:48:00 -05001300 put_filp(newfile1);
1301 put_unused_fd(fd1);
1302 goto out_release_both;
1303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Ulrich Drepper77d27202008-07-23 21:29:35 -07001305 err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
Al Virodb349502007-02-07 01:48:00 -05001306 if (unlikely(err < 0)) {
1307 goto out_fd2;
1308 }
1309
Ulrich Drepper77d27202008-07-23 21:29:35 -07001310 err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
Al Virodb349502007-02-07 01:48:00 -05001311 if (unlikely(err < 0)) {
1312 fput(newfile1);
1313 goto out_fd1;
1314 }
1315
Al Viro157cf642008-12-14 04:57:47 -05001316 audit_fd_pair(fd1, fd2);
Al Virodb349502007-02-07 01:48:00 -05001317 fd_install(fd1, newfile1);
1318 fd_install(fd2, newfile2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 /* fd1 and fd2 may be already another descriptors.
1320 * Not kernel problem.
1321 */
1322
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001323 err = put_user(fd1, &usockvec[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 if (!err)
1325 err = put_user(fd2, &usockvec[1]);
1326 if (!err)
1327 return 0;
1328
1329 sys_close(fd2);
1330 sys_close(fd1);
1331 return err;
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333out_release_both:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001334 sock_release(sock2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335out_release_1:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001336 sock_release(sock1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337out:
1338 return err;
Al Virodb349502007-02-07 01:48:00 -05001339
1340out_fd2:
1341 put_filp(newfile1);
1342 sock_release(sock1);
1343out_fd1:
1344 put_filp(newfile2);
1345 sock_release(sock2);
Al Virodb349502007-02-07 01:48:00 -05001346 put_unused_fd(fd1);
1347 put_unused_fd(fd2);
1348 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349}
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351/*
1352 * Bind a name to a socket. Nothing much to do here since it's
1353 * the protocol's responsibility to handle the local address.
1354 *
1355 * We move the socket address to kernel space before we call
1356 * the protocol layer (having also checked the address is ok).
1357 */
1358
1359asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1360{
1361 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001362 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001363 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001365 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001366 if (sock) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001367 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001368 if (err >= 0) {
1369 err = security_socket_bind(sock,
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001370 (struct sockaddr *)&address,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001371 addrlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001372 if (!err)
1373 err = sock->ops->bind(sock,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001374 (struct sockaddr *)
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001375 &address, addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001377 fput_light(sock->file, fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return err;
1380}
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382/*
1383 * Perform a listen. Basically, we allow the protocol to do anything
1384 * necessary for a listen, and if that works, we mark the socket as
1385 * ready for listening.
1386 */
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388asmlinkage long sys_listen(int fd, int backlog)
1389{
1390 struct socket *sock;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001391 int err, fput_needed;
Pavel Emelyanovb8e1f9b2007-12-08 00:12:33 -08001392 int somaxconn;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001393
1394 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1395 if (sock) {
Pavel Emelyanov8efa6e92008-03-31 19:41:14 -07001396 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
Pavel Emelyanovb8e1f9b2007-12-08 00:12:33 -08001397 if ((unsigned)backlog > somaxconn)
1398 backlog = somaxconn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 err = security_socket_listen(sock, backlog);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001401 if (!err)
1402 err = sock->ops->listen(sock, backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001404 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 }
1406 return err;
1407}
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409/*
1410 * For accept, we attempt to create a new socket, set up the link
1411 * with the client, wake up the client, then return the new
1412 * connected fd. We collect the address of the connector in kernel
1413 * space and move it to user at the very end. This is unclean because
1414 * we open the socket then return an error.
1415 *
1416 * 1003.1g adds the ability to recvmsg() to query connection pending
1417 * status to recvmsg. We need to add that support in a way thats
1418 * clean when we restucture accept also.
1419 */
1420
Ulrich Drepperde11def2008-11-19 15:36:14 -08001421asmlinkage long sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1422 int __user *upeer_addrlen, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
1424 struct socket *sock, *newsock;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001425 struct file *newfile;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001426 int err, len, newfd, fput_needed;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001427 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Ulrich Drepper77d27202008-07-23 21:29:35 -07001429 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001430 return -EINVAL;
1431
1432 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1433 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1434
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001435 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (!sock)
1437 goto out;
1438
1439 err = -ENFILE;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001440 if (!(newsock = sock_alloc()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 goto out_put;
1442
1443 newsock->type = sock->type;
1444 newsock->ops = sock->ops;
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 /*
1447 * We don't need try_module_get here, as the listening socket (sock)
1448 * has the protocol module (sock->ops->owner) held.
1449 */
1450 __module_get(newsock->ops->owner);
1451
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001452 newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001453 if (unlikely(newfd < 0)) {
1454 err = newfd;
David S. Miller9a1875e2006-04-01 12:48:36 -08001455 sock_release(newsock);
1456 goto out_put;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001457 }
1458
Ulrich Drepper77d27202008-07-23 21:29:35 -07001459 err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001460 if (err < 0)
Alexey Dobriyan79f4f642007-03-26 14:09:52 -07001461 goto out_fd_simple;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001462
Frank Filza79af592005-09-27 15:23:38 -07001463 err = security_socket_accept(sock, newsock);
1464 if (err)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001465 goto out_fd;
Frank Filza79af592005-09-27 15:23:38 -07001466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1468 if (err < 0)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001469 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
1471 if (upeer_sockaddr) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001472 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001473 &len, 2) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 err = -ECONNABORTED;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001475 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 }
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001477 err = move_addr_to_user((struct sockaddr *)&address,
1478 len, upeer_sockaddr, upeer_addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (err < 0)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001480 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 }
1482
1483 /* File flags are not inherited via accept() unlike another OSes. */
1484
David S. Miller39d8c1b2006-03-20 17:13:49 -08001485 fd_install(newfd, newfile);
1486 err = newfd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 security_socket_post_accept(sock, newsock);
1489
1490out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001491 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492out:
1493 return err;
Alexey Dobriyan79f4f642007-03-26 14:09:52 -07001494out_fd_simple:
1495 sock_release(newsock);
1496 put_filp(newfile);
1497 put_unused_fd(newfd);
1498 goto out_put;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001499out_fd:
David S. Miller9606a212006-04-01 01:00:14 -08001500 fput(newfile);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001501 put_unused_fd(newfd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 goto out_put;
1503}
1504
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001505asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr,
1506 int __user *upeer_addrlen)
1507{
Ulrich Drepperde11def2008-11-19 15:36:14 -08001508 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001509}
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511/*
1512 * Attempt to connect to a socket with the server address. The address
1513 * is in user space so we verify it is OK and move it to kernel space.
1514 *
1515 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1516 * break bindings
1517 *
1518 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1519 * other SEQPACKET protocols that take time to connect() as it doesn't
1520 * include the -EINPROGRESS status for such sockets.
1521 */
1522
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001523asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr,
1524 int addrlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
1526 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001527 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001528 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001530 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 if (!sock)
1532 goto out;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001533 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 if (err < 0)
1535 goto out_put;
1536
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001537 err =
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001538 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (err)
1540 goto out_put;
1541
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001542 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 sock->file->f_flags);
1544out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001545 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546out:
1547 return err;
1548}
1549
1550/*
1551 * Get the local address ('name') of a socket object. Move the obtained
1552 * name to user space.
1553 */
1554
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001555asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1556 int __user *usockaddr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001559 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001560 int len, err, fput_needed;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001561
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001562 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 if (!sock)
1564 goto out;
1565
1566 err = security_socket_getsockname(sock);
1567 if (err)
1568 goto out_put;
1569
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001570 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 if (err)
1572 goto out_put;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001573 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
1575out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001576 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577out:
1578 return err;
1579}
1580
1581/*
1582 * Get the remote address ('name') of a socket object. Move the obtained
1583 * name to user space.
1584 */
1585
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001586asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1587 int __user *usockaddr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
1589 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001590 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001591 int len, err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001593 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1594 if (sock != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 err = security_socket_getpeername(sock);
1596 if (err) {
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001597 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return err;
1599 }
1600
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001601 err =
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001602 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001603 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 if (!err)
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001605 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001606 usockaddr_len);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001607 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
1609 return err;
1610}
1611
1612/*
1613 * Send a datagram to a given address. We move the address into kernel
1614 * space and check the user space data area is readable before invoking
1615 * the protocol.
1616 */
1617
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001618asmlinkage long sys_sendto(int fd, void __user *buff, size_t len,
1619 unsigned flags, struct sockaddr __user *addr,
1620 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621{
1622 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001623 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 int err;
1625 struct msghdr msg;
1626 struct iovec iov;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001627 int fput_needed;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001628
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001629 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1630 if (!sock)
David S. Miller4387ff72007-02-08 15:06:08 -08001631 goto out;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001632
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001633 iov.iov_base = buff;
1634 iov.iov_len = len;
1635 msg.msg_name = NULL;
1636 msg.msg_iov = &iov;
1637 msg.msg_iovlen = 1;
1638 msg.msg_control = NULL;
1639 msg.msg_controllen = 0;
1640 msg.msg_namelen = 0;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001641 if (addr) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001642 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 if (err < 0)
1644 goto out_put;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001645 msg.msg_name = (struct sockaddr *)&address;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001646 msg.msg_namelen = addr_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 }
1648 if (sock->file->f_flags & O_NONBLOCK)
1649 flags |= MSG_DONTWAIT;
1650 msg.msg_flags = flags;
1651 err = sock_sendmsg(sock, &msg, len);
1652
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001653out_put:
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001654 fput_light(sock->file, fput_needed);
David S. Miller4387ff72007-02-08 15:06:08 -08001655out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 return err;
1657}
1658
1659/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001660 * Send a datagram down a socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 */
1662
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001663asmlinkage long sys_send(int fd, void __user *buff, size_t len, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664{
1665 return sys_sendto(fd, buff, len, flags, NULL, 0);
1666}
1667
1668/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001669 * Receive a frame from the socket and optionally record the address of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * sender. We verify the buffers are writable and if needed move the
1671 * sender address from kernel to user space.
1672 */
1673
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001674asmlinkage long sys_recvfrom(int fd, void __user *ubuf, size_t size,
1675 unsigned flags, struct sockaddr __user *addr,
1676 int __user *addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677{
1678 struct socket *sock;
1679 struct iovec iov;
1680 struct msghdr msg;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001681 struct sockaddr_storage address;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001682 int err, err2;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001683 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001685 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (!sock)
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001687 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001689 msg.msg_control = NULL;
1690 msg.msg_controllen = 0;
1691 msg.msg_iovlen = 1;
1692 msg.msg_iov = &iov;
1693 iov.iov_len = size;
1694 iov.iov_base = ubuf;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001695 msg.msg_name = (struct sockaddr *)&address;
1696 msg.msg_namelen = sizeof(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (sock->file->f_flags & O_NONBLOCK)
1698 flags |= MSG_DONTWAIT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001699 err = sock_recvmsg(sock, &msg, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001701 if (err >= 0 && addr != NULL) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001702 err2 = move_addr_to_user((struct sockaddr *)&address,
1703 msg.msg_namelen, addr, addr_len);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001704 if (err2 < 0)
1705 err = err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 }
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001707
1708 fput_light(sock->file, fput_needed);
David S. Miller4387ff72007-02-08 15:06:08 -08001709out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 return err;
1711}
1712
1713/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001714 * Receive a datagram from a socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 */
1716
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001717asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1718 unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719{
1720 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1721}
1722
1723/*
1724 * Set a socket option. Because we don't know the option lengths we have
1725 * to pass the user mode parameter for the protocols to sort out.
1726 */
1727
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001728asmlinkage long sys_setsockopt(int fd, int level, int optname,
1729 char __user *optval, int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001731 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 struct socket *sock;
1733
1734 if (optlen < 0)
1735 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001736
1737 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1738 if (sock != NULL) {
1739 err = security_socket_setsockopt(sock, level, optname);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001740 if (err)
1741 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 if (level == SOL_SOCKET)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001744 err =
1745 sock_setsockopt(sock, level, optname, optval,
1746 optlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001748 err =
1749 sock->ops->setsockopt(sock, level, optname, optval,
1750 optlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001751out_put:
1752 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
1754 return err;
1755}
1756
1757/*
1758 * Get a socket option. Because we don't know the option lengths we have
1759 * to pass a user mode parameter for the protocols to sort out.
1760 */
1761
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001762asmlinkage long sys_getsockopt(int fd, int level, int optname,
1763 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001765 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 struct socket *sock;
1767
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001768 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1769 if (sock != NULL) {
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001770 err = security_socket_getsockopt(sock, level, optname);
1771 if (err)
1772 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 if (level == SOL_SOCKET)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001775 err =
1776 sock_getsockopt(sock, level, optname, optval,
1777 optlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001779 err =
1780 sock->ops->getsockopt(sock, level, optname, optval,
1781 optlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001782out_put:
1783 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 }
1785 return err;
1786}
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788/*
1789 * Shutdown a socket.
1790 */
1791
Heiko Carstens754fe8d2009-01-14 14:14:09 +01001792SYSCALL_DEFINE2(shutdown, int, fd, int, how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001794 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 struct socket *sock;
1796
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001797 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1798 if (sock != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 err = security_socket_shutdown(sock, how);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001800 if (!err)
1801 err = sock->ops->shutdown(sock, how);
1802 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804 return err;
1805}
1806
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001807/* A couple of helpful macros for getting the address of the 32/64 bit
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 * fields which are the same type (int / unsigned) on our platforms.
1809 */
1810#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1811#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1812#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814/*
1815 * BSD sendmsg interface
1816 */
1817
1818asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1819{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001820 struct compat_msghdr __user *msg_compat =
1821 (struct compat_msghdr __user *)msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001823 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
Alex Williamsonb9d717a2005-09-26 14:28:02 -07001825 unsigned char ctl[sizeof(struct cmsghdr) + 20]
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001826 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1827 /* 20 is size of ipv6_pktinfo */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 unsigned char *ctl_buf = ctl;
1829 struct msghdr msg_sys;
1830 int err, ctl_len, iov_size, total_len;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001831 int fput_needed;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001832
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 err = -EFAULT;
1834 if (MSG_CMSG_COMPAT & flags) {
1835 if (get_compat_msghdr(&msg_sys, msg_compat))
1836 return -EFAULT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001837 }
1838 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 return -EFAULT;
1840
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001841 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001842 if (!sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 goto out;
1844
1845 /* do not move before msg_sys is valid */
1846 err = -EMSGSIZE;
1847 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1848 goto out_put;
1849
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001850 /* Check whether to allocate the iovec area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 err = -ENOMEM;
1852 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1853 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1854 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1855 if (!iov)
1856 goto out_put;
1857 }
1858
1859 /* This will also move the address data into kernel space */
1860 if (MSG_CMSG_COMPAT & flags) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001861 err = verify_compat_iovec(&msg_sys, iov,
1862 (struct sockaddr *)&address,
1863 VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 } else
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001865 err = verify_iovec(&msg_sys, iov,
1866 (struct sockaddr *)&address,
1867 VERIFY_READ);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001868 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 goto out_freeiov;
1870 total_len = err;
1871
1872 err = -ENOBUFS;
1873
1874 if (msg_sys.msg_controllen > INT_MAX)
1875 goto out_freeiov;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001876 ctl_len = msg_sys.msg_controllen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001878 err =
1879 cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1880 sizeof(ctl));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (err)
1882 goto out_freeiov;
1883 ctl_buf = msg_sys.msg_control;
Al Viro8920e8f2005-09-07 18:28:51 -07001884 ctl_len = msg_sys.msg_controllen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 } else if (ctl_len) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001886 if (ctl_len > sizeof(ctl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001888 if (ctl_buf == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 goto out_freeiov;
1890 }
1891 err = -EFAULT;
1892 /*
1893 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1894 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1895 * checking falls down on this.
1896 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001897 if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1898 ctl_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 goto out_freectl;
1900 msg_sys.msg_control = ctl_buf;
1901 }
1902 msg_sys.msg_flags = flags;
1903
1904 if (sock->file->f_flags & O_NONBLOCK)
1905 msg_sys.msg_flags |= MSG_DONTWAIT;
1906 err = sock_sendmsg(sock, &msg_sys, total_len);
1907
1908out_freectl:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001909 if (ctl_buf != ctl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1911out_freeiov:
1912 if (iov != iovstack)
1913 sock_kfree_s(sock->sk, iov, iov_size);
1914out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001915 fput_light(sock->file, fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001916out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 return err;
1918}
1919
1920/*
1921 * BSD recvmsg interface
1922 */
1923
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001924asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg,
1925 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001927 struct compat_msghdr __user *msg_compat =
1928 (struct compat_msghdr __user *)msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 struct socket *sock;
1930 struct iovec iovstack[UIO_FASTIOV];
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001931 struct iovec *iov = iovstack;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 struct msghdr msg_sys;
1933 unsigned long cmsg_ptr;
1934 int err, iov_size, total_len, len;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001935 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
1937 /* kernel mode address */
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001938 struct sockaddr_storage addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
1940 /* user mode address pointers */
1941 struct sockaddr __user *uaddr;
1942 int __user *uaddr_len;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (MSG_CMSG_COMPAT & flags) {
1945 if (get_compat_msghdr(&msg_sys, msg_compat))
1946 return -EFAULT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001947 }
1948 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1949 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001951 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 if (!sock)
1953 goto out;
1954
1955 err = -EMSGSIZE;
1956 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1957 goto out_put;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001958
1959 /* Check whether to allocate the iovec area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 err = -ENOMEM;
1961 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1962 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1963 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1964 if (!iov)
1965 goto out_put;
1966 }
1967
1968 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001969 * Save the user-mode address (verify_iovec will change the
1970 * kernel msghdr to use the kernel address space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001972
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -07001973 uaddr = (__force void __user *)msg_sys.msg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 uaddr_len = COMPAT_NAMELEN(msg);
1975 if (MSG_CMSG_COMPAT & flags) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001976 err = verify_compat_iovec(&msg_sys, iov,
1977 (struct sockaddr *)&addr,
1978 VERIFY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 } else
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001980 err = verify_iovec(&msg_sys, iov,
1981 (struct sockaddr *)&addr,
1982 VERIFY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (err < 0)
1984 goto out_freeiov;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001985 total_len = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987 cmsg_ptr = (unsigned long)msg_sys.msg_control;
Ulrich Drepper4a195422007-07-15 23:40:34 -07001988 msg_sys.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 if (sock->file->f_flags & O_NONBLOCK)
1991 flags |= MSG_DONTWAIT;
1992 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1993 if (err < 0)
1994 goto out_freeiov;
1995 len = err;
1996
1997 if (uaddr != NULL) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001998 err = move_addr_to_user((struct sockaddr *)&addr,
1999 msg_sys.msg_namelen, uaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002000 uaddr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (err < 0)
2002 goto out_freeiov;
2003 }
David S. Miller37f7f422005-09-16 16:51:01 -07002004 err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
2005 COMPAT_FLAGS(msg));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 if (err)
2007 goto out_freeiov;
2008 if (MSG_CMSG_COMPAT & flags)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002009 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 &msg_compat->msg_controllen);
2011 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002012 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 &msg->msg_controllen);
2014 if (err)
2015 goto out_freeiov;
2016 err = len;
2017
2018out_freeiov:
2019 if (iov != iovstack)
2020 sock_kfree_s(sock->sk, iov, iov_size);
2021out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08002022 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023out:
2024 return err;
2025}
2026
2027#ifdef __ARCH_WANT_SYS_SOCKETCALL
2028
2029/* Argument list sizes for sys_socketcall */
2030#define AL(x) ((x) * sizeof(unsigned long))
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002031static const unsigned char nargs[19]={
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002032 AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2033 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002034 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
Ulrich Drepperde11def2008-11-19 15:36:14 -08002035 AL(4)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002036};
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038#undef AL
2039
2040/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002041 * System call vectors.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 *
2043 * Argument checking cleaned up. Saved 20% in size.
2044 * This function doesn't need to set the kernel lock because
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002045 * it is set by the callees.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 */
2047
2048asmlinkage long sys_socketcall(int call, unsigned long __user *args)
2049{
2050 unsigned long a[6];
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002051 unsigned long a0, a1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 int err;
2053
Ulrich Drepperde11def2008-11-19 15:36:14 -08002054 if (call < 1 || call > SYS_ACCEPT4)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return -EINVAL;
2056
2057 /* copy_from_user should be SMP safe. */
2058 if (copy_from_user(a, args, nargs[call]))
2059 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01002060
Al Virof3298dc2008-12-10 03:16:51 -05002061 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01002062
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002063 a0 = a[0];
2064 a1 = a[1];
2065
2066 switch (call) {
2067 case SYS_SOCKET:
2068 err = sys_socket(a0, a1, a[2]);
2069 break;
2070 case SYS_BIND:
2071 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2072 break;
2073 case SYS_CONNECT:
2074 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2075 break;
2076 case SYS_LISTEN:
2077 err = sys_listen(a0, a1);
2078 break;
2079 case SYS_ACCEPT:
Ulrich Drepperde11def2008-11-19 15:36:14 -08002080 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2081 (int __user *)a[2], 0);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002082 break;
2083 case SYS_GETSOCKNAME:
2084 err =
2085 sys_getsockname(a0, (struct sockaddr __user *)a1,
2086 (int __user *)a[2]);
2087 break;
2088 case SYS_GETPEERNAME:
2089 err =
2090 sys_getpeername(a0, (struct sockaddr __user *)a1,
2091 (int __user *)a[2]);
2092 break;
2093 case SYS_SOCKETPAIR:
2094 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2095 break;
2096 case SYS_SEND:
2097 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2098 break;
2099 case SYS_SENDTO:
2100 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2101 (struct sockaddr __user *)a[4], a[5]);
2102 break;
2103 case SYS_RECV:
2104 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2105 break;
2106 case SYS_RECVFROM:
2107 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2108 (struct sockaddr __user *)a[4],
2109 (int __user *)a[5]);
2110 break;
2111 case SYS_SHUTDOWN:
2112 err = sys_shutdown(a0, a1);
2113 break;
2114 case SYS_SETSOCKOPT:
2115 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2116 break;
2117 case SYS_GETSOCKOPT:
2118 err =
2119 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2120 (int __user *)a[4]);
2121 break;
2122 case SYS_SENDMSG:
2123 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2124 break;
2125 case SYS_RECVMSG:
2126 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2127 break;
Ulrich Drepperde11def2008-11-19 15:36:14 -08002128 case SYS_ACCEPT4:
2129 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2130 (int __user *)a[2], a[3]);
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002131 break;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002132 default:
2133 err = -EINVAL;
2134 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 }
2136 return err;
2137}
2138
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002139#endif /* __ARCH_WANT_SYS_SOCKETCALL */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002141/**
2142 * sock_register - add a socket protocol handler
2143 * @ops: description of protocol
2144 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * This function is called by a protocol handler that wants to
2146 * advertise its address family, and have it linked into the
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002147 * socket interface. The value ops->family coresponds to the
2148 * socket system call protocol family.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 */
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002150int sock_register(const struct net_proto_family *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
2152 int err;
2153
2154 if (ops->family >= NPROTO) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002155 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2156 NPROTO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 return -ENOBUFS;
2158 }
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002159
2160 spin_lock(&net_family_lock);
2161 if (net_families[ops->family])
2162 err = -EEXIST;
2163 else {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002164 net_families[ops->family] = ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 err = 0;
2166 }
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002167 spin_unlock(&net_family_lock);
2168
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002169 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 return err;
2171}
2172
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002173/**
2174 * sock_unregister - remove a protocol handler
2175 * @family: protocol family to remove
2176 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 * This function is called by a protocol handler that wants to
2178 * remove its address family, and have it unlinked from the
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002179 * new socket creation.
2180 *
2181 * If protocol handler is a module, then it can use module reference
2182 * counts to protect against new references. If protocol handler is not
2183 * a module then it needs to provide its own protection in
2184 * the ops->create routine.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 */
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002186void sock_unregister(int family)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187{
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002188 BUG_ON(family < 0 || family >= NPROTO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002190 spin_lock(&net_family_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002191 net_families[family] = NULL;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002192 spin_unlock(&net_family_lock);
2193
2194 synchronize_rcu();
2195
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002196 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197}
2198
Andi Kleen77d76ea2005-12-22 12:43:42 -08002199static int __init sock_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
2201 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002202 * Initialize sock SLAB cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 sk_init();
2206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002208 * Initialize skbuff SLAB cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 */
2210 skb_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
2212 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002213 * Initialize the protocols module.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 */
2215
2216 init_inodecache();
2217 register_filesystem(&sock_fs_type);
2218 sock_mnt = kern_mount(&sock_fs_type);
Andi Kleen77d76ea2005-12-22 12:43:42 -08002219
2220 /* The real protocol initialization is performed in later initcalls.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 */
2222
2223#ifdef CONFIG_NETFILTER
2224 netfilter_init();
2225#endif
David S. Millercbeb3212005-12-22 12:58:55 -08002226
2227 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
2229
Andi Kleen77d76ea2005-12-22 12:43:42 -08002230core_initcall(sock_init); /* early initcall */
2231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232#ifdef CONFIG_PROC_FS
2233void socket_seq_show(struct seq_file *seq)
2234{
2235 int cpu;
2236 int counter = 0;
2237
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07002238 for_each_possible_cpu(cpu)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002239 counter += per_cpu(sockets_in_use, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 /* It can be negative, by the way. 8) */
2242 if (counter < 0)
2243 counter = 0;
2244
2245 seq_printf(seq, "sockets: used %d\n", counter);
2246}
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002247#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002249#ifdef CONFIG_COMPAT
2250static long compat_sock_ioctl(struct file *file, unsigned cmd,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002251 unsigned long arg)
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002252{
2253 struct socket *sock = file->private_data;
2254 int ret = -ENOIOCTLCMD;
David S. Miller87de87d2008-06-03 09:14:03 -07002255 struct sock *sk;
2256 struct net *net;
2257
2258 sk = sock->sk;
2259 net = sock_net(sk);
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002260
2261 if (sock->ops->compat_ioctl)
2262 ret = sock->ops->compat_ioctl(sock, cmd, arg);
2263
David S. Miller87de87d2008-06-03 09:14:03 -07002264 if (ret == -ENOIOCTLCMD &&
2265 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2266 ret = compat_wext_handle_ioctl(net, cmd, arg);
2267
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002268 return ret;
2269}
2270#endif
2271
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002272int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2273{
2274 return sock->ops->bind(sock, addr, addrlen);
2275}
2276
2277int kernel_listen(struct socket *sock, int backlog)
2278{
2279 return sock->ops->listen(sock, backlog);
2280}
2281
2282int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2283{
2284 struct sock *sk = sock->sk;
2285 int err;
2286
2287 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2288 newsock);
2289 if (err < 0)
2290 goto done;
2291
2292 err = sock->ops->accept(sock, *newsock, flags);
2293 if (err < 0) {
2294 sock_release(*newsock);
Tony Battersbyfa8705b2007-10-10 21:09:04 -07002295 *newsock = NULL;
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002296 goto done;
2297 }
2298
2299 (*newsock)->ops = sock->ops;
Wei Yongjun1b085342008-12-18 19:35:10 -08002300 __module_get((*newsock)->ops->owner);
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002301
2302done:
2303 return err;
2304}
2305
2306int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
YOSHIFUJI Hideaki4768fbc2007-02-09 23:25:31 +09002307 int flags)
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002308{
2309 return sock->ops->connect(sock, addr, addrlen, flags);
2310}
2311
2312int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2313 int *addrlen)
2314{
2315 return sock->ops->getname(sock, addr, addrlen, 0);
2316}
2317
2318int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2319 int *addrlen)
2320{
2321 return sock->ops->getname(sock, addr, addrlen, 1);
2322}
2323
2324int kernel_getsockopt(struct socket *sock, int level, int optname,
2325 char *optval, int *optlen)
2326{
2327 mm_segment_t oldfs = get_fs();
2328 int err;
2329
2330 set_fs(KERNEL_DS);
2331 if (level == SOL_SOCKET)
2332 err = sock_getsockopt(sock, level, optname, optval, optlen);
2333 else
2334 err = sock->ops->getsockopt(sock, level, optname, optval,
2335 optlen);
2336 set_fs(oldfs);
2337 return err;
2338}
2339
2340int kernel_setsockopt(struct socket *sock, int level, int optname,
2341 char *optval, int optlen)
2342{
2343 mm_segment_t oldfs = get_fs();
2344 int err;
2345
2346 set_fs(KERNEL_DS);
2347 if (level == SOL_SOCKET)
2348 err = sock_setsockopt(sock, level, optname, optval, optlen);
2349 else
2350 err = sock->ops->setsockopt(sock, level, optname, optval,
2351 optlen);
2352 set_fs(oldfs);
2353 return err;
2354}
2355
2356int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2357 size_t size, int flags)
2358{
2359 if (sock->ops->sendpage)
2360 return sock->ops->sendpage(sock, page, offset, size, flags);
2361
2362 return sock_no_sendpage(sock, page, offset, size, flags);
2363}
2364
2365int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2366{
2367 mm_segment_t oldfs = get_fs();
2368 int err;
2369
2370 set_fs(KERNEL_DS);
2371 err = sock->ops->ioctl(sock, cmd, arg);
2372 set_fs(oldfs);
2373
2374 return err;
2375}
2376
Trond Myklebust91cf45f2007-11-12 18:10:39 -08002377int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2378{
2379 return sock->ops->shutdown(sock, how);
2380}
2381
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382EXPORT_SYMBOL(sock_create);
2383EXPORT_SYMBOL(sock_create_kern);
2384EXPORT_SYMBOL(sock_create_lite);
2385EXPORT_SYMBOL(sock_map_fd);
2386EXPORT_SYMBOL(sock_recvmsg);
2387EXPORT_SYMBOL(sock_register);
2388EXPORT_SYMBOL(sock_release);
2389EXPORT_SYMBOL(sock_sendmsg);
2390EXPORT_SYMBOL(sock_unregister);
2391EXPORT_SYMBOL(sock_wake_async);
2392EXPORT_SYMBOL(sockfd_lookup);
2393EXPORT_SYMBOL(kernel_sendmsg);
2394EXPORT_SYMBOL(kernel_recvmsg);
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002395EXPORT_SYMBOL(kernel_bind);
2396EXPORT_SYMBOL(kernel_listen);
2397EXPORT_SYMBOL(kernel_accept);
2398EXPORT_SYMBOL(kernel_connect);
2399EXPORT_SYMBOL(kernel_getsockname);
2400EXPORT_SYMBOL(kernel_getpeername);
2401EXPORT_SYMBOL(kernel_getsockopt);
2402EXPORT_SYMBOL(kernel_setsockopt);
2403EXPORT_SYMBOL(kernel_sendpage);
2404EXPORT_SYMBOL(kernel_sock_ioctl);
Trond Myklebust91cf45f2007-11-12 18:10:39 -08002405EXPORT_SYMBOL(kernel_sock_shutdown);