xref: /linux/net/socket.c (revision d524dac9279b6a41ffdf7ff7958c577f2e387db6)
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/thread_info.h>
67 #include <linux/rcupdate.h>
68 #include <linux/netdevice.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/mutex.h>
72 #include <linux/wanrouter.h>
73 #include <linux/if_bridge.h>
74 #include <linux/if_frad.h>
75 #include <linux/if_vlan.h>
76 #include <linux/init.h>
77 #include <linux/poll.h>
78 #include <linux/cache.h>
79 #include <linux/module.h>
80 #include <linux/highmem.h>
81 #include <linux/mount.h>
82 #include <linux/security.h>
83 #include <linux/syscalls.h>
84 #include <linux/compat.h>
85 #include <linux/kmod.h>
86 #include <linux/audit.h>
87 #include <linux/wireless.h>
88 #include <linux/nsproxy.h>
89 #include <linux/magic.h>
90 #include <linux/slab.h>
91 
92 #include <asm/uaccess.h>
93 #include <asm/unistd.h>
94 
95 #include <net/compat.h>
96 #include <net/wext.h>
97 #include <net/cls_cgroup.h>
98 
99 #include <net/sock.h>
100 #include <linux/netfilter.h>
101 
102 #include <linux/if_tun.h>
103 #include <linux/ipv6_route.h>
104 #include <linux/route.h>
105 #include <linux/sockios.h>
106 #include <linux/atalk.h>
107 
108 static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
109 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
110 			 unsigned long nr_segs, loff_t pos);
111 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
112 			  unsigned long nr_segs, loff_t pos);
113 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
114 
115 static int sock_close(struct inode *inode, struct file *file);
116 static unsigned int sock_poll(struct file *file,
117 			      struct poll_table_struct *wait);
118 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
119 #ifdef CONFIG_COMPAT
120 static long compat_sock_ioctl(struct file *file,
121 			      unsigned int cmd, unsigned long arg);
122 #endif
123 static int sock_fasync(int fd, struct file *filp, int on);
124 static ssize_t sock_sendpage(struct file *file, struct page *page,
125 			     int offset, size_t size, loff_t *ppos, int more);
126 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
127 				struct pipe_inode_info *pipe, size_t len,
128 				unsigned int flags);
129 
130 /*
131  *	Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
132  *	in the operation structures but are done directly via the socketcall() multiplexor.
133  */
134 
135 static const struct file_operations socket_file_ops = {
136 	.owner =	THIS_MODULE,
137 	.llseek =	no_llseek,
138 	.aio_read =	sock_aio_read,
139 	.aio_write =	sock_aio_write,
140 	.poll =		sock_poll,
141 	.unlocked_ioctl = sock_ioctl,
142 #ifdef CONFIG_COMPAT
143 	.compat_ioctl = compat_sock_ioctl,
144 #endif
145 	.mmap =		sock_mmap,
146 	.open =		sock_no_open,	/* special open code to disallow open via /proc */
147 	.release =	sock_close,
148 	.fasync =	sock_fasync,
149 	.sendpage =	sock_sendpage,
150 	.splice_write = generic_splice_sendpage,
151 	.splice_read =	sock_splice_read,
152 };
153 
154 /*
155  *	The protocol list. Each protocol is registered in here.
156  */
157 
158 static DEFINE_SPINLOCK(net_family_lock);
159 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
160 
161 /*
162  *	Statistics counters of the socket lists
163  */
164 
165 static DEFINE_PER_CPU(int, sockets_in_use);
166 
167 /*
168  * Support routines.
169  * Move socket addresses back and forth across the kernel/user
170  * divide and look after the messy bits.
171  */
172 
173 /**
174  *	move_addr_to_kernel	-	copy a socket address into kernel space
175  *	@uaddr: Address in user space
176  *	@kaddr: Address in kernel space
177  *	@ulen: Length in user space
178  *
179  *	The address is copied into kernel space. If the provided address is
180  *	too long an error code of -EINVAL is returned. If the copy gives
181  *	invalid addresses -EFAULT is returned. On a success 0 is returned.
182  */
183 
184 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
185 {
186 	if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
187 		return -EINVAL;
188 	if (ulen == 0)
189 		return 0;
190 	if (copy_from_user(kaddr, uaddr, ulen))
191 		return -EFAULT;
192 	return audit_sockaddr(ulen, kaddr);
193 }
194 
195 /**
196  *	move_addr_to_user	-	copy an address to user space
197  *	@kaddr: kernel space address
198  *	@klen: length of address in kernel
199  *	@uaddr: user space address
200  *	@ulen: pointer to user length field
201  *
202  *	The value pointed to by ulen on entry is the buffer length available.
203  *	This is overwritten with the buffer space used. -EINVAL is returned
204  *	if an overlong buffer is specified or a negative buffer size. -EFAULT
205  *	is returned if either the buffer or the length field are not
206  *	accessible.
207  *	After copying the data up to the limit the user specifies, the true
208  *	length of the data is written over the length limit the user
209  *	specified. Zero is returned for a success.
210  */
211 
212 static int move_addr_to_user(struct sockaddr *kaddr, int klen,
213 			     void __user *uaddr, int __user *ulen)
214 {
215 	int err;
216 	int len;
217 
218 	err = get_user(len, ulen);
219 	if (err)
220 		return err;
221 	if (len > klen)
222 		len = klen;
223 	if (len < 0 || len > sizeof(struct sockaddr_storage))
224 		return -EINVAL;
225 	if (len) {
226 		if (audit_sockaddr(klen, kaddr))
227 			return -ENOMEM;
228 		if (copy_to_user(uaddr, kaddr, len))
229 			return -EFAULT;
230 	}
231 	/*
232 	 *      "fromlen shall refer to the value before truncation.."
233 	 *                      1003.1g
234 	 */
235 	return __put_user(klen, ulen);
236 }
237 
238 static struct kmem_cache *sock_inode_cachep __read_mostly;
239 
240 static struct inode *sock_alloc_inode(struct super_block *sb)
241 {
242 	struct socket_alloc *ei;
243 
244 	ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
245 	if (!ei)
246 		return NULL;
247 	ei->socket.wq = kmalloc(sizeof(struct socket_wq), GFP_KERNEL);
248 	if (!ei->socket.wq) {
249 		kmem_cache_free(sock_inode_cachep, ei);
250 		return NULL;
251 	}
252 	init_waitqueue_head(&ei->socket.wq->wait);
253 	ei->socket.wq->fasync_list = NULL;
254 
255 	ei->socket.state = SS_UNCONNECTED;
256 	ei->socket.flags = 0;
257 	ei->socket.ops = NULL;
258 	ei->socket.sk = NULL;
259 	ei->socket.file = NULL;
260 
261 	return &ei->vfs_inode;
262 }
263 
264 
265 
266 static void wq_free_rcu(struct rcu_head *head)
267 {
268 	struct socket_wq *wq = container_of(head, struct socket_wq, rcu);
269 
270 	kfree(wq);
271 }
272 
273 static void sock_destroy_inode(struct inode *inode)
274 {
275 	struct socket_alloc *ei;
276 
277 	ei = container_of(inode, struct socket_alloc, vfs_inode);
278 	call_rcu(&ei->socket.wq->rcu, wq_free_rcu);
279 	kmem_cache_free(sock_inode_cachep, ei);
280 }
281 
282 static void init_once(void *foo)
283 {
284 	struct socket_alloc *ei = (struct socket_alloc *)foo;
285 
286 	inode_init_once(&ei->vfs_inode);
287 }
288 
289 static int init_inodecache(void)
290 {
291 	sock_inode_cachep = kmem_cache_create("sock_inode_cache",
292 					      sizeof(struct socket_alloc),
293 					      0,
294 					      (SLAB_HWCACHE_ALIGN |
295 					       SLAB_RECLAIM_ACCOUNT |
296 					       SLAB_MEM_SPREAD),
297 					      init_once);
298 	if (sock_inode_cachep == NULL)
299 		return -ENOMEM;
300 	return 0;
301 }
302 
303 static const struct super_operations sockfs_ops = {
304 	.alloc_inode	= sock_alloc_inode,
305 	.destroy_inode	= sock_destroy_inode,
306 	.statfs		= simple_statfs,
307 };
308 
309 /*
310  * sockfs_dname() is called from d_path().
311  */
312 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
313 {
314 	return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
315 				dentry->d_inode->i_ino);
316 }
317 
318 static const struct dentry_operations sockfs_dentry_operations = {
319 	.d_dname  = sockfs_dname,
320 };
321 
322 static struct dentry *sockfs_mount(struct file_system_type *fs_type,
323 			 int flags, const char *dev_name, void *data)
324 {
325 	return mount_pseudo(fs_type, "socket:", &sockfs_ops,
326 		&sockfs_dentry_operations, SOCKFS_MAGIC);
327 }
328 
329 static struct vfsmount *sock_mnt __read_mostly;
330 
331 static struct file_system_type sock_fs_type = {
332 	.name =		"sockfs",
333 	.mount =	sockfs_mount,
334 	.kill_sb =	kill_anon_super,
335 };
336 
337 /*
338  *	Obtains the first available file descriptor and sets it up for use.
339  *
340  *	These functions create file structures and maps them to fd space
341  *	of the current process. On success it returns file descriptor
342  *	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 
354 static int sock_alloc_file(struct socket *sock, struct file **f, int flags)
355 {
356 	struct qstr name = { .name = "" };
357 	struct path path;
358 	struct file *file;
359 	int fd;
360 
361 	fd = get_unused_fd_flags(flags);
362 	if (unlikely(fd < 0))
363 		return fd;
364 
365 	path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
366 	if (unlikely(!path.dentry)) {
367 		put_unused_fd(fd);
368 		return -ENOMEM;
369 	}
370 	path.mnt = mntget(sock_mnt);
371 
372 	d_instantiate(path.dentry, SOCK_INODE(sock));
373 	SOCK_INODE(sock)->i_fop = &socket_file_ops;
374 
375 	file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
376 		  &socket_file_ops);
377 	if (unlikely(!file)) {
378 		/* drop dentry, keep inode */
379 		ihold(path.dentry->d_inode);
380 		path_put(&path);
381 		put_unused_fd(fd);
382 		return -ENFILE;
383 	}
384 
385 	sock->file = file;
386 	file->f_flags = O_RDWR | (flags & O_NONBLOCK);
387 	file->f_pos = 0;
388 	file->private_data = sock;
389 
390 	*f = file;
391 	return fd;
392 }
393 
394 int sock_map_fd(struct socket *sock, int flags)
395 {
396 	struct file *newfile;
397 	int fd = sock_alloc_file(sock, &newfile, flags);
398 
399 	if (likely(fd >= 0))
400 		fd_install(fd, newfile);
401 
402 	return fd;
403 }
404 EXPORT_SYMBOL(sock_map_fd);
405 
406 static struct socket *sock_from_file(struct file *file, int *err)
407 {
408 	if (file->f_op == &socket_file_ops)
409 		return file->private_data;	/* set in sock_map_fd */
410 
411 	*err = -ENOTSOCK;
412 	return NULL;
413 }
414 
415 /**
416  *	sockfd_lookup - Go from a file number to its socket slot
417  *	@fd: file handle
418  *	@err: pointer to an error code return
419  *
420  *	The file handle passed in is locked and the socket it is bound
421  *	too is returned. If an error occurs the err pointer is overwritten
422  *	with a negative errno code and NULL is returned. The function checks
423  *	for both invalid handles and passing a handle which is not a socket.
424  *
425  *	On a success the socket object pointer is returned.
426  */
427 
428 struct socket *sockfd_lookup(int fd, int *err)
429 {
430 	struct file *file;
431 	struct socket *sock;
432 
433 	file = fget(fd);
434 	if (!file) {
435 		*err = -EBADF;
436 		return NULL;
437 	}
438 
439 	sock = sock_from_file(file, err);
440 	if (!sock)
441 		fput(file);
442 	return sock;
443 }
444 EXPORT_SYMBOL(sockfd_lookup);
445 
446 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
447 {
448 	struct file *file;
449 	struct socket *sock;
450 
451 	*err = -EBADF;
452 	file = fget_light(fd, fput_needed);
453 	if (file) {
454 		sock = sock_from_file(file, err);
455 		if (sock)
456 			return sock;
457 		fput_light(file, *fput_needed);
458 	}
459 	return NULL;
460 }
461 
462 /**
463  *	sock_alloc	-	allocate a socket
464  *
465  *	Allocate a new inode and socket object. The two are bound together
466  *	and initialised. The socket is then returned. If we are out of inodes
467  *	NULL is returned.
468  */
469 
470 static struct socket *sock_alloc(void)
471 {
472 	struct inode *inode;
473 	struct socket *sock;
474 
475 	inode = new_inode(sock_mnt->mnt_sb);
476 	if (!inode)
477 		return NULL;
478 
479 	sock = SOCKET_I(inode);
480 
481 	kmemcheck_annotate_bitfield(sock, type);
482 	inode->i_ino = get_next_ino();
483 	inode->i_mode = S_IFSOCK | S_IRWXUGO;
484 	inode->i_uid = current_fsuid();
485 	inode->i_gid = current_fsgid();
486 
487 	percpu_add(sockets_in_use, 1);
488 	return sock;
489 }
490 
491 /*
492  *	In theory you can't get an open on this inode, but /proc provides
493  *	a back door. Remember to keep it shut otherwise you'll let the
494  *	creepy crawlies in.
495  */
496 
497 static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
498 {
499 	return -ENXIO;
500 }
501 
502 const struct file_operations bad_sock_fops = {
503 	.owner = THIS_MODULE,
504 	.open = sock_no_open,
505 	.llseek = noop_llseek,
506 };
507 
508 /**
509  *	sock_release	-	close a socket
510  *	@sock: socket to close
511  *
512  *	The socket is released from the protocol stack if it has a release
513  *	callback, and the inode is then released if the socket is bound to
514  *	an inode not a file.
515  */
516 
517 void sock_release(struct socket *sock)
518 {
519 	if (sock->ops) {
520 		struct module *owner = sock->ops->owner;
521 
522 		sock->ops->release(sock);
523 		sock->ops = NULL;
524 		module_put(owner);
525 	}
526 
527 	if (sock->wq->fasync_list)
528 		printk(KERN_ERR "sock_release: fasync list not empty!\n");
529 
530 	percpu_sub(sockets_in_use, 1);
531 	if (!sock->file) {
532 		iput(SOCK_INODE(sock));
533 		return;
534 	}
535 	sock->file = NULL;
536 }
537 EXPORT_SYMBOL(sock_release);
538 
539 int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags)
540 {
541 	*tx_flags = 0;
542 	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
543 		*tx_flags |= SKBTX_HW_TSTAMP;
544 	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
545 		*tx_flags |= SKBTX_SW_TSTAMP;
546 	return 0;
547 }
548 EXPORT_SYMBOL(sock_tx_timestamp);
549 
550 static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
551 				 struct msghdr *msg, size_t size)
552 {
553 	struct sock_iocb *si = kiocb_to_siocb(iocb);
554 	int err;
555 
556 	sock_update_classid(sock->sk);
557 
558 	si->sock = sock;
559 	si->scm = NULL;
560 	si->msg = msg;
561 	si->size = size;
562 
563 	err = security_socket_sendmsg(sock, msg, size);
564 	if (err)
565 		return err;
566 
567 	return sock->ops->sendmsg(iocb, sock, msg, size);
568 }
569 
570 int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
571 {
572 	struct kiocb iocb;
573 	struct sock_iocb siocb;
574 	int ret;
575 
576 	init_sync_kiocb(&iocb, NULL);
577 	iocb.private = &siocb;
578 	ret = __sock_sendmsg(&iocb, sock, msg, size);
579 	if (-EIOCBQUEUED == ret)
580 		ret = wait_on_sync_kiocb(&iocb);
581 	return ret;
582 }
583 EXPORT_SYMBOL(sock_sendmsg);
584 
585 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
586 		   struct kvec *vec, size_t num, size_t size)
587 {
588 	mm_segment_t oldfs = get_fs();
589 	int result;
590 
591 	set_fs(KERNEL_DS);
592 	/*
593 	 * the following is safe, since for compiler definitions of kvec and
594 	 * iovec are identical, yielding the same in-core layout and alignment
595 	 */
596 	msg->msg_iov = (struct iovec *)vec;
597 	msg->msg_iovlen = num;
598 	result = sock_sendmsg(sock, msg, size);
599 	set_fs(oldfs);
600 	return result;
601 }
602 EXPORT_SYMBOL(kernel_sendmsg);
603 
604 static int ktime2ts(ktime_t kt, struct timespec *ts)
605 {
606 	if (kt.tv64) {
607 		*ts = ktime_to_timespec(kt);
608 		return 1;
609 	} else {
610 		return 0;
611 	}
612 }
613 
614 /*
615  * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
616  */
617 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
618 	struct sk_buff *skb)
619 {
620 	int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
621 	struct timespec ts[3];
622 	int empty = 1;
623 	struct skb_shared_hwtstamps *shhwtstamps =
624 		skb_hwtstamps(skb);
625 
626 	/* Race occurred between timestamp enabling and packet
627 	   receiving.  Fill in the current time for now. */
628 	if (need_software_tstamp && skb->tstamp.tv64 == 0)
629 		__net_timestamp(skb);
630 
631 	if (need_software_tstamp) {
632 		if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
633 			struct timeval tv;
634 			skb_get_timestamp(skb, &tv);
635 			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
636 				 sizeof(tv), &tv);
637 		} else {
638 			skb_get_timestampns(skb, &ts[0]);
639 			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
640 				 sizeof(ts[0]), &ts[0]);
641 		}
642 	}
643 
644 
645 	memset(ts, 0, sizeof(ts));
646 	if (skb->tstamp.tv64 &&
647 	    sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
648 		skb_get_timestampns(skb, ts + 0);
649 		empty = 0;
650 	}
651 	if (shhwtstamps) {
652 		if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
653 		    ktime2ts(shhwtstamps->syststamp, ts + 1))
654 			empty = 0;
655 		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
656 		    ktime2ts(shhwtstamps->hwtstamp, ts + 2))
657 			empty = 0;
658 	}
659 	if (!empty)
660 		put_cmsg(msg, SOL_SOCKET,
661 			 SCM_TIMESTAMPING, sizeof(ts), &ts);
662 }
663 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
664 
665 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
666 				   struct sk_buff *skb)
667 {
668 	if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
669 		put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
670 			sizeof(__u32), &skb->dropcount);
671 }
672 
673 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
674 	struct sk_buff *skb)
675 {
676 	sock_recv_timestamp(msg, sk, skb);
677 	sock_recv_drops(msg, sk, skb);
678 }
679 EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
680 
681 static inline int __sock_recvmsg_nosec(struct kiocb *iocb, struct socket *sock,
682 				       struct msghdr *msg, size_t size, int flags)
683 {
684 	struct sock_iocb *si = kiocb_to_siocb(iocb);
685 
686 	sock_update_classid(sock->sk);
687 
688 	si->sock = sock;
689 	si->scm = NULL;
690 	si->msg = msg;
691 	si->size = size;
692 	si->flags = flags;
693 
694 	return sock->ops->recvmsg(iocb, sock, msg, size, flags);
695 }
696 
697 static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
698 				 struct msghdr *msg, size_t size, int flags)
699 {
700 	int err = security_socket_recvmsg(sock, msg, size, flags);
701 
702 	return err ?: __sock_recvmsg_nosec(iocb, sock, msg, size, flags);
703 }
704 
705 int sock_recvmsg(struct socket *sock, struct msghdr *msg,
706 		 size_t size, int flags)
707 {
708 	struct kiocb iocb;
709 	struct sock_iocb siocb;
710 	int ret;
711 
712 	init_sync_kiocb(&iocb, NULL);
713 	iocb.private = &siocb;
714 	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
715 	if (-EIOCBQUEUED == ret)
716 		ret = wait_on_sync_kiocb(&iocb);
717 	return ret;
718 }
719 EXPORT_SYMBOL(sock_recvmsg);
720 
721 static int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
722 			      size_t size, int flags)
723 {
724 	struct kiocb iocb;
725 	struct sock_iocb siocb;
726 	int ret;
727 
728 	init_sync_kiocb(&iocb, NULL);
729 	iocb.private = &siocb;
730 	ret = __sock_recvmsg_nosec(&iocb, sock, msg, size, flags);
731 	if (-EIOCBQUEUED == ret)
732 		ret = wait_on_sync_kiocb(&iocb);
733 	return ret;
734 }
735 
736 /**
737  * kernel_recvmsg - Receive a message from a socket (kernel space)
738  * @sock:       The socket to receive the message from
739  * @msg:        Received message
740  * @vec:        Input s/g array for message data
741  * @num:        Size of input s/g array
742  * @size:       Number of bytes to read
743  * @flags:      Message flags (MSG_DONTWAIT, etc...)
744  *
745  * On return the msg structure contains the scatter/gather array passed in the
746  * vec argument. The array is modified so that it consists of the unfilled
747  * portion of the original array.
748  *
749  * The returned value is the total number of bytes received, or an error.
750  */
751 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
752 		   struct kvec *vec, size_t num, size_t size, int flags)
753 {
754 	mm_segment_t oldfs = get_fs();
755 	int result;
756 
757 	set_fs(KERNEL_DS);
758 	/*
759 	 * the following is safe, since for compiler definitions of kvec and
760 	 * iovec are identical, yielding the same in-core layout and alignment
761 	 */
762 	msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
763 	result = sock_recvmsg(sock, msg, size, flags);
764 	set_fs(oldfs);
765 	return result;
766 }
767 EXPORT_SYMBOL(kernel_recvmsg);
768 
769 static void sock_aio_dtor(struct kiocb *iocb)
770 {
771 	kfree(iocb->private);
772 }
773 
774 static ssize_t sock_sendpage(struct file *file, struct page *page,
775 			     int offset, size_t size, loff_t *ppos, int more)
776 {
777 	struct socket *sock;
778 	int flags;
779 
780 	sock = file->private_data;
781 
782 	flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
783 	if (more)
784 		flags |= MSG_MORE;
785 
786 	return kernel_sendpage(sock, page, offset, size, flags);
787 }
788 
789 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
790 				struct pipe_inode_info *pipe, size_t len,
791 				unsigned int flags)
792 {
793 	struct socket *sock = file->private_data;
794 
795 	if (unlikely(!sock->ops->splice_read))
796 		return -EINVAL;
797 
798 	sock_update_classid(sock->sk);
799 
800 	return sock->ops->splice_read(sock, ppos, pipe, len, flags);
801 }
802 
803 static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
804 					 struct sock_iocb *siocb)
805 {
806 	if (!is_sync_kiocb(iocb)) {
807 		siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
808 		if (!siocb)
809 			return NULL;
810 		iocb->ki_dtor = sock_aio_dtor;
811 	}
812 
813 	siocb->kiocb = iocb;
814 	iocb->private = siocb;
815 	return siocb;
816 }
817 
818 static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
819 		struct file *file, const struct iovec *iov,
820 		unsigned long nr_segs)
821 {
822 	struct socket *sock = file->private_data;
823 	size_t size = 0;
824 	int i;
825 
826 	for (i = 0; i < nr_segs; i++)
827 		size += iov[i].iov_len;
828 
829 	msg->msg_name = NULL;
830 	msg->msg_namelen = 0;
831 	msg->msg_control = NULL;
832 	msg->msg_controllen = 0;
833 	msg->msg_iov = (struct iovec *)iov;
834 	msg->msg_iovlen = nr_segs;
835 	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
836 
837 	return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
838 }
839 
840 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
841 				unsigned long nr_segs, loff_t pos)
842 {
843 	struct sock_iocb siocb, *x;
844 
845 	if (pos != 0)
846 		return -ESPIPE;
847 
848 	if (iocb->ki_left == 0)	/* Match SYS5 behaviour */
849 		return 0;
850 
851 
852 	x = alloc_sock_iocb(iocb, &siocb);
853 	if (!x)
854 		return -ENOMEM;
855 	return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
856 }
857 
858 static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
859 			struct file *file, const struct iovec *iov,
860 			unsigned long nr_segs)
861 {
862 	struct socket *sock = file->private_data;
863 	size_t size = 0;
864 	int i;
865 
866 	for (i = 0; i < nr_segs; i++)
867 		size += iov[i].iov_len;
868 
869 	msg->msg_name = NULL;
870 	msg->msg_namelen = 0;
871 	msg->msg_control = NULL;
872 	msg->msg_controllen = 0;
873 	msg->msg_iov = (struct iovec *)iov;
874 	msg->msg_iovlen = nr_segs;
875 	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
876 	if (sock->type == SOCK_SEQPACKET)
877 		msg->msg_flags |= MSG_EOR;
878 
879 	return __sock_sendmsg(iocb, sock, msg, size);
880 }
881 
882 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
883 			  unsigned long nr_segs, loff_t pos)
884 {
885 	struct sock_iocb siocb, *x;
886 
887 	if (pos != 0)
888 		return -ESPIPE;
889 
890 	x = alloc_sock_iocb(iocb, &siocb);
891 	if (!x)
892 		return -ENOMEM;
893 
894 	return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
895 }
896 
897 /*
898  * Atomic setting of ioctl hooks to avoid race
899  * with module unload.
900  */
901 
902 static DEFINE_MUTEX(br_ioctl_mutex);
903 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
904 
905 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
906 {
907 	mutex_lock(&br_ioctl_mutex);
908 	br_ioctl_hook = hook;
909 	mutex_unlock(&br_ioctl_mutex);
910 }
911 EXPORT_SYMBOL(brioctl_set);
912 
913 static DEFINE_MUTEX(vlan_ioctl_mutex);
914 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
915 
916 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
917 {
918 	mutex_lock(&vlan_ioctl_mutex);
919 	vlan_ioctl_hook = hook;
920 	mutex_unlock(&vlan_ioctl_mutex);
921 }
922 EXPORT_SYMBOL(vlan_ioctl_set);
923 
924 static DEFINE_MUTEX(dlci_ioctl_mutex);
925 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
926 
927 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
928 {
929 	mutex_lock(&dlci_ioctl_mutex);
930 	dlci_ioctl_hook = hook;
931 	mutex_unlock(&dlci_ioctl_mutex);
932 }
933 EXPORT_SYMBOL(dlci_ioctl_set);
934 
935 static long sock_do_ioctl(struct net *net, struct socket *sock,
936 				 unsigned int cmd, unsigned long arg)
937 {
938 	int err;
939 	void __user *argp = (void __user *)arg;
940 
941 	err = sock->ops->ioctl(sock, cmd, arg);
942 
943 	/*
944 	 * If this ioctl is unknown try to hand it down
945 	 * to the NIC driver.
946 	 */
947 	if (err == -ENOIOCTLCMD)
948 		err = dev_ioctl(net, cmd, argp);
949 
950 	return err;
951 }
952 
953 /*
954  *	With an ioctl, arg may well be a user mode pointer, but we don't know
955  *	what to do with it - that's up to the protocol still.
956  */
957 
958 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
959 {
960 	struct socket *sock;
961 	struct sock *sk;
962 	void __user *argp = (void __user *)arg;
963 	int pid, err;
964 	struct net *net;
965 
966 	sock = file->private_data;
967 	sk = sock->sk;
968 	net = sock_net(sk);
969 	if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
970 		err = dev_ioctl(net, cmd, argp);
971 	} else
972 #ifdef CONFIG_WEXT_CORE
973 	if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
974 		err = dev_ioctl(net, cmd, argp);
975 	} else
976 #endif
977 		switch (cmd) {
978 		case FIOSETOWN:
979 		case SIOCSPGRP:
980 			err = -EFAULT;
981 			if (get_user(pid, (int __user *)argp))
982 				break;
983 			err = f_setown(sock->file, pid, 1);
984 			break;
985 		case FIOGETOWN:
986 		case SIOCGPGRP:
987 			err = put_user(f_getown(sock->file),
988 				       (int __user *)argp);
989 			break;
990 		case SIOCGIFBR:
991 		case SIOCSIFBR:
992 		case SIOCBRADDBR:
993 		case SIOCBRDELBR:
994 			err = -ENOPKG;
995 			if (!br_ioctl_hook)
996 				request_module("bridge");
997 
998 			mutex_lock(&br_ioctl_mutex);
999 			if (br_ioctl_hook)
1000 				err = br_ioctl_hook(net, cmd, argp);
1001 			mutex_unlock(&br_ioctl_mutex);
1002 			break;
1003 		case SIOCGIFVLAN:
1004 		case SIOCSIFVLAN:
1005 			err = -ENOPKG;
1006 			if (!vlan_ioctl_hook)
1007 				request_module("8021q");
1008 
1009 			mutex_lock(&vlan_ioctl_mutex);
1010 			if (vlan_ioctl_hook)
1011 				err = vlan_ioctl_hook(net, argp);
1012 			mutex_unlock(&vlan_ioctl_mutex);
1013 			break;
1014 		case SIOCADDDLCI:
1015 		case SIOCDELDLCI:
1016 			err = -ENOPKG;
1017 			if (!dlci_ioctl_hook)
1018 				request_module("dlci");
1019 
1020 			mutex_lock(&dlci_ioctl_mutex);
1021 			if (dlci_ioctl_hook)
1022 				err = dlci_ioctl_hook(cmd, argp);
1023 			mutex_unlock(&dlci_ioctl_mutex);
1024 			break;
1025 		default:
1026 			err = sock_do_ioctl(net, sock, cmd, arg);
1027 			break;
1028 		}
1029 	return err;
1030 }
1031 
1032 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1033 {
1034 	int err;
1035 	struct socket *sock = NULL;
1036 
1037 	err = security_socket_create(family, type, protocol, 1);
1038 	if (err)
1039 		goto out;
1040 
1041 	sock = sock_alloc();
1042 	if (!sock) {
1043 		err = -ENOMEM;
1044 		goto out;
1045 	}
1046 
1047 	sock->type = type;
1048 	err = security_socket_post_create(sock, family, type, protocol, 1);
1049 	if (err)
1050 		goto out_release;
1051 
1052 out:
1053 	*res = sock;
1054 	return err;
1055 out_release:
1056 	sock_release(sock);
1057 	sock = NULL;
1058 	goto out;
1059 }
1060 EXPORT_SYMBOL(sock_create_lite);
1061 
1062 /* No kernel lock held - perfect */
1063 static unsigned int sock_poll(struct file *file, poll_table *wait)
1064 {
1065 	struct socket *sock;
1066 
1067 	/*
1068 	 *      We can't return errors to poll, so it's either yes or no.
1069 	 */
1070 	sock = file->private_data;
1071 	return sock->ops->poll(file, sock, wait);
1072 }
1073 
1074 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1075 {
1076 	struct socket *sock = file->private_data;
1077 
1078 	return sock->ops->mmap(file, sock, vma);
1079 }
1080 
1081 static int sock_close(struct inode *inode, struct file *filp)
1082 {
1083 	/*
1084 	 *      It was possible the inode is NULL we were
1085 	 *      closing an unfinished socket.
1086 	 */
1087 
1088 	if (!inode) {
1089 		printk(KERN_DEBUG "sock_close: NULL inode\n");
1090 		return 0;
1091 	}
1092 	sock_release(SOCKET_I(inode));
1093 	return 0;
1094 }
1095 
1096 /*
1097  *	Update the socket async list
1098  *
1099  *	Fasync_list locking strategy.
1100  *
1101  *	1. fasync_list is modified only under process context socket lock
1102  *	   i.e. under semaphore.
1103  *	2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1104  *	   or under socket lock
1105  */
1106 
1107 static int sock_fasync(int fd, struct file *filp, int on)
1108 {
1109 	struct socket *sock = filp->private_data;
1110 	struct sock *sk = sock->sk;
1111 
1112 	if (sk == NULL)
1113 		return -EINVAL;
1114 
1115 	lock_sock(sk);
1116 
1117 	fasync_helper(fd, filp, on, &sock->wq->fasync_list);
1118 
1119 	if (!sock->wq->fasync_list)
1120 		sock_reset_flag(sk, SOCK_FASYNC);
1121 	else
1122 		sock_set_flag(sk, SOCK_FASYNC);
1123 
1124 	release_sock(sk);
1125 	return 0;
1126 }
1127 
1128 /* This function may be called only under socket lock or callback_lock or rcu_lock */
1129 
1130 int sock_wake_async(struct socket *sock, int how, int band)
1131 {
1132 	struct socket_wq *wq;
1133 
1134 	if (!sock)
1135 		return -1;
1136 	rcu_read_lock();
1137 	wq = rcu_dereference(sock->wq);
1138 	if (!wq || !wq->fasync_list) {
1139 		rcu_read_unlock();
1140 		return -1;
1141 	}
1142 	switch (how) {
1143 	case SOCK_WAKE_WAITD:
1144 		if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1145 			break;
1146 		goto call_kill;
1147 	case SOCK_WAKE_SPACE:
1148 		if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1149 			break;
1150 		/* fall through */
1151 	case SOCK_WAKE_IO:
1152 call_kill:
1153 		kill_fasync(&wq->fasync_list, SIGIO, band);
1154 		break;
1155 	case SOCK_WAKE_URG:
1156 		kill_fasync(&wq->fasync_list, SIGURG, band);
1157 	}
1158 	rcu_read_unlock();
1159 	return 0;
1160 }
1161 EXPORT_SYMBOL(sock_wake_async);
1162 
1163 int __sock_create(struct net *net, int family, int type, int protocol,
1164 			 struct socket **res, int kern)
1165 {
1166 	int err;
1167 	struct socket *sock;
1168 	const struct net_proto_family *pf;
1169 
1170 	/*
1171 	 *      Check protocol is in range
1172 	 */
1173 	if (family < 0 || family >= NPROTO)
1174 		return -EAFNOSUPPORT;
1175 	if (type < 0 || type >= SOCK_MAX)
1176 		return -EINVAL;
1177 
1178 	/* Compatibility.
1179 
1180 	   This uglymoron is moved from INET layer to here to avoid
1181 	   deadlock in module load.
1182 	 */
1183 	if (family == PF_INET && type == SOCK_PACKET) {
1184 		static int warned;
1185 		if (!warned) {
1186 			warned = 1;
1187 			printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1188 			       current->comm);
1189 		}
1190 		family = PF_PACKET;
1191 	}
1192 
1193 	err = security_socket_create(family, type, protocol, kern);
1194 	if (err)
1195 		return err;
1196 
1197 	/*
1198 	 *	Allocate the socket and allow the family to set things up. if
1199 	 *	the protocol is 0, the family is instructed to select an appropriate
1200 	 *	default.
1201 	 */
1202 	sock = sock_alloc();
1203 	if (!sock) {
1204 		if (net_ratelimit())
1205 			printk(KERN_WARNING "socket: no more sockets\n");
1206 		return -ENFILE;	/* Not exactly a match, but its the
1207 				   closest posix thing */
1208 	}
1209 
1210 	sock->type = type;
1211 
1212 #ifdef CONFIG_MODULES
1213 	/* Attempt to load a protocol module if the find failed.
1214 	 *
1215 	 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1216 	 * requested real, full-featured networking support upon configuration.
1217 	 * Otherwise module support will break!
1218 	 */
1219 	if (rcu_access_pointer(net_families[family]) == NULL)
1220 		request_module("net-pf-%d", family);
1221 #endif
1222 
1223 	rcu_read_lock();
1224 	pf = rcu_dereference(net_families[family]);
1225 	err = -EAFNOSUPPORT;
1226 	if (!pf)
1227 		goto out_release;
1228 
1229 	/*
1230 	 * We will call the ->create function, that possibly is in a loadable
1231 	 * module, so we have to bump that loadable module refcnt first.
1232 	 */
1233 	if (!try_module_get(pf->owner))
1234 		goto out_release;
1235 
1236 	/* Now protected by module ref count */
1237 	rcu_read_unlock();
1238 
1239 	err = pf->create(net, sock, protocol, kern);
1240 	if (err < 0)
1241 		goto out_module_put;
1242 
1243 	/*
1244 	 * Now to bump the refcnt of the [loadable] module that owns this
1245 	 * socket at sock_release time we decrement its refcnt.
1246 	 */
1247 	if (!try_module_get(sock->ops->owner))
1248 		goto out_module_busy;
1249 
1250 	/*
1251 	 * Now that we're done with the ->create function, the [loadable]
1252 	 * module can have its refcnt decremented
1253 	 */
1254 	module_put(pf->owner);
1255 	err = security_socket_post_create(sock, family, type, protocol, kern);
1256 	if (err)
1257 		goto out_sock_release;
1258 	*res = sock;
1259 
1260 	return 0;
1261 
1262 out_module_busy:
1263 	err = -EAFNOSUPPORT;
1264 out_module_put:
1265 	sock->ops = NULL;
1266 	module_put(pf->owner);
1267 out_sock_release:
1268 	sock_release(sock);
1269 	return err;
1270 
1271 out_release:
1272 	rcu_read_unlock();
1273 	goto out_sock_release;
1274 }
1275 EXPORT_SYMBOL(__sock_create);
1276 
1277 int sock_create(int family, int type, int protocol, struct socket **res)
1278 {
1279 	return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1280 }
1281 EXPORT_SYMBOL(sock_create);
1282 
1283 int sock_create_kern(int family, int type, int protocol, struct socket **res)
1284 {
1285 	return __sock_create(&init_net, family, type, protocol, res, 1);
1286 }
1287 EXPORT_SYMBOL(sock_create_kern);
1288 
1289 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1290 {
1291 	int retval;
1292 	struct socket *sock;
1293 	int flags;
1294 
1295 	/* Check the SOCK_* constants for consistency.  */
1296 	BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1297 	BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1298 	BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1299 	BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1300 
1301 	flags = type & ~SOCK_TYPE_MASK;
1302 	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1303 		return -EINVAL;
1304 	type &= SOCK_TYPE_MASK;
1305 
1306 	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1307 		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1308 
1309 	retval = sock_create(family, type, protocol, &sock);
1310 	if (retval < 0)
1311 		goto out;
1312 
1313 	retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1314 	if (retval < 0)
1315 		goto out_release;
1316 
1317 out:
1318 	/* It may be already another descriptor 8) Not kernel problem. */
1319 	return retval;
1320 
1321 out_release:
1322 	sock_release(sock);
1323 	return retval;
1324 }
1325 
1326 /*
1327  *	Create a pair of connected sockets.
1328  */
1329 
1330 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1331 		int __user *, usockvec)
1332 {
1333 	struct socket *sock1, *sock2;
1334 	int fd1, fd2, err;
1335 	struct file *newfile1, *newfile2;
1336 	int flags;
1337 
1338 	flags = type & ~SOCK_TYPE_MASK;
1339 	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1340 		return -EINVAL;
1341 	type &= SOCK_TYPE_MASK;
1342 
1343 	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1344 		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1345 
1346 	/*
1347 	 * Obtain the first socket and check if the underlying protocol
1348 	 * supports the socketpair call.
1349 	 */
1350 
1351 	err = sock_create(family, type, protocol, &sock1);
1352 	if (err < 0)
1353 		goto out;
1354 
1355 	err = sock_create(family, type, protocol, &sock2);
1356 	if (err < 0)
1357 		goto out_release_1;
1358 
1359 	err = sock1->ops->socketpair(sock1, sock2);
1360 	if (err < 0)
1361 		goto out_release_both;
1362 
1363 	fd1 = sock_alloc_file(sock1, &newfile1, flags);
1364 	if (unlikely(fd1 < 0)) {
1365 		err = fd1;
1366 		goto out_release_both;
1367 	}
1368 
1369 	fd2 = sock_alloc_file(sock2, &newfile2, flags);
1370 	if (unlikely(fd2 < 0)) {
1371 		err = fd2;
1372 		fput(newfile1);
1373 		put_unused_fd(fd1);
1374 		sock_release(sock2);
1375 		goto out;
1376 	}
1377 
1378 	audit_fd_pair(fd1, fd2);
1379 	fd_install(fd1, newfile1);
1380 	fd_install(fd2, newfile2);
1381 	/* fd1 and fd2 may be already another descriptors.
1382 	 * Not kernel problem.
1383 	 */
1384 
1385 	err = put_user(fd1, &usockvec[0]);
1386 	if (!err)
1387 		err = put_user(fd2, &usockvec[1]);
1388 	if (!err)
1389 		return 0;
1390 
1391 	sys_close(fd2);
1392 	sys_close(fd1);
1393 	return err;
1394 
1395 out_release_both:
1396 	sock_release(sock2);
1397 out_release_1:
1398 	sock_release(sock1);
1399 out:
1400 	return err;
1401 }
1402 
1403 /*
1404  *	Bind a name to a socket. Nothing much to do here since it's
1405  *	the protocol's responsibility to handle the local address.
1406  *
1407  *	We move the socket address to kernel space before we call
1408  *	the protocol layer (having also checked the address is ok).
1409  */
1410 
1411 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1412 {
1413 	struct socket *sock;
1414 	struct sockaddr_storage address;
1415 	int err, fput_needed;
1416 
1417 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1418 	if (sock) {
1419 		err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1420 		if (err >= 0) {
1421 			err = security_socket_bind(sock,
1422 						   (struct sockaddr *)&address,
1423 						   addrlen);
1424 			if (!err)
1425 				err = sock->ops->bind(sock,
1426 						      (struct sockaddr *)
1427 						      &address, addrlen);
1428 		}
1429 		fput_light(sock->file, fput_needed);
1430 	}
1431 	return err;
1432 }
1433 
1434 /*
1435  *	Perform a listen. Basically, we allow the protocol to do anything
1436  *	necessary for a listen, and if that works, we mark the socket as
1437  *	ready for listening.
1438  */
1439 
1440 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1441 {
1442 	struct socket *sock;
1443 	int err, fput_needed;
1444 	int somaxconn;
1445 
1446 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1447 	if (sock) {
1448 		somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1449 		if ((unsigned)backlog > somaxconn)
1450 			backlog = somaxconn;
1451 
1452 		err = security_socket_listen(sock, backlog);
1453 		if (!err)
1454 			err = sock->ops->listen(sock, backlog);
1455 
1456 		fput_light(sock->file, fput_needed);
1457 	}
1458 	return err;
1459 }
1460 
1461 /*
1462  *	For accept, we attempt to create a new socket, set up the link
1463  *	with the client, wake up the client, then return the new
1464  *	connected fd. We collect the address of the connector in kernel
1465  *	space and move it to user at the very end. This is unclean because
1466  *	we open the socket then return an error.
1467  *
1468  *	1003.1g adds the ability to recvmsg() to query connection pending
1469  *	status to recvmsg. We need to add that support in a way thats
1470  *	clean when we restucture accept also.
1471  */
1472 
1473 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1474 		int __user *, upeer_addrlen, int, flags)
1475 {
1476 	struct socket *sock, *newsock;
1477 	struct file *newfile;
1478 	int err, len, newfd, fput_needed;
1479 	struct sockaddr_storage address;
1480 
1481 	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1482 		return -EINVAL;
1483 
1484 	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1485 		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1486 
1487 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1488 	if (!sock)
1489 		goto out;
1490 
1491 	err = -ENFILE;
1492 	newsock = sock_alloc();
1493 	if (!newsock)
1494 		goto out_put;
1495 
1496 	newsock->type = sock->type;
1497 	newsock->ops = sock->ops;
1498 
1499 	/*
1500 	 * We don't need try_module_get here, as the listening socket (sock)
1501 	 * has the protocol module (sock->ops->owner) held.
1502 	 */
1503 	__module_get(newsock->ops->owner);
1504 
1505 	newfd = sock_alloc_file(newsock, &newfile, flags);
1506 	if (unlikely(newfd < 0)) {
1507 		err = newfd;
1508 		sock_release(newsock);
1509 		goto out_put;
1510 	}
1511 
1512 	err = security_socket_accept(sock, newsock);
1513 	if (err)
1514 		goto out_fd;
1515 
1516 	err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1517 	if (err < 0)
1518 		goto out_fd;
1519 
1520 	if (upeer_sockaddr) {
1521 		if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1522 					  &len, 2) < 0) {
1523 			err = -ECONNABORTED;
1524 			goto out_fd;
1525 		}
1526 		err = move_addr_to_user((struct sockaddr *)&address,
1527 					len, upeer_sockaddr, upeer_addrlen);
1528 		if (err < 0)
1529 			goto out_fd;
1530 	}
1531 
1532 	/* File flags are not inherited via accept() unlike another OSes. */
1533 
1534 	fd_install(newfd, newfile);
1535 	err = newfd;
1536 
1537 out_put:
1538 	fput_light(sock->file, fput_needed);
1539 out:
1540 	return err;
1541 out_fd:
1542 	fput(newfile);
1543 	put_unused_fd(newfd);
1544 	goto out_put;
1545 }
1546 
1547 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1548 		int __user *, upeer_addrlen)
1549 {
1550 	return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1551 }
1552 
1553 /*
1554  *	Attempt to connect to a socket with the server address.  The address
1555  *	is in user space so we verify it is OK and move it to kernel space.
1556  *
1557  *	For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1558  *	break bindings
1559  *
1560  *	NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1561  *	other SEQPACKET protocols that take time to connect() as it doesn't
1562  *	include the -EINPROGRESS status for such sockets.
1563  */
1564 
1565 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1566 		int, addrlen)
1567 {
1568 	struct socket *sock;
1569 	struct sockaddr_storage address;
1570 	int err, fput_needed;
1571 
1572 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1573 	if (!sock)
1574 		goto out;
1575 	err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1576 	if (err < 0)
1577 		goto out_put;
1578 
1579 	err =
1580 	    security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1581 	if (err)
1582 		goto out_put;
1583 
1584 	err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1585 				 sock->file->f_flags);
1586 out_put:
1587 	fput_light(sock->file, fput_needed);
1588 out:
1589 	return err;
1590 }
1591 
1592 /*
1593  *	Get the local address ('name') of a socket object. Move the obtained
1594  *	name to user space.
1595  */
1596 
1597 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1598 		int __user *, usockaddr_len)
1599 {
1600 	struct socket *sock;
1601 	struct sockaddr_storage address;
1602 	int len, err, fput_needed;
1603 
1604 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1605 	if (!sock)
1606 		goto out;
1607 
1608 	err = security_socket_getsockname(sock);
1609 	if (err)
1610 		goto out_put;
1611 
1612 	err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1613 	if (err)
1614 		goto out_put;
1615 	err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1616 
1617 out_put:
1618 	fput_light(sock->file, fput_needed);
1619 out:
1620 	return err;
1621 }
1622 
1623 /*
1624  *	Get the remote address ('name') of a socket object. Move the obtained
1625  *	name to user space.
1626  */
1627 
1628 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1629 		int __user *, usockaddr_len)
1630 {
1631 	struct socket *sock;
1632 	struct sockaddr_storage address;
1633 	int len, err, fput_needed;
1634 
1635 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1636 	if (sock != NULL) {
1637 		err = security_socket_getpeername(sock);
1638 		if (err) {
1639 			fput_light(sock->file, fput_needed);
1640 			return err;
1641 		}
1642 
1643 		err =
1644 		    sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1645 				       1);
1646 		if (!err)
1647 			err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1648 						usockaddr_len);
1649 		fput_light(sock->file, fput_needed);
1650 	}
1651 	return err;
1652 }
1653 
1654 /*
1655  *	Send a datagram to a given address. We move the address into kernel
1656  *	space and check the user space data area is readable before invoking
1657  *	the protocol.
1658  */
1659 
1660 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1661 		unsigned, flags, struct sockaddr __user *, addr,
1662 		int, addr_len)
1663 {
1664 	struct socket *sock;
1665 	struct sockaddr_storage address;
1666 	int err;
1667 	struct msghdr msg;
1668 	struct iovec iov;
1669 	int fput_needed;
1670 
1671 	if (len > INT_MAX)
1672 		len = INT_MAX;
1673 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1674 	if (!sock)
1675 		goto out;
1676 
1677 	iov.iov_base = buff;
1678 	iov.iov_len = len;
1679 	msg.msg_name = NULL;
1680 	msg.msg_iov = &iov;
1681 	msg.msg_iovlen = 1;
1682 	msg.msg_control = NULL;
1683 	msg.msg_controllen = 0;
1684 	msg.msg_namelen = 0;
1685 	if (addr) {
1686 		err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1687 		if (err < 0)
1688 			goto out_put;
1689 		msg.msg_name = (struct sockaddr *)&address;
1690 		msg.msg_namelen = addr_len;
1691 	}
1692 	if (sock->file->f_flags & O_NONBLOCK)
1693 		flags |= MSG_DONTWAIT;
1694 	msg.msg_flags = flags;
1695 	err = sock_sendmsg(sock, &msg, len);
1696 
1697 out_put:
1698 	fput_light(sock->file, fput_needed);
1699 out:
1700 	return err;
1701 }
1702 
1703 /*
1704  *	Send a datagram down a socket.
1705  */
1706 
1707 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1708 		unsigned, flags)
1709 {
1710 	return sys_sendto(fd, buff, len, flags, NULL, 0);
1711 }
1712 
1713 /*
1714  *	Receive a frame from the socket and optionally record the address of the
1715  *	sender. We verify the buffers are writable and if needed move the
1716  *	sender address from kernel to user space.
1717  */
1718 
1719 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1720 		unsigned, flags, struct sockaddr __user *, addr,
1721 		int __user *, addr_len)
1722 {
1723 	struct socket *sock;
1724 	struct iovec iov;
1725 	struct msghdr msg;
1726 	struct sockaddr_storage address;
1727 	int err, err2;
1728 	int fput_needed;
1729 
1730 	if (size > INT_MAX)
1731 		size = INT_MAX;
1732 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1733 	if (!sock)
1734 		goto out;
1735 
1736 	msg.msg_control = NULL;
1737 	msg.msg_controllen = 0;
1738 	msg.msg_iovlen = 1;
1739 	msg.msg_iov = &iov;
1740 	iov.iov_len = size;
1741 	iov.iov_base = ubuf;
1742 	msg.msg_name = (struct sockaddr *)&address;
1743 	msg.msg_namelen = sizeof(address);
1744 	if (sock->file->f_flags & O_NONBLOCK)
1745 		flags |= MSG_DONTWAIT;
1746 	err = sock_recvmsg(sock, &msg, size, flags);
1747 
1748 	if (err >= 0 && addr != NULL) {
1749 		err2 = move_addr_to_user((struct sockaddr *)&address,
1750 					 msg.msg_namelen, addr, addr_len);
1751 		if (err2 < 0)
1752 			err = err2;
1753 	}
1754 
1755 	fput_light(sock->file, fput_needed);
1756 out:
1757 	return err;
1758 }
1759 
1760 /*
1761  *	Receive a datagram from a socket.
1762  */
1763 
1764 asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1765 			 unsigned flags)
1766 {
1767 	return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1768 }
1769 
1770 /*
1771  *	Set a socket option. Because we don't know the option lengths we have
1772  *	to pass the user mode parameter for the protocols to sort out.
1773  */
1774 
1775 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1776 		char __user *, optval, int, optlen)
1777 {
1778 	int err, fput_needed;
1779 	struct socket *sock;
1780 
1781 	if (optlen < 0)
1782 		return -EINVAL;
1783 
1784 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1785 	if (sock != NULL) {
1786 		err = security_socket_setsockopt(sock, level, optname);
1787 		if (err)
1788 			goto out_put;
1789 
1790 		if (level == SOL_SOCKET)
1791 			err =
1792 			    sock_setsockopt(sock, level, optname, optval,
1793 					    optlen);
1794 		else
1795 			err =
1796 			    sock->ops->setsockopt(sock, level, optname, optval,
1797 						  optlen);
1798 out_put:
1799 		fput_light(sock->file, fput_needed);
1800 	}
1801 	return err;
1802 }
1803 
1804 /*
1805  *	Get a socket option. Because we don't know the option lengths we have
1806  *	to pass a user mode parameter for the protocols to sort out.
1807  */
1808 
1809 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1810 		char __user *, optval, int __user *, optlen)
1811 {
1812 	int err, fput_needed;
1813 	struct socket *sock;
1814 
1815 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1816 	if (sock != NULL) {
1817 		err = security_socket_getsockopt(sock, level, optname);
1818 		if (err)
1819 			goto out_put;
1820 
1821 		if (level == SOL_SOCKET)
1822 			err =
1823 			    sock_getsockopt(sock, level, optname, optval,
1824 					    optlen);
1825 		else
1826 			err =
1827 			    sock->ops->getsockopt(sock, level, optname, optval,
1828 						  optlen);
1829 out_put:
1830 		fput_light(sock->file, fput_needed);
1831 	}
1832 	return err;
1833 }
1834 
1835 /*
1836  *	Shutdown a socket.
1837  */
1838 
1839 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1840 {
1841 	int err, fput_needed;
1842 	struct socket *sock;
1843 
1844 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1845 	if (sock != NULL) {
1846 		err = security_socket_shutdown(sock, how);
1847 		if (!err)
1848 			err = sock->ops->shutdown(sock, how);
1849 		fput_light(sock->file, fput_needed);
1850 	}
1851 	return err;
1852 }
1853 
1854 /* A couple of helpful macros for getting the address of the 32/64 bit
1855  * fields which are the same type (int / unsigned) on our platforms.
1856  */
1857 #define COMPAT_MSG(msg, member)	((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1858 #define COMPAT_NAMELEN(msg)	COMPAT_MSG(msg, msg_namelen)
1859 #define COMPAT_FLAGS(msg)	COMPAT_MSG(msg, msg_flags)
1860 
1861 /*
1862  *	BSD sendmsg interface
1863  */
1864 
1865 SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1866 {
1867 	struct compat_msghdr __user *msg_compat =
1868 	    (struct compat_msghdr __user *)msg;
1869 	struct socket *sock;
1870 	struct sockaddr_storage address;
1871 	struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1872 	unsigned char ctl[sizeof(struct cmsghdr) + 20]
1873 	    __attribute__ ((aligned(sizeof(__kernel_size_t))));
1874 	/* 20 is size of ipv6_pktinfo */
1875 	unsigned char *ctl_buf = ctl;
1876 	struct msghdr msg_sys;
1877 	int err, ctl_len, iov_size, total_len;
1878 	int fput_needed;
1879 
1880 	err = -EFAULT;
1881 	if (MSG_CMSG_COMPAT & flags) {
1882 		if (get_compat_msghdr(&msg_sys, msg_compat))
1883 			return -EFAULT;
1884 	} else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1885 		return -EFAULT;
1886 
1887 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1888 	if (!sock)
1889 		goto out;
1890 
1891 	/* do not move before msg_sys is valid */
1892 	err = -EMSGSIZE;
1893 	if (msg_sys.msg_iovlen > UIO_MAXIOV)
1894 		goto out_put;
1895 
1896 	/* Check whether to allocate the iovec area */
1897 	err = -ENOMEM;
1898 	iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1899 	if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1900 		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1901 		if (!iov)
1902 			goto out_put;
1903 	}
1904 
1905 	/* This will also move the address data into kernel space */
1906 	if (MSG_CMSG_COMPAT & flags) {
1907 		err = verify_compat_iovec(&msg_sys, iov,
1908 					  (struct sockaddr *)&address,
1909 					  VERIFY_READ);
1910 	} else
1911 		err = verify_iovec(&msg_sys, iov,
1912 				   (struct sockaddr *)&address,
1913 				   VERIFY_READ);
1914 	if (err < 0)
1915 		goto out_freeiov;
1916 	total_len = err;
1917 
1918 	err = -ENOBUFS;
1919 
1920 	if (msg_sys.msg_controllen > INT_MAX)
1921 		goto out_freeiov;
1922 	ctl_len = msg_sys.msg_controllen;
1923 	if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1924 		err =
1925 		    cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1926 						     sizeof(ctl));
1927 		if (err)
1928 			goto out_freeiov;
1929 		ctl_buf = msg_sys.msg_control;
1930 		ctl_len = msg_sys.msg_controllen;
1931 	} else if (ctl_len) {
1932 		if (ctl_len > sizeof(ctl)) {
1933 			ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1934 			if (ctl_buf == NULL)
1935 				goto out_freeiov;
1936 		}
1937 		err = -EFAULT;
1938 		/*
1939 		 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1940 		 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1941 		 * checking falls down on this.
1942 		 */
1943 		if (copy_from_user(ctl_buf,
1944 				   (void __user __force *)msg_sys.msg_control,
1945 				   ctl_len))
1946 			goto out_freectl;
1947 		msg_sys.msg_control = ctl_buf;
1948 	}
1949 	msg_sys.msg_flags = flags;
1950 
1951 	if (sock->file->f_flags & O_NONBLOCK)
1952 		msg_sys.msg_flags |= MSG_DONTWAIT;
1953 	err = sock_sendmsg(sock, &msg_sys, total_len);
1954 
1955 out_freectl:
1956 	if (ctl_buf != ctl)
1957 		sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1958 out_freeiov:
1959 	if (iov != iovstack)
1960 		sock_kfree_s(sock->sk, iov, iov_size);
1961 out_put:
1962 	fput_light(sock->file, fput_needed);
1963 out:
1964 	return err;
1965 }
1966 
1967 static int __sys_recvmsg(struct socket *sock, struct msghdr __user *msg,
1968 			 struct msghdr *msg_sys, unsigned flags, int nosec)
1969 {
1970 	struct compat_msghdr __user *msg_compat =
1971 	    (struct compat_msghdr __user *)msg;
1972 	struct iovec iovstack[UIO_FASTIOV];
1973 	struct iovec *iov = iovstack;
1974 	unsigned long cmsg_ptr;
1975 	int err, iov_size, total_len, len;
1976 
1977 	/* kernel mode address */
1978 	struct sockaddr_storage addr;
1979 
1980 	/* user mode address pointers */
1981 	struct sockaddr __user *uaddr;
1982 	int __user *uaddr_len;
1983 
1984 	if (MSG_CMSG_COMPAT & flags) {
1985 		if (get_compat_msghdr(msg_sys, msg_compat))
1986 			return -EFAULT;
1987 	} else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
1988 		return -EFAULT;
1989 
1990 	err = -EMSGSIZE;
1991 	if (msg_sys->msg_iovlen > UIO_MAXIOV)
1992 		goto out;
1993 
1994 	/* Check whether to allocate the iovec area */
1995 	err = -ENOMEM;
1996 	iov_size = msg_sys->msg_iovlen * sizeof(struct iovec);
1997 	if (msg_sys->msg_iovlen > UIO_FASTIOV) {
1998 		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1999 		if (!iov)
2000 			goto out;
2001 	}
2002 
2003 	/*
2004 	 *      Save the user-mode address (verify_iovec will change the
2005 	 *      kernel msghdr to use the kernel address space)
2006 	 */
2007 
2008 	uaddr = (__force void __user *)msg_sys->msg_name;
2009 	uaddr_len = COMPAT_NAMELEN(msg);
2010 	if (MSG_CMSG_COMPAT & flags) {
2011 		err = verify_compat_iovec(msg_sys, iov,
2012 					  (struct sockaddr *)&addr,
2013 					  VERIFY_WRITE);
2014 	} else
2015 		err = verify_iovec(msg_sys, iov,
2016 				   (struct sockaddr *)&addr,
2017 				   VERIFY_WRITE);
2018 	if (err < 0)
2019 		goto out_freeiov;
2020 	total_len = err;
2021 
2022 	cmsg_ptr = (unsigned long)msg_sys->msg_control;
2023 	msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2024 
2025 	if (sock->file->f_flags & O_NONBLOCK)
2026 		flags |= MSG_DONTWAIT;
2027 	err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys,
2028 							  total_len, flags);
2029 	if (err < 0)
2030 		goto out_freeiov;
2031 	len = err;
2032 
2033 	if (uaddr != NULL) {
2034 		err = move_addr_to_user((struct sockaddr *)&addr,
2035 					msg_sys->msg_namelen, uaddr,
2036 					uaddr_len);
2037 		if (err < 0)
2038 			goto out_freeiov;
2039 	}
2040 	err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2041 			 COMPAT_FLAGS(msg));
2042 	if (err)
2043 		goto out_freeiov;
2044 	if (MSG_CMSG_COMPAT & flags)
2045 		err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2046 				 &msg_compat->msg_controllen);
2047 	else
2048 		err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2049 				 &msg->msg_controllen);
2050 	if (err)
2051 		goto out_freeiov;
2052 	err = len;
2053 
2054 out_freeiov:
2055 	if (iov != iovstack)
2056 		sock_kfree_s(sock->sk, iov, iov_size);
2057 out:
2058 	return err;
2059 }
2060 
2061 /*
2062  *	BSD recvmsg interface
2063  */
2064 
2065 SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
2066 		unsigned int, flags)
2067 {
2068 	int fput_needed, err;
2069 	struct msghdr msg_sys;
2070 	struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2071 
2072 	if (!sock)
2073 		goto out;
2074 
2075 	err = __sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2076 
2077 	fput_light(sock->file, fput_needed);
2078 out:
2079 	return err;
2080 }
2081 
2082 /*
2083  *     Linux recvmmsg interface
2084  */
2085 
2086 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2087 		   unsigned int flags, struct timespec *timeout)
2088 {
2089 	int fput_needed, err, datagrams;
2090 	struct socket *sock;
2091 	struct mmsghdr __user *entry;
2092 	struct compat_mmsghdr __user *compat_entry;
2093 	struct msghdr msg_sys;
2094 	struct timespec end_time;
2095 
2096 	if (timeout &&
2097 	    poll_select_set_timeout(&end_time, timeout->tv_sec,
2098 				    timeout->tv_nsec))
2099 		return -EINVAL;
2100 
2101 	datagrams = 0;
2102 
2103 	sock = sockfd_lookup_light(fd, &err, &fput_needed);
2104 	if (!sock)
2105 		return err;
2106 
2107 	err = sock_error(sock->sk);
2108 	if (err)
2109 		goto out_put;
2110 
2111 	entry = mmsg;
2112 	compat_entry = (struct compat_mmsghdr __user *)mmsg;
2113 
2114 	while (datagrams < vlen) {
2115 		/*
2116 		 * No need to ask LSM for more than the first datagram.
2117 		 */
2118 		if (MSG_CMSG_COMPAT & flags) {
2119 			err = __sys_recvmsg(sock, (struct msghdr __user *)compat_entry,
2120 					    &msg_sys, flags, datagrams);
2121 			if (err < 0)
2122 				break;
2123 			err = __put_user(err, &compat_entry->msg_len);
2124 			++compat_entry;
2125 		} else {
2126 			err = __sys_recvmsg(sock, (struct msghdr __user *)entry,
2127 					    &msg_sys, flags, datagrams);
2128 			if (err < 0)
2129 				break;
2130 			err = put_user(err, &entry->msg_len);
2131 			++entry;
2132 		}
2133 
2134 		if (err)
2135 			break;
2136 		++datagrams;
2137 
2138 		/* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2139 		if (flags & MSG_WAITFORONE)
2140 			flags |= MSG_DONTWAIT;
2141 
2142 		if (timeout) {
2143 			ktime_get_ts(timeout);
2144 			*timeout = timespec_sub(end_time, *timeout);
2145 			if (timeout->tv_sec < 0) {
2146 				timeout->tv_sec = timeout->tv_nsec = 0;
2147 				break;
2148 			}
2149 
2150 			/* Timeout, return less than vlen datagrams */
2151 			if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2152 				break;
2153 		}
2154 
2155 		/* Out of band data, return right away */
2156 		if (msg_sys.msg_flags & MSG_OOB)
2157 			break;
2158 	}
2159 
2160 out_put:
2161 	fput_light(sock->file, fput_needed);
2162 
2163 	if (err == 0)
2164 		return datagrams;
2165 
2166 	if (datagrams != 0) {
2167 		/*
2168 		 * We may return less entries than requested (vlen) if the
2169 		 * sock is non block and there aren't enough datagrams...
2170 		 */
2171 		if (err != -EAGAIN) {
2172 			/*
2173 			 * ... or  if recvmsg returns an error after we
2174 			 * received some datagrams, where we record the
2175 			 * error to return on the next call or if the
2176 			 * app asks about it using getsockopt(SO_ERROR).
2177 			 */
2178 			sock->sk->sk_err = -err;
2179 		}
2180 
2181 		return datagrams;
2182 	}
2183 
2184 	return err;
2185 }
2186 
2187 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2188 		unsigned int, vlen, unsigned int, flags,
2189 		struct timespec __user *, timeout)
2190 {
2191 	int datagrams;
2192 	struct timespec timeout_sys;
2193 
2194 	if (!timeout)
2195 		return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2196 
2197 	if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2198 		return -EFAULT;
2199 
2200 	datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2201 
2202 	if (datagrams > 0 &&
2203 	    copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2204 		datagrams = -EFAULT;
2205 
2206 	return datagrams;
2207 }
2208 
2209 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2210 /* Argument list sizes for sys_socketcall */
2211 #define AL(x) ((x) * sizeof(unsigned long))
2212 static const unsigned char nargs[20] = {
2213 	AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2214 	AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2215 	AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2216 	AL(4), AL(5)
2217 };
2218 
2219 #undef AL
2220 
2221 /*
2222  *	System call vectors.
2223  *
2224  *	Argument checking cleaned up. Saved 20% in size.
2225  *  This function doesn't need to set the kernel lock because
2226  *  it is set by the callees.
2227  */
2228 
2229 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2230 {
2231 	unsigned long a[6];
2232 	unsigned long a0, a1;
2233 	int err;
2234 	unsigned int len;
2235 
2236 	if (call < 1 || call > SYS_RECVMMSG)
2237 		return -EINVAL;
2238 
2239 	len = nargs[call];
2240 	if (len > sizeof(a))
2241 		return -EINVAL;
2242 
2243 	/* copy_from_user should be SMP safe. */
2244 	if (copy_from_user(a, args, len))
2245 		return -EFAULT;
2246 
2247 	audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2248 
2249 	a0 = a[0];
2250 	a1 = a[1];
2251 
2252 	switch (call) {
2253 	case SYS_SOCKET:
2254 		err = sys_socket(a0, a1, a[2]);
2255 		break;
2256 	case SYS_BIND:
2257 		err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2258 		break;
2259 	case SYS_CONNECT:
2260 		err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2261 		break;
2262 	case SYS_LISTEN:
2263 		err = sys_listen(a0, a1);
2264 		break;
2265 	case SYS_ACCEPT:
2266 		err = sys_accept4(a0, (struct sockaddr __user *)a1,
2267 				  (int __user *)a[2], 0);
2268 		break;
2269 	case SYS_GETSOCKNAME:
2270 		err =
2271 		    sys_getsockname(a0, (struct sockaddr __user *)a1,
2272 				    (int __user *)a[2]);
2273 		break;
2274 	case SYS_GETPEERNAME:
2275 		err =
2276 		    sys_getpeername(a0, (struct sockaddr __user *)a1,
2277 				    (int __user *)a[2]);
2278 		break;
2279 	case SYS_SOCKETPAIR:
2280 		err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2281 		break;
2282 	case SYS_SEND:
2283 		err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2284 		break;
2285 	case SYS_SENDTO:
2286 		err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2287 				 (struct sockaddr __user *)a[4], a[5]);
2288 		break;
2289 	case SYS_RECV:
2290 		err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2291 		break;
2292 	case SYS_RECVFROM:
2293 		err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2294 				   (struct sockaddr __user *)a[4],
2295 				   (int __user *)a[5]);
2296 		break;
2297 	case SYS_SHUTDOWN:
2298 		err = sys_shutdown(a0, a1);
2299 		break;
2300 	case SYS_SETSOCKOPT:
2301 		err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2302 		break;
2303 	case SYS_GETSOCKOPT:
2304 		err =
2305 		    sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2306 				   (int __user *)a[4]);
2307 		break;
2308 	case SYS_SENDMSG:
2309 		err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2310 		break;
2311 	case SYS_RECVMSG:
2312 		err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2313 		break;
2314 	case SYS_RECVMMSG:
2315 		err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2316 				   (struct timespec __user *)a[4]);
2317 		break;
2318 	case SYS_ACCEPT4:
2319 		err = sys_accept4(a0, (struct sockaddr __user *)a1,
2320 				  (int __user *)a[2], a[3]);
2321 		break;
2322 	default:
2323 		err = -EINVAL;
2324 		break;
2325 	}
2326 	return err;
2327 }
2328 
2329 #endif				/* __ARCH_WANT_SYS_SOCKETCALL */
2330 
2331 /**
2332  *	sock_register - add a socket protocol handler
2333  *	@ops: description of protocol
2334  *
2335  *	This function is called by a protocol handler that wants to
2336  *	advertise its address family, and have it linked into the
2337  *	socket interface. The value ops->family coresponds to the
2338  *	socket system call protocol family.
2339  */
2340 int sock_register(const struct net_proto_family *ops)
2341 {
2342 	int err;
2343 
2344 	if (ops->family >= NPROTO) {
2345 		printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2346 		       NPROTO);
2347 		return -ENOBUFS;
2348 	}
2349 
2350 	spin_lock(&net_family_lock);
2351 	if (rcu_dereference_protected(net_families[ops->family],
2352 				      lockdep_is_held(&net_family_lock)))
2353 		err = -EEXIST;
2354 	else {
2355 		rcu_assign_pointer(net_families[ops->family], ops);
2356 		err = 0;
2357 	}
2358 	spin_unlock(&net_family_lock);
2359 
2360 	printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2361 	return err;
2362 }
2363 EXPORT_SYMBOL(sock_register);
2364 
2365 /**
2366  *	sock_unregister - remove a protocol handler
2367  *	@family: protocol family to remove
2368  *
2369  *	This function is called by a protocol handler that wants to
2370  *	remove its address family, and have it unlinked from the
2371  *	new socket creation.
2372  *
2373  *	If protocol handler is a module, then it can use module reference
2374  *	counts to protect against new references. If protocol handler is not
2375  *	a module then it needs to provide its own protection in
2376  *	the ops->create routine.
2377  */
2378 void sock_unregister(int family)
2379 {
2380 	BUG_ON(family < 0 || family >= NPROTO);
2381 
2382 	spin_lock(&net_family_lock);
2383 	rcu_assign_pointer(net_families[family], NULL);
2384 	spin_unlock(&net_family_lock);
2385 
2386 	synchronize_rcu();
2387 
2388 	printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2389 }
2390 EXPORT_SYMBOL(sock_unregister);
2391 
2392 static int __init sock_init(void)
2393 {
2394 	int err;
2395 
2396 	/*
2397 	 *      Initialize sock SLAB cache.
2398 	 */
2399 
2400 	sk_init();
2401 
2402 	/*
2403 	 *      Initialize skbuff SLAB cache
2404 	 */
2405 	skb_init();
2406 
2407 	/*
2408 	 *      Initialize the protocols module.
2409 	 */
2410 
2411 	init_inodecache();
2412 
2413 	err = register_filesystem(&sock_fs_type);
2414 	if (err)
2415 		goto out_fs;
2416 	sock_mnt = kern_mount(&sock_fs_type);
2417 	if (IS_ERR(sock_mnt)) {
2418 		err = PTR_ERR(sock_mnt);
2419 		goto out_mount;
2420 	}
2421 
2422 	/* The real protocol initialization is performed in later initcalls.
2423 	 */
2424 
2425 #ifdef CONFIG_NETFILTER
2426 	netfilter_init();
2427 #endif
2428 
2429 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2430 	skb_timestamping_init();
2431 #endif
2432 
2433 out:
2434 	return err;
2435 
2436 out_mount:
2437 	unregister_filesystem(&sock_fs_type);
2438 out_fs:
2439 	goto out;
2440 }
2441 
2442 core_initcall(sock_init);	/* early initcall */
2443 
2444 #ifdef CONFIG_PROC_FS
2445 void socket_seq_show(struct seq_file *seq)
2446 {
2447 	int cpu;
2448 	int counter = 0;
2449 
2450 	for_each_possible_cpu(cpu)
2451 	    counter += per_cpu(sockets_in_use, cpu);
2452 
2453 	/* It can be negative, by the way. 8) */
2454 	if (counter < 0)
2455 		counter = 0;
2456 
2457 	seq_printf(seq, "sockets: used %d\n", counter);
2458 }
2459 #endif				/* CONFIG_PROC_FS */
2460 
2461 #ifdef CONFIG_COMPAT
2462 static int do_siocgstamp(struct net *net, struct socket *sock,
2463 			 unsigned int cmd, struct compat_timeval __user *up)
2464 {
2465 	mm_segment_t old_fs = get_fs();
2466 	struct timeval ktv;
2467 	int err;
2468 
2469 	set_fs(KERNEL_DS);
2470 	err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2471 	set_fs(old_fs);
2472 	if (!err) {
2473 		err = put_user(ktv.tv_sec, &up->tv_sec);
2474 		err |= __put_user(ktv.tv_usec, &up->tv_usec);
2475 	}
2476 	return err;
2477 }
2478 
2479 static int do_siocgstampns(struct net *net, struct socket *sock,
2480 			 unsigned int cmd, struct compat_timespec __user *up)
2481 {
2482 	mm_segment_t old_fs = get_fs();
2483 	struct timespec kts;
2484 	int err;
2485 
2486 	set_fs(KERNEL_DS);
2487 	err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2488 	set_fs(old_fs);
2489 	if (!err) {
2490 		err = put_user(kts.tv_sec, &up->tv_sec);
2491 		err |= __put_user(kts.tv_nsec, &up->tv_nsec);
2492 	}
2493 	return err;
2494 }
2495 
2496 static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2497 {
2498 	struct ifreq __user *uifr;
2499 	int err;
2500 
2501 	uifr = compat_alloc_user_space(sizeof(struct ifreq));
2502 	if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2503 		return -EFAULT;
2504 
2505 	err = dev_ioctl(net, SIOCGIFNAME, uifr);
2506 	if (err)
2507 		return err;
2508 
2509 	if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2510 		return -EFAULT;
2511 
2512 	return 0;
2513 }
2514 
2515 static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2516 {
2517 	struct compat_ifconf ifc32;
2518 	struct ifconf ifc;
2519 	struct ifconf __user *uifc;
2520 	struct compat_ifreq __user *ifr32;
2521 	struct ifreq __user *ifr;
2522 	unsigned int i, j;
2523 	int err;
2524 
2525 	if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2526 		return -EFAULT;
2527 
2528 	if (ifc32.ifcbuf == 0) {
2529 		ifc32.ifc_len = 0;
2530 		ifc.ifc_len = 0;
2531 		ifc.ifc_req = NULL;
2532 		uifc = compat_alloc_user_space(sizeof(struct ifconf));
2533 	} else {
2534 		size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2535 			sizeof(struct ifreq);
2536 		uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2537 		ifc.ifc_len = len;
2538 		ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2539 		ifr32 = compat_ptr(ifc32.ifcbuf);
2540 		for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2541 			if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2542 				return -EFAULT;
2543 			ifr++;
2544 			ifr32++;
2545 		}
2546 	}
2547 	if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2548 		return -EFAULT;
2549 
2550 	err = dev_ioctl(net, SIOCGIFCONF, uifc);
2551 	if (err)
2552 		return err;
2553 
2554 	if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2555 		return -EFAULT;
2556 
2557 	ifr = ifc.ifc_req;
2558 	ifr32 = compat_ptr(ifc32.ifcbuf);
2559 	for (i = 0, j = 0;
2560 	     i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2561 	     i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2562 		if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2563 			return -EFAULT;
2564 		ifr32++;
2565 		ifr++;
2566 	}
2567 
2568 	if (ifc32.ifcbuf == 0) {
2569 		/* Translate from 64-bit structure multiple to
2570 		 * a 32-bit one.
2571 		 */
2572 		i = ifc.ifc_len;
2573 		i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2574 		ifc32.ifc_len = i;
2575 	} else {
2576 		ifc32.ifc_len = i;
2577 	}
2578 	if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2579 		return -EFAULT;
2580 
2581 	return 0;
2582 }
2583 
2584 static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2585 {
2586 	struct ifreq __user *ifr;
2587 	u32 data;
2588 	void __user *datap;
2589 
2590 	ifr = compat_alloc_user_space(sizeof(*ifr));
2591 
2592 	if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2593 		return -EFAULT;
2594 
2595 	if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2596 		return -EFAULT;
2597 
2598 	datap = compat_ptr(data);
2599 	if (put_user(datap, &ifr->ifr_ifru.ifru_data))
2600 		return -EFAULT;
2601 
2602 	return dev_ioctl(net, SIOCETHTOOL, ifr);
2603 }
2604 
2605 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2606 {
2607 	void __user *uptr;
2608 	compat_uptr_t uptr32;
2609 	struct ifreq __user *uifr;
2610 
2611 	uifr = compat_alloc_user_space(sizeof(*uifr));
2612 	if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2613 		return -EFAULT;
2614 
2615 	if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2616 		return -EFAULT;
2617 
2618 	uptr = compat_ptr(uptr32);
2619 
2620 	if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2621 		return -EFAULT;
2622 
2623 	return dev_ioctl(net, SIOCWANDEV, uifr);
2624 }
2625 
2626 static int bond_ioctl(struct net *net, unsigned int cmd,
2627 			 struct compat_ifreq __user *ifr32)
2628 {
2629 	struct ifreq kifr;
2630 	struct ifreq __user *uifr;
2631 	mm_segment_t old_fs;
2632 	int err;
2633 	u32 data;
2634 	void __user *datap;
2635 
2636 	switch (cmd) {
2637 	case SIOCBONDENSLAVE:
2638 	case SIOCBONDRELEASE:
2639 	case SIOCBONDSETHWADDR:
2640 	case SIOCBONDCHANGEACTIVE:
2641 		if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2642 			return -EFAULT;
2643 
2644 		old_fs = get_fs();
2645 		set_fs(KERNEL_DS);
2646 		err = dev_ioctl(net, cmd, &kifr);
2647 		set_fs(old_fs);
2648 
2649 		return err;
2650 	case SIOCBONDSLAVEINFOQUERY:
2651 	case SIOCBONDINFOQUERY:
2652 		uifr = compat_alloc_user_space(sizeof(*uifr));
2653 		if (copy_in_user(&uifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2654 			return -EFAULT;
2655 
2656 		if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2657 			return -EFAULT;
2658 
2659 		datap = compat_ptr(data);
2660 		if (put_user(datap, &uifr->ifr_ifru.ifru_data))
2661 			return -EFAULT;
2662 
2663 		return dev_ioctl(net, cmd, uifr);
2664 	default:
2665 		return -EINVAL;
2666 	}
2667 }
2668 
2669 static int siocdevprivate_ioctl(struct net *net, unsigned int cmd,
2670 				 struct compat_ifreq __user *u_ifreq32)
2671 {
2672 	struct ifreq __user *u_ifreq64;
2673 	char tmp_buf[IFNAMSIZ];
2674 	void __user *data64;
2675 	u32 data32;
2676 
2677 	if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2678 			   IFNAMSIZ))
2679 		return -EFAULT;
2680 	if (__get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2681 		return -EFAULT;
2682 	data64 = compat_ptr(data32);
2683 
2684 	u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2685 
2686 	/* Don't check these user accesses, just let that get trapped
2687 	 * in the ioctl handler instead.
2688 	 */
2689 	if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2690 			 IFNAMSIZ))
2691 		return -EFAULT;
2692 	if (__put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2693 		return -EFAULT;
2694 
2695 	return dev_ioctl(net, cmd, u_ifreq64);
2696 }
2697 
2698 static int dev_ifsioc(struct net *net, struct socket *sock,
2699 			 unsigned int cmd, struct compat_ifreq __user *uifr32)
2700 {
2701 	struct ifreq __user *uifr;
2702 	int err;
2703 
2704 	uifr = compat_alloc_user_space(sizeof(*uifr));
2705 	if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2706 		return -EFAULT;
2707 
2708 	err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2709 
2710 	if (!err) {
2711 		switch (cmd) {
2712 		case SIOCGIFFLAGS:
2713 		case SIOCGIFMETRIC:
2714 		case SIOCGIFMTU:
2715 		case SIOCGIFMEM:
2716 		case SIOCGIFHWADDR:
2717 		case SIOCGIFINDEX:
2718 		case SIOCGIFADDR:
2719 		case SIOCGIFBRDADDR:
2720 		case SIOCGIFDSTADDR:
2721 		case SIOCGIFNETMASK:
2722 		case SIOCGIFPFLAGS:
2723 		case SIOCGIFTXQLEN:
2724 		case SIOCGMIIPHY:
2725 		case SIOCGMIIREG:
2726 			if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2727 				err = -EFAULT;
2728 			break;
2729 		}
2730 	}
2731 	return err;
2732 }
2733 
2734 static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2735 			struct compat_ifreq __user *uifr32)
2736 {
2737 	struct ifreq ifr;
2738 	struct compat_ifmap __user *uifmap32;
2739 	mm_segment_t old_fs;
2740 	int err;
2741 
2742 	uifmap32 = &uifr32->ifr_ifru.ifru_map;
2743 	err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2744 	err |= __get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2745 	err |= __get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2746 	err |= __get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2747 	err |= __get_user(ifr.ifr_map.irq, &uifmap32->irq);
2748 	err |= __get_user(ifr.ifr_map.dma, &uifmap32->dma);
2749 	err |= __get_user(ifr.ifr_map.port, &uifmap32->port);
2750 	if (err)
2751 		return -EFAULT;
2752 
2753 	old_fs = get_fs();
2754 	set_fs(KERNEL_DS);
2755 	err = dev_ioctl(net, cmd, (void __user *)&ifr);
2756 	set_fs(old_fs);
2757 
2758 	if (cmd == SIOCGIFMAP && !err) {
2759 		err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2760 		err |= __put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2761 		err |= __put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2762 		err |= __put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2763 		err |= __put_user(ifr.ifr_map.irq, &uifmap32->irq);
2764 		err |= __put_user(ifr.ifr_map.dma, &uifmap32->dma);
2765 		err |= __put_user(ifr.ifr_map.port, &uifmap32->port);
2766 		if (err)
2767 			err = -EFAULT;
2768 	}
2769 	return err;
2770 }
2771 
2772 static int compat_siocshwtstamp(struct net *net, struct compat_ifreq __user *uifr32)
2773 {
2774 	void __user *uptr;
2775 	compat_uptr_t uptr32;
2776 	struct ifreq __user *uifr;
2777 
2778 	uifr = compat_alloc_user_space(sizeof(*uifr));
2779 	if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2780 		return -EFAULT;
2781 
2782 	if (get_user(uptr32, &uifr32->ifr_data))
2783 		return -EFAULT;
2784 
2785 	uptr = compat_ptr(uptr32);
2786 
2787 	if (put_user(uptr, &uifr->ifr_data))
2788 		return -EFAULT;
2789 
2790 	return dev_ioctl(net, SIOCSHWTSTAMP, uifr);
2791 }
2792 
2793 struct rtentry32 {
2794 	u32		rt_pad1;
2795 	struct sockaddr rt_dst;         /* target address               */
2796 	struct sockaddr rt_gateway;     /* gateway addr (RTF_GATEWAY)   */
2797 	struct sockaddr rt_genmask;     /* target network mask (IP)     */
2798 	unsigned short	rt_flags;
2799 	short		rt_pad2;
2800 	u32		rt_pad3;
2801 	unsigned char	rt_tos;
2802 	unsigned char	rt_class;
2803 	short		rt_pad4;
2804 	short		rt_metric;      /* +1 for binary compatibility! */
2805 	/* char * */ u32 rt_dev;        /* forcing the device at add    */
2806 	u32		rt_mtu;         /* per route MTU/Window         */
2807 	u32		rt_window;      /* Window clamping              */
2808 	unsigned short  rt_irtt;        /* Initial RTT                  */
2809 };
2810 
2811 struct in6_rtmsg32 {
2812 	struct in6_addr		rtmsg_dst;
2813 	struct in6_addr		rtmsg_src;
2814 	struct in6_addr		rtmsg_gateway;
2815 	u32			rtmsg_type;
2816 	u16			rtmsg_dst_len;
2817 	u16			rtmsg_src_len;
2818 	u32			rtmsg_metric;
2819 	u32			rtmsg_info;
2820 	u32			rtmsg_flags;
2821 	s32			rtmsg_ifindex;
2822 };
2823 
2824 static int routing_ioctl(struct net *net, struct socket *sock,
2825 			 unsigned int cmd, void __user *argp)
2826 {
2827 	int ret;
2828 	void *r = NULL;
2829 	struct in6_rtmsg r6;
2830 	struct rtentry r4;
2831 	char devname[16];
2832 	u32 rtdev;
2833 	mm_segment_t old_fs = get_fs();
2834 
2835 	if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
2836 		struct in6_rtmsg32 __user *ur6 = argp;
2837 		ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
2838 			3 * sizeof(struct in6_addr));
2839 		ret |= __get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
2840 		ret |= __get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
2841 		ret |= __get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
2842 		ret |= __get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
2843 		ret |= __get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
2844 		ret |= __get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
2845 		ret |= __get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
2846 
2847 		r = (void *) &r6;
2848 	} else { /* ipv4 */
2849 		struct rtentry32 __user *ur4 = argp;
2850 		ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
2851 					3 * sizeof(struct sockaddr));
2852 		ret |= __get_user(r4.rt_flags, &(ur4->rt_flags));
2853 		ret |= __get_user(r4.rt_metric, &(ur4->rt_metric));
2854 		ret |= __get_user(r4.rt_mtu, &(ur4->rt_mtu));
2855 		ret |= __get_user(r4.rt_window, &(ur4->rt_window));
2856 		ret |= __get_user(r4.rt_irtt, &(ur4->rt_irtt));
2857 		ret |= __get_user(rtdev, &(ur4->rt_dev));
2858 		if (rtdev) {
2859 			ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
2860 			r4.rt_dev = devname; devname[15] = 0;
2861 		} else
2862 			r4.rt_dev = NULL;
2863 
2864 		r = (void *) &r4;
2865 	}
2866 
2867 	if (ret) {
2868 		ret = -EFAULT;
2869 		goto out;
2870 	}
2871 
2872 	set_fs(KERNEL_DS);
2873 	ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
2874 	set_fs(old_fs);
2875 
2876 out:
2877 	return ret;
2878 }
2879 
2880 /* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
2881  * for some operations; this forces use of the newer bridge-utils that
2882  * use compatiable ioctls
2883  */
2884 static int old_bridge_ioctl(compat_ulong_t __user *argp)
2885 {
2886 	compat_ulong_t tmp;
2887 
2888 	if (get_user(tmp, argp))
2889 		return -EFAULT;
2890 	if (tmp == BRCTL_GET_VERSION)
2891 		return BRCTL_VERSION + 1;
2892 	return -EINVAL;
2893 }
2894 
2895 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
2896 			 unsigned int cmd, unsigned long arg)
2897 {
2898 	void __user *argp = compat_ptr(arg);
2899 	struct sock *sk = sock->sk;
2900 	struct net *net = sock_net(sk);
2901 
2902 	if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
2903 		return siocdevprivate_ioctl(net, cmd, argp);
2904 
2905 	switch (cmd) {
2906 	case SIOCSIFBR:
2907 	case SIOCGIFBR:
2908 		return old_bridge_ioctl(argp);
2909 	case SIOCGIFNAME:
2910 		return dev_ifname32(net, argp);
2911 	case SIOCGIFCONF:
2912 		return dev_ifconf(net, argp);
2913 	case SIOCETHTOOL:
2914 		return ethtool_ioctl(net, argp);
2915 	case SIOCWANDEV:
2916 		return compat_siocwandev(net, argp);
2917 	case SIOCGIFMAP:
2918 	case SIOCSIFMAP:
2919 		return compat_sioc_ifmap(net, cmd, argp);
2920 	case SIOCBONDENSLAVE:
2921 	case SIOCBONDRELEASE:
2922 	case SIOCBONDSETHWADDR:
2923 	case SIOCBONDSLAVEINFOQUERY:
2924 	case SIOCBONDINFOQUERY:
2925 	case SIOCBONDCHANGEACTIVE:
2926 		return bond_ioctl(net, cmd, argp);
2927 	case SIOCADDRT:
2928 	case SIOCDELRT:
2929 		return routing_ioctl(net, sock, cmd, argp);
2930 	case SIOCGSTAMP:
2931 		return do_siocgstamp(net, sock, cmd, argp);
2932 	case SIOCGSTAMPNS:
2933 		return do_siocgstampns(net, sock, cmd, argp);
2934 	case SIOCSHWTSTAMP:
2935 		return compat_siocshwtstamp(net, argp);
2936 
2937 	case FIOSETOWN:
2938 	case SIOCSPGRP:
2939 	case FIOGETOWN:
2940 	case SIOCGPGRP:
2941 	case SIOCBRADDBR:
2942 	case SIOCBRDELBR:
2943 	case SIOCGIFVLAN:
2944 	case SIOCSIFVLAN:
2945 	case SIOCADDDLCI:
2946 	case SIOCDELDLCI:
2947 		return sock_ioctl(file, cmd, arg);
2948 
2949 	case SIOCGIFFLAGS:
2950 	case SIOCSIFFLAGS:
2951 	case SIOCGIFMETRIC:
2952 	case SIOCSIFMETRIC:
2953 	case SIOCGIFMTU:
2954 	case SIOCSIFMTU:
2955 	case SIOCGIFMEM:
2956 	case SIOCSIFMEM:
2957 	case SIOCGIFHWADDR:
2958 	case SIOCSIFHWADDR:
2959 	case SIOCADDMULTI:
2960 	case SIOCDELMULTI:
2961 	case SIOCGIFINDEX:
2962 	case SIOCGIFADDR:
2963 	case SIOCSIFADDR:
2964 	case SIOCSIFHWBROADCAST:
2965 	case SIOCDIFADDR:
2966 	case SIOCGIFBRDADDR:
2967 	case SIOCSIFBRDADDR:
2968 	case SIOCGIFDSTADDR:
2969 	case SIOCSIFDSTADDR:
2970 	case SIOCGIFNETMASK:
2971 	case SIOCSIFNETMASK:
2972 	case SIOCSIFPFLAGS:
2973 	case SIOCGIFPFLAGS:
2974 	case SIOCGIFTXQLEN:
2975 	case SIOCSIFTXQLEN:
2976 	case SIOCBRADDIF:
2977 	case SIOCBRDELIF:
2978 	case SIOCSIFNAME:
2979 	case SIOCGMIIPHY:
2980 	case SIOCGMIIREG:
2981 	case SIOCSMIIREG:
2982 		return dev_ifsioc(net, sock, cmd, argp);
2983 
2984 	case SIOCSARP:
2985 	case SIOCGARP:
2986 	case SIOCDARP:
2987 	case SIOCATMARK:
2988 		return sock_do_ioctl(net, sock, cmd, arg);
2989 	}
2990 
2991 	/* Prevent warning from compat_sys_ioctl, these always
2992 	 * result in -EINVAL in the native case anyway. */
2993 	switch (cmd) {
2994 	case SIOCRTMSG:
2995 	case SIOCGIFCOUNT:
2996 	case SIOCSRARP:
2997 	case SIOCGRARP:
2998 	case SIOCDRARP:
2999 	case SIOCSIFLINK:
3000 	case SIOCGIFSLAVE:
3001 	case SIOCSIFSLAVE:
3002 		return -EINVAL;
3003 	}
3004 
3005 	return -ENOIOCTLCMD;
3006 }
3007 
3008 static long compat_sock_ioctl(struct file *file, unsigned cmd,
3009 			      unsigned long arg)
3010 {
3011 	struct socket *sock = file->private_data;
3012 	int ret = -ENOIOCTLCMD;
3013 	struct sock *sk;
3014 	struct net *net;
3015 
3016 	sk = sock->sk;
3017 	net = sock_net(sk);
3018 
3019 	if (sock->ops->compat_ioctl)
3020 		ret = sock->ops->compat_ioctl(sock, cmd, arg);
3021 
3022 	if (ret == -ENOIOCTLCMD &&
3023 	    (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3024 		ret = compat_wext_handle_ioctl(net, cmd, arg);
3025 
3026 	if (ret == -ENOIOCTLCMD)
3027 		ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3028 
3029 	return ret;
3030 }
3031 #endif
3032 
3033 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3034 {
3035 	return sock->ops->bind(sock, addr, addrlen);
3036 }
3037 EXPORT_SYMBOL(kernel_bind);
3038 
3039 int kernel_listen(struct socket *sock, int backlog)
3040 {
3041 	return sock->ops->listen(sock, backlog);
3042 }
3043 EXPORT_SYMBOL(kernel_listen);
3044 
3045 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3046 {
3047 	struct sock *sk = sock->sk;
3048 	int err;
3049 
3050 	err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3051 			       newsock);
3052 	if (err < 0)
3053 		goto done;
3054 
3055 	err = sock->ops->accept(sock, *newsock, flags);
3056 	if (err < 0) {
3057 		sock_release(*newsock);
3058 		*newsock = NULL;
3059 		goto done;
3060 	}
3061 
3062 	(*newsock)->ops = sock->ops;
3063 	__module_get((*newsock)->ops->owner);
3064 
3065 done:
3066 	return err;
3067 }
3068 EXPORT_SYMBOL(kernel_accept);
3069 
3070 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3071 		   int flags)
3072 {
3073 	return sock->ops->connect(sock, addr, addrlen, flags);
3074 }
3075 EXPORT_SYMBOL(kernel_connect);
3076 
3077 int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3078 			 int *addrlen)
3079 {
3080 	return sock->ops->getname(sock, addr, addrlen, 0);
3081 }
3082 EXPORT_SYMBOL(kernel_getsockname);
3083 
3084 int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3085 			 int *addrlen)
3086 {
3087 	return sock->ops->getname(sock, addr, addrlen, 1);
3088 }
3089 EXPORT_SYMBOL(kernel_getpeername);
3090 
3091 int kernel_getsockopt(struct socket *sock, int level, int optname,
3092 			char *optval, int *optlen)
3093 {
3094 	mm_segment_t oldfs = get_fs();
3095 	char __user *uoptval;
3096 	int __user *uoptlen;
3097 	int err;
3098 
3099 	uoptval = (char __user __force *) optval;
3100 	uoptlen = (int __user __force *) optlen;
3101 
3102 	set_fs(KERNEL_DS);
3103 	if (level == SOL_SOCKET)
3104 		err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3105 	else
3106 		err = sock->ops->getsockopt(sock, level, optname, uoptval,
3107 					    uoptlen);
3108 	set_fs(oldfs);
3109 	return err;
3110 }
3111 EXPORT_SYMBOL(kernel_getsockopt);
3112 
3113 int kernel_setsockopt(struct socket *sock, int level, int optname,
3114 			char *optval, unsigned int optlen)
3115 {
3116 	mm_segment_t oldfs = get_fs();
3117 	char __user *uoptval;
3118 	int err;
3119 
3120 	uoptval = (char __user __force *) optval;
3121 
3122 	set_fs(KERNEL_DS);
3123 	if (level == SOL_SOCKET)
3124 		err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3125 	else
3126 		err = sock->ops->setsockopt(sock, level, optname, uoptval,
3127 					    optlen);
3128 	set_fs(oldfs);
3129 	return err;
3130 }
3131 EXPORT_SYMBOL(kernel_setsockopt);
3132 
3133 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3134 		    size_t size, int flags)
3135 {
3136 	sock_update_classid(sock->sk);
3137 
3138 	if (sock->ops->sendpage)
3139 		return sock->ops->sendpage(sock, page, offset, size, flags);
3140 
3141 	return sock_no_sendpage(sock, page, offset, size, flags);
3142 }
3143 EXPORT_SYMBOL(kernel_sendpage);
3144 
3145 int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3146 {
3147 	mm_segment_t oldfs = get_fs();
3148 	int err;
3149 
3150 	set_fs(KERNEL_DS);
3151 	err = sock->ops->ioctl(sock, cmd, arg);
3152 	set_fs(oldfs);
3153 
3154 	return err;
3155 }
3156 EXPORT_SYMBOL(kernel_sock_ioctl);
3157 
3158 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3159 {
3160 	return sock->ops->shutdown(sock, how);
3161 }
3162 EXPORT_SYMBOL(kernel_sock_shutdown);
3163