2 * NET An implementation of the SOCKET network access protocol.
4 * Version: @(#)socket.c 1.1.93 18/02/95
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
8 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
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
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
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
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
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
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
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.
55 * This module is effectively the top level interface to the BSD socket
58 * Based upon Swansea University Computer Society NET3.039
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>
91 #include <asm/uaccess.h>
92 #include <asm/unistd.h>
94 #include <net/compat.h>
98 #include <linux/netfilter.h>
100 static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
101 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
102 unsigned long nr_segs, loff_t pos);
103 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
104 unsigned long nr_segs, loff_t pos);
105 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
107 static int sock_close(struct inode *inode, struct file *file);
108 static unsigned int sock_poll(struct file *file,
109 struct poll_table_struct *wait);
110 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
112 static long compat_sock_ioctl(struct file *file,
113 unsigned int cmd, unsigned long arg);
115 static int sock_fasync(int fd, struct file *filp, int on);
116 static ssize_t sock_sendpage(struct file *file, struct page *page,
117 int offset, size_t size, loff_t *ppos, int more);
118 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
119 struct pipe_inode_info *pipe, size_t len,
123 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
124 * in the operation structures but are done directly via the socketcall() multiplexor.
127 static const struct file_operations socket_file_ops = {
128 .owner = THIS_MODULE,
130 .aio_read = sock_aio_read,
131 .aio_write = sock_aio_write,
133 .unlocked_ioctl = sock_ioctl,
135 .compat_ioctl = compat_sock_ioctl,
138 .open = sock_no_open, /* special open code to disallow open via /proc */
139 .release = sock_close,
140 .fasync = sock_fasync,
141 .sendpage = sock_sendpage,
142 .splice_write = generic_splice_sendpage,
143 .splice_read = sock_splice_read,
147 * The protocol list. Each protocol is registered in here.
150 static DEFINE_SPINLOCK(net_family_lock);
151 static const struct net_proto_family *net_families[NPROTO] __read_mostly;
154 * Statistics counters of the socket lists
157 static DEFINE_PER_CPU(int, sockets_in_use) = 0;
161 * Move socket addresses back and forth across the kernel/user
162 * divide and look after the messy bits.
165 #define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
166 16 for IP, 16 for IPX,
169 must be at least one bigger than
170 the AF_UNIX size (see net/unix/af_unix.c
175 * move_addr_to_kernel - copy a socket address into kernel space
176 * @uaddr: Address in user space
177 * @kaddr: Address in kernel space
178 * @ulen: Length in user space
180 * The address is copied into kernel space. If the provided address is
181 * too long an error code of -EINVAL is returned. If the copy gives
182 * invalid addresses -EFAULT is returned. On a success 0 is returned.
185 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
187 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
191 if (copy_from_user(kaddr, uaddr, ulen))
193 return audit_sockaddr(ulen, kaddr);
197 * move_addr_to_user - copy an address to user space
198 * @kaddr: kernel space address
199 * @klen: length of address in kernel
200 * @uaddr: user space address
201 * @ulen: pointer to user length field
203 * The value pointed to by ulen on entry is the buffer length available.
204 * This is overwritten with the buffer space used. -EINVAL is returned
205 * if an overlong buffer is specified or a negative buffer size. -EFAULT
206 * is returned if either the buffer or the length field are not
208 * After copying the data up to the limit the user specifies, the true
209 * length of the data is written over the length limit the user
210 * specified. Zero is returned for a success.
213 int move_addr_to_user(struct sockaddr *kaddr, int klen, void __user *uaddr,
219 err = get_user(len, ulen);
224 if (len < 0 || len > sizeof(struct sockaddr_storage))
227 if (audit_sockaddr(klen, kaddr))
229 if (copy_to_user(uaddr, kaddr, len))
233 * "fromlen shall refer to the value before truncation.."
236 return __put_user(klen, ulen);
239 static struct kmem_cache *sock_inode_cachep __read_mostly;
241 static struct inode *sock_alloc_inode(struct super_block *sb)
243 struct socket_alloc *ei;
245 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
248 init_waitqueue_head(&ei->socket.wait);
250 ei->socket.fasync_list = NULL;
251 ei->socket.state = SS_UNCONNECTED;
252 ei->socket.flags = 0;
253 ei->socket.ops = NULL;
254 ei->socket.sk = NULL;
255 ei->socket.file = NULL;
257 return &ei->vfs_inode;
260 static void sock_destroy_inode(struct inode *inode)
262 kmem_cache_free(sock_inode_cachep,
263 container_of(inode, struct socket_alloc, vfs_inode));
266 static void init_once(void *foo)
268 struct socket_alloc *ei = (struct socket_alloc *)foo;
270 inode_init_once(&ei->vfs_inode);
273 static int init_inodecache(void)
275 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
276 sizeof(struct socket_alloc),
278 (SLAB_HWCACHE_ALIGN |
279 SLAB_RECLAIM_ACCOUNT |
282 if (sock_inode_cachep == NULL)
287 static const struct super_operations sockfs_ops = {
288 .alloc_inode = sock_alloc_inode,
289 .destroy_inode =sock_destroy_inode,
290 .statfs = simple_statfs,
293 static int sockfs_get_sb(struct file_system_type *fs_type,
294 int flags, const char *dev_name, void *data,
295 struct vfsmount *mnt)
297 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
301 static struct vfsmount *sock_mnt __read_mostly;
303 static struct file_system_type sock_fs_type = {
305 .get_sb = sockfs_get_sb,
306 .kill_sb = kill_anon_super,
309 static int sockfs_delete_dentry(struct dentry *dentry)
312 * At creation time, we pretended this dentry was hashed
313 * (by clearing DCACHE_UNHASHED bit in d_flags)
314 * At delete time, we restore the truth : not hashed.
315 * (so that dput() can proceed correctly)
317 dentry->d_flags |= DCACHE_UNHASHED;
322 * sockfs_dname() is called from d_path().
324 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
326 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
327 dentry->d_inode->i_ino);
330 static const struct dentry_operations sockfs_dentry_operations = {
331 .d_delete = sockfs_delete_dentry,
332 .d_dname = sockfs_dname,
336 * Obtains the first available file descriptor and sets it up for use.
338 * These functions create file structures and maps them to fd space
339 * of the current process. On success it returns file descriptor
340 * and file struct implicitly stored in sock->file.
341 * Note that another thread may close file descriptor before we return
342 * from this function. We use the fact that now we do not refer
343 * to socket after mapping. If one day we will need it, this
344 * function will increment ref. count on file by 1.
346 * In any case returned fd MAY BE not valid!
347 * This race condition is unavoidable
348 * with shared fd spaces, we cannot solve it inside kernel,
349 * but we take care of internal coherence yet.
352 static int sock_alloc_fd(struct file **filep, int flags)
356 fd = get_unused_fd_flags(flags);
357 if (likely(fd >= 0)) {
358 struct file *file = get_empty_filp();
361 if (unlikely(!file)) {
370 static int sock_attach_fd(struct socket *sock, struct file *file, int flags)
372 struct dentry *dentry;
373 struct qstr name = { .name = "" };
375 dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
376 if (unlikely(!dentry))
379 dentry->d_op = &sockfs_dentry_operations;
381 * We dont want to push this dentry into global dentry hash table.
382 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
383 * This permits a working /proc/$pid/fd/XXX on sockets
385 dentry->d_flags &= ~DCACHE_UNHASHED;
386 d_instantiate(dentry, SOCK_INODE(sock));
389 init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE,
391 SOCK_INODE(sock)->i_fop = &socket_file_ops;
392 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
394 file->private_data = sock;
399 int sock_map_fd(struct socket *sock, int flags)
401 struct file *newfile;
402 int fd = sock_alloc_fd(&newfile, flags);
404 if (likely(fd >= 0)) {
405 int err = sock_attach_fd(sock, newfile, flags);
407 if (unlikely(err < 0)) {
412 fd_install(fd, newfile);
417 static struct socket *sock_from_file(struct file *file, int *err)
419 if (file->f_op == &socket_file_ops)
420 return file->private_data; /* set in sock_map_fd */
427 * sockfd_lookup - Go from a file number to its socket slot
429 * @err: pointer to an error code return
431 * The file handle passed in is locked and the socket it is bound
432 * too is returned. If an error occurs the err pointer is overwritten
433 * with a negative errno code and NULL is returned. The function checks
434 * for both invalid handles and passing a handle which is not a socket.
436 * On a success the socket object pointer is returned.
439 struct socket *sockfd_lookup(int fd, int *err)
450 sock = sock_from_file(file, err);
456 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
462 file = fget_light(fd, fput_needed);
464 sock = sock_from_file(file, err);
467 fput_light(file, *fput_needed);
473 * sock_alloc - allocate a socket
475 * Allocate a new inode and socket object. The two are bound together
476 * and initialised. The socket is then returned. If we are out of inodes
480 static struct socket *sock_alloc(void)
485 inode = new_inode(sock_mnt->mnt_sb);
489 sock = SOCKET_I(inode);
491 kmemcheck_annotate_bitfield(sock, type);
492 inode->i_mode = S_IFSOCK | S_IRWXUGO;
493 inode->i_uid = current_fsuid();
494 inode->i_gid = current_fsgid();
496 percpu_add(sockets_in_use, 1);
501 * In theory you can't get an open on this inode, but /proc provides
502 * a back door. Remember to keep it shut otherwise you'll let the
503 * creepy crawlies in.
506 static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
511 const struct file_operations bad_sock_fops = {
512 .owner = THIS_MODULE,
513 .open = sock_no_open,
517 * sock_release - close a socket
518 * @sock: socket to close
520 * The socket is released from the protocol stack if it has a release
521 * callback, and the inode is then released if the socket is bound to
522 * an inode not a file.
525 void sock_release(struct socket *sock)
528 struct module *owner = sock->ops->owner;
530 sock->ops->release(sock);
535 if (sock->fasync_list)
536 printk(KERN_ERR "sock_release: fasync list not empty!\n");
538 percpu_sub(sockets_in_use, 1);
540 iput(SOCK_INODE(sock));
546 int sock_tx_timestamp(struct msghdr *msg, struct sock *sk,
547 union skb_shared_tx *shtx)
550 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
552 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
556 EXPORT_SYMBOL(sock_tx_timestamp);
558 static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
559 struct msghdr *msg, size_t size)
561 struct sock_iocb *si = kiocb_to_siocb(iocb);
569 err = security_socket_sendmsg(sock, msg, size);
573 return sock->ops->sendmsg(iocb, sock, msg, size);
576 int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
579 struct sock_iocb siocb;
582 init_sync_kiocb(&iocb, NULL);
583 iocb.private = &siocb;
584 ret = __sock_sendmsg(&iocb, sock, msg, size);
585 if (-EIOCBQUEUED == ret)
586 ret = wait_on_sync_kiocb(&iocb);
590 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
591 struct kvec *vec, size_t num, size_t size)
593 mm_segment_t oldfs = get_fs();
598 * the following is safe, since for compiler definitions of kvec and
599 * iovec are identical, yielding the same in-core layout and alignment
601 msg->msg_iov = (struct iovec *)vec;
602 msg->msg_iovlen = num;
603 result = sock_sendmsg(sock, msg, size);
608 static int ktime2ts(ktime_t kt, struct timespec *ts)
611 *ts = ktime_to_timespec(kt);
619 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
621 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
624 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
625 struct timespec ts[3];
627 struct skb_shared_hwtstamps *shhwtstamps =
630 /* Race occurred between timestamp enabling and packet
631 receiving. Fill in the current time for now. */
632 if (need_software_tstamp && skb->tstamp.tv64 == 0)
633 __net_timestamp(skb);
635 if (need_software_tstamp) {
636 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
638 skb_get_timestamp(skb, &tv);
639 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
643 skb_get_timestampns(skb, &ts);
644 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
650 memset(ts, 0, sizeof(ts));
651 if (skb->tstamp.tv64 &&
652 sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
653 skb_get_timestampns(skb, ts + 0);
657 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
658 ktime2ts(shhwtstamps->syststamp, ts + 1))
660 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
661 ktime2ts(shhwtstamps->hwtstamp, ts + 2))
665 put_cmsg(msg, SOL_SOCKET,
666 SCM_TIMESTAMPING, sizeof(ts), &ts);
669 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
671 static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
672 struct msghdr *msg, size_t size, int flags)
675 struct sock_iocb *si = kiocb_to_siocb(iocb);
683 err = security_socket_recvmsg(sock, msg, size, flags);
687 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
690 int sock_recvmsg(struct socket *sock, struct msghdr *msg,
691 size_t size, int flags)
694 struct sock_iocb siocb;
697 init_sync_kiocb(&iocb, NULL);
698 iocb.private = &siocb;
699 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
700 if (-EIOCBQUEUED == ret)
701 ret = wait_on_sync_kiocb(&iocb);
705 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
706 struct kvec *vec, size_t num, size_t size, int flags)
708 mm_segment_t oldfs = get_fs();
713 * the following is safe, since for compiler definitions of kvec and
714 * iovec are identical, yielding the same in-core layout and alignment
716 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
717 result = sock_recvmsg(sock, msg, size, flags);
722 static void sock_aio_dtor(struct kiocb *iocb)
724 kfree(iocb->private);
727 static ssize_t sock_sendpage(struct file *file, struct page *page,
728 int offset, size_t size, loff_t *ppos, int more)
733 sock = file->private_data;
735 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
739 return kernel_sendpage(sock, page, offset, size, flags);
742 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
743 struct pipe_inode_info *pipe, size_t len,
746 struct socket *sock = file->private_data;
748 if (unlikely(!sock->ops->splice_read))
751 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
754 static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
755 struct sock_iocb *siocb)
757 if (!is_sync_kiocb(iocb)) {
758 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
761 iocb->ki_dtor = sock_aio_dtor;
765 iocb->private = siocb;
769 static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
770 struct file *file, const struct iovec *iov,
771 unsigned long nr_segs)
773 struct socket *sock = file->private_data;
777 for (i = 0; i < nr_segs; i++)
778 size += iov[i].iov_len;
780 msg->msg_name = NULL;
781 msg->msg_namelen = 0;
782 msg->msg_control = NULL;
783 msg->msg_controllen = 0;
784 msg->msg_iov = (struct iovec *)iov;
785 msg->msg_iovlen = nr_segs;
786 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
788 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
791 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
792 unsigned long nr_segs, loff_t pos)
794 struct sock_iocb siocb, *x;
799 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
803 x = alloc_sock_iocb(iocb, &siocb);
806 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
809 static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
810 struct file *file, const struct iovec *iov,
811 unsigned long nr_segs)
813 struct socket *sock = file->private_data;
817 for (i = 0; i < nr_segs; i++)
818 size += iov[i].iov_len;
820 msg->msg_name = NULL;
821 msg->msg_namelen = 0;
822 msg->msg_control = NULL;
823 msg->msg_controllen = 0;
824 msg->msg_iov = (struct iovec *)iov;
825 msg->msg_iovlen = nr_segs;
826 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
827 if (sock->type == SOCK_SEQPACKET)
828 msg->msg_flags |= MSG_EOR;
830 return __sock_sendmsg(iocb, sock, msg, size);
833 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
834 unsigned long nr_segs, loff_t pos)
836 struct sock_iocb siocb, *x;
841 x = alloc_sock_iocb(iocb, &siocb);
845 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
849 * Atomic setting of ioctl hooks to avoid race
850 * with module unload.
853 static DEFINE_MUTEX(br_ioctl_mutex);
854 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
856 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
858 mutex_lock(&br_ioctl_mutex);
859 br_ioctl_hook = hook;
860 mutex_unlock(&br_ioctl_mutex);
863 EXPORT_SYMBOL(brioctl_set);
865 static DEFINE_MUTEX(vlan_ioctl_mutex);
866 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
868 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
870 mutex_lock(&vlan_ioctl_mutex);
871 vlan_ioctl_hook = hook;
872 mutex_unlock(&vlan_ioctl_mutex);
875 EXPORT_SYMBOL(vlan_ioctl_set);
877 static DEFINE_MUTEX(dlci_ioctl_mutex);
878 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
880 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
882 mutex_lock(&dlci_ioctl_mutex);
883 dlci_ioctl_hook = hook;
884 mutex_unlock(&dlci_ioctl_mutex);
887 EXPORT_SYMBOL(dlci_ioctl_set);
890 * With an ioctl, arg may well be a user mode pointer, but we don't know
891 * what to do with it - that's up to the protocol still.
894 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
898 void __user *argp = (void __user *)arg;
902 sock = file->private_data;
905 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
906 err = dev_ioctl(net, cmd, argp);
908 #ifdef CONFIG_WIRELESS_EXT
909 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
910 err = dev_ioctl(net, cmd, argp);
912 #endif /* CONFIG_WIRELESS_EXT */
917 if (get_user(pid, (int __user *)argp))
919 err = f_setown(sock->file, pid, 1);
923 err = put_user(f_getown(sock->file),
932 request_module("bridge");
934 mutex_lock(&br_ioctl_mutex);
936 err = br_ioctl_hook(net, cmd, argp);
937 mutex_unlock(&br_ioctl_mutex);
942 if (!vlan_ioctl_hook)
943 request_module("8021q");
945 mutex_lock(&vlan_ioctl_mutex);
947 err = vlan_ioctl_hook(net, argp);
948 mutex_unlock(&vlan_ioctl_mutex);
953 if (!dlci_ioctl_hook)
954 request_module("dlci");
956 mutex_lock(&dlci_ioctl_mutex);
958 err = dlci_ioctl_hook(cmd, argp);
959 mutex_unlock(&dlci_ioctl_mutex);
962 err = sock->ops->ioctl(sock, cmd, arg);
965 * If this ioctl is unknown try to hand it down
968 if (err == -ENOIOCTLCMD)
969 err = dev_ioctl(net, cmd, argp);
975 int sock_create_lite(int family, int type, int protocol, struct socket **res)
978 struct socket *sock = NULL;
980 err = security_socket_create(family, type, protocol, 1);
991 err = security_socket_post_create(sock, family, type, protocol, 1);
1004 /* No kernel lock held - perfect */
1005 static unsigned int sock_poll(struct file *file, poll_table *wait)
1007 struct socket *sock;
1010 * We can't return errors to poll, so it's either yes or no.
1012 sock = file->private_data;
1013 return sock->ops->poll(file, sock, wait);
1016 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1018 struct socket *sock = file->private_data;
1020 return sock->ops->mmap(file, sock, vma);
1023 static int sock_close(struct inode *inode, struct file *filp)
1026 * It was possible the inode is NULL we were
1027 * closing an unfinished socket.
1031 printk(KERN_DEBUG "sock_close: NULL inode\n");
1034 sock_release(SOCKET_I(inode));
1039 * Update the socket async list
1041 * Fasync_list locking strategy.
1043 * 1. fasync_list is modified only under process context socket lock
1044 * i.e. under semaphore.
1045 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1046 * or under socket lock.
1047 * 3. fasync_list can be used from softirq context, so that
1048 * modification under socket lock have to be enhanced with
1049 * write_lock_bh(&sk->sk_callback_lock).
1053 static int sock_fasync(int fd, struct file *filp, int on)
1055 struct fasync_struct *fa, *fna = NULL, **prev;
1056 struct socket *sock;
1060 fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
1065 sock = filp->private_data;
1075 spin_lock(&filp->f_lock);
1077 filp->f_flags |= FASYNC;
1079 filp->f_flags &= ~FASYNC;
1080 spin_unlock(&filp->f_lock);
1082 prev = &(sock->fasync_list);
1084 for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1085 if (fa->fa_file == filp)
1090 write_lock_bh(&sk->sk_callback_lock);
1092 write_unlock_bh(&sk->sk_callback_lock);
1097 fna->fa_file = filp;
1099 fna->magic = FASYNC_MAGIC;
1100 fna->fa_next = sock->fasync_list;
1101 write_lock_bh(&sk->sk_callback_lock);
1102 sock->fasync_list = fna;
1103 write_unlock_bh(&sk->sk_callback_lock);
1106 write_lock_bh(&sk->sk_callback_lock);
1107 *prev = fa->fa_next;
1108 write_unlock_bh(&sk->sk_callback_lock);
1114 release_sock(sock->sk);
1118 /* This function may be called only under socket lock or callback_lock */
1120 int sock_wake_async(struct socket *sock, int how, int band)
1122 if (!sock || !sock->fasync_list)
1125 case SOCK_WAKE_WAITD:
1126 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1129 case SOCK_WAKE_SPACE:
1130 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1135 __kill_fasync(sock->fasync_list, SIGIO, band);
1138 __kill_fasync(sock->fasync_list, SIGURG, band);
1143 static int __sock_create(struct net *net, int family, int type, int protocol,
1144 struct socket **res, int kern)
1147 struct socket *sock;
1148 const struct net_proto_family *pf;
1151 * Check protocol is in range
1153 if (family < 0 || family >= NPROTO)
1154 return -EAFNOSUPPORT;
1155 if (type < 0 || type >= SOCK_MAX)
1160 This uglymoron is moved from INET layer to here to avoid
1161 deadlock in module load.
1163 if (family == PF_INET && type == SOCK_PACKET) {
1167 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1173 err = security_socket_create(family, type, protocol, kern);
1178 * Allocate the socket and allow the family to set things up. if
1179 * the protocol is 0, the family is instructed to select an appropriate
1182 sock = sock_alloc();
1184 if (net_ratelimit())
1185 printk(KERN_WARNING "socket: no more sockets\n");
1186 return -ENFILE; /* Not exactly a match, but its the
1187 closest posix thing */
1192 #ifdef CONFIG_MODULES
1193 /* Attempt to load a protocol module if the find failed.
1195 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1196 * requested real, full-featured networking support upon configuration.
1197 * Otherwise module support will break!
1199 if (net_families[family] == NULL)
1200 request_module("net-pf-%d", family);
1204 pf = rcu_dereference(net_families[family]);
1205 err = -EAFNOSUPPORT;
1210 * We will call the ->create function, that possibly is in a loadable
1211 * module, so we have to bump that loadable module refcnt first.
1213 if (!try_module_get(pf->owner))
1216 /* Now protected by module ref count */
1219 err = pf->create(net, sock, protocol);
1221 goto out_module_put;
1224 * Now to bump the refcnt of the [loadable] module that owns this
1225 * socket at sock_release time we decrement its refcnt.
1227 if (!try_module_get(sock->ops->owner))
1228 goto out_module_busy;
1231 * Now that we're done with the ->create function, the [loadable]
1232 * module can have its refcnt decremented
1234 module_put(pf->owner);
1235 err = security_socket_post_create(sock, family, type, protocol, kern);
1237 goto out_sock_release;
1243 err = -EAFNOSUPPORT;
1246 module_put(pf->owner);
1253 goto out_sock_release;
1256 int sock_create(int family, int type, int protocol, struct socket **res)
1258 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1261 int sock_create_kern(int family, int type, int protocol, struct socket **res)
1263 return __sock_create(&init_net, family, type, protocol, res, 1);
1266 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1269 struct socket *sock;
1272 /* Check the SOCK_* constants for consistency. */
1273 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1274 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1275 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1276 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1278 flags = type & ~SOCK_TYPE_MASK;
1279 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1281 type &= SOCK_TYPE_MASK;
1283 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1284 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1286 retval = sock_create(family, type, protocol, &sock);
1290 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1295 /* It may be already another descriptor 8) Not kernel problem. */
1304 * Create a pair of connected sockets.
1307 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1308 int __user *, usockvec)
1310 struct socket *sock1, *sock2;
1312 struct file *newfile1, *newfile2;
1315 flags = type & ~SOCK_TYPE_MASK;
1316 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1318 type &= SOCK_TYPE_MASK;
1320 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1321 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1324 * Obtain the first socket and check if the underlying protocol
1325 * supports the socketpair call.
1328 err = sock_create(family, type, protocol, &sock1);
1332 err = sock_create(family, type, protocol, &sock2);
1336 err = sock1->ops->socketpair(sock1, sock2);
1338 goto out_release_both;
1340 fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
1341 if (unlikely(fd1 < 0)) {
1343 goto out_release_both;
1346 fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
1347 if (unlikely(fd2 < 0)) {
1351 goto out_release_both;
1354 err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
1355 if (unlikely(err < 0)) {
1359 err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
1360 if (unlikely(err < 0)) {
1365 audit_fd_pair(fd1, fd2);
1366 fd_install(fd1, newfile1);
1367 fd_install(fd2, newfile2);
1368 /* fd1 and fd2 may be already another descriptors.
1369 * Not kernel problem.
1372 err = put_user(fd1, &usockvec[0]);
1374 err = put_user(fd2, &usockvec[1]);
1383 sock_release(sock2);
1385 sock_release(sock1);
1391 sock_release(sock1);
1394 sock_release(sock2);
1401 * Bind a name to a socket. Nothing much to do here since it's
1402 * the protocol's responsibility to handle the local address.
1404 * We move the socket address to kernel space before we call
1405 * the protocol layer (having also checked the address is ok).
1408 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1410 struct socket *sock;
1411 struct sockaddr_storage address;
1412 int err, fput_needed;
1414 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1416 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1418 err = security_socket_bind(sock,
1419 (struct sockaddr *)&address,
1422 err = sock->ops->bind(sock,
1426 fput_light(sock->file, fput_needed);
1432 * Perform a listen. Basically, we allow the protocol to do anything
1433 * necessary for a listen, and if that works, we mark the socket as
1434 * ready for listening.
1437 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1439 struct socket *sock;
1440 int err, fput_needed;
1443 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1445 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1446 if ((unsigned)backlog > somaxconn)
1447 backlog = somaxconn;
1449 err = security_socket_listen(sock, backlog);
1451 err = sock->ops->listen(sock, backlog);
1453 fput_light(sock->file, fput_needed);
1459 * For accept, we attempt to create a new socket, set up the link
1460 * with the client, wake up the client, then return the new
1461 * connected fd. We collect the address of the connector in kernel
1462 * space and move it to user at the very end. This is unclean because
1463 * we open the socket then return an error.
1465 * 1003.1g adds the ability to recvmsg() to query connection pending
1466 * status to recvmsg. We need to add that support in a way thats
1467 * clean when we restucture accept also.
1470 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1471 int __user *, upeer_addrlen, int, flags)
1473 struct socket *sock, *newsock;
1474 struct file *newfile;
1475 int err, len, newfd, fput_needed;
1476 struct sockaddr_storage address;
1478 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1481 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1482 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1484 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1489 if (!(newsock = sock_alloc()))
1492 newsock->type = sock->type;
1493 newsock->ops = sock->ops;
1496 * We don't need try_module_get here, as the listening socket (sock)
1497 * has the protocol module (sock->ops->owner) held.
1499 __module_get(newsock->ops->owner);
1501 newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
1502 if (unlikely(newfd < 0)) {
1504 sock_release(newsock);
1508 err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
1512 err = security_socket_accept(sock, newsock);
1516 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1520 if (upeer_sockaddr) {
1521 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1523 err = -ECONNABORTED;
1526 err = move_addr_to_user((struct sockaddr *)&address,
1527 len, upeer_sockaddr, upeer_addrlen);
1532 /* File flags are not inherited via accept() unlike another OSes. */
1534 fd_install(newfd, newfile);
1538 fput_light(sock->file, fput_needed);
1542 sock_release(newsock);
1544 put_unused_fd(newfd);
1548 put_unused_fd(newfd);
1552 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1553 int __user *, upeer_addrlen)
1555 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1559 * Attempt to connect to a socket with the server address. The address
1560 * is in user space so we verify it is OK and move it to kernel space.
1562 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1565 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1566 * other SEQPACKET protocols that take time to connect() as it doesn't
1567 * include the -EINPROGRESS status for such sockets.
1570 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1573 struct socket *sock;
1574 struct sockaddr_storage address;
1575 int err, fput_needed;
1577 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1580 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1585 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1589 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1590 sock->file->f_flags);
1592 fput_light(sock->file, fput_needed);
1598 * Get the local address ('name') of a socket object. Move the obtained
1599 * name to user space.
1602 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1603 int __user *, usockaddr_len)
1605 struct socket *sock;
1606 struct sockaddr_storage address;
1607 int len, err, fput_needed;
1609 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1613 err = security_socket_getsockname(sock);
1617 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1620 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1623 fput_light(sock->file, fput_needed);
1629 * Get the remote address ('name') of a socket object. Move the obtained
1630 * name to user space.
1633 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1634 int __user *, usockaddr_len)
1636 struct socket *sock;
1637 struct sockaddr_storage address;
1638 int len, err, fput_needed;
1640 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1642 err = security_socket_getpeername(sock);
1644 fput_light(sock->file, fput_needed);
1649 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1652 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1654 fput_light(sock->file, fput_needed);
1660 * Send a datagram to a given address. We move the address into kernel
1661 * space and check the user space data area is readable before invoking
1665 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1666 unsigned, flags, struct sockaddr __user *, addr,
1669 struct socket *sock;
1670 struct sockaddr_storage address;
1676 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1680 iov.iov_base = buff;
1682 msg.msg_name = NULL;
1685 msg.msg_control = NULL;
1686 msg.msg_controllen = 0;
1687 msg.msg_namelen = 0;
1689 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1692 msg.msg_name = (struct sockaddr *)&address;
1693 msg.msg_namelen = addr_len;
1695 if (sock->file->f_flags & O_NONBLOCK)
1696 flags |= MSG_DONTWAIT;
1697 msg.msg_flags = flags;
1698 err = sock_sendmsg(sock, &msg, len);
1701 fput_light(sock->file, fput_needed);
1707 * Send a datagram down a socket.
1710 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1713 return sys_sendto(fd, buff, len, flags, NULL, 0);
1717 * Receive a frame from the socket and optionally record the address of the
1718 * sender. We verify the buffers are writable and if needed move the
1719 * sender address from kernel to user space.
1722 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1723 unsigned, flags, struct sockaddr __user *, addr,
1724 int __user *, addr_len)
1726 struct socket *sock;
1729 struct sockaddr_storage address;
1733 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1737 msg.msg_control = NULL;
1738 msg.msg_controllen = 0;
1742 iov.iov_base = ubuf;
1743 msg.msg_name = (struct sockaddr *)&address;
1744 msg.msg_namelen = sizeof(address);
1745 if (sock->file->f_flags & O_NONBLOCK)
1746 flags |= MSG_DONTWAIT;
1747 err = sock_recvmsg(sock, &msg, size, flags);
1749 if (err >= 0 && addr != NULL) {
1750 err2 = move_addr_to_user((struct sockaddr *)&address,
1751 msg.msg_namelen, addr, addr_len);
1756 fput_light(sock->file, fput_needed);
1762 * Receive a datagram from a socket.
1765 asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1768 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1772 * Set a socket option. Because we don't know the option lengths we have
1773 * to pass the user mode parameter for the protocols to sort out.
1776 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1777 char __user *, optval, int, optlen)
1779 int err, fput_needed;
1780 struct socket *sock;
1785 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1787 err = security_socket_setsockopt(sock, level, optname);
1791 if (level == SOL_SOCKET)
1793 sock_setsockopt(sock, level, optname, optval,
1797 sock->ops->setsockopt(sock, level, optname, optval,
1800 fput_light(sock->file, fput_needed);
1806 * Get a socket option. Because we don't know the option lengths we have
1807 * to pass a user mode parameter for the protocols to sort out.
1810 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1811 char __user *, optval, int __user *, optlen)
1813 int err, fput_needed;
1814 struct socket *sock;
1816 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1818 err = security_socket_getsockopt(sock, level, optname);
1822 if (level == SOL_SOCKET)
1824 sock_getsockopt(sock, level, optname, optval,
1828 sock->ops->getsockopt(sock, level, optname, optval,
1831 fput_light(sock->file, fput_needed);
1837 * Shutdown a socket.
1840 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1842 int err, fput_needed;
1843 struct socket *sock;
1845 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1847 err = security_socket_shutdown(sock, how);
1849 err = sock->ops->shutdown(sock, how);
1850 fput_light(sock->file, fput_needed);
1855 /* A couple of helpful macros for getting the address of the 32/64 bit
1856 * fields which are the same type (int / unsigned) on our platforms.
1858 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1859 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1860 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1863 * BSD sendmsg interface
1866 SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1868 struct compat_msghdr __user *msg_compat =
1869 (struct compat_msghdr __user *)msg;
1870 struct socket *sock;
1871 struct sockaddr_storage address;
1872 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1873 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1874 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1875 /* 20 is size of ipv6_pktinfo */
1876 unsigned char *ctl_buf = ctl;
1877 struct msghdr msg_sys;
1878 int err, ctl_len, iov_size, total_len;
1882 if (MSG_CMSG_COMPAT & flags) {
1883 if (get_compat_msghdr(&msg_sys, msg_compat))
1886 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1889 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1893 /* do not move before msg_sys is valid */
1895 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1898 /* Check whether to allocate the iovec area */
1900 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1901 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1902 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1907 /* This will also move the address data into kernel space */
1908 if (MSG_CMSG_COMPAT & flags) {
1909 err = verify_compat_iovec(&msg_sys, iov,
1910 (struct sockaddr *)&address,
1913 err = verify_iovec(&msg_sys, iov,
1914 (struct sockaddr *)&address,
1922 if (msg_sys.msg_controllen > INT_MAX)
1924 ctl_len = msg_sys.msg_controllen;
1925 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1927 cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1931 ctl_buf = msg_sys.msg_control;
1932 ctl_len = msg_sys.msg_controllen;
1933 } else if (ctl_len) {
1934 if (ctl_len > sizeof(ctl)) {
1935 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1936 if (ctl_buf == NULL)
1941 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1942 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1943 * checking falls down on this.
1945 if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1948 msg_sys.msg_control = ctl_buf;
1950 msg_sys.msg_flags = flags;
1952 if (sock->file->f_flags & O_NONBLOCK)
1953 msg_sys.msg_flags |= MSG_DONTWAIT;
1954 err = sock_sendmsg(sock, &msg_sys, total_len);
1958 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1960 if (iov != iovstack)
1961 sock_kfree_s(sock->sk, iov, iov_size);
1963 fput_light(sock->file, fput_needed);
1969 * BSD recvmsg interface
1972 SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
1973 unsigned int, flags)
1975 struct compat_msghdr __user *msg_compat =
1976 (struct compat_msghdr __user *)msg;
1977 struct socket *sock;
1978 struct iovec iovstack[UIO_FASTIOV];
1979 struct iovec *iov = iovstack;
1980 struct msghdr msg_sys;
1981 unsigned long cmsg_ptr;
1982 int err, iov_size, total_len, len;
1985 /* kernel mode address */
1986 struct sockaddr_storage addr;
1988 /* user mode address pointers */
1989 struct sockaddr __user *uaddr;
1990 int __user *uaddr_len;
1992 if (MSG_CMSG_COMPAT & flags) {
1993 if (get_compat_msghdr(&msg_sys, msg_compat))
1996 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1999 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2004 if (msg_sys.msg_iovlen > UIO_MAXIOV)
2007 /* Check whether to allocate the iovec area */
2009 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
2010 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
2011 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2017 * Save the user-mode address (verify_iovec will change the
2018 * kernel msghdr to use the kernel address space)
2021 uaddr = (__force void __user *)msg_sys.msg_name;
2022 uaddr_len = COMPAT_NAMELEN(msg);
2023 if (MSG_CMSG_COMPAT & flags) {
2024 err = verify_compat_iovec(&msg_sys, iov,
2025 (struct sockaddr *)&addr,
2028 err = verify_iovec(&msg_sys, iov,
2029 (struct sockaddr *)&addr,
2035 cmsg_ptr = (unsigned long)msg_sys.msg_control;
2036 msg_sys.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2038 if (sock->file->f_flags & O_NONBLOCK)
2039 flags |= MSG_DONTWAIT;
2040 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
2045 if (uaddr != NULL) {
2046 err = move_addr_to_user((struct sockaddr *)&addr,
2047 msg_sys.msg_namelen, uaddr,
2052 err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
2056 if (MSG_CMSG_COMPAT & flags)
2057 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2058 &msg_compat->msg_controllen);
2060 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2061 &msg->msg_controllen);
2067 if (iov != iovstack)
2068 sock_kfree_s(sock->sk, iov, iov_size);
2070 fput_light(sock->file, fput_needed);
2075 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2077 /* Argument list sizes for sys_socketcall */
2078 #define AL(x) ((x) * sizeof(unsigned long))
2079 static const unsigned char nargs[19]={
2080 AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2081 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
2082 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
2089 * System call vectors.
2091 * Argument checking cleaned up. Saved 20% in size.
2092 * This function doesn't need to set the kernel lock because
2093 * it is set by the callees.
2096 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2099 unsigned long a0, a1;
2103 if (call < 1 || call > SYS_ACCEPT4)
2107 if (len > sizeof(a))
2110 /* copy_from_user should be SMP safe. */
2111 if (copy_from_user(a, args, len))
2114 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2121 err = sys_socket(a0, a1, a[2]);
2124 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2127 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2130 err = sys_listen(a0, a1);
2133 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2134 (int __user *)a[2], 0);
2136 case SYS_GETSOCKNAME:
2138 sys_getsockname(a0, (struct sockaddr __user *)a1,
2139 (int __user *)a[2]);
2141 case SYS_GETPEERNAME:
2143 sys_getpeername(a0, (struct sockaddr __user *)a1,
2144 (int __user *)a[2]);
2146 case SYS_SOCKETPAIR:
2147 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2150 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2153 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2154 (struct sockaddr __user *)a[4], a[5]);
2157 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2160 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2161 (struct sockaddr __user *)a[4],
2162 (int __user *)a[5]);
2165 err = sys_shutdown(a0, a1);
2167 case SYS_SETSOCKOPT:
2168 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2170 case SYS_GETSOCKOPT:
2172 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2173 (int __user *)a[4]);
2176 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2179 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2182 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2183 (int __user *)a[2], a[3]);
2192 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
2195 * sock_register - add a socket protocol handler
2196 * @ops: description of protocol
2198 * This function is called by a protocol handler that wants to
2199 * advertise its address family, and have it linked into the
2200 * socket interface. The value ops->family coresponds to the
2201 * socket system call protocol family.
2203 int sock_register(const struct net_proto_family *ops)
2207 if (ops->family >= NPROTO) {
2208 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2213 spin_lock(&net_family_lock);
2214 if (net_families[ops->family])
2217 net_families[ops->family] = ops;
2220 spin_unlock(&net_family_lock);
2222 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2227 * sock_unregister - remove a protocol handler
2228 * @family: protocol family to remove
2230 * This function is called by a protocol handler that wants to
2231 * remove its address family, and have it unlinked from the
2232 * new socket creation.
2234 * If protocol handler is a module, then it can use module reference
2235 * counts to protect against new references. If protocol handler is not
2236 * a module then it needs to provide its own protection in
2237 * the ops->create routine.
2239 void sock_unregister(int family)
2241 BUG_ON(family < 0 || family >= NPROTO);
2243 spin_lock(&net_family_lock);
2244 net_families[family] = NULL;
2245 spin_unlock(&net_family_lock);
2249 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2252 static int __init sock_init(void)
2255 * Initialize sock SLAB cache.
2261 * Initialize skbuff SLAB cache
2266 * Initialize the protocols module.
2270 register_filesystem(&sock_fs_type);
2271 sock_mnt = kern_mount(&sock_fs_type);
2273 /* The real protocol initialization is performed in later initcalls.
2276 #ifdef CONFIG_NETFILTER
2283 core_initcall(sock_init); /* early initcall */
2285 #ifdef CONFIG_PROC_FS
2286 void socket_seq_show(struct seq_file *seq)
2291 for_each_possible_cpu(cpu)
2292 counter += per_cpu(sockets_in_use, cpu);
2294 /* It can be negative, by the way. 8) */
2298 seq_printf(seq, "sockets: used %d\n", counter);
2300 #endif /* CONFIG_PROC_FS */
2302 #ifdef CONFIG_COMPAT
2303 static long compat_sock_ioctl(struct file *file, unsigned cmd,
2306 struct socket *sock = file->private_data;
2307 int ret = -ENOIOCTLCMD;
2314 if (sock->ops->compat_ioctl)
2315 ret = sock->ops->compat_ioctl(sock, cmd, arg);
2317 if (ret == -ENOIOCTLCMD &&
2318 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2319 ret = compat_wext_handle_ioctl(net, cmd, arg);
2325 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2327 return sock->ops->bind(sock, addr, addrlen);
2330 int kernel_listen(struct socket *sock, int backlog)
2332 return sock->ops->listen(sock, backlog);
2335 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2337 struct sock *sk = sock->sk;
2340 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2345 err = sock->ops->accept(sock, *newsock, flags);
2347 sock_release(*newsock);
2352 (*newsock)->ops = sock->ops;
2353 __module_get((*newsock)->ops->owner);
2359 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
2362 return sock->ops->connect(sock, addr, addrlen, flags);
2365 int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2368 return sock->ops->getname(sock, addr, addrlen, 0);
2371 int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2374 return sock->ops->getname(sock, addr, addrlen, 1);
2377 int kernel_getsockopt(struct socket *sock, int level, int optname,
2378 char *optval, int *optlen)
2380 mm_segment_t oldfs = get_fs();
2384 if (level == SOL_SOCKET)
2385 err = sock_getsockopt(sock, level, optname, optval, optlen);
2387 err = sock->ops->getsockopt(sock, level, optname, optval,
2393 int kernel_setsockopt(struct socket *sock, int level, int optname,
2394 char *optval, unsigned int optlen)
2396 mm_segment_t oldfs = get_fs();
2400 if (level == SOL_SOCKET)
2401 err = sock_setsockopt(sock, level, optname, optval, optlen);
2403 err = sock->ops->setsockopt(sock, level, optname, optval,
2409 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2410 size_t size, int flags)
2412 if (sock->ops->sendpage)
2413 return sock->ops->sendpage(sock, page, offset, size, flags);
2415 return sock_no_sendpage(sock, page, offset, size, flags);
2418 int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2420 mm_segment_t oldfs = get_fs();
2424 err = sock->ops->ioctl(sock, cmd, arg);
2430 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2432 return sock->ops->shutdown(sock, how);
2435 EXPORT_SYMBOL(sock_create);
2436 EXPORT_SYMBOL(sock_create_kern);
2437 EXPORT_SYMBOL(sock_create_lite);
2438 EXPORT_SYMBOL(sock_map_fd);
2439 EXPORT_SYMBOL(sock_recvmsg);
2440 EXPORT_SYMBOL(sock_register);
2441 EXPORT_SYMBOL(sock_release);
2442 EXPORT_SYMBOL(sock_sendmsg);
2443 EXPORT_SYMBOL(sock_unregister);
2444 EXPORT_SYMBOL(sock_wake_async);
2445 EXPORT_SYMBOL(sockfd_lookup);
2446 EXPORT_SYMBOL(kernel_sendmsg);
2447 EXPORT_SYMBOL(kernel_recvmsg);
2448 EXPORT_SYMBOL(kernel_bind);
2449 EXPORT_SYMBOL(kernel_listen);
2450 EXPORT_SYMBOL(kernel_accept);
2451 EXPORT_SYMBOL(kernel_connect);
2452 EXPORT_SYMBOL(kernel_getsockname);
2453 EXPORT_SYMBOL(kernel_getpeername);
2454 EXPORT_SYMBOL(kernel_getsockopt);
2455 EXPORT_SYMBOL(kernel_setsockopt);
2456 EXPORT_SYMBOL(kernel_sendpage);
2457 EXPORT_SYMBOL(kernel_sock_ioctl);
2458 EXPORT_SYMBOL(kernel_sock_shutdown);