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// 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/bpf-cgroup.h>
56#include <linux/ethtool.h>
57#include <linux/mm.h>
58#include <linux/socket.h>
59#include <linux/file.h>
60#include <linux/splice.h>
61#include <linux/net.h>
62#include <linux/interrupt.h>
63#include <linux/thread_info.h>
64#include <linux/rcupdate.h>
65#include <linux/netdevice.h>
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68#include <linux/mutex.h>
69#include <linux/if_bridge.h>
70#include <linux/if_vlan.h>
71#include <linux/ptp_classify.h>
72#include <linux/init.h>
73#include <linux/poll.h>
74#include <linux/cache.h>
75#include <linux/module.h>
76#include <linux/highmem.h>
77#include <linux/mount.h>
78#include <linux/pseudo_fs.h>
79#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/compat.h>
82#include <linux/kmod.h>
83#include <linux/audit.h>
84#include <linux/wireless.h>
85#include <linux/nsproxy.h>
86#include <linux/magic.h>
87#include <linux/slab.h>
88#include <linux/xattr.h>
89#include <linux/nospec.h>
90#include <linux/indirect_call_wrapper.h>
91#include <linux/io_uring/net.h>
92
93#include <linux/uaccess.h>
94#include <asm/unistd.h>
95
96#include <net/compat.h>
97#include <net/wext.h>
98#include <net/cls_cgroup.h>
99
100#include <net/sock.h>
101#include <linux/netfilter.h>
102
103#include <linux/if_tun.h>
104#include <linux/ipv6_route.h>
105#include <linux/route.h>
106#include <linux/termios.h>
107#include <linux/sockios.h>
108#include <net/busy_poll.h>
109#include <linux/errqueue.h>
110#include <linux/ptp_clock_kernel.h>
111#include <trace/events/sock.h>
112
113#ifdef CONFIG_NET_RX_BUSY_POLL
114unsigned int sysctl_net_busy_read __read_mostly;
115unsigned int sysctl_net_busy_poll __read_mostly;
116#endif
117
118static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
119static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
120static int sock_mmap(struct file *file, struct vm_area_struct *vma);
121
122static int sock_close(struct inode *inode, struct file *file);
123static __poll_t sock_poll(struct file *file,
124 struct poll_table_struct *wait);
125static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
126#ifdef CONFIG_COMPAT
127static long compat_sock_ioctl(struct file *file,
128 unsigned int cmd, unsigned long arg);
129#endif
130static int sock_fasync(int fd, struct file *filp, int on);
131static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
133 unsigned int flags);
134static void sock_splice_eof(struct file *file);
135
136#ifdef CONFIG_PROC_FS
137static void sock_show_fdinfo(struct seq_file *m, struct file *f)
138{
139 struct socket *sock = f->private_data;
140 const struct proto_ops *ops = READ_ONCE(sock->ops);
141
142 if (ops->show_fdinfo)
143 ops->show_fdinfo(m, sock);
144}
145#else
146#define sock_show_fdinfo NULL
147#endif
148
149/*
150 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
151 * in the operation structures but are done directly via the socketcall() multiplexor.
152 */
153
154static const struct file_operations socket_file_ops = {
155 .owner = THIS_MODULE,
156 .read_iter = sock_read_iter,
157 .write_iter = sock_write_iter,
158 .poll = sock_poll,
159 .unlocked_ioctl = sock_ioctl,
160#ifdef CONFIG_COMPAT
161 .compat_ioctl = compat_sock_ioctl,
162#endif
163 .uring_cmd = io_uring_cmd_sock,
164 .mmap = sock_mmap,
165 .release = sock_close,
166 .fasync = sock_fasync,
167 .splice_write = splice_to_socket,
168 .splice_read = sock_splice_read,
169 .splice_eof = sock_splice_eof,
170 .show_fdinfo = sock_show_fdinfo,
171};
172
173static const char * const pf_family_names[] = {
174 [PF_UNSPEC] = "PF_UNSPEC",
175 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
176 [PF_INET] = "PF_INET",
177 [PF_AX25] = "PF_AX25",
178 [PF_IPX] = "PF_IPX",
179 [PF_APPLETALK] = "PF_APPLETALK",
180 [PF_NETROM] = "PF_NETROM",
181 [PF_BRIDGE] = "PF_BRIDGE",
182 [PF_ATMPVC] = "PF_ATMPVC",
183 [PF_X25] = "PF_X25",
184 [PF_INET6] = "PF_INET6",
185 [PF_ROSE] = "PF_ROSE",
186 [PF_DECnet] = "PF_DECnet",
187 [PF_NETBEUI] = "PF_NETBEUI",
188 [PF_SECURITY] = "PF_SECURITY",
189 [PF_KEY] = "PF_KEY",
190 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
191 [PF_PACKET] = "PF_PACKET",
192 [PF_ASH] = "PF_ASH",
193 [PF_ECONET] = "PF_ECONET",
194 [PF_ATMSVC] = "PF_ATMSVC",
195 [PF_RDS] = "PF_RDS",
196 [PF_SNA] = "PF_SNA",
197 [PF_IRDA] = "PF_IRDA",
198 [PF_PPPOX] = "PF_PPPOX",
199 [PF_WANPIPE] = "PF_WANPIPE",
200 [PF_LLC] = "PF_LLC",
201 [PF_IB] = "PF_IB",
202 [PF_MPLS] = "PF_MPLS",
203 [PF_CAN] = "PF_CAN",
204 [PF_TIPC] = "PF_TIPC",
205 [PF_BLUETOOTH] = "PF_BLUETOOTH",
206 [PF_IUCV] = "PF_IUCV",
207 [PF_RXRPC] = "PF_RXRPC",
208 [PF_ISDN] = "PF_ISDN",
209 [PF_PHONET] = "PF_PHONET",
210 [PF_IEEE802154] = "PF_IEEE802154",
211 [PF_CAIF] = "PF_CAIF",
212 [PF_ALG] = "PF_ALG",
213 [PF_NFC] = "PF_NFC",
214 [PF_VSOCK] = "PF_VSOCK",
215 [PF_KCM] = "PF_KCM",
216 [PF_QIPCRTR] = "PF_QIPCRTR",
217 [PF_SMC] = "PF_SMC",
218 [PF_XDP] = "PF_XDP",
219 [PF_MCTP] = "PF_MCTP",
220};
221
222/*
223 * The protocol list. Each protocol is registered in here.
224 */
225
226static DEFINE_SPINLOCK(net_family_lock);
227static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
228
229/*
230 * Support routines.
231 * Move socket addresses back and forth across the kernel/user
232 * divide and look after the messy bits.
233 */
234
235/**
236 * move_addr_to_kernel - copy a socket address into kernel space
237 * @uaddr: Address in user space
238 * @kaddr: Address in kernel space
239 * @ulen: Length in user space
240 *
241 * The address is copied into kernel space. If the provided address is
242 * too long an error code of -EINVAL is returned. If the copy gives
243 * invalid addresses -EFAULT is returned. On a success 0 is returned.
244 */
245
246int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
247{
248 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
249 return -EINVAL;
250 if (ulen == 0)
251 return 0;
252 if (copy_from_user(kaddr, uaddr, ulen))
253 return -EFAULT;
254 return audit_sockaddr(ulen, kaddr);
255}
256
257/**
258 * move_addr_to_user - copy an address to user space
259 * @kaddr: kernel space address
260 * @klen: length of address in kernel
261 * @uaddr: user space address
262 * @ulen: pointer to user length field
263 *
264 * The value pointed to by ulen on entry is the buffer length available.
265 * This is overwritten with the buffer space used. -EINVAL is returned
266 * if an overlong buffer is specified or a negative buffer size. -EFAULT
267 * is returned if either the buffer or the length field are not
268 * accessible.
269 * After copying the data up to the limit the user specifies, the true
270 * length of the data is written over the length limit the user
271 * specified. Zero is returned for a success.
272 */
273
274static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
275 void __user *uaddr, int __user *ulen)
276{
277 int err;
278 int len;
279
280 BUG_ON(klen > sizeof(struct sockaddr_storage));
281 err = get_user(len, ulen);
282 if (err)
283 return err;
284 if (len > klen)
285 len = klen;
286 if (len < 0)
287 return -EINVAL;
288 if (len) {
289 if (audit_sockaddr(klen, kaddr))
290 return -ENOMEM;
291 if (copy_to_user(uaddr, kaddr, len))
292 return -EFAULT;
293 }
294 /*
295 * "fromlen shall refer to the value before truncation.."
296 * 1003.1g
297 */
298 return __put_user(klen, ulen);
299}
300
301static struct kmem_cache *sock_inode_cachep __ro_after_init;
302
303static struct inode *sock_alloc_inode(struct super_block *sb)
304{
305 struct socket_alloc *ei;
306
307 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
308 if (!ei)
309 return NULL;
310 init_waitqueue_head(&ei->socket.wq.wait);
311 ei->socket.wq.fasync_list = NULL;
312 ei->socket.wq.flags = 0;
313
314 ei->socket.state = SS_UNCONNECTED;
315 ei->socket.flags = 0;
316 ei->socket.ops = NULL;
317 ei->socket.sk = NULL;
318 ei->socket.file = NULL;
319
320 return &ei->vfs_inode;
321}
322
323static void sock_free_inode(struct inode *inode)
324{
325 struct socket_alloc *ei;
326
327 ei = container_of(inode, struct socket_alloc, vfs_inode);
328 kmem_cache_free(sock_inode_cachep, ei);
329}
330
331static void init_once(void *foo)
332{
333 struct socket_alloc *ei = (struct socket_alloc *)foo;
334
335 inode_init_once(&ei->vfs_inode);
336}
337
338static void init_inodecache(void)
339{
340 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
341 sizeof(struct socket_alloc),
342 0,
343 (SLAB_HWCACHE_ALIGN |
344 SLAB_RECLAIM_ACCOUNT |
345 SLAB_ACCOUNT),
346 init_once);
347 BUG_ON(sock_inode_cachep == NULL);
348}
349
350static const struct super_operations sockfs_ops = {
351 .alloc_inode = sock_alloc_inode,
352 .free_inode = sock_free_inode,
353 .statfs = simple_statfs,
354};
355
356/*
357 * sockfs_dname() is called from d_path().
358 */
359static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
360{
361 return dynamic_dname(buffer, buflen, "socket:[%lu]",
362 d_inode(dentry)->i_ino);
363}
364
365static const struct dentry_operations sockfs_dentry_operations = {
366 .d_dname = sockfs_dname,
367};
368
369static int sockfs_xattr_get(const struct xattr_handler *handler,
370 struct dentry *dentry, struct inode *inode,
371 const char *suffix, void *value, size_t size)
372{
373 if (value) {
374 if (dentry->d_name.len + 1 > size)
375 return -ERANGE;
376 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
377 }
378 return dentry->d_name.len + 1;
379}
380
381#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
382#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
383#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
384
385static const struct xattr_handler sockfs_xattr_handler = {
386 .name = XATTR_NAME_SOCKPROTONAME,
387 .get = sockfs_xattr_get,
388};
389
390static int sockfs_security_xattr_set(const struct xattr_handler *handler,
391 struct mnt_idmap *idmap,
392 struct dentry *dentry, struct inode *inode,
393 const char *suffix, const void *value,
394 size_t size, int flags)
395{
396 /* Handled by LSM. */
397 return -EAGAIN;
398}
399
400static const struct xattr_handler sockfs_security_xattr_handler = {
401 .prefix = XATTR_SECURITY_PREFIX,
402 .set = sockfs_security_xattr_set,
403};
404
405static const struct xattr_handler * const sockfs_xattr_handlers[] = {
406 &sockfs_xattr_handler,
407 &sockfs_security_xattr_handler,
408 NULL
409};
410
411static int sockfs_init_fs_context(struct fs_context *fc)
412{
413 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
414 if (!ctx)
415 return -ENOMEM;
416 ctx->ops = &sockfs_ops;
417 ctx->dops = &sockfs_dentry_operations;
418 ctx->xattr = sockfs_xattr_handlers;
419 return 0;
420}
421
422static struct vfsmount *sock_mnt __read_mostly;
423
424static struct file_system_type sock_fs_type = {
425 .name = "sockfs",
426 .init_fs_context = sockfs_init_fs_context,
427 .kill_sb = kill_anon_super,
428};
429
430/*
431 * Obtains the first available file descriptor and sets it up for use.
432 *
433 * These functions create file structures and maps them to fd space
434 * of the current process. On success it returns file descriptor
435 * and file struct implicitly stored in sock->file.
436 * Note that another thread may close file descriptor before we return
437 * from this function. We use the fact that now we do not refer
438 * to socket after mapping. If one day we will need it, this
439 * function will increment ref. count on file by 1.
440 *
441 * In any case returned fd MAY BE not valid!
442 * This race condition is unavoidable
443 * with shared fd spaces, we cannot solve it inside kernel,
444 * but we take care of internal coherence yet.
445 */
446
447/**
448 * sock_alloc_file - Bind a &socket to a &file
449 * @sock: socket
450 * @flags: file status flags
451 * @dname: protocol name
452 *
453 * Returns the &file bound with @sock, implicitly storing it
454 * in sock->file. If dname is %NULL, sets to "".
455 *
456 * On failure @sock is released, and an ERR pointer is returned.
457 *
458 * This function uses GFP_KERNEL internally.
459 */
460
461struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
462{
463 struct file *file;
464
465 if (!dname)
466 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
467
468 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
469 O_RDWR | (flags & O_NONBLOCK),
470 &socket_file_ops);
471 if (IS_ERR(file)) {
472 sock_release(sock);
473 return file;
474 }
475
476 file->f_mode |= FMODE_NOWAIT;
477 sock->file = file;
478 file->private_data = sock;
479 stream_open(SOCK_INODE(sock), file);
480 return file;
481}
482EXPORT_SYMBOL(sock_alloc_file);
483
484static int sock_map_fd(struct socket *sock, int flags)
485{
486 struct file *newfile;
487 int fd = get_unused_fd_flags(flags);
488 if (unlikely(fd < 0)) {
489 sock_release(sock);
490 return fd;
491 }
492
493 newfile = sock_alloc_file(sock, flags, NULL);
494 if (!IS_ERR(newfile)) {
495 fd_install(fd, newfile);
496 return fd;
497 }
498
499 put_unused_fd(fd);
500 return PTR_ERR(newfile);
501}
502
503/**
504 * sock_from_file - Return the &socket bounded to @file.
505 * @file: file
506 *
507 * On failure returns %NULL.
508 */
509
510struct socket *sock_from_file(struct file *file)
511{
512 if (likely(file->f_op == &socket_file_ops))
513 return file->private_data; /* set in sock_alloc_file */
514
515 return NULL;
516}
517EXPORT_SYMBOL(sock_from_file);
518
519/**
520 * sockfd_lookup - Go from a file number to its socket slot
521 * @fd: file handle
522 * @err: pointer to an error code return
523 *
524 * The file handle passed in is locked and the socket it is bound
525 * to is returned. If an error occurs the err pointer is overwritten
526 * with a negative errno code and NULL is returned. The function checks
527 * for both invalid handles and passing a handle which is not a socket.
528 *
529 * On a success the socket object pointer is returned.
530 */
531
532struct socket *sockfd_lookup(int fd, int *err)
533{
534 struct file *file;
535 struct socket *sock;
536
537 file = fget(fd);
538 if (!file) {
539 *err = -EBADF;
540 return NULL;
541 }
542
543 sock = sock_from_file(file);
544 if (!sock) {
545 *err = -ENOTSOCK;
546 fput(file);
547 }
548 return sock;
549}
550EXPORT_SYMBOL(sockfd_lookup);
551
552static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
553 size_t size)
554{
555 ssize_t len;
556 ssize_t used = 0;
557
558 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
559 if (len < 0)
560 return len;
561 used += len;
562 if (buffer) {
563 if (size < used)
564 return -ERANGE;
565 buffer += len;
566 }
567
568 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
569 used += len;
570 if (buffer) {
571 if (size < used)
572 return -ERANGE;
573 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
574 buffer += len;
575 }
576
577 return used;
578}
579
580static int sockfs_setattr(struct mnt_idmap *idmap,
581 struct dentry *dentry, struct iattr *iattr)
582{
583 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
584
585 if (!err && (iattr->ia_valid & ATTR_UID)) {
586 struct socket *sock = SOCKET_I(d_inode(dentry));
587
588 if (sock->sk)
589 sock->sk->sk_uid = iattr->ia_uid;
590 else
591 err = -ENOENT;
592 }
593
594 return err;
595}
596
597static const struct inode_operations sockfs_inode_ops = {
598 .listxattr = sockfs_listxattr,
599 .setattr = sockfs_setattr,
600};
601
602/**
603 * sock_alloc - allocate a socket
604 *
605 * Allocate a new inode and socket object. The two are bound together
606 * and initialised. The socket is then returned. If we are out of inodes
607 * NULL is returned. This functions uses GFP_KERNEL internally.
608 */
609
610struct socket *sock_alloc(void)
611{
612 struct inode *inode;
613 struct socket *sock;
614
615 inode = new_inode_pseudo(sock_mnt->mnt_sb);
616 if (!inode)
617 return NULL;
618
619 sock = SOCKET_I(inode);
620
621 inode->i_ino = get_next_ino();
622 inode->i_mode = S_IFSOCK | S_IRWXUGO;
623 inode->i_uid = current_fsuid();
624 inode->i_gid = current_fsgid();
625 inode->i_op = &sockfs_inode_ops;
626
627 return sock;
628}
629EXPORT_SYMBOL(sock_alloc);
630
631static void __sock_release(struct socket *sock, struct inode *inode)
632{
633 const struct proto_ops *ops = READ_ONCE(sock->ops);
634
635 if (ops) {
636 struct module *owner = ops->owner;
637
638 if (inode)
639 inode_lock(inode);
640 ops->release(sock);
641 sock->sk = NULL;
642 if (inode)
643 inode_unlock(inode);
644 sock->ops = NULL;
645 module_put(owner);
646 }
647
648 if (sock->wq.fasync_list)
649 pr_err("%s: fasync list not empty!\n", __func__);
650
651 if (!sock->file) {
652 iput(SOCK_INODE(sock));
653 return;
654 }
655 sock->file = NULL;
656}
657
658/**
659 * sock_release - close a socket
660 * @sock: socket to close
661 *
662 * The socket is released from the protocol stack if it has a release
663 * callback, and the inode is then released if the socket is bound to
664 * an inode not a file.
665 */
666void sock_release(struct socket *sock)
667{
668 __sock_release(sock, NULL);
669}
670EXPORT_SYMBOL(sock_release);
671
672void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags)
673{
674 u8 flags = *tx_flags;
675
676 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) {
677 flags |= SKBTX_HW_TSTAMP;
678
679 /* PTP hardware clocks can provide a free running cycle counter
680 * as a time base for virtual clocks. Tell driver to use the
681 * free running cycle counter for timestamp if socket is bound
682 * to virtual clock.
683 */
684 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
685 flags |= SKBTX_HW_TSTAMP_USE_CYCLES;
686 }
687
688 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
689 flags |= SKBTX_SW_TSTAMP;
690
691 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
692 flags |= SKBTX_SCHED_TSTAMP;
693
694 *tx_flags = flags;
695}
696EXPORT_SYMBOL(__sock_tx_timestamp);
697
698INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
699 size_t));
700INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
701 size_t));
702
703static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
704 int flags)
705{
706 trace_sock_send_length(sk, ret, 0);
707}
708
709static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
710{
711 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
712 inet_sendmsg, sock, msg,
713 msg_data_left(msg));
714 BUG_ON(ret == -EIOCBQUEUED);
715
716 if (trace_sock_send_length_enabled())
717 call_trace_sock_send_length(sock->sk, ret, 0);
718 return ret;
719}
720
721static int __sock_sendmsg(struct socket *sock, struct msghdr *msg)
722{
723 int err = security_socket_sendmsg(sock, msg,
724 msg_data_left(msg));
725
726 return err ?: sock_sendmsg_nosec(sock, msg);
727}
728
729/**
730 * sock_sendmsg - send a message through @sock
731 * @sock: socket
732 * @msg: message to send
733 *
734 * Sends @msg through @sock, passing through LSM.
735 * Returns the number of bytes sent, or an error code.
736 */
737int sock_sendmsg(struct socket *sock, struct msghdr *msg)
738{
739 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name;
740 struct sockaddr_storage address;
741 int save_len = msg->msg_namelen;
742 int ret;
743
744 if (msg->msg_name) {
745 memcpy(&address, msg->msg_name, msg->msg_namelen);
746 msg->msg_name = &address;
747 }
748
749 ret = __sock_sendmsg(sock, msg);
750 msg->msg_name = save_addr;
751 msg->msg_namelen = save_len;
752
753 return ret;
754}
755EXPORT_SYMBOL(sock_sendmsg);
756
757/**
758 * kernel_sendmsg - send a message through @sock (kernel-space)
759 * @sock: socket
760 * @msg: message header
761 * @vec: kernel vec
762 * @num: vec array length
763 * @size: total message data size
764 *
765 * Builds the message data with @vec and sends it through @sock.
766 * Returns the number of bytes sent, or an error code.
767 */
768
769int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
770 struct kvec *vec, size_t num, size_t size)
771{
772 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
773 return sock_sendmsg(sock, msg);
774}
775EXPORT_SYMBOL(kernel_sendmsg);
776
777/**
778 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
779 * @sk: sock
780 * @msg: message header
781 * @vec: output s/g array
782 * @num: output s/g array length
783 * @size: total message data size
784 *
785 * Builds the message data with @vec and sends it through @sock.
786 * Returns the number of bytes sent, or an error code.
787 * Caller must hold @sk.
788 */
789
790int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
791 struct kvec *vec, size_t num, size_t size)
792{
793 struct socket *sock = sk->sk_socket;
794 const struct proto_ops *ops = READ_ONCE(sock->ops);
795
796 if (!ops->sendmsg_locked)
797 return sock_no_sendmsg_locked(sk, msg, size);
798
799 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
800
801 return ops->sendmsg_locked(sk, msg, msg_data_left(msg));
802}
803EXPORT_SYMBOL(kernel_sendmsg_locked);
804
805static bool skb_is_err_queue(const struct sk_buff *skb)
806{
807 /* pkt_type of skbs enqueued on the error queue are set to
808 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
809 * in recvmsg, since skbs received on a local socket will never
810 * have a pkt_type of PACKET_OUTGOING.
811 */
812 return skb->pkt_type == PACKET_OUTGOING;
813}
814
815/* On transmit, software and hardware timestamps are returned independently.
816 * As the two skb clones share the hardware timestamp, which may be updated
817 * before the software timestamp is received, a hardware TX timestamp may be
818 * returned only if there is no software TX timestamp. Ignore false software
819 * timestamps, which may be made in the __sock_recv_timestamp() call when the
820 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
821 * hardware timestamp.
822 */
823static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
824{
825 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
826}
827
828static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
829{
830 bool cycles = READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_BIND_PHC;
831 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
832 struct net_device *orig_dev;
833 ktime_t hwtstamp;
834
835 rcu_read_lock();
836 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
837 if (orig_dev) {
838 *if_index = orig_dev->ifindex;
839 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
840 } else {
841 hwtstamp = shhwtstamps->hwtstamp;
842 }
843 rcu_read_unlock();
844
845 return hwtstamp;
846}
847
848static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
849 int if_index)
850{
851 struct scm_ts_pktinfo ts_pktinfo;
852 struct net_device *orig_dev;
853
854 if (!skb_mac_header_was_set(skb))
855 return;
856
857 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
858
859 if (!if_index) {
860 rcu_read_lock();
861 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
862 if (orig_dev)
863 if_index = orig_dev->ifindex;
864 rcu_read_unlock();
865 }
866 ts_pktinfo.if_index = if_index;
867
868 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
869 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
870 sizeof(ts_pktinfo), &ts_pktinfo);
871}
872
873/*
874 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
875 */
876void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
877 struct sk_buff *skb)
878{
879 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
880 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
881 struct scm_timestamping_internal tss;
882 int empty = 1, false_tstamp = 0;
883 struct skb_shared_hwtstamps *shhwtstamps =
884 skb_hwtstamps(skb);
885 int if_index;
886 ktime_t hwtstamp;
887 u32 tsflags;
888
889 /* Race occurred between timestamp enabling and packet
890 receiving. Fill in the current time for now. */
891 if (need_software_tstamp && skb->tstamp == 0) {
892 __net_timestamp(skb);
893 false_tstamp = 1;
894 }
895
896 if (need_software_tstamp) {
897 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
898 if (new_tstamp) {
899 struct __kernel_sock_timeval tv;
900
901 skb_get_new_timestamp(skb, &tv);
902 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
903 sizeof(tv), &tv);
904 } else {
905 struct __kernel_old_timeval tv;
906
907 skb_get_timestamp(skb, &tv);
908 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
909 sizeof(tv), &tv);
910 }
911 } else {
912 if (new_tstamp) {
913 struct __kernel_timespec ts;
914
915 skb_get_new_timestampns(skb, &ts);
916 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
917 sizeof(ts), &ts);
918 } else {
919 struct __kernel_old_timespec ts;
920
921 skb_get_timestampns(skb, &ts);
922 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
923 sizeof(ts), &ts);
924 }
925 }
926 }
927
928 memset(&tss, 0, sizeof(tss));
929 tsflags = READ_ONCE(sk->sk_tsflags);
930 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE &&
931 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE ||
932 skb_is_err_queue(skb) ||
933 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
934 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
935 empty = 0;
936 if (shhwtstamps &&
937 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE &&
938 (tsflags & SOF_TIMESTAMPING_RX_HARDWARE ||
939 skb_is_err_queue(skb) ||
940 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
941 !skb_is_swtx_tstamp(skb, false_tstamp)) {
942 if_index = 0;
943 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
944 hwtstamp = get_timestamp(sk, skb, &if_index);
945 else
946 hwtstamp = shhwtstamps->hwtstamp;
947
948 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
949 hwtstamp = ptp_convert_timestamp(&hwtstamp,
950 READ_ONCE(sk->sk_bind_phc));
951
952 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
953 empty = 0;
954
955 if ((tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
956 !skb_is_err_queue(skb))
957 put_ts_pktinfo(msg, skb, if_index);
958 }
959 }
960 if (!empty) {
961 if (sock_flag(sk, SOCK_TSTAMP_NEW))
962 put_cmsg_scm_timestamping64(msg, &tss);
963 else
964 put_cmsg_scm_timestamping(msg, &tss);
965
966 if (skb_is_err_queue(skb) && skb->len &&
967 SKB_EXT_ERR(skb)->opt_stats)
968 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
969 skb->len, skb->data);
970 }
971}
972EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
973
974#ifdef CONFIG_WIRELESS
975void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
976 struct sk_buff *skb)
977{
978 int ack;
979
980 if (!sock_flag(sk, SOCK_WIFI_STATUS))
981 return;
982 if (!skb->wifi_acked_valid)
983 return;
984
985 ack = skb->wifi_acked;
986
987 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
988}
989EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
990#endif
991
992static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
993 struct sk_buff *skb)
994{
995 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
996 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
997 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
998}
999
1000static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
1001 struct sk_buff *skb)
1002{
1003 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
1004 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
1005 __u32 mark = skb->mark;
1006
1007 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1008 }
1009}
1010
1011void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1012 struct sk_buff *skb)
1013{
1014 sock_recv_timestamp(msg, sk, skb);
1015 sock_recv_drops(msg, sk, skb);
1016 sock_recv_mark(msg, sk, skb);
1017}
1018EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1019
1020INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1021 size_t, int));
1022INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1023 size_t, int));
1024
1025static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1026{
1027 trace_sock_recv_length(sk, ret, flags);
1028}
1029
1030static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1031 int flags)
1032{
1033 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1034 inet6_recvmsg,
1035 inet_recvmsg, sock, msg,
1036 msg_data_left(msg), flags);
1037 if (trace_sock_recv_length_enabled())
1038 call_trace_sock_recv_length(sock->sk, ret, flags);
1039 return ret;
1040}
1041
1042/**
1043 * sock_recvmsg - receive a message from @sock
1044 * @sock: socket
1045 * @msg: message to receive
1046 * @flags: message flags
1047 *
1048 * Receives @msg from @sock, passing through LSM. Returns the total number
1049 * of bytes received, or an error.
1050 */
1051int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1052{
1053 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1054
1055 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1056}
1057EXPORT_SYMBOL(sock_recvmsg);
1058
1059/**
1060 * kernel_recvmsg - Receive a message from a socket (kernel space)
1061 * @sock: The socket to receive the message from
1062 * @msg: Received message
1063 * @vec: Input s/g array for message data
1064 * @num: Size of input s/g array
1065 * @size: Number of bytes to read
1066 * @flags: Message flags (MSG_DONTWAIT, etc...)
1067 *
1068 * On return the msg structure contains the scatter/gather array passed in the
1069 * vec argument. The array is modified so that it consists of the unfilled
1070 * portion of the original array.
1071 *
1072 * The returned value is the total number of bytes received, or an error.
1073 */
1074
1075int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1076 struct kvec *vec, size_t num, size_t size, int flags)
1077{
1078 msg->msg_control_is_user = false;
1079 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1080 return sock_recvmsg(sock, msg, flags);
1081}
1082EXPORT_SYMBOL(kernel_recvmsg);
1083
1084static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1085 struct pipe_inode_info *pipe, size_t len,
1086 unsigned int flags)
1087{
1088 struct socket *sock = file->private_data;
1089 const struct proto_ops *ops;
1090
1091 ops = READ_ONCE(sock->ops);
1092 if (unlikely(!ops->splice_read))
1093 return copy_splice_read(file, ppos, pipe, len, flags);
1094
1095 return ops->splice_read(sock, ppos, pipe, len, flags);
1096}
1097
1098static void sock_splice_eof(struct file *file)
1099{
1100 struct socket *sock = file->private_data;
1101 const struct proto_ops *ops;
1102
1103 ops = READ_ONCE(sock->ops);
1104 if (ops->splice_eof)
1105 ops->splice_eof(sock);
1106}
1107
1108static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1109{
1110 struct file *file = iocb->ki_filp;
1111 struct socket *sock = file->private_data;
1112 struct msghdr msg = {.msg_iter = *to,
1113 .msg_iocb = iocb};
1114 ssize_t res;
1115
1116 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1117 msg.msg_flags = MSG_DONTWAIT;
1118
1119 if (iocb->ki_pos != 0)
1120 return -ESPIPE;
1121
1122 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1123 return 0;
1124
1125 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1126 *to = msg.msg_iter;
1127 return res;
1128}
1129
1130static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1131{
1132 struct file *file = iocb->ki_filp;
1133 struct socket *sock = file->private_data;
1134 struct msghdr msg = {.msg_iter = *from,
1135 .msg_iocb = iocb};
1136 ssize_t res;
1137
1138 if (iocb->ki_pos != 0)
1139 return -ESPIPE;
1140
1141 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1142 msg.msg_flags = MSG_DONTWAIT;
1143
1144 if (sock->type == SOCK_SEQPACKET)
1145 msg.msg_flags |= MSG_EOR;
1146
1147 res = __sock_sendmsg(sock, &msg);
1148 *from = msg.msg_iter;
1149 return res;
1150}
1151
1152/*
1153 * Atomic setting of ioctl hooks to avoid race
1154 * with module unload.
1155 */
1156
1157static DEFINE_MUTEX(br_ioctl_mutex);
1158static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1159 unsigned int cmd, struct ifreq *ifr,
1160 void __user *uarg);
1161
1162void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1163 unsigned int cmd, struct ifreq *ifr,
1164 void __user *uarg))
1165{
1166 mutex_lock(&br_ioctl_mutex);
1167 br_ioctl_hook = hook;
1168 mutex_unlock(&br_ioctl_mutex);
1169}
1170EXPORT_SYMBOL(brioctl_set);
1171
1172int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1173 struct ifreq *ifr, void __user *uarg)
1174{
1175 int err = -ENOPKG;
1176
1177 if (!br_ioctl_hook)
1178 request_module("bridge");
1179
1180 mutex_lock(&br_ioctl_mutex);
1181 if (br_ioctl_hook)
1182 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1183 mutex_unlock(&br_ioctl_mutex);
1184
1185 return err;
1186}
1187
1188static DEFINE_MUTEX(vlan_ioctl_mutex);
1189static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1190
1191void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1192{
1193 mutex_lock(&vlan_ioctl_mutex);
1194 vlan_ioctl_hook = hook;
1195 mutex_unlock(&vlan_ioctl_mutex);
1196}
1197EXPORT_SYMBOL(vlan_ioctl_set);
1198
1199static long sock_do_ioctl(struct net *net, struct socket *sock,
1200 unsigned int cmd, unsigned long arg)
1201{
1202 const struct proto_ops *ops = READ_ONCE(sock->ops);
1203 struct ifreq ifr;
1204 bool need_copyout;
1205 int err;
1206 void __user *argp = (void __user *)arg;
1207 void __user *data;
1208
1209 err = ops->ioctl(sock, cmd, arg);
1210
1211 /*
1212 * If this ioctl is unknown try to hand it down
1213 * to the NIC driver.
1214 */
1215 if (err != -ENOIOCTLCMD)
1216 return err;
1217
1218 if (!is_socket_ioctl_cmd(cmd))
1219 return -ENOTTY;
1220
1221 if (get_user_ifreq(&ifr, &data, argp))
1222 return -EFAULT;
1223 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1224 if (!err && need_copyout)
1225 if (put_user_ifreq(&ifr, argp))
1226 return -EFAULT;
1227
1228 return err;
1229}
1230
1231/*
1232 * With an ioctl, arg may well be a user mode pointer, but we don't know
1233 * what to do with it - that's up to the protocol still.
1234 */
1235
1236static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1237{
1238 const struct proto_ops *ops;
1239 struct socket *sock;
1240 struct sock *sk;
1241 void __user *argp = (void __user *)arg;
1242 int pid, err;
1243 struct net *net;
1244
1245 sock = file->private_data;
1246 ops = READ_ONCE(sock->ops);
1247 sk = sock->sk;
1248 net = sock_net(sk);
1249 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1250 struct ifreq ifr;
1251 void __user *data;
1252 bool need_copyout;
1253 if (get_user_ifreq(&ifr, &data, argp))
1254 return -EFAULT;
1255 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1256 if (!err && need_copyout)
1257 if (put_user_ifreq(&ifr, argp))
1258 return -EFAULT;
1259 } else
1260#ifdef CONFIG_WEXT_CORE
1261 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1262 err = wext_handle_ioctl(net, cmd, argp);
1263 } else
1264#endif
1265 switch (cmd) {
1266 case FIOSETOWN:
1267 case SIOCSPGRP:
1268 err = -EFAULT;
1269 if (get_user(pid, (int __user *)argp))
1270 break;
1271 err = f_setown(sock->file, pid, 1);
1272 break;
1273 case FIOGETOWN:
1274 case SIOCGPGRP:
1275 err = put_user(f_getown(sock->file),
1276 (int __user *)argp);
1277 break;
1278 case SIOCGIFBR:
1279 case SIOCSIFBR:
1280 case SIOCBRADDBR:
1281 case SIOCBRDELBR:
1282 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1283 break;
1284 case SIOCGIFVLAN:
1285 case SIOCSIFVLAN:
1286 err = -ENOPKG;
1287 if (!vlan_ioctl_hook)
1288 request_module("8021q");
1289
1290 mutex_lock(&vlan_ioctl_mutex);
1291 if (vlan_ioctl_hook)
1292 err = vlan_ioctl_hook(net, argp);
1293 mutex_unlock(&vlan_ioctl_mutex);
1294 break;
1295 case SIOCGSKNS:
1296 err = -EPERM;
1297 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1298 break;
1299
1300 err = open_related_ns(&net->ns, get_net_ns);
1301 break;
1302 case SIOCGSTAMP_OLD:
1303 case SIOCGSTAMPNS_OLD:
1304 if (!ops->gettstamp) {
1305 err = -ENOIOCTLCMD;
1306 break;
1307 }
1308 err = ops->gettstamp(sock, argp,
1309 cmd == SIOCGSTAMP_OLD,
1310 !IS_ENABLED(CONFIG_64BIT));
1311 break;
1312 case SIOCGSTAMP_NEW:
1313 case SIOCGSTAMPNS_NEW:
1314 if (!ops->gettstamp) {
1315 err = -ENOIOCTLCMD;
1316 break;
1317 }
1318 err = ops->gettstamp(sock, argp,
1319 cmd == SIOCGSTAMP_NEW,
1320 false);
1321 break;
1322
1323 case SIOCGIFCONF:
1324 err = dev_ifconf(net, argp);
1325 break;
1326
1327 default:
1328 err = sock_do_ioctl(net, sock, cmd, arg);
1329 break;
1330 }
1331 return err;
1332}
1333
1334/**
1335 * sock_create_lite - creates a socket
1336 * @family: protocol family (AF_INET, ...)
1337 * @type: communication type (SOCK_STREAM, ...)
1338 * @protocol: protocol (0, ...)
1339 * @res: new socket
1340 *
1341 * Creates a new socket and assigns it to @res, passing through LSM.
1342 * The new socket initialization is not complete, see kernel_accept().
1343 * Returns 0 or an error. On failure @res is set to %NULL.
1344 * This function internally uses GFP_KERNEL.
1345 */
1346
1347int sock_create_lite(int family, int type, int protocol, struct socket **res)
1348{
1349 int err;
1350 struct socket *sock = NULL;
1351
1352 err = security_socket_create(family, type, protocol, 1);
1353 if (err)
1354 goto out;
1355
1356 sock = sock_alloc();
1357 if (!sock) {
1358 err = -ENOMEM;
1359 goto out;
1360 }
1361
1362 sock->type = type;
1363 err = security_socket_post_create(sock, family, type, protocol, 1);
1364 if (err)
1365 goto out_release;
1366
1367out:
1368 *res = sock;
1369 return err;
1370out_release:
1371 sock_release(sock);
1372 sock = NULL;
1373 goto out;
1374}
1375EXPORT_SYMBOL(sock_create_lite);
1376
1377/* No kernel lock held - perfect */
1378static __poll_t sock_poll(struct file *file, poll_table *wait)
1379{
1380 struct socket *sock = file->private_data;
1381 const struct proto_ops *ops = READ_ONCE(sock->ops);
1382 __poll_t events = poll_requested_events(wait), flag = 0;
1383
1384 if (!ops->poll)
1385 return 0;
1386
1387 if (sk_can_busy_loop(sock->sk)) {
1388 /* poll once if requested by the syscall */
1389 if (events & POLL_BUSY_LOOP)
1390 sk_busy_loop(sock->sk, 1);
1391
1392 /* if this socket can poll_ll, tell the system call */
1393 flag = POLL_BUSY_LOOP;
1394 }
1395
1396 return ops->poll(file, sock, wait) | flag;
1397}
1398
1399static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1400{
1401 struct socket *sock = file->private_data;
1402
1403 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1404}
1405
1406static int sock_close(struct inode *inode, struct file *filp)
1407{
1408 __sock_release(SOCKET_I(inode), inode);
1409 return 0;
1410}
1411
1412/*
1413 * Update the socket async list
1414 *
1415 * Fasync_list locking strategy.
1416 *
1417 * 1. fasync_list is modified only under process context socket lock
1418 * i.e. under semaphore.
1419 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1420 * or under socket lock
1421 */
1422
1423static int sock_fasync(int fd, struct file *filp, int on)
1424{
1425 struct socket *sock = filp->private_data;
1426 struct sock *sk = sock->sk;
1427 struct socket_wq *wq = &sock->wq;
1428
1429 if (sk == NULL)
1430 return -EINVAL;
1431
1432 lock_sock(sk);
1433 fasync_helper(fd, filp, on, &wq->fasync_list);
1434
1435 if (!wq->fasync_list)
1436 sock_reset_flag(sk, SOCK_FASYNC);
1437 else
1438 sock_set_flag(sk, SOCK_FASYNC);
1439
1440 release_sock(sk);
1441 return 0;
1442}
1443
1444/* This function may be called only under rcu_lock */
1445
1446int sock_wake_async(struct socket_wq *wq, int how, int band)
1447{
1448 if (!wq || !wq->fasync_list)
1449 return -1;
1450
1451 switch (how) {
1452 case SOCK_WAKE_WAITD:
1453 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1454 break;
1455 goto call_kill;
1456 case SOCK_WAKE_SPACE:
1457 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1458 break;
1459 fallthrough;
1460 case SOCK_WAKE_IO:
1461call_kill:
1462 kill_fasync(&wq->fasync_list, SIGIO, band);
1463 break;
1464 case SOCK_WAKE_URG:
1465 kill_fasync(&wq->fasync_list, SIGURG, band);
1466 }
1467
1468 return 0;
1469}
1470EXPORT_SYMBOL(sock_wake_async);
1471
1472/**
1473 * __sock_create - creates a socket
1474 * @net: net namespace
1475 * @family: protocol family (AF_INET, ...)
1476 * @type: communication type (SOCK_STREAM, ...)
1477 * @protocol: protocol (0, ...)
1478 * @res: new socket
1479 * @kern: boolean for kernel space sockets
1480 *
1481 * Creates a new socket and assigns it to @res, passing through LSM.
1482 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1483 * be set to true if the socket resides in kernel space.
1484 * This function internally uses GFP_KERNEL.
1485 */
1486
1487int __sock_create(struct net *net, int family, int type, int protocol,
1488 struct socket **res, int kern)
1489{
1490 int err;
1491 struct socket *sock;
1492 const struct net_proto_family *pf;
1493
1494 /*
1495 * Check protocol is in range
1496 */
1497 if (family < 0 || family >= NPROTO)
1498 return -EAFNOSUPPORT;
1499 if (type < 0 || type >= SOCK_MAX)
1500 return -EINVAL;
1501
1502 /* Compatibility.
1503
1504 This uglymoron is moved from INET layer to here to avoid
1505 deadlock in module load.
1506 */
1507 if (family == PF_INET && type == SOCK_PACKET) {
1508 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1509 current->comm);
1510 family = PF_PACKET;
1511 }
1512
1513 err = security_socket_create(family, type, protocol, kern);
1514 if (err)
1515 return err;
1516
1517 /*
1518 * Allocate the socket and allow the family to set things up. if
1519 * the protocol is 0, the family is instructed to select an appropriate
1520 * default.
1521 */
1522 sock = sock_alloc();
1523 if (!sock) {
1524 net_warn_ratelimited("socket: no more sockets\n");
1525 return -ENFILE; /* Not exactly a match, but its the
1526 closest posix thing */
1527 }
1528
1529 sock->type = type;
1530
1531#ifdef CONFIG_MODULES
1532 /* Attempt to load a protocol module if the find failed.
1533 *
1534 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1535 * requested real, full-featured networking support upon configuration.
1536 * Otherwise module support will break!
1537 */
1538 if (rcu_access_pointer(net_families[family]) == NULL)
1539 request_module("net-pf-%d", family);
1540#endif
1541
1542 rcu_read_lock();
1543 pf = rcu_dereference(net_families[family]);
1544 err = -EAFNOSUPPORT;
1545 if (!pf)
1546 goto out_release;
1547
1548 /*
1549 * We will call the ->create function, that possibly is in a loadable
1550 * module, so we have to bump that loadable module refcnt first.
1551 */
1552 if (!try_module_get(pf->owner))
1553 goto out_release;
1554
1555 /* Now protected by module ref count */
1556 rcu_read_unlock();
1557
1558 err = pf->create(net, sock, protocol, kern);
1559 if (err < 0) {
1560 /* ->create should release the allocated sock->sk object on error
1561 * and make sure sock->sk is set to NULL to avoid use-after-free
1562 */
1563 DEBUG_NET_WARN_ONCE(sock->sk,
1564 "%ps must clear sock->sk on failure, family: %d, type: %d, protocol: %d\n",
1565 pf->create, family, type, protocol);
1566 goto out_module_put;
1567 }
1568
1569 /*
1570 * Now to bump the refcnt of the [loadable] module that owns this
1571 * socket at sock_release time we decrement its refcnt.
1572 */
1573 if (!try_module_get(sock->ops->owner))
1574 goto out_module_busy;
1575
1576 /*
1577 * Now that we're done with the ->create function, the [loadable]
1578 * module can have its refcnt decremented
1579 */
1580 module_put(pf->owner);
1581 err = security_socket_post_create(sock, family, type, protocol, kern);
1582 if (err)
1583 goto out_sock_release;
1584 *res = sock;
1585
1586 return 0;
1587
1588out_module_busy:
1589 err = -EAFNOSUPPORT;
1590out_module_put:
1591 sock->ops = NULL;
1592 module_put(pf->owner);
1593out_sock_release:
1594 sock_release(sock);
1595 return err;
1596
1597out_release:
1598 rcu_read_unlock();
1599 goto out_sock_release;
1600}
1601EXPORT_SYMBOL(__sock_create);
1602
1603/**
1604 * sock_create - creates a socket
1605 * @family: protocol family (AF_INET, ...)
1606 * @type: communication type (SOCK_STREAM, ...)
1607 * @protocol: protocol (0, ...)
1608 * @res: new socket
1609 *
1610 * A wrapper around __sock_create().
1611 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1612 */
1613
1614int sock_create(int family, int type, int protocol, struct socket **res)
1615{
1616 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1617}
1618EXPORT_SYMBOL(sock_create);
1619
1620/**
1621 * sock_create_kern - creates a socket (kernel space)
1622 * @net: net namespace
1623 * @family: protocol family (AF_INET, ...)
1624 * @type: communication type (SOCK_STREAM, ...)
1625 * @protocol: protocol (0, ...)
1626 * @res: new socket
1627 *
1628 * A wrapper around __sock_create().
1629 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1630 */
1631
1632int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1633{
1634 return __sock_create(net, family, type, protocol, res, 1);
1635}
1636EXPORT_SYMBOL(sock_create_kern);
1637
1638static struct socket *__sys_socket_create(int family, int type, int protocol)
1639{
1640 struct socket *sock;
1641 int retval;
1642
1643 /* Check the SOCK_* constants for consistency. */
1644 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1645 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1646 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1647 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1648
1649 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1650 return ERR_PTR(-EINVAL);
1651 type &= SOCK_TYPE_MASK;
1652
1653 retval = sock_create(family, type, protocol, &sock);
1654 if (retval < 0)
1655 return ERR_PTR(retval);
1656
1657 return sock;
1658}
1659
1660struct file *__sys_socket_file(int family, int type, int protocol)
1661{
1662 struct socket *sock;
1663 int flags;
1664
1665 sock = __sys_socket_create(family, type, protocol);
1666 if (IS_ERR(sock))
1667 return ERR_CAST(sock);
1668
1669 flags = type & ~SOCK_TYPE_MASK;
1670 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1671 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1672
1673 return sock_alloc_file(sock, flags, NULL);
1674}
1675
1676/* A hook for bpf progs to attach to and update socket protocol.
1677 *
1678 * A static noinline declaration here could cause the compiler to
1679 * optimize away the function. A global noinline declaration will
1680 * keep the definition, but may optimize away the callsite.
1681 * Therefore, __weak is needed to ensure that the call is still
1682 * emitted, by telling the compiler that we don't know what the
1683 * function might eventually be.
1684 */
1685
1686__bpf_hook_start();
1687
1688__weak noinline int update_socket_protocol(int family, int type, int protocol)
1689{
1690 return protocol;
1691}
1692
1693__bpf_hook_end();
1694
1695int __sys_socket(int family, int type, int protocol)
1696{
1697 struct socket *sock;
1698 int flags;
1699
1700 sock = __sys_socket_create(family, type,
1701 update_socket_protocol(family, type, protocol));
1702 if (IS_ERR(sock))
1703 return PTR_ERR(sock);
1704
1705 flags = type & ~SOCK_TYPE_MASK;
1706 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1707 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1708
1709 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1710}
1711
1712SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1713{
1714 return __sys_socket(family, type, protocol);
1715}
1716
1717/*
1718 * Create a pair of connected sockets.
1719 */
1720
1721int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1722{
1723 struct socket *sock1, *sock2;
1724 int fd1, fd2, err;
1725 struct file *newfile1, *newfile2;
1726 int flags;
1727
1728 flags = type & ~SOCK_TYPE_MASK;
1729 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1730 return -EINVAL;
1731 type &= SOCK_TYPE_MASK;
1732
1733 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1734 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1735
1736 /*
1737 * reserve descriptors and make sure we won't fail
1738 * to return them to userland.
1739 */
1740 fd1 = get_unused_fd_flags(flags);
1741 if (unlikely(fd1 < 0))
1742 return fd1;
1743
1744 fd2 = get_unused_fd_flags(flags);
1745 if (unlikely(fd2 < 0)) {
1746 put_unused_fd(fd1);
1747 return fd2;
1748 }
1749
1750 err = put_user(fd1, &usockvec[0]);
1751 if (err)
1752 goto out;
1753
1754 err = put_user(fd2, &usockvec[1]);
1755 if (err)
1756 goto out;
1757
1758 /*
1759 * Obtain the first socket and check if the underlying protocol
1760 * supports the socketpair call.
1761 */
1762
1763 err = sock_create(family, type, protocol, &sock1);
1764 if (unlikely(err < 0))
1765 goto out;
1766
1767 err = sock_create(family, type, protocol, &sock2);
1768 if (unlikely(err < 0)) {
1769 sock_release(sock1);
1770 goto out;
1771 }
1772
1773 err = security_socket_socketpair(sock1, sock2);
1774 if (unlikely(err)) {
1775 sock_release(sock2);
1776 sock_release(sock1);
1777 goto out;
1778 }
1779
1780 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1781 if (unlikely(err < 0)) {
1782 sock_release(sock2);
1783 sock_release(sock1);
1784 goto out;
1785 }
1786
1787 newfile1 = sock_alloc_file(sock1, flags, NULL);
1788 if (IS_ERR(newfile1)) {
1789 err = PTR_ERR(newfile1);
1790 sock_release(sock2);
1791 goto out;
1792 }
1793
1794 newfile2 = sock_alloc_file(sock2, flags, NULL);
1795 if (IS_ERR(newfile2)) {
1796 err = PTR_ERR(newfile2);
1797 fput(newfile1);
1798 goto out;
1799 }
1800
1801 audit_fd_pair(fd1, fd2);
1802
1803 fd_install(fd1, newfile1);
1804 fd_install(fd2, newfile2);
1805 return 0;
1806
1807out:
1808 put_unused_fd(fd2);
1809 put_unused_fd(fd1);
1810 return err;
1811}
1812
1813SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1814 int __user *, usockvec)
1815{
1816 return __sys_socketpair(family, type, protocol, usockvec);
1817}
1818
1819int __sys_bind_socket(struct socket *sock, struct sockaddr_storage *address,
1820 int addrlen)
1821{
1822 int err;
1823
1824 err = security_socket_bind(sock, (struct sockaddr *)address,
1825 addrlen);
1826 if (!err)
1827 err = READ_ONCE(sock->ops)->bind(sock,
1828 (struct sockaddr *)address,
1829 addrlen);
1830 return err;
1831}
1832
1833/*
1834 * Bind a name to a socket. Nothing much to do here since it's
1835 * the protocol's responsibility to handle the local address.
1836 *
1837 * We move the socket address to kernel space before we call
1838 * the protocol layer (having also checked the address is ok).
1839 */
1840
1841int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1842{
1843 struct socket *sock;
1844 struct sockaddr_storage address;
1845 CLASS(fd, f)(fd);
1846 int err;
1847
1848 if (fd_empty(f))
1849 return -EBADF;
1850 sock = sock_from_file(fd_file(f));
1851 if (unlikely(!sock))
1852 return -ENOTSOCK;
1853
1854 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1855 if (unlikely(err))
1856 return err;
1857
1858 return __sys_bind_socket(sock, &address, addrlen);
1859}
1860
1861SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1862{
1863 return __sys_bind(fd, umyaddr, addrlen);
1864}
1865
1866/*
1867 * Perform a listen. Basically, we allow the protocol to do anything
1868 * necessary for a listen, and if that works, we mark the socket as
1869 * ready for listening.
1870 */
1871int __sys_listen_socket(struct socket *sock, int backlog)
1872{
1873 int somaxconn, err;
1874
1875 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1876 if ((unsigned int)backlog > somaxconn)
1877 backlog = somaxconn;
1878
1879 err = security_socket_listen(sock, backlog);
1880 if (!err)
1881 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1882 return err;
1883}
1884
1885int __sys_listen(int fd, int backlog)
1886{
1887 CLASS(fd, f)(fd);
1888 struct socket *sock;
1889
1890 if (fd_empty(f))
1891 return -EBADF;
1892 sock = sock_from_file(fd_file(f));
1893 if (unlikely(!sock))
1894 return -ENOTSOCK;
1895
1896 return __sys_listen_socket(sock, backlog);
1897}
1898
1899SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1900{
1901 return __sys_listen(fd, backlog);
1902}
1903
1904struct file *do_accept(struct file *file, struct proto_accept_arg *arg,
1905 struct sockaddr __user *upeer_sockaddr,
1906 int __user *upeer_addrlen, int flags)
1907{
1908 struct socket *sock, *newsock;
1909 struct file *newfile;
1910 int err, len;
1911 struct sockaddr_storage address;
1912 const struct proto_ops *ops;
1913
1914 sock = sock_from_file(file);
1915 if (!sock)
1916 return ERR_PTR(-ENOTSOCK);
1917
1918 newsock = sock_alloc();
1919 if (!newsock)
1920 return ERR_PTR(-ENFILE);
1921 ops = READ_ONCE(sock->ops);
1922
1923 newsock->type = sock->type;
1924 newsock->ops = ops;
1925
1926 /*
1927 * We don't need try_module_get here, as the listening socket (sock)
1928 * has the protocol module (sock->ops->owner) held.
1929 */
1930 __module_get(ops->owner);
1931
1932 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1933 if (IS_ERR(newfile))
1934 return newfile;
1935
1936 err = security_socket_accept(sock, newsock);
1937 if (err)
1938 goto out_fd;
1939
1940 arg->flags |= sock->file->f_flags;
1941 err = ops->accept(sock, newsock, arg);
1942 if (err < 0)
1943 goto out_fd;
1944
1945 if (upeer_sockaddr) {
1946 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1947 if (len < 0) {
1948 err = -ECONNABORTED;
1949 goto out_fd;
1950 }
1951 err = move_addr_to_user(&address,
1952 len, upeer_sockaddr, upeer_addrlen);
1953 if (err < 0)
1954 goto out_fd;
1955 }
1956
1957 /* File flags are not inherited via accept() unlike another OSes. */
1958 return newfile;
1959out_fd:
1960 fput(newfile);
1961 return ERR_PTR(err);
1962}
1963
1964static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
1965 int __user *upeer_addrlen, int flags)
1966{
1967 struct proto_accept_arg arg = { };
1968 struct file *newfile;
1969 int newfd;
1970
1971 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1972 return -EINVAL;
1973
1974 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1975 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1976
1977 newfd = get_unused_fd_flags(flags);
1978 if (unlikely(newfd < 0))
1979 return newfd;
1980
1981 newfile = do_accept(file, &arg, upeer_sockaddr, upeer_addrlen,
1982 flags);
1983 if (IS_ERR(newfile)) {
1984 put_unused_fd(newfd);
1985 return PTR_ERR(newfile);
1986 }
1987 fd_install(newfd, newfile);
1988 return newfd;
1989}
1990
1991/*
1992 * For accept, we attempt to create a new socket, set up the link
1993 * with the client, wake up the client, then return the new
1994 * connected fd. We collect the address of the connector in kernel
1995 * space and move it to user at the very end. This is unclean because
1996 * we open the socket then return an error.
1997 *
1998 * 1003.1g adds the ability to recvmsg() to query connection pending
1999 * status to recvmsg. We need to add that support in a way thats
2000 * clean when we restructure accept also.
2001 */
2002
2003int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
2004 int __user *upeer_addrlen, int flags)
2005{
2006 CLASS(fd, f)(fd);
2007
2008 if (fd_empty(f))
2009 return -EBADF;
2010 return __sys_accept4_file(fd_file(f), upeer_sockaddr,
2011 upeer_addrlen, flags);
2012}
2013
2014SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
2015 int __user *, upeer_addrlen, int, flags)
2016{
2017 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
2018}
2019
2020SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
2021 int __user *, upeer_addrlen)
2022{
2023 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
2024}
2025
2026/*
2027 * Attempt to connect to a socket with the server address. The address
2028 * is in user space so we verify it is OK and move it to kernel space.
2029 *
2030 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
2031 * break bindings
2032 *
2033 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
2034 * other SEQPACKET protocols that take time to connect() as it doesn't
2035 * include the -EINPROGRESS status for such sockets.
2036 */
2037
2038int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
2039 int addrlen, int file_flags)
2040{
2041 struct socket *sock;
2042 int err;
2043
2044 sock = sock_from_file(file);
2045 if (!sock) {
2046 err = -ENOTSOCK;
2047 goto out;
2048 }
2049
2050 err =
2051 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2052 if (err)
2053 goto out;
2054
2055 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2056 addrlen, sock->file->f_flags | file_flags);
2057out:
2058 return err;
2059}
2060
2061int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2062{
2063 struct sockaddr_storage address;
2064 CLASS(fd, f)(fd);
2065 int ret;
2066
2067 if (fd_empty(f))
2068 return -EBADF;
2069
2070 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2071 if (ret)
2072 return ret;
2073
2074 return __sys_connect_file(fd_file(f), &address, addrlen, 0);
2075}
2076
2077SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2078 int, addrlen)
2079{
2080 return __sys_connect(fd, uservaddr, addrlen);
2081}
2082
2083/*
2084 * Get the local address ('name') of a socket object. Move the obtained
2085 * name to user space.
2086 */
2087
2088int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2089 int __user *usockaddr_len)
2090{
2091 struct socket *sock;
2092 struct sockaddr_storage address;
2093 CLASS(fd, f)(fd);
2094 int err;
2095
2096 if (fd_empty(f))
2097 return -EBADF;
2098 sock = sock_from_file(fd_file(f));
2099 if (unlikely(!sock))
2100 return -ENOTSOCK;
2101
2102 err = security_socket_getsockname(sock);
2103 if (err)
2104 return err;
2105
2106 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2107 if (err < 0)
2108 return err;
2109
2110 /* "err" is actually length in this case */
2111 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2112}
2113
2114SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2115 int __user *, usockaddr_len)
2116{
2117 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2118}
2119
2120/*
2121 * Get the remote address ('name') of a socket object. Move the obtained
2122 * name to user space.
2123 */
2124
2125int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2126 int __user *usockaddr_len)
2127{
2128 struct socket *sock;
2129 struct sockaddr_storage address;
2130 CLASS(fd, f)(fd);
2131 int err;
2132
2133 if (fd_empty(f))
2134 return -EBADF;
2135 sock = sock_from_file(fd_file(f));
2136 if (unlikely(!sock))
2137 return -ENOTSOCK;
2138
2139 err = security_socket_getpeername(sock);
2140 if (err)
2141 return err;
2142
2143 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 1);
2144 if (err < 0)
2145 return err;
2146
2147 /* "err" is actually length in this case */
2148 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2149}
2150
2151SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2152 int __user *, usockaddr_len)
2153{
2154 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2155}
2156
2157/*
2158 * Send a datagram to a given address. We move the address into kernel
2159 * space and check the user space data area is readable before invoking
2160 * the protocol.
2161 */
2162int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2163 struct sockaddr __user *addr, int addr_len)
2164{
2165 struct socket *sock;
2166 struct sockaddr_storage address;
2167 int err;
2168 struct msghdr msg;
2169
2170 err = import_ubuf(ITER_SOURCE, buff, len, &msg.msg_iter);
2171 if (unlikely(err))
2172 return err;
2173
2174 CLASS(fd, f)(fd);
2175 if (fd_empty(f))
2176 return -EBADF;
2177 sock = sock_from_file(fd_file(f));
2178 if (unlikely(!sock))
2179 return -ENOTSOCK;
2180
2181 msg.msg_name = NULL;
2182 msg.msg_control = NULL;
2183 msg.msg_controllen = 0;
2184 msg.msg_namelen = 0;
2185 msg.msg_ubuf = NULL;
2186 if (addr) {
2187 err = move_addr_to_kernel(addr, addr_len, &address);
2188 if (err < 0)
2189 return err;
2190 msg.msg_name = (struct sockaddr *)&address;
2191 msg.msg_namelen = addr_len;
2192 }
2193 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2194 if (sock->file->f_flags & O_NONBLOCK)
2195 flags |= MSG_DONTWAIT;
2196 msg.msg_flags = flags;
2197 return __sock_sendmsg(sock, &msg);
2198}
2199
2200SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2201 unsigned int, flags, struct sockaddr __user *, addr,
2202 int, addr_len)
2203{
2204 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2205}
2206
2207/*
2208 * Send a datagram down a socket.
2209 */
2210
2211SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2212 unsigned int, flags)
2213{
2214 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2215}
2216
2217/*
2218 * Receive a frame from the socket and optionally record the address of the
2219 * sender. We verify the buffers are writable and if needed move the
2220 * sender address from kernel to user space.
2221 */
2222int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2223 struct sockaddr __user *addr, int __user *addr_len)
2224{
2225 struct sockaddr_storage address;
2226 struct msghdr msg = {
2227 /* Save some cycles and don't copy the address if not needed */
2228 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2229 };
2230 struct socket *sock;
2231 int err, err2;
2232
2233 err = import_ubuf(ITER_DEST, ubuf, size, &msg.msg_iter);
2234 if (unlikely(err))
2235 return err;
2236
2237 CLASS(fd, f)(fd);
2238
2239 if (fd_empty(f))
2240 return -EBADF;
2241 sock = sock_from_file(fd_file(f));
2242 if (unlikely(!sock))
2243 return -ENOTSOCK;
2244
2245 if (sock->file->f_flags & O_NONBLOCK)
2246 flags |= MSG_DONTWAIT;
2247 err = sock_recvmsg(sock, &msg, flags);
2248
2249 if (err >= 0 && addr != NULL) {
2250 err2 = move_addr_to_user(&address,
2251 msg.msg_namelen, addr, addr_len);
2252 if (err2 < 0)
2253 err = err2;
2254 }
2255 return err;
2256}
2257
2258SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2259 unsigned int, flags, struct sockaddr __user *, addr,
2260 int __user *, addr_len)
2261{
2262 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2263}
2264
2265/*
2266 * Receive a datagram from a socket.
2267 */
2268
2269SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2270 unsigned int, flags)
2271{
2272 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2273}
2274
2275static bool sock_use_custom_sol_socket(const struct socket *sock)
2276{
2277 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2278}
2279
2280int do_sock_setsockopt(struct socket *sock, bool compat, int level,
2281 int optname, sockptr_t optval, int optlen)
2282{
2283 const struct proto_ops *ops;
2284 char *kernel_optval = NULL;
2285 int err;
2286
2287 if (optlen < 0)
2288 return -EINVAL;
2289
2290 err = security_socket_setsockopt(sock, level, optname);
2291 if (err)
2292 goto out_put;
2293
2294 if (!compat)
2295 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2296 optval, &optlen,
2297 &kernel_optval);
2298 if (err < 0)
2299 goto out_put;
2300 if (err > 0) {
2301 err = 0;
2302 goto out_put;
2303 }
2304
2305 if (kernel_optval)
2306 optval = KERNEL_SOCKPTR(kernel_optval);
2307 ops = READ_ONCE(sock->ops);
2308 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2309 err = sock_setsockopt(sock, level, optname, optval, optlen);
2310 else if (unlikely(!ops->setsockopt))
2311 err = -EOPNOTSUPP;
2312 else
2313 err = ops->setsockopt(sock, level, optname, optval,
2314 optlen);
2315 kfree(kernel_optval);
2316out_put:
2317 return err;
2318}
2319EXPORT_SYMBOL(do_sock_setsockopt);
2320
2321/* Set a socket option. Because we don't know the option lengths we have
2322 * to pass the user mode parameter for the protocols to sort out.
2323 */
2324int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2325 int optlen)
2326{
2327 sockptr_t optval = USER_SOCKPTR(user_optval);
2328 bool compat = in_compat_syscall();
2329 struct socket *sock;
2330 CLASS(fd, f)(fd);
2331
2332 if (fd_empty(f))
2333 return -EBADF;
2334 sock = sock_from_file(fd_file(f));
2335 if (unlikely(!sock))
2336 return -ENOTSOCK;
2337
2338 return do_sock_setsockopt(sock, compat, level, optname, optval, optlen);
2339}
2340
2341SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2342 char __user *, optval, int, optlen)
2343{
2344 return __sys_setsockopt(fd, level, optname, optval, optlen);
2345}
2346
2347INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2348 int optname));
2349
2350int do_sock_getsockopt(struct socket *sock, bool compat, int level,
2351 int optname, sockptr_t optval, sockptr_t optlen)
2352{
2353 int max_optlen __maybe_unused = 0;
2354 const struct proto_ops *ops;
2355 int err;
2356
2357 err = security_socket_getsockopt(sock, level, optname);
2358 if (err)
2359 return err;
2360
2361 if (!compat)
2362 copy_from_sockptr(&max_optlen, optlen, sizeof(int));
2363
2364 ops = READ_ONCE(sock->ops);
2365 if (level == SOL_SOCKET) {
2366 err = sk_getsockopt(sock->sk, level, optname, optval, optlen);
2367 } else if (unlikely(!ops->getsockopt)) {
2368 err = -EOPNOTSUPP;
2369 } else {
2370 if (WARN_ONCE(optval.is_kernel || optlen.is_kernel,
2371 "Invalid argument type"))
2372 return -EOPNOTSUPP;
2373
2374 err = ops->getsockopt(sock, level, optname, optval.user,
2375 optlen.user);
2376 }
2377
2378 if (!compat)
2379 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2380 optval, optlen, max_optlen,
2381 err);
2382
2383 return err;
2384}
2385EXPORT_SYMBOL(do_sock_getsockopt);
2386
2387/*
2388 * Get a socket option. Because we don't know the option lengths we have
2389 * to pass a user mode parameter for the protocols to sort out.
2390 */
2391int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2392 int __user *optlen)
2393{
2394 struct socket *sock;
2395 CLASS(fd, f)(fd);
2396
2397 if (fd_empty(f))
2398 return -EBADF;
2399 sock = sock_from_file(fd_file(f));
2400 if (unlikely(!sock))
2401 return -ENOTSOCK;
2402
2403 return do_sock_getsockopt(sock, in_compat_syscall(), level, optname,
2404 USER_SOCKPTR(optval), USER_SOCKPTR(optlen));
2405}
2406
2407SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2408 char __user *, optval, int __user *, optlen)
2409{
2410 return __sys_getsockopt(fd, level, optname, optval, optlen);
2411}
2412
2413/*
2414 * Shutdown a socket.
2415 */
2416
2417int __sys_shutdown_sock(struct socket *sock, int how)
2418{
2419 int err;
2420
2421 err = security_socket_shutdown(sock, how);
2422 if (!err)
2423 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2424
2425 return err;
2426}
2427
2428int __sys_shutdown(int fd, int how)
2429{
2430 struct socket *sock;
2431 CLASS(fd, f)(fd);
2432
2433 if (fd_empty(f))
2434 return -EBADF;
2435 sock = sock_from_file(fd_file(f));
2436 if (unlikely(!sock))
2437 return -ENOTSOCK;
2438
2439 return __sys_shutdown_sock(sock, how);
2440}
2441
2442SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2443{
2444 return __sys_shutdown(fd, how);
2445}
2446
2447/* A couple of helpful macros for getting the address of the 32/64 bit
2448 * fields which are the same type (int / unsigned) on our platforms.
2449 */
2450#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2451#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2452#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2453
2454struct used_address {
2455 struct sockaddr_storage name;
2456 unsigned int name_len;
2457};
2458
2459int __copy_msghdr(struct msghdr *kmsg,
2460 struct user_msghdr *msg,
2461 struct sockaddr __user **save_addr)
2462{
2463 ssize_t err;
2464
2465 kmsg->msg_control_is_user = true;
2466 kmsg->msg_get_inq = 0;
2467 kmsg->msg_control_user = msg->msg_control;
2468 kmsg->msg_controllen = msg->msg_controllen;
2469 kmsg->msg_flags = msg->msg_flags;
2470
2471 kmsg->msg_namelen = msg->msg_namelen;
2472 if (!msg->msg_name)
2473 kmsg->msg_namelen = 0;
2474
2475 if (kmsg->msg_namelen < 0)
2476 return -EINVAL;
2477
2478 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2479 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2480
2481 if (save_addr)
2482 *save_addr = msg->msg_name;
2483
2484 if (msg->msg_name && kmsg->msg_namelen) {
2485 if (!save_addr) {
2486 err = move_addr_to_kernel(msg->msg_name,
2487 kmsg->msg_namelen,
2488 kmsg->msg_name);
2489 if (err < 0)
2490 return err;
2491 }
2492 } else {
2493 kmsg->msg_name = NULL;
2494 kmsg->msg_namelen = 0;
2495 }
2496
2497 if (msg->msg_iovlen > UIO_MAXIOV)
2498 return -EMSGSIZE;
2499
2500 kmsg->msg_iocb = NULL;
2501 kmsg->msg_ubuf = NULL;
2502 return 0;
2503}
2504
2505static int copy_msghdr_from_user(struct msghdr *kmsg,
2506 struct user_msghdr __user *umsg,
2507 struct sockaddr __user **save_addr,
2508 struct iovec **iov)
2509{
2510 struct user_msghdr msg;
2511 ssize_t err;
2512
2513 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2514 return -EFAULT;
2515
2516 err = __copy_msghdr(kmsg, &msg, save_addr);
2517 if (err)
2518 return err;
2519
2520 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2521 msg.msg_iov, msg.msg_iovlen,
2522 UIO_FASTIOV, iov, &kmsg->msg_iter);
2523 return err < 0 ? err : 0;
2524}
2525
2526static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2527 unsigned int flags, struct used_address *used_address,
2528 unsigned int allowed_msghdr_flags)
2529{
2530 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2531 __aligned(sizeof(__kernel_size_t));
2532 /* 20 is size of ipv6_pktinfo */
2533 unsigned char *ctl_buf = ctl;
2534 int ctl_len;
2535 ssize_t err;
2536
2537 err = -ENOBUFS;
2538
2539 if (msg_sys->msg_controllen > INT_MAX)
2540 goto out;
2541 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2542 ctl_len = msg_sys->msg_controllen;
2543 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2544 err =
2545 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2546 sizeof(ctl));
2547 if (err)
2548 goto out;
2549 ctl_buf = msg_sys->msg_control;
2550 ctl_len = msg_sys->msg_controllen;
2551 } else if (ctl_len) {
2552 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2553 CMSG_ALIGN(sizeof(struct cmsghdr)));
2554 if (ctl_len > sizeof(ctl)) {
2555 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2556 if (ctl_buf == NULL)
2557 goto out;
2558 }
2559 err = -EFAULT;
2560 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2561 goto out_freectl;
2562 msg_sys->msg_control = ctl_buf;
2563 msg_sys->msg_control_is_user = false;
2564 }
2565 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2566 msg_sys->msg_flags = flags;
2567
2568 if (sock->file->f_flags & O_NONBLOCK)
2569 msg_sys->msg_flags |= MSG_DONTWAIT;
2570 /*
2571 * If this is sendmmsg() and current destination address is same as
2572 * previously succeeded address, omit asking LSM's decision.
2573 * used_address->name_len is initialized to UINT_MAX so that the first
2574 * destination address never matches.
2575 */
2576 if (used_address && msg_sys->msg_name &&
2577 used_address->name_len == msg_sys->msg_namelen &&
2578 !memcmp(&used_address->name, msg_sys->msg_name,
2579 used_address->name_len)) {
2580 err = sock_sendmsg_nosec(sock, msg_sys);
2581 goto out_freectl;
2582 }
2583 err = __sock_sendmsg(sock, msg_sys);
2584 /*
2585 * If this is sendmmsg() and sending to current destination address was
2586 * successful, remember it.
2587 */
2588 if (used_address && err >= 0) {
2589 used_address->name_len = msg_sys->msg_namelen;
2590 if (msg_sys->msg_name)
2591 memcpy(&used_address->name, msg_sys->msg_name,
2592 used_address->name_len);
2593 }
2594
2595out_freectl:
2596 if (ctl_buf != ctl)
2597 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2598out:
2599 return err;
2600}
2601
2602static int sendmsg_copy_msghdr(struct msghdr *msg,
2603 struct user_msghdr __user *umsg, unsigned flags,
2604 struct iovec **iov)
2605{
2606 int err;
2607
2608 if (flags & MSG_CMSG_COMPAT) {
2609 struct compat_msghdr __user *msg_compat;
2610
2611 msg_compat = (struct compat_msghdr __user *) umsg;
2612 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2613 } else {
2614 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2615 }
2616 if (err < 0)
2617 return err;
2618
2619 return 0;
2620}
2621
2622static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2623 struct msghdr *msg_sys, unsigned int flags,
2624 struct used_address *used_address,
2625 unsigned int allowed_msghdr_flags)
2626{
2627 struct sockaddr_storage address;
2628 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2629 ssize_t err;
2630
2631 msg_sys->msg_name = &address;
2632
2633 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2634 if (err < 0)
2635 return err;
2636
2637 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2638 allowed_msghdr_flags);
2639 kfree(iov);
2640 return err;
2641}
2642
2643/*
2644 * BSD sendmsg interface
2645 */
2646long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2647 unsigned int flags)
2648{
2649 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2650}
2651
2652long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2653 bool forbid_cmsg_compat)
2654{
2655 struct msghdr msg_sys;
2656 struct socket *sock;
2657
2658 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2659 return -EINVAL;
2660
2661 CLASS(fd, f)(fd);
2662
2663 if (fd_empty(f))
2664 return -EBADF;
2665 sock = sock_from_file(fd_file(f));
2666 if (unlikely(!sock))
2667 return -ENOTSOCK;
2668
2669 return ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2670}
2671
2672SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2673{
2674 return __sys_sendmsg(fd, msg, flags, true);
2675}
2676
2677/*
2678 * Linux sendmmsg interface
2679 */
2680
2681int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2682 unsigned int flags, bool forbid_cmsg_compat)
2683{
2684 int err, datagrams;
2685 struct socket *sock;
2686 struct mmsghdr __user *entry;
2687 struct compat_mmsghdr __user *compat_entry;
2688 struct msghdr msg_sys;
2689 struct used_address used_address;
2690 unsigned int oflags = flags;
2691
2692 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2693 return -EINVAL;
2694
2695 if (vlen > UIO_MAXIOV)
2696 vlen = UIO_MAXIOV;
2697
2698 datagrams = 0;
2699
2700 CLASS(fd, f)(fd);
2701
2702 if (fd_empty(f))
2703 return -EBADF;
2704 sock = sock_from_file(fd_file(f));
2705 if (unlikely(!sock))
2706 return -ENOTSOCK;
2707
2708 used_address.name_len = UINT_MAX;
2709 entry = mmsg;
2710 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2711 err = 0;
2712 flags |= MSG_BATCH;
2713
2714 while (datagrams < vlen) {
2715 if (datagrams == vlen - 1)
2716 flags = oflags;
2717
2718 if (MSG_CMSG_COMPAT & flags) {
2719 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2720 &msg_sys, flags, &used_address, MSG_EOR);
2721 if (err < 0)
2722 break;
2723 err = __put_user(err, &compat_entry->msg_len);
2724 ++compat_entry;
2725 } else {
2726 err = ___sys_sendmsg(sock,
2727 (struct user_msghdr __user *)entry,
2728 &msg_sys, flags, &used_address, MSG_EOR);
2729 if (err < 0)
2730 break;
2731 err = put_user(err, &entry->msg_len);
2732 ++entry;
2733 }
2734
2735 if (err)
2736 break;
2737 ++datagrams;
2738 if (msg_data_left(&msg_sys))
2739 break;
2740 cond_resched();
2741 }
2742
2743 /* We only return an error if no datagrams were able to be sent */
2744 if (datagrams != 0)
2745 return datagrams;
2746
2747 return err;
2748}
2749
2750SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2751 unsigned int, vlen, unsigned int, flags)
2752{
2753 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2754}
2755
2756static int recvmsg_copy_msghdr(struct msghdr *msg,
2757 struct user_msghdr __user *umsg, unsigned flags,
2758 struct sockaddr __user **uaddr,
2759 struct iovec **iov)
2760{
2761 ssize_t err;
2762
2763 if (MSG_CMSG_COMPAT & flags) {
2764 struct compat_msghdr __user *msg_compat;
2765
2766 msg_compat = (struct compat_msghdr __user *) umsg;
2767 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2768 } else {
2769 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2770 }
2771 if (err < 0)
2772 return err;
2773
2774 return 0;
2775}
2776
2777static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2778 struct user_msghdr __user *msg,
2779 struct sockaddr __user *uaddr,
2780 unsigned int flags, int nosec)
2781{
2782 struct compat_msghdr __user *msg_compat =
2783 (struct compat_msghdr __user *) msg;
2784 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2785 struct sockaddr_storage addr;
2786 unsigned long cmsg_ptr;
2787 int len;
2788 ssize_t err;
2789
2790 msg_sys->msg_name = &addr;
2791 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2792 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2793
2794 /* We assume all kernel code knows the size of sockaddr_storage */
2795 msg_sys->msg_namelen = 0;
2796
2797 if (sock->file->f_flags & O_NONBLOCK)
2798 flags |= MSG_DONTWAIT;
2799
2800 if (unlikely(nosec))
2801 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2802 else
2803 err = sock_recvmsg(sock, msg_sys, flags);
2804
2805 if (err < 0)
2806 goto out;
2807 len = err;
2808
2809 if (uaddr != NULL) {
2810 err = move_addr_to_user(&addr,
2811 msg_sys->msg_namelen, uaddr,
2812 uaddr_len);
2813 if (err < 0)
2814 goto out;
2815 }
2816 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2817 COMPAT_FLAGS(msg));
2818 if (err)
2819 goto out;
2820 if (MSG_CMSG_COMPAT & flags)
2821 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2822 &msg_compat->msg_controllen);
2823 else
2824 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2825 &msg->msg_controllen);
2826 if (err)
2827 goto out;
2828 err = len;
2829out:
2830 return err;
2831}
2832
2833static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2834 struct msghdr *msg_sys, unsigned int flags, int nosec)
2835{
2836 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2837 /* user mode address pointers */
2838 struct sockaddr __user *uaddr;
2839 ssize_t err;
2840
2841 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2842 if (err < 0)
2843 return err;
2844
2845 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2846 kfree(iov);
2847 return err;
2848}
2849
2850/*
2851 * BSD recvmsg interface
2852 */
2853
2854long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2855 struct user_msghdr __user *umsg,
2856 struct sockaddr __user *uaddr, unsigned int flags)
2857{
2858 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2859}
2860
2861long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2862 bool forbid_cmsg_compat)
2863{
2864 struct msghdr msg_sys;
2865 struct socket *sock;
2866
2867 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2868 return -EINVAL;
2869
2870 CLASS(fd, f)(fd);
2871
2872 if (fd_empty(f))
2873 return -EBADF;
2874 sock = sock_from_file(fd_file(f));
2875 if (unlikely(!sock))
2876 return -ENOTSOCK;
2877
2878 return ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2879}
2880
2881SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2882 unsigned int, flags)
2883{
2884 return __sys_recvmsg(fd, msg, flags, true);
2885}
2886
2887/*
2888 * Linux recvmmsg interface
2889 */
2890
2891static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2892 unsigned int vlen, unsigned int flags,
2893 struct timespec64 *timeout)
2894{
2895 int err = 0, datagrams;
2896 struct socket *sock;
2897 struct mmsghdr __user *entry;
2898 struct compat_mmsghdr __user *compat_entry;
2899 struct msghdr msg_sys;
2900 struct timespec64 end_time;
2901 struct timespec64 timeout64;
2902
2903 if (timeout &&
2904 poll_select_set_timeout(&end_time, timeout->tv_sec,
2905 timeout->tv_nsec))
2906 return -EINVAL;
2907
2908 datagrams = 0;
2909
2910 CLASS(fd, f)(fd);
2911
2912 if (fd_empty(f))
2913 return -EBADF;
2914 sock = sock_from_file(fd_file(f));
2915 if (unlikely(!sock))
2916 return -ENOTSOCK;
2917
2918 if (likely(!(flags & MSG_ERRQUEUE))) {
2919 err = sock_error(sock->sk);
2920 if (err)
2921 return err;
2922 }
2923
2924 entry = mmsg;
2925 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2926
2927 while (datagrams < vlen) {
2928 /*
2929 * No need to ask LSM for more than the first datagram.
2930 */
2931 if (MSG_CMSG_COMPAT & flags) {
2932 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2933 &msg_sys, flags & ~MSG_WAITFORONE,
2934 datagrams);
2935 if (err < 0)
2936 break;
2937 err = __put_user(err, &compat_entry->msg_len);
2938 ++compat_entry;
2939 } else {
2940 err = ___sys_recvmsg(sock,
2941 (struct user_msghdr __user *)entry,
2942 &msg_sys, flags & ~MSG_WAITFORONE,
2943 datagrams);
2944 if (err < 0)
2945 break;
2946 err = put_user(err, &entry->msg_len);
2947 ++entry;
2948 }
2949
2950 if (err)
2951 break;
2952 ++datagrams;
2953
2954 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2955 if (flags & MSG_WAITFORONE)
2956 flags |= MSG_DONTWAIT;
2957
2958 if (timeout) {
2959 ktime_get_ts64(&timeout64);
2960 *timeout = timespec64_sub(end_time, timeout64);
2961 if (timeout->tv_sec < 0) {
2962 timeout->tv_sec = timeout->tv_nsec = 0;
2963 break;
2964 }
2965
2966 /* Timeout, return less than vlen datagrams */
2967 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2968 break;
2969 }
2970
2971 /* Out of band data, return right away */
2972 if (msg_sys.msg_flags & MSG_OOB)
2973 break;
2974 cond_resched();
2975 }
2976
2977 if (err == 0)
2978 return datagrams;
2979
2980 if (datagrams == 0)
2981 return err;
2982
2983 /*
2984 * We may return less entries than requested (vlen) if the
2985 * sock is non block and there aren't enough datagrams...
2986 */
2987 if (err != -EAGAIN) {
2988 /*
2989 * ... or if recvmsg returns an error after we
2990 * received some datagrams, where we record the
2991 * error to return on the next call or if the
2992 * app asks about it using getsockopt(SO_ERROR).
2993 */
2994 WRITE_ONCE(sock->sk->sk_err, -err);
2995 }
2996 return datagrams;
2997}
2998
2999int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
3000 unsigned int vlen, unsigned int flags,
3001 struct __kernel_timespec __user *timeout,
3002 struct old_timespec32 __user *timeout32)
3003{
3004 int datagrams;
3005 struct timespec64 timeout_sys;
3006
3007 if (timeout && get_timespec64(&timeout_sys, timeout))
3008 return -EFAULT;
3009
3010 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
3011 return -EFAULT;
3012
3013 if (!timeout && !timeout32)
3014 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
3015
3016 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
3017
3018 if (datagrams <= 0)
3019 return datagrams;
3020
3021 if (timeout && put_timespec64(&timeout_sys, timeout))
3022 datagrams = -EFAULT;
3023
3024 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
3025 datagrams = -EFAULT;
3026
3027 return datagrams;
3028}
3029
3030SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
3031 unsigned int, vlen, unsigned int, flags,
3032 struct __kernel_timespec __user *, timeout)
3033{
3034 if (flags & MSG_CMSG_COMPAT)
3035 return -EINVAL;
3036
3037 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
3038}
3039
3040#ifdef CONFIG_COMPAT_32BIT_TIME
3041SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
3042 unsigned int, vlen, unsigned int, flags,
3043 struct old_timespec32 __user *, timeout)
3044{
3045 if (flags & MSG_CMSG_COMPAT)
3046 return -EINVAL;
3047
3048 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
3049}
3050#endif
3051
3052#ifdef __ARCH_WANT_SYS_SOCKETCALL
3053/* Argument list sizes for sys_socketcall */
3054#define AL(x) ((x) * sizeof(unsigned long))
3055static const unsigned char nargs[21] = {
3056 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
3057 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
3058 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
3059 AL(4), AL(5), AL(4)
3060};
3061
3062#undef AL
3063
3064/*
3065 * System call vectors.
3066 *
3067 * Argument checking cleaned up. Saved 20% in size.
3068 * This function doesn't need to set the kernel lock because
3069 * it is set by the callees.
3070 */
3071
3072SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3073{
3074 unsigned long a[AUDITSC_ARGS];
3075 unsigned long a0, a1;
3076 int err;
3077 unsigned int len;
3078
3079 if (call < 1 || call > SYS_SENDMMSG)
3080 return -EINVAL;
3081 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3082
3083 len = nargs[call];
3084 if (len > sizeof(a))
3085 return -EINVAL;
3086
3087 /* copy_from_user should be SMP safe. */
3088 if (copy_from_user(a, args, len))
3089 return -EFAULT;
3090
3091 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3092 if (err)
3093 return err;
3094
3095 a0 = a[0];
3096 a1 = a[1];
3097
3098 switch (call) {
3099 case SYS_SOCKET:
3100 err = __sys_socket(a0, a1, a[2]);
3101 break;
3102 case SYS_BIND:
3103 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3104 break;
3105 case SYS_CONNECT:
3106 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3107 break;
3108 case SYS_LISTEN:
3109 err = __sys_listen(a0, a1);
3110 break;
3111 case SYS_ACCEPT:
3112 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3113 (int __user *)a[2], 0);
3114 break;
3115 case SYS_GETSOCKNAME:
3116 err =
3117 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3118 (int __user *)a[2]);
3119 break;
3120 case SYS_GETPEERNAME:
3121 err =
3122 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3123 (int __user *)a[2]);
3124 break;
3125 case SYS_SOCKETPAIR:
3126 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3127 break;
3128 case SYS_SEND:
3129 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3130 NULL, 0);
3131 break;
3132 case SYS_SENDTO:
3133 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3134 (struct sockaddr __user *)a[4], a[5]);
3135 break;
3136 case SYS_RECV:
3137 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3138 NULL, NULL);
3139 break;
3140 case SYS_RECVFROM:
3141 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3142 (struct sockaddr __user *)a[4],
3143 (int __user *)a[5]);
3144 break;
3145 case SYS_SHUTDOWN:
3146 err = __sys_shutdown(a0, a1);
3147 break;
3148 case SYS_SETSOCKOPT:
3149 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3150 a[4]);
3151 break;
3152 case SYS_GETSOCKOPT:
3153 err =
3154 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3155 (int __user *)a[4]);
3156 break;
3157 case SYS_SENDMSG:
3158 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3159 a[2], true);
3160 break;
3161 case SYS_SENDMMSG:
3162 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3163 a[3], true);
3164 break;
3165 case SYS_RECVMSG:
3166 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3167 a[2], true);
3168 break;
3169 case SYS_RECVMMSG:
3170 if (IS_ENABLED(CONFIG_64BIT))
3171 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3172 a[2], a[3],
3173 (struct __kernel_timespec __user *)a[4],
3174 NULL);
3175 else
3176 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3177 a[2], a[3], NULL,
3178 (struct old_timespec32 __user *)a[4]);
3179 break;
3180 case SYS_ACCEPT4:
3181 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3182 (int __user *)a[2], a[3]);
3183 break;
3184 default:
3185 err = -EINVAL;
3186 break;
3187 }
3188 return err;
3189}
3190
3191#endif /* __ARCH_WANT_SYS_SOCKETCALL */
3192
3193/**
3194 * sock_register - add a socket protocol handler
3195 * @ops: description of protocol
3196 *
3197 * This function is called by a protocol handler that wants to
3198 * advertise its address family, and have it linked into the
3199 * socket interface. The value ops->family corresponds to the
3200 * socket system call protocol family.
3201 */
3202int sock_register(const struct net_proto_family *ops)
3203{
3204 int err;
3205
3206 if (ops->family >= NPROTO) {
3207 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3208 return -ENOBUFS;
3209 }
3210
3211 spin_lock(&net_family_lock);
3212 if (rcu_dereference_protected(net_families[ops->family],
3213 lockdep_is_held(&net_family_lock)))
3214 err = -EEXIST;
3215 else {
3216 rcu_assign_pointer(net_families[ops->family], ops);
3217 err = 0;
3218 }
3219 spin_unlock(&net_family_lock);
3220
3221 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3222 return err;
3223}
3224EXPORT_SYMBOL(sock_register);
3225
3226/**
3227 * sock_unregister - remove a protocol handler
3228 * @family: protocol family to remove
3229 *
3230 * This function is called by a protocol handler that wants to
3231 * remove its address family, and have it unlinked from the
3232 * new socket creation.
3233 *
3234 * If protocol handler is a module, then it can use module reference
3235 * counts to protect against new references. If protocol handler is not
3236 * a module then it needs to provide its own protection in
3237 * the ops->create routine.
3238 */
3239void sock_unregister(int family)
3240{
3241 BUG_ON(family < 0 || family >= NPROTO);
3242
3243 spin_lock(&net_family_lock);
3244 RCU_INIT_POINTER(net_families[family], NULL);
3245 spin_unlock(&net_family_lock);
3246
3247 synchronize_rcu();
3248
3249 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3250}
3251EXPORT_SYMBOL(sock_unregister);
3252
3253bool sock_is_registered(int family)
3254{
3255 return family < NPROTO && rcu_access_pointer(net_families[family]);
3256}
3257
3258static int __init sock_init(void)
3259{
3260 int err;
3261 /*
3262 * Initialize the network sysctl infrastructure.
3263 */
3264 err = net_sysctl_init();
3265 if (err)
3266 goto out;
3267
3268 /*
3269 * Initialize skbuff SLAB cache
3270 */
3271 skb_init();
3272
3273 /*
3274 * Initialize the protocols module.
3275 */
3276
3277 init_inodecache();
3278
3279 err = register_filesystem(&sock_fs_type);
3280 if (err)
3281 goto out;
3282 sock_mnt = kern_mount(&sock_fs_type);
3283 if (IS_ERR(sock_mnt)) {
3284 err = PTR_ERR(sock_mnt);
3285 goto out_mount;
3286 }
3287
3288 /* The real protocol initialization is performed in later initcalls.
3289 */
3290
3291#ifdef CONFIG_NETFILTER
3292 err = netfilter_init();
3293 if (err)
3294 goto out;
3295#endif
3296
3297 ptp_classifier_init();
3298
3299out:
3300 return err;
3301
3302out_mount:
3303 unregister_filesystem(&sock_fs_type);
3304 goto out;
3305}
3306
3307core_initcall(sock_init); /* early initcall */
3308
3309#ifdef CONFIG_PROC_FS
3310void socket_seq_show(struct seq_file *seq)
3311{
3312 seq_printf(seq, "sockets: used %d\n",
3313 sock_inuse_get(seq->private));
3314}
3315#endif /* CONFIG_PROC_FS */
3316
3317/* Handle the fact that while struct ifreq has the same *layout* on
3318 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3319 * which are handled elsewhere, it still has different *size* due to
3320 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3321 * resulting in struct ifreq being 32 and 40 bytes respectively).
3322 * As a result, if the struct happens to be at the end of a page and
3323 * the next page isn't readable/writable, we get a fault. To prevent
3324 * that, copy back and forth to the full size.
3325 */
3326int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3327{
3328 if (in_compat_syscall()) {
3329 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3330
3331 memset(ifr, 0, sizeof(*ifr));
3332 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3333 return -EFAULT;
3334
3335 if (ifrdata)
3336 *ifrdata = compat_ptr(ifr32->ifr_data);
3337
3338 return 0;
3339 }
3340
3341 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3342 return -EFAULT;
3343
3344 if (ifrdata)
3345 *ifrdata = ifr->ifr_data;
3346
3347 return 0;
3348}
3349EXPORT_SYMBOL(get_user_ifreq);
3350
3351int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3352{
3353 size_t size = sizeof(*ifr);
3354
3355 if (in_compat_syscall())
3356 size = sizeof(struct compat_ifreq);
3357
3358 if (copy_to_user(arg, ifr, size))
3359 return -EFAULT;
3360
3361 return 0;
3362}
3363EXPORT_SYMBOL(put_user_ifreq);
3364
3365#ifdef CONFIG_COMPAT
3366static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3367{
3368 compat_uptr_t uptr32;
3369 struct ifreq ifr;
3370 void __user *saved;
3371 int err;
3372
3373 if (get_user_ifreq(&ifr, NULL, uifr32))
3374 return -EFAULT;
3375
3376 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3377 return -EFAULT;
3378
3379 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3380 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3381
3382 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3383 if (!err) {
3384 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3385 if (put_user_ifreq(&ifr, uifr32))
3386 err = -EFAULT;
3387 }
3388 return err;
3389}
3390
3391/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3392static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3393 struct compat_ifreq __user *u_ifreq32)
3394{
3395 struct ifreq ifreq;
3396 void __user *data;
3397
3398 if (!is_socket_ioctl_cmd(cmd))
3399 return -ENOTTY;
3400 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3401 return -EFAULT;
3402 ifreq.ifr_data = data;
3403
3404 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3405}
3406
3407static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3408 unsigned int cmd, unsigned long arg)
3409{
3410 void __user *argp = compat_ptr(arg);
3411 struct sock *sk = sock->sk;
3412 struct net *net = sock_net(sk);
3413 const struct proto_ops *ops;
3414
3415 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3416 return sock_ioctl(file, cmd, (unsigned long)argp);
3417
3418 switch (cmd) {
3419 case SIOCWANDEV:
3420 return compat_siocwandev(net, argp);
3421 case SIOCGSTAMP_OLD:
3422 case SIOCGSTAMPNS_OLD:
3423 ops = READ_ONCE(sock->ops);
3424 if (!ops->gettstamp)
3425 return -ENOIOCTLCMD;
3426 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3427 !COMPAT_USE_64BIT_TIME);
3428
3429 case SIOCETHTOOL:
3430 case SIOCBONDSLAVEINFOQUERY:
3431 case SIOCBONDINFOQUERY:
3432 case SIOCSHWTSTAMP:
3433 case SIOCGHWTSTAMP:
3434 return compat_ifr_data_ioctl(net, cmd, argp);
3435
3436 case FIOSETOWN:
3437 case SIOCSPGRP:
3438 case FIOGETOWN:
3439 case SIOCGPGRP:
3440 case SIOCBRADDBR:
3441 case SIOCBRDELBR:
3442 case SIOCGIFVLAN:
3443 case SIOCSIFVLAN:
3444 case SIOCGSKNS:
3445 case SIOCGSTAMP_NEW:
3446 case SIOCGSTAMPNS_NEW:
3447 case SIOCGIFCONF:
3448 case SIOCSIFBR:
3449 case SIOCGIFBR:
3450 return sock_ioctl(file, cmd, arg);
3451
3452 case SIOCGIFFLAGS:
3453 case SIOCSIFFLAGS:
3454 case SIOCGIFMAP:
3455 case SIOCSIFMAP:
3456 case SIOCGIFMETRIC:
3457 case SIOCSIFMETRIC:
3458 case SIOCGIFMTU:
3459 case SIOCSIFMTU:
3460 case SIOCGIFMEM:
3461 case SIOCSIFMEM:
3462 case SIOCGIFHWADDR:
3463 case SIOCSIFHWADDR:
3464 case SIOCADDMULTI:
3465 case SIOCDELMULTI:
3466 case SIOCGIFINDEX:
3467 case SIOCGIFADDR:
3468 case SIOCSIFADDR:
3469 case SIOCSIFHWBROADCAST:
3470 case SIOCDIFADDR:
3471 case SIOCGIFBRDADDR:
3472 case SIOCSIFBRDADDR:
3473 case SIOCGIFDSTADDR:
3474 case SIOCSIFDSTADDR:
3475 case SIOCGIFNETMASK:
3476 case SIOCSIFNETMASK:
3477 case SIOCSIFPFLAGS:
3478 case SIOCGIFPFLAGS:
3479 case SIOCGIFTXQLEN:
3480 case SIOCSIFTXQLEN:
3481 case SIOCBRADDIF:
3482 case SIOCBRDELIF:
3483 case SIOCGIFNAME:
3484 case SIOCSIFNAME:
3485 case SIOCGMIIPHY:
3486 case SIOCGMIIREG:
3487 case SIOCSMIIREG:
3488 case SIOCBONDENSLAVE:
3489 case SIOCBONDRELEASE:
3490 case SIOCBONDSETHWADDR:
3491 case SIOCBONDCHANGEACTIVE:
3492 case SIOCSARP:
3493 case SIOCGARP:
3494 case SIOCDARP:
3495 case SIOCOUTQ:
3496 case SIOCOUTQNSD:
3497 case SIOCATMARK:
3498 return sock_do_ioctl(net, sock, cmd, arg);
3499 }
3500
3501 return -ENOIOCTLCMD;
3502}
3503
3504static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3505 unsigned long arg)
3506{
3507 struct socket *sock = file->private_data;
3508 const struct proto_ops *ops = READ_ONCE(sock->ops);
3509 int ret = -ENOIOCTLCMD;
3510 struct sock *sk;
3511 struct net *net;
3512
3513 sk = sock->sk;
3514 net = sock_net(sk);
3515
3516 if (ops->compat_ioctl)
3517 ret = ops->compat_ioctl(sock, cmd, arg);
3518
3519 if (ret == -ENOIOCTLCMD &&
3520 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3521 ret = compat_wext_handle_ioctl(net, cmd, arg);
3522
3523 if (ret == -ENOIOCTLCMD)
3524 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3525
3526 return ret;
3527}
3528#endif
3529
3530/**
3531 * kernel_bind - bind an address to a socket (kernel space)
3532 * @sock: socket
3533 * @addr: address
3534 * @addrlen: length of address
3535 *
3536 * Returns 0 or an error.
3537 */
3538
3539int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3540{
3541 struct sockaddr_storage address;
3542
3543 memcpy(&address, addr, addrlen);
3544
3545 return READ_ONCE(sock->ops)->bind(sock, (struct sockaddr *)&address,
3546 addrlen);
3547}
3548EXPORT_SYMBOL(kernel_bind);
3549
3550/**
3551 * kernel_listen - move socket to listening state (kernel space)
3552 * @sock: socket
3553 * @backlog: pending connections queue size
3554 *
3555 * Returns 0 or an error.
3556 */
3557
3558int kernel_listen(struct socket *sock, int backlog)
3559{
3560 return READ_ONCE(sock->ops)->listen(sock, backlog);
3561}
3562EXPORT_SYMBOL(kernel_listen);
3563
3564/**
3565 * kernel_accept - accept a connection (kernel space)
3566 * @sock: listening socket
3567 * @newsock: new connected socket
3568 * @flags: flags
3569 *
3570 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3571 * If it fails, @newsock is guaranteed to be %NULL.
3572 * Returns 0 or an error.
3573 */
3574
3575int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3576{
3577 struct sock *sk = sock->sk;
3578 const struct proto_ops *ops = READ_ONCE(sock->ops);
3579 struct proto_accept_arg arg = {
3580 .flags = flags,
3581 .kern = true,
3582 };
3583 int err;
3584
3585 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3586 newsock);
3587 if (err < 0)
3588 goto done;
3589
3590 err = ops->accept(sock, *newsock, &arg);
3591 if (err < 0) {
3592 sock_release(*newsock);
3593 *newsock = NULL;
3594 goto done;
3595 }
3596
3597 (*newsock)->ops = ops;
3598 __module_get(ops->owner);
3599
3600done:
3601 return err;
3602}
3603EXPORT_SYMBOL(kernel_accept);
3604
3605/**
3606 * kernel_connect - connect a socket (kernel space)
3607 * @sock: socket
3608 * @addr: address
3609 * @addrlen: address length
3610 * @flags: flags (O_NONBLOCK, ...)
3611 *
3612 * For datagram sockets, @addr is the address to which datagrams are sent
3613 * by default, and the only address from which datagrams are received.
3614 * For stream sockets, attempts to connect to @addr.
3615 * Returns 0 or an error code.
3616 */
3617
3618int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3619 int flags)
3620{
3621 struct sockaddr_storage address;
3622
3623 memcpy(&address, addr, addrlen);
3624
3625 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address,
3626 addrlen, flags);
3627}
3628EXPORT_SYMBOL(kernel_connect);
3629
3630/**
3631 * kernel_getsockname - get the address which the socket is bound (kernel space)
3632 * @sock: socket
3633 * @addr: address holder
3634 *
3635 * Fills the @addr pointer with the address which the socket is bound.
3636 * Returns the length of the address in bytes or an error code.
3637 */
3638
3639int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3640{
3641 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3642}
3643EXPORT_SYMBOL(kernel_getsockname);
3644
3645/**
3646 * kernel_getpeername - get the address which the socket is connected (kernel space)
3647 * @sock: socket
3648 * @addr: address holder
3649 *
3650 * Fills the @addr pointer with the address which the socket is connected.
3651 * Returns the length of the address in bytes or an error code.
3652 */
3653
3654int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3655{
3656 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3657}
3658EXPORT_SYMBOL(kernel_getpeername);
3659
3660/**
3661 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3662 * @sock: socket
3663 * @how: connection part
3664 *
3665 * Returns 0 or an error.
3666 */
3667
3668int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3669{
3670 return READ_ONCE(sock->ops)->shutdown(sock, how);
3671}
3672EXPORT_SYMBOL(kernel_sock_shutdown);
3673
3674/**
3675 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3676 * @sk: socket
3677 *
3678 * This routine returns the IP overhead imposed by a socket i.e.
3679 * the length of the underlying IP header, depending on whether
3680 * this is an IPv4 or IPv6 socket and the length from IP options turned
3681 * on at the socket. Assumes that the caller has a lock on the socket.
3682 */
3683
3684u32 kernel_sock_ip_overhead(struct sock *sk)
3685{
3686 struct inet_sock *inet;
3687 struct ip_options_rcu *opt;
3688 u32 overhead = 0;
3689#if IS_ENABLED(CONFIG_IPV6)
3690 struct ipv6_pinfo *np;
3691 struct ipv6_txoptions *optv6 = NULL;
3692#endif /* IS_ENABLED(CONFIG_IPV6) */
3693
3694 if (!sk)
3695 return overhead;
3696
3697 switch (sk->sk_family) {
3698 case AF_INET:
3699 inet = inet_sk(sk);
3700 overhead += sizeof(struct iphdr);
3701 opt = rcu_dereference_protected(inet->inet_opt,
3702 sock_owned_by_user(sk));
3703 if (opt)
3704 overhead += opt->opt.optlen;
3705 return overhead;
3706#if IS_ENABLED(CONFIG_IPV6)
3707 case AF_INET6:
3708 np = inet6_sk(sk);
3709 overhead += sizeof(struct ipv6hdr);
3710 if (np)
3711 optv6 = rcu_dereference_protected(np->opt,
3712 sock_owned_by_user(sk));
3713 if (optv6)
3714 overhead += (optv6->opt_flen + optv6->opt_nflen);
3715 return overhead;
3716#endif /* IS_ENABLED(CONFIG_IPV6) */
3717 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3718 return overhead;
3719 }
3720}
3721EXPORT_SYMBOL(kernel_sock_ip_overhead);