PageRenderTime 117ms CodeModel.GetById 3ms app.highlight 98ms RepoModel.GetById 1ms app.codeStats 1ms

/kernel/linux-source-2.6.32/net/socket.c

https://bitbucket.org/ChuloChumo/sctp_thesis
C | 2462 lines | 1696 code | 357 blank | 409 comment | 267 complexity | d725fd4b152874cd5264c738470065a3 MD5 | raw file
   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
  91#include <asm/uaccess.h>
  92#include <asm/unistd.h>
  93
  94#include <net/compat.h>
  95#include <net/wext.h>
  96
  97#include <net/sock.h>
  98#include <linux/netfilter.h>
  99
 100static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
 101static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
 102			 unsigned long nr_segs, loff_t pos);
 103static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
 104			  unsigned long nr_segs, loff_t pos);
 105static int sock_mmap(struct file *file, struct vm_area_struct *vma);
 106
 107static int sock_close(struct inode *inode, struct file *file);
 108static unsigned int sock_poll(struct file *file,
 109			      struct poll_table_struct *wait);
 110static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 111#ifdef CONFIG_COMPAT
 112static long compat_sock_ioctl(struct file *file,
 113			      unsigned int cmd, unsigned long arg);
 114#endif
 115static int sock_fasync(int fd, struct file *filp, int on);
 116static ssize_t sock_sendpage(struct file *file, struct page *page,
 117			     int offset, size_t size, loff_t *ppos, int more);
 118static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 119			        struct pipe_inode_info *pipe, size_t len,
 120				unsigned int flags);
 121
 122/*
 123 *	Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
 124 *	in the operation structures but are done directly via the socketcall() multiplexor.
 125 */
 126
 127static const struct file_operations socket_file_ops = {
 128	.owner =	THIS_MODULE,
 129	.llseek =	no_llseek,
 130	.aio_read =	sock_aio_read,
 131	.aio_write =	sock_aio_write,
 132	.poll =		sock_poll,
 133	.unlocked_ioctl = sock_ioctl,
 134#ifdef CONFIG_COMPAT
 135	.compat_ioctl = compat_sock_ioctl,
 136#endif
 137	.mmap =		sock_mmap,
 138	.open =		sock_no_open,	/* special open code to disallow open via /proc */
 139	.release =	sock_close,
 140	.fasync =	sock_fasync,
 141	.sendpage =	sock_sendpage,
 142	.splice_write = generic_splice_sendpage,
 143	.splice_read =	sock_splice_read,
 144};
 145
 146/*
 147 *	The protocol list. Each protocol is registered in here.
 148 */
 149
 150static DEFINE_SPINLOCK(net_family_lock);
 151static const struct net_proto_family *net_families[NPROTO] __read_mostly;
 152
 153/*
 154 *	Statistics counters of the socket lists
 155 */
 156
 157static DEFINE_PER_CPU(int, sockets_in_use) = 0;
 158
 159/*
 160 * Support routines.
 161 * Move socket addresses back and forth across the kernel/user
 162 * divide and look after the messy bits.
 163 */
 164
 165#define MAX_SOCK_ADDR	128		/* 108 for Unix domain -
 166					   16 for IP, 16 for IPX,
 167					   24 for IPv6,
 168					   about 80 for AX.25
 169					   must be at least one bigger than
 170					   the AF_UNIX size (see net/unix/af_unix.c
 171					   :unix_mkname()).
 172					 */
 173
 174/**
 175 *	move_addr_to_kernel	-	copy a socket address into kernel space
 176 *	@uaddr: Address in user space
 177 *	@kaddr: Address in kernel space
 178 *	@ulen: Length in user space
 179 *
 180 *	The address is copied into kernel space. If the provided address is
 181 *	too long an error code of -EINVAL is returned. If the copy gives
 182 *	invalid addresses -EFAULT is returned. On a success 0 is returned.
 183 */
 184
 185int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
 186{
 187	if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
 188		return -EINVAL;
 189	if (ulen == 0)
 190		return 0;
 191	if (copy_from_user(kaddr, uaddr, ulen))
 192		return -EFAULT;
 193	return audit_sockaddr(ulen, kaddr);
 194}
 195
 196/**
 197 *	move_addr_to_user	-	copy an address to user space
 198 *	@kaddr: kernel space address
 199 *	@klen: length of address in kernel
 200 *	@uaddr: user space address
 201 *	@ulen: pointer to user length field
 202 *
 203 *	The value pointed to by ulen on entry is the buffer length available.
 204 *	This is overwritten with the buffer space used. -EINVAL is returned
 205 *	if an overlong buffer is specified or a negative buffer size. -EFAULT
 206 *	is returned if either the buffer or the length field are not
 207 *	accessible.
 208 *	After copying the data up to the limit the user specifies, the true
 209 *	length of the data is written over the length limit the user
 210 *	specified. Zero is returned for a success.
 211 */
 212
 213int move_addr_to_user(struct sockaddr *kaddr, int klen, void __user *uaddr,
 214		      int __user *ulen)
 215{
 216	int err;
 217	int len;
 218
 219	err = get_user(len, ulen);
 220	if (err)
 221		return err;
 222	if (len > klen)
 223		len = klen;
 224	if (len < 0 || len > sizeof(struct sockaddr_storage))
 225		return -EINVAL;
 226	if (len) {
 227		if (audit_sockaddr(klen, kaddr))
 228			return -ENOMEM;
 229		if (copy_to_user(uaddr, kaddr, len))
 230			return -EFAULT;
 231	}
 232	/*
 233	 *      "fromlen shall refer to the value before truncation.."
 234	 *                      1003.1g
 235	 */
 236	return __put_user(klen, ulen);
 237}
 238
 239static struct kmem_cache *sock_inode_cachep __read_mostly;
 240
 241static struct inode *sock_alloc_inode(struct super_block *sb)
 242{
 243	struct socket_alloc *ei;
 244
 245	ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
 246	if (!ei)
 247		return NULL;
 248	init_waitqueue_head(&ei->socket.wait);
 249
 250	ei->socket.fasync_list = NULL;
 251	ei->socket.state = SS_UNCONNECTED;
 252	ei->socket.flags = 0;
 253	ei->socket.ops = NULL;
 254	ei->socket.sk = NULL;
 255	ei->socket.file = NULL;
 256
 257	return &ei->vfs_inode;
 258}
 259
 260static void sock_destroy_inode(struct inode *inode)
 261{
 262	kmem_cache_free(sock_inode_cachep,
 263			container_of(inode, struct socket_alloc, vfs_inode));
 264}
 265
 266static void init_once(void *foo)
 267{
 268	struct socket_alloc *ei = (struct socket_alloc *)foo;
 269
 270	inode_init_once(&ei->vfs_inode);
 271}
 272
 273static int init_inodecache(void)
 274{
 275	sock_inode_cachep = kmem_cache_create("sock_inode_cache",
 276					      sizeof(struct socket_alloc),
 277					      0,
 278					      (SLAB_HWCACHE_ALIGN |
 279					       SLAB_RECLAIM_ACCOUNT |
 280					       SLAB_MEM_SPREAD),
 281					      init_once);
 282	if (sock_inode_cachep == NULL)
 283		return -ENOMEM;
 284	return 0;
 285}
 286
 287static const struct super_operations sockfs_ops = {
 288	.alloc_inode =	sock_alloc_inode,
 289	.destroy_inode =sock_destroy_inode,
 290	.statfs =	simple_statfs,
 291};
 292
 293static int sockfs_get_sb(struct file_system_type *fs_type,
 294			 int flags, const char *dev_name, void *data,
 295			 struct vfsmount *mnt)
 296{
 297	return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
 298			     mnt);
 299}
 300
 301static struct vfsmount *sock_mnt __read_mostly;
 302
 303static struct file_system_type sock_fs_type = {
 304	.name =		"sockfs",
 305	.get_sb =	sockfs_get_sb,
 306	.kill_sb =	kill_anon_super,
 307};
 308
 309static int sockfs_delete_dentry(struct dentry *dentry)
 310{
 311	/*
 312	 * At creation time, we pretended this dentry was hashed
 313	 * (by clearing DCACHE_UNHASHED bit in d_flags)
 314	 * At delete time, we restore the truth : not hashed.
 315	 * (so that dput() can proceed correctly)
 316	 */
 317	dentry->d_flags |= DCACHE_UNHASHED;
 318	return 0;
 319}
 320
 321/*
 322 * sockfs_dname() is called from d_path().
 323 */
 324static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
 325{
 326	return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
 327				dentry->d_inode->i_ino);
 328}
 329
 330static const struct dentry_operations sockfs_dentry_operations = {
 331	.d_delete = sockfs_delete_dentry,
 332	.d_dname  = sockfs_dname,
 333};
 334
 335/*
 336 *	Obtains the first available file descriptor and sets it up for use.
 337 *
 338 *	These functions create file structures and maps them to fd space
 339 *	of the current process. On success it returns file descriptor
 340 *	and file struct implicitly stored in sock->file.
 341 *	Note that another thread may close file descriptor before we return
 342 *	from this function. We use the fact that now we do not refer
 343 *	to socket after mapping. If one day we will need it, this
 344 *	function will increment ref. count on file by 1.
 345 *
 346 *	In any case returned fd MAY BE not valid!
 347 *	This race condition is unavoidable
 348 *	with shared fd spaces, we cannot solve it inside kernel,
 349 *	but we take care of internal coherence yet.
 350 */
 351
 352static int sock_alloc_fd(struct file **filep, int flags)
 353{
 354	int fd;
 355
 356	fd = get_unused_fd_flags(flags);
 357	if (likely(fd >= 0)) {
 358		struct file *file = get_empty_filp();
 359
 360		*filep = file;
 361		if (unlikely(!file)) {
 362			put_unused_fd(fd);
 363			return -ENFILE;
 364		}
 365	} else
 366		*filep = NULL;
 367	return fd;
 368}
 369
 370static int sock_attach_fd(struct socket *sock, struct file *file, int flags)
 371{
 372	struct dentry *dentry;
 373	struct qstr name = { .name = "" };
 374
 375	dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
 376	if (unlikely(!dentry))
 377		return -ENOMEM;
 378
 379	dentry->d_op = &sockfs_dentry_operations;
 380	/*
 381	 * We dont want to push this dentry into global dentry hash table.
 382	 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
 383	 * This permits a working /proc/$pid/fd/XXX on sockets
 384	 */
 385	dentry->d_flags &= ~DCACHE_UNHASHED;
 386	d_instantiate(dentry, SOCK_INODE(sock));
 387
 388	sock->file = file;
 389	init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE,
 390		  &socket_file_ops);
 391	SOCK_INODE(sock)->i_fop = &socket_file_ops;
 392	file->f_flags = O_RDWR | (flags & O_NONBLOCK);
 393	file->f_pos = 0;
 394	file->private_data = sock;
 395
 396	return 0;
 397}
 398
 399int sock_map_fd(struct socket *sock, int flags)
 400{
 401	struct file *newfile;
 402	int fd = sock_alloc_fd(&newfile, flags);
 403
 404	if (likely(fd >= 0)) {
 405		int err = sock_attach_fd(sock, newfile, flags);
 406
 407		if (unlikely(err < 0)) {
 408			put_filp(newfile);
 409			put_unused_fd(fd);
 410			return err;
 411		}
 412		fd_install(fd, newfile);
 413	}
 414	return fd;
 415}
 416
 417static struct socket *sock_from_file(struct file *file, int *err)
 418{
 419	if (file->f_op == &socket_file_ops)
 420		return file->private_data;	/* set in sock_map_fd */
 421
 422	*err = -ENOTSOCK;
 423	return NULL;
 424}
 425
 426/**
 427 *	sockfd_lookup	- 	Go from a file number to its socket slot
 428 *	@fd: file handle
 429 *	@err: pointer to an error code return
 430 *
 431 *	The file handle passed in is locked and the socket it is bound
 432 *	too is returned. If an error occurs the err pointer is overwritten
 433 *	with a negative errno code and NULL is returned. The function checks
 434 *	for both invalid handles and passing a handle which is not a socket.
 435 *
 436 *	On a success the socket object pointer is returned.
 437 */
 438
 439struct socket *sockfd_lookup(int fd, int *err)
 440{
 441	struct file *file;
 442	struct socket *sock;
 443
 444	file = fget(fd);
 445	if (!file) {
 446		*err = -EBADF;
 447		return NULL;
 448	}
 449
 450	sock = sock_from_file(file, err);
 451	if (!sock)
 452		fput(file);
 453	return sock;
 454}
 455
 456static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
 457{
 458	struct file *file;
 459	struct socket *sock;
 460
 461	*err = -EBADF;
 462	file = fget_light(fd, fput_needed);
 463	if (file) {
 464		sock = sock_from_file(file, err);
 465		if (sock)
 466			return sock;
 467		fput_light(file, *fput_needed);
 468	}
 469	return NULL;
 470}
 471
 472/**
 473 *	sock_alloc	-	allocate a socket
 474 *
 475 *	Allocate a new inode and socket object. The two are bound together
 476 *	and initialised. The socket is then returned. If we are out of inodes
 477 *	NULL is returned.
 478 */
 479
 480static struct socket *sock_alloc(void)
 481{
 482	struct inode *inode;
 483	struct socket *sock;
 484
 485	inode = new_inode(sock_mnt->mnt_sb);
 486	if (!inode)
 487		return NULL;
 488
 489	sock = SOCKET_I(inode);
 490
 491	kmemcheck_annotate_bitfield(sock, type);
 492	inode->i_mode = S_IFSOCK | S_IRWXUGO;
 493	inode->i_uid = current_fsuid();
 494	inode->i_gid = current_fsgid();
 495
 496	percpu_add(sockets_in_use, 1);
 497	return sock;
 498}
 499
 500/*
 501 *	In theory you can't get an open on this inode, but /proc provides
 502 *	a back door. Remember to keep it shut otherwise you'll let the
 503 *	creepy crawlies in.
 504 */
 505
 506static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
 507{
 508	return -ENXIO;
 509}
 510
 511const struct file_operations bad_sock_fops = {
 512	.owner = THIS_MODULE,
 513	.open = sock_no_open,
 514};
 515
 516/**
 517 *	sock_release	-	close a socket
 518 *	@sock: socket to close
 519 *
 520 *	The socket is released from the protocol stack if it has a release
 521 *	callback, and the inode is then released if the socket is bound to
 522 *	an inode not a file.
 523 */
 524
 525void sock_release(struct socket *sock)
 526{
 527	if (sock->ops) {
 528		struct module *owner = sock->ops->owner;
 529
 530		sock->ops->release(sock);
 531		sock->ops = NULL;
 532		module_put(owner);
 533	}
 534
 535	if (sock->fasync_list)
 536		printk(KERN_ERR "sock_release: fasync list not empty!\n");
 537
 538	percpu_sub(sockets_in_use, 1);
 539	if (!sock->file) {
 540		iput(SOCK_INODE(sock));
 541		return;
 542	}
 543	sock->file = NULL;
 544}
 545
 546int sock_tx_timestamp(struct msghdr *msg, struct sock *sk,
 547		      union skb_shared_tx *shtx)
 548{
 549	shtx->flags = 0;
 550	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 551		shtx->hardware = 1;
 552	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 553		shtx->software = 1;
 554	return 0;
 555}
 556EXPORT_SYMBOL(sock_tx_timestamp);
 557
 558static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
 559				 struct msghdr *msg, size_t size)
 560{
 561	struct sock_iocb *si = kiocb_to_siocb(iocb);
 562	int err;
 563
 564	si->sock = sock;
 565	si->scm = NULL;
 566	si->msg = msg;
 567	si->size = size;
 568
 569	err = security_socket_sendmsg(sock, msg, size);
 570	if (err)
 571		return err;
 572
 573	return sock->ops->sendmsg(iocb, sock, msg, size);
 574}
 575
 576int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
 577{
 578	struct kiocb iocb;
 579	struct sock_iocb siocb;
 580	int ret;
 581
 582	init_sync_kiocb(&iocb, NULL);
 583	iocb.private = &siocb;
 584	ret = __sock_sendmsg(&iocb, sock, msg, size);
 585	if (-EIOCBQUEUED == ret)
 586		ret = wait_on_sync_kiocb(&iocb);
 587	return ret;
 588}
 589
 590int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
 591		   struct kvec *vec, size_t num, size_t size)
 592{
 593	mm_segment_t oldfs = get_fs();
 594	int result;
 595
 596	set_fs(KERNEL_DS);
 597	/*
 598	 * the following is safe, since for compiler definitions of kvec and
 599	 * iovec are identical, yielding the same in-core layout and alignment
 600	 */
 601	msg->msg_iov = (struct iovec *)vec;
 602	msg->msg_iovlen = num;
 603	result = sock_sendmsg(sock, msg, size);
 604	set_fs(oldfs);
 605	return result;
 606}
 607
 608static int ktime2ts(ktime_t kt, struct timespec *ts)
 609{
 610	if (kt.tv64) {
 611		*ts = ktime_to_timespec(kt);
 612		return 1;
 613	} else {
 614		return 0;
 615	}
 616}
 617
 618/*
 619 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
 620 */
 621void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 622	struct sk_buff *skb)
 623{
 624	int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
 625	struct timespec ts[3];
 626	int empty = 1;
 627	struct skb_shared_hwtstamps *shhwtstamps =
 628		skb_hwtstamps(skb);
 629
 630	/* Race occurred between timestamp enabling and packet
 631	   receiving.  Fill in the current time for now. */
 632	if (need_software_tstamp && skb->tstamp.tv64 == 0)
 633		__net_timestamp(skb);
 634
 635	if (need_software_tstamp) {
 636		if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
 637			struct timeval tv;
 638			skb_get_timestamp(skb, &tv);
 639			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 640				 sizeof(tv), &tv);
 641		} else {
 642			struct timespec ts;
 643			skb_get_timestampns(skb, &ts);
 644			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
 645				 sizeof(ts), &ts);
 646		}
 647	}
 648
 649
 650	memset(ts, 0, sizeof(ts));
 651	if (skb->tstamp.tv64 &&
 652	    sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
 653		skb_get_timestampns(skb, ts + 0);
 654		empty = 0;
 655	}
 656	if (shhwtstamps) {
 657		if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
 658		    ktime2ts(shhwtstamps->syststamp, ts + 1))
 659			empty = 0;
 660		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
 661		    ktime2ts(shhwtstamps->hwtstamp, ts + 2))
 662			empty = 0;
 663	}
 664	if (!empty)
 665		put_cmsg(msg, SOL_SOCKET,
 666			 SCM_TIMESTAMPING, sizeof(ts), &ts);
 667}
 668
 669EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
 670
 671static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
 672				 struct msghdr *msg, size_t size, int flags)
 673{
 674	int err;
 675	struct sock_iocb *si = kiocb_to_siocb(iocb);
 676
 677	si->sock = sock;
 678	si->scm = NULL;
 679	si->msg = msg;
 680	si->size = size;
 681	si->flags = flags;
 682
 683	err = security_socket_recvmsg(sock, msg, size, flags);
 684	if (err)
 685		return err;
 686
 687	return sock->ops->recvmsg(iocb, sock, msg, size, flags);
 688}
 689
 690int sock_recvmsg(struct socket *sock, struct msghdr *msg,
 691		 size_t size, int flags)
 692{
 693	struct kiocb iocb;
 694	struct sock_iocb siocb;
 695	int ret;
 696
 697	init_sync_kiocb(&iocb, NULL);
 698	iocb.private = &siocb;
 699	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
 700	if (-EIOCBQUEUED == ret)
 701		ret = wait_on_sync_kiocb(&iocb);
 702	return ret;
 703}
 704
 705int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
 706		   struct kvec *vec, size_t num, size_t size, int flags)
 707{
 708	mm_segment_t oldfs = get_fs();
 709	int result;
 710
 711	set_fs(KERNEL_DS);
 712	/*
 713	 * the following is safe, since for compiler definitions of kvec and
 714	 * iovec are identical, yielding the same in-core layout and alignment
 715	 */
 716	msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
 717	result = sock_recvmsg(sock, msg, size, flags);
 718	set_fs(oldfs);
 719	return result;
 720}
 721
 722static void sock_aio_dtor(struct kiocb *iocb)
 723{
 724	kfree(iocb->private);
 725}
 726
 727static ssize_t sock_sendpage(struct file *file, struct page *page,
 728			     int offset, size_t size, loff_t *ppos, int more)
 729{
 730	struct socket *sock;
 731	int flags;
 732
 733	sock = file->private_data;
 734
 735	flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
 736	if (more)
 737		flags |= MSG_MORE;
 738
 739	return kernel_sendpage(sock, page, offset, size, flags);
 740}
 741
 742static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 743			        struct pipe_inode_info *pipe, size_t len,
 744				unsigned int flags)
 745{
 746	struct socket *sock = file->private_data;
 747
 748	if (unlikely(!sock->ops->splice_read))
 749		return -EINVAL;
 750
 751	return sock->ops->splice_read(sock, ppos, pipe, len, flags);
 752}
 753
 754static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
 755					 struct sock_iocb *siocb)
 756{
 757	if (!is_sync_kiocb(iocb)) {
 758		siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
 759		if (!siocb)
 760			return NULL;
 761		iocb->ki_dtor = sock_aio_dtor;
 762	}
 763
 764	siocb->kiocb = iocb;
 765	iocb->private = siocb;
 766	return siocb;
 767}
 768
 769static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
 770		struct file *file, const struct iovec *iov,
 771		unsigned long nr_segs)
 772{
 773	struct socket *sock = file->private_data;
 774	size_t size = 0;
 775	int i;
 776
 777	for (i = 0; i < nr_segs; i++)
 778		size += iov[i].iov_len;
 779
 780	msg->msg_name = NULL;
 781	msg->msg_namelen = 0;
 782	msg->msg_control = NULL;
 783	msg->msg_controllen = 0;
 784	msg->msg_iov = (struct iovec *)iov;
 785	msg->msg_iovlen = nr_segs;
 786	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 787
 788	return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
 789}
 790
 791static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
 792				unsigned long nr_segs, loff_t pos)
 793{
 794	struct sock_iocb siocb, *x;
 795
 796	if (pos != 0)
 797		return -ESPIPE;
 798
 799	if (iocb->ki_left == 0)	/* Match SYS5 behaviour */
 800		return 0;
 801
 802
 803	x = alloc_sock_iocb(iocb, &siocb);
 804	if (!x)
 805		return -ENOMEM;
 806	return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
 807}
 808
 809static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
 810			struct file *file, const struct iovec *iov,
 811			unsigned long nr_segs)
 812{
 813	struct socket *sock = file->private_data;
 814	size_t size = 0;
 815	int i;
 816
 817	for (i = 0; i < nr_segs; i++)
 818		size += iov[i].iov_len;
 819
 820	msg->msg_name = NULL;
 821	msg->msg_namelen = 0;
 822	msg->msg_control = NULL;
 823	msg->msg_controllen = 0;
 824	msg->msg_iov = (struct iovec *)iov;
 825	msg->msg_iovlen = nr_segs;
 826	msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 827	if (sock->type == SOCK_SEQPACKET)
 828		msg->msg_flags |= MSG_EOR;
 829
 830	return __sock_sendmsg(iocb, sock, msg, size);
 831}
 832
 833static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
 834			  unsigned long nr_segs, loff_t pos)
 835{
 836	struct sock_iocb siocb, *x;
 837
 838	if (pos != 0)
 839		return -ESPIPE;
 840
 841	x = alloc_sock_iocb(iocb, &siocb);
 842	if (!x)
 843		return -ENOMEM;
 844
 845	return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
 846}
 847
 848/*
 849 * Atomic setting of ioctl hooks to avoid race
 850 * with module unload.
 851 */
 852
 853static DEFINE_MUTEX(br_ioctl_mutex);
 854static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
 855
 856void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
 857{
 858	mutex_lock(&br_ioctl_mutex);
 859	br_ioctl_hook = hook;
 860	mutex_unlock(&br_ioctl_mutex);
 861}
 862
 863EXPORT_SYMBOL(brioctl_set);
 864
 865static DEFINE_MUTEX(vlan_ioctl_mutex);
 866static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
 867
 868void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
 869{
 870	mutex_lock(&vlan_ioctl_mutex);
 871	vlan_ioctl_hook = hook;
 872	mutex_unlock(&vlan_ioctl_mutex);
 873}
 874
 875EXPORT_SYMBOL(vlan_ioctl_set);
 876
 877static DEFINE_MUTEX(dlci_ioctl_mutex);
 878static int (*dlci_ioctl_hook) (unsigned int, void __user *);
 879
 880void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
 881{
 882	mutex_lock(&dlci_ioctl_mutex);
 883	dlci_ioctl_hook = hook;
 884	mutex_unlock(&dlci_ioctl_mutex);
 885}
 886
 887EXPORT_SYMBOL(dlci_ioctl_set);
 888
 889/*
 890 *	With an ioctl, arg may well be a user mode pointer, but we don't know
 891 *	what to do with it - that's up to the protocol still.
 892 */
 893
 894static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 895{
 896	struct socket *sock;
 897	struct sock *sk;
 898	void __user *argp = (void __user *)arg;
 899	int pid, err;
 900	struct net *net;
 901
 902	sock = file->private_data;
 903	sk = sock->sk;
 904	net = sock_net(sk);
 905	if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
 906		err = dev_ioctl(net, cmd, argp);
 907	} else
 908#ifdef CONFIG_WIRELESS_EXT
 909	if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
 910		err = dev_ioctl(net, cmd, argp);
 911	} else
 912#endif				/* CONFIG_WIRELESS_EXT */
 913		switch (cmd) {
 914		case FIOSETOWN:
 915		case SIOCSPGRP:
 916			err = -EFAULT;
 917			if (get_user(pid, (int __user *)argp))
 918				break;
 919			err = f_setown(sock->file, pid, 1);
 920			break;
 921		case FIOGETOWN:
 922		case SIOCGPGRP:
 923			err = put_user(f_getown(sock->file),
 924				       (int __user *)argp);
 925			break;
 926		case SIOCGIFBR:
 927		case SIOCSIFBR:
 928		case SIOCBRADDBR:
 929		case SIOCBRDELBR:
 930			err = -ENOPKG;
 931			if (!br_ioctl_hook)
 932				request_module("bridge");
 933
 934			mutex_lock(&br_ioctl_mutex);
 935			if (br_ioctl_hook)
 936				err = br_ioctl_hook(net, cmd, argp);
 937			mutex_unlock(&br_ioctl_mutex);
 938			break;
 939		case SIOCGIFVLAN:
 940		case SIOCSIFVLAN:
 941			err = -ENOPKG;
 942			if (!vlan_ioctl_hook)
 943				request_module("8021q");
 944
 945			mutex_lock(&vlan_ioctl_mutex);
 946			if (vlan_ioctl_hook)
 947				err = vlan_ioctl_hook(net, argp);
 948			mutex_unlock(&vlan_ioctl_mutex);
 949			break;
 950		case SIOCADDDLCI:
 951		case SIOCDELDLCI:
 952			err = -ENOPKG;
 953			if (!dlci_ioctl_hook)
 954				request_module("dlci");
 955
 956			mutex_lock(&dlci_ioctl_mutex);
 957			if (dlci_ioctl_hook)
 958				err = dlci_ioctl_hook(cmd, argp);
 959			mutex_unlock(&dlci_ioctl_mutex);
 960			break;
 961		default:
 962			err = sock->ops->ioctl(sock, cmd, arg);
 963
 964			/*
 965			 * If this ioctl is unknown try to hand it down
 966			 * to the NIC driver.
 967			 */
 968			if (err == -ENOIOCTLCMD)
 969				err = dev_ioctl(net, cmd, argp);
 970			break;
 971		}
 972	return err;
 973}
 974
 975int sock_create_lite(int family, int type, int protocol, struct socket **res)
 976{
 977	int err;
 978	struct socket *sock = NULL;
 979
 980	err = security_socket_create(family, type, protocol, 1);
 981	if (err)
 982		goto out;
 983
 984	sock = sock_alloc();
 985	if (!sock) {
 986		err = -ENOMEM;
 987		goto out;
 988	}
 989
 990	sock->type = type;
 991	err = security_socket_post_create(sock, family, type, protocol, 1);
 992	if (err)
 993		goto out_release;
 994
 995out:
 996	*res = sock;
 997	return err;
 998out_release:
 999	sock_release(sock);
1000	sock = NULL;
1001	goto out;
1002}
1003
1004/* No kernel lock held - perfect */
1005static unsigned int sock_poll(struct file *file, poll_table *wait)
1006{
1007	struct socket *sock;
1008
1009	/*
1010	 *      We can't return errors to poll, so it's either yes or no.
1011	 */
1012	sock = file->private_data;
1013	return sock->ops->poll(file, sock, wait);
1014}
1015
1016static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1017{
1018	struct socket *sock = file->private_data;
1019
1020	return sock->ops->mmap(file, sock, vma);
1021}
1022
1023static int sock_close(struct inode *inode, struct file *filp)
1024{
1025	/*
1026	 *      It was possible the inode is NULL we were
1027	 *      closing an unfinished socket.
1028	 */
1029
1030	if (!inode) {
1031		printk(KERN_DEBUG "sock_close: NULL inode\n");
1032		return 0;
1033	}
1034	sock_release(SOCKET_I(inode));
1035	return 0;
1036}
1037
1038/*
1039 *	Update the socket async list
1040 *
1041 *	Fasync_list locking strategy.
1042 *
1043 *	1. fasync_list is modified only under process context socket lock
1044 *	   i.e. under semaphore.
1045 *	2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1046 *	   or under socket lock.
1047 *	3. fasync_list can be used from softirq context, so that
1048 *	   modification under socket lock have to be enhanced with
1049 *	   write_lock_bh(&sk->sk_callback_lock).
1050 *							--ANK (990710)
1051 */
1052
1053static int sock_fasync(int fd, struct file *filp, int on)
1054{
1055	struct fasync_struct *fa, *fna = NULL, **prev;
1056	struct socket *sock;
1057	struct sock *sk;
1058
1059	if (on) {
1060		fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
1061		if (fna == NULL)
1062			return -ENOMEM;
1063	}
1064
1065	sock = filp->private_data;
1066
1067	sk = sock->sk;
1068	if (sk == NULL) {
1069		kfree(fna);
1070		return -EINVAL;
1071	}
1072
1073	lock_sock(sk);
1074
1075	spin_lock(&filp->f_lock);
1076	if (on)
1077		filp->f_flags |= FASYNC;
1078	else
1079		filp->f_flags &= ~FASYNC;
1080	spin_unlock(&filp->f_lock);
1081
1082	prev = &(sock->fasync_list);
1083
1084	for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1085		if (fa->fa_file == filp)
1086			break;
1087
1088	if (on) {
1089		if (fa != NULL) {
1090			write_lock_bh(&sk->sk_callback_lock);
1091			fa->fa_fd = fd;
1092			write_unlock_bh(&sk->sk_callback_lock);
1093
1094			kfree(fna);
1095			goto out;
1096		}
1097		fna->fa_file = filp;
1098		fna->fa_fd = fd;
1099		fna->magic = FASYNC_MAGIC;
1100		fna->fa_next = sock->fasync_list;
1101		write_lock_bh(&sk->sk_callback_lock);
1102		sock->fasync_list = fna;
1103		write_unlock_bh(&sk->sk_callback_lock);
1104	} else {
1105		if (fa != NULL) {
1106			write_lock_bh(&sk->sk_callback_lock);
1107			*prev = fa->fa_next;
1108			write_unlock_bh(&sk->sk_callback_lock);
1109			kfree(fa);
1110		}
1111	}
1112
1113out:
1114	release_sock(sock->sk);
1115	return 0;
1116}
1117
1118/* This function may be called only under socket lock or callback_lock */
1119
1120int sock_wake_async(struct socket *sock, int how, int band)
1121{
1122	if (!sock || !sock->fasync_list)
1123		return -1;
1124	switch (how) {
1125	case SOCK_WAKE_WAITD:
1126		if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1127			break;
1128		goto call_kill;
1129	case SOCK_WAKE_SPACE:
1130		if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1131			break;
1132		/* fall through */
1133	case SOCK_WAKE_IO:
1134call_kill:
1135		__kill_fasync(sock->fasync_list, SIGIO, band);
1136		break;
1137	case SOCK_WAKE_URG:
1138		__kill_fasync(sock->fasync_list, SIGURG, band);
1139	}
1140	return 0;
1141}
1142
1143static int __sock_create(struct net *net, int family, int type, int protocol,
1144			 struct socket **res, int kern)
1145{
1146	int err;
1147	struct socket *sock;
1148	const struct net_proto_family *pf;
1149
1150	/*
1151	 *      Check protocol is in range
1152	 */
1153	if (family < 0 || family >= NPROTO)
1154		return -EAFNOSUPPORT;
1155	if (type < 0 || type >= SOCK_MAX)
1156		return -EINVAL;
1157
1158	/* Compatibility.
1159
1160	   This uglymoron is moved from INET layer to here to avoid
1161	   deadlock in module load.
1162	 */
1163	if (family == PF_INET && type == SOCK_PACKET) {
1164		static int warned;
1165		if (!warned) {
1166			warned = 1;
1167			printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1168			       current->comm);
1169		}
1170		family = PF_PACKET;
1171	}
1172
1173	err = security_socket_create(family, type, protocol, kern);
1174	if (err)
1175		return err;
1176
1177	/*
1178	 *	Allocate the socket and allow the family to set things up. if
1179	 *	the protocol is 0, the family is instructed to select an appropriate
1180	 *	default.
1181	 */
1182	sock = sock_alloc();
1183	if (!sock) {
1184		if (net_ratelimit())
1185			printk(KERN_WARNING "socket: no more sockets\n");
1186		return -ENFILE;	/* Not exactly a match, but its the
1187				   closest posix thing */
1188	}
1189
1190	sock->type = type;
1191
1192#ifdef CONFIG_MODULES
1193	/* Attempt to load a protocol module if the find failed.
1194	 *
1195	 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1196	 * requested real, full-featured networking support upon configuration.
1197	 * Otherwise module support will break!
1198	 */
1199	if (net_families[family] == NULL)
1200		request_module("net-pf-%d", family);
1201#endif
1202
1203	rcu_read_lock();
1204	pf = rcu_dereference(net_families[family]);
1205	err = -EAFNOSUPPORT;
1206	if (!pf)
1207		goto out_release;
1208
1209	/*
1210	 * We will call the ->create function, that possibly is in a loadable
1211	 * module, so we have to bump that loadable module refcnt first.
1212	 */
1213	if (!try_module_get(pf->owner))
1214		goto out_release;
1215
1216	/* Now protected by module ref count */
1217	rcu_read_unlock();
1218
1219	err = pf->create(net, sock, protocol);
1220	if (err < 0)
1221		goto out_module_put;
1222
1223	/*
1224	 * Now to bump the refcnt of the [loadable] module that owns this
1225	 * socket at sock_release time we decrement its refcnt.
1226	 */
1227	if (!try_module_get(sock->ops->owner))
1228		goto out_module_busy;
1229
1230	/*
1231	 * Now that we're done with the ->create function, the [loadable]
1232	 * module can have its refcnt decremented
1233	 */
1234	module_put(pf->owner);
1235	err = security_socket_post_create(sock, family, type, protocol, kern);
1236	if (err)
1237		goto out_sock_release;
1238	*res = sock;
1239
1240	return 0;
1241
1242out_module_busy:
1243	err = -EAFNOSUPPORT;
1244out_module_put:
1245	sock->ops = NULL;
1246	module_put(pf->owner);
1247out_sock_release:
1248	sock_release(sock);
1249	return err;
1250
1251out_release:
1252	rcu_read_unlock();
1253	goto out_sock_release;
1254}
1255
1256int sock_create(int family, int type, int protocol, struct socket **res)
1257{
1258	return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1259}
1260
1261int sock_create_kern(int family, int type, int protocol, struct socket **res)
1262{
1263	return __sock_create(&init_net, family, type, protocol, res, 1);
1264}
1265
1266SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1267{
1268	int retval;
1269	struct socket *sock;
1270	int flags;
1271
1272	/* Check the SOCK_* constants for consistency.  */
1273	BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1274	BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1275	BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1276	BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1277
1278	flags = type & ~SOCK_TYPE_MASK;
1279	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1280		return -EINVAL;
1281	type &= SOCK_TYPE_MASK;
1282
1283	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1284		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1285
1286	retval = sock_create(family, type, protocol, &sock);
1287	if (retval < 0)
1288		goto out;
1289
1290	retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1291	if (retval < 0)
1292		goto out_release;
1293
1294out:
1295	/* It may be already another descriptor 8) Not kernel problem. */
1296	return retval;
1297
1298out_release:
1299	sock_release(sock);
1300	return retval;
1301}
1302
1303/*
1304 *	Create a pair of connected sockets.
1305 */
1306
1307SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1308		int __user *, usockvec)
1309{
1310	struct socket *sock1, *sock2;
1311	int fd1, fd2, err;
1312	struct file *newfile1, *newfile2;
1313	int flags;
1314
1315	flags = type & ~SOCK_TYPE_MASK;
1316	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1317		return -EINVAL;
1318	type &= SOCK_TYPE_MASK;
1319
1320	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1321		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1322
1323	/*
1324	 * Obtain the first socket and check if the underlying protocol
1325	 * supports the socketpair call.
1326	 */
1327
1328	err = sock_create(family, type, protocol, &sock1);
1329	if (err < 0)
1330		goto out;
1331
1332	err = sock_create(family, type, protocol, &sock2);
1333	if (err < 0)
1334		goto out_release_1;
1335
1336	err = sock1->ops->socketpair(sock1, sock2);
1337	if (err < 0)
1338		goto out_release_both;
1339
1340	fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
1341	if (unlikely(fd1 < 0)) {
1342		err = fd1;
1343		goto out_release_both;
1344	}
1345
1346	fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
1347	if (unlikely(fd2 < 0)) {
1348		err = fd2;
1349		put_filp(newfile1);
1350		put_unused_fd(fd1);
1351		goto out_release_both;
1352	}
1353
1354	err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
1355	if (unlikely(err < 0)) {
1356		goto out_fd2;
1357	}
1358
1359	err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
1360	if (unlikely(err < 0)) {
1361		fput(newfile1);
1362		goto out_fd1;
1363	}
1364
1365	audit_fd_pair(fd1, fd2);
1366	fd_install(fd1, newfile1);
1367	fd_install(fd2, newfile2);
1368	/* fd1 and fd2 may be already another descriptors.
1369	 * Not kernel problem.
1370	 */
1371
1372	err = put_user(fd1, &usockvec[0]);
1373	if (!err)
1374		err = put_user(fd2, &usockvec[1]);
1375	if (!err)
1376		return 0;
1377
1378	sys_close(fd2);
1379	sys_close(fd1);
1380	return err;
1381
1382out_release_both:
1383	sock_release(sock2);
1384out_release_1:
1385	sock_release(sock1);
1386out:
1387	return err;
1388
1389out_fd2:
1390	put_filp(newfile1);
1391	sock_release(sock1);
1392out_fd1:
1393	put_filp(newfile2);
1394	sock_release(sock2);
1395	put_unused_fd(fd1);
1396	put_unused_fd(fd2);
1397	goto out;
1398}
1399
1400/*
1401 *	Bind a name to a socket. Nothing much to do here since it's
1402 *	the protocol's responsibility to handle the local address.
1403 *
1404 *	We move the socket address to kernel space before we call
1405 *	the protocol layer (having also checked the address is ok).
1406 */
1407
1408SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1409{
1410	struct socket *sock;
1411	struct sockaddr_storage address;
1412	int err, fput_needed;
1413
1414	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1415	if (sock) {
1416		err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1417		if (err >= 0) {
1418			err = security_socket_bind(sock,
1419						   (struct sockaddr *)&address,
1420						   addrlen);
1421			if (!err)
1422				err = sock->ops->bind(sock,
1423						      (struct sockaddr *)
1424						      &address, addrlen);
1425		}
1426		fput_light(sock->file, fput_needed);
1427	}
1428	return err;
1429}
1430
1431/*
1432 *	Perform a listen. Basically, we allow the protocol to do anything
1433 *	necessary for a listen, and if that works, we mark the socket as
1434 *	ready for listening.
1435 */
1436
1437SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1438{
1439	struct socket *sock;
1440	int err, fput_needed;
1441	int somaxconn;
1442
1443	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1444	if (sock) {
1445		somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1446		if ((unsigned)backlog > somaxconn)
1447			backlog = somaxconn;
1448
1449		err = security_socket_listen(sock, backlog);
1450		if (!err)
1451			err = sock->ops->listen(sock, backlog);
1452
1453		fput_light(sock->file, fput_needed);
1454	}
1455	return err;
1456}
1457
1458/*
1459 *	For accept, we attempt to create a new socket, set up the link
1460 *	with the client, wake up the client, then return the new
1461 *	connected fd. We collect the address of the connector in kernel
1462 *	space and move it to user at the very end. This is unclean because
1463 *	we open the socket then return an error.
1464 *
1465 *	1003.1g adds the ability to recvmsg() to query connection pending
1466 *	status to recvmsg. We need to add that support in a way thats
1467 *	clean when we restucture accept also.
1468 */
1469
1470SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1471		int __user *, upeer_addrlen, int, flags)
1472{
1473	struct socket *sock, *newsock;
1474	struct file *newfile;
1475	int err, len, newfd, fput_needed;
1476	struct sockaddr_storage address;
1477
1478	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1479		return -EINVAL;
1480
1481	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1482		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1483
1484	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1485	if (!sock)
1486		goto out;
1487
1488	err = -ENFILE;
1489	if (!(newsock = sock_alloc()))
1490		goto out_put;
1491
1492	newsock->type = sock->type;
1493	newsock->ops = sock->ops;
1494
1495	/*
1496	 * We don't need try_module_get here, as the listening socket (sock)
1497	 * has the protocol module (sock->ops->owner) held.
1498	 */
1499	__module_get(newsock->ops->owner);
1500
1501	newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
1502	if (unlikely(newfd < 0)) {
1503		err = newfd;
1504		sock_release(newsock);
1505		goto out_put;
1506	}
1507
1508	err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
1509	if (err < 0)
1510		goto out_fd_simple;
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
1537out_put:
1538	fput_light(sock->file, fput_needed);
1539out:
1540	return err;
1541out_fd_simple:
1542	sock_release(newsock);
1543	put_filp(newfile);
1544	put_unused_fd(newfd);
1545	goto out_put;
1546out_fd:
1547	fput(newfile);
1548	put_unused_fd(newfd);
1549	goto out_put;
1550}
1551
1552SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1553		int __user *, upeer_addrlen)
1554{
1555	return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1556}
1557
1558/*
1559 *	Attempt to connect to a socket with the server address.  The address
1560 *	is in user space so we verify it is OK and move it to kernel space.
1561 *
1562 *	For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1563 *	break bindings
1564 *
1565 *	NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1566 *	other SEQPACKET protocols that take time to connect() as it doesn't
1567 *	include the -EINPROGRESS status for such sockets.
1568 */
1569
1570SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1571		int, addrlen)
1572{
1573	struct socket *sock;
1574	struct sockaddr_storage address;
1575	int err, fput_needed;
1576
1577	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1578	if (!sock)
1579		goto out;
1580	err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1581	if (err < 0)
1582		goto out_put;
1583
1584	err =
1585	    security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1586	if (err)
1587		goto out_put;
1588
1589	err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1590				 sock->file->f_flags);
1591out_put:
1592	fput_light(sock->file, fput_needed);
1593out:
1594	return err;
1595}
1596
1597/*
1598 *	Get the local address ('name') of a socket object. Move the obtained
1599 *	name to user space.
1600 */
1601
1602SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1603		int __user *, usockaddr_len)
1604{
1605	struct socket *sock;
1606	struct sockaddr_storage address;
1607	int len, err, fput_needed;
1608
1609	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1610	if (!sock)
1611		goto out;
1612
1613	err = security_socket_getsockname(sock);
1614	if (err)
1615		goto out_put;
1616
1617	err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1618	if (err)
1619		goto out_put;
1620	err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1621
1622out_put:
1623	fput_light(sock->file, fput_needed);
1624out:
1625	return err;
1626}
1627
1628/*
1629 *	Get the remote address ('name') of a socket object. Move the obtained
1630 *	name to user space.
1631 */
1632
1633SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1634		int __user *, usockaddr_len)
1635{
1636	struct socket *sock;
1637	struct sockaddr_storage address;
1638	int len, err, fput_needed;
1639
1640	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1641	if (sock != NULL) {
1642		err = security_socket_getpeername(sock);
1643		if (err) {
1644			fput_light(sock->file, fput_needed);
1645			return err;
1646		}
1647
1648		err =
1649		    sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1650				       1);
1651		if (!err)
1652			err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1653						usockaddr_len);
1654		fput_light(sock->file, fput_needed);
1655	}
1656	return err;
1657}
1658
1659/*
1660 *	Send a datagram to a given address. We move the address into kernel
1661 *	space and check the user space data area is readable before invoking
1662 *	the protocol.
1663 */
1664
1665SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1666		unsigned, flags, struct sockaddr __user *, addr,
1667		int, addr_len)
1668{
1669	struct socket *sock;
1670	struct sockaddr_storage address;
1671	int err;
1672	struct msghdr msg;
1673	struct iovec iov;
1674	int fput_needed;
1675
1676	if (len > INT_MAX)
1677		len = INT_MAX;
1678	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1679	if (!sock)
1680		goto out;
1681
1682	iov.iov_base = buff;
1683	iov.iov_len = len;
1684	msg.msg_name = NULL;
1685	msg.msg_iov = &iov;
1686	msg.msg_iovlen = 1;
1687	msg.msg_control = NULL;
1688	msg.msg_controllen = 0;
1689	msg.msg_namelen = 0;
1690	if (addr) {
1691		err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1692		if (err < 0)
1693			goto out_put;
1694		msg.msg_name = (struct sockaddr *)&address;
1695		msg.msg_namelen = addr_len;
1696	}
1697	if (sock->file->f_flags & O_NONBLOCK)
1698		flags |= MSG_DONTWAIT;
1699	msg.msg_flags = flags;
1700	err = sock_sendmsg(sock, &msg, len);
1701
1702out_put:
1703	fput_light(sock->file, fput_needed);
1704out:
1705	return err;
1706}
1707
1708/*
1709 *	Send a datagram down a socket.
1710 */
1711
1712SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1713		unsigned, flags)
1714{
1715	return sys_sendto(fd, buff, len, flags, NULL, 0);
1716}
1717
1718/*
1719 *	Receive a frame from the socket and optionally record the address of the
1720 *	sender. We verify the buffers are writable and if needed move the
1721 *	sender address from kernel to user space.
1722 */
1723
1724SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1725		unsigned, flags, struct sockaddr __user *, addr,
1726		int __user *, addr_len)
1727{
1728	struct socket *sock;
1729	struct iovec iov;
1730	struct msghdr msg;
1731	struct sockaddr_storage address;
1732	int err, err2;
1733	int fput_needed;
1734
1735	if (size > INT_MAX)
1736		size = INT_MAX;
1737	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1738	if (!sock)
1739		goto out;
1740
1741	msg.msg_control = NULL;
1742	msg.msg_controllen = 0;
1743	msg.msg_iovlen = 1;
1744	msg.msg_iov = &iov;
1745	iov.iov_len = size;
1746	iov.iov_base = ubuf;
1747	msg.msg_name = (struct sockaddr *)&address;
1748	msg.msg_namelen = sizeof(address);
1749	if (sock->file->f_flags & O_NONBLOCK)
1750		flags |= MSG_DONTWAIT;
1751	err = sock_recvmsg(sock, &msg, size, flags);
1752
1753	if (err >= 0 && addr != NULL) {
1754		err2 = move_addr_to_user((struct sockaddr *)&address,
1755					 msg.msg_namelen, addr, addr_len);
1756		if (err2 < 0)
1757			err = err2;
1758	}
1759
1760	fput_light(sock->file, fput_needed);
1761out:
1762	return err;
1763}
1764
1765/*
1766 *	Receive a datagram from a socket.
1767 */
1768
1769asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1770			 unsigned flags)
1771{
1772	return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1773}
1774
1775/*
1776 *	Set a socket option. Because we don't know the option lengths we have
1777 *	to pass the user mode parameter for the protocols to sort out.
1778 */
1779
1780SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1781		char __user *, optval, int, optlen)
1782{
1783	int err, fput_needed;
1784	struct socket *sock;
1785
1786	if (optlen < 0)
1787		return -EINVAL;
1788
1789	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1790	if (sock != NULL) {
1791		err = security_socket_setsockopt(sock, level, optname);
1792		if (err)
1793			goto out_put;
1794
1795		if (level == SOL_SOCKET)
1796			err =
1797			    sock_setsockopt(sock, level, optname, optval,
1798					    optlen);
1799		else
1800			err =
1801			    sock->ops->setsockopt(sock, level, optname, optval,
1802						  optlen);
1803out_put:
1804		fput_light(sock->file, fput_needed);
1805	}
1806	return err;
1807}
1808
1809/*
1810 *	Get a socket option. Because we don't know the option lengths we have
1811 *	to pass a user mode parameter for the protocols to sort out.
1812 */
1813
1814SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1815		char __user *, optval, int __user *, optlen)
1816{
1817	int err, fput_needed;
1818	struct socket *sock;
1819
1820	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1821	if (sock != NULL) {
1822		err = security_socket_getsockopt(sock, level, optname);
1823		if (err)
1824			goto out_put;
1825
1826		if (level == SOL_SOCKET)
1827			err =
1828			    sock_getsockopt(sock, level, optname, optval,
1829					    optlen);
1830		else
1831			err =
1832			    sock->ops->getsockopt(sock, level, optname, optval,
1833						  optlen);
1834out_put:
1835		fput_light(sock->file, fput_needed);
1836	}
1837	return err;
1838}
1839
1840/*
1841 *	Shutdown a socket.
1842 */
1843
1844SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1845{
1846	int err, fput_needed;
1847	struct socket *sock;
1848
1849	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1850	if (sock != NULL) {
1851		err = security_socket_shutdown(sock, how);
1852		if (!err)
1853			err = sock->ops->shutdown(sock, how);
1854		fput_light(sock->file, fput_needed);
1855	}
1856	return err;
1857}
1858
1859/* A couple of helpful macros for getting the address of the 32/64 bit
1860 * fields which are the same type (int / unsigned) on our platforms.
1861 */
1862#define COMPAT_MSG(msg, member)	((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1863#define COMPAT_NAMELEN(msg)	COMPAT_MSG(msg, msg_namelen)
1864#define COMPAT_FLAGS(msg)	COMPAT_MSG(msg, msg_flags)
1865
1866/*
1867 *	BSD sendmsg interface
1868 */
1869
1870SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1871{
1872	struct compat_msghdr __user *msg_compat =
1873	    (struct compat_msghdr __user *)msg;
1874	struct socket *sock;
1875	struct sockaddr_storage address;
1876	struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1877	unsigned char ctl[sizeof(struct cmsghdr) + 20]
1878	    __attribute__ ((aligned(sizeof(__kernel_size_t))));
1879	/* 20 is size of ipv6_pktinfo */
1880	unsigned char *ctl_buf = ctl;
1881	struct msghdr msg_sys;
1882	int err, ctl_len, iov_size, total_len;
1883	int fput_needed;
1884
1885	err = -EFAULT;
1886	if (MSG_CMSG_COMPAT & flags) {
1887		if (get_compat_msghdr(&msg_sys, msg_compat))
1888			return -EFAULT;
1889	}
1890	else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1891		return -EFAULT;
1892
1893	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1894	if (!sock)
1895		goto out;
1896
1897	/* do not move before msg_sys is valid */
1898	err = -EMSGSIZE;
1899	if (msg_sys.msg_iovlen > UIO_MAXIOV)
1900		goto out_put;
1901
1902	/* Check whether to allocate the iovec area */
1903	err = -ENOMEM;
1904	iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1905	if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1906		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1907		if (!iov)
1908			goto out_put;
1909	}
1910
1911	/* This will also move the address data into kernel space */
1912	if (MSG_CMSG_COMPAT & flags) {
1913		err = verify_compat_iovec(&msg_sys, iov,
1914					  (struct sockaddr *)&address,
1915					  VERIFY_READ);
1916	} else
1917		err = verify_iovec(&msg_sys, iov,
1918				   (struct sockaddr *)&address,
1919				   VERIFY_READ);
1920	if (err < 0)
1921		goto out_freeiov;
1922	total_len = err;
1923
1924	err = -ENOBUFS;
1925
1926	if (msg_sys.msg_controllen > INT_MAX)
1927		goto out_freeiov;
1928	ctl_len = msg_sys.msg_controllen;
1929	if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1930		err =
1931		    cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1932						     sizeof(ctl));
1933		if (err)
1934			goto out_freeiov;
1935		ctl_buf = msg_sys.msg_control;
1936		ctl_len = msg_sys.msg_controllen;
1937	} else if (ctl_len) {
1938		if (ctl_len > sizeof(ctl)) {
1939			ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1940			if (ctl_buf == NULL)
1941				goto out_freeiov;
1942		}
1943		err = -EFAULT;
1944		/*
1945		 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1946		 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1947		 * checking falls down on this.
1948		 */
1949		if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1950				   ctl_len))
1951			goto out_freectl;
1952		msg_sys.msg_control = ctl_buf;
1953	}
1954	msg_sys.msg_flags = flags;
1955
1956	if (sock->file->f_flags & O_NONBLOCK)
1957		msg_sys.msg_flags |= MSG_DONTWAIT;
1958	err = sock_sendmsg(sock, &msg_sys, total_len);
1959
1960out_freectl:
1961	if (ctl_buf != ctl)
1962		sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1963out_freeiov:
1964	if (iov != iovstack)
1965		sock_kfree_s(sock->sk, iov, iov_size);
1966out_put:
1967	fput_light(sock->file, fput_needed);
1968out:
1969	return err;
1970}
1971
1972/*
1973 *	BSD recvmsg interface
1974 */
1975
1976SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
1977		unsigned int, flags)
1978{
1979	struct compat_msghdr __user *msg_compat =
1980	    (struct compat_msghdr __user *)msg;
1981	struct socket *sock;
1982	struct iovec iovstack[UIO_FASTIOV];
1983	struct iovec *iov = iovstack;
1984	struct msghdr msg_sys;
1985	unsigned long cmsg_ptr;
1986	int err, iov_size, total_len, len;
1987	int fput_needed;
1988
1989	/* kernel mode address */
1990	struct sockaddr_storage addr;
1991
1992	/* user mode address pointers */
1993	struct sockaddr __user *uaddr;
1994	int __user *uaddr_len;
1995
1996	if (MSG_CMSG_COMPAT & flags) {
1997		if (get_compat_msghdr(&msg_sys, msg_compat))
1998			return -EFAULT;
1999	}
2000	else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
2001		return -EFAULT;
2002
2003	sock = sockfd_lookup_light(fd, &err, &fput_needed);
2004	if (!sock)
2005		goto out;
2006
2007	err = -EMSGSIZE;
2008	if (msg_sys.msg_iovlen > UIO_MAXIOV)
2009		goto out_put;
2010
2011	/* Check whether to allocate the iovec area */
2012	err = -ENOMEM;
2013	iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
2014	if (msg_sys.msg_iovlen > UIO_FASTIOV) {
2015		iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2016		if (!iov)
2017			goto out_put;
2018	}
2019
2020	/*
2021	 *      Save the user-mode address (verify_iovec will change the
2022	 *      kernel msghdr to use the kernel address space)
2023	 */
2024
2025	uaddr = (__force void __user *)msg_sys.msg_name;
2026	uaddr_len = COMPAT_NAMELEN(msg);
2027	if (MSG_CMSG_COMPAT & flags) {
2028		err = verify_compat_iovec(&msg_sys, iov,
2029					  (struct sockaddr *)&addr,
2030					  VERIFY_WRITE);
2031	} else
2032		err = verify_iovec(&msg_sys, iov,
2033				   (struct sockaddr *)&addr,
2034				   VERIFY_WRITE);
2035	if (err < 0)
2036		goto out_freeiov;
2037	total_len = err;
2038
2039	cmsg_ptr = (unsigned long)msg_sys.msg_control;
2040	msg_sys.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2041
2042	if (sock->file->f_flags & O_NONBLOCK)
2043		flags |= MSG_DONTWAIT;
2044	err = sock_recvmsg(sock, &msg_sys, total_len, flags);
2045	if (err < 0)
2046		goto out_freeiov;
2047	len = err;
2048
2049	if (uaddr != NULL) {
2050		err = move_addr_to_user((struct sockaddr *)&addr,
2051					msg_sys.msg_namelen, uaddr,
2052					uaddr_len);
2053		if (err < 0)
2054			goto out_freeiov;
2055	}
2056	err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
2057			 COMPAT_FLAGS(msg));
2058	if (err)
2059		goto out_freeiov;
2060	if (MSG_CMSG_COMPAT & flags)
2061		err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2062				 &msg_compat->msg_controllen);
2063	else
2064		err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2065				 &msg->msg_controllen);
2066	if (err)
2067		goto out_freeiov;
2068	err = len;
2069
2070out_freeiov:
2071	if (iov != iovstack)
2072		sock_kfree_s(sock->sk, iov, iov_size);
2073out_put:
2074	fput_light(sock->file, fput_needed);
2075out:
2076	return err;
2077}
2078
2079#ifdef __ARCH_WANT_SYS_SOCKETCALL
2080
2081/* Argument list sizes for sys_socketcall */
2082#define AL(x) ((x) * sizeof(unsigned long))
2083static const unsigned char nargs[19]={
2084	AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2085	AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
2086	AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
2087	AL(4)
2088};
2089
2090#undef AL
2091
2092/*
2093 *	System call vectors.
2094 *
2095 *	Argument checking cleaned up. Saved 20% in size.
2096 *  This function doesn't need to set the kernel lock because
2097 *  it is set by the callees.
2098 */
2099
2100SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2101{
2102	unsigned long a[6];
2103	unsigned long a0, a1;
2104	int err;
2105	unsigned int len;
2106
2107	if (call < 1 || call > SYS_ACCEPT4)
2108		return -EINVAL;
2109
2110	len = nargs[call];
2111	if (len > sizeof(a))
2112		return -EINVAL;
2113
2114	/* copy_from_user should be SMP safe. */
2115	if (copy_from_user(a, args, len))
2116		return -EFAULT;
2117
2118	audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2119
2120	a0 = a[0];
2121	a1 = a[1];
2122
2123	switch (call) {
2124	case SYS_SOCKET:
2125		err = sys_socket(a0, a1, a[2]);
2126		break;
2127	case SYS_BIND:
2128		err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2129		break;
2130	case SYS_CONNECT:
2131		err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2132		break;
2133	case SYS_LISTEN:
2134		err = sys_listen(a0, a1);
2135		break;
2136	case SYS_ACCEPT:
2137		err = sys_accept4(a0, (struct sockaddr __user *)a1,
2138				  (int __user *)a[2], 0);
2139		break;
2140	case SYS_GETSOCKNAME:
2141		err =
2142		    sys_getsockname(a0, (struct sockaddr __user *)a1,
2143				    (int __user *)a[2]);
2144		break;
2145	case SYS_GETPEERNAME:
2146		err =
2147		    sys_getpeername(a0, (struct sockaddr __user *)a1,
2148				    (int __user *)a[2]);
2149		break;
2150	case SYS_SOCKETPAIR:
2151		err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2152		break;
2153	case SYS_SEND:
2154		err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2155		break;
2156	case SYS_SENDTO:
2157		err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2158				 (struct sockaddr __user *)a[4], a[5]);
2159		break;
2160	case SYS_RECV:
2161		err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2162		break;
2163	case SYS_RECVFROM:
2164		err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2165				   (struct sockaddr __user *)a[4],
2166				   (int __user *)a[5]);
2167		break;
2168	case SYS_SHUTDOWN:
2169		err = sys_shutdown(a0, a1);
2170		break;
2171	case SYS_SETSOCKOPT:
2172		err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2173		break;
2174	case SYS_GETSOCKOPT:
2175		err =
2176		    sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2177				   (int __user *)a[4]);
2178		break;
2179	case SYS_SENDMSG:
2180		err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2181		break;
2182	case SYS_RECVMSG:
2183		err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2184		break;
2185	case SYS_ACCEPT4:
2186		err = sys_accept4(a0, (struct sockaddr __user *)a1,
2187				  (int __user *)a[2], a[3]);
2188		break;
2189	default:
2190		err = -EINVAL;
2191		break;
2192	}
2193	return err;
2194}
2195
2196#endif				/* __ARCH_WANT_SYS_SOCKETCALL */
2197
2198/**
2199 *	sock_register - add a socket protocol handler
2200 *	@ops: description of protocol
2201 *
2202 *	This function is called by a protocol handler that wants to
2203 *	advertise its address family, and have it linked into the
2204 *	socket interface. The value ops->family coresponds to the
2205 *	socket system call protocol family.
2206 */
2207int sock_register(const struct net_proto_family *ops)
2208{
2209	int err;
2210
2211	if (ops->family >= NPROTO) {
2212		printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2213		       NPROTO);
2214		return -ENOBUFS;
2215	}
2216
2217	spin_lock(&net_family_lock);
2218	if (net_families[ops->family])
2219		err = -EEXIST;
2220	else {
2221		net_families[ops->family] = ops;
2222		err = 0;
2223	}
2224	spin_unlock(&net_family_lock);
2225
2226	printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2227	return err;
2228}
2229
2230/**
2231 *	sock_unregister - remove a protocol handler
2232 *	@family: protocol family to remove
2233 *
2234 *	This function is called by a protocol handler that wants to
2235 *	remove its address family, and have it unlinked from the
2236 *	new socket creation.
2237 *
2238 *	If protocol handler is a module, then it can use module reference
2239 *	counts to protect against new references. If protocol handler is not
2240 *	a module then it needs to provide its own protection in
2241 *	the ops->create routine.
2242 */
2243void sock_unregister(int family)
2244{
2245	BUG_ON(family < 0 || family >= NPROTO);
2246
2247	spin_lock(&net_family_lock);
2248	net_families[family] = NULL;
2249	spin_unlock(&net_family_lock);
2250
2251	synchronize_rcu();
2252
2253	printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2254}
2255
2256static int __init sock_init(void)
2257{
2258	/*
2259	 *      Initialize sock SLAB cache.
2260	 */
2261
2262	sk_init();
2263
2264	/*
2265	 *      Initialize skbuff SLAB cache
2266	 */
2267	skb_init();
2268
2269	/*
2270	 *      Initialize the protocols module.
2271	 */
2272
2273	init_inodecache();
2274	register_filesystem(&sock_fs_type);
2275	sock_mnt = kern_mount(&sock_fs_type);
2276
2277	/* The real protocol initialization is performed in later initcalls.
2278	 */
2279
2280#ifdef CONFIG_NETFILTER
2281	netfilter_init();
2282#endif
2283
2284	return 0;
2285}
2286
2287core_initcall(sock_init);	/* early initcall */
2288
2289#ifdef CONFIG_PROC_FS
2290void socket_seq_show(struct seq_file *seq)
2291{
2292	int cpu;
2293	int counter = 0;
2294
2295	for_each_possible_cpu(cpu)
2296	    counter += per_cpu(sockets_in_use, cpu);
2297
2298	/* It can be negative, by the way. 8) */
2299	if (counter < 0)
2300		counter = 0;
2301
2302	seq_printf(seq, "sockets: used %d\n", counter);
2303}
2304#endif				/* CONFIG_PROC_FS */
2305
2306#ifdef CONFIG_COMPAT
2307static long compat_sock_ioctl(struct file *file, unsigned cmd,
2308			      unsigned long arg)
2309{
2310	struct socket *sock = file->private_data;
2311	int ret = -ENOIOCTLCMD;
2312	struct sock *sk;
2313	struct net *net;
2314
2315	sk = sock->sk;
2316	net = sock_net(sk);
2317
2318	if (sock->ops->compat_ioctl)
2319		ret = sock->ops->compat_ioctl(sock, cmd, arg);
2320
2321	if (ret == -ENOIOCTLCMD &&
2322	    (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2323		ret = compat_wext_handle_ioctl(net, cmd, arg);
2324
2325	return ret;
2326}
2327#endif
2328
2329int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2330{
2331	return sock->ops->bind(sock, addr, addrlen);
2332}
2333
2334int kernel_listen(struct socket *sock, int backlog)
2335{
2336	return sock->ops->listen(sock, backlog);
2337}
2338
2339int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2340{
2341	struct sock *sk = sock->sk;
2342	int err;
2343
2344	err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2345			       newsock);
2346	if (err < 0)
2347		goto done;
2348
2349	err = sock->ops->accept(sock, *newsock, flags);
2350	if (err < 0) {
2351		sock_release(*newsock);
2352		*newsock = NULL;
2353		goto done;
2354	}
2355
2356	(*newsock)->ops = sock->ops;
2357	__module_get((*newsock)->ops->owner);
2358
2359done:
2360	return err;
2361}
2362
2363int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
2364		   int flags)
2365{
2366	return sock->ops->connect(sock, addr, addrlen, flags);
2367}
2368
2369int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2370			 int *addrlen)
2371{
2372	return sock->ops->getname(sock, addr, addrlen, 0);
2373}
2374
2375int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2376			 int *addrlen)
2377{
2378	return sock->ops->getname(sock, addr, addrlen, 1);
2379}
2380
2381int kernel_getsockopt(struct socket *sock, int level, int optname,
2382			char *optval, int *optlen)
2383{
2384	mm_segment_t oldfs = get_fs();
2385	int err;
2386
2387	set_fs(KERNEL_DS);
2388	if (level == SOL_SOCKET)
2389		err = sock_getsockopt(sock, level, optname, optval, optlen);
2390	else
2391		err = sock->ops->getsockopt(sock, level, optname, optval,
2392					    optlen);
2393	set_fs(oldfs);
2394	return err;
2395}
2396
2397int kernel_setsockopt(struct socket *sock, int level, int optname,
2398			char *optval, unsigned int optlen)
2399{
2400	mm_segment_t oldfs = get_fs();
2401	int err;
2402
2403	set_fs(KERNEL_DS);
2404	if (level == SOL_SOCKET)
2405		err = sock_setsockopt(sock, level, optname, optval, optlen);
2406	else
2407		err = sock->ops->setsockopt(sock, level, optname, optval,
2408					    optlen);
2409	set_fs(oldfs);
2410	return err;
2411}
2412
2413int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2414		    size_t size, int flags)
2415{
2416	if (sock->ops->sendpage)
2417		return sock->ops->sendpage(sock, page, offset, size, flags);
2418
2419	return sock_no_sendpage(sock, page, offset, size, flags);
2420}
2421
2422int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2423{
2424	mm_segment_t oldfs = get_fs();
2425	int err;
2426
2427	set_fs(KERNEL_DS);
2428	err = sock->ops->ioctl(sock, cmd, arg);
2429	set_fs(oldfs);
2430
2431	return err;
2432}
2433
2434int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2435{
2436	return sock->ops->shutdown(sock, how);
2437}
2438
2439EXPORT_SYMBOL(sock_create);
2440EXPORT_SYMBOL(sock_create_kern);
2441EXPORT_SYMBOL(sock_create_lite);
2442EXPORT_SYMBOL(sock_map_fd);
2443EXPORT_SYMBOL(sock_recvmsg);
2444EXPORT_SYMBOL(sock_register);
2445EXPORT_SYMBOL(sock_release);
2446EXPORT_SYMBOL(sock_sendmsg);
2447EXPORT_SYMBOL(sock_unregister);
2448EXPORT_SYMBOL(sock_wake_async);
2449EXPORT_SYMBOL(sockfd_lookup);
2450EXPORT_SYMBOL(kernel_sendmsg);
2451EXPORT_SYMBOL(kernel_recvmsg);
2452EXPORT_SYMBOL(kernel_bind);
2453EXPORT_SYMBOL(kernel_listen);
2454EXPORT_SYMBOL(kernel_accept);
2455EXPORT_SYMBOL(kernel_connect);
2456EXPORT_SYMBOL(kernel_getsockname);
2457EXPORT_SYMBOL(kernel_getpeername);
2458EXPORT_SYMBOL(kernel_getsockopt);
2459EXPORT_SYMBOL(kernel_setsockopt);
2460EXPORT_SYMBOL(kernel_sendpage);
2461EXPORT_SYMBOL(kernel_sock_ioctl);
2462EXPORT_SYMBOL(kernel_sock_shutdown);