Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET An implementation of the SOCKET network access protocol.
4 *
5 * Version: @(#)socket.c 1.1.93 18/02/95
6 *
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
8 * Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * shutdown()
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
18 * top level.
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
23 * tty drivers).
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
26 * configurable.
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
35 * stuff.
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
41 * moment.
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
48 *
49 * This module is effectively the top level interface to the BSD socket
50 * paradigm.
51 *
52 * Based upon Swansea University Computer Society NET3.039
53 */
54
55#include <linux/ethtool.h>
56#include <linux/mm.h>
57#include <linux/socket.h>
58#include <linux/file.h>
59#include <linux/net.h>
60#include <linux/interrupt.h>
61#include <linux/thread_info.h>
62#include <linux/rcupdate.h>
63#include <linux/netdevice.h>
64#include <linux/proc_fs.h>
65#include <linux/seq_file.h>
66#include <linux/mutex.h>
67#include <linux/if_bridge.h>
68#include <linux/if_vlan.h>
69#include <linux/ptp_classify.h>
70#include <linux/init.h>
71#include <linux/poll.h>
72#include <linux/cache.h>
73#include <linux/module.h>
74#include <linux/highmem.h>
75#include <linux/mount.h>
76#include <linux/pseudo_fs.h>
77#include <linux/security.h>
78#include <linux/syscalls.h>
79#include <linux/compat.h>
80#include <linux/kmod.h>
81#include <linux/audit.h>
82#include <linux/wireless.h>
83#include <linux/nsproxy.h>
84#include <linux/magic.h>
85#include <linux/slab.h>
86#include <linux/xattr.h>
87#include <linux/nospec.h>
88#include <linux/indirect_call_wrapper.h>
89
90#include <linux/uaccess.h>
91#include <asm/unistd.h>
92
93#include <net/compat.h>
94#include <net/wext.h>
95#include <net/cls_cgroup.h>
96
97#include <net/sock.h>
98#include <linux/netfilter.h>
99
100#include <linux/if_tun.h>
101#include <linux/ipv6_route.h>
102#include <linux/route.h>
103#include <linux/termios.h>
104#include <linux/sockios.h>
105#include <net/busy_poll.h>
106#include <linux/errqueue.h>
107#include <linux/ptp_clock_kernel.h>
108
109#ifdef CONFIG_NET_RX_BUSY_POLL
110unsigned int sysctl_net_busy_read __read_mostly;
111unsigned int sysctl_net_busy_poll __read_mostly;
112#endif
113
114static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
115static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
116static int sock_mmap(struct file *file, struct vm_area_struct *vma);
117
118static int sock_close(struct inode *inode, struct file *file);
119static __poll_t sock_poll(struct file *file,
120 struct poll_table_struct *wait);
121static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
122#ifdef CONFIG_COMPAT
123static long compat_sock_ioctl(struct file *file,
124 unsigned int cmd, unsigned long arg);
125#endif
126static int sock_fasync(int fd, struct file *filp, int on);
127static ssize_t sock_sendpage(struct file *file, struct page *page,
128 int offset, size_t size, loff_t *ppos, int more);
129static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
130 struct pipe_inode_info *pipe, size_t len,
131 unsigned int flags);
132
133#ifdef CONFIG_PROC_FS
134static void sock_show_fdinfo(struct seq_file *m, struct file *f)
135{
136 struct socket *sock = f->private_data;
137
138 if (sock->ops->show_fdinfo)
139 sock->ops->show_fdinfo(m, sock);
140}
141#else
142#define sock_show_fdinfo NULL
143#endif
144
145/*
146 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
147 * in the operation structures but are done directly via the socketcall() multiplexor.
148 */
149
150static const struct file_operations socket_file_ops = {
151 .owner = THIS_MODULE,
152 .llseek = no_llseek,
153 .read_iter = sock_read_iter,
154 .write_iter = sock_write_iter,
155 .poll = sock_poll,
156 .unlocked_ioctl = sock_ioctl,
157#ifdef CONFIG_COMPAT
158 .compat_ioctl = compat_sock_ioctl,
159#endif
160 .mmap = sock_mmap,
161 .release = sock_close,
162 .fasync = sock_fasync,
163 .sendpage = sock_sendpage,
164 .splice_write = generic_splice_sendpage,
165 .splice_read = sock_splice_read,
166 .show_fdinfo = sock_show_fdinfo,
167};
168
169static const char * const pf_family_names[] = {
170 [PF_UNSPEC] = "PF_UNSPEC",
171 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
172 [PF_INET] = "PF_INET",
173 [PF_AX25] = "PF_AX25",
174 [PF_IPX] = "PF_IPX",
175 [PF_APPLETALK] = "PF_APPLETALK",
176 [PF_NETROM] = "PF_NETROM",
177 [PF_BRIDGE] = "PF_BRIDGE",
178 [PF_ATMPVC] = "PF_ATMPVC",
179 [PF_X25] = "PF_X25",
180 [PF_INET6] = "PF_INET6",
181 [PF_ROSE] = "PF_ROSE",
182 [PF_DECnet] = "PF_DECnet",
183 [PF_NETBEUI] = "PF_NETBEUI",
184 [PF_SECURITY] = "PF_SECURITY",
185 [PF_KEY] = "PF_KEY",
186 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
187 [PF_PACKET] = "PF_PACKET",
188 [PF_ASH] = "PF_ASH",
189 [PF_ECONET] = "PF_ECONET",
190 [PF_ATMSVC] = "PF_ATMSVC",
191 [PF_RDS] = "PF_RDS",
192 [PF_SNA] = "PF_SNA",
193 [PF_IRDA] = "PF_IRDA",
194 [PF_PPPOX] = "PF_PPPOX",
195 [PF_WANPIPE] = "PF_WANPIPE",
196 [PF_LLC] = "PF_LLC",
197 [PF_IB] = "PF_IB",
198 [PF_MPLS] = "PF_MPLS",
199 [PF_CAN] = "PF_CAN",
200 [PF_TIPC] = "PF_TIPC",
201 [PF_BLUETOOTH] = "PF_BLUETOOTH",
202 [PF_IUCV] = "PF_IUCV",
203 [PF_RXRPC] = "PF_RXRPC",
204 [PF_ISDN] = "PF_ISDN",
205 [PF_PHONET] = "PF_PHONET",
206 [PF_IEEE802154] = "PF_IEEE802154",
207 [PF_CAIF] = "PF_CAIF",
208 [PF_ALG] = "PF_ALG",
209 [PF_NFC] = "PF_NFC",
210 [PF_VSOCK] = "PF_VSOCK",
211 [PF_KCM] = "PF_KCM",
212 [PF_QIPCRTR] = "PF_QIPCRTR",
213 [PF_SMC] = "PF_SMC",
214 [PF_XDP] = "PF_XDP",
215};
216
217/*
218 * The protocol list. Each protocol is registered in here.
219 */
220
221static DEFINE_SPINLOCK(net_family_lock);
222static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
223
224/*
225 * Support routines.
226 * Move socket addresses back and forth across the kernel/user
227 * divide and look after the messy bits.
228 */
229
230/**
231 * move_addr_to_kernel - copy a socket address into kernel space
232 * @uaddr: Address in user space
233 * @kaddr: Address in kernel space
234 * @ulen: Length in user space
235 *
236 * The address is copied into kernel space. If the provided address is
237 * too long an error code of -EINVAL is returned. If the copy gives
238 * invalid addresses -EFAULT is returned. On a success 0 is returned.
239 */
240
241int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
242{
243 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
244 return -EINVAL;
245 if (ulen == 0)
246 return 0;
247 if (copy_from_user(kaddr, uaddr, ulen))
248 return -EFAULT;
249 return audit_sockaddr(ulen, kaddr);
250}
251
252/**
253 * move_addr_to_user - copy an address to user space
254 * @kaddr: kernel space address
255 * @klen: length of address in kernel
256 * @uaddr: user space address
257 * @ulen: pointer to user length field
258 *
259 * The value pointed to by ulen on entry is the buffer length available.
260 * This is overwritten with the buffer space used. -EINVAL is returned
261 * if an overlong buffer is specified or a negative buffer size. -EFAULT
262 * is returned if either the buffer or the length field are not
263 * accessible.
264 * After copying the data up to the limit the user specifies, the true
265 * length of the data is written over the length limit the user
266 * specified. Zero is returned for a success.
267 */
268
269static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
270 void __user *uaddr, int __user *ulen)
271{
272 int err;
273 int len;
274
275 BUG_ON(klen > sizeof(struct sockaddr_storage));
276 err = get_user(len, ulen);
277 if (err)
278 return err;
279 if (len > klen)
280 len = klen;
281 if (len < 0)
282 return -EINVAL;
283 if (len) {
284 if (audit_sockaddr(klen, kaddr))
285 return -ENOMEM;
286 if (copy_to_user(uaddr, kaddr, len))
287 return -EFAULT;
288 }
289 /*
290 * "fromlen shall refer to the value before truncation.."
291 * 1003.1g
292 */
293 return __put_user(klen, ulen);
294}
295
296static struct kmem_cache *sock_inode_cachep __ro_after_init;
297
298static struct inode *sock_alloc_inode(struct super_block *sb)
299{
300 struct socket_alloc *ei;
301
302 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
303 if (!ei)
304 return NULL;
305 init_waitqueue_head(&ei->socket.wq.wait);
306 ei->socket.wq.fasync_list = NULL;
307 ei->socket.wq.flags = 0;
308
309 ei->socket.state = SS_UNCONNECTED;
310 ei->socket.flags = 0;
311 ei->socket.ops = NULL;
312 ei->socket.sk = NULL;
313 ei->socket.file = NULL;
314
315 return &ei->vfs_inode;
316}
317
318static void sock_free_inode(struct inode *inode)
319{
320 struct socket_alloc *ei;
321
322 ei = container_of(inode, struct socket_alloc, vfs_inode);
323 kmem_cache_free(sock_inode_cachep, ei);
324}
325
326static void init_once(void *foo)
327{
328 struct socket_alloc *ei = (struct socket_alloc *)foo;
329
330 inode_init_once(&ei->vfs_inode);
331}
332
333static void init_inodecache(void)
334{
335 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
336 sizeof(struct socket_alloc),
337 0,
338 (SLAB_HWCACHE_ALIGN |
339 SLAB_RECLAIM_ACCOUNT |
340 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
341 init_once);
342 BUG_ON(sock_inode_cachep == NULL);
343}
344
345static const struct super_operations sockfs_ops = {
346 .alloc_inode = sock_alloc_inode,
347 .free_inode = sock_free_inode,
348 .statfs = simple_statfs,
349};
350
351/*
352 * sockfs_dname() is called from d_path().
353 */
354static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
355{
356 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
357 d_inode(dentry)->i_ino);
358}
359
360static const struct dentry_operations sockfs_dentry_operations = {
361 .d_dname = sockfs_dname,
362};
363
364static int sockfs_xattr_get(const struct xattr_handler *handler,
365 struct dentry *dentry, struct inode *inode,
366 const char *suffix, void *value, size_t size)
367{
368 if (value) {
369 if (dentry->d_name.len + 1 > size)
370 return -ERANGE;
371 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
372 }
373 return dentry->d_name.len + 1;
374}
375
376#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
377#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
378#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
379
380static const struct xattr_handler sockfs_xattr_handler = {
381 .name = XATTR_NAME_SOCKPROTONAME,
382 .get = sockfs_xattr_get,
383};
384
385static int sockfs_security_xattr_set(const struct xattr_handler *handler,
386 struct user_namespace *mnt_userns,
387 struct dentry *dentry, struct inode *inode,
388 const char *suffix, const void *value,
389 size_t size, int flags)
390{
391 /* Handled by LSM. */
392 return -EAGAIN;
393}
394
395static const struct xattr_handler sockfs_security_xattr_handler = {
396 .prefix = XATTR_SECURITY_PREFIX,
397 .set = sockfs_security_xattr_set,
398};
399
400static const struct xattr_handler *sockfs_xattr_handlers[] = {
401 &sockfs_xattr_handler,
402 &sockfs_security_xattr_handler,
403 NULL
404};
405
406static int sockfs_init_fs_context(struct fs_context *fc)
407{
408 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
409 if (!ctx)
410 return -ENOMEM;
411 ctx->ops = &sockfs_ops;
412 ctx->dops = &sockfs_dentry_operations;
413 ctx->xattr = sockfs_xattr_handlers;
414 return 0;
415}
416
417static struct vfsmount *sock_mnt __read_mostly;
418
419static struct file_system_type sock_fs_type = {
420 .name = "sockfs",
421 .init_fs_context = sockfs_init_fs_context,
422 .kill_sb = kill_anon_super,
423};
424
425/*
426 * Obtains the first available file descriptor and sets it up for use.
427 *
428 * These functions create file structures and maps them to fd space
429 * of the current process. On success it returns file descriptor
430 * and file struct implicitly stored in sock->file.
431 * Note that another thread may close file descriptor before we return
432 * from this function. We use the fact that now we do not refer
433 * to socket after mapping. If one day we will need it, this
434 * function will increment ref. count on file by 1.
435 *
436 * In any case returned fd MAY BE not valid!
437 * This race condition is unavoidable
438 * with shared fd spaces, we cannot solve it inside kernel,
439 * but we take care of internal coherence yet.
440 */
441
442/**
443 * sock_alloc_file - Bind a &socket to a &file
444 * @sock: socket
445 * @flags: file status flags
446 * @dname: protocol name
447 *
448 * Returns the &file bound with @sock, implicitly storing it
449 * in sock->file. If dname is %NULL, sets to "".
450 * On failure the return is a ERR pointer (see linux/err.h).
451 * This function uses GFP_KERNEL internally.
452 */
453
454struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
455{
456 struct file *file;
457
458 if (!dname)
459 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
460
461 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
462 O_RDWR | (flags & O_NONBLOCK),
463 &socket_file_ops);
464 if (IS_ERR(file)) {
465 sock_release(sock);
466 return file;
467 }
468
469 sock->file = file;
470 file->private_data = sock;
471 stream_open(SOCK_INODE(sock), file);
472 return file;
473}
474EXPORT_SYMBOL(sock_alloc_file);
475
476static int sock_map_fd(struct socket *sock, int flags)
477{
478 struct file *newfile;
479 int fd = get_unused_fd_flags(flags);
480 if (unlikely(fd < 0)) {
481 sock_release(sock);
482 return fd;
483 }
484
485 newfile = sock_alloc_file(sock, flags, NULL);
486 if (!IS_ERR(newfile)) {
487 fd_install(fd, newfile);
488 return fd;
489 }
490
491 put_unused_fd(fd);
492 return PTR_ERR(newfile);
493}
494
495/**
496 * sock_from_file - Return the &socket bounded to @file.
497 * @file: file
498 *
499 * On failure returns %NULL.
500 */
501
502struct socket *sock_from_file(struct file *file)
503{
504 if (file->f_op == &socket_file_ops)
505 return file->private_data; /* set in sock_map_fd */
506
507 return NULL;
508}
509EXPORT_SYMBOL(sock_from_file);
510
511/**
512 * sockfd_lookup - Go from a file number to its socket slot
513 * @fd: file handle
514 * @err: pointer to an error code return
515 *
516 * The file handle passed in is locked and the socket it is bound
517 * to is returned. If an error occurs the err pointer is overwritten
518 * with a negative errno code and NULL is returned. The function checks
519 * for both invalid handles and passing a handle which is not a socket.
520 *
521 * On a success the socket object pointer is returned.
522 */
523
524struct socket *sockfd_lookup(int fd, int *err)
525{
526 struct file *file;
527 struct socket *sock;
528
529 file = fget(fd);
530 if (!file) {
531 *err = -EBADF;
532 return NULL;
533 }
534
535 sock = sock_from_file(file);
536 if (!sock) {
537 *err = -ENOTSOCK;
538 fput(file);
539 }
540 return sock;
541}
542EXPORT_SYMBOL(sockfd_lookup);
543
544static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
545{
546 struct fd f = fdget(fd);
547 struct socket *sock;
548
549 *err = -EBADF;
550 if (f.file) {
551 sock = sock_from_file(f.file);
552 if (likely(sock)) {
553 *fput_needed = f.flags & FDPUT_FPUT;
554 return sock;
555 }
556 *err = -ENOTSOCK;
557 fdput(f);
558 }
559 return NULL;
560}
561
562static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
563 size_t size)
564{
565 ssize_t len;
566 ssize_t used = 0;
567
568 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
569 if (len < 0)
570 return len;
571 used += len;
572 if (buffer) {
573 if (size < used)
574 return -ERANGE;
575 buffer += len;
576 }
577
578 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
579 used += len;
580 if (buffer) {
581 if (size < used)
582 return -ERANGE;
583 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
584 buffer += len;
585 }
586
587 return used;
588}
589
590static int sockfs_setattr(struct user_namespace *mnt_userns,
591 struct dentry *dentry, struct iattr *iattr)
592{
593 int err = simple_setattr(&init_user_ns, dentry, iattr);
594
595 if (!err && (iattr->ia_valid & ATTR_UID)) {
596 struct socket *sock = SOCKET_I(d_inode(dentry));
597
598 if (sock->sk)
599 sock->sk->sk_uid = iattr->ia_uid;
600 else
601 err = -ENOENT;
602 }
603
604 return err;
605}
606
607static const struct inode_operations sockfs_inode_ops = {
608 .listxattr = sockfs_listxattr,
609 .setattr = sockfs_setattr,
610};
611
612/**
613 * sock_alloc - allocate a socket
614 *
615 * Allocate a new inode and socket object. The two are bound together
616 * and initialised. The socket is then returned. If we are out of inodes
617 * NULL is returned. This functions uses GFP_KERNEL internally.
618 */
619
620struct socket *sock_alloc(void)
621{
622 struct inode *inode;
623 struct socket *sock;
624
625 inode = new_inode_pseudo(sock_mnt->mnt_sb);
626 if (!inode)
627 return NULL;
628
629 sock = SOCKET_I(inode);
630
631 inode->i_ino = get_next_ino();
632 inode->i_mode = S_IFSOCK | S_IRWXUGO;
633 inode->i_uid = current_fsuid();
634 inode->i_gid = current_fsgid();
635 inode->i_op = &sockfs_inode_ops;
636
637 return sock;
638}
639EXPORT_SYMBOL(sock_alloc);
640
641static void __sock_release(struct socket *sock, struct inode *inode)
642{
643 if (sock->ops) {
644 struct module *owner = sock->ops->owner;
645
646 if (inode)
647 inode_lock(inode);
648 sock->ops->release(sock);
649 sock->sk = NULL;
650 if (inode)
651 inode_unlock(inode);
652 sock->ops = NULL;
653 module_put(owner);
654 }
655
656 if (sock->wq.fasync_list)
657 pr_err("%s: fasync list not empty!\n", __func__);
658
659 if (!sock->file) {
660 iput(SOCK_INODE(sock));
661 return;
662 }
663 sock->file = NULL;
664}
665
666/**
667 * sock_release - close a socket
668 * @sock: socket to close
669 *
670 * The socket is released from the protocol stack if it has a release
671 * callback, and the inode is then released if the socket is bound to
672 * an inode not a file.
673 */
674void sock_release(struct socket *sock)
675{
676 __sock_release(sock, NULL);
677}
678EXPORT_SYMBOL(sock_release);
679
680void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
681{
682 u8 flags = *tx_flags;
683
684 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
685 flags |= SKBTX_HW_TSTAMP;
686
687 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
688 flags |= SKBTX_SW_TSTAMP;
689
690 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
691 flags |= SKBTX_SCHED_TSTAMP;
692
693 *tx_flags = flags;
694}
695EXPORT_SYMBOL(__sock_tx_timestamp);
696
697INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
698 size_t));
699INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
700 size_t));
701static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
702{
703 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
704 inet_sendmsg, sock, msg,
705 msg_data_left(msg));
706 BUG_ON(ret == -EIOCBQUEUED);
707 return ret;
708}
709
710/**
711 * sock_sendmsg - send a message through @sock
712 * @sock: socket
713 * @msg: message to send
714 *
715 * Sends @msg through @sock, passing through LSM.
716 * Returns the number of bytes sent, or an error code.
717 */
718int sock_sendmsg(struct socket *sock, struct msghdr *msg)
719{
720 int err = security_socket_sendmsg(sock, msg,
721 msg_data_left(msg));
722
723 return err ?: sock_sendmsg_nosec(sock, msg);
724}
725EXPORT_SYMBOL(sock_sendmsg);
726
727/**
728 * kernel_sendmsg - send a message through @sock (kernel-space)
729 * @sock: socket
730 * @msg: message header
731 * @vec: kernel vec
732 * @num: vec array length
733 * @size: total message data size
734 *
735 * Builds the message data with @vec and sends it through @sock.
736 * Returns the number of bytes sent, or an error code.
737 */
738
739int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
740 struct kvec *vec, size_t num, size_t size)
741{
742 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
743 return sock_sendmsg(sock, msg);
744}
745EXPORT_SYMBOL(kernel_sendmsg);
746
747/**
748 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
749 * @sk: sock
750 * @msg: message header
751 * @vec: output s/g array
752 * @num: output s/g array length
753 * @size: total message data size
754 *
755 * Builds the message data with @vec and sends it through @sock.
756 * Returns the number of bytes sent, or an error code.
757 * Caller must hold @sk.
758 */
759
760int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
761 struct kvec *vec, size_t num, size_t size)
762{
763 struct socket *sock = sk->sk_socket;
764
765 if (!sock->ops->sendmsg_locked)
766 return sock_no_sendmsg_locked(sk, msg, size);
767
768 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
769
770 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
771}
772EXPORT_SYMBOL(kernel_sendmsg_locked);
773
774static bool skb_is_err_queue(const struct sk_buff *skb)
775{
776 /* pkt_type of skbs enqueued on the error queue are set to
777 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
778 * in recvmsg, since skbs received on a local socket will never
779 * have a pkt_type of PACKET_OUTGOING.
780 */
781 return skb->pkt_type == PACKET_OUTGOING;
782}
783
784/* On transmit, software and hardware timestamps are returned independently.
785 * As the two skb clones share the hardware timestamp, which may be updated
786 * before the software timestamp is received, a hardware TX timestamp may be
787 * returned only if there is no software TX timestamp. Ignore false software
788 * timestamps, which may be made in the __sock_recv_timestamp() call when the
789 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
790 * hardware timestamp.
791 */
792static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
793{
794 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
795}
796
797static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
798{
799 struct scm_ts_pktinfo ts_pktinfo;
800 struct net_device *orig_dev;
801
802 if (!skb_mac_header_was_set(skb))
803 return;
804
805 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
806
807 rcu_read_lock();
808 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
809 if (orig_dev)
810 ts_pktinfo.if_index = orig_dev->ifindex;
811 rcu_read_unlock();
812
813 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
814 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
815 sizeof(ts_pktinfo), &ts_pktinfo);
816}
817
818/*
819 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
820 */
821void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
822 struct sk_buff *skb)
823{
824 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
825 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
826 struct scm_timestamping_internal tss;
827
828 int empty = 1, false_tstamp = 0;
829 struct skb_shared_hwtstamps *shhwtstamps =
830 skb_hwtstamps(skb);
831
832 /* Race occurred between timestamp enabling and packet
833 receiving. Fill in the current time for now. */
834 if (need_software_tstamp && skb->tstamp == 0) {
835 __net_timestamp(skb);
836 false_tstamp = 1;
837 }
838
839 if (need_software_tstamp) {
840 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
841 if (new_tstamp) {
842 struct __kernel_sock_timeval tv;
843
844 skb_get_new_timestamp(skb, &tv);
845 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
846 sizeof(tv), &tv);
847 } else {
848 struct __kernel_old_timeval tv;
849
850 skb_get_timestamp(skb, &tv);
851 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
852 sizeof(tv), &tv);
853 }
854 } else {
855 if (new_tstamp) {
856 struct __kernel_timespec ts;
857
858 skb_get_new_timestampns(skb, &ts);
859 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
860 sizeof(ts), &ts);
861 } else {
862 struct __kernel_old_timespec ts;
863
864 skb_get_timestampns(skb, &ts);
865 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
866 sizeof(ts), &ts);
867 }
868 }
869 }
870
871 memset(&tss, 0, sizeof(tss));
872 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
873 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
874 empty = 0;
875 if (shhwtstamps &&
876 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
877 !skb_is_swtx_tstamp(skb, false_tstamp)) {
878 if (sk->sk_tsflags & SOF_TIMESTAMPING_BIND_PHC)
879 ptp_convert_timestamp(shhwtstamps, sk->sk_bind_phc);
880
881 if (ktime_to_timespec64_cond(shhwtstamps->hwtstamp,
882 tss.ts + 2)) {
883 empty = 0;
884
885 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
886 !skb_is_err_queue(skb))
887 put_ts_pktinfo(msg, skb);
888 }
889 }
890 if (!empty) {
891 if (sock_flag(sk, SOCK_TSTAMP_NEW))
892 put_cmsg_scm_timestamping64(msg, &tss);
893 else
894 put_cmsg_scm_timestamping(msg, &tss);
895
896 if (skb_is_err_queue(skb) && skb->len &&
897 SKB_EXT_ERR(skb)->opt_stats)
898 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
899 skb->len, skb->data);
900 }
901}
902EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
903
904void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
905 struct sk_buff *skb)
906{
907 int ack;
908
909 if (!sock_flag(sk, SOCK_WIFI_STATUS))
910 return;
911 if (!skb->wifi_acked_valid)
912 return;
913
914 ack = skb->wifi_acked;
915
916 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
917}
918EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
919
920static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
921 struct sk_buff *skb)
922{
923 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
924 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
925 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
926}
927
928void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
929 struct sk_buff *skb)
930{
931 sock_recv_timestamp(msg, sk, skb);
932 sock_recv_drops(msg, sk, skb);
933}
934EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
935
936INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
937 size_t, int));
938INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
939 size_t, int));
940static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
941 int flags)
942{
943 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
944 inet_recvmsg, sock, msg, msg_data_left(msg),
945 flags);
946}
947
948/**
949 * sock_recvmsg - receive a message from @sock
950 * @sock: socket
951 * @msg: message to receive
952 * @flags: message flags
953 *
954 * Receives @msg from @sock, passing through LSM. Returns the total number
955 * of bytes received, or an error.
956 */
957int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
958{
959 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
960
961 return err ?: sock_recvmsg_nosec(sock, msg, flags);
962}
963EXPORT_SYMBOL(sock_recvmsg);
964
965/**
966 * kernel_recvmsg - Receive a message from a socket (kernel space)
967 * @sock: The socket to receive the message from
968 * @msg: Received message
969 * @vec: Input s/g array for message data
970 * @num: Size of input s/g array
971 * @size: Number of bytes to read
972 * @flags: Message flags (MSG_DONTWAIT, etc...)
973 *
974 * On return the msg structure contains the scatter/gather array passed in the
975 * vec argument. The array is modified so that it consists of the unfilled
976 * portion of the original array.
977 *
978 * The returned value is the total number of bytes received, or an error.
979 */
980
981int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
982 struct kvec *vec, size_t num, size_t size, int flags)
983{
984 msg->msg_control_is_user = false;
985 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
986 return sock_recvmsg(sock, msg, flags);
987}
988EXPORT_SYMBOL(kernel_recvmsg);
989
990static ssize_t sock_sendpage(struct file *file, struct page *page,
991 int offset, size_t size, loff_t *ppos, int more)
992{
993 struct socket *sock;
994 int flags;
995
996 sock = file->private_data;
997
998 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
999 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
1000 flags |= more;
1001
1002 return kernel_sendpage(sock, page, offset, size, flags);
1003}
1004
1005static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1006 struct pipe_inode_info *pipe, size_t len,
1007 unsigned int flags)
1008{
1009 struct socket *sock = file->private_data;
1010
1011 if (unlikely(!sock->ops->splice_read))
1012 return generic_file_splice_read(file, ppos, pipe, len, flags);
1013
1014 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
1015}
1016
1017static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1018{
1019 struct file *file = iocb->ki_filp;
1020 struct socket *sock = file->private_data;
1021 struct msghdr msg = {.msg_iter = *to,
1022 .msg_iocb = iocb};
1023 ssize_t res;
1024
1025 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1026 msg.msg_flags = MSG_DONTWAIT;
1027
1028 if (iocb->ki_pos != 0)
1029 return -ESPIPE;
1030
1031 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1032 return 0;
1033
1034 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1035 *to = msg.msg_iter;
1036 return res;
1037}
1038
1039static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1040{
1041 struct file *file = iocb->ki_filp;
1042 struct socket *sock = file->private_data;
1043 struct msghdr msg = {.msg_iter = *from,
1044 .msg_iocb = iocb};
1045 ssize_t res;
1046
1047 if (iocb->ki_pos != 0)
1048 return -ESPIPE;
1049
1050 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1051 msg.msg_flags = MSG_DONTWAIT;
1052
1053 if (sock->type == SOCK_SEQPACKET)
1054 msg.msg_flags |= MSG_EOR;
1055
1056 res = sock_sendmsg(sock, &msg);
1057 *from = msg.msg_iter;
1058 return res;
1059}
1060
1061/*
1062 * Atomic setting of ioctl hooks to avoid race
1063 * with module unload.
1064 */
1065
1066static DEFINE_MUTEX(br_ioctl_mutex);
1067static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
1068
1069void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
1070{
1071 mutex_lock(&br_ioctl_mutex);
1072 br_ioctl_hook = hook;
1073 mutex_unlock(&br_ioctl_mutex);
1074}
1075EXPORT_SYMBOL(brioctl_set);
1076
1077static DEFINE_MUTEX(vlan_ioctl_mutex);
1078static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1079
1080void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1081{
1082 mutex_lock(&vlan_ioctl_mutex);
1083 vlan_ioctl_hook = hook;
1084 mutex_unlock(&vlan_ioctl_mutex);
1085}
1086EXPORT_SYMBOL(vlan_ioctl_set);
1087
1088static long sock_do_ioctl(struct net *net, struct socket *sock,
1089 unsigned int cmd, unsigned long arg)
1090{
1091 int err;
1092 void __user *argp = (void __user *)arg;
1093
1094 err = sock->ops->ioctl(sock, cmd, arg);
1095
1096 /*
1097 * If this ioctl is unknown try to hand it down
1098 * to the NIC driver.
1099 */
1100 if (err != -ENOIOCTLCMD)
1101 return err;
1102
1103 if (cmd == SIOCGIFCONF) {
1104 struct ifconf ifc;
1105 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
1106 return -EFAULT;
1107 rtnl_lock();
1108 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
1109 rtnl_unlock();
1110 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
1111 err = -EFAULT;
1112 } else if (is_socket_ioctl_cmd(cmd)) {
1113 struct ifreq ifr;
1114 bool need_copyout;
1115 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1116 return -EFAULT;
1117 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1118 if (!err && need_copyout)
1119 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1120 return -EFAULT;
1121 } else {
1122 err = -ENOTTY;
1123 }
1124 return err;
1125}
1126
1127/*
1128 * With an ioctl, arg may well be a user mode pointer, but we don't know
1129 * what to do with it - that's up to the protocol still.
1130 */
1131
1132static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1133{
1134 struct socket *sock;
1135 struct sock *sk;
1136 void __user *argp = (void __user *)arg;
1137 int pid, err;
1138 struct net *net;
1139
1140 sock = file->private_data;
1141 sk = sock->sk;
1142 net = sock_net(sk);
1143 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1144 struct ifreq ifr;
1145 bool need_copyout;
1146 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1147 return -EFAULT;
1148 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1149 if (!err && need_copyout)
1150 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1151 return -EFAULT;
1152 } else
1153#ifdef CONFIG_WEXT_CORE
1154 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1155 err = wext_handle_ioctl(net, cmd, argp);
1156 } else
1157#endif
1158 switch (cmd) {
1159 case FIOSETOWN:
1160 case SIOCSPGRP:
1161 err = -EFAULT;
1162 if (get_user(pid, (int __user *)argp))
1163 break;
1164 err = f_setown(sock->file, pid, 1);
1165 break;
1166 case FIOGETOWN:
1167 case SIOCGPGRP:
1168 err = put_user(f_getown(sock->file),
1169 (int __user *)argp);
1170 break;
1171 case SIOCGIFBR:
1172 case SIOCSIFBR:
1173 case SIOCBRADDBR:
1174 case SIOCBRDELBR:
1175 err = -ENOPKG;
1176 if (!br_ioctl_hook)
1177 request_module("bridge");
1178
1179 mutex_lock(&br_ioctl_mutex);
1180 if (br_ioctl_hook)
1181 err = br_ioctl_hook(net, cmd, argp);
1182 mutex_unlock(&br_ioctl_mutex);
1183 break;
1184 case SIOCGIFVLAN:
1185 case SIOCSIFVLAN:
1186 err = -ENOPKG;
1187 if (!vlan_ioctl_hook)
1188 request_module("8021q");
1189
1190 mutex_lock(&vlan_ioctl_mutex);
1191 if (vlan_ioctl_hook)
1192 err = vlan_ioctl_hook(net, argp);
1193 mutex_unlock(&vlan_ioctl_mutex);
1194 break;
1195 case SIOCGSKNS:
1196 err = -EPERM;
1197 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1198 break;
1199
1200 err = open_related_ns(&net->ns, get_net_ns);
1201 break;
1202 case SIOCGSTAMP_OLD:
1203 case SIOCGSTAMPNS_OLD:
1204 if (!sock->ops->gettstamp) {
1205 err = -ENOIOCTLCMD;
1206 break;
1207 }
1208 err = sock->ops->gettstamp(sock, argp,
1209 cmd == SIOCGSTAMP_OLD,
1210 !IS_ENABLED(CONFIG_64BIT));
1211 break;
1212 case SIOCGSTAMP_NEW:
1213 case SIOCGSTAMPNS_NEW:
1214 if (!sock->ops->gettstamp) {
1215 err = -ENOIOCTLCMD;
1216 break;
1217 }
1218 err = sock->ops->gettstamp(sock, argp,
1219 cmd == SIOCGSTAMP_NEW,
1220 false);
1221 break;
1222 default:
1223 err = sock_do_ioctl(net, sock, cmd, arg);
1224 break;
1225 }
1226 return err;
1227}
1228
1229/**
1230 * sock_create_lite - creates a socket
1231 * @family: protocol family (AF_INET, ...)
1232 * @type: communication type (SOCK_STREAM, ...)
1233 * @protocol: protocol (0, ...)
1234 * @res: new socket
1235 *
1236 * Creates a new socket and assigns it to @res, passing through LSM.
1237 * The new socket initialization is not complete, see kernel_accept().
1238 * Returns 0 or an error. On failure @res is set to %NULL.
1239 * This function internally uses GFP_KERNEL.
1240 */
1241
1242int sock_create_lite(int family, int type, int protocol, struct socket **res)
1243{
1244 int err;
1245 struct socket *sock = NULL;
1246
1247 err = security_socket_create(family, type, protocol, 1);
1248 if (err)
1249 goto out;
1250
1251 sock = sock_alloc();
1252 if (!sock) {
1253 err = -ENOMEM;
1254 goto out;
1255 }
1256
1257 sock->type = type;
1258 err = security_socket_post_create(sock, family, type, protocol, 1);
1259 if (err)
1260 goto out_release;
1261
1262out:
1263 *res = sock;
1264 return err;
1265out_release:
1266 sock_release(sock);
1267 sock = NULL;
1268 goto out;
1269}
1270EXPORT_SYMBOL(sock_create_lite);
1271
1272/* No kernel lock held - perfect */
1273static __poll_t sock_poll(struct file *file, poll_table *wait)
1274{
1275 struct socket *sock = file->private_data;
1276 __poll_t events = poll_requested_events(wait), flag = 0;
1277
1278 if (!sock->ops->poll)
1279 return 0;
1280
1281 if (sk_can_busy_loop(sock->sk)) {
1282 /* poll once if requested by the syscall */
1283 if (events & POLL_BUSY_LOOP)
1284 sk_busy_loop(sock->sk, 1);
1285
1286 /* if this socket can poll_ll, tell the system call */
1287 flag = POLL_BUSY_LOOP;
1288 }
1289
1290 return sock->ops->poll(file, sock, wait) | flag;
1291}
1292
1293static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1294{
1295 struct socket *sock = file->private_data;
1296
1297 return sock->ops->mmap(file, sock, vma);
1298}
1299
1300static int sock_close(struct inode *inode, struct file *filp)
1301{
1302 __sock_release(SOCKET_I(inode), inode);
1303 return 0;
1304}
1305
1306/*
1307 * Update the socket async list
1308 *
1309 * Fasync_list locking strategy.
1310 *
1311 * 1. fasync_list is modified only under process context socket lock
1312 * i.e. under semaphore.
1313 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1314 * or under socket lock
1315 */
1316
1317static int sock_fasync(int fd, struct file *filp, int on)
1318{
1319 struct socket *sock = filp->private_data;
1320 struct sock *sk = sock->sk;
1321 struct socket_wq *wq = &sock->wq;
1322
1323 if (sk == NULL)
1324 return -EINVAL;
1325
1326 lock_sock(sk);
1327 fasync_helper(fd, filp, on, &wq->fasync_list);
1328
1329 if (!wq->fasync_list)
1330 sock_reset_flag(sk, SOCK_FASYNC);
1331 else
1332 sock_set_flag(sk, SOCK_FASYNC);
1333
1334 release_sock(sk);
1335 return 0;
1336}
1337
1338/* This function may be called only under rcu_lock */
1339
1340int sock_wake_async(struct socket_wq *wq, int how, int band)
1341{
1342 if (!wq || !wq->fasync_list)
1343 return -1;
1344
1345 switch (how) {
1346 case SOCK_WAKE_WAITD:
1347 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1348 break;
1349 goto call_kill;
1350 case SOCK_WAKE_SPACE:
1351 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1352 break;
1353 fallthrough;
1354 case SOCK_WAKE_IO:
1355call_kill:
1356 kill_fasync(&wq->fasync_list, SIGIO, band);
1357 break;
1358 case SOCK_WAKE_URG:
1359 kill_fasync(&wq->fasync_list, SIGURG, band);
1360 }
1361
1362 return 0;
1363}
1364EXPORT_SYMBOL(sock_wake_async);
1365
1366/**
1367 * __sock_create - creates a socket
1368 * @net: net namespace
1369 * @family: protocol family (AF_INET, ...)
1370 * @type: communication type (SOCK_STREAM, ...)
1371 * @protocol: protocol (0, ...)
1372 * @res: new socket
1373 * @kern: boolean for kernel space sockets
1374 *
1375 * Creates a new socket and assigns it to @res, passing through LSM.
1376 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1377 * be set to true if the socket resides in kernel space.
1378 * This function internally uses GFP_KERNEL.
1379 */
1380
1381int __sock_create(struct net *net, int family, int type, int protocol,
1382 struct socket **res, int kern)
1383{
1384 int err;
1385 struct socket *sock;
1386 const struct net_proto_family *pf;
1387
1388 /*
1389 * Check protocol is in range
1390 */
1391 if (family < 0 || family >= NPROTO)
1392 return -EAFNOSUPPORT;
1393 if (type < 0 || type >= SOCK_MAX)
1394 return -EINVAL;
1395
1396 /* Compatibility.
1397
1398 This uglymoron is moved from INET layer to here to avoid
1399 deadlock in module load.
1400 */
1401 if (family == PF_INET && type == SOCK_PACKET) {
1402 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1403 current->comm);
1404 family = PF_PACKET;
1405 }
1406
1407 err = security_socket_create(family, type, protocol, kern);
1408 if (err)
1409 return err;
1410
1411 /*
1412 * Allocate the socket and allow the family to set things up. if
1413 * the protocol is 0, the family is instructed to select an appropriate
1414 * default.
1415 */
1416 sock = sock_alloc();
1417 if (!sock) {
1418 net_warn_ratelimited("socket: no more sockets\n");
1419 return -ENFILE; /* Not exactly a match, but its the
1420 closest posix thing */
1421 }
1422
1423 sock->type = type;
1424
1425#ifdef CONFIG_MODULES
1426 /* Attempt to load a protocol module if the find failed.
1427 *
1428 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1429 * requested real, full-featured networking support upon configuration.
1430 * Otherwise module support will break!
1431 */
1432 if (rcu_access_pointer(net_families[family]) == NULL)
1433 request_module("net-pf-%d", family);
1434#endif
1435
1436 rcu_read_lock();
1437 pf = rcu_dereference(net_families[family]);
1438 err = -EAFNOSUPPORT;
1439 if (!pf)
1440 goto out_release;
1441
1442 /*
1443 * We will call the ->create function, that possibly is in a loadable
1444 * module, so we have to bump that loadable module refcnt first.
1445 */
1446 if (!try_module_get(pf->owner))
1447 goto out_release;
1448
1449 /* Now protected by module ref count */
1450 rcu_read_unlock();
1451
1452 err = pf->create(net, sock, protocol, kern);
1453 if (err < 0)
1454 goto out_module_put;
1455
1456 /*
1457 * Now to bump the refcnt of the [loadable] module that owns this
1458 * socket at sock_release time we decrement its refcnt.
1459 */
1460 if (!try_module_get(sock->ops->owner))
1461 goto out_module_busy;
1462
1463 /*
1464 * Now that we're done with the ->create function, the [loadable]
1465 * module can have its refcnt decremented
1466 */
1467 module_put(pf->owner);
1468 err = security_socket_post_create(sock, family, type, protocol, kern);
1469 if (err)
1470 goto out_sock_release;
1471 *res = sock;
1472
1473 return 0;
1474
1475out_module_busy:
1476 err = -EAFNOSUPPORT;
1477out_module_put:
1478 sock->ops = NULL;
1479 module_put(pf->owner);
1480out_sock_release:
1481 sock_release(sock);
1482 return err;
1483
1484out_release:
1485 rcu_read_unlock();
1486 goto out_sock_release;
1487}
1488EXPORT_SYMBOL(__sock_create);
1489
1490/**
1491 * sock_create - creates a socket
1492 * @family: protocol family (AF_INET, ...)
1493 * @type: communication type (SOCK_STREAM, ...)
1494 * @protocol: protocol (0, ...)
1495 * @res: new socket
1496 *
1497 * A wrapper around __sock_create().
1498 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1499 */
1500
1501int sock_create(int family, int type, int protocol, struct socket **res)
1502{
1503 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1504}
1505EXPORT_SYMBOL(sock_create);
1506
1507/**
1508 * sock_create_kern - creates a socket (kernel space)
1509 * @net: net namespace
1510 * @family: protocol family (AF_INET, ...)
1511 * @type: communication type (SOCK_STREAM, ...)
1512 * @protocol: protocol (0, ...)
1513 * @res: new socket
1514 *
1515 * A wrapper around __sock_create().
1516 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1517 */
1518
1519int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1520{
1521 return __sock_create(net, family, type, protocol, res, 1);
1522}
1523EXPORT_SYMBOL(sock_create_kern);
1524
1525int __sys_socket(int family, int type, int protocol)
1526{
1527 int retval;
1528 struct socket *sock;
1529 int flags;
1530
1531 /* Check the SOCK_* constants for consistency. */
1532 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1533 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1534 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1535 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1536
1537 flags = type & ~SOCK_TYPE_MASK;
1538 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1539 return -EINVAL;
1540 type &= SOCK_TYPE_MASK;
1541
1542 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1543 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1544
1545 retval = sock_create(family, type, protocol, &sock);
1546 if (retval < 0)
1547 return retval;
1548
1549 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1550}
1551
1552SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1553{
1554 return __sys_socket(family, type, protocol);
1555}
1556
1557/*
1558 * Create a pair of connected sockets.
1559 */
1560
1561int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1562{
1563 struct socket *sock1, *sock2;
1564 int fd1, fd2, err;
1565 struct file *newfile1, *newfile2;
1566 int flags;
1567
1568 flags = type & ~SOCK_TYPE_MASK;
1569 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1570 return -EINVAL;
1571 type &= SOCK_TYPE_MASK;
1572
1573 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1574 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1575
1576 /*
1577 * reserve descriptors and make sure we won't fail
1578 * to return them to userland.
1579 */
1580 fd1 = get_unused_fd_flags(flags);
1581 if (unlikely(fd1 < 0))
1582 return fd1;
1583
1584 fd2 = get_unused_fd_flags(flags);
1585 if (unlikely(fd2 < 0)) {
1586 put_unused_fd(fd1);
1587 return fd2;
1588 }
1589
1590 err = put_user(fd1, &usockvec[0]);
1591 if (err)
1592 goto out;
1593
1594 err = put_user(fd2, &usockvec[1]);
1595 if (err)
1596 goto out;
1597
1598 /*
1599 * Obtain the first socket and check if the underlying protocol
1600 * supports the socketpair call.
1601 */
1602
1603 err = sock_create(family, type, protocol, &sock1);
1604 if (unlikely(err < 0))
1605 goto out;
1606
1607 err = sock_create(family, type, protocol, &sock2);
1608 if (unlikely(err < 0)) {
1609 sock_release(sock1);
1610 goto out;
1611 }
1612
1613 err = security_socket_socketpair(sock1, sock2);
1614 if (unlikely(err)) {
1615 sock_release(sock2);
1616 sock_release(sock1);
1617 goto out;
1618 }
1619
1620 err = sock1->ops->socketpair(sock1, sock2);
1621 if (unlikely(err < 0)) {
1622 sock_release(sock2);
1623 sock_release(sock1);
1624 goto out;
1625 }
1626
1627 newfile1 = sock_alloc_file(sock1, flags, NULL);
1628 if (IS_ERR(newfile1)) {
1629 err = PTR_ERR(newfile1);
1630 sock_release(sock2);
1631 goto out;
1632 }
1633
1634 newfile2 = sock_alloc_file(sock2, flags, NULL);
1635 if (IS_ERR(newfile2)) {
1636 err = PTR_ERR(newfile2);
1637 fput(newfile1);
1638 goto out;
1639 }
1640
1641 audit_fd_pair(fd1, fd2);
1642
1643 fd_install(fd1, newfile1);
1644 fd_install(fd2, newfile2);
1645 return 0;
1646
1647out:
1648 put_unused_fd(fd2);
1649 put_unused_fd(fd1);
1650 return err;
1651}
1652
1653SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1654 int __user *, usockvec)
1655{
1656 return __sys_socketpair(family, type, protocol, usockvec);
1657}
1658
1659/*
1660 * Bind a name to a socket. Nothing much to do here since it's
1661 * the protocol's responsibility to handle the local address.
1662 *
1663 * We move the socket address to kernel space before we call
1664 * the protocol layer (having also checked the address is ok).
1665 */
1666
1667int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1668{
1669 struct socket *sock;
1670 struct sockaddr_storage address;
1671 int err, fput_needed;
1672
1673 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1674 if (sock) {
1675 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1676 if (!err) {
1677 err = security_socket_bind(sock,
1678 (struct sockaddr *)&address,
1679 addrlen);
1680 if (!err)
1681 err = sock->ops->bind(sock,
1682 (struct sockaddr *)
1683 &address, addrlen);
1684 }
1685 fput_light(sock->file, fput_needed);
1686 }
1687 return err;
1688}
1689
1690SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1691{
1692 return __sys_bind(fd, umyaddr, addrlen);
1693}
1694
1695/*
1696 * Perform a listen. Basically, we allow the protocol to do anything
1697 * necessary for a listen, and if that works, we mark the socket as
1698 * ready for listening.
1699 */
1700
1701int __sys_listen(int fd, int backlog)
1702{
1703 struct socket *sock;
1704 int err, fput_needed;
1705 int somaxconn;
1706
1707 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1708 if (sock) {
1709 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1710 if ((unsigned int)backlog > somaxconn)
1711 backlog = somaxconn;
1712
1713 err = security_socket_listen(sock, backlog);
1714 if (!err)
1715 err = sock->ops->listen(sock, backlog);
1716
1717 fput_light(sock->file, fput_needed);
1718 }
1719 return err;
1720}
1721
1722SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1723{
1724 return __sys_listen(fd, backlog);
1725}
1726
1727int __sys_accept4_file(struct file *file, unsigned file_flags,
1728 struct sockaddr __user *upeer_sockaddr,
1729 int __user *upeer_addrlen, int flags,
1730 unsigned long nofile)
1731{
1732 struct socket *sock, *newsock;
1733 struct file *newfile;
1734 int err, len, newfd;
1735 struct sockaddr_storage address;
1736
1737 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1738 return -EINVAL;
1739
1740 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1741 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1742
1743 sock = sock_from_file(file);
1744 if (!sock) {
1745 err = -ENOTSOCK;
1746 goto out;
1747 }
1748
1749 err = -ENFILE;
1750 newsock = sock_alloc();
1751 if (!newsock)
1752 goto out;
1753
1754 newsock->type = sock->type;
1755 newsock->ops = sock->ops;
1756
1757 /*
1758 * We don't need try_module_get here, as the listening socket (sock)
1759 * has the protocol module (sock->ops->owner) held.
1760 */
1761 __module_get(newsock->ops->owner);
1762
1763 newfd = __get_unused_fd_flags(flags, nofile);
1764 if (unlikely(newfd < 0)) {
1765 err = newfd;
1766 sock_release(newsock);
1767 goto out;
1768 }
1769 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1770 if (IS_ERR(newfile)) {
1771 err = PTR_ERR(newfile);
1772 put_unused_fd(newfd);
1773 goto out;
1774 }
1775
1776 err = security_socket_accept(sock, newsock);
1777 if (err)
1778 goto out_fd;
1779
1780 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1781 false);
1782 if (err < 0)
1783 goto out_fd;
1784
1785 if (upeer_sockaddr) {
1786 len = newsock->ops->getname(newsock,
1787 (struct sockaddr *)&address, 2);
1788 if (len < 0) {
1789 err = -ECONNABORTED;
1790 goto out_fd;
1791 }
1792 err = move_addr_to_user(&address,
1793 len, upeer_sockaddr, upeer_addrlen);
1794 if (err < 0)
1795 goto out_fd;
1796 }
1797
1798 /* File flags are not inherited via accept() unlike another OSes. */
1799
1800 fd_install(newfd, newfile);
1801 err = newfd;
1802out:
1803 return err;
1804out_fd:
1805 fput(newfile);
1806 put_unused_fd(newfd);
1807 goto out;
1808
1809}
1810
1811/*
1812 * For accept, we attempt to create a new socket, set up the link
1813 * with the client, wake up the client, then return the new
1814 * connected fd. We collect the address of the connector in kernel
1815 * space and move it to user at the very end. This is unclean because
1816 * we open the socket then return an error.
1817 *
1818 * 1003.1g adds the ability to recvmsg() to query connection pending
1819 * status to recvmsg. We need to add that support in a way thats
1820 * clean when we restructure accept also.
1821 */
1822
1823int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1824 int __user *upeer_addrlen, int flags)
1825{
1826 int ret = -EBADF;
1827 struct fd f;
1828
1829 f = fdget(fd);
1830 if (f.file) {
1831 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1832 upeer_addrlen, flags,
1833 rlimit(RLIMIT_NOFILE));
1834 fdput(f);
1835 }
1836
1837 return ret;
1838}
1839
1840SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1841 int __user *, upeer_addrlen, int, flags)
1842{
1843 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1844}
1845
1846SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1847 int __user *, upeer_addrlen)
1848{
1849 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1850}
1851
1852/*
1853 * Attempt to connect to a socket with the server address. The address
1854 * is in user space so we verify it is OK and move it to kernel space.
1855 *
1856 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1857 * break bindings
1858 *
1859 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1860 * other SEQPACKET protocols that take time to connect() as it doesn't
1861 * include the -EINPROGRESS status for such sockets.
1862 */
1863
1864int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1865 int addrlen, int file_flags)
1866{
1867 struct socket *sock;
1868 int err;
1869
1870 sock = sock_from_file(file);
1871 if (!sock) {
1872 err = -ENOTSOCK;
1873 goto out;
1874 }
1875
1876 err =
1877 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1878 if (err)
1879 goto out;
1880
1881 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1882 sock->file->f_flags | file_flags);
1883out:
1884 return err;
1885}
1886
1887int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1888{
1889 int ret = -EBADF;
1890 struct fd f;
1891
1892 f = fdget(fd);
1893 if (f.file) {
1894 struct sockaddr_storage address;
1895
1896 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1897 if (!ret)
1898 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1899 fdput(f);
1900 }
1901
1902 return ret;
1903}
1904
1905SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1906 int, addrlen)
1907{
1908 return __sys_connect(fd, uservaddr, addrlen);
1909}
1910
1911/*
1912 * Get the local address ('name') of a socket object. Move the obtained
1913 * name to user space.
1914 */
1915
1916int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1917 int __user *usockaddr_len)
1918{
1919 struct socket *sock;
1920 struct sockaddr_storage address;
1921 int err, fput_needed;
1922
1923 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1924 if (!sock)
1925 goto out;
1926
1927 err = security_socket_getsockname(sock);
1928 if (err)
1929 goto out_put;
1930
1931 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1932 if (err < 0)
1933 goto out_put;
1934 /* "err" is actually length in this case */
1935 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1936
1937out_put:
1938 fput_light(sock->file, fput_needed);
1939out:
1940 return err;
1941}
1942
1943SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1944 int __user *, usockaddr_len)
1945{
1946 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1947}
1948
1949/*
1950 * Get the remote address ('name') of a socket object. Move the obtained
1951 * name to user space.
1952 */
1953
1954int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1955 int __user *usockaddr_len)
1956{
1957 struct socket *sock;
1958 struct sockaddr_storage address;
1959 int err, fput_needed;
1960
1961 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1962 if (sock != NULL) {
1963 err = security_socket_getpeername(sock);
1964 if (err) {
1965 fput_light(sock->file, fput_needed);
1966 return err;
1967 }
1968
1969 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1970 if (err >= 0)
1971 /* "err" is actually length in this case */
1972 err = move_addr_to_user(&address, err, usockaddr,
1973 usockaddr_len);
1974 fput_light(sock->file, fput_needed);
1975 }
1976 return err;
1977}
1978
1979SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1980 int __user *, usockaddr_len)
1981{
1982 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1983}
1984
1985/*
1986 * Send a datagram to a given address. We move the address into kernel
1987 * space and check the user space data area is readable before invoking
1988 * the protocol.
1989 */
1990int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1991 struct sockaddr __user *addr, int addr_len)
1992{
1993 struct socket *sock;
1994 struct sockaddr_storage address;
1995 int err;
1996 struct msghdr msg;
1997 struct iovec iov;
1998 int fput_needed;
1999
2000 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
2001 if (unlikely(err))
2002 return err;
2003 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2004 if (!sock)
2005 goto out;
2006
2007 msg.msg_name = NULL;
2008 msg.msg_control = NULL;
2009 msg.msg_controllen = 0;
2010 msg.msg_namelen = 0;
2011 if (addr) {
2012 err = move_addr_to_kernel(addr, addr_len, &address);
2013 if (err < 0)
2014 goto out_put;
2015 msg.msg_name = (struct sockaddr *)&address;
2016 msg.msg_namelen = addr_len;
2017 }
2018 if (sock->file->f_flags & O_NONBLOCK)
2019 flags |= MSG_DONTWAIT;
2020 msg.msg_flags = flags;
2021 err = sock_sendmsg(sock, &msg);
2022
2023out_put:
2024 fput_light(sock->file, fput_needed);
2025out:
2026 return err;
2027}
2028
2029SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2030 unsigned int, flags, struct sockaddr __user *, addr,
2031 int, addr_len)
2032{
2033 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2034}
2035
2036/*
2037 * Send a datagram down a socket.
2038 */
2039
2040SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2041 unsigned int, flags)
2042{
2043 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2044}
2045
2046/*
2047 * Receive a frame from the socket and optionally record the address of the
2048 * sender. We verify the buffers are writable and if needed move the
2049 * sender address from kernel to user space.
2050 */
2051int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2052 struct sockaddr __user *addr, int __user *addr_len)
2053{
2054 struct socket *sock;
2055 struct iovec iov;
2056 struct msghdr msg;
2057 struct sockaddr_storage address;
2058 int err, err2;
2059 int fput_needed;
2060
2061 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2062 if (unlikely(err))
2063 return err;
2064 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2065 if (!sock)
2066 goto out;
2067
2068 msg.msg_control = NULL;
2069 msg.msg_controllen = 0;
2070 /* Save some cycles and don't copy the address if not needed */
2071 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2072 /* We assume all kernel code knows the size of sockaddr_storage */
2073 msg.msg_namelen = 0;
2074 msg.msg_iocb = NULL;
2075 msg.msg_flags = 0;
2076 if (sock->file->f_flags & O_NONBLOCK)
2077 flags |= MSG_DONTWAIT;
2078 err = sock_recvmsg(sock, &msg, flags);
2079
2080 if (err >= 0 && addr != NULL) {
2081 err2 = move_addr_to_user(&address,
2082 msg.msg_namelen, addr, addr_len);
2083 if (err2 < 0)
2084 err = err2;
2085 }
2086
2087 fput_light(sock->file, fput_needed);
2088out:
2089 return err;
2090}
2091
2092SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2093 unsigned int, flags, struct sockaddr __user *, addr,
2094 int __user *, addr_len)
2095{
2096 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2097}
2098
2099/*
2100 * Receive a datagram from a socket.
2101 */
2102
2103SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2104 unsigned int, flags)
2105{
2106 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2107}
2108
2109static bool sock_use_custom_sol_socket(const struct socket *sock)
2110{
2111 const struct sock *sk = sock->sk;
2112
2113 /* Use sock->ops->setsockopt() for MPTCP */
2114 return IS_ENABLED(CONFIG_MPTCP) &&
2115 sk->sk_protocol == IPPROTO_MPTCP &&
2116 sk->sk_type == SOCK_STREAM &&
2117 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2118}
2119
2120/*
2121 * Set a socket option. Because we don't know the option lengths we have
2122 * to pass the user mode parameter for the protocols to sort out.
2123 */
2124int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2125 int optlen)
2126{
2127 sockptr_t optval = USER_SOCKPTR(user_optval);
2128 char *kernel_optval = NULL;
2129 int err, fput_needed;
2130 struct socket *sock;
2131
2132 if (optlen < 0)
2133 return -EINVAL;
2134
2135 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2136 if (!sock)
2137 return err;
2138
2139 err = security_socket_setsockopt(sock, level, optname);
2140 if (err)
2141 goto out_put;
2142
2143 if (!in_compat_syscall())
2144 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2145 user_optval, &optlen,
2146 &kernel_optval);
2147 if (err < 0)
2148 goto out_put;
2149 if (err > 0) {
2150 err = 0;
2151 goto out_put;
2152 }
2153
2154 if (kernel_optval)
2155 optval = KERNEL_SOCKPTR(kernel_optval);
2156 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2157 err = sock_setsockopt(sock, level, optname, optval, optlen);
2158 else if (unlikely(!sock->ops->setsockopt))
2159 err = -EOPNOTSUPP;
2160 else
2161 err = sock->ops->setsockopt(sock, level, optname, optval,
2162 optlen);
2163 kfree(kernel_optval);
2164out_put:
2165 fput_light(sock->file, fput_needed);
2166 return err;
2167}
2168
2169SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2170 char __user *, optval, int, optlen)
2171{
2172 return __sys_setsockopt(fd, level, optname, optval, optlen);
2173}
2174
2175INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2176 int optname));
2177
2178/*
2179 * Get a socket option. Because we don't know the option lengths we have
2180 * to pass a user mode parameter for the protocols to sort out.
2181 */
2182int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2183 int __user *optlen)
2184{
2185 int err, fput_needed;
2186 struct socket *sock;
2187 int max_optlen;
2188
2189 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2190 if (!sock)
2191 return err;
2192
2193 err = security_socket_getsockopt(sock, level, optname);
2194 if (err)
2195 goto out_put;
2196
2197 if (!in_compat_syscall())
2198 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2199
2200 if (level == SOL_SOCKET)
2201 err = sock_getsockopt(sock, level, optname, optval, optlen);
2202 else if (unlikely(!sock->ops->getsockopt))
2203 err = -EOPNOTSUPP;
2204 else
2205 err = sock->ops->getsockopt(sock, level, optname, optval,
2206 optlen);
2207
2208 if (!in_compat_syscall())
2209 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2210 optval, optlen, max_optlen,
2211 err);
2212out_put:
2213 fput_light(sock->file, fput_needed);
2214 return err;
2215}
2216
2217SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2218 char __user *, optval, int __user *, optlen)
2219{
2220 return __sys_getsockopt(fd, level, optname, optval, optlen);
2221}
2222
2223/*
2224 * Shutdown a socket.
2225 */
2226
2227int __sys_shutdown_sock(struct socket *sock, int how)
2228{
2229 int err;
2230
2231 err = security_socket_shutdown(sock, how);
2232 if (!err)
2233 err = sock->ops->shutdown(sock, how);
2234
2235 return err;
2236}
2237
2238int __sys_shutdown(int fd, int how)
2239{
2240 int err, fput_needed;
2241 struct socket *sock;
2242
2243 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2244 if (sock != NULL) {
2245 err = __sys_shutdown_sock(sock, how);
2246 fput_light(sock->file, fput_needed);
2247 }
2248 return err;
2249}
2250
2251SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2252{
2253 return __sys_shutdown(fd, how);
2254}
2255
2256/* A couple of helpful macros for getting the address of the 32/64 bit
2257 * fields which are the same type (int / unsigned) on our platforms.
2258 */
2259#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2260#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2261#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2262
2263struct used_address {
2264 struct sockaddr_storage name;
2265 unsigned int name_len;
2266};
2267
2268int __copy_msghdr_from_user(struct msghdr *kmsg,
2269 struct user_msghdr __user *umsg,
2270 struct sockaddr __user **save_addr,
2271 struct iovec __user **uiov, size_t *nsegs)
2272{
2273 struct user_msghdr msg;
2274 ssize_t err;
2275
2276 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2277 return -EFAULT;
2278
2279 kmsg->msg_control_is_user = true;
2280 kmsg->msg_control_user = msg.msg_control;
2281 kmsg->msg_controllen = msg.msg_controllen;
2282 kmsg->msg_flags = msg.msg_flags;
2283
2284 kmsg->msg_namelen = msg.msg_namelen;
2285 if (!msg.msg_name)
2286 kmsg->msg_namelen = 0;
2287
2288 if (kmsg->msg_namelen < 0)
2289 return -EINVAL;
2290
2291 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2292 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2293
2294 if (save_addr)
2295 *save_addr = msg.msg_name;
2296
2297 if (msg.msg_name && kmsg->msg_namelen) {
2298 if (!save_addr) {
2299 err = move_addr_to_kernel(msg.msg_name,
2300 kmsg->msg_namelen,
2301 kmsg->msg_name);
2302 if (err < 0)
2303 return err;
2304 }
2305 } else {
2306 kmsg->msg_name = NULL;
2307 kmsg->msg_namelen = 0;
2308 }
2309
2310 if (msg.msg_iovlen > UIO_MAXIOV)
2311 return -EMSGSIZE;
2312
2313 kmsg->msg_iocb = NULL;
2314 *uiov = msg.msg_iov;
2315 *nsegs = msg.msg_iovlen;
2316 return 0;
2317}
2318
2319static int copy_msghdr_from_user(struct msghdr *kmsg,
2320 struct user_msghdr __user *umsg,
2321 struct sockaddr __user **save_addr,
2322 struct iovec **iov)
2323{
2324 struct user_msghdr msg;
2325 ssize_t err;
2326
2327 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2328 &msg.msg_iovlen);
2329 if (err)
2330 return err;
2331
2332 err = import_iovec(save_addr ? READ : WRITE,
2333 msg.msg_iov, msg.msg_iovlen,
2334 UIO_FASTIOV, iov, &kmsg->msg_iter);
2335 return err < 0 ? err : 0;
2336}
2337
2338static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2339 unsigned int flags, struct used_address *used_address,
2340 unsigned int allowed_msghdr_flags)
2341{
2342 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2343 __aligned(sizeof(__kernel_size_t));
2344 /* 20 is size of ipv6_pktinfo */
2345 unsigned char *ctl_buf = ctl;
2346 int ctl_len;
2347 ssize_t err;
2348
2349 err = -ENOBUFS;
2350
2351 if (msg_sys->msg_controllen > INT_MAX)
2352 goto out;
2353 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2354 ctl_len = msg_sys->msg_controllen;
2355 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2356 err =
2357 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2358 sizeof(ctl));
2359 if (err)
2360 goto out;
2361 ctl_buf = msg_sys->msg_control;
2362 ctl_len = msg_sys->msg_controllen;
2363 } else if (ctl_len) {
2364 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2365 CMSG_ALIGN(sizeof(struct cmsghdr)));
2366 if (ctl_len > sizeof(ctl)) {
2367 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2368 if (ctl_buf == NULL)
2369 goto out;
2370 }
2371 err = -EFAULT;
2372 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2373 goto out_freectl;
2374 msg_sys->msg_control = ctl_buf;
2375 msg_sys->msg_control_is_user = false;
2376 }
2377 msg_sys->msg_flags = flags;
2378
2379 if (sock->file->f_flags & O_NONBLOCK)
2380 msg_sys->msg_flags |= MSG_DONTWAIT;
2381 /*
2382 * If this is sendmmsg() and current destination address is same as
2383 * previously succeeded address, omit asking LSM's decision.
2384 * used_address->name_len is initialized to UINT_MAX so that the first
2385 * destination address never matches.
2386 */
2387 if (used_address && msg_sys->msg_name &&
2388 used_address->name_len == msg_sys->msg_namelen &&
2389 !memcmp(&used_address->name, msg_sys->msg_name,
2390 used_address->name_len)) {
2391 err = sock_sendmsg_nosec(sock, msg_sys);
2392 goto out_freectl;
2393 }
2394 err = sock_sendmsg(sock, msg_sys);
2395 /*
2396 * If this is sendmmsg() and sending to current destination address was
2397 * successful, remember it.
2398 */
2399 if (used_address && err >= 0) {
2400 used_address->name_len = msg_sys->msg_namelen;
2401 if (msg_sys->msg_name)
2402 memcpy(&used_address->name, msg_sys->msg_name,
2403 used_address->name_len);
2404 }
2405
2406out_freectl:
2407 if (ctl_buf != ctl)
2408 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2409out:
2410 return err;
2411}
2412
2413int sendmsg_copy_msghdr(struct msghdr *msg,
2414 struct user_msghdr __user *umsg, unsigned flags,
2415 struct iovec **iov)
2416{
2417 int err;
2418
2419 if (flags & MSG_CMSG_COMPAT) {
2420 struct compat_msghdr __user *msg_compat;
2421
2422 msg_compat = (struct compat_msghdr __user *) umsg;
2423 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2424 } else {
2425 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2426 }
2427 if (err < 0)
2428 return err;
2429
2430 return 0;
2431}
2432
2433static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2434 struct msghdr *msg_sys, unsigned int flags,
2435 struct used_address *used_address,
2436 unsigned int allowed_msghdr_flags)
2437{
2438 struct sockaddr_storage address;
2439 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2440 ssize_t err;
2441
2442 msg_sys->msg_name = &address;
2443
2444 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2445 if (err < 0)
2446 return err;
2447
2448 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2449 allowed_msghdr_flags);
2450 kfree(iov);
2451 return err;
2452}
2453
2454/*
2455 * BSD sendmsg interface
2456 */
2457long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2458 unsigned int flags)
2459{
2460 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2461}
2462
2463long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2464 bool forbid_cmsg_compat)
2465{
2466 int fput_needed, err;
2467 struct msghdr msg_sys;
2468 struct socket *sock;
2469
2470 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2471 return -EINVAL;
2472
2473 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2474 if (!sock)
2475 goto out;
2476
2477 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2478
2479 fput_light(sock->file, fput_needed);
2480out:
2481 return err;
2482}
2483
2484SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2485{
2486 return __sys_sendmsg(fd, msg, flags, true);
2487}
2488
2489/*
2490 * Linux sendmmsg interface
2491 */
2492
2493int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2494 unsigned int flags, bool forbid_cmsg_compat)
2495{
2496 int fput_needed, err, datagrams;
2497 struct socket *sock;
2498 struct mmsghdr __user *entry;
2499 struct compat_mmsghdr __user *compat_entry;
2500 struct msghdr msg_sys;
2501 struct used_address used_address;
2502 unsigned int oflags = flags;
2503
2504 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2505 return -EINVAL;
2506
2507 if (vlen > UIO_MAXIOV)
2508 vlen = UIO_MAXIOV;
2509
2510 datagrams = 0;
2511
2512 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2513 if (!sock)
2514 return err;
2515
2516 used_address.name_len = UINT_MAX;
2517 entry = mmsg;
2518 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2519 err = 0;
2520 flags |= MSG_BATCH;
2521
2522 while (datagrams < vlen) {
2523 if (datagrams == vlen - 1)
2524 flags = oflags;
2525
2526 if (MSG_CMSG_COMPAT & flags) {
2527 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2528 &msg_sys, flags, &used_address, MSG_EOR);
2529 if (err < 0)
2530 break;
2531 err = __put_user(err, &compat_entry->msg_len);
2532 ++compat_entry;
2533 } else {
2534 err = ___sys_sendmsg(sock,
2535 (struct user_msghdr __user *)entry,
2536 &msg_sys, flags, &used_address, MSG_EOR);
2537 if (err < 0)
2538 break;
2539 err = put_user(err, &entry->msg_len);
2540 ++entry;
2541 }
2542
2543 if (err)
2544 break;
2545 ++datagrams;
2546 if (msg_data_left(&msg_sys))
2547 break;
2548 cond_resched();
2549 }
2550
2551 fput_light(sock->file, fput_needed);
2552
2553 /* We only return an error if no datagrams were able to be sent */
2554 if (datagrams != 0)
2555 return datagrams;
2556
2557 return err;
2558}
2559
2560SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2561 unsigned int, vlen, unsigned int, flags)
2562{
2563 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2564}
2565
2566int recvmsg_copy_msghdr(struct msghdr *msg,
2567 struct user_msghdr __user *umsg, unsigned flags,
2568 struct sockaddr __user **uaddr,
2569 struct iovec **iov)
2570{
2571 ssize_t err;
2572
2573 if (MSG_CMSG_COMPAT & flags) {
2574 struct compat_msghdr __user *msg_compat;
2575
2576 msg_compat = (struct compat_msghdr __user *) umsg;
2577 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2578 } else {
2579 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2580 }
2581 if (err < 0)
2582 return err;
2583
2584 return 0;
2585}
2586
2587static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2588 struct user_msghdr __user *msg,
2589 struct sockaddr __user *uaddr,
2590 unsigned int flags, int nosec)
2591{
2592 struct compat_msghdr __user *msg_compat =
2593 (struct compat_msghdr __user *) msg;
2594 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2595 struct sockaddr_storage addr;
2596 unsigned long cmsg_ptr;
2597 int len;
2598 ssize_t err;
2599
2600 msg_sys->msg_name = &addr;
2601 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2602 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2603
2604 /* We assume all kernel code knows the size of sockaddr_storage */
2605 msg_sys->msg_namelen = 0;
2606
2607 if (sock->file->f_flags & O_NONBLOCK)
2608 flags |= MSG_DONTWAIT;
2609
2610 if (unlikely(nosec))
2611 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2612 else
2613 err = sock_recvmsg(sock, msg_sys, flags);
2614
2615 if (err < 0)
2616 goto out;
2617 len = err;
2618
2619 if (uaddr != NULL) {
2620 err = move_addr_to_user(&addr,
2621 msg_sys->msg_namelen, uaddr,
2622 uaddr_len);
2623 if (err < 0)
2624 goto out;
2625 }
2626 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2627 COMPAT_FLAGS(msg));
2628 if (err)
2629 goto out;
2630 if (MSG_CMSG_COMPAT & flags)
2631 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2632 &msg_compat->msg_controllen);
2633 else
2634 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2635 &msg->msg_controllen);
2636 if (err)
2637 goto out;
2638 err = len;
2639out:
2640 return err;
2641}
2642
2643static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2644 struct msghdr *msg_sys, unsigned int flags, int nosec)
2645{
2646 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2647 /* user mode address pointers */
2648 struct sockaddr __user *uaddr;
2649 ssize_t err;
2650
2651 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2652 if (err < 0)
2653 return err;
2654
2655 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2656 kfree(iov);
2657 return err;
2658}
2659
2660/*
2661 * BSD recvmsg interface
2662 */
2663
2664long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2665 struct user_msghdr __user *umsg,
2666 struct sockaddr __user *uaddr, unsigned int flags)
2667{
2668 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2669}
2670
2671long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2672 bool forbid_cmsg_compat)
2673{
2674 int fput_needed, err;
2675 struct msghdr msg_sys;
2676 struct socket *sock;
2677
2678 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2679 return -EINVAL;
2680
2681 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2682 if (!sock)
2683 goto out;
2684
2685 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2686
2687 fput_light(sock->file, fput_needed);
2688out:
2689 return err;
2690}
2691
2692SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2693 unsigned int, flags)
2694{
2695 return __sys_recvmsg(fd, msg, flags, true);
2696}
2697
2698/*
2699 * Linux recvmmsg interface
2700 */
2701
2702static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2703 unsigned int vlen, unsigned int flags,
2704 struct timespec64 *timeout)
2705{
2706 int fput_needed, err, datagrams;
2707 struct socket *sock;
2708 struct mmsghdr __user *entry;
2709 struct compat_mmsghdr __user *compat_entry;
2710 struct msghdr msg_sys;
2711 struct timespec64 end_time;
2712 struct timespec64 timeout64;
2713
2714 if (timeout &&
2715 poll_select_set_timeout(&end_time, timeout->tv_sec,
2716 timeout->tv_nsec))
2717 return -EINVAL;
2718
2719 datagrams = 0;
2720
2721 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2722 if (!sock)
2723 return err;
2724
2725 if (likely(!(flags & MSG_ERRQUEUE))) {
2726 err = sock_error(sock->sk);
2727 if (err) {
2728 datagrams = err;
2729 goto out_put;
2730 }
2731 }
2732
2733 entry = mmsg;
2734 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2735
2736 while (datagrams < vlen) {
2737 /*
2738 * No need to ask LSM for more than the first datagram.
2739 */
2740 if (MSG_CMSG_COMPAT & flags) {
2741 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2742 &msg_sys, flags & ~MSG_WAITFORONE,
2743 datagrams);
2744 if (err < 0)
2745 break;
2746 err = __put_user(err, &compat_entry->msg_len);
2747 ++compat_entry;
2748 } else {
2749 err = ___sys_recvmsg(sock,
2750 (struct user_msghdr __user *)entry,
2751 &msg_sys, flags & ~MSG_WAITFORONE,
2752 datagrams);
2753 if (err < 0)
2754 break;
2755 err = put_user(err, &entry->msg_len);
2756 ++entry;
2757 }
2758
2759 if (err)
2760 break;
2761 ++datagrams;
2762
2763 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2764 if (flags & MSG_WAITFORONE)
2765 flags |= MSG_DONTWAIT;
2766
2767 if (timeout) {
2768 ktime_get_ts64(&timeout64);
2769 *timeout = timespec64_sub(end_time, timeout64);
2770 if (timeout->tv_sec < 0) {
2771 timeout->tv_sec = timeout->tv_nsec = 0;
2772 break;
2773 }
2774
2775 /* Timeout, return less than vlen datagrams */
2776 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2777 break;
2778 }
2779
2780 /* Out of band data, return right away */
2781 if (msg_sys.msg_flags & MSG_OOB)
2782 break;
2783 cond_resched();
2784 }
2785
2786 if (err == 0)
2787 goto out_put;
2788
2789 if (datagrams == 0) {
2790 datagrams = err;
2791 goto out_put;
2792 }
2793
2794 /*
2795 * We may return less entries than requested (vlen) if the
2796 * sock is non block and there aren't enough datagrams...
2797 */
2798 if (err != -EAGAIN) {
2799 /*
2800 * ... or if recvmsg returns an error after we
2801 * received some datagrams, where we record the
2802 * error to return on the next call or if the
2803 * app asks about it using getsockopt(SO_ERROR).
2804 */
2805 sock->sk->sk_err = -err;
2806 }
2807out_put:
2808 fput_light(sock->file, fput_needed);
2809
2810 return datagrams;
2811}
2812
2813int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2814 unsigned int vlen, unsigned int flags,
2815 struct __kernel_timespec __user *timeout,
2816 struct old_timespec32 __user *timeout32)
2817{
2818 int datagrams;
2819 struct timespec64 timeout_sys;
2820
2821 if (timeout && get_timespec64(&timeout_sys, timeout))
2822 return -EFAULT;
2823
2824 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2825 return -EFAULT;
2826
2827 if (!timeout && !timeout32)
2828 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2829
2830 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2831
2832 if (datagrams <= 0)
2833 return datagrams;
2834
2835 if (timeout && put_timespec64(&timeout_sys, timeout))
2836 datagrams = -EFAULT;
2837
2838 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2839 datagrams = -EFAULT;
2840
2841 return datagrams;
2842}
2843
2844SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2845 unsigned int, vlen, unsigned int, flags,
2846 struct __kernel_timespec __user *, timeout)
2847{
2848 if (flags & MSG_CMSG_COMPAT)
2849 return -EINVAL;
2850
2851 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2852}
2853
2854#ifdef CONFIG_COMPAT_32BIT_TIME
2855SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2856 unsigned int, vlen, unsigned int, flags,
2857 struct old_timespec32 __user *, timeout)
2858{
2859 if (flags & MSG_CMSG_COMPAT)
2860 return -EINVAL;
2861
2862 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2863}
2864#endif
2865
2866#ifdef __ARCH_WANT_SYS_SOCKETCALL
2867/* Argument list sizes for sys_socketcall */
2868#define AL(x) ((x) * sizeof(unsigned long))
2869static const unsigned char nargs[21] = {
2870 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2871 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2872 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2873 AL(4), AL(5), AL(4)
2874};
2875
2876#undef AL
2877
2878/*
2879 * System call vectors.
2880 *
2881 * Argument checking cleaned up. Saved 20% in size.
2882 * This function doesn't need to set the kernel lock because
2883 * it is set by the callees.
2884 */
2885
2886SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2887{
2888 unsigned long a[AUDITSC_ARGS];
2889 unsigned long a0, a1;
2890 int err;
2891 unsigned int len;
2892
2893 if (call < 1 || call > SYS_SENDMMSG)
2894 return -EINVAL;
2895 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2896
2897 len = nargs[call];
2898 if (len > sizeof(a))
2899 return -EINVAL;
2900
2901 /* copy_from_user should be SMP safe. */
2902 if (copy_from_user(a, args, len))
2903 return -EFAULT;
2904
2905 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2906 if (err)
2907 return err;
2908
2909 a0 = a[0];
2910 a1 = a[1];
2911
2912 switch (call) {
2913 case SYS_SOCKET:
2914 err = __sys_socket(a0, a1, a[2]);
2915 break;
2916 case SYS_BIND:
2917 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2918 break;
2919 case SYS_CONNECT:
2920 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2921 break;
2922 case SYS_LISTEN:
2923 err = __sys_listen(a0, a1);
2924 break;
2925 case SYS_ACCEPT:
2926 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2927 (int __user *)a[2], 0);
2928 break;
2929 case SYS_GETSOCKNAME:
2930 err =
2931 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2932 (int __user *)a[2]);
2933 break;
2934 case SYS_GETPEERNAME:
2935 err =
2936 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2937 (int __user *)a[2]);
2938 break;
2939 case SYS_SOCKETPAIR:
2940 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2941 break;
2942 case SYS_SEND:
2943 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2944 NULL, 0);
2945 break;
2946 case SYS_SENDTO:
2947 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2948 (struct sockaddr __user *)a[4], a[5]);
2949 break;
2950 case SYS_RECV:
2951 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2952 NULL, NULL);
2953 break;
2954 case SYS_RECVFROM:
2955 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2956 (struct sockaddr __user *)a[4],
2957 (int __user *)a[5]);
2958 break;
2959 case SYS_SHUTDOWN:
2960 err = __sys_shutdown(a0, a1);
2961 break;
2962 case SYS_SETSOCKOPT:
2963 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2964 a[4]);
2965 break;
2966 case SYS_GETSOCKOPT:
2967 err =
2968 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2969 (int __user *)a[4]);
2970 break;
2971 case SYS_SENDMSG:
2972 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2973 a[2], true);
2974 break;
2975 case SYS_SENDMMSG:
2976 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2977 a[3], true);
2978 break;
2979 case SYS_RECVMSG:
2980 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2981 a[2], true);
2982 break;
2983 case SYS_RECVMMSG:
2984 if (IS_ENABLED(CONFIG_64BIT))
2985 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2986 a[2], a[3],
2987 (struct __kernel_timespec __user *)a[4],
2988 NULL);
2989 else
2990 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2991 a[2], a[3], NULL,
2992 (struct old_timespec32 __user *)a[4]);
2993 break;
2994 case SYS_ACCEPT4:
2995 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2996 (int __user *)a[2], a[3]);
2997 break;
2998 default:
2999 err = -EINVAL;
3000 break;
3001 }
3002 return err;
3003}
3004
3005#endif /* __ARCH_WANT_SYS_SOCKETCALL */
3006
3007/**
3008 * sock_register - add a socket protocol handler
3009 * @ops: description of protocol
3010 *
3011 * This function is called by a protocol handler that wants to
3012 * advertise its address family, and have it linked into the
3013 * socket interface. The value ops->family corresponds to the
3014 * socket system call protocol family.
3015 */
3016int sock_register(const struct net_proto_family *ops)
3017{
3018 int err;
3019
3020 if (ops->family >= NPROTO) {
3021 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3022 return -ENOBUFS;
3023 }
3024
3025 spin_lock(&net_family_lock);
3026 if (rcu_dereference_protected(net_families[ops->family],
3027 lockdep_is_held(&net_family_lock)))
3028 err = -EEXIST;
3029 else {
3030 rcu_assign_pointer(net_families[ops->family], ops);
3031 err = 0;
3032 }
3033 spin_unlock(&net_family_lock);
3034
3035 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3036 return err;
3037}
3038EXPORT_SYMBOL(sock_register);
3039
3040/**
3041 * sock_unregister - remove a protocol handler
3042 * @family: protocol family to remove
3043 *
3044 * This function is called by a protocol handler that wants to
3045 * remove its address family, and have it unlinked from the
3046 * new socket creation.
3047 *
3048 * If protocol handler is a module, then it can use module reference
3049 * counts to protect against new references. If protocol handler is not
3050 * a module then it needs to provide its own protection in
3051 * the ops->create routine.
3052 */
3053void sock_unregister(int family)
3054{
3055 BUG_ON(family < 0 || family >= NPROTO);
3056
3057 spin_lock(&net_family_lock);
3058 RCU_INIT_POINTER(net_families[family], NULL);
3059 spin_unlock(&net_family_lock);
3060
3061 synchronize_rcu();
3062
3063 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3064}
3065EXPORT_SYMBOL(sock_unregister);
3066
3067bool sock_is_registered(int family)
3068{
3069 return family < NPROTO && rcu_access_pointer(net_families[family]);
3070}
3071
3072static int __init sock_init(void)
3073{
3074 int err;
3075 /*
3076 * Initialize the network sysctl infrastructure.
3077 */
3078 err = net_sysctl_init();
3079 if (err)
3080 goto out;
3081
3082 /*
3083 * Initialize skbuff SLAB cache
3084 */
3085 skb_init();
3086
3087 /*
3088 * Initialize the protocols module.
3089 */
3090
3091 init_inodecache();
3092
3093 err = register_filesystem(&sock_fs_type);
3094 if (err)
3095 goto out;
3096 sock_mnt = kern_mount(&sock_fs_type);
3097 if (IS_ERR(sock_mnt)) {
3098 err = PTR_ERR(sock_mnt);
3099 goto out_mount;
3100 }
3101
3102 /* The real protocol initialization is performed in later initcalls.
3103 */
3104
3105#ifdef CONFIG_NETFILTER
3106 err = netfilter_init();
3107 if (err)
3108 goto out;
3109#endif
3110
3111 ptp_classifier_init();
3112
3113out:
3114 return err;
3115
3116out_mount:
3117 unregister_filesystem(&sock_fs_type);
3118 goto out;
3119}
3120
3121core_initcall(sock_init); /* early initcall */
3122
3123#ifdef CONFIG_PROC_FS
3124void socket_seq_show(struct seq_file *seq)
3125{
3126 seq_printf(seq, "sockets: used %d\n",
3127 sock_inuse_get(seq->private));
3128}
3129#endif /* CONFIG_PROC_FS */
3130
3131#ifdef CONFIG_COMPAT
3132static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
3133{
3134 struct compat_ifconf ifc32;
3135 struct ifconf ifc;
3136 int err;
3137
3138 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
3139 return -EFAULT;
3140
3141 ifc.ifc_len = ifc32.ifc_len;
3142 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
3143
3144 rtnl_lock();
3145 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
3146 rtnl_unlock();
3147 if (err)
3148 return err;
3149
3150 ifc32.ifc_len = ifc.ifc_len;
3151 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
3152 return -EFAULT;
3153
3154 return 0;
3155}
3156
3157static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3158{
3159 compat_uptr_t uptr32;
3160 struct ifreq ifr;
3161 void __user *saved;
3162 int err;
3163
3164 if (copy_from_user(&ifr, uifr32, sizeof(struct compat_ifreq)))
3165 return -EFAULT;
3166
3167 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3168 return -EFAULT;
3169
3170 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3171 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3172
3173 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL);
3174 if (!err) {
3175 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3176 if (copy_to_user(uifr32, &ifr, sizeof(struct compat_ifreq)))
3177 err = -EFAULT;
3178 }
3179 return err;
3180}
3181
3182/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3183static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3184 struct compat_ifreq __user *u_ifreq32)
3185{
3186 struct ifreq ifreq;
3187 u32 data32;
3188
3189 if (!is_socket_ioctl_cmd(cmd))
3190 return -ENOTTY;
3191 if (copy_from_user(ifreq.ifr_name, u_ifreq32->ifr_name, IFNAMSIZ))
3192 return -EFAULT;
3193 if (get_user(data32, &u_ifreq32->ifr_data))
3194 return -EFAULT;
3195 ifreq.ifr_data = compat_ptr(data32);
3196
3197 return dev_ioctl(net, cmd, &ifreq, NULL);
3198}
3199
3200static int compat_ifreq_ioctl(struct net *net, struct socket *sock,
3201 unsigned int cmd,
3202 struct compat_ifreq __user *uifr32)
3203{
3204 struct ifreq __user *uifr;
3205 int err;
3206
3207 /* Handle the fact that while struct ifreq has the same *layout* on
3208 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3209 * which are handled elsewhere, it still has different *size* due to
3210 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3211 * resulting in struct ifreq being 32 and 40 bytes respectively).
3212 * As a result, if the struct happens to be at the end of a page and
3213 * the next page isn't readable/writable, we get a fault. To prevent
3214 * that, copy back and forth to the full size.
3215 */
3216
3217 uifr = compat_alloc_user_space(sizeof(*uifr));
3218 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
3219 return -EFAULT;
3220
3221 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
3222
3223 if (!err) {
3224 switch (cmd) {
3225 case SIOCGIFFLAGS:
3226 case SIOCGIFMETRIC:
3227 case SIOCGIFMTU:
3228 case SIOCGIFMEM:
3229 case SIOCGIFHWADDR:
3230 case SIOCGIFINDEX:
3231 case SIOCGIFADDR:
3232 case SIOCGIFBRDADDR:
3233 case SIOCGIFDSTADDR:
3234 case SIOCGIFNETMASK:
3235 case SIOCGIFPFLAGS:
3236 case SIOCGIFTXQLEN:
3237 case SIOCGMIIPHY:
3238 case SIOCGMIIREG:
3239 case SIOCGIFNAME:
3240 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
3241 err = -EFAULT;
3242 break;
3243 }
3244 }
3245 return err;
3246}
3247
3248static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
3249 struct compat_ifreq __user *uifr32)
3250{
3251 struct ifreq ifr;
3252 struct compat_ifmap __user *uifmap32;
3253 int err;
3254
3255 uifmap32 = &uifr32->ifr_ifru.ifru_map;
3256 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
3257 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3258 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3259 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3260 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
3261 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
3262 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
3263 if (err)
3264 return -EFAULT;
3265
3266 err = dev_ioctl(net, cmd, &ifr, NULL);
3267
3268 if (cmd == SIOCGIFMAP && !err) {
3269 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
3270 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3271 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3272 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3273 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
3274 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
3275 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
3276 if (err)
3277 err = -EFAULT;
3278 }
3279 return err;
3280}
3281
3282/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3283 * for some operations; this forces use of the newer bridge-utils that
3284 * use compatible ioctls
3285 */
3286static int old_bridge_ioctl(compat_ulong_t __user *argp)
3287{
3288 compat_ulong_t tmp;
3289
3290 if (get_user(tmp, argp))
3291 return -EFAULT;
3292 if (tmp == BRCTL_GET_VERSION)
3293 return BRCTL_VERSION + 1;
3294 return -EINVAL;
3295}
3296
3297static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3298 unsigned int cmd, unsigned long arg)
3299{
3300 void __user *argp = compat_ptr(arg);
3301 struct sock *sk = sock->sk;
3302 struct net *net = sock_net(sk);
3303
3304 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3305 return compat_ifr_data_ioctl(net, cmd, argp);
3306
3307 switch (cmd) {
3308 case SIOCSIFBR:
3309 case SIOCGIFBR:
3310 return old_bridge_ioctl(argp);
3311 case SIOCGIFCONF:
3312 return compat_dev_ifconf(net, argp);
3313 case SIOCWANDEV:
3314 return compat_siocwandev(net, argp);
3315 case SIOCGIFMAP:
3316 case SIOCSIFMAP:
3317 return compat_sioc_ifmap(net, cmd, argp);
3318 case SIOCGSTAMP_OLD:
3319 case SIOCGSTAMPNS_OLD:
3320 if (!sock->ops->gettstamp)
3321 return -ENOIOCTLCMD;
3322 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3323 !COMPAT_USE_64BIT_TIME);
3324
3325 case SIOCETHTOOL:
3326 case SIOCBONDSLAVEINFOQUERY:
3327 case SIOCBONDINFOQUERY:
3328 case SIOCSHWTSTAMP:
3329 case SIOCGHWTSTAMP:
3330 return compat_ifr_data_ioctl(net, cmd, argp);
3331
3332 case FIOSETOWN:
3333 case SIOCSPGRP:
3334 case FIOGETOWN:
3335 case SIOCGPGRP:
3336 case SIOCBRADDBR:
3337 case SIOCBRDELBR:
3338 case SIOCGIFVLAN:
3339 case SIOCSIFVLAN:
3340 case SIOCGSKNS:
3341 case SIOCGSTAMP_NEW:
3342 case SIOCGSTAMPNS_NEW:
3343 return sock_ioctl(file, cmd, arg);
3344
3345 case SIOCGIFFLAGS:
3346 case SIOCSIFFLAGS:
3347 case SIOCGIFMETRIC:
3348 case SIOCSIFMETRIC:
3349 case SIOCGIFMTU:
3350 case SIOCSIFMTU:
3351 case SIOCGIFMEM:
3352 case SIOCSIFMEM:
3353 case SIOCGIFHWADDR:
3354 case SIOCSIFHWADDR:
3355 case SIOCADDMULTI:
3356 case SIOCDELMULTI:
3357 case SIOCGIFINDEX:
3358 case SIOCGIFADDR:
3359 case SIOCSIFADDR:
3360 case SIOCSIFHWBROADCAST:
3361 case SIOCDIFADDR:
3362 case SIOCGIFBRDADDR:
3363 case SIOCSIFBRDADDR:
3364 case SIOCGIFDSTADDR:
3365 case SIOCSIFDSTADDR:
3366 case SIOCGIFNETMASK:
3367 case SIOCSIFNETMASK:
3368 case SIOCSIFPFLAGS:
3369 case SIOCGIFPFLAGS:
3370 case SIOCGIFTXQLEN:
3371 case SIOCSIFTXQLEN:
3372 case SIOCBRADDIF:
3373 case SIOCBRDELIF:
3374 case SIOCGIFNAME:
3375 case SIOCSIFNAME:
3376 case SIOCGMIIPHY:
3377 case SIOCGMIIREG:
3378 case SIOCSMIIREG:
3379 case SIOCBONDENSLAVE:
3380 case SIOCBONDRELEASE:
3381 case SIOCBONDSETHWADDR:
3382 case SIOCBONDCHANGEACTIVE:
3383 return compat_ifreq_ioctl(net, sock, cmd, argp);
3384
3385 case SIOCSARP:
3386 case SIOCGARP:
3387 case SIOCDARP:
3388 case SIOCOUTQ:
3389 case SIOCOUTQNSD:
3390 case SIOCATMARK:
3391 return sock_do_ioctl(net, sock, cmd, arg);
3392 }
3393
3394 return -ENOIOCTLCMD;
3395}
3396
3397static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3398 unsigned long arg)
3399{
3400 struct socket *sock = file->private_data;
3401 int ret = -ENOIOCTLCMD;
3402 struct sock *sk;
3403 struct net *net;
3404
3405 sk = sock->sk;
3406 net = sock_net(sk);
3407
3408 if (sock->ops->compat_ioctl)
3409 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3410
3411 if (ret == -ENOIOCTLCMD &&
3412 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3413 ret = compat_wext_handle_ioctl(net, cmd, arg);
3414
3415 if (ret == -ENOIOCTLCMD)
3416 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3417
3418 return ret;
3419}
3420#endif
3421
3422/**
3423 * kernel_bind - bind an address to a socket (kernel space)
3424 * @sock: socket
3425 * @addr: address
3426 * @addrlen: length of address
3427 *
3428 * Returns 0 or an error.
3429 */
3430
3431int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3432{
3433 return sock->ops->bind(sock, addr, addrlen);
3434}
3435EXPORT_SYMBOL(kernel_bind);
3436
3437/**
3438 * kernel_listen - move socket to listening state (kernel space)
3439 * @sock: socket
3440 * @backlog: pending connections queue size
3441 *
3442 * Returns 0 or an error.
3443 */
3444
3445int kernel_listen(struct socket *sock, int backlog)
3446{
3447 return sock->ops->listen(sock, backlog);
3448}
3449EXPORT_SYMBOL(kernel_listen);
3450
3451/**
3452 * kernel_accept - accept a connection (kernel space)
3453 * @sock: listening socket
3454 * @newsock: new connected socket
3455 * @flags: flags
3456 *
3457 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3458 * If it fails, @newsock is guaranteed to be %NULL.
3459 * Returns 0 or an error.
3460 */
3461
3462int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3463{
3464 struct sock *sk = sock->sk;
3465 int err;
3466
3467 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3468 newsock);
3469 if (err < 0)
3470 goto done;
3471
3472 err = sock->ops->accept(sock, *newsock, flags, true);
3473 if (err < 0) {
3474 sock_release(*newsock);
3475 *newsock = NULL;
3476 goto done;
3477 }
3478
3479 (*newsock)->ops = sock->ops;
3480 __module_get((*newsock)->ops->owner);
3481
3482done:
3483 return err;
3484}
3485EXPORT_SYMBOL(kernel_accept);
3486
3487/**
3488 * kernel_connect - connect a socket (kernel space)
3489 * @sock: socket
3490 * @addr: address
3491 * @addrlen: address length
3492 * @flags: flags (O_NONBLOCK, ...)
3493 *
3494 * For datagram sockets, @addr is the address to which datagrams are sent
3495 * by default, and the only address from which datagrams are received.
3496 * For stream sockets, attempts to connect to @addr.
3497 * Returns 0 or an error code.
3498 */
3499
3500int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3501 int flags)
3502{
3503 return sock->ops->connect(sock, addr, addrlen, flags);
3504}
3505EXPORT_SYMBOL(kernel_connect);
3506
3507/**
3508 * kernel_getsockname - get the address which the socket is bound (kernel space)
3509 * @sock: socket
3510 * @addr: address holder
3511 *
3512 * Fills the @addr pointer with the address which the socket is bound.
3513 * Returns 0 or an error code.
3514 */
3515
3516int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3517{
3518 return sock->ops->getname(sock, addr, 0);
3519}
3520EXPORT_SYMBOL(kernel_getsockname);
3521
3522/**
3523 * kernel_getpeername - get the address which the socket is connected (kernel space)
3524 * @sock: socket
3525 * @addr: address holder
3526 *
3527 * Fills the @addr pointer with the address which the socket is connected.
3528 * Returns 0 or an error code.
3529 */
3530
3531int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3532{
3533 return sock->ops->getname(sock, addr, 1);
3534}
3535EXPORT_SYMBOL(kernel_getpeername);
3536
3537/**
3538 * kernel_sendpage - send a &page through a socket (kernel space)
3539 * @sock: socket
3540 * @page: page
3541 * @offset: page offset
3542 * @size: total size in bytes
3543 * @flags: flags (MSG_DONTWAIT, ...)
3544 *
3545 * Returns the total amount sent in bytes or an error.
3546 */
3547
3548int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3549 size_t size, int flags)
3550{
3551 if (sock->ops->sendpage) {
3552 /* Warn in case the improper page to zero-copy send */
3553 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3554 return sock->ops->sendpage(sock, page, offset, size, flags);
3555 }
3556 return sock_no_sendpage(sock, page, offset, size, flags);
3557}
3558EXPORT_SYMBOL(kernel_sendpage);
3559
3560/**
3561 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3562 * @sk: sock
3563 * @page: page
3564 * @offset: page offset
3565 * @size: total size in bytes
3566 * @flags: flags (MSG_DONTWAIT, ...)
3567 *
3568 * Returns the total amount sent in bytes or an error.
3569 * Caller must hold @sk.
3570 */
3571
3572int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3573 size_t size, int flags)
3574{
3575 struct socket *sock = sk->sk_socket;
3576
3577 if (sock->ops->sendpage_locked)
3578 return sock->ops->sendpage_locked(sk, page, offset, size,
3579 flags);
3580
3581 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3582}
3583EXPORT_SYMBOL(kernel_sendpage_locked);
3584
3585/**
3586 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3587 * @sock: socket
3588 * @how: connection part
3589 *
3590 * Returns 0 or an error.
3591 */
3592
3593int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3594{
3595 return sock->ops->shutdown(sock, how);
3596}
3597EXPORT_SYMBOL(kernel_sock_shutdown);
3598
3599/**
3600 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3601 * @sk: socket
3602 *
3603 * This routine returns the IP overhead imposed by a socket i.e.
3604 * the length of the underlying IP header, depending on whether
3605 * this is an IPv4 or IPv6 socket and the length from IP options turned
3606 * on at the socket. Assumes that the caller has a lock on the socket.
3607 */
3608
3609u32 kernel_sock_ip_overhead(struct sock *sk)
3610{
3611 struct inet_sock *inet;
3612 struct ip_options_rcu *opt;
3613 u32 overhead = 0;
3614#if IS_ENABLED(CONFIG_IPV6)
3615 struct ipv6_pinfo *np;
3616 struct ipv6_txoptions *optv6 = NULL;
3617#endif /* IS_ENABLED(CONFIG_IPV6) */
3618
3619 if (!sk)
3620 return overhead;
3621
3622 switch (sk->sk_family) {
3623 case AF_INET:
3624 inet = inet_sk(sk);
3625 overhead += sizeof(struct iphdr);
3626 opt = rcu_dereference_protected(inet->inet_opt,
3627 sock_owned_by_user(sk));
3628 if (opt)
3629 overhead += opt->opt.optlen;
3630 return overhead;
3631#if IS_ENABLED(CONFIG_IPV6)
3632 case AF_INET6:
3633 np = inet6_sk(sk);
3634 overhead += sizeof(struct ipv6hdr);
3635 if (np)
3636 optv6 = rcu_dereference_protected(np->opt,
3637 sock_owned_by_user(sk));
3638 if (optv6)
3639 overhead += (optv6->opt_flen + optv6->opt_nflen);
3640 return overhead;
3641#endif /* IS_ENABLED(CONFIG_IPV6) */
3642 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3643 return overhead;
3644 }
3645}
3646EXPORT_SYMBOL(kernel_sock_ip_overhead);
1/*
2 * NET An implementation of the SOCKET network access protocol.
3 *
4 * Version: @(#)socket.c 1.1.93 18/02/95
5 *
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
7 * Ross Biro
8 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
9 *
10 * Fixes:
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
12 * shutdown()
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
17 * top level.
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
22 * tty drivers).
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
25 * configurable.
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
34 * stuff.
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
40 * moment.
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
45 * Tigran Aivazian : Made listen(2) backlog sanity checks
46 * protocol-independent
47 *
48 *
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
53 *
54 *
55 * This module is effectively the top level interface to the BSD socket
56 * paradigm.
57 *
58 * Based upon Swansea University Computer Society NET3.039
59 */
60
61#include <linux/mm.h>
62#include <linux/socket.h>
63#include <linux/file.h>
64#include <linux/net.h>
65#include <linux/interrupt.h>
66#include <linux/thread_info.h>
67#include <linux/rcupdate.h>
68#include <linux/netdevice.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
71#include <linux/mutex.h>
72#include <linux/wanrouter.h>
73#include <linux/if_bridge.h>
74#include <linux/if_frad.h>
75#include <linux/if_vlan.h>
76#include <linux/init.h>
77#include <linux/poll.h>
78#include <linux/cache.h>
79#include <linux/module.h>
80#include <linux/highmem.h>
81#include <linux/mount.h>
82#include <linux/security.h>
83#include <linux/syscalls.h>
84#include <linux/compat.h>
85#include <linux/kmod.h>
86#include <linux/audit.h>
87#include <linux/wireless.h>
88#include <linux/nsproxy.h>
89#include <linux/magic.h>
90#include <linux/slab.h>
91
92#include <asm/uaccess.h>
93#include <asm/unistd.h>
94
95#include <net/compat.h>
96#include <net/wext.h>
97#include <net/cls_cgroup.h>
98
99#include <net/sock.h>
100#include <linux/netfilter.h>
101
102#include <linux/if_tun.h>
103#include <linux/ipv6_route.h>
104#include <linux/route.h>
105#include <linux/sockios.h>
106#include <linux/atalk.h>
107
108static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
109static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
110 unsigned long nr_segs, loff_t pos);
111static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
112 unsigned long nr_segs, loff_t pos);
113static int sock_mmap(struct file *file, struct vm_area_struct *vma);
114
115static int sock_close(struct inode *inode, struct file *file);
116static unsigned int sock_poll(struct file *file,
117 struct poll_table_struct *wait);
118static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
119#ifdef CONFIG_COMPAT
120static long compat_sock_ioctl(struct file *file,
121 unsigned int cmd, unsigned long arg);
122#endif
123static int sock_fasync(int fd, struct file *filp, int on);
124static ssize_t sock_sendpage(struct file *file, struct page *page,
125 int offset, size_t size, loff_t *ppos, int more);
126static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
127 struct pipe_inode_info *pipe, size_t len,
128 unsigned int flags);
129
130/*
131 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
132 * in the operation structures but are done directly via the socketcall() multiplexor.
133 */
134
135static const struct file_operations socket_file_ops = {
136 .owner = THIS_MODULE,
137 .llseek = no_llseek,
138 .aio_read = sock_aio_read,
139 .aio_write = sock_aio_write,
140 .poll = sock_poll,
141 .unlocked_ioctl = sock_ioctl,
142#ifdef CONFIG_COMPAT
143 .compat_ioctl = compat_sock_ioctl,
144#endif
145 .mmap = sock_mmap,
146 .open = sock_no_open, /* special open code to disallow open via /proc */
147 .release = sock_close,
148 .fasync = sock_fasync,
149 .sendpage = sock_sendpage,
150 .splice_write = generic_splice_sendpage,
151 .splice_read = sock_splice_read,
152};
153
154/*
155 * The protocol list. Each protocol is registered in here.
156 */
157
158static DEFINE_SPINLOCK(net_family_lock);
159static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
160
161/*
162 * Statistics counters of the socket lists
163 */
164
165static DEFINE_PER_CPU(int, sockets_in_use);
166
167/*
168 * Support routines.
169 * Move socket addresses back and forth across the kernel/user
170 * divide and look after the messy bits.
171 */
172
173/**
174 * move_addr_to_kernel - copy a socket address into kernel space
175 * @uaddr: Address in user space
176 * @kaddr: Address in kernel space
177 * @ulen: Length in user space
178 *
179 * The address is copied into kernel space. If the provided address is
180 * too long an error code of -EINVAL is returned. If the copy gives
181 * invalid addresses -EFAULT is returned. On a success 0 is returned.
182 */
183
184int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
185{
186 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
187 return -EINVAL;
188 if (ulen == 0)
189 return 0;
190 if (copy_from_user(kaddr, uaddr, ulen))
191 return -EFAULT;
192 return audit_sockaddr(ulen, kaddr);
193}
194
195/**
196 * move_addr_to_user - copy an address to user space
197 * @kaddr: kernel space address
198 * @klen: length of address in kernel
199 * @uaddr: user space address
200 * @ulen: pointer to user length field
201 *
202 * The value pointed to by ulen on entry is the buffer length available.
203 * This is overwritten with the buffer space used. -EINVAL is returned
204 * if an overlong buffer is specified or a negative buffer size. -EFAULT
205 * is returned if either the buffer or the length field are not
206 * accessible.
207 * After copying the data up to the limit the user specifies, the true
208 * length of the data is written over the length limit the user
209 * specified. Zero is returned for a success.
210 */
211
212static int move_addr_to_user(struct sockaddr *kaddr, int klen,
213 void __user *uaddr, int __user *ulen)
214{
215 int err;
216 int len;
217
218 err = get_user(len, ulen);
219 if (err)
220 return err;
221 if (len > klen)
222 len = klen;
223 if (len < 0 || len > sizeof(struct sockaddr_storage))
224 return -EINVAL;
225 if (len) {
226 if (audit_sockaddr(klen, kaddr))
227 return -ENOMEM;
228 if (copy_to_user(uaddr, kaddr, len))
229 return -EFAULT;
230 }
231 /*
232 * "fromlen shall refer to the value before truncation.."
233 * 1003.1g
234 */
235 return __put_user(klen, ulen);
236}
237
238static struct kmem_cache *sock_inode_cachep __read_mostly;
239
240static struct inode *sock_alloc_inode(struct super_block *sb)
241{
242 struct socket_alloc *ei;
243 struct socket_wq *wq;
244
245 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
246 if (!ei)
247 return NULL;
248 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
249 if (!wq) {
250 kmem_cache_free(sock_inode_cachep, ei);
251 return NULL;
252 }
253 init_waitqueue_head(&wq->wait);
254 wq->fasync_list = NULL;
255 RCU_INIT_POINTER(ei->socket.wq, wq);
256
257 ei->socket.state = SS_UNCONNECTED;
258 ei->socket.flags = 0;
259 ei->socket.ops = NULL;
260 ei->socket.sk = NULL;
261 ei->socket.file = NULL;
262
263 return &ei->vfs_inode;
264}
265
266static void sock_destroy_inode(struct inode *inode)
267{
268 struct socket_alloc *ei;
269 struct socket_wq *wq;
270
271 ei = container_of(inode, struct socket_alloc, vfs_inode);
272 wq = rcu_dereference_protected(ei->socket.wq, 1);
273 kfree_rcu(wq, rcu);
274 kmem_cache_free(sock_inode_cachep, ei);
275}
276
277static void init_once(void *foo)
278{
279 struct socket_alloc *ei = (struct socket_alloc *)foo;
280
281 inode_init_once(&ei->vfs_inode);
282}
283
284static int init_inodecache(void)
285{
286 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
287 sizeof(struct socket_alloc),
288 0,
289 (SLAB_HWCACHE_ALIGN |
290 SLAB_RECLAIM_ACCOUNT |
291 SLAB_MEM_SPREAD),
292 init_once);
293 if (sock_inode_cachep == NULL)
294 return -ENOMEM;
295 return 0;
296}
297
298static const struct super_operations sockfs_ops = {
299 .alloc_inode = sock_alloc_inode,
300 .destroy_inode = sock_destroy_inode,
301 .statfs = simple_statfs,
302};
303
304/*
305 * sockfs_dname() is called from d_path().
306 */
307static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
308{
309 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
310 dentry->d_inode->i_ino);
311}
312
313static const struct dentry_operations sockfs_dentry_operations = {
314 .d_dname = sockfs_dname,
315};
316
317static struct dentry *sockfs_mount(struct file_system_type *fs_type,
318 int flags, const char *dev_name, void *data)
319{
320 return mount_pseudo(fs_type, "socket:", &sockfs_ops,
321 &sockfs_dentry_operations, SOCKFS_MAGIC);
322}
323
324static struct vfsmount *sock_mnt __read_mostly;
325
326static struct file_system_type sock_fs_type = {
327 .name = "sockfs",
328 .mount = sockfs_mount,
329 .kill_sb = kill_anon_super,
330};
331
332/*
333 * Obtains the first available file descriptor and sets it up for use.
334 *
335 * These functions create file structures and maps them to fd space
336 * of the current process. On success it returns file descriptor
337 * and file struct implicitly stored in sock->file.
338 * Note that another thread may close file descriptor before we return
339 * from this function. We use the fact that now we do not refer
340 * to socket after mapping. If one day we will need it, this
341 * function will increment ref. count on file by 1.
342 *
343 * In any case returned fd MAY BE not valid!
344 * This race condition is unavoidable
345 * with shared fd spaces, we cannot solve it inside kernel,
346 * but we take care of internal coherence yet.
347 */
348
349static int sock_alloc_file(struct socket *sock, struct file **f, int flags)
350{
351 struct qstr name = { .name = "" };
352 struct path path;
353 struct file *file;
354 int fd;
355
356 fd = get_unused_fd_flags(flags);
357 if (unlikely(fd < 0))
358 return fd;
359
360 path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
361 if (unlikely(!path.dentry)) {
362 put_unused_fd(fd);
363 return -ENOMEM;
364 }
365 path.mnt = mntget(sock_mnt);
366
367 d_instantiate(path.dentry, SOCK_INODE(sock));
368 SOCK_INODE(sock)->i_fop = &socket_file_ops;
369
370 file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
371 &socket_file_ops);
372 if (unlikely(!file)) {
373 /* drop dentry, keep inode */
374 ihold(path.dentry->d_inode);
375 path_put(&path);
376 put_unused_fd(fd);
377 return -ENFILE;
378 }
379
380 sock->file = file;
381 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
382 file->f_pos = 0;
383 file->private_data = sock;
384
385 *f = file;
386 return fd;
387}
388
389int sock_map_fd(struct socket *sock, int flags)
390{
391 struct file *newfile;
392 int fd = sock_alloc_file(sock, &newfile, flags);
393
394 if (likely(fd >= 0))
395 fd_install(fd, newfile);
396
397 return fd;
398}
399EXPORT_SYMBOL(sock_map_fd);
400
401static struct socket *sock_from_file(struct file *file, int *err)
402{
403 if (file->f_op == &socket_file_ops)
404 return file->private_data; /* set in sock_map_fd */
405
406 *err = -ENOTSOCK;
407 return NULL;
408}
409
410/**
411 * sockfd_lookup - Go from a file number to its socket slot
412 * @fd: file handle
413 * @err: pointer to an error code return
414 *
415 * The file handle passed in is locked and the socket it is bound
416 * too is returned. If an error occurs the err pointer is overwritten
417 * with a negative errno code and NULL is returned. The function checks
418 * for both invalid handles and passing a handle which is not a socket.
419 *
420 * On a success the socket object pointer is returned.
421 */
422
423struct socket *sockfd_lookup(int fd, int *err)
424{
425 struct file *file;
426 struct socket *sock;
427
428 file = fget(fd);
429 if (!file) {
430 *err = -EBADF;
431 return NULL;
432 }
433
434 sock = sock_from_file(file, err);
435 if (!sock)
436 fput(file);
437 return sock;
438}
439EXPORT_SYMBOL(sockfd_lookup);
440
441static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
442{
443 struct file *file;
444 struct socket *sock;
445
446 *err = -EBADF;
447 file = fget_light(fd, fput_needed);
448 if (file) {
449 sock = sock_from_file(file, err);
450 if (sock)
451 return sock;
452 fput_light(file, *fput_needed);
453 }
454 return NULL;
455}
456
457/**
458 * sock_alloc - allocate a socket
459 *
460 * Allocate a new inode and socket object. The two are bound together
461 * and initialised. The socket is then returned. If we are out of inodes
462 * NULL is returned.
463 */
464
465static struct socket *sock_alloc(void)
466{
467 struct inode *inode;
468 struct socket *sock;
469
470 inode = new_inode_pseudo(sock_mnt->mnt_sb);
471 if (!inode)
472 return NULL;
473
474 sock = SOCKET_I(inode);
475
476 kmemcheck_annotate_bitfield(sock, type);
477 inode->i_ino = get_next_ino();
478 inode->i_mode = S_IFSOCK | S_IRWXUGO;
479 inode->i_uid = current_fsuid();
480 inode->i_gid = current_fsgid();
481
482 percpu_add(sockets_in_use, 1);
483 return sock;
484}
485
486/*
487 * In theory you can't get an open on this inode, but /proc provides
488 * a back door. Remember to keep it shut otherwise you'll let the
489 * creepy crawlies in.
490 */
491
492static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
493{
494 return -ENXIO;
495}
496
497const struct file_operations bad_sock_fops = {
498 .owner = THIS_MODULE,
499 .open = sock_no_open,
500 .llseek = noop_llseek,
501};
502
503/**
504 * sock_release - close a socket
505 * @sock: socket to close
506 *
507 * The socket is released from the protocol stack if it has a release
508 * callback, and the inode is then released if the socket is bound to
509 * an inode not a file.
510 */
511
512void sock_release(struct socket *sock)
513{
514 if (sock->ops) {
515 struct module *owner = sock->ops->owner;
516
517 sock->ops->release(sock);
518 sock->ops = NULL;
519 module_put(owner);
520 }
521
522 if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
523 printk(KERN_ERR "sock_release: fasync list not empty!\n");
524
525 percpu_sub(sockets_in_use, 1);
526 if (!sock->file) {
527 iput(SOCK_INODE(sock));
528 return;
529 }
530 sock->file = NULL;
531}
532EXPORT_SYMBOL(sock_release);
533
534int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags)
535{
536 *tx_flags = 0;
537 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
538 *tx_flags |= SKBTX_HW_TSTAMP;
539 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
540 *tx_flags |= SKBTX_SW_TSTAMP;
541 return 0;
542}
543EXPORT_SYMBOL(sock_tx_timestamp);
544
545static inline int __sock_sendmsg_nosec(struct kiocb *iocb, struct socket *sock,
546 struct msghdr *msg, size_t size)
547{
548 struct sock_iocb *si = kiocb_to_siocb(iocb);
549
550 sock_update_classid(sock->sk);
551
552 si->sock = sock;
553 si->scm = NULL;
554 si->msg = msg;
555 si->size = size;
556
557 return sock->ops->sendmsg(iocb, sock, msg, size);
558}
559
560static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
561 struct msghdr *msg, size_t size)
562{
563 int err = security_socket_sendmsg(sock, msg, size);
564
565 return err ?: __sock_sendmsg_nosec(iocb, sock, msg, size);
566}
567
568int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
569{
570 struct kiocb iocb;
571 struct sock_iocb siocb;
572 int ret;
573
574 init_sync_kiocb(&iocb, NULL);
575 iocb.private = &siocb;
576 ret = __sock_sendmsg(&iocb, sock, msg, size);
577 if (-EIOCBQUEUED == ret)
578 ret = wait_on_sync_kiocb(&iocb);
579 return ret;
580}
581EXPORT_SYMBOL(sock_sendmsg);
582
583static int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg, size_t size)
584{
585 struct kiocb iocb;
586 struct sock_iocb siocb;
587 int ret;
588
589 init_sync_kiocb(&iocb, NULL);
590 iocb.private = &siocb;
591 ret = __sock_sendmsg_nosec(&iocb, sock, msg, size);
592 if (-EIOCBQUEUED == ret)
593 ret = wait_on_sync_kiocb(&iocb);
594 return ret;
595}
596
597int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
598 struct kvec *vec, size_t num, size_t size)
599{
600 mm_segment_t oldfs = get_fs();
601 int result;
602
603 set_fs(KERNEL_DS);
604 /*
605 * the following is safe, since for compiler definitions of kvec and
606 * iovec are identical, yielding the same in-core layout and alignment
607 */
608 msg->msg_iov = (struct iovec *)vec;
609 msg->msg_iovlen = num;
610 result = sock_sendmsg(sock, msg, size);
611 set_fs(oldfs);
612 return result;
613}
614EXPORT_SYMBOL(kernel_sendmsg);
615
616static int ktime2ts(ktime_t kt, struct timespec *ts)
617{
618 if (kt.tv64) {
619 *ts = ktime_to_timespec(kt);
620 return 1;
621 } else {
622 return 0;
623 }
624}
625
626/*
627 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
628 */
629void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
630 struct sk_buff *skb)
631{
632 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
633 struct timespec ts[3];
634 int empty = 1;
635 struct skb_shared_hwtstamps *shhwtstamps =
636 skb_hwtstamps(skb);
637
638 /* Race occurred between timestamp enabling and packet
639 receiving. Fill in the current time for now. */
640 if (need_software_tstamp && skb->tstamp.tv64 == 0)
641 __net_timestamp(skb);
642
643 if (need_software_tstamp) {
644 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
645 struct timeval tv;
646 skb_get_timestamp(skb, &tv);
647 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
648 sizeof(tv), &tv);
649 } else {
650 skb_get_timestampns(skb, &ts[0]);
651 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
652 sizeof(ts[0]), &ts[0]);
653 }
654 }
655
656
657 memset(ts, 0, sizeof(ts));
658 if (skb->tstamp.tv64 &&
659 sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
660 skb_get_timestampns(skb, ts + 0);
661 empty = 0;
662 }
663 if (shhwtstamps) {
664 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
665 ktime2ts(shhwtstamps->syststamp, ts + 1))
666 empty = 0;
667 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
668 ktime2ts(shhwtstamps->hwtstamp, ts + 2))
669 empty = 0;
670 }
671 if (!empty)
672 put_cmsg(msg, SOL_SOCKET,
673 SCM_TIMESTAMPING, sizeof(ts), &ts);
674}
675EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
676
677static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
678 struct sk_buff *skb)
679{
680 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
681 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
682 sizeof(__u32), &skb->dropcount);
683}
684
685void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
686 struct sk_buff *skb)
687{
688 sock_recv_timestamp(msg, sk, skb);
689 sock_recv_drops(msg, sk, skb);
690}
691EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
692
693static inline int __sock_recvmsg_nosec(struct kiocb *iocb, struct socket *sock,
694 struct msghdr *msg, size_t size, int flags)
695{
696 struct sock_iocb *si = kiocb_to_siocb(iocb);
697
698 sock_update_classid(sock->sk);
699
700 si->sock = sock;
701 si->scm = NULL;
702 si->msg = msg;
703 si->size = size;
704 si->flags = flags;
705
706 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
707}
708
709static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
710 struct msghdr *msg, size_t size, int flags)
711{
712 int err = security_socket_recvmsg(sock, msg, size, flags);
713
714 return err ?: __sock_recvmsg_nosec(iocb, sock, msg, size, flags);
715}
716
717int sock_recvmsg(struct socket *sock, struct msghdr *msg,
718 size_t size, int flags)
719{
720 struct kiocb iocb;
721 struct sock_iocb siocb;
722 int ret;
723
724 init_sync_kiocb(&iocb, NULL);
725 iocb.private = &siocb;
726 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
727 if (-EIOCBQUEUED == ret)
728 ret = wait_on_sync_kiocb(&iocb);
729 return ret;
730}
731EXPORT_SYMBOL(sock_recvmsg);
732
733static int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
734 size_t size, int flags)
735{
736 struct kiocb iocb;
737 struct sock_iocb siocb;
738 int ret;
739
740 init_sync_kiocb(&iocb, NULL);
741 iocb.private = &siocb;
742 ret = __sock_recvmsg_nosec(&iocb, sock, msg, size, flags);
743 if (-EIOCBQUEUED == ret)
744 ret = wait_on_sync_kiocb(&iocb);
745 return ret;
746}
747
748/**
749 * kernel_recvmsg - Receive a message from a socket (kernel space)
750 * @sock: The socket to receive the message from
751 * @msg: Received message
752 * @vec: Input s/g array for message data
753 * @num: Size of input s/g array
754 * @size: Number of bytes to read
755 * @flags: Message flags (MSG_DONTWAIT, etc...)
756 *
757 * On return the msg structure contains the scatter/gather array passed in the
758 * vec argument. The array is modified so that it consists of the unfilled
759 * portion of the original array.
760 *
761 * The returned value is the total number of bytes received, or an error.
762 */
763int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
764 struct kvec *vec, size_t num, size_t size, int flags)
765{
766 mm_segment_t oldfs = get_fs();
767 int result;
768
769 set_fs(KERNEL_DS);
770 /*
771 * the following is safe, since for compiler definitions of kvec and
772 * iovec are identical, yielding the same in-core layout and alignment
773 */
774 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
775 result = sock_recvmsg(sock, msg, size, flags);
776 set_fs(oldfs);
777 return result;
778}
779EXPORT_SYMBOL(kernel_recvmsg);
780
781static void sock_aio_dtor(struct kiocb *iocb)
782{
783 kfree(iocb->private);
784}
785
786static ssize_t sock_sendpage(struct file *file, struct page *page,
787 int offset, size_t size, loff_t *ppos, int more)
788{
789 struct socket *sock;
790 int flags;
791
792 sock = file->private_data;
793
794 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
795 if (more)
796 flags |= MSG_MORE;
797
798 return kernel_sendpage(sock, page, offset, size, flags);
799}
800
801static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
802 struct pipe_inode_info *pipe, size_t len,
803 unsigned int flags)
804{
805 struct socket *sock = file->private_data;
806
807 if (unlikely(!sock->ops->splice_read))
808 return -EINVAL;
809
810 sock_update_classid(sock->sk);
811
812 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
813}
814
815static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
816 struct sock_iocb *siocb)
817{
818 if (!is_sync_kiocb(iocb)) {
819 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
820 if (!siocb)
821 return NULL;
822 iocb->ki_dtor = sock_aio_dtor;
823 }
824
825 siocb->kiocb = iocb;
826 iocb->private = siocb;
827 return siocb;
828}
829
830static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
831 struct file *file, const struct iovec *iov,
832 unsigned long nr_segs)
833{
834 struct socket *sock = file->private_data;
835 size_t size = 0;
836 int i;
837
838 for (i = 0; i < nr_segs; i++)
839 size += iov[i].iov_len;
840
841 msg->msg_name = NULL;
842 msg->msg_namelen = 0;
843 msg->msg_control = NULL;
844 msg->msg_controllen = 0;
845 msg->msg_iov = (struct iovec *)iov;
846 msg->msg_iovlen = nr_segs;
847 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
848
849 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
850}
851
852static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
853 unsigned long nr_segs, loff_t pos)
854{
855 struct sock_iocb siocb, *x;
856
857 if (pos != 0)
858 return -ESPIPE;
859
860 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
861 return 0;
862
863
864 x = alloc_sock_iocb(iocb, &siocb);
865 if (!x)
866 return -ENOMEM;
867 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
868}
869
870static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
871 struct file *file, const struct iovec *iov,
872 unsigned long nr_segs)
873{
874 struct socket *sock = file->private_data;
875 size_t size = 0;
876 int i;
877
878 for (i = 0; i < nr_segs; i++)
879 size += iov[i].iov_len;
880
881 msg->msg_name = NULL;
882 msg->msg_namelen = 0;
883 msg->msg_control = NULL;
884 msg->msg_controllen = 0;
885 msg->msg_iov = (struct iovec *)iov;
886 msg->msg_iovlen = nr_segs;
887 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
888 if (sock->type == SOCK_SEQPACKET)
889 msg->msg_flags |= MSG_EOR;
890
891 return __sock_sendmsg(iocb, sock, msg, size);
892}
893
894static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
895 unsigned long nr_segs, loff_t pos)
896{
897 struct sock_iocb siocb, *x;
898
899 if (pos != 0)
900 return -ESPIPE;
901
902 x = alloc_sock_iocb(iocb, &siocb);
903 if (!x)
904 return -ENOMEM;
905
906 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
907}
908
909/*
910 * Atomic setting of ioctl hooks to avoid race
911 * with module unload.
912 */
913
914static DEFINE_MUTEX(br_ioctl_mutex);
915static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
916
917void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
918{
919 mutex_lock(&br_ioctl_mutex);
920 br_ioctl_hook = hook;
921 mutex_unlock(&br_ioctl_mutex);
922}
923EXPORT_SYMBOL(brioctl_set);
924
925static DEFINE_MUTEX(vlan_ioctl_mutex);
926static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
927
928void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
929{
930 mutex_lock(&vlan_ioctl_mutex);
931 vlan_ioctl_hook = hook;
932 mutex_unlock(&vlan_ioctl_mutex);
933}
934EXPORT_SYMBOL(vlan_ioctl_set);
935
936static DEFINE_MUTEX(dlci_ioctl_mutex);
937static int (*dlci_ioctl_hook) (unsigned int, void __user *);
938
939void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
940{
941 mutex_lock(&dlci_ioctl_mutex);
942 dlci_ioctl_hook = hook;
943 mutex_unlock(&dlci_ioctl_mutex);
944}
945EXPORT_SYMBOL(dlci_ioctl_set);
946
947static long sock_do_ioctl(struct net *net, struct socket *sock,
948 unsigned int cmd, unsigned long arg)
949{
950 int err;
951 void __user *argp = (void __user *)arg;
952
953 err = sock->ops->ioctl(sock, cmd, arg);
954
955 /*
956 * If this ioctl is unknown try to hand it down
957 * to the NIC driver.
958 */
959 if (err == -ENOIOCTLCMD)
960 err = dev_ioctl(net, cmd, argp);
961
962 return err;
963}
964
965/*
966 * With an ioctl, arg may well be a user mode pointer, but we don't know
967 * what to do with it - that's up to the protocol still.
968 */
969
970static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
971{
972 struct socket *sock;
973 struct sock *sk;
974 void __user *argp = (void __user *)arg;
975 int pid, err;
976 struct net *net;
977
978 sock = file->private_data;
979 sk = sock->sk;
980 net = sock_net(sk);
981 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
982 err = dev_ioctl(net, cmd, argp);
983 } else
984#ifdef CONFIG_WEXT_CORE
985 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
986 err = dev_ioctl(net, cmd, argp);
987 } else
988#endif
989 switch (cmd) {
990 case FIOSETOWN:
991 case SIOCSPGRP:
992 err = -EFAULT;
993 if (get_user(pid, (int __user *)argp))
994 break;
995 err = f_setown(sock->file, pid, 1);
996 break;
997 case FIOGETOWN:
998 case SIOCGPGRP:
999 err = put_user(f_getown(sock->file),
1000 (int __user *)argp);
1001 break;
1002 case SIOCGIFBR:
1003 case SIOCSIFBR:
1004 case SIOCBRADDBR:
1005 case SIOCBRDELBR:
1006 err = -ENOPKG;
1007 if (!br_ioctl_hook)
1008 request_module("bridge");
1009
1010 mutex_lock(&br_ioctl_mutex);
1011 if (br_ioctl_hook)
1012 err = br_ioctl_hook(net, cmd, argp);
1013 mutex_unlock(&br_ioctl_mutex);
1014 break;
1015 case SIOCGIFVLAN:
1016 case SIOCSIFVLAN:
1017 err = -ENOPKG;
1018 if (!vlan_ioctl_hook)
1019 request_module("8021q");
1020
1021 mutex_lock(&vlan_ioctl_mutex);
1022 if (vlan_ioctl_hook)
1023 err = vlan_ioctl_hook(net, argp);
1024 mutex_unlock(&vlan_ioctl_mutex);
1025 break;
1026 case SIOCADDDLCI:
1027 case SIOCDELDLCI:
1028 err = -ENOPKG;
1029 if (!dlci_ioctl_hook)
1030 request_module("dlci");
1031
1032 mutex_lock(&dlci_ioctl_mutex);
1033 if (dlci_ioctl_hook)
1034 err = dlci_ioctl_hook(cmd, argp);
1035 mutex_unlock(&dlci_ioctl_mutex);
1036 break;
1037 default:
1038 err = sock_do_ioctl(net, sock, cmd, arg);
1039 break;
1040 }
1041 return err;
1042}
1043
1044int sock_create_lite(int family, int type, int protocol, struct socket **res)
1045{
1046 int err;
1047 struct socket *sock = NULL;
1048
1049 err = security_socket_create(family, type, protocol, 1);
1050 if (err)
1051 goto out;
1052
1053 sock = sock_alloc();
1054 if (!sock) {
1055 err = -ENOMEM;
1056 goto out;
1057 }
1058
1059 sock->type = type;
1060 err = security_socket_post_create(sock, family, type, protocol, 1);
1061 if (err)
1062 goto out_release;
1063
1064out:
1065 *res = sock;
1066 return err;
1067out_release:
1068 sock_release(sock);
1069 sock = NULL;
1070 goto out;
1071}
1072EXPORT_SYMBOL(sock_create_lite);
1073
1074/* No kernel lock held - perfect */
1075static unsigned int sock_poll(struct file *file, poll_table *wait)
1076{
1077 struct socket *sock;
1078
1079 /*
1080 * We can't return errors to poll, so it's either yes or no.
1081 */
1082 sock = file->private_data;
1083 return sock->ops->poll(file, sock, wait);
1084}
1085
1086static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1087{
1088 struct socket *sock = file->private_data;
1089
1090 return sock->ops->mmap(file, sock, vma);
1091}
1092
1093static int sock_close(struct inode *inode, struct file *filp)
1094{
1095 /*
1096 * It was possible the inode is NULL we were
1097 * closing an unfinished socket.
1098 */
1099
1100 if (!inode) {
1101 printk(KERN_DEBUG "sock_close: NULL inode\n");
1102 return 0;
1103 }
1104 sock_release(SOCKET_I(inode));
1105 return 0;
1106}
1107
1108/*
1109 * Update the socket async list
1110 *
1111 * Fasync_list locking strategy.
1112 *
1113 * 1. fasync_list is modified only under process context socket lock
1114 * i.e. under semaphore.
1115 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1116 * or under socket lock
1117 */
1118
1119static int sock_fasync(int fd, struct file *filp, int on)
1120{
1121 struct socket *sock = filp->private_data;
1122 struct sock *sk = sock->sk;
1123 struct socket_wq *wq;
1124
1125 if (sk == NULL)
1126 return -EINVAL;
1127
1128 lock_sock(sk);
1129 wq = rcu_dereference_protected(sock->wq, sock_owned_by_user(sk));
1130 fasync_helper(fd, filp, on, &wq->fasync_list);
1131
1132 if (!wq->fasync_list)
1133 sock_reset_flag(sk, SOCK_FASYNC);
1134 else
1135 sock_set_flag(sk, SOCK_FASYNC);
1136
1137 release_sock(sk);
1138 return 0;
1139}
1140
1141/* This function may be called only under socket lock or callback_lock or rcu_lock */
1142
1143int sock_wake_async(struct socket *sock, int how, int band)
1144{
1145 struct socket_wq *wq;
1146
1147 if (!sock)
1148 return -1;
1149 rcu_read_lock();
1150 wq = rcu_dereference(sock->wq);
1151 if (!wq || !wq->fasync_list) {
1152 rcu_read_unlock();
1153 return -1;
1154 }
1155 switch (how) {
1156 case SOCK_WAKE_WAITD:
1157 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1158 break;
1159 goto call_kill;
1160 case SOCK_WAKE_SPACE:
1161 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1162 break;
1163 /* fall through */
1164 case SOCK_WAKE_IO:
1165call_kill:
1166 kill_fasync(&wq->fasync_list, SIGIO, band);
1167 break;
1168 case SOCK_WAKE_URG:
1169 kill_fasync(&wq->fasync_list, SIGURG, band);
1170 }
1171 rcu_read_unlock();
1172 return 0;
1173}
1174EXPORT_SYMBOL(sock_wake_async);
1175
1176int __sock_create(struct net *net, int family, int type, int protocol,
1177 struct socket **res, int kern)
1178{
1179 int err;
1180 struct socket *sock;
1181 const struct net_proto_family *pf;
1182
1183 /*
1184 * Check protocol is in range
1185 */
1186 if (family < 0 || family >= NPROTO)
1187 return -EAFNOSUPPORT;
1188 if (type < 0 || type >= SOCK_MAX)
1189 return -EINVAL;
1190
1191 /* Compatibility.
1192
1193 This uglymoron is moved from INET layer to here to avoid
1194 deadlock in module load.
1195 */
1196 if (family == PF_INET && type == SOCK_PACKET) {
1197 static int warned;
1198 if (!warned) {
1199 warned = 1;
1200 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1201 current->comm);
1202 }
1203 family = PF_PACKET;
1204 }
1205
1206 err = security_socket_create(family, type, protocol, kern);
1207 if (err)
1208 return err;
1209
1210 /*
1211 * Allocate the socket and allow the family to set things up. if
1212 * the protocol is 0, the family is instructed to select an appropriate
1213 * default.
1214 */
1215 sock = sock_alloc();
1216 if (!sock) {
1217 if (net_ratelimit())
1218 printk(KERN_WARNING "socket: no more sockets\n");
1219 return -ENFILE; /* Not exactly a match, but its the
1220 closest posix thing */
1221 }
1222
1223 sock->type = type;
1224
1225#ifdef CONFIG_MODULES
1226 /* Attempt to load a protocol module if the find failed.
1227 *
1228 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1229 * requested real, full-featured networking support upon configuration.
1230 * Otherwise module support will break!
1231 */
1232 if (rcu_access_pointer(net_families[family]) == NULL)
1233 request_module("net-pf-%d", family);
1234#endif
1235
1236 rcu_read_lock();
1237 pf = rcu_dereference(net_families[family]);
1238 err = -EAFNOSUPPORT;
1239 if (!pf)
1240 goto out_release;
1241
1242 /*
1243 * We will call the ->create function, that possibly is in a loadable
1244 * module, so we have to bump that loadable module refcnt first.
1245 */
1246 if (!try_module_get(pf->owner))
1247 goto out_release;
1248
1249 /* Now protected by module ref count */
1250 rcu_read_unlock();
1251
1252 err = pf->create(net, sock, protocol, kern);
1253 if (err < 0)
1254 goto out_module_put;
1255
1256 /*
1257 * Now to bump the refcnt of the [loadable] module that owns this
1258 * socket at sock_release time we decrement its refcnt.
1259 */
1260 if (!try_module_get(sock->ops->owner))
1261 goto out_module_busy;
1262
1263 /*
1264 * Now that we're done with the ->create function, the [loadable]
1265 * module can have its refcnt decremented
1266 */
1267 module_put(pf->owner);
1268 err = security_socket_post_create(sock, family, type, protocol, kern);
1269 if (err)
1270 goto out_sock_release;
1271 *res = sock;
1272
1273 return 0;
1274
1275out_module_busy:
1276 err = -EAFNOSUPPORT;
1277out_module_put:
1278 sock->ops = NULL;
1279 module_put(pf->owner);
1280out_sock_release:
1281 sock_release(sock);
1282 return err;
1283
1284out_release:
1285 rcu_read_unlock();
1286 goto out_sock_release;
1287}
1288EXPORT_SYMBOL(__sock_create);
1289
1290int sock_create(int family, int type, int protocol, struct socket **res)
1291{
1292 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1293}
1294EXPORT_SYMBOL(sock_create);
1295
1296int sock_create_kern(int family, int type, int protocol, struct socket **res)
1297{
1298 return __sock_create(&init_net, family, type, protocol, res, 1);
1299}
1300EXPORT_SYMBOL(sock_create_kern);
1301
1302SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1303{
1304 int retval;
1305 struct socket *sock;
1306 int flags;
1307
1308 /* Check the SOCK_* constants for consistency. */
1309 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1310 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1311 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1312 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1313
1314 flags = type & ~SOCK_TYPE_MASK;
1315 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1316 return -EINVAL;
1317 type &= SOCK_TYPE_MASK;
1318
1319 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1320 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1321
1322 retval = sock_create(family, type, protocol, &sock);
1323 if (retval < 0)
1324 goto out;
1325
1326 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1327 if (retval < 0)
1328 goto out_release;
1329
1330out:
1331 /* It may be already another descriptor 8) Not kernel problem. */
1332 return retval;
1333
1334out_release:
1335 sock_release(sock);
1336 return retval;
1337}
1338
1339/*
1340 * Create a pair of connected sockets.
1341 */
1342
1343SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1344 int __user *, usockvec)
1345{
1346 struct socket *sock1, *sock2;
1347 int fd1, fd2, err;
1348 struct file *newfile1, *newfile2;
1349 int flags;
1350
1351 flags = type & ~SOCK_TYPE_MASK;
1352 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1353 return -EINVAL;
1354 type &= SOCK_TYPE_MASK;
1355
1356 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1357 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1358
1359 /*
1360 * Obtain the first socket and check if the underlying protocol
1361 * supports the socketpair call.
1362 */
1363
1364 err = sock_create(family, type, protocol, &sock1);
1365 if (err < 0)
1366 goto out;
1367
1368 err = sock_create(family, type, protocol, &sock2);
1369 if (err < 0)
1370 goto out_release_1;
1371
1372 err = sock1->ops->socketpair(sock1, sock2);
1373 if (err < 0)
1374 goto out_release_both;
1375
1376 fd1 = sock_alloc_file(sock1, &newfile1, flags);
1377 if (unlikely(fd1 < 0)) {
1378 err = fd1;
1379 goto out_release_both;
1380 }
1381
1382 fd2 = sock_alloc_file(sock2, &newfile2, flags);
1383 if (unlikely(fd2 < 0)) {
1384 err = fd2;
1385 fput(newfile1);
1386 put_unused_fd(fd1);
1387 sock_release(sock2);
1388 goto out;
1389 }
1390
1391 audit_fd_pair(fd1, fd2);
1392 fd_install(fd1, newfile1);
1393 fd_install(fd2, newfile2);
1394 /* fd1 and fd2 may be already another descriptors.
1395 * Not kernel problem.
1396 */
1397
1398 err = put_user(fd1, &usockvec[0]);
1399 if (!err)
1400 err = put_user(fd2, &usockvec[1]);
1401 if (!err)
1402 return 0;
1403
1404 sys_close(fd2);
1405 sys_close(fd1);
1406 return err;
1407
1408out_release_both:
1409 sock_release(sock2);
1410out_release_1:
1411 sock_release(sock1);
1412out:
1413 return err;
1414}
1415
1416/*
1417 * Bind a name to a socket. Nothing much to do here since it's
1418 * the protocol's responsibility to handle the local address.
1419 *
1420 * We move the socket address to kernel space before we call
1421 * the protocol layer (having also checked the address is ok).
1422 */
1423
1424SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1425{
1426 struct socket *sock;
1427 struct sockaddr_storage address;
1428 int err, fput_needed;
1429
1430 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1431 if (sock) {
1432 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1433 if (err >= 0) {
1434 err = security_socket_bind(sock,
1435 (struct sockaddr *)&address,
1436 addrlen);
1437 if (!err)
1438 err = sock->ops->bind(sock,
1439 (struct sockaddr *)
1440 &address, addrlen);
1441 }
1442 fput_light(sock->file, fput_needed);
1443 }
1444 return err;
1445}
1446
1447/*
1448 * Perform a listen. Basically, we allow the protocol to do anything
1449 * necessary for a listen, and if that works, we mark the socket as
1450 * ready for listening.
1451 */
1452
1453SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1454{
1455 struct socket *sock;
1456 int err, fput_needed;
1457 int somaxconn;
1458
1459 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1460 if (sock) {
1461 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1462 if ((unsigned)backlog > somaxconn)
1463 backlog = somaxconn;
1464
1465 err = security_socket_listen(sock, backlog);
1466 if (!err)
1467 err = sock->ops->listen(sock, backlog);
1468
1469 fput_light(sock->file, fput_needed);
1470 }
1471 return err;
1472}
1473
1474/*
1475 * For accept, we attempt to create a new socket, set up the link
1476 * with the client, wake up the client, then return the new
1477 * connected fd. We collect the address of the connector in kernel
1478 * space and move it to user at the very end. This is unclean because
1479 * we open the socket then return an error.
1480 *
1481 * 1003.1g adds the ability to recvmsg() to query connection pending
1482 * status to recvmsg. We need to add that support in a way thats
1483 * clean when we restucture accept also.
1484 */
1485
1486SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1487 int __user *, upeer_addrlen, int, flags)
1488{
1489 struct socket *sock, *newsock;
1490 struct file *newfile;
1491 int err, len, newfd, fput_needed;
1492 struct sockaddr_storage address;
1493
1494 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1495 return -EINVAL;
1496
1497 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1498 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1499
1500 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1501 if (!sock)
1502 goto out;
1503
1504 err = -ENFILE;
1505 newsock = sock_alloc();
1506 if (!newsock)
1507 goto out_put;
1508
1509 newsock->type = sock->type;
1510 newsock->ops = sock->ops;
1511
1512 /*
1513 * We don't need try_module_get here, as the listening socket (sock)
1514 * has the protocol module (sock->ops->owner) held.
1515 */
1516 __module_get(newsock->ops->owner);
1517
1518 newfd = sock_alloc_file(newsock, &newfile, flags);
1519 if (unlikely(newfd < 0)) {
1520 err = newfd;
1521 sock_release(newsock);
1522 goto out_put;
1523 }
1524
1525 err = security_socket_accept(sock, newsock);
1526 if (err)
1527 goto out_fd;
1528
1529 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1530 if (err < 0)
1531 goto out_fd;
1532
1533 if (upeer_sockaddr) {
1534 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1535 &len, 2) < 0) {
1536 err = -ECONNABORTED;
1537 goto out_fd;
1538 }
1539 err = move_addr_to_user((struct sockaddr *)&address,
1540 len, upeer_sockaddr, upeer_addrlen);
1541 if (err < 0)
1542 goto out_fd;
1543 }
1544
1545 /* File flags are not inherited via accept() unlike another OSes. */
1546
1547 fd_install(newfd, newfile);
1548 err = newfd;
1549
1550out_put:
1551 fput_light(sock->file, fput_needed);
1552out:
1553 return err;
1554out_fd:
1555 fput(newfile);
1556 put_unused_fd(newfd);
1557 goto out_put;
1558}
1559
1560SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1561 int __user *, upeer_addrlen)
1562{
1563 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1564}
1565
1566/*
1567 * Attempt to connect to a socket with the server address. The address
1568 * is in user space so we verify it is OK and move it to kernel space.
1569 *
1570 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1571 * break bindings
1572 *
1573 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1574 * other SEQPACKET protocols that take time to connect() as it doesn't
1575 * include the -EINPROGRESS status for such sockets.
1576 */
1577
1578SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1579 int, addrlen)
1580{
1581 struct socket *sock;
1582 struct sockaddr_storage address;
1583 int err, fput_needed;
1584
1585 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1586 if (!sock)
1587 goto out;
1588 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1589 if (err < 0)
1590 goto out_put;
1591
1592 err =
1593 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1594 if (err)
1595 goto out_put;
1596
1597 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1598 sock->file->f_flags);
1599out_put:
1600 fput_light(sock->file, fput_needed);
1601out:
1602 return err;
1603}
1604
1605/*
1606 * Get the local address ('name') of a socket object. Move the obtained
1607 * name to user space.
1608 */
1609
1610SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1611 int __user *, usockaddr_len)
1612{
1613 struct socket *sock;
1614 struct sockaddr_storage address;
1615 int len, err, fput_needed;
1616
1617 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1618 if (!sock)
1619 goto out;
1620
1621 err = security_socket_getsockname(sock);
1622 if (err)
1623 goto out_put;
1624
1625 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1626 if (err)
1627 goto out_put;
1628 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1629
1630out_put:
1631 fput_light(sock->file, fput_needed);
1632out:
1633 return err;
1634}
1635
1636/*
1637 * Get the remote address ('name') of a socket object. Move the obtained
1638 * name to user space.
1639 */
1640
1641SYSCALL_DEFINE3(getpeername, 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 != NULL) {
1650 err = security_socket_getpeername(sock);
1651 if (err) {
1652 fput_light(sock->file, fput_needed);
1653 return err;
1654 }
1655
1656 err =
1657 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1658 1);
1659 if (!err)
1660 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1661 usockaddr_len);
1662 fput_light(sock->file, fput_needed);
1663 }
1664 return err;
1665}
1666
1667/*
1668 * Send a datagram to a given address. We move the address into kernel
1669 * space and check the user space data area is readable before invoking
1670 * the protocol.
1671 */
1672
1673SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1674 unsigned, flags, struct sockaddr __user *, addr,
1675 int, addr_len)
1676{
1677 struct socket *sock;
1678 struct sockaddr_storage address;
1679 int err;
1680 struct msghdr msg;
1681 struct iovec iov;
1682 int fput_needed;
1683
1684 if (len > INT_MAX)
1685 len = INT_MAX;
1686 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1687 if (!sock)
1688 goto out;
1689
1690 iov.iov_base = buff;
1691 iov.iov_len = len;
1692 msg.msg_name = NULL;
1693 msg.msg_iov = &iov;
1694 msg.msg_iovlen = 1;
1695 msg.msg_control = NULL;
1696 msg.msg_controllen = 0;
1697 msg.msg_namelen = 0;
1698 if (addr) {
1699 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1700 if (err < 0)
1701 goto out_put;
1702 msg.msg_name = (struct sockaddr *)&address;
1703 msg.msg_namelen = addr_len;
1704 }
1705 if (sock->file->f_flags & O_NONBLOCK)
1706 flags |= MSG_DONTWAIT;
1707 msg.msg_flags = flags;
1708 err = sock_sendmsg(sock, &msg, len);
1709
1710out_put:
1711 fput_light(sock->file, fput_needed);
1712out:
1713 return err;
1714}
1715
1716/*
1717 * Send a datagram down a socket.
1718 */
1719
1720SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1721 unsigned, flags)
1722{
1723 return sys_sendto(fd, buff, len, flags, NULL, 0);
1724}
1725
1726/*
1727 * Receive a frame from the socket and optionally record the address of the
1728 * sender. We verify the buffers are writable and if needed move the
1729 * sender address from kernel to user space.
1730 */
1731
1732SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1733 unsigned, flags, struct sockaddr __user *, addr,
1734 int __user *, addr_len)
1735{
1736 struct socket *sock;
1737 struct iovec iov;
1738 struct msghdr msg;
1739 struct sockaddr_storage address;
1740 int err, err2;
1741 int fput_needed;
1742
1743 if (size > INT_MAX)
1744 size = INT_MAX;
1745 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1746 if (!sock)
1747 goto out;
1748
1749 msg.msg_control = NULL;
1750 msg.msg_controllen = 0;
1751 msg.msg_iovlen = 1;
1752 msg.msg_iov = &iov;
1753 iov.iov_len = size;
1754 iov.iov_base = ubuf;
1755 msg.msg_name = (struct sockaddr *)&address;
1756 msg.msg_namelen = sizeof(address);
1757 if (sock->file->f_flags & O_NONBLOCK)
1758 flags |= MSG_DONTWAIT;
1759 err = sock_recvmsg(sock, &msg, size, flags);
1760
1761 if (err >= 0 && addr != NULL) {
1762 err2 = move_addr_to_user((struct sockaddr *)&address,
1763 msg.msg_namelen, addr, addr_len);
1764 if (err2 < 0)
1765 err = err2;
1766 }
1767
1768 fput_light(sock->file, fput_needed);
1769out:
1770 return err;
1771}
1772
1773/*
1774 * Receive a datagram from a socket.
1775 */
1776
1777asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1778 unsigned flags)
1779{
1780 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1781}
1782
1783/*
1784 * Set a socket option. Because we don't know the option lengths we have
1785 * to pass the user mode parameter for the protocols to sort out.
1786 */
1787
1788SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1789 char __user *, optval, int, optlen)
1790{
1791 int err, fput_needed;
1792 struct socket *sock;
1793
1794 if (optlen < 0)
1795 return -EINVAL;
1796
1797 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1798 if (sock != NULL) {
1799 err = security_socket_setsockopt(sock, level, optname);
1800 if (err)
1801 goto out_put;
1802
1803 if (level == SOL_SOCKET)
1804 err =
1805 sock_setsockopt(sock, level, optname, optval,
1806 optlen);
1807 else
1808 err =
1809 sock->ops->setsockopt(sock, level, optname, optval,
1810 optlen);
1811out_put:
1812 fput_light(sock->file, fput_needed);
1813 }
1814 return err;
1815}
1816
1817/*
1818 * Get a socket option. Because we don't know the option lengths we have
1819 * to pass a user mode parameter for the protocols to sort out.
1820 */
1821
1822SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1823 char __user *, optval, int __user *, optlen)
1824{
1825 int err, fput_needed;
1826 struct socket *sock;
1827
1828 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1829 if (sock != NULL) {
1830 err = security_socket_getsockopt(sock, level, optname);
1831 if (err)
1832 goto out_put;
1833
1834 if (level == SOL_SOCKET)
1835 err =
1836 sock_getsockopt(sock, level, optname, optval,
1837 optlen);
1838 else
1839 err =
1840 sock->ops->getsockopt(sock, level, optname, optval,
1841 optlen);
1842out_put:
1843 fput_light(sock->file, fput_needed);
1844 }
1845 return err;
1846}
1847
1848/*
1849 * Shutdown a socket.
1850 */
1851
1852SYSCALL_DEFINE2(shutdown, int, fd, int, how)
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_shutdown(sock, how);
1860 if (!err)
1861 err = sock->ops->shutdown(sock, how);
1862 fput_light(sock->file, fput_needed);
1863 }
1864 return err;
1865}
1866
1867/* A couple of helpful macros for getting the address of the 32/64 bit
1868 * fields which are the same type (int / unsigned) on our platforms.
1869 */
1870#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1871#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1872#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1873
1874struct used_address {
1875 struct sockaddr_storage name;
1876 unsigned int name_len;
1877};
1878
1879static int __sys_sendmsg(struct socket *sock, struct msghdr __user *msg,
1880 struct msghdr *msg_sys, unsigned flags,
1881 struct used_address *used_address)
1882{
1883 struct compat_msghdr __user *msg_compat =
1884 (struct compat_msghdr __user *)msg;
1885 struct sockaddr_storage address;
1886 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1887 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1888 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1889 /* 20 is size of ipv6_pktinfo */
1890 unsigned char *ctl_buf = ctl;
1891 int err, ctl_len, iov_size, total_len;
1892
1893 err = -EFAULT;
1894 if (MSG_CMSG_COMPAT & flags) {
1895 if (get_compat_msghdr(msg_sys, msg_compat))
1896 return -EFAULT;
1897 } else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
1898 return -EFAULT;
1899
1900 /* do not move before msg_sys is valid */
1901 err = -EMSGSIZE;
1902 if (msg_sys->msg_iovlen > UIO_MAXIOV)
1903 goto out;
1904
1905 /* Check whether to allocate the iovec area */
1906 err = -ENOMEM;
1907 iov_size = msg_sys->msg_iovlen * sizeof(struct iovec);
1908 if (msg_sys->msg_iovlen > UIO_FASTIOV) {
1909 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1910 if (!iov)
1911 goto out;
1912 }
1913
1914 /* This will also move the address data into kernel space */
1915 if (MSG_CMSG_COMPAT & flags) {
1916 err = verify_compat_iovec(msg_sys, iov,
1917 (struct sockaddr *)&address,
1918 VERIFY_READ);
1919 } else
1920 err = verify_iovec(msg_sys, iov,
1921 (struct sockaddr *)&address,
1922 VERIFY_READ);
1923 if (err < 0)
1924 goto out_freeiov;
1925 total_len = err;
1926
1927 err = -ENOBUFS;
1928
1929 if (msg_sys->msg_controllen > INT_MAX)
1930 goto out_freeiov;
1931 ctl_len = msg_sys->msg_controllen;
1932 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1933 err =
1934 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
1935 sizeof(ctl));
1936 if (err)
1937 goto out_freeiov;
1938 ctl_buf = msg_sys->msg_control;
1939 ctl_len = msg_sys->msg_controllen;
1940 } else if (ctl_len) {
1941 if (ctl_len > sizeof(ctl)) {
1942 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1943 if (ctl_buf == NULL)
1944 goto out_freeiov;
1945 }
1946 err = -EFAULT;
1947 /*
1948 * Careful! Before this, msg_sys->msg_control contains a user pointer.
1949 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1950 * checking falls down on this.
1951 */
1952 if (copy_from_user(ctl_buf,
1953 (void __user __force *)msg_sys->msg_control,
1954 ctl_len))
1955 goto out_freectl;
1956 msg_sys->msg_control = ctl_buf;
1957 }
1958 msg_sys->msg_flags = flags;
1959
1960 if (sock->file->f_flags & O_NONBLOCK)
1961 msg_sys->msg_flags |= MSG_DONTWAIT;
1962 /*
1963 * If this is sendmmsg() and current destination address is same as
1964 * previously succeeded address, omit asking LSM's decision.
1965 * used_address->name_len is initialized to UINT_MAX so that the first
1966 * destination address never matches.
1967 */
1968 if (used_address && msg_sys->msg_name &&
1969 used_address->name_len == msg_sys->msg_namelen &&
1970 !memcmp(&used_address->name, msg_sys->msg_name,
1971 used_address->name_len)) {
1972 err = sock_sendmsg_nosec(sock, msg_sys, total_len);
1973 goto out_freectl;
1974 }
1975 err = sock_sendmsg(sock, msg_sys, total_len);
1976 /*
1977 * If this is sendmmsg() and sending to current destination address was
1978 * successful, remember it.
1979 */
1980 if (used_address && err >= 0) {
1981 used_address->name_len = msg_sys->msg_namelen;
1982 if (msg_sys->msg_name)
1983 memcpy(&used_address->name, msg_sys->msg_name,
1984 used_address->name_len);
1985 }
1986
1987out_freectl:
1988 if (ctl_buf != ctl)
1989 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1990out_freeiov:
1991 if (iov != iovstack)
1992 sock_kfree_s(sock->sk, iov, iov_size);
1993out:
1994 return err;
1995}
1996
1997/*
1998 * BSD sendmsg interface
1999 */
2000
2001SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
2002{
2003 int fput_needed, err;
2004 struct msghdr msg_sys;
2005 struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2006
2007 if (!sock)
2008 goto out;
2009
2010 err = __sys_sendmsg(sock, msg, &msg_sys, flags, NULL);
2011
2012 fput_light(sock->file, fput_needed);
2013out:
2014 return err;
2015}
2016
2017/*
2018 * Linux sendmmsg interface
2019 */
2020
2021int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2022 unsigned int flags)
2023{
2024 int fput_needed, err, datagrams;
2025 struct socket *sock;
2026 struct mmsghdr __user *entry;
2027 struct compat_mmsghdr __user *compat_entry;
2028 struct msghdr msg_sys;
2029 struct used_address used_address;
2030
2031 if (vlen > UIO_MAXIOV)
2032 vlen = UIO_MAXIOV;
2033
2034 datagrams = 0;
2035
2036 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2037 if (!sock)
2038 return err;
2039
2040 used_address.name_len = UINT_MAX;
2041 entry = mmsg;
2042 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2043 err = 0;
2044
2045 while (datagrams < vlen) {
2046 if (MSG_CMSG_COMPAT & flags) {
2047 err = __sys_sendmsg(sock, (struct msghdr __user *)compat_entry,
2048 &msg_sys, flags, &used_address);
2049 if (err < 0)
2050 break;
2051 err = __put_user(err, &compat_entry->msg_len);
2052 ++compat_entry;
2053 } else {
2054 err = __sys_sendmsg(sock, (struct msghdr __user *)entry,
2055 &msg_sys, flags, &used_address);
2056 if (err < 0)
2057 break;
2058 err = put_user(err, &entry->msg_len);
2059 ++entry;
2060 }
2061
2062 if (err)
2063 break;
2064 ++datagrams;
2065 }
2066
2067 fput_light(sock->file, fput_needed);
2068
2069 /* We only return an error if no datagrams were able to be sent */
2070 if (datagrams != 0)
2071 return datagrams;
2072
2073 return err;
2074}
2075
2076SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2077 unsigned int, vlen, unsigned int, flags)
2078{
2079 return __sys_sendmmsg(fd, mmsg, vlen, flags);
2080}
2081
2082static int __sys_recvmsg(struct socket *sock, struct msghdr __user *msg,
2083 struct msghdr *msg_sys, unsigned flags, int nosec)
2084{
2085 struct compat_msghdr __user *msg_compat =
2086 (struct compat_msghdr __user *)msg;
2087 struct iovec iovstack[UIO_FASTIOV];
2088 struct iovec *iov = iovstack;
2089 unsigned long cmsg_ptr;
2090 int err, iov_size, total_len, len;
2091
2092 /* kernel mode address */
2093 struct sockaddr_storage addr;
2094
2095 /* user mode address pointers */
2096 struct sockaddr __user *uaddr;
2097 int __user *uaddr_len;
2098
2099 if (MSG_CMSG_COMPAT & flags) {
2100 if (get_compat_msghdr(msg_sys, msg_compat))
2101 return -EFAULT;
2102 } else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
2103 return -EFAULT;
2104
2105 err = -EMSGSIZE;
2106 if (msg_sys->msg_iovlen > UIO_MAXIOV)
2107 goto out;
2108
2109 /* Check whether to allocate the iovec area */
2110 err = -ENOMEM;
2111 iov_size = msg_sys->msg_iovlen * sizeof(struct iovec);
2112 if (msg_sys->msg_iovlen > UIO_FASTIOV) {
2113 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2114 if (!iov)
2115 goto out;
2116 }
2117
2118 /*
2119 * Save the user-mode address (verify_iovec will change the
2120 * kernel msghdr to use the kernel address space)
2121 */
2122
2123 uaddr = (__force void __user *)msg_sys->msg_name;
2124 uaddr_len = COMPAT_NAMELEN(msg);
2125 if (MSG_CMSG_COMPAT & flags) {
2126 err = verify_compat_iovec(msg_sys, iov,
2127 (struct sockaddr *)&addr,
2128 VERIFY_WRITE);
2129 } else
2130 err = verify_iovec(msg_sys, iov,
2131 (struct sockaddr *)&addr,
2132 VERIFY_WRITE);
2133 if (err < 0)
2134 goto out_freeiov;
2135 total_len = err;
2136
2137 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2138 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2139
2140 if (sock->file->f_flags & O_NONBLOCK)
2141 flags |= MSG_DONTWAIT;
2142 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys,
2143 total_len, flags);
2144 if (err < 0)
2145 goto out_freeiov;
2146 len = err;
2147
2148 if (uaddr != NULL) {
2149 err = move_addr_to_user((struct sockaddr *)&addr,
2150 msg_sys->msg_namelen, uaddr,
2151 uaddr_len);
2152 if (err < 0)
2153 goto out_freeiov;
2154 }
2155 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2156 COMPAT_FLAGS(msg));
2157 if (err)
2158 goto out_freeiov;
2159 if (MSG_CMSG_COMPAT & flags)
2160 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2161 &msg_compat->msg_controllen);
2162 else
2163 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2164 &msg->msg_controllen);
2165 if (err)
2166 goto out_freeiov;
2167 err = len;
2168
2169out_freeiov:
2170 if (iov != iovstack)
2171 sock_kfree_s(sock->sk, iov, iov_size);
2172out:
2173 return err;
2174}
2175
2176/*
2177 * BSD recvmsg interface
2178 */
2179
2180SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
2181 unsigned int, flags)
2182{
2183 int fput_needed, err;
2184 struct msghdr msg_sys;
2185 struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2186
2187 if (!sock)
2188 goto out;
2189
2190 err = __sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2191
2192 fput_light(sock->file, fput_needed);
2193out:
2194 return err;
2195}
2196
2197/*
2198 * Linux recvmmsg interface
2199 */
2200
2201int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2202 unsigned int flags, struct timespec *timeout)
2203{
2204 int fput_needed, err, datagrams;
2205 struct socket *sock;
2206 struct mmsghdr __user *entry;
2207 struct compat_mmsghdr __user *compat_entry;
2208 struct msghdr msg_sys;
2209 struct timespec end_time;
2210
2211 if (timeout &&
2212 poll_select_set_timeout(&end_time, timeout->tv_sec,
2213 timeout->tv_nsec))
2214 return -EINVAL;
2215
2216 datagrams = 0;
2217
2218 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2219 if (!sock)
2220 return err;
2221
2222 err = sock_error(sock->sk);
2223 if (err)
2224 goto out_put;
2225
2226 entry = mmsg;
2227 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2228
2229 while (datagrams < vlen) {
2230 /*
2231 * No need to ask LSM for more than the first datagram.
2232 */
2233 if (MSG_CMSG_COMPAT & flags) {
2234 err = __sys_recvmsg(sock, (struct msghdr __user *)compat_entry,
2235 &msg_sys, flags & ~MSG_WAITFORONE,
2236 datagrams);
2237 if (err < 0)
2238 break;
2239 err = __put_user(err, &compat_entry->msg_len);
2240 ++compat_entry;
2241 } else {
2242 err = __sys_recvmsg(sock, (struct msghdr __user *)entry,
2243 &msg_sys, flags & ~MSG_WAITFORONE,
2244 datagrams);
2245 if (err < 0)
2246 break;
2247 err = put_user(err, &entry->msg_len);
2248 ++entry;
2249 }
2250
2251 if (err)
2252 break;
2253 ++datagrams;
2254
2255 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2256 if (flags & MSG_WAITFORONE)
2257 flags |= MSG_DONTWAIT;
2258
2259 if (timeout) {
2260 ktime_get_ts(timeout);
2261 *timeout = timespec_sub(end_time, *timeout);
2262 if (timeout->tv_sec < 0) {
2263 timeout->tv_sec = timeout->tv_nsec = 0;
2264 break;
2265 }
2266
2267 /* Timeout, return less than vlen datagrams */
2268 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2269 break;
2270 }
2271
2272 /* Out of band data, return right away */
2273 if (msg_sys.msg_flags & MSG_OOB)
2274 break;
2275 }
2276
2277out_put:
2278 fput_light(sock->file, fput_needed);
2279
2280 if (err == 0)
2281 return datagrams;
2282
2283 if (datagrams != 0) {
2284 /*
2285 * We may return less entries than requested (vlen) if the
2286 * sock is non block and there aren't enough datagrams...
2287 */
2288 if (err != -EAGAIN) {
2289 /*
2290 * ... or if recvmsg returns an error after we
2291 * received some datagrams, where we record the
2292 * error to return on the next call or if the
2293 * app asks about it using getsockopt(SO_ERROR).
2294 */
2295 sock->sk->sk_err = -err;
2296 }
2297
2298 return datagrams;
2299 }
2300
2301 return err;
2302}
2303
2304SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2305 unsigned int, vlen, unsigned int, flags,
2306 struct timespec __user *, timeout)
2307{
2308 int datagrams;
2309 struct timespec timeout_sys;
2310
2311 if (!timeout)
2312 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2313
2314 if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2315 return -EFAULT;
2316
2317 datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2318
2319 if (datagrams > 0 &&
2320 copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2321 datagrams = -EFAULT;
2322
2323 return datagrams;
2324}
2325
2326#ifdef __ARCH_WANT_SYS_SOCKETCALL
2327/* Argument list sizes for sys_socketcall */
2328#define AL(x) ((x) * sizeof(unsigned long))
2329static const unsigned char nargs[21] = {
2330 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2331 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2332 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2333 AL(4), AL(5), AL(4)
2334};
2335
2336#undef AL
2337
2338/*
2339 * System call vectors.
2340 *
2341 * Argument checking cleaned up. Saved 20% in size.
2342 * This function doesn't need to set the kernel lock because
2343 * it is set by the callees.
2344 */
2345
2346SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2347{
2348 unsigned long a[6];
2349 unsigned long a0, a1;
2350 int err;
2351 unsigned int len;
2352
2353 if (call < 1 || call > SYS_SENDMMSG)
2354 return -EINVAL;
2355
2356 len = nargs[call];
2357 if (len > sizeof(a))
2358 return -EINVAL;
2359
2360 /* copy_from_user should be SMP safe. */
2361 if (copy_from_user(a, args, len))
2362 return -EFAULT;
2363
2364 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2365
2366 a0 = a[0];
2367 a1 = a[1];
2368
2369 switch (call) {
2370 case SYS_SOCKET:
2371 err = sys_socket(a0, a1, a[2]);
2372 break;
2373 case SYS_BIND:
2374 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2375 break;
2376 case SYS_CONNECT:
2377 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2378 break;
2379 case SYS_LISTEN:
2380 err = sys_listen(a0, a1);
2381 break;
2382 case SYS_ACCEPT:
2383 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2384 (int __user *)a[2], 0);
2385 break;
2386 case SYS_GETSOCKNAME:
2387 err =
2388 sys_getsockname(a0, (struct sockaddr __user *)a1,
2389 (int __user *)a[2]);
2390 break;
2391 case SYS_GETPEERNAME:
2392 err =
2393 sys_getpeername(a0, (struct sockaddr __user *)a1,
2394 (int __user *)a[2]);
2395 break;
2396 case SYS_SOCKETPAIR:
2397 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2398 break;
2399 case SYS_SEND:
2400 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2401 break;
2402 case SYS_SENDTO:
2403 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2404 (struct sockaddr __user *)a[4], a[5]);
2405 break;
2406 case SYS_RECV:
2407 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2408 break;
2409 case SYS_RECVFROM:
2410 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2411 (struct sockaddr __user *)a[4],
2412 (int __user *)a[5]);
2413 break;
2414 case SYS_SHUTDOWN:
2415 err = sys_shutdown(a0, a1);
2416 break;
2417 case SYS_SETSOCKOPT:
2418 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2419 break;
2420 case SYS_GETSOCKOPT:
2421 err =
2422 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2423 (int __user *)a[4]);
2424 break;
2425 case SYS_SENDMSG:
2426 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2427 break;
2428 case SYS_SENDMMSG:
2429 err = sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3]);
2430 break;
2431 case SYS_RECVMSG:
2432 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2433 break;
2434 case SYS_RECVMMSG:
2435 err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2436 (struct timespec __user *)a[4]);
2437 break;
2438 case SYS_ACCEPT4:
2439 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2440 (int __user *)a[2], a[3]);
2441 break;
2442 default:
2443 err = -EINVAL;
2444 break;
2445 }
2446 return err;
2447}
2448
2449#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2450
2451/**
2452 * sock_register - add a socket protocol handler
2453 * @ops: description of protocol
2454 *
2455 * This function is called by a protocol handler that wants to
2456 * advertise its address family, and have it linked into the
2457 * socket interface. The value ops->family coresponds to the
2458 * socket system call protocol family.
2459 */
2460int sock_register(const struct net_proto_family *ops)
2461{
2462 int err;
2463
2464 if (ops->family >= NPROTO) {
2465 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2466 NPROTO);
2467 return -ENOBUFS;
2468 }
2469
2470 spin_lock(&net_family_lock);
2471 if (rcu_dereference_protected(net_families[ops->family],
2472 lockdep_is_held(&net_family_lock)))
2473 err = -EEXIST;
2474 else {
2475 rcu_assign_pointer(net_families[ops->family], ops);
2476 err = 0;
2477 }
2478 spin_unlock(&net_family_lock);
2479
2480 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2481 return err;
2482}
2483EXPORT_SYMBOL(sock_register);
2484
2485/**
2486 * sock_unregister - remove a protocol handler
2487 * @family: protocol family to remove
2488 *
2489 * This function is called by a protocol handler that wants to
2490 * remove its address family, and have it unlinked from the
2491 * new socket creation.
2492 *
2493 * If protocol handler is a module, then it can use module reference
2494 * counts to protect against new references. If protocol handler is not
2495 * a module then it needs to provide its own protection in
2496 * the ops->create routine.
2497 */
2498void sock_unregister(int family)
2499{
2500 BUG_ON(family < 0 || family >= NPROTO);
2501
2502 spin_lock(&net_family_lock);
2503 rcu_assign_pointer(net_families[family], NULL);
2504 spin_unlock(&net_family_lock);
2505
2506 synchronize_rcu();
2507
2508 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2509}
2510EXPORT_SYMBOL(sock_unregister);
2511
2512static int __init sock_init(void)
2513{
2514 int err;
2515
2516 /*
2517 * Initialize sock SLAB cache.
2518 */
2519
2520 sk_init();
2521
2522 /*
2523 * Initialize skbuff SLAB cache
2524 */
2525 skb_init();
2526
2527 /*
2528 * Initialize the protocols module.
2529 */
2530
2531 init_inodecache();
2532
2533 err = register_filesystem(&sock_fs_type);
2534 if (err)
2535 goto out_fs;
2536 sock_mnt = kern_mount(&sock_fs_type);
2537 if (IS_ERR(sock_mnt)) {
2538 err = PTR_ERR(sock_mnt);
2539 goto out_mount;
2540 }
2541
2542 /* The real protocol initialization is performed in later initcalls.
2543 */
2544
2545#ifdef CONFIG_NETFILTER
2546 netfilter_init();
2547#endif
2548
2549#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2550 skb_timestamping_init();
2551#endif
2552
2553out:
2554 return err;
2555
2556out_mount:
2557 unregister_filesystem(&sock_fs_type);
2558out_fs:
2559 goto out;
2560}
2561
2562core_initcall(sock_init); /* early initcall */
2563
2564#ifdef CONFIG_PROC_FS
2565void socket_seq_show(struct seq_file *seq)
2566{
2567 int cpu;
2568 int counter = 0;
2569
2570 for_each_possible_cpu(cpu)
2571 counter += per_cpu(sockets_in_use, cpu);
2572
2573 /* It can be negative, by the way. 8) */
2574 if (counter < 0)
2575 counter = 0;
2576
2577 seq_printf(seq, "sockets: used %d\n", counter);
2578}
2579#endif /* CONFIG_PROC_FS */
2580
2581#ifdef CONFIG_COMPAT
2582static int do_siocgstamp(struct net *net, struct socket *sock,
2583 unsigned int cmd, struct compat_timeval __user *up)
2584{
2585 mm_segment_t old_fs = get_fs();
2586 struct timeval ktv;
2587 int err;
2588
2589 set_fs(KERNEL_DS);
2590 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2591 set_fs(old_fs);
2592 if (!err) {
2593 err = put_user(ktv.tv_sec, &up->tv_sec);
2594 err |= __put_user(ktv.tv_usec, &up->tv_usec);
2595 }
2596 return err;
2597}
2598
2599static int do_siocgstampns(struct net *net, struct socket *sock,
2600 unsigned int cmd, struct compat_timespec __user *up)
2601{
2602 mm_segment_t old_fs = get_fs();
2603 struct timespec kts;
2604 int err;
2605
2606 set_fs(KERNEL_DS);
2607 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2608 set_fs(old_fs);
2609 if (!err) {
2610 err = put_user(kts.tv_sec, &up->tv_sec);
2611 err |= __put_user(kts.tv_nsec, &up->tv_nsec);
2612 }
2613 return err;
2614}
2615
2616static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2617{
2618 struct ifreq __user *uifr;
2619 int err;
2620
2621 uifr = compat_alloc_user_space(sizeof(struct ifreq));
2622 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2623 return -EFAULT;
2624
2625 err = dev_ioctl(net, SIOCGIFNAME, uifr);
2626 if (err)
2627 return err;
2628
2629 if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2630 return -EFAULT;
2631
2632 return 0;
2633}
2634
2635static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2636{
2637 struct compat_ifconf ifc32;
2638 struct ifconf ifc;
2639 struct ifconf __user *uifc;
2640 struct compat_ifreq __user *ifr32;
2641 struct ifreq __user *ifr;
2642 unsigned int i, j;
2643 int err;
2644
2645 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2646 return -EFAULT;
2647
2648 if (ifc32.ifcbuf == 0) {
2649 ifc32.ifc_len = 0;
2650 ifc.ifc_len = 0;
2651 ifc.ifc_req = NULL;
2652 uifc = compat_alloc_user_space(sizeof(struct ifconf));
2653 } else {
2654 size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2655 sizeof(struct ifreq);
2656 uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2657 ifc.ifc_len = len;
2658 ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2659 ifr32 = compat_ptr(ifc32.ifcbuf);
2660 for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2661 if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2662 return -EFAULT;
2663 ifr++;
2664 ifr32++;
2665 }
2666 }
2667 if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2668 return -EFAULT;
2669
2670 err = dev_ioctl(net, SIOCGIFCONF, uifc);
2671 if (err)
2672 return err;
2673
2674 if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2675 return -EFAULT;
2676
2677 ifr = ifc.ifc_req;
2678 ifr32 = compat_ptr(ifc32.ifcbuf);
2679 for (i = 0, j = 0;
2680 i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2681 i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2682 if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2683 return -EFAULT;
2684 ifr32++;
2685 ifr++;
2686 }
2687
2688 if (ifc32.ifcbuf == 0) {
2689 /* Translate from 64-bit structure multiple to
2690 * a 32-bit one.
2691 */
2692 i = ifc.ifc_len;
2693 i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2694 ifc32.ifc_len = i;
2695 } else {
2696 ifc32.ifc_len = i;
2697 }
2698 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2699 return -EFAULT;
2700
2701 return 0;
2702}
2703
2704static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2705{
2706 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2707 bool convert_in = false, convert_out = false;
2708 size_t buf_size = ALIGN(sizeof(struct ifreq), 8);
2709 struct ethtool_rxnfc __user *rxnfc;
2710 struct ifreq __user *ifr;
2711 u32 rule_cnt = 0, actual_rule_cnt;
2712 u32 ethcmd;
2713 u32 data;
2714 int ret;
2715
2716 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2717 return -EFAULT;
2718
2719 compat_rxnfc = compat_ptr(data);
2720
2721 if (get_user(ethcmd, &compat_rxnfc->cmd))
2722 return -EFAULT;
2723
2724 /* Most ethtool structures are defined without padding.
2725 * Unfortunately struct ethtool_rxnfc is an exception.
2726 */
2727 switch (ethcmd) {
2728 default:
2729 break;
2730 case ETHTOOL_GRXCLSRLALL:
2731 /* Buffer size is variable */
2732 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2733 return -EFAULT;
2734 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2735 return -ENOMEM;
2736 buf_size += rule_cnt * sizeof(u32);
2737 /* fall through */
2738 case ETHTOOL_GRXRINGS:
2739 case ETHTOOL_GRXCLSRLCNT:
2740 case ETHTOOL_GRXCLSRULE:
2741 convert_out = true;
2742 /* fall through */
2743 case ETHTOOL_SRXCLSRLDEL:
2744 case ETHTOOL_SRXCLSRLINS:
2745 buf_size += sizeof(struct ethtool_rxnfc);
2746 convert_in = true;
2747 break;
2748 }
2749
2750 ifr = compat_alloc_user_space(buf_size);
2751 rxnfc = (void *)ifr + ALIGN(sizeof(struct ifreq), 8);
2752
2753 if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2754 return -EFAULT;
2755
2756 if (put_user(convert_in ? rxnfc : compat_ptr(data),
2757 &ifr->ifr_ifru.ifru_data))
2758 return -EFAULT;
2759
2760 if (convert_in) {
2761 /* We expect there to be holes between fs.m_ext and
2762 * fs.ring_cookie and at the end of fs, but nowhere else.
2763 */
2764 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
2765 sizeof(compat_rxnfc->fs.m_ext) !=
2766 offsetof(struct ethtool_rxnfc, fs.m_ext) +
2767 sizeof(rxnfc->fs.m_ext));
2768 BUILD_BUG_ON(
2769 offsetof(struct compat_ethtool_rxnfc, fs.location) -
2770 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2771 offsetof(struct ethtool_rxnfc, fs.location) -
2772 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2773
2774 if (copy_in_user(rxnfc, compat_rxnfc,
2775 (void *)(&rxnfc->fs.m_ext + 1) -
2776 (void *)rxnfc) ||
2777 copy_in_user(&rxnfc->fs.ring_cookie,
2778 &compat_rxnfc->fs.ring_cookie,
2779 (void *)(&rxnfc->fs.location + 1) -
2780 (void *)&rxnfc->fs.ring_cookie) ||
2781 copy_in_user(&rxnfc->rule_cnt, &compat_rxnfc->rule_cnt,
2782 sizeof(rxnfc->rule_cnt)))
2783 return -EFAULT;
2784 }
2785
2786 ret = dev_ioctl(net, SIOCETHTOOL, ifr);
2787 if (ret)
2788 return ret;
2789
2790 if (convert_out) {
2791 if (copy_in_user(compat_rxnfc, rxnfc,
2792 (const void *)(&rxnfc->fs.m_ext + 1) -
2793 (const void *)rxnfc) ||
2794 copy_in_user(&compat_rxnfc->fs.ring_cookie,
2795 &rxnfc->fs.ring_cookie,
2796 (const void *)(&rxnfc->fs.location + 1) -
2797 (const void *)&rxnfc->fs.ring_cookie) ||
2798 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2799 sizeof(rxnfc->rule_cnt)))
2800 return -EFAULT;
2801
2802 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2803 /* As an optimisation, we only copy the actual
2804 * number of rules that the underlying
2805 * function returned. Since Mallory might
2806 * change the rule count in user memory, we
2807 * check that it is less than the rule count
2808 * originally given (as the user buffer size),
2809 * which has been range-checked.
2810 */
2811 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2812 return -EFAULT;
2813 if (actual_rule_cnt < rule_cnt)
2814 rule_cnt = actual_rule_cnt;
2815 if (copy_in_user(&compat_rxnfc->rule_locs[0],
2816 &rxnfc->rule_locs[0],
2817 rule_cnt * sizeof(u32)))
2818 return -EFAULT;
2819 }
2820 }
2821
2822 return 0;
2823}
2824
2825static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2826{
2827 void __user *uptr;
2828 compat_uptr_t uptr32;
2829 struct ifreq __user *uifr;
2830
2831 uifr = compat_alloc_user_space(sizeof(*uifr));
2832 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2833 return -EFAULT;
2834
2835 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2836 return -EFAULT;
2837
2838 uptr = compat_ptr(uptr32);
2839
2840 if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2841 return -EFAULT;
2842
2843 return dev_ioctl(net, SIOCWANDEV, uifr);
2844}
2845
2846static int bond_ioctl(struct net *net, unsigned int cmd,
2847 struct compat_ifreq __user *ifr32)
2848{
2849 struct ifreq kifr;
2850 struct ifreq __user *uifr;
2851 mm_segment_t old_fs;
2852 int err;
2853 u32 data;
2854 void __user *datap;
2855
2856 switch (cmd) {
2857 case SIOCBONDENSLAVE:
2858 case SIOCBONDRELEASE:
2859 case SIOCBONDSETHWADDR:
2860 case SIOCBONDCHANGEACTIVE:
2861 if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2862 return -EFAULT;
2863
2864 old_fs = get_fs();
2865 set_fs(KERNEL_DS);
2866 err = dev_ioctl(net, cmd,
2867 (struct ifreq __user __force *) &kifr);
2868 set_fs(old_fs);
2869
2870 return err;
2871 case SIOCBONDSLAVEINFOQUERY:
2872 case SIOCBONDINFOQUERY:
2873 uifr = compat_alloc_user_space(sizeof(*uifr));
2874 if (copy_in_user(&uifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2875 return -EFAULT;
2876
2877 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2878 return -EFAULT;
2879
2880 datap = compat_ptr(data);
2881 if (put_user(datap, &uifr->ifr_ifru.ifru_data))
2882 return -EFAULT;
2883
2884 return dev_ioctl(net, cmd, uifr);
2885 default:
2886 return -EINVAL;
2887 }
2888}
2889
2890static int siocdevprivate_ioctl(struct net *net, unsigned int cmd,
2891 struct compat_ifreq __user *u_ifreq32)
2892{
2893 struct ifreq __user *u_ifreq64;
2894 char tmp_buf[IFNAMSIZ];
2895 void __user *data64;
2896 u32 data32;
2897
2898 if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2899 IFNAMSIZ))
2900 return -EFAULT;
2901 if (__get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2902 return -EFAULT;
2903 data64 = compat_ptr(data32);
2904
2905 u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2906
2907 /* Don't check these user accesses, just let that get trapped
2908 * in the ioctl handler instead.
2909 */
2910 if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2911 IFNAMSIZ))
2912 return -EFAULT;
2913 if (__put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2914 return -EFAULT;
2915
2916 return dev_ioctl(net, cmd, u_ifreq64);
2917}
2918
2919static int dev_ifsioc(struct net *net, struct socket *sock,
2920 unsigned int cmd, struct compat_ifreq __user *uifr32)
2921{
2922 struct ifreq __user *uifr;
2923 int err;
2924
2925 uifr = compat_alloc_user_space(sizeof(*uifr));
2926 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2927 return -EFAULT;
2928
2929 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2930
2931 if (!err) {
2932 switch (cmd) {
2933 case SIOCGIFFLAGS:
2934 case SIOCGIFMETRIC:
2935 case SIOCGIFMTU:
2936 case SIOCGIFMEM:
2937 case SIOCGIFHWADDR:
2938 case SIOCGIFINDEX:
2939 case SIOCGIFADDR:
2940 case SIOCGIFBRDADDR:
2941 case SIOCGIFDSTADDR:
2942 case SIOCGIFNETMASK:
2943 case SIOCGIFPFLAGS:
2944 case SIOCGIFTXQLEN:
2945 case SIOCGMIIPHY:
2946 case SIOCGMIIREG:
2947 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2948 err = -EFAULT;
2949 break;
2950 }
2951 }
2952 return err;
2953}
2954
2955static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2956 struct compat_ifreq __user *uifr32)
2957{
2958 struct ifreq ifr;
2959 struct compat_ifmap __user *uifmap32;
2960 mm_segment_t old_fs;
2961 int err;
2962
2963 uifmap32 = &uifr32->ifr_ifru.ifru_map;
2964 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2965 err |= __get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2966 err |= __get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2967 err |= __get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2968 err |= __get_user(ifr.ifr_map.irq, &uifmap32->irq);
2969 err |= __get_user(ifr.ifr_map.dma, &uifmap32->dma);
2970 err |= __get_user(ifr.ifr_map.port, &uifmap32->port);
2971 if (err)
2972 return -EFAULT;
2973
2974 old_fs = get_fs();
2975 set_fs(KERNEL_DS);
2976 err = dev_ioctl(net, cmd, (void __user __force *)&ifr);
2977 set_fs(old_fs);
2978
2979 if (cmd == SIOCGIFMAP && !err) {
2980 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2981 err |= __put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2982 err |= __put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2983 err |= __put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2984 err |= __put_user(ifr.ifr_map.irq, &uifmap32->irq);
2985 err |= __put_user(ifr.ifr_map.dma, &uifmap32->dma);
2986 err |= __put_user(ifr.ifr_map.port, &uifmap32->port);
2987 if (err)
2988 err = -EFAULT;
2989 }
2990 return err;
2991}
2992
2993static int compat_siocshwtstamp(struct net *net, struct compat_ifreq __user *uifr32)
2994{
2995 void __user *uptr;
2996 compat_uptr_t uptr32;
2997 struct ifreq __user *uifr;
2998
2999 uifr = compat_alloc_user_space(sizeof(*uifr));
3000 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
3001 return -EFAULT;
3002
3003 if (get_user(uptr32, &uifr32->ifr_data))
3004 return -EFAULT;
3005
3006 uptr = compat_ptr(uptr32);
3007
3008 if (put_user(uptr, &uifr->ifr_data))
3009 return -EFAULT;
3010
3011 return dev_ioctl(net, SIOCSHWTSTAMP, uifr);
3012}
3013
3014struct rtentry32 {
3015 u32 rt_pad1;
3016 struct sockaddr rt_dst; /* target address */
3017 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
3018 struct sockaddr rt_genmask; /* target network mask (IP) */
3019 unsigned short rt_flags;
3020 short rt_pad2;
3021 u32 rt_pad3;
3022 unsigned char rt_tos;
3023 unsigned char rt_class;
3024 short rt_pad4;
3025 short rt_metric; /* +1 for binary compatibility! */
3026 /* char * */ u32 rt_dev; /* forcing the device at add */
3027 u32 rt_mtu; /* per route MTU/Window */
3028 u32 rt_window; /* Window clamping */
3029 unsigned short rt_irtt; /* Initial RTT */
3030};
3031
3032struct in6_rtmsg32 {
3033 struct in6_addr rtmsg_dst;
3034 struct in6_addr rtmsg_src;
3035 struct in6_addr rtmsg_gateway;
3036 u32 rtmsg_type;
3037 u16 rtmsg_dst_len;
3038 u16 rtmsg_src_len;
3039 u32 rtmsg_metric;
3040 u32 rtmsg_info;
3041 u32 rtmsg_flags;
3042 s32 rtmsg_ifindex;
3043};
3044
3045static int routing_ioctl(struct net *net, struct socket *sock,
3046 unsigned int cmd, void __user *argp)
3047{
3048 int ret;
3049 void *r = NULL;
3050 struct in6_rtmsg r6;
3051 struct rtentry r4;
3052 char devname[16];
3053 u32 rtdev;
3054 mm_segment_t old_fs = get_fs();
3055
3056 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3057 struct in6_rtmsg32 __user *ur6 = argp;
3058 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3059 3 * sizeof(struct in6_addr));
3060 ret |= __get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3061 ret |= __get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3062 ret |= __get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3063 ret |= __get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3064 ret |= __get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3065 ret |= __get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3066 ret |= __get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3067
3068 r = (void *) &r6;
3069 } else { /* ipv4 */
3070 struct rtentry32 __user *ur4 = argp;
3071 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3072 3 * sizeof(struct sockaddr));
3073 ret |= __get_user(r4.rt_flags, &(ur4->rt_flags));
3074 ret |= __get_user(r4.rt_metric, &(ur4->rt_metric));
3075 ret |= __get_user(r4.rt_mtu, &(ur4->rt_mtu));
3076 ret |= __get_user(r4.rt_window, &(ur4->rt_window));
3077 ret |= __get_user(r4.rt_irtt, &(ur4->rt_irtt));
3078 ret |= __get_user(rtdev, &(ur4->rt_dev));
3079 if (rtdev) {
3080 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3081 r4.rt_dev = (char __user __force *)devname;
3082 devname[15] = 0;
3083 } else
3084 r4.rt_dev = NULL;
3085
3086 r = (void *) &r4;
3087 }
3088
3089 if (ret) {
3090 ret = -EFAULT;
3091 goto out;
3092 }
3093
3094 set_fs(KERNEL_DS);
3095 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3096 set_fs(old_fs);
3097
3098out:
3099 return ret;
3100}
3101
3102/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3103 * for some operations; this forces use of the newer bridge-utils that
3104 * use compatible ioctls
3105 */
3106static int old_bridge_ioctl(compat_ulong_t __user *argp)
3107{
3108 compat_ulong_t tmp;
3109
3110 if (get_user(tmp, argp))
3111 return -EFAULT;
3112 if (tmp == BRCTL_GET_VERSION)
3113 return BRCTL_VERSION + 1;
3114 return -EINVAL;
3115}
3116
3117static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3118 unsigned int cmd, unsigned long arg)
3119{
3120 void __user *argp = compat_ptr(arg);
3121 struct sock *sk = sock->sk;
3122 struct net *net = sock_net(sk);
3123
3124 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3125 return siocdevprivate_ioctl(net, cmd, argp);
3126
3127 switch (cmd) {
3128 case SIOCSIFBR:
3129 case SIOCGIFBR:
3130 return old_bridge_ioctl(argp);
3131 case SIOCGIFNAME:
3132 return dev_ifname32(net, argp);
3133 case SIOCGIFCONF:
3134 return dev_ifconf(net, argp);
3135 case SIOCETHTOOL:
3136 return ethtool_ioctl(net, argp);
3137 case SIOCWANDEV:
3138 return compat_siocwandev(net, argp);
3139 case SIOCGIFMAP:
3140 case SIOCSIFMAP:
3141 return compat_sioc_ifmap(net, cmd, argp);
3142 case SIOCBONDENSLAVE:
3143 case SIOCBONDRELEASE:
3144 case SIOCBONDSETHWADDR:
3145 case SIOCBONDSLAVEINFOQUERY:
3146 case SIOCBONDINFOQUERY:
3147 case SIOCBONDCHANGEACTIVE:
3148 return bond_ioctl(net, cmd, argp);
3149 case SIOCADDRT:
3150 case SIOCDELRT:
3151 return routing_ioctl(net, sock, cmd, argp);
3152 case SIOCGSTAMP:
3153 return do_siocgstamp(net, sock, cmd, argp);
3154 case SIOCGSTAMPNS:
3155 return do_siocgstampns(net, sock, cmd, argp);
3156 case SIOCSHWTSTAMP:
3157 return compat_siocshwtstamp(net, argp);
3158
3159 case FIOSETOWN:
3160 case SIOCSPGRP:
3161 case FIOGETOWN:
3162 case SIOCGPGRP:
3163 case SIOCBRADDBR:
3164 case SIOCBRDELBR:
3165 case SIOCGIFVLAN:
3166 case SIOCSIFVLAN:
3167 case SIOCADDDLCI:
3168 case SIOCDELDLCI:
3169 return sock_ioctl(file, cmd, arg);
3170
3171 case SIOCGIFFLAGS:
3172 case SIOCSIFFLAGS:
3173 case SIOCGIFMETRIC:
3174 case SIOCSIFMETRIC:
3175 case SIOCGIFMTU:
3176 case SIOCSIFMTU:
3177 case SIOCGIFMEM:
3178 case SIOCSIFMEM:
3179 case SIOCGIFHWADDR:
3180 case SIOCSIFHWADDR:
3181 case SIOCADDMULTI:
3182 case SIOCDELMULTI:
3183 case SIOCGIFINDEX:
3184 case SIOCGIFADDR:
3185 case SIOCSIFADDR:
3186 case SIOCSIFHWBROADCAST:
3187 case SIOCDIFADDR:
3188 case SIOCGIFBRDADDR:
3189 case SIOCSIFBRDADDR:
3190 case SIOCGIFDSTADDR:
3191 case SIOCSIFDSTADDR:
3192 case SIOCGIFNETMASK:
3193 case SIOCSIFNETMASK:
3194 case SIOCSIFPFLAGS:
3195 case SIOCGIFPFLAGS:
3196 case SIOCGIFTXQLEN:
3197 case SIOCSIFTXQLEN:
3198 case SIOCBRADDIF:
3199 case SIOCBRDELIF:
3200 case SIOCSIFNAME:
3201 case SIOCGMIIPHY:
3202 case SIOCGMIIREG:
3203 case SIOCSMIIREG:
3204 return dev_ifsioc(net, sock, cmd, argp);
3205
3206 case SIOCSARP:
3207 case SIOCGARP:
3208 case SIOCDARP:
3209 case SIOCATMARK:
3210 return sock_do_ioctl(net, sock, cmd, arg);
3211 }
3212
3213 /* Prevent warning from compat_sys_ioctl, these always
3214 * result in -EINVAL in the native case anyway. */
3215 switch (cmd) {
3216 case SIOCRTMSG:
3217 case SIOCGIFCOUNT:
3218 case SIOCSRARP:
3219 case SIOCGRARP:
3220 case SIOCDRARP:
3221 case SIOCSIFLINK:
3222 case SIOCGIFSLAVE:
3223 case SIOCSIFSLAVE:
3224 return -EINVAL;
3225 }
3226
3227 return -ENOIOCTLCMD;
3228}
3229
3230static long compat_sock_ioctl(struct file *file, unsigned cmd,
3231 unsigned long arg)
3232{
3233 struct socket *sock = file->private_data;
3234 int ret = -ENOIOCTLCMD;
3235 struct sock *sk;
3236 struct net *net;
3237
3238 sk = sock->sk;
3239 net = sock_net(sk);
3240
3241 if (sock->ops->compat_ioctl)
3242 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3243
3244 if (ret == -ENOIOCTLCMD &&
3245 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3246 ret = compat_wext_handle_ioctl(net, cmd, arg);
3247
3248 if (ret == -ENOIOCTLCMD)
3249 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3250
3251 return ret;
3252}
3253#endif
3254
3255int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3256{
3257 return sock->ops->bind(sock, addr, addrlen);
3258}
3259EXPORT_SYMBOL(kernel_bind);
3260
3261int kernel_listen(struct socket *sock, int backlog)
3262{
3263 return sock->ops->listen(sock, backlog);
3264}
3265EXPORT_SYMBOL(kernel_listen);
3266
3267int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3268{
3269 struct sock *sk = sock->sk;
3270 int err;
3271
3272 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3273 newsock);
3274 if (err < 0)
3275 goto done;
3276
3277 err = sock->ops->accept(sock, *newsock, flags);
3278 if (err < 0) {
3279 sock_release(*newsock);
3280 *newsock = NULL;
3281 goto done;
3282 }
3283
3284 (*newsock)->ops = sock->ops;
3285 __module_get((*newsock)->ops->owner);
3286
3287done:
3288 return err;
3289}
3290EXPORT_SYMBOL(kernel_accept);
3291
3292int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3293 int flags)
3294{
3295 return sock->ops->connect(sock, addr, addrlen, flags);
3296}
3297EXPORT_SYMBOL(kernel_connect);
3298
3299int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3300 int *addrlen)
3301{
3302 return sock->ops->getname(sock, addr, addrlen, 0);
3303}
3304EXPORT_SYMBOL(kernel_getsockname);
3305
3306int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3307 int *addrlen)
3308{
3309 return sock->ops->getname(sock, addr, addrlen, 1);
3310}
3311EXPORT_SYMBOL(kernel_getpeername);
3312
3313int kernel_getsockopt(struct socket *sock, int level, int optname,
3314 char *optval, int *optlen)
3315{
3316 mm_segment_t oldfs = get_fs();
3317 char __user *uoptval;
3318 int __user *uoptlen;
3319 int err;
3320
3321 uoptval = (char __user __force *) optval;
3322 uoptlen = (int __user __force *) optlen;
3323
3324 set_fs(KERNEL_DS);
3325 if (level == SOL_SOCKET)
3326 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3327 else
3328 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3329 uoptlen);
3330 set_fs(oldfs);
3331 return err;
3332}
3333EXPORT_SYMBOL(kernel_getsockopt);
3334
3335int kernel_setsockopt(struct socket *sock, int level, int optname,
3336 char *optval, unsigned int optlen)
3337{
3338 mm_segment_t oldfs = get_fs();
3339 char __user *uoptval;
3340 int err;
3341
3342 uoptval = (char __user __force *) optval;
3343
3344 set_fs(KERNEL_DS);
3345 if (level == SOL_SOCKET)
3346 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3347 else
3348 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3349 optlen);
3350 set_fs(oldfs);
3351 return err;
3352}
3353EXPORT_SYMBOL(kernel_setsockopt);
3354
3355int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3356 size_t size, int flags)
3357{
3358 sock_update_classid(sock->sk);
3359
3360 if (sock->ops->sendpage)
3361 return sock->ops->sendpage(sock, page, offset, size, flags);
3362
3363 return sock_no_sendpage(sock, page, offset, size, flags);
3364}
3365EXPORT_SYMBOL(kernel_sendpage);
3366
3367int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3368{
3369 mm_segment_t oldfs = get_fs();
3370 int err;
3371
3372 set_fs(KERNEL_DS);
3373 err = sock->ops->ioctl(sock, cmd, arg);
3374 set_fs(oldfs);
3375
3376 return err;
3377}
3378EXPORT_SYMBOL(kernel_sock_ioctl);
3379
3380int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3381{
3382 return sock->ops->shutdown(sock, how);
3383}
3384EXPORT_SYMBOL(kernel_sock_shutdown);