xref: /openbmc/linux/net/socket.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
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>
7  *		Ross Biro
8  *		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)
45  *		Tigran Aivazian	:	Made listen(2) backlog sanity checks
46  *					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
56  *	paradigm.
57  *
58  *	Based upon Swansea University Computer Society NET3.039
59  */
60 
61 #include <linux/mm.h>
62 #include <linux/socket.h>
63 #include <linux/file.h>
64 #include <linux/net.h>
65 #include <linux/interrupt.h>
66 #include <linux/rcupdate.h>
67 #include <linux/netdevice.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/mutex.h>
71 #include <linux/wanrouter.h>
72 #include <linux/if_bridge.h>
73 #include <linux/if_frad.h>
74 #include <linux/if_vlan.h>
75 #include <linux/init.h>
76 #include <linux/poll.h>
77 #include <linux/cache.h>
78 #include <linux/module.h>
79 #include <linux/highmem.h>
80 #include <linux/mount.h>
81 #include <linux/security.h>
82 #include <linux/syscalls.h>
83 #include <linux/compat.h>
84 #include <linux/kmod.h>
85 #include <linux/audit.h>
86 #include <linux/wireless.h>
87 
88 #include <asm/uaccess.h>
89 #include <asm/unistd.h>
90 
91 #include <net/compat.h>
92 
93 #include <net/sock.h>
94 #include <linux/netfilter.h>
95 
96 static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
97 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
98 			 unsigned long nr_segs, loff_t pos);
99 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
100 			  unsigned long nr_segs, loff_t pos);
101 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
102 
103 static int sock_close(struct inode *inode, struct file *file);
104 static unsigned int sock_poll(struct file *file,
105 			      struct poll_table_struct *wait);
106 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
107 #ifdef CONFIG_COMPAT
108 static long compat_sock_ioctl(struct file *file,
109 			      unsigned int cmd, unsigned long arg);
110 #endif
111 static int sock_fasync(int fd, struct file *filp, int on);
112 static ssize_t sock_sendpage(struct file *file, struct page *page,
113 			     int offset, size_t size, loff_t *ppos, int more);
114 
115 /*
116  *	Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
117  *	in the operation structures but are done directly via the socketcall() multiplexor.
118  */
119 
120 static const struct file_operations socket_file_ops = {
121 	.owner =	THIS_MODULE,
122 	.llseek =	no_llseek,
123 	.aio_read =	sock_aio_read,
124 	.aio_write =	sock_aio_write,
125 	.poll =		sock_poll,
126 	.unlocked_ioctl = sock_ioctl,
127 #ifdef CONFIG_COMPAT
128 	.compat_ioctl = compat_sock_ioctl,
129 #endif
130 	.mmap =		sock_mmap,
131 	.open =		sock_no_open,	/* special open code to disallow open via /proc */
132 	.release =	sock_close,
133 	.fasync =	sock_fasync,
134 	.sendpage =	sock_sendpage,
135 	.splice_write = generic_splice_sendpage,
136 };
137 
138 /*
139  *	The protocol list. Each protocol is registered in here.
140  */
141 
142 static DEFINE_SPINLOCK(net_family_lock);
143 static const struct net_proto_family *net_families[NPROTO] __read_mostly;
144 
145 /*
146  *	Statistics counters of the socket lists
147  */
148 
149 static DEFINE_PER_CPU(int, sockets_in_use) = 0;
150 
151 /*
152  * Support routines.
153  * Move socket addresses back and forth across the kernel/user
154  * divide and look after the messy bits.
155  */
156 
157 #define MAX_SOCK_ADDR	128		/* 108 for Unix domain -
158 					   16 for IP, 16 for IPX,
159 					   24 for IPv6,
160 					   about 80 for AX.25
161 					   must be at least one bigger than
162 					   the AF_UNIX size (see net/unix/af_unix.c
163 					   :unix_mkname()).
164 					 */
165 
166 /**
167  *	move_addr_to_kernel	-	copy a socket address into kernel space
168  *	@uaddr: Address in user space
169  *	@kaddr: Address in kernel space
170  *	@ulen: Length in user space
171  *
172  *	The address is copied into kernel space. If the provided address is
173  *	too long an error code of -EINVAL is returned. If the copy gives
174  *	invalid addresses -EFAULT is returned. On a success 0 is returned.
175  */
176 
177 int move_addr_to_kernel(void __user *uaddr, int ulen, void *kaddr)
178 {
179 	if (ulen < 0 || ulen > MAX_SOCK_ADDR)
180 		return -EINVAL;
181 	if (ulen == 0)
182 		return 0;
183 	if (copy_from_user(kaddr, uaddr, ulen))
184 		return -EFAULT;
185 	return audit_sockaddr(ulen, kaddr);
186 }
187 
188 /**
189  *	move_addr_to_user	-	copy an address to user space
190  *	@kaddr: kernel space address
191  *	@klen: length of address in kernel
192  *	@uaddr: user space address
193  *	@ulen: pointer to user length field
194  *
195  *	The value pointed to by ulen on entry is the buffer length available.
196  *	This is overwritten with the buffer space used. -EINVAL is returned
197  *	if an overlong buffer is specified or a negative buffer size. -EFAULT
198  *	is returned if either the buffer or the length field are not
199  *	accessible.
200  *	After copying the data up to the limit the user specifies, the true
201  *	length of the data is written over the length limit the user
202  *	specified. Zero is returned for a success.
203  */
204 
205 int move_addr_to_user(void *kaddr, int klen, void __user *uaddr,
206 		      int __user *ulen)
207 {
208 	int err;
209 	int len;
210 
211 	err = get_user(len, ulen);
212 	if (err)
213 		return err;
214 	if (len > klen)
215 		len = klen;
216 	if (len < 0 || len > MAX_SOCK_ADDR)
217 		return -EINVAL;
218 	if (len) {
219 		if (audit_sockaddr(klen, kaddr))
220 			return -ENOMEM;
221 		if (copy_to_user(uaddr, kaddr, len))
222 			return -EFAULT;
223 	}
224 	/*
225 	 *      "fromlen shall refer to the value before truncation.."
226 	 *                      1003.1g
227 	 */
228 	return __put_user(klen, ulen);
229 }
230 
231 #define SOCKFS_MAGIC 0x534F434B
232 
233 static struct kmem_cache *sock_inode_cachep __read_mostly;
234 
235 static struct inode *sock_alloc_inode(struct super_block *sb)
236 {
237 	struct socket_alloc *ei;
238 
239 	ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
240 	if (!ei)
241 		return NULL;
242 	init_waitqueue_head(&ei->socket.wait);
243 
244 	ei->socket.fasync_list = NULL;
245 	ei->socket.state = SS_UNCONNECTED;
246 	ei->socket.flags = 0;
247 	ei->socket.ops = NULL;
248 	ei->socket.sk = NULL;
249 	ei->socket.file = NULL;
250 
251 	return &ei->vfs_inode;
252 }
253 
254 static void sock_destroy_inode(struct inode *inode)
255 {
256 	kmem_cache_free(sock_inode_cachep,
257 			container_of(inode, struct socket_alloc, vfs_inode));
258 }
259 
260 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
261 {
262 	struct socket_alloc *ei = (struct socket_alloc *)foo;
263 
264 	if (flags & SLAB_CTOR_CONSTRUCTOR)
265 		inode_init_once(&ei->vfs_inode);
266 }
267 
268 static int init_inodecache(void)
269 {
270 	sock_inode_cachep = kmem_cache_create("sock_inode_cache",
271 					      sizeof(struct socket_alloc),
272 					      0,
273 					      (SLAB_HWCACHE_ALIGN |
274 					       SLAB_RECLAIM_ACCOUNT |
275 					       SLAB_MEM_SPREAD),
276 					      init_once,
277 					      NULL);
278 	if (sock_inode_cachep == NULL)
279 		return -ENOMEM;
280 	return 0;
281 }
282 
283 static struct super_operations sockfs_ops = {
284 	.alloc_inode =	sock_alloc_inode,
285 	.destroy_inode =sock_destroy_inode,
286 	.statfs =	simple_statfs,
287 };
288 
289 static int sockfs_get_sb(struct file_system_type *fs_type,
290 			 int flags, const char *dev_name, void *data,
291 			 struct vfsmount *mnt)
292 {
293 	return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
294 			     mnt);
295 }
296 
297 static struct vfsmount *sock_mnt __read_mostly;
298 
299 static struct file_system_type sock_fs_type = {
300 	.name =		"sockfs",
301 	.get_sb =	sockfs_get_sb,
302 	.kill_sb =	kill_anon_super,
303 };
304 
305 static int sockfs_delete_dentry(struct dentry *dentry)
306 {
307 	/*
308 	 * At creation time, we pretended this dentry was hashed
309 	 * (by clearing DCACHE_UNHASHED bit in d_flags)
310 	 * At delete time, we restore the truth : not hashed.
311 	 * (so that dput() can proceed correctly)
312 	 */
313 	dentry->d_flags |= DCACHE_UNHASHED;
314 	return 0;
315 }
316 
317 /*
318  * sockfs_dname() is called from d_path().
319  */
320 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
321 {
322 	return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
323 				dentry->d_inode->i_ino);
324 }
325 
326 static struct dentry_operations sockfs_dentry_operations = {
327 	.d_delete = sockfs_delete_dentry,
328 	.d_dname  = sockfs_dname,
329 };
330 
331 /*
332  *	Obtains the first available file descriptor and sets it up for use.
333  *
334  *	These functions create file structures and maps them to fd space
335  *	of the current process. On success it returns file descriptor
336  *	and file struct implicitly stored in sock->file.
337  *	Note that another thread may close file descriptor before we return
338  *	from this function. We use the fact that now we do not refer
339  *	to socket after mapping. If one day we will need it, this
340  *	function will increment ref. count on file by 1.
341  *
342  *	In any case returned fd MAY BE not valid!
343  *	This race condition is unavoidable
344  *	with shared fd spaces, we cannot solve it inside kernel,
345  *	but we take care of internal coherence yet.
346  */
347 
348 static int sock_alloc_fd(struct file **filep)
349 {
350 	int fd;
351 
352 	fd = get_unused_fd();
353 	if (likely(fd >= 0)) {
354 		struct file *file = get_empty_filp();
355 
356 		*filep = file;
357 		if (unlikely(!file)) {
358 			put_unused_fd(fd);
359 			return -ENFILE;
360 		}
361 	} else
362 		*filep = NULL;
363 	return fd;
364 }
365 
366 static int sock_attach_fd(struct socket *sock, struct file *file)
367 {
368 	struct qstr name = { .name = "" };
369 
370 	file->f_path.dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
371 	if (unlikely(!file->f_path.dentry))
372 		return -ENOMEM;
373 
374 	file->f_path.dentry->d_op = &sockfs_dentry_operations;
375 	/*
376 	 * We dont want to push this dentry into global dentry hash table.
377 	 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
378 	 * This permits a working /proc/$pid/fd/XXX on sockets
379 	 */
380 	file->f_path.dentry->d_flags &= ~DCACHE_UNHASHED;
381 	d_instantiate(file->f_path.dentry, SOCK_INODE(sock));
382 	file->f_path.mnt = mntget(sock_mnt);
383 	file->f_mapping = file->f_path.dentry->d_inode->i_mapping;
384 
385 	sock->file = file;
386 	file->f_op = SOCK_INODE(sock)->i_fop = &socket_file_ops;
387 	file->f_mode = FMODE_READ | FMODE_WRITE;
388 	file->f_flags = O_RDWR;
389 	file->f_pos = 0;
390 	file->private_data = sock;
391 
392 	return 0;
393 }
394 
395 int sock_map_fd(struct socket *sock)
396 {
397 	struct file *newfile;
398 	int fd = sock_alloc_fd(&newfile);
399 
400 	if (likely(fd >= 0)) {
401 		int err = sock_attach_fd(sock, newfile);
402 
403 		if (unlikely(err < 0)) {
404 			put_filp(newfile);
405 			put_unused_fd(fd);
406 			return err;
407 		}
408 		fd_install(fd, newfile);
409 	}
410 	return fd;
411 }
412 
413 static struct socket *sock_from_file(struct file *file, int *err)
414 {
415 	if (file->f_op == &socket_file_ops)
416 		return file->private_data;	/* set in sock_map_fd */
417 
418 	*err = -ENOTSOCK;
419 	return NULL;
420 }
421 
422 /**
423  *	sockfd_lookup	- 	Go from a file number to its socket slot
424  *	@fd: file handle
425  *	@err: pointer to an error code return
426  *
427  *	The file handle passed in is locked and the socket it is bound
428  *	too is returned. If an error occurs the err pointer is overwritten
429  *	with a negative errno code and NULL is returned. The function checks
430  *	for both invalid handles and passing a handle which is not a socket.
431  *
432  *	On a success the socket object pointer is returned.
433  */
434 
435 struct socket *sockfd_lookup(int fd, int *err)
436 {
437 	struct file *file;
438 	struct socket *sock;
439 
440 	file = fget(fd);
441 	if (!file) {
442 		*err = -EBADF;
443 		return NULL;
444 	}
445 
446 	sock = sock_from_file(file, err);
447 	if (!sock)
448 		fput(file);
449 	return sock;
450 }
451 
452 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
453 {
454 	struct file *file;
455 	struct socket *sock;
456 
457 	*err = -EBADF;
458 	file = fget_light(fd, fput_needed);
459 	if (file) {
460 		sock = sock_from_file(file, err);
461 		if (sock)
462 			return sock;
463 		fput_light(file, *fput_needed);
464 	}
465 	return NULL;
466 }
467 
468 /**
469  *	sock_alloc	-	allocate a socket
470  *
471  *	Allocate a new inode and socket object. The two are bound together
472  *	and initialised. The socket is then returned. If we are out of inodes
473  *	NULL is returned.
474  */
475 
476 static struct socket *sock_alloc(void)
477 {
478 	struct inode *inode;
479 	struct socket *sock;
480 
481 	inode = new_inode(sock_mnt->mnt_sb);
482 	if (!inode)
483 		return NULL;
484 
485 	sock = SOCKET_I(inode);
486 
487 	inode->i_mode = S_IFSOCK | S_IRWXUGO;
488 	inode->i_uid = current->fsuid;
489 	inode->i_gid = current->fsgid;
490 
491 	get_cpu_var(sockets_in_use)++;
492 	put_cpu_var(sockets_in_use);
493 	return sock;
494 }
495 
496 /*
497  *	In theory you can't get an open on this inode, but /proc provides
498  *	a back door. Remember to keep it shut otherwise you'll let the
499  *	creepy crawlies in.
500  */
501 
502 static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
503 {
504 	return -ENXIO;
505 }
506 
507 const struct file_operations bad_sock_fops = {
508 	.owner = THIS_MODULE,
509 	.open = sock_no_open,
510 };
511 
512 /**
513  *	sock_release	-	close a socket
514  *	@sock: socket to close
515  *
516  *	The socket is released from the protocol stack if it has a release
517  *	callback, and the inode is then released if the socket is bound to
518  *	an inode not a file.
519  */
520 
521 void sock_release(struct socket *sock)
522 {
523 	if (sock->ops) {
524 		struct module *owner = sock->ops->owner;
525 
526 		sock->ops->release(sock);
527 		sock->ops = NULL;
528 		module_put(owner);
529 	}
530 
531 	if (sock->fasync_list)
532 		printk(KERN_ERR "sock_release: fasync list not empty!\n");
533 
534 	get_cpu_var(sockets_in_use)--;
535 	put_cpu_var(sockets_in_use);
536 	if (!sock->file) {
537 		iput(SOCK_INODE(sock));
538 		return;
539 	}
540 	sock->file = NULL;
541 }
542 
543 static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
544 				 struct msghdr *msg, size_t size)
545 {
546 	struct sock_iocb *si = kiocb_to_siocb(iocb);
547 	int err;
548 
549 	si->sock = sock;
550 	si->scm = NULL;
551 	si->msg = msg;
552 	si->size = size;
553 
554 	err = security_socket_sendmsg(sock, msg, size);
555 	if (err)
556 		return err;
557 
558 	return sock->ops->sendmsg(iocb, sock, msg, size);
559 }
560 
561 int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
562 {
563 	struct kiocb iocb;
564 	struct sock_iocb siocb;
565 	int ret;
566 
567 	init_sync_kiocb(&iocb, NULL);
568 	iocb.private = &siocb;
569 	ret = __sock_sendmsg(&iocb, sock, msg, size);
570 	if (-EIOCBQUEUED == ret)
571 		ret = wait_on_sync_kiocb(&iocb);
572 	return ret;
573 }
574 
575 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
576 		   struct kvec *vec, size_t num, size_t size)
577 {
578 	mm_segment_t oldfs = get_fs();
579 	int result;
580 
581 	set_fs(KERNEL_DS);
582 	/*
583 	 * the following is safe, since for compiler definitions of kvec and
584 	 * iovec are identical, yielding the same in-core layout and alignment
585 	 */
586 	msg->msg_iov = (struct iovec *)vec;
587 	msg->msg_iovlen = num;
588 	result = sock_sendmsg(sock, msg, size);
589 	set_fs(oldfs);
590 	return result;
591 }
592 
593 /*
594  * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
595  */
596 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
597 	struct sk_buff *skb)
598 {
599 	ktime_t kt = skb->tstamp;
600 
601 	if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
602 		struct timeval tv;
603 		/* Race occurred between timestamp enabling and packet
604 		   receiving.  Fill in the current time for now. */
605 		if (kt.tv64 == 0)
606 			kt = ktime_get_real();
607 		skb->tstamp = kt;
608 		tv = ktime_to_timeval(kt);
609 		put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, sizeof(tv), &tv);
610 	} else {
611 		struct timespec ts;
612 		/* Race occurred between timestamp enabling and packet
613 		   receiving.  Fill in the current time for now. */
614 		if (kt.tv64 == 0)
615 			kt = ktime_get_real();
616 		skb->tstamp = kt;
617 		ts = ktime_to_timespec(kt);
618 		put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS, sizeof(ts), &ts);
619 	}
620 }
621 
622 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
623 
624 static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
625 				 struct msghdr *msg, size_t size, int flags)
626 {
627 	int err;
628 	struct sock_iocb *si = kiocb_to_siocb(iocb);
629 
630 	si->sock = sock;
631 	si->scm = NULL;
632 	si->msg = msg;
633 	si->size = size;
634 	si->flags = flags;
635 
636 	err = security_socket_recvmsg(sock, msg, size, flags);
637 	if (err)
638 		return err;
639 
640 	return sock->ops->recvmsg(iocb, sock, msg, size, flags);
641 }
642 
643 int sock_recvmsg(struct socket *sock, struct msghdr *msg,
644 		 size_t size, int flags)
645 {
646 	struct kiocb iocb;
647 	struct sock_iocb siocb;
648 	int ret;
649 
650 	init_sync_kiocb(&iocb, NULL);
651 	iocb.private = &siocb;
652 	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
653 	if (-EIOCBQUEUED == ret)
654 		ret = wait_on_sync_kiocb(&iocb);
655 	return ret;
656 }
657 
658 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
659 		   struct kvec *vec, size_t num, size_t size, int flags)
660 {
661 	mm_segment_t oldfs = get_fs();
662 	int result;
663 
664 	set_fs(KERNEL_DS);
665 	/*
666 	 * the following is safe, since for compiler definitions of kvec and
667 	 * iovec are identical, yielding the same in-core layout and alignment
668 	 */
669 	msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
670 	result = sock_recvmsg(sock, msg, size, flags);
671 	set_fs(oldfs);
672 	return result;
673 }
674 
675 static void sock_aio_dtor(struct kiocb *iocb)
676 {
677 	kfree(iocb->private);
678 }
679 
680 static ssize_t sock_sendpage(struct file *file, struct page *page,
681 			     int offset, size_t size, loff_t *ppos, int more)
682 {
683 	struct socket *sock;
684 	int flags;
685 
686 	sock = file->private_data;
687 
688 	flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
689 	if (more)
690 		flags |= MSG_MORE;
691 
692 	return sock->ops->sendpage(sock, page, offset, size, flags);
693 }
694 
695 static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
696 					 struct sock_iocb *siocb)
697 {
698 	if (!is_sync_kiocb(iocb)) {
699 		siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
700 		if (!siocb)
701 			return NULL;
702 		iocb->ki_dtor = sock_aio_dtor;
703 	}
704 
705 	siocb->kiocb = iocb;
706 	iocb->private = siocb;
707 	return siocb;
708 }
709 
710 static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
711 		struct file *file, const struct iovec *iov,
712 		unsigned long nr_segs)
713 {
714 	struct socket *sock = file->private_data;
715 	size_t size = 0;
716 	int i;
717 
718 	for (i = 0; i < nr_segs; i++)
719 		size += iov[i].iov_len;
720 
721 	msg->msg_name = NULL;
722 	msg->msg_namelen = 0;
723 	msg->msg_control = NULL;
724 	msg->msg_controllen = 0;
725 	msg->msg_iov = (struct iovec *)iov;
726 	msg->msg_iovlen = nr_segs;
727 	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
728 
729 	return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
730 }
731 
732 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
733 				unsigned long nr_segs, loff_t pos)
734 {
735 	struct sock_iocb siocb, *x;
736 
737 	if (pos != 0)
738 		return -ESPIPE;
739 
740 	if (iocb->ki_left == 0)	/* Match SYS5 behaviour */
741 		return 0;
742 
743 
744 	x = alloc_sock_iocb(iocb, &siocb);
745 	if (!x)
746 		return -ENOMEM;
747 	return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
748 }
749 
750 static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
751 			struct file *file, const struct iovec *iov,
752 			unsigned long nr_segs)
753 {
754 	struct socket *sock = file->private_data;
755 	size_t size = 0;
756 	int i;
757 
758 	for (i = 0; i < nr_segs; i++)
759 		size += iov[i].iov_len;
760 
761 	msg->msg_name = NULL;
762 	msg->msg_namelen = 0;
763 	msg->msg_control = NULL;
764 	msg->msg_controllen = 0;
765 	msg->msg_iov = (struct iovec *)iov;
766 	msg->msg_iovlen = nr_segs;
767 	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
768 	if (sock->type == SOCK_SEQPACKET)
769 		msg->msg_flags |= MSG_EOR;
770 
771 	return __sock_sendmsg(iocb, sock, msg, size);
772 }
773 
774 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
775 			  unsigned long nr_segs, loff_t pos)
776 {
777 	struct sock_iocb siocb, *x;
778 
779 	if (pos != 0)
780 		return -ESPIPE;
781 
782 	if (iocb->ki_left == 0)	/* Match SYS5 behaviour */
783 		return 0;
784 
785 	x = alloc_sock_iocb(iocb, &siocb);
786 	if (!x)
787 		return -ENOMEM;
788 
789 	return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
790 }
791 
792 /*
793  * Atomic setting of ioctl hooks to avoid race
794  * with module unload.
795  */
796 
797 static DEFINE_MUTEX(br_ioctl_mutex);
798 static int (*br_ioctl_hook) (unsigned int cmd, void __user *arg) = NULL;
799 
800 void brioctl_set(int (*hook) (unsigned int, void __user *))
801 {
802 	mutex_lock(&br_ioctl_mutex);
803 	br_ioctl_hook = hook;
804 	mutex_unlock(&br_ioctl_mutex);
805 }
806 
807 EXPORT_SYMBOL(brioctl_set);
808 
809 static DEFINE_MUTEX(vlan_ioctl_mutex);
810 static int (*vlan_ioctl_hook) (void __user *arg);
811 
812 void vlan_ioctl_set(int (*hook) (void __user *))
813 {
814 	mutex_lock(&vlan_ioctl_mutex);
815 	vlan_ioctl_hook = hook;
816 	mutex_unlock(&vlan_ioctl_mutex);
817 }
818 
819 EXPORT_SYMBOL(vlan_ioctl_set);
820 
821 static DEFINE_MUTEX(dlci_ioctl_mutex);
822 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
823 
824 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
825 {
826 	mutex_lock(&dlci_ioctl_mutex);
827 	dlci_ioctl_hook = hook;
828 	mutex_unlock(&dlci_ioctl_mutex);
829 }
830 
831 EXPORT_SYMBOL(dlci_ioctl_set);
832 
833 /*
834  *	With an ioctl, arg may well be a user mode pointer, but we don't know
835  *	what to do with it - that's up to the protocol still.
836  */
837 
838 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
839 {
840 	struct socket *sock;
841 	void __user *argp = (void __user *)arg;
842 	int pid, err;
843 
844 	sock = file->private_data;
845 	if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
846 		err = dev_ioctl(cmd, argp);
847 	} else
848 #ifdef CONFIG_WIRELESS_EXT
849 	if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
850 		err = dev_ioctl(cmd, argp);
851 	} else
852 #endif				/* CONFIG_WIRELESS_EXT */
853 		switch (cmd) {
854 		case FIOSETOWN:
855 		case SIOCSPGRP:
856 			err = -EFAULT;
857 			if (get_user(pid, (int __user *)argp))
858 				break;
859 			err = f_setown(sock->file, pid, 1);
860 			break;
861 		case FIOGETOWN:
862 		case SIOCGPGRP:
863 			err = put_user(f_getown(sock->file),
864 				       (int __user *)argp);
865 			break;
866 		case SIOCGIFBR:
867 		case SIOCSIFBR:
868 		case SIOCBRADDBR:
869 		case SIOCBRDELBR:
870 			err = -ENOPKG;
871 			if (!br_ioctl_hook)
872 				request_module("bridge");
873 
874 			mutex_lock(&br_ioctl_mutex);
875 			if (br_ioctl_hook)
876 				err = br_ioctl_hook(cmd, argp);
877 			mutex_unlock(&br_ioctl_mutex);
878 			break;
879 		case SIOCGIFVLAN:
880 		case SIOCSIFVLAN:
881 			err = -ENOPKG;
882 			if (!vlan_ioctl_hook)
883 				request_module("8021q");
884 
885 			mutex_lock(&vlan_ioctl_mutex);
886 			if (vlan_ioctl_hook)
887 				err = vlan_ioctl_hook(argp);
888 			mutex_unlock(&vlan_ioctl_mutex);
889 			break;
890 		case SIOCADDDLCI:
891 		case SIOCDELDLCI:
892 			err = -ENOPKG;
893 			if (!dlci_ioctl_hook)
894 				request_module("dlci");
895 
896 			if (dlci_ioctl_hook) {
897 				mutex_lock(&dlci_ioctl_mutex);
898 				err = dlci_ioctl_hook(cmd, argp);
899 				mutex_unlock(&dlci_ioctl_mutex);
900 			}
901 			break;
902 		default:
903 			err = sock->ops->ioctl(sock, cmd, arg);
904 
905 			/*
906 			 * If this ioctl is unknown try to hand it down
907 			 * to the NIC driver.
908 			 */
909 			if (err == -ENOIOCTLCMD)
910 				err = dev_ioctl(cmd, argp);
911 			break;
912 		}
913 	return err;
914 }
915 
916 int sock_create_lite(int family, int type, int protocol, struct socket **res)
917 {
918 	int err;
919 	struct socket *sock = NULL;
920 
921 	err = security_socket_create(family, type, protocol, 1);
922 	if (err)
923 		goto out;
924 
925 	sock = sock_alloc();
926 	if (!sock) {
927 		err = -ENOMEM;
928 		goto out;
929 	}
930 
931 	sock->type = type;
932 	err = security_socket_post_create(sock, family, type, protocol, 1);
933 	if (err)
934 		goto out_release;
935 
936 out:
937 	*res = sock;
938 	return err;
939 out_release:
940 	sock_release(sock);
941 	sock = NULL;
942 	goto out;
943 }
944 
945 /* No kernel lock held - perfect */
946 static unsigned int sock_poll(struct file *file, poll_table *wait)
947 {
948 	struct socket *sock;
949 
950 	/*
951 	 *      We can't return errors to poll, so it's either yes or no.
952 	 */
953 	sock = file->private_data;
954 	return sock->ops->poll(file, sock, wait);
955 }
956 
957 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
958 {
959 	struct socket *sock = file->private_data;
960 
961 	return sock->ops->mmap(file, sock, vma);
962 }
963 
964 static int sock_close(struct inode *inode, struct file *filp)
965 {
966 	/*
967 	 *      It was possible the inode is NULL we were
968 	 *      closing an unfinished socket.
969 	 */
970 
971 	if (!inode) {
972 		printk(KERN_DEBUG "sock_close: NULL inode\n");
973 		return 0;
974 	}
975 	sock_fasync(-1, filp, 0);
976 	sock_release(SOCKET_I(inode));
977 	return 0;
978 }
979 
980 /*
981  *	Update the socket async list
982  *
983  *	Fasync_list locking strategy.
984  *
985  *	1. fasync_list is modified only under process context socket lock
986  *	   i.e. under semaphore.
987  *	2. fasync_list is used under read_lock(&sk->sk_callback_lock)
988  *	   or under socket lock.
989  *	3. fasync_list can be used from softirq context, so that
990  *	   modification under socket lock have to be enhanced with
991  *	   write_lock_bh(&sk->sk_callback_lock).
992  *							--ANK (990710)
993  */
994 
995 static int sock_fasync(int fd, struct file *filp, int on)
996 {
997 	struct fasync_struct *fa, *fna = NULL, **prev;
998 	struct socket *sock;
999 	struct sock *sk;
1000 
1001 	if (on) {
1002 		fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
1003 		if (fna == NULL)
1004 			return -ENOMEM;
1005 	}
1006 
1007 	sock = filp->private_data;
1008 
1009 	sk = sock->sk;
1010 	if (sk == NULL) {
1011 		kfree(fna);
1012 		return -EINVAL;
1013 	}
1014 
1015 	lock_sock(sk);
1016 
1017 	prev = &(sock->fasync_list);
1018 
1019 	for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1020 		if (fa->fa_file == filp)
1021 			break;
1022 
1023 	if (on) {
1024 		if (fa != NULL) {
1025 			write_lock_bh(&sk->sk_callback_lock);
1026 			fa->fa_fd = fd;
1027 			write_unlock_bh(&sk->sk_callback_lock);
1028 
1029 			kfree(fna);
1030 			goto out;
1031 		}
1032 		fna->fa_file = filp;
1033 		fna->fa_fd = fd;
1034 		fna->magic = FASYNC_MAGIC;
1035 		fna->fa_next = sock->fasync_list;
1036 		write_lock_bh(&sk->sk_callback_lock);
1037 		sock->fasync_list = fna;
1038 		write_unlock_bh(&sk->sk_callback_lock);
1039 	} else {
1040 		if (fa != NULL) {
1041 			write_lock_bh(&sk->sk_callback_lock);
1042 			*prev = fa->fa_next;
1043 			write_unlock_bh(&sk->sk_callback_lock);
1044 			kfree(fa);
1045 		}
1046 	}
1047 
1048 out:
1049 	release_sock(sock->sk);
1050 	return 0;
1051 }
1052 
1053 /* This function may be called only under socket lock or callback_lock */
1054 
1055 int sock_wake_async(struct socket *sock, int how, int band)
1056 {
1057 	if (!sock || !sock->fasync_list)
1058 		return -1;
1059 	switch (how) {
1060 	case 1:
1061 
1062 		if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1063 			break;
1064 		goto call_kill;
1065 	case 2:
1066 		if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1067 			break;
1068 		/* fall through */
1069 	case 0:
1070 call_kill:
1071 		__kill_fasync(sock->fasync_list, SIGIO, band);
1072 		break;
1073 	case 3:
1074 		__kill_fasync(sock->fasync_list, SIGURG, band);
1075 	}
1076 	return 0;
1077 }
1078 
1079 static int __sock_create(int family, int type, int protocol,
1080 			 struct socket **res, int kern)
1081 {
1082 	int err;
1083 	struct socket *sock;
1084 	const struct net_proto_family *pf;
1085 
1086 	/*
1087 	 *      Check protocol is in range
1088 	 */
1089 	if (family < 0 || family >= NPROTO)
1090 		return -EAFNOSUPPORT;
1091 	if (type < 0 || type >= SOCK_MAX)
1092 		return -EINVAL;
1093 
1094 	/* Compatibility.
1095 
1096 	   This uglymoron is moved from INET layer to here to avoid
1097 	   deadlock in module load.
1098 	 */
1099 	if (family == PF_INET && type == SOCK_PACKET) {
1100 		static int warned;
1101 		if (!warned) {
1102 			warned = 1;
1103 			printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1104 			       current->comm);
1105 		}
1106 		family = PF_PACKET;
1107 	}
1108 
1109 	err = security_socket_create(family, type, protocol, kern);
1110 	if (err)
1111 		return err;
1112 
1113 	/*
1114 	 *	Allocate the socket and allow the family to set things up. if
1115 	 *	the protocol is 0, the family is instructed to select an appropriate
1116 	 *	default.
1117 	 */
1118 	sock = sock_alloc();
1119 	if (!sock) {
1120 		if (net_ratelimit())
1121 			printk(KERN_WARNING "socket: no more sockets\n");
1122 		return -ENFILE;	/* Not exactly a match, but its the
1123 				   closest posix thing */
1124 	}
1125 
1126 	sock->type = type;
1127 
1128 #if defined(CONFIG_KMOD)
1129 	/* Attempt to load a protocol module if the find failed.
1130 	 *
1131 	 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1132 	 * requested real, full-featured networking support upon configuration.
1133 	 * Otherwise module support will break!
1134 	 */
1135 	if (net_families[family] == NULL)
1136 		request_module("net-pf-%d", family);
1137 #endif
1138 
1139 	rcu_read_lock();
1140 	pf = rcu_dereference(net_families[family]);
1141 	err = -EAFNOSUPPORT;
1142 	if (!pf)
1143 		goto out_release;
1144 
1145 	/*
1146 	 * We will call the ->create function, that possibly is in a loadable
1147 	 * module, so we have to bump that loadable module refcnt first.
1148 	 */
1149 	if (!try_module_get(pf->owner))
1150 		goto out_release;
1151 
1152 	/* Now protected by module ref count */
1153 	rcu_read_unlock();
1154 
1155 	err = pf->create(sock, protocol);
1156 	if (err < 0)
1157 		goto out_module_put;
1158 
1159 	/*
1160 	 * Now to bump the refcnt of the [loadable] module that owns this
1161 	 * socket at sock_release time we decrement its refcnt.
1162 	 */
1163 	if (!try_module_get(sock->ops->owner))
1164 		goto out_module_busy;
1165 
1166 	/*
1167 	 * Now that we're done with the ->create function, the [loadable]
1168 	 * module can have its refcnt decremented
1169 	 */
1170 	module_put(pf->owner);
1171 	err = security_socket_post_create(sock, family, type, protocol, kern);
1172 	if (err)
1173 		goto out_release;
1174 	*res = sock;
1175 
1176 	return 0;
1177 
1178 out_module_busy:
1179 	err = -EAFNOSUPPORT;
1180 out_module_put:
1181 	sock->ops = NULL;
1182 	module_put(pf->owner);
1183 out_sock_release:
1184 	sock_release(sock);
1185 	return err;
1186 
1187 out_release:
1188 	rcu_read_unlock();
1189 	goto out_sock_release;
1190 }
1191 
1192 int sock_create(int family, int type, int protocol, struct socket **res)
1193 {
1194 	return __sock_create(family, type, protocol, res, 0);
1195 }
1196 
1197 int sock_create_kern(int family, int type, int protocol, struct socket **res)
1198 {
1199 	return __sock_create(family, type, protocol, res, 1);
1200 }
1201 
1202 asmlinkage long sys_socket(int family, int type, int protocol)
1203 {
1204 	int retval;
1205 	struct socket *sock;
1206 
1207 	retval = sock_create(family, type, protocol, &sock);
1208 	if (retval < 0)
1209 		goto out;
1210 
1211 	retval = sock_map_fd(sock);
1212 	if (retval < 0)
1213 		goto out_release;
1214 
1215 out:
1216 	/* It may be already another descriptor 8) Not kernel problem. */
1217 	return retval;
1218 
1219 out_release:
1220 	sock_release(sock);
1221 	return retval;
1222 }
1223 
1224 /*
1225  *	Create a pair of connected sockets.
1226  */
1227 
1228 asmlinkage long sys_socketpair(int family, int type, int protocol,
1229 			       int __user *usockvec)
1230 {
1231 	struct socket *sock1, *sock2;
1232 	int fd1, fd2, err;
1233 	struct file *newfile1, *newfile2;
1234 
1235 	/*
1236 	 * Obtain the first socket and check if the underlying protocol
1237 	 * supports the socketpair call.
1238 	 */
1239 
1240 	err = sock_create(family, type, protocol, &sock1);
1241 	if (err < 0)
1242 		goto out;
1243 
1244 	err = sock_create(family, type, protocol, &sock2);
1245 	if (err < 0)
1246 		goto out_release_1;
1247 
1248 	err = sock1->ops->socketpair(sock1, sock2);
1249 	if (err < 0)
1250 		goto out_release_both;
1251 
1252 	fd1 = sock_alloc_fd(&newfile1);
1253 	if (unlikely(fd1 < 0))
1254 		goto out_release_both;
1255 
1256 	fd2 = sock_alloc_fd(&newfile2);
1257 	if (unlikely(fd2 < 0)) {
1258 		put_filp(newfile1);
1259 		put_unused_fd(fd1);
1260 		goto out_release_both;
1261 	}
1262 
1263 	err = sock_attach_fd(sock1, newfile1);
1264 	if (unlikely(err < 0)) {
1265 		goto out_fd2;
1266 	}
1267 
1268 	err = sock_attach_fd(sock2, newfile2);
1269 	if (unlikely(err < 0)) {
1270 		fput(newfile1);
1271 		goto out_fd1;
1272 	}
1273 
1274 	err = audit_fd_pair(fd1, fd2);
1275 	if (err < 0) {
1276 		fput(newfile1);
1277 		fput(newfile2);
1278 		goto out_fd;
1279 	}
1280 
1281 	fd_install(fd1, newfile1);
1282 	fd_install(fd2, newfile2);
1283 	/* fd1 and fd2 may be already another descriptors.
1284 	 * Not kernel problem.
1285 	 */
1286 
1287 	err = put_user(fd1, &usockvec[0]);
1288 	if (!err)
1289 		err = put_user(fd2, &usockvec[1]);
1290 	if (!err)
1291 		return 0;
1292 
1293 	sys_close(fd2);
1294 	sys_close(fd1);
1295 	return err;
1296 
1297 out_release_both:
1298 	sock_release(sock2);
1299 out_release_1:
1300 	sock_release(sock1);
1301 out:
1302 	return err;
1303 
1304 out_fd2:
1305 	put_filp(newfile1);
1306 	sock_release(sock1);
1307 out_fd1:
1308 	put_filp(newfile2);
1309 	sock_release(sock2);
1310 out_fd:
1311 	put_unused_fd(fd1);
1312 	put_unused_fd(fd2);
1313 	goto out;
1314 }
1315 
1316 /*
1317  *	Bind a name to a socket. Nothing much to do here since it's
1318  *	the protocol's responsibility to handle the local address.
1319  *
1320  *	We move the socket address to kernel space before we call
1321  *	the protocol layer (having also checked the address is ok).
1322  */
1323 
1324 asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1325 {
1326 	struct socket *sock;
1327 	char address[MAX_SOCK_ADDR];
1328 	int err, fput_needed;
1329 
1330 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1331 	if (sock) {
1332 		err = move_addr_to_kernel(umyaddr, addrlen, address);
1333 		if (err >= 0) {
1334 			err = security_socket_bind(sock,
1335 						   (struct sockaddr *)address,
1336 						   addrlen);
1337 			if (!err)
1338 				err = sock->ops->bind(sock,
1339 						      (struct sockaddr *)
1340 						      address, addrlen);
1341 		}
1342 		fput_light(sock->file, fput_needed);
1343 	}
1344 	return err;
1345 }
1346 
1347 /*
1348  *	Perform a listen. Basically, we allow the protocol to do anything
1349  *	necessary for a listen, and if that works, we mark the socket as
1350  *	ready for listening.
1351  */
1352 
1353 int sysctl_somaxconn __read_mostly = SOMAXCONN;
1354 
1355 asmlinkage long sys_listen(int fd, int backlog)
1356 {
1357 	struct socket *sock;
1358 	int err, fput_needed;
1359 
1360 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1361 	if (sock) {
1362 		if ((unsigned)backlog > sysctl_somaxconn)
1363 			backlog = sysctl_somaxconn;
1364 
1365 		err = security_socket_listen(sock, backlog);
1366 		if (!err)
1367 			err = sock->ops->listen(sock, backlog);
1368 
1369 		fput_light(sock->file, fput_needed);
1370 	}
1371 	return err;
1372 }
1373 
1374 /*
1375  *	For accept, we attempt to create a new socket, set up the link
1376  *	with the client, wake up the client, then return the new
1377  *	connected fd. We collect the address of the connector in kernel
1378  *	space and move it to user at the very end. This is unclean because
1379  *	we open the socket then return an error.
1380  *
1381  *	1003.1g adds the ability to recvmsg() to query connection pending
1382  *	status to recvmsg. We need to add that support in a way thats
1383  *	clean when we restucture accept also.
1384  */
1385 
1386 asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr,
1387 			   int __user *upeer_addrlen)
1388 {
1389 	struct socket *sock, *newsock;
1390 	struct file *newfile;
1391 	int err, len, newfd, fput_needed;
1392 	char address[MAX_SOCK_ADDR];
1393 
1394 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1395 	if (!sock)
1396 		goto out;
1397 
1398 	err = -ENFILE;
1399 	if (!(newsock = sock_alloc()))
1400 		goto out_put;
1401 
1402 	newsock->type = sock->type;
1403 	newsock->ops = sock->ops;
1404 
1405 	/*
1406 	 * We don't need try_module_get here, as the listening socket (sock)
1407 	 * has the protocol module (sock->ops->owner) held.
1408 	 */
1409 	__module_get(newsock->ops->owner);
1410 
1411 	newfd = sock_alloc_fd(&newfile);
1412 	if (unlikely(newfd < 0)) {
1413 		err = newfd;
1414 		sock_release(newsock);
1415 		goto out_put;
1416 	}
1417 
1418 	err = sock_attach_fd(newsock, newfile);
1419 	if (err < 0)
1420 		goto out_fd_simple;
1421 
1422 	err = security_socket_accept(sock, newsock);
1423 	if (err)
1424 		goto out_fd;
1425 
1426 	err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1427 	if (err < 0)
1428 		goto out_fd;
1429 
1430 	if (upeer_sockaddr) {
1431 		if (newsock->ops->getname(newsock, (struct sockaddr *)address,
1432 					  &len, 2) < 0) {
1433 			err = -ECONNABORTED;
1434 			goto out_fd;
1435 		}
1436 		err = move_addr_to_user(address, len, upeer_sockaddr,
1437 					upeer_addrlen);
1438 		if (err < 0)
1439 			goto out_fd;
1440 	}
1441 
1442 	/* File flags are not inherited via accept() unlike another OSes. */
1443 
1444 	fd_install(newfd, newfile);
1445 	err = newfd;
1446 
1447 	security_socket_post_accept(sock, newsock);
1448 
1449 out_put:
1450 	fput_light(sock->file, fput_needed);
1451 out:
1452 	return err;
1453 out_fd_simple:
1454 	sock_release(newsock);
1455 	put_filp(newfile);
1456 	put_unused_fd(newfd);
1457 	goto out_put;
1458 out_fd:
1459 	fput(newfile);
1460 	put_unused_fd(newfd);
1461 	goto out_put;
1462 }
1463 
1464 /*
1465  *	Attempt to connect to a socket with the server address.  The address
1466  *	is in user space so we verify it is OK and move it to kernel space.
1467  *
1468  *	For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1469  *	break bindings
1470  *
1471  *	NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1472  *	other SEQPACKET protocols that take time to connect() as it doesn't
1473  *	include the -EINPROGRESS status for such sockets.
1474  */
1475 
1476 asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr,
1477 			    int addrlen)
1478 {
1479 	struct socket *sock;
1480 	char address[MAX_SOCK_ADDR];
1481 	int err, fput_needed;
1482 
1483 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1484 	if (!sock)
1485 		goto out;
1486 	err = move_addr_to_kernel(uservaddr, addrlen, address);
1487 	if (err < 0)
1488 		goto out_put;
1489 
1490 	err =
1491 	    security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1492 	if (err)
1493 		goto out_put;
1494 
1495 	err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1496 				 sock->file->f_flags);
1497 out_put:
1498 	fput_light(sock->file, fput_needed);
1499 out:
1500 	return err;
1501 }
1502 
1503 /*
1504  *	Get the local address ('name') of a socket object. Move the obtained
1505  *	name to user space.
1506  */
1507 
1508 asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1509 				int __user *usockaddr_len)
1510 {
1511 	struct socket *sock;
1512 	char address[MAX_SOCK_ADDR];
1513 	int len, err, fput_needed;
1514 
1515 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1516 	if (!sock)
1517 		goto out;
1518 
1519 	err = security_socket_getsockname(sock);
1520 	if (err)
1521 		goto out_put;
1522 
1523 	err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 0);
1524 	if (err)
1525 		goto out_put;
1526 	err = move_addr_to_user(address, len, usockaddr, usockaddr_len);
1527 
1528 out_put:
1529 	fput_light(sock->file, fput_needed);
1530 out:
1531 	return err;
1532 }
1533 
1534 /*
1535  *	Get the remote address ('name') of a socket object. Move the obtained
1536  *	name to user space.
1537  */
1538 
1539 asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1540 				int __user *usockaddr_len)
1541 {
1542 	struct socket *sock;
1543 	char address[MAX_SOCK_ADDR];
1544 	int len, err, fput_needed;
1545 
1546 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1547 	if (sock != NULL) {
1548 		err = security_socket_getpeername(sock);
1549 		if (err) {
1550 			fput_light(sock->file, fput_needed);
1551 			return err;
1552 		}
1553 
1554 		err =
1555 		    sock->ops->getname(sock, (struct sockaddr *)address, &len,
1556 				       1);
1557 		if (!err)
1558 			err = move_addr_to_user(address, len, usockaddr,
1559 						usockaddr_len);
1560 		fput_light(sock->file, fput_needed);
1561 	}
1562 	return err;
1563 }
1564 
1565 /*
1566  *	Send a datagram to a given address. We move the address into kernel
1567  *	space and check the user space data area is readable before invoking
1568  *	the protocol.
1569  */
1570 
1571 asmlinkage long sys_sendto(int fd, void __user *buff, size_t len,
1572 			   unsigned flags, struct sockaddr __user *addr,
1573 			   int addr_len)
1574 {
1575 	struct socket *sock;
1576 	char address[MAX_SOCK_ADDR];
1577 	int err;
1578 	struct msghdr msg;
1579 	struct iovec iov;
1580 	int fput_needed;
1581 	struct file *sock_file;
1582 
1583 	sock_file = fget_light(fd, &fput_needed);
1584 	err = -EBADF;
1585 	if (!sock_file)
1586 		goto out;
1587 
1588 	sock = sock_from_file(sock_file, &err);
1589 	if (!sock)
1590 		goto out_put;
1591 	iov.iov_base = buff;
1592 	iov.iov_len = len;
1593 	msg.msg_name = NULL;
1594 	msg.msg_iov = &iov;
1595 	msg.msg_iovlen = 1;
1596 	msg.msg_control = NULL;
1597 	msg.msg_controllen = 0;
1598 	msg.msg_namelen = 0;
1599 	if (addr) {
1600 		err = move_addr_to_kernel(addr, addr_len, address);
1601 		if (err < 0)
1602 			goto out_put;
1603 		msg.msg_name = address;
1604 		msg.msg_namelen = addr_len;
1605 	}
1606 	if (sock->file->f_flags & O_NONBLOCK)
1607 		flags |= MSG_DONTWAIT;
1608 	msg.msg_flags = flags;
1609 	err = sock_sendmsg(sock, &msg, len);
1610 
1611 out_put:
1612 	fput_light(sock_file, fput_needed);
1613 out:
1614 	return err;
1615 }
1616 
1617 /*
1618  *	Send a datagram down a socket.
1619  */
1620 
1621 asmlinkage long sys_send(int fd, void __user *buff, size_t len, unsigned flags)
1622 {
1623 	return sys_sendto(fd, buff, len, flags, NULL, 0);
1624 }
1625 
1626 /*
1627  *	Receive a frame from the socket and optionally record the address of the
1628  *	sender. We verify the buffers are writable and if needed move the
1629  *	sender address from kernel to user space.
1630  */
1631 
1632 asmlinkage long sys_recvfrom(int fd, void __user *ubuf, size_t size,
1633 			     unsigned flags, struct sockaddr __user *addr,
1634 			     int __user *addr_len)
1635 {
1636 	struct socket *sock;
1637 	struct iovec iov;
1638 	struct msghdr msg;
1639 	char address[MAX_SOCK_ADDR];
1640 	int err, err2;
1641 	struct file *sock_file;
1642 	int fput_needed;
1643 
1644 	sock_file = fget_light(fd, &fput_needed);
1645 	err = -EBADF;
1646 	if (!sock_file)
1647 		goto out;
1648 
1649 	sock = sock_from_file(sock_file, &err);
1650 	if (!sock)
1651 		goto out_put;
1652 
1653 	msg.msg_control = NULL;
1654 	msg.msg_controllen = 0;
1655 	msg.msg_iovlen = 1;
1656 	msg.msg_iov = &iov;
1657 	iov.iov_len = size;
1658 	iov.iov_base = ubuf;
1659 	msg.msg_name = address;
1660 	msg.msg_namelen = MAX_SOCK_ADDR;
1661 	if (sock->file->f_flags & O_NONBLOCK)
1662 		flags |= MSG_DONTWAIT;
1663 	err = sock_recvmsg(sock, &msg, size, flags);
1664 
1665 	if (err >= 0 && addr != NULL) {
1666 		err2 = move_addr_to_user(address, msg.msg_namelen, addr, addr_len);
1667 		if (err2 < 0)
1668 			err = err2;
1669 	}
1670 out_put:
1671 	fput_light(sock_file, fput_needed);
1672 out:
1673 	return err;
1674 }
1675 
1676 /*
1677  *	Receive a datagram from a socket.
1678  */
1679 
1680 asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1681 			 unsigned flags)
1682 {
1683 	return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1684 }
1685 
1686 /*
1687  *	Set a socket option. Because we don't know the option lengths we have
1688  *	to pass the user mode parameter for the protocols to sort out.
1689  */
1690 
1691 asmlinkage long sys_setsockopt(int fd, int level, int optname,
1692 			       char __user *optval, int optlen)
1693 {
1694 	int err, fput_needed;
1695 	struct socket *sock;
1696 
1697 	if (optlen < 0)
1698 		return -EINVAL;
1699 
1700 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1701 	if (sock != NULL) {
1702 		err = security_socket_setsockopt(sock, level, optname);
1703 		if (err)
1704 			goto out_put;
1705 
1706 		if (level == SOL_SOCKET)
1707 			err =
1708 			    sock_setsockopt(sock, level, optname, optval,
1709 					    optlen);
1710 		else
1711 			err =
1712 			    sock->ops->setsockopt(sock, level, optname, optval,
1713 						  optlen);
1714 out_put:
1715 		fput_light(sock->file, fput_needed);
1716 	}
1717 	return err;
1718 }
1719 
1720 /*
1721  *	Get a socket option. Because we don't know the option lengths we have
1722  *	to pass a user mode parameter for the protocols to sort out.
1723  */
1724 
1725 asmlinkage long sys_getsockopt(int fd, int level, int optname,
1726 			       char __user *optval, int __user *optlen)
1727 {
1728 	int err, fput_needed;
1729 	struct socket *sock;
1730 
1731 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1732 	if (sock != NULL) {
1733 		err = security_socket_getsockopt(sock, level, optname);
1734 		if (err)
1735 			goto out_put;
1736 
1737 		if (level == SOL_SOCKET)
1738 			err =
1739 			    sock_getsockopt(sock, level, optname, optval,
1740 					    optlen);
1741 		else
1742 			err =
1743 			    sock->ops->getsockopt(sock, level, optname, optval,
1744 						  optlen);
1745 out_put:
1746 		fput_light(sock->file, fput_needed);
1747 	}
1748 	return err;
1749 }
1750 
1751 /*
1752  *	Shutdown a socket.
1753  */
1754 
1755 asmlinkage long sys_shutdown(int fd, int how)
1756 {
1757 	int err, fput_needed;
1758 	struct socket *sock;
1759 
1760 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1761 	if (sock != NULL) {
1762 		err = security_socket_shutdown(sock, how);
1763 		if (!err)
1764 			err = sock->ops->shutdown(sock, how);
1765 		fput_light(sock->file, fput_needed);
1766 	}
1767 	return err;
1768 }
1769 
1770 /* A couple of helpful macros for getting the address of the 32/64 bit
1771  * fields which are the same type (int / unsigned) on our platforms.
1772  */
1773 #define COMPAT_MSG(msg, member)	((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1774 #define COMPAT_NAMELEN(msg)	COMPAT_MSG(msg, msg_namelen)
1775 #define COMPAT_FLAGS(msg)	COMPAT_MSG(msg, msg_flags)
1776 
1777 /*
1778  *	BSD sendmsg interface
1779  */
1780 
1781 asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1782 {
1783 	struct compat_msghdr __user *msg_compat =
1784 	    (struct compat_msghdr __user *)msg;
1785 	struct socket *sock;
1786 	char address[MAX_SOCK_ADDR];
1787 	struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1788 	unsigned char ctl[sizeof(struct cmsghdr) + 20]
1789 	    __attribute__ ((aligned(sizeof(__kernel_size_t))));
1790 	/* 20 is size of ipv6_pktinfo */
1791 	unsigned char *ctl_buf = ctl;
1792 	struct msghdr msg_sys;
1793 	int err, ctl_len, iov_size, total_len;
1794 	int fput_needed;
1795 
1796 	err = -EFAULT;
1797 	if (MSG_CMSG_COMPAT & flags) {
1798 		if (get_compat_msghdr(&msg_sys, msg_compat))
1799 			return -EFAULT;
1800 	}
1801 	else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1802 		return -EFAULT;
1803 
1804 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1805 	if (!sock)
1806 		goto out;
1807 
1808 	/* do not move before msg_sys is valid */
1809 	err = -EMSGSIZE;
1810 	if (msg_sys.msg_iovlen > UIO_MAXIOV)
1811 		goto out_put;
1812 
1813 	/* Check whether to allocate the iovec area */
1814 	err = -ENOMEM;
1815 	iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1816 	if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1817 		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1818 		if (!iov)
1819 			goto out_put;
1820 	}
1821 
1822 	/* This will also move the address data into kernel space */
1823 	if (MSG_CMSG_COMPAT & flags) {
1824 		err = verify_compat_iovec(&msg_sys, iov, address, VERIFY_READ);
1825 	} else
1826 		err = verify_iovec(&msg_sys, iov, address, VERIFY_READ);
1827 	if (err < 0)
1828 		goto out_freeiov;
1829 	total_len = err;
1830 
1831 	err = -ENOBUFS;
1832 
1833 	if (msg_sys.msg_controllen > INT_MAX)
1834 		goto out_freeiov;
1835 	ctl_len = msg_sys.msg_controllen;
1836 	if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1837 		err =
1838 		    cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1839 						     sizeof(ctl));
1840 		if (err)
1841 			goto out_freeiov;
1842 		ctl_buf = msg_sys.msg_control;
1843 		ctl_len = msg_sys.msg_controllen;
1844 	} else if (ctl_len) {
1845 		if (ctl_len > sizeof(ctl)) {
1846 			ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1847 			if (ctl_buf == NULL)
1848 				goto out_freeiov;
1849 		}
1850 		err = -EFAULT;
1851 		/*
1852 		 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1853 		 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1854 		 * checking falls down on this.
1855 		 */
1856 		if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1857 				   ctl_len))
1858 			goto out_freectl;
1859 		msg_sys.msg_control = ctl_buf;
1860 	}
1861 	msg_sys.msg_flags = flags;
1862 
1863 	if (sock->file->f_flags & O_NONBLOCK)
1864 		msg_sys.msg_flags |= MSG_DONTWAIT;
1865 	err = sock_sendmsg(sock, &msg_sys, total_len);
1866 
1867 out_freectl:
1868 	if (ctl_buf != ctl)
1869 		sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1870 out_freeiov:
1871 	if (iov != iovstack)
1872 		sock_kfree_s(sock->sk, iov, iov_size);
1873 out_put:
1874 	fput_light(sock->file, fput_needed);
1875 out:
1876 	return err;
1877 }
1878 
1879 /*
1880  *	BSD recvmsg interface
1881  */
1882 
1883 asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg,
1884 			    unsigned int flags)
1885 {
1886 	struct compat_msghdr __user *msg_compat =
1887 	    (struct compat_msghdr __user *)msg;
1888 	struct socket *sock;
1889 	struct iovec iovstack[UIO_FASTIOV];
1890 	struct iovec *iov = iovstack;
1891 	struct msghdr msg_sys;
1892 	unsigned long cmsg_ptr;
1893 	int err, iov_size, total_len, len;
1894 	int fput_needed;
1895 
1896 	/* kernel mode address */
1897 	char addr[MAX_SOCK_ADDR];
1898 
1899 	/* user mode address pointers */
1900 	struct sockaddr __user *uaddr;
1901 	int __user *uaddr_len;
1902 
1903 	if (MSG_CMSG_COMPAT & flags) {
1904 		if (get_compat_msghdr(&msg_sys, msg_compat))
1905 			return -EFAULT;
1906 	}
1907 	else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1908 		return -EFAULT;
1909 
1910 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1911 	if (!sock)
1912 		goto out;
1913 
1914 	err = -EMSGSIZE;
1915 	if (msg_sys.msg_iovlen > UIO_MAXIOV)
1916 		goto out_put;
1917 
1918 	/* Check whether to allocate the iovec area */
1919 	err = -ENOMEM;
1920 	iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1921 	if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1922 		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1923 		if (!iov)
1924 			goto out_put;
1925 	}
1926 
1927 	/*
1928 	 *      Save the user-mode address (verify_iovec will change the
1929 	 *      kernel msghdr to use the kernel address space)
1930 	 */
1931 
1932 	uaddr = (void __user *)msg_sys.msg_name;
1933 	uaddr_len = COMPAT_NAMELEN(msg);
1934 	if (MSG_CMSG_COMPAT & flags) {
1935 		err = verify_compat_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1936 	} else
1937 		err = verify_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1938 	if (err < 0)
1939 		goto out_freeiov;
1940 	total_len = err;
1941 
1942 	cmsg_ptr = (unsigned long)msg_sys.msg_control;
1943 	msg_sys.msg_flags = 0;
1944 	if (MSG_CMSG_COMPAT & flags)
1945 		msg_sys.msg_flags = MSG_CMSG_COMPAT;
1946 
1947 	if (sock->file->f_flags & O_NONBLOCK)
1948 		flags |= MSG_DONTWAIT;
1949 	err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1950 	if (err < 0)
1951 		goto out_freeiov;
1952 	len = err;
1953 
1954 	if (uaddr != NULL) {
1955 		err = move_addr_to_user(addr, msg_sys.msg_namelen, uaddr,
1956 					uaddr_len);
1957 		if (err < 0)
1958 			goto out_freeiov;
1959 	}
1960 	err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
1961 			 COMPAT_FLAGS(msg));
1962 	if (err)
1963 		goto out_freeiov;
1964 	if (MSG_CMSG_COMPAT & flags)
1965 		err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
1966 				 &msg_compat->msg_controllen);
1967 	else
1968 		err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
1969 				 &msg->msg_controllen);
1970 	if (err)
1971 		goto out_freeiov;
1972 	err = len;
1973 
1974 out_freeiov:
1975 	if (iov != iovstack)
1976 		sock_kfree_s(sock->sk, iov, iov_size);
1977 out_put:
1978 	fput_light(sock->file, fput_needed);
1979 out:
1980 	return err;
1981 }
1982 
1983 #ifdef __ARCH_WANT_SYS_SOCKETCALL
1984 
1985 /* Argument list sizes for sys_socketcall */
1986 #define AL(x) ((x) * sizeof(unsigned long))
1987 static const unsigned char nargs[18]={
1988 	AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
1989 	AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
1990 	AL(6),AL(2),AL(5),AL(5),AL(3),AL(3)
1991 };
1992 
1993 #undef AL
1994 
1995 /*
1996  *	System call vectors.
1997  *
1998  *	Argument checking cleaned up. Saved 20% in size.
1999  *  This function doesn't need to set the kernel lock because
2000  *  it is set by the callees.
2001  */
2002 
2003 asmlinkage long sys_socketcall(int call, unsigned long __user *args)
2004 {
2005 	unsigned long a[6];
2006 	unsigned long a0, a1;
2007 	int err;
2008 
2009 	if (call < 1 || call > SYS_RECVMSG)
2010 		return -EINVAL;
2011 
2012 	/* copy_from_user should be SMP safe. */
2013 	if (copy_from_user(a, args, nargs[call]))
2014 		return -EFAULT;
2015 
2016 	err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2017 	if (err)
2018 		return err;
2019 
2020 	a0 = a[0];
2021 	a1 = a[1];
2022 
2023 	switch (call) {
2024 	case SYS_SOCKET:
2025 		err = sys_socket(a0, a1, a[2]);
2026 		break;
2027 	case SYS_BIND:
2028 		err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2029 		break;
2030 	case SYS_CONNECT:
2031 		err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2032 		break;
2033 	case SYS_LISTEN:
2034 		err = sys_listen(a0, a1);
2035 		break;
2036 	case SYS_ACCEPT:
2037 		err =
2038 		    sys_accept(a0, (struct sockaddr __user *)a1,
2039 			       (int __user *)a[2]);
2040 		break;
2041 	case SYS_GETSOCKNAME:
2042 		err =
2043 		    sys_getsockname(a0, (struct sockaddr __user *)a1,
2044 				    (int __user *)a[2]);
2045 		break;
2046 	case SYS_GETPEERNAME:
2047 		err =
2048 		    sys_getpeername(a0, (struct sockaddr __user *)a1,
2049 				    (int __user *)a[2]);
2050 		break;
2051 	case SYS_SOCKETPAIR:
2052 		err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2053 		break;
2054 	case SYS_SEND:
2055 		err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2056 		break;
2057 	case SYS_SENDTO:
2058 		err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2059 				 (struct sockaddr __user *)a[4], a[5]);
2060 		break;
2061 	case SYS_RECV:
2062 		err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2063 		break;
2064 	case SYS_RECVFROM:
2065 		err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2066 				   (struct sockaddr __user *)a[4],
2067 				   (int __user *)a[5]);
2068 		break;
2069 	case SYS_SHUTDOWN:
2070 		err = sys_shutdown(a0, a1);
2071 		break;
2072 	case SYS_SETSOCKOPT:
2073 		err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2074 		break;
2075 	case SYS_GETSOCKOPT:
2076 		err =
2077 		    sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2078 				   (int __user *)a[4]);
2079 		break;
2080 	case SYS_SENDMSG:
2081 		err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2082 		break;
2083 	case SYS_RECVMSG:
2084 		err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2085 		break;
2086 	default:
2087 		err = -EINVAL;
2088 		break;
2089 	}
2090 	return err;
2091 }
2092 
2093 #endif				/* __ARCH_WANT_SYS_SOCKETCALL */
2094 
2095 /**
2096  *	sock_register - add a socket protocol handler
2097  *	@ops: description of protocol
2098  *
2099  *	This function is called by a protocol handler that wants to
2100  *	advertise its address family, and have it linked into the
2101  *	socket interface. The value ops->family coresponds to the
2102  *	socket system call protocol family.
2103  */
2104 int sock_register(const struct net_proto_family *ops)
2105 {
2106 	int err;
2107 
2108 	if (ops->family >= NPROTO) {
2109 		printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2110 		       NPROTO);
2111 		return -ENOBUFS;
2112 	}
2113 
2114 	spin_lock(&net_family_lock);
2115 	if (net_families[ops->family])
2116 		err = -EEXIST;
2117 	else {
2118 		net_families[ops->family] = ops;
2119 		err = 0;
2120 	}
2121 	spin_unlock(&net_family_lock);
2122 
2123 	printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2124 	return err;
2125 }
2126 
2127 /**
2128  *	sock_unregister - remove a protocol handler
2129  *	@family: protocol family to remove
2130  *
2131  *	This function is called by a protocol handler that wants to
2132  *	remove its address family, and have it unlinked from the
2133  *	new socket creation.
2134  *
2135  *	If protocol handler is a module, then it can use module reference
2136  *	counts to protect against new references. If protocol handler is not
2137  *	a module then it needs to provide its own protection in
2138  *	the ops->create routine.
2139  */
2140 void sock_unregister(int family)
2141 {
2142 	BUG_ON(family < 0 || family >= NPROTO);
2143 
2144 	spin_lock(&net_family_lock);
2145 	net_families[family] = NULL;
2146 	spin_unlock(&net_family_lock);
2147 
2148 	synchronize_rcu();
2149 
2150 	printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2151 }
2152 
2153 static int __init sock_init(void)
2154 {
2155 	/*
2156 	 *      Initialize sock SLAB cache.
2157 	 */
2158 
2159 	sk_init();
2160 
2161 	/*
2162 	 *      Initialize skbuff SLAB cache
2163 	 */
2164 	skb_init();
2165 
2166 	/*
2167 	 *      Initialize the protocols module.
2168 	 */
2169 
2170 	init_inodecache();
2171 	register_filesystem(&sock_fs_type);
2172 	sock_mnt = kern_mount(&sock_fs_type);
2173 
2174 	/* The real protocol initialization is performed in later initcalls.
2175 	 */
2176 
2177 #ifdef CONFIG_NETFILTER
2178 	netfilter_init();
2179 #endif
2180 
2181 	return 0;
2182 }
2183 
2184 core_initcall(sock_init);	/* early initcall */
2185 
2186 #ifdef CONFIG_PROC_FS
2187 void socket_seq_show(struct seq_file *seq)
2188 {
2189 	int cpu;
2190 	int counter = 0;
2191 
2192 	for_each_possible_cpu(cpu)
2193 	    counter += per_cpu(sockets_in_use, cpu);
2194 
2195 	/* It can be negative, by the way. 8) */
2196 	if (counter < 0)
2197 		counter = 0;
2198 
2199 	seq_printf(seq, "sockets: used %d\n", counter);
2200 }
2201 #endif				/* CONFIG_PROC_FS */
2202 
2203 #ifdef CONFIG_COMPAT
2204 static long compat_sock_ioctl(struct file *file, unsigned cmd,
2205 			      unsigned long arg)
2206 {
2207 	struct socket *sock = file->private_data;
2208 	int ret = -ENOIOCTLCMD;
2209 
2210 	if (sock->ops->compat_ioctl)
2211 		ret = sock->ops->compat_ioctl(sock, cmd, arg);
2212 
2213 	return ret;
2214 }
2215 #endif
2216 
2217 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2218 {
2219 	return sock->ops->bind(sock, addr, addrlen);
2220 }
2221 
2222 int kernel_listen(struct socket *sock, int backlog)
2223 {
2224 	return sock->ops->listen(sock, backlog);
2225 }
2226 
2227 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2228 {
2229 	struct sock *sk = sock->sk;
2230 	int err;
2231 
2232 	err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2233 			       newsock);
2234 	if (err < 0)
2235 		goto done;
2236 
2237 	err = sock->ops->accept(sock, *newsock, flags);
2238 	if (err < 0) {
2239 		sock_release(*newsock);
2240 		goto done;
2241 	}
2242 
2243 	(*newsock)->ops = sock->ops;
2244 
2245 done:
2246 	return err;
2247 }
2248 
2249 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
2250 		   int flags)
2251 {
2252 	return sock->ops->connect(sock, addr, addrlen, flags);
2253 }
2254 
2255 int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2256 			 int *addrlen)
2257 {
2258 	return sock->ops->getname(sock, addr, addrlen, 0);
2259 }
2260 
2261 int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2262 			 int *addrlen)
2263 {
2264 	return sock->ops->getname(sock, addr, addrlen, 1);
2265 }
2266 
2267 int kernel_getsockopt(struct socket *sock, int level, int optname,
2268 			char *optval, int *optlen)
2269 {
2270 	mm_segment_t oldfs = get_fs();
2271 	int err;
2272 
2273 	set_fs(KERNEL_DS);
2274 	if (level == SOL_SOCKET)
2275 		err = sock_getsockopt(sock, level, optname, optval, optlen);
2276 	else
2277 		err = sock->ops->getsockopt(sock, level, optname, optval,
2278 					    optlen);
2279 	set_fs(oldfs);
2280 	return err;
2281 }
2282 
2283 int kernel_setsockopt(struct socket *sock, int level, int optname,
2284 			char *optval, int optlen)
2285 {
2286 	mm_segment_t oldfs = get_fs();
2287 	int err;
2288 
2289 	set_fs(KERNEL_DS);
2290 	if (level == SOL_SOCKET)
2291 		err = sock_setsockopt(sock, level, optname, optval, optlen);
2292 	else
2293 		err = sock->ops->setsockopt(sock, level, optname, optval,
2294 					    optlen);
2295 	set_fs(oldfs);
2296 	return err;
2297 }
2298 
2299 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2300 		    size_t size, int flags)
2301 {
2302 	if (sock->ops->sendpage)
2303 		return sock->ops->sendpage(sock, page, offset, size, flags);
2304 
2305 	return sock_no_sendpage(sock, page, offset, size, flags);
2306 }
2307 
2308 int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2309 {
2310 	mm_segment_t oldfs = get_fs();
2311 	int err;
2312 
2313 	set_fs(KERNEL_DS);
2314 	err = sock->ops->ioctl(sock, cmd, arg);
2315 	set_fs(oldfs);
2316 
2317 	return err;
2318 }
2319 
2320 /* ABI emulation layers need these two */
2321 EXPORT_SYMBOL(move_addr_to_kernel);
2322 EXPORT_SYMBOL(move_addr_to_user);
2323 EXPORT_SYMBOL(sock_create);
2324 EXPORT_SYMBOL(sock_create_kern);
2325 EXPORT_SYMBOL(sock_create_lite);
2326 EXPORT_SYMBOL(sock_map_fd);
2327 EXPORT_SYMBOL(sock_recvmsg);
2328 EXPORT_SYMBOL(sock_register);
2329 EXPORT_SYMBOL(sock_release);
2330 EXPORT_SYMBOL(sock_sendmsg);
2331 EXPORT_SYMBOL(sock_unregister);
2332 EXPORT_SYMBOL(sock_wake_async);
2333 EXPORT_SYMBOL(sockfd_lookup);
2334 EXPORT_SYMBOL(kernel_sendmsg);
2335 EXPORT_SYMBOL(kernel_recvmsg);
2336 EXPORT_SYMBOL(kernel_bind);
2337 EXPORT_SYMBOL(kernel_listen);
2338 EXPORT_SYMBOL(kernel_accept);
2339 EXPORT_SYMBOL(kernel_connect);
2340 EXPORT_SYMBOL(kernel_getsockname);
2341 EXPORT_SYMBOL(kernel_getpeername);
2342 EXPORT_SYMBOL(kernel_getsockopt);
2343 EXPORT_SYMBOL(kernel_setsockopt);
2344 EXPORT_SYMBOL(kernel_sendpage);
2345 EXPORT_SYMBOL(kernel_sock_ioctl);
2346