개념 설명 전체
net/socket.c · v6.18.37
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, <[email protected]>
8 * Ross Biro
9 * Fred N. van Kempen, <[email protected]>
10 *
11 * Fixes:
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * shutdown()
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
18 * top level.
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
23 * tty drivers).
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
26 * configurable.
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
35 * stuff.
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
41 * moment.
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
48 *
49 * This module is effectively the top level interface to the BSD socket
50 * paradigm.
51 *
52 * Based upon Swansea University Computer Society NET3.039
53 */
54
55 #include <linux/bpf-cgroup.h>
56 #include <linux/ethtool.h>
57 #include <linux/mm.h>
58 #include <linux/socket.h>
59 #include <linux/file.h>
60 #include <linux/splice.h>
61 #include <linux/net.h>
62 #include <linux/interrupt.h>
63 #include <linux/thread_info.h>
64 #include <linux/rcupdate.h>
65 #include <linux/netdevice.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/mutex.h>
69 #include <linux/if_bridge.h>
70 #include <linux/if_vlan.h>
71 #include <linux/ptp_classify.h>
72 #include <linux/init.h>
73 #include <linux/poll.h>
74 #include <linux/cache.h>
75 #include <linux/module.h>
76 #include <linux/highmem.h>
77 #include <linux/mount.h>
78 #include <linux/pseudo_fs.h>
79 #include <linux/security.h>
80 #include <linux/syscalls.h>
81 #include <linux/compat.h>
82 #include <linux/kmod.h>
83 #include <linux/audit.h>
84 #include <linux/wireless.h>
85 #include <linux/nsproxy.h>
86 #include <linux/magic.h>
87 #include <linux/slab.h>
88 #include <linux/xattr.h>
89 #include <linux/nospec.h>
90 #include <linux/indirect_call_wrapper.h>
91 #include <linux/io_uring/net.h>
92
93 #include <linux/uaccess.h>
94 #include <asm/unistd.h>
95
96 #include <net/compat.h>
97 #include <net/wext.h>
98 #include <net/cls_cgroup.h>
99
100 #include <net/sock.h>
101 #include <linux/netfilter.h>
102
103 #include <linux/if_tun.h>
104 #include <linux/ipv6_route.h>
105 #include <linux/route.h>
106 #include <linux/termios.h>
107 #include <linux/sockios.h>
108 #include <net/busy_poll.h>
109 #include <linux/errqueue.h>
110 #include <linux/ptp_clock_kernel.h>
111 #include <trace/events/sock.h>
112
113 #include "core/dev.h"
114
115 #ifdef CONFIG_NET_RX_BUSY_POLL
116 unsigned int sysctl_net_busy_read __read_mostly;
117 unsigned int sysctl_net_busy_poll __read_mostly;
118 #endif
119
120 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
121 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
122 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
123
124 static int sock_close(struct inode *inode, struct file *file);
125 static __poll_t sock_poll(struct file *file,
126 struct poll_table_struct *wait);
127 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
128 #ifdef CONFIG_COMPAT
129 static long compat_sock_ioctl(struct file *file,
130 unsigned int cmd, unsigned long arg);
131 #endif
132 static int sock_fasync(int fd, struct file *filp, int on);
133 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
134 struct pipe_inode_info *pipe, size_t len,
135 unsigned int flags);
136 static void sock_splice_eof(struct file *file);
137
138 #ifdef CONFIG_PROC_FS
139 static void sock_show_fdinfo(struct seq_file *m, struct file *f)
140 {
141 struct socket *sock = f->private_data;
142 const struct proto_ops *ops = READ_ONCE(sock->ops);
143
144 if (ops->show_fdinfo)
145 ops->show_fdinfo(m, sock);
146 }
147 #else
148 #define sock_show_fdinfo NULL
149 #endif
150
151 /*
152 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
153 * in the operation structures but are done directly via the socketcall() multiplexor.
154 */
155
156 static const struct file_operations socket_file_ops = {
157 .owner = THIS_MODULE,
158 .read_iter = sock_read_iter,
159 .write_iter = sock_write_iter,
160 .poll = sock_poll,
161 .unlocked_ioctl = sock_ioctl,
162 #ifdef CONFIG_COMPAT
163 .compat_ioctl = compat_sock_ioctl,
164 #endif
165 .uring_cmd = io_uring_cmd_sock,
166 .mmap = sock_mmap,
167 .release = sock_close,
168 .fasync = sock_fasync,
169 .splice_write = splice_to_socket,
170 .splice_read = sock_splice_read,
171 .splice_eof = sock_splice_eof,
172 .show_fdinfo = sock_show_fdinfo,
173 };
174
175 static const char * const pf_family_names[] = {
176 [PF_UNSPEC] = "PF_UNSPEC",
177 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
178 [PF_INET] = "PF_INET",
179 [PF_AX25] = "PF_AX25",
180 [PF_IPX] = "PF_IPX",
181 [PF_APPLETALK] = "PF_APPLETALK",
182 [PF_NETROM] = "PF_NETROM",
183 [PF_BRIDGE] = "PF_BRIDGE",
184 [PF_ATMPVC] = "PF_ATMPVC",
185 [PF_X25] = "PF_X25",
186 [PF_INET6] = "PF_INET6",
187 [PF_ROSE] = "PF_ROSE",
188 [PF_DECnet] = "PF_DECnet",
189 [PF_NETBEUI] = "PF_NETBEUI",
190 [PF_SECURITY] = "PF_SECURITY",
191 [PF_KEY] = "PF_KEY",
192 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
193 [PF_PACKET] = "PF_PACKET",
194 [PF_ASH] = "PF_ASH",
195 [PF_ECONET] = "PF_ECONET",
196 [PF_ATMSVC] = "PF_ATMSVC",
197 [PF_RDS] = "PF_RDS",
198 [PF_SNA] = "PF_SNA",
199 [PF_IRDA] = "PF_IRDA",
200 [PF_PPPOX] = "PF_PPPOX",
201 [PF_WANPIPE] = "PF_WANPIPE",
202 [PF_LLC] = "PF_LLC",
203 [PF_IB] = "PF_IB",
204 [PF_MPLS] = "PF_MPLS",
205 [PF_CAN] = "PF_CAN",
206 [PF_TIPC] = "PF_TIPC",
207 [PF_BLUETOOTH] = "PF_BLUETOOTH",
208 [PF_IUCV] = "PF_IUCV",
209 [PF_RXRPC] = "PF_RXRPC",
210 [PF_ISDN] = "PF_ISDN",
211 [PF_PHONET] = "PF_PHONET",
212 [PF_IEEE802154] = "PF_IEEE802154",
213 [PF_CAIF] = "PF_CAIF",
214 [PF_ALG] = "PF_ALG",
215 [PF_NFC] = "PF_NFC",
216 [PF_VSOCK] = "PF_VSOCK",
217 [PF_KCM] = "PF_KCM",
218 [PF_QIPCRTR] = "PF_QIPCRTR",
219 [PF_SMC] = "PF_SMC",
220 [PF_XDP] = "PF_XDP",
221 [PF_MCTP] = "PF_MCTP",
222 };
223
224 /*
225 * The protocol list. Each protocol is registered in here.
226 */
227
228 static DEFINE_SPINLOCK(net_family_lock);
229 static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
230
231 /*
232 * Support routines.
233 * Move socket addresses back and forth across the kernel/user
234 * divide and look after the messy bits.
235 */
236
237 /**
238 * move_addr_to_kernel - copy a socket address into kernel space
239 * @uaddr: Address in user space
240 * @kaddr: Address in kernel space
241 * @ulen: Length in user space
242 *
243 * The address is copied into kernel space. If the provided address is
244 * too long an error code of -EINVAL is returned. If the copy gives
245 * invalid addresses -EFAULT is returned. On a success 0 is returned.
246 */
247
248 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
249 {
250 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
251 return -EINVAL;
252 if (ulen == 0)
253 return 0;
254 if (copy_from_user(kaddr, uaddr, ulen))
255 return -EFAULT;
256 return audit_sockaddr(ulen, kaddr);
257 }
258
259 /**
260 * move_addr_to_user - copy an address to user space
261 * @kaddr: kernel space address
262 * @klen: length of address in kernel
263 * @uaddr: user space address
264 * @ulen: pointer to user length field
265 *
266 * The value pointed to by ulen on entry is the buffer length available.
267 * This is overwritten with the buffer space used. -EINVAL is returned
268 * if an overlong buffer is specified or a negative buffer size. -EFAULT
269 * is returned if either the buffer or the length field are not
270 * accessible.
271 * After copying the data up to the limit the user specifies, the true
272 * length of the data is written over the length limit the user
273 * specified. Zero is returned for a success.
274 */
275
276 static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
277 void __user *uaddr, int __user *ulen)
278 {
279 int len;
280
281 BUG_ON(klen > sizeof(struct sockaddr_storage));
282
283 if (can_do_masked_user_access())
284 ulen = masked_user_access_begin(ulen);
285 else if (!user_access_begin(ulen, 4))
286 return -EFAULT;
287
288 unsafe_get_user(len, ulen, efault_end);
289
290 if (len > klen)
291 len = klen;
292 /*
293 * "fromlen shall refer to the value before truncation.."
294 * 1003.1g
295 */
296 if (len >= 0)
297 unsafe_put_user(klen, ulen, efault_end);
298
299 user_access_end();
300
301 if (len) {
302 if (len < 0)
303 return -EINVAL;
304 if (audit_sockaddr(klen, kaddr))
305 return -ENOMEM;
306 if (copy_to_user(uaddr, kaddr, len))
307 return -EFAULT;
308 }
309 return 0;
310
311 efault_end:
312 user_access_end();
313 return -EFAULT;
314 }
315
316 static struct kmem_cache *sock_inode_cachep __ro_after_init;
317
318 static struct inode *sock_alloc_inode(struct super_block *sb)
319 {
320 struct socket_alloc *ei;
321
322 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
323 if (!ei)
324 return NULL;
325 init_waitqueue_head(&ei->socket.wq.wait);
326 ei->socket.wq.fasync_list = NULL;
327 ei->socket.wq.flags = 0;
328
329 ei->socket.state = SS_UNCONNECTED;
330 ei->socket.flags = 0;
331 ei->socket.ops = NULL;
332 ei->socket.sk = NULL;
333 ei->socket.file = NULL;
334
335 return &ei->vfs_inode;
336 }
337
338 static void sock_free_inode(struct inode *inode)
339 {
340 struct socket_alloc *ei;
341
342 ei = container_of(inode, struct socket_alloc, vfs_inode);
343 kmem_cache_free(sock_inode_cachep, ei);
344 }
345
346 static void init_once(void *foo)
347 {
348 struct socket_alloc *ei = (struct socket_alloc *)foo;
349
350 inode_init_once(&ei->vfs_inode);
351 }
352
353 static void init_inodecache(void)
354 {
355 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
356 sizeof(struct socket_alloc),
357 0,
358 (SLAB_HWCACHE_ALIGN |
359 SLAB_RECLAIM_ACCOUNT |
360 SLAB_ACCOUNT),
361 init_once);
362 BUG_ON(sock_inode_cachep == NULL);
363 }
364
365 static const struct super_operations sockfs_ops = {
366 .alloc_inode = sock_alloc_inode,
367 .free_inode = sock_free_inode,
368 .statfs = simple_statfs,
369 };
370
371 /*
372 * sockfs_dname() is called from d_path().
373 */
374 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
375 {
376 return dynamic_dname(buffer, buflen, "socket:[%lu]",
377 d_inode(dentry)->i_ino);
378 }
379
380 static const struct dentry_operations sockfs_dentry_operations = {
381 .d_dname = sockfs_dname,
382 };
383
384 static int sockfs_xattr_get(const struct xattr_handler *handler,
385 struct dentry *dentry, struct inode *inode,
386 const char *suffix, void *value, size_t size)
387 {
388 if (value) {
389 if (dentry->d_name.len + 1 > size)
390 return -ERANGE;
391 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
392 }
393 return dentry->d_name.len + 1;
394 }
395
396 #define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
397 #define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
398 #define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
399
400 static const struct xattr_handler sockfs_xattr_handler = {
401 .name = XATTR_NAME_SOCKPROTONAME,
402 .get = sockfs_xattr_get,
403 };
404
405 static int sockfs_security_xattr_set(const struct xattr_handler *handler,
406 struct mnt_idmap *idmap,
407 struct dentry *dentry, struct inode *inode,
408 const char *suffix, const void *value,
409 size_t size, int flags)
410 {
411 /* Handled by LSM. */
412 return -EAGAIN;
413 }
414
415 static const struct xattr_handler sockfs_security_xattr_handler = {
416 .prefix = XATTR_SECURITY_PREFIX,
417 .set = sockfs_security_xattr_set,
418 };
419
420 static const struct xattr_handler * const sockfs_xattr_handlers[] = {
421 &sockfs_xattr_handler,
422 &sockfs_security_xattr_handler,
423 NULL
424 };
425
426 static int sockfs_init_fs_context(struct fs_context *fc)
427 {
428 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
429 if (!ctx)
430 return -ENOMEM;
431 ctx->ops = &sockfs_ops;
432 ctx->dops = &sockfs_dentry_operations;
433 ctx->xattr = sockfs_xattr_handlers;
434 return 0;
435 }
436
437 static struct vfsmount *sock_mnt __read_mostly;
438
439 static struct file_system_type sock_fs_type = {
440 .name = "sockfs",
441 .init_fs_context = sockfs_init_fs_context,
442 .kill_sb = kill_anon_super,
443 };
444
445 /*
446 * Obtains the first available file descriptor and sets it up for use.
447 *
448 * These functions create file structures and maps them to fd space
449 * of the current process. On success it returns file descriptor
450 * and file struct implicitly stored in sock->file.
451 * Note that another thread may close file descriptor before we return
452 * from this function. We use the fact that now we do not refer
453 * to socket after mapping. If one day we will need it, this
454 * function will increment ref. count on file by 1.
455 *
456 * In any case returned fd MAY BE not valid!
457 * This race condition is unavoidable
458 * with shared fd spaces, we cannot solve it inside kernel,
459 * but we take care of internal coherence yet.
460 */
461
462 /**
463 * sock_alloc_file - Bind a &socket to a &file
464 * @sock: socket
465 * @flags: file status flags
466 * @dname: protocol name
467 *
468 * Returns the &file bound with @sock, implicitly storing it
469 * in sock->file. If dname is %NULL, sets to "".
470 *
471 * On failure @sock is released, and an ERR pointer is returned.
472 *
473 * This function uses GFP_KERNEL internally.
474 */
475
476 struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
477 {
478 struct file *file;
479
480 if (!dname)
481 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
482
483 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
484 O_RDWR | (flags & O_NONBLOCK),
485 &socket_file_ops);
486 if (IS_ERR(file)) {
487 sock_release(sock);
488 return file;
489 }
490
491 file->f_mode |= FMODE_NOWAIT;
492 sock->file = file;
493 file->private_data = sock;
494 stream_open(SOCK_INODE(sock), file);
495 /*
496 * Disable permission and pre-content events, but enable legacy
497 * inotify events for legacy users.
498 */
499 file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
500 return file;
501 }
502 EXPORT_SYMBOL(sock_alloc_file);
503
504 static int sock_map_fd(struct socket *sock, int flags)
505 {
506 struct file *newfile;
507 int fd = get_unused_fd_flags(flags);
508 if (unlikely(fd < 0)) {
509 sock_release(sock);
510 return fd;
511 }
512
513 newfile = sock_alloc_file(sock, flags, NULL);
514 if (!IS_ERR(newfile)) {
515 fd_install(fd, newfile);
516 return fd;
517 }
518
519 put_unused_fd(fd);
520 return PTR_ERR(newfile);
521 }
522
523 /**
524 * sock_from_file - Return the &socket bounded to @file.
525 * @file: file
526 *
527 * On failure returns %NULL.
528 */
529
530 struct socket *sock_from_file(struct file *file)
531 {
532 if (likely(file->f_op == &socket_file_ops))
533 return file->private_data; /* set in sock_alloc_file */
534
535 return NULL;
536 }
537 EXPORT_SYMBOL(sock_from_file);
538
539 /**
540 * sockfd_lookup - Go from a file number to its socket slot
541 * @fd: file handle
542 * @err: pointer to an error code return
543 *
544 * The file handle passed in is locked and the socket it is bound
545 * to is returned. If an error occurs the err pointer is overwritten
546 * with a negative errno code and NULL is returned. The function checks
547 * for both invalid handles and passing a handle which is not a socket.
548 *
549 * On a success the socket object pointer is returned.
550 */
551
552 struct socket *sockfd_lookup(int fd, int *err)
553 {
554 struct file *file;
555 struct socket *sock;
556
557 file = fget(fd);
558 if (!file) {
559 *err = -EBADF;
560 return NULL;
561 }
562
563 sock = sock_from_file(file);
564 if (!sock) {
565 *err = -ENOTSOCK;
566 fput(file);
567 }
568 return sock;
569 }
570 EXPORT_SYMBOL(sockfd_lookup);
571
572 static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
573 size_t size)
574 {
575 ssize_t len;
576 ssize_t used = 0;
577
578 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
579 if (len < 0)
580 return len;
581 used += len;
582 if (buffer) {
583 if (size < used)
584 return -ERANGE;
585 buffer += len;
586 }
587
588 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
589 used += len;
590 if (buffer) {
591 if (size < used)
592 return -ERANGE;
593 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
594 buffer += len;
595 }
596
597 return used;
598 }
599
600 static int sockfs_setattr(struct mnt_idmap *idmap,
601 struct dentry *dentry, struct iattr *iattr)
602 {
603 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
604
605 if (!err && (iattr->ia_valid & ATTR_UID)) {
606 struct socket *sock = SOCKET_I(d_inode(dentry));
607
608 if (sock->sk) {
609 /* Paired with READ_ONCE() in sk_uid() */
610 WRITE_ONCE(sock->sk->sk_uid, iattr->ia_uid);
611 } else {
612 err = -ENOENT;
613 }
614 }
615
616 return err;
617 }
618
619 static const struct inode_operations sockfs_inode_ops = {
620 .listxattr = sockfs_listxattr,
621 .setattr = sockfs_setattr,
622 };
623
624 /**
625 * sock_alloc - allocate a socket
626 *
627 * Allocate a new inode and socket object. The two are bound together
628 * and initialised. The socket is then returned. If we are out of inodes
629 * NULL is returned. This functions uses GFP_KERNEL internally.
630 */
631
632 struct socket *sock_alloc(void)
633 {
634 struct inode *inode;
635 struct socket *sock;
636
637 inode = new_inode_pseudo(sock_mnt->mnt_sb);
638 if (!inode)
639 return NULL;
640
641 sock = SOCKET_I(inode);
642
643 inode->i_ino = get_next_ino();
644 inode->i_mode = S_IFSOCK | S_IRWXUGO;
645 inode->i_uid = current_fsuid();
646 inode->i_gid = current_fsgid();
647 inode->i_op = &sockfs_inode_ops;
648
649 return sock;
650 }
651 EXPORT_SYMBOL(sock_alloc);
652
653 static void __sock_release(struct socket *sock, struct inode *inode)
654 {
655 const struct proto_ops *ops = READ_ONCE(sock->ops);
656
657 if (ops) {
658 struct module *owner = ops->owner;
659
660 if (inode)
661 inode_lock(inode);
662 ops->release(sock);
663 sock->sk = NULL;
664 if (inode)
665 inode_unlock(inode);
666 sock->ops = NULL;
667 module_put(owner);
668 }
669
670 if (sock->wq.fasync_list)
671 pr_err("%s: fasync list not empty!\n", __func__);
672
673 if (!sock->file) {
674 iput(SOCK_INODE(sock));
675 return;
676 }
677 WRITE_ONCE(sock->file, NULL);
678 }
679
680 /**
681 * sock_release - close a socket
682 * @sock: socket to close
683 *
684 * The socket is released from the protocol stack if it has a release
685 * callback, and the inode is then released if the socket is bound to
686 * an inode not a file.
687 */
688 void sock_release(struct socket *sock)
689 {
690 __sock_release(sock, NULL);
691 }
692 EXPORT_SYMBOL(sock_release);
693
694 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags)
695 {
696 u8 flags = *tx_flags;
697
698 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
699 flags |= SKBTX_HW_TSTAMP_NOBPF;
700
701 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
702 flags |= SKBTX_SW_TSTAMP;
703
704 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
705 flags |= SKBTX_SCHED_TSTAMP;
706
707 if (tsflags & SOF_TIMESTAMPING_TX_COMPLETION)
708 flags |= SKBTX_COMPLETION_TSTAMP;
709
710 *tx_flags = flags;
711 }
712 EXPORT_SYMBOL(__sock_tx_timestamp);
713
714 INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
715 size_t));
716 INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
717 size_t));
718
719 static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
720 int flags)
721 {
722 trace_sock_send_length(sk, ret, 0);
723 }
724
725 static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
726 {
727 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
728 inet_sendmsg, sock, msg,
729 msg_data_left(msg));
730 BUG_ON(ret == -EIOCBQUEUED);
731
732 if (trace_sock_send_length_enabled())
733 call_trace_sock_send_length(sock->sk, ret, 0);
734 return ret;
735 }
736
737 static int __sock_sendmsg(struct socket *sock, struct msghdr *msg)
738 {
739 int err = security_socket_sendmsg(sock, msg,
740 msg_data_left(msg));
741
742 return err ?: sock_sendmsg_nosec(sock, msg);
743 }
744
745 /**
746 * sock_sendmsg - send a message through @sock
747 * @sock: socket
748 * @msg: message to send
749 *
750 * Sends @msg through @sock, passing through LSM.
751 * Returns the number of bytes sent, or an error code.
752 */
753 int sock_sendmsg(struct socket *sock, struct msghdr *msg)
754 {
755 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name;
756 struct sockaddr_storage address;
757 int save_len = msg->msg_namelen;
758 int ret;
759
760 if (msg->msg_name) {
761 memcpy(&address, msg->msg_name, msg->msg_namelen);
762 msg->msg_name = &address;
763 }
764
765 ret = __sock_sendmsg(sock, msg);
766 msg->msg_name = save_addr;
767 msg->msg_namelen = save_len;
768
769 return ret;
770 }
771 EXPORT_SYMBOL(sock_sendmsg);
772
773 /**
774 * kernel_sendmsg - send a message through @sock (kernel-space)
775 * @sock: socket
776 * @msg: message header
777 * @vec: kernel vec
778 * @num: vec array length
779 * @size: total message data size
780 *
781 * Builds the message data with @vec and sends it through @sock.
782 * Returns the number of bytes sent, or an error code.
783 */
784
785 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
786 struct kvec *vec, size_t num, size_t size)
787 {
788 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
789 return sock_sendmsg(sock, msg);
790 }
791 EXPORT_SYMBOL(kernel_sendmsg);
792
793 static bool skb_is_err_queue(const struct sk_buff *skb)
794 {
795 /* Error-queue skbs are marked as PACKET_OUTGOING in
796 * skb_set_err_queue() and use the destructor installed by
797 * sock_queue_err_skb(). PACKET_OUTGOING alone is not unique:
798 * AF_PACKET outgoing taps use the same pkt_type.
799 */
800 return skb->pkt_type == PACKET_OUTGOING &&
801 skb->destructor == sock_rmem_free;
802 }
803
804 /* On transmit, software and hardware timestamps are returned independently.
805 * As the two skb clones share the hardware timestamp, which may be updated
806 * before the software timestamp is received, a hardware TX timestamp may be
807 * returned only if there is no software TX timestamp. Ignore false software
808 * timestamps, which may be made in the __sock_recv_timestamp() call when the
809 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
810 * hardware timestamp.
811 */
812 static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
813 {
814 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
815 }
816
817 static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
818 {
819 bool cycles = READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_BIND_PHC;
820 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
821 struct net_device *orig_dev;
822 ktime_t hwtstamp;
823
824 rcu_read_lock();
825 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
826 if (orig_dev) {
827 *if_index = orig_dev->ifindex;
828 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
829 } else {
830 hwtstamp = shhwtstamps->hwtstamp;
831 }
832 rcu_read_unlock();
833
834 return hwtstamp;
835 }
836
837 static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
838 int if_index)
839 {
840 struct scm_ts_pktinfo ts_pktinfo;
841 struct net_device *orig_dev;
842
843 if (!skb_mac_header_was_set(skb))
844 return;
845
846 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
847
848 if (!if_index) {
849 rcu_read_lock();
850 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
851 if (orig_dev)
852 if_index = orig_dev->ifindex;
853 rcu_read_unlock();
854 }
855 ts_pktinfo.if_index = if_index;
856
857 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
858 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
859 sizeof(ts_pktinfo), &ts_pktinfo);
860 }
861
862 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk)
863 {
864 const struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
865 u32 tsflags = READ_ONCE(sk->sk_tsflags);
866
867 if (serr->ee.ee_errno != ENOMSG ||
868 serr->ee.ee_origin != SO_EE_ORIGIN_TIMESTAMPING)
869 return false;
870
871 /* software time stamp available and wanted */
872 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE) && skb->tstamp)
873 return true;
874 /* hardware time stamps available and wanted */
875 return (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
876 skb_hwtstamps(skb)->hwtstamp;
877 }
878
879 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk,
880 struct timespec64 *ts)
881 {
882 u32 tsflags = READ_ONCE(sk->sk_tsflags);
883 ktime_t hwtstamp;
884 int if_index = 0;
885
886 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
887 ktime_to_timespec64_cond(skb->tstamp, ts))
888 return SOF_TIMESTAMPING_TX_SOFTWARE;
889
890 if (!(tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) ||
891 skb_is_swtx_tstamp(skb, false))
892 return -ENOENT;
893
894 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
895 hwtstamp = get_timestamp(sk, skb, &if_index);
896 else
897 hwtstamp = skb_hwtstamps(skb)->hwtstamp;
898
899 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
900 hwtstamp = ptp_convert_timestamp(&hwtstamp,
901 READ_ONCE(sk->sk_bind_phc));
902 if (!ktime_to_timespec64_cond(hwtstamp, ts))
903 return -ENOENT;
904
905 return SOF_TIMESTAMPING_TX_HARDWARE;
906 }
907
908 /*
909 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
910 */
911 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
912 struct sk_buff *skb)
913 {
914 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
915 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
916 struct scm_timestamping_internal tss;
917 int empty = 1, false_tstamp = 0;
918 struct skb_shared_hwtstamps *shhwtstamps =
919 skb_hwtstamps(skb);
920 int if_index;
921 ktime_t hwtstamp;
922 u32 tsflags;
923
924 /* Race occurred between timestamp enabling and packet
925 receiving. Fill in the current time for now. */
926 if (need_software_tstamp && skb->tstamp == 0) {
927 __net_timestamp(skb);
928 false_tstamp = 1;
929 }
930
931 if (need_software_tstamp) {
932 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
933 if (new_tstamp) {
934 struct __kernel_sock_timeval tv;
935
936 skb_get_new_timestamp(skb, &tv);
937 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
938 sizeof(tv), &tv);
939 } else {
940 struct __kernel_old_timeval tv;
941
942 skb_get_timestamp(skb, &tv);
943 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
944 sizeof(tv), &tv);
945 }
946 } else {
947 if (new_tstamp) {
948 struct __kernel_timespec ts;
949
950 skb_get_new_timestampns(skb, &ts);
951 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
952 sizeof(ts), &ts);
953 } else {
954 struct __kernel_old_timespec ts;
955
956 skb_get_timestampns(skb, &ts);
957 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
958 sizeof(ts), &ts);
959 }
960 }
961 }
962
963 memset(&tss, 0, sizeof(tss));
964 tsflags = READ_ONCE(sk->sk_tsflags);
965 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE &&
966 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE ||
967 skb_is_err_queue(skb) ||
968 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
969 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
970 empty = 0;
971 if (shhwtstamps &&
972 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE &&
973 (tsflags & SOF_TIMESTAMPING_RX_HARDWARE ||
974 skb_is_err_queue(skb) ||
975 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
976 !skb_is_swtx_tstamp(skb, false_tstamp)) {
977 if_index = 0;
978 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
979 hwtstamp = get_timestamp(sk, skb, &if_index);
980 else
981 hwtstamp = shhwtstamps->hwtstamp;
982
983 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
984 hwtstamp = ptp_convert_timestamp(&hwtstamp,
985 READ_ONCE(sk->sk_bind_phc));
986
987 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
988 empty = 0;
989
990 if ((tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
991 !skb_is_err_queue(skb))
992 put_ts_pktinfo(msg, skb, if_index);
993 }
994 }
995 if (!empty) {
996 if (sock_flag(sk, SOCK_TSTAMP_NEW))
997 put_cmsg_scm_timestamping64(msg, &tss);
998 else
999 put_cmsg_scm_timestamping(msg, &tss);
1000
1001 if (skb_is_err_queue(skb) && skb->len &&
1002 SKB_EXT_ERR(skb)->opt_stats)
1003 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
1004 skb->len, skb->data);
1005 }
1006 }
1007 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
1008
1009 #ifdef CONFIG_WIRELESS
1010 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
1011 struct sk_buff *skb)
1012 {
1013 int ack;
1014
1015 if (!sock_flag(sk, SOCK_WIFI_STATUS))
1016 return;
1017 if (!skb->wifi_acked_valid)
1018 return;
1019
1020 ack = skb->wifi_acked;
1021
1022 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
1023 }
1024 EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
1025 #endif
1026
1027 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
1028 struct sk_buff *skb)
1029 {
1030 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
1031 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
1032 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
1033 }
1034
1035 static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
1036 struct sk_buff *skb)
1037 {
1038 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
1039 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
1040 __u32 mark = skb->mark;
1041
1042 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1043 }
1044 }
1045
1046 static void sock_recv_priority(struct msghdr *msg, struct sock *sk,
1047 struct sk_buff *skb)
1048 {
1049 if (sock_flag(sk, SOCK_RCVPRIORITY) && skb) {
1050 __u32 priority = skb->priority;
1051
1052 put_cmsg(msg, SOL_SOCKET, SO_PRIORITY, sizeof(__u32), &priority);
1053 }
1054 }
1055
1056 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1057 struct sk_buff *skb)
1058 {
1059 sock_recv_timestamp(msg, sk, skb);
1060 sock_recv_drops(msg, sk, skb);
1061 sock_recv_mark(msg, sk, skb);
1062 sock_recv_priority(msg, sk, skb);
1063 }
1064 EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1065
1066 INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1067 size_t, int));
1068 INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1069 size_t, int));
1070
1071 static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1072 {
1073 trace_sock_recv_length(sk, ret, flags);
1074 }
1075
1076 static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1077 int flags)
1078 {
1079 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1080 inet6_recvmsg,
1081 inet_recvmsg, sock, msg,
1082 msg_data_left(msg), flags);
1083 if (trace_sock_recv_length_enabled())
1084 call_trace_sock_recv_length(sock->sk, ret, flags);
1085 return ret;
1086 }
1087
1088 /**
1089 * sock_recvmsg - receive a message from @sock
1090 * @sock: socket
1091 * @msg: message to receive
1092 * @flags: message flags
1093 *
1094 * Receives @msg from @sock, passing through LSM. Returns the total number
1095 * of bytes received, or an error.
1096 */
1097 int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1098 {
1099 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1100
1101 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1102 }
1103 EXPORT_SYMBOL(sock_recvmsg);
1104
1105 /**
1106 * kernel_recvmsg - Receive a message from a socket (kernel space)
1107 * @sock: The socket to receive the message from
1108 * @msg: Received message
1109 * @vec: Input s/g array for message data
1110 * @num: Size of input s/g array
1111 * @size: Number of bytes to read
1112 * @flags: Message flags (MSG_DONTWAIT, etc...)
1113 *
1114 * On return the msg structure contains the scatter/gather array passed in the
1115 * vec argument. The array is modified so that it consists of the unfilled
1116 * portion of the original array.
1117 *
1118 * The returned value is the total number of bytes received, or an error.
1119 */
1120
1121 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1122 struct kvec *vec, size_t num, size_t size, int flags)
1123 {
1124 msg->msg_control_is_user = false;
1125 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1126 return sock_recvmsg(sock, msg, flags);
1127 }
1128 EXPORT_SYMBOL(kernel_recvmsg);
1129
1130 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1131 struct pipe_inode_info *pipe, size_t len,
1132 unsigned int flags)
1133 {
1134 struct socket *sock = file->private_data;
1135 const struct proto_ops *ops;
1136
1137 ops = READ_ONCE(sock->ops);
1138 if (unlikely(!ops->splice_read))
1139 return copy_splice_read(file, ppos, pipe, len, flags);
1140
1141 return ops->splice_read(sock, ppos, pipe, len, flags);
1142 }
1143
1144 static void sock_splice_eof(struct file *file)
1145 {
1146 struct socket *sock = file->private_data;
1147 const struct proto_ops *ops;
1148
1149 ops = READ_ONCE(sock->ops);
1150 if (ops->splice_eof)
1151 ops->splice_eof(sock);
1152 }
1153
1154 static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1155 {
1156 struct file *file = iocb->ki_filp;
1157 struct socket *sock = file->private_data;
1158 struct msghdr msg = {.msg_iter = *to,
1159 .msg_iocb = iocb};
1160 ssize_t res;
1161
1162 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1163 msg.msg_flags = MSG_DONTWAIT;
1164
1165 if (iocb->ki_pos != 0)
1166 return -ESPIPE;
1167
1168 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1169 return 0;
1170
1171 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1172 *to = msg.msg_iter;
1173 return res;
1174 }
1175
1176 static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1177 {
1178 struct file *file = iocb->ki_filp;
1179 struct socket *sock = file->private_data;
1180 struct msghdr msg = {.msg_iter = *from,
1181 .msg_iocb = iocb};
1182 ssize_t res;
1183
1184 if (iocb->ki_pos != 0)
1185 return -ESPIPE;
1186
1187 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1188 msg.msg_flags = MSG_DONTWAIT;
1189
1190 if (sock->type == SOCK_SEQPACKET)
1191 msg.msg_flags |= MSG_EOR;
1192
1193 if (iocb->ki_flags & IOCB_NOSIGNAL)
1194 msg.msg_flags |= MSG_NOSIGNAL;
1195
1196 res = __sock_sendmsg(sock, &msg);
1197 *from = msg.msg_iter;
1198 return res;
1199 }
1200
1201 /*
1202 * Atomic setting of ioctl hooks to avoid race
1203 * with module unload.
1204 */
1205
1206 static DEFINE_MUTEX(br_ioctl_mutex);
1207 static int (*br_ioctl_hook)(struct net *net, unsigned int cmd,
1208 void __user *uarg);
1209
1210 void brioctl_set(int (*hook)(struct net *net, unsigned int cmd,
1211 void __user *uarg))
1212 {
1213 mutex_lock(&br_ioctl_mutex);
1214 br_ioctl_hook = hook;
1215 mutex_unlock(&br_ioctl_mutex);
1216 }
1217 EXPORT_SYMBOL(brioctl_set);
1218
1219 int br_ioctl_call(struct net *net, unsigned int cmd, void __user *uarg)
1220 {
1221 int err = -ENOPKG;
1222
1223 if (!br_ioctl_hook)
1224 request_module("bridge");
1225
1226 mutex_lock(&br_ioctl_mutex);
1227 if (br_ioctl_hook)
1228 err = br_ioctl_hook(net, cmd, uarg);
1229 mutex_unlock(&br_ioctl_mutex);
1230
1231 return err;
1232 }
1233
1234 static DEFINE_MUTEX(vlan_ioctl_mutex);
1235 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1236
1237 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1238 {
1239 mutex_lock(&vlan_ioctl_mutex);
1240 vlan_ioctl_hook = hook;
1241 mutex_unlock(&vlan_ioctl_mutex);
1242 }
1243 EXPORT_SYMBOL(vlan_ioctl_set);
1244
1245 static long sock_do_ioctl(struct net *net, struct socket *sock,
1246 unsigned int cmd, unsigned long arg)
1247 {
1248 const struct proto_ops *ops = READ_ONCE(sock->ops);
1249 struct ifreq ifr;
1250 bool need_copyout;
1251 int err;
1252 void __user *argp = (void __user *)arg;
1253 void __user *data;
1254
1255 err = ops->ioctl(sock, cmd, arg);
1256
1257 /*
1258 * If this ioctl is unknown try to hand it down
1259 * to the NIC driver.
1260 */
1261 if (err != -ENOIOCTLCMD)
1262 return err;
1263
1264 if (!is_socket_ioctl_cmd(cmd))
1265 return -ENOTTY;
1266
1267 if (get_user_ifreq(&ifr, &data, argp))
1268 return -EFAULT;
1269 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1270 if (!err && need_copyout)
1271 if (put_user_ifreq(&ifr, argp))
1272 return -EFAULT;
1273
1274 return err;
1275 }
1276
1277 /*
1278 * With an ioctl, arg may well be a user mode pointer, but we don't know
1279 * what to do with it - that's up to the protocol still.
1280 */
1281
1282 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1283 {
1284 const struct proto_ops *ops;
1285 struct socket *sock;
1286 struct sock *sk;
1287 void __user *argp = (void __user *)arg;
1288 int pid, err;
1289 struct net *net;
1290
1291 sock = file->private_data;
1292 ops = READ_ONCE(sock->ops);
1293 sk = sock->sk;
1294 net = sock_net(sk);
1295 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1296 struct ifreq ifr;
1297 void __user *data;
1298 bool need_copyout;
1299 if (get_user_ifreq(&ifr, &data, argp))
1300 return -EFAULT;
1301 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1302 if (!err && need_copyout)
1303 if (put_user_ifreq(&ifr, argp))
1304 return -EFAULT;
1305 } else
1306 #ifdef CONFIG_WEXT_CORE
1307 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1308 err = wext_handle_ioctl(net, cmd, argp);
1309 } else
1310 #endif
1311 switch (cmd) {
1312 case FIOSETOWN:
1313 case SIOCSPGRP:
1314 err = -EFAULT;
1315 if (get_user(pid, (int __user *)argp))
1316 break;
1317 err = f_setown(sock->file, pid, 1);
1318 break;
1319 case FIOGETOWN:
1320 case SIOCGPGRP:
1321 err = put_user(f_getown(sock->file),
1322 (int __user *)argp);
1323 break;
1324 case SIOCGIFBR:
1325 case SIOCSIFBR:
1326 case SIOCBRADDBR:
1327 case SIOCBRDELBR:
1328 case SIOCBRADDIF:
1329 case SIOCBRDELIF:
1330 err = br_ioctl_call(net, cmd, argp);
1331 break;
1332 case SIOCGIFVLAN:
1333 case SIOCSIFVLAN:
1334 err = -ENOPKG;
1335 if (!vlan_ioctl_hook)
1336 request_module("8021q");
1337
1338 mutex_lock(&vlan_ioctl_mutex);
1339 if (vlan_ioctl_hook)
1340 err = vlan_ioctl_hook(net, argp);
1341 mutex_unlock(&vlan_ioctl_mutex);
1342 break;
1343 case SIOCGSKNS:
1344 err = -EPERM;
1345 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1346 break;
1347
1348 err = open_related_ns(&net->ns, get_net_ns);
1349 break;
1350 case SIOCGSTAMP_OLD:
1351 case SIOCGSTAMPNS_OLD:
1352 if (!ops->gettstamp) {
1353 err = -ENOIOCTLCMD;
1354 break;
1355 }
1356 err = ops->gettstamp(sock, argp,
1357 cmd == SIOCGSTAMP_OLD,
1358 !IS_ENABLED(CONFIG_64BIT));
1359 break;
1360 case SIOCGSTAMP_NEW:
1361 case SIOCGSTAMPNS_NEW:
1362 if (!ops->gettstamp) {
1363 err = -ENOIOCTLCMD;
1364 break;
1365 }
1366 err = ops->gettstamp(sock, argp,
1367 cmd == SIOCGSTAMP_NEW,
1368 false);
1369 break;
1370
1371 case SIOCGIFCONF:
1372 err = dev_ifconf(net, argp);
1373 break;
1374
1375 default:
1376 err = sock_do_ioctl(net, sock, cmd, arg);
1377 break;
1378 }
1379 return err;
1380 }
1381
1382 /**
1383 * sock_create_lite - creates a socket
1384 * @family: protocol family (AF_INET, ...)
1385 * @type: communication type (SOCK_STREAM, ...)
1386 * @protocol: protocol (0, ...)
1387 * @res: new socket
1388 *
1389 * Creates a new socket and assigns it to @res, passing through LSM.
1390 * The new socket initialization is not complete, see kernel_accept().
1391 * Returns 0 or an error. On failure @res is set to %NULL.
1392 * This function internally uses GFP_KERNEL.
1393 */
1394
1395 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1396 {
1397 int err;
1398 struct socket *sock = NULL;
1399
1400 err = security_socket_create(family, type, protocol, 1);
1401 if (err)
1402 goto out;
1403
1404 sock = sock_alloc();
1405 if (!sock) {
1406 err = -ENOMEM;
1407 goto out;
1408 }
1409
1410 sock->type = type;
1411 err = security_socket_post_create(sock, family, type, protocol, 1);
1412 if (err)
1413 goto out_release;
1414
1415 out:
1416 *res = sock;
1417 return err;
1418 out_release:
1419 sock_release(sock);
1420 sock = NULL;
1421 goto out;
1422 }
1423 EXPORT_SYMBOL(sock_create_lite);
1424
1425 /* No kernel lock held - perfect */
1426 static __poll_t sock_poll(struct file *file, poll_table *wait)
1427 {
1428 struct socket *sock = file->private_data;
1429 const struct proto_ops *ops = READ_ONCE(sock->ops);
1430 __poll_t events = poll_requested_events(wait), flag = 0;
1431
1432 if (!ops->poll)
1433 return 0;
1434
1435 if (sk_can_busy_loop(sock->sk)) {
1436 /* poll once if requested by the syscall */
1437 if (events & POLL_BUSY_LOOP)
1438 sk_busy_loop(sock->sk, 1);
1439
1440 /* if this socket can poll_ll, tell the system call */
1441 flag = POLL_BUSY_LOOP;
1442 }
1443
1444 return ops->poll(file, sock, wait) | flag;
1445 }
1446
1447 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1448 {
1449 struct socket *sock = file->private_data;
1450
1451 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1452 }
1453
1454 static int sock_close(struct inode *inode, struct file *filp)
1455 {
1456 __sock_release(SOCKET_I(inode), inode);
1457 return 0;
1458 }
1459
1460 /*
1461 * Update the socket async list
1462 *
1463 * Fasync_list locking strategy.
1464 *
1465 * 1. fasync_list is modified only under process context socket lock
1466 * i.e. under semaphore.
1467 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1468 * or under socket lock
1469 */
1470
1471 static int sock_fasync(int fd, struct file *filp, int on)
1472 {
1473 struct socket *sock = filp->private_data;
1474 struct sock *sk = sock->sk;
1475 struct socket_wq *wq = &sock->wq;
1476
1477 if (sk == NULL)
1478 return -EINVAL;
1479
1480 lock_sock(sk);
1481 fasync_helper(fd, filp, on, &wq->fasync_list);
1482
1483 if (!wq->fasync_list)
1484 sock_reset_flag(sk, SOCK_FASYNC);
1485 else
1486 sock_set_flag(sk, SOCK_FASYNC);
1487
1488 release_sock(sk);
1489 return 0;
1490 }
1491
1492 /* This function may be called only under rcu_lock */
1493
1494 int sock_wake_async(struct socket_wq *wq, int how, int band)
1495 {
1496 if (!wq || !wq->fasync_list)
1497 return -1;
1498
1499 switch (how) {
1500 case SOCK_WAKE_WAITD:
1501 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1502 break;
1503 goto call_kill;
1504 case SOCK_WAKE_SPACE:
1505 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1506 break;
1507 fallthrough;
1508 case SOCK_WAKE_IO:
1509 call_kill:
1510 kill_fasync(&wq->fasync_list, SIGIO, band);
1511 break;
1512 case SOCK_WAKE_URG:
1513 kill_fasync(&wq->fasync_list, SIGURG, band);
1514 }
1515
1516 return 0;
1517 }
1518 EXPORT_SYMBOL(sock_wake_async);
1519
1520 /**
1521 * __sock_create - creates a socket
1522 * @net: net namespace
1523 * @family: protocol family (AF_INET, ...)
1524 * @type: communication type (SOCK_STREAM, ...)
1525 * @protocol: protocol (0, ...)
1526 * @res: new socket
1527 * @kern: boolean for kernel space sockets
1528 *
1529 * Creates a new socket and assigns it to @res, passing through LSM.
1530 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1531 * be set to true if the socket resides in kernel space.
1532 * This function internally uses GFP_KERNEL.
1533 */
1534
1535 int __sock_create(struct net *net, int family, int type, int protocol,
1536 struct socket **res, int kern)
1537 {
1538 int err;
1539 struct socket *sock;
1540 const struct net_proto_family *pf;
1541
1542 /*
1543 * Check protocol is in range
1544 */
1545 if (family < 0 || family >= NPROTO)
1546 return -EAFNOSUPPORT;
1547 if (type < 0 || type >= SOCK_MAX)
1548 return -EINVAL;
1549
1550 /* Compatibility.
1551
1552 This uglymoron is moved from INET layer to here to avoid
1553 deadlock in module load.
1554 */
1555 if (family == PF_INET && type == SOCK_PACKET) {
1556 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1557 current->comm);
1558 family = PF_PACKET;
1559 }
1560
1561 err = security_socket_create(family, type, protocol, kern);
1562 if (err)
1563 return err;
1564
1565 /*
1566 * Allocate the socket and allow the family to set things up. if
1567 * the protocol is 0, the family is instructed to select an appropriate
1568 * default.
1569 */
1570 sock = sock_alloc();
1571 if (!sock) {
1572 net_warn_ratelimited("socket: no more sockets\n");
1573 return -ENFILE; /* Not exactly a match, but its the
1574 closest posix thing */
1575 }
1576
1577 sock->type = type;
1578
1579 #ifdef CONFIG_MODULES
1580 /* Attempt to load a protocol module if the find failed.
1581 *
1582 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1583 * requested real, full-featured networking support upon configuration.
1584 * Otherwise module support will break!
1585 */
1586 if (rcu_access_pointer(net_families[family]) == NULL)
1587 request_module("net-pf-%d", family);
1588 #endif
1589
1590 rcu_read_lock();
1591 pf = rcu_dereference(net_families[family]);
1592 err = -EAFNOSUPPORT;
1593 if (!pf)
1594 goto out_release;
1595
1596 /*
1597 * We will call the ->create function, that possibly is in a loadable
1598 * module, so we have to bump that loadable module refcnt first.
1599 */
1600 if (!try_module_get(pf->owner))
1601 goto out_release;
1602
1603 /* Now protected by module ref count */
1604 rcu_read_unlock();
1605
1606 err = pf->create(net, sock, protocol, kern);
1607 if (err < 0) {
1608 /* ->create should release the allocated sock->sk object on error
1609 * and make sure sock->sk is set to NULL to avoid use-after-free
1610 */
1611 DEBUG_NET_WARN_ONCE(sock->sk,
1612 "%ps must clear sock->sk on failure, family: %d, type: %d, protocol: %d\n",
1613 pf->create, family, type, protocol);
1614 goto out_module_put;
1615 }
1616
1617 /*
1618 * Now to bump the refcnt of the [loadable] module that owns this
1619 * socket at sock_release time we decrement its refcnt.
1620 */
1621 if (!try_module_get(sock->ops->owner))
1622 goto out_module_busy;
1623
1624 /*
1625 * Now that we're done with the ->create function, the [loadable]
1626 * module can have its refcnt decremented
1627 */
1628 module_put(pf->owner);
1629 err = security_socket_post_create(sock, family, type, protocol, kern);
1630 if (err)
1631 goto out_sock_release;
1632 *res = sock;
1633
1634 return 0;
1635
1636 out_module_busy:
1637 err = -EAFNOSUPPORT;
1638 out_module_put:
1639 sock->ops = NULL;
1640 module_put(pf->owner);
1641 out_sock_release:
1642 sock_release(sock);
1643 return err;
1644
1645 out_release:
1646 rcu_read_unlock();
1647 goto out_sock_release;
1648 }
1649 EXPORT_SYMBOL(__sock_create);
1650
1651 /**
1652 * sock_create - creates a socket
1653 * @family: protocol family (AF_INET, ...)
1654 * @type: communication type (SOCK_STREAM, ...)
1655 * @protocol: protocol (0, ...)
1656 * @res: new socket
1657 *
1658 * A wrapper around __sock_create().
1659 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1660 */
1661
1662 int sock_create(int family, int type, int protocol, struct socket **res)
1663 {
1664 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1665 }
1666 EXPORT_SYMBOL(sock_create);
1667
1668 /**
1669 * sock_create_kern - creates a socket (kernel space)
1670 * @net: net namespace
1671 * @family: protocol family (AF_INET, ...)
1672 * @type: communication type (SOCK_STREAM, ...)
1673 * @protocol: protocol (0, ...)
1674 * @res: new socket
1675 *
1676 * A wrapper around __sock_create().
1677 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1678 */
1679
1680 int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1681 {
1682 return __sock_create(net, family, type, protocol, res, 1);
1683 }
1684 EXPORT_SYMBOL(sock_create_kern);
1685
1686 static struct socket *__sys_socket_create(int family, int type, int protocol)
1687 {
1688 struct socket *sock;
1689 int retval;
1690
1691 /* Check the SOCK_* constants for consistency. */
1692 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1693 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1694 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1695 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1696
1697 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1698 return ERR_PTR(-EINVAL);
1699 type &= SOCK_TYPE_MASK;
1700
1701 retval = sock_create(family, type, protocol, &sock);
1702 if (retval < 0)
1703 return ERR_PTR(retval);
1704
1705 return sock;
1706 }
1707
1708 struct file *__sys_socket_file(int family, int type, int protocol)
1709 {
1710 struct socket *sock;
1711 int flags;
1712
1713 sock = __sys_socket_create(family, type, protocol);
1714 if (IS_ERR(sock))
1715 return ERR_CAST(sock);
1716
1717 flags = type & ~SOCK_TYPE_MASK;
1718 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1719 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1720
1721 return sock_alloc_file(sock, flags, NULL);
1722 }
1723
1724 /* A hook for bpf progs to attach to and update socket protocol.
1725 *
1726 * A static noinline declaration here could cause the compiler to
1727 * optimize away the function. A global noinline declaration will
1728 * keep the definition, but may optimize away the callsite.
1729 * Therefore, __weak is needed to ensure that the call is still
1730 * emitted, by telling the compiler that we don't know what the
1731 * function might eventually be.
1732 */
1733
1734 __bpf_hook_start();
1735
1736 __weak noinline int update_socket_protocol(int family, int type, int protocol)
1737 {
1738 return protocol;
1739 }
1740
1741 __bpf_hook_end();
1742
1743 int __sys_socket(int family, int type, int protocol)
1744 {
1745 struct socket *sock;
1746 int flags;
1747
1748 sock = __sys_socket_create(family, type,
1749 update_socket_protocol(family, type, protocol));
1750 if (IS_ERR(sock))
1751 return PTR_ERR(sock);
1752
1753 flags = type & ~SOCK_TYPE_MASK;
1754 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1755 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1756
1757 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1758 }
1759
1760 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1761 {
1762 return __sys_socket(family, type, protocol);
1763 }
1764
1765 /*
1766 * Create a pair of connected sockets.
1767 */
1768
1769 int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1770 {
1771 struct socket *sock1, *sock2;
1772 int fd1, fd2, err;
1773 struct file *newfile1, *newfile2;
1774 int flags;
1775
1776 flags = type & ~SOCK_TYPE_MASK;
1777 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1778 return -EINVAL;
1779 type &= SOCK_TYPE_MASK;
1780
1781 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1782 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1783
1784 /*
1785 * reserve descriptors and make sure we won't fail
1786 * to return them to userland.
1787 */
1788 fd1 = get_unused_fd_flags(flags);
1789 if (unlikely(fd1 < 0))
1790 return fd1;
1791
1792 fd2 = get_unused_fd_flags(flags);
1793 if (unlikely(fd2 < 0)) {
1794 put_unused_fd(fd1);
1795 return fd2;
1796 }
1797
1798 err = put_user(fd1, &usockvec[0]);
1799 if (err)
1800 goto out;
1801
1802 err = put_user(fd2, &usockvec[1]);
1803 if (err)
1804 goto out;
1805
1806 /*
1807 * Obtain the first socket and check if the underlying protocol
1808 * supports the socketpair call.
1809 */
1810
1811 err = sock_create(family, type, protocol, &sock1);
1812 if (unlikely(err < 0))
1813 goto out;
1814
1815 err = sock_create(family, type, protocol, &sock2);
1816 if (unlikely(err < 0)) {
1817 sock_release(sock1);
1818 goto out;
1819 }
1820
1821 err = security_socket_socketpair(sock1, sock2);
1822 if (unlikely(err)) {
1823 sock_release(sock2);
1824 sock_release(sock1);
1825 goto out;
1826 }
1827
1828 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1829 if (unlikely(err < 0)) {
1830 sock_release(sock2);
1831 sock_release(sock1);
1832 goto out;
1833 }
1834
1835 newfile1 = sock_alloc_file(sock1, flags, NULL);
1836 if (IS_ERR(newfile1)) {
1837 err = PTR_ERR(newfile1);
1838 sock_release(sock2);
1839 goto out;
1840 }
1841
1842 newfile2 = sock_alloc_file(sock2, flags, NULL);
1843 if (IS_ERR(newfile2)) {
1844 err = PTR_ERR(newfile2);
1845 fput(newfile1);
1846 goto out;
1847 }
1848
1849 audit_fd_pair(fd1, fd2);
1850
1851 fd_install(fd1, newfile1);
1852 fd_install(fd2, newfile2);
1853 return 0;
1854
1855 out:
1856 put_unused_fd(fd2);
1857 put_unused_fd(fd1);
1858 return err;
1859 }
1860
1861 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1862 int __user *, usockvec)
1863 {
1864 return __sys_socketpair(family, type, protocol, usockvec);
1865 }
1866
1867 int __sys_bind_socket(struct socket *sock, struct sockaddr_storage *address,
1868 int addrlen)
1869 {
1870 int err;
1871
1872 err = security_socket_bind(sock, (struct sockaddr *)address,
1873 addrlen);
1874 if (!err)
1875 err = READ_ONCE(sock->ops)->bind(sock,
1876 (struct sockaddr *)address,
1877 addrlen);
1878 return err;
1879 }
1880
1881 /*
1882 * Bind a name to a socket. Nothing much to do here since it's
1883 * the protocol's responsibility to handle the local address.
1884 *
1885 * We move the socket address to kernel space before we call
1886 * the protocol layer (having also checked the address is ok).
1887 */
1888
1889 int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1890 {
1891 struct socket *sock;
1892 struct sockaddr_storage address;
1893 CLASS(fd, f)(fd);
1894 int err;
1895
1896 if (fd_empty(f))
1897 return -EBADF;
1898 sock = sock_from_file(fd_file(f));
1899 if (unlikely(!sock))
1900 return -ENOTSOCK;
1901
1902 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1903 if (unlikely(err))
1904 return err;
1905
1906 return __sys_bind_socket(sock, &address, addrlen);
1907 }
1908
1909 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1910 {
1911 return __sys_bind(fd, umyaddr, addrlen);
1912 }
1913
1914 /*
1915 * Perform a listen. Basically, we allow the protocol to do anything
1916 * necessary for a listen, and if that works, we mark the socket as
1917 * ready for listening.
1918 */
1919 int __sys_listen_socket(struct socket *sock, int backlog)
1920 {
1921 int somaxconn, err;
1922
1923 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1924 if ((unsigned int)backlog > somaxconn)
1925 backlog = somaxconn;
1926
1927 err = security_socket_listen(sock, backlog);
1928 if (!err)
1929 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1930 return err;
1931 }
1932
1933 int __sys_listen(int fd, int backlog)
1934 {
1935 CLASS(fd, f)(fd);
1936 struct socket *sock;
1937
1938 if (fd_empty(f))
1939 return -EBADF;
1940 sock = sock_from_file(fd_file(f));
1941 if (unlikely(!sock))
1942 return -ENOTSOCK;
1943
1944 return __sys_listen_socket(sock, backlog);
1945 }
1946
1947 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1948 {
1949 return __sys_listen(fd, backlog);
1950 }
1951
1952 struct file *do_accept(struct file *file, struct proto_accept_arg *arg,
1953 struct sockaddr __user *upeer_sockaddr,
1954 int __user *upeer_addrlen, int flags)
1955 {
1956 struct socket *sock, *newsock;
1957 struct file *newfile;
1958 int err, len;
1959 struct sockaddr_storage address;
1960 const struct proto_ops *ops;
1961
1962 sock = sock_from_file(file);
1963 if (!sock)
1964 return ERR_PTR(-ENOTSOCK);
1965
1966 newsock = sock_alloc();
1967 if (!newsock)
1968 return ERR_PTR(-ENFILE);
1969 ops = READ_ONCE(sock->ops);
1970
1971 newsock->type = sock->type;
1972 newsock->ops = ops;
1973
1974 /*
1975 * We don't need try_module_get here, as the listening socket (sock)
1976 * has the protocol module (sock->ops->owner) held.
1977 */
1978 __module_get(ops->owner);
1979
1980 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1981 if (IS_ERR(newfile))
1982 return newfile;
1983
1984 err = security_socket_accept(sock, newsock);
1985 if (err)
1986 goto out_fd;
1987
1988 arg->flags |= sock->file->f_flags;
1989 err = ops->accept(sock, newsock, arg);
1990 if (err < 0)
1991 goto out_fd;
1992
1993 if (upeer_sockaddr) {
1994 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1995 if (len < 0) {
1996 err = -ECONNABORTED;
1997 goto out_fd;
1998 }
1999 err = move_addr_to_user(&address,
2000 len, upeer_sockaddr, upeer_addrlen);
2001 if (err < 0)
2002 goto out_fd;
2003 }
2004
2005 /* File flags are not inherited via accept() unlike another OSes. */
2006 return newfile;
2007 out_fd:
2008 fput(newfile);
2009 return ERR_PTR(err);
2010 }
2011
2012 static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
2013 int __user *upeer_addrlen, int flags)
2014 {
2015 struct proto_accept_arg arg = { };
2016 struct file *newfile;
2017 int newfd;
2018
2019 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
2020 return -EINVAL;
2021
2022 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
2023 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
2024
2025 newfd = get_unused_fd_flags(flags);
2026 if (unlikely(newfd < 0))
2027 return newfd;
2028
2029 newfile = do_accept(file, &arg, upeer_sockaddr, upeer_addrlen,
2030 flags);
2031 if (IS_ERR(newfile)) {
2032 put_unused_fd(newfd);
2033 return PTR_ERR(newfile);
2034 }
2035 fd_install(newfd, newfile);
2036 return newfd;
2037 }
2038
2039 /*
2040 * For accept, we attempt to create a new socket, set up the link
2041 * with the client, wake up the client, then return the new
2042 * connected fd. We collect the address of the connector in kernel
2043 * space and move it to user at the very end. This is unclean because
2044 * we open the socket then return an error.
2045 *
2046 * 1003.1g adds the ability to recvmsg() to query connection pending
2047 * status to recvmsg. We need to add that support in a way thats
2048 * clean when we restructure accept also.
2049 */
2050
2051 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
2052 int __user *upeer_addrlen, int flags)
2053 {
2054 CLASS(fd, f)(fd);
2055
2056 if (fd_empty(f))
2057 return -EBADF;
2058 return __sys_accept4_file(fd_file(f), upeer_sockaddr,
2059 upeer_addrlen, flags);
2060 }
2061
2062 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
2063 int __user *, upeer_addrlen, int, flags)
2064 {
2065 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
2066 }
2067
2068 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
2069 int __user *, upeer_addrlen)
2070 {
2071 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
2072 }
2073
2074 /*
2075 * Attempt to connect to a socket with the server address. The address
2076 * is in user space so we verify it is OK and move it to kernel space.
2077 *
2078 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
2079 * break bindings
2080 *
2081 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
2082 * other SEQPACKET protocols that take time to connect() as it doesn't
2083 * include the -EINPROGRESS status for such sockets.
2084 */
2085
2086 int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
2087 int addrlen, int file_flags)
2088 {
2089 struct socket *sock;
2090 int err;
2091
2092 sock = sock_from_file(file);
2093 if (!sock) {
2094 err = -ENOTSOCK;
2095 goto out;
2096 }
2097
2098 err =
2099 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2100 if (err)
2101 goto out;
2102
2103 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2104 addrlen, sock->file->f_flags | file_flags);
2105 out:
2106 return err;
2107 }
2108
2109 int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2110 {
2111 struct sockaddr_storage address;
2112 CLASS(fd, f)(fd);
2113 int ret;
2114
2115 if (fd_empty(f))
2116 return -EBADF;
2117
2118 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2119 if (ret)
2120 return ret;
2121
2122 return __sys_connect_file(fd_file(f), &address, addrlen, 0);
2123 }
2124
2125 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2126 int, addrlen)
2127 {
2128 return __sys_connect(fd, uservaddr, addrlen);
2129 }
2130
2131 /*
2132 * Get the local address ('name') of a socket object. Move the obtained
2133 * name to user space.
2134 */
2135
2136 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2137 int __user *usockaddr_len)
2138 {
2139 struct socket *sock;
2140 struct sockaddr_storage address;
2141 CLASS(fd, f)(fd);
2142 int err;
2143
2144 if (fd_empty(f))
2145 return -EBADF;
2146 sock = sock_from_file(fd_file(f));
2147 if (unlikely(!sock))
2148 return -ENOTSOCK;
2149
2150 err = security_socket_getsockname(sock);
2151 if (err)
2152 return err;
2153
2154 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2155 if (err < 0)
2156 return err;
2157
2158 /* "err" is actually length in this case */
2159 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2160 }
2161
2162 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2163 int __user *, usockaddr_len)
2164 {
2165 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2166 }
2167
2168 /*
2169 * Get the remote address ('name') of a socket object. Move the obtained
2170 * name to user space.
2171 */
2172
2173 int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2174 int __user *usockaddr_len)
2175 {
2176 struct socket *sock;
2177 struct sockaddr_storage address;
2178 CLASS(fd, f)(fd);
2179 int err;
2180
2181 if (fd_empty(f))
2182 return -EBADF;
2183 sock = sock_from_file(fd_file(f));
2184 if (unlikely(!sock))
2185 return -ENOTSOCK;
2186
2187 err = security_socket_getpeername(sock);
2188 if (err)
2189 return err;
2190
2191 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 1);
2192 if (err < 0)
2193 return err;
2194
2195 /* "err" is actually length in this case */
2196 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2197 }
2198
2199 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2200 int __user *, usockaddr_len)
2201 {
2202 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2203 }
2204
2205 /*
2206 * Send a datagram to a given address. We move the address into kernel
2207 * space and check the user space data area is readable before invoking
2208 * the protocol.
2209 */
2210 int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2211 struct sockaddr __user *addr, int addr_len)
2212 {
2213 struct socket *sock;
2214 struct sockaddr_storage address;
2215 int err;
2216 struct msghdr msg;
2217
2218 err = import_ubuf(ITER_SOURCE, buff, len, &msg.msg_iter);
2219 if (unlikely(err))
2220 return err;
2221
2222 CLASS(fd, f)(fd);
2223 if (fd_empty(f))
2224 return -EBADF;
2225 sock = sock_from_file(fd_file(f));
2226 if (unlikely(!sock))
2227 return -ENOTSOCK;
2228
2229 msg.msg_name = NULL;
2230 msg.msg_control = NULL;
2231 msg.msg_controllen = 0;
2232 msg.msg_namelen = 0;
2233 msg.msg_ubuf = NULL;
2234 if (addr) {
2235 err = move_addr_to_kernel(addr, addr_len, &address);
2236 if (err < 0)
2237 return err;
2238 msg.msg_name = (struct sockaddr *)&address;
2239 msg.msg_namelen = addr_len;
2240 }
2241 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2242 if (sock->file->f_flags & O_NONBLOCK)
2243 flags |= MSG_DONTWAIT;
2244 msg.msg_flags = flags;
2245 return __sock_sendmsg(sock, &msg);
2246 }
2247
2248 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2249 unsigned int, flags, struct sockaddr __user *, addr,
2250 int, addr_len)
2251 {
2252 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2253 }
2254
2255 /*
2256 * Send a datagram down a socket.
2257 */
2258
2259 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2260 unsigned int, flags)
2261 {
2262 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2263 }
2264
2265 /*
2266 * Receive a frame from the socket and optionally record the address of the
2267 * sender. We verify the buffers are writable and if needed move the
2268 * sender address from kernel to user space.
2269 */
2270 int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2271 struct sockaddr __user *addr, int __user *addr_len)
2272 {
2273 struct sockaddr_storage address;
2274 struct msghdr msg = {
2275 /* Save some cycles and don't copy the address if not needed */
2276 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2277 };
2278 struct socket *sock;
2279 int err, err2;
2280
2281 err = import_ubuf(ITER_DEST, ubuf, size, &msg.msg_iter);
2282 if (unlikely(err))
2283 return err;
2284
2285 CLASS(fd, f)(fd);
2286
2287 if (fd_empty(f))
2288 return -EBADF;
2289 sock = sock_from_file(fd_file(f));
2290 if (unlikely(!sock))
2291 return -ENOTSOCK;
2292
2293 if (sock->file->f_flags & O_NONBLOCK)
2294 flags |= MSG_DONTWAIT;
2295 err = sock_recvmsg(sock, &msg, flags);
2296
2297 if (err >= 0 && addr != NULL) {
2298 err2 = move_addr_to_user(&address,
2299 msg.msg_namelen, addr, addr_len);
2300 if (err2 < 0)
2301 err = err2;
2302 }
2303 return err;
2304 }
2305
2306 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2307 unsigned int, flags, struct sockaddr __user *, addr,
2308 int __user *, addr_len)
2309 {
2310 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2311 }
2312
2313 /*
2314 * Receive a datagram from a socket.
2315 */
2316
2317 SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2318 unsigned int, flags)
2319 {
2320 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2321 }
2322
2323 static bool sock_use_custom_sol_socket(const struct socket *sock)
2324 {
2325 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2326 }
2327
2328 int do_sock_setsockopt(struct socket *sock, bool compat, int level,
2329 int optname, sockptr_t optval, int optlen)
2330 {
2331 const struct proto_ops *ops;
2332 char *kernel_optval = NULL;
2333 int err;
2334
2335 if (optlen < 0)
2336 return -EINVAL;
2337
2338 err = security_socket_setsockopt(sock, level, optname);
2339 if (err)
2340 goto out_put;
2341
2342 if (!compat)
2343 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2344 optval, &optlen,
2345 &kernel_optval);
2346 if (err < 0)
2347 goto out_put;
2348 if (err > 0) {
2349 err = 0;
2350 goto out_put;
2351 }
2352
2353 if (kernel_optval)
2354 optval = KERNEL_SOCKPTR(kernel_optval);
2355 ops = READ_ONCE(sock->ops);
2356 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2357 err = sock_setsockopt(sock, level, optname, optval, optlen);
2358 else if (unlikely(!ops->setsockopt))
2359 err = -EOPNOTSUPP;
2360 else
2361 err = ops->setsockopt(sock, level, optname, optval,
2362 optlen);
2363 kfree(kernel_optval);
2364 out_put:
2365 return err;
2366 }
2367 EXPORT_SYMBOL(do_sock_setsockopt);
2368
2369 /* Set a socket option. Because we don't know the option lengths we have
2370 * to pass the user mode parameter for the protocols to sort out.
2371 */
2372 int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2373 int optlen)
2374 {
2375 sockptr_t optval = USER_SOCKPTR(user_optval);
2376 bool compat = in_compat_syscall();
2377 struct socket *sock;
2378 CLASS(fd, f)(fd);
2379
2380 if (fd_empty(f))
2381 return -EBADF;
2382 sock = sock_from_file(fd_file(f));
2383 if (unlikely(!sock))
2384 return -ENOTSOCK;
2385
2386 return do_sock_setsockopt(sock, compat, level, optname, optval, optlen);
2387 }
2388
2389 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2390 char __user *, optval, int, optlen)
2391 {
2392 return __sys_setsockopt(fd, level, optname, optval, optlen);
2393 }
2394
2395 INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2396 int optname));
2397
2398 int do_sock_getsockopt(struct socket *sock, bool compat, int level,
2399 int optname, sockptr_t optval, sockptr_t optlen)
2400 {
2401 int max_optlen __maybe_unused = 0;
2402 const struct proto_ops *ops;
2403 int err;
2404
2405 err = security_socket_getsockopt(sock, level, optname);
2406 if (err)
2407 return err;
2408
2409 if (!compat)
2410 copy_from_sockptr(&max_optlen, optlen, sizeof(int));
2411
2412 ops = READ_ONCE(sock->ops);
2413 if (level == SOL_SOCKET) {
2414 err = sk_getsockopt(sock->sk, level, optname, optval, optlen);
2415 } else if (unlikely(!ops->getsockopt)) {
2416 err = -EOPNOTSUPP;
2417 } else {
2418 if (WARN_ONCE(optval.is_kernel || optlen.is_kernel,
2419 "Invalid argument type"))
2420 return -EOPNOTSUPP;
2421
2422 err = ops->getsockopt(sock, level, optname, optval.user,
2423 optlen.user);
2424 }
2425
2426 if (!compat)
2427 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2428 optval, optlen, max_optlen,
2429 err);
2430
2431 return err;
2432 }
2433 EXPORT_SYMBOL(do_sock_getsockopt);
2434
2435 /*
2436 * Get a socket option. Because we don't know the option lengths we have
2437 * to pass a user mode parameter for the protocols to sort out.
2438 */
2439 int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2440 int __user *optlen)
2441 {
2442 struct socket *sock;
2443 CLASS(fd, f)(fd);
2444
2445 if (fd_empty(f))
2446 return -EBADF;
2447 sock = sock_from_file(fd_file(f));
2448 if (unlikely(!sock))
2449 return -ENOTSOCK;
2450
2451 return do_sock_getsockopt(sock, in_compat_syscall(), level, optname,
2452 USER_SOCKPTR(optval), USER_SOCKPTR(optlen));
2453 }
2454
2455 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2456 char __user *, optval, int __user *, optlen)
2457 {
2458 return __sys_getsockopt(fd, level, optname, optval, optlen);
2459 }
2460
2461 /*
2462 * Shutdown a socket.
2463 */
2464
2465 int __sys_shutdown_sock(struct socket *sock, int how)
2466 {
2467 int err;
2468
2469 err = security_socket_shutdown(sock, how);
2470 if (!err)
2471 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2472
2473 return err;
2474 }
2475
2476 int __sys_shutdown(int fd, int how)
2477 {
2478 struct socket *sock;
2479 CLASS(fd, f)(fd);
2480
2481 if (fd_empty(f))
2482 return -EBADF;
2483 sock = sock_from_file(fd_file(f));
2484 if (unlikely(!sock))
2485 return -ENOTSOCK;
2486
2487 return __sys_shutdown_sock(sock, how);
2488 }
2489
2490 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2491 {
2492 return __sys_shutdown(fd, how);
2493 }
2494
2495 /* A couple of helpful macros for getting the address of the 32/64 bit
2496 * fields which are the same type (int / unsigned) on our platforms.
2497 */
2498 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2499 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2500 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2501
2502 struct used_address {
2503 struct sockaddr_storage name;
2504 unsigned int name_len;
2505 };
2506
2507 int __copy_msghdr(struct msghdr *kmsg,
2508 struct user_msghdr *msg,
2509 struct sockaddr __user **save_addr)
2510 {
2511 ssize_t err;
2512
2513 kmsg->msg_control_is_user = true;
2514 kmsg->msg_get_inq = 0;
2515 kmsg->msg_control_user = msg->msg_control;
2516 kmsg->msg_controllen = msg->msg_controllen;
2517 kmsg->msg_flags = msg->msg_flags;
2518
2519 kmsg->msg_namelen = msg->msg_namelen;
2520 if (!msg->msg_name)
2521 kmsg->msg_namelen = 0;
2522
2523 if (kmsg->msg_namelen < 0)
2524 return -EINVAL;
2525
2526 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2527 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2528
2529 if (save_addr)
2530 *save_addr = msg->msg_name;
2531
2532 if (msg->msg_name && kmsg->msg_namelen) {
2533 if (!save_addr) {
2534 err = move_addr_to_kernel(msg->msg_name,
2535 kmsg->msg_namelen,
2536 kmsg->msg_name);
2537 if (err < 0)
2538 return err;
2539 }
2540 } else {
2541 kmsg->msg_name = NULL;
2542 kmsg->msg_namelen = 0;
2543 }
2544
2545 if (msg->msg_iovlen > UIO_MAXIOV)
2546 return -EMSGSIZE;
2547
2548 kmsg->msg_iocb = NULL;
2549 kmsg->msg_ubuf = NULL;
2550 return 0;
2551 }
2552
2553 static int copy_msghdr_from_user(struct msghdr *kmsg,
2554 struct user_msghdr __user *umsg,
2555 struct sockaddr __user **save_addr,
2556 struct iovec **iov)
2557 {
2558 struct user_msghdr msg;
2559 ssize_t err;
2560
2561 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2562 return -EFAULT;
2563
2564 err = __copy_msghdr(kmsg, &msg, save_addr);
2565 if (err)
2566 return err;
2567
2568 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2569 msg.msg_iov, msg.msg_iovlen,
2570 UIO_FASTIOV, iov, &kmsg->msg_iter);
2571 return err < 0 ? err : 0;
2572 }
2573
2574 static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2575 unsigned int flags, struct used_address *used_address,
2576 unsigned int allowed_msghdr_flags)
2577 {
2578 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2579 __aligned(sizeof(__kernel_size_t));
2580 /* 20 is size of ipv6_pktinfo */
2581 unsigned char *ctl_buf = ctl;
2582 int ctl_len;
2583 ssize_t err;
2584
2585 err = -ENOBUFS;
2586
2587 if (msg_sys->msg_controllen > INT_MAX)
2588 goto out;
2589 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2590 ctl_len = msg_sys->msg_controllen;
2591 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2592 err =
2593 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2594 sizeof(ctl));
2595 if (err)
2596 goto out;
2597 ctl_buf = msg_sys->msg_control;
2598 ctl_len = msg_sys->msg_controllen;
2599 } else if (ctl_len) {
2600 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2601 CMSG_ALIGN(sizeof(struct cmsghdr)));
2602 if (ctl_len > sizeof(ctl)) {
2603 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2604 if (ctl_buf == NULL)
2605 goto out;
2606 }
2607 err = -EFAULT;
2608 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2609 goto out_freectl;
2610 msg_sys->msg_control = ctl_buf;
2611 msg_sys->msg_control_is_user = false;
2612 }
2613 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2614 msg_sys->msg_flags = flags;
2615
2616 if (sock->file->f_flags & O_NONBLOCK)
2617 msg_sys->msg_flags |= MSG_DONTWAIT;
2618 /*
2619 * If this is sendmmsg() and current destination address is same as
2620 * previously succeeded address, omit asking LSM's decision.
2621 * used_address->name_len is initialized to UINT_MAX so that the first
2622 * destination address never matches.
2623 */
2624 if (used_address && msg_sys->msg_name &&
2625 used_address->name_len == msg_sys->msg_namelen &&
2626 !memcmp(&used_address->name, msg_sys->msg_name,
2627 used_address->name_len)) {
2628 err = sock_sendmsg_nosec(sock, msg_sys);
2629 goto out_freectl;
2630 }
2631 err = __sock_sendmsg(sock, msg_sys);
2632 /*
2633 * If this is sendmmsg() and sending to current destination address was
2634 * successful, remember it.
2635 */
2636 if (used_address && err >= 0) {
2637 used_address->name_len = msg_sys->msg_namelen;
2638 if (msg_sys->msg_name)
2639 memcpy(&used_address->name, msg_sys->msg_name,
2640 used_address->name_len);
2641 }
2642
2643 out_freectl:
2644 if (ctl_buf != ctl)
2645 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2646 out:
2647 return err;
2648 }
2649
2650 static int sendmsg_copy_msghdr(struct msghdr *msg,
2651 struct user_msghdr __user *umsg, unsigned flags,
2652 struct iovec **iov)
2653 {
2654 int err;
2655
2656 if (flags & MSG_CMSG_COMPAT) {
2657 struct compat_msghdr __user *msg_compat;
2658
2659 msg_compat = (struct compat_msghdr __user *) umsg;
2660 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2661 } else {
2662 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2663 }
2664 if (err < 0)
2665 return err;
2666
2667 return 0;
2668 }
2669
2670 static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2671 struct msghdr *msg_sys, unsigned int flags,
2672 struct used_address *used_address,
2673 unsigned int allowed_msghdr_flags)
2674 {
2675 struct sockaddr_storage address;
2676 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2677 ssize_t err;
2678
2679 msg_sys->msg_name = &address;
2680
2681 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2682 if (err < 0)
2683 return err;
2684
2685 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2686 allowed_msghdr_flags);
2687 kfree(iov);
2688 return err;
2689 }
2690
2691 /*
2692 * BSD sendmsg interface
2693 */
2694 long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2695 unsigned int flags)
2696 {
2697 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2698 }
2699
2700 long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2701 bool forbid_cmsg_compat)
2702 {
2703 struct msghdr msg_sys;
2704 struct socket *sock;
2705
2706 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2707 return -EINVAL;
2708
2709 CLASS(fd, f)(fd);
2710
2711 if (fd_empty(f))
2712 return -EBADF;
2713 sock = sock_from_file(fd_file(f));
2714 if (unlikely(!sock))
2715 return -ENOTSOCK;
2716
2717 return ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2718 }
2719
2720 SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2721 {
2722 return __sys_sendmsg(fd, msg, flags, true);
2723 }
2724
2725 /*
2726 * Linux sendmmsg interface
2727 */
2728
2729 int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2730 unsigned int flags, bool forbid_cmsg_compat)
2731 {
2732 int err, datagrams;
2733 struct socket *sock;
2734 struct mmsghdr __user *entry;
2735 struct compat_mmsghdr __user *compat_entry;
2736 struct msghdr msg_sys;
2737 struct used_address used_address;
2738 unsigned int oflags = flags;
2739
2740 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2741 return -EINVAL;
2742
2743 if (vlen > UIO_MAXIOV)
2744 vlen = UIO_MAXIOV;
2745
2746 datagrams = 0;
2747
2748 CLASS(fd, f)(fd);
2749
2750 if (fd_empty(f))
2751 return -EBADF;
2752 sock = sock_from_file(fd_file(f));
2753 if (unlikely(!sock))
2754 return -ENOTSOCK;
2755
2756 used_address.name_len = UINT_MAX;
2757 entry = mmsg;
2758 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2759 err = 0;
2760 flags |= MSG_BATCH;
2761
2762 while (datagrams < vlen) {
2763 if (datagrams == vlen - 1)
2764 flags = oflags;
2765
2766 if (MSG_CMSG_COMPAT & flags) {
2767 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2768 &msg_sys, flags, &used_address, MSG_EOR);
2769 if (err < 0)
2770 break;
2771 err = __put_user(err, &compat_entry->msg_len);
2772 ++compat_entry;
2773 } else {
2774 err = ___sys_sendmsg(sock,
2775 (struct user_msghdr __user *)entry,
2776 &msg_sys, flags, &used_address, MSG_EOR);
2777 if (err < 0)
2778 break;
2779 err = put_user(err, &entry->msg_len);
2780 ++entry;
2781 }
2782
2783 if (err)
2784 break;
2785 ++datagrams;
2786 if (msg_data_left(&msg_sys))
2787 break;
2788 cond_resched();
2789 }
2790
2791 /* We only return an error if no datagrams were able to be sent */
2792 if (datagrams != 0)
2793 return datagrams;
2794
2795 return err;
2796 }
2797
2798 SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2799 unsigned int, vlen, unsigned int, flags)
2800 {
2801 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2802 }
2803
2804 static int recvmsg_copy_msghdr(struct msghdr *msg,
2805 struct user_msghdr __user *umsg, unsigned flags,
2806 struct sockaddr __user **uaddr,
2807 struct iovec **iov)
2808 {
2809 ssize_t err;
2810
2811 if (MSG_CMSG_COMPAT & flags) {
2812 struct compat_msghdr __user *msg_compat;
2813
2814 msg_compat = (struct compat_msghdr __user *) umsg;
2815 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2816 } else {
2817 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2818 }
2819 if (err < 0)
2820 return err;
2821
2822 return 0;
2823 }
2824
2825 static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2826 struct user_msghdr __user *msg,
2827 struct sockaddr __user *uaddr,
2828 unsigned int flags, int nosec)
2829 {
2830 struct compat_msghdr __user *msg_compat =
2831 (struct compat_msghdr __user *) msg;
2832 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2833 struct sockaddr_storage addr;
2834 unsigned long cmsg_ptr;
2835 int len;
2836 ssize_t err;
2837
2838 msg_sys->msg_name = &addr;
2839 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2840 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2841
2842 /* We assume all kernel code knows the size of sockaddr_storage */
2843 msg_sys->msg_namelen = 0;
2844
2845 if (sock->file->f_flags & O_NONBLOCK)
2846 flags |= MSG_DONTWAIT;
2847
2848 if (unlikely(nosec))
2849 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2850 else
2851 err = sock_recvmsg(sock, msg_sys, flags);
2852
2853 if (err < 0)
2854 goto out;
2855 len = err;
2856
2857 if (uaddr != NULL) {
2858 err = move_addr_to_user(&addr,
2859 msg_sys->msg_namelen, uaddr,
2860 uaddr_len);
2861 if (err < 0)
2862 goto out;
2863 }
2864 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2865 COMPAT_FLAGS(msg));
2866 if (err)
2867 goto out;
2868 if (MSG_CMSG_COMPAT & flags)
2869 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2870 &msg_compat->msg_controllen);
2871 else
2872 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2873 &msg->msg_controllen);
2874 if (err)
2875 goto out;
2876 err = len;
2877 out:
2878 return err;
2879 }
2880
2881 static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2882 struct msghdr *msg_sys, unsigned int flags, int nosec)
2883 {
2884 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2885 /* user mode address pointers */
2886 struct sockaddr __user *uaddr;
2887 ssize_t err;
2888
2889 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2890 if (err < 0)
2891 return err;
2892
2893 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2894 kfree(iov);
2895 return err;
2896 }
2897
2898 /*
2899 * BSD recvmsg interface
2900 */
2901
2902 long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2903 struct user_msghdr __user *umsg,
2904 struct sockaddr __user *uaddr, unsigned int flags)
2905 {
2906 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2907 }
2908
2909 long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2910 bool forbid_cmsg_compat)
2911 {
2912 struct msghdr msg_sys;
2913 struct socket *sock;
2914
2915 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2916 return -EINVAL;
2917
2918 CLASS(fd, f)(fd);
2919
2920 if (fd_empty(f))
2921 return -EBADF;
2922 sock = sock_from_file(fd_file(f));
2923 if (unlikely(!sock))
2924 return -ENOTSOCK;
2925
2926 return ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2927 }
2928
2929 SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2930 unsigned int, flags)
2931 {
2932 return __sys_recvmsg(fd, msg, flags, true);
2933 }
2934
2935 /*
2936 * Linux recvmmsg interface
2937 */
2938
2939 static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2940 unsigned int vlen, unsigned int flags,
2941 struct timespec64 *timeout)
2942 {
2943 int err = 0, datagrams;
2944 struct socket *sock;
2945 struct mmsghdr __user *entry;
2946 struct compat_mmsghdr __user *compat_entry;
2947 struct msghdr msg_sys;
2948 struct timespec64 end_time;
2949 struct timespec64 timeout64;
2950
2951 if (timeout &&
2952 poll_select_set_timeout(&end_time, timeout->tv_sec,
2953 timeout->tv_nsec))
2954 return -EINVAL;
2955
2956 datagrams = 0;
2957
2958 CLASS(fd, f)(fd);
2959
2960 if (fd_empty(f))
2961 return -EBADF;
2962 sock = sock_from_file(fd_file(f));
2963 if (unlikely(!sock))
2964 return -ENOTSOCK;
2965
2966 if (likely(!(flags & MSG_ERRQUEUE))) {
2967 err = sock_error(sock->sk);
2968 if (err)
2969 return err;
2970 }
2971
2972 entry = mmsg;
2973 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2974
2975 while (datagrams < vlen) {
2976 /*
2977 * No need to ask LSM for more than the first datagram.
2978 */
2979 if (MSG_CMSG_COMPAT & flags) {
2980 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2981 &msg_sys, flags & ~MSG_WAITFORONE,
2982 datagrams);
2983 if (err < 0)
2984 break;
2985 err = __put_user(err, &compat_entry->msg_len);
2986 ++compat_entry;
2987 } else {
2988 err = ___sys_recvmsg(sock,
2989 (struct user_msghdr __user *)entry,
2990 &msg_sys, flags & ~MSG_WAITFORONE,
2991 datagrams);
2992 if (err < 0)
2993 break;
2994 err = put_user(err, &entry->msg_len);
2995 ++entry;
2996 }
2997
2998 if (err)
2999 break;
3000 ++datagrams;
3001
3002 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
3003 if (flags & MSG_WAITFORONE)
3004 flags |= MSG_DONTWAIT;
3005
3006 if (timeout) {
3007 ktime_get_ts64(&timeout64);
3008 *timeout = timespec64_sub(end_time, timeout64);
3009 if (timeout->tv_sec < 0) {
3010 timeout->tv_sec = timeout->tv_nsec = 0;
3011 break;
3012 }
3013
3014 /* Timeout, return less than vlen datagrams */
3015 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
3016 break;
3017 }
3018
3019 /* Out of band data, return right away */
3020 if (msg_sys.msg_flags & MSG_OOB)
3021 break;
3022 cond_resched();
3023 }
3024
3025 if (err == 0)
3026 return datagrams;
3027
3028 if (datagrams == 0)
3029 return err;
3030
3031 /*
3032 * We may return less entries than requested (vlen) if the
3033 * sock is non block and there aren't enough datagrams...
3034 */
3035 if (err != -EAGAIN) {
3036 /*
3037 * ... or if recvmsg returns an error after we
3038 * received some datagrams, where we record the
3039 * error to return on the next call or if the
3040 * app asks about it using getsockopt(SO_ERROR).
3041 */
3042 WRITE_ONCE(sock->sk->sk_err, -err);
3043 }
3044 return datagrams;
3045 }
3046
3047 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
3048 unsigned int vlen, unsigned int flags,
3049 struct __kernel_timespec __user *timeout,
3050 struct old_timespec32 __user *timeout32)
3051 {
3052 int datagrams;
3053 struct timespec64 timeout_sys;
3054
3055 if (timeout && get_timespec64(&timeout_sys, timeout))
3056 return -EFAULT;
3057
3058 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
3059 return -EFAULT;
3060
3061 if (!timeout && !timeout32)
3062 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
3063
3064 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
3065
3066 if (datagrams <= 0)
3067 return datagrams;
3068
3069 if (timeout && put_timespec64(&timeout_sys, timeout))
3070 datagrams = -EFAULT;
3071
3072 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
3073 datagrams = -EFAULT;
3074
3075 return datagrams;
3076 }
3077
3078 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
3079 unsigned int, vlen, unsigned int, flags,
3080 struct __kernel_timespec __user *, timeout)
3081 {
3082 if (flags & MSG_CMSG_COMPAT)
3083 return -EINVAL;
3084
3085 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
3086 }
3087
3088 #ifdef CONFIG_COMPAT_32BIT_TIME
3089 SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
3090 unsigned int, vlen, unsigned int, flags,
3091 struct old_timespec32 __user *, timeout)
3092 {
3093 if (flags & MSG_CMSG_COMPAT)
3094 return -EINVAL;
3095
3096 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
3097 }
3098 #endif
3099
3100 #ifdef __ARCH_WANT_SYS_SOCKETCALL
3101 /* Argument list sizes for sys_socketcall */
3102 #define AL(x) ((x) * sizeof(unsigned long))
3103 static const unsigned char nargs[21] = {
3104 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
3105 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
3106 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
3107 AL(4), AL(5), AL(4)
3108 };
3109
3110 #undef AL
3111
3112 /*
3113 * System call vectors.
3114 *
3115 * Argument checking cleaned up. Saved 20% in size.
3116 * This function doesn't need to set the kernel lock because
3117 * it is set by the callees.
3118 */
3119
3120 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3121 {
3122 unsigned long a[AUDITSC_ARGS];
3123 unsigned long a0, a1;
3124 int err;
3125 unsigned int len;
3126
3127 if (call < 1 || call > SYS_SENDMMSG)
3128 return -EINVAL;
3129 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3130
3131 len = nargs[call];
3132 if (len > sizeof(a))
3133 return -EINVAL;
3134
3135 /* copy_from_user should be SMP safe. */
3136 if (copy_from_user(a, args, len))
3137 return -EFAULT;
3138
3139 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3140 if (err)
3141 return err;
3142
3143 a0 = a[0];
3144 a1 = a[1];
3145
3146 switch (call) {
3147 case SYS_SOCKET:
3148 err = __sys_socket(a0, a1, a[2]);
3149 break;
3150 case SYS_BIND:
3151 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3152 break;
3153 case SYS_CONNECT:
3154 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3155 break;
3156 case SYS_LISTEN:
3157 err = __sys_listen(a0, a1);
3158 break;
3159 case SYS_ACCEPT:
3160 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3161 (int __user *)a[2], 0);
3162 break;
3163 case SYS_GETSOCKNAME:
3164 err =
3165 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3166 (int __user *)a[2]);
3167 break;
3168 case SYS_GETPEERNAME:
3169 err =
3170 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3171 (int __user *)a[2]);
3172 break;
3173 case SYS_SOCKETPAIR:
3174 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3175 break;
3176 case SYS_SEND:
3177 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3178 NULL, 0);
3179 break;
3180 case SYS_SENDTO:
3181 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3182 (struct sockaddr __user *)a[4], a[5]);
3183 break;
3184 case SYS_RECV:
3185 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3186 NULL, NULL);
3187 break;
3188 case SYS_RECVFROM:
3189 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3190 (struct sockaddr __user *)a[4],
3191 (int __user *)a[5]);
3192 break;
3193 case SYS_SHUTDOWN:
3194 err = __sys_shutdown(a0, a1);
3195 break;
3196 case SYS_SETSOCKOPT:
3197 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3198 a[4]);
3199 break;
3200 case SYS_GETSOCKOPT:
3201 err =
3202 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3203 (int __user *)a[4]);
3204 break;
3205 case SYS_SENDMSG:
3206 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3207 a[2], true);
3208 break;
3209 case SYS_SENDMMSG:
3210 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3211 a[3], true);
3212 break;
3213 case SYS_RECVMSG:
3214 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3215 a[2], true);
3216 break;
3217 case SYS_RECVMMSG:
3218 if (IS_ENABLED(CONFIG_64BIT))
3219 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3220 a[2], a[3],
3221 (struct __kernel_timespec __user *)a[4],
3222 NULL);
3223 else
3224 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3225 a[2], a[3], NULL,
3226 (struct old_timespec32 __user *)a[4]);
3227 break;
3228 case SYS_ACCEPT4:
3229 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3230 (int __user *)a[2], a[3]);
3231 break;
3232 default:
3233 err = -EINVAL;
3234 break;
3235 }
3236 return err;
3237 }
3238
3239 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
3240
3241 /**
3242 * sock_register - add a socket protocol handler
3243 * @ops: description of protocol
3244 *
3245 * This function is called by a protocol handler that wants to
3246 * advertise its address family, and have it linked into the
3247 * socket interface. The value ops->family corresponds to the
3248 * socket system call protocol family.
3249 */
3250 int sock_register(const struct net_proto_family *ops)
3251 {
3252 int err;
3253
3254 if (ops->family >= NPROTO) {
3255 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3256 return -ENOBUFS;
3257 }
3258
3259 spin_lock(&net_family_lock);
3260 if (rcu_dereference_protected(net_families[ops->family],
3261 lockdep_is_held(&net_family_lock)))
3262 err = -EEXIST;
3263 else {
3264 rcu_assign_pointer(net_families[ops->family], ops);
3265 err = 0;
3266 }
3267 spin_unlock(&net_family_lock);
3268
3269 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3270 return err;
3271 }
3272 EXPORT_SYMBOL(sock_register);
3273
3274 /**
3275 * sock_unregister - remove a protocol handler
3276 * @family: protocol family to remove
3277 *
3278 * This function is called by a protocol handler that wants to
3279 * remove its address family, and have it unlinked from the
3280 * new socket creation.
3281 *
3282 * If protocol handler is a module, then it can use module reference
3283 * counts to protect against new references. If protocol handler is not
3284 * a module then it needs to provide its own protection in
3285 * the ops->create routine.
3286 */
3287 void sock_unregister(int family)
3288 {
3289 BUG_ON(family < 0 || family >= NPROTO);
3290
3291 spin_lock(&net_family_lock);
3292 RCU_INIT_POINTER(net_families[family], NULL);
3293 spin_unlock(&net_family_lock);
3294
3295 synchronize_rcu();
3296
3297 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3298 }
3299 EXPORT_SYMBOL(sock_unregister);
3300
3301 bool sock_is_registered(int family)
3302 {
3303 return family < NPROTO && rcu_access_pointer(net_families[family]);
3304 }
3305
3306 static int __init sock_init(void)
3307 {
3308 int err;
3309 /*
3310 * Initialize the network sysctl infrastructure.
3311 */
3312 err = net_sysctl_init();
3313 if (err)
3314 goto out;
3315
3316 /*
3317 * Initialize skbuff SLAB cache
3318 */
3319 skb_init();
3320
3321 /*
3322 * Initialize the protocols module.
3323 */
3324
3325 init_inodecache();
3326
3327 err = register_filesystem(&sock_fs_type);
3328 if (err)
3329 goto out;
3330 sock_mnt = kern_mount(&sock_fs_type);
3331 if (IS_ERR(sock_mnt)) {
3332 err = PTR_ERR(sock_mnt);
3333 goto out_mount;
3334 }
3335
3336 /* The real protocol initialization is performed in later initcalls.
3337 */
3338
3339 #ifdef CONFIG_NETFILTER
3340 err = netfilter_init();
3341 if (err)
3342 goto out;
3343 #endif
3344
3345 ptp_classifier_init();
3346
3347 out:
3348 return err;
3349
3350 out_mount:
3351 unregister_filesystem(&sock_fs_type);
3352 goto out;
3353 }
3354
3355 core_initcall(sock_init); /* early initcall */
3356
3357 #ifdef CONFIG_PROC_FS
3358 void socket_seq_show(struct seq_file *seq)
3359 {
3360 seq_printf(seq, "sockets: used %d\n",
3361 sock_inuse_get(seq->private));
3362 }
3363 #endif /* CONFIG_PROC_FS */
3364
3365 /* Handle the fact that while struct ifreq has the same *layout* on
3366 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3367 * which are handled elsewhere, it still has different *size* due to
3368 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3369 * resulting in struct ifreq being 32 and 40 bytes respectively).
3370 * As a result, if the struct happens to be at the end of a page and
3371 * the next page isn't readable/writable, we get a fault. To prevent
3372 * that, copy back and forth to the full size.
3373 */
3374 int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3375 {
3376 if (in_compat_syscall()) {
3377 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3378
3379 memset(ifr, 0, sizeof(*ifr));
3380 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3381 return -EFAULT;
3382
3383 if (ifrdata)
3384 *ifrdata = compat_ptr(ifr32->ifr_data);
3385
3386 return 0;
3387 }
3388
3389 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3390 return -EFAULT;
3391
3392 if (ifrdata)
3393 *ifrdata = ifr->ifr_data;
3394
3395 return 0;
3396 }
3397 EXPORT_SYMBOL(get_user_ifreq);
3398
3399 int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3400 {
3401 size_t size = sizeof(*ifr);
3402
3403 if (in_compat_syscall())
3404 size = sizeof(struct compat_ifreq);
3405
3406 if (copy_to_user(arg, ifr, size))
3407 return -EFAULT;
3408
3409 return 0;
3410 }
3411 EXPORT_SYMBOL(put_user_ifreq);
3412
3413 #ifdef CONFIG_COMPAT
3414 static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3415 {
3416 compat_uptr_t uptr32;
3417 struct ifreq ifr;
3418 void __user *saved;
3419 int err;
3420
3421 if (get_user_ifreq(&ifr, NULL, uifr32))
3422 return -EFAULT;
3423
3424 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3425 return -EFAULT;
3426
3427 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3428 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3429
3430 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3431 if (!err) {
3432 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3433 if (put_user_ifreq(&ifr, uifr32))
3434 err = -EFAULT;
3435 }
3436 return err;
3437 }
3438
3439 /* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3440 static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3441 struct compat_ifreq __user *u_ifreq32)
3442 {
3443 struct ifreq ifreq;
3444 void __user *data;
3445
3446 if (!is_socket_ioctl_cmd(cmd))
3447 return -ENOTTY;
3448 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3449 return -EFAULT;
3450 ifreq.ifr_data = data;
3451
3452 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3453 }
3454
3455 static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3456 unsigned int cmd, unsigned long arg)
3457 {
3458 void __user *argp = compat_ptr(arg);
3459 struct sock *sk = sock->sk;
3460 struct net *net = sock_net(sk);
3461 const struct proto_ops *ops;
3462
3463 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3464 return sock_ioctl(file, cmd, (unsigned long)argp);
3465
3466 switch (cmd) {
3467 case SIOCWANDEV:
3468 return compat_siocwandev(net, argp);
3469 case SIOCGSTAMP_OLD:
3470 case SIOCGSTAMPNS_OLD:
3471 ops = READ_ONCE(sock->ops);
3472 if (!ops->gettstamp)
3473 return -ENOIOCTLCMD;
3474 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3475 !COMPAT_USE_64BIT_TIME);
3476
3477 case SIOCETHTOOL:
3478 case SIOCBONDSLAVEINFOQUERY:
3479 case SIOCBONDINFOQUERY:
3480 case SIOCSHWTSTAMP:
3481 case SIOCGHWTSTAMP:
3482 return compat_ifr_data_ioctl(net, cmd, argp);
3483
3484 case FIOSETOWN:
3485 case SIOCSPGRP:
3486 case FIOGETOWN:
3487 case SIOCGPGRP:
3488 case SIOCBRADDBR:
3489 case SIOCBRDELBR:
3490 case SIOCBRADDIF:
3491 case SIOCBRDELIF:
3492 case SIOCGIFVLAN:
3493 case SIOCSIFVLAN:
3494 case SIOCGSKNS:
3495 case SIOCGSTAMP_NEW:
3496 case SIOCGSTAMPNS_NEW:
3497 case SIOCGIFCONF:
3498 case SIOCSIFBR:
3499 case SIOCGIFBR:
3500 return sock_ioctl(file, cmd, arg);
3501
3502 case SIOCGIFFLAGS:
3503 case SIOCSIFFLAGS:
3504 case SIOCGIFMAP:
3505 case SIOCSIFMAP:
3506 case SIOCGIFMETRIC:
3507 case SIOCSIFMETRIC:
3508 case SIOCGIFMTU:
3509 case SIOCSIFMTU:
3510 case SIOCGIFMEM:
3511 case SIOCSIFMEM:
3512 case SIOCGIFHWADDR:
3513 case SIOCSIFHWADDR:
3514 case SIOCADDMULTI:
3515 case SIOCDELMULTI:
3516 case SIOCGIFINDEX:
3517 case SIOCGIFADDR:
3518 case SIOCSIFADDR:
3519 case SIOCSIFHWBROADCAST:
3520 case SIOCDIFADDR:
3521 case SIOCGIFBRDADDR:
3522 case SIOCSIFBRDADDR:
3523 case SIOCGIFDSTADDR:
3524 case SIOCSIFDSTADDR:
3525 case SIOCGIFNETMASK:
3526 case SIOCSIFNETMASK:
3527 case SIOCSIFPFLAGS:
3528 case SIOCGIFPFLAGS:
3529 case SIOCGIFTXQLEN:
3530 case SIOCSIFTXQLEN:
3531 case SIOCGIFNAME:
3532 case SIOCSIFNAME:
3533 case SIOCGMIIPHY:
3534 case SIOCGMIIREG:
3535 case SIOCSMIIREG:
3536 case SIOCBONDENSLAVE:
3537 case SIOCBONDRELEASE:
3538 case SIOCBONDSETHWADDR:
3539 case SIOCBONDCHANGEACTIVE:
3540 case SIOCSARP:
3541 case SIOCGARP:
3542 case SIOCDARP:
3543 case SIOCOUTQ:
3544 case SIOCOUTQNSD:
3545 case SIOCATMARK:
3546 return sock_do_ioctl(net, sock, cmd, arg);
3547 }
3548
3549 return -ENOIOCTLCMD;
3550 }
3551
3552 static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3553 unsigned long arg)
3554 {
3555 struct socket *sock = file->private_data;
3556 const struct proto_ops *ops = READ_ONCE(sock->ops);
3557 int ret = -ENOIOCTLCMD;
3558 struct sock *sk;
3559 struct net *net;
3560
3561 sk = sock->sk;
3562 net = sock_net(sk);
3563
3564 if (ops->compat_ioctl)
3565 ret = ops->compat_ioctl(sock, cmd, arg);
3566
3567 if (ret == -ENOIOCTLCMD &&
3568 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3569 ret = compat_wext_handle_ioctl(net, cmd, arg);
3570
3571 if (ret == -ENOIOCTLCMD)
3572 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3573
3574 return ret;
3575 }
3576 #endif
3577
3578 /**
3579 * kernel_bind - bind an address to a socket (kernel space)
3580 * @sock: socket
3581 * @addr: address
3582 * @addrlen: length of address
3583 *
3584 * Returns 0 or an error.
3585 */
3586
3587 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3588 {
3589 struct sockaddr_storage address;
3590
3591 memcpy(&address, addr, addrlen);
3592
3593 return READ_ONCE(sock->ops)->bind(sock, (struct sockaddr *)&address,
3594 addrlen);
3595 }
3596 EXPORT_SYMBOL(kernel_bind);
3597
3598 /**
3599 * kernel_listen - move socket to listening state (kernel space)
3600 * @sock: socket
3601 * @backlog: pending connections queue size
3602 *
3603 * Returns 0 or an error.
3604 */
3605
3606 int kernel_listen(struct socket *sock, int backlog)
3607 {
3608 return READ_ONCE(sock->ops)->listen(sock, backlog);
3609 }
3610 EXPORT_SYMBOL(kernel_listen);
3611
3612 /**
3613 * kernel_accept - accept a connection (kernel space)
3614 * @sock: listening socket
3615 * @newsock: new connected socket
3616 * @flags: flags
3617 *
3618 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3619 * If it fails, @newsock is guaranteed to be %NULL.
3620 * Returns 0 or an error.
3621 */
3622
3623 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3624 {
3625 struct sock *sk = sock->sk;
3626 const struct proto_ops *ops = READ_ONCE(sock->ops);
3627 struct proto_accept_arg arg = {
3628 .flags = flags,
3629 .kern = true,
3630 };
3631 int err;
3632
3633 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3634 newsock);
3635 if (err < 0)
3636 goto done;
3637
3638 err = ops->accept(sock, *newsock, &arg);
3639 if (err < 0) {
3640 sock_release(*newsock);
3641 *newsock = NULL;
3642 goto done;
3643 }
3644
3645 (*newsock)->ops = ops;
3646 __module_get(ops->owner);
3647
3648 done:
3649 return err;
3650 }
3651 EXPORT_SYMBOL(kernel_accept);
3652
3653 /**
3654 * kernel_connect - connect a socket (kernel space)
3655 * @sock: socket
3656 * @addr: address
3657 * @addrlen: address length
3658 * @flags: flags (O_NONBLOCK, ...)
3659 *
3660 * For datagram sockets, @addr is the address to which datagrams are sent
3661 * by default, and the only address from which datagrams are received.
3662 * For stream sockets, attempts to connect to @addr.
3663 * Returns 0 or an error code.
3664 */
3665
3666 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3667 int flags)
3668 {
3669 struct sockaddr_storage address;
3670
3671 memcpy(&address, addr, addrlen);
3672
3673 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address,
3674 addrlen, flags);
3675 }
3676 EXPORT_SYMBOL(kernel_connect);
3677
3678 /**
3679 * kernel_getsockname - get the address which the socket is bound (kernel space)
3680 * @sock: socket
3681 * @addr: address holder
3682 *
3683 * Fills the @addr pointer with the address which the socket is bound.
3684 * Returns the length of the address in bytes or an error code.
3685 */
3686
3687 int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3688 {
3689 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3690 }
3691 EXPORT_SYMBOL(kernel_getsockname);
3692
3693 /**
3694 * kernel_getpeername - get the address which the socket is connected (kernel space)
3695 * @sock: socket
3696 * @addr: address holder
3697 *
3698 * Fills the @addr pointer with the address which the socket is connected.
3699 * Returns the length of the address in bytes or an error code.
3700 */
3701
3702 int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3703 {
3704 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3705 }
3706 EXPORT_SYMBOL(kernel_getpeername);
3707
3708 /**
3709 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3710 * @sock: socket
3711 * @how: connection part
3712 *
3713 * Returns 0 or an error.
3714 */
3715
3716 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3717 {
3718 return READ_ONCE(sock->ops)->shutdown(sock, how);
3719 }
3720 EXPORT_SYMBOL(kernel_sock_shutdown);
3721
3722 /**
3723 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3724 * @sk: socket
3725 *
3726 * This routine returns the IP overhead imposed by a socket i.e.
3727 * the length of the underlying IP header, depending on whether
3728 * this is an IPv4 or IPv6 socket and the length from IP options turned
3729 * on at the socket. Assumes that the caller has a lock on the socket.
3730 */
3731
3732 u32 kernel_sock_ip_overhead(struct sock *sk)
3733 {
3734 struct inet_sock *inet;
3735 struct ip_options_rcu *opt;
3736 u32 overhead = 0;
3737 #if IS_ENABLED(CONFIG_IPV6)
3738 struct ipv6_pinfo *np;
3739 struct ipv6_txoptions *optv6 = NULL;
3740 #endif /* IS_ENABLED(CONFIG_IPV6) */
3741
3742 if (!sk)
3743 return overhead;
3744
3745 switch (sk->sk_family) {
3746 case AF_INET:
3747 inet = inet_sk(sk);
3748 overhead += sizeof(struct iphdr);
3749 opt = rcu_dereference_protected(inet->inet_opt,
3750 sock_owned_by_user(sk));
3751 if (opt)
3752 overhead += opt->opt.optlen;
3753 return overhead;
3754 #if IS_ENABLED(CONFIG_IPV6)
3755 case AF_INET6:
3756 np = inet6_sk(sk);
3757 overhead += sizeof(struct ipv6hdr);
3758 if (np)
3759 optv6 = rcu_dereference_protected(np->opt,
3760 sock_owned_by_user(sk));
3761 if (optv6)
3762 overhead += (optv6->opt_flen + optv6->opt_nflen);
3763 return overhead;
3764 #endif /* IS_ENABLED(CONFIG_IPV6) */
3765 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3766 return overhead;
3767 }
3768 }
3769 EXPORT_SYMBOL(kernel_sock_ip_overhead);