blob: 8ef8ba81b9e2048c93363f4a2f98f73b7778cdce [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>
Ulrich Drepperc019bbc2008-07-23 21:29:21 -070072#include <linux/thread_info.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <linux/wanrouter.h>
74#include <linux/if_bridge.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030075#include <linux/if_frad.h>
76#include <linux/if_vlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#include <linux/init.h>
78#include <linux/poll.h>
79#include <linux/cache.h>
80#include <linux/module.h>
81#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include <linux/mount.h>
83#include <linux/security.h>
84#include <linux/syscalls.h>
85#include <linux/compat.h>
86#include <linux/kmod.h>
David Woodhouse3ec3b2f2005-05-17 12:08:48 +010087#include <linux/audit.h>
Adrian Bunkd86b5e02006-01-21 00:46:55 +010088#include <linux/wireless.h>
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -070089#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91#include <asm/uaccess.h>
92#include <asm/unistd.h>
93
94#include <net/compat.h>
David S. Miller87de87d2008-06-03 09:14:03 -070095#include <net/wext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97#include <net/sock.h>
98#include <linux/netfilter.h>
99
100static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700101static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
102 unsigned long nr_segs, loff_t pos);
103static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
104 unsigned long nr_segs, loff_t pos);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700105static int sock_mmap(struct file *file, struct vm_area_struct *vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107static int sock_close(struct inode *inode, struct file *file);
108static unsigned int sock_poll(struct file *file,
109 struct poll_table_struct *wait);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700110static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800111#ifdef CONFIG_COMPAT
112static long compat_sock_ioctl(struct file *file,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700113 unsigned int cmd, unsigned long arg);
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800114#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115static int sock_fasync(int fd, struct file *filp, int on);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116static ssize_t sock_sendpage(struct file *file, struct page *page,
117 int offset, size_t size, loff_t *ppos, int more);
Jens Axboe9c55e012007-11-06 23:30:13 -0800118static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
119 struct pipe_inode_info *pipe, size_t len,
120 unsigned int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/*
123 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
124 * in the operation structures but are done directly via the socketcall() multiplexor.
125 */
126
Arjan van de Venda7071d2007-02-12 00:55:36 -0800127static const struct file_operations socket_file_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 .owner = THIS_MODULE,
129 .llseek = no_llseek,
130 .aio_read = sock_aio_read,
131 .aio_write = sock_aio_write,
132 .poll = sock_poll,
133 .unlocked_ioctl = sock_ioctl,
Shaun Pereira89bbfc92006-03-21 23:58:08 -0800134#ifdef CONFIG_COMPAT
135 .compat_ioctl = compat_sock_ioctl,
136#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 .mmap = sock_mmap,
138 .open = sock_no_open, /* special open code to disallow open via /proc */
139 .release = sock_close,
140 .fasync = sock_fasync,
Jens Axboe5274f052006-03-30 15:15:30 +0200141 .sendpage = sock_sendpage,
142 .splice_write = generic_splice_sendpage,
Jens Axboe9c55e012007-11-06 23:30:13 -0800143 .splice_read = sock_splice_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144};
145
146/*
147 * The protocol list. Each protocol is registered in here.
148 */
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150static DEFINE_SPINLOCK(net_family_lock);
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -0700151static const struct net_proto_family *net_families[NPROTO] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/*
154 * Statistics counters of the socket lists
155 */
156
157static DEFINE_PER_CPU(int, sockets_in_use) = 0;
158
159/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700160 * Support routines.
161 * Move socket addresses back and forth across the kernel/user
162 * divide and look after the messy bits.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 */
164
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700165#define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 16 for IP, 16 for IPX,
167 24 for IPv6,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700168 about 80 for AX.25
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 must be at least one bigger than
170 the AF_UNIX size (see net/unix/af_unix.c
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700171 :unix_mkname()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/**
175 * move_addr_to_kernel - copy a socket address into kernel space
176 * @uaddr: Address in user space
177 * @kaddr: Address in kernel space
178 * @ulen: Length in user space
179 *
180 * The address is copied into kernel space. If the provided address is
181 * too long an error code of -EINVAL is returned. If the copy gives
182 * invalid addresses -EFAULT is returned. On a success 0 is returned.
183 */
184
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700185int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700187 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700189 if (ulen == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 return 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700191 if (copy_from_user(kaddr, uaddr, ulen))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +0100193 return audit_sockaddr(ulen, kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194}
195
196/**
197 * move_addr_to_user - copy an address to user space
198 * @kaddr: kernel space address
199 * @klen: length of address in kernel
200 * @uaddr: user space address
201 * @ulen: pointer to user length field
202 *
203 * The value pointed to by ulen on entry is the buffer length available.
204 * This is overwritten with the buffer space used. -EINVAL is returned
205 * if an overlong buffer is specified or a negative buffer size. -EFAULT
206 * is returned if either the buffer or the length field are not
207 * accessible.
208 * After copying the data up to the limit the user specifies, the true
209 * length of the data is written over the length limit the user
210 * specified. Zero is returned for a success.
211 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700212
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700213int move_addr_to_user(struct sockaddr *kaddr, int klen, void __user *uaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700214 int __user *ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 int err;
217 int len;
218
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700219 err = get_user(len, ulen);
220 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 return err;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700222 if (len > klen)
223 len = klen;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -0700224 if (len < 0 || len > sizeof(struct sockaddr_storage))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700226 if (len) {
Steve Grubbd6fe3942006-03-30 12:20:22 -0500227 if (audit_sockaddr(klen, kaddr))
228 return -ENOMEM;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700229 if (copy_to_user(uaddr, kaddr, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return -EFAULT;
231 }
232 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700233 * "fromlen shall refer to the value before truncation.."
234 * 1003.1g
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
236 return __put_user(klen, ulen);
237}
238
239#define SOCKFS_MAGIC 0x534F434B
240
Christoph Lametere18b8902006-12-06 20:33:20 -0800241static struct kmem_cache *sock_inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243static struct inode *sock_alloc_inode(struct super_block *sb)
244{
245 struct socket_alloc *ei;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700246
Christoph Lametere94b1762006-12-06 20:33:17 -0800247 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (!ei)
249 return NULL;
250 init_waitqueue_head(&ei->socket.wait);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 ei->socket.fasync_list = NULL;
253 ei->socket.state = SS_UNCONNECTED;
254 ei->socket.flags = 0;
255 ei->socket.ops = NULL;
256 ei->socket.sk = NULL;
257 ei->socket.file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 return &ei->vfs_inode;
260}
261
262static void sock_destroy_inode(struct inode *inode)
263{
264 kmem_cache_free(sock_inode_cachep,
265 container_of(inode, struct socket_alloc, vfs_inode));
266}
267
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700268static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700270 struct socket_alloc *ei = (struct socket_alloc *)foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Christoph Lametera35afb82007-05-16 22:10:57 -0700272 inode_init_once(&ei->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275static int init_inodecache(void)
276{
277 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700278 sizeof(struct socket_alloc),
279 0,
280 (SLAB_HWCACHE_ALIGN |
281 SLAB_RECLAIM_ACCOUNT |
282 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +0900283 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (sock_inode_cachep == NULL)
285 return -ENOMEM;
286 return 0;
287}
288
289static struct super_operations sockfs_ops = {
290 .alloc_inode = sock_alloc_inode,
291 .destroy_inode =sock_destroy_inode,
292 .statfs = simple_statfs,
293};
294
David Howells454e2392006-06-23 02:02:57 -0700295static int sockfs_get_sb(struct file_system_type *fs_type,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700296 int flags, const char *dev_name, void *data,
297 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
David Howells454e2392006-06-23 02:02:57 -0700299 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
300 mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Eric Dumazetba899662005-08-26 12:05:31 -0700303static struct vfsmount *sock_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
305static struct file_system_type sock_fs_type = {
306 .name = "sockfs",
307 .get_sb = sockfs_get_sb,
308 .kill_sb = kill_anon_super,
309};
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311static int sockfs_delete_dentry(struct dentry *dentry)
312{
Eric Dumazet304e61e2006-12-06 20:38:49 -0800313 /*
314 * At creation time, we pretended this dentry was hashed
315 * (by clearing DCACHE_UNHASHED bit in d_flags)
316 * At delete time, we restore the truth : not hashed.
317 * (so that dput() can proceed correctly)
318 */
319 dentry->d_flags |= DCACHE_UNHASHED;
320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700322
323/*
324 * sockfs_dname() is called from d_path().
325 */
326static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
327{
328 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
329 dentry->d_inode->i_ino);
330}
331
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332static struct dentry_operations sockfs_dentry_operations = {
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700333 .d_delete = sockfs_delete_dentry,
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700334 .d_dname = sockfs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335};
336
337/*
338 * Obtains the first available file descriptor and sets it up for use.
339 *
David S. Miller39d8c1b2006-03-20 17:13:49 -0800340 * These functions create file structures and maps them to fd space
341 * of the current process. On success it returns file descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 * and file struct implicitly stored in sock->file.
343 * Note that another thread may close file descriptor before we return
344 * from this function. We use the fact that now we do not refer
345 * to socket after mapping. If one day we will need it, this
346 * function will increment ref. count on file by 1.
347 *
348 * In any case returned fd MAY BE not valid!
349 * This race condition is unavoidable
350 * with shared fd spaces, we cannot solve it inside kernel,
351 * but we take care of internal coherence yet.
352 */
353
Ulrich Dreppera677a032008-07-23 21:29:17 -0700354static int sock_alloc_fd(struct file **filep, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
356 int fd;
David S. Miller39d8c1b2006-03-20 17:13:49 -0800357
Ulrich Dreppera677a032008-07-23 21:29:17 -0700358 fd = get_unused_fd_flags(flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800359 if (likely(fd >= 0)) {
360 struct file *file = get_empty_filp();
361
362 *filep = file;
363 if (unlikely(!file)) {
364 put_unused_fd(fd);
365 return -ENFILE;
366 }
367 } else
368 *filep = NULL;
369 return fd;
370}
371
Ulrich Drepper77d27202008-07-23 21:29:35 -0700372static int sock_attach_fd(struct socket *sock, struct file *file, int flags)
David S. Miller39d8c1b2006-03-20 17:13:49 -0800373{
Dave Hansence8d2cd2007-10-16 23:31:13 -0700374 struct dentry *dentry;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700375 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Dave Hansence8d2cd2007-10-16 23:31:13 -0700377 dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
378 if (unlikely(!dentry))
David S. Miller39d8c1b2006-03-20 17:13:49 -0800379 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Dave Hansence8d2cd2007-10-16 23:31:13 -0700381 dentry->d_op = &sockfs_dentry_operations;
Eric Dumazet304e61e2006-12-06 20:38:49 -0800382 /*
383 * We dont want to push this dentry into global dentry hash table.
384 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
385 * This permits a working /proc/$pid/fd/XXX on sockets
386 */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700387 dentry->d_flags &= ~DCACHE_UNHASHED;
388 d_instantiate(dentry, SOCK_INODE(sock));
David S. Miller39d8c1b2006-03-20 17:13:49 -0800389
390 sock->file = file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700391 init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE,
392 &socket_file_ops);
393 SOCK_INODE(sock)->i_fop = &socket_file_ops;
Ulrich Drepper77d27202008-07-23 21:29:35 -0700394 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800395 file->f_pos = 0;
396 file->private_data = sock;
397
398 return 0;
399}
400
Ulrich Dreppera677a032008-07-23 21:29:17 -0700401int sock_map_fd(struct socket *sock, int flags)
David S. Miller39d8c1b2006-03-20 17:13:49 -0800402{
403 struct file *newfile;
Ulrich Dreppera677a032008-07-23 21:29:17 -0700404 int fd = sock_alloc_fd(&newfile, flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800405
406 if (likely(fd >= 0)) {
Ulrich Drepper77d27202008-07-23 21:29:35 -0700407 int err = sock_attach_fd(sock, newfile, flags);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800408
409 if (unlikely(err < 0)) {
410 put_filp(newfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 put_unused_fd(fd);
David S. Miller39d8c1b2006-03-20 17:13:49 -0800412 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
David S. Miller39d8c1b2006-03-20 17:13:49 -0800414 fd_install(fd, newfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 return fd;
417}
418
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800419static struct socket *sock_from_file(struct file *file, int *err)
420{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800421 if (file->f_op == &socket_file_ops)
422 return file->private_data; /* set in sock_map_fd */
423
Eric Dumazet23bb80d2007-02-08 14:59:57 -0800424 *err = -ENOTSOCK;
425 return NULL;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800426}
427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428/**
429 * sockfd_lookup - Go from a file number to its socket slot
430 * @fd: file handle
431 * @err: pointer to an error code return
432 *
433 * The file handle passed in is locked and the socket it is bound
434 * too is returned. If an error occurs the err pointer is overwritten
435 * with a negative errno code and NULL is returned. The function checks
436 * for both invalid handles and passing a handle which is not a socket.
437 *
438 * On a success the socket object pointer is returned.
439 */
440
441struct socket *sockfd_lookup(int fd, int *err)
442{
443 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 struct socket *sock;
445
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700446 file = fget(fd);
447 if (!file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 *err = -EBADF;
449 return NULL;
450 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700451
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800452 sock = sock_from_file(file, err);
453 if (!sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 fput(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 return sock;
456}
457
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800458static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
459{
460 struct file *file;
461 struct socket *sock;
462
Hua Zhong36725582006-04-19 15:25:02 -0700463 *err = -EBADF;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -0800464 file = fget_light(fd, fput_needed);
465 if (file) {
466 sock = sock_from_file(file, err);
467 if (sock)
468 return sock;
469 fput_light(file, *fput_needed);
470 }
471 return NULL;
472}
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/**
475 * sock_alloc - allocate a socket
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700476 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 * Allocate a new inode and socket object. The two are bound together
478 * and initialised. The socket is then returned. If we are out of inodes
479 * NULL is returned.
480 */
481
482static struct socket *sock_alloc(void)
483{
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700484 struct inode *inode;
485 struct socket *sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 inode = new_inode(sock_mnt->mnt_sb);
488 if (!inode)
489 return NULL;
490
491 sock = SOCKET_I(inode);
492
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700493 inode->i_mode = S_IFSOCK | S_IRWXUGO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 inode->i_uid = current->fsuid;
495 inode->i_gid = current->fsgid;
496
497 get_cpu_var(sockets_in_use)++;
498 put_cpu_var(sockets_in_use);
499 return sock;
500}
501
502/*
503 * In theory you can't get an open on this inode, but /proc provides
504 * a back door. Remember to keep it shut otherwise you'll let the
505 * creepy crawlies in.
506 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
509{
510 return -ENXIO;
511}
512
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800513const struct file_operations bad_sock_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 .owner = THIS_MODULE,
515 .open = sock_no_open,
516};
517
518/**
519 * sock_release - close a socket
520 * @sock: socket to close
521 *
522 * The socket is released from the protocol stack if it has a release
523 * callback, and the inode is then released if the socket is bound to
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700524 * an inode not a file.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527void sock_release(struct socket *sock)
528{
529 if (sock->ops) {
530 struct module *owner = sock->ops->owner;
531
532 sock->ops->release(sock);
533 sock->ops = NULL;
534 module_put(owner);
535 }
536
537 if (sock->fasync_list)
538 printk(KERN_ERR "sock_release: fasync list not empty!\n");
539
540 get_cpu_var(sockets_in_use)--;
541 put_cpu_var(sockets_in_use);
542 if (!sock->file) {
543 iput(SOCK_INODE(sock));
544 return;
545 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700546 sock->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700549static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 struct msghdr *msg, size_t size)
551{
552 struct sock_iocb *si = kiocb_to_siocb(iocb);
553 int err;
554
555 si->sock = sock;
556 si->scm = NULL;
557 si->msg = msg;
558 si->size = size;
559
560 err = security_socket_sendmsg(sock, msg, size);
561 if (err)
562 return err;
563
564 return sock->ops->sendmsg(iocb, sock, msg, size);
565}
566
567int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
568{
569 struct kiocb iocb;
570 struct sock_iocb siocb;
571 int ret;
572
573 init_sync_kiocb(&iocb, NULL);
574 iocb.private = &siocb;
575 ret = __sock_sendmsg(&iocb, sock, msg, size);
576 if (-EIOCBQUEUED == ret)
577 ret = wait_on_sync_kiocb(&iocb);
578 return ret;
579}
580
581int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
582 struct kvec *vec, size_t num, size_t size)
583{
584 mm_segment_t oldfs = get_fs();
585 int result;
586
587 set_fs(KERNEL_DS);
588 /*
589 * the following is safe, since for compiler definitions of kvec and
590 * iovec are identical, yielding the same in-core layout and alignment
591 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700592 msg->msg_iov = (struct iovec *)vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 msg->msg_iovlen = num;
594 result = sock_sendmsg(sock, msg, size);
595 set_fs(oldfs);
596 return result;
597}
598
Eric Dumazet92f37fd2007-03-25 22:14:49 -0700599/*
600 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
601 */
602void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
603 struct sk_buff *skb)
604{
605 ktime_t kt = skb->tstamp;
606
607 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
608 struct timeval tv;
609 /* Race occurred between timestamp enabling and packet
610 receiving. Fill in the current time for now. */
611 if (kt.tv64 == 0)
612 kt = ktime_get_real();
613 skb->tstamp = kt;
614 tv = ktime_to_timeval(kt);
615 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, sizeof(tv), &tv);
616 } else {
617 struct timespec ts;
618 /* Race occurred between timestamp enabling and packet
619 receiving. Fill in the current time for now. */
620 if (kt.tv64 == 0)
621 kt = ktime_get_real();
622 skb->tstamp = kt;
623 ts = ktime_to_timespec(kt);
624 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, sizeof(ts), &ts);
625 }
626}
627
Arnaldo Carvalho de Melo7c81fd82007-03-10 00:39:35 -0300628EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
629
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700630static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 struct msghdr *msg, size_t size, int flags)
632{
633 int err;
634 struct sock_iocb *si = kiocb_to_siocb(iocb);
635
636 si->sock = sock;
637 si->scm = NULL;
638 si->msg = msg;
639 si->size = size;
640 si->flags = flags;
641
642 err = security_socket_recvmsg(sock, msg, size, flags);
643 if (err)
644 return err;
645
646 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
647}
648
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700649int sock_recvmsg(struct socket *sock, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 size_t size, int flags)
651{
652 struct kiocb iocb;
653 struct sock_iocb siocb;
654 int ret;
655
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700656 init_sync_kiocb(&iocb, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 iocb.private = &siocb;
658 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
659 if (-EIOCBQUEUED == ret)
660 ret = wait_on_sync_kiocb(&iocb);
661 return ret;
662}
663
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700664int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
665 struct kvec *vec, size_t num, size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 mm_segment_t oldfs = get_fs();
668 int result;
669
670 set_fs(KERNEL_DS);
671 /*
672 * the following is safe, since for compiler definitions of kvec and
673 * iovec are identical, yielding the same in-core layout and alignment
674 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700675 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 result = sock_recvmsg(sock, msg, size, flags);
677 set_fs(oldfs);
678 return result;
679}
680
681static void sock_aio_dtor(struct kiocb *iocb)
682{
683 kfree(iocb->private);
684}
685
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300686static ssize_t sock_sendpage(struct file *file, struct page *page,
687 int offset, size_t size, loff_t *ppos, int more)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
689 struct socket *sock;
690 int flags;
691
Eric Dumazetb69aee02005-09-06 14:42:45 -0700692 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
695 if (more)
696 flags |= MSG_MORE;
697
698 return sock->ops->sendpage(sock, page, offset, size, flags);
699}
700
Jens Axboe9c55e012007-11-06 23:30:13 -0800701static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
702 struct pipe_inode_info *pipe, size_t len,
703 unsigned int flags)
704{
705 struct socket *sock = file->private_data;
706
Rémi Denis-Courmont997b37d2008-02-15 02:35:45 -0800707 if (unlikely(!sock->ops->splice_read))
708 return -EINVAL;
709
Jens Axboe9c55e012007-11-06 23:30:13 -0800710 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
711}
712
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800713static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700714 struct sock_iocb *siocb)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800715{
716 if (!is_sync_kiocb(iocb)) {
717 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
718 if (!siocb)
719 return NULL;
720 iocb->ki_dtor = sock_aio_dtor;
721 }
722
723 siocb->kiocb = iocb;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800724 iocb->private = siocb;
725 return siocb;
726}
727
728static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -0700729 struct file *file, const struct iovec *iov,
730 unsigned long nr_segs)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800731{
732 struct socket *sock = file->private_data;
733 size_t size = 0;
734 int i;
735
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700736 for (i = 0; i < nr_segs; i++)
737 size += iov[i].iov_len;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800738
739 msg->msg_name = NULL;
740 msg->msg_namelen = 0;
741 msg->msg_control = NULL;
742 msg->msg_controllen = 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700743 msg->msg_iov = (struct iovec *)iov;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800744 msg->msg_iovlen = nr_segs;
745 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
746
747 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
748}
749
Badari Pulavarty027445c2006-09-30 23:28:46 -0700750static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
751 unsigned long nr_segs, loff_t pos)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800752{
753 struct sock_iocb siocb, *x;
754
755 if (pos != 0)
756 return -ESPIPE;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700757
758 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800759 return 0;
760
Badari Pulavarty027445c2006-09-30 23:28:46 -0700761
762 x = alloc_sock_iocb(iocb, &siocb);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800763 if (!x)
764 return -ENOMEM;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700765 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800766}
767
768static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -0700769 struct file *file, const struct iovec *iov,
770 unsigned long nr_segs)
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800771{
772 struct socket *sock = file->private_data;
773 size_t size = 0;
774 int i;
775
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700776 for (i = 0; i < nr_segs; i++)
777 size += iov[i].iov_len;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800778
779 msg->msg_name = NULL;
780 msg->msg_namelen = 0;
781 msg->msg_control = NULL;
782 msg->msg_controllen = 0;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700783 msg->msg_iov = (struct iovec *)iov;
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800784 msg->msg_iovlen = nr_segs;
785 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
786 if (sock->type == SOCK_SEQPACKET)
787 msg->msg_flags |= MSG_EOR;
788
789 return __sock_sendmsg(iocb, sock, msg, size);
790}
791
Badari Pulavarty027445c2006-09-30 23:28:46 -0700792static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
793 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800795 struct sock_iocb siocb, *x;
796
797 if (pos != 0)
798 return -ESPIPE;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700799
Badari Pulavarty027445c2006-09-30 23:28:46 -0700800 x = alloc_sock_iocb(iocb, &siocb);
Christoph Hellwigce1d4d32005-12-22 21:08:46 -0800801 if (!x)
802 return -ENOMEM;
803
Badari Pulavarty027445c2006-09-30 23:28:46 -0700804 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805}
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807/*
808 * Atomic setting of ioctl hooks to avoid race
809 * with module unload.
810 */
811
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800812static DEFINE_MUTEX(br_ioctl_mutex);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700813static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Eric W. Biederman881d9662007-09-17 11:56:21 -0700815void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800817 mutex_lock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 br_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800819 mutex_unlock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822EXPORT_SYMBOL(brioctl_set);
823
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800824static DEFINE_MUTEX(vlan_ioctl_mutex);
Eric W. Biederman881d9662007-09-17 11:56:21 -0700825static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Eric W. Biederman881d9662007-09-17 11:56:21 -0700827void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800829 mutex_lock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 vlan_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800831 mutex_unlock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834EXPORT_SYMBOL(vlan_ioctl_set);
835
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800836static DEFINE_MUTEX(dlci_ioctl_mutex);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700837static int (*dlci_ioctl_hook) (unsigned int, void __user *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700839void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800841 mutex_lock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 dlci_ioctl_hook = hook;
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800843 mutex_unlock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844}
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846EXPORT_SYMBOL(dlci_ioctl_set);
847
848/*
849 * With an ioctl, arg may well be a user mode pointer, but we don't know
850 * what to do with it - that's up to the protocol still.
851 */
852
853static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
854{
855 struct socket *sock;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700856 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 void __user *argp = (void __user *)arg;
858 int pid, err;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700859 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Eric Dumazetb69aee02005-09-06 14:42:45 -0700861 sock = file->private_data;
Eric W. Biederman881d9662007-09-17 11:56:21 -0700862 sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900863 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
Eric W. Biederman881d9662007-09-17 11:56:21 -0700865 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 } else
Adrian Bunkd86b5e02006-01-21 00:46:55 +0100867#ifdef CONFIG_WIRELESS_EXT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
Eric W. Biederman881d9662007-09-17 11:56:21 -0700869 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 } else
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700871#endif /* CONFIG_WIRELESS_EXT */
872 switch (cmd) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 case FIOSETOWN:
874 case SIOCSPGRP:
875 err = -EFAULT;
876 if (get_user(pid, (int __user *)argp))
877 break;
878 err = f_setown(sock->file, pid, 1);
879 break;
880 case FIOGETOWN:
881 case SIOCGPGRP:
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700882 err = put_user(f_getown(sock->file),
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700883 (int __user *)argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 break;
885 case SIOCGIFBR:
886 case SIOCSIFBR:
887 case SIOCBRADDBR:
888 case SIOCBRDELBR:
889 err = -ENOPKG;
890 if (!br_ioctl_hook)
891 request_module("bridge");
892
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800893 mutex_lock(&br_ioctl_mutex);
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700894 if (br_ioctl_hook)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700895 err = br_ioctl_hook(net, cmd, argp);
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800896 mutex_unlock(&br_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 break;
898 case SIOCGIFVLAN:
899 case SIOCSIFVLAN:
900 err = -ENOPKG;
901 if (!vlan_ioctl_hook)
902 request_module("8021q");
903
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800904 mutex_lock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (vlan_ioctl_hook)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700906 err = vlan_ioctl_hook(net, argp);
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -0800907 mutex_unlock(&vlan_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 case SIOCADDDLCI:
910 case SIOCDELDLCI:
911 err = -ENOPKG;
912 if (!dlci_ioctl_hook)
913 request_module("dlci");
914
Pavel Emelyanov7512cbf2008-03-21 15:58:52 -0700915 mutex_lock(&dlci_ioctl_mutex);
916 if (dlci_ioctl_hook)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 err = dlci_ioctl_hook(cmd, argp);
Pavel Emelyanov7512cbf2008-03-21 15:58:52 -0700918 mutex_unlock(&dlci_ioctl_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 break;
920 default:
921 err = sock->ops->ioctl(sock, cmd, arg);
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -0800922
923 /*
924 * If this ioctl is unknown try to hand it down
925 * to the NIC driver.
926 */
927 if (err == -ENOIOCTLCMD)
Eric W. Biederman881d9662007-09-17 11:56:21 -0700928 err = dev_ioctl(net, cmd, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 break;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 return err;
932}
933
934int sock_create_lite(int family, int type, int protocol, struct socket **res)
935{
936 int err;
937 struct socket *sock = NULL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 err = security_socket_create(family, type, protocol, 1);
940 if (err)
941 goto out;
942
943 sock = sock_alloc();
944 if (!sock) {
945 err = -ENOMEM;
946 goto out;
947 }
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 sock->type = type;
Venkat Yekkirala7420ed22006-08-04 23:17:57 -0700950 err = security_socket_post_create(sock, family, type, protocol, 1);
951 if (err)
952 goto out_release;
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954out:
955 *res = sock;
956 return err;
Venkat Yekkirala7420ed22006-08-04 23:17:57 -0700957out_release:
958 sock_release(sock);
959 sock = NULL;
960 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
962
963/* No kernel lock held - perfect */
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700964static unsigned int sock_poll(struct file *file, poll_table *wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
966 struct socket *sock;
967
968 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700969 * We can't return errors to poll, so it's either yes or no.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 */
Eric Dumazetb69aee02005-09-06 14:42:45 -0700971 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return sock->ops->poll(file, sock, wait);
973}
974
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700975static int sock_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Eric Dumazetb69aee02005-09-06 14:42:45 -0700977 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 return sock->ops->mmap(file, sock, vma);
980}
981
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300982static int sock_close(struct inode *inode, struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
984 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700985 * It was possible the inode is NULL we were
986 * closing an unfinished socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 */
988
Stephen Hemminger89bddce2006-09-01 00:19:31 -0700989 if (!inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 printk(KERN_DEBUG "sock_close: NULL inode\n");
991 return 0;
992 }
993 sock_fasync(-1, filp, 0);
994 sock_release(SOCKET_I(inode));
995 return 0;
996}
997
998/*
999 * Update the socket async list
1000 *
1001 * Fasync_list locking strategy.
1002 *
1003 * 1. fasync_list is modified only under process context socket lock
1004 * i.e. under semaphore.
1005 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1006 * or under socket lock.
1007 * 3. fasync_list can be used from softirq context, so that
1008 * modification under socket lock have to be enhanced with
1009 * write_lock_bh(&sk->sk_callback_lock).
1010 * --ANK (990710)
1011 */
1012
1013static int sock_fasync(int fd, struct file *filp, int on)
1014{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001015 struct fasync_struct *fa, *fna = NULL, **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 struct socket *sock;
1017 struct sock *sk;
1018
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001019 if (on) {
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08001020 fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001021 if (fna == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 return -ENOMEM;
1023 }
1024
Eric Dumazetb69aee02005-09-06 14:42:45 -07001025 sock = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001027 sk = sock->sk;
1028 if (sk == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 kfree(fna);
1030 return -EINVAL;
1031 }
1032
1033 lock_sock(sk);
1034
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001035 prev = &(sock->fasync_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001037 for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1038 if (fa->fa_file == filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 break;
1040
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001041 if (on) {
1042 if (fa != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001044 fa->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 write_unlock_bh(&sk->sk_callback_lock);
1046
1047 kfree(fna);
1048 goto out;
1049 }
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001050 fna->fa_file = filp;
1051 fna->fa_fd = fd;
1052 fna->magic = FASYNC_MAGIC;
1053 fna->fa_next = sock->fasync_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001055 sock->fasync_list = fna;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 write_unlock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001057 } else {
1058 if (fa != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 write_lock_bh(&sk->sk_callback_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001060 *prev = fa->fa_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 write_unlock_bh(&sk->sk_callback_lock);
1062 kfree(fa);
1063 }
1064 }
1065
1066out:
1067 release_sock(sock->sk);
1068 return 0;
1069}
1070
1071/* This function may be called only under socket lock or callback_lock */
1072
1073int sock_wake_async(struct socket *sock, int how, int band)
1074{
1075 if (!sock || !sock->fasync_list)
1076 return -1;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001077 switch (how) {
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001078 case SOCK_WAKE_WAITD:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1080 break;
1081 goto call_kill;
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001082 case SOCK_WAKE_SPACE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1084 break;
1085 /* fall through */
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001086 case SOCK_WAKE_IO:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001087call_kill:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 __kill_fasync(sock->fasync_list, SIGIO, band);
1089 break;
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +08001090 case SOCK_WAKE_URG:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 __kill_fasync(sock->fasync_list, SIGURG, band);
1092 }
1093 return 0;
1094}
1095
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001096static int __sock_create(struct net *net, int family, int type, int protocol,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001097 struct socket **res, int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098{
1099 int err;
1100 struct socket *sock;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001101 const struct net_proto_family *pf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001104 * Check protocol is in range
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 */
1106 if (family < 0 || family >= NPROTO)
1107 return -EAFNOSUPPORT;
1108 if (type < 0 || type >= SOCK_MAX)
1109 return -EINVAL;
1110
1111 /* Compatibility.
1112
1113 This uglymoron is moved from INET layer to here to avoid
1114 deadlock in module load.
1115 */
1116 if (family == PF_INET && type == SOCK_PACKET) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001117 static int warned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if (!warned) {
1119 warned = 1;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001120 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1121 current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
1123 family = PF_PACKET;
1124 }
1125
1126 err = security_socket_create(family, type, protocol, kern);
1127 if (err)
1128 return err;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001129
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001130 /*
1131 * Allocate the socket and allow the family to set things up. if
1132 * the protocol is 0, the family is instructed to select an appropriate
1133 * default.
1134 */
1135 sock = sock_alloc();
1136 if (!sock) {
1137 if (net_ratelimit())
1138 printk(KERN_WARNING "socket: no more sockets\n");
1139 return -ENFILE; /* Not exactly a match, but its the
1140 closest posix thing */
1141 }
1142
1143 sock->type = type;
1144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145#if defined(CONFIG_KMOD)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001146 /* Attempt to load a protocol module if the find failed.
1147 *
1148 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 * requested real, full-featured networking support upon configuration.
1150 * Otherwise module support will break!
1151 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001152 if (net_families[family] == NULL)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001153 request_module("net-pf-%d", family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154#endif
1155
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001156 rcu_read_lock();
1157 pf = rcu_dereference(net_families[family]);
1158 err = -EAFNOSUPPORT;
1159 if (!pf)
1160 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
1162 /*
1163 * We will call the ->create function, that possibly is in a loadable
1164 * module, so we have to bump that loadable module refcnt first.
1165 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001166 if (!try_module_get(pf->owner))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 goto out_release;
1168
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001169 /* Now protected by module ref count */
1170 rcu_read_unlock();
1171
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001172 err = pf->create(net, sock, protocol);
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001173 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 goto out_module_put;
Frank Filza79af592005-09-27 15:23:38 -07001175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 /*
1177 * Now to bump the refcnt of the [loadable] module that owns this
1178 * socket at sock_release time we decrement its refcnt.
1179 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001180 if (!try_module_get(sock->ops->owner))
1181 goto out_module_busy;
1182
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 /*
1184 * Now that we're done with the ->create function, the [loadable]
1185 * module can have its refcnt decremented
1186 */
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001187 module_put(pf->owner);
Venkat Yekkirala7420ed22006-08-04 23:17:57 -07001188 err = security_socket_post_create(sock, family, type, protocol, kern);
1189 if (err)
Herbert Xu3b185522007-08-15 14:46:02 -07001190 goto out_sock_release;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001191 *res = sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001193 return 0;
1194
1195out_module_busy:
1196 err = -EAFNOSUPPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197out_module_put:
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001198 sock->ops = NULL;
1199 module_put(pf->owner);
1200out_sock_release:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 sock_release(sock);
Stephen Hemminger55737fd2006-09-01 00:23:39 -07001202 return err;
1203
1204out_release:
1205 rcu_read_unlock();
1206 goto out_sock_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209int sock_create(int family, int type, int protocol, struct socket **res)
1210{
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001211 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212}
1213
1214int sock_create_kern(int family, int type, int protocol, struct socket **res)
1215{
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -07001216 return __sock_create(&init_net, family, type, protocol, res, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
1219asmlinkage long sys_socket(int family, int type, int protocol)
1220{
1221 int retval;
1222 struct socket *sock;
Ulrich Dreppera677a032008-07-23 21:29:17 -07001223 int flags;
1224
Ulrich Dreppere38b36f2008-07-23 21:29:42 -07001225 /* Check the SOCK_* constants for consistency. */
1226 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1227 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1228 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1229 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1230
Ulrich Dreppera677a032008-07-23 21:29:17 -07001231 flags = type & ~SOCK_TYPE_MASK;
Ulrich Drepper77d27202008-07-23 21:29:35 -07001232 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Dreppera677a032008-07-23 21:29:17 -07001233 return -EINVAL;
1234 type &= SOCK_TYPE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001236 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1237 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 retval = sock_create(family, type, protocol, &sock);
1240 if (retval < 0)
1241 goto out;
1242
Ulrich Drepper77d27202008-07-23 21:29:35 -07001243 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (retval < 0)
1245 goto out_release;
1246
1247out:
1248 /* It may be already another descriptor 8) Not kernel problem. */
1249 return retval;
1250
1251out_release:
1252 sock_release(sock);
1253 return retval;
1254}
1255
1256/*
1257 * Create a pair of connected sockets.
1258 */
1259
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001260asmlinkage long sys_socketpair(int family, int type, int protocol,
1261 int __user *usockvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262{
1263 struct socket *sock1, *sock2;
1264 int fd1, fd2, err;
Al Virodb349502007-02-07 01:48:00 -05001265 struct file *newfile1, *newfile2;
Ulrich Dreppera677a032008-07-23 21:29:17 -07001266 int flags;
1267
1268 flags = type & ~SOCK_TYPE_MASK;
Ulrich Drepper77d27202008-07-23 21:29:35 -07001269 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Dreppera677a032008-07-23 21:29:17 -07001270 return -EINVAL;
1271 type &= SOCK_TYPE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001273 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1274 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 /*
1277 * Obtain the first socket and check if the underlying protocol
1278 * supports the socketpair call.
1279 */
1280
1281 err = sock_create(family, type, protocol, &sock1);
1282 if (err < 0)
1283 goto out;
1284
1285 err = sock_create(family, type, protocol, &sock2);
1286 if (err < 0)
1287 goto out_release_1;
1288
1289 err = sock1->ops->socketpair(sock1, sock2);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001290 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 goto out_release_both;
1292
Ulrich Dreppera677a032008-07-23 21:29:17 -07001293 fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
David S. Millerbf3c23d2007-10-29 21:54:02 -07001294 if (unlikely(fd1 < 0)) {
1295 err = fd1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 goto out_release_both;
David S. Millerbf3c23d2007-10-29 21:54:02 -07001297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Ulrich Dreppera677a032008-07-23 21:29:17 -07001299 fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
Al Virodb349502007-02-07 01:48:00 -05001300 if (unlikely(fd2 < 0)) {
David S. Millerbf3c23d2007-10-29 21:54:02 -07001301 err = fd2;
Al Virodb349502007-02-07 01:48:00 -05001302 put_filp(newfile1);
1303 put_unused_fd(fd1);
1304 goto out_release_both;
1305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Ulrich Drepper77d27202008-07-23 21:29:35 -07001307 err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
Al Virodb349502007-02-07 01:48:00 -05001308 if (unlikely(err < 0)) {
1309 goto out_fd2;
1310 }
1311
Ulrich Drepper77d27202008-07-23 21:29:35 -07001312 err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
Al Virodb349502007-02-07 01:48:00 -05001313 if (unlikely(err < 0)) {
1314 fput(newfile1);
1315 goto out_fd1;
1316 }
1317
1318 err = audit_fd_pair(fd1, fd2);
1319 if (err < 0) {
1320 fput(newfile1);
1321 fput(newfile2);
1322 goto out_fd;
1323 }
1324
1325 fd_install(fd1, newfile1);
1326 fd_install(fd2, newfile2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 /* fd1 and fd2 may be already another descriptors.
1328 * Not kernel problem.
1329 */
1330
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001331 err = put_user(fd1, &usockvec[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 if (!err)
1333 err = put_user(fd2, &usockvec[1]);
1334 if (!err)
1335 return 0;
1336
1337 sys_close(fd2);
1338 sys_close(fd1);
1339 return err;
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341out_release_both:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001342 sock_release(sock2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343out_release_1:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001344 sock_release(sock1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345out:
1346 return err;
Al Virodb349502007-02-07 01:48:00 -05001347
1348out_fd2:
1349 put_filp(newfile1);
1350 sock_release(sock1);
1351out_fd1:
1352 put_filp(newfile2);
1353 sock_release(sock2);
1354out_fd:
1355 put_unused_fd(fd1);
1356 put_unused_fd(fd2);
1357 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358}
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/*
1361 * Bind a name to a socket. Nothing much to do here since it's
1362 * the protocol's responsibility to handle the local address.
1363 *
1364 * We move the socket address to kernel space before we call
1365 * the protocol layer (having also checked the address is ok).
1366 */
1367
1368asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1369{
1370 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001371 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001372 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001374 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Stephen Hemmingere71a4782007-04-10 20:10:33 -07001375 if (sock) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001376 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001377 if (err >= 0) {
1378 err = security_socket_bind(sock,
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001379 (struct sockaddr *)&address,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001380 addrlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001381 if (!err)
1382 err = sock->ops->bind(sock,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001383 (struct sockaddr *)
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001384 &address, addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 }
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001386 fput_light(sock->file, fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return err;
1389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391/*
1392 * Perform a listen. Basically, we allow the protocol to do anything
1393 * necessary for a listen, and if that works, we mark the socket as
1394 * ready for listening.
1395 */
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397asmlinkage long sys_listen(int fd, int backlog)
1398{
1399 struct socket *sock;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001400 int err, fput_needed;
Pavel Emelyanovb8e1f9b2007-12-08 00:12:33 -08001401 int somaxconn;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001402
1403 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1404 if (sock) {
Pavel Emelyanov8efa6e92008-03-31 19:41:14 -07001405 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
Pavel Emelyanovb8e1f9b2007-12-08 00:12:33 -08001406 if ((unsigned)backlog > somaxconn)
1407 backlog = somaxconn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 err = security_socket_listen(sock, backlog);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001410 if (!err)
1411 err = sock->ops->listen(sock, backlog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001413 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 }
1415 return err;
1416}
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418/*
1419 * For accept, we attempt to create a new socket, set up the link
1420 * with the client, wake up the client, then return the new
1421 * connected fd. We collect the address of the connector in kernel
1422 * space and move it to user at the very end. This is unclean because
1423 * we open the socket then return an error.
1424 *
1425 * 1003.1g adds the ability to recvmsg() to query connection pending
1426 * status to recvmsg. We need to add that support in a way thats
1427 * clean when we restucture accept also.
1428 */
1429
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001430long do_accept(int fd, struct sockaddr __user *upeer_sockaddr,
1431 int __user *upeer_addrlen, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
1433 struct socket *sock, *newsock;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001434 struct file *newfile;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001435 int err, len, newfd, fput_needed;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001436 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Ulrich Drepper77d27202008-07-23 21:29:35 -07001438 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001439 return -EINVAL;
1440
1441 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1442 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1443
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001444 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 if (!sock)
1446 goto out;
1447
1448 err = -ENFILE;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001449 if (!(newsock = sock_alloc()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 goto out_put;
1451
1452 newsock->type = sock->type;
1453 newsock->ops = sock->ops;
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 /*
1456 * We don't need try_module_get here, as the listening socket (sock)
1457 * has the protocol module (sock->ops->owner) held.
1458 */
1459 __module_get(newsock->ops->owner);
1460
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001461 newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001462 if (unlikely(newfd < 0)) {
1463 err = newfd;
David S. Miller9a1875e2006-04-01 12:48:36 -08001464 sock_release(newsock);
1465 goto out_put;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001466 }
1467
Ulrich Drepper77d27202008-07-23 21:29:35 -07001468 err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001469 if (err < 0)
Alexey Dobriyan79f4f642007-03-26 14:09:52 -07001470 goto out_fd_simple;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001471
Frank Filza79af592005-09-27 15:23:38 -07001472 err = security_socket_accept(sock, newsock);
1473 if (err)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001474 goto out_fd;
Frank Filza79af592005-09-27 15:23:38 -07001475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1477 if (err < 0)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001478 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
1480 if (upeer_sockaddr) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001481 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001482 &len, 2) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 err = -ECONNABORTED;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001484 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001486 err = move_addr_to_user((struct sockaddr *)&address,
1487 len, upeer_sockaddr, upeer_addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 if (err < 0)
David S. Miller39d8c1b2006-03-20 17:13:49 -08001489 goto out_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
1491
1492 /* File flags are not inherited via accept() unlike another OSes. */
1493
David S. Miller39d8c1b2006-03-20 17:13:49 -08001494 fd_install(newfd, newfile);
1495 err = newfd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497 security_socket_post_accept(sock, newsock);
1498
1499out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001500 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501out:
1502 return err;
Alexey Dobriyan79f4f642007-03-26 14:09:52 -07001503out_fd_simple:
1504 sock_release(newsock);
1505 put_filp(newfile);
1506 put_unused_fd(newfd);
1507 goto out_put;
David S. Miller39d8c1b2006-03-20 17:13:49 -08001508out_fd:
David S. Miller9606a212006-04-01 01:00:14 -08001509 fput(newfile);
David S. Miller39d8c1b2006-03-20 17:13:49 -08001510 put_unused_fd(newfd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 goto out_put;
1512}
1513
Ulrich Drepperc019bbc2008-07-23 21:29:21 -07001514#ifdef HAVE_SET_RESTORE_SIGMASK
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001515asmlinkage long sys_paccept(int fd, struct sockaddr __user *upeer_sockaddr,
1516 int __user *upeer_addrlen,
1517 const sigset_t __user *sigmask,
1518 size_t sigsetsize, int flags)
1519{
1520 sigset_t ksigmask, sigsaved;
1521 int ret;
1522
1523 if (sigmask) {
1524 /* XXX: Don't preclude handling different sized sigset_t's. */
1525 if (sigsetsize != sizeof(sigset_t))
1526 return -EINVAL;
1527 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
1528 return -EFAULT;
1529
1530 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1531 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1532 }
1533
1534 ret = do_accept(fd, upeer_sockaddr, upeer_addrlen, flags);
1535
1536 if (ret < 0 && signal_pending(current)) {
1537 /*
1538 * Don't restore the signal mask yet. Let do_signal() deliver
1539 * the signal on the way back to userspace, before the signal
1540 * mask is restored.
1541 */
1542 if (sigmask) {
1543 memcpy(&current->saved_sigmask, &sigsaved,
1544 sizeof(sigsaved));
1545 set_restore_sigmask();
1546 }
1547 } else if (sigmask)
1548 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1549
1550 return ret;
1551}
Ulrich Drepperc019bbc2008-07-23 21:29:21 -07001552#else
1553asmlinkage long sys_paccept(int fd, struct sockaddr __user *upeer_sockaddr,
1554 int __user *upeer_addrlen,
1555 const sigset_t __user *sigmask,
1556 size_t sigsetsize, int flags)
1557{
1558 /* The platform does not support restoring the signal mask in the
1559 * return path. So we do not allow using paccept() with a signal
1560 * mask. */
1561 if (sigmask)
1562 return -EINVAL;
1563
1564 return do_accept(fd, upeer_sockaddr, upeer_addrlen, flags);
1565}
1566#endif
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07001567
1568asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr,
1569 int __user *upeer_addrlen)
1570{
1571 return do_accept(fd, upeer_sockaddr, upeer_addrlen, 0);
1572}
1573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574/*
1575 * Attempt to connect to a socket with the server address. The address
1576 * is in user space so we verify it is OK and move it to kernel space.
1577 *
1578 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1579 * break bindings
1580 *
1581 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1582 * other SEQPACKET protocols that take time to connect() as it doesn't
1583 * include the -EINPROGRESS status for such sockets.
1584 */
1585
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001586asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr,
1587 int addrlen)
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 err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001593 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (!sock)
1595 goto out;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001596 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if (err < 0)
1598 goto out_put;
1599
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001600 err =
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001601 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 if (err)
1603 goto out_put;
1604
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001605 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 sock->file->f_flags);
1607out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001608 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609out:
1610 return err;
1611}
1612
1613/*
1614 * Get the local address ('name') of a socket object. Move the obtained
1615 * name to user space.
1616 */
1617
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001618asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1619 int __user *usockaddr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620{
1621 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001622 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001623 int len, err, fput_needed;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001624
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001625 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 if (!sock)
1627 goto out;
1628
1629 err = security_socket_getsockname(sock);
1630 if (err)
1631 goto out_put;
1632
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001633 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 if (err)
1635 goto out_put;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001636 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
1638out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001639 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640out:
1641 return err;
1642}
1643
1644/*
1645 * Get the remote address ('name') of a socket object. Move the obtained
1646 * name to user space.
1647 */
1648
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001649asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1650 int __user *usockaddr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
1652 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001653 struct sockaddr_storage address;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001654 int len, err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001656 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1657 if (sock != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 err = security_socket_getpeername(sock);
1659 if (err) {
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001660 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return err;
1662 }
1663
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001664 err =
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001665 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001666 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (!err)
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001668 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001669 usockaddr_len);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001670 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 }
1672 return err;
1673}
1674
1675/*
1676 * Send a datagram to a given address. We move the address into kernel
1677 * space and check the user space data area is readable before invoking
1678 * the protocol.
1679 */
1680
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001681asmlinkage long sys_sendto(int fd, void __user *buff, size_t len,
1682 unsigned flags, struct sockaddr __user *addr,
1683 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684{
1685 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001686 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 int err;
1688 struct msghdr msg;
1689 struct iovec iov;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001690 int fput_needed;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001691
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001692 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1693 if (!sock)
David S. Miller4387ff72007-02-08 15:06:08 -08001694 goto out;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001695
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001696 iov.iov_base = buff;
1697 iov.iov_len = len;
1698 msg.msg_name = NULL;
1699 msg.msg_iov = &iov;
1700 msg.msg_iovlen = 1;
1701 msg.msg_control = NULL;
1702 msg.msg_controllen = 0;
1703 msg.msg_namelen = 0;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001704 if (addr) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001705 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 if (err < 0)
1707 goto out_put;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001708 msg.msg_name = (struct sockaddr *)&address;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001709 msg.msg_namelen = addr_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711 if (sock->file->f_flags & O_NONBLOCK)
1712 flags |= MSG_DONTWAIT;
1713 msg.msg_flags = flags;
1714 err = sock_sendmsg(sock, &msg, len);
1715
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001716out_put:
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001717 fput_light(sock->file, fput_needed);
David S. Miller4387ff72007-02-08 15:06:08 -08001718out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 return err;
1720}
1721
1722/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001723 * Send a datagram down a socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 */
1725
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001726asmlinkage long sys_send(int fd, void __user *buff, size_t len, unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
1728 return sys_sendto(fd, buff, len, flags, NULL, 0);
1729}
1730
1731/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001732 * Receive a frame from the socket and optionally record the address of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 * sender. We verify the buffers are writable and if needed move the
1734 * sender address from kernel to user space.
1735 */
1736
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001737asmlinkage long sys_recvfrom(int fd, void __user *ubuf, size_t size,
1738 unsigned flags, struct sockaddr __user *addr,
1739 int __user *addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
1741 struct socket *sock;
1742 struct iovec iov;
1743 struct msghdr msg;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001744 struct sockaddr_storage address;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001745 int err, err2;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001746 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001748 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 if (!sock)
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001750 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001752 msg.msg_control = NULL;
1753 msg.msg_controllen = 0;
1754 msg.msg_iovlen = 1;
1755 msg.msg_iov = &iov;
1756 iov.iov_len = size;
1757 iov.iov_base = ubuf;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001758 msg.msg_name = (struct sockaddr *)&address;
1759 msg.msg_namelen = sizeof(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 if (sock->file->f_flags & O_NONBLOCK)
1761 flags |= MSG_DONTWAIT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001762 err = sock_recvmsg(sock, &msg, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001764 if (err >= 0 && addr != NULL) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001765 err2 = move_addr_to_user((struct sockaddr *)&address,
1766 msg.msg_namelen, addr, addr_len);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001767 if (err2 < 0)
1768 err = err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 }
Pavel Emelyanovde0fa952007-11-14 16:01:43 -08001770
1771 fput_light(sock->file, fput_needed);
David S. Miller4387ff72007-02-08 15:06:08 -08001772out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return err;
1774}
1775
1776/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001777 * Receive a datagram from a socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 */
1779
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001780asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1781 unsigned flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782{
1783 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1784}
1785
1786/*
1787 * Set a socket option. Because we don't know the option lengths we have
1788 * to pass the user mode parameter for the protocols to sort out.
1789 */
1790
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001791asmlinkage long sys_setsockopt(int fd, int level, int optname,
1792 char __user *optval, int optlen)
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
1797 if (optlen < 0)
1798 return -EINVAL;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001799
1800 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1801 if (sock != NULL) {
1802 err = security_socket_setsockopt(sock, level, optname);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001803 if (err)
1804 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
1806 if (level == SOL_SOCKET)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001807 err =
1808 sock_setsockopt(sock, level, optname, optval,
1809 optlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001811 err =
1812 sock->ops->setsockopt(sock, level, optname, optval,
1813 optlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001814out_put:
1815 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 }
1817 return err;
1818}
1819
1820/*
1821 * Get a socket option. Because we don't know the option lengths we have
1822 * to pass a user mode parameter for the protocols to sort out.
1823 */
1824
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001825asmlinkage long sys_getsockopt(int fd, int level, int optname,
1826 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001828 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 struct socket *sock;
1830
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001831 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1832 if (sock != NULL) {
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001833 err = security_socket_getsockopt(sock, level, optname);
1834 if (err)
1835 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
1837 if (level == SOL_SOCKET)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001838 err =
1839 sock_getsockopt(sock, level, optname, optval,
1840 optlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001842 err =
1843 sock->ops->getsockopt(sock, level, optname, optval,
1844 optlen);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001845out_put:
1846 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 }
1848 return err;
1849}
1850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851/*
1852 * Shutdown a socket.
1853 */
1854
1855asmlinkage long sys_shutdown(int fd, int how)
1856{
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001857 int err, fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 struct socket *sock;
1859
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001860 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1861 if (sock != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 err = security_socket_shutdown(sock, how);
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001863 if (!err)
1864 err = sock->ops->shutdown(sock, how);
1865 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 }
1867 return err;
1868}
1869
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001870/* A couple of helpful macros for getting the address of the 32/64 bit
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 * fields which are the same type (int / unsigned) on our platforms.
1872 */
1873#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1874#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1875#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877/*
1878 * BSD sendmsg interface
1879 */
1880
1881asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1882{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001883 struct compat_msghdr __user *msg_compat =
1884 (struct compat_msghdr __user *)msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 struct socket *sock;
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001886 struct sockaddr_storage address;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
Alex Williamsonb9d717a2005-09-26 14:28:02 -07001888 unsigned char ctl[sizeof(struct cmsghdr) + 20]
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001889 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1890 /* 20 is size of ipv6_pktinfo */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 unsigned char *ctl_buf = ctl;
1892 struct msghdr msg_sys;
1893 int err, ctl_len, iov_size, total_len;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001894 int fput_needed;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 err = -EFAULT;
1897 if (MSG_CMSG_COMPAT & flags) {
1898 if (get_compat_msghdr(&msg_sys, msg_compat))
1899 return -EFAULT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001900 }
1901 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 return -EFAULT;
1903
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001904 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001905 if (!sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 goto out;
1907
1908 /* do not move before msg_sys is valid */
1909 err = -EMSGSIZE;
1910 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1911 goto out_put;
1912
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001913 /* Check whether to allocate the iovec area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 err = -ENOMEM;
1915 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1916 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1917 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1918 if (!iov)
1919 goto out_put;
1920 }
1921
1922 /* This will also move the address data into kernel space */
1923 if (MSG_CMSG_COMPAT & flags) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001924 err = verify_compat_iovec(&msg_sys, iov,
1925 (struct sockaddr *)&address,
1926 VERIFY_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 } else
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07001928 err = verify_iovec(&msg_sys, iov,
1929 (struct sockaddr *)&address,
1930 VERIFY_READ);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001931 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 goto out_freeiov;
1933 total_len = err;
1934
1935 err = -ENOBUFS;
1936
1937 if (msg_sys.msg_controllen > INT_MAX)
1938 goto out_freeiov;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001939 ctl_len = msg_sys.msg_controllen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001941 err =
1942 cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1943 sizeof(ctl));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 if (err)
1945 goto out_freeiov;
1946 ctl_buf = msg_sys.msg_control;
Al Viro8920e8f2005-09-07 18:28:51 -07001947 ctl_len = msg_sys.msg_controllen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 } else if (ctl_len) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001949 if (ctl_len > sizeof(ctl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001951 if (ctl_buf == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 goto out_freeiov;
1953 }
1954 err = -EFAULT;
1955 /*
1956 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1957 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1958 * checking falls down on this.
1959 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001960 if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1961 ctl_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 goto out_freectl;
1963 msg_sys.msg_control = ctl_buf;
1964 }
1965 msg_sys.msg_flags = flags;
1966
1967 if (sock->file->f_flags & O_NONBLOCK)
1968 msg_sys.msg_flags |= MSG_DONTWAIT;
1969 err = sock_sendmsg(sock, &msg_sys, total_len);
1970
1971out_freectl:
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001972 if (ctl_buf != ctl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1974out_freeiov:
1975 if (iov != iovstack)
1976 sock_kfree_s(sock->sk, iov, iov_size);
1977out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001978 fput_light(sock->file, fput_needed);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001979out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 return err;
1981}
1982
1983/*
1984 * BSD recvmsg interface
1985 */
1986
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001987asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg,
1988 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001990 struct compat_msghdr __user *msg_compat =
1991 (struct compat_msghdr __user *)msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 struct socket *sock;
1993 struct iovec iovstack[UIO_FASTIOV];
Stephen Hemminger89bddce2006-09-01 00:19:31 -07001994 struct iovec *iov = iovstack;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 struct msghdr msg_sys;
1996 unsigned long cmsg_ptr;
1997 int err, iov_size, total_len, len;
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08001998 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 /* kernel mode address */
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07002001 struct sockaddr_storage addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
2003 /* user mode address pointers */
2004 struct sockaddr __user *uaddr;
2005 int __user *uaddr_len;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002006
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 if (MSG_CMSG_COMPAT & flags) {
2008 if (get_compat_msghdr(&msg_sys, msg_compat))
2009 return -EFAULT;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002010 }
2011 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
2012 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08002014 sock = sockfd_lookup_light(fd, &err, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 if (!sock)
2016 goto out;
2017
2018 err = -EMSGSIZE;
2019 if (msg_sys.msg_iovlen > UIO_MAXIOV)
2020 goto out_put;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002021
2022 /* Check whether to allocate the iovec area */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 err = -ENOMEM;
2024 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
2025 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
2026 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2027 if (!iov)
2028 goto out_put;
2029 }
2030
2031 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002032 * Save the user-mode address (verify_iovec will change the
2033 * kernel msghdr to use the kernel address space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002035
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -07002036 uaddr = (__force void __user *)msg_sys.msg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 uaddr_len = COMPAT_NAMELEN(msg);
2038 if (MSG_CMSG_COMPAT & flags) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07002039 err = verify_compat_iovec(&msg_sys, iov,
2040 (struct sockaddr *)&addr,
2041 VERIFY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 } else
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07002043 err = verify_iovec(&msg_sys, iov,
2044 (struct sockaddr *)&addr,
2045 VERIFY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 if (err < 0)
2047 goto out_freeiov;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002048 total_len = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 cmsg_ptr = (unsigned long)msg_sys.msg_control;
Ulrich Drepper4a195422007-07-15 23:40:34 -07002051 msg_sys.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 if (sock->file->f_flags & O_NONBLOCK)
2054 flags |= MSG_DONTWAIT;
2055 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
2056 if (err < 0)
2057 goto out_freeiov;
2058 len = err;
2059
2060 if (uaddr != NULL) {
YOSHIFUJI Hideaki230b1832008-07-19 22:35:47 -07002061 err = move_addr_to_user((struct sockaddr *)&addr,
2062 msg_sys.msg_namelen, uaddr,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002063 uaddr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 if (err < 0)
2065 goto out_freeiov;
2066 }
David S. Miller37f7f422005-09-16 16:51:01 -07002067 err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
2068 COMPAT_FLAGS(msg));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (err)
2070 goto out_freeiov;
2071 if (MSG_CMSG_COMPAT & flags)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002072 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 &msg_compat->msg_controllen);
2074 else
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002075 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 &msg->msg_controllen);
2077 if (err)
2078 goto out_freeiov;
2079 err = len;
2080
2081out_freeiov:
2082 if (iov != iovstack)
2083 sock_kfree_s(sock->sk, iov, iov_size);
2084out_put:
Benjamin LaHaise6cb153c2006-03-20 22:27:12 -08002085 fput_light(sock->file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086out:
2087 return err;
2088}
2089
2090#ifdef __ARCH_WANT_SYS_SOCKETCALL
2091
2092/* Argument list sizes for sys_socketcall */
2093#define AL(x) ((x) * sizeof(unsigned long))
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002094static const unsigned char nargs[19]={
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002095 AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2096 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002097 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
2098 AL(6)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002099};
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101#undef AL
2102
2103/*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002104 * System call vectors.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 *
2106 * Argument checking cleaned up. Saved 20% in size.
2107 * This function doesn't need to set the kernel lock because
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002108 * it is set by the callees.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 */
2110
2111asmlinkage long sys_socketcall(int call, unsigned long __user *args)
2112{
2113 unsigned long a[6];
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002114 unsigned long a0, a1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 int err;
2116
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002117 if (call < 1 || call > SYS_PACCEPT)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 return -EINVAL;
2119
2120 /* copy_from_user should be SMP safe. */
2121 if (copy_from_user(a, args, nargs[call]))
2122 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01002123
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002124 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01002125 if (err)
2126 return err;
2127
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002128 a0 = a[0];
2129 a1 = a[1];
2130
2131 switch (call) {
2132 case SYS_SOCKET:
2133 err = sys_socket(a0, a1, a[2]);
2134 break;
2135 case SYS_BIND:
2136 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2137 break;
2138 case SYS_CONNECT:
2139 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2140 break;
2141 case SYS_LISTEN:
2142 err = sys_listen(a0, a1);
2143 break;
2144 case SYS_ACCEPT:
2145 err =
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002146 do_accept(a0, (struct sockaddr __user *)a1,
2147 (int __user *)a[2], 0);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002148 break;
2149 case SYS_GETSOCKNAME:
2150 err =
2151 sys_getsockname(a0, (struct sockaddr __user *)a1,
2152 (int __user *)a[2]);
2153 break;
2154 case SYS_GETPEERNAME:
2155 err =
2156 sys_getpeername(a0, (struct sockaddr __user *)a1,
2157 (int __user *)a[2]);
2158 break;
2159 case SYS_SOCKETPAIR:
2160 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2161 break;
2162 case SYS_SEND:
2163 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2164 break;
2165 case SYS_SENDTO:
2166 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2167 (struct sockaddr __user *)a[4], a[5]);
2168 break;
2169 case SYS_RECV:
2170 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2171 break;
2172 case SYS_RECVFROM:
2173 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2174 (struct sockaddr __user *)a[4],
2175 (int __user *)a[5]);
2176 break;
2177 case SYS_SHUTDOWN:
2178 err = sys_shutdown(a0, a1);
2179 break;
2180 case SYS_SETSOCKOPT:
2181 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2182 break;
2183 case SYS_GETSOCKOPT:
2184 err =
2185 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2186 (int __user *)a[4]);
2187 break;
2188 case SYS_SENDMSG:
2189 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2190 break;
2191 case SYS_RECVMSG:
2192 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2193 break;
Ulrich Drepperaaca0bd2008-07-23 21:29:20 -07002194 case SYS_PACCEPT:
2195 err =
2196 sys_paccept(a0, (struct sockaddr __user *)a1,
2197 (int __user *)a[2],
2198 (const sigset_t __user *) a[3],
2199 a[4], a[5]);
2200 break;
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002201 default:
2202 err = -EINVAL;
2203 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 }
2205 return err;
2206}
2207
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002208#endif /* __ARCH_WANT_SYS_SOCKETCALL */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002210/**
2211 * sock_register - add a socket protocol handler
2212 * @ops: description of protocol
2213 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 * This function is called by a protocol handler that wants to
2215 * advertise its address family, and have it linked into the
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002216 * socket interface. The value ops->family coresponds to the
2217 * socket system call protocol family.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 */
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002219int sock_register(const struct net_proto_family *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220{
2221 int err;
2222
2223 if (ops->family >= NPROTO) {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002224 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2225 NPROTO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 return -ENOBUFS;
2227 }
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002228
2229 spin_lock(&net_family_lock);
2230 if (net_families[ops->family])
2231 err = -EEXIST;
2232 else {
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002233 net_families[ops->family] = ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 err = 0;
2235 }
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002236 spin_unlock(&net_family_lock);
2237
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002238 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 return err;
2240}
2241
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002242/**
2243 * sock_unregister - remove a protocol handler
2244 * @family: protocol family to remove
2245 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 * This function is called by a protocol handler that wants to
2247 * remove its address family, and have it unlinked from the
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002248 * new socket creation.
2249 *
2250 * If protocol handler is a module, then it can use module reference
2251 * counts to protect against new references. If protocol handler is not
2252 * a module then it needs to provide its own protection in
2253 * the ops->create routine.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 */
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002255void sock_unregister(int family)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256{
Stephen Hemmingerf0fd27d2006-08-09 21:03:17 -07002257 BUG_ON(family < 0 || family >= NPROTO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002259 spin_lock(&net_family_lock);
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002260 net_families[family] = NULL;
Stephen Hemminger55737fd2006-09-01 00:23:39 -07002261 spin_unlock(&net_family_lock);
2262
2263 synchronize_rcu();
2264
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002265 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266}
2267
Andi Kleen77d76ea2005-12-22 12:43:42 -08002268static int __init sock_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269{
2270 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002271 * Initialize sock SLAB cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 */
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 sk_init();
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002277 * Initialize skbuff SLAB cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 */
2279 skb_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 /*
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002282 * Initialize the protocols module.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 */
2284
2285 init_inodecache();
2286 register_filesystem(&sock_fs_type);
2287 sock_mnt = kern_mount(&sock_fs_type);
Andi Kleen77d76ea2005-12-22 12:43:42 -08002288
2289 /* The real protocol initialization is performed in later initcalls.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 */
2291
2292#ifdef CONFIG_NETFILTER
2293 netfilter_init();
2294#endif
David S. Millercbeb3212005-12-22 12:58:55 -08002295
2296 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297}
2298
Andi Kleen77d76ea2005-12-22 12:43:42 -08002299core_initcall(sock_init); /* early initcall */
2300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301#ifdef CONFIG_PROC_FS
2302void socket_seq_show(struct seq_file *seq)
2303{
2304 int cpu;
2305 int counter = 0;
2306
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07002307 for_each_possible_cpu(cpu)
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002308 counter += per_cpu(sockets_in_use, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309
2310 /* It can be negative, by the way. 8) */
2311 if (counter < 0)
2312 counter = 0;
2313
2314 seq_printf(seq, "sockets: used %d\n", counter);
2315}
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002316#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002318#ifdef CONFIG_COMPAT
2319static long compat_sock_ioctl(struct file *file, unsigned cmd,
Stephen Hemminger89bddce2006-09-01 00:19:31 -07002320 unsigned long arg)
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002321{
2322 struct socket *sock = file->private_data;
2323 int ret = -ENOIOCTLCMD;
David S. Miller87de87d2008-06-03 09:14:03 -07002324 struct sock *sk;
2325 struct net *net;
2326
2327 sk = sock->sk;
2328 net = sock_net(sk);
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002329
2330 if (sock->ops->compat_ioctl)
2331 ret = sock->ops->compat_ioctl(sock, cmd, arg);
2332
David S. Miller87de87d2008-06-03 09:14:03 -07002333 if (ret == -ENOIOCTLCMD &&
2334 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2335 ret = compat_wext_handle_ioctl(net, cmd, arg);
2336
Shaun Pereira89bbfc92006-03-21 23:58:08 -08002337 return ret;
2338}
2339#endif
2340
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002341int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2342{
2343 return sock->ops->bind(sock, addr, addrlen);
2344}
2345
2346int kernel_listen(struct socket *sock, int backlog)
2347{
2348 return sock->ops->listen(sock, backlog);
2349}
2350
2351int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2352{
2353 struct sock *sk = sock->sk;
2354 int err;
2355
2356 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2357 newsock);
2358 if (err < 0)
2359 goto done;
2360
2361 err = sock->ops->accept(sock, *newsock, flags);
2362 if (err < 0) {
2363 sock_release(*newsock);
Tony Battersbyfa8705b2007-10-10 21:09:04 -07002364 *newsock = NULL;
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002365 goto done;
2366 }
2367
2368 (*newsock)->ops = sock->ops;
2369
2370done:
2371 return err;
2372}
2373
2374int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
YOSHIFUJI Hideaki4768fbc2007-02-09 23:25:31 +09002375 int flags)
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002376{
2377 return sock->ops->connect(sock, addr, addrlen, flags);
2378}
2379
2380int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2381 int *addrlen)
2382{
2383 return sock->ops->getname(sock, addr, addrlen, 0);
2384}
2385
2386int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2387 int *addrlen)
2388{
2389 return sock->ops->getname(sock, addr, addrlen, 1);
2390}
2391
2392int kernel_getsockopt(struct socket *sock, int level, int optname,
2393 char *optval, int *optlen)
2394{
2395 mm_segment_t oldfs = get_fs();
2396 int err;
2397
2398 set_fs(KERNEL_DS);
2399 if (level == SOL_SOCKET)
2400 err = sock_getsockopt(sock, level, optname, optval, optlen);
2401 else
2402 err = sock->ops->getsockopt(sock, level, optname, optval,
2403 optlen);
2404 set_fs(oldfs);
2405 return err;
2406}
2407
2408int kernel_setsockopt(struct socket *sock, int level, int optname,
2409 char *optval, int optlen)
2410{
2411 mm_segment_t oldfs = get_fs();
2412 int err;
2413
2414 set_fs(KERNEL_DS);
2415 if (level == SOL_SOCKET)
2416 err = sock_setsockopt(sock, level, optname, optval, optlen);
2417 else
2418 err = sock->ops->setsockopt(sock, level, optname, optval,
2419 optlen);
2420 set_fs(oldfs);
2421 return err;
2422}
2423
2424int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2425 size_t size, int flags)
2426{
2427 if (sock->ops->sendpage)
2428 return sock->ops->sendpage(sock, page, offset, size, flags);
2429
2430 return sock_no_sendpage(sock, page, offset, size, flags);
2431}
2432
2433int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2434{
2435 mm_segment_t oldfs = get_fs();
2436 int err;
2437
2438 set_fs(KERNEL_DS);
2439 err = sock->ops->ioctl(sock, cmd, arg);
2440 set_fs(oldfs);
2441
2442 return err;
2443}
2444
Trond Myklebust91cf45f2007-11-12 18:10:39 -08002445int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2446{
2447 return sock->ops->shutdown(sock, how);
2448}
2449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450EXPORT_SYMBOL(sock_create);
2451EXPORT_SYMBOL(sock_create_kern);
2452EXPORT_SYMBOL(sock_create_lite);
2453EXPORT_SYMBOL(sock_map_fd);
2454EXPORT_SYMBOL(sock_recvmsg);
2455EXPORT_SYMBOL(sock_register);
2456EXPORT_SYMBOL(sock_release);
2457EXPORT_SYMBOL(sock_sendmsg);
2458EXPORT_SYMBOL(sock_unregister);
2459EXPORT_SYMBOL(sock_wake_async);
2460EXPORT_SYMBOL(sockfd_lookup);
2461EXPORT_SYMBOL(kernel_sendmsg);
2462EXPORT_SYMBOL(kernel_recvmsg);
Sridhar Samudralaac5a4882006-08-07 20:57:31 -07002463EXPORT_SYMBOL(kernel_bind);
2464EXPORT_SYMBOL(kernel_listen);
2465EXPORT_SYMBOL(kernel_accept);
2466EXPORT_SYMBOL(kernel_connect);
2467EXPORT_SYMBOL(kernel_getsockname);
2468EXPORT_SYMBOL(kernel_getpeername);
2469EXPORT_SYMBOL(kernel_getsockopt);
2470EXPORT_SYMBOL(kernel_setsockopt);
2471EXPORT_SYMBOL(kernel_sendpage);
2472EXPORT_SYMBOL(kernel_sock_ioctl);
Trond Myklebust91cf45f2007-11-12 18:10:39 -08002473EXPORT_SYMBOL(kernel_sock_shutdown);