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