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/*
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/wanrouter.h>
73#include <linux/if_bridge.h>
74#include <linux/if_frad.h>
75#include <linux/if_vlan.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
92#include <asm/uaccess.h>
93#include <asm/unistd.h>
94
95#include <net/compat.h>
96#include <net/wext.h>
97#include <net/cls_cgroup.h>
98
99#include <net/sock.h>
100#include <linux/netfilter.h>
101
102#include <linux/if_tun.h>
103#include <linux/ipv6_route.h>
104#include <linux/route.h>
105#include <linux/sockios.h>
106#include <linux/atalk.h>
107
108static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
109static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
110 unsigned long nr_segs, loff_t pos);
111static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
112 unsigned long nr_segs, loff_t pos);
113static int sock_mmap(struct file *file, struct vm_area_struct *vma);
114
115static int sock_close(struct inode *inode, struct file *file);
116static unsigned int sock_poll(struct file *file,
117 struct poll_table_struct *wait);
118static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
119#ifdef CONFIG_COMPAT
120static long compat_sock_ioctl(struct file *file,
121 unsigned int cmd, unsigned long arg);
122#endif
123static int sock_fasync(int fd, struct file *filp, int on);
124static ssize_t sock_sendpage(struct file *file, struct page *page,
125 int offset, size_t size, loff_t *ppos, int more);
126static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
127 struct pipe_inode_info *pipe, size_t len,
128 unsigned int flags);
129
130/*
131 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
132 * in the operation structures but are done directly via the socketcall() multiplexor.
133 */
134
135static const struct file_operations socket_file_ops = {
136 .owner = THIS_MODULE,
137 .llseek = no_llseek,
138 .aio_read = sock_aio_read,
139 .aio_write = sock_aio_write,
140 .poll = sock_poll,
141 .unlocked_ioctl = sock_ioctl,
142#ifdef CONFIG_COMPAT
143 .compat_ioctl = compat_sock_ioctl,
144#endif
145 .mmap = sock_mmap,
146 .open = sock_no_open, /* special open code to disallow open via /proc */
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 * Statistics counters of the socket lists
163 */
164
165static DEFINE_PER_CPU(int, sockets_in_use);
166
167/*
168 * Support routines.
169 * Move socket addresses back and forth across the kernel/user
170 * divide and look after the messy bits.
171 */
172
173/**
174 * move_addr_to_kernel - copy a socket address into kernel space
175 * @uaddr: Address in user space
176 * @kaddr: Address in kernel space
177 * @ulen: Length in user space
178 *
179 * The address is copied into kernel space. If the provided address is
180 * too long an error code of -EINVAL is returned. If the copy gives
181 * invalid addresses -EFAULT is returned. On a success 0 is returned.
182 */
183
184int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr_storage *kaddr)
185{
186 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
187 return -EINVAL;
188 if (ulen == 0)
189 return 0;
190 if (copy_from_user(kaddr, uaddr, ulen))
191 return -EFAULT;
192 return audit_sockaddr(ulen, kaddr);
193}
194
195/**
196 * move_addr_to_user - copy an address to user space
197 * @kaddr: kernel space address
198 * @klen: length of address in kernel
199 * @uaddr: user space address
200 * @ulen: pointer to user length field
201 *
202 * The value pointed to by ulen on entry is the buffer length available.
203 * This is overwritten with the buffer space used. -EINVAL is returned
204 * if an overlong buffer is specified or a negative buffer size. -EFAULT
205 * is returned if either the buffer or the length field are not
206 * accessible.
207 * After copying the data up to the limit the user specifies, the true
208 * length of the data is written over the length limit the user
209 * specified. Zero is returned for a success.
210 */
211
212static int move_addr_to_user(struct sockaddr_storage *kaddr, int klen,
213 void __user *uaddr, int __user *ulen)
214{
215 int err;
216 int len;
217
218 err = get_user(len, ulen);
219 if (err)
220 return err;
221 if (len > klen)
222 len = klen;
223 if (len < 0 || len > sizeof(struct sockaddr_storage))
224 return -EINVAL;
225 if (len) {
226 if (audit_sockaddr(klen, kaddr))
227 return -ENOMEM;
228 if (copy_to_user(uaddr, kaddr, len))
229 return -EFAULT;
230 }
231 /*
232 * "fromlen shall refer to the value before truncation.."
233 * 1003.1g
234 */
235 return __put_user(klen, ulen);
236}
237
238static struct kmem_cache *sock_inode_cachep __read_mostly;
239
240static struct inode *sock_alloc_inode(struct super_block *sb)
241{
242 struct socket_alloc *ei;
243 struct socket_wq *wq;
244
245 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
246 if (!ei)
247 return NULL;
248 wq = kmalloc(sizeof(*wq), GFP_KERNEL);
249 if (!wq) {
250 kmem_cache_free(sock_inode_cachep, ei);
251 return NULL;
252 }
253 init_waitqueue_head(&wq->wait);
254 wq->fasync_list = NULL;
255 RCU_INIT_POINTER(ei->socket.wq, wq);
256
257 ei->socket.state = SS_UNCONNECTED;
258 ei->socket.flags = 0;
259 ei->socket.ops = NULL;
260 ei->socket.sk = NULL;
261 ei->socket.file = NULL;
262
263 return &ei->vfs_inode;
264}
265
266static void sock_destroy_inode(struct inode *inode)
267{
268 struct socket_alloc *ei;
269 struct socket_wq *wq;
270
271 ei = container_of(inode, struct socket_alloc, vfs_inode);
272 wq = rcu_dereference_protected(ei->socket.wq, 1);
273 kfree_rcu(wq, rcu);
274 kmem_cache_free(sock_inode_cachep, ei);
275}
276
277static void init_once(void *foo)
278{
279 struct socket_alloc *ei = (struct socket_alloc *)foo;
280
281 inode_init_once(&ei->vfs_inode);
282}
283
284static int init_inodecache(void)
285{
286 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
287 sizeof(struct socket_alloc),
288 0,
289 (SLAB_HWCACHE_ALIGN |
290 SLAB_RECLAIM_ACCOUNT |
291 SLAB_MEM_SPREAD),
292 init_once);
293 if (sock_inode_cachep == NULL)
294 return -ENOMEM;
295 return 0;
296}
297
298static const struct super_operations sockfs_ops = {
299 .alloc_inode = sock_alloc_inode,
300 .destroy_inode = sock_destroy_inode,
301 .statfs = simple_statfs,
302};
303
304/*
305 * sockfs_dname() is called from d_path().
306 */
307static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
308{
309 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
310 dentry->d_inode->i_ino);
311}
312
313static const struct dentry_operations sockfs_dentry_operations = {
314 .d_dname = sockfs_dname,
315};
316
317static struct dentry *sockfs_mount(struct file_system_type *fs_type,
318 int flags, const char *dev_name, void *data)
319{
320 return mount_pseudo(fs_type, "socket:", &sockfs_ops,
321 &sockfs_dentry_operations, SOCKFS_MAGIC);
322}
323
324static struct vfsmount *sock_mnt __read_mostly;
325
326static struct file_system_type sock_fs_type = {
327 .name = "sockfs",
328 .mount = sockfs_mount,
329 .kill_sb = kill_anon_super,
330};
331
332/*
333 * Obtains the first available file descriptor and sets it up for use.
334 *
335 * These functions create file structures and maps them to fd space
336 * of the current process. On success it returns file descriptor
337 * and file struct implicitly stored in sock->file.
338 * Note that another thread may close file descriptor before we return
339 * from this function. We use the fact that now we do not refer
340 * to socket after mapping. If one day we will need it, this
341 * function will increment ref. count on file by 1.
342 *
343 * In any case returned fd MAY BE not valid!
344 * This race condition is unavoidable
345 * with shared fd spaces, we cannot solve it inside kernel,
346 * but we take care of internal coherence yet.
347 */
348
349static int sock_alloc_file(struct socket *sock, struct file **f, int flags)
350{
351 struct qstr name = { .name = "" };
352 struct path path;
353 struct file *file;
354 int fd;
355
356 fd = get_unused_fd_flags(flags);
357 if (unlikely(fd < 0))
358 return fd;
359
360 path.dentry = d_alloc_pseudo(sock_mnt->mnt_sb, &name);
361 if (unlikely(!path.dentry)) {
362 put_unused_fd(fd);
363 return -ENOMEM;
364 }
365 path.mnt = mntget(sock_mnt);
366
367 d_instantiate(path.dentry, SOCK_INODE(sock));
368 SOCK_INODE(sock)->i_fop = &socket_file_ops;
369
370 file = alloc_file(&path, FMODE_READ | FMODE_WRITE,
371 &socket_file_ops);
372 if (unlikely(!file)) {
373 /* drop dentry, keep inode */
374 ihold(path.dentry->d_inode);
375 path_put(&path);
376 put_unused_fd(fd);
377 return -ENFILE;
378 }
379
380 sock->file = file;
381 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
382 file->f_pos = 0;
383 file->private_data = sock;
384
385 *f = file;
386 return fd;
387}
388
389int sock_map_fd(struct socket *sock, int flags)
390{
391 struct file *newfile;
392 int fd = sock_alloc_file(sock, &newfile, flags);
393
394 if (likely(fd >= 0))
395 fd_install(fd, newfile);
396
397 return fd;
398}
399EXPORT_SYMBOL(sock_map_fd);
400
401static struct socket *sock_from_file(struct file *file, int *err)
402{
403 if (file->f_op == &socket_file_ops)
404 return file->private_data; /* set in sock_map_fd */
405
406 *err = -ENOTSOCK;
407 return NULL;
408}
409
410/**
411 * sockfd_lookup - Go from a file number to its socket slot
412 * @fd: file handle
413 * @err: pointer to an error code return
414 *
415 * The file handle passed in is locked and the socket it is bound
416 * too is returned. If an error occurs the err pointer is overwritten
417 * with a negative errno code and NULL is returned. The function checks
418 * for both invalid handles and passing a handle which is not a socket.
419 *
420 * On a success the socket object pointer is returned.
421 */
422
423struct socket *sockfd_lookup(int fd, int *err)
424{
425 struct file *file;
426 struct socket *sock;
427
428 file = fget(fd);
429 if (!file) {
430 *err = -EBADF;
431 return NULL;
432 }
433
434 sock = sock_from_file(file, err);
435 if (!sock)
436 fput(file);
437 return sock;
438}
439EXPORT_SYMBOL(sockfd_lookup);
440
441static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
442{
443 struct file *file;
444 struct socket *sock;
445
446 *err = -EBADF;
447 file = fget_light(fd, fput_needed);
448 if (file) {
449 sock = sock_from_file(file, err);
450 if (sock)
451 return sock;
452 fput_light(file, *fput_needed);
453 }
454 return NULL;
455}
456
457/**
458 * sock_alloc - allocate a socket
459 *
460 * Allocate a new inode and socket object. The two are bound together
461 * and initialised. The socket is then returned. If we are out of inodes
462 * NULL is returned.
463 */
464
465static struct socket *sock_alloc(void)
466{
467 struct inode *inode;
468 struct socket *sock;
469
470 inode = new_inode_pseudo(sock_mnt->mnt_sb);
471 if (!inode)
472 return NULL;
473
474 sock = SOCKET_I(inode);
475
476 kmemcheck_annotate_bitfield(sock, type);
477 inode->i_ino = get_next_ino();
478 inode->i_mode = S_IFSOCK | S_IRWXUGO;
479 inode->i_uid = current_fsuid();
480 inode->i_gid = current_fsgid();
481
482 this_cpu_add(sockets_in_use, 1);
483 return sock;
484}
485
486/*
487 * In theory you can't get an open on this inode, but /proc provides
488 * a back door. Remember to keep it shut otherwise you'll let the
489 * creepy crawlies in.
490 */
491
492static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
493{
494 return -ENXIO;
495}
496
497const struct file_operations bad_sock_fops = {
498 .owner = THIS_MODULE,
499 .open = sock_no_open,
500 .llseek = noop_llseek,
501};
502
503/**
504 * sock_release - close a socket
505 * @sock: socket to close
506 *
507 * The socket is released from the protocol stack if it has a release
508 * callback, and the inode is then released if the socket is bound to
509 * an inode not a file.
510 */
511
512void sock_release(struct socket *sock)
513{
514 if (sock->ops) {
515 struct module *owner = sock->ops->owner;
516
517 sock->ops->release(sock);
518 sock->ops = NULL;
519 module_put(owner);
520 }
521
522 if (rcu_dereference_protected(sock->wq, 1)->fasync_list)
523 printk(KERN_ERR "sock_release: fasync list not empty!\n");
524
525 if (test_bit(SOCK_EXTERNALLY_ALLOCATED, &sock->flags))
526 return;
527
528 this_cpu_sub(sockets_in_use, 1);
529 if (!sock->file) {
530 iput(SOCK_INODE(sock));
531 return;
532 }
533 sock->file = NULL;
534}
535EXPORT_SYMBOL(sock_release);
536
537int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags)
538{
539 *tx_flags = 0;
540 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
541 *tx_flags |= SKBTX_HW_TSTAMP;
542 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
543 *tx_flags |= SKBTX_SW_TSTAMP;
544 if (sock_flag(sk, SOCK_WIFI_STATUS))
545 *tx_flags |= SKBTX_WIFI_STATUS;
546 return 0;
547}
548EXPORT_SYMBOL(sock_tx_timestamp);
549
550static inline int __sock_sendmsg_nosec(struct kiocb *iocb, struct socket *sock,
551 struct msghdr *msg, size_t size)
552{
553 struct sock_iocb *si = kiocb_to_siocb(iocb);
554
555 sock_update_classid(sock->sk);
556
557 sock_update_netprioidx(sock->sk);
558
559 si->sock = sock;
560 si->scm = NULL;
561 si->msg = msg;
562 si->size = size;
563
564 return sock->ops->sendmsg(iocb, sock, msg, size);
565}
566
567static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
568 struct msghdr *msg, size_t size)
569{
570 int err = security_socket_sendmsg(sock, msg, size);
571
572 return err ?: __sock_sendmsg_nosec(iocb, sock, msg, size);
573}
574
575int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
576{
577 struct kiocb iocb;
578 struct sock_iocb siocb;
579 int ret;
580
581 init_sync_kiocb(&iocb, NULL);
582 iocb.private = &siocb;
583 ret = __sock_sendmsg(&iocb, sock, msg, size);
584 if (-EIOCBQUEUED == ret)
585 ret = wait_on_sync_kiocb(&iocb);
586 return ret;
587}
588EXPORT_SYMBOL(sock_sendmsg);
589
590static int sock_sendmsg_nosec(struct socket *sock, struct msghdr *msg, size_t size)
591{
592 struct kiocb iocb;
593 struct sock_iocb siocb;
594 int ret;
595
596 init_sync_kiocb(&iocb, NULL);
597 iocb.private = &siocb;
598 ret = __sock_sendmsg_nosec(&iocb, sock, msg, size);
599 if (-EIOCBQUEUED == ret)
600 ret = wait_on_sync_kiocb(&iocb);
601 return ret;
602}
603
604int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
605 struct kvec *vec, size_t num, size_t size)
606{
607 mm_segment_t oldfs = get_fs();
608 int result;
609
610 set_fs(KERNEL_DS);
611 /*
612 * the following is safe, since for compiler definitions of kvec and
613 * iovec are identical, yielding the same in-core layout and alignment
614 */
615 msg->msg_iov = (struct iovec *)vec;
616 msg->msg_iovlen = num;
617 result = sock_sendmsg(sock, msg, size);
618 set_fs(oldfs);
619 return result;
620}
621EXPORT_SYMBOL(kernel_sendmsg);
622
623static int ktime2ts(ktime_t kt, struct timespec *ts)
624{
625 if (kt.tv64) {
626 *ts = ktime_to_timespec(kt);
627 return 1;
628 } else {
629 return 0;
630 }
631}
632
633/*
634 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
635 */
636void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
637 struct sk_buff *skb)
638{
639 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
640 struct timespec ts[3];
641 int empty = 1;
642 struct skb_shared_hwtstamps *shhwtstamps =
643 skb_hwtstamps(skb);
644
645 /* Race occurred between timestamp enabling and packet
646 receiving. Fill in the current time for now. */
647 if (need_software_tstamp && skb->tstamp.tv64 == 0)
648 __net_timestamp(skb);
649
650 if (need_software_tstamp) {
651 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
652 struct timeval tv;
653 skb_get_timestamp(skb, &tv);
654 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
655 sizeof(tv), &tv);
656 } else {
657 skb_get_timestampns(skb, &ts[0]);
658 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
659 sizeof(ts[0]), &ts[0]);
660 }
661 }
662
663
664 memset(ts, 0, sizeof(ts));
665 if (skb->tstamp.tv64 &&
666 sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
667 skb_get_timestampns(skb, ts + 0);
668 empty = 0;
669 }
670 if (shhwtstamps) {
671 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
672 ktime2ts(shhwtstamps->syststamp, ts + 1))
673 empty = 0;
674 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
675 ktime2ts(shhwtstamps->hwtstamp, ts + 2))
676 empty = 0;
677 }
678 if (!empty)
679 put_cmsg(msg, SOL_SOCKET,
680 SCM_TIMESTAMPING, sizeof(ts), &ts);
681}
682EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
683
684void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
685 struct sk_buff *skb)
686{
687 int ack;
688
689 if (!sock_flag(sk, SOCK_WIFI_STATUS))
690 return;
691 if (!skb->wifi_acked_valid)
692 return;
693
694 ack = skb->wifi_acked;
695
696 put_cmsg(msg, SOL_SOCKET, SCM_WIFI_STATUS, sizeof(ack), &ack);
697}
698EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
699
700static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
701 struct sk_buff *skb)
702{
703 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
704 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
705 sizeof(__u32), &skb->dropcount);
706}
707
708void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
709 struct sk_buff *skb)
710{
711 sock_recv_timestamp(msg, sk, skb);
712 sock_recv_drops(msg, sk, skb);
713}
714EXPORT_SYMBOL_GPL(__sock_recv_ts_and_drops);
715
716static inline int __sock_recvmsg_nosec(struct kiocb *iocb, struct socket *sock,
717 struct msghdr *msg, size_t size, int flags)
718{
719 struct sock_iocb *si = kiocb_to_siocb(iocb);
720
721 sock_update_classid(sock->sk);
722
723 si->sock = sock;
724 si->scm = NULL;
725 si->msg = msg;
726 si->size = size;
727 si->flags = flags;
728
729 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
730}
731
732static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
733 struct msghdr *msg, size_t size, int flags)
734{
735 int err = security_socket_recvmsg(sock, msg, size, flags);
736
737 return err ?: __sock_recvmsg_nosec(iocb, sock, msg, size, flags);
738}
739
740int sock_recvmsg(struct socket *sock, struct msghdr *msg,
741 size_t size, int flags)
742{
743 struct kiocb iocb;
744 struct sock_iocb siocb;
745 int ret;
746
747 init_sync_kiocb(&iocb, NULL);
748 iocb.private = &siocb;
749 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
750 if (-EIOCBQUEUED == ret)
751 ret = wait_on_sync_kiocb(&iocb);
752 return ret;
753}
754EXPORT_SYMBOL(sock_recvmsg);
755
756static int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
757 size_t size, int flags)
758{
759 struct kiocb iocb;
760 struct sock_iocb siocb;
761 int ret;
762
763 init_sync_kiocb(&iocb, NULL);
764 iocb.private = &siocb;
765 ret = __sock_recvmsg_nosec(&iocb, sock, msg, size, flags);
766 if (-EIOCBQUEUED == ret)
767 ret = wait_on_sync_kiocb(&iocb);
768 return ret;
769}
770
771/**
772 * kernel_recvmsg - Receive a message from a socket (kernel space)
773 * @sock: The socket to receive the message from
774 * @msg: Received message
775 * @vec: Input s/g array for message data
776 * @num: Size of input s/g array
777 * @size: Number of bytes to read
778 * @flags: Message flags (MSG_DONTWAIT, etc...)
779 *
780 * On return the msg structure contains the scatter/gather array passed in the
781 * vec argument. The array is modified so that it consists of the unfilled
782 * portion of the original array.
783 *
784 * The returned value is the total number of bytes received, or an error.
785 */
786int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
787 struct kvec *vec, size_t num, size_t size, int flags)
788{
789 mm_segment_t oldfs = get_fs();
790 int result;
791
792 set_fs(KERNEL_DS);
793 /*
794 * the following is safe, since for compiler definitions of kvec and
795 * iovec are identical, yielding the same in-core layout and alignment
796 */
797 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
798 result = sock_recvmsg(sock, msg, size, flags);
799 set_fs(oldfs);
800 return result;
801}
802EXPORT_SYMBOL(kernel_recvmsg);
803
804static void sock_aio_dtor(struct kiocb *iocb)
805{
806 kfree(iocb->private);
807}
808
809static ssize_t sock_sendpage(struct file *file, struct page *page,
810 int offset, size_t size, loff_t *ppos, int more)
811{
812 struct socket *sock;
813 int flags;
814
815 sock = file->private_data;
816
817 flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
818 /* more is a combination of MSG_MORE and MSG_SENDPAGE_NOTLAST */
819 flags |= more;
820
821 return kernel_sendpage(sock, page, offset, size, flags);
822}
823
824static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
825 struct pipe_inode_info *pipe, size_t len,
826 unsigned int flags)
827{
828 struct socket *sock = file->private_data;
829
830 if (unlikely(!sock->ops->splice_read))
831 return -EINVAL;
832
833 sock_update_classid(sock->sk);
834
835 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
836}
837
838static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
839 struct sock_iocb *siocb)
840{
841 if (!is_sync_kiocb(iocb)) {
842 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
843 if (!siocb)
844 return NULL;
845 iocb->ki_dtor = sock_aio_dtor;
846 }
847
848 siocb->kiocb = iocb;
849 iocb->private = siocb;
850 return siocb;
851}
852
853static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
854 struct file *file, const struct iovec *iov,
855 unsigned long nr_segs)
856{
857 struct socket *sock = file->private_data;
858 size_t size = 0;
859 int i;
860
861 for (i = 0; i < nr_segs; i++)
862 size += iov[i].iov_len;
863
864 msg->msg_name = NULL;
865 msg->msg_namelen = 0;
866 msg->msg_control = NULL;
867 msg->msg_controllen = 0;
868 msg->msg_iov = (struct iovec *)iov;
869 msg->msg_iovlen = nr_segs;
870 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
871
872 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
873}
874
875static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
876 unsigned long nr_segs, loff_t pos)
877{
878 struct sock_iocb siocb, *x;
879
880 if (pos != 0)
881 return -ESPIPE;
882
883 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
884 return 0;
885
886
887 x = alloc_sock_iocb(iocb, &siocb);
888 if (!x)
889 return -ENOMEM;
890 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
891}
892
893static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
894 struct file *file, const struct iovec *iov,
895 unsigned long nr_segs)
896{
897 struct socket *sock = file->private_data;
898 size_t size = 0;
899 int i;
900
901 for (i = 0; i < nr_segs; i++)
902 size += iov[i].iov_len;
903
904 msg->msg_name = NULL;
905 msg->msg_namelen = 0;
906 msg->msg_control = NULL;
907 msg->msg_controllen = 0;
908 msg->msg_iov = (struct iovec *)iov;
909 msg->msg_iovlen = nr_segs;
910 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
911 if (sock->type == SOCK_SEQPACKET)
912 msg->msg_flags |= MSG_EOR;
913
914 return __sock_sendmsg(iocb, sock, msg, size);
915}
916
917static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
918 unsigned long nr_segs, loff_t pos)
919{
920 struct sock_iocb siocb, *x;
921
922 if (pos != 0)
923 return -ESPIPE;
924
925 x = alloc_sock_iocb(iocb, &siocb);
926 if (!x)
927 return -ENOMEM;
928
929 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
930}
931
932/*
933 * Atomic setting of ioctl hooks to avoid race
934 * with module unload.
935 */
936
937static DEFINE_MUTEX(br_ioctl_mutex);
938static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg);
939
940void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
941{
942 mutex_lock(&br_ioctl_mutex);
943 br_ioctl_hook = hook;
944 mutex_unlock(&br_ioctl_mutex);
945}
946EXPORT_SYMBOL(brioctl_set);
947
948static DEFINE_MUTEX(vlan_ioctl_mutex);
949static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
950
951void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
952{
953 mutex_lock(&vlan_ioctl_mutex);
954 vlan_ioctl_hook = hook;
955 mutex_unlock(&vlan_ioctl_mutex);
956}
957EXPORT_SYMBOL(vlan_ioctl_set);
958
959static DEFINE_MUTEX(dlci_ioctl_mutex);
960static int (*dlci_ioctl_hook) (unsigned int, void __user *);
961
962void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
963{
964 mutex_lock(&dlci_ioctl_mutex);
965 dlci_ioctl_hook = hook;
966 mutex_unlock(&dlci_ioctl_mutex);
967}
968EXPORT_SYMBOL(dlci_ioctl_set);
969
970static long sock_do_ioctl(struct net *net, struct socket *sock,
971 unsigned int cmd, unsigned long arg)
972{
973 int err;
974 void __user *argp = (void __user *)arg;
975
976 err = sock->ops->ioctl(sock, cmd, arg);
977
978 /*
979 * If this ioctl is unknown try to hand it down
980 * to the NIC driver.
981 */
982 if (err == -ENOIOCTLCMD)
983 err = dev_ioctl(net, cmd, argp);
984
985 return err;
986}
987
988/*
989 * With an ioctl, arg may well be a user mode pointer, but we don't know
990 * what to do with it - that's up to the protocol still.
991 */
992
993static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
994{
995 struct socket *sock;
996 struct sock *sk;
997 void __user *argp = (void __user *)arg;
998 int pid, err;
999 struct net *net;
1000
1001 sock = file->private_data;
1002 sk = sock->sk;
1003 net = sock_net(sk);
1004 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
1005 err = dev_ioctl(net, cmd, argp);
1006 } else
1007#ifdef CONFIG_WEXT_CORE
1008 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
1009 err = dev_ioctl(net, cmd, argp);
1010 } else
1011#endif
1012 switch (cmd) {
1013 case FIOSETOWN:
1014 case SIOCSPGRP:
1015 err = -EFAULT;
1016 if (get_user(pid, (int __user *)argp))
1017 break;
1018 err = f_setown(sock->file, pid, 1);
1019 break;
1020 case FIOGETOWN:
1021 case SIOCGPGRP:
1022 err = put_user(f_getown(sock->file),
1023 (int __user *)argp);
1024 break;
1025 case SIOCGIFBR:
1026 case SIOCSIFBR:
1027 case SIOCBRADDBR:
1028 case SIOCBRDELBR:
1029 err = -ENOPKG;
1030 if (!br_ioctl_hook)
1031 request_module("bridge");
1032
1033 mutex_lock(&br_ioctl_mutex);
1034 if (br_ioctl_hook)
1035 err = br_ioctl_hook(net, cmd, argp);
1036 mutex_unlock(&br_ioctl_mutex);
1037 break;
1038 case SIOCGIFVLAN:
1039 case SIOCSIFVLAN:
1040 err = -ENOPKG;
1041 if (!vlan_ioctl_hook)
1042 request_module("8021q");
1043
1044 mutex_lock(&vlan_ioctl_mutex);
1045 if (vlan_ioctl_hook)
1046 err = vlan_ioctl_hook(net, argp);
1047 mutex_unlock(&vlan_ioctl_mutex);
1048 break;
1049 case SIOCADDDLCI:
1050 case SIOCDELDLCI:
1051 err = -ENOPKG;
1052 if (!dlci_ioctl_hook)
1053 request_module("dlci");
1054
1055 mutex_lock(&dlci_ioctl_mutex);
1056 if (dlci_ioctl_hook)
1057 err = dlci_ioctl_hook(cmd, argp);
1058 mutex_unlock(&dlci_ioctl_mutex);
1059 break;
1060 default:
1061 err = sock_do_ioctl(net, sock, cmd, arg);
1062 break;
1063 }
1064 return err;
1065}
1066
1067int sock_create_lite(int family, int type, int protocol, struct socket **res)
1068{
1069 int err;
1070 struct socket *sock = NULL;
1071
1072 err = security_socket_create(family, type, protocol, 1);
1073 if (err)
1074 goto out;
1075
1076 sock = sock_alloc();
1077 if (!sock) {
1078 err = -ENOMEM;
1079 goto out;
1080 }
1081
1082 sock->type = type;
1083 err = security_socket_post_create(sock, family, type, protocol, 1);
1084 if (err)
1085 goto out_release;
1086
1087out:
1088 *res = sock;
1089 return err;
1090out_release:
1091 sock_release(sock);
1092 sock = NULL;
1093 goto out;
1094}
1095EXPORT_SYMBOL(sock_create_lite);
1096
1097/* No kernel lock held - perfect */
1098static unsigned int sock_poll(struct file *file, poll_table *wait)
1099{
1100 struct socket *sock;
1101
1102 /*
1103 * We can't return errors to poll, so it's either yes or no.
1104 */
1105 sock = file->private_data;
1106 return sock->ops->poll(file, sock, wait);
1107}
1108
1109static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1110{
1111 struct socket *sock = file->private_data;
1112
1113 return sock->ops->mmap(file, sock, vma);
1114}
1115
1116static int sock_close(struct inode *inode, struct file *filp)
1117{
1118 /*
1119 * It was possible the inode is NULL we were
1120 * closing an unfinished socket.
1121 */
1122
1123 if (!inode) {
1124 printk(KERN_DEBUG "sock_close: NULL inode\n");
1125 return 0;
1126 }
1127 sock_release(SOCKET_I(inode));
1128 return 0;
1129}
1130
1131/*
1132 * Update the socket async list
1133 *
1134 * Fasync_list locking strategy.
1135 *
1136 * 1. fasync_list is modified only under process context socket lock
1137 * i.e. under semaphore.
1138 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1139 * or under socket lock
1140 */
1141
1142static int sock_fasync(int fd, struct file *filp, int on)
1143{
1144 struct socket *sock = filp->private_data;
1145 struct sock *sk = sock->sk;
1146 struct socket_wq *wq;
1147
1148 if (sk == NULL)
1149 return -EINVAL;
1150
1151 lock_sock(sk);
1152 wq = rcu_dereference_protected(sock->wq, sock_owned_by_user(sk));
1153 fasync_helper(fd, filp, on, &wq->fasync_list);
1154
1155 if (!wq->fasync_list)
1156 sock_reset_flag(sk, SOCK_FASYNC);
1157 else
1158 sock_set_flag(sk, SOCK_FASYNC);
1159
1160 release_sock(sk);
1161 return 0;
1162}
1163
1164/* This function may be called only under socket lock or callback_lock or rcu_lock */
1165
1166int sock_wake_async(struct socket *sock, int how, int band)
1167{
1168 struct socket_wq *wq;
1169
1170 if (!sock)
1171 return -1;
1172 rcu_read_lock();
1173 wq = rcu_dereference(sock->wq);
1174 if (!wq || !wq->fasync_list) {
1175 rcu_read_unlock();
1176 return -1;
1177 }
1178 switch (how) {
1179 case SOCK_WAKE_WAITD:
1180 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1181 break;
1182 goto call_kill;
1183 case SOCK_WAKE_SPACE:
1184 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1185 break;
1186 /* fall through */
1187 case SOCK_WAKE_IO:
1188call_kill:
1189 kill_fasync(&wq->fasync_list, SIGIO, band);
1190 break;
1191 case SOCK_WAKE_URG:
1192 kill_fasync(&wq->fasync_list, SIGURG, band);
1193 }
1194 rcu_read_unlock();
1195 return 0;
1196}
1197EXPORT_SYMBOL(sock_wake_async);
1198
1199int __sock_create(struct net *net, int family, int type, int protocol,
1200 struct socket **res, int kern)
1201{
1202 int err;
1203 struct socket *sock;
1204 const struct net_proto_family *pf;
1205
1206 /*
1207 * Check protocol is in range
1208 */
1209 if (family < 0 || family >= NPROTO)
1210 return -EAFNOSUPPORT;
1211 if (type < 0 || type >= SOCK_MAX)
1212 return -EINVAL;
1213
1214 /* Compatibility.
1215
1216 This uglymoron is moved from INET layer to here to avoid
1217 deadlock in module load.
1218 */
1219 if (family == PF_INET && type == SOCK_PACKET) {
1220 static int warned;
1221 if (!warned) {
1222 warned = 1;
1223 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1224 current->comm);
1225 }
1226 family = PF_PACKET;
1227 }
1228
1229 err = security_socket_create(family, type, protocol, kern);
1230 if (err)
1231 return err;
1232
1233 /*
1234 * Allocate the socket and allow the family to set things up. if
1235 * the protocol is 0, the family is instructed to select an appropriate
1236 * default.
1237 */
1238 sock = sock_alloc();
1239 if (!sock) {
1240 net_warn_ratelimited("socket: no more sockets\n");
1241 return -ENFILE; /* Not exactly a match, but its the
1242 closest posix thing */
1243 }
1244
1245 sock->type = type;
1246
1247#ifdef CONFIG_MODULES
1248 /* Attempt to load a protocol module if the find failed.
1249 *
1250 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1251 * requested real, full-featured networking support upon configuration.
1252 * Otherwise module support will break!
1253 */
1254 if (rcu_access_pointer(net_families[family]) == NULL)
1255 request_module("net-pf-%d", family);
1256#endif
1257
1258 rcu_read_lock();
1259 pf = rcu_dereference(net_families[family]);
1260 err = -EAFNOSUPPORT;
1261 if (!pf)
1262 goto out_release;
1263
1264 /*
1265 * We will call the ->create function, that possibly is in a loadable
1266 * module, so we have to bump that loadable module refcnt first.
1267 */
1268 if (!try_module_get(pf->owner))
1269 goto out_release;
1270
1271 /* Now protected by module ref count */
1272 rcu_read_unlock();
1273
1274 err = pf->create(net, sock, protocol, kern);
1275 if (err < 0)
1276 goto out_module_put;
1277
1278 /*
1279 * Now to bump the refcnt of the [loadable] module that owns this
1280 * socket at sock_release time we decrement its refcnt.
1281 */
1282 if (!try_module_get(sock->ops->owner))
1283 goto out_module_busy;
1284
1285 /*
1286 * Now that we're done with the ->create function, the [loadable]
1287 * module can have its refcnt decremented
1288 */
1289 module_put(pf->owner);
1290 err = security_socket_post_create(sock, family, type, protocol, kern);
1291 if (err)
1292 goto out_sock_release;
1293 *res = sock;
1294
1295 return 0;
1296
1297out_module_busy:
1298 err = -EAFNOSUPPORT;
1299out_module_put:
1300 sock->ops = NULL;
1301 module_put(pf->owner);
1302out_sock_release:
1303 sock_release(sock);
1304 return err;
1305
1306out_release:
1307 rcu_read_unlock();
1308 goto out_sock_release;
1309}
1310EXPORT_SYMBOL(__sock_create);
1311
1312int sock_create(int family, int type, int protocol, struct socket **res)
1313{
1314 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1315}
1316EXPORT_SYMBOL(sock_create);
1317
1318int sock_create_kern(int family, int type, int protocol, struct socket **res)
1319{
1320 return __sock_create(&init_net, family, type, protocol, res, 1);
1321}
1322EXPORT_SYMBOL(sock_create_kern);
1323
1324SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1325{
1326 int retval;
1327 struct socket *sock;
1328 int flags;
1329
1330 /* Check the SOCK_* constants for consistency. */
1331 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1332 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1333 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1334 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1335
1336 flags = type & ~SOCK_TYPE_MASK;
1337 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1338 return -EINVAL;
1339 type &= SOCK_TYPE_MASK;
1340
1341 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1342 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1343
1344 retval = sock_create(family, type, protocol, &sock);
1345 if (retval < 0)
1346 goto out;
1347
1348 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1349 if (retval < 0)
1350 goto out_release;
1351
1352out:
1353 /* It may be already another descriptor 8) Not kernel problem. */
1354 return retval;
1355
1356out_release:
1357 sock_release(sock);
1358 return retval;
1359}
1360
1361/*
1362 * Create a pair of connected sockets.
1363 */
1364
1365SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1366 int __user *, usockvec)
1367{
1368 struct socket *sock1, *sock2;
1369 int fd1, fd2, err;
1370 struct file *newfile1, *newfile2;
1371 int flags;
1372
1373 flags = type & ~SOCK_TYPE_MASK;
1374 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1375 return -EINVAL;
1376 type &= SOCK_TYPE_MASK;
1377
1378 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1379 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1380
1381 /*
1382 * Obtain the first socket and check if the underlying protocol
1383 * supports the socketpair call.
1384 */
1385
1386 err = sock_create(family, type, protocol, &sock1);
1387 if (err < 0)
1388 goto out;
1389
1390 err = sock_create(family, type, protocol, &sock2);
1391 if (err < 0)
1392 goto out_release_1;
1393
1394 err = sock1->ops->socketpair(sock1, sock2);
1395 if (err < 0)
1396 goto out_release_both;
1397
1398 fd1 = sock_alloc_file(sock1, &newfile1, flags);
1399 if (unlikely(fd1 < 0)) {
1400 err = fd1;
1401 goto out_release_both;
1402 }
1403
1404 fd2 = sock_alloc_file(sock2, &newfile2, flags);
1405 if (unlikely(fd2 < 0)) {
1406 err = fd2;
1407 fput(newfile1);
1408 put_unused_fd(fd1);
1409 sock_release(sock2);
1410 goto out;
1411 }
1412
1413 audit_fd_pair(fd1, fd2);
1414 fd_install(fd1, newfile1);
1415 fd_install(fd2, newfile2);
1416 /* fd1 and fd2 may be already another descriptors.
1417 * Not kernel problem.
1418 */
1419
1420 err = put_user(fd1, &usockvec[0]);
1421 if (!err)
1422 err = put_user(fd2, &usockvec[1]);
1423 if (!err)
1424 return 0;
1425
1426 sys_close(fd2);
1427 sys_close(fd1);
1428 return err;
1429
1430out_release_both:
1431 sock_release(sock2);
1432out_release_1:
1433 sock_release(sock1);
1434out:
1435 return err;
1436}
1437
1438/*
1439 * Bind a name to a socket. Nothing much to do here since it's
1440 * the protocol's responsibility to handle the local address.
1441 *
1442 * We move the socket address to kernel space before we call
1443 * the protocol layer (having also checked the address is ok).
1444 */
1445
1446SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1447{
1448 struct socket *sock;
1449 struct sockaddr_storage address;
1450 int err, fput_needed;
1451
1452 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1453 if (sock) {
1454 err = move_addr_to_kernel(umyaddr, addrlen, &address);
1455 if (err >= 0) {
1456 err = security_socket_bind(sock,
1457 (struct sockaddr *)&address,
1458 addrlen);
1459 if (!err)
1460 err = sock->ops->bind(sock,
1461 (struct sockaddr *)
1462 &address, addrlen);
1463 }
1464 fput_light(sock->file, fput_needed);
1465 }
1466 return err;
1467}
1468
1469/*
1470 * Perform a listen. Basically, we allow the protocol to do anything
1471 * necessary for a listen, and if that works, we mark the socket as
1472 * ready for listening.
1473 */
1474
1475SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1476{
1477 struct socket *sock;
1478 int err, fput_needed;
1479 int somaxconn;
1480
1481 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1482 if (sock) {
1483 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1484 if ((unsigned int)backlog > somaxconn)
1485 backlog = somaxconn;
1486
1487 err = security_socket_listen(sock, backlog);
1488 if (!err)
1489 err = sock->ops->listen(sock, backlog);
1490
1491 fput_light(sock->file, fput_needed);
1492 }
1493 return err;
1494}
1495
1496/*
1497 * For accept, we attempt to create a new socket, set up the link
1498 * with the client, wake up the client, then return the new
1499 * connected fd. We collect the address of the connector in kernel
1500 * space and move it to user at the very end. This is unclean because
1501 * we open the socket then return an error.
1502 *
1503 * 1003.1g adds the ability to recvmsg() to query connection pending
1504 * status to recvmsg. We need to add that support in a way thats
1505 * clean when we restucture accept also.
1506 */
1507
1508SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1509 int __user *, upeer_addrlen, int, flags)
1510{
1511 struct socket *sock, *newsock;
1512 struct file *newfile;
1513 int err, len, newfd, fput_needed;
1514 struct sockaddr_storage address;
1515
1516 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1517 return -EINVAL;
1518
1519 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1520 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1521
1522 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1523 if (!sock)
1524 goto out;
1525
1526 err = -ENFILE;
1527 newsock = sock_alloc();
1528 if (!newsock)
1529 goto out_put;
1530
1531 newsock->type = sock->type;
1532 newsock->ops = sock->ops;
1533
1534 /*
1535 * We don't need try_module_get here, as the listening socket (sock)
1536 * has the protocol module (sock->ops->owner) held.
1537 */
1538 __module_get(newsock->ops->owner);
1539
1540 newfd = sock_alloc_file(newsock, &newfile, flags);
1541 if (unlikely(newfd < 0)) {
1542 err = newfd;
1543 sock_release(newsock);
1544 goto out_put;
1545 }
1546
1547 err = security_socket_accept(sock, newsock);
1548 if (err)
1549 goto out_fd;
1550
1551 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1552 if (err < 0)
1553 goto out_fd;
1554
1555 if (upeer_sockaddr) {
1556 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1557 &len, 2) < 0) {
1558 err = -ECONNABORTED;
1559 goto out_fd;
1560 }
1561 err = move_addr_to_user(&address,
1562 len, upeer_sockaddr, upeer_addrlen);
1563 if (err < 0)
1564 goto out_fd;
1565 }
1566
1567 /* File flags are not inherited via accept() unlike another OSes. */
1568
1569 fd_install(newfd, newfile);
1570 err = newfd;
1571
1572out_put:
1573 fput_light(sock->file, fput_needed);
1574out:
1575 return err;
1576out_fd:
1577 fput(newfile);
1578 put_unused_fd(newfd);
1579 goto out_put;
1580}
1581
1582SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1583 int __user *, upeer_addrlen)
1584{
1585 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1586}
1587
1588/*
1589 * Attempt to connect to a socket with the server address. The address
1590 * is in user space so we verify it is OK and move it to kernel space.
1591 *
1592 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1593 * break bindings
1594 *
1595 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1596 * other SEQPACKET protocols that take time to connect() as it doesn't
1597 * include the -EINPROGRESS status for such sockets.
1598 */
1599
1600SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1601 int, addrlen)
1602{
1603 struct socket *sock;
1604 struct sockaddr_storage address;
1605 int err, fput_needed;
1606
1607 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1608 if (!sock)
1609 goto out;
1610 err = move_addr_to_kernel(uservaddr, addrlen, &address);
1611 if (err < 0)
1612 goto out_put;
1613
1614 err =
1615 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1616 if (err)
1617 goto out_put;
1618
1619 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1620 sock->file->f_flags);
1621out_put:
1622 fput_light(sock->file, fput_needed);
1623out:
1624 return err;
1625}
1626
1627/*
1628 * Get the local address ('name') of a socket object. Move the obtained
1629 * name to user space.
1630 */
1631
1632SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1633 int __user *, usockaddr_len)
1634{
1635 struct socket *sock;
1636 struct sockaddr_storage address;
1637 int len, err, fput_needed;
1638
1639 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1640 if (!sock)
1641 goto out;
1642
1643 err = security_socket_getsockname(sock);
1644 if (err)
1645 goto out_put;
1646
1647 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1648 if (err)
1649 goto out_put;
1650 err = move_addr_to_user(&address, len, usockaddr, usockaddr_len);
1651
1652out_put:
1653 fput_light(sock->file, fput_needed);
1654out:
1655 return err;
1656}
1657
1658/*
1659 * Get the remote address ('name') of a socket object. Move the obtained
1660 * name to user space.
1661 */
1662
1663SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1664 int __user *, usockaddr_len)
1665{
1666 struct socket *sock;
1667 struct sockaddr_storage address;
1668 int len, err, fput_needed;
1669
1670 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1671 if (sock != NULL) {
1672 err = security_socket_getpeername(sock);
1673 if (err) {
1674 fput_light(sock->file, fput_needed);
1675 return err;
1676 }
1677
1678 err =
1679 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1680 1);
1681 if (!err)
1682 err = move_addr_to_user(&address, len, usockaddr,
1683 usockaddr_len);
1684 fput_light(sock->file, fput_needed);
1685 }
1686 return err;
1687}
1688
1689/*
1690 * Send a datagram to a given address. We move the address into kernel
1691 * space and check the user space data area is readable before invoking
1692 * the protocol.
1693 */
1694
1695SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1696 unsigned int, flags, struct sockaddr __user *, addr,
1697 int, addr_len)
1698{
1699 struct socket *sock;
1700 struct sockaddr_storage address;
1701 int err;
1702 struct msghdr msg;
1703 struct iovec iov;
1704 int fput_needed;
1705
1706 if (len > INT_MAX)
1707 len = INT_MAX;
1708 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1709 if (!sock)
1710 goto out;
1711
1712 iov.iov_base = buff;
1713 iov.iov_len = len;
1714 msg.msg_name = NULL;
1715 msg.msg_iov = &iov;
1716 msg.msg_iovlen = 1;
1717 msg.msg_control = NULL;
1718 msg.msg_controllen = 0;
1719 msg.msg_namelen = 0;
1720 if (addr) {
1721 err = move_addr_to_kernel(addr, addr_len, &address);
1722 if (err < 0)
1723 goto out_put;
1724 msg.msg_name = (struct sockaddr *)&address;
1725 msg.msg_namelen = addr_len;
1726 }
1727 if (sock->file->f_flags & O_NONBLOCK)
1728 flags |= MSG_DONTWAIT;
1729 msg.msg_flags = flags;
1730 err = sock_sendmsg(sock, &msg, len);
1731
1732out_put:
1733 fput_light(sock->file, fput_needed);
1734out:
1735 return err;
1736}
1737
1738/*
1739 * Send a datagram down a socket.
1740 */
1741
1742SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1743 unsigned int, flags)
1744{
1745 return sys_sendto(fd, buff, len, flags, NULL, 0);
1746}
1747
1748/*
1749 * Receive a frame from the socket and optionally record the address of the
1750 * sender. We verify the buffers are writable and if needed move the
1751 * sender address from kernel to user space.
1752 */
1753
1754SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1755 unsigned int, flags, struct sockaddr __user *, addr,
1756 int __user *, addr_len)
1757{
1758 struct socket *sock;
1759 struct iovec iov;
1760 struct msghdr msg;
1761 struct sockaddr_storage address;
1762 int err, err2;
1763 int fput_needed;
1764
1765 if (size > INT_MAX)
1766 size = INT_MAX;
1767 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1768 if (!sock)
1769 goto out;
1770
1771 msg.msg_control = NULL;
1772 msg.msg_controllen = 0;
1773 msg.msg_iovlen = 1;
1774 msg.msg_iov = &iov;
1775 iov.iov_len = size;
1776 iov.iov_base = ubuf;
1777 msg.msg_name = (struct sockaddr *)&address;
1778 msg.msg_namelen = sizeof(address);
1779 if (sock->file->f_flags & O_NONBLOCK)
1780 flags |= MSG_DONTWAIT;
1781 err = sock_recvmsg(sock, &msg, size, flags);
1782
1783 if (err >= 0 && addr != NULL) {
1784 err2 = move_addr_to_user(&address,
1785 msg.msg_namelen, addr, addr_len);
1786 if (err2 < 0)
1787 err = err2;
1788 }
1789
1790 fput_light(sock->file, fput_needed);
1791out:
1792 return err;
1793}
1794
1795/*
1796 * Receive a datagram from a socket.
1797 */
1798
1799asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1800 unsigned int flags)
1801{
1802 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1803}
1804
1805/*
1806 * Set a socket option. Because we don't know the option lengths we have
1807 * to pass the user mode parameter for the protocols to sort out.
1808 */
1809
1810SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1811 char __user *, optval, int, optlen)
1812{
1813 int err, fput_needed;
1814 struct socket *sock;
1815
1816 if (optlen < 0)
1817 return -EINVAL;
1818
1819 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1820 if (sock != NULL) {
1821 err = security_socket_setsockopt(sock, level, optname);
1822 if (err)
1823 goto out_put;
1824
1825 if (level == SOL_SOCKET)
1826 err =
1827 sock_setsockopt(sock, level, optname, optval,
1828 optlen);
1829 else
1830 err =
1831 sock->ops->setsockopt(sock, level, optname, optval,
1832 optlen);
1833out_put:
1834 fput_light(sock->file, fput_needed);
1835 }
1836 return err;
1837}
1838
1839/*
1840 * Get a socket option. Because we don't know the option lengths we have
1841 * to pass a user mode parameter for the protocols to sort out.
1842 */
1843
1844SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1845 char __user *, optval, int __user *, optlen)
1846{
1847 int err, fput_needed;
1848 struct socket *sock;
1849
1850 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1851 if (sock != NULL) {
1852 err = security_socket_getsockopt(sock, level, optname);
1853 if (err)
1854 goto out_put;
1855
1856 if (level == SOL_SOCKET)
1857 err =
1858 sock_getsockopt(sock, level, optname, optval,
1859 optlen);
1860 else
1861 err =
1862 sock->ops->getsockopt(sock, level, optname, optval,
1863 optlen);
1864out_put:
1865 fput_light(sock->file, fput_needed);
1866 }
1867 return err;
1868}
1869
1870/*
1871 * Shutdown a socket.
1872 */
1873
1874SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1875{
1876 int err, fput_needed;
1877 struct socket *sock;
1878
1879 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1880 if (sock != NULL) {
1881 err = security_socket_shutdown(sock, how);
1882 if (!err)
1883 err = sock->ops->shutdown(sock, how);
1884 fput_light(sock->file, fput_needed);
1885 }
1886 return err;
1887}
1888
1889/* A couple of helpful macros for getting the address of the 32/64 bit
1890 * fields which are the same type (int / unsigned) on our platforms.
1891 */
1892#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1893#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1894#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1895
1896struct used_address {
1897 struct sockaddr_storage name;
1898 unsigned int name_len;
1899};
1900
1901static int __sys_sendmsg(struct socket *sock, struct msghdr __user *msg,
1902 struct msghdr *msg_sys, unsigned int flags,
1903 struct used_address *used_address)
1904{
1905 struct compat_msghdr __user *msg_compat =
1906 (struct compat_msghdr __user *)msg;
1907 struct sockaddr_storage address;
1908 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1909 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1910 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1911 /* 20 is size of ipv6_pktinfo */
1912 unsigned char *ctl_buf = ctl;
1913 int err, ctl_len, total_len;
1914
1915 err = -EFAULT;
1916 if (MSG_CMSG_COMPAT & flags) {
1917 if (get_compat_msghdr(msg_sys, msg_compat))
1918 return -EFAULT;
1919 } else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
1920 return -EFAULT;
1921
1922 if (msg_sys->msg_iovlen > UIO_FASTIOV) {
1923 err = -EMSGSIZE;
1924 if (msg_sys->msg_iovlen > UIO_MAXIOV)
1925 goto out;
1926 err = -ENOMEM;
1927 iov = kmalloc(msg_sys->msg_iovlen * sizeof(struct iovec),
1928 GFP_KERNEL);
1929 if (!iov)
1930 goto out;
1931 }
1932
1933 /* This will also move the address data into kernel space */
1934 if (MSG_CMSG_COMPAT & flags) {
1935 err = verify_compat_iovec(msg_sys, iov, &address, VERIFY_READ);
1936 } else
1937 err = verify_iovec(msg_sys, iov, &address, VERIFY_READ);
1938 if (err < 0)
1939 goto out_freeiov;
1940 total_len = err;
1941
1942 err = -ENOBUFS;
1943
1944 if (msg_sys->msg_controllen > INT_MAX)
1945 goto out_freeiov;
1946 ctl_len = msg_sys->msg_controllen;
1947 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1948 err =
1949 cmsghdr_from_user_compat_to_kern(msg_sys, sock->sk, ctl,
1950 sizeof(ctl));
1951 if (err)
1952 goto out_freeiov;
1953 ctl_buf = msg_sys->msg_control;
1954 ctl_len = msg_sys->msg_controllen;
1955 } else if (ctl_len) {
1956 if (ctl_len > sizeof(ctl)) {
1957 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1958 if (ctl_buf == NULL)
1959 goto out_freeiov;
1960 }
1961 err = -EFAULT;
1962 /*
1963 * Careful! Before this, msg_sys->msg_control contains a user pointer.
1964 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1965 * checking falls down on this.
1966 */
1967 if (copy_from_user(ctl_buf,
1968 (void __user __force *)msg_sys->msg_control,
1969 ctl_len))
1970 goto out_freectl;
1971 msg_sys->msg_control = ctl_buf;
1972 }
1973 msg_sys->msg_flags = flags;
1974
1975 if (sock->file->f_flags & O_NONBLOCK)
1976 msg_sys->msg_flags |= MSG_DONTWAIT;
1977 /*
1978 * If this is sendmmsg() and current destination address is same as
1979 * previously succeeded address, omit asking LSM's decision.
1980 * used_address->name_len is initialized to UINT_MAX so that the first
1981 * destination address never matches.
1982 */
1983 if (used_address && msg_sys->msg_name &&
1984 used_address->name_len == msg_sys->msg_namelen &&
1985 !memcmp(&used_address->name, msg_sys->msg_name,
1986 used_address->name_len)) {
1987 err = sock_sendmsg_nosec(sock, msg_sys, total_len);
1988 goto out_freectl;
1989 }
1990 err = sock_sendmsg(sock, msg_sys, total_len);
1991 /*
1992 * If this is sendmmsg() and sending to current destination address was
1993 * successful, remember it.
1994 */
1995 if (used_address && err >= 0) {
1996 used_address->name_len = msg_sys->msg_namelen;
1997 if (msg_sys->msg_name)
1998 memcpy(&used_address->name, msg_sys->msg_name,
1999 used_address->name_len);
2000 }
2001
2002out_freectl:
2003 if (ctl_buf != ctl)
2004 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
2005out_freeiov:
2006 if (iov != iovstack)
2007 kfree(iov);
2008out:
2009 return err;
2010}
2011
2012/*
2013 * BSD sendmsg interface
2014 */
2015
2016SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned int, flags)
2017{
2018 int fput_needed, err;
2019 struct msghdr msg_sys;
2020 struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2021
2022 if (!sock)
2023 goto out;
2024
2025 err = __sys_sendmsg(sock, msg, &msg_sys, flags, NULL);
2026
2027 fput_light(sock->file, fput_needed);
2028out:
2029 return err;
2030}
2031
2032/*
2033 * Linux sendmmsg interface
2034 */
2035
2036int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2037 unsigned int flags)
2038{
2039 int fput_needed, err, datagrams;
2040 struct socket *sock;
2041 struct mmsghdr __user *entry;
2042 struct compat_mmsghdr __user *compat_entry;
2043 struct msghdr msg_sys;
2044 struct used_address used_address;
2045
2046 if (vlen > UIO_MAXIOV)
2047 vlen = UIO_MAXIOV;
2048
2049 datagrams = 0;
2050
2051 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2052 if (!sock)
2053 return err;
2054
2055 used_address.name_len = UINT_MAX;
2056 entry = mmsg;
2057 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2058 err = 0;
2059
2060 while (datagrams < vlen) {
2061 if (MSG_CMSG_COMPAT & flags) {
2062 err = __sys_sendmsg(sock, (struct msghdr __user *)compat_entry,
2063 &msg_sys, flags, &used_address);
2064 if (err < 0)
2065 break;
2066 err = __put_user(err, &compat_entry->msg_len);
2067 ++compat_entry;
2068 } else {
2069 err = __sys_sendmsg(sock, (struct msghdr __user *)entry,
2070 &msg_sys, flags, &used_address);
2071 if (err < 0)
2072 break;
2073 err = put_user(err, &entry->msg_len);
2074 ++entry;
2075 }
2076
2077 if (err)
2078 break;
2079 ++datagrams;
2080 }
2081
2082 fput_light(sock->file, fput_needed);
2083
2084 /* We only return an error if no datagrams were able to be sent */
2085 if (datagrams != 0)
2086 return datagrams;
2087
2088 return err;
2089}
2090
2091SYSCALL_DEFINE4(sendmmsg, int, fd, struct mmsghdr __user *, mmsg,
2092 unsigned int, vlen, unsigned int, flags)
2093{
2094 return __sys_sendmmsg(fd, mmsg, vlen, flags);
2095}
2096
2097static int __sys_recvmsg(struct socket *sock, struct msghdr __user *msg,
2098 struct msghdr *msg_sys, unsigned int flags, int nosec)
2099{
2100 struct compat_msghdr __user *msg_compat =
2101 (struct compat_msghdr __user *)msg;
2102 struct iovec iovstack[UIO_FASTIOV];
2103 struct iovec *iov = iovstack;
2104 unsigned long cmsg_ptr;
2105 int err, total_len, len;
2106
2107 /* kernel mode address */
2108 struct sockaddr_storage addr;
2109
2110 /* user mode address pointers */
2111 struct sockaddr __user *uaddr;
2112 int __user *uaddr_len;
2113
2114 if (MSG_CMSG_COMPAT & flags) {
2115 if (get_compat_msghdr(msg_sys, msg_compat))
2116 return -EFAULT;
2117 } else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
2118 return -EFAULT;
2119
2120 if (msg_sys->msg_iovlen > UIO_FASTIOV) {
2121 err = -EMSGSIZE;
2122 if (msg_sys->msg_iovlen > UIO_MAXIOV)
2123 goto out;
2124 err = -ENOMEM;
2125 iov = kmalloc(msg_sys->msg_iovlen * sizeof(struct iovec),
2126 GFP_KERNEL);
2127 if (!iov)
2128 goto out;
2129 }
2130
2131 /*
2132 * Save the user-mode address (verify_iovec will change the
2133 * kernel msghdr to use the kernel address space)
2134 */
2135
2136 uaddr = (__force void __user *)msg_sys->msg_name;
2137 uaddr_len = COMPAT_NAMELEN(msg);
2138 if (MSG_CMSG_COMPAT & flags) {
2139 err = verify_compat_iovec(msg_sys, iov, &addr, VERIFY_WRITE);
2140 } else
2141 err = verify_iovec(msg_sys, iov, &addr, VERIFY_WRITE);
2142 if (err < 0)
2143 goto out_freeiov;
2144 total_len = err;
2145
2146 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2147 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2148
2149 if (sock->file->f_flags & O_NONBLOCK)
2150 flags |= MSG_DONTWAIT;
2151 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys,
2152 total_len, flags);
2153 if (err < 0)
2154 goto out_freeiov;
2155 len = err;
2156
2157 if (uaddr != NULL) {
2158 err = move_addr_to_user(&addr,
2159 msg_sys->msg_namelen, uaddr,
2160 uaddr_len);
2161 if (err < 0)
2162 goto out_freeiov;
2163 }
2164 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2165 COMPAT_FLAGS(msg));
2166 if (err)
2167 goto out_freeiov;
2168 if (MSG_CMSG_COMPAT & flags)
2169 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2170 &msg_compat->msg_controllen);
2171 else
2172 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2173 &msg->msg_controllen);
2174 if (err)
2175 goto out_freeiov;
2176 err = len;
2177
2178out_freeiov:
2179 if (iov != iovstack)
2180 kfree(iov);
2181out:
2182 return err;
2183}
2184
2185/*
2186 * BSD recvmsg interface
2187 */
2188
2189SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
2190 unsigned int, flags)
2191{
2192 int fput_needed, err;
2193 struct msghdr msg_sys;
2194 struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2195
2196 if (!sock)
2197 goto out;
2198
2199 err = __sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2200
2201 fput_light(sock->file, fput_needed);
2202out:
2203 return err;
2204}
2205
2206/*
2207 * Linux recvmmsg interface
2208 */
2209
2210int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2211 unsigned int flags, struct timespec *timeout)
2212{
2213 int fput_needed, err, datagrams;
2214 struct socket *sock;
2215 struct mmsghdr __user *entry;
2216 struct compat_mmsghdr __user *compat_entry;
2217 struct msghdr msg_sys;
2218 struct timespec end_time;
2219
2220 if (timeout &&
2221 poll_select_set_timeout(&end_time, timeout->tv_sec,
2222 timeout->tv_nsec))
2223 return -EINVAL;
2224
2225 datagrams = 0;
2226
2227 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2228 if (!sock)
2229 return err;
2230
2231 err = sock_error(sock->sk);
2232 if (err)
2233 goto out_put;
2234
2235 entry = mmsg;
2236 compat_entry = (struct compat_mmsghdr __user *)mmsg;
2237
2238 while (datagrams < vlen) {
2239 /*
2240 * No need to ask LSM for more than the first datagram.
2241 */
2242 if (MSG_CMSG_COMPAT & flags) {
2243 err = __sys_recvmsg(sock, (struct msghdr __user *)compat_entry,
2244 &msg_sys, flags & ~MSG_WAITFORONE,
2245 datagrams);
2246 if (err < 0)
2247 break;
2248 err = __put_user(err, &compat_entry->msg_len);
2249 ++compat_entry;
2250 } else {
2251 err = __sys_recvmsg(sock, (struct msghdr __user *)entry,
2252 &msg_sys, flags & ~MSG_WAITFORONE,
2253 datagrams);
2254 if (err < 0)
2255 break;
2256 err = put_user(err, &entry->msg_len);
2257 ++entry;
2258 }
2259
2260 if (err)
2261 break;
2262 ++datagrams;
2263
2264 /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
2265 if (flags & MSG_WAITFORONE)
2266 flags |= MSG_DONTWAIT;
2267
2268 if (timeout) {
2269 ktime_get_ts(timeout);
2270 *timeout = timespec_sub(end_time, *timeout);
2271 if (timeout->tv_sec < 0) {
2272 timeout->tv_sec = timeout->tv_nsec = 0;
2273 break;
2274 }
2275
2276 /* Timeout, return less than vlen datagrams */
2277 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2278 break;
2279 }
2280
2281 /* Out of band data, return right away */
2282 if (msg_sys.msg_flags & MSG_OOB)
2283 break;
2284 }
2285
2286out_put:
2287 fput_light(sock->file, fput_needed);
2288
2289 if (err == 0)
2290 return datagrams;
2291
2292 if (datagrams != 0) {
2293 /*
2294 * We may return less entries than requested (vlen) if the
2295 * sock is non block and there aren't enough datagrams...
2296 */
2297 if (err != -EAGAIN) {
2298 /*
2299 * ... or if recvmsg returns an error after we
2300 * received some datagrams, where we record the
2301 * error to return on the next call or if the
2302 * app asks about it using getsockopt(SO_ERROR).
2303 */
2304 sock->sk->sk_err = -err;
2305 }
2306
2307 return datagrams;
2308 }
2309
2310 return err;
2311}
2312
2313SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2314 unsigned int, vlen, unsigned int, flags,
2315 struct timespec __user *, timeout)
2316{
2317 int datagrams;
2318 struct timespec timeout_sys;
2319
2320 if (!timeout)
2321 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2322
2323 if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2324 return -EFAULT;
2325
2326 datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2327
2328 if (datagrams > 0 &&
2329 copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2330 datagrams = -EFAULT;
2331
2332 return datagrams;
2333}
2334
2335#ifdef __ARCH_WANT_SYS_SOCKETCALL
2336/* Argument list sizes for sys_socketcall */
2337#define AL(x) ((x) * sizeof(unsigned long))
2338static const unsigned char nargs[21] = {
2339 AL(0), AL(3), AL(3), AL(3), AL(2), AL(3),
2340 AL(3), AL(3), AL(4), AL(4), AL(4), AL(6),
2341 AL(6), AL(2), AL(5), AL(5), AL(3), AL(3),
2342 AL(4), AL(5), AL(4)
2343};
2344
2345#undef AL
2346
2347/*
2348 * System call vectors.
2349 *
2350 * Argument checking cleaned up. Saved 20% in size.
2351 * This function doesn't need to set the kernel lock because
2352 * it is set by the callees.
2353 */
2354
2355SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2356{
2357 unsigned long a[6];
2358 unsigned long a0, a1;
2359 int err;
2360 unsigned int len;
2361
2362 if (call < 1 || call > SYS_SENDMMSG)
2363 return -EINVAL;
2364
2365 len = nargs[call];
2366 if (len > sizeof(a))
2367 return -EINVAL;
2368
2369 /* copy_from_user should be SMP safe. */
2370 if (copy_from_user(a, args, len))
2371 return -EFAULT;
2372
2373 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2374
2375 a0 = a[0];
2376 a1 = a[1];
2377
2378 switch (call) {
2379 case SYS_SOCKET:
2380 err = sys_socket(a0, a1, a[2]);
2381 break;
2382 case SYS_BIND:
2383 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2384 break;
2385 case SYS_CONNECT:
2386 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2387 break;
2388 case SYS_LISTEN:
2389 err = sys_listen(a0, a1);
2390 break;
2391 case SYS_ACCEPT:
2392 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2393 (int __user *)a[2], 0);
2394 break;
2395 case SYS_GETSOCKNAME:
2396 err =
2397 sys_getsockname(a0, (struct sockaddr __user *)a1,
2398 (int __user *)a[2]);
2399 break;
2400 case SYS_GETPEERNAME:
2401 err =
2402 sys_getpeername(a0, (struct sockaddr __user *)a1,
2403 (int __user *)a[2]);
2404 break;
2405 case SYS_SOCKETPAIR:
2406 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2407 break;
2408 case SYS_SEND:
2409 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2410 break;
2411 case SYS_SENDTO:
2412 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2413 (struct sockaddr __user *)a[4], a[5]);
2414 break;
2415 case SYS_RECV:
2416 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2417 break;
2418 case SYS_RECVFROM:
2419 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2420 (struct sockaddr __user *)a[4],
2421 (int __user *)a[5]);
2422 break;
2423 case SYS_SHUTDOWN:
2424 err = sys_shutdown(a0, a1);
2425 break;
2426 case SYS_SETSOCKOPT:
2427 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2428 break;
2429 case SYS_GETSOCKOPT:
2430 err =
2431 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2432 (int __user *)a[4]);
2433 break;
2434 case SYS_SENDMSG:
2435 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2436 break;
2437 case SYS_SENDMMSG:
2438 err = sys_sendmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3]);
2439 break;
2440 case SYS_RECVMSG:
2441 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2442 break;
2443 case SYS_RECVMMSG:
2444 err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2445 (struct timespec __user *)a[4]);
2446 break;
2447 case SYS_ACCEPT4:
2448 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2449 (int __user *)a[2], a[3]);
2450 break;
2451 default:
2452 err = -EINVAL;
2453 break;
2454 }
2455 return err;
2456}
2457
2458#endif /* __ARCH_WANT_SYS_SOCKETCALL */
2459
2460/**
2461 * sock_register - add a socket protocol handler
2462 * @ops: description of protocol
2463 *
2464 * This function is called by a protocol handler that wants to
2465 * advertise its address family, and have it linked into the
2466 * socket interface. The value ops->family coresponds to the
2467 * socket system call protocol family.
2468 */
2469int sock_register(const struct net_proto_family *ops)
2470{
2471 int err;
2472
2473 if (ops->family >= NPROTO) {
2474 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2475 NPROTO);
2476 return -ENOBUFS;
2477 }
2478
2479 spin_lock(&net_family_lock);
2480 if (rcu_dereference_protected(net_families[ops->family],
2481 lockdep_is_held(&net_family_lock)))
2482 err = -EEXIST;
2483 else {
2484 rcu_assign_pointer(net_families[ops->family], ops);
2485 err = 0;
2486 }
2487 spin_unlock(&net_family_lock);
2488
2489 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2490 return err;
2491}
2492EXPORT_SYMBOL(sock_register);
2493
2494/**
2495 * sock_unregister - remove a protocol handler
2496 * @family: protocol family to remove
2497 *
2498 * This function is called by a protocol handler that wants to
2499 * remove its address family, and have it unlinked from the
2500 * new socket creation.
2501 *
2502 * If protocol handler is a module, then it can use module reference
2503 * counts to protect against new references. If protocol handler is not
2504 * a module then it needs to provide its own protection in
2505 * the ops->create routine.
2506 */
2507void sock_unregister(int family)
2508{
2509 BUG_ON(family < 0 || family >= NPROTO);
2510
2511 spin_lock(&net_family_lock);
2512 RCU_INIT_POINTER(net_families[family], NULL);
2513 spin_unlock(&net_family_lock);
2514
2515 synchronize_rcu();
2516
2517 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2518}
2519EXPORT_SYMBOL(sock_unregister);
2520
2521static int __init sock_init(void)
2522{
2523 int err;
2524 /*
2525 * Initialize the network sysctl infrastructure.
2526 */
2527 err = net_sysctl_init();
2528 if (err)
2529 goto out;
2530
2531 /*
2532 * Initialize sock SLAB cache.
2533 */
2534
2535 sk_init();
2536
2537 /*
2538 * Initialize skbuff SLAB cache
2539 */
2540 skb_init();
2541
2542 /*
2543 * Initialize the protocols module.
2544 */
2545
2546 init_inodecache();
2547
2548 err = register_filesystem(&sock_fs_type);
2549 if (err)
2550 goto out_fs;
2551 sock_mnt = kern_mount(&sock_fs_type);
2552 if (IS_ERR(sock_mnt)) {
2553 err = PTR_ERR(sock_mnt);
2554 goto out_mount;
2555 }
2556
2557 /* The real protocol initialization is performed in later initcalls.
2558 */
2559
2560#ifdef CONFIG_NETFILTER
2561 netfilter_init();
2562#endif
2563
2564#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2565 skb_timestamping_init();
2566#endif
2567
2568out:
2569 return err;
2570
2571out_mount:
2572 unregister_filesystem(&sock_fs_type);
2573out_fs:
2574 goto out;
2575}
2576
2577core_initcall(sock_init); /* early initcall */
2578
2579#ifdef CONFIG_PROC_FS
2580void socket_seq_show(struct seq_file *seq)
2581{
2582 int cpu;
2583 int counter = 0;
2584
2585 for_each_possible_cpu(cpu)
2586 counter += per_cpu(sockets_in_use, cpu);
2587
2588 /* It can be negative, by the way. 8) */
2589 if (counter < 0)
2590 counter = 0;
2591
2592 seq_printf(seq, "sockets: used %d\n", counter);
2593}
2594#endif /* CONFIG_PROC_FS */
2595
2596#ifdef CONFIG_COMPAT
2597static int do_siocgstamp(struct net *net, struct socket *sock,
2598 unsigned int cmd, void __user *up)
2599{
2600 mm_segment_t old_fs = get_fs();
2601 struct timeval ktv;
2602 int err;
2603
2604 set_fs(KERNEL_DS);
2605 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&ktv);
2606 set_fs(old_fs);
2607 if (!err)
2608 err = compat_put_timeval(&ktv, up);
2609
2610 return err;
2611}
2612
2613static int do_siocgstampns(struct net *net, struct socket *sock,
2614 unsigned int cmd, void __user *up)
2615{
2616 mm_segment_t old_fs = get_fs();
2617 struct timespec kts;
2618 int err;
2619
2620 set_fs(KERNEL_DS);
2621 err = sock_do_ioctl(net, sock, cmd, (unsigned long)&kts);
2622 set_fs(old_fs);
2623 if (!err)
2624 err = compat_put_timespec(&kts, up);
2625
2626 return err;
2627}
2628
2629static int dev_ifname32(struct net *net, struct compat_ifreq __user *uifr32)
2630{
2631 struct ifreq __user *uifr;
2632 int err;
2633
2634 uifr = compat_alloc_user_space(sizeof(struct ifreq));
2635 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2636 return -EFAULT;
2637
2638 err = dev_ioctl(net, SIOCGIFNAME, uifr);
2639 if (err)
2640 return err;
2641
2642 if (copy_in_user(uifr32, uifr, sizeof(struct compat_ifreq)))
2643 return -EFAULT;
2644
2645 return 0;
2646}
2647
2648static int dev_ifconf(struct net *net, struct compat_ifconf __user *uifc32)
2649{
2650 struct compat_ifconf ifc32;
2651 struct ifconf ifc;
2652 struct ifconf __user *uifc;
2653 struct compat_ifreq __user *ifr32;
2654 struct ifreq __user *ifr;
2655 unsigned int i, j;
2656 int err;
2657
2658 if (copy_from_user(&ifc32, uifc32, sizeof(struct compat_ifconf)))
2659 return -EFAULT;
2660
2661 memset(&ifc, 0, sizeof(ifc));
2662 if (ifc32.ifcbuf == 0) {
2663 ifc32.ifc_len = 0;
2664 ifc.ifc_len = 0;
2665 ifc.ifc_req = NULL;
2666 uifc = compat_alloc_user_space(sizeof(struct ifconf));
2667 } else {
2668 size_t len = ((ifc32.ifc_len / sizeof(struct compat_ifreq)) + 1) *
2669 sizeof(struct ifreq);
2670 uifc = compat_alloc_user_space(sizeof(struct ifconf) + len);
2671 ifc.ifc_len = len;
2672 ifr = ifc.ifc_req = (void __user *)(uifc + 1);
2673 ifr32 = compat_ptr(ifc32.ifcbuf);
2674 for (i = 0; i < ifc32.ifc_len; i += sizeof(struct compat_ifreq)) {
2675 if (copy_in_user(ifr, ifr32, sizeof(struct compat_ifreq)))
2676 return -EFAULT;
2677 ifr++;
2678 ifr32++;
2679 }
2680 }
2681 if (copy_to_user(uifc, &ifc, sizeof(struct ifconf)))
2682 return -EFAULT;
2683
2684 err = dev_ioctl(net, SIOCGIFCONF, uifc);
2685 if (err)
2686 return err;
2687
2688 if (copy_from_user(&ifc, uifc, sizeof(struct ifconf)))
2689 return -EFAULT;
2690
2691 ifr = ifc.ifc_req;
2692 ifr32 = compat_ptr(ifc32.ifcbuf);
2693 for (i = 0, j = 0;
2694 i + sizeof(struct compat_ifreq) <= ifc32.ifc_len && j < ifc.ifc_len;
2695 i += sizeof(struct compat_ifreq), j += sizeof(struct ifreq)) {
2696 if (copy_in_user(ifr32, ifr, sizeof(struct compat_ifreq)))
2697 return -EFAULT;
2698 ifr32++;
2699 ifr++;
2700 }
2701
2702 if (ifc32.ifcbuf == 0) {
2703 /* Translate from 64-bit structure multiple to
2704 * a 32-bit one.
2705 */
2706 i = ifc.ifc_len;
2707 i = ((i / sizeof(struct ifreq)) * sizeof(struct compat_ifreq));
2708 ifc32.ifc_len = i;
2709 } else {
2710 ifc32.ifc_len = i;
2711 }
2712 if (copy_to_user(uifc32, &ifc32, sizeof(struct compat_ifconf)))
2713 return -EFAULT;
2714
2715 return 0;
2716}
2717
2718static int ethtool_ioctl(struct net *net, struct compat_ifreq __user *ifr32)
2719{
2720 struct compat_ethtool_rxnfc __user *compat_rxnfc;
2721 bool convert_in = false, convert_out = false;
2722 size_t buf_size = ALIGN(sizeof(struct ifreq), 8);
2723 struct ethtool_rxnfc __user *rxnfc;
2724 struct ifreq __user *ifr;
2725 u32 rule_cnt = 0, actual_rule_cnt;
2726 u32 ethcmd;
2727 u32 data;
2728 int ret;
2729
2730 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2731 return -EFAULT;
2732
2733 compat_rxnfc = compat_ptr(data);
2734
2735 if (get_user(ethcmd, &compat_rxnfc->cmd))
2736 return -EFAULT;
2737
2738 /* Most ethtool structures are defined without padding.
2739 * Unfortunately struct ethtool_rxnfc is an exception.
2740 */
2741 switch (ethcmd) {
2742 default:
2743 break;
2744 case ETHTOOL_GRXCLSRLALL:
2745 /* Buffer size is variable */
2746 if (get_user(rule_cnt, &compat_rxnfc->rule_cnt))
2747 return -EFAULT;
2748 if (rule_cnt > KMALLOC_MAX_SIZE / sizeof(u32))
2749 return -ENOMEM;
2750 buf_size += rule_cnt * sizeof(u32);
2751 /* fall through */
2752 case ETHTOOL_GRXRINGS:
2753 case ETHTOOL_GRXCLSRLCNT:
2754 case ETHTOOL_GRXCLSRULE:
2755 case ETHTOOL_SRXCLSRLINS:
2756 convert_out = true;
2757 /* fall through */
2758 case ETHTOOL_SRXCLSRLDEL:
2759 buf_size += sizeof(struct ethtool_rxnfc);
2760 convert_in = true;
2761 break;
2762 }
2763
2764 ifr = compat_alloc_user_space(buf_size);
2765 rxnfc = (void *)ifr + ALIGN(sizeof(struct ifreq), 8);
2766
2767 if (copy_in_user(&ifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2768 return -EFAULT;
2769
2770 if (put_user(convert_in ? rxnfc : compat_ptr(data),
2771 &ifr->ifr_ifru.ifru_data))
2772 return -EFAULT;
2773
2774 if (convert_in) {
2775 /* We expect there to be holes between fs.m_ext and
2776 * fs.ring_cookie and at the end of fs, but nowhere else.
2777 */
2778 BUILD_BUG_ON(offsetof(struct compat_ethtool_rxnfc, fs.m_ext) +
2779 sizeof(compat_rxnfc->fs.m_ext) !=
2780 offsetof(struct ethtool_rxnfc, fs.m_ext) +
2781 sizeof(rxnfc->fs.m_ext));
2782 BUILD_BUG_ON(
2783 offsetof(struct compat_ethtool_rxnfc, fs.location) -
2784 offsetof(struct compat_ethtool_rxnfc, fs.ring_cookie) !=
2785 offsetof(struct ethtool_rxnfc, fs.location) -
2786 offsetof(struct ethtool_rxnfc, fs.ring_cookie));
2787
2788 if (copy_in_user(rxnfc, compat_rxnfc,
2789 (void *)(&rxnfc->fs.m_ext + 1) -
2790 (void *)rxnfc) ||
2791 copy_in_user(&rxnfc->fs.ring_cookie,
2792 &compat_rxnfc->fs.ring_cookie,
2793 (void *)(&rxnfc->fs.location + 1) -
2794 (void *)&rxnfc->fs.ring_cookie) ||
2795 copy_in_user(&rxnfc->rule_cnt, &compat_rxnfc->rule_cnt,
2796 sizeof(rxnfc->rule_cnt)))
2797 return -EFAULT;
2798 }
2799
2800 ret = dev_ioctl(net, SIOCETHTOOL, ifr);
2801 if (ret)
2802 return ret;
2803
2804 if (convert_out) {
2805 if (copy_in_user(compat_rxnfc, rxnfc,
2806 (const void *)(&rxnfc->fs.m_ext + 1) -
2807 (const void *)rxnfc) ||
2808 copy_in_user(&compat_rxnfc->fs.ring_cookie,
2809 &rxnfc->fs.ring_cookie,
2810 (const void *)(&rxnfc->fs.location + 1) -
2811 (const void *)&rxnfc->fs.ring_cookie) ||
2812 copy_in_user(&compat_rxnfc->rule_cnt, &rxnfc->rule_cnt,
2813 sizeof(rxnfc->rule_cnt)))
2814 return -EFAULT;
2815
2816 if (ethcmd == ETHTOOL_GRXCLSRLALL) {
2817 /* As an optimisation, we only copy the actual
2818 * number of rules that the underlying
2819 * function returned. Since Mallory might
2820 * change the rule count in user memory, we
2821 * check that it is less than the rule count
2822 * originally given (as the user buffer size),
2823 * which has been range-checked.
2824 */
2825 if (get_user(actual_rule_cnt, &rxnfc->rule_cnt))
2826 return -EFAULT;
2827 if (actual_rule_cnt < rule_cnt)
2828 rule_cnt = actual_rule_cnt;
2829 if (copy_in_user(&compat_rxnfc->rule_locs[0],
2830 &rxnfc->rule_locs[0],
2831 rule_cnt * sizeof(u32)))
2832 return -EFAULT;
2833 }
2834 }
2835
2836 return 0;
2837}
2838
2839static int compat_siocwandev(struct net *net, struct compat_ifreq __user *uifr32)
2840{
2841 void __user *uptr;
2842 compat_uptr_t uptr32;
2843 struct ifreq __user *uifr;
2844
2845 uifr = compat_alloc_user_space(sizeof(*uifr));
2846 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
2847 return -EFAULT;
2848
2849 if (get_user(uptr32, &uifr32->ifr_settings.ifs_ifsu))
2850 return -EFAULT;
2851
2852 uptr = compat_ptr(uptr32);
2853
2854 if (put_user(uptr, &uifr->ifr_settings.ifs_ifsu.raw_hdlc))
2855 return -EFAULT;
2856
2857 return dev_ioctl(net, SIOCWANDEV, uifr);
2858}
2859
2860static int bond_ioctl(struct net *net, unsigned int cmd,
2861 struct compat_ifreq __user *ifr32)
2862{
2863 struct ifreq kifr;
2864 struct ifreq __user *uifr;
2865 mm_segment_t old_fs;
2866 int err;
2867 u32 data;
2868 void __user *datap;
2869
2870 switch (cmd) {
2871 case SIOCBONDENSLAVE:
2872 case SIOCBONDRELEASE:
2873 case SIOCBONDSETHWADDR:
2874 case SIOCBONDCHANGEACTIVE:
2875 if (copy_from_user(&kifr, ifr32, sizeof(struct compat_ifreq)))
2876 return -EFAULT;
2877
2878 old_fs = get_fs();
2879 set_fs(KERNEL_DS);
2880 err = dev_ioctl(net, cmd,
2881 (struct ifreq __user __force *) &kifr);
2882 set_fs(old_fs);
2883
2884 return err;
2885 case SIOCBONDSLAVEINFOQUERY:
2886 case SIOCBONDINFOQUERY:
2887 uifr = compat_alloc_user_space(sizeof(*uifr));
2888 if (copy_in_user(&uifr->ifr_name, &ifr32->ifr_name, IFNAMSIZ))
2889 return -EFAULT;
2890
2891 if (get_user(data, &ifr32->ifr_ifru.ifru_data))
2892 return -EFAULT;
2893
2894 datap = compat_ptr(data);
2895 if (put_user(datap, &uifr->ifr_ifru.ifru_data))
2896 return -EFAULT;
2897
2898 return dev_ioctl(net, cmd, uifr);
2899 default:
2900 return -ENOIOCTLCMD;
2901 }
2902}
2903
2904static int siocdevprivate_ioctl(struct net *net, unsigned int cmd,
2905 struct compat_ifreq __user *u_ifreq32)
2906{
2907 struct ifreq __user *u_ifreq64;
2908 char tmp_buf[IFNAMSIZ];
2909 void __user *data64;
2910 u32 data32;
2911
2912 if (copy_from_user(&tmp_buf[0], &(u_ifreq32->ifr_ifrn.ifrn_name[0]),
2913 IFNAMSIZ))
2914 return -EFAULT;
2915 if (__get_user(data32, &u_ifreq32->ifr_ifru.ifru_data))
2916 return -EFAULT;
2917 data64 = compat_ptr(data32);
2918
2919 u_ifreq64 = compat_alloc_user_space(sizeof(*u_ifreq64));
2920
2921 /* Don't check these user accesses, just let that get trapped
2922 * in the ioctl handler instead.
2923 */
2924 if (copy_to_user(&u_ifreq64->ifr_ifrn.ifrn_name[0], &tmp_buf[0],
2925 IFNAMSIZ))
2926 return -EFAULT;
2927 if (__put_user(data64, &u_ifreq64->ifr_ifru.ifru_data))
2928 return -EFAULT;
2929
2930 return dev_ioctl(net, cmd, u_ifreq64);
2931}
2932
2933static int dev_ifsioc(struct net *net, struct socket *sock,
2934 unsigned int cmd, struct compat_ifreq __user *uifr32)
2935{
2936 struct ifreq __user *uifr;
2937 int err;
2938
2939 uifr = compat_alloc_user_space(sizeof(*uifr));
2940 if (copy_in_user(uifr, uifr32, sizeof(*uifr32)))
2941 return -EFAULT;
2942
2943 err = sock_do_ioctl(net, sock, cmd, (unsigned long)uifr);
2944
2945 if (!err) {
2946 switch (cmd) {
2947 case SIOCGIFFLAGS:
2948 case SIOCGIFMETRIC:
2949 case SIOCGIFMTU:
2950 case SIOCGIFMEM:
2951 case SIOCGIFHWADDR:
2952 case SIOCGIFINDEX:
2953 case SIOCGIFADDR:
2954 case SIOCGIFBRDADDR:
2955 case SIOCGIFDSTADDR:
2956 case SIOCGIFNETMASK:
2957 case SIOCGIFPFLAGS:
2958 case SIOCGIFTXQLEN:
2959 case SIOCGMIIPHY:
2960 case SIOCGMIIREG:
2961 if (copy_in_user(uifr32, uifr, sizeof(*uifr32)))
2962 err = -EFAULT;
2963 break;
2964 }
2965 }
2966 return err;
2967}
2968
2969static int compat_sioc_ifmap(struct net *net, unsigned int cmd,
2970 struct compat_ifreq __user *uifr32)
2971{
2972 struct ifreq ifr;
2973 struct compat_ifmap __user *uifmap32;
2974 mm_segment_t old_fs;
2975 int err;
2976
2977 uifmap32 = &uifr32->ifr_ifru.ifru_map;
2978 err = copy_from_user(&ifr, uifr32, sizeof(ifr.ifr_name));
2979 err |= __get_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2980 err |= __get_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2981 err |= __get_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2982 err |= __get_user(ifr.ifr_map.irq, &uifmap32->irq);
2983 err |= __get_user(ifr.ifr_map.dma, &uifmap32->dma);
2984 err |= __get_user(ifr.ifr_map.port, &uifmap32->port);
2985 if (err)
2986 return -EFAULT;
2987
2988 old_fs = get_fs();
2989 set_fs(KERNEL_DS);
2990 err = dev_ioctl(net, cmd, (void __user __force *)&ifr);
2991 set_fs(old_fs);
2992
2993 if (cmd == SIOCGIFMAP && !err) {
2994 err = copy_to_user(uifr32, &ifr, sizeof(ifr.ifr_name));
2995 err |= __put_user(ifr.ifr_map.mem_start, &uifmap32->mem_start);
2996 err |= __put_user(ifr.ifr_map.mem_end, &uifmap32->mem_end);
2997 err |= __put_user(ifr.ifr_map.base_addr, &uifmap32->base_addr);
2998 err |= __put_user(ifr.ifr_map.irq, &uifmap32->irq);
2999 err |= __put_user(ifr.ifr_map.dma, &uifmap32->dma);
3000 err |= __put_user(ifr.ifr_map.port, &uifmap32->port);
3001 if (err)
3002 err = -EFAULT;
3003 }
3004 return err;
3005}
3006
3007static int compat_siocshwtstamp(struct net *net, struct compat_ifreq __user *uifr32)
3008{
3009 void __user *uptr;
3010 compat_uptr_t uptr32;
3011 struct ifreq __user *uifr;
3012
3013 uifr = compat_alloc_user_space(sizeof(*uifr));
3014 if (copy_in_user(uifr, uifr32, sizeof(struct compat_ifreq)))
3015 return -EFAULT;
3016
3017 if (get_user(uptr32, &uifr32->ifr_data))
3018 return -EFAULT;
3019
3020 uptr = compat_ptr(uptr32);
3021
3022 if (put_user(uptr, &uifr->ifr_data))
3023 return -EFAULT;
3024
3025 return dev_ioctl(net, SIOCSHWTSTAMP, uifr);
3026}
3027
3028struct rtentry32 {
3029 u32 rt_pad1;
3030 struct sockaddr rt_dst; /* target address */
3031 struct sockaddr rt_gateway; /* gateway addr (RTF_GATEWAY) */
3032 struct sockaddr rt_genmask; /* target network mask (IP) */
3033 unsigned short rt_flags;
3034 short rt_pad2;
3035 u32 rt_pad3;
3036 unsigned char rt_tos;
3037 unsigned char rt_class;
3038 short rt_pad4;
3039 short rt_metric; /* +1 for binary compatibility! */
3040 /* char * */ u32 rt_dev; /* forcing the device at add */
3041 u32 rt_mtu; /* per route MTU/Window */
3042 u32 rt_window; /* Window clamping */
3043 unsigned short rt_irtt; /* Initial RTT */
3044};
3045
3046struct in6_rtmsg32 {
3047 struct in6_addr rtmsg_dst;
3048 struct in6_addr rtmsg_src;
3049 struct in6_addr rtmsg_gateway;
3050 u32 rtmsg_type;
3051 u16 rtmsg_dst_len;
3052 u16 rtmsg_src_len;
3053 u32 rtmsg_metric;
3054 u32 rtmsg_info;
3055 u32 rtmsg_flags;
3056 s32 rtmsg_ifindex;
3057};
3058
3059static int routing_ioctl(struct net *net, struct socket *sock,
3060 unsigned int cmd, void __user *argp)
3061{
3062 int ret;
3063 void *r = NULL;
3064 struct in6_rtmsg r6;
3065 struct rtentry r4;
3066 char devname[16];
3067 u32 rtdev;
3068 mm_segment_t old_fs = get_fs();
3069
3070 if (sock && sock->sk && sock->sk->sk_family == AF_INET6) { /* ipv6 */
3071 struct in6_rtmsg32 __user *ur6 = argp;
3072 ret = copy_from_user(&r6.rtmsg_dst, &(ur6->rtmsg_dst),
3073 3 * sizeof(struct in6_addr));
3074 ret |= __get_user(r6.rtmsg_type, &(ur6->rtmsg_type));
3075 ret |= __get_user(r6.rtmsg_dst_len, &(ur6->rtmsg_dst_len));
3076 ret |= __get_user(r6.rtmsg_src_len, &(ur6->rtmsg_src_len));
3077 ret |= __get_user(r6.rtmsg_metric, &(ur6->rtmsg_metric));
3078 ret |= __get_user(r6.rtmsg_info, &(ur6->rtmsg_info));
3079 ret |= __get_user(r6.rtmsg_flags, &(ur6->rtmsg_flags));
3080 ret |= __get_user(r6.rtmsg_ifindex, &(ur6->rtmsg_ifindex));
3081
3082 r = (void *) &r6;
3083 } else { /* ipv4 */
3084 struct rtentry32 __user *ur4 = argp;
3085 ret = copy_from_user(&r4.rt_dst, &(ur4->rt_dst),
3086 3 * sizeof(struct sockaddr));
3087 ret |= __get_user(r4.rt_flags, &(ur4->rt_flags));
3088 ret |= __get_user(r4.rt_metric, &(ur4->rt_metric));
3089 ret |= __get_user(r4.rt_mtu, &(ur4->rt_mtu));
3090 ret |= __get_user(r4.rt_window, &(ur4->rt_window));
3091 ret |= __get_user(r4.rt_irtt, &(ur4->rt_irtt));
3092 ret |= __get_user(rtdev, &(ur4->rt_dev));
3093 if (rtdev) {
3094 ret |= copy_from_user(devname, compat_ptr(rtdev), 15);
3095 r4.rt_dev = (char __user __force *)devname;
3096 devname[15] = 0;
3097 } else
3098 r4.rt_dev = NULL;
3099
3100 r = (void *) &r4;
3101 }
3102
3103 if (ret) {
3104 ret = -EFAULT;
3105 goto out;
3106 }
3107
3108 set_fs(KERNEL_DS);
3109 ret = sock_do_ioctl(net, sock, cmd, (unsigned long) r);
3110 set_fs(old_fs);
3111
3112out:
3113 return ret;
3114}
3115
3116/* Since old style bridge ioctl's endup using SIOCDEVPRIVATE
3117 * for some operations; this forces use of the newer bridge-utils that
3118 * use compatible ioctls
3119 */
3120static int old_bridge_ioctl(compat_ulong_t __user *argp)
3121{
3122 compat_ulong_t tmp;
3123
3124 if (get_user(tmp, argp))
3125 return -EFAULT;
3126 if (tmp == BRCTL_GET_VERSION)
3127 return BRCTL_VERSION + 1;
3128 return -EINVAL;
3129}
3130
3131static int compat_sock_ioctl_trans(struct file *file, struct socket *sock,
3132 unsigned int cmd, unsigned long arg)
3133{
3134 void __user *argp = compat_ptr(arg);
3135 struct sock *sk = sock->sk;
3136 struct net *net = sock_net(sk);
3137
3138 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
3139 return siocdevprivate_ioctl(net, cmd, argp);
3140
3141 switch (cmd) {
3142 case SIOCSIFBR:
3143 case SIOCGIFBR:
3144 return old_bridge_ioctl(argp);
3145 case SIOCGIFNAME:
3146 return dev_ifname32(net, argp);
3147 case SIOCGIFCONF:
3148 return dev_ifconf(net, argp);
3149 case SIOCETHTOOL:
3150 return ethtool_ioctl(net, argp);
3151 case SIOCWANDEV:
3152 return compat_siocwandev(net, argp);
3153 case SIOCGIFMAP:
3154 case SIOCSIFMAP:
3155 return compat_sioc_ifmap(net, cmd, argp);
3156 case SIOCBONDENSLAVE:
3157 case SIOCBONDRELEASE:
3158 case SIOCBONDSETHWADDR:
3159 case SIOCBONDSLAVEINFOQUERY:
3160 case SIOCBONDINFOQUERY:
3161 case SIOCBONDCHANGEACTIVE:
3162 return bond_ioctl(net, cmd, argp);
3163 case SIOCADDRT:
3164 case SIOCDELRT:
3165 return routing_ioctl(net, sock, cmd, argp);
3166 case SIOCGSTAMP:
3167 return do_siocgstamp(net, sock, cmd, argp);
3168 case SIOCGSTAMPNS:
3169 return do_siocgstampns(net, sock, cmd, argp);
3170 case SIOCSHWTSTAMP:
3171 return compat_siocshwtstamp(net, argp);
3172
3173 case FIOSETOWN:
3174 case SIOCSPGRP:
3175 case FIOGETOWN:
3176 case SIOCGPGRP:
3177 case SIOCBRADDBR:
3178 case SIOCBRDELBR:
3179 case SIOCGIFVLAN:
3180 case SIOCSIFVLAN:
3181 case SIOCADDDLCI:
3182 case SIOCDELDLCI:
3183 return sock_ioctl(file, cmd, arg);
3184
3185 case SIOCGIFFLAGS:
3186 case SIOCSIFFLAGS:
3187 case SIOCGIFMETRIC:
3188 case SIOCSIFMETRIC:
3189 case SIOCGIFMTU:
3190 case SIOCSIFMTU:
3191 case SIOCGIFMEM:
3192 case SIOCSIFMEM:
3193 case SIOCGIFHWADDR:
3194 case SIOCSIFHWADDR:
3195 case SIOCADDMULTI:
3196 case SIOCDELMULTI:
3197 case SIOCGIFINDEX:
3198 case SIOCGIFADDR:
3199 case SIOCSIFADDR:
3200 case SIOCSIFHWBROADCAST:
3201 case SIOCDIFADDR:
3202 case SIOCGIFBRDADDR:
3203 case SIOCSIFBRDADDR:
3204 case SIOCGIFDSTADDR:
3205 case SIOCSIFDSTADDR:
3206 case SIOCGIFNETMASK:
3207 case SIOCSIFNETMASK:
3208 case SIOCSIFPFLAGS:
3209 case SIOCGIFPFLAGS:
3210 case SIOCGIFTXQLEN:
3211 case SIOCSIFTXQLEN:
3212 case SIOCBRADDIF:
3213 case SIOCBRDELIF:
3214 case SIOCSIFNAME:
3215 case SIOCGMIIPHY:
3216 case SIOCGMIIREG:
3217 case SIOCSMIIREG:
3218 return dev_ifsioc(net, sock, cmd, argp);
3219
3220 case SIOCSARP:
3221 case SIOCGARP:
3222 case SIOCDARP:
3223 case SIOCATMARK:
3224 return sock_do_ioctl(net, sock, cmd, arg);
3225 }
3226
3227 return -ENOIOCTLCMD;
3228}
3229
3230static long compat_sock_ioctl(struct file *file, unsigned int cmd,
3231 unsigned long arg)
3232{
3233 struct socket *sock = file->private_data;
3234 int ret = -ENOIOCTLCMD;
3235 struct sock *sk;
3236 struct net *net;
3237
3238 sk = sock->sk;
3239 net = sock_net(sk);
3240
3241 if (sock->ops->compat_ioctl)
3242 ret = sock->ops->compat_ioctl(sock, cmd, arg);
3243
3244 if (ret == -ENOIOCTLCMD &&
3245 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
3246 ret = compat_wext_handle_ioctl(net, cmd, arg);
3247
3248 if (ret == -ENOIOCTLCMD)
3249 ret = compat_sock_ioctl_trans(file, sock, cmd, arg);
3250
3251 return ret;
3252}
3253#endif
3254
3255int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
3256{
3257 return sock->ops->bind(sock, addr, addrlen);
3258}
3259EXPORT_SYMBOL(kernel_bind);
3260
3261int kernel_listen(struct socket *sock, int backlog)
3262{
3263 return sock->ops->listen(sock, backlog);
3264}
3265EXPORT_SYMBOL(kernel_listen);
3266
3267int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
3268{
3269 struct sock *sk = sock->sk;
3270 int err;
3271
3272 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
3273 newsock);
3274 if (err < 0)
3275 goto done;
3276
3277 err = sock->ops->accept(sock, *newsock, flags);
3278 if (err < 0) {
3279 sock_release(*newsock);
3280 *newsock = NULL;
3281 goto done;
3282 }
3283
3284 (*newsock)->ops = sock->ops;
3285 __module_get((*newsock)->ops->owner);
3286
3287done:
3288 return err;
3289}
3290EXPORT_SYMBOL(kernel_accept);
3291
3292int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
3293 int flags)
3294{
3295 return sock->ops->connect(sock, addr, addrlen, flags);
3296}
3297EXPORT_SYMBOL(kernel_connect);
3298
3299int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
3300 int *addrlen)
3301{
3302 return sock->ops->getname(sock, addr, addrlen, 0);
3303}
3304EXPORT_SYMBOL(kernel_getsockname);
3305
3306int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
3307 int *addrlen)
3308{
3309 return sock->ops->getname(sock, addr, addrlen, 1);
3310}
3311EXPORT_SYMBOL(kernel_getpeername);
3312
3313int kernel_getsockopt(struct socket *sock, int level, int optname,
3314 char *optval, int *optlen)
3315{
3316 mm_segment_t oldfs = get_fs();
3317 char __user *uoptval;
3318 int __user *uoptlen;
3319 int err;
3320
3321 uoptval = (char __user __force *) optval;
3322 uoptlen = (int __user __force *) optlen;
3323
3324 set_fs(KERNEL_DS);
3325 if (level == SOL_SOCKET)
3326 err = sock_getsockopt(sock, level, optname, uoptval, uoptlen);
3327 else
3328 err = sock->ops->getsockopt(sock, level, optname, uoptval,
3329 uoptlen);
3330 set_fs(oldfs);
3331 return err;
3332}
3333EXPORT_SYMBOL(kernel_getsockopt);
3334
3335int kernel_setsockopt(struct socket *sock, int level, int optname,
3336 char *optval, unsigned int optlen)
3337{
3338 mm_segment_t oldfs = get_fs();
3339 char __user *uoptval;
3340 int err;
3341
3342 uoptval = (char __user __force *) optval;
3343
3344 set_fs(KERNEL_DS);
3345 if (level == SOL_SOCKET)
3346 err = sock_setsockopt(sock, level, optname, uoptval, optlen);
3347 else
3348 err = sock->ops->setsockopt(sock, level, optname, uoptval,
3349 optlen);
3350 set_fs(oldfs);
3351 return err;
3352}
3353EXPORT_SYMBOL(kernel_setsockopt);
3354
3355int kernel_sendpage(struct socket *sock, struct page *page, int offset,
3356 size_t size, int flags)
3357{
3358 sock_update_classid(sock->sk);
3359
3360 if (sock->ops->sendpage)
3361 return sock->ops->sendpage(sock, page, offset, size, flags);
3362
3363 return sock_no_sendpage(sock, page, offset, size, flags);
3364}
3365EXPORT_SYMBOL(kernel_sendpage);
3366
3367int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
3368{
3369 mm_segment_t oldfs = get_fs();
3370 int err;
3371
3372 set_fs(KERNEL_DS);
3373 err = sock->ops->ioctl(sock, cmd, arg);
3374 set_fs(oldfs);
3375
3376 return err;
3377}
3378EXPORT_SYMBOL(kernel_sock_ioctl);
3379
3380int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
3381{
3382 return sock->ops->shutdown(sock, how);
3383}
3384EXPORT_SYMBOL(kernel_sock_shutdown);