Loading...
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
93#include <linux/uaccess.h>
94#include <asm/unistd.h>
95
96#include <net/compat.h>
97#include <net/wext.h>
98#include <net/cls_cgroup.h>
99
100#include <net/sock.h>
101#include <linux/netfilter.h>
102
103#include <linux/if_tun.h>
104#include <linux/ipv6_route.h>
105#include <linux/route.h>
106#include <linux/sockios.h>
107#include <linux/atalk.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 unsigned int 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 * Statistics counters of the socket lists
167 */
168
169static DEFINE_PER_CPU(int, sockets_in_use);
170
171/*
172 * Support routines.
173 * Move socket addresses back and forth across the kernel/user
174 * divide and look after the messy bits.
175 */
176
177/**
178 * move_addr_to_kernel - copy a socket address into kernel space
179 * @uaddr: Address in user space
180 * @kaddr: Address in kernel space
181 * @ulen: Length in user space
182 *
183 * The address is copied into kernel space. If the provided address is
184 * too long an error code of -EINVAL is returned. If the copy gives
185 * invalid addresses -EFAULT is returned. On a success 0 is returned.
186 */
187
188int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
189{
190 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
191 return -EINVAL;
192 if (ulen == 0)
193 return 0;
194 if (copy_from_user(kaddr, uaddr, ulen))
195 return -EFAULT;
196 return audit_sockaddr(ulen, kaddr);
197}
198
199/**
200 * move_addr_to_user - copy an address to user space
201 * @kaddr: kernel space address
202 * @klen: length of address in kernel
203 * @uaddr: user space address
204 * @ulen: pointer to user length field
205 *
206 * The value pointed to by ulen on entry is the buffer length available.
207 * This is overwritten with the buffer space used. -EINVAL is returned
208 * if an overlong buffer is specified or a negative buffer size. -EFAULT
209 * is returned if either the buffer or the length field are not
210 * accessible.
211 * After copying the data up to the limit the user specifies, the true
212 * length of the data is written over the length limit the user
213 * specified. Zero is returned for a success.
214 */
215
216static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
217 void __user *uaddr, int __user *ulen)
218{
219 int err;
220 int len;
221
222 BUG_ON(klen > sizeof(struct sockaddr_storage));
223 err = get_user(len, ulen);
224 if (err)
225 return err;
226 if (len > klen)
227 len = klen;
228 if (len < 0)
229 return -EINVAL;
230 if (len) {
231 if (audit_sockaddr(klen, kaddr))
232 return -ENOMEM;
233 if (copy_to_user(uaddr, kaddr, len))
234 return -EFAULT;
235 }
236 /*
237 * "fromlen shall refer to the value before truncation.."
238 * 1003.1g
239 */
240 return __put_user(klen, ulen);
241}
242
243static struct kmem_cache *sock_inode_cachep __read_mostly;
244
245static struct inode *sock_alloc_inode(struct super_block *sb)
246{
247 struct socket_alloc *ei;
248 struct socket_wq *wq;
249
250 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
251 if (!ei)
252 return NULL;
253 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
254 if (!wq) {
255 kmem_cache_free(sock_inode_cachep, ei);
256 return NULL;
257 }
258 init_waitqueue_head(&wq->wait);
259 wq->fasync_list = NULL;
260 wq->flags = 0;
261 RCU_INIT_POINTER(ei->socket.wq, wq);
262
263 ei->socket.state = SS_UNCONNECTED;
264 ei->socket.flags = 0;
265 ei->socket.ops = NULL;
266 ei->socket.sk = NULL;
267 ei->socket.file = NULL;
268
269 return &ei->vfs_inode;
270}
271
272static void sock_destroy_inode(struct inode *inode)
273{
274 struct socket_alloc *ei;
275 struct socket_wq *wq;
276
277 ei = container_of(inode, struct socket_alloc, vfs_inode);
278 wq = rcu_dereference_protected(ei->socket.wq, 1);
279 kfree_rcu(wq, rcu);
280 kmem_cache_free(sock_inode_cachep, ei);
281}
282
283static void init_once(void *foo)
284{
285 struct socket_alloc *ei = (struct socket_alloc *)foo;
286
287 inode_init_once(&ei->vfs_inode);
288}
289
290static int init_inodecache(void)
291{
292 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
293 sizeof(struct socket_alloc),
294 0,
295 (SLAB_HWCACHE_ALIGN |
296 SLAB_RECLAIM_ACCOUNT |
297 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
298 init_once);
299 if (sock_inode_cachep == NULL)
300 return -ENOMEM;
301 return 0;
302}
303
304static const struct super_operations sockfs_ops = {
305 .alloc_inode = sock_alloc_inode,
306 .destroy_inode = sock_destroy_inode,
307 .statfs = simple_statfs,
308};
309
310/*
311 * sockfs_dname() is called from d_path().
312 */
313static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
314{
315 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
316 d_inode(dentry)->i_ino);
317}
318
319static const struct dentry_operations sockfs_dentry_operations = {
320 .d_dname = sockfs_dname,
321};
322
323static int sockfs_xattr_get(const struct xattr_handler *handler,
324 struct dentry *dentry, struct inode *inode,
325 const char *suffix, void *value, size_t size)
326{
327 if (value) {
328 if (dentry->d_name.len + 1 > size)
329 return -ERANGE;
330 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
331 }
332 return dentry->d_name.len + 1;
333}
334
335#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
336#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
337#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
338
339static const struct xattr_handler sockfs_xattr_handler = {
340 .name = XATTR_NAME_SOCKPROTONAME,
341 .get = sockfs_xattr_get,
342};
343
344static int sockfs_security_xattr_set(const struct xattr_handler *handler,
345 struct dentry *dentry, struct inode *inode,
346 const char *suffix, const void *value,
347 size_t size, int flags)
348{
349 /* Handled by LSM. */
350 return -EAGAIN;
351}
352
353static const struct xattr_handler sockfs_security_xattr_handler = {
354 .prefix = XATTR_SECURITY_PREFIX,
355 .set = sockfs_security_xattr_set,
356};
357
358static const struct xattr_handler *sockfs_xattr_handlers[] = {
359 &sockfs_xattr_handler,
360 &sockfs_security_xattr_handler,
361 NULL
362};
363
364static struct dentry *sockfs_mount(struct file_system_type *fs_type,
365 int flags, const char *dev_name, void *data)
366{
367 return mount_pseudo_xattr(fs_type, "socket:", &sockfs_ops,
368 sockfs_xattr_handlers,
369 &sockfs_dentry_operations, SOCKFS_MAGIC);
370}
371
372static struct vfsmount *sock_mnt __read_mostly;
373
374static struct file_system_type sock_fs_type = {
375 .name = "sockfs",
376 .mount = sockfs_mount,
377 .kill_sb = kill_anon_super,
378};
379
380/*
381 * Obtains the first available file descriptor and sets it up for use.
382 *
383 * These functions create file structures and maps them to fd space
384 * of the current process. On success it returns file descriptor
385 * and file struct implicitly stored in sock->file.
386 * Note that another thread may close file descriptor before we return
387 * from this function. We use the fact that now we do not refer
388 * to socket after mapping. If one day we will need it, this
389 * function will increment ref. count on file by 1.
390 *
391 * In any case returned fd MAY BE not valid!
392 * This race condition is unavoidable
393 * with shared fd spaces, we cannot solve it inside kernel,
394 * but we take care of internal coherence yet.
395 */
396
397struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
398{
399 struct qstr name = { .name = "" };
400 struct path path;
401 struct file *file;
402
403 if (dname) {
404 name.name = dname;
405 name.len = strlen(name.name);
406 } else if (sock->sk) {
407 name.name = sock->sk->sk_prot_creator->name;
408 name.len = strlen(name.name);
409 }
410 path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
411 if (unlikely(!path.dentry))
412 return ERR_PTR(-ENOMEM);
413 path.mnt = mntget(sock_mnt);
414
415 d_instantiate(path.dentry, SOCK_INODE(sock));
416
417 file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
418 &socket_file_ops);
419 if (IS_ERR(file)) {
420 /* drop dentry, keep inode */
421 ihold(d_inode(path.dentry));
422 path_put(&path);
423 return file;
424 }
425
426 sock->file = file;
427 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
428 file->private_data = sock;
429 return file;
430}
431EXPORT_SYMBOL(sock_alloc_file);
432
433static int sock_map_fd(struct socket *sock, int flags)
434{
435 struct file *newfile;
436 int fd = get_unused_fd_flags(flags);
437 if (unlikely(fd < 0))
438 return fd;
439
440 newfile = sock_alloc_file(sock, flags, NULL);
441 if (likely(!IS_ERR(newfile))) {
442 fd_install(fd, newfile);
443 return fd;
444 }
445
446 put_unused_fd(fd);
447 return PTR_ERR(newfile);
448}
449
450struct socket *sock_from_file(struct file *file, int *err)
451{
452 if (file->f_op == &socket_file_ops)
453 return file->private_data; /* set in sock_map_fd */
454
455 *err = -ENOTSOCK;
456 return NULL;
457}
458EXPORT_SYMBOL(sock_from_file);
459
460/**
461 * sockfd_lookup - Go from a file number to its socket slot
462 * @fd: file handle
463 * @err: pointer to an error code return
464 *
465 * The file handle passed in is locked and the socket it is bound
466 * too is returned. If an error occurs the err pointer is overwritten
467 * with a negative errno code and NULL is returned. The function checks
468 * for both invalid handles and passing a handle which is not a socket.
469 *
470 * On a success the socket object pointer is returned.
471 */
472
473struct socket *sockfd_lookup(int fd, int *err)
474{
475 struct file *file;
476 struct socket *sock;
477
478 file = fget(fd);
479 if (!file) {
480 *err = -EBADF;
481 return NULL;
482 }
483
484 sock = sock_from_file(file, err);
485 if (!sock)
486 fput(file);
487 return sock;
488}
489EXPORT_SYMBOL(sockfd_lookup);
490
491static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
492{
493 struct fd f = fdget(fd);
494 struct socket *sock;
495
496 *err = -EBADF;
497 if (f.file) {
498 sock = sock_from_file(f.file, err);
499 if (likely(sock)) {
500 *fput_needed = f.flags;
501 return sock;
502 }
503 fdput(f);
504 }
505 return NULL;
506}
507
508static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
509 size_t size)
510{
511 ssize_t len;
512 ssize_t used = 0;
513
514 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
515 if (len < 0)
516 return len;
517 used += len;
518 if (buffer) {
519 if (size < used)
520 return -ERANGE;
521 buffer += len;
522 }
523
524 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
525 used += len;
526 if (buffer) {
527 if (size < used)
528 return -ERANGE;
529 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
530 buffer += len;
531 }
532
533 return used;
534}
535
536static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
537{
538 int err = simple_setattr(dentry, iattr);
539
540 if (!err && (iattr->ia_valid & ATTR_UID)) {
541 struct socket *sock = SOCKET_I(d_inode(dentry));
542
543 sock->sk->sk_uid = iattr->ia_uid;
544 }
545
546 return err;
547}
548
549static const struct inode_operations sockfs_inode_ops = {
550 .listxattr = sockfs_listxattr,
551 .setattr = sockfs_setattr,
552};
553
554/**
555 * sock_alloc - allocate a socket
556 *
557 * Allocate a new inode and socket object. The two are bound together
558 * and initialised. The socket is then returned. If we are out of inodes
559 * NULL is returned.
560 */
561
562struct socket *sock_alloc(void)
563{
564 struct inode *inode;
565 struct socket *sock;
566
567 inode = new_inode_pseudo(sock_mnt->mnt_sb);
568 if (!inode)
569 return NULL;
570
571 sock = SOCKET_I(inode);
572
573 kmemcheck_annotate_bitfield(sock, type);
574 inode->i_ino = get_next_ino();
575 inode->i_mode = S_IFSOCK | S_IRWXUGO;
576 inode->i_uid = current_fsuid();
577 inode->i_gid = current_fsgid();
578 inode->i_op = &sockfs_inode_ops;
579
580 this_cpu_add(sockets_in_use, 1);
581 return sock;
582}
583EXPORT_SYMBOL(sock_alloc);
584
585/**
586 * sock_release - close a socket
587 * @sock: socket to close
588 *
589 * The socket is released from the protocol stack if it has a release
590 * callback, and the inode is then released if the socket is bound to
591 * an inode not a file.
592 */
593
594void sock_release(struct socket *sock)
595{
596 if (sock->ops) {
597 struct module *owner = sock->ops->owner;
598
599 sock->ops->release(sock);
600 sock->ops = NULL;
601 module_put(owner);
602 }
603
604 if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
605 pr_err("%s: fasync list not empty!\n", __func__);
606
607 this_cpu_sub(sockets_in_use, 1);
608 if (!sock->file) {
609 iput(SOCK_INODE(sock));
610 return;
611 }
612 sock->file = NULL;
613}
614EXPORT_SYMBOL(sock_release);
615
616void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
617{
618 u8 flags = *tx_flags;
619
620 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
621 flags |= SKBTX_HW_TSTAMP;
622
623 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
624 flags |= SKBTX_SW_TSTAMP;
625
626 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
627 flags |= SKBTX_SCHED_TSTAMP;
628
629 *tx_flags = flags;
630}
631EXPORT_SYMBOL(__sock_tx_timestamp);
632
633static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
634{
635 int ret = sock->ops->sendmsg(sock, msg, msg_data_left(msg));
636 BUG_ON(ret == -EIOCBQUEUED);
637 return ret;
638}
639
640int sock_sendmsg(struct socket *sock, struct msghdr *msg)
641{
642 int err = security_socket_sendmsg(sock, msg,
643 msg_data_left(msg));
644
645 return err ?: sock_sendmsg_nosec(sock, msg);
646}
647EXPORT_SYMBOL(sock_sendmsg);
648
649int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
650 struct kvec *vec, size_t num, size_t size)
651{
652 iov_iter_kvec(&msg->msg_iter, WRITE | ITER_KVEC, vec, num, size);
653 return sock_sendmsg(sock, msg);
654}
655EXPORT_SYMBOL(kernel_sendmsg);
656
657/*
658 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
659 */
660void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
661 struct sk_buff *skb)
662{
663 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
664 struct scm_timestamping tss;
665 int empty = 1;
666 struct skb_shared_hwtstamps *shhwtstamps =
667 skb_hwtstamps(skb);
668
669 /* Race occurred between timestamp enabling and packet
670 receiving. Fill in the current time for now. */
671 if (need_software_tstamp && skb->tstamp == 0)
672 __net_timestamp(skb);
673
674 if (need_software_tstamp) {
675 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
676 struct timeval tv;
677 skb_get_timestamp(skb, &tv);
678 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
679 sizeof(tv), &tv);
680 } else {
681 struct timespec ts;
682 skb_get_timestampns(skb, &ts);
683 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
684 sizeof(ts), &ts);
685 }
686 }
687
688 memset(&tss, 0, sizeof(tss));
689 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
690 ktime_to_timespec_cond(skb->tstamp, tss.ts + 0))
691 empty = 0;
692 if (shhwtstamps &&
693 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
694 ktime_to_timespec_cond(shhwtstamps->hwtstamp, tss.ts + 2))
695 empty = 0;
696 if (!empty) {
697 put_cmsg(msg, SOL_SOCKET,
698 SCM_TIMESTAMPING, sizeof(tss), &tss);
699
700 if (skb->len && (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS))
701 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
702 skb->len, skb->data);
703 }
704}
705EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
706
707void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
708 struct sk_buff *skb)
709{
710 int ack;
711
712 if (!sock_flag(sk, SOCK_WIFI_STATUS))
713 return;
714 if (!skb->wifi_acked_valid)
715 return;
716
717 ack = skb->wifi_acked;
718
719 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
720}
721EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
722
723static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
724 struct sk_buff *skb)
725{
726 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
727 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
728 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
729}
730
731void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
732 struct sk_buff *skb)
733{
734 sock_recv_timestamp(msg, sk, skb);
735 sock_recv_drops(msg, sk, skb);
736}
737EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
738
739static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
740 int flags)
741{
742 return sock->ops->recvmsg(sock, msg, msg_data_left(msg), flags);
743}
744
745int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
746{
747 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
748
749 return err ?: sock_recvmsg_nosec(sock, msg, flags);
750}
751EXPORT_SYMBOL(sock_recvmsg);
752
753/**
754 * kernel_recvmsg - Receive a message from a socket (kernel space)
755 * @sock: The socket to receive the message from
756 * @msg: Received message
757 * @vec: Input s/g array for message data
758 * @num: Size of input s/g array
759 * @size: Number of bytes to read
760 * @flags: Message flags (MSG_DONTWAIT, etc...)
761 *
762 * On return the msg structure contains the scatter/gather array passed in the
763 * vec argument. The array is modified so that it consists of the unfilled
764 * portion of the original array.
765 *
766 * The returned value is the total number of bytes received, or an error.
767 */
768int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
769 struct kvec *vec, size_t num, size_t size, int flags)
770{
771 mm_segment_t oldfs = get_fs();
772 int result;
773
774 iov_iter_kvec(&msg->msg_iter, READ | ITER_KVEC, vec, num, size);
775 set_fs(KERNEL_DS);
776 result = sock_recvmsg(sock, msg, flags);
777 set_fs(oldfs);
778 return result;
779}
780EXPORT_SYMBOL(kernel_recvmsg);
781
782static ssize_t sock_sendpage(struct file *file, struct page *page,
783 int offset, size_t size, loff_t *ppos, int more)
784{
785 struct socket *sock;
786 int flags;
787
788 sock = file->private_data;
789
790 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
791 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
792 flags |= more;
793
794 return kernel_sendpage(sock, page, offset, size, flags);
795}
796
797static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
798 struct pipe_inode_info *pipe, size_t len,
799 unsigned int flags)
800{
801 struct socket *sock = file->private_data;
802
803 if (unlikely(!sock->ops->splice_read))
804 return -EINVAL;
805
806 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
807}
808
809static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
810{
811 struct file *file = iocb->ki_filp;
812 struct socket *sock = file->private_data;
813 struct msghdr msg = {.msg_iter = *to,
814 .msg_iocb = iocb};
815 ssize_t res;
816
817 if (file->f_flags & O_NONBLOCK)
818 msg.msg_flags = MSG_DONTWAIT;
819
820 if (iocb->ki_pos != 0)
821 return -ESPIPE;
822
823 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
824 return 0;
825
826 res = sock_recvmsg(sock, &msg, msg.msg_flags);
827 *to = msg.msg_iter;
828 return res;
829}
830
831static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
832{
833 struct file *file = iocb->ki_filp;
834 struct socket *sock = file->private_data;
835 struct msghdr msg = {.msg_iter = *from,
836 .msg_iocb = iocb};
837 ssize_t res;
838
839 if (iocb->ki_pos != 0)
840 return -ESPIPE;
841
842 if (file->f_flags & O_NONBLOCK)
843 msg.msg_flags = MSG_DONTWAIT;
844
845 if (sock->type == SOCK_SEQPACKET)
846 msg.msg_flags |= MSG_EOR;
847
848 res = sock_sendmsg(sock, &msg);
849 *from = msg.msg_iter;
850 return res;
851}
852
853/*
854 * Atomic setting of ioctl hooks to avoid race
855 * with module unload.
856 */
857
858static DEFINE_MUTEX(br_ioctl_mutex);
859static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
860
861void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
862{
863 mutex_lock(&br_ioctl_mutex);
864 br_ioctl_hook = hook;
865 mutex_unlock(&br_ioctl_mutex);
866}
867EXPORT_SYMBOL(brioctl_set);
868
869static DEFINE_MUTEX(vlan_ioctl_mutex);
870static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
871
872void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
873{
874 mutex_lock(&vlan_ioctl_mutex);
875 vlan_ioctl_hook = hook;
876 mutex_unlock(&vlan_ioctl_mutex);
877}
878EXPORT_SYMBOL(vlan_ioctl_set);
879
880static DEFINE_MUTEX(dlci_ioctl_mutex);
881static int (*dlci_ioctl_hook) (unsigned int, void __user *);
882
883void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
884{
885 mutex_lock(&dlci_ioctl_mutex);
886 dlci_ioctl_hook = hook;
887 mutex_unlock(&dlci_ioctl_mutex);
888}
889EXPORT_SYMBOL(dlci_ioctl_set);
890
891static long sock_do_ioctl(struct net *net, struct socket *sock,
892 unsigned int cmd, unsigned long arg)
893{
894 int err;
895 void __user *argp = (void __user *)arg;
896
897 err = sock->ops->ioctl(sock, cmd, arg);
898
899 /*
900 * If this ioctl is unknown try to hand it down
901 * to the NIC driver.
902 */
903 if (err == -ENOIOCTLCMD)
904 err = dev_ioctl(net, cmd, argp);
905
906 return err;
907}
908
909/*
910 * With an ioctl, arg may well be a user mode pointer, but we don't know
911 * what to do with it - that's up to the protocol still.
912 */
913
914static struct ns_common *get_net_ns(struct ns_common *ns)
915{
916 return &get_net(container_of(ns, struct net, ns))->ns;
917}
918
919static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
920{
921 struct socket *sock;
922 struct sock *sk;
923 void __user *argp = (void __user *)arg;
924 int pid, err;
925 struct net *net;
926
927 sock = file->private_data;
928 sk = sock->sk;
929 net = sock_net(sk);
930 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
931 err = dev_ioctl(net, cmd, argp);
932 } else
933#ifdef CONFIG_WEXT_CORE
934 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
935 err = dev_ioctl(net, cmd, argp);
936 } else
937#endif
938 switch (cmd) {
939 case FIOSETOWN:
940 case SIOCSPGRP:
941 err = -EFAULT;
942 if (get_user(pid, (int __user *)argp))
943 break;
944 f_setown(sock->file, pid, 1);
945 err = 0;
946 break;
947 case FIOGETOWN:
948 case SIOCGPGRP:
949 err = put_user(f_getown(sock->file),
950 (int __user *)argp);
951 break;
952 case SIOCGIFBR:
953 case SIOCSIFBR:
954 case SIOCBRADDBR:
955 case SIOCBRDELBR:
956 err = -ENOPKG;
957 if (!br_ioctl_hook)
958 request_module("bridge");
959
960 mutex_lock(&br_ioctl_mutex);
961 if (br_ioctl_hook)
962 err = br_ioctl_hook(net, cmd, argp);
963 mutex_unlock(&br_ioctl_mutex);
964 break;
965 case SIOCGIFVLAN:
966 case SIOCSIFVLAN:
967 err = -ENOPKG;
968 if (!vlan_ioctl_hook)
969 request_module("8021q");
970
971 mutex_lock(&vlan_ioctl_mutex);
972 if (vlan_ioctl_hook)
973 err = vlan_ioctl_hook(net, argp);
974 mutex_unlock(&vlan_ioctl_mutex);
975 break;
976 case SIOCADDDLCI:
977 case SIOCDELDLCI:
978 err = -ENOPKG;
979 if (!dlci_ioctl_hook)
980 request_module("dlci");
981
982 mutex_lock(&dlci_ioctl_mutex);
983 if (dlci_ioctl_hook)
984 err = dlci_ioctl_hook(cmd, argp);
985 mutex_unlock(&dlci_ioctl_mutex);
986 break;
987 case SIOCGSKNS:
988 err = -EPERM;
989 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
990 break;
991
992 err = open_related_ns(&net->ns, get_net_ns);
993 break;
994 default:
995 err = sock_do_ioctl(net, sock, cmd, arg);
996 break;
997 }
998 return err;
999}
1000
1001int sock_create_lite(int family, int type, int protocol, struct socket **res)
1002{
1003 int err;
1004 struct socket *sock = NULL;
1005
1006 err = security_socket_create(family, type, protocol, 1);
1007 if (err)
1008 goto out;
1009
1010 sock = sock_alloc();
1011 if (!sock) {
1012 err = -ENOMEM;
1013 goto out;
1014 }
1015
1016 sock->type = type;
1017 err = security_socket_post_create(sock, family, type, protocol, 1);
1018 if (err)
1019 goto out_release;
1020
1021out:
1022 *res = sock;
1023 return err;
1024out_release:
1025 sock_release(sock);
1026 sock = NULL;
1027 goto out;
1028}
1029EXPORT_SYMBOL(sock_create_lite);
1030
1031/* No kernel lock held - perfect */
1032static unsigned int sock_poll(struct file *file, poll_table *wait)
1033{
1034 unsigned int busy_flag = 0;
1035 struct socket *sock;
1036
1037 /*
1038 * We can't return errors to poll, so it's either yes or no.
1039 */
1040 sock = file->private_data;
1041
1042 if (sk_can_busy_loop(sock->sk)) {
1043 /* this socket can poll_ll so tell the system call */
1044 busy_flag = POLL_BUSY_LOOP;
1045
1046 /* once, only if requested by syscall */
1047 if (wait && (wait->_key & POLL_BUSY_LOOP))
1048 sk_busy_loop(sock->sk, 1);
1049 }
1050
1051 return busy_flag | sock->ops->poll(file, sock, wait);
1052}
1053
1054static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1055{
1056 struct socket *sock = file->private_data;
1057
1058 return sock->ops->mmap(file, sock, vma);
1059}
1060
1061static int sock_close(struct inode *inode, struct file *filp)
1062{
1063 sock_release(SOCKET_I(inode));
1064 return 0;
1065}
1066
1067/*
1068 * Update the socket async list
1069 *
1070 * Fasync_list locking strategy.
1071 *
1072 * 1. fasync_list is modified only under process context socket lock
1073 * i.e. under semaphore.
1074 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1075 * or under socket lock
1076 */
1077
1078static int sock_fasync(int fd, struct file *filp, int on)
1079{
1080 struct socket *sock = filp->private_data;
1081 struct sock *sk = sock->sk;
1082 struct socket_wq *wq;
1083
1084 if (sk == NULL)
1085 return -EINVAL;
1086
1087 lock_sock(sk);
1088 wq = rcu_dereference_protected(sock->wq, lockdep_sock_is_held(sk));
1089 fasync_helper(fd, filp, on, &wq->fasync_list);
1090
1091 if (!wq->fasync_list)
1092 sock_reset_flag(sk, SOCK_FASYNC);
1093 else
1094 sock_set_flag(sk, SOCK_FASYNC);
1095
1096 release_sock(sk);
1097 return 0;
1098}
1099
1100/* This function may be called only under rcu_lock */
1101
1102int sock_wake_async(struct socket_wq *wq, int how, int band)
1103{
1104 if (!wq || !wq->fasync_list)
1105 return -1;
1106
1107 switch (how) {
1108 case SOCK_WAKE_WAITD:
1109 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1110 break;
1111 goto call_kill;
1112 case SOCK_WAKE_SPACE:
1113 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1114 break;
1115 /* fall through */
1116 case SOCK_WAKE_IO:
1117call_kill:
1118 kill_fasync(&wq->fasync_list, SIGIO, band);
1119 break;
1120 case SOCK_WAKE_URG:
1121 kill_fasync(&wq->fasync_list, SIGURG, band);
1122 }
1123
1124 return 0;
1125}
1126EXPORT_SYMBOL(sock_wake_async);
1127
1128int __sock_create(struct net *net, int family, int type, int protocol,
1129 struct socket **res, int kern)
1130{
1131 int err;
1132 struct socket *sock;
1133 const struct net_proto_family *pf;
1134
1135 /*
1136 * Check protocol is in range
1137 */
1138 if (family < 0 || family >= NPROTO)
1139 return -EAFNOSUPPORT;
1140 if (type < 0 || type >= SOCK_MAX)
1141 return -EINVAL;
1142
1143 /* Compatibility.
1144
1145 This uglymoron is moved from INET layer to here to avoid
1146 deadlock in module load.
1147 */
1148 if (family == PF_INET && type == SOCK_PACKET) {
1149 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1150 current->comm);
1151 family = PF_PACKET;
1152 }
1153
1154 err = security_socket_create(family, type, protocol, kern);
1155 if (err)
1156 return err;
1157
1158 /*
1159 * Allocate the socket and allow the family to set things up. if
1160 * the protocol is 0, the family is instructed to select an appropriate
1161 * default.
1162 */
1163 sock = sock_alloc();
1164 if (!sock) {
1165 net_warn_ratelimited("socket: no more sockets\n");
1166 return -ENFILE; /* Not exactly a match, but its the
1167 closest posix thing */
1168 }
1169
1170 sock->type = type;
1171
1172#ifdef CONFIG_MODULES
1173 /* Attempt to load a protocol module if the find failed.
1174 *
1175 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1176 * requested real, full-featured networking support upon configuration.
1177 * Otherwise module support will break!
1178 */
1179 if (rcu_access_pointer(net_families[family]) == NULL)
1180 request_module("net-pf-%d", family);
1181#endif
1182
1183 rcu_read_lock();
1184 pf = rcu_dereference(net_families[family]);
1185 err = -EAFNOSUPPORT;
1186 if (!pf)
1187 goto out_release;
1188
1189 /*
1190 * We will call the ->create function, that possibly is in a loadable
1191 * module, so we have to bump that loadable module refcnt first.
1192 */
1193 if (!try_module_get(pf->owner))
1194 goto out_release;
1195
1196 /* Now protected by module ref count */
1197 rcu_read_unlock();
1198
1199 err = pf->create(net, sock, protocol, kern);
1200 if (err < 0)
1201 goto out_module_put;
1202
1203 /*
1204 * Now to bump the refcnt of the [loadable] module that owns this
1205 * socket at sock_release time we decrement its refcnt.
1206 */
1207 if (!try_module_get(sock->ops->owner))
1208 goto out_module_busy;
1209
1210 /*
1211 * Now that we're done with the ->create function, the [loadable]
1212 * module can have its refcnt decremented
1213 */
1214 module_put(pf->owner);
1215 err = security_socket_post_create(sock, family, type, protocol, kern);
1216 if (err)
1217 goto out_sock_release;
1218 *res = sock;
1219
1220 return 0;
1221
1222out_module_busy:
1223 err = -EAFNOSUPPORT;
1224out_module_put:
1225 sock->ops = NULL;
1226 module_put(pf->owner);
1227out_sock_release:
1228 sock_release(sock);
1229 return err;
1230
1231out_release:
1232 rcu_read_unlock();
1233 goto out_sock_release;
1234}
1235EXPORT_SYMBOL(__sock_create);
1236
1237int sock_create(int family, int type, int protocol, struct socket **res)
1238{
1239 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1240}
1241EXPORT_SYMBOL(sock_create);
1242
1243int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1244{
1245 return __sock_create(net, family, type, protocol, res, 1);
1246}
1247EXPORT_SYMBOL(sock_create_kern);
1248
1249SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1250{
1251 int retval;
1252 struct socket *sock;
1253 int flags;
1254
1255 /* Check the SOCK_* constants for consistency. */
1256 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1257 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1258 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1259 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1260
1261 flags = type & ~SOCK_TYPE_MASK;
1262 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1263 return -EINVAL;
1264 type &= SOCK_TYPE_MASK;
1265
1266 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1267 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1268
1269 retval = sock_create(family, type, protocol, &sock);
1270 if (retval < 0)
1271 goto out;
1272
1273 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1274 if (retval < 0)
1275 goto out_release;
1276
1277out:
1278 /* It may be already another descriptor 8) Not kernel problem. */
1279 return retval;
1280
1281out_release:
1282 sock_release(sock);
1283 return retval;
1284}
1285
1286/*
1287 * Create a pair of connected sockets.
1288 */
1289
1290SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1291 int __user *, usockvec)
1292{
1293 struct socket *sock1, *sock2;
1294 int fd1, fd2, err;
1295 struct file *newfile1, *newfile2;
1296 int flags;
1297
1298 flags = type & ~SOCK_TYPE_MASK;
1299 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1300 return -EINVAL;
1301 type &= SOCK_TYPE_MASK;
1302
1303 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1304 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1305
1306 /*
1307 * Obtain the first socket and check if the underlying protocol
1308 * supports the socketpair call.
1309 */
1310
1311 err = sock_create(family, type, protocol, &sock1);
1312 if (err < 0)
1313 goto out;
1314
1315 err = sock_create(family, type, protocol, &sock2);
1316 if (err < 0)
1317 goto out_release_1;
1318
1319 err = sock1->ops->socketpair(sock1, sock2);
1320 if (err < 0)
1321 goto out_release_both;
1322
1323 fd1 = get_unused_fd_flags(flags);
1324 if (unlikely(fd1 < 0)) {
1325 err = fd1;
1326 goto out_release_both;
1327 }
1328
1329 fd2 = get_unused_fd_flags(flags);
1330 if (unlikely(fd2 < 0)) {
1331 err = fd2;
1332 goto out_put_unused_1;
1333 }
1334
1335 newfile1 = sock_alloc_file(sock1, flags, NULL);
1336 if (IS_ERR(newfile1)) {
1337 err = PTR_ERR(newfile1);
1338 goto out_put_unused_both;
1339 }
1340
1341 newfile2 = sock_alloc_file(sock2, flags, NULL);
1342 if (IS_ERR(newfile2)) {
1343 err = PTR_ERR(newfile2);
1344 goto out_fput_1;
1345 }
1346
1347 err = put_user(fd1, &usockvec[0]);
1348 if (err)
1349 goto out_fput_both;
1350
1351 err = put_user(fd2, &usockvec[1]);
1352 if (err)
1353 goto out_fput_both;
1354
1355 audit_fd_pair(fd1, fd2);
1356
1357 fd_install(fd1, newfile1);
1358 fd_install(fd2, newfile2);
1359 /* fd1 and fd2 may be already another descriptors.
1360 * Not kernel problem.
1361 */
1362
1363 return 0;
1364
1365out_fput_both:
1366 fput(newfile2);
1367 fput(newfile1);
1368 put_unused_fd(fd2);
1369 put_unused_fd(fd1);
1370 goto out;
1371
1372out_fput_1:
1373 fput(newfile1);
1374 put_unused_fd(fd2);
1375 put_unused_fd(fd1);
1376 sock_release(sock2);
1377 goto out;
1378
1379out_put_unused_both:
1380 put_unused_fd(fd2);
1381out_put_unused_1:
1382 put_unused_fd(fd1);
1383out_release_both:
1384 sock_release(sock2);
1385out_release_1:
1386 sock_release(sock1);
1387out:
1388 return err;
1389}
1390
1391/*
1392 * Bind a name to a socket. Nothing much to do here since it's
1393 * the protocol's responsibility to handle the local address.
1394 *
1395 * We move the socket address to kernel space before we call
1396 * the protocol layer (having also checked the address is ok).
1397 */
1398
1399SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1400{
1401 struct socket *sock;
1402 struct sockaddr_storage address;
1403 int err, fput_needed;
1404
1405 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1406 if (sock) {
1407 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1408 if (err >= 0) {
1409 err = security_socket_bind(sock,
1410 (struct sockaddr *)&address,
1411 addrlen);
1412 if (!err)
1413 err = sock->ops->bind(sock,
1414 (struct sockaddr *)
1415 &address, addrlen);
1416 }
1417 fput_light(sock->file, fput_needed);
1418 }
1419 return err;
1420}
1421
1422/*
1423 * Perform a listen. Basically, we allow the protocol to do anything
1424 * necessary for a listen, and if that works, we mark the socket as
1425 * ready for listening.
1426 */
1427
1428SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1429{
1430 struct socket *sock;
1431 int err, fput_needed;
1432 int somaxconn;
1433
1434 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1435 if (sock) {
1436 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1437 if ((unsigned int)backlog > somaxconn)
1438 backlog = somaxconn;
1439
1440 err = security_socket_listen(sock, backlog);
1441 if (!err)
1442 err = sock->ops->listen(sock, backlog);
1443
1444 fput_light(sock->file, fput_needed);
1445 }
1446 return err;
1447}
1448
1449/*
1450 * For accept, we attempt to create a new socket, set up the link
1451 * with the client, wake up the client, then return the new
1452 * connected fd. We collect the address of the connector in kernel
1453 * space and move it to user at the very end. This is unclean because
1454 * we open the socket then return an error.
1455 *
1456 * 1003.1g adds the ability to recvmsg() to query connection pending
1457 * status to recvmsg. We need to add that support in a way thats
1458 * clean when we restucture accept also.
1459 */
1460
1461SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1462 int __user *, upeer_addrlen, int, flags)
1463{
1464 struct socket *sock, *newsock;
1465 struct file *newfile;
1466 int err, len, newfd, fput_needed;
1467 struct sockaddr_storage address;
1468
1469 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1470 return -EINVAL;
1471
1472 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1473 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1474
1475 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1476 if (!sock)
1477 goto out;
1478
1479 err = -ENFILE;
1480 newsock = sock_alloc();
1481 if (!newsock)
1482 goto out_put;
1483
1484 newsock->type = sock->type;
1485 newsock->ops = sock->ops;
1486
1487 /*
1488 * We don't need try_module_get here, as the listening socket (sock)
1489 * has the protocol module (sock->ops->owner) held.
1490 */
1491 __module_get(newsock->ops->owner);
1492
1493 newfd = get_unused_fd_flags(flags);
1494 if (unlikely(newfd < 0)) {
1495 err = newfd;
1496 sock_release(newsock);
1497 goto out_put;
1498 }
1499 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1500 if (IS_ERR(newfile)) {
1501 err = PTR_ERR(newfile);
1502 put_unused_fd(newfd);
1503 sock_release(newsock);
1504 goto out_put;
1505 }
1506
1507 err = security_socket_accept(sock, newsock);
1508 if (err)
1509 goto out_fd;
1510
1511 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1512 if (err < 0)
1513 goto out_fd;
1514
1515 if (upeer_sockaddr) {
1516 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1517 &len, 2) < 0) {
1518 err = -ECONNABORTED;
1519 goto out_fd;
1520 }
1521 err = move_addr_to_user(&address,
1522 len, upeer_sockaddr, upeer_addrlen);
1523 if (err < 0)
1524 goto out_fd;
1525 }
1526
1527 /* File flags are not inherited via accept() unlike another OSes. */
1528
1529 fd_install(newfd, newfile);
1530 err = newfd;
1531
1532out_put:
1533 fput_light(sock->file, fput_needed);
1534out:
1535 return err;
1536out_fd:
1537 fput(newfile);
1538 put_unused_fd(newfd);
1539 goto out_put;
1540}
1541
1542SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1543 int __user *, upeer_addrlen)
1544{
1545 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1546}
1547
1548/*
1549 * Attempt to connect to a socket with the server address. The address
1550 * is in user space so we verify it is OK and move it to kernel space.
1551 *
1552 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1553 * break bindings
1554 *
1555 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1556 * other SEQPACKET protocols that take time to connect() as it doesn't
1557 * include the -EINPROGRESS status for such sockets.
1558 */
1559
1560SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1561 int, addrlen)
1562{
1563 struct socket *sock;
1564 struct sockaddr_storage address;
1565 int err, fput_needed;
1566
1567 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1568 if (!sock)
1569 goto out;
1570 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1571 if (err < 0)
1572 goto out_put;
1573
1574 err =
1575 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1576 if (err)
1577 goto out_put;
1578
1579 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1580 sock->file->f_flags);
1581out_put:
1582 fput_light(sock->file, fput_needed);
1583out:
1584 return err;
1585}
1586
1587/*
1588 * Get the local address ('name') of a socket object. Move the obtained
1589 * name to user space.
1590 */
1591
1592SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1593 int __user *, usockaddr_len)
1594{
1595 struct socket *sock;
1596 struct sockaddr_storage address;
1597 int len, err, fput_needed;
1598
1599 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1600 if (!sock)
1601 goto out;
1602
1603 err = security_socket_getsockname(sock);
1604 if (err)
1605 goto out_put;
1606
1607 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1608 if (err)
1609 goto out_put;
1610 err = move_addr_to_user(&address, len, usockaddr, usockaddr_len);
1611
1612out_put:
1613 fput_light(sock->file, fput_needed);
1614out:
1615 return err;
1616}
1617
1618/*
1619 * Get the remote address ('name') of a socket object. Move the obtained
1620 * name to user space.
1621 */
1622
1623SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1624 int __user *, usockaddr_len)
1625{
1626 struct socket *sock;
1627 struct sockaddr_storage address;
1628 int len, err, fput_needed;
1629
1630 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1631 if (sock != NULL) {
1632 err = security_socket_getpeername(sock);
1633 if (err) {
1634 fput_light(sock->file, fput_needed);
1635 return err;
1636 }
1637
1638 err =
1639 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1640 1);
1641 if (!err)
1642 err = move_addr_to_user(&address, len, usockaddr,
1643 usockaddr_len);
1644 fput_light(sock->file, fput_needed);
1645 }
1646 return err;
1647}
1648
1649/*
1650 * Send a datagram to a given address. We move the address into kernel
1651 * space and check the user space data area is readable before invoking
1652 * the protocol.
1653 */
1654
1655SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1656 unsigned int, flags, struct sockaddr __user *, addr,
1657 int, addr_len)
1658{
1659 struct socket *sock;
1660 struct sockaddr_storage address;
1661 int err;
1662 struct msghdr msg;
1663 struct iovec iov;
1664 int fput_needed;
1665
1666 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1667 if (unlikely(err))
1668 return err;
1669 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1670 if (!sock)
1671 goto out;
1672
1673 msg.msg_name = NULL;
1674 msg.msg_control = NULL;
1675 msg.msg_controllen = 0;
1676 msg.msg_namelen = 0;
1677 if (addr) {
1678 err = move_addr_to_kernel(addr, addr_len, &address);
1679 if (err < 0)
1680 goto out_put;
1681 msg.msg_name = (struct sockaddr *)&address;
1682 msg.msg_namelen = addr_len;
1683 }
1684 if (sock->file->f_flags & O_NONBLOCK)
1685 flags |= MSG_DONTWAIT;
1686 msg.msg_flags = flags;
1687 err = sock_sendmsg(sock, &msg);
1688
1689out_put:
1690 fput_light(sock->file, fput_needed);
1691out:
1692 return err;
1693}
1694
1695/*
1696 * Send a datagram down a socket.
1697 */
1698
1699SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1700 unsigned int, flags)
1701{
1702 return sys_sendto(fd, buff, len, flags, NULL, 0);
1703}
1704
1705/*
1706 * Receive a frame from the socket and optionally record the address of the
1707 * sender. We verify the buffers are writable and if needed move the
1708 * sender address from kernel to user space.
1709 */
1710
1711SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1712 unsigned int, flags, struct sockaddr __user *, addr,
1713 int __user *, addr_len)
1714{
1715 struct socket *sock;
1716 struct iovec iov;
1717 struct msghdr msg;
1718 struct sockaddr_storage address;
1719 int err, err2;
1720 int fput_needed;
1721
1722 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1723 if (unlikely(err))
1724 return err;
1725 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1726 if (!sock)
1727 goto out;
1728
1729 msg.msg_control = NULL;
1730 msg.msg_controllen = 0;
1731 /* Save some cycles and don't copy the address if not needed */
1732 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
1733 /* We assume all kernel code knows the size of sockaddr_storage */
1734 msg.msg_namelen = 0;
1735 msg.msg_iocb = NULL;
1736 if (sock->file->f_flags & O_NONBLOCK)
1737 flags |= MSG_DONTWAIT;
1738 err = sock_recvmsg(sock, &msg, flags);
1739
1740 if (err >= 0 && addr != NULL) {
1741 err2 = move_addr_to_user(&address,
1742 msg.msg_namelen, addr, addr_len);
1743 if (err2 < 0)
1744 err = err2;
1745 }
1746
1747 fput_light(sock->file, fput_needed);
1748out:
1749 return err;
1750}
1751
1752/*
1753 * Receive a datagram from a socket.
1754 */
1755
1756SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
1757 unsigned int, flags)
1758{
1759 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1760}
1761
1762/*
1763 * Set a socket option. Because we don't know the option lengths we have
1764 * to pass the user mode parameter for the protocols to sort out.
1765 */
1766
1767SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1768 char __user *, optval, int, optlen)
1769{
1770 int err, fput_needed;
1771 struct socket *sock;
1772
1773 if (optlen < 0)
1774 return -EINVAL;
1775
1776 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1777 if (sock != NULL) {
1778 err = security_socket_setsockopt(sock, level, optname);
1779 if (err)
1780 goto out_put;
1781
1782 if (level == SOL_SOCKET)
1783 err =
1784 sock_setsockopt(sock, level, optname, optval,
1785 optlen);
1786 else
1787 err =
1788 sock->ops->setsockopt(sock, level, optname, optval,
1789 optlen);
1790out_put:
1791 fput_light(sock->file, fput_needed);
1792 }
1793 return err;
1794}
1795
1796/*
1797 * Get a socket option. Because we don't know the option lengths we have
1798 * to pass a user mode parameter for the protocols to sort out.
1799 */
1800
1801SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1802 char __user *, optval, int __user *, optlen)
1803{
1804 int err, fput_needed;
1805 struct socket *sock;
1806
1807 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1808 if (sock != NULL) {
1809 err = security_socket_getsockopt(sock, level, optname);
1810 if (err)
1811 goto out_put;
1812
1813 if (level == SOL_SOCKET)
1814 err =
1815 sock_getsockopt(sock, level, optname, optval,
1816 optlen);
1817 else
1818 err =
1819 sock->ops->getsockopt(sock, level, optname, optval,
1820 optlen);
1821out_put:
1822 fput_light(sock->file, fput_needed);
1823 }
1824 return err;
1825}
1826
1827/*
1828 * Shutdown a socket.
1829 */
1830
1831SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1832{
1833 int err, fput_needed;
1834 struct socket *sock;
1835
1836 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1837 if (sock != NULL) {
1838 err = security_socket_shutdown(sock, how);
1839 if (!err)
1840 err = sock->ops->shutdown(sock, how);
1841 fput_light(sock->file, fput_needed);
1842 }
1843 return err;
1844}
1845
1846/* A couple of helpful macros for getting the address of the 32/64 bit
1847 * fields which are the same type (int / unsigned) on our platforms.
1848 */
1849#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1850#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1851#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1852
1853struct used_address {
1854 struct sockaddr_storage name;
1855 unsigned int name_len;
1856};
1857
1858static int copy_msghdr_from_user(struct msghdr *kmsg,
1859 struct user_msghdr __user *umsg,
1860 struct sockaddr __user **save_addr,
1861 struct iovec **iov)
1862{
1863 struct sockaddr __user *uaddr;
1864 struct iovec __user *uiov;
1865 size_t nr_segs;
1866 ssize_t err;
1867
1868 if (!access_ok(VERIFY_READ, umsg, sizeof(*umsg)) ||
1869 __get_user(uaddr, &umsg->msg_name) ||
1870 __get_user(kmsg->msg_namelen, &umsg->msg_namelen) ||
1871 __get_user(uiov, &umsg->msg_iov) ||
1872 __get_user(nr_segs, &umsg->msg_iovlen) ||
1873 __get_user(kmsg->msg_control, &umsg->msg_control) ||
1874 __get_user(kmsg->msg_controllen, &umsg->msg_controllen) ||
1875 __get_user(kmsg->msg_flags, &umsg->msg_flags))
1876 return -EFAULT;
1877
1878 if (!uaddr)
1879 kmsg->msg_namelen = 0;
1880
1881 if (kmsg->msg_namelen < 0)
1882 return -EINVAL;
1883
1884 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
1885 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
1886
1887 if (save_addr)
1888 *save_addr = uaddr;
1889
1890 if (uaddr && kmsg->msg_namelen) {
1891 if (!save_addr) {
1892 err = move_addr_to_kernel(uaddr, kmsg->msg_namelen,
1893 kmsg->msg_name);
1894 if (err < 0)
1895 return err;
1896 }
1897 } else {
1898 kmsg->msg_name = NULL;
1899 kmsg->msg_namelen = 0;
1900 }
1901
1902 if (nr_segs > UIO_MAXIOV)
1903 return -EMSGSIZE;
1904
1905 kmsg->msg_iocb = NULL;
1906
1907 return import_iovec(save_addr ? READ : WRITE, uiov, nr_segs,
1908 UIO_FASTIOV, iov, &kmsg->msg_iter);
1909}
1910
1911static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
1912 struct msghdr *msg_sys, unsigned int flags,
1913 struct used_address *used_address,
1914 unsigned int allowed_msghdr_flags)
1915{
1916 struct compat_msghdr __user *msg_compat =
1917 (struct compat_msghdr __user *)msg;
1918 struct sockaddr_storage address;
1919 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1920 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1921 __aligned(sizeof(__kernel_size_t));
1922 /* 20 is size of ipv6_pktinfo */
1923 unsigned char *ctl_buf = ctl;
1924 int ctl_len;
1925 ssize_t err;
1926
1927 msg_sys->msg_name = &address;
1928
1929 if (MSG_CMSG_COMPAT & flags)
1930 err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
1931 else
1932 err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
1933 if (err < 0)
1934 return err;
1935
1936 err = -ENOBUFS;
1937
1938 if (msg_sys->msg_controllen > INT_MAX)
1939 goto out_freeiov;
1940 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
1941 ctl_len = msg_sys->msg_controllen;
1942 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1943 err =
1944 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
1945 sizeof(ctl));
1946 if (err)
1947 goto out_freeiov;
1948 ctl_buf = msg_sys->msg_control;
1949 ctl_len = msg_sys->msg_controllen;
1950 } else if (ctl_len) {
1951 if (ctl_len > sizeof(ctl)) {
1952 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1953 if (ctl_buf == NULL)
1954 goto out_freeiov;
1955 }
1956 err = -EFAULT;
1957 /*
1958 * Careful! Before this, msg_sys->msg_control contains a user pointer.
1959 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1960 * checking falls down on this.
1961 */
1962 if (copy_from_user(ctl_buf,
1963 (void __user __force *)msg_sys->msg_control,
1964 ctl_len))
1965 goto out_freectl;
1966 msg_sys->msg_control = ctl_buf;
1967 }
1968 msg_sys->msg_flags = flags;
1969
1970 if (sock->file->f_flags & O_NONBLOCK)
1971 msg_sys->msg_flags |= MSG_DONTWAIT;
1972 /*
1973 * If this is sendmmsg() and current destination address is same as
1974 * previously succeeded address, omit asking LSM's decision.
1975 * used_address->name_len is initialized to UINT_MAX so that the first
1976 * destination address never matches.
1977 */
1978 if (used_address && msg_sys->msg_name &&
1979 used_address->name_len == msg_sys->msg_namelen &&
1980 !memcmp(&used_address->name, msg_sys->msg_name,
1981 used_address->name_len)) {
1982 err = sock_sendmsg_nosec(sock, msg_sys);
1983 goto out_freectl;
1984 }
1985 err = sock_sendmsg(sock, msg_sys);
1986 /*
1987 * If this is sendmmsg() and sending to current destination address was
1988 * successful, remember it.
1989 */
1990 if (used_address && err >= 0) {
1991 used_address->name_len = msg_sys->msg_namelen;
1992 if (msg_sys->msg_name)
1993 memcpy(&used_address->name, msg_sys->msg_name,
1994 used_address->name_len);
1995 }
1996
1997out_freectl:
1998 if (ctl_buf != ctl)
1999 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2000out_freeiov:
2001 kfree(iov);
2002 return err;
2003}
2004
2005/*
2006 * BSD sendmsg interface
2007 */
2008
2009long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned flags)
2010{
2011 int fput_needed, err;
2012 struct msghdr msg_sys;
2013 struct socket *sock;
2014
2015 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2016 if (!sock)
2017 goto out;
2018
2019 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2020
2021 fput_light(sock->file, fput_needed);
2022out:
2023 return err;
2024}
2025
2026SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2027{
2028 if (flags & MSG_CMSG_COMPAT)
2029 return -EINVAL;
2030 return __sys_sendmsg(fd, msg, flags);
2031}
2032
2033/*
2034 * Linux sendmmsg interface
2035 */
2036
2037int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2038 unsigned int flags)
2039{
2040 int fput_needed, err, datagrams;
2041 struct socket *sock;
2042 struct mmsghdr __user *entry;
2043 struct compat_mmsghdr __user *compat_entry;
2044 struct msghdr msg_sys;
2045 struct used_address used_address;
2046 unsigned int oflags = flags;
2047
2048 if (vlen > UIO_MAXIOV)
2049 vlen = UIO_MAXIOV;
2050
2051 datagrams = 0;
2052
2053 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2054 if (!sock)
2055 return err;
2056
2057 used_address.name_len = UINT_MAX;
2058 entry = mmsg;
2059 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2060 err = 0;
2061 flags |= MSG_BATCH;
2062
2063 while (datagrams < vlen) {
2064 if (datagrams == vlen - 1)
2065 flags = oflags;
2066
2067 if (MSG_CMSG_COMPAT & flags) {
2068 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2069 &msg_sys, flags, &used_address, MSG_EOR);
2070 if (err < 0)
2071 break;
2072 err = __put_user(err, &compat_entry->msg_len);
2073 ++compat_entry;
2074 } else {
2075 err = ___sys_sendmsg(sock,
2076 (struct user_msghdr __user *)entry,
2077 &msg_sys, flags, &used_address, MSG_EOR);
2078 if (err < 0)
2079 break;
2080 err = put_user(err, &entry->msg_len);
2081 ++entry;
2082 }
2083
2084 if (err)
2085 break;
2086 ++datagrams;
2087 if (msg_data_left(&msg_sys))
2088 break;
2089 cond_resched();
2090 }
2091
2092 fput_light(sock->file, fput_needed);
2093
2094 /* We only return an error if no datagrams were able to be sent */
2095 if (datagrams != 0)
2096 return datagrams;
2097
2098 return err;
2099}
2100
2101SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2102 unsigned int, vlen, unsigned int, flags)
2103{
2104 if (flags & MSG_CMSG_COMPAT)
2105 return -EINVAL;
2106 return __sys_sendmmsg(fd, mmsg, vlen, flags);
2107}
2108
2109static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2110 struct msghdr *msg_sys, unsigned int flags, int nosec)
2111{
2112 struct compat_msghdr __user *msg_compat =
2113 (struct compat_msghdr __user *)msg;
2114 struct iovec iovstack[UIO_FASTIOV];
2115 struct iovec *iov = iovstack;
2116 unsigned long cmsg_ptr;
2117 int len;
2118 ssize_t err;
2119
2120 /* kernel mode address */
2121 struct sockaddr_storage addr;
2122
2123 /* user mode address pointers */
2124 struct sockaddr __user *uaddr;
2125 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2126
2127 msg_sys->msg_name = &addr;
2128
2129 if (MSG_CMSG_COMPAT & flags)
2130 err = get_compat_msghdr(msg_sys, msg_compat, &uaddr, &iov);
2131 else
2132 err = copy_msghdr_from_user(msg_sys, msg, &uaddr, &iov);
2133 if (err < 0)
2134 return err;
2135
2136 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2137 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2138
2139 /* We assume all kernel code knows the size of sockaddr_storage */
2140 msg_sys->msg_namelen = 0;
2141
2142 if (sock->file->f_flags & O_NONBLOCK)
2143 flags |= MSG_DONTWAIT;
2144 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys, flags);
2145 if (err < 0)
2146 goto out_freeiov;
2147 len = err;
2148
2149 if (uaddr != NULL) {
2150 err = move_addr_to_user(&addr,
2151 msg_sys->msg_namelen, uaddr,
2152 uaddr_len);
2153 if (err < 0)
2154 goto out_freeiov;
2155 }
2156 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2157 COMPAT_FLAGS(msg));
2158 if (err)
2159 goto out_freeiov;
2160 if (MSG_CMSG_COMPAT & flags)
2161 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2162 &msg_compat->msg_controllen);
2163 else
2164 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2165 &msg->msg_controllen);
2166 if (err)
2167 goto out_freeiov;
2168 err = len;
2169
2170out_freeiov:
2171 kfree(iov);
2172 return err;
2173}
2174
2175/*
2176 * BSD recvmsg interface
2177 */
2178
2179long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned flags)
2180{
2181 int fput_needed, err;
2182 struct msghdr msg_sys;
2183 struct socket *sock;
2184
2185 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2186 if (!sock)
2187 goto out;
2188
2189 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2190
2191 fput_light(sock->file, fput_needed);
2192out:
2193 return err;
2194}
2195
2196SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2197 unsigned int, flags)
2198{
2199 if (flags & MSG_CMSG_COMPAT)
2200 return -EINVAL;
2201 return __sys_recvmsg(fd, msg, flags);
2202}
2203
2204/*
2205 * Linux recvmmsg interface
2206 */
2207
2208int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2209 unsigned int flags, struct timespec *timeout)
2210{
2211 int fput_needed, err, datagrams;
2212 struct socket *sock;
2213 struct mmsghdr __user *entry;
2214 struct compat_mmsghdr __user *compat_entry;
2215 struct msghdr msg_sys;
2216 struct timespec64 end_time;
2217 struct timespec64 timeout64;
2218
2219 if (timeout &&
2220 poll_select_set_timeout(&end_time, timeout->tv_sec,
2221 timeout->tv_nsec))
2222 return -EINVAL;
2223
2224 datagrams = 0;
2225
2226 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2227 if (!sock)
2228 return err;
2229
2230 err = sock_error(sock->sk);
2231 if (err) {
2232 datagrams = err;
2233 goto out_put;
2234 }
2235
2236 entry = mmsg;
2237 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2238
2239 while (datagrams < vlen) {
2240 /*
2241 * No need to ask LSM for more than the first datagram.
2242 */
2243 if (MSG_CMSG_COMPAT & flags) {
2244 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2245 &msg_sys, flags & ~MSG_WAITFORONE,
2246 datagrams);
2247 if (err < 0)
2248 break;
2249 err = __put_user(err, &compat_entry->msg_len);
2250 ++compat_entry;
2251 } else {
2252 err = ___sys_recvmsg(sock,
2253 (struct user_msghdr __user *)entry,
2254 &msg_sys, flags & ~MSG_WAITFORONE,
2255 datagrams);
2256 if (err < 0)
2257 break;
2258 err = put_user(err, &entry->msg_len);
2259 ++entry;
2260 }
2261
2262 if (err)
2263 break;
2264 ++datagrams;
2265
2266 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2267 if (flags & MSG_WAITFORONE)
2268 flags |= MSG_DONTWAIT;
2269
2270 if (timeout) {
2271 ktime_get_ts64(&timeout64);
2272 *timeout = timespec64_to_timespec(
2273 timespec64_sub(end_time, timeout64));
2274 if (timeout->tv_sec < 0) {
2275 timeout->tv_sec = timeout->tv_nsec = 0;
2276 break;
2277 }
2278
2279 /* Timeout, return less than vlen datagrams */
2280 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2281 break;
2282 }
2283
2284 /* Out of band data, return right away */
2285 if (msg_sys.msg_flags & MSG_OOB)
2286 break;
2287 cond_resched();
2288 }
2289
2290 if (err == 0)
2291 goto out_put;
2292
2293 if (datagrams == 0) {
2294 datagrams = err;
2295 goto out_put;
2296 }
2297
2298 /*
2299 * We may return less entries than requested (vlen) if the
2300 * sock is non block and there aren't enough datagrams...
2301 */
2302 if (err != -EAGAIN) {
2303 /*
2304 * ... or if recvmsg returns an error after we
2305 * received some datagrams, where we record the
2306 * error to return on the next call or if the
2307 * app asks about it using getsockopt(SO_ERROR).
2308 */
2309 sock->sk->sk_err = -err;
2310 }
2311out_put:
2312 fput_light(sock->file, fput_needed);
2313
2314 return datagrams;
2315}
2316
2317SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2318 unsigned int, vlen, unsigned int, flags,
2319 struct timespec __user *, timeout)
2320{
2321 int datagrams;
2322 struct timespec timeout_sys;
2323
2324 if (flags & MSG_CMSG_COMPAT)
2325 return -EINVAL;
2326
2327 if (!timeout)
2328 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2329
2330 if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2331 return -EFAULT;
2332
2333 datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2334
2335 if (datagrams > 0 &&
2336 copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2337 datagrams = -EFAULT;
2338
2339 return datagrams;
2340}
2341
2342#ifdef __ARCH_WANT_SYS_SOCKETCALL
2343/* Argument list sizes for sys_socketcall */
2344#define AL(x) ((x) * sizeof(unsigned long))
2345static const unsigned char nargs[21] = {
2346 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2347 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2348 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2349 AL(4), AL(5), AL(4)
2350};
2351
2352#undef AL
2353
2354/*
2355 * System call vectors.
2356 *
2357 * Argument checking cleaned up. Saved 20% in size.
2358 * This function doesn't need to set the kernel lock because
2359 * it is set by the callees.
2360 */
2361
2362SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2363{
2364 unsigned long a[AUDITSC_ARGS];
2365 unsigned long a0, a1;
2366 int err;
2367 unsigned int len;
2368
2369 if (call < 1 || call > SYS_SENDMMSG)
2370 return -EINVAL;
2371
2372 len = nargs[call];
2373 if (len > sizeof(a))
2374 return -EINVAL;
2375
2376 /* copy_from_user should be SMP safe. */
2377 if (copy_from_user(a, args, len))
2378 return -EFAULT;
2379
2380 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2381 if (err)
2382 return err;
2383
2384 a0 = a[0];
2385 a1 = a[1];
2386
2387 switch (call) {
2388 case SYS_SOCKET:
2389 err = sys_socket(a0, a1, a[2]);
2390 break;
2391 case SYS_BIND:
2392 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2393 break;
2394 case SYS_CONNECT:
2395 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2396 break;
2397 case SYS_LISTEN:
2398 err = sys_listen(a0, a1);
2399 break;
2400 case SYS_ACCEPT:
2401 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2402 (int __user *)a[2], 0);
2403 break;
2404 case SYS_GETSOCKNAME:
2405 err =
2406 sys_getsockname(a0, (struct sockaddr __user *)a1,
2407 (int __user *)a[2]);
2408 break;
2409 case SYS_GETPEERNAME:
2410 err =
2411 sys_getpeername(a0, (struct sockaddr __user *)a1,
2412 (int __user *)a[2]);
2413 break;
2414 case SYS_SOCKETPAIR:
2415 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2416 break;
2417 case SYS_SEND:
2418 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2419 break;
2420 case SYS_SENDTO:
2421 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2422 (struct sockaddr __user *)a[4], a[5]);
2423 break;
2424 case SYS_RECV:
2425 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2426 break;
2427 case SYS_RECVFROM:
2428 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2429 (struct sockaddr __user *)a[4],
2430 (int __user *)a[5]);
2431 break;
2432 case SYS_SHUTDOWN:
2433 err = sys_shutdown(a0, a1);
2434 break;
2435 case SYS_SETSOCKOPT:
2436 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2437 break;
2438 case SYS_GETSOCKOPT:
2439 err =
2440 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2441 (int __user *)a[4]);
2442 break;
2443 case SYS_SENDMSG:
2444 err = sys_sendmsg(a0, (struct user_msghdr __user *)a1, a[2]);
2445 break;
2446 case SYS_SENDMMSG:
2447 err = sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3]);
2448 break;
2449 case SYS_RECVMSG:
2450 err = sys_recvmsg(a0, (struct user_msghdr __user *)a1, a[2]);
2451 break;
2452 case SYS_RECVMMSG:
2453 err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2454 (struct timespec __user *)a[4]);
2455 break;
2456 case SYS_ACCEPT4:
2457 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2458 (int __user *)a[2], a[3]);
2459 break;
2460 default:
2461 err = -EINVAL;
2462 break;
2463 }
2464 return err;
2465}
2466
2467#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2468
2469/**
2470 * sock_register - add a socket protocol handler
2471 * @ops: description of protocol
2472 *
2473 * This function is called by a protocol handler that wants to
2474 * advertise its address family, and have it linked into the
2475 * socket interface. The value ops->family corresponds to the
2476 * socket system call protocol family.
2477 */
2478int sock_register(const struct net_proto_family *ops)
2479{
2480 int err;
2481
2482 if (ops->family >= NPROTO) {
2483 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2484 return -ENOBUFS;
2485 }
2486
2487 spin_lock(&net_family_lock);
2488 if (rcu_dereference_protected(net_families[ops->family],
2489 lockdep_is_held(&net_family_lock)))
2490 err = -EEXIST;
2491 else {
2492 rcu_assign_pointer(net_families[ops->family], ops);
2493 err = 0;
2494 }
2495 spin_unlock(&net_family_lock);
2496
2497 pr_info("NET: Registered protocol family %d\n", ops->family);
2498 return err;
2499}
2500EXPORT_SYMBOL(sock_register);
2501
2502/**
2503 * sock_unregister - remove a protocol handler
2504 * @family: protocol family to remove
2505 *
2506 * This function is called by a protocol handler that wants to
2507 * remove its address family, and have it unlinked from the
2508 * new socket creation.
2509 *
2510 * If protocol handler is a module, then it can use module reference
2511 * counts to protect against new references. If protocol handler is not
2512 * a module then it needs to provide its own protection in
2513 * the ops->create routine.
2514 */
2515void sock_unregister(int family)
2516{
2517 BUG_ON(family < 0 || family >= NPROTO);
2518
2519 spin_lock(&net_family_lock);
2520 RCU_INIT_POINTER(net_families[family], NULL);
2521 spin_unlock(&net_family_lock);
2522
2523 synchronize_rcu();
2524
2525 pr_info("NET: Unregistered protocol family %d\n", family);
2526}
2527EXPORT_SYMBOL(sock_unregister);
2528
2529static int __init sock_init(void)
2530{
2531 int err;
2532 /*
2533 * Initialize the network sysctl infrastructure.
2534 */
2535 err = net_sysctl_init();
2536 if (err)
2537 goto out;
2538
2539 /*
2540 * Initialize skbuff SLAB cache
2541 */
2542 skb_init();
2543
2544 /*
2545 * Initialize the protocols module.
2546 */
2547
2548 init_inodecache();
2549
2550 err = register_filesystem(&sock_fs_type);
2551 if (err)
2552 goto out_fs;
2553 sock_mnt = kern_mount(&sock_fs_type);
2554 if (IS_ERR(sock_mnt)) {
2555 err = PTR_ERR(sock_mnt);
2556 goto out_mount;
2557 }
2558
2559 /* The real protocol initialization is performed in later initcalls.
2560 */
2561
2562#ifdef CONFIG_NETFILTER
2563 err = netfilter_init();
2564 if (err)
2565 goto out;
2566#endif
2567
2568 ptp_classifier_init();
2569
2570out:
2571 return err;
2572
2573out_mount:
2574 unregister_filesystem(&sock_fs_type);
2575out_fs:
2576 goto out;
2577}
2578
2579core_initcall(sock_init); /* early initcall */
2580
2581#ifdef CONFIG_PROC_FS
2582void socket_seq_show(struct seq_file *seq)
2583{
2584 int cpu;
2585 int counter = 0;
2586
2587 for_each_possible_cpu(cpu)
2588 counter += per_cpu(sockets_in_use, cpu);
2589
2590 /* It can be negative, by the way. 8) */
2591 if (counter < 0)
2592 counter = 0;
2593
2594 seq_printf(seq, "sockets: used %d\n", counter);
2595}
2596#endif /* CONFIG_PROC_FS */
2597
2598#ifdef CONFIG_COMPAT
2599static int do_siocgstamp(struct net *net, struct socket *sock,
2600 unsigned int cmd, void __user *up)
2601{
2602 mm_segment_t old_fs = get_fs();
2603 struct timeval ktv;
2604 int err;
2605
2606 set_fs(KERNEL_DS);
2607 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2608 set_fs(old_fs);
2609 if (!err)
2610 err = compat_put_timeval(&ktv, up);
2611
2612 return err;
2613}
2614
2615static int do_siocgstampns(struct net *net, struct socket *sock,
2616 unsigned int cmd, void __user *up)
2617{
2618 mm_segment_t old_fs = get_fs();
2619 struct timespec kts;
2620 int err;
2621
2622 set_fs(KERNEL_DS);
2623 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2624 set_fs(old_fs);
2625 if (!err)
2626 err = compat_put_timespec(&kts, up);
2627
2628 return err;
2629}
2630
2631static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2632{
2633 struct ifreq __user *uifr;
2634 int err;
2635
2636 uifr = compat_alloc_user_space(sizeof(struct ifreq));
2637 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2638 return -EFAULT;
2639
2640 err = dev_ioctl(net, SIOCGIFNAME, uifr);
2641 if (err)
2642 return err;
2643
2644 if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2645 return -EFAULT;
2646
2647 return 0;
2648}
2649
2650static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2651{
2652 struct compat_ifconf ifc32;
2653 struct ifconf ifc;
2654 struct ifconf __user *uifc;
2655 struct compat_ifreq __user *ifr32;
2656 struct ifreq __user *ifr;
2657 unsigned int i, j;
2658 int err;
2659
2660 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2661 return -EFAULT;
2662
2663 memset(&ifc, 0, sizeof(ifc));
2664 if (ifc32.ifcbuf == 0) {
2665 ifc32.ifc_len = 0;
2666 ifc.ifc_len = 0;
2667 ifc.ifc_req = NULL;
2668 uifc = compat_alloc_user_space(sizeof(struct ifconf));
2669 } else {
2670 size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2671 sizeof(struct ifreq);
2672 uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2673 ifc.ifc_len = len;
2674 ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2675 ifr32 = compat_ptr(ifc32.ifcbuf);
2676 for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2677 if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2678 return -EFAULT;
2679 ifr++;
2680 ifr32++;
2681 }
2682 }
2683 if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2684 return -EFAULT;
2685
2686 err = dev_ioctl(net, SIOCGIFCONF, uifc);
2687 if (err)
2688 return err;
2689
2690 if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2691 return -EFAULT;
2692
2693 ifr = ifc.ifc_req;
2694 ifr32 = compat_ptr(ifc32.ifcbuf);
2695 for (i = 0, j = 0;
2696 i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2697 i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2698 if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2699 return -EFAULT;
2700 ifr32++;
2701 ifr++;
2702 }
2703
2704 if (ifc32.ifcbuf == 0) {
2705 /* Translate from 64-bit structure multiple to
2706 * a 32-bit one.
2707 */
2708 i = ifc.ifc_len;
2709 i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2710 ifc32.ifc_len = i;
2711 } else {
2712 ifc32.ifc_len = i;
2713 }
2714 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2715 return -EFAULT;
2716
2717 return 0;
2718}
2719
2720static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2721{
2722 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2723 bool convert_in = false, convert_out = false;
2724 size_t buf_size = ALIGN(sizeof(struct ifreq), 8);
2725 struct ethtool_rxnfc __user *rxnfc;
2726 struct ifreq __user *ifr;
2727 u32 rule_cnt = 0, actual_rule_cnt;
2728 u32 ethcmd;
2729 u32 data;
2730 int ret;
2731
2732 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2733 return -EFAULT;
2734
2735 compat_rxnfc = compat_ptr(data);
2736
2737 if (get_user(ethcmd, &compat_rxnfc->cmd))
2738 return -EFAULT;
2739
2740 /* Most ethtool structures are defined without padding.
2741 * Unfortunately struct ethtool_rxnfc is an exception.
2742 */
2743 switch (ethcmd) {
2744 default:
2745 break;
2746 case ETHTOOL_GRXCLSRLALL:
2747 /* Buffer size is variable */
2748 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2749 return -EFAULT;
2750 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2751 return -ENOMEM;
2752 buf_size += rule_cnt * sizeof(u32);
2753 /* fall through */
2754 case ETHTOOL_GRXRINGS:
2755 case ETHTOOL_GRXCLSRLCNT:
2756 case ETHTOOL_GRXCLSRULE:
2757 case ETHTOOL_SRXCLSRLINS:
2758 convert_out = true;
2759 /* fall through */
2760 case ETHTOOL_SRXCLSRLDEL:
2761 buf_size += sizeof(struct ethtool_rxnfc);
2762 convert_in = true;
2763 break;
2764 }
2765
2766 ifr = compat_alloc_user_space(buf_size);
2767 rxnfc = (void __user *)ifr + ALIGN(sizeof(struct ifreq), 8);
2768
2769 if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2770 return -EFAULT;
2771
2772 if (put_user(convert_in ? rxnfc : compat_ptr(data),
2773 &ifr->ifr_ifru.ifru_data))
2774 return -EFAULT;
2775
2776 if (convert_in) {
2777 /* We expect there to be holes between fs.m_ext and
2778 * fs.ring_cookie and at the end of fs, but nowhere else.
2779 */
2780 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
2781 sizeof(compat_rxnfc->fs.m_ext) !=
2782 offsetof(struct ethtool_rxnfc, fs.m_ext) +
2783 sizeof(rxnfc->fs.m_ext));
2784 BUILD_BUG_ON(
2785 offsetof(struct compat_ethtool_rxnfc, fs.location) -
2786 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2787 offsetof(struct ethtool_rxnfc, fs.location) -
2788 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2789
2790 if (copy_in_user(rxnfc, compat_rxnfc,
2791 (void __user *)(&rxnfc->fs.m_ext + 1) -
2792 (void __user *)rxnfc) ||
2793 copy_in_user(&rxnfc->fs.ring_cookie,
2794 &compat_rxnfc->fs.ring_cookie,
2795 (void __user *)(&rxnfc->fs.location + 1) -
2796 (void __user *)&rxnfc->fs.ring_cookie) ||
2797 copy_in_user(&rxnfc->rule_cnt, &compat_rxnfc->rule_cnt,
2798 sizeof(rxnfc->rule_cnt)))
2799 return -EFAULT;
2800 }
2801
2802 ret = dev_ioctl(net, SIOCETHTOOL, ifr);
2803 if (ret)
2804 return ret;
2805
2806 if (convert_out) {
2807 if (copy_in_user(compat_rxnfc, rxnfc,
2808 (const void __user *)(&rxnfc->fs.m_ext + 1) -
2809 (const void __user *)rxnfc) ||
2810 copy_in_user(&compat_rxnfc->fs.ring_cookie,
2811 &rxnfc->fs.ring_cookie,
2812 (const void __user *)(&rxnfc->fs.location + 1) -
2813 (const void __user *)&rxnfc->fs.ring_cookie) ||
2814 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2815 sizeof(rxnfc->rule_cnt)))
2816 return -EFAULT;
2817
2818 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2819 /* As an optimisation, we only copy the actual
2820 * number of rules that the underlying
2821 * function returned. Since Mallory might
2822 * change the rule count in user memory, we
2823 * check that it is less than the rule count
2824 * originally given (as the user buffer size),
2825 * which has been range-checked.
2826 */
2827 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2828 return -EFAULT;
2829 if (actual_rule_cnt < rule_cnt)
2830 rule_cnt = actual_rule_cnt;
2831 if (copy_in_user(&compat_rxnfc->rule_locs[0],
2832 &rxnfc->rule_locs[0],
2833 rule_cnt * sizeof(u32)))
2834 return -EFAULT;
2835 }
2836 }
2837
2838 return 0;
2839}
2840
2841static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2842{
2843 void __user *uptr;
2844 compat_uptr_t uptr32;
2845 struct ifreq __user *uifr;
2846
2847 uifr = compat_alloc_user_space(sizeof(*uifr));
2848 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2849 return -EFAULT;
2850
2851 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2852 return -EFAULT;
2853
2854 uptr = compat_ptr(uptr32);
2855
2856 if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2857 return -EFAULT;
2858
2859 return dev_ioctl(net, SIOCWANDEV, uifr);
2860}
2861
2862static int bond_ioctl(struct net *net, unsigned int cmd,
2863 struct compat_ifreq __user *ifr32)
2864{
2865 struct ifreq kifr;
2866 mm_segment_t old_fs;
2867 int err;
2868
2869 switch (cmd) {
2870 case SIOCBONDENSLAVE:
2871 case SIOCBONDRELEASE:
2872 case SIOCBONDSETHWADDR:
2873 case SIOCBONDCHANGEACTIVE:
2874 if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2875 return -EFAULT;
2876
2877 old_fs = get_fs();
2878 set_fs(KERNEL_DS);
2879 err = dev_ioctl(net, cmd,
2880 (struct ifreq __user __force *) &kifr);
2881 set_fs(old_fs);
2882
2883 return err;
2884 default:
2885 return -ENOIOCTLCMD;
2886 }
2887}
2888
2889/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
2890static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
2891 struct compat_ifreq __user *u_ifreq32)
2892{
2893 struct ifreq __user *u_ifreq64;
2894 char tmp_buf[IFNAMSIZ];
2895 void __user *data64;
2896 u32 data32;
2897
2898 if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2899 IFNAMSIZ))
2900 return -EFAULT;
2901 if (get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2902 return -EFAULT;
2903 data64 = compat_ptr(data32);
2904
2905 u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2906
2907 if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2908 IFNAMSIZ))
2909 return -EFAULT;
2910 if (put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2911 return -EFAULT;
2912
2913 return dev_ioctl(net, cmd, u_ifreq64);
2914}
2915
2916static int dev_ifsioc(struct net *net, struct socket *sock,
2917 unsigned int cmd, struct compat_ifreq __user *uifr32)
2918{
2919 struct ifreq __user *uifr;
2920 int err;
2921
2922 uifr = compat_alloc_user_space(sizeof(*uifr));
2923 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2924 return -EFAULT;
2925
2926 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2927
2928 if (!err) {
2929 switch (cmd) {
2930 case SIOCGIFFLAGS:
2931 case SIOCGIFMETRIC:
2932 case SIOCGIFMTU:
2933 case SIOCGIFMEM:
2934 case SIOCGIFHWADDR:
2935 case SIOCGIFINDEX:
2936 case SIOCGIFADDR:
2937 case SIOCGIFBRDADDR:
2938 case SIOCGIFDSTADDR:
2939 case SIOCGIFNETMASK:
2940 case SIOCGIFPFLAGS:
2941 case SIOCGIFTXQLEN:
2942 case SIOCGMIIPHY:
2943 case SIOCGMIIREG:
2944 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2945 err = -EFAULT;
2946 break;
2947 }
2948 }
2949 return err;
2950}
2951
2952static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2953 struct compat_ifreq __user *uifr32)
2954{
2955 struct ifreq ifr;
2956 struct compat_ifmap __user *uifmap32;
2957 mm_segment_t old_fs;
2958 int err;
2959
2960 uifmap32 = &uifr32->ifr_ifru.ifru_map;
2961 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2962 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2963 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2964 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2965 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
2966 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
2967 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
2968 if (err)
2969 return -EFAULT;
2970
2971 old_fs = get_fs();
2972 set_fs(KERNEL_DS);
2973 err = dev_ioctl(net, cmd, (void __user __force *)&ifr);
2974 set_fs(old_fs);
2975
2976 if (cmd == SIOCGIFMAP && !err) {
2977 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2978 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2979 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2980 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2981 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
2982 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
2983 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
2984 if (err)
2985 err = -EFAULT;
2986 }
2987 return err;
2988}
2989
2990struct rtentry32 {
2991 u32 rt_pad1;
2992 struct sockaddr rt_dst; /* target address */
2993 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
2994 struct sockaddr rt_genmask; /* target network mask (IP) */
2995 unsigned short rt_flags;
2996 short rt_pad2;
2997 u32 rt_pad3;
2998 unsigned char rt_tos;
2999 unsigned char rt_class;
3000 short rt_pad4;
3001 short rt_metric; /* +1 for binary compatibility! */
3002 /* char * */ u32 rt_dev; /* forcing the device at add */
3003 u32 rt_mtu; /* per route MTU/Window */
3004 u32 rt_window; /* Window clamping */
3005 unsigned short rt_irtt; /* Initial RTT */
3006};
3007
3008struct in6_rtmsg32 {
3009 struct in6_addr rtmsg_dst;
3010 struct in6_addr rtmsg_src;
3011 struct in6_addr rtmsg_gateway;
3012 u32 rtmsg_type;
3013 u16 rtmsg_dst_len;
3014 u16 rtmsg_src_len;
3015 u32 rtmsg_metric;
3016 u32 rtmsg_info;
3017 u32 rtmsg_flags;
3018 s32 rtmsg_ifindex;
3019};
3020
3021static int routing_ioctl(struct net *net, struct socket *sock,
3022 unsigned int cmd, void __user *argp)
3023{
3024 int ret;
3025 void *r = NULL;
3026 struct in6_rtmsg r6;
3027 struct rtentry r4;
3028 char devname[16];
3029 u32 rtdev;
3030 mm_segment_t old_fs = get_fs();
3031
3032 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3033 struct in6_rtmsg32 __user *ur6 = argp;
3034 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3035 3 * sizeof(struct in6_addr));
3036 ret |= get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3037 ret |= get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3038 ret |= get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3039 ret |= get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3040 ret |= get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3041 ret |= get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3042 ret |= get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3043
3044 r = (void *) &r6;
3045 } else { /* ipv4 */
3046 struct rtentry32 __user *ur4 = argp;
3047 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3048 3 * sizeof(struct sockaddr));
3049 ret |= get_user(r4.rt_flags, &(ur4->rt_flags));
3050 ret |= get_user(r4.rt_metric, &(ur4->rt_metric));
3051 ret |= get_user(r4.rt_mtu, &(ur4->rt_mtu));
3052 ret |= get_user(r4.rt_window, &(ur4->rt_window));
3053 ret |= get_user(r4.rt_irtt, &(ur4->rt_irtt));
3054 ret |= get_user(rtdev, &(ur4->rt_dev));
3055 if (rtdev) {
3056 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3057 r4.rt_dev = (char __user __force *)devname;
3058 devname[15] = 0;
3059 } else
3060 r4.rt_dev = NULL;
3061
3062 r = (void *) &r4;
3063 }
3064
3065 if (ret) {
3066 ret = -EFAULT;
3067 goto out;
3068 }
3069
3070 set_fs(KERNEL_DS);
3071 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3072 set_fs(old_fs);
3073
3074out:
3075 return ret;
3076}
3077
3078/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3079 * for some operations; this forces use of the newer bridge-utils that
3080 * use compatible ioctls
3081 */
3082static int old_bridge_ioctl(compat_ulong_t __user *argp)
3083{
3084 compat_ulong_t tmp;
3085
3086 if (get_user(tmp, argp))
3087 return -EFAULT;
3088 if (tmp == BRCTL_GET_VERSION)
3089 return BRCTL_VERSION + 1;
3090 return -EINVAL;
3091}
3092
3093static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3094 unsigned int cmd, unsigned long arg)
3095{
3096 void __user *argp = compat_ptr(arg);
3097 struct sock *sk = sock->sk;
3098 struct net *net = sock_net(sk);
3099
3100 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3101 return compat_ifr_data_ioctl(net, cmd, argp);
3102
3103 switch (cmd) {
3104 case SIOCSIFBR:
3105 case SIOCGIFBR:
3106 return old_bridge_ioctl(argp);
3107 case SIOCGIFNAME:
3108 return dev_ifname32(net, argp);
3109 case SIOCGIFCONF:
3110 return dev_ifconf(net, argp);
3111 case SIOCETHTOOL:
3112 return ethtool_ioctl(net, argp);
3113 case SIOCWANDEV:
3114 return compat_siocwandev(net, argp);
3115 case SIOCGIFMAP:
3116 case SIOCSIFMAP:
3117 return compat_sioc_ifmap(net, cmd, argp);
3118 case SIOCBONDENSLAVE:
3119 case SIOCBONDRELEASE:
3120 case SIOCBONDSETHWADDR:
3121 case SIOCBONDCHANGEACTIVE:
3122 return bond_ioctl(net, cmd, argp);
3123 case SIOCADDRT:
3124 case SIOCDELRT:
3125 return routing_ioctl(net, sock, cmd, argp);
3126 case SIOCGSTAMP:
3127 return do_siocgstamp(net, sock, cmd, argp);
3128 case SIOCGSTAMPNS:
3129 return do_siocgstampns(net, sock, cmd, argp);
3130 case SIOCBONDSLAVEINFOQUERY:
3131 case SIOCBONDINFOQUERY:
3132 case SIOCSHWTSTAMP:
3133 case SIOCGHWTSTAMP:
3134 return compat_ifr_data_ioctl(net, cmd, argp);
3135
3136 case FIOSETOWN:
3137 case SIOCSPGRP:
3138 case FIOGETOWN:
3139 case SIOCGPGRP:
3140 case SIOCBRADDBR:
3141 case SIOCBRDELBR:
3142 case SIOCGIFVLAN:
3143 case SIOCSIFVLAN:
3144 case SIOCADDDLCI:
3145 case SIOCDELDLCI:
3146 case SIOCGSKNS:
3147 return sock_ioctl(file, cmd, arg);
3148
3149 case SIOCGIFFLAGS:
3150 case SIOCSIFFLAGS:
3151 case SIOCGIFMETRIC:
3152 case SIOCSIFMETRIC:
3153 case SIOCGIFMTU:
3154 case SIOCSIFMTU:
3155 case SIOCGIFMEM:
3156 case SIOCSIFMEM:
3157 case SIOCGIFHWADDR:
3158 case SIOCSIFHWADDR:
3159 case SIOCADDMULTI:
3160 case SIOCDELMULTI:
3161 case SIOCGIFINDEX:
3162 case SIOCGIFADDR:
3163 case SIOCSIFADDR:
3164 case SIOCSIFHWBROADCAST:
3165 case SIOCDIFADDR:
3166 case SIOCGIFBRDADDR:
3167 case SIOCSIFBRDADDR:
3168 case SIOCGIFDSTADDR:
3169 case SIOCSIFDSTADDR:
3170 case SIOCGIFNETMASK:
3171 case SIOCSIFNETMASK:
3172 case SIOCSIFPFLAGS:
3173 case SIOCGIFPFLAGS:
3174 case SIOCGIFTXQLEN:
3175 case SIOCSIFTXQLEN:
3176 case SIOCBRADDIF:
3177 case SIOCBRDELIF:
3178 case SIOCSIFNAME:
3179 case SIOCGMIIPHY:
3180 case SIOCGMIIREG:
3181 case SIOCSMIIREG:
3182 return dev_ifsioc(net, sock, cmd, argp);
3183
3184 case SIOCSARP:
3185 case SIOCGARP:
3186 case SIOCDARP:
3187 case SIOCATMARK:
3188 return sock_do_ioctl(net, sock, cmd, arg);
3189 }
3190
3191 return -ENOIOCTLCMD;
3192}
3193
3194static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3195 unsigned long arg)
3196{
3197 struct socket *sock = file->private_data;
3198 int ret = -ENOIOCTLCMD;
3199 struct sock *sk;
3200 struct net *net;
3201
3202 sk = sock->sk;
3203 net = sock_net(sk);
3204
3205 if (sock->ops->compat_ioctl)
3206 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3207
3208 if (ret == -ENOIOCTLCMD &&
3209 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3210 ret = compat_wext_handle_ioctl(net, cmd, arg);
3211
3212 if (ret == -ENOIOCTLCMD)
3213 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3214
3215 return ret;
3216}
3217#endif
3218
3219int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3220{
3221 return sock->ops->bind(sock, addr, addrlen);
3222}
3223EXPORT_SYMBOL(kernel_bind);
3224
3225int kernel_listen(struct socket *sock, int backlog)
3226{
3227 return sock->ops->listen(sock, backlog);
3228}
3229EXPORT_SYMBOL(kernel_listen);
3230
3231int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3232{
3233 struct sock *sk = sock->sk;
3234 int err;
3235
3236 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3237 newsock);
3238 if (err < 0)
3239 goto done;
3240
3241 err = sock->ops->accept(sock, *newsock, flags);
3242 if (err < 0) {
3243 sock_release(*newsock);
3244 *newsock = NULL;
3245 goto done;
3246 }
3247
3248 (*newsock)->ops = sock->ops;
3249 __module_get((*newsock)->ops->owner);
3250
3251done:
3252 return err;
3253}
3254EXPORT_SYMBOL(kernel_accept);
3255
3256int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3257 int flags)
3258{
3259 return sock->ops->connect(sock, addr, addrlen, flags);
3260}
3261EXPORT_SYMBOL(kernel_connect);
3262
3263int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3264 int *addrlen)
3265{
3266 return sock->ops->getname(sock, addr, addrlen, 0);
3267}
3268EXPORT_SYMBOL(kernel_getsockname);
3269
3270int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3271 int *addrlen)
3272{
3273 return sock->ops->getname(sock, addr, addrlen, 1);
3274}
3275EXPORT_SYMBOL(kernel_getpeername);
3276
3277int kernel_getsockopt(struct socket *sock, int level, int optname,
3278 char *optval, int *optlen)
3279{
3280 mm_segment_t oldfs = get_fs();
3281 char __user *uoptval;
3282 int __user *uoptlen;
3283 int err;
3284
3285 uoptval = (char __user __force *) optval;
3286 uoptlen = (int __user __force *) optlen;
3287
3288 set_fs(KERNEL_DS);
3289 if (level == SOL_SOCKET)
3290 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3291 else
3292 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3293 uoptlen);
3294 set_fs(oldfs);
3295 return err;
3296}
3297EXPORT_SYMBOL(kernel_getsockopt);
3298
3299int kernel_setsockopt(struct socket *sock, int level, int optname,
3300 char *optval, unsigned int optlen)
3301{
3302 mm_segment_t oldfs = get_fs();
3303 char __user *uoptval;
3304 int err;
3305
3306 uoptval = (char __user __force *) optval;
3307
3308 set_fs(KERNEL_DS);
3309 if (level == SOL_SOCKET)
3310 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3311 else
3312 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3313 optlen);
3314 set_fs(oldfs);
3315 return err;
3316}
3317EXPORT_SYMBOL(kernel_setsockopt);
3318
3319int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3320 size_t size, int flags)
3321{
3322 if (sock->ops->sendpage)
3323 return sock->ops->sendpage(sock, page, offset, size, flags);
3324
3325 return sock_no_sendpage(sock, page, offset, size, flags);
3326}
3327EXPORT_SYMBOL(kernel_sendpage);
3328
3329int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3330{
3331 mm_segment_t oldfs = get_fs();
3332 int err;
3333
3334 set_fs(KERNEL_DS);
3335 err = sock->ops->ioctl(sock, cmd, arg);
3336 set_fs(oldfs);
3337
3338 return err;
3339}
3340EXPORT_SYMBOL(kernel_sock_ioctl);
3341
3342int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3343{
3344 return sock->ops->shutdown(sock, how);
3345}
3346EXPORT_SYMBOL(kernel_sock_shutdown);
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/sockios.h>
104#include <net/busy_poll.h>
105#include <linux/errqueue.h>
106
107#ifdef CONFIG_NET_RX_BUSY_POLL
108unsigned int sysctl_net_busy_read __read_mostly;
109unsigned int sysctl_net_busy_poll __read_mostly;
110#endif
111
112static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to);
113static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from);
114static int sock_mmap(struct file *file, struct vm_area_struct *vma);
115
116static int sock_close(struct inode *inode, struct file *file);
117static __poll_t sock_poll(struct file *file,
118 struct poll_table_struct *wait);
119static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
120#ifdef CONFIG_COMPAT
121static long compat_sock_ioctl(struct file *file,
122 unsigned int cmd, unsigned long arg);
123#endif
124static int sock_fasync(int fd, struct file *filp, int on);
125static ssize_t sock_sendpage(struct file *file, struct page *page,
126 int offset, size_t size, loff_t *ppos, int more);
127static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
128 struct pipe_inode_info *pipe, size_t len,
129 unsigned int flags);
130
131/*
132 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
133 * in the operation structures but are done directly via the socketcall() multiplexor.
134 */
135
136static const struct file_operations socket_file_ops = {
137 .owner = THIS_MODULE,
138 .llseek = no_llseek,
139 .read_iter = sock_read_iter,
140 .write_iter = sock_write_iter,
141 .poll = sock_poll,
142 .unlocked_ioctl = sock_ioctl,
143#ifdef CONFIG_COMPAT
144 .compat_ioctl = compat_sock_ioctl,
145#endif
146 .mmap = sock_mmap,
147 .release = sock_close,
148 .fasync = sock_fasync,
149 .sendpage = sock_sendpage,
150 .splice_write = generic_splice_sendpage,
151 .splice_read = sock_splice_read,
152};
153
154/*
155 * The protocol list. Each protocol is registered in here.
156 */
157
158static DEFINE_SPINLOCK(net_family_lock);
159static const struct net_proto_family __rcu *net_families[NPROTO] __read_mostly;
160
161/*
162 * Support routines.
163 * Move socket addresses back and forth across the kernel/user
164 * divide and look after the messy bits.
165 */
166
167/**
168 * move_addr_to_kernel - copy a socket address into kernel space
169 * @uaddr: Address in user space
170 * @kaddr: Address in kernel space
171 * @ulen: Length in user space
172 *
173 * The address is copied into kernel space. If the provided address is
174 * too long an error code of -EINVAL is returned. If the copy gives
175 * invalid addresses -EFAULT is returned. On a success 0 is returned.
176 */
177
178int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
179{
180 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
181 return -EINVAL;
182 if (ulen == 0)
183 return 0;
184 if (copy_from_user(kaddr, uaddr, ulen))
185 return -EFAULT;
186 return audit_sockaddr(ulen, kaddr);
187}
188
189/**
190 * move_addr_to_user - copy an address to user space
191 * @kaddr: kernel space address
192 * @klen: length of address in kernel
193 * @uaddr: user space address
194 * @ulen: pointer to user length field
195 *
196 * The value pointed to by ulen on entry is the buffer length available.
197 * This is overwritten with the buffer space used. -EINVAL is returned
198 * if an overlong buffer is specified or a negative buffer size. -EFAULT
199 * is returned if either the buffer or the length field are not
200 * accessible.
201 * After copying the data up to the limit the user specifies, the true
202 * length of the data is written over the length limit the user
203 * specified. Zero is returned for a success.
204 */
205
206static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
207 void __user *uaddr, int __user *ulen)
208{
209 int err;
210 int len;
211
212 BUG_ON(klen > sizeof(struct sockaddr_storage));
213 err = get_user(len, ulen);
214 if (err)
215 return err;
216 if (len > klen)
217 len = klen;
218 if (len < 0)
219 return -EINVAL;
220 if (len) {
221 if (audit_sockaddr(klen, kaddr))
222 return -ENOMEM;
223 if (copy_to_user(uaddr, kaddr, len))
224 return -EFAULT;
225 }
226 /*
227 * "fromlen shall refer to the value before truncation.."
228 * 1003.1g
229 */
230 return __put_user(klen, ulen);
231}
232
233static struct kmem_cache *sock_inode_cachep __ro_after_init;
234
235static struct inode *sock_alloc_inode(struct super_block *sb)
236{
237 struct socket_alloc *ei;
238
239 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
240 if (!ei)
241 return NULL;
242 init_waitqueue_head(&ei->socket.wq.wait);
243 ei->socket.wq.fasync_list = NULL;
244 ei->socket.wq.flags = 0;
245
246 ei->socket.state = SS_UNCONNECTED;
247 ei->socket.flags = 0;
248 ei->socket.ops = NULL;
249 ei->socket.sk = NULL;
250 ei->socket.file = NULL;
251
252 return &ei->vfs_inode;
253}
254
255static void sock_free_inode(struct inode *inode)
256{
257 struct socket_alloc *ei;
258
259 ei = container_of(inode, struct socket_alloc, vfs_inode);
260 kmem_cache_free(sock_inode_cachep, ei);
261}
262
263static void init_once(void *foo)
264{
265 struct socket_alloc *ei = (struct socket_alloc *)foo;
266
267 inode_init_once(&ei->vfs_inode);
268}
269
270static void init_inodecache(void)
271{
272 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
273 sizeof(struct socket_alloc),
274 0,
275 (SLAB_HWCACHE_ALIGN |
276 SLAB_RECLAIM_ACCOUNT |
277 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
278 init_once);
279 BUG_ON(sock_inode_cachep == NULL);
280}
281
282static const struct super_operations sockfs_ops = {
283 .alloc_inode = sock_alloc_inode,
284 .free_inode = sock_free_inode,
285 .statfs = simple_statfs,
286};
287
288/*
289 * sockfs_dname() is called from d_path().
290 */
291static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
292{
293 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
294 d_inode(dentry)->i_ino);
295}
296
297static const struct dentry_operations sockfs_dentry_operations = {
298 .d_dname = sockfs_dname,
299};
300
301static int sockfs_xattr_get(const struct xattr_handler *handler,
302 struct dentry *dentry, struct inode *inode,
303 const char *suffix, void *value, size_t size)
304{
305 if (value) {
306 if (dentry->d_name.len + 1 > size)
307 return -ERANGE;
308 memcpy(value, dentry->d_name.name, dentry->d_name.len + 1);
309 }
310 return dentry->d_name.len + 1;
311}
312
313#define XATTR_SOCKPROTONAME_SUFFIX "sockprotoname"
314#define XATTR_NAME_SOCKPROTONAME (XATTR_SYSTEM_PREFIX XATTR_SOCKPROTONAME_SUFFIX)
315#define XATTR_NAME_SOCKPROTONAME_LEN (sizeof(XATTR_NAME_SOCKPROTONAME)-1)
316
317static const struct xattr_handler sockfs_xattr_handler = {
318 .name = XATTR_NAME_SOCKPROTONAME,
319 .get = sockfs_xattr_get,
320};
321
322static int sockfs_security_xattr_set(const struct xattr_handler *handler,
323 struct dentry *dentry, struct inode *inode,
324 const char *suffix, const void *value,
325 size_t size, int flags)
326{
327 /* Handled by LSM. */
328 return -EAGAIN;
329}
330
331static const struct xattr_handler sockfs_security_xattr_handler = {
332 .prefix = XATTR_SECURITY_PREFIX,
333 .set = sockfs_security_xattr_set,
334};
335
336static const struct xattr_handler *sockfs_xattr_handlers[] = {
337 &sockfs_xattr_handler,
338 &sockfs_security_xattr_handler,
339 NULL
340};
341
342static int sockfs_init_fs_context(struct fs_context *fc)
343{
344 struct pseudo_fs_context *ctx = init_pseudo(fc, SOCKFS_MAGIC);
345 if (!ctx)
346 return -ENOMEM;
347 ctx->ops = &sockfs_ops;
348 ctx->dops = &sockfs_dentry_operations;
349 ctx->xattr = sockfs_xattr_handlers;
350 return 0;
351}
352
353static struct vfsmount *sock_mnt __read_mostly;
354
355static struct file_system_type sock_fs_type = {
356 .name = "sockfs",
357 .init_fs_context = sockfs_init_fs_context,
358 .kill_sb = kill_anon_super,
359};
360
361/*
362 * Obtains the first available file descriptor and sets it up for use.
363 *
364 * These functions create file structures and maps them to fd space
365 * of the current process. On success it returns file descriptor
366 * and file struct implicitly stored in sock->file.
367 * Note that another thread may close file descriptor before we return
368 * from this function. We use the fact that now we do not refer
369 * to socket after mapping. If one day we will need it, this
370 * function will increment ref. count on file by 1.
371 *
372 * In any case returned fd MAY BE not valid!
373 * This race condition is unavoidable
374 * with shared fd spaces, we cannot solve it inside kernel,
375 * but we take care of internal coherence yet.
376 */
377
378/**
379 * sock_alloc_file - Bind a &socket to a &file
380 * @sock: socket
381 * @flags: file status flags
382 * @dname: protocol name
383 *
384 * Returns the &file bound with @sock, implicitly storing it
385 * in sock->file. If dname is %NULL, sets to "".
386 * On failure the return is a ERR pointer (see linux/err.h).
387 * This function uses GFP_KERNEL internally.
388 */
389
390struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname)
391{
392 struct file *file;
393
394 if (!dname)
395 dname = sock->sk ? sock->sk->sk_prot_creator->name : "";
396
397 file = alloc_file_pseudo(SOCK_INODE(sock), sock_mnt, dname,
398 O_RDWR | (flags & O_NONBLOCK),
399 &socket_file_ops);
400 if (IS_ERR(file)) {
401 sock_release(sock);
402 return file;
403 }
404
405 sock->file = file;
406 file->private_data = sock;
407 return file;
408}
409EXPORT_SYMBOL(sock_alloc_file);
410
411static int sock_map_fd(struct socket *sock, int flags)
412{
413 struct file *newfile;
414 int fd = get_unused_fd_flags(flags);
415 if (unlikely(fd < 0)) {
416 sock_release(sock);
417 return fd;
418 }
419
420 newfile = sock_alloc_file(sock, flags, NULL);
421 if (!IS_ERR(newfile)) {
422 fd_install(fd, newfile);
423 return fd;
424 }
425
426 put_unused_fd(fd);
427 return PTR_ERR(newfile);
428}
429
430/**
431 * sock_from_file - Return the &socket bounded to @file.
432 * @file: file
433 * @err: pointer to an error code return
434 *
435 * On failure returns %NULL and assigns -ENOTSOCK to @err.
436 */
437
438struct socket *sock_from_file(struct file *file, int *err)
439{
440 if (file->f_op == &socket_file_ops)
441 return file->private_data; /* set in sock_map_fd */
442
443 *err = -ENOTSOCK;
444 return NULL;
445}
446EXPORT_SYMBOL(sock_from_file);
447
448/**
449 * sockfd_lookup - Go from a file number to its socket slot
450 * @fd: file handle
451 * @err: pointer to an error code return
452 *
453 * The file handle passed in is locked and the socket it is bound
454 * to is returned. If an error occurs the err pointer is overwritten
455 * with a negative errno code and NULL is returned. The function checks
456 * for both invalid handles and passing a handle which is not a socket.
457 *
458 * On a success the socket object pointer is returned.
459 */
460
461struct socket *sockfd_lookup(int fd, int *err)
462{
463 struct file *file;
464 struct socket *sock;
465
466 file = fget(fd);
467 if (!file) {
468 *err = -EBADF;
469 return NULL;
470 }
471
472 sock = sock_from_file(file, err);
473 if (!sock)
474 fput(file);
475 return sock;
476}
477EXPORT_SYMBOL(sockfd_lookup);
478
479static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
480{
481 struct fd f = fdget(fd);
482 struct socket *sock;
483
484 *err = -EBADF;
485 if (f.file) {
486 sock = sock_from_file(f.file, err);
487 if (likely(sock)) {
488 *fput_needed = f.flags;
489 return sock;
490 }
491 fdput(f);
492 }
493 return NULL;
494}
495
496static ssize_t sockfs_listxattr(struct dentry *dentry, char *buffer,
497 size_t size)
498{
499 ssize_t len;
500 ssize_t used = 0;
501
502 len = security_inode_listsecurity(d_inode(dentry), buffer, size);
503 if (len < 0)
504 return len;
505 used += len;
506 if (buffer) {
507 if (size < used)
508 return -ERANGE;
509 buffer += len;
510 }
511
512 len = (XATTR_NAME_SOCKPROTONAME_LEN + 1);
513 used += len;
514 if (buffer) {
515 if (size < used)
516 return -ERANGE;
517 memcpy(buffer, XATTR_NAME_SOCKPROTONAME, len);
518 buffer += len;
519 }
520
521 return used;
522}
523
524static int sockfs_setattr(struct dentry *dentry, struct iattr *iattr)
525{
526 int err = simple_setattr(dentry, iattr);
527
528 if (!err && (iattr->ia_valid & ATTR_UID)) {
529 struct socket *sock = SOCKET_I(d_inode(dentry));
530
531 if (sock->sk)
532 sock->sk->sk_uid = iattr->ia_uid;
533 else
534 err = -ENOENT;
535 }
536
537 return err;
538}
539
540static const struct inode_operations sockfs_inode_ops = {
541 .listxattr = sockfs_listxattr,
542 .setattr = sockfs_setattr,
543};
544
545/**
546 * sock_alloc - allocate a socket
547 *
548 * Allocate a new inode and socket object. The two are bound together
549 * and initialised. The socket is then returned. If we are out of inodes
550 * NULL is returned. This functions uses GFP_KERNEL internally.
551 */
552
553struct socket *sock_alloc(void)
554{
555 struct inode *inode;
556 struct socket *sock;
557
558 inode = new_inode_pseudo(sock_mnt->mnt_sb);
559 if (!inode)
560 return NULL;
561
562 sock = SOCKET_I(inode);
563
564 inode->i_ino = get_next_ino();
565 inode->i_mode = S_IFSOCK | S_IRWXUGO;
566 inode->i_uid = current_fsuid();
567 inode->i_gid = current_fsgid();
568 inode->i_op = &sockfs_inode_ops;
569
570 return sock;
571}
572EXPORT_SYMBOL(sock_alloc);
573
574/**
575 * sock_release - close a socket
576 * @sock: socket to close
577 *
578 * The socket is released from the protocol stack if it has a release
579 * callback, and the inode is then released if the socket is bound to
580 * an inode not a file.
581 */
582
583static void __sock_release(struct socket *sock, struct inode *inode)
584{
585 if (sock->ops) {
586 struct module *owner = sock->ops->owner;
587
588 if (inode)
589 inode_lock(inode);
590 sock->ops->release(sock);
591 sock->sk = NULL;
592 if (inode)
593 inode_unlock(inode);
594 sock->ops = NULL;
595 module_put(owner);
596 }
597
598 if (sock->wq.fasync_list)
599 pr_err("%s: fasync list not empty!\n", __func__);
600
601 if (!sock->file) {
602 iput(SOCK_INODE(sock));
603 return;
604 }
605 sock->file = NULL;
606}
607
608void sock_release(struct socket *sock)
609{
610 __sock_release(sock, NULL);
611}
612EXPORT_SYMBOL(sock_release);
613
614void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags)
615{
616 u8 flags = *tx_flags;
617
618 if (tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
619 flags |= SKBTX_HW_TSTAMP;
620
621 if (tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
622 flags |= SKBTX_SW_TSTAMP;
623
624 if (tsflags & SOF_TIMESTAMPING_TX_SCHED)
625 flags |= SKBTX_SCHED_TSTAMP;
626
627 *tx_flags = flags;
628}
629EXPORT_SYMBOL(__sock_tx_timestamp);
630
631INDIRECT_CALLABLE_DECLARE(int inet_sendmsg(struct socket *, struct msghdr *,
632 size_t));
633INDIRECT_CALLABLE_DECLARE(int inet6_sendmsg(struct socket *, struct msghdr *,
634 size_t));
635static inline int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg)
636{
637 int ret = INDIRECT_CALL_INET(sock->ops->sendmsg, inet6_sendmsg,
638 inet_sendmsg, sock, msg,
639 msg_data_left(msg));
640 BUG_ON(ret == -EIOCBQUEUED);
641 return ret;
642}
643
644/**
645 * sock_sendmsg - send a message through @sock
646 * @sock: socket
647 * @msg: message to send
648 *
649 * Sends @msg through @sock, passing through LSM.
650 * Returns the number of bytes sent, or an error code.
651 */
652int sock_sendmsg(struct socket *sock, struct msghdr *msg)
653{
654 int err = security_socket_sendmsg(sock, msg,
655 msg_data_left(msg));
656
657 return err ?: sock_sendmsg_nosec(sock, msg);
658}
659EXPORT_SYMBOL(sock_sendmsg);
660
661/**
662 * kernel_sendmsg - send a message through @sock (kernel-space)
663 * @sock: socket
664 * @msg: message header
665 * @vec: kernel vec
666 * @num: vec array length
667 * @size: total message data size
668 *
669 * Builds the message data with @vec and sends it through @sock.
670 * Returns the number of bytes sent, or an error code.
671 */
672
673int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
674 struct kvec *vec, size_t num, size_t size)
675{
676 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
677 return sock_sendmsg(sock, msg);
678}
679EXPORT_SYMBOL(kernel_sendmsg);
680
681/**
682 * kernel_sendmsg_locked - send a message through @sock (kernel-space)
683 * @sk: sock
684 * @msg: message header
685 * @vec: output s/g array
686 * @num: output s/g array length
687 * @size: total message data size
688 *
689 * Builds the message data with @vec and sends it through @sock.
690 * Returns the number of bytes sent, or an error code.
691 * Caller must hold @sk.
692 */
693
694int kernel_sendmsg_locked(struct sock *sk, struct msghdr *msg,
695 struct kvec *vec, size_t num, size_t size)
696{
697 struct socket *sock = sk->sk_socket;
698
699 if (!sock->ops->sendmsg_locked)
700 return sock_no_sendmsg_locked(sk, msg, size);
701
702 iov_iter_kvec(&msg->msg_iter, WRITE, vec, num, size);
703
704 return sock->ops->sendmsg_locked(sk, msg, msg_data_left(msg));
705}
706EXPORT_SYMBOL(kernel_sendmsg_locked);
707
708static bool skb_is_err_queue(const struct sk_buff *skb)
709{
710 /* pkt_type of skbs enqueued on the error queue are set to
711 * PACKET_OUTGOING in skb_set_err_queue(). This is only safe to do
712 * in recvmsg, since skbs received on a local socket will never
713 * have a pkt_type of PACKET_OUTGOING.
714 */
715 return skb->pkt_type == PACKET_OUTGOING;
716}
717
718/* On transmit, software and hardware timestamps are returned independently.
719 * As the two skb clones share the hardware timestamp, which may be updated
720 * before the software timestamp is received, a hardware TX timestamp may be
721 * returned only if there is no software TX timestamp. Ignore false software
722 * timestamps, which may be made in the __sock_recv_timestamp() call when the
723 * option SO_TIMESTAMP_OLD(NS) is enabled on the socket, even when the skb has a
724 * hardware timestamp.
725 */
726static bool skb_is_swtx_tstamp(const struct sk_buff *skb, int false_tstamp)
727{
728 return skb->tstamp && !false_tstamp && skb_is_err_queue(skb);
729}
730
731static void put_ts_pktinfo(struct msghdr *msg, struct sk_buff *skb)
732{
733 struct scm_ts_pktinfo ts_pktinfo;
734 struct net_device *orig_dev;
735
736 if (!skb_mac_header_was_set(skb))
737 return;
738
739 memset(&ts_pktinfo, 0, sizeof(ts_pktinfo));
740
741 rcu_read_lock();
742 orig_dev = dev_get_by_napi_id(skb_napi_id(skb));
743 if (orig_dev)
744 ts_pktinfo.if_index = orig_dev->ifindex;
745 rcu_read_unlock();
746
747 ts_pktinfo.pkt_length = skb->len - skb_mac_offset(skb);
748 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_PKTINFO,
749 sizeof(ts_pktinfo), &ts_pktinfo);
750}
751
752/*
753 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
754 */
755void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
756 struct sk_buff *skb)
757{
758 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
759 int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
760 struct scm_timestamping_internal tss;
761
762 int empty = 1, false_tstamp = 0;
763 struct skb_shared_hwtstamps *shhwtstamps =
764 skb_hwtstamps(skb);
765
766 /* Race occurred between timestamp enabling and packet
767 receiving. Fill in the current time for now. */
768 if (need_software_tstamp && skb->tstamp == 0) {
769 __net_timestamp(skb);
770 false_tstamp = 1;
771 }
772
773 if (need_software_tstamp) {
774 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
775 if (new_tstamp) {
776 struct __kernel_sock_timeval tv;
777
778 skb_get_new_timestamp(skb, &tv);
779 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
780 sizeof(tv), &tv);
781 } else {
782 struct __kernel_old_timeval tv;
783
784 skb_get_timestamp(skb, &tv);
785 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
786 sizeof(tv), &tv);
787 }
788 } else {
789 if (new_tstamp) {
790 struct __kernel_timespec ts;
791
792 skb_get_new_timestampns(skb, &ts);
793 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
794 sizeof(ts), &ts);
795 } else {
796 struct timespec ts;
797
798 skb_get_timestampns(skb, &ts);
799 put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
800 sizeof(ts), &ts);
801 }
802 }
803 }
804
805 memset(&tss, 0, sizeof(tss));
806 if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) &&
807 ktime_to_timespec64_cond(skb->tstamp, tss.ts + 0))
808 empty = 0;
809 if (shhwtstamps &&
810 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
811 !skb_is_swtx_tstamp(skb, false_tstamp) &&
812 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, tss.ts + 2)) {
813 empty = 0;
814 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_PKTINFO) &&
815 !skb_is_err_queue(skb))
816 put_ts_pktinfo(msg, skb);
817 }
818 if (!empty) {
819 if (sock_flag(sk, SOCK_TSTAMP_NEW))
820 put_cmsg_scm_timestamping64(msg, &tss);
821 else
822 put_cmsg_scm_timestamping(msg, &tss);
823
824 if (skb_is_err_queue(skb) && skb->len &&
825 SKB_EXT_ERR(skb)->opt_stats)
826 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPING_OPT_STATS,
827 skb->len, skb->data);
828 }
829}
830EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
831
832void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
833 struct sk_buff *skb)
834{
835 int ack;
836
837 if (!sock_flag(sk, SOCK_WIFI_STATUS))
838 return;
839 if (!skb->wifi_acked_valid)
840 return;
841
842 ack = skb->wifi_acked;
843
844 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
845}
846EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
847
848static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
849 struct sk_buff *skb)
850{
851 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
852 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
853 sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
854}
855
856void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
857 struct sk_buff *skb)
858{
859 sock_recv_timestamp(msg, sk, skb);
860 sock_recv_drops(msg, sk, skb);
861}
862EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
863
864INDIRECT_CALLABLE_DECLARE(int inet_recvmsg(struct socket *, struct msghdr *,
865 size_t, int));
866INDIRECT_CALLABLE_DECLARE(int inet6_recvmsg(struct socket *, struct msghdr *,
867 size_t, int));
868static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
869 int flags)
870{
871 return INDIRECT_CALL_INET(sock->ops->recvmsg, inet6_recvmsg,
872 inet_recvmsg, sock, msg, msg_data_left(msg),
873 flags);
874}
875
876/**
877 * sock_recvmsg - receive a message from @sock
878 * @sock: socket
879 * @msg: message to receive
880 * @flags: message flags
881 *
882 * Receives @msg from @sock, passing through LSM. Returns the total number
883 * of bytes received, or an error.
884 */
885int sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags)
886{
887 int err = security_socket_recvmsg(sock, msg, msg_data_left(msg), flags);
888
889 return err ?: sock_recvmsg_nosec(sock, msg, flags);
890}
891EXPORT_SYMBOL(sock_recvmsg);
892
893/**
894 * kernel_recvmsg - Receive a message from a socket (kernel space)
895 * @sock: The socket to receive the message from
896 * @msg: Received message
897 * @vec: Input s/g array for message data
898 * @num: Size of input s/g array
899 * @size: Number of bytes to read
900 * @flags: Message flags (MSG_DONTWAIT, etc...)
901 *
902 * On return the msg structure contains the scatter/gather array passed in the
903 * vec argument. The array is modified so that it consists of the unfilled
904 * portion of the original array.
905 *
906 * The returned value is the total number of bytes received, or an error.
907 */
908
909int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
910 struct kvec *vec, size_t num, size_t size, int flags)
911{
912 mm_segment_t oldfs = get_fs();
913 int result;
914
915 iov_iter_kvec(&msg->msg_iter, READ, vec, num, size);
916 set_fs(KERNEL_DS);
917 result = sock_recvmsg(sock, msg, flags);
918 set_fs(oldfs);
919 return result;
920}
921EXPORT_SYMBOL(kernel_recvmsg);
922
923static ssize_t sock_sendpage(struct file *file, struct page *page,
924 int offset, size_t size, loff_t *ppos, int more)
925{
926 struct socket *sock;
927 int flags;
928
929 sock = file->private_data;
930
931 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
932 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
933 flags |= more;
934
935 return kernel_sendpage(sock, page, offset, size, flags);
936}
937
938static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
939 struct pipe_inode_info *pipe, size_t len,
940 unsigned int flags)
941{
942 struct socket *sock = file->private_data;
943
944 if (unlikely(!sock->ops->splice_read))
945 return generic_file_splice_read(file, ppos, pipe, len, flags);
946
947 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
948}
949
950static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
951{
952 struct file *file = iocb->ki_filp;
953 struct socket *sock = file->private_data;
954 struct msghdr msg = {.msg_iter = *to,
955 .msg_iocb = iocb};
956 ssize_t res;
957
958 if (file->f_flags & O_NONBLOCK)
959 msg.msg_flags = MSG_DONTWAIT;
960
961 if (iocb->ki_pos != 0)
962 return -ESPIPE;
963
964 if (!iov_iter_count(to)) /* Match SYS5 behaviour */
965 return 0;
966
967 res = sock_recvmsg(sock, &msg, msg.msg_flags);
968 *to = msg.msg_iter;
969 return res;
970}
971
972static ssize_t sock_write_iter(struct kiocb *iocb, struct iov_iter *from)
973{
974 struct file *file = iocb->ki_filp;
975 struct socket *sock = file->private_data;
976 struct msghdr msg = {.msg_iter = *from,
977 .msg_iocb = iocb};
978 ssize_t res;
979
980 if (iocb->ki_pos != 0)
981 return -ESPIPE;
982
983 if (file->f_flags & O_NONBLOCK)
984 msg.msg_flags = MSG_DONTWAIT;
985
986 if (sock->type == SOCK_SEQPACKET)
987 msg.msg_flags |= MSG_EOR;
988
989 res = sock_sendmsg(sock, &msg);
990 *from = msg.msg_iter;
991 return res;
992}
993
994/*
995 * Atomic setting of ioctl hooks to avoid race
996 * with module unload.
997 */
998
999static DEFINE_MUTEX(br_ioctl_mutex);
1000static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
1001
1002void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
1003{
1004 mutex_lock(&br_ioctl_mutex);
1005 br_ioctl_hook = hook;
1006 mutex_unlock(&br_ioctl_mutex);
1007}
1008EXPORT_SYMBOL(brioctl_set);
1009
1010static DEFINE_MUTEX(vlan_ioctl_mutex);
1011static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
1012
1013void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
1014{
1015 mutex_lock(&vlan_ioctl_mutex);
1016 vlan_ioctl_hook = hook;
1017 mutex_unlock(&vlan_ioctl_mutex);
1018}
1019EXPORT_SYMBOL(vlan_ioctl_set);
1020
1021static DEFINE_MUTEX(dlci_ioctl_mutex);
1022static int (*dlci_ioctl_hook) (unsigned int, void __user *);
1023
1024void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
1025{
1026 mutex_lock(&dlci_ioctl_mutex);
1027 dlci_ioctl_hook = hook;
1028 mutex_unlock(&dlci_ioctl_mutex);
1029}
1030EXPORT_SYMBOL(dlci_ioctl_set);
1031
1032static long sock_do_ioctl(struct net *net, struct socket *sock,
1033 unsigned int cmd, unsigned long arg)
1034{
1035 int err;
1036 void __user *argp = (void __user *)arg;
1037
1038 err = sock->ops->ioctl(sock, cmd, arg);
1039
1040 /*
1041 * If this ioctl is unknown try to hand it down
1042 * to the NIC driver.
1043 */
1044 if (err != -ENOIOCTLCMD)
1045 return err;
1046
1047 if (cmd == SIOCGIFCONF) {
1048 struct ifconf ifc;
1049 if (copy_from_user(&ifc, argp, sizeof(struct ifconf)))
1050 return -EFAULT;
1051 rtnl_lock();
1052 err = dev_ifconf(net, &ifc, sizeof(struct ifreq));
1053 rtnl_unlock();
1054 if (!err && copy_to_user(argp, &ifc, sizeof(struct ifconf)))
1055 err = -EFAULT;
1056 } else {
1057 struct ifreq ifr;
1058 bool need_copyout;
1059 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1060 return -EFAULT;
1061 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1062 if (!err && need_copyout)
1063 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1064 return -EFAULT;
1065 }
1066 return err;
1067}
1068
1069/*
1070 * With an ioctl, arg may well be a user mode pointer, but we don't know
1071 * what to do with it - that's up to the protocol still.
1072 */
1073
1074/**
1075 * get_net_ns - increment the refcount of the network namespace
1076 * @ns: common namespace (net)
1077 *
1078 * Returns the net's common namespace.
1079 */
1080
1081struct ns_common *get_net_ns(struct ns_common *ns)
1082{
1083 return &get_net(container_of(ns, struct net, ns))->ns;
1084}
1085EXPORT_SYMBOL_GPL(get_net_ns);
1086
1087static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1088{
1089 struct socket *sock;
1090 struct sock *sk;
1091 void __user *argp = (void __user *)arg;
1092 int pid, err;
1093 struct net *net;
1094
1095 sock = file->private_data;
1096 sk = sock->sk;
1097 net = sock_net(sk);
1098 if (unlikely(cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))) {
1099 struct ifreq ifr;
1100 bool need_copyout;
1101 if (copy_from_user(&ifr, argp, sizeof(struct ifreq)))
1102 return -EFAULT;
1103 err = dev_ioctl(net, cmd, &ifr, &need_copyout);
1104 if (!err && need_copyout)
1105 if (copy_to_user(argp, &ifr, sizeof(struct ifreq)))
1106 return -EFAULT;
1107 } else
1108#ifdef CONFIG_WEXT_CORE
1109 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1110 err = wext_handle_ioctl(net, cmd, argp);
1111 } else
1112#endif
1113 switch (cmd) {
1114 case FIOSETOWN:
1115 case SIOCSPGRP:
1116 err = -EFAULT;
1117 if (get_user(pid, (int __user *)argp))
1118 break;
1119 err = f_setown(sock->file, pid, 1);
1120 break;
1121 case FIOGETOWN:
1122 case SIOCGPGRP:
1123 err = put_user(f_getown(sock->file),
1124 (int __user *)argp);
1125 break;
1126 case SIOCGIFBR:
1127 case SIOCSIFBR:
1128 case SIOCBRADDBR:
1129 case SIOCBRDELBR:
1130 err = -ENOPKG;
1131 if (!br_ioctl_hook)
1132 request_module("bridge");
1133
1134 mutex_lock(&br_ioctl_mutex);
1135 if (br_ioctl_hook)
1136 err = br_ioctl_hook(net, cmd, argp);
1137 mutex_unlock(&br_ioctl_mutex);
1138 break;
1139 case SIOCGIFVLAN:
1140 case SIOCSIFVLAN:
1141 err = -ENOPKG;
1142 if (!vlan_ioctl_hook)
1143 request_module("8021q");
1144
1145 mutex_lock(&vlan_ioctl_mutex);
1146 if (vlan_ioctl_hook)
1147 err = vlan_ioctl_hook(net, argp);
1148 mutex_unlock(&vlan_ioctl_mutex);
1149 break;
1150 case SIOCADDDLCI:
1151 case SIOCDELDLCI:
1152 err = -ENOPKG;
1153 if (!dlci_ioctl_hook)
1154 request_module("dlci");
1155
1156 mutex_lock(&dlci_ioctl_mutex);
1157 if (dlci_ioctl_hook)
1158 err = dlci_ioctl_hook(cmd, argp);
1159 mutex_unlock(&dlci_ioctl_mutex);
1160 break;
1161 case SIOCGSKNS:
1162 err = -EPERM;
1163 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1164 break;
1165
1166 err = open_related_ns(&net->ns, get_net_ns);
1167 break;
1168 case SIOCGSTAMP_OLD:
1169 case SIOCGSTAMPNS_OLD:
1170 if (!sock->ops->gettstamp) {
1171 err = -ENOIOCTLCMD;
1172 break;
1173 }
1174 err = sock->ops->gettstamp(sock, argp,
1175 cmd == SIOCGSTAMP_OLD,
1176 !IS_ENABLED(CONFIG_64BIT));
1177 break;
1178 case SIOCGSTAMP_NEW:
1179 case SIOCGSTAMPNS_NEW:
1180 if (!sock->ops->gettstamp) {
1181 err = -ENOIOCTLCMD;
1182 break;
1183 }
1184 err = sock->ops->gettstamp(sock, argp,
1185 cmd == SIOCGSTAMP_NEW,
1186 false);
1187 break;
1188 default:
1189 err = sock_do_ioctl(net, sock, cmd, arg);
1190 break;
1191 }
1192 return err;
1193}
1194
1195/**
1196 * sock_create_lite - creates a socket
1197 * @family: protocol family (AF_INET, ...)
1198 * @type: communication type (SOCK_STREAM, ...)
1199 * @protocol: protocol (0, ...)
1200 * @res: new socket
1201 *
1202 * Creates a new socket and assigns it to @res, passing through LSM.
1203 * The new socket initialization is not complete, see kernel_accept().
1204 * Returns 0 or an error. On failure @res is set to %NULL.
1205 * This function internally uses GFP_KERNEL.
1206 */
1207
1208int sock_create_lite(int family, int type, int protocol, struct socket **res)
1209{
1210 int err;
1211 struct socket *sock = NULL;
1212
1213 err = security_socket_create(family, type, protocol, 1);
1214 if (err)
1215 goto out;
1216
1217 sock = sock_alloc();
1218 if (!sock) {
1219 err = -ENOMEM;
1220 goto out;
1221 }
1222
1223 sock->type = type;
1224 err = security_socket_post_create(sock, family, type, protocol, 1);
1225 if (err)
1226 goto out_release;
1227
1228out:
1229 *res = sock;
1230 return err;
1231out_release:
1232 sock_release(sock);
1233 sock = NULL;
1234 goto out;
1235}
1236EXPORT_SYMBOL(sock_create_lite);
1237
1238/* No kernel lock held - perfect */
1239static __poll_t sock_poll(struct file *file, poll_table *wait)
1240{
1241 struct socket *sock = file->private_data;
1242 __poll_t events = poll_requested_events(wait), flag = 0;
1243
1244 if (!sock->ops->poll)
1245 return 0;
1246
1247 if (sk_can_busy_loop(sock->sk)) {
1248 /* poll once if requested by the syscall */
1249 if (events & POLL_BUSY_LOOP)
1250 sk_busy_loop(sock->sk, 1);
1251
1252 /* if this socket can poll_ll, tell the system call */
1253 flag = POLL_BUSY_LOOP;
1254 }
1255
1256 return sock->ops->poll(file, sock, wait) | flag;
1257}
1258
1259static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1260{
1261 struct socket *sock = file->private_data;
1262
1263 return sock->ops->mmap(file, sock, vma);
1264}
1265
1266static int sock_close(struct inode *inode, struct file *filp)
1267{
1268 __sock_release(SOCKET_I(inode), inode);
1269 return 0;
1270}
1271
1272/*
1273 * Update the socket async list
1274 *
1275 * Fasync_list locking strategy.
1276 *
1277 * 1. fasync_list is modified only under process context socket lock
1278 * i.e. under semaphore.
1279 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1280 * or under socket lock
1281 */
1282
1283static int sock_fasync(int fd, struct file *filp, int on)
1284{
1285 struct socket *sock = filp->private_data;
1286 struct sock *sk = sock->sk;
1287 struct socket_wq *wq = &sock->wq;
1288
1289 if (sk == NULL)
1290 return -EINVAL;
1291
1292 lock_sock(sk);
1293 fasync_helper(fd, filp, on, &wq->fasync_list);
1294
1295 if (!wq->fasync_list)
1296 sock_reset_flag(sk, SOCK_FASYNC);
1297 else
1298 sock_set_flag(sk, SOCK_FASYNC);
1299
1300 release_sock(sk);
1301 return 0;
1302}
1303
1304/* This function may be called only under rcu_lock */
1305
1306int sock_wake_async(struct socket_wq *wq, int how, int band)
1307{
1308 if (!wq || !wq->fasync_list)
1309 return -1;
1310
1311 switch (how) {
1312 case SOCK_WAKE_WAITD:
1313 if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags))
1314 break;
1315 goto call_kill;
1316 case SOCK_WAKE_SPACE:
1317 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags))
1318 break;
1319 /* fall through */
1320 case SOCK_WAKE_IO:
1321call_kill:
1322 kill_fasync(&wq->fasync_list, SIGIO, band);
1323 break;
1324 case SOCK_WAKE_URG:
1325 kill_fasync(&wq->fasync_list, SIGURG, band);
1326 }
1327
1328 return 0;
1329}
1330EXPORT_SYMBOL(sock_wake_async);
1331
1332/**
1333 * __sock_create - creates a socket
1334 * @net: net namespace
1335 * @family: protocol family (AF_INET, ...)
1336 * @type: communication type (SOCK_STREAM, ...)
1337 * @protocol: protocol (0, ...)
1338 * @res: new socket
1339 * @kern: boolean for kernel space sockets
1340 *
1341 * Creates a new socket and assigns it to @res, passing through LSM.
1342 * Returns 0 or an error. On failure @res is set to %NULL. @kern must
1343 * be set to true if the socket resides in kernel space.
1344 * This function internally uses GFP_KERNEL.
1345 */
1346
1347int __sock_create(struct net *net, int family, int type, int protocol,
1348 struct socket **res, int kern)
1349{
1350 int err;
1351 struct socket *sock;
1352 const struct net_proto_family *pf;
1353
1354 /*
1355 * Check protocol is in range
1356 */
1357 if (family < 0 || family >= NPROTO)
1358 return -EAFNOSUPPORT;
1359 if (type < 0 || type >= SOCK_MAX)
1360 return -EINVAL;
1361
1362 /* Compatibility.
1363
1364 This uglymoron is moved from INET layer to here to avoid
1365 deadlock in module load.
1366 */
1367 if (family == PF_INET && type == SOCK_PACKET) {
1368 pr_info_once("%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1369 current->comm);
1370 family = PF_PACKET;
1371 }
1372
1373 err = security_socket_create(family, type, protocol, kern);
1374 if (err)
1375 return err;
1376
1377 /*
1378 * Allocate the socket and allow the family to set things up. if
1379 * the protocol is 0, the family is instructed to select an appropriate
1380 * default.
1381 */
1382 sock = sock_alloc();
1383 if (!sock) {
1384 net_warn_ratelimited("socket: no more sockets\n");
1385 return -ENFILE; /* Not exactly a match, but its the
1386 closest posix thing */
1387 }
1388
1389 sock->type = type;
1390
1391#ifdef CONFIG_MODULES
1392 /* Attempt to load a protocol module if the find failed.
1393 *
1394 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1395 * requested real, full-featured networking support upon configuration.
1396 * Otherwise module support will break!
1397 */
1398 if (rcu_access_pointer(net_families[family]) == NULL)
1399 request_module("net-pf-%d", family);
1400#endif
1401
1402 rcu_read_lock();
1403 pf = rcu_dereference(net_families[family]);
1404 err = -EAFNOSUPPORT;
1405 if (!pf)
1406 goto out_release;
1407
1408 /*
1409 * We will call the ->create function, that possibly is in a loadable
1410 * module, so we have to bump that loadable module refcnt first.
1411 */
1412 if (!try_module_get(pf->owner))
1413 goto out_release;
1414
1415 /* Now protected by module ref count */
1416 rcu_read_unlock();
1417
1418 err = pf->create(net, sock, protocol, kern);
1419 if (err < 0)
1420 goto out_module_put;
1421
1422 /*
1423 * Now to bump the refcnt of the [loadable] module that owns this
1424 * socket at sock_release time we decrement its refcnt.
1425 */
1426 if (!try_module_get(sock->ops->owner))
1427 goto out_module_busy;
1428
1429 /*
1430 * Now that we're done with the ->create function, the [loadable]
1431 * module can have its refcnt decremented
1432 */
1433 module_put(pf->owner);
1434 err = security_socket_post_create(sock, family, type, protocol, kern);
1435 if (err)
1436 goto out_sock_release;
1437 *res = sock;
1438
1439 return 0;
1440
1441out_module_busy:
1442 err = -EAFNOSUPPORT;
1443out_module_put:
1444 sock->ops = NULL;
1445 module_put(pf->owner);
1446out_sock_release:
1447 sock_release(sock);
1448 return err;
1449
1450out_release:
1451 rcu_read_unlock();
1452 goto out_sock_release;
1453}
1454EXPORT_SYMBOL(__sock_create);
1455
1456/**
1457 * sock_create - creates a socket
1458 * @family: protocol family (AF_INET, ...)
1459 * @type: communication type (SOCK_STREAM, ...)
1460 * @protocol: protocol (0, ...)
1461 * @res: new socket
1462 *
1463 * A wrapper around __sock_create().
1464 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1465 */
1466
1467int sock_create(int family, int type, int protocol, struct socket **res)
1468{
1469 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1470}
1471EXPORT_SYMBOL(sock_create);
1472
1473/**
1474 * sock_create_kern - creates a socket (kernel space)
1475 * @net: net namespace
1476 * @family: protocol family (AF_INET, ...)
1477 * @type: communication type (SOCK_STREAM, ...)
1478 * @protocol: protocol (0, ...)
1479 * @res: new socket
1480 *
1481 * A wrapper around __sock_create().
1482 * Returns 0 or an error. This function internally uses GFP_KERNEL.
1483 */
1484
1485int sock_create_kern(struct net *net, int family, int type, int protocol, struct socket **res)
1486{
1487 return __sock_create(net, family, type, protocol, res, 1);
1488}
1489EXPORT_SYMBOL(sock_create_kern);
1490
1491int __sys_socket(int family, int type, int protocol)
1492{
1493 int retval;
1494 struct socket *sock;
1495 int flags;
1496
1497 /* Check the SOCK_* constants for consistency. */
1498 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1499 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1500 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1501 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1502
1503 flags = type & ~SOCK_TYPE_MASK;
1504 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1505 return -EINVAL;
1506 type &= SOCK_TYPE_MASK;
1507
1508 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1509 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1510
1511 retval = sock_create(family, type, protocol, &sock);
1512 if (retval < 0)
1513 return retval;
1514
1515 return sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1516}
1517
1518SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1519{
1520 return __sys_socket(family, type, protocol);
1521}
1522
1523/*
1524 * Create a pair of connected sockets.
1525 */
1526
1527int __sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1528{
1529 struct socket *sock1, *sock2;
1530 int fd1, fd2, err;
1531 struct file *newfile1, *newfile2;
1532 int flags;
1533
1534 flags = type & ~SOCK_TYPE_MASK;
1535 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1536 return -EINVAL;
1537 type &= SOCK_TYPE_MASK;
1538
1539 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1540 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1541
1542 /*
1543 * reserve descriptors and make sure we won't fail
1544 * to return them to userland.
1545 */
1546 fd1 = get_unused_fd_flags(flags);
1547 if (unlikely(fd1 < 0))
1548 return fd1;
1549
1550 fd2 = get_unused_fd_flags(flags);
1551 if (unlikely(fd2 < 0)) {
1552 put_unused_fd(fd1);
1553 return fd2;
1554 }
1555
1556 err = put_user(fd1, &usockvec[0]);
1557 if (err)
1558 goto out;
1559
1560 err = put_user(fd2, &usockvec[1]);
1561 if (err)
1562 goto out;
1563
1564 /*
1565 * Obtain the first socket and check if the underlying protocol
1566 * supports the socketpair call.
1567 */
1568
1569 err = sock_create(family, type, protocol, &sock1);
1570 if (unlikely(err < 0))
1571 goto out;
1572
1573 err = sock_create(family, type, protocol, &sock2);
1574 if (unlikely(err < 0)) {
1575 sock_release(sock1);
1576 goto out;
1577 }
1578
1579 err = security_socket_socketpair(sock1, sock2);
1580 if (unlikely(err)) {
1581 sock_release(sock2);
1582 sock_release(sock1);
1583 goto out;
1584 }
1585
1586 err = sock1->ops->socketpair(sock1, sock2);
1587 if (unlikely(err < 0)) {
1588 sock_release(sock2);
1589 sock_release(sock1);
1590 goto out;
1591 }
1592
1593 newfile1 = sock_alloc_file(sock1, flags, NULL);
1594 if (IS_ERR(newfile1)) {
1595 err = PTR_ERR(newfile1);
1596 sock_release(sock2);
1597 goto out;
1598 }
1599
1600 newfile2 = sock_alloc_file(sock2, flags, NULL);
1601 if (IS_ERR(newfile2)) {
1602 err = PTR_ERR(newfile2);
1603 fput(newfile1);
1604 goto out;
1605 }
1606
1607 audit_fd_pair(fd1, fd2);
1608
1609 fd_install(fd1, newfile1);
1610 fd_install(fd2, newfile2);
1611 return 0;
1612
1613out:
1614 put_unused_fd(fd2);
1615 put_unused_fd(fd1);
1616 return err;
1617}
1618
1619SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1620 int __user *, usockvec)
1621{
1622 return __sys_socketpair(family, type, protocol, usockvec);
1623}
1624
1625/*
1626 * Bind a name to a socket. Nothing much to do here since it's
1627 * the protocol's responsibility to handle the local address.
1628 *
1629 * We move the socket address to kernel space before we call
1630 * the protocol layer (having also checked the address is ok).
1631 */
1632
1633int __sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1634{
1635 struct socket *sock;
1636 struct sockaddr_storage address;
1637 int err, fput_needed;
1638
1639 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1640 if (sock) {
1641 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1642 if (!err) {
1643 err = security_socket_bind(sock,
1644 (struct sockaddr *)&address,
1645 addrlen);
1646 if (!err)
1647 err = sock->ops->bind(sock,
1648 (struct sockaddr *)
1649 &address, addrlen);
1650 }
1651 fput_light(sock->file, fput_needed);
1652 }
1653 return err;
1654}
1655
1656SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1657{
1658 return __sys_bind(fd, umyaddr, addrlen);
1659}
1660
1661/*
1662 * Perform a listen. Basically, we allow the protocol to do anything
1663 * necessary for a listen, and if that works, we mark the socket as
1664 * ready for listening.
1665 */
1666
1667int __sys_listen(int fd, int backlog)
1668{
1669 struct socket *sock;
1670 int err, fput_needed;
1671 int somaxconn;
1672
1673 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1674 if (sock) {
1675 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1676 if ((unsigned int)backlog > somaxconn)
1677 backlog = somaxconn;
1678
1679 err = security_socket_listen(sock, backlog);
1680 if (!err)
1681 err = sock->ops->listen(sock, backlog);
1682
1683 fput_light(sock->file, fput_needed);
1684 }
1685 return err;
1686}
1687
1688SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1689{
1690 return __sys_listen(fd, backlog);
1691}
1692
1693/*
1694 * For accept, we attempt to create a new socket, set up the link
1695 * with the client, wake up the client, then return the new
1696 * connected fd. We collect the address of the connector in kernel
1697 * space and move it to user at the very end. This is unclean because
1698 * we open the socket then return an error.
1699 *
1700 * 1003.1g adds the ability to recvmsg() to query connection pending
1701 * status to recvmsg. We need to add that support in a way thats
1702 * clean when we restructure accept also.
1703 */
1704
1705int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
1706 int __user *upeer_addrlen, int flags)
1707{
1708 struct socket *sock, *newsock;
1709 struct file *newfile;
1710 int err, len, newfd, fput_needed;
1711 struct sockaddr_storage address;
1712
1713 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1714 return -EINVAL;
1715
1716 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1717 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1718
1719 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1720 if (!sock)
1721 goto out;
1722
1723 err = -ENFILE;
1724 newsock = sock_alloc();
1725 if (!newsock)
1726 goto out_put;
1727
1728 newsock->type = sock->type;
1729 newsock->ops = sock->ops;
1730
1731 /*
1732 * We don't need try_module_get here, as the listening socket (sock)
1733 * has the protocol module (sock->ops->owner) held.
1734 */
1735 __module_get(newsock->ops->owner);
1736
1737 newfd = get_unused_fd_flags(flags);
1738 if (unlikely(newfd < 0)) {
1739 err = newfd;
1740 sock_release(newsock);
1741 goto out_put;
1742 }
1743 newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
1744 if (IS_ERR(newfile)) {
1745 err = PTR_ERR(newfile);
1746 put_unused_fd(newfd);
1747 goto out_put;
1748 }
1749
1750 err = security_socket_accept(sock, newsock);
1751 if (err)
1752 goto out_fd;
1753
1754 err = sock->ops->accept(sock, newsock, sock->file->f_flags, 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;
1775
1776out_put:
1777 fput_light(sock->file, fput_needed);
1778out:
1779 return err;
1780out_fd:
1781 fput(newfile);
1782 put_unused_fd(newfd);
1783 goto out_put;
1784}
1785
1786SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1787 int __user *, upeer_addrlen, int, flags)
1788{
1789 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, flags);
1790}
1791
1792SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1793 int __user *, upeer_addrlen)
1794{
1795 return __sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1796}
1797
1798/*
1799 * Attempt to connect to a socket with the server address. The address
1800 * is in user space so we verify it is OK and move it to kernel space.
1801 *
1802 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1803 * break bindings
1804 *
1805 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1806 * other SEQPACKET protocols that take time to connect() as it doesn't
1807 * include the -EINPROGRESS status for such sockets.
1808 */
1809
1810int __sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1811{
1812 struct socket *sock;
1813 struct sockaddr_storage address;
1814 int err, fput_needed;
1815
1816 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1817 if (!sock)
1818 goto out;
1819 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1820 if (err < 0)
1821 goto out_put;
1822
1823 err =
1824 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1825 if (err)
1826 goto out_put;
1827
1828 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1829 sock->file->f_flags);
1830out_put:
1831 fput_light(sock->file, fput_needed);
1832out:
1833 return err;
1834}
1835
1836SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1837 int, addrlen)
1838{
1839 return __sys_connect(fd, uservaddr, addrlen);
1840}
1841
1842/*
1843 * Get the local address ('name') of a socket object. Move the obtained
1844 * name to user space.
1845 */
1846
1847int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
1848 int __user *usockaddr_len)
1849{
1850 struct socket *sock;
1851 struct sockaddr_storage address;
1852 int err, fput_needed;
1853
1854 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1855 if (!sock)
1856 goto out;
1857
1858 err = security_socket_getsockname(sock);
1859 if (err)
1860 goto out_put;
1861
1862 err = sock->ops->getname(sock, (struct sockaddr *)&address, 0);
1863 if (err < 0)
1864 goto out_put;
1865 /* "err" is actually length in this case */
1866 err = move_addr_to_user(&address, err, usockaddr, usockaddr_len);
1867
1868out_put:
1869 fput_light(sock->file, fput_needed);
1870out:
1871 return err;
1872}
1873
1874SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1875 int __user *, usockaddr_len)
1876{
1877 return __sys_getsockname(fd, usockaddr, usockaddr_len);
1878}
1879
1880/*
1881 * Get the remote address ('name') of a socket object. Move the obtained
1882 * name to user space.
1883 */
1884
1885int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
1886 int __user *usockaddr_len)
1887{
1888 struct socket *sock;
1889 struct sockaddr_storage address;
1890 int err, fput_needed;
1891
1892 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1893 if (sock != NULL) {
1894 err = security_socket_getpeername(sock);
1895 if (err) {
1896 fput_light(sock->file, fput_needed);
1897 return err;
1898 }
1899
1900 err = sock->ops->getname(sock, (struct sockaddr *)&address, 1);
1901 if (err >= 0)
1902 /* "err" is actually length in this case */
1903 err = move_addr_to_user(&address, err, usockaddr,
1904 usockaddr_len);
1905 fput_light(sock->file, fput_needed);
1906 }
1907 return err;
1908}
1909
1910SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1911 int __user *, usockaddr_len)
1912{
1913 return __sys_getpeername(fd, usockaddr, usockaddr_len);
1914}
1915
1916/*
1917 * Send a datagram to a given address. We move the address into kernel
1918 * space and check the user space data area is readable before invoking
1919 * the protocol.
1920 */
1921int __sys_sendto(int fd, void __user *buff, size_t len, unsigned int flags,
1922 struct sockaddr __user *addr, int addr_len)
1923{
1924 struct socket *sock;
1925 struct sockaddr_storage address;
1926 int err;
1927 struct msghdr msg;
1928 struct iovec iov;
1929 int fput_needed;
1930
1931 err = import_single_range(WRITE, buff, len, &iov, &msg.msg_iter);
1932 if (unlikely(err))
1933 return err;
1934 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1935 if (!sock)
1936 goto out;
1937
1938 msg.msg_name = NULL;
1939 msg.msg_control = NULL;
1940 msg.msg_controllen = 0;
1941 msg.msg_namelen = 0;
1942 if (addr) {
1943 err = move_addr_to_kernel(addr, addr_len, &address);
1944 if (err < 0)
1945 goto out_put;
1946 msg.msg_name = (struct sockaddr *)&address;
1947 msg.msg_namelen = addr_len;
1948 }
1949 if (sock->file->f_flags & O_NONBLOCK)
1950 flags |= MSG_DONTWAIT;
1951 msg.msg_flags = flags;
1952 err = sock_sendmsg(sock, &msg);
1953
1954out_put:
1955 fput_light(sock->file, fput_needed);
1956out:
1957 return err;
1958}
1959
1960SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1961 unsigned int, flags, struct sockaddr __user *, addr,
1962 int, addr_len)
1963{
1964 return __sys_sendto(fd, buff, len, flags, addr, addr_len);
1965}
1966
1967/*
1968 * Send a datagram down a socket.
1969 */
1970
1971SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1972 unsigned int, flags)
1973{
1974 return __sys_sendto(fd, buff, len, flags, NULL, 0);
1975}
1976
1977/*
1978 * Receive a frame from the socket and optionally record the address of the
1979 * sender. We verify the buffers are writable and if needed move the
1980 * sender address from kernel to user space.
1981 */
1982int __sys_recvfrom(int fd, void __user *ubuf, size_t size, unsigned int flags,
1983 struct sockaddr __user *addr, int __user *addr_len)
1984{
1985 struct socket *sock;
1986 struct iovec iov;
1987 struct msghdr msg;
1988 struct sockaddr_storage address;
1989 int err, err2;
1990 int fput_needed;
1991
1992 err = import_single_range(READ, ubuf, size, &iov, &msg.msg_iter);
1993 if (unlikely(err))
1994 return err;
1995 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1996 if (!sock)
1997 goto out;
1998
1999 msg.msg_control = NULL;
2000 msg.msg_controllen = 0;
2001 /* Save some cycles and don't copy the address if not needed */
2002 msg.msg_name = addr ? (struct sockaddr *)&address : NULL;
2003 /* We assume all kernel code knows the size of sockaddr_storage */
2004 msg.msg_namelen = 0;
2005 msg.msg_iocb = NULL;
2006 msg.msg_flags = 0;
2007 if (sock->file->f_flags & O_NONBLOCK)
2008 flags |= MSG_DONTWAIT;
2009 err = sock_recvmsg(sock, &msg, flags);
2010
2011 if (err >= 0 && addr != NULL) {
2012 err2 = move_addr_to_user(&address,
2013 msg.msg_namelen, addr, addr_len);
2014 if (err2 < 0)
2015 err = err2;
2016 }
2017
2018 fput_light(sock->file, fput_needed);
2019out:
2020 return err;
2021}
2022
2023SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
2024 unsigned int, flags, struct sockaddr __user *, addr,
2025 int __user *, addr_len)
2026{
2027 return __sys_recvfrom(fd, ubuf, size, flags, addr, addr_len);
2028}
2029
2030/*
2031 * Receive a datagram from a socket.
2032 */
2033
2034SYSCALL_DEFINE4(recv, int, fd, void __user *, ubuf, size_t, size,
2035 unsigned int, flags)
2036{
2037 return __sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
2038}
2039
2040/*
2041 * Set a socket option. Because we don't know the option lengths we have
2042 * to pass the user mode parameter for the protocols to sort out.
2043 */
2044
2045static int __sys_setsockopt(int fd, int level, int optname,
2046 char __user *optval, int optlen)
2047{
2048 mm_segment_t oldfs = get_fs();
2049 char *kernel_optval = NULL;
2050 int err, fput_needed;
2051 struct socket *sock;
2052
2053 if (optlen < 0)
2054 return -EINVAL;
2055
2056 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2057 if (sock != NULL) {
2058 err = security_socket_setsockopt(sock, level, optname);
2059 if (err)
2060 goto out_put;
2061
2062 err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level,
2063 &optname, optval, &optlen,
2064 &kernel_optval);
2065
2066 if (err < 0) {
2067 goto out_put;
2068 } else if (err > 0) {
2069 err = 0;
2070 goto out_put;
2071 }
2072
2073 if (kernel_optval) {
2074 set_fs(KERNEL_DS);
2075 optval = (char __user __force *)kernel_optval;
2076 }
2077
2078 if (level == SOL_SOCKET)
2079 err =
2080 sock_setsockopt(sock, level, optname, optval,
2081 optlen);
2082 else
2083 err =
2084 sock->ops->setsockopt(sock, level, optname, optval,
2085 optlen);
2086
2087 if (kernel_optval) {
2088 set_fs(oldfs);
2089 kfree(kernel_optval);
2090 }
2091out_put:
2092 fput_light(sock->file, fput_needed);
2093 }
2094 return err;
2095}
2096
2097SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
2098 char __user *, optval, int, optlen)
2099{
2100 return __sys_setsockopt(fd, level, optname, optval, optlen);
2101}
2102
2103/*
2104 * Get a socket option. Because we don't know the option lengths we have
2105 * to pass a user mode parameter for the protocols to sort out.
2106 */
2107
2108static int __sys_getsockopt(int fd, int level, int optname,
2109 char __user *optval, int __user *optlen)
2110{
2111 int err, fput_needed;
2112 struct socket *sock;
2113 int max_optlen;
2114
2115 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2116 if (sock != NULL) {
2117 err = security_socket_getsockopt(sock, level, optname);
2118 if (err)
2119 goto out_put;
2120
2121 max_optlen = BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen);
2122
2123 if (level == SOL_SOCKET)
2124 err =
2125 sock_getsockopt(sock, level, optname, optval,
2126 optlen);
2127 else
2128 err =
2129 sock->ops->getsockopt(sock, level, optname, optval,
2130 optlen);
2131
2132 err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
2133 optval, optlen,
2134 max_optlen, err);
2135out_put:
2136 fput_light(sock->file, fput_needed);
2137 }
2138 return err;
2139}
2140
2141SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
2142 char __user *, optval, int __user *, optlen)
2143{
2144 return __sys_getsockopt(fd, level, optname, optval, optlen);
2145}
2146
2147/*
2148 * Shutdown a socket.
2149 */
2150
2151int __sys_shutdown(int fd, int how)
2152{
2153 int err, fput_needed;
2154 struct socket *sock;
2155
2156 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2157 if (sock != NULL) {
2158 err = security_socket_shutdown(sock, how);
2159 if (!err)
2160 err = sock->ops->shutdown(sock, how);
2161 fput_light(sock->file, fput_needed);
2162 }
2163 return err;
2164}
2165
2166SYSCALL_DEFINE2(shutdown, int, fd, int, how)
2167{
2168 return __sys_shutdown(fd, how);
2169}
2170
2171/* A couple of helpful macros for getting the address of the 32/64 bit
2172 * fields which are the same type (int / unsigned) on our platforms.
2173 */
2174#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
2175#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
2176#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
2177
2178struct used_address {
2179 struct sockaddr_storage name;
2180 unsigned int name_len;
2181};
2182
2183static int copy_msghdr_from_user(struct msghdr *kmsg,
2184 struct user_msghdr __user *umsg,
2185 struct sockaddr __user **save_addr,
2186 struct iovec **iov)
2187{
2188 struct user_msghdr msg;
2189 ssize_t err;
2190
2191 if (copy_from_user(&msg, umsg, sizeof(*umsg)))
2192 return -EFAULT;
2193
2194 kmsg->msg_control = (void __force *)msg.msg_control;
2195 kmsg->msg_controllen = msg.msg_controllen;
2196 kmsg->msg_flags = msg.msg_flags;
2197
2198 kmsg->msg_namelen = msg.msg_namelen;
2199 if (!msg.msg_name)
2200 kmsg->msg_namelen = 0;
2201
2202 if (kmsg->msg_namelen < 0)
2203 return -EINVAL;
2204
2205 if (kmsg->msg_namelen > sizeof(struct sockaddr_storage))
2206 kmsg->msg_namelen = sizeof(struct sockaddr_storage);
2207
2208 if (save_addr)
2209 *save_addr = msg.msg_name;
2210
2211 if (msg.msg_name && kmsg->msg_namelen) {
2212 if (!save_addr) {
2213 err = move_addr_to_kernel(msg.msg_name,
2214 kmsg->msg_namelen,
2215 kmsg->msg_name);
2216 if (err < 0)
2217 return err;
2218 }
2219 } else {
2220 kmsg->msg_name = NULL;
2221 kmsg->msg_namelen = 0;
2222 }
2223
2224 if (msg.msg_iovlen > UIO_MAXIOV)
2225 return -EMSGSIZE;
2226
2227 kmsg->msg_iocb = NULL;
2228
2229 err = import_iovec(save_addr ? READ : WRITE,
2230 msg.msg_iov, msg.msg_iovlen,
2231 UIO_FASTIOV, iov, &kmsg->msg_iter);
2232 return err < 0 ? err : 0;
2233}
2234
2235static int ___sys_sendmsg(struct socket *sock, struct user_msghdr __user *msg,
2236 struct msghdr *msg_sys, unsigned int flags,
2237 struct used_address *used_address,
2238 unsigned int allowed_msghdr_flags)
2239{
2240 struct compat_msghdr __user *msg_compat =
2241 (struct compat_msghdr __user *)msg;
2242 struct sockaddr_storage address;
2243 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
2244 unsigned char ctl[sizeof(struct cmsghdr) + 20]
2245 __aligned(sizeof(__kernel_size_t));
2246 /* 20 is size of ipv6_pktinfo */
2247 unsigned char *ctl_buf = ctl;
2248 int ctl_len;
2249 ssize_t err;
2250
2251 msg_sys->msg_name = &address;
2252
2253 if (MSG_CMSG_COMPAT & flags)
2254 err = get_compat_msghdr(msg_sys, msg_compat, NULL, &iov);
2255 else
2256 err = copy_msghdr_from_user(msg_sys, msg, NULL, &iov);
2257 if (err < 0)
2258 return err;
2259
2260 err = -ENOBUFS;
2261
2262 if (msg_sys->msg_controllen > INT_MAX)
2263 goto out_freeiov;
2264 flags |= (msg_sys->msg_flags & allowed_msghdr_flags);
2265 ctl_len = msg_sys->msg_controllen;
2266 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
2267 err =
2268 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
2269 sizeof(ctl));
2270 if (err)
2271 goto out_freeiov;
2272 ctl_buf = msg_sys->msg_control;
2273 ctl_len = msg_sys->msg_controllen;
2274 } else if (ctl_len) {
2275 BUILD_BUG_ON(sizeof(struct cmsghdr) !=
2276 CMSG_ALIGN(sizeof(struct cmsghdr)));
2277 if (ctl_len > sizeof(ctl)) {
2278 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
2279 if (ctl_buf == NULL)
2280 goto out_freeiov;
2281 }
2282 err = -EFAULT;
2283 /*
2284 * Careful! Before this, msg_sys->msg_control contains a user pointer.
2285 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
2286 * checking falls down on this.
2287 */
2288 if (copy_from_user(ctl_buf,
2289 (void __user __force *)msg_sys->msg_control,
2290 ctl_len))
2291 goto out_freectl;
2292 msg_sys->msg_control = ctl_buf;
2293 }
2294 msg_sys->msg_flags = flags;
2295
2296 if (sock->file->f_flags & O_NONBLOCK)
2297 msg_sys->msg_flags |= MSG_DONTWAIT;
2298 /*
2299 * If this is sendmmsg() and current destination address is same as
2300 * previously succeeded address, omit asking LSM's decision.
2301 * used_address->name_len is initialized to UINT_MAX so that the first
2302 * destination address never matches.
2303 */
2304 if (used_address && msg_sys->msg_name &&
2305 used_address->name_len == msg_sys->msg_namelen &&
2306 !memcmp(&used_address->name, msg_sys->msg_name,
2307 used_address->name_len)) {
2308 err = sock_sendmsg_nosec(sock, msg_sys);
2309 goto out_freectl;
2310 }
2311 err = sock_sendmsg(sock, msg_sys);
2312 /*
2313 * If this is sendmmsg() and sending to current destination address was
2314 * successful, remember it.
2315 */
2316 if (used_address && err >= 0) {
2317 used_address->name_len = msg_sys->msg_namelen;
2318 if (msg_sys->msg_name)
2319 memcpy(&used_address->name, msg_sys->msg_name,
2320 used_address->name_len);
2321 }
2322
2323out_freectl:
2324 if (ctl_buf != ctl)
2325 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2326out_freeiov:
2327 kfree(iov);
2328 return err;
2329}
2330
2331/*
2332 * BSD sendmsg interface
2333 */
2334long __sys_sendmsg_sock(struct socket *sock, struct user_msghdr __user *msg,
2335 unsigned int flags)
2336{
2337 struct msghdr msg_sys;
2338
2339 return ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2340}
2341
2342long __sys_sendmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2343 bool forbid_cmsg_compat)
2344{
2345 int fput_needed, err;
2346 struct msghdr msg_sys;
2347 struct socket *sock;
2348
2349 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2350 return -EINVAL;
2351
2352 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2353 if (!sock)
2354 goto out;
2355
2356 err = ___sys_sendmsg(sock, msg, &msg_sys, flags, NULL, 0);
2357
2358 fput_light(sock->file, fput_needed);
2359out:
2360 return err;
2361}
2362
2363SYSCALL_DEFINE3(sendmsg, int, fd, struct user_msghdr __user *, msg, unsigned int, flags)
2364{
2365 return __sys_sendmsg(fd, msg, flags, true);
2366}
2367
2368/*
2369 * Linux sendmmsg interface
2370 */
2371
2372int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2373 unsigned int flags, bool forbid_cmsg_compat)
2374{
2375 int fput_needed, err, datagrams;
2376 struct socket *sock;
2377 struct mmsghdr __user *entry;
2378 struct compat_mmsghdr __user *compat_entry;
2379 struct msghdr msg_sys;
2380 struct used_address used_address;
2381 unsigned int oflags = flags;
2382
2383 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2384 return -EINVAL;
2385
2386 if (vlen > UIO_MAXIOV)
2387 vlen = UIO_MAXIOV;
2388
2389 datagrams = 0;
2390
2391 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2392 if (!sock)
2393 return err;
2394
2395 used_address.name_len = UINT_MAX;
2396 entry = mmsg;
2397 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2398 err = 0;
2399 flags |= MSG_BATCH;
2400
2401 while (datagrams < vlen) {
2402 if (datagrams == vlen - 1)
2403 flags = oflags;
2404
2405 if (MSG_CMSG_COMPAT & flags) {
2406 err = ___sys_sendmsg(sock, (struct user_msghdr __user *)compat_entry,
2407 &msg_sys, flags, &used_address, MSG_EOR);
2408 if (err < 0)
2409 break;
2410 err = __put_user(err, &compat_entry->msg_len);
2411 ++compat_entry;
2412 } else {
2413 err = ___sys_sendmsg(sock,
2414 (struct user_msghdr __user *)entry,
2415 &msg_sys, flags, &used_address, MSG_EOR);
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 if (msg_data_left(&msg_sys))
2426 break;
2427 cond_resched();
2428 }
2429
2430 fput_light(sock->file, fput_needed);
2431
2432 /* We only return an error if no datagrams were able to be sent */
2433 if (datagrams != 0)
2434 return datagrams;
2435
2436 return err;
2437}
2438
2439SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2440 unsigned int, vlen, unsigned int, flags)
2441{
2442 return __sys_sendmmsg(fd, mmsg, vlen, flags, true);
2443}
2444
2445static int ___sys_recvmsg(struct socket *sock, struct user_msghdr __user *msg,
2446 struct msghdr *msg_sys, unsigned int flags, int nosec)
2447{
2448 struct compat_msghdr __user *msg_compat =
2449 (struct compat_msghdr __user *)msg;
2450 struct iovec iovstack[UIO_FASTIOV];
2451 struct iovec *iov = iovstack;
2452 unsigned long cmsg_ptr;
2453 int len;
2454 ssize_t err;
2455
2456 /* kernel mode address */
2457 struct sockaddr_storage addr;
2458
2459 /* user mode address pointers */
2460 struct sockaddr __user *uaddr;
2461 int __user *uaddr_len = COMPAT_NAMELEN(msg);
2462
2463 msg_sys->msg_name = &addr;
2464
2465 if (MSG_CMSG_COMPAT & flags)
2466 err = get_compat_msghdr(msg_sys, msg_compat, &uaddr, &iov);
2467 else
2468 err = copy_msghdr_from_user(msg_sys, msg, &uaddr, &iov);
2469 if (err < 0)
2470 return err;
2471
2472 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2473 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2474
2475 /* We assume all kernel code knows the size of sockaddr_storage */
2476 msg_sys->msg_namelen = 0;
2477
2478 if (sock->file->f_flags & O_NONBLOCK)
2479 flags |= MSG_DONTWAIT;
2480 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys, flags);
2481 if (err < 0)
2482 goto out_freeiov;
2483 len = err;
2484
2485 if (uaddr != NULL) {
2486 err = move_addr_to_user(&addr,
2487 msg_sys->msg_namelen, uaddr,
2488 uaddr_len);
2489 if (err < 0)
2490 goto out_freeiov;
2491 }
2492 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2493 COMPAT_FLAGS(msg));
2494 if (err)
2495 goto out_freeiov;
2496 if (MSG_CMSG_COMPAT & flags)
2497 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2498 &msg_compat->msg_controllen);
2499 else
2500 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2501 &msg->msg_controllen);
2502 if (err)
2503 goto out_freeiov;
2504 err = len;
2505
2506out_freeiov:
2507 kfree(iov);
2508 return err;
2509}
2510
2511/*
2512 * BSD recvmsg interface
2513 */
2514
2515long __sys_recvmsg_sock(struct socket *sock, struct user_msghdr __user *msg,
2516 unsigned int flags)
2517{
2518 struct msghdr msg_sys;
2519
2520 return ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2521}
2522
2523long __sys_recvmsg(int fd, struct user_msghdr __user *msg, unsigned int flags,
2524 bool forbid_cmsg_compat)
2525{
2526 int fput_needed, err;
2527 struct msghdr msg_sys;
2528 struct socket *sock;
2529
2530 if (forbid_cmsg_compat && (flags & MSG_CMSG_COMPAT))
2531 return -EINVAL;
2532
2533 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2534 if (!sock)
2535 goto out;
2536
2537 err = ___sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2538
2539 fput_light(sock->file, fput_needed);
2540out:
2541 return err;
2542}
2543
2544SYSCALL_DEFINE3(recvmsg, int, fd, struct user_msghdr __user *, msg,
2545 unsigned int, flags)
2546{
2547 return __sys_recvmsg(fd, msg, flags, true);
2548}
2549
2550/*
2551 * Linux recvmmsg interface
2552 */
2553
2554static int do_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2555 unsigned int vlen, unsigned int flags,
2556 struct timespec64 *timeout)
2557{
2558 int fput_needed, err, datagrams;
2559 struct socket *sock;
2560 struct mmsghdr __user *entry;
2561 struct compat_mmsghdr __user *compat_entry;
2562 struct msghdr msg_sys;
2563 struct timespec64 end_time;
2564 struct timespec64 timeout64;
2565
2566 if (timeout &&
2567 poll_select_set_timeout(&end_time, timeout->tv_sec,
2568 timeout->tv_nsec))
2569 return -EINVAL;
2570
2571 datagrams = 0;
2572
2573 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2574 if (!sock)
2575 return err;
2576
2577 if (likely(!(flags & MSG_ERRQUEUE))) {
2578 err = sock_error(sock->sk);
2579 if (err) {
2580 datagrams = err;
2581 goto out_put;
2582 }
2583 }
2584
2585 entry = mmsg;
2586 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2587
2588 while (datagrams < vlen) {
2589 /*
2590 * No need to ask LSM for more than the first datagram.
2591 */
2592 if (MSG_CMSG_COMPAT & flags) {
2593 err = ___sys_recvmsg(sock, (struct user_msghdr __user *)compat_entry,
2594 &msg_sys, flags & ~MSG_WAITFORONE,
2595 datagrams);
2596 if (err < 0)
2597 break;
2598 err = __put_user(err, &compat_entry->msg_len);
2599 ++compat_entry;
2600 } else {
2601 err = ___sys_recvmsg(sock,
2602 (struct user_msghdr __user *)entry,
2603 &msg_sys, flags & ~MSG_WAITFORONE,
2604 datagrams);
2605 if (err < 0)
2606 break;
2607 err = put_user(err, &entry->msg_len);
2608 ++entry;
2609 }
2610
2611 if (err)
2612 break;
2613 ++datagrams;
2614
2615 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2616 if (flags & MSG_WAITFORONE)
2617 flags |= MSG_DONTWAIT;
2618
2619 if (timeout) {
2620 ktime_get_ts64(&timeout64);
2621 *timeout = timespec64_sub(end_time, timeout64);
2622 if (timeout->tv_sec < 0) {
2623 timeout->tv_sec = timeout->tv_nsec = 0;
2624 break;
2625 }
2626
2627 /* Timeout, return less than vlen datagrams */
2628 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2629 break;
2630 }
2631
2632 /* Out of band data, return right away */
2633 if (msg_sys.msg_flags & MSG_OOB)
2634 break;
2635 cond_resched();
2636 }
2637
2638 if (err == 0)
2639 goto out_put;
2640
2641 if (datagrams == 0) {
2642 datagrams = err;
2643 goto out_put;
2644 }
2645
2646 /*
2647 * We may return less entries than requested (vlen) if the
2648 * sock is non block and there aren't enough datagrams...
2649 */
2650 if (err != -EAGAIN) {
2651 /*
2652 * ... or if recvmsg returns an error after we
2653 * received some datagrams, where we record the
2654 * error to return on the next call or if the
2655 * app asks about it using getsockopt(SO_ERROR).
2656 */
2657 sock->sk->sk_err = -err;
2658 }
2659out_put:
2660 fput_light(sock->file, fput_needed);
2661
2662 return datagrams;
2663}
2664
2665int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg,
2666 unsigned int vlen, unsigned int flags,
2667 struct __kernel_timespec __user *timeout,
2668 struct old_timespec32 __user *timeout32)
2669{
2670 int datagrams;
2671 struct timespec64 timeout_sys;
2672
2673 if (timeout && get_timespec64(&timeout_sys, timeout))
2674 return -EFAULT;
2675
2676 if (timeout32 && get_old_timespec32(&timeout_sys, timeout32))
2677 return -EFAULT;
2678
2679 if (!timeout && !timeout32)
2680 return do_recvmmsg(fd, mmsg, vlen, flags, NULL);
2681
2682 datagrams = do_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2683
2684 if (datagrams <= 0)
2685 return datagrams;
2686
2687 if (timeout && put_timespec64(&timeout_sys, timeout))
2688 datagrams = -EFAULT;
2689
2690 if (timeout32 && put_old_timespec32(&timeout_sys, timeout32))
2691 datagrams = -EFAULT;
2692
2693 return datagrams;
2694}
2695
2696SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2697 unsigned int, vlen, unsigned int, flags,
2698 struct __kernel_timespec __user *, timeout)
2699{
2700 if (flags & MSG_CMSG_COMPAT)
2701 return -EINVAL;
2702
2703 return __sys_recvmmsg(fd, mmsg, vlen, flags, timeout, NULL);
2704}
2705
2706#ifdef CONFIG_COMPAT_32BIT_TIME
2707SYSCALL_DEFINE5(recvmmsg_time32, int, fd, struct mmsghdr __user *, mmsg,
2708 unsigned int, vlen, unsigned int, flags,
2709 struct old_timespec32 __user *, timeout)
2710{
2711 if (flags & MSG_CMSG_COMPAT)
2712 return -EINVAL;
2713
2714 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL, timeout);
2715}
2716#endif
2717
2718#ifdef __ARCH_WANT_SYS_SOCKETCALL
2719/* Argument list sizes for sys_socketcall */
2720#define AL(x) ((x) * sizeof(unsigned long))
2721static const unsigned char nargs[21] = {
2722 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2723 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2724 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2725 AL(4), AL(5), AL(4)
2726};
2727
2728#undef AL
2729
2730/*
2731 * System call vectors.
2732 *
2733 * Argument checking cleaned up. Saved 20% in size.
2734 * This function doesn't need to set the kernel lock because
2735 * it is set by the callees.
2736 */
2737
2738SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2739{
2740 unsigned long a[AUDITSC_ARGS];
2741 unsigned long a0, a1;
2742 int err;
2743 unsigned int len;
2744
2745 if (call < 1 || call > SYS_SENDMMSG)
2746 return -EINVAL;
2747 call = array_index_nospec(call, SYS_SENDMMSG + 1);
2748
2749 len = nargs[call];
2750 if (len > sizeof(a))
2751 return -EINVAL;
2752
2753 /* copy_from_user should be SMP safe. */
2754 if (copy_from_user(a, args, len))
2755 return -EFAULT;
2756
2757 err = audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2758 if (err)
2759 return err;
2760
2761 a0 = a[0];
2762 a1 = a[1];
2763
2764 switch (call) {
2765 case SYS_SOCKET:
2766 err = __sys_socket(a0, a1, a[2]);
2767 break;
2768 case SYS_BIND:
2769 err = __sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2770 break;
2771 case SYS_CONNECT:
2772 err = __sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2773 break;
2774 case SYS_LISTEN:
2775 err = __sys_listen(a0, a1);
2776 break;
2777 case SYS_ACCEPT:
2778 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2779 (int __user *)a[2], 0);
2780 break;
2781 case SYS_GETSOCKNAME:
2782 err =
2783 __sys_getsockname(a0, (struct sockaddr __user *)a1,
2784 (int __user *)a[2]);
2785 break;
2786 case SYS_GETPEERNAME:
2787 err =
2788 __sys_getpeername(a0, (struct sockaddr __user *)a1,
2789 (int __user *)a[2]);
2790 break;
2791 case SYS_SOCKETPAIR:
2792 err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2793 break;
2794 case SYS_SEND:
2795 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2796 NULL, 0);
2797 break;
2798 case SYS_SENDTO:
2799 err = __sys_sendto(a0, (void __user *)a1, a[2], a[3],
2800 (struct sockaddr __user *)a[4], a[5]);
2801 break;
2802 case SYS_RECV:
2803 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2804 NULL, NULL);
2805 break;
2806 case SYS_RECVFROM:
2807 err = __sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2808 (struct sockaddr __user *)a[4],
2809 (int __user *)a[5]);
2810 break;
2811 case SYS_SHUTDOWN:
2812 err = __sys_shutdown(a0, a1);
2813 break;
2814 case SYS_SETSOCKOPT:
2815 err = __sys_setsockopt(a0, a1, a[2], (char __user *)a[3],
2816 a[4]);
2817 break;
2818 case SYS_GETSOCKOPT:
2819 err =
2820 __sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2821 (int __user *)a[4]);
2822 break;
2823 case SYS_SENDMSG:
2824 err = __sys_sendmsg(a0, (struct user_msghdr __user *)a1,
2825 a[2], true);
2826 break;
2827 case SYS_SENDMMSG:
2828 err = __sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2],
2829 a[3], true);
2830 break;
2831 case SYS_RECVMSG:
2832 err = __sys_recvmsg(a0, (struct user_msghdr __user *)a1,
2833 a[2], true);
2834 break;
2835 case SYS_RECVMMSG:
2836 if (IS_ENABLED(CONFIG_64BIT) || !IS_ENABLED(CONFIG_64BIT_TIME))
2837 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2838 a[2], a[3],
2839 (struct __kernel_timespec __user *)a[4],
2840 NULL);
2841 else
2842 err = __sys_recvmmsg(a0, (struct mmsghdr __user *)a1,
2843 a[2], a[3], NULL,
2844 (struct old_timespec32 __user *)a[4]);
2845 break;
2846 case SYS_ACCEPT4:
2847 err = __sys_accept4(a0, (struct sockaddr __user *)a1,
2848 (int __user *)a[2], a[3]);
2849 break;
2850 default:
2851 err = -EINVAL;
2852 break;
2853 }
2854 return err;
2855}
2856
2857#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2858
2859/**
2860 * sock_register - add a socket protocol handler
2861 * @ops: description of protocol
2862 *
2863 * This function is called by a protocol handler that wants to
2864 * advertise its address family, and have it linked into the
2865 * socket interface. The value ops->family corresponds to the
2866 * socket system call protocol family.
2867 */
2868int sock_register(const struct net_proto_family *ops)
2869{
2870 int err;
2871
2872 if (ops->family >= NPROTO) {
2873 pr_crit("protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2874 return -ENOBUFS;
2875 }
2876
2877 spin_lock(&net_family_lock);
2878 if (rcu_dereference_protected(net_families[ops->family],
2879 lockdep_is_held(&net_family_lock)))
2880 err = -EEXIST;
2881 else {
2882 rcu_assign_pointer(net_families[ops->family], ops);
2883 err = 0;
2884 }
2885 spin_unlock(&net_family_lock);
2886
2887 pr_info("NET: Registered protocol family %d\n", ops->family);
2888 return err;
2889}
2890EXPORT_SYMBOL(sock_register);
2891
2892/**
2893 * sock_unregister - remove a protocol handler
2894 * @family: protocol family to remove
2895 *
2896 * This function is called by a protocol handler that wants to
2897 * remove its address family, and have it unlinked from the
2898 * new socket creation.
2899 *
2900 * If protocol handler is a module, then it can use module reference
2901 * counts to protect against new references. If protocol handler is not
2902 * a module then it needs to provide its own protection in
2903 * the ops->create routine.
2904 */
2905void sock_unregister(int family)
2906{
2907 BUG_ON(family < 0 || family >= NPROTO);
2908
2909 spin_lock(&net_family_lock);
2910 RCU_INIT_POINTER(net_families[family], NULL);
2911 spin_unlock(&net_family_lock);
2912
2913 synchronize_rcu();
2914
2915 pr_info("NET: Unregistered protocol family %d\n", family);
2916}
2917EXPORT_SYMBOL(sock_unregister);
2918
2919bool sock_is_registered(int family)
2920{
2921 return family < NPROTO && rcu_access_pointer(net_families[family]);
2922}
2923
2924static int __init sock_init(void)
2925{
2926 int err;
2927 /*
2928 * Initialize the network sysctl infrastructure.
2929 */
2930 err = net_sysctl_init();
2931 if (err)
2932 goto out;
2933
2934 /*
2935 * Initialize skbuff SLAB cache
2936 */
2937 skb_init();
2938
2939 /*
2940 * Initialize the protocols module.
2941 */
2942
2943 init_inodecache();
2944
2945 err = register_filesystem(&sock_fs_type);
2946 if (err)
2947 goto out_fs;
2948 sock_mnt = kern_mount(&sock_fs_type);
2949 if (IS_ERR(sock_mnt)) {
2950 err = PTR_ERR(sock_mnt);
2951 goto out_mount;
2952 }
2953
2954 /* The real protocol initialization is performed in later initcalls.
2955 */
2956
2957#ifdef CONFIG_NETFILTER
2958 err = netfilter_init();
2959 if (err)
2960 goto out;
2961#endif
2962
2963 ptp_classifier_init();
2964
2965out:
2966 return err;
2967
2968out_mount:
2969 unregister_filesystem(&sock_fs_type);
2970out_fs:
2971 goto out;
2972}
2973
2974core_initcall(sock_init); /* early initcall */
2975
2976#ifdef CONFIG_PROC_FS
2977void socket_seq_show(struct seq_file *seq)
2978{
2979 seq_printf(seq, "sockets: used %d\n",
2980 sock_inuse_get(seq->private));
2981}
2982#endif /* CONFIG_PROC_FS */
2983
2984#ifdef CONFIG_COMPAT
2985static int compat_dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2986{
2987 struct compat_ifconf ifc32;
2988 struct ifconf ifc;
2989 int err;
2990
2991 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2992 return -EFAULT;
2993
2994 ifc.ifc_len = ifc32.ifc_len;
2995 ifc.ifc_req = compat_ptr(ifc32.ifcbuf);
2996
2997 rtnl_lock();
2998 err = dev_ifconf(net, &ifc, sizeof(struct compat_ifreq));
2999 rtnl_unlock();
3000 if (err)
3001 return err;
3002
3003 ifc32.ifc_len = ifc.ifc_len;
3004 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
3005 return -EFAULT;
3006
3007 return 0;
3008}
3009
3010static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
3011{
3012 struct compat_ethtool_rxnfc __user *compat_rxnfc;
3013 bool convert_in = false, convert_out = false;
3014 size_t buf_size = 0;
3015 struct ethtool_rxnfc __user *rxnfc = NULL;
3016 struct ifreq ifr;
3017 u32 rule_cnt = 0, actual_rule_cnt;
3018 u32 ethcmd;
3019 u32 data;
3020 int ret;
3021
3022 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
3023 return -EFAULT;
3024
3025 compat_rxnfc = compat_ptr(data);
3026
3027 if (get_user(ethcmd, &compat_rxnfc->cmd))
3028 return -EFAULT;
3029
3030 /* Most ethtool structures are defined without padding.
3031 * Unfortunately struct ethtool_rxnfc is an exception.
3032 */
3033 switch (ethcmd) {
3034 default:
3035 break;
3036 case ETHTOOL_GRXCLSRLALL:
3037 /* Buffer size is variable */
3038 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
3039 return -EFAULT;
3040 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
3041 return -ENOMEM;
3042 buf_size += rule_cnt * sizeof(u32);
3043 /* fall through */
3044 case ETHTOOL_GRXRINGS:
3045 case ETHTOOL_GRXCLSRLCNT:
3046 case ETHTOOL_GRXCLSRULE:
3047 case ETHTOOL_SRXCLSRLINS:
3048 convert_out = true;
3049 /* fall through */
3050 case ETHTOOL_SRXCLSRLDEL:
3051 buf_size += sizeof(struct ethtool_rxnfc);
3052 convert_in = true;
3053 rxnfc = compat_alloc_user_space(buf_size);
3054 break;
3055 }
3056
3057 if (copy_from_user(&ifr.ifr_name, &ifr32->ifr_name, IFNAMSIZ))
3058 return -EFAULT;
3059
3060 ifr.ifr_data = convert_in ? rxnfc : (void __user *)compat_rxnfc;
3061
3062 if (convert_in) {
3063 /* We expect there to be holes between fs.m_ext and
3064 * fs.ring_cookie and at the end of fs, but nowhere else.
3065 */
3066 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
3067 sizeof(compat_rxnfc->fs.m_ext) !=
3068 offsetof(struct ethtool_rxnfc, fs.m_ext) +
3069 sizeof(rxnfc->fs.m_ext));
3070 BUILD_BUG_ON(
3071 offsetof(struct compat_ethtool_rxnfc, fs.location) -
3072 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
3073 offsetof(struct ethtool_rxnfc, fs.location) -
3074 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
3075
3076 if (copy_in_user(rxnfc, compat_rxnfc,
3077 (void __user *)(&rxnfc->fs.m_ext + 1) -
3078 (void __user *)rxnfc) ||
3079 copy_in_user(&rxnfc->fs.ring_cookie,
3080 &compat_rxnfc->fs.ring_cookie,
3081 (void __user *)(&rxnfc->fs.location + 1) -
3082 (void __user *)&rxnfc->fs.ring_cookie))
3083 return -EFAULT;
3084 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3085 if (put_user(rule_cnt, &rxnfc->rule_cnt))
3086 return -EFAULT;
3087 } else if (copy_in_user(&rxnfc->rule_cnt,
3088 &compat_rxnfc->rule_cnt,
3089 sizeof(rxnfc->rule_cnt)))
3090 return -EFAULT;
3091 }
3092
3093 ret = dev_ioctl(net, SIOCETHTOOL, &ifr, NULL);
3094 if (ret)
3095 return ret;
3096
3097 if (convert_out) {
3098 if (copy_in_user(compat_rxnfc, rxnfc,
3099 (const void __user *)(&rxnfc->fs.m_ext + 1) -
3100 (const void __user *)rxnfc) ||
3101 copy_in_user(&compat_rxnfc->fs.ring_cookie,
3102 &rxnfc->fs.ring_cookie,
3103 (const void __user *)(&rxnfc->fs.location + 1) -
3104 (const void __user *)&rxnfc->fs.ring_cookie) ||
3105 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
3106 sizeof(rxnfc->rule_cnt)))
3107 return -EFAULT;
3108
3109 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
3110 /* As an optimisation, we only copy the actual
3111 * number of rules that the underlying
3112 * function returned. Since Mallory might
3113 * change the rule count in user memory, we
3114 * check that it is less than the rule count
3115 * originally given (as the user buffer size),
3116 * which has been range-checked.
3117 */
3118 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
3119 return -EFAULT;
3120 if (actual_rule_cnt < rule_cnt)
3121 rule_cnt = actual_rule_cnt;
3122 if (copy_in_user(&compat_rxnfc->rule_locs[0],
3123 &rxnfc->rule_locs[0],
3124 rule_cnt * sizeof(u32)))
3125 return -EFAULT;
3126 }
3127 }
3128
3129 return 0;
3130}
3131
3132static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
3133{
3134 compat_uptr_t uptr32;
3135 struct ifreq ifr;
3136 void __user *saved;
3137 int err;
3138
3139 if (copy_from_user(&ifr, uifr32, sizeof(struct compat_ifreq)))
3140 return -EFAULT;
3141
3142 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
3143 return -EFAULT;
3144
3145 saved = ifr.ifr_settings.ifs_ifsu.raw_hdlc;
3146 ifr.ifr_settings.ifs_ifsu.raw_hdlc = compat_ptr(uptr32);
3147
3148 err = dev_ioctl(net, SIOCWANDEV, &ifr, NULL);
3149 if (!err) {
3150 ifr.ifr_settings.ifs_ifsu.raw_hdlc = saved;
3151 if (copy_to_user(uifr32, &ifr, sizeof(struct compat_ifreq)))
3152 err = -EFAULT;
3153 }
3154 return err;
3155}
3156
3157/* Handle ioctls that use ifreq::ifr_data and just need struct ifreq converted */
3158static int compat_ifr_data_ioctl(struct net *net, unsigned int cmd,
3159 struct compat_ifreq __user *u_ifreq32)
3160{
3161 struct ifreq ifreq;
3162 u32 data32;
3163
3164 if (copy_from_user(ifreq.ifr_name, u_ifreq32->ifr_name, IFNAMSIZ))
3165 return -EFAULT;
3166 if (get_user(data32, &u_ifreq32->ifr_data))
3167 return -EFAULT;
3168 ifreq.ifr_data = compat_ptr(data32);
3169
3170 return dev_ioctl(net, cmd, &ifreq, NULL);
3171}
3172
3173static int compat_ifreq_ioctl(struct net *net, struct socket *sock,
3174 unsigned int cmd,
3175 struct compat_ifreq __user *uifr32)
3176{
3177 struct ifreq __user *uifr;
3178 int err;
3179
3180 /* Handle the fact that while struct ifreq has the same *layout* on
3181 * 32/64 for everything but ifreq::ifru_ifmap and ifreq::ifru_data,
3182 * which are handled elsewhere, it still has different *size* due to
3183 * ifreq::ifru_ifmap (which is 16 bytes on 32 bit, 24 bytes on 64-bit,
3184 * resulting in struct ifreq being 32 and 40 bytes respectively).
3185 * As a result, if the struct happens to be at the end of a page and
3186 * the next page isn't readable/writable, we get a fault. To prevent
3187 * that, copy back and forth to the full size.
3188 */
3189
3190 uifr = compat_alloc_user_space(sizeof(*uifr));
3191 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
3192 return -EFAULT;
3193
3194 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
3195
3196 if (!err) {
3197 switch (cmd) {
3198 case SIOCGIFFLAGS:
3199 case SIOCGIFMETRIC:
3200 case SIOCGIFMTU:
3201 case SIOCGIFMEM:
3202 case SIOCGIFHWADDR:
3203 case SIOCGIFINDEX:
3204 case SIOCGIFADDR:
3205 case SIOCGIFBRDADDR:
3206 case SIOCGIFDSTADDR:
3207 case SIOCGIFNETMASK:
3208 case SIOCGIFPFLAGS:
3209 case SIOCGIFTXQLEN:
3210 case SIOCGMIIPHY:
3211 case SIOCGMIIREG:
3212 case SIOCGIFNAME:
3213 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
3214 err = -EFAULT;
3215 break;
3216 }
3217 }
3218 return err;
3219}
3220
3221static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
3222 struct compat_ifreq __user *uifr32)
3223{
3224 struct ifreq ifr;
3225 struct compat_ifmap __user *uifmap32;
3226 int err;
3227
3228 uifmap32 = &uifr32->ifr_ifru.ifru_map;
3229 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
3230 err |= get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3231 err |= get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3232 err |= get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3233 err |= get_user(ifr.ifr_map.irq, &uifmap32->irq);
3234 err |= get_user(ifr.ifr_map.dma, &uifmap32->dma);
3235 err |= get_user(ifr.ifr_map.port, &uifmap32->port);
3236 if (err)
3237 return -EFAULT;
3238
3239 err = dev_ioctl(net, cmd, &ifr, NULL);
3240
3241 if (cmd == SIOCGIFMAP && !err) {
3242 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
3243 err |= put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
3244 err |= put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
3245 err |= put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
3246 err |= put_user(ifr.ifr_map.irq, &uifmap32->irq);
3247 err |= put_user(ifr.ifr_map.dma, &uifmap32->dma);
3248 err |= put_user(ifr.ifr_map.port, &uifmap32->port);
3249 if (err)
3250 err = -EFAULT;
3251 }
3252 return err;
3253}
3254
3255struct rtentry32 {
3256 u32 rt_pad1;
3257 struct sockaddr rt_dst; /* target address */
3258 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
3259 struct sockaddr rt_genmask; /* target network mask (IP) */
3260 unsigned short rt_flags;
3261 short rt_pad2;
3262 u32 rt_pad3;
3263 unsigned char rt_tos;
3264 unsigned char rt_class;
3265 short rt_pad4;
3266 short rt_metric; /* +1 for binary compatibility! */
3267 /* char * */ u32 rt_dev; /* forcing the device at add */
3268 u32 rt_mtu; /* per route MTU/Window */
3269 u32 rt_window; /* Window clamping */
3270 unsigned short rt_irtt; /* Initial RTT */
3271};
3272
3273struct in6_rtmsg32 {
3274 struct in6_addr rtmsg_dst;
3275 struct in6_addr rtmsg_src;
3276 struct in6_addr rtmsg_gateway;
3277 u32 rtmsg_type;
3278 u16 rtmsg_dst_len;
3279 u16 rtmsg_src_len;
3280 u32 rtmsg_metric;
3281 u32 rtmsg_info;
3282 u32 rtmsg_flags;
3283 s32 rtmsg_ifindex;
3284};
3285
3286static int routing_ioctl(struct net *net, struct socket *sock,
3287 unsigned int cmd, void __user *argp)
3288{
3289 int ret;
3290 void *r = NULL;
3291 struct in6_rtmsg r6;
3292 struct rtentry r4;
3293 char devname[16];
3294 u32 rtdev;
3295 mm_segment_t old_fs = get_fs();
3296
3297 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3298 struct in6_rtmsg32 __user *ur6 = argp;
3299 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3300 3 * sizeof(struct in6_addr));
3301 ret |= get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3302 ret |= get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3303 ret |= get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3304 ret |= get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3305 ret |= get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3306 ret |= get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3307 ret |= get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3308
3309 r = (void *) &r6;
3310 } else { /* ipv4 */
3311 struct rtentry32 __user *ur4 = argp;
3312 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3313 3 * sizeof(struct sockaddr));
3314 ret |= get_user(r4.rt_flags, &(ur4->rt_flags));
3315 ret |= get_user(r4.rt_metric, &(ur4->rt_metric));
3316 ret |= get_user(r4.rt_mtu, &(ur4->rt_mtu));
3317 ret |= get_user(r4.rt_window, &(ur4->rt_window));
3318 ret |= get_user(r4.rt_irtt, &(ur4->rt_irtt));
3319 ret |= get_user(rtdev, &(ur4->rt_dev));
3320 if (rtdev) {
3321 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3322 r4.rt_dev = (char __user __force *)devname;
3323 devname[15] = 0;
3324 } else
3325 r4.rt_dev = NULL;
3326
3327 r = (void *) &r4;
3328 }
3329
3330 if (ret) {
3331 ret = -EFAULT;
3332 goto out;
3333 }
3334
3335 set_fs(KERNEL_DS);
3336 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3337 set_fs(old_fs);
3338
3339out:
3340 return ret;
3341}
3342
3343/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3344 * for some operations; this forces use of the newer bridge-utils that
3345 * use compatible ioctls
3346 */
3347static int old_bridge_ioctl(compat_ulong_t __user *argp)
3348{
3349 compat_ulong_t tmp;
3350
3351 if (get_user(tmp, argp))
3352 return -EFAULT;
3353 if (tmp == BRCTL_GET_VERSION)
3354 return BRCTL_VERSION + 1;
3355 return -EINVAL;
3356}
3357
3358static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3359 unsigned int cmd, unsigned long arg)
3360{
3361 void __user *argp = compat_ptr(arg);
3362 struct sock *sk = sock->sk;
3363 struct net *net = sock_net(sk);
3364
3365 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3366 return compat_ifr_data_ioctl(net, cmd, argp);
3367
3368 switch (cmd) {
3369 case SIOCSIFBR:
3370 case SIOCGIFBR:
3371 return old_bridge_ioctl(argp);
3372 case SIOCGIFCONF:
3373 return compat_dev_ifconf(net, argp);
3374 case SIOCETHTOOL:
3375 return ethtool_ioctl(net, argp);
3376 case SIOCWANDEV:
3377 return compat_siocwandev(net, argp);
3378 case SIOCGIFMAP:
3379 case SIOCSIFMAP:
3380 return compat_sioc_ifmap(net, cmd, argp);
3381 case SIOCADDRT:
3382 case SIOCDELRT:
3383 return routing_ioctl(net, sock, cmd, argp);
3384 case SIOCGSTAMP_OLD:
3385 case SIOCGSTAMPNS_OLD:
3386 if (!sock->ops->gettstamp)
3387 return -ENOIOCTLCMD;
3388 return sock->ops->gettstamp(sock, argp, cmd == SIOCGSTAMP_OLD,
3389 !COMPAT_USE_64BIT_TIME);
3390
3391 case SIOCBONDSLAVEINFOQUERY:
3392 case SIOCBONDINFOQUERY:
3393 case SIOCSHWTSTAMP:
3394 case SIOCGHWTSTAMP:
3395 return compat_ifr_data_ioctl(net, cmd, argp);
3396
3397 case FIOSETOWN:
3398 case SIOCSPGRP:
3399 case FIOGETOWN:
3400 case SIOCGPGRP:
3401 case SIOCBRADDBR:
3402 case SIOCBRDELBR:
3403 case SIOCGIFVLAN:
3404 case SIOCSIFVLAN:
3405 case SIOCADDDLCI:
3406 case SIOCDELDLCI:
3407 case SIOCGSKNS:
3408 case SIOCGSTAMP_NEW:
3409 case SIOCGSTAMPNS_NEW:
3410 return sock_ioctl(file, cmd, arg);
3411
3412 case SIOCGIFFLAGS:
3413 case SIOCSIFFLAGS:
3414 case SIOCGIFMETRIC:
3415 case SIOCSIFMETRIC:
3416 case SIOCGIFMTU:
3417 case SIOCSIFMTU:
3418 case SIOCGIFMEM:
3419 case SIOCSIFMEM:
3420 case SIOCGIFHWADDR:
3421 case SIOCSIFHWADDR:
3422 case SIOCADDMULTI:
3423 case SIOCDELMULTI:
3424 case SIOCGIFINDEX:
3425 case SIOCGIFADDR:
3426 case SIOCSIFADDR:
3427 case SIOCSIFHWBROADCAST:
3428 case SIOCDIFADDR:
3429 case SIOCGIFBRDADDR:
3430 case SIOCSIFBRDADDR:
3431 case SIOCGIFDSTADDR:
3432 case SIOCSIFDSTADDR:
3433 case SIOCGIFNETMASK:
3434 case SIOCSIFNETMASK:
3435 case SIOCSIFPFLAGS:
3436 case SIOCGIFPFLAGS:
3437 case SIOCGIFTXQLEN:
3438 case SIOCSIFTXQLEN:
3439 case SIOCBRADDIF:
3440 case SIOCBRDELIF:
3441 case SIOCGIFNAME:
3442 case SIOCSIFNAME:
3443 case SIOCGMIIPHY:
3444 case SIOCGMIIREG:
3445 case SIOCSMIIREG:
3446 case SIOCBONDENSLAVE:
3447 case SIOCBONDRELEASE:
3448 case SIOCBONDSETHWADDR:
3449 case SIOCBONDCHANGEACTIVE:
3450 return compat_ifreq_ioctl(net, sock, cmd, argp);
3451
3452 case SIOCSARP:
3453 case SIOCGARP:
3454 case SIOCDARP:
3455 case SIOCATMARK:
3456 return sock_do_ioctl(net, sock, cmd, arg);
3457 }
3458
3459 return -ENOIOCTLCMD;
3460}
3461
3462static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3463 unsigned long arg)
3464{
3465 struct socket *sock = file->private_data;
3466 int ret = -ENOIOCTLCMD;
3467 struct sock *sk;
3468 struct net *net;
3469
3470 sk = sock->sk;
3471 net = sock_net(sk);
3472
3473 if (sock->ops->compat_ioctl)
3474 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3475
3476 if (ret == -ENOIOCTLCMD &&
3477 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3478 ret = compat_wext_handle_ioctl(net, cmd, arg);
3479
3480 if (ret == -ENOIOCTLCMD)
3481 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3482
3483 return ret;
3484}
3485#endif
3486
3487/**
3488 * kernel_bind - bind an address to a socket (kernel space)
3489 * @sock: socket
3490 * @addr: address
3491 * @addrlen: length of address
3492 *
3493 * Returns 0 or an error.
3494 */
3495
3496int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3497{
3498 return sock->ops->bind(sock, addr, addrlen);
3499}
3500EXPORT_SYMBOL(kernel_bind);
3501
3502/**
3503 * kernel_listen - move socket to listening state (kernel space)
3504 * @sock: socket
3505 * @backlog: pending connections queue size
3506 *
3507 * Returns 0 or an error.
3508 */
3509
3510int kernel_listen(struct socket *sock, int backlog)
3511{
3512 return sock->ops->listen(sock, backlog);
3513}
3514EXPORT_SYMBOL(kernel_listen);
3515
3516/**
3517 * kernel_accept - accept a connection (kernel space)
3518 * @sock: listening socket
3519 * @newsock: new connected socket
3520 * @flags: flags
3521 *
3522 * @flags must be SOCK_CLOEXEC, SOCK_NONBLOCK or 0.
3523 * If it fails, @newsock is guaranteed to be %NULL.
3524 * Returns 0 or an error.
3525 */
3526
3527int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3528{
3529 struct sock *sk = sock->sk;
3530 int err;
3531
3532 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3533 newsock);
3534 if (err < 0)
3535 goto done;
3536
3537 err = sock->ops->accept(sock, *newsock, flags, true);
3538 if (err < 0) {
3539 sock_release(*newsock);
3540 *newsock = NULL;
3541 goto done;
3542 }
3543
3544 (*newsock)->ops = sock->ops;
3545 __module_get((*newsock)->ops->owner);
3546
3547done:
3548 return err;
3549}
3550EXPORT_SYMBOL(kernel_accept);
3551
3552/**
3553 * kernel_connect - connect a socket (kernel space)
3554 * @sock: socket
3555 * @addr: address
3556 * @addrlen: address length
3557 * @flags: flags (O_NONBLOCK, ...)
3558 *
3559 * For datagram sockets, @addr is the addres to which datagrams are sent
3560 * by default, and the only address from which datagrams are received.
3561 * For stream sockets, attempts to connect to @addr.
3562 * Returns 0 or an error code.
3563 */
3564
3565int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3566 int flags)
3567{
3568 return sock->ops->connect(sock, addr, addrlen, flags);
3569}
3570EXPORT_SYMBOL(kernel_connect);
3571
3572/**
3573 * kernel_getsockname - get the address which the socket is bound (kernel space)
3574 * @sock: socket
3575 * @addr: address holder
3576 *
3577 * Fills the @addr pointer with the address which the socket is bound.
3578 * Returns 0 or an error code.
3579 */
3580
3581int kernel_getsockname(struct socket *sock, struct sockaddr *addr)
3582{
3583 return sock->ops->getname(sock, addr, 0);
3584}
3585EXPORT_SYMBOL(kernel_getsockname);
3586
3587/**
3588 * kernel_peername - get the address which the socket is connected (kernel space)
3589 * @sock: socket
3590 * @addr: address holder
3591 *
3592 * Fills the @addr pointer with the address which the socket is connected.
3593 * Returns 0 or an error code.
3594 */
3595
3596int kernel_getpeername(struct socket *sock, struct sockaddr *addr)
3597{
3598 return sock->ops->getname(sock, addr, 1);
3599}
3600EXPORT_SYMBOL(kernel_getpeername);
3601
3602/**
3603 * kernel_getsockopt - get a socket option (kernel space)
3604 * @sock: socket
3605 * @level: API level (SOL_SOCKET, ...)
3606 * @optname: option tag
3607 * @optval: option value
3608 * @optlen: option length
3609 *
3610 * Assigns the option length to @optlen.
3611 * Returns 0 or an error.
3612 */
3613
3614int kernel_getsockopt(struct socket *sock, int level, int optname,
3615 char *optval, int *optlen)
3616{
3617 mm_segment_t oldfs = get_fs();
3618 char __user *uoptval;
3619 int __user *uoptlen;
3620 int err;
3621
3622 uoptval = (char __user __force *) optval;
3623 uoptlen = (int __user __force *) optlen;
3624
3625 set_fs(KERNEL_DS);
3626 if (level == SOL_SOCKET)
3627 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3628 else
3629 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3630 uoptlen);
3631 set_fs(oldfs);
3632 return err;
3633}
3634EXPORT_SYMBOL(kernel_getsockopt);
3635
3636/**
3637 * kernel_setsockopt - set a socket option (kernel space)
3638 * @sock: socket
3639 * @level: API level (SOL_SOCKET, ...)
3640 * @optname: option tag
3641 * @optval: option value
3642 * @optlen: option length
3643 *
3644 * Returns 0 or an error.
3645 */
3646
3647int kernel_setsockopt(struct socket *sock, int level, int optname,
3648 char *optval, unsigned int optlen)
3649{
3650 mm_segment_t oldfs = get_fs();
3651 char __user *uoptval;
3652 int err;
3653
3654 uoptval = (char __user __force *) optval;
3655
3656 set_fs(KERNEL_DS);
3657 if (level == SOL_SOCKET)
3658 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3659 else
3660 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3661 optlen);
3662 set_fs(oldfs);
3663 return err;
3664}
3665EXPORT_SYMBOL(kernel_setsockopt);
3666
3667/**
3668 * kernel_sendpage - send a &page through a socket (kernel space)
3669 * @sock: socket
3670 * @page: page
3671 * @offset: page offset
3672 * @size: total size in bytes
3673 * @flags: flags (MSG_DONTWAIT, ...)
3674 *
3675 * Returns the total amount sent in bytes or an error.
3676 */
3677
3678int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3679 size_t size, int flags)
3680{
3681 if (sock->ops->sendpage)
3682 return sock->ops->sendpage(sock, page, offset, size, flags);
3683
3684 return sock_no_sendpage(sock, page, offset, size, flags);
3685}
3686EXPORT_SYMBOL(kernel_sendpage);
3687
3688/**
3689 * kernel_sendpage_locked - send a &page through the locked sock (kernel space)
3690 * @sk: sock
3691 * @page: page
3692 * @offset: page offset
3693 * @size: total size in bytes
3694 * @flags: flags (MSG_DONTWAIT, ...)
3695 *
3696 * Returns the total amount sent in bytes or an error.
3697 * Caller must hold @sk.
3698 */
3699
3700int kernel_sendpage_locked(struct sock *sk, struct page *page, int offset,
3701 size_t size, int flags)
3702{
3703 struct socket *sock = sk->sk_socket;
3704
3705 if (sock->ops->sendpage_locked)
3706 return sock->ops->sendpage_locked(sk, page, offset, size,
3707 flags);
3708
3709 return sock_no_sendpage_locked(sk, page, offset, size, flags);
3710}
3711EXPORT_SYMBOL(kernel_sendpage_locked);
3712
3713/**
3714 * kernel_shutdown - shut down part of a full-duplex connection (kernel space)
3715 * @sock: socket
3716 * @how: connection part
3717 *
3718 * Returns 0 or an error.
3719 */
3720
3721int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3722{
3723 return sock->ops->shutdown(sock, how);
3724}
3725EXPORT_SYMBOL(kernel_sock_shutdown);
3726
3727/**
3728 * kernel_sock_ip_overhead - returns the IP overhead imposed by a socket
3729 * @sk: socket
3730 *
3731 * This routine returns the IP overhead imposed by a socket i.e.
3732 * the length of the underlying IP header, depending on whether
3733 * this is an IPv4 or IPv6 socket and the length from IP options turned
3734 * on at the socket. Assumes that the caller has a lock on the socket.
3735 */
3736
3737u32 kernel_sock_ip_overhead(struct sock *sk)
3738{
3739 struct inet_sock *inet;
3740 struct ip_options_rcu *opt;
3741 u32 overhead = 0;
3742#if IS_ENABLED(CONFIG_IPV6)
3743 struct ipv6_pinfo *np;
3744 struct ipv6_txoptions *optv6 = NULL;
3745#endif /* IS_ENABLED(CONFIG_IPV6) */
3746
3747 if (!sk)
3748 return overhead;
3749
3750 switch (sk->sk_family) {
3751 case AF_INET:
3752 inet = inet_sk(sk);
3753 overhead += sizeof(struct iphdr);
3754 opt = rcu_dereference_protected(inet->inet_opt,
3755 sock_owned_by_user(sk));
3756 if (opt)
3757 overhead += opt->opt.optlen;
3758 return overhead;
3759#if IS_ENABLED(CONFIG_IPV6)
3760 case AF_INET6:
3761 np = inet6_sk(sk);
3762 overhead += sizeof(struct ipv6hdr);
3763 if (np)
3764 optv6 = rcu_dereference_protected(np->opt,
3765 sock_owned_by_user(sk));
3766 if (optv6)
3767 overhead += (optv6->opt_flen + optv6->opt_nflen);
3768 return overhead;
3769#endif /* IS_ENABLED(CONFIG_IPV6) */
3770 default: /* Returns 0 overhead if the socket is not ipv4 or ipv6 */
3771 return overhead;
3772 }
3773}
3774EXPORT_SYMBOL(kernel_sock_ip_overhead);