1/*
2 * NET An implementation of the SOCKET network access protocol.
3 *
4 * Version: @(#)socket.c 1.1.93 18/02/95
5 *
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
7 * Ross Biro
8 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
9 *
10 * Fixes:
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
12 * shutdown()
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
17 * top level.
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
22 * tty drivers).
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
25 * configurable.
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
34 * stuff.
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
40 * moment.
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
45 * Tigran Aivazian : Made listen(2) backlog sanity checks
46 * protocol-independent
47 *
48 *
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
53 *
54 *
55 * This module is effectively the top level interface to the BSD socket
56 * paradigm.
57 *
58 * Based upon Swansea University Computer Society NET3.039
59 */
60
61#include <linux/mm.h>
62#include <linux/socket.h>
63#include <linux/file.h>
64#include <linux/net.h>
65#include <linux/interrupt.h>
66#include <linux/thread_info.h>
67#include <linux/rcupdate.h>
68#include <linux/netdevice.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
71#include <linux/mutex.h>
72#include <linux/if_bridge.h>
73#include <linux/if_frad.h>
74#include <linux/if_vlan.h>
75#include <linux/ptp_classify.h>
76#include <linux/init.h>
77#include <linux/poll.h>
78#include <linux/cache.h>
79#include <linux/module.h>
80#include <linux/highmem.h>
81#include <linux/mount.h>
82#include <linux/security.h>
83#include <linux/syscalls.h>
84#include <linux/compat.h>
85#include <linux/kmod.h>
86#include <linux/audit.h>
87#include <linux/wireless.h>
88#include <linux/nsproxy.h>
89#include <linux/magic.h>
90#include <linux/slab.h>
91#include <linux/xattr.h>
92#include <linux/nospec.h>
93
94#include <linux/uaccess.h>
95#include <asm/unistd.h>
96
97#include <net/compat.h>
98#include <net/wext.h>
99#include <net/cls_cgroup.h>
100
101#include <net/sock.h>
102#include <linux/netfilter.h>
103
104#include <linux/if_tun.h>
105#include <linux/ipv6_route.h>
106#include <linux/route.h>
107#include <linux/sockios.h>
108#include <net/busy_poll.h>
109#include <linux/errqueue.h>
110
111#ifdef CONFIG_NET_RX_BUSY_POLL
112unsigned int sysctl_net_busy_read __read_mostly;
113unsigned int sysctl_net_busy_poll __read_mostly;
114#endif
115
116static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
117static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
118static int sock_mmap(struct file *file, struct vm_area_struct *vma);
119
120static int sock_close(struct inode *inode, struct file *file);
121static __poll_t sock_poll(struct file *file,
122 struct poll_table_struct *wait);
123static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
124#ifdef CONFIG_COMPAT
125static long compat_sock_ioctl(struct file *file,
126 unsigned int cmd, unsigned long arg);
127#endif
128static int sock_fasync(int fd, struct file *filp, int on);
129static ssize_t sock_sendpage(struct file *file, struct page *page,
130 int offset, size_t size, loff_t *ppos, int more);
131static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
132 struct pipe_inode_info *pipe, size_t len,
133 unsigned int flags);
134
135/*
136 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
137 * in the operation structures but are done directly via the socketcall() multiplexor.
138 */
139
140static const struct file_operations socket_file_ops = {
141 .owner = THIS_MODULE,
142 .llseek = no_llseek,
143 .read_iter = sock_read_iter,
144 .write_iter = sock_write_iter,
145 .poll = sock_poll,
146 .unlocked_ioctl = sock_ioctl,
147#ifdef CONFIG_COMPAT
148 .compat_ioctl = compat_sock_ioctl,
149#endif
150 .mmap = sock_mmap,
151 .release = sock_close,
152 .fasync = sock_fasync,
153 .sendpage = sock_sendpage,
154 .splice_write = generic_splice_sendpage,
155 .splice_read = sock_splice_read,
156};
157
158/*
159 * The protocol list. Each protocol is registered in here.
160 */
161
162static DEFINE_SPINLOCK(net_family_lock);
163static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
164
165/*
166 * Support routines.
167 * Move socket addresses back and forth across the kernel/user
168 * divide and look after the messy bits.
169 */
170
171/**
172 * move_addr_to_kernel - copy a socket address into kernel space
173 * @uaddr: Address in user space
174 * @kaddr: Address in kernel space
175 * @ulen: Length in user space
176 *
177 * The address is copied into kernel space. If the provided address is
178 * too long an error code of -EINVAL is returned. If the copy gives
179 * invalid addresses -EFAULT is returned. On a success 0 is returned.
180 */
181
182int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
183{
184 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
185 return -EINVAL;
186 if (ulen == 0)
187 return 0;
188 if (copy_from_user(kaddr, uaddr, ulen))
189 return -EFAULT;
190 return audit_sockaddr(ulen, kaddr);
191}
192
193/**
194 * move_addr_to_user - copy an address to user space
195 * @kaddr: kernel space address
196 * @klen: length of address in kernel
197 * @uaddr: user space address
198 * @ulen: pointer to user length field
199 *
200 * The value pointed to by ulen on entry is the buffer length available.
201 * This is overwritten with the buffer space used. -EINVAL is returned
202 * if an overlong buffer is specified or a negative buffer size. -EFAULT
203 * is returned if either the buffer or the length field are not
204 * accessible.
205 * After copying the data up to the limit the user specifies, the true
206 * length of the data is written over the length limit the user
207 * specified. Zero is returned for a success.
208 */
209
210static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
211 void __user *uaddr, int __user *ulen)
212{
213 int err;
214 int len;
215
216 BUG_ON(klen > sizeof(struct sockaddr_storage));
217 err = get_user(len, ulen);
218 if (err)
219 return err;
220 if (len > klen)
221 len = klen;
222 if (len < 0)
223 return -EINVAL;
224 if (len) {
225 if (audit_sockaddr(klen, kaddr))
226 return -ENOMEM;
227 if (copy_to_user(uaddr, kaddr, len))
228 return -EFAULT;
229 }
230 /*
231 * "fromlen shall refer to the value before truncation.."
232 * 1003.1g
233 */
234 return __put_user(klen, ulen);
235}
236
237static struct kmem_cache *sock_inode_cachep __ro_after_init;
238
239static struct inode *sock_alloc_inode(struct super_block *sb)
240{
241 struct socket_alloc *ei;
242 struct socket_wq *wq;
243
244 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
245 if (!ei)
246 return NULL;
247 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
248 if (!wq) {
249 kmem_cache_free(sock_inode_cachep, ei);
250 return NULL;
251 }
252 init_waitqueue_head(&wq->wait);
253 wq->fasync_list = NULL;
254 wq->flags = 0;
255 ei->socket.wq = wq;
256
257 ei->socket.state = SS_UNCONNECTED;
258 ei->socket.flags = 0;
259 ei->socket.ops = NULL;
260 ei->socket.sk = NULL;
261 ei->socket.file = NULL;
262
263 return &ei->vfs_inode;
264}
265
266static void sock_destroy_inode(struct inode *inode)
267{
268 struct socket_alloc *ei;
269
270 ei = container_of(inode, struct socket_alloc, vfs_inode);
271 kfree_rcu(ei->socket.wq, rcu);
272 kmem_cache_free(sock_inode_cachep, ei);
273}
274
275static void init_once(void *foo)
276{
277 struct socket_alloc *ei = (struct socket_alloc *)foo;
278
279 inode_init_once(&ei->vfs_inode);
280}
281
282static void init_inodecache(void)
283{
284 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
285 sizeof(struct socket_alloc),
286 0,
287 (SLAB_HWCACHE_ALIGN |
288 SLAB_RECLAIM_ACCOUNT |
289 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
290 init_once);
291 BUG_ON(sock_inode_cachep == NULL);
292}
293
294static const struct super_operations sockfs_ops = {
295 .alloc_inode = sock_alloc_inode,
296 .destroy_inode = sock_destroy_inode,
297 .statfs = simple_statfs,
298};
299
300/*
301 * sockfs_dname() is called from d_path().
302 */
303static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
304{
305 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
306 d_inode(dentry)->i_ino);
307}
308
309static const struct dentry_operations sockfs_dentry_operations = {
310 .d_dname = sockfs_dname,
311};
312
313static int sockfs_xattr_get(const struct xattr_handler *handler,
314 struct dentry *dentry, struct inode *inode,
315 const char *suffix, void *value, size_t size)
316{
317 if (value) {
318 if (dentry->d_name.len + 1 > size)
319 return -ERANGE;
320 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
321 }
322 return dentry->d_name.len + 1;
323}
324
325#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
326#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
327#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
328
329static const struct xattr_handler sockfs_xattr_handler = {
330 .name = XATTR_NAME_SOCKPROTONAME,
331 .get = sockfs_xattr_get,
332};
333
334static int sockfs_security_xattr_set(const struct xattr_handler *handler,
335 struct dentry *dentry, struct inode *inode,
336 const char *suffix, const void *value,
337 size_t size, int flags)
338{
339 /* Handled by LSM. */
340 return -EAGAIN;
341}
342
343static const struct xattr_handler sockfs_security_xattr_handler = {
344 .prefix = XATTR_SECURITY_PREFIX,
345 .set = sockfs_security_xattr_set,
346};
347
348static const struct xattr_handler *sockfs_xattr_handlers[] = {
349 &sockfs_xattr_handler,
350 &sockfs_security_xattr_handler,
351 NULL
352};
353
354static struct dentry *sockfs_mount(struct file_system_type *fs_type,
355 int flags, const char *dev_name, void *data)
356{
357 return mount_pseudo_xattr(fs_type, "socket:", &sockfs_ops,
358 sockfs_xattr_handlers,
359 &sockfs_dentry_operations, SOCKFS_MAGIC);
360}
361
362static struct vfsmount *sock_mnt __read_mostly;
363
364static struct file_system_type sock_fs_type = {
365 .name = "sockfs",
366 .mount = sockfs_mount,
367 .kill_sb = kill_anon_super,
368};
369
370/*
371 * Obtains the first available file descriptor and sets it up for use.
372 *
373 * These functions create file structures and maps them to fd space
374 * of the current process. On success it returns file descriptor
375 * and file struct implicitly stored in sock->file.
376 * Note that another thread may close file descriptor before we return
377 * from this function. We use the fact that now we do not refer
378 * to socket after mapping. If one day we will need it, this
379 * function will increment ref. count on file by 1.
380 *
381 * In any case returned fd MAY BE not valid!
382 * This race condition is unavoidable
383 * with shared fd spaces, we cannot solve it inside kernel,
384 * but we take care of internal coherence yet.
385 */
386
387struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
388{
389 struct file *file;
390
391 if (!dname)
392 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
393
394 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
395 O_RDWR | (flags & O_NONBLOCK),
396 &socket_file_ops);
397 if (IS_ERR(file)) {
398 sock_release(sock);
399 return file;
400 }
401
402 sock->file = file;
403 file->private_data = sock;
404 return file;
405}
406EXPORT_SYMBOL(sock_alloc_file);
407
408static int sock_map_fd(struct socket *sock, int flags)
409{
410 struct file *newfile;
411 int fd = get_unused_fd_flags(flags);
412 if (unlikely(fd < 0)) {
413 sock_release(sock);
414 return fd;
415 }
416
417 newfile = sock_alloc_file(sock, flags, NULL);
418 if (likely(!IS_ERR(newfile))) {
419 fd_install(fd, newfile);
420 return fd;
421 }
422
423 put_unused_fd(fd);
424 return PTR_ERR(newfile);
425}
426
427struct socket *sock_from_file(struct file *file, int *err)
428{
429 if (file->f_op == &socket_file_ops)
430 return file->private_data; /* set in sock_map_fd */
431
432 *err = -ENOTSOCK;
433 return NULL;
434}
435EXPORT_SYMBOL(sock_from_file);
436
437/**
438 * sockfd_lookup - Go from a file number to its socket slot
439 * @fd: file handle
440 * @err: pointer to an error code return
441 *
442 * The file handle passed in is locked and the socket it is bound
443 * to is returned. If an error occurs the err pointer is overwritten
444 * with a negative errno code and NULL is returned. The function checks
445 * for both invalid handles and passing a handle which is not a socket.
446 *
447 * On a success the socket object pointer is returned.
448 */
449
450struct socket *sockfd_lookup(int fd, int *err)
451{
452 struct file *file;
453 struct socket *sock;
454
455 file = fget(fd);
456 if (!file) {
457 *err = -EBADF;
458 return NULL;
459 }
460
461 sock = sock_from_file(file, err);
462 if (!sock)
463 fput(file);
464 return sock;
465}
466EXPORT_SYMBOL(sockfd_lookup);
467
468static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
469{
470 struct fd f = fdget(fd);
471 struct socket *sock;
472
473 *err = -EBADF;
474 if (f.file) {
475 sock = sock_from_file(f.file, err);
476 if (likely(sock)) {
477 *fput_needed = f.flags;
478 return sock;
479 }
480 fdput(f);
481 }
482 return NULL;
483}
484
485static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
486 size_t size)
487{
488 ssize_t len;
489 ssize_t used = 0;
490
491 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
492 if (len < 0)
493 return len;
494 used += len;
495 if (buffer) {
496 if (size < used)
497 return -ERANGE;
498 buffer += len;
499 }
500
501 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
502 used += len;
503 if (buffer) {
504 if (size < used)
505 return -ERANGE;
506 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
507 buffer += len;
508 }
509
510 return used;
511}
512
513static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
514{
515 int err = simple_setattr(dentry, iattr);
516
517 if (!err && (iattr->ia_valid & ATTR_UID)) {
518 struct socket *sock = SOCKET_I(d_inode(dentry));
519
520 if (sock->sk)
521 sock->sk->sk_uid = iattr->ia_uid;
522 else
523 err = -ENOENT;
524 }
525
526 return err;
527}
528
529static const struct inode_operations sockfs_inode_ops = {
530 .listxattr = sockfs_listxattr,
531 .setattr = sockfs_setattr,
532};
533
534/**
535 * sock_alloc - allocate a socket
536 *
537 * Allocate a new inode and socket object. The two are bound together
538 * and initialised. The socket is then returned. If we are out of inodes
539 * NULL is returned.
540 */
541
542struct socket *sock_alloc(void)
543{
544 struct inode *inode;
545 struct socket *sock;
546
547 inode = new_inode_pseudo(sock_mnt->mnt_sb);
548 if (!inode)
549 return NULL;
550
551 sock = SOCKET_I(inode);
552
553 inode->i_ino = get_next_ino();
554 inode->i_mode = S_IFSOCK | S_IRWXUGO;
555 inode->i_uid = current_fsuid();
556 inode->i_gid = current_fsgid();
557 inode->i_op = &sockfs_inode_ops;
558
559 return sock;
560}
561EXPORT_SYMBOL(sock_alloc);
562
563/**
564 * sock_release - close a socket
565 * @sock: socket to close
566 *
567 * The socket is released from the protocol stack if it has a release
568 * callback, and the inode is then released if the socket is bound to
569 * an inode not a file.
570 */
571
572static void __sock_release(struct socket *sock, struct inode *inode)
573{
574 if (sock->ops) {
575 struct module *owner = sock->ops->owner;
576
577 if (inode)
578 inode_lock(inode);
579 sock->ops->release(sock);
580 sock->sk = NULL;
581 if (inode)
582 inode_unlock(inode);
583 sock->ops = NULL;
584 module_put(owner);
585 }
586
587 if (sock->wq->fasync_list)
588 pr_err("%s: fasync list not empty!\n", __func__);
589
590 if (!sock->file) {
591 iput(SOCK_INODE(sock));
592 return;
593 }
594 sock->file = NULL;
595}
596
597void sock_release(struct socket *sock)
598{
599 __sock_release(sock, NULL);
600}
601EXPORT_SYMBOL(sock_release);
602
603void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
604{
605 u8 flags = *tx_flags;
606
607 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
608 flags |= SKBTX_HW_TSTAMP;
609
610 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
611 flags |= SKBTX_SW_TSTAMP;
612
613 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
614 flags |= SKBTX_SCHED_TSTAMP;
615
616 *tx_flags = flags;
617}
618EXPORT_SYMBOL(__sock_tx_timestamp);
619
620static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
621{
622 int ret = sock->ops->sendmsg(sock, msg, msg_data_left(msg));
623 BUG_ON(ret == -EIOCBQUEUED);
624 return ret;
625}
626
627int sock_sendmsg(struct socket *sock, struct msghdr *msg)
628{
629 int err = security_socket_sendmsg(sock, msg,
630 msg_data_left(msg));
631
632 return err ?: sock_sendmsg_nosec(sock, msg);
633}
634EXPORT_SYMBOL(sock_sendmsg);
635
636int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
637 struct kvec *vec, size_t num, size_t size)
638{
639 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
640 return sock_sendmsg(sock, msg);
641}
642EXPORT_SYMBOL(kernel_sendmsg);
643
644int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
645 struct kvec *vec, size_t num, size_t size)
646{
647 struct socket *sock = sk->sk_socket;
648
649 if (!sock->ops->sendmsg_locked)
650 return sock_no_sendmsg_locked(sk, msg, size);
651
652 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
653
654 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
655}
656EXPORT_SYMBOL(kernel_sendmsg_locked);
657
658static bool skb_is_err_queue(const struct sk_buff *skb)
659{
660 /* pkt_type of skbs enqueued on the error queue are set to
661 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
662 * in recvmsg, since skbs received on a local socket will never
663 * have a pkt_type of PACKET_OUTGOING.
664 */
665 return skb->pkt_type == PACKET_OUTGOING;
666}
667
668/* On transmit, software and hardware timestamps are returned independently.
669 * As the two skb clones share the hardware timestamp, which may be updated
670 * before the software timestamp is received, a hardware TX timestamp may be
671 * returned only if there is no software TX timestamp. Ignore false software
672 * timestamps, which may be made in the __sock_recv_timestamp() call when the
673 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
674 * hardware timestamp.
675 */
676static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
677{
678 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
679}
680
681static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
682{
683 struct scm_ts_pktinfo ts_pktinfo;
684 struct net_device *orig_dev;
685
686 if (!skb_mac_header_was_set(skb))
687 return;
688
689 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
690
691 rcu_read_lock();
692 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
693 if (orig_dev)
694 ts_pktinfo.if_index = orig_dev->ifindex;
695 rcu_read_unlock();
696
697 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
698 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
699 sizeof(ts_pktinfo), &ts_pktinfo);
700}
701
702/*
703 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
704 */
705void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
706 struct sk_buff *skb)
707{
708 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
709 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
710 struct scm_timestamping_internal tss;
711
712 int empty = 1, false_tstamp = 0;
713 struct skb_shared_hwtstamps *shhwtstamps =
714 skb_hwtstamps(skb);
715
716 /* Race occurred between timestamp enabling and packet
717 receiving. Fill in the current time for now. */
718 if (need_software_tstamp && skb->tstamp == 0) {
719 __net_timestamp(skb);
720 false_tstamp = 1;
721 }
722
723 if (need_software_tstamp) {
724 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
725 if (new_tstamp) {
726 struct __kernel_sock_timeval tv;
727
728 skb_get_new_timestamp(skb, &tv);
729 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
730 sizeof(tv), &tv);
731 } else {
732 struct __kernel_old_timeval tv;
733
734 skb_get_timestamp(skb, &tv);
735 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
736 sizeof(tv), &tv);
737 }
738 } else {
739 if (new_tstamp) {
740 struct __kernel_timespec ts;
741
742 skb_get_new_timestampns(skb, &ts);
743 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
744 sizeof(ts), &ts);
745 } else {
746 struct timespec ts;
747
748 skb_get_timestampns(skb, &ts);
749 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
750 sizeof(ts), &ts);
751 }
752 }
753 }
754
755 memset(&tss, 0, sizeof(tss));
756 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
757 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
758 empty = 0;
759 if (shhwtstamps &&
760 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
761 !skb_is_swtx_tstamp(skb, false_tstamp) &&
762 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
763 empty = 0;
764 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
765 !skb_is_err_queue(skb))
766 put_ts_pktinfo(msg, skb);
767 }
768 if (!empty) {
769 if (sock_flag(sk, SOCK_TSTAMP_NEW))
770 put_cmsg_scm_timestamping64(msg, &tss);
771 else
772 put_cmsg_scm_timestamping(msg, &tss);
773
774 if (skb_is_err_queue(skb) && skb->len &&
775 SKB_EXT_ERR(skb)->opt_stats)
776 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
777 skb->len, skb->data);
778 }
779}
780EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
781
782void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
783 struct sk_buff *skb)
784{
785 int ack;
786
787 if (!sock_flag(sk, SOCK_WIFI_STATUS))
788 return;
789 if (!skb->wifi_acked_valid)
790 return;
791
792 ack = skb->wifi_acked;
793
794 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
795}
796EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
797
798static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
799 struct sk_buff *skb)
800{
801 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
802 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
803 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
804}
805
806void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
807 struct sk_buff *skb)
808{
809 sock_recv_timestamp(msg, sk, skb);
810 sock_recv_drops(msg, sk, skb);
811}
812EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
813
814static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
815 int flags)
816{
817 return sock->ops->recvmsg(sock, msg, msg_data_left(msg), flags);
818}
819
820int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
821{
822 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
823
824 return err ?: sock_recvmsg_nosec(sock, msg, flags);
825}
826EXPORT_SYMBOL(sock_recvmsg);
827
828/**
829 * kernel_recvmsg - Receive a message from a socket (kernel space)
830 * @sock: The socket to receive the message from
831 * @msg: Received message
832 * @vec: Input s/g array for message data
833 * @num: Size of input s/g array
834 * @size: Number of bytes to read
835 * @flags: Message flags (MSG_DONTWAIT, etc...)
836 *
837 * On return the msg structure contains the scatter/gather array passed in the
838 * vec argument. The array is modified so that it consists of the unfilled
839 * portion of the original array.
840 *
841 * The returned value is the total number of bytes received, or an error.
842 */
843int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
844 struct kvec *vec, size_t num, size_t size, int flags)
845{
846 mm_segment_t oldfs = get_fs();
847 int result;
848
849 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
850 set_fs(KERNEL_DS);
851 result = sock_recvmsg(sock, msg, flags);
852 set_fs(oldfs);
853 return result;
854}
855EXPORT_SYMBOL(kernel_recvmsg);
856
857static ssize_t sock_sendpage(struct file *file, struct page *page,
858 int offset, size_t size, loff_t *ppos, int more)
859{
860 struct socket *sock;
861 int flags;
862
863 sock = file->private_data;
864
865 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
866 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
867 flags |= more;
868
869 return kernel_sendpage(sock, page, offset, size, flags);
870}
871
872static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
873 struct pipe_inode_info *pipe, size_t len,
874 unsigned int flags)
875{
876 struct socket *sock = file->private_data;
877
878 if (unlikely(!sock->ops->splice_read))
879 return generic_file_splice_read(file, ppos, pipe, len, flags);
880
881 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
882}
883
884static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
885{
886 struct file *file = iocb->ki_filp;
887 struct socket *sock = file->private_data;
888 struct msghdr msg = {.msg_iter = *to,
889 .msg_iocb = iocb};
890 ssize_t res;
891
892 if (file->f_flags & O_NONBLOCK)
893 msg.msg_flags = MSG_DONTWAIT;
894
895 if (iocb->ki_pos != 0)
896 return -ESPIPE;
897
898 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
899 return 0;
900
901 res = sock_recvmsg(sock, &msg, msg.msg_flags);
902 *to = msg.msg_iter;
903 return res;
904}
905
906static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
907{
908 struct file *file = iocb->ki_filp;
909 struct socket *sock = file->private_data;
910 struct msghdr msg = {.msg_iter = *from,
911 .msg_iocb = iocb};
912 ssize_t res;
913
914 if (iocb->ki_pos != 0)
915 return -ESPIPE;
916
917 if (file->f_flags & O_NONBLOCK)
918 msg.msg_flags = MSG_DONTWAIT;
919
920 if (sock->type == SOCK_SEQPACKET)
921 msg.msg_flags |= MSG_EOR;
922
923 res = sock_sendmsg(sock, &msg);
924 *from = msg.msg_iter;
925 return res;
926}
927
928/*
929 * Atomic setting of ioctl hooks to avoid race
930 * with module unload.
931 */
932
933static DEFINE_MUTEX(br_ioctl_mutex);
934static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
935
936void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
937{
938 mutex_lock(&br_ioctl_mutex);
939 br_ioctl_hook = hook;
940 mutex_unlock(&br_ioctl_mutex);
941}
942EXPORT_SYMBOL(brioctl_set);
943
944static DEFINE_MUTEX(vlan_ioctl_mutex);
945static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
946
947void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
948{
949 mutex_lock(&vlan_ioctl_mutex);
950 vlan_ioctl_hook = hook;
951 mutex_unlock(&vlan_ioctl_mutex);
952}
953EXPORT_SYMBOL(vlan_ioctl_set);
954
955static DEFINE_MUTEX(dlci_ioctl_mutex);
956static int (*dlci_ioctl_hook) (unsigned int, void __user *);
957
958void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
959{
960 mutex_lock(&dlci_ioctl_mutex);
961 dlci_ioctl_hook = hook;
962 mutex_unlock(&dlci_ioctl_mutex);
963}
964EXPORT_SYMBOL(dlci_ioctl_set);
965
966static long sock_do_ioctl(struct net *net, struct socket *sock,
967 unsigned int cmd, unsigned long arg)
968{
969 int err;
970 void __user *argp = (void __user *)arg;
971
972 err = sock->ops->ioctl(sock, cmd, arg);
973
974 /*
975 * If this ioctl is unknown try to hand it down
976 * to the NIC driver.
977 */
978 if (err != -ENOIOCTLCMD)
979 return err;
980
981 if (cmd == SIOCGIFCONF) {
982 struct ifconf ifc;
983 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
984 return -EFAULT;
985 rtnl_lock();
986 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
987 rtnl_unlock();
988 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
989 err = -EFAULT;
990 } else {
991 struct ifreq ifr;
992 bool need_copyout;
993 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
994 return -EFAULT;
995 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
996 if (!err && need_copyout)
997 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
998 return -EFAULT;
999 }
1000 return err;
1001}
1002
1003/*
1004 * With an ioctl, arg may well be a user mode pointer, but we don't know
1005 * what to do with it - that's up to the protocol still.
1006 */
1007
1008struct ns_common *get_net_ns(struct ns_common *ns)
1009{
1010 return &get_net(container_of(ns, struct net, ns))->ns;
1011}
1012EXPORT_SYMBOL_GPL(get_net_ns);
1013
1014static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1015{
1016 struct socket *sock;
1017 struct sock *sk;
1018 void __user *argp = (void __user *)arg;
1019 int pid, err;
1020 struct net *net;
1021
1022 sock = file->private_data;
1023 sk = sock->sk;
1024 net = sock_net(sk);
1025 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1026 struct ifreq ifr;
1027 bool need_copyout;
1028 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1029 return -EFAULT;
1030 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1031 if (!err && need_copyout)
1032 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1033 return -EFAULT;
1034 } else
1035#ifdef CONFIG_WEXT_CORE
1036 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1037 err = wext_handle_ioctl(net, cmd, argp);
1038 } else
1039#endif
1040 switch (cmd) {
1041 case FIOSETOWN:
1042 case SIOCSPGRP:
1043 err = -EFAULT;
1044 if (get_user(pid, (int __user *)argp))
1045 break;
1046 err = f_setown(sock->file, pid, 1);
1047 break;
1048 case FIOGETOWN:
1049 case SIOCGPGRP:
1050 err = put_user(f_getown(sock->file),
1051 (int __user *)argp);
1052 break;
1053 case SIOCGIFBR:
1054 case SIOCSIFBR:
1055 case SIOCBRADDBR:
1056 case SIOCBRDELBR:
1057 err = -ENOPKG;
1058 if (!br_ioctl_hook)
1059 request_module("bridge");
1060
1061 mutex_lock(&br_ioctl_mutex);
1062 if (br_ioctl_hook)
1063 err = br_ioctl_hook(net, cmd, argp);
1064 mutex_unlock(&br_ioctl_mutex);
1065 break;
1066 case SIOCGIFVLAN:
1067 case SIOCSIFVLAN:
1068 err = -ENOPKG;
1069 if (!vlan_ioctl_hook)
1070 request_module("8021q");
1071
1072 mutex_lock(&vlan_ioctl_mutex);
1073 if (vlan_ioctl_hook)
1074 err = vlan_ioctl_hook(net, argp);
1075 mutex_unlock(&vlan_ioctl_mutex);
1076 break;
1077 case SIOCADDDLCI:
1078 case SIOCDELDLCI:
1079 err = -ENOPKG;
1080 if (!dlci_ioctl_hook)
1081 request_module("dlci");
1082
1083 mutex_lock(&dlci_ioctl_mutex);
1084 if (dlci_ioctl_hook)
1085 err = dlci_ioctl_hook(cmd, argp);
1086 mutex_unlock(&dlci_ioctl_mutex);
1087 break;
1088 case SIOCGSKNS:
1089 err = -EPERM;
1090 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1091 break;
1092
1093 err = open_related_ns(&net->ns, get_net_ns);
1094 break;
1095 default:
1096 err = sock_do_ioctl(net, sock, cmd, arg);
1097 break;
1098 }
1099 return err;
1100}
1101
1102int sock_create_lite(int family, int type, int protocol, struct socket **res)
1103{
1104 int err;
1105 struct socket *sock = NULL;
1106
1107 err = security_socket_create(family, type, protocol, 1);
1108 if (err)
1109 goto out;
1110
1111 sock = sock_alloc();
1112 if (!sock) {
1113 err = -ENOMEM;
1114 goto out;
1115 }
1116
1117 sock->type = type;
1118 err = security_socket_post_create(sock, family, type, protocol, 1);
1119 if (err)
1120 goto out_release;
1121
1122out:
1123 *res = sock;
1124 return err;
1125out_release:
1126 sock_release(sock);
1127 sock = NULL;
1128 goto out;
1129}
1130EXPORT_SYMBOL(sock_create_lite);
1131
1132/* No kernel lock held - perfect */
1133static __poll_t sock_poll(struct file *file, poll_table *wait)
1134{
1135 struct socket *sock = file->private_data;
1136 __poll_t events = poll_requested_events(wait), flag = 0;
1137
1138 if (!sock->ops->poll)
1139 return 0;
1140
1141 if (sk_can_busy_loop(sock->sk)) {
1142 /* poll once if requested by the syscall */
1143 if (events & POLL_BUSY_LOOP)
1144 sk_busy_loop(sock->sk, 1);
1145
1146 /* if this socket can poll_ll, tell the system call */
1147 flag = POLL_BUSY_LOOP;
1148 }
1149
1150 return sock->ops->poll(file, sock, wait) | flag;
1151}
1152
1153static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1154{
1155 struct socket *sock = file->private_data;
1156
1157 return sock->ops->mmap(file, sock, vma);
1158}
1159
1160static int sock_close(struct inode *inode, struct file *filp)
1161{
1162 __sock_release(SOCKET_I(inode), inode);
1163 return 0;
1164}
1165
1166/*
1167 * Update the socket async list
1168 *
1169 * Fasync_list locking strategy.
1170 *
1171 * 1. fasync_list is modified only under process context socket lock
1172 * i.e. under semaphore.
1173 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1174 * or under socket lock
1175 */
1176
1177static int sock_fasync(int fd, struct file *filp, int on)
1178{
1179 struct socket *sock = filp->private_data;
1180 struct sock *sk = sock->sk;
1181 struct socket_wq *wq;
1182
1183 if (sk == NULL)
1184 return -EINVAL;
1185
1186 lock_sock(sk);
1187 wq = sock->wq;
1188 fasync_helper(fd, filp, on, &wq->fasync_list);
1189
1190 if (!wq->fasync_list)
1191 sock_reset_flag(sk, SOCK_FASYNC);
1192 else
1193 sock_set_flag(sk, SOCK_FASYNC);
1194
1195 release_sock(sk);
1196 return 0;
1197}
1198
1199/* This function may be called only under rcu_lock */
1200
1201int sock_wake_async(struct socket_wq *wq, int how, int band)
1202{
1203 if (!wq || !wq->fasync_list)
1204 return -1;
1205
1206 switch (how) {
1207 case SOCK_WAKE_WAITD:
1208 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1209 break;
1210 goto call_kill;
1211 case SOCK_WAKE_SPACE:
1212 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1213 break;
1214 /* fall through */
1215 case SOCK_WAKE_IO:
1216call_kill:
1217 kill_fasync(&wq->fasync_list, SIGIO, band);
1218 break;
1219 case SOCK_WAKE_URG:
1220 kill_fasync(&wq->fasync_list, SIGURG, band);
1221 }
1222
1223 return 0;
1224}
1225EXPORT_SYMBOL(sock_wake_async);
1226
1227int __sock_create(struct net *net, int family, int type, int protocol,
1228 struct socket **res, int kern)
1229{
1230 int err;
1231 struct socket *sock;
1232 const struct net_proto_family *pf;
1233
1234 /*
1235 * Check protocol is in range
1236 */
1237 if (family < 0 || family >= NPROTO)
1238 return -EAFNOSUPPORT;
1239 if (type < 0 || type >= SOCK_MAX)
1240 return -EINVAL;
1241
1242 /* Compatibility.
1243
1244 This uglymoron is moved from INET layer to here to avoid
1245 deadlock in module load.
1246 */
1247 if (family == PF_INET && type == SOCK_PACKET) {
1248 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1249 current->comm);
1250 family = PF_PACKET;
1251 }
1252
1253 err = security_socket_create(family, type, protocol, kern);
1254 if (err)
1255 return err;
1256
1257 /*
1258 * Allocate the socket and allow the family to set things up. if
1259 * the protocol is 0, the family is instructed to select an appropriate
1260 * default.
1261 */
1262 sock = sock_alloc();
1263 if (!sock) {
1264 net_warn_ratelimited("socket: no more sockets\n");
1265 return -ENFILE; /* Not exactly a match, but its the
1266 closest posix thing */
1267 }
1268
1269 sock->type = type;
1270
1271#ifdef CONFIG_MODULES
1272 /* Attempt to load a protocol module if the find failed.
1273 *
1274 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1275 * requested real, full-featured networking support upon configuration.
1276 * Otherwise module support will break!
1277 */
1278 if (rcu_access_pointer(net_families[family]) == NULL)
1279 request_module("net-pf-%d", family);
1280#endif
1281
1282 rcu_read_lock();
1283 pf = rcu_dereference(net_families[family]);
1284 err = -EAFNOSUPPORT;
1285 if (!pf)
1286 goto out_release;
1287
1288 /*
1289 * We will call the ->create function, that possibly is in a loadable
1290 * module, so we have to bump that loadable module refcnt first.
1291 */
1292 if (!try_module_get(pf->owner))
1293 goto out_release;
1294
1295 /* Now protected by module ref count */
1296 rcu_read_unlock();
1297
1298 err = pf->create(net, sock, protocol, kern);
1299 if (err < 0)
1300 goto out_module_put;
1301
1302 /*
1303 * Now to bump the refcnt of the [loadable] module that owns this
1304 * socket at sock_release time we decrement its refcnt.
1305 */
1306 if (!try_module_get(sock->ops->owner))
1307 goto out_module_busy;
1308
1309 /*
1310 * Now that we're done with the ->create function, the [loadable]
1311 * module can have its refcnt decremented
1312 */
1313 module_put(pf->owner);
1314 err = security_socket_post_create(sock, family, type, protocol, kern);
1315 if (err)
1316 goto out_sock_release;
1317 *res = sock;
1318
1319 return 0;
1320
1321out_module_busy:
1322 err = -EAFNOSUPPORT;
1323out_module_put:
1324 sock->ops = NULL;
1325 module_put(pf->owner);
1326out_sock_release:
1327 sock_release(sock);
1328 return err;
1329
1330out_release:
1331 rcu_read_unlock();
1332 goto out_sock_release;
1333}
1334EXPORT_SYMBOL(__sock_create);
1335
1336int sock_create(int family, int type, int protocol, struct socket **res)
1337{
1338 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1339}
1340EXPORT_SYMBOL(sock_create);
1341
1342int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1343{
1344 return __sock_create(net, family, type, protocol, res, 1);
1345}
1346EXPORT_SYMBOL(sock_create_kern);
1347
1348int __sys_socket(int family, int type, int protocol)
1349{
1350 int retval;
1351 struct socket *sock;
1352 int flags;
1353
1354 /* Check the SOCK_* constants for consistency. */
1355 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1356 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1357 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1358 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1359
1360 flags = type & ~SOCK_TYPE_MASK;
1361 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1362 return -EINVAL;
1363 type &= SOCK_TYPE_MASK;
1364
1365 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1366 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1367
1368 retval = sock_create(family, type, protocol, &sock);
1369 if (retval < 0)
1370 return retval;
1371
1372 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1373}
1374
1375SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1376{
1377 return __sys_socket(family, type, protocol);
1378}
1379
1380/*
1381 * Create a pair of connected sockets.
1382 */
1383
1384int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1385{
1386 struct socket *sock1, *sock2;
1387 int fd1, fd2, err;
1388 struct file *newfile1, *newfile2;
1389 int flags;
1390
1391 flags = type & ~SOCK_TYPE_MASK;
1392 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1393 return -EINVAL;
1394 type &= SOCK_TYPE_MASK;
1395
1396 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1397 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1398
1399 /*
1400 * reserve descriptors and make sure we won't fail
1401 * to return them to userland.
1402 */
1403 fd1 = get_unused_fd_flags(flags);
1404 if (unlikely(fd1 < 0))
1405 return fd1;
1406
1407 fd2 = get_unused_fd_flags(flags);
1408 if (unlikely(fd2 < 0)) {
1409 put_unused_fd(fd1);
1410 return fd2;
1411 }
1412
1413 err = put_user(fd1, &usockvec[0]);
1414 if (err)
1415 goto out;
1416
1417 err = put_user(fd2, &usockvec[1]);
1418 if (err)
1419 goto out;
1420
1421 /*
1422 * Obtain the first socket and check if the underlying protocol
1423 * supports the socketpair call.
1424 */
1425
1426 err = sock_create(family, type, protocol, &sock1);
1427 if (unlikely(err < 0))
1428 goto out;
1429
1430 err = sock_create(family, type, protocol, &sock2);
1431 if (unlikely(err < 0)) {
1432 sock_release(sock1);
1433 goto out;
1434 }
1435
1436 err = security_socket_socketpair(sock1, sock2);
1437 if (unlikely(err)) {
1438 sock_release(sock2);
1439 sock_release(sock1);
1440 goto out;
1441 }
1442
1443 err = sock1->ops->socketpair(sock1, sock2);
1444 if (unlikely(err < 0)) {
1445 sock_release(sock2);
1446 sock_release(sock1);
1447 goto out;
1448 }
1449
1450 newfile1 = sock_alloc_file(sock1, flags, NULL);
1451 if (IS_ERR(newfile1)) {
1452 err = PTR_ERR(newfile1);
1453 sock_release(sock2);
1454 goto out;
1455 }
1456
1457 newfile2 = sock_alloc_file(sock2, flags, NULL);
1458 if (IS_ERR(newfile2)) {
1459 err = PTR_ERR(newfile2);
1460 fput(newfile1);
1461 goto out;
1462 }
1463
1464 audit_fd_pair(fd1, fd2);
1465
1466 fd_install(fd1, newfile1);
1467 fd_install(fd2, newfile2);
1468 return 0;
1469
1470out:
1471 put_unused_fd(fd2);
1472 put_unused_fd(fd1);
1473 return err;
1474}
1475
1476SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1477 int __user *, usockvec)
1478{
1479 return __sys_socketpair(family, type, protocol, usockvec);
1480}
1481
1482/*
1483 * Bind a name to a socket. Nothing much to do here since it's
1484 * the protocol's responsibility to handle the local address.
1485 *
1486 * We move the socket address to kernel space before we call
1487 * the protocol layer (having also checked the address is ok).
1488 */
1489
1490int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1491{
1492 struct socket *sock;
1493 struct sockaddr_storage address;
1494 int err, fput_needed;
1495
1496 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1497 if (sock) {
1498 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1499 if (!err) {
1500 err = security_socket_bind(sock,
1501 (struct sockaddr *)&address,
1502 addrlen);
1503 if (!err)
1504 err = sock->ops->bind(sock,
1505 (struct sockaddr *)
1506 &address, addrlen);
1507 }
1508 fput_light(sock->file, fput_needed);
1509 }
1510 return err;
1511}
1512
1513SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1514{
1515 return __sys_bind(fd, umyaddr, addrlen);
1516}
1517
1518/*
1519 * Perform a listen. Basically, we allow the protocol to do anything
1520 * necessary for a listen, and if that works, we mark the socket as
1521 * ready for listening.
1522 */
1523
1524int __sys_listen(int fd, int backlog)
1525{
1526 struct socket *sock;
1527 int err, fput_needed;
1528 int somaxconn;
1529
1530 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1531 if (sock) {
1532 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1533 if ((unsigned int)backlog > somaxconn)
1534 backlog = somaxconn;
1535
1536 err = security_socket_listen(sock, backlog);
1537 if (!err)
1538 err = sock->ops->listen(sock, backlog);
1539
1540 fput_light(sock->file, fput_needed);
1541 }
1542 return err;
1543}
1544
1545SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1546{
1547 return __sys_listen(fd, backlog);
1548}
1549
1550/*
1551 * For accept, we attempt to create a new socket, set up the link
1552 * with the client, wake up the client, then return the new
1553 * connected fd. We collect the address of the connector in kernel
1554 * space and move it to user at the very end. This is unclean because
1555 * we open the socket then return an error.
1556 *
1557 * 1003.1g adds the ability to recvmsg() to query connection pending
1558 * status to recvmsg. We need to add that support in a way thats
1559 * clean when we restructure accept also.
1560 */
1561
1562int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1563 int __user *upeer_addrlen, int flags)
1564{
1565 struct socket *sock, *newsock;
1566 struct file *newfile;
1567 int err, len, newfd, fput_needed;
1568 struct sockaddr_storage address;
1569
1570 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1571 return -EINVAL;
1572
1573 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1574 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1575
1576 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1577 if (!sock)
1578 goto out;
1579
1580 err = -ENFILE;
1581 newsock = sock_alloc();
1582 if (!newsock)
1583 goto out_put;
1584
1585 newsock->type = sock->type;
1586 newsock->ops = sock->ops;
1587
1588 /*
1589 * We don't need try_module_get here, as the listening socket (sock)
1590 * has the protocol module (sock->ops->owner) held.
1591 */
1592 __module_get(newsock->ops->owner);
1593
1594 newfd = get_unused_fd_flags(flags);
1595 if (unlikely(newfd < 0)) {
1596 err = newfd;
1597 sock_release(newsock);
1598 goto out_put;
1599 }
1600 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1601 if (IS_ERR(newfile)) {
1602 err = PTR_ERR(newfile);
1603 put_unused_fd(newfd);
1604 goto out_put;
1605 }
1606
1607 err = security_socket_accept(sock, newsock);
1608 if (err)
1609 goto out_fd;
1610
1611 err = sock->ops->accept(sock, newsock, sock->file->f_flags, false);
1612 if (err < 0)
1613 goto out_fd;
1614
1615 if (upeer_sockaddr) {
1616 len = newsock->ops->getname(newsock,
1617 (struct sockaddr *)&address, 2);
1618 if (len < 0) {
1619 err = -ECONNABORTED;
1620 goto out_fd;
1621 }
1622 err = move_addr_to_user(&address,
1623 len, upeer_sockaddr, upeer_addrlen);
1624 if (err < 0)
1625 goto out_fd;
1626 }
1627
1628 /* File flags are not inherited via accept() unlike another OSes. */
1629
1630 fd_install(newfd, newfile);
1631 err = newfd;
1632
1633out_put:
1634 fput_light(sock->file, fput_needed);
1635out:
1636 return err;
1637out_fd:
1638 fput(newfile);
1639 put_unused_fd(newfd);
1640 goto out_put;
1641}
1642
1643SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1644 int __user *, upeer_addrlen, int, flags)
1645{
1646 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1647}
1648
1649SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1650 int __user *, upeer_addrlen)
1651{
1652 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1653}
1654
1655/*
1656 * Attempt to connect to a socket with the server address. The address
1657 * is in user space so we verify it is OK and move it to kernel space.
1658 *
1659 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1660 * break bindings
1661 *
1662 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1663 * other SEQPACKET protocols that take time to connect() as it doesn't
1664 * include the -EINPROGRESS status for such sockets.
1665 */
1666
1667int __sys_connect(int fd, struct sockaddr __user *uservaddr, 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 goto out;
1676 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1677 if (err < 0)
1678 goto out_put;
1679
1680 err =
1681 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1682 if (err)
1683 goto out_put;
1684
1685 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1686 sock->file->f_flags);
1687out_put:
1688 fput_light(sock->file, fput_needed);
1689out:
1690 return err;
1691}
1692
1693SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1694 int, addrlen)
1695{
1696 return __sys_connect(fd, uservaddr, addrlen);
1697}
1698
1699/*
1700 * Get the local address ('name') of a socket object. Move the obtained
1701 * name to user space.
1702 */
1703
1704int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1705 int __user *usockaddr_len)
1706{
1707 struct socket *sock;
1708 struct sockaddr_storage address;
1709 int err, fput_needed;
1710
1711 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1712 if (!sock)
1713 goto out;
1714
1715 err = security_socket_getsockname(sock);
1716 if (err)
1717 goto out_put;
1718
1719 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1720 if (err < 0)
1721 goto out_put;
1722 /* "err" is actually length in this case */
1723 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1724
1725out_put:
1726 fput_light(sock->file, fput_needed);
1727out:
1728 return err;
1729}
1730
1731SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1732 int __user *, usockaddr_len)
1733{
1734 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1735}
1736
1737/*
1738 * Get the remote address ('name') of a socket object. Move the obtained
1739 * name to user space.
1740 */
1741
1742int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1743 int __user *usockaddr_len)
1744{
1745 struct socket *sock;
1746 struct sockaddr_storage address;
1747 int err, fput_needed;
1748
1749 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1750 if (sock != NULL) {
1751 err = security_socket_getpeername(sock);
1752 if (err) {
1753 fput_light(sock->file, fput_needed);
1754 return err;
1755 }
1756
1757 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1758 if (err >= 0)
1759 /* "err" is actually length in this case */
1760 err = move_addr_to_user(&address, err, usockaddr,
1761 usockaddr_len);
1762 fput_light(sock->file, fput_needed);
1763 }
1764 return err;
1765}
1766
1767SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1768 int __user *, usockaddr_len)
1769{
1770 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1771}
1772
1773/*
1774 * Send a datagram to a given address. We move the address into kernel
1775 * space and check the user space data area is readable before invoking
1776 * the protocol.
1777 */
1778int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1779 struct sockaddr __user *addr, int addr_len)
1780{
1781 struct socket *sock;
1782 struct sockaddr_storage address;
1783 int err;
1784 struct msghdr msg;
1785 struct iovec iov;
1786 int fput_needed;
1787
1788 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1789 if (unlikely(err))
1790 return err;
1791 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1792 if (!sock)
1793 goto out;
1794
1795 msg.msg_name = NULL;
1796 msg.msg_control = NULL;
1797 msg.msg_controllen = 0;
1798 msg.msg_namelen = 0;
1799 if (addr) {
1800 err = move_addr_to_kernel(addr, addr_len, &address);
1801 if (err < 0)
1802 goto out_put;
1803 msg.msg_name = (struct sockaddr *)&address;
1804 msg.msg_namelen = addr_len;
1805 }
1806 if (sock->file->f_flags & O_NONBLOCK)
1807 flags |= MSG_DONTWAIT;
1808 msg.msg_flags = flags;
1809 err = sock_sendmsg(sock, &msg);
1810
1811out_put:
1812 fput_light(sock->file, fput_needed);
1813out:
1814 return err;
1815}
1816
1817SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1818 unsigned int, flags, struct sockaddr __user *, addr,
1819 int, addr_len)
1820{
1821 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
1822}
1823
1824/*
1825 * Send a datagram down a socket.
1826 */
1827
1828SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1829 unsigned int, flags)
1830{
1831 return __sys_sendto(fd, buff, len, flags, NULL, 0);
1832}
1833
1834/*
1835 * Receive a frame from the socket and optionally record the address of the
1836 * sender. We verify the buffers are writable and if needed move the
1837 * sender address from kernel to user space.
1838 */
1839int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
1840 struct sockaddr __user *addr, int __user *addr_len)
1841{
1842 struct socket *sock;
1843 struct iovec iov;
1844 struct msghdr msg;
1845 struct sockaddr_storage address;
1846 int err, err2;
1847 int fput_needed;
1848
1849 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1850 if (unlikely(err))
1851 return err;
1852 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1853 if (!sock)
1854 goto out;
1855
1856 msg.msg_control = NULL;
1857 msg.msg_controllen = 0;
1858 /* Save some cycles and don't copy the address if not needed */
1859 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
1860 /* We assume all kernel code knows the size of sockaddr_storage */
1861 msg.msg_namelen = 0;
1862 msg.msg_iocb = NULL;
1863 msg.msg_flags = 0;
1864 if (sock->file->f_flags & O_NONBLOCK)
1865 flags |= MSG_DONTWAIT;
1866 err = sock_recvmsg(sock, &msg, flags);
1867
1868 if (err >= 0 && addr != NULL) {
1869 err2 = move_addr_to_user(&address,
1870 msg.msg_namelen, addr, addr_len);
1871 if (err2 < 0)
1872 err = err2;
1873 }
1874
1875 fput_light(sock->file, fput_needed);
1876out:
1877 return err;
1878}
1879
1880SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1881 unsigned int, flags, struct sockaddr __user *, addr,
1882 int __user *, addr_len)
1883{
1884 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
1885}
1886
1887/*
1888 * Receive a datagram from a socket.
1889 */
1890
1891SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
1892 unsigned int, flags)
1893{
1894 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1895}
1896
1897/*
1898 * Set a socket option. Because we don't know the option lengths we have
1899 * to pass the user mode parameter for the protocols to sort out.
1900 */
1901
1902static int __sys_setsockopt(int fd, int level, int optname,
1903 char __user *optval, int optlen)
1904{
1905 int err, fput_needed;
1906 struct socket *sock;
1907
1908 if (optlen < 0)
1909 return -EINVAL;
1910
1911 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1912 if (sock != NULL) {
1913 err = security_socket_setsockopt(sock, level, optname);
1914 if (err)
1915 goto out_put;
1916
1917 if (level == SOL_SOCKET)
1918 err =
1919 sock_setsockopt(sock, level, optname, optval,
1920 optlen);
1921 else
1922 err =
1923 sock->ops->setsockopt(sock, level, optname, optval,
1924 optlen);
1925out_put:
1926 fput_light(sock->file, fput_needed);
1927 }
1928 return err;
1929}
1930
1931SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1932 char __user *, optval, int, optlen)
1933{
1934 return __sys_setsockopt(fd, level, optname, optval, optlen);
1935}
1936
1937/*
1938 * Get a socket option. Because we don't know the option lengths we have
1939 * to pass a user mode parameter for the protocols to sort out.
1940 */
1941
1942static int __sys_getsockopt(int fd, int level, int optname,
1943 char __user *optval, int __user *optlen)
1944{
1945 int err, fput_needed;
1946 struct socket *sock;
1947
1948 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1949 if (sock != NULL) {
1950 err = security_socket_getsockopt(sock, level, optname);
1951 if (err)
1952 goto out_put;
1953
1954 if (level == SOL_SOCKET)
1955 err =
1956 sock_getsockopt(sock, level, optname, optval,
1957 optlen);
1958 else
1959 err =
1960 sock->ops->getsockopt(sock, level, optname, optval,
1961 optlen);
1962out_put:
1963 fput_light(sock->file, fput_needed);
1964 }
1965 return err;
1966}
1967
1968SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1969 char __user *, optval, int __user *, optlen)
1970{
1971 return __sys_getsockopt(fd, level, optname, optval, optlen);
1972}
1973
1974/*
1975 * Shutdown a socket.
1976 */
1977
1978int __sys_shutdown(int fd, int how)
1979{
1980 int err, fput_needed;
1981 struct socket *sock;
1982
1983 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1984 if (sock != NULL) {
1985 err = security_socket_shutdown(sock, how);
1986 if (!err)
1987 err = sock->ops->shutdown(sock, how);
1988 fput_light(sock->file, fput_needed);
1989 }
1990 return err;
1991}
1992
1993SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1994{
1995 return __sys_shutdown(fd, how);
1996}
1997
1998/* A couple of helpful macros for getting the address of the 32/64 bit
1999 * fields which are the same type (int / unsigned) on our platforms.
2000 */
2001#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2002#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2003#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2004
2005struct used_address {
2006 struct sockaddr_storage name;
2007 unsigned int name_len;
2008};
2009
2010static int copy_msghdr_from_user(struct msghdr *kmsg,
2011 struct user_msghdr __user *umsg,
2012 struct sockaddr __user **save_addr,
2013 struct iovec **iov)
2014{
2015 struct user_msghdr msg;
2016 ssize_t err;
2017
2018 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2019 return -EFAULT;
2020
2021 kmsg->msg_control = (void __force *)msg.msg_control;
2022 kmsg->msg_controllen = msg.msg_controllen;
2023 kmsg->msg_flags = msg.msg_flags;
2024
2025 kmsg->msg_namelen = msg.msg_namelen;
2026 if (!msg.msg_name)
2027 kmsg->msg_namelen = 0;
2028
2029 if (kmsg->msg_namelen < 0)
2030 return -EINVAL;
2031
2032 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2033 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2034
2035 if (save_addr)
2036 *save_addr = msg.msg_name;
2037
2038 if (msg.msg_name && kmsg->msg_namelen) {
2039 if (!save_addr) {
2040 err = move_addr_to_kernel(msg.msg_name,
2041 kmsg->msg_namelen,
2042 kmsg->msg_name);
2043 if (err < 0)
2044 return err;
2045 }
2046 } else {
2047 kmsg->msg_name = NULL;
2048 kmsg->msg_namelen = 0;
2049 }
2050
2051 if (msg.msg_iovlen > UIO_MAXIOV)
2052 return -EMSGSIZE;
2053
2054 kmsg->msg_iocb = NULL;
2055
2056 return import_iovec(save_addr ? READ : WRITE,
2057 msg.msg_iov, msg.msg_iovlen,
2058 UIO_FASTIOV, iov, &kmsg->msg_iter);
2059}
2060
2061static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2062 struct msghdr *msg_sys, unsigned int flags,
2063 struct used_address *used_address,
2064 unsigned int allowed_msghdr_flags)
2065{
2066 struct compat_msghdr __user *msg_compat =
2067 (struct compat_msghdr __user *)msg;
2068 struct sockaddr_storage address;
2069 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2070 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2071 __aligned(sizeof(__kernel_size_t));
2072 /* 20 is size of ipv6_pktinfo */
2073 unsigned char *ctl_buf = ctl;
2074 int ctl_len;
2075 ssize_t err;
2076
2077 msg_sys->msg_name = &address;
2078
2079 if (MSG_CMSG_COMPAT & flags)
2080 err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
2081 else
2082 err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
2083 if (err < 0)
2084 return err;
2085
2086 err = -ENOBUFS;
2087
2088 if (msg_sys->msg_controllen > INT_MAX)
2089 goto out_freeiov;
2090 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2091 ctl_len = msg_sys->msg_controllen;
2092 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2093 err =
2094 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2095 sizeof(ctl));
2096 if (err)
2097 goto out_freeiov;
2098 ctl_buf = msg_sys->msg_control;
2099 ctl_len = msg_sys->msg_controllen;
2100 } else if (ctl_len) {
2101 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2102 CMSG_ALIGN(sizeof(struct cmsghdr)));
2103 if (ctl_len > sizeof(ctl)) {
2104 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2105 if (ctl_buf == NULL)
2106 goto out_freeiov;
2107 }
2108 err = -EFAULT;
2109 /*
2110 * Careful! Before this, msg_sys->msg_control contains a user pointer.
2111 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
2112 * checking falls down on this.
2113 */
2114 if (copy_from_user(ctl_buf,
2115 (void __user __force *)msg_sys->msg_control,
2116 ctl_len))
2117 goto out_freectl;
2118 msg_sys->msg_control = ctl_buf;
2119 }
2120 msg_sys->msg_flags = flags;
2121
2122 if (sock->file->f_flags & O_NONBLOCK)
2123 msg_sys->msg_flags |= MSG_DONTWAIT;
2124 /*
2125 * If this is sendmmsg() and current destination address is same as
2126 * previously succeeded address, omit asking LSM's decision.
2127 * used_address->name_len is initialized to UINT_MAX so that the first
2128 * destination address never matches.
2129 */
2130 if (used_address && msg_sys->msg_name &&
2131 used_address->name_len == msg_sys->msg_namelen &&
2132 !memcmp(&used_address->name, msg_sys->msg_name,
2133 used_address->name_len)) {
2134 err = sock_sendmsg_nosec(sock, msg_sys);
2135 goto out_freectl;
2136 }
2137 err = sock_sendmsg(sock, msg_sys);
2138 /*
2139 * If this is sendmmsg() and sending to current destination address was
2140 * successful, remember it.
2141 */
2142 if (used_address && err >= 0) {
2143 used_address->name_len = msg_sys->msg_namelen;
2144 if (msg_sys->msg_name)
2145 memcpy(&used_address->name, msg_sys->msg_name,
2146 used_address->name_len);
2147 }
2148
2149out_freectl:
2150 if (ctl_buf != ctl)
2151 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2152out_freeiov:
2153 kfree(iov);
2154 return err;
2155}
2156
2157/*
2158 * BSD sendmsg interface
2159 */
2160
2161long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2162 bool forbid_cmsg_compat)
2163{
2164 int fput_needed, err;
2165 struct msghdr msg_sys;
2166 struct socket *sock;
2167
2168 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2169 return -EINVAL;
2170
2171 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2172 if (!sock)
2173 goto out;
2174
2175 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2176
2177 fput_light(sock->file, fput_needed);
2178out:
2179 return err;
2180}
2181
2182SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2183{
2184 return __sys_sendmsg(fd, msg, flags, true);
2185}
2186
2187/*
2188 * Linux sendmmsg interface
2189 */
2190
2191int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2192 unsigned int flags, bool forbid_cmsg_compat)
2193{
2194 int fput_needed, err, datagrams;
2195 struct socket *sock;
2196 struct mmsghdr __user *entry;
2197 struct compat_mmsghdr __user *compat_entry;
2198 struct msghdr msg_sys;
2199 struct used_address used_address;
2200 unsigned int oflags = flags;
2201
2202 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2203 return -EINVAL;
2204
2205 if (vlen > UIO_MAXIOV)
2206 vlen = UIO_MAXIOV;
2207
2208 datagrams = 0;
2209
2210 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2211 if (!sock)
2212 return err;
2213
2214 used_address.name_len = UINT_MAX;
2215 entry = mmsg;
2216 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2217 err = 0;
2218 flags |= MSG_BATCH;
2219
2220 while (datagrams < vlen) {
2221 if (datagrams == vlen - 1)
2222 flags = oflags;
2223
2224 if (MSG_CMSG_COMPAT & flags) {
2225 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2226 &msg_sys, flags, &used_address, MSG_EOR);
2227 if (err < 0)
2228 break;
2229 err = __put_user(err, &compat_entry->msg_len);
2230 ++compat_entry;
2231 } else {
2232 err = ___sys_sendmsg(sock,
2233 (struct user_msghdr __user *)entry,
2234 &msg_sys, flags, &used_address, MSG_EOR);
2235 if (err < 0)
2236 break;
2237 err = put_user(err, &entry->msg_len);
2238 ++entry;
2239 }
2240
2241 if (err)
2242 break;
2243 ++datagrams;
2244 if (msg_data_left(&msg_sys))
2245 break;
2246 cond_resched();
2247 }
2248
2249 fput_light(sock->file, fput_needed);
2250
2251 /* We only return an error if no datagrams were able to be sent */
2252 if (datagrams != 0)
2253 return datagrams;
2254
2255 return err;
2256}
2257
2258SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2259 unsigned int, vlen, unsigned int, flags)
2260{
2261 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2262}
2263
2264static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2265 struct msghdr *msg_sys, unsigned int flags, int nosec)
2266{
2267 struct compat_msghdr __user *msg_compat =
2268 (struct compat_msghdr __user *)msg;
2269 struct iovec iovstack[UIO_FASTIOV];
2270 struct iovec *iov = iovstack;
2271 unsigned long cmsg_ptr;
2272 int len;
2273 ssize_t err;
2274
2275 /* kernel mode address */
2276 struct sockaddr_storage addr;
2277
2278 /* user mode address pointers */
2279 struct sockaddr __user *uaddr;
2280 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2281
2282 msg_sys->msg_name = &addr;
2283
2284 if (MSG_CMSG_COMPAT & flags)
2285 err = get_compat_msghdr(msg_sys, msg_compat, &uaddr, &iov);
2286 else
2287 err = copy_msghdr_from_user(msg_sys, msg, &uaddr, &iov);
2288 if (err < 0)
2289 return err;
2290
2291 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2292 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2293
2294 /* We assume all kernel code knows the size of sockaddr_storage */
2295 msg_sys->msg_namelen = 0;
2296
2297 if (sock->file->f_flags & O_NONBLOCK)
2298 flags |= MSG_DONTWAIT;
2299 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys, flags);
2300 if (err < 0)
2301 goto out_freeiov;
2302 len = err;
2303
2304 if (uaddr != NULL) {
2305 err = move_addr_to_user(&addr,
2306 msg_sys->msg_namelen, uaddr,
2307 uaddr_len);
2308 if (err < 0)
2309 goto out_freeiov;
2310 }
2311 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2312 COMPAT_FLAGS(msg));
2313 if (err)
2314 goto out_freeiov;
2315 if (MSG_CMSG_COMPAT & flags)
2316 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2317 &msg_compat->msg_controllen);
2318 else
2319 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2320 &msg->msg_controllen);
2321 if (err)
2322 goto out_freeiov;
2323 err = len;
2324
2325out_freeiov:
2326 kfree(iov);
2327 return err;
2328}
2329
2330/*
2331 * BSD recvmsg interface
2332 */
2333
2334long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2335 bool forbid_cmsg_compat)
2336{
2337 int fput_needed, err;
2338 struct msghdr msg_sys;
2339 struct socket *sock;
2340
2341 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2342 return -EINVAL;
2343
2344 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2345 if (!sock)
2346 goto out;
2347
2348 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2349
2350 fput_light(sock->file, fput_needed);
2351out:
2352 return err;
2353}
2354
2355SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2356 unsigned int, flags)
2357{
2358 return __sys_recvmsg(fd, msg, flags, true);
2359}
2360
2361/*
2362 * Linux recvmmsg interface
2363 */
2364
2365static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2366 unsigned int vlen, unsigned int flags,
2367 struct timespec64 *timeout)
2368{
2369 int fput_needed, err, datagrams;
2370 struct socket *sock;
2371 struct mmsghdr __user *entry;
2372 struct compat_mmsghdr __user *compat_entry;
2373 struct msghdr msg_sys;
2374 struct timespec64 end_time;
2375 struct timespec64 timeout64;
2376
2377 if (timeout &&
2378 poll_select_set_timeout(&end_time, timeout->tv_sec,
2379 timeout->tv_nsec))
2380 return -EINVAL;
2381
2382 datagrams = 0;
2383
2384 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2385 if (!sock)
2386 return err;
2387
2388 if (likely(!(flags & MSG_ERRQUEUE))) {
2389 err = sock_error(sock->sk);
2390 if (err) {
2391 datagrams = err;
2392 goto out_put;
2393 }
2394 }
2395
2396 entry = mmsg;
2397 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2398
2399 while (datagrams < vlen) {
2400 /*
2401 * No need to ask LSM for more than the first datagram.
2402 */
2403 if (MSG_CMSG_COMPAT & flags) {
2404 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2405 &msg_sys, flags & ~MSG_WAITFORONE,
2406 datagrams);
2407 if (err < 0)
2408 break;
2409 err = __put_user(err, &compat_entry->msg_len);
2410 ++compat_entry;
2411 } else {
2412 err = ___sys_recvmsg(sock,
2413 (struct user_msghdr __user *)entry,
2414 &msg_sys, flags & ~MSG_WAITFORONE,
2415 datagrams);
2416 if (err < 0)
2417 break;
2418 err = put_user(err, &entry->msg_len);
2419 ++entry;
2420 }
2421
2422 if (err)
2423 break;
2424 ++datagrams;
2425
2426 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2427 if (flags & MSG_WAITFORONE)
2428 flags |= MSG_DONTWAIT;
2429
2430 if (timeout) {
2431 ktime_get_ts64(&timeout64);
2432 *timeout = timespec64_sub(end_time, timeout64);
2433 if (timeout->tv_sec < 0) {
2434 timeout->tv_sec = timeout->tv_nsec = 0;
2435 break;
2436 }
2437
2438 /* Timeout, return less than vlen datagrams */
2439 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2440 break;
2441 }
2442
2443 /* Out of band data, return right away */
2444 if (msg_sys.msg_flags & MSG_OOB)
2445 break;
2446 cond_resched();
2447 }
2448
2449 if (err == 0)
2450 goto out_put;
2451
2452 if (datagrams == 0) {
2453 datagrams = err;
2454 goto out_put;
2455 }
2456
2457 /*
2458 * We may return less entries than requested (vlen) if the
2459 * sock is non block and there aren't enough datagrams...
2460 */
2461 if (err != -EAGAIN) {
2462 /*
2463 * ... or if recvmsg returns an error after we
2464 * received some datagrams, where we record the
2465 * error to return on the next call or if the
2466 * app asks about it using getsockopt(SO_ERROR).
2467 */
2468 sock->sk->sk_err = -err;
2469 }
2470out_put:
2471 fput_light(sock->file, fput_needed);
2472
2473 return datagrams;
2474}
2475
2476int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2477 unsigned int vlen, unsigned int flags,
2478 struct __kernel_timespec __user *timeout,
2479 struct old_timespec32 __user *timeout32)
2480{
2481 int datagrams;
2482 struct timespec64 timeout_sys;
2483
2484 if (timeout && get_timespec64(&timeout_sys, timeout))
2485 return -EFAULT;
2486
2487 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2488 return -EFAULT;
2489
2490 if (!timeout && !timeout32)
2491 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2492
2493 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2494
2495 if (datagrams <= 0)
2496 return datagrams;
2497
2498 if (timeout && put_timespec64(&timeout_sys, timeout))
2499 datagrams = -EFAULT;
2500
2501 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2502 datagrams = -EFAULT;
2503
2504 return datagrams;
2505}
2506
2507SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2508 unsigned int, vlen, unsigned int, flags,
2509 struct __kernel_timespec __user *, timeout)
2510{
2511 if (flags & MSG_CMSG_COMPAT)
2512 return -EINVAL;
2513
2514 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2515}
2516
2517#ifdef CONFIG_COMPAT_32BIT_TIME
2518SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2519 unsigned int, vlen, unsigned int, flags,
2520 struct old_timespec32 __user *, timeout)
2521{
2522 if (flags & MSG_CMSG_COMPAT)
2523 return -EINVAL;
2524
2525 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2526}
2527#endif
2528
2529#ifdef __ARCH_WANT_SYS_SOCKETCALL
2530/* Argument list sizes for sys_socketcall */
2531#define AL(x) ((x) * sizeof(unsigned long))
2532static const unsigned char nargs[21] = {
2533 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2534 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2535 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2536 AL(4), AL(5), AL(4)
2537};
2538
2539#undef AL
2540
2541/*
2542 * System call vectors.
2543 *
2544 * Argument checking cleaned up. Saved 20% in size.
2545 * This function doesn't need to set the kernel lock because
2546 * it is set by the callees.
2547 */
2548
2549SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2550{
2551 unsigned long a[AUDITSC_ARGS];
2552 unsigned long a0, a1;
2553 int err;
2554 unsigned int len;
2555
2556 if (call < 1 || call > SYS_SENDMMSG)
2557 return -EINVAL;
2558 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2559
2560 len = nargs[call];
2561 if (len > sizeof(a))
2562 return -EINVAL;
2563
2564 /* copy_from_user should be SMP safe. */
2565 if (copy_from_user(a, args, len))
2566 return -EFAULT;
2567
2568 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2569 if (err)
2570 return err;
2571
2572 a0 = a[0];
2573 a1 = a[1];
2574
2575 switch (call) {
2576 case SYS_SOCKET:
2577 err = __sys_socket(a0, a1, a[2]);
2578 break;
2579 case SYS_BIND:
2580 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2581 break;
2582 case SYS_CONNECT:
2583 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2584 break;
2585 case SYS_LISTEN:
2586 err = __sys_listen(a0, a1);
2587 break;
2588 case SYS_ACCEPT:
2589 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2590 (int __user *)a[2], 0);
2591 break;
2592 case SYS_GETSOCKNAME:
2593 err =
2594 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2595 (int __user *)a[2]);
2596 break;
2597 case SYS_GETPEERNAME:
2598 err =
2599 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2600 (int __user *)a[2]);
2601 break;
2602 case SYS_SOCKETPAIR:
2603 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2604 break;
2605 case SYS_SEND:
2606 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2607 NULL, 0);
2608 break;
2609 case SYS_SENDTO:
2610 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2611 (struct sockaddr __user *)a[4], a[5]);
2612 break;
2613 case SYS_RECV:
2614 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2615 NULL, NULL);
2616 break;
2617 case SYS_RECVFROM:
2618 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2619 (struct sockaddr __user *)a[4],
2620 (int __user *)a[5]);
2621 break;
2622 case SYS_SHUTDOWN:
2623 err = __sys_shutdown(a0, a1);
2624 break;
2625 case SYS_SETSOCKOPT:
2626 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2627 a[4]);
2628 break;
2629 case SYS_GETSOCKOPT:
2630 err =
2631 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2632 (int __user *)a[4]);
2633 break;
2634 case SYS_SENDMSG:
2635 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2636 a[2], true);
2637 break;
2638 case SYS_SENDMMSG:
2639 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2640 a[3], true);
2641 break;
2642 case SYS_RECVMSG:
2643 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2644 a[2], true);
2645 break;
2646 case SYS_RECVMMSG:
2647 if (IS_ENABLED(CONFIG_64BIT) || !IS_ENABLED(CONFIG_64BIT_TIME))
2648 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2649 a[2], a[3],
2650 (struct __kernel_timespec __user *)a[4],
2651 NULL);
2652 else
2653 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2654 a[2], a[3], NULL,
2655 (struct old_timespec32 __user *)a[4]);
2656 break;
2657 case SYS_ACCEPT4:
2658 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2659 (int __user *)a[2], a[3]);
2660 break;
2661 default:
2662 err = -EINVAL;
2663 break;
2664 }
2665 return err;
2666}
2667
2668#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2669
2670/**
2671 * sock_register - add a socket protocol handler
2672 * @ops: description of protocol
2673 *
2674 * This function is called by a protocol handler that wants to
2675 * advertise its address family, and have it linked into the
2676 * socket interface. The value ops->family corresponds to the
2677 * socket system call protocol family.
2678 */
2679int sock_register(const struct net_proto_family *ops)
2680{
2681 int err;
2682
2683 if (ops->family >= NPROTO) {
2684 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2685 return -ENOBUFS;
2686 }
2687
2688 spin_lock(&net_family_lock);
2689 if (rcu_dereference_protected(net_families[ops->family],
2690 lockdep_is_held(&net_family_lock)))
2691 err = -EEXIST;
2692 else {
2693 rcu_assign_pointer(net_families[ops->family], ops);
2694 err = 0;
2695 }
2696 spin_unlock(&net_family_lock);
2697
2698 pr_info("NET: Registered protocol family %d\n", ops->family);
2699 return err;
2700}
2701EXPORT_SYMBOL(sock_register);
2702
2703/**
2704 * sock_unregister - remove a protocol handler
2705 * @family: protocol family to remove
2706 *
2707 * This function is called by a protocol handler that wants to
2708 * remove its address family, and have it unlinked from the
2709 * new socket creation.
2710 *
2711 * If protocol handler is a module, then it can use module reference
2712 * counts to protect against new references. If protocol handler is not
2713 * a module then it needs to provide its own protection in
2714 * the ops->create routine.
2715 */
2716void sock_unregister(int family)
2717{
2718 BUG_ON(family < 0 || family >= NPROTO);
2719
2720 spin_lock(&net_family_lock);
2721 RCU_INIT_POINTER(net_families[family], NULL);
2722 spin_unlock(&net_family_lock);
2723
2724 synchronize_rcu();
2725
2726 pr_info("NET: Unregistered protocol family %d\n", family);
2727}
2728EXPORT_SYMBOL(sock_unregister);
2729
2730bool sock_is_registered(int family)
2731{
2732 return family < NPROTO && rcu_access_pointer(net_families[family]);
2733}
2734
2735static int __init sock_init(void)
2736{
2737 int err;
2738 /*
2739 * Initialize the network sysctl infrastructure.
2740 */
2741 err = net_sysctl_init();
2742 if (err)
2743 goto out;
2744
2745 /*
2746 * Initialize skbuff SLAB cache
2747 */
2748 skb_init();
2749
2750 /*
2751 * Initialize the protocols module.
2752 */
2753
2754 init_inodecache();
2755
2756 err = register_filesystem(&sock_fs_type);
2757 if (err)
2758 goto out_fs;
2759 sock_mnt = kern_mount(&sock_fs_type);
2760 if (IS_ERR(sock_mnt)) {
2761 err = PTR_ERR(sock_mnt);
2762 goto out_mount;
2763 }
2764
2765 /* The real protocol initialization is performed in later initcalls.
2766 */
2767
2768#ifdef CONFIG_NETFILTER
2769 err = netfilter_init();
2770 if (err)
2771 goto out;
2772#endif
2773
2774 ptp_classifier_init();
2775
2776out:
2777 return err;
2778
2779out_mount:
2780 unregister_filesystem(&sock_fs_type);
2781out_fs:
2782 goto out;
2783}
2784
2785core_initcall(sock_init); /* early initcall */
2786
2787#ifdef CONFIG_PROC_FS
2788void socket_seq_show(struct seq_file *seq)
2789{
2790 seq_printf(seq, "sockets: used %d\n",
2791 sock_inuse_get(seq->private));
2792}
2793#endif /* CONFIG_PROC_FS */
2794
2795#ifdef CONFIG_COMPAT
2796static int do_siocgstamp(struct net *net, struct socket *sock,
2797 unsigned int cmd, void __user *up)
2798{
2799 mm_segment_t old_fs = get_fs();
2800 struct timeval ktv;
2801 int err;
2802
2803 set_fs(KERNEL_DS);
2804 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2805 set_fs(old_fs);
2806 if (!err)
2807 err = compat_put_timeval(&ktv, up);
2808
2809 return err;
2810}
2811
2812static int do_siocgstampns(struct net *net, struct socket *sock,
2813 unsigned int cmd, void __user *up)
2814{
2815 mm_segment_t old_fs = get_fs();
2816 struct timespec kts;
2817 int err;
2818
2819 set_fs(KERNEL_DS);
2820 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2821 set_fs(old_fs);
2822 if (!err)
2823 err = compat_put_timespec(&kts, up);
2824
2825 return err;
2826}
2827
2828static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2829{
2830 struct compat_ifconf ifc32;
2831 struct ifconf ifc;
2832 int err;
2833
2834 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2835 return -EFAULT;
2836
2837 ifc.ifc_len = ifc32.ifc_len;
2838 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
2839
2840 rtnl_lock();
2841 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
2842 rtnl_unlock();
2843 if (err)
2844 return err;
2845
2846 ifc32.ifc_len = ifc.ifc_len;
2847 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2848 return -EFAULT;
2849
2850 return 0;
2851}
2852
2853static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2854{
2855 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2856 bool convert_in = false, convert_out = false;
2857 size_t buf_size = 0;
2858 struct ethtool_rxnfc __user *rxnfc = NULL;
2859 struct ifreq ifr;
2860 u32 rule_cnt = 0, actual_rule_cnt;
2861 u32 ethcmd;
2862 u32 data;
2863 int ret;
2864
2865 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2866 return -EFAULT;
2867
2868 compat_rxnfc = compat_ptr(data);
2869
2870 if (get_user(ethcmd, &compat_rxnfc->cmd))
2871 return -EFAULT;
2872
2873 /* Most ethtool structures are defined without padding.
2874 * Unfortunately struct ethtool_rxnfc is an exception.
2875 */
2876 switch (ethcmd) {
2877 default:
2878 break;
2879 case ETHTOOL_GRXCLSRLALL:
2880 /* Buffer size is variable */
2881 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2882 return -EFAULT;
2883 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2884 return -ENOMEM;
2885 buf_size += rule_cnt * sizeof(u32);
2886 /* fall through */
2887 case ETHTOOL_GRXRINGS:
2888 case ETHTOOL_GRXCLSRLCNT:
2889 case ETHTOOL_GRXCLSRULE:
2890 case ETHTOOL_SRXCLSRLINS:
2891 convert_out = true;
2892 /* fall through */
2893 case ETHTOOL_SRXCLSRLDEL:
2894 buf_size += sizeof(struct ethtool_rxnfc);
2895 convert_in = true;
2896 rxnfc = compat_alloc_user_space(buf_size);
2897 break;
2898 }
2899
2900 if (copy_from_user(&ifr.ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2901 return -EFAULT;
2902
2903 ifr.ifr_data = convert_in ? rxnfc : (void __user *)compat_rxnfc;
2904
2905 if (convert_in) {
2906 /* We expect there to be holes between fs.m_ext and
2907 * fs.ring_cookie and at the end of fs, but nowhere else.
2908 */
2909 BUILD_BUG_ON(offsetof(struct