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