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