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>
90 #include <asm/uaccess.h>
91 #include <asm/unistd.h>
93 #include <net/compat.h>
97 #include <linux/netfilter.h>
99 static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
100 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
101 unsigned long nr_segs, loff_t pos);
102 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
103 unsigned long nr_segs, loff_t pos);
104 static int sock_mmap(struct file *file, struct vm_area_struct *vma);
106 static int sock_close(struct inode *inode, struct file *file);
107 static unsigned int sock_poll(struct file *file,
108 struct poll_table_struct *wait);
109 static long sock_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
111 static long compat_sock_ioctl(struct file *file,
112 unsigned int cmd, unsigned long arg);
114 static int sock_fasync(int fd, struct file *filp, int on);
115 static ssize_t sock_sendpage(struct file *file, struct page *page,
116 int offset, size_t size, loff_t *ppos, int more);
117 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
118 struct pipe_inode_info *pipe, size_t len,
122 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
123 * in the operation structures but are done directly via the socketcall() multiplexor.
126 static const struct file_operations socket_file_ops = {
127 .owner = THIS_MODULE,
129 .aio_read = sock_aio_read,
130 .aio_write = sock_aio_write,
132 .unlocked_ioctl = sock_ioctl,
134 .compat_ioctl = compat_sock_ioctl,
137 .open = sock_no_open, /* special open code to disallow open via /proc */
138 .release = sock_close,
139 .fasync = sock_fasync,
140 .sendpage = sock_sendpage,
141 .splice_write = generic_splice_sendpage,
142 .splice_read = sock_splice_read,
146 * The protocol list. Each protocol is registered in here.
149 static DEFINE_SPINLOCK(net_family_lock);
150 static const struct net_proto_family *net_families[NPROTO] __read_mostly;
153 * Statistics counters of the socket lists
156 static DEFINE_PER_CPU(int, sockets_in_use) = 0;
160 * Move socket addresses back and forth across the kernel/user
161 * divide and look after the messy bits.
164 #define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
165 16 for IP, 16 for IPX,
168 must be at least one bigger than
169 the AF_UNIX size (see net/unix/af_unix.c
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
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.
184 int move_addr_to_kernel(void __user *uaddr, int ulen, struct sockaddr *kaddr)
186 if (ulen < 0 || ulen > sizeof(struct sockaddr_storage))
190 if (copy_from_user(kaddr, uaddr, ulen))
192 return audit_sockaddr(ulen, kaddr);
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
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
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.
212 int move_addr_to_user(struct sockaddr *kaddr, int klen, void __user *uaddr,
218 err = get_user(len, ulen);
223 if (len < 0 || len > sizeof(struct sockaddr_storage))
226 if (audit_sockaddr(klen, kaddr))
228 if (copy_to_user(uaddr, kaddr, len))
232 * "fromlen shall refer to the value before truncation.."
235 return __put_user(klen, ulen);
238 #define SOCKFS_MAGIC 0x534F434B
240 static struct kmem_cache *sock_inode_cachep __read_mostly;
242 static struct inode *sock_alloc_inode(struct super_block *sb)
244 struct socket_alloc *ei;
246 ei = kmem_cache_alloc(sock_inode_cachep, GFP_KERNEL);
249 init_waitqueue_head(&ei->socket.wait);
251 ei->socket.fasync_list = NULL;
252 ei->socket.state = SS_UNCONNECTED;
253 ei->socket.flags = 0;
254 ei->socket.ops = NULL;
255 ei->socket.sk = NULL;
256 ei->socket.file = NULL;
258 return &ei->vfs_inode;
261 static void sock_destroy_inode(struct inode *inode)
263 kmem_cache_free(sock_inode_cachep,
264 container_of(inode, struct socket_alloc, vfs_inode));
267 static void init_once(void *foo)
269 struct socket_alloc *ei = (struct socket_alloc *)foo;
271 inode_init_once(&ei->vfs_inode);
274 static int init_inodecache(void)
276 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
277 sizeof(struct socket_alloc),
279 (SLAB_HWCACHE_ALIGN |
280 SLAB_RECLAIM_ACCOUNT |
283 if (sock_inode_cachep == NULL)
288 static struct super_operations sockfs_ops = {
289 .alloc_inode = sock_alloc_inode,
290 .destroy_inode =sock_destroy_inode,
291 .statfs = simple_statfs,
294 static int sockfs_get_sb(struct file_system_type *fs_type,
295 int flags, const char *dev_name, void *data,
296 struct vfsmount *mnt)
298 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC,
302 static struct vfsmount *sock_mnt __read_mostly;
304 static struct file_system_type sock_fs_type = {
306 .get_sb = sockfs_get_sb,
307 .kill_sb = kill_anon_super,
310 static int sockfs_delete_dentry(struct dentry *dentry)
313 * At creation time, we pretended this dentry was hashed
314 * (by clearing DCACHE_UNHASHED bit in d_flags)
315 * At delete time, we restore the truth : not hashed.
316 * (so that dput() can proceed correctly)
318 dentry->d_flags |= DCACHE_UNHASHED;
323 * sockfs_dname() is called from d_path().
325 static char *sockfs_dname(struct dentry *dentry, char *buffer, int buflen)
327 return dynamic_dname(dentry, buffer, buflen, "socket:[%lu]",
328 dentry->d_inode->i_ino);
331 static struct dentry_operations sockfs_dentry_operations = {
332 .d_delete = sockfs_delete_dentry,
333 .d_dname = sockfs_dname,
337 * Obtains the first available file descriptor and sets it up for use.
339 * These functions create file structures and maps them to fd space
340 * of the current process. On success it returns file descriptor
341 * and file struct implicitly stored in sock->file.
342 * Note that another thread may close file descriptor before we return
343 * from this function. We use the fact that now we do not refer
344 * to socket after mapping. If one day we will need it, this
345 * function will increment ref. count on file by 1.
347 * In any case returned fd MAY BE not valid!
348 * This race condition is unavoidable
349 * with shared fd spaces, we cannot solve it inside kernel,
350 * but we take care of internal coherence yet.
353 static int sock_alloc_fd(struct file **filep, int flags)
357 fd = get_unused_fd_flags(flags);
358 if (likely(fd >= 0)) {
359 struct file *file = get_empty_filp();
362 if (unlikely(!file)) {
371 static int sock_attach_fd(struct socket *sock, struct file *file, int flags)
373 struct dentry *dentry;
374 struct qstr name = { .name = "" };
376 dentry = d_alloc(sock_mnt->mnt_sb->s_root, &name);
377 if (unlikely(!dentry))
380 dentry->d_op = &sockfs_dentry_operations;
382 * We dont want to push this dentry into global dentry hash table.
383 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
384 * This permits a working /proc/$pid/fd/XXX on sockets
386 dentry->d_flags &= ~DCACHE_UNHASHED;
387 d_instantiate(dentry, SOCK_INODE(sock));
390 init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE,
392 SOCK_INODE(sock)->i_fop = &socket_file_ops;
393 file->f_flags = O_RDWR | (flags & O_NONBLOCK);
395 file->private_data = sock;
400 int sock_map_fd(struct socket *sock, int flags)
402 struct file *newfile;
403 int fd = sock_alloc_fd(&newfile, flags);
405 if (likely(fd >= 0)) {
406 int err = sock_attach_fd(sock, newfile, flags);
408 if (unlikely(err < 0)) {
413 fd_install(fd, newfile);
418 static struct socket *sock_from_file(struct file *file, int *err)
420 if (file->f_op == &socket_file_ops)
421 return file->private_data; /* set in sock_map_fd */
428 * sockfd_lookup - Go from a file number to its socket slot
430 * @err: pointer to an error code return
432 * The file handle passed in is locked and the socket it is bound
433 * too is returned. If an error occurs the err pointer is overwritten
434 * with a negative errno code and NULL is returned. The function checks
435 * for both invalid handles and passing a handle which is not a socket.
437 * On a success the socket object pointer is returned.
440 struct socket *sockfd_lookup(int fd, int *err)
451 sock = sock_from_file(file, err);
457 static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)
463 file = fget_light(fd, fput_needed);
465 sock = sock_from_file(file, err);
468 fput_light(file, *fput_needed);
474 * sock_alloc - allocate a socket
476 * Allocate a new inode and socket object. The two are bound together
477 * and initialised. The socket is then returned. If we are out of inodes
481 static struct socket *sock_alloc(void)
486 inode = new_inode(sock_mnt->mnt_sb);
490 sock = SOCKET_I(inode);
492 inode->i_mode = S_IFSOCK | S_IRWXUGO;
493 inode->i_uid = current_fsuid();
494 inode->i_gid = current_fsgid();
496 get_cpu_var(sockets_in_use)++;
497 put_cpu_var(sockets_in_use);
502 * In theory you can't get an open on this inode, but /proc provides
503 * a back door. Remember to keep it shut otherwise you'll let the
504 * creepy crawlies in.
507 static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
512 const struct file_operations bad_sock_fops = {
513 .owner = THIS_MODULE,
514 .open = sock_no_open,
518 * sock_release - close a socket
519 * @sock: socket to close
521 * The socket is released from the protocol stack if it has a release
522 * callback, and the inode is then released if the socket is bound to
523 * an inode not a file.
526 void sock_release(struct socket *sock)
529 struct module *owner = sock->ops->owner;
531 sock->ops->release(sock);
536 if (sock->fasync_list)
537 printk(KERN_ERR "sock_release: fasync list not empty!\n");
539 get_cpu_var(sockets_in_use)--;
540 put_cpu_var(sockets_in_use);
542 iput(SOCK_INODE(sock));
548 int sock_tx_timestamp(struct msghdr *msg, struct sock *sk,
549 union skb_shared_tx *shtx)
552 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
554 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
558 EXPORT_SYMBOL(sock_tx_timestamp);
560 static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
561 struct msghdr *msg, size_t size)
563 struct sock_iocb *si = kiocb_to_siocb(iocb);
571 err = security_socket_sendmsg(sock, msg, size);
575 return sock->ops->sendmsg(iocb, sock, msg, size);
578 int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
581 struct sock_iocb siocb;
584 init_sync_kiocb(&iocb, NULL);
585 iocb.private = &siocb;
586 ret = __sock_sendmsg(&iocb, sock, msg, size);
587 if (-EIOCBQUEUED == ret)
588 ret = wait_on_sync_kiocb(&iocb);
592 int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
593 struct kvec *vec, size_t num, size_t size)
595 mm_segment_t oldfs = get_fs();
600 * the following is safe, since for compiler definitions of kvec and
601 * iovec are identical, yielding the same in-core layout and alignment
603 msg->msg_iov = (struct iovec *)vec;
604 msg->msg_iovlen = num;
605 result = sock_sendmsg(sock, msg, size);
610 static int ktime2ts(ktime_t kt, struct timespec *ts)
613 *ts = ktime_to_timespec(kt);
621 * called from sock_recv_timestamp() if sock_flag(sk, SOCK_RCVTSTAMP)
623 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
626 int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
627 struct timespec ts[3];
629 struct skb_shared_hwtstamps *shhwtstamps =
632 /* Race occurred between timestamp enabling and packet
633 receiving. Fill in the current time for now. */
634 if (need_software_tstamp && skb->tstamp.tv64 == 0)
635 __net_timestamp(skb);
637 if (need_software_tstamp) {
638 if (!sock_flag(sk, SOCK_RCVTSTAMPNS)) {
640 skb_get_timestamp(skb, &tv);
641 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
645 skb_get_timestampns(skb, &ts);
646 put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
652 memset(ts, 0, sizeof(ts));
653 if (skb->tstamp.tv64 &&
654 sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) {
655 skb_get_timestampns(skb, ts + 0);
659 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE) &&
660 ktime2ts(shhwtstamps->syststamp, ts + 1))
662 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
663 ktime2ts(shhwtstamps->hwtstamp, ts + 2))
667 put_cmsg(msg, SOL_SOCKET,
668 SCM_TIMESTAMPING, sizeof(ts), &ts);
671 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);
673 static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
674 struct msghdr *msg, size_t size, int flags)
677 struct sock_iocb *si = kiocb_to_siocb(iocb);
685 err = security_socket_recvmsg(sock, msg, size, flags);
689 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
692 int sock_recvmsg(struct socket *sock, struct msghdr *msg,
693 size_t size, int flags)
696 struct sock_iocb siocb;
699 init_sync_kiocb(&iocb, NULL);
700 iocb.private = &siocb;
701 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
702 if (-EIOCBQUEUED == ret)
703 ret = wait_on_sync_kiocb(&iocb);
707 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
708 struct kvec *vec, size_t num, size_t size, int flags)
710 mm_segment_t oldfs = get_fs();
715 * the following is safe, since for compiler definitions of kvec and
716 * iovec are identical, yielding the same in-core layout and alignment
718 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
719 result = sock_recvmsg(sock, msg, size, flags);
724 static void sock_aio_dtor(struct kiocb *iocb)
726 kfree(iocb->private);
729 static ssize_t sock_sendpage(struct file *file, struct page *page,
730 int offset, size_t size, loff_t *ppos, int more)
735 sock = file->private_data;
737 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
741 return sock->ops->sendpage(sock, page, offset, size, flags);
744 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
745 struct pipe_inode_info *pipe, size_t len,
748 struct socket *sock = file->private_data;
750 if (unlikely(!sock->ops->splice_read))
753 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
756 static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
757 struct sock_iocb *siocb)
759 if (!is_sync_kiocb(iocb)) {
760 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
763 iocb->ki_dtor = sock_aio_dtor;
767 iocb->private = siocb;
771 static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
772 struct file *file, const struct iovec *iov,
773 unsigned long nr_segs)
775 struct socket *sock = file->private_data;
779 for (i = 0; i < nr_segs; i++)
780 size += iov[i].iov_len;
782 msg->msg_name = NULL;
783 msg->msg_namelen = 0;
784 msg->msg_control = NULL;
785 msg->msg_controllen = 0;
786 msg->msg_iov = (struct iovec *)iov;
787 msg->msg_iovlen = nr_segs;
788 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
790 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
793 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
794 unsigned long nr_segs, loff_t pos)
796 struct sock_iocb siocb, *x;
801 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
805 x = alloc_sock_iocb(iocb, &siocb);
808 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
811 static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
812 struct file *file, const struct iovec *iov,
813 unsigned long nr_segs)
815 struct socket *sock = file->private_data;
819 for (i = 0; i < nr_segs; i++)
820 size += iov[i].iov_len;
822 msg->msg_name = NULL;
823 msg->msg_namelen = 0;
824 msg->msg_control = NULL;
825 msg->msg_controllen = 0;
826 msg->msg_iov = (struct iovec *)iov;
827 msg->msg_iovlen = nr_segs;
828 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
829 if (sock->type == SOCK_SEQPACKET)
830 msg->msg_flags |= MSG_EOR;
832 return __sock_sendmsg(iocb, sock, msg, size);
835 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
836 unsigned long nr_segs, loff_t pos)
838 struct sock_iocb siocb, *x;
843 x = alloc_sock_iocb(iocb, &siocb);
847 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
851 * Atomic setting of ioctl hooks to avoid race
852 * with module unload.
855 static DEFINE_MUTEX(br_ioctl_mutex);
856 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
858 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
860 mutex_lock(&br_ioctl_mutex);
861 br_ioctl_hook = hook;
862 mutex_unlock(&br_ioctl_mutex);
865 EXPORT_SYMBOL(brioctl_set);
867 static DEFINE_MUTEX(vlan_ioctl_mutex);
868 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
870 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
872 mutex_lock(&vlan_ioctl_mutex);
873 vlan_ioctl_hook = hook;
874 mutex_unlock(&vlan_ioctl_mutex);
877 EXPORT_SYMBOL(vlan_ioctl_set);
879 static DEFINE_MUTEX(dlci_ioctl_mutex);
880 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
882 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
884 mutex_lock(&dlci_ioctl_mutex);
885 dlci_ioctl_hook = hook;
886 mutex_unlock(&dlci_ioctl_mutex);
889 EXPORT_SYMBOL(dlci_ioctl_set);
892 * With an ioctl, arg may well be a user mode pointer, but we don't know
893 * what to do with it - that's up to the protocol still.
896 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
900 void __user *argp = (void __user *)arg;
904 sock = file->private_data;
907 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
908 err = dev_ioctl(net, cmd, argp);
910 #ifdef CONFIG_WIRELESS_EXT
911 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
912 err = dev_ioctl(net, cmd, argp);
914 #endif /* CONFIG_WIRELESS_EXT */
919 if (get_user(pid, (int __user *)argp))
921 err = f_setown(sock->file, pid, 1);
925 err = put_user(f_getown(sock->file),
934 request_module("bridge");
936 mutex_lock(&br_ioctl_mutex);
938 err = br_ioctl_hook(net, cmd, argp);
939 mutex_unlock(&br_ioctl_mutex);
944 if (!vlan_ioctl_hook)
945 request_module("8021q");
947 mutex_lock(&vlan_ioctl_mutex);
949 err = vlan_ioctl_hook(net, argp);
950 mutex_unlock(&vlan_ioctl_mutex);
955 if (!dlci_ioctl_hook)
956 request_module("dlci");
958 mutex_lock(&dlci_ioctl_mutex);
960 err = dlci_ioctl_hook(cmd, argp);
961 mutex_unlock(&dlci_ioctl_mutex);
964 err = sock->ops->ioctl(sock, cmd, arg);
967 * If this ioctl is unknown try to hand it down
970 if (err == -ENOIOCTLCMD)
971 err = dev_ioctl(net, cmd, argp);
977 int sock_create_lite(int family, int type, int protocol, struct socket **res)
980 struct socket *sock = NULL;
982 err = security_socket_create(family, type, protocol, 1);
993 err = security_socket_post_create(sock, family, type, protocol, 1);
1006 /* No kernel lock held - perfect */
1007 static unsigned int sock_poll(struct file *file, poll_table *wait)
1009 struct socket *sock;
1012 * We can't return errors to poll, so it's either yes or no.
1014 sock = file->private_data;
1015 return sock->ops->poll(file, sock, wait);
1018 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1020 struct socket *sock = file->private_data;
1022 return sock->ops->mmap(file, sock, vma);
1025 static int sock_close(struct inode *inode, struct file *filp)
1028 * It was possible the inode is NULL we were
1029 * closing an unfinished socket.
1033 printk(KERN_DEBUG "sock_close: NULL inode\n");
1036 sock_release(SOCKET_I(inode));
1041 * Update the socket async list
1043 * Fasync_list locking strategy.
1045 * 1. fasync_list is modified only under process context socket lock
1046 * i.e. under semaphore.
1047 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1048 * or under socket lock.
1049 * 3. fasync_list can be used from softirq context, so that
1050 * modification under socket lock have to be enhanced with
1051 * write_lock_bh(&sk->sk_callback_lock).
1055 static int sock_fasync(int fd, struct file *filp, int on)
1057 struct fasync_struct *fa, *fna = NULL, **prev;
1058 struct socket *sock;
1062 fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
1067 sock = filp->private_data;
1077 prev = &(sock->fasync_list);
1079 for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1080 if (fa->fa_file == filp)
1085 write_lock_bh(&sk->sk_callback_lock);
1087 write_unlock_bh(&sk->sk_callback_lock);
1092 fna->fa_file = filp;
1094 fna->magic = FASYNC_MAGIC;
1095 fna->fa_next = sock->fasync_list;
1096 write_lock_bh(&sk->sk_callback_lock);
1097 sock->fasync_list = fna;
1098 write_unlock_bh(&sk->sk_callback_lock);
1101 write_lock_bh(&sk->sk_callback_lock);
1102 *prev = fa->fa_next;
1103 write_unlock_bh(&sk->sk_callback_lock);
1109 release_sock(sock->sk);
1113 /* This function may be called only under socket lock or callback_lock */
1115 int sock_wake_async(struct socket *sock, int how, int band)
1117 if (!sock || !sock->fasync_list)
1120 case SOCK_WAKE_WAITD:
1121 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1124 case SOCK_WAKE_SPACE:
1125 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1130 __kill_fasync(sock->fasync_list, SIGIO, band);
1133 __kill_fasync(sock->fasync_list, SIGURG, band);
1138 static int __sock_create(struct net *net, int family, int type, int protocol,
1139 struct socket **res, int kern)
1142 struct socket *sock;
1143 const struct net_proto_family *pf;
1146 * Check protocol is in range
1148 if (family < 0 || family >= NPROTO)
1149 return -EAFNOSUPPORT;
1150 if (type < 0 || type >= SOCK_MAX)
1155 This uglymoron is moved from INET layer to here to avoid
1156 deadlock in module load.
1158 if (family == PF_INET && type == SOCK_PACKET) {
1162 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1168 err = security_socket_create(family, type, protocol, kern);
1173 * Allocate the socket and allow the family to set things up. if
1174 * the protocol is 0, the family is instructed to select an appropriate
1177 sock = sock_alloc();
1179 if (net_ratelimit())
1180 printk(KERN_WARNING "socket: no more sockets\n");
1181 return -ENFILE; /* Not exactly a match, but its the
1182 closest posix thing */
1187 #ifdef CONFIG_MODULES
1188 /* Attempt to load a protocol module if the find failed.
1190 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1191 * requested real, full-featured networking support upon configuration.
1192 * Otherwise module support will break!
1194 if (net_families[family] == NULL)
1195 request_module("net-pf-%d", family);
1199 pf = rcu_dereference(net_families[family]);
1200 err = -EAFNOSUPPORT;
1205 * We will call the ->create function, that possibly is in a loadable
1206 * module, so we have to bump that loadable module refcnt first.
1208 if (!try_module_get(pf->owner))
1211 /* Now protected by module ref count */
1214 err = pf->create(net, sock, protocol);
1216 goto out_module_put;
1219 * Now to bump the refcnt of the [loadable] module that owns this
1220 * socket at sock_release time we decrement its refcnt.
1222 if (!try_module_get(sock->ops->owner))
1223 goto out_module_busy;
1226 * Now that we're done with the ->create function, the [loadable]
1227 * module can have its refcnt decremented
1229 module_put(pf->owner);
1230 err = security_socket_post_create(sock, family, type, protocol, kern);
1232 goto out_sock_release;
1238 err = -EAFNOSUPPORT;
1241 module_put(pf->owner);
1248 goto out_sock_release;
1251 int sock_create(int family, int type, int protocol, struct socket **res)
1253 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1256 int sock_create_kern(int family, int type, int protocol, struct socket **res)
1258 return __sock_create(&init_net, family, type, protocol, res, 1);
1261 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1264 struct socket *sock;
1267 /* Check the SOCK_* constants for consistency. */
1268 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1269 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1270 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1271 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1273 flags = type & ~SOCK_TYPE_MASK;
1274 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1276 type &= SOCK_TYPE_MASK;
1278 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1279 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1281 retval = sock_create(family, type, protocol, &sock);
1285 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1290 /* It may be already another descriptor 8) Not kernel problem. */
1299 * Create a pair of connected sockets.
1302 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1303 int __user *, usockvec)
1305 struct socket *sock1, *sock2;
1307 struct file *newfile1, *newfile2;
1310 flags = type & ~SOCK_TYPE_MASK;
1311 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1313 type &= SOCK_TYPE_MASK;
1315 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1316 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1319 * Obtain the first socket and check if the underlying protocol
1320 * supports the socketpair call.
1323 err = sock_create(family, type, protocol, &sock1);
1327 err = sock_create(family, type, protocol, &sock2);
1331 err = sock1->ops->socketpair(sock1, sock2);
1333 goto out_release_both;
1335 fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
1336 if (unlikely(fd1 < 0)) {
1338 goto out_release_both;
1341 fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
1342 if (unlikely(fd2 < 0)) {
1346 goto out_release_both;
1349 err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
1350 if (unlikely(err < 0)) {
1354 err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
1355 if (unlikely(err < 0)) {
1360 audit_fd_pair(fd1, fd2);
1361 fd_install(fd1, newfile1);
1362 fd_install(fd2, newfile2);
1363 /* fd1 and fd2 may be already another descriptors.
1364 * Not kernel problem.
1367 err = put_user(fd1, &usockvec[0]);
1369 err = put_user(fd2, &usockvec[1]);
1378 sock_release(sock2);
1380 sock_release(sock1);
1386 sock_release(sock1);
1389 sock_release(sock2);
1396 * Bind a name to a socket. Nothing much to do here since it's
1397 * the protocol's responsibility to handle the local address.
1399 * We move the socket address to kernel space before we call
1400 * the protocol layer (having also checked the address is ok).
1403 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1405 struct socket *sock;
1406 struct sockaddr_storage address;
1407 int err, fput_needed;
1409 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1411 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1413 err = security_socket_bind(sock,
1414 (struct sockaddr *)&address,
1417 err = sock->ops->bind(sock,
1421 fput_light(sock->file, fput_needed);
1427 * Perform a listen. Basically, we allow the protocol to do anything
1428 * necessary for a listen, and if that works, we mark the socket as
1429 * ready for listening.
1432 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1434 struct socket *sock;
1435 int err, fput_needed;
1438 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1440 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1441 if ((unsigned)backlog > somaxconn)
1442 backlog = somaxconn;
1444 err = security_socket_listen(sock, backlog);
1446 err = sock->ops->listen(sock, backlog);
1448 fput_light(sock->file, fput_needed);
1454 * For accept, we attempt to create a new socket, set up the link
1455 * with the client, wake up the client, then return the new
1456 * connected fd. We collect the address of the connector in kernel
1457 * space and move it to user at the very end. This is unclean because
1458 * we open the socket then return an error.
1460 * 1003.1g adds the ability to recvmsg() to query connection pending
1461 * status to recvmsg. We need to add that support in a way thats
1462 * clean when we restucture accept also.
1465 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1466 int __user *, upeer_addrlen, int, flags)
1468 struct socket *sock, *newsock;
1469 struct file *newfile;
1470 int err, len, newfd, fput_needed;
1471 struct sockaddr_storage address;
1473 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1476 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1477 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1479 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1484 if (!(newsock = sock_alloc()))
1487 newsock->type = sock->type;
1488 newsock->ops = sock->ops;
1491 * We don't need try_module_get here, as the listening socket (sock)
1492 * has the protocol module (sock->ops->owner) held.
1494 __module_get(newsock->ops->owner);
1496 newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
1497 if (unlikely(newfd < 0)) {
1499 sock_release(newsock);
1503 err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
1507 err = security_socket_accept(sock, newsock);
1511 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1515 if (upeer_sockaddr) {
1516 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1518 err = -ECONNABORTED;
1521 err = move_addr_to_user((struct sockaddr *)&address,
1522 len, upeer_sockaddr, upeer_addrlen);
1527 /* File flags are not inherited via accept() unlike another OSes. */
1529 fd_install(newfd, newfile);
1532 security_socket_post_accept(sock, newsock);
1535 fput_light(sock->file, fput_needed);
1539 sock_release(newsock);
1541 put_unused_fd(newfd);
1545 put_unused_fd(newfd);
1549 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1550 int __user *, upeer_addrlen)
1552 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1556 * Attempt to connect to a socket with the server address. The address
1557 * is in user space so we verify it is OK and move it to kernel space.
1559 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1562 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1563 * other SEQPACKET protocols that take time to connect() as it doesn't
1564 * include the -EINPROGRESS status for such sockets.
1567 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1570 struct socket *sock;
1571 struct sockaddr_storage address;
1572 int err, fput_needed;
1574 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1577 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1582 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1586 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1587 sock->file->f_flags);
1589 fput_light(sock->file, fput_needed);
1595 * Get the local address ('name') of a socket object. Move the obtained
1596 * name to user space.
1599 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1600 int __user *, usockaddr_len)
1602 struct socket *sock;
1603 struct sockaddr_storage address;
1604 int len, err, fput_needed;
1606 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1610 err = security_socket_getsockname(sock);
1614 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1617 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1620 fput_light(sock->file, fput_needed);
1626 * Get the remote address ('name') of a socket object. Move the obtained
1627 * name to user space.
1630 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1631 int __user *, usockaddr_len)
1633 struct socket *sock;
1634 struct sockaddr_storage address;
1635 int len, err, fput_needed;
1637 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1639 err = security_socket_getpeername(sock);
1641 fput_light(sock->file, fput_needed);
1646 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1649 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1651 fput_light(sock->file, fput_needed);
1657 * Send a datagram to a given address. We move the address into kernel
1658 * space and check the user space data area is readable before invoking
1662 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1663 unsigned, flags, struct sockaddr __user *, addr,
1666 struct socket *sock;
1667 struct sockaddr_storage address;
1673 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1677 iov.iov_base = buff;
1679 msg.msg_name = NULL;
1682 msg.msg_control = NULL;
1683 msg.msg_controllen = 0;
1684 msg.msg_namelen = 0;
1686 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1689 msg.msg_name = (struct sockaddr *)&address;
1690 msg.msg_namelen = addr_len;
1692 if (sock->file->f_flags & O_NONBLOCK)
1693 flags |= MSG_DONTWAIT;
1694 msg.msg_flags = flags;
1695 err = sock_sendmsg(sock, &msg, len);
1698 fput_light(sock->file, fput_needed);
1704 * Send a datagram down a socket.
1707 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1710 return sys_sendto(fd, buff, len, flags, NULL, 0);
1714 * Receive a frame from the socket and optionally record the address of the
1715 * sender. We verify the buffers are writable and if needed move the
1716 * sender address from kernel to user space.
1719 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1720 unsigned, flags, struct sockaddr __user *, addr,
1721 int __user *, addr_len)
1723 struct socket *sock;
1726 struct sockaddr_storage address;
1730 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1734 msg.msg_control = NULL;
1735 msg.msg_controllen = 0;
1739 iov.iov_base = ubuf;
1740 msg.msg_name = (struct sockaddr *)&address;
1741 msg.msg_namelen = sizeof(address);
1742 if (sock->file->f_flags & O_NONBLOCK)
1743 flags |= MSG_DONTWAIT;
1744 err = sock_recvmsg(sock, &msg, size, flags);
1746 if (err >= 0 && addr != NULL) {
1747 err2 = move_addr_to_user((struct sockaddr *)&address,
1748 msg.msg_namelen, addr, addr_len);
1753 fput_light(sock->file, fput_needed);
1759 * Receive a datagram from a socket.
1762 asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1765 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1769 * Set a socket option. Because we don't know the option lengths we have
1770 * to pass the user mode parameter for the protocols to sort out.
1773 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1774 char __user *, optval, int, optlen)
1776 int err, fput_needed;
1777 struct socket *sock;
1782 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1784 err = security_socket_setsockopt(sock, level, optname);
1788 if (level == SOL_SOCKET)
1790 sock_setsockopt(sock, level, optname, optval,
1794 sock->ops->setsockopt(sock, level, optname, optval,
1797 fput_light(sock->file, fput_needed);
1803 * Get a socket option. Because we don't know the option lengths we have
1804 * to pass a user mode parameter for the protocols to sort out.
1807 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1808 char __user *, optval, int __user *, optlen)
1810 int err, fput_needed;
1811 struct socket *sock;
1813 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1815 err = security_socket_getsockopt(sock, level, optname);
1819 if (level == SOL_SOCKET)
1821 sock_getsockopt(sock, level, optname, optval,
1825 sock->ops->getsockopt(sock, level, optname, optval,
1828 fput_light(sock->file, fput_needed);
1834 * Shutdown a socket.
1837 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1839 int err, fput_needed;
1840 struct socket *sock;
1842 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1844 err = security_socket_shutdown(sock, how);
1846 err = sock->ops->shutdown(sock, how);
1847 fput_light(sock->file, fput_needed);
1852 /* A couple of helpful macros for getting the address of the 32/64 bit
1853 * fields which are the same type (int / unsigned) on our platforms.
1855 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1856 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1857 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1860 * BSD sendmsg interface
1863 SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1865 struct compat_msghdr __user *msg_compat =
1866 (struct compat_msghdr __user *)msg;
1867 struct socket *sock;
1868 struct sockaddr_storage address;
1869 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1870 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1871 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1872 /* 20 is size of ipv6_pktinfo */
1873 unsigned char *ctl_buf = ctl;
1874 struct msghdr msg_sys;
1875 int err, ctl_len, iov_size, total_len;
1879 if (MSG_CMSG_COMPAT & flags) {
1880 if (get_compat_msghdr(&msg_sys, msg_compat))
1883 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1886 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1890 /* do not move before msg_sys is valid */
1892 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1895 /* Check whether to allocate the iovec area */
1897 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1898 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1899 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1904 /* This will also move the address data into kernel space */
1905 if (MSG_CMSG_COMPAT & flags) {
1906 err = verify_compat_iovec(&msg_sys, iov,
1907 (struct sockaddr *)&address,
1910 err = verify_iovec(&msg_sys, iov,
1911 (struct sockaddr *)&address,
1919 if (msg_sys.msg_controllen > INT_MAX)
1921 ctl_len = msg_sys.msg_controllen;
1922 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1924 cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1928 ctl_buf = msg_sys.msg_control;
1929 ctl_len = msg_sys.msg_controllen;
1930 } else if (ctl_len) {
1931 if (ctl_len > sizeof(ctl)) {
1932 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1933 if (ctl_buf == NULL)
1938 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1939 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1940 * checking falls down on this.
1942 if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1945 msg_sys.msg_control = ctl_buf;
1947 msg_sys.msg_flags = flags;
1949 if (sock->file->f_flags & O_NONBLOCK)
1950 msg_sys.msg_flags |= MSG_DONTWAIT;
1951 err = sock_sendmsg(sock, &msg_sys, total_len);
1955 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1957 if (iov != iovstack)
1958 sock_kfree_s(sock->sk, iov, iov_size);
1960 fput_light(sock->file, fput_needed);
1966 * BSD recvmsg interface
1969 SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
1970 unsigned int, flags)
1972 struct compat_msghdr __user *msg_compat =
1973 (struct compat_msghdr __user *)msg;
1974 struct socket *sock;
1975 struct iovec iovstack[UIO_FASTIOV];
1976 struct iovec *iov = iovstack;
1977 struct msghdr msg_sys;
1978 unsigned long cmsg_ptr;
1979 int err, iov_size, total_len, len;
1982 /* kernel mode address */
1983 struct sockaddr_storage addr;
1985 /* user mode address pointers */
1986 struct sockaddr __user *uaddr;
1987 int __user *uaddr_len;
1989 if (MSG_CMSG_COMPAT & flags) {
1990 if (get_compat_msghdr(&msg_sys, msg_compat))
1993 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1996 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2001 if (msg_sys.msg_iovlen > UIO_MAXIOV)
2004 /* Check whether to allocate the iovec area */
2006 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
2007 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
2008 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2014 * Save the user-mode address (verify_iovec will change the
2015 * kernel msghdr to use the kernel address space)
2018 uaddr = (__force void __user *)msg_sys.msg_name;
2019 uaddr_len = COMPAT_NAMELEN(msg);
2020 if (MSG_CMSG_COMPAT & flags) {
2021 err = verify_compat_iovec(&msg_sys, iov,
2022 (struct sockaddr *)&addr,
2025 err = verify_iovec(&msg_sys, iov,
2026 (struct sockaddr *)&addr,
2032 cmsg_ptr = (unsigned long)msg_sys.msg_control;
2033 msg_sys.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2035 if (sock->file->f_flags & O_NONBLOCK)
2036 flags |= MSG_DONTWAIT;
2037 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
2042 if (uaddr != NULL) {
2043 err = move_addr_to_user((struct sockaddr *)&addr,
2044 msg_sys.msg_namelen, uaddr,
2049 err = __put_user((msg_sys.msg_flags & ~MSG_CMSG_COMPAT),
2053 if (MSG_CMSG_COMPAT & flags)
2054 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2055 &msg_compat->msg_controllen);
2057 err = __put_user((unsigned long)msg_sys.msg_control - cmsg_ptr,
2058 &msg->msg_controllen);
2064 if (iov != iovstack)
2065 sock_kfree_s(sock->sk, iov, iov_size);
2067 fput_light(sock->file, fput_needed);
2072 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2074 /* Argument list sizes for sys_socketcall */
2075 #define AL(x) ((x) * sizeof(unsigned long))
2076 static const unsigned char nargs[19]={
2077 AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2078 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
2079 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
2086 * System call vectors.
2088 * Argument checking cleaned up. Saved 20% in size.
2089 * This function doesn't need to set the kernel lock because
2090 * it is set by the callees.
2093 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2096 unsigned long a0, a1;
2099 if (call < 1 || call > SYS_ACCEPT4)
2102 /* copy_from_user should be SMP safe. */
2103 if (copy_from_user(a, args, nargs[call]))
2106 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2113 err = sys_socket(a0, a1, a[2]);
2116 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2119 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2122 err = sys_listen(a0, a1);
2125 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2126 (int __user *)a[2], 0);
2128 case SYS_GETSOCKNAME:
2130 sys_getsockname(a0, (struct sockaddr __user *)a1,
2131 (int __user *)a[2]);
2133 case SYS_GETPEERNAME:
2135 sys_getpeername(a0, (struct sockaddr __user *)a1,
2136 (int __user *)a[2]);
2138 case SYS_SOCKETPAIR:
2139 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2142 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2145 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2146 (struct sockaddr __user *)a[4], a[5]);
2149 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2152 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2153 (struct sockaddr __user *)a[4],
2154 (int __user *)a[5]);
2157 err = sys_shutdown(a0, a1);
2159 case SYS_SETSOCKOPT:
2160 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2162 case SYS_GETSOCKOPT:
2164 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2165 (int __user *)a[4]);
2168 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2171 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2174 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2175 (int __user *)a[2], a[3]);
2184 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
2187 * sock_register - add a socket protocol handler
2188 * @ops: description of protocol
2190 * This function is called by a protocol handler that wants to
2191 * advertise its address family, and have it linked into the
2192 * socket interface. The value ops->family coresponds to the
2193 * socket system call protocol family.
2195 int sock_register(const struct net_proto_family *ops)
2199 if (ops->family >= NPROTO) {
2200 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2205 spin_lock(&net_family_lock);
2206 if (net_families[ops->family])
2209 net_families[ops->family] = ops;
2212 spin_unlock(&net_family_lock);
2214 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2219 * sock_unregister - remove a protocol handler
2220 * @family: protocol family to remove
2222 * This function is called by a protocol handler that wants to
2223 * remove its address family, and have it unlinked from the
2224 * new socket creation.
2226 * If protocol handler is a module, then it can use module reference
2227 * counts to protect against new references. If protocol handler is not
2228 * a module then it needs to provide its own protection in
2229 * the ops->create routine.
2231 void sock_unregister(int family)
2233 BUG_ON(family < 0 || family >= NPROTO);
2235 spin_lock(&net_family_lock);
2236 net_families[family] = NULL;
2237 spin_unlock(&net_family_lock);
2241 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2244 static int __init sock_init(void)
2247 * Initialize sock SLAB cache.
2253 * Initialize skbuff SLAB cache
2258 * Initialize the protocols module.
2262 register_filesystem(&sock_fs_type);
2263 sock_mnt = kern_mount(&sock_fs_type);
2265 /* The real protocol initialization is performed in later initcalls.
2268 #ifdef CONFIG_NETFILTER
2275 core_initcall(sock_init); /* early initcall */
2277 #ifdef CONFIG_PROC_FS
2278 void socket_seq_show(struct seq_file *seq)
2283 for_each_possible_cpu(cpu)
2284 counter += per_cpu(sockets_in_use, cpu);
2286 /* It can be negative, by the way. 8) */
2290 seq_printf(seq, "sockets: used %d\n", counter);
2292 #endif /* CONFIG_PROC_FS */
2294 #ifdef CONFIG_COMPAT
2295 static long compat_sock_ioctl(struct file *file, unsigned cmd,
2298 struct socket *sock = file->private_data;
2299 int ret = -ENOIOCTLCMD;
2306 if (sock->ops->compat_ioctl)
2307 ret = sock->ops->compat_ioctl(sock, cmd, arg);
2309 if (ret == -ENOIOCTLCMD &&
2310 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2311 ret = compat_wext_handle_ioctl(net, cmd, arg);
2317 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2319 return sock->ops->bind(sock, addr, addrlen);
2322 int kernel_listen(struct socket *sock, int backlog)
2324 return sock->ops->listen(sock, backlog);
2327 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2329 struct sock *sk = sock->sk;
2332 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2337 err = sock->ops->accept(sock, *newsock, flags);
2339 sock_release(*newsock);
2344 (*newsock)->ops = sock->ops;
2345 __module_get((*newsock)->ops->owner);
2351 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
2354 return sock->ops->connect(sock, addr, addrlen, flags);
2357 int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2360 return sock->ops->getname(sock, addr, addrlen, 0);
2363 int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2366 return sock->ops->getname(sock, addr, addrlen, 1);
2369 int kernel_getsockopt(struct socket *sock, int level, int optname,
2370 char *optval, int *optlen)
2372 mm_segment_t oldfs = get_fs();
2376 if (level == SOL_SOCKET)
2377 err = sock_getsockopt(sock, level, optname, optval, optlen);
2379 err = sock->ops->getsockopt(sock, level, optname, optval,
2385 int kernel_setsockopt(struct socket *sock, int level, int optname,
2386 char *optval, int optlen)
2388 mm_segment_t oldfs = get_fs();
2392 if (level == SOL_SOCKET)
2393 err = sock_setsockopt(sock, level, optname, optval, optlen);
2395 err = sock->ops->setsockopt(sock, level, optname, optval,
2401 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2402 size_t size, int flags)
2404 if (sock->ops->sendpage)
2405 return sock->ops->sendpage(sock, page, offset, size, flags);
2407 return sock_no_sendpage(sock, page, offset, size, flags);
2410 int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2412 mm_segment_t oldfs = get_fs();
2416 err = sock->ops->ioctl(sock, cmd, arg);
2422 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2424 return sock->ops->shutdown(sock, how);
2427 EXPORT_SYMBOL(sock_create);
2428 EXPORT_SYMBOL(sock_create_kern);
2429 EXPORT_SYMBOL(sock_create_lite);
2430 EXPORT_SYMBOL(sock_map_fd);
2431 EXPORT_SYMBOL(sock_recvmsg);
2432 EXPORT_SYMBOL(sock_register);
2433 EXPORT_SYMBOL(sock_release);
2434 EXPORT_SYMBOL(sock_sendmsg);
2435 EXPORT_SYMBOL(sock_unregister);
2436 EXPORT_SYMBOL(sock_wake_async);
2437 EXPORT_SYMBOL(sockfd_lookup);
2438 EXPORT_SYMBOL(kernel_sendmsg);
2439 EXPORT_SYMBOL(kernel_recvmsg);
2440 EXPORT_SYMBOL(kernel_bind);
2441 EXPORT_SYMBOL(kernel_listen);
2442 EXPORT_SYMBOL(kernel_accept);
2443 EXPORT_SYMBOL(kernel_connect);
2444 EXPORT_SYMBOL(kernel_getsockname);
2445 EXPORT_SYMBOL(kernel_getpeername);
2446 EXPORT_SYMBOL(kernel_getsockopt);
2447 EXPORT_SYMBOL(kernel_setsockopt);
2448 EXPORT_SYMBOL(kernel_sendpage);
2449 EXPORT_SYMBOL(kernel_sock_ioctl);
2450 EXPORT_SYMBOL(kernel_sock_shutdown);