/net/socket.c

Large files files are truncated, but you can click here to view the full 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/if_bridge.h>
  73#include <linux/if_frad.h>
  74#include <linux/if_vlan.h>
  75#include <linux/ptp_classify.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#include <linux/xattr.h>
  92
  93#include <linux/uaccess.h>
  94#include <asm/unistd.h>
  95
  96#include <net/compat.h>
  97#include <net/wext.h>
  98#include <net/cls_cgroup.h>
  99
 100#include <net/sock.h>
 101#include <linux/netfilter.h>
 102
 103#include <linux/if_tun.h>
 104#include <linux/ipv6_route.h>
 105#include <linux/route.h>
 106#include <linux/sockios.h>
 107#include <linux/atalk.h>
 108#include <net/busy_poll.h>
 109#include <linux/errqueue.h>
 110
 111#ifdef CONFIG_NET_RX_BUSY_POLL
 112unsigned int sysctl_net_busy_read __read_mostly;
 113unsigned int sysctl_net_busy_poll __read_mostly;
 114#endif
 115
 116static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
 117static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
 118static int sock_mmap(struct file *file, struct vm_area_struct *vma);
 119
 120static int sock_close(struct inode *inode, struct file *file);
 121static unsigned int sock_poll(struct file *file,
 122			      struct poll_table_struct *wait);
 123static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 124#ifdef CONFIG_COMPAT
 125static long compat_sock_ioctl(struct file *file,
 126			      unsigned int cmd, unsigned long arg);
 127#endif
 128static int sock_fasync(int fd, struct file *filp, int on);
 129static ssize_t sock_sendpage(struct file *file, struct page *page,
 130			     int offset, size_t size, loff_t *ppos, int more);
 131static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 132				struct pipe_inode_info *pipe, size_t len,
 133				unsigned int flags);
 134
 135/*
 136 *	Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
 137 *	in the operation structures but are done directly via the socketcall() multiplexor.
 138 */
 139
 140static const struct file_operations socket_file_ops = {
 141	.owner =	THIS_MODULE,
 142	.llseek =	no_llseek,
 143	.read_iter =	sock_read_iter,
 144	.write_iter =	sock_write_iter,
 145	.poll =		sock_poll,
 146	.unlocked_ioctl = sock_ioctl,
 147#ifdef CONFIG_COMPAT
 148	.compat_ioctl = compat_sock_ioctl,
 149#endif
 150	.mmap =		sock_mmap,
 151	.release =	sock_close,
 152	.fasync =	sock_fasync,
 153	.sendpage =	sock_sendpage,
 154	.splice_write = generic_splice_sendpage,
 155	.splice_read =	sock_splice_read,
 156};
 157
 158/*
 159 *	The protocol list. Each protocol is registered in here.
 160 */
 161
 162static DEFINE_SPINLOCK(net_family_lock);
 163static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
 164
 165/*
 166 *	Statistics counters of the socket lists
 167 */
 168
 169static DEFINE_PER_CPU(int, sockets_in_use);
 170
 171/*
 172 * Support routines.
 173 * Move socket addresses back and forth across the kernel/user
 174 * divide and look after the messy bits.
 175 */
 176
 177/**
 178 *	move_addr_to_kernel	-	copy a socket address into kernel space
 179 *	@uaddr: Address in user space
 180 *	@kaddr: Address in kernel space
 181 *	@ulen: Length in user space
 182 *
 183 *	The address is copied into kernel space. If the provided address is
 184 *	too long an error code of -EINVAL is returned. If the copy gives
 185 *	invalid addresses -EFAULT is returned. On a success 0 is returned.
 186 */
 187
 188int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
 189{
 190	if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
 191		return -EINVAL;
 192	if (ulen == 0)
 193		return 0;
 194	if (copy_from_user(kaddr, uaddr, ulen))
 195		return -EFAULT;
 196	return audit_sockaddr(ulen, kaddr);
 197}
 198
 199/**
 200 *	move_addr_to_user	-	copy an address to user space
 201 *	@kaddr: kernel space address
 202 *	@klen: length of address in kernel
 203 *	@uaddr: user space address
 204 *	@ulen: pointer to user length field
 205 *
 206 *	The value pointed to by ulen on entry is the buffer length available.
 207 *	This is overwritten with the buffer space used. -EINVAL is returned
 208 *	if an overlong buffer is specified or a negative buffer size. -EFAULT
 209 *	is returned if either the buffer or the length field are not
 210 *	accessible.
 211 *	After copying the data up to the limit the user specifies, the true
 212 *	length of the data is written over the length limit the user
 213 *	specified. Zero is returned for a success.
 214 */
 215
 216static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
 217			     void __user *uaddr, int __user *ulen)
 218{
 219	int err;
 220	int len;
 221
 222	BUG_ON(klen > sizeof(struct sockaddr_storage));
 223	err = get_user(len, ulen);
 224	if (err)
 225		return err;
 226	if (len > klen)
 227		len = klen;
 228	if (len < 0)
 229		return -EINVAL;
 230	if (len) {
 231		if (audit_sockaddr(klen, kaddr))
 232			return -ENOMEM;
 233		if (copy_to_user(uaddr, kaddr, len))
 234			return -EFAULT;
 235	}
 236	/*
 237	 *      "fromlen shall refer to the value before truncation.."
 238	 *                      1003.1g
 239	 */
 240	return __put_user(klen, ulen);
 241}
 242
 243static struct kmem_cache *sock_inode_cachep __read_mostly;
 244
 245static struct inode *sock_alloc_inode(struct super_block *sb)
 246{
 247	struct socket_alloc *ei;
 248	struct socket_wq *wq;
 249
 250	ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
 251	if (!ei)
 252		return NULL;
 253	wq = kmalloc(sizeof(*wq), GFP_KERNEL);
 254	if (!wq) {
 255		kmem_cache_free(sock_inode_cachep, ei);
 256		return NULL;
 257	}
 258	init_waitqueue_head(&wq->wait);
 259	wq->fasync_list = NULL;
 260	wq->flags = 0;
 261	RCU_INIT_POINTER(ei->socket.wq, wq);
 262
 263	ei->socket.state = SS_UNCONNECTED;
 264	ei->socket.flags = 0;
 265	ei->socket.ops = NULL;
 266	ei->socket.sk = NULL;
 267	ei->socket.file = NULL;
 268
 269	return &ei->vfs_inode;
 270}
 271
 272static void sock_destroy_inode(struct inode *inode)
 273{
 274	struct socket_alloc *ei;
 275	struct socket_wq *wq;
 276
 277	ei = container_of(inode, struct socket_alloc, vfs_inode);
 278	wq = rcu_dereference_protected(ei->socket.wq, 1);
 279	kfree_rcu(wq, rcu);
 280	kmem_cache_free(sock_inode_cachep, ei);
 281}
 282
 283static void init_once(void *foo)
 284{
 285	struct socket_alloc *ei = (struct socket_alloc *)foo;
 286
 287	inode_init_once(&ei->vfs_inode);
 288}
 289
 290static void init_inodecache(void)
 291{
 292	sock_inode_cachep = kmem_cache_create("sock_inode_cache",
 293					      sizeof(struct socket_alloc),
 294					      0,
 295					      (SLAB_HWCACHE_ALIGN |
 296					       SLAB_RECLAIM_ACCOUNT |
 297					       SLAB_MEM_SPREAD | SLAB_ACCOUNT),
 298					      init_once);
 299	BUG_ON(sock_inode_cachep == NULL);
 300}
 301
 302static const struct super_operations sockfs_ops = {
 303	.alloc_inode	= sock_alloc_inode,
 304	.destroy_inode	= sock_destroy_inode,
 305	.statfs		= simple_statfs,
 306};
 307
 308/*
 309 * sockfs_dname() is called from d_path().
 310 */
 311static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
 312{
 313	return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
 314				d_inode(dentry)->i_ino);
 315}
 316
 317static const struct dentry_operations sockfs_dentry_operations = {
 318	.d_dname  = sockfs_dname,
 319};
 320
 321static int sockfs_xattr_get(const struct xattr_handler *handler,
 322			    struct dentry *dentry, struct inode *inode,
 323			    const char *suffix, void *value, size_t size)
 324{
 325	if (value) {
 326		if (dentry->d_name.len + 1 > size)
 327			return -ERANGE;
 328		memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
 329	}
 330	return dentry->d_name.len + 1;
 331}
 332
 333#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
 334#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
 335#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
 336
 337static const struct xattr_handler sockfs_xattr_handler = {
 338	.name = XATTR_NAME_SOCKPROTONAME,
 339	.get = sockfs_xattr_get,
 340};
 341
 342static int sockfs_security_xattr_set(const struct xattr_handler *handler,
 343				     struct dentry *dentry, struct inode *inode,
 344				     const char *suffix, const void *value,
 345				     size_t size, int flags)
 346{
 347	/* Handled by LSM. */
 348	return -EAGAIN;
 349}
 350
 351static const struct xattr_handler sockfs_security_xattr_handler = {
 352	.prefix = XATTR_SECURITY_PREFIX,
 353	.set = sockfs_security_xattr_set,
 354};
 355
 356static const struct xattr_handler *sockfs_xattr_handlers[] = {
 357	&sockfs_xattr_handler,
 358	&sockfs_security_xattr_handler,
 359	NULL
 360};
 361
 362static struct dentry *sockfs_mount(struct file_system_type *fs_type,
 363			 int flags, const char *dev_name, void *data)
 364{
 365	return mount_pseudo_xattr(fs_type, "socket:", &sockfs_ops,
 366				  sockfs_xattr_handlers,
 367				  &sockfs_dentry_operations, SOCKFS_MAGIC);
 368}
 369
 370static struct vfsmount *sock_mnt __read_mostly;
 371
 372static struct file_system_type sock_fs_type = {
 373	.name =		"sockfs",
 374	.mount =	sockfs_mount,
 375	.kill_sb =	kill_anon_super,
 376};
 377
 378/*
 379 *	Obtains the first available file descriptor and sets it up for use.
 380 *
 381 *	These functions create file structures and maps them to fd space
 382 *	of the current process. On success it returns file descriptor
 383 *	and file struct implicitly stored in sock->file.
 384 *	Note that another thread may close file descriptor before we return
 385 *	from this function. We use the fact that now we do not refer
 386 *	to socket after mapping. If one day we will need it, this
 387 *	function will increment ref. count on file by 1.
 388 *
 389 *	In any case returned fd MAY BE not valid!
 390 *	This race condition is unavoidable
 391 *	with shared fd spaces, we cannot solve it inside kernel,
 392 *	but we take care of internal coherence yet.
 393 */
 394
 395struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
 396{
 397	struct qstr name = { .name = "" };
 398	struct path path;
 399	struct file *file;
 400
 401	if (dname) {
 402		name.name = dname;
 403		name.len = strlen(name.name);
 404	} else if (sock->sk) {
 405		name.name = sock->sk->sk_prot_creator->name;
 406		name.len = strlen(name.name);
 407	}
 408	path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
 409	if (unlikely(!path.dentry))
 410		return ERR_PTR(-ENOMEM);
 411	path.mnt = mntget(sock_mnt);
 412
 413	d_instantiate(path.dentry, SOCK_INODE(sock));
 414
 415	file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
 416		  &socket_file_ops);
 417	if (IS_ERR(file)) {
 418		/* drop dentry, keep inode */
 419		ihold(d_inode(path.dentry));
 420		path_put(&path);
 421		return file;
 422	}
 423
 424	sock->file = file;
 425	file->f_flags = O_RDWR | (flags & O_NONBLOCK);
 426	file->private_data = sock;
 427	return file;
 428}
 429EXPORT_SYMBOL(sock_alloc_file);
 430
 431static int sock_map_fd(struct socket *sock, int flags)
 432{
 433	struct file *newfile;
 434	int fd = get_unused_fd_flags(flags);
 435	if (unlikely(fd < 0))
 436		return fd;
 437
 438	newfile = sock_alloc_file(sock, flags, NULL);
 439	if (likely(!IS_ERR(newfile))) {
 440		fd_install(fd, newfile);
 441		return fd;
 442	}
 443
 444	put_unused_fd(fd);
 445	return PTR_ERR(newfile);
 446}
 447
 448struct socket *sock_from_file(struct file *file, int *err)
 449{
 450	if (file->f_op == &socket_file_ops)
 451		return file->private_data;	/* set in sock_map_fd */
 452
 453	*err = -ENOTSOCK;
 454	return NULL;
 455}
 456EXPORT_SYMBOL(sock_from_file);
 457
 458/**
 459 *	sockfd_lookup - Go from a file number to its socket slot
 460 *	@fd: file handle
 461 *	@err: pointer to an error code return
 462 *
 463 *	The file handle passed in is locked and the socket it is bound
 464 *	to is returned. If an error occurs the err pointer is overwritten
 465 *	with a negative errno code and NULL is returned. The function checks
 466 *	for both invalid handles and passing a handle which is not a socket.
 467 *
 468 *	On a success the socket object pointer is returned.
 469 */
 470
 471struct socket *sockfd_lookup(int fd, int *err)
 472{
 473	struct file *file;
 474	struct socket *sock;
 475
 476	file = fget(fd);
 477	if (!file) {
 478		*err = -EBADF;
 479		return NULL;
 480	}
 481
 482	sock = sock_from_file(file, err);
 483	if (!sock)
 484		fput(file);
 485	return sock;
 486}
 487EXPORT_SYMBOL(sockfd_lookup);
 488
 489static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
 490{
 491	struct fd f = fdget(fd);
 492	struct socket *sock;
 493
 494	*err = -EBADF;
 495	if (f.file) {
 496		sock = sock_from_file(f.file, err);
 497		if (likely(sock)) {
 498			*fput_needed = f.flags;
 499			return sock;
 500		}
 501		fdput(f);
 502	}
 503	return NULL;
 504}
 505
 506static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
 507				size_t size)
 508{
 509	ssize_t len;
 510	ssize_t used = 0;
 511
 512	len = security_inode_listsecurity(d_inode(dentry), buffer, size);
 513	if (len < 0)
 514		return len;
 515	used += len;
 516	if (buffer) {
 517		if (size < used)
 518			return -ERANGE;
 519		buffer += len;
 520	}
 521
 522	len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
 523	used += len;
 524	if (buffer) {
 525		if (size < used)
 526			return -ERANGE;
 527		memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
 528		buffer += len;
 529	}
 530
 531	return used;
 532}
 533
 534static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
 535{
 536	int err = simple_setattr(dentry, iattr);
 537
 538	if (!err && (iattr->ia_valid & ATTR_UID)) {
 539		struct socket *sock = SOCKET_I(d_inode(dentry));
 540
 541		sock->sk->sk_uid = iattr->ia_uid;
 542	}
 543
 544	return err;
 545}
 546
 547static const struct inode_operations sockfs_inode_ops = {
 548	.listxattr = sockfs_listxattr,
 549	.setattr = sockfs_setattr,
 550};
 551
 552/**
 553 *	sock_alloc	-	allocate a socket
 554 *
 555 *	Allocate a new inode and socket object. The two are bound together
 556 *	and initialised. The socket is then returned. If we are out of inodes
 557 *	NULL is returned.
 558 */
 559
 560struct socket *sock_alloc(void)
 561{
 562	struct inode *inode;
 563	struct socket *sock;
 564
 565	inode = new_inode_pseudo(sock_mnt->mnt_sb);
 566	if (!inode)
 567		return NULL;
 568
 569	sock = SOCKET_I(inode);
 570
 571	kmemcheck_annotate_bitfield(sock, type);
 572	inode->i_ino = get_next_ino();
 573	inode->i_mode = S_IFSOCK | S_IRWXUGO;
 574	inode->i_uid = current_fsuid();
 575	inode->i_gid = current_fsgid();
 576	inode->i_op = &sockfs_inode_ops;
 577
 578	this_cpu_add(sockets_in_use, 1);
 579	return sock;
 580}
 581EXPORT_SYMBOL(sock_alloc);
 582
 583/**
 584 *	sock_release	-	close a socket
 585 *	@sock: socket to close
 586 *
 587 *	The socket is released from the protocol stack if it has a release
 588 *	callback, and the inode is then released if the socket is bound to
 589 *	an inode not a file.
 590 */
 591
 592void sock_release(struct socket *sock)
 593{
 594	if (sock->ops) {
 595		struct module *owner = sock->ops->owner;
 596
 597		sock->ops->release(sock);
 598		sock->ops = NULL;
 599		module_put(owner);
 600	}
 601
 602	if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
 603		pr_err("%s: fasync list not empty!\n", __func__);
 604
 605	this_cpu_sub(sockets_in_use, 1);
 606	if (!sock->file) {
 607		iput(SOCK_INODE(sock));
 608		return;
 609	}
 610	sock->file = NULL;
 611}
 612EXPORT_SYMBOL(sock_release);
 613
 614void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
 615{
 616	u8 flags = *tx_flags;
 617
 618	if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
 619		flags |= SKBTX_HW_TSTAMP;
 620
 621	if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
 622		flags |= SKBTX_SW_TSTAMP;
 623
 624	if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
 625		flags |= SKBTX_SCHED_TSTAMP;
 626
 627	*tx_flags = flags;
 628}
 629EXPORT_SYMBOL(__sock_tx_timestamp);
 630
 631static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
 632{
 633	int ret = sock->ops->sendmsg(sock, msg, msg_data_left(msg));
 634	BUG_ON(ret == -EIOCBQUEUED);
 635	return ret;
 636}
 637
 638int sock_sendmsg(struct socket *sock, struct msghdr *msg)
 639{
 640	int err = security_socket_sendmsg(sock, msg,
 641					  msg_data_left(msg));
 642
 643	return err ?: sock_sendmsg_nosec(sock, msg);
 644}
 645EXPORT_SYMBOL(sock_sendmsg);
 646
 647int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
 648		   struct kvec *vec, size_t num, size_t size)
 649{
 650	iov_iter_kvec(&msg->msg_iter, WRITE | ITER_KVEC, vec, num, size);
 651	return sock_sendmsg(sock, msg);
 652}
 653EXPORT_SYMBOL(kernel_sendmsg);
 654
 655static bool skb_is_err_queue(const struct sk_buff *skb)
 656{
 657	/* pkt_type of skbs enqueued on the error queue are set to
 658	 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
 659	 * in recvmsg, since skbs received on a local socket will never
 660	 * have a pkt_type of PACKET_OUTGOING.
 661	 */
 662	return skb->pkt_type == PACKET_OUTGOING;
 663}
 664
 665/* On transmit, software and hardware timestamps are returned independently.
 666 * As the two skb clones share the hardware timestamp, which may be updated
 667 * before the software timestamp is received, a hardware TX timestamp may be
 668 * returned only if there is no software TX timestamp. Ignore false software
 669 * timestamps, which may be made in the __sock_recv_timestamp() call when the
 670 * option SO_TIMESTAMP(NS) is enabled on the socket, even when the skb has a
 671 * hardware timestamp.
 672 */
 673static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
 674{
 675	return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
 676}
 677
 678static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
 679{
 680	struct scm_ts_pktinfo ts_pktinfo;
 681	struct net_device *orig_dev;
 682
 683	if (!skb_mac_header_was_set(skb))
 684		return;
 685
 686	memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
 687
 688	rcu_read_lock();
 689	orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
 690	if (orig_dev)
 691		ts_pktinfo.if_index = orig_dev->ifindex;
 692	rcu_read_unlock();
 693
 694	ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
 695	put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
 696		 sizeof(ts_pktinfo), &ts_pktinfo);
 697}
 698
 699/*
 700 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
 701 */
 702void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 703	struct sk_buff *skb)
 704{
 705	int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
 706	struct scm_timestamping tss;
 707	int empty = 1, false_tstamp = 0;
 708	struct skb_shared_hwtstamps *shhwtstamps =
 709		skb_hwtstamps(skb);
 710
 711	/* Race occurred between timestamp enabling and packet
 712	   receiving.  Fill in the current time for now. */
 713	if (need_software_tstamp && skb->tstamp == 0) {
 714		__net_timestamp(skb);
 715		false_tstamp = 1;
 716	}
 717
 718	if (need_software_tstamp) {
 719		if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
 720			struct timeval tv;
 721			skb_get_timestamp(skb, &tv);
 722			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 723				 sizeof(tv), &tv);
 724		} else {
 725			struct timespec ts;
 726			skb_get_timestampns(skb, &ts);
 727			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
 728				 sizeof(ts), &ts);
 729		}
 730	}
 731
 732	memset(&tss, 0, sizeof(tss));
 733	if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
 734	    ktime_to_timespec_cond(skb->tstamp, tss.ts + 0))
 735		empty = 0;
 736	if (shhwtstamps &&
 737	    (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 738	    !skb_is_swtx_tstamp(skb, false_tstamp) &&
 739	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
 740		empty = 0;
 741		if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
 742		    !skb_is_err_queue(skb))
 743			put_ts_pktinfo(msg, skb);
 744	}
 745	if (!empty) {
 746		put_cmsg(msg, SOL_SOCKET,
 747			 SCM_TIMESTAMPING, sizeof(tss), &tss);
 748
 749		if (skb_is_err_queue(skb) && skb->len &&
 750		    SKB_EXT_ERR(skb)->opt_stats)
 751			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
 752				 skb->len, skb->data);
 753	}
 754}
 755EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
 756
 757void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
 758	struct sk_buff *skb)
 759{
 760	int ack;
 761
 762	if (!sock_flag(sk, SOCK_WIFI_STATUS))
 763		return;
 764	if (!skb->wifi_acked_valid)
 765		return;
 766
 767	ack = skb->wifi_acked;
 768
 769	put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
 770}
 771EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
 772
 773static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
 774				   struct sk_buff *skb)
 775{
 776	if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
 777		put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
 778			sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
 779}
 780
 781void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
 782	struct sk_buff *skb)
 783{
 784	sock_recv_timestamp(msg, sk, skb);
 785	sock_recv_drops(msg, sk, skb);
 786}
 787EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
 788
 789static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
 790				     int flags)
 791{
 792	return sock->ops->recvmsg(sock, msg, msg_data_left(msg), flags);
 793}
 794
 795int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
 796{
 797	int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
 798
 799	return err ?: sock_recvmsg_nosec(sock, msg, flags);
 800}
 801EXPORT_SYMBOL(sock_recvmsg);
 802
 803/**
 804 * kernel_recvmsg - Receive a message from a socket (kernel space)
 805 * @sock:       The socket to receive the message from
 806 * @msg:        Received message
 807 * @vec:        Input s/g array for message data
 808 * @num:        Size of input s/g array
 809 * @size:       Number of bytes to read
 810 * @flags:      Message flags (MSG_DONTWAIT, etc...)
 811 *
 812 * On return the msg structure contains the scatter/gather array passed in the
 813 * vec argument. The array is modified so that it consists of the unfilled
 814 * portion of the original array.
 815 *
 816 * The returned value is the total number of bytes received, or an error.
 817 */
 818int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
 819		   struct kvec *vec, size_t num, size_t size, int flags)
 820{
 821	mm_segment_t oldfs = get_fs();
 822	int result;
 823
 824	iov_iter_kvec(&msg->msg_iter, READ | ITER_KVEC, vec, num, size);
 825	set_fs(KERNEL_DS);
 826	result = sock_recvmsg(sock, msg, flags);
 827	set_fs(oldfs);
 828	return result;
 829}
 830EXPORT_SYMBOL(kernel_recvmsg);
 831
 832static ssize_t sock_sendpage(struct file *file, struct page *page,
 833			     int offset, size_t size, loff_t *ppos, int more)
 834{
 835	struct socket *sock;
 836	int flags;
 837
 838	sock = file->private_data;
 839
 840	flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
 841	/* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
 842	flags |= more;
 843
 844	return kernel_sendpage(sock, page, offset, size, flags);
 845}
 846
 847static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
 848				struct pipe_inode_info *pipe, size_t len,
 849				unsigned int flags)
 850{
 851	struct socket *sock = file->private_data;
 852
 853	if (unlikely(!sock->ops->splice_read))
 854		return -EINVAL;
 855
 856	return sock->ops->splice_read(sock, ppos, pipe, len, flags);
 857}
 858
 859static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
 860{
 861	struct file *file = iocb->ki_filp;
 862	struct socket *sock = file->private_data;
 863	struct msghdr msg = {.msg_iter = *to,
 864			     .msg_iocb = iocb};
 865	ssize_t res;
 866
 867	if (file->f_flags & O_NONBLOCK)
 868		msg.msg_flags = MSG_DONTWAIT;
 869
 870	if (iocb->ki_pos != 0)
 871		return -ESPIPE;
 872
 873	if (!iov_iter_count(to))	/* Match SYS5 behaviour */
 874		return 0;
 875
 876	res = sock_recvmsg(sock, &msg, msg.msg_flags);
 877	*to = msg.msg_iter;
 878	return res;
 879}
 880
 881static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
 882{
 883	struct file *file = iocb->ki_filp;
 884	struct socket *sock = file->private_data;
 885	struct msghdr msg = {.msg_iter = *from,
 886			     .msg_iocb = iocb};
 887	ssize_t res;
 888
 889	if (iocb->ki_pos != 0)
 890		return -ESPIPE;
 891
 892	if (file->f_flags & O_NONBLOCK)
 893		msg.msg_flags = MSG_DONTWAIT;
 894
 895	if (sock->type == SOCK_SEQPACKET)
 896		msg.msg_flags |= MSG_EOR;
 897
 898	res = sock_sendmsg(sock, &msg);
 899	*from = msg.msg_iter;
 900	return res;
 901}
 902
 903/*
 904 * Atomic setting of ioctl hooks to avoid race
 905 * with module unload.
 906 */
 907
 908static DEFINE_MUTEX(br_ioctl_mutex);
 909static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
 910
 911void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
 912{
 913	mutex_lock(&br_ioctl_mutex);
 914	br_ioctl_hook = hook;
 915	mutex_unlock(&br_ioctl_mutex);
 916}
 917EXPORT_SYMBOL(brioctl_set);
 918
 919static DEFINE_MUTEX(vlan_ioctl_mutex);
 920static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
 921
 922void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
 923{
 924	mutex_lock(&vlan_ioctl_mutex);
 925	vlan_ioctl_hook = hook;
 926	mutex_unlock(&vlan_ioctl_mutex);
 927}
 928EXPORT_SYMBOL(vlan_ioctl_set);
 929
 930static DEFINE_MUTEX(dlci_ioctl_mutex);
 931static int (*dlci_ioctl_hook) (unsigned int, void __user *);
 932
 933void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
 934{
 935	mutex_lock(&dlci_ioctl_mutex);
 936	dlci_ioctl_hook = hook;
 937	mutex_unlock(&dlci_ioctl_mutex);
 938}
 939EXPORT_SYMBOL(dlci_ioctl_set);
 940
 941static long sock_do_ioctl(struct net *net, struct socket *sock,
 942				 unsigned int cmd, unsigned long arg)
 943{
 944	int err;
 945	void __user *argp = (void __user *)arg;
 946
 947	err = sock->ops->ioctl(sock, cmd, arg);
 948
 949	/*
 950	 * If this ioctl is unknown try to hand it down
 951	 * to the NIC driver.
 952	 */
 953	if (err == -ENOIOCTLCMD)
 954		err = dev_ioctl(net, cmd, argp);
 955
 956	return err;
 957}
 958
 959/*
 960 *	With an ioctl, arg may well be a user mode pointer, but we don't know
 961 *	what to do with it - that's up to the protocol still.
 962 */
 963
 964static struct ns_common *get_net_ns(struct ns_common *ns)
 965{
 966	return &get_net(container_of(ns, struct net, ns))->ns;
 967}
 968
 969static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 970{
 971	struct socket *sock;
 972	struct sock *sk;
 973	void __user *argp = (void __user *)arg;
 974	int pid, err;
 975	struct net *net;
 976
 977	sock = file->private_data;
 978	sk = sock->sk;
 979	net = sock_net(sk);
 980	if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
 981		err = dev_ioctl(net, cmd, argp);
 982	} else
 983#ifdef CONFIG_WEXT_CORE
 984	if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
 985		err = dev_ioctl(net, cmd, argp);
 986	} else
 987#endif
 988		switch (cmd) {
 989		case FIOSETOWN:
 990		case SIOCSPGRP:
 991			err = -EFAULT;
 992			if (get_user(pid, (int __user *)argp))
 993				break;
 994			err = f_setown(sock->file, pid, 1);
 995			break;
 996		case FIOGETOWN:
 997		case SIOCGPGRP:
 998			err = put_user(f_getown(sock->file),
 999				       (int __user *)argp);
1000			break;
1001		case SIOCGIFBR:
1002		case SIOCSIFBR:
1003		case SIOCBRADDBR:
1004		case SIOCBRDELBR:
1005			err = -ENOPKG;
1006			if (!br_ioctl_hook)
1007				request_module("bridge");
1008
1009			mutex_lock(&br_ioctl_mutex);
1010			if (br_ioctl_hook)
1011				err = br_ioctl_hook(net, cmd, argp);
1012			mutex_unlock(&br_ioctl_mutex);
1013			break;
1014		case SIOCGIFVLAN:
1015		case SIOCSIFVLAN:
1016			err = -ENOPKG;
1017			if (!vlan_ioctl_hook)
1018				request_module("8021q");
1019
1020			mutex_lock(&vlan_ioctl_mutex);
1021			if (vlan_ioctl_hook)
1022				err = vlan_ioctl_hook(net, argp);
1023			mutex_unlock(&vlan_ioctl_mutex);
1024			break;
1025		case SIOCADDDLCI:
1026		case SIOCDELDLCI:
1027			err = -ENOPKG;
1028			if (!dlci_ioctl_hook)
1029				request_module("dlci");
1030
1031			mutex_lock(&dlci_ioctl_mutex);
1032			if (dlci_ioctl_hook)
1033				err = dlci_ioctl_hook(cmd, argp);
1034			mutex_unlock(&dlci_ioctl_mutex);
1035			break;
1036		case SIOCGSKNS:
1037			err = -EPERM;
1038			if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1039				break;
1040
1041			err = open_related_ns(&net->ns, get_net_ns);
1042			break;
1043		default:
1044			err = sock_do_ioctl(net, sock, cmd, arg);
1045			break;
1046		}
1047	return err;
1048}
1049
1050int sock_create_lite(int family, int type, int protocol, struct socket **res)
1051{
1052	int err;
1053	struct socket *sock = NULL;
1054
1055	err = security_socket_create(family, type, protocol, 1);
1056	if (err)
1057		goto out;
1058
1059	sock = sock_alloc();
1060	if (!sock) {
1061		err = -ENOMEM;
1062		goto out;
1063	}
1064
1065	sock->type = type;
1066	err = security_socket_post_create(sock, family, type, protocol, 1);
1067	if (err)
1068		goto out_release;
1069
1070out:
1071	*res = sock;
1072	return err;
1073out_release:
1074	sock_release(sock);
1075	sock = NULL;
1076	goto out;
1077}
1078EXPORT_SYMBOL(sock_create_lite);
1079
1080/* No kernel lock held - perfect */
1081static unsigned int sock_poll(struct file *file, poll_table *wait)
1082{
1083	unsigned int busy_flag = 0;
1084	struct socket *sock;
1085
1086	/*
1087	 *      We can't return errors to poll, so it's either yes or no.
1088	 */
1089	sock = file->private_data;
1090
1091	if (sk_can_busy_loop(sock->sk)) {
1092		/* this socket can poll_ll so tell the system call */
1093		busy_flag = POLL_BUSY_LOOP;
1094
1095		/* once, only if requested by syscall */
1096		if (wait && (wait->_key & POLL_BUSY_LOOP))
1097			sk_busy_loop(sock->sk, 1);
1098	}
1099
1100	return busy_flag | sock->ops->poll(file, sock, wait);
1101}
1102
1103static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1104{
1105	struct socket *sock = file->private_data;
1106
1107	return sock->ops->mmap(file, sock, vma);
1108}
1109
1110static int sock_close(struct inode *inode, struct file *filp)
1111{
1112	sock_release(SOCKET_I(inode));
1113	return 0;
1114}
1115
1116/*
1117 *	Update the socket async list
1118 *
1119 *	Fasync_list locking strategy.
1120 *
1121 *	1. fasync_list is modified only under process context socket lock
1122 *	   i.e. under semaphore.
1123 *	2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1124 *	   or under socket lock
1125 */
1126
1127static int sock_fasync(int fd, struct file *filp, int on)
1128{
1129	struct socket *sock = filp->private_data;
1130	struct sock *sk = sock->sk;
1131	struct socket_wq *wq;
1132
1133	if (sk == NULL)
1134		return -EINVAL;
1135
1136	lock_sock(sk);
1137	wq = rcu_dereference_protected(sock->wq, lockdep_sock_is_held(sk));
1138	fasync_helper(fd, filp, on, &wq->fasync_list);
1139
1140	if (!wq->fasync_list)
1141		sock_reset_flag(sk, SOCK_FASYNC);
1142	else
1143		sock_set_flag(sk, SOCK_FASYNC);
1144
1145	release_sock(sk);
1146	return 0;
1147}
1148
1149/* This function may be called only under rcu_lock */
1150
1151int sock_wake_async(struct socket_wq *wq, int how, int band)
1152{
1153	if (!wq || !wq->fasync_list)
1154		return -1;
1155
1156	switch (how) {
1157	case SOCK_WAKE_WAITD:
1158		if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1159			break;
1160		goto call_kill;
1161	case SOCK_WAKE_SPACE:
1162		if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1163			break;
1164		/* fall through */
1165	case SOCK_WAKE_IO:
1166call_kill:
1167		kill_fasync(&wq->fasync_list, SIGIO, band);
1168		break;
1169	case SOCK_WAKE_URG:
1170		kill_fasync(&wq->fasync_list, SIGURG, band);
1171	}
1172
1173	return 0;
1174}
1175EXPORT_SYMBOL(sock_wake_async);
1176
1177int __sock_create(struct net *net, int family, int type, int protocol,
1178			 struct socket **res, int kern)
1179{
1180	int err;
1181	struct socket *sock;
1182	const struct net_proto_family *pf;
1183
1184	/*
1185	 *      Check protocol is in range
1186	 */
1187	if (family < 0 || family >= NPROTO)
1188		return -EAFNOSUPPORT;
1189	if (type < 0 || type >= SOCK_MAX)
1190		return -EINVAL;
1191
1192	/* Compatibility.
1193
1194	   This uglymoron is moved from INET layer to here to avoid
1195	   deadlock in module load.
1196	 */
1197	if (family == PF_INET && type == SOCK_PACKET) {
1198		pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1199			     current->comm);
1200		family = PF_PACKET;
1201	}
1202
1203	err = security_socket_create(family, type, protocol, kern);
1204	if (err)
1205		return err;
1206
1207	/*
1208	 *	Allocate the socket and allow the family to set things up. if
1209	 *	the protocol is 0, the family is instructed to select an appropriate
1210	 *	default.
1211	 */
1212	sock = sock_alloc();
1213	if (!sock) {
1214		net_warn_ratelimited("socket: no more sockets\n");
1215		return -ENFILE;	/* Not exactly a match, but its the
1216				   closest posix thing */
1217	}
1218
1219	sock->type = type;
1220
1221#ifdef CONFIG_MODULES
1222	/* Attempt to load a protocol module if the find failed.
1223	 *
1224	 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1225	 * requested real, full-featured networking support upon configuration.
1226	 * Otherwise module support will break!
1227	 */
1228	if (rcu_access_pointer(net_families[family]) == NULL)
1229		request_module("net-pf-%d", family);
1230#endif
1231
1232	rcu_read_lock();
1233	pf = rcu_dereference(net_families[family]);
1234	err = -EAFNOSUPPORT;
1235	if (!pf)
1236		goto out_release;
1237
1238	/*
1239	 * We will call the ->create function, that possibly is in a loadable
1240	 * module, so we have to bump that loadable module refcnt first.
1241	 */
1242	if (!try_module_get(pf->owner))
1243		goto out_release;
1244
1245	/* Now protected by module ref count */
1246	rcu_read_unlock();
1247
1248	err = pf->create(net, sock, protocol, kern);
1249	if (err < 0)
1250		goto out_module_put;
1251
1252	/*
1253	 * Now to bump the refcnt of the [loadable] module that owns this
1254	 * socket at sock_release time we decrement its refcnt.
1255	 */
1256	if (!try_module_get(sock->ops->owner))
1257		goto out_module_busy;
1258
1259	/*
1260	 * Now that we're done with the ->create function, the [loadable]
1261	 * module can have its refcnt decremented
1262	 */
1263	module_put(pf->owner);
1264	err = security_socket_post_create(sock, family, type, protocol, kern);
1265	if (err)
1266		goto out_sock_release;
1267	*res = sock;
1268
1269	return 0;
1270
1271out_module_busy:
1272	err = -EAFNOSUPPORT;
1273out_module_put:
1274	sock->ops = NULL;
1275	module_put(pf->owner);
1276out_sock_release:
1277	sock_release(sock);
1278	return err;
1279
1280out_release:
1281	rcu_read_unlock();
1282	goto out_sock_release;
1283}
1284EXPORT_SYMBOL(__sock_create);
1285
1286int sock_create(int family, int type, int protocol, struct socket **res)
1287{
1288	return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1289}
1290EXPORT_SYMBOL(sock_create);
1291
1292int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1293{
1294	return __sock_create(net, family, type, protocol, res, 1);
1295}
1296EXPORT_SYMBOL(sock_create_kern);
1297
1298SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1299{
1300	int retval;
1301	struct socket *sock;
1302	int flags;
1303
1304	/* Check the SOCK_* constants for consistency.  */
1305	BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1306	BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1307	BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1308	BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1309
1310	flags = type & ~SOCK_TYPE_MASK;
1311	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1312		return -EINVAL;
1313	type &= SOCK_TYPE_MASK;
1314
1315	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1316		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1317
1318	retval = sock_create(family, type, protocol, &sock);
1319	if (retval < 0)
1320		goto out;
1321
1322	retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1323	if (retval < 0)
1324		goto out_release;
1325
1326out:
1327	/* It may be already another descriptor 8) Not kernel problem. */
1328	return retval;
1329
1330out_release:
1331	sock_release(sock);
1332	return retval;
1333}
1334
1335/*
1336 *	Create a pair of connected sockets.
1337 */
1338
1339SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1340		int __user *, usockvec)
1341{
1342	struct socket *sock1, *sock2;
1343	int fd1, fd2, err;
1344	struct file *newfile1, *newfile2;
1345	int flags;
1346
1347	flags = type & ~SOCK_TYPE_MASK;
1348	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1349		return -EINVAL;
1350	type &= SOCK_TYPE_MASK;
1351
1352	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1353		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1354
1355	/*
1356	 * Obtain the first socket and check if the underlying protocol
1357	 * supports the socketpair call.
1358	 */
1359
1360	err = sock_create(family, type, protocol, &sock1);
1361	if (err < 0)
1362		goto out;
1363
1364	err = sock_create(family, type, protocol, &sock2);
1365	if (err < 0)
1366		goto out_release_1;
1367
1368	err = sock1->ops->socketpair(sock1, sock2);
1369	if (err < 0)
1370		goto out_release_both;
1371
1372	fd1 = get_unused_fd_flags(flags);
1373	if (unlikely(fd1 < 0)) {
1374		err = fd1;
1375		goto out_release_both;
1376	}
1377
1378	fd2 = get_unused_fd_flags(flags);
1379	if (unlikely(fd2 < 0)) {
1380		err = fd2;
1381		goto out_put_unused_1;
1382	}
1383
1384	newfile1 = sock_alloc_file(sock1, flags, NULL);
1385	if (IS_ERR(newfile1)) {
1386		err = PTR_ERR(newfile1);
1387		goto out_put_unused_both;
1388	}
1389
1390	newfile2 = sock_alloc_file(sock2, flags, NULL);
1391	if (IS_ERR(newfile2)) {
1392		err = PTR_ERR(newfile2);
1393		goto out_fput_1;
1394	}
1395
1396	err = put_user(fd1, &usockvec[0]);
1397	if (err)
1398		goto out_fput_both;
1399
1400	err = put_user(fd2, &usockvec[1]);
1401	if (err)
1402		goto out_fput_both;
1403
1404	audit_fd_pair(fd1, fd2);
1405
1406	fd_install(fd1, newfile1);
1407	fd_install(fd2, newfile2);
1408	/* fd1 and fd2 may be already another descriptors.
1409	 * Not kernel problem.
1410	 */
1411
1412	return 0;
1413
1414out_fput_both:
1415	fput(newfile2);
1416	fput(newfile1);
1417	put_unused_fd(fd2);
1418	put_unused_fd(fd1);
1419	goto out;
1420
1421out_fput_1:
1422	fput(newfile1);
1423	put_unused_fd(fd2);
1424	put_unused_fd(fd1);
1425	sock_release(sock2);
1426	goto out;
1427
1428out_put_unused_both:
1429	put_unused_fd(fd2);
1430out_put_unused_1:
1431	put_unused_fd(fd1);
1432out_release_both:
1433	sock_release(sock2);
1434out_release_1:
1435	sock_release(sock1);
1436out:
1437	return err;
1438}
1439
1440/*
1441 *	Bind a name to a socket. Nothing much to do here since it's
1442 *	the protocol's responsibility to handle the local address.
1443 *
1444 *	We move the socket address to kernel space before we call
1445 *	the protocol layer (having also checked the address is ok).
1446 */
1447
1448SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1449{
1450	struct socket *sock;
1451	struct sockaddr_storage address;
1452	int err, fput_needed;
1453
1454	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1455	if (sock) {
1456		err = move_addr_to_kernel(umyaddr, addrlen, &address);
1457		if (err >= 0) {
1458			err = security_socket_bind(sock,
1459						   (struct sockaddr *)&address,
1460						   addrlen);
1461			if (!err)
1462				err = sock->ops->bind(sock,
1463						      (struct sockaddr *)
1464						      &address, addrlen);
1465		}
1466		fput_light(sock->file, fput_needed);
1467	}
1468	return err;
1469}
1470
1471/*
1472 *	Perform a listen. Basically, we allow the protocol to do anything
1473 *	necessary for a listen, and if that works, we mark the socket as
1474 *	ready for listening.
1475 */
1476
1477SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1478{
1479	struct socket *sock;
1480	int err, fput_needed;
1481	int somaxconn;
1482
1483	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1484	if (sock) {
1485		somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1486		if ((unsigned int)backlog > somaxconn)
1487			backlog = somaxconn;
1488
1489		err = security_socket_listen(sock, backlog);
1490		if (!err)
1491			err = sock->ops->listen(sock, backlog);
1492
1493		fput_light(sock->file, fput_needed);
1494	}
1495	return err;
1496}
1497
1498/*
1499 *	For accept, we attempt to create a new socket, set up the link
1500 *	with the client, wake up the client, then return the new
1501 *	connected fd. We collect the address of the connector in kernel
1502 *	space and move it to user at the very end. This is unclean because
1503 *	we open the socket then return an error.
1504 *
1505 *	1003.1g adds the ability to recvmsg() to query connection pending
1506 *	status to recvmsg. We need to add that support in a way thats
1507 *	clean when we restucture accept also.
1508 */
1509
1510SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1511		int __user *, upeer_addrlen, int, flags)
1512{
1513	struct socket *sock, *newsock;
1514	struct file *newfile;
1515	int err, len, newfd, fput_needed;
1516	struct sockaddr_storage address;
1517
1518	if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1519		return -EINVAL;
1520
1521	if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1522		flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1523
1524	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1525	if (!sock)
1526		goto out;
1527
1528	err = -ENFILE;
1529	newsock = sock_alloc();
1530	if (!newsock)
1531		goto out_put;
1532
1533	newsock->type = sock->type;
1534	newsock->ops = sock->ops;
1535
1536	/*
1537	 * We don't need try_module_get here, as the listening socket (sock)
1538	 * has the protocol module (sock->ops->owner) held.
1539	 */
1540	__module_get(newsock->ops->owner);
1541
1542	newfd = get_unused_fd_flags(flags);
1543	if (unlikely(newfd < 0)) {
1544		err = newfd;
1545		sock_release(newsock);
1546		goto out_put;
1547	}
1548	newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1549	if (IS_ERR(newfile)) {
1550		err = PTR_ERR(newfile);
1551		put_unused_fd(newfd);
1552		sock_release(newsock);
1553		goto out_put;
1554	}
1555
1556	err = security_socket_accept(sock, newsock);
1557	if (err)
1558		goto out_fd;
1559
1560	err = sock->ops->accept(sock, newsock, sock->file->f_flags, false);
1561	if (err < 0)
1562		goto out_fd;
1563
1564	if (upeer_sockaddr) {
1565		if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1566					  &len, 2) < 0) {
1567			err = -ECONNABORTED;
1568			goto out_fd;
1569		}
1570		err = move_addr_to_user(&address,
1571					len, upeer_sockaddr, upeer_addrlen);
1572		if (err < 0)
1573			goto out_fd;
1574	}
1575
1576	/* File flags are not inherited via accept() unlike another OSes. */
1577
1578	fd_install(newfd, newfile);
1579	err = newfd;
1580
1581out_put:
1582	fput_light(sock->file, fput_needed);
1583out:
1584	return err;
1585out_fd:
1586	fput(newfile);
1587	put_unused_fd(newfd);
1588	goto out_put;
1589}
1590
1591SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1592		int __user *, upeer_addrlen)
1593{
1594	return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1595}
1596
1597/*
1598 *	Attempt to connect to a socket with the server address.  The address
1599 *	is in user space so we verify it is OK and move it to kernel space.
1600 *
1601 *	For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1602 *	break bindings
1603 *
1604 *	NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1605 *	other SEQPACKET protocols that take time to connect() as it doesn't
1606 *	include the -EINPROGRESS status for such sockets.
1607 */
1608
1609SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1610		int, addrlen)
1611{
1612	struct socket *sock;
1613	struct sockaddr_storage address;
1614	int err, fput_needed;
1615
1616	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1617	if (!sock)
1618		goto out;
1619	err = move_addr_to_kernel(uservaddr, addrlen, &address);
1620	if (err < 0)
1621		goto out_put;
1622
1623	err =
1624	    security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1625	if (err)
1626		goto out_put;
1627
1628	err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1629				 sock->file->f_flags);
1630out_put:
1631	fput_light(sock->file, fput_needed);
1632out:
1633	return err;
1634}
1635
1636/*
1637 *	Get the local address ('name') of a socket object. Move the obtained
1638 *	name to user space.
1639 */
1640
1641SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1642		int __user *, usockaddr_len)
1643{
1644	struct socket *sock;
1645	struct sockaddr_storage address;
1646	int len, err, fput_needed;
1647
1648	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1649	if (!sock)
1650		goto out;
1651
1652	err = security_socket_getsockname(sock);
1653	if (err)
1654		goto out_put;
1655
1656	err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1657	if (err)
1658		goto out_put;
1659	err = move_addr_to_user(&address, len, usockaddr, usockaddr_len);
1660
1661out_put:
1662	fput_light(sock->file, fput_needed);
1663out:
1664	return err;
1665}
1666
1667/*
1668 *	Get the remote address ('name') of a socket object. Move the obtained
1669 *	name to user space.
1670 */
1671
1672SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1673		int __user *, usockaddr_len)
1674{
1675	struct socket *sock;
1676	struct sockaddr_storage address;
1677	int len, err, fput_needed;
1678
1679	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1680	if (sock != NULL) {
1681		err = security_socket_getpeername(sock);
1682		if (err) {
1683			fput_light(sock->file, fput_needed);
1684			return err;
1685		}
1686
1687		err =
1688		    sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1689				       1);
1690		if (!err)
1691			err = move_addr_to_user(&address, len, usockaddr,
1692						usockaddr_len);
1693		fput_light(sock->file, fput_needed);
1694	}
1695	return err;
1696}
1697
1698/*
1699 *	Send a datagram to a given address. We move the address into kernel
1700 *	space and check the user space data area is readable before invoking
1701 *	the protocol.
1702 */
1703
1704SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1705		unsigned int, flags, struct sockaddr __user *, addr,
1706		int, addr_len)
1707{
1708	struct socket *sock;
1709	struct sockaddr_storage address;
1710	int err;
1711	struct msghdr msg;
1712	struct iovec iov;
1713	int fput_needed;
1714
1715	err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1716	if (unlikely(err))
1717		return err;
1718	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1719	if (!sock)
1720		goto out;
1721
1722	msg.msg_name = NULL;
1723	msg.msg_control = NULL;
1724	msg.msg_controllen = 0;
1725	msg.msg_namelen = 0;
1726	if (addr) {
1727		err = move_addr_to_kernel(addr, addr_len, &address);
1728		if (err < 0)
1729			goto out_put;
1730		msg.msg_name = (struct sockaddr *)&address;
1731		msg.msg_namelen = addr_len;
1732	}
1733	if (sock->file->f_flags & O_NONBLOCK)
1734		flags |= MSG_DONTWAIT;
1735	msg.msg_flags = flags;
1736	err = sock_sendmsg(sock, &msg);
1737
1738out_put:
1739	fput_light(sock->file, fput_needed);
1740out:
1741	return err;
1742}
1743
1744/*
1745 *	Send a datagram down a socket.
1746 */
1747
1748SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1749		unsigned int, flags)
1750{
1751	return sys_sendto(fd, buff, len, flags, NULL, 0);
1752}
1753
1754/*
1755 *	Receive a frame from the socket and optionally record the address of the
1756 *	sender. We verify the buffers are writable and if needed move the
1757 *	sender address from kernel to user space.
1758 */
1759
1760SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1761		unsigned int, flags, struct sockaddr __user *, addr,
1762		int __user *, addr_len)
1763{
1764	struct socket *sock;
1765	struct iovec iov;
1766	struct msghdr msg;
1767	struct sockaddr_storage address;
1768	int err, err2;
1769	int fput_needed;
1770
1771	err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1772	if (unlikely(err))
1773		return err;
1774	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1775	if (!sock)
1776		goto out;
1777
1778	msg.msg_control = NULL;
1779	msg.msg_controllen = 0;
1780	/* Save some cycles and don't copy the address if not needed */
1781	msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
1782	/* We assume all kernel code knows the size of sockaddr_storage */
1783	msg.msg_namelen = 0;
1784	msg.msg_iocb = NULL;
1785	msg.msg_flags = 0;
1786	if (sock->file->f_flags & O_NONBLOCK)
1787		flags |= MSG_DONTWAIT;
1788	err = sock_recvmsg(sock, &msg, flags);
1789
1790	if (err >= 0 && addr != NULL) {
1791		err2 = move_addr_to_user(&address,
1792					 msg.msg_namelen, addr, addr_len);
1793		if (err2 < 0)
1794			err = err2;
1795	}
1796
1797	fput_light(sock->file, fput_needed);
1798out:
1799	return err;
1800}
1801
1802/*
1803 *	Receive a datagram from a socket.
1804 */
1805
1806SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
1807		unsigned int, flags)
1808{
1809	return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1810}
1811
1812/*
1813 *	Set a socket option. Because we don't know the option lengths we have
1814 *	to pass the user mode parameter for the protocols to sort out.
1815 */
1816
1817SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1818		char __user *, optval, int, optlen)
1819{
1820	int err, fput_needed;
1821	struct socket *sock;
1822
1823	if (optlen < 0)
1824		return -EINVAL;
1825
1826	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1827	if (sock != NULL) {
1828		err = security_socket_setsockopt(sock, level, optname);
1829		if (err)
1830			goto out_put;
1831
1832		if (level == SOL_SOCKET)
1833			err =
1834			    sock_setsockopt(sock, level, optname, optval,
1835					    optlen);
1836		else
1837			err =
1838			    sock->ops->setsockopt(sock, level, optname, optval,
1839						  optlen);
1840out_put:
1841		fput_light(sock->file, fput_needed);
1842	}
1843	return err;
1844}
1845
1846/*
1847 *	Get a socket option. Because we don't know the option lengths we have
1848 *	to pass a user mode parameter for the protocols to sort out.
1849 */
1850
1851SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1852		char __user *, optval, int __user *, optlen)
1853{
1854	int err, fput_needed;
1855	struct socket *sock;
1856
1857	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1858	if (sock != NULL) {
1859		err = security_socket_getsockopt(sock, level, optname);
1860		if (err)
1861			goto out_put;
1862
1863		if (level == SOL_SOCKET)
1864			err =
1865			    sock_getsockopt(sock, level, optname, optval,
1866					    optlen);
1867		else
1868			err =
1869			    sock->ops->getsockopt(sock, level, optname, optval,
1870						  optlen);
1871out_put:
1872		fput_light(sock->file, fput_needed);
1873	}
1874	return err;
1875}
1876
1877/*
1878 *	Shutdown a socket.
1879 */
1880
1881SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1882{
1883	int err, fput_needed;
1884	struct socket *sock;
1885
1886	sock = sockfd_lookup_light(fd, &err, &fput_needed);
1887	if (sock != NULL) {
1888		err = security_socket_shutdown(sock, how);
1889		if (!err)
1890			err = sock->ops->shutdown(sock, how);
1891		fput_light(sock->file, fput_needed);
1892	}
1893	return err;
1894}
1895
1896/* A couple of helpful macros for getting the address of the 32/64 bit
1897 * fields which are the same type (int / unsigned) on our platforms.
1898 */
1899#define COMPAT_MSG(msg, member)	((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1900#define COMPAT_NAMELEN(msg)	COMPAT_MSG(msg, msg_namelen)
1901#define COMPAT_FLAGS(msg)	COMPAT_MSG(msg, msg_flags)
1902
1903struct used_address {
1904	struct sockaddr_storage name;
1905	unsigned int name_len;
1906};
1907
1908static int copy_msghdr_from_user(struct msghdr *kmsg,
1909				 struct user_msghdr __user *umsg,
1910				 struct sockaddr __user **save_addr,
1911				 struct iovec **iov)
1912{
1913	struct sockaddr __user *uaddr;
1914	struct iovec __user *uiov;
1915	size_t nr_segs;
1916	ssize_t err;
1917
1918	if (!access_ok(VERIFY_READ, umsg, sizeof(*umsg)) ||
1919	    __get_user(uaddr, &umsg->msg_name) ||
1920	    __get_user(kmsg->msg_namelen, &umsg->msg_namelen) ||
1921	    __get_user(uiov, &umsg->msg_iov) ||
1922	    __get_user(nr_segs, &umsg->msg_iovlen) ||
1923	    __get_user(kmsg->msg_control, &umsg->msg_control) ||
1924	    __get_user(kmsg->msg_controllen, &umsg->msg_controllen) ||
1925	    __get_user(kmsg->msg_flags, &umsg->msg_flags))
1926		return -EFAULT;
1927
1928	if (!uaddr)
1929		kmsg->msg_namelen = 0;
1930
1931	if (kmsg->msg_namelen < 0)
1932		return -EINVAL;
1933
1934	if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
1935		kmsg->msg_namelen = sizeof(struct sockaddr_storage);
1936
1937	if (save_addr)
1938		*save_addr = uaddr;
1939
1940	if (uaddr && kmsg->msg_namelen) {
1941		if (!save_addr) {
1942			err = move_addr_to_kernel(uaddr, kmsg->msg_namelen,
1943						  kmsg->msg_name);
1944			if (err < 0)
1945				return err;
1946		}
1947	} else {
1948		kmsg->msg_name = NULL;
1949		kmsg->msg_namelen = 0;
1950	}
1951
1952	if (nr_segs > UIO_MAXIOV)
1953		return -EMSGSIZE;
1954
1955	kmsg->msg_iocb = NULL;
1956
1957	return import_iovec(save_addr ? READ : WRITE, uiov, nr_segs,
1958			    UIO_FASTIOV, iov, &kmsg->msg_iter);
1959}
1960
1961static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
1962			 struct msghdr *msg_sys, unsigned int flags,
1963			 struct used_address *used_address,
1964			 unsigned int allowed_msghdr_flags)
1965{
1966	struct compat_msghdr __user *msg_compat =
1967	    (struct compat_msghdr __user *)msg;
1968	struct sockaddr_storage address;
1969	struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1970	unsigned char ctl[sizeof(struct cmsghdr) + 20]
1971				__aligned(sizeof(__kernel_size_t));
1972	/* 20 is size of ipv6_pktinfo */
1973	unsigned char *ctl_buf = ctl;
1974	int ctl_len;
1975	ssize_t err;
1976
1977	msg_sys->msg_name = &address;
1978
1979	if (MSG_CMSG_COMPAT & flags)
1980		err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
1981	else
1982		err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
1983	if (err < 0)
1984		return err;
1985
1986	err = -ENOBUFS;
1987
1988	if (msg_sys->msg_controllen > INT_MAX)
1989		goto out_freeiov;
1990	flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
1991	ctl_len = msg_sys->msg_controllen;
1992	if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1993		err =…