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