Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET An implementation of the SOCKET network access protocol.
4 *
5 * Version: @(#)socket.c 1.1.93 18/02/95
6 *
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
8 * Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * shutdown()
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
18 * top level.
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
23 * tty drivers).
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
26 * configurable.
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
35 * stuff.
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
41 * moment.
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
48 *
49 * This module is effectively the top level interface to the BSD socket
50 * paradigm.
51 *
52 * Based upon Swansea University Computer Society NET3.039
53 */
54
55#include <linux/bpf-cgroup.h>
56#include <linux/ethtool.h>
57#include <linux/mm.h>
58#include <linux/socket.h>
59#include <linux/file.h>
60#include <linux/splice.h>
61#include <linux/net.h>
62#include <linux/interrupt.h>
63#include <linux/thread_info.h>
64#include <linux/rcupdate.h>
65#include <linux/netdevice.h>
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68#include <linux/mutex.h>
69#include <linux/if_bridge.h>
70#include <linux/if_vlan.h>
71#include <linux/ptp_classify.h>
72#include <linux/init.h>
73#include <linux/poll.h>
74#include <linux/cache.h>
75#include <linux/module.h>
76#include <linux/highmem.h>
77#include <linux/mount.h>
78#include <linux/pseudo_fs.h>
79#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/compat.h>
82#include <linux/kmod.h>
83#include <linux/audit.h>
84#include <linux/wireless.h>
85#include <linux/nsproxy.h>
86#include <linux/magic.h>
87#include <linux/slab.h>
88#include <linux/xattr.h>
89#include <linux/nospec.h>
90#include <linux/indirect_call_wrapper.h>
91#include <linux/io_uring/net.h>
92
93#include <linux/uaccess.h>
94#include <asm/unistd.h>
95
96#include <net/compat.h>
97#include <net/wext.h>
98#include <net/cls_cgroup.h>
99
100#include <net/sock.h>
101#include <linux/netfilter.h>
102
103#include <linux/if_tun.h>
104#include <linux/ipv6_route.h>
105#include <linux/route.h>
106#include <linux/termios.h>
107#include <linux/sockios.h>
108#include <net/busy_poll.h>
109#include <linux/errqueue.h>
110#include <linux/ptp_clock_kernel.h>
111#include <trace/events/sock.h>
112
113#ifdef CONFIG_NET_RX_BUSY_POLL
114unsigned int sysctl_net_busy_read __read_mostly;
115unsigned int sysctl_net_busy_poll __read_mostly;
116#endif
117
118static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
119static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
120static int sock_mmap(struct file *file, struct vm_area_struct *vma);
121
122static int sock_close(struct inode *inode, struct file *file);
123static __poll_t sock_poll(struct file *file,
124 struct poll_table_struct *wait);
125static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
126#ifdef CONFIG_COMPAT
127static long compat_sock_ioctl(struct file *file,
128 unsigned int cmd, unsigned long arg);
129#endif
130static int sock_fasync(int fd, struct file *filp, int on);
131static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
133 unsigned int flags);
134static void sock_splice_eof(struct file *file);
135
136#ifdef CONFIG_PROC_FS
137static void sock_show_fdinfo(struct seq_file *m, struct file *f)
138{
139 struct socket *sock = f->private_data;
140 const struct proto_ops *ops = READ_ONCE(sock->ops);
141
142 if (ops->show_fdinfo)
143 ops->show_fdinfo(m, sock);
144}
145#else
146#define sock_show_fdinfo NULL
147#endif
148
149/*
150 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
151 * in the operation structures but are done directly via the socketcall() multiplexor.
152 */
153
154static const struct file_operations socket_file_ops = {
155 .owner = THIS_MODULE,
156 .read_iter = sock_read_iter,
157 .write_iter = sock_write_iter,
158 .poll = sock_poll,
159 .unlocked_ioctl = sock_ioctl,
160#ifdef CONFIG_COMPAT
161 .compat_ioctl = compat_sock_ioctl,
162#endif
163 .uring_cmd = io_uring_cmd_sock,
164 .mmap = sock_mmap,
165 .release = sock_close,
166 .fasync = sock_fasync,
167 .splice_write = splice_to_socket,
168 .splice_read = sock_splice_read,
169 .splice_eof = sock_splice_eof,
170 .show_fdinfo = sock_show_fdinfo,
171};
172
173static const char * const pf_family_names[] = {
174 [PF_UNSPEC] = "PF_UNSPEC",
175 [PF_UNIX] = "PF_UNIX/PF_LOCAL",
176 [PF_INET] = "PF_INET",
177 [PF_AX25] = "PF_AX25",
178 [PF_IPX] = "PF_IPX",
179 [PF_APPLETALK] = "PF_APPLETALK",
180 [PF_NETROM] = "PF_NETROM",
181 [PF_BRIDGE] = "PF_BRIDGE",
182 [PF_ATMPVC] = "PF_ATMPVC",
183 [PF_X25] = "PF_X25",
184 [PF_INET6] = "PF_INET6",
185 [PF_ROSE] = "PF_ROSE",
186 [PF_DECnet] = "PF_DECnet",
187 [PF_NETBEUI] = "PF_NETBEUI",
188 [PF_SECURITY] = "PF_SECURITY",
189 [PF_KEY] = "PF_KEY",
190 [PF_NETLINK] = "PF_NETLINK/PF_ROUTE",
191 [PF_PACKET] = "PF_PACKET",
192 [PF_ASH] = "PF_ASH",
193 [PF_ECONET] = "PF_ECONET",
194 [PF_ATMSVC] = "PF_ATMSVC",
195 [PF_RDS] = "PF_RDS",
196 [PF_SNA] = "PF_SNA",
197 [PF_IRDA] = "PF_IRDA",
198 [PF_PPPOX] = "PF_PPPOX",
199 [PF_WANPIPE] = "PF_WANPIPE",
200 [PF_LLC] = "PF_LLC",
201 [PF_IB] = "PF_IB",
202 [PF_MPLS] = "PF_MPLS",
203 [PF_CAN] = "PF_CAN",
204 [PF_TIPC] = "PF_TIPC",
205 [PF_BLUETOOTH] = "PF_BLUETOOTH",
206 [PF_IUCV] = "PF_IUCV",
207 [PF_RXRPC] = "PF_RXRPC",
208 [PF_ISDN] = "PF_ISDN",
209 [PF_PHONET] = "PF_PHONET",
210 [PF_IEEE802154] = "PF_IEEE802154",
211 [PF_CAIF] = "PF_CAIF",
212 [PF_ALG] = "PF_ALG",
213 [PF_NFC] = "PF_NFC",
214 [PF_VSOCK] = "PF_VSOCK",
215 [PF_KCM] = "PF_KCM",
216 [PF_QIPCRTR] = "PF_QIPCRTR",
217 [PF_SMC] = "PF_SMC",
218 [PF_XDP] = "PF_XDP",
219 [PF_MCTP] = "PF_MCTP",
220};
221
222/*
223 * The protocol list. Each protocol is registered in here.
224 */
225
226static DEFINE_SPINLOCK(net_family_lock);
227static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
228
229/*
230 * Support routines.
231 * Move socket addresses back and forth across the kernel/user
232 * divide and look after the messy bits.
233 */
234
235/**
236 * move_addr_to_kernel - copy a socket address into kernel space
237 * @uaddr: Address in user space
238 * @kaddr: Address in kernel space
239 * @ulen: Length in user space
240 *
241 * The address is copied into kernel space. If the provided address is
242 * too long an error code of -EINVAL is returned. If the copy gives
243 * invalid addresses -EFAULT is returned. On a success 0 is returned.
244 */
245
246int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
247{
248 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
249 return -EINVAL;
250 if (ulen == 0)
251 return 0;
252 if (copy_from_user(kaddr, uaddr, ulen))
253 return -EFAULT;
254 return audit_sockaddr(ulen, kaddr);
255}
256
257/**
258 * move_addr_to_user - copy an address to user space
259 * @kaddr: kernel space address
260 * @klen: length of address in kernel
261 * @uaddr: user space address
262 * @ulen: pointer to user length field
263 *
264 * The value pointed to by ulen on entry is the buffer length available.
265 * This is overwritten with the buffer space used. -EINVAL is returned
266 * if an overlong buffer is specified or a negative buffer size. -EFAULT
267 * is returned if either the buffer or the length field are not
268 * accessible.
269 * After copying the data up to the limit the user specifies, the true
270 * length of the data is written over the length limit the user
271 * specified. Zero is returned for a success.
272 */
273
274static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
275 void __user *uaddr, int __user *ulen)
276{
277 int err;
278 int len;
279
280 BUG_ON(klen > sizeof(struct sockaddr_storage));
281 err = get_user(len, ulen);
282 if (err)
283 return err;
284 if (len > klen)
285 len = klen;
286 if (len < 0)
287 return -EINVAL;
288 if (len) {
289 if (audit_sockaddr(klen, kaddr))
290 return -ENOMEM;
291 if (copy_to_user(uaddr, kaddr, len))
292 return -EFAULT;
293 }
294 /*
295 * "fromlen shall refer to the value before truncation.."
296 * 1003.1g
297 */
298 return __put_user(klen, ulen);
299}
300
301static struct kmem_cache *sock_inode_cachep __ro_after_init;
302
303static struct inode *sock_alloc_inode(struct super_block *sb)
304{
305 struct socket_alloc *ei;
306
307 ei = alloc_inode_sb(sb, sock_inode_cachep, GFP_KERNEL);
308 if (!ei)
309 return NULL;
310 init_waitqueue_head(&ei->socket.wq.wait);
311 ei->socket.wq.fasync_list = NULL;
312 ei->socket.wq.flags = 0;
313
314 ei->socket.state = SS_UNCONNECTED;
315 ei->socket.flags = 0;
316 ei->socket.ops = NULL;
317 ei->socket.sk = NULL;
318 ei->socket.file = NULL;
319
320 return &ei->vfs_inode;
321}
322
323static void sock_free_inode(struct inode *inode)
324{
325 struct socket_alloc *ei;
326
327 ei = container_of(inode, struct socket_alloc, vfs_inode);
328 kmem_cache_free(sock_inode_cachep, ei);
329}
330
331static void init_once(void *foo)
332{
333 struct socket_alloc *ei = (struct socket_alloc *)foo;
334
335 inode_init_once(&ei->vfs_inode);
336}
337
338static void init_inodecache(void)
339{
340 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
341 sizeof(struct socket_alloc),
342 0,
343 (SLAB_HWCACHE_ALIGN |
344 SLAB_RECLAIM_ACCOUNT |
345 SLAB_ACCOUNT),
346 init_once);
347 BUG_ON(sock_inode_cachep == NULL);
348}
349
350static const struct super_operations sockfs_ops = {
351 .alloc_inode = sock_alloc_inode,
352 .free_inode = sock_free_inode,
353 .statfs = simple_statfs,
354};
355
356/*
357 * sockfs_dname() is called from d_path().
358 */
359static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
360{
361 return dynamic_dname(buffer, buflen, "socket:[%lu]",
362 d_inode(dentry)->i_ino);
363}
364
365static const struct dentry_operations sockfs_dentry_operations = {
366 .d_dname = sockfs_dname,
367};
368
369static int sockfs_xattr_get(const struct xattr_handler *handler,
370 struct dentry *dentry, struct inode *inode,
371 const char *suffix, void *value, size_t size)
372{
373 if (value) {
374 if (dentry->d_name.len + 1 > size)
375 return -ERANGE;
376 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
377 }
378 return dentry->d_name.len + 1;
379}
380
381#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
382#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
383#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
384
385static const struct xattr_handler sockfs_xattr_handler = {
386 .name = XATTR_NAME_SOCKPROTONAME,
387 .get = sockfs_xattr_get,
388};
389
390static int sockfs_security_xattr_set(const struct xattr_handler *handler,
391 struct mnt_idmap *idmap,
392 struct dentry *dentry, struct inode *inode,
393 const char *suffix, const void *value,
394 size_t size, int flags)
395{
396 /* Handled by LSM. */
397 return -EAGAIN;
398}
399
400static const struct xattr_handler sockfs_security_xattr_handler = {
401 .prefix = XATTR_SECURITY_PREFIX,
402 .set = sockfs_security_xattr_set,
403};
404
405static const struct xattr_handler * const sockfs_xattr_handlers[] = {
406 &sockfs_xattr_handler,
407 &sockfs_security_xattr_handler,
408 NULL
409};
410
411static int sockfs_init_fs_context(struct fs_context *fc)
412{
413 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
414 if (!ctx)
415 return -ENOMEM;
416 ctx->ops = &sockfs_ops;
417 ctx->dops = &sockfs_dentry_operations;
418 ctx->xattr = sockfs_xattr_handlers;
419 return 0;
420}
421
422static struct vfsmount *sock_mnt __read_mostly;
423
424static struct file_system_type sock_fs_type = {
425 .name = "sockfs",
426 .init_fs_context = sockfs_init_fs_context,
427 .kill_sb = kill_anon_super,
428};
429
430/*
431 * Obtains the first available file descriptor and sets it up for use.
432 *
433 * These functions create file structures and maps them to fd space
434 * of the current process. On success it returns file descriptor
435 * and file struct implicitly stored in sock->file.
436 * Note that another thread may close file descriptor before we return
437 * from this function. We use the fact that now we do not refer
438 * to socket after mapping. If one day we will need it, this
439 * function will increment ref. count on file by 1.
440 *
441 * In any case returned fd MAY BE not valid!
442 * This race condition is unavoidable
443 * with shared fd spaces, we cannot solve it inside kernel,
444 * but we take care of internal coherence yet.
445 */
446
447/**
448 * sock_alloc_file - Bind a &socket to a &file
449 * @sock: socket
450 * @flags: file status flags
451 * @dname: protocol name
452 *
453 * Returns the &file bound with @sock, implicitly storing it
454 * in sock->file. If dname is %NULL, sets to "".
455 *
456 * On failure @sock is released, and an ERR pointer is returned.
457 *
458 * This function uses GFP_KERNEL internally.
459 */
460
461struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
462{
463 struct file *file;
464
465 if (!dname)
466 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
467
468 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
469 O_RDWR | (flags & O_NONBLOCK),
470 &socket_file_ops);
471 if (IS_ERR(file)) {
472 sock_release(sock);
473 return file;
474 }
475
476 file->f_mode |= FMODE_NOWAIT;
477 sock->file = file;
478 file->private_data = sock;
479 stream_open(SOCK_INODE(sock), file);
480 return file;
481}
482EXPORT_SYMBOL(sock_alloc_file);
483
484static int sock_map_fd(struct socket *sock, int flags)
485{
486 struct file *newfile;
487 int fd = get_unused_fd_flags(flags);
488 if (unlikely(fd < 0)) {
489 sock_release(sock);
490 return fd;
491 }
492
493 newfile = sock_alloc_file(sock, flags, NULL);
494 if (!IS_ERR(newfile)) {
495 fd_install(fd, newfile);
496 return fd;
497 }
498
499 put_unused_fd(fd);
500 return PTR_ERR(newfile);
501}
502
503/**
504 * sock_from_file - Return the &socket bounded to @file.
505 * @file: file
506 *
507 * On failure returns %NULL.
508 */
509
510struct socket *sock_from_file(struct file *file)
511{
512 if (likely(file->f_op == &socket_file_ops))
513 return file->private_data; /* set in sock_alloc_file */
514
515 return NULL;
516}
517EXPORT_SYMBOL(sock_from_file);
518
519/**
520 * sockfd_lookup - Go from a file number to its socket slot
521 * @fd: file handle
522 * @err: pointer to an error code return
523 *
524 * The file handle passed in is locked and the socket it is bound
525 * to is returned. If an error occurs the err pointer is overwritten
526 * with a negative errno code and NULL is returned. The function checks
527 * for both invalid handles and passing a handle which is not a socket.
528 *
529 * On a success the socket object pointer is returned.
530 */
531
532struct socket *sockfd_lookup(int fd, int *err)
533{
534 struct file *file;
535 struct socket *sock;
536
537 file = fget(fd);
538 if (!file) {
539 *err = -EBADF;
540 return NULL;
541 }
542
543 sock = sock_from_file(file);
544 if (!sock) {
545 *err = -ENOTSOCK;
546 fput(file);
547 }
548 return sock;
549}
550EXPORT_SYMBOL(sockfd_lookup);
551
552static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
553 size_t size)
554{
555 ssize_t len;
556 ssize_t used = 0;
557
558 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
559 if (len < 0)
560 return len;
561 used += len;
562 if (buffer) {
563 if (size < used)
564 return -ERANGE;
565 buffer += len;
566 }
567
568 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
569 used += len;
570 if (buffer) {
571 if (size < used)
572 return -ERANGE;
573 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
574 buffer += len;
575 }
576
577 return used;
578}
579
580static int sockfs_setattr(struct mnt_idmap *idmap,
581 struct dentry *dentry, struct iattr *iattr)
582{
583 int err = simple_setattr(&nop_mnt_idmap, dentry, iattr);
584
585 if (!err && (iattr->ia_valid & ATTR_UID)) {
586 struct socket *sock = SOCKET_I(d_inode(dentry));
587
588 if (sock->sk)
589 sock->sk->sk_uid = iattr->ia_uid;
590 else
591 err = -ENOENT;
592 }
593
594 return err;
595}
596
597static const struct inode_operations sockfs_inode_ops = {
598 .listxattr = sockfs_listxattr,
599 .setattr = sockfs_setattr,
600};
601
602/**
603 * sock_alloc - allocate a socket
604 *
605 * Allocate a new inode and socket object. The two are bound together
606 * and initialised. The socket is then returned. If we are out of inodes
607 * NULL is returned. This functions uses GFP_KERNEL internally.
608 */
609
610struct socket *sock_alloc(void)
611{
612 struct inode *inode;
613 struct socket *sock;
614
615 inode = new_inode_pseudo(sock_mnt->mnt_sb);
616 if (!inode)
617 return NULL;
618
619 sock = SOCKET_I(inode);
620
621 inode->i_ino = get_next_ino();
622 inode->i_mode = S_IFSOCK | S_IRWXUGO;
623 inode->i_uid = current_fsuid();
624 inode->i_gid = current_fsgid();
625 inode->i_op = &sockfs_inode_ops;
626
627 return sock;
628}
629EXPORT_SYMBOL(sock_alloc);
630
631static void __sock_release(struct socket *sock, struct inode *inode)
632{
633 const struct proto_ops *ops = READ_ONCE(sock->ops);
634
635 if (ops) {
636 struct module *owner = ops->owner;
637
638 if (inode)
639 inode_lock(inode);
640 ops->release(sock);
641 sock->sk = NULL;
642 if (inode)
643 inode_unlock(inode);
644 sock->ops = NULL;
645 module_put(owner);
646 }
647
648 if (sock->wq.fasync_list)
649 pr_err("%s: fasync list not empty!\n", __func__);
650
651 if (!sock->file) {
652 iput(SOCK_INODE(sock));
653 return;
654 }
655 sock->file = NULL;
656}
657
658/**
659 * sock_release - close a socket
660 * @sock: socket to close
661 *
662 * The socket is released from the protocol stack if it has a release
663 * callback, and the inode is then released if the socket is bound to
664 * an inode not a file.
665 */
666void sock_release(struct socket *sock)
667{
668 __sock_release(sock, NULL);
669}
670EXPORT_SYMBOL(sock_release);
671
672void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags)
673{
674 u8 flags = *tx_flags;
675
676 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE) {
677 flags |= SKBTX_HW_TSTAMP;
678
679 /* PTP hardware clocks can provide a free running cycle counter
680 * as a time base for virtual clocks. Tell driver to use the
681 * free running cycle counter for timestamp if socket is bound
682 * to virtual clock.
683 */
684 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
685 flags |= SKBTX_HW_TSTAMP_USE_CYCLES;
686 }
687
688 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
689 flags |= SKBTX_SW_TSTAMP;
690
691 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
692 flags |= SKBTX_SCHED_TSTAMP;
693
694 *tx_flags = flags;
695}
696EXPORT_SYMBOL(__sock_tx_timestamp);
697
698INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
699 size_t));
700INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
701 size_t));
702
703static noinline void call_trace_sock_send_length(struct sock *sk, int ret,
704 int flags)
705{
706 trace_sock_send_length(sk, ret, 0);
707}
708
709static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
710{
711 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->sendmsg, inet6_sendmsg,
712 inet_sendmsg, sock, msg,
713 msg_data_left(msg));
714 BUG_ON(ret == -EIOCBQUEUED);
715
716 if (trace_sock_send_length_enabled())
717 call_trace_sock_send_length(sock->sk, ret, 0);
718 return ret;
719}
720
721static int __sock_sendmsg(struct socket *sock, struct msghdr *msg)
722{
723 int err = security_socket_sendmsg(sock, msg,
724 msg_data_left(msg));
725
726 return err ?: sock_sendmsg_nosec(sock, msg);
727}
728
729/**
730 * sock_sendmsg - send a message through @sock
731 * @sock: socket
732 * @msg: message to send
733 *
734 * Sends @msg through @sock, passing through LSM.
735 * Returns the number of bytes sent, or an error code.
736 */
737int sock_sendmsg(struct socket *sock, struct msghdr *msg)
738{
739 struct sockaddr_storage *save_addr = (struct sockaddr_storage *)msg->msg_name;
740 struct sockaddr_storage address;
741 int save_len = msg->msg_namelen;
742 int ret;
743
744 if (msg->msg_name) {
745 memcpy(&address, msg->msg_name, msg->msg_namelen);
746 msg->msg_name = &address;
747 }
748
749 ret = __sock_sendmsg(sock, msg);
750 msg->msg_name = save_addr;
751 msg->msg_namelen = save_len;
752
753 return ret;
754}
755EXPORT_SYMBOL(sock_sendmsg);
756
757/**
758 * kernel_sendmsg - send a message through @sock (kernel-space)
759 * @sock: socket
760 * @msg: message header
761 * @vec: kernel vec
762 * @num: vec array length
763 * @size: total message data size
764 *
765 * Builds the message data with @vec and sends it through @sock.
766 * Returns the number of bytes sent, or an error code.
767 */
768
769int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
770 struct kvec *vec, size_t num, size_t size)
771{
772 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
773 return sock_sendmsg(sock, msg);
774}
775EXPORT_SYMBOL(kernel_sendmsg);
776
777/**
778 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
779 * @sk: sock
780 * @msg: message header
781 * @vec: output s/g array
782 * @num: output s/g array length
783 * @size: total message data size
784 *
785 * Builds the message data with @vec and sends it through @sock.
786 * Returns the number of bytes sent, or an error code.
787 * Caller must hold @sk.
788 */
789
790int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
791 struct kvec *vec, size_t num, size_t size)
792{
793 struct socket *sock = sk->sk_socket;
794 const struct proto_ops *ops = READ_ONCE(sock->ops);
795
796 if (!ops->sendmsg_locked)
797 return sock_no_sendmsg_locked(sk, msg, size);
798
799 iov_iter_kvec(&msg->msg_iter, ITER_SOURCE, vec, num, size);
800
801 return ops->sendmsg_locked(sk, msg, msg_data_left(msg));
802}
803EXPORT_SYMBOL(kernel_sendmsg_locked);
804
805static bool skb_is_err_queue(const struct sk_buff *skb)
806{
807 /* pkt_type of skbs enqueued on the error queue are set to
808 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
809 * in recvmsg, since skbs received on a local socket will never
810 * have a pkt_type of PACKET_OUTGOING.
811 */
812 return skb->pkt_type == PACKET_OUTGOING;
813}
814
815/* On transmit, software and hardware timestamps are returned independently.
816 * As the two skb clones share the hardware timestamp, which may be updated
817 * before the software timestamp is received, a hardware TX timestamp may be
818 * returned only if there is no software TX timestamp. Ignore false software
819 * timestamps, which may be made in the __sock_recv_timestamp() call when the
820 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
821 * hardware timestamp.
822 */
823static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
824{
825 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
826}
827
828static ktime_t get_timestamp(struct sock *sk, struct sk_buff *skb, int *if_index)
829{
830 bool cycles = READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_BIND_PHC;
831 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
832 struct net_device *orig_dev;
833 ktime_t hwtstamp;
834
835 rcu_read_lock();
836 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
837 if (orig_dev) {
838 *if_index = orig_dev->ifindex;
839 hwtstamp = netdev_get_tstamp(orig_dev, shhwtstamps, cycles);
840 } else {
841 hwtstamp = shhwtstamps->hwtstamp;
842 }
843 rcu_read_unlock();
844
845 return hwtstamp;
846}
847
848static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb,
849 int if_index)
850{
851 struct scm_ts_pktinfo ts_pktinfo;
852 struct net_device *orig_dev;
853
854 if (!skb_mac_header_was_set(skb))
855 return;
856
857 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
858
859 if (!if_index) {
860 rcu_read_lock();
861 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
862 if (orig_dev)
863 if_index = orig_dev->ifindex;
864 rcu_read_unlock();
865 }
866 ts_pktinfo.if_index = if_index;
867
868 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
869 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
870 sizeof(ts_pktinfo), &ts_pktinfo);
871}
872
873/*
874 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
875 */
876void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
877 struct sk_buff *skb)
878{
879 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
880 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
881 struct scm_timestamping_internal tss;
882 int empty = 1, false_tstamp = 0;
883 struct skb_shared_hwtstamps *shhwtstamps =
884 skb_hwtstamps(skb);
885 int if_index;
886 ktime_t hwtstamp;
887 u32 tsflags;
888
889 /* Race occurred between timestamp enabling and packet
890 receiving. Fill in the current time for now. */
891 if (need_software_tstamp && skb->tstamp == 0) {
892 __net_timestamp(skb);
893 false_tstamp = 1;
894 }
895
896 if (need_software_tstamp) {
897 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
898 if (new_tstamp) {
899 struct __kernel_sock_timeval tv;
900
901 skb_get_new_timestamp(skb, &tv);
902 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
903 sizeof(tv), &tv);
904 } else {
905 struct __kernel_old_timeval tv;
906
907 skb_get_timestamp(skb, &tv);
908 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
909 sizeof(tv), &tv);
910 }
911 } else {
912 if (new_tstamp) {
913 struct __kernel_timespec ts;
914
915 skb_get_new_timestampns(skb, &ts);
916 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
917 sizeof(ts), &ts);
918 } else {
919 struct __kernel_old_timespec ts;
920
921 skb_get_timestampns(skb, &ts);
922 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
923 sizeof(ts), &ts);
924 }
925 }
926 }
927
928 memset(&tss, 0, sizeof(tss));
929 tsflags = READ_ONCE(sk->sk_tsflags);
930 if ((tsflags & SOF_TIMESTAMPING_SOFTWARE &&
931 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE ||
932 skb_is_err_queue(skb) ||
933 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
934 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
935 empty = 0;
936 if (shhwtstamps &&
937 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE &&
938 (tsflags & SOF_TIMESTAMPING_RX_HARDWARE ||
939 skb_is_err_queue(skb) ||
940 !(tsflags & SOF_TIMESTAMPING_OPT_RX_FILTER))) &&
941 !skb_is_swtx_tstamp(skb, false_tstamp)) {
942 if_index = 0;
943 if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_NETDEV)
944 hwtstamp = get_timestamp(sk, skb, &if_index);
945 else
946 hwtstamp = shhwtstamps->hwtstamp;
947
948 if (tsflags & SOF_TIMESTAMPING_BIND_PHC)
949 hwtstamp = ptp_convert_timestamp(&hwtstamp,
950 READ_ONCE(sk->sk_bind_phc));
951
952 if (ktime_to_timespec64_cond(hwtstamp, tss.ts + 2)) {
953 empty = 0;
954
955 if ((tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
956 !skb_is_err_queue(skb))
957 put_ts_pktinfo(msg, skb, if_index);
958 }
959 }
960 if (!empty) {
961 if (sock_flag(sk, SOCK_TSTAMP_NEW))
962 put_cmsg_scm_timestamping64(msg, &tss);
963 else
964 put_cmsg_scm_timestamping(msg, &tss);
965
966 if (skb_is_err_queue(skb) && skb->len &&
967 SKB_EXT_ERR(skb)->opt_stats)
968 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
969 skb->len, skb->data);
970 }
971}
972EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
973
974#ifdef CONFIG_WIRELESS
975void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
976 struct sk_buff *skb)
977{
978 int ack;
979
980 if (!sock_flag(sk, SOCK_WIFI_STATUS))
981 return;
982 if (!skb->wifi_acked_valid)
983 return;
984
985 ack = skb->wifi_acked;
986
987 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
988}
989EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
990#endif
991
992static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
993 struct sk_buff *skb)
994{
995 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
996 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
997 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
998}
999
1000static void sock_recv_mark(struct msghdr *msg, struct sock *sk,
1001 struct sk_buff *skb)
1002{
1003 if (sock_flag(sk, SOCK_RCVMARK) && skb) {
1004 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */
1005 __u32 mark = skb->mark;
1006
1007 put_cmsg(msg, SOL_SOCKET, SO_MARK, sizeof(__u32), &mark);
1008 }
1009}
1010
1011void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
1012 struct sk_buff *skb)
1013{
1014 sock_recv_timestamp(msg, sk, skb);
1015 sock_recv_drops(msg, sk, skb);
1016 sock_recv_mark(msg, sk, skb);
1017}
1018EXPORT_SYMBOL_GPL(__sock_recv_cmsgs);
1019
1020INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
1021 size_t, int));
1022INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
1023 size_t, int));
1024
1025static noinline void call_trace_sock_recv_length(struct sock *sk, int ret, int flags)
1026{
1027 trace_sock_recv_length(sk, ret, flags);
1028}
1029
1030static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
1031 int flags)
1032{
1033 int ret = INDIRECT_CALL_INET(READ_ONCE(sock->ops)->recvmsg,
1034 inet6_recvmsg,
1035 inet_recvmsg, sock, msg,
1036 msg_data_left(msg), flags);
1037 if (trace_sock_recv_length_enabled())
1038 call_trace_sock_recv_length(sock->sk, ret, flags);
1039 return ret;
1040}
1041
1042/**
1043 * sock_recvmsg - receive a message from @sock
1044 * @sock: socket
1045 * @msg: message to receive
1046 * @flags: message flags
1047 *
1048 * Receives @msg from @sock, passing through LSM. Returns the total number
1049 * of bytes received, or an error.
1050 */
1051int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
1052{
1053 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
1054
1055 return err ?: sock_recvmsg_nosec(sock, msg, flags);
1056}
1057EXPORT_SYMBOL(sock_recvmsg);
1058
1059/**
1060 * kernel_recvmsg - Receive a message from a socket (kernel space)
1061 * @sock: The socket to receive the message from
1062 * @msg: Received message
1063 * @vec: Input s/g array for message data
1064 * @num: Size of input s/g array
1065 * @size: Number of bytes to read
1066 * @flags: Message flags (MSG_DONTWAIT, etc...)
1067 *
1068 * On return the msg structure contains the scatter/gather array passed in the
1069 * vec argument. The array is modified so that it consists of the unfilled
1070 * portion of the original array.
1071 *
1072 * The returned value is the total number of bytes received, or an error.
1073 */
1074
1075int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
1076 struct kvec *vec, size_t num, size_t size, int flags)
1077{
1078 msg->msg_control_is_user = false;
1079 iov_iter_kvec(&msg->msg_iter, ITER_DEST, vec, num, size);
1080 return sock_recvmsg(sock, msg, flags);
1081}
1082EXPORT_SYMBOL(kernel_recvmsg);
1083
1084static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
1085 struct pipe_inode_info *pipe, size_t len,
1086 unsigned int flags)
1087{
1088 struct socket *sock = file->private_data;
1089 const struct proto_ops *ops;
1090
1091 ops = READ_ONCE(sock->ops);
1092 if (unlikely(!ops->splice_read))
1093 return copy_splice_read(file, ppos, pipe, len, flags);
1094
1095 return ops->splice_read(sock, ppos, pipe, len, flags);
1096}
1097
1098static void sock_splice_eof(struct file *file)
1099{
1100 struct socket *sock = file->private_data;
1101 const struct proto_ops *ops;
1102
1103 ops = READ_ONCE(sock->ops);
1104 if (ops->splice_eof)
1105 ops->splice_eof(sock);
1106}
1107
1108static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
1109{
1110 struct file *file = iocb->ki_filp;
1111 struct socket *sock = file->private_data;
1112 struct msghdr msg = {.msg_iter = *to,
1113 .msg_iocb = iocb};
1114 ssize_t res;
1115
1116 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1117 msg.msg_flags = MSG_DONTWAIT;
1118
1119 if (iocb->ki_pos != 0)
1120 return -ESPIPE;
1121
1122 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
1123 return 0;
1124
1125 res = sock_recvmsg(sock, &msg, msg.msg_flags);
1126 *to = msg.msg_iter;
1127 return res;
1128}
1129
1130static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
1131{
1132 struct file *file = iocb->ki_filp;
1133 struct socket *sock = file->private_data;
1134 struct msghdr msg = {.msg_iter = *from,
1135 .msg_iocb = iocb};
1136 ssize_t res;
1137
1138 if (iocb->ki_pos != 0)
1139 return -ESPIPE;
1140
1141 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
1142 msg.msg_flags = MSG_DONTWAIT;
1143
1144 if (sock->type == SOCK_SEQPACKET)
1145 msg.msg_flags |= MSG_EOR;
1146
1147 res = __sock_sendmsg(sock, &msg);
1148 *from = msg.msg_iter;
1149 return res;
1150}
1151
1152/*
1153 * Atomic setting of ioctl hooks to avoid race
1154 * with module unload.
1155 */
1156
1157static DEFINE_MUTEX(br_ioctl_mutex);
1158static int (*br_ioctl_hook)(struct net *net, struct net_bridge *br,
1159 unsigned int cmd, struct ifreq *ifr,
1160 void __user *uarg);
1161
1162void brioctl_set(int (*hook)(struct net *net, struct net_bridge *br,
1163 unsigned int cmd, struct ifreq *ifr,
1164 void __user *uarg))
1165{
1166 mutex_lock(&br_ioctl_mutex);
1167 br_ioctl_hook = hook;
1168 mutex_unlock(&br_ioctl_mutex);
1169}
1170EXPORT_SYMBOL(brioctl_set);
1171
1172int br_ioctl_call(struct net *net, struct net_bridge *br, unsigned int cmd,
1173 struct ifreq *ifr, void __user *uarg)
1174{
1175 int err = -ENOPKG;
1176
1177 if (!br_ioctl_hook)
1178 request_module("bridge");
1179
1180 mutex_lock(&br_ioctl_mutex);
1181 if (br_ioctl_hook)
1182 err = br_ioctl_hook(net, br, cmd, ifr, uarg);
1183 mutex_unlock(&br_ioctl_mutex);
1184
1185 return err;
1186}
1187
1188static DEFINE_MUTEX(vlan_ioctl_mutex);
1189static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1190
1191void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1192{
1193 mutex_lock(&vlan_ioctl_mutex);
1194 vlan_ioctl_hook = hook;
1195 mutex_unlock(&vlan_ioctl_mutex);
1196}
1197EXPORT_SYMBOL(vlan_ioctl_set);
1198
1199static long sock_do_ioctl(struct net *net, struct socket *sock,
1200 unsigned int cmd, unsigned long arg)
1201{
1202 const struct proto_ops *ops = READ_ONCE(sock->ops);
1203 struct ifreq ifr;
1204 bool need_copyout;
1205 int err;
1206 void __user *argp = (void __user *)arg;
1207 void __user *data;
1208
1209 err = ops->ioctl(sock, cmd, arg);
1210
1211 /*
1212 * If this ioctl is unknown try to hand it down
1213 * to the NIC driver.
1214 */
1215 if (err != -ENOIOCTLCMD)
1216 return err;
1217
1218 if (!is_socket_ioctl_cmd(cmd))
1219 return -ENOTTY;
1220
1221 if (get_user_ifreq(&ifr, &data, argp))
1222 return -EFAULT;
1223 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1224 if (!err && need_copyout)
1225 if (put_user_ifreq(&ifr, argp))
1226 return -EFAULT;
1227
1228 return err;
1229}
1230
1231/*
1232 * With an ioctl, arg may well be a user mode pointer, but we don't know
1233 * what to do with it - that's up to the protocol still.
1234 */
1235
1236static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1237{
1238 const struct proto_ops *ops;
1239 struct socket *sock;
1240 struct sock *sk;
1241 void __user *argp = (void __user *)arg;
1242 int pid, err;
1243 struct net *net;
1244
1245 sock = file->private_data;
1246 ops = READ_ONCE(sock->ops);
1247 sk = sock->sk;
1248 net = sock_net(sk);
1249 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1250 struct ifreq ifr;
1251 void __user *data;
1252 bool need_copyout;
1253 if (get_user_ifreq(&ifr, &data, argp))
1254 return -EFAULT;
1255 err = dev_ioctl(net, cmd, &ifr, data, &need_copyout);
1256 if (!err && need_copyout)
1257 if (put_user_ifreq(&ifr, argp))
1258 return -EFAULT;
1259 } else
1260#ifdef CONFIG_WEXT_CORE
1261 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1262 err = wext_handle_ioctl(net, cmd, argp);
1263 } else
1264#endif
1265 switch (cmd) {
1266 case FIOSETOWN:
1267 case SIOCSPGRP:
1268 err = -EFAULT;
1269 if (get_user(pid, (int __user *)argp))
1270 break;
1271 err = f_setown(sock->file, pid, 1);
1272 break;
1273 case FIOGETOWN:
1274 case SIOCGPGRP:
1275 err = put_user(f_getown(sock->file),
1276 (int __user *)argp);
1277 break;
1278 case SIOCGIFBR:
1279 case SIOCSIFBR:
1280 case SIOCBRADDBR:
1281 case SIOCBRDELBR:
1282 err = br_ioctl_call(net, NULL, cmd, NULL, argp);
1283 break;
1284 case SIOCGIFVLAN:
1285 case SIOCSIFVLAN:
1286 err = -ENOPKG;
1287 if (!vlan_ioctl_hook)
1288 request_module("8021q");
1289
1290 mutex_lock(&vlan_ioctl_mutex);
1291 if (vlan_ioctl_hook)
1292 err = vlan_ioctl_hook(net, argp);
1293 mutex_unlock(&vlan_ioctl_mutex);
1294 break;
1295 case SIOCGSKNS:
1296 err = -EPERM;
1297 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1298 break;
1299
1300 err = open_related_ns(&net->ns, get_net_ns);
1301 break;
1302 case SIOCGSTAMP_OLD:
1303 case SIOCGSTAMPNS_OLD:
1304 if (!ops->gettstamp) {
1305 err = -ENOIOCTLCMD;
1306 break;
1307 }
1308 err = ops->gettstamp(sock, argp,
1309 cmd == SIOCGSTAMP_OLD,
1310 !IS_ENABLED(CONFIG_64BIT));
1311 break;
1312 case SIOCGSTAMP_NEW:
1313 case SIOCGSTAMPNS_NEW:
1314 if (!ops->gettstamp) {
1315 err = -ENOIOCTLCMD;
1316 break;
1317 }
1318 err = ops->gettstamp(sock, argp,
1319 cmd == SIOCGSTAMP_NEW,
1320 false);
1321 break;
1322
1323 case SIOCGIFCONF:
1324 err = dev_ifconf(net, argp);
1325 break;
1326
1327 default:
1328 err = sock_do_ioctl(net, sock, cmd, arg);
1329 break;
1330 }
1331 return err;
1332}
1333
1334/**
1335 * sock_create_lite - creates a socket
1336 * @family: protocol family (AF_INET, ...)
1337 * @type: communication type (SOCK_STREAM, ...)
1338 * @protocol: protocol (0, ...)
1339 * @res: new socket
1340 *
1341 * Creates a new socket and assigns it to @res, passing through LSM.
1342 * The new socket initialization is not complete, see kernel_accept().
1343 * Returns 0 or an error. On failure @res is set to %NULL.
1344 * This function internally uses GFP_KERNEL.
1345 */
1346
1347int sock_create_lite(int family, int type, int protocol, struct socket **res)
1348{
1349 int err;
1350 struct socket *sock = NULL;
1351
1352 err = security_socket_create(family, type, protocol, 1);
1353 if (err)
1354 goto out;
1355
1356 sock = sock_alloc();
1357 if (!sock) {
1358 err = -ENOMEM;
1359 goto out;
1360 }
1361
1362 sock->type = type;
1363 err = security_socket_post_create(sock, family, type, protocol, 1);
1364 if (err)
1365 goto out_release;
1366
1367out:
1368 *res = sock;
1369 return err;
1370out_release:
1371 sock_release(sock);
1372 sock = NULL;
1373 goto out;
1374}
1375EXPORT_SYMBOL(sock_create_lite);
1376
1377/* No kernel lock held - perfect */
1378static __poll_t sock_poll(struct file *file, poll_table *wait)
1379{
1380 struct socket *sock = file->private_data;
1381 const struct proto_ops *ops = READ_ONCE(sock->ops);
1382 __poll_t events = poll_requested_events(wait), flag = 0;
1383
1384 if (!ops->poll)
1385 return 0;
1386
1387 if (sk_can_busy_loop(sock->sk)) {
1388 /* poll once if requested by the syscall */
1389 if (events & POLL_BUSY_LOOP)
1390 sk_busy_loop(sock->sk, 1);
1391
1392 /* if this socket can poll_ll, tell the system call */
1393 flag = POLL_BUSY_LOOP;
1394 }
1395
1396 return ops->poll(file, sock, wait) | flag;
1397}
1398
1399static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1400{
1401 struct socket *sock = file->private_data;
1402
1403 return READ_ONCE(sock->ops)->mmap(file, sock, vma);
1404}
1405
1406static int sock_close(struct inode *inode, struct file *filp)
1407{
1408 __sock_release(SOCKET_I(inode), inode);
1409 return 0;
1410}
1411
1412/*
1413 * Update the socket async list
1414 *
1415 * Fasync_list locking strategy.
1416 *
1417 * 1. fasync_list is modified only under process context socket lock
1418 * i.e. under semaphore.
1419 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1420 * or under socket lock
1421 */
1422
1423static int sock_fasync(int fd, struct file *filp, int on)
1424{
1425 struct socket *sock = filp->private_data;
1426 struct sock *sk = sock->sk;
1427 struct socket_wq *wq = &sock->wq;
1428
1429 if (sk == NULL)
1430 return -EINVAL;
1431
1432 lock_sock(sk);
1433 fasync_helper(fd, filp, on, &wq->fasync_list);
1434
1435 if (!wq->fasync_list)
1436 sock_reset_flag(sk, SOCK_FASYNC);
1437 else
1438 sock_set_flag(sk, SOCK_FASYNC);
1439
1440 release_sock(sk);
1441 return 0;
1442}
1443
1444/* This function may be called only under rcu_lock */
1445
1446int sock_wake_async(struct socket_wq *wq, int how, int band)
1447{
1448 if (!wq || !wq->fasync_list)
1449 return -1;
1450
1451 switch (how) {
1452 case SOCK_WAKE_WAITD:
1453 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1454 break;
1455 goto call_kill;
1456 case SOCK_WAKE_SPACE:
1457 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1458 break;
1459 fallthrough;
1460 case SOCK_WAKE_IO:
1461call_kill:
1462 kill_fasync(&wq->fasync_list, SIGIO, band);
1463 break;
1464 case SOCK_WAKE_URG:
1465 kill_fasync(&wq->fasync_list, SIGURG, band);
1466 }
1467
1468 return 0;
1469}
1470EXPORT_SYMBOL(sock_wake_async);
1471
1472/**
1473 * __sock_create - creates a socket
1474 * @net: net namespace
1475 * @family: protocol family (AF_INET, ...)
1476 * @type: communication type (SOCK_STREAM, ...)
1477 * @protocol: protocol (0, ...)
1478 * @res: new socket
1479 * @kern: boolean for kernel space sockets
1480 *
1481 * Creates a new socket and assigns it to @res, passing through LSM.
1482 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1483 * be set to true if the socket resides in kernel space.
1484 * This function internally uses GFP_KERNEL.
1485 */
1486
1487int __sock_create(struct net *net, int family, int type, int protocol,
1488 struct socket **res, int kern)
1489{
1490 int err;
1491 struct socket *sock;
1492 const struct net_proto_family *pf;
1493
1494 /*
1495 * Check protocol is in range
1496 */
1497 if (family < 0 || family >= NPROTO)
1498 return -EAFNOSUPPORT;
1499 if (type < 0 || type >= SOCK_MAX)
1500 return -EINVAL;
1501
1502 /* Compatibility.
1503
1504 This uglymoron is moved from INET layer to here to avoid
1505 deadlock in module load.
1506 */
1507 if (family == PF_INET && type == SOCK_PACKET) {
1508 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1509 current->comm);
1510 family = PF_PACKET;
1511 }
1512
1513 err = security_socket_create(family, type, protocol, kern);
1514 if (err)
1515 return err;
1516
1517 /*
1518 * Allocate the socket and allow the family to set things up. if
1519 * the protocol is 0, the family is instructed to select an appropriate
1520 * default.
1521 */
1522 sock = sock_alloc();
1523 if (!sock) {
1524 net_warn_ratelimited("socket: no more sockets\n");
1525 return -ENFILE; /* Not exactly a match, but its the
1526 closest posix thing */
1527 }
1528
1529 sock->type = type;
1530
1531#ifdef CONFIG_MODULES
1532 /* Attempt to load a protocol module if the find failed.
1533 *
1534 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1535 * requested real, full-featured networking support upon configuration.
1536 * Otherwise module support will break!
1537 */
1538 if (rcu_access_pointer(net_families[family]) == NULL)
1539 request_module("net-pf-%d", family);
1540#endif
1541
1542 rcu_read_lock();
1543 pf = rcu_dereference(net_families[family]);
1544 err = -EAFNOSUPPORT;
1545 if (!pf)
1546 goto out_release;
1547
1548 /*
1549 * We will call the ->create function, that possibly is in a loadable
1550 * module, so we have to bump that loadable module refcnt first.
1551 */
1552 if (!try_module_get(pf->owner))
1553 goto out_release;
1554
1555 /* Now protected by module ref count */
1556 rcu_read_unlock();
1557
1558 err = pf->create(net, sock, protocol, kern);
1559 if (err < 0) {
1560 /* ->create should release the allocated sock->sk object on error
1561 * and make sure sock->sk is set to NULL to avoid use-after-free
1562 */
1563 DEBUG_NET_WARN_ONCE(sock->sk,
1564 "%ps must clear sock->sk on failure, family: %d, type: %d, protocol: %d\n",
1565 pf->create, family, type, protocol);
1566 goto out_module_put;
1567 }
1568
1569 /*
1570 * Now to bump the refcnt of the [loadable] module that owns this
1571 * socket at sock_release time we decrement its refcnt.
1572 */
1573 if (!try_module_get(sock->ops->owner))
1574 goto out_module_busy;
1575
1576 /*
1577 * Now that we're done with the ->create function, the [loadable]
1578 * module can have its refcnt decremented
1579 */
1580 module_put(pf->owner);
1581 err = security_socket_post_create(sock, family, type, protocol, kern);
1582 if (err)
1583 goto out_sock_release;
1584 *res = sock;
1585
1586 return 0;
1587
1588out_module_busy:
1589 err = -EAFNOSUPPORT;
1590out_module_put:
1591 sock->ops = NULL;
1592 module_put(pf->owner);
1593out_sock_release:
1594 sock_release(sock);
1595 return err;
1596
1597out_release:
1598 rcu_read_unlock();
1599 goto out_sock_release;
1600}
1601EXPORT_SYMBOL(__sock_create);
1602
1603/**
1604 * sock_create - creates a socket
1605 * @family: protocol family (AF_INET, ...)
1606 * @type: communication type (SOCK_STREAM, ...)
1607 * @protocol: protocol (0, ...)
1608 * @res: new socket
1609 *
1610 * A wrapper around __sock_create().
1611 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1612 */
1613
1614int sock_create(int family, int type, int protocol, struct socket **res)
1615{
1616 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1617}
1618EXPORT_SYMBOL(sock_create);
1619
1620/**
1621 * sock_create_kern - creates a socket (kernel space)
1622 * @net: net namespace
1623 * @family: protocol family (AF_INET, ...)
1624 * @type: communication type (SOCK_STREAM, ...)
1625 * @protocol: protocol (0, ...)
1626 * @res: new socket
1627 *
1628 * A wrapper around __sock_create().
1629 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1630 */
1631
1632int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1633{
1634 return __sock_create(net, family, type, protocol, res, 1);
1635}
1636EXPORT_SYMBOL(sock_create_kern);
1637
1638static struct socket *__sys_socket_create(int family, int type, int protocol)
1639{
1640 struct socket *sock;
1641 int retval;
1642
1643 /* Check the SOCK_* constants for consistency. */
1644 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1645 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1646 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1647 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1648
1649 if ((type & ~SOCK_TYPE_MASK) & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1650 return ERR_PTR(-EINVAL);
1651 type &= SOCK_TYPE_MASK;
1652
1653 retval = sock_create(family, type, protocol, &sock);
1654 if (retval < 0)
1655 return ERR_PTR(retval);
1656
1657 return sock;
1658}
1659
1660struct file *__sys_socket_file(int family, int type, int protocol)
1661{
1662 struct socket *sock;
1663 int flags;
1664
1665 sock = __sys_socket_create(family, type, protocol);
1666 if (IS_ERR(sock))
1667 return ERR_CAST(sock);
1668
1669 flags = type & ~SOCK_TYPE_MASK;
1670 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1671 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1672
1673 return sock_alloc_file(sock, flags, NULL);
1674}
1675
1676/* A hook for bpf progs to attach to and update socket protocol.
1677 *
1678 * A static noinline declaration here could cause the compiler to
1679 * optimize away the function. A global noinline declaration will
1680 * keep the definition, but may optimize away the callsite.
1681 * Therefore, __weak is needed to ensure that the call is still
1682 * emitted, by telling the compiler that we don't know what the
1683 * function might eventually be.
1684 */
1685
1686__bpf_hook_start();
1687
1688__weak noinline int update_socket_protocol(int family, int type, int protocol)
1689{
1690 return protocol;
1691}
1692
1693__bpf_hook_end();
1694
1695int __sys_socket(int family, int type, int protocol)
1696{
1697 struct socket *sock;
1698 int flags;
1699
1700 sock = __sys_socket_create(family, type,
1701 update_socket_protocol(family, type, protocol));
1702 if (IS_ERR(sock))
1703 return PTR_ERR(sock);
1704
1705 flags = type & ~SOCK_TYPE_MASK;
1706 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1707 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1708
1709 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1710}
1711
1712SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1713{
1714 return __sys_socket(family, type, protocol);
1715}
1716
1717/*
1718 * Create a pair of connected sockets.
1719 */
1720
1721int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1722{
1723 struct socket *sock1, *sock2;
1724 int fd1, fd2, err;
1725 struct file *newfile1, *newfile2;
1726 int flags;
1727
1728 flags = type & ~SOCK_TYPE_MASK;
1729 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1730 return -EINVAL;
1731 type &= SOCK_TYPE_MASK;
1732
1733 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1734 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1735
1736 /*
1737 * reserve descriptors and make sure we won't fail
1738 * to return them to userland.
1739 */
1740 fd1 = get_unused_fd_flags(flags);
1741 if (unlikely(fd1 < 0))
1742 return fd1;
1743
1744 fd2 = get_unused_fd_flags(flags);
1745 if (unlikely(fd2 < 0)) {
1746 put_unused_fd(fd1);
1747 return fd2;
1748 }
1749
1750 err = put_user(fd1, &usockvec[0]);
1751 if (err)
1752 goto out;
1753
1754 err = put_user(fd2, &usockvec[1]);
1755 if (err)
1756 goto out;
1757
1758 /*
1759 * Obtain the first socket and check if the underlying protocol
1760 * supports the socketpair call.
1761 */
1762
1763 err = sock_create(family, type, protocol, &sock1);
1764 if (unlikely(err < 0))
1765 goto out;
1766
1767 err = sock_create(family, type, protocol, &sock2);
1768 if (unlikely(err < 0)) {
1769 sock_release(sock1);
1770 goto out;
1771 }
1772
1773 err = security_socket_socketpair(sock1, sock2);
1774 if (unlikely(err)) {
1775 sock_release(sock2);
1776 sock_release(sock1);
1777 goto out;
1778 }
1779
1780 err = READ_ONCE(sock1->ops)->socketpair(sock1, sock2);
1781 if (unlikely(err < 0)) {
1782 sock_release(sock2);
1783 sock_release(sock1);
1784 goto out;
1785 }
1786
1787 newfile1 = sock_alloc_file(sock1, flags, NULL);
1788 if (IS_ERR(newfile1)) {
1789 err = PTR_ERR(newfile1);
1790 sock_release(sock2);
1791 goto out;
1792 }
1793
1794 newfile2 = sock_alloc_file(sock2, flags, NULL);
1795 if (IS_ERR(newfile2)) {
1796 err = PTR_ERR(newfile2);
1797 fput(newfile1);
1798 goto out;
1799 }
1800
1801 audit_fd_pair(fd1, fd2);
1802
1803 fd_install(fd1, newfile1);
1804 fd_install(fd2, newfile2);
1805 return 0;
1806
1807out:
1808 put_unused_fd(fd2);
1809 put_unused_fd(fd1);
1810 return err;
1811}
1812
1813SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1814 int __user *, usockvec)
1815{
1816 return __sys_socketpair(family, type, protocol, usockvec);
1817}
1818
1819int __sys_bind_socket(struct socket *sock, struct sockaddr_storage *address,
1820 int addrlen)
1821{
1822 int err;
1823
1824 err = security_socket_bind(sock, (struct sockaddr *)address,
1825 addrlen);
1826 if (!err)
1827 err = READ_ONCE(sock->ops)->bind(sock,
1828 (struct sockaddr *)address,
1829 addrlen);
1830 return err;
1831}
1832
1833/*
1834 * Bind a name to a socket. Nothing much to do here since it's
1835 * the protocol's responsibility to handle the local address.
1836 *
1837 * We move the socket address to kernel space before we call
1838 * the protocol layer (having also checked the address is ok).
1839 */
1840
1841int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1842{
1843 struct socket *sock;
1844 struct sockaddr_storage address;
1845 CLASS(fd, f)(fd);
1846 int err;
1847
1848 if (fd_empty(f))
1849 return -EBADF;
1850 sock = sock_from_file(fd_file(f));
1851 if (unlikely(!sock))
1852 return -ENOTSOCK;
1853
1854 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1855 if (unlikely(err))
1856 return err;
1857
1858 return __sys_bind_socket(sock, &address, addrlen);
1859}
1860
1861SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1862{
1863 return __sys_bind(fd, umyaddr, addrlen);
1864}
1865
1866/*
1867 * Perform a listen. Basically, we allow the protocol to do anything
1868 * necessary for a listen, and if that works, we mark the socket as
1869 * ready for listening.
1870 */
1871int __sys_listen_socket(struct socket *sock, int backlog)
1872{
1873 int somaxconn, err;
1874
1875 somaxconn = READ_ONCE(sock_net(sock->sk)->core.sysctl_somaxconn);
1876 if ((unsigned int)backlog > somaxconn)
1877 backlog = somaxconn;
1878
1879 err = security_socket_listen(sock, backlog);
1880 if (!err)
1881 err = READ_ONCE(sock->ops)->listen(sock, backlog);
1882 return err;
1883}
1884
1885int __sys_listen(int fd, int backlog)
1886{
1887 CLASS(fd, f)(fd);
1888 struct socket *sock;
1889
1890 if (fd_empty(f))
1891 return -EBADF;
1892 sock = sock_from_file(fd_file(f));
1893 if (unlikely(!sock))
1894 return -ENOTSOCK;
1895
1896 return __sys_listen_socket(sock, backlog);
1897}
1898
1899SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1900{
1901 return __sys_listen(fd, backlog);
1902}
1903
1904struct file *do_accept(struct file *file, struct proto_accept_arg *arg,
1905 struct sockaddr __user *upeer_sockaddr,
1906 int __user *upeer_addrlen, int flags)
1907{
1908 struct socket *sock, *newsock;
1909 struct file *newfile;
1910 int err, len;
1911 struct sockaddr_storage address;
1912 const struct proto_ops *ops;
1913
1914 sock = sock_from_file(file);
1915 if (!sock)
1916 return ERR_PTR(-ENOTSOCK);
1917
1918 newsock = sock_alloc();
1919 if (!newsock)
1920 return ERR_PTR(-ENFILE);
1921 ops = READ_ONCE(sock->ops);
1922
1923 newsock->type = sock->type;
1924 newsock->ops = ops;
1925
1926 /*
1927 * We don't need try_module_get here, as the listening socket (sock)
1928 * has the protocol module (sock->ops->owner) held.
1929 */
1930 __module_get(ops->owner);
1931
1932 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1933 if (IS_ERR(newfile))
1934 return newfile;
1935
1936 err = security_socket_accept(sock, newsock);
1937 if (err)
1938 goto out_fd;
1939
1940 arg->flags |= sock->file->f_flags;
1941 err = ops->accept(sock, newsock, arg);
1942 if (err < 0)
1943 goto out_fd;
1944
1945 if (upeer_sockaddr) {
1946 len = ops->getname(newsock, (struct sockaddr *)&address, 2);
1947 if (len < 0) {
1948 err = -ECONNABORTED;
1949 goto out_fd;
1950 }
1951 err = move_addr_to_user(&address,
1952 len, upeer_sockaddr, upeer_addrlen);
1953 if (err < 0)
1954 goto out_fd;
1955 }
1956
1957 /* File flags are not inherited via accept() unlike another OSes. */
1958 return newfile;
1959out_fd:
1960 fput(newfile);
1961 return ERR_PTR(err);
1962}
1963
1964static int __sys_accept4_file(struct file *file, struct sockaddr __user *upeer_sockaddr,
1965 int __user *upeer_addrlen, int flags)
1966{
1967 struct proto_accept_arg arg = { };
1968 struct file *newfile;
1969 int newfd;
1970
1971 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1972 return -EINVAL;
1973
1974 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1975 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1976
1977 newfd = get_unused_fd_flags(flags);
1978 if (unlikely(newfd < 0))
1979 return newfd;
1980
1981 newfile = do_accept(file, &arg, upeer_sockaddr, upeer_addrlen,
1982 flags);
1983 if (IS_ERR(newfile)) {
1984 put_unused_fd(newfd);
1985 return PTR_ERR(newfile);
1986 }
1987 fd_install(newfd, newfile);
1988 return newfd;
1989}
1990
1991/*
1992 * For accept, we attempt to create a new socket, set up the link
1993 * with the client, wake up the client, then return the new
1994 * connected fd. We collect the address of the connector in kernel
1995 * space and move it to user at the very end. This is unclean because
1996 * we open the socket then return an error.
1997 *
1998 * 1003.1g adds the ability to recvmsg() to query connection pending
1999 * status to recvmsg. We need to add that support in a way thats
2000 * clean when we restructure accept also.
2001 */
2002
2003int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
2004 int __user *upeer_addrlen, int flags)
2005{
2006 CLASS(fd, f)(fd);
2007
2008 if (fd_empty(f))
2009 return -EBADF;
2010 return __sys_accept4_file(fd_file(f), upeer_sockaddr,
2011 upeer_addrlen, flags);
2012}
2013
2014SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
2015 int __user *, upeer_addrlen, int, flags)
2016{
2017 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
2018}
2019
2020SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
2021 int __user *, upeer_addrlen)
2022{
2023 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
2024}
2025
2026/*
2027 * Attempt to connect to a socket with the server address. The address
2028 * is in user space so we verify it is OK and move it to kernel space.
2029 *
2030 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
2031 * break bindings
2032 *
2033 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
2034 * other SEQPACKET protocols that take time to connect() as it doesn't
2035 * include the -EINPROGRESS status for such sockets.
2036 */
2037
2038int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
2039 int addrlen, int file_flags)
2040{
2041 struct socket *sock;
2042 int err;
2043
2044 sock = sock_from_file(file);
2045 if (!sock) {
2046 err = -ENOTSOCK;
2047 goto out;
2048 }
2049
2050 err =
2051 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
2052 if (err)
2053 goto out;
2054
2055 err = READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)address,
2056 addrlen, sock->file->f_flags | file_flags);
2057out:
2058 return err;
2059}
2060
2061int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
2062{
2063 struct sockaddr_storage address;
2064 CLASS(fd, f)(fd);
2065 int ret;
2066
2067 if (fd_empty(f))
2068 return -EBADF;
2069
2070 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
2071 if (ret)
2072 return ret;
2073
2074 return __sys_connect_file(fd_file(f), &address, addrlen, 0);
2075}
2076
2077SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
2078 int, addrlen)
2079{
2080 return __sys_connect(fd, uservaddr, addrlen);
2081}
2082
2083/*
2084 * Get the local address ('name') of a socket object. Move the obtained
2085 * name to user space.
2086 */
2087
2088int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
2089 int __user *usockaddr_len)
2090{
2091 struct socket *sock;
2092 struct sockaddr_storage address;
2093 CLASS(fd, f)(fd);
2094 int err;
2095
2096 if (fd_empty(f))
2097 return -EBADF;
2098 sock = sock_from_file(fd_file(f));
2099 if (unlikely(!sock))
2100 return -ENOTSOCK;
2101
2102 err = security_socket_getsockname(sock);
2103 if (err)
2104 return err;
2105
2106 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
2107 if (err < 0)
2108 return err;
2109
2110 /* "err" is actually length in this case */
2111 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2112}
2113
2114SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
2115 int __user *, usockaddr_len)
2116{
2117 return __sys_getsockname(fd, usockaddr, usockaddr_len);
2118}
2119
2120/*
2121 * Get the remote address ('name') of a socket object. Move the obtained
2122 * name to user space.
2123 */
2124
2125int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
2126 int __user *usockaddr_len)
2127{
2128 struct socket *sock;
2129 struct sockaddr_storage address;
2130 CLASS(fd, f)(fd);
2131 int err;
2132
2133 if (fd_empty(f))
2134 return -EBADF;
2135 sock = sock_from_file(fd_file(f));
2136 if (unlikely(!sock))
2137 return -ENOTSOCK;
2138
2139 err = security_socket_getpeername(sock);
2140 if (err)
2141 return err;
2142
2143 err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 1);
2144 if (err < 0)
2145 return err;
2146
2147 /* "err" is actually length in this case */
2148 return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
2149}
2150
2151SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
2152 int __user *, usockaddr_len)
2153{
2154 return __sys_getpeername(fd, usockaddr, usockaddr_len);
2155}
2156
2157/*
2158 * Send a datagram to a given address. We move the address into kernel
2159 * space and check the user space data area is readable before invoking
2160 * the protocol.
2161 */
2162int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
2163 struct sockaddr __user *addr, int addr_len)
2164{
2165 struct socket *sock;
2166 struct sockaddr_storage address;
2167 int err;
2168 struct msghdr msg;
2169
2170 err = import_ubuf(ITER_SOURCE, buff, len, &msg.msg_iter);
2171 if (unlikely(err))
2172 return err;
2173
2174 CLASS(fd, f)(fd);
2175 if (fd_empty(f))
2176 return -EBADF;
2177 sock = sock_from_file(fd_file(f));
2178 if (unlikely(!sock))
2179 return -ENOTSOCK;
2180
2181 msg.msg_name = NULL;
2182 msg.msg_control = NULL;
2183 msg.msg_controllen = 0;
2184 msg.msg_namelen = 0;
2185 msg.msg_ubuf = NULL;
2186 if (addr) {
2187 err = move_addr_to_kernel(addr, addr_len, &address);
2188 if (err < 0)
2189 return err;
2190 msg.msg_name = (struct sockaddr *)&address;
2191 msg.msg_namelen = addr_len;
2192 }
2193 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2194 if (sock->file->f_flags & O_NONBLOCK)
2195 flags |= MSG_DONTWAIT;
2196 msg.msg_flags = flags;
2197 return __sock_sendmsg(sock, &msg);
2198}
2199
2200SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2201 unsigned int, flags, struct sockaddr __user *, addr,
2202 int, addr_len)
2203{
2204 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2205}
2206
2207/*
2208 * Send a datagram down a socket.
2209 */
2210
2211SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2212 unsigned int, flags)
2213{
2214 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2215}
2216
2217/*
2218 * Receive a frame from the socket and optionally record the address of the
2219 * sender. We verify the buffers are writable and if needed move the
2220 * sender address from kernel to user space.
2221 */
2222int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2223 struct sockaddr __user *addr, int __user *addr_len)
2224{
2225 struct sockaddr_storage address;
2226 struct msghdr msg = {
2227 /* Save some cycles and don't copy the address if not needed */
2228 .msg_name = addr ? (struct sockaddr *)&address : NULL,
2229 };
2230 struct socket *sock;
2231 int err, err2;
2232
2233 err = import_ubuf(ITER_DEST, ubuf, size, &msg.msg_iter);
2234 if (unlikely(err))
2235 return err;
2236
2237 CLASS(fd, f)(fd);
2238
2239 if (fd_empty(f))
2240 return -EBADF;
2241 sock = sock_from_file(fd_file(f));
2242 if (unlikely(!sock))
2243 return -ENOTSOCK;
2244
2245 if (sock->file->f_flags & O_NONBLOCK)
2246 flags |= MSG_DONTWAIT;
2247 err = sock_recvmsg(sock, &msg, flags);
2248
2249 if (err >= 0 && addr != NULL) {
2250 err2 = move_addr_to_user(&address,
2251 msg.msg_namelen, addr, addr_len);
2252 if (err2 < 0)
2253 err = err2;
2254 }
2255 return err;
2256}
2257
2258SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2259 unsigned int, flags, struct sockaddr __user *, addr,
2260 int __user *, addr_len)
2261{
2262 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2263}
2264
2265/*
2266 * Receive a datagram from a socket.
2267 */
2268
2269SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2270 unsigned int, flags)
2271{
2272 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2273}
2274
2275static bool sock_use_custom_sol_socket(const struct socket *sock)
2276{
2277 return test_bit(SOCK_CUSTOM_SOCKOPT, &sock->flags);
2278}
2279
2280int do_sock_setsockopt(struct socket *sock, bool compat, int level,
2281 int optname, sockptr_t optval, int optlen)
2282{
2283 const struct proto_ops *ops;
2284 char *kernel_optval = NULL;
2285 int err;
2286
2287 if (optlen < 0)
2288 return -EINVAL;
2289
2290 err = security_socket_setsockopt(sock, level, optname);
2291 if (err)
2292 goto out_put;
2293
2294 if (!compat)
2295 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2296 optval, &optlen,
2297 &kernel_optval);
2298 if (err < 0)
2299 goto out_put;
2300 if (err > 0) {
2301 err = 0;
2302 goto out_put;
2303 }
2304
2305 if (kernel_optval)
2306 optval = KERNEL_SOCKPTR(kernel_optval);
2307 ops = READ_ONCE(sock->ops);
2308 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2309 err = sock_setsockopt(sock, level, optname, optval, optlen);
2310 else if (unlikely(!ops->setsockopt))
2311 err = -EOPNOTSUPP;
2312 else
2313 err = ops->setsockopt(sock, level, optname, optval,
2314 optlen);
2315 kfree(kernel_optval);
2316out_put:
2317 return err;
2318}
2319EXPORT_SYMBOL(do_sock_setsockopt);
2320
2321/* Set a socket option. Because we don't know the option lengths we have
2322 * to pass the user mode parameter for the protocols to sort out.
2323 */
2324int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2325 int optlen)
2326{
2327 sockptr_t optval = USER_SOCKPTR(user_optval);
2328 bool compat = in_compat_syscall();
2329 struct socket *sock;
2330 CLASS(fd, f)(fd);
2331
2332 if (fd_empty(f))
2333 return -EBADF;
2334 sock = sock_from_file(fd_file(f));
2335 if (unlikely(!sock))
2336 return -ENOTSOCK;
2337
2338 return do_sock_setsockopt(sock, compat, level, optname, optval, optlen);
2339}
2340
2341SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2342 char __user *, optval, int, optlen)
2343{
2344 return __sys_setsockopt(fd, level, optname, optval, optlen);
2345}
2346
2347INDIRECT_CALLABLE_DECLARE(bool tcp_bpf_bypass_getsockopt(int level,
2348 int optname));
2349
2350int do_sock_getsockopt(struct socket *sock, bool compat, int level,
2351 int optname, sockptr_t optval, sockptr_t optlen)
2352{
2353 int max_optlen __maybe_unused = 0;
2354 const struct proto_ops *ops;
2355 int err;
2356
2357 err = security_socket_getsockopt(sock, level, optname);
2358 if (err)
2359 return err;
2360
2361 if (!compat)
2362 copy_from_sockptr(&max_optlen, optlen, sizeof(int));
2363
2364 ops = READ_ONCE(sock->ops);
2365 if (level == SOL_SOCKET) {
2366 err = sk_getsockopt(sock->sk, level, optname, optval, optlen);
2367 } else if (unlikely(!ops->getsockopt)) {
2368 err = -EOPNOTSUPP;
2369 } else {
2370 if (WARN_ONCE(optval.is_kernel || optlen.is_kernel,
2371 "Invalid argument type"))
2372 return -EOPNOTSUPP;
2373
2374 err = ops->getsockopt(sock, level, optname, optval.user,
2375 optlen.user);
2376 }
2377
2378 if (!compat)
2379 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2380 optval, optlen, max_optlen,
2381 err);
2382
2383 return err;
2384}
2385EXPORT_SYMBOL(do_sock_getsockopt);
2386
2387/*
2388 * Get a socket option. Because we don't know the option lengths we have
2389 * to pass a user mode parameter for the protocols to sort out.
2390 */
2391int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2392 int __user *optlen)
2393{
2394 struct socket *sock;
2395 CLASS(fd, f)(fd);
2396
2397 if (fd_empty(f))
2398 return -EBADF;
2399 sock = sock_from_file(fd_file(f));
2400 if (unlikely(!sock))
2401 return -ENOTSOCK;
2402
2403 return do_sock_getsockopt(sock, in_compat_syscall(), level, optname,
2404 USER_SOCKPTR(optval), USER_SOCKPTR(optlen));
2405}
2406
2407SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2408 char __user *, optval, int __user *, optlen)
2409{
2410 return __sys_getsockopt(fd, level, optname, optval, optlen);
2411}
2412
2413/*
2414 * Shutdown a socket.
2415 */
2416
2417int __sys_shutdown_sock(struct socket *sock, int how)
2418{
2419 int err;
2420
2421 err = security_socket_shutdown(sock, how);
2422 if (!err)
2423 err = READ_ONCE(sock->ops)->shutdown(sock, how);
2424
2425 return err;
2426}
2427
2428int __sys_shutdown(int fd, int how)
2429{
2430 struct socket *sock;
2431 CLASS(fd, f)(fd);
2432
2433 if (fd_empty(f))
2434 return -EBADF;
2435 sock = sock_from_file(fd_file(f));
2436 if (unlikely(!sock))
2437 return -ENOTSOCK;
2438
2439 return __sys_shutdown_sock(sock, how);
2440}
2441
2442SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2443{
2444 return __sys_shutdown(fd, how);
2445}
2446
2447/* A couple of helpful macros for getting the address of the 32/64 bit
2448 * fields which are the same type (int / unsigned) on our platforms.
2449 */
2450#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2451#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2452#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2453
2454struct used_address {
2455 struct sockaddr_storage name;
2456 unsigned int name_len;
2457};
2458
2459int __copy_msghdr(struct msghdr *kmsg,
2460 struct user_msghdr *msg,
2461 struct sockaddr __user **save_addr)
2462{
2463 ssize_t err;
2464
2465 kmsg->msg_control_is_user = true;
2466 kmsg->msg_get_inq = 0;
2467 kmsg->msg_control_user = msg->msg_control;
2468 kmsg->msg_controllen = msg->msg_controllen;
2469 kmsg->msg_flags = msg->msg_flags;
2470
2471 kmsg->msg_namelen = msg->msg_namelen;
2472 if (!msg->msg_name)
2473 kmsg->msg_namelen = 0;
2474
2475 if (kmsg->msg_namelen < 0)
2476 return -EINVAL;
2477
2478 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2479 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2480
2481 if (save_addr)
2482 *save_addr = msg->msg_name;
2483
2484 if (msg->msg_name && kmsg->msg_namelen) {
2485 if (!save_addr) {
2486 err = move_addr_to_kernel(msg->msg_name,
2487 kmsg->msg_namelen,
2488 kmsg->msg_name);
2489 if (err < 0)
2490 return err;
2491 }
2492 } else {
2493 kmsg->msg_name = NULL;
2494 kmsg->msg_namelen = 0;
2495 }
2496
2497 if (msg->msg_iovlen > UIO_MAXIOV)
2498 return -EMSGSIZE;
2499
2500 kmsg->msg_iocb = NULL;
2501 kmsg->msg_ubuf = NULL;
2502 return 0;
2503}
2504
2505static int copy_msghdr_from_user(struct msghdr *kmsg,
2506 struct user_msghdr __user *umsg,
2507 struct sockaddr __user **save_addr,
2508 struct iovec **iov)
2509{
2510 struct user_msghdr msg;
2511 ssize_t err;
2512
2513 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2514 return -EFAULT;
2515
2516 err = __copy_msghdr(kmsg, &msg, save_addr);
2517 if (err)
2518 return err;
2519
2520 err = import_iovec(save_addr ? ITER_DEST : ITER_SOURCE,
2521 msg.msg_iov, msg.msg_iovlen,
2522 UIO_FASTIOV, iov, &kmsg->msg_iter);
2523 return err < 0 ? err : 0;
2524}
2525
2526static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2527 unsigned int flags, struct used_address *used_address,
2528 unsigned int allowed_msghdr_flags)
2529{
2530 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2531 __aligned(sizeof(__kernel_size_t));
2532 /* 20 is size of ipv6_pktinfo */
2533 unsigned char *ctl_buf = ctl;
2534 int ctl_len;
2535 ssize_t err;
2536
2537 err = -ENOBUFS;
2538
2539 if (msg_sys->msg_controllen > INT_MAX)
2540 goto out;
2541 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2542 ctl_len = msg_sys->msg_controllen;
2543 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2544 err =
2545 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2546 sizeof(ctl));
2547 if (err)
2548 goto out;
2549 ctl_buf = msg_sys->msg_control;
2550 ctl_len = msg_sys->msg_controllen;
2551 } else if (ctl_len) {
2552 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2553 CMSG_ALIGN(sizeof(struct cmsghdr)));
2554 if (ctl_len > sizeof(ctl)) {
2555 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2556 if (ctl_buf == NULL)
2557 goto out;
2558 }
2559 err = -EFAULT;
2560 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2561 goto out_freectl;
2562 msg_sys->msg_control = ctl_buf;
2563 msg_sys->msg_control_is_user = false;
2564 }
2565 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
2566 msg_sys->msg_flags = flags;
2567
2568 if (sock->file->f_flags & O_NONBLOCK)
2569 msg_sys->msg_flags |= MSG_DONTWAIT;
2570 /*
2571 * If this is sendmmsg() and current destination address is same as
2572 * previously succeeded address, omit asking LSM's decision.
2573 * used_address->name_len is initialized to UINT_MAX so that the first
2574 * destination address never matches.
2575 */
2576 if (used_address && msg_sys->msg_name &&
2577 used_address->name_len == msg_sys->msg_namelen &&
2578 !memcmp(&used_address->name, msg_sys->msg_name,
2579 used_address->name_len)) {
2580 err = sock_sendmsg_nosec(sock, msg_sys);
2581 goto out_freectl;
2582 }
2583 err = __sock_sendmsg(sock, msg_sys);
2584 /*
2585 * If this is sendmmsg() and sending to current destination address was
2586 * successful, remember it.
2587 */
2588 if (used_address && err >= 0) {
2589 used_address->name_len = msg_sys->msg_namelen;
2590 if (msg_sys->msg_name)
2591 memcpy(&used_address->name, msg_sys->msg_name,
2592 used_address->name_len);
2593 }
2594
2595out_freectl:
2596 if (ctl_buf != ctl)
2597 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2598out:
2599 return err;
2600}
2601
2602static int sendmsg_copy_msghdr(struct msghdr *msg,
2603 struct user_msghdr __user *umsg, unsigned flags,
2604 struct iovec **iov)
2605{
2606 int err;
2607
2608 if (flags & MSG_CMSG_COMPAT) {
2609 struct compat_msghdr __user *msg_compat;
2610
2611 msg_compat = (struct compat_msghdr __user *) umsg;
2612 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2613 } else {
2614 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2615 }
2616 if (err < 0)
2617 return err;
2618
2619 return 0;
2620}
2621
2622static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2623 struct msghdr *msg_sys, unsigned int flags,
2624 struct used_address *used_address,
2625 unsigned int allowed_msghdr_flags)
2626{
2627 struct sockaddr_storage address;
2628 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2629 ssize_t err;
2630
2631 msg_sys->msg_name = &address;
2632
2633 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2634 if (err < 0)
2635 return err;
2636
2637 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2638 allowed_msghdr_flags);
2639 kfree(iov);
2640 return err;
2641}
2642
2643/*
2644 * BSD sendmsg interface
2645 */
2646long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2647 unsigned int flags)
2648{
2649 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2650}
2651
2652long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2653 bool forbid_cmsg_compat)
2654{
2655 struct msghdr msg_sys;
2656 struct socket *sock;
2657
2658 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2659 return -EINVAL;
2660
2661 CLASS(fd, f)(fd);
2662
2663 if (fd_empty(f))
2664 return -EBADF;
2665 sock = sock_from_file(fd_file(f));
2666 if (unlikely(!sock))
2667 return -ENOTSOCK;
2668
2669 return ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2670}
2671
2672SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2673{
2674 return __sys_sendmsg(fd, msg, flags, true);
2675}
2676
2677/*
2678 * Linux sendmmsg interface
2679 */
2680
2681int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2682 unsigned int flags, bool forbid_cmsg_compat)
2683{
2684 int err, datagrams;
2685 struct socket *sock;
2686 struct mmsghdr __user *entry;
2687 struct compat_mmsghdr __user *compat_entry;
2688 struct msghdr msg_sys;
2689 struct used_address used_address;
2690 unsigned int oflags = flags;
2691
2692 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2693 return -EINVAL;
2694
2695 if (vlen > UIO_MAXIOV)
2696 vlen = UIO_MAXIOV;
2697
2698 datagrams = 0;
2699
2700 CLASS(fd, f)(fd);
2701
2702 if (fd_empty(f))
2703 return -EBADF;
2704 sock = sock_from_file(fd_file(f));
2705 if (unlikely(!sock))
2706 return -ENOTSOCK;
2707
2708 used_address.name_len = UINT_MAX;
2709 entry = mmsg;
2710 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2711 err = 0;
2712 flags |= MSG_BATCH;
2713
2714 while (datagrams < vlen) {
2715 if (datagrams == vlen - 1)
2716 flags = oflags;
2717
2718 if (MSG_CMSG_COMPAT & flags) {
2719 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2720 &msg_sys, flags, &used_address, MSG_EOR);
2721 if (err < 0)
2722 break;
2723 err = __put_user(err, &compat_entry->msg_len);
2724 ++compat_entry;
2725 } else {
2726 err = ___sys_sendmsg(sock,
2727 (struct user_msghdr __user *)entry,
2728 &msg_sys, flags, &used_address, MSG_EOR);
2729 if (err < 0)
2730 break;
2731 err = put_user(err, &entry->msg_len);
2732 ++entry;
2733 }
2734
2735 if (err)
2736 break;
2737 ++datagrams;
2738 if (msg_data_left(&msg_sys))
2739 break;
2740 cond_resched();
2741 }
2742
2743 /* We only return an error if no datagrams were able to be sent */
2744 if (datagrams != 0)
2745 return datagrams;
2746
2747 return err;
2748}
2749
2750SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2751 unsigned int, vlen, unsigned int, flags)
2752{
2753 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2754}
2755
2756static int recvmsg_copy_msghdr(struct msghdr *msg,
2757 struct user_msghdr __user *umsg, unsigned flags,
2758 struct sockaddr __user **uaddr,
2759 struct iovec **iov)
2760{
2761 ssize_t err;
2762
2763 if (MSG_CMSG_COMPAT & flags) {
2764 struct compat_msghdr __user *msg_compat;
2765
2766 msg_compat = (struct compat_msghdr __user *) umsg;
2767 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2768 } else {
2769 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2770 }
2771 if (err < 0)
2772 return err;
2773
2774 return 0;
2775}
2776
2777static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2778 struct user_msghdr __user *msg,
2779 struct sockaddr __user *uaddr,
2780 unsigned int flags, int nosec)
2781{
2782 struct compat_msghdr __user *msg_compat =
2783 (struct compat_msghdr __user *) msg;
2784 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2785 struct sockaddr_storage addr;
2786 unsigned long cmsg_ptr;
2787 int len;
2788 ssize_t err;
2789
2790 msg_sys->msg_name = &addr;
2791 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2792 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2793
2794 /* We assume all kernel code knows the size of sockaddr_storage */
2795 msg_sys->msg_namelen = 0;
2796
2797 if (sock->file->f_flags & O_NONBLOCK)
2798 flags |= MSG_DONTWAIT;
2799
2800 if (unlikely(nosec))
2801 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2802 else
2803 err = sock_recvmsg(sock, msg_sys, flags);
2804
2805 if (err < 0)
2806 goto out;
2807 len = err;
2808
2809 if (uaddr != NULL) {
2810 err = move_addr_to_user(&addr,
2811 msg_sys->msg_namelen, uaddr,
2812 uaddr_len);
2813 if (err < 0)
2814 goto out;
2815 }
2816 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2817 COMPAT_FLAGS(msg));
2818 if (err)
2819 goto out;
2820 if (MSG_CMSG_COMPAT & flags)
2821 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2822 &msg_compat->msg_controllen);
2823 else
2824 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2825 &msg->msg_controllen);
2826 if (err)
2827 goto out;
2828 err = len;
2829out:
2830 return err;
2831}
2832
2833static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2834 struct msghdr *msg_sys, unsigned int flags, int nosec)
2835{
2836 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2837 /* user mode address pointers */
2838 struct sockaddr __user *uaddr;
2839 ssize_t err;
2840
2841 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2842 if (err < 0)
2843 return err;
2844
2845 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2846 kfree(iov);
2847 return err;
2848}
2849
2850/*
2851 * BSD recvmsg interface
2852 */
2853
2854long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2855 struct user_msghdr __user *umsg,
2856 struct sockaddr __user *uaddr, unsigned int flags)
2857{
2858 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2859}
2860
2861long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2862 bool forbid_cmsg_compat)
2863{
2864 struct msghdr msg_sys;
2865 struct socket *sock;
2866
2867 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2868 return -EINVAL;
2869
2870 CLASS(fd, f)(fd);
2871
2872 if (fd_empty(f))
2873 return -EBADF;
2874 sock = sock_from_file(fd_file(f));
2875 if (unlikely(!sock))
2876 return -ENOTSOCK;
2877
2878 return ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2879}
2880
2881SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2882 unsigned int, flags)
2883{
2884 return __sys_recvmsg(fd, msg, flags, true);
2885}
2886
2887/*
2888 * Linux recvmmsg interface
2889 */
2890
2891static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2892 unsigned int vlen, unsigned int flags,
2893 struct timespec64 *timeout)
2894{
2895 int err = 0, datagrams;
2896 struct socket *sock;
2897 struct mmsghdr __user *entry;
2898 struct compat_mmsghdr __user *compat_entry;
2899 struct msghdr msg_sys;
2900 struct timespec64 end_time;
2901 struct timespec64 timeout64;
2902
2903 if (timeout &&
2904 poll_select_set_timeout(&end_time, timeout->tv_sec,
2905 timeout->tv_nsec))
2906 return -EINVAL;
2907
2908 datagrams = 0;
2909
2910 CLASS(fd, f)(fd);
2911
2912 if (fd_empty(f))
2913 return -EBADF;
2914 sock = sock_from_file(fd_file(f));
2915 if (unlikely(!sock))
2916 return -ENOTSOCK;
2917
2918 if (likely(!(flags & MSG_ERRQUEUE))) {
2919 err = sock_error(sock->sk);
2920 if (err)
2921 return err;
2922 }
2923
2924 entry = mmsg;
2925 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2926
2927 while (datagrams < vlen) {
2928 /*
2929 * No need to ask LSM for more than the first datagram.
2930 */
2931 if (MSG_CMSG_COMPAT & flags) {
2932 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2933 &msg_sys, flags & ~MSG_WAITFORONE,
2934 datagrams);
2935 if (err < 0)
2936 break;
2937 err = __put_user(err, &compat_entry->msg_len);
2938 ++compat_entry;
2939 } else {
2940 err = ___sys_recvmsg(sock,
2941 (struct user_msghdr __user *)entry,
2942 &msg_sys, flags & ~MSG_WAITFORONE,
2943 datagrams);
2944 if (err < 0)
2945 break;
2946 err = put_user(err, &entry->msg_len);
2947 ++entry;
2948 }
2949
2950 if (err)
2951 break;
2952 ++datagrams;
2953
2954 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2955 if (flags & MSG_WAITFORONE)
2956 flags |= MSG_DONTWAIT;
2957
2958 if (timeout) {
2959 ktime_get_ts64(&timeout64);
2960 *timeout = timespec64_sub(end_time, timeout64);
2961 if (timeout->tv_sec < 0) {
2962 timeout->tv_sec = timeout->tv_nsec = 0;
2963 break;
2964 }
2965
2966 /* Timeout, return less than vlen datagrams */
2967 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2968 break;
2969 }
2970
2971 /* Out of band data, return right away */
2972 if (msg_sys.msg_flags & MSG_OOB)
2973 break;
2974 cond_resched();
2975 }
2976
2977 if (err == 0)
2978 return datagrams;
2979
2980 if (datagrams == 0)
2981 return err;
2982
2983 /*
2984 * We may return less entries than requested (vlen) if the
2985 * sock is non block and there aren't enough datagrams...
2986 */
2987 if (err != -EAGAIN) {
2988 /*
2989 * ... or if recvmsg returns an error after we
2990 * received some datagrams, where we record the
2991 * error to return on the next call or if the
2992 * app asks about it using getsockopt(SO_ERROR).
2993 */
2994 WRITE_ONCE(sock->sk->sk_err, -err);
2995 }
2996 return datagrams;
2997}
2998
2999int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
3000 unsigned int vlen, unsigned int flags,
3001 struct __kernel_timespec __user *timeout,
3002 struct old_timespec32 __user *timeout32)
3003{
3004 int datagrams;
3005 struct timespec64 timeout_sys;
3006
3007 if (timeout && get_timespec64(&timeout_sys, timeout))
3008 return -EFAULT;
3009
3010 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
3011 return -EFAULT;
3012
3013 if (!timeout && !timeout32)
3014 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
3015
3016 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
3017
3018 if (datagrams <= 0)
3019 return datagrams;
3020
3021 if (timeout && put_timespec64(&timeout_sys, timeout))
3022 datagrams = -EFAULT;
3023
3024 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
3025 datagrams = -EFAULT;
3026
3027 return datagrams;
3028}
3029
3030SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
3031 unsigned int, vlen, unsigned int, flags,
3032 struct __kernel_timespec __user *, timeout)
3033{
3034 if (flags & MSG_CMSG_COMPAT)
3035 return -EINVAL;
3036
3037 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
3038}
3039
3040#ifdef CONFIG_COMPAT_32BIT_TIME
3041SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
3042 unsigned int, vlen, unsigned int, flags,
3043 struct old_timespec32 __user *, timeout)
3044{
3045 if (flags & MSG_CMSG_COMPAT)
3046 return -EINVAL;
3047
3048 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
3049}
3050#endif
3051
3052#ifdef __ARCH_WANT_SYS_SOCKETCALL
3053/* Argument list sizes for sys_socketcall */
3054#define AL(x) ((x) * sizeof(unsigned long))
3055static const unsigned char nargs[21] = {
3056 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
3057 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
3058 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
3059 AL(4), AL(5), AL(4)
3060};
3061
3062#undef AL
3063
3064/*
3065 * System call vectors.
3066 *
3067 * Argument checking cleaned up. Saved 20% in size.
3068 * This function doesn't need to set the kernel lock because
3069 * it is set by the callees.
3070 */
3071
3072SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
3073{
3074 unsigned long a[AUDITSC_ARGS];
3075 unsigned long a0, a1;
3076 int err;
3077 unsigned int len;
3078
3079 if (call < 1 || call > SYS_SENDMMSG)
3080 return -EINVAL;
3081 call = array_index_nospec(call, SYS_SENDMMSG + 1);
3082
3083 len = nargs[call];
3084 if (len > sizeof(a))
3085 return -EINVAL;
3086
3087 /* copy_from_user should be SMP safe. */
3088 if (copy_from_user(a, args, len))
3089 return -EFAULT;
3090
3091 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
3092 if (err)
3093 return err;
3094
3095 a0 = a[0];
3096 a1 = a[1];
3097
3098 switch (call) {
3099 case SYS_SOCKET:
3100 err = __sys_socket(a0, a1, a[2]);
3101 break;
3102 case SYS_BIND:
3103 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
3104 break;
3105 case SYS_CONNECT:
3106 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
3107 break;
3108 case SYS_LISTEN:
3109 err = __sys_listen(a0, a1);
3110 break;
3111 case SYS_ACCEPT:
3112 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3113 (int __user *)a[2], 0);
3114 break;
3115 case SYS_GETSOCKNAME:
3116 err =
3117 __sys_getsockname(a0, (struct sockaddr __user *)a1,
3118 (int __user *)a[2]);
3119 break;
3120 case SYS_GETPEERNAME:
3121 err =
3122 __sys_getpeername(a0, (struct sockaddr __user *)a1,
3123 (int __user *)a[2]);
3124 break;
3125 case SYS_SOCKETPAIR:
3126 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
3127 break;
3128 case SYS_SEND:
3129 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3130 NULL, 0);
3131 break;
3132 case SYS_SENDTO:
3133 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
3134 (struct sockaddr __user *)a[4], a[5]);
3135 break;
3136 case SYS_RECV:
3137 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3138 NULL, NULL);
3139 break;
3140 case SYS_RECVFROM:
3141 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
3142 (struct sockaddr __user *)a[4],
3143 (int __user *)a[5]);
3144 break;
3145 case SYS_SHUTDOWN:
3146 err = __sys_shutdown(a0, a1);
3147 break;
3148 case SYS_SETSOCKOPT:
3149 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
3150 a[4]);
3151 break;
3152 case SYS_GETSOCKOPT:
3153 err =
3154 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
3155 (int __user *)a[4]);
3156 break;
3157 case SYS_SENDMSG:
3158 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
3159 a[2], true);
3160 break;
3161 case SYS_SENDMMSG:
3162 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
3163 a[3], true);
3164 break;
3165 case SYS_RECVMSG:
3166 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
3167 a[2], true);
3168 break;
3169 case SYS_RECVMMSG:
3170 if (IS_ENABLED(CONFIG_64BIT))
3171 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3172 a[2], a[3],
3173 (struct __kernel_timespec __user *)a[4],
3174 NULL);
3175 else
3176 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
3177 a[2], a[3], NULL,
3178 (struct old_timespec32 __user *)a[4]);
3179 break;
3180 case SYS_ACCEPT4:
3181 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
3182 (int __user *)a[2], a[3]);
3183 break;
3184 default:
3185 err = -EINVAL;
3186 break;
3187 }
3188 return err;
3189}
3190
3191#endif /* __ARCH_WANT_SYS_SOCKETCALL */
3192
3193/**
3194 * sock_register - add a socket protocol handler
3195 * @ops: description of protocol
3196 *
3197 * This function is called by a protocol handler that wants to
3198 * advertise its address family, and have it linked into the
3199 * socket interface. The value ops->family corresponds to the
3200 * socket system call protocol family.
3201 */
3202int sock_register(const struct net_proto_family *ops)
3203{
3204 int err;
3205
3206 if (ops->family >= NPROTO) {
3207 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
3208 return -ENOBUFS;
3209 }
3210
3211 spin_lock(&net_family_lock);
3212 if (rcu_dereference_protected(net_families[ops->family],
3213 lockdep_is_held(&net_family_lock)))
3214 err = -EEXIST;
3215 else {
3216 rcu_assign_pointer(net_families[ops->family], ops);
3217 err = 0;
3218 }
3219 spin_unlock(&net_family_lock);
3220
3221 pr_info("NET: Registered %s protocol family\n", pf_family_names[ops->family]);
3222 return err;
3223}
3224EXPORT_SYMBOL(sock_register);
3225
3226/**
3227 * sock_unregister - remove a protocol handler
3228 * @family: protocol family to remove
3229 *
3230 * This function is called by a protocol handler that wants to
3231 * remove its address family, and have it unlinked from the
3232 * new socket creation.
3233 *
3234 * If protocol handler is a module, then it can use module reference
3235 * counts to protect against new references. If protocol handler is not
3236 * a module then it needs to provide its own protection in
3237 * the ops->create routine.
3238 */
3239void sock_unregister(int family)
3240{
3241 BUG_ON(family < 0 || family >= NPROTO);
3242
3243 spin_lock(&net_family_lock);
3244 RCU_INIT_POINTER(net_families[family], NULL);
3245 spin_unlock(&net_family_lock);
3246
3247 synchronize_rcu();
3248
3249 pr_info("NET: Unregistered %s protocol family\n", pf_family_names[family]);
3250}
3251EXPORT_SYMBOL(sock_unregister);
3252
3253bool sock_is_registered(int family)
3254{
3255 return family < NPROTO && rcu_access_pointer(net_families[family]);
3256}
3257
3258static int __init sock_init(void)
3259{
3260 int err;
3261 /*
3262 * Initialize the network sysctl infrastructure.
3263 */
3264 err = net_sysctl_init();
3265 if (err)
3266 goto out;
3267
3268 /*
3269 * Initialize skbuff SLAB cache
3270 */
3271 skb_init();
3272
3273 /*
3274 * Initialize the protocols module.
3275 */
3276
3277 init_inodecache();
3278
3279 err = register_filesystem(&sock_fs_type);
3280 if (err)
3281 goto out;
3282 sock_mnt = kern_mount(&sock_fs_type);
3283 if (IS_ERR(sock_mnt)) {
3284 err = PTR_ERR(sock_mnt);
3285 goto out_mount;
3286 }
3287
3288 /* The real protocol initialization is performed in later initcalls.
3289 */
3290
3291#ifdef CONFIG_NETFILTER
3292 err = netfilter_init();
3293 if (err)
3294 goto out;
3295#endif
3296
3297 ptp_classifier_init();
3298
3299out:
3300 return err;
3301
3302out_mount:
3303 unregister_filesystem(&sock_fs_type);
3304 goto out;
3305}
3306
3307core_initcall(sock_init); /* early initcall */
3308
3309#ifdef CONFIG_PROC_FS
3310void socket_seq_show(struct seq_file *seq)
3311{
3312 seq_printf(seq, "sockets: used %d\n",
3313 sock_inuse_get(seq->private));
3314}
3315#endif /* CONFIG_PROC_FS */
3316
3317/* Handle the fact that while struct ifreq has the same *layout* on
3318 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3319 * which are handled elsewhere, it still has different *size* due to
3320 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3321 * resulting in struct ifreq being 32 and 40 bytes respectively).
3322 * As a result, if the struct happens to be at the end of a page and
3323 * the next page isn't readable/writable, we get a fault. To prevent
3324 * that, copy back and forth to the full size.
3325 */
3326int get_user_ifreq(struct ifreq *ifr, void __user **ifrdata, void __user *arg)
3327{
3328 if (in_compat_syscall()) {
3329 struct compat_ifreq *ifr32 = (struct compat_ifreq *)ifr;
3330
3331 memset(ifr, 0, sizeof(*ifr));
3332 if (copy_from_user(ifr32, arg, sizeof(*ifr32)))
3333 return -EFAULT;
3334
3335 if (ifrdata)
3336 *ifrdata = compat_ptr(ifr32->ifr_data);
3337
3338 return 0;
3339 }
3340
3341 if (copy_from_user(ifr, arg, sizeof(*ifr)))
3342 return -EFAULT;
3343
3344 if (ifrdata)
3345 *ifrdata = ifr->ifr_data;
3346
3347 return 0;
3348}
3349EXPORT_SYMBOL(get_user_ifreq);
3350
3351int put_user_ifreq(struct ifreq *ifr, void __user *arg)
3352{
3353 size_t size = sizeof(*ifr);
3354
3355 if (in_compat_syscall())
3356 size = sizeof(struct compat_ifreq);
3357
3358 if (copy_to_user(arg, ifr, size))
3359 return -EFAULT;
3360
3361 return 0;
3362}
3363EXPORT_SYMBOL(put_user_ifreq);
3364
3365#ifdef CONFIG_COMPAT
3366static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3367{
3368 compat_uptr_t uptr32;
3369 struct ifreq ifr;
3370 void __user *saved;
3371 int err;
3372
3373 if (get_user_ifreq(&ifr, NULL, uifr32))
3374 return -EFAULT;
3375
3376 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3377 return -EFAULT;
3378
3379 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3380 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3381
3382 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL, NULL);
3383 if (!err) {
3384 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3385 if (put_user_ifreq(&ifr, uifr32))
3386 err = -EFAULT;
3387 }
3388 return err;
3389}
3390
3391/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3392static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3393 struct compat_ifreq __user *u_ifreq32)
3394{
3395 struct ifreq ifreq;
3396 void __user *data;
3397
3398 if (!is_socket_ioctl_cmd(cmd))
3399 return -ENOTTY;
3400 if (get_user_ifreq(&ifreq, &data, u_ifreq32))
3401 return -EFAULT;
3402 ifreq.ifr_data = data;
3403
3404 return dev_ioctl(net, cmd, &ifreq, data, NULL);
3405}
3406
3407static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3408 unsigned int cmd, unsigned long arg)
3409{
3410 void __user *argp = compat_ptr(arg);
3411 struct sock *sk = sock->sk;
3412 struct net *net = sock_net(sk);
3413 const struct proto_ops *ops;
3414
3415 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3416 return sock_ioctl(file, cmd, (unsigned long)argp);
3417
3418 switch (cmd) {
3419 case SIOCWANDEV:
3420 return compat_siocwandev(net, argp);
3421 case SIOCGSTAMP_OLD:
3422 case SIOCGSTAMPNS_OLD:
3423 ops = READ_ONCE(sock->ops);
3424 if (!ops->gettstamp)
3425 return -ENOIOCTLCMD;
3426 return ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3427 !COMPAT_USE_64BIT_TIME);
3428
3429 case SIOCETHTOOL:
3430 case SIOCBONDSLAVEINFOQUERY:
3431 case SIOCBONDINFOQUERY:
3432 case SIOCSHWTSTAMP:
3433 case SIOCGHWTSTAMP:
3434 return compat_ifr_data_ioctl(net, cmd, argp);
3435
3436 case FIOSETOWN:
3437 case SIOCSPGRP:
3438 case FIOGETOWN:
3439 case SIOCGPGRP:
3440 case SIOCBRADDBR:
3441 case SIOCBRDELBR:
3442 case SIOCGIFVLAN:
3443 case SIOCSIFVLAN:
3444 case SIOCGSKNS:
3445 case SIOCGSTAMP_NEW:
3446 case SIOCGSTAMPNS_NEW:
3447 case SIOCGIFCONF:
3448 case SIOCSIFBR:
3449 case SIOCGIFBR:
3450 return sock_ioctl(file, cmd, arg);
3451
3452 case SIOCGIFFLAGS:
3453 case SIOCSIFFLAGS:
3454 case SIOCGIFMAP:
3455 case SIOCSIFMAP:
3456 case SIOCGIFMETRIC:
3457 case SIOCSIFMETRIC:
3458 case SIOCGIFMTU:
3459 case SIOCSIFMTU:
3460 case SIOCGIFMEM:
3461 case SIOCSIFMEM:
3462 case SIOCGIFHWADDR:
3463 case SIOCSIFHWADDR:
3464 case SIOCADDMULTI:
3465 case SIOCDELMULTI:
3466 case SIOCGIFINDEX:
3467 case SIOCGIFADDR:
3468 case SIOCSIFADDR:
3469 case SIOCSIFHWBROADCAST:
3470 case SIOCDIFADDR:
3471 case SIOCGIFBRDADDR:
3472 case SIOCSIFBRDADDR:
3473 case SIOCGIFDSTADDR:
3474 case SIOCSIFDSTADDR:
3475 case SIOCGIFNETMASK:
3476 case SIOCSIFNETMASK:
3477 case SIOCSIFPFLAGS:
3478 case SIOCGIFPFLAGS:
3479 case SIOCGIFTXQLEN:
3480 case SIOCSIFTXQLEN:
3481 case SIOCBRADDIF:
3482 case SIOCBRDELIF:
3483 case SIOCGIFNAME:
3484 case SIOCSIFNAME:
3485 case SIOCGMIIPHY:
3486 case SIOCGMIIREG:
3487 case SIOCSMIIREG:
3488 case SIOCBONDENSLAVE:
3489 case SIOCBONDRELEASE:
3490 case SIOCBONDSETHWADDR:
3491 case SIOCBONDCHANGEACTIVE:
3492 case SIOCSARP:
3493 case SIOCGARP:
3494 case SIOCDARP:
3495 case SIOCOUTQ:
3496 case SIOCOUTQNSD:
3497 case SIOCATMARK:
3498 return sock_do_ioctl(net, sock, cmd, arg);
3499 }
3500
3501 return -ENOIOCTLCMD;
3502}
3503
3504static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3505 unsigned long arg)
3506{
3507 struct socket *sock = file->private_data;
3508 const struct proto_ops *ops = READ_ONCE(sock->ops);
3509 int ret = -ENOIOCTLCMD;
3510 struct sock *sk;
3511 struct net *net;
3512
3513 sk = sock->sk;
3514 net = sock_net(sk);
3515
3516 if (ops->compat_ioctl)
3517 ret = ops->compat_ioctl(sock, cmd, arg);
3518
3519 if (ret == -ENOIOCTLCMD &&
3520 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3521 ret = compat_wext_handle_ioctl(net, cmd, arg);
3522
3523 if (ret == -ENOIOCTLCMD)
3524 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3525
3526 return ret;
3527}
3528#endif
3529
3530/**
3531 * kernel_bind - bind an address to a socket (kernel space)
3532 * @sock: socket
3533 * @addr: address
3534 * @addrlen: length of address
3535 *
3536 * Returns 0 or an error.
3537 */
3538
3539int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3540{
3541 struct sockaddr_storage address;
3542
3543 memcpy(&address, addr, addrlen);
3544
3545 return READ_ONCE(sock->ops)->bind(sock, (struct sockaddr *)&address,
3546 addrlen);
3547}
3548EXPORT_SYMBOL(kernel_bind);
3549
3550/**
3551 * kernel_listen - move socket to listening state (kernel space)
3552 * @sock: socket
3553 * @backlog: pending connections queue size
3554 *
3555 * Returns 0 or an error.
3556 */
3557
3558int kernel_listen(struct socket *sock, int backlog)
3559{
3560 return READ_ONCE(sock->ops)->listen(sock, backlog);
3561}
3562EXPORT_SYMBOL(kernel_listen);
3563
3564/**
3565 * kernel_accept - accept a connection (kernel space)
3566 * @sock: listening socket
3567 * @newsock: new connected socket
3568 * @flags: flags
3569 *
3570 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3571 * If it fails, @newsock is guaranteed to be %NULL.
3572 * Returns 0 or an error.
3573 */
3574
3575int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3576{
3577 struct sock *sk = sock->sk;
3578 const struct proto_ops *ops = READ_ONCE(sock->ops);
3579 struct proto_accept_arg arg = {
3580 .flags = flags,
3581 .kern = true,
3582 };
3583 int err;
3584
3585 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3586 newsock);
3587 if (err < 0)
3588 goto done;
3589
3590 err = ops->accept(sock, *newsock, &arg);
3591 if (err < 0) {
3592 sock_release(*newsock);
3593 *newsock = NULL;
3594 goto done;
3595 }
3596
3597 (*newsock)->ops = ops;
3598 __module_get(ops->owner);
3599
3600done:
3601 return err;
3602}
3603EXPORT_SYMBOL(kernel_accept);
3604
3605/**
3606 * kernel_connect - connect a socket (kernel space)
3607 * @sock: socket
3608 * @addr: address
3609 * @addrlen: address length
3610 * @flags: flags (O_NONBLOCK, ...)
3611 *
3612 * For datagram sockets, @addr is the address to which datagrams are sent
3613 * by default, and the only address from which datagrams are received.
3614 * For stream sockets, attempts to connect to @addr.
3615 * Returns 0 or an error code.
3616 */
3617
3618int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3619 int flags)
3620{
3621 struct sockaddr_storage address;
3622
3623 memcpy(&address, addr, addrlen);
3624
3625 return READ_ONCE(sock->ops)->connect(sock, (struct sockaddr *)&address,
3626 addrlen, flags);
3627}
3628EXPORT_SYMBOL(kernel_connect);
3629
3630/**
3631 * kernel_getsockname - get the address which the socket is bound (kernel space)
3632 * @sock: socket
3633 * @addr: address holder
3634 *
3635 * Fills the @addr pointer with the address which the socket is bound.
3636 * Returns the length of the address in bytes or an error code.
3637 */
3638
3639int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3640{
3641 return READ_ONCE(sock->ops)->getname(sock, addr, 0);
3642}
3643EXPORT_SYMBOL(kernel_getsockname);
3644
3645/**
3646 * kernel_getpeername - get the address which the socket is connected (kernel space)
3647 * @sock: socket
3648 * @addr: address holder
3649 *
3650 * Fills the @addr pointer with the address which the socket is connected.
3651 * Returns the length of the address in bytes or an error code.
3652 */
3653
3654int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3655{
3656 return READ_ONCE(sock->ops)->getname(sock, addr, 1);
3657}
3658EXPORT_SYMBOL(kernel_getpeername);
3659
3660/**
3661 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3662 * @sock: socket
3663 * @how: connection part
3664 *
3665 * Returns 0 or an error.
3666 */
3667
3668int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3669{
3670 return READ_ONCE(sock->ops)->shutdown(sock, how);
3671}
3672EXPORT_SYMBOL(kernel_sock_shutdown);
3673
3674/**
3675 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3676 * @sk: socket
3677 *
3678 * This routine returns the IP overhead imposed by a socket i.e.
3679 * the length of the underlying IP header, depending on whether
3680 * this is an IPv4 or IPv6 socket and the length from IP options turned
3681 * on at the socket. Assumes that the caller has a lock on the socket.
3682 */
3683
3684u32 kernel_sock_ip_overhead(struct sock *sk)
3685{
3686 struct inet_sock *inet;
3687 struct ip_options_rcu *opt;
3688 u32 overhead = 0;
3689#if IS_ENABLED(CONFIG_IPV6)
3690 struct ipv6_pinfo *np;
3691 struct ipv6_txoptions *optv6 = NULL;
3692#endif /* IS_ENABLED(CONFIG_IPV6) */
3693
3694 if (!sk)
3695 return overhead;
3696
3697 switch (sk->sk_family) {
3698 case AF_INET:
3699 inet = inet_sk(sk);
3700 overhead += sizeof(struct iphdr);
3701 opt = rcu_dereference_protected(inet->inet_opt,
3702 sock_owned_by_user(sk));
3703 if (opt)
3704 overhead += opt->opt.optlen;
3705 return overhead;
3706#if IS_ENABLED(CONFIG_IPV6)
3707 case AF_INET6:
3708 np = inet6_sk(sk);
3709 overhead += sizeof(struct ipv6hdr);
3710 if (np)
3711 optv6 = rcu_dereference_protected(np->opt,
3712 sock_owned_by_user(sk));
3713 if (optv6)
3714 overhead += (optv6->opt_flen + optv6->opt_nflen);
3715 return overhead;
3716#endif /* IS_ENABLED(CONFIG_IPV6) */
3717 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3718 return overhead;
3719 }
3720}
3721EXPORT_SYMBOL(kernel_sock_ip_overhead);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET An implementation of the SOCKET network access protocol.
4 *
5 * Version: @(#)socket.c 1.1.93 18/02/95
6 *
7 * Authors: Orest Zborowski, <obz@Kodak.COM>
8 * Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
13 * shutdown()
14 * Alan Cox : verify_area() fixes
15 * Alan Cox : Removed DDI
16 * Jonathan Kamens : SOCK_DGRAM reconnect bug
17 * Alan Cox : Moved a load of checks to the very
18 * top level.
19 * Alan Cox : Move address structures to/from user
20 * mode above the protocol layers.
21 * Rob Janssen : Allow 0 length sends.
22 * Alan Cox : Asynchronous I/O support (cribbed from the
23 * tty drivers).
24 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
25 * Jeff Uphoff : Made max number of sockets command-line
26 * configurable.
27 * Matti Aarnio : Made the number of sockets dynamic,
28 * to be allocated when needed, and mr.
29 * Uphoff's max is used as max to be
30 * allowed to allocate.
31 * Linus : Argh. removed all the socket allocation
32 * altogether: it's in the inode now.
33 * Alan Cox : Made sock_alloc()/sock_release() public
34 * for NetROM and future kernel nfsd type
35 * stuff.
36 * Alan Cox : sendmsg/recvmsg basics.
37 * Tom Dyas : Export net symbols.
38 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
39 * Alan Cox : Added thread locking to sys_* calls
40 * for sockets. May have errors at the
41 * moment.
42 * Kevin Buhr : Fixed the dumb errors in the above.
43 * Andi Kleen : Some small cleanups, optimizations,
44 * and fixed a copy_from_user() bug.
45 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
46 * Tigran Aivazian : Made listen(2) backlog sanity checks
47 * protocol-independent
48 *
49 * This module is effectively the top level interface to the BSD socket
50 * paradigm.
51 *
52 * Based upon Swansea University Computer Society NET3.039
53 */
54
55#include <linux/mm.h>
56#include <linux/socket.h>
57#include <linux/file.h>
58#include <linux/net.h>
59#include <linux/interrupt.h>
60#include <linux/thread_info.h>
61#include <linux/rcupdate.h>
62#include <linux/netdevice.h>
63#include <linux/proc_fs.h>
64#include <linux/seq_file.h>
65#include <linux/mutex.h>
66#include <linux/if_bridge.h>
67#include <linux/if_frad.h>
68#include <linux/if_vlan.h>
69#include <linux/ptp_classify.h>
70#include <linux/init.h>
71#include <linux/poll.h>
72#include <linux/cache.h>
73#include <linux/module.h>
74#include <linux/highmem.h>
75#include <linux/mount.h>
76#include <linux/pseudo_fs.h>
77#include <linux/security.h>
78#include <linux/syscalls.h>
79#include <linux/compat.h>
80#include <linux/kmod.h>
81#include <linux/audit.h>
82#include <linux/wireless.h>
83#include <linux/nsproxy.h>
84#include <linux/magic.h>
85#include <linux/slab.h>
86#include <linux/xattr.h>
87#include <linux/nospec.h>
88#include <linux/indirect_call_wrapper.h>
89
90#include <linux/uaccess.h>
91#include <asm/unistd.h>
92
93#include <net/compat.h>
94#include <net/wext.h>
95#include <net/cls_cgroup.h>
96
97#include <net/sock.h>
98#include <linux/netfilter.h>
99
100#include <linux/if_tun.h>
101#include <linux/ipv6_route.h>
102#include <linux/route.h>
103#include <linux/termios.h>
104#include <linux/sockios.h>
105#include <net/busy_poll.h>
106#include <linux/errqueue.h>
107
108#ifdef CONFIG_NET_RX_BUSY_POLL
109unsigned int sysctl_net_busy_read __read_mostly;
110unsigned int sysctl_net_busy_poll __read_mostly;
111#endif
112
113static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
114static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
115static int sock_mmap(struct file *file, struct vm_area_struct *vma);
116
117static int sock_close(struct inode *inode, struct file *file);
118static __poll_t sock_poll(struct file *file,
119 struct poll_table_struct *wait);
120static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
121#ifdef CONFIG_COMPAT
122static long compat_sock_ioctl(struct file *file,
123 unsigned int cmd, unsigned long arg);
124#endif
125static int sock_fasync(int fd, struct file *filp, int on);
126static ssize_t sock_sendpage(struct file *file, struct page *page,
127 int offset, size_t size, loff_t *ppos, int more);
128static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
129 struct pipe_inode_info *pipe, size_t len,
130 unsigned int flags);
131
132#ifdef CONFIG_PROC_FS
133static void sock_show_fdinfo(struct seq_file *m, struct file *f)
134{
135 struct socket *sock = f->private_data;
136
137 if (sock->ops->show_fdinfo)
138 sock->ops->show_fdinfo(m, sock);
139}
140#else
141#define sock_show_fdinfo NULL
142#endif
143
144/*
145 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
146 * in the operation structures but are done directly via the socketcall() multiplexor.
147 */
148
149static const struct file_operations socket_file_ops = {
150 .owner = THIS_MODULE,
151 .llseek = no_llseek,
152 .read_iter = sock_read_iter,
153 .write_iter = sock_write_iter,
154 .poll = sock_poll,
155 .unlocked_ioctl = sock_ioctl,
156#ifdef CONFIG_COMPAT
157 .compat_ioctl = compat_sock_ioctl,
158#endif
159 .mmap = sock_mmap,
160 .release = sock_close,
161 .fasync = sock_fasync,
162 .sendpage = sock_sendpage,
163 .splice_write = generic_splice_sendpage,
164 .splice_read = sock_splice_read,
165 .show_fdinfo = sock_show_fdinfo,
166};
167
168/*
169 * The protocol list. Each protocol is registered in here.
170 */
171
172static DEFINE_SPINLOCK(net_family_lock);
173static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
174
175/*
176 * Support routines.
177 * Move socket addresses back and forth across the kernel/user
178 * divide and look after the messy bits.
179 */
180
181/**
182 * move_addr_to_kernel - copy a socket address into kernel space
183 * @uaddr: Address in user space
184 * @kaddr: Address in kernel space
185 * @ulen: Length in user space
186 *
187 * The address is copied into kernel space. If the provided address is
188 * too long an error code of -EINVAL is returned. If the copy gives
189 * invalid addresses -EFAULT is returned. On a success 0 is returned.
190 */
191
192int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
193{
194 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
195 return -EINVAL;
196 if (ulen == 0)
197 return 0;
198 if (copy_from_user(kaddr, uaddr, ulen))
199 return -EFAULT;
200 return audit_sockaddr(ulen, kaddr);
201}
202
203/**
204 * move_addr_to_user - copy an address to user space
205 * @kaddr: kernel space address
206 * @klen: length of address in kernel
207 * @uaddr: user space address
208 * @ulen: pointer to user length field
209 *
210 * The value pointed to by ulen on entry is the buffer length available.
211 * This is overwritten with the buffer space used. -EINVAL is returned
212 * if an overlong buffer is specified or a negative buffer size. -EFAULT
213 * is returned if either the buffer or the length field are not
214 * accessible.
215 * After copying the data up to the limit the user specifies, the true
216 * length of the data is written over the length limit the user
217 * specified. Zero is returned for a success.
218 */
219
220static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
221 void __user *uaddr, int __user *ulen)
222{
223 int err;
224 int len;
225
226 BUG_ON(klen > sizeof(struct sockaddr_storage));
227 err = get_user(len, ulen);
228 if (err)
229 return err;
230 if (len > klen)
231 len = klen;
232 if (len < 0)
233 return -EINVAL;
234 if (len) {
235 if (audit_sockaddr(klen, kaddr))
236 return -ENOMEM;
237 if (copy_to_user(uaddr, kaddr, len))
238 return -EFAULT;
239 }
240 /*
241 * "fromlen shall refer to the value before truncation.."
242 * 1003.1g
243 */
244 return __put_user(klen, ulen);
245}
246
247static struct kmem_cache *sock_inode_cachep __ro_after_init;
248
249static struct inode *sock_alloc_inode(struct super_block *sb)
250{
251 struct socket_alloc *ei;
252
253 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
254 if (!ei)
255 return NULL;
256 init_waitqueue_head(&ei->socket.wq.wait);
257 ei->socket.wq.fasync_list = NULL;
258 ei->socket.wq.flags = 0;
259
260 ei->socket.state = SS_UNCONNECTED;
261 ei->socket.flags = 0;
262 ei->socket.ops = NULL;
263 ei->socket.sk = NULL;
264 ei->socket.file = NULL;
265
266 return &ei->vfs_inode;
267}
268
269static void sock_free_inode(struct inode *inode)
270{
271 struct socket_alloc *ei;
272
273 ei = container_of(inode, struct socket_alloc, vfs_inode);
274 kmem_cache_free(sock_inode_cachep, ei);
275}
276
277static void init_once(void *foo)
278{
279 struct socket_alloc *ei = (struct socket_alloc *)foo;
280
281 inode_init_once(&ei->vfs_inode);
282}
283
284static void init_inodecache(void)
285{
286 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
287 sizeof(struct socket_alloc),
288 0,
289 (SLAB_HWCACHE_ALIGN |
290 SLAB_RECLAIM_ACCOUNT |
291 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
292 init_once);
293 BUG_ON(sock_inode_cachep == NULL);
294}
295
296static const struct super_operations sockfs_ops = {
297 .alloc_inode = sock_alloc_inode,
298 .free_inode = sock_free_inode,
299 .statfs = simple_statfs,
300};
301
302/*
303 * sockfs_dname() is called from d_path().
304 */
305static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
306{
307 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
308 d_inode(dentry)->i_ino);
309}
310
311static const struct dentry_operations sockfs_dentry_operations = {
312 .d_dname = sockfs_dname,
313};
314
315static int sockfs_xattr_get(const struct xattr_handler *handler,
316 struct dentry *dentry, struct inode *inode,
317 const char *suffix, void *value, size_t size)
318{
319 if (value) {
320 if (dentry->d_name.len + 1 > size)
321 return -ERANGE;
322 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
323 }
324 return dentry->d_name.len + 1;
325}
326
327#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
328#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
329#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
330
331static const struct xattr_handler sockfs_xattr_handler = {
332 .name = XATTR_NAME_SOCKPROTONAME,
333 .get = sockfs_xattr_get,
334};
335
336static int sockfs_security_xattr_set(const struct xattr_handler *handler,
337 struct dentry *dentry, struct inode *inode,
338 const char *suffix, const void *value,
339 size_t size, int flags)
340{
341 /* Handled by LSM. */
342 return -EAGAIN;
343}
344
345static const struct xattr_handler sockfs_security_xattr_handler = {
346 .prefix = XATTR_SECURITY_PREFIX,
347 .set = sockfs_security_xattr_set,
348};
349
350static const struct xattr_handler *sockfs_xattr_handlers[] = {
351 &sockfs_xattr_handler,
352 &sockfs_security_xattr_handler,
353 NULL
354};
355
356static int sockfs_init_fs_context(struct fs_context *fc)
357{
358 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
359 if (!ctx)
360 return -ENOMEM;
361 ctx->ops = &sockfs_ops;
362 ctx->dops = &sockfs_dentry_operations;
363 ctx->xattr = sockfs_xattr_handlers;
364 return 0;
365}
366
367static struct vfsmount *sock_mnt __read_mostly;
368
369static struct file_system_type sock_fs_type = {
370 .name = "sockfs",
371 .init_fs_context = sockfs_init_fs_context,
372 .kill_sb = kill_anon_super,
373};
374
375/*
376 * Obtains the first available file descriptor and sets it up for use.
377 *
378 * These functions create file structures and maps them to fd space
379 * of the current process. On success it returns file descriptor
380 * and file struct implicitly stored in sock->file.
381 * Note that another thread may close file descriptor before we return
382 * from this function. We use the fact that now we do not refer
383 * to socket after mapping. If one day we will need it, this
384 * function will increment ref. count on file by 1.
385 *
386 * In any case returned fd MAY BE not valid!
387 * This race condition is unavoidable
388 * with shared fd spaces, we cannot solve it inside kernel,
389 * but we take care of internal coherence yet.
390 */
391
392/**
393 * sock_alloc_file - Bind a &socket to a &file
394 * @sock: socket
395 * @flags: file status flags
396 * @dname: protocol name
397 *
398 * Returns the &file bound with @sock, implicitly storing it
399 * in sock->file. If dname is %NULL, sets to "".
400 * On failure the return is a ERR pointer (see linux/err.h).
401 * This function uses GFP_KERNEL internally.
402 */
403
404struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
405{
406 struct file *file;
407
408 if (!dname)
409 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
410
411 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
412 O_RDWR | (flags & O_NONBLOCK),
413 &socket_file_ops);
414 if (IS_ERR(file)) {
415 sock_release(sock);
416 return file;
417 }
418
419 sock->file = file;
420 file->private_data = sock;
421 stream_open(SOCK_INODE(sock), file);
422 return file;
423}
424EXPORT_SYMBOL(sock_alloc_file);
425
426static int sock_map_fd(struct socket *sock, int flags)
427{
428 struct file *newfile;
429 int fd = get_unused_fd_flags(flags);
430 if (unlikely(fd < 0)) {
431 sock_release(sock);
432 return fd;
433 }
434
435 newfile = sock_alloc_file(sock, flags, NULL);
436 if (!IS_ERR(newfile)) {
437 fd_install(fd, newfile);
438 return fd;
439 }
440
441 put_unused_fd(fd);
442 return PTR_ERR(newfile);
443}
444
445/**
446 * sock_from_file - Return the &socket bounded to @file.
447 * @file: file
448 * @err: pointer to an error code return
449 *
450 * On failure returns %NULL and assigns -ENOTSOCK to @err.
451 */
452
453struct socket *sock_from_file(struct file *file, int *err)
454{
455 if (file->f_op == &socket_file_ops)
456 return file->private_data; /* set in sock_map_fd */
457
458 *err = -ENOTSOCK;
459 return NULL;
460}
461EXPORT_SYMBOL(sock_from_file);
462
463/**
464 * sockfd_lookup - Go from a file number to its socket slot
465 * @fd: file handle
466 * @err: pointer to an error code return
467 *
468 * The file handle passed in is locked and the socket it is bound
469 * to is returned. If an error occurs the err pointer is overwritten
470 * with a negative errno code and NULL is returned. The function checks
471 * for both invalid handles and passing a handle which is not a socket.
472 *
473 * On a success the socket object pointer is returned.
474 */
475
476struct socket *sockfd_lookup(int fd, int *err)
477{
478 struct file *file;
479 struct socket *sock;
480
481 file = fget(fd);
482 if (!file) {
483 *err = -EBADF;
484 return NULL;
485 }
486
487 sock = sock_from_file(file, err);
488 if (!sock)
489 fput(file);
490 return sock;
491}
492EXPORT_SYMBOL(sockfd_lookup);
493
494static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
495{
496 struct fd f = fdget(fd);
497 struct socket *sock;
498
499 *err = -EBADF;
500 if (f.file) {
501 sock = sock_from_file(f.file, err);
502 if (likely(sock)) {
503 *fput_needed = f.flags & FDPUT_FPUT;
504 return sock;
505 }
506 fdput(f);
507 }
508 return NULL;
509}
510
511static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
512 size_t size)
513{
514 ssize_t len;
515 ssize_t used = 0;
516
517 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
518 if (len < 0)
519 return len;
520 used += len;
521 if (buffer) {
522 if (size < used)
523 return -ERANGE;
524 buffer += len;
525 }
526
527 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
528 used += len;
529 if (buffer) {
530 if (size < used)
531 return -ERANGE;
532 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
533 buffer += len;
534 }
535
536 return used;
537}
538
539static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
540{
541 int err = simple_setattr(dentry, iattr);
542
543 if (!err && (iattr->ia_valid & ATTR_UID)) {
544 struct socket *sock = SOCKET_I(d_inode(dentry));
545
546 if (sock->sk)
547 sock->sk->sk_uid = iattr->ia_uid;
548 else
549 err = -ENOENT;
550 }
551
552 return err;
553}
554
555static const struct inode_operations sockfs_inode_ops = {
556 .listxattr = sockfs_listxattr,
557 .setattr = sockfs_setattr,
558};
559
560/**
561 * sock_alloc - allocate a socket
562 *
563 * Allocate a new inode and socket object. The two are bound together
564 * and initialised. The socket is then returned. If we are out of inodes
565 * NULL is returned. This functions uses GFP_KERNEL internally.
566 */
567
568struct socket *sock_alloc(void)
569{
570 struct inode *inode;
571 struct socket *sock;
572
573 inode = new_inode_pseudo(sock_mnt->mnt_sb);
574 if (!inode)
575 return NULL;
576
577 sock = SOCKET_I(inode);
578
579 inode->i_ino = get_next_ino();
580 inode->i_mode = S_IFSOCK | S_IRWXUGO;
581 inode->i_uid = current_fsuid();
582 inode->i_gid = current_fsgid();
583 inode->i_op = &sockfs_inode_ops;
584
585 return sock;
586}
587EXPORT_SYMBOL(sock_alloc);
588
589static void __sock_release(struct socket *sock, struct inode *inode)
590{
591 if (sock->ops) {
592 struct module *owner = sock->ops->owner;
593
594 if (inode)
595 inode_lock(inode);
596 sock->ops->release(sock);
597 sock->sk = NULL;
598 if (inode)
599 inode_unlock(inode);
600 sock->ops = NULL;
601 module_put(owner);
602 }
603
604 if (sock->wq.fasync_list)
605 pr_err("%s: fasync list not empty!\n", __func__);
606
607 if (!sock->file) {
608 iput(SOCK_INODE(sock));
609 return;
610 }
611 sock->file = NULL;
612}
613
614/**
615 * sock_release - close a socket
616 * @sock: socket to close
617 *
618 * The socket is released from the protocol stack if it has a release
619 * callback, and the inode is then released if the socket is bound to
620 * an inode not a file.
621 */
622void sock_release(struct socket *sock)
623{
624 __sock_release(sock, NULL);
625}
626EXPORT_SYMBOL(sock_release);
627
628void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
629{
630 u8 flags = *tx_flags;
631
632 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
633 flags |= SKBTX_HW_TSTAMP;
634
635 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
636 flags |= SKBTX_SW_TSTAMP;
637
638 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
639 flags |= SKBTX_SCHED_TSTAMP;
640
641 *tx_flags = flags;
642}
643EXPORT_SYMBOL(__sock_tx_timestamp);
644
645INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
646 size_t));
647INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
648 size_t));
649static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
650{
651 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
652 inet_sendmsg, sock, msg,
653 msg_data_left(msg));
654 BUG_ON(ret == -EIOCBQUEUED);
655 return ret;
656}
657
658/**
659 * sock_sendmsg - send a message through @sock
660 * @sock: socket
661 * @msg: message to send
662 *
663 * Sends @msg through @sock, passing through LSM.
664 * Returns the number of bytes sent, or an error code.
665 */
666int sock_sendmsg(struct socket *sock, struct msghdr *msg)
667{
668 int err = security_socket_sendmsg(sock, msg,
669 msg_data_left(msg));
670
671 return err ?: sock_sendmsg_nosec(sock, msg);
672}
673EXPORT_SYMBOL(sock_sendmsg);
674
675/**
676 * kernel_sendmsg - send a message through @sock (kernel-space)
677 * @sock: socket
678 * @msg: message header
679 * @vec: kernel vec
680 * @num: vec array length
681 * @size: total message data size
682 *
683 * Builds the message data with @vec and sends it through @sock.
684 * Returns the number of bytes sent, or an error code.
685 */
686
687int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
688 struct kvec *vec, size_t num, size_t size)
689{
690 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
691 return sock_sendmsg(sock, msg);
692}
693EXPORT_SYMBOL(kernel_sendmsg);
694
695/**
696 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
697 * @sk: sock
698 * @msg: message header
699 * @vec: output s/g array
700 * @num: output s/g array length
701 * @size: total message data size
702 *
703 * Builds the message data with @vec and sends it through @sock.
704 * Returns the number of bytes sent, or an error code.
705 * Caller must hold @sk.
706 */
707
708int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
709 struct kvec *vec, size_t num, size_t size)
710{
711 struct socket *sock = sk->sk_socket;
712
713 if (!sock->ops->sendmsg_locked)
714 return sock_no_sendmsg_locked(sk, msg, size);
715
716 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
717
718 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
719}
720EXPORT_SYMBOL(kernel_sendmsg_locked);
721
722static bool skb_is_err_queue(const struct sk_buff *skb)
723{
724 /* pkt_type of skbs enqueued on the error queue are set to
725 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
726 * in recvmsg, since skbs received on a local socket will never
727 * have a pkt_type of PACKET_OUTGOING.
728 */
729 return skb->pkt_type == PACKET_OUTGOING;
730}
731
732/* On transmit, software and hardware timestamps are returned independently.
733 * As the two skb clones share the hardware timestamp, which may be updated
734 * before the software timestamp is received, a hardware TX timestamp may be
735 * returned only if there is no software TX timestamp. Ignore false software
736 * timestamps, which may be made in the __sock_recv_timestamp() call when the
737 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
738 * hardware timestamp.
739 */
740static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
741{
742 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
743}
744
745static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
746{
747 struct scm_ts_pktinfo ts_pktinfo;
748 struct net_device *orig_dev;
749
750 if (!skb_mac_header_was_set(skb))
751 return;
752
753 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
754
755 rcu_read_lock();
756 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
757 if (orig_dev)
758 ts_pktinfo.if_index = orig_dev->ifindex;
759 rcu_read_unlock();
760
761 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
762 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
763 sizeof(ts_pktinfo), &ts_pktinfo);
764}
765
766/*
767 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
768 */
769void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
770 struct sk_buff *skb)
771{
772 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
773 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
774 struct scm_timestamping_internal tss;
775
776 int empty = 1, false_tstamp = 0;
777 struct skb_shared_hwtstamps *shhwtstamps =
778 skb_hwtstamps(skb);
779
780 /* Race occurred between timestamp enabling and packet
781 receiving. Fill in the current time for now. */
782 if (need_software_tstamp && skb->tstamp == 0) {
783 __net_timestamp(skb);
784 false_tstamp = 1;
785 }
786
787 if (need_software_tstamp) {
788 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
789 if (new_tstamp) {
790 struct __kernel_sock_timeval tv;
791
792 skb_get_new_timestamp(skb, &tv);
793 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
794 sizeof(tv), &tv);
795 } else {
796 struct __kernel_old_timeval tv;
797
798 skb_get_timestamp(skb, &tv);
799 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
800 sizeof(tv), &tv);
801 }
802 } else {
803 if (new_tstamp) {
804 struct __kernel_timespec ts;
805
806 skb_get_new_timestampns(skb, &ts);
807 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
808 sizeof(ts), &ts);
809 } else {
810 struct __kernel_old_timespec ts;
811
812 skb_get_timestampns(skb, &ts);
813 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
814 sizeof(ts), &ts);
815 }
816 }
817 }
818
819 memset(&tss, 0, sizeof(tss));
820 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
821 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
822 empty = 0;
823 if (shhwtstamps &&
824 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
825 !skb_is_swtx_tstamp(skb, false_tstamp) &&
826 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
827 empty = 0;
828 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
829 !skb_is_err_queue(skb))
830 put_ts_pktinfo(msg, skb);
831 }
832 if (!empty) {
833 if (sock_flag(sk, SOCK_TSTAMP_NEW))
834 put_cmsg_scm_timestamping64(msg, &tss);
835 else
836 put_cmsg_scm_timestamping(msg, &tss);
837
838 if (skb_is_err_queue(skb) && skb->len &&
839 SKB_EXT_ERR(skb)->opt_stats)
840 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
841 skb->len, skb->data);
842 }
843}
844EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
845
846void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
847 struct sk_buff *skb)
848{
849 int ack;
850
851 if (!sock_flag(sk, SOCK_WIFI_STATUS))
852 return;
853 if (!skb->wifi_acked_valid)
854 return;
855
856 ack = skb->wifi_acked;
857
858 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
859}
860EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
861
862static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
863 struct sk_buff *skb)
864{
865 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
866 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
867 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
868}
869
870void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
871 struct sk_buff *skb)
872{
873 sock_recv_timestamp(msg, sk, skb);
874 sock_recv_drops(msg, sk, skb);
875}
876EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
877
878INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
879 size_t, int));
880INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
881 size_t, int));
882static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
883 int flags)
884{
885 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
886 inet_recvmsg, sock, msg, msg_data_left(msg),
887 flags);
888}
889
890/**
891 * sock_recvmsg - receive a message from @sock
892 * @sock: socket
893 * @msg: message to receive
894 * @flags: message flags
895 *
896 * Receives @msg from @sock, passing through LSM. Returns the total number
897 * of bytes received, or an error.
898 */
899int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
900{
901 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
902
903 return err ?: sock_recvmsg_nosec(sock, msg, flags);
904}
905EXPORT_SYMBOL(sock_recvmsg);
906
907/**
908 * kernel_recvmsg - Receive a message from a socket (kernel space)
909 * @sock: The socket to receive the message from
910 * @msg: Received message
911 * @vec: Input s/g array for message data
912 * @num: Size of input s/g array
913 * @size: Number of bytes to read
914 * @flags: Message flags (MSG_DONTWAIT, etc...)
915 *
916 * On return the msg structure contains the scatter/gather array passed in the
917 * vec argument. The array is modified so that it consists of the unfilled
918 * portion of the original array.
919 *
920 * The returned value is the total number of bytes received, or an error.
921 */
922
923int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
924 struct kvec *vec, size_t num, size_t size, int flags)
925{
926 msg->msg_control_is_user = false;
927 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
928 return sock_recvmsg(sock, msg, flags);
929}
930EXPORT_SYMBOL(kernel_recvmsg);
931
932static ssize_t sock_sendpage(struct file *file, struct page *page,
933 int offset, size_t size, loff_t *ppos, int more)
934{
935 struct socket *sock;
936 int flags;
937
938 sock = file->private_data;
939
940 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
941 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
942 flags |= more;
943
944 return kernel_sendpage(sock, page, offset, size, flags);
945}
946
947static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
948 struct pipe_inode_info *pipe, size_t len,
949 unsigned int flags)
950{
951 struct socket *sock = file->private_data;
952
953 if (unlikely(!sock->ops->splice_read))
954 return generic_file_splice_read(file, ppos, pipe, len, flags);
955
956 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
957}
958
959static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
960{
961 struct file *file = iocb->ki_filp;
962 struct socket *sock = file->private_data;
963 struct msghdr msg = {.msg_iter = *to,
964 .msg_iocb = iocb};
965 ssize_t res;
966
967 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
968 msg.msg_flags = MSG_DONTWAIT;
969
970 if (iocb->ki_pos != 0)
971 return -ESPIPE;
972
973 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
974 return 0;
975
976 res = sock_recvmsg(sock, &msg, msg.msg_flags);
977 *to = msg.msg_iter;
978 return res;
979}
980
981static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
982{
983 struct file *file = iocb->ki_filp;
984 struct socket *sock = file->private_data;
985 struct msghdr msg = {.msg_iter = *from,
986 .msg_iocb = iocb};
987 ssize_t res;
988
989 if (iocb->ki_pos != 0)
990 return -ESPIPE;
991
992 if (file->f_flags & O_NONBLOCK || (iocb->ki_flags & IOCB_NOWAIT))
993 msg.msg_flags = MSG_DONTWAIT;
994
995 if (sock->type == SOCK_SEQPACKET)
996 msg.msg_flags |= MSG_EOR;
997
998 res = sock_sendmsg(sock, &msg);
999 *from = msg.msg_iter;
1000 return res;
1001}
1002
1003/*
1004 * Atomic setting of ioctl hooks to avoid race
1005 * with module unload.
1006 */
1007
1008static DEFINE_MUTEX(br_ioctl_mutex);
1009static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
1010
1011void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
1012{
1013 mutex_lock(&br_ioctl_mutex);
1014 br_ioctl_hook = hook;
1015 mutex_unlock(&br_ioctl_mutex);
1016}
1017EXPORT_SYMBOL(brioctl_set);
1018
1019static DEFINE_MUTEX(vlan_ioctl_mutex);
1020static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1021
1022void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1023{
1024 mutex_lock(&vlan_ioctl_mutex);
1025 vlan_ioctl_hook = hook;
1026 mutex_unlock(&vlan_ioctl_mutex);
1027}
1028EXPORT_SYMBOL(vlan_ioctl_set);
1029
1030static DEFINE_MUTEX(dlci_ioctl_mutex);
1031static int (*dlci_ioctl_hook) (unsigned int, void __user *);
1032
1033void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
1034{
1035 mutex_lock(&dlci_ioctl_mutex);
1036 dlci_ioctl_hook = hook;
1037 mutex_unlock(&dlci_ioctl_mutex);
1038}
1039EXPORT_SYMBOL(dlci_ioctl_set);
1040
1041static long sock_do_ioctl(struct net *net, struct socket *sock,
1042 unsigned int cmd, unsigned long arg)
1043{
1044 int err;
1045 void __user *argp = (void __user *)arg;
1046
1047 err = sock->ops->ioctl(sock, cmd, arg);
1048
1049 /*
1050 * If this ioctl is unknown try to hand it down
1051 * to the NIC driver.
1052 */
1053 if (err != -ENOIOCTLCMD)
1054 return err;
1055
1056 if (cmd == SIOCGIFCONF) {
1057 struct ifconf ifc;
1058 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
1059 return -EFAULT;
1060 rtnl_lock();
1061 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
1062 rtnl_unlock();
1063 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
1064 err = -EFAULT;
1065 } else {
1066 struct ifreq ifr;
1067 bool need_copyout;
1068 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1069 return -EFAULT;
1070 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1071 if (!err && need_copyout)
1072 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1073 return -EFAULT;
1074 }
1075 return err;
1076}
1077
1078/*
1079 * With an ioctl, arg may well be a user mode pointer, but we don't know
1080 * what to do with it - that's up to the protocol still.
1081 */
1082
1083/**
1084 * get_net_ns - increment the refcount of the network namespace
1085 * @ns: common namespace (net)
1086 *
1087 * Returns the net's common namespace.
1088 */
1089
1090struct ns_common *get_net_ns(struct ns_common *ns)
1091{
1092 return &get_net(container_of(ns, struct net, ns))->ns;
1093}
1094EXPORT_SYMBOL_GPL(get_net_ns);
1095
1096static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1097{
1098 struct socket *sock;
1099 struct sock *sk;
1100 void __user *argp = (void __user *)arg;
1101 int pid, err;
1102 struct net *net;
1103
1104 sock = file->private_data;
1105 sk = sock->sk;
1106 net = sock_net(sk);
1107 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1108 struct ifreq ifr;
1109 bool need_copyout;
1110 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1111 return -EFAULT;
1112 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1113 if (!err && need_copyout)
1114 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1115 return -EFAULT;
1116 } else
1117#ifdef CONFIG_WEXT_CORE
1118 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1119 err = wext_handle_ioctl(net, cmd, argp);
1120 } else
1121#endif
1122 switch (cmd) {
1123 case FIOSETOWN:
1124 case SIOCSPGRP:
1125 err = -EFAULT;
1126 if (get_user(pid, (int __user *)argp))
1127 break;
1128 err = f_setown(sock->file, pid, 1);
1129 break;
1130 case FIOGETOWN:
1131 case SIOCGPGRP:
1132 err = put_user(f_getown(sock->file),
1133 (int __user *)argp);
1134 break;
1135 case SIOCGIFBR:
1136 case SIOCSIFBR:
1137 case SIOCBRADDBR:
1138 case SIOCBRDELBR:
1139 err = -ENOPKG;
1140 if (!br_ioctl_hook)
1141 request_module("bridge");
1142
1143 mutex_lock(&br_ioctl_mutex);
1144 if (br_ioctl_hook)
1145 err = br_ioctl_hook(net, cmd, argp);
1146 mutex_unlock(&br_ioctl_mutex);
1147 break;
1148 case SIOCGIFVLAN:
1149 case SIOCSIFVLAN:
1150 err = -ENOPKG;
1151 if (!vlan_ioctl_hook)
1152 request_module("8021q");
1153
1154 mutex_lock(&vlan_ioctl_mutex);
1155 if (vlan_ioctl_hook)
1156 err = vlan_ioctl_hook(net, argp);
1157 mutex_unlock(&vlan_ioctl_mutex);
1158 break;
1159 case SIOCADDDLCI:
1160 case SIOCDELDLCI:
1161 err = -ENOPKG;
1162 if (!dlci_ioctl_hook)
1163 request_module("dlci");
1164
1165 mutex_lock(&dlci_ioctl_mutex);
1166 if (dlci_ioctl_hook)
1167 err = dlci_ioctl_hook(cmd, argp);
1168 mutex_unlock(&dlci_ioctl_mutex);
1169 break;
1170 case SIOCGSKNS:
1171 err = -EPERM;
1172 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1173 break;
1174
1175 err = open_related_ns(&net->ns, get_net_ns);
1176 break;
1177 case SIOCGSTAMP_OLD:
1178 case SIOCGSTAMPNS_OLD:
1179 if (!sock->ops->gettstamp) {
1180 err = -ENOIOCTLCMD;
1181 break;
1182 }
1183 err = sock->ops->gettstamp(sock, argp,
1184 cmd == SIOCGSTAMP_OLD,
1185 !IS_ENABLED(CONFIG_64BIT));
1186 break;
1187 case SIOCGSTAMP_NEW:
1188 case SIOCGSTAMPNS_NEW:
1189 if (!sock->ops->gettstamp) {
1190 err = -ENOIOCTLCMD;
1191 break;
1192 }
1193 err = sock->ops->gettstamp(sock, argp,
1194 cmd == SIOCGSTAMP_NEW,
1195 false);
1196 break;
1197 default:
1198 err = sock_do_ioctl(net, sock, cmd, arg);
1199 break;
1200 }
1201 return err;
1202}
1203
1204/**
1205 * sock_create_lite - creates a socket
1206 * @family: protocol family (AF_INET, ...)
1207 * @type: communication type (SOCK_STREAM, ...)
1208 * @protocol: protocol (0, ...)
1209 * @res: new socket
1210 *
1211 * Creates a new socket and assigns it to @res, passing through LSM.
1212 * The new socket initialization is not complete, see kernel_accept().
1213 * Returns 0 or an error. On failure @res is set to %NULL.
1214 * This function internally uses GFP_KERNEL.
1215 */
1216
1217int sock_create_lite(int family, int type, int protocol, struct socket **res)
1218{
1219 int err;
1220 struct socket *sock = NULL;
1221
1222 err = security_socket_create(family, type, protocol, 1);
1223 if (err)
1224 goto out;
1225
1226 sock = sock_alloc();
1227 if (!sock) {
1228 err = -ENOMEM;
1229 goto out;
1230 }
1231
1232 sock->type = type;
1233 err = security_socket_post_create(sock, family, type, protocol, 1);
1234 if (err)
1235 goto out_release;
1236
1237out:
1238 *res = sock;
1239 return err;
1240out_release:
1241 sock_release(sock);
1242 sock = NULL;
1243 goto out;
1244}
1245EXPORT_SYMBOL(sock_create_lite);
1246
1247/* No kernel lock held - perfect */
1248static __poll_t sock_poll(struct file *file, poll_table *wait)
1249{
1250 struct socket *sock = file->private_data;
1251 __poll_t events = poll_requested_events(wait), flag = 0;
1252
1253 if (!sock->ops->poll)
1254 return 0;
1255
1256 if (sk_can_busy_loop(sock->sk)) {
1257 /* poll once if requested by the syscall */
1258 if (events & POLL_BUSY_LOOP)
1259 sk_busy_loop(sock->sk, 1);
1260
1261 /* if this socket can poll_ll, tell the system call */
1262 flag = POLL_BUSY_LOOP;
1263 }
1264
1265 return sock->ops->poll(file, sock, wait) | flag;
1266}
1267
1268static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1269{
1270 struct socket *sock = file->private_data;
1271
1272 return sock->ops->mmap(file, sock, vma);
1273}
1274
1275static int sock_close(struct inode *inode, struct file *filp)
1276{
1277 __sock_release(SOCKET_I(inode), inode);
1278 return 0;
1279}
1280
1281/*
1282 * Update the socket async list
1283 *
1284 * Fasync_list locking strategy.
1285 *
1286 * 1. fasync_list is modified only under process context socket lock
1287 * i.e. under semaphore.
1288 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1289 * or under socket lock
1290 */
1291
1292static int sock_fasync(int fd, struct file *filp, int on)
1293{
1294 struct socket *sock = filp->private_data;
1295 struct sock *sk = sock->sk;
1296 struct socket_wq *wq = &sock->wq;
1297
1298 if (sk == NULL)
1299 return -EINVAL;
1300
1301 lock_sock(sk);
1302 fasync_helper(fd, filp, on, &wq->fasync_list);
1303
1304 if (!wq->fasync_list)
1305 sock_reset_flag(sk, SOCK_FASYNC);
1306 else
1307 sock_set_flag(sk, SOCK_FASYNC);
1308
1309 release_sock(sk);
1310 return 0;
1311}
1312
1313/* This function may be called only under rcu_lock */
1314
1315int sock_wake_async(struct socket_wq *wq, int how, int band)
1316{
1317 if (!wq || !wq->fasync_list)
1318 return -1;
1319
1320 switch (how) {
1321 case SOCK_WAKE_WAITD:
1322 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1323 break;
1324 goto call_kill;
1325 case SOCK_WAKE_SPACE:
1326 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1327 break;
1328 fallthrough;
1329 case SOCK_WAKE_IO:
1330call_kill:
1331 kill_fasync(&wq->fasync_list, SIGIO, band);
1332 break;
1333 case SOCK_WAKE_URG:
1334 kill_fasync(&wq->fasync_list, SIGURG, band);
1335 }
1336
1337 return 0;
1338}
1339EXPORT_SYMBOL(sock_wake_async);
1340
1341/**
1342 * __sock_create - creates a socket
1343 * @net: net namespace
1344 * @family: protocol family (AF_INET, ...)
1345 * @type: communication type (SOCK_STREAM, ...)
1346 * @protocol: protocol (0, ...)
1347 * @res: new socket
1348 * @kern: boolean for kernel space sockets
1349 *
1350 * Creates a new socket and assigns it to @res, passing through LSM.
1351 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1352 * be set to true if the socket resides in kernel space.
1353 * This function internally uses GFP_KERNEL.
1354 */
1355
1356int __sock_create(struct net *net, int family, int type, int protocol,
1357 struct socket **res, int kern)
1358{
1359 int err;
1360 struct socket *sock;
1361 const struct net_proto_family *pf;
1362
1363 /*
1364 * Check protocol is in range
1365 */
1366 if (family < 0 || family >= NPROTO)
1367 return -EAFNOSUPPORT;
1368 if (type < 0 || type >= SOCK_MAX)
1369 return -EINVAL;
1370
1371 /* Compatibility.
1372
1373 This uglymoron is moved from INET layer to here to avoid
1374 deadlock in module load.
1375 */
1376 if (family == PF_INET && type == SOCK_PACKET) {
1377 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1378 current->comm);
1379 family = PF_PACKET;
1380 }
1381
1382 err = security_socket_create(family, type, protocol, kern);
1383 if (err)
1384 return err;
1385
1386 /*
1387 * Allocate the socket and allow the family to set things up. if
1388 * the protocol is 0, the family is instructed to select an appropriate
1389 * default.
1390 */
1391 sock = sock_alloc();
1392 if (!sock) {
1393 net_warn_ratelimited("socket: no more sockets\n");
1394 return -ENFILE; /* Not exactly a match, but its the
1395 closest posix thing */
1396 }
1397
1398 sock->type = type;
1399
1400#ifdef CONFIG_MODULES
1401 /* Attempt to load a protocol module if the find failed.
1402 *
1403 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1404 * requested real, full-featured networking support upon configuration.
1405 * Otherwise module support will break!
1406 */
1407 if (rcu_access_pointer(net_families[family]) == NULL)
1408 request_module("net-pf-%d", family);
1409#endif
1410
1411 rcu_read_lock();
1412 pf = rcu_dereference(net_families[family]);
1413 err = -EAFNOSUPPORT;
1414 if (!pf)
1415 goto out_release;
1416
1417 /*
1418 * We will call the ->create function, that possibly is in a loadable
1419 * module, so we have to bump that loadable module refcnt first.
1420 */
1421 if (!try_module_get(pf->owner))
1422 goto out_release;
1423
1424 /* Now protected by module ref count */
1425 rcu_read_unlock();
1426
1427 err = pf->create(net, sock, protocol, kern);
1428 if (err < 0)
1429 goto out_module_put;
1430
1431 /*
1432 * Now to bump the refcnt of the [loadable] module that owns this
1433 * socket at sock_release time we decrement its refcnt.
1434 */
1435 if (!try_module_get(sock->ops->owner))
1436 goto out_module_busy;
1437
1438 /*
1439 * Now that we're done with the ->create function, the [loadable]
1440 * module can have its refcnt decremented
1441 */
1442 module_put(pf->owner);
1443 err = security_socket_post_create(sock, family, type, protocol, kern);
1444 if (err)
1445 goto out_sock_release;
1446 *res = sock;
1447
1448 return 0;
1449
1450out_module_busy:
1451 err = -EAFNOSUPPORT;
1452out_module_put:
1453 sock->ops = NULL;
1454 module_put(pf->owner);
1455out_sock_release:
1456 sock_release(sock);
1457 return err;
1458
1459out_release:
1460 rcu_read_unlock();
1461 goto out_sock_release;
1462}
1463EXPORT_SYMBOL(__sock_create);
1464
1465/**
1466 * sock_create - creates a socket
1467 * @family: protocol family (AF_INET, ...)
1468 * @type: communication type (SOCK_STREAM, ...)
1469 * @protocol: protocol (0, ...)
1470 * @res: new socket
1471 *
1472 * A wrapper around __sock_create().
1473 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1474 */
1475
1476int sock_create(int family, int type, int protocol, struct socket **res)
1477{
1478 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1479}
1480EXPORT_SYMBOL(sock_create);
1481
1482/**
1483 * sock_create_kern - creates a socket (kernel space)
1484 * @net: net namespace
1485 * @family: protocol family (AF_INET, ...)
1486 * @type: communication type (SOCK_STREAM, ...)
1487 * @protocol: protocol (0, ...)
1488 * @res: new socket
1489 *
1490 * A wrapper around __sock_create().
1491 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1492 */
1493
1494int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1495{
1496 return __sock_create(net, family, type, protocol, res, 1);
1497}
1498EXPORT_SYMBOL(sock_create_kern);
1499
1500int __sys_socket(int family, int type, int protocol)
1501{
1502 int retval;
1503 struct socket *sock;
1504 int flags;
1505
1506 /* Check the SOCK_* constants for consistency. */
1507 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1508 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1509 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1510 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1511
1512 flags = type & ~SOCK_TYPE_MASK;
1513 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1514 return -EINVAL;
1515 type &= SOCK_TYPE_MASK;
1516
1517 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1518 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1519
1520 retval = sock_create(family, type, protocol, &sock);
1521 if (retval < 0)
1522 return retval;
1523
1524 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1525}
1526
1527SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1528{
1529 return __sys_socket(family, type, protocol);
1530}
1531
1532/*
1533 * Create a pair of connected sockets.
1534 */
1535
1536int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1537{
1538 struct socket *sock1, *sock2;
1539 int fd1, fd2, err;
1540 struct file *newfile1, *newfile2;
1541 int flags;
1542
1543 flags = type & ~SOCK_TYPE_MASK;
1544 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1545 return -EINVAL;
1546 type &= SOCK_TYPE_MASK;
1547
1548 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1549 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1550
1551 /*
1552 * reserve descriptors and make sure we won't fail
1553 * to return them to userland.
1554 */
1555 fd1 = get_unused_fd_flags(flags);
1556 if (unlikely(fd1 < 0))
1557 return fd1;
1558
1559 fd2 = get_unused_fd_flags(flags);
1560 if (unlikely(fd2 < 0)) {
1561 put_unused_fd(fd1);
1562 return fd2;
1563 }
1564
1565 err = put_user(fd1, &usockvec[0]);
1566 if (err)
1567 goto out;
1568
1569 err = put_user(fd2, &usockvec[1]);
1570 if (err)
1571 goto out;
1572
1573 /*
1574 * Obtain the first socket and check if the underlying protocol
1575 * supports the socketpair call.
1576 */
1577
1578 err = sock_create(family, type, protocol, &sock1);
1579 if (unlikely(err < 0))
1580 goto out;
1581
1582 err = sock_create(family, type, protocol, &sock2);
1583 if (unlikely(err < 0)) {
1584 sock_release(sock1);
1585 goto out;
1586 }
1587
1588 err = security_socket_socketpair(sock1, sock2);
1589 if (unlikely(err)) {
1590 sock_release(sock2);
1591 sock_release(sock1);
1592 goto out;
1593 }
1594
1595 err = sock1->ops->socketpair(sock1, sock2);
1596 if (unlikely(err < 0)) {
1597 sock_release(sock2);
1598 sock_release(sock1);
1599 goto out;
1600 }
1601
1602 newfile1 = sock_alloc_file(sock1, flags, NULL);
1603 if (IS_ERR(newfile1)) {
1604 err = PTR_ERR(newfile1);
1605 sock_release(sock2);
1606 goto out;
1607 }
1608
1609 newfile2 = sock_alloc_file(sock2, flags, NULL);
1610 if (IS_ERR(newfile2)) {
1611 err = PTR_ERR(newfile2);
1612 fput(newfile1);
1613 goto out;
1614 }
1615
1616 audit_fd_pair(fd1, fd2);
1617
1618 fd_install(fd1, newfile1);
1619 fd_install(fd2, newfile2);
1620 return 0;
1621
1622out:
1623 put_unused_fd(fd2);
1624 put_unused_fd(fd1);
1625 return err;
1626}
1627
1628SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1629 int __user *, usockvec)
1630{
1631 return __sys_socketpair(family, type, protocol, usockvec);
1632}
1633
1634/*
1635 * Bind a name to a socket. Nothing much to do here since it's
1636 * the protocol's responsibility to handle the local address.
1637 *
1638 * We move the socket address to kernel space before we call
1639 * the protocol layer (having also checked the address is ok).
1640 */
1641
1642int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1643{
1644 struct socket *sock;
1645 struct sockaddr_storage address;
1646 int err, fput_needed;
1647
1648 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1649 if (sock) {
1650 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1651 if (!err) {
1652 err = security_socket_bind(sock,
1653 (struct sockaddr *)&address,
1654 addrlen);
1655 if (!err)
1656 err = sock->ops->bind(sock,
1657 (struct sockaddr *)
1658 &address, addrlen);
1659 }
1660 fput_light(sock->file, fput_needed);
1661 }
1662 return err;
1663}
1664
1665SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1666{
1667 return __sys_bind(fd, umyaddr, addrlen);
1668}
1669
1670/*
1671 * Perform a listen. Basically, we allow the protocol to do anything
1672 * necessary for a listen, and if that works, we mark the socket as
1673 * ready for listening.
1674 */
1675
1676int __sys_listen(int fd, int backlog)
1677{
1678 struct socket *sock;
1679 int err, fput_needed;
1680 int somaxconn;
1681
1682 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1683 if (sock) {
1684 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1685 if ((unsigned int)backlog > somaxconn)
1686 backlog = somaxconn;
1687
1688 err = security_socket_listen(sock, backlog);
1689 if (!err)
1690 err = sock->ops->listen(sock, backlog);
1691
1692 fput_light(sock->file, fput_needed);
1693 }
1694 return err;
1695}
1696
1697SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1698{
1699 return __sys_listen(fd, backlog);
1700}
1701
1702int __sys_accept4_file(struct file *file, unsigned file_flags,
1703 struct sockaddr __user *upeer_sockaddr,
1704 int __user *upeer_addrlen, int flags,
1705 unsigned long nofile)
1706{
1707 struct socket *sock, *newsock;
1708 struct file *newfile;
1709 int err, len, newfd;
1710 struct sockaddr_storage address;
1711
1712 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1713 return -EINVAL;
1714
1715 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1716 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1717
1718 sock = sock_from_file(file, &err);
1719 if (!sock)
1720 goto out;
1721
1722 err = -ENFILE;
1723 newsock = sock_alloc();
1724 if (!newsock)
1725 goto out;
1726
1727 newsock->type = sock->type;
1728 newsock->ops = sock->ops;
1729
1730 /*
1731 * We don't need try_module_get here, as the listening socket (sock)
1732 * has the protocol module (sock->ops->owner) held.
1733 */
1734 __module_get(newsock->ops->owner);
1735
1736 newfd = __get_unused_fd_flags(flags, nofile);
1737 if (unlikely(newfd < 0)) {
1738 err = newfd;
1739 sock_release(newsock);
1740 goto out;
1741 }
1742 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1743 if (IS_ERR(newfile)) {
1744 err = PTR_ERR(newfile);
1745 put_unused_fd(newfd);
1746 goto out;
1747 }
1748
1749 err = security_socket_accept(sock, newsock);
1750 if (err)
1751 goto out_fd;
1752
1753 err = sock->ops->accept(sock, newsock, sock->file->f_flags | file_flags,
1754 false);
1755 if (err < 0)
1756 goto out_fd;
1757
1758 if (upeer_sockaddr) {
1759 len = newsock->ops->getname(newsock,
1760 (struct sockaddr *)&address, 2);
1761 if (len < 0) {
1762 err = -ECONNABORTED;
1763 goto out_fd;
1764 }
1765 err = move_addr_to_user(&address,
1766 len, upeer_sockaddr, upeer_addrlen);
1767 if (err < 0)
1768 goto out_fd;
1769 }
1770
1771 /* File flags are not inherited via accept() unlike another OSes. */
1772
1773 fd_install(newfd, newfile);
1774 err = newfd;
1775out:
1776 return err;
1777out_fd:
1778 fput(newfile);
1779 put_unused_fd(newfd);
1780 goto out;
1781
1782}
1783
1784/*
1785 * For accept, we attempt to create a new socket, set up the link
1786 * with the client, wake up the client, then return the new
1787 * connected fd. We collect the address of the connector in kernel
1788 * space and move it to user at the very end. This is unclean because
1789 * we open the socket then return an error.
1790 *
1791 * 1003.1g adds the ability to recvmsg() to query connection pending
1792 * status to recvmsg. We need to add that support in a way thats
1793 * clean when we restructure accept also.
1794 */
1795
1796int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1797 int __user *upeer_addrlen, int flags)
1798{
1799 int ret = -EBADF;
1800 struct fd f;
1801
1802 f = fdget(fd);
1803 if (f.file) {
1804 ret = __sys_accept4_file(f.file, 0, upeer_sockaddr,
1805 upeer_addrlen, flags,
1806 rlimit(RLIMIT_NOFILE));
1807 fdput(f);
1808 }
1809
1810 return ret;
1811}
1812
1813SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1814 int __user *, upeer_addrlen, int, flags)
1815{
1816 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1817}
1818
1819SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1820 int __user *, upeer_addrlen)
1821{
1822 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1823}
1824
1825/*
1826 * Attempt to connect to a socket with the server address. The address
1827 * is in user space so we verify it is OK and move it to kernel space.
1828 *
1829 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1830 * break bindings
1831 *
1832 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1833 * other SEQPACKET protocols that take time to connect() as it doesn't
1834 * include the -EINPROGRESS status for such sockets.
1835 */
1836
1837int __sys_connect_file(struct file *file, struct sockaddr_storage *address,
1838 int addrlen, int file_flags)
1839{
1840 struct socket *sock;
1841 int err;
1842
1843 sock = sock_from_file(file, &err);
1844 if (!sock)
1845 goto out;
1846
1847 err =
1848 security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1849 if (err)
1850 goto out;
1851
1852 err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
1853 sock->file->f_flags | file_flags);
1854out:
1855 return err;
1856}
1857
1858int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1859{
1860 int ret = -EBADF;
1861 struct fd f;
1862
1863 f = fdget(fd);
1864 if (f.file) {
1865 struct sockaddr_storage address;
1866
1867 ret = move_addr_to_kernel(uservaddr, addrlen, &address);
1868 if (!ret)
1869 ret = __sys_connect_file(f.file, &address, addrlen, 0);
1870 fdput(f);
1871 }
1872
1873 return ret;
1874}
1875
1876SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1877 int, addrlen)
1878{
1879 return __sys_connect(fd, uservaddr, addrlen);
1880}
1881
1882/*
1883 * Get the local address ('name') of a socket object. Move the obtained
1884 * name to user space.
1885 */
1886
1887int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1888 int __user *usockaddr_len)
1889{
1890 struct socket *sock;
1891 struct sockaddr_storage address;
1892 int err, fput_needed;
1893
1894 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1895 if (!sock)
1896 goto out;
1897
1898 err = security_socket_getsockname(sock);
1899 if (err)
1900 goto out_put;
1901
1902 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1903 if (err < 0)
1904 goto out_put;
1905 /* "err" is actually length in this case */
1906 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1907
1908out_put:
1909 fput_light(sock->file, fput_needed);
1910out:
1911 return err;
1912}
1913
1914SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1915 int __user *, usockaddr_len)
1916{
1917 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1918}
1919
1920/*
1921 * Get the remote address ('name') of a socket object. Move the obtained
1922 * name to user space.
1923 */
1924
1925int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1926 int __user *usockaddr_len)
1927{
1928 struct socket *sock;
1929 struct sockaddr_storage address;
1930 int err, fput_needed;
1931
1932 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1933 if (sock != NULL) {
1934 err = security_socket_getpeername(sock);
1935 if (err) {
1936 fput_light(sock->file, fput_needed);
1937 return err;
1938 }
1939
1940 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1941 if (err >= 0)
1942 /* "err" is actually length in this case */
1943 err = move_addr_to_user(&address, err, usockaddr,
1944 usockaddr_len);
1945 fput_light(sock->file, fput_needed);
1946 }
1947 return err;
1948}
1949
1950SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1951 int __user *, usockaddr_len)
1952{
1953 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1954}
1955
1956/*
1957 * Send a datagram to a given address. We move the address into kernel
1958 * space and check the user space data area is readable before invoking
1959 * the protocol.
1960 */
1961int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1962 struct sockaddr __user *addr, int addr_len)
1963{
1964 struct socket *sock;
1965 struct sockaddr_storage address;
1966 int err;
1967 struct msghdr msg;
1968 struct iovec iov;
1969 int fput_needed;
1970
1971 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1972 if (unlikely(err))
1973 return err;
1974 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1975 if (!sock)
1976 goto out;
1977
1978 msg.msg_name = NULL;
1979 msg.msg_control = NULL;
1980 msg.msg_controllen = 0;
1981 msg.msg_namelen = 0;
1982 if (addr) {
1983 err = move_addr_to_kernel(addr, addr_len, &address);
1984 if (err < 0)
1985 goto out_put;
1986 msg.msg_name = (struct sockaddr *)&address;
1987 msg.msg_namelen = addr_len;
1988 }
1989 if (sock->file->f_flags & O_NONBLOCK)
1990 flags |= MSG_DONTWAIT;
1991 msg.msg_flags = flags;
1992 err = sock_sendmsg(sock, &msg);
1993
1994out_put:
1995 fput_light(sock->file, fput_needed);
1996out:
1997 return err;
1998}
1999
2000SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
2001 unsigned int, flags, struct sockaddr __user *, addr,
2002 int, addr_len)
2003{
2004 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
2005}
2006
2007/*
2008 * Send a datagram down a socket.
2009 */
2010
2011SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
2012 unsigned int, flags)
2013{
2014 return __sys_sendto(fd, buff, len, flags, NULL, 0);
2015}
2016
2017/*
2018 * Receive a frame from the socket and optionally record the address of the
2019 * sender. We verify the buffers are writable and if needed move the
2020 * sender address from kernel to user space.
2021 */
2022int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
2023 struct sockaddr __user *addr, int __user *addr_len)
2024{
2025 struct socket *sock;
2026 struct iovec iov;
2027 struct msghdr msg;
2028 struct sockaddr_storage address;
2029 int err, err2;
2030 int fput_needed;
2031
2032 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
2033 if (unlikely(err))
2034 return err;
2035 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2036 if (!sock)
2037 goto out;
2038
2039 msg.msg_control = NULL;
2040 msg.msg_controllen = 0;
2041 /* Save some cycles and don't copy the address if not needed */
2042 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2043 /* We assume all kernel code knows the size of sockaddr_storage */
2044 msg.msg_namelen = 0;
2045 msg.msg_iocb = NULL;
2046 msg.msg_flags = 0;
2047 if (sock->file->f_flags & O_NONBLOCK)
2048 flags |= MSG_DONTWAIT;
2049 err = sock_recvmsg(sock, &msg, flags);
2050
2051 if (err >= 0 && addr != NULL) {
2052 err2 = move_addr_to_user(&address,
2053 msg.msg_namelen, addr, addr_len);
2054 if (err2 < 0)
2055 err = err2;
2056 }
2057
2058 fput_light(sock->file, fput_needed);
2059out:
2060 return err;
2061}
2062
2063SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2064 unsigned int, flags, struct sockaddr __user *, addr,
2065 int __user *, addr_len)
2066{
2067 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2068}
2069
2070/*
2071 * Receive a datagram from a socket.
2072 */
2073
2074SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2075 unsigned int, flags)
2076{
2077 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2078}
2079
2080static bool sock_use_custom_sol_socket(const struct socket *sock)
2081{
2082 const struct sock *sk = sock->sk;
2083
2084 /* Use sock->ops->setsockopt() for MPTCP */
2085 return IS_ENABLED(CONFIG_MPTCP) &&
2086 sk->sk_protocol == IPPROTO_MPTCP &&
2087 sk->sk_type == SOCK_STREAM &&
2088 (sk->sk_family == AF_INET || sk->sk_family == AF_INET6);
2089}
2090
2091/*
2092 * Set a socket option. Because we don't know the option lengths we have
2093 * to pass the user mode parameter for the protocols to sort out.
2094 */
2095int __sys_setsockopt(int fd, int level, int optname, char __user *user_optval,
2096 int optlen)
2097{
2098 sockptr_t optval = USER_SOCKPTR(user_optval);
2099 char *kernel_optval = NULL;
2100 int err, fput_needed;
2101 struct socket *sock;
2102
2103 if (optlen < 0)
2104 return -EINVAL;
2105
2106 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2107 if (!sock)
2108 return err;
2109
2110 err = security_socket_setsockopt(sock, level, optname);
2111 if (err)
2112 goto out_put;
2113
2114 if (!in_compat_syscall())
2115 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level, &optname,
2116 user_optval, &optlen,
2117 &kernel_optval);
2118 if (err < 0)
2119 goto out_put;
2120 if (err > 0) {
2121 err = 0;
2122 goto out_put;
2123 }
2124
2125 if (kernel_optval)
2126 optval = KERNEL_SOCKPTR(kernel_optval);
2127 if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
2128 err = sock_setsockopt(sock, level, optname, optval, optlen);
2129 else if (unlikely(!sock->ops->setsockopt))
2130 err = -EOPNOTSUPP;
2131 else
2132 err = sock->ops->setsockopt(sock, level, optname, optval,
2133 optlen);
2134 kfree(kernel_optval);
2135out_put:
2136 fput_light(sock->file, fput_needed);
2137 return err;
2138}
2139
2140SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2141 char __user *, optval, int, optlen)
2142{
2143 return __sys_setsockopt(fd, level, optname, optval, optlen);
2144}
2145
2146/*
2147 * Get a socket option. Because we don't know the option lengths we have
2148 * to pass a user mode parameter for the protocols to sort out.
2149 */
2150int __sys_getsockopt(int fd, int level, int optname, char __user *optval,
2151 int __user *optlen)
2152{
2153 int err, fput_needed;
2154 struct socket *sock;
2155 int max_optlen;
2156
2157 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2158 if (!sock)
2159 return err;
2160
2161 err = security_socket_getsockopt(sock, level, optname);
2162 if (err)
2163 goto out_put;
2164
2165 if (!in_compat_syscall())
2166 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2167
2168 if (level == SOL_SOCKET)
2169 err = sock_getsockopt(sock, level, optname, optval, optlen);
2170 else if (unlikely(!sock->ops->getsockopt))
2171 err = -EOPNOTSUPP;
2172 else
2173 err = sock->ops->getsockopt(sock, level, optname, optval,
2174 optlen);
2175
2176 if (!in_compat_syscall())
2177 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2178 optval, optlen, max_optlen,
2179 err);
2180out_put:
2181 fput_light(sock->file, fput_needed);
2182 return err;
2183}
2184
2185SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2186 char __user *, optval, int __user *, optlen)
2187{
2188 return __sys_getsockopt(fd, level, optname, optval, optlen);
2189}
2190
2191/*
2192 * Shutdown a socket.
2193 */
2194
2195int __sys_shutdown(int fd, int how)
2196{
2197 int err, fput_needed;
2198 struct socket *sock;
2199
2200 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2201 if (sock != NULL) {
2202 err = security_socket_shutdown(sock, how);
2203 if (!err)
2204 err = sock->ops->shutdown(sock, how);
2205 fput_light(sock->file, fput_needed);
2206 }
2207 return err;
2208}
2209
2210SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2211{
2212 return __sys_shutdown(fd, how);
2213}
2214
2215/* A couple of helpful macros for getting the address of the 32/64 bit
2216 * fields which are the same type (int / unsigned) on our platforms.
2217 */
2218#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2219#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2220#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2221
2222struct used_address {
2223 struct sockaddr_storage name;
2224 unsigned int name_len;
2225};
2226
2227int __copy_msghdr_from_user(struct msghdr *kmsg,
2228 struct user_msghdr __user *umsg,
2229 struct sockaddr __user **save_addr,
2230 struct iovec __user **uiov, size_t *nsegs)
2231{
2232 struct user_msghdr msg;
2233 ssize_t err;
2234
2235 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2236 return -EFAULT;
2237
2238 kmsg->msg_control_is_user = true;
2239 kmsg->msg_control_user = msg.msg_control;
2240 kmsg->msg_controllen = msg.msg_controllen;
2241 kmsg->msg_flags = msg.msg_flags;
2242
2243 kmsg->msg_namelen = msg.msg_namelen;
2244 if (!msg.msg_name)
2245 kmsg->msg_namelen = 0;
2246
2247 if (kmsg->msg_namelen < 0)
2248 return -EINVAL;
2249
2250 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2251 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2252
2253 if (save_addr)
2254 *save_addr = msg.msg_name;
2255
2256 if (msg.msg_name && kmsg->msg_namelen) {
2257 if (!save_addr) {
2258 err = move_addr_to_kernel(msg.msg_name,
2259 kmsg->msg_namelen,
2260 kmsg->msg_name);
2261 if (err < 0)
2262 return err;
2263 }
2264 } else {
2265 kmsg->msg_name = NULL;
2266 kmsg->msg_namelen = 0;
2267 }
2268
2269 if (msg.msg_iovlen > UIO_MAXIOV)
2270 return -EMSGSIZE;
2271
2272 kmsg->msg_iocb = NULL;
2273 *uiov = msg.msg_iov;
2274 *nsegs = msg.msg_iovlen;
2275 return 0;
2276}
2277
2278static int copy_msghdr_from_user(struct msghdr *kmsg,
2279 struct user_msghdr __user *umsg,
2280 struct sockaddr __user **save_addr,
2281 struct iovec **iov)
2282{
2283 struct user_msghdr msg;
2284 ssize_t err;
2285
2286 err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
2287 &msg.msg_iovlen);
2288 if (err)
2289 return err;
2290
2291 err = import_iovec(save_addr ? READ : WRITE,
2292 msg.msg_iov, msg.msg_iovlen,
2293 UIO_FASTIOV, iov, &kmsg->msg_iter);
2294 return err < 0 ? err : 0;
2295}
2296
2297static int ____sys_sendmsg(struct socket *sock, struct msghdr *msg_sys,
2298 unsigned int flags, struct used_address *used_address,
2299 unsigned int allowed_msghdr_flags)
2300{
2301 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2302 __aligned(sizeof(__kernel_size_t));
2303 /* 20 is size of ipv6_pktinfo */
2304 unsigned char *ctl_buf = ctl;
2305 int ctl_len;
2306 ssize_t err;
2307
2308 err = -ENOBUFS;
2309
2310 if (msg_sys->msg_controllen > INT_MAX)
2311 goto out;
2312 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2313 ctl_len = msg_sys->msg_controllen;
2314 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2315 err =
2316 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2317 sizeof(ctl));
2318 if (err)
2319 goto out;
2320 ctl_buf = msg_sys->msg_control;
2321 ctl_len = msg_sys->msg_controllen;
2322 } else if (ctl_len) {
2323 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2324 CMSG_ALIGN(sizeof(struct cmsghdr)));
2325 if (ctl_len > sizeof(ctl)) {
2326 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2327 if (ctl_buf == NULL)
2328 goto out;
2329 }
2330 err = -EFAULT;
2331 if (copy_from_user(ctl_buf, msg_sys->msg_control_user, ctl_len))
2332 goto out_freectl;
2333 msg_sys->msg_control = ctl_buf;
2334 msg_sys->msg_control_is_user = false;
2335 }
2336 msg_sys->msg_flags = flags;
2337
2338 if (sock->file->f_flags & O_NONBLOCK)
2339 msg_sys->msg_flags |= MSG_DONTWAIT;
2340 /*
2341 * If this is sendmmsg() and current destination address is same as
2342 * previously succeeded address, omit asking LSM's decision.
2343 * used_address->name_len is initialized to UINT_MAX so that the first
2344 * destination address never matches.
2345 */
2346 if (used_address && msg_sys->msg_name &&
2347 used_address->name_len == msg_sys->msg_namelen &&
2348 !memcmp(&used_address->name, msg_sys->msg_name,
2349 used_address->name_len)) {
2350 err = sock_sendmsg_nosec(sock, msg_sys);
2351 goto out_freectl;
2352 }
2353 err = sock_sendmsg(sock, msg_sys);
2354 /*
2355 * If this is sendmmsg() and sending to current destination address was
2356 * successful, remember it.
2357 */
2358 if (used_address && err >= 0) {
2359 used_address->name_len = msg_sys->msg_namelen;
2360 if (msg_sys->msg_name)
2361 memcpy(&used_address->name, msg_sys->msg_name,
2362 used_address->name_len);
2363 }
2364
2365out_freectl:
2366 if (ctl_buf != ctl)
2367 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2368out:
2369 return err;
2370}
2371
2372int sendmsg_copy_msghdr(struct msghdr *msg,
2373 struct user_msghdr __user *umsg, unsigned flags,
2374 struct iovec **iov)
2375{
2376 int err;
2377
2378 if (flags & MSG_CMSG_COMPAT) {
2379 struct compat_msghdr __user *msg_compat;
2380
2381 msg_compat = (struct compat_msghdr __user *) umsg;
2382 err = get_compat_msghdr(msg, msg_compat, NULL, iov);
2383 } else {
2384 err = copy_msghdr_from_user(msg, umsg, NULL, iov);
2385 }
2386 if (err < 0)
2387 return err;
2388
2389 return 0;
2390}
2391
2392static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2393 struct msghdr *msg_sys, unsigned int flags,
2394 struct used_address *used_address,
2395 unsigned int allowed_msghdr_flags)
2396{
2397 struct sockaddr_storage address;
2398 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2399 ssize_t err;
2400
2401 msg_sys->msg_name = &address;
2402
2403 err = sendmsg_copy_msghdr(msg_sys, msg, flags, &iov);
2404 if (err < 0)
2405 return err;
2406
2407 err = ____sys_sendmsg(sock, msg_sys, flags, used_address,
2408 allowed_msghdr_flags);
2409 kfree(iov);
2410 return err;
2411}
2412
2413/*
2414 * BSD sendmsg interface
2415 */
2416long __sys_sendmsg_sock(struct socket *sock, struct msghdr *msg,
2417 unsigned int flags)
2418{
2419 /* disallow ancillary data requests from this path */
2420 if (msg->msg_control || msg->msg_controllen)
2421 return -EINVAL;
2422
2423 return ____sys_sendmsg(sock, msg, flags, NULL, 0);
2424}
2425
2426long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2427 bool forbid_cmsg_compat)
2428{
2429 int fput_needed, err;
2430 struct msghdr msg_sys;
2431 struct socket *sock;
2432
2433 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2434 return -EINVAL;
2435
2436 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2437 if (!sock)
2438 goto out;
2439
2440 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2441
2442 fput_light(sock->file, fput_needed);
2443out:
2444 return err;
2445}
2446
2447SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2448{
2449 return __sys_sendmsg(fd, msg, flags, true);
2450}
2451
2452/*
2453 * Linux sendmmsg interface
2454 */
2455
2456int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2457 unsigned int flags, bool forbid_cmsg_compat)
2458{
2459 int fput_needed, err, datagrams;
2460 struct socket *sock;
2461 struct mmsghdr __user *entry;
2462 struct compat_mmsghdr __user *compat_entry;
2463 struct msghdr msg_sys;
2464 struct used_address used_address;
2465 unsigned int oflags = flags;
2466
2467 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2468 return -EINVAL;
2469
2470 if (vlen > UIO_MAXIOV)
2471 vlen = UIO_MAXIOV;
2472
2473 datagrams = 0;
2474
2475 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2476 if (!sock)
2477 return err;
2478
2479 used_address.name_len = UINT_MAX;
2480 entry = mmsg;
2481 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2482 err = 0;
2483 flags |= MSG_BATCH;
2484
2485 while (datagrams < vlen) {
2486 if (datagrams == vlen - 1)
2487 flags = oflags;
2488
2489 if (MSG_CMSG_COMPAT & flags) {
2490 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2491 &msg_sys, flags, &used_address, MSG_EOR);
2492 if (err < 0)
2493 break;
2494 err = __put_user(err, &compat_entry->msg_len);
2495 ++compat_entry;
2496 } else {
2497 err = ___sys_sendmsg(sock,
2498 (struct user_msghdr __user *)entry,
2499 &msg_sys, flags, &used_address, MSG_EOR);
2500 if (err < 0)
2501 break;
2502 err = put_user(err, &entry->msg_len);
2503 ++entry;
2504 }
2505
2506 if (err)
2507 break;
2508 ++datagrams;
2509 if (msg_data_left(&msg_sys))
2510 break;
2511 cond_resched();
2512 }
2513
2514 fput_light(sock->file, fput_needed);
2515
2516 /* We only return an error if no datagrams were able to be sent */
2517 if (datagrams != 0)
2518 return datagrams;
2519
2520 return err;
2521}
2522
2523SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2524 unsigned int, vlen, unsigned int, flags)
2525{
2526 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2527}
2528
2529int recvmsg_copy_msghdr(struct msghdr *msg,
2530 struct user_msghdr __user *umsg, unsigned flags,
2531 struct sockaddr __user **uaddr,
2532 struct iovec **iov)
2533{
2534 ssize_t err;
2535
2536 if (MSG_CMSG_COMPAT & flags) {
2537 struct compat_msghdr __user *msg_compat;
2538
2539 msg_compat = (struct compat_msghdr __user *) umsg;
2540 err = get_compat_msghdr(msg, msg_compat, uaddr, iov);
2541 } else {
2542 err = copy_msghdr_from_user(msg, umsg, uaddr, iov);
2543 }
2544 if (err < 0)
2545 return err;
2546
2547 return 0;
2548}
2549
2550static int ____sys_recvmsg(struct socket *sock, struct msghdr *msg_sys,
2551 struct user_msghdr __user *msg,
2552 struct sockaddr __user *uaddr,
2553 unsigned int flags, int nosec)
2554{
2555 struct compat_msghdr __user *msg_compat =
2556 (struct compat_msghdr __user *) msg;
2557 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2558 struct sockaddr_storage addr;
2559 unsigned long cmsg_ptr;
2560 int len;
2561 ssize_t err;
2562
2563 msg_sys->msg_name = &addr;
2564 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2565 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2566
2567 /* We assume all kernel code knows the size of sockaddr_storage */
2568 msg_sys->msg_namelen = 0;
2569
2570 if (sock->file->f_flags & O_NONBLOCK)
2571 flags |= MSG_DONTWAIT;
2572
2573 if (unlikely(nosec))
2574 err = sock_recvmsg_nosec(sock, msg_sys, flags);
2575 else
2576 err = sock_recvmsg(sock, msg_sys, flags);
2577
2578 if (err < 0)
2579 goto out;
2580 len = err;
2581
2582 if (uaddr != NULL) {
2583 err = move_addr_to_user(&addr,
2584 msg_sys->msg_namelen, uaddr,
2585 uaddr_len);
2586 if (err < 0)
2587 goto out;
2588 }
2589 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2590 COMPAT_FLAGS(msg));
2591 if (err)
2592 goto out;
2593 if (MSG_CMSG_COMPAT & flags)
2594 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2595 &msg_compat->msg_controllen);
2596 else
2597 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2598 &msg->msg_controllen);
2599 if (err)
2600 goto out;
2601 err = len;
2602out:
2603 return err;
2604}
2605
2606static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2607 struct msghdr *msg_sys, unsigned int flags, int nosec)
2608{
2609 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2610 /* user mode address pointers */
2611 struct sockaddr __user *uaddr;
2612 ssize_t err;
2613
2614 err = recvmsg_copy_msghdr(msg_sys, msg, flags, &uaddr, &iov);
2615 if (err < 0)
2616 return err;
2617
2618 err = ____sys_recvmsg(sock, msg_sys, msg, uaddr, flags, nosec);
2619 kfree(iov);
2620 return err;
2621}
2622
2623/*
2624 * BSD recvmsg interface
2625 */
2626
2627long __sys_recvmsg_sock(struct socket *sock, struct msghdr *msg,
2628 struct user_msghdr __user *umsg,
2629 struct sockaddr __user *uaddr, unsigned int flags)
2630{
2631 /* disallow ancillary data requests from this path */
2632 if (msg->msg_control || msg->msg_controllen)
2633 return -EINVAL;
2634
2635 return ____sys_recvmsg(sock, msg, umsg, uaddr, flags, 0);
2636}
2637
2638long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2639 bool forbid_cmsg_compat)
2640{
2641 int fput_needed, err;
2642 struct msghdr msg_sys;
2643 struct socket *sock;
2644
2645 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2646 return -EINVAL;
2647
2648 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2649 if (!sock)
2650 goto out;
2651
2652 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2653
2654 fput_light(sock->file, fput_needed);
2655out:
2656 return err;
2657}
2658
2659SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2660 unsigned int, flags)
2661{
2662 return __sys_recvmsg(fd, msg, flags, true);
2663}
2664
2665/*
2666 * Linux recvmmsg interface
2667 */
2668
2669static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2670 unsigned int vlen, unsigned int flags,
2671 struct timespec64 *timeout)
2672{
2673 int fput_needed, err, datagrams;
2674 struct socket *sock;
2675 struct mmsghdr __user *entry;
2676 struct compat_mmsghdr __user *compat_entry;
2677 struct msghdr msg_sys;
2678 struct timespec64 end_time;
2679 struct timespec64 timeout64;
2680
2681 if (timeout &&
2682 poll_select_set_timeout(&end_time, timeout->tv_sec,
2683 timeout->tv_nsec))
2684 return -EINVAL;
2685
2686 datagrams = 0;
2687
2688 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2689 if (!sock)
2690 return err;
2691
2692 if (likely(!(flags & MSG_ERRQUEUE))) {
2693 err = sock_error(sock->sk);
2694 if (err) {
2695 datagrams = err;
2696 goto out_put;
2697 }
2698 }
2699
2700 entry = mmsg;
2701 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2702
2703 while (datagrams < vlen) {
2704 /*
2705 * No need to ask LSM for more than the first datagram.
2706 */
2707 if (MSG_CMSG_COMPAT & flags) {
2708 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2709 &msg_sys, flags & ~MSG_WAITFORONE,
2710 datagrams);
2711 if (err < 0)
2712 break;
2713 err = __put_user(err, &compat_entry->msg_len);
2714 ++compat_entry;
2715 } else {
2716 err = ___sys_recvmsg(sock,
2717 (struct user_msghdr __user *)entry,
2718 &msg_sys, flags & ~MSG_WAITFORONE,
2719 datagrams);
2720 if (err < 0)
2721 break;
2722 err = put_user(err, &entry->msg_len);
2723 ++entry;
2724 }
2725
2726 if (err)
2727 break;
2728 ++datagrams;
2729
2730 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2731 if (flags & MSG_WAITFORONE)
2732 flags |= MSG_DONTWAIT;
2733
2734 if (timeout) {
2735 ktime_get_ts64(&timeout64);
2736 *timeout = timespec64_sub(end_time, timeout64);
2737 if (timeout->tv_sec < 0) {
2738 timeout->tv_sec = timeout->tv_nsec = 0;
2739 break;
2740 }
2741
2742 /* Timeout, return less than vlen datagrams */
2743 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2744 break;
2745 }
2746
2747 /* Out of band data, return right away */
2748 if (msg_sys.msg_flags & MSG_OOB)
2749 break;
2750 cond_resched();
2751 }
2752
2753 if (err == 0)
2754 goto out_put;
2755
2756 if (datagrams == 0) {
2757 datagrams = err;
2758 goto out_put;
2759 }
2760
2761 /*
2762 * We may return less entries than requested (vlen) if the
2763 * sock is non block and there aren't enough datagrams...
2764 */
2765 if (err != -EAGAIN) {
2766 /*
2767 * ... or if recvmsg returns an error after we
2768 * received some datagrams, where we record the
2769 * error to return on the next call or if the
2770 * app asks about it using getsockopt(SO_ERROR).
2771 */
2772 sock->sk->sk_err = -err;
2773 }
2774out_put:
2775 fput_light(sock->file, fput_needed);
2776
2777 return datagrams;
2778}
2779
2780int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2781 unsigned int vlen, unsigned int flags,
2782 struct __kernel_timespec __user *timeout,
2783 struct old_timespec32 __user *timeout32)
2784{
2785 int datagrams;
2786 struct timespec64 timeout_sys;
2787
2788 if (timeout && get_timespec64(&timeout_sys, timeout))
2789 return -EFAULT;
2790
2791 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2792 return -EFAULT;
2793
2794 if (!timeout && !timeout32)
2795 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2796
2797 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2798
2799 if (datagrams <= 0)
2800 return datagrams;
2801
2802 if (timeout && put_timespec64(&timeout_sys, timeout))
2803 datagrams = -EFAULT;
2804
2805 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2806 datagrams = -EFAULT;
2807
2808 return datagrams;
2809}
2810
2811SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2812 unsigned int, vlen, unsigned int, flags,
2813 struct __kernel_timespec __user *, timeout)
2814{
2815 if (flags & MSG_CMSG_COMPAT)
2816 return -EINVAL;
2817
2818 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2819}
2820
2821#ifdef CONFIG_COMPAT_32BIT_TIME
2822SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2823 unsigned int, vlen, unsigned int, flags,
2824 struct old_timespec32 __user *, timeout)
2825{
2826 if (flags & MSG_CMSG_COMPAT)
2827 return -EINVAL;
2828
2829 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2830}
2831#endif
2832
2833#ifdef __ARCH_WANT_SYS_SOCKETCALL
2834/* Argument list sizes for sys_socketcall */
2835#define AL(x) ((x) * sizeof(unsigned long))
2836static const unsigned char nargs[21] = {
2837 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2838 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2839 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2840 AL(4), AL(5), AL(4)
2841};
2842
2843#undef AL
2844
2845/*
2846 * System call vectors.
2847 *
2848 * Argument checking cleaned up. Saved 20% in size.
2849 * This function doesn't need to set the kernel lock because
2850 * it is set by the callees.
2851 */
2852
2853SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2854{
2855 unsigned long a[AUDITSC_ARGS];
2856 unsigned long a0, a1;
2857 int err;
2858 unsigned int len;
2859
2860 if (call < 1 || call > SYS_SENDMMSG)
2861 return -EINVAL;
2862 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2863
2864 len = nargs[call];
2865 if (len > sizeof(a))
2866 return -EINVAL;
2867
2868 /* copy_from_user should be SMP safe. */
2869 if (copy_from_user(a, args, len))
2870 return -EFAULT;
2871
2872 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2873 if (err)
2874 return err;
2875
2876 a0 = a[0];
2877 a1 = a[1];
2878
2879 switch (call) {
2880 case SYS_SOCKET:
2881 err = __sys_socket(a0, a1, a[2]);
2882 break;
2883 case SYS_BIND:
2884 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2885 break;
2886 case SYS_CONNECT:
2887 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2888 break;
2889 case SYS_LISTEN:
2890 err = __sys_listen(a0, a1);
2891 break;
2892 case SYS_ACCEPT:
2893 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2894 (int __user *)a[2], 0);
2895 break;
2896 case SYS_GETSOCKNAME:
2897 err =
2898 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2899 (int __user *)a[2]);
2900 break;
2901 case SYS_GETPEERNAME:
2902 err =
2903 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2904 (int __user *)a[2]);
2905 break;
2906 case SYS_SOCKETPAIR:
2907 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2908 break;
2909 case SYS_SEND:
2910 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2911 NULL, 0);
2912 break;
2913 case SYS_SENDTO:
2914 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2915 (struct sockaddr __user *)a[4], a[5]);
2916 break;
2917 case SYS_RECV:
2918 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2919 NULL, NULL);
2920 break;
2921 case SYS_RECVFROM:
2922 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2923 (struct sockaddr __user *)a[4],
2924 (int __user *)a[5]);
2925 break;
2926 case SYS_SHUTDOWN:
2927 err = __sys_shutdown(a0, a1);
2928 break;
2929 case SYS_SETSOCKOPT:
2930 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2931 a[4]);
2932 break;
2933 case SYS_GETSOCKOPT:
2934 err =
2935 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2936 (int __user *)a[4]);
2937 break;
2938 case SYS_SENDMSG:
2939 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2940 a[2], true);
2941 break;
2942 case SYS_SENDMMSG:
2943 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2944 a[3], true);
2945 break;
2946 case SYS_RECVMSG:
2947 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2948 a[2], true);
2949 break;
2950 case SYS_RECVMMSG:
2951 if (IS_ENABLED(CONFIG_64BIT))
2952 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2953 a[2], a[3],
2954 (struct __kernel_timespec __user *)a[4],
2955 NULL);
2956 else
2957 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2958 a[2], a[3], NULL,
2959 (struct old_timespec32 __user *)a[4]);
2960 break;
2961 case SYS_ACCEPT4:
2962 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2963 (int __user *)a[2], a[3]);
2964 break;
2965 default:
2966 err = -EINVAL;
2967 break;
2968 }
2969 return err;
2970}
2971
2972#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2973
2974/**
2975 * sock_register - add a socket protocol handler
2976 * @ops: description of protocol
2977 *
2978 * This function is called by a protocol handler that wants to
2979 * advertise its address family, and have it linked into the
2980 * socket interface. The value ops->family corresponds to the
2981 * socket system call protocol family.
2982 */
2983int sock_register(const struct net_proto_family *ops)
2984{
2985 int err;
2986
2987 if (ops->family >= NPROTO) {
2988 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2989 return -ENOBUFS;
2990 }
2991
2992 spin_lock(&net_family_lock);
2993 if (rcu_dereference_protected(net_families[ops->family],
2994 lockdep_is_held(&net_family_lock)))
2995 err = -EEXIST;
2996 else {
2997 rcu_assign_pointer(net_families[ops->family], ops);
2998 err = 0;
2999 }
3000 spin_unlock(&net_family_lock);
3001
3002 pr_info("NET: Registered protocol family %d\n", ops->family);
3003 return err;
3004}
3005EXPORT_SYMBOL(sock_register);
3006
3007/**
3008 * sock_unregister - remove a protocol handler
3009 * @family: protocol family to remove
3010 *
3011 * This function is called by a protocol handler that wants to
3012 * remove its address family, and have it unlinked from the
3013 * new socket creation.
3014 *
3015 * If protocol handler is a module, then it can use module reference
3016 * counts to protect against new references. If protocol handler is not
3017 * a module then it needs to provide its own protection in
3018 * the ops->create routine.
3019 */
3020void sock_unregister(int family)
3021{
3022 BUG_ON(family < 0 || family >= NPROTO);
3023
3024 spin_lock(&net_family_lock);
3025 RCU_INIT_POINTER(net_families[family], NULL);
3026 spin_unlock(&net_family_lock);
3027
3028 synchronize_rcu();
3029
3030 pr_info("NET: Unregistered protocol family %d\n", family);
3031}
3032EXPORT_SYMBOL(sock_unregister);
3033
3034bool sock_is_registered(int family)
3035{
3036 return family < NPROTO && rcu_access_pointer(net_families[family]);
3037}
3038
3039static int __init sock_init(void)
3040{
3041 int err;
3042 /*
3043 * Initialize the network sysctl infrastructure.
3044 */
3045 err = net_sysctl_init();
3046 if (err)
3047 goto out;
3048
3049 /*
3050 * Initialize skbuff SLAB cache
3051 */
3052 skb_init();
3053
3054 /*
3055 * Initialize the protocols module.
3056 */
3057
3058 init_inodecache();
3059
3060 err = register_filesystem(&sock_fs_type);
3061 if (err)
3062 goto out;
3063 sock_mnt = kern_mount(&sock_fs_type);
3064 if (IS_ERR(sock_mnt)) {
3065 err = PTR_ERR(sock_mnt);
3066 goto out_mount;
3067 }
3068
3069 /* The real protocol initialization is performed in later initcalls.
3070 */
3071
3072#ifdef CONFIG_NETFILTER
3073 err = netfilter_init();
3074 if (err)
3075 goto out;
3076#endif
3077
3078 ptp_classifier_init();
3079
3080out:
3081 return err;
3082
3083out_mount:
3084 unregister_filesystem(&sock_fs_type);
3085 goto out;
3086}
3087
3088core_initcall(sock_init); /* early initcall */
3089
3090#ifdef CONFIG_PROC_FS
3091void socket_seq_show(struct seq_file *seq)
3092{
3093 seq_printf(seq, "sockets: used %d\n",
3094 sock_inuse_get(seq->private));
3095}
3096#endif /* CONFIG_PROC_FS */
3097
3098#ifdef CONFIG_COMPAT
3099static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
3100{
3101 struct compat_ifconf ifc32;
3102 struct ifconf ifc;
3103 int err;
3104
3105 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
3106 return -EFAULT;
3107
3108 ifc.ifc_len = ifc32.ifc_len;
3109 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
3110
3111 rtnl_lock();
3112 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
3113 rtnl_unlock();
3114 if (err)
3115 return err;
3116
3117 ifc32.ifc_len = ifc.ifc_len;
3118 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
3119 return -EFAULT;
3120
3121 return 0;
3122}
3123
3124static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
3125{
3126 struct compat_ethtool_rxnfc __user *compat_rxnfc;
3127 bool convert_in = false, convert_out = false;
3128 size_t buf_size = 0;
3129 struct ethtool_rxnfc __user *rxnfc = NULL;
3130 struct ifreq ifr;
3131 u32 rule_cnt = 0, actual_rule_cnt;
3132 u32 ethcmd;
3133 u32 data;
3134 int ret;
3135
3136 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
3137 return -EFAULT;
3138
3139 compat_rxnfc = compat_ptr(data);
3140
3141 if (get_user(ethcmd, &compat_rxnfc->cmd))
3142 return -EFAULT;
3143
3144 /* Most ethtool structures are defined without padding.
3145 * Unfortunately struct ethtool_rxnfc is an exception.
3146 */
3147 switch (ethcmd) {
3148 default:
3149 break;
3150 case ETHTOOL_GRXCLSRLALL:
3151 /* Buffer size is variable */
3152 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
3153 return -EFAULT;
3154 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
3155 return -ENOMEM;
3156 buf_size += rule_cnt * sizeof(u32);
3157 fallthrough;
3158 case ETHTOOL_GRXRINGS:
3159 case ETHTOOL_GRXCLSRLCNT:
3160 case ETHTOOL_GRXCLSRULE:
3161 case ETHTOOL_SRXCLSRLINS:
3162 convert_out = true;
3163 fallthrough;
3164 case ETHTOOL_SRXCLSRLDEL:
3165 buf_size += sizeof(struct ethtool_rxnfc);
3166 convert_in = true;
3167 rxnfc = compat_alloc_user_space(buf_size);
3168 break;
3169 }
3170
3171 if (copy_from_user(&ifr.ifr_name, &ifr32->ifr_name, IFNAMSIZ))
3172 return -EFAULT;
3173
3174 ifr.ifr_data = convert_in ? rxnfc : (void __user *)compat_rxnfc;
3175
3176 if (convert_in) {
3177 /* We expect there to be holes between fs.m_ext and
3178 * fs.ring_cookie and at the end of fs, but nowhere else.
3179 */
3180 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
3181 sizeof(compat_rxnfc->fs.m_ext) !=
3182 offsetof(struct ethtool_rxnfc, fs.m_ext) +
3183 sizeof(rxnfc->fs.m_ext));
3184 BUILD_BUG_ON(
3185 offsetof(struct compat_ethtool_rxnfc, fs.location) -
3186 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
3187 offsetof(struct ethtool_rxnfc, fs.location) -
3188 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
3189
3190 if (copy_in_user(rxnfc, compat_rxnfc,
3191 (void __user *)(&rxnfc->fs.m_ext + 1) -
3192 (void __user *)rxnfc) ||
3193 copy_in_user(&rxnfc->fs.ring_cookie,
3194 &compat_rxnfc->fs.ring_cookie,
3195 (void __user *)(&rxnfc->fs.location + 1) -
3196 (void __user *)&rxnfc->fs.ring_cookie))
3197 return -EFAULT;
3198 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3199 if (put_user(rule_cnt, &rxnfc->rule_cnt))
3200 return -EFAULT;
3201 } else if (copy_in_user(&rxnfc->rule_cnt,
3202 &compat_rxnfc->rule_cnt,
3203 sizeof(rxnfc->rule_cnt)))
3204 return -EFAULT;
3205 }
3206
3207 ret = dev_ioctl(net, SIOCETHTOOL, &ifr, NULL);
3208 if (ret)
3209 return ret;
3210
3211 if (convert_out) {
3212 if (copy_in_user(compat_rxnfc, rxnfc,
3213 (const void __user *)(&rxnfc->fs.m_ext + 1) -
3214 (const void __user *)rxnfc) ||
3215 copy_in_user(&compat_rxnfc->fs.ring_cookie,
3216 &rxnfc->fs.ring_cookie,
3217 (const void __user *)(&rxnfc->fs.location + 1) -
3218 (const void __user *)&rxnfc->fs.ring_cookie) ||
3219 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
3220 sizeof(rxnfc->rule_cnt)))
3221 return -EFAULT;
3222
3223 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3224 /* As an optimisation, we only copy the actual
3225 * number of rules that the underlying
3226 * function returned. Since Mallory might
3227 * change the rule count in user memory, we
3228 * check that it is less than the rule count
3229 * originally given (as the user buffer size),
3230 * which has been range-checked.
3231 */
3232 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
3233 return -EFAULT;
3234 if (actual_rule_cnt < rule_cnt)
3235 rule_cnt = actual_rule_cnt;
3236 if (copy_in_user(&compat_rxnfc->rule_locs[0],
3237 &rxnfc->rule_locs[0],
3238 rule_cnt * sizeof(u32)))
3239 return -EFAULT;
3240 }
3241 }
3242
3243 return 0;
3244}
3245
3246static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3247{
3248 compat_uptr_t uptr32;
3249 struct ifreq ifr;
3250 void __user *saved;
3251 int err;
3252
3253 if (copy_from_user(&ifr, uifr32, sizeof(struct compat_ifreq)))
3254 return -EFAULT;
3255
3256 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3257 return -EFAULT;
3258
3259 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3260 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3261
3262 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL);
3263 if (!err) {
3264 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3265 if (copy_to_user(uifr32, &ifr, sizeof(struct compat_ifreq)))
3266 err = -EFAULT;
3267 }
3268 return err;
3269}
3270
3271/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3272static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3273 struct compat_ifreq __user *u_ifreq32)
3274{
3275 struct ifreq ifreq;
3276 u32 data32;
3277
3278 if (copy_from_user(ifreq.ifr_name, u_ifreq32->ifr_name, IFNAMSIZ))
3279 return -EFAULT;
3280 if (get_user(data32, &u_ifreq32->ifr_data))
3281 return -EFAULT;
3282 ifreq.ifr_data = compat_ptr(data32);
3283
3284 return dev_ioctl(net, cmd, &ifreq, NULL);
3285}
3286
3287static int compat_ifreq_ioctl(struct net *net, struct socket *sock,
3288 unsigned int cmd,
3289 struct compat_ifreq __user *uifr32)
3290{
3291 struct ifreq __user *uifr;
3292 int err;
3293
3294 /* Handle the fact that while struct ifreq has the same *layout* on
3295 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3296 * which are handled elsewhere, it still has different *size* due to
3297 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3298 * resulting in struct ifreq being 32 and 40 bytes respectively).
3299 * As a result, if the struct happens to be at the end of a page and
3300 * the next page isn't readable/writable, we get a fault. To prevent
3301 * that, copy back and forth to the full size.
3302 */
3303
3304 uifr = compat_alloc_user_space(sizeof(*uifr));
3305 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
3306 return -EFAULT;
3307
3308 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
3309
3310 if (!err) {
3311 switch (cmd) {
3312 case SIOCGIFFLAGS:
3313 case SIOCGIFMETRIC:
3314 case SIOCGIFMTU:
3315 case SIOCGIFMEM:
3316 case SIOCGIFHWADDR:
3317 case SIOCGIFINDEX:
3318 case SIOCGIFADDR:
3319 case SIOCGIFBRDADDR:
3320 case SIOCGIFDSTADDR:
3321 case SIOCGIFNETMASK:
3322 case SIOCGIFPFLAGS:
3323 case SIOCGIFTXQLEN:
3324 case SIOCGMIIPHY:
3325 case SIOCGMIIREG:
3326 case SIOCGIFNAME:
3327 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
3328 err = -EFAULT;
3329 break;
3330 }
3331 }
3332 return err;
3333}
3334
3335static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
3336 struct compat_ifreq __user *uifr32)
3337{
3338 struct ifreq ifr;
3339 struct compat_ifmap __user *uifmap32;
3340 int err;
3341
3342 uifmap32 = &uifr32->ifr_ifru.ifru_map;
3343 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
3344 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3345 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3346 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3347 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
3348 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
3349 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
3350 if (err)
3351 return -EFAULT;
3352
3353 err = dev_ioctl(net, cmd, &ifr, NULL);
3354
3355 if (cmd == SIOCGIFMAP && !err) {
3356 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
3357 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3358 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3359 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3360 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
3361 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
3362 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
3363 if (err)
3364 err = -EFAULT;
3365 }
3366 return err;
3367}
3368
3369/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3370 * for some operations; this forces use of the newer bridge-utils that
3371 * use compatible ioctls
3372 */
3373static int old_bridge_ioctl(compat_ulong_t __user *argp)
3374{
3375 compat_ulong_t tmp;
3376
3377 if (get_user(tmp, argp))
3378 return -EFAULT;
3379 if (tmp == BRCTL_GET_VERSION)
3380 return BRCTL_VERSION + 1;
3381 return -EINVAL;
3382}
3383
3384static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3385 unsigned int cmd, unsigned long arg)
3386{
3387 void __user *argp = compat_ptr(arg);
3388 struct sock *sk = sock->sk;
3389 struct net *net = sock_net(sk);
3390
3391 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3392 return compat_ifr_data_ioctl(net, cmd, argp);
3393
3394 switch (cmd) {
3395 case SIOCSIFBR:
3396 case SIOCGIFBR:
3397 return old_bridge_ioctl(argp);
3398 case SIOCGIFCONF:
3399 return compat_dev_ifconf(net, argp);
3400 case SIOCETHTOOL:
3401 return ethtool_ioctl(net, argp);
3402 case SIOCWANDEV:
3403 return compat_siocwandev(net, argp);
3404 case SIOCGIFMAP:
3405 case SIOCSIFMAP:
3406 return compat_sioc_ifmap(net, cmd, argp);
3407 case SIOCGSTAMP_OLD:
3408 case SIOCGSTAMPNS_OLD:
3409 if (!sock->ops->gettstamp)
3410 return -ENOIOCTLCMD;
3411 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3412 !COMPAT_USE_64BIT_TIME);
3413
3414 case SIOCBONDSLAVEINFOQUERY:
3415 case SIOCBONDINFOQUERY:
3416 case SIOCSHWTSTAMP:
3417 case SIOCGHWTSTAMP:
3418 return compat_ifr_data_ioctl(net, cmd, argp);
3419
3420 case FIOSETOWN:
3421 case SIOCSPGRP:
3422 case FIOGETOWN:
3423 case SIOCGPGRP:
3424 case SIOCBRADDBR:
3425 case SIOCBRDELBR:
3426 case SIOCGIFVLAN:
3427 case SIOCSIFVLAN:
3428 case SIOCADDDLCI:
3429 case SIOCDELDLCI:
3430 case SIOCGSKNS:
3431 case SIOCGSTAMP_NEW:
3432 case SIOCGSTAMPNS_NEW:
3433 return sock_ioctl(file, cmd, arg);
3434
3435 case SIOCGIFFLAGS:
3436 case SIOCSIFFLAGS:
3437 case SIOCGIFMETRIC:
3438 case SIOCSIFMETRIC:
3439 case SIOCGIFMTU:
3440 case SIOCSIFMTU:
3441 case SIOCGIFMEM:
3442 case SIOCSIFMEM:
3443 case SIOCGIFHWADDR:
3444 case SIOCSIFHWADDR:
3445 case SIOCADDMULTI:
3446 case SIOCDELMULTI:
3447 case SIOCGIFINDEX:
3448 case SIOCGIFADDR:
3449 case SIOCSIFADDR:
3450 case SIOCSIFHWBROADCAST:
3451 case SIOCDIFADDR:
3452 case SIOCGIFBRDADDR:
3453 case SIOCSIFBRDADDR:
3454 case SIOCGIFDSTADDR:
3455 case SIOCSIFDSTADDR:
3456 case SIOCGIFNETMASK:
3457 case SIOCSIFNETMASK:
3458 case SIOCSIFPFLAGS:
3459 case SIOCGIFPFLAGS:
3460 case SIOCGIFTXQLEN:
3461 case SIOCSIFTXQLEN:
3462 case SIOCBRADDIF:
3463 case SIOCBRDELIF:
3464 case SIOCGIFNAME:
3465 case SIOCSIFNAME:
3466 case SIOCGMIIPHY:
3467 case SIOCGMIIREG:
3468 case SIOCSMIIREG:
3469 case SIOCBONDENSLAVE:
3470 case SIOCBONDRELEASE:
3471 case SIOCBONDSETHWADDR:
3472 case SIOCBONDCHANGEACTIVE:
3473 return compat_ifreq_ioctl(net, sock, cmd, argp);
3474
3475 case SIOCSARP:
3476 case SIOCGARP:
3477 case SIOCDARP:
3478 case SIOCOUTQ:
3479 case SIOCOUTQNSD:
3480 case SIOCATMARK:
3481 return sock_do_ioctl(net, sock, cmd, arg);
3482 }
3483
3484 return -ENOIOCTLCMD;
3485}
3486
3487static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3488 unsigned long arg)
3489{
3490 struct socket *sock = file->private_data;
3491 int ret = -ENOIOCTLCMD;
3492 struct sock *sk;
3493 struct net *net;
3494
3495 sk = sock->sk;
3496 net = sock_net(sk);
3497
3498 if (sock->ops->compat_ioctl)
3499 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3500
3501 if (ret == -ENOIOCTLCMD &&
3502 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3503 ret = compat_wext_handle_ioctl(net, cmd, arg);
3504
3505 if (ret == -ENOIOCTLCMD)
3506 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3507
3508 return ret;
3509}
3510#endif
3511
3512/**
3513 * kernel_bind - bind an address to a socket (kernel space)
3514 * @sock: socket
3515 * @addr: address
3516 * @addrlen: length of address
3517 *
3518 * Returns 0 or an error.
3519 */
3520
3521int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3522{
3523 return sock->ops->bind(sock, addr, addrlen);
3524}
3525EXPORT_SYMBOL(kernel_bind);
3526
3527/**
3528 * kernel_listen - move socket to listening state (kernel space)
3529 * @sock: socket
3530 * @backlog: pending connections queue size
3531 *
3532 * Returns 0 or an error.
3533 */
3534
3535int kernel_listen(struct socket *sock, int backlog)
3536{
3537 return sock->ops->listen(sock, backlog);
3538}
3539EXPORT_SYMBOL(kernel_listen);
3540
3541/**
3542 * kernel_accept - accept a connection (kernel space)
3543 * @sock: listening socket
3544 * @newsock: new connected socket
3545 * @flags: flags
3546 *
3547 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3548 * If it fails, @newsock is guaranteed to be %NULL.
3549 * Returns 0 or an error.
3550 */
3551
3552int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3553{
3554 struct sock *sk = sock->sk;
3555 int err;
3556
3557 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3558 newsock);
3559 if (err < 0)
3560 goto done;
3561
3562 err = sock->ops->accept(sock, *newsock, flags, true);
3563 if (err < 0) {
3564 sock_release(*newsock);
3565 *newsock = NULL;
3566 goto done;
3567 }
3568
3569 (*newsock)->ops = sock->ops;
3570 __module_get((*newsock)->ops->owner);
3571
3572done:
3573 return err;
3574}
3575EXPORT_SYMBOL(kernel_accept);
3576
3577/**
3578 * kernel_connect - connect a socket (kernel space)
3579 * @sock: socket
3580 * @addr: address
3581 * @addrlen: address length
3582 * @flags: flags (O_NONBLOCK, ...)
3583 *
3584 * For datagram sockets, @addr is the addres to which datagrams are sent
3585 * by default, and the only address from which datagrams are received.
3586 * For stream sockets, attempts to connect to @addr.
3587 * Returns 0 or an error code.
3588 */
3589
3590int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3591 int flags)
3592{
3593 return sock->ops->connect(sock, addr, addrlen, flags);
3594}
3595EXPORT_SYMBOL(kernel_connect);
3596
3597/**
3598 * kernel_getsockname - get the address which the socket is bound (kernel space)
3599 * @sock: socket
3600 * @addr: address holder
3601 *
3602 * Fills the @addr pointer with the address which the socket is bound.
3603 * Returns 0 or an error code.
3604 */
3605
3606int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3607{
3608 return sock->ops->getname(sock, addr, 0);
3609}
3610EXPORT_SYMBOL(kernel_getsockname);
3611
3612/**
3613 * kernel_getpeername - get the address which the socket is connected (kernel space)
3614 * @sock: socket
3615 * @addr: address holder
3616 *
3617 * Fills the @addr pointer with the address which the socket is connected.
3618 * Returns 0 or an error code.
3619 */
3620
3621int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3622{
3623 return sock->ops->getname(sock, addr, 1);
3624}
3625EXPORT_SYMBOL(kernel_getpeername);
3626
3627/**
3628 * kernel_sendpage - send a &page through a socket (kernel space)
3629 * @sock: socket
3630 * @page: page
3631 * @offset: page offset
3632 * @size: total size in bytes
3633 * @flags: flags (MSG_DONTWAIT, ...)
3634 *
3635 * Returns the total amount sent in bytes or an error.
3636 */
3637
3638int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3639 size_t size, int flags)
3640{
3641 if (sock->ops->sendpage) {
3642 /* Warn in case the improper page to zero-copy send */
3643 WARN_ONCE(!sendpage_ok(page), "improper page for zero-copy send");
3644 return sock->ops->sendpage(sock, page, offset, size, flags);
3645 }
3646 return sock_no_sendpage(sock, page, offset, size, flags);
3647}
3648EXPORT_SYMBOL(kernel_sendpage);
3649
3650/**
3651 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3652 * @sk: sock
3653 * @page: page
3654 * @offset: page offset
3655 * @size: total size in bytes
3656 * @flags: flags (MSG_DONTWAIT, ...)
3657 *
3658 * Returns the total amount sent in bytes or an error.
3659 * Caller must hold @sk.
3660 */
3661
3662int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3663 size_t size, int flags)
3664{
3665 struct socket *sock = sk->sk_socket;
3666
3667 if (sock->ops->sendpage_locked)
3668 return sock->ops->sendpage_locked(sk, page, offset, size,
3669 flags);
3670
3671 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3672}
3673EXPORT_SYMBOL(kernel_sendpage_locked);
3674
3675/**
3676 * kernel_sock_shutdown - shut down part of a full-duplex connection (kernel space)
3677 * @sock: socket
3678 * @how: connection part
3679 *
3680 * Returns 0 or an error.
3681 */
3682
3683int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3684{
3685 return sock->ops->shutdown(sock, how);
3686}
3687EXPORT_SYMBOL(kernel_sock_shutdown);
3688
3689/**
3690 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3691 * @sk: socket
3692 *
3693 * This routine returns the IP overhead imposed by a socket i.e.
3694 * the length of the underlying IP header, depending on whether
3695 * this is an IPv4 or IPv6 socket and the length from IP options turned
3696 * on at the socket. Assumes that the caller has a lock on the socket.
3697 */
3698
3699u32 kernel_sock_ip_overhead(struct sock *sk)
3700{
3701 struct inet_sock *inet;
3702 struct ip_options_rcu *opt;
3703 u32 overhead = 0;
3704#if IS_ENABLED(CONFIG_IPV6)
3705 struct ipv6_pinfo *np;
3706 struct ipv6_txoptions *optv6 = NULL;
3707#endif /* IS_ENABLED(CONFIG_IPV6) */
3708
3709 if (!sk)
3710 return overhead;
3711
3712 switch (sk->sk_family) {
3713 case AF_INET:
3714 inet = inet_sk(sk);
3715 overhead += sizeof(struct iphdr);
3716 opt = rcu_dereference_protected(inet->inet_opt,
3717 sock_owned_by_user(sk));
3718 if (opt)
3719 overhead += opt->opt.optlen;
3720 return overhead;
3721#if IS_ENABLED(CONFIG_IPV6)
3722 case AF_INET6:
3723 np = inet6_sk(sk);
3724 overhead += sizeof(struct ipv6hdr);
3725 if (np)
3726 optv6 = rcu_dereference_protected(np->opt,
3727 sock_owned_by_user(sk));
3728 if (optv6)
3729 overhead += (optv6->opt_flen + optv6->opt_nflen);
3730 return overhead;
3731#endif /* IS_ENABLED(CONFIG_IPV6) */
3732 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3733 return overhead;
3734 }
3735}
3736EXPORT_SYMBOL(kernel_sock_ip_overhead);